HMVMXR0.cpp@ 66250

Last change on this file since 66250 was 66052, checked in by vboxsync, 8 years ago
HMVMXR0.cpp: Disabled (for me only) annoying HMCPU_CF_IS_PENDING assertion triggered by bs3-cpu-generated-1.img and debugger.
Property svn:eol-style set to `native` Property svn:keywords set to `Author Date Id Revision`
File size: 587.8 KB

Line
1	/* $Id: HMVMXR0.cpp 66052 2017-03-10 19:21:28Z vboxsync $ */
2	/** @file
3	* HM VMX (Intel VT-x) - Host Context Ring-0.
4	*/
5
6	/*
7	* Copyright (C) 2012-2016 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18
19	/*********************************************************************************************************************************
20	* Header Files *
21	*********************************************************************************************************************************/
22	#define LOG_GROUP LOG_GROUP_HM
23	#include <iprt/x86.h>
24	#include <iprt/asm-amd64-x86.h>
25	#include <iprt/thread.h>
26
27	#include <VBox/vmm/pdmapi.h>
28	#include <VBox/vmm/dbgf.h>
29	#include <VBox/vmm/iem.h>
30	#include <VBox/vmm/iom.h>
31	#include <VBox/vmm/selm.h>
32	#include <VBox/vmm/tm.h>
33	#include <VBox/vmm/gim.h>
34	#include <VBox/vmm/apic.h>
35	#ifdef VBOX_WITH_REM
36	# include <VBox/vmm/rem.h>
37	#endif
38	#include "HMInternal.h"
39	#include <VBox/vmm/vm.h>
40	#include "HMVMXR0.h"
41	#include "dtrace/VBoxVMM.h"
42
43	#ifdef DEBUG_ramshankar
44	# define HMVMX_ALWAYS_SAVE_GUEST_RFLAGS
45	# define HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE
46	# define HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
47	# define HMVMX_ALWAYS_CHECK_GUEST_STATE
48	# define HMVMX_ALWAYS_TRAP_ALL_XCPTS
49	# define HMVMX_ALWAYS_TRAP_PF
50	# define HMVMX_ALWAYS_SWAP_FPU_STATE
51	# define HMVMX_ALWAYS_FLUSH_TLB
52	# define HMVMX_ALWAYS_SWAP_EFER
53	#endif
54
55
56	/*********************************************************************************************************************************
57	* Defined Constants And Macros *
58	*********************************************************************************************************************************/
59	/** Use the function table. */
60	#define HMVMX_USE_FUNCTION_TABLE
61
62	/** Determine which tagged-TLB flush handler to use. */
63	#define HMVMX_FLUSH_TAGGED_TLB_EPT_VPID 0
64	#define HMVMX_FLUSH_TAGGED_TLB_EPT 1
65	#define HMVMX_FLUSH_TAGGED_TLB_VPID 2
66	#define HMVMX_FLUSH_TAGGED_TLB_NONE 3
67
68	/** @name Updated-guest-state flags.
69	* @{ */
70	#define HMVMX_UPDATED_GUEST_RIP RT_BIT(0)
71	#define HMVMX_UPDATED_GUEST_RSP RT_BIT(1)
72	#define HMVMX_UPDATED_GUEST_RFLAGS RT_BIT(2)
73	#define HMVMX_UPDATED_GUEST_CR0 RT_BIT(3)
74	#define HMVMX_UPDATED_GUEST_CR3 RT_BIT(4)
75	#define HMVMX_UPDATED_GUEST_CR4 RT_BIT(5)
76	#define HMVMX_UPDATED_GUEST_GDTR RT_BIT(6)
77	#define HMVMX_UPDATED_GUEST_IDTR RT_BIT(7)
78	#define HMVMX_UPDATED_GUEST_LDTR RT_BIT(8)
79	#define HMVMX_UPDATED_GUEST_TR RT_BIT(9)
80	#define HMVMX_UPDATED_GUEST_SEGMENT_REGS RT_BIT(10)
81	#define HMVMX_UPDATED_GUEST_DR7 RT_BIT(11)
82	#define HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR RT_BIT(12)
83	#define HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR RT_BIT(13)
84	#define HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR RT_BIT(14)
85	#define HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS RT_BIT(15)
86	#define HMVMX_UPDATED_GUEST_LAZY_MSRS RT_BIT(16)
87	#define HMVMX_UPDATED_GUEST_ACTIVITY_STATE RT_BIT(17)
88	#define HMVMX_UPDATED_GUEST_INTR_STATE RT_BIT(18)
89	#define HMVMX_UPDATED_GUEST_APIC_STATE RT_BIT(19)
90	#define HMVMX_UPDATED_GUEST_ALL ( HMVMX_UPDATED_GUEST_RIP \
91	\| HMVMX_UPDATED_GUEST_RSP \
92	\| HMVMX_UPDATED_GUEST_RFLAGS \
93	\| HMVMX_UPDATED_GUEST_CR0 \
94	\| HMVMX_UPDATED_GUEST_CR3 \
95	\| HMVMX_UPDATED_GUEST_CR4 \
96	\| HMVMX_UPDATED_GUEST_GDTR \
97	\| HMVMX_UPDATED_GUEST_IDTR \
98	\| HMVMX_UPDATED_GUEST_LDTR \
99	\| HMVMX_UPDATED_GUEST_TR \
100	\| HMVMX_UPDATED_GUEST_SEGMENT_REGS \
101	\| HMVMX_UPDATED_GUEST_DR7 \
102	\| HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR \
103	\| HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR \
104	\| HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR \
105	\| HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS \
106	\| HMVMX_UPDATED_GUEST_LAZY_MSRS \
107	\| HMVMX_UPDATED_GUEST_ACTIVITY_STATE \
108	\| HMVMX_UPDATED_GUEST_INTR_STATE \
109	\| HMVMX_UPDATED_GUEST_APIC_STATE)
110	/** @} */
111
112	/** @name
113	* Flags to skip redundant reads of some common VMCS fields that are not part of
114	* the guest-CPU state but are in the transient structure.
115	*/
116	#define HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO RT_BIT(0)
117	#define HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE RT_BIT(1)
118	#define HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION RT_BIT(2)
119	#define HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN RT_BIT(3)
120	#define HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO RT_BIT(4)
121	#define HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE RT_BIT(5)
122	#define HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO RT_BIT(6)
123	/** @} */
124
125	/** @name
126	* States of the VMCS.
127	*
128	* This does not reflect all possible VMCS states but currently only those
129	* needed for maintaining the VMCS consistently even when thread-context hooks
130	* are used. Maybe later this can be extended (i.e. Nested Virtualization).
131	*/
132	#define HMVMX_VMCS_STATE_CLEAR RT_BIT(0)
133	#define HMVMX_VMCS_STATE_ACTIVE RT_BIT(1)
134	#define HMVMX_VMCS_STATE_LAUNCHED RT_BIT(2)
135	/** @} */
136
137	/**
138	* Exception bitmap mask for real-mode guests (real-on-v86).
139	*
140	* We need to intercept all exceptions manually except:
141	* - \#NM, \#MF handled in hmR0VmxLoadSharedCR0().
142	* - \#AC and \#DB are always intercepted to prevent the CPU from deadlocking
143	* due to bugs in Intel CPUs.
144	* - \#PF need not be intercepted even in real-mode if we have Nested Paging
145	* support.
146	*/
147	#define HMVMX_REAL_MODE_XCPT_MASK ( RT_BIT(X86_XCPT_DE) /* always: \| RT_BIT(X86_XCPT_DB) */ \| RT_BIT(X86_XCPT_NMI) \
148	\| RT_BIT(X86_XCPT_BP) \| RT_BIT(X86_XCPT_OF) \| RT_BIT(X86_XCPT_BR) \
149	\| RT_BIT(X86_XCPT_UD) /* RT_BIT(X86_XCPT_NM) */ \| RT_BIT(X86_XCPT_DF) \
150	\| RT_BIT(X86_XCPT_CO_SEG_OVERRUN) \| RT_BIT(X86_XCPT_TS) \| RT_BIT(X86_XCPT_NP) \
151	\| RT_BIT(X86_XCPT_SS) \| RT_BIT(X86_XCPT_GP) /* RT_BIT(X86_XCPT_PF) */ \
152	/* RT_BIT(X86_XCPT_MF) always: \| RT_BIT(X86_XCPT_AC) */ \| RT_BIT(X86_XCPT_MC) \
153	\| RT_BIT(X86_XCPT_XF))
154
155	/**
156	* Exception bitmap mask for all contributory exceptions.
157	*
158	* Page fault is deliberately excluded here as it's conditional as to whether
159	* it's contributory or benign. Page faults are handled separately.
160	*/
161	#define HMVMX_CONTRIBUTORY_XCPT_MASK ( RT_BIT(X86_XCPT_GP) \| RT_BIT(X86_XCPT_NP) \| RT_BIT(X86_XCPT_SS) \| RT_BIT(X86_XCPT_TS) \
162	\| RT_BIT(X86_XCPT_DE))
163
164	/** Maximum VM-instruction error number. */
165	#define HMVMX_INSTR_ERROR_MAX 28
166
167	/** Profiling macro. */
168	#ifdef HM_PROFILE_EXIT_DISPATCH
169	# define HMVMX_START_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitDispatch, ed)
170	# define HMVMX_STOP_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitDispatch, ed)
171	#else
172	# define HMVMX_START_EXIT_DISPATCH_PROF() do { } while (0)
173	# define HMVMX_STOP_EXIT_DISPATCH_PROF() do { } while (0)
174	#endif
175
176	/** Assert that preemption is disabled or covered by thread-context hooks. */
177	#define HMVMX_ASSERT_PREEMPT_SAFE() Assert( VMMR0ThreadCtxHookIsEnabled(pVCpu) \
178	\|\| !RTThreadPreemptIsEnabled(NIL_RTTHREAD));
179
180	/** Assert that we haven't migrated CPUs when thread-context hooks are not
181	* used. */
182	#define HMVMX_ASSERT_CPU_SAFE() AssertMsg( VMMR0ThreadCtxHookIsEnabled(pVCpu) \
183	\|\| pVCpu->hm.s.idEnteredCpu == RTMpCpuId(), \
184	("Illegal migration! Entered on CPU %u Current %u\n", \
185	pVCpu->hm.s.idEnteredCpu, RTMpCpuId())); \
186
187	/** Helper macro for VM-exit handlers called unexpectedly. */
188	#define HMVMX_RETURN_UNEXPECTED_EXIT() \
189	do { \
190	pVCpu->hm.s.u32HMError = pVmxTransient->uExitReason; \
191	return VERR_VMX_UNEXPECTED_EXIT; \
192	} while (0)
193
194
195	/*********************************************************************************************************************************
196	* Structures and Typedefs *
197	*********************************************************************************************************************************/
198	/**
199	* VMX transient state.
200	*
201	* A state structure for holding miscellaneous information across
202	* VMX non-root operation and restored after the transition.
203	*/
204	typedef struct VMXTRANSIENT
205	{
206	/** The host's rflags/eflags. */
207	RTCCUINTREG fEFlags;
208	#if HC_ARCH_BITS == 32
209	uint32_t u32Alignment0;
210	#endif
211	/** The guest's TPR value used for TPR shadowing. */
212	uint8_t u8GuestTpr;
213	/** Alignment. */
214	uint8_t abAlignment0[7];
215
216	/** The basic VM-exit reason. */
217	uint16_t uExitReason;
218	/** Alignment. */
219	uint16_t u16Alignment0;
220	/** The VM-exit interruption error code. */
221	uint32_t uExitIntErrorCode;
222	/** The VM-exit exit code qualification. */
223	uint64_t uExitQualification;
224
225	/** The VM-exit interruption-information field. */
226	uint32_t uExitIntInfo;
227	/** The VM-exit instruction-length field. */
228	uint32_t cbInstr;
229	/** The VM-exit instruction-information field. */
230	union
231	{
232	/** Plain unsigned int representation. */
233	uint32_t u;
234	/** INS and OUTS information. */
235	struct
236	{
237	uint32_t u7Reserved0 : 7;
238	/** The address size; 0=16-bit, 1=32-bit, 2=64-bit, rest undefined. */
239	uint32_t u3AddrSize : 3;
240	uint32_t u5Reserved1 : 5;
241	/** The segment register (X86_SREG_XXX). */
242	uint32_t iSegReg : 3;
243	uint32_t uReserved2 : 14;
244	} StrIo;
245	} ExitInstrInfo;
246	/** Whether the VM-entry failed or not. */
247	bool fVMEntryFailed;
248	/** Alignment. */
249	uint8_t abAlignment1[3];
250
251	/** The VM-entry interruption-information field. */
252	uint32_t uEntryIntInfo;
253	/** The VM-entry exception error code field. */
254	uint32_t uEntryXcptErrorCode;
255	/** The VM-entry instruction length field. */
256	uint32_t cbEntryInstr;
257
258	/** IDT-vectoring information field. */
259	uint32_t uIdtVectoringInfo;
260	/** IDT-vectoring error code. */
261	uint32_t uIdtVectoringErrorCode;
262
263	/** Mask of currently read VMCS fields; HMVMX_UPDATED_TRANSIENT_. /
264	uint32_t fVmcsFieldsRead;
265
266	/** Whether the guest FPU was active at the time of VM-exit. */
267	bool fWasGuestFPUStateActive;
268	/** Whether the guest debug state was active at the time of VM-exit. */
269	bool fWasGuestDebugStateActive;
270	/** Whether the hyper debug state was active at the time of VM-exit. */
271	bool fWasHyperDebugStateActive;
272	/** Whether TSC-offsetting should be setup before VM-entry. */
273	bool fUpdateTscOffsettingAndPreemptTimer;
274	/** Whether the VM-exit was caused by a page-fault during delivery of a
275	* contributory exception or a page-fault. */
276	bool fVectoringDoublePF;
277	/** Whether the VM-exit was caused by a page-fault during delivery of an
278	* external interrupt or NMI. */
279	bool fVectoringPF;
280	} VMXTRANSIENT;
281	AssertCompileMemberAlignment(VMXTRANSIENT, uExitReason, sizeof(uint64_t));
282	AssertCompileMemberAlignment(VMXTRANSIENT, uExitIntInfo, sizeof(uint64_t));
283	AssertCompileMemberAlignment(VMXTRANSIENT, uEntryIntInfo, sizeof(uint64_t));
284	AssertCompileMemberAlignment(VMXTRANSIENT, fWasGuestFPUStateActive, sizeof(uint64_t));
285	AssertCompileMemberSize(VMXTRANSIENT, ExitInstrInfo, sizeof(uint32_t));
286	/** Pointer to VMX transient state. */
287	typedef VMXTRANSIENT *PVMXTRANSIENT;
288
289
290	/**
291	* MSR-bitmap read permissions.
292	*/
293	typedef enum VMXMSREXITREAD
294	{
295	/** Reading this MSR causes a VM-exit. */
296	VMXMSREXIT_INTERCEPT_READ = 0xb,
297	/** Reading this MSR does not cause a VM-exit. */
298	VMXMSREXIT_PASSTHRU_READ
299	} VMXMSREXITREAD;
300	/** Pointer to MSR-bitmap read permissions. */
301	typedef VMXMSREXITREAD* PVMXMSREXITREAD;
302
303	/**
304	* MSR-bitmap write permissions.
305	*/
306	typedef enum VMXMSREXITWRITE
307	{
308	/** Writing to this MSR causes a VM-exit. */
309	VMXMSREXIT_INTERCEPT_WRITE = 0xd,
310	/** Writing to this MSR does not cause a VM-exit. */
311	VMXMSREXIT_PASSTHRU_WRITE
312	} VMXMSREXITWRITE;
313	/** Pointer to MSR-bitmap write permissions. */
314	typedef VMXMSREXITWRITE* PVMXMSREXITWRITE;
315
316
317	/**
318	* VMX VM-exit handler.
319	*
320	* @returns Strict VBox status code (i.e. informational status codes too).
321	* @param pVCpu The cross context virtual CPU structure.
322	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
323	* out-of-sync. Make sure to update the required
324	* fields before using them.
325	* @param pVmxTransient Pointer to the VMX-transient structure.
326	*/
327	#ifndef HMVMX_USE_FUNCTION_TABLE
328	typedef VBOXSTRICTRC FNVMXEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
329	#else
330	typedef DECLCALLBACK(VBOXSTRICTRC) FNVMXEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
331	/** Pointer to VM-exit handler. */
332	typedef FNVMXEXITHANDLER *PFNVMXEXITHANDLER;
333	#endif
334
335	/**
336	* VMX VM-exit handler, non-strict status code.
337	*
338	* This is generally the same as FNVMXEXITHANDLER, the NSRC bit is just FYI.
339	*
340	* @returns VBox status code, no informational status code returned.
341	* @param pVCpu The cross context virtual CPU structure.
342	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
343	* out-of-sync. Make sure to update the required
344	* fields before using them.
345	* @param pVmxTransient Pointer to the VMX-transient structure.
346	*
347	* @remarks This is not used on anything returning VERR_EM_INTERPRETER as the
348	* use of that status code will be replaced with VINF_EM_SOMETHING
349	* later when switching over to IEM.
350	*/
351	#ifndef HMVMX_USE_FUNCTION_TABLE
352	typedef int FNVMXEXITHANDLERNSRC(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
353	#else
354	typedef FNVMXEXITHANDLER FNVMXEXITHANDLERNSRC;
355	#endif
356
357
358	/*********************************************************************************************************************************
359	* Internal Functions *
360	*********************************************************************************************************************************/
361	static void hmR0VmxFlushEpt(PVMCPU pVCpu, VMXFLUSHEPT enmFlush);
362	static void hmR0VmxFlushVpid(PVM pVM, PVMCPU pVCpu, VMXFLUSHVPID enmFlush, RTGCPTR GCPtr);
363	static void hmR0VmxClearIntNmiWindowsVmcs(PVMCPU pVCpu);
364	static VBOXSTRICTRC hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint64_t u64IntInfo, uint32_t cbInstr,
365	uint32_t u32ErrCode, RTGCUINTREG GCPtrFaultAddress,
366	bool fStepping, uint32_t *puIntState);
367	#if HC_ARCH_BITS == 32
368	static int hmR0VmxInitVmcsReadCache(PVM pVM, PVMCPU pVCpu);
369	#endif
370	#ifndef HMVMX_USE_FUNCTION_TABLE
371	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, uint32_t rcReason);
372	# define HMVMX_EXIT_DECL DECLINLINE(VBOXSTRICTRC)
373	# define HMVMX_EXIT_NSRC_DECL DECLINLINE(int)
374	#else
375	# define HMVMX_EXIT_DECL static DECLCALLBACK(VBOXSTRICTRC)
376	# define HMVMX_EXIT_NSRC_DECL HMVMX_EXIT_DECL
377	#endif
378
379
380	/** @name VM-exit handlers.
381	* @{
382	*/
383	static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmi;
384	static FNVMXEXITHANDLER hmR0VmxExitExtInt;
385	static FNVMXEXITHANDLER hmR0VmxExitTripleFault;
386	static FNVMXEXITHANDLERNSRC hmR0VmxExitInitSignal;
387	static FNVMXEXITHANDLERNSRC hmR0VmxExitSipi;
388	static FNVMXEXITHANDLERNSRC hmR0VmxExitIoSmi;
389	static FNVMXEXITHANDLERNSRC hmR0VmxExitSmi;
390	static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindow;
391	static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindow;
392	static FNVMXEXITHANDLER hmR0VmxExitTaskSwitch;
393	static FNVMXEXITHANDLER hmR0VmxExitCpuid;
394	static FNVMXEXITHANDLER hmR0VmxExitGetsec;
395	static FNVMXEXITHANDLER hmR0VmxExitHlt;
396	static FNVMXEXITHANDLERNSRC hmR0VmxExitInvd;
397	static FNVMXEXITHANDLER hmR0VmxExitInvlpg;
398	static FNVMXEXITHANDLER hmR0VmxExitRdpmc;
399	static FNVMXEXITHANDLER hmR0VmxExitVmcall;
400	static FNVMXEXITHANDLER hmR0VmxExitRdtsc;
401	static FNVMXEXITHANDLERNSRC hmR0VmxExitRsm;
402	static FNVMXEXITHANDLERNSRC hmR0VmxExitSetPendingXcptUD;
403	static FNVMXEXITHANDLER hmR0VmxExitMovCRx;
404	static FNVMXEXITHANDLER hmR0VmxExitMovDRx;
405	static FNVMXEXITHANDLER hmR0VmxExitIoInstr;
406	static FNVMXEXITHANDLER hmR0VmxExitRdmsr;
407	static FNVMXEXITHANDLER hmR0VmxExitWrmsr;
408	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestState;
409	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrMsrLoad;
410	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrUndefined;
411	static FNVMXEXITHANDLER hmR0VmxExitMwait;
412	static FNVMXEXITHANDLER hmR0VmxExitMtf;
413	static FNVMXEXITHANDLER hmR0VmxExitMonitor;
414	static FNVMXEXITHANDLER hmR0VmxExitPause;
415	static FNVMXEXITHANDLERNSRC hmR0VmxExitErrMachineCheck;
416	static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThreshold;
417	static FNVMXEXITHANDLER hmR0VmxExitApicAccess;
418	static FNVMXEXITHANDLER hmR0VmxExitXdtrAccess;
419	static FNVMXEXITHANDLER hmR0VmxExitXdtrAccess;
420	static FNVMXEXITHANDLER hmR0VmxExitEptViolation;
421	static FNVMXEXITHANDLER hmR0VmxExitEptMisconfig;
422	static FNVMXEXITHANDLER hmR0VmxExitRdtscp;
423	static FNVMXEXITHANDLER hmR0VmxExitPreemptTimer;
424	static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvd;
425	static FNVMXEXITHANDLER hmR0VmxExitXsetbv;
426	static FNVMXEXITHANDLER hmR0VmxExitRdrand;
427	static FNVMXEXITHANDLER hmR0VmxExitInvpcid;
428	/** @} */
429
430	static int hmR0VmxExitXcptNM(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
431	static int hmR0VmxExitXcptPF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
432	static int hmR0VmxExitXcptMF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
433	static int hmR0VmxExitXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
434	static int hmR0VmxExitXcptBP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
435	static int hmR0VmxExitXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
436	static int hmR0VmxExitXcptAC(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
437	static int hmR0VmxExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
438	static uint32_t hmR0VmxCheckGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx);
439
440
441	/*********************************************************************************************************************************
442	* Global Variables *
443	*********************************************************************************************************************************/
444	#ifdef HMVMX_USE_FUNCTION_TABLE
445
446	/**
447	* VMX_EXIT dispatch table.
448	*/
449	static const PFNVMXEXITHANDLER g_apfnVMExitHandlers[VMX_EXIT_MAX + 1] =
450	{
451	/* 00 VMX_EXIT_XCPT_OR_NMI */ hmR0VmxExitXcptOrNmi,
452	/* 01 VMX_EXIT_EXT_INT */ hmR0VmxExitExtInt,
453	/* 02 VMX_EXIT_TRIPLE_FAULT */ hmR0VmxExitTripleFault,
454	/* 03 VMX_EXIT_INIT_SIGNAL */ hmR0VmxExitInitSignal,
455	/* 04 VMX_EXIT_SIPI */ hmR0VmxExitSipi,
456	/* 05 VMX_EXIT_IO_SMI */ hmR0VmxExitIoSmi,
457	/* 06 VMX_EXIT_SMI */ hmR0VmxExitSmi,
458	/* 07 VMX_EXIT_INT_WINDOW */ hmR0VmxExitIntWindow,
459	/* 08 VMX_EXIT_NMI_WINDOW */ hmR0VmxExitNmiWindow,
460	/* 09 VMX_EXIT_TASK_SWITCH */ hmR0VmxExitTaskSwitch,
461	/* 10 VMX_EXIT_CPUID */ hmR0VmxExitCpuid,
462	/* 11 VMX_EXIT_GETSEC */ hmR0VmxExitGetsec,
463	/* 12 VMX_EXIT_HLT */ hmR0VmxExitHlt,
464	/* 13 VMX_EXIT_INVD */ hmR0VmxExitInvd,
465	/* 14 VMX_EXIT_INVLPG */ hmR0VmxExitInvlpg,
466	/* 15 VMX_EXIT_RDPMC */ hmR0VmxExitRdpmc,
467	/* 16 VMX_EXIT_RDTSC */ hmR0VmxExitRdtsc,
468	/* 17 VMX_EXIT_RSM */ hmR0VmxExitRsm,
469	/* 18 VMX_EXIT_VMCALL */ hmR0VmxExitVmcall,
470	/* 19 VMX_EXIT_VMCLEAR */ hmR0VmxExitSetPendingXcptUD,
471	/* 20 VMX_EXIT_VMLAUNCH */ hmR0VmxExitSetPendingXcptUD,
472	/* 21 VMX_EXIT_VMPTRLD */ hmR0VmxExitSetPendingXcptUD,
473	/* 22 VMX_EXIT_VMPTRST */ hmR0VmxExitSetPendingXcptUD,
474	/* 23 VMX_EXIT_VMREAD */ hmR0VmxExitSetPendingXcptUD,
475	/* 24 VMX_EXIT_VMRESUME */ hmR0VmxExitSetPendingXcptUD,
476	/* 25 VMX_EXIT_VMWRITE */ hmR0VmxExitSetPendingXcptUD,
477	/* 26 VMX_EXIT_VMXOFF */ hmR0VmxExitSetPendingXcptUD,
478	/* 27 VMX_EXIT_VMXON */ hmR0VmxExitSetPendingXcptUD,
479	/* 28 VMX_EXIT_MOV_CRX */ hmR0VmxExitMovCRx,
480	/* 29 VMX_EXIT_MOV_DRX */ hmR0VmxExitMovDRx,
481	/* 30 VMX_EXIT_IO_INSTR */ hmR0VmxExitIoInstr,
482	/* 31 VMX_EXIT_RDMSR */ hmR0VmxExitRdmsr,
483	/* 32 VMX_EXIT_WRMSR */ hmR0VmxExitWrmsr,
484	/* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ hmR0VmxExitErrInvalidGuestState,
485	/* 34 VMX_EXIT_ERR_MSR_LOAD */ hmR0VmxExitErrMsrLoad,
486	/* 35 UNDEFINED */ hmR0VmxExitErrUndefined,
487	/* 36 VMX_EXIT_MWAIT */ hmR0VmxExitMwait,
488	/* 37 VMX_EXIT_MTF */ hmR0VmxExitMtf,
489	/* 38 UNDEFINED */ hmR0VmxExitErrUndefined,
490	/* 39 VMX_EXIT_MONITOR */ hmR0VmxExitMonitor,
491	/* 40 UNDEFINED */ hmR0VmxExitPause,
492	/* 41 VMX_EXIT_PAUSE */ hmR0VmxExitErrMachineCheck,
493	/* 42 VMX_EXIT_ERR_MACHINE_CHECK */ hmR0VmxExitErrUndefined,
494	/* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ hmR0VmxExitTprBelowThreshold,
495	/* 44 VMX_EXIT_APIC_ACCESS */ hmR0VmxExitApicAccess,
496	/* 45 UNDEFINED */ hmR0VmxExitErrUndefined,
497	/* 46 VMX_EXIT_XDTR_ACCESS */ hmR0VmxExitXdtrAccess,
498	/* 47 VMX_EXIT_TR_ACCESS */ hmR0VmxExitXdtrAccess,
499	/* 48 VMX_EXIT_EPT_VIOLATION */ hmR0VmxExitEptViolation,
500	/* 49 VMX_EXIT_EPT_MISCONFIG */ hmR0VmxExitEptMisconfig,
501	/* 50 VMX_EXIT_INVEPT */ hmR0VmxExitSetPendingXcptUD,
502	/* 51 VMX_EXIT_RDTSCP */ hmR0VmxExitRdtscp,
503	/* 52 VMX_EXIT_PREEMPT_TIMER */ hmR0VmxExitPreemptTimer,
504	/* 53 VMX_EXIT_INVVPID */ hmR0VmxExitSetPendingXcptUD,
505	/* 54 VMX_EXIT_WBINVD */ hmR0VmxExitWbinvd,
506	/* 55 VMX_EXIT_XSETBV */ hmR0VmxExitXsetbv,
507	/* 56 VMX_EXIT_APIC_WRITE */ hmR0VmxExitErrUndefined,
508	/* 57 VMX_EXIT_RDRAND */ hmR0VmxExitRdrand,
509	/* 58 VMX_EXIT_INVPCID */ hmR0VmxExitInvpcid,
510	/* 59 VMX_EXIT_VMFUNC */ hmR0VmxExitSetPendingXcptUD,
511	/* 60 VMX_EXIT_ENCLS */ hmR0VmxExitErrUndefined,
512	/* 61 VMX_EXIT_RDSEED / hmR0VmxExitErrUndefined, / only spurious exits, so undefined */
513	/* 62 VMX_EXIT_PML_FULL */ hmR0VmxExitErrUndefined,
514	/* 63 VMX_EXIT_XSAVES */ hmR0VmxExitSetPendingXcptUD,
515	/* 64 VMX_EXIT_XRSTORS */ hmR0VmxExitSetPendingXcptUD,
516	};
517	#endif /* HMVMX_USE_FUNCTION_TABLE */
518
519	#ifdef VBOX_STRICT
520	static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] =
521	{
522	/* 0 */ "(Not Used)",
523	/* 1 */ "VMCALL executed in VMX root operation.",
524	/* 2 */ "VMCLEAR with invalid physical address.",
525	/* 3 */ "VMCLEAR with VMXON pointer.",
526	/* 4 */ "VMLAUNCH with non-clear VMCS.",
527	/* 5 */ "VMRESUME with non-launched VMCS.",
528	/* 6 */ "VMRESUME after VMXOFF",
529	/* 7 */ "VM-entry with invalid control fields.",
530	/* 8 */ "VM-entry with invalid host state fields.",
531	/* 9 */ "VMPTRLD with invalid physical address.",
532	/* 10 */ "VMPTRLD with VMXON pointer.",
533	/* 11 */ "VMPTRLD with incorrect revision identifier.",
534	/* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.",
535	/* 13 */ "VMWRITE to read-only VMCS component.",
536	/* 14 */ "(Not Used)",
537	/* 15 */ "VMXON executed in VMX root operation.",
538	/* 16 */ "VM-entry with invalid executive-VMCS pointer.",
539	/* 17 */ "VM-entry with non-launched executing VMCS.",
540	/* 18 */ "VM-entry with executive-VMCS pointer not VMXON pointer.",
541	/* 19 */ "VMCALL with non-clear VMCS.",
542	/* 20 */ "VMCALL with invalid VM-exit control fields.",
543	/* 21 */ "(Not Used)",
544	/* 22 */ "VMCALL with incorrect MSEG revision identifier.",
545	/* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.",
546	/* 24 */ "VMCALL with invalid SMM-monitor features.",
547	/* 25 */ "VM-entry with invalid VM-execution control fields in executive VMCS.",
548	/* 26 */ "VM-entry with events blocked by MOV SS.",
549	/* 27 */ "(Not Used)",
550	/* 28 */ "Invalid operand to INVEPT/INVVPID."
551	};
552	#endif /* VBOX_STRICT */
553
554
555
556	/**
557	* Updates the VM's last error record.
558	*
559	* If there was a VMX instruction error, reads the error data from the VMCS and
560	* updates VCPU's last error record as well.
561	*
562	* @param pVM The cross context VM structure.
563	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
564	* Can be NULL if @a rc is not VERR_VMX_UNABLE_TO_START_VM or
565	* VERR_VMX_INVALID_VMCS_FIELD.
566	* @param rc The error code.
567	*/
568	static void hmR0VmxUpdateErrorRecord(PVM pVM, PVMCPU pVCpu, int rc)
569	{
570	AssertPtr(pVM);
571	if ( rc == VERR_VMX_INVALID_VMCS_FIELD
572	\|\| rc == VERR_VMX_UNABLE_TO_START_VM)
573	{
574	AssertPtrReturnVoid(pVCpu);
575	VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
576	}
577	pVM->hm.s.lLastError = rc;
578	}
579
580
581	/**
582	* Reads the VM-entry interruption-information field from the VMCS into the VMX
583	* transient structure.
584	*
585	* @returns VBox status code.
586	* @param pVmxTransient Pointer to the VMX transient structure.
587	*
588	* @remarks No-long-jump zone!!!
589	*/
590	DECLINLINE(int) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
591	{
592	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo);
593	AssertRCReturn(rc, rc);
594	return VINF_SUCCESS;
595	}
596
597
598	#ifdef VBOX_STRICT
599	/**
600	* Reads the VM-entry exception error code field from the VMCS into
601	* the VMX transient structure.
602	*
603	* @returns VBox status code.
604	* @param pVmxTransient Pointer to the VMX transient structure.
605	*
606	* @remarks No-long-jump zone!!!
607	*/
608	DECLINLINE(int) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
609	{
610	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode);
611	AssertRCReturn(rc, rc);
612	return VINF_SUCCESS;
613	}
614	#endif /* VBOX_STRICT */
615
616
617	#ifdef VBOX_STRICT
618	/**
619	* Reads the VM-entry exception error code field from the VMCS into
620	* the VMX transient structure.
621	*
622	* @returns VBox status code.
623	* @param pVmxTransient Pointer to the VMX transient structure.
624	*
625	* @remarks No-long-jump zone!!!
626	*/
627	DECLINLINE(int) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
628	{
629	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr);
630	AssertRCReturn(rc, rc);
631	return VINF_SUCCESS;
632	}
633	#endif /* VBOX_STRICT */
634
635
636	/**
637	* Reads the VM-exit interruption-information field from the VMCS into the VMX
638	* transient structure.
639	*
640	* @returns VBox status code.
641	* @param pVmxTransient Pointer to the VMX transient structure.
642	*/
643	DECLINLINE(int) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
644	{
645	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO))
646	{
647	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
648	AssertRCReturn(rc, rc);
649	pVmxTransient->fVmcsFieldsRead \|= HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO;
650	}
651	return VINF_SUCCESS;
652	}
653
654
655	/**
656	* Reads the VM-exit interruption error code from the VMCS into the VMX
657	* transient structure.
658	*
659	* @returns VBox status code.
660	* @param pVmxTransient Pointer to the VMX transient structure.
661	*/
662	DECLINLINE(int) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
663	{
664	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE))
665	{
666	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
667	AssertRCReturn(rc, rc);
668	pVmxTransient->fVmcsFieldsRead \|= HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE;
669	}
670	return VINF_SUCCESS;
671	}
672
673
674	/**
675	* Reads the VM-exit instruction length field from the VMCS into the VMX
676	* transient structure.
677	*
678	* @returns VBox status code.
679	* @param pVmxTransient Pointer to the VMX transient structure.
680	*/
681	DECLINLINE(int) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
682	{
683	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN))
684	{
685	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbInstr);
686	AssertRCReturn(rc, rc);
687	pVmxTransient->fVmcsFieldsRead \|= HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN;
688	}
689	return VINF_SUCCESS;
690	}
691
692
693	/**
694	* Reads the VM-exit instruction-information field from the VMCS into
695	* the VMX transient structure.
696	*
697	* @returns VBox status code.
698	* @param pVmxTransient Pointer to the VMX transient structure.
699	*/
700	DECLINLINE(int) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient)
701	{
702	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO))
703	{
704	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
705	AssertRCReturn(rc, rc);
706	pVmxTransient->fVmcsFieldsRead \|= HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO;
707	}
708	return VINF_SUCCESS;
709	}
710
711
712	/**
713	* Reads the exit code qualification from the VMCS into the VMX transient
714	* structure.
715	*
716	* @returns VBox status code.
717	* @param pVCpu The cross context virtual CPU structure of the
718	* calling EMT. (Required for the VMCS cache case.)
719	* @param pVmxTransient Pointer to the VMX transient structure.
720	*/
721	DECLINLINE(int) hmR0VmxReadExitQualificationVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
722	{
723	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION))
724	{
725	int rc = VMXReadVmcsGstN(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQualification); NOREF(pVCpu);
726	AssertRCReturn(rc, rc);
727	pVmxTransient->fVmcsFieldsRead \|= HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION;
728	}
729	return VINF_SUCCESS;
730	}
731
732
733	/**
734	* Reads the IDT-vectoring information field from the VMCS into the VMX
735	* transient structure.
736	*
737	* @returns VBox status code.
738	* @param pVmxTransient Pointer to the VMX transient structure.
739	*
740	* @remarks No-long-jump zone!!!
741	*/
742	DECLINLINE(int) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient)
743	{
744	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO))
745	{
746	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_INFO, &pVmxTransient->uIdtVectoringInfo);
747	AssertRCReturn(rc, rc);
748	pVmxTransient->fVmcsFieldsRead \|= HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO;
749	}
750	return VINF_SUCCESS;
751	}
752
753
754	/**
755	* Reads the IDT-vectoring error code from the VMCS into the VMX
756	* transient structure.
757	*
758	* @returns VBox status code.
759	* @param pVmxTransient Pointer to the VMX transient structure.
760	*/
761	DECLINLINE(int) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
762	{
763	if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE))
764	{
765	int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
766	AssertRCReturn(rc, rc);
767	pVmxTransient->fVmcsFieldsRead \|= HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE;
768	}
769	return VINF_SUCCESS;
770	}
771
772
773	/**
774	* Enters VMX root mode operation on the current CPU.
775	*
776	* @returns VBox status code.
777	* @param pVM The cross context VM structure. Can be
778	* NULL, after a resume.
779	* @param HCPhysCpuPage Physical address of the VMXON region.
780	* @param pvCpuPage Pointer to the VMXON region.
781	*/
782	static int hmR0VmxEnterRootMode(PVM pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage)
783	{
784	Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
785	Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
786	Assert(pvCpuPage);
787	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
788
789	if (pVM)
790	{
791	/* Write the VMCS revision dword to the VMXON region. */
792	(uint32_t )pvCpuPage = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo);
793	}
794
795	/* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */
796	RTCCUINTREG fEFlags = ASMIntDisableFlags();
797
798	/* Enable the VMX bit in CR4 if necessary. */
799	RTCCUINTREG uOldCr4 = SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
800
801	/* Enter VMX root mode. */
802	int rc = VMXEnable(HCPhysCpuPage);
803	if (RT_FAILURE(rc))
804	{
805	if (!(uOldCr4 & X86_CR4_VMXE))
806	SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
807
808	if (pVM)
809	pVM->hm.s.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
810	}
811
812	/* Restore interrupts. */
813	ASMSetFlags(fEFlags);
814	return rc;
815	}
816
817
818	/**
819	* Exits VMX root mode operation on the current CPU.
820	*
821	* @returns VBox status code.
822	*/
823	static int hmR0VmxLeaveRootMode(void)
824	{
825	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
826
827	/* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */
828	RTCCUINTREG fEFlags = ASMIntDisableFlags();
829
830	/* If we're for some reason not in VMX root mode, then don't leave it. */
831	RTCCUINTREG uHostCR4 = ASMGetCR4();
832
833	int rc;
834	if (uHostCR4 & X86_CR4_VMXE)
835	{
836	/* Exit VMX root mode and clear the VMX bit in CR4. */
837	VMXDisable();
838	SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
839	rc = VINF_SUCCESS;
840	}
841	else
842	rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE;
843
844	/* Restore interrupts. */
845	ASMSetFlags(fEFlags);
846	return rc;
847	}
848
849
850	/**
851	* Allocates and maps one physically contiguous page. The allocated page is
852	* zero'd out. (Used by various VT-x structures).
853	*
854	* @returns IPRT status code.
855	* @param pMemObj Pointer to the ring-0 memory object.
856	* @param ppVirt Where to store the virtual address of the
857	* allocation.
858	* @param pHCPhys Where to store the physical address of the
859	* allocation.
860	*/
861	DECLINLINE(int) hmR0VmxPageAllocZ(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
862	{
863	AssertPtrReturn(pMemObj, VERR_INVALID_PARAMETER);
864	AssertPtrReturn(ppVirt, VERR_INVALID_PARAMETER);
865	AssertPtrReturn(pHCPhys, VERR_INVALID_PARAMETER);
866
867	int rc = RTR0MemObjAllocCont(pMemObj, PAGE_SIZE, false /* fExecutable */);
868	if (RT_FAILURE(rc))
869	return rc;
870	ppVirt = RTR0MemObjAddress(pMemObj);
871	pHCPhys = RTR0MemObjGetPagePhysAddr(pMemObj, 0 /* iPage */);
872	ASMMemZero32(*ppVirt, PAGE_SIZE);
873	return VINF_SUCCESS;
874	}
875
876
877	/**
878	* Frees and unmaps an allocated physical page.
879	*
880	* @param pMemObj Pointer to the ring-0 memory object.
881	* @param ppVirt Where to re-initialize the virtual address of
882	* allocation as 0.
883	* @param pHCPhys Where to re-initialize the physical address of the
884	* allocation as 0.
885	*/
886	DECLINLINE(void) hmR0VmxPageFree(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
887	{
888	AssertPtr(pMemObj);
889	AssertPtr(ppVirt);
890	AssertPtr(pHCPhys);
891	if (*pMemObj != NIL_RTR0MEMOBJ)
892	{
893	int rc = RTR0MemObjFree(pMemObj, true / fFreeMappings */);
894	AssertRC(rc);
895	*pMemObj = NIL_RTR0MEMOBJ;
896	*ppVirt = 0;
897	*pHCPhys = 0;
898	}
899	}
900
901
902	/**
903	* Worker function to free VT-x related structures.
904	*
905	* @returns IPRT status code.
906	* @param pVM The cross context VM structure.
907	*/
908	static void hmR0VmxStructsFree(PVM pVM)
909	{
910	for (VMCPUID i = 0; i < pVM->cCpus; i++)
911	{
912	PVMCPU pVCpu = &pVM->aCpus[i];
913	AssertPtr(pVCpu);
914
915	hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjHostMsr, &pVCpu->hm.s.vmx.pvHostMsr, &pVCpu->hm.s.vmx.HCPhysHostMsr);
916	hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjGuestMsr, &pVCpu->hm.s.vmx.pvGuestMsr, &pVCpu->hm.s.vmx.HCPhysGuestMsr);
917
918	if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
919	hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjMsrBitmap, &pVCpu->hm.s.vmx.pvMsrBitmap, &pVCpu->hm.s.vmx.HCPhysMsrBitmap);
920
921	hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjVmcs, &pVCpu->hm.s.vmx.pvVmcs, &pVCpu->hm.s.vmx.HCPhysVmcs);
922	}
923
924	hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess, &pVM->hm.s.vmx.HCPhysApicAccess);
925	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
926	hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
927	#endif
928	}
929
930
931	/**
932	* Worker function to allocate VT-x related VM structures.
933	*
934	* @returns IPRT status code.
935	* @param pVM The cross context VM structure.
936	*/
937	static int hmR0VmxStructsAlloc(PVM pVM)
938	{
939	/*
940	* Initialize members up-front so we can cleanup properly on allocation failure.
941	*/
942	#define VMXLOCAL_INIT_VM_MEMOBJ(a_Name, a_VirtPrefix) \
943	pVM->hm.s.vmx.hMemObj##a_Name = NIL_RTR0MEMOBJ; \
944	pVM->hm.s.vmx.a_VirtPrefix##a_Name = 0; \
945	pVM->hm.s.vmx.HCPhys##a_Name = 0;
946
947	#define VMXLOCAL_INIT_VMCPU_MEMOBJ(a_Name, a_VirtPrefix) \
948	pVCpu->hm.s.vmx.hMemObj##a_Name = NIL_RTR0MEMOBJ; \
949	pVCpu->hm.s.vmx.a_VirtPrefix##a_Name = 0; \
950	pVCpu->hm.s.vmx.HCPhys##a_Name = 0;
951
952	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
953	VMXLOCAL_INIT_VM_MEMOBJ(Scratch, pv);
954	#endif
955	VMXLOCAL_INIT_VM_MEMOBJ(ApicAccess, pb);
956
957	AssertCompile(sizeof(VMCPUID) == sizeof(pVM->cCpus));
958	for (VMCPUID i = 0; i < pVM->cCpus; i++)
959	{
960	PVMCPU pVCpu = &pVM->aCpus[i];
961	VMXLOCAL_INIT_VMCPU_MEMOBJ(Vmcs, pv);
962	VMXLOCAL_INIT_VMCPU_MEMOBJ(MsrBitmap, pv);
963	VMXLOCAL_INIT_VMCPU_MEMOBJ(GuestMsr, pv);
964	VMXLOCAL_INIT_VMCPU_MEMOBJ(HostMsr, pv);
965	}
966	#undef VMXLOCAL_INIT_VMCPU_MEMOBJ
967	#undef VMXLOCAL_INIT_VM_MEMOBJ
968
969	/* The VMCS size cannot be more than 4096 bytes. See Intel spec. Appendix A.1 "Basic VMX Information". */
970	AssertReturnStmt(MSR_IA32_VMX_BASIC_INFO_VMCS_SIZE(pVM->hm.s.vmx.Msrs.u64BasicInfo) <= PAGE_SIZE,
971	(&pVM->aCpus[0])->hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE,
972	VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO);
973
974	/*
975	* Allocate all the VT-x structures.
976	*/
977	int rc = VINF_SUCCESS;
978	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
979	rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
980	if (RT_FAILURE(rc))
981	goto cleanup;
982	strcpy((char *)pVM->hm.s.vmx.pbScratch, "SCRATCH Magic");
983	(uint64_t )(pVM->hm.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef);
984	#endif
985
986	/* Allocate the APIC-access page for trapping APIC accesses from the guest. */
987	if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)
988	{
989	rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess,
990	&pVM->hm.s.vmx.HCPhysApicAccess);
991	if (RT_FAILURE(rc))
992	goto cleanup;
993	}
994
995	/*
996	* Initialize per-VCPU VT-x structures.
997	*/
998	for (VMCPUID i = 0; i < pVM->cCpus; i++)
999	{
1000	PVMCPU pVCpu = &pVM->aCpus[i];
1001	AssertPtr(pVCpu);
1002
1003	/* Allocate the VM control structure (VMCS). */
1004	rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjVmcs, &pVCpu->hm.s.vmx.pvVmcs, &pVCpu->hm.s.vmx.HCPhysVmcs);
1005	if (RT_FAILURE(rc))
1006	goto cleanup;
1007
1008	/* Get the allocated virtual-APIC page from the APIC device for transparent TPR accesses. */
1009	if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
1010	{
1011	rc = APICGetApicPageForCpu(pVCpu, &pVCpu->hm.s.vmx.HCPhysVirtApic, (PRTR0PTR)&pVCpu->hm.s.vmx.pbVirtApic,
1012	NULL /* pR3Ptr /, NULL / pRCPtr */);
1013	if (RT_FAILURE(rc))
1014	goto cleanup;
1015	}
1016
1017	/*
1018	* Allocate the MSR-bitmap if supported by the CPU. The MSR-bitmap is for
1019	* transparent accesses of specific MSRs.
1020	*
1021	* If the condition for enabling MSR bitmaps changes here, don't forget to
1022	* update HMAreMsrBitmapsAvailable().
1023	*/
1024	if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1025	{
1026	rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjMsrBitmap, &pVCpu->hm.s.vmx.pvMsrBitmap,
1027	&pVCpu->hm.s.vmx.HCPhysMsrBitmap);
1028	if (RT_FAILURE(rc))
1029	goto cleanup;
1030	ASMMemFill32(pVCpu->hm.s.vmx.pvMsrBitmap, PAGE_SIZE, UINT32_C(0xffffffff));
1031	}
1032
1033	/* Allocate the VM-entry MSR-load and VM-exit MSR-store page for the guest MSRs. */
1034	rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjGuestMsr, &pVCpu->hm.s.vmx.pvGuestMsr, &pVCpu->hm.s.vmx.HCPhysGuestMsr);
1035	if (RT_FAILURE(rc))
1036	goto cleanup;
1037
1038	/* Allocate the VM-exit MSR-load page for the host MSRs. */
1039	rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjHostMsr, &pVCpu->hm.s.vmx.pvHostMsr, &pVCpu->hm.s.vmx.HCPhysHostMsr);
1040	if (RT_FAILURE(rc))
1041	goto cleanup;
1042	}
1043
1044	return VINF_SUCCESS;
1045
1046	cleanup:
1047	hmR0VmxStructsFree(pVM);
1048	return rc;
1049	}
1050
1051
1052	/**
1053	* Does global VT-x initialization (called during module initialization).
1054	*
1055	* @returns VBox status code.
1056	*/
1057	VMMR0DECL(int) VMXR0GlobalInit(void)
1058	{
1059	#ifdef HMVMX_USE_FUNCTION_TABLE
1060	AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_apfnVMExitHandlers));
1061	# ifdef VBOX_STRICT
1062	for (unsigned i = 0; i < RT_ELEMENTS(g_apfnVMExitHandlers); i++)
1063	Assert(g_apfnVMExitHandlers[i]);
1064	# endif
1065	#endif
1066	return VINF_SUCCESS;
1067	}
1068
1069
1070	/**
1071	* Does global VT-x termination (called during module termination).
1072	*/
1073	VMMR0DECL(void) VMXR0GlobalTerm()
1074	{
1075	/* Nothing to do currently. */
1076	}
1077
1078
1079	/**
1080	* Sets up and activates VT-x on the current CPU.
1081	*
1082	* @returns VBox status code.
1083	* @param pCpu Pointer to the global CPU info struct.
1084	* @param pVM The cross context VM structure. Can be
1085	* NULL after a host resume operation.
1086	* @param pvCpuPage Pointer to the VMXON region (can be NULL if @a
1087	* fEnabledByHost is @c true).
1088	* @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if
1089	* @a fEnabledByHost is @c true).
1090	* @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to
1091	* enable VT-x on the host.
1092	* @param pvMsrs Opaque pointer to VMXMSRS struct.
1093	*/
1094	VMMR0DECL(int) VMXR0EnableCpu(PHMGLOBALCPUINFO pCpu, PVM pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
1095	void *pvMsrs)
1096	{
1097	Assert(pCpu);
1098	Assert(pvMsrs);
1099	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1100
1101	/* Enable VT-x if it's not already enabled by the host. */
1102	if (!fEnabledByHost)
1103	{
1104	int rc = hmR0VmxEnterRootMode(pVM, HCPhysCpuPage, pvCpuPage);
1105	if (RT_FAILURE(rc))
1106	return rc;
1107	}
1108
1109	/*
1110	* Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been using EPTPs) so
1111	* we don't retain any stale guest-physical mappings which won't get invalidated when flushing by VPID.
1112	*/
1113	PVMXMSRS pMsrs = (PVMXMSRS)pvMsrs;
1114	if (pMsrs->u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
1115	{
1116	hmR0VmxFlushEpt(NULL /* pVCpu */, VMXFLUSHEPT_ALL_CONTEXTS);
1117	pCpu->fFlushAsidBeforeUse = false;
1118	}
1119	else
1120	pCpu->fFlushAsidBeforeUse = true;
1121
1122	/* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */
1123	++pCpu->cTlbFlushes;
1124
1125	return VINF_SUCCESS;
1126	}
1127
1128
1129	/**
1130	* Deactivates VT-x on the current CPU.
1131	*
1132	* @returns VBox status code.
1133	* @param pCpu Pointer to the global CPU info struct.
1134	* @param pvCpuPage Pointer to the VMXON region.
1135	* @param HCPhysCpuPage Physical address of the VMXON region.
1136	*
1137	* @remarks This function should never be called when SUPR0EnableVTx() or
1138	* similar was used to enable VT-x on the host.
1139	*/
1140	VMMR0DECL(int) VMXR0DisableCpu(PHMGLOBALCPUINFO pCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
1141	{
1142	NOREF(pCpu);
1143	NOREF(pvCpuPage);
1144	NOREF(HCPhysCpuPage);
1145
1146	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1147	return hmR0VmxLeaveRootMode();
1148	}
1149
1150
1151	/**
1152	* Sets the permission bits for the specified MSR in the MSR bitmap.
1153	*
1154	* @param pVCpu The cross context virtual CPU structure.
1155	* @param uMsr The MSR value.
1156	* @param enmRead Whether reading this MSR causes a VM-exit.
1157	* @param enmWrite Whether writing this MSR causes a VM-exit.
1158	*/
1159	static void hmR0VmxSetMsrPermission(PVMCPU pVCpu, uint32_t uMsr, VMXMSREXITREAD enmRead, VMXMSREXITWRITE enmWrite)
1160	{
1161	int32_t iBit;
1162	uint8_t pbMsrBitmap = (uint8_t )pVCpu->hm.s.vmx.pvMsrBitmap;
1163
1164	/*
1165	* Layout:
1166	* 0x000 - 0x3ff - Low MSR read bits
1167	* 0x400 - 0x7ff - High MSR read bits
1168	* 0x800 - 0xbff - Low MSR write bits
1169	* 0xc00 - 0xfff - High MSR write bits
1170	*/
1171	if (uMsr <= 0x00001FFF)
1172	iBit = uMsr;
1173	else if (uMsr - UINT32_C(0xC0000000) <= UINT32_C(0x00001FFF))
1174	{
1175	iBit = uMsr - UINT32_C(0xC0000000);
1176	pbMsrBitmap += 0x400;
1177	}
1178	else
1179	AssertMsgFailedReturnVoid(("hmR0VmxSetMsrPermission: Invalid MSR %#RX32\n", uMsr));
1180
1181	Assert(iBit <= 0x1fff);
1182	if (enmRead == VMXMSREXIT_INTERCEPT_READ)
1183	ASMBitSet(pbMsrBitmap, iBit);
1184	else
1185	ASMBitClear(pbMsrBitmap, iBit);
1186
1187	if (enmWrite == VMXMSREXIT_INTERCEPT_WRITE)
1188	ASMBitSet(pbMsrBitmap + 0x800, iBit);
1189	else
1190	ASMBitClear(pbMsrBitmap + 0x800, iBit);
1191	}
1192
1193
1194	#ifdef VBOX_STRICT
1195	/**
1196	* Gets the permission bits for the specified MSR in the MSR bitmap.
1197	*
1198	* @returns VBox status code.
1199	* @retval VINF_SUCCESS if the specified MSR is found.
1200	* @retval VERR_NOT_FOUND if the specified MSR is not found.
1201	* @retval VERR_NOT_SUPPORTED if VT-x doesn't allow the MSR.
1202	*
1203	* @param pVCpu The cross context virtual CPU structure.
1204	* @param uMsr The MSR.
1205	* @param penmRead Where to store the read permissions.
1206	* @param penmWrite Where to store the write permissions.
1207	*/
1208	static int hmR0VmxGetMsrPermission(PVMCPU pVCpu, uint32_t uMsr, PVMXMSREXITREAD penmRead, PVMXMSREXITWRITE penmWrite)
1209	{
1210	AssertPtrReturn(penmRead, VERR_INVALID_PARAMETER);
1211	AssertPtrReturn(penmWrite, VERR_INVALID_PARAMETER);
1212	int32_t iBit;
1213	uint8_t pbMsrBitmap = (uint8_t )pVCpu->hm.s.vmx.pvMsrBitmap;
1214
1215	/* See hmR0VmxSetMsrPermission() for the layout. */
1216	if (uMsr <= 0x00001FFF)
1217	iBit = uMsr;
1218	else if ( uMsr >= 0xC0000000
1219	&& uMsr <= 0xC0001FFF)
1220	{
1221	iBit = (uMsr - 0xC0000000);
1222	pbMsrBitmap += 0x400;
1223	}
1224	else
1225	AssertMsgFailedReturn(("hmR0VmxGetMsrPermission: Invalid MSR %#RX32\n", uMsr), VERR_NOT_SUPPORTED);
1226
1227	Assert(iBit <= 0x1fff);
1228	if (ASMBitTest(pbMsrBitmap, iBit))
1229	*penmRead = VMXMSREXIT_INTERCEPT_READ;
1230	else
1231	*penmRead = VMXMSREXIT_PASSTHRU_READ;
1232
1233	if (ASMBitTest(pbMsrBitmap + 0x800, iBit))
1234	*penmWrite = VMXMSREXIT_INTERCEPT_WRITE;
1235	else
1236	*penmWrite = VMXMSREXIT_PASSTHRU_WRITE;
1237	return VINF_SUCCESS;
1238	}
1239	#endif /* VBOX_STRICT */
1240
1241
1242	/**
1243	* Updates the VMCS with the number of effective MSRs in the auto-load/store MSR
1244	* area.
1245	*
1246	* @returns VBox status code.
1247	* @param pVCpu The cross context virtual CPU structure.
1248	* @param cMsrs The number of MSRs.
1249	*/
1250	DECLINLINE(int) hmR0VmxSetAutoLoadStoreMsrCount(PVMCPU pVCpu, uint32_t cMsrs)
1251	{
1252	/* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */
1253	uint32_t const cMaxSupportedMsrs = MSR_IA32_VMX_MISC_MAX_MSR(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc);
1254	if (RT_UNLIKELY(cMsrs > cMaxSupportedMsrs))
1255	{
1256	LogRel(("CPU auto-load/store MSR count in VMCS exceeded cMsrs=%u Supported=%u.\n", cMsrs, cMaxSupportedMsrs));
1257	pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE;
1258	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
1259	}
1260
1261	/* Update number of guest MSRs to load/store across the world-switch. */
1262	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs);
1263	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs);
1264
1265	/* Update number of host MSRs to load after the world-switch. Identical to guest-MSR count as it's always paired. */
1266	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs);
1267	AssertRCReturn(rc, rc);
1268
1269	/* Update the VCPU's copy of the MSR count. */
1270	pVCpu->hm.s.vmx.cMsrs = cMsrs;
1271
1272	return VINF_SUCCESS;
1273	}
1274
1275
1276	/**
1277	* Adds a new (or updates the value of an existing) guest/host MSR
1278	* pair to be swapped during the world-switch as part of the
1279	* auto-load/store MSR area in the VMCS.
1280	*
1281	* @returns VBox status code.
1282	* @param pVCpu The cross context virtual CPU structure.
1283	* @param uMsr The MSR.
1284	* @param uGuestMsrValue Value of the guest MSR.
1285	* @param fUpdateHostMsr Whether to update the value of the host MSR if
1286	* necessary.
1287	* @param pfAddedAndUpdated Where to store whether the MSR was added -and-
1288	* its value was updated. Optional, can be NULL.
1289	*/
1290	static int hmR0VmxAddAutoLoadStoreMsr(PVMCPU pVCpu, uint32_t uMsr, uint64_t uGuestMsrValue, bool fUpdateHostMsr,
1291	bool *pfAddedAndUpdated)
1292	{
1293	PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1294	uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1295	uint32_t i;
1296	for (i = 0; i < cMsrs; i++)
1297	{
1298	if (pGuestMsr->u32Msr == uMsr)
1299	break;
1300	pGuestMsr++;
1301	}
1302
1303	bool fAdded = false;
1304	if (i == cMsrs)
1305	{
1306	++cMsrs;
1307	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, cMsrs);
1308	AssertMsgRCReturn(rc, ("hmR0VmxAddAutoLoadStoreMsr: Insufficient space to add MSR %u\n", uMsr), rc);
1309
1310	/* Now that we're swapping MSRs during the world-switch, allow the guest to read/write them without causing VM-exits. */
1311	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1312	hmR0VmxSetMsrPermission(pVCpu, uMsr, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
1313
1314	fAdded = true;
1315	}
1316
1317	/* Update the MSR values in the auto-load/store MSR area. */
1318	pGuestMsr->u32Msr = uMsr;
1319	pGuestMsr->u64Value = uGuestMsrValue;
1320
1321	/* Create/update the MSR slot in the host MSR area. */
1322	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1323	pHostMsr += i;
1324	pHostMsr->u32Msr = uMsr;
1325
1326	/*
1327	* Update the host MSR only when requested by the caller AND when we're
1328	* adding it to the auto-load/store area. Otherwise, it would have been
1329	* updated by hmR0VmxSaveHostMsrs(). We do this for performance reasons.
1330	*/
1331	bool fUpdatedMsrValue = false;
1332	if ( fAdded
1333	&& fUpdateHostMsr)
1334	{
1335	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1336	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1337	pHostMsr->u64Value = ASMRdMsr(pHostMsr->u32Msr);
1338	fUpdatedMsrValue = true;
1339	}
1340
1341	if (pfAddedAndUpdated)
1342	*pfAddedAndUpdated = fUpdatedMsrValue;
1343	return VINF_SUCCESS;
1344	}
1345
1346
1347	/**
1348	* Removes a guest/host MSR pair to be swapped during the world-switch from the
1349	* auto-load/store MSR area in the VMCS.
1350	*
1351	* @returns VBox status code.
1352	* @param pVCpu The cross context virtual CPU structure.
1353	* @param uMsr The MSR.
1354	*/
1355	static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPU pVCpu, uint32_t uMsr)
1356	{
1357	PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1358	uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1359	for (uint32_t i = 0; i < cMsrs; i++)
1360	{
1361	/* Find the MSR. */
1362	if (pGuestMsr->u32Msr == uMsr)
1363	{
1364	/* If it's the last MSR, simply reduce the count. */
1365	if (i == cMsrs - 1)
1366	{
1367	--cMsrs;
1368	break;
1369	}
1370
1371	/* Remove it by swapping the last MSR in place of it, and reducing the count. */
1372	PVMXAUTOMSR pLastGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1373	pLastGuestMsr += cMsrs - 1;
1374	pGuestMsr->u32Msr = pLastGuestMsr->u32Msr;
1375	pGuestMsr->u64Value = pLastGuestMsr->u64Value;
1376
1377	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1378	PVMXAUTOMSR pLastHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1379	pLastHostMsr += cMsrs - 1;
1380	pHostMsr->u32Msr = pLastHostMsr->u32Msr;
1381	pHostMsr->u64Value = pLastHostMsr->u64Value;
1382	--cMsrs;
1383	break;
1384	}
1385	pGuestMsr++;
1386	}
1387
1388	/* Update the VMCS if the count changed (meaning the MSR was found). */
1389	if (cMsrs != pVCpu->hm.s.vmx.cMsrs)
1390	{
1391	int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, cMsrs);
1392	AssertRCReturn(rc, rc);
1393
1394	/* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */
1395	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1396	hmR0VmxSetMsrPermission(pVCpu, uMsr, VMXMSREXIT_INTERCEPT_READ, VMXMSREXIT_INTERCEPT_WRITE);
1397
1398	Log4(("Removed MSR %#RX32 new cMsrs=%u\n", uMsr, pVCpu->hm.s.vmx.cMsrs));
1399	return VINF_SUCCESS;
1400	}
1401
1402	return VERR_NOT_FOUND;
1403	}
1404
1405
1406	/**
1407	* Checks if the specified guest MSR is part of the auto-load/store area in
1408	* the VMCS.
1409	*
1410	* @returns true if found, false otherwise.
1411	* @param pVCpu The cross context virtual CPU structure.
1412	* @param uMsr The MSR to find.
1413	*/
1414	static bool hmR0VmxIsAutoLoadStoreGuestMsr(PVMCPU pVCpu, uint32_t uMsr)
1415	{
1416	PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1417	uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1418
1419	for (uint32_t i = 0; i < cMsrs; i++, pGuestMsr++)
1420	{
1421	if (pGuestMsr->u32Msr == uMsr)
1422	return true;
1423	}
1424	return false;
1425	}
1426
1427
1428	/**
1429	* Updates the value of all host MSRs in the auto-load/store area in the VMCS.
1430	*
1431	* @param pVCpu The cross context virtual CPU structure.
1432	*
1433	* @remarks No-long-jump zone!!!
1434	*/
1435	static void hmR0VmxUpdateAutoLoadStoreHostMsrs(PVMCPU pVCpu)
1436	{
1437	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1438	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1439	PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1440	uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1441
1442	for (uint32_t i = 0; i < cMsrs; i++, pHostMsr++, pGuestMsr++)
1443	{
1444	AssertReturnVoid(pHostMsr->u32Msr == pGuestMsr->u32Msr);
1445
1446	/*
1447	* Performance hack for the host EFER MSR. We use the cached value rather than re-read it.
1448	* Strict builds will catch mismatches in hmR0VmxCheckAutoLoadStoreMsrs(). See @bugref{7368}.
1449	*/
1450	if (pHostMsr->u32Msr == MSR_K6_EFER)
1451	pHostMsr->u64Value = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostEfer;
1452	else
1453	pHostMsr->u64Value = ASMRdMsr(pHostMsr->u32Msr);
1454	}
1455
1456	pVCpu->hm.s.vmx.fUpdatedHostMsrs = true;
1457	}
1458
1459
1460	/**
1461	* Saves a set of host MSRs to allow read/write passthru access to the guest and
1462	* perform lazy restoration of the host MSRs while leaving VT-x.
1463	*
1464	* @param pVCpu The cross context virtual CPU structure.
1465	*
1466	* @remarks No-long-jump zone!!!
1467	*/
1468	static void hmR0VmxLazySaveHostMsrs(PVMCPU pVCpu)
1469	{
1470	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1471
1472	/*
1473	* Note: If you're adding MSRs here, make sure to update the MSR-bitmap permissions in hmR0VmxSetupProcCtls().
1474	*/
1475	if (!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST))
1476	{
1477	Assert(!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)); /* Guest MSRs better not be loaded now. */
1478	#if HC_ARCH_BITS == 64
1479	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
1480	{
1481	pVCpu->hm.s.vmx.u64HostLStarMsr = ASMRdMsr(MSR_K8_LSTAR);
1482	pVCpu->hm.s.vmx.u64HostStarMsr = ASMRdMsr(MSR_K6_STAR);
1483	pVCpu->hm.s.vmx.u64HostSFMaskMsr = ASMRdMsr(MSR_K8_SF_MASK);
1484	pVCpu->hm.s.vmx.u64HostKernelGSBaseMsr = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
1485	}
1486	#endif
1487	pVCpu->hm.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_SAVED_HOST;
1488	}
1489	}
1490
1491
1492	/**
1493	* Checks whether the MSR belongs to the set of guest MSRs that we restore
1494	* lazily while leaving VT-x.
1495	*
1496	* @returns true if it does, false otherwise.
1497	* @param pVCpu The cross context virtual CPU structure.
1498	* @param uMsr The MSR to check.
1499	*/
1500	static bool hmR0VmxIsLazyGuestMsr(PVMCPU pVCpu, uint32_t uMsr)
1501	{
1502	NOREF(pVCpu);
1503	#if HC_ARCH_BITS == 64
1504	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
1505	{
1506	switch (uMsr)
1507	{
1508	case MSR_K8_LSTAR:
1509	case MSR_K6_STAR:
1510	case MSR_K8_SF_MASK:
1511	case MSR_K8_KERNEL_GS_BASE:
1512	return true;
1513	}
1514	}
1515	#else
1516	RT_NOREF(pVCpu, uMsr);
1517	#endif
1518	return false;
1519	}
1520
1521
1522	/**
1523	* Saves a set of guest MSRs back into the guest-CPU context.
1524	*
1525	* @param pVCpu The cross context virtual CPU structure.
1526	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
1527	* out-of-sync. Make sure to update the required fields
1528	* before using them.
1529	*
1530	* @remarks No-long-jump zone!!!
1531	*/
1532	static void hmR0VmxLazySaveGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
1533	{
1534	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1535	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1536
1537	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
1538	{
1539	Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1540	#if HC_ARCH_BITS == 64
1541	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
1542	{
1543	pMixedCtx->msrLSTAR = ASMRdMsr(MSR_K8_LSTAR);
1544	pMixedCtx->msrSTAR = ASMRdMsr(MSR_K6_STAR);
1545	pMixedCtx->msrSFMASK = ASMRdMsr(MSR_K8_SF_MASK);
1546	pMixedCtx->msrKERNELGSBASE = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
1547	}
1548	#else
1549	NOREF(pMixedCtx);
1550	#endif
1551	}
1552	}
1553
1554
1555	/**
1556	* Loads a set of guests MSRs to allow read/passthru to the guest.
1557	*
1558	* The name of this function is slightly confusing. This function does NOT
1559	* postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a
1560	* common prefix for functions dealing with "lazy restoration" of the shared
1561	* MSRs.
1562	*
1563	* @param pVCpu The cross context virtual CPU structure.
1564	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
1565	* out-of-sync. Make sure to update the required fields
1566	* before using them.
1567	*
1568	* @remarks No-long-jump zone!!!
1569	*/
1570	static void hmR0VmxLazyLoadGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
1571	{
1572	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1573	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1574
1575	Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1576	#if HC_ARCH_BITS == 64
1577	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
1578	{
1579	/*
1580	* If the guest MSRs are not loaded -and- if all the guest MSRs are identical
1581	* to the MSRs on the CPU (which are the saved host MSRs, see assertion above) then
1582	* we can skip a few MSR writes.
1583	*
1584	* Otherwise, it implies either 1. they're not loaded, or 2. they're loaded but the
1585	* guest MSR values in the guest-CPU context might be different to what's currently
1586	* loaded in the CPU. In either case, we need to write the new guest MSR values to the
1587	* CPU, see @bugref{8728}.
1588	*/
1589	if ( !(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
1590	&& pMixedCtx->msrKERNELGSBASE == pVCpu->hm.s.vmx.u64HostKernelGSBaseMsr
1591	&& pMixedCtx->msrLSTAR == pVCpu->hm.s.vmx.u64HostLStarMsr
1592	&& pMixedCtx->msrSTAR == pVCpu->hm.s.vmx.u64HostStarMsr
1593	&& pMixedCtx->msrSFMASK == pVCpu->hm.s.vmx.u64HostSFMaskMsr)
1594	{
1595	#ifdef VBOX_STRICT
1596	Assert(ASMRdMsr(MSR_K8_KERNEL_GS_BASE) == pMixedCtx->msrKERNELGSBASE);
1597	Assert(ASMRdMsr(MSR_K8_LSTAR) == pMixedCtx->msrLSTAR);
1598	Assert(ASMRdMsr(MSR_K6_STAR) == pMixedCtx->msrSTAR);
1599	Assert(ASMRdMsr(MSR_K8_SF_MASK) == pMixedCtx->msrSFMASK);
1600	#endif
1601	}
1602	else
1603	{
1604	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pMixedCtx->msrKERNELGSBASE);
1605	ASMWrMsr(MSR_K8_LSTAR, pMixedCtx->msrLSTAR);
1606	ASMWrMsr(MSR_K6_STAR, pMixedCtx->msrSTAR);
1607	ASMWrMsr(MSR_K8_SF_MASK, pMixedCtx->msrSFMASK);
1608	}
1609	}
1610	#else
1611	RT_NOREF(pMixedCtx);
1612	#endif
1613	pVCpu->hm.s.vmx.fLazyMsrs \|= VMX_LAZY_MSRS_LOADED_GUEST;
1614	}
1615
1616
1617	/**
1618	* Performs lazy restoration of the set of host MSRs if they were previously
1619	* loaded with guest MSR values.
1620	*
1621	* @param pVCpu The cross context virtual CPU structure.
1622	*
1623	* @remarks No-long-jump zone!!!
1624	* @remarks The guest MSRs should have been saved back into the guest-CPU
1625	* context by hmR0VmxSaveGuestLazyMsrs()!!!
1626	*/
1627	static void hmR0VmxLazyRestoreHostMsrs(PVMCPU pVCpu)
1628	{
1629	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1630	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1631
1632	if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
1633	{
1634	Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1635	#if HC_ARCH_BITS == 64
1636	if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
1637	{
1638	ASMWrMsr(MSR_K8_LSTAR, pVCpu->hm.s.vmx.u64HostLStarMsr);
1639	ASMWrMsr(MSR_K6_STAR, pVCpu->hm.s.vmx.u64HostStarMsr);
1640	ASMWrMsr(MSR_K8_SF_MASK, pVCpu->hm.s.vmx.u64HostSFMaskMsr);
1641	ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pVCpu->hm.s.vmx.u64HostKernelGSBaseMsr);
1642	}
1643	#endif
1644	}
1645	pVCpu->hm.s.vmx.fLazyMsrs &= ~(VMX_LAZY_MSRS_LOADED_GUEST \| VMX_LAZY_MSRS_SAVED_HOST);
1646	}
1647
1648
1649	/**
1650	* Verifies that our cached values of the VMCS controls are all
1651	* consistent with what's actually present in the VMCS.
1652	*
1653	* @returns VBox status code.
1654	* @param pVCpu The cross context virtual CPU structure.
1655	*/
1656	static int hmR0VmxCheckVmcsCtls(PVMCPU pVCpu)
1657	{
1658	uint32_t u32Val;
1659	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
1660	AssertRCReturn(rc, rc);
1661	AssertMsgReturn(pVCpu->hm.s.vmx.u32EntryCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32EntryCtls, u32Val),
1662	VERR_VMX_ENTRY_CTLS_CACHE_INVALID);
1663
1664	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val);
1665	AssertRCReturn(rc, rc);
1666	AssertMsgReturn(pVCpu->hm.s.vmx.u32ExitCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ExitCtls, u32Val),
1667	VERR_VMX_EXIT_CTLS_CACHE_INVALID);
1668
1669	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val);
1670	AssertRCReturn(rc, rc);
1671	AssertMsgReturn(pVCpu->hm.s.vmx.u32PinCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32PinCtls, u32Val),
1672	VERR_VMX_PIN_EXEC_CTLS_CACHE_INVALID);
1673
1674	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val);
1675	AssertRCReturn(rc, rc);
1676	AssertMsgReturn(pVCpu->hm.s.vmx.u32ProcCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ProcCtls, u32Val),
1677	VERR_VMX_PROC_EXEC_CTLS_CACHE_INVALID);
1678
1679	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
1680	{
1681	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val);
1682	AssertRCReturn(rc, rc);
1683	AssertMsgReturn(pVCpu->hm.s.vmx.u32ProcCtls2 == u32Val,
1684	("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ProcCtls2, u32Val),
1685	VERR_VMX_PROC_EXEC2_CTLS_CACHE_INVALID);
1686	}
1687
1688	return VINF_SUCCESS;
1689	}
1690
1691
1692	#ifdef VBOX_STRICT
1693	/**
1694	* Verifies that our cached host EFER value has not changed
1695	* since we cached it.
1696	*
1697	* @param pVCpu The cross context virtual CPU structure.
1698	*/
1699	static void hmR0VmxCheckHostEferMsr(PVMCPU pVCpu)
1700	{
1701	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1702
1703	if (pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR)
1704	{
1705	uint64_t u64Val;
1706	int rc = VMXReadVmcs64(VMX_VMCS64_HOST_EFER_FULL, &u64Val);
1707	AssertRC(rc);
1708
1709	uint64_t u64HostEferMsr = ASMRdMsr(MSR_K6_EFER);
1710	AssertMsgReturnVoid(u64HostEferMsr == u64Val, ("u64HostEferMsr=%#RX64 u64Val=%#RX64\n", u64HostEferMsr, u64Val));
1711	}
1712	}
1713
1714
1715	/**
1716	* Verifies whether the guest/host MSR pairs in the auto-load/store area in the
1717	* VMCS are correct.
1718	*
1719	* @param pVCpu The cross context virtual CPU structure.
1720	*/
1721	static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPU pVCpu)
1722	{
1723	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1724
1725	/* Verify MSR counts in the VMCS are what we think it should be. */
1726	uint32_t cMsrs;
1727	int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cMsrs); AssertRC(rc);
1728	Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1729
1730	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cMsrs); AssertRC(rc);
1731	Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1732
1733	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cMsrs); AssertRC(rc);
1734	Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1735
1736	PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1737	PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1738	for (uint32_t i = 0; i < cMsrs; i++, pHostMsr++, pGuestMsr++)
1739	{
1740	/* Verify that the MSRs are paired properly and that the host MSR has the correct value. */
1741	AssertMsgReturnVoid(pHostMsr->u32Msr == pGuestMsr->u32Msr, ("HostMsr=%#RX32 GuestMsr=%#RX32 cMsrs=%u\n", pHostMsr->u32Msr,
1742	pGuestMsr->u32Msr, cMsrs));
1743
1744	uint64_t u64Msr = ASMRdMsr(pHostMsr->u32Msr);
1745	AssertMsgReturnVoid(pHostMsr->u64Value == u64Msr, ("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
1746	pHostMsr->u32Msr, pHostMsr->u64Value, u64Msr, cMsrs));
1747
1748	/* Verify that the permissions are as expected in the MSR bitmap. */
1749	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1750	{
1751	VMXMSREXITREAD enmRead;
1752	VMXMSREXITWRITE enmWrite;
1753	rc = hmR0VmxGetMsrPermission(pVCpu, pGuestMsr->u32Msr, &enmRead, &enmWrite);
1754	AssertMsgReturnVoid(rc == VINF_SUCCESS, ("hmR0VmxGetMsrPermission! failed. rc=%Rrc\n", rc));
1755	if (pGuestMsr->u32Msr == MSR_K6_EFER)
1756	{
1757	AssertMsgReturnVoid(enmRead == VMXMSREXIT_INTERCEPT_READ, ("Passthru read for EFER!?\n"));
1758	AssertMsgReturnVoid(enmWrite == VMXMSREXIT_INTERCEPT_WRITE, ("Passthru write for EFER!?\n"));
1759	}
1760	else
1761	{
1762	AssertMsgReturnVoid(enmRead == VMXMSREXIT_PASSTHRU_READ, ("u32Msr=%#RX32 cMsrs=%u No passthru read!\n",
1763	pGuestMsr->u32Msr, cMsrs));
1764	AssertMsgReturnVoid(enmWrite == VMXMSREXIT_PASSTHRU_WRITE, ("u32Msr=%#RX32 cMsrs=%u No passthru write!\n",
1765	pGuestMsr->u32Msr, cMsrs));
1766	}
1767	}
1768	}
1769	}
1770	#endif /* VBOX_STRICT */
1771
1772
1773	/**
1774	* Flushes the TLB using EPT.
1775	*
1776	* @returns VBox status code.
1777	* @param pVCpu The cross context virtual CPU structure of the calling
1778	* EMT. Can be NULL depending on @a enmFlush.
1779	* @param enmFlush Type of flush.
1780	*
1781	* @remarks Caller is responsible for making sure this function is called only
1782	* when NestedPaging is supported and providing @a enmFlush that is
1783	* supported by the CPU.
1784	* @remarks Can be called with interrupts disabled.
1785	*/
1786	static void hmR0VmxFlushEpt(PVMCPU pVCpu, VMXFLUSHEPT enmFlush)
1787	{
1788	uint64_t au64Descriptor[2];
1789	if (enmFlush == VMXFLUSHEPT_ALL_CONTEXTS)
1790	au64Descriptor[0] = 0;
1791	else
1792	{
1793	Assert(pVCpu);
1794	au64Descriptor[0] = pVCpu->hm.s.vmx.HCPhysEPTP;
1795	}
1796	au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */
1797
1798	int rc = VMXR0InvEPT(enmFlush, &au64Descriptor[0]);
1799	AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %RGv failed with %Rrc\n", enmFlush, pVCpu ? pVCpu->hm.s.vmx.HCPhysEPTP : 0,
1800	rc));
1801	if ( RT_SUCCESS(rc)
1802	&& pVCpu)
1803	{
1804	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging);
1805	}
1806	}
1807
1808
1809	/**
1810	* Flushes the TLB using VPID.
1811	*
1812	* @returns VBox status code.
1813	* @param pVM The cross context VM structure.
1814	* @param pVCpu The cross context virtual CPU structure of the calling
1815	* EMT. Can be NULL depending on @a enmFlush.
1816	* @param enmFlush Type of flush.
1817	* @param GCPtr Virtual address of the page to flush (can be 0 depending
1818	* on @a enmFlush).
1819	*
1820	* @remarks Can be called with interrupts disabled.
1821	*/
1822	static void hmR0VmxFlushVpid(PVM pVM, PVMCPU pVCpu, VMXFLUSHVPID enmFlush, RTGCPTR GCPtr)
1823	{
1824	NOREF(pVM);
1825	AssertPtr(pVM);
1826	Assert(pVM->hm.s.vmx.fVpid);
1827
1828	uint64_t au64Descriptor[2];
1829	if (enmFlush == VMXFLUSHVPID_ALL_CONTEXTS)
1830	{
1831	au64Descriptor[0] = 0;
1832	au64Descriptor[1] = 0;
1833	}
1834	else
1835	{
1836	AssertPtr(pVCpu);
1837	AssertMsg(pVCpu->hm.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
1838	AssertMsg(pVCpu->hm.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
1839	au64Descriptor[0] = pVCpu->hm.s.uCurrentAsid;
1840	au64Descriptor[1] = GCPtr;
1841	}
1842
1843	int rc = VMXR0InvVPID(enmFlush, &au64Descriptor[0]); NOREF(rc);
1844	AssertMsg(rc == VINF_SUCCESS,
1845	("VMXR0InvVPID %#x %u %RGv failed with %d\n", enmFlush, pVCpu ? pVCpu->hm.s.uCurrentAsid : 0, GCPtr, rc));
1846	if ( RT_SUCCESS(rc)
1847	&& pVCpu)
1848	{
1849	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
1850	}
1851	}
1852
1853
1854	/**
1855	* Invalidates a guest page by guest virtual address. Only relevant for
1856	* EPT/VPID, otherwise there is nothing really to invalidate.
1857	*
1858	* @returns VBox status code.
1859	* @param pVM The cross context VM structure.
1860	* @param pVCpu The cross context virtual CPU structure.
1861	* @param GCVirt Guest virtual address of the page to invalidate.
1862	*/
1863	VMMR0DECL(int) VMXR0InvalidatePage(PVM pVM, PVMCPU pVCpu, RTGCPTR GCVirt)
1864	{
1865	AssertPtr(pVM);
1866	AssertPtr(pVCpu);
1867	LogFlowFunc(("pVM=%p pVCpu=%p GCVirt=%RGv\n", pVM, pVCpu, GCVirt));
1868
1869	bool fFlushPending = VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_FLUSH);
1870	if (!fFlushPending)
1871	{
1872	/*
1873	* We must invalidate the guest TLB entry in either case, we cannot ignore it even for the EPT case
1874	* See @bugref{6043} and @bugref{6177}.
1875	*
1876	* Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*() as this
1877	* function maybe called in a loop with individual addresses.
1878	*/
1879	if (pVM->hm.s.vmx.fVpid)
1880	{
1881	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
1882	{
1883	hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_INDIV_ADDR, GCVirt);
1884	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
1885	}
1886	else
1887	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1888	}
1889	else if (pVM->hm.s.fNestedPaging)
1890	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1891	}
1892
1893	return VINF_SUCCESS;
1894	}
1895
1896
1897	/**
1898	* Invalidates a guest page by physical address. Only relevant for EPT/VPID,
1899	* otherwise there is nothing really to invalidate.
1900	*
1901	* @returns VBox status code.
1902	* @param pVM The cross context VM structure.
1903	* @param pVCpu The cross context virtual CPU structure.
1904	* @param GCPhys Guest physical address of the page to invalidate.
1905	*/
1906	VMMR0DECL(int) VMXR0InvalidatePhysPage(PVM pVM, PVMCPU pVCpu, RTGCPHYS GCPhys)
1907	{
1908	NOREF(pVM); NOREF(GCPhys);
1909	LogFlowFunc(("%RGp\n", GCPhys));
1910
1911	/*
1912	* We cannot flush a page by guest-physical address. invvpid takes only a linear address while invept only flushes
1913	* by EPT not individual addresses. We update the force flag here and flush before the next VM-entry in hmR0VmxFlushTLB*().
1914	* This function might be called in a loop. This should cause a flush-by-EPT if EPT is in use. See @bugref{6568}.
1915	*/
1916	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1917	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgPhys);
1918	return VINF_SUCCESS;
1919	}
1920
1921
1922	/**
1923	* Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the
1924	* case where neither EPT nor VPID is supported by the CPU.
1925	*
1926	* @param pVM The cross context VM structure.
1927	* @param pVCpu The cross context virtual CPU structure.
1928	* @param pCpu Pointer to the global HM struct.
1929	*
1930	* @remarks Called with interrupts disabled.
1931	*/
1932	static void hmR0VmxFlushTaggedTlbNone(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
1933	{
1934	AssertPtr(pVCpu);
1935	AssertPtr(pCpu);
1936	NOREF(pVM);
1937
1938	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
1939
1940	Assert(pCpu->idCpu != NIL_RTCPUID);
1941	pVCpu->hm.s.idLastCpu = pCpu->idCpu;
1942	pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
1943	pVCpu->hm.s.fForceTLBFlush = false;
1944	return;
1945	}
1946
1947
1948	/**
1949	* Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary.
1950	*
1951	* @param pVM The cross context VM structure.
1952	* @param pVCpu The cross context virtual CPU structure.
1953	* @param pCpu Pointer to the global HM CPU struct.
1954	* @remarks All references to "ASID" in this function pertains to "VPID" in
1955	* Intel's nomenclature. The reason is, to avoid confusion in compare
1956	* statements since the host-CPU copies are named "ASID".
1957	*
1958	* @remarks Called with interrupts disabled.
1959	*/
1960	static void hmR0VmxFlushTaggedTlbBoth(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
1961	{
1962	#ifdef VBOX_WITH_STATISTICS
1963	bool fTlbFlushed = false;
1964	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0)
1965	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \
1966	if (!fTlbFlushed) \
1967	STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \
1968	} while (0)
1969	#else
1970	# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0)
1971	# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0)
1972	#endif
1973
1974	AssertPtr(pVM);
1975	AssertPtr(pCpu);
1976	AssertPtr(pVCpu);
1977	Assert(pCpu->idCpu != NIL_RTCPUID);
1978
1979	AssertMsg(pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid,
1980	("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled."
1981	"fNestedPaging=%RTbool fVpid=%RTbool", pVM->hm.s.fNestedPaging, pVM->hm.s.vmx.fVpid));
1982
1983	/*
1984	* Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
1985	* If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB
1986	* or the host CPU is online after a suspend/resume, so we cannot reuse the current ASID anymore.
1987	*/
1988	if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
1989	\|\| pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
1990	{
1991	++pCpu->uCurrentAsid;
1992	if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
1993	{
1994	pCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */
1995	pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
1996	pCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
1997	}
1998
1999	pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid;
2000	pVCpu->hm.s.idLastCpu = pCpu->idCpu;
2001	pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
2002
2003	/*
2004	* Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also
2005	* invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}.
2006	*/
2007	hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
2008	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
2009	HMVMX_SET_TAGGED_TLB_FLUSHED();
2010	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH); /* Already flushed-by-EPT, skip doing it again below. */
2011	}
2012
2013	/* Check for explicit TLB flushes. */
2014	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
2015	{
2016	/*
2017	* Changes to the EPT paging structure by VMM requires flushing by EPT as the CPU creates
2018	* guest-physical (only EPT-tagged) mappings while traversing the EPT tables when EPT is in use.
2019	* Flushing by VPID will only flush linear (only VPID-tagged) and combined (EPT+VPID tagged) mappings
2020	* but not guest-physical mappings.
2021	* See Intel spec. 28.3.2 "Creating and Using Cached Translation Information". See @bugref{6568}.
2022	*/
2023	hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
2024	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2025	HMVMX_SET_TAGGED_TLB_FLUSHED();
2026	}
2027
2028	pVCpu->hm.s.fForceTLBFlush = false;
2029	HMVMX_UPDATE_FLUSH_SKIPPED_STAT();
2030
2031	Assert(pVCpu->hm.s.idLastCpu == pCpu->idCpu);
2032	Assert(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes);
2033	AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes,
2034	("Flush count mismatch for cpu %d (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
2035	AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
2036	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pCpu->idCpu,
2037	pCpu->uCurrentAsid, pCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
2038	AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
2039	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
2040
2041	/* Update VMCS with the VPID. */
2042	int rc = VMXWriteVmcs32(VMX_VMCS16_VPID, pVCpu->hm.s.uCurrentAsid);
2043	AssertRC(rc);
2044
2045	#undef HMVMX_SET_TAGGED_TLB_FLUSHED
2046	}
2047
2048
2049	/**
2050	* Flushes the tagged-TLB entries for EPT CPUs as necessary.
2051	*
2052	* @returns VBox status code.
2053	* @param pVM The cross context VM structure.
2054	* @param pVCpu The cross context virtual CPU structure.
2055	* @param pCpu Pointer to the global HM CPU struct.
2056	*
2057	* @remarks Called with interrupts disabled.
2058	*/
2059	static void hmR0VmxFlushTaggedTlbEpt(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2060	{
2061	AssertPtr(pVM);
2062	AssertPtr(pVCpu);
2063	AssertPtr(pCpu);
2064	Assert(pCpu->idCpu != NIL_RTCPUID);
2065	AssertMsg(pVM->hm.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with NestedPaging disabled."));
2066	AssertMsg(!pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID enabled."));
2067
2068	/*
2069	* Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
2070	* A change in the TLB flush count implies the host CPU is online after a suspend/resume.
2071	*/
2072	if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
2073	\|\| pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
2074	{
2075	pVCpu->hm.s.fForceTLBFlush = true;
2076	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
2077	}
2078
2079	/* Check for explicit TLB flushes. */
2080	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
2081	{
2082	pVCpu->hm.s.fForceTLBFlush = true;
2083	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2084	}
2085
2086	pVCpu->hm.s.idLastCpu = pCpu->idCpu;
2087	pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
2088
2089	if (pVCpu->hm.s.fForceTLBFlush)
2090	{
2091	hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
2092	pVCpu->hm.s.fForceTLBFlush = false;
2093	}
2094	}
2095
2096
2097	/**
2098	* Flushes the tagged-TLB entries for VPID CPUs as necessary.
2099	*
2100	* @returns VBox status code.
2101	* @param pVM The cross context VM structure.
2102	* @param pVCpu The cross context virtual CPU structure.
2103	* @param pCpu Pointer to the global HM CPU struct.
2104	*
2105	* @remarks Called with interrupts disabled.
2106	*/
2107	static void hmR0VmxFlushTaggedTlbVpid(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2108	{
2109	AssertPtr(pVM);
2110	AssertPtr(pVCpu);
2111	AssertPtr(pCpu);
2112	Assert(pCpu->idCpu != NIL_RTCPUID);
2113	AssertMsg(pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked with VPID disabled."));
2114	AssertMsg(!pVM->hm.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging enabled"));
2115
2116	/*
2117	* Force a TLB flush for the first world switch if the current CPU differs from the one we ran on last.
2118	* If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB
2119	* or the host CPU is online after a suspend/resume, so we cannot reuse the current ASID anymore.
2120	*/
2121	if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
2122	\|\| pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
2123	{
2124	pVCpu->hm.s.fForceTLBFlush = true;
2125	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
2126	}
2127
2128	/* Check for explicit TLB flushes. */
2129	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
2130	{
2131	/*
2132	* If we ever support VPID flush combinations other than ALL or SINGLE-context (see hmR0VmxSetupTaggedTlb())
2133	* we would need to explicitly flush in this case (add an fExplicitFlush = true here and change the
2134	* pCpu->fFlushAsidBeforeUse check below to include fExplicitFlush's too) - an obscure corner case.
2135	*/
2136	pVCpu->hm.s.fForceTLBFlush = true;
2137	STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2138	}
2139
2140	pVCpu->hm.s.idLastCpu = pCpu->idCpu;
2141	if (pVCpu->hm.s.fForceTLBFlush)
2142	{
2143	++pCpu->uCurrentAsid;
2144	if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
2145	{
2146	pCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */
2147	pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
2148	pCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
2149	}
2150
2151	pVCpu->hm.s.fForceTLBFlush = false;
2152	pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
2153	pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid;
2154	if (pCpu->fFlushAsidBeforeUse)
2155	{
2156	if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_SINGLE_CONTEXT)
2157	hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_SINGLE_CONTEXT, 0 /* GCPtr */);
2158	else if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_ALL_CONTEXTS)
2159	{
2160	hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_ALL_CONTEXTS, 0 /* GCPtr */);
2161	pCpu->fFlushAsidBeforeUse = false;
2162	}
2163	else
2164	{
2165	/* hmR0VmxSetupTaggedTlb() ensures we never get here. Paranoia. */
2166	AssertMsgFailed(("Unsupported VPID-flush context type.\n"));
2167	}
2168	}
2169	}
2170
2171	AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes,
2172	("Flush count mismatch for cpu %d (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
2173	AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
2174	("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pCpu->idCpu,
2175	pCpu->uCurrentAsid, pCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
2176	AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
2177	("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
2178
2179	int rc = VMXWriteVmcs32(VMX_VMCS16_VPID, pVCpu->hm.s.uCurrentAsid);
2180	AssertRC(rc);
2181	}
2182
2183
2184	/**
2185	* Flushes the guest TLB entry based on CPU capabilities.
2186	*
2187	* @param pVCpu The cross context virtual CPU structure.
2188	* @param pCpu Pointer to the global HM CPU struct.
2189	*/
2190	DECLINLINE(void) hmR0VmxFlushTaggedTlb(PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2191	{
2192	#ifdef HMVMX_ALWAYS_FLUSH_TLB
2193	VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2194	#endif
2195	PVM pVM = pVCpu->CTX_SUFF(pVM);
2196	switch (pVM->hm.s.vmx.uFlushTaggedTlb)
2197	{
2198	case HMVMX_FLUSH_TAGGED_TLB_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pVM, pVCpu, pCpu); break;
2199	case HMVMX_FLUSH_TAGGED_TLB_EPT: hmR0VmxFlushTaggedTlbEpt(pVM, pVCpu, pCpu); break;
2200	case HMVMX_FLUSH_TAGGED_TLB_VPID: hmR0VmxFlushTaggedTlbVpid(pVM, pVCpu, pCpu); break;
2201	case HMVMX_FLUSH_TAGGED_TLB_NONE: hmR0VmxFlushTaggedTlbNone(pVM, pVCpu, pCpu); break;
2202	default:
2203	AssertMsgFailed(("Invalid flush-tag function identifier\n"));
2204	break;
2205	}
2206
2207	/* Don't assert that VMCPU_FF_TLB_FLUSH should no longer be pending. It can be set by other EMTs. */
2208	}
2209
2210
2211	/**
2212	* Sets up the appropriate tagged TLB-flush level and handler for flushing guest
2213	* TLB entries from the host TLB before VM-entry.
2214	*
2215	* @returns VBox status code.
2216	* @param pVM The cross context VM structure.
2217	*/
2218	static int hmR0VmxSetupTaggedTlb(PVM pVM)
2219	{
2220	/*
2221	* Determine optimal flush type for Nested Paging.
2222	* We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup unrestricted
2223	* guest execution (see hmR3InitFinalizeR0()).
2224	*/
2225	if (pVM->hm.s.fNestedPaging)
2226	{
2227	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT)
2228	{
2229	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT)
2230	pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_SINGLE_CONTEXT;
2231	else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
2232	pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_ALL_CONTEXTS;
2233	else
2234	{
2235	/* Shouldn't happen. EPT is supported but no suitable flush-types supported. */
2236	pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2237	pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_FLUSH_TYPE_UNSUPPORTED;
2238	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2239	}
2240
2241	/* Make sure the write-back cacheable memory type for EPT is supported. */
2242	if (RT_UNLIKELY(!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB)))
2243	{
2244	pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2245	pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_MEM_TYPE_NOT_WB;
2246	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2247	}
2248
2249	/* EPT requires a page-walk length of 4. */
2250	if (RT_UNLIKELY(!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_PAGE_WALK_LENGTH_4)))
2251	{
2252	pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2253	pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_PAGE_WALK_LENGTH_UNSUPPORTED;
2254	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2255	}
2256	}
2257	else
2258	{
2259	/* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */
2260	pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2261	pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_INVEPT_UNAVAILABLE;
2262	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2263	}
2264	}
2265
2266	/*
2267	* Determine optimal flush type for VPID.
2268	*/
2269	if (pVM->hm.s.vmx.fVpid)
2270	{
2271	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID)
2272	{
2273	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT)
2274	pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_SINGLE_CONTEXT;
2275	else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS)
2276	pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_ALL_CONTEXTS;
2277	else
2278	{
2279	/* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */
2280	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
2281	LogRel(("hmR0VmxSetupTaggedTlb: Only INDIV_ADDR supported. Ignoring VPID.\n"));
2282	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
2283	LogRel(("hmR0VmxSetupTaggedTlb: Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
2284	pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NOT_SUPPORTED;
2285	pVM->hm.s.vmx.fVpid = false;
2286	}
2287	}
2288	else
2289	{
2290	/* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */
2291	Log4(("hmR0VmxSetupTaggedTlb: VPID supported without INVEPT support. Ignoring VPID.\n"));
2292	pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NOT_SUPPORTED;
2293	pVM->hm.s.vmx.fVpid = false;
2294	}
2295	}
2296
2297	/*
2298	* Setup the handler for flushing tagged-TLBs.
2299	*/
2300	if (pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid)
2301	pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_EPT_VPID;
2302	else if (pVM->hm.s.fNestedPaging)
2303	pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_EPT;
2304	else if (pVM->hm.s.vmx.fVpid)
2305	pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_VPID;
2306	else
2307	pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_NONE;
2308	return VINF_SUCCESS;
2309	}
2310
2311
2312	/**
2313	* Sets up pin-based VM-execution controls in the VMCS.
2314	*
2315	* @returns VBox status code.
2316	* @param pVM The cross context VM structure.
2317	* @param pVCpu The cross context virtual CPU structure.
2318	*/
2319	static int hmR0VmxSetupPinCtls(PVM pVM, PVMCPU pVCpu)
2320	{
2321	AssertPtr(pVM);
2322	AssertPtr(pVCpu);
2323
2324	uint32_t val = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0; /* Bits set here must always be set. */
2325	uint32_t zap = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1; /* Bits cleared here must always be cleared. */
2326
2327	val \|= VMX_VMCS_CTRL_PIN_EXEC_EXT_INT_EXIT /* External interrupts cause a VM-exit. */
2328	\| VMX_VMCS_CTRL_PIN_EXEC_NMI_EXIT; /* Non-maskable interrupts (NMIs) cause a VM-exit. */
2329
2330	if (pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
2331	val \|= VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI; /* Use virtual NMIs and virtual-NMI blocking features. */
2332
2333	/* Enable the VMX preemption timer. */
2334	if (pVM->hm.s.vmx.fUsePreemptTimer)
2335	{
2336	Assert(pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER);
2337	val \|= VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER;
2338	}
2339
2340	#if 0
2341	/* Enable posted-interrupt processing. */
2342	if (pVM->hm.s.fPostedIntrs)
2343	{
2344	Assert(pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_POSTED_INTR);
2345	Assert(pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT);
2346	val \|= VMX_VMCS_CTRL_PIN_EXEC_POSTED_INTR;
2347	}
2348	#endif
2349
2350	if ((val & zap) != val)
2351	{
2352	LogRel(("hmR0VmxSetupPinCtls: Invalid pin-based VM-execution controls combo! cpu=%#RX64 val=%#RX64 zap=%#RX64\n",
2353	pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0, val, zap));
2354	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC;
2355	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2356	}
2357
2358	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, val);
2359	AssertRCReturn(rc, rc);
2360
2361	pVCpu->hm.s.vmx.u32PinCtls = val;
2362	return rc;
2363	}
2364
2365
2366	/**
2367	* Sets up processor-based VM-execution controls in the VMCS.
2368	*
2369	* @returns VBox status code.
2370	* @param pVM The cross context VM structure.
2371	* @param pVCpu The cross context virtual CPU structure.
2372	*/
2373	static int hmR0VmxSetupProcCtls(PVM pVM, PVMCPU pVCpu)
2374	{
2375	AssertPtr(pVM);
2376	AssertPtr(pVCpu);
2377
2378	int rc = VERR_INTERNAL_ERROR_5;
2379	uint32_t val = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0; /* Bits set here must be set in the VMCS. */
2380	uint32_t zap = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
2381
2382	val \|= VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT /* HLT causes a VM-exit. */
2383	\| VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING /* Use TSC-offsetting. */
2384	\| VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */
2385	\| VMX_VMCS_CTRL_PROC_EXEC_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */
2386	\| VMX_VMCS_CTRL_PROC_EXEC_RDPMC_EXIT /* RDPMC causes a VM-exit. */
2387	\| VMX_VMCS_CTRL_PROC_EXEC_MONITOR_EXIT /* MONITOR causes a VM-exit. */
2388	\| VMX_VMCS_CTRL_PROC_EXEC_MWAIT_EXIT; /* MWAIT causes a VM-exit. */
2389
2390	/* We toggle VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */
2391	if ( !(pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT)
2392	\|\| (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0 & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT))
2393	{
2394	LogRel(("hmR0VmxSetupProcCtls: Unsupported VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT combo!"));
2395	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT;
2396	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2397	}
2398
2399	/* Without Nested Paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */
2400	if (!pVM->hm.s.fNestedPaging)
2401	{
2402	Assert(!pVM->hm.s.vmx.fUnrestrictedGuest); /* Paranoia. */
2403	val \|= VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT
2404	\| VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
2405	\| VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
2406	}
2407
2408	/* Use TPR shadowing if supported by the CPU. */
2409	if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
2410	{
2411	Assert(pVCpu->hm.s.vmx.HCPhysVirtApic);
2412	Assert(!(pVCpu->hm.s.vmx.HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */
2413	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, 0);
2414	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_VAPIC_PAGEADDR_FULL, pVCpu->hm.s.vmx.HCPhysVirtApic);
2415	AssertRCReturn(rc, rc);
2416
2417	val \|= VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */
2418	/* CR8 writes cause a VM-exit based on TPR threshold. */
2419	Assert(!(val & VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT));
2420	Assert(!(val & VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT));
2421	}
2422	else
2423	{
2424	/*
2425	* Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is invalid on 32-bit Intel CPUs.
2426	* Set this control only for 64-bit guests.
2427	*/
2428	if (pVM->hm.s.fAllow64BitGuests)
2429	{
2430	val \|= VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT /* CR8 reads cause a VM-exit. */
2431	\| VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT; /* CR8 writes cause a VM-exit. */
2432	}
2433	}
2434
2435	/* Use MSR-bitmaps if supported by the CPU. */
2436	if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
2437	{
2438	val \|= VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS;
2439
2440	Assert(pVCpu->hm.s.vmx.HCPhysMsrBitmap);
2441	Assert(!(pVCpu->hm.s.vmx.HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */
2442	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, pVCpu->hm.s.vmx.HCPhysMsrBitmap);
2443	AssertRCReturn(rc, rc);
2444
2445	/*
2446	* The guest can access the following MSRs (read, write) without causing VM-exits; they are loaded/stored
2447	* automatically using dedicated fields in the VMCS.
2448	*/
2449	hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_CS, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2450	hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_ESP, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2451	hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_EIP, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2452	hmR0VmxSetMsrPermission(pVCpu, MSR_K8_GS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2453	hmR0VmxSetMsrPermission(pVCpu, MSR_K8_FS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2454
2455	#if HC_ARCH_BITS == 64
2456	/*
2457	* Set passthru permissions for the following MSRs (mandatory for VT-x) required for 64-bit guests.
2458	*/
2459	if (pVM->hm.s.fAllow64BitGuests)
2460	{
2461	hmR0VmxSetMsrPermission(pVCpu, MSR_K8_LSTAR, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2462	hmR0VmxSetMsrPermission(pVCpu, MSR_K6_STAR, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2463	hmR0VmxSetMsrPermission(pVCpu, MSR_K8_SF_MASK, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2464	hmR0VmxSetMsrPermission(pVCpu, MSR_K8_KERNEL_GS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2465	}
2466	#endif
2467	/* Though MSR_IA32_PERF_GLOBAL_CTRL is saved/restored lazily, we want intercept reads/write to it for now. */
2468	}
2469
2470	/* Use the secondary processor-based VM-execution controls if supported by the CPU. */
2471	if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
2472	val \|= VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL;
2473
2474	if ((val & zap) != val)
2475	{
2476	LogRel(("hmR0VmxSetupProcCtls: Invalid processor-based VM-execution controls combo! cpu=%#RX64 val=%#RX64 zap=%#RX64\n",
2477	pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0, val, zap));
2478	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC;
2479	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2480	}
2481
2482	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, val);
2483	AssertRCReturn(rc, rc);
2484
2485	pVCpu->hm.s.vmx.u32ProcCtls = val;
2486
2487	/*
2488	* Secondary processor-based VM-execution controls.
2489	*/
2490	if (RT_LIKELY(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL))
2491	{
2492	val = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0; /* Bits set here must be set in the VMCS. */
2493	zap = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
2494
2495	if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT)
2496	val \|= VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT; /* WBINVD causes a VM-exit. */
2497
2498	if (pVM->hm.s.fNestedPaging)
2499	val \|= VMX_VMCS_CTRL_PROC_EXEC2_EPT; /* Enable EPT. */
2500	else
2501	{
2502	/*
2503	* Without Nested Paging, INVPCID should cause a VM-exit. Enabling this bit causes the CPU to refer to
2504	* VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT when INVPCID is executed by the guest.
2505	* See Intel spec. 25.4 "Changes to instruction behaviour in VMX non-root operation".
2506	*/
2507	if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_INVPCID)
2508	val \|= VMX_VMCS_CTRL_PROC_EXEC2_INVPCID;
2509	}
2510
2511	if (pVM->hm.s.vmx.fVpid)
2512	val \|= VMX_VMCS_CTRL_PROC_EXEC2_VPID; /* Enable VPID. */
2513
2514	if (pVM->hm.s.vmx.fUnrestrictedGuest)
2515	val \|= VMX_VMCS_CTRL_PROC_EXEC2_UNRESTRICTED_GUEST; /* Enable Unrestricted Execution. */
2516
2517	#if 0
2518	if (pVM->hm.s.fVirtApicRegs)
2519	{
2520	Assert(pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_APIC_REG_VIRT);
2521	val \|= VMX_VMCS_CTRL_PROC_EXEC2_APIC_REG_VIRT; /* Enable APIC-register virtualization. */
2522
2523	Assert(pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_INTR_DELIVERY);
2524	val \|= VMX_VMCS_CTRL_PROC_EXEC2_VIRT_INTR_DELIVERY; /* Enable virtual-interrupt delivery. */
2525	}
2526	#endif
2527
2528	/* Enable Virtual-APIC page accesses if supported by the CPU. This is essentially where the TPR shadow resides. */
2529	/** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be
2530	* done dynamically. */
2531	if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)
2532	{
2533	Assert(pVM->hm.s.vmx.HCPhysApicAccess);
2534	Assert(!(pVM->hm.s.vmx.HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */
2535	val \|= VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC; /* Virtualize APIC accesses. */
2536	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, pVM->hm.s.vmx.HCPhysApicAccess);
2537	AssertRCReturn(rc, rc);
2538	}
2539
2540	if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP)
2541	val \|= VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP; /* Enable RDTSCP support. */
2542
2543	if ( pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_PAUSE_LOOP_EXIT
2544	&& pVM->hm.s.vmx.cPleGapTicks
2545	&& pVM->hm.s.vmx.cPleWindowTicks)
2546	{
2547	val \|= VMX_VMCS_CTRL_PROC_EXEC2_PAUSE_LOOP_EXIT; /* Enable pause-loop exiting. */
2548
2549	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, pVM->hm.s.vmx.cPleGapTicks);
2550	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, pVM->hm.s.vmx.cPleWindowTicks);
2551	AssertRCReturn(rc, rc);
2552	}
2553
2554	if ((val & zap) != val)
2555	{
2556	LogRel(("hmR0VmxSetupProcCtls: Invalid secondary processor-based VM-execution controls combo! "
2557	"cpu=%#RX64 val=%#RX64 zap=%#RX64\n", pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0, val, zap));
2558	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC2;
2559	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2560	}
2561
2562	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, val);
2563	AssertRCReturn(rc, rc);
2564
2565	pVCpu->hm.s.vmx.u32ProcCtls2 = val;
2566	}
2567	else if (RT_UNLIKELY(pVM->hm.s.vmx.fUnrestrictedGuest))
2568	{
2569	LogRel(("hmR0VmxSetupProcCtls: Unrestricted Guest set as true when secondary processor-based VM-execution controls not "
2570	"available\n"));
2571	pVCpu->hm.s.u32HMError = VMX_UFC_INVALID_UX_COMBO;
2572	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2573	}
2574
2575	return VINF_SUCCESS;
2576	}
2577
2578
2579	/**
2580	* Sets up miscellaneous (everything other than Pin & Processor-based
2581	* VM-execution) control fields in the VMCS.
2582	*
2583	* @returns VBox status code.
2584	* @param pVM The cross context VM structure.
2585	* @param pVCpu The cross context virtual CPU structure.
2586	*/
2587	static int hmR0VmxSetupMiscCtls(PVM pVM, PVMCPU pVCpu)
2588	{
2589	NOREF(pVM);
2590	AssertPtr(pVM);
2591	AssertPtr(pVCpu);
2592
2593	int rc = VERR_GENERAL_FAILURE;
2594
2595	/* All fields are zero-initialized during allocation; but don't remove the commented block below. */
2596	#if 0
2597	/* All CR3 accesses cause VM-exits. Later we optimize CR3 accesses (see hmR0VmxLoadGuestCR3AndCR4())*/
2598	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, 0);
2599	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, 0);
2600
2601	/*
2602	* Set MASK & MATCH to 0. VMX checks if GuestPFErrCode & MASK == MATCH. If equal (in our case it always is)
2603	* and if the X86_XCPT_PF bit in the exception bitmap is set it causes a VM-exit, if clear doesn't cause an exit.
2604	* We thus use the exception bitmap to control it rather than use both.
2605	*/
2606	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, 0);
2607	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, 0);
2608
2609	/** @todo Explore possibility of using IO-bitmaps. */
2610	/* All IO & IOIO instructions cause VM-exits. */
2611	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_IO_BITMAP_A_FULL, 0);
2612	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_IO_BITMAP_B_FULL, 0);
2613
2614	/* Initialize the MSR-bitmap area. */
2615	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, 0);
2616	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, 0);
2617	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, 0);
2618	AssertRCReturn(rc, rc);
2619	#endif
2620
2621	/* Setup MSR auto-load/store area. */
2622	Assert(pVCpu->hm.s.vmx.HCPhysGuestMsr);
2623	Assert(!(pVCpu->hm.s.vmx.HCPhysGuestMsr & 0xf)); /* Lower 4 bits MBZ. */
2624	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, pVCpu->hm.s.vmx.HCPhysGuestMsr);
2625	rc \|= VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, pVCpu->hm.s.vmx.HCPhysGuestMsr);
2626	AssertRCReturn(rc, rc);
2627
2628	Assert(pVCpu->hm.s.vmx.HCPhysHostMsr);
2629	Assert(!(pVCpu->hm.s.vmx.HCPhysHostMsr & 0xf)); /* Lower 4 bits MBZ. */
2630	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, pVCpu->hm.s.vmx.HCPhysHostMsr);
2631	AssertRCReturn(rc, rc);
2632
2633	/* Set VMCS link pointer. Reserved for future use, must be -1. Intel spec. 24.4 "Guest-State Area". */
2634	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, UINT64_C(0xffffffffffffffff));
2635	AssertRCReturn(rc, rc);
2636
2637	/* All fields are zero-initialized during allocation; but don't remove the commented block below. */
2638	#if 0
2639	/* Setup debug controls */
2640	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, 0); /** @todo We don't support IA32_DEBUGCTL MSR. Should we? */
2641	rc \|= VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, 0);
2642	AssertRCReturn(rc, rc);
2643	#endif
2644
2645	return rc;
2646	}
2647
2648
2649	/**
2650	* Sets up the initial exception bitmap in the VMCS based on static conditions.
2651	*
2652	* We shall setup those exception intercepts that don't change during the
2653	* lifetime of the VM here. The rest are done dynamically while loading the
2654	* guest state.
2655	*
2656	* @returns VBox status code.
2657	* @param pVM The cross context VM structure.
2658	* @param pVCpu The cross context virtual CPU structure.
2659	*/
2660	static int hmR0VmxInitXcptBitmap(PVM pVM, PVMCPU pVCpu)
2661	{
2662	AssertPtr(pVM);
2663	AssertPtr(pVCpu);
2664
2665	LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
2666
2667	uint32_t u32XcptBitmap = 0;
2668
2669	/* Must always intercept #AC to prevent the guest from hanging the CPU. */
2670	u32XcptBitmap \|= RT_BIT_32(X86_XCPT_AC);
2671
2672	/* Because we need to maintain the DR6 state even when intercepting DRx reads
2673	and writes, and because recursive #DBs can cause the CPU hang, we must always
2674	intercept #DB. */
2675	u32XcptBitmap \|= RT_BIT_32(X86_XCPT_DB);
2676
2677	/* Without Nested Paging, #PF must cause a VM-exit so we can sync our shadow page tables. */
2678	if (!pVM->hm.s.fNestedPaging)
2679	u32XcptBitmap \|= RT_BIT(X86_XCPT_PF);
2680
2681	pVCpu->hm.s.vmx.u32XcptBitmap = u32XcptBitmap;
2682	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
2683	AssertRCReturn(rc, rc);
2684	return rc;
2685	}
2686
2687
2688	/**
2689	* Sets up the initial guest-state mask. The guest-state mask is consulted
2690	* before reading guest-state fields from the VMCS as VMREADs can be expensive
2691	* for the nested virtualization case (as it would cause a VM-exit).
2692	*
2693	* @param pVCpu The cross context virtual CPU structure.
2694	*/
2695	static int hmR0VmxInitUpdatedGuestStateMask(PVMCPU pVCpu)
2696	{
2697	/* Initially the guest-state is up-to-date as there is nothing in the VMCS. */
2698	HMVMXCPU_GST_RESET_TO(pVCpu, HMVMX_UPDATED_GUEST_ALL);
2699	return VINF_SUCCESS;
2700	}
2701
2702
2703	/**
2704	* Does per-VM VT-x initialization.
2705	*
2706	* @returns VBox status code.
2707	* @param pVM The cross context VM structure.
2708	*/
2709	VMMR0DECL(int) VMXR0InitVM(PVM pVM)
2710	{
2711	LogFlowFunc(("pVM=%p\n", pVM));
2712
2713	int rc = hmR0VmxStructsAlloc(pVM);
2714	if (RT_FAILURE(rc))
2715	{
2716	LogRel(("VMXR0InitVM: hmR0VmxStructsAlloc failed! rc=%Rrc\n", rc));
2717	return rc;
2718	}
2719
2720	return VINF_SUCCESS;
2721	}
2722
2723
2724	/**
2725	* Does per-VM VT-x termination.
2726	*
2727	* @returns VBox status code.
2728	* @param pVM The cross context VM structure.
2729	*/
2730	VMMR0DECL(int) VMXR0TermVM(PVM pVM)
2731	{
2732	LogFlowFunc(("pVM=%p\n", pVM));
2733
2734	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2735	if (pVM->hm.s.vmx.hMemObjScratch != NIL_RTR0MEMOBJ)
2736	ASMMemZero32(pVM->hm.s.vmx.pvScratch, PAGE_SIZE);
2737	#endif
2738	hmR0VmxStructsFree(pVM);
2739	return VINF_SUCCESS;
2740	}
2741
2742
2743	/**
2744	* Sets up the VM for execution under VT-x.
2745	* This function is only called once per-VM during initialization.
2746	*
2747	* @returns VBox status code.
2748	* @param pVM The cross context VM structure.
2749	*/
2750	VMMR0DECL(int) VMXR0SetupVM(PVM pVM)
2751	{
2752	AssertPtrReturn(pVM, VERR_INVALID_PARAMETER);
2753	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2754
2755	LogFlowFunc(("pVM=%p\n", pVM));
2756
2757	/*
2758	* Without UnrestrictedGuest, pRealModeTSS and pNonPagingModeEPTPageTable must always be allocated.
2759	* We no longer support the highly unlikely case of UnrestrictedGuest without pRealModeTSS. See hmR3InitFinalizeR0Intel().
2760	*/
2761	if ( !pVM->hm.s.vmx.fUnrestrictedGuest
2762	&& ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable
2763	\|\| !pVM->hm.s.vmx.pRealModeTSS))
2764	{
2765	LogRel(("VMXR0SetupVM: Invalid real-on-v86 state.\n"));
2766	return VERR_INTERNAL_ERROR;
2767	}
2768
2769	/* Initialize these always, see hmR3InitFinalizeR0().*/
2770	pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NONE;
2771	pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NONE;
2772
2773	/* Setup the tagged-TLB flush handlers. */
2774	int rc = hmR0VmxSetupTaggedTlb(pVM);
2775	if (RT_FAILURE(rc))
2776	{
2777	LogRel(("VMXR0SetupVM: hmR0VmxSetupTaggedTlb failed! rc=%Rrc\n", rc));
2778	return rc;
2779	}
2780
2781	/* Check if we can use the VMCS controls for swapping the EFER MSR. */
2782	Assert(!pVM->hm.s.vmx.fSupportsVmcsEfer);
2783	#if HC_ARCH_BITS == 64
2784	if ( (pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1 & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR)
2785	&& (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR)
2786	&& (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR))
2787	{
2788	pVM->hm.s.vmx.fSupportsVmcsEfer = true;
2789	}
2790	#endif
2791
2792	/* At least verify VMX is enabled, since we can't check if we're in VMX root mode without #GP'ing. */
2793	RTCCUINTREG uHostCR4 = ASMGetCR4();
2794	if (RT_UNLIKELY(!(uHostCR4 & X86_CR4_VMXE)))
2795	return VERR_VMX_NOT_IN_VMX_ROOT_MODE;
2796
2797	for (VMCPUID i = 0; i < pVM->cCpus; i++)
2798	{
2799	PVMCPU pVCpu = &pVM->aCpus[i];
2800	AssertPtr(pVCpu);
2801	AssertPtr(pVCpu->hm.s.vmx.pvVmcs);
2802
2803	/* Log the VCPU pointers, useful for debugging SMP VMs. */
2804	Log4(("VMXR0SetupVM: pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu));
2805
2806	/* Initialize the VM-exit history array with end-of-array markers (UINT16_MAX). */
2807	Assert(!pVCpu->hm.s.idxExitHistoryFree);
2808	HMCPU_EXIT_HISTORY_RESET(pVCpu);
2809
2810	/* Set revision dword at the beginning of the VMCS structure. */
2811	(uint32_t )pVCpu->hm.s.vmx.pvVmcs = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo);
2812
2813	/* Initialize our VMCS region in memory, set the VMCS launch state to "clear". */
2814	rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2815	AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXClearVmcs failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2816	hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2817
2818	/* Load this VMCS as the current VMCS. */
2819	rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2820	AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXActivateVmcs failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2821	hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2822
2823	rc = hmR0VmxSetupPinCtls(pVM, pVCpu);
2824	AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupPinCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2825	hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2826
2827	rc = hmR0VmxSetupProcCtls(pVM, pVCpu);
2828	AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupProcCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2829	hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2830
2831	rc = hmR0VmxSetupMiscCtls(pVM, pVCpu);
2832	AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupMiscCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2833	hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2834
2835	rc = hmR0VmxInitXcptBitmap(pVM, pVCpu);
2836	AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitXcptBitmap failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2837	hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2838
2839	rc = hmR0VmxInitUpdatedGuestStateMask(pVCpu);
2840	AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitUpdatedGuestStateMask failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2841	hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2842
2843	#if HC_ARCH_BITS == 32
2844	rc = hmR0VmxInitVmcsReadCache(pVM, pVCpu);
2845	AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitVmcsReadCache failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2846	hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2847	#endif
2848
2849	/* Re-sync the CPU's internal data into our VMCS memory region & reset the launch state to "clear". */
2850	rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2851	AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXClearVmcs(2) failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2852	hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2853
2854	pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
2855
2856	hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc);
2857	}
2858
2859	return VINF_SUCCESS;
2860	}
2861
2862
2863	/**
2864	* Saves the host control registers (CR0, CR3, CR4) into the host-state area in
2865	* the VMCS.
2866	*
2867	* @returns VBox status code.
2868	* @param pVM The cross context VM structure.
2869	* @param pVCpu The cross context virtual CPU structure.
2870	*/
2871	DECLINLINE(int) hmR0VmxSaveHostControlRegs(PVM pVM, PVMCPU pVCpu)
2872	{
2873	NOREF(pVM); NOREF(pVCpu);
2874
2875	RTCCUINTREG uReg = ASMGetCR0();
2876	int rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR0, uReg);
2877	AssertRCReturn(rc, rc);
2878
2879	uReg = ASMGetCR3();
2880	rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR3, uReg);
2881	AssertRCReturn(rc, rc);
2882
2883	uReg = ASMGetCR4();
2884	rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR4, uReg);
2885	AssertRCReturn(rc, rc);
2886	return rc;
2887	}
2888
2889
2890	#if HC_ARCH_BITS == 64
2891	/**
2892	* Macro for adjusting host segment selectors to satisfy VT-x's VM-entry
2893	* requirements. See hmR0VmxSaveHostSegmentRegs().
2894	*/
2895	# define VMXLOCAL_ADJUST_HOST_SEG(seg, selValue) \
2896	if ((selValue) & (X86_SEL_RPL \| X86_SEL_LDT)) \
2897	{ \
2898	bool fValidSelector = true; \
2899	if ((selValue) & X86_SEL_LDT) \
2900	{ \
2901	uint32_t uAttr = ASMGetSegAttr((selValue)); \
2902	fValidSelector = RT_BOOL(uAttr != UINT32_MAX && (uAttr & X86_DESC_P)); \
2903	} \
2904	if (fValidSelector) \
2905	{ \
2906	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_##seg; \
2907	pVCpu->hm.s.vmx.RestoreHost.uHostSel##seg = (selValue); \
2908	} \
2909	(selValue) = 0; \
2910	}
2911	#endif
2912
2913
2914	/**
2915	* Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into
2916	* the host-state area in the VMCS.
2917	*
2918	* @returns VBox status code.
2919	* @param pVM The cross context VM structure.
2920	* @param pVCpu The cross context virtual CPU structure.
2921	*/
2922	DECLINLINE(int) hmR0VmxSaveHostSegmentRegs(PVM pVM, PVMCPU pVCpu)
2923	{
2924	int rc = VERR_INTERNAL_ERROR_5;
2925
2926	#if HC_ARCH_BITS == 64
2927	/*
2928	* If we've executed guest code using VT-x, the host-state bits will be messed up. We
2929	* should -not- save the messed up state without restoring the original host-state. See @bugref{7240}.
2930	*
2931	* This apparently can happen (most likely the FPU changes), deal with it rather than asserting.
2932	* Was observed booting Solaris10u10 32-bit guest.
2933	*/
2934	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
2935	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
2936	{
2937	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags,
2938	pVCpu->idCpu));
2939	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
2940	}
2941	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
2942	#else
2943	RT_NOREF(pVCpu);
2944	#endif
2945
2946	/*
2947	* Host DS, ES, FS and GS segment registers.
2948	*/
2949	#if HC_ARCH_BITS == 64
2950	RTSEL uSelDS = ASMGetDS();
2951	RTSEL uSelES = ASMGetES();
2952	RTSEL uSelFS = ASMGetFS();
2953	RTSEL uSelGS = ASMGetGS();
2954	#else
2955	RTSEL uSelDS = 0;
2956	RTSEL uSelES = 0;
2957	RTSEL uSelFS = 0;
2958	RTSEL uSelGS = 0;
2959	#endif
2960
2961	/*
2962	* Host CS and SS segment registers.
2963	*/
2964	RTSEL uSelCS = ASMGetCS();
2965	RTSEL uSelSS = ASMGetSS();
2966
2967	/*
2968	* Host TR segment register.
2969	*/
2970	RTSEL uSelTR = ASMGetTR();
2971
2972	#if HC_ARCH_BITS == 64
2973	/*
2974	* Determine if the host segment registers are suitable for VT-x. Otherwise use zero to gain VM-entry and restore them
2975	* before we get preempted. See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers".
2976	*/
2977	VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
2978	VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
2979	VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
2980	VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
2981	# undef VMXLOCAL_ADJUST_HOST_SEG
2982	#endif
2983
2984	/* Verification based on Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers" */
2985	Assert(!(uSelCS & X86_SEL_RPL)); Assert(!(uSelCS & X86_SEL_LDT));
2986	Assert(!(uSelSS & X86_SEL_RPL)); Assert(!(uSelSS & X86_SEL_LDT));
2987	Assert(!(uSelDS & X86_SEL_RPL)); Assert(!(uSelDS & X86_SEL_LDT));
2988	Assert(!(uSelES & X86_SEL_RPL)); Assert(!(uSelES & X86_SEL_LDT));
2989	Assert(!(uSelFS & X86_SEL_RPL)); Assert(!(uSelFS & X86_SEL_LDT));
2990	Assert(!(uSelGS & X86_SEL_RPL)); Assert(!(uSelGS & X86_SEL_LDT));
2991	Assert(!(uSelTR & X86_SEL_RPL)); Assert(!(uSelTR & X86_SEL_LDT));
2992	Assert(uSelCS);
2993	Assert(uSelTR);
2994
2995	/* Assertion is right but we would not have updated u32ExitCtls yet. */
2996	#if 0
2997	if (!(pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE))
2998	Assert(uSelSS != 0);
2999	#endif
3000
3001	/* Write these host selector fields into the host-state area in the VMCS. */
3002	rc = VMXWriteVmcs32(VMX_VMCS16_HOST_CS_SEL, uSelCS);
3003	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_SS_SEL, uSelSS);
3004	#if HC_ARCH_BITS == 64
3005	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_DS_SEL, uSelDS);
3006	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_ES_SEL, uSelES);
3007	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_FS_SEL, uSelFS);
3008	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_GS_SEL, uSelGS);
3009	#else
3010	NOREF(uSelDS);
3011	NOREF(uSelES);
3012	NOREF(uSelFS);
3013	NOREF(uSelGS);
3014	#endif
3015	rc \|= VMXWriteVmcs32(VMX_VMCS16_HOST_TR_SEL, uSelTR);
3016	AssertRCReturn(rc, rc);
3017
3018	/*
3019	* Host GDTR and IDTR.
3020	*/
3021	RTGDTR Gdtr;
3022	RTIDTR Idtr;
3023	RT_ZERO(Gdtr);
3024	RT_ZERO(Idtr);
3025	ASMGetGDTR(&Gdtr);
3026	ASMGetIDTR(&Idtr);
3027	rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, Gdtr.pGdt);
3028	rc \|= VMXWriteVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, Idtr.pIdt);
3029	AssertRCReturn(rc, rc);
3030
3031	#if HC_ARCH_BITS == 64
3032	/*
3033	* Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps them to the
3034	* maximum limit (0xffff) on every VM-exit.
3035	*/
3036	if (Gdtr.cbGdt != 0xffff)
3037	{
3038	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDTR;
3039	AssertCompile(sizeof(Gdtr) == sizeof(X86XDTR64));
3040	memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
3041	}
3042
3043	/*
3044	* IDT limit is effectively capped at 0xfff. (See Intel spec. 6.14.1 "64-Bit Mode IDT"
3045	* and Intel spec. 6.2 "Exception and Interrupt Vectors".) Therefore if the host has the limit as 0xfff, VT-x
3046	* bloating the limit to 0xffff shouldn't cause any different CPU behavior. However, several hosts either insists
3047	* on 0xfff being the limit (Windows Patch Guard) or uses the limit for other purposes (darwin puts the CPU ID in there
3048	* but botches sidt alignment in at least one consumer). So, we're only allowing IDTR.LIMIT to be left at 0xffff on
3049	* hosts where we are pretty sure it won't cause trouble.
3050	*/
3051	# if defined(RT_OS_LINUX) \|\| defined(RT_OS_SOLARIS)
3052	if (Idtr.cbIdt < 0x0fff)
3053	# else
3054	if (Idtr.cbIdt != 0xffff)
3055	# endif
3056	{
3057	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_IDTR;
3058	AssertCompile(sizeof(Idtr) == sizeof(X86XDTR64));
3059	memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostIdtr, &Idtr, sizeof(X86XDTR64));
3060	}
3061	#endif
3062
3063	/*
3064	* Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI and RPL bits
3065	* is effectively what the CPU does for "scaling by 8". TI is always 0 and RPL should be too in most cases.
3066	*/
3067	AssertMsgReturn((uSelTR \| X86_SEL_RPL_LDT) <= Gdtr.cbGdt,
3068	("hmR0VmxSaveHostSegmentRegs: TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, Gdtr.cbGdt),
3069	VERR_VMX_INVALID_HOST_STATE);
3070
3071	PCX86DESCHC pDesc = (PCX86DESCHC)(Gdtr.pGdt + (uSelTR & X86_SEL_MASK));
3072	#if HC_ARCH_BITS == 64
3073	uintptr_t uTRBase = X86DESC64_BASE(pDesc);
3074
3075	/*
3076	* VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on all VM-exits.
3077	* The type is the same for 64-bit busy TSS[1]. The limit needs manual restoration if the host has something else.
3078	* Task switching is not supported in 64-bit mode[2], but the limit still matters as IOPM is supported in 64-bit mode.
3079	* Restoring the limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0.
3080	*
3081	* [1] See Intel spec. 3.5 "System Descriptor Types".
3082	* [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode".
3083	*/
3084	Assert(pDesc->System.u4Type == 11);
3085	if ( pDesc->System.u16LimitLow != 0x67
3086	\|\| pDesc->System.u4LimitHigh)
3087	{
3088	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_SEL_TR;
3089	/* If the host has made GDT read-only, we would need to temporarily toggle CR0.WP before writing the GDT. */
3090	if (pVM->hm.s.fHostKernelFeatures & SUPKERNELFEATURES_GDT_READ_ONLY)
3091	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_GDT_READ_ONLY;
3092	pVCpu->hm.s.vmx.RestoreHost.uHostSelTR = uSelTR;
3093
3094	/* Store the GDTR here as we need it while restoring TR. */
3095	memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
3096	}
3097	#else
3098	NOREF(pVM);
3099	uintptr_t uTRBase = X86DESC_BASE(pDesc);
3100	#endif
3101	rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_TR_BASE, uTRBase);
3102	AssertRCReturn(rc, rc);
3103
3104	/*
3105	* Host FS base and GS base.
3106	*/
3107	#if HC_ARCH_BITS == 64
3108	uint64_t u64FSBase = ASMRdMsr(MSR_K8_FS_BASE);
3109	uint64_t u64GSBase = ASMRdMsr(MSR_K8_GS_BASE);
3110	rc = VMXWriteVmcs64(VMX_VMCS_HOST_FS_BASE, u64FSBase);
3111	rc \|= VMXWriteVmcs64(VMX_VMCS_HOST_GS_BASE, u64GSBase);
3112	AssertRCReturn(rc, rc);
3113
3114	/* Store the base if we have to restore FS or GS manually as we need to restore the base as well. */
3115	if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_FS)
3116	pVCpu->hm.s.vmx.RestoreHost.uHostFSBase = u64FSBase;
3117	if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_GS)
3118	pVCpu->hm.s.vmx.RestoreHost.uHostGSBase = u64GSBase;
3119	#endif
3120	return rc;
3121	}
3122
3123
3124	/**
3125	* Saves certain host MSRs in the VM-exit MSR-load area and some in the
3126	* host-state area of the VMCS. Theses MSRs will be automatically restored on
3127	* the host after every successful VM-exit.
3128	*
3129	* @returns VBox status code.
3130	* @param pVM The cross context VM structure.
3131	* @param pVCpu The cross context virtual CPU structure.
3132	*
3133	* @remarks No-long-jump zone!!!
3134	*/
3135	DECLINLINE(int) hmR0VmxSaveHostMsrs(PVM pVM, PVMCPU pVCpu)
3136	{
3137	NOREF(pVM);
3138
3139	AssertPtr(pVCpu);
3140	AssertPtr(pVCpu->hm.s.vmx.pvHostMsr);
3141
3142	/*
3143	* Save MSRs that we restore lazily (due to preemption or transition to ring-3)
3144	* rather than swapping them on every VM-entry.
3145	*/
3146	hmR0VmxLazySaveHostMsrs(pVCpu);
3147
3148	/*
3149	* Host Sysenter MSRs.
3150	*/
3151	int rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS));
3152	#if HC_ARCH_BITS == 32
3153	rc \|= VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr_Low(MSR_IA32_SYSENTER_ESP));
3154	rc \|= VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr_Low(MSR_IA32_SYSENTER_EIP));
3155	#else
3156	rc \|= VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP));
3157	rc \|= VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP));
3158	#endif
3159	AssertRCReturn(rc, rc);
3160
3161	/*
3162	* Host EFER MSR.
3163	* If the CPU supports the newer VMCS controls for managing EFER, use it.
3164	* Otherwise it's done as part of auto-load/store MSR area in the VMCS, see hmR0VmxLoadGuestMsrs().
3165	*/
3166	if (pVM->hm.s.vmx.fSupportsVmcsEfer)
3167	{
3168	rc = VMXWriteVmcs64(VMX_VMCS64_HOST_EFER_FULL, pVM->hm.s.vmx.u64HostEfer);
3169	AssertRCReturn(rc, rc);
3170	}
3171
3172	/** @todo IA32_PERF_GLOBALCTRL, IA32_PAT also see
3173	* hmR0VmxLoadGuestExitCtls() !! */
3174
3175	return rc;
3176	}
3177
3178
3179	/**
3180	* Figures out if we need to swap the EFER MSR which is particularly expensive.
3181	*
3182	* We check all relevant bits. For now, that's everything besides LMA/LME, as
3183	* these two bits are handled by VM-entry, see hmR0VmxLoadGuestExitCtls() and
3184	* hmR0VMxLoadGuestEntryCtls().
3185	*
3186	* @returns true if we need to load guest EFER, false otherwise.
3187	* @param pVCpu The cross context virtual CPU structure.
3188	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3189	* out-of-sync. Make sure to update the required fields
3190	* before using them.
3191	*
3192	* @remarks Requires EFER, CR4.
3193	* @remarks No-long-jump zone!!!
3194	*/
3195	static bool hmR0VmxShouldSwapEferMsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3196	{
3197	#ifdef HMVMX_ALWAYS_SWAP_EFER
3198	return true;
3199	#endif
3200
3201	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
3202	/* For 32-bit hosts running 64-bit guests, we always swap EFER in the world-switcher. Nothing to do here. */
3203	if (CPUMIsGuestInLongMode(pVCpu))
3204	return false;
3205	#endif
3206
3207	PVM pVM = pVCpu->CTX_SUFF(pVM);
3208	uint64_t u64HostEfer = pVM->hm.s.vmx.u64HostEfer;
3209	uint64_t u64GuestEfer = pMixedCtx->msrEFER;
3210
3211	/*
3212	* For 64-bit guests, if EFER.SCE bit differs, we need to swap to ensure that the
3213	* guest's SYSCALL behaviour isn't screwed. See @bugref{7386}.
3214	*/
3215	if ( CPUMIsGuestInLongMode(pVCpu)
3216	&& (u64GuestEfer & MSR_K6_EFER_SCE) != (u64HostEfer & MSR_K6_EFER_SCE))
3217	{
3218	return true;
3219	}
3220
3221	/*
3222	* If the guest uses PAE and EFER.NXE bit differs, we need to swap EFER as it
3223	* affects guest paging. 64-bit paging implies CR4.PAE as well.
3224	* See Intel spec. 4.5 "IA-32e Paging" and Intel spec. 4.1.1 "Three Paging Modes".
3225	*/
3226	if ( (pMixedCtx->cr4 & X86_CR4_PAE)
3227	&& (pMixedCtx->cr0 & X86_CR0_PG)
3228	&& (u64GuestEfer & MSR_K6_EFER_NXE) != (u64HostEfer & MSR_K6_EFER_NXE))
3229	{
3230	/* Assert that host is PAE capable. */
3231	Assert(pVM->hm.s.cpuid.u32AMDFeatureEDX & X86_CPUID_EXT_FEATURE_EDX_NX);
3232	return true;
3233	}
3234
3235	/** @todo Check the latest Intel spec. for any other bits,
3236	* like SMEP/SMAP? */
3237	return false;
3238	}
3239
3240
3241	/**
3242	* Sets up VM-entry controls in the VMCS. These controls can affect things done
3243	* on VM-exit; e.g. "load debug controls", see Intel spec. 24.8.1 "VM-entry
3244	* controls".
3245	*
3246	* @returns VBox status code.
3247	* @param pVCpu The cross context virtual CPU structure.
3248	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3249	* out-of-sync. Make sure to update the required fields
3250	* before using them.
3251	*
3252	* @remarks Requires EFER.
3253	* @remarks No-long-jump zone!!!
3254	*/
3255	DECLINLINE(int) hmR0VmxLoadGuestEntryCtls(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3256	{
3257	int rc = VINF_SUCCESS;
3258	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_ENTRY_CTLS))
3259	{
3260	PVM pVM = pVCpu->CTX_SUFF(pVM);
3261	uint32_t val = pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0; /* Bits set here must be set in the VMCS. */
3262	uint32_t zap = pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3263
3264	/* Load debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x capable CPUs only supports the 1-setting of this bit. */
3265	val \|= VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG;
3266
3267	/* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry. */
3268	if (CPUMIsGuestInLongModeEx(pMixedCtx))
3269	{
3270	val \|= VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST;
3271	Log4(("Load[%RU32]: VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST\n", pVCpu->idCpu));
3272	}
3273	else
3274	Assert(!(val & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST));
3275
3276	/* If the CPU supports the newer VMCS controls for managing guest/host EFER, use it. */
3277	if ( pVM->hm.s.vmx.fSupportsVmcsEfer
3278	&& hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
3279	{
3280	val \|= VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR;
3281	Log4(("Load[%RU32]: VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR\n", pVCpu->idCpu));
3282	}
3283
3284	/*
3285	* The following should -not- be set (since we're not in SMM mode):
3286	* - VMX_VMCS_CTRL_ENTRY_ENTRY_SMM
3287	* - VMX_VMCS_CTRL_ENTRY_DEACTIVATE_DUALMON
3288	*/
3289
3290	/** @todo VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR,
3291	* VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR. */
3292
3293	if ((val & zap) != val)
3294	{
3295	LogRel(("hmR0VmxLoadGuestEntryCtls: Invalid VM-entry controls combo! cpu=%RX64 val=%RX64 zap=%RX64\n",
3296	pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0, val, zap));
3297	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY;
3298	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3299	}
3300
3301	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, val);
3302	AssertRCReturn(rc, rc);
3303
3304	pVCpu->hm.s.vmx.u32EntryCtls = val;
3305	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_ENTRY_CTLS);
3306	}
3307	return rc;
3308	}
3309
3310
3311	/**
3312	* Sets up the VM-exit controls in the VMCS.
3313	*
3314	* @returns VBox status code.
3315	* @param pVCpu The cross context virtual CPU structure.
3316	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3317	* out-of-sync. Make sure to update the required fields
3318	* before using them.
3319	*
3320	* @remarks Requires EFER.
3321	*/
3322	DECLINLINE(int) hmR0VmxLoadGuestExitCtls(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3323	{
3324	NOREF(pMixedCtx);
3325
3326	int rc = VINF_SUCCESS;
3327	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_EXIT_CTLS))
3328	{
3329	PVM pVM = pVCpu->CTX_SUFF(pVM);
3330	uint32_t val = pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0; /* Bits set here must be set in the VMCS. */
3331	uint32_t zap = pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3332
3333	/* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only supported the 1-setting of this bit. */
3334	val \|= VMX_VMCS_CTRL_EXIT_SAVE_DEBUG;
3335
3336	/*
3337	* Set the host long mode active (EFER.LMA) bit (which Intel calls "Host address-space size") if necessary.
3338	* On VM-exit, VT-x sets both the host EFER.LMA and EFER.LME bit to this value. See assertion in hmR0VmxSaveHostMsrs().
3339	*/
3340	#if HC_ARCH_BITS == 64
3341	val \|= VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE;
3342	Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE\n", pVCpu->idCpu));
3343	#else
3344	Assert( pVCpu->hm.s.vmx.pfnStartVM == VMXR0SwitcherStartVM64
3345	\|\| pVCpu->hm.s.vmx.pfnStartVM == VMXR0StartVM32);
3346	/* Set the host address-space size based on the switcher, not guest state. See @bugref{8432}. */
3347	if (pVCpu->hm.s.vmx.pfnStartVM == VMXR0SwitcherStartVM64)
3348	{
3349	/* The switcher returns to long mode, EFER is managed by the switcher. */
3350	val \|= VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE;
3351	Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE\n", pVCpu->idCpu));
3352	}
3353	else
3354	Assert(!(val & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE));
3355	#endif
3356
3357	/* If the newer VMCS fields for managing EFER exists, use it. */
3358	if ( pVM->hm.s.vmx.fSupportsVmcsEfer
3359	&& hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
3360	{
3361	val \|= VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR
3362	\| VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR;
3363	Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR, VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR\n", pVCpu->idCpu));
3364	}
3365
3366	/* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */
3367	Assert(!(val & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT));
3368
3369	/** @todo VMX_VMCS_CTRL_EXIT_LOAD_PERF_MSR,
3370	* VMX_VMCS_CTRL_EXIT_SAVE_GUEST_PAT_MSR,
3371	* VMX_VMCS_CTRL_EXIT_LOAD_HOST_PAT_MSR. */
3372
3373	if ( pVM->hm.s.vmx.fUsePreemptTimer
3374	&& (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER))
3375	val \|= VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER;
3376
3377	if ((val & zap) != val)
3378	{
3379	LogRel(("hmR0VmxSetupProcCtls: Invalid VM-exit controls combo! cpu=%RX64 val=%RX64 zap=%RX64\n",
3380	pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0, val, zap));
3381	pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT;
3382	return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3383	}
3384
3385	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, val);
3386	AssertRCReturn(rc, rc);
3387
3388	pVCpu->hm.s.vmx.u32ExitCtls = val;
3389	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_EXIT_CTLS);
3390	}
3391	return rc;
3392	}
3393
3394
3395	/**
3396	* Sets the TPR threshold in the VMCS.
3397	*
3398	* @returns VBox status code.
3399	* @param pVCpu The cross context virtual CPU structure.
3400	* @param u32TprThreshold The TPR threshold (task-priority class only).
3401	*/
3402	DECLINLINE(int) hmR0VmxApicSetTprThreshold(PVMCPU pVCpu, uint32_t u32TprThreshold)
3403	{
3404	Assert(!(u32TprThreshold & 0xfffffff0)); /* Bits 31:4 MBZ. */
3405	Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW); RT_NOREF_PV(pVCpu);
3406	return VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
3407	}
3408
3409
3410	/**
3411	* Loads the guest APIC and related state.
3412	*
3413	* @returns VBox status code.
3414	* @param pVCpu The cross context virtual CPU structure.
3415	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3416	* out-of-sync. Make sure to update the required fields
3417	* before using them.
3418	*
3419	* @remarks No-long-jump zone!!!
3420	*/
3421	DECLINLINE(int) hmR0VmxLoadGuestApicState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3422	{
3423	NOREF(pMixedCtx);
3424
3425	int rc = VINF_SUCCESS;
3426	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE))
3427	{
3428	if ( PDMHasApic(pVCpu->CTX_SUFF(pVM))
3429	&& APICIsEnabled(pVCpu))
3430	{
3431	/*
3432	* Setup TPR shadowing.
3433	*/
3434	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
3435	{
3436	Assert(pVCpu->hm.s.vmx.HCPhysVirtApic);
3437
3438	bool fPendingIntr = false;
3439	uint8_t u8Tpr = 0;
3440	uint8_t u8PendingIntr = 0;
3441	rc = APICGetTpr(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr);
3442	AssertRCReturn(rc, rc);
3443
3444	/*
3445	* If there are interrupts pending but masked by the TPR, instruct VT-x to cause a TPR-below-threshold VM-exit
3446	* when the guest lowers its TPR below the priority of the pending interrupt so we can deliver the interrupt.
3447	* If there are no interrupts pending, set threshold to 0 to not cause any TPR-below-threshold VM-exits.
3448	*/
3449	pVCpu->hm.s.vmx.pbVirtApic[XAPIC_OFF_TPR] = u8Tpr;
3450	uint32_t u32TprThreshold = 0;
3451	if (fPendingIntr)
3452	{
3453	/* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR (which is the Task-Priority Class). */
3454	const uint8_t u8PendingPriority = u8PendingIntr >> 4;
3455	const uint8_t u8TprPriority = u8Tpr >> 4;
3456	if (u8PendingPriority <= u8TprPriority)
3457	u32TprThreshold = u8PendingPriority;
3458	}
3459
3460	rc = hmR0VmxApicSetTprThreshold(pVCpu, u32TprThreshold);
3461	AssertRCReturn(rc, rc);
3462	}
3463	}
3464	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
3465	}
3466
3467	return rc;
3468	}
3469
3470
3471	/**
3472	* Gets the guest's interruptibility-state ("interrupt shadow" as AMD calls it).
3473	*
3474	* @returns Guest's interruptibility-state.
3475	* @param pVCpu The cross context virtual CPU structure.
3476	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3477	* out-of-sync. Make sure to update the required fields
3478	* before using them.
3479	*
3480	* @remarks No-long-jump zone!!!
3481	*/
3482	DECLINLINE(uint32_t) hmR0VmxGetGuestIntrState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3483	{
3484	/*
3485	* Check if we should inhibit interrupt delivery due to instructions like STI and MOV SS.
3486	*/
3487	uint32_t uIntrState = 0;
3488	if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
3489	{
3490	/* If inhibition is active, RIP & RFLAGS should've been accessed (i.e. read previously from the VMCS or from ring-3). */
3491	AssertMsg(HMVMXCPU_GST_IS_SET(pVCpu, HMVMX_UPDATED_GUEST_RIP \| HMVMX_UPDATED_GUEST_RFLAGS),
3492	("%#x\n", HMVMXCPU_GST_VALUE(pVCpu)));
3493	if (pMixedCtx->rip == EMGetInhibitInterruptsPC(pVCpu))
3494	{
3495	if (pMixedCtx->eflags.Bits.u1IF)
3496	uIntrState = VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI;
3497	else
3498	uIntrState = VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS;
3499	}
3500	else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
3501	{
3502	/*
3503	* We can clear the inhibit force flag as even if we go back to the recompiler without executing guest code in
3504	* VT-x, the flag's condition to be cleared is met and thus the cleared state is correct.
3505	*/
3506	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
3507	}
3508	}
3509
3510	/*
3511	* NMIs to the guest are blocked after an NMI is injected until the guest executes an IRET. We only
3512	* bother with virtual-NMI blocking when we have support for virtual NMIs in the CPU, otherwise
3513	* setting this would block host-NMIs and IRET will not clear the blocking.
3514	*
3515	* See Intel spec. 26.6.1 "Interruptibility state". See @bugref{7445}.
3516	*/
3517	if ( VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS)
3518	&& (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI))
3519	{
3520	uIntrState \|= VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI;
3521	}
3522
3523	return uIntrState;
3524	}
3525
3526
3527	/**
3528	* Loads the guest's interruptibility-state into the guest-state area in the
3529	* VMCS.
3530	*
3531	* @returns VBox status code.
3532	* @param pVCpu The cross context virtual CPU structure.
3533	* @param uIntrState The interruptibility-state to set.
3534	*/
3535	static int hmR0VmxLoadGuestIntrState(PVMCPU pVCpu, uint32_t uIntrState)
3536	{
3537	NOREF(pVCpu);
3538	AssertMsg(!(uIntrState & 0xfffffff0), ("%#x\n", uIntrState)); /* Bits 31:4 MBZ. */
3539	Assert((uIntrState & 0x3) != 0x3); /* Block-by-STI and MOV SS cannot be simultaneously set. */
3540	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, uIntrState);
3541	AssertRC(rc);
3542	return rc;
3543	}
3544
3545
3546	/**
3547	* Loads the exception intercepts required for guest execution in the VMCS.
3548	*
3549	* @returns VBox status code.
3550	* @param pVCpu The cross context virtual CPU structure.
3551	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3552	* out-of-sync. Make sure to update the required fields
3553	* before using them.
3554	*/
3555	static int hmR0VmxLoadGuestXcptIntercepts(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3556	{
3557	NOREF(pMixedCtx);
3558	int rc = VINF_SUCCESS;
3559	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS))
3560	{
3561	/* The remaining exception intercepts are handled elsewhere, e.g. in hmR0VmxLoadSharedCR0(). */
3562	if (pVCpu->hm.s.fGIMTrapXcptUD)
3563	pVCpu->hm.s.vmx.u32XcptBitmap \|= RT_BIT(X86_XCPT_UD);
3564	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
3565	else
3566	pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_UD);
3567	#endif
3568
3569	Assert(pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT_32(X86_XCPT_AC));
3570	Assert(pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT_32(X86_XCPT_DB));
3571
3572	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
3573	AssertRCReturn(rc, rc);
3574
3575	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
3576	Log4(("Load[%RU32]: VMX_VMCS32_CTRL_EXCEPTION_BITMAP=%#RX64 fContextUseFlags=%#RX32\n", pVCpu->idCpu,
3577	pVCpu->hm.s.vmx.u32XcptBitmap, HMCPU_CF_VALUE(pVCpu)));
3578	}
3579	return rc;
3580	}
3581
3582
3583	/**
3584	* Loads the guest's RIP into the guest-state area in the VMCS.
3585	*
3586	* @returns VBox status code.
3587	* @param pVCpu The cross context virtual CPU structure.
3588	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3589	* out-of-sync. Make sure to update the required fields
3590	* before using them.
3591	*
3592	* @remarks No-long-jump zone!!!
3593	*/
3594	static int hmR0VmxLoadGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3595	{
3596	int rc = VINF_SUCCESS;
3597	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RIP))
3598	{
3599	rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RIP, pMixedCtx->rip);
3600	AssertRCReturn(rc, rc);
3601
3602	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RIP);
3603	Log4(("Load[%RU32]: VMX_VMCS_GUEST_RIP=%#RX64 fContextUseFlags=%#RX32\n", pVCpu->idCpu, pMixedCtx->rip,
3604	HMCPU_CF_VALUE(pVCpu)));
3605	}
3606	return rc;
3607	}
3608
3609
3610	/**
3611	* Loads the guest's RSP into the guest-state area in the VMCS.
3612	*
3613	* @returns VBox status code.
3614	* @param pVCpu The cross context virtual CPU structure.
3615	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3616	* out-of-sync. Make sure to update the required fields
3617	* before using them.
3618	*
3619	* @remarks No-long-jump zone!!!
3620	*/
3621	static int hmR0VmxLoadGuestRsp(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3622	{
3623	int rc = VINF_SUCCESS;
3624	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RSP))
3625	{
3626	rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RSP, pMixedCtx->rsp);
3627	AssertRCReturn(rc, rc);
3628
3629	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RSP);
3630	Log4(("Load[%RU32]: VMX_VMCS_GUEST_RSP=%#RX64\n", pVCpu->idCpu, pMixedCtx->rsp));
3631	}
3632	return rc;
3633	}
3634
3635
3636	/**
3637	* Loads the guest's RFLAGS into the guest-state area in the VMCS.
3638	*
3639	* @returns VBox status code.
3640	* @param pVCpu The cross context virtual CPU structure.
3641	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3642	* out-of-sync. Make sure to update the required fields
3643	* before using them.
3644	*
3645	* @remarks No-long-jump zone!!!
3646	*/
3647	static int hmR0VmxLoadGuestRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3648	{
3649	int rc = VINF_SUCCESS;
3650	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS))
3651	{
3652	/* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ).
3653	Let us assert it as such and use 32-bit VMWRITE. */
3654	Assert(!(pMixedCtx->rflags.u64 >> 32));
3655	X86EFLAGS Eflags = pMixedCtx->eflags;
3656	/** @todo r=bird: There shall be no need to OR in X86_EFL_1 here, nor
3657	* shall there be any reason for clearing bits 63:22, 15, 5 and 3.
3658	* These will never be cleared/set, unless some other part of the VMM
3659	* code is buggy - in which case we're better of finding and fixing
3660	* those bugs than hiding them. */
3661	Assert(Eflags.u32 & X86_EFL_RA1_MASK);
3662	Assert(!(Eflags.u32 & ~(X86_EFL_1 \| X86_EFL_LIVE_MASK)));
3663	Eflags.u32 &= VMX_EFLAGS_RESERVED_0; /* Bits 22-31, 15, 5 & 3 MBZ. */
3664	Eflags.u32 \|= VMX_EFLAGS_RESERVED_1; /* Bit 1 MB1. */
3665
3666	/*
3667	* If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so we can restore them on VM-exit.
3668	* Modify the real-mode guest's eflags so that VT-x can run the real-mode guest code under Virtual 8086 mode.
3669	*/
3670	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
3671	{
3672	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
3673	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
3674	pVCpu->hm.s.vmx.RealMode.Eflags.u32 = Eflags.u32; /* Save the original eflags of the real-mode guest. */
3675	Eflags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */
3676	Eflags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */
3677	}
3678
3679	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_RFLAGS, Eflags.u32);
3680	AssertRCReturn(rc, rc);
3681
3682	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RFLAGS);
3683	Log4(("Load[%RU32]: VMX_VMCS_GUEST_RFLAGS=%#RX32\n", pVCpu->idCpu, Eflags.u32));
3684	}
3685	return rc;
3686	}
3687
3688
3689	/**
3690	* Loads the guest RIP, RSP and RFLAGS into the guest-state area in the VMCS.
3691	*
3692	* @returns VBox status code.
3693	* @param pVCpu The cross context virtual CPU structure.
3694	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3695	* out-of-sync. Make sure to update the required fields
3696	* before using them.
3697	*
3698	* @remarks No-long-jump zone!!!
3699	*/
3700	DECLINLINE(int) hmR0VmxLoadGuestRipRspRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3701	{
3702	int rc = hmR0VmxLoadGuestRip(pVCpu, pMixedCtx);
3703	rc \|= hmR0VmxLoadGuestRsp(pVCpu, pMixedCtx);
3704	rc \|= hmR0VmxLoadGuestRflags(pVCpu, pMixedCtx);
3705	AssertRCReturn(rc, rc);
3706	return rc;
3707	}
3708
3709
3710	/**
3711	* Loads the guest CR0 control register into the guest-state area in the VMCS.
3712	* CR0 is partially shared with the host and we have to consider the FPU bits.
3713	*
3714	* @returns VBox status code.
3715	* @param pVCpu The cross context virtual CPU structure.
3716	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3717	* out-of-sync. Make sure to update the required fields
3718	* before using them.
3719	*
3720	* @remarks No-long-jump zone!!!
3721	*/
3722	static int hmR0VmxLoadSharedCR0(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3723	{
3724	/*
3725	* Guest CR0.
3726	* Guest FPU.
3727	*/
3728	int rc = VINF_SUCCESS;
3729	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0))
3730	{
3731	Assert(!(pMixedCtx->cr0 >> 32));
3732	uint32_t u32GuestCR0 = pMixedCtx->cr0;
3733	PVM pVM = pVCpu->CTX_SUFF(pVM);
3734
3735	/* The guest's view (read access) of its CR0 is unblemished. */
3736	rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, u32GuestCR0);
3737	AssertRCReturn(rc, rc);
3738	Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR0_READ_SHADOW=%#RX32\n", pVCpu->idCpu, u32GuestCR0));
3739
3740	/* Setup VT-x's view of the guest CR0. */
3741	/* Minimize VM-exits due to CR3 changes when we have NestedPaging. */
3742	if (pVM->hm.s.fNestedPaging)
3743	{
3744	if (CPUMIsGuestPagingEnabledEx(pMixedCtx))
3745	{
3746	/* The guest has paging enabled, let it access CR3 without causing a VM-exit if supported. */
3747	pVCpu->hm.s.vmx.u32ProcCtls &= ~( VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
3748	\| VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT);
3749	}
3750	else
3751	{
3752	/* The guest doesn't have paging enabled, make CR3 access cause a VM-exit to update our shadow. */
3753	pVCpu->hm.s.vmx.u32ProcCtls \|= VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
3754	\| VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
3755	}
3756
3757	/* If we have unrestricted guest execution, we never have to intercept CR3 reads. */
3758	if (pVM->hm.s.vmx.fUnrestrictedGuest)
3759	pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
3760
3761	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
3762	AssertRCReturn(rc, rc);
3763	}
3764	else
3765	u32GuestCR0 \|= X86_CR0_WP; /* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */
3766
3767	/*
3768	* Guest FPU bits.
3769	* Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be set on the first
3770	* CPUs to support VT-x and no mention of with regards to UX in VM-entry checks.
3771	*/
3772	u32GuestCR0 \|= X86_CR0_NE;
3773	bool fInterceptNM = false;
3774	if (CPUMIsGuestFPUStateActive(pVCpu))
3775	{
3776	fInterceptNM = false; /* Guest FPU active, no need to VM-exit on #NM. */
3777	/* The guest should still get #NM exceptions when it expects it to, so we should not clear TS & MP bits here.
3778	We're only concerned about -us- not intercepting #NMs when the guest-FPU is active. Not the guest itself! */
3779	}
3780	else
3781	{
3782	fInterceptNM = true; /* Guest FPU inactive, VM-exit on #NM for lazy FPU loading. */
3783	u32GuestCR0 \|= X86_CR0_TS /* Guest can task switch quickly and do lazy FPU syncing. */
3784	\| X86_CR0_MP; /* FWAIT/WAIT should not ignore CR0.TS and should generate #NM. */
3785	}
3786
3787	/* Catch floating point exceptions if we need to report them to the guest in a different way. */
3788	bool fInterceptMF = false;
3789	if (!(pMixedCtx->cr0 & X86_CR0_NE))
3790	fInterceptMF = true;
3791
3792	/* Finally, intercept all exceptions as we cannot directly inject them in real-mode, see hmR0VmxInjectEventVmcs(). */
3793	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
3794	{
3795	Assert(PDMVmmDevHeapIsEnabled(pVM));
3796	Assert(pVM->hm.s.vmx.pRealModeTSS);
3797	pVCpu->hm.s.vmx.u32XcptBitmap \|= HMVMX_REAL_MODE_XCPT_MASK;
3798	fInterceptNM = true;
3799	fInterceptMF = true;
3800	}
3801	else
3802	{
3803	/* For now, cleared here as mode-switches can happen outside HM/VT-x. See @bugref{7626#c11}. */
3804	pVCpu->hm.s.vmx.u32XcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK;
3805	}
3806	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
3807
3808	if (fInterceptNM)
3809	pVCpu->hm.s.vmx.u32XcptBitmap \|= RT_BIT(X86_XCPT_NM);
3810	else
3811	pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_NM);
3812
3813	if (fInterceptMF)
3814	pVCpu->hm.s.vmx.u32XcptBitmap \|= RT_BIT(X86_XCPT_MF);
3815	else
3816	pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_MF);
3817
3818	/* Additional intercepts for debugging, define these yourself explicitly. */
3819	#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
3820	pVCpu->hm.s.vmx.u32XcptBitmap \|= 0
3821	\| RT_BIT(X86_XCPT_BP)
3822	\| RT_BIT(X86_XCPT_DE)
3823	\| RT_BIT(X86_XCPT_NM)
3824	\| RT_BIT(X86_XCPT_TS)
3825	\| RT_BIT(X86_XCPT_UD)
3826	\| RT_BIT(X86_XCPT_NP)
3827	\| RT_BIT(X86_XCPT_SS)
3828	\| RT_BIT(X86_XCPT_GP)
3829	\| RT_BIT(X86_XCPT_PF)
3830	\| RT_BIT(X86_XCPT_MF)
3831	;
3832	#elif defined(HMVMX_ALWAYS_TRAP_PF)
3833	pVCpu->hm.s.vmx.u32XcptBitmap \|= RT_BIT(X86_XCPT_PF);
3834	#endif
3835
3836	Assert(pVM->hm.s.fNestedPaging \|\| (pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT(X86_XCPT_PF)));
3837
3838	/* Set/clear the CR0 specific bits along with their exceptions (PE, PG, CD, NW). */
3839	uint32_t uSetCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
3840	uint32_t uZapCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
3841	if (pVM->hm.s.vmx.fUnrestrictedGuest) /* Exceptions for unrestricted-guests for fixed CR0 bits (PE, PG). */
3842	uSetCR0 &= ~(X86_CR0_PE \| X86_CR0_PG);
3843	else
3844	Assert((uSetCR0 & (X86_CR0_PE \| X86_CR0_PG)) == (X86_CR0_PE \| X86_CR0_PG));
3845
3846	u32GuestCR0 \|= uSetCR0;
3847	u32GuestCR0 &= uZapCR0;
3848	u32GuestCR0 &= ~(X86_CR0_CD \| X86_CR0_NW); /* Always enable caching. */
3849
3850	/* Write VT-x's view of the guest CR0 into the VMCS. */
3851	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR0, u32GuestCR0);
3852	AssertRCReturn(rc, rc);
3853	Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR0=%#RX32 (uSetCR0=%#RX32 uZapCR0=%#RX32)\n", pVCpu->idCpu, u32GuestCR0, uSetCR0,
3854	uZapCR0));
3855
3856	/*
3857	* CR0 is shared between host and guest along with a CR0 read shadow. Therefore, certain bits must not be changed
3858	* by the guest because VT-x ignores saving/restoring them (namely CD, ET, NW) and for certain other bits
3859	* we want to be notified immediately of guest CR0 changes (e.g. PG to update our shadow page tables).
3860	*/
3861	uint32_t u32CR0Mask = 0;
3862	u32CR0Mask = X86_CR0_PE
3863	\| X86_CR0_NE
3864	\| X86_CR0_WP
3865	\| X86_CR0_PG
3866	\| X86_CR0_ET /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.ET */
3867	\| X86_CR0_CD /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.CD */
3868	\| X86_CR0_NW; /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.NW */
3869
3870	/** @todo Avoid intercepting CR0.PE with unrestricted guests. Fix PGM
3871	* enmGuestMode to be in-sync with the current mode. See @bugref{6398}
3872	* and @bugref{6944}. */
3873	#if 0
3874	if (pVM->hm.s.vmx.fUnrestrictedGuest)
3875	u32CR0Mask &= ~X86_CR0_PE;
3876	#endif
3877	if (pVM->hm.s.fNestedPaging)
3878	u32CR0Mask &= ~X86_CR0_WP;
3879
3880	/* If the guest FPU state is active, don't need to VM-exit on writes to FPU related bits in CR0. */
3881	if (fInterceptNM)
3882	{
3883	u32CR0Mask \|= X86_CR0_TS
3884	\| X86_CR0_MP;
3885	}
3886
3887	/* Write the CR0 mask into the VMCS and update the VCPU's copy of the current CR0 mask. */
3888	pVCpu->hm.s.vmx.u32CR0Mask = u32CR0Mask;
3889	rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_MASK, u32CR0Mask);
3890	AssertRCReturn(rc, rc);
3891	Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR0_MASK=%#RX32\n", pVCpu->idCpu, u32CR0Mask));
3892
3893	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR0);
3894	}
3895	return rc;
3896	}
3897
3898
3899	/**
3900	* Loads the guest control registers (CR3, CR4) into the guest-state area
3901	* in the VMCS.
3902	*
3903	* @returns VBox strict status code.
3904	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
3905	* without unrestricted guest access and the VMMDev is not presently
3906	* mapped (e.g. EFI32).
3907	*
3908	* @param pVCpu The cross context virtual CPU structure.
3909	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
3910	* out-of-sync. Make sure to update the required fields
3911	* before using them.
3912	*
3913	* @remarks No-long-jump zone!!!
3914	*/
3915	static VBOXSTRICTRC hmR0VmxLoadGuestCR3AndCR4(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3916	{
3917	int rc = VINF_SUCCESS;
3918	PVM pVM = pVCpu->CTX_SUFF(pVM);
3919
3920	/*
3921	* Guest CR2.
3922	* It's always loaded in the assembler code. Nothing to do here.
3923	*/
3924
3925	/*
3926	* Guest CR3.
3927	*/
3928	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR3))
3929	{
3930	RTGCPHYS GCPhysGuestCR3 = NIL_RTGCPHYS;
3931	if (pVM->hm.s.fNestedPaging)
3932	{
3933	pVCpu->hm.s.vmx.HCPhysEPTP = PGMGetHyperCR3(pVCpu);
3934
3935	/* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */
3936	Assert(pVCpu->hm.s.vmx.HCPhysEPTP);
3937	Assert(!(pVCpu->hm.s.vmx.HCPhysEPTP & UINT64_C(0xfff0000000000000)));
3938	Assert(!(pVCpu->hm.s.vmx.HCPhysEPTP & 0xfff));
3939
3940	/* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */
3941	pVCpu->hm.s.vmx.HCPhysEPTP \|= VMX_EPT_MEMTYPE_WB
3942	\| (VMX_EPT_PAGE_WALK_LENGTH_DEFAULT << VMX_EPT_PAGE_WALK_LENGTH_SHIFT);
3943
3944	/* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */
3945	AssertMsg( ((pVCpu->hm.s.vmx.HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */
3946	&& ((pVCpu->hm.s.vmx.HCPhysEPTP >> 7) & 0x1f) == 0, /* Bits 7:11 MBZ. */
3947	("EPTP %#RX64\n", pVCpu->hm.s.vmx.HCPhysEPTP));
3948	AssertMsg( !((pVCpu->hm.s.vmx.HCPhysEPTP >> 6) & 0x01) /* Bit 6 (EPT accessed & dirty bit). */
3949	\|\| (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EPT_ACCESS_DIRTY),
3950	("EPTP accessed/dirty bit not supported by CPU but set %#RX64\n", pVCpu->hm.s.vmx.HCPhysEPTP));
3951
3952	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVCpu->hm.s.vmx.HCPhysEPTP);
3953	AssertRCReturn(rc, rc);
3954	Log4(("Load[%RU32]: VMX_VMCS64_CTRL_EPTP_FULL=%#RX64\n", pVCpu->idCpu, pVCpu->hm.s.vmx.HCPhysEPTP));
3955
3956	if ( pVM->hm.s.vmx.fUnrestrictedGuest
3957	\|\| CPUMIsGuestPagingEnabledEx(pMixedCtx))
3958	{
3959	/* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */
3960	if (CPUMIsGuestInPAEModeEx(pMixedCtx))
3961	{
3962	rc = PGMGstGetPaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
3963	AssertRCReturn(rc, rc);
3964	rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pVCpu->hm.s.aPdpes[0].u);
3965	rc \|= VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pVCpu->hm.s.aPdpes[1].u);
3966	rc \|= VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pVCpu->hm.s.aPdpes[2].u);
3967	rc \|= VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pVCpu->hm.s.aPdpes[3].u);
3968	AssertRCReturn(rc, rc);
3969	}
3970
3971	/* The guest's view of its CR3 is unblemished with Nested Paging when the guest is using paging or we
3972	have Unrestricted Execution to handle the guest when it's not using paging. */
3973	GCPhysGuestCR3 = pMixedCtx->cr3;
3974	}
3975	else
3976	{
3977	/*
3978	* The guest is not using paging, but the CPU (VT-x) has to. While the guest thinks it accesses physical memory
3979	* directly, we use our identity-mapped page table to map guest-linear to guest-physical addresses.
3980	* EPT takes care of translating it to host-physical addresses.
3981	*/
3982	RTGCPHYS GCPhys;
3983	Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable);
3984
3985	/* We obtain it here every time as the guest could have relocated this PCI region. */
3986	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
3987	if (RT_SUCCESS(rc))
3988	{ /* likely */ }
3989	else if (rc == VERR_PDM_DEV_HEAP_R3_TO_GCPHYS)
3990	{
3991	Log4(("Load[%RU32]: VERR_PDM_DEV_HEAP_R3_TO_GCPHYS -> VINF_EM_RESCHEDULE_REM\n", pVCpu->idCpu));
3992	return VINF_EM_RESCHEDULE_REM; /* We cannot execute now, switch to REM/IEM till the guest maps in VMMDev. */
3993	}
3994	else
3995	AssertMsgFailedReturn(("%Rrc\n", rc), rc);
3996
3997	GCPhysGuestCR3 = GCPhys;
3998	}
3999
4000	Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR3=%#RGp (GstN)\n", pVCpu->idCpu, GCPhysGuestCR3));
4001	rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_CR3, GCPhysGuestCR3);
4002	}
4003	else
4004	{
4005	/* Non-nested paging case, just use the hypervisor's CR3. */
4006	RTHCPHYS HCPhysGuestCR3 = PGMGetHyperCR3(pVCpu);
4007
4008	Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR3=%#RHv (HstN)\n", pVCpu->idCpu, HCPhysGuestCR3));
4009	rc = VMXWriteVmcsHstN(VMX_VMCS_GUEST_CR3, HCPhysGuestCR3);
4010	}
4011	AssertRCReturn(rc, rc);
4012
4013	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR3);
4014	}
4015
4016	/*
4017	* Guest CR4.
4018	* ASSUMES this is done everytime we get in from ring-3! (XCR0)
4019	*/
4020	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR4))
4021	{
4022	Assert(!(pMixedCtx->cr4 >> 32));
4023	uint32_t u32GuestCR4 = pMixedCtx->cr4;
4024
4025	/* The guest's view of its CR4 is unblemished. */
4026	rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, u32GuestCR4);
4027	AssertRCReturn(rc, rc);
4028	Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR4_READ_SHADOW=%#RX32\n", pVCpu->idCpu, u32GuestCR4));
4029
4030	/* Setup VT-x's view of the guest CR4. */
4031	/*
4032	* If we're emulating real-mode using virtual-8086 mode, we want to redirect software interrupts to the 8086 program
4033	* interrupt handler. Clear the VME bit (the interrupt redirection bitmap is already all 0, see hmR3InitFinalizeR0())
4034	* See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode".
4035	*/
4036	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4037	{
4038	Assert(pVM->hm.s.vmx.pRealModeTSS);
4039	Assert(PDMVmmDevHeapIsEnabled(pVM));
4040	u32GuestCR4 &= ~X86_CR4_VME;
4041	}
4042
4043	if (pVM->hm.s.fNestedPaging)
4044	{
4045	if ( !CPUMIsGuestPagingEnabledEx(pMixedCtx)
4046	&& !pVM->hm.s.vmx.fUnrestrictedGuest)
4047	{
4048	/* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */
4049	u32GuestCR4 \|= X86_CR4_PSE;
4050	/* Our identity mapping is a 32-bit page directory. */
4051	u32GuestCR4 &= ~X86_CR4_PAE;
4052	}
4053	/* else use guest CR4.*/
4054	}
4055	else
4056	{
4057	/*
4058	* The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host
4059	* paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables.
4060	*/
4061	switch (pVCpu->hm.s.enmShadowMode)
4062	{
4063	case PGMMODE_REAL: /* Real-mode. */
4064	case PGMMODE_PROTECTED: /* Protected mode without paging. */
4065	case PGMMODE_32_BIT: /* 32-bit paging. */
4066	{
4067	u32GuestCR4 &= ~X86_CR4_PAE;
4068	break;
4069	}
4070
4071	case PGMMODE_PAE: /* PAE paging. */
4072	case PGMMODE_PAE_NX: /* PAE paging with NX. */
4073	{
4074	u32GuestCR4 \|= X86_CR4_PAE;
4075	break;
4076	}
4077
4078	case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
4079	case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
4080	#ifdef VBOX_ENABLE_64_BITS_GUESTS
4081	break;
4082	#endif
4083	default:
4084	AssertFailed();
4085	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
4086	}
4087	}
4088
4089	/* We need to set and clear the CR4 specific bits here (mainly the X86_CR4_VMXE bit). */
4090	uint64_t uSetCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
4091	uint64_t uZapCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
4092	u32GuestCR4 \|= uSetCR4;
4093	u32GuestCR4 &= uZapCR4;
4094
4095	/* Write VT-x's view of the guest CR4 into the VMCS. */
4096	Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR4=%#RX32 (Set=%#RX32 Zap=%#RX32)\n", pVCpu->idCpu, u32GuestCR4, uSetCR4, uZapCR4));
4097	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR4, u32GuestCR4);
4098	AssertRCReturn(rc, rc);
4099
4100	/* Setup CR4 mask. CR4 flags owned by the host, if the guest attempts to change them, that would cause a VM-exit. */
4101	uint32_t u32CR4Mask = X86_CR4_VME
4102	\| X86_CR4_PAE
4103	\| X86_CR4_PGE
4104	\| X86_CR4_PSE
4105	\| X86_CR4_VMXE;
4106	if (pVM->cpum.ro.HostFeatures.fXSaveRstor)
4107	u32CR4Mask \|= X86_CR4_OSXSAVE;
4108	pVCpu->hm.s.vmx.u32CR4Mask = u32CR4Mask;
4109	rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_MASK, u32CR4Mask);
4110	AssertRCReturn(rc, rc);
4111
4112	/* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
4113	pVCpu->hm.s.fLoadSaveGuestXcr0 = (pMixedCtx->cr4 & X86_CR4_OSXSAVE) && pMixedCtx->aXcr[0] != ASMGetXcr0();
4114
4115	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR4);
4116	}
4117	return rc;
4118	}
4119
4120
4121	/**
4122	* Loads the guest debug registers into the guest-state area in the VMCS.
4123	*
4124	* This also sets up whether \#DB and MOV DRx accesses cause VM-exits.
4125	*
4126	* The guest debug bits are partially shared with the host (e.g. DR6, DR0-3).
4127	*
4128	* @returns VBox status code.
4129	* @param pVCpu The cross context virtual CPU structure.
4130	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
4131	* out-of-sync. Make sure to update the required fields
4132	* before using them.
4133	*
4134	* @remarks No-long-jump zone!!!
4135	*/
4136	static int hmR0VmxLoadSharedDebugState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4137	{
4138	if (!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG))
4139	return VINF_SUCCESS;
4140
4141	#ifdef VBOX_STRICT
4142	/* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */
4143	if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
4144	{
4145	/* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */
4146	Assert((pMixedCtx->dr[7] & (X86_DR7_MBZ_MASK \| X86_DR7_RAZ_MASK)) == 0); /* Bits 63:32, 15, 14, 12, 11 are reserved. */
4147	Assert((pMixedCtx->dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK); /* Bit 10 is reserved (RA1). */
4148	}
4149	#endif
4150
4151	int rc;
4152	PVM pVM = pVCpu->CTX_SUFF(pVM);
4153	bool fSteppingDB = false;
4154	bool fInterceptMovDRx = false;
4155	if (pVCpu->hm.s.fSingleInstruction)
4156	{
4157	/* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */
4158	if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG)
4159	{
4160	pVCpu->hm.s.vmx.u32ProcCtls \|= VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG;
4161	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
4162	AssertRCReturn(rc, rc);
4163	Assert(fSteppingDB == false);
4164	}
4165	else
4166	{
4167	pMixedCtx->eflags.u32 \|= X86_EFL_TF;
4168	pVCpu->hm.s.fClearTrapFlag = true;
4169	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
4170	fSteppingDB = true;
4171	}
4172	}
4173
4174	if ( fSteppingDB
4175	\|\| (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
4176	{
4177	/*
4178	* Use the combined guest and host DRx values found in the hypervisor
4179	* register set because the debugger has breakpoints active or someone
4180	* is single stepping on the host side without a monitor trap flag.
4181	*
4182	* Note! DBGF expects a clean DR6 state before executing guest code.
4183	*/
4184	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
4185	if ( CPUMIsGuestInLongModeEx(pMixedCtx)
4186	&& !CPUMIsHyperDebugStateActivePending(pVCpu))
4187	{
4188	CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
4189	Assert(CPUMIsHyperDebugStateActivePending(pVCpu));
4190	Assert(!CPUMIsGuestDebugStateActivePending(pVCpu));
4191	}
4192	else
4193	#endif
4194	if (!CPUMIsHyperDebugStateActive(pVCpu))
4195	{
4196	CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
4197	Assert(CPUMIsHyperDebugStateActive(pVCpu));
4198	Assert(!CPUMIsGuestDebugStateActive(pVCpu));
4199	}
4200
4201	/* Update DR7. (The other DRx values are handled by CPUM one way or the other.) */
4202	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)CPUMGetHyperDR7(pVCpu));
4203	AssertRCReturn(rc, rc);
4204
4205	pVCpu->hm.s.fUsingHyperDR7 = true;
4206	fInterceptMovDRx = true;
4207	}
4208	else
4209	{
4210	/*
4211	* If the guest has enabled debug registers, we need to load them prior to
4212	* executing guest code so they'll trigger at the right time.
4213	*/
4214	if (pMixedCtx->dr[7] & (X86_DR7_ENABLED_MASK \| X86_DR7_GD)) /** @todo Why GD? */
4215	{
4216	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
4217	if ( CPUMIsGuestInLongModeEx(pMixedCtx)
4218	&& !CPUMIsGuestDebugStateActivePending(pVCpu))
4219	{
4220	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
4221	Assert(CPUMIsGuestDebugStateActivePending(pVCpu));
4222	Assert(!CPUMIsHyperDebugStateActivePending(pVCpu));
4223	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
4224	}
4225	else
4226	#endif
4227	if (!CPUMIsGuestDebugStateActive(pVCpu))
4228	{
4229	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
4230	Assert(CPUMIsGuestDebugStateActive(pVCpu));
4231	Assert(!CPUMIsHyperDebugStateActive(pVCpu));
4232	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
4233	}
4234	Assert(!fInterceptMovDRx);
4235	}
4236	/*
4237	* If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we
4238	* must intercept #DB in order to maintain a correct DR6 guest value, and
4239	* because we need to intercept it to prevent nested #DBs from hanging the
4240	* CPU, we end up always having to intercept it. See hmR0VmxInitXcptBitmap.
4241	*/
4242	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
4243	else if ( !CPUMIsGuestDebugStateActivePending(pVCpu)
4244	&& !CPUMIsGuestDebugStateActive(pVCpu))
4245	#else
4246	else if (!CPUMIsGuestDebugStateActive(pVCpu))
4247	#endif
4248	{
4249	fInterceptMovDRx = true;
4250	}
4251
4252	/* Update guest DR7. */
4253	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, pMixedCtx->dr[7]);
4254	AssertRCReturn(rc, rc);
4255
4256	pVCpu->hm.s.fUsingHyperDR7 = false;
4257	}
4258
4259	/*
4260	* Update the processor-based VM-execution controls regarding intercepting MOV DRx instructions.
4261	*/
4262	if (fInterceptMovDRx)
4263	pVCpu->hm.s.vmx.u32ProcCtls \|= VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
4264	else
4265	pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
4266	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
4267	AssertRCReturn(rc, rc);
4268
4269	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_DEBUG);
4270	return VINF_SUCCESS;
4271	}
4272
4273
4274	#ifdef VBOX_STRICT
4275	/**
4276	* Strict function to validate segment registers.
4277	*
4278	* @remarks ASSUMES CR0 is up to date.
4279	*/
4280	static void hmR0VmxValidateSegmentRegs(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
4281	{
4282	/* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers". */
4283	/* NOTE: The reason we check for attribute value 0 and not just the unusable bit here is because hmR0VmxWriteSegmentReg()
4284	* only updates the VMCS' copy of the value with the unusable bit and doesn't change the guest-context value. */
4285	if ( !pVM->hm.s.vmx.fUnrestrictedGuest
4286	&& ( !CPUMIsGuestInRealModeEx(pCtx)
4287	&& !CPUMIsGuestInV86ModeEx(pCtx)))
4288	{
4289	/* Protected mode checks */
4290	/* CS */
4291	Assert(pCtx->cs.Attr.n.u1Present);
4292	Assert(!(pCtx->cs.Attr.u & 0xf00));
4293	Assert(!(pCtx->cs.Attr.u & 0xfffe0000));
4294	Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff
4295	\|\| !(pCtx->cs.Attr.n.u1Granularity));
4296	Assert( !(pCtx->cs.u32Limit & 0xfff00000)
4297	\|\| (pCtx->cs.Attr.n.u1Granularity));
4298	/* CS cannot be loaded with NULL in protected mode. */
4299	Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS? */
4300	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
4301	Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl);
4302	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
4303	Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl);
4304	else
4305	AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl));
4306	/* SS */
4307	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
4308	Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL));
4309	if ( !(pCtx->cr0 & X86_CR0_PE)
4310	\|\| pCtx->cs.Attr.n.u4Type == 3)
4311	{
4312	Assert(!pCtx->ss.Attr.n.u2Dpl);
4313	}
4314	if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
4315	{
4316	Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
4317	Assert(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7);
4318	Assert(pCtx->ss.Attr.n.u1Present);
4319	Assert(!(pCtx->ss.Attr.u & 0xf00));
4320	Assert(!(pCtx->ss.Attr.u & 0xfffe0000));
4321	Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff
4322	\|\| !(pCtx->ss.Attr.n.u1Granularity));
4323	Assert( !(pCtx->ss.u32Limit & 0xfff00000)
4324	\|\| (pCtx->ss.Attr.n.u1Granularity));
4325	}
4326	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxWriteSegmentReg(). */
4327	if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
4328	{
4329	Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4330	Assert(pCtx->ds.Attr.n.u1Present);
4331	Assert(pCtx->ds.Attr.n.u4Type > 11 \|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL));
4332	Assert(!(pCtx->ds.Attr.u & 0xf00));
4333	Assert(!(pCtx->ds.Attr.u & 0xfffe0000));
4334	Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff
4335	\|\| !(pCtx->ds.Attr.n.u1Granularity));
4336	Assert( !(pCtx->ds.u32Limit & 0xfff00000)
4337	\|\| (pCtx->ds.Attr.n.u1Granularity));
4338	Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4339	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ));
4340	}
4341	if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
4342	{
4343	Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4344	Assert(pCtx->es.Attr.n.u1Present);
4345	Assert(pCtx->es.Attr.n.u4Type > 11 \|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL));
4346	Assert(!(pCtx->es.Attr.u & 0xf00));
4347	Assert(!(pCtx->es.Attr.u & 0xfffe0000));
4348	Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff
4349	\|\| !(pCtx->es.Attr.n.u1Granularity));
4350	Assert( !(pCtx->es.u32Limit & 0xfff00000)
4351	\|\| (pCtx->es.Attr.n.u1Granularity));
4352	Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4353	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ));
4354	}
4355	if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
4356	{
4357	Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4358	Assert(pCtx->fs.Attr.n.u1Present);
4359	Assert(pCtx->fs.Attr.n.u4Type > 11 \|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL));
4360	Assert(!(pCtx->fs.Attr.u & 0xf00));
4361	Assert(!(pCtx->fs.Attr.u & 0xfffe0000));
4362	Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff
4363	\|\| !(pCtx->fs.Attr.n.u1Granularity));
4364	Assert( !(pCtx->fs.u32Limit & 0xfff00000)
4365	\|\| (pCtx->fs.Attr.n.u1Granularity));
4366	Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4367	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ));
4368	}
4369	if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
4370	{
4371	Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4372	Assert(pCtx->gs.Attr.n.u1Present);
4373	Assert(pCtx->gs.Attr.n.u4Type > 11 \|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL));
4374	Assert(!(pCtx->gs.Attr.u & 0xf00));
4375	Assert(!(pCtx->gs.Attr.u & 0xfffe0000));
4376	Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff
4377	\|\| !(pCtx->gs.Attr.n.u1Granularity));
4378	Assert( !(pCtx->gs.u32Limit & 0xfff00000)
4379	\|\| (pCtx->gs.Attr.n.u1Granularity));
4380	Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4381	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ));
4382	}
4383	/* 64-bit capable CPUs. */
4384	# if HC_ARCH_BITS == 64
4385	Assert(!(pCtx->cs.u64Base >> 32));
4386	Assert(!pCtx->ss.Attr.u \|\| !(pCtx->ss.u64Base >> 32));
4387	Assert(!pCtx->ds.Attr.u \|\| !(pCtx->ds.u64Base >> 32));
4388	Assert(!pCtx->es.Attr.u \|\| !(pCtx->es.u64Base >> 32));
4389	# endif
4390	}
4391	else if ( CPUMIsGuestInV86ModeEx(pCtx)
4392	\|\| ( CPUMIsGuestInRealModeEx(pCtx)
4393	&& !pVM->hm.s.vmx.fUnrestrictedGuest))
4394	{
4395	/* Real and v86 mode checks. */
4396	/* hmR0VmxWriteSegmentReg() writes the modified in VMCS. We want what we're feeding to VT-x. */
4397	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
4398	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4399	{
4400	u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3; u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3;
4401	}
4402	else
4403	{
4404	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u;
4405	u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
4406	}
4407
4408	/* CS */
4409	AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel));
4410	Assert(pCtx->cs.u32Limit == 0xffff);
4411	Assert(u32CSAttr == 0xf3);
4412	/* SS */
4413	Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4);
4414	Assert(pCtx->ss.u32Limit == 0xffff);
4415	Assert(u32SSAttr == 0xf3);
4416	/* DS */
4417	Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4);
4418	Assert(pCtx->ds.u32Limit == 0xffff);
4419	Assert(u32DSAttr == 0xf3);
4420	/* ES */
4421	Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4);
4422	Assert(pCtx->es.u32Limit == 0xffff);
4423	Assert(u32ESAttr == 0xf3);
4424	/* FS */
4425	Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4);
4426	Assert(pCtx->fs.u32Limit == 0xffff);
4427	Assert(u32FSAttr == 0xf3);
4428	/* GS */
4429	Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4);
4430	Assert(pCtx->gs.u32Limit == 0xffff);
4431	Assert(u32GSAttr == 0xf3);
4432	/* 64-bit capable CPUs. */
4433	# if HC_ARCH_BITS == 64
4434	Assert(!(pCtx->cs.u64Base >> 32));
4435	Assert(!u32SSAttr \|\| !(pCtx->ss.u64Base >> 32));
4436	Assert(!u32DSAttr \|\| !(pCtx->ds.u64Base >> 32));
4437	Assert(!u32ESAttr \|\| !(pCtx->es.u64Base >> 32));
4438	# endif
4439	}
4440	}
4441	#endif /* VBOX_STRICT */
4442
4443
4444	/**
4445	* Writes a guest segment register into the guest-state area in the VMCS.
4446	*
4447	* @returns VBox status code.
4448	* @param pVCpu The cross context virtual CPU structure.
4449	* @param idxSel Index of the selector in the VMCS.
4450	* @param idxLimit Index of the segment limit in the VMCS.
4451	* @param idxBase Index of the segment base in the VMCS.
4452	* @param idxAccess Index of the access rights of the segment in the VMCS.
4453	* @param pSelReg Pointer to the segment selector.
4454	*
4455	* @remarks No-long-jump zone!!!
4456	*/
4457	static int hmR0VmxWriteSegmentReg(PVMCPU pVCpu, uint32_t idxSel, uint32_t idxLimit, uint32_t idxBase,
4458	uint32_t idxAccess, PCPUMSELREG pSelReg)
4459	{
4460	int rc = VMXWriteVmcs32(idxSel, pSelReg->Sel); /* 16-bit guest selector field. */
4461	rc \|= VMXWriteVmcs32(idxLimit, pSelReg->u32Limit); /* 32-bit guest segment limit field. */
4462	rc \|= VMXWriteVmcsGstN(idxBase, pSelReg->u64Base); /* Natural width guest segment base field.*/
4463	AssertRCReturn(rc, rc);
4464
4465	uint32_t u32Access = pSelReg->Attr.u;
4466	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4467	{
4468	/* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */
4469	u32Access = 0xf3;
4470	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
4471	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
4472	}
4473	else
4474	{
4475	/*
4476	* The way to differentiate between whether this is really a null selector or was just a selector loaded with 0 in
4477	* real-mode is using the segment attributes. A selector loaded in real-mode with the value 0 is valid and usable in
4478	* protected-mode and we should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures NULL selectors
4479	* loaded in protected-mode have their attribute as 0.
4480	*/
4481	if (!u32Access)
4482	u32Access = X86DESCATTR_UNUSABLE;
4483	}
4484
4485	/* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */
4486	AssertMsg((u32Access & X86DESCATTR_UNUSABLE) \|\| (u32Access & X86_SEL_TYPE_ACCESSED),
4487	("Access bit not set for usable segment. idx=%#x sel=%#x attr %#x\n", idxBase, pSelReg, pSelReg->Attr.u));
4488
4489	rc = VMXWriteVmcs32(idxAccess, u32Access); /* 32-bit guest segment access-rights field. */
4490	AssertRCReturn(rc, rc);
4491	return rc;
4492	}
4493
4494
4495	/**
4496	* Loads the guest segment registers, GDTR, IDTR, LDTR, (TR, FS and GS bases)
4497	* into the guest-state area in the VMCS.
4498	*
4499	* @returns VBox status code.
4500	* @param pVCpu The cross context virtual CPU structure.
4501	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
4502	* out-of-sync. Make sure to update the required fields
4503	* before using them.
4504	*
4505	* @remarks ASSUMES pMixedCtx->cr0 is up to date (strict builds validation).
4506	* @remarks No-long-jump zone!!!
4507	*/
4508	static int hmR0VmxLoadGuestSegmentRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4509	{
4510	int rc = VERR_INTERNAL_ERROR_5;
4511	PVM pVM = pVCpu->CTX_SUFF(pVM);
4512
4513	/*
4514	* Guest Segment registers: CS, SS, DS, ES, FS, GS.
4515	*/
4516	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS))
4517	{
4518	/* Save the segment attributes for real-on-v86 mode hack, so we can restore them on VM-exit. */
4519	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4520	{
4521	pVCpu->hm.s.vmx.RealMode.AttrCS.u = pMixedCtx->cs.Attr.u;
4522	pVCpu->hm.s.vmx.RealMode.AttrSS.u = pMixedCtx->ss.Attr.u;
4523	pVCpu->hm.s.vmx.RealMode.AttrDS.u = pMixedCtx->ds.Attr.u;
4524	pVCpu->hm.s.vmx.RealMode.AttrES.u = pMixedCtx->es.Attr.u;
4525	pVCpu->hm.s.vmx.RealMode.AttrFS.u = pMixedCtx->fs.Attr.u;
4526	pVCpu->hm.s.vmx.RealMode.AttrGS.u = pMixedCtx->gs.Attr.u;
4527	}
4528
4529	#ifdef VBOX_WITH_REM
4530	if (!pVM->hm.s.vmx.fUnrestrictedGuest)
4531	{
4532	Assert(pVM->hm.s.vmx.pRealModeTSS);
4533	AssertCompile(PGMMODE_REAL < PGMMODE_PROTECTED);
4534	if ( pVCpu->hm.s.vmx.fWasInRealMode
4535	&& PGMGetGuestMode(pVCpu) >= PGMMODE_PROTECTED)
4536	{
4537	/* Signal that the recompiler must flush its code-cache as the guest -may- rewrite code it will later execute
4538	in real-mode (e.g. OpenBSD 4.0) */
4539	REMFlushTBs(pVM);
4540	Log4(("Load[%RU32]: Switch to protected mode detected!\n", pVCpu->idCpu));
4541	pVCpu->hm.s.vmx.fWasInRealMode = false;
4542	}
4543	}
4544	#endif
4545	rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_CS_SEL, VMX_VMCS32_GUEST_CS_LIMIT, VMX_VMCS_GUEST_CS_BASE,
4546	VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS, &pMixedCtx->cs);
4547	AssertRCReturn(rc, rc);
4548	rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_SS_SEL, VMX_VMCS32_GUEST_SS_LIMIT, VMX_VMCS_GUEST_SS_BASE,
4549	VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS, &pMixedCtx->ss);
4550	AssertRCReturn(rc, rc);
4551	rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_DS_SEL, VMX_VMCS32_GUEST_DS_LIMIT, VMX_VMCS_GUEST_DS_BASE,
4552	VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS, &pMixedCtx->ds);
4553	AssertRCReturn(rc, rc);
4554	rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_ES_SEL, VMX_VMCS32_GUEST_ES_LIMIT, VMX_VMCS_GUEST_ES_BASE,
4555	VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS, &pMixedCtx->es);
4556	AssertRCReturn(rc, rc);
4557	rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FS_SEL, VMX_VMCS32_GUEST_FS_LIMIT, VMX_VMCS_GUEST_FS_BASE,
4558	VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS, &pMixedCtx->fs);
4559	AssertRCReturn(rc, rc);
4560	rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_GS_SEL, VMX_VMCS32_GUEST_GS_LIMIT, VMX_VMCS_GUEST_GS_BASE,
4561	VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS, &pMixedCtx->gs);
4562	AssertRCReturn(rc, rc);
4563
4564	#ifdef VBOX_STRICT
4565	/* Validate. */
4566	hmR0VmxValidateSegmentRegs(pVM, pVCpu, pMixedCtx);
4567	#endif
4568
4569	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS);
4570	Log4(("Load[%RU32]: CS=%#RX16 Base=%#RX64 Limit=%#RX32 Attr=%#RX32\n", pVCpu->idCpu, pMixedCtx->cs.Sel,
4571	pMixedCtx->cs.u64Base, pMixedCtx->cs.u32Limit, pMixedCtx->cs.Attr.u));
4572	}
4573
4574	/*
4575	* Guest TR.
4576	*/
4577	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_TR))
4578	{
4579	/*
4580	* Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is achieved
4581	* using the interrupt redirection bitmap (all bits cleared to let the guest handle INT-n's) in the TSS.
4582	* See hmR3InitFinalizeR0() to see how pRealModeTSS is setup.
4583	*/
4584	uint16_t u16Sel = 0;
4585	uint32_t u32Limit = 0;
4586	uint64_t u64Base = 0;
4587	uint32_t u32AccessRights = 0;
4588
4589	if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4590	{
4591	u16Sel = pMixedCtx->tr.Sel;
4592	u32Limit = pMixedCtx->tr.u32Limit;
4593	u64Base = pMixedCtx->tr.u64Base;
4594	u32AccessRights = pMixedCtx->tr.Attr.u;
4595	}
4596	else
4597	{
4598	Assert(pVM->hm.s.vmx.pRealModeTSS);
4599	Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMR3CanExecuteGuest() -XXX- what about inner loop changes? */
4600
4601	/* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */
4602	RTGCPHYS GCPhys;
4603	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
4604	AssertRCReturn(rc, rc);
4605
4606	X86DESCATTR DescAttr;
4607	DescAttr.u = 0;
4608	DescAttr.n.u1Present = 1;
4609	DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
4610
4611	u16Sel = 0;
4612	u32Limit = HM_VTX_TSS_SIZE;
4613	u64Base = GCPhys; /* in real-mode phys = virt. */
4614	u32AccessRights = DescAttr.u;
4615	}
4616
4617	/* Validate. */
4618	Assert(!(u16Sel & RT_BIT(2)));
4619	AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY
4620	\|\| (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights));
4621	AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights));
4622	Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/
4623	Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/
4624	Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */
4625	Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */
4626	Assert( (u32Limit & 0xfff) == 0xfff
4627	\|\| !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */
4628	Assert( !(pMixedCtx->tr.u32Limit & 0xfff00000)
4629	\|\| (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */
4630
4631	rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_TR_SEL, u16Sel);
4632	rc \|= VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit);
4633	rc \|= VMXWriteVmcsGstN(VMX_VMCS_GUEST_TR_BASE, u64Base);
4634	rc \|= VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights);
4635	AssertRCReturn(rc, rc);
4636
4637	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_TR);
4638	Log4(("Load[%RU32]: VMX_VMCS_GUEST_TR_BASE=%#RX64\n", pVCpu->idCpu, u64Base));
4639	}
4640
4641	/*
4642	* Guest GDTR.
4643	*/
4644	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_GDTR))
4645	{
4646	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pMixedCtx->gdtr.cbGdt);
4647	rc \|= VMXWriteVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, pMixedCtx->gdtr.pGdt);
4648	AssertRCReturn(rc, rc);
4649
4650	/* Validate. */
4651	Assert(!(pMixedCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */
4652
4653	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_GDTR);
4654	Log4(("Load[%RU32]: VMX_VMCS_GUEST_GDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->gdtr.pGdt));
4655	}
4656
4657	/*
4658	* Guest LDTR.
4659	*/
4660	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LDTR))
4661	{
4662	/* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */
4663	uint32_t u32Access = 0;
4664	if (!pMixedCtx->ldtr.Attr.u)
4665	u32Access = X86DESCATTR_UNUSABLE;
4666	else
4667	u32Access = pMixedCtx->ldtr.Attr.u;
4668
4669	rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_LDTR_SEL, pMixedCtx->ldtr.Sel);
4670	rc \|= VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pMixedCtx->ldtr.u32Limit);
4671	rc \|= VMXWriteVmcsGstN(VMX_VMCS_GUEST_LDTR_BASE, pMixedCtx->ldtr.u64Base);
4672	rc \|= VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access);
4673	AssertRCReturn(rc, rc);
4674
4675	/* Validate. */
4676	if (!(u32Access & X86DESCATTR_UNUSABLE))
4677	{
4678	Assert(!(pMixedCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */
4679	Assert(pMixedCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */
4680	Assert(!pMixedCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */
4681	Assert(pMixedCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */
4682	Assert(!pMixedCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */
4683	Assert(!(pMixedCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */
4684	Assert( (pMixedCtx->ldtr.u32Limit & 0xfff) == 0xfff
4685	\|\| !pMixedCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */
4686	Assert( !(pMixedCtx->ldtr.u32Limit & 0xfff00000)
4687	\|\| pMixedCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */
4688	}
4689
4690	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_LDTR);
4691	Log4(("Load[%RU32]: VMX_VMCS_GUEST_LDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->ldtr.u64Base));
4692	}
4693
4694	/*
4695	* Guest IDTR.
4696	*/
4697	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_IDTR))
4698	{
4699	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pMixedCtx->idtr.cbIdt);
4700	rc \|= VMXWriteVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, pMixedCtx->idtr.pIdt);
4701	AssertRCReturn(rc, rc);
4702
4703	/* Validate. */
4704	Assert(!(pMixedCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */
4705
4706	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_IDTR);
4707	Log4(("Load[%RU32]: VMX_VMCS_GUEST_IDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->idtr.pIdt));
4708	}
4709
4710	return VINF_SUCCESS;
4711	}
4712
4713
4714	/**
4715	* Loads certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store
4716	* areas.
4717	*
4718	* These MSRs will automatically be loaded to the host CPU on every successful
4719	* VM-entry and stored from the host CPU on every successful VM-exit. This also
4720	* creates/updates MSR slots for the host MSRs. The actual host MSR values are
4721	* -not- updated here for performance reasons. See hmR0VmxSaveHostMsrs().
4722	*
4723	* Also loads the sysenter MSRs into the guest-state area in the VMCS.
4724	*
4725	* @returns VBox status code.
4726	* @param pVCpu The cross context virtual CPU structure.
4727	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
4728	* out-of-sync. Make sure to update the required fields
4729	* before using them.
4730	*
4731	* @remarks No-long-jump zone!!!
4732	*/
4733	static int hmR0VmxLoadGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4734	{
4735	AssertPtr(pVCpu);
4736	AssertPtr(pVCpu->hm.s.vmx.pvGuestMsr);
4737
4738	/*
4739	* MSRs that we use the auto-load/store MSR area in the VMCS.
4740	*/
4741	PVM pVM = pVCpu->CTX_SUFF(pVM);
4742	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS))
4743	{
4744	/* For 64-bit hosts, we load/restore them lazily, see hmR0VmxLazyLoadGuestMsrs(). */
4745	#if HC_ARCH_BITS == 32
4746	if (pVM->hm.s.fAllow64BitGuests)
4747	{
4748	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_LSTAR, pMixedCtx->msrLSTAR, false, NULL);
4749	rc \|= hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K6_STAR, pMixedCtx->msrSTAR, false, NULL);
4750	rc \|= hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_SF_MASK, pMixedCtx->msrSFMASK, false, NULL);
4751	rc \|= hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_KERNEL_GS_BASE, pMixedCtx->msrKERNELGSBASE, false, NULL);
4752	AssertRCReturn(rc, rc);
4753	# ifdef LOG_ENABLED
4754	PVMXAUTOMSR pMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
4755	for (uint32_t i = 0; i < pVCpu->hm.s.vmx.cMsrs; i++, pMsr++)
4756	{
4757	Log4(("Load[%RU32]: MSR[%RU32]: u32Msr=%#RX32 u64Value=%#RX64\n", pVCpu->idCpu, i, pMsr->u32Msr,
4758	pMsr->u64Value));
4759	}
4760	# endif
4761	}
4762	#endif
4763	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
4764	}
4765
4766	/*
4767	* Guest Sysenter MSRs.
4768	* These flags are only set when MSR-bitmaps are not supported by the CPU and we cause
4769	* VM-exits on WRMSRs for these MSRs.
4770	*/
4771	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR))
4772	{
4773	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pMixedCtx->SysEnter.cs); AssertRCReturn(rc, rc);
4774	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR);
4775	}
4776
4777	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR))
4778	{
4779	int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, pMixedCtx->SysEnter.eip); AssertRCReturn(rc, rc);
4780	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
4781	}
4782
4783	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR))
4784	{
4785	int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, pMixedCtx->SysEnter.esp); AssertRCReturn(rc, rc);
4786	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
4787	}
4788
4789	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_EFER_MSR))
4790	{
4791	if (hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
4792	{
4793	/*
4794	* If the CPU supports VMCS controls for swapping EFER, use it. Otherwise, we have no option
4795	* but to use the auto-load store MSR area in the VMCS for swapping EFER. See @bugref{7368}.
4796	*/
4797	if (pVM->hm.s.vmx.fSupportsVmcsEfer)
4798	{
4799	int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_EFER_FULL, pMixedCtx->msrEFER);
4800	AssertRCReturn(rc,rc);
4801	Log4(("Load[%RU32]: VMX_VMCS64_GUEST_EFER_FULL=%#RX64\n", pVCpu->idCpu, pMixedCtx->msrEFER));
4802	}
4803	else
4804	{
4805	int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K6_EFER, pMixedCtx->msrEFER, false /* fUpdateHostMsr */,
4806	NULL /* pfAddedAndUpdated */);
4807	AssertRCReturn(rc, rc);
4808
4809	/* We need to intercept reads too, see @bugref{7386#c16}. */
4810	if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
4811	hmR0VmxSetMsrPermission(pVCpu, MSR_K6_EFER, VMXMSREXIT_INTERCEPT_READ, VMXMSREXIT_INTERCEPT_WRITE);
4812	Log4(("Load[%RU32]: MSR[--]: u32Msr=%#RX32 u64Value=%#RX64 cMsrs=%u\n", pVCpu->idCpu, MSR_K6_EFER,
4813	pMixedCtx->msrEFER, pVCpu->hm.s.vmx.cMsrs));
4814	}
4815	}
4816	else if (!pVM->hm.s.vmx.fSupportsVmcsEfer)
4817	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, MSR_K6_EFER);
4818	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_EFER_MSR);
4819	}
4820
4821	return VINF_SUCCESS;
4822	}
4823
4824
4825	/**
4826	* Loads the guest activity state into the guest-state area in the VMCS.
4827	*
4828	* @returns VBox status code.
4829	* @param pVCpu The cross context virtual CPU structure.
4830	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
4831	* out-of-sync. Make sure to update the required fields
4832	* before using them.
4833	*
4834	* @remarks No-long-jump zone!!!
4835	*/
4836	static int hmR0VmxLoadGuestActivityState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4837	{
4838	NOREF(pMixedCtx);
4839	/** @todo See if we can make use of other states, e.g.
4840	* VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN or HLT. */
4841	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_ACTIVITY_STATE))
4842	{
4843	int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, VMX_VMCS_GUEST_ACTIVITY_ACTIVE);
4844	AssertRCReturn(rc, rc);
4845
4846	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_ACTIVITY_STATE);
4847	}
4848	return VINF_SUCCESS;
4849	}
4850
4851
4852	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
4853	/**
4854	* Check if guest state allows safe use of 32-bit switcher again.
4855	*
4856	* Segment bases and protected mode structures must be 32-bit addressable
4857	* because the 32-bit switcher will ignore high dword when writing these VMCS
4858	* fields. See @bugref{8432} for details.
4859	*
4860	* @returns true if safe, false if must continue to use the 64-bit switcher.
4861	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
4862	* out-of-sync. Make sure to update the required fields
4863	* before using them.
4864	*
4865	* @remarks No-long-jump zone!!!
4866	*/
4867	static bool hmR0VmxIs32BitSwitcherSafe(PCPUMCTX pMixedCtx)
4868	{
4869	if (pMixedCtx->gdtr.pGdt & UINT64_C(0xffffffff00000000))
4870	return false;
4871	if (pMixedCtx->idtr.pIdt & UINT64_C(0xffffffff00000000))
4872	return false;
4873	if (pMixedCtx->ldtr.u64Base & UINT64_C(0xffffffff00000000))
4874	return false;
4875	if (pMixedCtx->tr.u64Base & UINT64_C(0xffffffff00000000))
4876	return false;
4877	if (pMixedCtx->es.u64Base & UINT64_C(0xffffffff00000000))
4878	return false;
4879	if (pMixedCtx->cs.u64Base & UINT64_C(0xffffffff00000000))
4880	return false;
4881	if (pMixedCtx->ss.u64Base & UINT64_C(0xffffffff00000000))
4882	return false;
4883	if (pMixedCtx->ds.u64Base & UINT64_C(0xffffffff00000000))
4884	return false;
4885	if (pMixedCtx->fs.u64Base & UINT64_C(0xffffffff00000000))
4886	return false;
4887	if (pMixedCtx->gs.u64Base & UINT64_C(0xffffffff00000000))
4888	return false;
4889	/* All good, bases are 32-bit. */
4890	return true;
4891	}
4892	#endif
4893
4894
4895	/**
4896	* Sets up the appropriate function to run guest code.
4897	*
4898	* @returns VBox status code.
4899	* @param pVCpu The cross context virtual CPU structure.
4900	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
4901	* out-of-sync. Make sure to update the required fields
4902	* before using them.
4903	*
4904	* @remarks No-long-jump zone!!!
4905	*/
4906	static int hmR0VmxSetupVMRunHandler(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4907	{
4908	if (CPUMIsGuestInLongModeEx(pMixedCtx))
4909	{
4910	#ifndef VBOX_ENABLE_64_BITS_GUESTS
4911	return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
4912	#endif
4913	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests); /* Guaranteed by hmR3InitFinalizeR0(). */
4914	#if HC_ARCH_BITS == 32
4915	/* 32-bit host. We need to switch to 64-bit before running the 64-bit guest. */
4916	if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0SwitcherStartVM64)
4917	{
4918	if (pVCpu->hm.s.vmx.pfnStartVM != NULL) /* Very first entry would have saved host-state already, ignore it. */
4919	{
4920	/* Currently, all mode changes sends us back to ring-3, so these should be set. See @bugref{6944}. */
4921	AssertMsg(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_VMX_EXIT_CTLS
4922	\| HM_CHANGED_VMX_ENTRY_CTLS
4923	\| HM_CHANGED_GUEST_EFER_MSR), ("flags=%#x\n", HMCPU_CF_VALUE(pVCpu)));
4924	}
4925	pVCpu->hm.s.vmx.pfnStartVM = VMXR0SwitcherStartVM64;
4926
4927	/* Mark that we've switched to 64-bit handler, we can't safely switch back to 32-bit for
4928	the rest of the VM run (until VM reset). See @bugref{8432#c7}. */
4929	pVCpu->hm.s.vmx.fSwitchedTo64on32 = true;
4930	Log4(("Load[%RU32]: hmR0VmxSetupVMRunHandler: selected 64-bit switcher\n", pVCpu->idCpu));
4931	}
4932	#else
4933	/* 64-bit host. */
4934	pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM64;
4935	#endif
4936	}
4937	else
4938	{
4939	/* Guest is not in long mode, use the 32-bit handler. */
4940	#if HC_ARCH_BITS == 32
4941	if ( pVCpu->hm.s.vmx.pfnStartVM != VMXR0StartVM32
4942	&& !pVCpu->hm.s.vmx.fSwitchedTo64on32 /* If set, guest mode change does not imply switcher change. */
4943	&& pVCpu->hm.s.vmx.pfnStartVM != NULL) /* Very first entry would have saved host-state already, ignore it. */
4944	{
4945	/* Currently, all mode changes sends us back to ring-3, so these should be set. See @bugref{6944}. */
4946	AssertMsg(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_VMX_EXIT_CTLS
4947	\| HM_CHANGED_VMX_ENTRY_CTLS
4948	\| HM_CHANGED_GUEST_EFER_MSR), ("flags=%#x\n", HMCPU_CF_VALUE(pVCpu)));
4949	}
4950	# ifdef VBOX_ENABLE_64_BITS_GUESTS
4951	/*
4952	* Keep using the 64-bit switcher even though we're in 32-bit because of bad Intel design, see @bugref{8432#c7}.
4953	* If real-on-v86 mode is active, clear the 64-bit switcher flag because now we know the guest is in a sane
4954	* state where it's safe to use the 32-bit switcher. Otherwise check the guest state if it's safe to use
4955	* the much faster 32-bit switcher again.
4956	*/
4957	if (!pVCpu->hm.s.vmx.fSwitchedTo64on32)
4958	{
4959	if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0StartVM32)
4960	Log4(("Load[%RU32]: hmR0VmxSetupVMRunHandler: selected 32-bit switcher\n", pVCpu->idCpu));
4961	pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4962	}
4963	else
4964	{
4965	Assert(pVCpu->hm.s.vmx.pfnStartVM == VMXR0SwitcherStartVM64);
4966	if ( pVCpu->hm.s.vmx.RealMode.fRealOnV86Active
4967	\|\| hmR0VmxIs32BitSwitcherSafe(pMixedCtx))
4968	{
4969	pVCpu->hm.s.vmx.fSwitchedTo64on32 = false;
4970	pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4971	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_EFER_MSR
4972	\| HM_CHANGED_VMX_ENTRY_CTLS
4973	\| HM_CHANGED_VMX_EXIT_CTLS
4974	\| HM_CHANGED_HOST_CONTEXT);
4975	Log4(("Load[%RU32]: hmR0VmxSetupVMRunHandler: selected 32-bit switcher (safe)\n", pVCpu->idCpu));
4976	}
4977	}
4978	# else
4979	pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4980	# endif
4981	#else
4982	pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4983	#endif
4984	}
4985	Assert(pVCpu->hm.s.vmx.pfnStartVM);
4986	return VINF_SUCCESS;
4987	}
4988
4989
4990	/**
4991	* Wrapper for running the guest code in VT-x.
4992	*
4993	* @returns VBox status code, no informational status codes.
4994	* @param pVM The cross context VM structure.
4995	* @param pVCpu The cross context virtual CPU structure.
4996	* @param pCtx Pointer to the guest-CPU context.
4997	*
4998	* @remarks No-long-jump zone!!!
4999	*/
5000	DECLINLINE(int) hmR0VmxRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
5001	{
5002	/*
5003	* 64-bit Windows uses XMM registers in the kernel as the Microsoft compiler expresses floating-point operations
5004	* using SSE instructions. Some XMM registers (XMM6-XMM15) are callee-saved and thus the need for this XMM wrapper.
5005	* Refer MSDN docs. "Configuring Programs for 64-bit / x64 Software Conventions / Register Usage" for details.
5006	*/
5007	bool const fResumeVM = RT_BOOL(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_LAUNCHED);
5008	/** @todo Add stats for resume vs launch. */
5009	#ifdef VBOX_WITH_KERNEL_USING_XMM
5010	int rc = hmR0VMXStartVMWrapXMM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu, pVCpu->hm.s.vmx.pfnStartVM);
5011	#else
5012	int rc = pVCpu->hm.s.vmx.pfnStartVM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu);
5013	#endif
5014	AssertMsg(rc <= VINF_SUCCESS, ("%Rrc\n", rc));
5015	return rc;
5016	}
5017
5018
5019	/**
5020	* Reports world-switch error and dumps some useful debug info.
5021	*
5022	* @param pVM The cross context VM structure.
5023	* @param pVCpu The cross context virtual CPU structure.
5024	* @param rcVMRun The return code from VMLAUNCH/VMRESUME.
5025	* @param pCtx Pointer to the guest-CPU context.
5026	* @param pVmxTransient Pointer to the VMX transient structure (only
5027	* exitReason updated).
5028	*/
5029	static void hmR0VmxReportWorldSwitchError(PVM pVM, PVMCPU pVCpu, int rcVMRun, PCPUMCTX pCtx, PVMXTRANSIENT pVmxTransient)
5030	{
5031	Assert(pVM);
5032	Assert(pVCpu);
5033	Assert(pCtx);
5034	Assert(pVmxTransient);
5035	HMVMX_ASSERT_PREEMPT_SAFE();
5036
5037	Log4(("VM-entry failure: %Rrc\n", rcVMRun));
5038	switch (rcVMRun)
5039	{
5040	case VERR_VMX_INVALID_VMXON_PTR:
5041	AssertFailed();
5042	break;
5043	case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */
5044	case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */
5045	{
5046	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason);
5047	rc \|= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
5048	rc \|= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
5049	AssertRC(rc);
5050
5051	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
5052	/* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted().
5053	Cannot do it here as we may have been long preempted. */
5054
5055	#ifdef VBOX_STRICT
5056	Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason,
5057	pVmxTransient->uExitReason));
5058	Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQualification));
5059	Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError));
5060	if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX)
5061	Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError]));
5062	else
5063	Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX));
5064	Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu));
5065	Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu));
5066
5067	/* VMX control bits. */
5068	uint32_t u32Val;
5069	uint64_t u64Val;
5070	RTHCUINTREG uHCReg;
5071	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val); AssertRC(rc);
5072	Log4(("VMX_VMCS32_CTRL_PIN_EXEC %#RX32\n", u32Val));
5073	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val); AssertRC(rc);
5074	Log4(("VMX_VMCS32_CTRL_PROC_EXEC %#RX32\n", u32Val));
5075	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
5076	{
5077	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val); AssertRC(rc);
5078	Log4(("VMX_VMCS32_CTRL_PROC_EXEC2 %#RX32\n", u32Val));
5079	}
5080	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val); AssertRC(rc);
5081	Log4(("VMX_VMCS32_CTRL_ENTRY %#RX32\n", u32Val));
5082	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val); AssertRC(rc);
5083	Log4(("VMX_VMCS32_CTRL_EXIT %#RX32\n", u32Val));
5084	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, &u32Val); AssertRC(rc);
5085	Log4(("VMX_VMCS32_CTRL_CR3_TARGET_COUNT %#RX32\n", u32Val));
5086	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32Val); AssertRC(rc);
5087	Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", u32Val));
5088	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &u32Val); AssertRC(rc);
5089	Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", u32Val));
5090	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &u32Val); AssertRC(rc);
5091	Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %u\n", u32Val));
5092	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, &u32Val); AssertRC(rc);
5093	Log4(("VMX_VMCS32_CTRL_TPR_THRESHOLD %u\n", u32Val));
5094	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &u32Val); AssertRC(rc);
5095	Log4(("VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT %u (guest MSRs)\n", u32Val));
5096	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &u32Val); AssertRC(rc);
5097	Log4(("VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT %u (host MSRs)\n", u32Val));
5098	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &u32Val); AssertRC(rc);
5099	Log4(("VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT %u (guest MSRs)\n", u32Val));
5100	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val); AssertRC(rc);
5101	Log4(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP %#RX32\n", u32Val));
5102	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, &u32Val); AssertRC(rc);
5103	Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK %#RX32\n", u32Val));
5104	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, &u32Val); AssertRC(rc);
5105	Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH %#RX32\n", u32Val));
5106	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc);
5107	Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg));
5108	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc);
5109	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
5110	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc);
5111	Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg));
5112	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc);
5113	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
5114	if (pVM->hm.s.fNestedPaging)
5115	{
5116	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
5117	Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
5118	}
5119
5120	/* Guest bits. */
5121	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val); AssertRC(rc);
5122	Log4(("Old Guest Rip %#RX64 New %#RX64\n", pCtx->rip, u64Val));
5123	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val); AssertRC(rc);
5124	Log4(("Old Guest Rsp %#RX64 New %#RX64\n", pCtx->rsp, u64Val));
5125	rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Val); AssertRC(rc);
5126	Log4(("Old Guest Rflags %#RX32 New %#RX32\n", pCtx->eflags.u32, u32Val));
5127	if (pVM->hm.s.vmx.fVpid)
5128	{
5129	rc = VMXReadVmcs32(VMX_VMCS16_VPID, &u32Val); AssertRC(rc);
5130	Log4(("VMX_VMCS16_VPID %u\n", u32Val));
5131	}
5132
5133	/* Host bits. */
5134	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR0, &uHCReg); AssertRC(rc);
5135	Log4(("Host CR0 %#RHr\n", uHCReg));
5136	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR3, &uHCReg); AssertRC(rc);
5137	Log4(("Host CR3 %#RHr\n", uHCReg));
5138	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR4, &uHCReg); AssertRC(rc);
5139	Log4(("Host CR4 %#RHr\n", uHCReg));
5140
5141	RTGDTR HostGdtr;
5142	PCX86DESCHC pDesc;
5143	ASMGetGDTR(&HostGdtr);
5144	rc = VMXReadVmcs32(VMX_VMCS16_HOST_CS_SEL, &u32Val); AssertRC(rc);
5145	Log4(("Host CS %#08x\n", u32Val));
5146	if (u32Val < HostGdtr.cbGdt)
5147	{
5148	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5149	hmR0DumpDescriptor(pDesc, u32Val, "CS: ");
5150	}
5151
5152	rc = VMXReadVmcs32(VMX_VMCS16_HOST_DS_SEL, &u32Val); AssertRC(rc);
5153	Log4(("Host DS %#08x\n", u32Val));
5154	if (u32Val < HostGdtr.cbGdt)
5155	{
5156	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5157	hmR0DumpDescriptor(pDesc, u32Val, "DS: ");
5158	}
5159
5160	rc = VMXReadVmcs32(VMX_VMCS16_HOST_ES_SEL, &u32Val); AssertRC(rc);
5161	Log4(("Host ES %#08x\n", u32Val));
5162	if (u32Val < HostGdtr.cbGdt)
5163	{
5164	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5165	hmR0DumpDescriptor(pDesc, u32Val, "ES: ");
5166	}
5167
5168	rc = VMXReadVmcs32(VMX_VMCS16_HOST_FS_SEL, &u32Val); AssertRC(rc);
5169	Log4(("Host FS %#08x\n", u32Val));
5170	if (u32Val < HostGdtr.cbGdt)
5171	{
5172	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5173	hmR0DumpDescriptor(pDesc, u32Val, "FS: ");
5174	}
5175
5176	rc = VMXReadVmcs32(VMX_VMCS16_HOST_GS_SEL, &u32Val); AssertRC(rc);
5177	Log4(("Host GS %#08x\n", u32Val));
5178	if (u32Val < HostGdtr.cbGdt)
5179	{
5180	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5181	hmR0DumpDescriptor(pDesc, u32Val, "GS: ");
5182	}
5183
5184	rc = VMXReadVmcs32(VMX_VMCS16_HOST_SS_SEL, &u32Val); AssertRC(rc);
5185	Log4(("Host SS %#08x\n", u32Val));
5186	if (u32Val < HostGdtr.cbGdt)
5187	{
5188	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5189	hmR0DumpDescriptor(pDesc, u32Val, "SS: ");
5190	}
5191
5192	rc = VMXReadVmcs32(VMX_VMCS16_HOST_TR_SEL, &u32Val); AssertRC(rc);
5193	Log4(("Host TR %#08x\n", u32Val));
5194	if (u32Val < HostGdtr.cbGdt)
5195	{
5196	pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5197	hmR0DumpDescriptor(pDesc, u32Val, "TR: ");
5198	}
5199
5200	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_TR_BASE, &uHCReg); AssertRC(rc);
5201	Log4(("Host TR Base %#RHv\n", uHCReg));
5202	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, &uHCReg); AssertRC(rc);
5203	Log4(("Host GDTR Base %#RHv\n", uHCReg));
5204	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, &uHCReg); AssertRC(rc);
5205	Log4(("Host IDTR Base %#RHv\n", uHCReg));
5206	rc = VMXReadVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, &u32Val); AssertRC(rc);
5207	Log4(("Host SYSENTER CS %#08x\n", u32Val));
5208	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_EIP, &uHCReg); AssertRC(rc);
5209	Log4(("Host SYSENTER EIP %#RHv\n", uHCReg));
5210	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_ESP, &uHCReg); AssertRC(rc);
5211	Log4(("Host SYSENTER ESP %#RHv\n", uHCReg));
5212	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RSP, &uHCReg); AssertRC(rc);
5213	Log4(("Host RSP %#RHv\n", uHCReg));
5214	rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RIP, &uHCReg); AssertRC(rc);
5215	Log4(("Host RIP %#RHv\n", uHCReg));
5216	# if HC_ARCH_BITS == 64
5217	Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER)));
5218	Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR)));
5219	Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR)));
5220	Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR)));
5221	Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK)));
5222	Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE)));
5223	# endif
5224	#endif /* VBOX_STRICT */
5225	break;
5226	}
5227
5228	default:
5229	/* Impossible */
5230	AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun));
5231	break;
5232	}
5233	NOREF(pVM); NOREF(pCtx);
5234	}
5235
5236
5237	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
5238	#ifndef VMX_USE_CACHED_VMCS_ACCESSES
5239	# error "VMX_USE_CACHED_VMCS_ACCESSES not defined when it should be!"
5240	#endif
5241	#ifdef VBOX_STRICT
5242	static bool hmR0VmxIsValidWriteField(uint32_t idxField)
5243	{
5244	switch (idxField)
5245	{
5246	case VMX_VMCS_GUEST_RIP:
5247	case VMX_VMCS_GUEST_RSP:
5248	case VMX_VMCS_GUEST_SYSENTER_EIP:
5249	case VMX_VMCS_GUEST_SYSENTER_ESP:
5250	case VMX_VMCS_GUEST_GDTR_BASE:
5251	case VMX_VMCS_GUEST_IDTR_BASE:
5252	case VMX_VMCS_GUEST_CS_BASE:
5253	case VMX_VMCS_GUEST_DS_BASE:
5254	case VMX_VMCS_GUEST_ES_BASE:
5255	case VMX_VMCS_GUEST_FS_BASE:
5256	case VMX_VMCS_GUEST_GS_BASE:
5257	case VMX_VMCS_GUEST_SS_BASE:
5258	case VMX_VMCS_GUEST_LDTR_BASE:
5259	case VMX_VMCS_GUEST_TR_BASE:
5260	case VMX_VMCS_GUEST_CR3:
5261	return true;
5262	}
5263	return false;
5264	}
5265
5266	static bool hmR0VmxIsValidReadField(uint32_t idxField)
5267	{
5268	switch (idxField)
5269	{
5270	/* Read-only fields. */
5271	case VMX_VMCS_RO_EXIT_QUALIFICATION:
5272	return true;
5273	}
5274	/* Remaining readable fields should also be writable. */
5275	return hmR0VmxIsValidWriteField(idxField);
5276	}
5277	#endif /* VBOX_STRICT */
5278
5279
5280	/**
5281	* Executes the specified handler in 64-bit mode.
5282	*
5283	* @returns VBox status code (no informational status codes).
5284	* @param pVM The cross context VM structure.
5285	* @param pVCpu The cross context virtual CPU structure.
5286	* @param pCtx Pointer to the guest CPU context.
5287	* @param enmOp The operation to perform.
5288	* @param cParams Number of parameters.
5289	* @param paParam Array of 32-bit parameters.
5290	*/
5291	VMMR0DECL(int) VMXR0Execute64BitsHandler(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, HM64ON32OP enmOp,
5292	uint32_t cParams, uint32_t *paParam)
5293	{
5294	NOREF(pCtx);
5295
5296	AssertReturn(pVM->hm.s.pfnHost32ToGuest64R0, VERR_HM_NO_32_TO_64_SWITCHER);
5297	Assert(enmOp > HM64ON32OP_INVALID && enmOp < HM64ON32OP_END);
5298	Assert(pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Write.aField));
5299	Assert(pVCpu->hm.s.vmx.VMCSCache.Read.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Read.aField));
5300
5301	#ifdef VBOX_STRICT
5302	for (uint32_t i = 0; i < pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries; i++)
5303	Assert(hmR0VmxIsValidWriteField(pVCpu->hm.s.vmx.VMCSCache.Write.aField[i]));
5304
5305	for (uint32_t i = 0; i <pVCpu->hm.s.vmx.VMCSCache.Read.cValidEntries; i++)
5306	Assert(hmR0VmxIsValidReadField(pVCpu->hm.s.vmx.VMCSCache.Read.aField[i]));
5307	#endif
5308
5309	/* Disable interrupts. */
5310	RTCCUINTREG fOldEFlags = ASMIntDisableFlags();
5311
5312	#ifdef VBOX_WITH_VMMR0_DISABLE_LAPIC_NMI
5313	RTCPUID idHostCpu = RTMpCpuId();
5314	CPUMR0SetLApic(pVCpu, idHostCpu);
5315	#endif
5316
5317	PHMGLOBALCPUINFO pCpu = hmR0GetCurrentCpu();
5318	RTHCPHYS HCPhysCpuPage = pCpu->HCPhysMemObj;
5319
5320	/* Clear VMCS. Marking it inactive, clearing implementation-specific data and writing VMCS data back to memory. */
5321	VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
5322	pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
5323
5324	/* Leave VMX Root Mode. */
5325	VMXDisable();
5326
5327	SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
5328
5329	CPUMSetHyperESP(pVCpu, VMMGetStackRC(pVCpu));
5330	CPUMSetHyperEIP(pVCpu, enmOp);
5331	for (int i = (int)cParams - 1; i >= 0; i--)
5332	CPUMPushHyper(pVCpu, paParam[i]);
5333
5334	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatWorldSwitch3264, z);
5335
5336	/* Call the switcher. */
5337	int rc = pVM->hm.s.pfnHost32ToGuest64R0(pVM, RT_OFFSETOF(VM, aCpus[pVCpu->idCpu].cpum) - RT_OFFSETOF(VM, cpum));
5338	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatWorldSwitch3264, z);
5339
5340	/** @todo replace with hmR0VmxEnterRootMode() and hmR0VmxLeaveRootMode(). */
5341	/* Make sure the VMX instructions don't cause #UD faults. */
5342	SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
5343
5344	/* Re-enter VMX Root Mode */
5345	int rc2 = VMXEnable(HCPhysCpuPage);
5346	if (RT_FAILURE(rc2))
5347	{
5348	SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
5349	ASMSetFlags(fOldEFlags);
5350	pVM->hm.s.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
5351	return rc2;
5352	}
5353
5354	rc2 = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
5355	AssertRC(rc2);
5356	pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE;
5357	Assert(!(ASMGetFlags() & X86_EFL_IF));
5358	ASMSetFlags(fOldEFlags);
5359	return rc;
5360	}
5361
5362
5363	/**
5364	* Prepares for and executes VMLAUNCH (64-bit guests) for 32-bit hosts
5365	* supporting 64-bit guests.
5366	*
5367	* @returns VBox status code.
5368	* @param fResume Whether to VMLAUNCH or VMRESUME.
5369	* @param pCtx Pointer to the guest-CPU context.
5370	* @param pCache Pointer to the VMCS cache.
5371	* @param pVM The cross context VM structure.
5372	* @param pVCpu The cross context virtual CPU structure.
5373	*/
5374	DECLASM(int) VMXR0SwitcherStartVM64(RTHCUINT fResume, PCPUMCTX pCtx, PVMCSCACHE pCache, PVM pVM, PVMCPU pVCpu)
5375	{
5376	NOREF(fResume);
5377
5378	PHMGLOBALCPUINFO pCpu = hmR0GetCurrentCpu();
5379	RTHCPHYS HCPhysCpuPage = pCpu->HCPhysMemObj;
5380
5381	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5382	pCache->uPos = 1;
5383	pCache->interPD = PGMGetInterPaeCR3(pVM);
5384	pCache->pSwitcher = (uint64_t)pVM->hm.s.pfnHost32ToGuest64R0;
5385	#endif
5386
5387	#if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES)
5388	pCache->TestIn.HCPhysCpuPage = 0;
5389	pCache->TestIn.HCPhysVmcs = 0;
5390	pCache->TestIn.pCache = 0;
5391	pCache->TestOut.HCPhysVmcs = 0;
5392	pCache->TestOut.pCache = 0;
5393	pCache->TestOut.pCtx = 0;
5394	pCache->TestOut.eflags = 0;
5395	#else
5396	NOREF(pCache);
5397	#endif
5398
5399	uint32_t aParam[10];
5400	aParam[0] = (uint32_t)(HCPhysCpuPage); /* Param 1: VMXON physical address - Lo. */
5401	aParam[1] = (uint32_t)(HCPhysCpuPage >> 32); /* Param 1: VMXON physical address - Hi. */
5402	aParam[2] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs); /* Param 2: VMCS physical address - Lo. */
5403	aParam[3] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs >> 32); /* Param 2: VMCS physical address - Hi. */
5404	aParam[4] = VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache);
5405	aParam[5] = 0;
5406	aParam[6] = VM_RC_ADDR(pVM, pVM);
5407	aParam[7] = 0;
5408	aParam[8] = VM_RC_ADDR(pVM, pVCpu);
5409	aParam[9] = 0;
5410
5411	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5412	pCtx->dr[4] = pVM->hm.s.vmx.pScratchPhys + 16 + 8;
5413	(uint32_t )(pVM->hm.s.vmx.pScratch + 16 + 8) = 1;
5414	#endif
5415	int rc = VMXR0Execute64BitsHandler(pVM, pVCpu, pCtx, HM64ON32OP_VMXRCStartVM64, RT_ELEMENTS(aParam), &aParam[0]);
5416
5417	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5418	Assert((uint32_t )(pVM->hm.s.vmx.pScratch + 16 + 8) == 5);
5419	Assert(pCtx->dr[4] == 10);
5420	(uint32_t )(pVM->hm.s.vmx.pScratch + 16 + 8) = 0xff;
5421	#endif
5422
5423	#if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES)
5424	AssertMsg(pCache->TestIn.HCPhysCpuPage == HCPhysCpuPage, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysCpuPage, HCPhysCpuPage));
5425	AssertMsg(pCache->TestIn.HCPhysVmcs == pVCpu->hm.s.vmx.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
5426	pVCpu->hm.s.vmx.HCPhysVmcs));
5427	AssertMsg(pCache->TestIn.HCPhysVmcs == pCache->TestOut.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
5428	pCache->TestOut.HCPhysVmcs));
5429	AssertMsg(pCache->TestIn.pCache == pCache->TestOut.pCache, ("%RGv vs %RGv\n", pCache->TestIn.pCache,
5430	pCache->TestOut.pCache));
5431	AssertMsg(pCache->TestIn.pCache == VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache),
5432	("%RGv vs %RGv\n", pCache->TestIn.pCache, VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache)));
5433	AssertMsg(pCache->TestIn.pCtx == pCache->TestOut.pCtx, ("%RGv vs %RGv\n", pCache->TestIn.pCtx,
5434	pCache->TestOut.pCtx));
5435	Assert(!(pCache->TestOut.eflags & X86_EFL_IF));
5436	#endif
5437	return rc;
5438	}
5439
5440
5441	/**
5442	* Initialize the VMCS-Read cache.
5443	*
5444	* The VMCS cache is used for 32-bit hosts running 64-bit guests (except 32-bit
5445	* Darwin which runs with 64-bit paging in 32-bit mode) for 64-bit fields that
5446	* cannot be accessed in 32-bit mode. Some 64-bit fields -can- be accessed
5447	* (those that have a 32-bit FULL & HIGH part).
5448	*
5449	* @returns VBox status code.
5450	* @param pVM The cross context VM structure.
5451	* @param pVCpu The cross context virtual CPU structure.
5452	*/
5453	static int hmR0VmxInitVmcsReadCache(PVM pVM, PVMCPU pVCpu)
5454	{
5455	#define VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, idxField) \
5456	{ \
5457	Assert(pCache->Read.aField[idxField##_CACHE_IDX] == 0); \
5458	pCache->Read.aField[idxField##_CACHE_IDX] = idxField; \
5459	pCache->Read.aFieldVal[idxField##_CACHE_IDX] = 0; \
5460	++cReadFields; \
5461	}
5462
5463	AssertPtr(pVM);
5464	AssertPtr(pVCpu);
5465	PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache;
5466	uint32_t cReadFields = 0;
5467
5468	/*
5469	* Don't remove the #if 0'd fields in this code. They're listed here for consistency
5470	* and serve to indicate exceptions to the rules.
5471	*/
5472
5473	/* Guest-natural selector base fields. */
5474	#if 0
5475	/* These are 32-bit in practice. See Intel spec. 2.5 "Control Registers". */
5476	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR0);
5477	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR4);
5478	#endif
5479	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_ES_BASE);
5480	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CS_BASE);
5481	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SS_BASE);
5482	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_DS_BASE);
5483	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_FS_BASE);
5484	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GS_BASE);
5485	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_LDTR_BASE);
5486	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_TR_BASE);
5487	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GDTR_BASE);
5488	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_IDTR_BASE);
5489	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RSP);
5490	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RIP);
5491	#if 0
5492	/* Unused natural width guest-state fields. */
5493	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS);
5494	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3); /* Handled in Nested Paging case */
5495	#endif
5496	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_ESP);
5497	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_EIP);
5498
5499	/* 64-bit guest-state fields; unused as we use two 32-bit VMREADs for these 64-bit fields (using "FULL" and "HIGH" fields). */
5500	#if 0
5501	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL);
5502	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_DEBUGCTL_FULL);
5503	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PAT_FULL);
5504	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_EFER_FULL);
5505	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL);
5506	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE0_FULL);
5507	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE1_FULL);
5508	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE2_FULL);
5509	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE3_FULL);
5510	#endif
5511
5512	/* Natural width guest-state fields. */
5513	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_QUALIFICATION);
5514	#if 0
5515	/* Currently unused field. */
5516	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_GUEST_LINEAR_ADDR);
5517	#endif
5518
5519	if (pVM->hm.s.fNestedPaging)
5520	{
5521	VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3);
5522	AssertMsg(cReadFields == VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields,
5523	VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX));
5524	pCache->Read.cValidEntries = VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX;
5525	}
5526	else
5527	{
5528	AssertMsg(cReadFields == VMX_VMCS_MAX_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields, VMX_VMCS_MAX_CACHE_IDX));
5529	pCache->Read.cValidEntries = VMX_VMCS_MAX_CACHE_IDX;
5530	}
5531
5532	#undef VMXLOCAL_INIT_READ_CACHE_FIELD
5533	return VINF_SUCCESS;
5534	}
5535
5536
5537	/**
5538	* Writes a field into the VMCS. This can either directly invoke a VMWRITE or
5539	* queue up the VMWRITE by using the VMCS write cache (on 32-bit hosts, except
5540	* darwin, running 64-bit guests).
5541	*
5542	* @returns VBox status code.
5543	* @param pVCpu The cross context virtual CPU structure.
5544	* @param idxField The VMCS field encoding.
5545	* @param u64Val 16, 32 or 64-bit value.
5546	*/
5547	VMMR0DECL(int) VMXWriteVmcs64Ex(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val)
5548	{
5549	int rc;
5550	switch (idxField)
5551	{
5552	/*
5553	* These fields consists of a "FULL" and a "HIGH" part which can be written to individually.
5554	*/
5555	/* 64-bit Control fields. */
5556	case VMX_VMCS64_CTRL_IO_BITMAP_A_FULL:
5557	case VMX_VMCS64_CTRL_IO_BITMAP_B_FULL:
5558	case VMX_VMCS64_CTRL_MSR_BITMAP_FULL:
5559	case VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL:
5560	case VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL:
5561	case VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL:
5562	case VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL:
5563	case VMX_VMCS64_CTRL_TSC_OFFSET_FULL:
5564	case VMX_VMCS64_CTRL_VAPIC_PAGEADDR_FULL:
5565	case VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL:
5566	case VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL:
5567	case VMX_VMCS64_CTRL_EPTP_FULL:
5568	case VMX_VMCS64_CTRL_EPTP_LIST_FULL:
5569	/* 64-bit Guest-state fields. */
5570	case VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL:
5571	case VMX_VMCS64_GUEST_DEBUGCTL_FULL:
5572	case VMX_VMCS64_GUEST_PAT_FULL:
5573	case VMX_VMCS64_GUEST_EFER_FULL:
5574	case VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL:
5575	case VMX_VMCS64_GUEST_PDPTE0_FULL:
5576	case VMX_VMCS64_GUEST_PDPTE1_FULL:
5577	case VMX_VMCS64_GUEST_PDPTE2_FULL:
5578	case VMX_VMCS64_GUEST_PDPTE3_FULL:
5579	/* 64-bit Host-state fields. */
5580	case VMX_VMCS64_HOST_PAT_FULL:
5581	case VMX_VMCS64_HOST_EFER_FULL:
5582	case VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL:
5583	{
5584	rc = VMXWriteVmcs32(idxField, u64Val);
5585	rc \|= VMXWriteVmcs32(idxField + 1, (uint32_t)(u64Val >> 32));
5586	break;
5587	}
5588
5589	/*
5590	* These fields do not have high and low parts. Queue up the VMWRITE by using the VMCS write-cache (for 64-bit
5591	* values). When we switch the host to 64-bit mode for running 64-bit guests, these VMWRITEs get executed then.
5592	*/
5593	/* Natural-width Guest-state fields. */
5594	case VMX_VMCS_GUEST_CR3:
5595	case VMX_VMCS_GUEST_ES_BASE:
5596	case VMX_VMCS_GUEST_CS_BASE:
5597	case VMX_VMCS_GUEST_SS_BASE:
5598	case VMX_VMCS_GUEST_DS_BASE:
5599	case VMX_VMCS_GUEST_FS_BASE:
5600	case VMX_VMCS_GUEST_GS_BASE:
5601	case VMX_VMCS_GUEST_LDTR_BASE:
5602	case VMX_VMCS_GUEST_TR_BASE:
5603	case VMX_VMCS_GUEST_GDTR_BASE:
5604	case VMX_VMCS_GUEST_IDTR_BASE:
5605	case VMX_VMCS_GUEST_RSP:
5606	case VMX_VMCS_GUEST_RIP:
5607	case VMX_VMCS_GUEST_SYSENTER_ESP:
5608	case VMX_VMCS_GUEST_SYSENTER_EIP:
5609	{
5610	if (!(u64Val >> 32))
5611	{
5612	/* If this field is 64-bit, VT-x will zero out the top bits. */
5613	rc = VMXWriteVmcs32(idxField, (uint32_t)u64Val);
5614	}
5615	else
5616	{
5617	/* Assert that only the 32->64 switcher case should ever come here. */
5618	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests);
5619	rc = VMXWriteCachedVmcsEx(pVCpu, idxField, u64Val);
5620	}
5621	break;
5622	}
5623
5624	default:
5625	{
5626	AssertMsgFailed(("VMXWriteVmcs64Ex: Invalid field %#RX32 (pVCpu=%p u64Val=%#RX64)\n", idxField, pVCpu, u64Val));
5627	rc = VERR_INVALID_PARAMETER;
5628	break;
5629	}
5630	}
5631	AssertRCReturn(rc, rc);
5632	return rc;
5633	}
5634
5635
5636	/**
5637	* Queue up a VMWRITE by using the VMCS write cache.
5638	* This is only used on 32-bit hosts (except darwin) for 64-bit guests.
5639	*
5640	* @param pVCpu The cross context virtual CPU structure.
5641	* @param idxField The VMCS field encoding.
5642	* @param u64Val 16, 32 or 64-bit value.
5643	*/
5644	VMMR0DECL(int) VMXWriteCachedVmcsEx(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val)
5645	{
5646	AssertPtr(pVCpu);
5647	PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache;
5648
5649	AssertMsgReturn(pCache->Write.cValidEntries < VMCSCACHE_MAX_ENTRY - 1,
5650	("entries=%u\n", pCache->Write.cValidEntries), VERR_ACCESS_DENIED);
5651
5652	/* Make sure there are no duplicates. */
5653	for (uint32_t i = 0; i < pCache->Write.cValidEntries; i++)
5654	{
5655	if (pCache->Write.aField[i] == idxField)
5656	{
5657	pCache->Write.aFieldVal[i] = u64Val;
5658	return VINF_SUCCESS;
5659	}
5660	}
5661
5662	pCache->Write.aField[pCache->Write.cValidEntries] = idxField;
5663	pCache->Write.aFieldVal[pCache->Write.cValidEntries] = u64Val;
5664	pCache->Write.cValidEntries++;
5665	return VINF_SUCCESS;
5666	}
5667	#endif /* HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) */
5668
5669
5670	/**
5671	* Sets up the usage of TSC-offsetting and updates the VMCS.
5672	*
5673	* If offsetting is not possible, cause VM-exits on RDTSC(P)s. Also sets up the
5674	* VMX preemption timer.
5675	*
5676	* @returns VBox status code.
5677	* @param pVM The cross context VM structure.
5678	* @param pVCpu The cross context virtual CPU structure.
5679	*
5680	* @remarks No-long-jump zone!!!
5681	*/
5682	static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVM pVM, PVMCPU pVCpu)
5683	{
5684	int rc;
5685	bool fOffsettedTsc;
5686	bool fParavirtTsc;
5687	if (pVM->hm.s.vmx.fUsePreemptTimer)
5688	{
5689	uint64_t cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVM, pVCpu, &pVCpu->hm.s.vmx.u64TSCOffset,
5690	&fOffsettedTsc, &fParavirtTsc);
5691
5692	/* Make sure the returned values have sane upper and lower boundaries. */
5693	uint64_t u64CpuHz = SUPGetCpuHzFromGipBySetIndex(g_pSUPGlobalInfoPage, pVCpu->iHostCpuSet);
5694	cTicksToDeadline = RT_MIN(cTicksToDeadline, u64CpuHz / 64); /* 1/64th of a second */
5695	cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 2048); /* 1/2048th of a second */
5696	cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift;
5697
5698	uint32_t cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16);
5699	rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_PREEMPT_TIMER_VALUE, cPreemptionTickCount); AssertRC(rc);
5700	}
5701	else
5702	fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &pVCpu->hm.s.vmx.u64TSCOffset, &fParavirtTsc);
5703
5704	/** @todo later optimize this to be done elsewhere and not before every
5705	* VM-entry. */
5706	if (fParavirtTsc)
5707	{
5708	/* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
5709	information before every VM-entry, hence disable it for performance sake. */
5710	#if 0
5711	rc = GIMR0UpdateParavirtTsc(pVM, 0 /* u64Offset */);
5712	AssertRC(rc);
5713	#endif
5714	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
5715	}
5716
5717	if (fOffsettedTsc && RT_LIKELY(!pVCpu->hm.s.fDebugWantRdTscExit))
5718	{
5719	/* Note: VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT takes precedence over TSC_OFFSET, applies to RDTSCP too. */
5720	rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, pVCpu->hm.s.vmx.u64TSCOffset); AssertRC(rc);
5721
5722	pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT;
5723	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc);
5724	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
5725	}
5726	else
5727	{
5728	/* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */
5729	pVCpu->hm.s.vmx.u32ProcCtls \|= VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT;
5730	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc);
5731	STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
5732	}
5733	}
5734
5735
5736	/**
5737	* Determines if an exception is a contributory exception.
5738	*
5739	* Contributory exceptions are ones which can cause double-faults unless the
5740	* original exception was a benign exception. Page-fault is intentionally not
5741	* included here as it's a conditional contributory exception.
5742	*
5743	* @returns true if the exception is contributory, false otherwise.
5744	* @param uVector The exception vector.
5745	*/
5746	DECLINLINE(bool) hmR0VmxIsContributoryXcpt(const uint32_t uVector)
5747	{
5748	switch (uVector)
5749	{
5750	case X86_XCPT_GP:
5751	case X86_XCPT_SS:
5752	case X86_XCPT_NP:
5753	case X86_XCPT_TS:
5754	case X86_XCPT_DE:
5755	return true;
5756	default:
5757	break;
5758	}
5759	return false;
5760	}
5761
5762
5763	/**
5764	* Sets an event as a pending event to be injected into the guest.
5765	*
5766	* @param pVCpu The cross context virtual CPU structure.
5767	* @param u32IntInfo The VM-entry interruption-information field.
5768	* @param cbInstr The VM-entry instruction length in bytes (for software
5769	* interrupts, exceptions and privileged software
5770	* exceptions).
5771	* @param u32ErrCode The VM-entry exception error code.
5772	* @param GCPtrFaultAddress The fault-address (CR2) in case it's a
5773	* page-fault.
5774	*
5775	* @remarks Statistics counter assumes this is a guest event being injected or
5776	* re-injected into the guest, i.e. 'StatInjectPendingReflect' is
5777	* always incremented.
5778	*/
5779	DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPU pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode,
5780	RTGCUINTPTR GCPtrFaultAddress)
5781	{
5782	Assert(!pVCpu->hm.s.Event.fPending);
5783	pVCpu->hm.s.Event.fPending = true;
5784	pVCpu->hm.s.Event.u64IntInfo = u32IntInfo;
5785	pVCpu->hm.s.Event.u32ErrCode = u32ErrCode;
5786	pVCpu->hm.s.Event.cbInstr = cbInstr;
5787	pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
5788	}
5789
5790
5791	/**
5792	* Sets a double-fault (\#DF) exception as pending-for-injection into the VM.
5793	*
5794	* @param pVCpu The cross context virtual CPU structure.
5795	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
5796	* out-of-sync. Make sure to update the required fields
5797	* before using them.
5798	*/
5799	DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
5800	{
5801	NOREF(pMixedCtx);
5802	uint32_t u32IntInfo = X86_XCPT_DF \| VMX_EXIT_INTERRUPTION_INFO_VALID;
5803	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
5804	u32IntInfo \|= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
5805	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
5806	}
5807
5808
5809	/**
5810	* Handle a condition that occurred while delivering an event through the guest
5811	* IDT.
5812	*
5813	* @returns Strict VBox status code (i.e. informational status codes too).
5814	* @retval VINF_SUCCESS if we should continue handling the VM-exit.
5815	* @retval VINF_HM_DOUBLE_FAULT if a \#DF condition was detected and we ought
5816	* to continue execution of the guest which will delivery the \#DF.
5817	* @retval VINF_EM_RESET if we detected a triple-fault condition.
5818	* @retval VERR_EM_GUEST_CPU_HANG if we detected a guest CPU hang.
5819	*
5820	* @param pVCpu The cross context virtual CPU structure.
5821	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
5822	* out-of-sync. Make sure to update the required fields
5823	* before using them.
5824	* @param pVmxTransient Pointer to the VMX transient structure.
5825	*
5826	* @remarks No-long-jump zone!!!
5827	*/
5828	static VBOXSTRICTRC hmR0VmxCheckExitDueToEventDelivery(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
5829	{
5830	uint32_t uExitVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo);
5831
5832	int rc2 = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient); AssertRCReturn(rc2, rc2);
5833	rc2 = hmR0VmxReadExitIntInfoVmcs(pVmxTransient); AssertRCReturn(rc2, rc2);
5834
5835	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
5836	if (VMX_IDT_VECTORING_INFO_VALID(pVmxTransient->uIdtVectoringInfo))
5837	{
5838	uint32_t uIdtVectorType = VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo);
5839	uint32_t uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo);
5840
5841	typedef enum
5842	{
5843	VMXREFLECTXCPT_XCPT, /* Reflect the exception to the guest or for further evaluation by VMM. */
5844	VMXREFLECTXCPT_DF, /* Reflect the exception as a double-fault to the guest. */
5845	VMXREFLECTXCPT_TF, /* Indicate a triple faulted state to the VMM. */
5846	VMXREFLECTXCPT_HANG, /* Indicate bad VM trying to deadlock the CPU. */
5847	VMXREFLECTXCPT_NONE /* Nothing to reflect. */
5848	} VMXREFLECTXCPT;
5849
5850	/* See Intel spec. 30.7.1.1 "Reflecting Exceptions to Guest Software". */
5851	VMXREFLECTXCPT enmReflect = VMXREFLECTXCPT_NONE;
5852	if (VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo))
5853	{
5854	if (uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT)
5855	{
5856	enmReflect = VMXREFLECTXCPT_XCPT;
5857	#ifdef VBOX_STRICT
5858	if ( hmR0VmxIsContributoryXcpt(uIdtVector)
5859	&& uExitVector == X86_XCPT_PF)
5860	{
5861	Log4(("IDT: vcpu[%RU32] Contributory #PF uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5862	}
5863	#endif
5864	if ( uExitVector == X86_XCPT_PF
5865	&& uIdtVector == X86_XCPT_PF)
5866	{
5867	pVmxTransient->fVectoringDoublePF = true;
5868	Log4(("IDT: vcpu[%RU32] Vectoring Double #PF uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5869	}
5870	else if ( uExitVector == X86_XCPT_AC
5871	&& uIdtVector == X86_XCPT_AC)
5872	{
5873	enmReflect = VMXREFLECTXCPT_HANG;
5874	Log4(("IDT: Nested #AC - Bad guest\n"));
5875	}
5876	else if ( (pVCpu->hm.s.vmx.u32XcptBitmap & HMVMX_CONTRIBUTORY_XCPT_MASK)
5877	&& hmR0VmxIsContributoryXcpt(uExitVector)
5878	&& ( hmR0VmxIsContributoryXcpt(uIdtVector)
5879	\|\| uIdtVector == X86_XCPT_PF))
5880	{
5881	enmReflect = VMXREFLECTXCPT_DF;
5882	}
5883	else if (uIdtVector == X86_XCPT_DF)
5884	enmReflect = VMXREFLECTXCPT_TF;
5885	}
5886	else if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT
5887	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI)
5888	{
5889	/*
5890	* Ignore software interrupts (INT n), software exceptions (#BP, #OF) and
5891	* privileged software exception (#DB from ICEBP) as they reoccur when restarting the instruction.
5892	*/
5893	enmReflect = VMXREFLECTXCPT_XCPT;
5894
5895	if (uExitVector == X86_XCPT_PF)
5896	{
5897	pVmxTransient->fVectoringPF = true;
5898	Log4(("IDT: vcpu[%RU32] Vectoring #PF due to Ext-Int/NMI. uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5899	}
5900	}
5901	}
5902	else if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
5903	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT
5904	\|\| uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI)
5905	{
5906	/*
5907	* If event delivery caused an EPT violation/misconfig or APIC access VM-exit, then the VM-exit
5908	* interruption-information will not be valid as it's not an exception and we end up here. In such cases,
5909	* it is sufficient to reflect the original exception to the guest after handling the VM-exit.
5910	*/
5911	enmReflect = VMXREFLECTXCPT_XCPT;
5912	}
5913
5914	/*
5915	* On CPUs that support Virtual NMIs, if this VM-exit (be it an exception or EPT violation/misconfig etc.) occurred
5916	* while delivering the NMI, we need to clear the block-by-NMI field in the guest interruptibility-state before
5917	* re-delivering the NMI after handling the VM-exit. Otherwise the subsequent VM-entry would fail.
5918	*
5919	* See Intel spec. 30.7.1.2 "Resuming Guest Software after Handling an Exception". See @bugref{7445}.
5920	*/
5921	if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
5922	&& enmReflect == VMXREFLECTXCPT_XCPT
5923	&& (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
5924	&& VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
5925	{
5926	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
5927	}
5928
5929	switch (enmReflect)
5930	{
5931	case VMXREFLECTXCPT_XCPT:
5932	{
5933	Assert( uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_SW_INT
5934	&& uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
5935	&& uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT);
5936
5937	uint32_t u32ErrCode = 0;
5938	if (VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uIdtVectoringInfo))
5939	{
5940	rc2 = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
5941	AssertRCReturn(rc2, rc2);
5942	u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
5943	}
5944
5945	/* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF. See hmR0VmxExitXcptPF(). */
5946	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect);
5947	hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo),
5948	0 /* cbInstr */, u32ErrCode, pMixedCtx->cr2);
5949	rcStrict = VINF_SUCCESS;
5950	Log4(("IDT: vcpu[%RU32] Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->idCpu,
5951	pVCpu->hm.s.Event.u64IntInfo, pVCpu->hm.s.Event.u32ErrCode));
5952
5953	break;
5954	}
5955
5956	case VMXREFLECTXCPT_DF:
5957	{
5958	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect);
5959	hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
5960	rcStrict = VINF_HM_DOUBLE_FAULT;
5961	Log4(("IDT: vcpu[%RU32] Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->idCpu,
5962	pVCpu->hm.s.Event.u64IntInfo, uIdtVector, uExitVector));
5963
5964	break;
5965	}
5966
5967	case VMXREFLECTXCPT_TF:
5968	{
5969	rcStrict = VINF_EM_RESET;
5970	Log4(("IDT: vcpu[%RU32] Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", pVCpu->idCpu, uIdtVector,
5971	uExitVector));
5972	break;
5973	}
5974
5975	case VMXREFLECTXCPT_HANG:
5976	{
5977	rcStrict = VERR_EM_GUEST_CPU_HANG;
5978	break;
5979	}
5980
5981	default:
5982	Assert(rcStrict == VINF_SUCCESS);
5983	break;
5984	}
5985	}
5986	else if ( VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo)
5987	&& VMX_EXIT_INTERRUPTION_INFO_NMI_UNBLOCK_IRET(pVmxTransient->uExitIntInfo)
5988	&& uExitVector != X86_XCPT_DF
5989	&& (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI))
5990	{
5991	/*
5992	* Execution of IRET caused this fault when NMI blocking was in effect (i.e we're in the guest NMI handler).
5993	* We need to set the block-by-NMI field so that NMIs remain blocked until the IRET execution is restarted.
5994	* See Intel spec. 30.7.1.2 "Resuming guest software after handling an exception".
5995	*/
5996	if (!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
5997	{
5998	Log4(("hmR0VmxCheckExitDueToEventDelivery: vcpu[%RU32] Setting VMCPU_FF_BLOCK_NMIS. Valid=%RTbool uExitReason=%u\n",
5999	pVCpu->idCpu, VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo), pVmxTransient->uExitReason));
6000	VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
6001	}
6002	}
6003
6004	Assert( rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_HM_DOUBLE_FAULT
6005	\|\| rcStrict == VINF_EM_RESET \|\| rcStrict == VERR_EM_GUEST_CPU_HANG);
6006	return rcStrict;
6007	}
6008
6009
6010	/**
6011	* Saves the guest's CR0 register from the VMCS into the guest-CPU context.
6012	*
6013	* @returns VBox status code.
6014	* @param pVCpu The cross context virtual CPU structure.
6015	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6016	* out-of-sync. Make sure to update the required fields
6017	* before using them.
6018	*
6019	* @remarks No-long-jump zone!!!
6020	*/
6021	static int hmR0VmxSaveGuestCR0(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6022	{
6023	NOREF(pMixedCtx);
6024
6025	/*
6026	* While in the middle of saving guest-CR0, we could get preempted and re-invoked from the preemption hook,
6027	* see hmR0VmxLeave(). Safer to just make this code non-preemptible.
6028	*/
6029	VMMRZCallRing3Disable(pVCpu);
6030	HM_DISABLE_PREEMPT();
6031
6032	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0))
6033	{
6034	#ifndef DEBUG_bird /** @todo this triggers running bs3-cpu-generated-1.img with --debug-command-line
6035	* and 'dbgc-init' containing:
6036	* sxe "xcpt_de"
6037	* sxe "xcpt_bp"
6038	* sxi "xcpt_gp"
6039	* sxi "xcpt_ss"
6040	* sxi "xcpt_np"
6041	*/
6042	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0));
6043	#endif
6044	uint32_t uVal = 0;
6045	uint32_t uShadow = 0;
6046	int rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &uVal);
6047	rc \|= VMXReadVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uShadow);
6048	AssertRCReturn(rc, rc);
6049
6050	uVal = (uShadow & pVCpu->hm.s.vmx.u32CR0Mask) \| (uVal & ~pVCpu->hm.s.vmx.u32CR0Mask);
6051	CPUMSetGuestCR0(pVCpu, uVal);
6052	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0);
6053	}
6054
6055	HM_RESTORE_PREEMPT();
6056	VMMRZCallRing3Enable(pVCpu);
6057	return VINF_SUCCESS;
6058	}
6059
6060
6061	/**
6062	* Saves the guest's CR4 register from the VMCS into the guest-CPU context.
6063	*
6064	* @returns VBox status code.
6065	* @param pVCpu The cross context virtual CPU structure.
6066	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6067	* out-of-sync. Make sure to update the required fields
6068	* before using them.
6069	*
6070	* @remarks No-long-jump zone!!!
6071	*/
6072	static int hmR0VmxSaveGuestCR4(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6073	{
6074	NOREF(pMixedCtx);
6075
6076	int rc = VINF_SUCCESS;
6077	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4))
6078	{
6079	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR4));
6080	uint32_t uVal = 0;
6081	uint32_t uShadow = 0;
6082	rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &uVal);
6083	rc \|= VMXReadVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uShadow);
6084	AssertRCReturn(rc, rc);
6085
6086	uVal = (uShadow & pVCpu->hm.s.vmx.u32CR4Mask) \| (uVal & ~pVCpu->hm.s.vmx.u32CR4Mask);
6087	CPUMSetGuestCR4(pVCpu, uVal);
6088	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4);
6089	}
6090	return rc;
6091	}
6092
6093
6094	/**
6095	* Saves the guest's RIP register from the VMCS into the guest-CPU context.
6096	*
6097	* @returns VBox status code.
6098	* @param pVCpu The cross context virtual CPU structure.
6099	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6100	* out-of-sync. Make sure to update the required fields
6101	* before using them.
6102	*
6103	* @remarks No-long-jump zone!!!
6104	*/
6105	static int hmR0VmxSaveGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6106	{
6107	int rc = VINF_SUCCESS;
6108	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP))
6109	{
6110	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RIP));
6111	uint64_t u64Val = 0;
6112	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val);
6113	AssertRCReturn(rc, rc);
6114
6115	pMixedCtx->rip = u64Val;
6116	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP);
6117	}
6118	return rc;
6119	}
6120
6121
6122	/**
6123	* Saves the guest's RSP register from the VMCS into the guest-CPU context.
6124	*
6125	* @returns VBox status code.
6126	* @param pVCpu The cross context virtual CPU structure.
6127	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6128	* out-of-sync. Make sure to update the required fields
6129	* before using them.
6130	*
6131	* @remarks No-long-jump zone!!!
6132	*/
6133	static int hmR0VmxSaveGuestRsp(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6134	{
6135	int rc = VINF_SUCCESS;
6136	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RSP))
6137	{
6138	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RSP));
6139	uint64_t u64Val = 0;
6140	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val);
6141	AssertRCReturn(rc, rc);
6142
6143	pMixedCtx->rsp = u64Val;
6144	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RSP);
6145	}
6146	return rc;
6147	}
6148
6149
6150	/**
6151	* Saves the guest's RFLAGS from the VMCS into the guest-CPU context.
6152	*
6153	* @returns VBox status code.
6154	* @param pVCpu The cross context virtual CPU structure.
6155	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6156	* out-of-sync. Make sure to update the required fields
6157	* before using them.
6158	*
6159	* @remarks No-long-jump zone!!!
6160	*/
6161	static int hmR0VmxSaveGuestRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6162	{
6163	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS))
6164	{
6165	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS));
6166	uint32_t uVal = 0;
6167	int rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &uVal);
6168	AssertRCReturn(rc, rc);
6169
6170	pMixedCtx->eflags.u32 = uVal;
6171	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) /* Undo our real-on-v86-mode changes to eflags if necessary. */
6172	{
6173	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
6174	Log4(("Saving real-mode EFLAGS VT-x view=%#RX32\n", pMixedCtx->eflags.u32));
6175
6176	pMixedCtx->eflags.Bits.u1VM = 0;
6177	pMixedCtx->eflags.Bits.u2IOPL = pVCpu->hm.s.vmx.RealMode.Eflags.Bits.u2IOPL;
6178	}
6179
6180	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS);
6181	}
6182	return VINF_SUCCESS;
6183	}
6184
6185
6186	/**
6187	* Wrapper for saving the guest's RIP, RSP and RFLAGS from the VMCS into the
6188	* guest-CPU context.
6189	*/
6190	DECLINLINE(int) hmR0VmxSaveGuestRipRspRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6191	{
6192	int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6193	rc \|= hmR0VmxSaveGuestRsp(pVCpu, pMixedCtx);
6194	rc \|= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
6195	return rc;
6196	}
6197
6198
6199	/**
6200	* Saves the guest's interruptibility-state ("interrupt shadow" as AMD calls it)
6201	* from the guest-state area in the VMCS.
6202	*
6203	* @param pVCpu The cross context virtual CPU structure.
6204	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6205	* out-of-sync. Make sure to update the required fields
6206	* before using them.
6207	*
6208	* @remarks No-long-jump zone!!!
6209	*/
6210	static void hmR0VmxSaveGuestIntrState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6211	{
6212	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_INTR_STATE))
6213	{
6214	uint32_t uIntrState = 0;
6215	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
6216	AssertRC(rc);
6217
6218	if (!uIntrState)
6219	{
6220	if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
6221	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
6222
6223	if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6224	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
6225	}
6226	else
6227	{
6228	if (uIntrState & ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS
6229	\| VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI))
6230	{
6231	rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6232	AssertRC(rc);
6233	rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); /* for hmR0VmxGetGuestIntrState(). */
6234	AssertRC(rc);
6235
6236	EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip);
6237	Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS));
6238	}
6239	else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
6240	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
6241
6242	if (uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI)
6243	{
6244	if (!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6245	VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
6246	}
6247	else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6248	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
6249	}
6250
6251	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_INTR_STATE);
6252	}
6253	}
6254
6255
6256	/**
6257	* Saves the guest's activity state.
6258	*
6259	* @returns VBox status code.
6260	* @param pVCpu The cross context virtual CPU structure.
6261	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6262	* out-of-sync. Make sure to update the required fields
6263	* before using them.
6264	*
6265	* @remarks No-long-jump zone!!!
6266	*/
6267	static int hmR0VmxSaveGuestActivityState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6268	{
6269	NOREF(pMixedCtx);
6270	/* Nothing to do for now until we make use of different guest-CPU activity state. Just update the flag. */
6271	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_ACTIVITY_STATE);
6272	return VINF_SUCCESS;
6273	}
6274
6275
6276	/**
6277	* Saves the guest SYSENTER MSRs (SYSENTER_CS, SYSENTER_EIP, SYSENTER_ESP) from
6278	* the current VMCS into the guest-CPU context.
6279	*
6280	* @returns VBox status code.
6281	* @param pVCpu The cross context virtual CPU structure.
6282	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6283	* out-of-sync. Make sure to update the required fields
6284	* before using them.
6285	*
6286	* @remarks No-long-jump zone!!!
6287	*/
6288	static int hmR0VmxSaveGuestSysenterMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6289	{
6290	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR))
6291	{
6292	Assert(!HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR));
6293	uint32_t u32Val = 0;
6294	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val); AssertRCReturn(rc, rc);
6295	pMixedCtx->SysEnter.cs = u32Val;
6296	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR);
6297	}
6298
6299	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR))
6300	{
6301	Assert(!HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR));
6302	uint64_t u64Val = 0;
6303	int rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, &u64Val); AssertRCReturn(rc, rc);
6304	pMixedCtx->SysEnter.eip = u64Val;
6305	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR);
6306	}
6307	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR))
6308	{
6309	Assert(!HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR));
6310	uint64_t u64Val = 0;
6311	int rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, &u64Val); AssertRCReturn(rc, rc);
6312	pMixedCtx->SysEnter.esp = u64Val;
6313	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR);
6314	}
6315	return VINF_SUCCESS;
6316	}
6317
6318
6319	/**
6320	* Saves the set of guest MSRs (that we restore lazily while leaving VT-x) from
6321	* the CPU back into the guest-CPU context.
6322	*
6323	* @returns VBox status code.
6324	* @param pVCpu The cross context virtual CPU structure.
6325	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6326	* out-of-sync. Make sure to update the required fields
6327	* before using them.
6328	*
6329	* @remarks No-long-jump zone!!!
6330	*/
6331	static int hmR0VmxSaveGuestLazyMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6332	{
6333	/* Since this can be called from our preemption hook it's safer to make the guest-MSRs update non-preemptible. */
6334	VMMRZCallRing3Disable(pVCpu);
6335	HM_DISABLE_PREEMPT();
6336
6337	/* Doing the check here ensures we don't overwrite already-saved guest MSRs from a preemption hook. */
6338	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS))
6339	{
6340	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS));
6341	hmR0VmxLazySaveGuestMsrs(pVCpu, pMixedCtx);
6342	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS);
6343	}
6344
6345	HM_RESTORE_PREEMPT();
6346	VMMRZCallRing3Enable(pVCpu);
6347
6348	return VINF_SUCCESS;
6349	}
6350
6351
6352	/**
6353	* Saves the auto load/store'd guest MSRs from the current VMCS into
6354	* the guest-CPU context.
6355	*
6356	* @returns VBox status code.
6357	* @param pVCpu The cross context virtual CPU structure.
6358	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6359	* out-of-sync. Make sure to update the required fields
6360	* before using them.
6361	*
6362	* @remarks No-long-jump zone!!!
6363	*/
6364	static int hmR0VmxSaveGuestAutoLoadStoreMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6365	{
6366	if (HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS))
6367	return VINF_SUCCESS;
6368
6369	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS));
6370	PVMXAUTOMSR pMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
6371	uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
6372	Log4(("hmR0VmxSaveGuestAutoLoadStoreMsrs: cMsrs=%u\n", cMsrs));
6373	for (uint32_t i = 0; i < cMsrs; i++, pMsr++)
6374	{
6375	switch (pMsr->u32Msr)
6376	{
6377	case MSR_K8_TSC_AUX: CPUMR0SetGuestTscAux(pVCpu, pMsr->u64Value); break;
6378	case MSR_K8_LSTAR: pMixedCtx->msrLSTAR = pMsr->u64Value; break;
6379	case MSR_K6_STAR: pMixedCtx->msrSTAR = pMsr->u64Value; break;
6380	case MSR_K8_SF_MASK: pMixedCtx->msrSFMASK = pMsr->u64Value; break;
6381	case MSR_K8_KERNEL_GS_BASE: pMixedCtx->msrKERNELGSBASE = pMsr->u64Value; break;
6382	case MSR_K6_EFER: /* Nothing to do here since we intercept writes, see hmR0VmxLoadGuestMsrs(). */
6383	break;
6384
6385	default:
6386	{
6387	AssertMsgFailed(("Unexpected MSR in auto-load/store area. uMsr=%#RX32 cMsrs=%u\n", pMsr->u32Msr, cMsrs));
6388	pVCpu->hm.s.u32HMError = pMsr->u32Msr;
6389	return VERR_HM_UNEXPECTED_LD_ST_MSR;
6390	}
6391	}
6392	}
6393
6394	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS);
6395	return VINF_SUCCESS;
6396	}
6397
6398
6399	/**
6400	* Saves the guest control registers from the current VMCS into the guest-CPU
6401	* context.
6402	*
6403	* @returns VBox status code.
6404	* @param pVCpu The cross context virtual CPU structure.
6405	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6406	* out-of-sync. Make sure to update the required fields
6407	* before using them.
6408	*
6409	* @remarks No-long-jump zone!!!
6410	*/
6411	static int hmR0VmxSaveGuestControlRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6412	{
6413	/* Guest CR0. Guest FPU. */
6414	int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6415	AssertRCReturn(rc, rc);
6416
6417	/* Guest CR4. */
6418	rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
6419	AssertRCReturn(rc, rc);
6420
6421	/* Guest CR2 - updated always during the world-switch or in #PF. */
6422	/* Guest CR3. Only changes with Nested Paging. This must be done -after- saving CR0 and CR4 from the guest! */
6423	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR3))
6424	{
6425	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR3));
6426	Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
6427	Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4));
6428
6429	PVM pVM = pVCpu->CTX_SUFF(pVM);
6430	if ( pVM->hm.s.vmx.fUnrestrictedGuest
6431	\|\| ( pVM->hm.s.fNestedPaging
6432	&& CPUMIsGuestPagingEnabledEx(pMixedCtx)))
6433	{
6434	uint64_t u64Val = 0;
6435	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_CR3, &u64Val);
6436	if (pMixedCtx->cr3 != u64Val)
6437	{
6438	CPUMSetGuestCR3(pVCpu, u64Val);
6439	if (VMMRZCallRing3IsEnabled(pVCpu))
6440	{
6441	PGMUpdateCR3(pVCpu, u64Val);
6442	Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6443	}
6444	else
6445	{
6446	/* Set the force flag to inform PGM about it when necessary. It is cleared by PGMUpdateCR3().*/
6447	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
6448	}
6449	}
6450
6451	/* If the guest is in PAE mode, sync back the PDPE's into the guest state. */
6452	if (CPUMIsGuestInPAEModeEx(pMixedCtx)) /* Reads CR0, CR4 and EFER MSR (EFER is always up-to-date). */
6453	{
6454	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &pVCpu->hm.s.aPdpes[0].u);
6455	rc \|= VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &pVCpu->hm.s.aPdpes[1].u);
6456	rc \|= VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &pVCpu->hm.s.aPdpes[2].u);
6457	rc \|= VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &pVCpu->hm.s.aPdpes[3].u);
6458	AssertRCReturn(rc, rc);
6459
6460	if (VMMRZCallRing3IsEnabled(pVCpu))
6461	{
6462	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6463	Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6464	}
6465	else
6466	{
6467	/* Set the force flag to inform PGM about it when necessary. It is cleared by PGMGstUpdatePaePdpes(). */
6468	VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES);
6469	}
6470	}
6471	}
6472
6473	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR3);
6474	}
6475
6476	/*
6477	* Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> hmR0VmxCallRing3Callback()
6478	* -> VMMRZCallRing3Disable() -> hmR0VmxSaveGuestState() -> Set VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp
6479	* -> continue with VM-exit handling -> hmR0VmxSaveGuestControlRegs() and here we are.
6480	*
6481	* The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus
6482	* if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that
6483	* calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should
6484	* -NOT- check if HMVMX_UPDATED_GUEST_CR3 is already set or not!
6485	*
6486	* The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here.
6487	*/
6488	if (VMMRZCallRing3IsEnabled(pVCpu))
6489	{
6490	if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
6491	PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
6492
6493	if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
6494	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6495
6496	Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6497	Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6498	}
6499
6500	return rc;
6501	}
6502
6503
6504	/**
6505	* Reads a guest segment register from the current VMCS into the guest-CPU
6506	* context.
6507	*
6508	* @returns VBox status code.
6509	* @param pVCpu The cross context virtual CPU structure.
6510	* @param idxSel Index of the selector in the VMCS.
6511	* @param idxLimit Index of the segment limit in the VMCS.
6512	* @param idxBase Index of the segment base in the VMCS.
6513	* @param idxAccess Index of the access rights of the segment in the VMCS.
6514	* @param pSelReg Pointer to the segment selector.
6515	*
6516	* @remarks No-long-jump zone!!!
6517	* @remarks Never call this function directly!!! Use the VMXLOCAL_READ_SEG()
6518	* macro as that takes care of whether to read from the VMCS cache or
6519	* not.
6520	*/
6521	DECLINLINE(int) hmR0VmxReadSegmentReg(PVMCPU pVCpu, uint32_t idxSel, uint32_t idxLimit, uint32_t idxBase, uint32_t idxAccess,
6522	PCPUMSELREG pSelReg)
6523	{
6524	NOREF(pVCpu);
6525
6526	uint32_t u32Val = 0;
6527	int rc = VMXReadVmcs32(idxSel, &u32Val);
6528	AssertRCReturn(rc, rc);
6529	pSelReg->Sel = (uint16_t)u32Val;
6530	pSelReg->ValidSel = (uint16_t)u32Val;
6531	pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
6532
6533	rc = VMXReadVmcs32(idxLimit, &u32Val);
6534	AssertRCReturn(rc, rc);
6535	pSelReg->u32Limit = u32Val;
6536
6537	uint64_t u64Val = 0;
6538	rc = VMXReadVmcsGstNByIdxVal(idxBase, &u64Val);
6539	AssertRCReturn(rc, rc);
6540	pSelReg->u64Base = u64Val;
6541
6542	rc = VMXReadVmcs32(idxAccess, &u32Val);
6543	AssertRCReturn(rc, rc);
6544	pSelReg->Attr.u = u32Val;
6545
6546	/*
6547	* If VT-x marks the segment as unusable, most other bits remain undefined:
6548	* - For CS the L, D and G bits have meaning.
6549	* - For SS the DPL has meaning (it -is- the CPL for Intel and VBox).
6550	* - For the remaining data segments no bits are defined.
6551	*
6552	* The present bit and the unusable bit has been observed to be set at the
6553	* same time (the selector was supposed to be invalid as we started executing
6554	* a V8086 interrupt in ring-0).
6555	*
6556	* What should be important for the rest of the VBox code, is that the P bit is
6557	* cleared. Some of the other VBox code recognizes the unusable bit, but
6558	* AMD-V certainly don't, and REM doesn't really either. So, to be on the
6559	* safe side here, we'll strip off P and other bits we don't care about. If
6560	* any code breaks because Attr.u != 0 when Sel < 4, it should be fixed.
6561	*
6562	* See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
6563	*/
6564	if (pSelReg->Attr.u & X86DESCATTR_UNUSABLE)
6565	{
6566	Assert(idxSel != VMX_VMCS16_GUEST_TR_SEL); /* TR is the only selector that can never be unusable. */
6567
6568	/* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */
6569	pSelReg->Attr.u &= X86DESCATTR_UNUSABLE \| X86DESCATTR_L \| X86DESCATTR_D \| X86DESCATTR_G
6570	\| X86DESCATTR_DPL \| X86DESCATTR_TYPE \| X86DESCATTR_DT;
6571
6572	Log4(("hmR0VmxReadSegmentReg: Unusable idxSel=%#x attr=%#x -> %#x\n", idxSel, u32Val, pSelReg->Attr.u));
6573	#ifdef DEBUG_bird
6574	AssertMsg((u32Val & ~X86DESCATTR_P) == pSelReg->Attr.u,
6575	("%#x: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n",
6576	idxSel, u32Val, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit));
6577	#endif
6578	}
6579	return VINF_SUCCESS;
6580	}
6581
6582
6583	#ifdef VMX_USE_CACHED_VMCS_ACCESSES
6584	# define VMXLOCAL_READ_SEG(Sel, CtxSel) \
6585	hmR0VmxReadSegmentReg(pVCpu, VMX_VMCS16_GUEST_##Sel##_SEL, VMX_VMCS32_GUEST_##Sel##_LIMIT, \
6586	VMX_VMCS_GUEST_##Sel##_BASE_CACHE_IDX, VMX_VMCS32_GUEST_##Sel##_ACCESS_RIGHTS, &pMixedCtx->CtxSel)
6587	#else
6588	# define VMXLOCAL_READ_SEG(Sel, CtxSel) \
6589	hmR0VmxReadSegmentReg(pVCpu, VMX_VMCS16_GUEST_##Sel##_SEL, VMX_VMCS32_GUEST_##Sel##_LIMIT, \
6590	VMX_VMCS_GUEST_##Sel##_BASE, VMX_VMCS32_GUEST_##Sel##_ACCESS_RIGHTS, &pMixedCtx->CtxSel)
6591	#endif
6592
6593
6594	/**
6595	* Saves the guest segment registers from the current VMCS into the guest-CPU
6596	* context.
6597	*
6598	* @returns VBox status code.
6599	* @param pVCpu The cross context virtual CPU structure.
6600	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6601	* out-of-sync. Make sure to update the required fields
6602	* before using them.
6603	*
6604	* @remarks No-long-jump zone!!!
6605	*/
6606	static int hmR0VmxSaveGuestSegmentRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6607	{
6608	/* Guest segment registers. */
6609	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SEGMENT_REGS))
6610	{
6611	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS));
6612	int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6613	AssertRCReturn(rc, rc);
6614
6615	rc = VMXLOCAL_READ_SEG(CS, cs);
6616	rc \|= VMXLOCAL_READ_SEG(SS, ss);
6617	rc \|= VMXLOCAL_READ_SEG(DS, ds);
6618	rc \|= VMXLOCAL_READ_SEG(ES, es);
6619	rc \|= VMXLOCAL_READ_SEG(FS, fs);
6620	rc \|= VMXLOCAL_READ_SEG(GS, gs);
6621	AssertRCReturn(rc, rc);
6622
6623	/* Restore segment attributes for real-on-v86 mode hack. */
6624	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
6625	{
6626	pMixedCtx->cs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrCS.u;
6627	pMixedCtx->ss.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrSS.u;
6628	pMixedCtx->ds.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrDS.u;
6629	pMixedCtx->es.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrES.u;
6630	pMixedCtx->fs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrFS.u;
6631	pMixedCtx->gs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrGS.u;
6632	}
6633	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SEGMENT_REGS);
6634	}
6635
6636	return VINF_SUCCESS;
6637	}
6638
6639
6640	/**
6641	* Saves the guest descriptor table registers and task register from the current
6642	* VMCS into the guest-CPU context.
6643	*
6644	* @returns VBox status code.
6645	* @param pVCpu The cross context virtual CPU structure.
6646	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6647	* out-of-sync. Make sure to update the required fields
6648	* before using them.
6649	*
6650	* @remarks No-long-jump zone!!!
6651	*/
6652	static int hmR0VmxSaveGuestTableRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6653	{
6654	int rc = VINF_SUCCESS;
6655
6656	/* Guest LDTR. */
6657	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LDTR))
6658	{
6659	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LDTR));
6660	rc = VMXLOCAL_READ_SEG(LDTR, ldtr);
6661	AssertRCReturn(rc, rc);
6662	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LDTR);
6663	}
6664
6665	/* Guest GDTR. */
6666	uint64_t u64Val = 0;
6667	uint32_t u32Val = 0;
6668	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_GDTR))
6669	{
6670	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_GDTR));
6671	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
6672	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRCReturn(rc, rc);
6673	pMixedCtx->gdtr.pGdt = u64Val;
6674	pMixedCtx->gdtr.cbGdt = u32Val;
6675	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_GDTR);
6676	}
6677
6678	/* Guest IDTR. */
6679	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_IDTR))
6680	{
6681	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_IDTR));
6682	rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
6683	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRCReturn(rc, rc);
6684	pMixedCtx->idtr.pIdt = u64Val;
6685	pMixedCtx->idtr.cbIdt = u32Val;
6686	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_IDTR);
6687	}
6688
6689	/* Guest TR. */
6690	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_TR))
6691	{
6692	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_TR));
6693	rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6694	AssertRCReturn(rc, rc);
6695
6696	/* For real-mode emulation using virtual-8086 mode we have the fake TSS (pRealModeTSS) in TR, don't save the fake one. */
6697	if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
6698	{
6699	rc = VMXLOCAL_READ_SEG(TR, tr);
6700	AssertRCReturn(rc, rc);
6701	}
6702	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_TR);
6703	}
6704	return rc;
6705	}
6706
6707	#undef VMXLOCAL_READ_SEG
6708
6709
6710	/**
6711	* Saves the guest debug-register DR7 from the current VMCS into the guest-CPU
6712	* context.
6713	*
6714	* @returns VBox status code.
6715	* @param pVCpu The cross context virtual CPU structure.
6716	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6717	* out-of-sync. Make sure to update the required fields
6718	* before using them.
6719	*
6720	* @remarks No-long-jump zone!!!
6721	*/
6722	static int hmR0VmxSaveGuestDR7(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6723	{
6724	if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_DR7))
6725	{
6726	if (!pVCpu->hm.s.fUsingHyperDR7)
6727	{
6728	/* Upper 32-bits are always zero. See Intel spec. 2.7.3 "Loading and Storing Debug Registers". */
6729	uint32_t u32Val;
6730	int rc = VMXReadVmcs32(VMX_VMCS_GUEST_DR7, &u32Val); AssertRCReturn(rc, rc);
6731	pMixedCtx->dr[7] = u32Val;
6732	}
6733
6734	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_DR7);
6735	}
6736	return VINF_SUCCESS;
6737	}
6738
6739
6740	/**
6741	* Saves the guest APIC state from the current VMCS into the guest-CPU context.
6742	*
6743	* @returns VBox status code.
6744	* @param pVCpu The cross context virtual CPU structure.
6745	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6746	* out-of-sync. Make sure to update the required fields
6747	* before using them.
6748	*
6749	* @remarks No-long-jump zone!!!
6750	*/
6751	static int hmR0VmxSaveGuestApicState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6752	{
6753	NOREF(pMixedCtx);
6754
6755	/* Updating TPR is already done in hmR0VmxPostRunGuest(). Just update the flag. */
6756	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_APIC_STATE);
6757	return VINF_SUCCESS;
6758	}
6759
6760
6761	/**
6762	* Saves the entire guest state from the currently active VMCS into the
6763	* guest-CPU context.
6764	*
6765	* This essentially VMREADs all guest-data.
6766	*
6767	* @returns VBox status code.
6768	* @param pVCpu The cross context virtual CPU structure.
6769	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
6770	* out-of-sync. Make sure to update the required fields
6771	* before using them.
6772	*/
6773	static int hmR0VmxSaveGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6774	{
6775	Assert(pVCpu);
6776	Assert(pMixedCtx);
6777
6778	if (HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL)
6779	return VINF_SUCCESS;
6780
6781	/* Though we can longjmp to ring-3 due to log-flushes here and get recalled
6782	again on the ring-3 callback path, there is no real need to. */
6783	if (VMMRZCallRing3IsEnabled(pVCpu))
6784	VMMR0LogFlushDisable(pVCpu);
6785	else
6786	Assert(VMMR0IsLogFlushDisabled(pVCpu));
6787	Log4Func(("vcpu[%RU32]\n", pVCpu->idCpu));
6788
6789	int rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
6790	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestRipRspRflags failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6791
6792	rc = hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6793	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestControlRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6794
6795	rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
6796	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestSegmentRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6797
6798	rc = hmR0VmxSaveGuestTableRegs(pVCpu, pMixedCtx);
6799	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestTableRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6800
6801	rc = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
6802	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestDR7 failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6803
6804	rc = hmR0VmxSaveGuestSysenterMsrs(pVCpu, pMixedCtx);
6805	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestSysenterMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6806
6807	rc = hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
6808	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestLazyMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6809
6810	rc = hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
6811	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestAutoLoadStoreMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6812
6813	rc = hmR0VmxSaveGuestActivityState(pVCpu, pMixedCtx);
6814	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestActivityState failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6815
6816	rc = hmR0VmxSaveGuestApicState(pVCpu, pMixedCtx);
6817	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestApicState failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6818
6819	AssertMsg(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL,
6820	("Missed guest state bits while saving state; missing %RX32 (got %RX32, want %RX32) - check log for any previous errors!\n",
6821	HMVMX_UPDATED_GUEST_ALL ^ HMVMXCPU_GST_VALUE(pVCpu), HMVMXCPU_GST_VALUE(pVCpu), HMVMX_UPDATED_GUEST_ALL));
6822
6823	if (VMMRZCallRing3IsEnabled(pVCpu))
6824	VMMR0LogFlushEnable(pVCpu);
6825
6826	return VINF_SUCCESS;
6827	}
6828
6829
6830	/**
6831	* Saves basic guest registers needed for IEM instruction execution.
6832	*
6833	* @returns VBox status code (OR-able).
6834	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
6835	* @param pMixedCtx Pointer to the CPU context of the guest.
6836	* @param fMemory Whether the instruction being executed operates on
6837	* memory or not. Only CR0 is synced up if clear.
6838	* @param fNeedRsp Need RSP (any instruction working on GPRs or stack).
6839	*/
6840	static int hmR0VmxSaveGuestRegsForIemExec(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fMemory, bool fNeedRsp)
6841	{
6842	/*
6843	* We assume all general purpose registers other than RSP are available.
6844	*
6845	* - RIP is a must, as it will be incremented or otherwise changed.
6846	* - RFLAGS are always required to figure the CPL.
6847	* - RSP isn't always required, however it's a GPR, so frequently required.
6848	* - SS and CS are the only segment register needed if IEM doesn't do memory
6849	* access (CPL + 16/32/64-bit mode), but we can only get all segment registers.
6850	* - CR0 is always required by IEM for the CPL, while CR3 and CR4 will only
6851	* be required for memory accesses.
6852	*
6853	* Note! Before IEM dispatches an exception, it will call us to sync in everything.
6854	*/
6855	int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6856	rc \|= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
6857	if (fNeedRsp)
6858	rc \|= hmR0VmxSaveGuestRsp(pVCpu, pMixedCtx);
6859	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
6860	if (!fMemory)
6861	rc \|= hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6862	else
6863	rc \|= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6864	AssertRCReturn(rc, rc);
6865	return rc;
6866	}
6867
6868
6869	/**
6870	* Ensures that we've got a complete basic guest-context.
6871	*
6872	* This excludes the FPU, SSE, AVX, and similar extended state. The interface
6873	* is for the interpreter.
6874	*
6875	* @returns VBox status code.
6876	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
6877	* @param pMixedCtx Pointer to the guest-CPU context which may have data
6878	* needing to be synced in.
6879	* @thread EMT(pVCpu)
6880	*/
6881	VMMR0_INT_DECL(int) HMR0EnsureCompleteBasicContext(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6882	{
6883	/* Note! Since this is only applicable to VT-x, the implementation is placed
6884	in the VT-x part of the sources instead of the generic stuff. */
6885	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported)
6886	{
6887	int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
6888	/*
6889	* For now, imply that the caller might change everything too. Do this after
6890	* saving the guest state so as to not trigger assertions.
6891	*/
6892	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
6893	return rc;
6894	}
6895	return VINF_SUCCESS;
6896	}
6897
6898
6899	/**
6900	* Check per-VM and per-VCPU force flag actions that require us to go back to
6901	* ring-3 for one reason or another.
6902	*
6903	* @returns Strict VBox status code (i.e. informational status codes too)
6904	* @retval VINF_SUCCESS if we don't have any actions that require going back to
6905	* ring-3.
6906	* @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
6907	* @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
6908	* interrupts)
6909	* @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
6910	* all EMTs to be in ring-3.
6911	* @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
6912	* @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
6913	* to the EM loop.
6914	*
6915	* @param pVM The cross context VM structure.
6916	* @param pVCpu The cross context virtual CPU structure.
6917	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
6918	* out-of-sync. Make sure to update the required fields
6919	* before using them.
6920	* @param fStepping Running in hmR0VmxRunGuestCodeStep().
6921	*/
6922	static VBOXSTRICTRC hmR0VmxCheckForceFlags(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fStepping)
6923	{
6924	Assert(VMMRZCallRing3IsEnabled(pVCpu));
6925
6926	/*
6927	* Anything pending? Should be more likely than not if we're doing a good job.
6928	*/
6929	if ( !fStepping
6930	? !VM_FF_IS_PENDING(pVM, VM_FF_HP_R0_PRE_HM_MASK)
6931	&& !VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HP_R0_PRE_HM_MASK)
6932	: !VM_FF_IS_PENDING(pVM, VM_FF_HP_R0_PRE_HM_STEP_MASK)
6933	&& !VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
6934	return VINF_SUCCESS;
6935
6936	/* We need the control registers now, make sure the guest-CPU context is updated. */
6937	int rc3 = hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6938	AssertRCReturn(rc3, rc3);
6939
6940	/* Pending HM CR3 sync. */
6941	if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
6942	{
6943	int rc2 = PGMUpdateCR3(pVCpu, pMixedCtx->cr3);
6944	AssertMsgReturn(rc2 == VINF_SUCCESS \|\| rc2 == VINF_PGM_SYNC_CR3,
6945	("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS);
6946	Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6947	}
6948
6949	/* Pending HM PAE PDPEs. */
6950	if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
6951	{
6952	PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6953	Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6954	}
6955
6956	/* Pending PGM C3 sync. */
6957	if (VMCPU_FF_IS_PENDING(pVCpu,VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
6958	{
6959	VBOXSTRICTRC rcStrict2 = PGMSyncCR3(pVCpu, pMixedCtx->cr0, pMixedCtx->cr3, pMixedCtx->cr4,
6960	VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
6961	if (rcStrict2 != VINF_SUCCESS)
6962	{
6963	AssertRC(VBOXSTRICTRC_VAL(rcStrict2));
6964	Log4(("hmR0VmxCheckForceFlags: PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", VBOXSTRICTRC_VAL(rcStrict2)));
6965	return rcStrict2;
6966	}
6967	}
6968
6969	/* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
6970	if ( VM_FF_IS_PENDING(pVM, VM_FF_HM_TO_R3_MASK)
6971	\|\| VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
6972	{
6973	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
6974	int rc2 = RT_UNLIKELY(VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_NO_MEMORY : VINF_EM_RAW_TO_R3;
6975	Log4(("hmR0VmxCheckForceFlags: HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc2));
6976	return rc2;
6977	}
6978
6979	/* Pending VM request packets, such as hardware interrupts. */
6980	if ( VM_FF_IS_PENDING(pVM, VM_FF_REQUEST)
6981	\|\| VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_REQUEST))
6982	{
6983	Log4(("hmR0VmxCheckForceFlags: Pending VM request forcing us back to ring-3\n"));
6984	return VINF_EM_PENDING_REQUEST;
6985	}
6986
6987	/* Pending PGM pool flushes. */
6988	if (VM_FF_IS_PENDING(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
6989	{
6990	Log4(("hmR0VmxCheckForceFlags: PGM pool flush pending forcing us back to ring-3\n"));
6991	return VINF_PGM_POOL_FLUSH_PENDING;
6992	}
6993
6994	/* Pending DMA requests. */
6995	if (VM_FF_IS_PENDING(pVM, VM_FF_PDM_DMA))
6996	{
6997	Log4(("hmR0VmxCheckForceFlags: Pending DMA request forcing us back to ring-3\n"));
6998	return VINF_EM_RAW_TO_R3;
6999	}
7000
7001	return VINF_SUCCESS;
7002	}
7003
7004
7005	/**
7006	* Converts any TRPM trap into a pending HM event. This is typically used when
7007	* entering from ring-3 (not longjmp returns).
7008	*
7009	* @param pVCpu The cross context virtual CPU structure.
7010	*/
7011	static void hmR0VmxTrpmTrapToPendingEvent(PVMCPU pVCpu)
7012	{
7013	Assert(TRPMHasTrap(pVCpu));
7014	Assert(!pVCpu->hm.s.Event.fPending);
7015
7016	uint8_t uVector;
7017	TRPMEVENT enmTrpmEvent;
7018	RTGCUINT uErrCode;
7019	RTGCUINTPTR GCPtrFaultAddress;
7020	uint8_t cbInstr;
7021
7022	int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr);
7023	AssertRC(rc);
7024
7025	/* Refer Intel spec. 24.8.3 "VM-entry Controls for Event Injection" for the format of u32IntInfo. */
7026	uint32_t u32IntInfo = uVector \| VMX_EXIT_INTERRUPTION_INFO_VALID;
7027	if (enmTrpmEvent == TRPM_TRAP)
7028	{
7029	switch (uVector)
7030	{
7031	case X86_XCPT_NMI:
7032	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7033	break;
7034
7035	case X86_XCPT_BP:
7036	case X86_XCPT_OF:
7037	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7038	break;
7039
7040	case X86_XCPT_PF:
7041	case X86_XCPT_DF:
7042	case X86_XCPT_TS:
7043	case X86_XCPT_NP:
7044	case X86_XCPT_SS:
7045	case X86_XCPT_GP:
7046	case X86_XCPT_AC:
7047	u32IntInfo \|= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7048	/* fall thru */
7049	default:
7050	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7051	break;
7052	}
7053	}
7054	else if (enmTrpmEvent == TRPM_HARDWARE_INT)
7055	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7056	else if (enmTrpmEvent == TRPM_SOFTWARE_INT)
7057	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7058	else
7059	AssertMsgFailed(("Invalid TRPM event type %d\n", enmTrpmEvent));
7060
7061	rc = TRPMResetTrap(pVCpu);
7062	AssertRC(rc);
7063	Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n",
7064	u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress));
7065
7066	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress);
7067	}
7068
7069
7070	/**
7071	* Converts the pending HM event into a TRPM trap.
7072	*
7073	* @param pVCpu The cross context virtual CPU structure.
7074	*/
7075	static void hmR0VmxPendingEventToTrpmTrap(PVMCPU pVCpu)
7076	{
7077	Assert(pVCpu->hm.s.Event.fPending);
7078
7079	uint32_t uVectorType = VMX_IDT_VECTORING_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
7080	uint32_t uVector = VMX_IDT_VECTORING_INFO_VECTOR(pVCpu->hm.s.Event.u64IntInfo);
7081	bool fErrorCodeValid = VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVCpu->hm.s.Event.u64IntInfo);
7082	uint32_t uErrorCode = pVCpu->hm.s.Event.u32ErrCode;
7083
7084	/* If a trap was already pending, we did something wrong! */
7085	Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo /, NULL / pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
7086
7087	TRPMEVENT enmTrapType;
7088	switch (uVectorType)
7089	{
7090	case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT:
7091	enmTrapType = TRPM_HARDWARE_INT;
7092	break;
7093
7094	case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
7095	enmTrapType = TRPM_SOFTWARE_INT;
7096	break;
7097
7098	case VMX_IDT_VECTORING_INFO_TYPE_NMI:
7099	case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
7100	case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT: /* #BP and #OF */
7101	case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT:
7102	enmTrapType = TRPM_TRAP;
7103	break;
7104
7105	default:
7106	AssertMsgFailed(("Invalid trap type %#x\n", uVectorType));
7107	enmTrapType = TRPM_32BIT_HACK;
7108	break;
7109	}
7110
7111	Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType));
7112
7113	int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
7114	AssertRC(rc);
7115
7116	if (fErrorCodeValid)
7117	TRPMSetErrorCode(pVCpu, uErrorCode);
7118
7119	if ( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
7120	&& uVector == X86_XCPT_PF)
7121	{
7122	TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
7123	}
7124	else if ( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
7125	\|\| uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
7126	\|\| uVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT)
7127	{
7128	AssertMsg( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
7129	\|\| (uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF),
7130	("Invalid vector: uVector=%#x uVectorType=%#x\n", uVector, uVectorType));
7131	TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
7132	}
7133
7134	/* Clear any pending events from the VMCS. */
7135	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0); AssertRC(rc);
7136	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, 0); AssertRC(rc);
7137
7138	/* We're now done converting the pending event. */
7139	pVCpu->hm.s.Event.fPending = false;
7140	}
7141
7142
7143	/**
7144	* Does the necessary state syncing before returning to ring-3 for any reason
7145	* (longjmp, preemption, voluntary exits to ring-3) from VT-x.
7146	*
7147	* @returns VBox status code.
7148	* @param pVCpu The cross context virtual CPU structure.
7149	* @param pMixedCtx Pointer to the guest-CPU context. The data may
7150	* be out-of-sync. Make sure to update the required
7151	* fields before using them.
7152	* @param fSaveGuestState Whether to save the guest state or not.
7153	*
7154	* @remarks No-long-jmp zone!!!
7155	*/
7156	static int hmR0VmxLeave(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fSaveGuestState)
7157	{
7158	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
7159	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
7160
7161	RTCPUID idCpu = RTMpCpuId();
7162	Log4Func(("HostCpuId=%u\n", idCpu));
7163
7164	/*
7165	* !!! IMPORTANT !!!
7166	* If you modify code here, check whether hmR0VmxCallRing3Callback() needs to be updated too.
7167	*/
7168
7169	/* Save the guest state if necessary. */
7170	if ( fSaveGuestState
7171	&& HMVMXCPU_GST_VALUE(pVCpu) != HMVMX_UPDATED_GUEST_ALL)
7172	{
7173	int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
7174	AssertRCReturn(rc, rc);
7175	Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
7176	}
7177
7178	/* Restore host FPU state if necessary and resync on next R0 reentry .*/
7179	if (CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu))
7180	{
7181	/* We shouldn't reload CR0 without saving it first. */
7182	if (!fSaveGuestState)
7183	{
7184	int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
7185	AssertRCReturn(rc, rc);
7186	}
7187	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
7188	}
7189
7190	/* Restore host debug registers if necessary and resync on next R0 reentry. */
7191	#ifdef VBOX_STRICT
7192	if (CPUMIsHyperDebugStateActive(pVCpu))
7193	Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT);
7194	#endif
7195	if (CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */))
7196	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
7197	Assert(!CPUMIsGuestDebugStateActive(pVCpu) && !CPUMIsGuestDebugStateActivePending(pVCpu));
7198	Assert(!CPUMIsHyperDebugStateActive(pVCpu) && !CPUMIsHyperDebugStateActivePending(pVCpu));
7199
7200	#if HC_ARCH_BITS == 64
7201	/* Restore host-state bits that VT-x only restores partially. */
7202	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
7203	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
7204	{
7205	Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags, idCpu));
7206	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
7207	}
7208	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
7209	#endif
7210
7211	/* Restore the lazy host MSRs as we're leaving VT-x context. */
7212	if (pVCpu->hm.s.vmx.fLazyMsrs)
7213	{
7214	/* We shouldn't reload the guest MSRs without saving it first. */
7215	if (!fSaveGuestState)
7216	{
7217	int rc = hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
7218	AssertRCReturn(rc, rc);
7219	}
7220	Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS));
7221	hmR0VmxLazyRestoreHostMsrs(pVCpu);
7222	Assert(!pVCpu->hm.s.vmx.fLazyMsrs);
7223	}
7224
7225	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
7226	pVCpu->hm.s.vmx.fUpdatedHostMsrs = false;
7227
7228	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
7229	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatLoadGuestState);
7230	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit1);
7231	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit2);
7232	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO);
7233	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx);
7234	STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi);
7235	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
7236
7237	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
7238
7239	/** @todo This partially defeats the purpose of having preemption hooks.
7240	* The problem is, deregistering the hooks should be moved to a place that
7241	* lasts until the EMT is about to be destroyed not everytime while leaving HM
7242	* context.
7243	*/
7244	if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_ACTIVE)
7245	{
7246	int rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
7247	AssertRCReturn(rc, rc);
7248
7249	pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
7250	Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu));
7251	}
7252	Assert(!(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_LAUNCHED));
7253	NOREF(idCpu);
7254
7255	return VINF_SUCCESS;
7256	}
7257
7258
7259	/**
7260	* Leaves the VT-x session.
7261	*
7262	* @returns VBox status code.
7263	* @param pVCpu The cross context virtual CPU structure.
7264	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7265	* out-of-sync. Make sure to update the required fields
7266	* before using them.
7267	*
7268	* @remarks No-long-jmp zone!!!
7269	*/
7270	DECLINLINE(int) hmR0VmxLeaveSession(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7271	{
7272	HM_DISABLE_PREEMPT();
7273	HMVMX_ASSERT_CPU_SAFE();
7274	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
7275	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
7276
7277	/* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
7278	and done this from the VMXR0ThreadCtxCallback(). */
7279	if (!pVCpu->hm.s.fLeaveDone)
7280	{
7281	int rc2 = hmR0VmxLeave(pVCpu, pMixedCtx, true /* fSaveGuestState */);
7282	AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT(), rc2);
7283	pVCpu->hm.s.fLeaveDone = true;
7284	}
7285	Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
7286
7287	/*
7288	* !!! IMPORTANT !!!
7289	* If you modify code here, make sure to check whether hmR0VmxCallRing3Callback() needs to be updated too.
7290	*/
7291
7292	/* Deregister hook now that we've left HM context before re-enabling preemption. */
7293	/** @todo Deregistering here means we need to VMCLEAR always
7294	* (longjmp/exit-to-r3) in VT-x which is not efficient. */
7295	/** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
7296	VMMR0ThreadCtxHookDisable(pVCpu);
7297
7298	/* Leave HM context. This takes care of local init (term). */
7299	int rc = HMR0LeaveCpu(pVCpu);
7300
7301	HM_RESTORE_PREEMPT();
7302	return rc;
7303	}
7304
7305
7306	/**
7307	* Does the necessary state syncing before doing a longjmp to ring-3.
7308	*
7309	* @returns VBox status code.
7310	* @param pVCpu The cross context virtual CPU structure.
7311	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7312	* out-of-sync. Make sure to update the required fields
7313	* before using them.
7314	*
7315	* @remarks No-long-jmp zone!!!
7316	*/
7317	DECLINLINE(int) hmR0VmxLongJmpToRing3(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7318	{
7319	return hmR0VmxLeaveSession(pVCpu, pMixedCtx);
7320	}
7321
7322
7323	/**
7324	* Take necessary actions before going back to ring-3.
7325	*
7326	* An action requires us to go back to ring-3. This function does the necessary
7327	* steps before we can safely return to ring-3. This is not the same as longjmps
7328	* to ring-3, this is voluntary and prepares the guest so it may continue
7329	* executing outside HM (recompiler/IEM).
7330	*
7331	* @returns VBox status code.
7332	* @param pVM The cross context VM structure.
7333	* @param pVCpu The cross context virtual CPU structure.
7334	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7335	* out-of-sync. Make sure to update the required fields
7336	* before using them.
7337	* @param rcExit The reason for exiting to ring-3. Can be
7338	* VINF_VMM_UNKNOWN_RING3_CALL.
7339	*/
7340	static int hmR0VmxExitToRing3(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, VBOXSTRICTRC rcExit)
7341	{
7342	Assert(pVM);
7343	Assert(pVCpu);
7344	Assert(pMixedCtx);
7345	HMVMX_ASSERT_PREEMPT_SAFE();
7346
7347	if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR))
7348	{
7349	VMXGetActivatedVmcs(&pVCpu->hm.s.vmx.LastError.u64VMCSPhys);
7350	pVCpu->hm.s.vmx.LastError.u32VMCSRevision = (uint32_t )pVCpu->hm.s.vmx.pvVmcs;
7351	pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
7352	/* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */
7353	}
7354
7355	/* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
7356	VMMRZCallRing3Disable(pVCpu);
7357	Log4(("hmR0VmxExitToRing3: pVCpu=%p idCpu=%RU32 rcExit=%d\n", pVCpu, pVCpu->idCpu, VBOXSTRICTRC_VAL(rcExit)));
7358
7359	/* We need to do this only while truly exiting the "inner loop" back to ring-3 and -not- for any longjmp to ring3. */
7360	if (pVCpu->hm.s.Event.fPending)
7361	{
7362	hmR0VmxPendingEventToTrpmTrap(pVCpu);
7363	Assert(!pVCpu->hm.s.Event.fPending);
7364	}
7365
7366	/* Clear interrupt-window and NMI-window controls as we re-evaluate it when we return from ring-3. */
7367	hmR0VmxClearIntNmiWindowsVmcs(pVCpu);
7368
7369	/* If we're emulating an instruction, we shouldn't have any TRPM traps pending
7370	and if we're injecting an event we should have a TRPM trap pending. */
7371	AssertMsg(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT \|\| TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
7372	#ifndef DEBUG_bird /* Triggered after firing an NMI against NT4SP1, possibly a tripple fault in progress. */
7373	AssertMsg(rcExit != VINF_EM_RAW_EMULATE_INSTR \|\| !TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
7374	#endif
7375
7376	/* Save guest state and restore host state bits. */
7377	int rc = hmR0VmxLeaveSession(pVCpu, pMixedCtx);
7378	AssertRCReturn(rc, rc);
7379	STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
7380	/* Thread-context hooks are unregistered at this point!!! */
7381
7382	/* Sync recompiler state. */
7383	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
7384	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
7385	\| CPUM_CHANGED_LDTR
7386	\| CPUM_CHANGED_GDTR
7387	\| CPUM_CHANGED_IDTR
7388	\| CPUM_CHANGED_TR
7389	\| CPUM_CHANGED_HIDDEN_SEL_REGS);
7390	Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
7391	if ( pVM->hm.s.fNestedPaging
7392	&& CPUMIsGuestPagingEnabledEx(pMixedCtx))
7393	{
7394	CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
7395	}
7396
7397	Assert(!pVCpu->hm.s.fClearTrapFlag);
7398
7399	/* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
7400	if (rcExit != VINF_EM_RAW_INTERRUPT)
7401	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
7402
7403	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
7404
7405	/* We do -not- want any longjmp notifications after this! We must return to ring-3 ASAP. */
7406	VMMRZCallRing3RemoveNotification(pVCpu);
7407	VMMRZCallRing3Enable(pVCpu);
7408
7409	return rc;
7410	}
7411
7412
7413	/**
7414	* VMMRZCallRing3() callback wrapper which saves the guest state before we
7415	* longjump to ring-3 and possibly get preempted.
7416	*
7417	* @returns VBox status code.
7418	* @param pVCpu The cross context virtual CPU structure.
7419	* @param enmOperation The operation causing the ring-3 longjump.
7420	* @param pvUser Opaque pointer to the guest-CPU context. The data
7421	* may be out-of-sync. Make sure to update the required
7422	* fields before using them.
7423	*/
7424	static DECLCALLBACK(int) hmR0VmxCallRing3Callback(PVMCPU pVCpu, VMMCALLRING3 enmOperation, void *pvUser)
7425	{
7426	if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION)
7427	{
7428	/*
7429	* !!! IMPORTANT !!!
7430	* If you modify code here, check whether hmR0VmxLeave() and hmR0VmxLeaveSession() needs to be updated too.
7431	* This is a stripped down version which gets out ASAP, trying to not trigger any further assertions.
7432	*/
7433	VMMRZCallRing3RemoveNotification(pVCpu);
7434	VMMRZCallRing3Disable(pVCpu);
7435	RTTHREADPREEMPTSTATE PreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
7436	RTThreadPreemptDisable(&PreemptState);
7437
7438	CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
7439	CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
7440
7441	#if HC_ARCH_BITS == 64
7442	/* Restore host-state bits that VT-x only restores partially. */
7443	if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
7444	&& (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
7445	VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
7446	pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
7447	#endif
7448	/* Restore the lazy host MSRs as we're leaving VT-x context. */
7449	if (pVCpu->hm.s.vmx.fLazyMsrs)
7450	hmR0VmxLazyRestoreHostMsrs(pVCpu);
7451
7452	/* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
7453	pVCpu->hm.s.vmx.fUpdatedHostMsrs = false;
7454	VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
7455	if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_ACTIVE)
7456	{
7457	VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
7458	pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
7459	}
7460
7461	/** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
7462	VMMR0ThreadCtxHookDisable(pVCpu);
7463	HMR0LeaveCpu(pVCpu);
7464	RTThreadPreemptRestore(&PreemptState);
7465	return VINF_SUCCESS;
7466	}
7467
7468	Assert(pVCpu);
7469	Assert(pvUser);
7470	Assert(VMMRZCallRing3IsEnabled(pVCpu));
7471	HMVMX_ASSERT_PREEMPT_SAFE();
7472
7473	VMMRZCallRing3Disable(pVCpu);
7474	Assert(VMMR0IsLogFlushDisabled(pVCpu));
7475
7476	Log4(("hmR0VmxCallRing3Callback->hmR0VmxLongJmpToRing3 pVCpu=%p idCpu=%RU32 enmOperation=%d\n", pVCpu, pVCpu->idCpu,
7477	enmOperation));
7478
7479	int rc = hmR0VmxLongJmpToRing3(pVCpu, (PCPUMCTX)pvUser);
7480	AssertRCReturn(rc, rc);
7481
7482	VMMRZCallRing3Enable(pVCpu);
7483	return VINF_SUCCESS;
7484	}
7485
7486
7487	/**
7488	* Sets the interrupt-window exiting control in the VMCS which instructs VT-x to
7489	* cause a VM-exit as soon as the guest is in a state to receive interrupts.
7490	*
7491	* @param pVCpu The cross context virtual CPU structure.
7492	*/
7493	DECLINLINE(void) hmR0VmxSetIntWindowExitVmcs(PVMCPU pVCpu)
7494	{
7495	if (RT_LIKELY(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT))
7496	{
7497	if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT))
7498	{
7499	pVCpu->hm.s.vmx.u32ProcCtls \|= VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT;
7500	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7501	AssertRC(rc);
7502	Log4(("Setup interrupt-window exiting\n"));
7503	}
7504	} /* else we will deliver interrupts whenever the guest exits next and is in a state to receive events. */
7505	}
7506
7507
7508	/**
7509	* Clears the interrupt-window exiting control in the VMCS.
7510	*
7511	* @param pVCpu The cross context virtual CPU structure.
7512	*/
7513	DECLINLINE(void) hmR0VmxClearIntWindowExitVmcs(PVMCPU pVCpu)
7514	{
7515	Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT);
7516	pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT;
7517	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7518	AssertRC(rc);
7519	Log4(("Cleared interrupt-window exiting\n"));
7520	}
7521
7522
7523	/**
7524	* Sets the NMI-window exiting control in the VMCS which instructs VT-x to
7525	* cause a VM-exit as soon as the guest is in a state to receive NMIs.
7526	*
7527	* @param pVCpu The cross context virtual CPU structure.
7528	*/
7529	DECLINLINE(void) hmR0VmxSetNmiWindowExitVmcs(PVMCPU pVCpu)
7530	{
7531	if (RT_LIKELY(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT))
7532	{
7533	if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT))
7534	{
7535	pVCpu->hm.s.vmx.u32ProcCtls \|= VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT;
7536	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7537	AssertRC(rc);
7538	Log4(("Setup NMI-window exiting\n"));
7539	}
7540	} /* else we will deliver NMIs whenever we VM-exit next, even possibly nesting NMIs. Can't be helped on ancient CPUs. */
7541	}
7542
7543
7544	/**
7545	* Clears the NMI-window exiting control in the VMCS.
7546	*
7547	* @param pVCpu The cross context virtual CPU structure.
7548	*/
7549	DECLINLINE(void) hmR0VmxClearNmiWindowExitVmcs(PVMCPU pVCpu)
7550	{
7551	Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT);
7552	pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT;
7553	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7554	AssertRC(rc);
7555	Log4(("Cleared NMI-window exiting\n"));
7556	}
7557
7558
7559	/**
7560	* Evaluates the event to be delivered to the guest and sets it as the pending
7561	* event.
7562	*
7563	* @returns The VT-x guest-interruptibility state.
7564	* @param pVCpu The cross context virtual CPU structure.
7565	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7566	* out-of-sync. Make sure to update the required fields
7567	* before using them.
7568	*/
7569	static uint32_t hmR0VmxEvaluatePendingEvent(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7570	{
7571	/* Get the current interruptibility-state of the guest and then figure out what can be injected. */
7572	uint32_t const uIntrState = hmR0VmxGetGuestIntrState(pVCpu, pMixedCtx);
7573	bool const fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7574	bool const fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7575	bool const fBlockNmi = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI);
7576
7577	Assert(!fBlockSti \|\| HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS));
7578	Assert(!(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
7579	Assert(!fBlockSti \|\| pMixedCtx->eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
7580	Assert(!TRPMHasTrap(pVCpu));
7581
7582	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UPDATE_APIC))
7583	APICUpdatePendingInterrupts(pVCpu);
7584
7585	/*
7586	* Toggling of interrupt force-flags here is safe since we update TRPM on premature exits
7587	* to ring-3 before executing guest code, see hmR0VmxExitToRing3(). We must NOT restore these force-flags.
7588	*/
7589	/** @todo SMI. SMIs take priority over NMIs. */
7590	if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_NMI)) /* NMI. NMIs take priority over regular interrupts. */
7591	{
7592	/* On some CPUs block-by-STI also blocks NMIs. See Intel spec. 26.3.1.5 "Checks On Guest Non-Register State". */
7593	if ( !pVCpu->hm.s.Event.fPending
7594	&& !fBlockNmi
7595	&& !fBlockSti
7596	&& !fBlockMovSS)
7597	{
7598	Log4(("Pending NMI vcpu[%RU32]\n", pVCpu->idCpu));
7599	uint32_t u32IntInfo = X86_XCPT_NMI \| VMX_EXIT_INTERRUPTION_INFO_VALID;
7600	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7601
7602	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7603	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
7604	}
7605	else
7606	hmR0VmxSetNmiWindowExitVmcs(pVCpu);
7607	}
7608	/*
7609	* Check if the guest can receive external interrupts (PIC/APIC). Once PDMGetInterrupt() returns
7610	* a valid interrupt we must- deliver the interrupt. We can no longer re-request it from the APIC.
7611	*/
7612	else if ( VMCPU_FF_IS_PENDING(pVCpu, (VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC))
7613	&& !pVCpu->hm.s.fSingleInstruction)
7614	{
7615	Assert(!DBGFIsStepping(pVCpu));
7616	int rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
7617	AssertRC(rc);
7618	bool const fBlockInt = !(pMixedCtx->eflags.u32 & X86_EFL_IF);
7619	if ( !pVCpu->hm.s.Event.fPending
7620	&& !fBlockInt
7621	&& !fBlockSti
7622	&& !fBlockMovSS)
7623	{
7624	uint8_t u8Interrupt;
7625	rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
7626	if (RT_SUCCESS(rc))
7627	{
7628	Log4(("Pending interrupt vcpu[%RU32] u8Interrupt=%#x \n", pVCpu->idCpu, u8Interrupt));
7629	uint32_t u32IntInfo = u8Interrupt \| VMX_EXIT_INTERRUPTION_INFO_VALID;
7630	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7631
7632	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrfaultAddress */);
7633	}
7634	else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
7635	{
7636	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
7637	hmR0VmxApicSetTprThreshold(pVCpu, u8Interrupt >> 4);
7638	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
7639
7640	/*
7641	* If the CPU doesn't have TPR shadowing, we will always get a VM-exit on TPR changes and
7642	* APICSetTpr() will end up setting the VMCPU_FF_INTERRUPT_APIC if required, so there is no
7643	* need to re-set this force-flag here.
7644	*/
7645	}
7646	else
7647	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
7648	}
7649	else
7650	hmR0VmxSetIntWindowExitVmcs(pVCpu);
7651	}
7652
7653	return uIntrState;
7654	}
7655
7656
7657	/**
7658	* Sets a pending-debug exception to be delivered to the guest if the guest is
7659	* single-stepping in the VMCS.
7660	*
7661	* @param pVCpu The cross context virtual CPU structure.
7662	*/
7663	DECLINLINE(void) hmR0VmxSetPendingDebugXcptVmcs(PVMCPU pVCpu)
7664	{
7665	Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS)); NOREF(pVCpu);
7666	int rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, VMX_VMCS_GUEST_DEBUG_EXCEPTIONS_BS);
7667	AssertRC(rc);
7668	}
7669
7670
7671	/**
7672	* Injects any pending events into the guest if the guest is in a state to
7673	* receive them.
7674	*
7675	* @returns Strict VBox status code (i.e. informational status codes too).
7676	* @param pVCpu The cross context virtual CPU structure.
7677	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7678	* out-of-sync. Make sure to update the required fields
7679	* before using them.
7680	* @param uIntrState The VT-x guest-interruptibility state.
7681	* @param fStepping Running in hmR0VmxRunGuestCodeStep() and we should
7682	* return VINF_EM_DBG_STEPPED if the event was
7683	* dispatched directly.
7684	*/
7685	static VBOXSTRICTRC hmR0VmxInjectPendingEvent(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t uIntrState, bool fStepping)
7686	{
7687	HMVMX_ASSERT_PREEMPT_SAFE();
7688	Assert(VMMRZCallRing3IsEnabled(pVCpu));
7689
7690	bool fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7691	bool fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7692
7693	Assert(!fBlockSti \|\| HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS));
7694	Assert(!(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
7695	Assert(!fBlockSti \|\| pMixedCtx->eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
7696	Assert(!TRPMHasTrap(pVCpu));
7697
7698	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
7699	if (pVCpu->hm.s.Event.fPending)
7700	{
7701	/*
7702	* Do -not- clear any interrupt-window exiting control here. We might have an interrupt
7703	* pending even while injecting an event and in this case, we want a VM-exit as soon as
7704	* the guest is ready for the next interrupt, see @bugref{6208#c45}.
7705	*
7706	* See Intel spec. 26.6.5 "Interrupt-Window Exiting and Virtual-Interrupt Delivery".
7707	*/
7708	uint32_t const uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
7709	#ifdef VBOX_STRICT
7710	if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
7711	{
7712	bool const fBlockInt = !(pMixedCtx->eflags.u32 & X86_EFL_IF);
7713	Assert(!fBlockInt);
7714	Assert(!fBlockSti);
7715	Assert(!fBlockMovSS);
7716	}
7717	else if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
7718	{
7719	bool const fBlockNmi = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI);
7720	Assert(!fBlockSti);
7721	Assert(!fBlockMovSS);
7722	Assert(!fBlockNmi);
7723	}
7724	#endif
7725	Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64 Type=%#x\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo,
7726	(uint8_t)uIntType));
7727	rcStrict = hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, pVCpu->hm.s.Event.u64IntInfo, pVCpu->hm.s.Event.cbInstr,
7728	pVCpu->hm.s.Event.u32ErrCode, pVCpu->hm.s.Event.GCPtrFaultAddress,
7729	fStepping, &uIntrState);
7730	AssertRCReturn(VBOXSTRICTRC_VAL(rcStrict), rcStrict);
7731
7732	/* Update the interruptibility-state as it could have been changed by
7733	hmR0VmxInjectEventVmcs() (e.g. real-on-v86 guest injecting software interrupts) */
7734	fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7735	fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7736
7737	if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
7738	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
7739	else
7740	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
7741	}
7742
7743	/* Deliver pending debug exception if the guest is single-stepping. Evaluate and set the BS bit. */
7744	if ( fBlockSti
7745	\|\| fBlockMovSS)
7746	{
7747	if (!pVCpu->hm.s.fSingleInstruction)
7748	{
7749	/*
7750	* The pending-debug exceptions field is cleared on all VM-exits except VMX_EXIT_TPR_BELOW_THRESHOLD,
7751	* VMX_EXIT_MTF, VMX_EXIT_APIC_WRITE and VMX_EXIT_VIRTUALIZED_EOI.
7752	* See Intel spec. 27.3.4 "Saving Non-Register State".
7753	*/
7754	Assert(!DBGFIsStepping(pVCpu));
7755	int rc2 = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
7756	AssertRCReturn(rc2, rc2);
7757	if (pMixedCtx->eflags.Bits.u1TF)
7758	hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
7759	}
7760	else if (pMixedCtx->eflags.Bits.u1TF)
7761	{
7762	/*
7763	* We are single-stepping in the hypervisor debugger using EFLAGS.TF. Clear interrupt inhibition as setting the
7764	* BS bit would mean delivering a #DB to the guest upon VM-entry when it shouldn't be.
7765	*/
7766	Assert(!(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG));
7767	uIntrState = 0;
7768	}
7769	}
7770
7771	/*
7772	* There's no need to clear the VM-entry interruption-information field here if we're not injecting anything.
7773	* VT-x clears the valid bit on every VM-exit. See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
7774	*/
7775	int rc2 = hmR0VmxLoadGuestIntrState(pVCpu, uIntrState);
7776	AssertRC(rc2);
7777
7778	Assert(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping));
7779	NOREF(fBlockMovSS); NOREF(fBlockSti);
7780	return rcStrict;
7781	}
7782
7783
7784	/**
7785	* Sets an invalid-opcode (\#UD) exception as pending-for-injection into the VM.
7786	*
7787	* @param pVCpu The cross context virtual CPU structure.
7788	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7789	* out-of-sync. Make sure to update the required fields
7790	* before using them.
7791	*/
7792	DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7793	{
7794	NOREF(pMixedCtx);
7795	uint32_t u32IntInfo = X86_XCPT_UD \| VMX_EXIT_INTERRUPTION_INFO_VALID;
7796	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7797	}
7798
7799
7800	/**
7801	* Injects a double-fault (\#DF) exception into the VM.
7802	*
7803	* @returns Strict VBox status code (i.e. informational status codes too).
7804	* @param pVCpu The cross context virtual CPU structure.
7805	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7806	* out-of-sync. Make sure to update the required fields
7807	* before using them.
7808	* @param fStepping Whether we're running in hmR0VmxRunGuestCodeStep()
7809	* and should return VINF_EM_DBG_STEPPED if the event
7810	* is injected directly (register modified by us, not
7811	* by hardware on VM-entry).
7812	* @param puIntrState Pointer to the current guest interruptibility-state.
7813	* This interruptibility-state will be updated if
7814	* necessary. This cannot not be NULL.
7815	*/
7816	DECLINLINE(VBOXSTRICTRC) hmR0VmxInjectXcptDF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fStepping, uint32_t *puIntrState)
7817	{
7818	uint32_t u32IntInfo = X86_XCPT_DF \| VMX_EXIT_INTERRUPTION_INFO_VALID;
7819	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7820	u32IntInfo \|= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7821	return hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */,
7822	fStepping, puIntrState);
7823	}
7824
7825
7826	/**
7827	* Sets a debug (\#DB) exception as pending-for-injection into the VM.
7828	*
7829	* @param pVCpu The cross context virtual CPU structure.
7830	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7831	* out-of-sync. Make sure to update the required fields
7832	* before using them.
7833	*/
7834	DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7835	{
7836	NOREF(pMixedCtx);
7837	uint32_t u32IntInfo = X86_XCPT_DB \| VMX_EXIT_INTERRUPTION_INFO_VALID;
7838	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7839	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, 0 / u32ErrCode /, 0 / GCPtrFaultAddress */);
7840	}
7841
7842
7843	/**
7844	* Sets an overflow (\#OF) exception as pending-for-injection into the VM.
7845	*
7846	* @param pVCpu The cross context virtual CPU structure.
7847	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7848	* out-of-sync. Make sure to update the required fields
7849	* before using them.
7850	* @param cbInstr The value of RIP that is to be pushed on the guest
7851	* stack.
7852	*/
7853	DECLINLINE(void) hmR0VmxSetPendingXcptOF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t cbInstr)
7854	{
7855	NOREF(pMixedCtx);
7856	uint32_t u32IntInfo = X86_XCPT_OF \| VMX_EXIT_INTERRUPTION_INFO_VALID;
7857	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7858	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, 0 /* u32ErrCode /, 0 / GCPtrFaultAddress */);
7859	}
7860
7861
7862	/**
7863	* Injects a general-protection (\#GP) fault into the VM.
7864	*
7865	* @returns Strict VBox status code (i.e. informational status codes too).
7866	* @param pVCpu The cross context virtual CPU structure.
7867	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7868	* out-of-sync. Make sure to update the required fields
7869	* before using them.
7870	* @param fErrorCodeValid Whether the error code is valid (depends on the CPU
7871	* mode, i.e. in real-mode it's not valid).
7872	* @param u32ErrorCode The error code associated with the \#GP.
7873	* @param fStepping Whether we're running in
7874	* hmR0VmxRunGuestCodeStep() and should return
7875	* VINF_EM_DBG_STEPPED if the event is injected
7876	* directly (register modified by us, not by
7877	* hardware on VM-entry).
7878	* @param puIntrState Pointer to the current guest interruptibility-state.
7879	* This interruptibility-state will be updated if
7880	* necessary. This cannot not be NULL.
7881	*/
7882	DECLINLINE(VBOXSTRICTRC) hmR0VmxInjectXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fErrorCodeValid, uint32_t u32ErrorCode,
7883	bool fStepping, uint32_t *puIntrState)
7884	{
7885	uint32_t u32IntInfo = X86_XCPT_GP \| VMX_EXIT_INTERRUPTION_INFO_VALID;
7886	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7887	if (fErrorCodeValid)
7888	u32IntInfo \|= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7889	return hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, u32IntInfo, 0 /* cbInstr /, u32ErrorCode, 0 / GCPtrFaultAddress */,
7890	fStepping, puIntrState);
7891	}
7892
7893
7894	#if 0 /* unused */
7895	/**
7896	* Sets a general-protection (\#GP) exception as pending-for-injection into the
7897	* VM.
7898	*
7899	* @param pVCpu The cross context virtual CPU structure.
7900	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7901	* out-of-sync. Make sure to update the required fields
7902	* before using them.
7903	* @param u32ErrorCode The error code associated with the \#GP.
7904	*/
7905	DECLINLINE(void) hmR0VmxSetPendingXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t u32ErrorCode)
7906	{
7907	NOREF(pMixedCtx);
7908	uint32_t u32IntInfo = X86_XCPT_GP \| VMX_EXIT_INTERRUPTION_INFO_VALID;
7909	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7910	u32IntInfo \|= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7911	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr /, u32ErrorCode, 0 / GCPtrFaultAddress */);
7912	}
7913	#endif /* unused */
7914
7915
7916	/**
7917	* Sets a software interrupt (INTn) as pending-for-injection into the VM.
7918	*
7919	* @param pVCpu The cross context virtual CPU structure.
7920	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
7921	* out-of-sync. Make sure to update the required fields
7922	* before using them.
7923	* @param uVector The software interrupt vector number.
7924	* @param cbInstr The value of RIP that is to be pushed on the guest
7925	* stack.
7926	*/
7927	DECLINLINE(void) hmR0VmxSetPendingIntN(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint16_t uVector, uint32_t cbInstr)
7928	{
7929	NOREF(pMixedCtx);
7930	uint32_t u32IntInfo = uVector \| VMX_EXIT_INTERRUPTION_INFO_VALID;
7931	if ( uVector == X86_XCPT_BP
7932	\|\| uVector == X86_XCPT_OF)
7933	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7934	else
7935	u32IntInfo \|= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7936	hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, 0 /* u32ErrCode /, 0 / GCPtrFaultAddress */);
7937	}
7938
7939
7940	/**
7941	* Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's
7942	* stack.
7943	*
7944	* @returns Strict VBox status code (i.e. informational status codes too).
7945	* @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault.
7946	* @param pVM The cross context VM structure.
7947	* @param pMixedCtx Pointer to the guest-CPU context.
7948	* @param uValue The value to push to the guest stack.
7949	*/
7950	DECLINLINE(VBOXSTRICTRC) hmR0VmxRealModeGuestStackPush(PVM pVM, PCPUMCTX pMixedCtx, uint16_t uValue)
7951	{
7952	/*
7953	* The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in
7954	* virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
7955	* See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound".
7956	*/
7957	if (pMixedCtx->sp == 1)
7958	return VINF_EM_RESET;
7959	pMixedCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */
7960	int rc = PGMPhysSimpleWriteGCPhys(pVM, pMixedCtx->ss.u64Base + pMixedCtx->sp, &uValue, sizeof(uint16_t));
7961	AssertRC(rc);
7962	return rc;
7963	}
7964
7965
7966	/**
7967	* Injects an event into the guest upon VM-entry by updating the relevant fields
7968	* in the VM-entry area in the VMCS.
7969	*
7970	* @returns Strict VBox status code (i.e. informational status codes too).
7971	* @retval VINF_SUCCESS if the event is successfully injected into the VMCS.
7972	* @retval VINF_EM_RESET if event injection resulted in a triple-fault.
7973	*
7974	* @param pVCpu The cross context virtual CPU structure.
7975	* @param pMixedCtx Pointer to the guest-CPU context. The data may
7976	* be out-of-sync. Make sure to update the required
7977	* fields before using them.
7978	* @param u64IntInfo The VM-entry interruption-information field.
7979	* @param cbInstr The VM-entry instruction length in bytes (for
7980	* software interrupts, exceptions and privileged
7981	* software exceptions).
7982	* @param u32ErrCode The VM-entry exception error code.
7983	* @param GCPtrFaultAddress The page-fault address for \#PF exceptions.
7984	* @param puIntrState Pointer to the current guest interruptibility-state.
7985	* This interruptibility-state will be updated if
7986	* necessary. This cannot not be NULL.
7987	* @param fStepping Whether we're running in
7988	* hmR0VmxRunGuestCodeStep() and should return
7989	* VINF_EM_DBG_STEPPED if the event is injected
7990	* directly (register modified by us, not by
7991	* hardware on VM-entry).
7992	*
7993	* @remarks Requires CR0!
7994	*/
7995	static VBOXSTRICTRC hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint64_t u64IntInfo, uint32_t cbInstr,
7996	uint32_t u32ErrCode, RTGCUINTREG GCPtrFaultAddress, bool fStepping,
7997	uint32_t *puIntrState)
7998	{
7999	/* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */
8000	AssertMsg(u64IntInfo >> 32 == 0, ("%#RX64\n", u64IntInfo));
8001	Assert(puIntrState);
8002	uint32_t u32IntInfo = (uint32_t)u64IntInfo;
8003
8004	uint32_t const uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(u32IntInfo);
8005	uint32_t const uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(u32IntInfo);
8006
8007	#ifdef VBOX_STRICT
8008	/* Validate the error-code-valid bit for hardware exceptions. */
8009	if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT)
8010	{
8011	switch (uVector)
8012	{
8013	case X86_XCPT_PF:
8014	case X86_XCPT_DF:
8015	case X86_XCPT_TS:
8016	case X86_XCPT_NP:
8017	case X86_XCPT_SS:
8018	case X86_XCPT_GP:
8019	case X86_XCPT_AC:
8020	AssertMsg(VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(u32IntInfo),
8021	("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector));
8022	/* fall thru */
8023	default:
8024	break;
8025	}
8026	}
8027	#endif
8028
8029	/* Cannot inject an NMI when block-by-MOV SS is in effect. */
8030	Assert( uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI
8031	\|\| !(*puIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS));
8032
8033	STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedIrqsR0[uVector & MASK_INJECT_IRQ_STAT]);
8034
8035	/* We require CR0 to check if the guest is in real-mode. */
8036	int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
8037	AssertRCReturn(rc, rc);
8038
8039	/*
8040	* Hardware interrupts & exceptions cannot be delivered through the software interrupt redirection bitmap to the real
8041	* mode task in virtual-8086 mode. We must jump to the interrupt handler in the (real-mode) guest.
8042	* See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode" for interrupt & exception classes.
8043	* See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling.
8044	*/
8045	if (CPUMIsGuestInRealModeEx(pMixedCtx))
8046	{
8047	PVM pVM = pVCpu->CTX_SUFF(pVM);
8048	if (!pVM->hm.s.vmx.fUnrestrictedGuest)
8049	{
8050	Assert(PDMVmmDevHeapIsEnabled(pVM));
8051	Assert(pVM->hm.s.vmx.pRealModeTSS);
8052
8053	/* We require RIP, RSP, RFLAGS, CS, IDTR. Save the required ones from the VMCS. */
8054	rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
8055	rc \|= hmR0VmxSaveGuestTableRegs(pVCpu, pMixedCtx);
8056	rc \|= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
8057	AssertRCReturn(rc, rc);
8058	Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP));
8059
8060	/* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */
8061	size_t const cbIdtEntry = sizeof(X86IDTR16);
8062	if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pMixedCtx->idtr.cbIdt)
8063	{
8064	/* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */
8065	if (uVector == X86_XCPT_DF)
8066	return VINF_EM_RESET;
8067
8068	/* If we're injecting a #GP with no valid IDT entry, inject a double-fault. */
8069	if (uVector == X86_XCPT_GP)
8070	return hmR0VmxInjectXcptDF(pVCpu, pMixedCtx, fStepping, puIntrState);
8071
8072	/* If we're injecting an interrupt/exception with no valid IDT entry, inject a general-protection fault. */
8073	/* No error codes for exceptions in real-mode. See Intel spec. 20.1.4 "Interrupt and Exception Handling" */
8074	return hmR0VmxInjectXcptGP(pVCpu, pMixedCtx, false /* fErrCodeValid /, 0 / u32ErrCode */,
8075	fStepping, puIntrState);
8076	}
8077
8078	/* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */
8079	uint16_t uGuestIp = pMixedCtx->ip;
8080	if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT)
8081	{
8082	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF);
8083	/* #BP and #OF are both benign traps, we need to resume the next instruction. */
8084	uGuestIp = pMixedCtx->ip + (uint16_t)cbInstr;
8085	}
8086	else if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT)
8087	uGuestIp = pMixedCtx->ip + (uint16_t)cbInstr;
8088
8089	/* Get the code segment selector and offset from the IDT entry for the interrupt handler. */
8090	X86IDTR16 IdtEntry;
8091	RTGCPHYS GCPhysIdtEntry = (RTGCPHYS)pMixedCtx->idtr.pIdt + uVector * cbIdtEntry;
8092	rc = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry);
8093	AssertRCReturn(rc, rc);
8094
8095	/* Construct the stack frame for the interrupt/exception handler. */
8096	VBOXSTRICTRC rcStrict;
8097	rcStrict = hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, pMixedCtx->eflags.u32);
8098	if (rcStrict == VINF_SUCCESS)
8099	rcStrict = hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, pMixedCtx->cs.Sel);
8100	if (rcStrict == VINF_SUCCESS)
8101	rcStrict = hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, uGuestIp);
8102
8103	/* Clear the required eflag bits and jump to the interrupt/exception handler. */
8104	if (rcStrict == VINF_SUCCESS)
8105	{
8106	pMixedCtx->eflags.u32 &= ~(X86_EFL_IF \| X86_EFL_TF \| X86_EFL_RF \| X86_EFL_AC);
8107	pMixedCtx->rip = IdtEntry.offSel;
8108	pMixedCtx->cs.Sel = IdtEntry.uSel;
8109	pMixedCtx->cs.ValidSel = IdtEntry.uSel;
8110	pMixedCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry;
8111	if ( uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT
8112	&& uVector == X86_XCPT_PF)
8113	pMixedCtx->cr2 = GCPtrFaultAddress;
8114
8115	/* If any other guest-state bits are changed here, make sure to update
8116	hmR0VmxPreRunGuestCommitted() when thread-context hooks are used. */
8117	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS
8118	\| HM_CHANGED_GUEST_RIP
8119	\| HM_CHANGED_GUEST_RFLAGS
8120	\| HM_CHANGED_GUEST_RSP);
8121
8122	/* We're clearing interrupts, which means no block-by-STI interrupt-inhibition. */
8123	if (*puIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
8124	{
8125	Assert( uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI
8126	&& uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT);
8127	Log4(("Clearing inhibition due to STI.\n"));
8128	*puIntrState &= ~VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI;
8129	}
8130	Log4(("Injecting real-mode: u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x Eflags=%#x CS:EIP=%04x:%04x\n",
8131	u32IntInfo, u32ErrCode, cbInstr, pMixedCtx->eflags.u, pMixedCtx->cs.Sel, pMixedCtx->eip));
8132
8133	/* The event has been truly dispatched. Mark it as no longer pending so we don't attempt to 'undo'
8134	it, if we are returning to ring-3 before executing guest code. */
8135	pVCpu->hm.s.Event.fPending = false;
8136
8137	/* Make hmR0VmxPreRunGuest return if we're stepping since we've changed cs:rip. */
8138	if (fStepping)
8139	rcStrict = VINF_EM_DBG_STEPPED;
8140	}
8141	AssertMsg(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
8142	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8143	return rcStrict;
8144	}
8145
8146	/*
8147	* For unrestricted execution enabled CPUs running real-mode guests, we must not set the deliver-error-code bit.
8148	* See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
8149	*/
8150	u32IntInfo &= ~VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
8151	}
8152
8153	/* Validate. */
8154	Assert(VMX_EXIT_INTERRUPTION_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */
8155	Assert(!VMX_EXIT_INTERRUPTION_INFO_NMI_UNBLOCK_IRET(u32IntInfo)); /* Bit 12 MBZ. */
8156	Assert(!(u32IntInfo & 0x7ffff000)); /* Bits 30:12 MBZ. */
8157
8158	/* Inject. */
8159	rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo);
8160	if (VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(u32IntInfo))
8161	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode);
8162	rc \|= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr);
8163
8164	if ( VMX_EXIT_INTERRUPTION_INFO_TYPE(u32IntInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT
8165	&& uVector == X86_XCPT_PF)
8166	pMixedCtx->cr2 = GCPtrFaultAddress;
8167
8168	Log4(("Injecting vcpu[%RU32] u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x pMixedCtx->uCR2=%#RX64\n", pVCpu->idCpu,
8169	u32IntInfo, u32ErrCode, cbInstr, pMixedCtx->cr2));
8170
8171	AssertRCReturn(rc, rc);
8172	return VINF_SUCCESS;
8173	}
8174
8175
8176	/**
8177	* Clears the interrupt-window exiting control in the VMCS and if necessary
8178	* clears the current event in the VMCS as well.
8179	*
8180	* @returns VBox status code.
8181	* @param pVCpu The cross context virtual CPU structure.
8182	*
8183	* @remarks Use this function only to clear events that have not yet been
8184	* delivered to the guest but are injected in the VMCS!
8185	* @remarks No-long-jump zone!!!
8186	*/
8187	static void hmR0VmxClearIntNmiWindowsVmcs(PVMCPU pVCpu)
8188	{
8189	Log4Func(("vcpu[%d]\n", pVCpu->idCpu));
8190
8191	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT)
8192	hmR0VmxClearIntWindowExitVmcs(pVCpu);
8193
8194	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT)
8195	hmR0VmxClearNmiWindowExitVmcs(pVCpu);
8196	}
8197
8198
8199	/**
8200	* Enters the VT-x session.
8201	*
8202	* @returns VBox status code.
8203	* @param pVM The cross context VM structure.
8204	* @param pVCpu The cross context virtual CPU structure.
8205	* @param pCpu Pointer to the CPU info struct.
8206	*/
8207	VMMR0DECL(int) VMXR0Enter(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
8208	{
8209	AssertPtr(pVM);
8210	AssertPtr(pVCpu);
8211	Assert(pVM->hm.s.vmx.fSupported);
8212	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8213	NOREF(pCpu); NOREF(pVM);
8214
8215	LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8216	Assert(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_HOST_GUEST_SHARED_STATE));
8217
8218	#ifdef VBOX_STRICT
8219	/* At least verify VMX is enabled, since we can't check if we're in VMX root mode without #GP'ing. */
8220	RTCCUINTREG uHostCR4 = ASMGetCR4();
8221	if (!(uHostCR4 & X86_CR4_VMXE))
8222	{
8223	LogRel(("VMXR0Enter: X86_CR4_VMXE bit in CR4 is not set!\n"));
8224	return VERR_VMX_X86_CR4_VMXE_CLEARED;
8225	}
8226	#endif
8227
8228	/*
8229	* Load the VCPU's VMCS as the current (and active) one.
8230	*/
8231	Assert(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_CLEAR);
8232	int rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
8233	if (RT_FAILURE(rc))
8234	return rc;
8235
8236	pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE;
8237	pVCpu->hm.s.fLeaveDone = false;
8238	Log4Func(("Activated Vmcs. HostCpuId=%u\n", RTMpCpuId()));
8239
8240	return VINF_SUCCESS;
8241	}
8242
8243
8244	/**
8245	* The thread-context callback (only on platforms which support it).
8246	*
8247	* @param enmEvent The thread-context event.
8248	* @param pVCpu The cross context virtual CPU structure.
8249	* @param fGlobalInit Whether global VT-x/AMD-V init. was used.
8250	* @thread EMT(pVCpu)
8251	*/
8252	VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPU pVCpu, bool fGlobalInit)
8253	{
8254	NOREF(fGlobalInit);
8255
8256	switch (enmEvent)
8257	{
8258	case RTTHREADCTXEVENT_OUT:
8259	{
8260	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8261	Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
8262	VMCPU_ASSERT_EMT(pVCpu);
8263
8264	PCPUMCTX pMixedCtx = CPUMQueryGuestCtxPtr(pVCpu);
8265
8266	/* No longjmps (logger flushes, locks) in this fragile context. */
8267	VMMRZCallRing3Disable(pVCpu);
8268	Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId()));
8269
8270	/*
8271	* Restore host-state (FPU, debug etc.)
8272	*/
8273	if (!pVCpu->hm.s.fLeaveDone)
8274	{
8275	/* Do -not- save guest-state here as we might already be in the middle of saving it (esp. bad if we are
8276	holding the PGM lock while saving the guest state (see hmR0VmxSaveGuestControlRegs()). */
8277	hmR0VmxLeave(pVCpu, pMixedCtx, false /* fSaveGuestState */);
8278	pVCpu->hm.s.fLeaveDone = true;
8279	}
8280
8281	/* Leave HM context, takes care of local init (term). */
8282	int rc = HMR0LeaveCpu(pVCpu);
8283	AssertRC(rc); NOREF(rc);
8284
8285	/* Restore longjmp state. */
8286	VMMRZCallRing3Enable(pVCpu);
8287	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
8288	break;
8289	}
8290
8291	case RTTHREADCTXEVENT_IN:
8292	{
8293	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8294	Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
8295	VMCPU_ASSERT_EMT(pVCpu);
8296
8297	/* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */
8298	VMMRZCallRing3Disable(pVCpu);
8299	Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId()));
8300
8301	/* Initialize the bare minimum state required for HM. This takes care of
8302	initializing VT-x if necessary (onlined CPUs, local init etc.) */
8303	int rc = HMR0EnterCpu(pVCpu);
8304	AssertRC(rc);
8305	Assert(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_HOST_GUEST_SHARED_STATE));
8306
8307	/* Load the active VMCS as the current one. */
8308	if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_CLEAR)
8309	{
8310	rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
8311	AssertRC(rc); NOREF(rc);
8312	pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE;
8313	Log4Func(("Resumed: Activated Vmcs. HostCpuId=%u\n", RTMpCpuId()));
8314	}
8315	pVCpu->hm.s.fLeaveDone = false;
8316
8317	/* Restore longjmp state. */
8318	VMMRZCallRing3Enable(pVCpu);
8319	break;
8320	}
8321
8322	default:
8323	break;
8324	}
8325	}
8326
8327
8328	/**
8329	* Saves the host state in the VMCS host-state.
8330	* Sets up the VM-exit MSR-load area.
8331	*
8332	* The CPU state will be loaded from these fields on every successful VM-exit.
8333	*
8334	* @returns VBox status code.
8335	* @param pVM The cross context VM structure.
8336	* @param pVCpu The cross context virtual CPU structure.
8337	*
8338	* @remarks No-long-jump zone!!!
8339	*/
8340	static int hmR0VmxSaveHostState(PVM pVM, PVMCPU pVCpu)
8341	{
8342	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8343
8344	int rc = VINF_SUCCESS;
8345	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT))
8346	{
8347	rc = hmR0VmxSaveHostControlRegs(pVM, pVCpu);
8348	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostControlRegisters failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8349
8350	rc = hmR0VmxSaveHostSegmentRegs(pVM, pVCpu);
8351	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostSegmentRegisters failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8352
8353	rc = hmR0VmxSaveHostMsrs(pVM, pVCpu);
8354	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostMsrs failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8355
8356	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_HOST_CONTEXT);
8357	}
8358	return rc;
8359	}
8360
8361
8362	/**
8363	* Saves the host state in the VMCS host-state.
8364	*
8365	* @returns VBox status code.
8366	* @param pVM The cross context VM structure.
8367	* @param pVCpu The cross context virtual CPU structure.
8368	*
8369	* @remarks No-long-jump zone!!!
8370	*/
8371	VMMR0DECL(int) VMXR0SaveHostState(PVM pVM, PVMCPU pVCpu)
8372	{
8373	AssertPtr(pVM);
8374	AssertPtr(pVCpu);
8375
8376	LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8377
8378	/* Save the host state here while entering HM context. When thread-context hooks are used, we might get preempted
8379	and have to resave the host state but most of the time we won't be, so do it here before we disable interrupts. */
8380	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8381	return hmR0VmxSaveHostState(pVM, pVCpu);
8382	}
8383
8384
8385	/**
8386	* Loads the guest state into the VMCS guest-state area.
8387	*
8388	* The will typically be done before VM-entry when the guest-CPU state and the
8389	* VMCS state may potentially be out of sync.
8390	*
8391	* Sets up the VM-entry MSR-load and VM-exit MSR-store areas. Sets up the
8392	* VM-entry controls.
8393	* Sets up the appropriate VMX non-root function to execute guest code based on
8394	* the guest CPU mode.
8395	*
8396	* @returns VBox strict status code.
8397	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
8398	* without unrestricted guest access and the VMMDev is not presently
8399	* mapped (e.g. EFI32).
8400	*
8401	* @param pVM The cross context VM structure.
8402	* @param pVCpu The cross context virtual CPU structure.
8403	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
8404	* out-of-sync. Make sure to update the required fields
8405	* before using them.
8406	*
8407	* @remarks No-long-jump zone!!!
8408	*/
8409	static VBOXSTRICTRC hmR0VmxLoadGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
8410	{
8411	AssertPtr(pVM);
8412	AssertPtr(pVCpu);
8413	AssertPtr(pMixedCtx);
8414	HMVMX_ASSERT_PREEMPT_SAFE();
8415
8416	LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8417
8418	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestState, x);
8419
8420	/* Determine real-on-v86 mode. */
8421	pVCpu->hm.s.vmx.RealMode.fRealOnV86Active = false;
8422	if ( !pVM->hm.s.vmx.fUnrestrictedGuest
8423	&& CPUMIsGuestInRealModeEx(pMixedCtx))
8424	{
8425	pVCpu->hm.s.vmx.RealMode.fRealOnV86Active = true;
8426	}
8427
8428	/*
8429	* Load the guest-state into the VMCS.
8430	* Any ordering dependency among the sub-functions below must be explicitly stated using comments.
8431	* Ideally, assert that the cross-dependent bits are up-to-date at the point of using it.
8432	*/
8433	int rc = hmR0VmxSetupVMRunHandler(pVCpu, pMixedCtx);
8434	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSetupVMRunHandler! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8435
8436	/* This needs to be done after hmR0VmxSetupVMRunHandler() as changing pfnStartVM may require VM-entry control updates. */
8437	rc = hmR0VmxLoadGuestEntryCtls(pVCpu, pMixedCtx);
8438	AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestEntryCtls! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8439
8440	/* This needs to be done after hmR0VmxSetupVMRunHandler() as changing pfnStartVM may require VM-exit control updates. */
8441	rc = hmR0VmxLoadGuestExitCtls(pVCpu, pMixedCtx);
8442	AssertLogRelMsgRCReturn(rc, ("hmR0VmxSetupExitCtls failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8443
8444	rc = hmR0VmxLoadGuestActivityState(pVCpu, pMixedCtx);
8445	AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestActivityState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8446
8447	VBOXSTRICTRC rcStrict = hmR0VmxLoadGuestCR3AndCR4(pVCpu, pMixedCtx);
8448	if (rcStrict == VINF_SUCCESS)
8449	{ /* likely */ }
8450	else
8451	{
8452	Assert(rcStrict == VINF_EM_RESCHEDULE_REM \|\| RT_FAILURE_NP(rcStrict));
8453	return rcStrict;
8454	}
8455
8456	/* Assumes pMixedCtx->cr0 is up-to-date (strict builds require CR0 for segment register validation checks). */
8457	rc = hmR0VmxLoadGuestSegmentRegs(pVCpu, pMixedCtx);
8458	AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestSegmentRegs: rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8459
8460	/* This needs to be done after hmR0VmxLoadGuestEntryCtls() and hmR0VmxLoadGuestExitCtls() as it may alter controls if we
8461	determine we don't have to swap EFER after all. */
8462	rc = hmR0VmxLoadGuestMsrs(pVCpu, pMixedCtx);
8463	AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestMsrs! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8464
8465	rc = hmR0VmxLoadGuestApicState(pVCpu, pMixedCtx);
8466	AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestApicState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8467
8468	rc = hmR0VmxLoadGuestXcptIntercepts(pVCpu, pMixedCtx);
8469	AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestXcptIntercepts! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8470
8471	/*
8472	* Loading Rflags here is fine, even though Rflags.TF might depend on guest debug state (which is not loaded here).
8473	* It is re-evaluated and updated if necessary in hmR0VmxLoadSharedState().
8474	*/
8475	rc = hmR0VmxLoadGuestRipRspRflags(pVCpu, pMixedCtx);
8476	AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestRipRspRflags! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8477
8478	/* Clear any unused and reserved bits. */
8479	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR2);
8480
8481	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestState, x);
8482	return rc;
8483	}
8484
8485
8486	/**
8487	* Loads the state shared between the host and guest into the VMCS.
8488	*
8489	* @param pVM The cross context VM structure.
8490	* @param pVCpu The cross context virtual CPU structure.
8491	* @param pCtx Pointer to the guest-CPU context.
8492	*
8493	* @remarks No-long-jump zone!!!
8494	*/
8495	static void hmR0VmxLoadSharedState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
8496	{
8497	NOREF(pVM);
8498
8499	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8500	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8501
8502	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0))
8503	{
8504	int rc = hmR0VmxLoadSharedCR0(pVCpu, pCtx);
8505	AssertRC(rc);
8506	}
8507
8508	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG))
8509	{
8510	int rc = hmR0VmxLoadSharedDebugState(pVCpu, pCtx);
8511	AssertRC(rc);
8512
8513	/* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */
8514	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS))
8515	{
8516	rc = hmR0VmxLoadGuestRflags(pVCpu, pCtx);
8517	AssertRC(rc);
8518	}
8519	}
8520
8521	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS))
8522	{
8523	hmR0VmxLazyLoadGuestMsrs(pVCpu, pCtx);
8524	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS);
8525	}
8526
8527	/* Loading CR0, debug state might have changed intercepts, update VMCS. */
8528	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS))
8529	{
8530	Assert(pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT_32(X86_XCPT_AC));
8531	Assert(pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT_32(X86_XCPT_DB));
8532	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
8533	AssertRC(rc);
8534	HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
8535	}
8536
8537	AssertMsg(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE),
8538	("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8539	}
8540
8541
8542	/**
8543	* Worker for loading the guest-state bits in the inner VT-x execution loop.
8544	*
8545	* @returns Strict VBox status code (i.e. informational status codes too).
8546	* @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
8547	* without unrestricted guest access and the VMMDev is not presently
8548	* mapped (e.g. EFI32).
8549	*
8550	* @param pVM The cross context VM structure.
8551	* @param pVCpu The cross context virtual CPU structure.
8552	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
8553	* out-of-sync. Make sure to update the required fields
8554	* before using them.
8555	*
8556	* @remarks No-long-jump zone!!!
8557	*/
8558	static VBOXSTRICTRC hmR0VmxLoadGuestStateOptimal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
8559	{
8560	HMVMX_ASSERT_PREEMPT_SAFE();
8561	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8562	Assert(VMMR0IsLogFlushDisabled(pVCpu));
8563
8564	Log5(("LoadFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8565	#ifdef HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
8566	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
8567	#endif
8568
8569	/*
8570	* RIP is what changes the most often and hence if it's the only bit needing to be
8571	* updated, we shall handle it early for performance reasons.
8572	*/
8573	VBOXSTRICTRC rcStrict = VINF_SUCCESS;
8574	if (HMCPU_CF_IS_SET_ONLY(pVCpu, HM_CHANGED_GUEST_RIP))
8575	{
8576	rcStrict = hmR0VmxLoadGuestRip(pVCpu, pMixedCtx);
8577	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8578	{ /* likely */}
8579	else
8580	{
8581	AssertMsgFailedReturn(("hmR0VmxLoadGuestStateOptimal: hmR0VmxLoadGuestRip failed! rc=%Rrc\n",
8582	VBOXSTRICTRC_VAL(rcStrict)), rcStrict);
8583	}
8584	STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadMinimal);
8585	}
8586	else if (HMCPU_CF_VALUE(pVCpu))
8587	{
8588	rcStrict = hmR0VmxLoadGuestState(pVM, pVCpu, pMixedCtx);
8589	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8590	{ /* likely */}
8591	else
8592	{
8593	AssertMsg(rcStrict == VINF_EM_RESCHEDULE_REM,
8594	("hmR0VmxLoadGuestStateOptimal: hmR0VmxLoadGuestState failed! rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8595	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8596	return rcStrict;
8597	}
8598	STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadFull);
8599	}
8600
8601	/* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */
8602	AssertMsg( !HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_ALL_GUEST)
8603	\|\| HMCPU_CF_IS_PENDING_ONLY(pVCpu, HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_HOST_GUEST_SHARED_STATE),
8604	("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8605	return rcStrict;
8606	}
8607
8608
8609	/**
8610	* Does the preparations before executing guest code in VT-x.
8611	*
8612	* This may cause longjmps to ring-3 and may even result in rescheduling to the
8613	* recompiler/IEM. We must be cautious what we do here regarding committing
8614	* guest-state information into the VMCS assuming we assuredly execute the
8615	* guest in VT-x mode.
8616	*
8617	* If we fall back to the recompiler/IEM after updating the VMCS and clearing
8618	* the common-state (TRPM/forceflags), we must undo those changes so that the
8619	* recompiler/IEM can (and should) use them when it resumes guest execution.
8620	* Otherwise such operations must be done when we can no longer exit to ring-3.
8621	*
8622	* @returns Strict VBox status code (i.e. informational status codes too).
8623	* @retval VINF_SUCCESS if we can proceed with running the guest, interrupts
8624	* have been disabled.
8625	* @retval VINF_EM_RESET if a triple-fault occurs while injecting a
8626	* double-fault into the guest.
8627	* @retval VINF_EM_DBG_STEPPED if @a fStepping is true and an event was
8628	* dispatched directly.
8629	* @retval VINF_* scheduling changes, we have to go back to ring-3.
8630	*
8631	* @param pVM The cross context VM structure.
8632	* @param pVCpu The cross context virtual CPU structure.
8633	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
8634	* out-of-sync. Make sure to update the required fields
8635	* before using them.
8636	* @param pVmxTransient Pointer to the VMX transient structure.
8637	* @param fStepping Set if called from hmR0VmxRunGuestCodeStep(). Makes
8638	* us ignore some of the reasons for returning to
8639	* ring-3, and return VINF_EM_DBG_STEPPED if event
8640	* dispatching took place.
8641	*/
8642	static VBOXSTRICTRC hmR0VmxPreRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, bool fStepping)
8643	{
8644	Assert(VMMRZCallRing3IsEnabled(pVCpu));
8645
8646	#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
8647	PGMRZDynMapFlushAutoSet(pVCpu);
8648	#endif
8649
8650	/* Check force flag actions that might require us to go back to ring-3. */
8651	VBOXSTRICTRC rcStrict = hmR0VmxCheckForceFlags(pVM, pVCpu, pMixedCtx, fStepping);
8652	if (rcStrict == VINF_SUCCESS)
8653	{ /* FFs doesn't get set all the time. */ }
8654	else
8655	return rcStrict;
8656
8657	#ifndef IEM_VERIFICATION_MODE_FULL
8658	/*
8659	* Setup the virtualized-APIC accesses.
8660	*
8661	* Note! This can cause a longjumps to R3 due to the acquisition of the PGM lock
8662	* in both PGMHandlerPhysicalReset() and IOMMMIOMapMMIOHCPage(), see @bugref{8721}.
8663	*
8664	* This is the reason we do it here and not in hmR0VmxLoadGuestState().
8665	*/
8666	if ( !pVCpu->hm.s.vmx.u64MsrApicBase
8667	&& (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)
8668	&& PDMHasApic(pVM))
8669	{
8670	uint64_t const u64MsrApicBase = APICGetBaseMsrNoCheck(pVCpu);
8671	Assert(u64MsrApicBase);
8672	Assert(pVM->hm.s.vmx.HCPhysApicAccess);
8673
8674	RTGCPHYS const GCPhysApicBase = u64MsrApicBase & PAGE_BASE_GC_MASK;
8675
8676	/* Unalias any existing mapping. */
8677	int rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase);
8678	AssertRCReturn(rc, rc);
8679
8680	/* Map the HC APIC-access page in place of the MMIO page, also updates the shadow page tables if necessary. */
8681	Log4(("hmR0VmxPreRunGuest: VCPU%u: Mapped HC APIC-access page at %#RGp\n", pVCpu->idCpu, GCPhysApicBase));
8682	rc = IOMMMIOMapMMIOHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hm.s.vmx.HCPhysApicAccess, X86_PTE_RW \| X86_PTE_P);
8683	AssertRCReturn(rc, rc);
8684
8685	/* Update the per-VCPU cache of the APIC base MSR. */
8686	pVCpu->hm.s.vmx.u64MsrApicBase = u64MsrApicBase;
8687	}
8688	#endif /* !IEM_VERIFICATION_MODE_FULL */
8689
8690	if (TRPMHasTrap(pVCpu))
8691	hmR0VmxTrpmTrapToPendingEvent(pVCpu);
8692	uint32_t uIntrState = hmR0VmxEvaluatePendingEvent(pVCpu, pMixedCtx);
8693
8694	/*
8695	* Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus needs to be done with
8696	* longjmps or interrupts + preemption enabled. Event injection might also result in triple-faulting the VM.
8697	*/
8698	rcStrict = hmR0VmxInjectPendingEvent(pVCpu, pMixedCtx, uIntrState, fStepping);
8699	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8700	{ /* likely */ }
8701	else
8702	{
8703	AssertMsg(rcStrict == VINF_EM_RESET \|\| (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
8704	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8705	return rcStrict;
8706	}
8707
8708	/*
8709	* No longjmps to ring-3 from this point on!!!
8710	* Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
8711	* This also disables flushing of the R0-logger instance (if any).
8712	*/
8713	VMMRZCallRing3Disable(pVCpu);
8714
8715	/*
8716	* Load the guest state bits.
8717	*
8718	* We cannot perform longjmps while loading the guest state because we do not preserve the
8719	* host/guest state (although the VMCS will be preserved) across longjmps which can cause
8720	* CPU migration.
8721	*
8722	* If we are injecting events to a real-on-v86 mode guest, we will have to update
8723	* RIP and some segment registers, i.e. hmR0VmxInjectPendingEvent()->hmR0VmxInjectEventVmcs().
8724	* Hence, loading of the guest state needs to be done -after- injection of events.
8725	*/
8726	rcStrict = hmR0VmxLoadGuestStateOptimal(pVM, pVCpu, pMixedCtx);
8727	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8728	{ /* likely */ }
8729	else
8730	{
8731	VMMRZCallRing3Enable(pVCpu);
8732	return rcStrict;
8733	}
8734
8735	/*
8736	* We disable interrupts so that we don't miss any interrupts that would flag preemption (IPI/timers etc.)
8737	* when thread-context hooks aren't used and we've been running with preemption disabled for a while.
8738	*
8739	* We need to check for force-flags that could've possible been altered since we last checked them (e.g.
8740	* by PDMGetInterrupt() leaving the PDM critical section, see @bugref{6398}).
8741	*
8742	* We also check a couple of other force-flags as a last opportunity to get the EMT back to ring-3 before
8743	* executing guest code.
8744	*/
8745	pVmxTransient->fEFlags = ASMIntDisableFlags();
8746
8747	if ( ( !VM_FF_IS_PENDING(pVM, VM_FF_EMT_RENDEZVOUS \| VM_FF_TM_VIRTUAL_SYNC)
8748	&& !VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
8749	\|\| ( fStepping /* Optimized for the non-stepping case, so a bit of unnecessary work when stepping. */
8750	&& !VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK & ~(VMCPU_FF_TIMER \| VMCPU_FF_PDM_CRITSECT))) )
8751	{
8752	if (!RTThreadPreemptIsPending(NIL_RTTHREAD))
8753	{
8754	pVCpu->hm.s.Event.fPending = false;
8755
8756	/*
8757	* We've injected any pending events. This is really the point of no return (to ring-3).
8758	*
8759	* Note! The caller expects to continue with interrupts & longjmps disabled on successful
8760	* returns from this function, so don't enable them here.
8761	*/
8762	return VINF_SUCCESS;
8763	}
8764
8765	STAM_COUNTER_INC(&pVCpu->hm.s.StatPendingHostIrq);
8766	rcStrict = VINF_EM_RAW_INTERRUPT;
8767	}
8768	else
8769	{
8770	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
8771	rcStrict = VINF_EM_RAW_TO_R3;
8772	}
8773
8774	ASMSetFlags(pVmxTransient->fEFlags);
8775	VMMRZCallRing3Enable(pVCpu);
8776
8777	return rcStrict;
8778	}
8779
8780
8781	/**
8782	* Prepares to run guest code in VT-x and we've committed to doing so. This
8783	* means there is no backing out to ring-3 or anywhere else at this
8784	* point.
8785	*
8786	* @param pVM The cross context VM structure.
8787	* @param pVCpu The cross context virtual CPU structure.
8788	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
8789	* out-of-sync. Make sure to update the required fields
8790	* before using them.
8791	* @param pVmxTransient Pointer to the VMX transient structure.
8792	*
8793	* @remarks Called with preemption disabled.
8794	* @remarks No-long-jump zone!!!
8795	*/
8796	static void hmR0VmxPreRunGuestCommitted(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
8797	{
8798	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8799	Assert(VMMR0IsLogFlushDisabled(pVCpu));
8800	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8801
8802	/*
8803	* Indicate start of guest execution and where poking EMT out of guest-context is recognized.
8804	*/
8805	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
8806	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC);
8807
8808	#ifdef HMVMX_ALWAYS_SWAP_FPU_STATE
8809	if (!CPUMIsGuestFPUStateActive(pVCpu))
8810	if (CPUMR0LoadGuestFPU(pVM, pVCpu) == VINF_CPUM_HOST_CR0_MODIFIED)
8811	HMCPU_CF_SET(pVCpu, HM_CHANGED_HOST_CONTEXT);
8812	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8813	#endif
8814
8815	if ( pVCpu->hm.s.fPreloadGuestFpu
8816	&& !CPUMIsGuestFPUStateActive(pVCpu))
8817	{
8818	if (CPUMR0LoadGuestFPU(pVM, pVCpu) == VINF_CPUM_HOST_CR0_MODIFIED)
8819	HMCPU_CF_SET(pVCpu, HM_CHANGED_HOST_CONTEXT);
8820	Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
8821	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8822	}
8823
8824	/*
8825	* Lazy-update of the host MSRs values in the auto-load/store MSR area.
8826	*/
8827	if ( !pVCpu->hm.s.vmx.fUpdatedHostMsrs
8828	&& pVCpu->hm.s.vmx.cMsrs > 0)
8829	{
8830	hmR0VmxUpdateAutoLoadStoreHostMsrs(pVCpu);
8831	}
8832
8833	/*
8834	* Load the host state bits as we may've been preempted (only happens when
8835	* thread-context hooks are used or when hmR0VmxSetupVMRunHandler() changes pfnStartVM).
8836	* Note that the 64-on-32 switcher saves the (64-bit) host state into the VMCS and
8837	* if we change the switcher back to 32-bit, we must save the 32-bit host state here.
8838	* See @bugref{8432}.
8839	*/
8840	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT))
8841	{
8842	int rc = hmR0VmxSaveHostState(pVM, pVCpu);
8843	AssertRC(rc);
8844	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreemptSaveHostState);
8845	}
8846	Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT));
8847
8848	/*
8849	* Load the state shared between host and guest (FPU, debug, lazy MSRs).
8850	*/
8851	if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE))
8852	hmR0VmxLoadSharedState(pVM, pVCpu, pMixedCtx);
8853	AssertMsg(!HMCPU_CF_VALUE(pVCpu), ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8854
8855	/* Store status of the shared guest-host state at the time of VM-entry. */
8856	#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
8857	if (CPUMIsGuestInLongModeEx(pMixedCtx))
8858	{
8859	pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActivePending(pVCpu);
8860	pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActivePending(pVCpu);
8861	}
8862	else
8863	#endif
8864	{
8865	pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
8866	pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
8867	}
8868	pVmxTransient->fWasGuestFPUStateActive = CPUMIsGuestFPUStateActive(pVCpu);
8869
8870	/*
8871	* Cache the TPR-shadow for checking on every VM-exit if it might have changed.
8872	*/
8873	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
8874	pVmxTransient->u8GuestTpr = pVCpu->hm.s.vmx.pbVirtApic[XAPIC_OFF_TPR];
8875
8876	PHMGLOBALCPUINFO pCpu = hmR0GetCurrentCpu();
8877	RTCPUID idCurrentCpu = pCpu->idCpu;
8878	if ( pVmxTransient->fUpdateTscOffsettingAndPreemptTimer
8879	\|\| idCurrentCpu != pVCpu->hm.s.idLastCpu)
8880	{
8881	hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVM, pVCpu);
8882	pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = false;
8883	}
8884
8885	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
8886	hmR0VmxFlushTaggedTlb(pVCpu, pCpu); /* Invalidate the appropriate guest entries from the TLB. */
8887	Assert(idCurrentCpu == pVCpu->hm.s.idLastCpu);
8888	pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Update the error reporting info. with the current host CPU. */
8889
8890	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
8891
8892	TMNotifyStartOfExecution(pVCpu); /* Finally, notify TM to resume its clocks as we're about
8893	to start executing. */
8894
8895	/*
8896	* Load the TSC_AUX MSR when we are not intercepting RDTSCP.
8897	*/
8898	if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP)
8899	{
8900	if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT))
8901	{
8902	bool fMsrUpdated;
8903	int rc2 = hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
8904	AssertRC(rc2);
8905	Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS));
8906
8907	rc2 = hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_TSC_AUX, CPUMR0GetGuestTscAux(pVCpu), true /* fUpdateHostMsr */,
8908	&fMsrUpdated);
8909	AssertRC(rc2);
8910	Assert(fMsrUpdated \|\| pVCpu->hm.s.vmx.fUpdatedHostMsrs);
8911
8912	/* Finally, mark that all host MSR values are updated so we don't redo it without leaving VT-x. See @bugref{6956}. */
8913	pVCpu->hm.s.vmx.fUpdatedHostMsrs = true;
8914	}
8915	else
8916	{
8917	hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, MSR_K8_TSC_AUX);
8918	Assert(!pVCpu->hm.s.vmx.cMsrs \|\| pVCpu->hm.s.vmx.fUpdatedHostMsrs);
8919	}
8920	}
8921
8922	#ifdef VBOX_STRICT
8923	hmR0VmxCheckAutoLoadStoreMsrs(pVCpu);
8924	hmR0VmxCheckHostEferMsr(pVCpu);
8925	AssertRC(hmR0VmxCheckVmcsCtls(pVCpu));
8926	#endif
8927	#ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE
8928	uint32_t uInvalidReason = hmR0VmxCheckGuestState(pVM, pVCpu, pMixedCtx);
8929	if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND)
8930	Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason));
8931	#endif
8932	}
8933
8934
8935	/**
8936	* Performs some essential restoration of state after running guest code in
8937	* VT-x.
8938	*
8939	* @param pVM The cross context VM structure.
8940	* @param pVCpu The cross context virtual CPU structure.
8941	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
8942	* out-of-sync. Make sure to update the required fields
8943	* before using them.
8944	* @param pVmxTransient Pointer to the VMX transient structure.
8945	* @param rcVMRun Return code of VMLAUNCH/VMRESUME.
8946	*
8947	* @remarks Called with interrupts disabled, and returns with interrupts enabled!
8948	*
8949	* @remarks No-long-jump zone!!! This function will however re-enable longjmps
8950	* unconditionally when it is safe to do so.
8951	*/
8952	static void hmR0VmxPostRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, int rcVMRun)
8953	{
8954	NOREF(pVM);
8955
8956	Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8957
8958	ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
8959	ASMAtomicIncU32(&pVCpu->hm.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
8960	HMVMXCPU_GST_RESET_TO(pVCpu, 0); /* Exits/longjmps to ring-3 requires saving the guest state. */
8961	pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */
8962	pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
8963	pVmxTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
8964
8965	if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT))
8966	TMCpuTickSetLastSeen(pVCpu, ASMReadTSC() + pVCpu->hm.s.vmx.u64TSCOffset);
8967
8968	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatExit1, x);
8969	TMNotifyEndOfExecution(pVCpu); /* Notify TM that the guest is no longer running. */
8970	Assert(!ASMIntAreEnabled());
8971	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
8972
8973	#ifdef HMVMX_ALWAYS_SWAP_FPU_STATE
8974	if (CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVM, pVCpu))
8975	{
8976	hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
8977	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8978	}
8979	#endif
8980
8981	#if HC_ARCH_BITS == 64
8982	pVCpu->hm.s.vmx.fRestoreHostFlags \|= VMX_RESTORE_HOST_REQUIRED; /* Host state messed up by VT-x, we must restore. */
8983	#endif
8984	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
8985	/* The 64-on-32 switcher maintains uVmcsState on its own and we need to leave it alone here. */
8986	if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0SwitcherStartVM64)
8987	pVCpu->hm.s.vmx.uVmcsState \|= HMVMX_VMCS_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
8988	#else
8989	pVCpu->hm.s.vmx.uVmcsState \|= HMVMX_VMCS_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
8990	#endif
8991	#ifdef VBOX_STRICT
8992	hmR0VmxCheckHostEferMsr(pVCpu); /* Verify that VMRUN/VMLAUNCH didn't modify host EFER. */
8993	#endif
8994	ASMSetFlags(pVmxTransient->fEFlags); /* Enable interrupts. */
8995	VMMRZCallRing3Enable(pVCpu); /* It is now safe to do longjmps to ring-3!!! */
8996
8997	/* Save the basic VM-exit reason. Refer Intel spec. 24.9.1 "Basic VM-exit Information". */
8998	uint32_t uExitReason;
8999	int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
9000	rc \|= hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
9001	AssertRC(rc);
9002	pVmxTransient->uExitReason = (uint16_t)VMX_EXIT_REASON_BASIC(uExitReason);
9003	pVmxTransient->fVMEntryFailed = VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uEntryIntInfo);
9004
9005	/* If the VMLAUNCH/VMRESUME failed, we can bail out early. This does -not- cover VMX_EXIT_ERR_. /
9006	if (RT_UNLIKELY(rcVMRun != VINF_SUCCESS))
9007	{
9008	Log4(("VM-entry failure: pVCpu=%p idCpu=%RU32 rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", pVCpu, pVCpu->idCpu, rcVMRun,
9009	pVmxTransient->fVMEntryFailed));
9010	return;
9011	}
9012
9013	/*
9014	* Update the VM-exit history array here even if the VM-entry failed due to:
9015	* - Invalid guest state.
9016	* - MSR loading.
9017	* - Machine-check event.
9018	*
9019	* In any of the above cases we will still have a "valid" VM-exit reason
9020	* despite @a fVMEntryFailed being false.
9021	*
9022	* See Intel spec. 26.7 "VM-Entry failures during or after loading guest state".
9023	*/
9024	HMCPU_EXIT_HISTORY_ADD(pVCpu, pVmxTransient->uExitReason);
9025
9026	if (RT_LIKELY(!pVmxTransient->fVMEntryFailed))
9027	{
9028	/** @todo We can optimize this by only syncing with our force-flags when
9029	* really needed and keeping the VMCS state as it is for most
9030	* VM-exits. */
9031	/* Update the guest interruptibility-state from the VMCS. */
9032	hmR0VmxSaveGuestIntrState(pVCpu, pMixedCtx);
9033
9034	#if defined(HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE) \|\| defined(HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE)
9035	rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
9036	AssertRC(rc);
9037	#elif defined(HMVMX_ALWAYS_SAVE_GUEST_RFLAGS)
9038	rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
9039	AssertRC(rc);
9040	#endif
9041
9042	/*
9043	* Sync the TPR shadow with our APIC state.
9044	*/
9045	if ( (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
9046	&& pVmxTransient->u8GuestTpr != pVCpu->hm.s.vmx.pbVirtApic[XAPIC_OFF_TPR])
9047	{
9048	rc = APICSetTpr(pVCpu, pVCpu->hm.s.vmx.pbVirtApic[XAPIC_OFF_TPR]);
9049	AssertRC(rc);
9050	HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
9051	}
9052	}
9053	}
9054
9055
9056	/**
9057	* Runs the guest code using VT-x the normal way.
9058	*
9059	* @returns VBox status code.
9060	* @param pVM The cross context VM structure.
9061	* @param pVCpu The cross context virtual CPU structure.
9062	* @param pCtx Pointer to the guest-CPU context.
9063	*
9064	* @note Mostly the same as hmR0VmxRunGuestCodeStep().
9065	*/
9066	static VBOXSTRICTRC hmR0VmxRunGuestCodeNormal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
9067	{
9068	VMXTRANSIENT VmxTransient;
9069	VmxTransient.fUpdateTscOffsettingAndPreemptTimer = true;
9070	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
9071	uint32_t cLoops = 0;
9072
9073	for (;; cLoops++)
9074	{
9075	Assert(!HMR0SuspendPending());
9076	HMVMX_ASSERT_CPU_SAFE();
9077
9078	/* Preparatory work for running guest code, this may force us to return
9079	to ring-3. This bugger disables interrupts on VINF_SUCCESS! */
9080	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
9081	rcStrict = hmR0VmxPreRunGuest(pVM, pVCpu, pCtx, &VmxTransient, false /* fStepping */);
9082	if (rcStrict != VINF_SUCCESS)
9083	break;
9084
9085	hmR0VmxPreRunGuestCommitted(pVM, pVCpu, pCtx, &VmxTransient);
9086	int rcRun = hmR0VmxRunGuest(pVM, pVCpu, pCtx);
9087	/* The guest-CPU context is now outdated, 'pCtx' is to be treated as 'pMixedCtx' from this point on!!! */
9088
9089	/* Restore any residual host-state and save any bits shared between host
9090	and guest into the guest-CPU state. Re-enables interrupts! */
9091	hmR0VmxPostRunGuest(pVM, pVCpu, pCtx, &VmxTransient, rcRun);
9092
9093	/* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
9094	if (RT_SUCCESS(rcRun))
9095	{ /* very likely */ }
9096	else
9097	{
9098	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x);
9099	hmR0VmxReportWorldSwitchError(pVM, pVCpu, rcRun, pCtx, &VmxTransient);
9100	return rcRun;
9101	}
9102
9103	/* Profile the VM-exit. */
9104	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
9105	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
9106	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
9107	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
9108	HMVMX_START_EXIT_DISPATCH_PROF();
9109
9110	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, pCtx, VmxTransient.uExitReason);
9111
9112	/* Handle the VM-exit. */
9113	#ifdef HMVMX_USE_FUNCTION_TABLE
9114	rcStrict = g_apfnVMExitHandlers[VmxTransient.uExitReason](pVCpu, pCtx, &VmxTransient);
9115	#else
9116	rcStrict = hmR0VmxHandleExit(pVCpu, pCtx, &VmxTransient, VmxTransient.uExitReason);
9117	#endif
9118	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
9119	if (rcStrict == VINF_SUCCESS)
9120	{
9121	if (cLoops <= pVM->hm.s.cMaxResumeLoops)
9122	continue; /* likely */
9123	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
9124	rcStrict = VINF_EM_RAW_INTERRUPT;
9125	}
9126	break;
9127	}
9128
9129	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
9130	return rcStrict;
9131	}
9132
9133
9134
9135	/** @name Execution loop for single stepping, DBGF events and expensive Dtrace
9136	* probes.
9137	*
9138	* The following few functions and associated structure contains the bloat
9139	* necessary for providing detailed debug events and dtrace probes as well as
9140	* reliable host side single stepping. This works on the principle of
9141	* "subclassing" the normal execution loop and workers. We replace the loop
9142	* method completely and override selected helpers to add necessary adjustments
9143	* to their core operation.
9144	*
9145	* The goal is to keep the "parent" code lean and mean, so as not to sacrifice
9146	* any performance for debug and analysis features.
9147	*
9148	* @{
9149	*/
9150
9151	/**
9152	* Transient per-VCPU debug state of VMCS and related info. we save/restore in
9153	* the debug run loop.
9154	*/
9155	typedef struct VMXRUNDBGSTATE
9156	{
9157	/** The RIP we started executing at. This is for detecting that we stepped. */
9158	uint64_t uRipStart;
9159	/** The CS we started executing with. */
9160	uint16_t uCsStart;
9161
9162	/** Whether we've actually modified the 1st execution control field. */
9163	bool fModifiedProcCtls : 1;
9164	/** Whether we've actually modified the 2nd execution control field. */
9165	bool fModifiedProcCtls2 : 1;
9166	/** Whether we've actually modified the exception bitmap. */
9167	bool fModifiedXcptBitmap : 1;
9168
9169	/** We desire the modified the CR0 mask to be cleared. */
9170	bool fClearCr0Mask : 1;
9171	/** We desire the modified the CR4 mask to be cleared. */
9172	bool fClearCr4Mask : 1;
9173	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC. */
9174	uint32_t fCpe1Extra;
9175	/** Stuff we do not want in VMX_VMCS32_CTRL_PROC_EXEC. */
9176	uint32_t fCpe1Unwanted;
9177	/** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC2. */
9178	uint32_t fCpe2Extra;
9179	/** Extra stuff we need in VMX_VMCS32_CTRL_EXCEPTION_BITMAP. */
9180	uint32_t bmXcptExtra;
9181	/** The sequence number of the Dtrace provider settings the state was
9182	* configured against. */
9183	uint32_t uDtraceSettingsSeqNo;
9184	/** VM-exits to check (one bit per VM-exit). */
9185	uint32_t bmExitsToCheck[3];
9186
9187	/** The initial VMX_VMCS32_CTRL_PROC_EXEC value (helps with restore). */
9188	uint32_t fProcCtlsInitial;
9189	/** The initial VMX_VMCS32_CTRL_PROC_EXEC2 value (helps with restore). */
9190	uint32_t fProcCtls2Initial;
9191	/** The initial VMX_VMCS32_CTRL_EXCEPTION_BITMAP value (helps with restore). */
9192	uint32_t bmXcptInitial;
9193	} VMXRUNDBGSTATE;
9194	AssertCompileMemberSize(VMXRUNDBGSTATE, bmExitsToCheck, (VMX_EXIT_MAX + 1 + 31) / 32 * 4);
9195	typedef VMXRUNDBGSTATE *PVMXRUNDBGSTATE;
9196
9197
9198	/**
9199	* Initializes the VMXRUNDBGSTATE structure.
9200	*
9201	* @param pVCpu The cross context virtual CPU structure of the
9202	* calling EMT.
9203	* @param pCtx The CPU register context to go with @a pVCpu.
9204	* @param pDbgState The structure to initialize.
9205	*/
9206	DECLINLINE(void) hmR0VmxRunDebugStateInit(PVMCPU pVCpu, PCCPUMCTX pCtx, PVMXRUNDBGSTATE pDbgState)
9207	{
9208	pDbgState->uRipStart = pCtx->rip;
9209	pDbgState->uCsStart = pCtx->cs.Sel;
9210
9211	pDbgState->fModifiedProcCtls = false;
9212	pDbgState->fModifiedProcCtls2 = false;
9213	pDbgState->fModifiedXcptBitmap = false;
9214	pDbgState->fClearCr0Mask = false;
9215	pDbgState->fClearCr4Mask = false;
9216	pDbgState->fCpe1Extra = 0;
9217	pDbgState->fCpe1Unwanted = 0;
9218	pDbgState->fCpe2Extra = 0;
9219	pDbgState->bmXcptExtra = 0;
9220	pDbgState->fProcCtlsInitial = pVCpu->hm.s.vmx.u32ProcCtls;
9221	pDbgState->fProcCtls2Initial = pVCpu->hm.s.vmx.u32ProcCtls2;
9222	pDbgState->bmXcptInitial = pVCpu->hm.s.vmx.u32XcptBitmap;
9223	}
9224
9225
9226	/**
9227	* Updates the VMSC fields with changes requested by @a pDbgState.
9228	*
9229	* This is performed after hmR0VmxPreRunGuestDebugStateUpdate as well
9230	* immediately before executing guest code, i.e. when interrupts are disabled.
9231	* We don't check status codes here as we cannot easily assert or return in the
9232	* latter case.
9233	*
9234	* @param pVCpu The cross context virtual CPU structure.
9235	* @param pDbgState The debug state.
9236	*/
9237	DECLINLINE(void) hmR0VmxPreRunGuestDebugStateApply(PVMCPU pVCpu, PVMXRUNDBGSTATE pDbgState)
9238	{
9239	/*
9240	* Ensure desired flags in VMCS control fields are set.
9241	* (Ignoring write failure here, as we're committed and it's just debug extras.)
9242	*
9243	* Note! We load the shadow CR0 & CR4 bits when we flag the clearing, so
9244	* there should be no stale data in pCtx at this point.
9245	*/
9246	if ( (pVCpu->hm.s.vmx.u32ProcCtls & pDbgState->fCpe1Extra) != pDbgState->fCpe1Extra
9247	\|\| (pVCpu->hm.s.vmx.u32ProcCtls & pDbgState->fCpe1Unwanted))
9248	{
9249	pVCpu->hm.s.vmx.u32ProcCtls \|= pDbgState->fCpe1Extra;
9250	pVCpu->hm.s.vmx.u32ProcCtls &= ~pDbgState->fCpe1Unwanted;
9251	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
9252	Log6(("hmR0VmxRunDebugStateRevert: VMX_VMCS32_CTRL_PROC_EXEC: %#RX32\n", pVCpu->hm.s.vmx.u32ProcCtls));
9253	pDbgState->fModifiedProcCtls = true;
9254	}
9255
9256	if ((pVCpu->hm.s.vmx.u32ProcCtls2 & pDbgState->fCpe2Extra) != pDbgState->fCpe2Extra)
9257	{
9258	pVCpu->hm.s.vmx.u32ProcCtls2 \|= pDbgState->fCpe2Extra;
9259	VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pVCpu->hm.s.vmx.u32ProcCtls2);
9260	Log6(("hmR0VmxRunDebugStateRevert: VMX_VMCS32_CTRL_PROC_EXEC2: %#RX32\n", pVCpu->hm.s.vmx.u32ProcCtls2));
9261	pDbgState->fModifiedProcCtls2 = true;
9262	}
9263
9264	if ((pVCpu->hm.s.vmx.u32XcptBitmap & pDbgState->bmXcptExtra) != pDbgState->bmXcptExtra)
9265	{
9266	pVCpu->hm.s.vmx.u32XcptBitmap \|= pDbgState->bmXcptExtra;
9267	VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
9268	Log6(("hmR0VmxRunDebugStateRevert: VMX_VMCS32_CTRL_EXCEPTION_BITMAP: %#RX32\n", pVCpu->hm.s.vmx.u32XcptBitmap));
9269	pDbgState->fModifiedXcptBitmap = true;
9270	}
9271
9272	if (pDbgState->fClearCr0Mask && pVCpu->hm.s.vmx.u32CR0Mask != 0)
9273	{
9274	pVCpu->hm.s.vmx.u32CR0Mask = 0;
9275	VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_MASK, 0);
9276	Log6(("hmR0VmxRunDebugStateRevert: VMX_VMCS_CTRL_CR0_MASK: 0\n"));
9277	}
9278
9279	if (pDbgState->fClearCr4Mask && pVCpu->hm.s.vmx.u32CR4Mask != 0)
9280	{
9281	pVCpu->hm.s.vmx.u32CR4Mask = 0;
9282	VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_MASK, 0);
9283	Log6(("hmR0VmxRunDebugStateRevert: VMX_VMCS_CTRL_CR4_MASK: 0\n"));
9284	}
9285	}
9286
9287
9288	DECLINLINE(VBOXSTRICTRC) hmR0VmxRunDebugStateRevert(PVMCPU pVCpu, PVMXRUNDBGSTATE pDbgState, VBOXSTRICTRC rcStrict)
9289	{
9290	/*
9291	* Restore VM-exit control settings as we may not reenter this function the
9292	* next time around.
9293	*/
9294	/* We reload the initial value, trigger what we can of recalculations the
9295	next time around. From the looks of things, that's all that's required atm. */
9296	if (pDbgState->fModifiedProcCtls)
9297	{
9298	if (!(pDbgState->fProcCtlsInitial & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT) && CPUMIsHyperDebugStateActive(pVCpu))
9299	pDbgState->fProcCtlsInitial \|= VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT; /* Avoid assertion in hmR0VmxLeave */
9300	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pDbgState->fProcCtlsInitial);
9301	AssertRCReturn(rc2, rc2);
9302	pVCpu->hm.s.vmx.u32ProcCtls = pDbgState->fProcCtlsInitial;
9303	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0 \| HM_CHANGED_GUEST_DEBUG);
9304	}
9305
9306	/* We're currently the only ones messing with this one, so just restore the
9307	cached value and reload the field. */
9308	if ( pDbgState->fModifiedProcCtls2
9309	&& pVCpu->hm.s.vmx.u32ProcCtls2 != pDbgState->fProcCtls2Initial)
9310	{
9311	int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pDbgState->fProcCtls2Initial);
9312	AssertRCReturn(rc2, rc2);
9313	pVCpu->hm.s.vmx.u32ProcCtls2 = pDbgState->fProcCtls2Initial;
9314	}
9315
9316	/* If we've modified the exception bitmap, we restore it and trigger
9317	reloading and partial recalculation the next time around. */
9318	if (pDbgState->fModifiedXcptBitmap)
9319	{
9320	pVCpu->hm.s.vmx.u32XcptBitmap = pDbgState->bmXcptInitial;
9321	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS \| HM_CHANGED_GUEST_CR0);
9322	}
9323
9324	/* We assume hmR0VmxLoadSharedCR0 will recalculate and load the CR0 mask. */
9325	if (pDbgState->fClearCr0Mask)
9326	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
9327
9328	/* We assume hmR0VmxLoadGuestCR3AndCR4 will recalculate and load the CR4 mask. */
9329	if (pDbgState->fClearCr4Mask)
9330	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR4);
9331
9332	return rcStrict;
9333	}
9334
9335
9336	/**
9337	* Configures VM-exit controls for current DBGF and DTrace settings.
9338	*
9339	* This updates @a pDbgState and the VMCS execution control fields to reflect
9340	* the necessary VM-exits demanded by DBGF and DTrace.
9341	*
9342	* @param pVM The cross context VM structure.
9343	* @param pVCpu The cross context virtual CPU structure.
9344	* @param pCtx Pointer to the guest-CPU context.
9345	* @param pDbgState The debug state.
9346	* @param pVmxTransient Pointer to the VMX transient structure. May update
9347	* fUpdateTscOffsettingAndPreemptTimer.
9348	*/
9349	static void hmR0VmxPreRunGuestDebugStateUpdate(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx,
9350	PVMXRUNDBGSTATE pDbgState, PVMXTRANSIENT pVmxTransient)
9351	{
9352	/*
9353	* Take down the dtrace serial number so we can spot changes.
9354	*/
9355	pDbgState->uDtraceSettingsSeqNo = VBOXVMM_GET_SETTINGS_SEQ_NO();
9356	ASMCompilerBarrier();
9357
9358	/*
9359	* We'll rebuild most of the middle block of data members (holding the
9360	* current settings) as we go along here, so start by clearing it all.
9361	*/
9362	pDbgState->bmXcptExtra = 0;
9363	pDbgState->fCpe1Extra = 0;
9364	pDbgState->fCpe1Unwanted = 0;
9365	pDbgState->fCpe2Extra = 0;
9366	for (unsigned i = 0; i < RT_ELEMENTS(pDbgState->bmExitsToCheck); i++)
9367	pDbgState->bmExitsToCheck[i] = 0;
9368
9369	/*
9370	* Software interrupts (INT XXh) - no idea how to trigger these...
9371	*/
9372	if ( DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_INTERRUPT_SOFTWARE)
9373	\|\| VBOXVMM_INT_SOFTWARE_ENABLED())
9374	{
9375	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
9376	}
9377
9378	/*
9379	* INT3 breakpoints - triggered by #BP exceptions.
9380	*/
9381	if (pVM->dbgf.ro.cEnabledInt3Breakpoints > 0)
9382	pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
9383
9384	/*
9385	* Exception bitmap and XCPT events+probes.
9386	*/
9387	for (int iXcpt = 0; iXcpt < (DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST + 1); iXcpt++)
9388	if (DBGF_IS_EVENT_ENABLED(pVM, (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + iXcpt)))
9389	pDbgState->bmXcptExtra \|= RT_BIT_32(iXcpt);
9390
9391	if (VBOXVMM_XCPT_DE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DE);
9392	if (VBOXVMM_XCPT_DB_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DB);
9393	if (VBOXVMM_XCPT_BP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BP);
9394	if (VBOXVMM_XCPT_OF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_OF);
9395	if (VBOXVMM_XCPT_BR_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_BR);
9396	if (VBOXVMM_XCPT_UD_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_UD);
9397	if (VBOXVMM_XCPT_NM_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NM);
9398	if (VBOXVMM_XCPT_DF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_DF);
9399	if (VBOXVMM_XCPT_TS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_TS);
9400	if (VBOXVMM_XCPT_NP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_NP);
9401	if (VBOXVMM_XCPT_SS_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SS);
9402	if (VBOXVMM_XCPT_GP_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_GP);
9403	if (VBOXVMM_XCPT_PF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_PF);
9404	if (VBOXVMM_XCPT_MF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_MF);
9405	if (VBOXVMM_XCPT_AC_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_AC);
9406	if (VBOXVMM_XCPT_XF_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_XF);
9407	if (VBOXVMM_XCPT_VE_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_VE);
9408	if (VBOXVMM_XCPT_SX_ENABLED()) pDbgState->bmXcptExtra \|= RT_BIT_32(X86_XCPT_SX);
9409
9410	if (pDbgState->bmXcptExtra)
9411	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
9412
9413	/*
9414	* Process events and probes for VM-exits, making sure we get the wanted VM-exits.
9415	*
9416	* Note! This is the reverse of waft hmR0VmxHandleExitDtraceEvents does.
9417	* So, when adding/changing/removing please don't forget to update it.
9418	*
9419	* Some of the macros are picking up local variables to save horizontal space,
9420	* (being able to see it in a table is the lesser evil here).
9421	*/
9422	#define IS_EITHER_ENABLED(a_pVM, a_EventSubName) \
9423	( DBGF_IS_EVENT_ENABLED(a_pVM, RT_CONCAT(DBGFEVENT_, a_EventSubName)) \
9424	\|\| RT_CONCAT3(VBOXVMM_, a_EventSubName, _ENABLED)() )
9425	#define SET_ONLY_XBM_IF_EITHER_EN(a_EventSubName, a_uExit) \
9426	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
9427	{ AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
9428	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
9429	} else do { } while (0)
9430	#define SET_CPE1_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec) \
9431	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
9432	{ \
9433	(pDbgState)->fCpe1Extra \|= (a_fCtrlProcExec); \
9434	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
9435	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
9436	} else do { } while (0)
9437	#define SET_CPEU_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fUnwantedCtrlProcExec) \
9438	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
9439	{ \
9440	(pDbgState)->fCpe1Unwanted \|= (a_fUnwantedCtrlProcExec); \
9441	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
9442	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
9443	} else do { } while (0)
9444	#define SET_CPE2_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec2) \
9445	if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
9446	{ \
9447	(pDbgState)->fCpe2Extra \|= (a_fCtrlProcExec2); \
9448	AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
9449	ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
9450	} else do { } while (0)
9451
9452	SET_ONLY_XBM_IF_EITHER_EN(EXIT_TASK_SWITCH, VMX_EXIT_TASK_SWITCH); /* unconditional */
9453	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_VIOLATION, VMX_EXIT_EPT_VIOLATION); /* unconditional */
9454	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_MISCONFIG, VMX_EXIT_EPT_MISCONFIG); /* unconditional (unless #VE) */
9455	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_ACCESS, VMX_EXIT_APIC_ACCESS); /* feature dependent, nothing to enable here */
9456	SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_WRITE, VMX_EXIT_APIC_WRITE); /* feature dependent, nothing to enable here */
9457
9458	SET_ONLY_XBM_IF_EITHER_EN(INSTR_CPUID, VMX_EXIT_CPUID); /* unconditional */
9459	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CPUID, VMX_EXIT_CPUID);
9460	SET_ONLY_XBM_IF_EITHER_EN(INSTR_GETSEC, VMX_EXIT_GETSEC); /* unconditional */
9461	SET_ONLY_XBM_IF_EITHER_EN( EXIT_GETSEC, VMX_EXIT_GETSEC);
9462	SET_CPE1_XBM_IF_EITHER_EN(INSTR_HALT, VMX_EXIT_HLT, VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT); /* paranoia */
9463	SET_ONLY_XBM_IF_EITHER_EN( EXIT_HALT, VMX_EXIT_HLT);
9464	SET_ONLY_XBM_IF_EITHER_EN(INSTR_INVD, VMX_EXIT_INVD); /* unconditional */
9465	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVD, VMX_EXIT_INVD);
9466	SET_CPE1_XBM_IF_EITHER_EN(INSTR_INVLPG, VMX_EXIT_INVLPG, VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT);
9467	SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVLPG, VMX_EXIT_INVLPG);
9468	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDPMC, VMX_EXIT_RDPMC, VMX_VMCS_CTRL_PROC_EXEC_RDPMC_EXIT);
9469	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDPMC, VMX_EXIT_RDPMC);
9470	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSC, VMX_EXIT_RDTSC, VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT);
9471	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSC, VMX_EXIT_RDTSC);
9472	SET_ONLY_XBM_IF_EITHER_EN(INSTR_RSM, VMX_EXIT_RSM); /* unconditional */
9473	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RSM, VMX_EXIT_RSM);
9474	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMM_CALL, VMX_EXIT_VMCALL); /* unconditional */
9475	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMM_CALL, VMX_EXIT_VMCALL);
9476	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMCLEAR, VMX_EXIT_VMCLEAR); /* unconditional */
9477	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMCLEAR, VMX_EXIT_VMCLEAR);
9478	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH); /* unconditional */
9479	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH);
9480	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRLD, VMX_EXIT_VMPTRLD); /* unconditional */
9481	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRLD, VMX_EXIT_VMPTRLD);
9482	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRST, VMX_EXIT_VMPTRST); /* unconditional */
9483	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRST, VMX_EXIT_VMPTRST);
9484	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMREAD, VMX_EXIT_VMREAD); /* unconditional */
9485	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMREAD, VMX_EXIT_VMREAD);
9486	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMRESUME, VMX_EXIT_VMRESUME); /* unconditional */
9487	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMRESUME, VMX_EXIT_VMRESUME);
9488	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMWRITE, VMX_EXIT_VMWRITE); /* unconditional */
9489	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMWRITE, VMX_EXIT_VMWRITE);
9490	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXOFF, VMX_EXIT_VMXOFF); /* unconditional */
9491	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXOFF, VMX_EXIT_VMXOFF);
9492	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXON, VMX_EXIT_VMXON); /* unconditional */
9493	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXON, VMX_EXIT_VMXON);
9494
9495	if ( IS_EITHER_ENABLED(pVM, INSTR_CRX_READ)
9496	\|\| IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
9497	{
9498	int rc2 = hmR0VmxSaveGuestCR0(pVCpu, pCtx);
9499	rc2 \|= hmR0VmxSaveGuestCR4(pVCpu, pCtx);
9500	rc2 \|= hmR0VmxSaveGuestApicState(pVCpu, pCtx);
9501	AssertRC(rc2);
9502
9503	#if 0 /** @todo fix me */
9504	pDbgState->fClearCr0Mask = true;
9505	pDbgState->fClearCr4Mask = true;
9506	#endif
9507	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_READ))
9508	pDbgState->fCpe1Extra \|= VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT \| VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT;
9509	if (IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
9510	pDbgState->fCpe1Extra \|= VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT \| VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT;
9511	pDbgState->fCpe1Unwanted \|= VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW; /* risky? */
9512	/* Note! We currently don't use VMX_VMCS32_CTRL_CR3_TARGET_COUNT. It would
9513	require clearing here and in the loop if we start using it. */
9514	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_CRX);
9515	}
9516	else
9517	{
9518	if (pDbgState->fClearCr0Mask)
9519	{
9520	pDbgState->fClearCr0Mask = false;
9521	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
9522	}
9523	if (pDbgState->fClearCr4Mask)
9524	{
9525	pDbgState->fClearCr4Mask = false;
9526	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR4);
9527	}
9528	}
9529	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_READ, VMX_EXIT_MOV_CRX);
9530	SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_WRITE, VMX_EXIT_MOV_CRX);
9531
9532	if ( IS_EITHER_ENABLED(pVM, INSTR_DRX_READ)
9533	\|\| IS_EITHER_ENABLED(pVM, INSTR_DRX_WRITE))
9534	{
9535	/** @todo later, need to fix handler as it assumes this won't usually happen. */
9536	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_DRX);
9537	}
9538	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_READ, VMX_EXIT_MOV_DRX);
9539	SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_WRITE, VMX_EXIT_MOV_DRX);
9540
9541	SET_CPEU_XBM_IF_EITHER_EN(INSTR_RDMSR, VMX_EXIT_RDMSR, VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS); /* risky clearing this? */
9542	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDMSR, VMX_EXIT_RDMSR);
9543	SET_CPEU_XBM_IF_EITHER_EN(INSTR_WRMSR, VMX_EXIT_WRMSR, VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS);
9544	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WRMSR, VMX_EXIT_WRMSR);
9545	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MWAIT, VMX_EXIT_MWAIT, VMX_VMCS_CTRL_PROC_EXEC_MWAIT_EXIT); /* paranoia */
9546	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MWAIT, VMX_EXIT_MWAIT);
9547	SET_CPE1_XBM_IF_EITHER_EN(INSTR_MONITOR, VMX_EXIT_MONITOR, VMX_VMCS_CTRL_PROC_EXEC_MONITOR_EXIT); /* paranoia */
9548	SET_ONLY_XBM_IF_EITHER_EN( EXIT_MONITOR, VMX_EXIT_MONITOR);
9549	#if 0 /** @todo too slow, fix handler. */
9550	SET_CPE1_XBM_IF_EITHER_EN(INSTR_PAUSE, VMX_EXIT_PAUSE, VMX_VMCS_CTRL_PROC_EXEC_PAUSE_EXIT);
9551	#endif
9552	SET_ONLY_XBM_IF_EITHER_EN( EXIT_PAUSE, VMX_EXIT_PAUSE);
9553
9554	if ( IS_EITHER_ENABLED(pVM, INSTR_SGDT)
9555	\|\| IS_EITHER_ENABLED(pVM, INSTR_SIDT)
9556	\|\| IS_EITHER_ENABLED(pVM, INSTR_LGDT)
9557	\|\| IS_EITHER_ENABLED(pVM, INSTR_LIDT))
9558	{
9559	pDbgState->fCpe2Extra \|= VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT;
9560	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XDTR_ACCESS);
9561	}
9562	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SGDT, VMX_EXIT_XDTR_ACCESS);
9563	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SIDT, VMX_EXIT_XDTR_ACCESS);
9564	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LGDT, VMX_EXIT_XDTR_ACCESS);
9565	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LIDT, VMX_EXIT_XDTR_ACCESS);
9566
9567	if ( IS_EITHER_ENABLED(pVM, INSTR_SLDT)
9568	\|\| IS_EITHER_ENABLED(pVM, INSTR_STR)
9569	\|\| IS_EITHER_ENABLED(pVM, INSTR_LLDT)
9570	\|\| IS_EITHER_ENABLED(pVM, INSTR_LTR))
9571	{
9572	pDbgState->fCpe2Extra \|= VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT;
9573	ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_TR_ACCESS);
9574	}
9575	SET_ONLY_XBM_IF_EITHER_EN( EXIT_SLDT, VMX_EXIT_TR_ACCESS);
9576	SET_ONLY_XBM_IF_EITHER_EN( EXIT_STR, VMX_EXIT_TR_ACCESS);
9577	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LLDT, VMX_EXIT_TR_ACCESS);
9578	SET_ONLY_XBM_IF_EITHER_EN( EXIT_LTR, VMX_EXIT_TR_ACCESS);
9579
9580	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVEPT, VMX_EXIT_INVEPT); /* unconditional */
9581	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVEPT, VMX_EXIT_INVEPT);
9582	SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSCP, VMX_EXIT_RDTSCP, VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT);
9583	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSCP, VMX_EXIT_RDTSCP);
9584	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVVPID, VMX_EXIT_INVVPID); /* unconditional */
9585	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVVPID, VMX_EXIT_INVVPID);
9586	SET_CPE2_XBM_IF_EITHER_EN(INSTR_WBINVD, VMX_EXIT_WBINVD, VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT);
9587	SET_ONLY_XBM_IF_EITHER_EN( EXIT_WBINVD, VMX_EXIT_WBINVD);
9588	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSETBV, VMX_EXIT_XSETBV); /* unconditional */
9589	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XSETBV, VMX_EXIT_XSETBV);
9590	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDRAND, VMX_EXIT_RDRAND, VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT);
9591	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDRAND, VMX_EXIT_RDRAND);
9592	SET_CPE1_XBM_IF_EITHER_EN(INSTR_VMX_INVPCID, VMX_EXIT_INVPCID, VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT);
9593	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVPCID, VMX_EXIT_INVPCID);
9594	SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMFUNC, VMX_EXIT_VMFUNC); /* unconditional for the current setup */
9595	SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMFUNC, VMX_EXIT_VMFUNC);
9596	SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDSEED, VMX_EXIT_RDSEED, VMX_VMCS_CTRL_PROC_EXEC2_RDSEED_EXIT);
9597	SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDSEED, VMX_EXIT_RDSEED);
9598	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSAVES, VMX_EXIT_XSAVES); /* unconditional (enabled by host, guest cfg) */
9599	SET_ONLY_XBM_IF_EITHER_EN(EXIT_XSAVES, VMX_EXIT_XSAVES);
9600	SET_ONLY_XBM_IF_EITHER_EN(INSTR_XRSTORS, VMX_EXIT_XRSTORS); /* unconditional (enabled by host, guest cfg) */
9601	SET_ONLY_XBM_IF_EITHER_EN( EXIT_XRSTORS, VMX_EXIT_XRSTORS);
9602
9603	#undef IS_EITHER_ENABLED
9604	#undef SET_ONLY_XBM_IF_EITHER_EN
9605	#undef SET_CPE1_XBM_IF_EITHER_EN
9606	#undef SET_CPEU_XBM_IF_EITHER_EN
9607	#undef SET_CPE2_XBM_IF_EITHER_EN
9608
9609	/*
9610	* Sanitize the control stuff.
9611	*/
9612	pDbgState->fCpe2Extra &= pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1;
9613	if (pDbgState->fCpe2Extra)
9614	pDbgState->fCpe1Extra \|= VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL;
9615	pDbgState->fCpe1Extra &= pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1;
9616	pDbgState->fCpe1Unwanted &= ~pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0;
9617	if (pVCpu->hm.s.fDebugWantRdTscExit != RT_BOOL(pDbgState->fCpe1Extra & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT))
9618	{
9619	pVCpu->hm.s.fDebugWantRdTscExit ^= true;
9620	pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
9621	}
9622
9623	Log6(("HM: debug state: cpe1=%#RX32 cpeu=%#RX32 cpe2=%#RX32%s%s\n",
9624	pDbgState->fCpe1Extra, pDbgState->fCpe1Unwanted, pDbgState->fCpe2Extra,
9625	pDbgState->fClearCr0Mask ? " clr-cr0" : "",
9626	pDbgState->fClearCr4Mask ? " clr-cr4" : ""));
9627	}
9628
9629
9630	/**
9631	* Fires off DBGF events and dtrace probes for a VM-exit, when it's
9632	* appropriate.
9633	*
9634	* The caller has checked the VM-exit against the
9635	* VMXRUNDBGSTATE::bmExitsToCheck bitmap. The caller has checked for NMIs
9636	* already, so we don't have to do that either.
9637	*
9638	* @returns Strict VBox status code (i.e. informational status codes too).
9639	* @param pVM The cross context VM structure.
9640	* @param pVCpu The cross context virtual CPU structure.
9641	* @param pMixedCtx Pointer to the guest-CPU context.
9642	* @param pVmxTransient Pointer to the VMX-transient structure.
9643	* @param uExitReason The VM-exit reason.
9644	*
9645	* @remarks The name of this function is displayed by dtrace, so keep it short
9646	* and to the point. No longer than 33 chars long, please.
9647	*/
9648	static VBOXSTRICTRC hmR0VmxHandleExitDtraceEvents(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx,
9649	PVMXTRANSIENT pVmxTransient, uint32_t uExitReason)
9650	{
9651	/*
9652	* Translate the event into a DBGF event (enmEvent + uEventArg) and at the
9653	* same time check whether any corresponding Dtrace event is enabled (fDtrace).
9654	*
9655	* Note! This is the reverse operation of what hmR0VmxPreRunGuestDebugStateUpdate
9656	* does. Must add/change/remove both places. Same ordering, please.
9657	*
9658	* Added/removed events must also be reflected in the next section
9659	* where we dispatch dtrace events.
9660	*/
9661	bool fDtrace1 = false;
9662	bool fDtrace2 = false;
9663	DBGFEVENTTYPE enmEvent1 = DBGFEVENT_END;
9664	DBGFEVENTTYPE enmEvent2 = DBGFEVENT_END;
9665	uint32_t uEventArg = 0;
9666	#define SET_EXIT(a_EventSubName) \
9667	do { \
9668	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
9669	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
9670	} while (0)
9671	#define SET_BOTH(a_EventSubName) \
9672	do { \
9673	enmEvent1 = RT_CONCAT(DBGFEVENT_INSTR_, a_EventSubName); \
9674	enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
9675	fDtrace1 = RT_CONCAT3(VBOXVMM_INSTR_, a_EventSubName, _ENABLED)(); \
9676	fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
9677	} while (0)
9678	switch (uExitReason)
9679	{
9680	case VMX_EXIT_MTF:
9681	return hmR0VmxExitMtf(pVCpu, pMixedCtx, pVmxTransient);
9682
9683	case VMX_EXIT_XCPT_OR_NMI:
9684	{
9685	uint8_t const idxVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo);
9686	switch (VMX_EXIT_INTERRUPTION_INFO_TYPE(pVmxTransient->uExitIntInfo))
9687	{
9688	case VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT:
9689	case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT:
9690	case VMX_EXIT_INTERRUPTION_INFO_TYPE_PRIV_SW_XCPT:
9691	if (idxVector <= (unsigned)(DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST))
9692	{
9693	if (VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uExitIntInfo))
9694	{
9695	hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
9696	uEventArg = pVmxTransient->uExitIntErrorCode;
9697	}
9698	enmEvent1 = (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + idxVector);
9699	switch (enmEvent1)
9700	{
9701	case DBGFEVENT_XCPT_DE: fDtrace1 = VBOXVMM_XCPT_DE_ENABLED(); break;
9702	case DBGFEVENT_XCPT_DB: fDtrace1 = VBOXVMM_XCPT_DB_ENABLED(); break;
9703	case DBGFEVENT_XCPT_BP: fDtrace1 = VBOXVMM_XCPT_BP_ENABLED(); break;
9704	case DBGFEVENT_XCPT_OF: fDtrace1 = VBOXVMM_XCPT_OF_ENABLED(); break;
9705	case DBGFEVENT_XCPT_BR: fDtrace1 = VBOXVMM_XCPT_BR_ENABLED(); break;
9706	case DBGFEVENT_XCPT_UD: fDtrace1 = VBOXVMM_XCPT_UD_ENABLED(); break;
9707	case DBGFEVENT_XCPT_NM: fDtrace1 = VBOXVMM_XCPT_NM_ENABLED(); break;
9708	case DBGFEVENT_XCPT_DF: fDtrace1 = VBOXVMM_XCPT_DF_ENABLED(); break;
9709	case DBGFEVENT_XCPT_TS: fDtrace1 = VBOXVMM_XCPT_TS_ENABLED(); break;
9710	case DBGFEVENT_XCPT_NP: fDtrace1 = VBOXVMM_XCPT_NP_ENABLED(); break;
9711	case DBGFEVENT_XCPT_SS: fDtrace1 = VBOXVMM_XCPT_SS_ENABLED(); break;
9712	case DBGFEVENT_XCPT_GP: fDtrace1 = VBOXVMM_XCPT_GP_ENABLED(); break;
9713	case DBGFEVENT_XCPT_PF: fDtrace1 = VBOXVMM_XCPT_PF_ENABLED(); break;
9714	case DBGFEVENT_XCPT_MF: fDtrace1 = VBOXVMM_XCPT_MF_ENABLED(); break;
9715	case DBGFEVENT_XCPT_AC: fDtrace1 = VBOXVMM_XCPT_AC_ENABLED(); break;
9716	case DBGFEVENT_XCPT_XF: fDtrace1 = VBOXVMM_XCPT_XF_ENABLED(); break;
9717	case DBGFEVENT_XCPT_VE: fDtrace1 = VBOXVMM_XCPT_VE_ENABLED(); break;
9718	case DBGFEVENT_XCPT_SX: fDtrace1 = VBOXVMM_XCPT_SX_ENABLED(); break;
9719	default: break;
9720	}
9721	}
9722	else
9723	AssertFailed();
9724	break;
9725
9726	case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT:
9727	uEventArg = idxVector;
9728	enmEvent1 = DBGFEVENT_INTERRUPT_SOFTWARE;
9729	fDtrace1 = VBOXVMM_INT_SOFTWARE_ENABLED();
9730	break;
9731	}
9732	break;
9733	}
9734
9735	case VMX_EXIT_TRIPLE_FAULT:
9736	enmEvent1 = DBGFEVENT_TRIPLE_FAULT;
9737	//fDtrace1 = VBOXVMM_EXIT_TRIPLE_FAULT_ENABLED();
9738	break;
9739	case VMX_EXIT_TASK_SWITCH: SET_EXIT(TASK_SWITCH); break;
9740	case VMX_EXIT_EPT_VIOLATION: SET_EXIT(VMX_EPT_VIOLATION); break;
9741	case VMX_EXIT_EPT_MISCONFIG: SET_EXIT(VMX_EPT_MISCONFIG); break;
9742	case VMX_EXIT_APIC_ACCESS: SET_EXIT(VMX_VAPIC_ACCESS); break;
9743	case VMX_EXIT_APIC_WRITE: SET_EXIT(VMX_VAPIC_WRITE); break;
9744
9745	/* Instruction specific VM-exits: */
9746	case VMX_EXIT_CPUID: SET_BOTH(CPUID); break;
9747	case VMX_EXIT_GETSEC: SET_BOTH(GETSEC); break;
9748	case VMX_EXIT_HLT: SET_BOTH(HALT); break;
9749	case VMX_EXIT_INVD: SET_BOTH(INVD); break;
9750	case VMX_EXIT_INVLPG: SET_BOTH(INVLPG); break;
9751	case VMX_EXIT_RDPMC: SET_BOTH(RDPMC); break;
9752	case VMX_EXIT_RDTSC: SET_BOTH(RDTSC); break;
9753	case VMX_EXIT_RSM: SET_BOTH(RSM); break;
9754	case VMX_EXIT_VMCALL: SET_BOTH(VMM_CALL); break;
9755	case VMX_EXIT_VMCLEAR: SET_BOTH(VMX_VMCLEAR); break;
9756	case VMX_EXIT_VMLAUNCH: SET_BOTH(VMX_VMLAUNCH); break;
9757	case VMX_EXIT_VMPTRLD: SET_BOTH(VMX_VMPTRLD); break;
9758	case VMX_EXIT_VMPTRST: SET_BOTH(VMX_VMPTRST); break;
9759	case VMX_EXIT_VMREAD: SET_BOTH(VMX_VMREAD); break;
9760	case VMX_EXIT_VMRESUME: SET_BOTH(VMX_VMRESUME); break;
9761	case VMX_EXIT_VMWRITE: SET_BOTH(VMX_VMWRITE); break;
9762	case VMX_EXIT_VMXOFF: SET_BOTH(VMX_VMXOFF); break;
9763	case VMX_EXIT_VMXON: SET_BOTH(VMX_VMXON); break;
9764	case VMX_EXIT_MOV_CRX:
9765	hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
9766	/** @todo r=bird: I feel these macros aren't very descriptive and needs to be at least 30 chars longer! ;-)
9767	* Sensible abbreviations strongly recommended here because even with 130 columns this stuff get too wide! */
9768	if ( VMX_EXIT_QUALIFICATION_CRX_ACCESS(pVmxTransient->uExitQualification)
9769	== VMX_EXIT_QUALIFICATION_CRX_ACCESS_READ)
9770	SET_BOTH(CRX_READ);
9771	else
9772	SET_BOTH(CRX_WRITE);
9773	uEventArg = VMX_EXIT_QUALIFICATION_CRX_REGISTER(pVmxTransient->uExitQualification);
9774	break;
9775	case VMX_EXIT_MOV_DRX:
9776	hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
9777	if ( VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification)
9778	== VMX_EXIT_QUALIFICATION_DRX_DIRECTION_READ)
9779	SET_BOTH(DRX_READ);
9780	else
9781	SET_BOTH(DRX_WRITE);
9782	uEventArg = VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification);
9783	break;
9784	case VMX_EXIT_RDMSR: SET_BOTH(RDMSR); break;
9785	case VMX_EXIT_WRMSR: SET_BOTH(WRMSR); break;
9786	case VMX_EXIT_MWAIT: SET_BOTH(MWAIT); break;
9787	case VMX_EXIT_MONITOR: SET_BOTH(MONITOR); break;
9788	case VMX_EXIT_PAUSE: SET_BOTH(PAUSE); break;
9789	case VMX_EXIT_XDTR_ACCESS:
9790	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
9791	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_XDTR_INSINFO_INSTR_ID))
9792	{
9793	case VMX_XDTR_INSINFO_II_SGDT: SET_BOTH(SGDT); break;
9794	case VMX_XDTR_INSINFO_II_SIDT: SET_BOTH(SIDT); break;
9795	case VMX_XDTR_INSINFO_II_LGDT: SET_BOTH(LGDT); break;
9796	case VMX_XDTR_INSINFO_II_LIDT: SET_BOTH(LIDT); break;
9797	}
9798	break;
9799
9800	case VMX_EXIT_TR_ACCESS:
9801	hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
9802	switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_YYTR_INSINFO_INSTR_ID))
9803	{
9804	case VMX_YYTR_INSINFO_II_SLDT: SET_BOTH(SLDT); break;
9805	case VMX_YYTR_INSINFO_II_STR: SET_BOTH(STR); break;
9806	case VMX_YYTR_INSINFO_II_LLDT: SET_BOTH(LLDT); break;
9807	case VMX_YYTR_INSINFO_II_LTR: SET_BOTH(LTR); break;
9808	}
9809	break;
9810
9811	case VMX_EXIT_INVEPT: SET_BOTH(VMX_INVEPT); break;
9812	case VMX_EXIT_RDTSCP: SET_BOTH(RDTSCP); break;
9813	case VMX_EXIT_INVVPID: SET_BOTH(VMX_INVVPID); break;
9814	case VMX_EXIT_WBINVD: SET_BOTH(WBINVD); break;
9815	case VMX_EXIT_XSETBV: SET_BOTH(XSETBV); break;
9816	case VMX_EXIT_RDRAND: SET_BOTH(RDRAND); break;
9817	case VMX_EXIT_INVPCID: SET_BOTH(VMX_INVPCID); break;
9818	case VMX_EXIT_VMFUNC: SET_BOTH(VMX_VMFUNC); break;
9819	case VMX_EXIT_RDSEED: SET_BOTH(RDSEED); break;
9820	case VMX_EXIT_XSAVES: SET_BOTH(XSAVES); break;
9821	case VMX_EXIT_XRSTORS: SET_BOTH(XRSTORS); break;
9822
9823	/* Events that aren't relevant at this point. */
9824	case VMX_EXIT_EXT_INT:
9825	case VMX_EXIT_INT_WINDOW:
9826	case VMX_EXIT_NMI_WINDOW:
9827	case VMX_EXIT_TPR_BELOW_THRESHOLD:
9828	case VMX_EXIT_PREEMPT_TIMER:
9829	case VMX_EXIT_IO_INSTR:
9830	break;
9831
9832	/* Errors and unexpected events. */
9833	case VMX_EXIT_INIT_SIGNAL:
9834	case VMX_EXIT_SIPI:
9835	case VMX_EXIT_IO_SMI:
9836	case VMX_EXIT_SMI:
9837	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
9838	case VMX_EXIT_ERR_MSR_LOAD:
9839	case VMX_EXIT_ERR_MACHINE_CHECK:
9840	break;
9841
9842	default:
9843	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
9844	break;
9845	}
9846	#undef SET_BOTH
9847	#undef SET_EXIT
9848
9849	/*
9850	* Dtrace tracepoints go first. We do them here at once so we don't
9851	* have to copy the guest state saving and stuff a few dozen times.
9852	* Down side is that we've got to repeat the switch, though this time
9853	* we use enmEvent since the probes are a subset of what DBGF does.
9854	*/
9855	if (fDtrace1 \|\| fDtrace2)
9856	{
9857	hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
9858	hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
9859	switch (enmEvent1)
9860	{
9861	/** @todo consider which extra parameters would be helpful for each probe. */
9862	case DBGFEVENT_END: break;
9863	case DBGFEVENT_XCPT_DE: VBOXVMM_XCPT_DE(pVCpu, pMixedCtx); break;
9864	case DBGFEVENT_XCPT_DB: VBOXVMM_XCPT_DB(pVCpu, pMixedCtx, pMixedCtx->dr[6]); break;
9865	case DBGFEVENT_XCPT_BP: VBOXVMM_XCPT_BP(pVCpu, pMixedCtx); break;
9866	case DBGFEVENT_XCPT_OF: VBOXVMM_XCPT_OF(pVCpu, pMixedCtx); break;
9867	case DBGFEVENT_XCPT_BR: VBOXVMM_XCPT_BR(pVCpu, pMixedCtx); break;
9868	case DBGFEVENT_XCPT_UD: VBOXVMM_XCPT_UD(pVCpu, pMixedCtx); break;
9869	case DBGFEVENT_XCPT_NM: VBOXVMM_XCPT_NM(pVCpu, pMixedCtx); break;
9870	case DBGFEVENT_XCPT_DF: VBOXVMM_XCPT_DF(pVCpu, pMixedCtx); break;
9871	case DBGFEVENT_XCPT_TS: VBOXVMM_XCPT_TS(pVCpu, pMixedCtx, uEventArg); break;
9872	case DBGFEVENT_XCPT_NP: VBOXVMM_XCPT_NP(pVCpu, pMixedCtx, uEventArg); break;
9873	case DBGFEVENT_XCPT_SS: VBOXVMM_XCPT_SS(pVCpu, pMixedCtx, uEventArg); break;
9874	case DBGFEVENT_XCPT_GP: VBOXVMM_XCPT_GP(pVCpu, pMixedCtx, uEventArg); break;
9875	case DBGFEVENT_XCPT_PF: VBOXVMM_XCPT_PF(pVCpu, pMixedCtx, uEventArg, pMixedCtx->cr2); break;
9876	case DBGFEVENT_XCPT_MF: VBOXVMM_XCPT_MF(pVCpu, pMixedCtx); break;
9877	case DBGFEVENT_XCPT_AC: VBOXVMM_XCPT_AC(pVCpu, pMixedCtx); break;
9878	case DBGFEVENT_XCPT_XF: VBOXVMM_XCPT_XF(pVCpu, pMixedCtx); break;
9879	case DBGFEVENT_XCPT_VE: VBOXVMM_XCPT_VE(pVCpu, pMixedCtx); break;
9880	case DBGFEVENT_XCPT_SX: VBOXVMM_XCPT_SX(pVCpu, pMixedCtx, uEventArg); break;
9881	case DBGFEVENT_INTERRUPT_SOFTWARE: VBOXVMM_INT_SOFTWARE(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9882	case DBGFEVENT_INSTR_CPUID: VBOXVMM_INSTR_CPUID(pVCpu, pMixedCtx, pMixedCtx->eax, pMixedCtx->ecx); break;
9883	case DBGFEVENT_INSTR_GETSEC: VBOXVMM_INSTR_GETSEC(pVCpu, pMixedCtx); break;
9884	case DBGFEVENT_INSTR_HALT: VBOXVMM_INSTR_HALT(pVCpu, pMixedCtx); break;
9885	case DBGFEVENT_INSTR_INVD: VBOXVMM_INSTR_INVD(pVCpu, pMixedCtx); break;
9886	case DBGFEVENT_INSTR_INVLPG: VBOXVMM_INSTR_INVLPG(pVCpu, pMixedCtx); break;
9887	case DBGFEVENT_INSTR_RDPMC: VBOXVMM_INSTR_RDPMC(pVCpu, pMixedCtx); break;
9888	case DBGFEVENT_INSTR_RDTSC: VBOXVMM_INSTR_RDTSC(pVCpu, pMixedCtx); break;
9889	case DBGFEVENT_INSTR_RSM: VBOXVMM_INSTR_RSM(pVCpu, pMixedCtx); break;
9890	case DBGFEVENT_INSTR_CRX_READ: VBOXVMM_INSTR_CRX_READ(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9891	case DBGFEVENT_INSTR_CRX_WRITE: VBOXVMM_INSTR_CRX_WRITE(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9892	case DBGFEVENT_INSTR_DRX_READ: VBOXVMM_INSTR_DRX_READ(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9893	case DBGFEVENT_INSTR_DRX_WRITE: VBOXVMM_INSTR_DRX_WRITE(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9894	case DBGFEVENT_INSTR_RDMSR: VBOXVMM_INSTR_RDMSR(pVCpu, pMixedCtx, pMixedCtx->ecx); break;
9895	case DBGFEVENT_INSTR_WRMSR: VBOXVMM_INSTR_WRMSR(pVCpu, pMixedCtx, pMixedCtx->ecx,
9896	RT_MAKE_U64(pMixedCtx->eax, pMixedCtx->edx)); break;
9897	case DBGFEVENT_INSTR_MWAIT: VBOXVMM_INSTR_MWAIT(pVCpu, pMixedCtx); break;
9898	case DBGFEVENT_INSTR_MONITOR: VBOXVMM_INSTR_MONITOR(pVCpu, pMixedCtx); break;
9899	case DBGFEVENT_INSTR_PAUSE: VBOXVMM_INSTR_PAUSE(pVCpu, pMixedCtx); break;
9900	case DBGFEVENT_INSTR_SGDT: VBOXVMM_INSTR_SGDT(pVCpu, pMixedCtx); break;
9901	case DBGFEVENT_INSTR_SIDT: VBOXVMM_INSTR_SIDT(pVCpu, pMixedCtx); break;
9902	case DBGFEVENT_INSTR_LGDT: VBOXVMM_INSTR_LGDT(pVCpu, pMixedCtx); break;
9903	case DBGFEVENT_INSTR_LIDT: VBOXVMM_INSTR_LIDT(pVCpu, pMixedCtx); break;
9904	case DBGFEVENT_INSTR_SLDT: VBOXVMM_INSTR_SLDT(pVCpu, pMixedCtx); break;
9905	case DBGFEVENT_INSTR_STR: VBOXVMM_INSTR_STR(pVCpu, pMixedCtx); break;
9906	case DBGFEVENT_INSTR_LLDT: VBOXVMM_INSTR_LLDT(pVCpu, pMixedCtx); break;
9907	case DBGFEVENT_INSTR_LTR: VBOXVMM_INSTR_LTR(pVCpu, pMixedCtx); break;
9908	case DBGFEVENT_INSTR_RDTSCP: VBOXVMM_INSTR_RDTSCP(pVCpu, pMixedCtx); break;
9909	case DBGFEVENT_INSTR_WBINVD: VBOXVMM_INSTR_WBINVD(pVCpu, pMixedCtx); break;
9910	case DBGFEVENT_INSTR_XSETBV: VBOXVMM_INSTR_XSETBV(pVCpu, pMixedCtx); break;
9911	case DBGFEVENT_INSTR_RDRAND: VBOXVMM_INSTR_RDRAND(pVCpu, pMixedCtx); break;
9912	case DBGFEVENT_INSTR_RDSEED: VBOXVMM_INSTR_RDSEED(pVCpu, pMixedCtx); break;
9913	case DBGFEVENT_INSTR_XSAVES: VBOXVMM_INSTR_XSAVES(pVCpu, pMixedCtx); break;
9914	case DBGFEVENT_INSTR_XRSTORS: VBOXVMM_INSTR_XRSTORS(pVCpu, pMixedCtx); break;
9915	case DBGFEVENT_INSTR_VMM_CALL: VBOXVMM_INSTR_VMM_CALL(pVCpu, pMixedCtx); break;
9916	case DBGFEVENT_INSTR_VMX_VMCLEAR: VBOXVMM_INSTR_VMX_VMCLEAR(pVCpu, pMixedCtx); break;
9917	case DBGFEVENT_INSTR_VMX_VMLAUNCH: VBOXVMM_INSTR_VMX_VMLAUNCH(pVCpu, pMixedCtx); break;
9918	case DBGFEVENT_INSTR_VMX_VMPTRLD: VBOXVMM_INSTR_VMX_VMPTRLD(pVCpu, pMixedCtx); break;
9919	case DBGFEVENT_INSTR_VMX_VMPTRST: VBOXVMM_INSTR_VMX_VMPTRST(pVCpu, pMixedCtx); break;
9920	case DBGFEVENT_INSTR_VMX_VMREAD: VBOXVMM_INSTR_VMX_VMREAD(pVCpu, pMixedCtx); break;
9921	case DBGFEVENT_INSTR_VMX_VMRESUME: VBOXVMM_INSTR_VMX_VMRESUME(pVCpu, pMixedCtx); break;
9922	case DBGFEVENT_INSTR_VMX_VMWRITE: VBOXVMM_INSTR_VMX_VMWRITE(pVCpu, pMixedCtx); break;
9923	case DBGFEVENT_INSTR_VMX_VMXOFF: VBOXVMM_INSTR_VMX_VMXOFF(pVCpu, pMixedCtx); break;
9924	case DBGFEVENT_INSTR_VMX_VMXON: VBOXVMM_INSTR_VMX_VMXON(pVCpu, pMixedCtx); break;
9925	case DBGFEVENT_INSTR_VMX_INVEPT: VBOXVMM_INSTR_VMX_INVEPT(pVCpu, pMixedCtx); break;
9926	case DBGFEVENT_INSTR_VMX_INVVPID: VBOXVMM_INSTR_VMX_INVVPID(pVCpu, pMixedCtx); break;
9927	case DBGFEVENT_INSTR_VMX_INVPCID: VBOXVMM_INSTR_VMX_INVPCID(pVCpu, pMixedCtx); break;
9928	case DBGFEVENT_INSTR_VMX_VMFUNC: VBOXVMM_INSTR_VMX_VMFUNC(pVCpu, pMixedCtx); break;
9929	default: AssertMsgFailed(("enmEvent1=%d uExitReason=%d\n", enmEvent1, uExitReason)); break;
9930	}
9931	switch (enmEvent2)
9932	{
9933	/** @todo consider which extra parameters would be helpful for each probe. */
9934	case DBGFEVENT_END: break;
9935	case DBGFEVENT_EXIT_TASK_SWITCH: VBOXVMM_EXIT_TASK_SWITCH(pVCpu, pMixedCtx); break;
9936	case DBGFEVENT_EXIT_CPUID: VBOXVMM_EXIT_CPUID(pVCpu, pMixedCtx, pMixedCtx->eax, pMixedCtx->ecx); break;
9937	case DBGFEVENT_EXIT_GETSEC: VBOXVMM_EXIT_GETSEC(pVCpu, pMixedCtx); break;
9938	case DBGFEVENT_EXIT_HALT: VBOXVMM_EXIT_HALT(pVCpu, pMixedCtx); break;
9939	case DBGFEVENT_EXIT_INVD: VBOXVMM_EXIT_INVD(pVCpu, pMixedCtx); break;
9940	case DBGFEVENT_EXIT_INVLPG: VBOXVMM_EXIT_INVLPG(pVCpu, pMixedCtx); break;
9941	case DBGFEVENT_EXIT_RDPMC: VBOXVMM_EXIT_RDPMC(pVCpu, pMixedCtx); break;
9942	case DBGFEVENT_EXIT_RDTSC: VBOXVMM_EXIT_RDTSC(pVCpu, pMixedCtx); break;
9943	case DBGFEVENT_EXIT_RSM: VBOXVMM_EXIT_RSM(pVCpu, pMixedCtx); break;
9944	case DBGFEVENT_EXIT_CRX_READ: VBOXVMM_EXIT_CRX_READ(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9945	case DBGFEVENT_EXIT_CRX_WRITE: VBOXVMM_EXIT_CRX_WRITE(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9946	case DBGFEVENT_EXIT_DRX_READ: VBOXVMM_EXIT_DRX_READ(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9947	case DBGFEVENT_EXIT_DRX_WRITE: VBOXVMM_EXIT_DRX_WRITE(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9948	case DBGFEVENT_EXIT_RDMSR: VBOXVMM_EXIT_RDMSR(pVCpu, pMixedCtx, pMixedCtx->ecx); break;
9949	case DBGFEVENT_EXIT_WRMSR: VBOXVMM_EXIT_WRMSR(pVCpu, pMixedCtx, pMixedCtx->ecx,
9950	RT_MAKE_U64(pMixedCtx->eax, pMixedCtx->edx)); break;
9951	case DBGFEVENT_EXIT_MWAIT: VBOXVMM_EXIT_MWAIT(pVCpu, pMixedCtx); break;
9952	case DBGFEVENT_EXIT_MONITOR: VBOXVMM_EXIT_MONITOR(pVCpu, pMixedCtx); break;
9953	case DBGFEVENT_EXIT_PAUSE: VBOXVMM_EXIT_PAUSE(pVCpu, pMixedCtx); break;
9954	case DBGFEVENT_EXIT_SGDT: VBOXVMM_EXIT_SGDT(pVCpu, pMixedCtx); break;
9955	case DBGFEVENT_EXIT_SIDT: VBOXVMM_EXIT_SIDT(pVCpu, pMixedCtx); break;
9956	case DBGFEVENT_EXIT_LGDT: VBOXVMM_EXIT_LGDT(pVCpu, pMixedCtx); break;
9957	case DBGFEVENT_EXIT_LIDT: VBOXVMM_EXIT_LIDT(pVCpu, pMixedCtx); break;
9958	case DBGFEVENT_EXIT_SLDT: VBOXVMM_EXIT_SLDT(pVCpu, pMixedCtx); break;
9959	case DBGFEVENT_EXIT_STR: VBOXVMM_EXIT_STR(pVCpu, pMixedCtx); break;
9960	case DBGFEVENT_EXIT_LLDT: VBOXVMM_EXIT_LLDT(pVCpu, pMixedCtx); break;
9961	case DBGFEVENT_EXIT_LTR: VBOXVMM_EXIT_LTR(pVCpu, pMixedCtx); break;
9962	case DBGFEVENT_EXIT_RDTSCP: VBOXVMM_EXIT_RDTSCP(pVCpu, pMixedCtx); break;
9963	case DBGFEVENT_EXIT_WBINVD: VBOXVMM_EXIT_WBINVD(pVCpu, pMixedCtx); break;
9964	case DBGFEVENT_EXIT_XSETBV: VBOXVMM_EXIT_XSETBV(pVCpu, pMixedCtx); break;
9965	case DBGFEVENT_EXIT_RDRAND: VBOXVMM_EXIT_RDRAND(pVCpu, pMixedCtx); break;
9966	case DBGFEVENT_EXIT_RDSEED: VBOXVMM_EXIT_RDSEED(pVCpu, pMixedCtx); break;
9967	case DBGFEVENT_EXIT_XSAVES: VBOXVMM_EXIT_XSAVES(pVCpu, pMixedCtx); break;
9968	case DBGFEVENT_EXIT_XRSTORS: VBOXVMM_EXIT_XRSTORS(pVCpu, pMixedCtx); break;
9969	case DBGFEVENT_EXIT_VMM_CALL: VBOXVMM_EXIT_VMM_CALL(pVCpu, pMixedCtx); break;
9970	case DBGFEVENT_EXIT_VMX_VMCLEAR: VBOXVMM_EXIT_VMX_VMCLEAR(pVCpu, pMixedCtx); break;
9971	case DBGFEVENT_EXIT_VMX_VMLAUNCH: VBOXVMM_EXIT_VMX_VMLAUNCH(pVCpu, pMixedCtx); break;
9972	case DBGFEVENT_EXIT_VMX_VMPTRLD: VBOXVMM_EXIT_VMX_VMPTRLD(pVCpu, pMixedCtx); break;
9973	case DBGFEVENT_EXIT_VMX_VMPTRST: VBOXVMM_EXIT_VMX_VMPTRST(pVCpu, pMixedCtx); break;
9974	case DBGFEVENT_EXIT_VMX_VMREAD: VBOXVMM_EXIT_VMX_VMREAD(pVCpu, pMixedCtx); break;
9975	case DBGFEVENT_EXIT_VMX_VMRESUME: VBOXVMM_EXIT_VMX_VMRESUME(pVCpu, pMixedCtx); break;
9976	case DBGFEVENT_EXIT_VMX_VMWRITE: VBOXVMM_EXIT_VMX_VMWRITE(pVCpu, pMixedCtx); break;
9977	case DBGFEVENT_EXIT_VMX_VMXOFF: VBOXVMM_EXIT_VMX_VMXOFF(pVCpu, pMixedCtx); break;
9978	case DBGFEVENT_EXIT_VMX_VMXON: VBOXVMM_EXIT_VMX_VMXON(pVCpu, pMixedCtx); break;
9979	case DBGFEVENT_EXIT_VMX_INVEPT: VBOXVMM_EXIT_VMX_INVEPT(pVCpu, pMixedCtx); break;
9980	case DBGFEVENT_EXIT_VMX_INVVPID: VBOXVMM_EXIT_VMX_INVVPID(pVCpu, pMixedCtx); break;
9981	case DBGFEVENT_EXIT_VMX_INVPCID: VBOXVMM_EXIT_VMX_INVPCID(pVCpu, pMixedCtx); break;
9982	case DBGFEVENT_EXIT_VMX_VMFUNC: VBOXVMM_EXIT_VMX_VMFUNC(pVCpu, pMixedCtx); break;
9983	case DBGFEVENT_EXIT_VMX_EPT_MISCONFIG: VBOXVMM_EXIT_VMX_EPT_MISCONFIG(pVCpu, pMixedCtx); break;
9984	case DBGFEVENT_EXIT_VMX_EPT_VIOLATION: VBOXVMM_EXIT_VMX_EPT_VIOLATION(pVCpu, pMixedCtx); break;
9985	case DBGFEVENT_EXIT_VMX_VAPIC_ACCESS: VBOXVMM_EXIT_VMX_VAPIC_ACCESS(pVCpu, pMixedCtx); break;
9986	case DBGFEVENT_EXIT_VMX_VAPIC_WRITE: VBOXVMM_EXIT_VMX_VAPIC_WRITE(pVCpu, pMixedCtx); break;
9987	default: AssertMsgFailed(("enmEvent2=%d uExitReason=%d\n", enmEvent2, uExitReason)); break;
9988	}
9989	}
9990
9991	/*
9992	* Fire of the DBGF event, if enabled (our check here is just a quick one,
9993	* the DBGF call will do a full check).
9994	*
9995	* Note! DBGF sets DBGFEVENT_INTERRUPT_SOFTWARE in the bitmap.
9996	* Note! If we have to events, we prioritize the first, i.e. the instruction
9997	* one, in order to avoid event nesting.
9998	*/
9999	if ( enmEvent1 != DBGFEVENT_END
10000	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent1))
10001	{
10002	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArg(pVM, pVCpu, enmEvent1, uEventArg, DBGFEVENTCTX_HM);
10003	if (rcStrict != VINF_SUCCESS)
10004	return rcStrict;
10005	}
10006	else if ( enmEvent2 != DBGFEVENT_END
10007	&& DBGF_IS_EVENT_ENABLED(pVM, enmEvent2))
10008	{
10009	VBOXSTRICTRC rcStrict = DBGFEventGenericWithArg(pVM, pVCpu, enmEvent2, uEventArg, DBGFEVENTCTX_HM);
10010	if (rcStrict != VINF_SUCCESS)
10011	return rcStrict;
10012	}
10013
10014	return VINF_SUCCESS;
10015	}
10016
10017
10018	/**
10019	* Single-stepping VM-exit filtering.
10020	*
10021	* This is preprocessing the VM-exits and deciding whether we've gotten far
10022	* enough to return VINF_EM_DBG_STEPPED already. If not, normal VM-exit
10023	* handling is performed.
10024	*
10025	* @returns Strict VBox status code (i.e. informational status codes too).
10026	* @param pVM The cross context VM structure.
10027	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
10028	* @param pMixedCtx Pointer to the guest-CPU context. The data may be
10029	* out-of-sync. Make sure to update the required
10030	* fields before using them.
10031	* @param pVmxTransient Pointer to the VMX-transient structure.
10032	* @param uExitReason The VM-exit reason.
10033	* @param pDbgState The debug state.
10034	*/
10035	DECLINLINE(VBOXSTRICTRC) hmR0VmxRunDebugHandleExit(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient,
10036	uint32_t uExitReason, PVMXRUNDBGSTATE pDbgState)
10037	{
10038	/*
10039	* Expensive (saves context) generic dtrace VM-exit probe.
10040	*/
10041	if (!VBOXVMM_R0_HMVMX_VMEXIT_ENABLED())
10042	{ /* more likely */ }
10043	else
10044	{
10045	hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
10046	hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
10047	VBOXVMM_R0_HMVMX_VMEXIT(pVCpu, pMixedCtx, pVmxTransient->uExitReason, pVmxTransient->uExitQualification);
10048	}
10049
10050	/*
10051	* Check for host NMI, just to get that out of the way.
10052	*/
10053	if (uExitReason != VMX_EXIT_XCPT_OR_NMI)
10054	{ /* normally likely */ }
10055	else
10056	{
10057	int rc2 = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
10058	AssertRCReturn(rc2, rc2);
10059	uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVmxTransient->uExitIntInfo);
10060	if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
10061	return hmR0VmxExitXcptOrNmi(pVCpu, pMixedCtx, pVmxTransient);
10062	}
10063
10064	/*
10065	* Check for single stepping event if we're stepping.
10066	*/
10067	if (pVCpu->hm.s.fSingleInstruction)
10068	{
10069	switch (uExitReason)
10070	{
10071	case VMX_EXIT_MTF:
10072	return hmR0VmxExitMtf(pVCpu, pMixedCtx, pVmxTransient);
10073
10074	/* Various events: */
10075	case VMX_EXIT_XCPT_OR_NMI:
10076	case VMX_EXIT_EXT_INT:
10077	case VMX_EXIT_TRIPLE_FAULT:
10078	case VMX_EXIT_INT_WINDOW:
10079	case VMX_EXIT_NMI_WINDOW:
10080	case VMX_EXIT_TASK_SWITCH:
10081	case VMX_EXIT_TPR_BELOW_THRESHOLD:
10082	case VMX_EXIT_APIC_ACCESS:
10083	case VMX_EXIT_EPT_VIOLATION:
10084	case VMX_EXIT_EPT_MISCONFIG:
10085	case VMX_EXIT_PREEMPT_TIMER:
10086
10087	/* Instruction specific VM-exits: */
10088	case VMX_EXIT_CPUID:
10089	case VMX_EXIT_GETSEC:
10090	case VMX_EXIT_HLT:
10091	case VMX_EXIT_INVD:
10092	case VMX_EXIT_INVLPG:
10093	case VMX_EXIT_RDPMC:
10094	case VMX_EXIT_RDTSC:
10095	case VMX_EXIT_RSM:
10096	case VMX_EXIT_VMCALL:
10097	case VMX_EXIT_VMCLEAR:
10098	case VMX_EXIT_VMLAUNCH:
10099	case VMX_EXIT_VMPTRLD:
10100	case VMX_EXIT_VMPTRST:
10101	case VMX_EXIT_VMREAD:
10102	case VMX_EXIT_VMRESUME:
10103	case VMX_EXIT_VMWRITE:
10104	case VMX_EXIT_VMXOFF:
10105	case VMX_EXIT_VMXON:
10106	case VMX_EXIT_MOV_CRX:
10107	case VMX_EXIT_MOV_DRX:
10108	case VMX_EXIT_IO_INSTR:
10109	case VMX_EXIT_RDMSR:
10110	case VMX_EXIT_WRMSR:
10111	case VMX_EXIT_MWAIT:
10112	case VMX_EXIT_MONITOR:
10113	case VMX_EXIT_PAUSE:
10114	case VMX_EXIT_XDTR_ACCESS:
10115	case VMX_EXIT_TR_ACCESS:
10116	case VMX_EXIT_INVEPT:
10117	case VMX_EXIT_RDTSCP:
10118	case VMX_EXIT_INVVPID:
10119	case VMX_EXIT_WBINVD:
10120	case VMX_EXIT_XSETBV:
10121	case VMX_EXIT_RDRAND:
10122	case VMX_EXIT_INVPCID:
10123	case VMX_EXIT_VMFUNC:
10124	case VMX_EXIT_RDSEED:
10125	case VMX_EXIT_XSAVES:
10126	case VMX_EXIT_XRSTORS:
10127	{
10128	int rc2 = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
10129	rc2 \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10130	AssertRCReturn(rc2, rc2);
10131	if ( pMixedCtx->rip != pDbgState->uRipStart
10132	\|\| pMixedCtx->cs.Sel != pDbgState->uCsStart)
10133	return VINF_EM_DBG_STEPPED;
10134	break;
10135	}
10136
10137	/* Errors and unexpected events: */
10138	case VMX_EXIT_INIT_SIGNAL:
10139	case VMX_EXIT_SIPI:
10140	case VMX_EXIT_IO_SMI:
10141	case VMX_EXIT_SMI:
10142	case VMX_EXIT_ERR_INVALID_GUEST_STATE:
10143	case VMX_EXIT_ERR_MSR_LOAD:
10144	case VMX_EXIT_ERR_MACHINE_CHECK:
10145	case VMX_EXIT_APIC_WRITE: /* Some talk about this being fault like, so I guess we must process it? */
10146	break;
10147
10148	default:
10149	AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
10150	break;
10151	}
10152	}
10153
10154	/*
10155	* Check for debugger event breakpoints and dtrace probes.
10156	*/
10157	if ( uExitReason < RT_ELEMENTS(pDbgState->bmExitsToCheck) * 32U
10158	&& ASMBitTest(pDbgState->bmExitsToCheck, uExitReason) )
10159	{
10160	VBOXSTRICTRC rcStrict = hmR0VmxHandleExitDtraceEvents(pVM, pVCpu, pMixedCtx, pVmxTransient, uExitReason);
10161	if (rcStrict != VINF_SUCCESS)
10162	return rcStrict;
10163	}
10164
10165	/*
10166	* Normal processing.
10167	*/
10168	#ifdef HMVMX_USE_FUNCTION_TABLE
10169	return g_apfnVMExitHandlers[uExitReason](pVCpu, pMixedCtx, pVmxTransient);
10170	#else
10171	return hmR0VmxHandleExit(pVCpu, pMixedCtx, pVmxTransient, uExitReason);
10172	#endif
10173	}
10174
10175
10176	/**
10177	* Single steps guest code using VT-x.
10178	*
10179	* @returns Strict VBox status code (i.e. informational status codes too).
10180	* @param pVM The cross context VM structure.
10181	* @param pVCpu The cross context virtual CPU structure.
10182	* @param pCtx Pointer to the guest-CPU context.
10183	*
10184	* @note Mostly the same as hmR0VmxRunGuestCodeNormal().
10185	*/
10186	static VBOXSTRICTRC hmR0VmxRunGuestCodeDebug(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
10187	{
10188	VMXTRANSIENT VmxTransient;
10189	VmxTransient.fUpdateTscOffsettingAndPreemptTimer = true;
10190
10191	/* Set HMCPU indicators. */
10192	bool const fSavedSingleInstruction = pVCpu->hm.s.fSingleInstruction;
10193	pVCpu->hm.s.fSingleInstruction = pVCpu->hm.s.fSingleInstruction \|\| DBGFIsStepping(pVCpu);
10194	pVCpu->hm.s.fDebugWantRdTscExit = false;
10195	pVCpu->hm.s.fUsingDebugLoop = true;
10196
10197	/* State we keep to help modify and later restore the VMCS fields we alter, and for detecting steps. */
10198	VMXRUNDBGSTATE DbgState;
10199	hmR0VmxRunDebugStateInit(pVCpu, pCtx, &DbgState);
10200	hmR0VmxPreRunGuestDebugStateUpdate(pVM, pVCpu, pCtx, &DbgState, &VmxTransient);
10201
10202	/*
10203	* The loop.
10204	*/
10205	VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
10206	for (uint32_t cLoops = 0; ; cLoops++)
10207	{
10208	Assert(!HMR0SuspendPending());
10209	HMVMX_ASSERT_CPU_SAFE();
10210	bool fStepping = pVCpu->hm.s.fSingleInstruction;
10211
10212	/*
10213	* Preparatory work for running guest code, this may force us to return
10214	* to ring-3. This bugger disables interrupts on VINF_SUCCESS!
10215	*/
10216	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
10217	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &DbgState); /* Set up execute controls the next to can respond to. */
10218	rcStrict = hmR0VmxPreRunGuest(pVM, pVCpu, pCtx, &VmxTransient, fStepping);
10219	if (rcStrict != VINF_SUCCESS)
10220	break;
10221
10222	hmR0VmxPreRunGuestCommitted(pVM, pVCpu, pCtx, &VmxTransient);
10223	hmR0VmxPreRunGuestDebugStateApply(pVCpu, &DbgState); /* Override any obnoxious code in the above two calls. */
10224
10225	/*
10226	* Now we can run the guest code.
10227	*/
10228	int rcRun = hmR0VmxRunGuest(pVM, pVCpu, pCtx);
10229
10230	/* The guest-CPU context is now outdated, 'pCtx' is to be treated as 'pMixedCtx' from this point on!!! */
10231
10232	/*
10233	* Restore any residual host-state and save any bits shared between host
10234	* and guest into the guest-CPU state. Re-enables interrupts!
10235	*/
10236	hmR0VmxPostRunGuest(pVM, pVCpu, pCtx, &VmxTransient, rcRun);
10237
10238	/* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
10239	if (RT_SUCCESS(rcRun))
10240	{ /* very likely */ }
10241	else
10242	{
10243	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x);
10244	hmR0VmxReportWorldSwitchError(pVM, pVCpu, rcRun, pCtx, &VmxTransient);
10245	return rcRun;
10246	}
10247
10248	/* Profile the VM-exit. */
10249	AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
10250	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
10251	STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
10252	STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
10253	HMVMX_START_EXIT_DISPATCH_PROF();
10254
10255	VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, pCtx, VmxTransient.uExitReason);
10256
10257	/*
10258	* Handle the VM-exit - we quit earlier on certain VM-exits, see hmR0VmxHandleExitDebug().
10259	*/
10260	rcStrict = hmR0VmxRunDebugHandleExit(pVM, pVCpu, pCtx, &VmxTransient, VmxTransient.uExitReason, &DbgState);
10261	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
10262	if (rcStrict != VINF_SUCCESS)
10263	break;
10264	if (cLoops > pVM->hm.s.cMaxResumeLoops)
10265	{
10266	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
10267	rcStrict = VINF_EM_RAW_INTERRUPT;
10268	break;
10269	}
10270
10271	/*
10272	* Stepping: Did the RIP change, if so, consider it a single step.
10273	* Otherwise, make sure one of the TFs gets set.
10274	*/
10275	if (fStepping)
10276	{
10277	int rc2 = hmR0VmxSaveGuestRip(pVCpu, pCtx);
10278	rc2 \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pCtx);
10279	AssertRCReturn(rc2, rc2);
10280	if ( pCtx->rip != DbgState.uRipStart
10281	\|\| pCtx->cs.Sel != DbgState.uCsStart)
10282	{
10283	rcStrict = VINF_EM_DBG_STEPPED;
10284	break;
10285	}
10286	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
10287	}
10288
10289	/*
10290	* Update when dtrace settings changes (DBGF kicks us, so no need to check).
10291	*/
10292	if (VBOXVMM_GET_SETTINGS_SEQ_NO() != DbgState.uDtraceSettingsSeqNo)
10293	hmR0VmxPreRunGuestDebugStateUpdate(pVM, pVCpu, pCtx, &DbgState, &VmxTransient);
10294	}
10295
10296	/*
10297	* Clear the X86_EFL_TF if necessary.
10298	*/
10299	if (pVCpu->hm.s.fClearTrapFlag)
10300	{
10301	int rc2 = hmR0VmxSaveGuestRflags(pVCpu, pCtx);
10302	AssertRCReturn(rc2, rc2);
10303	pVCpu->hm.s.fClearTrapFlag = false;
10304	pCtx->eflags.Bits.u1TF = 0;
10305	}
10306	/** @todo there seems to be issues with the resume flag when the monitor trap
10307	* flag is pending without being used. Seen early in bios init when
10308	* accessing APIC page in protected mode. */
10309
10310	/*
10311	* Restore VM-exit control settings as we may not reenter this function the
10312	* next time around.
10313	*/
10314	rcStrict = hmR0VmxRunDebugStateRevert(pVCpu, &DbgState, rcStrict);
10315
10316	/* Restore HMCPU indicators. */
10317	pVCpu->hm.s.fUsingDebugLoop = false;
10318	pVCpu->hm.s.fDebugWantRdTscExit = false;
10319	pVCpu->hm.s.fSingleInstruction = fSavedSingleInstruction;
10320
10321	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
10322	return rcStrict;
10323	}
10324
10325
10326	/** @} */
10327
10328
10329	/**
10330	* Checks if any expensive dtrace probes are enabled and we should go to the
10331	* debug loop.
10332	*
10333	* @returns true if we should use debug loop, false if not.
10334	*/
10335	static bool hmR0VmxAnyExpensiveProbesEnabled(void)
10336	{
10337	/* It's probably faster to OR the raw 32-bit counter variables together.
10338	Since the variables are in an array and the probes are next to one
10339	another (more or less), we have good locality. So, better read
10340	eight-nine cache lines ever time and only have one conditional, than
10341	128+ conditionals, right? */
10342	return ( VBOXVMM_R0_HMVMX_VMEXIT_ENABLED_RAW() /* expensive too due to context */
10343	\| VBOXVMM_XCPT_DE_ENABLED_RAW()
10344	\| VBOXVMM_XCPT_DB_ENABLED_RAW()
10345	\| VBOXVMM_XCPT_BP_ENABLED_RAW()
10346	\| VBOXVMM_XCPT_OF_ENABLED_RAW()
10347	\| VBOXVMM_XCPT_BR_ENABLED_RAW()
10348	\| VBOXVMM_XCPT_UD_ENABLED_RAW()
10349	\| VBOXVMM_XCPT_NM_ENABLED_RAW()
10350	\| VBOXVMM_XCPT_DF_ENABLED_RAW()
10351	\| VBOXVMM_XCPT_TS_ENABLED_RAW()
10352	\| VBOXVMM_XCPT_NP_ENABLED_RAW()
10353	\| VBOXVMM_XCPT_SS_ENABLED_RAW()
10354	\| VBOXVMM_XCPT_GP_ENABLED_RAW()
10355	\| VBOXVMM_XCPT_PF_ENABLED_RAW()
10356	\| VBOXVMM_XCPT_MF_ENABLED_RAW()
10357	\| VBOXVMM_XCPT_AC_ENABLED_RAW()
10358	\| VBOXVMM_XCPT_XF_ENABLED_RAW()
10359	\| VBOXVMM_XCPT_VE_ENABLED_RAW()
10360	\| VBOXVMM_XCPT_SX_ENABLED_RAW()
10361	\| VBOXVMM_INT_SOFTWARE_ENABLED_RAW()
10362	\| VBOXVMM_INT_HARDWARE_ENABLED_RAW()
10363	) != 0
10364	\|\| ( VBOXVMM_INSTR_HALT_ENABLED_RAW()
10365	\| VBOXVMM_INSTR_MWAIT_ENABLED_RAW()
10366	\| VBOXVMM_INSTR_MONITOR_ENABLED_RAW()
10367	\| VBOXVMM_INSTR_CPUID_ENABLED_RAW()
10368	\| VBOXVMM_INSTR_INVD_ENABLED_RAW()
10369	\| VBOXVMM_INSTR_WBINVD_ENABLED_RAW()
10370	\| VBOXVMM_INSTR_INVLPG_ENABLED_RAW()
10371	\| VBOXVMM_INSTR_RDTSC_ENABLED_RAW()
10372	\| VBOXVMM_INSTR_RDTSCP_ENABLED_RAW()
10373	\| VBOXVMM_INSTR_RDPMC_ENABLED_RAW()
10374	\| VBOXVMM_INSTR_RDMSR_ENABLED_RAW()
10375	\| VBOXVMM_INSTR_WRMSR_ENABLED_RAW()
10376	\| VBOXVMM_INSTR_CRX_READ_ENABLED_RAW()
10377	\| VBOXVMM_INSTR_CRX_WRITE_ENABLED_RAW()
10378	\| VBOXVMM_INSTR_DRX_READ_ENABLED_RAW()
10379	\| VBOXVMM_INSTR_DRX_WRITE_ENABLED_RAW()
10380	\| VBOXVMM_INSTR_PAUSE_ENABLED_RAW()
10381	\| VBOXVMM_INSTR_XSETBV_ENABLED_RAW()
10382	\| VBOXVMM_INSTR_SIDT_ENABLED_RAW()
10383	\| VBOXVMM_INSTR_LIDT_ENABLED_RAW()
10384	\| VBOXVMM_INSTR_SGDT_ENABLED_RAW()
10385	\| VBOXVMM_INSTR_LGDT_ENABLED_RAW()
10386	\| VBOXVMM_INSTR_SLDT_ENABLED_RAW()
10387	\| VBOXVMM_INSTR_LLDT_ENABLED_RAW()
10388	\| VBOXVMM_INSTR_STR_ENABLED_RAW()
10389	\| VBOXVMM_INSTR_LTR_ENABLED_RAW()
10390	\| VBOXVMM_INSTR_GETSEC_ENABLED_RAW()
10391	\| VBOXVMM_INSTR_RSM_ENABLED_RAW()
10392	\| VBOXVMM_INSTR_RDRAND_ENABLED_RAW()
10393	\| VBOXVMM_INSTR_RDSEED_ENABLED_RAW()
10394	\| VBOXVMM_INSTR_XSAVES_ENABLED_RAW()
10395	\| VBOXVMM_INSTR_XRSTORS_ENABLED_RAW()
10396	\| VBOXVMM_INSTR_VMM_CALL_ENABLED_RAW()
10397	\| VBOXVMM_INSTR_VMX_VMCLEAR_ENABLED_RAW()
10398	\| VBOXVMM_INSTR_VMX_VMLAUNCH_ENABLED_RAW()
10399	\| VBOXVMM_INSTR_VMX_VMPTRLD_ENABLED_RAW()
10400	\| VBOXVMM_INSTR_VMX_VMPTRST_ENABLED_RAW()
10401	\| VBOXVMM_INSTR_VMX_VMREAD_ENABLED_RAW()
10402	\| VBOXVMM_INSTR_VMX_VMRESUME_ENABLED_RAW()
10403	\| VBOXVMM_INSTR_VMX_VMWRITE_ENABLED_RAW()
10404	\| VBOXVMM_INSTR_VMX_VMXOFF_ENABLED_RAW()
10405	\| VBOXVMM_INSTR_VMX_VMXON_ENABLED_RAW()
10406	\| VBOXVMM_INSTR_VMX_VMFUNC_ENABLED_RAW()
10407	\| VBOXVMM_INSTR_VMX_INVEPT_ENABLED_RAW()
10408	\| VBOXVMM_INSTR_VMX_INVVPID_ENABLED_RAW()
10409	\| VBOXVMM_INSTR_VMX_INVPCID_ENABLED_RAW()
10410	) != 0
10411	\|\| ( VBOXVMM_EXIT_TASK_SWITCH_ENABLED_RAW()
10412	\| VBOXVMM_EXIT_HALT_ENABLED_RAW()
10413	\| VBOXVMM_EXIT_MWAIT_ENABLED_RAW()
10414	\| VBOXVMM_EXIT_MONITOR_ENABLED_RAW()
10415	\| VBOXVMM_EXIT_CPUID_ENABLED_RAW()
10416	\| VBOXVMM_EXIT_INVD_ENABLED_RAW()
10417	\| VBOXVMM_EXIT_WBINVD_ENABLED_RAW()
10418	\| VBOXVMM_EXIT_INVLPG_ENABLED_RAW()
10419	\| VBOXVMM_EXIT_RDTSC_ENABLED_RAW()
10420	\| VBOXVMM_EXIT_RDTSCP_ENABLED_RAW()
10421	\| VBOXVMM_EXIT_RDPMC_ENABLED_RAW()
10422	\| VBOXVMM_EXIT_RDMSR_ENABLED_RAW()
10423	\| VBOXVMM_EXIT_WRMSR_ENABLED_RAW()
10424	\| VBOXVMM_EXIT_CRX_READ_ENABLED_RAW()
10425	\| VBOXVMM_EXIT_CRX_WRITE_ENABLED_RAW()
10426	\| VBOXVMM_EXIT_DRX_READ_ENABLED_RAW()
10427	\| VBOXVMM_EXIT_DRX_WRITE_ENABLED_RAW()
10428	\| VBOXVMM_EXIT_PAUSE_ENABLED_RAW()
10429	\| VBOXVMM_EXIT_XSETBV_ENABLED_RAW()
10430	\| VBOXVMM_EXIT_SIDT_ENABLED_RAW()
10431	\| VBOXVMM_EXIT_LIDT_ENABLED_RAW()
10432	\| VBOXVMM_EXIT_SGDT_ENABLED_RAW()
10433	\| VBOXVMM_EXIT_LGDT_ENABLED_RAW()
10434	\| VBOXVMM_EXIT_SLDT_ENABLED_RAW()
10435	\| VBOXVMM_EXIT_LLDT_ENABLED_RAW()
10436	\| VBOXVMM_EXIT_STR_ENABLED_RAW()
10437	\| VBOXVMM_EXIT_LTR_ENABLED_RAW()
10438	\| VBOXVMM_EXIT_GETSEC_ENABLED_RAW()
10439	\| VBOXVMM_EXIT_RSM_ENABLED_RAW()
10440	\| VBOXVMM_EXIT_RDRAND_ENABLED_RAW()
10441	\| VBOXVMM_EXIT_RDSEED_ENABLED_RAW()
10442	\| VBOXVMM_EXIT_XSAVES_ENABLED_RAW()
10443	\| VBOXVMM_EXIT_XRSTORS_ENABLED_RAW()
10444	\| VBOXVMM_EXIT_VMM_CALL_ENABLED_RAW()
10445	\| VBOXVMM_EXIT_VMX_VMCLEAR_ENABLED_RAW()
10446	\| VBOXVMM_EXIT_VMX_VMLAUNCH_ENABLED_RAW()
10447	\| VBOXVMM_EXIT_VMX_VMPTRLD_ENABLED_RAW()
10448	\| VBOXVMM_EXIT_VMX_VMPTRST_ENABLED_RAW()
10449	\| VBOXVMM_EXIT_VMX_VMREAD_ENABLED_RAW()
10450	\| VBOXVMM_EXIT_VMX_VMRESUME_ENABLED_RAW()
10451	\| VBOXVMM_EXIT_VMX_VMWRITE_ENABLED_RAW()
10452	\| VBOXVMM_EXIT_VMX_VMXOFF_ENABLED_RAW()
10453	\| VBOXVMM_EXIT_VMX_VMXON_ENABLED_RAW()
10454	\| VBOXVMM_EXIT_VMX_VMFUNC_ENABLED_RAW()
10455	\| VBOXVMM_EXIT_VMX_INVEPT_ENABLED_RAW()
10456	\| VBOXVMM_EXIT_VMX_INVVPID_ENABLED_RAW()
10457	\| VBOXVMM_EXIT_VMX_INVPCID_ENABLED_RAW()
10458	\| VBOXVMM_EXIT_VMX_EPT_VIOLATION_ENABLED_RAW()
10459	\| VBOXVMM_EXIT_VMX_EPT_MISCONFIG_ENABLED_RAW()
10460	\| VBOXVMM_EXIT_VMX_VAPIC_ACCESS_ENABLED_RAW()
10461	\| VBOXVMM_EXIT_VMX_VAPIC_WRITE_ENABLED_RAW()
10462	) != 0;
10463	}
10464
10465
10466	/**
10467	* Runs the guest code using VT-x.
10468	*
10469	* @returns Strict VBox status code (i.e. informational status codes too).
10470	* @param pVM The cross context VM structure.
10471	* @param pVCpu The cross context virtual CPU structure.
10472	* @param pCtx Pointer to the guest-CPU context.
10473	*/
10474	VMMR0DECL(VBOXSTRICTRC) VMXR0RunGuestCode(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
10475	{
10476	Assert(VMMRZCallRing3IsEnabled(pVCpu));
10477	Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
10478	HMVMX_ASSERT_PREEMPT_SAFE();
10479
10480	VMMRZCallRing3SetNotification(pVCpu, hmR0VmxCallRing3Callback, pCtx);
10481
10482	VBOXSTRICTRC rcStrict;
10483	if ( !pVCpu->hm.s.fUseDebugLoop
10484	&& (!VBOXVMM_ANY_PROBES_ENABLED() \|\| !hmR0VmxAnyExpensiveProbesEnabled())
10485	&& !DBGFIsStepping(pVCpu)
10486	&& !pVM->dbgf.ro.cEnabledInt3Breakpoints)
10487	rcStrict = hmR0VmxRunGuestCodeNormal(pVM, pVCpu, pCtx);
10488	else
10489	rcStrict = hmR0VmxRunGuestCodeDebug(pVM, pVCpu, pCtx);
10490
10491	if (rcStrict == VERR_EM_INTERPRETER)
10492	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
10493	else if (rcStrict == VINF_EM_RESET)
10494	rcStrict = VINF_EM_TRIPLE_FAULT;
10495
10496	int rc2 = hmR0VmxExitToRing3(pVM, pVCpu, pCtx, rcStrict);
10497	if (RT_FAILURE(rc2))
10498	{
10499	pVCpu->hm.s.u32HMError = (uint32_t)VBOXSTRICTRC_VAL(rcStrict);
10500	rcStrict = rc2;
10501	}
10502	Assert(!VMMRZCallRing3IsNotificationSet(pVCpu));
10503	return rcStrict;
10504	}
10505
10506
10507	#ifndef HMVMX_USE_FUNCTION_TABLE
10508	DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, uint32_t rcReason)
10509	{
10510	# ifdef DEBUG_ramshankar
10511	# define RETURN_EXIT_CALL(a_CallExpr) \
10512	do { \
10513	int rc2 = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRC(rc2); \
10514	VBOXSTRICTRC rcStrict = a_CallExpr; \
10515	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST); \
10516	return rcStrict; \
10517	} while (0)
10518	# else
10519	# define RETURN_EXIT_CALL(a_CallExpr) return a_CallExpr
10520	# endif
10521	switch (rcReason)
10522	{
10523	case VMX_EXIT_EPT_MISCONFIG: RETURN_EXIT_CALL(hmR0VmxExitEptMisconfig(pVCpu, pMixedCtx, pVmxTransient));
10524	case VMX_EXIT_EPT_VIOLATION: RETURN_EXIT_CALL(hmR0VmxExitEptViolation(pVCpu, pMixedCtx, pVmxTransient));
10525	case VMX_EXIT_IO_INSTR: RETURN_EXIT_CALL(hmR0VmxExitIoInstr(pVCpu, pMixedCtx, pVmxTransient));
10526	case VMX_EXIT_CPUID: RETURN_EXIT_CALL(hmR0VmxExitCpuid(pVCpu, pMixedCtx, pVmxTransient));
10527	case VMX_EXIT_RDTSC: RETURN_EXIT_CALL(hmR0VmxExitRdtsc(pVCpu, pMixedCtx, pVmxTransient));
10528	case VMX_EXIT_RDTSCP: RETURN_EXIT_CALL(hmR0VmxExitRdtscp(pVCpu, pMixedCtx, pVmxTransient));
10529	case VMX_EXIT_APIC_ACCESS: RETURN_EXIT_CALL(hmR0VmxExitApicAccess(pVCpu, pMixedCtx, pVmxTransient));
10530	case VMX_EXIT_XCPT_OR_NMI: RETURN_EXIT_CALL(hmR0VmxExitXcptOrNmi(pVCpu, pMixedCtx, pVmxTransient));
10531	case VMX_EXIT_MOV_CRX: RETURN_EXIT_CALL(hmR0VmxExitMovCRx(pVCpu, pMixedCtx, pVmxTransient));
10532	case VMX_EXIT_EXT_INT: RETURN_EXIT_CALL(hmR0VmxExitExtInt(pVCpu, pMixedCtx, pVmxTransient));
10533	case VMX_EXIT_INT_WINDOW: RETURN_EXIT_CALL(hmR0VmxExitIntWindow(pVCpu, pMixedCtx, pVmxTransient));
10534	case VMX_EXIT_MWAIT: RETURN_EXIT_CALL(hmR0VmxExitMwait(pVCpu, pMixedCtx, pVmxTransient));
10535	case VMX_EXIT_MONITOR: RETURN_EXIT_CALL(hmR0VmxExitMonitor(pVCpu, pMixedCtx, pVmxTransient));
10536	case VMX_EXIT_TASK_SWITCH: RETURN_EXIT_CALL(hmR0VmxExitTaskSwitch(pVCpu, pMixedCtx, pVmxTransient));
10537	case VMX_EXIT_PREEMPT_TIMER: RETURN_EXIT_CALL(hmR0VmxExitPreemptTimer(pVCpu, pMixedCtx, pVmxTransient));
10538	case VMX_EXIT_RDMSR: RETURN_EXIT_CALL(hmR0VmxExitRdmsr(pVCpu, pMixedCtx, pVmxTransient));
10539	case VMX_EXIT_WRMSR: RETURN_EXIT_CALL(hmR0VmxExitWrmsr(pVCpu, pMixedCtx, pVmxTransient));
10540	case VMX_EXIT_MOV_DRX: RETURN_EXIT_CALL(hmR0VmxExitMovDRx(pVCpu, pMixedCtx, pVmxTransient));
10541	case VMX_EXIT_TPR_BELOW_THRESHOLD: RETURN_EXIT_CALL(hmR0VmxExitTprBelowThreshold(pVCpu, pMixedCtx, pVmxTransient));
10542	case VMX_EXIT_HLT: RETURN_EXIT_CALL(hmR0VmxExitHlt(pVCpu, pMixedCtx, pVmxTransient));
10543	case VMX_EXIT_INVD: RETURN_EXIT_CALL(hmR0VmxExitInvd(pVCpu, pMixedCtx, pVmxTransient));
10544	case VMX_EXIT_INVLPG: RETURN_EXIT_CALL(hmR0VmxExitInvlpg(pVCpu, pMixedCtx, pVmxTransient));
10545	case VMX_EXIT_RSM: RETURN_EXIT_CALL(hmR0VmxExitRsm(pVCpu, pMixedCtx, pVmxTransient));
10546	case VMX_EXIT_MTF: RETURN_EXIT_CALL(hmR0VmxExitMtf(pVCpu, pMixedCtx, pVmxTransient));
10547	case VMX_EXIT_PAUSE: RETURN_EXIT_CALL(hmR0VmxExitPause(pVCpu, pMixedCtx, pVmxTransient));
10548	case VMX_EXIT_XDTR_ACCESS: RETURN_EXIT_CALL(hmR0VmxExitXdtrAccess(pVCpu, pMixedCtx, pVmxTransient));
10549	case VMX_EXIT_TR_ACCESS: RETURN_EXIT_CALL(hmR0VmxExitXdtrAccess(pVCpu, pMixedCtx, pVmxTransient));
10550	case VMX_EXIT_WBINVD: RETURN_EXIT_CALL(hmR0VmxExitWbinvd(pVCpu, pMixedCtx, pVmxTransient));
10551	case VMX_EXIT_XSETBV: RETURN_EXIT_CALL(hmR0VmxExitXsetbv(pVCpu, pMixedCtx, pVmxTransient));
10552	case VMX_EXIT_RDRAND: RETURN_EXIT_CALL(hmR0VmxExitRdrand(pVCpu, pMixedCtx, pVmxTransient));
10553	case VMX_EXIT_INVPCID: RETURN_EXIT_CALL(hmR0VmxExitInvpcid(pVCpu, pMixedCtx, pVmxTransient));
10554	case VMX_EXIT_GETSEC: RETURN_EXIT_CALL(hmR0VmxExitGetsec(pVCpu, pMixedCtx, pVmxTransient));
10555	case VMX_EXIT_RDPMC: RETURN_EXIT_CALL(hmR0VmxExitRdpmc(pVCpu, pMixedCtx, pVmxTransient));
10556	case VMX_EXIT_VMCALL: RETURN_EXIT_CALL(hmR0VmxExitVmcall(pVCpu, pMixedCtx, pVmxTransient));
10557
10558	case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFault(pVCpu, pMixedCtx, pVmxTransient);
10559	case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindow(pVCpu, pMixedCtx, pVmxTransient);
10560	case VMX_EXIT_INIT_SIGNAL: return hmR0VmxExitInitSignal(pVCpu, pMixedCtx, pVmxTransient);
10561	case VMX_EXIT_SIPI: return hmR0VmxExitSipi(pVCpu, pMixedCtx, pVmxTransient);
10562	case VMX_EXIT_IO_SMI: return hmR0VmxExitIoSmi(pVCpu, pMixedCtx, pVmxTransient);
10563	case VMX_EXIT_SMI: return hmR0VmxExitSmi(pVCpu, pMixedCtx, pVmxTransient);
10564	case VMX_EXIT_ERR_MSR_LOAD: return hmR0VmxExitErrMsrLoad(pVCpu, pMixedCtx, pVmxTransient);
10565	case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestState(pVCpu, pMixedCtx, pVmxTransient);
10566	case VMX_EXIT_ERR_MACHINE_CHECK: return hmR0VmxExitErrMachineCheck(pVCpu, pMixedCtx, pVmxTransient);
10567
10568	case VMX_EXIT_VMCLEAR:
10569	case VMX_EXIT_VMLAUNCH:
10570	case VMX_EXIT_VMPTRLD:
10571	case VMX_EXIT_VMPTRST:
10572	case VMX_EXIT_VMREAD:
10573	case VMX_EXIT_VMRESUME:
10574	case VMX_EXIT_VMWRITE:
10575	case VMX_EXIT_VMXOFF:
10576	case VMX_EXIT_VMXON:
10577	case VMX_EXIT_INVEPT:
10578	case VMX_EXIT_INVVPID:
10579	case VMX_EXIT_VMFUNC:
10580	case VMX_EXIT_XSAVES:
10581	case VMX_EXIT_XRSTORS:
10582	return hmR0VmxExitSetPendingXcptUD(pVCpu, pMixedCtx, pVmxTransient);
10583	case VMX_EXIT_ENCLS:
10584	case VMX_EXIT_RDSEED: /* only spurious VM-exits, so undefined */
10585	case VMX_EXIT_PML_FULL:
10586	default:
10587	return hmR0VmxExitErrUndefined(pVCpu, pMixedCtx, pVmxTransient);
10588	}
10589	#undef RETURN_EXIT_CALL
10590	}
10591	#endif /* !HMVMX_USE_FUNCTION_TABLE */
10592
10593
10594	#ifdef VBOX_STRICT
10595	/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
10596	# define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \
10597	RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
10598
10599	# define HMVMX_ASSERT_PREEMPT_CPUID() \
10600	do { \
10601	RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
10602	AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
10603	} while (0)
10604
10605	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS() \
10606	do { \
10607	AssertPtr(pVCpu); \
10608	AssertPtr(pMixedCtx); \
10609	AssertPtr(pVmxTransient); \
10610	Assert(pVmxTransient->fVMEntryFailed == false); \
10611	Assert(ASMIntAreEnabled()); \
10612	HMVMX_ASSERT_PREEMPT_SAFE(); \
10613	HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \
10614	Log4Func(("vcpu[%RU32] -v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v\n", pVCpu->idCpu)); \
10615	HMVMX_ASSERT_PREEMPT_SAFE(); \
10616	if (VMMR0IsLogFlushDisabled(pVCpu)) \
10617	HMVMX_ASSERT_PREEMPT_CPUID(); \
10618	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
10619	} while (0)
10620
10621	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS() \
10622	do { \
10623	Log4Func(("\n")); \
10624	} while (0)
10625	#else /* nonstrict builds: */
10626	# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS() \
10627	do { \
10628	HMVMX_STOP_EXIT_DISPATCH_PROF(); \
10629	NOREF(pVCpu); NOREF(pMixedCtx); NOREF(pVmxTransient); \
10630	} while (0)
10631	# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS() do { } while (0)
10632	#endif
10633
10634
10635	/**
10636	* Advances the guest RIP by the specified number of bytes.
10637	*
10638	* @param pVCpu The cross context virtual CPU structure.
10639	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
10640	* out-of-sync. Make sure to update the required fields
10641	* before using them.
10642	* @param cbInstr Number of bytes to advance the RIP by.
10643	*
10644	* @remarks No-long-jump zone!!!
10645	*/
10646	DECLINLINE(void) hmR0VmxAdvanceGuestRipBy(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t cbInstr)
10647	{
10648	/* Advance the RIP. */
10649	pMixedCtx->rip += cbInstr;
10650	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
10651
10652	/* Update interrupt inhibition. */
10653	if ( VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)
10654	&& pMixedCtx->rip != EMGetInhibitInterruptsPC(pVCpu))
10655	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
10656	}
10657
10658
10659	/**
10660	* Advances the guest RIP after reading it from the VMCS.
10661	*
10662	* @returns VBox status code, no informational status codes.
10663	* @param pVCpu The cross context virtual CPU structure.
10664	* @param pMixedCtx Pointer to the guest-CPU context. The data maybe
10665	* out-of-sync. Make sure to update the required fields
10666	* before using them.
10667	* @param pVmxTransient Pointer to the VMX transient structure.
10668	*
10669	* @remarks No-long-jump zone!!!
10670	*/
10671	static int hmR0VmxAdvanceGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10672	{
10673	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
10674	rc \|= hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
10675	rc \|= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10676	AssertRCReturn(rc, rc);
10677
10678	hmR0VmxAdvanceGuestRipBy(pVCpu, pMixedCtx, pVmxTransient->cbInstr);
10679
10680	/*
10681	* Deliver a debug exception to the guest if it is single-stepping. Don't directly inject a #DB but use the
10682	* pending debug exception field as it takes care of priority of events.
10683	*
10684	* See Intel spec. 32.2.1 "Debug Exceptions".
10685	*/
10686	if ( !pVCpu->hm.s.fSingleInstruction
10687	&& pMixedCtx->eflags.Bits.u1TF)
10688	hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
10689
10690	return VINF_SUCCESS;
10691	}
10692
10693
10694	/**
10695	* Tries to determine what part of the guest-state VT-x has deemed as invalid
10696	* and update error record fields accordingly.
10697	*
10698	* @return VMX_IGS_* return codes.
10699	* @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything
10700	* wrong with the guest state.
10701	*
10702	* @param pVM The cross context VM structure.
10703	* @param pVCpu The cross context virtual CPU structure.
10704	* @param pCtx Pointer to the guest-CPU state.
10705	*
10706	* @remarks This function assumes our cache of the VMCS controls
10707	* are valid, i.e. hmR0VmxCheckVmcsCtls() succeeded.
10708	*/
10709	static uint32_t hmR0VmxCheckGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
10710	{
10711	#define HMVMX_ERROR_BREAK(err) { uError = (err); break; }
10712	#define HMVMX_CHECK_BREAK(expr, err) if (!(expr)) { \
10713	uError = (err); \
10714	break; \
10715	} else do { } while (0)
10716
10717	int rc;
10718	uint32_t uError = VMX_IGS_ERROR;
10719	uint32_t u32Val;
10720	bool fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuest;
10721
10722	do
10723	{
10724	/*
10725	* CR0.
10726	*/
10727	uint32_t uSetCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
10728	uint32_t uZapCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
10729	/* Exceptions for unrestricted-guests for fixed CR0 bits (PE, PG).
10730	See Intel spec. 26.3.1 "Checks on Guest Control Registers, Debug Registers and MSRs." */
10731	if (fUnrestrictedGuest)
10732	uSetCR0 &= ~(X86_CR0_PE \| X86_CR0_PG);
10733
10734	uint32_t u32GuestCR0;
10735	rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32GuestCR0);
10736	AssertRCBreak(rc);
10737	HMVMX_CHECK_BREAK((u32GuestCR0 & uSetCR0) == uSetCR0, VMX_IGS_CR0_FIXED1);
10738	HMVMX_CHECK_BREAK(!(u32GuestCR0 & ~uZapCR0), VMX_IGS_CR0_FIXED0);
10739	if ( !fUnrestrictedGuest
10740	&& (u32GuestCR0 & X86_CR0_PG)
10741	&& !(u32GuestCR0 & X86_CR0_PE))
10742	{
10743	HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO);
10744	}
10745
10746	/*
10747	* CR4.
10748	*/
10749	uint64_t uSetCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
10750	uint64_t uZapCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 \| pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
10751
10752	uint32_t u32GuestCR4;
10753	rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &u32GuestCR4);
10754	AssertRCBreak(rc);
10755	HMVMX_CHECK_BREAK((u32GuestCR4 & uSetCR4) == uSetCR4, VMX_IGS_CR4_FIXED1);
10756	HMVMX_CHECK_BREAK(!(u32GuestCR4 & ~uZapCR4), VMX_IGS_CR4_FIXED0);
10757
10758	/*
10759	* IA32_DEBUGCTL MSR.
10760	*/
10761	uint64_t u64Val;
10762	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val);
10763	AssertRCBreak(rc);
10764	if ( (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
10765	&& (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */
10766	{
10767	HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED);
10768	}
10769	uint64_t u64DebugCtlMsr = u64Val;
10770
10771	#ifdef VBOX_STRICT
10772	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
10773	AssertRCBreak(rc);
10774	Assert(u32Val == pVCpu->hm.s.vmx.u32EntryCtls);
10775	#endif
10776	bool const fLongModeGuest = RT_BOOL(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST);
10777
10778	/*
10779	* RIP and RFLAGS.
10780	*/
10781	uint32_t u32Eflags;
10782	#if HC_ARCH_BITS == 64
10783	rc = VMXReadVmcs64(VMX_VMCS_GUEST_RIP, &u64Val);
10784	AssertRCBreak(rc);
10785	/* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */
10786	if ( !fLongModeGuest
10787	\|\| !pCtx->cs.Attr.n.u1Long)
10788	{
10789	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID);
10790	}
10791	/** @todo If the processor supports N < 64 linear-address bits, bits 63:N
10792	* must be identical if the "IA-32e mode guest" VM-entry
10793	* control is 1 and CS.L is 1. No check applies if the
10794	* CPU supports 64 linear-address bits. */
10795
10796	/* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */
10797	rc = VMXReadVmcs64(VMX_VMCS_GUEST_RFLAGS, &u64Val);
10798	AssertRCBreak(rc);
10799	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */
10800	VMX_IGS_RFLAGS_RESERVED);
10801	HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
10802	u32Eflags = u64Val;
10803	#else
10804	rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Eflags);
10805	AssertRCBreak(rc);
10806	HMVMX_CHECK_BREAK(!(u32Eflags & 0xffc08028), VMX_IGS_RFLAGS_RESERVED); /* Bit 31:22, Bit 15, 5, 3 MBZ. */
10807	HMVMX_CHECK_BREAK((u32Eflags & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
10808	#endif
10809
10810	if ( fLongModeGuest
10811	\|\| ( fUnrestrictedGuest
10812	&& !(u32GuestCR0 & X86_CR0_PE)))
10813	{
10814	HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID);
10815	}
10816
10817	uint32_t u32EntryInfo;
10818	rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
10819	AssertRCBreak(rc);
10820	if ( VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo)
10821	&& VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
10822	{
10823	HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID);
10824	}
10825
10826	/*
10827	* 64-bit checks.
10828	*/
10829	#if HC_ARCH_BITS == 64
10830	if (fLongModeGuest)
10831	{
10832	HMVMX_CHECK_BREAK(u32GuestCR0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE);
10833	HMVMX_CHECK_BREAK(u32GuestCR4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE);
10834	}
10835
10836	if ( !fLongModeGuest
10837	&& (u32GuestCR4 & X86_CR4_PCIDE))
10838	{
10839	HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE);
10840	}
10841
10842	/** @todo CR3 field must be such that bits 63:52 and bits in the range
10843	* 51:32 beyond the processor's physical-address width are 0. */
10844
10845	if ( (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
10846	&& (pCtx->dr[7] & X86_DR7_MBZ_MASK))
10847	{
10848	HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED);
10849	}
10850
10851	rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val);
10852	AssertRCBreak(rc);
10853	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL);
10854
10855	rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val);
10856	AssertRCBreak(rc);
10857	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL);
10858	#endif
10859
10860	/*
10861	* PERF_GLOBAL MSR.
10862	*/
10863	if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR)
10864	{
10865	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val);
10866	AssertRCBreak(rc);
10867	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)),
10868	VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */
10869	}
10870
10871	/*
10872	* PAT MSR.
10873	*/
10874	if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR)
10875	{
10876	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val);
10877	AssertRCBreak(rc);
10878	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED);
10879	for (unsigned i = 0; i < 8; i++)
10880	{
10881	uint8_t u8Val = (u64Val & 0xff);
10882	if ( u8Val != 0 /* UC */
10883	&& u8Val != 1 /* WC */
10884	&& u8Val != 4 /* WT */
10885	&& u8Val != 5 /* WP */
10886	&& u8Val != 6 /* WB */
10887	&& u8Val != 7 /* UC- */)
10888	{
10889	HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID);
10890	}
10891	u64Val >>= 8;
10892	}
10893	}
10894
10895	/*
10896	* EFER MSR.
10897	*/
10898	if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR)
10899	{
10900	Assert(pVM->hm.s.vmx.fSupportsVmcsEfer);
10901	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val);
10902	AssertRCBreak(rc);
10903	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)),
10904	VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */
10905	HMVMX_CHECK_BREAK(RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL( pVCpu->hm.s.vmx.u32EntryCtls
10906	& VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST),
10907	VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH);
10908	HMVMX_CHECK_BREAK( fUnrestrictedGuest
10909	\|\| !(u32GuestCR0 & X86_CR0_PG)
10910	\|\| RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(u64Val & MSR_K6_EFER_LME),
10911	VMX_IGS_EFER_LMA_LME_MISMATCH);
10912	}
10913
10914	/*
10915	* Segment registers.
10916	*/
10917	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10918	\|\| !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID);
10919	if (!(u32Eflags & X86_EFL_VM))
10920	{
10921	/* CS */
10922	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID);
10923	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED);
10924	HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED);
10925	HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff
10926	\|\| !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
10927	HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000)
10928	\|\| (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
10929	/* CS cannot be loaded with NULL in protected mode. */
10930	HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE);
10931	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID);
10932	if (pCtx->cs.Attr.n.u4Type == 9 \|\| pCtx->cs.Attr.n.u4Type == 11)
10933	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL);
10934	else if (pCtx->cs.Attr.n.u4Type == 13 \|\| pCtx->cs.Attr.n.u4Type == 15)
10935	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH);
10936	else if (pVM->hm.s.vmx.fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3)
10937	HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID);
10938	else
10939	HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID);
10940
10941	/* SS */
10942	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10943	\|\| (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL);
10944	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL);
10945	if ( !(pCtx->cr0 & X86_CR0_PE)
10946	\|\| pCtx->cs.Attr.n.u4Type == 3)
10947	{
10948	HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID);
10949	}
10950	if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
10951	{
10952	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 \|\| pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID);
10953	HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID);
10954	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED);
10955	HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED);
10956	HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff
10957	\|\| !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
10958	HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000)
10959	\|\| (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
10960	}
10961
10962	/* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxWriteSegmentReg(). */
10963	if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
10964	{
10965	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID);
10966	HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID);
10967	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10968	\|\| pCtx->ds.Attr.n.u4Type > 11
10969	\|\| pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
10970	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED);
10971	HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED);
10972	HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff
10973	\|\| !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
10974	HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000)
10975	\|\| (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
10976	HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10977	\|\| (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID);
10978	}
10979	if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
10980	{
10981	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID);
10982	HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID);
10983	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10984	\|\| pCtx->es.Attr.n.u4Type > 11
10985	\|\| pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
10986	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED);
10987	HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED);
10988	HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff
10989	\|\| !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
10990	HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000)
10991	\|\| (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
10992	HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10993	\|\| (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID);
10994	}
10995	if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
10996	{
10997	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID);
10998	HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID);
10999	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
11000	\|\| pCtx->fs.Attr.n.u4Type > 11
11001	\|\| pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL);
11002	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED);
11003	HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED);
11004	HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff
11005	\|\| !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
11006	HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000)
11007	\|\| (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
11008	HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
11009	\|\| (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID);
11010	}
11011	if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
11012	{
11013	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID);
11014	HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID);
11015	HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
11016	\|\| pCtx->gs.Attr.n.u4Type > 11
11017	\|\| pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL);
11018	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED);
11019	HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED);
11020	HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff
11021	\|\| !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
11022	HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000)
11023	\|\| (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
11024	HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
11025	\|\| (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID);
11026	}
11027	/* 64-bit capable CPUs. */
11028	#if HC_ARCH_BITS == 64
11029	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
11030	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
11031	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
11032	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
11033	HMVMX_CHECK_BREAK(!(pCtx->cs.u64Base >> 32), VMX_IGS_LONGMODE_CS_BASE_INVALID);
11034	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !(pCtx->ss.u64Base >> 32),
11035	VMX_IGS_LONGMODE_SS_BASE_INVALID);
11036	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !(pCtx->ds.u64Base >> 32),
11037	VMX_IGS_LONGMODE_DS_BASE_INVALID);
11038	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !(pCtx->es.u64Base >> 32),
11039	VMX_IGS_LONGMODE_ES_BASE_INVALID);
11040	#endif
11041	}
11042	else
11043	{
11044	/* V86 mode checks. */
11045	uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
11046	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
11047	{
11048	u32CSAttr = 0xf3; u32SSAttr = 0xf3;
11049	u32DSAttr = 0xf3; u32ESAttr = 0xf3;
11050	u32FSAttr = 0xf3; u32GSAttr = 0xf3;
11051	}
11052	else
11053	{
11054	u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u;
11055	u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u;
11056	u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
11057	}
11058
11059	/* CS */
11060	HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID);
11061	HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID);
11062	HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID);
11063	/* SS */
11064	HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID);
11065	HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID);
11066	HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID);
11067	/* DS */
11068	HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID);
11069	HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID);
11070	HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID);
11071	/* ES */
11072	HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID);
11073	HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID);
11074	HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID);
11075	/* FS */
11076	HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID);
11077	HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID);
11078	HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID);
11079	/* GS */
11080	HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID);
11081	HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID);
11082	HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID);
11083	/* 64-bit capable CPUs. */
11084	#if HC_ARCH_BITS == 64
11085	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
11086	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
11087	HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
11088	\|\| X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
11089	HMVMX_CHECK_BREAK(!(pCtx->cs.u64Base >> 32), VMX_IGS_LONGMODE_CS_BASE_INVALID);
11090	HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) \|\| !(pCtx->ss.u64Base >> 32),
11091	VMX_IGS_LONGMODE_SS_BASE_INVALID);
11092	HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) \|\| !(pCtx->ds.u64Base >> 32),
11093	VMX_IGS_LONGMODE_DS_BASE_INVALID);
11094	HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) \|\| !(pCtx->es.u64Base >> 32),
11095	VMX_IGS_LONGMODE_ES_BASE_INVALID);
11096	#endif
11097	}
11098
11099	/*
11100	* TR.
11101	*/
11102	HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID);
11103	/* 64-bit capable CPUs. */
11104	#if HC_ARCH_BITS == 64
11105	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL);
11106	#endif
11107	if (fLongModeGuest)
11108	{
11109	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */
11110	VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID);
11111	}
11112	else
11113	{
11114	HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */
11115	\|\| pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/
11116	VMX_IGS_TR_ATTR_TYPE_INVALID);
11117	}
11118	HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID);
11119	HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID);
11120	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */
11121	HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff
11122	\|\| !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
11123	HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000)
11124	\|\| (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
11125	HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE);
11126
11127	/*
11128	* GDTR and IDTR.
11129	*/
11130	#if HC_ARCH_BITS == 64
11131	rc = VMXReadVmcs64(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
11132	AssertRCBreak(rc);
11133	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL);
11134
11135	rc = VMXReadVmcs64(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
11136	AssertRCBreak(rc);
11137	HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL);
11138	#endif
11139
11140	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
11141	AssertRCBreak(rc);
11142	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
11143
11144	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
11145	AssertRCBreak(rc);
11146	HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
11147
11148	/*
11149	* Guest Non-Register State.
11150	*/
11151	/* Activity State. */
11152	uint32_t u32ActivityState;
11153	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState);
11154	AssertRCBreak(rc);
11155	HMVMX_CHECK_BREAK( !u32ActivityState
11156	\|\| (u32ActivityState & MSR_IA32_VMX_MISC_ACTIVITY_STATES(pVM->hm.s.vmx.Msrs.u64Misc)),
11157	VMX_IGS_ACTIVITY_STATE_INVALID);
11158	HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl)
11159	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID);
11160	uint32_t u32IntrState;
11161	rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &u32IntrState);
11162	AssertRCBreak(rc);
11163	if ( u32IntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS
11164	\|\| u32IntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
11165	{
11166	HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID);
11167	}
11168
11169	/** @todo Activity state and injecting interrupts. Left as a todo since we
11170	* currently don't use activity states but ACTIVE. */
11171
11172	HMVMX_CHECK_BREAK( !(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_ENTRY_SMM)
11173	\|\| u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID);
11174
11175	/* Guest interruptibility-state. */
11176	HMVMX_CHECK_BREAK(!(u32IntrState & 0xfffffff0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED);
11177	HMVMX_CHECK_BREAK((u32IntrState & ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI
11178	\| VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS))
11179	!= ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI
11180	\| VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
11181	VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID);
11182	HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF)
11183	\|\| !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI),
11184	VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID);
11185	if (VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo))
11186	{
11187	if (VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
11188	{
11189	HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
11190	&& !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
11191	VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID);
11192	}
11193	else if (VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
11194	{
11195	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
11196	VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID);
11197	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI),
11198	VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID);
11199	}
11200	}
11201	/** @todo Assumes the processor is not in SMM. */
11202	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI),
11203	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID);
11204	HMVMX_CHECK_BREAK( !(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_ENTRY_SMM)
11205	\|\| (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI),
11206	VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID);
11207	if ( (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
11208	&& VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo)
11209	&& VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
11210	{
11211	HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI),
11212	VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID);
11213	}
11214
11215	/* Pending debug exceptions. */
11216	#if HC_ARCH_BITS == 64
11217	rc = VMXReadVmcs64(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, &u64Val);
11218	AssertRCBreak(rc);
11219	/* Bits 63:15, Bit 13, Bits 11:4 MBZ. */
11220	HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED);
11221	u32Val = u64Val; /* For pending debug exceptions checks below. */
11222	#else
11223	rc = VMXReadVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, &u32Val);
11224	AssertRCBreak(rc);
11225	/* Bits 31:15, Bit 13, Bits 11:4 MBZ. */
11226	HMVMX_CHECK_BREAK(!(u32Val & 0xffffaff0), VMX_IGS_PENDING_DEBUG_RESERVED);
11227	#endif
11228
11229	if ( (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
11230	\|\| (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS)
11231	\|\| u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
11232	{
11233	if ( (u32Eflags & X86_EFL_TF)
11234	&& !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
11235	{
11236	/* Bit 14 is PendingDebug.BS. */
11237	HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET);
11238	}
11239	if ( !(u32Eflags & X86_EFL_TF)
11240	\|\| (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
11241	{
11242	/* Bit 14 is PendingDebug.BS. */
11243	HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR);
11244	}
11245	}
11246
11247	/* VMCS link pointer. */
11248	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val);
11249	AssertRCBreak(rc);
11250	if (u64Val != UINT64_C(0xffffffffffffffff))
11251	{
11252	HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED);
11253	/** @todo Bits beyond the processor's physical-address width MBZ. */
11254	/** @todo 32-bit located in memory referenced by value of this field (as a
11255	* physical address) must contain the processor's VMCS revision ID. */
11256	/** @todo SMM checks. */
11257	}
11258
11259	/** @todo Checks on Guest Page-Directory-Pointer-Table Entries when guest is
11260	* not using Nested Paging? */
11261	if ( pVM->hm.s.fNestedPaging
11262	&& !fLongModeGuest
11263	&& CPUMIsGuestInPAEModeEx(pCtx))
11264	{
11265	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &u64Val);
11266	AssertRCBreak(rc);
11267	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
11268
11269	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &u64Val);
11270	AssertRCBreak(rc);
11271	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
11272
11273	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &u64Val);
11274	AssertRCBreak(rc);
11275	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
11276
11277	rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &u64Val);
11278	AssertRCBreak(rc);
11279	HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
11280	}
11281
11282	/* Shouldn't happen but distinguish it from AssertRCBreak() errors. */
11283	if (uError == VMX_IGS_ERROR)
11284	uError = VMX_IGS_REASON_NOT_FOUND;
11285	} while (0);
11286
11287	pVCpu->hm.s.u32HMError = uError;
11288	return uError;
11289
11290	#undef HMVMX_ERROR_BREAK
11291	#undef HMVMX_CHECK_BREAK
11292	}
11293
11294	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
11295	/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
11296	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
11297
11298	/** @name VM-exit handlers.
11299	* @{
11300	*/
11301
11302	/**
11303	* VM-exit handler for external interrupts (VMX_EXIT_EXT_INT).
11304	*/
11305	HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11306	{
11307	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11308	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
11309	/* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */
11310	if (VMMR0ThreadCtxHookIsEnabled(pVCpu))
11311	return VINF_SUCCESS;
11312	return VINF_EM_RAW_INTERRUPT;
11313	}
11314
11315
11316	/**
11317	* VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI).
11318	*/
11319	HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11320	{
11321	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11322	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3);
11323
11324	int rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11325	AssertRCReturn(rc, rc);
11326
11327	uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVmxTransient->uExitIntInfo);
11328	Assert( !(pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT)
11329	&& uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT);
11330	Assert(VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
11331
11332	if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
11333	{
11334	/*
11335	* This cannot be a guest NMI as the only way for the guest to receive an NMI is if we injected it ourselves and
11336	* anything we inject is not going to cause a VM-exit directly for the event being injected.
11337	* See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery".
11338	*
11339	* Dispatch the NMI to the host. See Intel spec. 27.5.5 "Updating Non-Register State".
11340	*/
11341	VMXDispatchHostNmi();
11342	STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
11343	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
11344	return VINF_SUCCESS;
11345	}
11346
11347	/* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
11348	VBOXSTRICTRC rcStrictRc1 = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
11349	if (RT_UNLIKELY(rcStrictRc1 == VINF_SUCCESS))
11350	{ /* likely */ }
11351	else
11352	{
11353	if (rcStrictRc1 == VINF_HM_DOUBLE_FAULT)
11354	rcStrictRc1 = VINF_SUCCESS;
11355	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
11356	return rcStrictRc1;
11357	}
11358
11359	uint32_t uExitIntInfo = pVmxTransient->uExitIntInfo;
11360	uint32_t uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(uExitIntInfo);
11361	switch (uIntType)
11362	{
11363	case VMX_EXIT_INTERRUPTION_INFO_TYPE_PRIV_SW_XCPT: /* Privileged software exception. (#DB from ICEBP) */
11364	Assert(uVector == X86_XCPT_DB);
11365	/* fall thru */
11366	case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT: /* Software exception. (#BP or #OF) */
11367	Assert(uVector == X86_XCPT_BP \|\| uVector == X86_XCPT_OF \|\| uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_PRIV_SW_XCPT);
11368	/* fall thru */
11369	case VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT:
11370	{
11371	/*
11372	* If there's any exception caused as a result of event injection, go back to
11373	* the interpreter. The page-fault case is complicated and we manually handle
11374	* any currently pending event in hmR0VmxExitXcptPF. Nested #ACs are already
11375	* handled in hmR0VmxCheckExitDueToEventDelivery.
11376	*/
11377	if (!pVCpu->hm.s.Event.fPending)
11378	{ /* likely */ }
11379	else if ( uVector != X86_XCPT_PF
11380	&& uVector != X86_XCPT_AC)
11381	{
11382	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingInterpret);
11383	rc = VERR_EM_INTERPRETER;
11384	break;
11385	}
11386
11387	switch (uVector)
11388	{
11389	case X86_XCPT_PF: rc = hmR0VmxExitXcptPF(pVCpu, pMixedCtx, pVmxTransient); break;
11390	case X86_XCPT_GP: rc = hmR0VmxExitXcptGP(pVCpu, pMixedCtx, pVmxTransient); break;
11391	case X86_XCPT_NM: rc = hmR0VmxExitXcptNM(pVCpu, pMixedCtx, pVmxTransient); break;
11392	case X86_XCPT_MF: rc = hmR0VmxExitXcptMF(pVCpu, pMixedCtx, pVmxTransient); break;
11393	case X86_XCPT_DB: rc = hmR0VmxExitXcptDB(pVCpu, pMixedCtx, pVmxTransient); break;
11394	case X86_XCPT_BP: rc = hmR0VmxExitXcptBP(pVCpu, pMixedCtx, pVmxTransient); break;
11395	case X86_XCPT_AC: rc = hmR0VmxExitXcptAC(pVCpu, pMixedCtx, pVmxTransient); break;
11396
11397	case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF);
11398	rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11399	case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE);
11400	rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11401	case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD);
11402	rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11403	case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS);
11404	rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11405	case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP);
11406	rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11407	case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS);
11408	rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11409	default:
11410	{
11411	rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11412	AssertRCReturn(rc, rc);
11413
11414	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
11415	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
11416	{
11417	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
11418	Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
11419	Assert(CPUMIsGuestInRealModeEx(pMixedCtx));
11420
11421	rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11422	rc \|= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11423	AssertRCReturn(rc, rc);
11424	hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(uExitIntInfo),
11425	pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode,
11426	0 /* GCPtrFaultAddress */);
11427	AssertRCReturn(rc, rc);
11428	}
11429	else
11430	{
11431	AssertMsgFailed(("Unexpected VM-exit caused by exception %#x\n", uVector));
11432	pVCpu->hm.s.u32HMError = uVector;
11433	rc = VERR_VMX_UNEXPECTED_EXCEPTION;
11434	}
11435	break;
11436	}
11437	}
11438	break;
11439	}
11440
11441	default:
11442	{
11443	pVCpu->hm.s.u32HMError = uExitIntInfo;
11444	rc = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
11445	AssertMsgFailed(("Unexpected interruption info %#x\n", VMX_EXIT_INTERRUPTION_INFO_TYPE(uExitIntInfo)));
11446	break;
11447	}
11448	}
11449	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
11450	return rc;
11451	}
11452
11453
11454	/**
11455	* VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
11456	*/
11457	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindow(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11458	{
11459	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11460
11461	/* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */
11462	hmR0VmxClearIntWindowExitVmcs(pVCpu);
11463
11464	/* Deliver the pending interrupts via hmR0VmxEvaluatePendingEvent() and resume guest execution. */
11465	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
11466	return VINF_SUCCESS;
11467	}
11468
11469
11470	/**
11471	* VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
11472	*/
11473	HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindow(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11474	{
11475	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11476	if (RT_UNLIKELY(!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT)))
11477	{
11478	AssertMsgFailed(("Unexpected NMI-window exit.\n"));
11479	HMVMX_RETURN_UNEXPECTED_EXIT();
11480	}
11481
11482	Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS));
11483
11484	/*
11485	* If block-by-STI is set when we get this VM-exit, it means the CPU doesn't block NMIs following STI.
11486	* It is therefore safe to unblock STI and deliver the NMI ourselves. See @bugref{7445}.
11487	*/
11488	uint32_t uIntrState = 0;
11489	int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
11490	AssertRCReturn(rc, rc);
11491
11492	bool const fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
11493	if ( fBlockSti
11494	&& VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
11495	{
11496	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
11497	}
11498
11499	/* Indicate that we no longer need to VM-exit when the guest is ready to receive NMIs, it is now ready */
11500	hmR0VmxClearNmiWindowExitVmcs(pVCpu);
11501
11502	/* Deliver the pending NMI via hmR0VmxEvaluatePendingEvent() and resume guest execution. */
11503	return VINF_SUCCESS;
11504	}
11505
11506
11507	/**
11508	* VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
11509	*/
11510	HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvd(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11511	{
11512	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11513	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWbinvd);
11514	return hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11515	}
11516
11517
11518	/**
11519	* VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit.
11520	*/
11521	HMVMX_EXIT_NSRC_DECL hmR0VmxExitInvd(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11522	{
11523	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11524	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvd);
11525	return hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11526	}
11527
11528
11529	/**
11530	* VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit.
11531	*/
11532	HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11533	{
11534	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11535	PVM pVM = pVCpu->CTX_SUFF(pVM);
11536	int rc = EMInterpretCpuId(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11537	if (RT_LIKELY(rc == VINF_SUCCESS))
11538	{
11539	rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11540	Assert(pVmxTransient->cbInstr == 2);
11541	}
11542	else
11543	{
11544	AssertMsgFailed(("hmR0VmxExitCpuid: EMInterpretCpuId failed with %Rrc\n", rc));
11545	rc = VERR_EM_INTERPRETER;
11546	}
11547	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCpuid);
11548	return rc;
11549	}
11550
11551
11552	/**
11553	* VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit.
11554	*/
11555	HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11556	{
11557	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11558	int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
11559	AssertRCReturn(rc, rc);
11560
11561	if (pMixedCtx->cr4 & X86_CR4_SMXE)
11562	return VINF_EM_RAW_EMULATE_INSTR;
11563
11564	AssertMsgFailed(("hmR0VmxExitGetsec: unexpected VM-exit when CR4.SMXE is 0.\n"));
11565	HMVMX_RETURN_UNEXPECTED_EXIT();
11566	}
11567
11568
11569	/**
11570	* VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
11571	*/
11572	HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11573	{
11574	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11575	int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
11576	AssertRCReturn(rc, rc);
11577
11578	PVM pVM = pVCpu->CTX_SUFF(pVM);
11579	rc = EMInterpretRdtsc(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11580	if (RT_LIKELY(rc == VINF_SUCCESS))
11581	{
11582	rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11583	Assert(pVmxTransient->cbInstr == 2);
11584	/* If we get a spurious VM-exit when offsetting is enabled, we must reset offsetting on VM-reentry. See @bugref{6634}. */
11585	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING)
11586	pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
11587	}
11588	else
11589	rc = VERR_EM_INTERPRETER;
11590	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc);
11591	return rc;
11592	}
11593
11594
11595	/**
11596	* VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
11597	*/
11598	HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11599	{
11600	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11601	int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
11602	rc \|= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx); /* For MSR_K8_TSC_AUX */
11603	AssertRCReturn(rc, rc);
11604
11605	PVM pVM = pVCpu->CTX_SUFF(pVM);
11606	rc = EMInterpretRdtscp(pVM, pVCpu, pMixedCtx);
11607	if (RT_SUCCESS(rc))
11608	{
11609	rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11610	Assert(pVmxTransient->cbInstr == 3);
11611	/* If we get a spurious VM-exit when offsetting is enabled, we must reset offsetting on VM-reentry. See @bugref{6634}. */
11612	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING)
11613	pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
11614	}
11615	else
11616	{
11617	AssertMsgFailed(("hmR0VmxExitRdtscp: EMInterpretRdtscp failed with %Rrc\n", rc));
11618	rc = VERR_EM_INTERPRETER;
11619	}
11620	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc);
11621	return rc;
11622	}
11623
11624
11625	/**
11626	* VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
11627	*/
11628	HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11629	{
11630	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11631	int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
11632	rc \|= hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11633	AssertRCReturn(rc, rc);
11634
11635	PVM pVM = pVCpu->CTX_SUFF(pVM);
11636	rc = EMInterpretRdpmc(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11637	if (RT_LIKELY(rc == VINF_SUCCESS))
11638	{
11639	rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11640	Assert(pVmxTransient->cbInstr == 2);
11641	}
11642	else
11643	{
11644	AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
11645	rc = VERR_EM_INTERPRETER;
11646	}
11647	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdpmc);
11648	return rc;
11649	}
11650
11651
11652	/**
11653	* VM-exit handler for VMCALL (VMX_EXIT_VMCALL). Unconditional VM-exit.
11654	*/
11655	HMVMX_EXIT_DECL hmR0VmxExitVmcall(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11656	{
11657	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11658	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitVmcall);
11659
11660	VBOXSTRICTRC rcStrict = VERR_VMX_IPE_3;
11661	if (pVCpu->hm.s.fHypercallsEnabled)
11662	{
11663	#if 0
11664	int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11665	#else
11666	/* Aggressive state sync. for now. */
11667	int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
11668	rc \|= hmR0VmxSaveGuestRflags(pVCpu,pMixedCtx); /* For CPL checks in gimHvHypercall() & gimKvmHypercall() */
11669	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); /* For long-mode checks in gimKvmHypercall(). */
11670	AssertRCReturn(rc, rc);
11671	#endif
11672
11673	/* Perform the hypercall. */
11674	rcStrict = GIMHypercall(pVCpu, pMixedCtx);
11675	if (rcStrict == VINF_SUCCESS)
11676	{
11677	rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11678	AssertRCReturn(rc, rc);
11679	}
11680	else
11681	Assert( rcStrict == VINF_GIM_R3_HYPERCALL
11682	\|\| rcStrict == VINF_GIM_HYPERCALL_CONTINUING
11683	\|\| RT_FAILURE(VBOXSTRICTRC_VAL(rcStrict)));
11684
11685	/* If the hypercall changes anything other than guest's general-purpose registers,
11686	we would need to reload the guest changed bits here before VM-entry. */
11687	}
11688	else
11689	Log4(("hmR0VmxExitVmcall: Hypercalls not enabled\n"));
11690
11691	/* If hypercalls are disabled or the hypercall failed for some reason, raise #UD and continue. */
11692	if (RT_FAILURE(VBOXSTRICTRC_VAL(rcStrict)))
11693	{
11694	hmR0VmxSetPendingXcptUD(pVCpu, pMixedCtx);
11695	rcStrict = VINF_SUCCESS;
11696	}
11697
11698	return rcStrict;
11699	}
11700
11701
11702	/**
11703	* VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
11704	*/
11705	HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11706	{
11707	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11708	PVM pVM = pVCpu->CTX_SUFF(pVM);
11709	Assert(!pVM->hm.s.fNestedPaging \|\| pVCpu->hm.s.fUsingDebugLoop);
11710
11711	int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11712	rc \|= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
11713	AssertRCReturn(rc, rc);
11714
11715	VBOXSTRICTRC rcStrict = EMInterpretInvlpg(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), pVmxTransient->uExitQualification);
11716	if (RT_LIKELY(rcStrict == VINF_SUCCESS))
11717	rcStrict = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11718	else
11719	AssertMsg(rcStrict == VERR_EM_INTERPRETER, ("hmR0VmxExitInvlpg: EMInterpretInvlpg %#RX64 failed with %Rrc\n",
11720	pVmxTransient->uExitQualification, VBOXSTRICTRC_VAL(rcStrict)));
11721	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvlpg);
11722	return rcStrict;
11723	}
11724
11725
11726	/**
11727	* VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
11728	*/
11729	HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11730	{
11731	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11732	int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11733	rc \|= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
11734	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11735	AssertRCReturn(rc, rc);
11736
11737	PVM pVM = pVCpu->CTX_SUFF(pVM);
11738	rc = EMInterpretMonitor(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11739	if (RT_LIKELY(rc == VINF_SUCCESS))
11740	rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11741	else
11742	{
11743	AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMonitor: EMInterpretMonitor failed with %Rrc\n", rc));
11744	rc = VERR_EM_INTERPRETER;
11745	}
11746	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMonitor);
11747	return rc;
11748	}
11749
11750
11751	/**
11752	* VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
11753	*/
11754	HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11755	{
11756	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11757	int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11758	rc \|= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
11759	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11760	AssertRCReturn(rc, rc);
11761
11762	PVM pVM = pVCpu->CTX_SUFF(pVM);
11763	VBOXSTRICTRC rc2 = EMInterpretMWait(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11764	rc = VBOXSTRICTRC_VAL(rc2);
11765	if (RT_LIKELY( rc == VINF_SUCCESS
11766	\|\| rc == VINF_EM_HALT))
11767	{
11768	int rc3 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11769	AssertRCReturn(rc3, rc3);
11770
11771	if ( rc == VINF_EM_HALT
11772	&& EMMonitorWaitShouldContinue(pVCpu, pMixedCtx))
11773	{
11774	rc = VINF_SUCCESS;
11775	}
11776	}
11777	else
11778	{
11779	AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMwait: EMInterpretMWait failed with %Rrc\n", rc));
11780	rc = VERR_EM_INTERPRETER;
11781	}
11782	AssertMsg(rc == VINF_SUCCESS \|\| rc == VINF_EM_HALT \|\| rc == VERR_EM_INTERPRETER,
11783	("hmR0VmxExitMwait: failed, invalid error code %Rrc\n", rc));
11784	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMwait);
11785	return rc;
11786	}
11787
11788
11789	/**
11790	* VM-exit handler for RSM (VMX_EXIT_RSM). Unconditional VM-exit.
11791	*/
11792	HMVMX_EXIT_NSRC_DECL hmR0VmxExitRsm(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11793	{
11794	/*
11795	* Execution of RSM outside of SMM mode causes #UD regardless of VMX root or VMX non-root mode. In theory, we should never
11796	* get this VM-exit. This can happen only if dual-monitor treatment of SMI and VMX is enabled, which can (only?) be done by
11797	* executing VMCALL in VMX root operation. If we get here, something funny is going on.
11798	* See Intel spec. "33.15.5 Enabling the Dual-Monitor Treatment".
11799	*/
11800	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11801	AssertMsgFailed(("Unexpected RSM VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
11802	HMVMX_RETURN_UNEXPECTED_EXIT();
11803	}
11804
11805
11806	/**
11807	* VM-exit handler for SMI (VMX_EXIT_SMI). Unconditional VM-exit.
11808	*/
11809	HMVMX_EXIT_NSRC_DECL hmR0VmxExitSmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11810	{
11811	/*
11812	* This can only happen if we support dual-monitor treatment of SMI, which can be activated by executing VMCALL in VMX
11813	* root operation. Only an STM (SMM transfer monitor) would get this VM-exit when we (the executive monitor) execute a VMCALL
11814	* in VMX root mode or receive an SMI. If we get here, something funny is going on.
11815	* See Intel spec. "33.15.6 Activating the Dual-Monitor Treatment" and Intel spec. 25.3 "Other Causes of VM-Exits"
11816	*/
11817	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11818	AssertMsgFailed(("Unexpected SMI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
11819	HMVMX_RETURN_UNEXPECTED_EXIT();
11820	}
11821
11822
11823	/**
11824	* VM-exit handler for IO SMI (VMX_EXIT_IO_SMI). Unconditional VM-exit.
11825	*/
11826	HMVMX_EXIT_NSRC_DECL hmR0VmxExitIoSmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11827	{
11828	/* Same treatment as VMX_EXIT_SMI. See comment in hmR0VmxExitSmi(). */
11829	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11830	AssertMsgFailed(("Unexpected IO SMI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
11831	HMVMX_RETURN_UNEXPECTED_EXIT();
11832	}
11833
11834
11835	/**
11836	* VM-exit handler for SIPI (VMX_EXIT_SIPI). Conditional VM-exit.
11837	*/
11838	HMVMX_EXIT_NSRC_DECL hmR0VmxExitSipi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11839	{
11840	/*
11841	* SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest activity state is used. We currently
11842	* don't make use of it (see hmR0VmxLoadGuestActivityState()) as our guests don't have direct access to the host LAPIC.
11843	* See Intel spec. 25.3 "Other Causes of VM-exits".
11844	*/
11845	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11846	AssertMsgFailed(("Unexpected SIPI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
11847	HMVMX_RETURN_UNEXPECTED_EXIT();
11848	}
11849
11850
11851	/**
11852	* VM-exit handler for INIT signal (VMX_EXIT_INIT_SIGNAL). Unconditional
11853	* VM-exit.
11854	*/
11855	HMVMX_EXIT_NSRC_DECL hmR0VmxExitInitSignal(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11856	{
11857	/*
11858	* INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM.
11859	* See Intel spec. 33.14.1 Default Treatment of SMI Delivery" and Intel spec. 29.3 "VMX Instructions" for "VMXON".
11860	*
11861	* It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these VM-exits.
11862	* See Intel spec. "23.8 Restrictions on VMX operation".
11863	*/
11864	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11865	return VINF_SUCCESS;
11866	}
11867
11868
11869	/**
11870	* VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional
11871	* VM-exit.
11872	*/
11873	HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11874	{
11875	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11876	return VINF_EM_RESET;
11877	}
11878
11879
11880	/**
11881	* VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
11882	*/
11883	HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11884	{
11885	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11886	Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT);
11887
11888	int rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11889	AssertRCReturn(rc, rc);
11890
11891	if (EMShouldContinueAfterHalt(pVCpu, pMixedCtx)) /* Requires eflags. */
11892	rc = VINF_SUCCESS;
11893	else
11894	rc = VINF_EM_HALT;
11895
11896	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt);
11897	if (rc != VINF_SUCCESS)
11898	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
11899	return rc;
11900	}
11901
11902
11903	/**
11904	* VM-exit handler for instructions that result in a \#UD exception delivered to
11905	* the guest.
11906	*/
11907	HMVMX_EXIT_NSRC_DECL hmR0VmxExitSetPendingXcptUD(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11908	{
11909	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11910	hmR0VmxSetPendingXcptUD(pVCpu, pMixedCtx);
11911	return VINF_SUCCESS;
11912	}
11913
11914
11915	/**
11916	* VM-exit handler for expiry of the VMX preemption timer.
11917	*/
11918	HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11919	{
11920	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11921
11922	/* If the preemption-timer has expired, reinitialize the preemption timer on next VM-entry. */
11923	pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
11924
11925	/* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */
11926	PVM pVM = pVCpu->CTX_SUFF(pVM);
11927	bool fTimersPending = TMTimerPollBool(pVM, pVCpu);
11928	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer);
11929	return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS;
11930	}
11931
11932
11933	/**
11934	* VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit.
11935	*/
11936	HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11937	{
11938	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11939
11940	int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11941	rc \|= hmR0VmxSaveGuestRegsForIemExec(pVCpu, pMixedCtx, false /fMemory/, false /fNeedRsp/);
11942	rc \|= hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
11943	AssertRCReturn(rc, rc);
11944
11945	VBOXSTRICTRC rcStrict = IEMExecDecodedXsetbv(pVCpu, pVmxTransient->cbInstr);
11946	HMCPU_CF_SET(pVCpu, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS : HM_CHANGED_ALL_GUEST);
11947
11948	pVCpu->hm.s.fLoadSaveGuestXcr0 = (pMixedCtx->cr4 & X86_CR4_OSXSAVE) && pMixedCtx->aXcr[0] != ASMGetXcr0();
11949
11950	return rcStrict;
11951	}
11952
11953
11954	/**
11955	* VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
11956	*/
11957	HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11958	{
11959	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11960
11961	/* The guest should not invalidate the host CPU's TLBs, fallback to interpreter. */
11962	/** @todo implement EMInterpretInvpcid() */
11963	return VERR_EM_INTERPRETER;
11964	}
11965
11966
11967	/**
11968	* VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE).
11969	* Error VM-exit.
11970	*/
11971	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrInvalidGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11972	{
11973	int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11974	AssertRCReturn(rc, rc);
11975
11976	rc = hmR0VmxCheckVmcsCtls(pVCpu);
11977	AssertRCReturn(rc, rc);
11978
11979	uint32_t uInvalidReason = hmR0VmxCheckGuestState(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx);
11980	NOREF(uInvalidReason);
11981
11982	#ifdef VBOX_STRICT
11983	uint32_t uIntrState;
11984	RTHCUINTREG uHCReg;
11985	uint64_t u64Val;
11986	uint32_t u32Val;
11987
11988	rc = hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
11989	rc \|= hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient);
11990	rc \|= hmR0VmxReadEntryInstrLenVmcs(pVmxTransient);
11991	rc \|= VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
11992	AssertRCReturn(rc, rc);
11993
11994	Log4(("uInvalidReason %u\n", uInvalidReason));
11995	Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo));
11996	Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode));
11997	Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr));
11998	Log4(("VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE %#RX32\n", uIntrState));
11999
12000	rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32Val); AssertRC(rc);
12001	Log4(("VMX_VMCS_GUEST_CR0 %#RX32\n", u32Val));
12002	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc);
12003	Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg));
12004	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc);
12005	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
12006	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc);
12007	Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg));
12008	rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc);
12009	Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
12010	rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
12011	Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
12012	#else
12013	NOREF(pVmxTransient);
12014	#endif
12015
12016	hmDumpRegs(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx);
12017	return VERR_VMX_INVALID_GUEST_STATE;
12018	}
12019
12020
12021	/**
12022	* VM-exit handler for VM-entry failure due to an MSR-load
12023	* (VMX_EXIT_ERR_MSR_LOAD). Error VM-exit.
12024	*/
12025	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrMsrLoad(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12026	{
12027	NOREF(pVmxTransient);
12028	AssertMsgFailed(("Unexpected MSR-load exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); NOREF(pMixedCtx);
12029	HMVMX_RETURN_UNEXPECTED_EXIT();
12030	}
12031
12032
12033	/**
12034	* VM-exit handler for VM-entry failure due to a machine-check event
12035	* (VMX_EXIT_ERR_MACHINE_CHECK). Error VM-exit.
12036	*/
12037	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrMachineCheck(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12038	{
12039	NOREF(pVmxTransient);
12040	AssertMsgFailed(("Unexpected machine-check event exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); NOREF(pMixedCtx);
12041	HMVMX_RETURN_UNEXPECTED_EXIT();
12042	}
12043
12044
12045	/**
12046	* VM-exit handler for all undefined reasons. Should never ever happen.. in
12047	* theory.
12048	*/
12049	HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrUndefined(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12050	{
12051	AssertMsgFailed(("Huh!? Undefined VM-exit reason %d. pVCpu=%p pMixedCtx=%p\n", pVmxTransient->uExitReason, pVCpu, pMixedCtx));
12052	NOREF(pVCpu); NOREF(pMixedCtx); NOREF(pVmxTransient);
12053	return VERR_VMX_UNDEFINED_EXIT_CODE;
12054	}
12055
12056
12057	/**
12058	* VM-exit handler for XDTR (LGDT, SGDT, LIDT, SIDT) accesses
12059	* (VMX_EXIT_XDTR_ACCESS) and LDT and TR access (LLDT, LTR, SLDT, STR).
12060	* Conditional VM-exit.
12061	*/
12062	HMVMX_EXIT_DECL hmR0VmxExitXdtrAccess(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12063	{
12064	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12065
12066	/* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT. */
12067	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitXdtrAccess);
12068	if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT)
12069	return VERR_EM_INTERPRETER;
12070	AssertMsgFailed(("Unexpected XDTR access. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
12071	HMVMX_RETURN_UNEXPECTED_EXIT();
12072	}
12073
12074
12075	/**
12076	* VM-exit handler for RDRAND (VMX_EXIT_RDRAND). Conditional VM-exit.
12077	*/
12078	HMVMX_EXIT_DECL hmR0VmxExitRdrand(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12079	{
12080	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12081
12082	/* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT. */
12083	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdrand);
12084	if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT)
12085	return VERR_EM_INTERPRETER;
12086	AssertMsgFailed(("Unexpected RDRAND exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
12087	HMVMX_RETURN_UNEXPECTED_EXIT();
12088	}
12089
12090
12091	/**
12092	* VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
12093	*/
12094	HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12095	{
12096	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12097
12098	/* EMInterpretRdmsr() requires CR0, Eflags and SS segment register. */
12099	int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
12100	rc \|= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
12101	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12102	if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
12103	{
12104	rc \|= hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
12105	rc \|= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
12106	}
12107	AssertRCReturn(rc, rc);
12108	Log4(("ecx=%#RX32\n", pMixedCtx->ecx));
12109
12110	#ifdef VBOX_STRICT
12111	if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
12112	{
12113	if ( hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx)
12114	&& pMixedCtx->ecx != MSR_K6_EFER)
12115	{
12116	AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
12117	pMixedCtx->ecx));
12118	HMVMX_RETURN_UNEXPECTED_EXIT();
12119	}
12120	if (hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
12121	{
12122	VMXMSREXITREAD enmRead;
12123	VMXMSREXITWRITE enmWrite;
12124	int rc2 = hmR0VmxGetMsrPermission(pVCpu, pMixedCtx->ecx, &enmRead, &enmWrite);
12125	AssertRCReturn(rc2, rc2);
12126	if (enmRead == VMXMSREXIT_PASSTHRU_READ)
12127	{
12128	AssertMsgFailed(("Unexpected RDMSR for a passthru lazy-restore MSR. ecx=%#RX32\n", pMixedCtx->ecx));
12129	HMVMX_RETURN_UNEXPECTED_EXIT();
12130	}
12131	}
12132	}
12133	#endif
12134
12135	PVM pVM = pVCpu->CTX_SUFF(pVM);
12136	rc = EMInterpretRdmsr(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
12137	AssertMsg(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER,
12138	("hmR0VmxExitRdmsr: failed, invalid error code %Rrc\n", rc));
12139	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
12140	if (RT_SUCCESS(rc))
12141	{
12142	rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
12143	Assert(pVmxTransient->cbInstr == 2);
12144	}
12145	return rc;
12146	}
12147
12148
12149	/**
12150	* VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
12151	*/
12152	HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12153	{
12154	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12155	PVM pVM = pVCpu->CTX_SUFF(pVM);
12156	int rc = VINF_SUCCESS;
12157
12158	/* EMInterpretWrmsr() requires CR0, EFLAGS and SS segment register. */
12159	rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
12160	rc \|= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
12161	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12162	if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
12163	{
12164	rc \|= hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
12165	rc \|= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
12166	}
12167	AssertRCReturn(rc, rc);
12168	Log4(("ecx=%#RX32 edx:eax=%#RX32:%#RX32\n", pMixedCtx->ecx, pMixedCtx->edx, pMixedCtx->eax));
12169
12170	rc = EMInterpretWrmsr(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
12171	AssertMsg(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER, ("hmR0VmxExitWrmsr: failed, invalid error code %Rrc\n", rc));
12172	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
12173
12174	if (RT_SUCCESS(rc))
12175	{
12176	rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
12177
12178	/* If this is an X2APIC WRMSR access, update the APIC state as well. */
12179	if ( pMixedCtx->ecx == MSR_IA32_APICBASE
12180	\|\| ( pMixedCtx->ecx >= MSR_IA32_X2APIC_START
12181	&& pMixedCtx->ecx <= MSR_IA32_X2APIC_END))
12182	{
12183	/*
12184	* We've already saved the APIC related guest-state (TPR) in hmR0VmxPostRunGuest(). When full APIC register
12185	* virtualization is implemented we'll have to make sure APIC state is saved from the VMCS before
12186	* EMInterpretWrmsr() changes it.
12187	*/
12188	HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
12189	}
12190	else if (pMixedCtx->ecx == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */
12191	pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
12192	else if (pMixedCtx->ecx == MSR_K6_EFER)
12193	{
12194	/*
12195	* If the guest touches EFER we need to update the VM-Entry and VM-Exit controls as well,
12196	* even if it is -not- touching bits that cause paging mode changes (LMA/LME). We care about
12197	* the other bits as well, SCE and NXE. See @bugref{7368}.
12198	*/
12199	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_EFER_MSR \| HM_CHANGED_VMX_ENTRY_CTLS \| HM_CHANGED_VMX_EXIT_CTLS);
12200	}
12201
12202	/* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not supported. */
12203	if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
12204	{
12205	switch (pMixedCtx->ecx)
12206	{
12207	/*
12208	* For SYSENTER CS, EIP, ESP MSRs, we set both the flags here so we don't accidentally
12209	* overwrite the changed guest-CPU context value while going to ring-3, see @bufref{8745}.
12210	*/
12211	case MSR_IA32_SYSENTER_CS:
12212	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR);
12213	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR);
12214	break;
12215	case MSR_IA32_SYSENTER_EIP:
12216	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
12217	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR);
12218	break;
12219	case MSR_IA32_SYSENTER_ESP:
12220	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
12221	HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR);
12222	break;
12223	case MSR_K8_FS_BASE: /* fall thru */
12224	case MSR_K8_GS_BASE: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS); break;
12225	case MSR_K6_EFER: /* already handled above */ break;
12226	default:
12227	{
12228	if (hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx))
12229	HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
12230	else if (hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
12231	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS);
12232	break;
12233	}
12234	}
12235	}
12236	#ifdef VBOX_STRICT
12237	else
12238	{
12239	/* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */
12240	switch (pMixedCtx->ecx)
12241	{
12242	case MSR_IA32_SYSENTER_CS:
12243	case MSR_IA32_SYSENTER_EIP:
12244	case MSR_IA32_SYSENTER_ESP:
12245	case MSR_K8_FS_BASE:
12246	case MSR_K8_GS_BASE:
12247	{
12248	AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", pMixedCtx->ecx));
12249	HMVMX_RETURN_UNEXPECTED_EXIT();
12250	}
12251
12252	/* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */
12253	default:
12254	{
12255	if (hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx))
12256	{
12257	/* EFER writes are always intercepted, see hmR0VmxLoadGuestMsrs(). */
12258	if (pMixedCtx->ecx != MSR_K6_EFER)
12259	{
12260	AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
12261	pMixedCtx->ecx));
12262	HMVMX_RETURN_UNEXPECTED_EXIT();
12263	}
12264	}
12265
12266	if (hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
12267	{
12268	VMXMSREXITREAD enmRead;
12269	VMXMSREXITWRITE enmWrite;
12270	int rc2 = hmR0VmxGetMsrPermission(pVCpu, pMixedCtx->ecx, &enmRead, &enmWrite);
12271	AssertRCReturn(rc2, rc2);
12272	if (enmWrite == VMXMSREXIT_PASSTHRU_WRITE)
12273	{
12274	AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", pMixedCtx->ecx));
12275	HMVMX_RETURN_UNEXPECTED_EXIT();
12276	}
12277	}
12278	break;
12279	}
12280	}
12281	}
12282	#endif /* VBOX_STRICT */
12283	}
12284	return rc;
12285	}
12286
12287
12288	/**
12289	* VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
12290	*/
12291	HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12292	{
12293	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12294
12295	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPause);
12296	return VINF_EM_RAW_INTERRUPT;
12297	}
12298
12299
12300	/**
12301	* VM-exit handler for when the TPR value is lowered below the specified
12302	* threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
12303	*/
12304	HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThreshold(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12305	{
12306	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12307	Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW);
12308
12309	/*
12310	* The TPR shadow would've been synced with the APIC TPR in hmR0VmxPostRunGuest(). We'll re-evaluate
12311	* pending interrupts and inject them before the next VM-entry so we can just continue execution here.
12312	*/
12313	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold);
12314	return VINF_SUCCESS;
12315	}
12316
12317
12318	/**
12319	* VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional
12320	* VM-exit.
12321	*
12322	* @retval VINF_SUCCESS when guest execution can continue.
12323	* @retval VINF_PGM_CHANGE_MODE when shadow paging mode changed, back to ring-3.
12324	* @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3.
12325	* @retval VERR_EM_INTERPRETER when something unexpected happened, fallback to
12326	* interpreter.
12327	*/
12328	HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12329	{
12330	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12331	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2);
12332	int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12333	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
12334	AssertRCReturn(rc, rc);
12335
12336	RTGCUINTPTR const uExitQualification = pVmxTransient->uExitQualification;
12337	uint32_t const uAccessType = VMX_EXIT_QUALIFICATION_CRX_ACCESS(uExitQualification);
12338	PVM pVM = pVCpu->CTX_SUFF(pVM);
12339	VBOXSTRICTRC rcStrict;
12340	rc = hmR0VmxSaveGuestRegsForIemExec(pVCpu, pMixedCtx, false /fMemory/, true /fNeedRsp/);
12341	switch (uAccessType)
12342	{
12343	case VMX_EXIT_QUALIFICATION_CRX_ACCESS_WRITE: /* MOV to CRx */
12344	{
12345	rc \|= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
12346	AssertRCReturn(rc, rc);
12347
12348	rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, pVmxTransient->cbInstr,
12349	VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification),
12350	VMX_EXIT_QUALIFICATION_CRX_GENREG(uExitQualification));
12351	AssertMsg( rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_IEM_RAISED_XCPT \|\| rcStrict == VINF_PGM_CHANGE_MODE
12352	\|\| rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
12353	switch (VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification))
12354	{
12355	case 0: /* CR0 */
12356	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
12357	Log4(("CRX CR0 write rcStrict=%Rrc CR0=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->cr0));
12358	break;
12359	case 2: /* CR2 */
12360	/* Nothing to do here, CR2 it's not part of the VMCS. */
12361	break;
12362	case 3: /* CR3 */
12363	Assert(!pVM->hm.s.fNestedPaging \|\| !CPUMIsGuestPagingEnabledEx(pMixedCtx) \|\| pVCpu->hm.s.fUsingDebugLoop);
12364	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR3);
12365	Log4(("CRX CR3 write rcStrict=%Rrc CR3=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->cr3));
12366	break;
12367	case 4: /* CR4 */
12368	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR4);
12369	Log4(("CRX CR4 write rc=%Rrc CR4=%#RX64 fLoadSaveGuestXcr0=%u\n",
12370	VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->cr4, pVCpu->hm.s.fLoadSaveGuestXcr0));
12371	break;
12372	case 8: /* CR8 */
12373	Assert(!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW));
12374	/* CR8 contains the APIC TPR. Was updated by IEMExecDecodedMovCRxWrite(). */
12375	HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
12376	break;
12377	default:
12378	AssertMsgFailed(("Invalid CRx register %#x\n", VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)));
12379	break;
12380	}
12381
12382	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxWrite[VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)]);
12383	break;
12384	}
12385
12386	case VMX_EXIT_QUALIFICATION_CRX_ACCESS_READ: /* MOV from CRx */
12387	{
12388	rc \|= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
12389	AssertRCReturn(rc, rc);
12390
12391	Assert( !pVM->hm.s.fNestedPaging
12392	\|\| !CPUMIsGuestPagingEnabledEx(pMixedCtx)
12393	\|\| pVCpu->hm.s.fUsingDebugLoop
12394	\|\| VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification) != 3);
12395
12396	/* CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */
12397	Assert( VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification) != 8
12398	\|\| !(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW));
12399
12400	rcStrict = IEMExecDecodedMovCRxRead(pVCpu, pVmxTransient->cbInstr,
12401	VMX_EXIT_QUALIFICATION_CRX_GENREG(uExitQualification),
12402	VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification));
12403	AssertMsg(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
12404	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxRead[VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)]);
12405	Log4(("CRX CR%d Read access rcStrict=%Rrc\n", VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification),
12406	VBOXSTRICTRC_VAL(rcStrict)));
12407	break;
12408	}
12409
12410	case VMX_EXIT_QUALIFICATION_CRX_ACCESS_CLTS: /* CLTS (Clear Task-Switch Flag in CR0) */
12411	{
12412	AssertRCReturn(rc, rc);
12413	rcStrict = IEMExecDecodedClts(pVCpu, pVmxTransient->cbInstr);
12414	AssertMsg(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
12415	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
12416	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts);
12417	Log4(("CRX CLTS rcStrict=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
12418	break;
12419	}
12420
12421	case VMX_EXIT_QUALIFICATION_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */
12422	{
12423	AssertRCReturn(rc, rc);
12424	rcStrict = IEMExecDecodedLmsw(pVCpu, pVmxTransient->cbInstr,
12425	VMX_EXIT_QUALIFICATION_CRX_LMSW_DATA(uExitQualification));
12426	AssertMsg(rcStrict == VINF_SUCCESS \|\| rcStrict == VINF_IEM_RAISED_XCPT \|\| rcStrict == VINF_PGM_CHANGE_MODE,
12427	("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
12428	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
12429	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw);
12430	Log4(("CRX LMSW rcStrict=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
12431	break;
12432	}
12433
12434	default:
12435	AssertMsgFailedReturn(("Invalid access-type in Mov CRx VM-exit qualification %#x\n", uAccessType),
12436	VERR_VMX_UNEXPECTED_EXCEPTION);
12437	}
12438
12439	HMCPU_CF_SET(pVCpu, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS : HM_CHANGED_ALL_GUEST);
12440	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2);
12441	NOREF(pVM);
12442	return rcStrict;
12443	}
12444
12445
12446	/**
12447	* VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional
12448	* VM-exit.
12449	*/
12450	HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12451	{
12452	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12453	STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1);
12454
12455	int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12456	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
12457	rc \|= hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
12458	rc \|= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); /* Eflag checks in EMInterpretDisasCurrent(). */
12459	rc \|= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); /* CR0 checks & PGM* in EMInterpretDisasCurrent(). */
12460	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); /* SELM checks in EMInterpretDisasCurrent(). */
12461	/* EFER also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */
12462	AssertRCReturn(rc, rc);
12463
12464	/* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */
12465	uint32_t uIOPort = VMX_EXIT_QUALIFICATION_IO_PORT(pVmxTransient->uExitQualification);
12466	uint8_t uIOWidth = VMX_EXIT_QUALIFICATION_IO_WIDTH(pVmxTransient->uExitQualification);
12467	bool fIOWrite = ( VMX_EXIT_QUALIFICATION_IO_DIRECTION(pVmxTransient->uExitQualification)
12468	== VMX_EXIT_QUALIFICATION_IO_DIRECTION_OUT);
12469	bool fIOString = VMX_EXIT_QUALIFICATION_IO_IS_STRING(pVmxTransient->uExitQualification);
12470	bool fGstStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
12471	bool fDbgStepping = pVCpu->hm.s.fSingleInstruction;
12472	AssertReturn(uIOWidth <= 3 && uIOWidth != 2, VERR_VMX_IPE_1);
12473
12474	/* I/O operation lookup arrays. */
12475	static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses. */
12476	static uint32_t const s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving the result (in AL/AX/EAX). */
12477
12478	VBOXSTRICTRC rcStrict;
12479	uint32_t const cbValue = s_aIOSizes[uIOWidth];
12480	uint32_t const cbInstr = pVmxTransient->cbInstr;
12481	bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */
12482	PVM pVM = pVCpu->CTX_SUFF(pVM);
12483	if (fIOString)
12484	{
12485	#ifdef VBOX_WITH_2ND_IEM_STEP /* This used to gurus with debian 32-bit guest without NP (on ATA reads).
12486	See @bugref{5752#c158}. Should work now. */
12487	/*
12488	* INS/OUTS - I/O String instruction.
12489	*
12490	* Use instruction-information if available, otherwise fall back on
12491	* interpreting the instruction.
12492	*/
12493	Log4(("CS:RIP=%04x:%08RX64 %#06x/%u %c str\n", pMixedCtx->cs.Sel, pMixedCtx->rip, uIOPort, cbValue,
12494	fIOWrite ? 'w' : 'r'));
12495	AssertReturn(pMixedCtx->dx == uIOPort, VERR_VMX_IPE_2);
12496	if (MSR_IA32_VMX_BASIC_INFO_VMCS_INS_OUTS(pVM->hm.s.vmx.Msrs.u64BasicInfo))
12497	{
12498	int rc2 = hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12499	/** @todo optimize this, IEM should request the additional state if it needs it (GP, PF, ++). */
12500	rc2 \|= hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
12501	AssertRCReturn(rc2, rc2);
12502	AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_VMX_IPE_3);
12503	AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2);
12504	IEMMODE enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize;
12505	bool fRep = VMX_EXIT_QUALIFICATION_IO_IS_REP(pVmxTransient->uExitQualification);
12506	if (fIOWrite)
12507	{
12508	rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr,
12509	pVmxTransient->ExitInstrInfo.StrIo.iSegReg, true /fIoChecked/);
12510	}
12511	else
12512	{
12513	/*
12514	* The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES.
12515	* Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS.
12516	* See Intel Instruction spec. for "INS".
12517	* See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS".
12518	*/
12519	rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr, true /fIoChecked/);
12520	}
12521	}
12522	else
12523	{
12524	/** @todo optimize this, IEM should request the additional state if it needs it (GP, PF, ++). */
12525	int rc2 = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
12526	AssertRCReturn(rc2, rc2);
12527	rcStrict = IEMExecOne(pVCpu);
12528	}
12529	/** @todo IEM needs to be setting these flags somehow. */
12530	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
12531	fUpdateRipAlready = true;
12532	#else
12533	PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
12534	rcStrict = EMInterpretDisasCurrent(pVM, pVCpu, pDis, NULL /* pcbInstr */);
12535	if (RT_SUCCESS(rcStrict))
12536	{
12537	if (fIOWrite)
12538	{
12539	rcStrict = IOMInterpretOUTSEx(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), uIOPort, pDis->fPrefix,
12540	(DISCPUMODE)pDis->uAddrMode, cbValue);
12541	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringWrite);
12542	}
12543	else
12544	{
12545	rcStrict = IOMInterpretINSEx(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), uIOPort, pDis->fPrefix,
12546	(DISCPUMODE)pDis->uAddrMode, cbValue);
12547	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringRead);
12548	}
12549	}
12550	else
12551	{
12552	AssertMsg(rcStrict == VERR_EM_INTERPRETER, ("rcStrict=%Rrc RIP=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict),
12553	pMixedCtx->rip));
12554	rcStrict = VINF_EM_RAW_EMULATE_INSTR;
12555	}
12556	#endif
12557	}
12558	else
12559	{
12560	/*
12561	* IN/OUT - I/O instruction.
12562	*/
12563	Log4(("CS:RIP=%04x:%08RX64 %#06x/%u %c\n", pMixedCtx->cs.Sel, pMixedCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
12564	uint32_t const uAndVal = s_aIOOpAnd[uIOWidth];
12565	Assert(!VMX_EXIT_QUALIFICATION_IO_IS_REP(pVmxTransient->uExitQualification));
12566	if (fIOWrite)
12567	{
12568	rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pMixedCtx->eax & uAndVal, cbValue);
12569	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
12570	}
12571	else
12572	{
12573	uint32_t u32Result = 0;
12574	rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue);
12575	if (IOM_SUCCESS(rcStrict))
12576	{
12577	/* Save result of I/O IN instr. in AL/AX/EAX. */
12578	pMixedCtx->eax = (pMixedCtx->eax & ~uAndVal) \| (u32Result & uAndVal);
12579	}
12580	else if (rcStrict == VINF_IOM_R3_IOPORT_READ)
12581	HMR0SavePendingIOPortRead(pVCpu, pMixedCtx->rip, pMixedCtx->rip + cbInstr, uIOPort, uAndVal, cbValue);
12582	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
12583	}
12584	}
12585
12586	if (IOM_SUCCESS(rcStrict))
12587	{
12588	if (!fUpdateRipAlready)
12589	{
12590	hmR0VmxAdvanceGuestRipBy(pVCpu, pMixedCtx, cbInstr);
12591	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
12592	}
12593
12594	/*
12595	* INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru while booting Fedora 17 64-bit guest.
12596	* See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ.
12597	*/
12598	if (fIOString)
12599	{
12600	/** @todo Single-step for INS/OUTS with REP prefix? */
12601	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
12602	}
12603	else if ( !fDbgStepping
12604	&& fGstStepping)
12605	{
12606	hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
12607	}
12608
12609	/*
12610	* If any I/O breakpoints are armed, we need to check if one triggered
12611	* and take appropriate action.
12612	* Note that the I/O breakpoint type is undefined if CR4.DE is 0.
12613	*/
12614	int rc2 = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
12615	AssertRCReturn(rc2, rc2);
12616
12617	/** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
12618	* execution engines about whether hyper BPs and such are pending. */
12619	uint32_t const uDr7 = pMixedCtx->dr[7];
12620	if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
12621	&& X86_DR7_ANY_RW_IO(uDr7)
12622	&& (pMixedCtx->cr4 & X86_CR4_DE))
12623	\|\| DBGFBpIsHwIoArmed(pVM)))
12624	{
12625	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
12626
12627	/* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
12628	VMMRZCallRing3Disable(pVCpu);
12629	HM_DISABLE_PREEMPT();
12630
12631	bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /* fDr6 */);
12632
12633	VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pMixedCtx, uIOPort, cbValue);
12634	if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
12635	{
12636	/* Raise #DB. */
12637	if (fIsGuestDbgActive)
12638	ASMSetDR6(pMixedCtx->dr[6]);
12639	if (pMixedCtx->dr[7] != uDr7)
12640	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
12641
12642	hmR0VmxSetPendingXcptDB(pVCpu, pMixedCtx);
12643	}
12644	/* rcStrict is VINF_SUCCESS, VINF_IOM_R3_IOPORT_COMMIT_WRITE, or in [VINF_EM_FIRST..VINF_EM_LAST],
12645	however we can ditch VINF_IOM_R3_IOPORT_COMMIT_WRITE as it has VMCPU_FF_IOM as backup. */
12646	else if ( rcStrict2 != VINF_SUCCESS
12647	&& (rcStrict == VINF_SUCCESS \|\| rcStrict2 < rcStrict))
12648	rcStrict = rcStrict2;
12649	AssertCompile(VINF_EM_LAST < VINF_IOM_R3_IOPORT_COMMIT_WRITE);
12650
12651	HM_RESTORE_PREEMPT();
12652	VMMRZCallRing3Enable(pVCpu);
12653	}
12654	}
12655
12656	#ifdef VBOX_STRICT
12657	if (rcStrict == VINF_IOM_R3_IOPORT_READ)
12658	Assert(!fIOWrite);
12659	else if (rcStrict == VINF_IOM_R3_IOPORT_WRITE \|\| rcStrict == VINF_IOM_R3_IOPORT_COMMIT_WRITE)
12660	Assert(fIOWrite);
12661	else
12662	{
12663	#if 0 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
12664	* statuses, that the VMM device and some others may return. See
12665	* IOM_SUCCESS() for guidance. */
12666	AssertMsg( RT_FAILURE(rcStrict)
12667	\|\| rcStrict == VINF_SUCCESS
12668	\|\| rcStrict == VINF_EM_RAW_EMULATE_INSTR
12669	\|\| rcStrict == VINF_EM_DBG_BREAKPOINT
12670	\|\| rcStrict == VINF_EM_RAW_GUEST_TRAP
12671	\|\| rcStrict == VINF_EM_RAW_TO_R3
12672	\|\| rcStrict == VINF_TRPM_XCPT_DISPATCHED, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
12673	#endif
12674	}
12675	#endif
12676
12677	STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1);
12678	return rcStrict;
12679	}
12680
12681
12682	/**
12683	* VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional
12684	* VM-exit.
12685	*/
12686	HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12687	{
12688	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12689
12690	/* Check if this task-switch occurred while delivery an event through the guest IDT. */
12691	int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12692	AssertRCReturn(rc, rc);
12693	if (VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE_IDT)
12694	{
12695	rc = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
12696	AssertRCReturn(rc, rc);
12697	if (VMX_IDT_VECTORING_INFO_VALID(pVmxTransient->uIdtVectoringInfo))
12698	{
12699	uint32_t uIntType = VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo);
12700
12701	uint32_t uVector = VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo);
12702	bool fErrorCodeValid = VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uIdtVectoringInfo);
12703
12704	/* Save it as a pending event and it'll be converted to a TRPM event on the way out to ring-3. */
12705	Assert(!pVCpu->hm.s.Event.fPending);
12706	pVCpu->hm.s.Event.fPending = true;
12707	pVCpu->hm.s.Event.u64IntInfo = pVmxTransient->uIdtVectoringInfo;
12708	rc = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
12709	AssertRCReturn(rc, rc);
12710	if (fErrorCodeValid)
12711	pVCpu->hm.s.Event.u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
12712	else
12713	pVCpu->hm.s.Event.u32ErrCode = 0;
12714	if ( uIntType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
12715	&& uVector == X86_XCPT_PF)
12716	{
12717	pVCpu->hm.s.Event.GCPtrFaultAddress = pMixedCtx->cr2;
12718	}
12719
12720	Log4(("Pending event on TaskSwitch uIntType=%#x uVector=%#x\n", uIntType, uVector));
12721	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
12722	return VINF_EM_RAW_INJECT_TRPM_EVENT;
12723	}
12724	}
12725
12726	/* Fall back to the interpreter to emulate the task-switch. */
12727	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
12728	return VERR_EM_INTERPRETER;
12729	}
12730
12731
12732	/**
12733	* VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit.
12734	*/
12735	HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12736	{
12737	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12738	Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG);
12739	pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG;
12740	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
12741	AssertRCReturn(rc, rc);
12742	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMtf);
12743	return VINF_EM_DBG_STEPPED;
12744	}
12745
12746
12747	/**
12748	* VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit.
12749	*/
12750	HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12751	{
12752	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12753
12754	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess);
12755
12756	/* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
12757	VBOXSTRICTRC rcStrict1 = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
12758	if (RT_LIKELY(rcStrict1 == VINF_SUCCESS))
12759	{
12760	/* For some crazy guest, if an event delivery causes an APIC-access VM-exit, go to instruction emulation. */
12761	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
12762	{
12763	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingInterpret);
12764	return VERR_EM_INTERPRETER;
12765	}
12766	}
12767	else
12768	{
12769	if (rcStrict1 == VINF_HM_DOUBLE_FAULT)
12770	rcStrict1 = VINF_SUCCESS;
12771	return rcStrict1;
12772	}
12773
12774	#if 0
12775	/** @todo Investigate if IOMMMIOPhysHandler() requires a lot of state, for now
12776	* just sync the whole thing. */
12777	int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
12778	#else
12779	/* Aggressive state sync. for now. */
12780	int rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
12781	rc \|= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
12782	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12783	#endif
12784	rc \|= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12785	AssertRCReturn(rc, rc);
12786
12787	/* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */
12788	uint32_t uAccessType = VMX_EXIT_QUALIFICATION_APIC_ACCESS_TYPE(pVmxTransient->uExitQualification);
12789	VBOXSTRICTRC rcStrict2;
12790	switch (uAccessType)
12791	{
12792	case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE:
12793	case VMX_APIC_ACCESS_TYPE_LINEAR_READ:
12794	{
12795	AssertMsg( !(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
12796	\|\| VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification) != XAPIC_OFF_TPR,
12797	("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n"));
12798
12799	RTGCPHYS GCPhys = pVCpu->hm.s.vmx.u64MsrApicBase; /* Always up-to-date, u64MsrApicBase is not part of the VMCS. */
12800	GCPhys &= PAGE_BASE_GC_MASK;
12801	GCPhys += VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification);
12802	PVM pVM = pVCpu->CTX_SUFF(pVM);
12803	Log4(("ApicAccess uAccessType=%#x GCPhys=%#RGp Off=%#x\n", uAccessType, GCPhys,
12804	VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification)));
12805
12806	rcStrict2 = IOMMMIOPhysHandler(pVM, pVCpu,
12807	uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ ? 0 : X86_TRAP_PF_RW,
12808	CPUMCTX2CORE(pMixedCtx), GCPhys);
12809	Log4(("ApicAccess rcStrict2=%d\n", VBOXSTRICTRC_VAL(rcStrict2)));
12810	if ( rcStrict2 == VINF_SUCCESS
12811	\|\| rcStrict2 == VERR_PAGE_TABLE_NOT_PRESENT
12812	\|\| rcStrict2 == VERR_PAGE_NOT_PRESENT)
12813	{
12814	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
12815	\| HM_CHANGED_GUEST_RSP
12816	\| HM_CHANGED_GUEST_RFLAGS
12817	\| HM_CHANGED_VMX_GUEST_APIC_STATE);
12818	rcStrict2 = VINF_SUCCESS;
12819	}
12820	break;
12821	}
12822
12823	default:
12824	Log4(("ApicAccess uAccessType=%#x\n", uAccessType));
12825	rcStrict2 = VINF_EM_RAW_EMULATE_INSTR;
12826	break;
12827	}
12828
12829	if (rcStrict2 != VINF_SUCCESS)
12830	STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchApicAccessToR3);
12831	return rcStrict2;
12832	}
12833
12834
12835	/**
12836	* VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional
12837	* VM-exit.
12838	*/
12839	HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12840	{
12841	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12842
12843	/* We should -not- get this VM-exit if the guest's debug registers were active. */
12844	if (pVmxTransient->fWasGuestDebugStateActive)
12845	{
12846	AssertMsgFailed(("Unexpected MOV DRx exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
12847	HMVMX_RETURN_UNEXPECTED_EXIT();
12848	}
12849
12850	if ( !pVCpu->hm.s.fSingleInstruction
12851	&& !pVmxTransient->fWasHyperDebugStateActive)
12852	{
12853	Assert(!DBGFIsStepping(pVCpu));
12854	Assert(pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT_32(X86_XCPT_DB));
12855
12856	/* Don't intercept MOV DRx any more. */
12857	pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
12858	int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
12859	AssertRCReturn(rc, rc);
12860
12861	/* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */
12862	VMMRZCallRing3Disable(pVCpu);
12863	HM_DISABLE_PREEMPT();
12864
12865	/* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
12866	CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
12867	Assert(CPUMIsGuestDebugStateActive(pVCpu) \|\| HC_ARCH_BITS == 32);
12868
12869	HM_RESTORE_PREEMPT();
12870	VMMRZCallRing3Enable(pVCpu);
12871
12872	#ifdef VBOX_WITH_STATISTICS
12873	rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12874	AssertRCReturn(rc, rc);
12875	if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
12876	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
12877	else
12878	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
12879	#endif
12880	STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
12881	return VINF_SUCCESS;
12882	}
12883
12884	/*
12885	* EMInterpretDRx[Write\|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER, CS. EFER is always up-to-date.
12886	* Update the segment registers and DR7 from the CPU.
12887	*/
12888	int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12889	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12890	rc \|= hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
12891	AssertRCReturn(rc, rc);
12892	Log4(("CS:RIP=%04x:%08RX64\n", pMixedCtx->cs.Sel, pMixedCtx->rip));
12893
12894	PVM pVM = pVCpu->CTX_SUFF(pVM);
12895	if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
12896	{
12897	rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx),
12898	VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification),
12899	VMX_EXIT_QUALIFICATION_DRX_GENREG(pVmxTransient->uExitQualification));
12900	if (RT_SUCCESS(rc))
12901	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
12902	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
12903	}
12904	else
12905	{
12906	rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx),
12907	VMX_EXIT_QUALIFICATION_DRX_GENREG(pVmxTransient->uExitQualification),
12908	VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification));
12909	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
12910	}
12911
12912	Assert(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER);
12913	if (RT_SUCCESS(rc))
12914	{
12915	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
12916	AssertRCReturn(rc2, rc2);
12917	return VINF_SUCCESS;
12918	}
12919	return rc;
12920	}
12921
12922
12923	/**
12924	* VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG).
12925	* Conditional VM-exit.
12926	*/
12927	HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12928	{
12929	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12930	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
12931
12932	/* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
12933	VBOXSTRICTRC rcStrict1 = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
12934	if (RT_LIKELY(rcStrict1 == VINF_SUCCESS))
12935	{
12936	/* If event delivery causes an EPT misconfig (MMIO), go back to instruction emulation as otherwise
12937	injecting the original pending event would most likely cause the same EPT misconfig VM-exit. */
12938	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
12939	{
12940	STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingInterpret);
12941	return VERR_EM_INTERPRETER;
12942	}
12943	}
12944	else
12945	{
12946	if (rcStrict1 == VINF_HM_DOUBLE_FAULT)
12947	rcStrict1 = VINF_SUCCESS;
12948	return rcStrict1;
12949	}
12950
12951	RTGCPHYS GCPhys = 0;
12952	int rc = VMXReadVmcs64(VMX_VMCS64_EXIT_GUEST_PHYS_ADDR_FULL, &GCPhys);
12953
12954	#if 0
12955	rc \|= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); /** @todo Can we do better? */
12956	#else
12957	/* Aggressive state sync. for now. */
12958	rc \|= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
12959	rc \|= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
12960	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12961	#endif
12962	AssertRCReturn(rc, rc);
12963
12964	/*
12965	* If we succeed, resume guest execution.
12966	* If we fail in interpreting the instruction because we couldn't get the guest physical address
12967	* of the page containing the instruction via the guest's page tables (we would invalidate the guest page
12968	* in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
12969	* weird case. See @bugref{6043}.
12970	*/
12971	PVM pVM = pVCpu->CTX_SUFF(pVM);
12972	VBOXSTRICTRC rcStrict2 = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pMixedCtx), GCPhys, UINT32_MAX);
12973	Log4(("EPT misconfig at %#RGp RIP=%#RX64 rc=%Rrc\n", GCPhys, pMixedCtx->rip, VBOXSTRICTRC_VAL(rcStrict2)));
12974	if ( rcStrict2 == VINF_SUCCESS
12975	\|\| rcStrict2 == VERR_PAGE_TABLE_NOT_PRESENT
12976	\|\| rcStrict2 == VERR_PAGE_NOT_PRESENT)
12977	{
12978	/* Successfully handled MMIO operation. */
12979	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
12980	\| HM_CHANGED_GUEST_RSP
12981	\| HM_CHANGED_GUEST_RFLAGS
12982	\| HM_CHANGED_VMX_GUEST_APIC_STATE);
12983	return VINF_SUCCESS;
12984	}
12985	return rcStrict2;
12986	}
12987
12988
12989	/**
12990	* VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional
12991	* VM-exit.
12992	*/
12993	HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12994	{
12995	HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12996	Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
12997
12998	/* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
12999	VBOXSTRICTRC rcStrict1 = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
13000	if (RT_LIKELY(rcStrict1 == VINF_SUCCESS))
13001	{
13002	/* In the unlikely case that the EPT violation happened as a result of delivering an event, log it. */
13003	if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
13004	Log4(("EPT violation with an event pending u64IntInfo=%#RX64\n", pVCpu->hm.s.Event.u64IntInfo));
13005	}
13006	else
13007	{
13008	if (rcStrict1 == VINF_HM_DOUBLE_FAULT)
13009	rcStrict1 = VINF_SUCCESS;
13010	return rcStrict1;
13011	}
13012
13013	RTGCPHYS GCPhys = 0;
13014	int rc = VMXReadVmcs64(VMX_VMCS64_EXIT_GUEST_PHYS_ADDR_FULL, &GCPhys);
13015	rc \|= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
13016	#if 0
13017	rc \|= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); /** @todo Can we do better? */
13018	#else
13019	/* Aggressive state sync. for now. */
13020	rc \|= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
13021	rc \|= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
13022	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
13023	#endif
13024	AssertRCReturn(rc, rc);
13025
13026	/* Intel spec. Table 27-7 "Exit Qualifications for EPT violations". */
13027	AssertMsg(((pVmxTransient->uExitQualification >> 7) & 3) != 2, ("%#RX64", pVmxTransient->uExitQualification));
13028
13029	RTGCUINT uErrorCode = 0;
13030	if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_INSTR_FETCH)
13031	uErrorCode \|= X86_TRAP_PF_ID;
13032	if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_DATA_WRITE)
13033	uErrorCode \|= X86_TRAP_PF_RW;
13034	if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_ENTRY_PRESENT)
13035	uErrorCode \|= X86_TRAP_PF_P;
13036
13037	TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode);
13038
13039	Log4(("EPT violation %#x at %#RX64 ErrorCode %#x CS:RIP=%04x:%08RX64\n", pVmxTransient->uExitQualification, GCPhys,
13040	uErrorCode, pMixedCtx->cs.Sel, pMixedCtx->rip));
13041
13042	/* Handle the pagefault trap for the nested shadow table. */
13043	PVM pVM = pVCpu->CTX_SUFF(pVM);
13044	VBOXSTRICTRC rcStrict2 = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pMixedCtx), GCPhys);
13045	TRPMResetTrap(pVCpu);
13046
13047	/* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */
13048	if ( rcStrict2 == VINF_SUCCESS
13049	\|\| rcStrict2 == VERR_PAGE_TABLE_NOT_PRESENT
13050	\|\| rcStrict2 == VERR_PAGE_NOT_PRESENT)
13051	{
13052	/* Successfully synced our nested page tables. */
13053	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf);
13054	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
13055	\| HM_CHANGED_GUEST_RSP
13056	\| HM_CHANGED_GUEST_RFLAGS);
13057	return VINF_SUCCESS;
13058	}
13059
13060	Log4(("EPT return to ring-3 rcStrict2=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict2)));
13061	return rcStrict2;
13062	}
13063
13064	/** @} */
13065
13066	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13067	/* -=-=-=-=-=-=-=-=-=- VM-exit Exception Handlers -=-=-=-=-=-=-=-=-=-=- */
13068	/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-= */
13069
13070	/** @name VM-exit exception handlers.
13071	* @{
13072	*/
13073
13074	/**
13075	* VM-exit exception handler for \#MF (Math Fault: floating point exception).
13076	*/
13077	static int hmR0VmxExitXcptMF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13078	{
13079	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13080	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
13081
13082	int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
13083	AssertRCReturn(rc, rc);
13084
13085	if (!(pMixedCtx->cr0 & X86_CR0_NE))
13086	{
13087	/* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
13088	rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
13089
13090	/** @todo r=ramshankar: The Intel spec. does -not- specify that this VM-exit
13091	* provides VM-exit instruction length. If this causes problem later,
13092	* disassemble the instruction like it's done on AMD-V. */
13093	int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
13094	AssertRCReturn(rc2, rc2);
13095	return rc;
13096	}
13097
13098	hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13099	pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13100	return rc;
13101	}
13102
13103
13104	/**
13105	* VM-exit exception handler for \#BP (Breakpoint exception).
13106	*/
13107	static int hmR0VmxExitXcptBP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13108	{
13109	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13110	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
13111
13112	/** @todo Try optimize this by not saving the entire guest state unless
13113	* really needed. */
13114	int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
13115	AssertRCReturn(rc, rc);
13116
13117	PVM pVM = pVCpu->CTX_SUFF(pVM);
13118	rc = DBGFRZTrap03Handler(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
13119	if (rc == VINF_EM_RAW_GUEST_TRAP)
13120	{
13121	rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
13122	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13123	rc \|= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13124	AssertRCReturn(rc, rc);
13125
13126	hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13127	pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13128	}
13129
13130	Assert(rc == VINF_SUCCESS \|\| rc == VINF_EM_RAW_GUEST_TRAP \|\| rc == VINF_EM_DBG_BREAKPOINT);
13131	return rc;
13132	}
13133
13134
13135	/**
13136	* VM-exit exception handler for \#AC (alignment check exception).
13137	*/
13138	static int hmR0VmxExitXcptAC(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13139	{
13140	RT_NOREF_PV(pMixedCtx);
13141	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13142
13143	/*
13144	* Re-inject it. We'll detect any nesting before getting here.
13145	*/
13146	int rc = hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13147	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13148	AssertRCReturn(rc, rc);
13149	Assert(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO);
13150
13151	hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13152	pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13153	return VINF_SUCCESS;
13154	}
13155
13156
13157	/**
13158	* VM-exit exception handler for \#DB (Debug exception).
13159	*/
13160	static int hmR0VmxExitXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13161	{
13162	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13163	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
13164	Log6(("XcptDB\n"));
13165
13166	/*
13167	* Get the DR6-like values from the VM-exit qualification and pass it to DBGF
13168	* for processing.
13169	*/
13170	int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
13171	AssertRCReturn(rc, rc);
13172
13173	/* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */
13174	uint64_t uDR6 = X86_DR6_INIT_VAL;
13175	uDR6 \|= ( pVmxTransient->uExitQualification
13176	& (X86_DR6_B0 \| X86_DR6_B1 \| X86_DR6_B2 \| X86_DR6_B3 \| X86_DR6_BD \| X86_DR6_BS));
13177
13178	rc = DBGFRZTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pMixedCtx), uDR6, pVCpu->hm.s.fSingleInstruction);
13179	if (rc == VINF_EM_RAW_GUEST_TRAP)
13180	{
13181	/*
13182	* The exception was for the guest. Update DR6, DR7.GD and
13183	* IA32_DEBUGCTL.LBR before forwarding it.
13184	* (See Intel spec. 27.1 "Architectural State before a VM-Exit".)
13185	*/
13186	VMMRZCallRing3Disable(pVCpu);
13187	HM_DISABLE_PREEMPT();
13188
13189	pMixedCtx->dr[6] &= ~X86_DR6_B_MASK;
13190	pMixedCtx->dr[6] \|= uDR6;
13191	if (CPUMIsGuestDebugStateActive(pVCpu))
13192	ASMSetDR6(pMixedCtx->dr[6]);
13193
13194	HM_RESTORE_PREEMPT();
13195	VMMRZCallRing3Enable(pVCpu);
13196
13197	rc = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
13198	AssertRCReturn(rc, rc);
13199
13200	/* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
13201	pMixedCtx->dr[7] &= ~X86_DR7_GD;
13202
13203	/* Paranoia. */
13204	pMixedCtx->dr[7] &= ~X86_DR7_RAZ_MASK;
13205	pMixedCtx->dr[7] \|= X86_DR7_RA1_MASK;
13206
13207	rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)pMixedCtx->dr[7]);
13208	AssertRCReturn(rc, rc);
13209
13210	/*
13211	* Raise #DB in the guest.
13212	*
13213	* It is important to reflect what the VM-exit gave us (preserving the interruption-type) rather than use
13214	* hmR0VmxSetPendingXcptDB() as the #DB could've been raised while executing ICEBP and not the 'normal' #DB.
13215	* Thus it -may- trigger different handling in the CPU (like skipped DPL checks). See @bugref{6398}.
13216	*
13217	* Since ICEBP isn't documented on Intel, see AMD spec. 15.20 "Event Injection".
13218	*/
13219	rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
13220	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13221	rc \|= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13222	AssertRCReturn(rc, rc);
13223	hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13224	pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13225	return VINF_SUCCESS;
13226	}
13227
13228	/*
13229	* Not a guest trap, must be a hypervisor related debug event then.
13230	* Update DR6 in case someone is interested in it.
13231	*/
13232	AssertMsg(rc == VINF_EM_DBG_STEPPED \|\| rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc));
13233	AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5);
13234	CPUMSetHyperDR6(pVCpu, uDR6);
13235
13236	return rc;
13237	}
13238
13239
13240	/**
13241	* VM-exit exception handler for \#NM (Device-not-available exception: floating
13242	* point exception).
13243	*/
13244	static int hmR0VmxExitXcptNM(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13245	{
13246	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13247
13248	/* We require CR0 and EFER. EFER is always up-to-date. */
13249	int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
13250	AssertRCReturn(rc, rc);
13251
13252	/* We're playing with the host CPU state here, have to disable preemption or longjmp. */
13253	VMMRZCallRing3Disable(pVCpu);
13254	HM_DISABLE_PREEMPT();
13255
13256	/* If the guest FPU was active at the time of the #NM VM-exit, then it's a guest fault. */
13257	if (pVmxTransient->fWasGuestFPUStateActive)
13258	{
13259	rc = VINF_EM_RAW_GUEST_TRAP;
13260	Assert(CPUMIsGuestFPUStateActive(pVCpu) \|\| HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0));
13261	}
13262	else
13263	{
13264	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
13265	Assert(!pVmxTransient->fWasGuestFPUStateActive \|\| pVCpu->hm.s.fUsingDebugLoop);
13266	#endif
13267	rc = CPUMR0Trap07Handler(pVCpu->CTX_SUFF(pVM), pVCpu);
13268	Assert( rc == VINF_EM_RAW_GUEST_TRAP
13269	\|\| ((rc == VINF_SUCCESS \|\| rc == VINF_CPUM_HOST_CR0_MODIFIED) && CPUMIsGuestFPUStateActive(pVCpu)));
13270	if (rc == VINF_CPUM_HOST_CR0_MODIFIED)
13271	HMCPU_CF_SET(pVCpu, HM_CHANGED_HOST_CONTEXT);
13272	}
13273
13274	HM_RESTORE_PREEMPT();
13275	VMMRZCallRing3Enable(pVCpu);
13276
13277	if (rc == VINF_SUCCESS \|\| rc == VINF_CPUM_HOST_CR0_MODIFIED)
13278	{
13279	/* Guest FPU state was activated, we'll want to change CR0 FPU intercepts before the next VM-reentry. */
13280	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
13281	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowNM);
13282	pVCpu->hm.s.fPreloadGuestFpu = true;
13283	}
13284	else
13285	{
13286	/* Forward #NM to the guest. */
13287	Assert(rc == VINF_EM_RAW_GUEST_TRAP);
13288	rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
13289	AssertRCReturn(rc, rc);
13290	hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13291	pVmxTransient->cbInstr, 0 /* error code /, 0 / GCPtrFaultAddress */);
13292	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNM);
13293	}
13294
13295	return VINF_SUCCESS;
13296	}
13297
13298
13299	/**
13300	* VM-exit exception handler for \#GP (General-protection exception).
13301	*
13302	* @remarks Requires pVmxTransient->uExitIntInfo to be up-to-date.
13303	*/
13304	static int hmR0VmxExitXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13305	{
13306	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13307	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
13308
13309	int rc;
13310	if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
13311	{ /* likely */ }
13312	else
13313	{
13314	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
13315	Assert(pVCpu->hm.s.fUsingDebugLoop);
13316	#endif
13317	/* If the guest is not in real-mode or we have unrestricted execution support, reflect #GP to the guest. */
13318	rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
13319	rc \|= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13320	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13321	rc \|= hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
13322	AssertRCReturn(rc, rc);
13323	Log4(("#GP Gst: CS:RIP %04x:%08RX64 ErrorCode=%#x CR0=%#RX64 CPL=%u TR=%#04x\n", pMixedCtx->cs.Sel, pMixedCtx->rip,
13324	pVmxTransient->uExitIntErrorCode, pMixedCtx->cr0, CPUMGetGuestCPL(pVCpu), pMixedCtx->tr.Sel));
13325	hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13326	pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13327	return rc;
13328	}
13329
13330	Assert(CPUMIsGuestInRealModeEx(pMixedCtx));
13331	Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest);
13332
13333	/* EMInterpretDisasCurrent() requires a lot of the state, save the entire state. */
13334	rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
13335	AssertRCReturn(rc, rc);
13336
13337	PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
13338	uint32_t cbOp = 0;
13339	PVM pVM = pVCpu->CTX_SUFF(pVM);
13340	bool fDbgStepping = pVCpu->hm.s.fSingleInstruction;
13341	rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
13342	if (RT_SUCCESS(rc))
13343	{
13344	rc = VINF_SUCCESS;
13345	Assert(cbOp == pDis->cbInstr);
13346	Log4(("#GP Disas OpCode=%u CS:EIP %04x:%04RX64\n", pDis->pCurInstr->uOpcode, pMixedCtx->cs.Sel, pMixedCtx->rip));
13347	switch (pDis->pCurInstr->uOpcode)
13348	{
13349	case OP_CLI:
13350	{
13351	pMixedCtx->eflags.Bits.u1IF = 0;
13352	pMixedCtx->eflags.Bits.u1RF = 0;
13353	pMixedCtx->rip += pDis->cbInstr;
13354	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
13355	if ( !fDbgStepping
13356	&& pMixedCtx->eflags.Bits.u1TF)
13357	hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
13358	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCli);
13359	break;
13360	}
13361
13362	case OP_STI:
13363	{
13364	bool fOldIF = pMixedCtx->eflags.Bits.u1IF;
13365	pMixedCtx->eflags.Bits.u1IF = 1;
13366	pMixedCtx->eflags.Bits.u1RF = 0;
13367	pMixedCtx->rip += pDis->cbInstr;
13368	if (!fOldIF)
13369	{
13370	EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip);
13371	Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS));
13372	}
13373	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
13374	if ( !fDbgStepping
13375	&& pMixedCtx->eflags.Bits.u1TF)
13376	hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
13377	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitSti);
13378	break;
13379	}
13380
13381	case OP_HLT:
13382	{
13383	rc = VINF_EM_HALT;
13384	pMixedCtx->rip += pDis->cbInstr;
13385	pMixedCtx->eflags.Bits.u1RF = 0;
13386	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP \| HM_CHANGED_GUEST_RFLAGS);
13387	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt);
13388	break;
13389	}
13390
13391	case OP_POPF:
13392	{
13393	Log4(("POPF CS:EIP %04x:%04RX64\n", pMixedCtx->cs.Sel, pMixedCtx->rip));
13394	uint32_t cbParm;
13395	uint32_t uMask;
13396	bool fGstStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
13397	if (pDis->fPrefix & DISPREFIX_OPSIZE)
13398	{
13399	cbParm = 4;
13400	uMask = 0xffffffff;
13401	}
13402	else
13403	{
13404	cbParm = 2;
13405	uMask = 0xffff;
13406	}
13407
13408	/* Get the stack pointer & pop the contents of the stack onto Eflags. */
13409	RTGCPTR GCPtrStack = 0;
13410	X86EFLAGS Eflags;
13411	Eflags.u32 = 0;
13412	rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), pMixedCtx->esp & uMask, SELMTOFLAT_FLAGS_CPL0,
13413	&GCPtrStack);
13414	if (RT_SUCCESS(rc))
13415	{
13416	Assert(sizeof(Eflags.u32) >= cbParm);
13417	rc = VBOXSTRICTRC_TODO(PGMPhysRead(pVM, (RTGCPHYS)GCPtrStack, &Eflags.u32, cbParm, PGMACCESSORIGIN_HM));
13418	AssertMsg(rc == VINF_SUCCESS, ("%Rrc\n", rc)); /** @todo allow strict return codes here */
13419	}
13420	if (RT_FAILURE(rc))
13421	{
13422	rc = VERR_EM_INTERPRETER;
13423	break;
13424	}
13425	Log4(("POPF %#x -> %#RX64 mask=%#x RIP=%#RX64\n", Eflags.u, pMixedCtx->rsp, uMask, pMixedCtx->rip));
13426	pMixedCtx->eflags.u32 = (pMixedCtx->eflags.u32 & ~((X86_EFL_POPF_BITS & uMask) \| X86_EFL_RF))
13427	\| (Eflags.u32 & X86_EFL_POPF_BITS & uMask);
13428	pMixedCtx->esp += cbParm;
13429	pMixedCtx->esp &= uMask;
13430	pMixedCtx->rip += pDis->cbInstr;
13431	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
13432	\| HM_CHANGED_GUEST_RSP
13433	\| HM_CHANGED_GUEST_RFLAGS);
13434	/* Generate a pending-debug exception when the guest stepping over POPF regardless of how
13435	POPF restores EFLAGS.TF. */
13436	if ( !fDbgStepping
13437	&& fGstStepping)
13438	hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
13439	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPopf);
13440	break;
13441	}
13442
13443	case OP_PUSHF:
13444	{
13445	uint32_t cbParm;
13446	uint32_t uMask;
13447	if (pDis->fPrefix & DISPREFIX_OPSIZE)
13448	{
13449	cbParm = 4;
13450	uMask = 0xffffffff;
13451	}
13452	else
13453	{
13454	cbParm = 2;
13455	uMask = 0xffff;
13456	}
13457
13458	/* Get the stack pointer & push the contents of eflags onto the stack. */
13459	RTGCPTR GCPtrStack = 0;
13460	rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), (pMixedCtx->esp - cbParm) & uMask,
13461	SELMTOFLAT_FLAGS_CPL0, &GCPtrStack);
13462	if (RT_FAILURE(rc))
13463	{
13464	rc = VERR_EM_INTERPRETER;
13465	break;
13466	}
13467	X86EFLAGS Eflags = pMixedCtx->eflags;
13468	/* The RF & VM bits are cleared on image stored on stack; see Intel Instruction reference for PUSHF. */
13469	Eflags.Bits.u1RF = 0;
13470	Eflags.Bits.u1VM = 0;
13471
13472	rc = VBOXSTRICTRC_TODO(PGMPhysWrite(pVM, (RTGCPHYS)GCPtrStack, &Eflags.u, cbParm, PGMACCESSORIGIN_HM));
13473	if (RT_UNLIKELY(rc != VINF_SUCCESS))
13474	{
13475	AssertMsgFailed(("%Rrc\n", rc)); /** @todo allow strict return codes here */
13476	rc = VERR_EM_INTERPRETER;
13477	break;
13478	}
13479	Log4(("PUSHF %#x -> %#RGv\n", Eflags.u, GCPtrStack));
13480	pMixedCtx->esp -= cbParm;
13481	pMixedCtx->esp &= uMask;
13482	pMixedCtx->rip += pDis->cbInstr;
13483	pMixedCtx->eflags.Bits.u1RF = 0;
13484	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
13485	\| HM_CHANGED_GUEST_RSP
13486	\| HM_CHANGED_GUEST_RFLAGS);
13487	if ( !fDbgStepping
13488	&& pMixedCtx->eflags.Bits.u1TF)
13489	hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
13490	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPushf);
13491	break;
13492	}
13493
13494	case OP_IRET:
13495	{
13496	/** @todo Handle 32-bit operand sizes and check stack limits. See Intel
13497	* instruction reference. */
13498	RTGCPTR GCPtrStack = 0;
13499	uint32_t uMask = 0xffff;
13500	bool fGstStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
13501	uint16_t aIretFrame[3];
13502	if (pDis->fPrefix & (DISPREFIX_OPSIZE \| DISPREFIX_ADDRSIZE))
13503	{
13504	rc = VERR_EM_INTERPRETER;
13505	break;
13506	}
13507	rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), pMixedCtx->esp & uMask, SELMTOFLAT_FLAGS_CPL0,
13508	&GCPtrStack);
13509	if (RT_SUCCESS(rc))
13510	{
13511	rc = VBOXSTRICTRC_TODO(PGMPhysRead(pVM, (RTGCPHYS)GCPtrStack, &aIretFrame[0], sizeof(aIretFrame),
13512	PGMACCESSORIGIN_HM));
13513	AssertMsg(rc == VINF_SUCCESS, ("%Rrc\n", rc)); /** @todo allow strict return codes here */
13514	}
13515	if (RT_FAILURE(rc))
13516	{
13517	rc = VERR_EM_INTERPRETER;
13518	break;
13519	}
13520	pMixedCtx->eip = 0;
13521	pMixedCtx->ip = aIretFrame[0];
13522	pMixedCtx->cs.Sel = aIretFrame[1];
13523	pMixedCtx->cs.ValidSel = aIretFrame[1];
13524	pMixedCtx->cs.u64Base = (uint64_t)pMixedCtx->cs.Sel << 4;
13525	pMixedCtx->eflags.u32 = (pMixedCtx->eflags.u32 & ((UINT32_C(0xffff0000) \| X86_EFL_1) & ~X86_EFL_RF))
13526	\| (aIretFrame[2] & X86_EFL_POPF_BITS & uMask);
13527	pMixedCtx->sp += sizeof(aIretFrame);
13528	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
13529	\| HM_CHANGED_GUEST_SEGMENT_REGS
13530	\| HM_CHANGED_GUEST_RSP
13531	\| HM_CHANGED_GUEST_RFLAGS);
13532	/* Generate a pending-debug exception when stepping over IRET regardless of how IRET modifies EFLAGS.TF. */
13533	if ( !fDbgStepping
13534	&& fGstStepping)
13535	hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
13536	Log4(("IRET %#RX32 to %04x:%04x\n", GCPtrStack, pMixedCtx->cs.Sel, pMixedCtx->ip));
13537	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIret);
13538	break;
13539	}
13540
13541	case OP_INT:
13542	{
13543	uint16_t uVector = pDis->Param1.uValue & 0xff;
13544	hmR0VmxSetPendingIntN(pVCpu, pMixedCtx, uVector, pDis->cbInstr);
13545	/* INT clears EFLAGS.TF, we must not set any pending debug exceptions here. */
13546	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInt);
13547	break;
13548	}
13549
13550	case OP_INTO:
13551	{
13552	if (pMixedCtx->eflags.Bits.u1OF)
13553	{
13554	hmR0VmxSetPendingXcptOF(pVCpu, pMixedCtx, pDis->cbInstr);
13555	/* INTO clears EFLAGS.TF, we must not set any pending debug exceptions here. */
13556	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInt);
13557	}
13558	else
13559	{
13560	pMixedCtx->eflags.Bits.u1RF = 0;
13561	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
13562	}
13563	break;
13564	}
13565
13566	default:
13567	{
13568	pMixedCtx->eflags.Bits.u1RF = 0; /* This is correct most of the time... */
13569	VBOXSTRICTRC rc2 = EMInterpretInstructionDisasState(pVCpu, pDis, CPUMCTX2CORE(pMixedCtx), 0 /* pvFault */,
13570	EMCODETYPE_SUPERVISOR);
13571	rc = VBOXSTRICTRC_VAL(rc2);
13572	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
13573	/** @todo We have to set pending-debug exceptions here when the guest is
13574	* single-stepping depending on the instruction that was interpreted. */
13575	Log4(("#GP rc=%Rrc\n", rc));
13576	break;
13577	}
13578	}
13579	}
13580	else
13581	rc = VERR_EM_INTERPRETER;
13582
13583	AssertMsg(rc == VINF_SUCCESS \|\| rc == VERR_EM_INTERPRETER \|\| rc == VINF_PGM_CHANGE_MODE \|\| rc == VINF_EM_HALT,
13584	("#GP Unexpected rc=%Rrc\n", rc));
13585	return rc;
13586	}
13587
13588
13589	/**
13590	* VM-exit exception handler wrapper for generic exceptions. Simply re-injects
13591	* the exception reported in the VMX transient structure back into the VM.
13592	*
13593	* @remarks Requires uExitIntInfo in the VMX transient structure to be
13594	* up-to-date.
13595	*/
13596	static int hmR0VmxExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13597	{
13598	RT_NOREF_PV(pMixedCtx);
13599	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13600	#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
13601	Assert(pVCpu->hm.s.fUsingDebugLoop);
13602	#endif
13603
13604	/* Re-inject the exception into the guest. This cannot be a double-fault condition which would have been handled in
13605	hmR0VmxCheckExitDueToEventDelivery(). */
13606	int rc = hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13607	rc \|= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13608	AssertRCReturn(rc, rc);
13609	Assert(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO);
13610
13611	#ifdef DEBUG_ramshankar
13612	rc \|= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
13613	uint8_t uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo);
13614	Log(("hmR0VmxExitXcptGeneric: Reinjecting Xcpt. uVector=%#x cs:rip=%#04x:%#RX64\n", uVector, pCtx->cs.Sel, pCtx->rip));
13615	#endif
13616
13617	hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13618	pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13619	return VINF_SUCCESS;
13620	}
13621
13622
13623	/**
13624	* VM-exit exception handler for \#PF (Page-fault exception).
13625	*/
13626	static int hmR0VmxExitXcptPF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13627	{
13628	HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13629	PVM pVM = pVCpu->CTX_SUFF(pVM);
13630	int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
13631	rc \|= hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
13632	rc \|= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13633	AssertRCReturn(rc, rc);
13634
13635	if (!pVM->hm.s.fNestedPaging)
13636	{ /* likely */ }
13637	else
13638	{
13639	#if !defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) && !defined(HMVMX_ALWAYS_TRAP_PF)
13640	Assert(pVCpu->hm.s.fUsingDebugLoop);
13641	#endif
13642	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
13643	if (RT_LIKELY(!pVmxTransient->fVectoringDoublePF))
13644	{
13645	pMixedCtx->cr2 = pVmxTransient->uExitQualification; /* Update here in case we go back to ring-3 before injection. */
13646	hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13647	0 /* cbInstr */, pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQualification);
13648	}
13649	else
13650	{
13651	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
13652	hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
13653	Log4(("Pending #DF due to vectoring #PF. NP\n"));
13654	}
13655	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
13656	return rc;
13657	}
13658
13659	/* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
13660	of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
13661	if (pVmxTransient->fVectoringPF)
13662	{
13663	Assert(pVCpu->hm.s.Event.fPending);
13664	return VINF_EM_RAW_INJECT_TRPM_EVENT;
13665	}
13666
13667	rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
13668	AssertRCReturn(rc, rc);
13669
13670	Log4(("#PF: cr2=%#RX64 cs:rip=%#04x:%#RX64 uErrCode %#RX32 cr3=%#RX64\n", pVmxTransient->uExitQualification,
13671	pMixedCtx->cs.Sel, pMixedCtx->rip, pVmxTransient->uExitIntErrorCode, pMixedCtx->cr3));
13672
13673	TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQualification, (RTGCUINT)pVmxTransient->uExitIntErrorCode);
13674	rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pMixedCtx),
13675	(RTGCPTR)pVmxTransient->uExitQualification);
13676
13677	Log4(("#PF: rc=%Rrc\n", rc));
13678	if (rc == VINF_SUCCESS)
13679	{
13680	#if 0
13681	/* Successfully synced shadow pages tables or emulated an MMIO instruction. */
13682	/** @todo this isn't quite right, what if guest does lgdt with some MMIO
13683	* memory? We don't update the whole state here... */
13684	HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
13685	\| HM_CHANGED_GUEST_RSP
13686	\| HM_CHANGED_GUEST_RFLAGS
13687	\| HM_CHANGED_VMX_GUEST_APIC_STATE);
13688	#else
13689	/*
13690	* This is typically a shadow page table sync or a MMIO instruction. But we may have
13691	* emulated something like LTR or a far jump. Any part of the CPU context may have changed.
13692	*/
13693	/** @todo take advantage of CPUM changed flags instead of brute forcing. */
13694	HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
13695	#endif
13696	TRPMResetTrap(pVCpu);
13697	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
13698	return rc;
13699	}
13700
13701	if (rc == VINF_EM_RAW_GUEST_TRAP)
13702	{
13703	if (!pVmxTransient->fVectoringDoublePF)
13704	{
13705	/* It's a guest page fault and needs to be reflected to the guest. */
13706	uint32_t uGstErrorCode = TRPMGetErrorCode(pVCpu);
13707	TRPMResetTrap(pVCpu);
13708	pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */
13709	pMixedCtx->cr2 = pVmxTransient->uExitQualification; /* Update here in case we go back to ring-3 before injection. */
13710	hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13711	0 /* cbInstr */, uGstErrorCode, pVmxTransient->uExitQualification);
13712	}
13713	else
13714	{
13715	/* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
13716	TRPMResetTrap(pVCpu);
13717	pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */
13718	hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
13719	Log4(("#PF: Pending #DF due to vectoring #PF\n"));
13720	}
13721
13722	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
13723	return VINF_SUCCESS;
13724	}
13725
13726	TRPMResetTrap(pVCpu);
13727	STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
13728	return rc;
13729	}
13730
13731	/** @} */
13732

Note: See TracBrowser for help on using the repository browser.

source: vbox/trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp@ 66250

Download in other formats: