VirtualBox

source: vbox/trunk/src/VBox/VMM/VMMR3/HM.cpp@ 72555

Last change on this file since 72555 was 72555, checked in by vboxsync, 7 years ago

EM,NEM,HM: Started on generic exit history, only hooked up to NEM/win. bugref:9044
  • Property svn:eol-style set to native
  • Property svn:keywords set to Id Revision
File size: 169.8 KB
Line 
1/* $Id: HM.cpp 72555 2018-06-14 21:28:31Z vboxsync $ */
2/** @file
3 * HM - Intel/AMD VM Hardware Support Manager.
4 */
5
6/*
7 * Copyright (C) 2006-2017 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/** @page pg_hm HM - Hardware Assisted Virtualization Manager
19 *
20 * The HM manages guest execution using the VT-x and AMD-V CPU hardware
21 * extensions.
22 *
23 * {summary of what HM does}
24 *
25 * Hardware assisted virtualization manager was originally abbreviated HWACCM,
26 * however that was cumbersome to write and parse for such a central component,
27 * so it was shortened to HM when refactoring the code in the 4.3 development
28 * cycle.
29 *
30 * {add sections with more details}
31 *
32 * @sa @ref grp_hm
33 */
34
35
36/*********************************************************************************************************************************
37* Header Files *
38*********************************************************************************************************************************/
39#define LOG_GROUP LOG_GROUP_HM
40#include <VBox/vmm/cpum.h>
41#include <VBox/vmm/stam.h>
42#include <VBox/vmm/mm.h>
43#include <VBox/vmm/pdmapi.h>
44#include <VBox/vmm/pgm.h>
45#include <VBox/vmm/ssm.h>
46#include <VBox/vmm/trpm.h>
47#include <VBox/vmm/dbgf.h>
48#include <VBox/vmm/iom.h>
49#include <VBox/vmm/iem.h>
50#include <VBox/vmm/patm.h>
51#include <VBox/vmm/csam.h>
52#include <VBox/vmm/selm.h>
53#include <VBox/vmm/nem.h>
54#ifdef VBOX_WITH_REM
55# include <VBox/vmm/rem.h>
56#endif
57#include <VBox/vmm/hm_vmx.h>
58#include <VBox/vmm/hm_svm.h>
59#include "HMInternal.h"
60#include <VBox/vmm/vm.h>
61#include <VBox/vmm/uvm.h>
62#include <VBox/err.h>
63#include <VBox/param.h>
64
65#include <iprt/assert.h>
66#include <VBox/log.h>
67#include <iprt/asm.h>
68#include <iprt/asm-amd64-x86.h>
69#include <iprt/env.h>
70#include <iprt/thread.h>
71
72
73/*********************************************************************************************************************************
74* Global Variables *
75*********************************************************************************************************************************/
76#define EXIT_REASON(def, val, str) #def " - " #val " - " str
77#define EXIT_REASON_NIL() NULL
78/** Exit reason descriptions for VT-x, used to describe statistics. */
79static const char * const g_apszVTxExitReasons[MAX_EXITREASON_STAT] =
80{
81 EXIT_REASON(VMX_EXIT_XCPT_OR_NMI , 0, "Exception or non-maskable interrupt (NMI)."),
82 EXIT_REASON(VMX_EXIT_EXT_INT , 1, "External interrupt."),
83 EXIT_REASON(VMX_EXIT_TRIPLE_FAULT , 2, "Triple fault."),
84 EXIT_REASON(VMX_EXIT_INIT_SIGNAL , 3, "INIT signal."),
85 EXIT_REASON(VMX_EXIT_SIPI , 4, "Start-up IPI (SIPI)."),
86 EXIT_REASON(VMX_EXIT_IO_SMI_IRQ , 5, "I/O system-management interrupt (SMI)."),
87 EXIT_REASON(VMX_EXIT_SMI_IRQ , 6, "Other SMI."),
88 EXIT_REASON(VMX_EXIT_INT_WINDOW , 7, "Interrupt window."),
89 EXIT_REASON(VMX_EXIT_NMI_WINDOW , 8, "NMI window."),
90 EXIT_REASON(VMX_EXIT_TASK_SWITCH , 9, "Task switch."),
91 EXIT_REASON(VMX_EXIT_CPUID , 10, "CPUID instruction."),
92 EXIT_REASON(VMX_EXIT_GETSEC , 11, "GETSEC instrunction."),
93 EXIT_REASON(VMX_EXIT_HLT , 12, "HLT instruction."),
94 EXIT_REASON(VMX_EXIT_INVD , 13, "INVD instruction."),
95 EXIT_REASON(VMX_EXIT_INVLPG , 14, "INVLPG instruction."),
96 EXIT_REASON(VMX_EXIT_RDPMC , 15, "RDPMCinstruction."),
97 EXIT_REASON(VMX_EXIT_RDTSC , 16, "RDTSC instruction."),
98 EXIT_REASON(VMX_EXIT_RSM , 17, "RSM instruction in SMM."),
99 EXIT_REASON(VMX_EXIT_VMCALL , 18, "VMCALL instruction."),
100 EXIT_REASON(VMX_EXIT_VMCLEAR , 19, "VMCLEAR instruction."),
101 EXIT_REASON(VMX_EXIT_VMLAUNCH , 20, "VMLAUNCH instruction."),
102 EXIT_REASON(VMX_EXIT_VMPTRLD , 21, "VMPTRLD instruction."),
103 EXIT_REASON(VMX_EXIT_VMPTRST , 22, "VMPTRST instruction."),
104 EXIT_REASON(VMX_EXIT_VMREAD , 23, "VMREAD instruction."),
105 EXIT_REASON(VMX_EXIT_VMRESUME , 24, "VMRESUME instruction."),
106 EXIT_REASON(VMX_EXIT_VMWRITE , 25, "VMWRITE instruction."),
107 EXIT_REASON(VMX_EXIT_VMXOFF , 26, "VMXOFF instruction."),
108 EXIT_REASON(VMX_EXIT_VMXON , 27, "VMXON instruction."),
109 EXIT_REASON(VMX_EXIT_MOV_CRX , 28, "Control-register accesses."),
110 EXIT_REASON(VMX_EXIT_MOV_DRX , 29, "Debug-register accesses."),
111 EXIT_REASON(VMX_EXIT_PORT_IO , 30, "I/O instruction."),
112 EXIT_REASON(VMX_EXIT_RDMSR , 31, "RDMSR instruction."),
113 EXIT_REASON(VMX_EXIT_WRMSR , 32, "WRMSR instruction."),
114 EXIT_REASON(VMX_EXIT_ERR_INVALID_GUEST_STATE, 33, "VM-entry failure due to invalid guest state."),
115 EXIT_REASON(VMX_EXIT_ERR_MSR_LOAD , 34, "VM-entry failure due to MSR loading."),
116 EXIT_REASON_NIL(),
117 EXIT_REASON(VMX_EXIT_MWAIT , 36, "MWAIT instruction."),
118 EXIT_REASON(VMX_EXIT_MTF , 37, "Monitor Trap Flag."),
119 EXIT_REASON_NIL(),
120 EXIT_REASON(VMX_EXIT_MONITOR , 39, "MONITOR instruction."),
121 EXIT_REASON(VMX_EXIT_PAUSE , 40, "PAUSE instruction."),
122 EXIT_REASON(VMX_EXIT_ERR_MACHINE_CHECK , 41, "VM-entry failure due to machine-check."),
123 EXIT_REASON_NIL(),
124 EXIT_REASON(VMX_EXIT_TPR_BELOW_THRESHOLD , 43, "TPR below threshold (MOV to CR8)."),
125 EXIT_REASON(VMX_EXIT_APIC_ACCESS , 44, "APIC access."),
126 EXIT_REASON(VMX_EXIT_VIRTUALIZED_EOI , 45, "Virtualized EOI."),
127 EXIT_REASON(VMX_EXIT_XDTR_ACCESS , 46, "GDTR/IDTR access using LGDT/SGDT/LIDT/SIDT."),
128 EXIT_REASON(VMX_EXIT_TR_ACCESS , 47, "LDTR/TR access using LLDT/SLDT/LTR/STR."),
129 EXIT_REASON(VMX_EXIT_EPT_VIOLATION , 48, "EPT violation."),
130 EXIT_REASON(VMX_EXIT_EPT_MISCONFIG , 49, "EPT misconfiguration."),
131 EXIT_REASON(VMX_EXIT_INVEPT , 50, "INVEPT instruction."),
132 EXIT_REASON(VMX_EXIT_RDTSCP , 51, "RDTSCP instruction."),
133 EXIT_REASON(VMX_EXIT_PREEMPT_TIMER , 52, "VMX-preemption timer expired."),
134 EXIT_REASON(VMX_EXIT_INVVPID , 53, "INVVPID instruction."),
135 EXIT_REASON(VMX_EXIT_WBINVD , 54, "WBINVD instruction."),
136 EXIT_REASON(VMX_EXIT_XSETBV , 55, "XSETBV instruction."),
137 EXIT_REASON(VMX_EXIT_APIC_WRITE , 56, "APIC write completed to virtual-APIC page."),
138 EXIT_REASON(VMX_EXIT_RDRAND , 57, "RDRAND instruction."),
139 EXIT_REASON(VMX_EXIT_INVPCID , 58, "INVPCID instruction."),
140 EXIT_REASON(VMX_EXIT_VMFUNC , 59, "VMFUNC instruction."),
141 EXIT_REASON(VMX_EXIT_ENCLS , 60, "ENCLS instrunction."),
142 EXIT_REASON(VMX_EXIT_RDSEED , 61, "RDSEED instruction."),
143 EXIT_REASON(VMX_EXIT_PML_FULL , 62, "Page-modification log full."),
144 EXIT_REASON(VMX_EXIT_XSAVES , 63, "XSAVES instruction."),
145 EXIT_REASON(VMX_EXIT_XRSTORS , 64, "XRSTORS instruction.")
146};
147/** Array index of the last valid VT-x exit reason. */
148#define MAX_EXITREASON_VTX 64
149
150/** A partial list of Exit reason descriptions for AMD-V, used to describe
151 * statistics.
152 *
153 * @note AMD-V have annoyingly large gaps (e.g. \#NPF VMEXIT comes at 1024),
154 * this array doesn't contain the entire set of exit reasons, we
155 * handle them via hmSvmGetSpecialExitReasonDesc(). */
156static const char * const g_apszAmdVExitReasons[MAX_EXITREASON_STAT] =
157{
158 EXIT_REASON(SVM_EXIT_READ_CR0 , 0, "Read CR0."),
159 EXIT_REASON(SVM_EXIT_READ_CR1 , 1, "Read CR1."),
160 EXIT_REASON(SVM_EXIT_READ_CR2 , 2, "Read CR2."),
161 EXIT_REASON(SVM_EXIT_READ_CR3 , 3, "Read CR3."),
162 EXIT_REASON(SVM_EXIT_READ_CR4 , 4, "Read CR4."),
163 EXIT_REASON(SVM_EXIT_READ_CR5 , 5, "Read CR5."),
164 EXIT_REASON(SVM_EXIT_READ_CR6 , 6, "Read CR6."),
165 EXIT_REASON(SVM_EXIT_READ_CR7 , 7, "Read CR7."),
166 EXIT_REASON(SVM_EXIT_READ_CR8 , 8, "Read CR8."),
167 EXIT_REASON(SVM_EXIT_READ_CR9 , 9, "Read CR9."),
168 EXIT_REASON(SVM_EXIT_READ_CR10 , 10, "Read CR10."),
169 EXIT_REASON(SVM_EXIT_READ_CR11 , 11, "Read CR11."),
170 EXIT_REASON(SVM_EXIT_READ_CR12 , 12, "Read CR12."),
171 EXIT_REASON(SVM_EXIT_READ_CR13 , 13, "Read CR13."),
172 EXIT_REASON(SVM_EXIT_READ_CR14 , 14, "Read CR14."),
173 EXIT_REASON(SVM_EXIT_READ_CR15 , 15, "Read CR15."),
174 EXIT_REASON(SVM_EXIT_WRITE_CR0 , 16, "Write CR0."),
175 EXIT_REASON(SVM_EXIT_WRITE_CR1 , 17, "Write CR1."),
176 EXIT_REASON(SVM_EXIT_WRITE_CR2 , 18, "Write CR2."),
177 EXIT_REASON(SVM_EXIT_WRITE_CR3 , 19, "Write CR3."),
178 EXIT_REASON(SVM_EXIT_WRITE_CR4 , 20, "Write CR4."),
179 EXIT_REASON(SVM_EXIT_WRITE_CR5 , 21, "Write CR5."),
180 EXIT_REASON(SVM_EXIT_WRITE_CR6 , 22, "Write CR6."),
181 EXIT_REASON(SVM_EXIT_WRITE_CR7 , 23, "Write CR7."),
182 EXIT_REASON(SVM_EXIT_WRITE_CR8 , 24, "Write CR8."),
183 EXIT_REASON(SVM_EXIT_WRITE_CR9 , 25, "Write CR9."),
184 EXIT_REASON(SVM_EXIT_WRITE_CR10 , 26, "Write CR10."),
185 EXIT_REASON(SVM_EXIT_WRITE_CR11 , 27, "Write CR11."),
186 EXIT_REASON(SVM_EXIT_WRITE_CR12 , 28, "Write CR12."),
187 EXIT_REASON(SVM_EXIT_WRITE_CR13 , 29, "Write CR13."),
188 EXIT_REASON(SVM_EXIT_WRITE_CR14 , 30, "Write CR14."),
189 EXIT_REASON(SVM_EXIT_WRITE_CR15 , 31, "Write CR15."),
190 EXIT_REASON(SVM_EXIT_READ_DR0 , 32, "Read DR0."),
191 EXIT_REASON(SVM_EXIT_READ_DR1 , 33, "Read DR1."),
192 EXIT_REASON(SVM_EXIT_READ_DR2 , 34, "Read DR2."),
193 EXIT_REASON(SVM_EXIT_READ_DR3 , 35, "Read DR3."),
194 EXIT_REASON(SVM_EXIT_READ_DR4 , 36, "Read DR4."),
195 EXIT_REASON(SVM_EXIT_READ_DR5 , 37, "Read DR5."),
196 EXIT_REASON(SVM_EXIT_READ_DR6 , 38, "Read DR6."),
197 EXIT_REASON(SVM_EXIT_READ_DR7 , 39, "Read DR7."),
198 EXIT_REASON(SVM_EXIT_READ_DR8 , 40, "Read DR8."),
199 EXIT_REASON(SVM_EXIT_READ_DR9 , 41, "Read DR9."),
200 EXIT_REASON(SVM_EXIT_READ_DR10 , 42, "Read DR10."),
201 EXIT_REASON(SVM_EXIT_READ_DR11 , 43, "Read DR11"),
202 EXIT_REASON(SVM_EXIT_READ_DR12 , 44, "Read DR12."),
203 EXIT_REASON(SVM_EXIT_READ_DR13 , 45, "Read DR13."),
204 EXIT_REASON(SVM_EXIT_READ_DR14 , 46, "Read DR14."),
205 EXIT_REASON(SVM_EXIT_READ_DR15 , 47, "Read DR15."),
206 EXIT_REASON(SVM_EXIT_WRITE_DR0 , 48, "Write DR0."),
207 EXIT_REASON(SVM_EXIT_WRITE_DR1 , 49, "Write DR1."),
208 EXIT_REASON(SVM_EXIT_WRITE_DR2 , 50, "Write DR2."),
209 EXIT_REASON(SVM_EXIT_WRITE_DR3 , 51, "Write DR3."),
210 EXIT_REASON(SVM_EXIT_WRITE_DR4 , 52, "Write DR4."),
211 EXIT_REASON(SVM_EXIT_WRITE_DR5 , 53, "Write DR5."),
212 EXIT_REASON(SVM_EXIT_WRITE_DR6 , 54, "Write DR6."),
213 EXIT_REASON(SVM_EXIT_WRITE_DR7 , 55, "Write DR7."),
214 EXIT_REASON(SVM_EXIT_WRITE_DR8 , 56, "Write DR8."),
215 EXIT_REASON(SVM_EXIT_WRITE_DR9 , 57, "Write DR9."),
216 EXIT_REASON(SVM_EXIT_WRITE_DR10 , 58, "Write DR10."),
217 EXIT_REASON(SVM_EXIT_WRITE_DR11 , 59, "Write DR11."),
218 EXIT_REASON(SVM_EXIT_WRITE_DR12 , 60, "Write DR12."),
219 EXIT_REASON(SVM_EXIT_WRITE_DR13 , 61, "Write DR13."),
220 EXIT_REASON(SVM_EXIT_WRITE_DR14 , 62, "Write DR14."),
221 EXIT_REASON(SVM_EXIT_WRITE_DR15 , 63, "Write DR15."),
222 EXIT_REASON(SVM_EXIT_XCPT_0 , 64, "Exception 0 (#DE)."),
223 EXIT_REASON(SVM_EXIT_XCPT_1 , 65, "Exception 1 (#DB)."),
224 EXIT_REASON(SVM_EXIT_XCPT_2 , 66, "Exception 2 (#NMI)."),
225 EXIT_REASON(SVM_EXIT_XCPT_3 , 67, "Exception 3 (#BP)."),
226 EXIT_REASON(SVM_EXIT_XCPT_4 , 68, "Exception 4 (#OF)."),
227 EXIT_REASON(SVM_EXIT_XCPT_5 , 69, "Exception 5 (#BR)."),
228 EXIT_REASON(SVM_EXIT_XCPT_6 , 70, "Exception 6 (#UD)."),
229 EXIT_REASON(SVM_EXIT_XCPT_7 , 71, "Exception 7 (#NM)."),
230 EXIT_REASON(SVM_EXIT_XCPT_8 , 72, "Exception 8 (#DF)."),
231 EXIT_REASON(SVM_EXIT_XCPT_9 , 73, "Exception 9 (#CO_SEG_OVERRUN)."),
232 EXIT_REASON(SVM_EXIT_XCPT_10 , 74, "Exception 10 (#TS)."),
233 EXIT_REASON(SVM_EXIT_XCPT_11 , 75, "Exception 11 (#NP)."),
234 EXIT_REASON(SVM_EXIT_XCPT_12 , 76, "Exception 12 (#SS)."),
235 EXIT_REASON(SVM_EXIT_XCPT_13 , 77, "Exception 13 (#GP)."),
236 EXIT_REASON(SVM_EXIT_XCPT_14 , 78, "Exception 14 (#PF)."),
237 EXIT_REASON(SVM_EXIT_XCPT_15 , 79, "Exception 15 (0x0f)."),
238 EXIT_REASON(SVM_EXIT_XCPT_16 , 80, "Exception 16 (#MF)."),
239 EXIT_REASON(SVM_EXIT_XCPT_17 , 81, "Exception 17 (#AC)."),
240 EXIT_REASON(SVM_EXIT_XCPT_18 , 82, "Exception 18 (#MC)."),
241 EXIT_REASON(SVM_EXIT_XCPT_19 , 83, "Exception 19 (#XF)."),
242 EXIT_REASON(SVM_EXIT_XCPT_20 , 84, "Exception 20 (#VE)."),
243 EXIT_REASON(SVM_EXIT_XCPT_21 , 85, "Exception 22 (0x15)."),
244 EXIT_REASON(SVM_EXIT_XCPT_22 , 86, "Exception 22 (0x16)."),
245 EXIT_REASON(SVM_EXIT_XCPT_23 , 87, "Exception 23 (0x17)."),
246 EXIT_REASON(SVM_EXIT_XCPT_24 , 88, "Exception 24 (0x18)."),
247 EXIT_REASON(SVM_EXIT_XCPT_25 , 89, "Exception 25 (0x19)."),
248 EXIT_REASON(SVM_EXIT_XCPT_26 , 90, "Exception 26 (0x1A)."),
249 EXIT_REASON(SVM_EXIT_XCPT_27 , 91, "Exception 27 (0x1B)."),
250 EXIT_REASON(SVM_EXIT_XCPT_28 , 92, "Exception 28 (0x1C)."),
251 EXIT_REASON(SVM_EXIT_XCPT_29 , 93, "Exception 29 (0x1D)."),
252 EXIT_REASON(SVM_EXIT_XCPT_30 , 94, "Exception 30 (#SX)."),
253 EXIT_REASON(SVM_EXIT_XCPT_31 , 95, "Exception 31 (0x1F)."),
254 EXIT_REASON(SVM_EXIT_INTR , 96, "Physical maskable interrupt (host)."),
255 EXIT_REASON(SVM_EXIT_NMI , 97, "Physical non-maskable interrupt (host)."),
256 EXIT_REASON(SVM_EXIT_SMI , 98, "System management interrupt (host)."),
257 EXIT_REASON(SVM_EXIT_INIT , 99, "Physical INIT signal (host)."),
258 EXIT_REASON(SVM_EXIT_VINTR , 100, "Virtual interrupt-window exit."),
259 EXIT_REASON(SVM_EXIT_CR0_SEL_WRITE, 101, "Write to CR0 that changed any bits other than CR0.TS or CR0.MP."),
260 EXIT_REASON(SVM_EXIT_IDTR_READ , 102, "Read IDTR"),
261 EXIT_REASON(SVM_EXIT_GDTR_READ , 103, "Read GDTR"),
262 EXIT_REASON(SVM_EXIT_LDTR_READ , 104, "Read LDTR."),
263 EXIT_REASON(SVM_EXIT_TR_READ , 105, "Read TR."),
264 EXIT_REASON(SVM_EXIT_IDTR_WRITE , 106, "Write IDTR."),
265 EXIT_REASON(SVM_EXIT_GDTR_WRITE , 107, "Write GDTR."),
266 EXIT_REASON(SVM_EXIT_LDTR_WRITE , 108, "Write LDTR."),
267 EXIT_REASON(SVM_EXIT_TR_WRITE , 109, "Write TR."),
268 EXIT_REASON(SVM_EXIT_RDTSC , 110, "RDTSC instruction."),
269 EXIT_REASON(SVM_EXIT_RDPMC , 111, "RDPMC instruction."),
270 EXIT_REASON(SVM_EXIT_PUSHF , 112, "PUSHF instruction."),
271 EXIT_REASON(SVM_EXIT_POPF , 113, "POPF instruction."),
272 EXIT_REASON(SVM_EXIT_CPUID , 114, "CPUID instruction."),
273 EXIT_REASON(SVM_EXIT_RSM , 115, "RSM instruction."),
274 EXIT_REASON(SVM_EXIT_IRET , 116, "IRET instruction."),
275 EXIT_REASON(SVM_EXIT_SWINT , 117, "Software interrupt (INTn instructions)."),
276 EXIT_REASON(SVM_EXIT_INVD , 118, "INVD instruction."),
277 EXIT_REASON(SVM_EXIT_PAUSE , 119, "PAUSE instruction."),
278 EXIT_REASON(SVM_EXIT_HLT , 120, "HLT instruction."),
279 EXIT_REASON(SVM_EXIT_INVLPG , 121, "INVLPG instruction."),
280 EXIT_REASON(SVM_EXIT_INVLPGA , 122, "INVLPGA instruction."),
281 EXIT_REASON(SVM_EXIT_IOIO , 123, "IN/OUT accessing protected port."),
282 EXIT_REASON(SVM_EXIT_MSR , 124, "RDMSR or WRMSR access to protected MSR."),
283 EXIT_REASON(SVM_EXIT_TASK_SWITCH , 125, "Task switch."),
284 EXIT_REASON(SVM_EXIT_FERR_FREEZE , 126, "Legacy FPU handling enabled; CPU frozen in an x87/mmx instr. waiting for interrupt."),
285 EXIT_REASON(SVM_EXIT_SHUTDOWN , 127, "Shutdown."),
286 EXIT_REASON(SVM_EXIT_VMRUN , 128, "VMRUN instruction."),
287 EXIT_REASON(SVM_EXIT_VMMCALL , 129, "VMCALL instruction."),
288 EXIT_REASON(SVM_EXIT_VMLOAD , 130, "VMLOAD instruction."),
289 EXIT_REASON(SVM_EXIT_VMSAVE , 131, "VMSAVE instruction."),
290 EXIT_REASON(SVM_EXIT_STGI , 132, "STGI instruction."),
291 EXIT_REASON(SVM_EXIT_CLGI , 133, "CLGI instruction."),
292 EXIT_REASON(SVM_EXIT_SKINIT , 134, "SKINIT instruction."),
293 EXIT_REASON(SVM_EXIT_RDTSCP , 135, "RDTSCP instruction."),
294 EXIT_REASON(SVM_EXIT_ICEBP , 136, "ICEBP instruction."),
295 EXIT_REASON(SVM_EXIT_WBINVD , 137, "WBINVD instruction."),
296 EXIT_REASON(SVM_EXIT_MONITOR , 138, "MONITOR instruction."),
297 EXIT_REASON(SVM_EXIT_MWAIT , 139, "MWAIT instruction."),
298 EXIT_REASON(SVM_EXIT_MWAIT_ARMED , 140, "MWAIT instruction when armed."),
299 EXIT_REASON(SVM_EXIT_XSETBV , 141, "XSETBV instruction."),
300};
301/** Array index of the last valid AMD-V exit reason. */
302#define MAX_EXITREASON_AMDV 141
303
304/** Special exit reasons not covered in the array above. */
305#define SVM_EXIT_REASON_NPF EXIT_REASON(SVM_EXIT_NPF , 1024, "Nested Page Fault.")
306#define SVM_EXIT_REASON_AVIC_INCOMPLETE_IPI EXIT_REASON(SVM_EXIT_AVIC_INCOMPLETE_IPI, 1025, "AVIC - Incomplete IPI delivery.")
307#define SVM_EXIT_REASON_AVIC_NOACCEL EXIT_REASON(SVM_EXIT_AVIC_NOACCEL , 1026, "AVIC - Unhandled register.")
308
309/**
310 * Gets the SVM exit reason if it's one of the reasons not present in the @c
311 * g_apszAmdVExitReasons array.
312 *
313 * @returns The exit reason or NULL if unknown.
314 * @param uExit The exit.
315 */
316DECLINLINE(const char *) hmSvmGetSpecialExitReasonDesc(uint16_t uExit)
317{
318 switch (uExit)
319 {
320 case SVM_EXIT_NPF: return SVM_EXIT_REASON_NPF;
321 case SVM_EXIT_AVIC_INCOMPLETE_IPI: return SVM_EXIT_REASON_AVIC_INCOMPLETE_IPI;
322 case SVM_EXIT_AVIC_NOACCEL: return SVM_EXIT_REASON_AVIC_NOACCEL;
323 }
324 return EXIT_REASON_NIL();
325}
326#undef EXIT_REASON_NIL
327#undef EXIT_REASON
328
329/** @def HMVMX_REPORT_FEATURE
330 * Reports VT-x feature to the release log.
331 *
332 * @param allowed1 Mask of allowed feature bits.
333 * @param disallowed0 Mask of disallowed feature bits.
334 * @param strdesc The description string to report.
335 * @param featflag Mask of the feature to report.
336 */
337#define HMVMX_REPORT_FEATURE(allowed1, disallowed0, strdesc, featflag) \
338 do { \
339 if ((allowed1) & (featflag)) \
340 { \
341 if ((disallowed0) & (featflag)) \
342 LogRel(("HM: " strdesc " (must be set)\n")); \
343 else \
344 LogRel(("HM: " strdesc "\n")); \
345 } \
346 else \
347 LogRel(("HM: " strdesc " (must be cleared)\n")); \
348 } while (0)
349
350/** @def HMVMX_REPORT_ALLOWED_FEATURE
351 * Reports an allowed VT-x feature to the release log.
352 *
353 * @param allowed1 Mask of allowed feature bits.
354 * @param strdesc The description string to report.
355 * @param featflag Mask of the feature to report.
356 */
357#define HMVMX_REPORT_ALLOWED_FEATURE(allowed1, strdesc, featflag) \
358 do { \
359 if ((allowed1) & (featflag)) \
360 LogRel(("HM: " strdesc "\n")); \
361 else \
362 LogRel(("HM: " strdesc " not supported\n")); \
363 } while (0)
364
365/** @def HMVMX_REPORT_MSR_CAPABILITY
366 * Reports MSR feature capability.
367 *
368 * @param msrcaps Mask of MSR feature bits.
369 * @param strdesc The description string to report.
370 * @param cap Mask of the feature to report.
371 */
372#define HMVMX_REPORT_MSR_CAPABILITY(msrcaps, strdesc, cap) \
373 do { \
374 if ((msrcaps) & (cap)) \
375 LogRel(("HM: " strdesc "\n")); \
376 } while (0)
377
378
379/*********************************************************************************************************************************
380* Internal Functions *
381*********************************************************************************************************************************/
382static DECLCALLBACK(int) hmR3Save(PVM pVM, PSSMHANDLE pSSM);
383static DECLCALLBACK(int) hmR3Load(PVM pVM, PSSMHANDLE pSSM, uint32_t uVersion, uint32_t uPass);
384static DECLCALLBACK(void) hmR3InfoSvmNstGstVmcbCache(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs);
385static DECLCALLBACK(void) hmR3InfoExitHistory(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs);
386static DECLCALLBACK(void) hmR3InfoEventPending(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs);
387static int hmR3InitCPU(PVM pVM);
388static int hmR3InitFinalizeR0(PVM pVM);
389static int hmR3InitFinalizeR0Intel(PVM pVM);
390static int hmR3InitFinalizeR0Amd(PVM pVM);
391static int hmR3TermCPU(PVM pVM);
392
393
394
395/**
396 * Initializes the HM.
397 *
398 * This is the very first component to really do init after CFGM so that we can
399 * establish the predominat execution engine for the VM prior to initializing
400 * other modules. It takes care of NEM initialization if needed (HM disabled or
401 * not available in HW).
402 *
403 * If VT-x or AMD-V hardware isn't available, HM will try fall back on a native
404 * hypervisor API via NEM, and then back on raw-mode if that isn't available
405 * either. The fallback to raw-mode will not happen if /HM/HMForced is set
406 * (like for guest using SMP or 64-bit as well as for complicated guest like OS
407 * X, OS/2 and others).
408 *
409 * Note that a lot of the set up work is done in ring-0 and thus postponed till
410 * the ring-3 and ring-0 callback to HMR3InitCompleted.
411 *
412 * @returns VBox status code.
413 * @param pVM The cross context VM structure.
414 *
415 * @remarks Be careful with what we call here, since most of the VMM components
416 * are uninitialized.
417 */
418VMMR3_INT_DECL(int) HMR3Init(PVM pVM)
419{
420 LogFlow(("HMR3Init\n"));
421
422 /*
423 * Assert alignment and sizes.
424 */
425 AssertCompileMemberAlignment(VM, hm.s, 32);
426 AssertCompile(sizeof(pVM->hm.s) <= sizeof(pVM->hm.padding));
427
428 /*
429 * Register the saved state data unit.
430 */
431 int rc = SSMR3RegisterInternal(pVM, "HWACCM", 0, HM_SAVED_STATE_VERSION, sizeof(HM),
432 NULL, NULL, NULL,
433 NULL, hmR3Save, NULL,
434 NULL, hmR3Load, NULL);
435 if (RT_FAILURE(rc))
436 return rc;
437
438 /*
439 * Register info handlers.
440 */
441 rc = DBGFR3InfoRegisterInternalEx(pVM, "exithistory", "Dumps the HM VM-exit history.", hmR3InfoExitHistory,
442 DBGFINFO_FLAGS_ALL_EMTS);
443 AssertRCReturn(rc, rc);
444
445 rc = DBGFR3InfoRegisterInternalEx(pVM, "hmeventpending", "Dumps the pending HM event.", hmR3InfoEventPending,
446 DBGFINFO_FLAGS_ALL_EMTS);
447 AssertRCReturn(rc, rc);
448
449 rc = DBGFR3InfoRegisterInternalEx(pVM, "svmvmcbcache", "Dumps the HM SVM nested-guest VMCB cache.",
450 hmR3InfoSvmNstGstVmcbCache, DBGFINFO_FLAGS_ALL_EMTS);
451 AssertRCReturn(rc, rc);
452
453 /*
454 * Read configuration.
455 */
456 PCFGMNODE pCfgHm = CFGMR3GetChild(CFGMR3GetRoot(pVM), "HM/");
457
458 /*
459 * Validate the HM settings.
460 */
461 rc = CFGMR3ValidateConfig(pCfgHm, "/HM/",
462 "HMForced"
463 "|UseNEMInstead"
464 "|FallbackToNEM"
465 "|EnableNestedPaging"
466 "|EnableUX"
467 "|EnableLargePages"
468 "|EnableVPID"
469 "|IBPBOnVMExit"
470 "|IBPBOnVMEntry"
471 "|SpecCtrlByHost"
472 "|TPRPatchingEnabled"
473 "|64bitEnabled"
474 "|Exclusive"
475 "|MaxResumeLoops"
476 "|VmxPleGap"
477 "|VmxPleWindow"
478 "|UseVmxPreemptTimer"
479 "|SvmPauseFilter"
480 "|SvmPauseFilterThreshold"
481 "|SvmVirtVmsaveVmload"
482 "|SvmVGif",
483 "" /* pszValidNodes */, "HM" /* pszWho */, 0 /* uInstance */);
484 if (RT_FAILURE(rc))
485 return rc;
486
487 /** @cfgm{/HM/HMForced, bool, false}
488 * Forces hardware virtualization, no falling back on raw-mode. HM must be
489 * enabled, i.e. /HMEnabled must be true. */
490 bool fHMForced;
491#ifdef VBOX_WITH_RAW_MODE
492 rc = CFGMR3QueryBoolDef(pCfgHm, "HMForced", &fHMForced, false);
493 AssertRCReturn(rc, rc);
494 AssertLogRelMsgReturn(!fHMForced || pVM->fHMEnabled, ("Configuration error: HM forced but not enabled!\n"),
495 VERR_INVALID_PARAMETER);
496# if defined(RT_OS_DARWIN)
497 if (pVM->fHMEnabled)
498 fHMForced = true;
499# endif
500 AssertLogRelMsgReturn(pVM->cCpus == 1 || pVM->fHMEnabled, ("Configuration error: SMP requires HM to be enabled!\n"),
501 VERR_INVALID_PARAMETER);
502 if (pVM->cCpus > 1)
503 fHMForced = true;
504#else /* !VBOX_WITH_RAW_MODE */
505 AssertRelease(pVM->fHMEnabled);
506 fHMForced = true;
507#endif /* !VBOX_WITH_RAW_MODE */
508
509 /** @cfgm{/HM/UseNEMInstead, bool, true}
510 * Don't use HM, use NEM instead. */
511 bool fUseNEMInstead = false;
512 rc = CFGMR3QueryBoolDef(pCfgHm, "UseNEMInstead", &fUseNEMInstead, false);
513 AssertRCReturn(rc, rc);
514 if (fUseNEMInstead && pVM->fHMEnabled)
515 {
516 LogRel(("HM: Setting fHMEnabled to false because fUseNEMInstead is set.\n"));
517 pVM->fHMEnabled = false;
518 }
519
520 /** @cfgm{/HM/FallbackToNEM, bool, true}
521 * Enables fallback on NEM. */
522 bool fFallbackToNEM = true;
523 rc = CFGMR3QueryBoolDef(pCfgHm, "FallbackToNEM", &fFallbackToNEM, true);
524 AssertRCReturn(rc, rc);
525
526 /** @cfgm{/HM/EnableNestedPaging, bool, false}
527 * Enables nested paging (aka extended page tables). */
528 rc = CFGMR3QueryBoolDef(pCfgHm, "EnableNestedPaging", &pVM->hm.s.fAllowNestedPaging, false);
529 AssertRCReturn(rc, rc);
530
531 /** @cfgm{/HM/EnableUX, bool, true}
532 * Enables the VT-x unrestricted execution feature. */
533 rc = CFGMR3QueryBoolDef(pCfgHm, "EnableUX", &pVM->hm.s.vmx.fAllowUnrestricted, true);
534 AssertRCReturn(rc, rc);
535
536 /** @cfgm{/HM/EnableLargePages, bool, false}
537 * Enables using large pages (2 MB) for guest memory, thus saving on (nested)
538 * page table walking and maybe better TLB hit rate in some cases. */
539 rc = CFGMR3QueryBoolDef(pCfgHm, "EnableLargePages", &pVM->hm.s.fLargePages, false);
540 AssertRCReturn(rc, rc);
541
542 /** @cfgm{/HM/EnableVPID, bool, false}
543 * Enables the VT-x VPID feature. */
544 rc = CFGMR3QueryBoolDef(pCfgHm, "EnableVPID", &pVM->hm.s.vmx.fAllowVpid, false);
545 AssertRCReturn(rc, rc);
546
547 /** @cfgm{/HM/TPRPatchingEnabled, bool, false}
548 * Enables TPR patching for 32-bit windows guests with IO-APIC. */
549 rc = CFGMR3QueryBoolDef(pCfgHm, "TPRPatchingEnabled", &pVM->hm.s.fTprPatchingAllowed, false);
550 AssertRCReturn(rc, rc);
551
552 /** @cfgm{/HM/64bitEnabled, bool, 32-bit:false, 64-bit:true}
553 * Enables AMD64 cpu features.
554 * On 32-bit hosts this isn't default and require host CPU support. 64-bit hosts
555 * already have the support. */
556#ifdef VBOX_ENABLE_64_BITS_GUESTS
557 rc = CFGMR3QueryBoolDef(pCfgHm, "64bitEnabled", &pVM->hm.s.fAllow64BitGuests, HC_ARCH_BITS == 64);
558 AssertLogRelRCReturn(rc, rc);
559#else
560 pVM->hm.s.fAllow64BitGuests = false;
561#endif
562
563 /** @cfgm{/HM/VmxPleGap, uint32_t, 0}
564 * The pause-filter exiting gap in TSC ticks. When the number of ticks between
565 * two successive PAUSE instructions exceeds VmxPleGap, the CPU considers the
566 * latest PAUSE instruction to be start of a new PAUSE loop.
567 */
568 rc = CFGMR3QueryU32Def(pCfgHm, "VmxPleGap", &pVM->hm.s.vmx.cPleGapTicks, 0);
569 AssertRCReturn(rc, rc);
570
571 /** @cfgm{/HM/VmxPleWindow, uint32_t, 0}
572 * The pause-filter exiting window in TSC ticks. When the number of ticks
573 * between the current PAUSE instruction and first PAUSE of a loop exceeds
574 * VmxPleWindow, a VM-exit is triggered.
575 *
576 * Setting VmxPleGap and VmxPleGap to 0 disables pause-filter exiting.
577 */
578 rc = CFGMR3QueryU32Def(pCfgHm, "VmxPleWindow", &pVM->hm.s.vmx.cPleWindowTicks, 0);
579 AssertRCReturn(rc, rc);
580
581 /** @cfgm{/HM/SvmPauseFilterCount, uint16_t, 0}
582 * A counter that is decrement each time a PAUSE instruction is executed by the
583 * guest. When the counter is 0, a \#VMEXIT is triggered.
584 *
585 * Setting SvmPauseFilterCount to 0 disables pause-filter exiting.
586 */
587 rc = CFGMR3QueryU16Def(pCfgHm, "SvmPauseFilter", &pVM->hm.s.svm.cPauseFilter, 0);
588 AssertRCReturn(rc, rc);
589
590 /** @cfgm{/HM/SvmPauseFilterThreshold, uint16_t, 0}
591 * The pause filter threshold in ticks. When the elapsed time (in ticks) between
592 * two successive PAUSE instructions exceeds SvmPauseFilterThreshold, the
593 * PauseFilter count is reset to its initial value. However, if PAUSE is
594 * executed PauseFilter times within PauseFilterThreshold ticks, a VM-exit will
595 * be triggered.
596 *
597 * Requires SvmPauseFilterCount to be non-zero for pause-filter threshold to be
598 * activated.
599 */
600 rc = CFGMR3QueryU16Def(pCfgHm, "SvmPauseFilterThreshold", &pVM->hm.s.svm.cPauseFilterThresholdTicks, 0);
601 AssertRCReturn(rc, rc);
602
603 /** @cfgm{/HM/SvmVirtVmsaveVmload, bool, true}
604 * Whether to make use of virtualized VMSAVE/VMLOAD feature of the CPU if it's
605 * available. */
606 rc = CFGMR3QueryBoolDef(pCfgHm, "SvmVirtVmsaveVmload", &pVM->hm.s.svm.fVirtVmsaveVmload, true);
607 AssertRCReturn(rc, rc);
608
609 /** @cfgm{/HM/SvmVGif, bool, true}
610 * Whether to make use of Virtual GIF (Global Interrupt Flag) feature of the CPU
611 * if it's available. */
612 rc = CFGMR3QueryBoolDef(pCfgHm, "SvmVGif", &pVM->hm.s.svm.fVGif, true);
613 AssertRCReturn(rc, rc);
614
615 /** @cfgm{/HM/Exclusive, bool}
616 * Determines the init method for AMD-V and VT-x. If set to true, HM will do a
617 * global init for each host CPU. If false, we do local init each time we wish
618 * to execute guest code.
619 *
620 * On Windows, default is false due to the higher risk of conflicts with other
621 * hypervisors.
622 *
623 * On Mac OS X, this setting is ignored since the code does not handle local
624 * init when it utilizes the OS provided VT-x function, SUPR0EnableVTx().
625 */
626#if defined(RT_OS_DARWIN)
627 pVM->hm.s.fGlobalInit = true;
628#else
629 rc = CFGMR3QueryBoolDef(pCfgHm, "Exclusive", &pVM->hm.s.fGlobalInit,
630# if defined(RT_OS_WINDOWS)
631 false
632# else
633 true
634# endif
635 );
636 AssertLogRelRCReturn(rc, rc);
637#endif
638
639 /** @cfgm{/HM/MaxResumeLoops, uint32_t}
640 * The number of times to resume guest execution before we forcibly return to
641 * ring-3. The return value of RTThreadPreemptIsPendingTrusty in ring-0
642 * determines the default value. */
643 rc = CFGMR3QueryU32Def(pCfgHm, "MaxResumeLoops", &pVM->hm.s.cMaxResumeLoops, 0 /* set by R0 later */);
644 AssertLogRelRCReturn(rc, rc);
645
646 /** @cfgm{/HM/UseVmxPreemptTimer, bool}
647 * Whether to make use of the VMX-preemption timer feature of the CPU if it's
648 * available. */
649 rc = CFGMR3QueryBoolDef(pCfgHm, "UseVmxPreemptTimer", &pVM->hm.s.vmx.fUsePreemptTimer, true);
650 AssertLogRelRCReturn(rc, rc);
651
652 /** @cfgm{/HM/IBPBOnVMExit, bool}
653 * Costly paranoia setting. */
654 rc = CFGMR3QueryBoolDef(pCfgHm, "IBPBOnVMExit", &pVM->hm.s.fIbpbOnVmExit, false);
655 AssertLogRelRCReturn(rc, rc);
656
657 /** @cfgm{/HM/IBPBOnVMEntry, bool}
658 * Costly paranoia setting. */
659 rc = CFGMR3QueryBoolDef(pCfgHm, "IBPBOnVMEntry", &pVM->hm.s.fIbpbOnVmEntry, false);
660 AssertLogRelRCReturn(rc, rc);
661
662 /** @cfgm{/HM/SpecCtrlByHost, bool}
663 * Another expensive paranoia setting. */
664 rc = CFGMR3QueryBoolDef(pCfgHm, "SpecCtrlByHost", &pVM->hm.s.fSpecCtrlByHost, false);
665 AssertLogRelRCReturn(rc, rc);
666
667 /*
668 * Check if VT-x or AMD-v support according to the users wishes.
669 */
670 /** @todo SUPR3QueryVTCaps won't catch VERR_VMX_IN_VMX_ROOT_MODE or
671 * VERR_SVM_IN_USE. */
672 if (pVM->fHMEnabled)
673 {
674 uint32_t fCaps;
675 rc = SUPR3QueryVTCaps(&fCaps);
676 if (RT_SUCCESS(rc))
677 {
678 if (fCaps & SUPVTCAPS_AMD_V)
679 {
680 LogRel(("HM: HMR3Init: AMD-V%s\n", fCaps & SUPVTCAPS_NESTED_PAGING ? " w/ nested paging" : ""));
681 pVM->hm.s.svm.fSupported = true;
682 VM_SET_MAIN_EXECUTION_ENGINE(pVM, VM_EXEC_ENGINE_HW_VIRT);
683 }
684 else if (fCaps & SUPVTCAPS_VT_X)
685 {
686 const char *pszWhy;
687 rc = SUPR3QueryVTxSupported(&pszWhy);
688 if (RT_SUCCESS(rc))
689 {
690 LogRel(("HM: HMR3Init: VT-x%s%s%s\n",
691 fCaps & SUPVTCAPS_NESTED_PAGING ? " w/ nested paging" : "",
692 fCaps & SUPVTCAPS_VTX_UNRESTRICTED_GUEST ? " and unrestricted guest execution" : "",
693 (fCaps & (SUPVTCAPS_NESTED_PAGING | SUPVTCAPS_VTX_UNRESTRICTED_GUEST)) ? " hw support" : ""));
694 pVM->hm.s.vmx.fSupported = true;
695 VM_SET_MAIN_EXECUTION_ENGINE(pVM, VM_EXEC_ENGINE_HW_VIRT);
696 }
697 else
698 {
699 /*
700 * Before failing, try fallback to NEM if we're allowed to do that.
701 */
702 pVM->fHMEnabled = false;
703 Assert(pVM->bMainExecutionEngine == VM_EXEC_ENGINE_NOT_SET);
704 if (fFallbackToNEM)
705 {
706 LogRel(("HM: HMR3Init: Attempting fall back to NEM: The host kernel does not support VT-x - %s\n", pszWhy));
707 int rc2 = NEMR3Init(pVM, true /*fFallback*/, fHMForced);
708
709 ASMCompilerBarrier(); /* NEMR3Init may have changed bMainExecutionEngine. */
710 if ( RT_SUCCESS(rc2)
711 && pVM->bMainExecutionEngine != VM_EXEC_ENGINE_NOT_SET)
712 rc = VINF_SUCCESS;
713 }
714 if (RT_FAILURE(rc))
715 {
716 if (fHMForced)
717 return VMSetError(pVM, rc, RT_SRC_POS, "The host kernel does not support VT-x: %s\n", pszWhy);
718
719 /* Fall back to raw-mode. */
720 LogRel(("HM: HMR3Init: Falling back to raw-mode: The host kernel does not support VT-x - %s\n", pszWhy));
721 VM_SET_MAIN_EXECUTION_ENGINE(pVM, VM_EXEC_ENGINE_RAW_MODE);
722 }
723 }
724 }
725 else
726 AssertLogRelMsgFailedReturn(("SUPR3QueryVTCaps didn't return either AMD-V or VT-x flag set (%#x)!\n", fCaps),
727 VERR_INTERNAL_ERROR_5);
728
729 /*
730 * Do we require a little bit or raw-mode for 64-bit guest execution?
731 */
732 pVM->fHMNeedRawModeCtx = HC_ARCH_BITS == 32
733 && pVM->fHMEnabled
734 && pVM->hm.s.fAllow64BitGuests;
735
736 /*
737 * Disable nested paging and unrestricted guest execution now if they're
738 * configured so that CPUM can make decisions based on our configuration.
739 */
740 Assert(!pVM->hm.s.fNestedPaging);
741 if (pVM->hm.s.fAllowNestedPaging)
742 {
743 if (fCaps & SUPVTCAPS_NESTED_PAGING)
744 pVM->hm.s.fNestedPaging = true;
745 else
746 pVM->hm.s.fAllowNestedPaging = false;
747 }
748
749 if (fCaps & SUPVTCAPS_VT_X)
750 {
751 Assert(!pVM->hm.s.vmx.fUnrestrictedGuest);
752 if (pVM->hm.s.vmx.fAllowUnrestricted)
753 {
754 if ( (fCaps & SUPVTCAPS_VTX_UNRESTRICTED_GUEST)
755 && pVM->hm.s.fNestedPaging)
756 pVM->hm.s.vmx.fUnrestrictedGuest = true;
757 else
758 pVM->hm.s.vmx.fAllowUnrestricted = false;
759 }
760 }
761 }
762 else
763 {
764 const char *pszMsg;
765 switch (rc)
766 {
767 case VERR_UNSUPPORTED_CPU: pszMsg = "Unknown CPU, VT-x or AMD-v features cannot be ascertained"; break;
768 case VERR_VMX_NO_VMX: pszMsg = "VT-x is not available"; break;
769 case VERR_VMX_MSR_VMX_DISABLED: pszMsg = "VT-x is disabled in the BIOS"; break;
770 case VERR_VMX_MSR_ALL_VMX_DISABLED: pszMsg = "VT-x is disabled in the BIOS for all CPU modes"; break;
771 case VERR_VMX_MSR_LOCKING_FAILED: pszMsg = "Failed to enable and lock VT-x features"; break;
772 case VERR_SVM_NO_SVM: pszMsg = "AMD-V is not available"; break;
773 case VERR_SVM_DISABLED: pszMsg = "AMD-V is disabled in the BIOS (or by the host OS)"; break;
774 default:
775 return VMSetError(pVM, rc, RT_SRC_POS, "SUPR3QueryVTCaps failed with %Rrc", rc);
776 }
777
778 /*
779 * Before failing, try fallback to NEM if we're allowed to do that.
780 */
781 pVM->fHMEnabled = false;
782 if (fFallbackToNEM)
783 {
784 LogRel(("HM: HMR3Init: Attempting fall back to NEM: %s\n", pszMsg));
785 int rc2 = NEMR3Init(pVM, true /*fFallback*/, fHMForced);
786 ASMCompilerBarrier(); /* NEMR3Init may have changed bMainExecutionEngine. */
787 if ( RT_SUCCESS(rc2)
788 && pVM->bMainExecutionEngine != VM_EXEC_ENGINE_NOT_SET)
789 rc = VINF_SUCCESS;
790 }
791 if (RT_FAILURE(rc))
792 {
793 if (fHMForced)
794 return VM_SET_ERROR(pVM, rc, pszMsg);
795
796 LogRel(("HM: HMR3Init: Falling back to raw-mode: %s\n", pszMsg));
797 VM_SET_MAIN_EXECUTION_ENGINE(pVM, VM_EXEC_ENGINE_RAW_MODE);
798 }
799 }
800 }
801 else
802 {
803 /*
804 * Disabled HM mean raw-mode, unless NEM is supposed to be used.
805 */
806 if (!fUseNEMInstead)
807 VM_SET_MAIN_EXECUTION_ENGINE(pVM, VM_EXEC_ENGINE_RAW_MODE);
808 else
809 {
810 rc = NEMR3Init(pVM, false /*fFallback*/, true);
811 ASMCompilerBarrier(); /* NEMR3Init may have changed bMainExecutionEngine. */
812 if (RT_FAILURE(rc))
813 return rc;
814 }
815 }
816
817 return VINF_SUCCESS;
818}
819
820
821/**
822 * Initializes the per-VCPU HM.
823 *
824 * @returns VBox status code.
825 * @param pVM The cross context VM structure.
826 */
827static int hmR3InitCPU(PVM pVM)
828{
829 LogFlow(("HMR3InitCPU\n"));
830
831 if (!HMIsEnabled(pVM))
832 return VINF_SUCCESS;
833
834 for (VMCPUID i = 0; i < pVM->cCpus; i++)
835 {
836 PVMCPU pVCpu = &pVM->aCpus[i];
837 pVCpu->hm.s.fActive = false;
838 }
839
840#ifdef VBOX_WITH_STATISTICS
841 STAM_REG(pVM, &pVM->hm.s.StatTprPatchSuccess, STAMTYPE_COUNTER, "/HM/TPR/Patch/Success", STAMUNIT_OCCURENCES, "Number of times an instruction was successfully patched.");
842 STAM_REG(pVM, &pVM->hm.s.StatTprPatchFailure, STAMTYPE_COUNTER, "/HM/TPR/Patch/Failed", STAMUNIT_OCCURENCES, "Number of unsuccessful patch attempts.");
843 STAM_REG(pVM, &pVM->hm.s.StatTprReplaceSuccessCr8, STAMTYPE_COUNTER, "/HM/TPR/Replace/SuccessCR8",STAMUNIT_OCCURENCES, "Number of instruction replacements by MOV CR8.");
844 STAM_REG(pVM, &pVM->hm.s.StatTprReplaceSuccessVmc, STAMTYPE_COUNTER, "/HM/TPR/Replace/SuccessVMC",STAMUNIT_OCCURENCES, "Number of instruction replacements by VMMCALL.");
845 STAM_REG(pVM, &pVM->hm.s.StatTprReplaceFailure, STAMTYPE_COUNTER, "/HM/TPR/Replace/Failed", STAMUNIT_OCCURENCES, "Number of unsuccessful replace attempts.");
846#endif
847
848 /*
849 * Statistics.
850 */
851 for (VMCPUID i = 0; i < pVM->cCpus; i++)
852 {
853 PVMCPU pVCpu = &pVM->aCpus[i];
854 int rc;
855
856#ifdef VBOX_WITH_STATISTICS
857 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatPoke, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
858 "Profiling of RTMpPokeCpu",
859 "/PROF/CPU%d/HM/Poke", i);
860 AssertRC(rc);
861 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatSpinPoke, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
862 "Profiling of poke wait",
863 "/PROF/CPU%d/HM/PokeWait", i);
864 AssertRC(rc);
865 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatSpinPokeFailed, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
866 "Profiling of poke wait when RTMpPokeCpu fails",
867 "/PROF/CPU%d/HM/PokeWaitFailed", i);
868 AssertRC(rc);
869 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatEntry, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
870 "Profiling of VMXR0RunGuestCode entry",
871 "/PROF/CPU%d/HM/StatEntry", i);
872 AssertRC(rc);
873 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExit1, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
874 "Profiling of VMXR0RunGuestCode exit part 1",
875 "/PROF/CPU%d/HM/SwitchFromGC_1", i);
876 AssertRC(rc);
877 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExit2, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
878 "Profiling of VMXR0RunGuestCode exit part 2",
879 "/PROF/CPU%d/HM/SwitchFromGC_2", i);
880 AssertRC(rc);
881
882 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitIO, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
883 "I/O",
884 "/PROF/CPU%d/HM/SwitchFromGC_2/IO", i);
885 AssertRC(rc);
886 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitMovCRx, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
887 "MOV CRx",
888 "/PROF/CPU%d/HM/SwitchFromGC_2/MovCRx", i);
889 AssertRC(rc);
890 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitXcptNmi, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
891 "Exceptions, NMIs",
892 "/PROF/CPU%d/HM/SwitchFromGC_2/XcptNmi", i);
893 AssertRC(rc);
894
895 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatLoadGuestState, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
896 "Profiling of VMXR0LoadGuestState",
897 "/PROF/CPU%d/HM/StatLoadGuestState", i);
898 AssertRC(rc);
899 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatLoadGuestFpuState, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
900 "Profiling of CPUMR0LoadGuestFPU",
901 "/PROF/CPU%d/HM/StatLoadGuestFpuState", i);
902 AssertRC(rc);
903 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatInGC, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
904 "Profiling of VMLAUNCH/VMRESUME.",
905 "/PROF/CPU%d/HM/InGC", i);
906 AssertRC(rc);
907
908# if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
909 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatWorldSwitch3264, STAMTYPE_PROFILE, STAMVISIBILITY_USED,
910 STAMUNIT_TICKS_PER_CALL, "Profiling of the 32/64 switcher.",
911 "/PROF/CPU%d/HM/Switcher3264", i);
912 AssertRC(rc);
913# endif
914
915# ifdef HM_PROFILE_EXIT_DISPATCH
916 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitDispatch, STAMTYPE_PROFILE_ADV, STAMVISIBILITY_USED,
917 STAMUNIT_TICKS_PER_CALL, "Profiling the dispatching of exit handlers.",
918 "/PROF/CPU%d/HM/ExitDispatch", i);
919 AssertRC(rc);
920# endif
921
922#endif
923# define HM_REG_COUNTER(a, b, desc) \
924 rc = STAMR3RegisterF(pVM, a, STAMTYPE_COUNTER, STAMVISIBILITY_ALWAYS, STAMUNIT_OCCURENCES, desc, b, i); \
925 AssertRC(rc);
926
927#ifdef VBOX_WITH_STATISTICS
928 HM_REG_COUNTER(&pVCpu->hm.s.StatExitAll, "/HM/CPU%d/Exit/All", "Exits (total).");
929 HM_REG_COUNTER(&pVCpu->hm.s.StatExitShadowNM, "/HM/CPU%d/Exit/Trap/Shw/#NM", "Shadow #NM (device not available, no math co-processor) exception.");
930 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestNM, "/HM/CPU%d/Exit/Trap/Gst/#NM", "Guest #NM (device not available, no math co-processor) exception.");
931 HM_REG_COUNTER(&pVCpu->hm.s.StatExitShadowPF, "/HM/CPU%d/Exit/Trap/Shw/#PF", "Shadow #PF (page fault) exception.");
932 HM_REG_COUNTER(&pVCpu->hm.s.StatExitShadowPFEM, "/HM/CPU%d/Exit/Trap/Shw/#PF-EM", "#PF (page fault) exception going back to ring-3 for emulating the instruction.");
933 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestPF, "/HM/CPU%d/Exit/Trap/Gst/#PF", "Guest #PF (page fault) exception.");
934 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestUD, "/HM/CPU%d/Exit/Trap/Gst/#UD", "Guest #UD (undefined opcode) exception.");
935 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestSS, "/HM/CPU%d/Exit/Trap/Gst/#SS", "Guest #SS (stack-segment fault) exception.");
936 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestNP, "/HM/CPU%d/Exit/Trap/Gst/#NP", "Guest #NP (segment not present) exception.");
937 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestGP, "/HM/CPU%d/Exit/Trap/Gst/#GP", "Guest #GP (general protection) exception.");
938 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestMF, "/HM/CPU%d/Exit/Trap/Gst/#MF", "Guest #MF (x87 FPU error, math fault) exception.");
939 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestDE, "/HM/CPU%d/Exit/Trap/Gst/#DE", "Guest #DE (divide error) exception.");
940 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestDB, "/HM/CPU%d/Exit/Trap/Gst/#DB", "Guest #DB (debug) exception.");
941 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestBP, "/HM/CPU%d/Exit/Trap/Gst/#BP", "Guest #BP (breakpoint) exception.");
942 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestXF, "/HM/CPU%d/Exit/Trap/Gst/#XF", "Guest #XF (extended math fault, SIMD FPU) exception.");
943 HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestXcpUnk, "/HM/CPU%d/Exit/Trap/Gst/Other", "Other guest exceptions.");
944 HM_REG_COUNTER(&pVCpu->hm.s.StatExitInvlpg, "/HM/CPU%d/Exit/Instr/Invlpg", "Guest attempted to execute INVLPG.");
945 HM_REG_COUNTER(&pVCpu->hm.s.StatExitInvd, "/HM/CPU%d/Exit/Instr/Invd", "Guest attempted to execute INVD.");
946 HM_REG_COUNTER(&pVCpu->hm.s.StatExitWbinvd, "/HM/CPU%d/Exit/Instr/Wbinvd", "Guest attempted to execute WBINVD.");
947 HM_REG_COUNTER(&pVCpu->hm.s.StatExitPause, "/HM/CPU%d/Exit/Instr/Pause", "Guest attempted to execute PAUSE.");
948 HM_REG_COUNTER(&pVCpu->hm.s.StatExitCpuid, "/HM/CPU%d/Exit/Instr/Cpuid", "Guest attempted to execute CPUID.");
949 HM_REG_COUNTER(&pVCpu->hm.s.StatExitRdtsc, "/HM/CPU%d/Exit/Instr/Rdtsc", "Guest attempted to execute RDTSC.");
950 HM_REG_COUNTER(&pVCpu->hm.s.StatExitRdtscp, "/HM/CPU%d/Exit/Instr/Rdtscp", "Guest attempted to execute RDTSCP.");
951 HM_REG_COUNTER(&pVCpu->hm.s.StatExitRdpmc, "/HM/CPU%d/Exit/Instr/Rdpmc", "Guest attempted to execute RDPMC.");
952 HM_REG_COUNTER(&pVCpu->hm.s.StatExitRdrand, "/HM/CPU%d/Exit/Instr/Rdrand", "Guest attempted to execute RDRAND.");
953 HM_REG_COUNTER(&pVCpu->hm.s.StatExitRdmsr, "/HM/CPU%d/Exit/Instr/Rdmsr", "Guest attempted to execute RDMSR.");
954 HM_REG_COUNTER(&pVCpu->hm.s.StatExitWrmsr, "/HM/CPU%d/Exit/Instr/Wrmsr", "Guest attempted to execute WRMSR.");
955 HM_REG_COUNTER(&pVCpu->hm.s.StatExitMwait, "/HM/CPU%d/Exit/Instr/Mwait", "Guest attempted to execute MWAIT.");
956 HM_REG_COUNTER(&pVCpu->hm.s.StatExitMonitor, "/HM/CPU%d/Exit/Instr/Monitor", "Guest attempted to execute MONITOR.");
957 HM_REG_COUNTER(&pVCpu->hm.s.StatExitDRxWrite, "/HM/CPU%d/Exit/Instr/DR/Write", "Guest attempted to write a debug register.");
958 HM_REG_COUNTER(&pVCpu->hm.s.StatExitDRxRead, "/HM/CPU%d/Exit/Instr/DR/Read", "Guest attempted to read a debug register.");
959 HM_REG_COUNTER(&pVCpu->hm.s.StatExitClts, "/HM/CPU%d/Exit/Instr/CLTS", "Guest attempted to execute CLTS.");
960 HM_REG_COUNTER(&pVCpu->hm.s.StatExitLmsw, "/HM/CPU%d/Exit/Instr/LMSW", "Guest attempted to execute LMSW.");
961 HM_REG_COUNTER(&pVCpu->hm.s.StatExitCli, "/HM/CPU%d/Exit/Instr/Cli", "Guest attempted to execute CLI.");
962 HM_REG_COUNTER(&pVCpu->hm.s.StatExitSti, "/HM/CPU%d/Exit/Instr/Sti", "Guest attempted to execute STI.");
963 HM_REG_COUNTER(&pVCpu->hm.s.StatExitPushf, "/HM/CPU%d/Exit/Instr/Pushf", "Guest attempted to execute PUSHF.");
964 HM_REG_COUNTER(&pVCpu->hm.s.StatExitPopf, "/HM/CPU%d/Exit/Instr/Popf", "Guest attempted to execute POPF.");
965 HM_REG_COUNTER(&pVCpu->hm.s.StatExitIret, "/HM/CPU%d/Exit/Instr/Iret", "Guest attempted to execute IRET.");
966 HM_REG_COUNTER(&pVCpu->hm.s.StatExitInt, "/HM/CPU%d/Exit/Instr/Int", "Guest attempted to execute INT.");
967 HM_REG_COUNTER(&pVCpu->hm.s.StatExitHlt, "/HM/CPU%d/Exit/Instr/Hlt", "Guest attempted to execute HLT.");
968 HM_REG_COUNTER(&pVCpu->hm.s.StatExitXdtrAccess, "/HM/CPU%d/Exit/Instr/XdtrAccess", "Guest attempted to access descriptor table register (GDTR, IDTR, LDTR).");
969 HM_REG_COUNTER(&pVCpu->hm.s.StatExitIOWrite, "/HM/CPU%d/Exit/IO/Write", "I/O write.");
970 HM_REG_COUNTER(&pVCpu->hm.s.StatExitIORead, "/HM/CPU%d/Exit/IO/Read", "I/O read.");
971 HM_REG_COUNTER(&pVCpu->hm.s.StatExitIOStringWrite, "/HM/CPU%d/Exit/IO/WriteString", "String I/O write.");
972 HM_REG_COUNTER(&pVCpu->hm.s.StatExitIOStringRead, "/HM/CPU%d/Exit/IO/ReadString", "String I/O read.");
973 HM_REG_COUNTER(&pVCpu->hm.s.StatExitIntWindow, "/HM/CPU%d/Exit/IntWindow", "Interrupt-window exit. Guest is ready to receive interrupts again.");
974 HM_REG_COUNTER(&pVCpu->hm.s.StatExitExtInt, "/HM/CPU%d/Exit/ExtInt", "Host interrupt received.");
975#endif
976 HM_REG_COUNTER(&pVCpu->hm.s.StatExitHostNmiInGC, "/HM/CPU%d/Exit/HostNmiInGC", "Host NMI received while in guest context.");
977#ifdef VBOX_WITH_STATISTICS
978 HM_REG_COUNTER(&pVCpu->hm.s.StatExitPreemptTimer, "/HM/CPU%d/Exit/PreemptTimer", "VMX-preemption timer expired.");
979 HM_REG_COUNTER(&pVCpu->hm.s.StatExitTprBelowThreshold, "/HM/CPU%d/Exit/TprBelowThreshold", "TPR lowered below threshold by the guest.");
980 HM_REG_COUNTER(&pVCpu->hm.s.StatExitTaskSwitch, "/HM/CPU%d/Exit/TaskSwitch", "Guest attempted a task switch.");
981 HM_REG_COUNTER(&pVCpu->hm.s.StatExitMtf, "/HM/CPU%d/Exit/MonitorTrapFlag", "Monitor Trap Flag.");
982 HM_REG_COUNTER(&pVCpu->hm.s.StatExitApicAccess, "/HM/CPU%d/Exit/ApicAccess", "APIC access. Guest attempted to access memory at a physical address on the APIC-access page.");
983
984 HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchTprMaskedIrq, "/HM/CPU%d/Switch/TprMaskedIrq", "PDMGetInterrupt() signals TPR masks pending Irq.");
985 HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchGuestIrq, "/HM/CPU%d/Switch/IrqPending", "PDMGetInterrupt() cleared behind our back!?!.");
986 HM_REG_COUNTER(&pVCpu->hm.s.StatPendingHostIrq, "/HM/CPU%d/Switch/PendingHostIrq", "Exit to ring-3 due to pending host interrupt before executing guest code.");
987 HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchHmToR3FF, "/HM/CPU%d/Switch/HmToR3FF", "Exit to ring-3 due to pending timers, EMT rendezvous, critical section etc.");
988 HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchExitToR3, "/HM/CPU%d/Switch/ExitToR3", "Exit to ring-3 (total).");
989 HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchLongJmpToR3, "/HM/CPU%d/Switch/LongJmpToR3", "Longjump to ring-3.");
990 HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchMaxResumeLoops, "/HM/CPU%d/Switch/MaxResumeToR3", "Maximum VMRESUME inner-loop counter reached.");
991 HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchHltToR3, "/HM/CPU%d/Switch/HltToR3", "HLT causing us to go to ring-3.");
992 HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchApicAccessToR3, "/HM/CPU%d/Switch/ApicAccessToR3", "APIC access causing us to go to ring-3.");
993#endif
994 HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchPreempt, "/HM/CPU%d/Switch/Preempting", "EMT has been preempted while in HM context.");
995#ifdef VBOX_WITH_STATISTICS
996 HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchPreemptSaveHostState, "/HM/CPU%d/Switch/SaveHostState", "Preemption caused us to resave host state.");
997
998 HM_REG_COUNTER(&pVCpu->hm.s.StatInjectInterrupt, "/HM/CPU%d/EventInject/Interrupt", "Injected an external interrupt into the guest.");
999 HM_REG_COUNTER(&pVCpu->hm.s.StatInjectXcpt, "/HM/CPU%d/EventInject/Trap", "Injected an exception into the guest.");
1000 HM_REG_COUNTER(&pVCpu->hm.s.StatInjectPendingReflect, "/HM/CPU%d/EventInject/PendingReflect", "Reflecting an exception (or #DF) caused due to event injection.");
1001 HM_REG_COUNTER(&pVCpu->hm.s.StatInjectPendingInterpret, "/HM/CPU%d/EventInject/PendingInterpret", "Falling to interpreter for handling exception caused due to event injection.");
1002
1003 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushPage, "/HM/CPU%d/Flush/Page", "Invalidating a guest page on all guest CPUs.");
1004 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushPageManual, "/HM/CPU%d/Flush/Page/Virt", "Invalidating a guest page using guest-virtual address.");
1005 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushPhysPageManual, "/HM/CPU%d/Flush/Page/Phys", "Invalidating a guest page using guest-physical address.");
1006 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushTlb, "/HM/CPU%d/Flush/TLB", "Forcing a full guest-TLB flush (ring-0).");
1007 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushTlbManual, "/HM/CPU%d/Flush/TLB/Manual", "Request a full guest-TLB flush.");
1008 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushTlbWorldSwitch, "/HM/CPU%d/Flush/TLB/CpuSwitch", "Forcing a full guest-TLB flush due to host-CPU reschedule or ASID-limit hit by another guest-VCPU.");
1009 HM_REG_COUNTER(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch, "/HM/CPU%d/Flush/TLB/Skipped", "No TLB flushing required.");
1010 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushEntire, "/HM/CPU%d/Flush/TLB/Entire", "Flush the entire TLB (host + guest).");
1011 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushAsid, "/HM/CPU%d/Flush/TLB/ASID", "Flushed guest-TLB entries for the current VPID.");
1012 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushNestedPaging, "/HM/CPU%d/Flush/TLB/NestedPaging", "Flushed guest-TLB entries for the current EPT.");
1013 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushTlbInvlpgVirt, "/HM/CPU%d/Flush/TLB/InvlpgVirt", "Invalidated a guest-TLB entry for a guest-virtual address.");
1014 HM_REG_COUNTER(&pVCpu->hm.s.StatFlushTlbInvlpgPhys, "/HM/CPU%d/Flush/TLB/InvlpgPhys", "Currently not possible, flushes entire guest-TLB.");
1015 HM_REG_COUNTER(&pVCpu->hm.s.StatTlbShootdown, "/HM/CPU%d/Flush/Shootdown/Page", "Inter-VCPU request to flush queued guest page.");
1016 HM_REG_COUNTER(&pVCpu->hm.s.StatTlbShootdownFlush, "/HM/CPU%d/Flush/Shootdown/TLB", "Inter-VCPU request to flush entire guest-TLB.");
1017
1018 HM_REG_COUNTER(&pVCpu->hm.s.StatTscParavirt, "/HM/CPU%d/TSC/Paravirt", "Paravirtualized TSC in effect.");
1019 HM_REG_COUNTER(&pVCpu->hm.s.StatTscOffset, "/HM/CPU%d/TSC/Offset", "TSC offsetting is in effect.");
1020 HM_REG_COUNTER(&pVCpu->hm.s.StatTscIntercept, "/HM/CPU%d/TSC/Intercept", "Intercept TSC accesses.");
1021
1022 HM_REG_COUNTER(&pVCpu->hm.s.StatDRxArmed, "/HM/CPU%d/Debug/Armed", "Loaded guest-debug state while loading guest-state.");
1023 HM_REG_COUNTER(&pVCpu->hm.s.StatDRxContextSwitch, "/HM/CPU%d/Debug/ContextSwitch", "Loaded guest-debug state on MOV DRx.");
1024 HM_REG_COUNTER(&pVCpu->hm.s.StatDRxIoCheck, "/HM/CPU%d/Debug/IOCheck", "Checking for I/O breakpoint.");
1025
1026 HM_REG_COUNTER(&pVCpu->hm.s.StatLoadMinimal, "/HM/CPU%d/Load/Minimal", "VM-entry loading minimal guest-state.");
1027 HM_REG_COUNTER(&pVCpu->hm.s.StatLoadFull, "/HM/CPU%d/Load/Full", "VM-entry loading the full guest-state.");
1028 HM_REG_COUNTER(&pVCpu->hm.s.StatLoadFull, "/HM/CPU%d/Load/GuestFpu", "VM-entry loading the guest-FPU state.");
1029
1030 HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadRmSelBase, "/HM/CPU%d/VMXCheck/RMSelBase", "Could not use VMX due to unsuitable real-mode selector base.");
1031 HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadRmSelLimit, "/HM/CPU%d/VMXCheck/RMSelLimit", "Could not use VMX due to unsuitable real-mode selector limit.");
1032 HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckRmOk, "/HM/CPU%d/VMXCheck/VMX_RM", "VMX execution in real (V86) mode OK.");
1033 HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadSel, "/HM/CPU%d/VMXCheck/Selector", "Could not use VMX due to unsuitable selector.");
1034 HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadRpl, "/HM/CPU%d/VMXCheck/RPL", "Could not use VMX due to unsuitable RPL.");
1035 HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadLdt, "/HM/CPU%d/VMXCheck/LDT", "Could not use VMX due to unsuitable LDT.");
1036 HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadTr, "/HM/CPU%d/VMXCheck/TR", "Could not use VMX due to unsuitable TR.");
1037 HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckPmOk, "/HM/CPU%d/VMXCheck/VMX_PM", "VMX execution in protected mode OK.");
1038
1039#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
1040 HM_REG_COUNTER(&pVCpu->hm.s.StatFpu64SwitchBack, "/HM/CPU%d/Switch64/Fpu", "Saving guest FPU/XMM state.");
1041 HM_REG_COUNTER(&pVCpu->hm.s.StatDebug64SwitchBack, "/HM/CPU%d/Switch64/Debug", "Saving guest debug state.");
1042#endif
1043
1044 for (unsigned j = 0; j < RT_ELEMENTS(pVCpu->hm.s.StatExitCRxWrite); j++)
1045 {
1046 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitCRxWrite[j], STAMTYPE_COUNTER, STAMVISIBILITY_USED,
1047 STAMUNIT_OCCURENCES, "Profiling of CRx writes",
1048 "/HM/CPU%d/Exit/Instr/CR/Write/%x", i, j);
1049 AssertRC(rc);
1050 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitCRxRead[j], STAMTYPE_COUNTER, STAMVISIBILITY_USED,
1051 STAMUNIT_OCCURENCES, "Profiling of CRx reads",
1052 "/HM/CPU%d/Exit/Instr/CR/Read/%x", i, j);
1053 AssertRC(rc);
1054 }
1055
1056#undef HM_REG_COUNTER
1057
1058 const char *const *papszDesc = ASMIsIntelCpu() || ASMIsViaCentaurCpu() ? &g_apszVTxExitReasons[0]
1059 : &g_apszAmdVExitReasons[0];
1060
1061 /*
1062 * Guest Exit reason stats.
1063 */
1064 pVCpu->hm.s.paStatExitReason = NULL;
1065 rc = MMHyperAlloc(pVM, MAX_EXITREASON_STAT * sizeof(*pVCpu->hm.s.paStatExitReason), 0 /* uAlignment */, MM_TAG_HM,
1066 (void **)&pVCpu->hm.s.paStatExitReason);
1067 AssertRCReturn(rc, rc);
1068 for (int j = 0; j < MAX_EXITREASON_STAT; j++)
1069 {
1070 if (papszDesc[j])
1071 {
1072 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.paStatExitReason[j], STAMTYPE_COUNTER, STAMVISIBILITY_USED,
1073 STAMUNIT_OCCURENCES, papszDesc[j], "/HM/CPU%d/Exit/Reason/%02x", i, j);
1074 AssertRCReturn(rc, rc);
1075 }
1076 }
1077 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitReasonNpf, STAMTYPE_COUNTER, STAMVISIBILITY_USED, STAMUNIT_OCCURENCES,
1078 "Nested page fault", "/HM/CPU%d/Exit/Reason/#NPF", i);
1079 AssertRCReturn(rc, rc);
1080 pVCpu->hm.s.paStatExitReasonR0 = MMHyperR3ToR0(pVM, pVCpu->hm.s.paStatExitReason);
1081# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE
1082 Assert(pVCpu->hm.s.paStatExitReasonR0 != NIL_RTR0PTR || !HMIsEnabled(pVM));
1083# else
1084 Assert(pVCpu->hm.s.paStatExitReasonR0 != NIL_RTR0PTR);
1085# endif
1086
1087#ifdef VBOX_WITH_NESTED_HWVIRT_SVM
1088 /*
1089 * Nested-guest Exit reason stats.
1090 */
1091 pVCpu->hm.s.paStatNestedExitReason = NULL;
1092 rc = MMHyperAlloc(pVM, MAX_EXITREASON_STAT * sizeof(*pVCpu->hm.s.paStatNestedExitReason), 0 /* uAlignment */, MM_TAG_HM,
1093 (void **)&pVCpu->hm.s.paStatNestedExitReason);
1094 AssertRCReturn(rc, rc);
1095 for (int j = 0; j < MAX_EXITREASON_STAT; j++)
1096 {
1097 if (papszDesc[j])
1098 {
1099 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.paStatNestedExitReason[j], STAMTYPE_COUNTER, STAMVISIBILITY_USED,
1100 STAMUNIT_OCCURENCES, papszDesc[j], "/HM/CPU%d/NestedExit/Reason/%02x", i, j);
1101 AssertRC(rc);
1102 }
1103 }
1104 rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatNestedExitReasonNpf, STAMTYPE_COUNTER, STAMVISIBILITY_USED,
1105 STAMUNIT_OCCURENCES, "Nested page fault", "/HM/CPU%d/NestedExit/Reason/#NPF", i);
1106 AssertRCReturn(rc, rc);
1107 pVCpu->hm.s.paStatNestedExitReasonR0 = MMHyperR3ToR0(pVM, pVCpu->hm.s.paStatNestedExitReason);
1108# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE
1109 Assert(pVCpu->hm.s.paStatNestedExitReasonR0 != NIL_RTR0PTR || !HMIsEnabled(pVM));
1110# else
1111 Assert(pVCpu->hm.s.paStatNestedExitReasonR0 != NIL_RTR0PTR);
1112# endif
1113#endif
1114
1115 /*
1116 * Injected events stats.
1117 */
1118 rc = MMHyperAlloc(pVM, sizeof(STAMCOUNTER) * 256, 8, MM_TAG_HM, (void **)&pVCpu->hm.s.paStatInjectedIrqs);
1119 AssertRCReturn(rc, rc);
1120 pVCpu->hm.s.paStatInjectedIrqsR0 = MMHyperR3ToR0(pVM, pVCpu->hm.s.paStatInjectedIrqs);
1121# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE
1122 Assert(pVCpu->hm.s.paStatInjectedIrqsR0 != NIL_RTR0PTR || !HMIsEnabled(pVM));
1123# else
1124 Assert(pVCpu->hm.s.paStatInjectedIrqsR0 != NIL_RTR0PTR);
1125# endif
1126 for (unsigned j = 0; j < 255; j++)
1127 {
1128 STAMR3RegisterF(pVM, &pVCpu->hm.s.paStatInjectedIrqs[j], STAMTYPE_COUNTER, STAMVISIBILITY_USED, STAMUNIT_OCCURENCES,
1129 "Injected event.",
1130 (j < 0x20) ? "/HM/CPU%d/EventInject/InjectTrap/%02X" : "/HM/CPU%d/EventInject/InjectIRQ/%02X", i, j);
1131 }
1132
1133#endif /* VBOX_WITH_STATISTICS */
1134 }
1135
1136#ifdef VBOX_WITH_CRASHDUMP_MAGIC
1137 /*
1138 * Magic marker for searching in crash dumps.
1139 */
1140 for (VMCPUID i = 0; i < pVM->cCpus; i++)
1141 {
1142 PVMCPU pVCpu = &pVM->aCpus[i];
1143
1144 PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache;
1145 strcpy((char *)pCache->aMagic, "VMCSCACHE Magic");
1146 pCache->uMagic = UINT64_C(0xDEADBEEFDEADBEEF);
1147 }
1148#endif
1149
1150 return VINF_SUCCESS;
1151}
1152
1153
1154/**
1155 * Called when a init phase has completed.
1156 *
1157 * @returns VBox status code.
1158 * @param pVM The cross context VM structure.
1159 * @param enmWhat The phase that completed.
1160 */
1161VMMR3_INT_DECL(int) HMR3InitCompleted(PVM pVM, VMINITCOMPLETED enmWhat)
1162{
1163 switch (enmWhat)
1164 {
1165 case VMINITCOMPLETED_RING3:
1166 return hmR3InitCPU(pVM);
1167 case VMINITCOMPLETED_RING0:
1168 return hmR3InitFinalizeR0(pVM);
1169 default:
1170 return VINF_SUCCESS;
1171 }
1172}
1173
1174
1175/**
1176 * Turns off normal raw mode features.
1177 *
1178 * @param pVM The cross context VM structure.
1179 */
1180static void hmR3DisableRawMode(PVM pVM)
1181{
1182 /* Reinit the paging mode to force the new shadow mode. */
1183 for (VMCPUID i = 0; i < pVM->cCpus; i++)
1184 {
1185 PVMCPU pVCpu = &pVM->aCpus[i];
1186
1187 PGMR3ChangeMode(pVM, pVCpu, PGMMODE_REAL);
1188 }
1189}
1190
1191
1192/**
1193 * Initialize VT-x or AMD-V.
1194 *
1195 * @returns VBox status code.
1196 * @param pVM The cross context VM structure.
1197 */
1198static int hmR3InitFinalizeR0(PVM pVM)
1199{
1200 int rc;
1201
1202 if (!HMIsEnabled(pVM))
1203 return VINF_SUCCESS;
1204
1205 /*
1206 * Hack to allow users to work around broken BIOSes that incorrectly set
1207 * EFER.SVME, which makes us believe somebody else is already using AMD-V.
1208 */
1209 if ( !pVM->hm.s.vmx.fSupported
1210 && !pVM->hm.s.svm.fSupported
1211 && pVM->hm.s.lLastError == VERR_SVM_IN_USE /* implies functional AMD-V */
1212 && RTEnvExist("VBOX_HWVIRTEX_IGNORE_SVM_IN_USE"))
1213 {
1214 LogRel(("HM: VBOX_HWVIRTEX_IGNORE_SVM_IN_USE active!\n"));
1215 pVM->hm.s.svm.fSupported = true;
1216 pVM->hm.s.svm.fIgnoreInUseError = true;
1217 pVM->hm.s.lLastError = VINF_SUCCESS;
1218 }
1219
1220 /*
1221 * Report ring-0 init errors.
1222 */
1223 if ( !pVM->hm.s.vmx.fSupported
1224 && !pVM->hm.s.svm.fSupported)
1225 {
1226 LogRel(("HM: Failed to initialize VT-x / AMD-V: %Rrc\n", pVM->hm.s.lLastError));
1227 LogRel(("HM: VMX MSR_IA32_FEATURE_CONTROL=%RX64\n", pVM->hm.s.vmx.Msrs.u64FeatureCtrl));
1228 switch (pVM->hm.s.lLastError)
1229 {
1230 case VERR_VMX_IN_VMX_ROOT_MODE:
1231 return VM_SET_ERROR(pVM, VERR_VMX_IN_VMX_ROOT_MODE, "VT-x is being used by another hypervisor");
1232 case VERR_VMX_NO_VMX:
1233 return VM_SET_ERROR(pVM, VERR_VMX_NO_VMX, "VT-x is not available");
1234 case VERR_VMX_MSR_VMX_DISABLED:
1235 return VM_SET_ERROR(pVM, VERR_VMX_MSR_VMX_DISABLED, "VT-x is disabled in the BIOS");
1236 case VERR_VMX_MSR_ALL_VMX_DISABLED:
1237 return VM_SET_ERROR(pVM, VERR_VMX_MSR_ALL_VMX_DISABLED, "VT-x is disabled in the BIOS for all CPU modes");
1238 case VERR_VMX_MSR_LOCKING_FAILED:
1239 return VM_SET_ERROR(pVM, VERR_VMX_MSR_LOCKING_FAILED, "Failed to lock VT-x features while trying to enable VT-x");
1240 case VERR_VMX_MSR_VMX_ENABLE_FAILED:
1241 return VM_SET_ERROR(pVM, VERR_VMX_MSR_VMX_ENABLE_FAILED, "Failed to enable VT-x features");
1242 case VERR_VMX_MSR_SMX_VMX_ENABLE_FAILED:
1243 return VM_SET_ERROR(pVM, VERR_VMX_MSR_SMX_VMX_ENABLE_FAILED, "Failed to enable VT-x features in SMX mode");
1244
1245 case VERR_SVM_IN_USE:
1246 return VM_SET_ERROR(pVM, VERR_SVM_IN_USE, "AMD-V is being used by another hypervisor");
1247 case VERR_SVM_NO_SVM:
1248 return VM_SET_ERROR(pVM, VERR_SVM_NO_SVM, "AMD-V is not available");
1249 case VERR_SVM_DISABLED:
1250 return VM_SET_ERROR(pVM, VERR_SVM_DISABLED, "AMD-V is disabled in the BIOS");
1251 }
1252 return VMSetError(pVM, pVM->hm.s.lLastError, RT_SRC_POS, "HM ring-0 init failed: %Rrc", pVM->hm.s.lLastError);
1253 }
1254
1255 /*
1256 * Enable VT-x or AMD-V on all host CPUs.
1257 */
1258 rc = SUPR3CallVMMR0Ex(pVM->pVMR0, 0 /*idCpu*/, VMMR0_DO_HM_ENABLE, 0, NULL);
1259 if (RT_FAILURE(rc))
1260 {
1261 LogRel(("HM: Failed to enable, error %Rrc\n", rc));
1262 HMR3CheckError(pVM, rc);
1263 return rc;
1264 }
1265
1266 /*
1267 * No TPR patching is required when the IO-APIC is not enabled for this VM.
1268 * (Main should have taken care of this already)
1269 */
1270 if (!PDMHasIoApic(pVM))
1271 {
1272 Assert(!pVM->hm.s.fTprPatchingAllowed); /* paranoia */
1273 pVM->hm.s.fTprPatchingAllowed = false;
1274 }
1275
1276 /*
1277 * Sync options.
1278 */
1279 /** @todo Move this out of of CPUMCTX and into some ring-0 only HM structure.
1280 * That will require a little bit of work, of course. */
1281 for (VMCPUID iCpu = 0; iCpu < pVM->cCpus; iCpu++)
1282 {
1283 PVMCPU pVCpu = &pVM->aCpus[iCpu];
1284 PCPUMCTX pCpuCtx = CPUMQueryGuestCtxPtr(pVCpu);
1285 pCpuCtx->fWorldSwitcher &= ~(CPUMCTX_WSF_IBPB_EXIT | CPUMCTX_WSF_IBPB_ENTRY);
1286 if (pVM->cpum.ro.HostFeatures.fIbpb)
1287 {
1288 if (pVM->hm.s.fIbpbOnVmExit)
1289 pCpuCtx->fWorldSwitcher |= CPUMCTX_WSF_IBPB_EXIT;
1290 if (pVM->hm.s.fIbpbOnVmEntry)
1291 pCpuCtx->fWorldSwitcher |= CPUMCTX_WSF_IBPB_ENTRY;
1292 }
1293 if (iCpu == 0)
1294 LogRel(("HM: fWorldSwitcher=%#x (fIbpbOnVmExit=%RTbool fIbpbOnVmEntry=%RTbool)\n",
1295 pCpuCtx->fWorldSwitcher, pVM->hm.s.fIbpbOnVmExit, pVM->hm.s.fIbpbOnVmEntry));
1296 }
1297
1298 /*
1299 * Do the vendor specific initialization
1300 *
1301 * Note! We disable release log buffering here since we're doing relatively
1302 * lot of logging and doesn't want to hit the disk with each LogRel
1303 * statement.
1304 */
1305 AssertLogRelReturn(!pVM->hm.s.fInitialized, VERR_HM_IPE_5);
1306 bool fOldBuffered = RTLogRelSetBuffering(true /*fBuffered*/);
1307 if (pVM->hm.s.vmx.fSupported)
1308 rc = hmR3InitFinalizeR0Intel(pVM);
1309 else
1310 rc = hmR3InitFinalizeR0Amd(pVM);
1311 LogRel(("HM: VT-x/AMD-V init method: %s\n", (pVM->hm.s.fGlobalInit) ? "GLOBAL" : "LOCAL"));
1312 RTLogRelSetBuffering(fOldBuffered);
1313 pVM->hm.s.fInitialized = true;
1314
1315 return rc;
1316}
1317
1318
1319/**
1320 * @callback_method_impl{FNPDMVMMDEVHEAPNOTIFY}
1321 */
1322static DECLCALLBACK(void) hmR3VmmDevHeapNotify(PVM pVM, void *pvAllocation, RTGCPHYS GCPhysAllocation)
1323{
1324 NOREF(pVM);
1325 NOREF(pvAllocation);
1326 NOREF(GCPhysAllocation);
1327}
1328
1329
1330/**
1331 * Finish VT-x initialization (after ring-0 init).
1332 *
1333 * @returns VBox status code.
1334 * @param pVM The cross context VM structure.
1335 */
1336static int hmR3InitFinalizeR0Intel(PVM pVM)
1337{
1338 int rc;
1339
1340 Log(("pVM->hm.s.vmx.fSupported = %d\n", pVM->hm.s.vmx.fSupported));
1341 AssertLogRelReturn(pVM->hm.s.vmx.Msrs.u64FeatureCtrl != 0, VERR_HM_IPE_4);
1342
1343 uint64_t val;
1344 uint64_t zap;
1345
1346 LogRel(("HM: Using VT-x implementation 2.0\n"));
1347 LogRel(("HM: Host CR4 = %#RX64\n", pVM->hm.s.vmx.u64HostCr4));
1348 LogRel(("HM: Host EFER = %#RX64\n", pVM->hm.s.vmx.u64HostEfer));
1349 LogRel(("HM: MSR_IA32_SMM_MONITOR_CTL = %#RX64\n", pVM->hm.s.vmx.u64HostSmmMonitorCtl));
1350 LogRel(("HM: MSR_IA32_FEATURE_CONTROL = %#RX64\n", pVM->hm.s.vmx.Msrs.u64FeatureCtrl));
1351 if (!(pVM->hm.s.vmx.Msrs.u64FeatureCtrl & MSR_IA32_FEATURE_CONTROL_LOCK))
1352 LogRel(("HM: IA32_FEATURE_CONTROL lock bit not set, possibly bad hardware!\n"));
1353 LogRel(("HM: MSR_IA32_VMX_BASIC_INFO = %#RX64\n", pVM->hm.s.vmx.Msrs.u64BasicInfo));
1354 LogRel(("HM: VMCS id = %#x\n", MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo)));
1355 LogRel(("HM: VMCS size = %u bytes\n", MSR_IA32_VMX_BASIC_INFO_VMCS_SIZE(pVM->hm.s.vmx.Msrs.u64BasicInfo)));
1356 LogRel(("HM: VMCS physical address limit = %s\n", MSR_IA32_VMX_BASIC_INFO_VMCS_PHYS_WIDTH(pVM->hm.s.vmx.Msrs.u64BasicInfo) ? "< 4 GB" : "None"));
1357 LogRel(("HM: VMCS memory type = %#x\n", MSR_IA32_VMX_BASIC_INFO_VMCS_MEM_TYPE(pVM->hm.s.vmx.Msrs.u64BasicInfo)));
1358 LogRel(("HM: Dual-monitor treatment support = %RTbool\n", MSR_IA32_VMX_BASIC_INFO_VMCS_DUAL_MON(pVM->hm.s.vmx.Msrs.u64BasicInfo)));
1359 LogRel(("HM: OUTS & INS instruction-info = %RTbool\n", MSR_IA32_VMX_BASIC_INFO_VMCS_INS_OUTS(pVM->hm.s.vmx.Msrs.u64BasicInfo)));
1360 LogRel(("HM: Supports true capability MSRs = %RTbool\n", MSR_IA32_VMX_BASIC_INFO_TRUE_CONTROLS(pVM->hm.s.vmx.Msrs.u64BasicInfo)));
1361 LogRel(("HM: Max resume loops = %u\n", pVM->hm.s.cMaxResumeLoops));
1362
1363 LogRel(("HM: MSR_IA32_VMX_PINBASED_CTLS = %#RX64\n", pVM->hm.s.vmx.Msrs.VmxPinCtls.u));
1364 val = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1;
1365 zap = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0;
1366 HMVMX_REPORT_FEATURE(val, zap, "EXT_INT_EXIT", VMX_VMCS_CTRL_PIN_EXEC_EXT_INT_EXIT);
1367 HMVMX_REPORT_FEATURE(val, zap, "NMI_EXIT", VMX_VMCS_CTRL_PIN_EXEC_NMI_EXIT);
1368 HMVMX_REPORT_FEATURE(val, zap, "VIRTUAL_NMI", VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI);
1369 HMVMX_REPORT_FEATURE(val, zap, "PREEMPT_TIMER", VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER);
1370 HMVMX_REPORT_FEATURE(val, zap, "POSTED_INTR", VMX_VMCS_CTRL_PIN_EXEC_POSTED_INTR);
1371
1372 LogRel(("HM: MSR_IA32_VMX_PROCBASED_CTLS = %#RX64\n", pVM->hm.s.vmx.Msrs.VmxProcCtls.u));
1373 val = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1;
1374 zap = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0;
1375 HMVMX_REPORT_FEATURE(val, zap, "INT_WINDOW_EXIT", VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT);
1376 HMVMX_REPORT_FEATURE(val, zap, "USE_TSC_OFFSETTING", VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING);
1377 HMVMX_REPORT_FEATURE(val, zap, "HLT_EXIT", VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT);
1378 HMVMX_REPORT_FEATURE(val, zap, "INVLPG_EXIT", VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT);
1379 HMVMX_REPORT_FEATURE(val, zap, "MWAIT_EXIT", VMX_VMCS_CTRL_PROC_EXEC_MWAIT_EXIT);
1380 HMVMX_REPORT_FEATURE(val, zap, "RDPMC_EXIT", VMX_VMCS_CTRL_PROC_EXEC_RDPMC_EXIT);
1381 HMVMX_REPORT_FEATURE(val, zap, "RDTSC_EXIT", VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT);
1382 HMVMX_REPORT_FEATURE(val, zap, "CR3_LOAD_EXIT", VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT);
1383 HMVMX_REPORT_FEATURE(val, zap, "CR3_STORE_EXIT", VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT);
1384 HMVMX_REPORT_FEATURE(val, zap, "CR8_LOAD_EXIT", VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT);
1385 HMVMX_REPORT_FEATURE(val, zap, "CR8_STORE_EXIT", VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT);
1386 HMVMX_REPORT_FEATURE(val, zap, "USE_TPR_SHADOW", VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW);
1387 HMVMX_REPORT_FEATURE(val, zap, "NMI_WINDOW_EXIT", VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT);
1388 HMVMX_REPORT_FEATURE(val, zap, "MOV_DR_EXIT", VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT);
1389 HMVMX_REPORT_FEATURE(val, zap, "UNCOND_IO_EXIT", VMX_VMCS_CTRL_PROC_EXEC_UNCOND_IO_EXIT);
1390 HMVMX_REPORT_FEATURE(val, zap, "USE_IO_BITMAPS", VMX_VMCS_CTRL_PROC_EXEC_USE_IO_BITMAPS);
1391 HMVMX_REPORT_FEATURE(val, zap, "MONITOR_TRAP_FLAG", VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG);
1392 HMVMX_REPORT_FEATURE(val, zap, "USE_MSR_BITMAPS", VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS);
1393 HMVMX_REPORT_FEATURE(val, zap, "MONITOR_EXIT", VMX_VMCS_CTRL_PROC_EXEC_MONITOR_EXIT);
1394 HMVMX_REPORT_FEATURE(val, zap, "PAUSE_EXIT", VMX_VMCS_CTRL_PROC_EXEC_PAUSE_EXIT);
1395 HMVMX_REPORT_FEATURE(val, zap, "USE_SECONDARY_EXEC_CTRL", VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL);
1396 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
1397 {
1398 LogRel(("HM: MSR_IA32_VMX_PROCBASED_CTLS2 = %#RX64\n", pVM->hm.s.vmx.Msrs.VmxProcCtls2.u));
1399 val = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1;
1400 zap = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0;
1401 HMVMX_REPORT_FEATURE(val, zap, "VIRT_APIC", VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC);
1402 HMVMX_REPORT_FEATURE(val, zap, "EPT", VMX_VMCS_CTRL_PROC_EXEC2_EPT);
1403 HMVMX_REPORT_FEATURE(val, zap, "DESCRIPTOR_TABLE_EXIT", VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT);
1404 HMVMX_REPORT_FEATURE(val, zap, "RDTSCP", VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP);
1405 HMVMX_REPORT_FEATURE(val, zap, "VIRT_X2APIC", VMX_VMCS_CTRL_PROC_EXEC2_VIRT_X2APIC);
1406 HMVMX_REPORT_FEATURE(val, zap, "VPID", VMX_VMCS_CTRL_PROC_EXEC2_VPID);
1407 HMVMX_REPORT_FEATURE(val, zap, "WBINVD_EXIT", VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT);
1408 HMVMX_REPORT_FEATURE(val, zap, "UNRESTRICTED_GUEST", VMX_VMCS_CTRL_PROC_EXEC2_UNRESTRICTED_GUEST);
1409 HMVMX_REPORT_FEATURE(val, zap, "APIC_REG_VIRT", VMX_VMCS_CTRL_PROC_EXEC2_APIC_REG_VIRT);
1410 HMVMX_REPORT_FEATURE(val, zap, "VIRT_INTR_DELIVERY", VMX_VMCS_CTRL_PROC_EXEC2_VIRT_INTR_DELIVERY);
1411 HMVMX_REPORT_FEATURE(val, zap, "PAUSE_LOOP_EXIT", VMX_VMCS_CTRL_PROC_EXEC2_PAUSE_LOOP_EXIT);
1412 HMVMX_REPORT_FEATURE(val, zap, "RDRAND_EXIT", VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT);
1413 HMVMX_REPORT_FEATURE(val, zap, "INVPCID", VMX_VMCS_CTRL_PROC_EXEC2_INVPCID);
1414 HMVMX_REPORT_FEATURE(val, zap, "VMFUNC", VMX_VMCS_CTRL_PROC_EXEC2_VMFUNC);
1415 HMVMX_REPORT_FEATURE(val, zap, "VMCS_SHADOWING", VMX_VMCS_CTRL_PROC_EXEC2_VMCS_SHADOWING);
1416 HMVMX_REPORT_FEATURE(val, zap, "ENCLS_EXIT", VMX_VMCS_CTRL_PROC_EXEC2_ENCLS_EXIT);
1417 HMVMX_REPORT_FEATURE(val, zap, "RDSEED_EXIT", VMX_VMCS_CTRL_PROC_EXEC2_RDSEED_EXIT);
1418 HMVMX_REPORT_FEATURE(val, zap, "PML", VMX_VMCS_CTRL_PROC_EXEC2_PML);
1419 HMVMX_REPORT_FEATURE(val, zap, "EPT_VE", VMX_VMCS_CTRL_PROC_EXEC2_EPT_VE);
1420 HMVMX_REPORT_FEATURE(val, zap, "CONCEAL_FROM_PT", VMX_VMCS_CTRL_PROC_EXEC2_CONCEAL_FROM_PT);
1421 HMVMX_REPORT_FEATURE(val, zap, "XSAVES_XRSTORS", VMX_VMCS_CTRL_PROC_EXEC2_XSAVES_XRSTORS);
1422 HMVMX_REPORT_FEATURE(val, zap, "TSC_SCALING", VMX_VMCS_CTRL_PROC_EXEC2_TSC_SCALING);
1423 }
1424
1425 LogRel(("HM: MSR_IA32_VMX_ENTRY_CTLS = %#RX64\n", pVM->hm.s.vmx.Msrs.VmxEntry.u));
1426 val = pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1;
1427 zap = pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0;
1428 HMVMX_REPORT_FEATURE(val, zap, "LOAD_DEBUG", VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG);
1429 HMVMX_REPORT_FEATURE(val, zap, "IA32E_MODE_GUEST", VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST);
1430 HMVMX_REPORT_FEATURE(val, zap, "ENTRY_SMM", VMX_VMCS_CTRL_ENTRY_ENTRY_SMM);
1431 HMVMX_REPORT_FEATURE(val, zap, "DEACTIVATE_DUALMON", VMX_VMCS_CTRL_ENTRY_DEACTIVATE_DUALMON);
1432 HMVMX_REPORT_FEATURE(val, zap, "LOAD_GUEST_PERF_MSR", VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR);
1433 HMVMX_REPORT_FEATURE(val, zap, "LOAD_GUEST_PAT_MSR", VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR);
1434 HMVMX_REPORT_FEATURE(val, zap, "LOAD_GUEST_EFER_MSR", VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR);
1435
1436 LogRel(("HM: MSR_IA32_VMX_EXIT_CTLS = %#RX64\n", pVM->hm.s.vmx.Msrs.VmxExit.u));
1437 val = pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1;
1438 zap = pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0;
1439 HMVMX_REPORT_FEATURE(val, zap, "SAVE_DEBUG", VMX_VMCS_CTRL_EXIT_SAVE_DEBUG);
1440 HMVMX_REPORT_FEATURE(val, zap, "HOST_ADDR_SPACE_SIZE", VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE);
1441 HMVMX_REPORT_FEATURE(val, zap, "LOAD_PERF_MSR", VMX_VMCS_CTRL_EXIT_LOAD_PERF_MSR);
1442 HMVMX_REPORT_FEATURE(val, zap, "ACK_EXT_INT", VMX_VMCS_CTRL_EXIT_ACK_EXT_INT);
1443 HMVMX_REPORT_FEATURE(val, zap, "SAVE_GUEST_PAT_MSR", VMX_VMCS_CTRL_EXIT_SAVE_GUEST_PAT_MSR);
1444 HMVMX_REPORT_FEATURE(val, zap, "LOAD_HOST_PAT_MSR", VMX_VMCS_CTRL_EXIT_LOAD_HOST_PAT_MSR);
1445 HMVMX_REPORT_FEATURE(val, zap, "SAVE_GUEST_EFER_MSR", VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR);
1446 HMVMX_REPORT_FEATURE(val, zap, "LOAD_HOST_EFER_MSR", VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR);
1447 HMVMX_REPORT_FEATURE(val, zap, "SAVE_VMX_PREEMPT_TIMER", VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER);
1448
1449 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps)
1450 {
1451 val = pVM->hm.s.vmx.Msrs.u64EptVpidCaps;
1452 LogRel(("HM: MSR_IA32_VMX_EPT_VPID_CAP = %#RX64\n", val));
1453 HMVMX_REPORT_MSR_CAPABILITY(val, "RWX_X_ONLY", MSR_IA32_VMX_EPT_VPID_CAP_RWX_X_ONLY);
1454 HMVMX_REPORT_MSR_CAPABILITY(val, "PAGE_WALK_LENGTH_4", MSR_IA32_VMX_EPT_VPID_CAP_PAGE_WALK_LENGTH_4);
1455 HMVMX_REPORT_MSR_CAPABILITY(val, "EMT_UC", MSR_IA32_VMX_EPT_VPID_CAP_EMT_UC);
1456 HMVMX_REPORT_MSR_CAPABILITY(val, "EMT_WB", MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB);
1457 HMVMX_REPORT_MSR_CAPABILITY(val, "PDE_2M", MSR_IA32_VMX_EPT_VPID_CAP_PDE_2M);
1458 HMVMX_REPORT_MSR_CAPABILITY(val, "PDPTE_1G", MSR_IA32_VMX_EPT_VPID_CAP_PDPTE_1G);
1459 HMVMX_REPORT_MSR_CAPABILITY(val, "INVEPT", MSR_IA32_VMX_EPT_VPID_CAP_INVEPT);
1460 HMVMX_REPORT_MSR_CAPABILITY(val, "EPT_ACCESS_DIRTY", MSR_IA32_VMX_EPT_VPID_CAP_EPT_ACCESS_DIRTY);
1461 HMVMX_REPORT_MSR_CAPABILITY(val, "INVEPT_SINGLE_CONTEXT", MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT);
1462 HMVMX_REPORT_MSR_CAPABILITY(val, "INVEPT_ALL_CONTEXTS", MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS);
1463 HMVMX_REPORT_MSR_CAPABILITY(val, "INVVPID", MSR_IA32_VMX_EPT_VPID_CAP_INVVPID);
1464 HMVMX_REPORT_MSR_CAPABILITY(val, "INVVPID_INDIV_ADDR", MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR);
1465 HMVMX_REPORT_MSR_CAPABILITY(val, "INVVPID_SINGLE_CONTEXT", MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT);
1466 HMVMX_REPORT_MSR_CAPABILITY(val, "INVVPID_ALL_CONTEXTS", MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS);
1467 HMVMX_REPORT_MSR_CAPABILITY(val, "INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS", MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS);
1468 }
1469
1470 val = pVM->hm.s.vmx.Msrs.u64Misc;
1471 LogRel(("HM: MSR_IA32_VMX_MISC = %#RX64\n", val));
1472 if (MSR_IA32_VMX_MISC_PREEMPT_TSC_BIT(val) == pVM->hm.s.vmx.cPreemptTimerShift)
1473 LogRel(("HM: PREEMPT_TSC_BIT = %#x\n", MSR_IA32_VMX_MISC_PREEMPT_TSC_BIT(val)));
1474 else
1475 {
1476 LogRel(("HM: PREEMPT_TSC_BIT = %#x - erratum detected, using %#x instead\n",
1477 MSR_IA32_VMX_MISC_PREEMPT_TSC_BIT(val), pVM->hm.s.vmx.cPreemptTimerShift));
1478 }
1479
1480 LogRel(("HM: STORE_EFERLMA_VMEXIT = %RTbool\n", RT_BOOL(MSR_IA32_VMX_MISC_STORE_EFERLMA_VMEXIT(val))));
1481 LogRel(("HM: ACTIVITY_STATES = %#x\n", MSR_IA32_VMX_MISC_ACTIVITY_STATES(val)));
1482 LogRel(("HM: CR3_TARGET = %#x\n", MSR_IA32_VMX_MISC_CR3_TARGET(val)));
1483 LogRel(("HM: MAX_MSR = %u\n", MSR_IA32_VMX_MISC_MAX_MSR(val)));
1484 LogRel(("HM: RDMSR_SMBASE_MSR_SMM = %RTbool\n", RT_BOOL(MSR_IA32_VMX_MISC_RDMSR_SMBASE_MSR_SMM(val))));
1485 LogRel(("HM: SMM_MONITOR_CTL_B2 = %RTbool\n", RT_BOOL(MSR_IA32_VMX_MISC_SMM_MONITOR_CTL_B2(val))));
1486 LogRel(("HM: VMWRITE_VMEXIT_INFO = %RTbool\n", RT_BOOL(MSR_IA32_VMX_MISC_VMWRITE_VMEXIT_INFO(val))));
1487 LogRel(("HM: MSEG_ID = %#x\n", MSR_IA32_VMX_MISC_MSEG_ID(val)));
1488
1489 /* Paranoia */
1490 AssertRelease(MSR_IA32_VMX_MISC_MAX_MSR(pVM->hm.s.vmx.Msrs.u64Misc) >= 512);
1491
1492 LogRel(("HM: MSR_IA32_VMX_CR0_FIXED0 = %#RX64\n", pVM->hm.s.vmx.Msrs.u64Cr0Fixed0));
1493 LogRel(("HM: MSR_IA32_VMX_CR0_FIXED1 = %#RX64\n", pVM->hm.s.vmx.Msrs.u64Cr0Fixed1));
1494 LogRel(("HM: MSR_IA32_VMX_CR4_FIXED0 = %#RX64\n", pVM->hm.s.vmx.Msrs.u64Cr4Fixed0));
1495 LogRel(("HM: MSR_IA32_VMX_CR4_FIXED1 = %#RX64\n", pVM->hm.s.vmx.Msrs.u64Cr4Fixed1));
1496
1497 val = pVM->hm.s.vmx.Msrs.u64VmcsEnum;
1498 LogRel(("HM: MSR_IA32_VMX_VMCS_ENUM = %#RX64\n", val));
1499 LogRel(("HM: HIGHEST_INDEX = %#x\n", MSR_IA32_VMX_VMCS_ENUM_HIGHEST_INDEX(val)));
1500
1501 val = pVM->hm.s.vmx.Msrs.u64Vmfunc;
1502 if (val)
1503 {
1504 LogRel(("HM: MSR_IA32_VMX_VMFUNC = %#RX64\n", val));
1505 HMVMX_REPORT_ALLOWED_FEATURE(val, "EPTP_SWITCHING", VMX_VMCS_CTRL_VMFUNC_EPTP_SWITCHING);
1506 }
1507
1508 LogRel(("HM: APIC-access page physaddr = %#RHp\n", pVM->hm.s.vmx.HCPhysApicAccess));
1509
1510 for (VMCPUID i = 0; i < pVM->cCpus; i++)
1511 {
1512 LogRel(("HM: VCPU%3d: MSR bitmap physaddr = %#RHp\n", i, pVM->aCpus[i].hm.s.vmx.HCPhysMsrBitmap));
1513 LogRel(("HM: VCPU%3d: VMCS physaddr = %#RHp\n", i, pVM->aCpus[i].hm.s.vmx.HCPhysVmcs));
1514 }
1515
1516 /*
1517 * EPT and unhampered guest execution are determined in HMR3Init, verify the sanity of that.
1518 */
1519 AssertLogRelReturn( !pVM->hm.s.fNestedPaging
1520 || (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_EPT),
1521 VERR_HM_IPE_1);
1522 AssertLogRelReturn( !pVM->hm.s.vmx.fUnrestrictedGuest
1523 || ( (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_UNRESTRICTED_GUEST)
1524 && pVM->hm.s.fNestedPaging),
1525 VERR_HM_IPE_1);
1526
1527 /*
1528 * Enable VPID if configured and supported.
1529 */
1530 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VPID)
1531 pVM->hm.s.vmx.fVpid = pVM->hm.s.vmx.fAllowVpid;
1532
1533#if 0
1534 /*
1535 * Enable APIC register virtualization and virtual-interrupt delivery if supported.
1536 */
1537 if ( (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_APIC_REG_VIRT)
1538 && (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_INTR_DELIVERY))
1539 pVM->hm.s.fVirtApicRegs = true;
1540
1541 /*
1542 * Enable posted-interrupt processing if supported.
1543 */
1544 /** @todo Add and query IPRT API for host OS support for posted-interrupt IPI
1545 * here. */
1546 if ( (pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_POSTED_INTR)
1547 && (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT))
1548 pVM->hm.s.fPostedIntrs = true;
1549#endif
1550
1551 /*
1552 * Disallow RDTSCP in the guest if there is no secondary process-based VM execution controls as otherwise
1553 * RDTSCP would cause a #UD. There might be no CPUs out there where this happens, as RDTSCP was introduced
1554 * in Nehalems and secondary VM exec. controls should be supported in all of them, but nonetheless it's Intel...
1555 */
1556 if ( !(pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
1557 && CPUMR3GetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_RDTSCP))
1558 {
1559 CPUMR3ClearGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_RDTSCP);
1560 LogRel(("HM: Disabled RDTSCP\n"));
1561 }
1562
1563 if (!pVM->hm.s.vmx.fUnrestrictedGuest)
1564 {
1565 /* Allocate three pages for the TSS we need for real mode emulation. (2 pages for the IO bitmap) */
1566 rc = PDMR3VmmDevHeapAlloc(pVM, HM_VTX_TOTAL_DEVHEAP_MEM, hmR3VmmDevHeapNotify, (RTR3PTR *)&pVM->hm.s.vmx.pRealModeTSS);
1567 if (RT_SUCCESS(rc))
1568 {
1569 /* The IO bitmap starts right after the virtual interrupt redirection bitmap.
1570 Refer Intel spec. 20.3.3 "Software Interrupt Handling in Virtual-8086 mode"
1571 esp. Figure 20-5.*/
1572 ASMMemZero32(pVM->hm.s.vmx.pRealModeTSS, sizeof(*pVM->hm.s.vmx.pRealModeTSS));
1573 pVM->hm.s.vmx.pRealModeTSS->offIoBitmap = sizeof(*pVM->hm.s.vmx.pRealModeTSS);
1574
1575 /* Bit set to 0 means software interrupts are redirected to the
1576 8086 program interrupt handler rather than switching to
1577 protected-mode handler. */
1578 memset(pVM->hm.s.vmx.pRealModeTSS->IntRedirBitmap, 0, sizeof(pVM->hm.s.vmx.pRealModeTSS->IntRedirBitmap));
1579
1580 /* Allow all port IO, so that port IO instructions do not cause
1581 exceptions and would instead cause a VM-exit (based on VT-x's
1582 IO bitmap which we currently configure to always cause an exit). */
1583 memset(pVM->hm.s.vmx.pRealModeTSS + 1, 0, PAGE_SIZE * 2);
1584 *((unsigned char *)pVM->hm.s.vmx.pRealModeTSS + HM_VTX_TSS_SIZE - 2) = 0xff;
1585
1586 /*
1587 * Construct a 1024 element page directory with 4 MB pages for
1588 * the identity mapped page table used in real and protected mode
1589 * without paging with EPT.
1590 */
1591 pVM->hm.s.vmx.pNonPagingModeEPTPageTable = (PX86PD)((char *)pVM->hm.s.vmx.pRealModeTSS + PAGE_SIZE * 3);
1592 for (uint32_t i = 0; i < X86_PG_ENTRIES; i++)
1593 {
1594 pVM->hm.s.vmx.pNonPagingModeEPTPageTable->a[i].u = _4M * i;
1595 pVM->hm.s.vmx.pNonPagingModeEPTPageTable->a[i].u |= X86_PDE4M_P | X86_PDE4M_RW | X86_PDE4M_US
1596 | X86_PDE4M_A | X86_PDE4M_D | X86_PDE4M_PS
1597 | X86_PDE4M_G;
1598 }
1599
1600 /* We convert it here every time as PCI regions could be reconfigured. */
1601 if (PDMVmmDevHeapIsEnabled(pVM))
1602 {
1603 RTGCPHYS GCPhys;
1604 rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
1605 AssertRCReturn(rc, rc);
1606 LogRel(("HM: Real Mode TSS guest physaddr = %#RGp\n", GCPhys));
1607
1608 rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
1609 AssertRCReturn(rc, rc);
1610 LogRel(("HM: Non-Paging Mode EPT CR3 = %#RGp\n", GCPhys));
1611 }
1612 }
1613 else
1614 {
1615 LogRel(("HM: No real mode VT-x support (PDMR3VMMDevHeapAlloc returned %Rrc)\n", rc));
1616 pVM->hm.s.vmx.pRealModeTSS = NULL;
1617 pVM->hm.s.vmx.pNonPagingModeEPTPageTable = NULL;
1618 return VMSetError(pVM, rc, RT_SRC_POS,
1619 "HM failure: No real mode VT-x support (PDMR3VMMDevHeapAlloc returned %Rrc)", rc);
1620 }
1621 }
1622
1623 LogRel((pVM->hm.s.fAllow64BitGuests
1624 ? "HM: Guest support: 32-bit and 64-bit\n"
1625 : "HM: Guest support: 32-bit only\n"));
1626
1627 /*
1628 * Call ring-0 to set up the VM.
1629 */
1630 rc = SUPR3CallVMMR0Ex(pVM->pVMR0, 0 /* idCpu */, VMMR0_DO_HM_SETUP_VM, 0 /* u64Arg */, NULL /* pReqHdr */);
1631 if (rc != VINF_SUCCESS)
1632 {
1633 AssertMsgFailed(("%Rrc\n", rc));
1634 LogRel(("HM: VMX setup failed with rc=%Rrc!\n", rc));
1635 for (VMCPUID i = 0; i < pVM->cCpus; i++)
1636 {
1637 PVMCPU pVCpu = &pVM->aCpus[i];
1638 LogRel(("HM: CPU[%u] Last instruction error %#x\n", i, pVCpu->hm.s.vmx.LastError.u32InstrError));
1639 LogRel(("HM: CPU[%u] HM error %#x (%u)\n", i, pVCpu->hm.s.u32HMError, pVCpu->hm.s.u32HMError));
1640 }
1641 HMR3CheckError(pVM, rc);
1642 return VMSetError(pVM, rc, RT_SRC_POS, "VT-x setup failed: %Rrc", rc);
1643 }
1644
1645 LogRel(("HM: Supports VMCS EFER fields = %RTbool\n", pVM->hm.s.vmx.fSupportsVmcsEfer));
1646 LogRel(("HM: Enabled VMX\n"));
1647 pVM->hm.s.vmx.fEnabled = true;
1648
1649 hmR3DisableRawMode(pVM); /** @todo make this go away! */
1650
1651 /*
1652 * Change the CPU features.
1653 */
1654 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_SEP);
1655 if (pVM->hm.s.fAllow64BitGuests)
1656 {
1657 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_PAE);
1658 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_LONG_MODE);
1659 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_SYSCALL); /* 64 bits only on Intel CPUs */
1660 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_LAHF);
1661 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_NX);
1662 }
1663 /* Turn on NXE if PAE has been enabled *and* the host has turned on NXE
1664 (we reuse the host EFER in the switcher). */
1665 /** @todo this needs to be fixed properly!! */
1666 else if (CPUMR3GetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_PAE))
1667 {
1668 if (pVM->hm.s.vmx.u64HostEfer & MSR_K6_EFER_NXE)
1669 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_NX);
1670 else
1671 LogRel(("HM: NX not enabled on the host, unavailable to PAE guest\n"));
1672 }
1673
1674 /*
1675 * Log configuration details.
1676 */
1677 if (pVM->hm.s.fNestedPaging)
1678 {
1679 LogRel(("HM: Enabled nested paging\n"));
1680 if (pVM->hm.s.vmx.enmFlushEpt == VMXFLUSHEPT_SINGLE_CONTEXT)
1681 LogRel(("HM: EPT flush type = VMXFLUSHEPT_SINGLE_CONTEXT\n"));
1682 else if (pVM->hm.s.vmx.enmFlushEpt == VMXFLUSHEPT_ALL_CONTEXTS)
1683 LogRel(("HM: EPT flush type = VMXFLUSHEPT_ALL_CONTEXTS\n"));
1684 else if (pVM->hm.s.vmx.enmFlushEpt == VMXFLUSHEPT_NOT_SUPPORTED)
1685 LogRel(("HM: EPT flush type = VMXFLUSHEPT_NOT_SUPPORTED\n"));
1686 else
1687 LogRel(("HM: EPT flush type = %d\n", pVM->hm.s.vmx.enmFlushEpt));
1688
1689 if (pVM->hm.s.vmx.fUnrestrictedGuest)
1690 LogRel(("HM: Enabled unrestricted guest execution\n"));
1691
1692#if HC_ARCH_BITS == 64
1693 if (pVM->hm.s.fLargePages)
1694 {
1695 /* Use large (2 MB) pages for our EPT PDEs where possible. */
1696 PGMSetLargePageUsage(pVM, true);
1697 LogRel(("HM: Enabled large page support\n"));
1698 }
1699#endif
1700 }
1701 else
1702 Assert(!pVM->hm.s.vmx.fUnrestrictedGuest);
1703
1704 if (pVM->hm.s.fVirtApicRegs)
1705 LogRel(("HM: Enabled APIC-register virtualization support\n"));
1706
1707 if (pVM->hm.s.fPostedIntrs)
1708 LogRel(("HM: Enabled posted-interrupt processing support\n"));
1709
1710 if (pVM->hm.s.vmx.fVpid)
1711 {
1712 LogRel(("HM: Enabled VPID\n"));
1713 if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_INDIV_ADDR)
1714 LogRel(("HM: VPID flush type = VMXFLUSHVPID_INDIV_ADDR\n"));
1715 else if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_SINGLE_CONTEXT)
1716 LogRel(("HM: VPID flush type = VMXFLUSHVPID_SINGLE_CONTEXT\n"));
1717 else if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_ALL_CONTEXTS)
1718 LogRel(("HM: VPID flush type = VMXFLUSHVPID_ALL_CONTEXTS\n"));
1719 else if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
1720 LogRel(("HM: VPID flush type = VMXFLUSHVPID_SINGLE_CONTEXT_RETAIN_GLOBALS\n"));
1721 else
1722 LogRel(("HM: VPID flush type = %d\n", pVM->hm.s.vmx.enmFlushVpid));
1723 }
1724 else if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_NOT_SUPPORTED)
1725 LogRel(("HM: Ignoring VPID capabilities of CPU\n"));
1726
1727 if (pVM->hm.s.vmx.fUsePreemptTimer)
1728 LogRel(("HM: Enabled VMX-preemption timer (cPreemptTimerShift=%u)\n", pVM->hm.s.vmx.cPreemptTimerShift));
1729 else
1730 LogRel(("HM: Disabled VMX-preemption timer\n"));
1731
1732 return VINF_SUCCESS;
1733}
1734
1735
1736/**
1737 * Finish AMD-V initialization (after ring-0 init).
1738 *
1739 * @returns VBox status code.
1740 * @param pVM The cross context VM structure.
1741 */
1742static int hmR3InitFinalizeR0Amd(PVM pVM)
1743{
1744 Log(("pVM->hm.s.svm.fSupported = %d\n", pVM->hm.s.svm.fSupported));
1745
1746 LogRel(("HM: Using AMD-V implementation 2.0\n"));
1747
1748 uint32_t u32Family;
1749 uint32_t u32Model;
1750 uint32_t u32Stepping;
1751 if (HMAmdIsSubjectToErratum170(&u32Family, &u32Model, &u32Stepping))
1752 LogRel(("HM: AMD Cpu with erratum 170 family %#x model %#x stepping %#x\n", u32Family, u32Model, u32Stepping));
1753 LogRel(("HM: Max resume loops = %u\n", pVM->hm.s.cMaxResumeLoops));
1754 LogRel(("HM: CPUID 0x80000001.u32AMDFeatureECX = %#RX32\n", pVM->hm.s.cpuid.u32AMDFeatureECX));
1755 LogRel(("HM: CPUID 0x80000001.u32AMDFeatureEDX = %#RX32\n", pVM->hm.s.cpuid.u32AMDFeatureEDX));
1756 LogRel(("HM: AMD HWCR MSR = %#RX64\n", pVM->hm.s.svm.u64MsrHwcr));
1757 LogRel(("HM: AMD-V revision = %#x\n", pVM->hm.s.svm.u32Rev));
1758 LogRel(("HM: AMD-V max ASID = %RU32\n", pVM->hm.s.uMaxAsid));
1759 LogRel(("HM: AMD-V features = %#x\n", pVM->hm.s.svm.u32Features));
1760
1761 /*
1762 * Enumerate AMD-V features.
1763 */
1764 static const struct { uint32_t fFlag; const char *pszName; } s_aSvmFeatures[] =
1765 {
1766#define HMSVM_REPORT_FEATURE(a_StrDesc, a_Define) { a_Define, a_StrDesc }
1767 HMSVM_REPORT_FEATURE("NESTED_PAGING", X86_CPUID_SVM_FEATURE_EDX_NESTED_PAGING),
1768 HMSVM_REPORT_FEATURE("LBR_VIRT", X86_CPUID_SVM_FEATURE_EDX_LBR_VIRT),
1769 HMSVM_REPORT_FEATURE("SVM_LOCK", X86_CPUID_SVM_FEATURE_EDX_SVM_LOCK),
1770 HMSVM_REPORT_FEATURE("NRIP_SAVE", X86_CPUID_SVM_FEATURE_EDX_NRIP_SAVE),
1771 HMSVM_REPORT_FEATURE("TSC_RATE_MSR", X86_CPUID_SVM_FEATURE_EDX_TSC_RATE_MSR),
1772 HMSVM_REPORT_FEATURE("VMCB_CLEAN", X86_CPUID_SVM_FEATURE_EDX_VMCB_CLEAN),
1773 HMSVM_REPORT_FEATURE("FLUSH_BY_ASID", X86_CPUID_SVM_FEATURE_EDX_FLUSH_BY_ASID),
1774 HMSVM_REPORT_FEATURE("DECODE_ASSISTS", X86_CPUID_SVM_FEATURE_EDX_DECODE_ASSISTS),
1775 HMSVM_REPORT_FEATURE("PAUSE_FILTER", X86_CPUID_SVM_FEATURE_EDX_PAUSE_FILTER),
1776 HMSVM_REPORT_FEATURE("PAUSE_FILTER_THRESHOLD", X86_CPUID_SVM_FEATURE_EDX_PAUSE_FILTER_THRESHOLD),
1777 HMSVM_REPORT_FEATURE("AVIC", X86_CPUID_SVM_FEATURE_EDX_AVIC),
1778 HMSVM_REPORT_FEATURE("VIRT_VMSAVE_VMLOAD", X86_CPUID_SVM_FEATURE_EDX_VIRT_VMSAVE_VMLOAD),
1779 HMSVM_REPORT_FEATURE("VGIF", X86_CPUID_SVM_FEATURE_EDX_VGIF),
1780#undef HMSVM_REPORT_FEATURE
1781 };
1782
1783 uint32_t fSvmFeatures = pVM->hm.s.svm.u32Features;
1784 for (unsigned i = 0; i < RT_ELEMENTS(s_aSvmFeatures); i++)
1785 if (fSvmFeatures & s_aSvmFeatures[i].fFlag)
1786 {
1787 LogRel(("HM: %s\n", s_aSvmFeatures[i].pszName));
1788 fSvmFeatures &= ~s_aSvmFeatures[i].fFlag;
1789 }
1790 if (fSvmFeatures)
1791 for (unsigned iBit = 0; iBit < 32; iBit++)
1792 if (RT_BIT_32(iBit) & fSvmFeatures)
1793 LogRel(("HM: Reserved bit %u\n", iBit));
1794
1795 /*
1796 * Nested paging is determined in HMR3Init, verify the sanity of that.
1797 */
1798 AssertLogRelReturn( !pVM->hm.s.fNestedPaging
1799 || (pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_NESTED_PAGING),
1800 VERR_HM_IPE_1);
1801
1802#if 0
1803 /** @todo Add and query IPRT API for host OS support for posted-interrupt IPI
1804 * here. */
1805 if (RTR0IsPostIpiSupport())
1806 pVM->hm.s.fPostedIntrs = true;
1807#endif
1808
1809 /*
1810 * Call ring-0 to set up the VM.
1811 */
1812 int rc = SUPR3CallVMMR0Ex(pVM->pVMR0, 0 /*idCpu*/, VMMR0_DO_HM_SETUP_VM, 0, NULL);
1813 if (rc != VINF_SUCCESS)
1814 {
1815 AssertMsgFailed(("%Rrc\n", rc));
1816 LogRel(("HM: AMD-V setup failed with rc=%Rrc!\n", rc));
1817 return VMSetError(pVM, rc, RT_SRC_POS, "AMD-V setup failed: %Rrc", rc);
1818 }
1819
1820 LogRel(("HM: Enabled SVM\n"));
1821 pVM->hm.s.svm.fEnabled = true;
1822
1823 if (pVM->hm.s.fNestedPaging)
1824 {
1825 LogRel(("HM: Enabled nested paging\n"));
1826
1827 /*
1828 * Enable large pages (2 MB) if applicable.
1829 */
1830#if HC_ARCH_BITS == 64
1831 if (pVM->hm.s.fLargePages)
1832 {
1833 PGMSetLargePageUsage(pVM, true);
1834 LogRel(("HM: Enabled large page support\n"));
1835 }
1836#endif
1837 }
1838
1839 if (pVM->hm.s.fVirtApicRegs)
1840 LogRel(("HM: Enabled APIC-register virtualization support\n"));
1841
1842 if (pVM->hm.s.fPostedIntrs)
1843 LogRel(("HM: Enabled posted-interrupt processing support\n"));
1844
1845 hmR3DisableRawMode(pVM);
1846
1847 /*
1848 * Change the CPU features.
1849 */
1850 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_SEP);
1851 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_SYSCALL);
1852 if (pVM->hm.s.fAllow64BitGuests)
1853 {
1854 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_PAE);
1855 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_LONG_MODE);
1856 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_NX);
1857 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_LAHF);
1858 }
1859 /* Turn on NXE if PAE has been enabled. */
1860 else if (CPUMR3GetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_PAE))
1861 CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_NX);
1862
1863 LogRel(("HM: %s TPR patching\n", (pVM->hm.s.fTprPatchingAllowed) ? "Enabled" : "Disabled"));
1864
1865 LogRel((pVM->hm.s.fAllow64BitGuests
1866 ? "HM: Guest support: 32-bit and 64-bit\n"
1867 : "HM: Guest support: 32-bit only\n"));
1868
1869 return VINF_SUCCESS;
1870}
1871
1872
1873/**
1874 * Applies relocations to data and code managed by this
1875 * component. This function will be called at init and
1876 * whenever the VMM need to relocate it self inside the GC.
1877 *
1878 * @param pVM The cross context VM structure.
1879 */
1880VMMR3_INT_DECL(void) HMR3Relocate(PVM pVM)
1881{
1882 Log(("HMR3Relocate to %RGv\n", MMHyperGetArea(pVM, 0)));
1883
1884 /* Fetch the current paging mode during the relocate callback during state loading. */
1885 if (VMR3GetState(pVM) == VMSTATE_LOADING)
1886 {
1887 for (VMCPUID i = 0; i < pVM->cCpus; i++)
1888 {
1889 PVMCPU pVCpu = &pVM->aCpus[i];
1890 pVCpu->hm.s.enmShadowMode = PGMGetShadowMode(pVCpu);
1891 }
1892 }
1893#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
1894 if (HMIsEnabled(pVM))
1895 {
1896 switch (PGMGetHostMode(pVM))
1897 {
1898 case PGMMODE_32_BIT:
1899 pVM->hm.s.pfnHost32ToGuest64R0 = VMMR3GetHostToGuestSwitcher(pVM, VMMSWITCHER_32_TO_AMD64);
1900 break;
1901
1902 case PGMMODE_PAE:
1903 case PGMMODE_PAE_NX:
1904 pVM->hm.s.pfnHost32ToGuest64R0 = VMMR3GetHostToGuestSwitcher(pVM, VMMSWITCHER_PAE_TO_AMD64);
1905 break;
1906
1907 default:
1908 AssertFailed();
1909 break;
1910 }
1911 }
1912#endif
1913 return;
1914}
1915
1916
1917/**
1918 * Notification callback which is called whenever there is a chance that a CR3
1919 * value might have changed.
1920 *
1921 * This is called by PGM.
1922 *
1923 * @param pVM The cross context VM structure.
1924 * @param pVCpu The cross context virtual CPU structure.
1925 * @param enmShadowMode New shadow paging mode.
1926 * @param enmGuestMode New guest paging mode.
1927 */
1928VMMR3_INT_DECL(void) HMR3PagingModeChanged(PVM pVM, PVMCPU pVCpu, PGMMODE enmShadowMode, PGMMODE enmGuestMode)
1929{
1930 RT_NOREF_PV(pVM);
1931
1932 /* Ignore page mode changes during state loading. */
1933 if (VMR3GetState(pVCpu->pVMR3) == VMSTATE_LOADING)
1934 return;
1935
1936 pVCpu->hm.s.enmShadowMode = enmShadowMode;
1937
1938 /*
1939 * If the guest left protected mode VMX execution, we'll have to be
1940 * extra careful if/when the guest switches back to protected mode.
1941 */
1942 if (enmGuestMode == PGMMODE_REAL)
1943 pVCpu->hm.s.vmx.fWasInRealMode = true;
1944
1945 Log4(("HMR3PagingModeChanged: Guest paging mode '%s', shadow paging mode '%s'\n", PGMGetModeName(enmGuestMode),
1946 PGMGetModeName(enmShadowMode)));
1947}
1948
1949
1950/**
1951 * Terminates the HM.
1952 *
1953 * Termination means cleaning up and freeing all resources,
1954 * the VM itself is, at this point, powered off or suspended.
1955 *
1956 * @returns VBox status code.
1957 * @param pVM The cross context VM structure.
1958 */
1959VMMR3_INT_DECL(int) HMR3Term(PVM pVM)
1960{
1961 if (pVM->hm.s.vmx.pRealModeTSS)
1962 {
1963 PDMR3VmmDevHeapFree(pVM, pVM->hm.s.vmx.pRealModeTSS);
1964 pVM->hm.s.vmx.pRealModeTSS = 0;
1965 }
1966 hmR3TermCPU(pVM);
1967 return 0;
1968}
1969
1970
1971/**
1972 * Terminates the per-VCPU HM.
1973 *
1974 * @returns VBox status code.
1975 * @param pVM The cross context VM structure.
1976 */
1977static int hmR3TermCPU(PVM pVM)
1978{
1979 for (VMCPUID i = 0; i < pVM->cCpus; i++)
1980 {
1981 PVMCPU pVCpu = &pVM->aCpus[i]; NOREF(pVCpu);
1982
1983#ifdef VBOX_WITH_STATISTICS
1984 if (pVCpu->hm.s.paStatExitReason)
1985 {
1986 MMHyperFree(pVM, pVCpu->hm.s.paStatExitReason);
1987 pVCpu->hm.s.paStatExitReason = NULL;
1988 pVCpu->hm.s.paStatExitReasonR0 = NIL_RTR0PTR;
1989 }
1990 if (pVCpu->hm.s.paStatInjectedIrqs)
1991 {
1992 MMHyperFree(pVM, pVCpu->hm.s.paStatInjectedIrqs);
1993 pVCpu->hm.s.paStatInjectedIrqs = NULL;
1994 pVCpu->hm.s.paStatInjectedIrqsR0 = NIL_RTR0PTR;
1995 }
1996#endif
1997
1998#ifdef VBOX_WITH_CRASHDUMP_MAGIC
1999 memset(pVCpu->hm.s.vmx.VMCSCache.aMagic, 0, sizeof(pVCpu->hm.s.vmx.VMCSCache.aMagic));
2000 pVCpu->hm.s.vmx.VMCSCache.uMagic = 0;
2001 pVCpu->hm.s.vmx.VMCSCache.uPos = 0xffffffff;
2002#endif
2003 }
2004 return 0;
2005}
2006
2007
2008/**
2009 * Resets a virtual CPU.
2010 *
2011 * Used by HMR3Reset and CPU hot plugging.
2012 *
2013 * @param pVCpu The cross context virtual CPU structure to reset.
2014 */
2015VMMR3_INT_DECL(void) HMR3ResetCpu(PVMCPU pVCpu)
2016{
2017 /* Sync. entire state on VM reset R0-reentry. It's safe to reset
2018 the HM flags here, all other EMTs are in ring-3. See VMR3Reset(). */
2019 HMCPU_CF_RESET_TO(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_ALL_GUEST);
2020
2021 pVCpu->hm.s.vmx.u32CR0Mask = 0;
2022 pVCpu->hm.s.vmx.u32CR4Mask = 0;
2023 pVCpu->hm.s.fActive = false;
2024 pVCpu->hm.s.Event.fPending = false;
2025 pVCpu->hm.s.vmx.fWasInRealMode = true;
2026 pVCpu->hm.s.vmx.u64MsrApicBase = 0;
2027 pVCpu->hm.s.vmx.fSwitchedTo64on32 = false;
2028
2029 /* Reset the contents of the read cache. */
2030 PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache;
2031 for (unsigned j = 0; j < pCache->Read.cValidEntries; j++)
2032 pCache->Read.aFieldVal[j] = 0;
2033
2034#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2035 /* Magic marker for searching in crash dumps. */
2036 strcpy((char *)pCache->aMagic, "VMCSCACHE Magic");
2037 pCache->uMagic = UINT64_C(0xDEADBEEFDEADBEEF);
2038#endif
2039}
2040
2041
2042/**
2043 * The VM is being reset.
2044 *
2045 * For the HM component this means that any GDT/LDT/TSS monitors
2046 * needs to be removed.
2047 *
2048 * @param pVM The cross context VM structure.
2049 */
2050VMMR3_INT_DECL(void) HMR3Reset(PVM pVM)
2051{
2052 LogFlow(("HMR3Reset:\n"));
2053
2054 if (HMIsEnabled(pVM))
2055 hmR3DisableRawMode(pVM);
2056
2057 for (VMCPUID i = 0; i < pVM->cCpus; i++)
2058 {
2059 PVMCPU pVCpu = &pVM->aCpus[i];
2060
2061 HMR3ResetCpu(pVCpu);
2062 }
2063
2064 /* Clear all patch information. */
2065 pVM->hm.s.pGuestPatchMem = 0;
2066 pVM->hm.s.pFreeGuestPatchMem = 0;
2067 pVM->hm.s.cbGuestPatchMem = 0;
2068 pVM->hm.s.cPatches = 0;
2069 pVM->hm.s.PatchTree = 0;
2070 pVM->hm.s.fTPRPatchingActive = false;
2071 ASMMemZero32(pVM->hm.s.aPatches, sizeof(pVM->hm.s.aPatches));
2072}
2073
2074
2075/**
2076 * Callback to patch a TPR instruction (vmmcall or mov cr8).
2077 *
2078 * @returns VBox strict status code.
2079 * @param pVM The cross context VM structure.
2080 * @param pVCpu The cross context virtual CPU structure of the calling EMT.
2081 * @param pvUser Unused.
2082 */
2083static DECLCALLBACK(VBOXSTRICTRC) hmR3RemovePatches(PVM pVM, PVMCPU pVCpu, void *pvUser)
2084{
2085 VMCPUID idCpu = (VMCPUID)(uintptr_t)pvUser;
2086
2087 /* Only execute the handler on the VCPU the original patch request was issued. */
2088 if (pVCpu->idCpu != idCpu)
2089 return VINF_SUCCESS;
2090
2091 Log(("hmR3RemovePatches\n"));
2092 for (unsigned i = 0; i < pVM->hm.s.cPatches; i++)
2093 {
2094 uint8_t abInstr[15];
2095 PHMTPRPATCH pPatch = &pVM->hm.s.aPatches[i];
2096 RTGCPTR pInstrGC = (RTGCPTR)pPatch->Core.Key;
2097 int rc;
2098
2099#ifdef LOG_ENABLED
2100 char szOutput[256];
2101
2102 rc = DBGFR3DisasInstrEx(pVM->pUVM, pVCpu->idCpu, CPUMGetGuestCS(pVCpu), pInstrGC, DBGF_DISAS_FLAGS_DEFAULT_MODE,
2103 szOutput, sizeof(szOutput), NULL);
2104 if (RT_SUCCESS(rc))
2105 Log(("Patched instr: %s\n", szOutput));
2106#endif
2107
2108 /* Check if the instruction is still the same. */
2109 rc = PGMPhysSimpleReadGCPtr(pVCpu, abInstr, pInstrGC, pPatch->cbNewOp);
2110 if (rc != VINF_SUCCESS)
2111 {
2112 Log(("Patched code removed? (rc=%Rrc0\n", rc));
2113 continue; /* swapped out or otherwise removed; skip it. */
2114 }
2115
2116 if (memcmp(abInstr, pPatch->aNewOpcode, pPatch->cbNewOp))
2117 {
2118 Log(("Patched instruction was changed! (rc=%Rrc0\n", rc));
2119 continue; /* skip it. */
2120 }
2121
2122 rc = PGMPhysSimpleWriteGCPtr(pVCpu, pInstrGC, pPatch->aOpcode, pPatch->cbOp);
2123 AssertRC(rc);
2124
2125#ifdef LOG_ENABLED
2126 rc = DBGFR3DisasInstrEx(pVM->pUVM, pVCpu->idCpu, CPUMGetGuestCS(pVCpu), pInstrGC, DBGF_DISAS_FLAGS_DEFAULT_MODE,
2127 szOutput, sizeof(szOutput), NULL);
2128 if (RT_SUCCESS(rc))
2129 Log(("Original instr: %s\n", szOutput));
2130#endif
2131 }
2132 pVM->hm.s.cPatches = 0;
2133 pVM->hm.s.PatchTree = 0;
2134 pVM->hm.s.pFreeGuestPatchMem = pVM->hm.s.pGuestPatchMem;
2135 pVM->hm.s.fTPRPatchingActive = false;
2136 return VINF_SUCCESS;
2137}
2138
2139
2140/**
2141 * Worker for enabling patching in a VT-x/AMD-V guest.
2142 *
2143 * @returns VBox status code.
2144 * @param pVM The cross context VM structure.
2145 * @param idCpu VCPU to execute hmR3RemovePatches on.
2146 * @param pPatchMem Patch memory range.
2147 * @param cbPatchMem Size of the memory range.
2148 */
2149static int hmR3EnablePatching(PVM pVM, VMCPUID idCpu, RTRCPTR pPatchMem, unsigned cbPatchMem)
2150{
2151 int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONE_BY_ONE, hmR3RemovePatches, (void *)(uintptr_t)idCpu);
2152 AssertRC(rc);
2153
2154 pVM->hm.s.pGuestPatchMem = pPatchMem;
2155 pVM->hm.s.pFreeGuestPatchMem = pPatchMem;
2156 pVM->hm.s.cbGuestPatchMem = cbPatchMem;
2157 return VINF_SUCCESS;
2158}
2159
2160
2161/**
2162 * Enable patching in a VT-x/AMD-V guest
2163 *
2164 * @returns VBox status code.
2165 * @param pVM The cross context VM structure.
2166 * @param pPatchMem Patch memory range.
2167 * @param cbPatchMem Size of the memory range.
2168 */
2169VMMR3_INT_DECL(int) HMR3EnablePatching(PVM pVM, RTGCPTR pPatchMem, unsigned cbPatchMem)
2170{
2171 VM_ASSERT_EMT(pVM);
2172 Log(("HMR3EnablePatching %RGv size %x\n", pPatchMem, cbPatchMem));
2173 if (pVM->cCpus > 1)
2174 {
2175 /* We own the IOM lock here and could cause a deadlock by waiting for a VCPU that is blocking on the IOM lock. */
2176 int rc = VMR3ReqCallNoWait(pVM, VMCPUID_ANY_QUEUE,
2177 (PFNRT)hmR3EnablePatching, 4, pVM, VMMGetCpuId(pVM), (RTRCPTR)pPatchMem, cbPatchMem);
2178 AssertRC(rc);
2179 return rc;
2180 }
2181 return hmR3EnablePatching(pVM, VMMGetCpuId(pVM), (RTRCPTR)pPatchMem, cbPatchMem);
2182}
2183
2184
2185/**
2186 * Disable patching in a VT-x/AMD-V guest.
2187 *
2188 * @returns VBox status code.
2189 * @param pVM The cross context VM structure.
2190 * @param pPatchMem Patch memory range.
2191 * @param cbPatchMem Size of the memory range.
2192 */
2193VMMR3_INT_DECL(int) HMR3DisablePatching(PVM pVM, RTGCPTR pPatchMem, unsigned cbPatchMem)
2194{
2195 Log(("HMR3DisablePatching %RGv size %x\n", pPatchMem, cbPatchMem));
2196 RT_NOREF2(pPatchMem, cbPatchMem);
2197
2198 Assert(pVM->hm.s.pGuestPatchMem == pPatchMem);
2199 Assert(pVM->hm.s.cbGuestPatchMem == cbPatchMem);
2200
2201 /** @todo Potential deadlock when other VCPUs are waiting on the IOM lock (we own it)!! */
2202 int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONE_BY_ONE, hmR3RemovePatches,
2203 (void *)(uintptr_t)VMMGetCpuId(pVM));
2204 AssertRC(rc);
2205
2206 pVM->hm.s.pGuestPatchMem = 0;
2207 pVM->hm.s.pFreeGuestPatchMem = 0;
2208 pVM->hm.s.cbGuestPatchMem = 0;
2209 pVM->hm.s.fTPRPatchingActive = false;
2210 return VINF_SUCCESS;
2211}
2212
2213
2214/**
2215 * Callback to patch a TPR instruction (vmmcall or mov cr8).
2216 *
2217 * @returns VBox strict status code.
2218 * @param pVM The cross context VM structure.
2219 * @param pVCpu The cross context virtual CPU structure of the calling EMT.
2220 * @param pvUser User specified CPU context.
2221 *
2222 */
2223static DECLCALLBACK(VBOXSTRICTRC) hmR3ReplaceTprInstr(PVM pVM, PVMCPU pVCpu, void *pvUser)
2224{
2225 /*
2226 * Only execute the handler on the VCPU the original patch request was
2227 * issued. (The other CPU(s) might not yet have switched to protected
2228 * mode, nor have the correct memory context.)
2229 */
2230 VMCPUID idCpu = (VMCPUID)(uintptr_t)pvUser;
2231 if (pVCpu->idCpu != idCpu)
2232 return VINF_SUCCESS;
2233
2234 /*
2235 * We're racing other VCPUs here, so don't try patch the instruction twice
2236 * and make sure there is still room for our patch record.
2237 */
2238 PCPUMCTX pCtx = CPUMQueryGuestCtxPtr(pVCpu);
2239 PHMTPRPATCH pPatch = (PHMTPRPATCH)RTAvloU32Get(&pVM->hm.s.PatchTree, (AVLOU32KEY)pCtx->eip);
2240 if (pPatch)
2241 {
2242 Log(("hmR3ReplaceTprInstr: already patched %RGv\n", pCtx->rip));
2243 return VINF_SUCCESS;
2244 }
2245 uint32_t const idx = pVM->hm.s.cPatches;
2246 if (idx >= RT_ELEMENTS(pVM->hm.s.aPatches))
2247 {
2248 Log(("hmR3ReplaceTprInstr: no available patch slots (%RGv)\n", pCtx->rip));
2249 return VINF_SUCCESS;
2250 }
2251 pPatch = &pVM->hm.s.aPatches[idx];
2252
2253 Log(("hmR3ReplaceTprInstr: rip=%RGv idxPatch=%u\n", pCtx->rip, idx));
2254
2255 /*
2256 * Disassembler the instruction and get cracking.
2257 */
2258 DBGFR3_DISAS_INSTR_CUR_LOG(pVCpu, "hmR3ReplaceTprInstr");
2259 PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
2260 uint32_t cbOp;
2261 int rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
2262 AssertRC(rc);
2263 if ( rc == VINF_SUCCESS
2264 && pDis->pCurInstr->uOpcode == OP_MOV
2265 && cbOp >= 3)
2266 {
2267 static uint8_t const s_abVMMCall[3] = { 0x0f, 0x01, 0xd9 };
2268
2269 rc = PGMPhysSimpleReadGCPtr(pVCpu, pPatch->aOpcode, pCtx->rip, cbOp);
2270 AssertRC(rc);
2271
2272 pPatch->cbOp = cbOp;
2273
2274 if (pDis->Param1.fUse == DISUSE_DISPLACEMENT32)
2275 {
2276 /* write. */
2277 if (pDis->Param2.fUse == DISUSE_REG_GEN32)
2278 {
2279 pPatch->enmType = HMTPRINSTR_WRITE_REG;
2280 pPatch->uSrcOperand = pDis->Param2.Base.idxGenReg;
2281 Log(("hmR3ReplaceTprInstr: HMTPRINSTR_WRITE_REG %u\n", pDis->Param2.Base.idxGenReg));
2282 }
2283 else
2284 {
2285 Assert(pDis->Param2.fUse == DISUSE_IMMEDIATE32);
2286 pPatch->enmType = HMTPRINSTR_WRITE_IMM;
2287 pPatch->uSrcOperand = pDis->Param2.uValue;
2288 Log(("hmR3ReplaceTprInstr: HMTPRINSTR_WRITE_IMM %#llx\n", pDis->Param2.uValue));
2289 }
2290 rc = PGMPhysSimpleWriteGCPtr(pVCpu, pCtx->rip, s_abVMMCall, sizeof(s_abVMMCall));
2291 AssertRC(rc);
2292
2293 memcpy(pPatch->aNewOpcode, s_abVMMCall, sizeof(s_abVMMCall));
2294 pPatch->cbNewOp = sizeof(s_abVMMCall);
2295 STAM_COUNTER_INC(&pVM->hm.s.StatTprReplaceSuccessVmc);
2296 }
2297 else
2298 {
2299 /*
2300 * TPR Read.
2301 *
2302 * Found:
2303 * mov eax, dword [fffe0080] (5 bytes)
2304 * Check if next instruction is:
2305 * shr eax, 4
2306 */
2307 Assert(pDis->Param1.fUse == DISUSE_REG_GEN32);
2308
2309 uint8_t const idxMmioReg = pDis->Param1.Base.idxGenReg;
2310 uint8_t const cbOpMmio = cbOp;
2311 uint64_t const uSavedRip = pCtx->rip;
2312
2313 pCtx->rip += cbOp;
2314 rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
2315 DBGFR3_DISAS_INSTR_CUR_LOG(pVCpu, "Following read");
2316 pCtx->rip = uSavedRip;
2317
2318 if ( rc == VINF_SUCCESS
2319 && pDis->pCurInstr->uOpcode == OP_SHR
2320 && pDis->Param1.fUse == DISUSE_REG_GEN32
2321 && pDis->Param1.Base.idxGenReg == idxMmioReg
2322 && pDis->Param2.fUse == DISUSE_IMMEDIATE8
2323 && pDis->Param2.uValue == 4
2324 && cbOpMmio + cbOp < sizeof(pVM->hm.s.aPatches[idx].aOpcode))
2325 {
2326 uint8_t abInstr[15];
2327
2328 /* Replacing the two instructions above with an AMD-V specific lock-prefixed 32-bit MOV CR8 instruction so as to
2329 access CR8 in 32-bit mode and not cause a #VMEXIT. */
2330 rc = PGMPhysSimpleReadGCPtr(pVCpu, &pPatch->aOpcode, pCtx->rip, cbOpMmio + cbOp);
2331 AssertRC(rc);
2332
2333 pPatch->cbOp = cbOpMmio + cbOp;
2334
2335 /* 0xF0, 0x0F, 0x20, 0xC0 = mov eax, cr8 */
2336 abInstr[0] = 0xF0;
2337 abInstr[1] = 0x0F;
2338 abInstr[2] = 0x20;
2339 abInstr[3] = 0xC0 | pDis->Param1.Base.idxGenReg;
2340 for (unsigned i = 4; i < pPatch->cbOp; i++)
2341 abInstr[i] = 0x90; /* nop */
2342
2343 rc = PGMPhysSimpleWriteGCPtr(pVCpu, pCtx->rip, abInstr, pPatch->cbOp);
2344 AssertRC(rc);
2345
2346 memcpy(pPatch->aNewOpcode, abInstr, pPatch->cbOp);
2347 pPatch->cbNewOp = pPatch->cbOp;
2348 STAM_COUNTER_INC(&pVM->hm.s.StatTprReplaceSuccessCr8);
2349
2350 Log(("Acceptable read/shr candidate!\n"));
2351 pPatch->enmType = HMTPRINSTR_READ_SHR4;
2352 }
2353 else
2354 {
2355 pPatch->enmType = HMTPRINSTR_READ;
2356 pPatch->uDstOperand = idxMmioReg;
2357
2358 rc = PGMPhysSimpleWriteGCPtr(pVCpu, pCtx->rip, s_abVMMCall, sizeof(s_abVMMCall));
2359 AssertRC(rc);
2360
2361 memcpy(pPatch->aNewOpcode, s_abVMMCall, sizeof(s_abVMMCall));
2362 pPatch->cbNewOp = sizeof(s_abVMMCall);
2363 STAM_COUNTER_INC(&pVM->hm.s.StatTprReplaceSuccessVmc);
2364 Log(("hmR3ReplaceTprInstr: HMTPRINSTR_READ %u\n", pPatch->uDstOperand));
2365 }
2366 }
2367
2368 pPatch->Core.Key = pCtx->eip;
2369 rc = RTAvloU32Insert(&pVM->hm.s.PatchTree, &pPatch->Core);
2370 AssertRC(rc);
2371
2372 pVM->hm.s.cPatches++;
2373 return VINF_SUCCESS;
2374 }
2375
2376 /*
2377 * Save invalid patch, so we will not try again.
2378 */
2379 Log(("hmR3ReplaceTprInstr: Failed to patch instr!\n"));
2380 pPatch->Core.Key = pCtx->eip;
2381 pPatch->enmType = HMTPRINSTR_INVALID;
2382 rc = RTAvloU32Insert(&pVM->hm.s.PatchTree, &pPatch->Core);
2383 AssertRC(rc);
2384 pVM->hm.s.cPatches++;
2385 STAM_COUNTER_INC(&pVM->hm.s.StatTprReplaceFailure);
2386 return VINF_SUCCESS;
2387}
2388
2389
2390/**
2391 * Callback to patch a TPR instruction (jump to generated code).
2392 *
2393 * @returns VBox strict status code.
2394 * @param pVM The cross context VM structure.
2395 * @param pVCpu The cross context virtual CPU structure of the calling EMT.
2396 * @param pvUser User specified CPU context.
2397 *
2398 */
2399static DECLCALLBACK(VBOXSTRICTRC) hmR3PatchTprInstr(PVM pVM, PVMCPU pVCpu, void *pvUser)
2400{
2401 /*
2402 * Only execute the handler on the VCPU the original patch request was
2403 * issued. (The other CPU(s) might not yet have switched to protected
2404 * mode, nor have the correct memory context.)
2405 */
2406 VMCPUID idCpu = (VMCPUID)(uintptr_t)pvUser;
2407 if (pVCpu->idCpu != idCpu)
2408 return VINF_SUCCESS;
2409
2410 /*
2411 * We're racing other VCPUs here, so don't try patch the instruction twice
2412 * and make sure there is still room for our patch record.
2413 */
2414 PCPUMCTX pCtx = CPUMQueryGuestCtxPtr(pVCpu);
2415 PHMTPRPATCH pPatch = (PHMTPRPATCH)RTAvloU32Get(&pVM->hm.s.PatchTree, (AVLOU32KEY)pCtx->eip);
2416 if (pPatch)
2417 {
2418 Log(("hmR3PatchTprInstr: already patched %RGv\n", pCtx->rip));
2419 return VINF_SUCCESS;
2420 }
2421 uint32_t const idx = pVM->hm.s.cPatches;
2422 if (idx >= RT_ELEMENTS(pVM->hm.s.aPatches))
2423 {
2424 Log(("hmR3PatchTprInstr: no available patch slots (%RGv)\n", pCtx->rip));
2425 return VINF_SUCCESS;
2426 }
2427 pPatch = &pVM->hm.s.aPatches[idx];
2428
2429 Log(("hmR3PatchTprInstr: rip=%RGv idxPatch=%u\n", pCtx->rip, idx));
2430 DBGFR3_DISAS_INSTR_CUR_LOG(pVCpu, "hmR3PatchTprInstr");
2431
2432 /*
2433 * Disassemble the instruction and get cracking.
2434 */
2435 PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
2436 uint32_t cbOp;
2437 int rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
2438 AssertRC(rc);
2439 if ( rc == VINF_SUCCESS
2440 && pDis->pCurInstr->uOpcode == OP_MOV
2441 && cbOp >= 5)
2442 {
2443 uint8_t aPatch[64];
2444 uint32_t off = 0;
2445
2446 rc = PGMPhysSimpleReadGCPtr(pVCpu, pPatch->aOpcode, pCtx->rip, cbOp);
2447 AssertRC(rc);
2448
2449 pPatch->cbOp = cbOp;
2450 pPatch->enmType = HMTPRINSTR_JUMP_REPLACEMENT;
2451
2452 if (pDis->Param1.fUse == DISUSE_DISPLACEMENT32)
2453 {
2454 /*
2455 * TPR write:
2456 *
2457 * push ECX [51]
2458 * push EDX [52]
2459 * push EAX [50]
2460 * xor EDX,EDX [31 D2]
2461 * mov EAX,EAX [89 C0]
2462 * or
2463 * mov EAX,0000000CCh [B8 CC 00 00 00]
2464 * mov ECX,0C0000082h [B9 82 00 00 C0]
2465 * wrmsr [0F 30]
2466 * pop EAX [58]
2467 * pop EDX [5A]
2468 * pop ECX [59]
2469 * jmp return_address [E9 return_address]
2470 *
2471 */
2472 bool fUsesEax = (pDis->Param2.fUse == DISUSE_REG_GEN32 && pDis->Param2.Base.idxGenReg == DISGREG_EAX);
2473
2474 aPatch[off++] = 0x51; /* push ecx */
2475 aPatch[off++] = 0x52; /* push edx */
2476 if (!fUsesEax)
2477 aPatch[off++] = 0x50; /* push eax */
2478 aPatch[off++] = 0x31; /* xor edx, edx */
2479 aPatch[off++] = 0xD2;
2480 if (pDis->Param2.fUse == DISUSE_REG_GEN32)
2481 {
2482 if (!fUsesEax)
2483 {
2484 aPatch[off++] = 0x89; /* mov eax, src_reg */
2485 aPatch[off++] = MAKE_MODRM(3, pDis->Param2.Base.idxGenReg, DISGREG_EAX);
2486 }
2487 }
2488 else
2489 {
2490 Assert(pDis->Param2.fUse == DISUSE_IMMEDIATE32);
2491 aPatch[off++] = 0xB8; /* mov eax, immediate */
2492 *(uint32_t *)&aPatch[off] = pDis->Param2.uValue;
2493 off += sizeof(uint32_t);
2494 }
2495 aPatch[off++] = 0xB9; /* mov ecx, 0xc0000082 */
2496 *(uint32_t *)&aPatch[off] = MSR_K8_LSTAR;
2497 off += sizeof(uint32_t);
2498
2499 aPatch[off++] = 0x0F; /* wrmsr */
2500 aPatch[off++] = 0x30;
2501 if (!fUsesEax)
2502 aPatch[off++] = 0x58; /* pop eax */
2503 aPatch[off++] = 0x5A; /* pop edx */
2504 aPatch[off++] = 0x59; /* pop ecx */
2505 }
2506 else
2507 {
2508 /*
2509 * TPR read:
2510 *
2511 * push ECX [51]
2512 * push EDX [52]
2513 * push EAX [50]
2514 * mov ECX,0C0000082h [B9 82 00 00 C0]
2515 * rdmsr [0F 32]
2516 * mov EAX,EAX [89 C0]
2517 * pop EAX [58]
2518 * pop EDX [5A]
2519 * pop ECX [59]
2520 * jmp return_address [E9 return_address]
2521 *
2522 */
2523 Assert(pDis->Param1.fUse == DISUSE_REG_GEN32);
2524
2525 if (pDis->Param1.Base.idxGenReg != DISGREG_ECX)
2526 aPatch[off++] = 0x51; /* push ecx */
2527 if (pDis->Param1.Base.idxGenReg != DISGREG_EDX )
2528 aPatch[off++] = 0x52; /* push edx */
2529 if (pDis->Param1.Base.idxGenReg != DISGREG_EAX)
2530 aPatch[off++] = 0x50; /* push eax */
2531
2532 aPatch[off++] = 0x31; /* xor edx, edx */
2533 aPatch[off++] = 0xD2;
2534
2535 aPatch[off++] = 0xB9; /* mov ecx, 0xc0000082 */
2536 *(uint32_t *)&aPatch[off] = MSR_K8_LSTAR;
2537 off += sizeof(uint32_t);
2538
2539 aPatch[off++] = 0x0F; /* rdmsr */
2540 aPatch[off++] = 0x32;
2541
2542 if (pDis->Param1.Base.idxGenReg != DISGREG_EAX)
2543 {
2544 aPatch[off++] = 0x89; /* mov dst_reg, eax */
2545 aPatch[off++] = MAKE_MODRM(3, DISGREG_EAX, pDis->Param1.Base.idxGenReg);
2546 }
2547
2548 if (pDis->Param1.Base.idxGenReg != DISGREG_EAX)
2549 aPatch[off++] = 0x58; /* pop eax */
2550 if (pDis->Param1.Base.idxGenReg != DISGREG_EDX )
2551 aPatch[off++] = 0x5A; /* pop edx */
2552 if (pDis->Param1.Base.idxGenReg != DISGREG_ECX)
2553 aPatch[off++] = 0x59; /* pop ecx */
2554 }
2555 aPatch[off++] = 0xE9; /* jmp return_address */
2556 *(RTRCUINTPTR *)&aPatch[off] = ((RTRCUINTPTR)pCtx->eip + cbOp) - ((RTRCUINTPTR)pVM->hm.s.pFreeGuestPatchMem + off + 4);
2557 off += sizeof(RTRCUINTPTR);
2558
2559 if (pVM->hm.s.pFreeGuestPatchMem + off <= pVM->hm.s.pGuestPatchMem + pVM->hm.s.cbGuestPatchMem)
2560 {
2561 /* Write new code to the patch buffer. */
2562 rc = PGMPhysSimpleWriteGCPtr(pVCpu, pVM->hm.s.pFreeGuestPatchMem, aPatch, off);
2563 AssertRC(rc);
2564
2565#ifdef LOG_ENABLED
2566 uint32_t cbCurInstr;
2567 for (RTGCPTR GCPtrInstr = pVM->hm.s.pFreeGuestPatchMem;
2568 GCPtrInstr < pVM->hm.s.pFreeGuestPatchMem + off;
2569 GCPtrInstr += RT_MAX(cbCurInstr, 1))
2570 {
2571 char szOutput[256];
2572 rc = DBGFR3DisasInstrEx(pVM->pUVM, pVCpu->idCpu, pCtx->cs.Sel, GCPtrInstr, DBGF_DISAS_FLAGS_DEFAULT_MODE,
2573 szOutput, sizeof(szOutput), &cbCurInstr);
2574 if (RT_SUCCESS(rc))
2575 Log(("Patch instr %s\n", szOutput));
2576 else
2577 Log(("%RGv: rc=%Rrc\n", GCPtrInstr, rc));
2578 }
2579#endif
2580
2581 pPatch->aNewOpcode[0] = 0xE9;
2582 *(RTRCUINTPTR *)&pPatch->aNewOpcode[1] = ((RTRCUINTPTR)pVM->hm.s.pFreeGuestPatchMem) - ((RTRCUINTPTR)pCtx->eip + 5);
2583
2584 /* Overwrite the TPR instruction with a jump. */
2585 rc = PGMPhysSimpleWriteGCPtr(pVCpu, pCtx->eip, pPatch->aNewOpcode, 5);
2586 AssertRC(rc);
2587
2588 DBGFR3_DISAS_INSTR_CUR_LOG(pVCpu, "Jump");
2589
2590 pVM->hm.s.pFreeGuestPatchMem += off;
2591 pPatch->cbNewOp = 5;
2592
2593 pPatch->Core.Key = pCtx->eip;
2594 rc = RTAvloU32Insert(&pVM->hm.s.PatchTree, &pPatch->Core);
2595 AssertRC(rc);
2596
2597 pVM->hm.s.cPatches++;
2598 pVM->hm.s.fTPRPatchingActive = true;
2599 STAM_COUNTER_INC(&pVM->hm.s.StatTprPatchSuccess);
2600 return VINF_SUCCESS;
2601 }
2602
2603 Log(("Ran out of space in our patch buffer!\n"));
2604 }
2605 else
2606 Log(("hmR3PatchTprInstr: Failed to patch instr!\n"));
2607
2608
2609 /*
2610 * Save invalid patch, so we will not try again.
2611 */
2612 pPatch = &pVM->hm.s.aPatches[idx];
2613 pPatch->Core.Key = pCtx->eip;
2614 pPatch->enmType = HMTPRINSTR_INVALID;
2615 rc = RTAvloU32Insert(&pVM->hm.s.PatchTree, &pPatch->Core);
2616 AssertRC(rc);
2617 pVM->hm.s.cPatches++;
2618 STAM_COUNTER_INC(&pVM->hm.s.StatTprPatchFailure);
2619 return VINF_SUCCESS;
2620}
2621
2622
2623/**
2624 * Attempt to patch TPR mmio instructions.
2625 *
2626 * @returns VBox status code.
2627 * @param pVM The cross context VM structure.
2628 * @param pVCpu The cross context virtual CPU structure.
2629 * @param pCtx Pointer to the guest CPU context.
2630 */
2631VMMR3_INT_DECL(int) HMR3PatchTprInstr(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
2632{
2633 NOREF(pCtx);
2634 int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONE_BY_ONE,
2635 pVM->hm.s.pGuestPatchMem ? hmR3PatchTprInstr : hmR3ReplaceTprInstr,
2636 (void *)(uintptr_t)pVCpu->idCpu);
2637 AssertRC(rc);
2638 return rc;
2639}
2640
2641
2642/**
2643 * Checks if a code selector (CS) is suitable for execution
2644 * within VMX when unrestricted execution isn't available.
2645 *
2646 * @returns true if selector is suitable for VMX, otherwise
2647 * false.
2648 * @param pSel Pointer to the selector to check (CS).
2649 * @param uStackDpl The CPL, aka the DPL of the stack segment.
2650 */
2651static bool hmR3IsCodeSelectorOkForVmx(PCPUMSELREG pSel, unsigned uStackDpl)
2652{
2653 /*
2654 * Segment must be an accessed code segment, it must be present and it must
2655 * be usable.
2656 * Note! These are all standard requirements and if CS holds anything else
2657 * we've got buggy code somewhere!
2658 */
2659 AssertCompile(X86DESCATTR_TYPE == 0xf);
2660 AssertMsgReturn( (pSel->Attr.u & (X86_SEL_TYPE_ACCESSED | X86_SEL_TYPE_CODE | X86DESCATTR_DT | X86DESCATTR_P | X86DESCATTR_UNUSABLE))
2661 == (X86_SEL_TYPE_ACCESSED | X86_SEL_TYPE_CODE | X86DESCATTR_DT | X86DESCATTR_P),
2662 ("%#x\n", pSel->Attr.u),
2663 false);
2664
2665 /* For conforming segments, CS.DPL must be <= SS.DPL, while CS.DPL
2666 must equal SS.DPL for non-confroming segments.
2667 Note! This is also a hard requirement like above. */
2668 AssertMsgReturn( pSel->Attr.n.u4Type & X86_SEL_TYPE_CONF
2669 ? pSel->Attr.n.u2Dpl <= uStackDpl
2670 : pSel->Attr.n.u2Dpl == uStackDpl,
2671 ("u4Type=%#x u2Dpl=%u uStackDpl=%u\n", pSel->Attr.n.u4Type, pSel->Attr.n.u2Dpl, uStackDpl),
2672 false);
2673
2674 /*
2675 * The following two requirements are VT-x specific:
2676 * - G bit must be set if any high limit bits are set.
2677 * - G bit must be clear if any low limit bits are clear.
2678 */
2679 if ( ((pSel->u32Limit & 0xfff00000) == 0x00000000 || pSel->Attr.n.u1Granularity)
2680 && ((pSel->u32Limit & 0x00000fff) == 0x00000fff || !pSel->Attr.n.u1Granularity) )
2681 return true;
2682 return false;
2683}
2684
2685
2686/**
2687 * Checks if a data selector (DS/ES/FS/GS) is suitable for
2688 * execution within VMX when unrestricted execution isn't
2689 * available.
2690 *
2691 * @returns true if selector is suitable for VMX, otherwise
2692 * false.
2693 * @param pSel Pointer to the selector to check
2694 * (DS/ES/FS/GS).
2695 */
2696static bool hmR3IsDataSelectorOkForVmx(PCPUMSELREG pSel)
2697{
2698 /*
2699 * Unusable segments are OK. These days they should be marked as such, as
2700 * but as an alternative we for old saved states and AMD<->VT-x migration
2701 * we also treat segments with all the attributes cleared as unusable.
2702 */
2703 if (pSel->Attr.n.u1Unusable || !pSel->Attr.u)
2704 return true;
2705
2706 /** @todo tighten these checks. Will require CPUM load adjusting. */
2707
2708 /* Segment must be accessed. */
2709 if (pSel->Attr.u & X86_SEL_TYPE_ACCESSED)
2710 {
2711 /* Code segments must also be readable. */
2712 if ( !(pSel->Attr.u & X86_SEL_TYPE_CODE)
2713 || (pSel->Attr.u & X86_SEL_TYPE_READ))
2714 {
2715 /* The S bit must be set. */
2716 if (pSel->Attr.n.u1DescType)
2717 {
2718 /* Except for conforming segments, DPL >= RPL. */
2719 if ( pSel->Attr.n.u2Dpl >= (pSel->Sel & X86_SEL_RPL)
2720 || pSel->Attr.n.u4Type >= X86_SEL_TYPE_ER_ACC)
2721 {
2722 /* Segment must be present. */
2723 if (pSel->Attr.n.u1Present)
2724 {
2725 /*
2726 * The following two requirements are VT-x specific:
2727 * - G bit must be set if any high limit bits are set.
2728 * - G bit must be clear if any low limit bits are clear.
2729 */
2730 if ( ((pSel->u32Limit & 0xfff00000) == 0x00000000 || pSel->Attr.n.u1Granularity)
2731 && ((pSel->u32Limit & 0x00000fff) == 0x00000fff || !pSel->Attr.n.u1Granularity) )
2732 return true;
2733 }
2734 }
2735 }
2736 }
2737 }
2738
2739 return false;
2740}
2741
2742
2743/**
2744 * Checks if the stack selector (SS) is suitable for execution
2745 * within VMX when unrestricted execution isn't available.
2746 *
2747 * @returns true if selector is suitable for VMX, otherwise
2748 * false.
2749 * @param pSel Pointer to the selector to check (SS).
2750 */
2751static bool hmR3IsStackSelectorOkForVmx(PCPUMSELREG pSel)
2752{
2753 /*
2754 * Unusable segments are OK. These days they should be marked as such, as
2755 * but as an alternative we for old saved states and AMD<->VT-x migration
2756 * we also treat segments with all the attributes cleared as unusable.
2757 */
2758 /** @todo r=bird: actually all zeroes isn't gonna cut it... SS.DPL == CPL. */
2759 if (pSel->Attr.n.u1Unusable || !pSel->Attr.u)
2760 return true;
2761
2762 /*
2763 * Segment must be an accessed writable segment, it must be present.
2764 * Note! These are all standard requirements and if SS holds anything else
2765 * we've got buggy code somewhere!
2766 */
2767 AssertCompile(X86DESCATTR_TYPE == 0xf);
2768 AssertMsgReturn( (pSel->Attr.u & (X86_SEL_TYPE_ACCESSED | X86_SEL_TYPE_WRITE | X86DESCATTR_DT | X86DESCATTR_P | X86_SEL_TYPE_CODE))
2769 == (X86_SEL_TYPE_ACCESSED | X86_SEL_TYPE_WRITE | X86DESCATTR_DT | X86DESCATTR_P),
2770 ("%#x\n", pSel->Attr.u),
2771 false);
2772
2773 /* DPL must equal RPL.
2774 Note! This is also a hard requirement like above. */
2775 AssertMsgReturn(pSel->Attr.n.u2Dpl == (pSel->Sel & X86_SEL_RPL),
2776 ("u2Dpl=%u Sel=%#x\n", pSel->Attr.n.u2Dpl, pSel->Sel),
2777 false);
2778
2779 /*
2780 * The following two requirements are VT-x specific:
2781 * - G bit must be set if any high limit bits are set.
2782 * - G bit must be clear if any low limit bits are clear.
2783 */
2784 if ( ((pSel->u32Limit & 0xfff00000) == 0x00000000 || pSel->Attr.n.u1Granularity)
2785 && ((pSel->u32Limit & 0x00000fff) == 0x00000fff || !pSel->Attr.n.u1Granularity) )
2786 return true;
2787 return false;
2788}
2789
2790
2791/**
2792 * Force execution of the current IO code in the recompiler.
2793 *
2794 * @returns VBox status code.
2795 * @param pVM The cross context VM structure.
2796 * @param pCtx Partial VM execution context.
2797 */
2798VMMR3_INT_DECL(int) HMR3EmulateIoBlock(PVM pVM, PCPUMCTX pCtx)
2799{
2800 PVMCPU pVCpu = VMMGetCpu(pVM);
2801
2802 Assert(HMIsEnabled(pVM));
2803 Log(("HMR3EmulateIoBlock\n"));
2804
2805 /* This is primarily intended to speed up Grub, so we don't care about paged protected mode. */
2806 if (HMCanEmulateIoBlockEx(pCtx))
2807 {
2808 Log(("HMR3EmulateIoBlock -> enabled\n"));
2809 pVCpu->hm.s.EmulateIoBlock.fEnabled = true;
2810 pVCpu->hm.s.EmulateIoBlock.GCPtrFunctionEip = pCtx->rip;
2811 pVCpu->hm.s.EmulateIoBlock.cr0 = pCtx->cr0;
2812 return VINF_EM_RESCHEDULE_REM;
2813 }
2814 return VINF_SUCCESS;
2815}
2816
2817
2818/**
2819 * Checks if we can currently use hardware accelerated raw mode.
2820 *
2821 * @returns true if we can currently use hardware acceleration, otherwise false.
2822 * @param pVM The cross context VM structure.
2823 * @param pCtx Partial VM execution context.
2824 */
2825VMMR3DECL(bool) HMR3CanExecuteGuest(PVM pVM, PCPUMCTX pCtx)
2826{
2827 PVMCPU pVCpu = VMMGetCpu(pVM);
2828
2829 Assert(HMIsEnabled(pVM));
2830
2831#ifdef VBOX_WITH_NESTED_HWVIRT_ONLY_IN_IEM
2832 if (CPUMIsGuestInNestedHwVirtMode(pCtx))
2833 {
2834 Log(("HMR3CanExecuteGuest: In nested-guest mode - returning false"));
2835 return false;
2836 }
2837#endif
2838
2839 /* If we're still executing the IO code, then return false. */
2840 if ( RT_UNLIKELY(pVCpu->hm.s.EmulateIoBlock.fEnabled)
2841 && pCtx->rip < pVCpu->hm.s.EmulateIoBlock.GCPtrFunctionEip + 0x200
2842 && pCtx->rip > pVCpu->hm.s.EmulateIoBlock.GCPtrFunctionEip - 0x200
2843 && pCtx->cr0 == pVCpu->hm.s.EmulateIoBlock.cr0)
2844 return false;
2845
2846 pVCpu->hm.s.EmulateIoBlock.fEnabled = false;
2847
2848 /* AMD-V supports real & protected mode with or without paging. */
2849 if (pVM->hm.s.svm.fEnabled)
2850 {
2851 pVCpu->hm.s.fActive = true;
2852 return true;
2853 }
2854
2855 pVCpu->hm.s.fActive = false;
2856
2857 /* Note! The context supplied by REM is partial. If we add more checks here, be sure to verify that REM provides this info! */
2858 Assert( (pVM->hm.s.vmx.fUnrestrictedGuest && !pVM->hm.s.vmx.pRealModeTSS)
2859 || (!pVM->hm.s.vmx.fUnrestrictedGuest && pVM->hm.s.vmx.pRealModeTSS));
2860
2861 bool fSupportsRealMode = pVM->hm.s.vmx.fUnrestrictedGuest || PDMVmmDevHeapIsEnabled(pVM);
2862 if (!pVM->hm.s.vmx.fUnrestrictedGuest)
2863 {
2864 /*
2865 * The VMM device heap is a requirement for emulating real mode or protected mode without paging with the unrestricted
2866 * guest execution feature is missing (VT-x only).
2867 */
2868 if (fSupportsRealMode)
2869 {
2870 if (CPUMIsGuestInRealModeEx(pCtx))
2871 {
2872 /* In V86 mode (VT-x or not), the CPU enforces real-mode compatible selector
2873 * bases and limits, i.e. limit must be 64K and base must be selector * 16.
2874 * If this is not true, we cannot execute real mode as V86 and have to fall
2875 * back to emulation.
2876 */
2877 if ( pCtx->cs.Sel != (pCtx->cs.u64Base >> 4)
2878 || pCtx->ds.Sel != (pCtx->ds.u64Base >> 4)
2879 || pCtx->es.Sel != (pCtx->es.u64Base >> 4)
2880 || pCtx->ss.Sel != (pCtx->ss.u64Base >> 4)
2881 || pCtx->fs.Sel != (pCtx->fs.u64Base >> 4)
2882 || pCtx->gs.Sel != (pCtx->gs.u64Base >> 4))
2883 {
2884 STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxCheckBadRmSelBase);
2885 return false;
2886 }
2887 if ( (pCtx->cs.u32Limit != 0xffff)
2888 || (pCtx->ds.u32Limit != 0xffff)
2889 || (pCtx->es.u32Limit != 0xffff)
2890 || (pCtx->ss.u32Limit != 0xffff)
2891 || (pCtx->fs.u32Limit != 0xffff)
2892 || (pCtx->gs.u32Limit != 0xffff))
2893 {
2894 STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxCheckBadRmSelLimit);
2895 return false;
2896 }
2897 STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxCheckRmOk);
2898 }
2899 else
2900 {
2901 /* Verify the requirements for executing code in protected
2902 mode. VT-x can't handle the CPU state right after a switch
2903 from real to protected mode. (all sorts of RPL & DPL assumptions). */
2904 if (pVCpu->hm.s.vmx.fWasInRealMode)
2905 {
2906 /** @todo If guest is in V86 mode, these checks should be different! */
2907 if ((pCtx->cs.Sel & X86_SEL_RPL) != (pCtx->ss.Sel & X86_SEL_RPL))
2908 {
2909 STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxCheckBadRpl);
2910 return false;
2911 }
2912 if ( !hmR3IsCodeSelectorOkForVmx(&pCtx->cs, pCtx->ss.Attr.n.u2Dpl)
2913 || !hmR3IsDataSelectorOkForVmx(&pCtx->ds)
2914 || !hmR3IsDataSelectorOkForVmx(&pCtx->es)
2915 || !hmR3IsDataSelectorOkForVmx(&pCtx->fs)
2916 || !hmR3IsDataSelectorOkForVmx(&pCtx->gs)
2917 || !hmR3IsStackSelectorOkForVmx(&pCtx->ss))
2918 {
2919 STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxCheckBadSel);
2920 return false;
2921 }
2922 }
2923 /* VT-x also chokes on invalid TR or LDTR selectors (minix). */
2924 if (pCtx->gdtr.cbGdt)
2925 {
2926 if ((pCtx->tr.Sel | X86_SEL_RPL_LDT) > pCtx->gdtr.cbGdt)
2927 {
2928 STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxCheckBadTr);
2929 return false;
2930 }
2931 else if ((pCtx->ldtr.Sel | X86_SEL_RPL_LDT) > pCtx->gdtr.cbGdt)
2932 {
2933 STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxCheckBadLdt);
2934 return false;
2935 }
2936 }
2937 STAM_COUNTER_INC(&pVCpu->hm.s.StatVmxCheckPmOk);
2938 }
2939 }
2940 else
2941 {
2942 if ( !CPUMIsGuestInLongModeEx(pCtx)
2943 && !pVM->hm.s.vmx.fUnrestrictedGuest)
2944 {
2945 if ( !pVM->hm.s.fNestedPaging /* Requires a fake PD for real *and* protected mode without paging - stored in the VMM device heap */
2946 || CPUMIsGuestInRealModeEx(pCtx)) /* Requires a fake TSS for real mode - stored in the VMM device heap */
2947 return false;
2948
2949 /* Too early for VT-x; Solaris guests will fail with a guru meditation otherwise; same for XP. */
2950 if (pCtx->idtr.pIdt == 0 || pCtx->idtr.cbIdt == 0 || pCtx->tr.Sel == 0)
2951 return false;
2952
2953 /* The guest is about to complete the switch to protected mode. Wait a bit longer. */
2954 /* Windows XP; switch to protected mode; all selectors are marked not present in the
2955 * hidden registers (possible recompiler bug; see load_seg_vm) */
2956 if (pCtx->cs.Attr.n.u1Present == 0)
2957 return false;
2958 if (pCtx->ss.Attr.n.u1Present == 0)
2959 return false;
2960
2961 /* Windows XP: possible same as above, but new recompiler requires new heuristics?
2962 VT-x doesn't seem to like something about the guest state and this stuff avoids it. */
2963 /** @todo This check is actually wrong, it doesn't take the direction of the
2964 * stack segment into account. But, it does the job for now. */
2965 if (pCtx->rsp >= pCtx->ss.u32Limit)
2966 return false;
2967 }
2968 }
2969 }
2970
2971 if (pVM->hm.s.vmx.fEnabled)
2972 {
2973 uint32_t mask;
2974
2975 /* if bit N is set in cr0_fixed0, then it must be set in the guest's cr0. */
2976 mask = (uint32_t)pVM->hm.s.vmx.Msrs.u64Cr0Fixed0;
2977 /* Note: We ignore the NE bit here on purpose; see vmmr0\hmr0.cpp for details. */
2978 mask &= ~X86_CR0_NE;
2979
2980 if (fSupportsRealMode)
2981 {
2982 /* Note: We ignore the PE & PG bits here on purpose; we emulate real and protected mode without paging. */
2983 mask &= ~(X86_CR0_PG|X86_CR0_PE);
2984 }
2985 else
2986 {
2987 /* We support protected mode without paging using identity mapping. */
2988 mask &= ~X86_CR0_PG;
2989 }
2990 if ((pCtx->cr0 & mask) != mask)
2991 return false;
2992
2993 /* if bit N is cleared in cr0_fixed1, then it must be zero in the guest's cr0. */
2994 mask = (uint32_t)~pVM->hm.s.vmx.Msrs.u64Cr0Fixed1;
2995 if ((pCtx->cr0 & mask) != 0)
2996 return false;
2997
2998 /* if bit N is set in cr4_fixed0, then it must be set in the guest's cr4. */
2999 mask = (uint32_t)pVM->hm.s.vmx.Msrs.u64Cr4Fixed0;
3000 mask &= ~X86_CR4_VMXE;
3001 if ((pCtx->cr4 & mask) != mask)
3002 return false;
3003
3004 /* if bit N is cleared in cr4_fixed1, then it must be zero in the guest's cr4. */
3005 mask = (uint32_t)~pVM->hm.s.vmx.Msrs.u64Cr4Fixed1;
3006 if ((pCtx->cr4 & mask) != 0)
3007 return false;
3008
3009 pVCpu->hm.s.fActive = true;
3010 return true;
3011 }
3012
3013 return false;
3014}
3015
3016
3017/**
3018 * Checks if we need to reschedule due to VMM device heap changes.
3019 *
3020 * @returns true if a reschedule is required, otherwise false.
3021 * @param pVM The cross context VM structure.
3022 * @param pCtx VM execution context.
3023 */
3024VMMR3_INT_DECL(bool) HMR3IsRescheduleRequired(PVM pVM, PCPUMCTX pCtx)
3025{
3026 /*
3027 * The VMM device heap is a requirement for emulating real-mode or protected-mode without paging
3028 * when the unrestricted guest execution feature is missing (VT-x only).
3029 */
3030 if ( pVM->hm.s.vmx.fEnabled
3031 && !pVM->hm.s.vmx.fUnrestrictedGuest
3032 && CPUMIsGuestInRealModeEx(pCtx)
3033 && !PDMVmmDevHeapIsEnabled(pVM))
3034 {
3035 return true;
3036 }
3037
3038 return false;
3039}
3040
3041
3042/**
3043 * Noticiation callback from DBGF when interrupt breakpoints or generic debug
3044 * event settings changes.
3045 *
3046 * DBGF will call HMR3NotifyDebugEventChangedPerCpu on each CPU afterwards, this
3047 * function is just updating the VM globals.
3048 *
3049 * @param pVM The VM cross context VM structure.
3050 * @thread EMT(0)
3051 */
3052VMMR3_INT_DECL(void) HMR3NotifyDebugEventChanged(PVM pVM)
3053{
3054 /* Interrupts. */
3055 bool fUseDebugLoop = pVM->dbgf.ro.cSoftIntBreakpoints > 0
3056 || pVM->dbgf.ro.cHardIntBreakpoints > 0;
3057
3058 /* CPU Exceptions. */
3059 for (DBGFEVENTTYPE enmEvent = DBGFEVENT_XCPT_FIRST;
3060 !fUseDebugLoop && enmEvent <= DBGFEVENT_XCPT_LAST;
3061 enmEvent = (DBGFEVENTTYPE)(enmEvent + 1))
3062 fUseDebugLoop = DBGF_IS_EVENT_ENABLED(pVM, enmEvent);
3063
3064 /* Common VM exits. */
3065 for (DBGFEVENTTYPE enmEvent = DBGFEVENT_EXIT_FIRST;
3066 !fUseDebugLoop && enmEvent <= DBGFEVENT_EXIT_LAST_COMMON;
3067 enmEvent = (DBGFEVENTTYPE)(enmEvent + 1))
3068 fUseDebugLoop = DBGF_IS_EVENT_ENABLED(pVM, enmEvent);
3069
3070 /* Vendor specific VM exits. */
3071 if (HMR3IsVmxEnabled(pVM->pUVM))
3072 for (DBGFEVENTTYPE enmEvent = DBGFEVENT_EXIT_VMX_FIRST;
3073 !fUseDebugLoop && enmEvent <= DBGFEVENT_EXIT_VMX_LAST;
3074 enmEvent = (DBGFEVENTTYPE)(enmEvent + 1))
3075 fUseDebugLoop = DBGF_IS_EVENT_ENABLED(pVM, enmEvent);
3076 else
3077 for (DBGFEVENTTYPE enmEvent = DBGFEVENT_EXIT_SVM_FIRST;
3078 !fUseDebugLoop && enmEvent <= DBGFEVENT_EXIT_SVM_LAST;
3079 enmEvent = (DBGFEVENTTYPE)(enmEvent + 1))
3080 fUseDebugLoop = DBGF_IS_EVENT_ENABLED(pVM, enmEvent);
3081
3082 /* Done. */
3083 pVM->hm.s.fUseDebugLoop = fUseDebugLoop;
3084}
3085
3086
3087/**
3088 * Follow up notification callback to HMR3NotifyDebugEventChanged for each CPU.
3089 *
3090 * HM uses this to combine the decision made by HMR3NotifyDebugEventChanged with
3091 * per CPU settings.
3092 *
3093 * @param pVM The VM cross context VM structure.
3094 * @param pVCpu The cross context virtual CPU structure of the calling EMT.
3095 */
3096VMMR3_INT_DECL(void) HMR3NotifyDebugEventChangedPerCpu(PVM pVM, PVMCPU pVCpu)
3097{
3098 pVCpu->hm.s.fUseDebugLoop = pVCpu->hm.s.fSingleInstruction | pVM->hm.s.fUseDebugLoop;
3099}
3100
3101
3102/**
3103 * Notification from EM about a rescheduling into hardware assisted execution
3104 * mode.
3105 *
3106 * @param pVCpu The cross context virtual CPU structure of the calling EMT.
3107 */
3108VMMR3_INT_DECL(void) HMR3NotifyScheduled(PVMCPU pVCpu)
3109{
3110 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
3111}
3112
3113
3114/**
3115 * Notification from EM about returning from instruction emulation (REM / EM).
3116 *
3117 * @param pVCpu The cross context virtual CPU structure.
3118 */
3119VMMR3_INT_DECL(void) HMR3NotifyEmulated(PVMCPU pVCpu)
3120{
3121 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
3122}
3123
3124
3125/**
3126 * Checks if we are currently using hardware acceleration.
3127 *
3128 * @returns true if hardware acceleration is being used, otherwise false.
3129 * @param pVCpu The cross context virtual CPU structure.
3130 */
3131VMMR3_INT_DECL(bool) HMR3IsActive(PVMCPU pVCpu)
3132{
3133 return pVCpu->hm.s.fActive;
3134}
3135
3136
3137/**
3138 * External interface for querying whether hardware acceleration is enabled.
3139 *
3140 * @returns true if VT-x or AMD-V is being used, otherwise false.
3141 * @param pUVM The user mode VM handle.
3142 * @sa HMIsEnabled, HMIsEnabledNotMacro.
3143 */
3144VMMR3DECL(bool) HMR3IsEnabled(PUVM pUVM)
3145{
3146 UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
3147 PVM pVM = pUVM->pVM;
3148 VM_ASSERT_VALID_EXT_RETURN(pVM, false);
3149 return pVM->fHMEnabled; /* Don't use the macro as the GUI may query us very very early. */
3150}
3151
3152
3153/**
3154 * External interface for querying whether VT-x is being used.
3155 *
3156 * @returns true if VT-x is being used, otherwise false.
3157 * @param pUVM The user mode VM handle.
3158 * @sa HMR3IsSvmEnabled, HMIsEnabled
3159 */
3160VMMR3DECL(bool) HMR3IsVmxEnabled(PUVM pUVM)
3161{
3162 UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
3163 PVM pVM = pUVM->pVM;
3164 VM_ASSERT_VALID_EXT_RETURN(pVM, false);
3165 return pVM->hm.s.vmx.fEnabled
3166 && pVM->hm.s.vmx.fSupported
3167 && pVM->fHMEnabled;
3168}
3169
3170
3171/**
3172 * External interface for querying whether AMD-V is being used.
3173 *
3174 * @returns true if VT-x is being used, otherwise false.
3175 * @param pUVM The user mode VM handle.
3176 * @sa HMR3IsVmxEnabled, HMIsEnabled
3177 */
3178VMMR3DECL(bool) HMR3IsSvmEnabled(PUVM pUVM)
3179{
3180 UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
3181 PVM pVM = pUVM->pVM;
3182 VM_ASSERT_VALID_EXT_RETURN(pVM, false);
3183 return pVM->hm.s.svm.fEnabled
3184 && pVM->hm.s.svm.fSupported
3185 && pVM->fHMEnabled;
3186}
3187
3188
3189/**
3190 * Checks if we are currently using nested paging.
3191 *
3192 * @returns true if nested paging is being used, otherwise false.
3193 * @param pUVM The user mode VM handle.
3194 */
3195VMMR3DECL(bool) HMR3IsNestedPagingActive(PUVM pUVM)
3196{
3197 UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
3198 PVM pVM = pUVM->pVM;
3199 VM_ASSERT_VALID_EXT_RETURN(pVM, false);
3200 return pVM->hm.s.fNestedPaging;
3201}
3202
3203
3204/**
3205 * Checks if virtualized APIC registers is enabled.
3206 *
3207 * When enabled this feature allows the hardware to access most of the
3208 * APIC registers in the virtual-APIC page without causing VM-exits. See
3209 * Intel spec. 29.1.1 "Virtualized APIC Registers".
3210 *
3211 * @returns true if virtualized APIC registers is enabled, otherwise
3212 * false.
3213 * @param pUVM The user mode VM handle.
3214 */
3215VMMR3DECL(bool) HMR3IsVirtApicRegsEnabled(PUVM pUVM)
3216{
3217 UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
3218 PVM pVM = pUVM->pVM;
3219 VM_ASSERT_VALID_EXT_RETURN(pVM, false);
3220 return pVM->hm.s.fVirtApicRegs;
3221}
3222
3223
3224/**
3225 * Checks if APIC posted-interrupt processing is enabled.
3226 *
3227 * This returns whether we can deliver interrupts to the guest without
3228 * leaving guest-context by updating APIC state from host-context.
3229 *
3230 * @returns true if APIC posted-interrupt processing is enabled,
3231 * otherwise false.
3232 * @param pUVM The user mode VM handle.
3233 */
3234VMMR3DECL(bool) HMR3IsPostedIntrsEnabled(PUVM pUVM)
3235{
3236 UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
3237 PVM pVM = pUVM->pVM;
3238 VM_ASSERT_VALID_EXT_RETURN(pVM, false);
3239 return pVM->hm.s.fPostedIntrs;
3240}
3241
3242
3243/**
3244 * Checks if we are currently using VPID in VT-x mode.
3245 *
3246 * @returns true if VPID is being used, otherwise false.
3247 * @param pUVM The user mode VM handle.
3248 */
3249VMMR3DECL(bool) HMR3IsVpidActive(PUVM pUVM)
3250{
3251 UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
3252 PVM pVM = pUVM->pVM;
3253 VM_ASSERT_VALID_EXT_RETURN(pVM, false);
3254 return pVM->hm.s.vmx.fVpid;
3255}
3256
3257
3258/**
3259 * Checks if we are currently using VT-x unrestricted execution,
3260 * aka UX.
3261 *
3262 * @returns true if UX is being used, otherwise false.
3263 * @param pUVM The user mode VM handle.
3264 */
3265VMMR3DECL(bool) HMR3IsUXActive(PUVM pUVM)
3266{
3267 UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
3268 PVM pVM = pUVM->pVM;
3269 VM_ASSERT_VALID_EXT_RETURN(pVM, false);
3270 return pVM->hm.s.vmx.fUnrestrictedGuest;
3271}
3272
3273
3274/**
3275 * Checks if internal events are pending. In that case we are not allowed to dispatch interrupts.
3276 *
3277 * @returns true if an internal event is pending, otherwise false.
3278 * @param pVCpu The cross context virtual CPU structure.
3279 */
3280VMMR3_INT_DECL(bool) HMR3IsEventPending(PVMCPU pVCpu)
3281{
3282 return HMIsEnabled(pVCpu->pVMR3) && pVCpu->hm.s.Event.fPending;
3283}
3284
3285
3286/**
3287 * Checks if the VMX-preemption timer is being used.
3288 *
3289 * @returns true if the VMX-preemption timer is being used, otherwise false.
3290 * @param pVM The cross context VM structure.
3291 */
3292VMMR3_INT_DECL(bool) HMR3IsVmxPreemptionTimerUsed(PVM pVM)
3293{
3294 return HMIsEnabled(pVM)
3295 && pVM->hm.s.vmx.fEnabled
3296 && pVM->hm.s.vmx.fUsePreemptTimer;
3297}
3298
3299
3300/**
3301 * Restart an I/O instruction that was refused in ring-0
3302 *
3303 * @returns Strict VBox status code. Informational status codes other than the one documented
3304 * here are to be treated as internal failure. Use IOM_SUCCESS() to check for success.
3305 * @retval VINF_SUCCESS Success.
3306 * @retval VINF_EM_FIRST-VINF_EM_LAST Success with some exceptions (see IOM_SUCCESS()), the
3307 * status code must be passed on to EM.
3308 * @retval VERR_NOT_FOUND if no pending I/O instruction.
3309 *
3310 * @param pVM The cross context VM structure.
3311 * @param pVCpu The cross context virtual CPU structure.
3312 * @param pCtx Pointer to the guest CPU context.
3313 */
3314VMMR3_INT_DECL(VBOXSTRICTRC) HMR3RestartPendingIOInstr(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
3315{
3316 /*
3317 * Check if we've got relevant data pending.
3318 */
3319 HMPENDINGIO enmType = pVCpu->hm.s.PendingIO.enmType;
3320 if (enmType == HMPENDINGIO_INVALID)
3321 return VERR_NOT_FOUND;
3322 pVCpu->hm.s.PendingIO.enmType = HMPENDINGIO_INVALID;
3323 if (pVCpu->hm.s.PendingIO.GCPtrRip != pCtx->rip)
3324 return VERR_NOT_FOUND;
3325
3326 /*
3327 * Execute pending I/O.
3328 */
3329 VBOXSTRICTRC rcStrict;
3330 switch (enmType)
3331 {
3332 case HMPENDINGIO_PORT_READ:
3333 {
3334 uint32_t uAndVal = pVCpu->hm.s.PendingIO.s.Port.uAndVal;
3335 uint32_t u32Val = 0;
3336
3337 rcStrict = IOMIOPortRead(pVM, pVCpu, pVCpu->hm.s.PendingIO.s.Port.uPort, &u32Val,
3338 pVCpu->hm.s.PendingIO.s.Port.cbSize);
3339 if (IOM_SUCCESS(rcStrict))
3340 {
3341 /* Write back to the EAX register. */
3342 pCtx->eax = (pCtx->eax & ~uAndVal) | (u32Val & uAndVal);
3343 pCtx->rip = pVCpu->hm.s.PendingIO.GCPtrRipNext;
3344 }
3345 break;
3346 }
3347
3348 default:
3349 AssertLogRelFailedReturn(VERR_HM_UNKNOWN_IO_INSTRUCTION);
3350 }
3351
3352 if (IOM_SUCCESS(rcStrict))
3353 {
3354 /*
3355 * Check for I/O breakpoints.
3356 */
3357 uint32_t const uDr7 = pCtx->dr[7];
3358 if ( ( (uDr7 & X86_DR7_ENABLED_MASK)
3359 && X86_DR7_ANY_RW_IO(uDr7)
3360 && (pCtx->cr4 & X86_CR4_DE))
3361 || DBGFBpIsHwIoArmed(pVM))
3362 {
3363 VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pCtx, pVCpu->hm.s.PendingIO.s.Port.uPort,
3364 pVCpu->hm.s.PendingIO.s.Port.cbSize);
3365 if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
3366 rcStrict2 = TRPMAssertTrap(pVCpu, X86_XCPT_DB, TRPM_TRAP);
3367 /* rcStrict is VINF_SUCCESS or in [VINF_EM_FIRST..VINF_EM_LAST]. */
3368 else if (rcStrict2 != VINF_SUCCESS && (rcStrict == VINF_SUCCESS || rcStrict2 < rcStrict))
3369 rcStrict = rcStrict2;
3370 }
3371 }
3372 return rcStrict;
3373}
3374
3375
3376/**
3377 * Check fatal VT-x/AMD-V error and produce some meaningful
3378 * log release message.
3379 *
3380 * @param pVM The cross context VM structure.
3381 * @param iStatusCode VBox status code.
3382 */
3383VMMR3_INT_DECL(void) HMR3CheckError(PVM pVM, int iStatusCode)
3384{
3385 for (VMCPUID i = 0; i < pVM->cCpus; i++)
3386 {
3387 PVMCPU pVCpu = &pVM->aCpus[i];
3388 switch (iStatusCode)
3389 {
3390 /** @todo r=ramshankar: Are all EMTs out of ring-0 at this point!? If not, we
3391 * might be getting inaccurate values for non-guru'ing EMTs. */
3392 case VERR_VMX_INVALID_VMCS_FIELD:
3393 break;
3394
3395 case VERR_VMX_INVALID_VMCS_PTR:
3396 LogRel(("HM: VERR_VMX_INVALID_VMCS_PTR:\n"));
3397 LogRel(("HM: CPU[%u] Current pointer %#RGp vs %#RGp\n", i, pVCpu->hm.s.vmx.LastError.u64VMCSPhys,
3398 pVCpu->hm.s.vmx.HCPhysVmcs));
3399 LogRel(("HM: CPU[%u] Current VMCS version %#x\n", i, pVCpu->hm.s.vmx.LastError.u32VMCSRevision));
3400 LogRel(("HM: CPU[%u] Entered Host Cpu %u\n", i, pVCpu->hm.s.vmx.LastError.idEnteredCpu));
3401 LogRel(("HM: CPU[%u] Current Host Cpu %u\n", i, pVCpu->hm.s.vmx.LastError.idCurrentCpu));
3402 break;
3403
3404 case VERR_VMX_UNABLE_TO_START_VM:
3405 LogRel(("HM: VERR_VMX_UNABLE_TO_START_VM:\n"));
3406 LogRel(("HM: CPU[%u] Instruction error %#x\n", i, pVCpu->hm.s.vmx.LastError.u32InstrError));
3407 LogRel(("HM: CPU[%u] Exit reason %#x\n", i, pVCpu->hm.s.vmx.LastError.u32ExitReason));
3408
3409 if ( pVM->aCpus[i].hm.s.vmx.LastError.u32InstrError == VMX_ERROR_VMLAUCH_NON_CLEAR_VMCS
3410 || pVM->aCpus[i].hm.s.vmx.LastError.u32InstrError == VMX_ERROR_VMRESUME_NON_LAUNCHED_VMCS)
3411 {
3412 LogRel(("HM: CPU[%u] Entered Host Cpu %u\n", i, pVCpu->hm.s.vmx.LastError.idEnteredCpu));
3413 LogRel(("HM: CPU[%u] Current Host Cpu %u\n", i, pVCpu->hm.s.vmx.LastError.idCurrentCpu));
3414 }
3415 else if (pVM->aCpus[i].hm.s.vmx.LastError.u32InstrError == VMX_ERROR_VMENTRY_INVALID_CONTROL_FIELDS)
3416 {
3417 LogRel(("HM: CPU[%u] PinCtls %#RX32\n", i, pVCpu->hm.s.vmx.u32PinCtls));
3418 LogRel(("HM: CPU[%u] ProcCtls %#RX32\n", i, pVCpu->hm.s.vmx.u32ProcCtls));
3419 LogRel(("HM: CPU[%u] ProcCtls2 %#RX32\n", i, pVCpu->hm.s.vmx.u32ProcCtls2));
3420 LogRel(("HM: CPU[%u] EntryCtls %#RX32\n", i, pVCpu->hm.s.vmx.u32EntryCtls));
3421 LogRel(("HM: CPU[%u] ExitCtls %#RX32\n", i, pVCpu->hm.s.vmx.u32ExitCtls));
3422 LogRel(("HM: CPU[%u] HCPhysMsrBitmap %#RHp\n", i, pVCpu->hm.s.vmx.HCPhysMsrBitmap));
3423 LogRel(("HM: CPU[%u] HCPhysGuestMsr %#RHp\n", i, pVCpu->hm.s.vmx.HCPhysGuestMsr));
3424 LogRel(("HM: CPU[%u] HCPhysHostMsr %#RHp\n", i, pVCpu->hm.s.vmx.HCPhysHostMsr));
3425 LogRel(("HM: CPU[%u] cMsrs %u\n", i, pVCpu->hm.s.vmx.cMsrs));
3426 }
3427 /** @todo Log VM-entry event injection control fields
3428 * VMX_VMCS_CTRL_ENTRY_IRQ_INFO, VMX_VMCS_CTRL_ENTRY_EXCEPTION_ERRCODE
3429 * and VMX_VMCS_CTRL_ENTRY_INSTR_LENGTH from the VMCS. */
3430 break;
3431
3432 /* The guru will dump the HM error and exit history. Nothing extra to report for these errors. */
3433 case VERR_VMX_INVALID_VMXON_PTR:
3434 case VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO:
3435 case VERR_VMX_INVALID_GUEST_STATE:
3436 case VERR_VMX_UNEXPECTED_EXIT:
3437 case VERR_SVM_UNKNOWN_EXIT:
3438 case VERR_SVM_UNEXPECTED_EXIT:
3439 case VERR_SVM_UNEXPECTED_PATCH_TYPE:
3440 case VERR_SVM_UNEXPECTED_XCPT_EXIT:
3441 case VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE:
3442 break;
3443 }
3444 }
3445
3446 if (iStatusCode == VERR_VMX_UNABLE_TO_START_VM)
3447 {
3448 LogRel(("HM: VERR_VMX_UNABLE_TO_START_VM: VM-entry allowed %#RX32\n", pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1));
3449 LogRel(("HM: VERR_VMX_UNABLE_TO_START_VM: VM-entry disallowed %#RX32\n", pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0));
3450 }
3451 else if (iStatusCode == VERR_VMX_INVALID_VMXON_PTR)
3452 LogRel(("HM: HCPhysVmxEnableError = %#RHp\n", pVM->hm.s.vmx.HCPhysVmxEnableError));
3453}
3454
3455
3456/**
3457 * Execute state save operation.
3458 *
3459 * Save only data that cannot be re-loaded while entering HM ring-0 code. This
3460 * is because we always save the VM state from ring-3 and thus most HM state
3461 * will be re-synced dynamically at runtime and don't need to be part of the VM
3462 * saved state.
3463 *
3464 * @returns VBox status code.
3465 * @param pVM The cross context VM structure.
3466 * @param pSSM SSM operation handle.
3467 */
3468static DECLCALLBACK(int) hmR3Save(PVM pVM, PSSMHANDLE pSSM)
3469{
3470 int rc;
3471
3472 Log(("hmR3Save:\n"));
3473
3474 for (VMCPUID i = 0; i < pVM->cCpus; i++)
3475 {
3476 Assert(!pVM->aCpus[i].hm.s.Event.fPending);
3477 if (pVM->cpum.ro.GuestFeatures.fSvm)
3478 {
3479 PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVM->aCpus[i].hm.s.svm.NstGstVmcbCache;
3480 rc = SSMR3PutU16(pSSM, pVmcbNstGstCache->u16InterceptRdCRx);
3481 rc |= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16InterceptWrCRx);
3482 rc |= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16InterceptRdDRx);
3483 rc |= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16InterceptWrDRx);
3484 rc |= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16PauseFilterThreshold);
3485 rc |= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16PauseFilterCount);
3486 rc |= SSMR3PutU32(pSSM, pVmcbNstGstCache->u32InterceptXcpt);
3487 rc |= SSMR3PutU64(pSSM, pVmcbNstGstCache->u64InterceptCtrl);
3488 rc |= SSMR3PutU64(pSSM, pVmcbNstGstCache->u64TSCOffset);
3489 rc |= SSMR3PutBool(pSSM, pVmcbNstGstCache->fVIntrMasking);
3490 rc |= SSMR3PutBool(pSSM, pVmcbNstGstCache->fNestedPaging);
3491 rc |= SSMR3PutBool(pSSM, pVmcbNstGstCache->fLbrVirt);
3492 AssertRCReturn(rc, rc);
3493 }
3494 }
3495
3496 /* Save the guest patch data. */
3497 rc = SSMR3PutGCPtr(pSSM, pVM->hm.s.pGuestPatchMem);
3498 rc |= SSMR3PutGCPtr(pSSM, pVM->hm.s.pFreeGuestPatchMem);
3499 rc |= SSMR3PutU32(pSSM, pVM->hm.s.cbGuestPatchMem);
3500
3501 /* Store all the guest patch records too. */
3502 rc |= SSMR3PutU32(pSSM, pVM->hm.s.cPatches);
3503 AssertRCReturn(rc, rc);
3504
3505 for (uint32_t i = 0; i < pVM->hm.s.cPatches; i++)
3506 {
3507 AssertCompileSize(HMTPRINSTR, 4);
3508 PCHMTPRPATCH pPatch = &pVM->hm.s.aPatches[i];
3509 rc = SSMR3PutU32(pSSM, pPatch->Core.Key);
3510 rc |= SSMR3PutMem(pSSM, pPatch->aOpcode, sizeof(pPatch->aOpcode));
3511 rc |= SSMR3PutU32(pSSM, pPatch->cbOp);
3512 rc |= SSMR3PutMem(pSSM, pPatch->aNewOpcode, sizeof(pPatch->aNewOpcode));
3513 rc |= SSMR3PutU32(pSSM, pPatch->cbNewOp);
3514 rc |= SSMR3PutU32(pSSM, (uint32_t)pPatch->enmType);
3515 rc |= SSMR3PutU32(pSSM, pPatch->uSrcOperand);
3516 rc |= SSMR3PutU32(pSSM, pPatch->uDstOperand);
3517 rc |= SSMR3PutU32(pSSM, pPatch->pJumpTarget);
3518 rc |= SSMR3PutU32(pSSM, pPatch->cFaults);
3519 AssertRCReturn(rc, rc);
3520 }
3521
3522 return VINF_SUCCESS;
3523}
3524
3525
3526/**
3527 * Execute state load operation.
3528 *
3529 * @returns VBox status code.
3530 * @param pVM The cross context VM structure.
3531 * @param pSSM SSM operation handle.
3532 * @param uVersion Data layout version.
3533 * @param uPass The data pass.
3534 */
3535static DECLCALLBACK(int) hmR3Load(PVM pVM, PSSMHANDLE pSSM, uint32_t uVersion, uint32_t uPass)
3536{
3537 int rc;
3538
3539 LogFlowFunc(("uVersion=%u\n", uVersion));
3540 Assert(uPass == SSM_PASS_FINAL); NOREF(uPass);
3541
3542 /*
3543 * Validate version.
3544 */
3545 if ( uVersion != HM_SAVED_STATE_VERSION_SVM_NESTED_HWVIRT
3546 && uVersion != HM_SAVED_STATE_VERSION_TPR_PATCHING
3547 && uVersion != HM_SAVED_STATE_VERSION_NO_TPR_PATCHING
3548 && uVersion != HM_SAVED_STATE_VERSION_2_0_X)
3549 {
3550 AssertMsgFailed(("hmR3Load: Invalid version uVersion=%d!\n", uVersion));
3551 return VERR_SSM_UNSUPPORTED_DATA_UNIT_VERSION;
3552 }
3553
3554 /*
3555 * Load per-VCPU state.
3556 */
3557 for (VMCPUID i = 0; i < pVM->cCpus; i++)
3558 {
3559 if (uVersion >= HM_SAVED_STATE_VERSION_SVM_NESTED_HWVIRT)
3560 {
3561 /* Load the SVM nested hw.virt state if the VM is configured for it. */
3562 if (pVM->cpum.ro.GuestFeatures.fSvm)
3563 {
3564 PSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVM->aCpus[i].hm.s.svm.NstGstVmcbCache;
3565 rc = SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16InterceptRdCRx);
3566 rc |= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16InterceptWrCRx);
3567 rc |= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16InterceptRdDRx);
3568 rc |= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16InterceptWrDRx);
3569 rc |= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16PauseFilterThreshold);
3570 rc |= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16PauseFilterCount);
3571 rc |= SSMR3GetU32(pSSM, &pVmcbNstGstCache->u32InterceptXcpt);
3572 rc |= SSMR3GetU64(pSSM, &pVmcbNstGstCache->u64InterceptCtrl);
3573 rc |= SSMR3GetU64(pSSM, &pVmcbNstGstCache->u64TSCOffset);
3574 rc |= SSMR3GetBool(pSSM, &pVmcbNstGstCache->fVIntrMasking);
3575 rc |= SSMR3GetBool(pSSM, &pVmcbNstGstCache->fNestedPaging);
3576 rc |= SSMR3GetBool(pSSM, &pVmcbNstGstCache->fLbrVirt);
3577 AssertRCReturn(rc, rc);
3578 }
3579 }
3580 else
3581 {
3582 /* Pending HM event (obsolete for a long time since TPRM holds the info.) */
3583 rc = SSMR3GetU32(pSSM, &pVM->aCpus[i].hm.s.Event.fPending);
3584 rc |= SSMR3GetU32(pSSM, &pVM->aCpus[i].hm.s.Event.u32ErrCode);
3585 rc |= SSMR3GetU64(pSSM, &pVM->aCpus[i].hm.s.Event.u64IntInfo);
3586
3587 /* VMX fWasInRealMode related data. */
3588 uint32_t uDummy;
3589 rc |= SSMR3GetU32(pSSM, &uDummy); AssertRCReturn(rc, rc);
3590 rc |= SSMR3GetU32(pSSM, &uDummy); AssertRCReturn(rc, rc);
3591 rc |= SSMR3GetU32(pSSM, &uDummy); AssertRCReturn(rc, rc);
3592 AssertRCReturn(rc, rc);
3593 }
3594 }
3595
3596 /*
3597 * Load TPR patching data.
3598 */
3599 if (uVersion >= HM_SAVED_STATE_VERSION_TPR_PATCHING)
3600 {
3601 rc = SSMR3GetGCPtr(pSSM, &pVM->hm.s.pGuestPatchMem);
3602 rc |= SSMR3GetGCPtr(pSSM, &pVM->hm.s.pFreeGuestPatchMem);
3603 rc |= SSMR3GetU32(pSSM, &pVM->hm.s.cbGuestPatchMem);
3604
3605 /* Fetch all TPR patch records. */
3606 rc |= SSMR3GetU32(pSSM, &pVM->hm.s.cPatches);
3607 AssertRCReturn(rc, rc);
3608 for (uint32_t i = 0; i < pVM->hm.s.cPatches; i++)
3609 {
3610 PHMTPRPATCH pPatch = &pVM->hm.s.aPatches[i];
3611 rc = SSMR3GetU32(pSSM, &pPatch->Core.Key);
3612 rc |= SSMR3GetMem(pSSM, pPatch->aOpcode, sizeof(pPatch->aOpcode));
3613 rc |= SSMR3GetU32(pSSM, &pPatch->cbOp);
3614 rc |= SSMR3GetMem(pSSM, pPatch->aNewOpcode, sizeof(pPatch->aNewOpcode));
3615 rc |= SSMR3GetU32(pSSM, &pPatch->cbNewOp);
3616 rc |= SSMR3GetU32(pSSM, (uint32_t *)&pPatch->enmType);
3617
3618 if (pPatch->enmType == HMTPRINSTR_JUMP_REPLACEMENT)
3619 pVM->hm.s.fTPRPatchingActive = true;
3620 Assert(pPatch->enmType == HMTPRINSTR_JUMP_REPLACEMENT || pVM->hm.s.fTPRPatchingActive == false);
3621
3622 rc |= SSMR3GetU32(pSSM, &pPatch->uSrcOperand);
3623 rc |= SSMR3GetU32(pSSM, &pPatch->uDstOperand);
3624 rc |= SSMR3GetU32(pSSM, &pPatch->cFaults);
3625 rc |= SSMR3GetU32(pSSM, &pPatch->pJumpTarget);
3626 AssertRCReturn(rc, rc);
3627
3628 LogFlow(("hmR3Load: patch %d\n", i));
3629 LogFlow(("Key = %x\n", pPatch->Core.Key));
3630 LogFlow(("cbOp = %d\n", pPatch->cbOp));
3631 LogFlow(("cbNewOp = %d\n", pPatch->cbNewOp));
3632 LogFlow(("type = %d\n", pPatch->enmType));
3633 LogFlow(("srcop = %d\n", pPatch->uSrcOperand));
3634 LogFlow(("dstop = %d\n", pPatch->uDstOperand));
3635 LogFlow(("cFaults = %d\n", pPatch->cFaults));
3636 LogFlow(("target = %x\n", pPatch->pJumpTarget));
3637
3638 rc = RTAvloU32Insert(&pVM->hm.s.PatchTree, &pPatch->Core);
3639 AssertRCReturn(rc, rc);
3640 }
3641 }
3642
3643 return VINF_SUCCESS;
3644}
3645
3646
3647/**
3648 * Gets the name of a VT-x exit code.
3649 *
3650 * @returns Pointer to read only string if @a uExit is known, otherwise NULL.
3651 * @param uExit The VT-x exit to name.
3652 */
3653VMMR3DECL(const char *) HMR3GetVmxExitName(uint32_t uExit)
3654{
3655 if (uExit < RT_ELEMENTS(g_apszVTxExitReasons))
3656 return g_apszVTxExitReasons[uExit];
3657 return NULL;
3658}
3659
3660
3661/**
3662 * Gets the name of an AMD-V exit code.
3663 *
3664 * @returns Pointer to read only string if @a uExit is known, otherwise NULL.
3665 * @param uExit The AMD-V exit to name.
3666 */
3667VMMR3DECL(const char *) HMR3GetSvmExitName(uint32_t uExit)
3668{
3669 if (uExit < RT_ELEMENTS(g_apszAmdVExitReasons))
3670 return g_apszAmdVExitReasons[uExit];
3671 return hmSvmGetSpecialExitReasonDesc(uExit);
3672}
3673
3674
3675/**
3676 * Displays the guest VM-exit history.
3677 *
3678 * @param pVM The cross context VM structure.
3679 * @param pHlp The info helper functions.
3680 * @param pszArgs Arguments, ignored.
3681 */
3682static DECLCALLBACK(void) hmR3InfoExitHistory(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs)
3683{
3684 NOREF(pszArgs);
3685 PVMCPU pVCpu = VMMGetCpu(pVM);
3686 if (!pVCpu)
3687 pVCpu = &pVM->aCpus[0];
3688
3689 if (HMIsEnabled(pVM))
3690 {
3691 bool const fIsVtx = pVM->hm.s.vmx.fSupported;
3692 const char * const *papszDesc;
3693 unsigned cMaxExitDesc;
3694 if (fIsVtx)
3695 {
3696 cMaxExitDesc = MAX_EXITREASON_VTX;
3697 papszDesc = &g_apszVTxExitReasons[0];
3698 pHlp->pfnPrintf(pHlp, "CPU[%u]: VT-x VM-exit history:\n", pVCpu->idCpu);
3699 }
3700 else
3701 {
3702 cMaxExitDesc = MAX_EXITREASON_AMDV;
3703 papszDesc = &g_apszAmdVExitReasons[0];
3704 pHlp->pfnPrintf(pHlp, "CPU[%u]: AMD-V #VMEXIT history:\n", pVCpu->idCpu);
3705 }
3706
3707 pHlp->pfnPrintf(pHlp, " idxExitHistoryFree = %u\n", pVCpu->hm.s.idxExitHistoryFree);
3708 unsigned const idxLast = pVCpu->hm.s.idxExitHistoryFree > 0 ?
3709 pVCpu->hm.s.idxExitHistoryFree - 1 :
3710 RT_ELEMENTS(pVCpu->hm.s.auExitHistory) - 1;
3711 for (unsigned i = 0; i < RT_ELEMENTS(pVCpu->hm.s.auExitHistory); i++)
3712 {
3713 uint16_t const uExit = pVCpu->hm.s.auExitHistory[i];
3714 const char *pszExit = NULL;
3715 if (uExit <= cMaxExitDesc)
3716 pszExit = papszDesc[uExit];
3717 else if (!fIsVtx)
3718 pszExit = hmSvmGetSpecialExitReasonDesc(uExit);
3719 else
3720 pszExit = NULL;
3721
3722 pHlp->pfnPrintf(pHlp, " auExitHistory[%2u] = 0x%04x %s %s\n", i, uExit, pszExit,
3723 idxLast == i ? "<-- Latest exit" : "");
3724 }
3725 pHlp->pfnPrintf(pHlp, "HM error = %#x (%u)\n", pVCpu->hm.s.u32HMError, pVCpu->hm.s.u32HMError);
3726 }
3727 else
3728 pHlp->pfnPrintf(pHlp, "HM is not enabled for this VM!\n");
3729}
3730
3731
3732/**
3733 * Displays the HM pending event.
3734 *
3735 * @param pVM The cross context VM structure.
3736 * @param pHlp The info helper functions.
3737 * @param pszArgs Arguments, ignored.
3738 */
3739static DECLCALLBACK(void) hmR3InfoEventPending(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs)
3740{
3741 NOREF(pszArgs);
3742 PVMCPU pVCpu = VMMGetCpu(pVM);
3743 if (!pVCpu)
3744 pVCpu = &pVM->aCpus[0];
3745
3746 if (HMIsEnabled(pVM))
3747 {
3748 pHlp->pfnPrintf(pHlp, "CPU[%u]: HM event (fPending=%RTbool)\n", pVCpu->idCpu, pVCpu->hm.s.Event.fPending);
3749 if (pVCpu->hm.s.Event.fPending)
3750 {
3751 pHlp->pfnPrintf(pHlp, " u64IntInfo = %#RX64\n", pVCpu->hm.s.Event.u64IntInfo);
3752 pHlp->pfnPrintf(pHlp, " u32ErrCode = %#RX64\n", pVCpu->hm.s.Event.u32ErrCode);
3753 pHlp->pfnPrintf(pHlp, " cbInstr = %u bytes\n", pVCpu->hm.s.Event.cbInstr);
3754 pHlp->pfnPrintf(pHlp, " GCPtrFaultAddress = %#RGp\n", pVCpu->hm.s.Event.GCPtrFaultAddress);
3755 }
3756 }
3757 else
3758 pHlp->pfnPrintf(pHlp, "HM is not enabled for this VM!\n");
3759}
3760
3761
3762/**
3763 * Displays the SVM nested-guest VMCB cache.
3764 *
3765 * @param pVM The cross context VM structure.
3766 * @param pHlp The info helper functions.
3767 * @param pszArgs Arguments, ignored.
3768 */
3769static DECLCALLBACK(void) hmR3InfoSvmNstGstVmcbCache(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs)
3770{
3771 NOREF(pszArgs);
3772 PVMCPU pVCpu = VMMGetCpu(pVM);
3773 if (!pVCpu)
3774 pVCpu = &pVM->aCpus[0];
3775
3776 bool const fSvmEnabled = HMR3IsSvmEnabled(pVM->pUVM);
3777 if ( fSvmEnabled
3778 && pVM->cpum.ro.GuestFeatures.fSvm)
3779 {
3780 PCCPUMCTX pCtx = CPUMQueryGuestCtxPtr(pVCpu);
3781 PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVCpu->hm.s.svm.NstGstVmcbCache;
3782 pHlp->pfnPrintf(pHlp, "CPU[%u]: HM SVM nested-guest VMCB cache\n", pVCpu->idCpu);
3783 pHlp->pfnPrintf(pHlp, " fHMCachedVmcb = %#RTbool\n", pCtx->hwvirt.svm.fHMCachedVmcb);
3784 pHlp->pfnPrintf(pHlp, " u16InterceptRdCRx = %#RX16\n", pVmcbNstGstCache->u16InterceptRdCRx);
3785 pHlp->pfnPrintf(pHlp, " u16InterceptWrCRx = %#RX16\n", pVmcbNstGstCache->u16InterceptWrCRx);
3786 pHlp->pfnPrintf(pHlp, " u16InterceptRdDRx = %#RX16\n", pVmcbNstGstCache->u16InterceptRdDRx);
3787 pHlp->pfnPrintf(pHlp, " u16InterceptWrDRx = %#RX16\n", pVmcbNstGstCache->u16InterceptWrDRx);
3788 pHlp->pfnPrintf(pHlp, " u16PauseFilterThreshold = %#RX16\n", pVmcbNstGstCache->u16PauseFilterThreshold);
3789 pHlp->pfnPrintf(pHlp, " u16PauseFilterCount = %#RX16\n", pVmcbNstGstCache->u16PauseFilterCount);
3790 pHlp->pfnPrintf(pHlp, " u32InterceptXcpt = %#RX32\n", pVmcbNstGstCache->u32InterceptXcpt);
3791 pHlp->pfnPrintf(pHlp, " u64InterceptCtrl = %#RX64\n", pVmcbNstGstCache->u64InterceptCtrl);
3792 pHlp->pfnPrintf(pHlp, " u64TSCOffset = %#RX64\n", pVmcbNstGstCache->u64TSCOffset);
3793 pHlp->pfnPrintf(pHlp, " fVIntrMasking = %RTbool\n", pVmcbNstGstCache->fVIntrMasking);
3794 pHlp->pfnPrintf(pHlp, " fNestedPaging = %RTbool\n", pVmcbNstGstCache->fNestedPaging);
3795 pHlp->pfnPrintf(pHlp, " fLbrVirt = %RTbool\n", pVmcbNstGstCache->fLbrVirt);
3796 }
3797 else
3798 {
3799 if (!fSvmEnabled)
3800 pHlp->pfnPrintf(pHlp, "HM SVM is not enabled for this VM!\n");
3801 else
3802 pHlp->pfnPrintf(pHlp, "SVM feature is not exposed to the guest!\n");
3803 }
3804}
3805
Note: See TracBrowser for help on using the repository browser.

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette