HMR0.cpp@ 95140

Last change on this file since 95140 was 93963, checked in by vboxsync, 3 years ago
VMM: Nested VMX: bugref:10092 Add HM ring-0 API for querying transient VMX/SVM info.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 69.3 KB

Line
1	/* $Id: HMR0.cpp 93963 2022-02-28 08:39:08Z vboxsync $ */
2	/** @file
3	* Hardware Assisted Virtualization Manager (HM) - Host Context Ring-0.
4	*/
5
6	/*
7	* Copyright (C) 2006-2022 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	#define VMCPU_INCL_CPUM_GST_CTX
24	#include <VBox/vmm/hm.h>
25	#include <VBox/vmm/pgm.h>
26	#include "HMInternal.h"
27	#include <VBox/vmm/vmcc.h>
28	#include <VBox/vmm/hm_svm.h>
29	#include <VBox/vmm/hmvmxinline.h>
30	#include <VBox/err.h>
31	#include <VBox/log.h>
32	#include <iprt/assert.h>
33	#include <iprt/asm.h>
34	#include <iprt/asm-amd64-x86.h>
35	#include <iprt/cpuset.h>
36	#include <iprt/mem.h>
37	#include <iprt/memobj.h>
38	#include <iprt/once.h>
39	#include <iprt/param.h>
40	#include <iprt/power.h>
41	#include <iprt/string.h>
42	#include <iprt/thread.h>
43	#include <iprt/x86.h>
44	#include "HMVMXR0.h"
45	#include "HMSVMR0.h"
46
47
48	/*********************************************************************************************************************************
49	* Internal Functions *
50	*********************************************************************************************************************************/
51	static DECLCALLBACK(void) hmR0EnableCpuCallback(RTCPUID idCpu, void pvUser1, void pvUser2);
52	static DECLCALLBACK(void) hmR0DisableCpuCallback(RTCPUID idCpu, void pvUser1, void pvUser2);
53	static DECLCALLBACK(void) hmR0PowerCallback(RTPOWEREVENT enmEvent, void *pvUser);
54	static DECLCALLBACK(void) hmR0MpEventCallback(RTMPEVENT enmEvent, RTCPUID idCpu, void *pvData);
55
56
57	/*********************************************************************************************************************************
58	* Structures and Typedefs *
59	*********************************************************************************************************************************/
60	/**
61	* This is used to manage the status code of a RTMpOnAll in HM.
62	*/
63	typedef struct HMR0FIRSTRC
64	{
65	/** The status code. */
66	int32_t volatile rc;
67	/** The ID of the CPU reporting the first failure. */
68	RTCPUID volatile idCpu;
69	} HMR0FIRSTRC;
70	/** Pointer to a first return code structure. */
71	typedef HMR0FIRSTRC *PHMR0FIRSTRC;
72
73	/**
74	* Ring-0 method table for AMD-V and VT-x specific operations.
75	*/
76	typedef struct HMR0VTABLE
77	{
78	DECLR0CALLBACKMEMBER(int, pfnEnterSession, (PVMCPUCC pVCpu));
79	DECLR0CALLBACKMEMBER(void, pfnThreadCtxCallback, (RTTHREADCTXEVENT enmEvent, PVMCPUCC pVCpu, bool fGlobalInit));
80	DECLR0CALLBACKMEMBER(int, pfnAssertionCallback, (PVMCPUCC pVCpu));
81	DECLR0CALLBACKMEMBER(int, pfnExportHostState, (PVMCPUCC pVCpu));
82	DECLR0CALLBACKMEMBER(VBOXSTRICTRC, pfnRunGuestCode, (PVMCPUCC pVCpu));
83	DECLR0CALLBACKMEMBER(int, pfnEnableCpu, (PHMPHYSCPU pHostCpu, PVMCC pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage,
84	bool fEnabledByHost, PCSUPHWVIRTMSRS pHwvirtMsrs));
85	DECLR0CALLBACKMEMBER(int, pfnDisableCpu, (PHMPHYSCPU pHostCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage));
86	DECLR0CALLBACKMEMBER(int, pfnInitVM, (PVMCC pVM));
87	DECLR0CALLBACKMEMBER(int, pfnTermVM, (PVMCC pVM));
88	DECLR0CALLBACKMEMBER(int, pfnSetupVM, (PVMCC pVM));
89	} HMR0VTABLE;
90
91
92	/*********************************************************************************************************************************
93	* Global Variables *
94	*********************************************************************************************************************************/
95	/** The active ring-0 HM operations (copied from one of the table at init). */
96	static HMR0VTABLE g_HmR0Ops;
97	/** Indicates whether the host is suspending or not. We'll refuse a few
98	* actions when the host is being suspended to speed up the suspending and
99	* avoid trouble. */
100	static bool volatile g_fHmSuspended;
101	/** If set, VT-x/AMD-V is enabled globally at init time, otherwise it's
102	* enabled and disabled each time it's used to execute guest code. */
103	static bool g_fHmGlobalInit;
104	/** Host kernel flags that HM might need to know (SUPKERNELFEATURES_XXX). */
105	uint32_t g_fHmHostKernelFeatures;
106	/** Maximum allowed ASID/VPID (inclusive).
107	* @todo r=bird: This is exclusive for VT-x according to source code comment.
108	* Couldn't immediately find any docs on AMD-V, but suspect it is
109	* exclusive there as well given how hmR0SvmFlushTaggedTlb() use it. */
110	uint32_t g_uHmMaxAsid;
111
112
113	/** Set if VT-x (VMX) is supported by the CPU. */
114	bool g_fHmVmxSupported = false;
115	/** VMX: Whether we're using the preemption timer or not. */
116	bool g_fHmVmxUsePreemptTimer;
117	/** VMX: The shift mask employed by the VMX-Preemption timer. */
118	uint8_t g_cHmVmxPreemptTimerShift;
119	/** VMX: Set if swapping EFER is supported. */
120	bool g_fHmVmxSupportsVmcsEfer = false;
121	/** VMX: Whether we're using SUPR0EnableVTx or not. */
122	static bool g_fHmVmxUsingSUPR0EnableVTx = false;
123	/** VMX: Set if we've called SUPR0EnableVTx(true) and should disable it during
124	* module termination. */
125	static bool g_fHmVmxCalledSUPR0EnableVTx = false;
126	/** VMX: Host CR4 value (set by ring-0 VMX init) */
127	uint64_t g_uHmVmxHostCr4;
128	/** VMX: Host EFER value (set by ring-0 VMX init) */
129	uint64_t g_uHmVmxHostMsrEfer;
130	/** VMX: Host SMM monitor control (used for logging/diagnostics) */
131	uint64_t g_uHmVmxHostSmmMonitorCtl;
132
133
134	/** Set if AMD-V is supported by the CPU. */
135	bool g_fHmSvmSupported = false;
136	/** SVM revision. */
137	uint32_t g_uHmSvmRev;
138	/** SVM feature bits from cpuid 0x8000000a */
139	uint32_t g_fHmSvmFeatures;
140
141
142	/** MSRs. */
143	SUPHWVIRTMSRS g_HmMsrs;
144
145	/** Last recorded error code during HM ring-0 init. */
146	static int32_t g_rcHmInit = VINF_SUCCESS;
147
148	/** Per CPU globals. */
149	static HMPHYSCPU g_aHmCpuInfo[RTCPUSET_MAX_CPUS];
150
151	/** Whether we've already initialized all CPUs.
152	* @remarks We could check the EnableAllCpusOnce state, but this is
153	* simpler and hopefully easier to understand. */
154	static bool g_fHmEnabled = false;
155	/** Serialize initialization in HMR0EnableAllCpus. */
156	static RTONCE g_HmEnableAllCpusOnce = RTONCE_INITIALIZER;
157
158
159	/** HM ring-0 operations for VT-x. */
160	static HMR0VTABLE const g_HmR0OpsVmx =
161	{
162	/* .pfnEnterSession = */ VMXR0Enter,
163	/* .pfnThreadCtxCallback = */ VMXR0ThreadCtxCallback,
164	/* .pfnAssertionCallback = */ VMXR0AssertionCallback,
165	/* .pfnExportHostState = */ VMXR0ExportHostState,
166	/* .pfnRunGuestCode = */ VMXR0RunGuestCode,
167	/* .pfnEnableCpu = */ VMXR0EnableCpu,
168	/* .pfnDisableCpu = */ VMXR0DisableCpu,
169	/* .pfnInitVM = */ VMXR0InitVM,
170	/* .pfnTermVM = */ VMXR0TermVM,
171	/* .pfnSetupVM = */ VMXR0SetupVM,
172	};
173
174	/** HM ring-0 operations for AMD-V. */
175	static HMR0VTABLE const g_HmR0OpsSvm =
176	{
177	/* .pfnEnterSession = */ SVMR0Enter,
178	/* .pfnThreadCtxCallback = */ SVMR0ThreadCtxCallback,
179	/* .pfnAssertionCallback = */ SVMR0AssertionCallback,
180	/* .pfnExportHostState = */ SVMR0ExportHostState,
181	/* .pfnRunGuestCode = */ SVMR0RunGuestCode,
182	/* .pfnEnableCpu = */ SVMR0EnableCpu,
183	/* .pfnDisableCpu = */ SVMR0DisableCpu,
184	/* .pfnInitVM = */ SVMR0InitVM,
185	/* .pfnTermVM = */ SVMR0TermVM,
186	/* .pfnSetupVM = */ SVMR0SetupVM,
187	};
188
189
190	/** @name Dummy callback handlers for when neither VT-x nor AMD-V is supported.
191	* @{ */
192
193	static DECLCALLBACK(int) hmR0DummyEnter(PVMCPUCC pVCpu)
194	{
195	RT_NOREF(pVCpu);
196	return VINF_SUCCESS;
197	}
198
199	static DECLCALLBACK(void) hmR0DummyThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPUCC pVCpu, bool fGlobalInit)
200	{
201	RT_NOREF(enmEvent, pVCpu, fGlobalInit);
202	}
203
204	static DECLCALLBACK(int) hmR0DummyEnableCpu(PHMPHYSCPU pHostCpu, PVMCC pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage,
205	bool fEnabledBySystem, PCSUPHWVIRTMSRS pHwvirtMsrs)
206	{
207	RT_NOREF(pHostCpu, pVM, pvCpuPage, HCPhysCpuPage, fEnabledBySystem, pHwvirtMsrs);
208	return VINF_SUCCESS;
209	}
210
211	static DECLCALLBACK(int) hmR0DummyDisableCpu(PHMPHYSCPU pHostCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
212	{
213	RT_NOREF(pHostCpu, pvCpuPage, HCPhysCpuPage);
214	return VINF_SUCCESS;
215	}
216
217	static DECLCALLBACK(int) hmR0DummyInitVM(PVMCC pVM)
218	{
219	RT_NOREF(pVM);
220	return VINF_SUCCESS;
221	}
222
223	static DECLCALLBACK(int) hmR0DummyTermVM(PVMCC pVM)
224	{
225	RT_NOREF(pVM);
226	return VINF_SUCCESS;
227	}
228
229	static DECLCALLBACK(int) hmR0DummySetupVM(PVMCC pVM)
230	{
231	RT_NOREF(pVM);
232	return VINF_SUCCESS;
233	}
234
235	static DECLCALLBACK(int) hmR0DummyAssertionCallback(PVMCPUCC pVCpu)
236	{
237	RT_NOREF(pVCpu);
238	return VINF_SUCCESS;
239	}
240
241	static DECLCALLBACK(VBOXSTRICTRC) hmR0DummyRunGuestCode(PVMCPUCC pVCpu)
242	{
243	RT_NOREF(pVCpu);
244	return VERR_NOT_SUPPORTED;
245	}
246
247	static DECLCALLBACK(int) hmR0DummyExportHostState(PVMCPUCC pVCpu)
248	{
249	RT_NOREF(pVCpu);
250	return VINF_SUCCESS;
251	}
252
253	/** Dummy ops. */
254	static HMR0VTABLE const g_HmR0OpsDummy =
255	{
256	/* .pfnEnterSession = */ hmR0DummyEnter,
257	/* .pfnThreadCtxCallback = */ hmR0DummyThreadCtxCallback,
258	/* .pfnAssertionCallback = */ hmR0DummyAssertionCallback,
259	/* .pfnExportHostState = */ hmR0DummyExportHostState,
260	/* .pfnRunGuestCode = */ hmR0DummyRunGuestCode,
261	/* .pfnEnableCpu = */ hmR0DummyEnableCpu,
262	/* .pfnDisableCpu = */ hmR0DummyDisableCpu,
263	/* .pfnInitVM = */ hmR0DummyInitVM,
264	/* .pfnTermVM = */ hmR0DummyTermVM,
265	/* .pfnSetupVM = */ hmR0DummySetupVM,
266	};
267
268	/** @} */
269
270
271	/**
272	* Initializes a first return code structure.
273	*
274	* @param pFirstRc The structure to init.
275	*/
276	static void hmR0FirstRcInit(PHMR0FIRSTRC pFirstRc)
277	{
278	pFirstRc->rc = VINF_SUCCESS;
279	pFirstRc->idCpu = NIL_RTCPUID;
280	}
281
282
283	/**
284	* Try set the status code (success ignored).
285	*
286	* @param pFirstRc The first return code structure.
287	* @param rc The status code.
288	*/
289	static void hmR0FirstRcSetStatus(PHMR0FIRSTRC pFirstRc, int rc)
290	{
291	if ( RT_FAILURE(rc)
292	&& ASMAtomicCmpXchgS32(&pFirstRc->rc, rc, VINF_SUCCESS))
293	pFirstRc->idCpu = RTMpCpuId();
294	}
295
296
297	/**
298	* Get the status code of a first return code structure.
299	*
300	* @returns The status code; VINF_SUCCESS or error status, no informational or
301	* warning errors.
302	* @param pFirstRc The first return code structure.
303	*/
304	static int hmR0FirstRcGetStatus(PHMR0FIRSTRC pFirstRc)
305	{
306	return pFirstRc->rc;
307	}
308
309
310	#ifdef VBOX_STRICT
311	# ifndef DEBUG_bird
312	/**
313	* Get the CPU ID on which the failure status code was reported.
314	*
315	* @returns The CPU ID, NIL_RTCPUID if no failure was reported.
316	* @param pFirstRc The first return code structure.
317	*/
318	static RTCPUID hmR0FirstRcGetCpuId(PHMR0FIRSTRC pFirstRc)
319	{
320	return pFirstRc->idCpu;
321	}
322	# endif
323	#endif /* VBOX_STRICT */
324
325
326
327	/**
328	* Worker function used by hmR0PowerCallback() and HMR0Init() to initalize VT-x
329	* on a CPU.
330	*
331	* @param idCpu The identifier for the CPU the function is called on.
332	* @param pvUser1 Pointer to the first RC structure.
333	* @param pvUser2 Ignored.
334	*/
335	static DECLCALLBACK(void) hmR0InitIntelCpu(RTCPUID idCpu, void pvUser1, void pvUser2)
336	{
337	PHMR0FIRSTRC pFirstRc = (PHMR0FIRSTRC)pvUser1;
338	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
339	Assert(idCpu == (RTCPUID)RTMpCpuIdToSetIndex(idCpu)); /** @todo fix idCpu == index assumption (rainy day) */
340	NOREF(idCpu); NOREF(pvUser2);
341
342	int rc = SUPR0GetVmxUsability(NULL /* pfIsSmxModeAmbiguous */);
343	hmR0FirstRcSetStatus(pFirstRc, rc);
344	}
345
346
347	/**
348	* Intel specific initialization code.
349	*
350	* @returns VBox status code (will only fail if out of memory).
351	*/
352	static int hmR0InitIntel(void)
353	{
354	/* Read this MSR now as it may be useful for error reporting when initializing VT-x fails. */
355	g_HmMsrs.u.vmx.u64FeatCtrl = ASMRdMsr(MSR_IA32_FEATURE_CONTROL);
356
357	/*
358	* First try use native kernel API for controlling VT-x.
359	* (This is only supported by some Mac OS X kernels atm.)
360	*/
361	int rc;
362	g_rcHmInit = rc = SUPR0EnableVTx(true /* fEnable */);
363	g_fHmVmxUsingSUPR0EnableVTx = rc != VERR_NOT_SUPPORTED;
364	if (g_fHmVmxUsingSUPR0EnableVTx)
365	{
366	AssertLogRelMsg(rc == VINF_SUCCESS \|\| rc == VERR_VMX_IN_VMX_ROOT_MODE \|\| rc == VERR_VMX_NO_VMX, ("%Rrc\n", rc));
367	if (RT_SUCCESS(rc))
368	{
369	g_fHmVmxSupported = true;
370	rc = SUPR0EnableVTx(false /* fEnable */);
371	AssertLogRelRC(rc);
372	rc = VINF_SUCCESS;
373	}
374	}
375	else
376	{
377	HMR0FIRSTRC FirstRc;
378	hmR0FirstRcInit(&FirstRc);
379	g_rcHmInit = rc = RTMpOnAll(hmR0InitIntelCpu, &FirstRc, NULL);
380	if (RT_SUCCESS(rc))
381	g_rcHmInit = rc = hmR0FirstRcGetStatus(&FirstRc);
382	}
383
384	if (RT_SUCCESS(rc))
385	{
386	/* Read CR4 and EFER for logging/diagnostic purposes. */
387	g_uHmVmxHostCr4 = ASMGetCR4();
388	g_uHmVmxHostMsrEfer = ASMRdMsr(MSR_K6_EFER);
389
390	/* Get VMX MSRs (and feature control MSR) for determining VMX features we can ultimately use. */
391	SUPR0GetHwvirtMsrs(&g_HmMsrs, SUPVTCAPS_VT_X, false /* fForce */);
392
393	/*
394	* Nested KVM workaround: Intel SDM section 34.15.5 describes that
395	* MSR_IA32_SMM_MONITOR_CTL depends on bit 49 of MSR_IA32_VMX_BASIC while
396	* table 35-2 says that this MSR is available if either VMX or SMX is supported.
397	*/
398	uint64_t const uVmxBasicMsr = g_HmMsrs.u.vmx.u64Basic;
399	if (RT_BF_GET(uVmxBasicMsr, VMX_BF_BASIC_DUAL_MON))
400	g_uHmVmxHostSmmMonitorCtl = ASMRdMsr(MSR_IA32_SMM_MONITOR_CTL);
401
402	/* Initialize VPID - 16 bits ASID. */
403	g_uHmMaxAsid = 0x10000; /* exclusive */
404
405	/*
406	* If the host OS has not enabled VT-x for us, try enter VMX root mode
407	* to really verify if VT-x is usable.
408	*/
409	if (!g_fHmVmxUsingSUPR0EnableVTx)
410	{
411	/* Allocate a temporary VMXON region. */
412	RTR0MEMOBJ hScatchMemObj;
413	rc = RTR0MemObjAllocCont(&hScatchMemObj, HOST_PAGE_SIZE, false /* fExecutable */);
414	if (RT_FAILURE(rc))
415	{
416	LogRel(("hmR0InitIntel: RTR0MemObjAllocCont(,HOST_PAGE_SIZE,false) -> %Rrc\n", rc));
417	return rc;
418	}
419	void *pvScatchPage = RTR0MemObjAddress(hScatchMemObj);
420	RTHCPHYS const HCPhysScratchPage = RTR0MemObjGetPagePhysAddr(hScatchMemObj, 0);
421	RT_BZERO(pvScatchPage, HOST_PAGE_SIZE);
422
423	/* Set revision dword at the beginning of the VMXON structure. */
424	(uint32_t )pvScatchPage = RT_BF_GET(uVmxBasicMsr, VMX_BF_BASIC_VMCS_ID);
425
426	/* Make sure we don't get rescheduled to another CPU during this probe. */
427	RTCCUINTREG const fEFlags = ASMIntDisableFlags();
428
429	/* Enable CR4.VMXE if it isn't already set. */
430	RTCCUINTREG const uOldCr4 = SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
431
432	/*
433	* The only way of checking if we're in VMX root mode or not is to try and enter it.
434	* There is no instruction or control bit that tells us if we're in VMX root mode.
435	* Therefore, try and enter VMX root mode here.
436	*/
437	rc = VMXEnable(HCPhysScratchPage);
438	if (RT_SUCCESS(rc))
439	{
440	g_fHmVmxSupported = true;
441	VMXDisable();
442	}
443	else
444	{
445	/*
446	* KVM leaves the CPU in VMX root mode. Not only is this not allowed,
447	* it will crash the host when we enter raw mode, because:
448	*
449	* (a) clearing X86_CR4_VMXE in CR4 causes a #GP (we no longer modify
450	* this bit), and
451	* (b) turning off paging causes a #GP (unavoidable when switching
452	* from long to 32 bits mode or 32 bits to PAE).
453	*
454	* They should fix their code, but until they do we simply refuse to run.
455	*/
456	g_rcHmInit = VERR_VMX_IN_VMX_ROOT_MODE;
457	Assert(g_fHmVmxSupported == false);
458	}
459
460	/* Restore CR4.VMXE if it wasn't set prior to us setting it above. */
461	if (!(uOldCr4 & X86_CR4_VMXE))
462	SUPR0ChangeCR4(0 /* fOrMask */, ~(uint64_t)X86_CR4_VMXE);
463
464	/* Restore interrupts. */
465	ASMSetFlags(fEFlags);
466
467	RTR0MemObjFree(hScatchMemObj, false);
468	}
469
470	if (g_fHmVmxSupported)
471	{
472	rc = VMXR0GlobalInit();
473	if (RT_SUCCESS(rc))
474	{
475	/*
476	* Install the VT-x methods.
477	*/
478	g_HmR0Ops = g_HmR0OpsVmx;
479
480	/*
481	* Check for the VMX-Preemption Timer and adjust for the "VMX-Preemption
482	* Timer Does Not Count Down at the Rate Specified" CPU erratum.
483	*/
484	if (g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_PREEMPT_TIMER)
485	{
486	g_fHmVmxUsePreemptTimer = true;
487	g_cHmVmxPreemptTimerShift = RT_BF_GET(g_HmMsrs.u.vmx.u64Misc, VMX_BF_MISC_PREEMPT_TIMER_TSC);
488	if (HMIsSubjectToVmxPreemptTimerErratum())
489	g_cHmVmxPreemptTimerShift = 0; /* This is about right most of the time here. */
490	}
491	else
492	g_fHmVmxUsePreemptTimer = false;
493
494	/*
495	* Check for EFER swapping support.
496	*/
497	g_fHmVmxSupportsVmcsEfer = (g_HmMsrs.u.vmx.EntryCtls.n.allowed1 & VMX_ENTRY_CTLS_LOAD_EFER_MSR)
498	&& (g_HmMsrs.u.vmx.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_LOAD_EFER_MSR)
499	&& (g_HmMsrs.u.vmx.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_SAVE_EFER_MSR);
500	}
501	else
502	{
503	g_rcHmInit = rc;
504	g_fHmVmxSupported = false;
505	}
506	}
507	}
508	#ifdef LOG_ENABLED
509	else
510	SUPR0Printf("hmR0InitIntelCpu failed with rc=%Rrc\n", g_rcHmInit);
511	#endif
512	return VINF_SUCCESS;
513	}
514
515
516	/**
517	* Worker function used by hmR0PowerCallback() and HMR0Init() to initalize AMD-V
518	* on a CPU.
519	*
520	* @param idCpu The identifier for the CPU the function is called on.
521	* @param pvUser1 Pointer to the first RC structure.
522	* @param pvUser2 Ignored.
523	*/
524	static DECLCALLBACK(void) hmR0InitAmdCpu(RTCPUID idCpu, void pvUser1, void pvUser2)
525	{
526	PHMR0FIRSTRC pFirstRc = (PHMR0FIRSTRC)pvUser1;
527	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
528	Assert(idCpu == (RTCPUID)RTMpCpuIdToSetIndex(idCpu)); /** @todo fix idCpu == index assumption (rainy day) */
529	NOREF(idCpu); NOREF(pvUser2);
530
531	int rc = SUPR0GetSvmUsability(true /* fInitSvm */);
532	hmR0FirstRcSetStatus(pFirstRc, rc);
533	}
534
535
536	/**
537	* AMD-specific initialization code.
538	*
539	* @returns VBox status code (will only fail if out of memory).
540	*/
541	static int hmR0InitAmd(void)
542	{
543	/* Call the global AMD-V initialization routine (should only fail in out-of-memory situations). */
544	int rc = SVMR0GlobalInit();
545	if (RT_SUCCESS(rc))
546	{
547	/*
548	* Install the AMD-V methods.
549	*/
550	g_HmR0Ops = g_HmR0OpsSvm;
551
552	/* Query AMD features. */
553	uint32_t u32Dummy;
554	ASMCpuId(0x8000000a, &g_uHmSvmRev, &g_uHmMaxAsid, &u32Dummy, &g_fHmSvmFeatures);
555
556	/*
557	* We need to check if AMD-V has been properly initialized on all CPUs.
558	* Some BIOSes might do a poor job.
559	*/
560	HMR0FIRSTRC FirstRc;
561	hmR0FirstRcInit(&FirstRc);
562	rc = RTMpOnAll(hmR0InitAmdCpu, &FirstRc, NULL);
563	AssertRC(rc);
564	if (RT_SUCCESS(rc))
565	rc = hmR0FirstRcGetStatus(&FirstRc);
566	#ifndef DEBUG_bird
567	AssertMsg(rc == VINF_SUCCESS \|\| rc == VERR_SVM_IN_USE,
568	("hmR0InitAmdCpu failed for cpu %d with rc=%Rrc\n", hmR0FirstRcGetCpuId(&FirstRc), rc));
569	#endif
570	if (RT_SUCCESS(rc))
571	{
572	SUPR0GetHwvirtMsrs(&g_HmMsrs, SUPVTCAPS_AMD_V, false /* fForce */);
573	g_fHmSvmSupported = true;
574	}
575	else
576	{
577	g_rcHmInit = rc;
578	if (rc == VERR_SVM_DISABLED \|\| rc == VERR_SVM_IN_USE)
579	rc = VINF_SUCCESS; /* Don't fail if AMD-V is disabled or in use. */
580	}
581	}
582	else
583	g_rcHmInit = rc;
584	return rc;
585	}
586
587
588	/**
589	* Does global Ring-0 HM initialization (at module init).
590	*
591	* @returns VBox status code.
592	*/
593	VMMR0_INT_DECL(int) HMR0Init(void)
594	{
595	/*
596	* Initialize the globals.
597	*/
598	g_fHmEnabled = false;
599	for (unsigned i = 0; i < RT_ELEMENTS(g_aHmCpuInfo); i++)
600	{
601	g_aHmCpuInfo[i].idCpu = NIL_RTCPUID;
602	g_aHmCpuInfo[i].hMemObj = NIL_RTR0MEMOBJ;
603	g_aHmCpuInfo[i].HCPhysMemObj = NIL_RTHCPHYS;
604	g_aHmCpuInfo[i].pvMemObj = NULL;
605	#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
606	g_aHmCpuInfo[i].n.svm.hNstGstMsrpm = NIL_RTR0MEMOBJ;
607	g_aHmCpuInfo[i].n.svm.HCPhysNstGstMsrpm = NIL_RTHCPHYS;
608	g_aHmCpuInfo[i].n.svm.pvNstGstMsrpm = NULL;
609	#endif
610	}
611
612	/* Fill in all callbacks with placeholders. */
613	g_HmR0Ops = g_HmR0OpsDummy;
614
615	/* Default is global VT-x/AMD-V init. */
616	g_fHmGlobalInit = true;
617
618	g_fHmVmxSupported = false;
619	g_fHmSvmSupported = false;
620	g_uHmMaxAsid = 0;
621
622	/*
623	* Get host kernel features that HM might need to know in order
624	* to co-operate and function properly with the host OS (e.g. SMAP).
625	*/
626	g_fHmHostKernelFeatures = SUPR0GetKernelFeatures();
627
628	/*
629	* Make sure aCpuInfo is big enough for all the CPUs on this system.
630	*/
631	if (RTMpGetArraySize() > RT_ELEMENTS(g_aHmCpuInfo))
632	{
633	LogRel(("HM: Too many real CPUs/cores/threads - %u, max %u\n", RTMpGetArraySize(), RT_ELEMENTS(g_aHmCpuInfo)));
634	return VERR_TOO_MANY_CPUS;
635	}
636
637	/*
638	* Check for VT-x or AMD-V support.
639	* Return failure only in out-of-memory situations.
640	*/
641	uint32_t fCaps = 0;
642	int rc = SUPR0GetVTSupport(&fCaps);
643	if (RT_SUCCESS(rc))
644	{
645	if (fCaps & SUPVTCAPS_VT_X)
646	rc = hmR0InitIntel();
647	else
648	{
649	Assert(fCaps & SUPVTCAPS_AMD_V);
650	rc = hmR0InitAmd();
651	}
652	if (RT_SUCCESS(rc))
653	{
654	/*
655	* Register notification callbacks that we can use to disable/enable CPUs
656	* when brought offline/online or suspending/resuming.
657	*/
658	if (!g_fHmVmxUsingSUPR0EnableVTx)
659	{
660	rc = RTMpNotificationRegister(hmR0MpEventCallback, NULL);
661	if (RT_SUCCESS(rc))
662	{
663	rc = RTPowerNotificationRegister(hmR0PowerCallback, NULL);
664	if (RT_FAILURE(rc))
665	RTMpNotificationDeregister(hmR0MpEventCallback, NULL);
666	}
667	if (RT_FAILURE(rc))
668	{
669	/* There shouldn't be any per-cpu allocations at this point,
670	so just have to call SVMR0GlobalTerm and VMXR0GlobalTerm. */
671	if (fCaps & SUPVTCAPS_VT_X)
672	VMXR0GlobalTerm();
673	else
674	SVMR0GlobalTerm();
675	g_HmR0Ops = g_HmR0OpsDummy;
676	g_rcHmInit = rc;
677	g_fHmSvmSupported = false;
678	g_fHmVmxSupported = false;
679	}
680	}
681	}
682	}
683	else
684	{
685	g_rcHmInit = rc;
686	rc = VINF_SUCCESS; /* We return success here because module init shall not fail if HM fails to initialize. */
687	}
688	return rc;
689	}
690
691
692	/**
693	* Does global Ring-0 HM termination (at module termination).
694	*
695	* @returns VBox status code (ignored).
696	*/
697	VMMR0_INT_DECL(int) HMR0Term(void)
698	{
699	int rc;
700	if ( g_fHmVmxSupported
701	&& g_fHmVmxUsingSUPR0EnableVTx)
702	{
703	/*
704	* Simple if the host OS manages VT-x.
705	*/
706	Assert(g_fHmGlobalInit);
707
708	if (g_fHmVmxCalledSUPR0EnableVTx)
709	{
710	rc = SUPR0EnableVTx(false /* fEnable */);
711	g_fHmVmxCalledSUPR0EnableVTx = false;
712	}
713	else
714	rc = VINF_SUCCESS;
715
716	for (unsigned iCpu = 0; iCpu < RT_ELEMENTS(g_aHmCpuInfo); iCpu++)
717	{
718	g_aHmCpuInfo[iCpu].fConfigured = false;
719	Assert(g_aHmCpuInfo[iCpu].hMemObj == NIL_RTR0MEMOBJ);
720	}
721	}
722	else
723	{
724	Assert(!g_fHmVmxSupported \|\| !g_fHmVmxUsingSUPR0EnableVTx);
725
726	/* Doesn't really matter if this fails. */
727	RTMpNotificationDeregister(hmR0MpEventCallback, NULL);
728	RTPowerNotificationDeregister(hmR0PowerCallback, NULL);
729	rc = VINF_SUCCESS;
730
731	/*
732	* Disable VT-x/AMD-V on all CPUs if we enabled it before.
733	*/
734	if (g_fHmGlobalInit)
735	{
736	HMR0FIRSTRC FirstRc;
737	hmR0FirstRcInit(&FirstRc);
738	rc = RTMpOnAll(hmR0DisableCpuCallback, NULL /* pvUser 1 */, &FirstRc);
739	Assert(RT_SUCCESS(rc) \|\| rc == VERR_NOT_SUPPORTED);
740	if (RT_SUCCESS(rc))
741	rc = hmR0FirstRcGetStatus(&FirstRc);
742	}
743
744	/*
745	* Free the per-cpu pages used for VT-x and AMD-V.
746	*/
747	for (unsigned i = 0; i < RT_ELEMENTS(g_aHmCpuInfo); i++)
748	{
749	if (g_aHmCpuInfo[i].hMemObj != NIL_RTR0MEMOBJ)
750	{
751	RTR0MemObjFree(g_aHmCpuInfo[i].hMemObj, false);
752	g_aHmCpuInfo[i].hMemObj = NIL_RTR0MEMOBJ;
753	g_aHmCpuInfo[i].HCPhysMemObj = NIL_RTHCPHYS;
754	g_aHmCpuInfo[i].pvMemObj = NULL;
755	}
756	#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
757	if (g_aHmCpuInfo[i].n.svm.hNstGstMsrpm != NIL_RTR0MEMOBJ)
758	{
759	RTR0MemObjFree(g_aHmCpuInfo[i].n.svm.hNstGstMsrpm, false);
760	g_aHmCpuInfo[i].n.svm.hNstGstMsrpm = NIL_RTR0MEMOBJ;
761	g_aHmCpuInfo[i].n.svm.HCPhysNstGstMsrpm = NIL_RTHCPHYS;
762	g_aHmCpuInfo[i].n.svm.pvNstGstMsrpm = NULL;
763	}
764	#endif
765	}
766	}
767
768	/** @todo This needs cleaning up. There's no matching
769	* hmR0TermIntel()/hmR0TermAmd() and all the VT-x/AMD-V specific bits
770	* should move into their respective modules. */
771	/* Finally, call global VT-x/AMD-V termination. */
772	if (g_fHmVmxSupported)
773	VMXR0GlobalTerm();
774	else if (g_fHmSvmSupported)
775	SVMR0GlobalTerm();
776
777	return rc;
778	}
779
780
781	/**
782	* Enable VT-x or AMD-V on the current CPU
783	*
784	* @returns VBox status code.
785	* @param pVM The cross context VM structure. Can be NULL.
786	* @param idCpu The identifier for the CPU the function is called on.
787	*
788	* @remarks Maybe called with interrupts disabled!
789	*/
790	static int hmR0EnableCpu(PVMCC pVM, RTCPUID idCpu)
791	{
792	PHMPHYSCPU pHostCpu = &g_aHmCpuInfo[idCpu];
793
794	Assert(idCpu == (RTCPUID)RTMpCpuIdToSetIndex(idCpu)); /** @todo fix idCpu == index assumption (rainy day) */
795	Assert(idCpu < RT_ELEMENTS(g_aHmCpuInfo));
796	Assert(!pHostCpu->fConfigured);
797	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
798
799	pHostCpu->idCpu = idCpu;
800	/* Do NOT reset cTlbFlushes here, see @bugref{6255}. */
801
802	int rc;
803	if ( g_fHmVmxSupported
804	&& g_fHmVmxUsingSUPR0EnableVTx)
805	rc = g_HmR0Ops.pfnEnableCpu(pHostCpu, pVM, NULL /* pvCpuPage */, NIL_RTHCPHYS, true, &g_HmMsrs);
806	else
807	{
808	AssertLogRelMsgReturn(pHostCpu->hMemObj != NIL_RTR0MEMOBJ, ("hmR0EnableCpu failed idCpu=%u.\n", idCpu), VERR_HM_IPE_1);
809	rc = g_HmR0Ops.pfnEnableCpu(pHostCpu, pVM, pHostCpu->pvMemObj, pHostCpu->HCPhysMemObj, false, &g_HmMsrs);
810	}
811	if (RT_SUCCESS(rc))
812	pHostCpu->fConfigured = true;
813	return rc;
814	}
815
816
817	/**
818	* Worker function passed to RTMpOnAll() that is to be called on all CPUs.
819	*
820	* @param idCpu The identifier for the CPU the function is called on.
821	* @param pvUser1 Opaque pointer to the VM (can be NULL!).
822	* @param pvUser2 The 2nd user argument.
823	*/
824	static DECLCALLBACK(void) hmR0EnableCpuCallback(RTCPUID idCpu, void pvUser1, void pvUser2)
825	{
826	PVMCC pVM = (PVMCC)pvUser1; /* can be NULL! */
827	PHMR0FIRSTRC pFirstRc = (PHMR0FIRSTRC)pvUser2;
828	AssertReturnVoid(g_fHmGlobalInit);
829	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
830	hmR0FirstRcSetStatus(pFirstRc, hmR0EnableCpu(pVM, idCpu));
831	}
832
833
834	/**
835	* RTOnce callback employed by HMR0EnableAllCpus.
836	*
837	* @returns VBox status code.
838	* @param pvUser Pointer to the VM.
839	*/
840	static DECLCALLBACK(int32_t) hmR0EnableAllCpuOnce(void *pvUser)
841	{
842	PVMCC pVM = (PVMCC)pvUser;
843
844	/*
845	* Indicate that we've initialized.
846	*
847	* Note! There is a potential race between this function and the suspend
848	* notification. Kind of unlikely though, so ignored for now.
849	*/
850	AssertReturn(!g_fHmEnabled, VERR_HM_ALREADY_ENABLED_IPE);
851	ASMAtomicWriteBool(&g_fHmEnabled, true);
852
853	/*
854	* The global init variable is set by the first VM.
855	*/
856	g_fHmGlobalInit = pVM->hm.s.fGlobalInit;
857
858	#ifdef VBOX_STRICT
859	for (unsigned i = 0; i < RT_ELEMENTS(g_aHmCpuInfo); i++)
860	{
861	Assert(g_aHmCpuInfo[i].hMemObj == NIL_RTR0MEMOBJ);
862	Assert(g_aHmCpuInfo[i].HCPhysMemObj == NIL_RTHCPHYS);
863	Assert(g_aHmCpuInfo[i].pvMemObj == NULL);
864	Assert(!g_aHmCpuInfo[i].fConfigured);
865	Assert(!g_aHmCpuInfo[i].cTlbFlushes);
866	Assert(!g_aHmCpuInfo[i].uCurrentAsid);
867	# ifdef VBOX_WITH_NESTED_HWVIRT_SVM
868	Assert(g_aHmCpuInfo[i].n.svm.hNstGstMsrpm == NIL_RTR0MEMOBJ);
869	Assert(g_aHmCpuInfo[i].n.svm.HCPhysNstGstMsrpm == NIL_RTHCPHYS);
870	Assert(g_aHmCpuInfo[i].n.svm.pvNstGstMsrpm == NULL);
871	# endif
872	}
873	#endif
874
875	int rc;
876	if ( g_fHmVmxSupported
877	&& g_fHmVmxUsingSUPR0EnableVTx)
878	{
879	/*
880	* Global VT-x initialization API (only darwin for now).
881	*/
882	rc = SUPR0EnableVTx(true /* fEnable */);
883	if (RT_SUCCESS(rc))
884	{
885	g_fHmVmxCalledSUPR0EnableVTx = true;
886	/* If the host provides a VT-x init API, then we'll rely on that for global init. */
887	g_fHmGlobalInit = pVM->hm.s.fGlobalInit = true;
888	}
889	else
890	AssertMsgFailed(("hmR0EnableAllCpuOnce/SUPR0EnableVTx: rc=%Rrc\n", rc));
891	}
892	else
893	{
894	/*
895	* We're doing the job ourselves.
896	*/
897	/* Allocate one page per cpu for the global VT-x and AMD-V pages */
898	for (unsigned i = 0; i < RT_ELEMENTS(g_aHmCpuInfo); i++)
899	{
900	Assert(g_aHmCpuInfo[i].hMemObj == NIL_RTR0MEMOBJ);
901	#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
902	Assert(g_aHmCpuInfo[i].n.svm.hNstGstMsrpm == NIL_RTR0MEMOBJ);
903	#endif
904	if (RTMpIsCpuPossible(RTMpCpuIdFromSetIndex(i)))
905	{
906	/** @todo NUMA */
907	rc = RTR0MemObjAllocCont(&g_aHmCpuInfo[i].hMemObj, HOST_PAGE_SIZE, false /* executable R0 mapping */);
908	AssertLogRelRCReturn(rc, rc);
909
910	g_aHmCpuInfo[i].HCPhysMemObj = RTR0MemObjGetPagePhysAddr(g_aHmCpuInfo[i].hMemObj, 0);
911	Assert(g_aHmCpuInfo[i].HCPhysMemObj != NIL_RTHCPHYS);
912	Assert(!(g_aHmCpuInfo[i].HCPhysMemObj & HOST_PAGE_OFFSET_MASK));
913
914	g_aHmCpuInfo[i].pvMemObj = RTR0MemObjAddress(g_aHmCpuInfo[i].hMemObj);
915	AssertPtr(g_aHmCpuInfo[i].pvMemObj);
916	RT_BZERO(g_aHmCpuInfo[i].pvMemObj, HOST_PAGE_SIZE);
917
918	#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
919	rc = RTR0MemObjAllocCont(&g_aHmCpuInfo[i].n.svm.hNstGstMsrpm, SVM_MSRPM_PAGES << X86_PAGE_4K_SHIFT,
920	false /* executable R0 mapping */);
921	AssertLogRelRCReturn(rc, rc);
922
923	g_aHmCpuInfo[i].n.svm.HCPhysNstGstMsrpm = RTR0MemObjGetPagePhysAddr(g_aHmCpuInfo[i].n.svm.hNstGstMsrpm, 0);
924	Assert(g_aHmCpuInfo[i].n.svm.HCPhysNstGstMsrpm != NIL_RTHCPHYS);
925	Assert(!(g_aHmCpuInfo[i].n.svm.HCPhysNstGstMsrpm & HOST_PAGE_OFFSET_MASK));
926
927	g_aHmCpuInfo[i].n.svm.pvNstGstMsrpm = RTR0MemObjAddress(g_aHmCpuInfo[i].n.svm.hNstGstMsrpm);
928	AssertPtr(g_aHmCpuInfo[i].n.svm.pvNstGstMsrpm);
929	ASMMemFill32(g_aHmCpuInfo[i].n.svm.pvNstGstMsrpm, SVM_MSRPM_PAGES << X86_PAGE_4K_SHIFT, UINT32_C(0xffffffff));
930	#endif
931	}
932	}
933
934	rc = VINF_SUCCESS;
935	}
936
937	if ( RT_SUCCESS(rc)
938	&& g_fHmGlobalInit)
939	{
940	/* First time, so initialize each cpu/core. */
941	HMR0FIRSTRC FirstRc;
942	hmR0FirstRcInit(&FirstRc);
943	rc = RTMpOnAll(hmR0EnableCpuCallback, (void *)pVM, &FirstRc);
944	if (RT_SUCCESS(rc))
945	rc = hmR0FirstRcGetStatus(&FirstRc);
946	}
947
948	return rc;
949	}
950
951
952	/**
953	* Sets up HM on all cpus.
954	*
955	* @returns VBox status code.
956	* @param pVM The cross context VM structure.
957	*/
958	VMMR0_INT_DECL(int) HMR0EnableAllCpus(PVMCC pVM)
959	{
960	/* Make sure we don't touch HM after we've disabled HM in preparation of a suspend. */
961	if (ASMAtomicReadBool(&g_fHmSuspended))
962	return VERR_HM_SUSPEND_PENDING;
963
964	return RTOnce(&g_HmEnableAllCpusOnce, hmR0EnableAllCpuOnce, pVM);
965	}
966
967
968	/**
969	* Disable VT-x or AMD-V on the current CPU.
970	*
971	* @returns VBox status code.
972	* @param idCpu The identifier for the CPU this function is called on.
973	*
974	* @remarks Must be called with preemption disabled.
975	*/
976	static int hmR0DisableCpu(RTCPUID idCpu)
977	{
978	PHMPHYSCPU pHostCpu = &g_aHmCpuInfo[idCpu];
979
980	Assert(!g_fHmVmxSupported \|\| !g_fHmVmxUsingSUPR0EnableVTx);
981	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
982	Assert(idCpu == (RTCPUID)RTMpCpuIdToSetIndex(idCpu)); /** @todo fix idCpu == index assumption (rainy day) */
983	Assert(idCpu < RT_ELEMENTS(g_aHmCpuInfo));
984	Assert(!pHostCpu->fConfigured \|\| pHostCpu->hMemObj != NIL_RTR0MEMOBJ);
985	AssertRelease(idCpu == RTMpCpuId());
986
987	if (pHostCpu->hMemObj == NIL_RTR0MEMOBJ)
988	return pHostCpu->fConfigured ? VERR_NO_MEMORY : VINF_SUCCESS /* not initialized. */;
989	AssertPtr(pHostCpu->pvMemObj);
990	Assert(pHostCpu->HCPhysMemObj != NIL_RTHCPHYS);
991
992	int rc;
993	if (pHostCpu->fConfigured)
994	{
995	rc = g_HmR0Ops.pfnDisableCpu(pHostCpu, pHostCpu->pvMemObj, pHostCpu->HCPhysMemObj);
996	AssertRCReturn(rc, rc);
997
998	pHostCpu->fConfigured = false;
999	pHostCpu->idCpu = NIL_RTCPUID;
1000	}
1001	else
1002	rc = VINF_SUCCESS; /* nothing to do */
1003	return rc;
1004	}
1005
1006
1007	/**
1008	* Worker function passed to RTMpOnAll() that is to be called on the target
1009	* CPUs.
1010	*
1011	* @param idCpu The identifier for the CPU the function is called on.
1012	* @param pvUser1 The 1st user argument.
1013	* @param pvUser2 Opaque pointer to the FirstRc.
1014	*/
1015	static DECLCALLBACK(void) hmR0DisableCpuCallback(RTCPUID idCpu, void pvUser1, void pvUser2)
1016	{
1017	PHMR0FIRSTRC pFirstRc = (PHMR0FIRSTRC)pvUser2; NOREF(pvUser1);
1018	AssertReturnVoid(g_fHmGlobalInit);
1019	hmR0FirstRcSetStatus(pFirstRc, hmR0DisableCpu(idCpu));
1020	}
1021
1022
1023	/**
1024	* Worker function passed to RTMpOnSpecific() that is to be called on the target
1025	* CPU.
1026	*
1027	* @param idCpu The identifier for the CPU the function is called on.
1028	* @param pvUser1 Null, not used.
1029	* @param pvUser2 Null, not used.
1030	*/
1031	static DECLCALLBACK(void) hmR0DisableCpuOnSpecificCallback(RTCPUID idCpu, void pvUser1, void pvUser2)
1032	{
1033	NOREF(pvUser1);
1034	NOREF(pvUser2);
1035	hmR0DisableCpu(idCpu);
1036	}
1037
1038
1039	/**
1040	* Callback function invoked when a cpu goes online or offline.
1041	*
1042	* @param enmEvent The Mp event.
1043	* @param idCpu The identifier for the CPU the function is called on.
1044	* @param pvData Opaque data (PVMCC pointer).
1045	*/
1046	static DECLCALLBACK(void) hmR0MpEventCallback(RTMPEVENT enmEvent, RTCPUID idCpu, void *pvData)
1047	{
1048	NOREF(pvData);
1049	Assert(!g_fHmVmxSupported \|\| !g_fHmVmxUsingSUPR0EnableVTx);
1050
1051	/*
1052	* We only care about uninitializing a CPU that is going offline. When a
1053	* CPU comes online, the initialization is done lazily in HMR0Enter().
1054	*/
1055	switch (enmEvent)
1056	{
1057	case RTMPEVENT_OFFLINE:
1058	{
1059	RTTHREADPREEMPTSTATE PreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
1060	RTThreadPreemptDisable(&PreemptState);
1061	if (idCpu == RTMpCpuId())
1062	{
1063	int rc = hmR0DisableCpu(idCpu);
1064	AssertRC(rc);
1065	RTThreadPreemptRestore(&PreemptState);
1066	}
1067	else
1068	{
1069	RTThreadPreemptRestore(&PreemptState);
1070	RTMpOnSpecific(idCpu, hmR0DisableCpuOnSpecificCallback, NULL /* pvUser1 /, NULL / pvUser2 */);
1071	}
1072	break;
1073	}
1074
1075	default:
1076	break;
1077	}
1078	}
1079
1080
1081	/**
1082	* Called whenever a system power state change occurs.
1083	*
1084	* @param enmEvent The Power event.
1085	* @param pvUser User argument.
1086	*/
1087	static DECLCALLBACK(void) hmR0PowerCallback(RTPOWEREVENT enmEvent, void *pvUser)
1088	{
1089	NOREF(pvUser);
1090	Assert(!g_fHmVmxSupported \|\| !g_fHmVmxUsingSUPR0EnableVTx);
1091
1092	#ifdef LOG_ENABLED
1093	if (enmEvent == RTPOWEREVENT_SUSPEND)
1094	SUPR0Printf("hmR0PowerCallback RTPOWEREVENT_SUSPEND\n");
1095	else
1096	SUPR0Printf("hmR0PowerCallback RTPOWEREVENT_RESUME\n");
1097	#endif
1098
1099	if (enmEvent == RTPOWEREVENT_SUSPEND)
1100	ASMAtomicWriteBool(&g_fHmSuspended, true);
1101
1102	if (g_fHmEnabled)
1103	{
1104	int rc;
1105	HMR0FIRSTRC FirstRc;
1106	hmR0FirstRcInit(&FirstRc);
1107
1108	if (enmEvent == RTPOWEREVENT_SUSPEND)
1109	{
1110	if (g_fHmGlobalInit)
1111	{
1112	/* Turn off VT-x or AMD-V on all CPUs. */
1113	rc = RTMpOnAll(hmR0DisableCpuCallback, NULL /* pvUser 1 */, &FirstRc);
1114	Assert(RT_SUCCESS(rc) \|\| rc == VERR_NOT_SUPPORTED);
1115	}
1116	/* else nothing to do here for the local init case */
1117	}
1118	else
1119	{
1120	/* Reinit the CPUs from scratch as the suspend state might have
1121	messed with the MSRs. (lousy BIOSes as usual) */
1122	if (g_fHmVmxSupported)
1123	rc = RTMpOnAll(hmR0InitIntelCpu, &FirstRc, NULL);
1124	else
1125	rc = RTMpOnAll(hmR0InitAmdCpu, &FirstRc, NULL);
1126	Assert(RT_SUCCESS(rc) \|\| rc == VERR_NOT_SUPPORTED);
1127	if (RT_SUCCESS(rc))
1128	rc = hmR0FirstRcGetStatus(&FirstRc);
1129	#ifdef LOG_ENABLED
1130	if (RT_FAILURE(rc))
1131	SUPR0Printf("hmR0PowerCallback hmR0InitXxxCpu failed with %Rc\n", rc);
1132	#endif
1133	if (g_fHmGlobalInit)
1134	{
1135	/* Turn VT-x or AMD-V back on on all CPUs. */
1136	rc = RTMpOnAll(hmR0EnableCpuCallback, NULL /* pVM /, &FirstRc / output ignored */);
1137	Assert(RT_SUCCESS(rc) \|\| rc == VERR_NOT_SUPPORTED);
1138	}
1139	/* else nothing to do here for the local init case */
1140	}
1141	}
1142
1143	if (enmEvent == RTPOWEREVENT_RESUME)
1144	ASMAtomicWriteBool(&g_fHmSuspended, false);
1145	}
1146
1147
1148	/**
1149	* Does ring-0 per-VM HM initialization.
1150	*
1151	* This will call the CPU specific init. routine which may initialize and allocate
1152	* resources for virtual CPUs.
1153	*
1154	* @returns VBox status code.
1155	* @param pVM The cross context VM structure.
1156	*
1157	* @remarks This is called after HMR3Init(), see vmR3CreateU() and
1158	* vmR3InitRing3().
1159	*/
1160	VMMR0_INT_DECL(int) HMR0InitVM(PVMCC pVM)
1161	{
1162	AssertCompile(sizeof(pVM->hm.s) <= sizeof(pVM->hm.padding));
1163	AssertCompile(sizeof(pVM->hmr0.s) <= sizeof(pVM->hmr0.padding));
1164	AssertCompile(sizeof(pVM->aCpus[0].hm.s) <= sizeof(pVM->aCpus[0].hm.padding));
1165	AssertCompile(sizeof(pVM->aCpus[0].hmr0.s) <= sizeof(pVM->aCpus[0].hmr0.padding));
1166	AssertReturn(pVM, VERR_INVALID_PARAMETER);
1167
1168	/* Make sure we don't touch HM after we've disabled HM in preparation of a suspend. */
1169	if (ASMAtomicReadBool(&g_fHmSuspended))
1170	return VERR_HM_SUSPEND_PENDING;
1171
1172	/*
1173	* Copy globals to the VM structure.
1174	*/
1175	Assert(!(pVM->hm.s.vmx.fSupported && pVM->hm.s.svm.fSupported));
1176	if (pVM->hm.s.vmx.fSupported)
1177	{
1178	pVM->hmr0.s.vmx.fUsePreemptTimer = pVM->hm.s.vmx.fUsePreemptTimerCfg && g_fHmVmxUsePreemptTimer;
1179	pVM->hm.s.vmx.fUsePreemptTimerCfg = pVM->hmr0.s.vmx.fUsePreemptTimer;
1180	pVM->hm.s.vmx.cPreemptTimerShift = g_cHmVmxPreemptTimerShift;
1181	pVM->hm.s.ForR3.vmx.u64HostCr4 = g_uHmVmxHostCr4;
1182	pVM->hm.s.ForR3.vmx.u64HostMsrEfer = g_uHmVmxHostMsrEfer;
1183	pVM->hm.s.ForR3.vmx.u64HostSmmMonitorCtl = g_uHmVmxHostSmmMonitorCtl;
1184	pVM->hm.s.ForR3.vmx.u64HostFeatCtrl = g_HmMsrs.u.vmx.u64FeatCtrl;
1185	HMGetVmxMsrsFromHwvirtMsrs(&g_HmMsrs, &pVM->hm.s.ForR3.vmx.Msrs);
1186	/* If you need to tweak host MSRs for testing VMX R0 code, do it here. */
1187
1188	/* Enable VPID if supported and configured. */
1189	if (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VPID)
1190	pVM->hm.s.ForR3.vmx.fVpid = pVM->hmr0.s.vmx.fVpid = pVM->hm.s.vmx.fAllowVpid; /* Can be overridden by CFGM in HMR3Init(). */
1191
1192	/* Use VMCS shadowing if supported. */
1193	pVM->hmr0.s.vmx.fUseVmcsShadowing = pVM->cpum.ro.GuestFeatures.fVmx
1194	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VMCS_SHADOWING);
1195	pVM->hm.s.ForR3.vmx.fUseVmcsShadowing = pVM->hmr0.s.vmx.fUseVmcsShadowing;
1196
1197	/* Use the VMCS controls for swapping the EFER MSR if supported. */
1198	pVM->hm.s.ForR3.vmx.fSupportsVmcsEfer = g_fHmVmxSupportsVmcsEfer;
1199
1200	#if 0
1201	/* Enable APIC register virtualization and virtual-interrupt delivery if supported. */
1202	if ( (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_APIC_REG_VIRT)
1203	&& (g_HmMsrs.u.vmx.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_VIRT_INTR_DELIVERY))
1204	pVM->hm.s.fVirtApicRegs = true;
1205
1206	/* Enable posted-interrupt processing if supported. */
1207	/** @todo Add and query IPRT API for host OS support for posted-interrupt IPI
1208	* here. */
1209	if ( (g_HmMsrs.u.vmx.PinCtls.n.allowed1 & VMX_PIN_CTLS_POSTED_INT)
1210	&& (g_HmMsrs.u.vmx.ExitCtls.n.allowed1 & VMX_EXIT_CTLS_ACK_EXT_INT))
1211	pVM->hm.s.fPostedIntrs = true;
1212	#endif
1213	}
1214	else if (pVM->hm.s.svm.fSupported)
1215	{
1216	pVM->hm.s.ForR3.svm.u32Rev = g_uHmSvmRev;
1217	pVM->hm.s.ForR3.svm.fFeatures = g_fHmSvmFeatures;
1218	pVM->hm.s.ForR3.svm.u64MsrHwcr = g_HmMsrs.u.svm.u64MsrHwcr;
1219	/* If you need to tweak host MSRs for testing SVM R0 code, do it here. */
1220	}
1221	pVM->hm.s.ForR3.rcInit = g_rcHmInit;
1222	pVM->hm.s.ForR3.uMaxAsid = g_uHmMaxAsid;
1223
1224	/*
1225	* Set default maximum inner loops in ring-0 before returning to ring-3.
1226	* Can be overriden using CFGM.
1227	*/
1228	uint32_t cMaxResumeLoops = pVM->hm.s.cMaxResumeLoopsCfg;
1229	if (!cMaxResumeLoops)
1230	{
1231	cMaxResumeLoops = 1024;
1232	if (RTThreadPreemptIsPendingTrusty())
1233	cMaxResumeLoops = 8192;
1234	}
1235	else if (cMaxResumeLoops > 16384)
1236	cMaxResumeLoops = 16384;
1237	else if (cMaxResumeLoops < 32)
1238	cMaxResumeLoops = 32;
1239	pVM->hm.s.cMaxResumeLoopsCfg = pVM->hmr0.s.cMaxResumeLoops = cMaxResumeLoops;
1240
1241	/*
1242	* Initialize some per-VCPU fields.
1243	*/
1244	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
1245	{
1246	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
1247	pVCpu->hmr0.s.idEnteredCpu = NIL_RTCPUID;
1248	pVCpu->hmr0.s.idLastCpu = NIL_RTCPUID;
1249
1250	/* We'll aways increment this the first time (host uses ASID 0). */
1251	AssertReturn(!pVCpu->hmr0.s.uCurrentAsid, VERR_HM_IPE_3);
1252	}
1253
1254	/*
1255	* Configure defences against spectre and other CPU bugs.
1256	*/
1257	uint32_t fWorldSwitcher = 0;
1258	uint32_t cLastStdLeaf = ASMCpuId_EAX(0);
1259	if (cLastStdLeaf >= 0x00000007 && RTX86IsValidStdRange(cLastStdLeaf))
1260	{
1261	uint32_t uEdx = 0;
1262	ASMCpuIdExSlow(0x00000007, 0, 0, 0, NULL, NULL, NULL, &uEdx);
1263
1264	if (uEdx & X86_CPUID_STEXT_FEATURE_EDX_IBRS_IBPB)
1265	{
1266	if (pVM->hm.s.fIbpbOnVmExit)
1267	fWorldSwitcher \|= HM_WSF_IBPB_EXIT;
1268	if (pVM->hm.s.fIbpbOnVmEntry)
1269	fWorldSwitcher \|= HM_WSF_IBPB_ENTRY;
1270	}
1271	if (uEdx & X86_CPUID_STEXT_FEATURE_EDX_FLUSH_CMD)
1272	{
1273	if (pVM->hm.s.fL1dFlushOnVmEntry)
1274	fWorldSwitcher \|= HM_WSF_L1D_ENTRY;
1275	else if (pVM->hm.s.fL1dFlushOnSched)
1276	fWorldSwitcher \|= HM_WSF_L1D_SCHED;
1277	}
1278	if (uEdx & X86_CPUID_STEXT_FEATURE_EDX_MD_CLEAR)
1279	{
1280	if (pVM->hm.s.fMdsClearOnVmEntry)
1281	fWorldSwitcher \|= HM_WSF_MDS_ENTRY;
1282	else if (pVM->hm.s.fMdsClearOnSched)
1283	fWorldSwitcher \|= HM_WSF_MDS_SCHED;
1284	}
1285	}
1286	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
1287	{
1288	PVMCPUCC pVCpu = VMCC_GET_CPU(pVM, idCpu);
1289	pVCpu->hmr0.s.fWorldSwitcher = fWorldSwitcher;
1290	}
1291	pVM->hm.s.ForR3.fWorldSwitcher = fWorldSwitcher;
1292
1293
1294	/*
1295	* Call the hardware specific initialization method.
1296	*/
1297	return g_HmR0Ops.pfnInitVM(pVM);
1298	}
1299
1300
1301	/**
1302	* Does ring-0 per VM HM termination.
1303	*
1304	* @returns VBox status code.
1305	* @param pVM The cross context VM structure.
1306	*/
1307	VMMR0_INT_DECL(int) HMR0TermVM(PVMCC pVM)
1308	{
1309	Log(("HMR0TermVM: %p\n", pVM));
1310	AssertReturn(pVM, VERR_INVALID_PARAMETER);
1311
1312	/*
1313	* Call the hardware specific method.
1314	*
1315	* Note! We might be preparing for a suspend, so the pfnTermVM() functions should probably not
1316	* mess with VT-x/AMD-V features on the CPU, currently all they do is free memory so this is safe.
1317	*/
1318	return g_HmR0Ops.pfnTermVM(pVM);
1319	}
1320
1321
1322	/**
1323	* Sets up a VT-x or AMD-V session.
1324	*
1325	* This is mostly about setting up the hardware VM state.
1326	*
1327	* @returns VBox status code.
1328	* @param pVM The cross context VM structure.
1329	*/
1330	VMMR0_INT_DECL(int) HMR0SetupVM(PVMCC pVM)
1331	{
1332	Log(("HMR0SetupVM: %p\n", pVM));
1333	AssertReturn(pVM, VERR_INVALID_PARAMETER);
1334
1335	/* Make sure we don't touch HM after we've disabled HM in preparation of a suspend. */
1336	AssertReturn(!ASMAtomicReadBool(&g_fHmSuspended), VERR_HM_SUSPEND_PENDING);
1337
1338	/* On first entry we'll sync everything. */
1339	VMCC_FOR_EACH_VMCPU_STMT(pVM, pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_ALL_GUEST);
1340
1341	/*
1342	* Call the hardware specific setup VM method. This requires the CPU to be
1343	* enabled for AMD-V/VT-x and preemption to be prevented.
1344	*/
1345	RTTHREADPREEMPTSTATE PreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
1346	RTThreadPreemptDisable(&PreemptState);
1347	RTCPUID const idCpu = RTMpCpuId();
1348
1349	/* Enable VT-x or AMD-V if local init is required. */
1350	int rc;
1351	if (!g_fHmGlobalInit)
1352	{
1353	Assert(!g_fHmVmxSupported \|\| !g_fHmVmxUsingSUPR0EnableVTx);
1354	rc = hmR0EnableCpu(pVM, idCpu);
1355	if (RT_FAILURE(rc))
1356	{
1357	RTThreadPreemptRestore(&PreemptState);
1358	return rc;
1359	}
1360	}
1361
1362	/* Setup VT-x or AMD-V. */
1363	rc = g_HmR0Ops.pfnSetupVM(pVM);
1364
1365	/* Disable VT-x or AMD-V if local init was done before. */
1366	if (!g_fHmGlobalInit)
1367	{
1368	Assert(!g_fHmVmxSupported \|\| !g_fHmVmxUsingSUPR0EnableVTx);
1369	int rc2 = hmR0DisableCpu(idCpu);
1370	AssertRC(rc2);
1371	}
1372
1373	RTThreadPreemptRestore(&PreemptState);
1374	return rc;
1375	}
1376
1377
1378	/**
1379	* Notification callback before an assertion longjump and guru mediation.
1380	*
1381	* @returns VBox status code.
1382	* @param pVCpu The cross context virtual CPU structure.
1383	* @param pvUser User argument, currently unused, NULL.
1384	*/
1385	static DECLCALLBACK(int) hmR0AssertionCallback(PVMCPUCC pVCpu, void *pvUser)
1386	{
1387	RT_NOREF(pvUser);
1388	Assert(pVCpu);
1389	Assert(g_HmR0Ops.pfnAssertionCallback);
1390	return g_HmR0Ops.pfnAssertionCallback(pVCpu);
1391	}
1392
1393
1394	/**
1395	* Turns on HM on the CPU if necessary and initializes the bare minimum state
1396	* required for entering HM context.
1397	*
1398	* @returns VBox status code.
1399	* @param pVCpu The cross context virtual CPU structure.
1400	*
1401	* @remarks No-long-jump zone!!!
1402	*/
1403	VMMR0_INT_DECL(int) hmR0EnterCpu(PVMCPUCC pVCpu)
1404	{
1405	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1406
1407	int rc = VINF_SUCCESS;
1408	RTCPUID const idCpu = RTMpCpuId();
1409	PHMPHYSCPU pHostCpu = &g_aHmCpuInfo[idCpu];
1410	AssertPtr(pHostCpu);
1411
1412	/* Enable VT-x or AMD-V if local init is required, or enable if it's a freshly onlined CPU. */
1413	if (!pHostCpu->fConfigured)
1414	rc = hmR0EnableCpu(pVCpu->CTX_SUFF(pVM), idCpu);
1415
1416	/* Register a callback to fire prior to performing a longjmp to ring-3 so HM can disable VT-x/AMD-V if needed. */
1417	VMMR0AssertionSetNotification(pVCpu, hmR0AssertionCallback, NULL /pvUser/);
1418
1419	/* Reload host-state (back from ring-3/migrated CPUs) and shared guest/host bits. */
1420	if (g_fHmVmxSupported)
1421	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE;
1422	else
1423	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_SVM_HOST_GUEST_SHARED_STATE;
1424
1425	Assert(pHostCpu->idCpu == idCpu && pHostCpu->idCpu != NIL_RTCPUID);
1426	pVCpu->hmr0.s.idEnteredCpu = idCpu;
1427	return rc;
1428	}
1429
1430
1431	/**
1432	* Enters the VT-x or AMD-V session.
1433	*
1434	* @returns VBox status code.
1435	* @param pVCpu The cross context virtual CPU structure.
1436	*
1437	* @remarks This is called with preemption disabled.
1438	*/
1439	VMMR0_INT_DECL(int) HMR0Enter(PVMCPUCC pVCpu)
1440	{
1441	/* Make sure we can't enter a session after we've disabled HM in preparation of a suspend. */
1442	AssertReturn(!ASMAtomicReadBool(&g_fHmSuspended), VERR_HM_SUSPEND_PENDING);
1443	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1444
1445	/* Load the bare minimum state required for entering HM. */
1446	int rc = hmR0EnterCpu(pVCpu);
1447	if (RT_SUCCESS(rc))
1448	{
1449	if (g_fHmVmxSupported)
1450	Assert( (pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE))
1451	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_VMX_HOST_GUEST_SHARED_STATE));
1452	else
1453	Assert( (pVCpu->hm.s.fCtxChanged & (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_SVM_HOST_GUEST_SHARED_STATE))
1454	== (HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_SVM_HOST_GUEST_SHARED_STATE));
1455
1456	/* Keep track of the CPU owning the VMCS for debugging scheduling weirdness and ring-3 calls. */
1457	rc = g_HmR0Ops.pfnEnterSession(pVCpu);
1458	AssertMsgRCReturnStmt(rc, ("rc=%Rrc pVCpu=%p\n", rc, pVCpu), pVCpu->hmr0.s.idEnteredCpu = NIL_RTCPUID, rc);
1459
1460	/* Exports the host-state as we may be resuming code after a longjmp and quite
1461	possibly now be scheduled on a different CPU. */
1462	rc = g_HmR0Ops.pfnExportHostState(pVCpu);
1463	AssertMsgRCReturnStmt(rc, ("rc=%Rrc pVCpu=%p\n", rc, pVCpu), pVCpu->hmr0.s.idEnteredCpu = NIL_RTCPUID, rc);
1464	}
1465	return rc;
1466	}
1467
1468
1469	/**
1470	* Deinitializes the bare minimum state used for HM context and if necessary
1471	* disable HM on the CPU.
1472	*
1473	* @returns VBox status code.
1474	* @param pVCpu The cross context virtual CPU structure.
1475	*
1476	* @remarks No-long-jump zone!!!
1477	*/
1478	VMMR0_INT_DECL(int) HMR0LeaveCpu(PVMCPUCC pVCpu)
1479	{
1480	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1481	VMCPU_ASSERT_EMT_RETURN(pVCpu, VERR_HM_WRONG_CPU);
1482
1483	RTCPUID const idCpu = RTMpCpuId();
1484	PCHMPHYSCPU pHostCpu = &g_aHmCpuInfo[idCpu];
1485
1486	if ( !g_fHmGlobalInit
1487	&& pHostCpu->fConfigured)
1488	{
1489	int rc = hmR0DisableCpu(idCpu);
1490	AssertRCReturn(rc, rc);
1491	Assert(!pHostCpu->fConfigured);
1492	Assert(pHostCpu->idCpu == NIL_RTCPUID);
1493
1494	/* For obtaining a non-zero ASID/VPID on next re-entry. */
1495	pVCpu->hmr0.s.idLastCpu = NIL_RTCPUID;
1496	}
1497
1498	/* Clear it while leaving HM context, hmPokeCpuForTlbFlush() relies on this. */
1499	pVCpu->hmr0.s.idEnteredCpu = NIL_RTCPUID;
1500
1501	/* De-register the longjmp-to-ring 3 callback now that we have reliquished hardware resources. */
1502	VMMR0AssertionRemoveNotification(pVCpu);
1503	return VINF_SUCCESS;
1504	}
1505
1506
1507	/**
1508	* Thread-context hook for HM.
1509	*
1510	* This is used together with RTThreadCtxHookCreate() on platforms which
1511	* supports it, and directly from VMMR0EmtPrepareForBlocking() and
1512	* VMMR0EmtResumeAfterBlocking() on platforms which don't.
1513	*
1514	* @param enmEvent The thread-context event.
1515	* @param pvUser Opaque pointer to the VMCPU.
1516	*/
1517	VMMR0_INT_DECL(void) HMR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, void *pvUser)
1518	{
1519	PVMCPUCC pVCpu = (PVMCPUCC)pvUser;
1520	Assert(pVCpu);
1521	Assert(g_HmR0Ops.pfnThreadCtxCallback);
1522
1523	g_HmR0Ops.pfnThreadCtxCallback(enmEvent, pVCpu, g_fHmGlobalInit);
1524	}
1525
1526
1527	/**
1528	* Runs guest code in a hardware accelerated VM.
1529	*
1530	* @returns Strict VBox status code. (VBOXSTRICTRC isn't used because it's
1531	* called from setjmp assembly.)
1532	* @param pVM The cross context VM structure.
1533	* @param pVCpu The cross context virtual CPU structure.
1534	*
1535	* @remarks Can be called with preemption enabled if thread-context hooks are
1536	* used!!!
1537	*/
1538	VMMR0_INT_DECL(int) HMR0RunGuestCode(PVMCC pVM, PVMCPUCC pVCpu)
1539	{
1540	RT_NOREF(pVM);
1541
1542	#ifdef VBOX_STRICT
1543	/* With thread-context hooks we would be running this code with preemption enabled. */
1544	if (!RTThreadPreemptIsEnabled(NIL_RTTHREAD))
1545	{
1546	PCHMPHYSCPU pHostCpu = &g_aHmCpuInfo[RTMpCpuId()];
1547	Assert(!VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL));
1548	Assert(pHostCpu->fConfigured);
1549	AssertReturn(!ASMAtomicReadBool(&g_fHmSuspended), VERR_HM_SUSPEND_PENDING);
1550	}
1551	#endif
1552
1553	VBOXSTRICTRC rcStrict = g_HmR0Ops.pfnRunGuestCode(pVCpu);
1554	return VBOXSTRICTRC_VAL(rcStrict);
1555	}
1556
1557
1558	/**
1559	* Notification from CPUM that it has unloaded the guest FPU/SSE/AVX state from
1560	* the host CPU and that guest access to it must be intercepted.
1561	*
1562	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
1563	*/
1564	VMMR0_INT_DECL(void) HMR0NotifyCpumUnloadedGuestFpuState(PVMCPUCC pVCpu)
1565	{
1566	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_GUEST_CR0);
1567	}
1568
1569
1570	/**
1571	* Notification from CPUM that it has modified the host CR0 (because of FPU).
1572	*
1573	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
1574	*/
1575	VMMR0_INT_DECL(void) HMR0NotifyCpumModifiedHostCr0(PVMCPUCC pVCpu)
1576	{
1577	ASMAtomicUoOrU64(&pVCpu->hm.s.fCtxChanged, HM_CHANGED_HOST_CONTEXT);
1578	}
1579
1580
1581	/**
1582	* Returns suspend status of the host.
1583	*
1584	* @returns Suspend pending or not.
1585	*/
1586	VMMR0_INT_DECL(bool) HMR0SuspendPending(void)
1587	{
1588	return ASMAtomicReadBool(&g_fHmSuspended);
1589	}
1590
1591
1592	/**
1593	* Invalidates a guest page from the host TLB.
1594	*
1595	* @param pVCpu The cross context virtual CPU structure.
1596	* @param GCVirt Page to invalidate.
1597	*/
1598	VMMR0_INT_DECL(int) HMR0InvalidatePage(PVMCPUCC pVCpu, RTGCPTR GCVirt)
1599	{
1600	PVMCC pVM = pVCpu->CTX_SUFF(pVM);
1601	if (pVM->hm.s.vmx.fSupported)
1602	return VMXR0InvalidatePage(pVCpu, GCVirt);
1603	return SVMR0InvalidatePage(pVCpu, GCVirt);
1604	}
1605
1606
1607	/**
1608	* Returns the cpu structure for the current cpu.
1609	* Keep in mind that there is no guarantee it will stay the same (long jumps to ring 3!!!).
1610	*
1611	* @returns The cpu structure pointer.
1612	*/
1613	VMMR0_INT_DECL(PHMPHYSCPU) hmR0GetCurrentCpu(void)
1614	{
1615	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1616	RTCPUID const idCpu = RTMpCpuId();
1617	Assert(idCpu < RT_ELEMENTS(g_aHmCpuInfo));
1618	return &g_aHmCpuInfo[idCpu];
1619	}
1620
1621
1622	/**
1623	* Interface for importing state on demand (used by IEM).
1624	*
1625	* @returns VBox status code.
1626	* @param pVCpu The cross context CPU structure.
1627	* @param fWhat What to import, CPUMCTX_EXTRN_XXX.
1628	*/
1629	VMMR0_INT_DECL(int) HMR0ImportStateOnDemand(PVMCPUCC pVCpu, uint64_t fWhat)
1630	{
1631	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported)
1632	return VMXR0ImportStateOnDemand(pVCpu, fWhat);
1633	return SVMR0ImportStateOnDemand(pVCpu, fWhat);
1634	}
1635
1636
1637	/**
1638	* Gets HM VM-exit auxiliary information.
1639	*
1640	* @returns VBox status code.
1641	* @param pVCpu The cross context CPU structure.
1642	* @param pHmExitAux Where to store the auxiliary info.
1643	* @param fWhat What to get, see HMVMX_READ_XXX. This is ignored/unused
1644	* on AMD-V.
1645	*
1646	* @remarks Currently this works only when executing a nested-guest using
1647	* hardware-assisted execution as it's where the auxiliary information is
1648	* required outside of HM. In the future we can make this available while
1649	* executing a regular (non-nested) guest if necessary.
1650	*/
1651	VMMR0_INT_DECL(int) HMR0GetExitAuxInfo(PVMCPUCC pVCpu, PHMEXITAUX pHmExitAux, uint32_t fWhat)
1652	{
1653	Assert(pHmExitAux);
1654	Assert(!(fWhat & ~HMVMX_READ_VALID_MASK));
1655	if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported)
1656	return VMXR0GetExitAuxInfo(pVCpu, &pHmExitAux->Vmx, fWhat);
1657	return SVMR0GetExitAuxInfo(pVCpu, &pHmExitAux->Svm);
1658	}
1659
1660
1661	#ifdef VBOX_STRICT
1662
1663	/**
1664	* Dumps a descriptor.
1665	*
1666	* @param pDesc Descriptor to dump.
1667	* @param Sel The selector.
1668	* @param pszSel The name of the selector.
1669	*/
1670	VMMR0_INT_DECL(void) hmR0DumpDescriptor(PCX86DESCHC pDesc, RTSEL Sel, const char *pszSel)
1671	{
1672	/*
1673	* Make variable description string.
1674	*/
1675	static struct
1676	{
1677	unsigned cch;
1678	const char *psz;
1679	} const s_aTypes[32] =
1680	{
1681	# define STRENTRY(str) { sizeof(str) - 1, str }
1682
1683	/* system */
1684	# if HC_ARCH_BITS == 64
1685	STRENTRY("Reserved0 "), /* 0x00 */
1686	STRENTRY("Reserved1 "), /* 0x01 */
1687	STRENTRY("LDT "), /* 0x02 */
1688	STRENTRY("Reserved3 "), /* 0x03 */
1689	STRENTRY("Reserved4 "), /* 0x04 */
1690	STRENTRY("Reserved5 "), /* 0x05 */
1691	STRENTRY("Reserved6 "), /* 0x06 */
1692	STRENTRY("Reserved7 "), /* 0x07 */
1693	STRENTRY("Reserved8 "), /* 0x08 */
1694	STRENTRY("TSS64Avail "), /* 0x09 */
1695	STRENTRY("ReservedA "), /* 0x0a */
1696	STRENTRY("TSS64Busy "), /* 0x0b */
1697	STRENTRY("Call64 "), /* 0x0c */
1698	STRENTRY("ReservedD "), /* 0x0d */
1699	STRENTRY("Int64 "), /* 0x0e */
1700	STRENTRY("Trap64 "), /* 0x0f */
1701	# else
1702	STRENTRY("Reserved0 "), /* 0x00 */
1703	STRENTRY("TSS16Avail "), /* 0x01 */
1704	STRENTRY("LDT "), /* 0x02 */
1705	STRENTRY("TSS16Busy "), /* 0x03 */
1706	STRENTRY("Call16 "), /* 0x04 */
1707	STRENTRY("Task "), /* 0x05 */
1708	STRENTRY("Int16 "), /* 0x06 */
1709	STRENTRY("Trap16 "), /* 0x07 */
1710	STRENTRY("Reserved8 "), /* 0x08 */
1711	STRENTRY("TSS32Avail "), /* 0x09 */
1712	STRENTRY("ReservedA "), /* 0x0a */
1713	STRENTRY("TSS32Busy "), /* 0x0b */
1714	STRENTRY("Call32 "), /* 0x0c */
1715	STRENTRY("ReservedD "), /* 0x0d */
1716	STRENTRY("Int32 "), /* 0x0e */
1717	STRENTRY("Trap32 "), /* 0x0f */
1718	# endif
1719	/* non system */
1720	STRENTRY("DataRO "), /* 0x10 */
1721	STRENTRY("DataRO Accessed "), /* 0x11 */
1722	STRENTRY("DataRW "), /* 0x12 */
1723	STRENTRY("DataRW Accessed "), /* 0x13 */
1724	STRENTRY("DataDownRO "), /* 0x14 */
1725	STRENTRY("DataDownRO Accessed "), /* 0x15 */
1726	STRENTRY("DataDownRW "), /* 0x16 */
1727	STRENTRY("DataDownRW Accessed "), /* 0x17 */
1728	STRENTRY("CodeEO "), /* 0x18 */
1729	STRENTRY("CodeEO Accessed "), /* 0x19 */
1730	STRENTRY("CodeER "), /* 0x1a */
1731	STRENTRY("CodeER Accessed "), /* 0x1b */
1732	STRENTRY("CodeConfEO "), /* 0x1c */
1733	STRENTRY("CodeConfEO Accessed "), /* 0x1d */
1734	STRENTRY("CodeConfER "), /* 0x1e */
1735	STRENTRY("CodeConfER Accessed ") /* 0x1f */
1736	# undef SYSENTRY
1737	};
1738	# define ADD_STR(psz, pszAdd) do { strcpy(psz, pszAdd); psz += strlen(pszAdd); } while (0)
1739	char szMsg[128];
1740	char *psz = &szMsg[0];
1741	unsigned i = pDesc->Gen.u1DescType << 4 \| pDesc->Gen.u4Type;
1742	memcpy(psz, s_aTypes[i].psz, s_aTypes[i].cch);
1743	psz += s_aTypes[i].cch;
1744
1745	if (pDesc->Gen.u1Present)
1746	ADD_STR(psz, "Present ");
1747	else
1748	ADD_STR(psz, "Not-Present ");
1749	# if HC_ARCH_BITS == 64
1750	if (pDesc->Gen.u1Long)
1751	ADD_STR(psz, "64-bit ");
1752	else
1753	ADD_STR(psz, "Comp ");
1754	# else
1755	if (pDesc->Gen.u1Granularity)
1756	ADD_STR(psz, "Page ");
1757	if (pDesc->Gen.u1DefBig)
1758	ADD_STR(psz, "32-bit ");
1759	else
1760	ADD_STR(psz, "16-bit ");
1761	# endif
1762	# undef ADD_STR
1763	*psz = '\0';
1764
1765	/*
1766	* Limit and Base and format the output.
1767	*/
1768	#ifdef LOG_ENABLED
1769	uint32_t u32Limit = X86DESC_LIMIT_G(pDesc);
1770
1771	# if HC_ARCH_BITS == 64
1772	uint64_t const u64Base = X86DESC64_BASE(pDesc);
1773	Log((" %s { %#04x - %#RX64 %#RX64 - base=%#RX64 limit=%#08x dpl=%d } %s\n", pszSel,
1774	Sel, pDesc->au64[0], pDesc->au64[1], u64Base, u32Limit, pDesc->Gen.u2Dpl, szMsg));
1775	# else
1776	uint32_t const u32Base = X86DESC_BASE(pDesc);
1777	Log((" %s { %#04x - %#08x %#08x - base=%#08x limit=%#08x dpl=%d } %s\n", pszSel,
1778	Sel, pDesc->au32[0], pDesc->au32[1], u32Base, u32Limit, pDesc->Gen.u2Dpl, szMsg));
1779	# endif
1780	#else
1781	NOREF(Sel); NOREF(pszSel);
1782	#endif
1783	}
1784
1785
1786	/**
1787	* Formats a full register dump.
1788	*
1789	* @param pVCpu The cross context virtual CPU structure.
1790	* @param fFlags The dumping flags (HM_DUMP_REG_FLAGS_XXX).
1791	*/
1792	VMMR0_INT_DECL(void) hmR0DumpRegs(PVMCPUCC pVCpu, uint32_t fFlags)
1793	{
1794	/*
1795	* Format the flags.
1796	*/
1797	static struct
1798	{
1799	const char *pszSet;
1800	const char *pszClear;
1801	uint32_t fFlag;
1802	} const s_aFlags[] =
1803	{
1804	{ "vip", NULL, X86_EFL_VIP },
1805	{ "vif", NULL, X86_EFL_VIF },
1806	{ "ac", NULL, X86_EFL_AC },
1807	{ "vm", NULL, X86_EFL_VM },
1808	{ "rf", NULL, X86_EFL_RF },
1809	{ "nt", NULL, X86_EFL_NT },
1810	{ "ov", "nv", X86_EFL_OF },
1811	{ "dn", "up", X86_EFL_DF },
1812	{ "ei", "di", X86_EFL_IF },
1813	{ "tf", NULL, X86_EFL_TF },
1814	{ "nt", "pl", X86_EFL_SF },
1815	{ "nz", "zr", X86_EFL_ZF },
1816	{ "ac", "na", X86_EFL_AF },
1817	{ "po", "pe", X86_EFL_PF },
1818	{ "cy", "nc", X86_EFL_CF },
1819	};
1820	char szEFlags[80];
1821	char *psz = szEFlags;
1822	PCCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
1823	uint32_t uEFlags = pCtx->eflags.u32;
1824	for (unsigned i = 0; i < RT_ELEMENTS(s_aFlags); i++)
1825	{
1826	const char *pszAdd = s_aFlags[i].fFlag & uEFlags ? s_aFlags[i].pszSet : s_aFlags[i].pszClear;
1827	if (pszAdd)
1828	{
1829	strcpy(psz, pszAdd);
1830	psz += strlen(pszAdd);
1831	*psz++ = ' ';
1832	}
1833	}
1834	psz[-1] = '\0';
1835
1836	if (fFlags & HM_DUMP_REG_FLAGS_GPRS)
1837	{
1838	/*
1839	* Format the registers.
1840	*/
1841	if (CPUMIsGuestIn64BitCode(pVCpu))
1842	{
1843	Log(("rax=%016RX64 rbx=%016RX64 rcx=%016RX64 rdx=%016RX64\n"
1844	"rsi=%016RX64 rdi=%016RX64 r8 =%016RX64 r9 =%016RX64\n"
1845	"r10=%016RX64 r11=%016RX64 r12=%016RX64 r13=%016RX64\n"
1846	"r14=%016RX64 r15=%016RX64\n"
1847	"rip=%016RX64 rsp=%016RX64 rbp=%016RX64 iopl=%d %*s\n"
1848	"cs={%04x base=%016RX64 limit=%08x flags=%08x}\n"
1849	"ds={%04x base=%016RX64 limit=%08x flags=%08x}\n"
1850	"es={%04x base=%016RX64 limit=%08x flags=%08x}\n"
1851	"fs={%04x base=%016RX64 limit=%08x flags=%08x}\n"
1852	"gs={%04x base=%016RX64 limit=%08x flags=%08x}\n"
1853	"ss={%04x base=%016RX64 limit=%08x flags=%08x}\n"
1854	"cr0=%016RX64 cr2=%016RX64 cr3=%016RX64 cr4=%016RX64\n"
1855	"dr0=%016RX64 dr1=%016RX64 dr2=%016RX64 dr3=%016RX64\n"
1856	"dr4=%016RX64 dr5=%016RX64 dr6=%016RX64 dr7=%016RX64\n"
1857	"gdtr=%016RX64:%04x idtr=%016RX64:%04x eflags=%08x\n"
1858	"ldtr={%04x base=%08RX64 limit=%08x flags=%08x}\n"
1859	"tr ={%04x base=%08RX64 limit=%08x flags=%08x}\n"
1860	"SysEnter={cs=%04llx eip=%08llx esp=%08llx}\n"
1861	,
1862	pCtx->rax, pCtx->rbx, pCtx->rcx, pCtx->rdx, pCtx->rsi, pCtx->rdi,
1863	pCtx->r8, pCtx->r9, pCtx->r10, pCtx->r11, pCtx->r12, pCtx->r13,
1864	pCtx->r14, pCtx->r15,
1865	pCtx->rip, pCtx->rsp, pCtx->rbp, X86_EFL_GET_IOPL(uEFlags), 31, szEFlags,
1866	pCtx->cs.Sel, pCtx->cs.u64Base, pCtx->cs.u32Limit, pCtx->cs.Attr.u,
1867	pCtx->ds.Sel, pCtx->ds.u64Base, pCtx->ds.u32Limit, pCtx->ds.Attr.u,
1868	pCtx->es.Sel, pCtx->es.u64Base, pCtx->es.u32Limit, pCtx->es.Attr.u,
1869	pCtx->fs.Sel, pCtx->fs.u64Base, pCtx->fs.u32Limit, pCtx->fs.Attr.u,
1870	pCtx->gs.Sel, pCtx->gs.u64Base, pCtx->gs.u32Limit, pCtx->gs.Attr.u,
1871	pCtx->ss.Sel, pCtx->ss.u64Base, pCtx->ss.u32Limit, pCtx->ss.Attr.u,
1872	pCtx->cr0, pCtx->cr2, pCtx->cr3, pCtx->cr4,
1873	pCtx->dr[0], pCtx->dr[1], pCtx->dr[2], pCtx->dr[3],
1874	pCtx->dr[4], pCtx->dr[5], pCtx->dr[6], pCtx->dr[7],
1875	pCtx->gdtr.pGdt, pCtx->gdtr.cbGdt, pCtx->idtr.pIdt, pCtx->idtr.cbIdt, uEFlags,
1876	pCtx->ldtr.Sel, pCtx->ldtr.u64Base, pCtx->ldtr.u32Limit, pCtx->ldtr.Attr.u,
1877	pCtx->tr.Sel, pCtx->tr.u64Base, pCtx->tr.u32Limit, pCtx->tr.Attr.u,
1878	pCtx->SysEnter.cs, pCtx->SysEnter.eip, pCtx->SysEnter.esp));
1879	}
1880	else
1881	Log(("eax=%08x ebx=%08x ecx=%08x edx=%08x esi=%08x edi=%08x\n"
1882	"eip=%08x esp=%08x ebp=%08x iopl=%d %*s\n"
1883	"cs={%04x base=%016RX64 limit=%08x flags=%08x} dr0=%08RX64 dr1=%08RX64\n"
1884	"ds={%04x base=%016RX64 limit=%08x flags=%08x} dr2=%08RX64 dr3=%08RX64\n"
1885	"es={%04x base=%016RX64 limit=%08x flags=%08x} dr4=%08RX64 dr5=%08RX64\n"
1886	"fs={%04x base=%016RX64 limit=%08x flags=%08x} dr6=%08RX64 dr7=%08RX64\n"
1887	"gs={%04x base=%016RX64 limit=%08x flags=%08x} cr0=%08RX64 cr2=%08RX64\n"
1888	"ss={%04x base=%016RX64 limit=%08x flags=%08x} cr3=%08RX64 cr4=%08RX64\n"
1889	"gdtr=%016RX64:%04x idtr=%016RX64:%04x eflags=%08x\n"
1890	"ldtr={%04x base=%08RX64 limit=%08x flags=%08x}\n"
1891	"tr ={%04x base=%08RX64 limit=%08x flags=%08x}\n"
1892	"SysEnter={cs=%04llx eip=%08llx esp=%08llx}\n"
1893	,
1894	pCtx->eax, pCtx->ebx, pCtx->ecx, pCtx->edx, pCtx->esi, pCtx->edi,
1895	pCtx->eip, pCtx->esp, pCtx->ebp, X86_EFL_GET_IOPL(uEFlags), 31, szEFlags,
1896	pCtx->cs.Sel, pCtx->cs.u64Base, pCtx->cs.u32Limit, pCtx->cs.Attr.u, pCtx->dr[0], pCtx->dr[1],
1897	pCtx->ds.Sel, pCtx->ds.u64Base, pCtx->ds.u32Limit, pCtx->ds.Attr.u, pCtx->dr[2], pCtx->dr[3],
1898	pCtx->es.Sel, pCtx->es.u64Base, pCtx->es.u32Limit, pCtx->es.Attr.u, pCtx->dr[4], pCtx->dr[5],
1899	pCtx->fs.Sel, pCtx->fs.u64Base, pCtx->fs.u32Limit, pCtx->fs.Attr.u, pCtx->dr[6], pCtx->dr[7],
1900	pCtx->gs.Sel, pCtx->gs.u64Base, pCtx->gs.u32Limit, pCtx->gs.Attr.u, pCtx->cr0, pCtx->cr2,
1901	pCtx->ss.Sel, pCtx->ss.u64Base, pCtx->ss.u32Limit, pCtx->ss.Attr.u, pCtx->cr3, pCtx->cr4,
1902	pCtx->gdtr.pGdt, pCtx->gdtr.cbGdt, pCtx->idtr.pIdt, pCtx->idtr.cbIdt, uEFlags,
1903	pCtx->ldtr.Sel, pCtx->ldtr.u64Base, pCtx->ldtr.u32Limit, pCtx->ldtr.Attr.u,
1904	pCtx->tr.Sel, pCtx->tr.u64Base, pCtx->tr.u32Limit, pCtx->tr.Attr.u,
1905	pCtx->SysEnter.cs, pCtx->SysEnter.eip, pCtx->SysEnter.esp));
1906	}
1907
1908	if (fFlags & HM_DUMP_REG_FLAGS_FPU)
1909	{
1910	PCX86FXSTATE pFpuCtx = &pCtx->XState.x87;
1911	Log(("FPU:\n"
1912	"FCW=%04x FSW=%04x FTW=%02x\n"
1913	"FOP=%04x FPUIP=%08x CS=%04x Rsrvd1=%04x\n"
1914	"FPUDP=%04x DS=%04x Rsvrd2=%04x MXCSR=%08x MXCSR_MASK=%08x\n"
1915	,
1916	pFpuCtx->FCW, pFpuCtx->FSW, pFpuCtx->FTW,
1917	pFpuCtx->FOP, pFpuCtx->FPUIP, pFpuCtx->CS, pFpuCtx->Rsrvd1,
1918	pFpuCtx->FPUDP, pFpuCtx->DS, pFpuCtx->Rsrvd2,
1919	pFpuCtx->MXCSR, pFpuCtx->MXCSR_MASK));
1920	NOREF(pFpuCtx);
1921	}
1922
1923	if (fFlags & HM_DUMP_REG_FLAGS_MSRS)
1924	{
1925	Log(("MSR:\n"
1926	"EFER =%016RX64\n"
1927	"PAT =%016RX64\n"
1928	"STAR =%016RX64\n"
1929	"CSTAR =%016RX64\n"
1930	"LSTAR =%016RX64\n"
1931	"SFMASK =%016RX64\n"
1932	"KERNELGSBASE =%016RX64\n",
1933	pCtx->msrEFER,
1934	pCtx->msrPAT,
1935	pCtx->msrSTAR,
1936	pCtx->msrCSTAR,
1937	pCtx->msrLSTAR,
1938	pCtx->msrSFMASK,
1939	pCtx->msrKERNELGSBASE));
1940	}
1941	}
1942
1943	#endif /* VBOX_STRICT */
1944

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

source: vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp@ 95140

Download in other formats: