VirtualBox

source: vbox/trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp@ 52573

Last change on this file since 52573 was 52573, checked in by vboxsync, 10 years ago

VMM/HMVMXR0: Fix a wrong invalid-guest-state check with regards to EFER.LMA and LME.
  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 497.8 KB
Line 
1/* $Id: HMVMXR0.cpp 52573 2014-09-02 16:16:48Z vboxsync $ */
2/** @file
3 * HM VMX (Intel VT-x) - Host Context Ring-0.
4 */
5
6/*
7 * Copyright (C) 2012-2014 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 "HMInternal.h"
27#include <VBox/vmm/vm.h>
28#include "HMVMXR0.h"
29#include <VBox/vmm/pdmapi.h>
30#include <VBox/vmm/dbgf.h>
31#include <VBox/vmm/iem.h>
32#include <VBox/vmm/iom.h>
33#include <VBox/vmm/selm.h>
34#include <VBox/vmm/tm.h>
35#include <VBox/vmm/gim.h>
36#ifdef VBOX_WITH_REM
37# include <VBox/vmm/rem.h>
38#endif
39#ifdef DEBUG_ramshankar
40# define HMVMX_ALWAYS_SAVE_GUEST_RFLAGS
41# define HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE
42# define HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
43# define HMVMX_ALWAYS_CHECK_GUEST_STATE
44# define HMVMX_ALWAYS_TRAP_ALL_XCPTS
45# define HMVMX_ALWAYS_TRAP_PF
46# define HMVMX_ALWAYS_SWAP_FPU_STATE
47# define HMVMX_ALWAYS_FLUSH_TLB
48# define HMVMX_ALWAYS_SWAP_EFER
49#endif
50
51
52/*******************************************************************************
53* Defined Constants And Macros *
54*******************************************************************************/
55#if defined(RT_ARCH_AMD64)
56# define HMVMX_IS_64BIT_HOST_MODE() (true)
57typedef RTHCUINTREG HMVMXHCUINTREG;
58#elif defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
59extern "C" uint32_t g_fVMXIs64bitHost;
60# define HMVMX_IS_64BIT_HOST_MODE() (g_fVMXIs64bitHost != 0)
61typedef uint64_t HMVMXHCUINTREG;
62#else
63# define HMVMX_IS_64BIT_HOST_MODE() (false)
64typedef RTHCUINTREG HMVMXHCUINTREG;
65#endif
66
67/** Use the function table. */
68#define HMVMX_USE_FUNCTION_TABLE
69
70/** Determine which tagged-TLB flush handler to use. */
71#define HMVMX_FLUSH_TAGGED_TLB_EPT_VPID 0
72#define HMVMX_FLUSH_TAGGED_TLB_EPT 1
73#define HMVMX_FLUSH_TAGGED_TLB_VPID 2
74#define HMVMX_FLUSH_TAGGED_TLB_NONE 3
75
76/** @name Updated-guest-state flags.
77 * @{ */
78#define HMVMX_UPDATED_GUEST_RIP RT_BIT(0)
79#define HMVMX_UPDATED_GUEST_RSP RT_BIT(1)
80#define HMVMX_UPDATED_GUEST_RFLAGS RT_BIT(2)
81#define HMVMX_UPDATED_GUEST_CR0 RT_BIT(3)
82#define HMVMX_UPDATED_GUEST_CR3 RT_BIT(4)
83#define HMVMX_UPDATED_GUEST_CR4 RT_BIT(5)
84#define HMVMX_UPDATED_GUEST_GDTR RT_BIT(6)
85#define HMVMX_UPDATED_GUEST_IDTR RT_BIT(7)
86#define HMVMX_UPDATED_GUEST_LDTR RT_BIT(8)
87#define HMVMX_UPDATED_GUEST_TR RT_BIT(9)
88#define HMVMX_UPDATED_GUEST_SEGMENT_REGS RT_BIT(10)
89#define HMVMX_UPDATED_GUEST_DEBUG RT_BIT(11)
90#define HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR RT_BIT(12)
91#define HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR RT_BIT(13)
92#define HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR RT_BIT(14)
93#define HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS RT_BIT(15)
94#define HMVMX_UPDATED_GUEST_LAZY_MSRS RT_BIT(16)
95#define HMVMX_UPDATED_GUEST_ACTIVITY_STATE RT_BIT(17)
96#define HMVMX_UPDATED_GUEST_INTR_STATE RT_BIT(18)
97#define HMVMX_UPDATED_GUEST_APIC_STATE RT_BIT(19)
98#define HMVMX_UPDATED_GUEST_ALL ( HMVMX_UPDATED_GUEST_RIP \
99 | HMVMX_UPDATED_GUEST_RSP \
100 | HMVMX_UPDATED_GUEST_RFLAGS \
101 | HMVMX_UPDATED_GUEST_CR0 \
102 | HMVMX_UPDATED_GUEST_CR3 \
103 | HMVMX_UPDATED_GUEST_CR4 \
104 | HMVMX_UPDATED_GUEST_GDTR \
105 | HMVMX_UPDATED_GUEST_IDTR \
106 | HMVMX_UPDATED_GUEST_LDTR \
107 | HMVMX_UPDATED_GUEST_TR \
108 | HMVMX_UPDATED_GUEST_SEGMENT_REGS \
109 | HMVMX_UPDATED_GUEST_DEBUG \
110 | HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR \
111 | HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR \
112 | HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR \
113 | HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS \
114 | HMVMX_UPDATED_GUEST_LAZY_MSRS \
115 | HMVMX_UPDATED_GUEST_ACTIVITY_STATE \
116 | HMVMX_UPDATED_GUEST_INTR_STATE \
117 | HMVMX_UPDATED_GUEST_APIC_STATE)
118/** @} */
119
120/** @name
121 * Flags to skip redundant reads of some common VMCS fields that are not part of
122 * the guest-CPU state but are in the transient structure.
123 */
124#define HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO RT_BIT(0)
125#define HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE RT_BIT(1)
126#define HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION RT_BIT(2)
127#define HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN RT_BIT(3)
128#define HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO RT_BIT(4)
129#define HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE RT_BIT(5)
130#define HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO RT_BIT(6)
131/** @} */
132
133/** @name
134 * States of the VMCS.
135 *
136 * This does not reflect all possible VMCS states but currently only those
137 * needed for maintaining the VMCS consistently even when thread-context hooks
138 * are used. Maybe later this can be extended (i.e. Nested Virtualization).
139 */
140#define HMVMX_VMCS_STATE_CLEAR RT_BIT(0)
141#define HMVMX_VMCS_STATE_ACTIVE RT_BIT(1)
142#define HMVMX_VMCS_STATE_LAUNCHED RT_BIT(2)
143/** @} */
144
145/**
146 * Exception bitmap mask for real-mode guests (real-on-v86).
147 *
148 * We need to intercept all exceptions manually (except #PF). #NM is also
149 * handled separately, see hmR0VmxLoadSharedCR0(). #PF need not be intercepted
150 * even in real-mode if we have Nested Paging support.
151 */
152#define HMVMX_REAL_MODE_XCPT_MASK ( RT_BIT(X86_XCPT_DE) | RT_BIT(X86_XCPT_DB) | RT_BIT(X86_XCPT_NMI) \
153 | RT_BIT(X86_XCPT_BP) | RT_BIT(X86_XCPT_OF) | RT_BIT(X86_XCPT_BR) \
154 | RT_BIT(X86_XCPT_UD) /* RT_BIT(X86_XCPT_NM) */ | RT_BIT(X86_XCPT_DF) \
155 | RT_BIT(X86_XCPT_CO_SEG_OVERRUN) | RT_BIT(X86_XCPT_TS) | RT_BIT(X86_XCPT_NP) \
156 | RT_BIT(X86_XCPT_SS) | RT_BIT(X86_XCPT_GP) /* RT_BIT(X86_XCPT_PF) */ \
157 | RT_BIT(X86_XCPT_MF) | RT_BIT(X86_XCPT_AC) | RT_BIT(X86_XCPT_MC) \
158 | RT_BIT(X86_XCPT_XF))
159
160/**
161 * Exception bitmap mask for all contributory exceptions.
162 *
163 * Page fault is deliberately excluded here as it's conditional as to whether
164 * it's contributory or benign. Page faults are handled separately.
165 */
166#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) \
167 | RT_BIT(X86_XCPT_DE))
168
169/** Maximum VM-instruction error number. */
170#define HMVMX_INSTR_ERROR_MAX 28
171
172/** Profiling macro. */
173#ifdef HM_PROFILE_EXIT_DISPATCH
174# define HMVMX_START_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitDispatch, ed)
175# define HMVMX_STOP_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitDispatch, ed)
176#else
177# define HMVMX_START_EXIT_DISPATCH_PROF() do { } while (0)
178# define HMVMX_STOP_EXIT_DISPATCH_PROF() do { } while (0)
179#endif
180
181/** Assert that preemption is disabled or covered by thread-context hooks. */
182#define HMVMX_ASSERT_PREEMPT_SAFE() Assert( VMMR0ThreadCtxHooksAreRegistered(pVCpu) \
183 || !RTThreadPreemptIsEnabled(NIL_RTTHREAD));
184
185/** Assert that we haven't migrated CPUs when thread-context hooks are not
186 * used. */
187#define HMVMX_ASSERT_CPU_SAFE() AssertMsg( VMMR0ThreadCtxHooksAreRegistered(pVCpu) \
188 || pVCpu->hm.s.idEnteredCpu == RTMpCpuId(), \
189 ("Illegal migration! Entered on CPU %u Current %u\n", \
190 pVCpu->hm.s.idEnteredCpu, RTMpCpuId())); \
191
192/** Helper macro for VM-exit handlers called unexpectedly. */
193#define HMVMX_RETURN_UNEXPECTED_EXIT() \
194 do { \
195 pVCpu->hm.s.u32HMError = pVmxTransient->uExitReason; \
196 return VERR_VMX_UNEXPECTED_EXIT; \
197 } while (0)
198
199
200/*******************************************************************************
201* Structures and Typedefs *
202*******************************************************************************/
203/**
204 * VMX transient state.
205 *
206 * A state structure for holding miscellaneous information across
207 * VMX non-root operation and restored after the transition.
208 */
209typedef struct VMXTRANSIENT
210{
211 /** The host's rflags/eflags. */
212 RTCCUINTREG uEflags;
213#if HC_ARCH_BITS == 32
214 uint32_t u32Alignment0;
215#endif
216 /** The guest's TPR value used for TPR shadowing. */
217 uint8_t u8GuestTpr;
218 /** Alignment. */
219 uint8_t abAlignment0[7];
220
221 /** The basic VM-exit reason. */
222 uint16_t uExitReason;
223 /** Alignment. */
224 uint16_t u16Alignment0;
225 /** The VM-exit interruption error code. */
226 uint32_t uExitIntErrorCode;
227 /** The VM-exit exit qualification. */
228 uint64_t uExitQualification;
229
230 /** The VM-exit interruption-information field. */
231 uint32_t uExitIntInfo;
232 /** The VM-exit instruction-length field. */
233 uint32_t cbInstr;
234 /** The VM-exit instruction-information field. */
235 union
236 {
237 /** Plain unsigned int representation. */
238 uint32_t u;
239 /** INS and OUTS information. */
240 struct
241 {
242 uint32_t u6Reserved0 : 7;
243 /** The address size; 0=16-bit, 1=32-bit, 2=64-bit, rest undefined. */
244 uint32_t u3AddrSize : 3;
245 uint32_t u5Reserved1 : 5;
246 /** The segment register (X86_SREG_XXX). */
247 uint32_t iSegReg : 3;
248 uint32_t uReserved2 : 14;
249 } StrIo;
250 } ExitInstrInfo;
251 /** Whether the VM-entry failed or not. */
252 bool fVMEntryFailed;
253 /** Alignment. */
254 uint8_t abAlignment1[3];
255
256 /** The VM-entry interruption-information field. */
257 uint32_t uEntryIntInfo;
258 /** The VM-entry exception error code field. */
259 uint32_t uEntryXcptErrorCode;
260 /** The VM-entry instruction length field. */
261 uint32_t cbEntryInstr;
262
263 /** IDT-vectoring information field. */
264 uint32_t uIdtVectoringInfo;
265 /** IDT-vectoring error code. */
266 uint32_t uIdtVectoringErrorCode;
267
268 /** Mask of currently read VMCS fields; HMVMX_UPDATED_TRANSIENT_*. */
269 uint32_t fVmcsFieldsRead;
270
271 /** Whether the guest FPU was active at the time of VM-exit. */
272 bool fWasGuestFPUStateActive;
273 /** Whether the guest debug state was active at the time of VM-exit. */
274 bool fWasGuestDebugStateActive;
275 /** Whether the hyper debug state was active at the time of VM-exit. */
276 bool fWasHyperDebugStateActive;
277 /** Whether TSC-offsetting should be setup before VM-entry. */
278 bool fUpdateTscOffsettingAndPreemptTimer;
279 /** Whether the VM-exit was caused by a page-fault during delivery of a
280 * contributory exception or a page-fault. */
281 bool fVectoringPF;
282} VMXTRANSIENT;
283AssertCompileMemberAlignment(VMXTRANSIENT, uExitReason, sizeof(uint64_t));
284AssertCompileMemberAlignment(VMXTRANSIENT, uExitIntInfo, sizeof(uint64_t));
285AssertCompileMemberAlignment(VMXTRANSIENT, uEntryIntInfo, sizeof(uint64_t));
286AssertCompileMemberAlignment(VMXTRANSIENT, fWasGuestFPUStateActive, sizeof(uint64_t));
287AssertCompileMemberSize(VMXTRANSIENT, ExitInstrInfo, sizeof(uint32_t));
288/** Pointer to VMX transient state. */
289typedef VMXTRANSIENT *PVMXTRANSIENT;
290
291
292/**
293 * MSR-bitmap read permissions.
294 */
295typedef enum VMXMSREXITREAD
296{
297 /** Reading this MSR causes a VM-exit. */
298 VMXMSREXIT_INTERCEPT_READ = 0xb,
299 /** Reading this MSR does not cause a VM-exit. */
300 VMXMSREXIT_PASSTHRU_READ
301} VMXMSREXITREAD;
302/** Pointer to MSR-bitmap read permissions. */
303typedef VMXMSREXITREAD* PVMXMSREXITREAD;
304
305/**
306 * MSR-bitmap write permissions.
307 */
308typedef enum VMXMSREXITWRITE
309{
310 /** Writing to this MSR causes a VM-exit. */
311 VMXMSREXIT_INTERCEPT_WRITE = 0xd,
312 /** Writing to this MSR does not cause a VM-exit. */
313 VMXMSREXIT_PASSTHRU_WRITE
314} VMXMSREXITWRITE;
315/** Pointer to MSR-bitmap write permissions. */
316typedef VMXMSREXITWRITE* PVMXMSREXITWRITE;
317
318
319/**
320 * VMX VM-exit handler.
321 *
322 * @returns VBox status code.
323 * @param pVCpu Pointer to the VMCPU.
324 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
325 * out-of-sync. Make sure to update the required
326 * fields before using them.
327 * @param pVmxTransient Pointer to the VMX-transient structure.
328 */
329#ifndef HMVMX_USE_FUNCTION_TABLE
330typedef int FNVMXEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
331#else
332typedef DECLCALLBACK(int) FNVMXEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
333/** Pointer to VM-exit handler. */
334typedef FNVMXEXITHANDLER *PFNVMXEXITHANDLER;
335#endif
336
337
338/*******************************************************************************
339* Internal Functions *
340*******************************************************************************/
341static void hmR0VmxFlushEpt(PVMCPU pVCpu, VMXFLUSHEPT enmFlush);
342static void hmR0VmxFlushVpid(PVM pVM, PVMCPU pVCpu, VMXFLUSHVPID enmFlush, RTGCPTR GCPtr);
343static int hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint64_t u64IntInfo, uint32_t cbInstr,
344 uint32_t u32ErrCode, RTGCUINTREG GCPtrFaultAddress, uint32_t *puIntState);
345#if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
346static int hmR0VmxInitVmcsReadCache(PVM pVM, PVMCPU pVCpu);
347#endif
348#ifndef HMVMX_USE_FUNCTION_TABLE
349DECLINLINE(int) hmR0VmxHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, uint32_t rcReason);
350# define HMVMX_EXIT_DECL static int
351#else
352# define HMVMX_EXIT_DECL static DECLCALLBACK(int)
353#endif
354
355/** @name VM-exit handlers.
356 * @{
357 */
358static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmi;
359static FNVMXEXITHANDLER hmR0VmxExitExtInt;
360static FNVMXEXITHANDLER hmR0VmxExitTripleFault;
361static FNVMXEXITHANDLER hmR0VmxExitInitSignal;
362static FNVMXEXITHANDLER hmR0VmxExitSipi;
363static FNVMXEXITHANDLER hmR0VmxExitIoSmi;
364static FNVMXEXITHANDLER hmR0VmxExitSmi;
365static FNVMXEXITHANDLER hmR0VmxExitIntWindow;
366static FNVMXEXITHANDLER hmR0VmxExitNmiWindow;
367static FNVMXEXITHANDLER hmR0VmxExitTaskSwitch;
368static FNVMXEXITHANDLER hmR0VmxExitCpuid;
369static FNVMXEXITHANDLER hmR0VmxExitGetsec;
370static FNVMXEXITHANDLER hmR0VmxExitHlt;
371static FNVMXEXITHANDLER hmR0VmxExitInvd;
372static FNVMXEXITHANDLER hmR0VmxExitInvlpg;
373static FNVMXEXITHANDLER hmR0VmxExitRdpmc;
374static FNVMXEXITHANDLER hmR0VmxExitVmcall;
375static FNVMXEXITHANDLER hmR0VmxExitRdtsc;
376static FNVMXEXITHANDLER hmR0VmxExitRsm;
377static FNVMXEXITHANDLER hmR0VmxExitSetPendingXcptUD;
378static FNVMXEXITHANDLER hmR0VmxExitMovCRx;
379static FNVMXEXITHANDLER hmR0VmxExitMovDRx;
380static FNVMXEXITHANDLER hmR0VmxExitIoInstr;
381static FNVMXEXITHANDLER hmR0VmxExitRdmsr;
382static FNVMXEXITHANDLER hmR0VmxExitWrmsr;
383static FNVMXEXITHANDLER hmR0VmxExitErrInvalidGuestState;
384static FNVMXEXITHANDLER hmR0VmxExitErrMsrLoad;
385static FNVMXEXITHANDLER hmR0VmxExitErrUndefined;
386static FNVMXEXITHANDLER hmR0VmxExitMwait;
387static FNVMXEXITHANDLER hmR0VmxExitMtf;
388static FNVMXEXITHANDLER hmR0VmxExitMonitor;
389static FNVMXEXITHANDLER hmR0VmxExitPause;
390static FNVMXEXITHANDLER hmR0VmxExitErrMachineCheck;
391static FNVMXEXITHANDLER hmR0VmxExitTprBelowThreshold;
392static FNVMXEXITHANDLER hmR0VmxExitApicAccess;
393static FNVMXEXITHANDLER hmR0VmxExitXdtrAccess;
394static FNVMXEXITHANDLER hmR0VmxExitXdtrAccess;
395static FNVMXEXITHANDLER hmR0VmxExitEptViolation;
396static FNVMXEXITHANDLER hmR0VmxExitEptMisconfig;
397static FNVMXEXITHANDLER hmR0VmxExitRdtscp;
398static FNVMXEXITHANDLER hmR0VmxExitPreemptTimer;
399static FNVMXEXITHANDLER hmR0VmxExitWbinvd;
400static FNVMXEXITHANDLER hmR0VmxExitXsetbv;
401static FNVMXEXITHANDLER hmR0VmxExitRdrand;
402static FNVMXEXITHANDLER hmR0VmxExitInvpcid;
403/** @} */
404
405static int hmR0VmxExitXcptNM(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
406static int hmR0VmxExitXcptPF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
407static int hmR0VmxExitXcptMF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
408static int hmR0VmxExitXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
409static int hmR0VmxExitXcptBP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
410static int hmR0VmxExitXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
411#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
412static int hmR0VmxExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
413#endif
414static uint32_t hmR0VmxCheckGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx);
415
416/*******************************************************************************
417* Global Variables *
418*******************************************************************************/
419#ifdef HMVMX_USE_FUNCTION_TABLE
420
421/**
422 * VMX_EXIT dispatch table.
423 */
424static const PFNVMXEXITHANDLER g_apfnVMExitHandlers[VMX_EXIT_MAX + 1] =
425{
426 /* 00 VMX_EXIT_XCPT_OR_NMI */ hmR0VmxExitXcptOrNmi,
427 /* 01 VMX_EXIT_EXT_INT */ hmR0VmxExitExtInt,
428 /* 02 VMX_EXIT_TRIPLE_FAULT */ hmR0VmxExitTripleFault,
429 /* 03 VMX_EXIT_INIT_SIGNAL */ hmR0VmxExitInitSignal,
430 /* 04 VMX_EXIT_SIPI */ hmR0VmxExitSipi,
431 /* 05 VMX_EXIT_IO_SMI */ hmR0VmxExitIoSmi,
432 /* 06 VMX_EXIT_SMI */ hmR0VmxExitSmi,
433 /* 07 VMX_EXIT_INT_WINDOW */ hmR0VmxExitIntWindow,
434 /* 08 VMX_EXIT_NMI_WINDOW */ hmR0VmxExitNmiWindow,
435 /* 09 VMX_EXIT_TASK_SWITCH */ hmR0VmxExitTaskSwitch,
436 /* 10 VMX_EXIT_CPUID */ hmR0VmxExitCpuid,
437 /* 11 VMX_EXIT_GETSEC */ hmR0VmxExitGetsec,
438 /* 12 VMX_EXIT_HLT */ hmR0VmxExitHlt,
439 /* 13 VMX_EXIT_INVD */ hmR0VmxExitInvd,
440 /* 14 VMX_EXIT_INVLPG */ hmR0VmxExitInvlpg,
441 /* 15 VMX_EXIT_RDPMC */ hmR0VmxExitRdpmc,
442 /* 16 VMX_EXIT_RDTSC */ hmR0VmxExitRdtsc,
443 /* 17 VMX_EXIT_RSM */ hmR0VmxExitRsm,
444 /* 18 VMX_EXIT_VMCALL */ hmR0VmxExitVmcall,
445 /* 19 VMX_EXIT_VMCLEAR */ hmR0VmxExitSetPendingXcptUD,
446 /* 20 VMX_EXIT_VMLAUNCH */ hmR0VmxExitSetPendingXcptUD,
447 /* 21 VMX_EXIT_VMPTRLD */ hmR0VmxExitSetPendingXcptUD,
448 /* 22 VMX_EXIT_VMPTRST */ hmR0VmxExitSetPendingXcptUD,
449 /* 23 VMX_EXIT_VMREAD */ hmR0VmxExitSetPendingXcptUD,
450 /* 24 VMX_EXIT_VMRESUME */ hmR0VmxExitSetPendingXcptUD,
451 /* 25 VMX_EXIT_VMWRITE */ hmR0VmxExitSetPendingXcptUD,
452 /* 26 VMX_EXIT_VMXOFF */ hmR0VmxExitSetPendingXcptUD,
453 /* 27 VMX_EXIT_VMXON */ hmR0VmxExitSetPendingXcptUD,
454 /* 28 VMX_EXIT_MOV_CRX */ hmR0VmxExitMovCRx,
455 /* 29 VMX_EXIT_MOV_DRX */ hmR0VmxExitMovDRx,
456 /* 30 VMX_EXIT_IO_INSTR */ hmR0VmxExitIoInstr,
457 /* 31 VMX_EXIT_RDMSR */ hmR0VmxExitRdmsr,
458 /* 32 VMX_EXIT_WRMSR */ hmR0VmxExitWrmsr,
459 /* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ hmR0VmxExitErrInvalidGuestState,
460 /* 34 VMX_EXIT_ERR_MSR_LOAD */ hmR0VmxExitErrMsrLoad,
461 /* 35 UNDEFINED */ hmR0VmxExitErrUndefined,
462 /* 36 VMX_EXIT_MWAIT */ hmR0VmxExitMwait,
463 /* 37 VMX_EXIT_MTF */ hmR0VmxExitMtf,
464 /* 38 UNDEFINED */ hmR0VmxExitErrUndefined,
465 /* 39 VMX_EXIT_MONITOR */ hmR0VmxExitMonitor,
466 /* 40 UNDEFINED */ hmR0VmxExitPause,
467 /* 41 VMX_EXIT_PAUSE */ hmR0VmxExitErrMachineCheck,
468 /* 42 VMX_EXIT_ERR_MACHINE_CHECK */ hmR0VmxExitErrUndefined,
469 /* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ hmR0VmxExitTprBelowThreshold,
470 /* 44 VMX_EXIT_APIC_ACCESS */ hmR0VmxExitApicAccess,
471 /* 45 UNDEFINED */ hmR0VmxExitErrUndefined,
472 /* 46 VMX_EXIT_XDTR_ACCESS */ hmR0VmxExitXdtrAccess,
473 /* 47 VMX_EXIT_TR_ACCESS */ hmR0VmxExitXdtrAccess,
474 /* 48 VMX_EXIT_EPT_VIOLATION */ hmR0VmxExitEptViolation,
475 /* 49 VMX_EXIT_EPT_MISCONFIG */ hmR0VmxExitEptMisconfig,
476 /* 50 VMX_EXIT_INVEPT */ hmR0VmxExitSetPendingXcptUD,
477 /* 51 VMX_EXIT_RDTSCP */ hmR0VmxExitRdtscp,
478 /* 52 VMX_EXIT_PREEMPT_TIMER */ hmR0VmxExitPreemptTimer,
479 /* 53 VMX_EXIT_INVVPID */ hmR0VmxExitSetPendingXcptUD,
480 /* 54 VMX_EXIT_WBINVD */ hmR0VmxExitWbinvd,
481 /* 55 VMX_EXIT_XSETBV */ hmR0VmxExitXsetbv,
482 /* 56 UNDEFINED */ hmR0VmxExitErrUndefined,
483 /* 57 VMX_EXIT_RDRAND */ hmR0VmxExitRdrand,
484 /* 58 VMX_EXIT_INVPCID */ hmR0VmxExitInvpcid,
485 /* 59 VMX_EXIT_VMFUNC */ hmR0VmxExitSetPendingXcptUD
486};
487#endif /* HMVMX_USE_FUNCTION_TABLE */
488
489#ifdef VBOX_STRICT
490static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] =
491{
492 /* 0 */ "(Not Used)",
493 /* 1 */ "VMCALL executed in VMX root operation.",
494 /* 2 */ "VMCLEAR with invalid physical address.",
495 /* 3 */ "VMCLEAR with VMXON pointer.",
496 /* 4 */ "VMLAUNCH with non-clear VMCS.",
497 /* 5 */ "VMRESUME with non-launched VMCS.",
498 /* 6 */ "VMRESUME after VMXOFF",
499 /* 7 */ "VM entry with invalid control fields.",
500 /* 8 */ "VM entry with invalid host state fields.",
501 /* 9 */ "VMPTRLD with invalid physical address.",
502 /* 10 */ "VMPTRLD with VMXON pointer.",
503 /* 11 */ "VMPTRLD with incorrect revision identifier.",
504 /* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.",
505 /* 13 */ "VMWRITE to read-only VMCS component.",
506 /* 14 */ "(Not Used)",
507 /* 15 */ "VMXON executed in VMX root operation.",
508 /* 16 */ "VM entry with invalid executive-VMCS pointer.",
509 /* 17 */ "VM entry with non-launched executing VMCS.",
510 /* 18 */ "VM entry with executive-VMCS pointer not VMXON pointer.",
511 /* 19 */ "VMCALL with non-clear VMCS.",
512 /* 20 */ "VMCALL with invalid VM-exit control fields.",
513 /* 21 */ "(Not Used)",
514 /* 22 */ "VMCALL with incorrect MSEG revision identifier.",
515 /* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.",
516 /* 24 */ "VMCALL with invalid SMM-monitor features.",
517 /* 25 */ "VM entry with invalid VM-execution control fields in executive VMCS.",
518 /* 26 */ "VM entry with events blocked by MOV SS.",
519 /* 27 */ "(Not Used)",
520 /* 28 */ "Invalid operand to INVEPT/INVVPID."
521};
522#endif /* VBOX_STRICT */
523
524
525
526/**
527 * Updates the VM's last error record. If there was a VMX instruction error,
528 * reads the error data from the VMCS and updates VCPU's last error record as
529 * well.
530 *
531 * @param pVM Pointer to the VM.
532 * @param pVCpu Pointer to the VMCPU (can be NULL if @a rc is not
533 * VERR_VMX_UNABLE_TO_START_VM or
534 * VERR_VMX_INVALID_VMCS_FIELD).
535 * @param rc The error code.
536 */
537static void hmR0VmxUpdateErrorRecord(PVM pVM, PVMCPU pVCpu, int rc)
538{
539 AssertPtr(pVM);
540 if ( rc == VERR_VMX_INVALID_VMCS_FIELD
541 || rc == VERR_VMX_UNABLE_TO_START_VM)
542 {
543 AssertPtrReturnVoid(pVCpu);
544 VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
545 }
546 pVM->hm.s.lLastError = rc;
547}
548
549
550/**
551 * Reads the VM-entry interruption-information field from the VMCS into the VMX
552 * transient structure.
553 *
554 * @returns VBox status code.
555 * @param pVmxTransient Pointer to the VMX transient structure.
556 *
557 * @remarks No-long-jump zone!!!
558 */
559DECLINLINE(int) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
560{
561 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo);
562 AssertRCReturn(rc, rc);
563 return VINF_SUCCESS;
564}
565
566
567/**
568 * Reads the VM-entry exception error code field from the VMCS into
569 * the VMX transient structure.
570 *
571 * @returns VBox status code.
572 * @param pVmxTransient Pointer to the VMX transient structure.
573 *
574 * @remarks No-long-jump zone!!!
575 */
576DECLINLINE(int) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
577{
578 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode);
579 AssertRCReturn(rc, rc);
580 return VINF_SUCCESS;
581}
582
583
584/**
585 * Reads the VM-entry exception error code field from the VMCS into
586 * the VMX transient structure.
587 *
588 * @returns VBox status code.
589 * @param pVmxTransient Pointer to the VMX transient structure.
590 *
591 * @remarks No-long-jump zone!!!
592 */
593DECLINLINE(int) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
594{
595 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr);
596 AssertRCReturn(rc, rc);
597 return VINF_SUCCESS;
598}
599
600
601/**
602 * Reads the VM-exit interruption-information field from the VMCS into the VMX
603 * transient structure.
604 *
605 * @returns VBox status code.
606 * @param pVmxTransient Pointer to the VMX transient structure.
607 */
608DECLINLINE(int) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
609{
610 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO))
611 {
612 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
613 AssertRCReturn(rc, rc);
614 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO;
615 }
616 return VINF_SUCCESS;
617}
618
619
620/**
621 * Reads the VM-exit interruption error code from the VMCS into the VMX
622 * transient structure.
623 *
624 * @returns VBox status code.
625 * @param pVmxTransient Pointer to the VMX transient structure.
626 */
627DECLINLINE(int) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
628{
629 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE))
630 {
631 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
632 AssertRCReturn(rc, rc);
633 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE;
634 }
635 return VINF_SUCCESS;
636}
637
638
639/**
640 * Reads the VM-exit instruction length field from the VMCS into the VMX
641 * transient structure.
642 *
643 * @returns VBox status code.
644 * @param pVCpu Pointer to the VMCPU.
645 * @param pVmxTransient Pointer to the VMX transient structure.
646 */
647DECLINLINE(int) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
648{
649 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN))
650 {
651 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbInstr);
652 AssertRCReturn(rc, rc);
653 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN;
654 }
655 return VINF_SUCCESS;
656}
657
658
659/**
660 * Reads the VM-exit instruction-information field from the VMCS into
661 * the VMX transient structure.
662 *
663 * @returns VBox status code.
664 * @param pVmxTransient Pointer to the VMX transient structure.
665 */
666DECLINLINE(int) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient)
667{
668 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO))
669 {
670 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
671 AssertRCReturn(rc, rc);
672 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO;
673 }
674 return VINF_SUCCESS;
675}
676
677
678/**
679 * Reads the exit qualification from the VMCS into the VMX transient structure.
680 *
681 * @returns VBox status code.
682 * @param pVCpu Pointer to the VMCPU (required for the VMCS cache
683 * case).
684 * @param pVmxTransient Pointer to the VMX transient structure.
685 */
686DECLINLINE(int) hmR0VmxReadExitQualificationVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
687{
688 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION))
689 {
690 int rc = VMXReadVmcsGstN(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQualification); NOREF(pVCpu);
691 AssertRCReturn(rc, rc);
692 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION;
693 }
694 return VINF_SUCCESS;
695}
696
697
698/**
699 * Reads the IDT-vectoring information field from the VMCS into the VMX
700 * transient structure.
701 *
702 * @returns VBox status code.
703 * @param pVmxTransient Pointer to the VMX transient structure.
704 *
705 * @remarks No-long-jump zone!!!
706 */
707DECLINLINE(int) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient)
708{
709 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO))
710 {
711 int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_INFO, &pVmxTransient->uIdtVectoringInfo);
712 AssertRCReturn(rc, rc);
713 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO;
714 }
715 return VINF_SUCCESS;
716}
717
718
719/**
720 * Reads the IDT-vectoring error code from the VMCS into the VMX
721 * transient structure.
722 *
723 * @returns VBox status code.
724 * @param pVmxTransient Pointer to the VMX transient structure.
725 */
726DECLINLINE(int) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
727{
728 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE))
729 {
730 int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
731 AssertRCReturn(rc, rc);
732 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE;
733 }
734 return VINF_SUCCESS;
735}
736
737
738/**
739 * Enters VMX root mode operation on the current CPU.
740 *
741 * @returns VBox status code.
742 * @param pVM Pointer to the VM (optional, can be NULL, after
743 * a resume).
744 * @param HCPhysCpuPage Physical address of the VMXON region.
745 * @param pvCpuPage Pointer to the VMXON region.
746 */
747static int hmR0VmxEnterRootMode(PVM pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage)
748{
749 Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
750 Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
751 Assert(pvCpuPage);
752 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
753
754 if (pVM)
755 {
756 /* Write the VMCS revision dword to the VMXON region. */
757 *(uint32_t *)pvCpuPage = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo);
758 }
759
760 /* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */
761 RTCCUINTREG uEflags = ASMIntDisableFlags();
762
763 /* Enable the VMX bit in CR4 if necessary. */
764 RTCCUINTREG uCr4 = ASMGetCR4();
765 if (!(uCr4 & X86_CR4_VMXE))
766 ASMSetCR4(uCr4 | X86_CR4_VMXE);
767
768 /* Enter VMX root mode. */
769 int rc = VMXEnable(HCPhysCpuPage);
770 if (RT_FAILURE(rc))
771 ASMSetCR4(uCr4);
772
773 /* Restore interrupts. */
774 ASMSetFlags(uEflags);
775 return rc;
776}
777
778
779/**
780 * Exits VMX root mode operation on the current CPU.
781 *
782 * @returns VBox status code.
783 */
784static int hmR0VmxLeaveRootMode(void)
785{
786 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
787
788 /* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */
789 RTCCUINTREG uEflags = ASMIntDisableFlags();
790
791 /* If we're for some reason not in VMX root mode, then don't leave it. */
792 RTCCUINTREG uHostCR4 = ASMGetCR4();
793
794 int rc;
795 if (uHostCR4 & X86_CR4_VMXE)
796 {
797 /* Exit VMX root mode and clear the VMX bit in CR4. */
798 VMXDisable();
799 ASMSetCR4(uHostCR4 & ~X86_CR4_VMXE);
800 rc = VINF_SUCCESS;
801 }
802 else
803 rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE;
804
805 /* Restore interrupts. */
806 ASMSetFlags(uEflags);
807 return rc;
808}
809
810
811/**
812 * Allocates and maps one physically contiguous page. The allocated page is
813 * zero'd out. (Used by various VT-x structures).
814 *
815 * @returns IPRT status code.
816 * @param pMemObj Pointer to the ring-0 memory object.
817 * @param ppVirt Where to store the virtual address of the
818 * allocation.
819 * @param pPhys Where to store the physical address of the
820 * allocation.
821 */
822DECLINLINE(int) hmR0VmxPageAllocZ(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
823{
824 AssertPtrReturn(pMemObj, VERR_INVALID_PARAMETER);
825 AssertPtrReturn(ppVirt, VERR_INVALID_PARAMETER);
826 AssertPtrReturn(pHCPhys, VERR_INVALID_PARAMETER);
827
828 int rc = RTR0MemObjAllocCont(pMemObj, PAGE_SIZE, false /* fExecutable */);
829 if (RT_FAILURE(rc))
830 return rc;
831 *ppVirt = RTR0MemObjAddress(*pMemObj);
832 *pHCPhys = RTR0MemObjGetPagePhysAddr(*pMemObj, 0 /* iPage */);
833 ASMMemZero32(*ppVirt, PAGE_SIZE);
834 return VINF_SUCCESS;
835}
836
837
838/**
839 * Frees and unmaps an allocated physical page.
840 *
841 * @param pMemObj Pointer to the ring-0 memory object.
842 * @param ppVirt Where to re-initialize the virtual address of
843 * allocation as 0.
844 * @param pHCPhys Where to re-initialize the physical address of the
845 * allocation as 0.
846 */
847DECLINLINE(void) hmR0VmxPageFree(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
848{
849 AssertPtr(pMemObj);
850 AssertPtr(ppVirt);
851 AssertPtr(pHCPhys);
852 if (*pMemObj != NIL_RTR0MEMOBJ)
853 {
854 int rc = RTR0MemObjFree(*pMemObj, true /* fFreeMappings */);
855 AssertRC(rc);
856 *pMemObj = NIL_RTR0MEMOBJ;
857 *ppVirt = 0;
858 *pHCPhys = 0;
859 }
860}
861
862
863/**
864 * Worker function to free VT-x related structures.
865 *
866 * @returns IPRT status code.
867 * @param pVM Pointer to the VM.
868 */
869static void hmR0VmxStructsFree(PVM pVM)
870{
871 for (VMCPUID i = 0; i < pVM->cCpus; i++)
872 {
873 PVMCPU pVCpu = &pVM->aCpus[i];
874 AssertPtr(pVCpu);
875
876 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjHostMsr, &pVCpu->hm.s.vmx.pvHostMsr, &pVCpu->hm.s.vmx.HCPhysHostMsr);
877 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjGuestMsr, &pVCpu->hm.s.vmx.pvGuestMsr, &pVCpu->hm.s.vmx.HCPhysGuestMsr);
878
879 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
880 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjMsrBitmap, &pVCpu->hm.s.vmx.pvMsrBitmap, &pVCpu->hm.s.vmx.HCPhysMsrBitmap);
881
882 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjVirtApic, (PRTR0PTR)&pVCpu->hm.s.vmx.pbVirtApic, &pVCpu->hm.s.vmx.HCPhysVirtApic);
883 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjVmcs, &pVCpu->hm.s.vmx.pvVmcs, &pVCpu->hm.s.vmx.HCPhysVmcs);
884 }
885
886 hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess, &pVM->hm.s.vmx.HCPhysApicAccess);
887#ifdef VBOX_WITH_CRASHDUMP_MAGIC
888 hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
889#endif
890}
891
892
893/**
894 * Worker function to allocate VT-x related VM structures.
895 *
896 * @returns IPRT status code.
897 * @param pVM Pointer to the VM.
898 */
899static int hmR0VmxStructsAlloc(PVM pVM)
900{
901 /*
902 * Initialize members up-front so we can cleanup properly on allocation failure.
903 */
904#define VMXLOCAL_INIT_VM_MEMOBJ(a_Name, a_VirtPrefix) \
905 pVM->hm.s.vmx.hMemObj##a_Name = NIL_RTR0MEMOBJ; \
906 pVM->hm.s.vmx.a_VirtPrefix##a_Name = 0; \
907 pVM->hm.s.vmx.HCPhys##a_Name = 0;
908
909#define VMXLOCAL_INIT_VMCPU_MEMOBJ(a_Name, a_VirtPrefix) \
910 pVCpu->hm.s.vmx.hMemObj##a_Name = NIL_RTR0MEMOBJ; \
911 pVCpu->hm.s.vmx.a_VirtPrefix##a_Name = 0; \
912 pVCpu->hm.s.vmx.HCPhys##a_Name = 0;
913
914#ifdef VBOX_WITH_CRASHDUMP_MAGIC
915 VMXLOCAL_INIT_VM_MEMOBJ(Scratch, pv);
916#endif
917 VMXLOCAL_INIT_VM_MEMOBJ(ApicAccess, pb);
918
919 AssertCompile(sizeof(VMCPUID) == sizeof(pVM->cCpus));
920 for (VMCPUID i = 0; i < pVM->cCpus; i++)
921 {
922 PVMCPU pVCpu = &pVM->aCpus[i];
923 VMXLOCAL_INIT_VMCPU_MEMOBJ(Vmcs, pv);
924 VMXLOCAL_INIT_VMCPU_MEMOBJ(VirtApic, pb);
925 VMXLOCAL_INIT_VMCPU_MEMOBJ(MsrBitmap, pv);
926 VMXLOCAL_INIT_VMCPU_MEMOBJ(GuestMsr, pv);
927 VMXLOCAL_INIT_VMCPU_MEMOBJ(HostMsr, pv);
928 }
929#undef VMXLOCAL_INIT_VMCPU_MEMOBJ
930#undef VMXLOCAL_INIT_VM_MEMOBJ
931
932 /* The VMCS size cannot be more than 4096 bytes. See Intel spec. Appendix A.1 "Basic VMX Information". */
933 AssertReturnStmt(MSR_IA32_VMX_BASIC_INFO_VMCS_SIZE(pVM->hm.s.vmx.Msrs.u64BasicInfo) <= PAGE_SIZE,
934 (&pVM->aCpus[0])->hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE,
935 VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO);
936
937 /*
938 * Allocate all the VT-x structures.
939 */
940 int rc = VINF_SUCCESS;
941#ifdef VBOX_WITH_CRASHDUMP_MAGIC
942 rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
943 if (RT_FAILURE(rc))
944 goto cleanup;
945 strcpy((char *)pVM->hm.s.vmx.pbScratch, "SCRATCH Magic");
946 *(uint64_t *)(pVM->hm.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef);
947#endif
948
949 /* Allocate the APIC-access page for trapping APIC accesses from the guest. */
950 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)
951 {
952 rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess,
953 &pVM->hm.s.vmx.HCPhysApicAccess);
954 if (RT_FAILURE(rc))
955 goto cleanup;
956 }
957
958 /*
959 * Initialize per-VCPU VT-x structures.
960 */
961 for (VMCPUID i = 0; i < pVM->cCpus; i++)
962 {
963 PVMCPU pVCpu = &pVM->aCpus[i];
964 AssertPtr(pVCpu);
965
966 /* Allocate the VM control structure (VMCS). */
967 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjVmcs, &pVCpu->hm.s.vmx.pvVmcs, &pVCpu->hm.s.vmx.HCPhysVmcs);
968 if (RT_FAILURE(rc))
969 goto cleanup;
970
971 /* Allocate the Virtual-APIC page for transparent TPR accesses. */
972 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
973 {
974 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjVirtApic, (PRTR0PTR)&pVCpu->hm.s.vmx.pbVirtApic,
975 &pVCpu->hm.s.vmx.HCPhysVirtApic);
976 if (RT_FAILURE(rc))
977 goto cleanup;
978 }
979
980 /*
981 * Allocate the MSR-bitmap if supported by the CPU. The MSR-bitmap is for
982 * transparent accesses of specific MSRs.
983 *
984 * If the condition for enabling MSR bitmaps changes here, don't forget to
985 * update HMIsMsrBitmapsAvailable().
986 */
987 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
988 {
989 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjMsrBitmap, &pVCpu->hm.s.vmx.pvMsrBitmap,
990 &pVCpu->hm.s.vmx.HCPhysMsrBitmap);
991 if (RT_FAILURE(rc))
992 goto cleanup;
993 ASMMemFill32(pVCpu->hm.s.vmx.pvMsrBitmap, PAGE_SIZE, UINT32_C(0xffffffff));
994 }
995
996 /* Allocate the VM-entry MSR-load and VM-exit MSR-store page for the guest MSRs. */
997 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjGuestMsr, &pVCpu->hm.s.vmx.pvGuestMsr, &pVCpu->hm.s.vmx.HCPhysGuestMsr);
998 if (RT_FAILURE(rc))
999 goto cleanup;
1000
1001 /* Allocate the VM-exit MSR-load page for the host MSRs. */
1002 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjHostMsr, &pVCpu->hm.s.vmx.pvHostMsr, &pVCpu->hm.s.vmx.HCPhysHostMsr);
1003 if (RT_FAILURE(rc))
1004 goto cleanup;
1005 }
1006
1007 return VINF_SUCCESS;
1008
1009cleanup:
1010 hmR0VmxStructsFree(pVM);
1011 return rc;
1012}
1013
1014
1015/**
1016 * Does global VT-x initialization (called during module initialization).
1017 *
1018 * @returns VBox status code.
1019 */
1020VMMR0DECL(int) VMXR0GlobalInit(void)
1021{
1022#ifdef HMVMX_USE_FUNCTION_TABLE
1023 AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_apfnVMExitHandlers));
1024# ifdef VBOX_STRICT
1025 for (unsigned i = 0; i < RT_ELEMENTS(g_apfnVMExitHandlers); i++)
1026 Assert(g_apfnVMExitHandlers[i]);
1027# endif
1028#endif
1029 return VINF_SUCCESS;
1030}
1031
1032
1033/**
1034 * Does global VT-x termination (called during module termination).
1035 */
1036VMMR0DECL(void) VMXR0GlobalTerm()
1037{
1038 /* Nothing to do currently. */
1039}
1040
1041
1042/**
1043 * Sets up and activates VT-x on the current CPU.
1044 *
1045 * @returns VBox status code.
1046 * @param pCpu Pointer to the global CPU info struct.
1047 * @param pVM Pointer to the VM (can be NULL after a host resume
1048 * operation).
1049 * @param pvCpuPage Pointer to the VMXON region (can be NULL if @a
1050 * fEnabledByHost is true).
1051 * @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if
1052 * @a fEnabledByHost is true).
1053 * @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to
1054 * enable VT-x on the host.
1055 * @param pvMsrs Opaque pointer to VMXMSRS struct.
1056 */
1057VMMR0DECL(int) VMXR0EnableCpu(PHMGLOBALCPUINFO pCpu, PVM pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
1058 void *pvMsrs)
1059{
1060 Assert(pCpu);
1061 Assert(pvMsrs);
1062 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1063
1064 /* Enable VT-x if it's not already enabled by the host. */
1065 if (!fEnabledByHost)
1066 {
1067 int rc = hmR0VmxEnterRootMode(pVM, HCPhysCpuPage, pvCpuPage);
1068 if (RT_FAILURE(rc))
1069 return rc;
1070 }
1071
1072 /*
1073 * Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been using EPTPs) so
1074 * we don't retain any stale guest-physical mappings which won't get invalidated when flushing by VPID.
1075 */
1076 PVMXMSRS pMsrs = (PVMXMSRS)pvMsrs;
1077 if (pMsrs->u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
1078 {
1079 hmR0VmxFlushEpt(NULL /* pVCpu */, VMXFLUSHEPT_ALL_CONTEXTS);
1080 pCpu->fFlushAsidBeforeUse = false;
1081 }
1082 else
1083 pCpu->fFlushAsidBeforeUse = true;
1084
1085 /* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */
1086 ++pCpu->cTlbFlushes;
1087
1088 return VINF_SUCCESS;
1089}
1090
1091
1092/**
1093 * Deactivates VT-x on the current CPU.
1094 *
1095 * @returns VBox status code.
1096 * @param pCpu Pointer to the global CPU info struct.
1097 * @param pvCpuPage Pointer to the VMXON region.
1098 * @param HCPhysCpuPage Physical address of the VMXON region.
1099 *
1100 * @remarks This function should never be called when SUPR0EnableVTx() or
1101 * similar was used to enable VT-x on the host.
1102 */
1103VMMR0DECL(int) VMXR0DisableCpu(PHMGLOBALCPUINFO pCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
1104{
1105 NOREF(pCpu);
1106 NOREF(pvCpuPage);
1107 NOREF(HCPhysCpuPage);
1108
1109 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1110 return hmR0VmxLeaveRootMode();
1111}
1112
1113
1114/**
1115 * Sets the permission bits for the specified MSR in the MSR bitmap.
1116 *
1117 * @param pVCpu Pointer to the VMCPU.
1118 * @param uMSR The MSR value.
1119 * @param enmRead Whether reading this MSR causes a VM-exit.
1120 * @param enmWrite Whether writing this MSR causes a VM-exit.
1121 */
1122static void hmR0VmxSetMsrPermission(PVMCPU pVCpu, uint32_t uMsr, VMXMSREXITREAD enmRead, VMXMSREXITWRITE enmWrite)
1123{
1124 int32_t iBit;
1125 uint8_t *pbMsrBitmap = (uint8_t *)pVCpu->hm.s.vmx.pvMsrBitmap;
1126
1127 /*
1128 * Layout:
1129 * 0x000 - 0x3ff - Low MSR read bits
1130 * 0x400 - 0x7ff - High MSR read bits
1131 * 0x800 - 0xbff - Low MSR write bits
1132 * 0xc00 - 0xfff - High MSR write bits
1133 */
1134 if (uMsr <= 0x00001FFF)
1135 iBit = uMsr;
1136 else if ( uMsr >= 0xC0000000
1137 && uMsr <= 0xC0001FFF)
1138 {
1139 iBit = (uMsr - 0xC0000000);
1140 pbMsrBitmap += 0x400;
1141 }
1142 else
1143 {
1144 AssertMsgFailed(("hmR0VmxSetMsrPermission: Invalid MSR %#RX32\n", uMsr));
1145 return;
1146 }
1147
1148 Assert(iBit <= 0x1fff);
1149 if (enmRead == VMXMSREXIT_INTERCEPT_READ)
1150 ASMBitSet(pbMsrBitmap, iBit);
1151 else
1152 ASMBitClear(pbMsrBitmap, iBit);
1153
1154 if (enmWrite == VMXMSREXIT_INTERCEPT_WRITE)
1155 ASMBitSet(pbMsrBitmap + 0x800, iBit);
1156 else
1157 ASMBitClear(pbMsrBitmap + 0x800, iBit);
1158}
1159
1160
1161#ifdef VBOX_STRICT
1162/**
1163 * Gets the permission bits for the specified MSR in the MSR bitmap.
1164 *
1165 * @returns VBox status code.
1166 * @retval VINF_SUCCESS if the specified MSR is found.
1167 * @retval VERR_NOT_FOUND if the specified MSR is not found.
1168 * @retval VERR_NOT_SUPPORTED if VT-x doesn't allow the MSR.
1169 *
1170 * @param pVCpu Pointer to the VMCPU.
1171 * @param uMsr The MSR.
1172 * @param penmRead Where to store the read permissions.
1173 * @param penmWrite Where to store the write permissions.
1174 */
1175static int hmR0VmxGetMsrPermission(PVMCPU pVCpu, uint32_t uMsr, PVMXMSREXITREAD penmRead, PVMXMSREXITWRITE penmWrite)
1176{
1177 AssertPtrReturn(penmRead, VERR_INVALID_PARAMETER);
1178 AssertPtrReturn(penmWrite, VERR_INVALID_PARAMETER);
1179 int32_t iBit;
1180 uint8_t *pbMsrBitmap = (uint8_t *)pVCpu->hm.s.vmx.pvMsrBitmap;
1181
1182 /* See hmR0VmxSetMsrPermission() for the layout. */
1183 if (uMsr <= 0x00001FFF)
1184 iBit = uMsr;
1185 else if ( uMsr >= 0xC0000000
1186 && uMsr <= 0xC0001FFF)
1187 {
1188 iBit = (uMsr - 0xC0000000);
1189 pbMsrBitmap += 0x400;
1190 }
1191 else
1192 {
1193 AssertMsgFailed(("hmR0VmxGetMsrPermission: Invalid MSR %#RX32\n", uMsr));
1194 return VERR_NOT_SUPPORTED;
1195 }
1196
1197 Assert(iBit <= 0x1fff);
1198 if (ASMBitTest(pbMsrBitmap, iBit))
1199 *penmRead = VMXMSREXIT_INTERCEPT_READ;
1200 else
1201 *penmRead = VMXMSREXIT_PASSTHRU_READ;
1202
1203 if (ASMBitTest(pbMsrBitmap + 0x800, iBit))
1204 *penmWrite = VMXMSREXIT_INTERCEPT_WRITE;
1205 else
1206 *penmWrite = VMXMSREXIT_PASSTHRU_WRITE;
1207 return VINF_SUCCESS;
1208}
1209#endif /* VBOX_STRICT */
1210
1211
1212/**
1213 * Updates the VMCS with the number of effective MSRs in the auto-load/store MSR
1214 * area.
1215 *
1216 * @returns VBox status code.
1217 * @param pVCpu Pointer to the VMCPU.
1218 * @param cMsrs The number of MSRs.
1219 */
1220DECLINLINE(int) hmR0VmxSetAutoLoadStoreMsrCount(PVMCPU pVCpu, uint32_t cMsrs)
1221{
1222 /* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */
1223 uint32_t const cMaxSupportedMsrs = MSR_IA32_VMX_MISC_MAX_MSR(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc);
1224 if (RT_UNLIKELY(cMsrs > cMaxSupportedMsrs))
1225 {
1226 LogRel(("CPU auto-load/store MSR count in VMCS exceeded cMsrs=%u Supported=%u.\n", cMsrs, cMaxSupportedMsrs));
1227 pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE;
1228 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
1229 }
1230
1231 /* Update number of guest MSRs to load/store across the world-switch. */
1232 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs); AssertRCReturn(rc, rc);
1233 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs); AssertRCReturn(rc, rc);
1234
1235 /* Update number of host MSRs to load after the world-switch. Identical to guest-MSR count as it's always paired. */
1236 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs); AssertRCReturn(rc, rc);
1237
1238 /* Update the VCPU's copy of the MSR count. */
1239 pVCpu->hm.s.vmx.cMsrs = cMsrs;
1240
1241 return VINF_SUCCESS;
1242}
1243
1244
1245/**
1246 * Adds a new (or updates the value of an existing) guest/host MSR
1247 * pair to be swapped during the world-switch as part of the
1248 * auto-load/store MSR area in the VMCS.
1249 *
1250 * @returns true if the MSR was added -and- its value was updated, false
1251 * otherwise.
1252 * @param pVCpu Pointer to the VMCPU.
1253 * @param uMsr The MSR.
1254 * @param uGuestMsr Value of the guest MSR.
1255 * @param fUpdateHostMsr Whether to update the value of the host MSR if
1256 * necessary.
1257 */
1258static bool hmR0VmxAddAutoLoadStoreMsr(PVMCPU pVCpu, uint32_t uMsr, uint64_t uGuestMsrValue, bool fUpdateHostMsr)
1259{
1260 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1261 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1262 uint32_t i;
1263 for (i = 0; i < cMsrs; i++)
1264 {
1265 if (pGuestMsr->u32Msr == uMsr)
1266 break;
1267 pGuestMsr++;
1268 }
1269
1270 bool fAdded = false;
1271 if (i == cMsrs)
1272 {
1273 ++cMsrs;
1274 int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, cMsrs);
1275 AssertRC(rc);
1276
1277 /* Now that we're swapping MSRs during the world-switch, allow the guest to read/write them without causing VM-exits. */
1278 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1279 hmR0VmxSetMsrPermission(pVCpu, uMsr, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
1280
1281 fAdded = true;
1282 }
1283
1284 /* Update the MSR values in the auto-load/store MSR area. */
1285 pGuestMsr->u32Msr = uMsr;
1286 pGuestMsr->u64Value = uGuestMsrValue;
1287
1288 /* Create/update the MSR slot in the host MSR area. */
1289 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1290 pHostMsr += i;
1291 pHostMsr->u32Msr = uMsr;
1292
1293 /*
1294 * Update the host MSR only when requested by the caller AND when we're
1295 * adding it to the auto-load/store area. Otherwise, it would have been
1296 * updated by hmR0VmxSaveHostMsrs(). We do this for performance reasons.
1297 */
1298 bool fUpdatedMsrValue = false;
1299 if ( fAdded
1300 && fUpdateHostMsr)
1301 {
1302 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1303 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1304 pHostMsr->u64Value = ASMRdMsr(pHostMsr->u32Msr);
1305 fUpdatedMsrValue = true;
1306 }
1307
1308 return fUpdatedMsrValue;
1309}
1310
1311
1312/**
1313 * Removes a guest/host MSR pair to be swapped during the world-switch from the
1314 * auto-load/store MSR area in the VMCS.
1315 *
1316 * @returns VBox status code.
1317 * @param pVCpu Pointer to the VMCPU.
1318 * @param uMsr The MSR.
1319 */
1320static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPU pVCpu, uint32_t uMsr)
1321{
1322 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1323 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1324 for (uint32_t i = 0; i < cMsrs; i++)
1325 {
1326 /* Find the MSR. */
1327 if (pGuestMsr->u32Msr == uMsr)
1328 {
1329 /* If it's the last MSR, simply reduce the count. */
1330 if (i == cMsrs - 1)
1331 {
1332 --cMsrs;
1333 break;
1334 }
1335
1336 /* Remove it by swapping the last MSR in place of it, and reducing the count. */
1337 PVMXAUTOMSR pLastGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1338 pLastGuestMsr += cMsrs - 1;
1339 pGuestMsr->u32Msr = pLastGuestMsr->u32Msr;
1340 pGuestMsr->u64Value = pLastGuestMsr->u64Value;
1341
1342 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1343 PVMXAUTOMSR pLastHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1344 pLastHostMsr += cMsrs - 1;
1345 pHostMsr->u32Msr = pLastHostMsr->u32Msr;
1346 pHostMsr->u64Value = pLastHostMsr->u64Value;
1347 --cMsrs;
1348 break;
1349 }
1350 pGuestMsr++;
1351 }
1352
1353 /* Update the VMCS if the count changed (meaning the MSR was found). */
1354 if (cMsrs != pVCpu->hm.s.vmx.cMsrs)
1355 {
1356 int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, cMsrs);
1357 AssertRCReturn(rc, rc);
1358
1359 /* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */
1360 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1361 hmR0VmxSetMsrPermission(pVCpu, uMsr, VMXMSREXIT_INTERCEPT_READ, VMXMSREXIT_INTERCEPT_WRITE);
1362
1363 Log4(("Removed MSR %#RX32 new cMsrs=%u\n", uMsr, pVCpu->hm.s.vmx.cMsrs));
1364 return VINF_SUCCESS;
1365 }
1366
1367 return VERR_NOT_FOUND;
1368}
1369
1370
1371/**
1372 * Checks if the specified guest MSR is part of the auto-load/store area in
1373 * the VMCS.
1374 *
1375 * @returns true if found, false otherwise.
1376 * @param pVCpu Pointer to the VMCPU.
1377 * @param uMsr The MSR to find.
1378 */
1379static bool hmR0VmxIsAutoLoadStoreGuestMsr(PVMCPU pVCpu, uint32_t uMsr)
1380{
1381 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1382 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1383
1384 for (uint32_t i = 0; i < cMsrs; i++, pGuestMsr++)
1385 {
1386 if (pGuestMsr->u32Msr == uMsr)
1387 return true;
1388 }
1389 return false;
1390}
1391
1392
1393/**
1394 * Updates the value of all host MSRs in the auto-load/store area in the VMCS.
1395 *
1396 * @param pVCpu Pointer to the VMCPU.
1397 *
1398 * @remarks No-long-jump zone!!!
1399 */
1400static void hmR0VmxUpdateAutoLoadStoreHostMsrs(PVMCPU pVCpu)
1401{
1402 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1403 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1404 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1405 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1406
1407 for (uint32_t i = 0; i < cMsrs; i++, pHostMsr++, pGuestMsr++)
1408 {
1409 AssertReturnVoid(pHostMsr->u32Msr == pGuestMsr->u32Msr);
1410
1411 /*
1412 * Performance hack for the host EFER MSR. We use the cached value rather than re-read it.
1413 * Strict builds will catch mismatches in hmR0VmxCheckAutoLoadStoreMsrs(). See @bugref{7368}.
1414 */
1415 if (pHostMsr->u32Msr == MSR_K6_EFER)
1416 pHostMsr->u64Value = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostEfer;
1417 else
1418 pHostMsr->u64Value = ASMRdMsr(pHostMsr->u32Msr);
1419 }
1420
1421 pVCpu->hm.s.vmx.fUpdatedHostMsrs = true;
1422}
1423
1424
1425#if HC_ARCH_BITS == 64
1426/**
1427 * Saves a set of host MSRs to allow read/write passthru access to the guest and
1428 * perform lazy restoration of the host MSRs while leaving VT-x.
1429 *
1430 * @param pVCpu Pointer to the VMCPU.
1431 *
1432 * @remarks No-long-jump zone!!!
1433 */
1434static void hmR0VmxLazySaveHostMsrs(PVMCPU pVCpu)
1435{
1436 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1437
1438 /*
1439 * Note: If you're adding MSRs here, make sure to update the MSR-bitmap permissions in hmR0VmxSetupProcCtls().
1440 */
1441 if (!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
1442 {
1443 pVCpu->hm.s.vmx.u64HostLStarMsr = ASMRdMsr(MSR_K8_LSTAR);
1444 pVCpu->hm.s.vmx.u64HostStarMsr = ASMRdMsr(MSR_K6_STAR);
1445 pVCpu->hm.s.vmx.u64HostSFMaskMsr = ASMRdMsr(MSR_K8_SF_MASK);
1446 pVCpu->hm.s.vmx.u64HostKernelGSBaseMsr = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
1447 pVCpu->hm.s.vmx.fLazyMsrs |= VMX_LAZY_MSRS_SAVED_HOST;
1448 }
1449}
1450
1451
1452/**
1453 * Checks whether the MSR belongs to the set of guest MSRs that we restore
1454 * lazily while leaving VT-x.
1455 *
1456 * @returns true if it does, false otherwise.
1457 * @param pVCpu Pointer to the VMCPU.
1458 * @param uMsr The MSR to check.
1459 */
1460static bool hmR0VmxIsLazyGuestMsr(PVMCPU pVCpu, uint32_t uMsr)
1461{
1462 NOREF(pVCpu);
1463 switch (uMsr)
1464 {
1465 case MSR_K8_LSTAR:
1466 case MSR_K6_STAR:
1467 case MSR_K8_SF_MASK:
1468 case MSR_K8_KERNEL_GS_BASE:
1469 return true;
1470 }
1471 return false;
1472}
1473
1474
1475/**
1476 * Saves a set of guest MSRs back into the guest-CPU context.
1477 *
1478 * @param pVCpu Pointer to the VMCPU.
1479 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
1480 * out-of-sync. Make sure to update the required fields
1481 * before using them.
1482 *
1483 * @remarks No-long-jump zone!!!
1484 */
1485static void hmR0VmxLazySaveGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
1486{
1487 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1488 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1489
1490 if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
1491 {
1492 Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1493 pMixedCtx->msrLSTAR = ASMRdMsr(MSR_K8_LSTAR);
1494 pMixedCtx->msrSTAR = ASMRdMsr(MSR_K6_STAR);
1495 pMixedCtx->msrSFMASK = ASMRdMsr(MSR_K8_SF_MASK);
1496 pMixedCtx->msrKERNELGSBASE = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
1497 }
1498}
1499
1500
1501/**
1502 * Loads a set of guests MSRs to allow read/passthru to the guest.
1503 *
1504 * The name of this function is slightly confusing. This function does NOT
1505 * postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a
1506 * common prefix for functions dealing with "lazy restoration" of the shared
1507 * MSRs.
1508 *
1509 * @param pVCpu Pointer to the VMCPU.
1510 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
1511 * out-of-sync. Make sure to update the required fields
1512 * before using them.
1513 *
1514 * @remarks No-long-jump zone!!!
1515 */
1516static void hmR0VmxLazyLoadGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
1517{
1518 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1519 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1520
1521#define VMXLOCAL_LAZY_LOAD_GUEST_MSR(uMsr, a_GuestMsr, a_HostMsr) \
1522 do { \
1523 if (pMixedCtx->msr##a_GuestMsr != pVCpu->hm.s.vmx.u64Host##a_HostMsr##Msr) \
1524 ASMWrMsr(uMsr, pMixedCtx->msr##a_GuestMsr); \
1525 else \
1526 Assert(ASMRdMsr(uMsr) == pVCpu->hm.s.vmx.u64Host##a_HostMsr##Msr); \
1527 } while (0)
1528
1529 Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1530 if (!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
1531 {
1532 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K8_LSTAR, LSTAR, LStar);
1533 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K6_STAR, STAR, Star);
1534 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K8_SF_MASK, SFMASK, SFMask);
1535 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K8_KERNEL_GS_BASE, KERNELGSBASE, KernelGSBase);
1536 pVCpu->hm.s.vmx.fLazyMsrs |= VMX_LAZY_MSRS_LOADED_GUEST;
1537 }
1538 else
1539 {
1540 ASMWrMsr(MSR_K8_LSTAR, pMixedCtx->msrLSTAR);
1541 ASMWrMsr(MSR_K6_STAR, pMixedCtx->msrSTAR);
1542 ASMWrMsr(MSR_K8_SF_MASK, pMixedCtx->msrSFMASK);
1543 ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pMixedCtx->msrKERNELGSBASE);
1544 }
1545
1546#undef VMXLOCAL_LAZY_LOAD_GUEST_MSR
1547}
1548
1549
1550/**
1551 * Performs lazy restoration of the set of host MSRs if they were previously
1552 * loaded with guest MSR values.
1553 *
1554 * @param pVCpu Pointer to the VMCPU.
1555 *
1556 * @remarks No-long-jump zone!!!
1557 * @remarks The guest MSRs should have been saved back into the guest-CPU
1558 * context by hmR0VmxSaveGuestLazyMsrs()!!!
1559 */
1560static void hmR0VmxLazyRestoreHostMsrs(PVMCPU pVCpu)
1561{
1562 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1563 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1564
1565 if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
1566 {
1567 Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1568 ASMWrMsr(MSR_K8_LSTAR, pVCpu->hm.s.vmx.u64HostLStarMsr);
1569 ASMWrMsr(MSR_K6_STAR, pVCpu->hm.s.vmx.u64HostStarMsr);
1570 ASMWrMsr(MSR_K8_SF_MASK, pVCpu->hm.s.vmx.u64HostSFMaskMsr);
1571 ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pVCpu->hm.s.vmx.u64HostKernelGSBaseMsr);
1572 }
1573 pVCpu->hm.s.vmx.fLazyMsrs &= ~(VMX_LAZY_MSRS_LOADED_GUEST | VMX_LAZY_MSRS_SAVED_HOST);
1574}
1575#endif /* HC_ARCH_BITS == 64 */
1576
1577
1578/**
1579 * Verifies that our cached values of the VMCS controls are all
1580 * consistent with what's actually present in the VMCS.
1581 *
1582 * @returns VBox status code.
1583 * @param pVCpu Pointer to the VMCPU.
1584 */
1585static int hmR0VmxCheckVmcsCtls(PVMCPU pVCpu)
1586{
1587 uint32_t u32Val;
1588 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
1589 AssertRCReturn(rc, rc);
1590 AssertMsgReturn(pVCpu->hm.s.vmx.u32EntryCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32EntryCtls, u32Val),
1591 VERR_VMX_ENTRY_CTLS_CACHE_INVALID);
1592
1593 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val);
1594 AssertRCReturn(rc, rc);
1595 AssertMsgReturn(pVCpu->hm.s.vmx.u32ExitCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ExitCtls, u32Val),
1596 VERR_VMX_EXIT_CTLS_CACHE_INVALID);
1597
1598 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val);
1599 AssertRCReturn(rc, rc);
1600 AssertMsgReturn(pVCpu->hm.s.vmx.u32PinCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32PinCtls, u32Val),
1601 VERR_VMX_PIN_EXEC_CTLS_CACHE_INVALID);
1602
1603 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val);
1604 AssertRCReturn(rc, rc);
1605 AssertMsgReturn(pVCpu->hm.s.vmx.u32ProcCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ProcCtls, u32Val),
1606 VERR_VMX_PROC_EXEC_CTLS_CACHE_INVALID);
1607
1608 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
1609 {
1610 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val);
1611 AssertRCReturn(rc, rc);
1612 AssertMsgReturn(pVCpu->hm.s.vmx.u32ProcCtls2 == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ProcCtls2, u32Val),
1613 VERR_VMX_PROC_EXEC2_CTLS_CACHE_INVALID);
1614 }
1615
1616 return VINF_SUCCESS;
1617}
1618
1619
1620#ifdef VBOX_STRICT
1621/**
1622 * Verifies that our cached host EFER value has not changed
1623 * since we cached it.
1624 *
1625 * @param pVCpu Pointer to the VMCPU.
1626 */
1627static void hmR0VmxCheckHostEferMsr(PVMCPU pVCpu)
1628{
1629 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1630
1631 if (pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR)
1632 {
1633 uint64_t u64Val;
1634 int rc = VMXReadVmcs64(VMX_VMCS64_HOST_FIELD_EFER_FULL, &u64Val);
1635 AssertRC(rc);
1636
1637 uint64_t u64HostEferMsr = ASMRdMsr(MSR_K6_EFER);
1638 AssertMsgReturnVoid(u64HostEferMsr == u64Val, ("u64HostEferMsr=%#RX64 u64Val=%#RX64\n", u64HostEferMsr, u64Val));
1639 }
1640}
1641
1642
1643/**
1644 * Verifies whether the guest/host MSR pairs in the auto-load/store area in the
1645 * VMCS are correct.
1646 *
1647 * @param pVCpu Pointer to the VMCPU.
1648 */
1649static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPU pVCpu)
1650{
1651 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1652
1653 /* Verify MSR counts in the VMCS are what we think it should be. */
1654 uint32_t cMsrs;
1655 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cMsrs); AssertRC(rc);
1656 Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1657
1658 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cMsrs); AssertRC(rc);
1659 Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1660
1661 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cMsrs); AssertRC(rc);
1662 Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1663
1664 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1665 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1666 for (uint32_t i = 0; i < cMsrs; i++, pHostMsr++, pGuestMsr++)
1667 {
1668 /* Verify that the MSRs are paired properly and that the host MSR has the correct value. */
1669 AssertMsgReturnVoid(pHostMsr->u32Msr == pGuestMsr->u32Msr, ("HostMsr=%#RX32 GuestMsr=%#RX32 cMsrs=%u\n", pHostMsr->u32Msr,
1670 pGuestMsr->u32Msr, cMsrs));
1671
1672 uint64_t u64Msr = ASMRdMsr(pHostMsr->u32Msr);
1673 AssertMsgReturnVoid(pHostMsr->u64Value == u64Msr, ("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
1674 pHostMsr->u32Msr, pHostMsr->u64Value, u64Msr, cMsrs));
1675
1676 /* Verify that the permissions are as expected in the MSR bitmap. */
1677 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1678 {
1679 VMXMSREXITREAD enmRead;
1680 VMXMSREXITWRITE enmWrite;
1681 rc = hmR0VmxGetMsrPermission(pVCpu, pGuestMsr->u32Msr, &enmRead, &enmWrite);
1682 AssertMsgReturnVoid(rc == VINF_SUCCESS, ("hmR0VmxGetMsrPermission! failed. rc=%Rrc\n", rc));
1683 if (pGuestMsr->u32Msr == MSR_K6_EFER)
1684 {
1685 AssertMsgReturnVoid(enmRead == VMXMSREXIT_INTERCEPT_READ, ("Passthru read for EFER!?\n"));
1686 AssertMsgReturnVoid(enmWrite == VMXMSREXIT_INTERCEPT_WRITE, ("Passthru write for EFER!?\n"));
1687 }
1688 else
1689 {
1690 AssertMsgReturnVoid(enmRead == VMXMSREXIT_PASSTHRU_READ, ("u32Msr=%#RX32 cMsrs=%u No passthru read!\n",
1691 pGuestMsr->u32Msr, cMsrs));
1692 AssertMsgReturnVoid(enmWrite == VMXMSREXIT_PASSTHRU_WRITE, ("u32Msr=%#RX32 cMsrs=%u No passthru write!\n",
1693 pGuestMsr->u32Msr, cMsrs));
1694 }
1695 }
1696 }
1697}
1698#endif /* VBOX_STRICT */
1699
1700
1701/**
1702 * Flushes the TLB using EPT.
1703 *
1704 * @returns VBox status code.
1705 * @param pVCpu Pointer to the VMCPU (can be NULL depending on @a
1706 * enmFlush).
1707 * @param enmFlush Type of flush.
1708 *
1709 * @remarks Caller is responsible for making sure this function is called only
1710 * when NestedPaging is supported and providing @a enmFlush that is
1711 * supported by the CPU.
1712 * @remarks Can be called with interrupts disabled.
1713 */
1714static void hmR0VmxFlushEpt(PVMCPU pVCpu, VMXFLUSHEPT enmFlush)
1715{
1716 uint64_t au64Descriptor[2];
1717 if (enmFlush == VMXFLUSHEPT_ALL_CONTEXTS)
1718 au64Descriptor[0] = 0;
1719 else
1720 {
1721 Assert(pVCpu);
1722 au64Descriptor[0] = pVCpu->hm.s.vmx.HCPhysEPTP;
1723 }
1724 au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */
1725
1726 int rc = VMXR0InvEPT(enmFlush, &au64Descriptor[0]);
1727 AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %RGv failed with %Rrc\n", enmFlush, pVCpu ? pVCpu->hm.s.vmx.HCPhysEPTP : 0,
1728 rc));
1729 if ( RT_SUCCESS(rc)
1730 && pVCpu)
1731 {
1732 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging);
1733 }
1734}
1735
1736
1737/**
1738 * Flushes the TLB using VPID.
1739 *
1740 * @returns VBox status code.
1741 * @param pVM Pointer to the VM.
1742 * @param pVCpu Pointer to the VMCPU (can be NULL depending on @a
1743 * enmFlush).
1744 * @param enmFlush Type of flush.
1745 * @param GCPtr Virtual address of the page to flush (can be 0 depending
1746 * on @a enmFlush).
1747 *
1748 * @remarks Can be called with interrupts disabled.
1749 */
1750static void hmR0VmxFlushVpid(PVM pVM, PVMCPU pVCpu, VMXFLUSHVPID enmFlush, RTGCPTR GCPtr)
1751{
1752 NOREF(pVM);
1753 AssertPtr(pVM);
1754 Assert(pVM->hm.s.vmx.fVpid);
1755
1756 uint64_t au64Descriptor[2];
1757 if (enmFlush == VMXFLUSHVPID_ALL_CONTEXTS)
1758 {
1759 au64Descriptor[0] = 0;
1760 au64Descriptor[1] = 0;
1761 }
1762 else
1763 {
1764 AssertPtr(pVCpu);
1765 AssertMsg(pVCpu->hm.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
1766 AssertMsg(pVCpu->hm.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
1767 au64Descriptor[0] = pVCpu->hm.s.uCurrentAsid;
1768 au64Descriptor[1] = GCPtr;
1769 }
1770
1771 int rc = VMXR0InvVPID(enmFlush, &au64Descriptor[0]); NOREF(rc);
1772 AssertMsg(rc == VINF_SUCCESS,
1773 ("VMXR0InvVPID %#x %u %RGv failed with %d\n", enmFlush, pVCpu ? pVCpu->hm.s.uCurrentAsid : 0, GCPtr, rc));
1774 if ( RT_SUCCESS(rc)
1775 && pVCpu)
1776 {
1777 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
1778 }
1779}
1780
1781
1782/**
1783 * Invalidates a guest page by guest virtual address. Only relevant for
1784 * EPT/VPID, otherwise there is nothing really to invalidate.
1785 *
1786 * @returns VBox status code.
1787 * @param pVM Pointer to the VM.
1788 * @param pVCpu Pointer to the VMCPU.
1789 * @param GCVirt Guest virtual address of the page to invalidate.
1790 */
1791VMMR0DECL(int) VMXR0InvalidatePage(PVM pVM, PVMCPU pVCpu, RTGCPTR GCVirt)
1792{
1793 AssertPtr(pVM);
1794 AssertPtr(pVCpu);
1795 LogFlowFunc(("pVM=%p pVCpu=%p GCVirt=%RGv\n", pVM, pVCpu, GCVirt));
1796
1797 bool fFlushPending = VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_FLUSH);
1798 if (!fFlushPending)
1799 {
1800 /*
1801 * We must invalidate the guest TLB entry in either case, we cannot ignore it even for the EPT case
1802 * See @bugref{6043} and @bugref{6177}.
1803 *
1804 * Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*() as this
1805 * function maybe called in a loop with individual addresses.
1806 */
1807 if (pVM->hm.s.vmx.fVpid)
1808 {
1809 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
1810 {
1811 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_INDIV_ADDR, GCVirt);
1812 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
1813 }
1814 else
1815 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1816 }
1817 else if (pVM->hm.s.fNestedPaging)
1818 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1819 }
1820
1821 return VINF_SUCCESS;
1822}
1823
1824
1825/**
1826 * Invalidates a guest page by physical address. Only relevant for EPT/VPID,
1827 * otherwise there is nothing really to invalidate.
1828 *
1829 * @returns VBox status code.
1830 * @param pVM Pointer to the VM.
1831 * @param pVCpu Pointer to the VMCPU.
1832 * @param GCPhys Guest physical address of the page to invalidate.
1833 */
1834VMMR0DECL(int) VMXR0InvalidatePhysPage(PVM pVM, PVMCPU pVCpu, RTGCPHYS GCPhys)
1835{
1836 NOREF(pVM); NOREF(GCPhys);
1837 LogFlowFunc(("%RGp\n", GCPhys));
1838
1839 /*
1840 * We cannot flush a page by guest-physical address. invvpid takes only a linear address while invept only flushes
1841 * by EPT not individual addresses. We update the force flag here and flush before the next VM-entry in hmR0VmxFlushTLB*().
1842 * This function might be called in a loop. This should cause a flush-by-EPT if EPT is in use. See @bugref{6568}.
1843 */
1844 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1845 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgPhys);
1846 return VINF_SUCCESS;
1847}
1848
1849
1850/**
1851 * Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the
1852 * case where neither EPT nor VPID is supported by the CPU.
1853 *
1854 * @param pVM Pointer to the VM.
1855 * @param pVCpu Pointer to the VMCPU.
1856 * @param pCpu Pointer to the global HM struct.
1857 *
1858 * @remarks Called with interrupts disabled.
1859 */
1860static void hmR0VmxFlushTaggedTlbNone(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
1861{
1862 AssertPtr(pVCpu);
1863 AssertPtr(pCpu);
1864 NOREF(pVM);
1865
1866 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
1867
1868 /** @todo TLB shootdown is currently not used. See hmQueueInvlPage(). */
1869#if 0
1870 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN);
1871 pVCpu->hm.s.TlbShootdown.cPages = 0;
1872#endif
1873
1874 Assert(pCpu->idCpu != NIL_RTCPUID);
1875 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
1876 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
1877 pVCpu->hm.s.fForceTLBFlush = false;
1878 return;
1879}
1880
1881
1882/**
1883 * Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary.
1884 *
1885 * @param pVM Pointer to the VM.
1886 * @param pVCpu Pointer to the VMCPU.
1887 * @param pCpu Pointer to the global HM CPU struct.
1888 * @remarks All references to "ASID" in this function pertains to "VPID" in
1889 * Intel's nomenclature. The reason is, to avoid confusion in compare
1890 * statements since the host-CPU copies are named "ASID".
1891 *
1892 * @remarks Called with interrupts disabled.
1893 */
1894static void hmR0VmxFlushTaggedTlbBoth(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
1895{
1896#ifdef VBOX_WITH_STATISTICS
1897 bool fTlbFlushed = false;
1898# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0)
1899# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \
1900 if (!fTlbFlushed) \
1901 STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \
1902 } while (0)
1903#else
1904# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0)
1905# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0)
1906#endif
1907
1908 AssertPtr(pVM);
1909 AssertPtr(pCpu);
1910 AssertPtr(pVCpu);
1911 Assert(pCpu->idCpu != NIL_RTCPUID);
1912
1913 AssertMsg(pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid,
1914 ("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled."
1915 "fNestedPaging=%RTbool fVpid=%RTbool", pVM->hm.s.fNestedPaging, pVM->hm.s.vmx.fVpid));
1916
1917 /*
1918 * Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
1919 * If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB
1920 * or the host CPU is online after a suspend/resume, so we cannot reuse the current ASID anymore.
1921 */
1922 if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
1923 || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
1924 {
1925 ++pCpu->uCurrentAsid;
1926 if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
1927 {
1928 pCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */
1929 pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
1930 pCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
1931 }
1932
1933 pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid;
1934 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
1935 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
1936
1937 /*
1938 * Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also
1939 * invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}.
1940 */
1941 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
1942 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
1943 HMVMX_SET_TAGGED_TLB_FLUSHED();
1944 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH); /* Already flushed-by-EPT, skip doing it again below. */
1945 }
1946
1947 /* Check for explicit TLB shootdowns. */
1948 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
1949 {
1950 /*
1951 * Changes to the EPT paging structure by VMM requires flushing by EPT as the CPU creates
1952 * guest-physical (only EPT-tagged) mappings while traversing the EPT tables when EPT is in use.
1953 * Flushing by VPID will only flush linear (only VPID-tagged) and combined (EPT+VPID tagged) mappings
1954 * but not guest-physical mappings.
1955 * See Intel spec. 28.3.2 "Creating and Using Cached Translation Information". See @bugref{6568}.
1956 */
1957 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
1958 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
1959 HMVMX_SET_TAGGED_TLB_FLUSHED();
1960 }
1961
1962 /** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere. See hmQueueInvlPage()
1963 * where it is commented out. Support individual entry flushing
1964 * someday. */
1965#if 0
1966 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN))
1967 {
1968 STAM_COUNTER_INC(&pVCpu->hm.s.StatTlbShootdown);
1969
1970 /*
1971 * Flush individual guest entries using VPID from the TLB or as little as possible with EPT
1972 * as supported by the CPU.
1973 */
1974 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
1975 {
1976 for (uint32_t i = 0; i < pVCpu->hm.s.TlbShootdown.cPages; i++)
1977 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_INDIV_ADDR, pVCpu->hm.s.TlbShootdown.aPages[i]);
1978 }
1979 else
1980 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
1981
1982 HMVMX_SET_TAGGED_TLB_FLUSHED();
1983 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN);
1984 pVCpu->hm.s.TlbShootdown.cPages = 0;
1985 }
1986#endif
1987
1988 pVCpu->hm.s.fForceTLBFlush = false;
1989
1990 HMVMX_UPDATE_FLUSH_SKIPPED_STAT();
1991
1992 Assert(pVCpu->hm.s.idLastCpu == pCpu->idCpu);
1993 Assert(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes);
1994 AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes,
1995 ("Flush count mismatch for cpu %d (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
1996 AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
1997 ("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pCpu->idCpu,
1998 pCpu->uCurrentAsid, pCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
1999 AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
2000 ("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
2001
2002 /* Update VMCS with the VPID. */
2003 int rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_VPID, pVCpu->hm.s.uCurrentAsid);
2004 AssertRC(rc);
2005
2006#undef HMVMX_SET_TAGGED_TLB_FLUSHED
2007}
2008
2009
2010/**
2011 * Flushes the tagged-TLB entries for EPT CPUs as necessary.
2012 *
2013 * @returns VBox status code.
2014 * @param pVM Pointer to the VM.
2015 * @param pVCpu Pointer to the VMCPU.
2016 * @param pCpu Pointer to the global HM CPU struct.
2017 *
2018 * @remarks Called with interrupts disabled.
2019 */
2020static void hmR0VmxFlushTaggedTlbEpt(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2021{
2022 AssertPtr(pVM);
2023 AssertPtr(pVCpu);
2024 AssertPtr(pCpu);
2025 Assert(pCpu->idCpu != NIL_RTCPUID);
2026 AssertMsg(pVM->hm.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with NestedPaging disabled."));
2027 AssertMsg(!pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID enabled."));
2028
2029 /*
2030 * Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
2031 * A change in the TLB flush count implies the host CPU is online after a suspend/resume.
2032 */
2033 if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
2034 || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
2035 {
2036 pVCpu->hm.s.fForceTLBFlush = true;
2037 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
2038 }
2039
2040 /* Check for explicit TLB shootdown flushes. */
2041 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
2042 {
2043 pVCpu->hm.s.fForceTLBFlush = true;
2044 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2045 }
2046
2047 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
2048 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
2049
2050 if (pVCpu->hm.s.fForceTLBFlush)
2051 {
2052 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
2053 pVCpu->hm.s.fForceTLBFlush = false;
2054 }
2055 /** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere. See hmQueueInvlPage()
2056 * where it is commented out. Support individual entry flushing
2057 * someday. */
2058#if 0
2059 else
2060 {
2061 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN))
2062 {
2063 /* We cannot flush individual entries without VPID support. Flush using EPT. */
2064 STAM_COUNTER_INC(&pVCpu->hm.s.StatTlbShootdown);
2065 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
2066 }
2067 else
2068 STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch);
2069
2070 pVCpu->hm.s.TlbShootdown.cPages = 0;
2071 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN);
2072 }
2073#endif
2074}
2075
2076
2077/**
2078 * Flushes the tagged-TLB entries for VPID CPUs as necessary.
2079 *
2080 * @returns VBox status code.
2081 * @param pVM Pointer to the VM.
2082 * @param pVCpu Pointer to the VMCPU.
2083 * @param pCpu Pointer to the global HM CPU struct.
2084 *
2085 * @remarks Called with interrupts disabled.
2086 */
2087static void hmR0VmxFlushTaggedTlbVpid(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2088{
2089 AssertPtr(pVM);
2090 AssertPtr(pVCpu);
2091 AssertPtr(pCpu);
2092 Assert(pCpu->idCpu != NIL_RTCPUID);
2093 AssertMsg(pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked with VPID disabled."));
2094 AssertMsg(!pVM->hm.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging enabled"));
2095
2096 /*
2097 * Force a TLB flush for the first world switch if the current CPU differs from the one we ran on last.
2098 * If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB
2099 * or the host CPU is online after a suspend/resume, so we cannot reuse the current ASID anymore.
2100 */
2101 if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
2102 || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
2103 {
2104 pVCpu->hm.s.fForceTLBFlush = true;
2105 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
2106 }
2107
2108 /* Check for explicit TLB shootdown flushes. */
2109 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
2110 {
2111 /*
2112 * If we ever support VPID flush combinations other than ALL or SINGLE-context (see hmR0VmxSetupTaggedTlb())
2113 * we would need to explicitly flush in this case (add an fExplicitFlush = true here and change the
2114 * pCpu->fFlushAsidBeforeUse check below to include fExplicitFlush's too) - an obscure corner case.
2115 */
2116 pVCpu->hm.s.fForceTLBFlush = true;
2117 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2118 }
2119
2120 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
2121 if (pVCpu->hm.s.fForceTLBFlush)
2122 {
2123 ++pCpu->uCurrentAsid;
2124 if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
2125 {
2126 pCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */
2127 pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
2128 pCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
2129 }
2130
2131 pVCpu->hm.s.fForceTLBFlush = false;
2132 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
2133 pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid;
2134 if (pCpu->fFlushAsidBeforeUse)
2135 {
2136 if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_SINGLE_CONTEXT)
2137 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_SINGLE_CONTEXT, 0 /* GCPtr */);
2138 else if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_ALL_CONTEXTS)
2139 {
2140 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_ALL_CONTEXTS, 0 /* GCPtr */);
2141 pCpu->fFlushAsidBeforeUse = false;
2142 }
2143 else
2144 {
2145 /* hmR0VmxSetupTaggedTlb() ensures we never get here. Paranoia. */
2146 AssertMsgFailed(("Unsupported VPID-flush context type.\n"));
2147 }
2148 }
2149 }
2150 /** @todo We never set VMCPU_FF_TLB_SHOOTDOWN anywhere. See hmQueueInvlPage()
2151 * where it is commented out. Support individual entry flushing
2152 * someday. */
2153#if 0
2154 else
2155 {
2156 AssertMsg(pVCpu->hm.s.uCurrentAsid && pCpu->uCurrentAsid,
2157 ("hm->uCurrentAsid=%lu hm->cTlbFlushes=%lu cpu->uCurrentAsid=%lu cpu->cTlbFlushes=%lu\n",
2158 pVCpu->hm.s.uCurrentAsid, pVCpu->hm.s.cTlbFlushes,
2159 pCpu->uCurrentAsid, pCpu->cTlbFlushes));
2160
2161 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN))
2162 {
2163 /* Flush individual guest entries using VPID or as little as possible with EPT as supported by the CPU. */
2164 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
2165 {
2166 for (uint32_t i = 0; i < pVCpu->hm.s.TlbShootdown.cPages; i++)
2167 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_INDIV_ADDR, pVCpu->hm.s.TlbShootdown.aPages[i]);
2168 }
2169 else
2170 hmR0VmxFlushVpid(pVM, pVCpu, pVM->hm.s.vmx.enmFlushVpid, 0 /* GCPtr */);
2171
2172 pVCpu->hm.s.TlbShootdown.cPages = 0;
2173 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_SHOOTDOWN);
2174 }
2175 else
2176 STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch);
2177 }
2178#endif
2179
2180 AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes,
2181 ("Flush count mismatch for cpu %d (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
2182 AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
2183 ("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pCpu->idCpu,
2184 pCpu->uCurrentAsid, pCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
2185 AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
2186 ("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
2187
2188 int rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_VPID, pVCpu->hm.s.uCurrentAsid);
2189 AssertRC(rc);
2190}
2191
2192
2193/**
2194 * Flushes the guest TLB entry based on CPU capabilities.
2195 *
2196 * @param pVCpu Pointer to the VMCPU.
2197 * @param pCpu Pointer to the global HM CPU struct.
2198 */
2199DECLINLINE(void) hmR0VmxFlushTaggedTlb(PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2200{
2201#ifdef HMVMX_ALWAYS_FLUSH_TLB
2202 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2203#endif
2204 PVM pVM = pVCpu->CTX_SUFF(pVM);
2205 switch (pVM->hm.s.vmx.uFlushTaggedTlb)
2206 {
2207 case HMVMX_FLUSH_TAGGED_TLB_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pVM, pVCpu, pCpu); break;
2208 case HMVMX_FLUSH_TAGGED_TLB_EPT: hmR0VmxFlushTaggedTlbEpt(pVM, pVCpu, pCpu); break;
2209 case HMVMX_FLUSH_TAGGED_TLB_VPID: hmR0VmxFlushTaggedTlbVpid(pVM, pVCpu, pCpu); break;
2210 case HMVMX_FLUSH_TAGGED_TLB_NONE: hmR0VmxFlushTaggedTlbNone(pVM, pVCpu, pCpu); break;
2211 default:
2212 AssertMsgFailed(("Invalid flush-tag function identifier\n"));
2213 break;
2214 }
2215
2216 /* VMCPU_FF_TLB_SHOOTDOWN is unused. */
2217 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_SHOOTDOWN));
2218
2219 /* Don't assert that VMCPU_FF_TLB_FLUSH should no longer be pending. It can be set by other EMTs. */
2220}
2221
2222
2223/**
2224 * Sets up the appropriate tagged TLB-flush level and handler for flushing guest
2225 * TLB entries from the host TLB before VM-entry.
2226 *
2227 * @returns VBox status code.
2228 * @param pVM Pointer to the VM.
2229 */
2230static int hmR0VmxSetupTaggedTlb(PVM pVM)
2231{
2232 /*
2233 * Determine optimal flush type for Nested Paging.
2234 * We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup unrestricted
2235 * guest execution (see hmR3InitFinalizeR0()).
2236 */
2237 if (pVM->hm.s.fNestedPaging)
2238 {
2239 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT)
2240 {
2241 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT)
2242 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_SINGLE_CONTEXT;
2243 else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
2244 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_ALL_CONTEXTS;
2245 else
2246 {
2247 /* Shouldn't happen. EPT is supported but no suitable flush-types supported. */
2248 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2249 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2250 }
2251
2252 /* Make sure the write-back cacheable memory type for EPT is supported. */
2253 if (!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB))
2254 {
2255 LogRel(("hmR0VmxSetupTaggedTlb: Unsupported EPTP memory type %#x.\n", pVM->hm.s.vmx.Msrs.u64EptVpidCaps));
2256 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2257 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2258 }
2259 }
2260 else
2261 {
2262 /* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */
2263 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2264 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2265 }
2266 }
2267
2268 /*
2269 * Determine optimal flush type for VPID.
2270 */
2271 if (pVM->hm.s.vmx.fVpid)
2272 {
2273 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID)
2274 {
2275 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT)
2276 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_SINGLE_CONTEXT;
2277 else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS)
2278 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_ALL_CONTEXTS;
2279 else
2280 {
2281 /* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */
2282 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
2283 LogRel(("hmR0VmxSetupTaggedTlb: Only INDIV_ADDR supported. Ignoring VPID.\n"));
2284 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
2285 LogRel(("hmR0VmxSetupTaggedTlb: Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
2286 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NOT_SUPPORTED;
2287 pVM->hm.s.vmx.fVpid = false;
2288 }
2289 }
2290 else
2291 {
2292 /* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */
2293 Log4(("hmR0VmxSetupTaggedTlb: VPID supported without INVEPT support. Ignoring VPID.\n"));
2294 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NOT_SUPPORTED;
2295 pVM->hm.s.vmx.fVpid = false;
2296 }
2297 }
2298
2299 /*
2300 * Setup the handler for flushing tagged-TLBs.
2301 */
2302 if (pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid)
2303 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_EPT_VPID;
2304 else if (pVM->hm.s.fNestedPaging)
2305 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_EPT;
2306 else if (pVM->hm.s.vmx.fVpid)
2307 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_VPID;
2308 else
2309 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_NONE;
2310 return VINF_SUCCESS;
2311}
2312
2313
2314/**
2315 * Sets up pin-based VM-execution controls in the VMCS.
2316 *
2317 * @returns VBox status code.
2318 * @param pVM Pointer to the VM.
2319 * @param pVCpu Pointer to the VMCPU.
2320 */
2321static int hmR0VmxSetupPinCtls(PVM pVM, PVMCPU pVCpu)
2322{
2323 AssertPtr(pVM);
2324 AssertPtr(pVCpu);
2325
2326 uint32_t val = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0; /* Bits set here must always be set. */
2327 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1; /* Bits cleared here must always be cleared. */
2328
2329 val |= VMX_VMCS_CTRL_PIN_EXEC_EXT_INT_EXIT /* External interrupts cause a VM-exit. */
2330 | VMX_VMCS_CTRL_PIN_EXEC_NMI_EXIT; /* Non-maskable interrupts (NMIs) cause a VM-exit. */
2331
2332 if (pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
2333 val |= VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI; /* Use virtual NMIs and virtual-NMI blocking features. */
2334
2335 /* Enable the VMX preemption timer. */
2336 if (pVM->hm.s.vmx.fUsePreemptTimer)
2337 {
2338 Assert(pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER);
2339 val |= VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER;
2340 }
2341
2342 if ((val & zap) != val)
2343 {
2344 LogRel(("hmR0VmxSetupPinCtls: invalid pin-based VM-execution controls combo! cpu=%#RX64 val=%#RX64 zap=%#RX64\n",
2345 pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0, val, zap));
2346 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC;
2347 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2348 }
2349
2350 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, val);
2351 AssertRCReturn(rc, rc);
2352
2353 pVCpu->hm.s.vmx.u32PinCtls = val;
2354 return rc;
2355}
2356
2357
2358/**
2359 * Sets up processor-based VM-execution controls in the VMCS.
2360 *
2361 * @returns VBox status code.
2362 * @param pVM Pointer to the VM.
2363 * @param pVMCPU Pointer to the VMCPU.
2364 */
2365static int hmR0VmxSetupProcCtls(PVM pVM, PVMCPU pVCpu)
2366{
2367 AssertPtr(pVM);
2368 AssertPtr(pVCpu);
2369
2370 int rc = VERR_INTERNAL_ERROR_5;
2371 uint32_t val = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0; /* Bits set here must be set in the VMCS. */
2372 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
2373
2374 val |= VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT /* HLT causes a VM-exit. */
2375 | VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING /* Use TSC-offsetting. */
2376 | VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */
2377 | VMX_VMCS_CTRL_PROC_EXEC_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */
2378 | VMX_VMCS_CTRL_PROC_EXEC_RDPMC_EXIT /* RDPMC causes a VM-exit. */
2379 | VMX_VMCS_CTRL_PROC_EXEC_MONITOR_EXIT /* MONITOR causes a VM-exit. */
2380 | VMX_VMCS_CTRL_PROC_EXEC_MWAIT_EXIT; /* MWAIT causes a VM-exit. */
2381
2382 /* We toggle VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */
2383 if ( !(pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT)
2384 || (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0 & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT))
2385 {
2386 LogRel(("hmR0VmxSetupProcCtls: unsupported VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT combo!"));
2387 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT;
2388 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2389 }
2390
2391 /* Without Nested Paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */
2392 if (!pVM->hm.s.fNestedPaging)
2393 {
2394 Assert(!pVM->hm.s.vmx.fUnrestrictedGuest); /* Paranoia. */
2395 val |= VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT
2396 | VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
2397 | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
2398 }
2399
2400 /* Use TPR shadowing if supported by the CPU. */
2401 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
2402 {
2403 Assert(pVCpu->hm.s.vmx.HCPhysVirtApic);
2404 Assert(!(pVCpu->hm.s.vmx.HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */
2405 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, 0);
2406 rc |= VMXWriteVmcs64(VMX_VMCS64_CTRL_VAPIC_PAGEADDR_FULL, pVCpu->hm.s.vmx.HCPhysVirtApic);
2407 AssertRCReturn(rc, rc);
2408
2409 val |= VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */
2410 /* CR8 writes cause a VM-exit based on TPR threshold. */
2411 Assert(!(val & VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT));
2412 Assert(!(val & VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT));
2413 }
2414 else
2415 {
2416 /*
2417 * Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is invalid on 32-bit Intel CPUs.
2418 * Set this control only for 64-bit guests.
2419 */
2420 if (pVM->hm.s.fAllow64BitGuests)
2421 {
2422 val |= VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT /* CR8 reads cause a VM-exit. */
2423 | VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT; /* CR8 writes cause a VM-exit. */
2424 }
2425 }
2426
2427 /* Use MSR-bitmaps if supported by the CPU. */
2428 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
2429 {
2430 val |= VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS;
2431
2432 Assert(pVCpu->hm.s.vmx.HCPhysMsrBitmap);
2433 Assert(!(pVCpu->hm.s.vmx.HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */
2434 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, pVCpu->hm.s.vmx.HCPhysMsrBitmap);
2435 AssertRCReturn(rc, rc);
2436
2437 /*
2438 * The guest can access the following MSRs (read, write) without causing VM-exits; they are loaded/stored
2439 * automatically using dedicated fields in the VMCS.
2440 */
2441 hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_CS, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2442 hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_ESP, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2443 hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_EIP, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2444 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_GS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2445 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_FS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2446
2447#if HC_ARCH_BITS == 64
2448 /*
2449 * Set passthru permissions for the following MSRs (mandatory for VT-x) required for 64-bit guests.
2450 */
2451 if (pVM->hm.s.fAllow64BitGuests)
2452 {
2453 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_LSTAR, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2454 hmR0VmxSetMsrPermission(pVCpu, MSR_K6_STAR, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2455 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_SF_MASK, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2456 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_KERNEL_GS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2457 }
2458#endif
2459 }
2460
2461 /* Use the secondary processor-based VM-execution controls if supported by the CPU. */
2462 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
2463 val |= VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL;
2464
2465 if ((val & zap) != val)
2466 {
2467 LogRel(("hmR0VmxSetupProcCtls: invalid processor-based VM-execution controls combo! cpu=%#RX64 val=%#RX64 zap=%#RX64\n",
2468 pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0, val, zap));
2469 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC;
2470 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2471 }
2472
2473 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, val);
2474 AssertRCReturn(rc, rc);
2475
2476 pVCpu->hm.s.vmx.u32ProcCtls = val;
2477
2478 /*
2479 * Secondary processor-based VM-execution controls.
2480 */
2481 if (RT_LIKELY(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL))
2482 {
2483 val = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0; /* Bits set here must be set in the VMCS. */
2484 zap = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
2485
2486 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT)
2487 val |= VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT; /* WBINVD causes a VM-exit. */
2488
2489 if (pVM->hm.s.fNestedPaging)
2490 val |= VMX_VMCS_CTRL_PROC_EXEC2_EPT; /* Enable EPT. */
2491 else
2492 {
2493 /*
2494 * Without Nested Paging, INVPCID should cause a VM-exit. Enabling this bit causes the CPU to refer to
2495 * VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT when INVPCID is executed by the guest.
2496 * See Intel spec. 25.4 "Changes to instruction behaviour in VMX non-root operation".
2497 */
2498 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_INVPCID)
2499 val |= VMX_VMCS_CTRL_PROC_EXEC2_INVPCID;
2500 }
2501
2502 if (pVM->hm.s.vmx.fVpid)
2503 val |= VMX_VMCS_CTRL_PROC_EXEC2_VPID; /* Enable VPID. */
2504
2505 if (pVM->hm.s.vmx.fUnrestrictedGuest)
2506 val |= VMX_VMCS_CTRL_PROC_EXEC2_UNRESTRICTED_GUEST; /* Enable Unrestricted Execution. */
2507
2508 /* Enable Virtual-APIC page accesses if supported by the CPU. This is essentially where the TPR shadow resides. */
2509 /** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be
2510 * done dynamically. */
2511 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)
2512 {
2513 Assert(pVM->hm.s.vmx.HCPhysApicAccess);
2514 Assert(!(pVM->hm.s.vmx.HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */
2515 val |= VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC; /* Virtualize APIC accesses. */
2516 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, pVM->hm.s.vmx.HCPhysApicAccess);
2517 AssertRCReturn(rc, rc);
2518 }
2519
2520 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP)
2521 val |= VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP; /* Enable RDTSCP support. */
2522
2523 if ((val & zap) != val)
2524 {
2525 LogRel(("hmR0VmxSetupProcCtls: invalid secondary processor-based VM-execution controls combo! "
2526 "cpu=%#RX64 val=%#RX64 zap=%#RX64\n", pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0, val, zap));
2527 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC2;
2528 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2529 }
2530
2531 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, val);
2532 AssertRCReturn(rc, rc);
2533
2534 pVCpu->hm.s.vmx.u32ProcCtls2 = val;
2535 }
2536 else if (RT_UNLIKELY(pVM->hm.s.vmx.fUnrestrictedGuest))
2537 {
2538 LogRel(("hmR0VmxSetupProcCtls: Unrestricted Guest set as true when secondary processor-based VM-execution controls not "
2539 "available\n"));
2540 pVCpu->hm.s.u32HMError = VMX_UFC_INVALID_UX_COMBO;
2541 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2542 }
2543
2544 return VINF_SUCCESS;
2545}
2546
2547
2548/**
2549 * Sets up miscellaneous (everything other than Pin & Processor-based
2550 * VM-execution) control fields in the VMCS.
2551 *
2552 * @returns VBox status code.
2553 * @param pVM Pointer to the VM.
2554 * @param pVCpu Pointer to the VMCPU.
2555 */
2556static int hmR0VmxSetupMiscCtls(PVM pVM, PVMCPU pVCpu)
2557{
2558 NOREF(pVM);
2559 AssertPtr(pVM);
2560 AssertPtr(pVCpu);
2561
2562 int rc = VERR_GENERAL_FAILURE;
2563
2564 /* All fields are zero-initialized during allocation; but don't remove the commented block below. */
2565#if 0
2566 /* All CR3 accesses cause VM-exits. Later we optimize CR3 accesses (see hmR0VmxLoadGuestCR3AndCR4())*/
2567 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, 0); AssertRCReturn(rc, rc);
2568 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, 0); AssertRCReturn(rc, rc);
2569
2570 /*
2571 * Set MASK & MATCH to 0. VMX checks if GuestPFErrCode & MASK == MATCH. If equal (in our case it always is)
2572 * 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.
2573 * We thus use the exception bitmap to control it rather than use both.
2574 */
2575 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, 0); AssertRCReturn(rc, rc);
2576 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, 0); AssertRCReturn(rc, rc);
2577
2578 /** @todo Explore possibility of using IO-bitmaps. */
2579 /* All IO & IOIO instructions cause VM-exits. */
2580 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_IO_BITMAP_A_FULL, 0); AssertRCReturn(rc, rc);
2581 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_IO_BITMAP_B_FULL, 0); AssertRCReturn(rc, rc);
2582
2583 /* Initialize the MSR-bitmap area. */
2584 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, 0); AssertRCReturn(rc, rc);
2585 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, 0); AssertRCReturn(rc, rc);
2586 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, 0); AssertRCReturn(rc, rc);
2587#endif
2588
2589 /* Setup MSR auto-load/store area. */
2590 Assert(pVCpu->hm.s.vmx.HCPhysGuestMsr);
2591 Assert(!(pVCpu->hm.s.vmx.HCPhysGuestMsr & 0xf)); /* Lower 4 bits MBZ. */
2592 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, pVCpu->hm.s.vmx.HCPhysGuestMsr);
2593 AssertRCReturn(rc, rc);
2594 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, pVCpu->hm.s.vmx.HCPhysGuestMsr);
2595 AssertRCReturn(rc, rc);
2596
2597 Assert(pVCpu->hm.s.vmx.HCPhysHostMsr);
2598 Assert(!(pVCpu->hm.s.vmx.HCPhysHostMsr & 0xf)); /* Lower 4 bits MBZ. */
2599 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, pVCpu->hm.s.vmx.HCPhysHostMsr);
2600 AssertRCReturn(rc, rc);
2601
2602 /* Set VMCS link pointer. Reserved for future use, must be -1. Intel spec. 24.4 "Guest-State Area". */
2603 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, UINT64_C(0xffffffffffffffff));
2604 AssertRCReturn(rc, rc);
2605
2606 /* All fields are zero-initialized during allocation; but don't remove the commented block below. */
2607#if 0
2608 /* Setup debug controls */
2609 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, 0); /** @todo We don't support IA32_DEBUGCTL MSR. Should we? */
2610 AssertRCReturn(rc, rc);
2611 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, 0);
2612 AssertRCReturn(rc, rc);
2613#endif
2614
2615 return rc;
2616}
2617
2618
2619/**
2620 * Sets up the initial exception bitmap in the VMCS based on static conditions
2621 * (i.e. conditions that cannot ever change after starting the VM).
2622 *
2623 * @returns VBox status code.
2624 * @param pVM Pointer to the VM.
2625 * @param pVCpu Pointer to the VMCPU.
2626 */
2627static int hmR0VmxInitXcptBitmap(PVM pVM, PVMCPU pVCpu)
2628{
2629 AssertPtr(pVM);
2630 AssertPtr(pVCpu);
2631
2632 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
2633
2634 uint32_t u32XcptBitmap = 0;
2635
2636 /* Without Nested Paging, #PF must cause a VM-exit so we can sync our shadow page tables. */
2637 if (!pVM->hm.s.fNestedPaging)
2638 u32XcptBitmap |= RT_BIT(X86_XCPT_PF);
2639
2640 pVCpu->hm.s.vmx.u32XcptBitmap = u32XcptBitmap;
2641 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
2642 AssertRCReturn(rc, rc);
2643 return rc;
2644}
2645
2646
2647/**
2648 * Sets up the initial guest-state mask. The guest-state mask is consulted
2649 * before reading guest-state fields from the VMCS as VMREADs can be expensive
2650 * for the nested virtualization case (as it would cause a VM-exit).
2651 *
2652 * @param pVCpu Pointer to the VMCPU.
2653 */
2654static int hmR0VmxInitUpdatedGuestStateMask(PVMCPU pVCpu)
2655{
2656 /* Initially the guest-state is up-to-date as there is nothing in the VMCS. */
2657 HMVMXCPU_GST_RESET_TO(pVCpu, HMVMX_UPDATED_GUEST_ALL);
2658 return VINF_SUCCESS;
2659}
2660
2661
2662/**
2663 * Does per-VM VT-x initialization.
2664 *
2665 * @returns VBox status code.
2666 * @param pVM Pointer to the VM.
2667 */
2668VMMR0DECL(int) VMXR0InitVM(PVM pVM)
2669{
2670 LogFlowFunc(("pVM=%p\n", pVM));
2671
2672 int rc = hmR0VmxStructsAlloc(pVM);
2673 if (RT_FAILURE(rc))
2674 {
2675 LogRel(("VMXR0InitVM: hmR0VmxStructsAlloc failed! rc=%Rrc\n", rc));
2676 return rc;
2677 }
2678
2679 return VINF_SUCCESS;
2680}
2681
2682
2683/**
2684 * Does per-VM VT-x termination.
2685 *
2686 * @returns VBox status code.
2687 * @param pVM Pointer to the VM.
2688 */
2689VMMR0DECL(int) VMXR0TermVM(PVM pVM)
2690{
2691 LogFlowFunc(("pVM=%p\n", pVM));
2692
2693#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2694 if (pVM->hm.s.vmx.hMemObjScratch != NIL_RTR0MEMOBJ)
2695 ASMMemZero32(pVM->hm.s.vmx.pvScratch, PAGE_SIZE);
2696#endif
2697 hmR0VmxStructsFree(pVM);
2698 return VINF_SUCCESS;
2699}
2700
2701
2702/**
2703 * Sets up the VM for execution under VT-x.
2704 * This function is only called once per-VM during initialization.
2705 *
2706 * @returns VBox status code.
2707 * @param pVM Pointer to the VM.
2708 */
2709VMMR0DECL(int) VMXR0SetupVM(PVM pVM)
2710{
2711 AssertPtrReturn(pVM, VERR_INVALID_PARAMETER);
2712 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2713
2714 LogFlowFunc(("pVM=%p\n", pVM));
2715
2716 /*
2717 * Without UnrestrictedGuest, pRealModeTSS and pNonPagingModeEPTPageTable *must* always be allocated.
2718 * We no longer support the highly unlikely case of UnrestrictedGuest without pRealModeTSS. See hmR3InitFinalizeR0Intel().
2719 */
2720 if ( !pVM->hm.s.vmx.fUnrestrictedGuest
2721 && ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable
2722 || !pVM->hm.s.vmx.pRealModeTSS))
2723 {
2724 LogRel(("VMXR0SetupVM: invalid real-on-v86 state.\n"));
2725 return VERR_INTERNAL_ERROR;
2726 }
2727
2728#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
2729 /*
2730 * This is for the darwin 32-bit/PAE kernels trying to execute 64-bit guests. We don't bother with
2731 * the 32<->64 switcher in this case. This is a rare, legacy use-case with barely any test coverage.
2732 */
2733 if ( pVM->hm.s.fAllow64BitGuests
2734 && !HMVMX_IS_64BIT_HOST_MODE())
2735 {
2736 LogRel(("VMXR0SetupVM: Unsupported guest and host paging mode combination.\n"));
2737 return VERR_PGM_UNSUPPORTED_HOST_PAGING_MODE;
2738 }
2739#endif
2740
2741 /* Initialize these always, see hmR3InitFinalizeR0().*/
2742 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NONE;
2743 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NONE;
2744
2745 /* Setup the tagged-TLB flush handlers. */
2746 int rc = hmR0VmxSetupTaggedTlb(pVM);
2747 if (RT_FAILURE(rc))
2748 {
2749 LogRel(("VMXR0SetupVM: hmR0VmxSetupTaggedTlb failed! rc=%Rrc\n", rc));
2750 return rc;
2751 }
2752
2753 /* Check if we can use the VMCS controls for swapping the EFER MSR. */
2754 Assert(!pVM->hm.s.vmx.fSupportsVmcsEfer);
2755#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
2756 if ( HMVMX_IS_64BIT_HOST_MODE()
2757 && (pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1 & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR)
2758 && (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR)
2759 && (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR))
2760 {
2761 pVM->hm.s.vmx.fSupportsVmcsEfer = true;
2762 }
2763#endif
2764
2765 for (VMCPUID i = 0; i < pVM->cCpus; i++)
2766 {
2767 PVMCPU pVCpu = &pVM->aCpus[i];
2768 AssertPtr(pVCpu);
2769 AssertPtr(pVCpu->hm.s.vmx.pvVmcs);
2770
2771 /* Log the VCPU pointers, useful for debugging SMP VMs. */
2772 Log4(("VMXR0SetupVM: pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu));
2773
2774 /* Initialize the VM-exit history array with end-of-array markers (UINT16_MAX). */
2775 Assert(!pVCpu->hm.s.idxExitHistoryFree);
2776 HMCPU_EXIT_HISTORY_RESET(pVCpu);
2777
2778 /* Set revision dword at the beginning of the VMCS structure. */
2779 *(uint32_t *)pVCpu->hm.s.vmx.pvVmcs = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo);
2780
2781 /* Initialize our VMCS region in memory, set the VMCS launch state to "clear". */
2782 rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2783 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXClearVmcs failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2784 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2785
2786 /* Load this VMCS as the current VMCS. */
2787 rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2788 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXActivateVmcs failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2789 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2790
2791 rc = hmR0VmxSetupPinCtls(pVM, pVCpu);
2792 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupPinCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2793 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2794
2795 rc = hmR0VmxSetupProcCtls(pVM, pVCpu);
2796 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupProcCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2797 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2798
2799 rc = hmR0VmxSetupMiscCtls(pVM, pVCpu);
2800 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupMiscCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2801 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2802
2803 rc = hmR0VmxInitXcptBitmap(pVM, pVCpu);
2804 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitXcptBitmap failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2805 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2806
2807 rc = hmR0VmxInitUpdatedGuestStateMask(pVCpu);
2808 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitUpdatedGuestStateMask failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2809 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2810
2811#if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
2812 rc = hmR0VmxInitVmcsReadCache(pVM, pVCpu);
2813 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitVmcsReadCache failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2814 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2815#endif
2816
2817 /* Re-sync the CPU's internal data into our VMCS memory region & reset the launch state to "clear". */
2818 rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2819 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXClearVmcs(2) failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2820 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2821
2822 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
2823
2824 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc);
2825 }
2826
2827 return VINF_SUCCESS;
2828}
2829
2830
2831/**
2832 * Saves the host control registers (CR0, CR3, CR4) into the host-state area in
2833 * the VMCS.
2834 *
2835 * @returns VBox status code.
2836 * @param pVM Pointer to the VM.
2837 * @param pVCpu Pointer to the VMCPU.
2838 */
2839DECLINLINE(int) hmR0VmxSaveHostControlRegs(PVM pVM, PVMCPU pVCpu)
2840{
2841 NOREF(pVM); NOREF(pVCpu);
2842
2843 RTCCUINTREG uReg = ASMGetCR0();
2844 int rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR0, uReg);
2845 AssertRCReturn(rc, rc);
2846
2847#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
2848 /* For the darwin 32-bit hybrid kernel, we need the 64-bit CR3 as it uses 64-bit paging. */
2849 if (HMVMX_IS_64BIT_HOST_MODE())
2850 {
2851 uint64_t uRegCR3 = HMR0Get64bitCR3();
2852 rc = VMXWriteVmcs64(VMX_VMCS_HOST_CR3, uRegCR3);
2853 }
2854 else
2855#endif
2856 {
2857 uReg = ASMGetCR3();
2858 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR3, uReg);
2859 }
2860 AssertRCReturn(rc, rc);
2861
2862 uReg = ASMGetCR4();
2863 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR4, uReg);
2864 AssertRCReturn(rc, rc);
2865 return rc;
2866}
2867
2868
2869#if HC_ARCH_BITS == 64
2870/**
2871 * Macro for adjusting host segment selectors to satisfy VT-x's VM-entry
2872 * requirements. See hmR0VmxSaveHostSegmentRegs().
2873 */
2874# define VMXLOCAL_ADJUST_HOST_SEG(seg, selValue) \
2875 if ((selValue) & (X86_SEL_RPL | X86_SEL_LDT)) \
2876 { \
2877 bool fValidSelector = true; \
2878 if ((selValue) & X86_SEL_LDT) \
2879 { \
2880 uint32_t uAttr = ASMGetSegAttr((selValue)); \
2881 fValidSelector = RT_BOOL(uAttr != UINT32_MAX && (uAttr & X86_DESC_P)); \
2882 } \
2883 if (fValidSelector) \
2884 { \
2885 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_SEL_##seg; \
2886 pVCpu->hm.s.vmx.RestoreHost.uHostSel##seg = (selValue); \
2887 } \
2888 (selValue) = 0; \
2889 }
2890#endif
2891
2892
2893/**
2894 * Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into
2895 * the host-state area in the VMCS.
2896 *
2897 * @returns VBox status code.
2898 * @param pVM Pointer to the VM.
2899 * @param pVCpu Pointer to the VMCPU.
2900 */
2901DECLINLINE(int) hmR0VmxSaveHostSegmentRegs(PVM pVM, PVMCPU pVCpu)
2902{
2903 int rc = VERR_INTERNAL_ERROR_5;
2904
2905#if HC_ARCH_BITS == 64
2906 /*
2907 * If we've executed guest code using VT-x, the host-state bits will be messed up. We
2908 * should -not- save the messed up state without restoring the original host-state. See @bugref{7240}.
2909 */
2910 AssertMsgReturn(!(pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED),
2911 ("Re-saving host-state after executing guest code without leaving VT-x!\n"), VERR_WRONG_ORDER);
2912#endif
2913
2914 /*
2915 * Host DS, ES, FS and GS segment registers.
2916 */
2917#if HC_ARCH_BITS == 64
2918 RTSEL uSelDS = ASMGetDS();
2919 RTSEL uSelES = ASMGetES();
2920 RTSEL uSelFS = ASMGetFS();
2921 RTSEL uSelGS = ASMGetGS();
2922#else
2923 RTSEL uSelDS = 0;
2924 RTSEL uSelES = 0;
2925 RTSEL uSelFS = 0;
2926 RTSEL uSelGS = 0;
2927#endif
2928
2929 /* Recalculate which host-state bits need to be manually restored. */
2930 pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
2931
2932 /*
2933 * Host CS and SS segment registers.
2934 */
2935#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
2936 RTSEL uSelCS;
2937 RTSEL uSelSS;
2938 if (HMVMX_IS_64BIT_HOST_MODE())
2939 {
2940 uSelCS = (RTSEL)(uintptr_t)&SUPR0Abs64bitKernelCS;
2941 uSelSS = (RTSEL)(uintptr_t)&SUPR0Abs64bitKernelSS;
2942 }
2943 else
2944 {
2945 /* Seems darwin uses the LDT (TI flag is set) in the CS & SS selectors which VT-x doesn't like. */
2946 uSelCS = (RTSEL)(uintptr_t)&SUPR0AbsKernelCS;
2947 uSelSS = (RTSEL)(uintptr_t)&SUPR0AbsKernelSS;
2948 }
2949#else
2950 RTSEL uSelCS = ASMGetCS();
2951 RTSEL uSelSS = ASMGetSS();
2952#endif
2953
2954 /*
2955 * Host TR segment register.
2956 */
2957 RTSEL uSelTR = ASMGetTR();
2958
2959#if HC_ARCH_BITS == 64
2960 /*
2961 * Determine if the host segment registers are suitable for VT-x. Otherwise use zero to gain VM-entry and restore them
2962 * before we get preempted. See Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers".
2963 */
2964 VMXLOCAL_ADJUST_HOST_SEG(DS, uSelDS);
2965 VMXLOCAL_ADJUST_HOST_SEG(ES, uSelES);
2966 VMXLOCAL_ADJUST_HOST_SEG(FS, uSelFS);
2967 VMXLOCAL_ADJUST_HOST_SEG(GS, uSelGS);
2968# undef VMXLOCAL_ADJUST_HOST_SEG
2969#endif
2970
2971 /* Verification based on Intel spec. 26.2.3 "Checks on Host Segment and Descriptor-Table Registers" */
2972 Assert(!(uSelCS & X86_SEL_RPL)); Assert(!(uSelCS & X86_SEL_LDT));
2973 Assert(!(uSelSS & X86_SEL_RPL)); Assert(!(uSelSS & X86_SEL_LDT));
2974 Assert(!(uSelDS & X86_SEL_RPL)); Assert(!(uSelDS & X86_SEL_LDT));
2975 Assert(!(uSelES & X86_SEL_RPL)); Assert(!(uSelES & X86_SEL_LDT));
2976 Assert(!(uSelFS & X86_SEL_RPL)); Assert(!(uSelFS & X86_SEL_LDT));
2977 Assert(!(uSelGS & X86_SEL_RPL)); Assert(!(uSelGS & X86_SEL_LDT));
2978 Assert(!(uSelTR & X86_SEL_RPL)); Assert(!(uSelTR & X86_SEL_LDT));
2979 Assert(uSelCS);
2980 Assert(uSelTR);
2981
2982 /* Assertion is right but we would not have updated u32ExitCtls yet. */
2983#if 0
2984 if (!(pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE))
2985 Assert(uSelSS != 0);
2986#endif
2987
2988 /* Write these host selector fields into the host-state area in the VMCS. */
2989 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_CS, uSelCS); AssertRCReturn(rc, rc);
2990 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_SS, uSelSS); AssertRCReturn(rc, rc);
2991#if HC_ARCH_BITS == 64
2992 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_DS, uSelDS); AssertRCReturn(rc, rc);
2993 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_ES, uSelES); AssertRCReturn(rc, rc);
2994 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_FS, uSelFS); AssertRCReturn(rc, rc);
2995 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_GS, uSelGS); AssertRCReturn(rc, rc);
2996#endif
2997 rc = VMXWriteVmcs32(VMX_VMCS16_HOST_FIELD_TR, uSelTR); AssertRCReturn(rc, rc);
2998
2999 /*
3000 * Host GDTR and IDTR.
3001 */
3002 RTGDTR Gdtr;
3003 RT_ZERO(Gdtr);
3004#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
3005 if (HMVMX_IS_64BIT_HOST_MODE())
3006 {
3007 X86XDTR64 Gdtr64;
3008 X86XDTR64 Idtr64;
3009 HMR0Get64bitGdtrAndIdtr(&Gdtr64, &Idtr64);
3010 rc = VMXWriteVmcs64(VMX_VMCS_HOST_GDTR_BASE, Gdtr64.uAddr); AssertRCReturn(rc, rc);
3011 rc = VMXWriteVmcs64(VMX_VMCS_HOST_IDTR_BASE, Idtr64.uAddr); AssertRCReturn(rc, rc);
3012
3013 Gdtr.cbGdt = Gdtr64.cb;
3014 Gdtr.pGdt = (uintptr_t)Gdtr64.uAddr;
3015 }
3016 else
3017#endif
3018 {
3019 RTIDTR Idtr;
3020 ASMGetGDTR(&Gdtr);
3021 ASMGetIDTR(&Idtr);
3022 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, Gdtr.pGdt); AssertRCReturn(rc, rc);
3023 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, Idtr.pIdt); AssertRCReturn(rc, rc);
3024
3025#if HC_ARCH_BITS == 64
3026 /*
3027 * Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps them to the
3028 * maximum limit (0xffff) on every VM-exit.
3029 */
3030 if (Gdtr.cbGdt != 0xffff)
3031 {
3032 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_GDTR;
3033 AssertCompile(sizeof(Gdtr) == sizeof(X86XDTR64));
3034 memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
3035 }
3036
3037 /*
3038 * IDT limit is effectively capped at 0xfff. (See Intel spec. 6.14.1 "64-Bit Mode IDT"
3039 * and Intel spec. 6.2 "Exception and Interrupt Vectors".) Therefore if the host has the limit as 0xfff, VT-x
3040 * bloating the limit to 0xffff shouldn't cause any different CPU behavior. However, several hosts either insists
3041 * on 0xfff being the limit (Windows Patch Guard) or uses the limit for other purposes (darwin puts the CPU ID in there
3042 * but botches sidt alignment in at least one consumer). So, we're only allowing IDTR.LIMIT to be left at 0xffff on
3043 * hosts where we are pretty sure it won't cause trouble.
3044 */
3045# if defined(RT_OS_LINUX) || defined(RT_OS_SOLARIS)
3046 if (Idtr.cbIdt < 0x0fff)
3047# else
3048 if (Idtr.cbIdt != 0xffff)
3049# endif
3050 {
3051 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_IDTR;
3052 AssertCompile(sizeof(Idtr) == sizeof(X86XDTR64));
3053 memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostIdtr, &Idtr, sizeof(X86XDTR64));
3054 }
3055#endif
3056 }
3057
3058 /*
3059 * Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI and RPL bits
3060 * is effectively what the CPU does for "scaling by 8". TI is always 0 and RPL should be too in most cases.
3061 */
3062 if ((uSelTR | X86_SEL_RPL_LDT) > Gdtr.cbGdt)
3063 {
3064 AssertMsgFailed(("hmR0VmxSaveHostSegmentRegs: TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, Gdtr.cbGdt));
3065 return VERR_VMX_INVALID_HOST_STATE;
3066 }
3067
3068 PCX86DESCHC pDesc = (PCX86DESCHC)(Gdtr.pGdt + (uSelTR & X86_SEL_MASK));
3069#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
3070 if (HMVMX_IS_64BIT_HOST_MODE())
3071 {
3072 /* We need the 64-bit TR base for hybrid darwin. */
3073 uint64_t u64TRBase = X86DESC64_BASE((PX86DESC64)pDesc);
3074 rc = VMXWriteVmcs64(VMX_VMCS_HOST_TR_BASE, u64TRBase);
3075 }
3076 else
3077#endif
3078 {
3079 uintptr_t uTRBase;
3080#if HC_ARCH_BITS == 64
3081 uTRBase = X86DESC64_BASE(pDesc);
3082
3083 /*
3084 * VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on all VM-exits.
3085 * The type is the same for 64-bit busy TSS[1]. The limit needs manual restoration if the host has something else.
3086 * Task switching is not supported in 64-bit mode[2], but the limit still matters as IOPM is supported in 64-bit mode.
3087 * Restoring the limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0.
3088 *
3089 * [1] See Intel spec. 3.5 "System Descriptor Types".
3090 * [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode".
3091 */
3092 Assert(pDesc->System.u4Type == 11);
3093 if ( pDesc->System.u16LimitLow != 0x67
3094 || pDesc->System.u4LimitHigh)
3095 {
3096 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_SEL_TR;
3097 /* If the host has made GDT read-only, we would need to temporarily toggle CR0.WP before writing the GDT. */
3098 if (pVM->hm.s.uHostKernelFeatures & SUPKERNELFEATURES_GDT_READ_ONLY)
3099 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_GDT_READ_ONLY;
3100 pVCpu->hm.s.vmx.RestoreHost.uHostSelTR = uSelTR;
3101
3102 /* Store the GDTR here as we need it while restoring TR. */
3103 memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
3104 }
3105#else
3106 uTRBase = X86DESC_BASE(pDesc);
3107#endif
3108 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_TR_BASE, uTRBase);
3109 }
3110 AssertRCReturn(rc, rc);
3111
3112 /*
3113 * Host FS base and GS base.
3114 */
3115#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
3116 if (HMVMX_IS_64BIT_HOST_MODE())
3117 {
3118 uint64_t u64FSBase = ASMRdMsr(MSR_K8_FS_BASE);
3119 uint64_t u64GSBase = ASMRdMsr(MSR_K8_GS_BASE);
3120 rc = VMXWriteVmcs64(VMX_VMCS_HOST_FS_BASE, u64FSBase); AssertRCReturn(rc, rc);
3121 rc = VMXWriteVmcs64(VMX_VMCS_HOST_GS_BASE, u64GSBase); AssertRCReturn(rc, rc);
3122
3123# if HC_ARCH_BITS == 64
3124 /* Store the base if we have to restore FS or GS manually as we need to restore the base as well. */
3125 if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_FS)
3126 pVCpu->hm.s.vmx.RestoreHost.uHostFSBase = u64FSBase;
3127 if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_GS)
3128 pVCpu->hm.s.vmx.RestoreHost.uHostGSBase = u64GSBase;
3129# endif
3130 }
3131#endif
3132 return rc;
3133}
3134
3135
3136/**
3137 * Saves certain host MSRs in the VM-Exit MSR-load area and some in the
3138 * host-state area of the VMCS. Theses MSRs will be automatically restored on
3139 * the host after every successful VM-exit.
3140 *
3141 * @returns VBox status code.
3142 * @param pVM Pointer to the VM.
3143 * @param pVCpu Pointer to the VMCPU.
3144 *
3145 * @remarks No-long-jump zone!!!
3146 */
3147DECLINLINE(int) hmR0VmxSaveHostMsrs(PVM pVM, PVMCPU pVCpu)
3148{
3149 NOREF(pVM);
3150
3151 AssertPtr(pVCpu);
3152 AssertPtr(pVCpu->hm.s.vmx.pvHostMsr);
3153
3154 int rc = VINF_SUCCESS;
3155#if HC_ARCH_BITS == 64
3156 if (pVM->hm.s.fAllow64BitGuests)
3157 hmR0VmxLazySaveHostMsrs(pVCpu);
3158#endif
3159
3160 /*
3161 * Host Sysenter MSRs.
3162 */
3163 rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS));
3164 AssertRCReturn(rc, rc);
3165#ifdef VBOX_WITH_HYBRID_32BIT_KERNEL
3166 if (HMVMX_IS_64BIT_HOST_MODE())
3167 {
3168 rc = VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP));
3169 AssertRCReturn(rc, rc);
3170 rc = VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP));
3171 }
3172 else
3173 {
3174 rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr_Low(MSR_IA32_SYSENTER_ESP));
3175 AssertRCReturn(rc, rc);
3176 rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr_Low(MSR_IA32_SYSENTER_EIP));
3177 }
3178#elif HC_ARCH_BITS == 32
3179 rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr_Low(MSR_IA32_SYSENTER_ESP));
3180 AssertRCReturn(rc, rc);
3181 rc = VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr_Low(MSR_IA32_SYSENTER_EIP));
3182#else
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#endif
3187 AssertRCReturn(rc, rc);
3188
3189 /*
3190 * Host EFER MSR.
3191 * If the CPU supports the newer VMCS controls for managing EFER, use it.
3192 * Otherwise it's done as part of auto-load/store MSR area in the VMCS, see hmR0VmxLoadGuestMsrs().
3193 */
3194 if (pVM->hm.s.vmx.fSupportsVmcsEfer)
3195 {
3196 rc = VMXWriteVmcs64(VMX_VMCS64_HOST_FIELD_EFER_FULL, pVM->hm.s.vmx.u64HostEfer);
3197 AssertRCReturn(rc, rc);
3198 }
3199
3200 /** @todo IA32_PERF_GLOBALCTRL, IA32_PAT also see
3201 * hmR0VmxLoadGuestExitCtls() !! */
3202
3203 return rc;
3204}
3205
3206
3207/**
3208 * Figures out if we need to swap the EFER MSR which is
3209 * particularly expensive.
3210 *
3211 * We check all relevant bits. For now, that's everything
3212 * besides LMA/LME, as these two bits are handled by VM-entry,
3213 * see hmR0VmxLoadGuestExitCtls() and
3214 * hmR0VMxLoadGuestEntryCtls().
3215 *
3216 * @returns true if we need to load guest EFER, false otherwise.
3217 * @param pVCpu Pointer to the VMCPU.
3218 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3219 * out-of-sync. Make sure to update the required fields
3220 * before using them.
3221 *
3222 * @remarks Requires EFER, CR4.
3223 * @remarks No-long-jump zone!!!
3224 */
3225static bool hmR0VmxShouldSwapEferMsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3226{
3227#ifdef HMVMX_ALWAYS_SWAP_EFER
3228 return true;
3229#endif
3230
3231#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
3232 /* For 32-bit hosts running 64-bit guests, we always swap EFER in the world-switcher. Nothing to do here. */
3233 if (CPUMIsGuestInLongMode(pVCpu))
3234 return false;
3235#endif
3236
3237 PVM pVM = pVCpu->CTX_SUFF(pVM);
3238 uint64_t u64HostEfer = pVM->hm.s.vmx.u64HostEfer;
3239 uint64_t u64GuestEfer = pMixedCtx->msrEFER;
3240
3241 /*
3242 * For 64-bit guests, if EFER.SCE bit differs, we need to swap to ensure that the
3243 * guest's SYSCALL behaviour isn't screwed. See @bugref{7386}.
3244 */
3245 if ( CPUMIsGuestInLongMode(pVCpu)
3246 && (u64GuestEfer & MSR_K6_EFER_SCE) != (u64HostEfer & MSR_K6_EFER_SCE))
3247 {
3248 return true;
3249 }
3250
3251 /*
3252 * If the guest uses PAE and EFER.NXE bit differs, we need to swap EFER as it .
3253 * affects guest paging. 64-bit paging implies CR4.PAE as well.
3254 * See Intel spec. 4.5 "IA-32e Paging" and Intel spec. 4.1.1 "Three Paging Modes".
3255 */
3256 if ( (pMixedCtx->cr4 & X86_CR4_PAE)
3257 && (pMixedCtx->cr0 & X86_CR0_PG)
3258 && (u64GuestEfer & MSR_K6_EFER_NXE) != (u64HostEfer & MSR_K6_EFER_NXE))
3259 {
3260 /* Assert that host is PAE capable. */
3261 Assert(pVM->hm.s.cpuid.u32AMDFeatureEDX & X86_CPUID_EXT_FEATURE_EDX_NX);
3262 return true;
3263 }
3264
3265 /** @todo Check the latest Intel spec. for any other bits,
3266 * like SMEP/SMAP? */
3267 return false;
3268}
3269
3270
3271/**
3272 * Sets up VM-entry controls in the VMCS. These controls can affect things done
3273 * on VM-exit; e.g. "load debug controls", see Intel spec. 24.8.1 "VM-entry
3274 * controls".
3275 *
3276 * @returns VBox status code.
3277 * @param pVCpu Pointer to the VMCPU.
3278 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3279 * out-of-sync. Make sure to update the required fields
3280 * before using them.
3281 *
3282 * @remarks Requires EFER.
3283 * @remarks No-long-jump zone!!!
3284 */
3285DECLINLINE(int) hmR0VmxLoadGuestEntryCtls(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3286{
3287 int rc = VINF_SUCCESS;
3288 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_ENTRY_CTLS))
3289 {
3290 PVM pVM = pVCpu->CTX_SUFF(pVM);
3291 uint32_t val = pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0; /* Bits set here must be set in the VMCS. */
3292 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3293
3294 /* Load debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x capable CPUs only supports the 1-setting of this bit. */
3295 val |= VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG;
3296
3297 /* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry. */
3298 if (CPUMIsGuestInLongModeEx(pMixedCtx))
3299 {
3300 val |= VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST;
3301 Log4(("Load[%RU32]: VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST\n", pVCpu->idCpu));
3302 }
3303 else
3304 Assert(!(val & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST));
3305
3306 /* If the CPU supports the newer VMCS controls for managing guest/host EFER, use it. */
3307 if ( pVM->hm.s.vmx.fSupportsVmcsEfer
3308 && hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
3309 {
3310 val |= VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR;
3311 Log4(("Load[%RU32]: VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR\n", pVCpu->idCpu));
3312 }
3313
3314 /*
3315 * The following should -not- be set (since we're not in SMM mode):
3316 * - VMX_VMCS_CTRL_ENTRY_ENTRY_SMM
3317 * - VMX_VMCS_CTRL_ENTRY_DEACTIVATE_DUALMON
3318 */
3319
3320 /** @todo VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR,
3321 * VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR. */
3322
3323 if ((val & zap) != val)
3324 {
3325 LogRel(("hmR0VmxLoadGuestEntryCtls: invalid VM-entry controls combo! cpu=%RX64 val=%RX64 zap=%RX64\n",
3326 pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0, val, zap));
3327 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY;
3328 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3329 }
3330
3331 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, val);
3332 AssertRCReturn(rc, rc);
3333
3334 pVCpu->hm.s.vmx.u32EntryCtls = val;
3335 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_ENTRY_CTLS);
3336 }
3337 return rc;
3338}
3339
3340
3341/**
3342 * Sets up the VM-exit controls in the VMCS.
3343 *
3344 * @returns VBox status code.
3345 * @param pVM Pointer to the VM.
3346 * @param pVCpu Pointer to the VMCPU.
3347 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3348 * out-of-sync. Make sure to update the required fields
3349 * before using them.
3350 *
3351 * @remarks Requires EFER.
3352 */
3353DECLINLINE(int) hmR0VmxLoadGuestExitCtls(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3354{
3355 NOREF(pMixedCtx);
3356
3357 int rc = VINF_SUCCESS;
3358 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_EXIT_CTLS))
3359 {
3360 PVM pVM = pVCpu->CTX_SUFF(pVM);
3361 uint32_t val = pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0; /* Bits set here must be set in the VMCS. */
3362 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3363
3364 /* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only supported the 1-setting of this bit. */
3365 val |= VMX_VMCS_CTRL_EXIT_SAVE_DEBUG;
3366
3367 /*
3368 * Set the host long mode active (EFER.LMA) bit (which Intel calls "Host address-space size") if necessary.
3369 * On VM-exit, VT-x sets both the host EFER.LMA and EFER.LME bit to this value. See assertion in hmR0VmxSaveHostMsrs().
3370 */
3371#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
3372 if (HMVMX_IS_64BIT_HOST_MODE())
3373 {
3374 val |= VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE;
3375 Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE\n", pVCpu->idCpu));
3376 }
3377 else
3378 Assert(!(val & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE));
3379#else
3380 if (CPUMIsGuestInLongModeEx(pMixedCtx))
3381 {
3382 /* The switcher returns to long mode, EFER is managed by the switcher. */
3383 val |= VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE;
3384 Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE\n", pVCpu->idCpu));
3385 }
3386 else
3387 Assert(!(val & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE));
3388#endif /* HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL) */
3389
3390 /* If the newer VMCS fields for managing EFER exists, use it. */
3391 if ( pVM->hm.s.vmx.fSupportsVmcsEfer
3392 && hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
3393 {
3394 val |= VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR
3395 | VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR;
3396 Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR, VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR\n", pVCpu->idCpu));
3397 }
3398
3399 /* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */
3400 Assert(!(val & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT));
3401
3402 /** @todo VMX_VMCS_CTRL_EXIT_LOAD_PERF_MSR,
3403 * VMX_VMCS_CTRL_EXIT_SAVE_GUEST_PAT_MSR,
3404 * VMX_VMCS_CTRL_EXIT_LOAD_HOST_PAT_MSR. */
3405
3406 if (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER)
3407 val |= VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER;
3408
3409 if ((val & zap) != val)
3410 {
3411 LogRel(("hmR0VmxSetupProcCtls: invalid VM-exit controls combo! cpu=%RX64 val=%RX64 zap=%RX64\n",
3412 pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0, val, zap));
3413 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT;
3414 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3415 }
3416
3417 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, val);
3418 AssertRCReturn(rc, rc);
3419
3420 pVCpu->hm.s.vmx.u32ExitCtls = val;
3421 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_EXIT_CTLS);
3422 }
3423 return rc;
3424}
3425
3426
3427/**
3428 * Loads the guest APIC and related state.
3429 *
3430 * @returns VBox status code.
3431 * @param pVM Pointer to the VM.
3432 * @param pVCpu Pointer to the VMCPU.
3433 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3434 * out-of-sync. Make sure to update the required fields
3435 * before using them.
3436 */
3437DECLINLINE(int) hmR0VmxLoadGuestApicState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3438{
3439 NOREF(pMixedCtx);
3440
3441 int rc = VINF_SUCCESS;
3442 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE))
3443 {
3444 /* Setup TPR shadowing. Also setup TPR patching for 32-bit guests. */
3445 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
3446 {
3447 Assert(pVCpu->hm.s.vmx.HCPhysVirtApic);
3448
3449 bool fPendingIntr = false;
3450 uint8_t u8Tpr = 0;
3451 uint8_t u8PendingIntr = 0;
3452 rc = PDMApicGetTPR(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr);
3453 AssertRCReturn(rc, rc);
3454
3455 /*
3456 * If there are external interrupts pending but masked by the TPR value, instruct VT-x to cause a VM-exit when
3457 * the guest lowers its TPR below the highest-priority pending interrupt and we can deliver the interrupt.
3458 * If there are no external interrupts pending, set threshold to 0 to not cause a VM-exit. We will eventually deliver
3459 * the interrupt when we VM-exit for other reasons.
3460 */
3461 pVCpu->hm.s.vmx.pbVirtApic[0x80] = u8Tpr; /* Offset 0x80 is TPR in the APIC MMIO range. */
3462 uint32_t u32TprThreshold = 0;
3463 if (fPendingIntr)
3464 {
3465 /* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR (which is the Task-Priority Class). */
3466 const uint8_t u8PendingPriority = (u8PendingIntr >> 4) & 0xf;
3467 const uint8_t u8TprPriority = (u8Tpr >> 4) & 0xf;
3468 if (u8PendingPriority <= u8TprPriority)
3469 u32TprThreshold = u8PendingPriority;
3470 else
3471 u32TprThreshold = u8TprPriority; /* Required for Vista 64-bit guest, see @bugref{6398}. */
3472 }
3473 Assert(!(u32TprThreshold & 0xfffffff0)); /* Bits 31:4 MBZ. */
3474
3475 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
3476 AssertRCReturn(rc, rc);
3477 }
3478
3479 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
3480 }
3481 return rc;
3482}
3483
3484
3485/**
3486 * Gets the guest's interruptibility-state ("interrupt shadow" as AMD calls it).
3487 *
3488 * @returns Guest's interruptibility-state.
3489 * @param pVCpu Pointer to the VMCPU.
3490 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3491 * out-of-sync. Make sure to update the required fields
3492 * before using them.
3493 *
3494 * @remarks No-long-jump zone!!!
3495 */
3496DECLINLINE(uint32_t) hmR0VmxGetGuestIntrState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3497{
3498 /*
3499 * Check if we should inhibit interrupt delivery due to instructions like STI and MOV SS.
3500 */
3501 uint32_t uIntrState = 0;
3502 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
3503 {
3504 /* If inhibition is active, RIP & RFLAGS should've been accessed (i.e. read previously from the VMCS or from ring-3). */
3505 AssertMsg(HMVMXCPU_GST_IS_SET(pVCpu, HMVMX_UPDATED_GUEST_RIP | HMVMX_UPDATED_GUEST_RFLAGS),
3506 ("%#x\n", HMVMXCPU_GST_VALUE(pVCpu)));
3507 if (pMixedCtx->rip == EMGetInhibitInterruptsPC(pVCpu))
3508 {
3509 if (pMixedCtx->eflags.Bits.u1IF)
3510 uIntrState = VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI;
3511 else
3512 uIntrState = VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS;
3513 }
3514 /* else: Although we can clear the force-flag here, let's keep this side-effects free. */
3515 }
3516
3517 /*
3518 * NMIs to the guest are blocked after an NMI is injected until the guest executes an IRET. We only
3519 * bother with virtual-NMI blocking when we have support for virtual NMIs in the CPU, otherwise
3520 * setting this would block host-NMIs and IRET will not clear the blocking.
3521 *
3522 * See Intel spec. 26.6.1 "Interruptibility state". See @bugref{7445}.
3523 */
3524 if ( VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS)
3525 && (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI))
3526 {
3527 uIntrState |= VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI;
3528 }
3529
3530 return uIntrState;
3531}
3532
3533
3534/**
3535 * Loads the guest's interruptibility-state into the guest-state area in the
3536 * VMCS.
3537 *
3538 * @returns VBox status code.
3539 * @param pVCpu Pointer to the VMCPU.
3540 * @param uIntrState The interruptibility-state to set.
3541 */
3542static int hmR0VmxLoadGuestIntrState(PVMCPU pVCpu, uint32_t uIntrState)
3543{
3544 NOREF(pVCpu);
3545 AssertMsg(!(uIntrState & 0xfffffff0), ("%#x\n", uIntrState)); /* Bits 31:4 MBZ. */
3546 Assert((uIntrState & 0x3) != 0x3); /* Block-by-STI and MOV SS cannot be simultaneously set. */
3547 int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, uIntrState);
3548 AssertRCReturn(rc, rc);
3549 return rc;
3550}
3551
3552
3553/**
3554 * Loads the guest's RIP into the guest-state area in the VMCS.
3555 *
3556 * @returns VBox status code.
3557 * @param pVCpu Pointer to the VMCPU.
3558 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3559 * out-of-sync. Make sure to update the required fields
3560 * before using them.
3561 *
3562 * @remarks No-long-jump zone!!!
3563 */
3564static int hmR0VmxLoadGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3565{
3566 int rc = VINF_SUCCESS;
3567 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RIP))
3568 {
3569 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RIP, pMixedCtx->rip);
3570 AssertRCReturn(rc, rc);
3571
3572 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RIP);
3573 Log4(("Load[%RU32]: VMX_VMCS_GUEST_RIP=%#RX64 fContextUseFlags=%#RX32\n", pVCpu->idCpu, pMixedCtx->rip,
3574 HMCPU_CF_VALUE(pVCpu)));
3575 }
3576 return rc;
3577}
3578
3579
3580/**
3581 * Loads the guest's RSP into the guest-state area in the VMCS.
3582 *
3583 * @returns VBox status code.
3584 * @param pVCpu Pointer to the VMCPU.
3585 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3586 * out-of-sync. Make sure to update the required fields
3587 * before using them.
3588 *
3589 * @remarks No-long-jump zone!!!
3590 */
3591static int hmR0VmxLoadGuestRsp(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3592{
3593 int rc = VINF_SUCCESS;
3594 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RSP))
3595 {
3596 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RSP, pMixedCtx->rsp);
3597 AssertRCReturn(rc, rc);
3598
3599 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RSP);
3600 Log4(("Load[%RU32]: VMX_VMCS_GUEST_RSP=%#RX64\n", pVCpu->idCpu, pMixedCtx->rsp));
3601 }
3602 return rc;
3603}
3604
3605
3606/**
3607 * Loads the guest's RFLAGS into the guest-state area in the VMCS.
3608 *
3609 * @returns VBox status code.
3610 * @param pVCpu Pointer to the VMCPU.
3611 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3612 * out-of-sync. Make sure to update the required fields
3613 * before using them.
3614 *
3615 * @remarks No-long-jump zone!!!
3616 */
3617static int hmR0VmxLoadGuestRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3618{
3619 int rc = VINF_SUCCESS;
3620 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS))
3621 {
3622 /* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ).
3623 Let us assert it as such and use 32-bit VMWRITE. */
3624 Assert(!(pMixedCtx->rflags.u64 >> 32));
3625 X86EFLAGS Eflags = pMixedCtx->eflags;
3626 Eflags.u32 &= VMX_EFLAGS_RESERVED_0; /* Bits 22-31, 15, 5 & 3 MBZ. */
3627 Eflags.u32 |= VMX_EFLAGS_RESERVED_1; /* Bit 1 MB1. */
3628
3629 /*
3630 * If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so we can restore them on VM-exit.
3631 * Modify the real-mode guest's eflags so that VT-x can run the real-mode guest code under Virtual 8086 mode.
3632 */
3633 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
3634 {
3635 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
3636 Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
3637 pVCpu->hm.s.vmx.RealMode.Eflags.u32 = Eflags.u32; /* Save the original eflags of the real-mode guest. */
3638 Eflags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */
3639 Eflags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */
3640 }
3641
3642 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_RFLAGS, Eflags.u32);
3643 AssertRCReturn(rc, rc);
3644
3645 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RFLAGS);
3646 Log4(("Load[%RU32]: VMX_VMCS_GUEST_RFLAGS=%#RX32\n", pVCpu->idCpu, Eflags.u32));
3647 }
3648 return rc;
3649}
3650
3651
3652/**
3653 * Loads the guest RIP, RSP and RFLAGS into the guest-state area in the VMCS.
3654 *
3655 * @returns VBox status code.
3656 * @param pVCpu Pointer to the VMCPU.
3657 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3658 * out-of-sync. Make sure to update the required fields
3659 * before using them.
3660 *
3661 * @remarks No-long-jump zone!!!
3662 */
3663DECLINLINE(int) hmR0VmxLoadGuestRipRspRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3664{
3665 int rc = hmR0VmxLoadGuestRip(pVCpu, pMixedCtx);
3666 AssertRCReturn(rc, rc);
3667 rc = hmR0VmxLoadGuestRsp(pVCpu, pMixedCtx);
3668 AssertRCReturn(rc, rc);
3669 rc = hmR0VmxLoadGuestRflags(pVCpu, pMixedCtx);
3670 AssertRCReturn(rc, rc);
3671 return rc;
3672}
3673
3674
3675/**
3676 * Loads the guest CR0 control register into the guest-state area in the VMCS.
3677 * CR0 is partially shared with the host and we have to consider the FPU bits.
3678 *
3679 * @returns VBox status code.
3680 * @param pVM Pointer to the VM.
3681 * @param pVCpu Pointer to the VMCPU.
3682 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3683 * out-of-sync. Make sure to update the required fields
3684 * before using them.
3685 *
3686 * @remarks No-long-jump zone!!!
3687 */
3688static int hmR0VmxLoadSharedCR0(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3689{
3690 /*
3691 * Guest CR0.
3692 * Guest FPU.
3693 */
3694 int rc = VINF_SUCCESS;
3695 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0))
3696 {
3697 Assert(!(pMixedCtx->cr0 >> 32));
3698 uint32_t u32GuestCR0 = pMixedCtx->cr0;
3699 PVM pVM = pVCpu->CTX_SUFF(pVM);
3700
3701 /* The guest's view (read access) of its CR0 is unblemished. */
3702 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, u32GuestCR0);
3703 AssertRCReturn(rc, rc);
3704 Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR0_READ_SHADOW=%#RX32\n", pVCpu->idCpu, u32GuestCR0));
3705
3706 /* Setup VT-x's view of the guest CR0. */
3707 /* Minimize VM-exits due to CR3 changes when we have NestedPaging. */
3708 if (pVM->hm.s.fNestedPaging)
3709 {
3710 if (CPUMIsGuestPagingEnabledEx(pMixedCtx))
3711 {
3712 /* The guest has paging enabled, let it access CR3 without causing a VM-exit if supported. */
3713 pVCpu->hm.s.vmx.u32ProcCtls &= ~( VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
3714 | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT);
3715 }
3716 else
3717 {
3718 /* The guest doesn't have paging enabled, make CR3 access cause a VM-exit to update our shadow. */
3719 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
3720 | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
3721 }
3722
3723 /* If we have unrestricted guest execution, we never have to intercept CR3 reads. */
3724 if (pVM->hm.s.vmx.fUnrestrictedGuest)
3725 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
3726
3727 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
3728 AssertRCReturn(rc, rc);
3729 }
3730 else
3731 u32GuestCR0 |= X86_CR0_WP; /* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */
3732
3733 /*
3734 * Guest FPU bits.
3735 * Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be set on the first
3736 * CPUs to support VT-x and no mention of with regards to UX in VM-entry checks.
3737 */
3738 u32GuestCR0 |= X86_CR0_NE;
3739 bool fInterceptNM = false;
3740 if (CPUMIsGuestFPUStateActive(pVCpu))
3741 {
3742 fInterceptNM = false; /* Guest FPU active, no need to VM-exit on #NM. */
3743 /* The guest should still get #NM exceptions when it expects it to, so we should not clear TS & MP bits here.
3744 We're only concerned about -us- not intercepting #NMs when the guest-FPU is active. Not the guest itself! */
3745 }
3746 else
3747 {
3748 fInterceptNM = true; /* Guest FPU inactive, VM-exit on #NM for lazy FPU loading. */
3749 u32GuestCR0 |= X86_CR0_TS /* Guest can task switch quickly and do lazy FPU syncing. */
3750 | X86_CR0_MP; /* FWAIT/WAIT should not ignore CR0.TS and should generate #NM. */
3751 }
3752
3753 /* Catch floating point exceptions if we need to report them to the guest in a different way. */
3754 bool fInterceptMF = false;
3755 if (!(pMixedCtx->cr0 & X86_CR0_NE))
3756 fInterceptMF = true;
3757
3758 /* Finally, intercept all exceptions as we cannot directly inject them in real-mode, see hmR0VmxInjectEventVmcs(). */
3759 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
3760 {
3761 Assert(PDMVmmDevHeapIsEnabled(pVM));
3762 Assert(pVM->hm.s.vmx.pRealModeTSS);
3763 pVCpu->hm.s.vmx.u32XcptBitmap |= HMVMX_REAL_MODE_XCPT_MASK;
3764 fInterceptNM = true;
3765 fInterceptMF = true;
3766 }
3767 else
3768 pVCpu->hm.s.vmx.u32XcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK;
3769
3770 if (fInterceptNM)
3771 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_NM);
3772 else
3773 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_NM);
3774
3775 if (fInterceptMF)
3776 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_MF);
3777 else
3778 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_MF);
3779
3780 /* Additional intercepts for debugging, define these yourself explicitly. */
3781#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
3782 pVCpu->hm.s.vmx.u32XcptBitmap |= 0
3783 | RT_BIT(X86_XCPT_BP)
3784 | RT_BIT(X86_XCPT_DB)
3785 | RT_BIT(X86_XCPT_DE)
3786 | RT_BIT(X86_XCPT_NM)
3787 | RT_BIT(X86_XCPT_TS)
3788 | RT_BIT(X86_XCPT_UD)
3789 | RT_BIT(X86_XCPT_NP)
3790 | RT_BIT(X86_XCPT_SS)
3791 | RT_BIT(X86_XCPT_GP)
3792 | RT_BIT(X86_XCPT_PF)
3793 | RT_BIT(X86_XCPT_MF)
3794 ;
3795#elif defined(HMVMX_ALWAYS_TRAP_PF)
3796 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_PF);
3797#endif
3798
3799 Assert(pVM->hm.s.fNestedPaging || (pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT(X86_XCPT_PF)));
3800
3801 /* Set/clear the CR0 specific bits along with their exceptions (PE, PG, CD, NW). */
3802 uint32_t uSetCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
3803 uint32_t uZapCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
3804 if (pVM->hm.s.vmx.fUnrestrictedGuest) /* Exceptions for unrestricted-guests for fixed CR0 bits (PE, PG). */
3805 uSetCR0 &= ~(X86_CR0_PE | X86_CR0_PG);
3806 else
3807 Assert((uSetCR0 & (X86_CR0_PE | X86_CR0_PG)) == (X86_CR0_PE | X86_CR0_PG));
3808
3809 u32GuestCR0 |= uSetCR0;
3810 u32GuestCR0 &= uZapCR0;
3811 u32GuestCR0 &= ~(X86_CR0_CD | X86_CR0_NW); /* Always enable caching. */
3812
3813 /* Write VT-x's view of the guest CR0 into the VMCS and update the exception bitmap. */
3814 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR0, u32GuestCR0);
3815 AssertRCReturn(rc, rc);
3816 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
3817 AssertRCReturn(rc, rc);
3818 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR0=%#RX32 (uSetCR0=%#RX32 uZapCR0=%#RX32)\n", pVCpu->idCpu, u32GuestCR0, uSetCR0,
3819 uZapCR0));
3820
3821 /*
3822 * CR0 is shared between host and guest along with a CR0 read shadow. Therefore, certain bits must not be changed
3823 * by the guest because VT-x ignores saving/restoring them (namely CD, ET, NW) and for certain other bits
3824 * we want to be notified immediately of guest CR0 changes (e.g. PG to update our shadow page tables).
3825 */
3826 uint32_t u32CR0Mask = 0;
3827 u32CR0Mask = X86_CR0_PE
3828 | X86_CR0_NE
3829 | X86_CR0_WP
3830 | X86_CR0_PG
3831 | X86_CR0_ET /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.ET */
3832 | X86_CR0_CD /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.CD */
3833 | X86_CR0_NW; /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.NW */
3834
3835 /** @todo Avoid intercepting CR0.PE with unrestricted guests. Fix PGM
3836 * enmGuestMode to be in-sync with the current mode. See @bugref{6398}
3837 * and @bugref{6944}. */
3838#if 0
3839 if (pVM->hm.s.vmx.fUnrestrictedGuest)
3840 u32CR0Mask &= ~X86_CR0_PE;
3841#endif
3842 if (pVM->hm.s.fNestedPaging)
3843 u32CR0Mask &= ~X86_CR0_WP;
3844
3845 /* If the guest FPU state is active, don't need to VM-exit on writes to FPU related bits in CR0. */
3846 if (fInterceptNM)
3847 {
3848 u32CR0Mask |= X86_CR0_TS
3849 | X86_CR0_MP;
3850 }
3851
3852 /* Write the CR0 mask into the VMCS and update the VCPU's copy of the current CR0 mask. */
3853 pVCpu->hm.s.vmx.u32CR0Mask = u32CR0Mask;
3854 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_MASK, u32CR0Mask);
3855 AssertRCReturn(rc, rc);
3856 Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR0_MASK=%#RX32\n", pVCpu->idCpu, u32CR0Mask));
3857
3858 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR0);
3859 }
3860 return rc;
3861}
3862
3863
3864/**
3865 * Loads the guest control registers (CR3, CR4) into the guest-state area
3866 * in the VMCS.
3867 *
3868 * @returns VBox status code.
3869 * @param pVM Pointer to the VM.
3870 * @param pVCpu Pointer to the VMCPU.
3871 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3872 * out-of-sync. Make sure to update the required fields
3873 * before using them.
3874 *
3875 * @remarks No-long-jump zone!!!
3876 */
3877static int hmR0VmxLoadGuestCR3AndCR4(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3878{
3879 int rc = VINF_SUCCESS;
3880 PVM pVM = pVCpu->CTX_SUFF(pVM);
3881
3882 /*
3883 * Guest CR2.
3884 * It's always loaded in the assembler code. Nothing to do here.
3885 */
3886
3887 /*
3888 * Guest CR3.
3889 */
3890 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR3))
3891 {
3892 RTGCPHYS GCPhysGuestCR3 = NIL_RTGCPHYS;
3893 if (pVM->hm.s.fNestedPaging)
3894 {
3895 pVCpu->hm.s.vmx.HCPhysEPTP = PGMGetHyperCR3(pVCpu);
3896
3897 /* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */
3898 Assert(pVCpu->hm.s.vmx.HCPhysEPTP);
3899 Assert(!(pVCpu->hm.s.vmx.HCPhysEPTP & UINT64_C(0xfff0000000000000)));
3900 Assert(!(pVCpu->hm.s.vmx.HCPhysEPTP & 0xfff));
3901
3902 /* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */
3903 pVCpu->hm.s.vmx.HCPhysEPTP |= VMX_EPT_MEMTYPE_WB
3904 | (VMX_EPT_PAGE_WALK_LENGTH_DEFAULT << VMX_EPT_PAGE_WALK_LENGTH_SHIFT);
3905
3906 /* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */
3907 AssertMsg( ((pVCpu->hm.s.vmx.HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */
3908 && ((pVCpu->hm.s.vmx.HCPhysEPTP >> 6) & 0x3f) == 0, /* Bits 6:11 MBZ. */
3909 ("EPTP %#RX64\n", pVCpu->hm.s.vmx.HCPhysEPTP));
3910
3911 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVCpu->hm.s.vmx.HCPhysEPTP);
3912 AssertRCReturn(rc, rc);
3913 Log4(("Load[%RU32]: VMX_VMCS64_CTRL_EPTP_FULL=%#RX64\n", pVCpu->idCpu, pVCpu->hm.s.vmx.HCPhysEPTP));
3914
3915 if ( pVM->hm.s.vmx.fUnrestrictedGuest
3916 || CPUMIsGuestPagingEnabledEx(pMixedCtx))
3917 {
3918 /* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */
3919 if (CPUMIsGuestInPAEModeEx(pMixedCtx))
3920 {
3921 rc = PGMGstGetPaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]); AssertRCReturn(rc, rc);
3922 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pVCpu->hm.s.aPdpes[0].u); AssertRCReturn(rc, rc);
3923 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pVCpu->hm.s.aPdpes[1].u); AssertRCReturn(rc, rc);
3924 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pVCpu->hm.s.aPdpes[2].u); AssertRCReturn(rc, rc);
3925 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pVCpu->hm.s.aPdpes[3].u); AssertRCReturn(rc, rc);
3926 }
3927
3928 /* The guest's view of its CR3 is unblemished with Nested Paging when the guest is using paging or we
3929 have Unrestricted Execution to handle the guest when it's not using paging. */
3930 GCPhysGuestCR3 = pMixedCtx->cr3;
3931 }
3932 else
3933 {
3934 /*
3935 * The guest is not using paging, but the CPU (VT-x) has to. While the guest thinks it accesses physical memory
3936 * directly, we use our identity-mapped page table to map guest-linear to guest-physical addresses.
3937 * EPT takes care of translating it to host-physical addresses.
3938 */
3939 RTGCPHYS GCPhys;
3940 Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable);
3941 Assert(PDMVmmDevHeapIsEnabled(pVM));
3942
3943 /* We obtain it here every time as the guest could have relocated this PCI region. */
3944 rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
3945 AssertRCReturn(rc, rc);
3946
3947 GCPhysGuestCR3 = GCPhys;
3948 }
3949
3950 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR3=%#RGv (GstN)\n", pVCpu->idCpu, GCPhysGuestCR3));
3951 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_CR3, GCPhysGuestCR3);
3952 }
3953 else
3954 {
3955 /* Non-nested paging case, just use the hypervisor's CR3. */
3956 RTHCPHYS HCPhysGuestCR3 = PGMGetHyperCR3(pVCpu);
3957
3958 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR3=%#RHv (HstN)\n", pVCpu->idCpu, HCPhysGuestCR3));
3959 rc = VMXWriteVmcsHstN(VMX_VMCS_GUEST_CR3, HCPhysGuestCR3);
3960 }
3961 AssertRCReturn(rc, rc);
3962
3963 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR3);
3964 }
3965
3966 /*
3967 * Guest CR4.
3968 */
3969 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR4))
3970 {
3971 Assert(!(pMixedCtx->cr4 >> 32));
3972 uint32_t u32GuestCR4 = pMixedCtx->cr4;
3973
3974 /* The guest's view of its CR4 is unblemished. */
3975 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, u32GuestCR4);
3976 AssertRCReturn(rc, rc);
3977 Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR4_READ_SHADOW=%#RX32\n", pVCpu->idCpu, u32GuestCR4));
3978
3979 /* Setup VT-x's view of the guest CR4. */
3980 /*
3981 * If we're emulating real-mode using virtual-8086 mode, we want to redirect software interrupts to the 8086 program
3982 * interrupt handler. Clear the VME bit (the interrupt redirection bitmap is already all 0, see hmR3InitFinalizeR0())
3983 * See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode".
3984 */
3985 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
3986 {
3987 Assert(pVM->hm.s.vmx.pRealModeTSS);
3988 Assert(PDMVmmDevHeapIsEnabled(pVM));
3989 u32GuestCR4 &= ~X86_CR4_VME;
3990 }
3991
3992 if (pVM->hm.s.fNestedPaging)
3993 {
3994 if ( !CPUMIsGuestPagingEnabledEx(pMixedCtx)
3995 && !pVM->hm.s.vmx.fUnrestrictedGuest)
3996 {
3997 /* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */
3998 u32GuestCR4 |= X86_CR4_PSE;
3999 /* Our identity mapping is a 32-bit page directory. */
4000 u32GuestCR4 &= ~X86_CR4_PAE;
4001 }
4002 /* else use guest CR4.*/
4003 }
4004 else
4005 {
4006 /*
4007 * The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host
4008 * paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables.
4009 */
4010 switch (pVCpu->hm.s.enmShadowMode)
4011 {
4012 case PGMMODE_REAL: /* Real-mode. */
4013 case PGMMODE_PROTECTED: /* Protected mode without paging. */
4014 case PGMMODE_32_BIT: /* 32-bit paging. */
4015 {
4016 u32GuestCR4 &= ~X86_CR4_PAE;
4017 break;
4018 }
4019
4020 case PGMMODE_PAE: /* PAE paging. */
4021 case PGMMODE_PAE_NX: /* PAE paging with NX. */
4022 {
4023 u32GuestCR4 |= X86_CR4_PAE;
4024 break;
4025 }
4026
4027 case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
4028 case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
4029#ifdef VBOX_ENABLE_64_BITS_GUESTS
4030 break;
4031#endif
4032 default:
4033 AssertFailed();
4034 return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
4035 }
4036 }
4037
4038 /* We need to set and clear the CR4 specific bits here (mainly the X86_CR4_VMXE bit). */
4039 uint64_t uSetCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
4040 uint64_t uZapCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
4041 u32GuestCR4 |= uSetCR4;
4042 u32GuestCR4 &= uZapCR4;
4043
4044 /* Write VT-x's view of the guest CR4 into the VMCS. */
4045 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR4=%#RX32 (Set=%#RX32 Zap=%#RX32)\n", pVCpu->idCpu, u32GuestCR4, uSetCR4, uZapCR4));
4046 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR4, u32GuestCR4);
4047 AssertRCReturn(rc, rc);
4048
4049 /* Setup CR4 mask. CR4 flags owned by the host, if the guest attempts to change them, that would cause a VM-exit. */
4050 uint32_t u32CR4Mask = 0;
4051 u32CR4Mask = X86_CR4_VME
4052 | X86_CR4_PAE
4053 | X86_CR4_PGE
4054 | X86_CR4_PSE
4055 | X86_CR4_VMXE;
4056 pVCpu->hm.s.vmx.u32CR4Mask = u32CR4Mask;
4057 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_MASK, u32CR4Mask);
4058 AssertRCReturn(rc, rc);
4059
4060 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR4);
4061 }
4062 return rc;
4063}
4064
4065
4066/**
4067 * Loads the guest debug registers into the guest-state area in the VMCS.
4068 * This also sets up whether #DB and MOV DRx accesses cause VM-exits.
4069 *
4070 * The guest debug bits are partially shared with the host (e.g. DR6, DR0-3).
4071 *
4072 * @returns VBox status code.
4073 * @param pVCpu Pointer to the VMCPU.
4074 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4075 * out-of-sync. Make sure to update the required fields
4076 * before using them.
4077 *
4078 * @remarks No-long-jump zone!!!
4079 */
4080static int hmR0VmxLoadSharedDebugState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4081{
4082 if (!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG))
4083 return VINF_SUCCESS;
4084
4085#ifdef VBOX_STRICT
4086 /* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */
4087 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
4088 {
4089 /* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */
4090 Assert((pMixedCtx->dr[7] & (X86_DR7_MBZ_MASK | X86_DR7_RAZ_MASK)) == 0); /* Bits 63:32, 15, 14, 12, 11 are reserved. */
4091 Assert((pMixedCtx->dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK); /* Bit 10 is reserved (RA1). */
4092 }
4093#endif
4094
4095 int rc;
4096 PVM pVM = pVCpu->CTX_SUFF(pVM);
4097 bool fInterceptDB = false;
4098 bool fInterceptMovDRx = false;
4099 if ( pVCpu->hm.s.fSingleInstruction
4100 || DBGFIsStepping(pVCpu))
4101 {
4102 /* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */
4103 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG)
4104 {
4105 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG;
4106 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
4107 AssertRCReturn(rc, rc);
4108 Assert(fInterceptDB == false);
4109 }
4110 else
4111 {
4112 pMixedCtx->eflags.u32 |= X86_EFL_TF;
4113 pVCpu->hm.s.fClearTrapFlag = true;
4114 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
4115 fInterceptDB = true;
4116 }
4117 }
4118
4119 if ( fInterceptDB
4120 || (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
4121 {
4122 /*
4123 * Use the combined guest and host DRx values found in the hypervisor
4124 * register set because the debugger has breakpoints active or someone
4125 * is single stepping on the host side without a monitor trap flag.
4126 *
4127 * Note! DBGF expects a clean DR6 state before executing guest code.
4128 */
4129#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4130 if ( CPUMIsGuestInLongModeEx(pMixedCtx)
4131 && !CPUMIsHyperDebugStateActivePending(pVCpu))
4132 {
4133 CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
4134 Assert(CPUMIsHyperDebugStateActivePending(pVCpu));
4135 Assert(!CPUMIsGuestDebugStateActivePending(pVCpu));
4136 }
4137 else
4138#endif
4139 if (!CPUMIsHyperDebugStateActive(pVCpu))
4140 {
4141 CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
4142 Assert(CPUMIsHyperDebugStateActive(pVCpu));
4143 Assert(!CPUMIsGuestDebugStateActive(pVCpu));
4144 }
4145
4146 /* Update DR7. (The other DRx values are handled by CPUM one way or the other.) */
4147 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)CPUMGetHyperDR7(pVCpu));
4148 AssertRCReturn(rc, rc);
4149
4150 pVCpu->hm.s.fUsingHyperDR7 = true;
4151 fInterceptDB = true;
4152 fInterceptMovDRx = true;
4153 }
4154 else
4155 {
4156 /*
4157 * If the guest has enabled debug registers, we need to load them prior to
4158 * executing guest code so they'll trigger at the right time.
4159 */
4160 if (pMixedCtx->dr[7] & (X86_DR7_ENABLED_MASK | X86_DR7_GD)) /** @todo Why GD? */
4161 {
4162#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4163 if ( CPUMIsGuestInLongModeEx(pMixedCtx)
4164 && !CPUMIsGuestDebugStateActivePending(pVCpu))
4165 {
4166 CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
4167 Assert(CPUMIsGuestDebugStateActivePending(pVCpu));
4168 Assert(!CPUMIsHyperDebugStateActivePending(pVCpu));
4169 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
4170 }
4171 else
4172#endif
4173 if (!CPUMIsGuestDebugStateActive(pVCpu))
4174 {
4175 CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
4176 Assert(CPUMIsGuestDebugStateActive(pVCpu));
4177 Assert(!CPUMIsHyperDebugStateActive(pVCpu));
4178 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
4179 }
4180 Assert(!fInterceptDB);
4181 Assert(!fInterceptMovDRx);
4182 }
4183 /*
4184 * If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we
4185 * must intercept #DB in order to maintain a correct DR6 guest value.
4186 */
4187#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4188 else if ( !CPUMIsGuestDebugStateActivePending(pVCpu)
4189 && !CPUMIsGuestDebugStateActive(pVCpu))
4190#else
4191 else if (!CPUMIsGuestDebugStateActive(pVCpu))
4192#endif
4193 {
4194 fInterceptMovDRx = true;
4195 fInterceptDB = true;
4196 }
4197
4198 /* Update guest DR7. */
4199 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, pMixedCtx->dr[7]);
4200 AssertRCReturn(rc, rc);
4201
4202 pVCpu->hm.s.fUsingHyperDR7 = false;
4203 }
4204
4205 /*
4206 * Update the exception bitmap regarding intercepting #DB generated by the guest.
4207 */
4208 if (fInterceptDB)
4209 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_DB);
4210 else if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4211 {
4212#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
4213 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_DB);
4214#endif
4215 }
4216 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
4217 AssertRCReturn(rc, rc);
4218
4219 /*
4220 * Update the processor-based VM-execution controls regarding intercepting MOV DRx instructions.
4221 */
4222 if (fInterceptMovDRx)
4223 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
4224 else
4225 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
4226 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
4227 AssertRCReturn(rc, rc);
4228
4229 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_DEBUG);
4230 return VINF_SUCCESS;
4231}
4232
4233
4234#ifdef VBOX_STRICT
4235/**
4236 * Strict function to validate segment registers.
4237 *
4238 * @remarks ASSUMES CR0 is up to date.
4239 */
4240static void hmR0VmxValidateSegmentRegs(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
4241{
4242 /* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers". */
4243 /* NOTE: The reason we check for attribute value 0 and not just the unusable bit here is because hmR0VmxWriteSegmentReg()
4244 * only updates the VMCS' copy of the value with the unusable bit and doesn't change the guest-context value. */
4245 if ( !pVM->hm.s.vmx.fUnrestrictedGuest
4246 && ( !CPUMIsGuestInRealModeEx(pCtx)
4247 && !CPUMIsGuestInV86ModeEx(pCtx)))
4248 {
4249 /* Protected mode checks */
4250 /* CS */
4251 Assert(pCtx->cs.Attr.n.u1Present);
4252 Assert(!(pCtx->cs.Attr.u & 0xf00));
4253 Assert(!(pCtx->cs.Attr.u & 0xfffe0000));
4254 Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff
4255 || !(pCtx->cs.Attr.n.u1Granularity));
4256 Assert( !(pCtx->cs.u32Limit & 0xfff00000)
4257 || (pCtx->cs.Attr.n.u1Granularity));
4258 /* CS cannot be loaded with NULL in protected mode. */
4259 Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS?!? */
4260 if (pCtx->cs.Attr.n.u4Type == 9 || pCtx->cs.Attr.n.u4Type == 11)
4261 Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl);
4262 else if (pCtx->cs.Attr.n.u4Type == 13 || pCtx->cs.Attr.n.u4Type == 15)
4263 Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl);
4264 else
4265 AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl));
4266 /* SS */
4267 Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
4268 Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL));
4269 if ( !(pCtx->cr0 & X86_CR0_PE)
4270 || pCtx->cs.Attr.n.u4Type == 3)
4271 {
4272 Assert(!pCtx->ss.Attr.n.u2Dpl);
4273 }
4274 if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
4275 {
4276 Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
4277 Assert(pCtx->ss.Attr.n.u4Type == 3 || pCtx->ss.Attr.n.u4Type == 7);
4278 Assert(pCtx->ss.Attr.n.u1Present);
4279 Assert(!(pCtx->ss.Attr.u & 0xf00));
4280 Assert(!(pCtx->ss.Attr.u & 0xfffe0000));
4281 Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff
4282 || !(pCtx->ss.Attr.n.u1Granularity));
4283 Assert( !(pCtx->ss.u32Limit & 0xfff00000)
4284 || (pCtx->ss.Attr.n.u1Granularity));
4285 }
4286 /* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxWriteSegmentReg(). */
4287 if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
4288 {
4289 Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4290 Assert(pCtx->ds.Attr.n.u1Present);
4291 Assert(pCtx->ds.Attr.n.u4Type > 11 || pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL));
4292 Assert(!(pCtx->ds.Attr.u & 0xf00));
4293 Assert(!(pCtx->ds.Attr.u & 0xfffe0000));
4294 Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff
4295 || !(pCtx->ds.Attr.n.u1Granularity));
4296 Assert( !(pCtx->ds.u32Limit & 0xfff00000)
4297 || (pCtx->ds.Attr.n.u1Granularity));
4298 Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4299 || (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ));
4300 }
4301 if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
4302 {
4303 Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4304 Assert(pCtx->es.Attr.n.u1Present);
4305 Assert(pCtx->es.Attr.n.u4Type > 11 || pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL));
4306 Assert(!(pCtx->es.Attr.u & 0xf00));
4307 Assert(!(pCtx->es.Attr.u & 0xfffe0000));
4308 Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff
4309 || !(pCtx->es.Attr.n.u1Granularity));
4310 Assert( !(pCtx->es.u32Limit & 0xfff00000)
4311 || (pCtx->es.Attr.n.u1Granularity));
4312 Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4313 || (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ));
4314 }
4315 if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
4316 {
4317 Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4318 Assert(pCtx->fs.Attr.n.u1Present);
4319 Assert(pCtx->fs.Attr.n.u4Type > 11 || pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL));
4320 Assert(!(pCtx->fs.Attr.u & 0xf00));
4321 Assert(!(pCtx->fs.Attr.u & 0xfffe0000));
4322 Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff
4323 || !(pCtx->fs.Attr.n.u1Granularity));
4324 Assert( !(pCtx->fs.u32Limit & 0xfff00000)
4325 || (pCtx->fs.Attr.n.u1Granularity));
4326 Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4327 || (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ));
4328 }
4329 if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
4330 {
4331 Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4332 Assert(pCtx->gs.Attr.n.u1Present);
4333 Assert(pCtx->gs.Attr.n.u4Type > 11 || pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL));
4334 Assert(!(pCtx->gs.Attr.u & 0xf00));
4335 Assert(!(pCtx->gs.Attr.u & 0xfffe0000));
4336 Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff
4337 || !(pCtx->gs.Attr.n.u1Granularity));
4338 Assert( !(pCtx->gs.u32Limit & 0xfff00000)
4339 || (pCtx->gs.Attr.n.u1Granularity));
4340 Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4341 || (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ));
4342 }
4343 /* 64-bit capable CPUs. */
4344# if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4345 if (HMVMX_IS_64BIT_HOST_MODE())
4346 {
4347 Assert(!(pCtx->cs.u64Base >> 32));
4348 Assert(!pCtx->ss.Attr.u || !(pCtx->ss.u64Base >> 32));
4349 Assert(!pCtx->ds.Attr.u || !(pCtx->ds.u64Base >> 32));
4350 Assert(!pCtx->es.Attr.u || !(pCtx->es.u64Base >> 32));
4351 }
4352# endif
4353 }
4354 else if ( CPUMIsGuestInV86ModeEx(pCtx)
4355 || ( CPUMIsGuestInRealModeEx(pCtx)
4356 && !pVM->hm.s.vmx.fUnrestrictedGuest))
4357 {
4358 /* Real and v86 mode checks. */
4359 /* hmR0VmxWriteSegmentReg() writes the modified in VMCS. We want what we're feeding to VT-x. */
4360 uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
4361 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4362 {
4363 u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3; u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3;
4364 }
4365 else
4366 {
4367 u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u;
4368 u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
4369 }
4370
4371 /* CS */
4372 AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel));
4373 Assert(pCtx->cs.u32Limit == 0xffff);
4374 Assert(u32CSAttr == 0xf3);
4375 /* SS */
4376 Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4);
4377 Assert(pCtx->ss.u32Limit == 0xffff);
4378 Assert(u32SSAttr == 0xf3);
4379 /* DS */
4380 Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4);
4381 Assert(pCtx->ds.u32Limit == 0xffff);
4382 Assert(u32DSAttr == 0xf3);
4383 /* ES */
4384 Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4);
4385 Assert(pCtx->es.u32Limit == 0xffff);
4386 Assert(u32ESAttr == 0xf3);
4387 /* FS */
4388 Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4);
4389 Assert(pCtx->fs.u32Limit == 0xffff);
4390 Assert(u32FSAttr == 0xf3);
4391 /* GS */
4392 Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4);
4393 Assert(pCtx->gs.u32Limit == 0xffff);
4394 Assert(u32GSAttr == 0xf3);
4395 /* 64-bit capable CPUs. */
4396# if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4397 if (HMVMX_IS_64BIT_HOST_MODE())
4398 {
4399 Assert(!(pCtx->cs.u64Base >> 32));
4400 Assert(!u32SSAttr || !(pCtx->ss.u64Base >> 32));
4401 Assert(!u32DSAttr || !(pCtx->ds.u64Base >> 32));
4402 Assert(!u32ESAttr || !(pCtx->es.u64Base >> 32));
4403 }
4404# endif
4405 }
4406}
4407#endif /* VBOX_STRICT */
4408
4409
4410/**
4411 * Writes a guest segment register into the guest-state area in the VMCS.
4412 *
4413 * @returns VBox status code.
4414 * @param pVCpu Pointer to the VMCPU.
4415 * @param idxSel Index of the selector in the VMCS.
4416 * @param idxLimit Index of the segment limit in the VMCS.
4417 * @param idxBase Index of the segment base in the VMCS.
4418 * @param idxAccess Index of the access rights of the segment in the VMCS.
4419 * @param pSelReg Pointer to the segment selector.
4420 *
4421 * @remarks No-long-jump zone!!!
4422 */
4423static int hmR0VmxWriteSegmentReg(PVMCPU pVCpu, uint32_t idxSel, uint32_t idxLimit, uint32_t idxBase,
4424 uint32_t idxAccess, PCPUMSELREG pSelReg)
4425{
4426 int rc = VMXWriteVmcs32(idxSel, pSelReg->Sel); /* 16-bit guest selector field. */
4427 AssertRCReturn(rc, rc);
4428 rc = VMXWriteVmcs32(idxLimit, pSelReg->u32Limit); /* 32-bit guest segment limit field. */
4429 AssertRCReturn(rc, rc);
4430 rc = VMXWriteVmcsGstN(idxBase, pSelReg->u64Base); /* Natural width guest segment base field.*/
4431 AssertRCReturn(rc, rc);
4432
4433 uint32_t u32Access = pSelReg->Attr.u;
4434 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4435 {
4436 /* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */
4437 u32Access = 0xf3;
4438 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
4439 Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
4440 }
4441 else
4442 {
4443 /*
4444 * The way to differentiate between whether this is really a null selector or was just a selector loaded with 0 in
4445 * real-mode is using the segment attributes. A selector loaded in real-mode with the value 0 is valid and usable in
4446 * protected-mode and we should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures NULL selectors
4447 * loaded in protected-mode have their attribute as 0.
4448 */
4449 if (!u32Access)
4450 u32Access = X86DESCATTR_UNUSABLE;
4451 }
4452
4453 /* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */
4454 AssertMsg((u32Access & X86DESCATTR_UNUSABLE) || (u32Access & X86_SEL_TYPE_ACCESSED),
4455 ("Access bit not set for usable segment. idx=%#x sel=%#x attr %#x\n", idxBase, pSelReg, pSelReg->Attr.u));
4456
4457 rc = VMXWriteVmcs32(idxAccess, u32Access); /* 32-bit guest segment access-rights field. */
4458 AssertRCReturn(rc, rc);
4459 return rc;
4460}
4461
4462
4463/**
4464 * Loads the guest segment registers, GDTR, IDTR, LDTR, (TR, FS and GS bases)
4465 * into the guest-state area in the VMCS.
4466 *
4467 * @returns VBox status code.
4468 * @param pVM Pointer to the VM.
4469 * @param pVCPU Pointer to the VMCPU.
4470 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4471 * out-of-sync. Make sure to update the required fields
4472 * before using them.
4473 *
4474 * @remarks ASSUMES pMixedCtx->cr0 is up to date (strict builds validation).
4475 * @remarks No-long-jump zone!!!
4476 */
4477static int hmR0VmxLoadGuestSegmentRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4478{
4479 int rc = VERR_INTERNAL_ERROR_5;
4480 PVM pVM = pVCpu->CTX_SUFF(pVM);
4481
4482 /*
4483 * Guest Segment registers: CS, SS, DS, ES, FS, GS.
4484 */
4485 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS))
4486 {
4487 /* Save the segment attributes for real-on-v86 mode hack, so we can restore them on VM-exit. */
4488 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4489 {
4490 pVCpu->hm.s.vmx.RealMode.AttrCS.u = pMixedCtx->cs.Attr.u;
4491 pVCpu->hm.s.vmx.RealMode.AttrSS.u = pMixedCtx->ss.Attr.u;
4492 pVCpu->hm.s.vmx.RealMode.AttrDS.u = pMixedCtx->ds.Attr.u;
4493 pVCpu->hm.s.vmx.RealMode.AttrES.u = pMixedCtx->es.Attr.u;
4494 pVCpu->hm.s.vmx.RealMode.AttrFS.u = pMixedCtx->fs.Attr.u;
4495 pVCpu->hm.s.vmx.RealMode.AttrGS.u = pMixedCtx->gs.Attr.u;
4496 }
4497
4498#ifdef VBOX_WITH_REM
4499 if (!pVM->hm.s.vmx.fUnrestrictedGuest)
4500 {
4501 Assert(pVM->hm.s.vmx.pRealModeTSS);
4502 AssertCompile(PGMMODE_REAL < PGMMODE_PROTECTED);
4503 if ( pVCpu->hm.s.vmx.fWasInRealMode
4504 && PGMGetGuestMode(pVCpu) >= PGMMODE_PROTECTED)
4505 {
4506 /* Signal that the recompiler must flush its code-cache as the guest -may- rewrite code it will later execute
4507 in real-mode (e.g. OpenBSD 4.0) */
4508 REMFlushTBs(pVM);
4509 Log4(("Load[%RU32]: Switch to protected mode detected!\n", pVCpu->idCpu));
4510 pVCpu->hm.s.vmx.fWasInRealMode = false;
4511 }
4512 }
4513#endif
4514 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_CS, VMX_VMCS32_GUEST_CS_LIMIT, VMX_VMCS_GUEST_CS_BASE,
4515 VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS, &pMixedCtx->cs);
4516 AssertRCReturn(rc, rc);
4517 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_SS, VMX_VMCS32_GUEST_SS_LIMIT, VMX_VMCS_GUEST_SS_BASE,
4518 VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS, &pMixedCtx->ss);
4519 AssertRCReturn(rc, rc);
4520 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_DS, VMX_VMCS32_GUEST_DS_LIMIT, VMX_VMCS_GUEST_DS_BASE,
4521 VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS, &pMixedCtx->ds);
4522 AssertRCReturn(rc, rc);
4523 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_ES, VMX_VMCS32_GUEST_ES_LIMIT, VMX_VMCS_GUEST_ES_BASE,
4524 VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS, &pMixedCtx->es);
4525 AssertRCReturn(rc, rc);
4526 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_FS, VMX_VMCS32_GUEST_FS_LIMIT, VMX_VMCS_GUEST_FS_BASE,
4527 VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS, &pMixedCtx->fs);
4528 AssertRCReturn(rc, rc);
4529 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_GS, VMX_VMCS32_GUEST_GS_LIMIT, VMX_VMCS_GUEST_GS_BASE,
4530 VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS, &pMixedCtx->gs);
4531 AssertRCReturn(rc, rc);
4532
4533#ifdef VBOX_STRICT
4534 /* Validate. */
4535 hmR0VmxValidateSegmentRegs(pVM, pVCpu, pMixedCtx);
4536#endif
4537
4538 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS);
4539 Log4(("Load[%RU32]: CS=%#RX16 Base=%#RX64 Limit=%#RX32 Attr=%#RX32\n", pVCpu->idCpu, pMixedCtx->cs.Sel,
4540 pMixedCtx->cs.u64Base, pMixedCtx->cs.u32Limit, pMixedCtx->cs.Attr.u));
4541 }
4542
4543 /*
4544 * Guest TR.
4545 */
4546 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_TR))
4547 {
4548 /*
4549 * Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is achieved
4550 * using the interrupt redirection bitmap (all bits cleared to let the guest handle INT-n's) in the TSS.
4551 * See hmR3InitFinalizeR0() to see how pRealModeTSS is setup.
4552 */
4553 uint16_t u16Sel = 0;
4554 uint32_t u32Limit = 0;
4555 uint64_t u64Base = 0;
4556 uint32_t u32AccessRights = 0;
4557
4558 if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4559 {
4560 u16Sel = pMixedCtx->tr.Sel;
4561 u32Limit = pMixedCtx->tr.u32Limit;
4562 u64Base = pMixedCtx->tr.u64Base;
4563 u32AccessRights = pMixedCtx->tr.Attr.u;
4564 }
4565 else
4566 {
4567 Assert(pVM->hm.s.vmx.pRealModeTSS);
4568 Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMR3CanExecuteGuest() -XXX- what about inner loop changes? */
4569
4570 /* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */
4571 RTGCPHYS GCPhys;
4572 rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
4573 AssertRCReturn(rc, rc);
4574
4575 X86DESCATTR DescAttr;
4576 DescAttr.u = 0;
4577 DescAttr.n.u1Present = 1;
4578 DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
4579
4580 u16Sel = 0;
4581 u32Limit = HM_VTX_TSS_SIZE;
4582 u64Base = GCPhys; /* in real-mode phys = virt. */
4583 u32AccessRights = DescAttr.u;
4584 }
4585
4586 /* Validate. */
4587 Assert(!(u16Sel & RT_BIT(2)));
4588 AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY
4589 || (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights));
4590 AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights));
4591 Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/
4592 Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/
4593 Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */
4594 Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */
4595 Assert( (u32Limit & 0xfff) == 0xfff
4596 || !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */
4597 Assert( !(pMixedCtx->tr.u32Limit & 0xfff00000)
4598 || (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */
4599
4600 rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_TR, u16Sel); AssertRCReturn(rc, rc);
4601 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit); AssertRCReturn(rc, rc);
4602 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_TR_BASE, u64Base); AssertRCReturn(rc, rc);
4603 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights); AssertRCReturn(rc, rc);
4604
4605 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_TR);
4606 Log4(("Load[%RU32]: VMX_VMCS_GUEST_TR_BASE=%#RX64\n", pVCpu->idCpu, u64Base));
4607 }
4608
4609 /*
4610 * Guest GDTR.
4611 */
4612 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_GDTR))
4613 {
4614 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pMixedCtx->gdtr.cbGdt); AssertRCReturn(rc, rc);
4615 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, pMixedCtx->gdtr.pGdt); AssertRCReturn(rc, rc);
4616
4617 /* Validate. */
4618 Assert(!(pMixedCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */
4619
4620 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_GDTR);
4621 Log4(("Load[%RU32]: VMX_VMCS_GUEST_GDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->gdtr.pGdt));
4622 }
4623
4624 /*
4625 * Guest LDTR.
4626 */
4627 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LDTR))
4628 {
4629 /* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */
4630 uint32_t u32Access = 0;
4631 if (!pMixedCtx->ldtr.Attr.u)
4632 u32Access = X86DESCATTR_UNUSABLE;
4633 else
4634 u32Access = pMixedCtx->ldtr.Attr.u;
4635
4636 rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_FIELD_LDTR, pMixedCtx->ldtr.Sel); AssertRCReturn(rc, rc);
4637 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pMixedCtx->ldtr.u32Limit); AssertRCReturn(rc, rc);
4638 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_LDTR_BASE, pMixedCtx->ldtr.u64Base); AssertRCReturn(rc, rc);
4639 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access); AssertRCReturn(rc, rc);
4640
4641 /* Validate. */
4642 if (!(u32Access & X86DESCATTR_UNUSABLE))
4643 {
4644 Assert(!(pMixedCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */
4645 Assert(pMixedCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */
4646 Assert(!pMixedCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */
4647 Assert(pMixedCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */
4648 Assert(!pMixedCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */
4649 Assert(!(pMixedCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */
4650 Assert( (pMixedCtx->ldtr.u32Limit & 0xfff) == 0xfff
4651 || !pMixedCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */
4652 Assert( !(pMixedCtx->ldtr.u32Limit & 0xfff00000)
4653 || pMixedCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */
4654 }
4655
4656 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_LDTR);
4657 Log4(("Load[%RU32]: VMX_VMCS_GUEST_LDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->ldtr.u64Base));
4658 }
4659
4660 /*
4661 * Guest IDTR.
4662 */
4663 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_IDTR))
4664 {
4665 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pMixedCtx->idtr.cbIdt); AssertRCReturn(rc, rc);
4666 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, pMixedCtx->idtr.pIdt); AssertRCReturn(rc, rc);
4667
4668 /* Validate. */
4669 Assert(!(pMixedCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */
4670
4671 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_IDTR);
4672 Log4(("Load[%RU32]: VMX_VMCS_GUEST_IDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->idtr.pIdt));
4673 }
4674
4675 return VINF_SUCCESS;
4676}
4677
4678
4679/**
4680 * Loads certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store
4681 * areas. These MSRs will automatically be loaded to the host CPU on every
4682 * successful VM entry and stored from the host CPU on every successful VM-exit.
4683 *
4684 * This also creates/updates MSR slots for the host MSRs. The actual host
4685 * MSR values are -not- updated here for performance reasons. See
4686 * hmR0VmxSaveHostMsrs().
4687 *
4688 * Also loads the sysenter MSRs into the guest-state area in the VMCS.
4689 *
4690 * @returns VBox status code.
4691 * @param pVCpu Pointer to the VMCPU.
4692 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4693 * out-of-sync. Make sure to update the required fields
4694 * before using them.
4695 *
4696 * @remarks No-long-jump zone!!!
4697 */
4698static int hmR0VmxLoadGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4699{
4700 AssertPtr(pVCpu);
4701 AssertPtr(pVCpu->hm.s.vmx.pvGuestMsr);
4702
4703 /*
4704 * MSRs that we use the auto-load/store MSR area in the VMCS.
4705 */
4706 PVM pVM = pVCpu->CTX_SUFF(pVM);
4707 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS))
4708 {
4709 /* For 64-bit hosts, we load/restore them lazily, see hmR0VmxLazyLoadGuestMsrs(). */
4710#if HC_ARCH_BITS == 32 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4711 if (pVM->hm.s.fAllow64BitGuests)
4712 {
4713 hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_LSTAR, pMixedCtx->msrLSTAR, false /* fUpdateHostMsr */);
4714 hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K6_STAR, pMixedCtx->msrSTAR, false /* fUpdateHostMsr */);
4715 hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_SF_MASK, pMixedCtx->msrSFMASK, false /* fUpdateHostMsr */);
4716 hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_KERNEL_GS_BASE, pMixedCtx->msrKERNELGSBASE, false /* fUpdateHostMsr */);
4717# ifdef DEBUG
4718 PVMXAUTOMSR pMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
4719 for (uint32_t i = 0; i < pVCpu->hm.s.vmx.cMsrs; i++, pMsr++)
4720 {
4721 Log4(("Load[%RU32]: MSR[%RU32]: u32Msr=%#RX32 u64Value=%#RX64\n", pVCpu->idCpu, i, pMsr->u32Msr,
4722 pMsr->u64Value));
4723 }
4724# endif
4725 }
4726#endif
4727 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
4728 }
4729
4730 /*
4731 * Guest Sysenter MSRs.
4732 * These flags are only set when MSR-bitmaps are not supported by the CPU and we cause
4733 * VM-exits on WRMSRs for these MSRs.
4734 */
4735 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR))
4736 {
4737 int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pMixedCtx->SysEnter.cs); AssertRCReturn(rc, rc);
4738 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR);
4739 }
4740
4741 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR))
4742 {
4743 int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, pMixedCtx->SysEnter.eip); AssertRCReturn(rc, rc);
4744 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
4745 }
4746
4747 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR))
4748 {
4749 int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, pMixedCtx->SysEnter.esp); AssertRCReturn(rc, rc);
4750 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
4751 }
4752
4753 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_EFER_MSR))
4754 {
4755 if (hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
4756 {
4757 /*
4758 * If the CPU supports VMCS controls for swapping EFER, use it. Otherwise, we have no option
4759 * but to use the auto-load store MSR area in the VMCS for swapping EFER. See @bugref{7368}.
4760 */
4761 if (pVM->hm.s.vmx.fSupportsVmcsEfer)
4762 {
4763 int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_EFER_FULL, pMixedCtx->msrEFER);
4764 AssertRCReturn(rc,rc);
4765 Log4(("Load[%RU32]: VMX_VMCS64_GUEST_EFER_FULL=%#RX64\n", pVCpu->idCpu, pMixedCtx->msrEFER));
4766 }
4767 else
4768 {
4769 hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K6_EFER, pMixedCtx->msrEFER, false /* fUpdateHostMsr */);
4770 /* We need to intercept reads too, see @bugref{7386} comment #16. */
4771 hmR0VmxSetMsrPermission(pVCpu, MSR_K6_EFER, VMXMSREXIT_INTERCEPT_READ, VMXMSREXIT_INTERCEPT_WRITE);
4772 Log4(("Load[%RU32]: MSR[--]: u32Msr=%#RX32 u64Value=%#RX64 cMsrs=%u\n", pVCpu->idCpu, MSR_K6_EFER,
4773 pMixedCtx->msrEFER, pVCpu->hm.s.vmx.cMsrs));
4774 }
4775 }
4776 else if (!pVM->hm.s.vmx.fSupportsVmcsEfer)
4777 hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, MSR_K6_EFER);
4778 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_EFER_MSR);
4779 }
4780
4781 return VINF_SUCCESS;
4782}
4783
4784
4785/**
4786 * Loads the guest activity state into the guest-state area in the VMCS.
4787 *
4788 * @returns VBox status code.
4789 * @param pVCpu Pointer to the VMCPU.
4790 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4791 * out-of-sync. Make sure to update the required fields
4792 * before using them.
4793 *
4794 * @remarks No-long-jump zone!!!
4795 */
4796static int hmR0VmxLoadGuestActivityState(PVMCPU pVCpu, PCPUMCTX pCtx)
4797{
4798 NOREF(pCtx);
4799 /** @todo See if we can make use of other states, e.g.
4800 * VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN or HLT. */
4801 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_ACTIVITY_STATE))
4802 {
4803 int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, VMX_VMCS_GUEST_ACTIVITY_ACTIVE);
4804 AssertRCReturn(rc, rc);
4805
4806 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_ACTIVITY_STATE);
4807 }
4808 return VINF_SUCCESS;
4809}
4810
4811
4812/**
4813 * Sets up the appropriate function to run guest code.
4814 *
4815 * @returns VBox status code.
4816 * @param pVCpu Pointer to the VMCPU.
4817 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4818 * out-of-sync. Make sure to update the required fields
4819 * before using them.
4820 *
4821 * @remarks No-long-jump zone!!!
4822 */
4823static int hmR0VmxSetupVMRunHandler(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4824{
4825 if (CPUMIsGuestInLongModeEx(pMixedCtx))
4826 {
4827#ifndef VBOX_ENABLE_64_BITS_GUESTS
4828 return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
4829#endif
4830 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests); /* Guaranteed by hmR3InitFinalizeR0(). */
4831#if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4832 /* 32-bit host. We need to switch to 64-bit before running the 64-bit guest. */
4833 if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0SwitcherStartVM64)
4834 {
4835 if (pVCpu->hm.s.vmx.pfnStartVM != NULL) /* Very first entry would have saved host-state already, ignore it. */
4836 {
4837 /* Currently, all mode changes sends us back to ring-3, so these should be set. See @bugref{6944}. */
4838 AssertMsg(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT
4839 | HM_CHANGED_VMX_EXIT_CTLS
4840 | HM_CHANGED_VMX_ENTRY_CTLS
4841 | HM_CHANGED_GUEST_EFER_MSR), ("flags=%#x\n", HMCPU_CF_VALUE(pVCpu)));
4842 }
4843 pVCpu->hm.s.vmx.pfnStartVM = VMXR0SwitcherStartVM64;
4844 }
4845#else
4846 /* 64-bit host or hybrid host. */
4847 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM64;
4848#endif
4849 }
4850 else
4851 {
4852 /* Guest is not in long mode, use the 32-bit handler. */
4853#if HC_ARCH_BITS == 32 && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
4854 if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0StartVM32)
4855 {
4856 if (pVCpu->hm.s.vmx.pfnStartVM != NULL) /* Very first entry would have saved host-state already, ignore it. */
4857 {
4858 /* Currently, all mode changes sends us back to ring-3, so these should be set. See @bugref{6944}. */
4859 AssertMsg(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT
4860 | HM_CHANGED_VMX_EXIT_CTLS
4861 | HM_CHANGED_VMX_ENTRY_CTLS
4862 | HM_CHANGED_GUEST_EFER_MSR), ("flags=%#x\n", HMCPU_CF_VALUE(pVCpu)));
4863 }
4864 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4865 }
4866#else
4867 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4868#endif
4869 }
4870 Assert(pVCpu->hm.s.vmx.pfnStartVM);
4871 return VINF_SUCCESS;
4872}
4873
4874
4875/**
4876 * Wrapper for running the guest code in VT-x.
4877 *
4878 * @returns VBox strict status code.
4879 * @param pVM Pointer to the VM.
4880 * @param pVCpu Pointer to the VMCPU.
4881 * @param pCtx Pointer to the guest-CPU context.
4882 *
4883 * @remarks No-long-jump zone!!!
4884 */
4885DECLINLINE(int) hmR0VmxRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
4886{
4887 /*
4888 * 64-bit Windows uses XMM registers in the kernel as the Microsoft compiler expresses floating-point operations
4889 * using SSE instructions. Some XMM registers (XMM6-XMM15) are callee-saved and thus the need for this XMM wrapper.
4890 * Refer MSDN docs. "Configuring Programs for 64-bit / x64 Software Conventions / Register Usage" for details.
4891 */
4892 bool const fResumeVM = RT_BOOL(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_LAUNCHED);
4893 /** @todo Add stats for resume vs launch. */
4894#ifdef VBOX_WITH_KERNEL_USING_XMM
4895 return HMR0VMXStartVMWrapXMM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu, pVCpu->hm.s.vmx.pfnStartVM);
4896#else
4897 return pVCpu->hm.s.vmx.pfnStartVM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu);
4898#endif
4899}
4900
4901
4902/**
4903 * Reports world-switch error and dumps some useful debug info.
4904 *
4905 * @param pVM Pointer to the VM.
4906 * @param pVCpu Pointer to the VMCPU.
4907 * @param rcVMRun The return code from VMLAUNCH/VMRESUME.
4908 * @param pCtx Pointer to the guest-CPU context.
4909 * @param pVmxTransient Pointer to the VMX transient structure (only
4910 * exitReason updated).
4911 */
4912static void hmR0VmxReportWorldSwitchError(PVM pVM, PVMCPU pVCpu, int rcVMRun, PCPUMCTX pCtx, PVMXTRANSIENT pVmxTransient)
4913{
4914 Assert(pVM);
4915 Assert(pVCpu);
4916 Assert(pCtx);
4917 Assert(pVmxTransient);
4918 HMVMX_ASSERT_PREEMPT_SAFE();
4919
4920 Log4(("VM-entry failure: %Rrc\n", rcVMRun));
4921 switch (rcVMRun)
4922 {
4923 case VERR_VMX_INVALID_VMXON_PTR:
4924 AssertFailed();
4925 break;
4926 case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */
4927 case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */
4928 {
4929 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason);
4930 rc |= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
4931 rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
4932 AssertRC(rc);
4933
4934 pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
4935 /* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted().
4936 Cannot do it here as we may have been long preempted. */
4937
4938#ifdef VBOX_STRICT
4939 Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason,
4940 pVmxTransient->uExitReason));
4941 Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQualification));
4942 Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError));
4943 if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX)
4944 Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError]));
4945 else
4946 Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX));
4947 Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu));
4948 Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu));
4949
4950 /* VMX control bits. */
4951 uint32_t u32Val;
4952 uint64_t u64Val;
4953 HMVMXHCUINTREG uHCReg;
4954 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val); AssertRC(rc);
4955 Log4(("VMX_VMCS32_CTRL_PIN_EXEC %#RX32\n", u32Val));
4956 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val); AssertRC(rc);
4957 Log4(("VMX_VMCS32_CTRL_PROC_EXEC %#RX32\n", u32Val));
4958 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val); AssertRC(rc);
4959 Log4(("VMX_VMCS32_CTRL_PROC_EXEC2 %#RX32\n", u32Val));
4960 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val); AssertRC(rc);
4961 Log4(("VMX_VMCS32_CTRL_ENTRY %#RX32\n", u32Val));
4962 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val); AssertRC(rc);
4963 Log4(("VMX_VMCS32_CTRL_EXIT %#RX32\n", u32Val));
4964 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, &u32Val); AssertRC(rc);
4965 Log4(("VMX_VMCS32_CTRL_CR3_TARGET_COUNT %#RX32\n", u32Val));
4966 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32Val); AssertRC(rc);
4967 Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", u32Val));
4968 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &u32Val); AssertRC(rc);
4969 Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", u32Val));
4970 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &u32Val); AssertRC(rc);
4971 Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %u\n", u32Val));
4972 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, &u32Val); AssertRC(rc);
4973 Log4(("VMX_VMCS32_CTRL_TPR_THRESHOLD %u\n", u32Val));
4974 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &u32Val); AssertRC(rc);
4975 Log4(("VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT %u (guest MSRs)\n", u32Val));
4976 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &u32Val); AssertRC(rc);
4977 Log4(("VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT %u (host MSRs)\n", u32Val));
4978 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &u32Val); AssertRC(rc);
4979 Log4(("VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT %u (guest MSRs)\n", u32Val));
4980 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val); AssertRC(rc);
4981 Log4(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP %#RX32\n", u32Val));
4982 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, &u32Val); AssertRC(rc);
4983 Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK %#RX32\n", u32Val));
4984 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, &u32Val); AssertRC(rc);
4985 Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH %#RX32\n", u32Val));
4986 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc);
4987 Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg));
4988 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc);
4989 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
4990 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc);
4991 Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg));
4992 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc);
4993 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
4994 rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
4995 Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
4996
4997 /* Guest bits. */
4998 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val); AssertRC(rc);
4999 Log4(("Old Guest Rip %#RX64 New %#RX64\n", pCtx->rip, u64Val));
5000 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val); AssertRC(rc);
5001 Log4(("Old Guest Rsp %#RX64 New %#RX64\n", pCtx->rsp, u64Val));
5002 rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Val); AssertRC(rc);
5003 Log4(("Old Guest Rflags %#RX32 New %#RX32\n", pCtx->eflags.u32, u32Val));
5004 rc = VMXReadVmcs32(VMX_VMCS16_GUEST_FIELD_VPID, &u32Val); AssertRC(rc);
5005 Log4(("VMX_VMCS16_GUEST_FIELD_VPID %u\n", u32Val));
5006
5007 /* Host bits. */
5008 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR0, &uHCReg); AssertRC(rc);
5009 Log4(("Host CR0 %#RHr\n", uHCReg));
5010 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR3, &uHCReg); AssertRC(rc);
5011 Log4(("Host CR3 %#RHr\n", uHCReg));
5012 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR4, &uHCReg); AssertRC(rc);
5013 Log4(("Host CR4 %#RHr\n", uHCReg));
5014
5015 RTGDTR HostGdtr;
5016 PCX86DESCHC pDesc;
5017 ASMGetGDTR(&HostGdtr);
5018 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_CS, &u32Val); AssertRC(rc);
5019 Log4(("Host CS %#08x\n", u32Val));
5020 if (u32Val < HostGdtr.cbGdt)
5021 {
5022 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5023 HMR0DumpDescriptor(pDesc, u32Val, "CS: ");
5024 }
5025
5026 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_DS, &u32Val); AssertRC(rc);
5027 Log4(("Host DS %#08x\n", u32Val));
5028 if (u32Val < HostGdtr.cbGdt)
5029 {
5030 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5031 HMR0DumpDescriptor(pDesc, u32Val, "DS: ");
5032 }
5033
5034 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_ES, &u32Val); AssertRC(rc);
5035 Log4(("Host ES %#08x\n", u32Val));
5036 if (u32Val < HostGdtr.cbGdt)
5037 {
5038 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5039 HMR0DumpDescriptor(pDesc, u32Val, "ES: ");
5040 }
5041
5042 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_FS, &u32Val); AssertRC(rc);
5043 Log4(("Host FS %#08x\n", u32Val));
5044 if (u32Val < HostGdtr.cbGdt)
5045 {
5046 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5047 HMR0DumpDescriptor(pDesc, u32Val, "FS: ");
5048 }
5049
5050 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_GS, &u32Val); AssertRC(rc);
5051 Log4(("Host GS %#08x\n", u32Val));
5052 if (u32Val < HostGdtr.cbGdt)
5053 {
5054 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5055 HMR0DumpDescriptor(pDesc, u32Val, "GS: ");
5056 }
5057
5058 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_SS, &u32Val); AssertRC(rc);
5059 Log4(("Host SS %#08x\n", u32Val));
5060 if (u32Val < HostGdtr.cbGdt)
5061 {
5062 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5063 HMR0DumpDescriptor(pDesc, u32Val, "SS: ");
5064 }
5065
5066 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FIELD_TR, &u32Val); AssertRC(rc);
5067 Log4(("Host TR %#08x\n", u32Val));
5068 if (u32Val < HostGdtr.cbGdt)
5069 {
5070 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5071 HMR0DumpDescriptor(pDesc, u32Val, "TR: ");
5072 }
5073
5074 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_TR_BASE, &uHCReg); AssertRC(rc);
5075 Log4(("Host TR Base %#RHv\n", uHCReg));
5076 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, &uHCReg); AssertRC(rc);
5077 Log4(("Host GDTR Base %#RHv\n", uHCReg));
5078 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, &uHCReg); AssertRC(rc);
5079 Log4(("Host IDTR Base %#RHv\n", uHCReg));
5080 rc = VMXReadVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, &u32Val); AssertRC(rc);
5081 Log4(("Host SYSENTER CS %#08x\n", u32Val));
5082 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_EIP, &uHCReg); AssertRC(rc);
5083 Log4(("Host SYSENTER EIP %#RHv\n", uHCReg));
5084 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_ESP, &uHCReg); AssertRC(rc);
5085 Log4(("Host SYSENTER ESP %#RHv\n", uHCReg));
5086 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RSP, &uHCReg); AssertRC(rc);
5087 Log4(("Host RSP %#RHv\n", uHCReg));
5088 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RIP, &uHCReg); AssertRC(rc);
5089 Log4(("Host RIP %#RHv\n", uHCReg));
5090# if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
5091 if (HMVMX_IS_64BIT_HOST_MODE())
5092 {
5093 Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER)));
5094 Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR)));
5095 Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR)));
5096 Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR)));
5097 Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK)));
5098 Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE)));
5099 }
5100# endif
5101#endif /* VBOX_STRICT */
5102 break;
5103 }
5104
5105 default:
5106 /* Impossible */
5107 AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun));
5108 break;
5109 }
5110 NOREF(pVM); NOREF(pCtx);
5111}
5112
5113
5114#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
5115#ifndef VMX_USE_CACHED_VMCS_ACCESSES
5116# error "VMX_USE_CACHED_VMCS_ACCESSES not defined when it should be!"
5117#endif
5118#ifdef VBOX_STRICT
5119static bool hmR0VmxIsValidWriteField(uint32_t idxField)
5120{
5121 switch (idxField)
5122 {
5123 case VMX_VMCS_GUEST_RIP:
5124 case VMX_VMCS_GUEST_RSP:
5125 case VMX_VMCS_GUEST_SYSENTER_EIP:
5126 case VMX_VMCS_GUEST_SYSENTER_ESP:
5127 case VMX_VMCS_GUEST_GDTR_BASE:
5128 case VMX_VMCS_GUEST_IDTR_BASE:
5129 case VMX_VMCS_GUEST_CS_BASE:
5130 case VMX_VMCS_GUEST_DS_BASE:
5131 case VMX_VMCS_GUEST_ES_BASE:
5132 case VMX_VMCS_GUEST_FS_BASE:
5133 case VMX_VMCS_GUEST_GS_BASE:
5134 case VMX_VMCS_GUEST_SS_BASE:
5135 case VMX_VMCS_GUEST_LDTR_BASE:
5136 case VMX_VMCS_GUEST_TR_BASE:
5137 case VMX_VMCS_GUEST_CR3:
5138 return true;
5139 }
5140 return false;
5141}
5142
5143static bool hmR0VmxIsValidReadField(uint32_t idxField)
5144{
5145 switch (idxField)
5146 {
5147 /* Read-only fields. */
5148 case VMX_VMCS_RO_EXIT_QUALIFICATION:
5149 return true;
5150 }
5151 /* Remaining readable fields should also be writable. */
5152 return hmR0VmxIsValidWriteField(idxField);
5153}
5154#endif /* VBOX_STRICT */
5155
5156
5157/**
5158 * Executes the specified handler in 64-bit mode.
5159 *
5160 * @returns VBox status code.
5161 * @param pVM Pointer to the VM.
5162 * @param pVCpu Pointer to the VMCPU.
5163 * @param pCtx Pointer to the guest CPU context.
5164 * @param enmOp The operation to perform.
5165 * @param cbParam Number of parameters.
5166 * @param paParam Array of 32-bit parameters.
5167 */
5168VMMR0DECL(int) VMXR0Execute64BitsHandler(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, HM64ON32OP enmOp, uint32_t cbParam,
5169 uint32_t *paParam)
5170{
5171 int rc, rc2;
5172 PHMGLOBALCPUINFO pCpu;
5173 RTHCPHYS HCPhysCpuPage;
5174 RTCCUINTREG uOldEflags;
5175
5176 AssertReturn(pVM->hm.s.pfnHost32ToGuest64R0, VERR_HM_NO_32_TO_64_SWITCHER);
5177 Assert(enmOp > HM64ON32OP_INVALID && enmOp < HM64ON32OP_END);
5178 Assert(pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Write.aField));
5179 Assert(pVCpu->hm.s.vmx.VMCSCache.Read.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Read.aField));
5180
5181#ifdef VBOX_STRICT
5182 for (uint32_t i = 0; i < pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries; i++)
5183 Assert(hmR0VmxIsValidWriteField(pVCpu->hm.s.vmx.VMCSCache.Write.aField[i]));
5184
5185 for (uint32_t i = 0; i <pVCpu->hm.s.vmx.VMCSCache.Read.cValidEntries; i++)
5186 Assert(hmR0VmxIsValidReadField(pVCpu->hm.s.vmx.VMCSCache.Read.aField[i]));
5187#endif
5188
5189 /* Disable interrupts. */
5190 uOldEflags = ASMIntDisableFlags();
5191
5192#ifdef VBOX_WITH_VMMR0_DISABLE_LAPIC_NMI
5193 RTCPUID idHostCpu = RTMpCpuId();
5194 CPUMR0SetLApic(pVCpu, idHostCpu);
5195#endif
5196
5197 pCpu = HMR0GetCurrentCpu();
5198 HCPhysCpuPage = RTR0MemObjGetPagePhysAddr(pCpu->hMemObj, 0);
5199
5200 /* Clear VMCS. Marking it inactive, clearing implementation-specific data and writing VMCS data back to memory. */
5201 VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
5202
5203 /* Leave VMX Root Mode. */
5204 VMXDisable();
5205
5206 ASMSetCR4(ASMGetCR4() & ~X86_CR4_VMXE);
5207
5208 CPUMSetHyperESP(pVCpu, VMMGetStackRC(pVCpu));
5209 CPUMSetHyperEIP(pVCpu, enmOp);
5210 for (int i = (int)cbParam - 1; i >= 0; i--)
5211 CPUMPushHyper(pVCpu, paParam[i]);
5212
5213 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatWorldSwitch3264, z);
5214
5215 /* Call the switcher. */
5216 rc = pVM->hm.s.pfnHost32ToGuest64R0(pVM, RT_OFFSETOF(VM, aCpus[pVCpu->idCpu].cpum) - RT_OFFSETOF(VM, cpum));
5217 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatWorldSwitch3264, z);
5218
5219 /** @todo replace with hmR0VmxEnterRootMode() and hmR0VmxLeaveRootMode(). */
5220 /* Make sure the VMX instructions don't cause #UD faults. */
5221 ASMSetCR4(ASMGetCR4() | X86_CR4_VMXE);
5222
5223 /* Re-enter VMX Root Mode */
5224 rc2 = VMXEnable(HCPhysCpuPage);
5225 if (RT_FAILURE(rc2))
5226 {
5227 ASMSetCR4(ASMGetCR4() & ~X86_CR4_VMXE);
5228 ASMSetFlags(uOldEflags);
5229 return rc2;
5230 }
5231
5232 rc2 = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
5233 AssertRC(rc2);
5234 Assert(!(ASMGetFlags() & X86_EFL_IF));
5235 ASMSetFlags(uOldEflags);
5236 return rc;
5237}
5238
5239
5240/**
5241 * Prepares for and executes VMLAUNCH (64-bit guests) for 32-bit hosts
5242 * supporting 64-bit guests.
5243 *
5244 * @returns VBox status code.
5245 * @param fResume Whether to VMLAUNCH or VMRESUME.
5246 * @param pCtx Pointer to the guest-CPU context.
5247 * @param pCache Pointer to the VMCS cache.
5248 * @param pVM Pointer to the VM.
5249 * @param pVCpu Pointer to the VMCPU.
5250 */
5251DECLASM(int) VMXR0SwitcherStartVM64(RTHCUINT fResume, PCPUMCTX pCtx, PVMCSCACHE pCache, PVM pVM, PVMCPU pVCpu)
5252{
5253 uint32_t aParam[6];
5254 PHMGLOBALCPUINFO pCpu = NULL;
5255 RTHCPHYS HCPhysCpuPage = 0;
5256 int rc = VERR_INTERNAL_ERROR_5;
5257
5258 pCpu = HMR0GetCurrentCpu();
5259 HCPhysCpuPage = RTR0MemObjGetPagePhysAddr(pCpu->hMemObj, 0);
5260
5261#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5262 pCache->uPos = 1;
5263 pCache->interPD = PGMGetInterPaeCR3(pVM);
5264 pCache->pSwitcher = (uint64_t)pVM->hm.s.pfnHost32ToGuest64R0;
5265#endif
5266
5267#if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES)
5268 pCache->TestIn.HCPhysCpuPage = 0;
5269 pCache->TestIn.HCPhysVmcs = 0;
5270 pCache->TestIn.pCache = 0;
5271 pCache->TestOut.HCPhysVmcs = 0;
5272 pCache->TestOut.pCache = 0;
5273 pCache->TestOut.pCtx = 0;
5274 pCache->TestOut.eflags = 0;
5275#endif
5276
5277 aParam[0] = (uint32_t)(HCPhysCpuPage); /* Param 1: VMXON physical address - Lo. */
5278 aParam[1] = (uint32_t)(HCPhysCpuPage >> 32); /* Param 1: VMXON physical address - Hi. */
5279 aParam[2] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs); /* Param 2: VMCS physical address - Lo. */
5280 aParam[3] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs >> 32); /* Param 2: VMCS physical address - Hi. */
5281 aParam[4] = VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache);
5282 aParam[5] = 0;
5283
5284#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5285 pCtx->dr[4] = pVM->hm.s.vmx.pScratchPhys + 16 + 8;
5286 *(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) = 1;
5287#endif
5288 rc = VMXR0Execute64BitsHandler(pVM, pVCpu, pCtx, HM64ON32OP_VMXRCStartVM64, 6, &aParam[0]);
5289
5290#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5291 Assert(*(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) == 5);
5292 Assert(pCtx->dr[4] == 10);
5293 *(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) = 0xff;
5294#endif
5295
5296#if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES)
5297 AssertMsg(pCache->TestIn.HCPhysCpuPage == HCPhysCpuPage, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysCpuPage, HCPhysCpuPage));
5298 AssertMsg(pCache->TestIn.HCPhysVmcs == pVCpu->hm.s.vmx.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
5299 pVCpu->hm.s.vmx.HCPhysVmcs));
5300 AssertMsg(pCache->TestIn.HCPhysVmcs == pCache->TestOut.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
5301 pCache->TestOut.HCPhysVmcs));
5302 AssertMsg(pCache->TestIn.pCache == pCache->TestOut.pCache, ("%RGv vs %RGv\n", pCache->TestIn.pCache,
5303 pCache->TestOut.pCache));
5304 AssertMsg(pCache->TestIn.pCache == VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache),
5305 ("%RGv vs %RGv\n", pCache->TestIn.pCache, VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache)));
5306 AssertMsg(pCache->TestIn.pCtx == pCache->TestOut.pCtx, ("%RGv vs %RGv\n", pCache->TestIn.pCtx,
5307 pCache->TestOut.pCtx));
5308 Assert(!(pCache->TestOut.eflags & X86_EFL_IF));
5309#endif
5310 return rc;
5311}
5312
5313
5314/**
5315 * Initialize the VMCS-Read cache. The VMCS cache is used for 32-bit hosts
5316 * running 64-bit guests (except 32-bit Darwin which runs with 64-bit paging in
5317 * 32-bit mode) for 64-bit fields that cannot be accessed in 32-bit mode. Some
5318 * 64-bit fields -can- be accessed (those that have a 32-bit FULL & HIGH part).
5319 *
5320 * @returns VBox status code.
5321 * @param pVM Pointer to the VM.
5322 * @param pVCpu Pointer to the VMCPU.
5323 */
5324static int hmR0VmxInitVmcsReadCache(PVM pVM, PVMCPU pVCpu)
5325{
5326#define VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, idxField) \
5327{ \
5328 Assert(pCache->Read.aField[idxField##_CACHE_IDX] == 0); \
5329 pCache->Read.aField[idxField##_CACHE_IDX] = idxField; \
5330 pCache->Read.aFieldVal[idxField##_CACHE_IDX] = 0; \
5331 ++cReadFields; \
5332}
5333
5334 AssertPtr(pVM);
5335 AssertPtr(pVCpu);
5336 PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache;
5337 uint32_t cReadFields = 0;
5338
5339 /*
5340 * Don't remove the #if 0'd fields in this code. They're listed here for consistency
5341 * and serve to indicate exceptions to the rules.
5342 */
5343
5344 /* Guest-natural selector base fields. */
5345#if 0
5346 /* These are 32-bit in practice. See Intel spec. 2.5 "Control Registers". */
5347 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR0);
5348 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR4);
5349#endif
5350 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_ES_BASE);
5351 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CS_BASE);
5352 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SS_BASE);
5353 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_DS_BASE);
5354 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_FS_BASE);
5355 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GS_BASE);
5356 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_LDTR_BASE);
5357 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_TR_BASE);
5358 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GDTR_BASE);
5359 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_IDTR_BASE);
5360 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RSP);
5361 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RIP);
5362#if 0
5363 /* Unused natural width guest-state fields. */
5364 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS);
5365 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3); /* Handled in Nested Paging case */
5366#endif
5367 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_ESP);
5368 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_EIP);
5369
5370 /* 64-bit guest-state fields; unused as we use two 32-bit VMREADs for these 64-bit fields (using "FULL" and "HIGH" fields). */
5371#if 0
5372 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL);
5373 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_DEBUGCTL_FULL);
5374 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PAT_FULL);
5375 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_EFER_FULL);
5376 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL);
5377 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE0_FULL);
5378 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE1_FULL);
5379 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE2_FULL);
5380 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE3_FULL);
5381#endif
5382
5383 /* Natural width guest-state fields. */
5384 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_QUALIFICATION);
5385#if 0
5386 /* Currently unused field. */
5387 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_GUEST_LINEAR_ADDR);
5388#endif
5389
5390 if (pVM->hm.s.fNestedPaging)
5391 {
5392 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3);
5393 AssertMsg(cReadFields == VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields,
5394 VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX));
5395 pCache->Read.cValidEntries = VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX;
5396 }
5397 else
5398 {
5399 AssertMsg(cReadFields == VMX_VMCS_MAX_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields, VMX_VMCS_MAX_CACHE_IDX));
5400 pCache->Read.cValidEntries = VMX_VMCS_MAX_CACHE_IDX;
5401 }
5402
5403#undef VMXLOCAL_INIT_READ_CACHE_FIELD
5404 return VINF_SUCCESS;
5405}
5406
5407
5408/**
5409 * Writes a field into the VMCS. This can either directly invoke a VMWRITE or
5410 * queue up the VMWRITE by using the VMCS write cache (on 32-bit hosts, except
5411 * darwin, running 64-bit guests).
5412 *
5413 * @returns VBox status code.
5414 * @param pVCpu Pointer to the VMCPU.
5415 * @param idxField The VMCS field encoding.
5416 * @param u64Val 16, 32 or 64-bit value.
5417 */
5418VMMR0DECL(int) VMXWriteVmcs64Ex(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val)
5419{
5420 int rc;
5421 switch (idxField)
5422 {
5423 /*
5424 * These fields consists of a "FULL" and a "HIGH" part which can be written to individually.
5425 */
5426 /* 64-bit Control fields. */
5427 case VMX_VMCS64_CTRL_IO_BITMAP_A_FULL:
5428 case VMX_VMCS64_CTRL_IO_BITMAP_B_FULL:
5429 case VMX_VMCS64_CTRL_MSR_BITMAP_FULL:
5430 case VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL:
5431 case VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL:
5432 case VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL:
5433 case VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL:
5434 case VMX_VMCS64_CTRL_TSC_OFFSET_FULL:
5435 case VMX_VMCS64_CTRL_VAPIC_PAGEADDR_FULL:
5436 case VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL:
5437 case VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL:
5438 case VMX_VMCS64_CTRL_EPTP_FULL:
5439 case VMX_VMCS64_CTRL_EPTP_LIST_FULL:
5440 /* 64-bit Guest-state fields. */
5441 case VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL:
5442 case VMX_VMCS64_GUEST_DEBUGCTL_FULL:
5443 case VMX_VMCS64_GUEST_PAT_FULL:
5444 case VMX_VMCS64_GUEST_EFER_FULL:
5445 case VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL:
5446 case VMX_VMCS64_GUEST_PDPTE0_FULL:
5447 case VMX_VMCS64_GUEST_PDPTE1_FULL:
5448 case VMX_VMCS64_GUEST_PDPTE2_FULL:
5449 case VMX_VMCS64_GUEST_PDPTE3_FULL:
5450 /* 64-bit Host-state fields. */
5451 case VMX_VMCS64_HOST_FIELD_PAT_FULL:
5452 case VMX_VMCS64_HOST_FIELD_EFER_FULL:
5453 case VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL:
5454 {
5455 rc = VMXWriteVmcs32(idxField, u64Val);
5456 rc |= VMXWriteVmcs32(idxField + 1, (uint32_t)(u64Val >> 32));
5457 break;
5458 }
5459
5460 /*
5461 * These fields do not have high and low parts. Queue up the VMWRITE by using the VMCS write-cache (for 64-bit
5462 * values). When we switch the host to 64-bit mode for running 64-bit guests, these VMWRITEs get executed then.
5463 */
5464 /* Natural-width Guest-state fields. */
5465 case VMX_VMCS_GUEST_CR3:
5466 case VMX_VMCS_GUEST_ES_BASE:
5467 case VMX_VMCS_GUEST_CS_BASE:
5468 case VMX_VMCS_GUEST_SS_BASE:
5469 case VMX_VMCS_GUEST_DS_BASE:
5470 case VMX_VMCS_GUEST_FS_BASE:
5471 case VMX_VMCS_GUEST_GS_BASE:
5472 case VMX_VMCS_GUEST_LDTR_BASE:
5473 case VMX_VMCS_GUEST_TR_BASE:
5474 case VMX_VMCS_GUEST_GDTR_BASE:
5475 case VMX_VMCS_GUEST_IDTR_BASE:
5476 case VMX_VMCS_GUEST_RSP:
5477 case VMX_VMCS_GUEST_RIP:
5478 case VMX_VMCS_GUEST_SYSENTER_ESP:
5479 case VMX_VMCS_GUEST_SYSENTER_EIP:
5480 {
5481 if (!(u64Val >> 32))
5482 {
5483 /* If this field is 64-bit, VT-x will zero out the top bits. */
5484 rc = VMXWriteVmcs32(idxField, (uint32_t)u64Val);
5485 }
5486 else
5487 {
5488 /* Assert that only the 32->64 switcher case should ever come here. */
5489 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests);
5490 rc = VMXWriteCachedVmcsEx(pVCpu, idxField, u64Val);
5491 }
5492 break;
5493 }
5494
5495 default:
5496 {
5497 AssertMsgFailed(("VMXWriteVmcs64Ex: Invalid field %#RX32 (pVCpu=%p u64Val=%#RX64)\n", idxField, pVCpu, u64Val));
5498 rc = VERR_INVALID_PARAMETER;
5499 break;
5500 }
5501 }
5502 AssertRCReturn(rc, rc);
5503 return rc;
5504}
5505
5506
5507/**
5508 * Queue up a VMWRITE by using the VMCS write cache. This is only used on 32-bit
5509 * hosts (except darwin) for 64-bit guests.
5510 *
5511 * @param pVCpu Pointer to the VMCPU.
5512 * @param idxField The VMCS field encoding.
5513 * @param u64Val 16, 32 or 64-bit value.
5514 */
5515VMMR0DECL(int) VMXWriteCachedVmcsEx(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val)
5516{
5517 AssertPtr(pVCpu);
5518 PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache;
5519
5520 AssertMsgReturn(pCache->Write.cValidEntries < VMCSCACHE_MAX_ENTRY - 1,
5521 ("entries=%u\n", pCache->Write.cValidEntries), VERR_ACCESS_DENIED);
5522
5523 /* Make sure there are no duplicates. */
5524 for (uint32_t i = 0; i < pCache->Write.cValidEntries; i++)
5525 {
5526 if (pCache->Write.aField[i] == idxField)
5527 {
5528 pCache->Write.aFieldVal[i] = u64Val;
5529 return VINF_SUCCESS;
5530 }
5531 }
5532
5533 pCache->Write.aField[pCache->Write.cValidEntries] = idxField;
5534 pCache->Write.aFieldVal[pCache->Write.cValidEntries] = u64Val;
5535 pCache->Write.cValidEntries++;
5536 return VINF_SUCCESS;
5537}
5538
5539/* Enable later when the assembly code uses these as callbacks. */
5540#if 0
5541/*
5542 * Loads the VMCS write-cache into the CPU (by executing VMWRITEs).
5543 *
5544 * @param pVCpu Pointer to the VMCPU.
5545 * @param pCache Pointer to the VMCS cache.
5546 *
5547 * @remarks No-long-jump zone!!!
5548 */
5549VMMR0DECL(void) VMXWriteCachedVmcsLoad(PVMCPU pVCpu, PVMCSCACHE pCache)
5550{
5551 AssertPtr(pCache);
5552 for (uint32_t i = 0; i < pCache->Write.cValidEntries; i++)
5553 {
5554 int rc = VMXWriteVmcs64(pCache->Write.aField[i], pCache->Write.aFieldVal[i]);
5555 AssertRC(rc);
5556 }
5557 pCache->Write.cValidEntries = 0;
5558}
5559
5560
5561/**
5562 * Stores the VMCS read-cache from the CPU (by executing VMREADs).
5563 *
5564 * @param pVCpu Pointer to the VMCPU.
5565 * @param pCache Pointer to the VMCS cache.
5566 *
5567 * @remarks No-long-jump zone!!!
5568 */
5569VMMR0DECL(void) VMXReadCachedVmcsStore(PVMCPU pVCpu, PVMCSCACHE pCache)
5570{
5571 AssertPtr(pCache);
5572 for (uint32_t i = 0; i < pCache->Read.cValidEntries; i++)
5573 {
5574 int rc = VMXReadVmcs64(pCache->Read.aField[i], &pCache->Read.aFieldVal[i]);
5575 AssertRC(rc);
5576 }
5577}
5578#endif
5579#endif /* HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL) */
5580
5581
5582/**
5583 * Sets up the usage of TSC-offsetting and updates the VMCS. If offsetting is
5584 * not possible, cause VM-exits on RDTSC(P)s. Also sets up the VMX preemption
5585 * timer.
5586 *
5587 * @returns VBox status code.
5588 * @param pVCpu Pointer to the VMCPU.
5589 *
5590 * @remarks No-long-jump zone!!!
5591 */
5592static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVMCPU pVCpu)
5593{
5594 int rc = VERR_INTERNAL_ERROR_5;
5595 bool fOffsettedTsc = false;
5596 bool fParavirtTsc = false;
5597 PVM pVM = pVCpu->CTX_SUFF(pVM);
5598 if (pVM->hm.s.vmx.fUsePreemptTimer)
5599 {
5600 uint64_t cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVCpu, &fOffsettedTsc, &fParavirtTsc,
5601 &pVCpu->hm.s.vmx.u64TSCOffset);
5602
5603 /* Make sure the returned values have sane upper and lower boundaries. */
5604 uint64_t u64CpuHz = SUPGetCpuHzFromGIP(g_pSUPGlobalInfoPage);
5605 cTicksToDeadline = RT_MIN(cTicksToDeadline, u64CpuHz / 64); /* 1/64th of a second */
5606 cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 2048); /* 1/2048th of a second */
5607 cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift;
5608
5609 uint32_t cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16);
5610 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_PREEMPT_TIMER_VALUE, cPreemptionTickCount); AssertRC(rc);
5611 }
5612 else
5613 fOffsettedTsc = TMCpuTickCanUseRealTSC(pVCpu, &pVCpu->hm.s.vmx.u64TSCOffset, &fParavirtTsc);
5614
5615#if 1
5616 if (fParavirtTsc)
5617 {
5618#if 1
5619 uint64_t const u64CurTsc = ASMReadTSC();
5620 uint64_t const u64LastTick = TMCpuTickGetLastSeen(pVCpu);
5621 if (u64CurTsc + pVCpu->hm.s.vmx.u64TSCOffset < u64LastTick)
5622 {
5623 pVCpu->hm.s.vmx.u64TSCOffset = (u64LastTick - u64CurTsc);
5624 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffsetAdjusted);
5625 }
5626
5627 Assert(u64CurTsc + pVCpu->hm.s.vmx.u64TSCOffset >= u64LastTick);
5628#endif
5629 rc = GIMR0UpdateParavirtTsc(pVM, pVCpu->hm.s.vmx.u64TSCOffset);
5630 AssertRC(rc);
5631 /* Note: VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT takes precedence over TSC_OFFSET, applies to RDTSCP too. */
5632 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, 0); AssertRC(rc);
5633
5634 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT;
5635 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc);
5636 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
5637 }
5638 else
5639#else
5640 if (fParavirtTsc)
5641 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
5642#endif
5643 if (fOffsettedTsc)
5644 {
5645 uint64_t u64CurTSC = ASMReadTSC();
5646 if (u64CurTSC + pVCpu->hm.s.vmx.u64TSCOffset >= TMCpuTickGetLastSeen(pVCpu))
5647 {
5648 /* Note: VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT takes precedence over TSC_OFFSET, applies to RDTSCP too. */
5649 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, pVCpu->hm.s.vmx.u64TSCOffset); AssertRC(rc);
5650
5651 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT;
5652 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc);
5653 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
5654 }
5655 else
5656 {
5657 /* VM-exit on RDTSC(P) as we would otherwise pass decreasing TSC values to the guest. */
5658 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT;
5659 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc);
5660 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscInterceptOverFlow);
5661 }
5662 }
5663 else
5664 {
5665 /* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */
5666 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT;
5667 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc);
5668 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
5669 }
5670}
5671
5672
5673/**
5674 * Determines if an exception is a contributory exception. Contributory
5675 * exceptions are ones which can cause double-faults. Page-fault is
5676 * intentionally not included here as it's a conditional contributory exception.
5677 *
5678 * @returns true if the exception is contributory, false otherwise.
5679 * @param uVector The exception vector.
5680 */
5681DECLINLINE(bool) hmR0VmxIsContributoryXcpt(const uint32_t uVector)
5682{
5683 switch (uVector)
5684 {
5685 case X86_XCPT_GP:
5686 case X86_XCPT_SS:
5687 case X86_XCPT_NP:
5688 case X86_XCPT_TS:
5689 case X86_XCPT_DE:
5690 return true;
5691 default:
5692 break;
5693 }
5694 return false;
5695}
5696
5697
5698/**
5699 * Sets an event as a pending event to be injected into the guest.
5700 *
5701 * @param pVCpu Pointer to the VMCPU.
5702 * @param u32IntInfo The VM-entry interruption-information field.
5703 * @param cbInstr The VM-entry instruction length in bytes (for software
5704 * interrupts, exceptions and privileged software
5705 * exceptions).
5706 * @param u32ErrCode The VM-entry exception error code.
5707 * @param GCPtrFaultAddress The fault-address (CR2) in case it's a
5708 * page-fault.
5709 *
5710 * @remarks Statistics counter assumes this is a guest event being injected or
5711 * re-injected into the guest, i.e. 'StatInjectPendingReflect' is
5712 * always incremented.
5713 */
5714DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPU pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode,
5715 RTGCUINTPTR GCPtrFaultAddress)
5716{
5717 Assert(!pVCpu->hm.s.Event.fPending);
5718 pVCpu->hm.s.Event.fPending = true;
5719 pVCpu->hm.s.Event.u64IntInfo = u32IntInfo;
5720 pVCpu->hm.s.Event.u32ErrCode = u32ErrCode;
5721 pVCpu->hm.s.Event.cbInstr = cbInstr;
5722 pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
5723
5724 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect);
5725}
5726
5727
5728/**
5729 * Sets a double-fault (#DF) exception as pending-for-injection into the VM.
5730 *
5731 * @param pVCpu Pointer to the VMCPU.
5732 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
5733 * out-of-sync. Make sure to update the required fields
5734 * before using them.
5735 */
5736DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
5737{
5738 NOREF(pMixedCtx);
5739 uint32_t u32IntInfo = X86_XCPT_DF | VMX_EXIT_INTERRUPTION_INFO_VALID;
5740 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
5741 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
5742 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
5743}
5744
5745
5746/**
5747 * Handle a condition that occurred while delivering an event through the guest
5748 * IDT.
5749 *
5750 * @returns VBox status code (informational error codes included).
5751 * @retval VINF_SUCCESS if we should continue handling the VM-exit.
5752 * @retval VINF_HM_DOUBLE_FAULT if a #DF condition was detected and we ought to
5753 * continue execution of the guest which will delivery the #DF.
5754 * @retval VINF_EM_RESET if we detected a triple-fault condition.
5755 *
5756 * @param pVCpu Pointer to the VMCPU.
5757 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
5758 * out-of-sync. Make sure to update the required fields
5759 * before using them.
5760 * @param pVmxTransient Pointer to the VMX transient structure.
5761 *
5762 * @remarks No-long-jump zone!!!
5763 */
5764static int hmR0VmxCheckExitDueToEventDelivery(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
5765{
5766 uint32_t uExitVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo);
5767
5768 int rc = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
5769 AssertRCReturn(rc, rc);
5770 if (VMX_IDT_VECTORING_INFO_VALID(pVmxTransient->uIdtVectoringInfo))
5771 {
5772 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
5773 AssertRCReturn(rc, rc);
5774
5775 uint32_t uIdtVectorType = VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo);
5776 uint32_t uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo);
5777
5778 typedef enum
5779 {
5780 VMXREFLECTXCPT_XCPT, /* Reflect the exception to the guest or for further evaluation by VMM. */
5781 VMXREFLECTXCPT_DF, /* Reflect the exception as a double-fault to the guest. */
5782 VMXREFLECTXCPT_TF, /* Indicate a triple faulted state to the VMM. */
5783 VMXREFLECTXCPT_NONE /* Nothing to reflect. */
5784 } VMXREFLECTXCPT;
5785
5786 /* See Intel spec. 30.7.1.1 "Reflecting Exceptions to Guest Software". */
5787 VMXREFLECTXCPT enmReflect = VMXREFLECTXCPT_NONE;
5788 if (VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo))
5789 {
5790 if (uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT)
5791 {
5792 enmReflect = VMXREFLECTXCPT_XCPT;
5793#ifdef VBOX_STRICT
5794 if ( hmR0VmxIsContributoryXcpt(uIdtVector)
5795 && uExitVector == X86_XCPT_PF)
5796 {
5797 Log4(("IDT: vcpu[%RU32] Contributory #PF uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5798 }
5799#endif
5800 if ( uExitVector == X86_XCPT_PF
5801 && uIdtVector == X86_XCPT_PF)
5802 {
5803 pVmxTransient->fVectoringPF = true;
5804 Log4(("IDT: vcpu[%RU32] Vectoring #PF uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5805 }
5806 else if ( (pVCpu->hm.s.vmx.u32XcptBitmap & HMVMX_CONTRIBUTORY_XCPT_MASK)
5807 && hmR0VmxIsContributoryXcpt(uExitVector)
5808 && ( hmR0VmxIsContributoryXcpt(uIdtVector)
5809 || uIdtVector == X86_XCPT_PF))
5810 {
5811 enmReflect = VMXREFLECTXCPT_DF;
5812 }
5813 else if (uIdtVector == X86_XCPT_DF)
5814 enmReflect = VMXREFLECTXCPT_TF;
5815 }
5816 else if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
5817 || uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT
5818 || uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI)
5819 {
5820 /*
5821 * Ignore software interrupts (INT n), software exceptions (#BP, #OF) and
5822 * privileged software exception (#DB from ICEBP) as they reoccur when restarting the instruction.
5823 */
5824 enmReflect = VMXREFLECTXCPT_XCPT;
5825 }
5826 }
5827 else if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
5828 || uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT
5829 || uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI)
5830 {
5831 /*
5832 * If event delivery caused an EPT violation/misconfig or APIC access VM-exit, then the VM-exit
5833 * interruption-information will not be valid as it's not an exception and we end up here. In such cases,
5834 * it is sufficient to reflect the original exception to the guest after handling the VM-exit.
5835 */
5836 enmReflect = VMXREFLECTXCPT_XCPT;
5837 }
5838
5839 /*
5840 * On CPUs that support Virtual NMIs, if this VM-exit (be it an exception or EPT violation/misconfig etc.) occurred
5841 * while delivering the NMI, we need to clear the block-by-NMI field in the guest interruptibility-state before
5842 * re-delivering the NMI after handling the VM-exit. Otherwise the subsequent VM-entry would fail.
5843 *
5844 * See Intel spec. 30.7.1.2 "Resuming Guest Software after Handling an Exception". See @bugref{7445}.
5845 */
5846 if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
5847 && enmReflect == VMXREFLECTXCPT_XCPT
5848 && (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
5849 && VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
5850 {
5851 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
5852 }
5853
5854 switch (enmReflect)
5855 {
5856 case VMXREFLECTXCPT_XCPT:
5857 {
5858 Assert( uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_SW_INT
5859 && uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
5860 && uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT);
5861
5862 uint32_t u32ErrCode = 0;
5863 if (VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uIdtVectoringInfo))
5864 {
5865 rc = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
5866 AssertRCReturn(rc, rc);
5867 u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
5868 }
5869
5870 /* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF. See hmR0VmxExitXcptPF(). */
5871 hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo),
5872 0 /* cbInstr */, u32ErrCode, pMixedCtx->cr2);
5873 rc = VINF_SUCCESS;
5874 Log4(("IDT: vcpu[%RU32] Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->idCpu,
5875 pVCpu->hm.s.Event.u64IntInfo, pVCpu->hm.s.Event.u32ErrCode));
5876
5877 break;
5878 }
5879
5880 case VMXREFLECTXCPT_DF:
5881 {
5882 hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
5883 rc = VINF_HM_DOUBLE_FAULT;
5884 Log4(("IDT: vcpu[%RU32] Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->idCpu,
5885 pVCpu->hm.s.Event.u64IntInfo, uIdtVector, uExitVector));
5886
5887 break;
5888 }
5889
5890 case VMXREFLECTXCPT_TF:
5891 {
5892 rc = VINF_EM_RESET;
5893 Log4(("IDT: vcpu[%RU32] Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", pVCpu->idCpu, uIdtVector,
5894 uExitVector));
5895 break;
5896 }
5897
5898 default:
5899 Assert(rc == VINF_SUCCESS);
5900 break;
5901 }
5902 }
5903 else if ( VMX_EXIT_INTERRUPTION_INFO_NMI_UNBLOCK_IRET(pVmxTransient->uExitIntInfo)
5904 && uExitVector != X86_XCPT_DF
5905 && (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI))
5906 {
5907 /*
5908 * Execution of IRET caused this fault when NMI blocking was in effect (i.e we're in the guest NMI handler).
5909 * We need to set the block-by-NMI field so that NMIs remain blocked until the IRET execution is restarted.
5910 * See Intel spec. 30.7.1.2 "Resuming guest software after handling an exception".
5911 */
5912 if (!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
5913 VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
5914 }
5915
5916 Assert(rc == VINF_SUCCESS || rc == VINF_HM_DOUBLE_FAULT || rc == VINF_EM_RESET);
5917 return rc;
5918}
5919
5920
5921/**
5922 * Saves the guest's CR0 register from the VMCS into the guest-CPU context.
5923 *
5924 * @returns VBox status code.
5925 * @param pVCpu Pointer to the VMCPU.
5926 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
5927 * out-of-sync. Make sure to update the required fields
5928 * before using them.
5929 *
5930 * @remarks No-long-jump zone!!!
5931 */
5932static int hmR0VmxSaveGuestCR0(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
5933{
5934 NOREF(pMixedCtx);
5935
5936 /*
5937 * While in the middle of saving guest-CR0, we could get preempted and re-invoked from the preemption hook,
5938 * see hmR0VmxLeave(). Safer to just make this code non-preemptible.
5939 */
5940 VMMRZCallRing3Disable(pVCpu);
5941 HM_DISABLE_PREEMPT_IF_NEEDED();
5942
5943 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0))
5944 {
5945 uint32_t uVal = 0;
5946 int rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &uVal);
5947 AssertRCReturn(rc, rc);
5948
5949 uint32_t uShadow = 0;
5950 rc = VMXReadVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uShadow);
5951 AssertRCReturn(rc, rc);
5952
5953 uVal = (uShadow & pVCpu->hm.s.vmx.u32CR0Mask) | (uVal & ~pVCpu->hm.s.vmx.u32CR0Mask);
5954 CPUMSetGuestCR0(pVCpu, uVal);
5955 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0);
5956 }
5957
5958 HM_RESTORE_PREEMPT_IF_NEEDED();
5959 VMMRZCallRing3Enable(pVCpu);
5960 return VINF_SUCCESS;
5961}
5962
5963
5964/**
5965 * Saves the guest's CR4 register from the VMCS into the guest-CPU context.
5966 *
5967 * @returns VBox status code.
5968 * @param pVCpu Pointer to the VMCPU.
5969 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
5970 * out-of-sync. Make sure to update the required fields
5971 * before using them.
5972 *
5973 * @remarks No-long-jump zone!!!
5974 */
5975static int hmR0VmxSaveGuestCR4(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
5976{
5977 NOREF(pMixedCtx);
5978
5979 int rc = VINF_SUCCESS;
5980 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4))
5981 {
5982 uint32_t uVal = 0;
5983 uint32_t uShadow = 0;
5984 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &uVal);
5985 AssertRCReturn(rc, rc);
5986 rc = VMXReadVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uShadow);
5987 AssertRCReturn(rc, rc);
5988
5989 uVal = (uShadow & pVCpu->hm.s.vmx.u32CR4Mask) | (uVal & ~pVCpu->hm.s.vmx.u32CR4Mask);
5990 CPUMSetGuestCR4(pVCpu, uVal);
5991 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4);
5992 }
5993 return rc;
5994}
5995
5996
5997/**
5998 * Saves the guest's RIP register from the VMCS into the guest-CPU context.
5999 *
6000 * @returns VBox status code.
6001 * @param pVCpu Pointer to the VMCPU.
6002 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6003 * out-of-sync. Make sure to update the required fields
6004 * before using them.
6005 *
6006 * @remarks No-long-jump zone!!!
6007 */
6008static int hmR0VmxSaveGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6009{
6010 int rc = VINF_SUCCESS;
6011 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP))
6012 {
6013 uint64_t u64Val = 0;
6014 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val);
6015 AssertRCReturn(rc, rc);
6016
6017 pMixedCtx->rip = u64Val;
6018 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP);
6019 }
6020 return rc;
6021}
6022
6023
6024/**
6025 * Saves the guest's RSP register from the VMCS into the guest-CPU context.
6026 *
6027 * @returns VBox status code.
6028 * @param pVCpu Pointer to the VMCPU.
6029 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6030 * out-of-sync. Make sure to update the required fields
6031 * before using them.
6032 *
6033 * @remarks No-long-jump zone!!!
6034 */
6035static int hmR0VmxSaveGuestRsp(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6036{
6037 int rc = VINF_SUCCESS;
6038 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RSP))
6039 {
6040 uint64_t u64Val = 0;
6041 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val);
6042 AssertRCReturn(rc, rc);
6043
6044 pMixedCtx->rsp = u64Val;
6045 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RSP);
6046 }
6047 return rc;
6048}
6049
6050
6051/**
6052 * Saves the guest's RFLAGS from the VMCS into the guest-CPU context.
6053 *
6054 * @returns VBox status code.
6055 * @param pVCpu Pointer to the VMCPU.
6056 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6057 * out-of-sync. Make sure to update the required fields
6058 * before using them.
6059 *
6060 * @remarks No-long-jump zone!!!
6061 */
6062static int hmR0VmxSaveGuestRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6063{
6064 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS))
6065 {
6066 uint32_t uVal = 0;
6067 int rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &uVal);
6068 AssertRCReturn(rc, rc);
6069
6070 pMixedCtx->eflags.u32 = uVal;
6071 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) /* Undo our real-on-v86-mode changes to eflags if necessary. */
6072 {
6073 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
6074 Log4(("Saving real-mode EFLAGS VT-x view=%#RX32\n", pMixedCtx->eflags.u32));
6075
6076 pMixedCtx->eflags.Bits.u1VM = 0;
6077 pMixedCtx->eflags.Bits.u2IOPL = pVCpu->hm.s.vmx.RealMode.Eflags.Bits.u2IOPL;
6078 }
6079
6080 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS);
6081 }
6082 return VINF_SUCCESS;
6083}
6084
6085
6086/**
6087 * Wrapper for saving the guest's RIP, RSP and RFLAGS from the VMCS into the
6088 * guest-CPU context.
6089 */
6090DECLINLINE(int) hmR0VmxSaveGuestRipRspRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6091{
6092 int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6093 rc |= hmR0VmxSaveGuestRsp(pVCpu, pMixedCtx);
6094 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
6095 return rc;
6096}
6097
6098
6099/**
6100 * Saves the guest's interruptibility-state ("interrupt shadow" as AMD calls it)
6101 * from the guest-state area in the VMCS.
6102 *
6103 * @param pVCpu Pointer to the VMCPU.
6104 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6105 * out-of-sync. Make sure to update the required fields
6106 * before using them.
6107 *
6108 * @remarks No-long-jump zone!!!
6109 */
6110static void hmR0VmxSaveGuestIntrState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6111{
6112 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_INTR_STATE))
6113 {
6114 uint32_t uIntrState = 0;
6115 int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
6116 AssertRC(rc);
6117
6118 if (!uIntrState)
6119 {
6120 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
6121 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
6122
6123 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6124 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
6125 }
6126 else
6127 {
6128 if (uIntrState & ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS
6129 | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI))
6130 {
6131 rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6132 AssertRC(rc);
6133 rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); /* for hmR0VmxGetGuestIntrState(). */
6134 AssertRC(rc);
6135
6136 EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip);
6137 Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS));
6138 }
6139 else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
6140 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
6141
6142 if (uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI)
6143 {
6144 if (!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6145 VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
6146 }
6147 else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6148 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
6149 }
6150
6151 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_INTR_STATE);
6152 }
6153}
6154
6155
6156/**
6157 * Saves the guest's activity state.
6158 *
6159 * @returns VBox status code.
6160 * @param pVCpu Pointer to the VMCPU.
6161 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6162 * out-of-sync. Make sure to update the required fields
6163 * before using them.
6164 *
6165 * @remarks No-long-jump zone!!!
6166 */
6167static int hmR0VmxSaveGuestActivityState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6168{
6169 NOREF(pMixedCtx);
6170 /* Nothing to do for now until we make use of different guest-CPU activity state. Just update the flag. */
6171 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_ACTIVITY_STATE);
6172 return VINF_SUCCESS;
6173}
6174
6175
6176/**
6177 * Saves the guest SYSENTER MSRs (SYSENTER_CS, SYSENTER_EIP, SYSENTER_ESP) from
6178 * the current VMCS into the guest-CPU context.
6179 *
6180 * @returns VBox status code.
6181 * @param pVCpu Pointer to the VMCPU.
6182 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6183 * out-of-sync. Make sure to update the required fields
6184 * before using them.
6185 *
6186 * @remarks No-long-jump zone!!!
6187 */
6188static int hmR0VmxSaveGuestSysenterMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6189{
6190 int rc = VINF_SUCCESS;
6191 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR))
6192 {
6193 uint32_t u32Val = 0;
6194 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val); AssertRCReturn(rc, rc);
6195 pMixedCtx->SysEnter.cs = u32Val;
6196 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR);
6197 }
6198
6199 uint64_t u64Val = 0;
6200 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR))
6201 {
6202 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, &u64Val); AssertRCReturn(rc, rc);
6203 pMixedCtx->SysEnter.eip = u64Val;
6204 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR);
6205 }
6206 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR))
6207 {
6208 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, &u64Val); AssertRCReturn(rc, rc);
6209 pMixedCtx->SysEnter.esp = u64Val;
6210 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR);
6211 }
6212 return rc;
6213}
6214
6215
6216/**
6217 * Saves the set of guest MSRs (that we restore lazily while leaving VT-x) from
6218 * the CPU back into the guest-CPU context.
6219 *
6220 * @returns VBox status code.
6221 * @param pVCpu Pointer to the VMCPU.
6222 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6223 * out-of-sync. Make sure to update the required fields
6224 * before using them.
6225 *
6226 * @remarks No-long-jump zone!!!
6227 */
6228static int hmR0VmxSaveGuestLazyMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6229{
6230#if HC_ARCH_BITS == 64
6231 if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
6232 {
6233 /* Since this can be called from our preemption hook it's safer to make the guest-MSRs update non-preemptible. */
6234 VMMRZCallRing3Disable(pVCpu);
6235 HM_DISABLE_PREEMPT_IF_NEEDED();
6236
6237 /* Doing the check here ensures we don't overwrite already-saved guest MSRs from a preemption hook. */
6238 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS))
6239 {
6240 hmR0VmxLazySaveGuestMsrs(pVCpu, pMixedCtx);
6241 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS);
6242 }
6243
6244 HM_RESTORE_PREEMPT_IF_NEEDED();
6245 VMMRZCallRing3Enable(pVCpu);
6246 }
6247 else
6248 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS);
6249#else
6250 NOREF(pMixedCtx);
6251 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS);
6252#endif
6253
6254 return VINF_SUCCESS;
6255}
6256
6257
6258/**
6259 * Saves the auto load/store'd guest MSRs from the current VMCS into
6260 * the guest-CPU context.
6261 *
6262 * @returns VBox status code.
6263 * @param pVCpu Pointer to the VMCPU.
6264 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6265 * out-of-sync. Make sure to update the required fields
6266 * before using them.
6267 *
6268 * @remarks No-long-jump zone!!!
6269 */
6270static int hmR0VmxSaveGuestAutoLoadStoreMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6271{
6272 if (HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS))
6273 return VINF_SUCCESS;
6274
6275 PVMXAUTOMSR pMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
6276 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
6277 Log4(("hmR0VmxSaveGuestAutoLoadStoreMsrs: cMsrs=%u\n", cMsrs));
6278 for (uint32_t i = 0; i < cMsrs; i++, pMsr++)
6279 {
6280 switch (pMsr->u32Msr)
6281 {
6282 case MSR_K8_TSC_AUX: CPUMR0SetGuestTscAux(pVCpu, pMsr->u64Value); break;
6283 case MSR_K8_LSTAR: pMixedCtx->msrLSTAR = pMsr->u64Value; break;
6284 case MSR_K6_STAR: pMixedCtx->msrSTAR = pMsr->u64Value; break;
6285 case MSR_K8_SF_MASK: pMixedCtx->msrSFMASK = pMsr->u64Value; break;
6286 case MSR_K8_KERNEL_GS_BASE: pMixedCtx->msrKERNELGSBASE = pMsr->u64Value; break;
6287 case MSR_K6_EFER: /* Nothing to do here since we intercept writes, see hmR0VmxLoadGuestMsrs(). */
6288 break;
6289
6290 default:
6291 {
6292 AssertMsgFailed(("Unexpected MSR in auto-load/store area. uMsr=%#RX32 cMsrs=%u\n", pMsr->u32Msr, cMsrs));
6293 pVCpu->hm.s.u32HMError = pMsr->u32Msr;
6294 return VERR_HM_UNEXPECTED_LD_ST_MSR;
6295 }
6296 }
6297 }
6298
6299 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS);
6300 return VINF_SUCCESS;
6301}
6302
6303
6304/**
6305 * Saves the guest control registers from the current VMCS into the guest-CPU
6306 * context.
6307 *
6308 * @returns VBox status code.
6309 * @param pVCpu Pointer to the VMCPU.
6310 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6311 * out-of-sync. Make sure to update the required fields
6312 * before using them.
6313 *
6314 * @remarks No-long-jump zone!!!
6315 */
6316static int hmR0VmxSaveGuestControlRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6317{
6318 /* Guest CR0. Guest FPU. */
6319 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6320 AssertRCReturn(rc, rc);
6321
6322 /* Guest CR4. */
6323 rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
6324 AssertRCReturn(rc, rc);
6325
6326 /* Guest CR2 - updated always during the world-switch or in #PF. */
6327 /* Guest CR3. Only changes with Nested Paging. This must be done -after- saving CR0 and CR4 from the guest! */
6328 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR3))
6329 {
6330 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
6331 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4));
6332
6333 PVM pVM = pVCpu->CTX_SUFF(pVM);
6334 if ( pVM->hm.s.vmx.fUnrestrictedGuest
6335 || ( pVM->hm.s.fNestedPaging
6336 && CPUMIsGuestPagingEnabledEx(pMixedCtx)))
6337 {
6338 uint64_t u64Val = 0;
6339 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_CR3, &u64Val);
6340 if (pMixedCtx->cr3 != u64Val)
6341 {
6342 CPUMSetGuestCR3(pVCpu, u64Val);
6343 if (VMMRZCallRing3IsEnabled(pVCpu))
6344 {
6345 PGMUpdateCR3(pVCpu, u64Val);
6346 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6347 }
6348 else
6349 {
6350 /* Set the force flag to inform PGM about it when necessary. It is cleared by PGMUpdateCR3().*/
6351 VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
6352 }
6353 }
6354
6355 /* If the guest is in PAE mode, sync back the PDPE's into the guest state. */
6356 if (CPUMIsGuestInPAEModeEx(pMixedCtx)) /* Reads CR0, CR4 and EFER MSR (EFER is always up-to-date). */
6357 {
6358 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &pVCpu->hm.s.aPdpes[0].u); AssertRCReturn(rc, rc);
6359 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &pVCpu->hm.s.aPdpes[1].u); AssertRCReturn(rc, rc);
6360 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &pVCpu->hm.s.aPdpes[2].u); AssertRCReturn(rc, rc);
6361 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &pVCpu->hm.s.aPdpes[3].u); AssertRCReturn(rc, rc);
6362
6363 if (VMMRZCallRing3IsEnabled(pVCpu))
6364 {
6365 PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6366 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6367 }
6368 else
6369 {
6370 /* Set the force flag to inform PGM about it when necessary. It is cleared by PGMGstUpdatePaePdpes(). */
6371 VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES);
6372 }
6373 }
6374 }
6375
6376 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR3);
6377 }
6378
6379 /*
6380 * Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> hmR0VmxCallRing3Callback()
6381 * -> VMMRZCallRing3Disable() -> hmR0VmxSaveGuestState() -> Set VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp
6382 * -> continue with VM-exit handling -> hmR0VmxSaveGuestControlRegs() and here we are.
6383 *
6384 * The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus
6385 * if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that
6386 * calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should
6387 * -NOT- check if HMVMX_UPDATED_GUEST_CR3 is already set or not!
6388 *
6389 * The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here.
6390 */
6391 if (VMMRZCallRing3IsEnabled(pVCpu))
6392 {
6393 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
6394 PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
6395
6396 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
6397 PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6398
6399 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6400 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6401 }
6402
6403 return rc;
6404}
6405
6406
6407/**
6408 * Reads a guest segment register from the current VMCS into the guest-CPU
6409 * context.
6410 *
6411 * @returns VBox status code.
6412 * @param pVCpu Pointer to the VMCPU.
6413 * @param idxSel Index of the selector in the VMCS.
6414 * @param idxLimit Index of the segment limit in the VMCS.
6415 * @param idxBase Index of the segment base in the VMCS.
6416 * @param idxAccess Index of the access rights of the segment in the VMCS.
6417 * @param pSelReg Pointer to the segment selector.
6418 *
6419 * @remarks No-long-jump zone!!!
6420 * @remarks Never call this function directly!!! Use the VMXLOCAL_READ_SEG()
6421 * macro as that takes care of whether to read from the VMCS cache or
6422 * not.
6423 */
6424DECLINLINE(int) hmR0VmxReadSegmentReg(PVMCPU pVCpu, uint32_t idxSel, uint32_t idxLimit, uint32_t idxBase, uint32_t idxAccess,
6425 PCPUMSELREG pSelReg)
6426{
6427 NOREF(pVCpu);
6428
6429 uint32_t u32Val = 0;
6430 int rc = VMXReadVmcs32(idxSel, &u32Val);
6431 AssertRCReturn(rc, rc);
6432 pSelReg->Sel = (uint16_t)u32Val;
6433 pSelReg->ValidSel = (uint16_t)u32Val;
6434 pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
6435
6436 rc = VMXReadVmcs32(idxLimit, &u32Val);
6437 AssertRCReturn(rc, rc);
6438 pSelReg->u32Limit = u32Val;
6439
6440 uint64_t u64Val = 0;
6441 rc = VMXReadVmcsGstNByIdxVal(idxBase, &u64Val);
6442 AssertRCReturn(rc, rc);
6443 pSelReg->u64Base = u64Val;
6444
6445 rc = VMXReadVmcs32(idxAccess, &u32Val);
6446 AssertRCReturn(rc, rc);
6447 pSelReg->Attr.u = u32Val;
6448
6449 /*
6450 * If VT-x marks the segment as unusable, most other bits remain undefined:
6451 * - For CS the L, D and G bits have meaning.
6452 * - For SS the DPL has meaning (it -is- the CPL for Intel and VBox).
6453 * - For the remaining data segments no bits are defined.
6454 *
6455 * The present bit and the unusable bit has been observed to be set at the
6456 * same time (the selector was supposed to be invalid as we started executing
6457 * a V8086 interrupt in ring-0).
6458 *
6459 * What should be important for the rest of the VBox code, is that the P bit is
6460 * cleared. Some of the other VBox code recognizes the unusable bit, but
6461 * AMD-V certainly don't, and REM doesn't really either. So, to be on the
6462 * safe side here, we'll strip off P and other bits we don't care about. If
6463 * any code breaks because Attr.u != 0 when Sel < 4, it should be fixed.
6464 *
6465 * See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
6466 */
6467 if (pSelReg->Attr.u & X86DESCATTR_UNUSABLE)
6468 {
6469 Assert(idxSel != VMX_VMCS16_GUEST_FIELD_TR); /* TR is the only selector that can never be unusable. */
6470
6471 /* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */
6472 pSelReg->Attr.u &= X86DESCATTR_UNUSABLE | X86DESCATTR_L | X86DESCATTR_D | X86DESCATTR_G
6473 | X86DESCATTR_DPL | X86DESCATTR_TYPE | X86DESCATTR_DT;
6474
6475 Log4(("hmR0VmxReadSegmentReg: Unusable idxSel=%#x attr=%#x -> %#x\n", idxSel, u32Val, pSelReg->Attr.u));
6476#ifdef DEBUG_bird
6477 AssertMsg((u32Val & ~X86DESCATTR_P) == pSelReg->Attr.u,
6478 ("%#x: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n",
6479 idxSel, u32Val, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit));
6480#endif
6481 }
6482 return VINF_SUCCESS;
6483}
6484
6485
6486#ifdef VMX_USE_CACHED_VMCS_ACCESSES
6487# define VMXLOCAL_READ_SEG(Sel, CtxSel) \
6488 hmR0VmxReadSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_##Sel, VMX_VMCS32_GUEST_##Sel##_LIMIT, \
6489 VMX_VMCS_GUEST_##Sel##_BASE_CACHE_IDX, VMX_VMCS32_GUEST_##Sel##_ACCESS_RIGHTS, &pMixedCtx->CtxSel)
6490#else
6491# define VMXLOCAL_READ_SEG(Sel, CtxSel) \
6492 hmR0VmxReadSegmentReg(pVCpu, VMX_VMCS16_GUEST_FIELD_##Sel, VMX_VMCS32_GUEST_##Sel##_LIMIT, \
6493 VMX_VMCS_GUEST_##Sel##_BASE, VMX_VMCS32_GUEST_##Sel##_ACCESS_RIGHTS, &pMixedCtx->CtxSel)
6494#endif
6495
6496
6497/**
6498 * Saves the guest segment registers from the current VMCS into the guest-CPU
6499 * context.
6500 *
6501 * @returns VBox status code.
6502 * @param pVCpu Pointer to the VMCPU.
6503 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6504 * out-of-sync. Make sure to update the required fields
6505 * before using them.
6506 *
6507 * @remarks No-long-jump zone!!!
6508 */
6509static int hmR0VmxSaveGuestSegmentRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6510{
6511 /* Guest segment registers. */
6512 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SEGMENT_REGS))
6513 {
6514 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); AssertRCReturn(rc, rc);
6515 rc = VMXLOCAL_READ_SEG(CS, cs); AssertRCReturn(rc, rc);
6516 rc = VMXLOCAL_READ_SEG(SS, ss); AssertRCReturn(rc, rc);
6517 rc = VMXLOCAL_READ_SEG(DS, ds); AssertRCReturn(rc, rc);
6518 rc = VMXLOCAL_READ_SEG(ES, es); AssertRCReturn(rc, rc);
6519 rc = VMXLOCAL_READ_SEG(FS, fs); AssertRCReturn(rc, rc);
6520 rc = VMXLOCAL_READ_SEG(GS, gs); AssertRCReturn(rc, rc);
6521
6522 /* Restore segment attributes for real-on-v86 mode hack. */
6523 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
6524 {
6525 pMixedCtx->cs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrCS.u;
6526 pMixedCtx->ss.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrSS.u;
6527 pMixedCtx->ds.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrDS.u;
6528 pMixedCtx->es.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrES.u;
6529 pMixedCtx->fs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrFS.u;
6530 pMixedCtx->gs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrGS.u;
6531 }
6532 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SEGMENT_REGS);
6533 }
6534
6535 return VINF_SUCCESS;
6536}
6537
6538
6539/**
6540 * Saves the guest descriptor table registers and task register from the current
6541 * VMCS into the guest-CPU context.
6542 *
6543 * @returns VBox status code.
6544 * @param pVCpu Pointer to the VMCPU.
6545 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6546 * out-of-sync. Make sure to update the required fields
6547 * before using them.
6548 *
6549 * @remarks No-long-jump zone!!!
6550 */
6551static int hmR0VmxSaveGuestTableRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6552{
6553 int rc = VINF_SUCCESS;
6554
6555 /* Guest LDTR. */
6556 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LDTR))
6557 {
6558 rc = VMXLOCAL_READ_SEG(LDTR, ldtr);
6559 AssertRCReturn(rc, rc);
6560 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LDTR);
6561 }
6562
6563 /* Guest GDTR. */
6564 uint64_t u64Val = 0;
6565 uint32_t u32Val = 0;
6566 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_GDTR))
6567 {
6568 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, &u64Val); AssertRCReturn(rc, rc);
6569 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRCReturn(rc, rc);
6570 pMixedCtx->gdtr.pGdt = u64Val;
6571 pMixedCtx->gdtr.cbGdt = u32Val;
6572 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_GDTR);
6573 }
6574
6575 /* Guest IDTR. */
6576 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_IDTR))
6577 {
6578 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, &u64Val); AssertRCReturn(rc, rc);
6579 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRCReturn(rc, rc);
6580 pMixedCtx->idtr.pIdt = u64Val;
6581 pMixedCtx->idtr.cbIdt = u32Val;
6582 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_IDTR);
6583 }
6584
6585 /* Guest TR. */
6586 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_TR))
6587 {
6588 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6589 AssertRCReturn(rc, rc);
6590
6591 /* For real-mode emulation using virtual-8086 mode we have the fake TSS (pRealModeTSS) in TR, don't save the fake one. */
6592 if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
6593 {
6594 rc = VMXLOCAL_READ_SEG(TR, tr);
6595 AssertRCReturn(rc, rc);
6596 }
6597 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_TR);
6598 }
6599 return rc;
6600}
6601
6602#undef VMXLOCAL_READ_SEG
6603
6604
6605/**
6606 * Saves the guest debug-register DR7 from the current VMCS into the guest-CPU
6607 * context.
6608 *
6609 * @returns VBox status code.
6610 * @param pVCpu Pointer to the VMCPU.
6611 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6612 * out-of-sync. Make sure to update the required fields
6613 * before using them.
6614 *
6615 * @remarks No-long-jump zone!!!
6616 */
6617static int hmR0VmxSaveGuestDR7(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6618{
6619 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_DEBUG))
6620 {
6621 if (!pVCpu->hm.s.fUsingHyperDR7)
6622 {
6623 /* Upper 32-bits are always zero. See Intel spec. 2.7.3 "Loading and Storing Debug Registers". */
6624 uint32_t u32Val;
6625 int rc = VMXReadVmcs32(VMX_VMCS_GUEST_DR7, &u32Val); AssertRCReturn(rc, rc);
6626 pMixedCtx->dr[7] = u32Val;
6627 }
6628
6629 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_DEBUG);
6630 }
6631 return VINF_SUCCESS;
6632}
6633
6634
6635/**
6636 * Saves the guest APIC state from the current VMCS into the guest-CPU context.
6637 *
6638 * @returns VBox status code.
6639 * @param pVCpu Pointer to the VMCPU.
6640 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6641 * out-of-sync. Make sure to update the required fields
6642 * before using them.
6643 *
6644 * @remarks No-long-jump zone!!!
6645 */
6646static int hmR0VmxSaveGuestApicState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6647{
6648 NOREF(pMixedCtx);
6649
6650 /* Updating TPR is already done in hmR0VmxPostRunGuest(). Just update the flag. */
6651 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_APIC_STATE);
6652 return VINF_SUCCESS;
6653}
6654
6655
6656/**
6657 * Saves the entire guest state from the currently active VMCS into the
6658 * guest-CPU context. This essentially VMREADs all guest-data.
6659 *
6660 * @returns VBox status code.
6661 * @param pVCpu Pointer to the VMCPU.
6662 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
6663 * out-of-sync. Make sure to update the required fields
6664 * before using them.
6665 */
6666static int hmR0VmxSaveGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6667{
6668 Assert(pVCpu);
6669 Assert(pMixedCtx);
6670
6671 if (HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL)
6672 return VINF_SUCCESS;
6673
6674 /* Though we can longjmp to ring-3 due to log-flushes here and get recalled
6675 again on the ring-3 callback path, there is no real need to. */
6676 if (VMMRZCallRing3IsEnabled(pVCpu))
6677 VMMR0LogFlushDisable(pVCpu);
6678 else
6679 Assert(VMMR0IsLogFlushDisabled(pVCpu));
6680 Log4Func(("vcpu[%RU32]\n", pVCpu->idCpu));
6681
6682 int rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
6683 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestRipRspRflags failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6684
6685 rc = hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6686 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestControlRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6687
6688 rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
6689 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestSegmentRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6690
6691 rc = hmR0VmxSaveGuestTableRegs(pVCpu, pMixedCtx);
6692 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestTableRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6693
6694 rc = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
6695 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestDR7 failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6696
6697 rc = hmR0VmxSaveGuestSysenterMsrs(pVCpu, pMixedCtx);
6698 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestSysenterMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6699
6700 rc = hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
6701 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestLazyMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6702
6703 rc = hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
6704 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestAutoLoadStoreMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6705
6706 rc = hmR0VmxSaveGuestActivityState(pVCpu, pMixedCtx);
6707 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestActivityState failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6708
6709 rc = hmR0VmxSaveGuestApicState(pVCpu, pMixedCtx);
6710 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestApicState failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6711
6712 AssertMsg(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL,
6713 ("Missed guest state bits while saving state; residue %RX32\n", HMVMXCPU_GST_VALUE(pVCpu)));
6714
6715 if (VMMRZCallRing3IsEnabled(pVCpu))
6716 VMMR0LogFlushEnable(pVCpu);
6717
6718 return rc;
6719}
6720
6721
6722/**
6723 * Check per-VM and per-VCPU force flag actions that require us to go back to
6724 * ring-3 for one reason or another.
6725 *
6726 * @returns VBox status code (information status code included).
6727 * @retval VINF_SUCCESS if we don't have any actions that require going back to
6728 * ring-3.
6729 * @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
6730 * @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
6731 * interrupts)
6732 * @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
6733 * all EMTs to be in ring-3.
6734 * @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
6735 * @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
6736 * to the EM loop.
6737 *
6738 * @param pVM Pointer to the VM.
6739 * @param pVCpu Pointer to the VMCPU.
6740 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
6741 * out-of-sync. Make sure to update the required fields
6742 * before using them.
6743 */
6744static int hmR0VmxCheckForceFlags(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6745{
6746 Assert(VMMRZCallRing3IsEnabled(pVCpu));
6747
6748 if ( VM_FF_IS_PENDING(pVM, !pVCpu->hm.s.fSingleInstruction
6749 ? VM_FF_HP_R0_PRE_HM_MASK : VM_FF_HP_R0_PRE_HM_STEP_MASK)
6750 || VMCPU_FF_IS_PENDING(pVCpu, !pVCpu->hm.s.fSingleInstruction
6751 ? VMCPU_FF_HP_R0_PRE_HM_MASK : VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
6752 {
6753 /* We need the control registers now, make sure the guest-CPU context is updated. */
6754 int rc3 = hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6755 AssertRCReturn(rc3, rc3);
6756
6757 /* Pending HM CR3 sync. */
6758 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
6759 {
6760 int rc2 = PGMUpdateCR3(pVCpu, pMixedCtx->cr3);
6761 AssertMsgReturn(rc2 == VINF_SUCCESS || rc2 == VINF_PGM_SYNC_CR3,
6762 ("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS);
6763 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6764 }
6765
6766 /* Pending HM PAE PDPEs. */
6767 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
6768 {
6769 PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6770 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6771 }
6772
6773 /* Pending PGM C3 sync. */
6774 if (VMCPU_FF_IS_PENDING(pVCpu,VMCPU_FF_PGM_SYNC_CR3 | VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
6775 {
6776 int rc2 = PGMSyncCR3(pVCpu, pMixedCtx->cr0, pMixedCtx->cr3, pMixedCtx->cr4,
6777 VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
6778 if (rc2 != VINF_SUCCESS)
6779 {
6780 AssertRC(rc2);
6781 Log4(("hmR0VmxCheckForceFlags: PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", rc2));
6782 return rc2;
6783 }
6784 }
6785
6786 /* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
6787 if ( VM_FF_IS_PENDING(pVM, VM_FF_HM_TO_R3_MASK)
6788 || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
6789 {
6790 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
6791 int rc2 = RT_UNLIKELY(VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_NO_MEMORY : VINF_EM_RAW_TO_R3;
6792 Log4(("hmR0VmxCheckForceFlags: HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc2));
6793 return rc2;
6794 }
6795
6796 /* Pending VM request packets, such as hardware interrupts. */
6797 if ( VM_FF_IS_PENDING(pVM, VM_FF_REQUEST)
6798 || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_REQUEST))
6799 {
6800 Log4(("hmR0VmxCheckForceFlags: Pending VM request forcing us back to ring-3\n"));
6801 return VINF_EM_PENDING_REQUEST;
6802 }
6803
6804 /* Pending PGM pool flushes. */
6805 if (VM_FF_IS_PENDING(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
6806 {
6807 Log4(("hmR0VmxCheckForceFlags: PGM pool flush pending forcing us back to ring-3\n"));
6808 return VINF_PGM_POOL_FLUSH_PENDING;
6809 }
6810
6811 /* Pending DMA requests. */
6812 if (VM_FF_IS_PENDING(pVM, VM_FF_PDM_DMA))
6813 {
6814 Log4(("hmR0VmxCheckForceFlags: Pending DMA request forcing us back to ring-3\n"));
6815 return VINF_EM_RAW_TO_R3;
6816 }
6817 }
6818
6819 return VINF_SUCCESS;
6820}
6821
6822
6823/**
6824 * Converts any TRPM trap into a pending HM event. This is typically used when
6825 * entering from ring-3 (not longjmp returns).
6826 *
6827 * @param pVCpu Pointer to the VMCPU.
6828 */
6829static void hmR0VmxTrpmTrapToPendingEvent(PVMCPU pVCpu)
6830{
6831 Assert(TRPMHasTrap(pVCpu));
6832 Assert(!pVCpu->hm.s.Event.fPending);
6833
6834 uint8_t uVector;
6835 TRPMEVENT enmTrpmEvent;
6836 RTGCUINT uErrCode;
6837 RTGCUINTPTR GCPtrFaultAddress;
6838 uint8_t cbInstr;
6839
6840 int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr);
6841 AssertRC(rc);
6842
6843 /* Refer Intel spec. 24.8.3 "VM-entry Controls for Event Injection" for the format of u32IntInfo. */
6844 uint32_t u32IntInfo = uVector | VMX_EXIT_INTERRUPTION_INFO_VALID;
6845 if (enmTrpmEvent == TRPM_TRAP)
6846 {
6847 switch (uVector)
6848 {
6849 case X86_XCPT_NMI:
6850 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
6851 break;
6852
6853 case X86_XCPT_BP:
6854 case X86_XCPT_OF:
6855 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
6856 break;
6857
6858 case X86_XCPT_PF:
6859 case X86_XCPT_DF:
6860 case X86_XCPT_TS:
6861 case X86_XCPT_NP:
6862 case X86_XCPT_SS:
6863 case X86_XCPT_GP:
6864 case X86_XCPT_AC:
6865 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
6866 /* no break! */
6867 default:
6868 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
6869 break;
6870 }
6871 }
6872 else if (enmTrpmEvent == TRPM_HARDWARE_INT)
6873 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
6874 else if (enmTrpmEvent == TRPM_SOFTWARE_INT)
6875 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
6876 else
6877 AssertMsgFailed(("Invalid TRPM event type %d\n", enmTrpmEvent));
6878
6879 rc = TRPMResetTrap(pVCpu);
6880 AssertRC(rc);
6881 Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n",
6882 u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress));
6883
6884 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress);
6885 STAM_COUNTER_DEC(&pVCpu->hm.s.StatInjectPendingReflect);
6886}
6887
6888
6889/**
6890 * Converts any pending HM event into a TRPM trap. Typically used when leaving
6891 * VT-x to execute any instruction.
6892 *
6893 * @param pvCpu Pointer to the VMCPU.
6894 */
6895static void hmR0VmxPendingEventToTrpmTrap(PVMCPU pVCpu)
6896{
6897 Assert(pVCpu->hm.s.Event.fPending);
6898
6899 uint32_t uVectorType = VMX_IDT_VECTORING_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
6900 uint32_t uVector = VMX_IDT_VECTORING_INFO_VECTOR(pVCpu->hm.s.Event.u64IntInfo);
6901 bool fErrorCodeValid = VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVCpu->hm.s.Event.u64IntInfo);
6902 uint32_t uErrorCode = pVCpu->hm.s.Event.u32ErrCode;
6903
6904 /* If a trap was already pending, we did something wrong! */
6905 Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo */, NULL /* pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
6906
6907 TRPMEVENT enmTrapType;
6908 switch (uVectorType)
6909 {
6910 case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT:
6911 enmTrapType = TRPM_HARDWARE_INT;
6912 break;
6913
6914 case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
6915 enmTrapType = TRPM_SOFTWARE_INT;
6916 break;
6917
6918 case VMX_IDT_VECTORING_INFO_TYPE_NMI:
6919 case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
6920 case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT: /* #BP and #OF */
6921 case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT:
6922 enmTrapType = TRPM_TRAP;
6923 break;
6924
6925 default:
6926 AssertMsgFailed(("Invalid trap type %#x\n", uVectorType));
6927 enmTrapType = TRPM_32BIT_HACK;
6928 break;
6929 }
6930
6931 Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType));
6932
6933 int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
6934 AssertRC(rc);
6935
6936 if (fErrorCodeValid)
6937 TRPMSetErrorCode(pVCpu, uErrorCode);
6938
6939 if ( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
6940 && uVector == X86_XCPT_PF)
6941 {
6942 TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
6943 }
6944 else if ( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
6945 || uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
6946 || uVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT)
6947 {
6948 AssertMsg( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
6949 || (uVector == X86_XCPT_BP || uVector == X86_XCPT_OF),
6950 ("Invalid vector: uVector=%#x uVectorType=%#x\n", uVector, uVectorType));
6951 TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
6952 }
6953 pVCpu->hm.s.Event.fPending = false;
6954}
6955
6956
6957/**
6958 * Does the necessary state syncing before returning to ring-3 for any reason
6959 * (longjmp, preemption, voluntary exits to ring-3) from VT-x.
6960 *
6961 * @returns VBox status code.
6962 * @param pVM Pointer to the VM.
6963 * @param pVCpu Pointer to the VMCPU.
6964 * @param pMixedCtx Pointer to the guest-CPU context. The data may
6965 * be out-of-sync. Make sure to update the required
6966 * fields before using them.
6967 * @param fSaveGuestState Whether to save the guest state or not.
6968 *
6969 * @remarks No-long-jmp zone!!!
6970 */
6971static int hmR0VmxLeave(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fSaveGuestState)
6972{
6973 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
6974 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
6975
6976 RTCPUID idCpu = RTMpCpuId();
6977 Log4Func(("HostCpuId=%u\n", idCpu));
6978
6979 /*
6980 * !!! IMPORTANT !!!
6981 * If you modify code here, make sure to check whether hmR0VmxCallRing3Callback() needs to be updated too.
6982 */
6983
6984 /* Save the guest state if necessary. */
6985 if ( fSaveGuestState
6986 && HMVMXCPU_GST_VALUE(pVCpu) != HMVMX_UPDATED_GUEST_ALL)
6987 {
6988 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
6989 AssertRCReturn(rc, rc);
6990 Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
6991 }
6992
6993 /* Restore host FPU state if necessary and resync on next R0 reentry .*/
6994 if (CPUMIsGuestFPUStateActive(pVCpu))
6995 {
6996 /* We shouldn't reload CR0 without saving it first. */
6997 if (!fSaveGuestState)
6998 {
6999 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
7000 AssertRCReturn(rc, rc);
7001 }
7002 CPUMR0SaveGuestFPU(pVM, pVCpu, pMixedCtx);
7003 Assert(!CPUMIsGuestFPUStateActive(pVCpu));
7004 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
7005 }
7006
7007 /* Restore host debug registers if necessary and resync on next R0 reentry. */
7008#ifdef VBOX_STRICT
7009 if (CPUMIsHyperDebugStateActive(pVCpu))
7010 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT);
7011#endif
7012 if (CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */))
7013 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
7014 Assert(!CPUMIsGuestDebugStateActive(pVCpu) && !CPUMIsGuestDebugStateActivePending(pVCpu));
7015 Assert(!CPUMIsHyperDebugStateActive(pVCpu) && !CPUMIsHyperDebugStateActivePending(pVCpu));
7016
7017#if HC_ARCH_BITS == 64
7018 /* Restore host-state bits that VT-x only restores partially. */
7019 if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
7020 && (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
7021 {
7022 Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags, idCpu));
7023 VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
7024 }
7025 pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
7026#endif
7027
7028#if HC_ARCH_BITS == 64
7029 /* Restore the lazy host MSRs as we're leaving VT-x context. */
7030 if ( pVM->hm.s.fAllow64BitGuests
7031 && pVCpu->hm.s.vmx.fLazyMsrs)
7032 {
7033 /* We shouldn't reload the guest MSRs without saving it first. */
7034 if (!fSaveGuestState)
7035 {
7036 int rc = hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
7037 AssertRCReturn(rc, rc);
7038 }
7039 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS));
7040 hmR0VmxLazyRestoreHostMsrs(pVCpu);
7041 Assert(!pVCpu->hm.s.vmx.fLazyMsrs);
7042 }
7043#endif
7044
7045 /* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
7046 pVCpu->hm.s.vmx.fUpdatedHostMsrs = false;
7047
7048 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
7049 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatLoadGuestState);
7050 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit1);
7051 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit2);
7052 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO);
7053 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx);
7054 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi);
7055 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
7056
7057 VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
7058
7059 /** @todo This partially defeats the purpose of having preemption hooks.
7060 * The problem is, deregistering the hooks should be moved to a place that
7061 * lasts until the EMT is about to be destroyed not everytime while leaving HM
7062 * context.
7063 */
7064 if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_ACTIVE)
7065 {
7066 int rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
7067 AssertRCReturn(rc, rc);
7068
7069 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
7070 Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu));
7071 }
7072 Assert(!(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_LAUNCHED));
7073 NOREF(idCpu);
7074
7075 return VINF_SUCCESS;
7076}
7077
7078
7079/**
7080 * Leaves the VT-x session.
7081 *
7082 * @returns VBox status code.
7083 * @param pVM Pointer to the VM.
7084 * @param pVCpu Pointer to the VMCPU.
7085 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7086 * out-of-sync. Make sure to update the required fields
7087 * before using them.
7088 *
7089 * @remarks No-long-jmp zone!!!
7090 */
7091DECLINLINE(int) hmR0VmxLeaveSession(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7092{
7093 HM_DISABLE_PREEMPT_IF_NEEDED();
7094 HMVMX_ASSERT_CPU_SAFE();
7095 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
7096 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
7097
7098 /* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
7099 and done this from the VMXR0ThreadCtxCallback(). */
7100 if (!pVCpu->hm.s.fLeaveDone)
7101 {
7102 int rc2 = hmR0VmxLeave(pVM, pVCpu, pMixedCtx, true /* fSaveGuestState */);
7103 AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT_IF_NEEDED(), rc2);
7104 pVCpu->hm.s.fLeaveDone = true;
7105 }
7106 Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
7107
7108 /*
7109 * !!! IMPORTANT !!!
7110 * If you modify code here, make sure to check whether hmR0VmxCallRing3Callback() needs to be updated too.
7111 */
7112
7113 /* Deregister hook now that we've left HM context before re-enabling preemption. */
7114 /** @todo This is bad. Deregistering here means we need to VMCLEAR always
7115 * (longjmp/exit-to-r3) in VT-x which is not efficient. */
7116 if (VMMR0ThreadCtxHooksAreRegistered(pVCpu))
7117 VMMR0ThreadCtxHooksDeregister(pVCpu);
7118
7119 /* Leave HM context. This takes care of local init (term). */
7120 int rc = HMR0LeaveCpu(pVCpu);
7121
7122 HM_RESTORE_PREEMPT_IF_NEEDED();
7123
7124 return rc;
7125}
7126
7127
7128/**
7129 * Does the necessary state syncing before doing a longjmp to ring-3.
7130 *
7131 * @returns VBox status code.
7132 * @param pVM Pointer to the VM.
7133 * @param pVCpu Pointer to the VMCPU.
7134 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7135 * out-of-sync. Make sure to update the required fields
7136 * before using them.
7137 *
7138 * @remarks No-long-jmp zone!!!
7139 */
7140DECLINLINE(int) hmR0VmxLongJmpToRing3(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7141{
7142 return hmR0VmxLeaveSession(pVM, pVCpu, pMixedCtx);
7143}
7144
7145
7146/**
7147 * Take necessary actions before going back to ring-3.
7148 *
7149 * An action requires us to go back to ring-3. This function does the necessary
7150 * steps before we can safely return to ring-3. This is not the same as longjmps
7151 * to ring-3, this is voluntary and prepares the guest so it may continue
7152 * executing outside HM (recompiler/IEM).
7153 *
7154 * @returns VBox status code.
7155 * @param pVM Pointer to the VM.
7156 * @param pVCpu Pointer to the VMCPU.
7157 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7158 * out-of-sync. Make sure to update the required fields
7159 * before using them.
7160 * @param rcExit The reason for exiting to ring-3. Can be
7161 * VINF_VMM_UNKNOWN_RING3_CALL.
7162 */
7163static int hmR0VmxExitToRing3(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, int rcExit)
7164{
7165 Assert(pVM);
7166 Assert(pVCpu);
7167 Assert(pMixedCtx);
7168 HMVMX_ASSERT_PREEMPT_SAFE();
7169
7170 if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR))
7171 {
7172 VMXGetActivatedVmcs(&pVCpu->hm.s.vmx.LastError.u64VMCSPhys);
7173 pVCpu->hm.s.vmx.LastError.u32VMCSRevision = *(uint32_t *)pVCpu->hm.s.vmx.pvVmcs;
7174 pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
7175 /* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */
7176 }
7177
7178 /* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
7179 VMMRZCallRing3Disable(pVCpu);
7180 Log4(("hmR0VmxExitToRing3: pVCpu=%p idCpu=%RU32 rcExit=%d\n", pVCpu, pVCpu->idCpu, rcExit));
7181
7182 /* We need to do this only while truly exiting the "inner loop" back to ring-3 and -not- for any longjmp to ring3. */
7183 if (pVCpu->hm.s.Event.fPending)
7184 {
7185 hmR0VmxPendingEventToTrpmTrap(pVCpu);
7186 Assert(!pVCpu->hm.s.Event.fPending);
7187 }
7188
7189 /* Save guest state and restore host state bits. */
7190 int rc = hmR0VmxLeaveSession(pVM, pVCpu, pMixedCtx);
7191 AssertRCReturn(rc, rc);
7192 STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
7193
7194 /* Sync recompiler state. */
7195 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
7196 CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
7197 | CPUM_CHANGED_LDTR
7198 | CPUM_CHANGED_GDTR
7199 | CPUM_CHANGED_IDTR
7200 | CPUM_CHANGED_TR
7201 | CPUM_CHANGED_HIDDEN_SEL_REGS);
7202 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
7203 if ( pVM->hm.s.fNestedPaging
7204 && CPUMIsGuestPagingEnabledEx(pMixedCtx))
7205 {
7206 CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
7207 }
7208
7209 Assert(!pVCpu->hm.s.fClearTrapFlag);
7210
7211 /* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
7212 if (rcExit != VINF_EM_RAW_INTERRUPT)
7213 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
7214
7215 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
7216
7217 /* We do -not- want any longjmp notifications after this! We must return to ring-3 ASAP. */
7218 VMMRZCallRing3RemoveNotification(pVCpu);
7219 VMMRZCallRing3Enable(pVCpu);
7220
7221 return rc;
7222}
7223
7224
7225/**
7226 * VMMRZCallRing3() callback wrapper which saves the guest state before we
7227 * longjump to ring-3 and possibly get preempted.
7228 *
7229 * @returns VBox status code.
7230 * @param pVCpu Pointer to the VMCPU.
7231 * @param enmOperation The operation causing the ring-3 longjump.
7232 * @param pvUser Opaque pointer to the guest-CPU context. The data
7233 * may be out-of-sync. Make sure to update the required
7234 * fields before using them.
7235 */
7236DECLCALLBACK(int) hmR0VmxCallRing3Callback(PVMCPU pVCpu, VMMCALLRING3 enmOperation, void *pvUser)
7237{
7238 if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION)
7239 {
7240 /*
7241 * !!! IMPORTANT !!!
7242 * If you modify code here, make sure to check whether hmR0VmxLeave() and hmR0VmxLeaveSession() needs
7243 * to be updated too. This is a stripped down version which gets out ASAP trying to not trigger any assertion.
7244 */
7245 VMMRZCallRing3RemoveNotification(pVCpu);
7246 VMMRZCallRing3Disable(pVCpu);
7247 HM_DISABLE_PREEMPT_IF_NEEDED();
7248
7249 PVM pVM = pVCpu->CTX_SUFF(pVM);
7250 if (CPUMIsGuestFPUStateActive(pVCpu))
7251 CPUMR0SaveGuestFPU(pVM, pVCpu, (PCPUMCTX)pvUser);
7252
7253 CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
7254
7255#if HC_ARCH_BITS == 64
7256 /* Restore host-state bits that VT-x only restores partially. */
7257 if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
7258 && (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
7259 VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
7260 pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
7261
7262 /* Restore the lazy host MSRs as we're leaving VT-x context. */
7263 if ( pVM->hm.s.fAllow64BitGuests
7264 && pVCpu->hm.s.vmx.fLazyMsrs)
7265 {
7266 hmR0VmxLazyRestoreHostMsrs(pVCpu);
7267 }
7268#endif
7269 /* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
7270 pVCpu->hm.s.vmx.fUpdatedHostMsrs = false;
7271 VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
7272 if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_ACTIVE)
7273 {
7274 VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
7275 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
7276 }
7277
7278 if (VMMR0ThreadCtxHooksAreRegistered(pVCpu))
7279 VMMR0ThreadCtxHooksDeregister(pVCpu);
7280
7281 HMR0LeaveCpu(pVCpu);
7282 HM_RESTORE_PREEMPT_IF_NEEDED();
7283 return VINF_SUCCESS;
7284 }
7285
7286 Assert(pVCpu);
7287 Assert(pvUser);
7288 Assert(VMMRZCallRing3IsEnabled(pVCpu));
7289 HMVMX_ASSERT_PREEMPT_SAFE();
7290
7291 VMMRZCallRing3Disable(pVCpu);
7292 Assert(VMMR0IsLogFlushDisabled(pVCpu));
7293
7294 Log4(("hmR0VmxCallRing3Callback->hmR0VmxLongJmpToRing3 pVCpu=%p idCpu=%RU32\n enmOperation=%d", pVCpu, pVCpu->idCpu,
7295 enmOperation));
7296
7297 int rc = hmR0VmxLongJmpToRing3(pVCpu->CTX_SUFF(pVM), pVCpu, (PCPUMCTX)pvUser);
7298 AssertRCReturn(rc, rc);
7299
7300 VMMRZCallRing3Enable(pVCpu);
7301 return VINF_SUCCESS;
7302}
7303
7304
7305/**
7306 * Sets the interrupt-window exiting control in the VMCS which instructs VT-x to
7307 * cause a VM-exit as soon as the guest is in a state to receive interrupts.
7308 *
7309 * @param pVCpu Pointer to the VMCPU.
7310 */
7311DECLINLINE(void) hmR0VmxSetIntWindowExitVmcs(PVMCPU pVCpu)
7312{
7313 if (RT_LIKELY(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT))
7314 {
7315 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT))
7316 {
7317 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT;
7318 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7319 AssertRC(rc);
7320 Log4(("Setup interrupt-window exiting\n"));
7321 }
7322 } /* else we will deliver interrupts whenever the guest exits next and is in a state to receive events. */
7323}
7324
7325
7326/**
7327 * Clears the interrupt-window exiting control in the VMCS.
7328 *
7329 * @param pVCpu Pointer to the VMCPU.
7330 */
7331DECLINLINE(void) hmR0VmxClearIntWindowExitVmcs(PVMCPU pVCpu)
7332{
7333 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT);
7334 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT;
7335 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7336 AssertRC(rc);
7337 Log4(("Cleared interrupt-window exiting\n"));
7338}
7339
7340
7341/**
7342 * Sets the NMI-window exiting control in the VMCS which instructs VT-x to
7343 * cause a VM-exit as soon as the guest is in a state to receive NMIs.
7344 *
7345 * @param pVCpu Pointer to the VMCPU.
7346 */
7347DECLINLINE(void) hmR0VmxSetNmiWindowExitVmcs(PVMCPU pVCpu)
7348{
7349 if (RT_LIKELY(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT))
7350 {
7351 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT))
7352 {
7353 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT;
7354 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7355 AssertRC(rc);
7356 Log4(("Setup NMI-window exiting\n"));
7357 }
7358 } /* else we will deliver NMIs whenever we VM-exit next, even possibly nesting NMIs. Can't be helped on ancient CPUs. */
7359}
7360
7361
7362/**
7363 * Clears the NMI-window exiting control in the VMCS.
7364 *
7365 * @param pVCpu Pointer to the VMCPU.
7366 */
7367DECLINLINE(void) hmR0VmxClearNmiWindowExitVmcs(PVMCPU pVCpu)
7368{
7369 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT);
7370 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT;
7371 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7372 AssertRC(rc);
7373 Log4(("Cleared NMI-window exiting\n"));
7374}
7375
7376
7377/**
7378 * Evaluates the event to be delivered to the guest and sets it as the pending
7379 * event.
7380 *
7381 * @param pVCpu Pointer to the VMCPU.
7382 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7383 * out-of-sync. Make sure to update the required fields
7384 * before using them.
7385 */
7386static void hmR0VmxEvaluatePendingEvent(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7387{
7388 Assert(!pVCpu->hm.s.Event.fPending);
7389
7390 /* Get the current interruptibility-state of the guest and then figure out what can be injected. */
7391 uint32_t const uIntrState = hmR0VmxGetGuestIntrState(pVCpu, pMixedCtx);
7392 bool const fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7393 bool const fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7394 bool const fBlockNmi = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI);
7395
7396 Assert(!fBlockSti || HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS));
7397 Assert(!(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
7398 Assert(!fBlockSti || pMixedCtx->eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
7399 Assert(!TRPMHasTrap(pVCpu));
7400
7401 /*
7402 * Toggling of interrupt force-flags here is safe since we update TRPM on premature exits
7403 * to ring-3 before executing guest code, see hmR0VmxExitToRing3(). We must NOT restore these force-flags.
7404 */
7405 /** @todo SMI. SMIs take priority over NMIs. */
7406 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_NMI)) /* NMI. NMIs take priority over regular interrupts. */
7407 {
7408 /* On some CPUs block-by-STI also blocks NMIs. See Intel spec. 26.3.1.5 "Checks On Guest Non-Register State". */
7409 if ( !fBlockNmi
7410 && !fBlockSti
7411 && !fBlockMovSS)
7412 {
7413 Log4(("Pending NMI vcpu[%RU32]\n", pVCpu->idCpu));
7414 uint32_t u32IntInfo = X86_XCPT_NMI | VMX_EXIT_INTERRUPTION_INFO_VALID;
7415 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7416
7417 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7418 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
7419 }
7420 else
7421 hmR0VmxSetNmiWindowExitVmcs(pVCpu);
7422 }
7423 /*
7424 * Check if the guest can receive external interrupts (PIC/APIC). Once we do PDMGetInterrupt() we -must- deliver
7425 * the interrupt ASAP. We must not execute any guest code until we inject the interrupt.
7426 */
7427 else if ( VMCPU_FF_IS_PENDING(pVCpu, (VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC))
7428 && !pVCpu->hm.s.fSingleInstruction)
7429 {
7430 int rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
7431 AssertRC(rc);
7432 bool const fBlockInt = !(pMixedCtx->eflags.u32 & X86_EFL_IF);
7433 if ( !fBlockInt
7434 && !fBlockSti
7435 && !fBlockMovSS)
7436 {
7437 uint8_t u8Interrupt;
7438 rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
7439 if (RT_SUCCESS(rc))
7440 {
7441 Log4(("Pending interrupt vcpu[%RU32] u8Interrupt=%#x \n", pVCpu->idCpu, u8Interrupt));
7442 uint32_t u32IntInfo = u8Interrupt | VMX_EXIT_INTERRUPTION_INFO_VALID;
7443 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7444
7445 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrfaultAddress */);
7446 }
7447 else
7448 {
7449 /** @todo Does this actually happen? If not turn it into an assertion. */
7450 Assert(!VMCPU_FF_IS_PENDING(pVCpu, (VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC)));
7451 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
7452 }
7453 }
7454 else
7455 hmR0VmxSetIntWindowExitVmcs(pVCpu);
7456 }
7457}
7458
7459
7460/**
7461 * Sets a pending-debug exception to be delivered to the guest if the guest is
7462 * single-stepping.
7463 *
7464 * @param pVCpu Pointer to the VMCPU.
7465 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7466 * out-of-sync. Make sure to update the required fields
7467 * before using them.
7468 */
7469DECLINLINE(void) hmR0VmxSetPendingDebugXcpt(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7470{
7471 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS));
7472 if (pMixedCtx->eflags.Bits.u1TF) /* We don't have any IA32_DEBUGCTL MSR for guests. Treat as all bits 0. */
7473 {
7474 int rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, VMX_VMCS_GUEST_DEBUG_EXCEPTIONS_BS);
7475 AssertRC(rc);
7476 }
7477}
7478
7479
7480/**
7481 * Injects any pending events into the guest if the guest is in a state to
7482 * receive them.
7483 *
7484 * @returns VBox status code (informational status codes included).
7485 * @param pVCpu Pointer to the VMCPU.
7486 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7487 * out-of-sync. Make sure to update the required fields
7488 * before using them.
7489 */
7490static int hmR0VmxInjectPendingEvent(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7491{
7492 HMVMX_ASSERT_PREEMPT_SAFE();
7493 Assert(VMMRZCallRing3IsEnabled(pVCpu));
7494
7495 /* Get the current interruptibility-state of the guest and then figure out what can be injected. */
7496 uint32_t uIntrState = hmR0VmxGetGuestIntrState(pVCpu, pMixedCtx);
7497 bool fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7498 bool fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7499
7500 Assert(!fBlockSti || HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS));
7501 Assert(!(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
7502 Assert(!fBlockSti || pMixedCtx->eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
7503 Assert(!TRPMHasTrap(pVCpu));
7504
7505 int rc = VINF_SUCCESS;
7506 if (pVCpu->hm.s.Event.fPending)
7507 {
7508 /*
7509 * Clear any interrupt-window exiting control if we're going to inject an interrupt. Saves one extra
7510 * VM-exit in situations where we previously setup interrupt-window exiting but got other VM-exits and
7511 * ended up enabling interrupts outside VT-x.
7512 */
7513 uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
7514 if ( (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT)
7515 && uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
7516 {
7517 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT);
7518 hmR0VmxClearIntWindowExitVmcs(pVCpu);
7519 }
7520
7521#ifdef VBOX_STRICT
7522 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
7523 {
7524 bool const fBlockInt = !(pMixedCtx->eflags.u32 & X86_EFL_IF);
7525 Assert(!fBlockInt);
7526 Assert(!fBlockSti);
7527 Assert(!fBlockMovSS);
7528 }
7529 else if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
7530 {
7531 bool fBlockNmi = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI);
7532 Assert(!fBlockSti);
7533 Assert(!fBlockMovSS);
7534 Assert(!fBlockNmi);
7535 }
7536#endif
7537 Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64 Type=%#x\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo,
7538 (uint8_t)uIntType));
7539 rc = hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, pVCpu->hm.s.Event.u64IntInfo, pVCpu->hm.s.Event.cbInstr,
7540 pVCpu->hm.s.Event.u32ErrCode, pVCpu->hm.s.Event.GCPtrFaultAddress, &uIntrState);
7541 AssertRCReturn(rc, rc);
7542
7543 /* Update the interruptibility-state as it could have been changed by
7544 hmR0VmxInjectEventVmcs() (e.g. real-on-v86 guest injecting software interrupts) */
7545 fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7546 fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7547
7548#ifdef VBOX_WITH_STATISTICS
7549 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
7550 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
7551 else
7552 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
7553#endif
7554 }
7555
7556 /* Deliver pending debug exception if the guest is single-stepping. Evaluate and set the BS bit. */
7557 if ( fBlockSti
7558 || fBlockMovSS)
7559 {
7560 if ( !pVCpu->hm.s.fSingleInstruction
7561 && !DBGFIsStepping(pVCpu))
7562 {
7563 /*
7564 * The pending-debug exceptions field is cleared on all VM-exits except VMX_EXIT_TPR_BELOW_THRESHOLD,
7565 * VMX_EXIT_MTF, VMX_EXIT_APIC_WRITE and VMX_EXIT_VIRTUALIZED_EOI.
7566 * See Intel spec. 27.3.4 "Saving Non-Register State".
7567 */
7568 int rc2 = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
7569 AssertRCReturn(rc2, rc2);
7570 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
7571 }
7572 else if (pMixedCtx->eflags.Bits.u1TF)
7573 {
7574 /*
7575 * We are single-stepping in the hypervisor debugger using EFLAGS.TF. Clear interrupt inhibition as setting the
7576 * BS bit would mean delivering a #DB to the guest upon VM-entry when it shouldn't be.
7577 */
7578 Assert(!(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG));
7579 uIntrState = 0;
7580 }
7581 }
7582
7583 /*
7584 * There's no need to clear the VM entry-interruption information field here if we're not injecting anything.
7585 * VT-x clears the valid bit on every VM-exit. See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
7586 */
7587 int rc2 = hmR0VmxLoadGuestIntrState(pVCpu, uIntrState);
7588 AssertRC(rc2);
7589
7590 Assert(rc == VINF_SUCCESS || rc == VINF_EM_RESET);
7591 NOREF(fBlockMovSS); NOREF(fBlockSti);
7592 return rc;
7593}
7594
7595
7596/**
7597 * Sets an invalid-opcode (#UD) exception as pending-for-injection into the VM.
7598 *
7599 * @param pVCpu Pointer to the VMCPU.
7600 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7601 * out-of-sync. Make sure to update the required fields
7602 * before using them.
7603 */
7604DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7605{
7606 NOREF(pMixedCtx);
7607 uint32_t u32IntInfo = X86_XCPT_UD | VMX_EXIT_INTERRUPTION_INFO_VALID;
7608 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7609}
7610
7611
7612/**
7613 * Injects a double-fault (#DF) exception into the VM.
7614 *
7615 * @returns VBox status code (informational status code included).
7616 * @param pVCpu Pointer to the VMCPU.
7617 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7618 * out-of-sync. Make sure to update the required fields
7619 * before using them.
7620 */
7621DECLINLINE(int) hmR0VmxInjectXcptDF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t *puIntrState)
7622{
7623 uint32_t u32IntInfo = X86_XCPT_DF | VMX_EXIT_INTERRUPTION_INFO_VALID;
7624 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7625 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7626 return hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */,
7627 puIntrState);
7628}
7629
7630
7631/**
7632 * Sets a debug (#DB) exception as pending-for-injection into the VM.
7633 *
7634 * @param pVCpu Pointer to the VMCPU.
7635 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7636 * out-of-sync. Make sure to update the required fields
7637 * before using them.
7638 */
7639DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7640{
7641 NOREF(pMixedCtx);
7642 uint32_t u32IntInfo = X86_XCPT_DB | VMX_EXIT_INTERRUPTION_INFO_VALID;
7643 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7644 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7645}
7646
7647
7648/**
7649 * Sets an overflow (#OF) exception as pending-for-injection into the VM.
7650 *
7651 * @param pVCpu Pointer to the VMCPU.
7652 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7653 * out-of-sync. Make sure to update the required fields
7654 * before using them.
7655 * @param cbInstr The value of RIP that is to be pushed on the guest
7656 * stack.
7657 */
7658DECLINLINE(void) hmR0VmxSetPendingXcptOF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t cbInstr)
7659{
7660 NOREF(pMixedCtx);
7661 uint32_t u32IntInfo = X86_XCPT_OF | VMX_EXIT_INTERRUPTION_INFO_VALID;
7662 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7663 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7664}
7665
7666
7667/**
7668 * Injects a general-protection (#GP) fault into the VM.
7669 *
7670 * @returns VBox status code (informational status code included).
7671 * @param pVCpu Pointer to the VMCPU.
7672 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7673 * out-of-sync. Make sure to update the required fields
7674 * before using them.
7675 * @param fErrorCodeValid Whether the error code is valid (depends on the CPU
7676 * mode, i.e. in real-mode it's not valid).
7677 * @param u32ErrorCode The error code associated with the #GP.
7678 */
7679DECLINLINE(int) hmR0VmxInjectXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fErrorCodeValid, uint32_t u32ErrorCode,
7680 uint32_t *puIntrState)
7681{
7682 uint32_t u32IntInfo = X86_XCPT_GP | VMX_EXIT_INTERRUPTION_INFO_VALID;
7683 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7684 if (fErrorCodeValid)
7685 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7686 return hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, u32IntInfo, 0 /* cbInstr */, u32ErrorCode, 0 /* GCPtrFaultAddress */,
7687 puIntrState);
7688}
7689
7690
7691/**
7692 * Sets a general-protection (#GP) exception as pending-for-injection into the
7693 * VM.
7694 *
7695 * @param pVCpu Pointer to the VMCPU.
7696 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7697 * out-of-sync. Make sure to update the required fields
7698 * before using them.
7699 * @param u32ErrorCode The error code associated with the #GP.
7700 */
7701DECLINLINE(void) hmR0VmxSetPendingXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t u32ErrorCode)
7702{
7703 NOREF(pMixedCtx);
7704 uint32_t u32IntInfo = X86_XCPT_GP | VMX_EXIT_INTERRUPTION_INFO_VALID;
7705 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7706 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7707 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, u32ErrorCode, 0 /* GCPtrFaultAddress */);
7708}
7709
7710
7711/**
7712 * Sets a software interrupt (INTn) as pending-for-injection into the VM.
7713 *
7714 * @param pVCpu Pointer to the VMCPU.
7715 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7716 * out-of-sync. Make sure to update the required fields
7717 * before using them.
7718 * @param uVector The software interrupt vector number.
7719 * @param cbInstr The value of RIP that is to be pushed on the guest
7720 * stack.
7721 */
7722DECLINLINE(void) hmR0VmxSetPendingIntN(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint16_t uVector, uint32_t cbInstr)
7723{
7724 NOREF(pMixedCtx);
7725 uint32_t u32IntInfo = uVector | VMX_EXIT_INTERRUPTION_INFO_VALID;
7726 if ( uVector == X86_XCPT_BP
7727 || uVector == X86_XCPT_OF)
7728 {
7729 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7730 }
7731 else
7732 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7733 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7734}
7735
7736
7737/**
7738 * Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's
7739 * stack.
7740 *
7741 * @returns VBox status code (information status code included).
7742 * @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault.
7743 * @param pVM Pointer to the VM.
7744 * @param pMixedCtx Pointer to the guest-CPU context.
7745 * @param uValue The value to push to the guest stack.
7746 */
7747DECLINLINE(int) hmR0VmxRealModeGuestStackPush(PVM pVM, PCPUMCTX pMixedCtx, uint16_t uValue)
7748{
7749 /*
7750 * The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in
7751 * virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
7752 * See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound".
7753 */
7754 if (pMixedCtx->sp == 1)
7755 return VINF_EM_RESET;
7756 pMixedCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */
7757 int rc = PGMPhysSimpleWriteGCPhys(pVM, pMixedCtx->ss.u64Base + pMixedCtx->sp, &uValue, sizeof(uint16_t));
7758 AssertRCReturn(rc, rc);
7759 return rc;
7760}
7761
7762
7763/**
7764 * Injects an event into the guest upon VM-entry by updating the relevant fields
7765 * in the VM-entry area in the VMCS.
7766 *
7767 * @returns VBox status code (informational error codes included).
7768 * @retval VINF_SUCCESS if the event is successfully injected into the VMCS.
7769 * @retval VINF_EM_RESET if event injection resulted in a triple-fault.
7770 *
7771 * @param pVCpu Pointer to the VMCPU.
7772 * @param pMixedCtx Pointer to the guest-CPU context. The data may
7773 * be out-of-sync. Make sure to update the required
7774 * fields before using them.
7775 * @param u64IntInfo The VM-entry interruption-information field.
7776 * @param cbInstr The VM-entry instruction length in bytes (for
7777 * software interrupts, exceptions and privileged
7778 * software exceptions).
7779 * @param u32ErrCode The VM-entry exception error code.
7780 * @param GCPtrFaultAddress The page-fault address for #PF exceptions.
7781 * @param puIntrState Pointer to the current guest interruptibility-state.
7782 * This interruptibility-state will be updated if
7783 * necessary. This cannot not be NULL.
7784 *
7785 * @remarks Requires CR0!
7786 * @remarks No-long-jump zone!!!
7787 */
7788static int hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint64_t u64IntInfo, uint32_t cbInstr,
7789 uint32_t u32ErrCode, RTGCUINTREG GCPtrFaultAddress, uint32_t *puIntrState)
7790{
7791 /* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */
7792 AssertMsg(u64IntInfo >> 32 == 0, ("%#RX64\n", u64IntInfo));
7793 Assert(puIntrState);
7794 uint32_t u32IntInfo = (uint32_t)u64IntInfo;
7795
7796 uint32_t const uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(u32IntInfo);
7797 uint32_t const uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(u32IntInfo);
7798
7799#ifdef VBOX_STRICT
7800 /* Validate the error-code-valid bit for hardware exceptions. */
7801 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT)
7802 {
7803 switch (uVector)
7804 {
7805 case X86_XCPT_PF:
7806 case X86_XCPT_DF:
7807 case X86_XCPT_TS:
7808 case X86_XCPT_NP:
7809 case X86_XCPT_SS:
7810 case X86_XCPT_GP:
7811 case X86_XCPT_AC:
7812 AssertMsg(VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(u32IntInfo),
7813 ("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector));
7814 /* fallthru */
7815 default:
7816 break;
7817 }
7818 }
7819#endif
7820
7821 /* Cannot inject an NMI when block-by-MOV SS is in effect. */
7822 Assert( uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI
7823 || !(*puIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS));
7824
7825 STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedIrqsR0[uVector & MASK_INJECT_IRQ_STAT]);
7826
7827 /* We require CR0 to check if the guest is in real-mode. */
7828 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
7829 AssertRCReturn(rc, rc);
7830
7831 /*
7832 * Hardware interrupts & exceptions cannot be delivered through the software interrupt redirection bitmap to the real
7833 * mode task in virtual-8086 mode. We must jump to the interrupt handler in the (real-mode) guest.
7834 * See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode" for interrupt & exception classes.
7835 * See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling.
7836 */
7837 if (CPUMIsGuestInRealModeEx(pMixedCtx))
7838 {
7839 PVM pVM = pVCpu->CTX_SUFF(pVM);
7840 if (!pVM->hm.s.vmx.fUnrestrictedGuest)
7841 {
7842 Assert(PDMVmmDevHeapIsEnabled(pVM));
7843 Assert(pVM->hm.s.vmx.pRealModeTSS);
7844
7845 /* We require RIP, RSP, RFLAGS, CS, IDTR. Save the required ones from the VMCS. */
7846 rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
7847 rc |= hmR0VmxSaveGuestTableRegs(pVCpu, pMixedCtx);
7848 rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
7849 AssertRCReturn(rc, rc);
7850 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP));
7851
7852 /* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */
7853 size_t const cbIdtEntry = sizeof(X86IDTR16);
7854 if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pMixedCtx->idtr.cbIdt)
7855 {
7856 /* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */
7857 if (uVector == X86_XCPT_DF)
7858 return VINF_EM_RESET;
7859 else if (uVector == X86_XCPT_GP)
7860 {
7861 /* If we're injecting a #GP with no valid IDT entry, inject a double-fault. */
7862 return hmR0VmxInjectXcptDF(pVCpu, pMixedCtx, puIntrState);
7863 }
7864
7865 /* If we're injecting an interrupt/exception with no valid IDT entry, inject a general-protection fault. */
7866 /* No error codes for exceptions in real-mode. See Intel spec. 20.1.4 "Interrupt and Exception Handling" */
7867 return hmR0VmxInjectXcptGP(pVCpu, pMixedCtx, false /* fErrCodeValid */, 0 /* u32ErrCode */, puIntrState);
7868 }
7869
7870 /* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */
7871 uint16_t uGuestIp = pMixedCtx->ip;
7872 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT)
7873 {
7874 Assert(uVector == X86_XCPT_BP || uVector == X86_XCPT_OF);
7875 /* #BP and #OF are both benign traps, we need to resume the next instruction. */
7876 uGuestIp = pMixedCtx->ip + (uint16_t)cbInstr;
7877 }
7878 else if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT)
7879 uGuestIp = pMixedCtx->ip + (uint16_t)cbInstr;
7880
7881 /* Get the code segment selector and offset from the IDT entry for the interrupt handler. */
7882 X86IDTR16 IdtEntry;
7883 RTGCPHYS GCPhysIdtEntry = (RTGCPHYS)pMixedCtx->idtr.pIdt + uVector * cbIdtEntry;
7884 rc = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry);
7885 AssertRCReturn(rc, rc);
7886
7887 /* Construct the stack frame for the interrupt/exception handler. */
7888 rc = hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, pMixedCtx->eflags.u32);
7889 rc |= hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, pMixedCtx->cs.Sel);
7890 rc |= hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, uGuestIp);
7891 AssertRCReturn(rc, rc);
7892
7893 /* Clear the required eflag bits and jump to the interrupt/exception handler. */
7894 if (rc == VINF_SUCCESS)
7895 {
7896 pMixedCtx->eflags.u32 &= ~(X86_EFL_IF | X86_EFL_TF | X86_EFL_RF | X86_EFL_AC);
7897 pMixedCtx->rip = IdtEntry.offSel;
7898 pMixedCtx->cs.Sel = IdtEntry.uSel;
7899 pMixedCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry;
7900 if ( uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT
7901 && uVector == X86_XCPT_PF)
7902 {
7903 pMixedCtx->cr2 = GCPtrFaultAddress;
7904 }
7905
7906 /* If any other guest-state bits are changed here, make sure to update
7907 hmR0VmxPreRunGuestCommitted() when thread-context hooks are used. */
7908 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS
7909 | HM_CHANGED_GUEST_RIP
7910 | HM_CHANGED_GUEST_RFLAGS
7911 | HM_CHANGED_GUEST_RSP);
7912
7913 /* We're clearing interrupts, which means no block-by-STI interrupt-inhibition. */
7914 if (*puIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
7915 {
7916 Assert( uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI
7917 && uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT);
7918 Log4(("Clearing inhibition due to STI.\n"));
7919 *puIntrState &= ~VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI;
7920 }
7921 Log4(("Injecting real-mode: u32IntInfo=%#x u32ErrCode=%#x instrlen=%#x\n", u32IntInfo, u32ErrCode, cbInstr));
7922
7923 /* The event has been truly dispatched. Mark it as no longer pending so we don't attempt to 'undo'
7924 it, if we are returning to ring-3 before executing guest code. */
7925 pVCpu->hm.s.Event.fPending = false;
7926 }
7927 Assert(rc == VINF_SUCCESS || rc == VINF_EM_RESET);
7928 return rc;
7929 }
7930 else
7931 {
7932 /*
7933 * For unrestricted execution enabled CPUs running real-mode guests, we must not set the deliver-error-code bit.
7934 * See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
7935 */
7936 u32IntInfo &= ~VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7937 }
7938 }
7939
7940 /* Validate. */
7941 Assert(VMX_EXIT_INTERRUPTION_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */
7942 Assert(!VMX_EXIT_INTERRUPTION_INFO_NMI_UNBLOCK_IRET(u32IntInfo)); /* Bit 12 MBZ. */
7943 Assert(!(u32IntInfo & 0x7ffff000)); /* Bits 30:12 MBZ. */
7944
7945 /* Inject. */
7946 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo);
7947 if (VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(u32IntInfo))
7948 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode);
7949 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr);
7950
7951 if ( VMX_EXIT_INTERRUPTION_INFO_TYPE(u32IntInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT
7952 && uVector == X86_XCPT_PF)
7953 {
7954 pMixedCtx->cr2 = GCPtrFaultAddress;
7955 }
7956
7957 Log4(("Injecting vcpu[%RU32] u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x pMixedCtx->uCR2=%#RX64\n", pVCpu->idCpu,
7958 u32IntInfo, u32ErrCode, cbInstr, pMixedCtx->cr2));
7959
7960 AssertRCReturn(rc, rc);
7961 return rc;
7962}
7963
7964
7965/**
7966 * Clears the interrupt-window exiting control in the VMCS and if necessary
7967 * clears the current event in the VMCS as well.
7968 *
7969 * @returns VBox status code.
7970 * @param pVCpu Pointer to the VMCPU.
7971 *
7972 * @remarks Use this function only to clear events that have not yet been
7973 * delivered to the guest but are injected in the VMCS!
7974 * @remarks No-long-jump zone!!!
7975 */
7976static void hmR0VmxClearEventVmcs(PVMCPU pVCpu)
7977{
7978 int rc;
7979 Log4Func(("vcpu[%d]\n", pVCpu->idCpu));
7980
7981 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT)
7982 {
7983 hmR0VmxClearIntWindowExitVmcs(pVCpu);
7984 Assert(!pVCpu->hm.s.Event.fPending);
7985 }
7986
7987 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT)
7988 {
7989 hmR0VmxClearNmiWindowExitVmcs(pVCpu);
7990 Assert(!pVCpu->hm.s.Event.fPending);
7991 }
7992
7993 if (!pVCpu->hm.s.Event.fPending)
7994 return;
7995
7996#ifdef VBOX_STRICT
7997 uint32_t u32EntryInfo;
7998 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
7999 AssertRC(rc);
8000 Assert(VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo));
8001#endif
8002
8003 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0);
8004 AssertRC(rc);
8005
8006 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, 0);
8007 AssertRC(rc);
8008
8009 /* We deliberately don't clear "hm.s.Event.fPending" here, it's taken
8010 care of in hmR0VmxExitToRing3() converting the pending event to TRPM. */
8011}
8012
8013
8014/**
8015 * Enters the VT-x session.
8016 *
8017 * @returns VBox status code.
8018 * @param pVM Pointer to the VM.
8019 * @param pVCpu Pointer to the VMCPU.
8020 * @param pCpu Pointer to the CPU info struct.
8021 */
8022VMMR0DECL(int) VMXR0Enter(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
8023{
8024 AssertPtr(pVM);
8025 AssertPtr(pVCpu);
8026 Assert(pVM->hm.s.vmx.fSupported);
8027 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8028 NOREF(pCpu); NOREF(pVM);
8029
8030 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8031 Assert(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE));
8032
8033#ifdef VBOX_STRICT
8034 /* Make sure we're in VMX root mode. */
8035 RTCCUINTREG u32HostCR4 = ASMGetCR4();
8036 if (!(u32HostCR4 & X86_CR4_VMXE))
8037 {
8038 LogRel(("VMXR0Enter: X86_CR4_VMXE bit in CR4 is not set!\n"));
8039 return VERR_VMX_X86_CR4_VMXE_CLEARED;
8040 }
8041#endif
8042
8043 /*
8044 * Load the VCPU's VMCS as the current (and active) one.
8045 */
8046 Assert(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_CLEAR);
8047 int rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
8048 if (RT_FAILURE(rc))
8049 return rc;
8050
8051 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE;
8052 pVCpu->hm.s.fLeaveDone = false;
8053 Log4Func(("Activated Vmcs. HostCpuId=%u\n", RTMpCpuId()));
8054
8055 return VINF_SUCCESS;
8056}
8057
8058
8059/**
8060 * The thread-context callback (only on platforms which support it).
8061 *
8062 * @param enmEvent The thread-context event.
8063 * @param pVCpu Pointer to the VMCPU.
8064 * @param fGlobalInit Whether global VT-x/AMD-V init. was used.
8065 * @thread EMT(pVCpu)
8066 */
8067VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPU pVCpu, bool fGlobalInit)
8068{
8069 NOREF(fGlobalInit);
8070
8071 switch (enmEvent)
8072 {
8073 case RTTHREADCTXEVENT_PREEMPTING:
8074 {
8075 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8076 Assert(VMMR0ThreadCtxHooksAreRegistered(pVCpu));
8077 VMCPU_ASSERT_EMT(pVCpu);
8078
8079 PVM pVM = pVCpu->CTX_SUFF(pVM);
8080 PCPUMCTX pMixedCtx = CPUMQueryGuestCtxPtr(pVCpu);
8081
8082 /* No longjmps (logger flushes, locks) in this fragile context. */
8083 VMMRZCallRing3Disable(pVCpu);
8084 Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId()));
8085
8086 /*
8087 * Restore host-state (FPU, debug etc.)
8088 */
8089 if (!pVCpu->hm.s.fLeaveDone)
8090 {
8091 /* Do -not- save guest-state here as we might already be in the middle of saving it (esp. bad if we are
8092 holding the PGM lock while saving the guest state (see hmR0VmxSaveGuestControlRegs()). */
8093 hmR0VmxLeave(pVM, pVCpu, pMixedCtx, false /* fSaveGuestState */);
8094 pVCpu->hm.s.fLeaveDone = true;
8095 }
8096
8097 /* Leave HM context, takes care of local init (term). */
8098 int rc = HMR0LeaveCpu(pVCpu);
8099 AssertRC(rc); NOREF(rc);
8100
8101 /* Restore longjmp state. */
8102 VMMRZCallRing3Enable(pVCpu);
8103 STAM_COUNTER_INC(&pVCpu->hm.s.StatPreemptPreempting);
8104 break;
8105 }
8106
8107 case RTTHREADCTXEVENT_RESUMED:
8108 {
8109 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8110 Assert(VMMR0ThreadCtxHooksAreRegistered(pVCpu));
8111 VMCPU_ASSERT_EMT(pVCpu);
8112
8113 /* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */
8114 VMMRZCallRing3Disable(pVCpu);
8115 Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId()));
8116
8117 /* Initialize the bare minimum state required for HM. This takes care of
8118 initializing VT-x if necessary (onlined CPUs, local init etc.) */
8119 int rc = HMR0EnterCpu(pVCpu);
8120 AssertRC(rc);
8121 Assert(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE));
8122
8123 /* Load the active VMCS as the current one. */
8124 if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_CLEAR)
8125 {
8126 rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
8127 AssertRC(rc); NOREF(rc);
8128 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE;
8129 Log4Func(("Resumed: Activated Vmcs. HostCpuId=%u\n", RTMpCpuId()));
8130 }
8131 pVCpu->hm.s.fLeaveDone = false;
8132
8133 /* Restore longjmp state. */
8134 VMMRZCallRing3Enable(pVCpu);
8135 break;
8136 }
8137
8138 default:
8139 break;
8140 }
8141}
8142
8143
8144/**
8145 * Saves the host state in the VMCS host-state.
8146 * Sets up the VM-exit MSR-load area.
8147 *
8148 * The CPU state will be loaded from these fields on every successful VM-exit.
8149 *
8150 * @returns VBox status code.
8151 * @param pVM Pointer to the VM.
8152 * @param pVCpu Pointer to the VMCPU.
8153 *
8154 * @remarks No-long-jump zone!!!
8155 */
8156static int hmR0VmxSaveHostState(PVM pVM, PVMCPU pVCpu)
8157{
8158 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8159
8160 if (!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT))
8161 return VINF_SUCCESS;
8162
8163 int rc = hmR0VmxSaveHostControlRegs(pVM, pVCpu);
8164 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostControlRegisters failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8165
8166 rc = hmR0VmxSaveHostSegmentRegs(pVM, pVCpu);
8167 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostSegmentRegisters failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8168
8169 rc = hmR0VmxSaveHostMsrs(pVM, pVCpu);
8170 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostMsrs failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8171
8172 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_HOST_CONTEXT);
8173 return rc;
8174}
8175
8176
8177/**
8178 * Saves the host state in the VMCS host-state.
8179 *
8180 * @returns VBox status code.
8181 * @param pVM Pointer to the VM.
8182 * @param pVCpu Pointer to the VMCPU.
8183 *
8184 * @remarks No-long-jump zone!!!
8185 */
8186VMMR0DECL(int) VMXR0SaveHostState(PVM pVM, PVMCPU pVCpu)
8187{
8188 AssertPtr(pVM);
8189 AssertPtr(pVCpu);
8190
8191 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8192
8193 /* Save the host state here while entering HM context. When thread-context hooks are used, we might get preempted
8194 and have to resave the host state but most of the time we won't be, so do it here before we disable interrupts. */
8195 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8196 return hmR0VmxSaveHostState(pVM, pVCpu);
8197}
8198
8199
8200/**
8201 * Loads the guest state into the VMCS guest-state area. The CPU state will be
8202 * loaded from these fields on every successful VM-entry.
8203 *
8204 * Sets up the VM-entry MSR-load and VM-exit MSR-store areas.
8205 * Sets up the VM-entry controls.
8206 * Sets up the appropriate VMX non-root function to execute guest code based on
8207 * the guest CPU mode.
8208 *
8209 * @returns VBox status code.
8210 * @param pVM Pointer to the VM.
8211 * @param pVCpu Pointer to the VMCPU.
8212 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8213 * out-of-sync. Make sure to update the required fields
8214 * before using them.
8215 *
8216 * @remarks No-long-jump zone!!!
8217 */
8218static int hmR0VmxLoadGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
8219{
8220 AssertPtr(pVM);
8221 AssertPtr(pVCpu);
8222 AssertPtr(pMixedCtx);
8223 HMVMX_ASSERT_PREEMPT_SAFE();
8224
8225 VMMRZCallRing3Disable(pVCpu);
8226 Assert(VMMR0IsLogFlushDisabled(pVCpu));
8227
8228 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8229
8230 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestState, x);
8231
8232 /* Determine real-on-v86 mode. */
8233 pVCpu->hm.s.vmx.RealMode.fRealOnV86Active = false;
8234 if ( !pVM->hm.s.vmx.fUnrestrictedGuest
8235 && CPUMIsGuestInRealModeEx(pMixedCtx))
8236 {
8237 pVCpu->hm.s.vmx.RealMode.fRealOnV86Active = true;
8238 }
8239
8240 /*
8241 * Load the guest-state into the VMCS.
8242 * Any ordering dependency among the sub-functions below must be explicitly stated using comments.
8243 * Ideally, assert that the cross-dependent bits are up-to-date at the point of using it.
8244 */
8245 int rc = hmR0VmxSetupVMRunHandler(pVCpu, pMixedCtx);
8246 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSetupVMRunHandler! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8247
8248 /* This needs to be done after hmR0VmxSetupVMRunHandler() as changing pfnStartVM may require VM-entry control updates. */
8249 rc = hmR0VmxLoadGuestEntryCtls(pVCpu, pMixedCtx);
8250 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestEntryCtls! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8251
8252 /* This needs to be done after hmR0VmxSetupVMRunHandler() as changing pfnStartVM may require VM-exit control updates. */
8253 rc = hmR0VmxLoadGuestExitCtls(pVCpu, pMixedCtx);
8254 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSetupExitCtls failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8255
8256 rc = hmR0VmxLoadGuestActivityState(pVCpu, pMixedCtx);
8257 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestActivityState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8258
8259 rc = hmR0VmxLoadGuestCR3AndCR4(pVCpu, pMixedCtx);
8260 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestCR3AndCR4: rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8261
8262 /* Assumes pMixedCtx->cr0 is up-to-date (strict builds require CR0 for segment register validation checks). */
8263 rc = hmR0VmxLoadGuestSegmentRegs(pVCpu, pMixedCtx);
8264 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestSegmentRegs: rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8265
8266 /* This needs to be done after hmR0VmxLoadGuestEntryCtls() and hmR0VmxLoadGuestExitCtls() as it may alter controls if we
8267 determine we don't have to swap EFER after all. */
8268 rc = hmR0VmxLoadGuestMsrs(pVCpu, pMixedCtx);
8269 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadSharedMsrs! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8270
8271 rc = hmR0VmxLoadGuestApicState(pVCpu, pMixedCtx);
8272 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestApicState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8273
8274 /*
8275 * Loading Rflags here is fine, even though Rflags.TF might depend on guest debug state (which is not loaded here).
8276 * It is re-evaluated and updated if necessary in hmR0VmxLoadSharedState().
8277 */
8278 rc = hmR0VmxLoadGuestRipRspRflags(pVCpu, pMixedCtx);
8279 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestRipRspRflags! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8280
8281 /* Clear any unused and reserved bits. */
8282 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR2);
8283
8284 VMMRZCallRing3Enable(pVCpu);
8285
8286 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestState, x);
8287 return rc;
8288}
8289
8290
8291/**
8292 * Loads the state shared between the host and guest into the VMCS.
8293 *
8294 * @param pVM Pointer to the VM.
8295 * @param pVCpu Pointer to the VMCPU.
8296 * @param pCtx Pointer to the guest-CPU context.
8297 *
8298 * @remarks No-long-jump zone!!!
8299 */
8300static void hmR0VmxLoadSharedState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
8301{
8302 NOREF(pVM);
8303
8304 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8305 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8306
8307 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0))
8308 {
8309 int rc = hmR0VmxLoadSharedCR0(pVCpu, pCtx);
8310 AssertRC(rc);
8311 }
8312
8313 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG))
8314 {
8315 int rc = hmR0VmxLoadSharedDebugState(pVCpu, pCtx);
8316 AssertRC(rc);
8317
8318 /* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */
8319 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS))
8320 {
8321 rc = hmR0VmxLoadGuestRflags(pVCpu, pCtx);
8322 AssertRC(rc);
8323 }
8324 }
8325
8326 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS))
8327 {
8328#if HC_ARCH_BITS == 64
8329 if (pVM->hm.s.fAllow64BitGuests)
8330 hmR0VmxLazyLoadGuestMsrs(pVCpu, pCtx);
8331#endif
8332 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS);
8333 }
8334
8335 AssertMsg(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE),
8336 ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8337}
8338
8339
8340/**
8341 * Worker for loading the guest-state bits in the inner VT-x execution loop.
8342 *
8343 * @param pVM Pointer to the VM.
8344 * @param pVCpu Pointer to the VMCPU.
8345 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8346 * out-of-sync. Make sure to update the required fields
8347 * before using them.
8348 */
8349DECLINLINE(void) hmR0VmxLoadGuestStateOptimal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
8350{
8351 HMVMX_ASSERT_PREEMPT_SAFE();
8352
8353 Log5(("LoadFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8354#ifdef HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
8355 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
8356#endif
8357
8358 if (HMCPU_CF_IS_SET_ONLY(pVCpu, HM_CHANGED_GUEST_RIP))
8359 {
8360 int rc = hmR0VmxLoadGuestRip(pVCpu, pMixedCtx);
8361 AssertRC(rc);
8362 STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadMinimal);
8363 }
8364 else if (HMCPU_CF_VALUE(pVCpu))
8365 {
8366 int rc = hmR0VmxLoadGuestState(pVM, pVCpu, pMixedCtx);
8367 AssertRC(rc);
8368 STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadFull);
8369 }
8370
8371 /* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */
8372 AssertMsg( !HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_ALL_GUEST)
8373 || HMCPU_CF_IS_PENDING_ONLY(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE),
8374 ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8375}
8376
8377
8378/**
8379 * Does the preparations before executing guest code in VT-x.
8380 *
8381 * This may cause longjmps to ring-3 and may even result in rescheduling to the
8382 * recompiler. We must be cautious what we do here regarding committing
8383 * guest-state information into the VMCS assuming we assuredly execute the
8384 * guest in VT-x mode. If we fall back to the recompiler after updating the VMCS
8385 * and clearing the common-state (TRPM/forceflags), we must undo those changes
8386 * so that the recompiler can (and should) use them when it resumes guest
8387 * execution. Otherwise such operations must be done when we can no longer
8388 * exit to ring-3.
8389 *
8390 * @returns Strict VBox status code.
8391 * @retval VINF_SUCCESS if we can proceed with running the guest, interrupts
8392 * have been disabled.
8393 * @retval VINF_EM_RESET if a triple-fault occurs while injecting a
8394 * double-fault into the guest.
8395 * @retval VINF_* scheduling changes, we have to go back to ring-3.
8396 *
8397 * @param pVM Pointer to the VM.
8398 * @param pVCpu Pointer to the VMCPU.
8399 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8400 * out-of-sync. Make sure to update the required fields
8401 * before using them.
8402 * @param pVmxTransient Pointer to the VMX transient structure.
8403 */
8404static int hmR0VmxPreRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
8405{
8406 Assert(VMMRZCallRing3IsEnabled(pVCpu));
8407
8408#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
8409 PGMRZDynMapFlushAutoSet(pVCpu);
8410#endif
8411
8412 /* Check force flag actions that might require us to go back to ring-3. */
8413 int rc = hmR0VmxCheckForceFlags(pVM, pVCpu, pMixedCtx);
8414 if (rc != VINF_SUCCESS)
8415 return rc;
8416
8417#ifndef IEM_VERIFICATION_MODE_FULL
8418 /* Setup the Virtualized APIC accesses. pMixedCtx->msrApicBase is always up-to-date. It's not part of the VMCS. */
8419 if ( pVCpu->hm.s.vmx.u64MsrApicBase != pMixedCtx->msrApicBase
8420 && (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC))
8421 {
8422 Assert(pVM->hm.s.vmx.HCPhysApicAccess);
8423 RTGCPHYS GCPhysApicBase;
8424 GCPhysApicBase = pMixedCtx->msrApicBase;
8425 GCPhysApicBase &= PAGE_BASE_GC_MASK;
8426
8427 /* Unalias any existing mapping. */
8428 rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase);
8429 AssertRCReturn(rc, rc);
8430
8431 /* Map the HC APIC-access page into the GC space, this also updates the shadow page tables if necessary. */
8432 Log4(("Mapped HC APIC-access page into GC: GCPhysApicBase=%#RGv\n", GCPhysApicBase));
8433 rc = IOMMMIOMapMMIOHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hm.s.vmx.HCPhysApicAccess, X86_PTE_RW | X86_PTE_P);
8434 AssertRCReturn(rc, rc);
8435
8436 pVCpu->hm.s.vmx.u64MsrApicBase = pMixedCtx->msrApicBase;
8437 }
8438#endif /* !IEM_VERIFICATION_MODE_FULL */
8439
8440 if (TRPMHasTrap(pVCpu))
8441 hmR0VmxTrpmTrapToPendingEvent(pVCpu);
8442 else if (!pVCpu->hm.s.Event.fPending)
8443 hmR0VmxEvaluatePendingEvent(pVCpu, pMixedCtx);
8444
8445 /*
8446 * Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus needs to be done with
8447 * longjmps or interrupts + preemption enabled. Event injection might also result in triple-faulting the VM.
8448 */
8449 rc = hmR0VmxInjectPendingEvent(pVCpu, pMixedCtx);
8450 if (RT_UNLIKELY(rc != VINF_SUCCESS))
8451 {
8452 Assert(rc == VINF_EM_RESET);
8453 return rc;
8454 }
8455
8456 /*
8457 * Load the guest state bits, we can handle longjmps/getting preempted here.
8458 *
8459 * If we are injecting events to a real-on-v86 mode guest, we will have to update
8460 * RIP and some segment registers, i.e. hmR0VmxInjectPendingEvent()->hmR0VmxInjectEventVmcs().
8461 * Hence, this needs to be done -after- injection of events.
8462 */
8463 hmR0VmxLoadGuestStateOptimal(pVM, pVCpu, pMixedCtx);
8464
8465 /*
8466 * No longjmps to ring-3 from this point on!!!
8467 * Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
8468 * This also disables flushing of the R0-logger instance (if any).
8469 */
8470 VMMRZCallRing3Disable(pVCpu);
8471
8472 /*
8473 * We disable interrupts so that we don't miss any interrupts that would flag preemption (IPI/timers etc.)
8474 * when thread-context hooks aren't used and we've been running with preemption disabled for a while.
8475 *
8476 * We need to check for force-flags that could've possible been altered since we last checked them (e.g.
8477 * by PDMGetInterrupt() leaving the PDM critical section, see @bugref{6398}).
8478 *
8479 * We also check a couple of other force-flags as a last opportunity to get the EMT back to ring-3 before
8480 * executing guest code.
8481 */
8482 pVmxTransient->uEflags = ASMIntDisableFlags();
8483 if ( VM_FF_IS_PENDING(pVM, VM_FF_EMT_RENDEZVOUS | VM_FF_TM_VIRTUAL_SYNC)
8484 || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
8485 {
8486 hmR0VmxClearEventVmcs(pVCpu);
8487 ASMSetFlags(pVmxTransient->uEflags);
8488 VMMRZCallRing3Enable(pVCpu);
8489 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
8490 return VINF_EM_RAW_TO_R3;
8491 }
8492
8493 if (RTThreadPreemptIsPending(NIL_RTTHREAD))
8494 {
8495 hmR0VmxClearEventVmcs(pVCpu);
8496 ASMSetFlags(pVmxTransient->uEflags);
8497 VMMRZCallRing3Enable(pVCpu);
8498 STAM_COUNTER_INC(&pVCpu->hm.s.StatPendingHostIrq);
8499 return VINF_EM_RAW_INTERRUPT;
8500 }
8501
8502 /* We've injected any pending events. This is really the point of no return (to ring-3). */
8503 pVCpu->hm.s.Event.fPending = false;
8504
8505 return VINF_SUCCESS;
8506}
8507
8508
8509/**
8510 * Prepares to run guest code in VT-x and we've committed to doing so. This
8511 * means there is no backing out to ring-3 or anywhere else at this
8512 * point.
8513 *
8514 * @param pVM Pointer to the VM.
8515 * @param pVCpu Pointer to the VMCPU.
8516 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8517 * out-of-sync. Make sure to update the required fields
8518 * before using them.
8519 * @param pVmxTransient Pointer to the VMX transient structure.
8520 *
8521 * @remarks Called with preemption disabled.
8522 * @remarks No-long-jump zone!!!
8523 */
8524static void hmR0VmxPreRunGuestCommitted(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
8525{
8526 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8527 Assert(VMMR0IsLogFlushDisabled(pVCpu));
8528 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8529
8530 VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
8531 VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC); /* Indicate the start of guest execution. */
8532
8533#ifdef HMVMX_ALWAYS_SWAP_FPU_STATE
8534 if (!CPUMIsGuestFPUStateActive(pVCpu))
8535 CPUMR0LoadGuestFPU(pVM, pVCpu, pMixedCtx);
8536 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8537#endif
8538
8539 if ( pVCpu->hm.s.fUseGuestFpu
8540 && !CPUMIsGuestFPUStateActive(pVCpu))
8541 {
8542 CPUMR0LoadGuestFPU(pVM, pVCpu, pMixedCtx);
8543 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
8544 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8545 }
8546
8547 /*
8548 * Lazy-update of the host MSRs values in the auto-load/store MSR area.
8549 */
8550 if ( !pVCpu->hm.s.vmx.fUpdatedHostMsrs
8551 && pVCpu->hm.s.vmx.cMsrs > 0)
8552 {
8553 hmR0VmxUpdateAutoLoadStoreHostMsrs(pVCpu);
8554 }
8555
8556 /*
8557 * Load the host state bits as we may've been preempted (only happens when
8558 * thread-context hooks are used or when hmR0VmxSetupVMRunHandler() changes pfnStartVM).
8559 */
8560 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT))
8561 {
8562 /* This ASSUMES that pfnStartVM has been set up already. */
8563 int rc = hmR0VmxSaveHostState(pVM, pVCpu);
8564 AssertRC(rc);
8565 STAM_COUNTER_INC(&pVCpu->hm.s.StatPreemptSaveHostState);
8566 }
8567 Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT));
8568
8569 /*
8570 * Load the state shared between host and guest (FPU, debug, lazy MSRs).
8571 */
8572 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE))
8573 hmR0VmxLoadSharedState(pVM, pVCpu, pMixedCtx);
8574 AssertMsg(!HMCPU_CF_VALUE(pVCpu), ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8575
8576 /* Store status of the shared guest-host state at the time of VM-entry. */
8577#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS) && !defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
8578 if (CPUMIsGuestInLongModeEx(pMixedCtx))
8579 {
8580 pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActivePending(pVCpu);
8581 pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActivePending(pVCpu);
8582 }
8583 else
8584#endif
8585 {
8586 pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
8587 pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
8588 }
8589 pVmxTransient->fWasGuestFPUStateActive = CPUMIsGuestFPUStateActive(pVCpu);
8590
8591 /*
8592 * Cache the TPR-shadow for checking on every VM-exit if it might have changed.
8593 */
8594 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
8595 pVmxTransient->u8GuestTpr = pVCpu->hm.s.vmx.pbVirtApic[0x80];
8596
8597 PHMGLOBALCPUINFO pCpu = HMR0GetCurrentCpu();
8598 RTCPUID idCurrentCpu = pCpu->idCpu;
8599 if ( pVmxTransient->fUpdateTscOffsettingAndPreemptTimer
8600 || idCurrentCpu != pVCpu->hm.s.idLastCpu)
8601 {
8602 hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVCpu);
8603 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = false;
8604 }
8605
8606 ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB-shootdowns, set this across the world switch. */
8607 hmR0VmxFlushTaggedTlb(pVCpu, pCpu); /* Invalidate the appropriate guest entries from the TLB. */
8608 Assert(idCurrentCpu == pVCpu->hm.s.idLastCpu);
8609 pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Update the error reporting info. with the current host CPU. */
8610
8611 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
8612
8613 TMNotifyStartOfExecution(pVCpu); /* Finally, notify TM to resume its clocks as we're about
8614 to start executing. */
8615
8616 /*
8617 * Load the TSC_AUX MSR when we are not intercepting RDTSCP.
8618 */
8619 if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP)
8620 {
8621 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT))
8622 {
8623 int rc2 = hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
8624 AssertRC(rc2);
8625 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS));
8626 bool fMsrUpdated = hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_TSC_AUX, CPUMR0GetGuestTscAux(pVCpu),
8627 true /* fUpdateHostMsr */);
8628 Assert(fMsrUpdated || pVCpu->hm.s.vmx.fUpdatedHostMsrs);
8629 /* Finally, mark that all host MSR values are updated so we don't redo it without leaving VT-x. See @bugref{6956}. */
8630 pVCpu->hm.s.vmx.fUpdatedHostMsrs = true;
8631 }
8632 else
8633 {
8634 hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, MSR_K8_TSC_AUX);
8635 Assert(!pVCpu->hm.s.vmx.cMsrs || pVCpu->hm.s.vmx.fUpdatedHostMsrs);
8636 }
8637 }
8638
8639#ifdef VBOX_STRICT
8640 hmR0VmxCheckAutoLoadStoreMsrs(pVCpu);
8641 hmR0VmxCheckHostEferMsr(pVCpu);
8642 AssertRC(hmR0VmxCheckVmcsCtls(pVCpu));
8643#endif
8644#ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE
8645 uint32_t uInvalidReason = hmR0VmxCheckGuestState(pVM, pVCpu, pMixedCtx);
8646 if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND)
8647 Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason));
8648#endif
8649}
8650
8651
8652/**
8653 * Performs some essential restoration of state after running guest code in
8654 * VT-x.
8655 *
8656 * @param pVM Pointer to the VM.
8657 * @param pVCpu Pointer to the VMCPU.
8658 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
8659 * out-of-sync. Make sure to update the required fields
8660 * before using them.
8661 * @param pVmxTransient Pointer to the VMX transient structure.
8662 * @param rcVMRun Return code of VMLAUNCH/VMRESUME.
8663 *
8664 * @remarks Called with interrupts disabled, and returns with interrups enabled!
8665 *
8666 * @remarks No-long-jump zone!!! This function will however re-enable longjmps
8667 * unconditionally when it is safe to do so.
8668 */
8669static void hmR0VmxPostRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, int rcVMRun)
8670{
8671 NOREF(pVM);
8672
8673 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8674
8675 ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB-shootdowns. */
8676 ASMAtomicIncU32(&pVCpu->hm.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for TLB-shootdowns. */
8677 HMVMXCPU_GST_RESET_TO(pVCpu, 0); /* Exits/longjmps to ring-3 requires saving the guest state. */
8678 pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */
8679 pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
8680
8681 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT))
8682 {
8683 /** @todo Find a way to fix hardcoding a guestimate. */
8684 TMCpuTickSetLastSeen(pVCpu, ASMReadTSC()
8685 + pVCpu->hm.s.vmx.u64TSCOffset - 0x400 /* guestimate of world switch overhead in clock ticks */);
8686 }
8687
8688 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatExit1, x);
8689 TMNotifyEndOfExecution(pVCpu); /* Notify TM that the guest is no longer running. */
8690 Assert(!(ASMGetFlags() & X86_EFL_IF));
8691 VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
8692
8693#ifdef HMVMX_ALWAYS_SWAP_FPU_STATE
8694 if (CPUMIsGuestFPUStateActive(pVCpu))
8695 {
8696 hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
8697 CPUMR0SaveGuestFPU(pVM, pVCpu, pMixedCtx);
8698 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8699 }
8700#endif
8701
8702#if HC_ARCH_BITS == 64
8703 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_REQUIRED; /* Host state messed up by VT-x, we must restore. */
8704#endif
8705 pVCpu->hm.s.vmx.uVmcsState |= HMVMX_VMCS_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
8706#ifdef VBOX_STRICT
8707 hmR0VmxCheckHostEferMsr(pVCpu); /* Verify that VMRUN/VMLAUNCH didn't modify host EFER. */
8708#endif
8709 ASMSetFlags(pVmxTransient->uEflags); /* Enable interrupts. */
8710 VMMRZCallRing3Enable(pVCpu); /* It is now safe to do longjmps to ring-3!!! */
8711
8712 /* Save the basic VM-exit reason. Refer Intel spec. 24.9.1 "Basic VM-exit Information". */
8713 uint32_t uExitReason;
8714 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
8715 rc |= hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
8716 AssertRC(rc);
8717 pVmxTransient->uExitReason = (uint16_t)VMX_EXIT_REASON_BASIC(uExitReason);
8718 pVmxTransient->fVMEntryFailed = VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uEntryIntInfo);
8719
8720 /* Update the VM-exit history array. */
8721 HMCPU_EXIT_HISTORY_ADD(pVCpu, pVmxTransient->uExitReason);
8722
8723 /* If the VMLAUNCH/VMRESUME failed, we can bail out early. This does -not- cover VMX_EXIT_ERR_*. */
8724 if (RT_UNLIKELY(rcVMRun != VINF_SUCCESS))
8725 {
8726 Log4(("VM-entry failure: pVCpu=%p idCpu=%RU32 rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", pVCpu, pVCpu->idCpu, rcVMRun,
8727 pVmxTransient->fVMEntryFailed));
8728 return;
8729 }
8730
8731 if (RT_LIKELY(!pVmxTransient->fVMEntryFailed))
8732 {
8733 /** @todo We can optimize this by only syncing with our force-flags when
8734 * really needed and keeping the VMCS state as it is for most
8735 * VM-exits. */
8736 /* Update the guest interruptibility-state from the VMCS. */
8737 hmR0VmxSaveGuestIntrState(pVCpu, pMixedCtx);
8738
8739#if defined(HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE) || defined(HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE)
8740 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
8741 AssertRC(rc);
8742#elif defined(HMVMX_ALWAYS_SAVE_GUEST_RFLAGS)
8743 rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
8744 AssertRC(rc);
8745#endif
8746
8747 /*
8748 * If the TPR was raised by the guest, it wouldn't cause a VM-exit immediately. Instead we sync the TPR lazily whenever
8749 * we eventually get a VM-exit for any reason. This maybe expensive as PDMApicSetTPR() can longjmp to ring-3 and which is
8750 * why it's done here as it's easier and no less efficient to deal with it here than making hmR0VmxSaveGuestState()
8751 * cope with longjmps safely (see VMCPU_FF_HM_UPDATE_CR3 handling).
8752 */
8753 if ( (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
8754 && pVmxTransient->u8GuestTpr != pVCpu->hm.s.vmx.pbVirtApic[0x80])
8755 {
8756 rc = PDMApicSetTPR(pVCpu, pVCpu->hm.s.vmx.pbVirtApic[0x80]);
8757 AssertRC(rc);
8758 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
8759 }
8760 }
8761}
8762
8763
8764/**
8765 * Runs the guest code using VT-x the normal way.
8766 *
8767 * @returns VBox status code.
8768 * @param pVM Pointer to the VM.
8769 * @param pVCpu Pointer to the VMCPU.
8770 * @param pCtx Pointer to the guest-CPU context.
8771 *
8772 * @note Mostly the same as hmR0VmxRunGuestCodeStep().
8773 */
8774static int hmR0VmxRunGuestCodeNormal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
8775{
8776 VMXTRANSIENT VmxTransient;
8777 VmxTransient.fUpdateTscOffsettingAndPreemptTimer = true;
8778 int rc = VERR_INTERNAL_ERROR_5;
8779 uint32_t cLoops = 0;
8780
8781 for (;; cLoops++)
8782 {
8783 Assert(!HMR0SuspendPending());
8784 HMVMX_ASSERT_CPU_SAFE();
8785
8786 /* Preparatory work for running guest code, this may force us to return
8787 to ring-3. This bugger disables interrupts on VINF_SUCCESS! */
8788 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
8789 rc = hmR0VmxPreRunGuest(pVM, pVCpu, pCtx, &VmxTransient);
8790 if (rc != VINF_SUCCESS)
8791 break;
8792
8793 hmR0VmxPreRunGuestCommitted(pVM, pVCpu, pCtx, &VmxTransient);
8794 rc = hmR0VmxRunGuest(pVM, pVCpu, pCtx);
8795 /* The guest-CPU context is now outdated, 'pCtx' is to be treated as 'pMixedCtx' from this point on!!! */
8796
8797 /* Restore any residual host-state and save any bits shared between host
8798 and guest into the guest-CPU state. Re-enables interrupts! */
8799 hmR0VmxPostRunGuest(pVM, pVCpu, pCtx, &VmxTransient, rc);
8800
8801 /* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
8802 if (RT_UNLIKELY(rc != VINF_SUCCESS))
8803 {
8804 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x);
8805 hmR0VmxReportWorldSwitchError(pVM, pVCpu, rc, pCtx, &VmxTransient);
8806 return rc;
8807 }
8808
8809 /* Handle the VM-exit. */
8810 AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
8811 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
8812 STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
8813 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
8814 HMVMX_START_EXIT_DISPATCH_PROF();
8815#ifdef HMVMX_USE_FUNCTION_TABLE
8816 rc = g_apfnVMExitHandlers[VmxTransient.uExitReason](pVCpu, pCtx, &VmxTransient);
8817#else
8818 rc = hmR0VmxHandleExit(pVCpu, pCtx, &VmxTransient, VmxTransient.uExitReason);
8819#endif
8820 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
8821 if (rc != VINF_SUCCESS)
8822 break;
8823 else if (cLoops > pVM->hm.s.cMaxResumeLoops)
8824 {
8825 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMaxResume);
8826 rc = VINF_EM_RAW_INTERRUPT;
8827 break;
8828 }
8829 }
8830
8831 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
8832 return rc;
8833}
8834
8835
8836/**
8837 * Single steps guest code using VT-x.
8838 *
8839 * @returns VBox status code.
8840 * @param pVM Pointer to the VM.
8841 * @param pVCpu Pointer to the VMCPU.
8842 * @param pCtx Pointer to the guest-CPU context.
8843 *
8844 * @note Mostly the same as hmR0VmxRunGuestCodeNormal().
8845 */
8846static int hmR0VmxRunGuestCodeStep(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
8847{
8848 VMXTRANSIENT VmxTransient;
8849 VmxTransient.fUpdateTscOffsettingAndPreemptTimer = true;
8850 int rc = VERR_INTERNAL_ERROR_5;
8851 uint32_t cLoops = 0;
8852 uint16_t uCsStart = pCtx->cs.Sel;
8853 uint64_t uRipStart = pCtx->rip;
8854
8855 for (;; cLoops++)
8856 {
8857 Assert(!HMR0SuspendPending());
8858 HMVMX_ASSERT_CPU_SAFE();
8859
8860 /* Preparatory work for running guest code, this may force us to return
8861 to ring-3. This bugger disables interrupts on VINF_SUCCESS! */
8862 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
8863 rc = hmR0VmxPreRunGuest(pVM, pVCpu, pCtx, &VmxTransient);
8864 if (rc != VINF_SUCCESS)
8865 break;
8866
8867 hmR0VmxPreRunGuestCommitted(pVM, pVCpu, pCtx, &VmxTransient);
8868 rc = hmR0VmxRunGuest(pVM, pVCpu, pCtx);
8869 /* The guest-CPU context is now outdated, 'pCtx' is to be treated as 'pMixedCtx' from this point on!!! */
8870
8871 /* Restore any residual host-state and save any bits shared between host
8872 and guest into the guest-CPU state. Re-enables interrupts! */
8873 hmR0VmxPostRunGuest(pVM, pVCpu, pCtx, &VmxTransient, rc);
8874
8875 /* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
8876 if (RT_UNLIKELY(rc != VINF_SUCCESS))
8877 {
8878 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x);
8879 hmR0VmxReportWorldSwitchError(pVM, pVCpu, rc, pCtx, &VmxTransient);
8880 return rc;
8881 }
8882
8883 /* Handle the VM-exit. */
8884 AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
8885 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
8886 STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
8887 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
8888 HMVMX_START_EXIT_DISPATCH_PROF();
8889#ifdef HMVMX_USE_FUNCTION_TABLE
8890 rc = g_apfnVMExitHandlers[VmxTransient.uExitReason](pVCpu, pCtx, &VmxTransient);
8891#else
8892 rc = hmR0VmxHandleExit(pVCpu, pCtx, &VmxTransient, VmxTransient.uExitReason);
8893#endif
8894 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
8895 if (rc != VINF_SUCCESS)
8896 break;
8897 else if (cLoops > pVM->hm.s.cMaxResumeLoops)
8898 {
8899 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMaxResume);
8900 rc = VINF_EM_RAW_INTERRUPT;
8901 break;
8902 }
8903
8904 /*
8905 * Did the RIP change, if so, consider it a single step.
8906 * Otherwise, make sure one of the TFs gets set.
8907 */
8908 int rc2 = hmR0VmxSaveGuestRip(pVCpu, pCtx);
8909 rc2 |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pCtx);
8910 AssertRCReturn(rc2, rc2);
8911 if ( pCtx->rip != uRipStart
8912 || pCtx->cs.Sel != uCsStart)
8913 {
8914 rc = VINF_EM_DBG_STEPPED;
8915 break;
8916 }
8917 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
8918 }
8919
8920 /*
8921 * Clear the X86_EFL_TF if necessary.
8922 */
8923 if (pVCpu->hm.s.fClearTrapFlag)
8924 {
8925 int rc2 = hmR0VmxSaveGuestRflags(pVCpu, pCtx);
8926 AssertRCReturn(rc2, rc2);
8927 pVCpu->hm.s.fClearTrapFlag = false;
8928 pCtx->eflags.Bits.u1TF = 0;
8929 }
8930 /** @todo there seems to be issues with the resume flag when the monitor trap
8931 * flag is pending without being used. Seen early in bios init when
8932 * accessing APIC page in protected mode. */
8933
8934 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
8935 return rc;
8936}
8937
8938
8939/**
8940 * Runs the guest code using VT-x.
8941 *
8942 * @returns VBox status code.
8943 * @param pVM Pointer to the VM.
8944 * @param pVCpu Pointer to the VMCPU.
8945 * @param pCtx Pointer to the guest-CPU context.
8946 */
8947VMMR0DECL(int) VMXR0RunGuestCode(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
8948{
8949 Assert(VMMRZCallRing3IsEnabled(pVCpu));
8950 Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
8951 HMVMX_ASSERT_PREEMPT_SAFE();
8952
8953 VMMRZCallRing3SetNotification(pVCpu, hmR0VmxCallRing3Callback, pCtx);
8954
8955 int rc;
8956 if (!pVCpu->hm.s.fSingleInstruction && !DBGFIsStepping(pVCpu))
8957 rc = hmR0VmxRunGuestCodeNormal(pVM, pVCpu, pCtx);
8958 else
8959 rc = hmR0VmxRunGuestCodeStep(pVM, pVCpu, pCtx);
8960
8961 if (rc == VERR_EM_INTERPRETER)
8962 rc = VINF_EM_RAW_EMULATE_INSTR;
8963 else if (rc == VINF_EM_RESET)
8964 rc = VINF_EM_TRIPLE_FAULT;
8965
8966 int rc2 = hmR0VmxExitToRing3(pVM, pVCpu, pCtx, rc);
8967 if (RT_FAILURE(rc2))
8968 {
8969 pVCpu->hm.s.u32HMError = rc;
8970 rc = rc2;
8971 }
8972 Assert(!VMMRZCallRing3IsNotificationSet(pVCpu));
8973 return rc;
8974}
8975
8976
8977#ifndef HMVMX_USE_FUNCTION_TABLE
8978DECLINLINE(int) hmR0VmxHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, uint32_t rcReason)
8979{
8980#ifdef DEBUG_ramshankar
8981# define SVVMCS() do { int rc2 = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRC(rc2); } while (0)
8982# define LDVMCS() do { HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST); } while (0)
8983#endif
8984 int rc;
8985 switch (rcReason)
8986 {
8987 case VMX_EXIT_EPT_MISCONFIG: /* SVVMCS(); */ rc = hmR0VmxExitEptMisconfig(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8988 case VMX_EXIT_EPT_VIOLATION: /* SVVMCS(); */ rc = hmR0VmxExitEptViolation(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8989 case VMX_EXIT_IO_INSTR: /* SVVMCS(); */ rc = hmR0VmxExitIoInstr(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8990 case VMX_EXIT_CPUID: /* SVVMCS(); */ rc = hmR0VmxExitCpuid(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8991 case VMX_EXIT_RDTSC: /* SVVMCS(); */ rc = hmR0VmxExitRdtsc(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8992 case VMX_EXIT_RDTSCP: /* SVVMCS(); */ rc = hmR0VmxExitRdtscp(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8993 case VMX_EXIT_APIC_ACCESS: /* SVVMCS(); */ rc = hmR0VmxExitApicAccess(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8994 case VMX_EXIT_XCPT_OR_NMI: /* SVVMCS(); */ rc = hmR0VmxExitXcptOrNmi(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8995 case VMX_EXIT_MOV_CRX: /* SVVMCS(); */ rc = hmR0VmxExitMovCRx(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8996 case VMX_EXIT_EXT_INT: /* SVVMCS(); */ rc = hmR0VmxExitExtInt(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8997 case VMX_EXIT_INT_WINDOW: /* SVVMCS(); */ rc = hmR0VmxExitIntWindow(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8998 case VMX_EXIT_MWAIT: /* SVVMCS(); */ rc = hmR0VmxExitMwait(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
8999 case VMX_EXIT_MONITOR: /* SVVMCS(); */ rc = hmR0VmxExitMonitor(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9000 case VMX_EXIT_TASK_SWITCH: /* SVVMCS(); */ rc = hmR0VmxExitTaskSwitch(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9001 case VMX_EXIT_PREEMPT_TIMER: /* SVVMCS(); */ rc = hmR0VmxExitPreemptTimer(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9002 case VMX_EXIT_RDMSR: /* SVVMCS(); */ rc = hmR0VmxExitRdmsr(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9003 case VMX_EXIT_WRMSR: /* SVVMCS(); */ rc = hmR0VmxExitWrmsr(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9004 case VMX_EXIT_MOV_DRX: /* SVVMCS(); */ rc = hmR0VmxExitMovDRx(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9005 case VMX_EXIT_TPR_BELOW_THRESHOLD: /* SVVMCS(); */ rc = hmR0VmxExitTprBelowThreshold(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9006 case VMX_EXIT_HLT: /* SVVMCS(); */ rc = hmR0VmxExitHlt(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9007 case VMX_EXIT_INVD: /* SVVMCS(); */ rc = hmR0VmxExitInvd(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9008 case VMX_EXIT_INVLPG: /* SVVMCS(); */ rc = hmR0VmxExitInvlpg(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9009 case VMX_EXIT_RSM: /* SVVMCS(); */ rc = hmR0VmxExitRsm(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9010 case VMX_EXIT_MTF: /* SVVMCS(); */ rc = hmR0VmxExitMtf(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9011 case VMX_EXIT_PAUSE: /* SVVMCS(); */ rc = hmR0VmxExitPause(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9012 case VMX_EXIT_XDTR_ACCESS: /* SVVMCS(); */ rc = hmR0VmxExitXdtrAccess(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9013 case VMX_EXIT_TR_ACCESS: /* SVVMCS(); */ rc = hmR0VmxExitXdtrAccess(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9014 case VMX_EXIT_WBINVD: /* SVVMCS(); */ rc = hmR0VmxExitWbinvd(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9015 case VMX_EXIT_XSETBV: /* SVVMCS(); */ rc = hmR0VmxExitXsetbv(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9016 case VMX_EXIT_RDRAND: /* SVVMCS(); */ rc = hmR0VmxExitRdrand(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9017 case VMX_EXIT_INVPCID: /* SVVMCS(); */ rc = hmR0VmxExitInvpcid(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9018 case VMX_EXIT_GETSEC: /* SVVMCS(); */ rc = hmR0VmxExitGetsec(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9019 case VMX_EXIT_RDPMC: /* SVVMCS(); */ rc = hmR0VmxExitRdpmc(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9020 case VMX_EXIT_VMCALL: /* SVVMCS(); */ rc = hmR0VmxExitVmcall(pVCpu, pMixedCtx, pVmxTransient); /* LDVMCS(); */ break;
9021
9022 case VMX_EXIT_TRIPLE_FAULT: rc = hmR0VmxExitTripleFault(pVCpu, pMixedCtx, pVmxTransient); break;
9023 case VMX_EXIT_NMI_WINDOW: rc = hmR0VmxExitNmiWindow(pVCpu, pMixedCtx, pVmxTransient); break;
9024 case VMX_EXIT_INIT_SIGNAL: rc = hmR0VmxExitInitSignal(pVCpu, pMixedCtx, pVmxTransient); break;
9025 case VMX_EXIT_SIPI: rc = hmR0VmxExitSipi(pVCpu, pMixedCtx, pVmxTransient); break;
9026 case VMX_EXIT_IO_SMI: rc = hmR0VmxExitIoSmi(pVCpu, pMixedCtx, pVmxTransient); break;
9027 case VMX_EXIT_SMI: rc = hmR0VmxExitSmi(pVCpu, pMixedCtx, pVmxTransient); break;
9028 case VMX_EXIT_ERR_MSR_LOAD: rc = hmR0VmxExitErrMsrLoad(pVCpu, pMixedCtx, pVmxTransient); break;
9029 case VMX_EXIT_ERR_INVALID_GUEST_STATE: rc = hmR0VmxExitErrInvalidGuestState(pVCpu, pMixedCtx, pVmxTransient); break;
9030 case VMX_EXIT_ERR_MACHINE_CHECK: rc = hmR0VmxExitErrMachineCheck(pVCpu, pMixedCtx, pVmxTransient); break;
9031
9032 case VMX_EXIT_VMCLEAR:
9033 case VMX_EXIT_VMLAUNCH:
9034 case VMX_EXIT_VMPTRLD:
9035 case VMX_EXIT_VMPTRST:
9036 case VMX_EXIT_VMREAD:
9037 case VMX_EXIT_VMRESUME:
9038 case VMX_EXIT_VMWRITE:
9039 case VMX_EXIT_VMXOFF:
9040 case VMX_EXIT_VMXON:
9041 case VMX_EXIT_INVEPT:
9042 case VMX_EXIT_INVVPID:
9043 case VMX_EXIT_VMFUNC:
9044 rc = hmR0VmxExitSetPendingXcptUD(pVCpu, pMixedCtx, pVmxTransient);
9045 break;
9046 default:
9047 rc = hmR0VmxExitErrUndefined(pVCpu, pMixedCtx, pVmxTransient);
9048 break;
9049 }
9050 return rc;
9051}
9052#endif
9053
9054#ifdef DEBUG
9055/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
9056# define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \
9057 RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
9058
9059# define HMVMX_ASSERT_PREEMPT_CPUID() \
9060 do \
9061 { \
9062 RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
9063 AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
9064 } while (0)
9065
9066# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS() \
9067 do { \
9068 AssertPtr(pVCpu); \
9069 AssertPtr(pMixedCtx); \
9070 AssertPtr(pVmxTransient); \
9071 Assert(pVmxTransient->fVMEntryFailed == false); \
9072 Assert(ASMIntAreEnabled()); \
9073 HMVMX_ASSERT_PREEMPT_SAFE(); \
9074 HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \
9075 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)); \
9076 HMVMX_ASSERT_PREEMPT_SAFE(); \
9077 if (VMMR0IsLogFlushDisabled(pVCpu)) \
9078 HMVMX_ASSERT_PREEMPT_CPUID(); \
9079 HMVMX_STOP_EXIT_DISPATCH_PROF(); \
9080 } while (0)
9081
9082# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS() \
9083 do { \
9084 Log4Func(("\n")); \
9085 } while (0)
9086#else /* Release builds */
9087# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS() \
9088 do { \
9089 HMVMX_STOP_EXIT_DISPATCH_PROF(); \
9090 NOREF(pVCpu); NOREF(pMixedCtx); NOREF(pVmxTransient); \
9091 } while (0)
9092# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS() do { } while (0)
9093#endif
9094
9095
9096/**
9097 * Advances the guest RIP after reading it from the VMCS.
9098 *
9099 * @returns VBox status code.
9100 * @param pVCpu Pointer to the VMCPU.
9101 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
9102 * out-of-sync. Make sure to update the required fields
9103 * before using them.
9104 * @param pVmxTransient Pointer to the VMX transient structure.
9105 *
9106 * @remarks No-long-jump zone!!!
9107 */
9108DECLINLINE(int) hmR0VmxAdvanceGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
9109{
9110 int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
9111 rc |= hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
9112 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
9113 AssertRCReturn(rc, rc);
9114
9115 pMixedCtx->rip += pVmxTransient->cbInstr;
9116 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
9117
9118 /*
9119 * Deliver a debug exception to the guest if it is single-stepping. Don't directly inject a #DB but use the
9120 * pending debug exception field as it takes care of priority of events.
9121 *
9122 * See Intel spec. 32.2.1 "Debug Exceptions".
9123 */
9124 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
9125
9126 return rc;
9127}
9128
9129
9130/**
9131 * Tries to determine what part of the guest-state VT-x has deemed as invalid
9132 * and update error record fields accordingly.
9133 *
9134 * @return VMX_IGS_* return codes.
9135 * @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything
9136 * wrong with the guest state.
9137 *
9138 * @param pVM Pointer to the VM.
9139 * @param pVCpu Pointer to the VMCPU.
9140 * @param pCtx Pointer to the guest-CPU state.
9141 *
9142 * @remarks This function assumes our cache of the VMCS controls
9143 * are valid, i.e. hmR0VmxCheckVmcsCtls() succeeded.
9144 */
9145static uint32_t hmR0VmxCheckGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
9146{
9147#define HMVMX_ERROR_BREAK(err) { uError = (err); break; }
9148#define HMVMX_CHECK_BREAK(expr, err) if (!(expr)) { \
9149 uError = (err); \
9150 break; \
9151 } else do { } while (0)
9152
9153 int rc;
9154 uint32_t uError = VMX_IGS_ERROR;
9155 uint32_t u32Val;
9156 bool fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuest;
9157
9158 do
9159 {
9160 /*
9161 * CR0.
9162 */
9163 uint32_t uSetCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
9164 uint32_t uZapCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
9165 /* Exceptions for unrestricted-guests for fixed CR0 bits (PE, PG).
9166 See Intel spec. 26.3.1 "Checks on guest Guest Control Registers, Debug Registers and MSRs." */
9167 if (fUnrestrictedGuest)
9168 uSetCR0 &= ~(X86_CR0_PE | X86_CR0_PG);
9169
9170 uint32_t u32GuestCR0;
9171 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32GuestCR0);
9172 AssertRCBreak(rc);
9173 HMVMX_CHECK_BREAK((u32GuestCR0 & uSetCR0) == uSetCR0, VMX_IGS_CR0_FIXED1);
9174 HMVMX_CHECK_BREAK(!(u32GuestCR0 & ~uZapCR0), VMX_IGS_CR0_FIXED0);
9175 if ( !fUnrestrictedGuest
9176 && (u32GuestCR0 & X86_CR0_PG)
9177 && !(u32GuestCR0 & X86_CR0_PE))
9178 {
9179 HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO);
9180 }
9181
9182 /*
9183 * CR4.
9184 */
9185 uint64_t uSetCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
9186 uint64_t uZapCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
9187
9188 uint32_t u32GuestCR4;
9189 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &u32GuestCR4);
9190 AssertRCBreak(rc);
9191 HMVMX_CHECK_BREAK((u32GuestCR4 & uSetCR4) == uSetCR4, VMX_IGS_CR4_FIXED1);
9192 HMVMX_CHECK_BREAK(!(u32GuestCR4 & ~uZapCR4), VMX_IGS_CR4_FIXED0);
9193
9194 /*
9195 * IA32_DEBUGCTL MSR.
9196 */
9197 uint64_t u64Val;
9198 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val);
9199 AssertRCBreak(rc);
9200 if ( (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
9201 && (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */
9202 {
9203 HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED);
9204 }
9205 uint64_t u64DebugCtlMsr = u64Val;
9206
9207#ifdef VBOX_STRICT
9208 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
9209 AssertRCBreak(rc);
9210 Assert(u32Val == pVCpu->hm.s.vmx.u32EntryCtls);
9211#endif
9212 bool const fLongModeGuest = RT_BOOL(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST);
9213
9214 /*
9215 * RIP and RFLAGS.
9216 */
9217 uint32_t u32Eflags;
9218#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9219 if (HMVMX_IS_64BIT_HOST_MODE())
9220 {
9221 rc = VMXReadVmcs64(VMX_VMCS_GUEST_RIP, &u64Val);
9222 AssertRCBreak(rc);
9223 /* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */
9224 if ( !fLongModeGuest
9225 || !pCtx->cs.Attr.n.u1Long)
9226 {
9227 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID);
9228 }
9229 /** @todo If the processor supports N < 64 linear-address bits, bits 63:N
9230 * must be identical if the "IA-32e mode guest" VM-entry
9231 * control is 1 and CS.L is 1. No check applies if the
9232 * CPU supports 64 linear-address bits. */
9233
9234 /* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */
9235 rc = VMXReadVmcs64(VMX_VMCS_GUEST_RFLAGS, &u64Val);
9236 AssertRCBreak(rc);
9237 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */
9238 VMX_IGS_RFLAGS_RESERVED);
9239 HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
9240 u32Eflags = u64Val;
9241 }
9242 else
9243#endif
9244 {
9245 rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Eflags);
9246 AssertRCBreak(rc);
9247 HMVMX_CHECK_BREAK(!(u32Eflags & 0xffc08028), VMX_IGS_RFLAGS_RESERVED); /* Bit 31:22, Bit 15, 5, 3 MBZ. */
9248 HMVMX_CHECK_BREAK((u32Eflags & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
9249 }
9250
9251 if ( fLongModeGuest
9252 || ( fUnrestrictedGuest
9253 && !(u32GuestCR0 & X86_CR0_PE)))
9254 {
9255 HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID);
9256 }
9257
9258 uint32_t u32EntryInfo;
9259 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
9260 AssertRCBreak(rc);
9261 if ( VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo)
9262 && VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
9263 {
9264 HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID);
9265 }
9266
9267 /*
9268 * 64-bit checks.
9269 */
9270#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9271 if (HMVMX_IS_64BIT_HOST_MODE())
9272 {
9273 if ( fLongModeGuest
9274 && !fUnrestrictedGuest)
9275 {
9276 HMVMX_CHECK_BREAK(u32GuestCR0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE);
9277 HMVMX_CHECK_BREAK(u32GuestCR4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE);
9278 }
9279
9280 if ( !fLongModeGuest
9281 && (u32GuestCR4 & X86_CR4_PCIDE))
9282 {
9283 HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE);
9284 }
9285
9286 /** @todo CR3 field must be such that bits 63:52 and bits in the range
9287 * 51:32 beyond the processor's physical-address width are 0. */
9288
9289 if ( (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
9290 && (pCtx->dr[7] & X86_DR7_MBZ_MASK))
9291 {
9292 HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED);
9293 }
9294
9295 rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val);
9296 AssertRCBreak(rc);
9297 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL);
9298
9299 rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val);
9300 AssertRCBreak(rc);
9301 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL);
9302 }
9303#endif
9304
9305 /*
9306 * PERF_GLOBAL MSR.
9307 */
9308 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR)
9309 {
9310 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val);
9311 AssertRCBreak(rc);
9312 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)),
9313 VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */
9314 }
9315
9316 /*
9317 * PAT MSR.
9318 */
9319 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR)
9320 {
9321 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val);
9322 AssertRCBreak(rc);
9323 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED);
9324 for (unsigned i = 0; i < 8; i++)
9325 {
9326 uint8_t u8Val = (u64Val & 0x7);
9327 if ( u8Val != 0 /* UC */
9328 || u8Val != 1 /* WC */
9329 || u8Val != 4 /* WT */
9330 || u8Val != 5 /* WP */
9331 || u8Val != 6 /* WB */
9332 || u8Val != 7 /* UC- */)
9333 {
9334 HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID);
9335 }
9336 u64Val >>= 3;
9337 }
9338 }
9339
9340 /*
9341 * EFER MSR.
9342 */
9343 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR)
9344 {
9345 Assert(pVM->hm.s.vmx.fSupportsVmcsEfer);
9346 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val);
9347 AssertRCBreak(rc);
9348 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)),
9349 VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */
9350 HMVMX_CHECK_BREAK(RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST),
9351 VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH);
9352 HMVMX_CHECK_BREAK( fUnrestrictedGuest
9353 || !(u32GuestCR0 & X86_CR0_PG)
9354 || RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(u64Val & MSR_K6_EFER_LME),
9355 VMX_IGS_EFER_LMA_LME_MISMATCH);
9356 }
9357
9358 /*
9359 * Segment registers.
9360 */
9361 HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9362 || !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID);
9363 if (!(u32Eflags & X86_EFL_VM))
9364 {
9365 /* CS */
9366 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID);
9367 HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED);
9368 HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED);
9369 HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff
9370 || !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9371 HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000)
9372 || (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
9373 /* CS cannot be loaded with NULL in protected mode. */
9374 HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE);
9375 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID);
9376 if (pCtx->cs.Attr.n.u4Type == 9 || pCtx->cs.Attr.n.u4Type == 11)
9377 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL);
9378 else if (pCtx->cs.Attr.n.u4Type == 13 || pCtx->cs.Attr.n.u4Type == 15)
9379 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH);
9380 else if (pVM->hm.s.vmx.fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3)
9381 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID);
9382 else
9383 HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID);
9384
9385 /* SS */
9386 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9387 || (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL);
9388 HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL);
9389 if ( !(pCtx->cr0 & X86_CR0_PE)
9390 || pCtx->cs.Attr.n.u4Type == 3)
9391 {
9392 HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID);
9393 }
9394 if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
9395 {
9396 HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 || pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID);
9397 HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID);
9398 HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED);
9399 HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED);
9400 HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff
9401 || !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9402 HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000)
9403 || (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
9404 }
9405
9406 /* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxWriteSegmentReg(). */
9407 if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
9408 {
9409 HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID);
9410 HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID);
9411 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9412 || pCtx->ds.Attr.n.u4Type > 11
9413 || pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9414 HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED);
9415 HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED);
9416 HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff
9417 || !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9418 HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000)
9419 || (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
9420 HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9421 || (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID);
9422 }
9423 if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
9424 {
9425 HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID);
9426 HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID);
9427 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9428 || pCtx->es.Attr.n.u4Type > 11
9429 || pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
9430 HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED);
9431 HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED);
9432 HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff
9433 || !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9434 HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000)
9435 || (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
9436 HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9437 || (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID);
9438 }
9439 if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
9440 {
9441 HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID);
9442 HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID);
9443 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9444 || pCtx->fs.Attr.n.u4Type > 11
9445 || pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL);
9446 HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED);
9447 HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED);
9448 HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff
9449 || !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
9450 HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000)
9451 || (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
9452 HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9453 || (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID);
9454 }
9455 if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
9456 {
9457 HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID);
9458 HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID);
9459 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
9460 || pCtx->gs.Attr.n.u4Type > 11
9461 || pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL);
9462 HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED);
9463 HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED);
9464 HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff
9465 || !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
9466 HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000)
9467 || (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
9468 HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
9469 || (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID);
9470 }
9471 /* 64-bit capable CPUs. */
9472#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9473 if (HMVMX_IS_64BIT_HOST_MODE())
9474 {
9475 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
9476 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
9477 HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9478 || X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
9479 HMVMX_CHECK_BREAK(!(pCtx->cs.u64Base >> 32), VMX_IGS_LONGMODE_CS_BASE_INVALID);
9480 HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ss.u64Base >> 32),
9481 VMX_IGS_LONGMODE_SS_BASE_INVALID);
9482 HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ds.u64Base >> 32),
9483 VMX_IGS_LONGMODE_DS_BASE_INVALID);
9484 HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->es.u64Base >> 32),
9485 VMX_IGS_LONGMODE_ES_BASE_INVALID);
9486 }
9487#endif
9488 }
9489 else
9490 {
9491 /* V86 mode checks. */
9492 uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
9493 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
9494 {
9495 u32CSAttr = 0xf3; u32SSAttr = 0xf3;
9496 u32DSAttr = 0xf3; u32ESAttr = 0xf3;
9497 u32FSAttr = 0xf3; u32GSAttr = 0xf3;
9498 }
9499 else
9500 {
9501 u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u;
9502 u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u;
9503 u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
9504 }
9505
9506 /* CS */
9507 HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID);
9508 HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID);
9509 HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID);
9510 /* SS */
9511 HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID);
9512 HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID);
9513 HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID);
9514 /* DS */
9515 HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID);
9516 HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID);
9517 HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID);
9518 /* ES */
9519 HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID);
9520 HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID);
9521 HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID);
9522 /* FS */
9523 HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID);
9524 HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID);
9525 HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID);
9526 /* GS */
9527 HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID);
9528 HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID);
9529 HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID);
9530 /* 64-bit capable CPUs. */
9531#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9532 if (HMVMX_IS_64BIT_HOST_MODE())
9533 {
9534 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
9535 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
9536 HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
9537 || X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
9538 HMVMX_CHECK_BREAK(!(pCtx->cs.u64Base >> 32), VMX_IGS_LONGMODE_CS_BASE_INVALID);
9539 HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ss.u64Base >> 32),
9540 VMX_IGS_LONGMODE_SS_BASE_INVALID);
9541 HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ds.u64Base >> 32),
9542 VMX_IGS_LONGMODE_DS_BASE_INVALID);
9543 HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->es.u64Base >> 32),
9544 VMX_IGS_LONGMODE_ES_BASE_INVALID);
9545 }
9546#endif
9547 }
9548
9549 /*
9550 * TR.
9551 */
9552 HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID);
9553 /* 64-bit capable CPUs. */
9554#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9555 if (HMVMX_IS_64BIT_HOST_MODE())
9556 {
9557 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL);
9558 }
9559#endif
9560 if (fLongModeGuest)
9561 {
9562 HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */
9563 VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID);
9564 }
9565 else
9566 {
9567 HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */
9568 || pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/
9569 VMX_IGS_TR_ATTR_TYPE_INVALID);
9570 }
9571 HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID);
9572 HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID);
9573 HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */
9574 HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff
9575 || !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
9576 HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000)
9577 || (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
9578 HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE);
9579
9580 /*
9581 * GDTR and IDTR.
9582 */
9583#if HC_ARCH_BITS == 64 || defined(VBOX_WITH_HYBRID_32BIT_KERNEL)
9584 if (HMVMX_IS_64BIT_HOST_MODE())
9585 {
9586 rc = VMXReadVmcs64(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
9587 AssertRCBreak(rc);
9588 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL);
9589
9590 rc = VMXReadVmcs64(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
9591 AssertRCBreak(rc);
9592 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL);
9593 }
9594#endif
9595
9596 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
9597 AssertRCBreak(rc);
9598 HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
9599
9600 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
9601 AssertRCBreak(rc);
9602 HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
9603
9604 /*
9605 * Guest Non-Register State.
9606 */
9607 /* Activity State. */
9608 uint32_t u32ActivityState;
9609 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState);
9610 AssertRCBreak(rc);
9611 HMVMX_CHECK_BREAK( !u32ActivityState
9612 || (u32ActivityState & MSR_IA32_VMX_MISC_ACTIVITY_STATES(pVM->hm.s.vmx.Msrs.u64Misc)),
9613 VMX_IGS_ACTIVITY_STATE_INVALID);
9614 HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl)
9615 || u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID);
9616 uint32_t u32IntrState;
9617 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &u32IntrState);
9618 AssertRCBreak(rc);
9619 if ( u32IntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS
9620 || u32IntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
9621 {
9622 HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID);
9623 }
9624
9625 /** @todo Activity state and injecting interrupts. Left as a todo since we
9626 * currently don't use activity states but ACTIVE. */
9627
9628 HMVMX_CHECK_BREAK( !(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_ENTRY_SMM)
9629 || u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID);
9630
9631 /* Guest interruptibility-state. */
9632 HMVMX_CHECK_BREAK(!(u32IntrState & 0xfffffff0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED);
9633 HMVMX_CHECK_BREAK((u32IntrState & ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI
9634 | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS))
9635 != ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI
9636 | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
9637 VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID);
9638 HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF)
9639 || !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI),
9640 VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID);
9641 if (VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo))
9642 {
9643 if (VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
9644 {
9645 HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
9646 && !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
9647 VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID);
9648 }
9649 else if (VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
9650 {
9651 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
9652 VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID);
9653 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI),
9654 VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID);
9655 }
9656 }
9657 /** @todo Assumes the processor is not in SMM. */
9658 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI),
9659 VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID);
9660 HMVMX_CHECK_BREAK( !(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_ENTRY_SMM)
9661 || (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI),
9662 VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID);
9663 if ( (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
9664 && VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo)
9665 && VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
9666 {
9667 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI),
9668 VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID);
9669 }
9670
9671 /* Pending debug exceptions. */
9672 if (HMVMX_IS_64BIT_HOST_MODE())
9673 {
9674 rc = VMXReadVmcs64(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, &u64Val);
9675 AssertRCBreak(rc);
9676 /* Bits 63:15, Bit 13, Bits 11:4 MBZ. */
9677 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED);
9678 u32Val = u64Val; /* For pending debug exceptions checks below. */
9679 }
9680 else
9681 {
9682 rc = VMXReadVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, &u32Val);
9683 AssertRCBreak(rc);
9684 /* Bits 31:15, Bit 13, Bits 11:4 MBZ. */
9685 HMVMX_CHECK_BREAK(!(u64Val & 0xffffaff0), VMX_IGS_PENDING_DEBUG_RESERVED);
9686 }
9687
9688 if ( (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
9689 || (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS)
9690 || u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
9691 {
9692 if ( (u32Eflags & X86_EFL_TF)
9693 && !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
9694 {
9695 /* Bit 14 is PendingDebug.BS. */
9696 HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET);
9697 }
9698 if ( !(u32Eflags & X86_EFL_TF)
9699 || (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
9700 {
9701 /* Bit 14 is PendingDebug.BS. */
9702 HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR);
9703 }
9704 }
9705
9706 /* VMCS link pointer. */
9707 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val);
9708 AssertRCBreak(rc);
9709 if (u64Val != UINT64_C(0xffffffffffffffff))
9710 {
9711 HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED);
9712 /** @todo Bits beyond the processor's physical-address width MBZ. */
9713 /** @todo 32-bit located in memory referenced by value of this field (as a
9714 * physical address) must contain the processor's VMCS revision ID. */
9715 /** @todo SMM checks. */
9716 }
9717
9718 /** @todo Checks on Guest Page-Directory-Pointer-Table Entries when guest is
9719 * not using Nested Paging? */
9720 if ( pVM->hm.s.fNestedPaging
9721 && !fLongModeGuest
9722 && CPUMIsGuestInPAEModeEx(pCtx))
9723 {
9724 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &u64Val);
9725 AssertRCBreak(rc);
9726 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
9727
9728 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &u64Val);
9729 AssertRCBreak(rc);
9730 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
9731
9732 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &u64Val);
9733 AssertRCBreak(rc);
9734 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
9735
9736 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &u64Val);
9737 AssertRCBreak(rc);
9738 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
9739 }
9740
9741 /* Shouldn't happen but distinguish it from AssertRCBreak() errors. */
9742 if (uError == VMX_IGS_ERROR)
9743 uError = VMX_IGS_REASON_NOT_FOUND;
9744 } while (0);
9745
9746 pVCpu->hm.s.u32HMError = uError;
9747 return uError;
9748
9749#undef HMVMX_ERROR_BREAK
9750#undef HMVMX_CHECK_BREAK
9751}
9752
9753/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
9754/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
9755/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
9756
9757/** @name VM-exit handlers.
9758 * @{
9759 */
9760
9761/**
9762 * VM-exit handler for external interrupts (VMX_EXIT_EXT_INT).
9763 */
9764HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
9765{
9766 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
9767 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
9768 /* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */
9769 if (VMMR0ThreadCtxHooksAreRegistered(pVCpu))
9770 return VINF_SUCCESS;
9771 return VINF_EM_RAW_INTERRUPT;
9772}
9773
9774
9775/**
9776 * VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI).
9777 */
9778HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
9779{
9780 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
9781 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3);
9782
9783 int rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
9784 AssertRCReturn(rc, rc);
9785
9786 uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVmxTransient->uExitIntInfo);
9787 Assert( !(pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT)
9788 && uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT);
9789 Assert(VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
9790
9791 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
9792 {
9793 /*
9794 * This cannot be a guest NMI as the only way for the guest to receive an NMI is if we injected it ourselves and
9795 * anything we inject is not going to cause a VM-exit directly for the event being injected.
9796 * See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery".
9797 *
9798 * Dispatch the NMI to the host. See Intel spec. 27.5.5 "Updating Non-Register State".
9799 */
9800 VMXDispatchHostNmi();
9801 STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
9802 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
9803 return VINF_SUCCESS;
9804 }
9805
9806 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
9807 rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
9808 if (RT_UNLIKELY(rc == VINF_HM_DOUBLE_FAULT))
9809 {
9810 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
9811 return VINF_SUCCESS;
9812 }
9813 else if (RT_UNLIKELY(rc == VINF_EM_RESET))
9814 {
9815 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
9816 return rc;
9817 }
9818
9819 uint32_t uExitIntInfo = pVmxTransient->uExitIntInfo;
9820 uint32_t uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(uExitIntInfo);
9821 switch (uIntType)
9822 {
9823 case VMX_EXIT_INTERRUPTION_INFO_TYPE_PRIV_SW_XCPT: /* Privileged software exception. (#DB from ICEBP) */
9824 Assert(uVector == X86_XCPT_DB);
9825 /* no break */
9826 case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT: /* Software exception. (#BP or #OF) */
9827 Assert(uVector == X86_XCPT_BP || uVector == X86_XCPT_OF || uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_PRIV_SW_XCPT);
9828 /* no break */
9829 case VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT:
9830 {
9831 switch (uVector)
9832 {
9833 case X86_XCPT_PF: rc = hmR0VmxExitXcptPF(pVCpu, pMixedCtx, pVmxTransient); break;
9834 case X86_XCPT_GP: rc = hmR0VmxExitXcptGP(pVCpu, pMixedCtx, pVmxTransient); break;
9835 case X86_XCPT_NM: rc = hmR0VmxExitXcptNM(pVCpu, pMixedCtx, pVmxTransient); break;
9836 case X86_XCPT_MF: rc = hmR0VmxExitXcptMF(pVCpu, pMixedCtx, pVmxTransient); break;
9837 case X86_XCPT_DB: rc = hmR0VmxExitXcptDB(pVCpu, pMixedCtx, pVmxTransient); break;
9838 case X86_XCPT_BP: rc = hmR0VmxExitXcptBP(pVCpu, pMixedCtx, pVmxTransient); break;
9839#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
9840 case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF);
9841 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
9842 case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE);
9843 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
9844 case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD);
9845 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
9846 case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS);
9847 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
9848 case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP);
9849 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
9850 case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS);
9851 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
9852#endif
9853 default:
9854 {
9855 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
9856 AssertRCReturn(rc, rc);
9857
9858 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
9859 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
9860 {
9861 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
9862 Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
9863 Assert(CPUMIsGuestInRealModeEx(pMixedCtx));
9864
9865 rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
9866 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
9867 AssertRCReturn(rc, rc);
9868 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(uExitIntInfo),
9869 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode,
9870 0 /* GCPtrFaultAddress */);
9871 AssertRCReturn(rc, rc);
9872 }
9873 else
9874 {
9875 AssertMsgFailed(("Unexpected VM-exit caused by exception %#x\n", uVector));
9876 pVCpu->hm.s.u32HMError = uVector;
9877 rc = VERR_VMX_UNEXPECTED_EXCEPTION;
9878 }
9879 break;
9880 }
9881 }
9882 break;
9883 }
9884
9885 default:
9886 {
9887 pVCpu->hm.s.u32HMError = uExitIntInfo;
9888 rc = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
9889 AssertMsgFailed(("Unexpected interruption info %#x\n", VMX_EXIT_INTERRUPTION_INFO_TYPE(uExitIntInfo)));
9890 break;
9891 }
9892 }
9893 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
9894 return rc;
9895}
9896
9897
9898/**
9899 * VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
9900 */
9901HMVMX_EXIT_DECL hmR0VmxExitIntWindow(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
9902{
9903 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
9904
9905 /* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */
9906 hmR0VmxClearIntWindowExitVmcs(pVCpu);
9907
9908 /* Deliver the pending interrupts via hmR0VmxEvaluatePendingEvent() and resume guest execution. */
9909 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
9910 return VINF_SUCCESS;
9911}
9912
9913
9914/**
9915 * VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
9916 */
9917HMVMX_EXIT_DECL hmR0VmxExitNmiWindow(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
9918{
9919 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
9920 if (RT_UNLIKELY(!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT)))
9921 {
9922 AssertMsgFailed(("Unexpected NMI-window exit.\n"));
9923 HMVMX_RETURN_UNEXPECTED_EXIT();
9924 }
9925
9926 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS));
9927
9928 /*
9929 * If block-by-STI is set when we get this VM-exit, it means the CPU doesn't block NMIs following STI.
9930 * It is therefore safe to unblock STI and deliver the NMI ourselves. See @bugref{7445}.
9931 */
9932 uint32_t uIntrState = 0;
9933 int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
9934 AssertRCReturn(rc, rc);
9935
9936 bool const fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
9937 if ( fBlockSti
9938 && VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
9939 {
9940 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
9941 }
9942
9943 /* Indicate that we no longer need to VM-exit when the guest is ready to receive NMIs, it is now ready */
9944 hmR0VmxClearNmiWindowExitVmcs(pVCpu);
9945
9946 /* Deliver the pending NMI via hmR0VmxEvaluatePendingEvent() and resume guest execution. */
9947 return VINF_SUCCESS;
9948}
9949
9950
9951/**
9952 * VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
9953 */
9954HMVMX_EXIT_DECL hmR0VmxExitWbinvd(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
9955{
9956 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
9957 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWbinvd);
9958 return hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
9959}
9960
9961
9962/**
9963 * VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit.
9964 */
9965HMVMX_EXIT_DECL hmR0VmxExitInvd(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
9966{
9967 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
9968 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvd);
9969 return hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
9970}
9971
9972
9973/**
9974 * VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit.
9975 */
9976HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
9977{
9978 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
9979 PVM pVM = pVCpu->CTX_SUFF(pVM);
9980 int rc = EMInterpretCpuId(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
9981 if (RT_LIKELY(rc == VINF_SUCCESS))
9982 {
9983 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
9984 Assert(pVmxTransient->cbInstr == 2);
9985 }
9986 else
9987 {
9988 AssertMsgFailed(("hmR0VmxExitCpuid: EMInterpretCpuId failed with %Rrc\n", rc));
9989 rc = VERR_EM_INTERPRETER;
9990 }
9991 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCpuid);
9992 return rc;
9993}
9994
9995
9996/**
9997 * VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit.
9998 */
9999HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10000{
10001 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10002 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
10003 AssertRCReturn(rc, rc);
10004
10005 if (pMixedCtx->cr4 & X86_CR4_SMXE)
10006 return VINF_EM_RAW_EMULATE_INSTR;
10007
10008 AssertMsgFailed(("hmR0VmxExitGetsec: unexpected VM-exit when CR4.SMXE is 0.\n"));
10009 HMVMX_RETURN_UNEXPECTED_EXIT();
10010}
10011
10012
10013/**
10014 * VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
10015 */
10016HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10017{
10018 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10019 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); /** @todo review if CR4 is really required by EM. */
10020 AssertRCReturn(rc, rc);
10021
10022 PVM pVM = pVCpu->CTX_SUFF(pVM);
10023 rc = EMInterpretRdtsc(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10024 if (RT_LIKELY(rc == VINF_SUCCESS))
10025 {
10026 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10027 Assert(pVmxTransient->cbInstr == 2);
10028 /* If we get a spurious VM-exit when offsetting is enabled, we must reset offsetting on VM-reentry. See @bugref{6634}. */
10029 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING)
10030 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
10031 }
10032 else
10033 {
10034 AssertMsgFailed(("hmR0VmxExitRdtsc: EMInterpretRdtsc failed with %Rrc\n", rc));
10035 rc = VERR_EM_INTERPRETER;
10036 }
10037 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc);
10038 return rc;
10039}
10040
10041
10042/**
10043 * VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
10044 */
10045HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10046{
10047 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10048 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); /** @todo review if CR4 is really required by EM. */
10049 rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx); /* For MSR_K8_TSC_AUX */
10050 AssertRCReturn(rc, rc);
10051
10052 PVM pVM = pVCpu->CTX_SUFF(pVM);
10053 rc = EMInterpretRdtscp(pVM, pVCpu, pMixedCtx);
10054 if (RT_LIKELY(rc == VINF_SUCCESS))
10055 {
10056 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10057 Assert(pVmxTransient->cbInstr == 3);
10058 /* If we get a spurious VM-exit when offsetting is enabled, we must reset offsetting on VM-reentry. See @bugref{6634}. */
10059 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING)
10060 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
10061 }
10062 else
10063 {
10064 AssertMsgFailed(("hmR0VmxExitRdtscp: EMInterpretRdtscp failed with %Rrc\n", rc));
10065 rc = VERR_EM_INTERPRETER;
10066 }
10067 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc);
10068 return rc;
10069}
10070
10071
10072/**
10073 * VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
10074 */
10075HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10076{
10077 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10078 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx); /** @todo review if CR4 is really required by EM. */
10079 rc |= hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx); /** @todo review if CR0 is really required by EM. */
10080 AssertRCReturn(rc, rc);
10081
10082 PVM pVM = pVCpu->CTX_SUFF(pVM);
10083 rc = EMInterpretRdpmc(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10084 if (RT_LIKELY(rc == VINF_SUCCESS))
10085 {
10086 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10087 Assert(pVmxTransient->cbInstr == 2);
10088 }
10089 else
10090 {
10091 AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
10092 rc = VERR_EM_INTERPRETER;
10093 }
10094 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdpmc);
10095 return rc;
10096}
10097
10098
10099/**
10100 * VM-exit handler for VMCALL (VMX_EXIT_VMCALL). Unconditional VM-exit.
10101 */
10102HMVMX_EXIT_DECL hmR0VmxExitVmcall(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10103{
10104 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10105
10106 int rc = VERR_NOT_SUPPORTED;
10107 if (GIMAreHypercallsEnabled(pVCpu))
10108 {
10109 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
10110 AssertRCReturn(rc, rc);
10111
10112 rc = GIMHypercall(pVCpu, pMixedCtx);
10113 }
10114 if (rc != VINF_SUCCESS)
10115 {
10116 hmR0VmxSetPendingXcptUD(pVCpu, pMixedCtx);
10117 rc = VINF_SUCCESS;
10118 }
10119
10120 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitVmcall);
10121 return rc;
10122}
10123
10124
10125/**
10126 * VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
10127 */
10128HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10129{
10130 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10131 PVM pVM = pVCpu->CTX_SUFF(pVM);
10132 Assert(!pVM->hm.s.fNestedPaging);
10133
10134 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
10135 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
10136 AssertRCReturn(rc, rc);
10137
10138 VBOXSTRICTRC rc2 = EMInterpretInvlpg(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), pVmxTransient->uExitQualification);
10139 rc = VBOXSTRICTRC_VAL(rc2);
10140 if (RT_LIKELY(rc == VINF_SUCCESS))
10141 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10142 else
10143 {
10144 AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitInvlpg: EMInterpretInvlpg %#RX64 failed with %Rrc\n",
10145 pVmxTransient->uExitQualification, rc));
10146 }
10147 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvlpg);
10148 return rc;
10149}
10150
10151
10152/**
10153 * VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
10154 */
10155HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10156{
10157 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10158 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10159 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10160 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10161 AssertRCReturn(rc, rc);
10162
10163 PVM pVM = pVCpu->CTX_SUFF(pVM);
10164 rc = EMInterpretMonitor(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10165 if (RT_LIKELY(rc == VINF_SUCCESS))
10166 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10167 else
10168 {
10169 AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMonitor: EMInterpretMonitor failed with %Rrc\n", rc));
10170 rc = VERR_EM_INTERPRETER;
10171 }
10172 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMonitor);
10173 return rc;
10174}
10175
10176
10177/**
10178 * VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
10179 */
10180HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10181{
10182 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10183 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10184 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10185 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10186 AssertRCReturn(rc, rc);
10187
10188 PVM pVM = pVCpu->CTX_SUFF(pVM);
10189 VBOXSTRICTRC rc2 = EMInterpretMWait(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10190 rc = VBOXSTRICTRC_VAL(rc2);
10191 if (RT_LIKELY( rc == VINF_SUCCESS
10192 || rc == VINF_EM_HALT))
10193 {
10194 int rc3 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10195 AssertRCReturn(rc3, rc3);
10196
10197 if ( rc == VINF_EM_HALT
10198 && EMMonitorWaitShouldContinue(pVCpu, pMixedCtx))
10199 {
10200 rc = VINF_SUCCESS;
10201 }
10202 }
10203 else
10204 {
10205 AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMwait: EMInterpretMWait failed with %Rrc\n", rc));
10206 rc = VERR_EM_INTERPRETER;
10207 }
10208 AssertMsg(rc == VINF_SUCCESS || rc == VINF_EM_HALT || rc == VERR_EM_INTERPRETER,
10209 ("hmR0VmxExitMwait: failed, invalid error code %Rrc\n", rc));
10210 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMwait);
10211 return rc;
10212}
10213
10214
10215/**
10216 * VM-exit handler for RSM (VMX_EXIT_RSM). Unconditional VM-exit.
10217 */
10218HMVMX_EXIT_DECL hmR0VmxExitRsm(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10219{
10220 /*
10221 * Execution of RSM outside of SMM mode causes #UD regardless of VMX root or VMX non-root mode. In theory, we should never
10222 * get this VM-exit. This can happen only if dual-monitor treatment of SMI and VMX is enabled, which can (only?) be done by
10223 * executing VMCALL in VMX root operation. If we get here, something funny is going on.
10224 * See Intel spec. "33.15.5 Enabling the Dual-Monitor Treatment".
10225 */
10226 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10227 AssertMsgFailed(("Unexpected RSM VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10228 HMVMX_RETURN_UNEXPECTED_EXIT();
10229}
10230
10231
10232/**
10233 * VM-exit handler for SMI (VMX_EXIT_SMI). Unconditional VM-exit.
10234 */
10235HMVMX_EXIT_DECL hmR0VmxExitSmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10236{
10237 /*
10238 * This can only happen if we support dual-monitor treatment of SMI, which can be activated by executing VMCALL in VMX
10239 * root operation. Only an STM (SMM transfer monitor) would get this exit when we (the executive monitor) execute a VMCALL
10240 * in VMX root mode or receive an SMI. If we get here, something funny is going on.
10241 * See Intel spec. "33.15.6 Activating the Dual-Monitor Treatment" and Intel spec. 25.3 "Other Causes of VM-Exits"
10242 */
10243 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10244 AssertMsgFailed(("Unexpected SMI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10245 HMVMX_RETURN_UNEXPECTED_EXIT();
10246}
10247
10248
10249/**
10250 * VM-exit handler for IO SMI (VMX_EXIT_IO_SMI). Unconditional VM-exit.
10251 */
10252HMVMX_EXIT_DECL hmR0VmxExitIoSmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10253{
10254 /* Same treatment as VMX_EXIT_SMI. See comment in hmR0VmxExitSmi(). */
10255 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10256 AssertMsgFailed(("Unexpected IO SMI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10257 HMVMX_RETURN_UNEXPECTED_EXIT();
10258}
10259
10260
10261/**
10262 * VM-exit handler for SIPI (VMX_EXIT_SIPI). Conditional VM-exit.
10263 */
10264HMVMX_EXIT_DECL hmR0VmxExitSipi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10265{
10266 /*
10267 * SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest activity state is used. We currently
10268 * don't make use of it (see hmR0VmxLoadGuestActivityState()) as our guests don't have direct access to the host LAPIC.
10269 * See Intel spec. 25.3 "Other Causes of VM-exits".
10270 */
10271 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10272 AssertMsgFailed(("Unexpected SIPI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10273 HMVMX_RETURN_UNEXPECTED_EXIT();
10274}
10275
10276
10277/**
10278 * VM-exit handler for INIT signal (VMX_EXIT_INIT_SIGNAL). Unconditional
10279 * VM-exit.
10280 */
10281HMVMX_EXIT_DECL hmR0VmxExitInitSignal(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10282{
10283 /*
10284 * INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM.
10285 * See Intel spec. 33.14.1 Default Treatment of SMI Delivery" and Intel spec. 29.3 "VMX Instructions" for "VMXON".
10286 *
10287 * It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these VM-exits.
10288 * See Intel spec. "23.8 Restrictions on VMX operation".
10289 */
10290 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10291 return VINF_SUCCESS;
10292}
10293
10294
10295/**
10296 * VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional
10297 * VM-exit.
10298 */
10299HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10300{
10301 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10302 return VINF_EM_RESET;
10303}
10304
10305
10306/**
10307 * VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
10308 */
10309HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10310{
10311 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10312 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT);
10313 int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
10314 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10315 AssertRCReturn(rc, rc);
10316
10317 pMixedCtx->rip++;
10318 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
10319 if (EMShouldContinueAfterHalt(pVCpu, pMixedCtx)) /* Requires eflags. */
10320 rc = VINF_SUCCESS;
10321 else
10322 rc = VINF_EM_HALT;
10323
10324 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt);
10325 return rc;
10326}
10327
10328
10329/**
10330 * VM-exit handler for instructions that result in a #UD exception delivered to
10331 * the guest.
10332 */
10333HMVMX_EXIT_DECL hmR0VmxExitSetPendingXcptUD(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10334{
10335 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10336 hmR0VmxSetPendingXcptUD(pVCpu, pMixedCtx);
10337 return VINF_SUCCESS;
10338}
10339
10340
10341/**
10342 * VM-exit handler for expiry of the VMX preemption timer.
10343 */
10344HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10345{
10346 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10347
10348 /* If the preemption-timer has expired, reinitialize the preemption timer on next VM-entry. */
10349 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
10350
10351 /* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */
10352 PVM pVM = pVCpu->CTX_SUFF(pVM);
10353 bool fTimersPending = TMTimerPollBool(pVM, pVCpu);
10354 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer);
10355 return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS;
10356}
10357
10358
10359/**
10360 * VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit.
10361 */
10362HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10363{
10364 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10365
10366 /* We expose XSETBV to the guest, fallback to the recompiler for emulation. */
10367 /** @todo check if XSETBV is supported by the recompiler. */
10368 return VERR_EM_INTERPRETER;
10369}
10370
10371
10372/**
10373 * VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
10374 */
10375HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10376{
10377 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10378
10379 /* The guest should not invalidate the host CPU's TLBs, fallback to recompiler. */
10380 /** @todo implement EMInterpretInvpcid() */
10381 return VERR_EM_INTERPRETER;
10382}
10383
10384
10385/**
10386 * VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE).
10387 * Error VM-exit.
10388 */
10389HMVMX_EXIT_DECL hmR0VmxExitErrInvalidGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10390{
10391 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
10392 AssertRCReturn(rc, rc);
10393
10394 rc = hmR0VmxCheckVmcsCtls(pVCpu);
10395 AssertRCReturn(rc, rc);
10396
10397 uint32_t uInvalidReason = hmR0VmxCheckGuestState(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx);
10398 NOREF(uInvalidReason);
10399
10400#ifdef VBOX_STRICT
10401 uint32_t uIntrState;
10402 HMVMXHCUINTREG uHCReg;
10403 uint64_t u64Val;
10404 uint32_t u32Val;
10405
10406 rc = hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
10407 rc |= hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient);
10408 rc |= hmR0VmxReadEntryInstrLenVmcs(pVmxTransient);
10409 rc |= VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
10410 AssertRCReturn(rc, rc);
10411
10412 Log4(("uInvalidReason %u\n", uInvalidReason));
10413 Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo));
10414 Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode));
10415 Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr));
10416 Log4(("VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE %#RX32\n", uIntrState));
10417
10418 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32Val); AssertRC(rc);
10419 Log4(("VMX_VMCS_GUEST_CR0 %#RX32\n", u32Val));
10420 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc);
10421 Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg));
10422 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc);
10423 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
10424 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc);
10425 Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg));
10426 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc);
10427 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
10428 rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
10429 Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
10430#else
10431 NOREF(pVmxTransient);
10432#endif
10433
10434 HMDumpRegs(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx);
10435 return VERR_VMX_INVALID_GUEST_STATE;
10436}
10437
10438
10439/**
10440 * VM-exit handler for VM-entry failure due to an MSR-load
10441 * (VMX_EXIT_ERR_MSR_LOAD). Error VM-exit.
10442 */
10443HMVMX_EXIT_DECL hmR0VmxExitErrMsrLoad(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10444{
10445 NOREF(pVmxTransient);
10446 AssertMsgFailed(("Unexpected MSR-load exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); NOREF(pMixedCtx);
10447 HMVMX_RETURN_UNEXPECTED_EXIT();
10448}
10449
10450
10451/**
10452 * VM-exit handler for VM-entry failure due to a machine-check event
10453 * (VMX_EXIT_ERR_MACHINE_CHECK). Error VM-exit.
10454 */
10455HMVMX_EXIT_DECL hmR0VmxExitErrMachineCheck(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10456{
10457 NOREF(pVmxTransient);
10458 AssertMsgFailed(("Unexpected machine-check event exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); NOREF(pMixedCtx);
10459 HMVMX_RETURN_UNEXPECTED_EXIT();
10460}
10461
10462
10463/**
10464 * VM-exit handler for all undefined reasons. Should never ever happen.. in
10465 * theory.
10466 */
10467HMVMX_EXIT_DECL hmR0VmxExitErrUndefined(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10468{
10469 AssertMsgFailed(("Huh!? Undefined VM-exit reason %d. pVCpu=%p pMixedCtx=%p\n", pVmxTransient->uExitReason, pVCpu, pMixedCtx));
10470 NOREF(pVCpu); NOREF(pMixedCtx); NOREF(pVmxTransient);
10471 return VERR_VMX_UNDEFINED_EXIT_CODE;
10472}
10473
10474
10475/**
10476 * VM-exit handler for XDTR (LGDT, SGDT, LIDT, SIDT) accesses
10477 * (VMX_EXIT_XDTR_ACCESS) and LDT and TR access (LLDT, LTR, SLDT, STR).
10478 * Conditional VM-exit.
10479 */
10480HMVMX_EXIT_DECL hmR0VmxExitXdtrAccess(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10481{
10482 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10483
10484 /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT. */
10485 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitXdtrAccess);
10486 if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT)
10487 return VERR_EM_INTERPRETER;
10488 AssertMsgFailed(("Unexpected XDTR access. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10489 HMVMX_RETURN_UNEXPECTED_EXIT();
10490}
10491
10492
10493/**
10494 * VM-exit handler for RDRAND (VMX_EXIT_RDRAND). Conditional VM-exit.
10495 */
10496HMVMX_EXIT_DECL hmR0VmxExitRdrand(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10497{
10498 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10499
10500 /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT. */
10501 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdrand);
10502 if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT)
10503 return VERR_EM_INTERPRETER;
10504 AssertMsgFailed(("Unexpected RDRAND exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10505 HMVMX_RETURN_UNEXPECTED_EXIT();
10506}
10507
10508
10509/**
10510 * VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
10511 */
10512HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10513{
10514 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10515
10516 /* EMInterpretRdmsr() requires CR0, Eflags and SS segment register. */
10517 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10518 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10519 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10520 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
10521 {
10522 rc |= hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
10523 rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
10524 }
10525 AssertRCReturn(rc, rc);
10526 Log4(("CS:RIP=%04x:%#RX64 ECX=%X\n", pMixedCtx->cs.Sel, pMixedCtx->rip, pMixedCtx->ecx));
10527
10528#ifdef VBOX_STRICT
10529 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
10530 {
10531 if ( hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx)
10532 && pMixedCtx->ecx != MSR_K6_EFER)
10533 {
10534 AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n", pMixedCtx->ecx));
10535 HMVMX_RETURN_UNEXPECTED_EXIT();
10536 }
10537# if HC_ARCH_BITS == 64
10538 if ( pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests
10539 && hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
10540 {
10541 AssertMsgFailed(("Unexpected RDMSR for a passthru lazy-restore MSR. ecx=%#RX32\n", pMixedCtx->ecx));
10542 HMVMX_RETURN_UNEXPECTED_EXIT();
10543 }
10544# endif
10545 }
10546#endif
10547
10548 PVM pVM = pVCpu->CTX_SUFF(pVM);
10549 rc = EMInterpretRdmsr(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10550 AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER,
10551 ("hmR0VmxExitRdmsr: failed, invalid error code %Rrc\n", rc));
10552 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
10553 if (RT_LIKELY(rc == VINF_SUCCESS))
10554 {
10555 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10556 Assert(pVmxTransient->cbInstr == 2);
10557 }
10558 return rc;
10559}
10560
10561
10562/**
10563 * VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
10564 */
10565HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10566{
10567 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10568 PVM pVM = pVCpu->CTX_SUFF(pVM);
10569 int rc = VINF_SUCCESS;
10570
10571 /* EMInterpretWrmsr() requires CR0, EFLAGS and SS segment register. */
10572 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10573 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10574 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10575 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
10576 {
10577 rc |= hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
10578 rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
10579 }
10580 AssertRCReturn(rc, rc);
10581 Log4(("ecx=%#RX32\n", pMixedCtx->ecx));
10582
10583 rc = EMInterpretWrmsr(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
10584 AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER, ("hmR0VmxExitWrmsr: failed, invalid error code %Rrc\n", rc));
10585 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
10586
10587 if (RT_LIKELY(rc == VINF_SUCCESS))
10588 {
10589 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10590
10591 /* If this is an X2APIC WRMSR access, update the APIC state as well. */
10592 if ( pMixedCtx->ecx >= MSR_IA32_X2APIC_START
10593 && pMixedCtx->ecx <= MSR_IA32_X2APIC_END)
10594 {
10595 /* We've already saved the APIC related guest-state (TPR) in hmR0VmxPostRunGuest(). When full APIC register
10596 * virtualization is implemented we'll have to make sure APIC state is saved from the VMCS before
10597 EMInterpretWrmsr() changes it. */
10598 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
10599 }
10600 else if (pMixedCtx->ecx == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */
10601 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
10602 else if (pMixedCtx->ecx == MSR_K6_EFER)
10603 {
10604 /*
10605 * If the guest touches EFER we need to update the VM-Entry and VM-Exit controls as well,
10606 * even if it is -not- touching bits that cause paging mode changes (LMA/LME). We care about
10607 * the other bits as well, SCE and NXE. See @bugref{7368}.
10608 */
10609 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_EFER_MSR | HM_CHANGED_VMX_ENTRY_CTLS | HM_CHANGED_VMX_EXIT_CTLS);
10610 }
10611
10612 /* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not supported. */
10613 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
10614 {
10615 switch (pMixedCtx->ecx)
10616 {
10617 case MSR_IA32_SYSENTER_CS: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
10618 case MSR_IA32_SYSENTER_EIP: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
10619 case MSR_IA32_SYSENTER_ESP: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
10620 case MSR_K8_FS_BASE: /* no break */
10621 case MSR_K8_GS_BASE: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS); break;
10622 case MSR_K6_EFER: /* already handled above */ break;
10623 default:
10624 {
10625 if (hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx))
10626 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
10627#if HC_ARCH_BITS == 64
10628 else if (hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
10629 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS);
10630#endif
10631 break;
10632 }
10633 }
10634 }
10635#ifdef VBOX_STRICT
10636 else
10637 {
10638 /* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */
10639 switch (pMixedCtx->ecx)
10640 {
10641 case MSR_IA32_SYSENTER_CS:
10642 case MSR_IA32_SYSENTER_EIP:
10643 case MSR_IA32_SYSENTER_ESP:
10644 case MSR_K8_FS_BASE:
10645 case MSR_K8_GS_BASE:
10646 {
10647 AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", pMixedCtx->ecx));
10648 HMVMX_RETURN_UNEXPECTED_EXIT();
10649 }
10650
10651 /* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */
10652 default:
10653 {
10654 if (hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx))
10655 {
10656 /* EFER writes are always intercepted, see hmR0VmxLoadGuestMsrs(). */
10657 if (pMixedCtx->ecx != MSR_K6_EFER)
10658 {
10659 AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
10660 pMixedCtx->ecx));
10661 HMVMX_RETURN_UNEXPECTED_EXIT();
10662 }
10663 }
10664
10665#if HC_ARCH_BITS == 64
10666 if (hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
10667 {
10668 AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", pMixedCtx->ecx));
10669 HMVMX_RETURN_UNEXPECTED_EXIT();
10670 }
10671#endif
10672 break;
10673 }
10674 }
10675 }
10676#endif /* VBOX_STRICT */
10677 }
10678 return rc;
10679}
10680
10681
10682/**
10683 * VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
10684 */
10685HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10686{
10687 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10688
10689 /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC_PAUSE_EXIT. */
10690 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPause);
10691 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_PAUSE_EXIT)
10692 return VERR_EM_INTERPRETER;
10693 AssertMsgFailed(("Unexpected PAUSE exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
10694 HMVMX_RETURN_UNEXPECTED_EXIT();
10695}
10696
10697
10698/**
10699 * VM-exit handler for when the TPR value is lowered below the specified
10700 * threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
10701 */
10702HMVMX_EXIT_DECL hmR0VmxExitTprBelowThreshold(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10703{
10704 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10705 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW);
10706
10707 /*
10708 * The TPR has already been updated, see hmR0VMXPostRunGuest(). RIP is also updated as part of the VM-exit by VT-x. Update
10709 * the threshold in the VMCS, deliver the pending interrupt via hmR0VmxPreRunGuest()->hmR0VmxInjectPendingEvent() and
10710 * resume guest execution.
10711 */
10712 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
10713 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold);
10714 return VINF_SUCCESS;
10715}
10716
10717
10718/**
10719 * VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional
10720 * VM-exit.
10721 *
10722 * @retval VINF_SUCCESS when guest execution can continue.
10723 * @retval VINF_PGM_CHANGE_MODE when shadow paging mode changed, back to ring-3.
10724 * @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3.
10725 * @retval VERR_EM_INTERPRETER when something unexpected happened, fallback to
10726 * recompiler.
10727 */
10728HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10729{
10730 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10731 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2);
10732 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
10733 AssertRCReturn(rc, rc);
10734
10735 RTGCUINTPTR const uExitQualification = pVmxTransient->uExitQualification;
10736 uint32_t const uAccessType = VMX_EXIT_QUALIFICATION_CRX_ACCESS(uExitQualification);
10737 PVM pVM = pVCpu->CTX_SUFF(pVM);
10738 switch (uAccessType)
10739 {
10740 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_WRITE: /* MOV to CRx */
10741 {
10742#if 0
10743 /* EMInterpretCRxWrite() references a lot of guest state (EFER, RFLAGS, Segment Registers, etc.) Sync entire state */
10744 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
10745#else
10746 rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
10747 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
10748 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10749#endif
10750 AssertRCReturn(rc, rc);
10751
10752 rc = EMInterpretCRxWrite(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx),
10753 VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification),
10754 VMX_EXIT_QUALIFICATION_CRX_GENREG(uExitQualification));
10755 Assert(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER || rc == VINF_PGM_CHANGE_MODE || rc == VINF_PGM_SYNC_CR3);
10756
10757 switch (VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification))
10758 {
10759 case 0: /* CR0 */
10760 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
10761 Log4(("CRX CR0 write rc=%d CR0=%#RX64\n", rc, pMixedCtx->cr0));
10762 break;
10763 case 2: /* CR2 */
10764 /* Nothing to do here, CR2 it's not part of the VMCS. */
10765 break;
10766 case 3: /* CR3 */
10767 Assert(!pVM->hm.s.fNestedPaging || !CPUMIsGuestPagingEnabledEx(pMixedCtx));
10768 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR3);
10769 Log4(("CRX CR3 write rc=%d CR3=%#RX64\n", rc, pMixedCtx->cr3));
10770 break;
10771 case 4: /* CR4 */
10772 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR4);
10773 Log4(("CRX CR4 write rc=%d CR4=%#RX64\n", rc, pMixedCtx->cr4));
10774 break;
10775 case 8: /* CR8 */
10776 Assert(!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW));
10777 /* CR8 contains the APIC TPR. Was updated by EMInterpretCRxWrite(). */
10778 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
10779 break;
10780 default:
10781 AssertMsgFailed(("Invalid CRx register %#x\n", VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)));
10782 break;
10783 }
10784
10785 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxWrite[VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)]);
10786 break;
10787 }
10788
10789 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_READ: /* MOV from CRx */
10790 {
10791 /* EMInterpretCRxRead() requires EFER MSR, CS. */
10792 rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10793 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
10794 AssertRCReturn(rc, rc);
10795 Assert( !pVM->hm.s.fNestedPaging
10796 || !CPUMIsGuestPagingEnabledEx(pMixedCtx)
10797 || VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification) != 3);
10798
10799 /* CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */
10800 Assert( VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification) != 8
10801 || !(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW));
10802
10803 rc = EMInterpretCRxRead(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx),
10804 VMX_EXIT_QUALIFICATION_CRX_GENREG(uExitQualification),
10805 VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification));
10806 Assert(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER);
10807 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxRead[VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)]);
10808 Log4(("CRX CR%d Read access rc=%d\n", VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification), rc));
10809 break;
10810 }
10811
10812 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_CLTS: /* CLTS (Clear Task-Switch Flag in CR0) */
10813 {
10814 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10815 AssertRCReturn(rc, rc);
10816 rc = EMInterpretCLTS(pVM, pVCpu);
10817 AssertRCReturn(rc, rc);
10818 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
10819 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts);
10820 Log4(("CRX CLTS write rc=%d\n", rc));
10821 break;
10822 }
10823
10824 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */
10825 {
10826 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
10827 AssertRCReturn(rc, rc);
10828 rc = EMInterpretLMSW(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), VMX_EXIT_QUALIFICATION_CRX_LMSW_DATA(uExitQualification));
10829 if (RT_LIKELY(rc == VINF_SUCCESS))
10830 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
10831 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw);
10832 Log4(("CRX LMSW write rc=%d\n", rc));
10833 break;
10834 }
10835
10836 default:
10837 {
10838 AssertMsgFailed(("Invalid access-type in Mov CRx exit qualification %#x\n", uAccessType));
10839 rc = VERR_VMX_UNEXPECTED_EXCEPTION;
10840 }
10841 }
10842
10843 /* Validate possible error codes. */
10844 Assert(rc == VINF_SUCCESS || rc == VINF_PGM_CHANGE_MODE || rc == VERR_EM_INTERPRETER || rc == VINF_PGM_SYNC_CR3
10845 || rc == VERR_VMX_UNEXPECTED_EXCEPTION);
10846 if (RT_SUCCESS(rc))
10847 {
10848 int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
10849 AssertRCReturn(rc2, rc2);
10850 }
10851
10852 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2);
10853 return rc;
10854}
10855
10856
10857/**
10858 * VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional
10859 * VM-exit.
10860 */
10861HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10862{
10863 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
10864 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1);
10865
10866 int rc2 = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
10867 rc2 |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
10868 rc2 |= hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
10869 rc2 |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); /* Eflag checks in EMInterpretDisasCurrent(). */
10870 rc2 |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); /* CR0 checks & PGM* in EMInterpretDisasCurrent(). */
10871 rc2 |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); /* SELM checks in EMInterpretDisasCurrent(). */
10872 /* EFER also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */
10873 AssertRCReturn(rc2, rc2);
10874
10875 /* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */
10876 uint32_t uIOPort = VMX_EXIT_QUALIFICATION_IO_PORT(pVmxTransient->uExitQualification);
10877 uint8_t uIOWidth = VMX_EXIT_QUALIFICATION_IO_WIDTH(pVmxTransient->uExitQualification);
10878 bool fIOWrite = ( VMX_EXIT_QUALIFICATION_IO_DIRECTION(pVmxTransient->uExitQualification)
10879 == VMX_EXIT_QUALIFICATION_IO_DIRECTION_OUT);
10880 bool fIOString = VMX_EXIT_QUALIFICATION_IO_IS_STRING(pVmxTransient->uExitQualification);
10881 bool fStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
10882 AssertReturn(uIOWidth <= 3 && uIOWidth != 2, VERR_VMX_IPE_1);
10883
10884 /* I/O operation lookup arrays. */
10885 static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses. */
10886 static uint32_t const s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving the result (in AL/AX/EAX). */
10887
10888 VBOXSTRICTRC rcStrict;
10889 uint32_t const cbValue = s_aIOSizes[uIOWidth];
10890 uint32_t const cbInstr = pVmxTransient->cbInstr;
10891 bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */
10892 PVM pVM = pVCpu->CTX_SUFF(pVM);
10893 if (fIOString)
10894 {
10895#if 0 /* Not yet ready. IEM gurus with debian 32-bit guest without NP (on ATA reads). See @bugref{5752#c158} */
10896 /*
10897 * INS/OUTS - I/O String instruction.
10898 *
10899 * Use instruction-information if available, otherwise fall back on
10900 * interpreting the instruction.
10901 */
10902 Log4(("CS:RIP=%04x:%#RX64 %#06x/%u %c str\n", pMixedCtx->cs.Sel, pMixedCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
10903 AssertReturn(pMixedCtx->dx == uIOPort, VERR_VMX_IPE_2);
10904 if (MSR_IA32_VMX_BASIC_INFO_VMCS_INS_OUTS(pVM->hm.s.vmx.Msrs.u64BasicInfo))
10905 {
10906 rc2 = hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
10907 /** @todo optimize this, IEM should request the additional state if it needs it (GP, PF, ++). */
10908 rc2 |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
10909 AssertRCReturn(rc2, rc2);
10910 AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_VMX_IPE_3);
10911 AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2);
10912 IEMMODE enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize;
10913 bool fRep = VMX_EXIT_QUALIFICATION_IO_IS_REP(pVmxTransient->uExitQualification);
10914 if (fIOWrite)
10915 {
10916 rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr,
10917 pVmxTransient->ExitInstrInfo.StrIo.iSegReg);
10918 }
10919 else
10920 {
10921 /*
10922 * The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES.
10923 * Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS.
10924 * See Intel Instruction spec. for "INS".
10925 * See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS".
10926 */
10927 rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr);
10928 }
10929 }
10930 else
10931 {
10932 /** @todo optimize this, IEM should request the additional state if it needs it (GP, PF, ++). */
10933 rc2 = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
10934 AssertRCReturn(rc2, rc2);
10935 rcStrict = IEMExecOne(pVCpu);
10936 }
10937 /** @todo IEM needs to be setting these flags somehow. */
10938 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
10939 fUpdateRipAlready = true;
10940#else
10941 PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
10942 rcStrict = EMInterpretDisasCurrent(pVM, pVCpu, pDis, NULL /* pcbInstr */);
10943 if (RT_SUCCESS(rcStrict))
10944 {
10945 if (fIOWrite)
10946 {
10947 rcStrict = IOMInterpretOUTSEx(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), uIOPort, pDis->fPrefix,
10948 (DISCPUMODE)pDis->uAddrMode, cbValue);
10949 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringWrite);
10950 }
10951 else
10952 {
10953 rcStrict = IOMInterpretINSEx(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), uIOPort, pDis->fPrefix,
10954 (DISCPUMODE)pDis->uAddrMode, cbValue);
10955 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringRead);
10956 }
10957 }
10958 else
10959 {
10960 AssertMsg(rcStrict == VERR_EM_INTERPRETER, ("rcStrict=%Rrc RIP %#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->rip));
10961 rcStrict = VINF_EM_RAW_EMULATE_INSTR;
10962 }
10963#endif
10964 }
10965 else
10966 {
10967 /*
10968 * IN/OUT - I/O instruction.
10969 */
10970 Log4(("CS:RIP=%04x:%#RX64 %#06x/%u %c\n", pMixedCtx->cs.Sel, pMixedCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
10971 uint32_t const uAndVal = s_aIOOpAnd[uIOWidth];
10972 Assert(!VMX_EXIT_QUALIFICATION_IO_IS_REP(pVmxTransient->uExitQualification));
10973 if (fIOWrite)
10974 {
10975 rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pMixedCtx->eax & uAndVal, cbValue);
10976 if (rcStrict == VINF_IOM_R3_IOPORT_WRITE)
10977 HMR0SavePendingIOPortWrite(pVCpu, pMixedCtx->rip, pMixedCtx->rip + cbInstr, uIOPort, uAndVal, cbValue);
10978 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
10979 }
10980 else
10981 {
10982 uint32_t u32Result = 0;
10983 rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue);
10984 if (IOM_SUCCESS(rcStrict))
10985 {
10986 /* Save result of I/O IN instr. in AL/AX/EAX. */
10987 pMixedCtx->eax = (pMixedCtx->eax & ~uAndVal) | (u32Result & uAndVal);
10988 }
10989 else if (rcStrict == VINF_IOM_R3_IOPORT_READ)
10990 HMR0SavePendingIOPortRead(pVCpu, pMixedCtx->rip, pMixedCtx->rip + cbInstr, uIOPort, uAndVal, cbValue);
10991 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
10992 }
10993 }
10994
10995 if (IOM_SUCCESS(rcStrict))
10996 {
10997 if (!fUpdateRipAlready)
10998 {
10999 pMixedCtx->rip += cbInstr;
11000 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
11001 }
11002
11003 /*
11004 * INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru while booting Fedora 17 64-bit guest.
11005 * See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ.
11006 */
11007 if (fIOString)
11008 {
11009 /** @todo Single-step for INS/OUTS with REP prefix? */
11010 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
11011 }
11012 else if (fStepping)
11013 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
11014
11015 /*
11016 * If any I/O breakpoints are armed, we need to check if one triggered
11017 * and take appropriate action.
11018 * Note that the I/O breakpoint type is undefined if CR4.DE is 0.
11019 */
11020 rc2 = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
11021 AssertRCReturn(rc2, rc2);
11022
11023 /** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
11024 * execution engines about whether hyper BPs and such are pending. */
11025 uint32_t const uDr7 = pMixedCtx->dr[7];
11026 if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
11027 && X86_DR7_ANY_RW_IO(uDr7)
11028 && (pMixedCtx->cr4 & X86_CR4_DE))
11029 || DBGFBpIsHwIoArmed(pVM)))
11030 {
11031 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
11032
11033 /* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
11034 VMMRZCallRing3Disable(pVCpu);
11035 HM_DISABLE_PREEMPT_IF_NEEDED();
11036
11037 bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /*fDr6*/);
11038
11039 VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pMixedCtx, uIOPort, cbValue);
11040 if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
11041 {
11042 /* Raise #DB. */
11043 if (fIsGuestDbgActive)
11044 ASMSetDR6(pMixedCtx->dr[6]);
11045 if (pMixedCtx->dr[7] != uDr7)
11046 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
11047
11048 hmR0VmxSetPendingXcptDB(pVCpu, pMixedCtx);
11049 }
11050 /* rcStrict is VINF_SUCCESS or in [VINF_EM_FIRST..VINF_EM_LAST]. */
11051 else if ( rcStrict2 != VINF_SUCCESS
11052 && (rcStrict == VINF_SUCCESS || rcStrict2 < rcStrict))
11053 rcStrict = rcStrict2;
11054
11055 HM_RESTORE_PREEMPT_IF_NEEDED();
11056 VMMRZCallRing3Enable(pVCpu);
11057 }
11058 }
11059
11060#ifdef DEBUG
11061 if (rcStrict == VINF_IOM_R3_IOPORT_READ)
11062 Assert(!fIOWrite);
11063 else if (rcStrict == VINF_IOM_R3_IOPORT_WRITE)
11064 Assert(fIOWrite);
11065 else
11066 {
11067 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
11068 * statuses, that the VMM device and some others may return. See
11069 * IOM_SUCCESS() for guidance. */
11070 AssertMsg( RT_FAILURE(rcStrict)
11071 || rcStrict == VINF_SUCCESS
11072 || rcStrict == VINF_EM_RAW_EMULATE_INSTR
11073 || rcStrict == VINF_EM_DBG_BREAKPOINT
11074 || rcStrict == VINF_EM_RAW_GUEST_TRAP
11075 || rcStrict == VINF_TRPM_XCPT_DISPATCHED, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
11076 }
11077#endif
11078
11079 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1);
11080 return VBOXSTRICTRC_TODO(rcStrict);
11081}
11082
11083
11084/**
11085 * VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional
11086 * VM-exit.
11087 */
11088HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11089{
11090 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11091
11092 /* Check if this task-switch occurred while delivery an event through the guest IDT. */
11093 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11094 AssertRCReturn(rc, rc);
11095 if (VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE_IDT)
11096 {
11097 rc = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
11098 AssertRCReturn(rc, rc);
11099 if (VMX_IDT_VECTORING_INFO_VALID(pVmxTransient->uIdtVectoringInfo))
11100 {
11101 uint32_t uIntType = VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo);
11102
11103 uint32_t uVector = VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo);
11104 bool fErrorCodeValid = VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uIdtVectoringInfo);
11105
11106 /* Save it as a pending event and it'll be converted to a TRPM event on the way out to ring-3. */
11107 Assert(!pVCpu->hm.s.Event.fPending);
11108 pVCpu->hm.s.Event.fPending = true;
11109 pVCpu->hm.s.Event.u64IntInfo = pVmxTransient->uIdtVectoringInfo;
11110 rc = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
11111 AssertRCReturn(rc, rc);
11112 if (fErrorCodeValid)
11113 pVCpu->hm.s.Event.u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
11114 else
11115 pVCpu->hm.s.Event.u32ErrCode = 0;
11116 if ( uIntType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
11117 && uVector == X86_XCPT_PF)
11118 {
11119 pVCpu->hm.s.Event.GCPtrFaultAddress = pMixedCtx->cr2;
11120 }
11121
11122 Log4(("Pending event on TaskSwitch uIntType=%#x uVector=%#x\n", uIntType, uVector));
11123 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
11124 return VINF_EM_RAW_INJECT_TRPM_EVENT;
11125 }
11126 }
11127
11128 /** @todo Emulate task switch someday, currently just going back to ring-3 for
11129 * emulation. */
11130 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
11131 return VERR_EM_INTERPRETER;
11132}
11133
11134
11135/**
11136 * VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit.
11137 */
11138HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11139{
11140 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11141 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG);
11142 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG;
11143 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
11144 AssertRCReturn(rc, rc);
11145 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMtf);
11146 return VINF_EM_DBG_STEPPED;
11147}
11148
11149
11150/**
11151 * VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit.
11152 */
11153HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11154{
11155 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11156
11157 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
11158 int rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
11159 if (RT_UNLIKELY(rc == VINF_HM_DOUBLE_FAULT))
11160 return VINF_SUCCESS;
11161 else if (RT_UNLIKELY(rc == VINF_EM_RESET))
11162 return rc;
11163
11164#if 0
11165 /** @todo Investigate if IOMMMIOPhysHandler() requires a lot of state, for now
11166 * just sync the whole thing. */
11167 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11168#else
11169 /* Aggressive state sync. for now. */
11170 rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
11171 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
11172 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11173#endif
11174 rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11175 AssertRCReturn(rc, rc);
11176
11177 /* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */
11178 uint32_t uAccessType = VMX_EXIT_QUALIFICATION_APIC_ACCESS_TYPE(pVmxTransient->uExitQualification);
11179 switch (uAccessType)
11180 {
11181 case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE:
11182 case VMX_APIC_ACCESS_TYPE_LINEAR_READ:
11183 {
11184 if ( (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
11185 && VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification) == 0x80)
11186 {
11187 AssertMsgFailed(("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n"));
11188 }
11189
11190 RTGCPHYS GCPhys = pMixedCtx->msrApicBase; /* Always up-to-date, msrApicBase is not part of the VMCS. */
11191 GCPhys &= PAGE_BASE_GC_MASK;
11192 GCPhys += VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification);
11193 PVM pVM = pVCpu->CTX_SUFF(pVM);
11194 Log4(("ApicAccess uAccessType=%#x GCPhys=%#RGv Off=%#x\n", uAccessType, GCPhys,
11195 VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification)));
11196
11197 VBOXSTRICTRC rc2 = IOMMMIOPhysHandler(pVM, pVCpu,
11198 (uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ) ? 0 : X86_TRAP_PF_RW,
11199 CPUMCTX2CORE(pMixedCtx), GCPhys);
11200 rc = VBOXSTRICTRC_VAL(rc2);
11201 Log4(("ApicAccess rc=%d\n", rc));
11202 if ( rc == VINF_SUCCESS
11203 || rc == VERR_PAGE_TABLE_NOT_PRESENT
11204 || rc == VERR_PAGE_NOT_PRESENT)
11205 {
11206 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
11207 | HM_CHANGED_GUEST_RSP
11208 | HM_CHANGED_GUEST_RFLAGS
11209 | HM_CHANGED_VMX_GUEST_APIC_STATE);
11210 rc = VINF_SUCCESS;
11211 }
11212 break;
11213 }
11214
11215 default:
11216 Log4(("ApicAccess uAccessType=%#x\n", uAccessType));
11217 rc = VINF_EM_RAW_EMULATE_INSTR;
11218 break;
11219 }
11220
11221 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess);
11222 return rc;
11223}
11224
11225
11226/**
11227 * VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional
11228 * VM-exit.
11229 */
11230HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11231{
11232 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11233
11234 /* We should -not- get this VM-exit if the guest's debug registers were active. */
11235 if (pVmxTransient->fWasGuestDebugStateActive)
11236 {
11237 AssertMsgFailed(("Unexpected MOV DRx exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
11238 HMVMX_RETURN_UNEXPECTED_EXIT();
11239 }
11240
11241 int rc = VERR_INTERNAL_ERROR_5;
11242 if ( !DBGFIsStepping(pVCpu)
11243 && !pVCpu->hm.s.fSingleInstruction
11244 && !pVmxTransient->fWasHyperDebugStateActive)
11245 {
11246 /* Don't intercept MOV DRx and #DB any more. */
11247 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
11248 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
11249 AssertRCReturn(rc, rc);
11250
11251 if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
11252 {
11253#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
11254 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_DB);
11255 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
11256 AssertRCReturn(rc, rc);
11257#endif
11258 }
11259
11260 /* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */
11261 VMMRZCallRing3Disable(pVCpu);
11262 HM_DISABLE_PREEMPT_IF_NEEDED();
11263
11264 /* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
11265 CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
11266 Assert(CPUMIsGuestDebugStateActive(pVCpu) || HC_ARCH_BITS == 32);
11267
11268 HM_RESTORE_PREEMPT_IF_NEEDED();
11269 VMMRZCallRing3Enable(pVCpu);
11270
11271#ifdef VBOX_WITH_STATISTICS
11272 rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11273 AssertRCReturn(rc, rc);
11274 if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
11275 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
11276 else
11277 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
11278#endif
11279 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
11280 return VINF_SUCCESS;
11281 }
11282
11283 /*
11284 * EMInterpretDRx[Write|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER, CS. EFER is always up-to-date.
11285 * Update the segment registers and DR7 from the CPU.
11286 */
11287 rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11288 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11289 rc |= hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
11290 AssertRCReturn(rc, rc);
11291 Log4(("CS:RIP=%04x:%#RX64\n", pMixedCtx->cs.Sel, pMixedCtx->rip));
11292
11293 PVM pVM = pVCpu->CTX_SUFF(pVM);
11294 if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
11295 {
11296 rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx),
11297 VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification),
11298 VMX_EXIT_QUALIFICATION_DRX_GENREG(pVmxTransient->uExitQualification));
11299 if (RT_SUCCESS(rc))
11300 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
11301 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
11302 }
11303 else
11304 {
11305 rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx),
11306 VMX_EXIT_QUALIFICATION_DRX_GENREG(pVmxTransient->uExitQualification),
11307 VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification));
11308 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
11309 }
11310
11311 Assert(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER);
11312 if (RT_SUCCESS(rc))
11313 {
11314 int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11315 AssertRCReturn(rc2, rc2);
11316 }
11317 return rc;
11318}
11319
11320
11321/**
11322 * VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG).
11323 * Conditional VM-exit.
11324 */
11325HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11326{
11327 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11328 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
11329
11330 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
11331 int rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
11332 if (RT_UNLIKELY(rc == VINF_HM_DOUBLE_FAULT))
11333 return VINF_SUCCESS;
11334 else if (RT_UNLIKELY(rc == VINF_EM_RESET))
11335 return rc;
11336
11337 RTGCPHYS GCPhys = 0;
11338 rc = VMXReadVmcs64(VMX_VMCS64_EXIT_GUEST_PHYS_ADDR_FULL, &GCPhys);
11339
11340#if 0
11341 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); /** @todo Can we do better? */
11342#else
11343 /* Aggressive state sync. for now. */
11344 rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
11345 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
11346 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11347#endif
11348 AssertRCReturn(rc, rc);
11349
11350 /*
11351 * If we succeed, resume guest execution.
11352 * If we fail in interpreting the instruction because we couldn't get the guest physical address
11353 * of the page containing the instruction via the guest's page tables (we would invalidate the guest page
11354 * in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
11355 * weird case. See @bugref{6043}.
11356 */
11357 PVM pVM = pVCpu->CTX_SUFF(pVM);
11358 VBOXSTRICTRC rc2 = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pMixedCtx), GCPhys, UINT32_MAX);
11359 rc = VBOXSTRICTRC_VAL(rc2);
11360 Log4(("EPT misconfig at %#RGv RIP=%#RX64 rc=%d\n", GCPhys, pMixedCtx->rip, rc));
11361 if ( rc == VINF_SUCCESS
11362 || rc == VERR_PAGE_TABLE_NOT_PRESENT
11363 || rc == VERR_PAGE_NOT_PRESENT)
11364 {
11365 /* Successfully handled MMIO operation. */
11366 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
11367 | HM_CHANGED_GUEST_RSP
11368 | HM_CHANGED_GUEST_RFLAGS
11369 | HM_CHANGED_VMX_GUEST_APIC_STATE);
11370 rc = VINF_SUCCESS;
11371 }
11372 return rc;
11373}
11374
11375
11376/**
11377 * VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional
11378 * VM-exit.
11379 */
11380HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11381{
11382 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11383 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
11384
11385 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
11386 int rc = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
11387 if (RT_UNLIKELY(rc == VINF_HM_DOUBLE_FAULT))
11388 return VINF_SUCCESS;
11389 else if (RT_UNLIKELY(rc == VINF_EM_RESET))
11390 return rc;
11391
11392 RTGCPHYS GCPhys = 0;
11393 rc = VMXReadVmcs64(VMX_VMCS64_EXIT_GUEST_PHYS_ADDR_FULL, &GCPhys);
11394 rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11395#if 0
11396 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); /** @todo Can we do better? */
11397#else
11398 /* Aggressive state sync. for now. */
11399 rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
11400 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
11401 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11402#endif
11403 AssertRCReturn(rc, rc);
11404
11405 /* Intel spec. Table 27-7 "Exit Qualifications for EPT violations". */
11406 AssertMsg(((pVmxTransient->uExitQualification >> 7) & 3) != 2, ("%#RX64", pVmxTransient->uExitQualification));
11407
11408 RTGCUINT uErrorCode = 0;
11409 if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_INSTR_FETCH)
11410 uErrorCode |= X86_TRAP_PF_ID;
11411 if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_DATA_WRITE)
11412 uErrorCode |= X86_TRAP_PF_RW;
11413 if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_ENTRY_PRESENT)
11414 uErrorCode |= X86_TRAP_PF_P;
11415
11416 TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode);
11417
11418 Log4(("EPT violation %#x at %#RX64 ErrorCode %#x CS:EIP=%04x:%#RX64\n", pVmxTransient->uExitQualification, GCPhys,
11419 uErrorCode, pMixedCtx->cs.Sel, pMixedCtx->rip));
11420
11421 /* Handle the pagefault trap for the nested shadow table. */
11422 PVM pVM = pVCpu->CTX_SUFF(pVM);
11423 rc = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pMixedCtx), GCPhys);
11424 TRPMResetTrap(pVCpu);
11425
11426 /* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */
11427 if ( rc == VINF_SUCCESS
11428 || rc == VERR_PAGE_TABLE_NOT_PRESENT
11429 || rc == VERR_PAGE_NOT_PRESENT)
11430 {
11431 /* Successfully synced our nested page tables. */
11432 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf);
11433 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
11434 | HM_CHANGED_GUEST_RSP
11435 | HM_CHANGED_GUEST_RFLAGS);
11436 return VINF_SUCCESS;
11437 }
11438
11439 Log4(("EPT return to ring-3 rc=%Rrc\n", rc));
11440 return rc;
11441}
11442
11443/** @} */
11444
11445/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-= */
11446/* -=-=-=-=-=-=-=-=-=- VM-exit Exception Handlers -=-=-=-=-=-=-=-=-=-=- */
11447/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-= */
11448
11449/** @name VM-exit exception handlers.
11450 * @{
11451 */
11452
11453/**
11454 * VM-exit exception handler for #MF (Math Fault: floating point exception).
11455 */
11456static int hmR0VmxExitXcptMF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11457{
11458 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11459 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
11460
11461 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11462 AssertRCReturn(rc, rc);
11463
11464 if (!(pMixedCtx->cr0 & X86_CR0_NE))
11465 {
11466 /* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
11467 rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
11468
11469 /** @todo r=ramshankar: The Intel spec. does -not- specify that this VM-exit
11470 * provides VM-exit instruction length. If this causes problem later,
11471 * disassemble the instruction like it's done on AMD-V. */
11472 int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11473 AssertRCReturn(rc2, rc2);
11474 return rc;
11475 }
11476
11477 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11478 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
11479 return rc;
11480}
11481
11482
11483/**
11484 * VM-exit exception handler for #BP (Breakpoint exception).
11485 */
11486static int hmR0VmxExitXcptBP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11487{
11488 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11489 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
11490
11491 /** @todo Try optimize this by not saving the entire guest state unless
11492 * really needed. */
11493 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11494 AssertRCReturn(rc, rc);
11495
11496 PVM pVM = pVCpu->CTX_SUFF(pVM);
11497 rc = DBGFRZTrap03Handler(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11498 if (rc == VINF_EM_RAW_GUEST_TRAP)
11499 {
11500 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11501 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11502 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11503 AssertRCReturn(rc, rc);
11504
11505 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11506 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
11507 }
11508
11509 Assert(rc == VINF_SUCCESS || rc == VINF_EM_RAW_GUEST_TRAP || rc == VINF_EM_DBG_BREAKPOINT);
11510 return rc;
11511}
11512
11513
11514/**
11515 * VM-exit exception handler for #DB (Debug exception).
11516 */
11517static int hmR0VmxExitXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11518{
11519 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11520 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
11521 Log6(("XcptDB\n"));
11522
11523 /*
11524 * Get the DR6-like values from the exit qualification and pass it to DBGF
11525 * for processing.
11526 */
11527 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11528 AssertRCReturn(rc, rc);
11529
11530 /* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */
11531 uint64_t uDR6 = X86_DR6_INIT_VAL;
11532 uDR6 |= ( pVmxTransient->uExitQualification
11533 & (X86_DR6_B0 | X86_DR6_B1 | X86_DR6_B2 | X86_DR6_B3 | X86_DR6_BD | X86_DR6_BS));
11534
11535 rc = DBGFRZTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pMixedCtx), uDR6, pVCpu->hm.s.fSingleInstruction);
11536 if (rc == VINF_EM_RAW_GUEST_TRAP)
11537 {
11538 /*
11539 * The exception was for the guest. Update DR6, DR7.GD and
11540 * IA32_DEBUGCTL.LBR before forwarding it.
11541 * (See Intel spec. 27.1 "Architectural State before a VM-Exit".)
11542 */
11543 VMMRZCallRing3Disable(pVCpu);
11544 HM_DISABLE_PREEMPT_IF_NEEDED();
11545
11546 pMixedCtx->dr[6] &= ~X86_DR6_B_MASK;
11547 pMixedCtx->dr[6] |= uDR6;
11548 if (CPUMIsGuestDebugStateActive(pVCpu))
11549 ASMSetDR6(pMixedCtx->dr[6]);
11550
11551 HM_RESTORE_PREEMPT_IF_NEEDED();
11552 VMMRZCallRing3Enable(pVCpu);
11553
11554 rc = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
11555 AssertRCReturn(rc, rc);
11556
11557 /* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
11558 pMixedCtx->dr[7] &= ~X86_DR7_GD;
11559
11560 /* Paranoia. */
11561 pMixedCtx->dr[7] &= ~X86_DR7_RAZ_MASK;
11562 pMixedCtx->dr[7] |= X86_DR7_RA1_MASK;
11563
11564 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)pMixedCtx->dr[7]);
11565 AssertRCReturn(rc, rc);
11566
11567 /*
11568 * Raise #DB in the guest.
11569 *
11570 * It is important to reflect what the VM-exit gave us (preserving the interruption-type) rather than use
11571 * hmR0VmxSetPendingXcptDB() as the #DB could've been raised while executing ICEBP and not the 'normal' #DB.
11572 * Thus it -may- trigger different handling in the CPU (like skipped DPL checks). See @bugref{6398}.
11573 *
11574 * Since ICEBP isn't documented on Intel, see AMD spec. 15.20 "Event Injection".
11575 */
11576 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11577 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11578 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11579 AssertRCReturn(rc, rc);
11580 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11581 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
11582 return VINF_SUCCESS;
11583 }
11584
11585 /*
11586 * Not a guest trap, must be a hypervisor related debug event then.
11587 * Update DR6 in case someone is interested in it.
11588 */
11589 AssertMsg(rc == VINF_EM_DBG_STEPPED || rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc));
11590 AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5);
11591 CPUMSetHyperDR6(pVCpu, uDR6);
11592
11593 return rc;
11594}
11595
11596
11597/**
11598 * VM-exit exception handler for #NM (Device-not-available exception: floating
11599 * point exception).
11600 */
11601static int hmR0VmxExitXcptNM(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11602{
11603 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11604
11605 /* We require CR0 and EFER. EFER is always up-to-date. */
11606 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11607 AssertRCReturn(rc, rc);
11608
11609 /* We're playing with the host CPU state here, have to disable preemption or longjmp. */
11610 VMMRZCallRing3Disable(pVCpu);
11611 HM_DISABLE_PREEMPT_IF_NEEDED();
11612
11613 /* If the guest FPU was active at the time of the #NM exit, then it's a guest fault. */
11614 if (pVmxTransient->fWasGuestFPUStateActive)
11615 {
11616 rc = VINF_EM_RAW_GUEST_TRAP;
11617 Assert(CPUMIsGuestFPUStateActive(pVCpu) || HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0));
11618 }
11619 else
11620 {
11621#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
11622 Assert(!pVmxTransient->fWasGuestFPUStateActive);
11623#endif
11624 rc = CPUMR0Trap07Handler(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx);
11625 Assert(rc == VINF_EM_RAW_GUEST_TRAP || (rc == VINF_SUCCESS && CPUMIsGuestFPUStateActive(pVCpu)));
11626 }
11627
11628 HM_RESTORE_PREEMPT_IF_NEEDED();
11629 VMMRZCallRing3Enable(pVCpu);
11630
11631 if (rc == VINF_SUCCESS)
11632 {
11633 /* Guest FPU state was activated, we'll want to change CR0 FPU intercepts before the next VM-reentry. */
11634 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
11635 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowNM);
11636 pVCpu->hm.s.fUseGuestFpu = true;
11637 }
11638 else
11639 {
11640 /* Forward #NM to the guest. */
11641 Assert(rc == VINF_EM_RAW_GUEST_TRAP);
11642 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11643 AssertRCReturn(rc, rc);
11644 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11645 pVmxTransient->cbInstr, 0 /* error code */, 0 /* GCPtrFaultAddress */);
11646 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNM);
11647 }
11648
11649 return VINF_SUCCESS;
11650}
11651
11652
11653/**
11654 * VM-exit exception handler for #GP (General-protection exception).
11655 *
11656 * @remarks Requires pVmxTransient->uExitIntInfo to be up-to-date.
11657 */
11658static int hmR0VmxExitXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11659{
11660 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11661 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
11662
11663 int rc = VERR_INTERNAL_ERROR_5;
11664 if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
11665 {
11666#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
11667 /* If the guest is not in real-mode or we have unrestricted execution support, reflect #GP to the guest. */
11668 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11669 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11670 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11671 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11672 AssertRCReturn(rc, rc);
11673 Log4(("#GP Gst: CS:RIP %04x:%#RX64 ErrorCode=%#x CR0=%#RX64 CPL=%u TR=%#04x\n", pMixedCtx->cs.Sel, pMixedCtx->rip,
11674 pVmxTransient->uExitIntErrorCode, pMixedCtx->cr0, CPUMGetGuestCPL(pVCpu), pMixedCtx->tr.Sel));
11675 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11676 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
11677 return rc;
11678#else
11679 /* We don't intercept #GP. */
11680 AssertMsgFailed(("Unexpected VM-exit caused by #GP exception\n"));
11681 NOREF(pVmxTransient);
11682 return VERR_VMX_UNEXPECTED_EXCEPTION;
11683#endif
11684 }
11685
11686 Assert(CPUMIsGuestInRealModeEx(pMixedCtx));
11687 Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest);
11688
11689 /* EMInterpretDisasCurrent() requires a lot of the state, save the entire state. */
11690 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11691 AssertRCReturn(rc, rc);
11692
11693 PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
11694 uint32_t cbOp = 0;
11695 PVM pVM = pVCpu->CTX_SUFF(pVM);
11696 rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
11697 if (RT_SUCCESS(rc))
11698 {
11699 rc = VINF_SUCCESS;
11700 Assert(cbOp == pDis->cbInstr);
11701 Log4(("#GP Disas OpCode=%u CS:EIP %04x:%#RX64\n", pDis->pCurInstr->uOpcode, pMixedCtx->cs.Sel, pMixedCtx->rip));
11702 switch (pDis->pCurInstr->uOpcode)
11703 {
11704 case OP_CLI:
11705 {
11706 pMixedCtx->eflags.Bits.u1IF = 0;
11707 pMixedCtx->rip += pDis->cbInstr;
11708 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS);
11709 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
11710 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCli);
11711 break;
11712 }
11713
11714 case OP_STI:
11715 {
11716 pMixedCtx->eflags.Bits.u1IF = 1;
11717 pMixedCtx->rip += pDis->cbInstr;
11718 EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip);
11719 Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS));
11720 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS);
11721 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
11722 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitSti);
11723 break;
11724 }
11725
11726 case OP_HLT:
11727 {
11728 rc = VINF_EM_HALT;
11729 pMixedCtx->rip += pDis->cbInstr;
11730 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
11731 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt);
11732 break;
11733 }
11734
11735 case OP_POPF:
11736 {
11737 Log4(("POPF CS:RIP %04x:%#RX64\n", pMixedCtx->cs.Sel, pMixedCtx->rip));
11738 uint32_t cbParm;
11739 uint32_t uMask;
11740 bool fStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
11741 if (pDis->fPrefix & DISPREFIX_OPSIZE)
11742 {
11743 cbParm = 4;
11744 uMask = 0xffffffff;
11745 }
11746 else
11747 {
11748 cbParm = 2;
11749 uMask = 0xffff;
11750 }
11751
11752 /* Get the stack pointer & pop the contents of the stack onto Eflags. */
11753 RTGCPTR GCPtrStack = 0;
11754 X86EFLAGS Eflags;
11755 rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), pMixedCtx->esp & uMask, SELMTOFLAT_FLAGS_CPL0,
11756 &GCPtrStack);
11757 if (RT_SUCCESS(rc))
11758 {
11759 Assert(sizeof(Eflags.u32) >= cbParm);
11760 Eflags.u32 = 0;
11761 rc = PGMPhysRead(pVM, (RTGCPHYS)GCPtrStack, &Eflags.u32, cbParm);
11762 }
11763 if (RT_FAILURE(rc))
11764 {
11765 rc = VERR_EM_INTERPRETER;
11766 break;
11767 }
11768 Log4(("POPF %#x -> %#RX64 mask=%#x RIP=%#RX64\n", Eflags.u, pMixedCtx->rsp, uMask, pMixedCtx->rip));
11769 pMixedCtx->eflags.u32 = (pMixedCtx->eflags.u32 & ~(X86_EFL_POPF_BITS & uMask))
11770 | (Eflags.u32 & X86_EFL_POPF_BITS & uMask);
11771 pMixedCtx->eflags.Bits.u1RF = 0; /* The RF bit is always cleared by POPF; see Intel Instruction reference. */
11772 pMixedCtx->esp += cbParm;
11773 pMixedCtx->esp &= uMask;
11774 pMixedCtx->rip += pDis->cbInstr;
11775 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
11776 | HM_CHANGED_GUEST_RSP
11777 | HM_CHANGED_GUEST_RFLAGS);
11778 /* Generate a pending-debug exception when stepping over POPF regardless of how POPF modifies EFLAGS.TF. */
11779 if (fStepping)
11780 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
11781
11782 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPopf);
11783 break;
11784 }
11785
11786 case OP_PUSHF:
11787 {
11788 uint32_t cbParm;
11789 uint32_t uMask;
11790 if (pDis->fPrefix & DISPREFIX_OPSIZE)
11791 {
11792 cbParm = 4;
11793 uMask = 0xffffffff;
11794 }
11795 else
11796 {
11797 cbParm = 2;
11798 uMask = 0xffff;
11799 }
11800
11801 /* Get the stack pointer & push the contents of eflags onto the stack. */
11802 RTGCPTR GCPtrStack = 0;
11803 rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), (pMixedCtx->esp - cbParm) & uMask,
11804 SELMTOFLAT_FLAGS_CPL0, &GCPtrStack);
11805 if (RT_FAILURE(rc))
11806 {
11807 rc = VERR_EM_INTERPRETER;
11808 break;
11809 }
11810 X86EFLAGS Eflags = pMixedCtx->eflags;
11811 /* The RF & VM bits are cleared on image stored on stack; see Intel Instruction reference for PUSHF. */
11812 Eflags.Bits.u1RF = 0;
11813 Eflags.Bits.u1VM = 0;
11814
11815 rc = PGMPhysWrite(pVM, (RTGCPHYS)GCPtrStack, &Eflags.u, cbParm);
11816 if (RT_FAILURE(rc))
11817 {
11818 rc = VERR_EM_INTERPRETER;
11819 break;
11820 }
11821 Log4(("PUSHF %#x -> %#RGv\n", Eflags.u, GCPtrStack));
11822 pMixedCtx->esp -= cbParm;
11823 pMixedCtx->esp &= uMask;
11824 pMixedCtx->rip += pDis->cbInstr;
11825 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RSP);
11826 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
11827 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPushf);
11828 break;
11829 }
11830
11831 case OP_IRET:
11832 {
11833 /** @todo Handle 32-bit operand sizes and check stack limits. See Intel
11834 * instruction reference. */
11835 RTGCPTR GCPtrStack = 0;
11836 uint32_t uMask = 0xffff;
11837 bool fStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
11838 uint16_t aIretFrame[3];
11839 if (pDis->fPrefix & (DISPREFIX_OPSIZE | DISPREFIX_ADDRSIZE))
11840 {
11841 rc = VERR_EM_INTERPRETER;
11842 break;
11843 }
11844 rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), pMixedCtx->esp & uMask, SELMTOFLAT_FLAGS_CPL0,
11845 &GCPtrStack);
11846 if (RT_SUCCESS(rc))
11847 rc = PGMPhysRead(pVM, (RTGCPHYS)GCPtrStack, &aIretFrame[0], sizeof(aIretFrame));
11848 if (RT_FAILURE(rc))
11849 {
11850 rc = VERR_EM_INTERPRETER;
11851 break;
11852 }
11853 pMixedCtx->eip = 0;
11854 pMixedCtx->ip = aIretFrame[0];
11855 pMixedCtx->cs.Sel = aIretFrame[1];
11856 pMixedCtx->cs.ValidSel = aIretFrame[1];
11857 pMixedCtx->cs.u64Base = (uint64_t)pMixedCtx->cs.Sel << 4;
11858 pMixedCtx->eflags.u32 = (pMixedCtx->eflags.u32 & ~(X86_EFL_POPF_BITS & uMask))
11859 | (aIretFrame[2] & X86_EFL_POPF_BITS & uMask);
11860 pMixedCtx->sp += sizeof(aIretFrame);
11861 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
11862 | HM_CHANGED_GUEST_SEGMENT_REGS
11863 | HM_CHANGED_GUEST_RSP
11864 | HM_CHANGED_GUEST_RFLAGS);
11865 /* Generate a pending-debug exception when stepping over IRET regardless of how IRET modifies EFLAGS.TF. */
11866 if (fStepping)
11867 hmR0VmxSetPendingDebugXcpt(pVCpu, pMixedCtx);
11868 Log4(("IRET %#RX32 to %04x:%x\n", GCPtrStack, pMixedCtx->cs.Sel, pMixedCtx->ip));
11869 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIret);
11870 break;
11871 }
11872
11873 case OP_INT:
11874 {
11875 uint16_t uVector = pDis->Param1.uValue & 0xff;
11876 hmR0VmxSetPendingIntN(pVCpu, pMixedCtx, uVector, pDis->cbInstr);
11877 /* INT clears EFLAGS.TF, we mustn't set any pending debug exceptions here. */
11878 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInt);
11879 break;
11880 }
11881
11882 case OP_INTO:
11883 {
11884 if (pMixedCtx->eflags.Bits.u1OF)
11885 {
11886 hmR0VmxSetPendingXcptOF(pVCpu, pMixedCtx, pDis->cbInstr);
11887 /* INTO clears EFLAGS.TF, we mustn't set any pending debug exceptions here. */
11888 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInt);
11889 }
11890 break;
11891 }
11892
11893 default:
11894 {
11895 VBOXSTRICTRC rc2 = EMInterpretInstructionDisasState(pVCpu, pDis, CPUMCTX2CORE(pMixedCtx), 0 /* pvFault */,
11896 EMCODETYPE_SUPERVISOR);
11897 rc = VBOXSTRICTRC_VAL(rc2);
11898 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
11899 /** @todo We have to set pending-debug exceptions here when the guest is
11900 * single-stepping depending on the instruction that was interpreted. */
11901 Log4(("#GP rc=%Rrc\n", rc));
11902 break;
11903 }
11904 }
11905 }
11906 else
11907 rc = VERR_EM_INTERPRETER;
11908
11909 AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER || rc == VINF_PGM_CHANGE_MODE || rc == VINF_EM_HALT,
11910 ("#GP Unexpected rc=%Rrc\n", rc));
11911 return rc;
11912}
11913
11914
11915#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
11916/**
11917 * VM-exit exception handler wrapper for generic exceptions. Simply re-injects
11918 * the exception reported in the VMX transient structure back into the VM.
11919 *
11920 * @remarks Requires uExitIntInfo in the VMX transient structure to be
11921 * up-to-date.
11922 */
11923static int hmR0VmxExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11924{
11925 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11926
11927 /* Re-inject the exception into the guest. This cannot be a double-fault condition which would have been handled in
11928 hmR0VmxCheckExitDueToEventDelivery(). */
11929 int rc = hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11930 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11931 AssertRCReturn(rc, rc);
11932 Assert(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO);
11933
11934#ifdef DEBUG_ramshankar
11935 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11936 uint8_t uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo);
11937 Log(("hmR0VmxExitXcptGeneric: Reinjecting Xcpt. uVector=%#x cs:rip=%#04x:%#RX64\n", uVector, pCtx->cs.Sel, pCtx->rip));
11938#endif
11939
11940 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11941 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
11942 return VINF_SUCCESS;
11943}
11944#endif
11945
11946
11947/**
11948 * VM-exit exception handler for #PF (Page-fault exception).
11949 */
11950static int hmR0VmxExitXcptPF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11951{
11952 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
11953 PVM pVM = pVCpu->CTX_SUFF(pVM);
11954 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11955 rc |= hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11956 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11957 AssertRCReturn(rc, rc);
11958
11959#if defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) || defined(HMVMX_ALWAYS_TRAP_PF)
11960 if (pVM->hm.s.fNestedPaging)
11961 {
11962 pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
11963 if (RT_LIKELY(!pVmxTransient->fVectoringPF))
11964 {
11965 pMixedCtx->cr2 = pVmxTransient->uExitQualification; /* Update here in case we go back to ring-3 before injection. */
11966 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
11967 0 /* cbInstr */, pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQualification);
11968 }
11969 else
11970 {
11971 /* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
11972 hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
11973 Log4(("Pending #DF due to vectoring #PF. NP\n"));
11974 }
11975 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
11976 return rc;
11977 }
11978#else
11979 Assert(!pVM->hm.s.fNestedPaging);
11980 NOREF(pVM);
11981#endif
11982
11983 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11984 AssertRCReturn(rc, rc);
11985
11986 Log4(("#PF: cr2=%#RX64 cs:rip=%#04x:%#RX64 uErrCode %#RX32 cr3=%#RX64\n", pVmxTransient->uExitQualification,
11987 pMixedCtx->cs.Sel, pMixedCtx->rip, pVmxTransient->uExitIntErrorCode, pMixedCtx->cr3));
11988
11989 TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQualification, (RTGCUINT)pVmxTransient->uExitIntErrorCode);
11990 rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pMixedCtx),
11991 (RTGCPTR)pVmxTransient->uExitQualification);
11992
11993 Log4(("#PF: rc=%Rrc\n", rc));
11994 if (rc == VINF_SUCCESS)
11995 {
11996 /* Successfully synced shadow pages tables or emulated an MMIO instruction. */
11997 /** @todo this isn't quite right, what if guest does lgdt with some MMIO
11998 * memory? We don't update the whole state here... */
11999 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
12000 | HM_CHANGED_GUEST_RSP
12001 | HM_CHANGED_GUEST_RFLAGS
12002 | HM_CHANGED_VMX_GUEST_APIC_STATE);
12003 TRPMResetTrap(pVCpu);
12004 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
12005 return rc;
12006 }
12007 else if (rc == VINF_EM_RAW_GUEST_TRAP)
12008 {
12009 if (!pVmxTransient->fVectoringPF)
12010 {
12011 /* It's a guest page fault and needs to be reflected to the guest. */
12012 uint32_t uGstErrorCode = TRPMGetErrorCode(pVCpu);
12013 TRPMResetTrap(pVCpu);
12014 pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */
12015 pMixedCtx->cr2 = pVmxTransient->uExitQualification; /* Update here in case we go back to ring-3 before injection. */
12016 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
12017 0 /* cbInstr */, uGstErrorCode, pVmxTransient->uExitQualification);
12018 }
12019 else
12020 {
12021 /* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
12022 TRPMResetTrap(pVCpu);
12023 pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */
12024 hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
12025 Log4(("#PF: Pending #DF due to vectoring #PF\n"));
12026 }
12027
12028 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
12029 return VINF_SUCCESS;
12030 }
12031
12032 TRPMResetTrap(pVCpu);
12033 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
12034 return rc;
12035}
12036
12037/** @} */
12038
Note: See TracBrowser for help on using the repository browser.

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