VirtualBox

source: vbox/trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp@ 56815

Last change on this file since 56815 was 56815, checked in by vboxsync, 10 years ago

VMM/HMVMXR0: nit.
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 513.7 KB
Line 
1/* $Id: HMVMXR0.cpp 56815 2015-07-06 11:25:42Z vboxsync $ */
2/** @file
3 * HM VMX (Intel VT-x) - Host Context Ring-0.
4 */
5
6/*
7 * Copyright (C) 2012-2015 Oracle Corporation
8 *
9 * This file is part of VirtualBox Open Source Edition (OSE), as
10 * available from http://www.virtualbox.org. This file is free software;
11 * you can redistribute it and/or modify it under the terms of the GNU
12 * General Public License (GPL) as published by the Free Software
13 * Foundation, in version 2 as it comes in the "COPYING" file of the
14 * VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15 * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16 */
17
18/*******************************************************************************
19* Header Files *
20*******************************************************************************/
21#define LOG_GROUP LOG_GROUP_HM
22#include <iprt/x86.h>
23#include <iprt/asm-amd64-x86.h>
24#include <iprt/thread.h>
25
26#include <VBox/vmm/pdmapi.h>
27#include <VBox/vmm/dbgf.h>
28#include <VBox/vmm/iem.h>
29#include <VBox/vmm/iom.h>
30#include <VBox/vmm/selm.h>
31#include <VBox/vmm/tm.h>
32#include <VBox/vmm/gim.h>
33#ifdef VBOX_WITH_REM
34# include <VBox/vmm/rem.h>
35#endif
36#include "HMInternal.h"
37#include <VBox/vmm/vm.h>
38#include "HMVMXR0.h"
39#include "dtrace/VBoxVMM.h"
40
41#ifdef DEBUG_ramshankar
42# define HMVMX_ALWAYS_SAVE_GUEST_RFLAGS
43# define HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE
44# define HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
45# define HMVMX_ALWAYS_CHECK_GUEST_STATE
46# define HMVMX_ALWAYS_TRAP_ALL_XCPTS
47# define HMVMX_ALWAYS_TRAP_PF
48# define HMVMX_ALWAYS_SWAP_FPU_STATE
49# define HMVMX_ALWAYS_FLUSH_TLB
50# define HMVMX_ALWAYS_SWAP_EFER
51#endif
52
53
54/*******************************************************************************
55* Defined Constants And Macros *
56*******************************************************************************/
57#if defined(RT_ARCH_AMD64)
58# define HMVMX_IS_64BIT_HOST_MODE() (true)
59typedef RTHCUINTREG HMVMXHCUINTREG;
60#elif defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
61extern "C" uint32_t g_fVMXIs64bitHost;
62# define HMVMX_IS_64BIT_HOST_MODE() (g_fVMXIs64bitHost != 0)
63typedef uint64_t HMVMXHCUINTREG;
64#else
65# define HMVMX_IS_64BIT_HOST_MODE() (false)
66typedef RTHCUINTREG HMVMXHCUINTREG;
67#endif
68
69/** Use the function table. */
70#define HMVMX_USE_FUNCTION_TABLE
71
72/** Determine which tagged-TLB flush handler to use. */
73#define HMVMX_FLUSH_TAGGED_TLB_EPT_VPID 0
74#define HMVMX_FLUSH_TAGGED_TLB_EPT 1
75#define HMVMX_FLUSH_TAGGED_TLB_VPID 2
76#define HMVMX_FLUSH_TAGGED_TLB_NONE 3
77
78/** @name Updated-guest-state flags.
79 * @{ */
80#define HMVMX_UPDATED_GUEST_RIP RT_BIT(0)
81#define HMVMX_UPDATED_GUEST_RSP RT_BIT(1)
82#define HMVMX_UPDATED_GUEST_RFLAGS RT_BIT(2)
83#define HMVMX_UPDATED_GUEST_CR0 RT_BIT(3)
84#define HMVMX_UPDATED_GUEST_CR3 RT_BIT(4)
85#define HMVMX_UPDATED_GUEST_CR4 RT_BIT(5)
86#define HMVMX_UPDATED_GUEST_GDTR RT_BIT(6)
87#define HMVMX_UPDATED_GUEST_IDTR RT_BIT(7)
88#define HMVMX_UPDATED_GUEST_LDTR RT_BIT(8)
89#define HMVMX_UPDATED_GUEST_TR RT_BIT(9)
90#define HMVMX_UPDATED_GUEST_SEGMENT_REGS RT_BIT(10)
91#define HMVMX_UPDATED_GUEST_DEBUG RT_BIT(11)
92#define HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR RT_BIT(12)
93#define HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR RT_BIT(13)
94#define HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR RT_BIT(14)
95#define HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS RT_BIT(15)
96#define HMVMX_UPDATED_GUEST_LAZY_MSRS RT_BIT(16)
97#define HMVMX_UPDATED_GUEST_ACTIVITY_STATE RT_BIT(17)
98#define HMVMX_UPDATED_GUEST_INTR_STATE RT_BIT(18)
99#define HMVMX_UPDATED_GUEST_APIC_STATE RT_BIT(19)
100#define HMVMX_UPDATED_GUEST_ALL ( HMVMX_UPDATED_GUEST_RIP \
101 | HMVMX_UPDATED_GUEST_RSP \
102 | HMVMX_UPDATED_GUEST_RFLAGS \
103 | HMVMX_UPDATED_GUEST_CR0 \
104 | HMVMX_UPDATED_GUEST_CR3 \
105 | HMVMX_UPDATED_GUEST_CR4 \
106 | HMVMX_UPDATED_GUEST_GDTR \
107 | HMVMX_UPDATED_GUEST_IDTR \
108 | HMVMX_UPDATED_GUEST_LDTR \
109 | HMVMX_UPDATED_GUEST_TR \
110 | HMVMX_UPDATED_GUEST_SEGMENT_REGS \
111 | HMVMX_UPDATED_GUEST_DEBUG \
112 | HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR \
113 | HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR \
114 | HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR \
115 | HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS \
116 | HMVMX_UPDATED_GUEST_LAZY_MSRS \
117 | HMVMX_UPDATED_GUEST_ACTIVITY_STATE \
118 | HMVMX_UPDATED_GUEST_INTR_STATE \
119 | HMVMX_UPDATED_GUEST_APIC_STATE)
120/** @} */
121
122/** @name
123 * Flags to skip redundant reads of some common VMCS fields that are not part of
124 * the guest-CPU state but are in the transient structure.
125 */
126#define HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO RT_BIT(0)
127#define HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE RT_BIT(1)
128#define HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION RT_BIT(2)
129#define HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN RT_BIT(3)
130#define HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO RT_BIT(4)
131#define HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE RT_BIT(5)
132#define HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO RT_BIT(6)
133/** @} */
134
135/** @name
136 * States of the VMCS.
137 *
138 * This does not reflect all possible VMCS states but currently only those
139 * needed for maintaining the VMCS consistently even when thread-context hooks
140 * are used. Maybe later this can be extended (i.e. Nested Virtualization).
141 */
142#define HMVMX_VMCS_STATE_CLEAR RT_BIT(0)
143#define HMVMX_VMCS_STATE_ACTIVE RT_BIT(1)
144#define HMVMX_VMCS_STATE_LAUNCHED RT_BIT(2)
145/** @} */
146
147/**
148 * Exception bitmap mask for real-mode guests (real-on-v86).
149 *
150 * We need to intercept all exceptions manually except:
151 * - #NM, #MF handled in hmR0VmxLoadSharedCR0().
152 * - #DB handled in hmR0VmxLoadSharedDebugState().
153 * - #PF need not be intercepted even in real-mode if we have Nested Paging
154 * support.
155 */
156#define HMVMX_REAL_MODE_XCPT_MASK ( RT_BIT(X86_XCPT_DE) /* RT_BIT(X86_XCPT_DB) */ | RT_BIT(X86_XCPT_NMI) \
157 | RT_BIT(X86_XCPT_BP) | RT_BIT(X86_XCPT_OF) | RT_BIT(X86_XCPT_BR) \
158 | RT_BIT(X86_XCPT_UD) /* RT_BIT(X86_XCPT_NM) */ | RT_BIT(X86_XCPT_DF) \
159 | RT_BIT(X86_XCPT_CO_SEG_OVERRUN) | RT_BIT(X86_XCPT_TS) | RT_BIT(X86_XCPT_NP) \
160 | RT_BIT(X86_XCPT_SS) | RT_BIT(X86_XCPT_GP) /* RT_BIT(X86_XCPT_PF) */ \
161 /* RT_BIT(X86_XCPT_MF) */ | RT_BIT(X86_XCPT_AC) | RT_BIT(X86_XCPT_MC) \
162 | RT_BIT(X86_XCPT_XF))
163
164/**
165 * Exception bitmap mask for all contributory exceptions.
166 *
167 * Page fault is deliberately excluded here as it's conditional as to whether
168 * it's contributory or benign. Page faults are handled separately.
169 */
170#define HMVMX_CONTRIBUTORY_XCPT_MASK ( RT_BIT(X86_XCPT_GP) | RT_BIT(X86_XCPT_NP) | RT_BIT(X86_XCPT_SS) | RT_BIT(X86_XCPT_TS) \
171 | RT_BIT(X86_XCPT_DE))
172
173/** Maximum VM-instruction error number. */
174#define HMVMX_INSTR_ERROR_MAX 28
175
176/** Profiling macro. */
177#ifdef HM_PROFILE_EXIT_DISPATCH
178# define HMVMX_START_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitDispatch, ed)
179# define HMVMX_STOP_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitDispatch, ed)
180#else
181# define HMVMX_START_EXIT_DISPATCH_PROF() do { } while (0)
182# define HMVMX_STOP_EXIT_DISPATCH_PROF() do { } while (0)
183#endif
184
185/** Assert that preemption is disabled or covered by thread-context hooks. */
186#define HMVMX_ASSERT_PREEMPT_SAFE() Assert( VMMR0ThreadCtxHookIsEnabled(pVCpu) \
187 || !RTThreadPreemptIsEnabled(NIL_RTTHREAD));
188
189/** Assert that we haven't migrated CPUs when thread-context hooks are not
190 * used. */
191#define HMVMX_ASSERT_CPU_SAFE() AssertMsg( VMMR0ThreadCtxHookIsEnabled(pVCpu) \
192 || pVCpu->hm.s.idEnteredCpu == RTMpCpuId(), \
193 ("Illegal migration! Entered on CPU %u Current %u\n", \
194 pVCpu->hm.s.idEnteredCpu, RTMpCpuId())); \
195
196/** Helper macro for VM-exit handlers called unexpectedly. */
197#define HMVMX_RETURN_UNEXPECTED_EXIT() \
198 do { \
199 pVCpu->hm.s.u32HMError = pVmxTransient->uExitReason; \
200 return VERR_VMX_UNEXPECTED_EXIT; \
201 } while (0)
202
203
204/*******************************************************************************
205* Structures and Typedefs *
206*******************************************************************************/
207/**
208 * VMX transient state.
209 *
210 * A state structure for holding miscellaneous information across
211 * VMX non-root operation and restored after the transition.
212 */
213typedef struct VMXTRANSIENT
214{
215 /** The host's rflags/eflags. */
216 RTCCUINTREG fEFlags;
217#if HC_ARCH_BITS == 32
218 uint32_t u32Alignment0;
219#endif
220 /** The guest's TPR value used for TPR shadowing. */
221 uint8_t u8GuestTpr;
222 /** Alignment. */
223 uint8_t abAlignment0[7];
224
225 /** The basic VM-exit reason. */
226 uint16_t uExitReason;
227 /** Alignment. */
228 uint16_t u16Alignment0;
229 /** The VM-exit interruption error code. */
230 uint32_t uExitIntErrorCode;
231 /** The VM-exit exit code qualification. */
232 uint64_t uExitQualification;
233
234 /** The VM-exit interruption-information field. */
235 uint32_t uExitIntInfo;
236 /** The VM-exit instruction-length field. */
237 uint32_t cbInstr;
238 /** The VM-exit instruction-information field. */
239 union
240 {
241 /** Plain unsigned int representation. */
242 uint32_t u;
243 /** INS and OUTS information. */
244 struct
245 {
246 uint32_t u6Reserved0 : 7;
247 /** The address size; 0=16-bit, 1=32-bit, 2=64-bit, rest undefined. */
248 uint32_t u3AddrSize : 3;
249 uint32_t u5Reserved1 : 5;
250 /** The segment register (X86_SREG_XXX). */
251 uint32_t iSegReg : 3;
252 uint32_t uReserved2 : 14;
253 } StrIo;
254 } ExitInstrInfo;
255 /** Whether the VM-entry failed or not. */
256 bool fVMEntryFailed;
257 /** Alignment. */
258 uint8_t abAlignment1[3];
259
260 /** The VM-entry interruption-information field. */
261 uint32_t uEntryIntInfo;
262 /** The VM-entry exception error code field. */
263 uint32_t uEntryXcptErrorCode;
264 /** The VM-entry instruction length field. */
265 uint32_t cbEntryInstr;
266
267 /** IDT-vectoring information field. */
268 uint32_t uIdtVectoringInfo;
269 /** IDT-vectoring error code. */
270 uint32_t uIdtVectoringErrorCode;
271
272 /** Mask of currently read VMCS fields; HMVMX_UPDATED_TRANSIENT_*. */
273 uint32_t fVmcsFieldsRead;
274
275 /** Whether the guest FPU was active at the time of VM-exit. */
276 bool fWasGuestFPUStateActive;
277 /** Whether the guest debug state was active at the time of VM-exit. */
278 bool fWasGuestDebugStateActive;
279 /** Whether the hyper debug state was active at the time of VM-exit. */
280 bool fWasHyperDebugStateActive;
281 /** Whether TSC-offsetting should be setup before VM-entry. */
282 bool fUpdateTscOffsettingAndPreemptTimer;
283 /** Whether the VM-exit was caused by a page-fault during delivery of a
284 * contributory exception or a page-fault. */
285 bool fVectoringDoublePF;
286 /** Whether the VM-exit was caused by a page-fault during delivery of an
287 * external interrupt or NMI. */
288 bool fVectoringPF;
289} VMXTRANSIENT;
290AssertCompileMemberAlignment(VMXTRANSIENT, uExitReason, sizeof(uint64_t));
291AssertCompileMemberAlignment(VMXTRANSIENT, uExitIntInfo, sizeof(uint64_t));
292AssertCompileMemberAlignment(VMXTRANSIENT, uEntryIntInfo, sizeof(uint64_t));
293AssertCompileMemberAlignment(VMXTRANSIENT, fWasGuestFPUStateActive, sizeof(uint64_t));
294AssertCompileMemberSize(VMXTRANSIENT, ExitInstrInfo, sizeof(uint32_t));
295/** Pointer to VMX transient state. */
296typedef VMXTRANSIENT *PVMXTRANSIENT;
297
298
299/**
300 * MSR-bitmap read permissions.
301 */
302typedef enum VMXMSREXITREAD
303{
304 /** Reading this MSR causes a VM-exit. */
305 VMXMSREXIT_INTERCEPT_READ = 0xb,
306 /** Reading this MSR does not cause a VM-exit. */
307 VMXMSREXIT_PASSTHRU_READ
308} VMXMSREXITREAD;
309/** Pointer to MSR-bitmap read permissions. */
310typedef VMXMSREXITREAD* PVMXMSREXITREAD;
311
312/**
313 * MSR-bitmap write permissions.
314 */
315typedef enum VMXMSREXITWRITE
316{
317 /** Writing to this MSR causes a VM-exit. */
318 VMXMSREXIT_INTERCEPT_WRITE = 0xd,
319 /** Writing to this MSR does not cause a VM-exit. */
320 VMXMSREXIT_PASSTHRU_WRITE
321} VMXMSREXITWRITE;
322/** Pointer to MSR-bitmap write permissions. */
323typedef VMXMSREXITWRITE* PVMXMSREXITWRITE;
324
325
326/**
327 * VMX VM-exit handler.
328 *
329 * @returns VBox status code.
330 * @param pVCpu Pointer to the VMCPU.
331 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
332 * out-of-sync. Make sure to update the required
333 * fields before using them.
334 * @param pVmxTransient Pointer to the VMX-transient structure.
335 */
336#ifndef HMVMX_USE_FUNCTION_TABLE
337typedef int FNVMXEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
338#else
339typedef DECLCALLBACK(int) FNVMXEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
340/** Pointer to VM-exit handler. */
341typedef FNVMXEXITHANDLER *PFNVMXEXITHANDLER;
342#endif
343
344
345/*******************************************************************************
346* Internal Functions *
347*******************************************************************************/
348static void hmR0VmxFlushEpt(PVMCPU pVCpu, VMXFLUSHEPT enmFlush);
349static void hmR0VmxFlushVpid(PVM pVM, PVMCPU pVCpu, VMXFLUSHVPID enmFlush, RTGCPTR GCPtr);
350static int hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint64_t u64IntInfo, uint32_t cbInstr,
351 uint32_t u32ErrCode, RTGCUINTREG GCPtrFaultAddress,
352 bool fStepping, uint32_t *puIntState);
353#if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
354static int hmR0VmxInitVmcsReadCache(PVM pVM, PVMCPU pVCpu);
355#endif
356#ifndef HMVMX_USE_FUNCTION_TABLE
357DECLINLINE(int) hmR0VmxHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, uint32_t rcReason);
358# define HMVMX_EXIT_DECL static int
359#else
360# define HMVMX_EXIT_DECL static DECLCALLBACK(int)
361#endif
362DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitStep(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient,
363 uint32_t uExitReason, uint16_t uCsStart, uint64_t uRipStart);
364
365/** @name VM-exit handlers.
366 * @{
367 */
368static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmi;
369static FNVMXEXITHANDLER hmR0VmxExitExtInt;
370static FNVMXEXITHANDLER hmR0VmxExitTripleFault;
371static FNVMXEXITHANDLER hmR0VmxExitInitSignal;
372static FNVMXEXITHANDLER hmR0VmxExitSipi;
373static FNVMXEXITHANDLER hmR0VmxExitIoSmi;
374static FNVMXEXITHANDLER hmR0VmxExitSmi;
375static FNVMXEXITHANDLER hmR0VmxExitIntWindow;
376static FNVMXEXITHANDLER hmR0VmxExitNmiWindow;
377static FNVMXEXITHANDLER hmR0VmxExitTaskSwitch;
378static FNVMXEXITHANDLER hmR0VmxExitCpuid;
379static FNVMXEXITHANDLER hmR0VmxExitGetsec;
380static FNVMXEXITHANDLER hmR0VmxExitHlt;
381static FNVMXEXITHANDLER hmR0VmxExitInvd;
382static FNVMXEXITHANDLER hmR0VmxExitInvlpg;
383static FNVMXEXITHANDLER hmR0VmxExitRdpmc;
384static FNVMXEXITHANDLER hmR0VmxExitVmcall;
385static FNVMXEXITHANDLER hmR0VmxExitRdtsc;
386static FNVMXEXITHANDLER hmR0VmxExitRsm;
387static FNVMXEXITHANDLER hmR0VmxExitSetPendingXcptUD;
388static FNVMXEXITHANDLER hmR0VmxExitMovCRx;
389static FNVMXEXITHANDLER hmR0VmxExitMovDRx;
390static FNVMXEXITHANDLER hmR0VmxExitIoInstr;
391static FNVMXEXITHANDLER hmR0VmxExitRdmsr;
392static FNVMXEXITHANDLER hmR0VmxExitWrmsr;
393static FNVMXEXITHANDLER hmR0VmxExitErrInvalidGuestState;
394static FNVMXEXITHANDLER hmR0VmxExitErrMsrLoad;
395static FNVMXEXITHANDLER hmR0VmxExitErrUndefined;
396static FNVMXEXITHANDLER hmR0VmxExitMwait;
397static FNVMXEXITHANDLER hmR0VmxExitMtf;
398static FNVMXEXITHANDLER hmR0VmxExitMonitor;
399static FNVMXEXITHANDLER hmR0VmxExitPause;
400static FNVMXEXITHANDLER hmR0VmxExitErrMachineCheck;
401static FNVMXEXITHANDLER hmR0VmxExitTprBelowThreshold;
402static FNVMXEXITHANDLER hmR0VmxExitApicAccess;
403static FNVMXEXITHANDLER hmR0VmxExitXdtrAccess;
404static FNVMXEXITHANDLER hmR0VmxExitXdtrAccess;
405static FNVMXEXITHANDLER hmR0VmxExitEptViolation;
406static FNVMXEXITHANDLER hmR0VmxExitEptMisconfig;
407static FNVMXEXITHANDLER hmR0VmxExitRdtscp;
408static FNVMXEXITHANDLER hmR0VmxExitPreemptTimer;
409static FNVMXEXITHANDLER hmR0VmxExitWbinvd;
410static FNVMXEXITHANDLER hmR0VmxExitXsetbv;
411static FNVMXEXITHANDLER hmR0VmxExitRdrand;
412static FNVMXEXITHANDLER hmR0VmxExitInvpcid;
413/** @} */
414
415static int hmR0VmxExitXcptNM(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
416static int hmR0VmxExitXcptPF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
417static int hmR0VmxExitXcptMF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
418static int hmR0VmxExitXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
419static int hmR0VmxExitXcptBP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
420static int hmR0VmxExitXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
421#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
422static int hmR0VmxExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
423#endif
424static uint32_t hmR0VmxCheckGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx);
425
426/*******************************************************************************
427* Global Variables *
428*******************************************************************************/
429#ifdef HMVMX_USE_FUNCTION_TABLE
430
431/**
432 * VMX_EXIT dispatch table.
433 */
434static const PFNVMXEXITHANDLER g_apfnVMExitHandlers[VMX_EXIT_MAX + 1] =
435{
436 /* 00 VMX_EXIT_XCPT_OR_NMI */ hmR0VmxExitXcptOrNmi,
437 /* 01 VMX_EXIT_EXT_INT */ hmR0VmxExitExtInt,
438 /* 02 VMX_EXIT_TRIPLE_FAULT */ hmR0VmxExitTripleFault,
439 /* 03 VMX_EXIT_INIT_SIGNAL */ hmR0VmxExitInitSignal,
440 /* 04 VMX_EXIT_SIPI */ hmR0VmxExitSipi,
441 /* 05 VMX_EXIT_IO_SMI */ hmR0VmxExitIoSmi,
442 /* 06 VMX_EXIT_SMI */ hmR0VmxExitSmi,
443 /* 07 VMX_EXIT_INT_WINDOW */ hmR0VmxExitIntWindow,
444 /* 08 VMX_EXIT_NMI_WINDOW */ hmR0VmxExitNmiWindow,
445 /* 09 VMX_EXIT_TASK_SWITCH */ hmR0VmxExitTaskSwitch,
446 /* 10 VMX_EXIT_CPUID */ hmR0VmxExitCpuid,
447 /* 11 VMX_EXIT_GETSEC */ hmR0VmxExitGetsec,
448 /* 12 VMX_EXIT_HLT */ hmR0VmxExitHlt,
449 /* 13 VMX_EXIT_INVD */ hmR0VmxExitInvd,
450 /* 14 VMX_EXIT_INVLPG */ hmR0VmxExitInvlpg,
451 /* 15 VMX_EXIT_RDPMC */ hmR0VmxExitRdpmc,
452 /* 16 VMX_EXIT_RDTSC */ hmR0VmxExitRdtsc,
453 /* 17 VMX_EXIT_RSM */ hmR0VmxExitRsm,
454 /* 18 VMX_EXIT_VMCALL */ hmR0VmxExitVmcall,
455 /* 19 VMX_EXIT_VMCLEAR */ hmR0VmxExitSetPendingXcptUD,
456 /* 20 VMX_EXIT_VMLAUNCH */ hmR0VmxExitSetPendingXcptUD,
457 /* 21 VMX_EXIT_VMPTRLD */ hmR0VmxExitSetPendingXcptUD,
458 /* 22 VMX_EXIT_VMPTRST */ hmR0VmxExitSetPendingXcptUD,
459 /* 23 VMX_EXIT_VMREAD */ hmR0VmxExitSetPendingXcptUD,
460 /* 24 VMX_EXIT_VMRESUME */ hmR0VmxExitSetPendingXcptUD,
461 /* 25 VMX_EXIT_VMWRITE */ hmR0VmxExitSetPendingXcptUD,
462 /* 26 VMX_EXIT_VMXOFF */ hmR0VmxExitSetPendingXcptUD,
463 /* 27 VMX_EXIT_VMXON */ hmR0VmxExitSetPendingXcptUD,
464 /* 28 VMX_EXIT_MOV_CRX */ hmR0VmxExitMovCRx,
465 /* 29 VMX_EXIT_MOV_DRX */ hmR0VmxExitMovDRx,
466 /* 30 VMX_EXIT_IO_INSTR */ hmR0VmxExitIoInstr,
467 /* 31 VMX_EXIT_RDMSR */ hmR0VmxExitRdmsr,
468 /* 32 VMX_EXIT_WRMSR */ hmR0VmxExitWrmsr,
469 /* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ hmR0VmxExitErrInvalidGuestState,
470 /* 34 VMX_EXIT_ERR_MSR_LOAD */ hmR0VmxExitErrMsrLoad,
471 /* 35 UNDEFINED */ hmR0VmxExitErrUndefined,
472 /* 36 VMX_EXIT_MWAIT */ hmR0VmxExitMwait,
473 /* 37 VMX_EXIT_MTF */ hmR0VmxExitMtf,
474 /* 38 UNDEFINED */ hmR0VmxExitErrUndefined,
475 /* 39 VMX_EXIT_MONITOR */ hmR0VmxExitMonitor,
476 /* 40 UNDEFINED */ hmR0VmxExitPause,
477 /* 41 VMX_EXIT_PAUSE */ hmR0VmxExitErrMachineCheck,
478 /* 42 VMX_EXIT_ERR_MACHINE_CHECK */ hmR0VmxExitErrUndefined,
479 /* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ hmR0VmxExitTprBelowThreshold,
480 /* 44 VMX_EXIT_APIC_ACCESS */ hmR0VmxExitApicAccess,
481 /* 45 UNDEFINED */ hmR0VmxExitErrUndefined,
482 /* 46 VMX_EXIT_XDTR_ACCESS */ hmR0VmxExitXdtrAccess,
483 /* 47 VMX_EXIT_TR_ACCESS */ hmR0VmxExitXdtrAccess,
484 /* 48 VMX_EXIT_EPT_VIOLATION */ hmR0VmxExitEptViolation,
485 /* 49 VMX_EXIT_EPT_MISCONFIG */ hmR0VmxExitEptMisconfig,
486 /* 50 VMX_EXIT_INVEPT */ hmR0VmxExitSetPendingXcptUD,
487 /* 51 VMX_EXIT_RDTSCP */ hmR0VmxExitRdtscp,
488 /* 52 VMX_EXIT_PREEMPT_TIMER */ hmR0VmxExitPreemptTimer,
489 /* 53 VMX_EXIT_INVVPID */ hmR0VmxExitSetPendingXcptUD,
490 /* 54 VMX_EXIT_WBINVD */ hmR0VmxExitWbinvd,
491 /* 55 VMX_EXIT_XSETBV */ hmR0VmxExitXsetbv,
492 /* 56 UNDEFINED */ hmR0VmxExitErrUndefined,
493 /* 57 VMX_EXIT_RDRAND */ hmR0VmxExitRdrand,
494 /* 58 VMX_EXIT_INVPCID */ hmR0VmxExitInvpcid,
495 /* 59 VMX_EXIT_VMFUNC */ hmR0VmxExitSetPendingXcptUD,
496 /* 60 VMX_EXIT_RESERVED_60 */ hmR0VmxExitErrUndefined,
497 /* 61 VMX_EXIT_RDSEED */ hmR0VmxExitErrUndefined, /* only spurious exits, so undefined */
498 /* 62 VMX_EXIT_RESERVED_62 */ hmR0VmxExitErrUndefined,
499 /* 63 VMX_EXIT_XSAVES */ hmR0VmxExitSetPendingXcptUD,
500 /* 64 VMX_EXIT_XRSTORS */ hmR0VmxExitSetPendingXcptUD,
501};
502#endif /* HMVMX_USE_FUNCTION_TABLE */
503
504#ifdef VBOX_STRICT
505static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] =
506{
507 /* 0 */ "(Not Used)",
508 /* 1 */ "VMCALL executed in VMX root operation.",
509 /* 2 */ "VMCLEAR with invalid physical address.",
510 /* 3 */ "VMCLEAR with VMXON pointer.",
511 /* 4 */ "VMLAUNCH with non-clear VMCS.",
512 /* 5 */ "VMRESUME with non-launched VMCS.",
513 /* 6 */ "VMRESUME after VMXOFF",
514 /* 7 */ "VM-entry with invalid control fields.",
515 /* 8 */ "VM-entry with invalid host state fields.",
516 /* 9 */ "VMPTRLD with invalid physical address.",
517 /* 10 */ "VMPTRLD with VMXON pointer.",
518 /* 11 */ "VMPTRLD with incorrect revision identifier.",
519 /* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.",
520 /* 13 */ "VMWRITE to read-only VMCS component.",
521 /* 14 */ "(Not Used)",
522 /* 15 */ "VMXON executed in VMX root operation.",
523 /* 16 */ "VM-entry with invalid executive-VMCS pointer.",
524 /* 17 */ "VM-entry with non-launched executing VMCS.",
525 /* 18 */ "VM-entry with executive-VMCS pointer not VMXON pointer.",
526 /* 19 */ "VMCALL with non-clear VMCS.",
527 /* 20 */ "VMCALL with invalid VM-exit control fields.",
528 /* 21 */ "(Not Used)",
529 /* 22 */ "VMCALL with incorrect MSEG revision identifier.",
530 /* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.",
531 /* 24 */ "VMCALL with invalid SMM-monitor features.",
532 /* 25 */ "VM-entry with invalid VM-execution control fields in executive VMCS.",
533 /* 26 */ "VM-entry with events blocked by MOV SS.",
534 /* 27 */ "(Not Used)",
535 /* 28 */ "Invalid operand to INVEPT/INVVPID."
536};
537#endif /* VBOX_STRICT */
538
539
540
541/**
542 * Updates the VM's last error record. If there was a VMX instruction error,
543 * reads the error data from the VMCS and updates VCPU's last error record as
544 * well.
545 *
546 * @param pVM Pointer to the VM.
547 * @param pVCpu Pointer to the VMCPU (can be NULL if @a rc is not
548 * VERR_VMX_UNABLE_TO_START_VM or
549 * VERR_VMX_INVALID_VMCS_FIELD).
550 * @param rc The error code.
551 */
552static void hmR0VmxUpdateErrorRecord(PVM pVM, PVMCPU pVCpu, int rc)
553{
554 AssertPtr(pVM);
555 if ( rc == VERR_VMX_INVALID_VMCS_FIELD
556 || rc == VERR_VMX_UNABLE_TO_START_VM)
557 {
558 AssertPtrReturnVoid(pVCpu);
559 VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
560 }
561 pVM->hm.s.lLastError = rc;
562}
563
564
565/**
566 * Reads the VM-entry interruption-information field from the VMCS into the VMX
567 * transient structure.
568 *
569 * @returns VBox status code.
570 * @param pVmxTransient Pointer to the VMX transient structure.
571 *
572 * @remarks No-long-jump zone!!!
573 */
574DECLINLINE(int) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
575{
576 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo);
577 AssertRCReturn(rc, rc);
578 return VINF_SUCCESS;
579}
580
581
582/**
583 * Reads the VM-entry exception error code field from the VMCS into
584 * the VMX transient structure.
585 *
586 * @returns VBox status code.
587 * @param pVmxTransient Pointer to the VMX transient structure.
588 *
589 * @remarks No-long-jump zone!!!
590 */
591DECLINLINE(int) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
592{
593 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode);
594 AssertRCReturn(rc, rc);
595 return VINF_SUCCESS;
596}
597
598
599/**
600 * Reads the VM-entry exception error code field from the VMCS into
601 * the VMX transient structure.
602 *
603 * @returns VBox status code.
604 * @param pVmxTransient Pointer to the VMX transient structure.
605 *
606 * @remarks No-long-jump zone!!!
607 */
608DECLINLINE(int) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
609{
610 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr);
611 AssertRCReturn(rc, rc);
612 return VINF_SUCCESS;
613}
614
615
616/**
617 * Reads the VM-exit interruption-information field from the VMCS into the VMX
618 * transient structure.
619 *
620 * @returns VBox status code.
621 * @param pVmxTransient Pointer to the VMX transient structure.
622 */
623DECLINLINE(int) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
624{
625 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO))
626 {
627 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
628 AssertRCReturn(rc, rc);
629 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO;
630 }
631 return VINF_SUCCESS;
632}
633
634
635/**
636 * Reads the VM-exit interruption error code from the VMCS into the VMX
637 * transient structure.
638 *
639 * @returns VBox status code.
640 * @param pVmxTransient Pointer to the VMX transient structure.
641 */
642DECLINLINE(int) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
643{
644 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE))
645 {
646 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
647 AssertRCReturn(rc, rc);
648 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE;
649 }
650 return VINF_SUCCESS;
651}
652
653
654/**
655 * Reads the VM-exit instruction length field from the VMCS into the VMX
656 * transient structure.
657 *
658 * @returns VBox status code.
659 * @param pVCpu Pointer to the VMCPU.
660 * @param pVmxTransient Pointer to the VMX transient structure.
661 */
662DECLINLINE(int) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
663{
664 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN))
665 {
666 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbInstr);
667 AssertRCReturn(rc, rc);
668 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN;
669 }
670 return VINF_SUCCESS;
671}
672
673
674/**
675 * Reads the VM-exit instruction-information field from the VMCS into
676 * the VMX transient structure.
677 *
678 * @returns VBox status code.
679 * @param pVmxTransient Pointer to the VMX transient structure.
680 */
681DECLINLINE(int) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient)
682{
683 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO))
684 {
685 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
686 AssertRCReturn(rc, rc);
687 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO;
688 }
689 return VINF_SUCCESS;
690}
691
692
693/**
694 * Reads the exit code qualification from the VMCS into the VMX transient
695 * structure.
696 *
697 * @returns VBox status code.
698 * @param pVCpu Pointer to the VMCPU (required for the VMCS cache
699 * case).
700 * @param pVmxTransient Pointer to the VMX transient structure.
701 */
702DECLINLINE(int) hmR0VmxReadExitQualificationVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
703{
704 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION))
705 {
706 int rc = VMXReadVmcsGstN(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQualification); NOREF(pVCpu);
707 AssertRCReturn(rc, rc);
708 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION;
709 }
710 return VINF_SUCCESS;
711}
712
713
714/**
715 * Reads the IDT-vectoring information field from the VMCS into the VMX
716 * transient structure.
717 *
718 * @returns VBox status code.
719 * @param pVmxTransient Pointer to the VMX transient structure.
720 *
721 * @remarks No-long-jump zone!!!
722 */
723DECLINLINE(int) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient)
724{
725 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO))
726 {
727 int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_INFO, &pVmxTransient->uIdtVectoringInfo);
728 AssertRCReturn(rc, rc);
729 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO;
730 }
731 return VINF_SUCCESS;
732}
733
734
735/**
736 * Reads the IDT-vectoring error code from the VMCS into the VMX
737 * transient structure.
738 *
739 * @returns VBox status code.
740 * @param pVmxTransient Pointer to the VMX transient structure.
741 */
742DECLINLINE(int) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
743{
744 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE))
745 {
746 int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
747 AssertRCReturn(rc, rc);
748 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE;
749 }
750 return VINF_SUCCESS;
751}
752
753
754/**
755 * Enters VMX root mode operation on the current CPU.
756 *
757 * @returns VBox status code.
758 * @param pVM Pointer to the VM (optional, can be NULL, after
759 * a resume).
760 * @param HCPhysCpuPage Physical address of the VMXON region.
761 * @param pvCpuPage Pointer to the VMXON region.
762 */
763static int hmR0VmxEnterRootMode(PVM pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage)
764{
765 Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
766 Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
767 Assert(pvCpuPage);
768 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
769
770 if (pVM)
771 {
772 /* Write the VMCS revision dword to the VMXON region. */
773 *(uint32_t *)pvCpuPage = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo);
774 }
775
776 /* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */
777 RTCCUINTREG fEFlags = ASMIntDisableFlags();
778
779 /* Enable the VMX bit in CR4 if necessary. */
780 RTCCUINTREG uOldCr4 = SUPR0ChangeCR4(X86_CR4_VMXE, ~0);
781
782 /* Enter VMX root mode. */
783 int rc = VMXEnable(HCPhysCpuPage);
784 if (RT_FAILURE(rc))
785 {
786 if (!(uOldCr4 & X86_CR4_VMXE))
787 SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
788
789 if (pVM)
790 pVM->hm.s.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
791 }
792
793 /* Restore interrupts. */
794 ASMSetFlags(fEFlags);
795 return rc;
796}
797
798
799/**
800 * Exits VMX root mode operation on the current CPU.
801 *
802 * @returns VBox status code.
803 */
804static int hmR0VmxLeaveRootMode(void)
805{
806 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
807
808 /* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */
809 RTCCUINTREG fEFlags = ASMIntDisableFlags();
810
811 /* If we're for some reason not in VMX root mode, then don't leave it. */
812 RTCCUINTREG uHostCR4 = ASMGetCR4();
813
814 int rc;
815 if (uHostCR4 & X86_CR4_VMXE)
816 {
817 /* Exit VMX root mode and clear the VMX bit in CR4. */
818 VMXDisable();
819 SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
820 rc = VINF_SUCCESS;
821 }
822 else
823 rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE;
824
825 /* Restore interrupts. */
826 ASMSetFlags(fEFlags);
827 return rc;
828}
829
830
831/**
832 * Allocates and maps one physically contiguous page. The allocated page is
833 * zero'd out. (Used by various VT-x structures).
834 *
835 * @returns IPRT status code.
836 * @param pMemObj Pointer to the ring-0 memory object.
837 * @param ppVirt Where to store the virtual address of the
838 * allocation.
839 * @param pPhys Where to store the physical address of the
840 * allocation.
841 */
842DECLINLINE(int) hmR0VmxPageAllocZ(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
843{
844 AssertPtrReturn(pMemObj, VERR_INVALID_PARAMETER);
845 AssertPtrReturn(ppVirt, VERR_INVALID_PARAMETER);
846 AssertPtrReturn(pHCPhys, VERR_INVALID_PARAMETER);
847
848 int rc = RTR0MemObjAllocCont(pMemObj, PAGE_SIZE, false /* fExecutable */);
849 if (RT_FAILURE(rc))
850 return rc;
851 *ppVirt = RTR0MemObjAddress(*pMemObj);
852 *pHCPhys = RTR0MemObjGetPagePhysAddr(*pMemObj, 0 /* iPage */);
853 ASMMemZero32(*ppVirt, PAGE_SIZE);
854 return VINF_SUCCESS;
855}
856
857
858/**
859 * Frees and unmaps an allocated physical page.
860 *
861 * @param pMemObj Pointer to the ring-0 memory object.
862 * @param ppVirt Where to re-initialize the virtual address of
863 * allocation as 0.
864 * @param pHCPhys Where to re-initialize the physical address of the
865 * allocation as 0.
866 */
867DECLINLINE(void) hmR0VmxPageFree(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
868{
869 AssertPtr(pMemObj);
870 AssertPtr(ppVirt);
871 AssertPtr(pHCPhys);
872 if (*pMemObj != NIL_RTR0MEMOBJ)
873 {
874 int rc = RTR0MemObjFree(*pMemObj, true /* fFreeMappings */);
875 AssertRC(rc);
876 *pMemObj = NIL_RTR0MEMOBJ;
877 *ppVirt = 0;
878 *pHCPhys = 0;
879 }
880}
881
882
883/**
884 * Worker function to free VT-x related structures.
885 *
886 * @returns IPRT status code.
887 * @param pVM Pointer to the VM.
888 */
889static void hmR0VmxStructsFree(PVM pVM)
890{
891 for (VMCPUID i = 0; i < pVM->cCpus; i++)
892 {
893 PVMCPU pVCpu = &pVM->aCpus[i];
894 AssertPtr(pVCpu);
895
896 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjHostMsr, &pVCpu->hm.s.vmx.pvHostMsr, &pVCpu->hm.s.vmx.HCPhysHostMsr);
897 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjGuestMsr, &pVCpu->hm.s.vmx.pvGuestMsr, &pVCpu->hm.s.vmx.HCPhysGuestMsr);
898
899 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
900 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjMsrBitmap, &pVCpu->hm.s.vmx.pvMsrBitmap, &pVCpu->hm.s.vmx.HCPhysMsrBitmap);
901
902 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjVirtApic, (PRTR0PTR)&pVCpu->hm.s.vmx.pbVirtApic, &pVCpu->hm.s.vmx.HCPhysVirtApic);
903 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjVmcs, &pVCpu->hm.s.vmx.pvVmcs, &pVCpu->hm.s.vmx.HCPhysVmcs);
904 }
905
906 hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess, &pVM->hm.s.vmx.HCPhysApicAccess);
907#ifdef VBOX_WITH_CRASHDUMP_MAGIC
908 hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
909#endif
910}
911
912
913/**
914 * Worker function to allocate VT-x related VM structures.
915 *
916 * @returns IPRT status code.
917 * @param pVM Pointer to the VM.
918 */
919static int hmR0VmxStructsAlloc(PVM pVM)
920{
921 /*
922 * Initialize members up-front so we can cleanup properly on allocation failure.
923 */
924#define VMXLOCAL_INIT_VM_MEMOBJ(a_Name, a_VirtPrefix) \
925 pVM->hm.s.vmx.hMemObj##a_Name = NIL_RTR0MEMOBJ; \
926 pVM->hm.s.vmx.a_VirtPrefix##a_Name = 0; \
927 pVM->hm.s.vmx.HCPhys##a_Name = 0;
928
929#define VMXLOCAL_INIT_VMCPU_MEMOBJ(a_Name, a_VirtPrefix) \
930 pVCpu->hm.s.vmx.hMemObj##a_Name = NIL_RTR0MEMOBJ; \
931 pVCpu->hm.s.vmx.a_VirtPrefix##a_Name = 0; \
932 pVCpu->hm.s.vmx.HCPhys##a_Name = 0;
933
934#ifdef VBOX_WITH_CRASHDUMP_MAGIC
935 VMXLOCAL_INIT_VM_MEMOBJ(Scratch, pv);
936#endif
937 VMXLOCAL_INIT_VM_MEMOBJ(ApicAccess, pb);
938
939 AssertCompile(sizeof(VMCPUID) == sizeof(pVM->cCpus));
940 for (VMCPUID i = 0; i < pVM->cCpus; i++)
941 {
942 PVMCPU pVCpu = &pVM->aCpus[i];
943 VMXLOCAL_INIT_VMCPU_MEMOBJ(Vmcs, pv);
944 VMXLOCAL_INIT_VMCPU_MEMOBJ(VirtApic, pb);
945 VMXLOCAL_INIT_VMCPU_MEMOBJ(MsrBitmap, pv);
946 VMXLOCAL_INIT_VMCPU_MEMOBJ(GuestMsr, pv);
947 VMXLOCAL_INIT_VMCPU_MEMOBJ(HostMsr, pv);
948 }
949#undef VMXLOCAL_INIT_VMCPU_MEMOBJ
950#undef VMXLOCAL_INIT_VM_MEMOBJ
951
952 /* The VMCS size cannot be more than 4096 bytes. See Intel spec. Appendix A.1 "Basic VMX Information". */
953 AssertReturnStmt(MSR_IA32_VMX_BASIC_INFO_VMCS_SIZE(pVM->hm.s.vmx.Msrs.u64BasicInfo) <= PAGE_SIZE,
954 (&pVM->aCpus[0])->hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE,
955 VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO);
956
957 /*
958 * Allocate all the VT-x structures.
959 */
960 int rc = VINF_SUCCESS;
961#ifdef VBOX_WITH_CRASHDUMP_MAGIC
962 rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
963 if (RT_FAILURE(rc))
964 goto cleanup;
965 strcpy((char *)pVM->hm.s.vmx.pbScratch, "SCRATCH Magic");
966 *(uint64_t *)(pVM->hm.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef);
967#endif
968
969 /* Allocate the APIC-access page for trapping APIC accesses from the guest. */
970 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)
971 {
972 rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess,
973 &pVM->hm.s.vmx.HCPhysApicAccess);
974 if (RT_FAILURE(rc))
975 goto cleanup;
976 }
977
978 /*
979 * Initialize per-VCPU VT-x structures.
980 */
981 for (VMCPUID i = 0; i < pVM->cCpus; i++)
982 {
983 PVMCPU pVCpu = &pVM->aCpus[i];
984 AssertPtr(pVCpu);
985
986 /* Allocate the VM control structure (VMCS). */
987 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjVmcs, &pVCpu->hm.s.vmx.pvVmcs, &pVCpu->hm.s.vmx.HCPhysVmcs);
988 if (RT_FAILURE(rc))
989 goto cleanup;
990
991 /* Allocate the Virtual-APIC page for transparent TPR accesses. */
992 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
993 {
994 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjVirtApic, (PRTR0PTR)&pVCpu->hm.s.vmx.pbVirtApic,
995 &pVCpu->hm.s.vmx.HCPhysVirtApic);
996 if (RT_FAILURE(rc))
997 goto cleanup;
998 }
999
1000 /*
1001 * Allocate the MSR-bitmap if supported by the CPU. The MSR-bitmap is for
1002 * transparent accesses of specific MSRs.
1003 *
1004 * If the condition for enabling MSR bitmaps changes here, don't forget to
1005 * update HMAreMsrBitmapsAvailable().
1006 */
1007 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1008 {
1009 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjMsrBitmap, &pVCpu->hm.s.vmx.pvMsrBitmap,
1010 &pVCpu->hm.s.vmx.HCPhysMsrBitmap);
1011 if (RT_FAILURE(rc))
1012 goto cleanup;
1013 ASMMemFill32(pVCpu->hm.s.vmx.pvMsrBitmap, PAGE_SIZE, UINT32_C(0xffffffff));
1014 }
1015
1016 /* Allocate the VM-entry MSR-load and VM-exit MSR-store page for the guest MSRs. */
1017 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjGuestMsr, &pVCpu->hm.s.vmx.pvGuestMsr, &pVCpu->hm.s.vmx.HCPhysGuestMsr);
1018 if (RT_FAILURE(rc))
1019 goto cleanup;
1020
1021 /* Allocate the VM-exit MSR-load page for the host MSRs. */
1022 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjHostMsr, &pVCpu->hm.s.vmx.pvHostMsr, &pVCpu->hm.s.vmx.HCPhysHostMsr);
1023 if (RT_FAILURE(rc))
1024 goto cleanup;
1025 }
1026
1027 return VINF_SUCCESS;
1028
1029cleanup:
1030 hmR0VmxStructsFree(pVM);
1031 return rc;
1032}
1033
1034
1035/**
1036 * Does global VT-x initialization (called during module initialization).
1037 *
1038 * @returns VBox status code.
1039 */
1040VMMR0DECL(int) VMXR0GlobalInit(void)
1041{
1042#ifdef HMVMX_USE_FUNCTION_TABLE
1043 AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_apfnVMExitHandlers));
1044# ifdef VBOX_STRICT
1045 for (unsigned i = 0; i < RT_ELEMENTS(g_apfnVMExitHandlers); i++)
1046 Assert(g_apfnVMExitHandlers[i]);
1047# endif
1048#endif
1049 return VINF_SUCCESS;
1050}
1051
1052
1053/**
1054 * Does global VT-x termination (called during module termination).
1055 */
1056VMMR0DECL(void) VMXR0GlobalTerm()
1057{
1058 /* Nothing to do currently. */
1059}
1060
1061
1062/**
1063 * Sets up and activates VT-x on the current CPU.
1064 *
1065 * @returns VBox status code.
1066 * @param pCpu Pointer to the global CPU info struct.
1067 * @param pVM Pointer to the VM (can be NULL after a host resume
1068 * operation).
1069 * @param pvCpuPage Pointer to the VMXON region (can be NULL if @a
1070 * fEnabledByHost is true).
1071 * @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if
1072 * @a fEnabledByHost is true).
1073 * @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to
1074 * enable VT-x on the host.
1075 * @param pvMsrs Opaque pointer to VMXMSRS struct.
1076 */
1077VMMR0DECL(int) VMXR0EnableCpu(PHMGLOBALCPUINFO pCpu, PVM pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
1078 void *pvMsrs)
1079{
1080 Assert(pCpu);
1081 Assert(pvMsrs);
1082 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1083
1084 /* Enable VT-x if it's not already enabled by the host. */
1085 if (!fEnabledByHost)
1086 {
1087 int rc = hmR0VmxEnterRootMode(pVM, HCPhysCpuPage, pvCpuPage);
1088 if (RT_FAILURE(rc))
1089 return rc;
1090 }
1091
1092 /*
1093 * Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been using EPTPs) so
1094 * we don't retain any stale guest-physical mappings which won't get invalidated when flushing by VPID.
1095 */
1096 PVMXMSRS pMsrs = (PVMXMSRS)pvMsrs;
1097 if (pMsrs->u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
1098 {
1099 hmR0VmxFlushEpt(NULL /* pVCpu */, VMXFLUSHEPT_ALL_CONTEXTS);
1100 pCpu->fFlushAsidBeforeUse = false;
1101 }
1102 else
1103 pCpu->fFlushAsidBeforeUse = true;
1104
1105 /* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */
1106 ++pCpu->cTlbFlushes;
1107
1108 return VINF_SUCCESS;
1109}
1110
1111
1112/**
1113 * Deactivates VT-x on the current CPU.
1114 *
1115 * @returns VBox status code.
1116 * @param pCpu Pointer to the global CPU info struct.
1117 * @param pvCpuPage Pointer to the VMXON region.
1118 * @param HCPhysCpuPage Physical address of the VMXON region.
1119 *
1120 * @remarks This function should never be called when SUPR0EnableVTx() or
1121 * similar was used to enable VT-x on the host.
1122 */
1123VMMR0DECL(int) VMXR0DisableCpu(PHMGLOBALCPUINFO pCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
1124{
1125 NOREF(pCpu);
1126 NOREF(pvCpuPage);
1127 NOREF(HCPhysCpuPage);
1128
1129 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1130 return hmR0VmxLeaveRootMode();
1131}
1132
1133
1134/**
1135 * Sets the permission bits for the specified MSR in the MSR bitmap.
1136 *
1137 * @param pVCpu Pointer to the VMCPU.
1138 * @param uMSR The MSR value.
1139 * @param enmRead Whether reading this MSR causes a VM-exit.
1140 * @param enmWrite Whether writing this MSR causes a VM-exit.
1141 */
1142static void hmR0VmxSetMsrPermission(PVMCPU pVCpu, uint32_t uMsr, VMXMSREXITREAD enmRead, VMXMSREXITWRITE enmWrite)
1143{
1144 int32_t iBit;
1145 uint8_t *pbMsrBitmap = (uint8_t *)pVCpu->hm.s.vmx.pvMsrBitmap;
1146
1147 /*
1148 * Layout:
1149 * 0x000 - 0x3ff - Low MSR read bits
1150 * 0x400 - 0x7ff - High MSR read bits
1151 * 0x800 - 0xbff - Low MSR write bits
1152 * 0xc00 - 0xfff - High MSR write bits
1153 */
1154 if (uMsr <= 0x00001FFF)
1155 iBit = uMsr;
1156 else if ( uMsr >= 0xC0000000
1157 && uMsr <= 0xC0001FFF)
1158 {
1159 iBit = (uMsr - 0xC0000000);
1160 pbMsrBitmap += 0x400;
1161 }
1162 else
1163 AssertMsgFailedReturnVoid(("hmR0VmxSetMsrPermission: Invalid MSR %#RX32\n", uMsr));
1164
1165 Assert(iBit <= 0x1fff);
1166 if (enmRead == VMXMSREXIT_INTERCEPT_READ)
1167 ASMBitSet(pbMsrBitmap, iBit);
1168 else
1169 ASMBitClear(pbMsrBitmap, iBit);
1170
1171 if (enmWrite == VMXMSREXIT_INTERCEPT_WRITE)
1172 ASMBitSet(pbMsrBitmap + 0x800, iBit);
1173 else
1174 ASMBitClear(pbMsrBitmap + 0x800, iBit);
1175}
1176
1177
1178#ifdef VBOX_STRICT
1179/**
1180 * Gets the permission bits for the specified MSR in the MSR bitmap.
1181 *
1182 * @returns VBox status code.
1183 * @retval VINF_SUCCESS if the specified MSR is found.
1184 * @retval VERR_NOT_FOUND if the specified MSR is not found.
1185 * @retval VERR_NOT_SUPPORTED if VT-x doesn't allow the MSR.
1186 *
1187 * @param pVCpu Pointer to the VMCPU.
1188 * @param uMsr The MSR.
1189 * @param penmRead Where to store the read permissions.
1190 * @param penmWrite Where to store the write permissions.
1191 */
1192static int hmR0VmxGetMsrPermission(PVMCPU pVCpu, uint32_t uMsr, PVMXMSREXITREAD penmRead, PVMXMSREXITWRITE penmWrite)
1193{
1194 AssertPtrReturn(penmRead, VERR_INVALID_PARAMETER);
1195 AssertPtrReturn(penmWrite, VERR_INVALID_PARAMETER);
1196 int32_t iBit;
1197 uint8_t *pbMsrBitmap = (uint8_t *)pVCpu->hm.s.vmx.pvMsrBitmap;
1198
1199 /* See hmR0VmxSetMsrPermission() for the layout. */
1200 if (uMsr <= 0x00001FFF)
1201 iBit = uMsr;
1202 else if ( uMsr >= 0xC0000000
1203 && uMsr <= 0xC0001FFF)
1204 {
1205 iBit = (uMsr - 0xC0000000);
1206 pbMsrBitmap += 0x400;
1207 }
1208 else
1209 AssertMsgFailedReturn(("hmR0VmxGetMsrPermission: Invalid MSR %#RX32\n", uMsr), VERR_NOT_SUPPORTED);
1210
1211 Assert(iBit <= 0x1fff);
1212 if (ASMBitTest(pbMsrBitmap, iBit))
1213 *penmRead = VMXMSREXIT_INTERCEPT_READ;
1214 else
1215 *penmRead = VMXMSREXIT_PASSTHRU_READ;
1216
1217 if (ASMBitTest(pbMsrBitmap + 0x800, iBit))
1218 *penmWrite = VMXMSREXIT_INTERCEPT_WRITE;
1219 else
1220 *penmWrite = VMXMSREXIT_PASSTHRU_WRITE;
1221 return VINF_SUCCESS;
1222}
1223#endif /* VBOX_STRICT */
1224
1225
1226/**
1227 * Updates the VMCS with the number of effective MSRs in the auto-load/store MSR
1228 * area.
1229 *
1230 * @returns VBox status code.
1231 * @param pVCpu Pointer to the VMCPU.
1232 * @param cMsrs The number of MSRs.
1233 */
1234DECLINLINE(int) hmR0VmxSetAutoLoadStoreMsrCount(PVMCPU pVCpu, uint32_t cMsrs)
1235{
1236 /* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */
1237 uint32_t const cMaxSupportedMsrs = MSR_IA32_VMX_MISC_MAX_MSR(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc);
1238 if (RT_UNLIKELY(cMsrs > cMaxSupportedMsrs))
1239 {
1240 LogRel(("CPU auto-load/store MSR count in VMCS exceeded cMsrs=%u Supported=%u.\n", cMsrs, cMaxSupportedMsrs));
1241 pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE;
1242 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
1243 }
1244
1245 /* Update number of guest MSRs to load/store across the world-switch. */
1246 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs); AssertRCReturn(rc, rc);
1247 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs); AssertRCReturn(rc, rc);
1248
1249 /* Update number of host MSRs to load after the world-switch. Identical to guest-MSR count as it's always paired. */
1250 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs); AssertRCReturn(rc, rc);
1251
1252 /* Update the VCPU's copy of the MSR count. */
1253 pVCpu->hm.s.vmx.cMsrs = cMsrs;
1254
1255 return VINF_SUCCESS;
1256}
1257
1258
1259/**
1260 * Adds a new (or updates the value of an existing) guest/host MSR
1261 * pair to be swapped during the world-switch as part of the
1262 * auto-load/store MSR area in the VMCS.
1263 *
1264 * @returns VBox status code.
1265 * @param pVCpu Pointer to the VMCPU.
1266 * @param uMsr The MSR.
1267 * @param uGuestMsr Value of the guest MSR.
1268 * @param fUpdateHostMsr Whether to update the value of the host MSR if
1269 * necessary.
1270 * @param pfAddedAndUpdated Where to store whether the MSR was added -and-
1271 * its value was updated. Optional, can be NULL.
1272 */
1273static int hmR0VmxAddAutoLoadStoreMsr(PVMCPU pVCpu, uint32_t uMsr, uint64_t uGuestMsrValue, bool fUpdateHostMsr,
1274 bool *pfAddedAndUpdated)
1275{
1276 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1277 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1278 uint32_t i;
1279 for (i = 0; i < cMsrs; i++)
1280 {
1281 if (pGuestMsr->u32Msr == uMsr)
1282 break;
1283 pGuestMsr++;
1284 }
1285
1286 bool fAdded = false;
1287 if (i == cMsrs)
1288 {
1289 ++cMsrs;
1290 int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, cMsrs);
1291 AssertMsgRCReturn(rc, ("hmR0VmxAddAutoLoadStoreMsr: Insufficient space to add MSR %u\n", uMsr), rc);
1292
1293 /* Now that we're swapping MSRs during the world-switch, allow the guest to read/write them without causing VM-exits. */
1294 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1295 hmR0VmxSetMsrPermission(pVCpu, uMsr, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
1296
1297 fAdded = true;
1298 }
1299
1300 /* Update the MSR values in the auto-load/store MSR area. */
1301 pGuestMsr->u32Msr = uMsr;
1302 pGuestMsr->u64Value = uGuestMsrValue;
1303
1304 /* Create/update the MSR slot in the host MSR area. */
1305 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1306 pHostMsr += i;
1307 pHostMsr->u32Msr = uMsr;
1308
1309 /*
1310 * Update the host MSR only when requested by the caller AND when we're
1311 * adding it to the auto-load/store area. Otherwise, it would have been
1312 * updated by hmR0VmxSaveHostMsrs(). We do this for performance reasons.
1313 */
1314 bool fUpdatedMsrValue = false;
1315 if ( fAdded
1316 && fUpdateHostMsr)
1317 {
1318 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1319 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1320 pHostMsr->u64Value = ASMRdMsr(pHostMsr->u32Msr);
1321 fUpdatedMsrValue = true;
1322 }
1323
1324 if (pfAddedAndUpdated)
1325 *pfAddedAndUpdated = fUpdatedMsrValue;
1326 return VINF_SUCCESS;
1327}
1328
1329
1330/**
1331 * Removes a guest/host MSR pair to be swapped during the world-switch from the
1332 * auto-load/store MSR area in the VMCS.
1333 *
1334 * @returns VBox status code.
1335 * @param pVCpu Pointer to the VMCPU.
1336 * @param uMsr The MSR.
1337 */
1338static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPU pVCpu, uint32_t uMsr)
1339{
1340 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1341 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1342 for (uint32_t i = 0; i < cMsrs; i++)
1343 {
1344 /* Find the MSR. */
1345 if (pGuestMsr->u32Msr == uMsr)
1346 {
1347 /* If it's the last MSR, simply reduce the count. */
1348 if (i == cMsrs - 1)
1349 {
1350 --cMsrs;
1351 break;
1352 }
1353
1354 /* Remove it by swapping the last MSR in place of it, and reducing the count. */
1355 PVMXAUTOMSR pLastGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1356 pLastGuestMsr += cMsrs - 1;
1357 pGuestMsr->u32Msr = pLastGuestMsr->u32Msr;
1358 pGuestMsr->u64Value = pLastGuestMsr->u64Value;
1359
1360 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1361 PVMXAUTOMSR pLastHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1362 pLastHostMsr += cMsrs - 1;
1363 pHostMsr->u32Msr = pLastHostMsr->u32Msr;
1364 pHostMsr->u64Value = pLastHostMsr->u64Value;
1365 --cMsrs;
1366 break;
1367 }
1368 pGuestMsr++;
1369 }
1370
1371 /* Update the VMCS if the count changed (meaning the MSR was found). */
1372 if (cMsrs != pVCpu->hm.s.vmx.cMsrs)
1373 {
1374 int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, cMsrs);
1375 AssertRCReturn(rc, rc);
1376
1377 /* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */
1378 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1379 hmR0VmxSetMsrPermission(pVCpu, uMsr, VMXMSREXIT_INTERCEPT_READ, VMXMSREXIT_INTERCEPT_WRITE);
1380
1381 Log4(("Removed MSR %#RX32 new cMsrs=%u\n", uMsr, pVCpu->hm.s.vmx.cMsrs));
1382 return VINF_SUCCESS;
1383 }
1384
1385 return VERR_NOT_FOUND;
1386}
1387
1388
1389/**
1390 * Checks if the specified guest MSR is part of the auto-load/store area in
1391 * the VMCS.
1392 *
1393 * @returns true if found, false otherwise.
1394 * @param pVCpu Pointer to the VMCPU.
1395 * @param uMsr The MSR to find.
1396 */
1397static bool hmR0VmxIsAutoLoadStoreGuestMsr(PVMCPU pVCpu, uint32_t uMsr)
1398{
1399 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1400 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1401
1402 for (uint32_t i = 0; i < cMsrs; i++, pGuestMsr++)
1403 {
1404 if (pGuestMsr->u32Msr == uMsr)
1405 return true;
1406 }
1407 return false;
1408}
1409
1410
1411/**
1412 * Updates the value of all host MSRs in the auto-load/store area in the VMCS.
1413 *
1414 * @param pVCpu Pointer to the VMCPU.
1415 *
1416 * @remarks No-long-jump zone!!!
1417 */
1418static void hmR0VmxUpdateAutoLoadStoreHostMsrs(PVMCPU pVCpu)
1419{
1420 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1421 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1422 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1423 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1424
1425 for (uint32_t i = 0; i < cMsrs; i++, pHostMsr++, pGuestMsr++)
1426 {
1427 AssertReturnVoid(pHostMsr->u32Msr == pGuestMsr->u32Msr);
1428
1429 /*
1430 * Performance hack for the host EFER MSR. We use the cached value rather than re-read it.
1431 * Strict builds will catch mismatches in hmR0VmxCheckAutoLoadStoreMsrs(). See @bugref{7368}.
1432 */
1433 if (pHostMsr->u32Msr == MSR_K6_EFER)
1434 pHostMsr->u64Value = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostEfer;
1435 else
1436 pHostMsr->u64Value = ASMRdMsr(pHostMsr->u32Msr);
1437 }
1438
1439 pVCpu->hm.s.vmx.fUpdatedHostMsrs = true;
1440}
1441
1442
1443#if HC_ARCH_BITS == 64
1444/**
1445 * Saves a set of host MSRs to allow read/write passthru access to the guest and
1446 * perform lazy restoration of the host MSRs while leaving VT-x.
1447 *
1448 * @param pVCpu Pointer to the VMCPU.
1449 *
1450 * @remarks No-long-jump zone!!!
1451 */
1452static void hmR0VmxLazySaveHostMsrs(PVMCPU pVCpu)
1453{
1454 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1455
1456 /*
1457 * Note: If you're adding MSRs here, make sure to update the MSR-bitmap permissions in hmR0VmxSetupProcCtls().
1458 */
1459 if (!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
1460 {
1461 pVCpu->hm.s.vmx.u64HostLStarMsr = ASMRdMsr(MSR_K8_LSTAR);
1462 pVCpu->hm.s.vmx.u64HostStarMsr = ASMRdMsr(MSR_K6_STAR);
1463 pVCpu->hm.s.vmx.u64HostSFMaskMsr = ASMRdMsr(MSR_K8_SF_MASK);
1464 pVCpu->hm.s.vmx.u64HostKernelGSBaseMsr = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
1465 pVCpu->hm.s.vmx.fLazyMsrs |= VMX_LAZY_MSRS_SAVED_HOST;
1466 }
1467}
1468
1469
1470/**
1471 * Checks whether the MSR belongs to the set of guest MSRs that we restore
1472 * lazily while leaving VT-x.
1473 *
1474 * @returns true if it does, false otherwise.
1475 * @param pVCpu Pointer to the VMCPU.
1476 * @param uMsr The MSR to check.
1477 */
1478static bool hmR0VmxIsLazyGuestMsr(PVMCPU pVCpu, uint32_t uMsr)
1479{
1480 NOREF(pVCpu);
1481 switch (uMsr)
1482 {
1483 case MSR_K8_LSTAR:
1484 case MSR_K6_STAR:
1485 case MSR_K8_SF_MASK:
1486 case MSR_K8_KERNEL_GS_BASE:
1487 return true;
1488 }
1489 return false;
1490}
1491
1492
1493/**
1494 * Saves a set of guest MSRs back into the guest-CPU context.
1495 *
1496 * @param pVCpu Pointer to the VMCPU.
1497 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
1498 * out-of-sync. Make sure to update the required fields
1499 * before using them.
1500 *
1501 * @remarks No-long-jump zone!!!
1502 */
1503static void hmR0VmxLazySaveGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
1504{
1505 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1506 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1507
1508 if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
1509 {
1510 Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1511 pMixedCtx->msrLSTAR = ASMRdMsr(MSR_K8_LSTAR);
1512 pMixedCtx->msrSTAR = ASMRdMsr(MSR_K6_STAR);
1513 pMixedCtx->msrSFMASK = ASMRdMsr(MSR_K8_SF_MASK);
1514 pMixedCtx->msrKERNELGSBASE = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
1515 }
1516}
1517
1518
1519/**
1520 * Loads a set of guests MSRs to allow read/passthru to the guest.
1521 *
1522 * The name of this function is slightly confusing. This function does NOT
1523 * postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a
1524 * common prefix for functions dealing with "lazy restoration" of the shared
1525 * MSRs.
1526 *
1527 * @param pVCpu Pointer to the VMCPU.
1528 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
1529 * out-of-sync. Make sure to update the required fields
1530 * before using them.
1531 *
1532 * @remarks No-long-jump zone!!!
1533 */
1534static void hmR0VmxLazyLoadGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
1535{
1536 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1537 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1538
1539#define VMXLOCAL_LAZY_LOAD_GUEST_MSR(uMsr, a_GuestMsr, a_HostMsr) \
1540 do { \
1541 if (pMixedCtx->msr##a_GuestMsr != pVCpu->hm.s.vmx.u64Host##a_HostMsr##Msr) \
1542 ASMWrMsr(uMsr, pMixedCtx->msr##a_GuestMsr); \
1543 else \
1544 Assert(ASMRdMsr(uMsr) == pVCpu->hm.s.vmx.u64Host##a_HostMsr##Msr); \
1545 } while (0)
1546
1547 Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1548 if (!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
1549 {
1550 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K8_LSTAR, LSTAR, LStar);
1551 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K6_STAR, STAR, Star);
1552 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K8_SF_MASK, SFMASK, SFMask);
1553 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K8_KERNEL_GS_BASE, KERNELGSBASE, KernelGSBase);
1554 pVCpu->hm.s.vmx.fLazyMsrs |= VMX_LAZY_MSRS_LOADED_GUEST;
1555 }
1556 else
1557 {
1558 ASMWrMsr(MSR_K8_LSTAR, pMixedCtx->msrLSTAR);
1559 ASMWrMsr(MSR_K6_STAR, pMixedCtx->msrSTAR);
1560 ASMWrMsr(MSR_K8_SF_MASK, pMixedCtx->msrSFMASK);
1561 ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pMixedCtx->msrKERNELGSBASE);
1562 }
1563
1564#undef VMXLOCAL_LAZY_LOAD_GUEST_MSR
1565}
1566
1567
1568/**
1569 * Performs lazy restoration of the set of host MSRs if they were previously
1570 * loaded with guest MSR values.
1571 *
1572 * @param pVCpu Pointer to the VMCPU.
1573 *
1574 * @remarks No-long-jump zone!!!
1575 * @remarks The guest MSRs should have been saved back into the guest-CPU
1576 * context by hmR0VmxSaveGuestLazyMsrs()!!!
1577 */
1578static void hmR0VmxLazyRestoreHostMsrs(PVMCPU pVCpu)
1579{
1580 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1581 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1582
1583 if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
1584 {
1585 Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1586 ASMWrMsr(MSR_K8_LSTAR, pVCpu->hm.s.vmx.u64HostLStarMsr);
1587 ASMWrMsr(MSR_K6_STAR, pVCpu->hm.s.vmx.u64HostStarMsr);
1588 ASMWrMsr(MSR_K8_SF_MASK, pVCpu->hm.s.vmx.u64HostSFMaskMsr);
1589 ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pVCpu->hm.s.vmx.u64HostKernelGSBaseMsr);
1590 }
1591 pVCpu->hm.s.vmx.fLazyMsrs &= ~(VMX_LAZY_MSRS_LOADED_GUEST | VMX_LAZY_MSRS_SAVED_HOST);
1592}
1593#endif /* HC_ARCH_BITS == 64 */
1594
1595
1596/**
1597 * Verifies that our cached values of the VMCS controls are all
1598 * consistent with what's actually present in the VMCS.
1599 *
1600 * @returns VBox status code.
1601 * @param pVCpu Pointer to the VMCPU.
1602 */
1603static int hmR0VmxCheckVmcsCtls(PVMCPU pVCpu)
1604{
1605 uint32_t u32Val;
1606 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
1607 AssertRCReturn(rc, rc);
1608 AssertMsgReturn(pVCpu->hm.s.vmx.u32EntryCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32EntryCtls, u32Val),
1609 VERR_VMX_ENTRY_CTLS_CACHE_INVALID);
1610
1611 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val);
1612 AssertRCReturn(rc, rc);
1613 AssertMsgReturn(pVCpu->hm.s.vmx.u32ExitCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ExitCtls, u32Val),
1614 VERR_VMX_EXIT_CTLS_CACHE_INVALID);
1615
1616 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val);
1617 AssertRCReturn(rc, rc);
1618 AssertMsgReturn(pVCpu->hm.s.vmx.u32PinCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32PinCtls, u32Val),
1619 VERR_VMX_PIN_EXEC_CTLS_CACHE_INVALID);
1620
1621 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val);
1622 AssertRCReturn(rc, rc);
1623 AssertMsgReturn(pVCpu->hm.s.vmx.u32ProcCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ProcCtls, u32Val),
1624 VERR_VMX_PROC_EXEC_CTLS_CACHE_INVALID);
1625
1626 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
1627 {
1628 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val);
1629 AssertRCReturn(rc, rc);
1630 AssertMsgReturn(pVCpu->hm.s.vmx.u32ProcCtls2 == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ProcCtls2, u32Val),
1631 VERR_VMX_PROC_EXEC2_CTLS_CACHE_INVALID);
1632 }
1633
1634 return VINF_SUCCESS;
1635}
1636
1637
1638#ifdef VBOX_STRICT
1639/**
1640 * Verifies that our cached host EFER value has not changed
1641 * since we cached it.
1642 *
1643 * @param pVCpu Pointer to the VMCPU.
1644 */
1645static void hmR0VmxCheckHostEferMsr(PVMCPU pVCpu)
1646{
1647 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1648
1649 if (pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR)
1650 {
1651 uint64_t u64Val;
1652 int rc = VMXReadVmcs64(VMX_VMCS64_HOST_FIELD_EFER_FULL, &u64Val);
1653 AssertRC(rc);
1654
1655 uint64_t u64HostEferMsr = ASMRdMsr(MSR_K6_EFER);
1656 AssertMsgReturnVoid(u64HostEferMsr == u64Val, ("u64HostEferMsr=%#RX64 u64Val=%#RX64\n", u64HostEferMsr, u64Val));
1657 }
1658}
1659
1660
1661/**
1662 * Verifies whether the guest/host MSR pairs in the auto-load/store area in the
1663 * VMCS are correct.
1664 *
1665 * @param pVCpu Pointer to the VMCPU.
1666 */
1667static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPU pVCpu)
1668{
1669 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1670
1671 /* Verify MSR counts in the VMCS are what we think it should be. */
1672 uint32_t cMsrs;
1673 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cMsrs); AssertRC(rc);
1674 Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1675
1676 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cMsrs); AssertRC(rc);
1677 Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1678
1679 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cMsrs); AssertRC(rc);
1680 Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1681
1682 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1683 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1684 for (uint32_t i = 0; i < cMsrs; i++, pHostMsr++, pGuestMsr++)
1685 {
1686 /* Verify that the MSRs are paired properly and that the host MSR has the correct value. */
1687 AssertMsgReturnVoid(pHostMsr->u32Msr == pGuestMsr->u32Msr, ("HostMsr=%#RX32 GuestMsr=%#RX32 cMsrs=%u\n", pHostMsr->u32Msr,
1688 pGuestMsr->u32Msr, cMsrs));
1689
1690 uint64_t u64Msr = ASMRdMsr(pHostMsr->u32Msr);
1691 AssertMsgReturnVoid(pHostMsr->u64Value == u64Msr, ("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
1692 pHostMsr->u32Msr, pHostMsr->u64Value, u64Msr, cMsrs));
1693
1694 /* Verify that the permissions are as expected in the MSR bitmap. */
1695 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1696 {
1697 VMXMSREXITREAD enmRead;
1698 VMXMSREXITWRITE enmWrite;
1699 rc = hmR0VmxGetMsrPermission(pVCpu, pGuestMsr->u32Msr, &enmRead, &enmWrite);
1700 AssertMsgReturnVoid(rc == VINF_SUCCESS, ("hmR0VmxGetMsrPermission! failed. rc=%Rrc\n", rc));
1701 if (pGuestMsr->u32Msr == MSR_K6_EFER)
1702 {
1703 AssertMsgReturnVoid(enmRead == VMXMSREXIT_INTERCEPT_READ, ("Passthru read for EFER!?\n"));
1704 AssertMsgReturnVoid(enmWrite == VMXMSREXIT_INTERCEPT_WRITE, ("Passthru write for EFER!?\n"));
1705 }
1706 else
1707 {
1708 AssertMsgReturnVoid(enmRead == VMXMSREXIT_PASSTHRU_READ, ("u32Msr=%#RX32 cMsrs=%u No passthru read!\n",
1709 pGuestMsr->u32Msr, cMsrs));
1710 AssertMsgReturnVoid(enmWrite == VMXMSREXIT_PASSTHRU_WRITE, ("u32Msr=%#RX32 cMsrs=%u No passthru write!\n",
1711 pGuestMsr->u32Msr, cMsrs));
1712 }
1713 }
1714 }
1715}
1716#endif /* VBOX_STRICT */
1717
1718
1719/**
1720 * Flushes the TLB using EPT.
1721 *
1722 * @returns VBox status code.
1723 * @param pVCpu Pointer to the VMCPU (can be NULL depending on @a
1724 * enmFlush).
1725 * @param enmFlush Type of flush.
1726 *
1727 * @remarks Caller is responsible for making sure this function is called only
1728 * when NestedPaging is supported and providing @a enmFlush that is
1729 * supported by the CPU.
1730 * @remarks Can be called with interrupts disabled.
1731 */
1732static void hmR0VmxFlushEpt(PVMCPU pVCpu, VMXFLUSHEPT enmFlush)
1733{
1734 uint64_t au64Descriptor[2];
1735 if (enmFlush == VMXFLUSHEPT_ALL_CONTEXTS)
1736 au64Descriptor[0] = 0;
1737 else
1738 {
1739 Assert(pVCpu);
1740 au64Descriptor[0] = pVCpu->hm.s.vmx.HCPhysEPTP;
1741 }
1742 au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */
1743
1744 int rc = VMXR0InvEPT(enmFlush, &au64Descriptor[0]);
1745 AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %RGv failed with %Rrc\n", enmFlush, pVCpu ? pVCpu->hm.s.vmx.HCPhysEPTP : 0,
1746 rc));
1747 if ( RT_SUCCESS(rc)
1748 && pVCpu)
1749 {
1750 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging);
1751 }
1752}
1753
1754
1755/**
1756 * Flushes the TLB using VPID.
1757 *
1758 * @returns VBox status code.
1759 * @param pVM Pointer to the VM.
1760 * @param pVCpu Pointer to the VMCPU (can be NULL depending on @a
1761 * enmFlush).
1762 * @param enmFlush Type of flush.
1763 * @param GCPtr Virtual address of the page to flush (can be 0 depending
1764 * on @a enmFlush).
1765 *
1766 * @remarks Can be called with interrupts disabled.
1767 */
1768static void hmR0VmxFlushVpid(PVM pVM, PVMCPU pVCpu, VMXFLUSHVPID enmFlush, RTGCPTR GCPtr)
1769{
1770 NOREF(pVM);
1771 AssertPtr(pVM);
1772 Assert(pVM->hm.s.vmx.fVpid);
1773
1774 uint64_t au64Descriptor[2];
1775 if (enmFlush == VMXFLUSHVPID_ALL_CONTEXTS)
1776 {
1777 au64Descriptor[0] = 0;
1778 au64Descriptor[1] = 0;
1779 }
1780 else
1781 {
1782 AssertPtr(pVCpu);
1783 AssertMsg(pVCpu->hm.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
1784 AssertMsg(pVCpu->hm.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
1785 au64Descriptor[0] = pVCpu->hm.s.uCurrentAsid;
1786 au64Descriptor[1] = GCPtr;
1787 }
1788
1789 int rc = VMXR0InvVPID(enmFlush, &au64Descriptor[0]); NOREF(rc);
1790 AssertMsg(rc == VINF_SUCCESS,
1791 ("VMXR0InvVPID %#x %u %RGv failed with %d\n", enmFlush, pVCpu ? pVCpu->hm.s.uCurrentAsid : 0, GCPtr, rc));
1792 if ( RT_SUCCESS(rc)
1793 && pVCpu)
1794 {
1795 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
1796 }
1797}
1798
1799
1800/**
1801 * Invalidates a guest page by guest virtual address. Only relevant for
1802 * EPT/VPID, otherwise there is nothing really to invalidate.
1803 *
1804 * @returns VBox status code.
1805 * @param pVM Pointer to the VM.
1806 * @param pVCpu Pointer to the VMCPU.
1807 * @param GCVirt Guest virtual address of the page to invalidate.
1808 */
1809VMMR0DECL(int) VMXR0InvalidatePage(PVM pVM, PVMCPU pVCpu, RTGCPTR GCVirt)
1810{
1811 AssertPtr(pVM);
1812 AssertPtr(pVCpu);
1813 LogFlowFunc(("pVM=%p pVCpu=%p GCVirt=%RGv\n", pVM, pVCpu, GCVirt));
1814
1815 bool fFlushPending = VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_FLUSH);
1816 if (!fFlushPending)
1817 {
1818 /*
1819 * We must invalidate the guest TLB entry in either case, we cannot ignore it even for the EPT case
1820 * See @bugref{6043} and @bugref{6177}.
1821 *
1822 * Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*() as this
1823 * function maybe called in a loop with individual addresses.
1824 */
1825 if (pVM->hm.s.vmx.fVpid)
1826 {
1827 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
1828 {
1829 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_INDIV_ADDR, GCVirt);
1830 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
1831 }
1832 else
1833 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1834 }
1835 else if (pVM->hm.s.fNestedPaging)
1836 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1837 }
1838
1839 return VINF_SUCCESS;
1840}
1841
1842
1843/**
1844 * Invalidates a guest page by physical address. Only relevant for EPT/VPID,
1845 * otherwise there is nothing really to invalidate.
1846 *
1847 * @returns VBox status code.
1848 * @param pVM Pointer to the VM.
1849 * @param pVCpu Pointer to the VMCPU.
1850 * @param GCPhys Guest physical address of the page to invalidate.
1851 */
1852VMMR0DECL(int) VMXR0InvalidatePhysPage(PVM pVM, PVMCPU pVCpu, RTGCPHYS GCPhys)
1853{
1854 NOREF(pVM); NOREF(GCPhys);
1855 LogFlowFunc(("%RGp\n", GCPhys));
1856
1857 /*
1858 * We cannot flush a page by guest-physical address. invvpid takes only a linear address while invept only flushes
1859 * by EPT not individual addresses. We update the force flag here and flush before the next VM-entry in hmR0VmxFlushTLB*().
1860 * This function might be called in a loop. This should cause a flush-by-EPT if EPT is in use. See @bugref{6568}.
1861 */
1862 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1863 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgPhys);
1864 return VINF_SUCCESS;
1865}
1866
1867
1868/**
1869 * Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the
1870 * case where neither EPT nor VPID is supported by the CPU.
1871 *
1872 * @param pVM Pointer to the VM.
1873 * @param pVCpu Pointer to the VMCPU.
1874 * @param pCpu Pointer to the global HM struct.
1875 *
1876 * @remarks Called with interrupts disabled.
1877 */
1878static void hmR0VmxFlushTaggedTlbNone(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
1879{
1880 AssertPtr(pVCpu);
1881 AssertPtr(pCpu);
1882 NOREF(pVM);
1883
1884 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
1885
1886 /** @todo TLB shootdown is currently not used. See hmQueueInvlPage(). */
1887#if 0
1888 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN);
1889 pVCpu->hm.s.TlbShootdown.cPages = 0;
1890#endif
1891
1892 Assert(pCpu->idCpu != NIL_RTCPUID);
1893 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
1894 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
1895 pVCpu->hm.s.fForceTLBFlush = false;
1896 return;
1897}
1898
1899
1900/**
1901 * Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary.
1902 *
1903 * @param pVM Pointer to the VM.
1904 * @param pVCpu Pointer to the VMCPU.
1905 * @param pCpu Pointer to the global HM CPU struct.
1906 * @remarks All references to "ASID" in this function pertains to "VPID" in
1907 * Intel's nomenclature. The reason is, to avoid confusion in compare
1908 * statements since the host-CPU copies are named "ASID".
1909 *
1910 * @remarks Called with interrupts disabled.
1911 */
1912static void hmR0VmxFlushTaggedTlbBoth(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
1913{
1914#ifdef VBOX_WITH_STATISTICS
1915 bool fTlbFlushed = false;
1916# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0)
1917# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \
1918 if (!fTlbFlushed) \
1919 STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \
1920 } while (0)
1921#else
1922# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0)
1923# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0)
1924#endif
1925
1926 AssertPtr(pVM);
1927 AssertPtr(pCpu);
1928 AssertPtr(pVCpu);
1929 Assert(pCpu->idCpu != NIL_RTCPUID);
1930
1931 AssertMsg(pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid,
1932 ("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled."
1933 "fNestedPaging=%RTbool fVpid=%RTbool", pVM->hm.s.fNestedPaging, pVM->hm.s.vmx.fVpid));
1934
1935 /*
1936 * Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
1937 * If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB
1938 * or the host CPU is online after a suspend/resume, so we cannot reuse the current ASID anymore.
1939 */
1940 if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
1941 || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
1942 {
1943 ++pCpu->uCurrentAsid;
1944 if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
1945 {
1946 pCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */
1947 pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
1948 pCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
1949 }
1950
1951 pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid;
1952 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
1953 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
1954
1955 /*
1956 * Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also
1957 * invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}.
1958 */
1959 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
1960 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
1961 HMVMX_SET_TAGGED_TLB_FLUSHED();
1962 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH); /* Already flushed-by-EPT, skip doing it again below. */
1963 }
1964
1965 /* Check for explicit TLB shootdowns. */
1966 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
1967 {
1968 /*
1969 * Changes to the EPT paging structure by VMM requires flushing by EPT as the CPU creates
1970 * guest-physical (only EPT-tagged) mappings while traversing the EPT tables when EPT is in use.
1971 * Flushing by VPID will only flush linear (only VPID-tagged) and combined (EPT+VPID tagged) mappings
1972 * but not guest-physical mappings.
1973 * See Intel spec. 28.3.2 "Creating and Using Cached Translation Information". See @bugref{6568}.
1974 */
1975 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
1976 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
1977 HMVMX_SET_TAGGED_TLB_FLUSHED();
1978 }
1979
1980 /** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere. See hmQueueInvlPage()
1981 * where it is commented out. Support individual entry flushing
1982 * someday. */
1983#if 0
1984 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN))
1985 {
1986 STAM_COUNTER_INC(&pVCpu->hm.s.StatTlbShootdown);
1987
1988 /*
1989 * Flush individual guest entries using VPID from the TLB or as little as possible with EPT
1990 * as supported by the CPU.
1991 */
1992 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
1993 {
1994 for (uint32_t i = 0; i < pVCpu->hm.s.TlbShootdown.cPages; i++)
1995 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_INDIV_ADDR, pVCpu->hm.s.TlbShootdown.aPages[i]);
1996 }
1997 else
1998 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
1999
2000 HMVMX_SET_TAGGED_TLB_FLUSHED();
2001 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN);
2002 pVCpu->hm.s.TlbShootdown.cPages = 0;
2003 }
2004#endif
2005
2006 pVCpu->hm.s.fForceTLBFlush = false;
2007
2008 HMVMX_UPDATE_FLUSH_SKIPPED_STAT();
2009
2010 Assert(pVCpu->hm.s.idLastCpu == pCpu->idCpu);
2011 Assert(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes);
2012 AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes,
2013 ("Flush count mismatch for cpu %d (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
2014 AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
2015 ("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pCpu->idCpu,
2016 pCpu->uCurrentAsid, pCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
2017 AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
2018 ("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
2019
2020 /* Update VMCS with the VPID. */
2021 int rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_VPID, pVCpu->hm.s.uCurrentAsid);
2022 AssertRC(rc);
2023
2024#undef HMVMX_SET_TAGGED_TLB_FLUSHED
2025}
2026
2027
2028/**
2029 * Flushes the tagged-TLB entries for EPT CPUs as necessary.
2030 *
2031 * @returns VBox status code.
2032 * @param pVM Pointer to the VM.
2033 * @param pVCpu Pointer to the VMCPU.
2034 * @param pCpu Pointer to the global HM CPU struct.
2035 *
2036 * @remarks Called with interrupts disabled.
2037 */
2038static void hmR0VmxFlushTaggedTlbEpt(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2039{
2040 AssertPtr(pVM);
2041 AssertPtr(pVCpu);
2042 AssertPtr(pCpu);
2043 Assert(pCpu->idCpu != NIL_RTCPUID);
2044 AssertMsg(pVM->hm.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with NestedPaging disabled."));
2045 AssertMsg(!pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID enabled."));
2046
2047 /*
2048 * Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
2049 * A change in the TLB flush count implies the host CPU is online after a suspend/resume.
2050 */
2051 if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
2052 || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
2053 {
2054 pVCpu->hm.s.fForceTLBFlush = true;
2055 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
2056 }
2057
2058 /* Check for explicit TLB shootdown flushes. */
2059 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
2060 {
2061 pVCpu->hm.s.fForceTLBFlush = true;
2062 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2063 }
2064
2065 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
2066 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
2067
2068 if (pVCpu->hm.s.fForceTLBFlush)
2069 {
2070 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
2071 pVCpu->hm.s.fForceTLBFlush = false;
2072 }
2073 /** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere. See hmQueueInvlPage()
2074 * where it is commented out. Support individual entry flushing
2075 * someday. */
2076#if 0
2077 else
2078 {
2079 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN))
2080 {
2081 /* We cannot flush individual entries without VPID support. Flush using EPT. */
2082 STAM_COUNTER_INC(&pVCpu->hm.s.StatTlbShootdown);
2083 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
2084 }
2085 else
2086 STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch);
2087
2088 pVCpu->hm.s.TlbShootdown.cPages = 0;
2089 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN);
2090 }
2091#endif
2092}
2093
2094
2095/**
2096 * Flushes the tagged-TLB entries for VPID CPUs as necessary.
2097 *
2098 * @returns VBox status code.
2099 * @param pVM Pointer to the VM.
2100 * @param pVCpu Pointer to the VMCPU.
2101 * @param pCpu Pointer to the global HM CPU struct.
2102 *
2103 * @remarks Called with interrupts disabled.
2104 */
2105static void hmR0VmxFlushTaggedTlbVpid(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2106{
2107 AssertPtr(pVM);
2108 AssertPtr(pVCpu);
2109 AssertPtr(pCpu);
2110 Assert(pCpu->idCpu != NIL_RTCPUID);
2111 AssertMsg(pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked with VPID disabled."));
2112 AssertMsg(!pVM->hm.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging enabled"));
2113
2114 /*
2115 * Force a TLB flush for the first world switch if the current CPU differs from the one we ran on last.
2116 * If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB
2117 * or the host CPU is online after a suspend/resume, so we cannot reuse the current ASID anymore.
2118 */
2119 if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
2120 || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
2121 {
2122 pVCpu->hm.s.fForceTLBFlush = true;
2123 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
2124 }
2125
2126 /* Check for explicit TLB shootdown flushes. */
2127 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
2128 {
2129 /*
2130 * If we ever support VPID flush combinations other than ALL or SINGLE-context (see hmR0VmxSetupTaggedTlb())
2131 * we would need to explicitly flush in this case (add an fExplicitFlush = true here and change the
2132 * pCpu->fFlushAsidBeforeUse check below to include fExplicitFlush's too) - an obscure corner case.
2133 */
2134 pVCpu->hm.s.fForceTLBFlush = true;
2135 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2136 }
2137
2138 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
2139 if (pVCpu->hm.s.fForceTLBFlush)
2140 {
2141 ++pCpu->uCurrentAsid;
2142 if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
2143 {
2144 pCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */
2145 pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
2146 pCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
2147 }
2148
2149 pVCpu->hm.s.fForceTLBFlush = false;
2150 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
2151 pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid;
2152 if (pCpu->fFlushAsidBeforeUse)
2153 {
2154 if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_SINGLE_CONTEXT)
2155 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_SINGLE_CONTEXT, 0 /* GCPtr */);
2156 else if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_ALL_CONTEXTS)
2157 {
2158 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_ALL_CONTEXTS, 0 /* GCPtr */);
2159 pCpu->fFlushAsidBeforeUse = false;
2160 }
2161 else
2162 {
2163 /* hmR0VmxSetupTaggedTlb() ensures we never get here. Paranoia. */
2164 AssertMsgFailed(("Unsupported VPID-flush context type.\n"));
2165 }
2166 }
2167 }
2168 /** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere. See hmQueueInvlPage()
2169 * where it is commented out. Support individual entry flushing
2170 * someday. */
2171#if 0
2172 else
2173 {
2174 AssertMsg(pVCpu->hm.s.uCurrentAsid && pCpu->uCurrentAsid,
2175 ("hm->uCurrentAsid=%lu hm->cTlbFlushes=%lu cpu->uCurrentAsid=%lu cpu->cTlbFlushes=%lu\n",
2176 pVCpu->hm.s.uCurrentAsid, pVCpu->hm.s.cTlbFlushes,
2177 pCpu->uCurrentAsid, pCpu->cTlbFlushes));
2178
2179 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN))
2180 {
2181 /* Flush individual guest entries using VPID or as little as possible with EPT as supported by the CPU. */
2182 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
2183 {
2184 for (uint32_t i = 0; i < pVCpu->hm.s.TlbShootdown.cPages; i++)
2185 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_INDIV_ADDR, pVCpu->hm.s.TlbShootdown.aPages[i]);
2186 }
2187 else
2188 hmR0VmxFlushVpid(pVM, pVCpu, pVM->hm.s.vmx.enmFlushVpid, 0 /* GCPtr */);
2189
2190 pVCpu->hm.s.TlbShootdown.cPages = 0;
2191 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN);
2192 }
2193 else
2194 STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch);
2195 }
2196#endif
2197
2198 AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes,
2199 ("Flush count mismatch for cpu %d (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
2200 AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
2201 ("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pCpu->idCpu,
2202 pCpu->uCurrentAsid, pCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
2203 AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
2204 ("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
2205
2206 int rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_VPID, pVCpu->hm.s.uCurrentAsid);
2207 AssertRC(rc);
2208}
2209
2210
2211/**
2212 * Flushes the guest TLB entry based on CPU capabilities.
2213 *
2214 * @param pVCpu Pointer to the VMCPU.
2215 * @param pCpu Pointer to the global HM CPU struct.
2216 */
2217DECLINLINE(void) hmR0VmxFlushTaggedTlb(PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2218{
2219#ifdef HMVMX_ALWAYS_FLUSH_TLB
2220 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2221#endif
2222 PVM pVM = pVCpu->CTX_SUFF(pVM);
2223 switch (pVM->hm.s.vmx.uFlushTaggedTlb)
2224 {
2225 case HMVMX_FLUSH_TAGGED_TLB_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pVM, pVCpu, pCpu); break;
2226 case HMVMX_FLUSH_TAGGED_TLB_EPT: hmR0VmxFlushTaggedTlbEpt(pVM, pVCpu, pCpu); break;
2227 case HMVMX_FLUSH_TAGGED_TLB_VPID: hmR0VmxFlushTaggedTlbVpid(pVM, pVCpu, pCpu); break;
2228 case HMVMX_FLUSH_TAGGED_TLB_NONE: hmR0VmxFlushTaggedTlbNone(pVM, pVCpu, pCpu); break;
2229 default:
2230 AssertMsgFailed(("Invalid flush-tag function identifier\n"));
2231 break;
2232 }
2233
2234 /* VMCPU_FF_TLB_SHOOTDOWN is unused. */
2235 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN));
2236
2237 /* Don't assert that VMCPU_FF_TLB_FLUSH should no longer be pending. It can be set by other EMTs. */
2238}
2239
2240
2241/**
2242 * Sets up the appropriate tagged TLB-flush level and handler for flushing guest
2243 * TLB entries from the host TLB before VM-entry.
2244 *
2245 * @returns VBox status code.
2246 * @param pVM Pointer to the VM.
2247 */
2248static int hmR0VmxSetupTaggedTlb(PVM pVM)
2249{
2250 /*
2251 * Determine optimal flush type for Nested Paging.
2252 * We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup unrestricted
2253 * guest execution (see hmR3InitFinalizeR0()).
2254 */
2255 if (pVM->hm.s.fNestedPaging)
2256 {
2257 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT)
2258 {
2259 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT)
2260 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_SINGLE_CONTEXT;
2261 else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
2262 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_ALL_CONTEXTS;
2263 else
2264 {
2265 /* Shouldn't happen. EPT is supported but no suitable flush-types supported. */
2266 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2267 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2268 }
2269
2270 /* Make sure the write-back cacheable memory type for EPT is supported. */
2271 if (!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB))
2272 {
2273 LogRel(("hmR0VmxSetupTaggedTlb: Unsupported EPTP memory type %#x.\n", pVM->hm.s.vmx.Msrs.u64EptVpidCaps));
2274 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2275 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2276 }
2277 }
2278 else
2279 {
2280 /* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */
2281 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2282 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2283 }
2284 }
2285
2286 /*
2287 * Determine optimal flush type for VPID.
2288 */
2289 if (pVM->hm.s.vmx.fVpid)
2290 {
2291 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID)
2292 {
2293 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT)
2294 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_SINGLE_CONTEXT;
2295 else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS)
2296 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_ALL_CONTEXTS;
2297 else
2298 {
2299 /* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */
2300 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
2301 LogRel(("hmR0VmxSetupTaggedTlb: Only INDIV_ADDR supported. Ignoring VPID.\n"));
2302 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
2303 LogRel(("hmR0VmxSetupTaggedTlb: Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
2304 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NOT_SUPPORTED;
2305 pVM->hm.s.vmx.fVpid = false;
2306 }
2307 }
2308 else
2309 {
2310 /* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */
2311 Log4(("hmR0VmxSetupTaggedTlb: VPID supported without INVEPT support. Ignoring VPID.\n"));
2312 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NOT_SUPPORTED;
2313 pVM->hm.s.vmx.fVpid = false;
2314 }
2315 }
2316
2317 /*
2318 * Setup the handler for flushing tagged-TLBs.
2319 */
2320 if (pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid)
2321 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_EPT_VPID;
2322 else if (pVM->hm.s.fNestedPaging)
2323 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_EPT;
2324 else if (pVM->hm.s.vmx.fVpid)
2325 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_VPID;
2326 else
2327 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_NONE;
2328 return VINF_SUCCESS;
2329}
2330
2331
2332/**
2333 * Sets up pin-based VM-execution controls in the VMCS.
2334 *
2335 * @returns VBox status code.
2336 * @param pVM Pointer to the VM.
2337 * @param pVCpu Pointer to the VMCPU.
2338 */
2339static int hmR0VmxSetupPinCtls(PVM pVM, PVMCPU pVCpu)
2340{
2341 AssertPtr(pVM);
2342 AssertPtr(pVCpu);
2343
2344 uint32_t val = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0; /* Bits set here must always be set. */
2345 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1; /* Bits cleared here must always be cleared. */
2346
2347 val |= VMX_VMCS_CTRL_PIN_EXEC_EXT_INT_EXIT /* External interrupts cause a VM-exit. */
2348 | VMX_VMCS_CTRL_PIN_EXEC_NMI_EXIT; /* Non-maskable interrupts (NMIs) cause a VM-exit. */
2349
2350 if (pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
2351 val |= VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI; /* Use virtual NMIs and virtual-NMI blocking features. */
2352
2353 /* Enable the VMX preemption timer. */
2354 if (pVM->hm.s.vmx.fUsePreemptTimer)
2355 {
2356 Assert(pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER);
2357 val |= VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER;
2358 }
2359
2360 if ((val & zap) != val)
2361 {
2362 LogRel(("hmR0VmxSetupPinCtls: invalid pin-based VM-execution controls combo! cpu=%#RX64 val=%#RX64 zap=%#RX64\n",
2363 pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0, val, zap));
2364 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC;
2365 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2366 }
2367
2368 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, val);
2369 AssertRCReturn(rc, rc);
2370
2371 pVCpu->hm.s.vmx.u32PinCtls = val;
2372 return rc;
2373}
2374
2375
2376/**
2377 * Sets up processor-based VM-execution controls in the VMCS.
2378 *
2379 * @returns VBox status code.
2380 * @param pVM Pointer to the VM.
2381 * @param pVMCPU Pointer to the VMCPU.
2382 */
2383static int hmR0VmxSetupProcCtls(PVM pVM, PVMCPU pVCpu)
2384{
2385 AssertPtr(pVM);
2386 AssertPtr(pVCpu);
2387
2388 int rc = VERR_INTERNAL_ERROR_5;
2389 uint32_t val = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0; /* Bits set here must be set in the VMCS. */
2390 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
2391
2392 val |= VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT /* HLT causes a VM-exit. */
2393 | VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING /* Use TSC-offsetting. */
2394 | VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */
2395 | VMX_VMCS_CTRL_PROC_EXEC_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */
2396 | VMX_VMCS_CTRL_PROC_EXEC_RDPMC_EXIT /* RDPMC causes a VM-exit. */
2397 | VMX_VMCS_CTRL_PROC_EXEC_MONITOR_EXIT /* MONITOR causes a VM-exit. */
2398 | VMX_VMCS_CTRL_PROC_EXEC_MWAIT_EXIT; /* MWAIT causes a VM-exit. */
2399
2400 /* We toggle VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */
2401 if ( !(pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT)
2402 || (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0 & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT))
2403 {
2404 LogRel(("hmR0VmxSetupProcCtls: unsupported VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT combo!"));
2405 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT;
2406 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2407 }
2408
2409 /* Without Nested Paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */
2410 if (!pVM->hm.s.fNestedPaging)
2411 {
2412 Assert(!pVM->hm.s.vmx.fUnrestrictedGuest); /* Paranoia. */
2413 val |= VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT
2414 | VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
2415 | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
2416 }
2417
2418 /* Use TPR shadowing if supported by the CPU. */
2419 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
2420 {
2421 Assert(pVCpu->hm.s.vmx.HCPhysVirtApic);
2422 Assert(!(pVCpu->hm.s.vmx.HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */
2423 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, 0);
2424 rc |= VMXWriteVmcs64(VMX_VMCS64_CTRL_VAPIC_PAGEADDR_FULL, pVCpu->hm.s.vmx.HCPhysVirtApic);
2425 AssertRCReturn(rc, rc);
2426
2427 val |= VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */
2428 /* CR8 writes cause a VM-exit based on TPR threshold. */
2429 Assert(!(val & VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT));
2430 Assert(!(val & VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT));
2431 }
2432 else
2433 {
2434 /*
2435 * Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is invalid on 32-bit Intel CPUs.
2436 * Set this control only for 64-bit guests.
2437 */
2438 if (pVM->hm.s.fAllow64BitGuests)
2439 {
2440 val |= VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT /* CR8 reads cause a VM-exit. */
2441 | VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT; /* CR8 writes cause a VM-exit. */
2442 }
2443 }
2444
2445 /* Use MSR-bitmaps if supported by the CPU. */
2446 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
2447 {
2448 val |= VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS;
2449
2450 Assert(pVCpu->hm.s.vmx.HCPhysMsrBitmap);
2451 Assert(!(pVCpu->hm.s.vmx.HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */
2452 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, pVCpu->hm.s.vmx.HCPhysMsrBitmap);
2453 AssertRCReturn(rc, rc);
2454
2455 /*
2456 * The guest can access the following MSRs (read, write) without causing VM-exits; they are loaded/stored
2457 * automatically using dedicated fields in the VMCS.
2458 */
2459 hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_CS, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2460 hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_ESP, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2461 hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_EIP, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2462 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_GS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2463 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_FS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2464
2465#if HC_ARCH_BITS == 64
2466 /*
2467 * Set passthru permissions for the following MSRs (mandatory for VT-x) required for 64-bit guests.
2468 */
2469 if (pVM->hm.s.fAllow64BitGuests)
2470 {
2471 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_LSTAR, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2472 hmR0VmxSetMsrPermission(pVCpu, MSR_K6_STAR, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2473 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_SF_MASK, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2474 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_KERNEL_GS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2475 }
2476#endif
2477 }
2478
2479 /* Use the secondary processor-based VM-execution controls if supported by the CPU. */
2480 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
2481 val |= VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL;
2482
2483 if ((val & zap) != val)
2484 {
2485 LogRel(("hmR0VmxSetupProcCtls: invalid processor-based VM-execution controls combo! cpu=%#RX64 val=%#RX64 zap=%#RX64\n",
2486 pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0, val, zap));
2487 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC;
2488 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2489 }
2490
2491 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, val);
2492 AssertRCReturn(rc, rc);
2493
2494 pVCpu->hm.s.vmx.u32ProcCtls = val;
2495
2496 /*
2497 * Secondary processor-based VM-execution controls.
2498 */
2499 if (RT_LIKELY(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL))
2500 {
2501 val = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0; /* Bits set here must be set in the VMCS. */
2502 zap = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
2503
2504 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT)
2505 val |= VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT; /* WBINVD causes a VM-exit. */
2506
2507 if (pVM->hm.s.fNestedPaging)
2508 val |= VMX_VMCS_CTRL_PROC_EXEC2_EPT; /* Enable EPT. */
2509 else
2510 {
2511 /*
2512 * Without Nested Paging, INVPCID should cause a VM-exit. Enabling this bit causes the CPU to refer to
2513 * VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT when INVPCID is executed by the guest.
2514 * See Intel spec. 25.4 "Changes to instruction behaviour in VMX non-root operation".
2515 */
2516 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_INVPCID)
2517 val |= VMX_VMCS_CTRL_PROC_EXEC2_INVPCID;
2518 }
2519
2520 if (pVM->hm.s.vmx.fVpid)
2521 val |= VMX_VMCS_CTRL_PROC_EXEC2_VPID; /* Enable VPID. */
2522
2523 if (pVM->hm.s.vmx.fUnrestrictedGuest)
2524 val |= VMX_VMCS_CTRL_PROC_EXEC2_UNRESTRICTED_GUEST; /* Enable Unrestricted Execution. */
2525
2526 /* Enable Virtual-APIC page accesses if supported by the CPU. This is essentially where the TPR shadow resides. */
2527 /** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be
2528 * done dynamically. */
2529 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)
2530 {
2531 Assert(pVM->hm.s.vmx.HCPhysApicAccess);
2532 Assert(!(pVM->hm.s.vmx.HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */
2533 val |= VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC; /* Virtualize APIC accesses. */
2534 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, pVM->hm.s.vmx.HCPhysApicAccess);
2535 AssertRCReturn(rc, rc);
2536 }
2537
2538 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP)
2539 val |= VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP; /* Enable RDTSCP support. */
2540
2541 if ((val & zap) != val)
2542 {
2543 LogRel(("hmR0VmxSetupProcCtls: invalid secondary processor-based VM-execution controls combo! "
2544 "cpu=%#RX64 val=%#RX64 zap=%#RX64\n", pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0, val, zap));
2545 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC2;
2546 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2547 }
2548
2549 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, val);
2550 AssertRCReturn(rc, rc);
2551
2552 pVCpu->hm.s.vmx.u32ProcCtls2 = val;
2553 }
2554 else if (RT_UNLIKELY(pVM->hm.s.vmx.fUnrestrictedGuest))
2555 {
2556 LogRel(("hmR0VmxSetupProcCtls: Unrestricted Guest set as true when secondary processor-based VM-execution controls not "
2557 "available\n"));
2558 pVCpu->hm.s.u32HMError = VMX_UFC_INVALID_UX_COMBO;
2559 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2560 }
2561
2562 return VINF_SUCCESS;
2563}
2564
2565
2566/**
2567 * Sets up miscellaneous (everything other than Pin & Processor-based
2568 * VM-execution) control fields in the VMCS.
2569 *
2570 * @returns VBox status code.
2571 * @param pVM Pointer to the VM.
2572 * @param pVCpu Pointer to the VMCPU.
2573 */
2574static int hmR0VmxSetupMiscCtls(PVM pVM, PVMCPU pVCpu)
2575{
2576 NOREF(pVM);
2577 AssertPtr(pVM);
2578 AssertPtr(pVCpu);
2579
2580 int rc = VERR_GENERAL_FAILURE;
2581
2582 /* All fields are zero-initialized during allocation; but don't remove the commented block below. */
2583#if 0
2584 /* All CR3 accesses cause VM-exits. Later we optimize CR3 accesses (see hmR0VmxLoadGuestCR3AndCR4())*/
2585 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, 0); AssertRCReturn(rc, rc);
2586 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, 0); AssertRCReturn(rc, rc);
2587
2588 /*
2589 * Set MASK & MATCH to 0. VMX checks if GuestPFErrCode & MASK == MATCH. If equal (in our case it always is)
2590 * and if the X86_XCPT_PF bit in the exception bitmap is set it causes a VM-exit, if clear doesn't cause an exit.
2591 * We thus use the exception bitmap to control it rather than use both.
2592 */
2593 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, 0); AssertRCReturn(rc, rc);
2594 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, 0); AssertRCReturn(rc, rc);
2595
2596 /** @todo Explore possibility of using IO-bitmaps. */
2597 /* All IO & IOIO instructions cause VM-exits. */
2598 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_IO_BITMAP_A_FULL, 0); AssertRCReturn(rc, rc);
2599 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_IO_BITMAP_B_FULL, 0); AssertRCReturn(rc, rc);
2600
2601 /* Initialize the MSR-bitmap area. */
2602 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, 0); AssertRCReturn(rc, rc);
2603 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, 0); AssertRCReturn(rc, rc);
2604 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, 0); AssertRCReturn(rc, rc);
2605#endif
2606
2607 /* Setup MSR auto-load/store area. */
2608 Assert(pVCpu->hm.s.vmx.HCPhysGuestMsr);
2609 Assert(!(pVCpu->hm.s.vmx.HCPhysGuestMsr & 0xf)); /* Lower 4 bits MBZ. */
2610 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, pVCpu->hm.s.vmx.HCPhysGuestMsr);
2611 AssertRCReturn(rc, rc);
2612 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, pVCpu->hm.s.vmx.HCPhysGuestMsr);
2613 AssertRCReturn(rc, rc);
2614
2615 Assert(pVCpu->hm.s.vmx.HCPhysHostMsr);
2616 Assert(!(pVCpu->hm.s.vmx.HCPhysHostMsr & 0xf)); /* Lower 4 bits MBZ. */
2617 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, pVCpu->hm.s.vmx.HCPhysHostMsr);
2618 AssertRCReturn(rc, rc);
2619
2620 /* Set VMCS link pointer. Reserved for future use, must be -1. Intel spec. 24.4 "Guest-State Area". */
2621 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, UINT64_C(0xffffffffffffffff));
2622 AssertRCReturn(rc, rc);
2623
2624 /* All fields are zero-initialized during allocation; but don't remove the commented block below. */
2625#if 0
2626 /* Setup debug controls */
2627 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, 0); /** @todo We don't support IA32_DEBUGCTL MSR. Should we? */
2628 AssertRCReturn(rc, rc);
2629 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, 0);
2630 AssertRCReturn(rc, rc);
2631#endif
2632
2633 return rc;
2634}
2635
2636
2637/**
2638 * Sets up the initial exception bitmap in the VMCS based on static conditions.
2639 *
2640 * @returns VBox status code.
2641 * @param pVM Pointer to the VM.
2642 * @param pVCpu Pointer to the VMCPU.
2643 */
2644static int hmR0VmxInitXcptBitmap(PVM pVM, PVMCPU pVCpu)
2645{
2646 AssertPtr(pVM);
2647 AssertPtr(pVCpu);
2648
2649 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
2650
2651 uint32_t u32XcptBitmap = pVCpu->hm.s.fGIMTrapXcptUD ? RT_BIT(X86_XCPT_UD) : 0;
2652
2653 /* Without Nested Paging, #PF must cause a VM-exit so we can sync our shadow page tables. */
2654 if (!pVM->hm.s.fNestedPaging)
2655 u32XcptBitmap |= RT_BIT(X86_XCPT_PF);
2656
2657 pVCpu->hm.s.vmx.u32XcptBitmap = u32XcptBitmap;
2658 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
2659 AssertRCReturn(rc, rc);
2660 return rc;
2661}
2662
2663
2664/**
2665 * Sets up the initial guest-state mask. The guest-state mask is consulted
2666 * before reading guest-state fields from the VMCS as VMREADs can be expensive
2667 * for the nested virtualization case (as it would cause a VM-exit).
2668 *
2669 * @param pVCpu Pointer to the VMCPU.
2670 */
2671static int hmR0VmxInitUpdatedGuestStateMask(PVMCPU pVCpu)
2672{
2673 /* Initially the guest-state is up-to-date as there is nothing in the VMCS. */
2674 HMVMXCPU_GST_RESET_TO(pVCpu, HMVMX_UPDATED_GUEST_ALL);
2675 return VINF_SUCCESS;
2676}
2677
2678
2679/**
2680 * Does per-VM VT-x initialization.
2681 *
2682 * @returns VBox status code.
2683 * @param pVM Pointer to the VM.
2684 */
2685VMMR0DECL(int) VMXR0InitVM(PVM pVM)
2686{
2687 LogFlowFunc(("pVM=%p\n", pVM));
2688
2689 int rc = hmR0VmxStructsAlloc(pVM);
2690 if (RT_FAILURE(rc))
2691 {
2692 LogRel(("VMXR0InitVM: hmR0VmxStructsAlloc failed! rc=%Rrc\n", rc));
2693 return rc;
2694 }
2695
2696 return VINF_SUCCESS;
2697}
2698
2699
2700/**
2701 * Does per-VM VT-x termination.
2702 *
2703 * @returns VBox status code.
2704 * @param pVM Pointer to the VM.
2705 */
2706VMMR0DECL(int) VMXR0TermVM(PVM pVM)
2707{
2708 LogFlowFunc(("pVM=%p\n", pVM));
2709
2710#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2711 if (pVM->hm.s.vmx.hMemObjScratch != NIL_RTR0MEMOBJ)
2712 ASMMemZero32(pVM->hm.s.vmx.pvScratch, PAGE_SIZE);
2713#endif
2714 hmR0VmxStructsFree(pVM);
2715 return VINF_SUCCESS;
2716}
2717
2718
2719/**
2720 * Sets up the VM for execution under VT-x.
2721 * This function is only called once per-VM during initialization.
2722 *
2723 * @returns VBox status code.
2724 * @param pVM Pointer to the VM.
2725 */
2726VMMR0DECL(int) VMXR0SetupVM(PVM pVM)
2727{
2728 AssertPtrReturn(pVM, VERR_INVALID_PARAMETER);
2729 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2730
2731 LogFlowFunc(("pVM=%p\n", pVM));
2732
2733 /*
2734 * Without UnrestrictedGuest, pRealModeTSS and pNonPagingModeEPTPageTable *must* always be allocated.
2735 * We no longer support the highly unlikely case of UnrestrictedGuest without pRealModeTSS. See hmR3InitFinalizeR0Intel().
2736 */
2737 if ( !pVM->hm.s.vmx.fUnrestrictedGuest
2738 && ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable
2739 || !pVM->hm.s.vmx.pRealModeTSS))
2740 {
2741 LogRel(("VMXR0SetupVM: invalid real-on-v86 state.\n"));
2742 return VERR_INTERNAL_ERROR;
2743 }
2744
2745#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
2746 /*
2747 * This is for the darwin 32-bit/PAE kernels trying to execute 64-bit guests. We don't bother with
2748 * the 32<->64 switcher in this case. This is a rare, legacy use-case with barely any test coverage.
2749 */
2750 if ( pVM->hm.s.fAllow64BitGuests
2751 && !HMVMX_IS_64BIT_HOST_MODE())
2752 {
2753 LogRel(("VMXR0SetupVM: Unsupported guest and host paging mode combination.\n"));
2754 return VERR_PGM_UNSUPPORTED_HOST_PAGING_MODE;
2755 }
2756#endif
2757
2758 /* Initialize these always, see hmR3InitFinalizeR0().*/
2759 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NONE;
2760 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NONE;
2761
2762 /* Setup the tagged-TLB flush handlers. */
2763 int rc = hmR0VmxSetupTaggedTlb(pVM);
2764 if (RT_FAILURE(rc))
2765 {
2766 LogRel(("VMXR0SetupVM: hmR0VmxSetupTaggedTlb failed! rc=%Rrc\n", rc));
2767 return rc;
2768 }
2769
2770 /* Check if we can use the VMCS controls for swapping the EFER MSR. */
2771 Assert(!pVM->hm.s.vmx.fSupportsVmcsEfer);
2772#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
2773 if ( HMVMX_IS_64BIT_HOST_MODE()
2774 && (pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1 & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR)
2775 && (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR)
2776 && (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR))
2777 {
2778 pVM->hm.s.vmx.fSupportsVmcsEfer = true;
2779 }
2780#endif
2781
2782 for (VMCPUID i = 0; i < pVM->cCpus; i++)
2783 {
2784 PVMCPU pVCpu = &pVM->aCpus[i];
2785 AssertPtr(pVCpu);
2786 AssertPtr(pVCpu->hm.s.vmx.pvVmcs);
2787
2788 /* Log the VCPU pointers, useful for debugging SMP VMs. */
2789 Log4(("VMXR0SetupVM: pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu));
2790
2791 /* Initialize the VM-exit history array with end-of-array markers (UINT16_MAX). */
2792 Assert(!pVCpu->hm.s.idxExitHistoryFree);
2793 HMCPU_EXIT_HISTORY_RESET(pVCpu);
2794
2795 /* Set revision dword at the beginning of the VMCS structure. */
2796 *(uint32_t *)pVCpu->hm.s.vmx.pvVmcs = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo);
2797
2798 /* Initialize our VMCS region in memory, set the VMCS launch state to "clear". */
2799 rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2800 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXClearVmcs failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2801 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2802
2803 /* Load this VMCS as the current VMCS. */
2804 rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2805 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXActivateVmcs failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2806 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2807
2808 rc = hmR0VmxSetupPinCtls(pVM, pVCpu);
2809 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupPinCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2810 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2811
2812 rc = hmR0VmxSetupProcCtls(pVM, pVCpu);
2813 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupProcCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2814 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2815
2816 rc = hmR0VmxSetupMiscCtls(pVM, pVCpu);
2817 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupMiscCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2818 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2819
2820 rc = hmR0VmxInitXcptBitmap(pVM, pVCpu);
2821 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitXcptBitmap failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2822 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2823
2824 rc = hmR0VmxInitUpdatedGuestStateMask(pVCpu);
2825 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitUpdatedGuestStateMask failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2826 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2827
2828#if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
2829 rc = hmR0VmxInitVmcsReadCache(pVM, pVCpu);
2830 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitVmcsReadCache failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2831 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2832#endif
2833
2834 /* Re-sync the CPU's internal data into our VMCS memory region & reset the launch state to "clear". */
2835 rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2836 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXClearVmcs(2) failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2837 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2838
2839 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
2840
2841 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc);
2842 }
2843
2844 return VINF_SUCCESS;
2845}
2846
2847
2848/**
2849 * Saves the host control registers (CR0, CR3, CR4) into the host-state area in
2850 * the VMCS.
2851 *
2852 * @returns VBox status code.
2853 * @param pVM Pointer to the VM.
2854 * @param pVCpu Pointer to the VMCPU.
2855 */
2856DECLINLINE(int) hmR0VmxSaveHostControlRegs(PVM pVM, PVMCPU pVCpu)
2857{
2858 NOREF(pVM); NOREF(pVCpu);
2859
2860 RTCCUINTREG uReg = ASMGetCR0();
2861 int rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR0, uReg);
2862 AssertRCReturn(rc, rc);
2863
2864#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
2865 /* For the darwin 32-bit hybrid kernel, we need the 64-bit CR3 as it uses 64-bit paging. */
2866 if (HMVMX_IS_64BIT_HOST_MODE())
2867 {
2868 uint64_t uRegCR3 = HMR0Get64bitCR3();
2869 rc = VMXWriteVmcs64(VMX_VMCS_HOST_CR3, uRegCR3);
2870 }
2871 else
2872#endif
2873 {
2874 uReg = ASMGetCR3();
2875 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR3, uReg);
2876 }
2877 AssertRCReturn(rc, rc);
2878
2879 uReg = ASMGetCR4();
2880 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR4, uReg);
2881 AssertRCReturn(rc, rc);
2882 return rc;
2883}
2884
2885
2886#if HC_ARCH_BITS == 64
2887/**
2888 * Macro for adjusting host segment selectors to satisfy VT-x's VM-entry
2889 * requirements. See hmR0VmxSaveHostSegmentRegs().
2890 */
2891# define VMXLOCAL_ADJUST_HOST_SEG(seg, selValue) \
2892 if ((selValue) & (X86_SEL_RPL | X86_SEL_LDT)) \
2893 { \
2894 bool fValidSelector = true; \
2895 if ((selValue) & X86_SEL_LDT) \
2896 { \
2897 uint32_t uAttr = ASMGetSegAttr((selValue)); \
2898 fValidSelector = RT_BOOL(uAttr != UINT32_MAX && (uAttr & X86_DESC_P)); \
2899 } \
2900 if (fValidSelector) \
2901 { \
2902 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_SEL_##seg; \
2903 pVCpu->hm.s.vmx.RestoreHost.uHostSel##seg = (selValue); \
2904 } \
2905 (selValue) = 0; \
2906 }
2907#endif
2908
2909
2910/**
2911 * Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into
2912 * the host-state area in the VMCS.
2913 *
2914 * @returns VBox status code.
2915 * @param pVM Pointer to the VM.
2916 * @param pVCpu Pointer to the VMCPU.
2917 */
2918DECLINLINE(int) hmR0VmxSaveHostSegmentRegs(PVM pVM, PVMCPU pVCpu)
2919{
2920 int rc = VERR_INTERNAL_ERROR_5;
2921
2922#if HC_ARCH_BITS == 64
2923 /*
2924 * If we've executed guest code using VT-x, the host-state bits will be messed up. We
2925 * should -not- save the messed up state without restoring the original host-state. See @bugref{7240}.
2926 */
2927 AssertMsgReturn(!(pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED),
2928 ("Re-saving host-state after executing guest code without leaving VT-x!\n"), VERR_WRONG_ORDER);
2929#endif
2930
2931 /*
2932 * Host DS, ES, FS and GS segment registers.
2933 */
2934#if HC_ARCH_BITS == 64
2935 RTSEL uSelDS = ASMGetDS();
2936 RTSEL uSelES = ASMGetES();
2937 RTSEL uSelFS = ASMGetFS();
2938 RTSEL uSelGS = ASMGetGS();
2939#else
2940 RTSEL uSelDS = 0;
2941 RTSEL uSelES = 0;
2942 RTSEL uSelFS = 0;
2943 RTSEL uSelGS = 0;
2944#endif
2945
2946 /* Recalculate which host-state bits need to be manually restored. */
2947 pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
2948
2949 /*
2950 * Host CS and SS segment registers.
2951 */
2952#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
2953 RTSEL uSelCS;
2954 RTSEL uSelSS;
2955 if (HMVMX_IS_64BIT_HOST_MODE())
2956 {
2957 uSelCS = (RTSEL)(uintptr_t)&SUPR0Abs64bitKernelCS;
2958 uSelSS = (RTSEL)(uintptr_t)&SUPR0Abs64bitKernelSS;
2959 }
2960 else
2961 {
2962 /* Seems darwin uses the LDT (TI flag is set) in the CS & SS selectors which VT-x doesn't like. */
2963 uSelCS = (RTSEL)(uintptr_t)&SUPR0AbsKernelCS;
2964 uSelSS = (RTSEL)(uintptr_t)&SUPR0AbsKernelSS;
2965 }
2966#else
2967 RTSEL uSelCS = ASMGetCS();
2968 RTSEL uSelSS = ASMGetSS();
2969#endif
2970
2971 /*
2972 * Host TR segment register.
2973 */
2974 RTSEL uSelTR = ASMGetTR();
2975
2976#if HC_ARCH_BITS == 64
2977 /*
2978 * Determine if the host segment registers are suitable for VT-x. Otherwise use zero to gain VM-entry and restore them
2979 * before we get preempted. See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers".
2980 */
2981 VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
2982 VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
2983 VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
2984 VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
2985# undef VMXLOCAL_ADJUST_HOST_SEG
2986#endif
2987
2988 /* Verification based on Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers" */
2989 Assert(!(uSelCS & X86_SEL_RPL)); Assert(!(uSelCS & X86_SEL_LDT));
2990 Assert(!(uSelSS & X86_SEL_RPL)); Assert(!(uSelSS & X86_SEL_LDT));
2991 Assert(!(uSelDS & X86_SEL_RPL)); Assert(!(uSelDS & X86_SEL_LDT));
2992 Assert(!(uSelES & X86_SEL_RPL)); Assert(!(uSelES & X86_SEL_LDT));
2993 Assert(!(uSelFS & X86_SEL_RPL)); Assert(!(uSelFS & X86_SEL_LDT));
2994 Assert(!(uSelGS & X86_SEL_RPL)); Assert(!(uSelGS & X86_SEL_LDT));
2995 Assert(!(uSelTR & X86_SEL_RPL)); Assert(!(uSelTR & X86_SEL_LDT));
2996 Assert(uSelCS);
2997 Assert(uSelTR);
2998
2999 /* Assertion is right but we would not have updated u32ExitCtls yet. */
3000#if 0
3001 if (!(pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE))
3002 Assert(uSelSS != 0);
3003#endif
3004
3005 /* Write these host selector fields into the host-state area in the VMCS. */
3006 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_CS, uSelCS); AssertRCReturn(rc, rc);
3007 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_SS, uSelSS); AssertRCReturn(rc, rc);
3008#if HC_ARCH_BITS == 64
3009 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_DS, uSelDS); AssertRCReturn(rc, rc);
3010 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_ES, uSelES); AssertRCReturn(rc, rc);
3011 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_FS, uSelFS); AssertRCReturn(rc, rc);
3012 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_GS, uSelGS); AssertRCReturn(rc, rc);
3013#endif
3014 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_TR, uSelTR); AssertRCReturn(rc, rc);
3015
3016 /*
3017 * Host GDTR and IDTR.
3018 */
3019 RTGDTR Gdtr;
3020 RT_ZERO(Gdtr);
3021#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
3022 if (HMVMX_IS_64BIT_HOST_MODE())
3023 {
3024 X86XDTR64 Gdtr64;
3025 X86XDTR64 Idtr64;
3026 HMR0Get64bitGdtrAndIdtr(&Gdtr64, &Idtr64);
3027 rc = VMXWriteVmcs64(VMX_VMCS_HOST_GDTR_BASE, Gdtr64.uAddr); AssertRCReturn(rc, rc);
3028 rc = VMXWriteVmcs64(VMX_VMCS_HOST_IDTR_BASE, Idtr64.uAddr); AssertRCReturn(rc, rc);
3029
3030 Gdtr.cbGdt = Gdtr64.cb;
3031 Gdtr.pGdt = (uintptr_t)Gdtr64.uAddr;
3032 }
3033 else
3034#endif
3035 {
3036 RTIDTR Idtr;
3037 ASMGetGDTR(&Gdtr);
3038 ASMGetIDTR(&Idtr);
3039 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, Gdtr.pGdt); AssertRCReturn(rc, rc);
3040 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, Idtr.pIdt); AssertRCReturn(rc, rc);
3041
3042#if HC_ARCH_BITS == 64
3043 /*
3044 * Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps them to the
3045 * maximum limit (0xffff) on every VM-exit.
3046 */
3047 if (Gdtr.cbGdt != 0xffff)
3048 {
3049 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_GDTR;
3050 AssertCompile(sizeof(Gdtr) == sizeof(X86XDTR64));
3051 memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
3052 }
3053
3054 /*
3055 * IDT limit is effectively capped at 0xfff. (See Intel spec. 6.14.1 "64-Bit Mode IDT"
3056 * and Intel spec. 6.2 "Exception and Interrupt Vectors".) Therefore if the host has the limit as 0xfff, VT-x
3057 * bloating the limit to 0xffff shouldn't cause any different CPU behavior. However, several hosts either insists
3058 * on 0xfff being the limit (Windows Patch Guard) or uses the limit for other purposes (darwin puts the CPU ID in there
3059 * but botches sidt alignment in at least one consumer). So, we're only allowing IDTR.LIMIT to be left at 0xffff on
3060 * hosts where we are pretty sure it won't cause trouble.
3061 */
3062# if defined(RT_OS_LINUX) || defined(RT_OS_SOLARIS)
3063 if (Idtr.cbIdt < 0x0fff)
3064# else
3065 if (Idtr.cbIdt != 0xffff)
3066# endif
3067 {
3068 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_IDTR;
3069 AssertCompile(sizeof(Idtr) == sizeof(X86XDTR64));
3070 memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostIdtr, &Idtr, sizeof(X86XDTR64));
3071 }
3072#endif
3073 }
3074
3075 /*
3076 * Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI and RPL bits
3077 * is effectively what the CPU does for "scaling by 8". TI is always 0 and RPL should be too in most cases.
3078 */
3079 AssertMsgReturn((uSelTR | X86_SEL_RPL_LDT) <= Gdtr.cbGdt,
3080 ("hmR0VmxSaveHostSegmentRegs: TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, Gdtr.cbGdt),
3081 VERR_VMX_INVALID_HOST_STATE);
3082
3083 PCX86DESCHC pDesc = (PCX86DESCHC)(Gdtr.pGdt + (uSelTR & X86_SEL_MASK));
3084#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
3085 if (HMVMX_IS_64BIT_HOST_MODE())
3086 {
3087 /* We need the 64-bit TR base for hybrid darwin. */
3088 uint64_t u64TRBase = X86DESC64_BASE((PX86DESC64)pDesc);
3089 rc = VMXWriteVmcs64(VMX_VMCS_HOST_TR_BASE, u64TRBase);
3090 }
3091 else
3092#endif
3093 {
3094 uintptr_t uTRBase;
3095#if HC_ARCH_BITS == 64
3096 uTRBase = X86DESC64_BASE(pDesc);
3097
3098 /*
3099 * VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on all VM-exits.
3100 * The type is the same for 64-bit busy TSS[1]. The limit needs manual restoration if the host has something else.
3101 * Task switching is not supported in 64-bit mode[2], but the limit still matters as IOPM is supported in 64-bit mode.
3102 * Restoring the limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0.
3103 *
3104 * [1] See Intel spec. 3.5 "System Descriptor Types".
3105 * [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode".
3106 */
3107 Assert(pDesc->System.u4Type == 11);
3108 if ( pDesc->System.u16LimitLow != 0x67
3109 || pDesc->System.u4LimitHigh)
3110 {
3111 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_SEL_TR;
3112 /* If the host has made GDT read-only, we would need to temporarily toggle CR0.WP before writing the GDT. */
3113 if (pVM->hm.s.uHostKernelFeatures & SUPKERNELFEATURES_GDT_READ_ONLY)
3114 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_GDT_READ_ONLY;
3115 pVCpu->hm.s.vmx.RestoreHost.uHostSelTR = uSelTR;
3116
3117 /* Store the GDTR here as we need it while restoring TR. */
3118 memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
3119 }
3120#else
3121 uTRBase = X86DESC_BASE(pDesc);
3122#endif
3123 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_TR_BASE, uTRBase);
3124 }
3125 AssertRCReturn(rc, rc);
3126
3127 /*
3128 * Host FS base and GS base.
3129 */
3130#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
3131 if (HMVMX_IS_64BIT_HOST_MODE())
3132 {
3133 uint64_t u64FSBase = ASMRdMsr(MSR_K8_FS_BASE);
3134 uint64_t u64GSBase = ASMRdMsr(MSR_K8_GS_BASE);
3135 rc = VMXWriteVmcs64(VMX_VMCS_HOST_FS_BASE, u64FSBase); AssertRCReturn(rc, rc);
3136 rc = VMXWriteVmcs64(VMX_VMCS_HOST_GS_BASE, u64GSBase); AssertRCReturn(rc, rc);
3137
3138# if HC_ARCH_BITS == 64
3139 /* Store the base if we have to restore FS or GS manually as we need to restore the base as well. */
3140 if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_FS)
3141 pVCpu->hm.s.vmx.RestoreHost.uHostFSBase = u64FSBase;
3142 if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_GS)
3143 pVCpu->hm.s.vmx.RestoreHost.uHostGSBase = u64GSBase;
3144# endif
3145 }
3146#endif
3147 return rc;
3148}
3149
3150
3151/**
3152 * Saves certain host MSRs in the VM-Exit MSR-load area and some in the
3153 * host-state area of the VMCS. Theses MSRs will be automatically restored on
3154 * the host after every successful VM-exit.
3155 *
3156 * @returns VBox status code.
3157 * @param pVM Pointer to the VM.
3158 * @param pVCpu Pointer to the VMCPU.
3159 *
3160 * @remarks No-long-jump zone!!!
3161 */
3162DECLINLINE(int) hmR0VmxSaveHostMsrs(PVM pVM, PVMCPU pVCpu)
3163{
3164 NOREF(pVM);
3165
3166 AssertPtr(pVCpu);
3167 AssertPtr(pVCpu->hm.s.vmx.pvHostMsr);
3168
3169 int rc = VINF_SUCCESS;
3170#if HC_ARCH_BITS == 64
3171 if (pVM->hm.s.fAllow64BitGuests)
3172 hmR0VmxLazySaveHostMsrs(pVCpu);
3173#endif
3174
3175 /*
3176 * Host Sysenter MSRs.
3177 */
3178 rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS));
3179 AssertRCReturn(rc, rc);
3180#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
3181 if (HMVMX_IS_64BIT_HOST_MODE())
3182 {
3183 rc = VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP));
3184 AssertRCReturn(rc, rc);
3185 rc = VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP));
3186 }
3187 else
3188 {
3189 rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr_Low(MSR_IA32_SYSENTER_ESP));
3190 AssertRCReturn(rc, rc);
3191 rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr_Low(MSR_IA32_SYSENTER_EIP));
3192 }
3193#elif HC_ARCH_BITS == 32
3194 rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr_Low(MSR_IA32_SYSENTER_ESP));
3195 AssertRCReturn(rc, rc);
3196 rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr_Low(MSR_IA32_SYSENTER_EIP));
3197#else
3198 rc = VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP));
3199 AssertRCReturn(rc, rc);
3200 rc = VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP));
3201#endif
3202 AssertRCReturn(rc, rc);
3203
3204 /*
3205 * Host EFER MSR.
3206 * If the CPU supports the newer VMCS controls for managing EFER, use it.
3207 * Otherwise it's done as part of auto-load/store MSR area in the VMCS, see hmR0VmxLoadGuestMsrs().
3208 */
3209 if (pVM->hm.s.vmx.fSupportsVmcsEfer)
3210 {
3211 rc = VMXWriteVmcs64(VMX_VMCS64_HOST_FIELD_EFER_FULL, pVM->hm.s.vmx.u64HostEfer);
3212 AssertRCReturn(rc, rc);
3213 }
3214
3215 /** @todo IA32_PERF_GLOBALCTRL, IA32_PAT also see
3216 * hmR0VmxLoadGuestExitCtls() !! */
3217
3218 return rc;
3219}
3220
3221
3222/**
3223 * Figures out if we need to swap the EFER MSR which is
3224 * particularly expensive.
3225 *
3226 * We check all relevant bits. For now, that's everything
3227 * besides LMA/LME, as these two bits are handled by VM-entry,
3228 * see hmR0VmxLoadGuestExitCtls() and
3229 * hmR0VMxLoadGuestEntryCtls().
3230 *
3231 * @returns true if we need to load guest EFER, false otherwise.
3232 * @param pVCpu Pointer to the VMCPU.
3233 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3234 * out-of-sync. Make sure to update the required fields
3235 * before using them.
3236 *
3237 * @remarks Requires EFER, CR4.
3238 * @remarks No-long-jump zone!!!
3239 */
3240static bool hmR0VmxShouldSwapEferMsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3241{
3242#ifdef HMVMX_ALWAYS_SWAP_EFER
3243 return true;
3244#endif
3245
3246#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
3247 /* For 32-bit hosts running 64-bit guests, we always swap EFER in the world-switcher. Nothing to do here. */
3248 if (CPUMIsGuestInLongMode(pVCpu))
3249 return false;
3250#endif
3251
3252 PVM pVM = pVCpu->CTX_SUFF(pVM);
3253 uint64_t u64HostEfer = pVM->hm.s.vmx.u64HostEfer;
3254 uint64_t u64GuestEfer = pMixedCtx->msrEFER;
3255
3256 /*
3257 * For 64-bit guests, if EFER.SCE bit differs, we need to swap to ensure that the
3258 * guest's SYSCALL behaviour isn't screwed. See @bugref{7386}.
3259 */
3260 if ( CPUMIsGuestInLongMode(pVCpu)
3261 && (u64GuestEfer & MSR_K6_EFER_SCE) != (u64HostEfer & MSR_K6_EFER_SCE))
3262 {
3263 return true;
3264 }
3265
3266 /*
3267 * If the guest uses PAE and EFER.NXE bit differs, we need to swap EFER as it
3268 * affects guest paging. 64-bit paging implies CR4.PAE as well.
3269 * See Intel spec. 4.5 "IA-32e Paging" and Intel spec. 4.1.1 "Three Paging Modes".
3270 */
3271 if ( (pMixedCtx->cr4 & X86_CR4_PAE)
3272 && (pMixedCtx->cr0 & X86_CR0_PG)
3273 && (u64GuestEfer & MSR_K6_EFER_NXE) != (u64HostEfer & MSR_K6_EFER_NXE))
3274 {
3275 /* Assert that host is PAE capable. */
3276 Assert(pVM->hm.s.cpuid.u32AMDFeatureEDX & X86_CPUID_EXT_FEATURE_EDX_NX);
3277 return true;
3278 }
3279
3280 /** @todo Check the latest Intel spec. for any other bits,
3281 * like SMEP/SMAP? */
3282 return false;
3283}
3284
3285
3286/**
3287 * Sets up VM-entry controls in the VMCS. These controls can affect things done
3288 * on VM-exit; e.g. "load debug controls", see Intel spec. 24.8.1 "VM-entry
3289 * controls".
3290 *
3291 * @returns VBox status code.
3292 * @param pVCpu Pointer to the VMCPU.
3293 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3294 * out-of-sync. Make sure to update the required fields
3295 * before using them.
3296 *
3297 * @remarks Requires EFER.
3298 * @remarks No-long-jump zone!!!
3299 */
3300DECLINLINE(int) hmR0VmxLoadGuestEntryCtls(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3301{
3302 int rc = VINF_SUCCESS;
3303 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_ENTRY_CTLS))
3304 {
3305 PVM pVM = pVCpu->CTX_SUFF(pVM);
3306 uint32_t val = pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0; /* Bits set here must be set in the VMCS. */
3307 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3308
3309 /* Load debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x capable CPUs only supports the 1-setting of this bit. */
3310 val |= VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG;
3311
3312 /* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry. */
3313 if (CPUMIsGuestInLongModeEx(pMixedCtx))
3314 {
3315 val |= VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST;
3316 Log4(("Load[%RU32]: VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST\n", pVCpu->idCpu));
3317 }
3318 else
3319 Assert(!(val & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST));
3320
3321 /* If the CPU supports the newer VMCS controls for managing guest/host EFER, use it. */
3322 if ( pVM->hm.s.vmx.fSupportsVmcsEfer
3323 && hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
3324 {
3325 val |= VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR;
3326 Log4(("Load[%RU32]: VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR\n", pVCpu->idCpu));
3327 }
3328
3329 /*
3330 * The following should -not- be set (since we're not in SMM mode):
3331 * - VMX_VMCS_CTRL_ENTRY_ENTRY_SMM
3332 * - VMX_VMCS_CTRL_ENTRY_DEACTIVATE_DUALMON
3333 */
3334
3335 /** @todo VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR,
3336 * VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR. */
3337
3338 if ((val & zap) != val)
3339 {
3340 LogRel(("hmR0VmxLoadGuestEntryCtls: invalid VM-entry controls combo! cpu=%RX64 val=%RX64 zap=%RX64\n",
3341 pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0, val, zap));
3342 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY;
3343 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3344 }
3345
3346 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, val);
3347 AssertRCReturn(rc, rc);
3348
3349 pVCpu->hm.s.vmx.u32EntryCtls = val;
3350 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_ENTRY_CTLS);
3351 }
3352 return rc;
3353}
3354
3355
3356/**
3357 * Sets up the VM-exit controls in the VMCS.
3358 *
3359 * @returns VBox status code.
3360 * @param pVM Pointer to the VM.
3361 * @param pVCpu Pointer to the VMCPU.
3362 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3363 * out-of-sync. Make sure to update the required fields
3364 * before using them.
3365 *
3366 * @remarks Requires EFER.
3367 */
3368DECLINLINE(int) hmR0VmxLoadGuestExitCtls(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3369{
3370 NOREF(pMixedCtx);
3371
3372 int rc = VINF_SUCCESS;
3373 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_EXIT_CTLS))
3374 {
3375 PVM pVM = pVCpu->CTX_SUFF(pVM);
3376 uint32_t val = pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0; /* Bits set here must be set in the VMCS. */
3377 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3378
3379 /* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only supported the 1-setting of this bit. */
3380 val |= VMX_VMCS_CTRL_EXIT_SAVE_DEBUG;
3381
3382 /*
3383 * Set the host long mode active (EFER.LMA) bit (which Intel calls "Host address-space size") if necessary.
3384 * On VM-exit, VT-x sets both the host EFER.LMA and EFER.LME bit to this value. See assertion in hmR0VmxSaveHostMsrs().
3385 */
3386#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
3387 if (HMVMX_IS_64BIT_HOST_MODE())
3388 {
3389 val |= VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE;
3390 Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE\n", pVCpu->idCpu));
3391 }
3392 else
3393 Assert(!(val & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE));
3394#else
3395 if (CPUMIsGuestInLongModeEx(pMixedCtx))
3396 {
3397 /* The switcher returns to long mode, EFER is managed by the switcher. */
3398 val |= VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE;
3399 Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE\n", pVCpu->idCpu));
3400 }
3401 else
3402 Assert(!(val & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE));
3403#endif /* HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) */
3404
3405 /* If the newer VMCS fields for managing EFER exists, use it. */
3406 if ( pVM->hm.s.vmx.fSupportsVmcsEfer
3407 && hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
3408 {
3409 val |= VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR
3410 | VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR;
3411 Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR, VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR\n", pVCpu->idCpu));
3412 }
3413
3414 /* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */
3415 Assert(!(val & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT));
3416
3417 /** @todo VMX_VMCS_CTRL_EXIT_LOAD_PERF_MSR,
3418 * VMX_VMCS_CTRL_EXIT_SAVE_GUEST_PAT_MSR,
3419 * VMX_VMCS_CTRL_EXIT_LOAD_HOST_PAT_MSR. */
3420
3421 if ( pVM->hm.s.vmx.fUsePreemptTimer
3422 && (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER))
3423 val |= VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER;
3424
3425 if ((val & zap) != val)
3426 {
3427 LogRel(("hmR0VmxSetupProcCtls: invalid VM-exit controls combo! cpu=%RX64 val=%RX64 zap=%RX64\n",
3428 pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0, val, zap));
3429 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT;
3430 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3431 }
3432
3433 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, val);
3434 AssertRCReturn(rc, rc);
3435
3436 pVCpu->hm.s.vmx.u32ExitCtls = val;
3437 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_EXIT_CTLS);
3438 }
3439 return rc;
3440}
3441
3442
3443/**
3444 * Loads the guest APIC and related state.
3445 *
3446 * @returns VBox status code.
3447 * @param pVM Pointer to the VM.
3448 * @param pVCpu Pointer to the VMCPU.
3449 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3450 * out-of-sync. Make sure to update the required fields
3451 * before using them.
3452 */
3453DECLINLINE(int) hmR0VmxLoadGuestApicState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3454{
3455 NOREF(pMixedCtx);
3456
3457 int rc = VINF_SUCCESS;
3458 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE))
3459 {
3460 /* Setup TPR shadowing. Also setup TPR patching for 32-bit guests. */
3461 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
3462 {
3463 Assert(pVCpu->hm.s.vmx.HCPhysVirtApic);
3464
3465 bool fPendingIntr = false;
3466 uint8_t u8Tpr = 0;
3467 uint8_t u8PendingIntr = 0;
3468 rc = PDMApicGetTPR(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr);
3469 AssertRCReturn(rc, rc);
3470
3471 /*
3472 * If there are external interrupts pending but masked by the TPR value, instruct VT-x to cause a VM-exit when
3473 * the guest lowers its TPR below the highest-priority pending interrupt and we can deliver the interrupt.
3474 * If there are no external interrupts pending, set threshold to 0 to not cause a VM-exit. We will eventually deliver
3475 * the interrupt when we VM-exit for other reasons.
3476 */
3477 pVCpu->hm.s.vmx.pbVirtApic[0x80] = u8Tpr; /* Offset 0x80 is TPR in the APIC MMIO range. */
3478 uint32_t u32TprThreshold = 0;
3479 if (fPendingIntr)
3480 {
3481 /* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR (which is the Task-Priority Class). */
3482 const uint8_t u8PendingPriority = (u8PendingIntr >> 4) & 0xf;
3483 const uint8_t u8TprPriority = (u8Tpr >> 4) & 0xf;
3484 if (u8PendingPriority <= u8TprPriority)
3485 u32TprThreshold = u8PendingPriority;
3486 else
3487 u32TprThreshold = u8TprPriority; /* Required for Vista 64-bit guest, see @bugref{6398}. */
3488 }
3489 Assert(!(u32TprThreshold & 0xfffffff0)); /* Bits 31:4 MBZ. */
3490
3491 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
3492 AssertRCReturn(rc, rc);
3493 }
3494
3495 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
3496 }
3497 return rc;
3498}
3499
3500
3501/**
3502 * Gets the guest's interruptibility-state ("interrupt shadow" as AMD calls it).
3503 *
3504 * @returns Guest's interruptibility-state.
3505 * @param pVCpu Pointer to the VMCPU.
3506 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3507 * out-of-sync. Make sure to update the required fields
3508 * before using them.
3509 *
3510 * @remarks No-long-jump zone!!!
3511 */
3512DECLINLINE(uint32_t) hmR0VmxGetGuestIntrState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3513{
3514 /*
3515 * Check if we should inhibit interrupt delivery due to instructions like STI and MOV SS.
3516 */
3517 uint32_t uIntrState = 0;
3518 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
3519 {
3520 /* If inhibition is active, RIP & RFLAGS should've been accessed (i.e. read previously from the VMCS or from ring-3). */
3521 AssertMsg(HMVMXCPU_GST_IS_SET(pVCpu, HMVMX_UPDATED_GUEST_RIP | HMVMX_UPDATED_GUEST_RFLAGS),
3522 ("%#x\n", HMVMXCPU_GST_VALUE(pVCpu)));
3523 if (pMixedCtx->rip == EMGetInhibitInterruptsPC(pVCpu))
3524 {
3525 if (pMixedCtx->eflags.Bits.u1IF)
3526 uIntrState = VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI;
3527 else
3528 uIntrState = VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS;
3529 }
3530 /* else: Although we can clear the force-flag here, let's keep this side-effects free. */
3531 }
3532
3533 /*
3534 * NMIs to the guest are blocked after an NMI is injected until the guest executes an IRET. We only
3535 * bother with virtual-NMI blocking when we have support for virtual NMIs in the CPU, otherwise
3536 * setting this would block host-NMIs and IRET will not clear the blocking.
3537 *
3538 * See Intel spec. 26.6.1 "Interruptibility state". See @bugref{7445}.
3539 */
3540 if ( VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS)
3541 && (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI))
3542 {
3543 uIntrState |= VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI;
3544 }
3545
3546 return uIntrState;
3547}
3548
3549
3550/**
3551 * Loads the guest's interruptibility-state into the guest-state area in the
3552 * VMCS.
3553 *
3554 * @returns VBox status code.
3555 * @param pVCpu Pointer to the VMCPU.
3556 * @param uIntrState The interruptibility-state to set.
3557 */
3558static int hmR0VmxLoadGuestIntrState(PVMCPU pVCpu, uint32_t uIntrState)
3559{
3560 NOREF(pVCpu);
3561 AssertMsg(!(uIntrState & 0xfffffff0), ("%#x\n", uIntrState)); /* Bits 31:4 MBZ. */
3562 Assert((uIntrState & 0x3) != 0x3); /* Block-by-STI and MOV SS cannot be simultaneously set. */
3563 int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, uIntrState);
3564 AssertRCReturn(rc, rc);
3565 return rc;
3566}
3567
3568
3569/**
3570 * Loads the exception intercepts required for guest execution in the VMCS.
3571 *
3572 * @returns VBox status code.
3573 * @param pVCpu Pointer to the VMCPU.
3574 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3575 * out-of-sync. Make sure to update the required fields
3576 * before using them.
3577 */
3578static int hmR0VmxLoadGuestXcptIntercepts(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3579{
3580 NOREF(pMixedCtx);
3581 int rc = VINF_SUCCESS;
3582 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS))
3583 {
3584 /* The remaining exception intercepts are handled elsewhere, e.g. in hmR0VmxLoadSharedCR0(). */
3585 if (pVCpu->hm.s.fGIMTrapXcptUD)
3586 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_UD);
3587 else
3588 {
3589#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
3590 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_UD);
3591#endif
3592 }
3593
3594 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
3595 AssertRCReturn(rc, rc);
3596
3597 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
3598 Log4(("Load[%RU32]: VMX_VMCS32_CTRL_EXCEPTION_BITMAP=%#RX64 fContextUseFlags=%#RX32\n", pVCpu->idCpu,
3599 pVCpu->hm.s.vmx.u32XcptBitmap, HMCPU_CF_VALUE(pVCpu)));
3600 }
3601 return rc;
3602}
3603
3604
3605/**
3606 * Loads the guest's RIP into the guest-state area in the VMCS.
3607 *
3608 * @returns VBox status code.
3609 * @param pVCpu Pointer to the VMCPU.
3610 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3611 * out-of-sync. Make sure to update the required fields
3612 * before using them.
3613 *
3614 * @remarks No-long-jump zone!!!
3615 */
3616static int hmR0VmxLoadGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3617{
3618 int rc = VINF_SUCCESS;
3619 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RIP))
3620 {
3621 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RIP, pMixedCtx->rip);
3622 AssertRCReturn(rc, rc);
3623
3624 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RIP);
3625 Log4(("Load[%RU32]: VMX_VMCS_GUEST_RIP=%#RX64 fContextUseFlags=%#RX32\n", pVCpu->idCpu, pMixedCtx->rip,
3626 HMCPU_CF_VALUE(pVCpu)));
3627 }
3628 return rc;
3629}
3630
3631
3632/**
3633 * Loads the guest's RSP into the guest-state area in the VMCS.
3634 *
3635 * @returns VBox status code.
3636 * @param pVCpu Pointer to the VMCPU.
3637 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3638 * out-of-sync. Make sure to update the required fields
3639 * before using them.
3640 *
3641 * @remarks No-long-jump zone!!!
3642 */
3643static int hmR0VmxLoadGuestRsp(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3644{
3645 int rc = VINF_SUCCESS;
3646 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RSP))
3647 {
3648 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RSP, pMixedCtx->rsp);
3649 AssertRCReturn(rc, rc);
3650
3651 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RSP);
3652 Log4(("Load[%RU32]: VMX_VMCS_GUEST_RSP=%#RX64\n", pVCpu->idCpu, pMixedCtx->rsp));
3653 }
3654 return rc;
3655}
3656
3657
3658/**
3659 * Loads the guest's RFLAGS into the guest-state area in the VMCS.
3660 *
3661 * @returns VBox status code.
3662 * @param pVCpu Pointer to the VMCPU.
3663 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3664 * out-of-sync. Make sure to update the required fields
3665 * before using them.
3666 *
3667 * @remarks No-long-jump zone!!!
3668 */
3669static int hmR0VmxLoadGuestRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3670{
3671 int rc = VINF_SUCCESS;
3672 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS))
3673 {
3674 /* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ).
3675 Let us assert it as such and use 32-bit VMWRITE. */
3676 Assert(!(pMixedCtx->rflags.u64 >> 32));
3677 X86EFLAGS Eflags = pMixedCtx->eflags;
3678 /** @todo r=bird: There shall be no need to OR in X86_EFL_1 here, nor
3679 * shall there be any reason for clearing bits 63:22, 15, 5 and 3.
3680 * These will never be cleared/set, unless some other part of the VMM
3681 * code is buggy - in which case we're better of finding and fixing
3682 * those bugs than hiding them. */
3683 Assert(Eflags.u32 & X86_EFL_RA1_MASK);
3684 Assert(!(Eflags.u32 & ~(X86_EFL_1 | X86_EFL_LIVE_MASK)));
3685 Eflags.u32 &= VMX_EFLAGS_RESERVED_0; /* Bits 22-31, 15, 5 & 3 MBZ. */
3686 Eflags.u32 |= VMX_EFLAGS_RESERVED_1; /* Bit 1 MB1. */
3687
3688 /*
3689 * If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so we can restore them on VM-exit.
3690 * Modify the real-mode guest's eflags so that VT-x can run the real-mode guest code under Virtual 8086 mode.
3691 */
3692 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
3693 {
3694 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
3695 Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
3696 pVCpu->hm.s.vmx.RealMode.Eflags.u32 = Eflags.u32; /* Save the original eflags of the real-mode guest. */
3697 Eflags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */
3698 Eflags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */
3699 }
3700
3701 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_RFLAGS, Eflags.u32);
3702 AssertRCReturn(rc, rc);
3703
3704 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RFLAGS);
3705 Log4(("Load[%RU32]: VMX_VMCS_GUEST_RFLAGS=%#RX32\n", pVCpu->idCpu, Eflags.u32));
3706 }
3707 return rc;
3708}
3709
3710
3711/**
3712 * Loads the guest RIP, RSP and RFLAGS into the guest-state area in the VMCS.
3713 *
3714 * @returns VBox status code.
3715 * @param pVCpu Pointer to the VMCPU.
3716 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3717 * out-of-sync. Make sure to update the required fields
3718 * before using them.
3719 *
3720 * @remarks No-long-jump zone!!!
3721 */
3722DECLINLINE(int) hmR0VmxLoadGuestRipRspRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3723{
3724 int rc = hmR0VmxLoadGuestRip(pVCpu, pMixedCtx);
3725 AssertRCReturn(rc, rc);
3726 rc = hmR0VmxLoadGuestRsp(pVCpu, pMixedCtx);
3727 AssertRCReturn(rc, rc);
3728 rc = hmR0VmxLoadGuestRflags(pVCpu, pMixedCtx);
3729 AssertRCReturn(rc, rc);
3730 return rc;
3731}
3732
3733
3734/**
3735 * Loads the guest CR0 control register into the guest-state area in the VMCS.
3736 * CR0 is partially shared with the host and we have to consider the FPU bits.
3737 *
3738 * @returns VBox status code.
3739 * @param pVM Pointer to the VM.
3740 * @param pVCpu Pointer to the VMCPU.
3741 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3742 * out-of-sync. Make sure to update the required fields
3743 * before using them.
3744 *
3745 * @remarks No-long-jump zone!!!
3746 */
3747static int hmR0VmxLoadSharedCR0(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3748{
3749 /*
3750 * Guest CR0.
3751 * Guest FPU.
3752 */
3753 int rc = VINF_SUCCESS;
3754 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0))
3755 {
3756 Assert(!(pMixedCtx->cr0 >> 32));
3757 uint32_t u32GuestCR0 = pMixedCtx->cr0;
3758 PVM pVM = pVCpu->CTX_SUFF(pVM);
3759
3760 /* The guest's view (read access) of its CR0 is unblemished. */
3761 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, u32GuestCR0);
3762 AssertRCReturn(rc, rc);
3763 Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR0_READ_SHADOW=%#RX32\n", pVCpu->idCpu, u32GuestCR0));
3764
3765 /* Setup VT-x's view of the guest CR0. */
3766 /* Minimize VM-exits due to CR3 changes when we have NestedPaging. */
3767 if (pVM->hm.s.fNestedPaging)
3768 {
3769 if (CPUMIsGuestPagingEnabledEx(pMixedCtx))
3770 {
3771 /* The guest has paging enabled, let it access CR3 without causing a VM-exit if supported. */
3772 pVCpu->hm.s.vmx.u32ProcCtls &= ~( VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
3773 | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT);
3774 }
3775 else
3776 {
3777 /* The guest doesn't have paging enabled, make CR3 access cause a VM-exit to update our shadow. */
3778 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
3779 | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
3780 }
3781
3782 /* If we have unrestricted guest execution, we never have to intercept CR3 reads. */
3783 if (pVM->hm.s.vmx.fUnrestrictedGuest)
3784 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
3785
3786 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
3787 AssertRCReturn(rc, rc);
3788 }
3789 else
3790 u32GuestCR0 |= X86_CR0_WP; /* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */
3791
3792 /*
3793 * Guest FPU bits.
3794 * Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be set on the first
3795 * CPUs to support VT-x and no mention of with regards to UX in VM-entry checks.
3796 */
3797 u32GuestCR0 |= X86_CR0_NE;
3798 bool fInterceptNM = false;
3799 if (CPUMIsGuestFPUStateActive(pVCpu))
3800 {
3801 fInterceptNM = false; /* Guest FPU active, no need to VM-exit on #NM. */
3802 /* The guest should still get #NM exceptions when it expects it to, so we should not clear TS & MP bits here.
3803 We're only concerned about -us- not intercepting #NMs when the guest-FPU is active. Not the guest itself! */
3804 }
3805 else
3806 {
3807 fInterceptNM = true; /* Guest FPU inactive, VM-exit on #NM for lazy FPU loading. */
3808 u32GuestCR0 |= X86_CR0_TS /* Guest can task switch quickly and do lazy FPU syncing. */
3809 | X86_CR0_MP; /* FWAIT/WAIT should not ignore CR0.TS and should generate #NM. */
3810 }
3811
3812 /* Catch floating point exceptions if we need to report them to the guest in a different way. */
3813 bool fInterceptMF = false;
3814 if (!(pMixedCtx->cr0 & X86_CR0_NE))
3815 fInterceptMF = true;
3816
3817 /* Finally, intercept all exceptions as we cannot directly inject them in real-mode, see hmR0VmxInjectEventVmcs(). */
3818 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
3819 {
3820 Assert(PDMVmmDevHeapIsEnabled(pVM));
3821 Assert(pVM->hm.s.vmx.pRealModeTSS);
3822 pVCpu->hm.s.vmx.u32XcptBitmap |= HMVMX_REAL_MODE_XCPT_MASK;
3823 fInterceptNM = true;
3824 fInterceptMF = true;
3825 }
3826 else
3827 {
3828 /* For now, cleared here as mode-switches can happen outside HM/VT-x. See @bugref{7626} comment #11. */
3829 pVCpu->hm.s.vmx.u32XcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK;
3830 }
3831 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
3832
3833 if (fInterceptNM)
3834 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_NM);
3835 else
3836 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_NM);
3837
3838 if (fInterceptMF)
3839 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_MF);
3840 else
3841 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_MF);
3842
3843 /* Additional intercepts for debugging, define these yourself explicitly. */
3844#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
3845 pVCpu->hm.s.vmx.u32XcptBitmap |= 0
3846 | RT_BIT(X86_XCPT_BP)
3847 | RT_BIT(X86_XCPT_DB)
3848 | RT_BIT(X86_XCPT_DE)
3849 | RT_BIT(X86_XCPT_NM)
3850 | RT_BIT(X86_XCPT_TS)
3851 | RT_BIT(X86_XCPT_UD)
3852 | RT_BIT(X86_XCPT_NP)
3853 | RT_BIT(X86_XCPT_SS)
3854 | RT_BIT(X86_XCPT_GP)
3855 | RT_BIT(X86_XCPT_PF)
3856 | RT_BIT(X86_XCPT_MF)
3857 ;
3858#elif defined(HMVMX_ALWAYS_TRAP_PF)
3859 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_PF);
3860#endif
3861
3862 Assert(pVM->hm.s.fNestedPaging || (pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT(X86_XCPT_PF)));
3863
3864 /* Set/clear the CR0 specific bits along with their exceptions (PE, PG, CD, NW). */
3865 uint32_t uSetCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
3866 uint32_t uZapCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
3867 if (pVM->hm.s.vmx.fUnrestrictedGuest) /* Exceptions for unrestricted-guests for fixed CR0 bits (PE, PG). */
3868 uSetCR0 &= ~(X86_CR0_PE | X86_CR0_PG);
3869 else
3870 Assert((uSetCR0 & (X86_CR0_PE | X86_CR0_PG)) == (X86_CR0_PE | X86_CR0_PG));
3871
3872 u32GuestCR0 |= uSetCR0;
3873 u32GuestCR0 &= uZapCR0;
3874 u32GuestCR0 &= ~(X86_CR0_CD | X86_CR0_NW); /* Always enable caching. */
3875
3876 /* Write VT-x's view of the guest CR0 into the VMCS. */
3877 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR0, u32GuestCR0);
3878 AssertRCReturn(rc, rc);
3879 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR0=%#RX32 (uSetCR0=%#RX32 uZapCR0=%#RX32)\n", pVCpu->idCpu, u32GuestCR0, uSetCR0,
3880 uZapCR0));
3881
3882 /*
3883 * CR0 is shared between host and guest along with a CR0 read shadow. Therefore, certain bits must not be changed
3884 * by the guest because VT-x ignores saving/restoring them (namely CD, ET, NW) and for certain other bits
3885 * we want to be notified immediately of guest CR0 changes (e.g. PG to update our shadow page tables).
3886 */
3887 uint32_t u32CR0Mask = 0;
3888 u32CR0Mask = X86_CR0_PE
3889 | X86_CR0_NE
3890 | X86_CR0_WP
3891 | X86_CR0_PG
3892 | X86_CR0_ET /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.ET */
3893 | X86_CR0_CD /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.CD */
3894 | X86_CR0_NW; /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.NW */
3895
3896 /** @todo Avoid intercepting CR0.PE with unrestricted guests. Fix PGM
3897 * enmGuestMode to be in-sync with the current mode. See @bugref{6398}
3898 * and @bugref{6944}. */
3899#if 0
3900 if (pVM->hm.s.vmx.fUnrestrictedGuest)
3901 u32CR0Mask &= ~X86_CR0_PE;
3902#endif
3903 if (pVM->hm.s.fNestedPaging)
3904 u32CR0Mask &= ~X86_CR0_WP;
3905
3906 /* If the guest FPU state is active, don't need to VM-exit on writes to FPU related bits in CR0. */
3907 if (fInterceptNM)
3908 {
3909 u32CR0Mask |= X86_CR0_TS
3910 | X86_CR0_MP;
3911 }
3912
3913 /* Write the CR0 mask into the VMCS and update the VCPU's copy of the current CR0 mask. */
3914 pVCpu->hm.s.vmx.u32CR0Mask = u32CR0Mask;
3915 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_MASK, u32CR0Mask);
3916 AssertRCReturn(rc, rc);
3917 Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR0_MASK=%#RX32\n", pVCpu->idCpu, u32CR0Mask));
3918
3919 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR0);
3920 }
3921 return rc;
3922}
3923
3924
3925/**
3926 * Loads the guest control registers (CR3, CR4) into the guest-state area
3927 * in the VMCS.
3928 *
3929 * @returns VBox status code.
3930 * @param pVM Pointer to the VM.
3931 * @param pVCpu Pointer to the VMCPU.
3932 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3933 * out-of-sync. Make sure to update the required fields
3934 * before using them.
3935 *
3936 * @remarks No-long-jump zone!!!
3937 */
3938static int hmR0VmxLoadGuestCR3AndCR4(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3939{
3940 int rc = VINF_SUCCESS;
3941 PVM pVM = pVCpu->CTX_SUFF(pVM);
3942
3943 /*
3944 * Guest CR2.
3945 * It's always loaded in the assembler code. Nothing to do here.
3946 */
3947
3948 /*
3949 * Guest CR3.
3950 */
3951 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR3))
3952 {
3953 RTGCPHYS GCPhysGuestCR3 = NIL_RTGCPHYS;
3954 if (pVM->hm.s.fNestedPaging)
3955 {
3956 pVCpu->hm.s.vmx.HCPhysEPTP = PGMGetHyperCR3(pVCpu);
3957
3958 /* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */
3959 Assert(pVCpu->hm.s.vmx.HCPhysEPTP);
3960 Assert(!(pVCpu->hm.s.vmx.HCPhysEPTP & UINT64_C(0xfff0000000000000)));
3961 Assert(!(pVCpu->hm.s.vmx.HCPhysEPTP & 0xfff));
3962
3963 /* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */
3964 pVCpu->hm.s.vmx.HCPhysEPTP |= VMX_EPT_MEMTYPE_WB
3965 | (VMX_EPT_PAGE_WALK_LENGTH_DEFAULT << VMX_EPT_PAGE_WALK_LENGTH_SHIFT);
3966
3967 /* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */
3968 AssertMsg( ((pVCpu->hm.s.vmx.HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */
3969 && ((pVCpu->hm.s.vmx.HCPhysEPTP >> 6) & 0x3f) == 0, /* Bits 6:11 MBZ. */
3970 ("EPTP %#RX64\n", pVCpu->hm.s.vmx.HCPhysEPTP));
3971
3972 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVCpu->hm.s.vmx.HCPhysEPTP);
3973 AssertRCReturn(rc, rc);
3974 Log4(("Load[%RU32]: VMX_VMCS64_CTRL_EPTP_FULL=%#RX64\n", pVCpu->idCpu, pVCpu->hm.s.vmx.HCPhysEPTP));
3975
3976 if ( pVM->hm.s.vmx.fUnrestrictedGuest
3977 || CPUMIsGuestPagingEnabledEx(pMixedCtx))
3978 {
3979 /* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */
3980 if (CPUMIsGuestInPAEModeEx(pMixedCtx))
3981 {
3982 rc = PGMGstGetPaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]); AssertRCReturn(rc, rc);
3983 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pVCpu->hm.s.aPdpes[0].u); AssertRCReturn(rc, rc);
3984 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pVCpu->hm.s.aPdpes[1].u); AssertRCReturn(rc, rc);
3985 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pVCpu->hm.s.aPdpes[2].u); AssertRCReturn(rc, rc);
3986 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pVCpu->hm.s.aPdpes[3].u); AssertRCReturn(rc, rc);
3987 }
3988
3989 /* The guest's view of its CR3 is unblemished with Nested Paging when the guest is using paging or we
3990 have Unrestricted Execution to handle the guest when it's not using paging. */
3991 GCPhysGuestCR3 = pMixedCtx->cr3;
3992 }
3993 else
3994 {
3995 /*
3996 * The guest is not using paging, but the CPU (VT-x) has to. While the guest thinks it accesses physical memory
3997 * directly, we use our identity-mapped page table to map guest-linear to guest-physical addresses.
3998 * EPT takes care of translating it to host-physical addresses.
3999 */
4000 RTGCPHYS GCPhys;
4001 Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable);
4002 Assert(PDMVmmDevHeapIsEnabled(pVM));
4003
4004 /* We obtain it here every time as the guest could have relocated this PCI region. */
4005 rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
4006 AssertRCReturn(rc, rc);
4007
4008 GCPhysGuestCR3 = GCPhys;
4009 }
4010
4011 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR3=%#RGv (GstN)\n", pVCpu->idCpu, GCPhysGuestCR3));
4012 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_CR3, GCPhysGuestCR3);
4013 }
4014 else
4015 {
4016 /* Non-nested paging case, just use the hypervisor's CR3. */
4017 RTHCPHYS HCPhysGuestCR3 = PGMGetHyperCR3(pVCpu);
4018
4019 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR3=%#RHv (HstN)\n", pVCpu->idCpu, HCPhysGuestCR3));
4020 rc = VMXWriteVmcsHstN(VMX_VMCS_GUEST_CR3, HCPhysGuestCR3);
4021 }
4022 AssertRCReturn(rc, rc);
4023
4024 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR3);
4025 }
4026
4027 /*
4028 * Guest CR4.
4029 * ASSUMES this is done everytime we get in from ring-3! (XCR0)
4030 */
4031 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR4))
4032 {
4033 Assert(!(pMixedCtx->cr4 >> 32));
4034 uint32_t u32GuestCR4 = pMixedCtx->cr4;
4035
4036 /* The guest's view of its CR4 is unblemished. */
4037 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, u32GuestCR4);
4038 AssertRCReturn(rc, rc);
4039 Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR4_READ_SHADOW=%#RX32\n", pVCpu->idCpu, u32GuestCR4));
4040
4041 /* Setup VT-x's view of the guest CR4. */
4042 /*
4043 * If we're emulating real-mode using virtual-8086 mode, we want to redirect software interrupts to the 8086 program
4044 * interrupt handler. Clear the VME bit (the interrupt redirection bitmap is already all 0, see hmR3InitFinalizeR0())
4045 * See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode".
4046 */
4047 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4048 {
4049 Assert(pVM->hm.s.vmx.pRealModeTSS);
4050 Assert(PDMVmmDevHeapIsEnabled(pVM));
4051 u32GuestCR4 &= ~X86_CR4_VME;
4052 }
4053
4054 if (pVM->hm.s.fNestedPaging)
4055 {
4056 if ( !CPUMIsGuestPagingEnabledEx(pMixedCtx)
4057 && !pVM->hm.s.vmx.fUnrestrictedGuest)
4058 {
4059 /* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */
4060 u32GuestCR4 |= X86_CR4_PSE;
4061 /* Our identity mapping is a 32-bit page directory. */
4062 u32GuestCR4 &= ~X86_CR4_PAE;
4063 }
4064 /* else use guest CR4.*/
4065 }
4066 else
4067 {
4068 /*
4069 * The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host
4070 * paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables.
4071 */
4072 switch (pVCpu->hm.s.enmShadowMode)
4073 {
4074 case PGMMODE_REAL: /* Real-mode. */
4075 case PGMMODE_PROTECTED: /* Protected mode without paging. */
4076 case PGMMODE_32_BIT: /* 32-bit paging. */
4077 {
4078 u32GuestCR4 &= ~X86_CR4_PAE;
4079 break;
4080 }
4081
4082 case PGMMODE_PAE: /* PAE paging. */
4083 case PGMMODE_PAE_NX: /* PAE paging with NX. */
4084 {
4085 u32GuestCR4 |= X86_CR4_PAE;
4086 break;
4087 }
4088
4089 case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
4090 case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
4091#ifdef VBOX_ENABLE_64_BITS_GUESTS
4092 break;
4093#endif
4094 default:
4095 AssertFailed();
4096 return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
4097 }
4098 }
4099
4100 /* We need to set and clear the CR4 specific bits here (mainly the X86_CR4_VMXE bit). */
4101 uint64_t uSetCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
4102 uint64_t uZapCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
4103 u32GuestCR4 |= uSetCR4;
4104 u32GuestCR4 &= uZapCR4;
4105
4106 /* Write VT-x's view of the guest CR4 into the VMCS. */
4107 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR4=%#RX32 (Set=%#RX32 Zap=%#RX32)\n", pVCpu->idCpu, u32GuestCR4, uSetCR4, uZapCR4));
4108 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR4, u32GuestCR4);
4109 AssertRCReturn(rc, rc);
4110
4111 /* Setup CR4 mask. CR4 flags owned by the host, if the guest attempts to change them, that would cause a VM-exit. */
4112 uint32_t u32CR4Mask = X86_CR4_VME
4113 | X86_CR4_PAE
4114 | X86_CR4_PGE
4115 | X86_CR4_PSE
4116 | X86_CR4_VMXE;
4117 if (pVM->cpum.ro.HostFeatures.fXSaveRstor)
4118 u32CR4Mask |= X86_CR4_OSXSAVE;
4119 pVCpu->hm.s.vmx.u32CR4Mask = u32CR4Mask;
4120 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_MASK, u32CR4Mask);
4121 AssertRCReturn(rc, rc);
4122
4123 /* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
4124 pVCpu->hm.s.fLoadSaveGuestXcr0 = (pMixedCtx->cr4 & X86_CR4_OSXSAVE) && pMixedCtx->aXcr[0] != ASMGetXcr0();
4125
4126 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR4);
4127 }
4128 return rc;
4129}
4130
4131
4132/**
4133 * Loads the guest debug registers into the guest-state area in the VMCS.
4134 * This also sets up whether #DB and MOV DRx accesses cause VM-exits.
4135 *
4136 * The guest debug bits are partially shared with the host (e.g. DR6, DR0-3).
4137 *
4138 * @returns VBox status code.
4139 * @param pVCpu Pointer to the VMCPU.
4140 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4141 * out-of-sync. Make sure to update the required fields
4142 * before using them.
4143 *
4144 * @remarks No-long-jump zone!!!
4145 */
4146static int hmR0VmxLoadSharedDebugState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4147{
4148 if (!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG))
4149 return VINF_SUCCESS;
4150
4151#ifdef VBOX_STRICT
4152 /* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */
4153 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
4154 {
4155 /* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */
4156 Assert((pMixedCtx->dr[7] & (X86_DR7_MBZ_MASK | X86_DR7_RAZ_MASK)) == 0); /* Bits 63:32, 15, 14, 12, 11 are reserved. */
4157 Assert((pMixedCtx->dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK); /* Bit 10 is reserved (RA1). */
4158 }
4159#endif
4160
4161 int rc;
4162 PVM pVM = pVCpu->CTX_SUFF(pVM);
4163 bool fInterceptDB = false;
4164 bool fInterceptMovDRx = false;
4165 if ( pVCpu->hm.s.fSingleInstruction
4166 || DBGFIsStepping(pVCpu))
4167 {
4168 /* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */
4169 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG)
4170 {
4171 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG;
4172 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
4173 AssertRCReturn(rc, rc);
4174 Assert(fInterceptDB == false);
4175 }
4176 else
4177 {
4178 pMixedCtx->eflags.u32 |= X86_EFL_TF;
4179 pVCpu->hm.s.fClearTrapFlag = true;
4180 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
4181 fInterceptDB = true;
4182 }
4183 }
4184
4185 if ( fInterceptDB
4186 || (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
4187 {
4188 /*
4189 * Use the combined guest and host DRx values found in the hypervisor
4190 * register set because the debugger has breakpoints active or someone
4191 * is single stepping on the host side without a monitor trap flag.
4192 *
4193 * Note! DBGF expects a clean DR6 state before executing guest code.
4194 */
4195#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4196 if ( CPUMIsGuestInLongModeEx(pMixedCtx)
4197 && !CPUMIsHyperDebugStateActivePending(pVCpu))
4198 {
4199 CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
4200 Assert(CPUMIsHyperDebugStateActivePending(pVCpu));
4201 Assert(!CPUMIsGuestDebugStateActivePending(pVCpu));
4202 }
4203 else
4204#endif
4205 if (!CPUMIsHyperDebugStateActive(pVCpu))
4206 {
4207 CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
4208 Assert(CPUMIsHyperDebugStateActive(pVCpu));
4209 Assert(!CPUMIsGuestDebugStateActive(pVCpu));
4210 }
4211
4212 /* Update DR7. (The other DRx values are handled by CPUM one way or the other.) */
4213 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)CPUMGetHyperDR7(pVCpu));
4214 AssertRCReturn(rc, rc);
4215
4216 pVCpu->hm.s.fUsingHyperDR7 = true;
4217 fInterceptDB = true;
4218 fInterceptMovDRx = true;
4219 }
4220 else
4221 {
4222 /*
4223 * If the guest has enabled debug registers, we need to load them prior to
4224 * executing guest code so they'll trigger at the right time.
4225 */
4226 if (pMixedCtx->dr[7] & (X86_DR7_ENABLED_MASK | X86_DR7_GD)) /** @todo Why GD? */
4227 {
4228#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4229 if ( CPUMIsGuestInLongModeEx(pMixedCtx)
4230 && !CPUMIsGuestDebugStateActivePending(pVCpu))
4231 {
4232 CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
4233 Assert(CPUMIsGuestDebugStateActivePending(pVCpu));
4234 Assert(!CPUMIsHyperDebugStateActivePending(pVCpu));
4235 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
4236 }
4237 else
4238#endif
4239 if (!CPUMIsGuestDebugStateActive(pVCpu))
4240 {
4241 CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
4242 Assert(CPUMIsGuestDebugStateActive(pVCpu));
4243 Assert(!CPUMIsHyperDebugStateActive(pVCpu));
4244 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
4245 }
4246 Assert(!fInterceptDB);
4247 Assert(!fInterceptMovDRx);
4248 }
4249 /*
4250 * If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we
4251 * must intercept #DB in order to maintain a correct DR6 guest value.
4252 */
4253#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4254 else if ( !CPUMIsGuestDebugStateActivePending(pVCpu)
4255 && !CPUMIsGuestDebugStateActive(pVCpu))
4256#else
4257 else if (!CPUMIsGuestDebugStateActive(pVCpu))
4258#endif
4259 {
4260 fInterceptMovDRx = true;
4261 fInterceptDB = true;
4262 }
4263
4264 /* Update guest DR7. */
4265 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, pMixedCtx->dr[7]);
4266 AssertRCReturn(rc, rc);
4267
4268 pVCpu->hm.s.fUsingHyperDR7 = false;
4269 }
4270
4271 /*
4272 * Update the exception bitmap regarding intercepting #DB generated by the guest.
4273 */
4274 if ( fInterceptDB
4275 || pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4276 {
4277 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_DB);
4278 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
4279 }
4280 else
4281 {
4282#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
4283 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_DB);
4284 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
4285#endif
4286 }
4287
4288 /*
4289 * Update the processor-based VM-execution controls regarding intercepting MOV DRx instructions.
4290 */
4291 if (fInterceptMovDRx)
4292 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
4293 else
4294 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
4295 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
4296 AssertRCReturn(rc, rc);
4297
4298 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_DEBUG);
4299 return VINF_SUCCESS;
4300}
4301
4302
4303#ifdef VBOX_STRICT
4304/**
4305 * Strict function to validate segment registers.
4306 *
4307 * @remarks ASSUMES CR0 is up to date.
4308 */
4309static void hmR0VmxValidateSegmentRegs(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
4310{
4311 /* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers". */
4312 /* NOTE: The reason we check for attribute value 0 and not just the unusable bit here is because hmR0VmxWriteSegmentReg()
4313 * only updates the VMCS' copy of the value with the unusable bit and doesn't change the guest-context value. */
4314 if ( !pVM->hm.s.vmx.fUnrestrictedGuest
4315 && ( !CPUMIsGuestInRealModeEx(pCtx)
4316 && !CPUMIsGuestInV86ModeEx(pCtx)))
4317 {
4318 /* Protected mode checks */
4319 /* CS */
4320 Assert(pCtx->cs.Attr.n.u1Present);
4321 Assert(!(pCtx->cs.Attr.u & 0xf00));
4322 Assert(!(pCtx->cs.Attr.u & 0xfffe0000));
4323 Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff
4324 || !(pCtx->cs.Attr.n.u1Granularity));
4325 Assert( !(pCtx->cs.u32Limit & 0xfff00000)
4326 || (pCtx->cs.Attr.n.u1Granularity));
4327 /* CS cannot be loaded with NULL in protected mode. */
4328 Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS?!? */
4329 if (pCtx->cs.Attr.n.u4Type == 9 || pCtx->cs.Attr.n.u4Type == 11)
4330 Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl);
4331 else if (pCtx->cs.Attr.n.u4Type == 13 || pCtx->cs.Attr.n.u4Type == 15)
4332 Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl);
4333 else
4334 AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl));
4335 /* SS */
4336 Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
4337 Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL));
4338 if ( !(pCtx->cr0 & X86_CR0_PE)
4339 || pCtx->cs.Attr.n.u4Type == 3)
4340 {
4341 Assert(!pCtx->ss.Attr.n.u2Dpl);
4342 }
4343 if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
4344 {
4345 Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
4346 Assert(pCtx->ss.Attr.n.u4Type == 3 || pCtx->ss.Attr.n.u4Type == 7);
4347 Assert(pCtx->ss.Attr.n.u1Present);
4348 Assert(!(pCtx->ss.Attr.u & 0xf00));
4349 Assert(!(pCtx->ss.Attr.u & 0xfffe0000));
4350 Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff
4351 || !(pCtx->ss.Attr.n.u1Granularity));
4352 Assert( !(pCtx->ss.u32Limit & 0xfff00000)
4353 || (pCtx->ss.Attr.n.u1Granularity));
4354 }
4355 /* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxWriteSegmentReg(). */
4356 if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
4357 {
4358 Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4359 Assert(pCtx->ds.Attr.n.u1Present);
4360 Assert(pCtx->ds.Attr.n.u4Type > 11 || pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL));
4361 Assert(!(pCtx->ds.Attr.u & 0xf00));
4362 Assert(!(pCtx->ds.Attr.u & 0xfffe0000));
4363 Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff
4364 || !(pCtx->ds.Attr.n.u1Granularity));
4365 Assert( !(pCtx->ds.u32Limit & 0xfff00000)
4366 || (pCtx->ds.Attr.n.u1Granularity));
4367 Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4368 || (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ));
4369 }
4370 if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
4371 {
4372 Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4373 Assert(pCtx->es.Attr.n.u1Present);
4374 Assert(pCtx->es.Attr.n.u4Type > 11 || pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL));
4375 Assert(!(pCtx->es.Attr.u & 0xf00));
4376 Assert(!(pCtx->es.Attr.u & 0xfffe0000));
4377 Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff
4378 || !(pCtx->es.Attr.n.u1Granularity));
4379 Assert( !(pCtx->es.u32Limit & 0xfff00000)
4380 || (pCtx->es.Attr.n.u1Granularity));
4381 Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4382 || (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ));
4383 }
4384 if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
4385 {
4386 Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4387 Assert(pCtx->fs.Attr.n.u1Present);
4388 Assert(pCtx->fs.Attr.n.u4Type > 11 || pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL));
4389 Assert(!(pCtx->fs.Attr.u & 0xf00));
4390 Assert(!(pCtx->fs.Attr.u & 0xfffe0000));
4391 Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff
4392 || !(pCtx->fs.Attr.n.u1Granularity));
4393 Assert( !(pCtx->fs.u32Limit & 0xfff00000)
4394 || (pCtx->fs.Attr.n.u1Granularity));
4395 Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4396 || (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ));
4397 }
4398 if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
4399 {
4400 Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4401 Assert(pCtx->gs.Attr.n.u1Present);
4402 Assert(pCtx->gs.Attr.n.u4Type > 11 || pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL));
4403 Assert(!(pCtx->gs.Attr.u & 0xf00));
4404 Assert(!(pCtx->gs.Attr.u & 0xfffe0000));
4405 Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff
4406 || !(pCtx->gs.Attr.n.u1Granularity));
4407 Assert( !(pCtx->gs.u32Limit & 0xfff00000)
4408 || (pCtx->gs.Attr.n.u1Granularity));
4409 Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4410 || (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ));
4411 }
4412 /* 64-bit capable CPUs. */
4413# if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4414 if (HMVMX_IS_64BIT_HOST_MODE())
4415 {
4416 Assert(!(pCtx->cs.u64Base >> 32));
4417 Assert(!pCtx->ss.Attr.u || !(pCtx->ss.u64Base >> 32));
4418 Assert(!pCtx->ds.Attr.u || !(pCtx->ds.u64Base >> 32));
4419 Assert(!pCtx->es.Attr.u || !(pCtx->es.u64Base >> 32));
4420 }
4421# endif
4422 }
4423 else if ( CPUMIsGuestInV86ModeEx(pCtx)
4424 || ( CPUMIsGuestInRealModeEx(pCtx)
4425 && !pVM->hm.s.vmx.fUnrestrictedGuest))
4426 {
4427 /* Real and v86 mode checks. */
4428 /* hmR0VmxWriteSegmentReg() writes the modified in VMCS. We want what we're feeding to VT-x. */
4429 uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
4430 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4431 {
4432 u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3; u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3;
4433 }
4434 else
4435 {
4436 u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u;
4437 u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
4438 }
4439
4440 /* CS */
4441 AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel));
4442 Assert(pCtx->cs.u32Limit == 0xffff);
4443 Assert(u32CSAttr == 0xf3);
4444 /* SS */
4445 Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4);
4446 Assert(pCtx->ss.u32Limit == 0xffff);
4447 Assert(u32SSAttr == 0xf3);
4448 /* DS */
4449 Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4);
4450 Assert(pCtx->ds.u32Limit == 0xffff);
4451 Assert(u32DSAttr == 0xf3);
4452 /* ES */
4453 Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4);
4454 Assert(pCtx->es.u32Limit == 0xffff);
4455 Assert(u32ESAttr == 0xf3);
4456 /* FS */
4457 Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4);
4458 Assert(pCtx->fs.u32Limit == 0xffff);
4459 Assert(u32FSAttr == 0xf3);
4460 /* GS */
4461 Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4);
4462 Assert(pCtx->gs.u32Limit == 0xffff);
4463 Assert(u32GSAttr == 0xf3);
4464 /* 64-bit capable CPUs. */
4465# if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4466 if (HMVMX_IS_64BIT_HOST_MODE())
4467 {
4468 Assert(!(pCtx->cs.u64Base >> 32));
4469 Assert(!u32SSAttr || !(pCtx->ss.u64Base >> 32));
4470 Assert(!u32DSAttr || !(pCtx->ds.u64Base >> 32));
4471 Assert(!u32ESAttr || !(pCtx->es.u64Base >> 32));
4472 }
4473# endif
4474 }
4475}
4476#endif /* VBOX_STRICT */
4477
4478
4479/**
4480 * Writes a guest segment register into the guest-state area in the VMCS.
4481 *
4482 * @returns VBox status code.
4483 * @param pVCpu Pointer to the VMCPU.
4484 * @param idxSel Index of the selector in the VMCS.
4485 * @param idxLimit Index of the segment limit in the VMCS.
4486 * @param idxBase Index of the segment base in the VMCS.
4487 * @param idxAccess Index of the access rights of the segment in the VMCS.
4488 * @param pSelReg Pointer to the segment selector.
4489 *
4490 * @remarks No-long-jump zone!!!
4491 */
4492static int hmR0VmxWriteSegmentReg(PVMCPU pVCpu, uint32_t idxSel, uint32_t idxLimit, uint32_t idxBase,
4493 uint32_t idxAccess, PCPUMSELREG pSelReg)
4494{
4495 int rc = VMXWriteVmcs32(idxSel, pSelReg->Sel); /* 16-bit guest selector field. */
4496 AssertRCReturn(rc, rc);
4497 rc = VMXWriteVmcs32(idxLimit, pSelReg->u32Limit); /* 32-bit guest segment limit field. */
4498 AssertRCReturn(rc, rc);
4499 rc = VMXWriteVmcsGstN(idxBase, pSelReg->u64Base); /* Natural width guest segment base field.*/
4500 AssertRCReturn(rc, rc);
4501
4502 uint32_t u32Access = pSelReg->Attr.u;
4503 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4504 {
4505 /* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */
4506 u32Access = 0xf3;
4507 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
4508 Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
4509 }
4510 else
4511 {
4512 /*
4513 * The way to differentiate between whether this is really a null selector or was just a selector loaded with 0 in
4514 * real-mode is using the segment attributes. A selector loaded in real-mode with the value 0 is valid and usable in
4515 * protected-mode and we should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures NULL selectors
4516 * loaded in protected-mode have their attribute as 0.
4517 */
4518 if (!u32Access)
4519 u32Access = X86DESCATTR_UNUSABLE;
4520 }
4521
4522 /* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */
4523 AssertMsg((u32Access & X86DESCATTR_UNUSABLE) || (u32Access & X86_SEL_TYPE_ACCESSED),
4524 ("Access bit not set for usable segment. idx=%#x sel=%#x attr %#x\n", idxBase, pSelReg, pSelReg->Attr.u));
4525
4526 rc = VMXWriteVmcs32(idxAccess, u32Access); /* 32-bit guest segment access-rights field. */
4527 AssertRCReturn(rc, rc);
4528 return rc;
4529}
4530
4531
4532/**
4533 * Loads the guest segment registers, GDTR, IDTR, LDTR, (TR, FS and GS bases)
4534 * into the guest-state area in the VMCS.
4535 *
4536 * @returns VBox status code.
4537 * @param pVM Pointer to the VM.
4538 * @param pVCPU Pointer to the VMCPU.
4539 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4540 * out-of-sync. Make sure to update the required fields
4541 * before using them.
4542 *
4543 * @remarks ASSUMES pMixedCtx->cr0 is up to date (strict builds validation).
4544 * @remarks No-long-jump zone!!!
4545 */
4546static int hmR0VmxLoadGuestSegmentRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4547{
4548 int rc = VERR_INTERNAL_ERROR_5;
4549 PVM pVM = pVCpu->CTX_SUFF(pVM);
4550
4551 /*
4552 * Guest Segment registers: CS, SS, DS, ES, FS, GS.
4553 */
4554 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS))
4555 {
4556 /* Save the segment attributes for real-on-v86 mode hack, so we can restore them on VM-exit. */
4557 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4558 {
4559 pVCpu->hm.s.vmx.RealMode.AttrCS.u = pMixedCtx->cs.Attr.u;
4560 pVCpu->hm.s.vmx.RealMode.AttrSS.u = pMixedCtx->ss.Attr.u;
4561 pVCpu->hm.s.vmx.RealMode.AttrDS.u = pMixedCtx->ds.Attr.u;
4562 pVCpu->hm.s.vmx.RealMode.AttrES.u = pMixedCtx->es.Attr.u;
4563 pVCpu->hm.s.vmx.RealMode.AttrFS.u = pMixedCtx->fs.Attr.u;
4564 pVCpu->hm.s.vmx.RealMode.AttrGS.u = pMixedCtx->gs.Attr.u;
4565 }
4566
4567#ifdef VBOX_WITH_REM
4568 if (!pVM->hm.s.vmx.fUnrestrictedGuest)
4569 {
4570 Assert(pVM->hm.s.vmx.pRealModeTSS);
4571 AssertCompile(PGMMODE_REAL < PGMMODE_PROTECTED);
4572 if ( pVCpu->hm.s.vmx.fWasInRealMode
4573 && PGMGetGuestMode(pVCpu) >= PGMMODE_PROTECTED)
4574 {
4575 /* Signal that the recompiler must flush its code-cache as the guest -may- rewrite code it will later execute
4576 in real-mode (e.g. OpenBSD 4.0) */
4577 REMFlushTBs(pVM);
4578 Log4(("Load[%RU32]: Switch to protected mode detected!\n", pVCpu->idCpu));
4579 pVCpu->hm.s.vmx.fWasInRealMode = false;
4580 }
4581 }
4582#endif
4583 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_CS, VMX_VMCS32_GUEST_CS_LIMIT, VMX_VMCS_GUEST_CS_BASE,
4584 VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS, &pMixedCtx->cs);
4585 AssertRCReturn(rc, rc);
4586 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_SS, VMX_VMCS32_GUEST_SS_LIMIT, VMX_VMCS_GUEST_SS_BASE,
4587 VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS, &pMixedCtx->ss);
4588 AssertRCReturn(rc, rc);
4589 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_DS, VMX_VMCS32_GUEST_DS_LIMIT, VMX_VMCS_GUEST_DS_BASE,
4590 VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS, &pMixedCtx->ds);
4591 AssertRCReturn(rc, rc);
4592 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_ES, VMX_VMCS32_GUEST_ES_LIMIT, VMX_VMCS_GUEST_ES_BASE,
4593 VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS, &pMixedCtx->es);
4594 AssertRCReturn(rc, rc);
4595 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_FS, VMX_VMCS32_GUEST_FS_LIMIT, VMX_VMCS_GUEST_FS_BASE,
4596 VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS, &pMixedCtx->fs);
4597 AssertRCReturn(rc, rc);
4598 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_GS, VMX_VMCS32_GUEST_GS_LIMIT, VMX_VMCS_GUEST_GS_BASE,
4599 VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS, &pMixedCtx->gs);
4600 AssertRCReturn(rc, rc);
4601
4602#ifdef VBOX_STRICT
4603 /* Validate. */
4604 hmR0VmxValidateSegmentRegs(pVM, pVCpu, pMixedCtx);
4605#endif
4606
4607 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS);
4608 Log4(("Load[%RU32]: CS=%#RX16 Base=%#RX64 Limit=%#RX32 Attr=%#RX32\n", pVCpu->idCpu, pMixedCtx->cs.Sel,
4609 pMixedCtx->cs.u64Base, pMixedCtx->cs.u32Limit, pMixedCtx->cs.Attr.u));
4610 }
4611
4612 /*
4613 * Guest TR.
4614 */
4615 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_TR))
4616 {
4617 /*
4618 * Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is achieved
4619 * using the interrupt redirection bitmap (all bits cleared to let the guest handle INT-n's) in the TSS.
4620 * See hmR3InitFinalizeR0() to see how pRealModeTSS is setup.
4621 */
4622 uint16_t u16Sel = 0;
4623 uint32_t u32Limit = 0;
4624 uint64_t u64Base = 0;
4625 uint32_t u32AccessRights = 0;
4626
4627 if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4628 {
4629 u16Sel = pMixedCtx->tr.Sel;
4630 u32Limit = pMixedCtx->tr.u32Limit;
4631 u64Base = pMixedCtx->tr.u64Base;
4632 u32AccessRights = pMixedCtx->tr.Attr.u;
4633 }
4634 else
4635 {
4636 Assert(pVM->hm.s.vmx.pRealModeTSS);
4637 Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMR3CanExecuteGuest() -XXX- what about inner loop changes? */
4638
4639 /* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */
4640 RTGCPHYS GCPhys;
4641 rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
4642 AssertRCReturn(rc, rc);
4643
4644 X86DESCATTR DescAttr;
4645 DescAttr.u = 0;
4646 DescAttr.n.u1Present = 1;
4647 DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
4648
4649 u16Sel = 0;
4650 u32Limit = HM_VTX_TSS_SIZE;
4651 u64Base = GCPhys; /* in real-mode phys = virt. */
4652 u32AccessRights = DescAttr.u;
4653 }
4654
4655 /* Validate. */
4656 Assert(!(u16Sel & RT_BIT(2)));
4657 AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY
4658 || (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights));
4659 AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights));
4660 Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/
4661 Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/
4662 Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */
4663 Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */
4664 Assert( (u32Limit & 0xfff) == 0xfff
4665 || !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */
4666 Assert( !(pMixedCtx->tr.u32Limit & 0xfff00000)
4667 || (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */
4668
4669 rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_TR, u16Sel); AssertRCReturn(rc, rc);
4670 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit); AssertRCReturn(rc, rc);
4671 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_TR_BASE, u64Base); AssertRCReturn(rc, rc);
4672 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights); AssertRCReturn(rc, rc);
4673
4674 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_TR);
4675 Log4(("Load[%RU32]: VMX_VMCS_GUEST_TR_BASE=%#RX64\n", pVCpu->idCpu, u64Base));
4676 }
4677
4678 /*
4679 * Guest GDTR.
4680 */
4681 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_GDTR))
4682 {
4683 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pMixedCtx->gdtr.cbGdt); AssertRCReturn(rc, rc);
4684 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, pMixedCtx->gdtr.pGdt); AssertRCReturn(rc, rc);
4685
4686 /* Validate. */
4687 Assert(!(pMixedCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */
4688
4689 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_GDTR);
4690 Log4(("Load[%RU32]: VMX_VMCS_GUEST_GDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->gdtr.pGdt));
4691 }
4692
4693 /*
4694 * Guest LDTR.
4695 */
4696 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LDTR))
4697 {
4698 /* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */
4699 uint32_t u32Access = 0;
4700 if (!pMixedCtx->ldtr.Attr.u)
4701 u32Access = X86DESCATTR_UNUSABLE;
4702 else
4703 u32Access = pMixedCtx->ldtr.Attr.u;
4704
4705 rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_LDTR, pMixedCtx->ldtr.Sel); AssertRCReturn(rc, rc);
4706 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pMixedCtx->ldtr.u32Limit); AssertRCReturn(rc, rc);
4707 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_LDTR_BASE, pMixedCtx->ldtr.u64Base); AssertRCReturn(rc, rc);
4708 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access); AssertRCReturn(rc, rc);
4709
4710 /* Validate. */
4711 if (!(u32Access & X86DESCATTR_UNUSABLE))
4712 {
4713 Assert(!(pMixedCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */
4714 Assert(pMixedCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */
4715 Assert(!pMixedCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */
4716 Assert(pMixedCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */
4717 Assert(!pMixedCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */
4718 Assert(!(pMixedCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */
4719 Assert( (pMixedCtx->ldtr.u32Limit & 0xfff) == 0xfff
4720 || !pMixedCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */
4721 Assert( !(pMixedCtx->ldtr.u32Limit & 0xfff00000)
4722 || pMixedCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */
4723 }
4724
4725 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_LDTR);
4726 Log4(("Load[%RU32]: VMX_VMCS_GUEST_LDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->ldtr.u64Base));
4727 }
4728
4729 /*
4730 * Guest IDTR.
4731 */
4732 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_IDTR))
4733 {
4734 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pMixedCtx->idtr.cbIdt); AssertRCReturn(rc, rc);
4735 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, pMixedCtx->idtr.pIdt); AssertRCReturn(rc, rc);
4736
4737 /* Validate. */
4738 Assert(!(pMixedCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */
4739
4740 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_IDTR);
4741 Log4(("Load[%RU32]: VMX_VMCS_GUEST_IDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->idtr.pIdt));
4742 }
4743
4744 return VINF_SUCCESS;
4745}
4746
4747
4748/**
4749 * Loads certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store
4750 * areas.
4751 *
4752 * These MSRs will automatically be loaded to the host CPU on every successful
4753 * VM-entry and stored from the host CPU on every successful VM-exit. This also
4754 * creates/updates MSR slots for the host MSRs. The actual host MSR values are
4755 * -not- updated here for performance reasons. See hmR0VmxSaveHostMsrs().
4756 *
4757 * Also loads the sysenter MSRs into the guest-state area in the VMCS.
4758 *
4759 * @returns VBox status code.
4760 * @param pVCpu Pointer to the VMCPU.
4761 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4762 * out-of-sync. Make sure to update the required fields
4763 * before using them.
4764 *
4765 * @remarks No-long-jump zone!!!
4766 */
4767static int hmR0VmxLoadGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4768{
4769 AssertPtr(pVCpu);
4770 AssertPtr(pVCpu->hm.s.vmx.pvGuestMsr);
4771
4772 /*
4773 * MSRs that we use the auto-load/store MSR area in the VMCS.
4774 */
4775 PVM pVM = pVCpu->CTX_SUFF(pVM);
4776 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS))
4777 {
4778 /* For 64-bit hosts, we load/restore them lazily, see hmR0VmxLazyLoadGuestMsrs(). */
4779#if HC_ARCH_BITS == 32 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4780 if (pVM->hm.s.fAllow64BitGuests)
4781 {
4782 int rc = VINF_SUCCESS;
4783 rc |= hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_LSTAR, pMixedCtx->msrLSTAR, false, NULL);
4784 rc |= hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K6_STAR, pMixedCtx->msrSTAR, false, NULL);
4785 rc |= hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_SF_MASK, pMixedCtx->msrSFMASK, false, NULL);
4786 rc |= hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_KERNEL_GS_BASE, pMixedCtx->msrKERNELGSBASE, false, NULL);
4787 AssertRCReturn(rc, rc);
4788#ifdef DEBUG
4789 PVMXAUTOMSR pMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
4790 for (uint32_t i = 0; i < pVCpu->hm.s.vmx.cMsrs; i++, pMsr++)
4791 {
4792 Log4(("Load[%RU32]: MSR[%RU32]: u32Msr=%#RX32 u64Value=%#RX64\n", pVCpu->idCpu, i, pMsr->u32Msr,
4793 pMsr->u64Value));
4794 }
4795# endif
4796 }
4797#endif
4798 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
4799 }
4800
4801 /*
4802 * Guest Sysenter MSRs.
4803 * These flags are only set when MSR-bitmaps are not supported by the CPU and we cause
4804 * VM-exits on WRMSRs for these MSRs.
4805 */
4806 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR))
4807 {
4808 int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pMixedCtx->SysEnter.cs); AssertRCReturn(rc, rc);
4809 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR);
4810 }
4811
4812 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR))
4813 {
4814 int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, pMixedCtx->SysEnter.eip); AssertRCReturn(rc, rc);
4815 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
4816 }
4817
4818 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR))
4819 {
4820 int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, pMixedCtx->SysEnter.esp); AssertRCReturn(rc, rc);
4821 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
4822 }
4823
4824 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_EFER_MSR))
4825 {
4826 if (hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
4827 {
4828 /*
4829 * If the CPU supports VMCS controls for swapping EFER, use it. Otherwise, we have no option
4830 * but to use the auto-load store MSR area in the VMCS for swapping EFER. See @bugref{7368}.
4831 */
4832 if (pVM->hm.s.vmx.fSupportsVmcsEfer)
4833 {
4834 int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_EFER_FULL, pMixedCtx->msrEFER);
4835 AssertRCReturn(rc,rc);
4836 Log4(("Load[%RU32]: VMX_VMCS64_GUEST_EFER_FULL=%#RX64\n", pVCpu->idCpu, pMixedCtx->msrEFER));
4837 }
4838 else
4839 {
4840 int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K6_EFER, pMixedCtx->msrEFER, false /* fUpdateHostMsr */,
4841 NULL /* pfAddedAndUpdated */);
4842 AssertRCReturn(rc, rc);
4843
4844 /* We need to intercept reads too, see @bugref{7386} comment #16. */
4845 hmR0VmxSetMsrPermission(pVCpu, MSR_K6_EFER, VMXMSREXIT_INTERCEPT_READ, VMXMSREXIT_INTERCEPT_WRITE);
4846 Log4(("Load[%RU32]: MSR[--]: u32Msr=%#RX32 u64Value=%#RX64 cMsrs=%u\n", pVCpu->idCpu, MSR_K6_EFER,
4847 pMixedCtx->msrEFER, pVCpu->hm.s.vmx.cMsrs));
4848 }
4849 }
4850 else if (!pVM->hm.s.vmx.fSupportsVmcsEfer)
4851 hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, MSR_K6_EFER);
4852 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_EFER_MSR);
4853 }
4854
4855 return VINF_SUCCESS;
4856}
4857
4858
4859/**
4860 * Loads the guest activity state into the guest-state area in the VMCS.
4861 *
4862 * @returns VBox status code.
4863 * @param pVCpu Pointer to the VMCPU.
4864 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4865 * out-of-sync. Make sure to update the required fields
4866 * before using them.
4867 *
4868 * @remarks No-long-jump zone!!!
4869 */
4870static int hmR0VmxLoadGuestActivityState(PVMCPU pVCpu, PCPUMCTX pCtx)
4871{
4872 NOREF(pCtx);
4873 /** @todo See if we can make use of other states, e.g.
4874 * VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN or HLT. */
4875 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_ACTIVITY_STATE))
4876 {
4877 int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, VMX_VMCS_GUEST_ACTIVITY_ACTIVE);
4878 AssertRCReturn(rc, rc);
4879
4880 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_ACTIVITY_STATE);
4881 }
4882 return VINF_SUCCESS;
4883}
4884
4885
4886/**
4887 * Sets up the appropriate function to run guest code.
4888 *
4889 * @returns VBox status code.
4890 * @param pVCpu Pointer to the VMCPU.
4891 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4892 * out-of-sync. Make sure to update the required fields
4893 * before using them.
4894 *
4895 * @remarks No-long-jump zone!!!
4896 */
4897static int hmR0VmxSetupVMRunHandler(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4898{
4899 if (CPUMIsGuestInLongModeEx(pMixedCtx))
4900 {
4901#ifndef VBOX_ENABLE_64_BITS_GUESTS
4902 return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
4903#endif
4904 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests); /* Guaranteed by hmR3InitFinalizeR0(). */
4905#if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4906 /* 32-bit host. We need to switch to 64-bit before running the 64-bit guest. */
4907 if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0SwitcherStartVM64)
4908 {
4909 if (pVCpu->hm.s.vmx.pfnStartVM != NULL) /* Very first entry would have saved host-state already, ignore it. */
4910 {
4911 /* Currently, all mode changes sends us back to ring-3, so these should be set. See @bugref{6944}. */
4912 AssertMsg(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_VMX_EXIT_CTLS
4913 | HM_CHANGED_VMX_ENTRY_CTLS
4914 | HM_CHANGED_GUEST_EFER_MSR), ("flags=%#x\n", HMCPU_CF_VALUE(pVCpu)));
4915 }
4916 pVCpu->hm.s.vmx.pfnStartVM = VMXR0SwitcherStartVM64;
4917 }
4918#else
4919 /* 64-bit host or hybrid host. */
4920 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM64;
4921#endif
4922 }
4923 else
4924 {
4925 /* Guest is not in long mode, use the 32-bit handler. */
4926#if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4927 if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0StartVM32)
4928 {
4929 if (pVCpu->hm.s.vmx.pfnStartVM != NULL) /* Very first entry would have saved host-state already, ignore it. */
4930 {
4931 /* Currently, all mode changes sends us back to ring-3, so these should be set. See @bugref{6944}. */
4932 AssertMsg(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_VMX_EXIT_CTLS
4933 | HM_CHANGED_VMX_ENTRY_CTLS
4934 | HM_CHANGED_GUEST_EFER_MSR), ("flags=%#x\n", HMCPU_CF_VALUE(pVCpu)));
4935 }
4936 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4937 }
4938#else
4939 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4940#endif
4941 }
4942 Assert(pVCpu->hm.s.vmx.pfnStartVM);
4943 return VINF_SUCCESS;
4944}
4945
4946
4947/**
4948 * Wrapper for running the guest code in VT-x.
4949 *
4950 * @returns VBox strict status code.
4951 * @param pVM Pointer to the VM.
4952 * @param pVCpu Pointer to the VMCPU.
4953 * @param pCtx Pointer to the guest-CPU context.
4954 *
4955 * @remarks No-long-jump zone!!!
4956 */
4957DECLINLINE(int) hmR0VmxRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
4958{
4959 /*
4960 * 64-bit Windows uses XMM registers in the kernel as the Microsoft compiler expresses floating-point operations
4961 * using SSE instructions. Some XMM registers (XMM6-XMM15) are callee-saved and thus the need for this XMM wrapper.
4962 * Refer MSDN docs. "Configuring Programs for 64-bit / x64 Software Conventions / Register Usage" for details.
4963 */
4964 bool const fResumeVM = RT_BOOL(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_LAUNCHED);
4965 /** @todo Add stats for resume vs launch. */
4966#ifdef VBOX_WITH_KERNEL_USING_XMM
4967 return HMR0VMXStartVMWrapXMM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu, pVCpu->hm.s.vmx.pfnStartVM);
4968#else
4969 return pVCpu->hm.s.vmx.pfnStartVM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu);
4970#endif
4971}
4972
4973
4974/**
4975 * Reports world-switch error and dumps some useful debug info.
4976 *
4977 * @param pVM Pointer to the VM.
4978 * @param pVCpu Pointer to the VMCPU.
4979 * @param rcVMRun The return code from VMLAUNCH/VMRESUME.
4980 * @param pCtx Pointer to the guest-CPU context.
4981 * @param pVmxTransient Pointer to the VMX transient structure (only
4982 * exitReason updated).
4983 */
4984static void hmR0VmxReportWorldSwitchError(PVM pVM, PVMCPU pVCpu, int rcVMRun, PCPUMCTX pCtx, PVMXTRANSIENT pVmxTransient)
4985{
4986 Assert(pVM);
4987 Assert(pVCpu);
4988 Assert(pCtx);
4989 Assert(pVmxTransient);
4990 HMVMX_ASSERT_PREEMPT_SAFE();
4991
4992 Log4(("VM-entry failure: %Rrc\n", rcVMRun));
4993 switch (rcVMRun)
4994 {
4995 case VERR_VMX_INVALID_VMXON_PTR:
4996 AssertFailed();
4997 break;
4998 case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */
4999 case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */
5000 {
5001 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason);
5002 rc |= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
5003 rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
5004 AssertRC(rc);
5005
5006 pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
5007 /* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted().
5008 Cannot do it here as we may have been long preempted. */
5009
5010#ifdef VBOX_STRICT
5011 Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason,
5012 pVmxTransient->uExitReason));
5013 Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQualification));
5014 Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError));
5015 if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX)
5016 Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError]));
5017 else
5018 Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX));
5019 Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu));
5020 Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu));
5021
5022 /* VMX control bits. */
5023 uint32_t u32Val;
5024 uint64_t u64Val;
5025 HMVMXHCUINTREG uHCReg;
5026 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val); AssertRC(rc);
5027 Log4(("VMX_VMCS32_CTRL_PIN_EXEC %#RX32\n", u32Val));
5028 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val); AssertRC(rc);
5029 Log4(("VMX_VMCS32_CTRL_PROC_EXEC %#RX32\n", u32Val));
5030 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val); AssertRC(rc);
5031 Log4(("VMX_VMCS32_CTRL_PROC_EXEC2 %#RX32\n", u32Val));
5032 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val); AssertRC(rc);
5033 Log4(("VMX_VMCS32_CTRL_ENTRY %#RX32\n", u32Val));
5034 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val); AssertRC(rc);
5035 Log4(("VMX_VMCS32_CTRL_EXIT %#RX32\n", u32Val));
5036 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, &u32Val); AssertRC(rc);
5037 Log4(("VMX_VMCS32_CTRL_CR3_TARGET_COUNT %#RX32\n", u32Val));
5038 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32Val); AssertRC(rc);
5039 Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", u32Val));
5040 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &u32Val); AssertRC(rc);
5041 Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", u32Val));
5042 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &u32Val); AssertRC(rc);
5043 Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %u\n", u32Val));
5044 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, &u32Val); AssertRC(rc);
5045 Log4(("VMX_VMCS32_CTRL_TPR_THRESHOLD %u\n", u32Val));
5046 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &u32Val); AssertRC(rc);
5047 Log4(("VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT %u (guest MSRs)\n", u32Val));
5048 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &u32Val); AssertRC(rc);
5049 Log4(("VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT %u (host MSRs)\n", u32Val));
5050 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &u32Val); AssertRC(rc);
5051 Log4(("VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT %u (guest MSRs)\n", u32Val));
5052 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val); AssertRC(rc);
5053 Log4(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP %#RX32\n", u32Val));
5054 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, &u32Val); AssertRC(rc);
5055 Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK %#RX32\n", u32Val));
5056 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, &u32Val); AssertRC(rc);
5057 Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH %#RX32\n", u32Val));
5058 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc);
5059 Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg));
5060 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc);
5061 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
5062 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc);
5063 Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg));
5064 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc);
5065 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
5066 rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
5067 Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
5068
5069 /* Guest bits. */
5070 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val); AssertRC(rc);
5071 Log4(("Old Guest Rip %#RX64 New %#RX64\n", pCtx->rip, u64Val));
5072 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val); AssertRC(rc);
5073 Log4(("Old Guest Rsp %#RX64 New %#RX64\n", pCtx->rsp, u64Val));
5074 rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Val); AssertRC(rc);
5075 Log4(("Old Guest Rflags %#RX32 New %#RX32\n", pCtx->eflags.u32, u32Val));
5076 rc = VMXReadVmcs32(VMX_VMCS16_GUEST_FIELD_VPID, &u32Val); AssertRC(rc);
5077 Log4(("VMX_VMCS16_GUEST_FIELD_VPID %u\n", u32Val));
5078
5079 /* Host bits. */
5080 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR0, &uHCReg); AssertRC(rc);
5081 Log4(("Host CR0 %#RHr\n", uHCReg));
5082 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR3, &uHCReg); AssertRC(rc);
5083 Log4(("Host CR3 %#RHr\n", uHCReg));
5084 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR4, &uHCReg); AssertRC(rc);
5085 Log4(("Host CR4 %#RHr\n", uHCReg));
5086
5087 RTGDTR HostGdtr;
5088 PCX86DESCHC pDesc;
5089 ASMGetGDTR(&HostGdtr);
5090 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_CS, &u32Val); AssertRC(rc);
5091 Log4(("Host CS %#08x\n", u32Val));
5092 if (u32Val < HostGdtr.cbGdt)
5093 {
5094 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5095 HMR0DumpDescriptor(pDesc, u32Val, "CS: ");
5096 }
5097
5098 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_DS, &u32Val); AssertRC(rc);
5099 Log4(("Host DS %#08x\n", u32Val));
5100 if (u32Val < HostGdtr.cbGdt)
5101 {
5102 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5103 HMR0DumpDescriptor(pDesc, u32Val, "DS: ");
5104 }
5105
5106 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_ES, &u32Val); AssertRC(rc);
5107 Log4(("Host ES %#08x\n", u32Val));
5108 if (u32Val < HostGdtr.cbGdt)
5109 {
5110 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5111 HMR0DumpDescriptor(pDesc, u32Val, "ES: ");
5112 }
5113
5114 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_FS, &u32Val); AssertRC(rc);
5115 Log4(("Host FS %#08x\n", u32Val));
5116 if (u32Val < HostGdtr.cbGdt)
5117 {
5118 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5119 HMR0DumpDescriptor(pDesc, u32Val, "FS: ");
5120 }
5121
5122 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_GS, &u32Val); AssertRC(rc);
5123 Log4(("Host GS %#08x\n", u32Val));
5124 if (u32Val < HostGdtr.cbGdt)
5125 {
5126 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5127 HMR0DumpDescriptor(pDesc, u32Val, "GS: ");
5128 }
5129
5130 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_SS, &u32Val); AssertRC(rc);
5131 Log4(("Host SS %#08x\n", u32Val));
5132 if (u32Val < HostGdtr.cbGdt)
5133 {
5134 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5135 HMR0DumpDescriptor(pDesc, u32Val, "SS: ");
5136 }
5137
5138 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_TR, &u32Val); AssertRC(rc);
5139 Log4(("Host TR %#08x\n", u32Val));
5140 if (u32Val < HostGdtr.cbGdt)
5141 {
5142 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5143 HMR0DumpDescriptor(pDesc, u32Val, "TR: ");
5144 }
5145
5146 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_TR_BASE, &uHCReg); AssertRC(rc);
5147 Log4(("Host TR Base %#RHv\n", uHCReg));
5148 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, &uHCReg); AssertRC(rc);
5149 Log4(("Host GDTR Base %#RHv\n", uHCReg));
5150 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, &uHCReg); AssertRC(rc);
5151 Log4(("Host IDTR Base %#RHv\n", uHCReg));
5152 rc = VMXReadVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, &u32Val); AssertRC(rc);
5153 Log4(("Host SYSENTER CS %#08x\n", u32Val));
5154 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_EIP, &uHCReg); AssertRC(rc);
5155 Log4(("Host SYSENTER EIP %#RHv\n", uHCReg));
5156 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_ESP, &uHCReg); AssertRC(rc);
5157 Log4(("Host SYSENTER ESP %#RHv\n", uHCReg));
5158 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RSP, &uHCReg); AssertRC(rc);
5159 Log4(("Host RSP %#RHv\n", uHCReg));
5160 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RIP, &uHCReg); AssertRC(rc);
5161 Log4(("Host RIP %#RHv\n", uHCReg));
5162# if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
5163 if (HMVMX_IS_64BIT_HOST_MODE())
5164 {
5165 Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER)));
5166 Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR)));
5167 Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR)));
5168 Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR)));
5169 Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK)));
5170 Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE)));
5171 }
5172# endif
5173#endif /* VBOX_STRICT */
5174 break;
5175 }
5176
5177 default:
5178 /* Impossible */
5179 AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun));
5180 break;
5181 }
5182 NOREF(pVM); NOREF(pCtx);
5183}
5184
5185
5186#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
5187#ifndef VMX_USE_CACHED_VMCS_ACCESSES
5188# error "VMX_USE_CACHED_VMCS_ACCESSES not defined when it should be!"
5189#endif
5190#ifdef VBOX_STRICT
5191static bool hmR0VmxIsValidWriteField(uint32_t idxField)
5192{
5193 switch (idxField)
5194 {
5195 case VMX_VMCS_GUEST_RIP:
5196 case VMX_VMCS_GUEST_RSP:
5197 case VMX_VMCS_GUEST_SYSENTER_EIP:
5198 case VMX_VMCS_GUEST_SYSENTER_ESP:
5199 case VMX_VMCS_GUEST_GDTR_BASE:
5200 case VMX_VMCS_GUEST_IDTR_BASE:
5201 case VMX_VMCS_GUEST_CS_BASE:
5202 case VMX_VMCS_GUEST_DS_BASE:
5203 case VMX_VMCS_GUEST_ES_BASE:
5204 case VMX_VMCS_GUEST_FS_BASE:
5205 case VMX_VMCS_GUEST_GS_BASE:
5206 case VMX_VMCS_GUEST_SS_BASE:
5207 case VMX_VMCS_GUEST_LDTR_BASE:
5208 case VMX_VMCS_GUEST_TR_BASE:
5209 case VMX_VMCS_GUEST_CR3:
5210 return true;
5211 }
5212 return false;
5213}
5214
5215static bool hmR0VmxIsValidReadField(uint32_t idxField)
5216{
5217 switch (idxField)
5218 {
5219 /* Read-only fields. */
5220 case VMX_VMCS_RO_EXIT_QUALIFICATION:
5221 return true;
5222 }
5223 /* Remaining readable fields should also be writable. */
5224 return hmR0VmxIsValidWriteField(idxField);
5225}
5226#endif /* VBOX_STRICT */
5227
5228
5229/**
5230 * Executes the specified handler in 64-bit mode.
5231 *
5232 * @returns VBox status code.
5233 * @param pVM Pointer to the VM.
5234 * @param pVCpu Pointer to the VMCPU.
5235 * @param pCtx Pointer to the guest CPU context.
5236 * @param enmOp The operation to perform.
5237 * @param cParams Number of parameters.
5238 * @param paParam Array of 32-bit parameters.
5239 */
5240VMMR0DECL(int) VMXR0Execute64BitsHandler(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, HM64ON32OP enmOp,
5241 uint32_t cParams, uint32_t *paParam)
5242{
5243 int rc, rc2;
5244 PHMGLOBALCPUINFO pCpu;
5245 RTHCPHYS HCPhysCpuPage;
5246 RTCCUINTREG fOldEFlags;
5247
5248 AssertReturn(pVM->hm.s.pfnHost32ToGuest64R0, VERR_HM_NO_32_TO_64_SWITCHER);
5249 Assert(enmOp > HM64ON32OP_INVALID && enmOp < HM64ON32OP_END);
5250 Assert(pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Write.aField));
5251 Assert(pVCpu->hm.s.vmx.VMCSCache.Read.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Read.aField));
5252
5253#ifdef VBOX_STRICT
5254 for (uint32_t i = 0; i < pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries; i++)
5255 Assert(hmR0VmxIsValidWriteField(pVCpu->hm.s.vmx.VMCSCache.Write.aField[i]));
5256
5257 for (uint32_t i = 0; i <pVCpu->hm.s.vmx.VMCSCache.Read.cValidEntries; i++)
5258 Assert(hmR0VmxIsValidReadField(pVCpu->hm.s.vmx.VMCSCache.Read.aField[i]));
5259#endif
5260
5261 /* Disable interrupts. */
5262 fOldEFlags = ASMIntDisableFlags();
5263
5264#ifdef VBOX_WITH_VMMR0_DISABLE_LAPIC_NMI
5265 RTCPUID idHostCpu = RTMpCpuId();
5266 CPUMR0SetLApic(pVCpu, idHostCpu);
5267#endif
5268
5269 pCpu = HMR0GetCurrentCpu();
5270 HCPhysCpuPage = RTR0MemObjGetPagePhysAddr(pCpu->hMemObj, 0);
5271
5272 /* Clear VMCS. Marking it inactive, clearing implementation-specific data and writing VMCS data back to memory. */
5273 VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
5274
5275 /* Leave VMX Root Mode. */
5276 VMXDisable();
5277
5278 SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
5279
5280 CPUMSetHyperESP(pVCpu, VMMGetStackRC(pVCpu));
5281 CPUMSetHyperEIP(pVCpu, enmOp);
5282 for (int i = (int)cParams - 1; i >= 0; i--)
5283 CPUMPushHyper(pVCpu, paParam[i]);
5284
5285 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatWorldSwitch3264, z);
5286
5287 /* Call the switcher. */
5288 rc = pVM->hm.s.pfnHost32ToGuest64R0(pVM, RT_OFFSETOF(VM, aCpus[pVCpu->idCpu].cpum) - RT_OFFSETOF(VM, cpum));
5289 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatWorldSwitch3264, z);
5290
5291 /** @todo replace with hmR0VmxEnterRootMode() and hmR0VmxLeaveRootMode(). */
5292 /* Make sure the VMX instructions don't cause #UD faults. */
5293 SUPR0ChangeCR4(X86_CR4_VMXE, ~0);
5294
5295 /* Re-enter VMX Root Mode */
5296 rc2 = VMXEnable(HCPhysCpuPage);
5297 if (RT_FAILURE(rc2))
5298 {
5299 SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
5300 ASMSetFlags(fOldEFlags);
5301 pVM->hm.s.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
5302 return rc2;
5303 }
5304
5305 rc2 = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
5306 AssertRC(rc2);
5307 Assert(!(ASMGetFlags() & X86_EFL_IF));
5308 ASMSetFlags(fOldEFlags);
5309 return rc;
5310}
5311
5312
5313/**
5314 * Prepares for and executes VMLAUNCH (64-bit guests) for 32-bit hosts
5315 * supporting 64-bit guests.
5316 *
5317 * @returns VBox status code.
5318 * @param fResume Whether to VMLAUNCH or VMRESUME.
5319 * @param pCtx Pointer to the guest-CPU context.
5320 * @param pCache Pointer to the VMCS cache.
5321 * @param pVM Pointer to the VM.
5322 * @param pVCpu Pointer to the VMCPU.
5323 */
5324DECLASM(int) VMXR0SwitcherStartVM64(RTHCUINT fResume, PCPUMCTX pCtx, PVMCSCACHE pCache, PVM pVM, PVMCPU pVCpu)
5325{
5326 PHMGLOBALCPUINFO pCpu = NULL;
5327 RTHCPHYS HCPhysCpuPage = 0;
5328 int rc = VERR_INTERNAL_ERROR_5;
5329
5330 pCpu = HMR0GetCurrentCpu();
5331 HCPhysCpuPage = RTR0MemObjGetPagePhysAddr(pCpu->hMemObj, 0);
5332
5333#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5334 pCache->uPos = 1;
5335 pCache->interPD = PGMGetInterPaeCR3(pVM);
5336 pCache->pSwitcher = (uint64_t)pVM->hm.s.pfnHost32ToGuest64R0;
5337#endif
5338
5339#if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES)
5340 pCache->TestIn.HCPhysCpuPage = 0;
5341 pCache->TestIn.HCPhysVmcs = 0;
5342 pCache->TestIn.pCache = 0;
5343 pCache->TestOut.HCPhysVmcs = 0;
5344 pCache->TestOut.pCache = 0;
5345 pCache->TestOut.pCtx = 0;
5346 pCache->TestOut.eflags = 0;
5347#endif
5348
5349 uint32_t aParam[10];
5350 aParam[0] = (uint32_t)(HCPhysCpuPage); /* Param 1: VMXON physical address - Lo. */
5351 aParam[1] = (uint32_t)(HCPhysCpuPage >> 32); /* Param 1: VMXON physical address - Hi. */
5352 aParam[2] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs); /* Param 2: VMCS physical address - Lo. */
5353 aParam[3] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs >> 32); /* Param 2: VMCS physical address - Hi. */
5354 aParam[4] = VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache);
5355 aParam[5] = 0;
5356 aParam[6] = VM_RC_ADDR(pVM, pVM);
5357 aParam[7] = 0;
5358 aParam[8] = VM_RC_ADDR(pVM, pVCpu);
5359 aParam[9] = 0;
5360
5361#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5362 pCtx->dr[4] = pVM->hm.s.vmx.pScratchPhys + 16 + 8;
5363 *(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) = 1;
5364#endif
5365 rc = VMXR0Execute64BitsHandler(pVM, pVCpu, pCtx, HM64ON32OP_VMXRCStartVM64, RT_ELEMENTS(aParam), &aParam[0]);
5366
5367#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5368 Assert(*(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) == 5);
5369 Assert(pCtx->dr[4] == 10);
5370 *(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) = 0xff;
5371#endif
5372
5373#if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES)
5374 AssertMsg(pCache->TestIn.HCPhysCpuPage == HCPhysCpuPage, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysCpuPage, HCPhysCpuPage));
5375 AssertMsg(pCache->TestIn.HCPhysVmcs == pVCpu->hm.s.vmx.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
5376 pVCpu->hm.s.vmx.HCPhysVmcs));
5377 AssertMsg(pCache->TestIn.HCPhysVmcs == pCache->TestOut.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
5378 pCache->TestOut.HCPhysVmcs));
5379 AssertMsg(pCache->TestIn.pCache == pCache->TestOut.pCache, ("%RGv vs %RGv\n", pCache->TestIn.pCache,
5380 pCache->TestOut.pCache));
5381 AssertMsg(pCache->TestIn.pCache == VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache),
5382 ("%RGv vs %RGv\n", pCache->TestIn.pCache, VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache)));
5383 AssertMsg(pCache->TestIn.pCtx == pCache->TestOut.pCtx, ("%RGv vs %RGv\n", pCache->TestIn.pCtx,
5384 pCache->TestOut.pCtx));
5385 Assert(!(pCache->TestOut.eflags & X86_EFL_IF));
5386#endif
5387 return rc;
5388}
5389
5390
5391/**
5392 * Initialize the VMCS-Read cache.
5393 *
5394 * The VMCS cache is used for 32-bit hosts running 64-bit guests (except 32-bit
5395 * Darwin which runs with 64-bit paging in 32-bit mode) for 64-bit fields that
5396 * cannot be accessed in 32-bit mode. Some 64-bit fields -can- be accessed
5397 * (those that have a 32-bit FULL & HIGH part).
5398 *
5399 * @returns VBox status code.
5400 * @param pVM Pointer to the VM.
5401 * @param pVCpu Pointer to the VMCPU.
5402 */
5403static int hmR0VmxInitVmcsReadCache(PVM pVM, PVMCPU pVCpu)
5404{
5405#define VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, idxField) \
5406{ \
5407 Assert(pCache->Read.aField[idxField##_CACHE_IDX] == 0); \
5408 pCache->Read.aField[idxField##_CACHE_IDX] = idxField; \
5409 pCache->Read.aFieldVal[idxField##_CACHE_IDX] = 0; \
5410 ++cReadFields; \
5411}
5412
5413 AssertPtr(pVM);
5414 AssertPtr(pVCpu);
5415 PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache;
5416 uint32_t cReadFields = 0;
5417
5418 /*
5419 * Don't remove the #if 0'd fields in this code. They're listed here for consistency
5420 * and serve to indicate exceptions to the rules.
5421 */
5422
5423 /* Guest-natural selector base fields. */
5424#if 0
5425 /* These are 32-bit in practice. See Intel spec. 2.5 "Control Registers". */
5426 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR0);
5427 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR4);
5428#endif
5429 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_ES_BASE);
5430 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CS_BASE);
5431 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SS_BASE);
5432 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_DS_BASE);
5433 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_FS_BASE);
5434 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GS_BASE);
5435 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_LDTR_BASE);
5436 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_TR_BASE);
5437 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GDTR_BASE);
5438 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_IDTR_BASE);
5439 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RSP);
5440 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RIP);
5441#if 0
5442 /* Unused natural width guest-state fields. */
5443 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS);
5444 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3); /* Handled in Nested Paging case */
5445#endif
5446 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_ESP);
5447 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_EIP);
5448
5449 /* 64-bit guest-state fields; unused as we use two 32-bit VMREADs for these 64-bit fields (using "FULL" and "HIGH" fields). */
5450#if 0
5451 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL);
5452 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_DEBUGCTL_FULL);
5453 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PAT_FULL);
5454 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_EFER_FULL);
5455 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL);
5456 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE0_FULL);
5457 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE1_FULL);
5458 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE2_FULL);
5459 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE3_FULL);
5460#endif
5461
5462 /* Natural width guest-state fields. */
5463 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_QUALIFICATION);
5464#if 0
5465 /* Currently unused field. */
5466 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_GUEST_LINEAR_ADDR);
5467#endif
5468
5469 if (pVM->hm.s.fNestedPaging)
5470 {
5471 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3);
5472 AssertMsg(cReadFields == VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields,
5473 VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX));
5474 pCache->Read.cValidEntries = VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX;
5475 }
5476 else
5477 {
5478 AssertMsg(cReadFields == VMX_VMCS_MAX_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields, VMX_VMCS_MAX_CACHE_IDX));
5479 pCache->Read.cValidEntries = VMX_VMCS_MAX_CACHE_IDX;
5480 }
5481
5482#undef VMXLOCAL_INIT_READ_CACHE_FIELD
5483 return VINF_SUCCESS;
5484}
5485
5486
5487/**
5488 * Writes a field into the VMCS. This can either directly invoke a VMWRITE or
5489 * queue up the VMWRITE by using the VMCS write cache (on 32-bit hosts, except
5490 * darwin, running 64-bit guests).
5491 *
5492 * @returns VBox status code.
5493 * @param pVCpu Pointer to the VMCPU.
5494 * @param idxField The VMCS field encoding.
5495 * @param u64Val 16, 32 or 64-bit value.
5496 */
5497VMMR0DECL(int) VMXWriteVmcs64Ex(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val)
5498{
5499 int rc;
5500 switch (idxField)
5501 {
5502 /*
5503 * These fields consists of a "FULL" and a "HIGH" part which can be written to individually.
5504 */
5505 /* 64-bit Control fields. */
5506 case VMX_VMCS64_CTRL_IO_BITMAP_A_FULL:
5507 case VMX_VMCS64_CTRL_IO_BITMAP_B_FULL:
5508 case VMX_VMCS64_CTRL_MSR_BITMAP_FULL:
5509 case VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL:
5510 case VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL:
5511 case VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL:
5512 case VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL:
5513 case VMX_VMCS64_CTRL_TSC_OFFSET_FULL:
5514 case VMX_VMCS64_CTRL_VAPIC_PAGEADDR_FULL:
5515 case VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL:
5516 case VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL:
5517 case VMX_VMCS64_CTRL_EPTP_FULL:
5518 case VMX_VMCS64_CTRL_EPTP_LIST_FULL:
5519 /* 64-bit Guest-state fields. */
5520 case VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL:
5521 case VMX_VMCS64_GUEST_DEBUGCTL_FULL:
5522 case VMX_VMCS64_GUEST_PAT_FULL:
5523 case VMX_VMCS64_GUEST_EFER_FULL:
5524 case VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL:
5525 case VMX_VMCS64_GUEST_PDPTE0_FULL:
5526 case VMX_VMCS64_GUEST_PDPTE1_FULL:
5527 case VMX_VMCS64_GUEST_PDPTE2_FULL:
5528 case VMX_VMCS64_GUEST_PDPTE3_FULL:
5529 /* 64-bit Host-state fields. */
5530 case VMX_VMCS64_HOST_FIELD_PAT_FULL:
5531 case VMX_VMCS64_HOST_FIELD_EFER_FULL:
5532 case VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL:
5533 {
5534 rc = VMXWriteVmcs32(idxField, u64Val);
5535 rc |= VMXWriteVmcs32(idxField + 1, (uint32_t)(u64Val >> 32));
5536 break;
5537 }
5538
5539 /*
5540 * These fields do not have high and low parts. Queue up the VMWRITE by using the VMCS write-cache (for 64-bit
5541 * values). When we switch the host to 64-bit mode for running 64-bit guests, these VMWRITEs get executed then.
5542 */
5543 /* Natural-width Guest-state fields. */
5544 case VMX_VMCS_GUEST_CR3:
5545 case VMX_VMCS_GUEST_ES_BASE:
5546 case VMX_VMCS_GUEST_CS_BASE:
5547 case VMX_VMCS_GUEST_SS_BASE:
5548 case VMX_VMCS_GUEST_DS_BASE:
5549 case VMX_VMCS_GUEST_FS_BASE:
5550 case VMX_VMCS_GUEST_GS_BASE:
5551 case VMX_VMCS_GUEST_LDTR_BASE:
5552 case VMX_VMCS_GUEST_TR_BASE:
5553 case VMX_VMCS_GUEST_GDTR_BASE:
5554 case VMX_VMCS_GUEST_IDTR_BASE:
5555 case VMX_VMCS_GUEST_RSP:
5556 case VMX_VMCS_GUEST_RIP:
5557 case VMX_VMCS_GUEST_SYSENTER_ESP:
5558 case VMX_VMCS_GUEST_SYSENTER_EIP:
5559 {
5560 if (!(u64Val >> 32))
5561 {
5562 /* If this field is 64-bit, VT-x will zero out the top bits. */
5563 rc = VMXWriteVmcs32(idxField, (uint32_t)u64Val);
5564 }
5565 else
5566 {
5567 /* Assert that only the 32->64 switcher case should ever come here. */
5568 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests);
5569 rc = VMXWriteCachedVmcsEx(pVCpu, idxField, u64Val);
5570 }
5571 break;
5572 }
5573
5574 default:
5575 {
5576 AssertMsgFailed(("VMXWriteVmcs64Ex: Invalid field %#RX32 (pVCpu=%p u64Val=%#RX64)\n", idxField, pVCpu, u64Val));
5577 rc = VERR_INVALID_PARAMETER;
5578 break;
5579 }
5580 }
5581 AssertRCReturn(rc, rc);
5582 return rc;
5583}
5584
5585
5586/**
5587 * Queue up a VMWRITE by using the VMCS write cache.
5588 * This is only used on 32-bit hosts (except darwin) for 64-bit guests.
5589 *
5590 * @param pVCpu Pointer to the VMCPU.
5591 * @param idxField The VMCS field encoding.
5592 * @param u64Val 16, 32 or 64-bit value.
5593 */
5594VMMR0DECL(int) VMXWriteCachedVmcsEx(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val)
5595{
5596 AssertPtr(pVCpu);
5597 PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache;
5598
5599 AssertMsgReturn(pCache->Write.cValidEntries < VMCSCACHE_MAX_ENTRY - 1,
5600 ("entries=%u\n", pCache->Write.cValidEntries), VERR_ACCESS_DENIED);
5601
5602 /* Make sure there are no duplicates. */
5603 for (uint32_t i = 0; i < pCache->Write.cValidEntries; i++)
5604 {
5605 if (pCache->Write.aField[i] == idxField)
5606 {
5607 pCache->Write.aFieldVal[i] = u64Val;
5608 return VINF_SUCCESS;
5609 }
5610 }
5611
5612 pCache->Write.aField[pCache->Write.cValidEntries] = idxField;
5613 pCache->Write.aFieldVal[pCache->Write.cValidEntries] = u64Val;
5614 pCache->Write.cValidEntries++;
5615 return VINF_SUCCESS;
5616}
5617
5618/* Enable later when the assembly code uses these as callbacks. */
5619#if 0
5620/*
5621 * Loads the VMCS write-cache into the CPU (by executing VMWRITEs).
5622 *
5623 * @param pVCpu Pointer to the VMCPU.
5624 * @param pCache Pointer to the VMCS cache.
5625 *
5626 * @remarks No-long-jump zone!!!
5627 */
5628VMMR0DECL(void) VMXWriteCachedVmcsLoad(PVMCPU pVCpu, PVMCSCACHE pCache)
5629{
5630 AssertPtr(pCache);
5631 for (uint32_t i = 0; i < pCache->Write.cValidEntries; i++)
5632 {
5633 int rc = VMXWriteVmcs64(pCache->Write.aField[i], pCache->Write.aFieldVal[i]);
5634 AssertRC(rc);
5635 }
5636 pCache->Write.cValidEntries = 0;
5637}
5638
5639
5640/**
5641 * Stores the VMCS read-cache from the CPU (by executing VMREADs).
5642 *
5643 * @param pVCpu Pointer to the VMCPU.
5644 * @param pCache Pointer to the VMCS cache.
5645 *
5646 * @remarks No-long-jump zone!!!
5647 */
5648VMMR0DECL(void) VMXReadCachedVmcsStore(PVMCPU pVCpu, PVMCSCACHE pCache)
5649{
5650 AssertPtr(pCache);
5651 for (uint32_t i = 0; i < pCache->Read.cValidEntries; i++)
5652 {
5653 int rc = VMXReadVmcs64(pCache->Read.aField[i], &pCache->Read.aFieldVal[i]);
5654 AssertRC(rc);
5655 }
5656}
5657#endif
5658#endif /* HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) */
5659
5660
5661/**
5662 * Sets up the usage of TSC-offsetting and updates the VMCS.
5663 *
5664 * If offsetting is not possible, cause VM-exits on RDTSC(P)s. Also sets up the
5665 * VMX preemption timer.
5666 *
5667 * @returns VBox status code.
5668 * @param pVM Pointer to the cross context VM structure.
5669 * @param pVCpu Pointer to the VMCPU.
5670 *
5671 * @remarks No-long-jump zone!!!
5672 */
5673static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVM pVM, PVMCPU pVCpu)
5674{
5675 int rc;
5676 bool fOffsettedTsc;
5677 bool fParavirtTsc;
5678 if (pVM->hm.s.vmx.fUsePreemptTimer)
5679 {
5680 uint64_t cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVM, pVCpu, &pVCpu->hm.s.vmx.u64TSCOffset,
5681 &fOffsettedTsc, &fParavirtTsc);
5682
5683 /* Make sure the returned values have sane upper and lower boundaries. */
5684 uint64_t u64CpuHz = SUPGetCpuHzFromGipBySetIndex(g_pSUPGlobalInfoPage, pVCpu->iHostCpuSet);
5685 cTicksToDeadline = RT_MIN(cTicksToDeadline, u64CpuHz / 64); /* 1/64th of a second */
5686 cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 2048); /* 1/2048th of a second */
5687 cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift;
5688
5689 uint32_t cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16);
5690 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_PREEMPT_TIMER_VALUE, cPreemptionTickCount); AssertRC(rc);
5691 }
5692 else
5693 fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &pVCpu->hm.s.vmx.u64TSCOffset, &fParavirtTsc);
5694
5695 /** @todo later optimize this to be done elsewhere and not before every
5696 * VM-entry. */
5697 if (fParavirtTsc)
5698 {
5699 rc = GIMR0UpdateParavirtTsc(pVM, 0 /* u64Offset */);
5700 AssertRC(rc);
5701 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
5702 }
5703
5704 if (fOffsettedTsc)
5705 {
5706 /* Note: VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT takes precedence over TSC_OFFSET, applies to RDTSCP too. */
5707 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, pVCpu->hm.s.vmx.u64TSCOffset); AssertRC(rc);
5708
5709 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT;
5710 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc);
5711 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
5712 }
5713 else
5714 {
5715 /* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */
5716 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT;
5717 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc);
5718 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
5719 }
5720}
5721
5722
5723/**
5724 * Determines if an exception is a contributory exception.
5725 *
5726 * Contributory exceptions are ones which can cause double-faults unless the
5727 * original exception was a benign exception. Page-fault is intentionally not
5728 * included here as it's a conditional contributory exception.
5729 *
5730 * @returns true if the exception is contributory, false otherwise.
5731 * @param uVector The exception vector.
5732 */
5733DECLINLINE(bool) hmR0VmxIsContributoryXcpt(const uint32_t uVector)
5734{
5735 switch (uVector)
5736 {
5737 case X86_XCPT_GP:
5738 case X86_XCPT_SS:
5739 case X86_XCPT_NP:
5740 case X86_XCPT_TS:
5741 case X86_XCPT_DE:
5742 return true;
5743 default:
5744 break;
5745 }
5746 return false;
5747}
5748
5749
5750/**
5751 * Sets an event as a pending event to be injected into the guest.
5752 *
5753 * @param pVCpu Pointer to the VMCPU.
5754 * @param u32IntInfo The VM-entry interruption-information field.
5755 * @param cbInstr The VM-entry instruction length in bytes (for software
5756 * interrupts, exceptions and privileged software
5757 * exceptions).
5758 * @param u32ErrCode The VM-entry exception error code.
5759 * @param GCPtrFaultAddress The fault-address (CR2) in case it's a
5760 * page-fault.
5761 *
5762 * @remarks Statistics counter assumes this is a guest event being injected or
5763 * re-injected into the guest, i.e. 'StatInjectPendingReflect' is
5764 * always incremented.
5765 */
5766DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPU pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode,
5767 RTGCUINTPTR GCPtrFaultAddress)
5768{
5769 Assert(!pVCpu->hm.s.Event.fPending);
5770 pVCpu->hm.s.Event.fPending = true;
5771 pVCpu->hm.s.Event.u64IntInfo = u32IntInfo;
5772 pVCpu->hm.s.Event.u32ErrCode = u32ErrCode;
5773 pVCpu->hm.s.Event.cbInstr = cbInstr;
5774 pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
5775
5776 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect);
5777}
5778
5779
5780/**
5781 * Sets a double-fault (#DF) exception as pending-for-injection into the VM.
5782 *
5783 * @param pVCpu Pointer to the VMCPU.
5784 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
5785 * out-of-sync. Make sure to update the required fields
5786 * before using them.
5787 */
5788DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
5789{
5790 NOREF(pMixedCtx);
5791 uint32_t u32IntInfo = X86_XCPT_DF | VMX_EXIT_INTERRUPTION_INFO_VALID;
5792 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
5793 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
5794 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
5795}
5796
5797
5798/**
5799 * Handle a condition that occurred while delivering an event through the guest
5800 * IDT.
5801 *
5802 * @returns VBox status code (informational error codes included).
5803 * @retval VINF_SUCCESS if we should continue handling the VM-exit.
5804 * @retval VINF_HM_DOUBLE_FAULT if a #DF condition was detected and we ought to
5805 * continue execution of the guest which will delivery the #DF.
5806 * @retval VINF_EM_RESET if we detected a triple-fault condition.
5807 *
5808 * @param pVCpu Pointer to the VMCPU.
5809 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
5810 * out-of-sync. Make sure to update the required fields
5811 * before using them.
5812 * @param pVmxTransient Pointer to the VMX transient structure.
5813 *
5814 * @remarks No-long-jump zone!!!
5815 */
5816static int hmR0VmxCheckExitDueToEventDelivery(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
5817{
5818 uint32_t uExitVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo);
5819
5820 int rc = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
5821 AssertRCReturn(rc, rc);
5822 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
5823 AssertRCReturn(rc, rc);
5824
5825 if (VMX_IDT_VECTORING_INFO_VALID(pVmxTransient->uIdtVectoringInfo))
5826 {
5827 uint32_t uIdtVectorType = VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo);
5828 uint32_t uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo);
5829
5830 typedef enum
5831 {
5832 VMXREFLECTXCPT_XCPT, /* Reflect the exception to the guest or for further evaluation by VMM. */
5833 VMXREFLECTXCPT_DF, /* Reflect the exception as a double-fault to the guest. */
5834 VMXREFLECTXCPT_TF, /* Indicate a triple faulted state to the VMM. */
5835 VMXREFLECTXCPT_NONE /* Nothing to reflect. */
5836 } VMXREFLECTXCPT;
5837
5838 /* See Intel spec. 30.7.1.1 "Reflecting Exceptions to Guest Software". */
5839 VMXREFLECTXCPT enmReflect = VMXREFLECTXCPT_NONE;
5840 if (VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo))
5841 {
5842 if (uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT)
5843 {
5844 enmReflect = VMXREFLECTXCPT_XCPT;
5845#ifdef VBOX_STRICT
5846 if ( hmR0VmxIsContributoryXcpt(uIdtVector)
5847 && uExitVector == X86_XCPT_PF)
5848 {
5849 Log4(("IDT: vcpu[%RU32] Contributory #PF uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5850 }
5851#endif
5852 if ( uExitVector == X86_XCPT_PF
5853 && uIdtVector == X86_XCPT_PF)
5854 {
5855 pVmxTransient->fVectoringDoublePF = true;
5856 Log4(("IDT: vcpu[%RU32] Vectoring Double #PF uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5857 }
5858 else if ( (pVCpu->hm.s.vmx.u32XcptBitmap & HMVMX_CONTRIBUTORY_XCPT_MASK)
5859 && hmR0VmxIsContributoryXcpt(uExitVector)
5860 && ( hmR0VmxIsContributoryXcpt(uIdtVector)
5861 || uIdtVector == X86_XCPT_PF))
5862 {
5863 enmReflect = VMXREFLECTXCPT_DF;
5864 }
5865 else if (uIdtVector == X86_XCPT_DF)
5866 enmReflect = VMXREFLECTXCPT_TF;
5867 }
5868 else if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT
5869 || uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI)
5870 {
5871 /*
5872 * Ignore software interrupts (INT n), software exceptions (#BP, #OF) and
5873 * privileged software exception (#DB from ICEBP) as they reoccur when restarting the instruction.
5874 */
5875 enmReflect = VMXREFLECTXCPT_XCPT;
5876
5877 if (uExitVector == X86_XCPT_PF)
5878 {
5879 pVmxTransient->fVectoringPF = true;
5880 Log4(("IDT: vcpu[%RU32] Vectoring #PF due to Ext-Int/NMI. uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5881 }
5882 }
5883 }
5884 else if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
5885 || uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT
5886 || uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI)
5887 {
5888 /*
5889 * If event delivery caused an EPT violation/misconfig or APIC access VM-exit, then the VM-exit
5890 * interruption-information will not be valid as it's not an exception and we end up here. In such cases,
5891 * it is sufficient to reflect the original exception to the guest after handling the VM-exit.
5892 */
5893 enmReflect = VMXREFLECTXCPT_XCPT;
5894 }
5895
5896 /*
5897 * On CPUs that support Virtual NMIs, if this VM-exit (be it an exception or EPT violation/misconfig etc.) occurred
5898 * while delivering the NMI, we need to clear the block-by-NMI field in the guest interruptibility-state before
5899 * re-delivering the NMI after handling the VM-exit. Otherwise the subsequent VM-entry would fail.
5900 *
5901 * See Intel spec. 30.7.1.2 "Resuming Guest Software after Handling an Exception". See @bugref{7445}.
5902 */
5903 if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
5904 && enmReflect == VMXREFLECTXCPT_XCPT
5905 && (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
5906 && VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
5907 {
5908 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
5909 }
5910
5911 switch (enmReflect)
5912 {
5913 case VMXREFLECTXCPT_XCPT:
5914 {
5915 Assert( uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_SW_INT
5916 && uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
5917 && uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT);
5918
5919 uint32_t u32ErrCode = 0;
5920 if (VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uIdtVectoringInfo))
5921 {
5922 rc = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
5923 AssertRCReturn(rc, rc);
5924 u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
5925 }
5926
5927 /* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF. See hmR0VmxExitXcptPF(). */
5928 hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo),
5929 0 /* cbInstr */, u32ErrCode, pMixedCtx->cr2);
5930 rc = VINF_SUCCESS;
5931 Log4(("IDT: vcpu[%RU32] Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->idCpu,
5932 pVCpu->hm.s.Event.u64IntInfo, pVCpu->hm.s.Event.u32ErrCode));
5933
5934 break;
5935 }
5936
5937 case VMXREFLECTXCPT_DF:
5938 {
5939 hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
5940 rc = VINF_HM_DOUBLE_FAULT;
5941 Log4(("IDT: vcpu[%RU32] Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->idCpu,
5942 pVCpu->hm.s.Event.u64IntInfo, uIdtVector, uExitVector));
5943
5944 break;
5945 }
5946
5947 case VMXREFLECTXCPT_TF:
5948 {
5949 rc = VINF_EM_RESET;
5950 Log4(("IDT: vcpu[%RU32] Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", pVCpu->idCpu, uIdtVector,
5951 uExitVector));
5952 break;
5953 }
5954
5955 default:
5956 Assert(rc == VINF_SUCCESS);
5957 break;
5958 }
5959 }
5960 else if ( VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo)
5961 && VMX_EXIT_INTERRUPTION_INFO_NMI_UNBLOCK_IRET(pVmxTransient->uExitIntInfo)
5962 && uExitVector != X86_XCPT_DF
5963 && (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI))
5964 {
5965 /*
5966 * Execution of IRET caused this fault when NMI blocking was in effect (i.e we're in the guest NMI handler).
5967 * We need to set the block-by-NMI field so that NMIs remain blocked until the IRET execution is restarted.
5968 * See Intel spec. 30.7.1.2 "Resuming guest software after handling an exception".
5969 */
5970 if (!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
5971 {
5972 Log4(("hmR0VmxCheckExitDueToEventDelivery: vcpu[%RU32] Setting VMCPU_FF_BLOCK_NMIS. Valid=%RTbool uExitReason=%u\n",
5973 pVCpu->idCpu, VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo), pVmxTransient->uExitReason));
5974 VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
5975 }
5976 }
5977
5978 Assert(rc == VINF_SUCCESS || rc == VINF_HM_DOUBLE_FAULT || rc == VINF_EM_RESET);
5979 return rc;
5980}
5981
5982
5983/**
5984 * Saves the guest's CR0 register from the VMCS into the guest-CPU context.
5985 *
5986 * @returns VBox status code.
5987 * @param pVCpu Pointer to the VMCPU.
5988 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
5989 * out-of-sync. Make sure to update the required fields
5990 * before using them.
5991 *
5992 * @remarks No-long-jump zone!!!
5993 */
5994static int hmR0VmxSaveGuestCR0(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
5995{
5996 NOREF(pMixedCtx);
5997
5998 /*
5999 * While in the middle of saving guest-CR0, we could get preempted and re-invoked from the preemption hook,
6000 * see hmR0VmxLeave(). Safer to just make this code non-preemptible.
6001 */
6002 VMMRZCallRing3Disable(pVCpu);
6003 HM_DISABLE_PREEMPT();
6004
6005 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0))
6006 {
6007 uint32_t uVal = 0;
6008 int rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &uVal);
6009 AssertRCReturn(rc, rc);
6010
6011 uint32_t uShadow = 0;
6012 rc = VMXReadVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uShadow);
6013 AssertRCReturn(rc, rc);
6014
6015 uVal = (uShadow & pVCpu->hm.s.vmx.u32CR0Mask) | (uVal & ~pVCpu->hm.s.vmx.u32CR0Mask);
6016 CPUMSetGuestCR0(pVCpu, uVal);
6017 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0);
6018 }
6019
6020 HM_RESTORE_PREEMPT();
6021 VMMRZCallRing3Enable(pVCpu);
6022 return VINF_SUCCESS;
6023}
6024
6025
6026/**
6027 * Saves the guest's CR4 register from the VMCS into the guest-CPU context.
6028 *
6029 * @returns VBox status code.
6030 * @param pVCpu Pointer to the VMCPU.
6031 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6032 * out-of-sync. Make sure to update the required fields
6033 * before using them.
6034 *
6035 * @remarks No-long-jump zone!!!
6036 */
6037static int hmR0VmxSaveGuestCR4(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6038{
6039 NOREF(pMixedCtx);
6040
6041 int rc = VINF_SUCCESS;
6042 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4))
6043 {
6044 uint32_t uVal = 0;
6045 uint32_t uShadow = 0;
6046 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &uVal);
6047 AssertRCReturn(rc, rc);
6048 rc = VMXReadVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uShadow);
6049 AssertRCReturn(rc, rc);
6050
6051 uVal = (uShadow & pVCpu->hm.s.vmx.u32CR4Mask) | (uVal & ~pVCpu->hm.s.vmx.u32CR4Mask);
6052 CPUMSetGuestCR4(pVCpu, uVal);
6053 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4);
6054 }
6055 return rc;
6056}
6057
6058
6059/**
6060 * Saves the guest's RIP register from the VMCS into the guest-CPU context.
6061 *
6062 * @returns VBox status code.
6063 * @param pVCpu Pointer to the VMCPU.
6064 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6065 * out-of-sync. Make sure to update the required fields
6066 * before using them.
6067 *
6068 * @remarks No-long-jump zone!!!
6069 */
6070static int hmR0VmxSaveGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6071{
6072 int rc = VINF_SUCCESS;
6073 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP))
6074 {
6075 uint64_t u64Val = 0;
6076 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val);
6077 AssertRCReturn(rc, rc);
6078
6079 pMixedCtx->rip = u64Val;
6080 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP);
6081 }
6082 return rc;
6083}
6084
6085
6086/**
6087 * Saves the guest's RSP register from the VMCS into the guest-CPU context.
6088 *
6089 * @returns VBox status code.
6090 * @param pVCpu Pointer to the VMCPU.
6091 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6092 * out-of-sync. Make sure to update the required fields
6093 * before using them.
6094 *
6095 * @remarks No-long-jump zone!!!
6096 */
6097static int hmR0VmxSaveGuestRsp(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6098{
6099 int rc = VINF_SUCCESS;
6100 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RSP))
6101 {
6102 uint64_t u64Val = 0;
6103 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val);
6104 AssertRCReturn(rc, rc);
6105
6106 pMixedCtx->rsp = u64Val;
6107 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RSP);
6108 }
6109 return rc;
6110}
6111
6112
6113/**
6114 * Saves the guest's RFLAGS from the VMCS into the guest-CPU context.
6115 *
6116 * @returns VBox status code.
6117 * @param pVCpu Pointer to the VMCPU.
6118 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6119 * out-of-sync. Make sure to update the required fields
6120 * before using them.
6121 *
6122 * @remarks No-long-jump zone!!!
6123 */
6124static int hmR0VmxSaveGuestRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6125{
6126 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS))
6127 {
6128 uint32_t uVal = 0;
6129 int rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &uVal);
6130 AssertRCReturn(rc, rc);
6131
6132 pMixedCtx->eflags.u32 = uVal;
6133 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) /* Undo our real-on-v86-mode changes to eflags if necessary. */
6134 {
6135 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
6136 Log4(("Saving real-mode EFLAGS VT-x view=%#RX32\n", pMixedCtx->eflags.u32));
6137
6138 pMixedCtx->eflags.Bits.u1VM = 0;
6139 pMixedCtx->eflags.Bits.u2IOPL = pVCpu->hm.s.vmx.RealMode.Eflags.Bits.u2IOPL;
6140 }
6141
6142 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS);
6143 }
6144 return VINF_SUCCESS;
6145}
6146
6147
6148/**
6149 * Wrapper for saving the guest's RIP, RSP and RFLAGS from the VMCS into the
6150 * guest-CPU context.
6151 */
6152DECLINLINE(int) hmR0VmxSaveGuestRipRspRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6153{
6154 int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6155 rc |= hmR0VmxSaveGuestRsp(pVCpu, pMixedCtx);
6156 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
6157 return rc;
6158}
6159
6160
6161/**
6162 * Saves the guest's interruptibility-state ("interrupt shadow" as AMD calls it)
6163 * from the guest-state area in the VMCS.
6164 *
6165 * @param pVCpu Pointer to the VMCPU.
6166 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6167 * out-of-sync. Make sure to update the required fields
6168 * before using them.
6169 *
6170 * @remarks No-long-jump zone!!!
6171 */
6172static void hmR0VmxSaveGuestIntrState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6173{
6174 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_INTR_STATE))
6175 {
6176 uint32_t uIntrState = 0;
6177 int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
6178 AssertRC(rc);
6179
6180 if (!uIntrState)
6181 {
6182 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
6183 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
6184
6185 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6186 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
6187 }
6188 else
6189 {
6190 if (uIntrState & ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS
6191 | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI))
6192 {
6193 rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6194 AssertRC(rc);
6195 rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); /* for hmR0VmxGetGuestIntrState(). */
6196 AssertRC(rc);
6197
6198 EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip);
6199 Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS));
6200 }
6201 else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
6202 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
6203
6204 if (uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI)
6205 {
6206 if (!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6207 VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
6208 }
6209 else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6210 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
6211 }
6212
6213 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_INTR_STATE);
6214 }
6215}
6216
6217
6218/**
6219 * Saves the guest's activity state.
6220 *
6221 * @returns VBox status code.
6222 * @param pVCpu Pointer to the VMCPU.
6223 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6224 * out-of-sync. Make sure to update the required fields
6225 * before using them.
6226 *
6227 * @remarks No-long-jump zone!!!
6228 */
6229static int hmR0VmxSaveGuestActivityState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6230{
6231 NOREF(pMixedCtx);
6232 /* Nothing to do for now until we make use of different guest-CPU activity state. Just update the flag. */
6233 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_ACTIVITY_STATE);
6234 return VINF_SUCCESS;
6235}
6236
6237
6238/**
6239 * Saves the guest SYSENTER MSRs (SYSENTER_CS, SYSENTER_EIP, SYSENTER_ESP) from
6240 * the current VMCS into the guest-CPU context.
6241 *
6242 * @returns VBox status code.
6243 * @param pVCpu Pointer to the VMCPU.
6244 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6245 * out-of-sync. Make sure to update the required fields
6246 * before using them.
6247 *
6248 * @remarks No-long-jump zone!!!
6249 */
6250static int hmR0VmxSaveGuestSysenterMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6251{
6252 int rc = VINF_SUCCESS;
6253 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR))
6254 {
6255 uint32_t u32Val = 0;
6256 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val); AssertRCReturn(rc, rc);
6257 pMixedCtx->SysEnter.cs = u32Val;
6258 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR);
6259 }
6260
6261 uint64_t u64Val = 0;
6262 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR))
6263 {
6264 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, &u64Val); AssertRCReturn(rc, rc);
6265 pMixedCtx->SysEnter.eip = u64Val;
6266 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR);
6267 }
6268 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR))
6269 {
6270 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, &u64Val); AssertRCReturn(rc, rc);
6271 pMixedCtx->SysEnter.esp = u64Val;
6272 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR);
6273 }
6274 return rc;
6275}
6276
6277
6278/**
6279 * Saves the set of guest MSRs (that we restore lazily while leaving VT-x) from
6280 * the CPU back into the guest-CPU context.
6281 *
6282 * @returns VBox status code.
6283 * @param pVCpu Pointer to the VMCPU.
6284 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6285 * out-of-sync. Make sure to update the required fields
6286 * before using them.
6287 *
6288 * @remarks No-long-jump zone!!!
6289 */
6290static int hmR0VmxSaveGuestLazyMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6291{
6292#if HC_ARCH_BITS == 64
6293 if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
6294 {
6295 /* Since this can be called from our preemption hook it's safer to make the guest-MSRs update non-preemptible. */
6296 VMMRZCallRing3Disable(pVCpu);
6297 HM_DISABLE_PREEMPT();
6298
6299 /* Doing the check here ensures we don't overwrite already-saved guest MSRs from a preemption hook. */
6300 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS))
6301 {
6302 hmR0VmxLazySaveGuestMsrs(pVCpu, pMixedCtx);
6303 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS);
6304 }
6305
6306 HM_RESTORE_PREEMPT();
6307 VMMRZCallRing3Enable(pVCpu);
6308 }
6309 else
6310 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS);
6311#else
6312 NOREF(pMixedCtx);
6313 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS);
6314#endif
6315
6316 return VINF_SUCCESS;
6317}
6318
6319
6320/**
6321 * Saves the auto load/store'd guest MSRs from the current VMCS into
6322 * the guest-CPU context.
6323 *
6324 * @returns VBox status code.
6325 * @param pVCpu Pointer to the VMCPU.
6326 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6327 * out-of-sync. Make sure to update the required fields
6328 * before using them.
6329 *
6330 * @remarks No-long-jump zone!!!
6331 */
6332static int hmR0VmxSaveGuestAutoLoadStoreMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6333{
6334 if (HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS))
6335 return VINF_SUCCESS;
6336
6337 PVMXAUTOMSR pMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
6338 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
6339 Log4(("hmR0VmxSaveGuestAutoLoadStoreMsrs: cMsrs=%u\n", cMsrs));
6340 for (uint32_t i = 0; i < cMsrs; i++, pMsr++)
6341 {
6342 switch (pMsr->u32Msr)
6343 {
6344 case MSR_K8_TSC_AUX: CPUMR0SetGuestTscAux(pVCpu, pMsr->u64Value); break;
6345 case MSR_K8_LSTAR: pMixedCtx->msrLSTAR = pMsr->u64Value; break;
6346 case MSR_K6_STAR: pMixedCtx->msrSTAR = pMsr->u64Value; break;
6347 case MSR_K8_SF_MASK: pMixedCtx->msrSFMASK = pMsr->u64Value; break;
6348 case MSR_K8_KERNEL_GS_BASE: pMixedCtx->msrKERNELGSBASE = pMsr->u64Value; break;
6349 case MSR_K6_EFER: /* Nothing to do here since we intercept writes, see hmR0VmxLoadGuestMsrs(). */
6350 break;
6351
6352 default:
6353 {
6354 AssertMsgFailed(("Unexpected MSR in auto-load/store area. uMsr=%#RX32 cMsrs=%u\n", pMsr->u32Msr, cMsrs));
6355 pVCpu->hm.s.u32HMError = pMsr->u32Msr;
6356 return VERR_HM_UNEXPECTED_LD_ST_MSR;
6357 }
6358 }
6359 }
6360
6361 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS);
6362 return VINF_SUCCESS;
6363}
6364
6365
6366/**
6367 * Saves the guest control registers from the current VMCS into the guest-CPU
6368 * context.
6369 *
6370 * @returns VBox status code.
6371 * @param pVCpu Pointer to the VMCPU.
6372 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6373 * out-of-sync. Make sure to update the required fields
6374 * before using them.
6375 *
6376 * @remarks No-long-jump zone!!!
6377 */
6378static int hmR0VmxSaveGuestControlRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6379{
6380 /* Guest CR0. Guest FPU. */
6381 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6382 AssertRCReturn(rc, rc);
6383
6384 /* Guest CR4. */
6385 rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
6386 AssertRCReturn(rc, rc);
6387
6388 /* Guest CR2 - updated always during the world-switch or in #PF. */
6389 /* Guest CR3. Only changes with Nested Paging. This must be done -after- saving CR0 and CR4 from the guest! */
6390 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR3))
6391 {
6392 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
6393 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4));
6394
6395 PVM pVM = pVCpu->CTX_SUFF(pVM);
6396 if ( pVM->hm.s.vmx.fUnrestrictedGuest
6397 || ( pVM->hm.s.fNestedPaging
6398 && CPUMIsGuestPagingEnabledEx(pMixedCtx)))
6399 {
6400 uint64_t u64Val = 0;
6401 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_CR3, &u64Val);
6402 if (pMixedCtx->cr3 != u64Val)
6403 {
6404 CPUMSetGuestCR3(pVCpu, u64Val);
6405 if (VMMRZCallRing3IsEnabled(pVCpu))
6406 {
6407 PGMUpdateCR3(pVCpu, u64Val);
6408 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6409 }
6410 else
6411 {
6412 /* Set the force flag to inform PGM about it when necessary. It is cleared by PGMUpdateCR3().*/
6413 VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
6414 }
6415 }
6416
6417 /* If the guest is in PAE mode, sync back the PDPE's into the guest state. */
6418 if (CPUMIsGuestInPAEModeEx(pMixedCtx)) /* Reads CR0, CR4 and EFER MSR (EFER is always up-to-date). */
6419 {
6420 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &pVCpu->hm.s.aPdpes[0].u); AssertRCReturn(rc, rc);
6421 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &pVCpu->hm.s.aPdpes[1].u); AssertRCReturn(rc, rc);
6422 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &pVCpu->hm.s.aPdpes[2].u); AssertRCReturn(rc, rc);
6423 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &pVCpu->hm.s.aPdpes[3].u); AssertRCReturn(rc, rc);
6424
6425 if (VMMRZCallRing3IsEnabled(pVCpu))
6426 {
6427 PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6428 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6429 }
6430 else
6431 {
6432 /* Set the force flag to inform PGM about it when necessary. It is cleared by PGMGstUpdatePaePdpes(). */
6433 VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES);
6434 }
6435 }
6436 }
6437
6438 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR3);
6439 }
6440
6441 /*
6442 * Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> hmR0VmxCallRing3Callback()
6443 * -> VMMRZCallRing3Disable() -> hmR0VmxSaveGuestState() -> Set VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp
6444 * -> continue with VM-exit handling -> hmR0VmxSaveGuestControlRegs() and here we are.
6445 *
6446 * The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus
6447 * if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that
6448 * calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should
6449 * -NOT- check if HMVMX_UPDATED_GUEST_CR3 is already set or not!
6450 *
6451 * The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here.
6452 */
6453 if (VMMRZCallRing3IsEnabled(pVCpu))
6454 {
6455 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
6456 PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
6457
6458 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
6459 PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6460
6461 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6462 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6463 }
6464
6465 return rc;
6466}
6467
6468
6469/**
6470 * Reads a guest segment register from the current VMCS into the guest-CPU
6471 * context.
6472 *
6473 * @returns VBox status code.
6474 * @param pVCpu Pointer to the VMCPU.
6475 * @param idxSel Index of the selector in the VMCS.
6476 * @param idxLimit Index of the segment limit in the VMCS.
6477 * @param idxBase Index of the segment base in the VMCS.
6478 * @param idxAccess Index of the access rights of the segment in the VMCS.
6479 * @param pSelReg Pointer to the segment selector.
6480 *
6481 * @remarks No-long-jump zone!!!
6482 * @remarks Never call this function directly!!! Use the VMXLOCAL_READ_SEG()
6483 * macro as that takes care of whether to read from the VMCS cache or
6484 * not.
6485 */
6486DECLINLINE(int) hmR0VmxReadSegmentReg(PVMCPU pVCpu, uint32_t idxSel, uint32_t idxLimit, uint32_t idxBase, uint32_t idxAccess,
6487 PCPUMSELREG pSelReg)
6488{
6489 NOREF(pVCpu);
6490
6491 uint32_t u32Val = 0;
6492 int rc = VMXReadVmcs32(idxSel, &u32Val);
6493 AssertRCReturn(rc, rc);
6494 pSelReg->Sel = (uint16_t)u32Val;
6495 pSelReg->ValidSel = (uint16_t)u32Val;
6496 pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
6497
6498 rc = VMXReadVmcs32(idxLimit, &u32Val);
6499 AssertRCReturn(rc, rc);
6500 pSelReg->u32Limit = u32Val;
6501
6502 uint64_t u64Val = 0;
6503 rc = VMXReadVmcsGstNByIdxVal(idxBase, &u64Val);
6504 AssertRCReturn(rc, rc);
6505 pSelReg->u64Base = u64Val;
6506
6507 rc = VMXReadVmcs32(idxAccess, &u32Val);
6508 AssertRCReturn(rc, rc);
6509 pSelReg->Attr.u = u32Val;
6510
6511 /*
6512 * If VT-x marks the segment as unusable, most other bits remain undefined:
6513 * - For CS the L, D and G bits have meaning.
6514 * - For SS the DPL has meaning (it -is- the CPL for Intel and VBox).
6515 * - For the remaining data segments no bits are defined.
6516 *
6517 * The present bit and the unusable bit has been observed to be set at the
6518 * same time (the selector was supposed to be invalid as we started executing
6519 * a V8086 interrupt in ring-0).
6520 *
6521 * What should be important for the rest of the VBox code, is that the P bit is
6522 * cleared. Some of the other VBox code recognizes the unusable bit, but
6523 * AMD-V certainly don't, and REM doesn't really either. So, to be on the
6524 * safe side here, we'll strip off P and other bits we don't care about. If
6525 * any code breaks because Attr.u != 0 when Sel < 4, it should be fixed.
6526 *
6527 * See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
6528 */
6529 if (pSelReg->Attr.u & X86DESCATTR_UNUSABLE)
6530 {
6531 Assert(idxSel != VMX_VMCS16_GUEST_FIELD_TR); /* TR is the only selector that can never be unusable. */
6532
6533 /* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */
6534 pSelReg->Attr.u &= X86DESCATTR_UNUSABLE | X86DESCATTR_L | X86DESCATTR_D | X86DESCATTR_G
6535 | X86DESCATTR_DPL | X86DESCATTR_TYPE | X86DESCATTR_DT;
6536
6537 Log4(("hmR0VmxReadSegmentReg: Unusable idxSel=%#x attr=%#x -> %#x\n", idxSel, u32Val, pSelReg->Attr.u));
6538#ifdef DEBUG_bird
6539 AssertMsg((u32Val & ~X86DESCATTR_P) == pSelReg->Attr.u,
6540 ("%#x: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n",
6541 idxSel, u32Val, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit));
6542#endif
6543 }
6544 return VINF_SUCCESS;
6545}
6546
6547
6548#ifdef VMX_USE_CACHED_VMCS_ACCESSES
6549# define VMXLOCAL_READ_SEG(Sel, CtxSel) \
6550 hmR0VmxReadSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_##Sel, VMX_VMCS32_GUEST_##Sel##_LIMIT, \
6551 VMX_VMCS_GUEST_##Sel##_BASE_CACHE_IDX, VMX_VMCS32_GUEST_##Sel##_ACCESS_RIGHTS, &pMixedCtx->CtxSel)
6552#else
6553# define VMXLOCAL_READ_SEG(Sel, CtxSel) \
6554 hmR0VmxReadSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_##Sel, VMX_VMCS32_GUEST_##Sel##_LIMIT, \
6555 VMX_VMCS_GUEST_##Sel##_BASE, VMX_VMCS32_GUEST_##Sel##_ACCESS_RIGHTS, &pMixedCtx->CtxSel)
6556#endif
6557
6558
6559/**
6560 * Saves the guest segment registers from the current VMCS into the guest-CPU
6561 * context.
6562 *
6563 * @returns VBox status code.
6564 * @param pVCpu Pointer to the VMCPU.
6565 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6566 * out-of-sync. Make sure to update the required fields
6567 * before using them.
6568 *
6569 * @remarks No-long-jump zone!!!
6570 */
6571static int hmR0VmxSaveGuestSegmentRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6572{
6573 /* Guest segment registers. */
6574 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SEGMENT_REGS))
6575 {
6576 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc);
6577 rc = VMXLOCAL_READ_SEG(CS, cs); AssertRCReturn(rc, rc);
6578 rc = VMXLOCAL_READ_SEG(SS, ss); AssertRCReturn(rc, rc);
6579 rc = VMXLOCAL_READ_SEG(DS, ds); AssertRCReturn(rc, rc);
6580 rc = VMXLOCAL_READ_SEG(ES, es); AssertRCReturn(rc, rc);
6581 rc = VMXLOCAL_READ_SEG(FS, fs); AssertRCReturn(rc, rc);
6582 rc = VMXLOCAL_READ_SEG(GS, gs); AssertRCReturn(rc, rc);
6583
6584 /* Restore segment attributes for real-on-v86 mode hack. */
6585 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
6586 {
6587 pMixedCtx->cs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrCS.u;
6588 pMixedCtx->ss.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrSS.u;
6589 pMixedCtx->ds.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrDS.u;
6590 pMixedCtx->es.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrES.u;
6591 pMixedCtx->fs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrFS.u;
6592 pMixedCtx->gs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrGS.u;
6593 }
6594 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SEGMENT_REGS);
6595 }
6596
6597 return VINF_SUCCESS;
6598}
6599
6600
6601/**
6602 * Saves the guest descriptor table registers and task register from the current
6603 * VMCS into the guest-CPU context.
6604 *
6605 * @returns VBox status code.
6606 * @param pVCpu Pointer to the VMCPU.
6607 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6608 * out-of-sync. Make sure to update the required fields
6609 * before using them.
6610 *
6611 * @remarks No-long-jump zone!!!
6612 */
6613static int hmR0VmxSaveGuestTableRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6614{
6615 int rc = VINF_SUCCESS;
6616
6617 /* Guest LDTR. */
6618 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LDTR))
6619 {
6620 rc = VMXLOCAL_READ_SEG(LDTR, ldtr);
6621 AssertRCReturn(rc, rc);
6622 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LDTR);
6623 }
6624
6625 /* Guest GDTR. */
6626 uint64_t u64Val = 0;
6627 uint32_t u32Val = 0;
6628 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_GDTR))
6629 {
6630 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, &u64Val); AssertRCReturn(rc, rc);
6631 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRCReturn(rc, rc);
6632 pMixedCtx->gdtr.pGdt = u64Val;
6633 pMixedCtx->gdtr.cbGdt = u32Val;
6634 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_GDTR);
6635 }
6636
6637 /* Guest IDTR. */
6638 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_IDTR))
6639 {
6640 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, &u64Val); AssertRCReturn(rc, rc);
6641 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRCReturn(rc, rc);
6642 pMixedCtx->idtr.pIdt = u64Val;
6643 pMixedCtx->idtr.cbIdt = u32Val;
6644 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_IDTR);
6645 }
6646
6647 /* Guest TR. */
6648 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_TR))
6649 {
6650 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6651 AssertRCReturn(rc, rc);
6652
6653 /* For real-mode emulation using virtual-8086 mode we have the fake TSS (pRealModeTSS) in TR, don't save the fake one. */
6654 if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
6655 {
6656 rc = VMXLOCAL_READ_SEG(TR, tr);
6657 AssertRCReturn(rc, rc);
6658 }
6659 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_TR);
6660 }
6661 return rc;
6662}
6663
6664#undef VMXLOCAL_READ_SEG
6665
6666
6667/**
6668 * Saves the guest debug-register DR7 from the current VMCS into the guest-CPU
6669 * context.
6670 *
6671 * @returns VBox status code.
6672 * @param pVCpu Pointer to the VMCPU.
6673 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6674 * out-of-sync. Make sure to update the required fields
6675 * before using them.
6676 *
6677 * @remarks No-long-jump zone!!!
6678 */
6679static int hmR0VmxSaveGuestDR7(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6680{
6681 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_DEBUG))
6682 {
6683 if (!pVCpu->hm.s.fUsingHyperDR7)
6684 {
6685 /* Upper 32-bits are always zero. See Intel spec. 2.7.3 "Loading and Storing Debug Registers". */
6686 uint32_t u32Val;
6687 int rc = VMXReadVmcs32(VMX_VMCS_GUEST_DR7, &u32Val); AssertRCReturn(rc, rc);
6688 pMixedCtx->dr[7] = u32Val;
6689 }
6690
6691 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_DEBUG);
6692 }
6693 return VINF_SUCCESS;
6694}
6695
6696
6697/**
6698 * Saves the guest APIC state from the current VMCS into the guest-CPU context.
6699 *
6700 * @returns VBox status code.
6701 * @param pVCpu Pointer to the VMCPU.
6702 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6703 * out-of-sync. Make sure to update the required fields
6704 * before using them.
6705 *
6706 * @remarks No-long-jump zone!!!
6707 */
6708static int hmR0VmxSaveGuestApicState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6709{
6710 NOREF(pMixedCtx);
6711
6712 /* Updating TPR is already done in hmR0VmxPostRunGuest(). Just update the flag. */
6713 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_APIC_STATE);
6714 return VINF_SUCCESS;
6715}
6716
6717
6718/**
6719 * Saves the entire guest state from the currently active VMCS into the
6720 * guest-CPU context.
6721 *
6722 * This essentially VMREADs all guest-data.
6723 *
6724 * @returns VBox status code.
6725 * @param pVCpu Pointer to the VMCPU.
6726 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
6727 * out-of-sync. Make sure to update the required fields
6728 * before using them.
6729 */
6730static int hmR0VmxSaveGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6731{
6732 Assert(pVCpu);
6733 Assert(pMixedCtx);
6734
6735 if (HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL)
6736 return VINF_SUCCESS;
6737
6738 /* Though we can longjmp to ring-3 due to log-flushes here and get recalled
6739 again on the ring-3 callback path, there is no real need to. */
6740 if (VMMRZCallRing3IsEnabled(pVCpu))
6741 VMMR0LogFlushDisable(pVCpu);
6742 else
6743 Assert(VMMR0IsLogFlushDisabled(pVCpu));
6744 Log4Func(("vcpu[%RU32]\n", pVCpu->idCpu));
6745
6746 int rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
6747 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestRipRspRflags failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6748
6749 rc = hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6750 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestControlRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6751
6752 rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
6753 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestSegmentRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6754
6755 rc = hmR0VmxSaveGuestTableRegs(pVCpu, pMixedCtx);
6756 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestTableRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6757
6758 rc = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
6759 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestDR7 failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6760
6761 rc = hmR0VmxSaveGuestSysenterMsrs(pVCpu, pMixedCtx);
6762 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestSysenterMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6763
6764 rc = hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
6765 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestLazyMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6766
6767 rc = hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
6768 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestAutoLoadStoreMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6769
6770 rc = hmR0VmxSaveGuestActivityState(pVCpu, pMixedCtx);
6771 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestActivityState failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6772
6773 rc = hmR0VmxSaveGuestApicState(pVCpu, pMixedCtx);
6774 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestApicState failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6775
6776 AssertMsg(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL,
6777 ("Missed guest state bits while saving state; residue %RX32\n", HMVMXCPU_GST_VALUE(pVCpu)));
6778
6779 if (VMMRZCallRing3IsEnabled(pVCpu))
6780 VMMR0LogFlushEnable(pVCpu);
6781
6782 return VINF_SUCCESS;
6783}
6784
6785
6786/**
6787 * Saves basic guest registers needed for IEM instruction execution.
6788 *
6789 * @returns VBox status code (OR-able).
6790 * @param pVCpu Pointer to the cross context CPU data for the calling
6791 * EMT.
6792 * @param pMixedCtx Pointer to the CPU context of the guest.
6793 * @param fMemory Whether the instruction being executed operates on
6794 * memory or not. Only CR0 is synced up if clear.
6795 * @param fNeedRsp Need RSP (any instruction working on GPRs or stack).
6796 */
6797static int hmR0VmxSaveGuestRegsForIemExec(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fMemory, bool fNeedRsp)
6798{
6799 /*
6800 * We assume all general purpose registers other than RSP are available.
6801 *
6802 * RIP is a must, as it will be incremented or otherwise changed.
6803 *
6804 * RFLAGS are always required to figure the CPL.
6805 *
6806 * RSP isn't always required, however it's a GPR, so frequently required.
6807 *
6808 * SS and CS are the only segment register needed if IEM doesn't do memory
6809 * access (CPL + 16/32/64-bit mode), but we can only get all segment registers.
6810 *
6811 * CR0 is always required by IEM for the CPL, while CR3 and CR4 will only
6812 * be required for memory accesses.
6813 *
6814 * Note! Before IEM dispatches an exception, it will call us to sync in everything.
6815 */
6816 int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6817 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
6818 if (fNeedRsp)
6819 rc |= hmR0VmxSaveGuestRsp(pVCpu, pMixedCtx);
6820 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
6821 if (!fMemory)
6822 rc |= hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6823 else
6824 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6825 return rc;
6826}
6827
6828
6829/**
6830 * Ensures that we've got a complete basic guest-context.
6831 *
6832 * This excludes the FPU, SSE, AVX, and similar extended state. The interface
6833 * is for the interpreter.
6834 *
6835 * @returns VBox status code.
6836 * @param pVCpu Pointer to the VMCPU of the calling EMT.
6837 * @param pMixedCtx Pointer to the guest-CPU context which may have data
6838 * needing to be synced in.
6839 * @thread EMT(pVCpu)
6840 */
6841VMMR0_INT_DECL(int) HMR0EnsureCompleteBasicContext(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6842{
6843 /* Note! Since this is only applicable to VT-x, the implementation is placed
6844 in the VT-x part of the sources instead of the generic stuff. */
6845 if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported)
6846 return hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
6847 return VINF_SUCCESS;
6848}
6849
6850
6851/**
6852 * Check per-VM and per-VCPU force flag actions that require us to go back to
6853 * ring-3 for one reason or another.
6854 *
6855 * @returns VBox status code (information status code included).
6856 * @retval VINF_SUCCESS if we don't have any actions that require going back to
6857 * ring-3.
6858 * @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
6859 * @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
6860 * interrupts)
6861 * @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
6862 * all EMTs to be in ring-3.
6863 * @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
6864 * @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
6865 * to the EM loop.
6866 *
6867 * @param pVM Pointer to the VM.
6868 * @param pVCpu Pointer to the VMCPU.
6869 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
6870 * out-of-sync. Make sure to update the required fields
6871 * before using them.
6872 */
6873static int hmR0VmxCheckForceFlags(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6874{
6875 Assert(VMMRZCallRing3IsEnabled(pVCpu));
6876
6877 if ( VM_FF_IS_PENDING(pVM, !pVCpu->hm.s.fSingleInstruction
6878 ? VM_FF_HP_R0_PRE_HM_MASK : VM_FF_HP_R0_PRE_HM_STEP_MASK)
6879 || VMCPU_FF_IS_PENDING(pVCpu, !pVCpu->hm.s.fSingleInstruction
6880 ? VMCPU_FF_HP_R0_PRE_HM_MASK : VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
6881 {
6882 /* We need the control registers now, make sure the guest-CPU context is updated. */
6883 int rc3 = hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6884 AssertRCReturn(rc3, rc3);
6885
6886 /* Pending HM CR3 sync. */
6887 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
6888 {
6889 int rc2 = PGMUpdateCR3(pVCpu, pMixedCtx->cr3);
6890 AssertMsgReturn(rc2 == VINF_SUCCESS || rc2 == VINF_PGM_SYNC_CR3,
6891 ("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS);
6892 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6893 }
6894
6895 /* Pending HM PAE PDPEs. */
6896 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
6897 {
6898 PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6899 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6900 }
6901
6902 /* Pending PGM C3 sync. */
6903 if (VMCPU_FF_IS_PENDING(pVCpu,VMCPU_FF_PGM_SYNC_CR3 | VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
6904 {
6905 int rc2 = PGMSyncCR3(pVCpu, pMixedCtx->cr0, pMixedCtx->cr3, pMixedCtx->cr4,
6906 VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
6907 if (rc2 != VINF_SUCCESS)
6908 {
6909 AssertRC(rc2);
6910 Log4(("hmR0VmxCheckForceFlags: PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", rc2));
6911 return rc2;
6912 }
6913 }
6914
6915 /* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
6916 if ( VM_FF_IS_PENDING(pVM, VM_FF_HM_TO_R3_MASK)
6917 || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
6918 {
6919 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
6920 int rc2 = RT_UNLIKELY(VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_NO_MEMORY : VINF_EM_RAW_TO_R3;
6921 Log4(("hmR0VmxCheckForceFlags: HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc2));
6922 return rc2;
6923 }
6924
6925 /* Pending VM request packets, such as hardware interrupts. */
6926 if ( VM_FF_IS_PENDING(pVM, VM_FF_REQUEST)
6927 || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_REQUEST))
6928 {
6929 Log4(("hmR0VmxCheckForceFlags: Pending VM request forcing us back to ring-3\n"));
6930 return VINF_EM_PENDING_REQUEST;
6931 }
6932
6933 /* Pending PGM pool flushes. */
6934 if (VM_FF_IS_PENDING(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
6935 {
6936 Log4(("hmR0VmxCheckForceFlags: PGM pool flush pending forcing us back to ring-3\n"));
6937 return VINF_PGM_POOL_FLUSH_PENDING;
6938 }
6939
6940 /* Pending DMA requests. */
6941 if (VM_FF_IS_PENDING(pVM, VM_FF_PDM_DMA))
6942 {
6943 Log4(("hmR0VmxCheckForceFlags: Pending DMA request forcing us back to ring-3\n"));
6944 return VINF_EM_RAW_TO_R3;
6945 }
6946 }
6947
6948 return VINF_SUCCESS;
6949}
6950
6951
6952/**
6953 * Converts any TRPM trap into a pending HM event. This is typically used when
6954 * entering from ring-3 (not longjmp returns).
6955 *
6956 * @param pVCpu Pointer to the VMCPU.
6957 */
6958static void hmR0VmxTrpmTrapToPendingEvent(PVMCPU pVCpu)
6959{
6960 Assert(TRPMHasTrap(pVCpu));
6961 Assert(!pVCpu->hm.s.Event.fPending);
6962
6963 uint8_t uVector;
6964 TRPMEVENT enmTrpmEvent;
6965 RTGCUINT uErrCode;
6966 RTGCUINTPTR GCPtrFaultAddress;
6967 uint8_t cbInstr;
6968
6969 int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr);
6970 AssertRC(rc);
6971
6972 /* Refer Intel spec. 24.8.3 "VM-entry Controls for Event Injection" for the format of u32IntInfo. */
6973 uint32_t u32IntInfo = uVector | VMX_EXIT_INTERRUPTION_INFO_VALID;
6974 if (enmTrpmEvent == TRPM_TRAP)
6975 {
6976 switch (uVector)
6977 {
6978 case X86_XCPT_NMI:
6979 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
6980 break;
6981
6982 case X86_XCPT_BP:
6983 case X86_XCPT_OF:
6984 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
6985 break;
6986
6987 case X86_XCPT_PF:
6988 case X86_XCPT_DF:
6989 case X86_XCPT_TS:
6990 case X86_XCPT_NP:
6991 case X86_XCPT_SS:
6992 case X86_XCPT_GP:
6993 case X86_XCPT_AC:
6994 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
6995 /* no break! */
6996 default:
6997 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
6998 break;
6999 }
7000 }
7001 else if (enmTrpmEvent == TRPM_HARDWARE_INT)
7002 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7003 else if (enmTrpmEvent == TRPM_SOFTWARE_INT)
7004 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7005 else
7006 AssertMsgFailed(("Invalid TRPM event type %d\n", enmTrpmEvent));
7007
7008 rc = TRPMResetTrap(pVCpu);
7009 AssertRC(rc);
7010 Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n",
7011 u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress));
7012
7013 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress);
7014 STAM_COUNTER_DEC(&pVCpu->hm.s.StatInjectPendingReflect);
7015}
7016
7017
7018/**
7019 * Converts any pending HM event into a TRPM trap. Typically used when leaving
7020 * VT-x to execute any instruction.
7021 *
7022 * @param pvCpu Pointer to the VMCPU.
7023 */
7024static void hmR0VmxPendingEventToTrpmTrap(PVMCPU pVCpu)
7025{
7026 Assert(pVCpu->hm.s.Event.fPending);
7027
7028 uint32_t uVectorType = VMX_IDT_VECTORING_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
7029 uint32_t uVector = VMX_IDT_VECTORING_INFO_VECTOR(pVCpu->hm.s.Event.u64IntInfo);
7030 bool fErrorCodeValid = VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVCpu->hm.s.Event.u64IntInfo);
7031 uint32_t uErrorCode = pVCpu->hm.s.Event.u32ErrCode;
7032
7033 /* If a trap was already pending, we did something wrong! */
7034 Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo */, NULL /* pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
7035
7036 TRPMEVENT enmTrapType;
7037 switch (uVectorType)
7038 {
7039 case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT:
7040 enmTrapType = TRPM_HARDWARE_INT;
7041 break;
7042
7043 case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
7044 enmTrapType = TRPM_SOFTWARE_INT;
7045 break;
7046
7047 case VMX_IDT_VECTORING_INFO_TYPE_NMI:
7048 case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
7049 case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT: /* #BP and #OF */
7050 case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT:
7051 enmTrapType = TRPM_TRAP;
7052 break;
7053
7054 default:
7055 AssertMsgFailed(("Invalid trap type %#x\n", uVectorType));
7056 enmTrapType = TRPM_32BIT_HACK;
7057 break;
7058 }
7059
7060 Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType));
7061
7062 int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
7063 AssertRC(rc);
7064
7065 if (fErrorCodeValid)
7066 TRPMSetErrorCode(pVCpu, uErrorCode);
7067
7068 if ( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
7069 && uVector == X86_XCPT_PF)
7070 {
7071 TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
7072 }
7073 else if ( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
7074 || uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
7075 || uVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT)
7076 {
7077 AssertMsg( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
7078 || (uVector == X86_XCPT_BP || uVector == X86_XCPT_OF),
7079 ("Invalid vector: uVector=%#x uVectorType=%#x\n", uVector, uVectorType));
7080 TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
7081 }
7082 pVCpu->hm.s.Event.fPending = false;
7083}
7084
7085
7086/**
7087 * Does the necessary state syncing before returning to ring-3 for any reason
7088 * (longjmp, preemption, voluntary exits to ring-3) from VT-x.
7089 *
7090 * @returns VBox status code.
7091 * @param pVM Pointer to the VM.
7092 * @param pVCpu Pointer to the VMCPU.
7093 * @param pMixedCtx Pointer to the guest-CPU context. The data may
7094 * be out-of-sync. Make sure to update the required
7095 * fields before using them.
7096 * @param fSaveGuestState Whether to save the guest state or not.
7097 *
7098 * @remarks No-long-jmp zone!!!
7099 */
7100static int hmR0VmxLeave(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fSaveGuestState)
7101{
7102 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
7103 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
7104
7105 RTCPUID idCpu = RTMpCpuId();
7106 Log4Func(("HostCpuId=%u\n", idCpu));
7107
7108 /*
7109 * !!! IMPORTANT !!!
7110 * If you modify code here, check whether hmR0VmxCallRing3Callback() needs to be updated too.
7111 */
7112
7113 /* Save the guest state if necessary. */
7114 if ( fSaveGuestState
7115 && HMVMXCPU_GST_VALUE(pVCpu) != HMVMX_UPDATED_GUEST_ALL)
7116 {
7117 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
7118 AssertRCReturn(rc, rc);
7119 Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
7120 }
7121
7122 /* Restore host FPU state if necessary and resync on next R0 reentry .*/
7123 if (CPUMIsGuestFPUStateActive(pVCpu))
7124 {
7125 /* We shouldn't reload CR0 without saving it first. */
7126 if (!fSaveGuestState)
7127 {
7128 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
7129 AssertRCReturn(rc, rc);
7130 }
7131 CPUMR0SaveGuestFPU(pVM, pVCpu, pMixedCtx);
7132 Assert(!CPUMIsGuestFPUStateActive(pVCpu));
7133 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
7134 }
7135
7136 /* Restore host debug registers if necessary and resync on next R0 reentry. */
7137#ifdef VBOX_STRICT
7138 if (CPUMIsHyperDebugStateActive(pVCpu))
7139 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT);
7140#endif
7141 if (CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */))
7142 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
7143 Assert(!CPUMIsGuestDebugStateActive(pVCpu) && !CPUMIsGuestDebugStateActivePending(pVCpu));
7144 Assert(!CPUMIsHyperDebugStateActive(pVCpu) && !CPUMIsHyperDebugStateActivePending(pVCpu));
7145
7146#if HC_ARCH_BITS == 64
7147 /* Restore host-state bits that VT-x only restores partially. */
7148 if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
7149 && (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
7150 {
7151 Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags, idCpu));
7152 VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
7153 }
7154 pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
7155#endif
7156
7157#if HC_ARCH_BITS == 64
7158 /* Restore the lazy host MSRs as we're leaving VT-x context. */
7159 if ( pVM->hm.s.fAllow64BitGuests
7160 && pVCpu->hm.s.vmx.fLazyMsrs)
7161 {
7162 /* We shouldn't reload the guest MSRs without saving it first. */
7163 if (!fSaveGuestState)
7164 {
7165 int rc = hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
7166 AssertRCReturn(rc, rc);
7167 }
7168 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS));
7169 hmR0VmxLazyRestoreHostMsrs(pVCpu);
7170 Assert(!pVCpu->hm.s.vmx.fLazyMsrs);
7171 }
7172#endif
7173
7174 /* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
7175 pVCpu->hm.s.vmx.fUpdatedHostMsrs = false;
7176
7177 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
7178 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatLoadGuestState);
7179 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit1);
7180 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit2);
7181 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO);
7182 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx);
7183 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi);
7184 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
7185
7186 VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
7187
7188 /** @todo This partially defeats the purpose of having preemption hooks.
7189 * The problem is, deregistering the hooks should be moved to a place that
7190 * lasts until the EMT is about to be destroyed not everytime while leaving HM
7191 * context.
7192 */
7193 if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_ACTIVE)
7194 {
7195 int rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
7196 AssertRCReturn(rc, rc);
7197
7198 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
7199 Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu));
7200 }
7201 Assert(!(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_LAUNCHED));
7202 NOREF(idCpu);
7203
7204 return VINF_SUCCESS;
7205}
7206
7207
7208/**
7209 * Leaves the VT-x session.
7210 *
7211 * @returns VBox status code.
7212 * @param pVM Pointer to the VM.
7213 * @param pVCpu Pointer to the VMCPU.
7214 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7215 * out-of-sync. Make sure to update the required fields
7216 * before using them.
7217 *
7218 * @remarks No-long-jmp zone!!!
7219 */
7220DECLINLINE(int) hmR0VmxLeaveSession(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7221{
7222 HM_DISABLE_PREEMPT();
7223 HMVMX_ASSERT_CPU_SAFE();
7224 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
7225 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
7226
7227 /* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
7228 and done this from the VMXR0ThreadCtxCallback(). */
7229 if (!pVCpu->hm.s.fLeaveDone)
7230 {
7231 int rc2 = hmR0VmxLeave(pVM, pVCpu, pMixedCtx, true /* fSaveGuestState */);
7232 AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT(), rc2);
7233 pVCpu->hm.s.fLeaveDone = true;
7234 }
7235 Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
7236
7237 /*
7238 * !!! IMPORTANT !!!
7239 * If you modify code here, make sure to check whether hmR0VmxCallRing3Callback() needs to be updated too.
7240 */
7241
7242 /* Deregister hook now that we've left HM context before re-enabling preemption. */
7243 /** @todo Deregistering here means we need to VMCLEAR always
7244 * (longjmp/exit-to-r3) in VT-x which is not efficient. */
7245 /** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
7246 VMMR0ThreadCtxHookDisable(pVCpu);
7247
7248 /* Leave HM context. This takes care of local init (term). */
7249 int rc = HMR0LeaveCpu(pVCpu);
7250
7251 HM_RESTORE_PREEMPT();
7252 return rc;
7253}
7254
7255
7256/**
7257 * Does the necessary state syncing before doing a longjmp to ring-3.
7258 *
7259 * @returns VBox status code.
7260 * @param pVM Pointer to the VM.
7261 * @param pVCpu Pointer to the VMCPU.
7262 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7263 * out-of-sync. Make sure to update the required fields
7264 * before using them.
7265 *
7266 * @remarks No-long-jmp zone!!!
7267 */
7268DECLINLINE(int) hmR0VmxLongJmpToRing3(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7269{
7270 return hmR0VmxLeaveSession(pVM, pVCpu, pMixedCtx);
7271}
7272
7273
7274/**
7275 * Take necessary actions before going back to ring-3.
7276 *
7277 * An action requires us to go back to ring-3. This function does the necessary
7278 * steps before we can safely return to ring-3. This is not the same as longjmps
7279 * to ring-3, this is voluntary and prepares the guest so it may continue
7280 * executing outside HM (recompiler/IEM).
7281 *
7282 * @returns VBox status code.
7283 * @param pVM Pointer to the VM.
7284 * @param pVCpu Pointer to the VMCPU.
7285 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7286 * out-of-sync. Make sure to update the required fields
7287 * before using them.
7288 * @param rcExit The reason for exiting to ring-3. Can be
7289 * VINF_VMM_UNKNOWN_RING3_CALL.
7290 */
7291static int hmR0VmxExitToRing3(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, int rcExit)
7292{
7293 Assert(pVM);
7294 Assert(pVCpu);
7295 Assert(pMixedCtx);
7296 HMVMX_ASSERT_PREEMPT_SAFE();
7297
7298 if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR))
7299 {
7300 VMXGetActivatedVmcs(&pVCpu->hm.s.vmx.LastError.u64VMCSPhys);
7301 pVCpu->hm.s.vmx.LastError.u32VMCSRevision = *(uint32_t *)pVCpu->hm.s.vmx.pvVmcs;
7302 pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
7303 /* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */
7304 }
7305
7306 /* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
7307 VMMRZCallRing3Disable(pVCpu);
7308 Log4(("hmR0VmxExitToRing3: pVCpu=%p idCpu=%RU32 rcExit=%d\n", pVCpu, pVCpu->idCpu, rcExit));
7309
7310 /* We need to do this only while truly exiting the "inner loop" back to ring-3 and -not- for any longjmp to ring3. */
7311 if (pVCpu->hm.s.Event.fPending)
7312 {
7313 hmR0VmxPendingEventToTrpmTrap(pVCpu);
7314 Assert(!pVCpu->hm.s.Event.fPending);
7315 }
7316
7317 /* If we're emulating an instruction, we shouldn't have any TRPM traps pending
7318 and if we're injecting an event we should have a TRPM trap pending. */
7319 AssertMsg(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT || TRPMHasTrap(pVCpu), ("%Rrc\n", rcExit));
7320 AssertMsg(rcExit != VINF_EM_RAW_EMULATE_INSTR || !TRPMHasTrap(pVCpu), ("%Rrc\n", rcExit));
7321
7322 /* Save guest state and restore host state bits. */
7323 int rc = hmR0VmxLeaveSession(pVM, pVCpu, pMixedCtx);
7324 AssertRCReturn(rc, rc);
7325 STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
7326 /* Thread-context hooks are unregistered at this point!!! */
7327
7328 /* Sync recompiler state. */
7329 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
7330 CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
7331 | CPUM_CHANGED_LDTR
7332 | CPUM_CHANGED_GDTR
7333 | CPUM_CHANGED_IDTR
7334 | CPUM_CHANGED_TR
7335 | CPUM_CHANGED_HIDDEN_SEL_REGS);
7336 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
7337 if ( pVM->hm.s.fNestedPaging
7338 && CPUMIsGuestPagingEnabledEx(pMixedCtx))
7339 {
7340 CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
7341 }
7342
7343 Assert(!pVCpu->hm.s.fClearTrapFlag);
7344
7345 /* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
7346 if (rcExit != VINF_EM_RAW_INTERRUPT)
7347 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
7348
7349 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
7350
7351 /* We do -not- want any longjmp notifications after this! We must return to ring-3 ASAP. */
7352 VMMRZCallRing3RemoveNotification(pVCpu);
7353 VMMRZCallRing3Enable(pVCpu);
7354
7355 return rc;
7356}
7357
7358
7359/**
7360 * VMMRZCallRing3() callback wrapper which saves the guest state before we
7361 * longjump to ring-3 and possibly get preempted.
7362 *
7363 * @returns VBox status code.
7364 * @param pVCpu Pointer to the VMCPU.
7365 * @param enmOperation The operation causing the ring-3 longjump.
7366 * @param pvUser Opaque pointer to the guest-CPU context. The data
7367 * may be out-of-sync. Make sure to update the required
7368 * fields before using them.
7369 */
7370DECLCALLBACK(int) hmR0VmxCallRing3Callback(PVMCPU pVCpu, VMMCALLRING3 enmOperation, void *pvUser)
7371{
7372 if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION)
7373 {
7374 /*
7375 * !!! IMPORTANT !!!
7376 * If you modify code here, check whether hmR0VmxLeave() and hmR0VmxLeaveSession() needs to be updated too.
7377 * This is a stripped down version which gets out ASAP, trying to not trigger any further assertions.
7378 */
7379 VMMRZCallRing3RemoveNotification(pVCpu);
7380 VMMRZCallRing3Disable(pVCpu);
7381 RTTHREADPREEMPTSTATE PreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
7382 RTThreadPreemptDisable(&PreemptState);
7383
7384 PVM pVM = pVCpu->CTX_SUFF(pVM);
7385 if (CPUMIsGuestFPUStateActive(pVCpu))
7386 CPUMR0SaveGuestFPU(pVM, pVCpu, (PCPUMCTX)pvUser);
7387
7388 CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
7389
7390#if HC_ARCH_BITS == 64
7391 /* Restore host-state bits that VT-x only restores partially. */
7392 if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
7393 && (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
7394 VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
7395 pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
7396
7397 /* Restore the lazy host MSRs as we're leaving VT-x context. */
7398 if ( pVM->hm.s.fAllow64BitGuests
7399 && pVCpu->hm.s.vmx.fLazyMsrs)
7400 hmR0VmxLazyRestoreHostMsrs(pVCpu);
7401#endif
7402 /* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
7403 pVCpu->hm.s.vmx.fUpdatedHostMsrs = false;
7404 VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
7405 if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_ACTIVE)
7406 {
7407 VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
7408 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
7409 }
7410
7411 /** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
7412 VMMR0ThreadCtxHookDisable(pVCpu);
7413 HMR0LeaveCpu(pVCpu);
7414 RTThreadPreemptRestore(&PreemptState);
7415 return VINF_SUCCESS;
7416 }
7417
7418 Assert(pVCpu);
7419 Assert(pvUser);
7420 Assert(VMMRZCallRing3IsEnabled(pVCpu));
7421 HMVMX_ASSERT_PREEMPT_SAFE();
7422
7423 VMMRZCallRing3Disable(pVCpu);
7424 Assert(VMMR0IsLogFlushDisabled(pVCpu));
7425
7426 Log4(("hmR0VmxCallRing3Callback->hmR0VmxLongJmpToRing3 pVCpu=%p idCpu=%RU32 enmOperation=%d\n", pVCpu, pVCpu->idCpu,
7427 enmOperation));
7428
7429 int rc = hmR0VmxLongJmpToRing3(pVCpu->CTX_SUFF(pVM), pVCpu, (PCPUMCTX)pvUser);
7430 AssertRCReturn(rc, rc);
7431
7432 VMMRZCallRing3Enable(pVCpu);
7433 return VINF_SUCCESS;
7434}
7435
7436
7437/**
7438 * Sets the interrupt-window exiting control in the VMCS which instructs VT-x to
7439 * cause a VM-exit as soon as the guest is in a state to receive interrupts.
7440 *
7441 * @param pVCpu Pointer to the VMCPU.
7442 */
7443DECLINLINE(void) hmR0VmxSetIntWindowExitVmcs(PVMCPU pVCpu)
7444{
7445 if (RT_LIKELY(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT))
7446 {
7447 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT))
7448 {
7449 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT;
7450 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7451 AssertRC(rc);
7452 Log4(("Setup interrupt-window exiting\n"));
7453 }
7454 } /* else we will deliver interrupts whenever the guest exits next and is in a state to receive events. */
7455}
7456
7457
7458/**
7459 * Clears the interrupt-window exiting control in the VMCS.
7460 *
7461 * @param pVCpu Pointer to the VMCPU.
7462 */
7463DECLINLINE(void) hmR0VmxClearIntWindowExitVmcs(PVMCPU pVCpu)
7464{
7465 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT);
7466 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT;
7467 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7468 AssertRC(rc);
7469 Log4(("Cleared interrupt-window exiting\n"));
7470}
7471
7472
7473/**
7474 * Sets the NMI-window exiting control in the VMCS which instructs VT-x to
7475 * cause a VM-exit as soon as the guest is in a state to receive NMIs.
7476 *
7477 * @param pVCpu Pointer to the VMCPU.
7478 */
7479DECLINLINE(void) hmR0VmxSetNmiWindowExitVmcs(PVMCPU pVCpu)
7480{
7481 if (RT_LIKELY(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT))
7482 {
7483 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT))
7484 {
7485 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT;
7486 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7487 AssertRC(rc);
7488 Log4(("Setup NMI-window exiting\n"));
7489 }
7490 } /* else we will deliver NMIs whenever we VM-exit next, even possibly nesting NMIs. Can't be helped on ancient CPUs. */
7491}
7492
7493
7494/**
7495 * Clears the NMI-window exiting control in the VMCS.
7496 *
7497 * @param pVCpu Pointer to the VMCPU.
7498 */
7499DECLINLINE(void) hmR0VmxClearNmiWindowExitVmcs(PVMCPU pVCpu)
7500{
7501 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT);
7502 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT;
7503 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7504 AssertRC(rc);
7505 Log4(("Cleared NMI-window exiting\n"));
7506}
7507
7508
7509/**
7510 * Evaluates the event to be delivered to the guest and sets it as the pending
7511 * event.
7512 *
7513 * @param pVCpu Pointer to the VMCPU.
7514 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7515 * out-of-sync. Make sure to update the required fields
7516 * before using them.
7517 */
7518static void hmR0VmxEvaluatePendingEvent(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7519{
7520 Assert(!pVCpu->hm.s.Event.fPending);
7521
7522 /* Get the current interruptibility-state of the guest and then figure out what can be injected. */
7523 uint32_t const uIntrState = hmR0VmxGetGuestIntrState(pVCpu, pMixedCtx);
7524 bool const fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7525 bool const fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7526 bool const fBlockNmi = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI);
7527
7528 Assert(!fBlockSti || HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS));
7529 Assert(!(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
7530 Assert(!fBlockSti || pMixedCtx->eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
7531 Assert(!TRPMHasTrap(pVCpu));
7532
7533 /*
7534 * Toggling of interrupt force-flags here is safe since we update TRPM on premature exits
7535 * to ring-3 before executing guest code, see hmR0VmxExitToRing3(). We must NOT restore these force-flags.
7536 */
7537 /** @todo SMI. SMIs take priority over NMIs. */
7538 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_NMI)) /* NMI. NMIs take priority over regular interrupts. */
7539 {
7540 /* On some CPUs block-by-STI also blocks NMIs. See Intel spec. 26.3.1.5 "Checks On Guest Non-Register State". */
7541 if ( !fBlockNmi
7542 && !fBlockSti
7543 && !fBlockMovSS)
7544 {
7545 Log4(("Pending NMI vcpu[%RU32]\n", pVCpu->idCpu));
7546 uint32_t u32IntInfo = X86_XCPT_NMI | VMX_EXIT_INTERRUPTION_INFO_VALID;
7547 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7548
7549 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7550 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
7551 }
7552 else
7553 hmR0VmxSetNmiWindowExitVmcs(pVCpu);
7554 }
7555 /*
7556 * Check if the guest can receive external interrupts (PIC/APIC). Once we do PDMGetInterrupt() we -must- deliver
7557 * the interrupt ASAP. We must not execute any guest code until we inject the interrupt.
7558 */
7559 else if ( VMCPU_FF_IS_PENDING(pVCpu, (VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC))
7560 && !pVCpu->hm.s.fSingleInstruction)
7561 {
7562 int rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
7563 AssertRC(rc);
7564 bool const fBlockInt = !(pMixedCtx->eflags.u32 & X86_EFL_IF);
7565 if ( !fBlockInt
7566 && !fBlockSti
7567 && !fBlockMovSS)
7568 {
7569 uint8_t u8Interrupt;
7570 rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
7571 if (RT_SUCCESS(rc))
7572 {
7573 Log4(("Pending interrupt vcpu[%RU32] u8Interrupt=%#x \n", pVCpu->idCpu, u8Interrupt));
7574 uint32_t u32IntInfo = u8Interrupt | VMX_EXIT_INTERRUPTION_INFO_VALID;
7575 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7576
7577 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrfaultAddress */);
7578 }
7579 else
7580 {
7581 /** @todo Does this actually happen? If not turn it into an assertion. */
7582 Assert(!VMCPU_FF_IS_PENDING(pVCpu, (VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC)));
7583 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
7584 }
7585 }
7586 else
7587 hmR0VmxSetIntWindowExitVmcs(pVCpu);
7588 }
7589}
7590
7591
7592/**
7593 * Sets a pending-debug exception to be delivered to the guest if the guest is
7594 * single-stepping.
7595 *
7596 * @param pVCpu Pointer to the VMCPU.
7597 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7598 * out-of-sync. Make sure to update the required fields
7599 * before using them.
7600 */
7601DECLINLINE(void) hmR0VmxSetPendingDebugXcpt(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7602{
7603 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS));
7604 if (pMixedCtx->eflags.Bits.u1TF) /* We don't have any IA32_DEBUGCTL MSR for guests. Treat as all bits 0. */
7605 {
7606 int rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, VMX_VMCS_GUEST_DEBUG_EXCEPTIONS_BS);
7607 AssertRC(rc);
7608 }
7609}
7610
7611
7612/**
7613 * Injects any pending events into the guest if the guest is in a state to
7614 * receive them.
7615 *
7616 * @returns VBox status code (informational status codes included).
7617 * @param pVCpu Pointer to the VMCPU.
7618 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7619 * out-of-sync. Make sure to update the required fields
7620 * before using them.
7621 * @param fStepping Running in hmR0VmxRunGuestCodeStep() and we should
7622 * return VINF_EM_DBG_STEPPED if the event was
7623 * dispatched directly.
7624 */
7625static int hmR0VmxInjectPendingEvent(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fStepping)
7626{
7627 HMVMX_ASSERT_PREEMPT_SAFE();
7628 Assert(VMMRZCallRing3IsEnabled(pVCpu));
7629
7630 /* Get the current interruptibility-state of the guest and then figure out what can be injected. */
7631 uint32_t uIntrState = hmR0VmxGetGuestIntrState(pVCpu, pMixedCtx);
7632 bool fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7633 bool fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7634
7635 Assert(!fBlockSti || HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS));
7636 Assert(!(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
7637 Assert(!fBlockSti || pMixedCtx->eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
7638 Assert(!TRPMHasTrap(pVCpu));
7639
7640 int rc = VINF_SUCCESS;
7641 if (pVCpu->hm.s.Event.fPending)
7642 {
7643 /*
7644 * Clear any interrupt-window exiting control if we're going to inject an interrupt. Saves one extra
7645 * VM-exit in situations where we previously setup interrupt-window exiting but got other VM-exits and
7646 * ended up enabling interrupts outside VT-x.
7647 */
7648 uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
7649 if ( (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT)
7650 && uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
7651 {
7652 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT);
7653 hmR0VmxClearIntWindowExitVmcs(pVCpu);
7654 }
7655
7656#ifdef VBOX_STRICT
7657 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
7658 {
7659 bool const fBlockInt = !(pMixedCtx->eflags.u32 & X86_EFL_IF);
7660 Assert(!fBlockInt);
7661 Assert(!fBlockSti);
7662 Assert(!fBlockMovSS);
7663 }
7664 else if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
7665 {
7666 bool const fBlockNmi = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI);
7667 Assert(!fBlockSti);
7668 Assert(!fBlockMovSS);
7669 Assert(!fBlockNmi);
7670 }
7671#endif
7672 Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64 Type=%#x\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo,
7673 (uint8_t)uIntType));
7674 rc = hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, pVCpu->hm.s.Event.u64IntInfo, pVCpu->hm.s.Event.cbInstr,
7675 pVCpu->hm.s.Event.u32ErrCode, pVCpu->hm.s.Event.GCPtrFaultAddress, fStepping, &uIntrState);
7676 AssertRCReturn(rc, rc);
7677
7678 /* Update the interruptibility-state as it could have been changed by
7679 hmR0VmxInjectEventVmcs() (e.g. real-on-v86 guest injecting software interrupts) */
7680 fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7681 fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7682
7683#ifdef VBOX_WITH_STATISTICS
7684 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
7685 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
7686 else
7687 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
7688#endif
7689 }
7690
7691 /* Deliver pending debug exception if the guest is single-stepping. Evaluate and set the BS bit. */
7692 if ( fBlockSti
7693 || fBlockMovSS)
7694 {
7695 if ( !pVCpu->hm.s.fSingleInstruction
7696 && !DBGFIsStepping(pVCpu))
7697 {
7698 /*
7699 * The pending-debug exceptions field is cleared on all VM-exits except VMX_EXIT_TPR_BELOW_THRESHOLD,
7700 * VMX_EXIT_MTF, VMX_EXIT_APIC_WRITE and VMX_EXIT_VIRTUALIZED_EOI.
7701 * See Intel spec. 27.3.4 "Saving Non-Register State".
7702 */
7703 int rc2 = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
7704 AssertRCReturn(rc2, rc2);
7705 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
7706 }
7707 else if (pMixedCtx->eflags.Bits.u1TF)
7708 {
7709 /*
7710 * We are single-stepping in the hypervisor debugger using EFLAGS.TF. Clear interrupt inhibition as setting the
7711 * BS bit would mean delivering a #DB to the guest upon VM-entry when it shouldn't be.
7712 */
7713 Assert(!(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG));
7714 uIntrState = 0;
7715 }
7716 }
7717
7718 /*
7719 * There's no need to clear the VM-entry interruption-information field here if we're not injecting anything.
7720 * VT-x clears the valid bit on every VM-exit. See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
7721 */
7722 int rc2 = hmR0VmxLoadGuestIntrState(pVCpu, uIntrState);
7723 AssertRC(rc2);
7724
7725 Assert(rc == VINF_SUCCESS || rc == VINF_EM_RESET || (rc == VINF_EM_DBG_STEPPED && fStepping));
7726 NOREF(fBlockMovSS); NOREF(fBlockSti);
7727 return rc;
7728}
7729
7730
7731/**
7732 * Sets an invalid-opcode (#UD) exception as pending-for-injection into the VM.
7733 *
7734 * @param pVCpu Pointer to the VMCPU.
7735 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7736 * out-of-sync. Make sure to update the required fields
7737 * before using them.
7738 */
7739DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7740{
7741 NOREF(pMixedCtx);
7742 uint32_t u32IntInfo = X86_XCPT_UD | VMX_EXIT_INTERRUPTION_INFO_VALID;
7743 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7744}
7745
7746
7747/**
7748 * Injects a double-fault (#DF) exception into the VM.
7749 *
7750 * @returns VBox status code (informational status code included).
7751 * @param pVCpu Pointer to the VMCPU.
7752 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7753 * out-of-sync. Make sure to update the required fields
7754 * before using them.
7755 * @param fStepping Whether we're running in hmR0VmxRunGuestCodeStep()
7756 * and should return VINF_EM_DBG_STEPPED if the event
7757 * is injected directly (register modified by us, not
7758 * by hardware on VM-entry).
7759 * @param puIntrState Pointer to the current guest interruptibility-state.
7760 * This interruptibility-state will be updated if
7761 * necessary. This cannot not be NULL.
7762 */
7763DECLINLINE(int) hmR0VmxInjectXcptDF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fStepping, uint32_t *puIntrState)
7764{
7765 uint32_t u32IntInfo = X86_XCPT_DF | VMX_EXIT_INTERRUPTION_INFO_VALID;
7766 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7767 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7768 return hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */,
7769 fStepping, puIntrState);
7770}
7771
7772
7773/**
7774 * Sets a debug (#DB) exception as pending-for-injection into the VM.
7775 *
7776 * @param pVCpu Pointer to the VMCPU.
7777 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7778 * out-of-sync. Make sure to update the required fields
7779 * before using them.
7780 */
7781DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7782{
7783 NOREF(pMixedCtx);
7784 uint32_t u32IntInfo = X86_XCPT_DB | VMX_EXIT_INTERRUPTION_INFO_VALID;
7785 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7786 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7787}
7788
7789
7790/**
7791 * Sets an overflow (#OF) exception as pending-for-injection into the VM.
7792 *
7793 * @param pVCpu Pointer to the VMCPU.
7794 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7795 * out-of-sync. Make sure to update the required fields
7796 * before using them.
7797 * @param cbInstr The value of RIP that is to be pushed on the guest
7798 * stack.
7799 */
7800DECLINLINE(void) hmR0VmxSetPendingXcptOF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t cbInstr)
7801{
7802 NOREF(pMixedCtx);
7803 uint32_t u32IntInfo = X86_XCPT_OF | VMX_EXIT_INTERRUPTION_INFO_VALID;
7804 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7805 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7806}
7807
7808
7809/**
7810 * Injects a general-protection (#GP) fault into the VM.
7811 *
7812 * @returns VBox status code (informational status code included).
7813 * @param pVCpu Pointer to the VMCPU.
7814 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7815 * out-of-sync. Make sure to update the required fields
7816 * before using them.
7817 * @param fErrorCodeValid Whether the error code is valid (depends on the CPU
7818 * mode, i.e. in real-mode it's not valid).
7819 * @param u32ErrorCode The error code associated with the #GP.
7820 * @param fStepping Whether we're running in
7821 * hmR0VmxRunGuestCodeStep() and should return
7822 * VINF_EM_DBG_STEPPED if the event is injected
7823 * directly (register modified by us, not by
7824 * hardware on VM-entry).
7825 * @param puIntrState Pointer to the current guest interruptibility-state.
7826 * This interruptibility-state will be updated if
7827 * necessary. This cannot not be NULL.
7828 */
7829DECLINLINE(int) hmR0VmxInjectXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fErrorCodeValid, uint32_t u32ErrorCode,
7830 bool fStepping, uint32_t *puIntrState)
7831{
7832 uint32_t u32IntInfo = X86_XCPT_GP | VMX_EXIT_INTERRUPTION_INFO_VALID;
7833 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7834 if (fErrorCodeValid)
7835 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7836 return hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, u32IntInfo, 0 /* cbInstr */, u32ErrorCode, 0 /* GCPtrFaultAddress */,
7837 fStepping, puIntrState);
7838}
7839
7840
7841/**
7842 * Sets a general-protection (#GP) exception as pending-for-injection into the
7843 * VM.
7844 *
7845 * @param pVCpu Pointer to the VMCPU.
7846 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7847 * out-of-sync. Make sure to update the required fields
7848 * before using them.
7849 * @param u32ErrorCode The error code associated with the #GP.
7850 */
7851DECLINLINE(void) hmR0VmxSetPendingXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t u32ErrorCode)
7852{
7853 NOREF(pMixedCtx);
7854 uint32_t u32IntInfo = X86_XCPT_GP | VMX_EXIT_INTERRUPTION_INFO_VALID;
7855 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7856 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7857 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, u32ErrorCode, 0 /* GCPtrFaultAddress */);
7858}
7859
7860
7861/**
7862 * Sets a software interrupt (INTn) as pending-for-injection into the VM.
7863 *
7864 * @param pVCpu Pointer to the VMCPU.
7865 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7866 * out-of-sync. Make sure to update the required fields
7867 * before using them.
7868 * @param uVector The software interrupt vector number.
7869 * @param cbInstr The value of RIP that is to be pushed on the guest
7870 * stack.
7871 */
7872DECLINLINE(void) hmR0VmxSetPendingIntN(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint16_t uVector, uint32_t cbInstr)
7873{
7874 NOREF(pMixedCtx);
7875 uint32_t u32IntInfo = uVector | VMX_EXIT_INTERRUPTION_INFO_VALID;
7876 if ( uVector == X86_XCPT_BP
7877 || uVector == X86_XCPT_OF)
7878 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7879 else
7880 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7881 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7882}
7883
7884
7885/**
7886 * Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's
7887 * stack.
7888 *
7889 * @returns VBox status code (information status code included).
7890 * @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault.
7891 * @param pVM Pointer to the VM.
7892 * @param pMixedCtx Pointer to the guest-CPU context.
7893 * @param uValue The value to push to the guest stack.
7894 */
7895DECLINLINE(int) hmR0VmxRealModeGuestStackPush(PVM pVM, PCPUMCTX pMixedCtx, uint16_t uValue)
7896{
7897 /*
7898 * The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in
7899 * virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
7900 * See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound".
7901 */
7902 if (pMixedCtx->sp == 1)
7903 return VINF_EM_RESET;
7904 pMixedCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */
7905 int rc = PGMPhysSimpleWriteGCPhys(pVM, pMixedCtx->ss.u64Base + pMixedCtx->sp, &uValue, sizeof(uint16_t));
7906 AssertRCReturn(rc, rc);
7907 return rc;
7908}
7909
7910
7911/**
7912 * Injects an event into the guest upon VM-entry by updating the relevant fields
7913 * in the VM-entry area in the VMCS.
7914 *
7915 * @returns VBox status code (informational error codes included).
7916 * @retval VINF_SUCCESS if the event is successfully injected into the VMCS.
7917 * @retval VINF_EM_RESET if event injection resulted in a triple-fault.
7918 *
7919 * @param pVCpu Pointer to the VMCPU.
7920 * @param pMixedCtx Pointer to the guest-CPU context. The data may
7921 * be out-of-sync. Make sure to update the required
7922 * fields before using them.
7923 * @param u64IntInfo The VM-entry interruption-information field.
7924 * @param cbInstr The VM-entry instruction length in bytes (for
7925 * software interrupts, exceptions and privileged
7926 * software exceptions).
7927 * @param u32ErrCode The VM-entry exception error code.
7928 * @param GCPtrFaultAddress The page-fault address for #PF exceptions.
7929 * @param puIntrState Pointer to the current guest interruptibility-state.
7930 * This interruptibility-state will be updated if
7931 * necessary. This cannot not be NULL.
7932 * @param fStepping Whether we're running in
7933 * hmR0VmxRunGuestCodeStep() and should return
7934 * VINF_EM_DBG_STEPPED if the event is injected
7935 * directly (register modified by us, not by
7936 * hardware on VM-entry).
7937 *
7938 * @remarks Requires CR0!
7939 * @remarks No-long-jump zone!!!
7940 */
7941static int hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint64_t u64IntInfo, uint32_t cbInstr,
7942 uint32_t u32ErrCode, RTGCUINTREG GCPtrFaultAddress, bool fStepping, uint32_t *puIntrState)
7943{
7944 /* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */
7945 AssertMsg(u64IntInfo >> 32 == 0, ("%#RX64\n", u64IntInfo));
7946 Assert(puIntrState);
7947 uint32_t u32IntInfo = (uint32_t)u64IntInfo;
7948
7949 uint32_t const uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(u32IntInfo);
7950 uint32_t const uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(u32IntInfo);
7951
7952#ifdef VBOX_STRICT
7953 /* Validate the error-code-valid bit for hardware exceptions. */
7954 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT)
7955 {
7956 switch (uVector)
7957 {
7958 case X86_XCPT_PF:
7959 case X86_XCPT_DF:
7960 case X86_XCPT_TS:
7961 case X86_XCPT_NP:
7962 case X86_XCPT_SS:
7963 case X86_XCPT_GP:
7964 case X86_XCPT_AC:
7965 AssertMsg(VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(u32IntInfo),
7966 ("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector));
7967 /* fallthru */
7968 default:
7969 break;
7970 }
7971 }
7972#endif
7973
7974 /* Cannot inject an NMI when block-by-MOV SS is in effect. */
7975 Assert( uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI
7976 || !(*puIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS));
7977
7978 STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedIrqsR0[uVector & MASK_INJECT_IRQ_STAT]);
7979
7980 /* We require CR0 to check if the guest is in real-mode. */
7981 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
7982 AssertRCReturn(rc, rc);
7983
7984 /*
7985 * Hardware interrupts & exceptions cannot be delivered through the software interrupt redirection bitmap to the real
7986 * mode task in virtual-8086 mode. We must jump to the interrupt handler in the (real-mode) guest.
7987 * See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode" for interrupt & exception classes.
7988 * See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling.
7989 */
7990 if (CPUMIsGuestInRealModeEx(pMixedCtx))
7991 {
7992 PVM pVM = pVCpu->CTX_SUFF(pVM);
7993 if (!pVM->hm.s.vmx.fUnrestrictedGuest)
7994 {
7995 Assert(PDMVmmDevHeapIsEnabled(pVM));
7996 Assert(pVM->hm.s.vmx.pRealModeTSS);
7997
7998 /* We require RIP, RSP, RFLAGS, CS, IDTR. Save the required ones from the VMCS. */
7999 rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
8000 rc |= hmR0VmxSaveGuestTableRegs(pVCpu, pMixedCtx);
8001 rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
8002 AssertRCReturn(rc, rc);
8003 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP));
8004
8005 /* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */
8006 size_t const cbIdtEntry = sizeof(X86IDTR16);
8007 if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pMixedCtx->idtr.cbIdt)
8008 {
8009 /* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */
8010 if (uVector == X86_XCPT_DF)
8011 return VINF_EM_RESET;
8012
8013 /* If we're injecting a #GP with no valid IDT entry, inject a double-fault. */
8014 if (uVector == X86_XCPT_GP)
8015 return hmR0VmxInjectXcptDF(pVCpu, pMixedCtx, fStepping, puIntrState);
8016
8017 /* If we're injecting an interrupt/exception with no valid IDT entry, inject a general-protection fault. */
8018 /* No error codes for exceptions in real-mode. See Intel spec. 20.1.4 "Interrupt and Exception Handling" */
8019 return hmR0VmxInjectXcptGP(pVCpu, pMixedCtx, false /* fErrCodeValid */, 0 /* u32ErrCode */,
8020 fStepping, puIntrState);
8021 }
8022
8023 /* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */
8024 uint16_t uGuestIp = pMixedCtx->ip;
8025 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT)
8026 {
8027 Assert(uVector == X86_XCPT_BP || uVector == X86_XCPT_OF);
8028 /* #BP and #OF are both benign traps, we need to resume the next instruction. */
8029 uGuestIp = pMixedCtx->ip + (uint16_t)cbInstr;
8030 }
8031 else if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT)
8032 uGuestIp = pMixedCtx->ip + (uint16_t)cbInstr;
8033
8034 /* Get the code segment selector and offset from the IDT entry for the interrupt handler. */
8035 X86IDTR16 IdtEntry;
8036 RTGCPHYS GCPhysIdtEntry = (RTGCPHYS)pMixedCtx->idtr.pIdt + uVector * cbIdtEntry;
8037 rc = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry);
8038 AssertRCReturn(rc, rc);
8039
8040 /* Construct the stack frame for the interrupt/exception handler. */
8041 rc = hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, pMixedCtx->eflags.u32);
8042 rc |= hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, pMixedCtx->cs.Sel);
8043 rc |= hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, uGuestIp);
8044 AssertRCReturn(rc, rc);
8045
8046 /* Clear the required eflag bits and jump to the interrupt/exception handler. */
8047 if (rc == VINF_SUCCESS)
8048 {
8049 pMixedCtx->eflags.u32 &= ~(X86_EFL_IF | X86_EFL_TF | X86_EFL_RF | X86_EFL_AC);
8050 pMixedCtx->rip = IdtEntry.offSel;
8051 pMixedCtx->cs.Sel = IdtEntry.uSel;
8052 pMixedCtx->cs.ValidSel = IdtEntry.uSel;
8053 pMixedCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry;
8054 if ( uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT
8055 && uVector == X86_XCPT_PF)
8056 pMixedCtx->cr2 = GCPtrFaultAddress;
8057
8058 /* If any other guest-state bits are changed here, make sure to update
8059 hmR0VmxPreRunGuestCommitted() when thread-context hooks are used. */
8060 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS
8061 | HM_CHANGED_GUEST_RIP
8062 | HM_CHANGED_GUEST_RFLAGS
8063 | HM_CHANGED_GUEST_RSP);
8064
8065 /* We're clearing interrupts, which means no block-by-STI interrupt-inhibition. */
8066 if (*puIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
8067 {
8068 Assert( uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI
8069 && uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT);
8070 Log4(("Clearing inhibition due to STI.\n"));
8071 *puIntrState &= ~VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI;
8072 }
8073 Log4(("Injecting real-mode: u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x Eflags=%#x CS:EIP=%04x:%04x\n",
8074 u32IntInfo, u32ErrCode, cbInstr, pMixedCtx->eflags.u, pMixedCtx->cs.Sel, pMixedCtx->eip));
8075
8076 /* The event has been truly dispatched. Mark it as no longer pending so we don't attempt to 'undo'
8077 it, if we are returning to ring-3 before executing guest code. */
8078 pVCpu->hm.s.Event.fPending = false;
8079
8080 /* Make hmR0VmxPreRunGuest return if we're stepping since we've changed cs:rip. */
8081 if (fStepping)
8082 rc = VINF_EM_DBG_STEPPED;
8083 }
8084 Assert(rc == VINF_SUCCESS || rc == VINF_EM_RESET || (rc == VINF_EM_DBG_STEPPED && fStepping));
8085 return rc;
8086 }
8087
8088 /*
8089 * For unrestricted execution enabled CPUs running real-mode guests, we must not set the deliver-error-code bit.
8090 * See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
8091 */
8092 u32IntInfo &= ~VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
8093 }
8094
8095 /* Validate. */
8096 Assert(VMX_EXIT_INTERRUPTION_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */
8097 Assert(!VMX_EXIT_INTERRUPTION_INFO_NMI_UNBLOCK_IRET(u32IntInfo)); /* Bit 12 MBZ. */
8098 Assert(!(u32IntInfo & 0x7ffff000)); /* Bits 30:12 MBZ. */
8099
8100 /* Inject. */
8101 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo);
8102 if (VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(u32IntInfo))
8103 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode);
8104 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr);
8105
8106 if ( VMX_EXIT_INTERRUPTION_INFO_TYPE(u32IntInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT
8107 && uVector == X86_XCPT_PF)
8108 pMixedCtx->cr2 = GCPtrFaultAddress;
8109
8110 Log4(("Injecting vcpu[%RU32] u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x pMixedCtx->uCR2=%#RX64\n", pVCpu->idCpu,
8111 u32IntInfo, u32ErrCode, cbInstr, pMixedCtx->cr2));
8112
8113 AssertRCReturn(rc, rc);
8114 return rc;
8115}
8116
8117
8118/**
8119 * Clears the interrupt-window exiting control in the VMCS and if necessary
8120 * clears the current event in the VMCS as well.
8121 *
8122 * @returns VBox status code.
8123 * @param pVCpu Pointer to the VMCPU.
8124 *
8125 * @remarks Use this function only to clear events that have not yet been
8126 * delivered to the guest but are injected in the VMCS!
8127 * @remarks No-long-jump zone!!!
8128 */
8129static void hmR0VmxClearEventVmcs(PVMCPU pVCpu)
8130{
8131 int rc;
8132 Log4Func(("vcpu[%d]\n", pVCpu->idCpu));
8133
8134 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT)
8135 hmR0VmxClearIntWindowExitVmcs(pVCpu);
8136
8137 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT)
8138 hmR0VmxClearNmiWindowExitVmcs(pVCpu);
8139
8140 if (!pVCpu->hm.s.Event.fPending)
8141 return;
8142
8143#ifdef VBOX_STRICT
8144 uint32_t u32EntryInfo;
8145 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
8146 AssertRC(rc);
8147 Assert(VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo));
8148#endif
8149
8150 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0);
8151 AssertRC(rc);
8152
8153 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, 0);
8154 AssertRC(rc);
8155
8156 /* We deliberately don't clear "hm.s.Event.fPending" here, it's taken
8157 care of in hmR0VmxExitToRing3() converting the pending event to TRPM. */
8158}
8159
8160
8161/**
8162 * Enters the VT-x session.
8163 *
8164 * @returns VBox status code.
8165 * @param pVM Pointer to the VM.
8166 * @param pVCpu Pointer to the VMCPU.
8167 * @param pCpu Pointer to the CPU info struct.
8168 */
8169VMMR0DECL(int) VMXR0Enter(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
8170{
8171 AssertPtr(pVM);
8172 AssertPtr(pVCpu);
8173 Assert(pVM->hm.s.vmx.fSupported);
8174 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8175 NOREF(pCpu); NOREF(pVM);
8176
8177 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8178 Assert(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE));
8179
8180#ifdef VBOX_STRICT
8181 /* Make sure we're in VMX root mode. */
8182 RTCCUINTREG u32HostCR4 = ASMGetCR4();
8183 if (!(u32HostCR4 & X86_CR4_VMXE))
8184 {
8185 LogRel(("VMXR0Enter: X86_CR4_VMXE bit in CR4 is not set!\n"));
8186 return VERR_VMX_X86_CR4_VMXE_CLEARED;
8187 }
8188#endif
8189
8190 /*
8191 * Load the VCPU's VMCS as the current (and active) one.
8192 */
8193 Assert(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_CLEAR);
8194 int rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
8195 if (RT_FAILURE(rc))
8196 return rc;
8197
8198 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE;
8199 pVCpu->hm.s.fLeaveDone = false;
8200 Log4Func(("Activated Vmcs. HostCpuId=%u\n", RTMpCpuId()));
8201
8202 return VINF_SUCCESS;
8203}
8204
8205
8206/**
8207 * The thread-context callback (only on platforms which support it).
8208 *
8209 * @param enmEvent The thread-context event.
8210 * @param pVCpu Pointer to the VMCPU.
8211 * @param fGlobalInit Whether global VT-x/AMD-V init. was used.
8212 * @thread EMT(pVCpu)
8213 */
8214VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPU pVCpu, bool fGlobalInit)
8215{
8216 NOREF(fGlobalInit);
8217
8218 switch (enmEvent)
8219 {
8220 case RTTHREADCTXEVENT_OUT:
8221 {
8222 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8223 Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
8224 VMCPU_ASSERT_EMT(pVCpu);
8225
8226 PVM pVM = pVCpu->CTX_SUFF(pVM);
8227 PCPUMCTX pMixedCtx = CPUMQueryGuestCtxPtr(pVCpu);
8228
8229 /* No longjmps (logger flushes, locks) in this fragile context. */
8230 VMMRZCallRing3Disable(pVCpu);
8231 Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId()));
8232
8233 /*
8234 * Restore host-state (FPU, debug etc.)
8235 */
8236 if (!pVCpu->hm.s.fLeaveDone)
8237 {
8238 /* Do -not- save guest-state here as we might already be in the middle of saving it (esp. bad if we are
8239 holding the PGM lock while saving the guest state (see hmR0VmxSaveGuestControlRegs()). */
8240 hmR0VmxLeave(pVM, pVCpu, pMixedCtx, false /* fSaveGuestState */);
8241 pVCpu->hm.s.fLeaveDone = true;
8242 }
8243
8244 /* Leave HM context, takes care of local init (term). */
8245 int rc = HMR0LeaveCpu(pVCpu);
8246 AssertRC(rc); NOREF(rc);
8247
8248 /* Restore longjmp state. */
8249 VMMRZCallRing3Enable(pVCpu);
8250 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
8251 break;
8252 }
8253
8254 case RTTHREADCTXEVENT_IN:
8255 {
8256 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8257 Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
8258 VMCPU_ASSERT_EMT(pVCpu);
8259
8260 /* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */
8261 VMMRZCallRing3Disable(pVCpu);
8262 Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId()));
8263
8264 /* Initialize the bare minimum state required for HM. This takes care of
8265 initializing VT-x if necessary (onlined CPUs, local init etc.) */
8266 int rc = HMR0EnterCpu(pVCpu);
8267 AssertRC(rc);
8268 Assert(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE));
8269
8270 /* Load the active VMCS as the current one. */
8271 if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_CLEAR)
8272 {
8273 rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
8274 AssertRC(rc); NOREF(rc);
8275 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE;
8276 Log4Func(("Resumed: Activated Vmcs. HostCpuId=%u\n", RTMpCpuId()));
8277 }
8278 pVCpu->hm.s.fLeaveDone = false;
8279
8280 /* Restore longjmp state. */
8281 VMMRZCallRing3Enable(pVCpu);
8282 break;
8283 }
8284
8285 default:
8286 break;
8287 }
8288}
8289
8290
8291/**
8292 * Saves the host state in the VMCS host-state.
8293 * Sets up the VM-exit MSR-load area.
8294 *
8295 * The CPU state will be loaded from these fields on every successful VM-exit.
8296 *
8297 * @returns VBox status code.
8298 * @param pVM Pointer to the VM.
8299 * @param pVCpu Pointer to the VMCPU.
8300 *
8301 * @remarks No-long-jump zone!!!
8302 */
8303static int hmR0VmxSaveHostState(PVM pVM, PVMCPU pVCpu)
8304{
8305 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8306
8307 if (!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT))
8308 return VINF_SUCCESS;
8309
8310 int rc = hmR0VmxSaveHostControlRegs(pVM, pVCpu);
8311 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostControlRegisters failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8312
8313 rc = hmR0VmxSaveHostSegmentRegs(pVM, pVCpu);
8314 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostSegmentRegisters failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8315
8316 rc = hmR0VmxSaveHostMsrs(pVM, pVCpu);
8317 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostMsrs failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8318
8319 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_HOST_CONTEXT);
8320 return rc;
8321}
8322
8323
8324/**
8325 * Saves the host state in the VMCS host-state.
8326 *
8327 * @returns VBox status code.
8328 * @param pVM Pointer to the VM.
8329 * @param pVCpu Pointer to the VMCPU.
8330 *
8331 * @remarks No-long-jump zone!!!
8332 */
8333VMMR0DECL(int) VMXR0SaveHostState(PVM pVM, PVMCPU pVCpu)
8334{
8335 AssertPtr(pVM);
8336 AssertPtr(pVCpu);
8337
8338 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8339
8340 /* Save the host state here while entering HM context. When thread-context hooks are used, we might get preempted
8341 and have to resave the host state but most of the time we won't be, so do it here before we disable interrupts. */
8342 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8343 return hmR0VmxSaveHostState(pVM, pVCpu);
8344}
8345
8346
8347/**
8348 * Loads the guest state into the VMCS guest-state area.
8349 *
8350 * The will typically be done before VM-entry when the guest-CPU state and the
8351 * VMCS state may potentially be out of sync.
8352 *
8353 * Sets up the VM-entry MSR-load and VM-exit MSR-store areas. Sets up the
8354 * VM-entry controls.
8355 * Sets up the appropriate VMX non-root function to execute guest code based on
8356 * the guest CPU mode.
8357 *
8358 * @returns VBox status code.
8359 * @param pVM Pointer to the VM.
8360 * @param pVCpu Pointer to the VMCPU.
8361 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8362 * out-of-sync. Make sure to update the required fields
8363 * before using them.
8364 *
8365 * @remarks No-long-jump zone!!!
8366 */
8367static int hmR0VmxLoadGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
8368{
8369 AssertPtr(pVM);
8370 AssertPtr(pVCpu);
8371 AssertPtr(pMixedCtx);
8372 HMVMX_ASSERT_PREEMPT_SAFE();
8373
8374 VMMRZCallRing3Disable(pVCpu);
8375 Assert(VMMR0IsLogFlushDisabled(pVCpu));
8376
8377 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8378
8379 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestState, x);
8380
8381 /* Determine real-on-v86 mode. */
8382 pVCpu->hm.s.vmx.RealMode.fRealOnV86Active = false;
8383 if ( !pVM->hm.s.vmx.fUnrestrictedGuest
8384 && CPUMIsGuestInRealModeEx(pMixedCtx))
8385 {
8386 pVCpu->hm.s.vmx.RealMode.fRealOnV86Active = true;
8387 }
8388
8389 /*
8390 * Load the guest-state into the VMCS.
8391 * Any ordering dependency among the sub-functions below must be explicitly stated using comments.
8392 * Ideally, assert that the cross-dependent bits are up-to-date at the point of using it.
8393 */
8394 int rc = hmR0VmxSetupVMRunHandler(pVCpu, pMixedCtx);
8395 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSetupVMRunHandler! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8396
8397 /* This needs to be done after hmR0VmxSetupVMRunHandler() as changing pfnStartVM may require VM-entry control updates. */
8398 rc = hmR0VmxLoadGuestEntryCtls(pVCpu, pMixedCtx);
8399 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestEntryCtls! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8400
8401 /* This needs to be done after hmR0VmxSetupVMRunHandler() as changing pfnStartVM may require VM-exit control updates. */
8402 rc = hmR0VmxLoadGuestExitCtls(pVCpu, pMixedCtx);
8403 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSetupExitCtls failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8404
8405 rc = hmR0VmxLoadGuestActivityState(pVCpu, pMixedCtx);
8406 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestActivityState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8407
8408 rc = hmR0VmxLoadGuestCR3AndCR4(pVCpu, pMixedCtx);
8409 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestCR3AndCR4: rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8410
8411 /* Assumes pMixedCtx->cr0 is up-to-date (strict builds require CR0 for segment register validation checks). */
8412 rc = hmR0VmxLoadGuestSegmentRegs(pVCpu, pMixedCtx);
8413 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestSegmentRegs: rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8414
8415 /* This needs to be done after hmR0VmxLoadGuestEntryCtls() and hmR0VmxLoadGuestExitCtls() as it may alter controls if we
8416 determine we don't have to swap EFER after all. */
8417 rc = hmR0VmxLoadGuestMsrs(pVCpu, pMixedCtx);
8418 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadSharedMsrs! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8419
8420 rc = hmR0VmxLoadGuestApicState(pVCpu, pMixedCtx);
8421 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestApicState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8422
8423 rc = hmR0VmxLoadGuestXcptIntercepts(pVCpu, pMixedCtx);
8424 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestXcptIntercepts! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8425
8426 /*
8427 * Loading Rflags here is fine, even though Rflags.TF might depend on guest debug state (which is not loaded here).
8428 * It is re-evaluated and updated if necessary in hmR0VmxLoadSharedState().
8429 */
8430 rc = hmR0VmxLoadGuestRipRspRflags(pVCpu, pMixedCtx);
8431 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestRipRspRflags! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8432
8433 /* Clear any unused and reserved bits. */
8434 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR2);
8435
8436 VMMRZCallRing3Enable(pVCpu);
8437
8438 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestState, x);
8439 return rc;
8440}
8441
8442
8443/**
8444 * Loads the state shared between the host and guest into the VMCS.
8445 *
8446 * @param pVM Pointer to the VM.
8447 * @param pVCpu Pointer to the VMCPU.
8448 * @param pCtx Pointer to the guest-CPU context.
8449 *
8450 * @remarks No-long-jump zone!!!
8451 */
8452static void hmR0VmxLoadSharedState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
8453{
8454 NOREF(pVM);
8455
8456 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8457 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8458
8459 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0))
8460 {
8461 int rc = hmR0VmxLoadSharedCR0(pVCpu, pCtx);
8462 AssertRC(rc);
8463 }
8464
8465 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG))
8466 {
8467 int rc = hmR0VmxLoadSharedDebugState(pVCpu, pCtx);
8468 AssertRC(rc);
8469
8470 /* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */
8471 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS))
8472 {
8473 rc = hmR0VmxLoadGuestRflags(pVCpu, pCtx);
8474 AssertRC(rc);
8475 }
8476 }
8477
8478 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS))
8479 {
8480#if HC_ARCH_BITS == 64
8481 if (pVM->hm.s.fAllow64BitGuests)
8482 hmR0VmxLazyLoadGuestMsrs(pVCpu, pCtx);
8483#endif
8484 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS);
8485 }
8486
8487 /* Loading CR0, debug state might have changed intercepts, update VMCS. */
8488 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS))
8489 {
8490 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
8491 AssertRC(rc);
8492 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
8493 }
8494
8495 AssertMsg(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE),
8496 ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8497}
8498
8499
8500/**
8501 * Worker for loading the guest-state bits in the inner VT-x execution loop.
8502 *
8503 * @param pVM Pointer to the VM.
8504 * @param pVCpu Pointer to the VMCPU.
8505 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8506 * out-of-sync. Make sure to update the required fields
8507 * before using them.
8508 */
8509DECLINLINE(void) hmR0VmxLoadGuestStateOptimal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
8510{
8511 HMVMX_ASSERT_PREEMPT_SAFE();
8512
8513 Log5(("LoadFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8514#ifdef HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
8515 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
8516#endif
8517
8518 if (HMCPU_CF_IS_SET_ONLY(pVCpu, HM_CHANGED_GUEST_RIP))
8519 {
8520 int rc = hmR0VmxLoadGuestRip(pVCpu, pMixedCtx);
8521 AssertRC(rc);
8522 STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadMinimal);
8523 }
8524 else if (HMCPU_CF_VALUE(pVCpu))
8525 {
8526 int rc = hmR0VmxLoadGuestState(pVM, pVCpu, pMixedCtx);
8527 AssertRC(rc);
8528 STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadFull);
8529 }
8530
8531 /* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */
8532 AssertMsg( !HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_ALL_GUEST)
8533 || HMCPU_CF_IS_PENDING_ONLY(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE),
8534 ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8535}
8536
8537
8538/**
8539 * Does the preparations before executing guest code in VT-x.
8540 *
8541 * This may cause longjmps to ring-3 and may even result in rescheduling to the
8542 * recompiler/IEM. We must be cautious what we do here regarding committing
8543 * guest-state information into the VMCS assuming we assuredly execute the
8544 * guest in VT-x mode.
8545 *
8546 * If we fall back to the recompiler/IEM after updating the VMCS and clearing
8547 * the common-state (TRPM/forceflags), we must undo those changes so that the
8548 * recompiler/IEM can (and should) use them when it resumes guest execution.
8549 * Otherwise such operations must be done when we can no longer exit to ring-3.
8550 *
8551 * @returns Strict VBox status code.
8552 * @retval VINF_SUCCESS if we can proceed with running the guest, interrupts
8553 * have been disabled.
8554 * @retval VINF_EM_RESET if a triple-fault occurs while injecting a
8555 * double-fault into the guest.
8556 * @retval VINF_EM_DBG_STEPPED if @a fStepping is true and an event was
8557 * dispatched directly.
8558 * @retval VINF_* scheduling changes, we have to go back to ring-3.
8559 *
8560 * @param pVM Pointer to the VM.
8561 * @param pVCpu Pointer to the VMCPU.
8562 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8563 * out-of-sync. Make sure to update the required fields
8564 * before using them.
8565 * @param pVmxTransient Pointer to the VMX transient structure.
8566 * @param fStepping Set if called from hmR0VmxRunGuestCodeStep(). Makes
8567 * us ignore some of the reasons for returning to
8568 * ring-3, and return VINF_EM_DBG_STEPPED if event
8569 * dispatching took place.
8570 */
8571static int hmR0VmxPreRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, bool fStepping)
8572{
8573 Assert(VMMRZCallRing3IsEnabled(pVCpu));
8574
8575#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
8576 PGMRZDynMapFlushAutoSet(pVCpu);
8577#endif
8578
8579 /* Check force flag actions that might require us to go back to ring-3. */
8580 int rc = hmR0VmxCheckForceFlags(pVM, pVCpu, pMixedCtx);
8581 if (rc != VINF_SUCCESS)
8582 return rc;
8583
8584#ifndef IEM_VERIFICATION_MODE_FULL
8585 /* Setup the Virtualized APIC accesses. pMixedCtx->msrApicBase is always up-to-date. It's not part of the VMCS. */
8586 if ( pVCpu->hm.s.vmx.u64MsrApicBase != pMixedCtx->msrApicBase
8587 && (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC))
8588 {
8589 Assert(pVM->hm.s.vmx.HCPhysApicAccess);
8590 RTGCPHYS GCPhysApicBase;
8591 GCPhysApicBase = pMixedCtx->msrApicBase;
8592 GCPhysApicBase &= PAGE_BASE_GC_MASK;
8593
8594 /* Unalias any existing mapping. */
8595 rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase);
8596 AssertRCReturn(rc, rc);
8597
8598 /* Map the HC APIC-access page into the GC space, this also updates the shadow page tables if necessary. */
8599 Log4(("Mapped HC APIC-access page into GC: GCPhysApicBase=%#RGv\n", GCPhysApicBase));
8600 rc = IOMMMIOMapMMIOHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hm.s.vmx.HCPhysApicAccess, X86_PTE_RW | X86_PTE_P);
8601 AssertRCReturn(rc, rc);
8602
8603 pVCpu->hm.s.vmx.u64MsrApicBase = pMixedCtx->msrApicBase;
8604 }
8605#endif /* !IEM_VERIFICATION_MODE_FULL */
8606
8607 if (TRPMHasTrap(pVCpu))
8608 hmR0VmxTrpmTrapToPendingEvent(pVCpu);
8609 else if (!pVCpu->hm.s.Event.fPending)
8610 hmR0VmxEvaluatePendingEvent(pVCpu, pMixedCtx);
8611
8612 /*
8613 * Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus needs to be done with
8614 * longjmps or interrupts + preemption enabled. Event injection might also result in triple-faulting the VM.
8615 */
8616 rc = hmR0VmxInjectPendingEvent(pVCpu, pMixedCtx, fStepping);
8617 if (RT_UNLIKELY(rc != VINF_SUCCESS))
8618 {
8619 Assert(rc == VINF_EM_RESET || (rc == VINF_EM_DBG_STEPPED && fStepping));
8620 return rc;
8621 }
8622
8623 /*
8624 * Load the guest state bits, we can handle longjmps/getting preempted here.
8625 *
8626 * If we are injecting events to a real-on-v86 mode guest, we will have to update
8627 * RIP and some segment registers, i.e. hmR0VmxInjectPendingEvent()->hmR0VmxInjectEventVmcs().
8628 * Hence, this needs to be done -after- injection of events.
8629 */
8630 hmR0VmxLoadGuestStateOptimal(pVM, pVCpu, pMixedCtx);
8631
8632 /*
8633 * No longjmps to ring-3 from this point on!!!
8634 * Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
8635 * This also disables flushing of the R0-logger instance (if any).
8636 */
8637 VMMRZCallRing3Disable(pVCpu);
8638
8639 /*
8640 * We disable interrupts so that we don't miss any interrupts that would flag preemption (IPI/timers etc.)
8641 * when thread-context hooks aren't used and we've been running with preemption disabled for a while.
8642 *
8643 * We need to check for force-flags that could've possible been altered since we last checked them (e.g.
8644 * by PDMGetInterrupt() leaving the PDM critical section, see @bugref{6398}).
8645 *
8646 * We also check a couple of other force-flags as a last opportunity to get the EMT back to ring-3 before
8647 * executing guest code.
8648 */
8649 pVmxTransient->fEFlags = ASMIntDisableFlags();
8650 if ( ( VM_FF_IS_PENDING(pVM, VM_FF_EMT_RENDEZVOUS | VM_FF_TM_VIRTUAL_SYNC)
8651 || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
8652 && ( !fStepping /* Optimized for the non-stepping case, of course. */
8653 || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK & ~(VMCPU_FF_TIMER | VMCPU_FF_PDM_CRITSECT))) )
8654 {
8655 hmR0VmxClearEventVmcs(pVCpu);
8656 ASMSetFlags(pVmxTransient->fEFlags);
8657 VMMRZCallRing3Enable(pVCpu);
8658 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
8659 return VINF_EM_RAW_TO_R3;
8660 }
8661
8662 if (RTThreadPreemptIsPending(NIL_RTTHREAD))
8663 {
8664 hmR0VmxClearEventVmcs(pVCpu);
8665 ASMSetFlags(pVmxTransient->fEFlags);
8666 VMMRZCallRing3Enable(pVCpu);
8667 STAM_COUNTER_INC(&pVCpu->hm.s.StatPendingHostIrq);
8668 return VINF_EM_RAW_INTERRUPT;
8669 }
8670
8671 /* We've injected any pending events. This is really the point of no return (to ring-3). */
8672 pVCpu->hm.s.Event.fPending = false;
8673
8674 return VINF_SUCCESS;
8675}
8676
8677
8678/**
8679 * Prepares to run guest code in VT-x and we've committed to doing so. This
8680 * means there is no backing out to ring-3 or anywhere else at this
8681 * point.
8682 *
8683 * @param pVM Pointer to the VM.
8684 * @param pVCpu Pointer to the VMCPU.
8685 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8686 * out-of-sync. Make sure to update the required fields
8687 * before using them.
8688 * @param pVmxTransient Pointer to the VMX transient structure.
8689 *
8690 * @remarks Called with preemption disabled.
8691 * @remarks No-long-jump zone!!!
8692 */
8693static void hmR0VmxPreRunGuestCommitted(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
8694{
8695 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8696 Assert(VMMR0IsLogFlushDisabled(pVCpu));
8697 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8698
8699 VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
8700 VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC); /* Indicate the start of guest execution. */
8701
8702#ifdef HMVMX_ALWAYS_SWAP_FPU_STATE
8703 if (!CPUMIsGuestFPUStateActive(pVCpu))
8704 CPUMR0LoadGuestFPU(pVM, pVCpu, pMixedCtx);
8705 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8706#endif
8707
8708 if ( pVCpu->hm.s.fPreloadGuestFpu
8709 && !CPUMIsGuestFPUStateActive(pVCpu))
8710 {
8711 CPUMR0LoadGuestFPU(pVM, pVCpu, pMixedCtx);
8712 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
8713 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8714 }
8715
8716 /*
8717 * Lazy-update of the host MSRs values in the auto-load/store MSR area.
8718 */
8719 if ( !pVCpu->hm.s.vmx.fUpdatedHostMsrs
8720 && pVCpu->hm.s.vmx.cMsrs > 0)
8721 {
8722 hmR0VmxUpdateAutoLoadStoreHostMsrs(pVCpu);
8723 }
8724
8725 /*
8726 * Load the host state bits as we may've been preempted (only happens when
8727 * thread-context hooks are used or when hmR0VmxSetupVMRunHandler() changes pfnStartVM).
8728 */
8729 /** @todo Why should hmR0VmxSetupVMRunHandler() changing pfnStartVM have
8730 * any effect to the host state needing to be saved? */
8731 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT))
8732 {
8733 /* This ASSUMES that pfnStartVM has been set up already. */
8734 int rc = hmR0VmxSaveHostState(pVM, pVCpu);
8735 AssertRC(rc);
8736 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreemptSaveHostState);
8737 }
8738 Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT));
8739
8740 /*
8741 * Load the state shared between host and guest (FPU, debug, lazy MSRs).
8742 */
8743 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE))
8744 hmR0VmxLoadSharedState(pVM, pVCpu, pMixedCtx);
8745 AssertMsg(!HMCPU_CF_VALUE(pVCpu), ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8746
8747 /* Store status of the shared guest-host state at the time of VM-entry. */
8748#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
8749 if (CPUMIsGuestInLongModeEx(pMixedCtx))
8750 {
8751 pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActivePending(pVCpu);
8752 pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActivePending(pVCpu);
8753 }
8754 else
8755#endif
8756 {
8757 pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
8758 pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
8759 }
8760 pVmxTransient->fWasGuestFPUStateActive = CPUMIsGuestFPUStateActive(pVCpu);
8761
8762 /*
8763 * Cache the TPR-shadow for checking on every VM-exit if it might have changed.
8764 */
8765 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
8766 pVmxTransient->u8GuestTpr = pVCpu->hm.s.vmx.pbVirtApic[0x80];
8767
8768 PHMGLOBALCPUINFO pCpu = HMR0GetCurrentCpu();
8769 RTCPUID idCurrentCpu = pCpu->idCpu;
8770 if ( pVmxTransient->fUpdateTscOffsettingAndPreemptTimer
8771 || idCurrentCpu != pVCpu->hm.s.idLastCpu)
8772 {
8773 hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVM, pVCpu);
8774 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = false;
8775 }
8776
8777 ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB-shootdowns, set this across the world switch. */
8778 hmR0VmxFlushTaggedTlb(pVCpu, pCpu); /* Invalidate the appropriate guest entries from the TLB. */
8779 Assert(idCurrentCpu == pVCpu->hm.s.idLastCpu);
8780 pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Update the error reporting info. with the current host CPU. */
8781
8782 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
8783
8784 TMNotifyStartOfExecution(pVCpu); /* Finally, notify TM to resume its clocks as we're about
8785 to start executing. */
8786
8787 /*
8788 * Load the TSC_AUX MSR when we are not intercepting RDTSCP.
8789 */
8790 if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP)
8791 {
8792 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT))
8793 {
8794 bool fMsrUpdated;
8795 int rc2 = hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
8796 AssertRC(rc2);
8797 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS));
8798
8799 rc2 = hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_TSC_AUX, CPUMR0GetGuestTscAux(pVCpu), true /* fUpdateHostMsr */,
8800 &fMsrUpdated);
8801 AssertRC(rc2);
8802 Assert(fMsrUpdated || pVCpu->hm.s.vmx.fUpdatedHostMsrs);
8803
8804 /* Finally, mark that all host MSR values are updated so we don't redo it without leaving VT-x. See @bugref{6956}. */
8805 pVCpu->hm.s.vmx.fUpdatedHostMsrs = true;
8806 }
8807 else
8808 {
8809 hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, MSR_K8_TSC_AUX);
8810 Assert(!pVCpu->hm.s.vmx.cMsrs || pVCpu->hm.s.vmx.fUpdatedHostMsrs);
8811 }
8812 }
8813
8814#ifdef VBOX_STRICT
8815 hmR0VmxCheckAutoLoadStoreMsrs(pVCpu);
8816 hmR0VmxCheckHostEferMsr(pVCpu);
8817 AssertRC(hmR0VmxCheckVmcsCtls(pVCpu));
8818#endif
8819#ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE
8820 uint32_t uInvalidReason = hmR0VmxCheckGuestState(pVM, pVCpu, pMixedCtx);
8821 if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND)
8822 Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason));
8823#endif
8824}
8825
8826
8827/**
8828 * Performs some essential restoration of state after running guest code in
8829 * VT-x.
8830 *
8831 * @param pVM Pointer to the VM.
8832 * @param pVCpu Pointer to the VMCPU.
8833 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
8834 * out-of-sync. Make sure to update the required fields
8835 * before using them.
8836 * @param pVmxTransient Pointer to the VMX transient structure.
8837 * @param rcVMRun Return code of VMLAUNCH/VMRESUME.
8838 *
8839 * @remarks Called with interrupts disabled, and returns with interrups enabled!
8840 *
8841 * @remarks No-long-jump zone!!! This function will however re-enable longjmps
8842 * unconditionally when it is safe to do so.
8843 */
8844static void hmR0VmxPostRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, int rcVMRun)
8845{
8846 NOREF(pVM);
8847
8848 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8849
8850 ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB-shootdowns. */
8851 ASMAtomicIncU32(&pVCpu->hm.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for TLB-shootdowns. */
8852 HMVMXCPU_GST_RESET_TO(pVCpu, 0); /* Exits/longjmps to ring-3 requires saving the guest state. */
8853 pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */
8854 pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
8855 pVmxTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
8856
8857 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT))
8858 TMCpuTickSetLastSeen(pVCpu, ASMReadTSC() + pVCpu->hm.s.vmx.u64TSCOffset);
8859
8860 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatExit1, x);
8861 TMNotifyEndOfExecution(pVCpu); /* Notify TM that the guest is no longer running. */
8862 Assert(!ASMIntAreEnabled());
8863 VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
8864
8865#ifdef HMVMX_ALWAYS_SWAP_FPU_STATE
8866 if (CPUMIsGuestFPUStateActive(pVCpu))
8867 {
8868 hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
8869 CPUMR0SaveGuestFPU(pVM, pVCpu, pMixedCtx);
8870 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8871 }
8872#endif
8873
8874#if HC_ARCH_BITS == 64
8875 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_REQUIRED; /* Host state messed up by VT-x, we must restore. */
8876#endif
8877 pVCpu->hm.s.vmx.uVmcsState |= HMVMX_VMCS_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
8878#ifdef VBOX_STRICT
8879 hmR0VmxCheckHostEferMsr(pVCpu); /* Verify that VMRUN/VMLAUNCH didn't modify host EFER. */
8880#endif
8881 ASMSetFlags(pVmxTransient->fEFlags); /* Enable interrupts. */
8882 VMMRZCallRing3Enable(pVCpu); /* It is now safe to do longjmps to ring-3!!! */
8883
8884 /* Save the basic VM-exit reason. Refer Intel spec. 24.9.1 "Basic VM-exit Information". */
8885 uint32_t uExitReason;
8886 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
8887 rc |= hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
8888 AssertRC(rc);
8889 pVmxTransient->uExitReason = (uint16_t)VMX_EXIT_REASON_BASIC(uExitReason);
8890 pVmxTransient->fVMEntryFailed = VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uEntryIntInfo);
8891
8892 /* Update the VM-exit history array. */
8893 HMCPU_EXIT_HISTORY_ADD(pVCpu, pVmxTransient->uExitReason);
8894
8895 /* If the VMLAUNCH/VMRESUME failed, we can bail out early. This does -not- cover VMX_EXIT_ERR_*. */
8896 if (RT_UNLIKELY(rcVMRun != VINF_SUCCESS))
8897 {
8898 Log4(("VM-entry failure: pVCpu=%p idCpu=%RU32 rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", pVCpu, pVCpu->idCpu, rcVMRun,
8899 pVmxTransient->fVMEntryFailed));
8900 return;
8901 }
8902
8903 if (RT_LIKELY(!pVmxTransient->fVMEntryFailed))
8904 {
8905 /** @todo We can optimize this by only syncing with our force-flags when
8906 * really needed and keeping the VMCS state as it is for most
8907 * VM-exits. */
8908 /* Update the guest interruptibility-state from the VMCS. */
8909 hmR0VmxSaveGuestIntrState(pVCpu, pMixedCtx);
8910
8911#if defined(HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE) || defined(HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE)
8912 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
8913 AssertRC(rc);
8914#elif defined(HMVMX_ALWAYS_SAVE_GUEST_RFLAGS)
8915 rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
8916 AssertRC(rc);
8917#endif
8918
8919 /*
8920 * If the TPR was raised by the guest, it wouldn't cause a VM-exit immediately. Instead we sync the TPR lazily whenever
8921 * we eventually get a VM-exit for any reason. This maybe expensive as PDMApicSetTPR() can longjmp to ring-3 and which is
8922 * why it's done here as it's easier and no less efficient to deal with it here than making hmR0VmxSaveGuestState()
8923 * cope with longjmps safely (see VMCPU_FF_HM_UPDATE_CR3 handling).
8924 */
8925 if ( (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
8926 && pVmxTransient->u8GuestTpr != pVCpu->hm.s.vmx.pbVirtApic[0x80])
8927 {
8928 rc = PDMApicSetTPR(pVCpu, pVCpu->hm.s.vmx.pbVirtApic[0x80]);
8929 AssertRC(rc);
8930 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
8931 }
8932 }
8933}
8934
8935
8936/**
8937 * Runs the guest code using VT-x the normal way.
8938 *
8939 * @returns VBox status code.
8940 * @param pVM Pointer to the VM.
8941 * @param pVCpu Pointer to the VMCPU.
8942 * @param pCtx Pointer to the guest-CPU context.
8943 *
8944 * @note Mostly the same as hmR0VmxRunGuestCodeStep().
8945 */
8946static int hmR0VmxRunGuestCodeNormal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
8947{
8948 VMXTRANSIENT VmxTransient;
8949 VmxTransient.fUpdateTscOffsettingAndPreemptTimer = true;
8950 int rc = VERR_INTERNAL_ERROR_5;
8951 uint32_t cLoops = 0;
8952
8953 for (;; cLoops++)
8954 {
8955 Assert(!HMR0SuspendPending());
8956 HMVMX_ASSERT_CPU_SAFE();
8957
8958 /* Preparatory work for running guest code, this may force us to return
8959 to ring-3. This bugger disables interrupts on VINF_SUCCESS! */
8960 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
8961 rc = hmR0VmxPreRunGuest(pVM, pVCpu, pCtx, &VmxTransient, false /* fStepping */);
8962 if (rc != VINF_SUCCESS)
8963 break;
8964
8965 hmR0VmxPreRunGuestCommitted(pVM, pVCpu, pCtx, &VmxTransient);
8966 rc = hmR0VmxRunGuest(pVM, pVCpu, pCtx);
8967 /* The guest-CPU context is now outdated, 'pCtx' is to be treated as 'pMixedCtx' from this point on!!! */
8968
8969 /* Restore any residual host-state and save any bits shared between host
8970 and guest into the guest-CPU state. Re-enables interrupts! */
8971 hmR0VmxPostRunGuest(pVM, pVCpu, pCtx, &VmxTransient, rc);
8972
8973 /* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
8974 if (RT_UNLIKELY(rc != VINF_SUCCESS))
8975 {
8976 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x);
8977 hmR0VmxReportWorldSwitchError(pVM, pVCpu, rc, pCtx, &VmxTransient);
8978 return rc;
8979 }
8980
8981 /* Profile the VM-exit. */
8982 AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
8983 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
8984 STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
8985 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
8986 HMVMX_START_EXIT_DISPATCH_PROF();
8987
8988 VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, pCtx, VmxTransient.uExitReason);
8989 if (RT_UNLIKELY(VBOXVMM_R0_HMVMX_VMEXIT_ENABLED()))
8990 {
8991 hmR0VmxReadExitQualificationVmcs(pVCpu, &VmxTransient);
8992 hmR0VmxSaveGuestState(pVCpu, pCtx);
8993 VBOXVMM_R0_HMVMX_VMEXIT(pVCpu, pCtx, VmxTransient.uExitReason, VmxTransient.uExitQualification);
8994 }
8995
8996 /* Handle the VM-exit. */
8997#ifdef HMVMX_USE_FUNCTION_TABLE
8998 rc = g_apfnVMExitHandlers[VmxTransient.uExitReason](pVCpu, pCtx, &VmxTransient);
8999#else
9000 rc = hmR0VmxHandleExit(pVCpu, pCtx, &VmxTransient, VmxTransient.uExitReason);
9001#endif
9002 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
9003 if (rc != VINF_SUCCESS)
9004 break;
9005 if (cLoops > pVM->hm.s.cMaxResumeLoops)
9006 {
9007 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
9008 rc = VINF_EM_RAW_INTERRUPT;
9009 break;
9010 }
9011 }
9012
9013 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
9014 return rc;
9015}
9016
9017
9018/**
9019 * Single steps guest code using VT-x.
9020 *
9021 * @returns VBox status code.
9022 * @param pVM Pointer to the VM.
9023 * @param pVCpu Pointer to the VMCPU.
9024 * @param pCtx Pointer to the guest-CPU context.
9025 *
9026 * @note Mostly the same as hmR0VmxRunGuestCodeNormal().
9027 */
9028static int hmR0VmxRunGuestCodeStep(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
9029{
9030 VMXTRANSIENT VmxTransient;
9031 VmxTransient.fUpdateTscOffsettingAndPreemptTimer = true;
9032 VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
9033 uint32_t cLoops = 0;
9034 uint16_t uCsStart = pCtx->cs.Sel;
9035 uint64_t uRipStart = pCtx->rip;
9036
9037 for (;; cLoops++)
9038 {
9039 Assert(!HMR0SuspendPending());
9040 HMVMX_ASSERT_CPU_SAFE();
9041
9042 /* Preparatory work for running guest code, this may force us to return
9043 to ring-3. This bugger disables interrupts on VINF_SUCCESS! */
9044 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
9045 rcStrict = hmR0VmxPreRunGuest(pVM, pVCpu, pCtx, &VmxTransient, true /* fStepping */);
9046 if (rcStrict != VINF_SUCCESS)
9047 break;
9048
9049 hmR0VmxPreRunGuestCommitted(pVM, pVCpu, pCtx, &VmxTransient);
9050 rcStrict = hmR0VmxRunGuest(pVM, pVCpu, pCtx);
9051 /* The guest-CPU context is now outdated, 'pCtx' is to be treated as 'pMixedCtx' from this point on!!! */
9052
9053 /* Restore any residual host-state and save any bits shared between host
9054 and guest into the guest-CPU state. Re-enables interrupts! */
9055 hmR0VmxPostRunGuest(pVM, pVCpu, pCtx, &VmxTransient, VBOXSTRICTRC_TODO(rcStrict));
9056
9057 /* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
9058 if (RT_UNLIKELY(rcStrict != VINF_SUCCESS))
9059 {
9060 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x);
9061 hmR0VmxReportWorldSwitchError(pVM, pVCpu, VBOXSTRICTRC_TODO(rcStrict), pCtx, &VmxTransient);
9062 return VBOXSTRICTRC_TODO(rcStrict);
9063 }
9064
9065 /* Profile the VM-exit. */
9066 AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
9067 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
9068 STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
9069 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
9070 HMVMX_START_EXIT_DISPATCH_PROF();
9071
9072 VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, pCtx, VmxTransient.uExitReason);
9073 if (RT_UNLIKELY(VBOXVMM_R0_HMVMX_VMEXIT_ENABLED()))
9074 {
9075 hmR0VmxReadExitQualificationVmcs(pVCpu, &VmxTransient);
9076 hmR0VmxSaveGuestState(pVCpu, pCtx);
9077 VBOXVMM_R0_HMVMX_VMEXIT(pVCpu, pCtx, VmxTransient.uExitReason, VmxTransient.uExitQualification);
9078 }
9079
9080 /* Handle the VM-exit - we quit earlier on certain VM-exits, see hmR0VmxHandleExitStep(). */
9081 rcStrict = hmR0VmxHandleExitStep(pVCpu, pCtx, &VmxTransient, VmxTransient.uExitReason, uCsStart, uRipStart);
9082 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
9083 if (rcStrict != VINF_SUCCESS)
9084 break;
9085 if (cLoops > pVM->hm.s.cMaxResumeLoops)
9086 {
9087 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
9088 rcStrict = VINF_EM_RAW_INTERRUPT;
9089 break;
9090 }
9091
9092 /*
9093 * Did the RIP change, if so, consider it a single step.
9094 * Otherwise, make sure one of the TFs gets set.
9095 */
9096 int rc2 = hmR0VmxSaveGuestRip(pVCpu, pCtx);
9097 rc2 |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pCtx);
9098 AssertRCReturn(rc2, rc2);
9099 if ( pCtx->rip != uRipStart
9100 || pCtx->cs.Sel != uCsStart)
9101 {
9102 rcStrict = VINF_EM_DBG_STEPPED;
9103 break;
9104 }
9105 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
9106 }
9107
9108 /*
9109 * Clear the X86_EFL_TF if necessary.
9110 */
9111 if (pVCpu->hm.s.fClearTrapFlag)
9112 {
9113 int rc2 = hmR0VmxSaveGuestRflags(pVCpu, pCtx);
9114 AssertRCReturn(rc2, rc2);
9115 pVCpu->hm.s.fClearTrapFlag = false;
9116 pCtx->eflags.Bits.u1TF = 0;
9117 }
9118 /** @todo there seems to be issues with the resume flag when the monitor trap
9119 * flag is pending without being used. Seen early in bios init when
9120 * accessing APIC page in protected mode. */
9121
9122 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
9123 return VBOXSTRICTRC_TODO(rcStrict);
9124}
9125
9126
9127/**
9128 * Runs the guest code using VT-x.
9129 *
9130 * @returns VBox status code.
9131 * @param pVM Pointer to the VM.
9132 * @param pVCpu Pointer to the VMCPU.
9133 * @param pCtx Pointer to the guest-CPU context.
9134 */
9135VMMR0DECL(int) VMXR0RunGuestCode(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
9136{
9137 Assert(VMMRZCallRing3IsEnabled(pVCpu));
9138 Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
9139 HMVMX_ASSERT_PREEMPT_SAFE();
9140
9141 VMMRZCallRing3SetNotification(pVCpu, hmR0VmxCallRing3Callback, pCtx);
9142
9143 int rc;
9144 if (!pVCpu->hm.s.fSingleInstruction && !DBGFIsStepping(pVCpu))
9145 rc = hmR0VmxRunGuestCodeNormal(pVM, pVCpu, pCtx);
9146 else
9147 rc = hmR0VmxRunGuestCodeStep(pVM, pVCpu, pCtx);
9148
9149 if (rc == VERR_EM_INTERPRETER)
9150 rc = VINF_EM_RAW_EMULATE_INSTR;
9151 else if (rc == VINF_EM_RESET)
9152 rc = VINF_EM_TRIPLE_FAULT;
9153
9154 int rc2 = hmR0VmxExitToRing3(pVM, pVCpu, pCtx, rc);
9155 if (RT_FAILURE(rc2))
9156 {
9157 pVCpu->hm.s.u32HMError = rc;
9158 rc = rc2;
9159 }
9160 Assert(!VMMRZCallRing3IsNotificationSet(pVCpu));
9161 return rc;
9162}
9163
9164
9165#ifndef HMVMX_USE_FUNCTION_TABLE
9166DECLINLINE(int) hmR0VmxHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, uint32_t rcReason)
9167{
9168#ifdef DEBUG_ramshankar
9169# define SVVMCS() do { int rc2 = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRC(rc2); } while (0)
9170# define LDVMCS() do { HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST); } while (0)
9171#endif
9172 int rc;
9173 switch (rcReason)
9174 {
9175 case VMX_EXIT_EPT_MISCONFIG: /* SVVMCS(); */ rc = hmR0VmxExitEptMisconfig(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9176 case VMX_EXIT_EPT_VIOLATION: /* SVVMCS(); */ rc = hmR0VmxExitEptViolation(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9177 case VMX_EXIT_IO_INSTR: /* SVVMCS(); */ rc = hmR0VmxExitIoInstr(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9178 case VMX_EXIT_CPUID: /* SVVMCS(); */ rc = hmR0VmxExitCpuid(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9179 case VMX_EXIT_RDTSC: /* SVVMCS(); */ rc = hmR0VmxExitRdtsc(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9180 case VMX_EXIT_RDTSCP: /* SVVMCS(); */ rc = hmR0VmxExitRdtscp(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9181 case VMX_EXIT_APIC_ACCESS: /* SVVMCS(); */ rc = hmR0VmxExitApicAccess(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9182 case VMX_EXIT_XCPT_OR_NMI: /* SVVMCS(); */ rc = hmR0VmxExitXcptOrNmi(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9183 case VMX_EXIT_MOV_CRX: /* SVVMCS(); */ rc = hmR0VmxExitMovCRx(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9184 case VMX_EXIT_EXT_INT: /* SVVMCS(); */ rc = hmR0VmxExitExtInt(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9185 case VMX_EXIT_INT_WINDOW: /* SVVMCS(); */ rc = hmR0VmxExitIntWindow(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9186 case VMX_EXIT_MWAIT: /* SVVMCS(); */ rc = hmR0VmxExitMwait(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9187 case VMX_EXIT_MONITOR: /* SVVMCS(); */ rc = hmR0VmxExitMonitor(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9188 case VMX_EXIT_TASK_SWITCH: /* SVVMCS(); */ rc = hmR0VmxExitTaskSwitch(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9189 case VMX_EXIT_PREEMPT_TIMER: /* SVVMCS(); */ rc = hmR0VmxExitPreemptTimer(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9190 case VMX_EXIT_RDMSR: /* SVVMCS(); */ rc = hmR0VmxExitRdmsr(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9191 case VMX_EXIT_WRMSR: /* SVVMCS(); */ rc = hmR0VmxExitWrmsr(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9192 case VMX_EXIT_MOV_DRX: /* SVVMCS(); */ rc = hmR0VmxExitMovDRx(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9193 case VMX_EXIT_TPR_BELOW_THRESHOLD: /* SVVMCS(); */ rc = hmR0VmxExitTprBelowThreshold(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9194 case VMX_EXIT_HLT: /* SVVMCS(); */ rc = hmR0VmxExitHlt(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9195 case VMX_EXIT_INVD: /* SVVMCS(); */ rc = hmR0VmxExitInvd(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9196 case VMX_EXIT_INVLPG: /* SVVMCS(); */ rc = hmR0VmxExitInvlpg(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9197 case VMX_EXIT_RSM: /* SVVMCS(); */ rc = hmR0VmxExitRsm(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9198 case VMX_EXIT_MTF: /* SVVMCS(); */ rc = hmR0VmxExitMtf(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9199 case VMX_EXIT_PAUSE: /* SVVMCS(); */ rc = hmR0VmxExitPause(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9200 case VMX_EXIT_XDTR_ACCESS: /* SVVMCS(); */ rc = hmR0VmxExitXdtrAccess(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9201 case VMX_EXIT_TR_ACCESS: /* SVVMCS(); */ rc = hmR0VmxExitXdtrAccess(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9202 case VMX_EXIT_WBINVD: /* SVVMCS(); */ rc = hmR0VmxExitWbinvd(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9203 case VMX_EXIT_XSETBV: /* SVVMCS(); */ rc = hmR0VmxExitXsetbv(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9204 case VMX_EXIT_RDRAND: /* SVVMCS(); */ rc = hmR0VmxExitRdrand(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9205 case VMX_EXIT_INVPCID: /* SVVMCS(); */ rc = hmR0VmxExitInvpcid(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9206 case VMX_EXIT_GETSEC: /* SVVMCS(); */ rc = hmR0VmxExitGetsec(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9207 case VMX_EXIT_RDPMC: /* SVVMCS(); */ rc = hmR0VmxExitRdpmc(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9208 case VMX_EXIT_VMCALL: /* SVVMCS(); */ rc = hmR0VmxExitVmcall(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9209
9210 case VMX_EXIT_TRIPLE_FAULT: rc = hmR0VmxExitTripleFault(pVCpu, pMixedCtx, pVmxTransient); break;
9211 case VMX_EXIT_NMI_WINDOW: rc = hmR0VmxExitNmiWindow(pVCpu, pMixedCtx, pVmxTransient); break;
9212 case VMX_EXIT_INIT_SIGNAL: rc = hmR0VmxExitInitSignal(pVCpu, pMixedCtx, pVmxTransient); break;
9213 case VMX_EXIT_SIPI: rc = hmR0VmxExitSipi(pVCpu, pMixedCtx, pVmxTransient); break;
9214 case VMX_EXIT_IO_SMI: rc = hmR0VmxExitIoSmi(pVCpu, pMixedCtx, pVmxTransient); break;
9215 case VMX_EXIT_SMI: rc = hmR0VmxExitSmi(pVCpu, pMixedCtx, pVmxTransient); break;
9216 case VMX_EXIT_ERR_MSR_LOAD: rc = hmR0VmxExitErrMsrLoad(pVCpu, pMixedCtx, pVmxTransient); break;
9217 case VMX_EXIT_ERR_INVALID_GUEST_STATE: rc = hmR0VmxExitErrInvalidGuestState(pVCpu, pMixedCtx, pVmxTransient); break;
9218 case VMX_EXIT_ERR_MACHINE_CHECK: rc = hmR0VmxExitErrMachineCheck(pVCpu, pMixedCtx, pVmxTransient); break;
9219
9220 case VMX_EXIT_VMCLEAR:
9221 case VMX_EXIT_VMLAUNCH:
9222 case VMX_EXIT_VMPTRLD:
9223 case VMX_EXIT_VMPTRST:
9224 case VMX_EXIT_VMREAD:
9225 case VMX_EXIT_VMRESUME:
9226 case VMX_EXIT_VMWRITE:
9227 case VMX_EXIT_VMXOFF:
9228 case VMX_EXIT_VMXON:
9229 case VMX_EXIT_INVEPT:
9230 case VMX_EXIT_INVVPID:
9231 case VMX_EXIT_VMFUNC:
9232 case VMX_EXIT_XSAVES:
9233 case VMX_EXIT_XRSTORS:
9234 rc = hmR0VmxExitSetPendingXcptUD(pVCpu, pMixedCtx, pVmxTransient);
9235 break;
9236 case VMX_EXIT_RESERVED_60:
9237 case VMX_EXIT_RDSEED: /* only spurious exits, so undefined */
9238 case VMX_EXIT_RESERVED_62:
9239 default:
9240 rc = hmR0VmxExitErrUndefined(pVCpu, pMixedCtx, pVmxTransient);
9241 break;
9242 }
9243 return rc;
9244}
9245#endif /* !HMVMX_USE_FUNCTION_TABLE */
9246
9247
9248/**
9249 * Single-stepping VM-exit filtering.
9250 *
9251 * This is preprocessing the exits and deciding whether we've gotten far enough
9252 * to return VINF_EM_DBG_STEPPED already. If not, normal VM-exit handling is
9253 * performed.
9254 *
9255 * @returns Strict VBox status code.
9256 * @param pVCpu The virtual CPU of the calling EMT.
9257 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
9258 * out-of-sync. Make sure to update the required
9259 * fields before using them.
9260 * @param pVmxTransient Pointer to the VMX-transient structure.
9261 * @param uExitReason The VM-exit reason.
9262 */
9263DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExitStep(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient,
9264 uint32_t uExitReason, uint16_t uCsStart, uint64_t uRipStart)
9265{
9266 switch (uExitReason)
9267 {
9268 case VMX_EXIT_XCPT_OR_NMI:
9269 {
9270 /* Check for host NMI. */
9271 int rc2 = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
9272 AssertRCReturn(rc2, rc2);
9273 uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVmxTransient->uExitIntInfo);
9274 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
9275 return hmR0VmxExitXcptOrNmi(pVCpu, pMixedCtx, pVmxTransient);
9276 /* fall thru */
9277 }
9278
9279 case VMX_EXIT_EPT_MISCONFIG:
9280 case VMX_EXIT_TRIPLE_FAULT:
9281 case VMX_EXIT_APIC_ACCESS:
9282 case VMX_EXIT_TPR_BELOW_THRESHOLD:
9283 case VMX_EXIT_TASK_SWITCH:
9284
9285 /* Instruction specific VM-exits: */
9286 case VMX_EXIT_IO_INSTR:
9287 case VMX_EXIT_CPUID:
9288 case VMX_EXIT_RDTSC:
9289 case VMX_EXIT_RDTSCP:
9290 case VMX_EXIT_MOV_CRX:
9291 case VMX_EXIT_MWAIT:
9292 case VMX_EXIT_MONITOR:
9293 case VMX_EXIT_RDMSR:
9294 case VMX_EXIT_WRMSR:
9295 case VMX_EXIT_MOV_DRX:
9296 case VMX_EXIT_HLT:
9297 case VMX_EXIT_INVD:
9298 case VMX_EXIT_INVLPG:
9299 case VMX_EXIT_RSM:
9300 case VMX_EXIT_PAUSE:
9301 case VMX_EXIT_XDTR_ACCESS:
9302 case VMX_EXIT_TR_ACCESS:
9303 case VMX_EXIT_WBINVD:
9304 case VMX_EXIT_XSETBV:
9305 case VMX_EXIT_RDRAND:
9306 case VMX_EXIT_INVPCID:
9307 case VMX_EXIT_GETSEC:
9308 case VMX_EXIT_RDPMC:
9309 case VMX_EXIT_VMCALL:
9310 case VMX_EXIT_VMCLEAR:
9311 case VMX_EXIT_VMLAUNCH:
9312 case VMX_EXIT_VMPTRLD:
9313 case VMX_EXIT_VMPTRST:
9314 case VMX_EXIT_VMREAD:
9315 case VMX_EXIT_VMRESUME:
9316 case VMX_EXIT_VMWRITE:
9317 case VMX_EXIT_VMXOFF:
9318 case VMX_EXIT_VMXON:
9319 case VMX_EXIT_INVEPT:
9320 case VMX_EXIT_INVVPID:
9321 case VMX_EXIT_VMFUNC:
9322 {
9323 int rc2 = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
9324 rc2 |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
9325 AssertRCReturn(rc2, rc2);
9326 if ( pMixedCtx->rip != uRipStart
9327 || pMixedCtx->cs.Sel != uCsStart)
9328 return VINF_EM_DBG_STEPPED;
9329 break;
9330 }
9331 }
9332
9333 /*
9334 * Normal processing.
9335 */
9336#ifdef HMVMX_USE_FUNCTION_TABLE
9337 return g_apfnVMExitHandlers[uExitReason](pVCpu, pMixedCtx, pVmxTransient);
9338#else
9339 return hmR0VmxHandleExit(pVCpu, pMixedCtx, pVmxTransient, uExitReason);
9340#endif
9341}
9342
9343
9344#ifdef DEBUG
9345/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
9346# define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \
9347 RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
9348
9349# define HMVMX_ASSERT_PREEMPT_CPUID() \
9350 do \
9351 { \
9352 RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
9353 AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
9354 } while (0)
9355
9356# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS() \
9357 do { \
9358 AssertPtr(pVCpu); \
9359 AssertPtr(pMixedCtx); \
9360 AssertPtr(pVmxTransient); \
9361 Assert(pVmxTransient->fVMEntryFailed == false); \
9362 Assert(ASMIntAreEnabled()); \
9363 HMVMX_ASSERT_PREEMPT_SAFE(); \
9364 HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \
9365 Log4Func(("vcpu[%RU32] -v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v-v\n", pVCpu->idCpu)); \
9366 HMVMX_ASSERT_PREEMPT_SAFE(); \
9367 if (VMMR0IsLogFlushDisabled(pVCpu)) \
9368 HMVMX_ASSERT_PREEMPT_CPUID(); \
9369 HMVMX_STOP_EXIT_DISPATCH_PROF(); \
9370 } while (0)
9371
9372# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS() \
9373 do { \
9374 Log4Func(("\n")); \
9375 } while (0)
9376#else /* Release builds */
9377# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS() \
9378 do { \
9379 HMVMX_STOP_EXIT_DISPATCH_PROF(); \
9380 NOREF(pVCpu); NOREF(pMixedCtx); NOREF(pVmxTransient); \
9381 } while (0)
9382# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS() do { } while (0)
9383#endif
9384
9385
9386/**
9387 * Advances the guest RIP after reading it from the VMCS.
9388 *
9389 * @returns VBox status code.
9390 * @param pVCpu Pointer to the VMCPU.
9391 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
9392 * out-of-sync. Make sure to update the required fields
9393 * before using them.
9394 * @param pVmxTransient Pointer to the VMX transient structure.
9395 *
9396 * @remarks No-long-jump zone!!!
9397 */
9398DECLINLINE(int) hmR0VmxAdvanceGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
9399{
9400 int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
9401 rc |= hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
9402 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
9403 AssertRCReturn(rc, rc);
9404
9405 pMixedCtx->rip += pVmxTransient->cbInstr;
9406 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
9407
9408 /*
9409 * Deliver a debug exception to the guest if it is single-stepping. Don't directly inject a #DB but use the
9410 * pending debug exception field as it takes care of priority of events.
9411 *
9412 * See Intel spec. 32.2.1 "Debug Exceptions".
9413 */
9414 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
9415
9416 return rc;
9417}
9418
9419
9420/**
9421 * Tries to determine what part of the guest-state VT-x has deemed as invalid
9422 * and update error record fields accordingly.
9423 *
9424 * @return VMX_IGS_* return codes.
9425 * @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything
9426 * wrong with the guest state.
9427 *
9428 * @param pVM Pointer to the VM.
9429 * @param pVCpu Pointer to the VMCPU.
9430 * @param pCtx Pointer to the guest-CPU state.
9431 *
9432 * @remarks This function assumes our cache of the VMCS controls
9433 * are valid, i.e. hmR0VmxCheckVmcsCtls() succeeded.
9434 */
9435static uint32_t hmR0VmxCheckGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
9436{
9437#define HMVMX_ERROR_BREAK(err) { uError = (err); break; }
9438#define HMVMX_CHECK_BREAK(expr, err) if (!(expr)) { \
9439 uError = (err); \
9440 break; \
9441 } else do { } while (0)
9442
9443 int rc;
9444 uint32_t uError = VMX_IGS_ERROR;
9445 uint32_t u32Val;
9446 bool fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuest;
9447
9448 do
9449 {
9450 /*
9451 * CR0.
9452 */
9453 uint32_t uSetCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
9454 uint32_t uZapCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
9455 /* Exceptions for unrestricted-guests for fixed CR0 bits (PE, PG).
9456 See Intel spec. 26.3.1 "Checks on Guest Control Registers, Debug Registers and MSRs." */
9457 if (fUnrestrictedGuest)
9458 uSetCR0 &= ~(X86_CR0_PE | X86_CR0_PG);
9459
9460 uint32_t u32GuestCR0;
9461 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32GuestCR0);
9462 AssertRCBreak(rc);
9463 HMVMX_CHECK_BREAK((u32GuestCR0 & uSetCR0) == uSetCR0, VMX_IGS_CR0_FIXED1);
9464 HMVMX_CHECK_BREAK(!(u32GuestCR0 & ~uZapCR0), VMX_IGS_CR0_FIXED0);
9465 if ( !fUnrestrictedGuest
9466 && (u32GuestCR0 & X86_CR0_PG)
9467 && !(u32GuestCR0 & X86_CR0_PE))
9468 {
9469 HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO);
9470 }
9471
9472 /*
9473 * CR4.
9474 */
9475 uint64_t uSetCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
9476 uint64_t uZapCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
9477
9478 uint32_t u32GuestCR4;
9479 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &u32GuestCR4);
9480 AssertRCBreak(rc);
9481 HMVMX_CHECK_BREAK((u32GuestCR4 & uSetCR4) == uSetCR4, VMX_IGS_CR4_FIXED1);
9482 HMVMX_CHECK_BREAK(!(u32GuestCR4 & ~uZapCR4), VMX_IGS_CR4_FIXED0);
9483
9484 /*
9485 * IA32_DEBUGCTL MSR.
9486 */
9487 uint64_t u64Val;
9488 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val);
9489 AssertRCBreak(rc);
9490 if ( (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
9491 && (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */
9492 {
9493 HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED);
9494 }
9495 uint64_t u64DebugCtlMsr = u64Val;
9496
9497#ifdef VBOX_STRICT
9498 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
9499 AssertRCBreak(rc);
9500 Assert(u32Val == pVCpu->hm.s.vmx.u32EntryCtls);
9501#endif
9502 bool const fLongModeGuest = RT_BOOL(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST);
9503
9504 /*
9505 * RIP and RFLAGS.
9506 */
9507 uint32_t u32Eflags;
9508#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9509 if (HMVMX_IS_64BIT_HOST_MODE())
9510 {
9511 rc = VMXReadVmcs64(VMX_VMCS_GUEST_RIP, &u64Val);
9512 AssertRCBreak(rc);
9513 /* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */
9514 if ( !fLongModeGuest
9515 || !pCtx->cs.Attr.n.u1Long)
9516 {
9517 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID);
9518 }
9519 /** @todo If the processor supports N < 64 linear-address bits, bits 63:N
9520 * must be identical if the "IA-32e mode guest" VM-entry
9521 * control is 1 and CS.L is 1. No check applies if the
9522 * CPU supports 64 linear-address bits. */
9523
9524 /* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */
9525 rc = VMXReadVmcs64(VMX_VMCS_GUEST_RFLAGS, &u64Val);
9526 AssertRCBreak(rc);
9527 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */
9528 VMX_IGS_RFLAGS_RESERVED);
9529 HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
9530 u32Eflags = u64Val;
9531 }
9532 else
9533#endif
9534 {
9535 rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Eflags);
9536 AssertRCBreak(rc);
9537 HMVMX_CHECK_BREAK(!(u32Eflags & 0xffc08028), VMX_IGS_RFLAGS_RESERVED); /* Bit 31:22, Bit 15, 5, 3 MBZ. */
9538 HMVMX_CHECK_BREAK((u32Eflags & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
9539 }
9540
9541 if ( fLongModeGuest
9542 || ( fUnrestrictedGuest
9543 && !(u32GuestCR0 & X86_CR0_PE)))
9544 {
9545 HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID);
9546 }
9547
9548 uint32_t u32EntryInfo;
9549 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
9550 AssertRCBreak(rc);
9551 if ( VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo)
9552 && VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
9553 {
9554 HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID);
9555 }
9556
9557 /*
9558 * 64-bit checks.
9559 */
9560#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9561 if (HMVMX_IS_64BIT_HOST_MODE())
9562 {
9563 if (fLongModeGuest)
9564 {
9565 HMVMX_CHECK_BREAK(u32GuestCR0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE);
9566 HMVMX_CHECK_BREAK(u32GuestCR4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE);
9567 }
9568
9569 if ( !fLongModeGuest
9570 && (u32GuestCR4 & X86_CR4_PCIDE))
9571 {
9572 HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE);
9573 }
9574
9575 /** @todo CR3 field must be such that bits 63:52 and bits in the range
9576 * 51:32 beyond the processor's physical-address width are 0. */
9577
9578 if ( (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
9579 && (pCtx->dr[7] & X86_DR7_MBZ_MASK))
9580 {
9581 HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED);
9582 }
9583
9584 rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val);
9585 AssertRCBreak(rc);
9586 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL);
9587
9588 rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val);
9589 AssertRCBreak(rc);
9590 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL);
9591 }
9592#endif
9593
9594 /*
9595 * PERF_GLOBAL MSR.
9596 */
9597 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR)
9598 {
9599 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val);
9600 AssertRCBreak(rc);
9601 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)),
9602 VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */
9603 }
9604
9605 /*
9606 * PAT MSR.
9607 */
9608 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR)
9609 {
9610 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val);
9611 AssertRCBreak(rc);
9612 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED);
9613 for (unsigned i = 0; i < 8; i++)
9614 {
9615 uint8_t u8Val = (u64Val & 0xff);
9616 if ( u8Val != 0 /* UC */
9617 && u8Val != 1 /* WC */
9618 && u8Val != 4 /* WT */
9619 && u8Val != 5 /* WP */
9620 && u8Val != 6 /* WB */
9621 && u8Val != 7 /* UC- */)
9622 {
9623 HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID);
9624 }
9625 u64Val >>= 8;
9626 }
9627 }
9628
9629 /*
9630 * EFER MSR.
9631 */
9632 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR)
9633 {
9634 Assert(pVM->hm.s.vmx.fSupportsVmcsEfer);
9635 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val);
9636 AssertRCBreak(rc);
9637 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)),
9638 VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */
9639 HMVMX_CHECK_BREAK(RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST),
9640 VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH);
9641 HMVMX_CHECK_BREAK( fUnrestrictedGuest
9642 || !(u32GuestCR0 & X86_CR0_PG)
9643 || RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(u64Val & MSR_K6_EFER_LME),
9644 VMX_IGS_EFER_LMA_LME_MISMATCH);
9645 }
9646
9647 /*
9648 * Segment registers.
9649 */
9650 HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9651 || !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID);
9652 if (!(u32Eflags & X86_EFL_VM))
9653 {
9654 /* CS */
9655 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID);
9656 HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED);
9657 HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED);
9658 HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff
9659 || !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9660 HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000)
9661 || (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9662 /* CS cannot be loaded with NULL in protected mode. */
9663 HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE);
9664 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID);
9665 if (pCtx->cs.Attr.n.u4Type == 9 || pCtx->cs.Attr.n.u4Type == 11)
9666 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL);
9667 else if (pCtx->cs.Attr.n.u4Type == 13 || pCtx->cs.Attr.n.u4Type == 15)
9668 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH);
9669 else if (pVM->hm.s.vmx.fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3)
9670 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID);
9671 else
9672 HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID);
9673
9674 /* SS */
9675 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9676 || (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL);
9677 HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL);
9678 if ( !(pCtx->cr0 & X86_CR0_PE)
9679 || pCtx->cs.Attr.n.u4Type == 3)
9680 {
9681 HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID);
9682 }
9683 if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
9684 {
9685 HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 || pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID);
9686 HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID);
9687 HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED);
9688 HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED);
9689 HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff
9690 || !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9691 HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000)
9692 || (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9693 }
9694
9695 /* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxWriteSegmentReg(). */
9696 if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
9697 {
9698 HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID);
9699 HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID);
9700 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9701 || pCtx->ds.Attr.n.u4Type > 11
9702 || pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9703 HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED);
9704 HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED);
9705 HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff
9706 || !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9707 HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000)
9708 || (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9709 HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9710 || (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID);
9711 }
9712 if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
9713 {
9714 HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID);
9715 HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID);
9716 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9717 || pCtx->es.Attr.n.u4Type > 11
9718 || pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9719 HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED);
9720 HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED);
9721 HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff
9722 || !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9723 HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000)
9724 || (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9725 HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9726 || (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID);
9727 }
9728 if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
9729 {
9730 HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID);
9731 HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID);
9732 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9733 || pCtx->fs.Attr.n.u4Type > 11
9734 || pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL);
9735 HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED);
9736 HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED);
9737 HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff
9738 || !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
9739 HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000)
9740 || (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
9741 HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9742 || (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID);
9743 }
9744 if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
9745 {
9746 HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID);
9747 HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID);
9748 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9749 || pCtx->gs.Attr.n.u4Type > 11
9750 || pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL);
9751 HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED);
9752 HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED);
9753 HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff
9754 || !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
9755 HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000)
9756 || (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
9757 HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9758 || (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID);
9759 }
9760 /* 64-bit capable CPUs. */
9761#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9762 if (HMVMX_IS_64BIT_HOST_MODE())
9763 {
9764 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
9765 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
9766 HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9767 || X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
9768 HMVMX_CHECK_BREAK(!(pCtx->cs.u64Base >> 32), VMX_IGS_LONGMODE_CS_BASE_INVALID);
9769 HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ss.u64Base >> 32),
9770 VMX_IGS_LONGMODE_SS_BASE_INVALID);
9771 HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ds.u64Base >> 32),
9772 VMX_IGS_LONGMODE_DS_BASE_INVALID);
9773 HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->es.u64Base >> 32),
9774 VMX_IGS_LONGMODE_ES_BASE_INVALID);
9775 }
9776#endif
9777 }
9778 else
9779 {
9780 /* V86 mode checks. */
9781 uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
9782 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
9783 {
9784 u32CSAttr = 0xf3; u32SSAttr = 0xf3;
9785 u32DSAttr = 0xf3; u32ESAttr = 0xf3;
9786 u32FSAttr = 0xf3; u32GSAttr = 0xf3;
9787 }
9788 else
9789 {
9790 u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u;
9791 u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u;
9792 u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
9793 }
9794
9795 /* CS */
9796 HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID);
9797 HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID);
9798 HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID);
9799 /* SS */
9800 HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID);
9801 HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID);
9802 HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID);
9803 /* DS */
9804 HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID);
9805 HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID);
9806 HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID);
9807 /* ES */
9808 HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID);
9809 HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID);
9810 HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID);
9811 /* FS */
9812 HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID);
9813 HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID);
9814 HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID);
9815 /* GS */
9816 HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID);
9817 HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID);
9818 HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID);
9819 /* 64-bit capable CPUs. */
9820#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9821 if (HMVMX_IS_64BIT_HOST_MODE())
9822 {
9823 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
9824 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
9825 HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9826 || X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
9827 HMVMX_CHECK_BREAK(!(pCtx->cs.u64Base >> 32), VMX_IGS_LONGMODE_CS_BASE_INVALID);
9828 HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ss.u64Base >> 32),
9829 VMX_IGS_LONGMODE_SS_BASE_INVALID);
9830 HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ds.u64Base >> 32),
9831 VMX_IGS_LONGMODE_DS_BASE_INVALID);
9832 HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->es.u64Base >> 32),
9833 VMX_IGS_LONGMODE_ES_BASE_INVALID);
9834 }
9835#endif
9836 }
9837
9838 /*
9839 * TR.
9840 */
9841 HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID);
9842 /* 64-bit capable CPUs. */
9843#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9844 if (HMVMX_IS_64BIT_HOST_MODE())
9845 {
9846 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL);
9847 }
9848#endif
9849 if (fLongModeGuest)
9850 {
9851 HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */
9852 VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID);
9853 }
9854 else
9855 {
9856 HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */
9857 || pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/
9858 VMX_IGS_TR_ATTR_TYPE_INVALID);
9859 }
9860 HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID);
9861 HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID);
9862 HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */
9863 HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff
9864 || !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
9865 HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000)
9866 || (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
9867 HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE);
9868
9869 /*
9870 * GDTR and IDTR.
9871 */
9872#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9873 if (HMVMX_IS_64BIT_HOST_MODE())
9874 {
9875 rc = VMXReadVmcs64(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
9876 AssertRCBreak(rc);
9877 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL);
9878
9879 rc = VMXReadVmcs64(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
9880 AssertRCBreak(rc);
9881 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL);
9882 }
9883#endif
9884
9885 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
9886 AssertRCBreak(rc);
9887 HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
9888
9889 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
9890 AssertRCBreak(rc);
9891 HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
9892
9893 /*
9894 * Guest Non-Register State.
9895 */
9896 /* Activity State. */
9897 uint32_t u32ActivityState;
9898 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState);
9899 AssertRCBreak(rc);
9900 HMVMX_CHECK_BREAK( !u32ActivityState
9901 || (u32ActivityState & MSR_IA32_VMX_MISC_ACTIVITY_STATES(pVM->hm.s.vmx.Msrs.u64Misc)),
9902 VMX_IGS_ACTIVITY_STATE_INVALID);
9903 HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl)
9904 || u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID);
9905 uint32_t u32IntrState;
9906 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &u32IntrState);
9907 AssertRCBreak(rc);
9908 if ( u32IntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS
9909 || u32IntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
9910 {
9911 HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID);
9912 }
9913
9914 /** @todo Activity state and injecting interrupts. Left as a todo since we
9915 * currently don't use activity states but ACTIVE. */
9916
9917 HMVMX_CHECK_BREAK( !(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_ENTRY_SMM)
9918 || u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID);
9919
9920 /* Guest interruptibility-state. */
9921 HMVMX_CHECK_BREAK(!(u32IntrState & 0xfffffff0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED);
9922 HMVMX_CHECK_BREAK((u32IntrState & ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI
9923 | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS))
9924 != ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI
9925 | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
9926 VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID);
9927 HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF)
9928 || !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI),
9929 VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID);
9930 if (VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo))
9931 {
9932 if (VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
9933 {
9934 HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
9935 && !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
9936 VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID);
9937 }
9938 else if (VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
9939 {
9940 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
9941 VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID);
9942 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI),
9943 VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID);
9944 }
9945 }
9946 /** @todo Assumes the processor is not in SMM. */
9947 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI),
9948 VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID);
9949 HMVMX_CHECK_BREAK( !(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_ENTRY_SMM)
9950 || (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI),
9951 VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID);
9952 if ( (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
9953 && VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo)
9954 && VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
9955 {
9956 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI),
9957 VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID);
9958 }
9959
9960 /* Pending debug exceptions. */
9961 if (HMVMX_IS_64BIT_HOST_MODE())
9962 {
9963 rc = VMXReadVmcs64(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, &u64Val);
9964 AssertRCBreak(rc);
9965 /* Bits 63:15, Bit 13, Bits 11:4 MBZ. */
9966 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED);
9967 u32Val = u64Val; /* For pending debug exceptions checks below. */
9968 }
9969 else
9970 {
9971 rc = VMXReadVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, &u32Val);
9972 AssertRCBreak(rc);
9973 /* Bits 31:15, Bit 13, Bits 11:4 MBZ. */
9974 HMVMX_CHECK_BREAK(!(u64Val & 0xffffaff0), VMX_IGS_PENDING_DEBUG_RESERVED);
9975 }
9976
9977 if ( (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
9978 || (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS)
9979 || u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
9980 {
9981 if ( (u32Eflags & X86_EFL_TF)
9982 && !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
9983 {
9984 /* Bit 14 is PendingDebug.BS. */
9985 HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET);
9986 }
9987 if ( !(u32Eflags & X86_EFL_TF)
9988 || (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
9989 {
9990 /* Bit 14 is PendingDebug.BS. */
9991 HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR);
9992 }
9993 }
9994
9995 /* VMCS link pointer. */
9996 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val);
9997 AssertRCBreak(rc);
9998 if (u64Val != UINT64_C(0xffffffffffffffff))
9999 {
10000 HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED);
10001 /** @todo Bits beyond the processor's physical-address width MBZ. */
10002 /** @todo 32-bit located in memory referenced by value of this field (as a
10003 * physical address) must contain the processor's VMCS revision ID. */
10004 /** @todo SMM checks. */
10005 }
10006
10007 /** @todo Checks on Guest Page-Directory-Pointer-Table Entries when guest is
10008 * not using Nested Paging? */
10009 if ( pVM->hm.s.fNestedPaging
10010 && !fLongModeGuest
10011 && CPUMIsGuestInPAEModeEx(pCtx))
10012 {
10013 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &u64Val);
10014 AssertRCBreak(rc);
10015 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10016
10017 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &u64Val);
10018 AssertRCBreak(rc);
10019 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10020
10021 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &u64Val);
10022 AssertRCBreak(rc);
10023 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10024
10025 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &u64Val);
10026 AssertRCBreak(rc);
10027 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
10028 }
10029
10030 /* Shouldn't happen but distinguish it from AssertRCBreak() errors. */
10031 if (uError == VMX_IGS_ERROR)
10032 uError = VMX_IGS_REASON_NOT_FOUND;
10033 } while (0);
10034
10035 pVCpu->hm.s.u32HMError = uError;
10036 return uError;
10037
10038#undef HMVMX_ERROR_BREAK
10039#undef HMVMX_CHECK_BREAK
10040}
10041
10042/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
10043/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
10044/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
10045
10046/** @name VM-exit handlers.
10047 * @{
10048 */
10049
10050/**
10051 * VM-exit handler for external interrupts (VMX_EXIT_EXT_INT).
10052 */
10053HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10054{
10055 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10056 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
10057 /* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */
10058 if (VMMR0ThreadCtxHookIsEnabled(pVCpu))
10059 return VINF_SUCCESS;
10060 return VINF_EM_RAW_INTERRUPT;
10061}
10062
10063
10064/**
10065 * VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI).
10066 */
10067HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10068{
10069 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10070 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3);
10071
10072 int rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
10073 AssertRCReturn(rc, rc);
10074
10075 uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVmxTransient->uExitIntInfo);
10076 Assert( !(pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT)
10077 && uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT);
10078 Assert(VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
10079
10080 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
10081 {
10082 /*
10083 * This cannot be a guest NMI as the only way for the guest to receive an NMI is if we injected it ourselves and
10084 * anything we inject is not going to cause a VM-exit directly for the event being injected.
10085 * See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery".
10086 *
10087 * Dispatch the NMI to the host. See Intel spec. 27.5.5 "Updating Non-Register State".
10088 */
10089 VMXDispatchHostNmi();
10090 STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
10091 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
10092 return VINF_SUCCESS;
10093 }
10094
10095 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
10096 rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
10097 if (RT_UNLIKELY(rc != VINF_SUCCESS))
10098 {
10099 if (rc == VINF_HM_DOUBLE_FAULT)
10100 rc = VINF_SUCCESS;
10101 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
10102 return rc;
10103 }
10104
10105 uint32_t uExitIntInfo = pVmxTransient->uExitIntInfo;
10106 uint32_t uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(uExitIntInfo);
10107 switch (uIntType)
10108 {
10109 case VMX_EXIT_INTERRUPTION_INFO_TYPE_PRIV_SW_XCPT: /* Privileged software exception. (#DB from ICEBP) */
10110 Assert(uVector == X86_XCPT_DB);
10111 /* no break */
10112 case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT: /* Software exception. (#BP or #OF) */
10113 Assert(uVector == X86_XCPT_BP || uVector == X86_XCPT_OF || uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_PRIV_SW_XCPT);
10114 /* no break */
10115 case VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT:
10116 {
10117 switch (uVector)
10118 {
10119 case X86_XCPT_PF: rc = hmR0VmxExitXcptPF(pVCpu, pMixedCtx, pVmxTransient); break;
10120 case X86_XCPT_GP: rc = hmR0VmxExitXcptGP(pVCpu, pMixedCtx, pVmxTransient); break;
10121 case X86_XCPT_NM: rc = hmR0VmxExitXcptNM(pVCpu, pMixedCtx, pVmxTransient); break;
10122 case X86_XCPT_MF: rc = hmR0VmxExitXcptMF(pVCpu, pMixedCtx, pVmxTransient); break;
10123 case X86_XCPT_DB: rc = hmR0VmxExitXcptDB(pVCpu, pMixedCtx, pVmxTransient); break;
10124 case X86_XCPT_BP: rc = hmR0VmxExitXcptBP(pVCpu, pMixedCtx, pVmxTransient); break;
10125#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
10126 case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF);
10127 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
10128 case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE);
10129 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
10130 case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD);
10131 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
10132 case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS);
10133 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
10134 case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP);
10135 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
10136 case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS);
10137 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
10138#endif
10139 default:
10140 {
10141 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10142 AssertRCReturn(rc, rc);
10143
10144 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
10145 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
10146 {
10147 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
10148 Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
10149 Assert(CPUMIsGuestInRealModeEx(pMixedCtx));
10150
10151 rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
10152 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
10153 AssertRCReturn(rc, rc);
10154 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(uExitIntInfo),
10155 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode,
10156 0 /* GCPtrFaultAddress */);
10157 AssertRCReturn(rc, rc);
10158 }
10159 else
10160 {
10161 AssertMsgFailed(("Unexpected VM-exit caused by exception %#x\n", uVector));
10162 pVCpu->hm.s.u32HMError = uVector;
10163 rc = VERR_VMX_UNEXPECTED_EXCEPTION;
10164 }
10165 break;
10166 }
10167 }
10168 break;
10169 }
10170
10171 default:
10172 {
10173 pVCpu->hm.s.u32HMError = uExitIntInfo;
10174 rc = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
10175 AssertMsgFailed(("Unexpected interruption info %#x\n", VMX_EXIT_INTERRUPTION_INFO_TYPE(uExitIntInfo)));
10176 break;
10177 }
10178 }
10179 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
10180 return rc;
10181}
10182
10183
10184/**
10185 * VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
10186 */
10187HMVMX_EXIT_DECL hmR0VmxExitIntWindow(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10188{
10189 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10190
10191 /* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */
10192 hmR0VmxClearIntWindowExitVmcs(pVCpu);
10193
10194 /* Deliver the pending interrupts via hmR0VmxEvaluatePendingEvent() and resume guest execution. */
10195 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
10196 return VINF_SUCCESS;
10197}
10198
10199
10200/**
10201 * VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
10202 */
10203HMVMX_EXIT_DECL hmR0VmxExitNmiWindow(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10204{
10205 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10206 if (RT_UNLIKELY(!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT)))
10207 {
10208 AssertMsgFailed(("Unexpected NMI-window exit.\n"));
10209 HMVMX_RETURN_UNEXPECTED_EXIT();
10210 }
10211
10212 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS));
10213
10214 /*
10215 * If block-by-STI is set when we get this VM-exit, it means the CPU doesn't block NMIs following STI.
10216 * It is therefore safe to unblock STI and deliver the NMI ourselves. See @bugref{7445}.
10217 */
10218 uint32_t uIntrState = 0;
10219 int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
10220 AssertRCReturn(rc, rc);
10221
10222 bool const fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
10223 if ( fBlockSti
10224 && VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
10225 {
10226 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
10227 }
10228
10229 /* Indicate that we no longer need to VM-exit when the guest is ready to receive NMIs, it is now ready */
10230 hmR0VmxClearNmiWindowExitVmcs(pVCpu);
10231
10232 /* Deliver the pending NMI via hmR0VmxEvaluatePendingEvent() and resume guest execution. */
10233 return VINF_SUCCESS;
10234}
10235
10236
10237/**
10238 * VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
10239 */
10240HMVMX_EXIT_DECL hmR0VmxExitWbinvd(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10241{
10242 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10243 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWbinvd);
10244 return hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10245}
10246
10247
10248/**
10249 * VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit.
10250 */
10251HMVMX_EXIT_DECL hmR0VmxExitInvd(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10252{
10253 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10254 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvd);
10255 return hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10256}
10257
10258
10259/**
10260 * VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit.
10261 */
10262HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10263{
10264 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10265 PVM pVM = pVCpu->CTX_SUFF(pVM);
10266 int rc = EMInterpretCpuId(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10267 if (RT_LIKELY(rc == VINF_SUCCESS))
10268 {
10269 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10270 Assert(pVmxTransient->cbInstr == 2);
10271 }
10272 else
10273 {
10274 AssertMsgFailed(("hmR0VmxExitCpuid: EMInterpretCpuId failed with %Rrc\n", rc));
10275 rc = VERR_EM_INTERPRETER;
10276 }
10277 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCpuid);
10278 return rc;
10279}
10280
10281
10282/**
10283 * VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit.
10284 */
10285HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10286{
10287 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10288 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
10289 AssertRCReturn(rc, rc);
10290
10291 if (pMixedCtx->cr4 & X86_CR4_SMXE)
10292 return VINF_EM_RAW_EMULATE_INSTR;
10293
10294 AssertMsgFailed(("hmR0VmxExitGetsec: unexpected VM-exit when CR4.SMXE is 0.\n"));
10295 HMVMX_RETURN_UNEXPECTED_EXIT();
10296}
10297
10298
10299/**
10300 * VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
10301 */
10302HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10303{
10304 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10305 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); /** @todo review if CR4 is really required by EM. */
10306 AssertRCReturn(rc, rc);
10307
10308 PVM pVM = pVCpu->CTX_SUFF(pVM);
10309 rc = EMInterpretRdtsc(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10310 if (RT_LIKELY(rc == VINF_SUCCESS))
10311 {
10312 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10313 Assert(pVmxTransient->cbInstr == 2);
10314 /* If we get a spurious VM-exit when offsetting is enabled, we must reset offsetting on VM-reentry. See @bugref{6634}. */
10315 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING)
10316 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
10317 }
10318 else
10319 rc = VERR_EM_INTERPRETER;
10320 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc);
10321 return rc;
10322}
10323
10324
10325/**
10326 * VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
10327 */
10328HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10329{
10330 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10331 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); /** @todo review if CR4 is really required by EM. */
10332 rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx); /* For MSR_K8_TSC_AUX */
10333 AssertRCReturn(rc, rc);
10334
10335 PVM pVM = pVCpu->CTX_SUFF(pVM);
10336 rc = EMInterpretRdtscp(pVM, pVCpu, pMixedCtx);
10337 if (RT_LIKELY(rc == VINF_SUCCESS))
10338 {
10339 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10340 Assert(pVmxTransient->cbInstr == 3);
10341 /* If we get a spurious VM-exit when offsetting is enabled, we must reset offsetting on VM-reentry. See @bugref{6634}. */
10342 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING)
10343 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
10344 }
10345 else
10346 {
10347 AssertMsgFailed(("hmR0VmxExitRdtscp: EMInterpretRdtscp failed with %Rrc\n", rc));
10348 rc = VERR_EM_INTERPRETER;
10349 }
10350 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc);
10351 return rc;
10352}
10353
10354
10355/**
10356 * VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
10357 */
10358HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10359{
10360 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10361 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); /** @todo review if CR4 is really required by EM. */
10362 rc |= hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); /** @todo review if CR0 is really required by EM. */
10363 AssertRCReturn(rc, rc);
10364
10365 PVM pVM = pVCpu->CTX_SUFF(pVM);
10366 rc = EMInterpretRdpmc(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10367 if (RT_LIKELY(rc == VINF_SUCCESS))
10368 {
10369 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10370 Assert(pVmxTransient->cbInstr == 2);
10371 }
10372 else
10373 {
10374 AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
10375 rc = VERR_EM_INTERPRETER;
10376 }
10377 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdpmc);
10378 return rc;
10379}
10380
10381
10382/**
10383 * VM-exit handler for VMCALL (VMX_EXIT_VMCALL). Unconditional VM-exit.
10384 */
10385HMVMX_EXIT_DECL hmR0VmxExitVmcall(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10386{
10387 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10388 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitVmcall);
10389
10390 if (pVCpu->hm.s.fHypercallsEnabled)
10391 {
10392#if 0
10393 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
10394 AssertRCReturn(rc, rc);
10395#else
10396 /* Aggressive state sync. for now. */
10397 int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
10398 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); /* For long-mode checks in gimKvmHypercall(). */
10399#endif
10400 AssertRCReturn(rc, rc);
10401
10402 rc = GIMHypercall(pVCpu, pMixedCtx);
10403 if (RT_SUCCESS(rc))
10404 {
10405 /* If the hypercall changes anything other than guest general-purpose registers,
10406 we would need to reload the guest changed bits here before VM-reentry. */
10407 hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10408 return VINF_SUCCESS;
10409 }
10410 }
10411
10412 hmR0VmxSetPendingXcptUD(pVCpu, pMixedCtx);
10413 return VINF_SUCCESS;
10414}
10415
10416
10417/**
10418 * VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
10419 */
10420HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10421{
10422 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10423 PVM pVM = pVCpu->CTX_SUFF(pVM);
10424 Assert(!pVM->hm.s.fNestedPaging);
10425
10426 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
10427 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
10428 AssertRCReturn(rc, rc);
10429
10430 VBOXSTRICTRC rc2 = EMInterpretInvlpg(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), pVmxTransient->uExitQualification);
10431 rc = VBOXSTRICTRC_VAL(rc2);
10432 if (RT_LIKELY(rc == VINF_SUCCESS))
10433 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10434 else
10435 {
10436 AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitInvlpg: EMInterpretInvlpg %#RX64 failed with %Rrc\n",
10437 pVmxTransient->uExitQualification, rc));
10438 }
10439 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvlpg);
10440 return rc;
10441}
10442
10443
10444/**
10445 * VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
10446 */
10447HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10448{
10449 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10450 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10451 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10452 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10453 AssertRCReturn(rc, rc);
10454
10455 PVM pVM = pVCpu->CTX_SUFF(pVM);
10456 rc = EMInterpretMonitor(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10457 if (RT_LIKELY(rc == VINF_SUCCESS))
10458 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10459 else
10460 {
10461 AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMonitor: EMInterpretMonitor failed with %Rrc\n", rc));
10462 rc = VERR_EM_INTERPRETER;
10463 }
10464 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMonitor);
10465 return rc;
10466}
10467
10468
10469/**
10470 * VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
10471 */
10472HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10473{
10474 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10475 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10476 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10477 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10478 AssertRCReturn(rc, rc);
10479
10480 PVM pVM = pVCpu->CTX_SUFF(pVM);
10481 VBOXSTRICTRC rc2 = EMInterpretMWait(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10482 rc = VBOXSTRICTRC_VAL(rc2);
10483 if (RT_LIKELY( rc == VINF_SUCCESS
10484 || rc == VINF_EM_HALT))
10485 {
10486 int rc3 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10487 AssertRCReturn(rc3, rc3);
10488
10489 if ( rc == VINF_EM_HALT
10490 && EMMonitorWaitShouldContinue(pVCpu, pMixedCtx))
10491 {
10492 rc = VINF_SUCCESS;
10493 }
10494 }
10495 else
10496 {
10497 AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMwait: EMInterpretMWait failed with %Rrc\n", rc));
10498 rc = VERR_EM_INTERPRETER;
10499 }
10500 AssertMsg(rc == VINF_SUCCESS || rc == VINF_EM_HALT || rc == VERR_EM_INTERPRETER,
10501 ("hmR0VmxExitMwait: failed, invalid error code %Rrc\n", rc));
10502 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMwait);
10503 return rc;
10504}
10505
10506
10507/**
10508 * VM-exit handler for RSM (VMX_EXIT_RSM). Unconditional VM-exit.
10509 */
10510HMVMX_EXIT_DECL hmR0VmxExitRsm(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10511{
10512 /*
10513 * Execution of RSM outside of SMM mode causes #UD regardless of VMX root or VMX non-root mode. In theory, we should never
10514 * get this VM-exit. This can happen only if dual-monitor treatment of SMI and VMX is enabled, which can (only?) be done by
10515 * executing VMCALL in VMX root operation. If we get here, something funny is going on.
10516 * See Intel spec. "33.15.5 Enabling the Dual-Monitor Treatment".
10517 */
10518 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10519 AssertMsgFailed(("Unexpected RSM VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10520 HMVMX_RETURN_UNEXPECTED_EXIT();
10521}
10522
10523
10524/**
10525 * VM-exit handler for SMI (VMX_EXIT_SMI). Unconditional VM-exit.
10526 */
10527HMVMX_EXIT_DECL hmR0VmxExitSmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10528{
10529 /*
10530 * This can only happen if we support dual-monitor treatment of SMI, which can be activated by executing VMCALL in VMX
10531 * root operation. Only an STM (SMM transfer monitor) would get this VM-exit when we (the executive monitor) execute a VMCALL
10532 * in VMX root mode or receive an SMI. If we get here, something funny is going on.
10533 * See Intel spec. "33.15.6 Activating the Dual-Monitor Treatment" and Intel spec. 25.3 "Other Causes of VM-Exits"
10534 */
10535 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10536 AssertMsgFailed(("Unexpected SMI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10537 HMVMX_RETURN_UNEXPECTED_EXIT();
10538}
10539
10540
10541/**
10542 * VM-exit handler for IO SMI (VMX_EXIT_IO_SMI). Unconditional VM-exit.
10543 */
10544HMVMX_EXIT_DECL hmR0VmxExitIoSmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10545{
10546 /* Same treatment as VMX_EXIT_SMI. See comment in hmR0VmxExitSmi(). */
10547 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10548 AssertMsgFailed(("Unexpected IO SMI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10549 HMVMX_RETURN_UNEXPECTED_EXIT();
10550}
10551
10552
10553/**
10554 * VM-exit handler for SIPI (VMX_EXIT_SIPI). Conditional VM-exit.
10555 */
10556HMVMX_EXIT_DECL hmR0VmxExitSipi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10557{
10558 /*
10559 * SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest activity state is used. We currently
10560 * don't make use of it (see hmR0VmxLoadGuestActivityState()) as our guests don't have direct access to the host LAPIC.
10561 * See Intel spec. 25.3 "Other Causes of VM-exits".
10562 */
10563 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10564 AssertMsgFailed(("Unexpected SIPI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10565 HMVMX_RETURN_UNEXPECTED_EXIT();
10566}
10567
10568
10569/**
10570 * VM-exit handler for INIT signal (VMX_EXIT_INIT_SIGNAL). Unconditional
10571 * VM-exit.
10572 */
10573HMVMX_EXIT_DECL hmR0VmxExitInitSignal(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10574{
10575 /*
10576 * INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM.
10577 * See Intel spec. 33.14.1 Default Treatment of SMI Delivery" and Intel spec. 29.3 "VMX Instructions" for "VMXON".
10578 *
10579 * It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these VM-exits.
10580 * See Intel spec. "23.8 Restrictions on VMX operation".
10581 */
10582 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10583 return VINF_SUCCESS;
10584}
10585
10586
10587/**
10588 * VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional
10589 * VM-exit.
10590 */
10591HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10592{
10593 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10594 return VINF_EM_RESET;
10595}
10596
10597
10598/**
10599 * VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
10600 */
10601HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10602{
10603 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10604 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT);
10605 int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
10606 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10607 AssertRCReturn(rc, rc);
10608
10609 pMixedCtx->rip++;
10610 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
10611 if (EMShouldContinueAfterHalt(pVCpu, pMixedCtx)) /* Requires eflags. */
10612 rc = VINF_SUCCESS;
10613 else
10614 rc = VINF_EM_HALT;
10615
10616 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt);
10617 if (rc != VINF_SUCCESS)
10618 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
10619 return rc;
10620}
10621
10622
10623/**
10624 * VM-exit handler for instructions that result in a #UD exception delivered to
10625 * the guest.
10626 */
10627HMVMX_EXIT_DECL hmR0VmxExitSetPendingXcptUD(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10628{
10629 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10630 hmR0VmxSetPendingXcptUD(pVCpu, pMixedCtx);
10631 return VINF_SUCCESS;
10632}
10633
10634
10635/**
10636 * VM-exit handler for expiry of the VMX preemption timer.
10637 */
10638HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10639{
10640 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10641
10642 /* If the preemption-timer has expired, reinitialize the preemption timer on next VM-entry. */
10643 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
10644
10645 /* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */
10646 PVM pVM = pVCpu->CTX_SUFF(pVM);
10647 bool fTimersPending = TMTimerPollBool(pVM, pVCpu);
10648 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer);
10649 return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS;
10650}
10651
10652
10653/**
10654 * VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit.
10655 */
10656HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10657{
10658 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10659
10660 int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
10661 rc |= hmR0VmxSaveGuestRegsForIemExec(pVCpu, pMixedCtx, false /*fMemory*/, false /*fNeedRsp*/);
10662 rc |= hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
10663 AssertRCReturn(rc, rc);
10664
10665 VBOXSTRICTRC rcStrict = IEMExecDecodedXsetbv(pVCpu, pVmxTransient->cbInstr);
10666 HMCPU_CF_SET(pVCpu, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS : HM_CHANGED_ALL_GUEST);
10667
10668 pVCpu->hm.s.fLoadSaveGuestXcr0 = (pMixedCtx->cr4 & X86_CR4_OSXSAVE) && pMixedCtx->aXcr[0] != ASMGetXcr0();
10669
10670 return VBOXSTRICTRC_TODO(rcStrict);
10671}
10672
10673
10674/**
10675 * VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
10676 */
10677HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10678{
10679 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10680
10681 /* The guest should not invalidate the host CPU's TLBs, fallback to interpreter. */
10682 /** @todo implement EMInterpretInvpcid() */
10683 return VERR_EM_INTERPRETER;
10684}
10685
10686
10687/**
10688 * VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE).
10689 * Error VM-exit.
10690 */
10691HMVMX_EXIT_DECL hmR0VmxExitErrInvalidGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10692{
10693 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
10694 AssertRCReturn(rc, rc);
10695
10696 rc = hmR0VmxCheckVmcsCtls(pVCpu);
10697 AssertRCReturn(rc, rc);
10698
10699 uint32_t uInvalidReason = hmR0VmxCheckGuestState(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx);
10700 NOREF(uInvalidReason);
10701
10702#ifdef VBOX_STRICT
10703 uint32_t uIntrState;
10704 HMVMXHCUINTREG uHCReg;
10705 uint64_t u64Val;
10706 uint32_t u32Val;
10707
10708 rc = hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
10709 rc |= hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient);
10710 rc |= hmR0VmxReadEntryInstrLenVmcs(pVmxTransient);
10711 rc |= VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
10712 AssertRCReturn(rc, rc);
10713
10714 Log4(("uInvalidReason %u\n", uInvalidReason));
10715 Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo));
10716 Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode));
10717 Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr));
10718 Log4(("VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE %#RX32\n", uIntrState));
10719
10720 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32Val); AssertRC(rc);
10721 Log4(("VMX_VMCS_GUEST_CR0 %#RX32\n", u32Val));
10722 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc);
10723 Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg));
10724 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc);
10725 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
10726 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc);
10727 Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg));
10728 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc);
10729 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
10730 rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
10731 Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
10732#else
10733 NOREF(pVmxTransient);
10734#endif
10735
10736 HMDumpRegs(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx);
10737 return VERR_VMX_INVALID_GUEST_STATE;
10738}
10739
10740
10741/**
10742 * VM-exit handler for VM-entry failure due to an MSR-load
10743 * (VMX_EXIT_ERR_MSR_LOAD). Error VM-exit.
10744 */
10745HMVMX_EXIT_DECL hmR0VmxExitErrMsrLoad(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10746{
10747 NOREF(pVmxTransient);
10748 AssertMsgFailed(("Unexpected MSR-load exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); NOREF(pMixedCtx);
10749 HMVMX_RETURN_UNEXPECTED_EXIT();
10750}
10751
10752
10753/**
10754 * VM-exit handler for VM-entry failure due to a machine-check event
10755 * (VMX_EXIT_ERR_MACHINE_CHECK). Error VM-exit.
10756 */
10757HMVMX_EXIT_DECL hmR0VmxExitErrMachineCheck(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10758{
10759 NOREF(pVmxTransient);
10760 AssertMsgFailed(("Unexpected machine-check event exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); NOREF(pMixedCtx);
10761 HMVMX_RETURN_UNEXPECTED_EXIT();
10762}
10763
10764
10765/**
10766 * VM-exit handler for all undefined reasons. Should never ever happen.. in
10767 * theory.
10768 */
10769HMVMX_EXIT_DECL hmR0VmxExitErrUndefined(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10770{
10771 AssertMsgFailed(("Huh!? Undefined VM-exit reason %d. pVCpu=%p pMixedCtx=%p\n", pVmxTransient->uExitReason, pVCpu, pMixedCtx));
10772 NOREF(pVCpu); NOREF(pMixedCtx); NOREF(pVmxTransient);
10773 return VERR_VMX_UNDEFINED_EXIT_CODE;
10774}
10775
10776
10777/**
10778 * VM-exit handler for XDTR (LGDT, SGDT, LIDT, SIDT) accesses
10779 * (VMX_EXIT_XDTR_ACCESS) and LDT and TR access (LLDT, LTR, SLDT, STR).
10780 * Conditional VM-exit.
10781 */
10782HMVMX_EXIT_DECL hmR0VmxExitXdtrAccess(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10783{
10784 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10785
10786 /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT. */
10787 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitXdtrAccess);
10788 if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT)
10789 return VERR_EM_INTERPRETER;
10790 AssertMsgFailed(("Unexpected XDTR access. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10791 HMVMX_RETURN_UNEXPECTED_EXIT();
10792}
10793
10794
10795/**
10796 * VM-exit handler for RDRAND (VMX_EXIT_RDRAND). Conditional VM-exit.
10797 */
10798HMVMX_EXIT_DECL hmR0VmxExitRdrand(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10799{
10800 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10801
10802 /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT. */
10803 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdrand);
10804 if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT)
10805 return VERR_EM_INTERPRETER;
10806 AssertMsgFailed(("Unexpected RDRAND exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10807 HMVMX_RETURN_UNEXPECTED_EXIT();
10808}
10809
10810
10811/**
10812 * VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
10813 */
10814HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10815{
10816 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10817
10818 /* EMInterpretRdmsr() requires CR0, Eflags and SS segment register. */
10819 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10820 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10821 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10822 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
10823 {
10824 rc |= hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
10825 rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
10826 }
10827 AssertRCReturn(rc, rc);
10828 Log4(("ecx=%#RX32\n", pMixedCtx->ecx));
10829
10830#ifdef VBOX_STRICT
10831 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
10832 {
10833 if ( hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx)
10834 && pMixedCtx->ecx != MSR_K6_EFER)
10835 {
10836 AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n", pMixedCtx->ecx));
10837 HMVMX_RETURN_UNEXPECTED_EXIT();
10838 }
10839# if HC_ARCH_BITS == 64
10840 if ( pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests
10841 && hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
10842 {
10843 AssertMsgFailed(("Unexpected RDMSR for a passthru lazy-restore MSR. ecx=%#RX32\n", pMixedCtx->ecx));
10844 HMVMX_RETURN_UNEXPECTED_EXIT();
10845 }
10846# endif
10847 }
10848#endif
10849
10850 PVM pVM = pVCpu->CTX_SUFF(pVM);
10851 rc = EMInterpretRdmsr(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10852 AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER,
10853 ("hmR0VmxExitRdmsr: failed, invalid error code %Rrc\n", rc));
10854 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
10855 if (RT_LIKELY(rc == VINF_SUCCESS))
10856 {
10857 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10858 Assert(pVmxTransient->cbInstr == 2);
10859 }
10860 return rc;
10861}
10862
10863
10864/**
10865 * VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
10866 */
10867HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10868{
10869 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10870 PVM pVM = pVCpu->CTX_SUFF(pVM);
10871 int rc = VINF_SUCCESS;
10872
10873 /* EMInterpretWrmsr() requires CR0, EFLAGS and SS segment register. */
10874 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10875 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10876 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10877 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
10878 {
10879 rc |= hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
10880 rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
10881 }
10882 AssertRCReturn(rc, rc);
10883 Log4(("ecx=%#RX32 edx:eax=%#RX32:%#RX32\n", pMixedCtx->ecx, pMixedCtx->edx, pMixedCtx->eax));
10884
10885 rc = EMInterpretWrmsr(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10886 AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER, ("hmR0VmxExitWrmsr: failed, invalid error code %Rrc\n", rc));
10887 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
10888
10889 if (RT_LIKELY(rc == VINF_SUCCESS))
10890 {
10891 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10892
10893 /* If this is an X2APIC WRMSR access, update the APIC state as well. */
10894 if ( pMixedCtx->ecx >= MSR_IA32_X2APIC_START
10895 && pMixedCtx->ecx <= MSR_IA32_X2APIC_END)
10896 {
10897 /* We've already saved the APIC related guest-state (TPR) in hmR0VmxPostRunGuest(). When full APIC register
10898 * virtualization is implemented we'll have to make sure APIC state is saved from the VMCS before
10899 EMInterpretWrmsr() changes it. */
10900 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
10901 }
10902 else if (pMixedCtx->ecx == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */
10903 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
10904 else if (pMixedCtx->ecx == MSR_K6_EFER)
10905 {
10906 /*
10907 * If the guest touches EFER we need to update the VM-Entry and VM-Exit controls as well,
10908 * even if it is -not- touching bits that cause paging mode changes (LMA/LME). We care about
10909 * the other bits as well, SCE and NXE. See @bugref{7368}.
10910 */
10911 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_EFER_MSR | HM_CHANGED_VMX_ENTRY_CTLS | HM_CHANGED_VMX_EXIT_CTLS);
10912 }
10913
10914 /* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not supported. */
10915 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
10916 {
10917 switch (pMixedCtx->ecx)
10918 {
10919 case MSR_IA32_SYSENTER_CS: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
10920 case MSR_IA32_SYSENTER_EIP: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
10921 case MSR_IA32_SYSENTER_ESP: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
10922 case MSR_K8_FS_BASE: /* no break */
10923 case MSR_K8_GS_BASE: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS); break;
10924 case MSR_K6_EFER: /* already handled above */ break;
10925 default:
10926 {
10927 if (hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx))
10928 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
10929#if HC_ARCH_BITS == 64
10930 else if (hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
10931 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS);
10932#endif
10933 break;
10934 }
10935 }
10936 }
10937#ifdef VBOX_STRICT
10938 else
10939 {
10940 /* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */
10941 switch (pMixedCtx->ecx)
10942 {
10943 case MSR_IA32_SYSENTER_CS:
10944 case MSR_IA32_SYSENTER_EIP:
10945 case MSR_IA32_SYSENTER_ESP:
10946 case MSR_K8_FS_BASE:
10947 case MSR_K8_GS_BASE:
10948 {
10949 AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", pMixedCtx->ecx));
10950 HMVMX_RETURN_UNEXPECTED_EXIT();
10951 }
10952
10953 /* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */
10954 default:
10955 {
10956 if (hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx))
10957 {
10958 /* EFER writes are always intercepted, see hmR0VmxLoadGuestMsrs(). */
10959 if (pMixedCtx->ecx != MSR_K6_EFER)
10960 {
10961 AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
10962 pMixedCtx->ecx));
10963 HMVMX_RETURN_UNEXPECTED_EXIT();
10964 }
10965 }
10966
10967#if HC_ARCH_BITS == 64
10968 if (hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
10969 {
10970 AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", pMixedCtx->ecx));
10971 HMVMX_RETURN_UNEXPECTED_EXIT();
10972 }
10973#endif
10974 break;
10975 }
10976 }
10977 }
10978#endif /* VBOX_STRICT */
10979 }
10980 return rc;
10981}
10982
10983
10984/**
10985 * VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
10986 */
10987HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10988{
10989 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10990
10991 /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC_PAUSE_EXIT. */
10992 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPause);
10993 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_PAUSE_EXIT)
10994 return VERR_EM_INTERPRETER;
10995 AssertMsgFailed(("Unexpected PAUSE exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10996 HMVMX_RETURN_UNEXPECTED_EXIT();
10997}
10998
10999
11000/**
11001 * VM-exit handler for when the TPR value is lowered below the specified
11002 * threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
11003 */
11004HMVMX_EXIT_DECL hmR0VmxExitTprBelowThreshold(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11005{
11006 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11007 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW);
11008
11009 /*
11010 * The TPR has already been updated, see hmR0VMXPostRunGuest(). RIP is also updated as part of the VM-exit by VT-x. Update
11011 * the threshold in the VMCS, deliver the pending interrupt via hmR0VmxPreRunGuest()->hmR0VmxInjectPendingEvent() and
11012 * resume guest execution.
11013 */
11014 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
11015 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold);
11016 return VINF_SUCCESS;
11017}
11018
11019
11020/**
11021 * VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional
11022 * VM-exit.
11023 *
11024 * @retval VINF_SUCCESS when guest execution can continue.
11025 * @retval VINF_PGM_CHANGE_MODE when shadow paging mode changed, back to ring-3.
11026 * @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3.
11027 * @retval VERR_EM_INTERPRETER when something unexpected happened, fallback to
11028 * interpreter.
11029 */
11030HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11031{
11032 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11033 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2);
11034 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11035 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11036 AssertRCReturn(rc, rc);
11037
11038 RTGCUINTPTR const uExitQualification = pVmxTransient->uExitQualification;
11039 uint32_t const uAccessType = VMX_EXIT_QUALIFICATION_CRX_ACCESS(uExitQualification);
11040 PVM pVM = pVCpu->CTX_SUFF(pVM);
11041 VBOXSTRICTRC rcStrict;
11042 rc = hmR0VmxSaveGuestRegsForIemExec(pVCpu, pMixedCtx, false /*fMemory*/, true /*fNeedRsp*/);
11043 switch (uAccessType)
11044 {
11045 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_WRITE: /* MOV to CRx */
11046 {
11047 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
11048 AssertRCReturn(rc, rc);
11049
11050 rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, pVmxTransient->cbInstr,
11051 VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification),
11052 VMX_EXIT_QUALIFICATION_CRX_GENREG(uExitQualification));
11053 AssertMsg(rcStrict == VINF_SUCCESS || rcStrict == VINF_IEM_RAISED_XCPT || rcStrict == VINF_PGM_CHANGE_MODE
11054 || rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
11055 switch (VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification))
11056 {
11057 case 0: /* CR0 */
11058 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
11059 Log4(("CRX CR0 write rcStrict=%Rrc CR0=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->cr0));
11060 break;
11061 case 2: /* CR2 */
11062 /* Nothing to do here, CR2 it's not part of the VMCS. */
11063 break;
11064 case 3: /* CR3 */
11065 Assert(!pVM->hm.s.fNestedPaging || !CPUMIsGuestPagingEnabledEx(pMixedCtx));
11066 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR3);
11067 Log4(("CRX CR3 write rcStrict=%Rrc CR3=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->cr3));
11068 break;
11069 case 4: /* CR4 */
11070 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR4);
11071 Log4(("CRX CR4 write rc=%Rrc CR4=%#RX64 fLoadSaveGuestXcr0=%u\n",
11072 VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->cr4, pVCpu->hm.s.fLoadSaveGuestXcr0));
11073 break;
11074 case 8: /* CR8 */
11075 Assert(!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW));
11076 /* CR8 contains the APIC TPR. Was updated by IEMExecDecodedMovCRxWrite(). */
11077 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
11078 break;
11079 default:
11080 AssertMsgFailed(("Invalid CRx register %#x\n", VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)));
11081 break;
11082 }
11083
11084 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxWrite[VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)]);
11085 break;
11086 }
11087
11088 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_READ: /* MOV from CRx */
11089 {
11090 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
11091 AssertRCReturn(rc, rc);
11092
11093 Assert( !pVM->hm.s.fNestedPaging
11094 || !CPUMIsGuestPagingEnabledEx(pMixedCtx)
11095 || VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification) != 3);
11096
11097 /* CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */
11098 Assert( VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification) != 8
11099 || !(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW));
11100
11101 rcStrict = IEMExecDecodedMovCRxRead(pVCpu, pVmxTransient->cbInstr,
11102 VMX_EXIT_QUALIFICATION_CRX_GENREG(uExitQualification),
11103 VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification));
11104 AssertMsg(rcStrict == VINF_SUCCESS || rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
11105 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxRead[VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)]);
11106 Log4(("CRX CR%d Read access rcStrict=%Rrc\n", VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification),
11107 VBOXSTRICTRC_VAL(rcStrict)));
11108 break;
11109 }
11110
11111 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_CLTS: /* CLTS (Clear Task-Switch Flag in CR0) */
11112 {
11113 AssertRCReturn(rc, rc);
11114 rcStrict = IEMExecDecodedClts(pVCpu, pVmxTransient->cbInstr);
11115 AssertMsg(rcStrict == VINF_SUCCESS || rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
11116 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
11117 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts);
11118 Log4(("CRX CLTS rcStrict=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
11119 break;
11120 }
11121
11122 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */
11123 {
11124 AssertRCReturn(rc, rc);
11125 rcStrict = IEMExecDecodedLmsw(pVCpu, pVmxTransient->cbInstr,
11126 VMX_EXIT_QUALIFICATION_CRX_LMSW_DATA(uExitQualification));
11127 AssertMsg(rcStrict == VINF_SUCCESS || rcStrict == VINF_IEM_RAISED_XCPT || rcStrict == VINF_PGM_CHANGE_MODE, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
11128 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw);
11129 Log4(("CRX LMSW rcStrict=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
11130 break;
11131 }
11132
11133 default:
11134 AssertMsgFailedReturn(("Invalid access-type in Mov CRx VM-exit qualification %#x\n", uAccessType),
11135 VERR_VMX_UNEXPECTED_EXCEPTION);
11136 }
11137
11138 HMCPU_CF_SET(pVCpu, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS : HM_CHANGED_ALL_GUEST);
11139 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2);
11140 return VBOXSTRICTRC_TODO(rcStrict);
11141}
11142
11143
11144/**
11145 * VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional
11146 * VM-exit.
11147 */
11148HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11149{
11150 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11151 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1);
11152
11153 int rc2 = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11154 rc2 |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11155 rc2 |= hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
11156 rc2 |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); /* Eflag checks in EMInterpretDisasCurrent(). */
11157 rc2 |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); /* CR0 checks & PGM* in EMInterpretDisasCurrent(). */
11158 rc2 |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); /* SELM checks in EMInterpretDisasCurrent(). */
11159 /* EFER also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */
11160 AssertRCReturn(rc2, rc2);
11161
11162 /* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */
11163 uint32_t uIOPort = VMX_EXIT_QUALIFICATION_IO_PORT(pVmxTransient->uExitQualification);
11164 uint8_t uIOWidth = VMX_EXIT_QUALIFICATION_IO_WIDTH(pVmxTransient->uExitQualification);
11165 bool fIOWrite = ( VMX_EXIT_QUALIFICATION_IO_DIRECTION(pVmxTransient->uExitQualification)
11166 == VMX_EXIT_QUALIFICATION_IO_DIRECTION_OUT);
11167 bool fIOString = VMX_EXIT_QUALIFICATION_IO_IS_STRING(pVmxTransient->uExitQualification);
11168 bool fStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
11169 AssertReturn(uIOWidth <= 3 && uIOWidth != 2, VERR_VMX_IPE_1);
11170
11171 /* I/O operation lookup arrays. */
11172 static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses. */
11173 static uint32_t const s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving the result (in AL/AX/EAX). */
11174
11175 VBOXSTRICTRC rcStrict;
11176 uint32_t const cbValue = s_aIOSizes[uIOWidth];
11177 uint32_t const cbInstr = pVmxTransient->cbInstr;
11178 bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */
11179 PVM pVM = pVCpu->CTX_SUFF(pVM);
11180 if (fIOString)
11181 {
11182#ifdef VBOX_WITH_2ND_IEM_STEP /* This used to gurus with debian 32-bit guest without NP (on ATA reads). See @bugref{5752#c158}. Should work now. */
11183 /*
11184 * INS/OUTS - I/O String instruction.
11185 *
11186 * Use instruction-information if available, otherwise fall back on
11187 * interpreting the instruction.
11188 */
11189 Log4(("CS:RIP=%04x:%08RX64 %#06x/%u %c str\n", pMixedCtx->cs.Sel, pMixedCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
11190 AssertReturn(pMixedCtx->dx == uIOPort, VERR_VMX_IPE_2);
11191 if (MSR_IA32_VMX_BASIC_INFO_VMCS_INS_OUTS(pVM->hm.s.vmx.Msrs.u64BasicInfo))
11192 {
11193 rc2 = hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
11194 /** @todo optimize this, IEM should request the additional state if it needs it (GP, PF, ++). */
11195 rc2 |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11196 AssertRCReturn(rc2, rc2);
11197 AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_VMX_IPE_3);
11198 AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2);
11199 IEMMODE enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize;
11200 bool fRep = VMX_EXIT_QUALIFICATION_IO_IS_REP(pVmxTransient->uExitQualification);
11201 if (fIOWrite)
11202 {
11203 rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr,
11204 pVmxTransient->ExitInstrInfo.StrIo.iSegReg);
11205 }
11206 else
11207 {
11208 /*
11209 * The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES.
11210 * Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS.
11211 * See Intel Instruction spec. for "INS".
11212 * See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS".
11213 */
11214 rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr);
11215 }
11216 }
11217 else
11218 {
11219 /** @todo optimize this, IEM should request the additional state if it needs it (GP, PF, ++). */
11220 rc2 = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11221 AssertRCReturn(rc2, rc2);
11222 rcStrict = IEMExecOne(pVCpu);
11223 }
11224 /** @todo IEM needs to be setting these flags somehow. */
11225 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
11226 fUpdateRipAlready = true;
11227#else
11228 PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
11229 rcStrict = EMInterpretDisasCurrent(pVM, pVCpu, pDis, NULL /* pcbInstr */);
11230 if (RT_SUCCESS(rcStrict))
11231 {
11232 if (fIOWrite)
11233 {
11234 rcStrict = IOMInterpretOUTSEx(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), uIOPort, pDis->fPrefix,
11235 (DISCPUMODE)pDis->uAddrMode, cbValue);
11236 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringWrite);
11237 }
11238 else
11239 {
11240 rcStrict = IOMInterpretINSEx(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), uIOPort, pDis->fPrefix,
11241 (DISCPUMODE)pDis->uAddrMode, cbValue);
11242 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringRead);
11243 }
11244 }
11245 else
11246 {
11247 AssertMsg(rcStrict == VERR_EM_INTERPRETER, ("rcStrict=%Rrc RIP %#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->rip));
11248 rcStrict = VINF_EM_RAW_EMULATE_INSTR;
11249 }
11250#endif
11251 }
11252 else
11253 {
11254 /*
11255 * IN/OUT - I/O instruction.
11256 */
11257 Log4(("CS:RIP=%04x:%08RX64 %#06x/%u %c\n", pMixedCtx->cs.Sel, pMixedCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
11258 uint32_t const uAndVal = s_aIOOpAnd[uIOWidth];
11259 Assert(!VMX_EXIT_QUALIFICATION_IO_IS_REP(pVmxTransient->uExitQualification));
11260 if (fIOWrite)
11261 {
11262 rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pMixedCtx->eax & uAndVal, cbValue);
11263 if (rcStrict == VINF_IOM_R3_IOPORT_WRITE)
11264 HMR0SavePendingIOPortWrite(pVCpu, pMixedCtx->rip, pMixedCtx->rip + cbInstr, uIOPort, uAndVal, cbValue);
11265 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
11266 }
11267 else
11268 {
11269 uint32_t u32Result = 0;
11270 rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue);
11271 if (IOM_SUCCESS(rcStrict))
11272 {
11273 /* Save result of I/O IN instr. in AL/AX/EAX. */
11274 pMixedCtx->eax = (pMixedCtx->eax & ~uAndVal) | (u32Result & uAndVal);
11275 }
11276 else if (rcStrict == VINF_IOM_R3_IOPORT_READ)
11277 HMR0SavePendingIOPortRead(pVCpu, pMixedCtx->rip, pMixedCtx->rip + cbInstr, uIOPort, uAndVal, cbValue);
11278 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
11279 }
11280 }
11281
11282 if (IOM_SUCCESS(rcStrict))
11283 {
11284 if (!fUpdateRipAlready)
11285 {
11286 pMixedCtx->rip += cbInstr;
11287 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
11288 }
11289
11290 /*
11291 * INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru while booting Fedora 17 64-bit guest.
11292 * See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ.
11293 */
11294 if (fIOString)
11295 {
11296 /** @todo Single-step for INS/OUTS with REP prefix? */
11297 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
11298 }
11299 else if (fStepping)
11300 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
11301
11302 /*
11303 * If any I/O breakpoints are armed, we need to check if one triggered
11304 * and take appropriate action.
11305 * Note that the I/O breakpoint type is undefined if CR4.DE is 0.
11306 */
11307 rc2 = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
11308 AssertRCReturn(rc2, rc2);
11309
11310 /** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
11311 * execution engines about whether hyper BPs and such are pending. */
11312 uint32_t const uDr7 = pMixedCtx->dr[7];
11313 if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
11314 && X86_DR7_ANY_RW_IO(uDr7)
11315 && (pMixedCtx->cr4 & X86_CR4_DE))
11316 || DBGFBpIsHwIoArmed(pVM)))
11317 {
11318 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
11319
11320 /* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
11321 VMMRZCallRing3Disable(pVCpu);
11322 HM_DISABLE_PREEMPT();
11323
11324 bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /* fDr6 */);
11325
11326 VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pMixedCtx, uIOPort, cbValue);
11327 if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
11328 {
11329 /* Raise #DB. */
11330 if (fIsGuestDbgActive)
11331 ASMSetDR6(pMixedCtx->dr[6]);
11332 if (pMixedCtx->dr[7] != uDr7)
11333 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
11334
11335 hmR0VmxSetPendingXcptDB(pVCpu, pMixedCtx);
11336 }
11337 /* rcStrict is VINF_SUCCESS or in [VINF_EM_FIRST..VINF_EM_LAST]. */
11338 else if ( rcStrict2 != VINF_SUCCESS
11339 && (rcStrict == VINF_SUCCESS || rcStrict2 < rcStrict))
11340 rcStrict = rcStrict2;
11341
11342 HM_RESTORE_PREEMPT();
11343 VMMRZCallRing3Enable(pVCpu);
11344 }
11345 }
11346
11347#ifdef DEBUG
11348 if (rcStrict == VINF_IOM_R3_IOPORT_READ)
11349 Assert(!fIOWrite);
11350 else if (rcStrict == VINF_IOM_R3_IOPORT_WRITE)
11351 Assert(fIOWrite);
11352 else
11353 {
11354 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
11355 * statuses, that the VMM device and some others may return. See
11356 * IOM_SUCCESS() for guidance. */
11357 AssertMsg( RT_FAILURE(rcStrict)
11358 || rcStrict == VINF_SUCCESS
11359 || rcStrict == VINF_EM_RAW_EMULATE_INSTR
11360 || rcStrict == VINF_EM_DBG_BREAKPOINT
11361 || rcStrict == VINF_EM_RAW_GUEST_TRAP
11362 || rcStrict == VINF_EM_RAW_TO_R3
11363 || rcStrict == VINF_TRPM_XCPT_DISPATCHED, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
11364 }
11365#endif
11366
11367 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1);
11368 return VBOXSTRICTRC_TODO(rcStrict);
11369}
11370
11371
11372/**
11373 * VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional
11374 * VM-exit.
11375 */
11376HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11377{
11378 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11379
11380 /* Check if this task-switch occurred while delivery an event through the guest IDT. */
11381 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11382 AssertRCReturn(rc, rc);
11383 if (VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE_IDT)
11384 {
11385 rc = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
11386 AssertRCReturn(rc, rc);
11387 if (VMX_IDT_VECTORING_INFO_VALID(pVmxTransient->uIdtVectoringInfo))
11388 {
11389 uint32_t uIntType = VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo);
11390
11391 uint32_t uVector = VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo);
11392 bool fErrorCodeValid = VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uIdtVectoringInfo);
11393
11394 /* Save it as a pending event and it'll be converted to a TRPM event on the way out to ring-3. */
11395 Assert(!pVCpu->hm.s.Event.fPending);
11396 pVCpu->hm.s.Event.fPending = true;
11397 pVCpu->hm.s.Event.u64IntInfo = pVmxTransient->uIdtVectoringInfo;
11398 rc = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
11399 AssertRCReturn(rc, rc);
11400 if (fErrorCodeValid)
11401 pVCpu->hm.s.Event.u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
11402 else
11403 pVCpu->hm.s.Event.u32ErrCode = 0;
11404 if ( uIntType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
11405 && uVector == X86_XCPT_PF)
11406 {
11407 pVCpu->hm.s.Event.GCPtrFaultAddress = pMixedCtx->cr2;
11408 }
11409
11410 Log4(("Pending event on TaskSwitch uIntType=%#x uVector=%#x\n", uIntType, uVector));
11411 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
11412 return VINF_EM_RAW_INJECT_TRPM_EVENT;
11413 }
11414 }
11415
11416 /** @todo Emulate task switch someday, currently just going back to ring-3 for
11417 * emulation. */
11418 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
11419 return VERR_EM_INTERPRETER;
11420}
11421
11422
11423/**
11424 * VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit.
11425 */
11426HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11427{
11428 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11429 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG);
11430 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG;
11431 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
11432 AssertRCReturn(rc, rc);
11433 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMtf);
11434 return VINF_EM_DBG_STEPPED;
11435}
11436
11437
11438/**
11439 * VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit.
11440 */
11441HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11442{
11443 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11444
11445 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
11446 int rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
11447 if (RT_UNLIKELY(rc != VINF_SUCCESS))
11448 {
11449 if (rc == VINF_HM_DOUBLE_FAULT)
11450 rc = VINF_SUCCESS;
11451 return rc;
11452 }
11453
11454#if 0
11455 /** @todo Investigate if IOMMMIOPhysHandler() requires a lot of state, for now
11456 * just sync the whole thing. */
11457 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11458#else
11459 /* Aggressive state sync. for now. */
11460 rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
11461 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
11462 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11463#endif
11464 rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11465 AssertRCReturn(rc, rc);
11466
11467 /* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */
11468 uint32_t uAccessType = VMX_EXIT_QUALIFICATION_APIC_ACCESS_TYPE(pVmxTransient->uExitQualification);
11469 switch (uAccessType)
11470 {
11471 case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE:
11472 case VMX_APIC_ACCESS_TYPE_LINEAR_READ:
11473 {
11474 AssertMsg( !(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
11475 || VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification) != 0x80,
11476 ("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n"));
11477
11478 RTGCPHYS GCPhys = pMixedCtx->msrApicBase; /* Always up-to-date, msrApicBase is not part of the VMCS. */
11479 GCPhys &= PAGE_BASE_GC_MASK;
11480 GCPhys += VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification);
11481 PVM pVM = pVCpu->CTX_SUFF(pVM);
11482 Log4(("ApicAccess uAccessType=%#x GCPhys=%#RGv Off=%#x\n", uAccessType, GCPhys,
11483 VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification)));
11484
11485 VBOXSTRICTRC rc2 = IOMMMIOPhysHandler(pVM, pVCpu,
11486 uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ ? 0 : X86_TRAP_PF_RW,
11487 CPUMCTX2CORE(pMixedCtx), GCPhys);
11488 rc = VBOXSTRICTRC_VAL(rc2);
11489 Log4(("ApicAccess rc=%d\n", rc));
11490 if ( rc == VINF_SUCCESS
11491 || rc == VERR_PAGE_TABLE_NOT_PRESENT
11492 || rc == VERR_PAGE_NOT_PRESENT)
11493 {
11494 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
11495 | HM_CHANGED_GUEST_RSP
11496 | HM_CHANGED_GUEST_RFLAGS
11497 | HM_CHANGED_VMX_GUEST_APIC_STATE);
11498 rc = VINF_SUCCESS;
11499 }
11500 break;
11501 }
11502
11503 default:
11504 Log4(("ApicAccess uAccessType=%#x\n", uAccessType));
11505 rc = VINF_EM_RAW_EMULATE_INSTR;
11506 break;
11507 }
11508
11509 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess);
11510 if (rc != VINF_SUCCESS)
11511 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchApicAccessToR3);
11512 return rc;
11513}
11514
11515
11516/**
11517 * VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional
11518 * VM-exit.
11519 */
11520HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11521{
11522 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11523
11524 /* We should -not- get this VM-exit if the guest's debug registers were active. */
11525 if (pVmxTransient->fWasGuestDebugStateActive)
11526 {
11527 AssertMsgFailed(("Unexpected MOV DRx exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
11528 HMVMX_RETURN_UNEXPECTED_EXIT();
11529 }
11530
11531 int rc = VERR_INTERNAL_ERROR_5;
11532 if ( !DBGFIsStepping(pVCpu)
11533 && !pVCpu->hm.s.fSingleInstruction
11534 && !pVmxTransient->fWasHyperDebugStateActive)
11535 {
11536 /* Don't intercept MOV DRx and #DB any more. */
11537 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
11538 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
11539 AssertRCReturn(rc, rc);
11540
11541 if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
11542 {
11543#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
11544 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_DB);
11545 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
11546#endif
11547 }
11548
11549 /* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */
11550 VMMRZCallRing3Disable(pVCpu);
11551 HM_DISABLE_PREEMPT();
11552
11553 /* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
11554 CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
11555 Assert(CPUMIsGuestDebugStateActive(pVCpu) || HC_ARCH_BITS == 32);
11556
11557 HM_RESTORE_PREEMPT();
11558 VMMRZCallRing3Enable(pVCpu);
11559
11560#ifdef VBOX_WITH_STATISTICS
11561 rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11562 AssertRCReturn(rc, rc);
11563 if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
11564 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
11565 else
11566 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
11567#endif
11568 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
11569 return VINF_SUCCESS;
11570 }
11571
11572 /*
11573 * EMInterpretDRx[Write|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER, CS. EFER is always up-to-date.
11574 * Update the segment registers and DR7 from the CPU.
11575 */
11576 rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11577 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11578 rc |= hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
11579 AssertRCReturn(rc, rc);
11580 Log4(("CS:RIP=%04x:%08RX64\n", pMixedCtx->cs.Sel, pMixedCtx->rip));
11581
11582 PVM pVM = pVCpu->CTX_SUFF(pVM);
11583 if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
11584 {
11585 rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx),
11586 VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification),
11587 VMX_EXIT_QUALIFICATION_DRX_GENREG(pVmxTransient->uExitQualification));
11588 if (RT_SUCCESS(rc))
11589 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
11590 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
11591 }
11592 else
11593 {
11594 rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx),
11595 VMX_EXIT_QUALIFICATION_DRX_GENREG(pVmxTransient->uExitQualification),
11596 VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification));
11597 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
11598 }
11599
11600 Assert(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER);
11601 if (RT_SUCCESS(rc))
11602 {
11603 int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11604 AssertRCReturn(rc2, rc2);
11605 }
11606 return rc;
11607}
11608
11609
11610/**
11611 * VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG).
11612 * Conditional VM-exit.
11613 */
11614HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11615{
11616 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11617 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
11618
11619 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
11620 int rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
11621 if (RT_UNLIKELY(rc != VINF_SUCCESS))
11622 {
11623 if (rc == VINF_HM_DOUBLE_FAULT)
11624 rc = VINF_SUCCESS;
11625 return rc;
11626 }
11627
11628 RTGCPHYS GCPhys = 0;
11629 rc = VMXReadVmcs64(VMX_VMCS64_EXIT_GUEST_PHYS_ADDR_FULL, &GCPhys);
11630
11631#if 0
11632 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); /** @todo Can we do better? */
11633#else
11634 /* Aggressive state sync. for now. */
11635 rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
11636 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
11637 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11638#endif
11639 AssertRCReturn(rc, rc);
11640
11641 /*
11642 * If we succeed, resume guest execution.
11643 * If we fail in interpreting the instruction because we couldn't get the guest physical address
11644 * of the page containing the instruction via the guest's page tables (we would invalidate the guest page
11645 * in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
11646 * weird case. See @bugref{6043}.
11647 */
11648 PVM pVM = pVCpu->CTX_SUFF(pVM);
11649 VBOXSTRICTRC rc2 = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pMixedCtx), GCPhys, UINT32_MAX);
11650 rc = VBOXSTRICTRC_VAL(rc2);
11651 Log4(("EPT misconfig at %#RGv RIP=%#RX64 rc=%d\n", GCPhys, pMixedCtx->rip, rc));
11652 if ( rc == VINF_SUCCESS
11653 || rc == VERR_PAGE_TABLE_NOT_PRESENT
11654 || rc == VERR_PAGE_NOT_PRESENT)
11655 {
11656 /* Successfully handled MMIO operation. */
11657 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
11658 | HM_CHANGED_GUEST_RSP
11659 | HM_CHANGED_GUEST_RFLAGS
11660 | HM_CHANGED_VMX_GUEST_APIC_STATE);
11661 rc = VINF_SUCCESS;
11662 }
11663 return rc;
11664}
11665
11666
11667/**
11668 * VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional
11669 * VM-exit.
11670 */
11671HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11672{
11673 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11674 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
11675
11676 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
11677 int rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
11678 if (RT_UNLIKELY(rc != VINF_SUCCESS))
11679 {
11680 if (rc == VINF_HM_DOUBLE_FAULT)
11681 rc = VINF_SUCCESS;
11682 return rc;
11683 }
11684
11685 RTGCPHYS GCPhys = 0;
11686 rc = VMXReadVmcs64(VMX_VMCS64_EXIT_GUEST_PHYS_ADDR_FULL, &GCPhys);
11687 rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11688#if 0
11689 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); /** @todo Can we do better? */
11690#else
11691 /* Aggressive state sync. for now. */
11692 rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
11693 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
11694 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11695#endif
11696 AssertRCReturn(rc, rc);
11697
11698 /* Intel spec. Table 27-7 "Exit Qualifications for EPT violations". */
11699 AssertMsg(((pVmxTransient->uExitQualification >> 7) & 3) != 2, ("%#RX64", pVmxTransient->uExitQualification));
11700
11701 RTGCUINT uErrorCode = 0;
11702 if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_INSTR_FETCH)
11703 uErrorCode |= X86_TRAP_PF_ID;
11704 if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_DATA_WRITE)
11705 uErrorCode |= X86_TRAP_PF_RW;
11706 if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_ENTRY_PRESENT)
11707 uErrorCode |= X86_TRAP_PF_P;
11708
11709 TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode);
11710
11711 Log4(("EPT violation %#x at %#RX64 ErrorCode %#x CS:RIP=%04x:%08RX64\n", pVmxTransient->uExitQualification, GCPhys,
11712 uErrorCode, pMixedCtx->cs.Sel, pMixedCtx->rip));
11713
11714 /* Handle the pagefault trap for the nested shadow table. */
11715 PVM pVM = pVCpu->CTX_SUFF(pVM);
11716 rc = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pMixedCtx), GCPhys);
11717 TRPMResetTrap(pVCpu);
11718
11719 /* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */
11720 if ( rc == VINF_SUCCESS
11721 || rc == VERR_PAGE_TABLE_NOT_PRESENT
11722 || rc == VERR_PAGE_NOT_PRESENT)
11723 {
11724 /* Successfully synced our nested page tables. */
11725 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf);
11726 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
11727 | HM_CHANGED_GUEST_RSP
11728 | HM_CHANGED_GUEST_RFLAGS);
11729 return VINF_SUCCESS;
11730 }
11731
11732 Log4(("EPT return to ring-3 rc=%Rrc\n", rc));
11733 return rc;
11734}
11735
11736/** @} */
11737
11738/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-= */
11739/* -=-=-=-=-=-=-=-=-=- VM-exit Exception Handlers -=-=-=-=-=-=-=-=-=-=- */
11740/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-= */
11741
11742/** @name VM-exit exception handlers.
11743 * @{
11744 */
11745
11746/**
11747 * VM-exit exception handler for #MF (Math Fault: floating point exception).
11748 */
11749static int hmR0VmxExitXcptMF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11750{
11751 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11752 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
11753
11754 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11755 AssertRCReturn(rc, rc);
11756
11757 if (!(pMixedCtx->cr0 & X86_CR0_NE))
11758 {
11759 /* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
11760 rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
11761
11762 /** @todo r=ramshankar: The Intel spec. does -not- specify that this VM-exit
11763 * provides VM-exit instruction length. If this causes problem later,
11764 * disassemble the instruction like it's done on AMD-V. */
11765 int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11766 AssertRCReturn(rc2, rc2);
11767 return rc;
11768 }
11769
11770 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11771 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
11772 return rc;
11773}
11774
11775
11776/**
11777 * VM-exit exception handler for #BP (Breakpoint exception).
11778 */
11779static int hmR0VmxExitXcptBP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11780{
11781 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11782 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
11783
11784 /** @todo Try optimize this by not saving the entire guest state unless
11785 * really needed. */
11786 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11787 AssertRCReturn(rc, rc);
11788
11789 PVM pVM = pVCpu->CTX_SUFF(pVM);
11790 rc = DBGFRZTrap03Handler(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11791 if (rc == VINF_EM_RAW_GUEST_TRAP)
11792 {
11793 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11794 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11795 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11796 AssertRCReturn(rc, rc);
11797
11798 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11799 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
11800 }
11801
11802 Assert(rc == VINF_SUCCESS || rc == VINF_EM_RAW_GUEST_TRAP || rc == VINF_EM_DBG_BREAKPOINT);
11803 return rc;
11804}
11805
11806
11807/**
11808 * VM-exit exception handler for #DB (Debug exception).
11809 */
11810static int hmR0VmxExitXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11811{
11812 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11813 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
11814 Log6(("XcptDB\n"));
11815
11816 /*
11817 * Get the DR6-like values from the VM-exit qualification and pass it to DBGF
11818 * for processing.
11819 */
11820 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11821 AssertRCReturn(rc, rc);
11822
11823 /* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */
11824 uint64_t uDR6 = X86_DR6_INIT_VAL;
11825 uDR6 |= ( pVmxTransient->uExitQualification
11826 & (X86_DR6_B0 | X86_DR6_B1 | X86_DR6_B2 | X86_DR6_B3 | X86_DR6_BD | X86_DR6_BS));
11827
11828 rc = DBGFRZTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pMixedCtx), uDR6, pVCpu->hm.s.fSingleInstruction);
11829 if (rc == VINF_EM_RAW_GUEST_TRAP)
11830 {
11831 /*
11832 * The exception was for the guest. Update DR6, DR7.GD and
11833 * IA32_DEBUGCTL.LBR before forwarding it.
11834 * (See Intel spec. 27.1 "Architectural State before a VM-Exit".)
11835 */
11836 VMMRZCallRing3Disable(pVCpu);
11837 HM_DISABLE_PREEMPT();
11838
11839 pMixedCtx->dr[6] &= ~X86_DR6_B_MASK;
11840 pMixedCtx->dr[6] |= uDR6;
11841 if (CPUMIsGuestDebugStateActive(pVCpu))
11842 ASMSetDR6(pMixedCtx->dr[6]);
11843
11844 HM_RESTORE_PREEMPT();
11845 VMMRZCallRing3Enable(pVCpu);
11846
11847 rc = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
11848 AssertRCReturn(rc, rc);
11849
11850 /* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
11851 pMixedCtx->dr[7] &= ~X86_DR7_GD;
11852
11853 /* Paranoia. */
11854 pMixedCtx->dr[7] &= ~X86_DR7_RAZ_MASK;
11855 pMixedCtx->dr[7] |= X86_DR7_RA1_MASK;
11856
11857 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)pMixedCtx->dr[7]);
11858 AssertRCReturn(rc, rc);
11859
11860 /*
11861 * Raise #DB in the guest.
11862 *
11863 * It is important to reflect what the VM-exit gave us (preserving the interruption-type) rather than use
11864 * hmR0VmxSetPendingXcptDB() as the #DB could've been raised while executing ICEBP and not the 'normal' #DB.
11865 * Thus it -may- trigger different handling in the CPU (like skipped DPL checks). See @bugref{6398}.
11866 *
11867 * Since ICEBP isn't documented on Intel, see AMD spec. 15.20 "Event Injection".
11868 */
11869 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11870 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11871 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11872 AssertRCReturn(rc, rc);
11873 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11874 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
11875 return VINF_SUCCESS;
11876 }
11877
11878 /*
11879 * Not a guest trap, must be a hypervisor related debug event then.
11880 * Update DR6 in case someone is interested in it.
11881 */
11882 AssertMsg(rc == VINF_EM_DBG_STEPPED || rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc));
11883 AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5);
11884 CPUMSetHyperDR6(pVCpu, uDR6);
11885
11886 return rc;
11887}
11888
11889
11890/**
11891 * VM-exit exception handler for #NM (Device-not-available exception: floating
11892 * point exception).
11893 */
11894static int hmR0VmxExitXcptNM(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11895{
11896 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11897
11898 /* We require CR0 and EFER. EFER is always up-to-date. */
11899 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11900 AssertRCReturn(rc, rc);
11901
11902 /* We're playing with the host CPU state here, have to disable preemption or longjmp. */
11903 VMMRZCallRing3Disable(pVCpu);
11904 HM_DISABLE_PREEMPT();
11905
11906 /* If the guest FPU was active at the time of the #NM exit, then it's a guest fault. */
11907 if (pVmxTransient->fWasGuestFPUStateActive)
11908 {
11909 rc = VINF_EM_RAW_GUEST_TRAP;
11910 Assert(CPUMIsGuestFPUStateActive(pVCpu) || HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0));
11911 }
11912 else
11913 {
11914#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
11915 Assert(!pVmxTransient->fWasGuestFPUStateActive);
11916#endif
11917 rc = CPUMR0Trap07Handler(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx);
11918 Assert(rc == VINF_EM_RAW_GUEST_TRAP || (rc == VINF_SUCCESS && CPUMIsGuestFPUStateActive(pVCpu)));
11919 }
11920
11921 HM_RESTORE_PREEMPT();
11922 VMMRZCallRing3Enable(pVCpu);
11923
11924 if (rc == VINF_SUCCESS)
11925 {
11926 /* Guest FPU state was activated, we'll want to change CR0 FPU intercepts before the next VM-reentry. */
11927 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
11928 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowNM);
11929 pVCpu->hm.s.fPreloadGuestFpu = true;
11930 }
11931 else
11932 {
11933 /* Forward #NM to the guest. */
11934 Assert(rc == VINF_EM_RAW_GUEST_TRAP);
11935 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11936 AssertRCReturn(rc, rc);
11937 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11938 pVmxTransient->cbInstr, 0 /* error code */, 0 /* GCPtrFaultAddress */);
11939 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNM);
11940 }
11941
11942 return VINF_SUCCESS;
11943}
11944
11945
11946/**
11947 * VM-exit exception handler for #GP (General-protection exception).
11948 *
11949 * @remarks Requires pVmxTransient->uExitIntInfo to be up-to-date.
11950 */
11951static int hmR0VmxExitXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11952{
11953 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11954 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
11955
11956 int rc = VERR_INTERNAL_ERROR_5;
11957 if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
11958 {
11959#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
11960 /* If the guest is not in real-mode or we have unrestricted execution support, reflect #GP to the guest. */
11961 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11962 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11963 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11964 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11965 AssertRCReturn(rc, rc);
11966 Log4(("#GP Gst: CS:RIP %04x:%08RX64 ErrorCode=%#x CR0=%#RX64 CPL=%u TR=%#04x\n", pMixedCtx->cs.Sel, pMixedCtx->rip,
11967 pVmxTransient->uExitIntErrorCode, pMixedCtx->cr0, CPUMGetGuestCPL(pVCpu), pMixedCtx->tr.Sel));
11968 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11969 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
11970 return rc;
11971#else
11972 /* We don't intercept #GP. */
11973 AssertMsgFailed(("Unexpected VM-exit caused by #GP exception\n"));
11974 NOREF(pVmxTransient);
11975 return VERR_VMX_UNEXPECTED_EXCEPTION;
11976#endif
11977 }
11978
11979 Assert(CPUMIsGuestInRealModeEx(pMixedCtx));
11980 Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest);
11981
11982 /* EMInterpretDisasCurrent() requires a lot of the state, save the entire state. */
11983 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11984 AssertRCReturn(rc, rc);
11985
11986 PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
11987 uint32_t cbOp = 0;
11988 PVM pVM = pVCpu->CTX_SUFF(pVM);
11989 rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
11990 if (RT_SUCCESS(rc))
11991 {
11992 rc = VINF_SUCCESS;
11993 Assert(cbOp == pDis->cbInstr);
11994 Log4(("#GP Disas OpCode=%u CS:EIP %04x:%04RX64\n", pDis->pCurInstr->uOpcode, pMixedCtx->cs.Sel, pMixedCtx->rip));
11995 switch (pDis->pCurInstr->uOpcode)
11996 {
11997 case OP_CLI:
11998 {
11999 pMixedCtx->eflags.Bits.u1IF = 0;
12000 pMixedCtx->eflags.Bits.u1RF = 0;
12001 pMixedCtx->rip += pDis->cbInstr;
12002 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS);
12003 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
12004 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCli);
12005 break;
12006 }
12007
12008 case OP_STI:
12009 {
12010 bool fOldIF = pMixedCtx->eflags.Bits.u1IF;
12011 pMixedCtx->eflags.Bits.u1IF = 1;
12012 pMixedCtx->eflags.Bits.u1RF = 0;
12013 pMixedCtx->rip += pDis->cbInstr;
12014 if (!fOldIF)
12015 {
12016 EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip);
12017 Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS));
12018 }
12019 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS);
12020 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
12021 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitSti);
12022 break;
12023 }
12024
12025 case OP_HLT:
12026 {
12027 rc = VINF_EM_HALT;
12028 pMixedCtx->rip += pDis->cbInstr;
12029 pMixedCtx->eflags.Bits.u1RF = 0;
12030 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS);
12031 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt);
12032 break;
12033 }
12034
12035 case OP_POPF:
12036 {
12037 Log4(("POPF CS:EIP %04x:%04RX64\n", pMixedCtx->cs.Sel, pMixedCtx->rip));
12038 uint32_t cbParm;
12039 uint32_t uMask;
12040 bool fStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
12041 if (pDis->fPrefix & DISPREFIX_OPSIZE)
12042 {
12043 cbParm = 4;
12044 uMask = 0xffffffff;
12045 }
12046 else
12047 {
12048 cbParm = 2;
12049 uMask = 0xffff;
12050 }
12051
12052 /* Get the stack pointer & pop the contents of the stack onto Eflags. */
12053 RTGCPTR GCPtrStack = 0;
12054 X86EFLAGS Eflags;
12055 rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), pMixedCtx->esp & uMask, SELMTOFLAT_FLAGS_CPL0,
12056 &GCPtrStack);
12057 if (RT_SUCCESS(rc))
12058 {
12059 Assert(sizeof(Eflags.u32) >= cbParm);
12060 Eflags.u32 = 0;
12061 rc = VBOXSTRICTRC_TODO(PGMPhysRead(pVM, (RTGCPHYS)GCPtrStack, &Eflags.u32, cbParm, PGMACCESSORIGIN_HM));
12062 AssertMsg(rc == VINF_SUCCESS, ("%Rrc\n", rc)); /** @todo allow strict return codes here */
12063 }
12064 if (RT_FAILURE(rc))
12065 {
12066 rc = VERR_EM_INTERPRETER;
12067 break;
12068 }
12069 Log4(("POPF %#x -> %#RX64 mask=%#x RIP=%#RX64\n", Eflags.u, pMixedCtx->rsp, uMask, pMixedCtx->rip));
12070 pMixedCtx->eflags.u32 = (pMixedCtx->eflags.u32 & ~((X86_EFL_POPF_BITS & uMask) | X86_EFL_RF))
12071 | (Eflags.u32 & X86_EFL_POPF_BITS & uMask);
12072 pMixedCtx->esp += cbParm;
12073 pMixedCtx->esp &= uMask;
12074 pMixedCtx->rip += pDis->cbInstr;
12075 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
12076 | HM_CHANGED_GUEST_RSP
12077 | HM_CHANGED_GUEST_RFLAGS);
12078 /* Generate a pending-debug exception when stepping over POPF regardless of how POPF modifies EFLAGS.TF. */
12079 if (fStepping)
12080 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
12081
12082 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPopf);
12083 break;
12084 }
12085
12086 case OP_PUSHF:
12087 {
12088 uint32_t cbParm;
12089 uint32_t uMask;
12090 if (pDis->fPrefix & DISPREFIX_OPSIZE)
12091 {
12092 cbParm = 4;
12093 uMask = 0xffffffff;
12094 }
12095 else
12096 {
12097 cbParm = 2;
12098 uMask = 0xffff;
12099 }
12100
12101 /* Get the stack pointer & push the contents of eflags onto the stack. */
12102 RTGCPTR GCPtrStack = 0;
12103 rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), (pMixedCtx->esp - cbParm) & uMask,
12104 SELMTOFLAT_FLAGS_CPL0, &GCPtrStack);
12105 if (RT_FAILURE(rc))
12106 {
12107 rc = VERR_EM_INTERPRETER;
12108 break;
12109 }
12110 X86EFLAGS Eflags = pMixedCtx->eflags;
12111 /* The RF & VM bits are cleared on image stored on stack; see Intel Instruction reference for PUSHF. */
12112 Eflags.Bits.u1RF = 0;
12113 Eflags.Bits.u1VM = 0;
12114
12115 rc = VBOXSTRICTRC_TODO(PGMPhysWrite(pVM, (RTGCPHYS)GCPtrStack, &Eflags.u, cbParm, PGMACCESSORIGIN_HM));
12116 if (RT_UNLIKELY(rc != VINF_SUCCESS))
12117 {
12118 AssertMsgFailed(("%Rrc\n", rc)); /** @todo allow strict return codes here */
12119 rc = VERR_EM_INTERPRETER;
12120 break;
12121 }
12122 Log4(("PUSHF %#x -> %#RGv\n", Eflags.u, GCPtrStack));
12123 pMixedCtx->esp -= cbParm;
12124 pMixedCtx->esp &= uMask;
12125 pMixedCtx->rip += pDis->cbInstr;
12126 pMixedCtx->eflags.Bits.u1RF = 0;
12127 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
12128 | HM_CHANGED_GUEST_RSP
12129 | HM_CHANGED_GUEST_RFLAGS);
12130 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
12131 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPushf);
12132 break;
12133 }
12134
12135 case OP_IRET:
12136 {
12137 /** @todo Handle 32-bit operand sizes and check stack limits. See Intel
12138 * instruction reference. */
12139 RTGCPTR GCPtrStack = 0;
12140 uint32_t uMask = 0xffff;
12141 bool fStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
12142 uint16_t aIretFrame[3];
12143 if (pDis->fPrefix & (DISPREFIX_OPSIZE | DISPREFIX_ADDRSIZE))
12144 {
12145 rc = VERR_EM_INTERPRETER;
12146 break;
12147 }
12148 rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), pMixedCtx->esp & uMask, SELMTOFLAT_FLAGS_CPL0,
12149 &GCPtrStack);
12150 if (RT_SUCCESS(rc))
12151 {
12152 rc = VBOXSTRICTRC_TODO(PGMPhysRead(pVM, (RTGCPHYS)GCPtrStack, &aIretFrame[0], sizeof(aIretFrame),
12153 PGMACCESSORIGIN_HM));
12154 AssertMsg(rc == VINF_SUCCESS, ("%Rrc\n", rc)); /** @todo allow strict return codes here */
12155 }
12156 if (RT_FAILURE(rc))
12157 {
12158 rc = VERR_EM_INTERPRETER;
12159 break;
12160 }
12161 pMixedCtx->eip = 0;
12162 pMixedCtx->ip = aIretFrame[0];
12163 pMixedCtx->cs.Sel = aIretFrame[1];
12164 pMixedCtx->cs.ValidSel = aIretFrame[1];
12165 pMixedCtx->cs.u64Base = (uint64_t)pMixedCtx->cs.Sel << 4;
12166 pMixedCtx->eflags.u32 = (pMixedCtx->eflags.u32 & ((UINT32_C(0xffff0000) | X86_EFL_1) & ~X86_EFL_RF))
12167 | (aIretFrame[2] & X86_EFL_POPF_BITS & uMask);
12168 pMixedCtx->sp += sizeof(aIretFrame);
12169 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
12170 | HM_CHANGED_GUEST_SEGMENT_REGS
12171 | HM_CHANGED_GUEST_RSP
12172 | HM_CHANGED_GUEST_RFLAGS);
12173 /* Generate a pending-debug exception when stepping over IRET regardless of how IRET modifies EFLAGS.TF. */
12174 if (fStepping)
12175 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
12176 Log4(("IRET %#RX32 to %04x:%04x\n", GCPtrStack, pMixedCtx->cs.Sel, pMixedCtx->ip));
12177 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIret);
12178 break;
12179 }
12180
12181 case OP_INT:
12182 {
12183 uint16_t uVector = pDis->Param1.uValue & 0xff;
12184 hmR0VmxSetPendingIntN(pVCpu, pMixedCtx, uVector, pDis->cbInstr);
12185 /* INT clears EFLAGS.TF, we mustn't set any pending debug exceptions here. */
12186 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInt);
12187 break;
12188 }
12189
12190 case OP_INTO:
12191 {
12192 if (pMixedCtx->eflags.Bits.u1OF)
12193 {
12194 hmR0VmxSetPendingXcptOF(pVCpu, pMixedCtx, pDis->cbInstr);
12195 /* INTO clears EFLAGS.TF, we mustn't set any pending debug exceptions here. */
12196 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInt);
12197 }
12198 else
12199 {
12200 pMixedCtx->eflags.Bits.u1RF = 0;
12201 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
12202 }
12203 break;
12204 }
12205
12206 default:
12207 {
12208 pMixedCtx->eflags.Bits.u1RF = 0; /* This is correct most of the time... */
12209 VBOXSTRICTRC rc2 = EMInterpretInstructionDisasState(pVCpu, pDis, CPUMCTX2CORE(pMixedCtx), 0 /* pvFault */,
12210 EMCODETYPE_SUPERVISOR);
12211 rc = VBOXSTRICTRC_VAL(rc2);
12212 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
12213 /** @todo We have to set pending-debug exceptions here when the guest is
12214 * single-stepping depending on the instruction that was interpreted. */
12215 Log4(("#GP rc=%Rrc\n", rc));
12216 break;
12217 }
12218 }
12219 }
12220 else
12221 rc = VERR_EM_INTERPRETER;
12222
12223 AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER || rc == VINF_PGM_CHANGE_MODE || rc == VINF_EM_HALT,
12224 ("#GP Unexpected rc=%Rrc\n", rc));
12225 return rc;
12226}
12227
12228
12229#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
12230/**
12231 * VM-exit exception handler wrapper for generic exceptions. Simply re-injects
12232 * the exception reported in the VMX transient structure back into the VM.
12233 *
12234 * @remarks Requires uExitIntInfo in the VMX transient structure to be
12235 * up-to-date.
12236 */
12237static int hmR0VmxExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12238{
12239 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
12240
12241 /* Re-inject the exception into the guest. This cannot be a double-fault condition which would have been handled in
12242 hmR0VmxCheckExitDueToEventDelivery(). */
12243 int rc = hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
12244 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
12245 AssertRCReturn(rc, rc);
12246 Assert(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO);
12247
12248#ifdef DEBUG_ramshankar
12249 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12250 uint8_t uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo);
12251 Log(("hmR0VmxExitXcptGeneric: Reinjecting Xcpt. uVector=%#x cs:rip=%#04x:%#RX64\n", uVector, pCtx->cs.Sel, pCtx->rip));
12252#endif
12253
12254 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
12255 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
12256 return VINF_SUCCESS;
12257}
12258#endif
12259
12260
12261/**
12262 * VM-exit exception handler for #PF (Page-fault exception).
12263 */
12264static int hmR0VmxExitXcptPF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12265{
12266 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
12267 PVM pVM = pVCpu->CTX_SUFF(pVM);
12268 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12269 rc |= hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
12270 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
12271 AssertRCReturn(rc, rc);
12272
12273#if defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) || defined(HMVMX_ALWAYS_TRAP_PF)
12274 if (pVM->hm.s.fNestedPaging)
12275 {
12276 pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
12277 if (RT_LIKELY(!pVmxTransient->fVectoringDoublePF))
12278 {
12279 pMixedCtx->cr2 = pVmxTransient->uExitQualification; /* Update here in case we go back to ring-3 before injection. */
12280 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
12281 0 /* cbInstr */, pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQualification);
12282 }
12283 else
12284 {
12285 /* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
12286 hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
12287 Log4(("Pending #DF due to vectoring #PF. NP\n"));
12288 }
12289 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
12290 return rc;
12291 }
12292#else
12293 Assert(!pVM->hm.s.fNestedPaging);
12294 NOREF(pVM);
12295#endif
12296
12297 /* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
12298 of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
12299 if (pVmxTransient->fVectoringPF)
12300 {
12301 Assert(pVCpu->hm.s.Event.fPending);
12302 return VINF_EM_RAW_INJECT_TRPM_EVENT;
12303 }
12304
12305 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
12306 AssertRCReturn(rc, rc);
12307
12308 Log4(("#PF: cr2=%#RX64 cs:rip=%#04x:%#RX64 uErrCode %#RX32 cr3=%#RX64\n", pVmxTransient->uExitQualification,
12309 pMixedCtx->cs.Sel, pMixedCtx->rip, pVmxTransient->uExitIntErrorCode, pMixedCtx->cr3));
12310
12311 TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQualification, (RTGCUINT)pVmxTransient->uExitIntErrorCode);
12312 rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pMixedCtx),
12313 (RTGCPTR)pVmxTransient->uExitQualification);
12314
12315 Log4(("#PF: rc=%Rrc\n", rc));
12316 if (rc == VINF_SUCCESS)
12317 {
12318 /* Successfully synced shadow pages tables or emulated an MMIO instruction. */
12319 /** @todo this isn't quite right, what if guest does lgdt with some MMIO
12320 * memory? We don't update the whole state here... */
12321 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
12322 | HM_CHANGED_GUEST_RSP
12323 | HM_CHANGED_GUEST_RFLAGS
12324 | HM_CHANGED_VMX_GUEST_APIC_STATE);
12325 TRPMResetTrap(pVCpu);
12326 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
12327 return rc;
12328 }
12329
12330 if (rc == VINF_EM_RAW_GUEST_TRAP)
12331 {
12332 if (!pVmxTransient->fVectoringDoublePF)
12333 {
12334 /* It's a guest page fault and needs to be reflected to the guest. */
12335 uint32_t uGstErrorCode = TRPMGetErrorCode(pVCpu);
12336 TRPMResetTrap(pVCpu);
12337 pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */
12338 pMixedCtx->cr2 = pVmxTransient->uExitQualification; /* Update here in case we go back to ring-3 before injection. */
12339 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
12340 0 /* cbInstr */, uGstErrorCode, pVmxTransient->uExitQualification);
12341 }
12342 else
12343 {
12344 /* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
12345 TRPMResetTrap(pVCpu);
12346 pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */
12347 hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
12348 Log4(("#PF: Pending #DF due to vectoring #PF\n"));
12349 }
12350
12351 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
12352 return VINF_SUCCESS;
12353 }
12354
12355 TRPMResetTrap(pVCpu);
12356 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
12357 return rc;
12358}
12359
12360/** @} */
12361
Note: See TracBrowser for help on using the repository browser.

© 2025 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette