VirtualBox

source: vbox/trunk/src/VBox/VMM/VMMR0/HMVMXR0.cpp@ 64882

Last change on this file since 64882 was 64882, checked in by vboxsync, 8 years ago

VMM/HMVMXR0: Avoid setting HM_CHANGED_VMX_GUEST_APIC_STATE in TprBelowThreshold VM-exit as it should've already been set in hmR0VmxPostRunGuest.

  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 584.1 KB
Line 
1/* $Id: HMVMXR0.cpp 64882 2016-12-15 15:17:35Z vboxsync $ */
2/** @file
3 * HM VMX (Intel VT-x) - Host Context Ring-0.
4 */
5
6/*
7 * Copyright (C) 2012-2016 Oracle Corporation
8 *
9 * This file is part of VirtualBox Open Source Edition (OSE), as
10 * available from http://www.virtualbox.org. This file is free software;
11 * you can redistribute it and/or modify it under the terms of the GNU
12 * General Public License (GPL) as published by the Free Software
13 * Foundation, in version 2 as it comes in the "COPYING" file of the
14 * VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15 * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16 */
17
18
19/*********************************************************************************************************************************
20* Header Files *
21*********************************************************************************************************************************/
22#define LOG_GROUP LOG_GROUP_HM
23#include <iprt/x86.h>
24#include <iprt/asm-amd64-x86.h>
25#include <iprt/thread.h>
26
27#include <VBox/vmm/pdmapi.h>
28#include <VBox/vmm/dbgf.h>
29#include <VBox/vmm/iem.h>
30#include <VBox/vmm/iom.h>
31#include <VBox/vmm/selm.h>
32#include <VBox/vmm/tm.h>
33#include <VBox/vmm/gim.h>
34#include <VBox/vmm/apic.h>
35#ifdef VBOX_WITH_REM
36# include <VBox/vmm/rem.h>
37#endif
38#include "HMInternal.h"
39#include <VBox/vmm/vm.h>
40#include "HMVMXR0.h"
41#include "dtrace/VBoxVMM.h"
42
43#ifdef DEBUG_ramshankar
44# define HMVMX_ALWAYS_SAVE_GUEST_RFLAGS
45# define HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE
46# define HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
47# define HMVMX_ALWAYS_CHECK_GUEST_STATE
48# define HMVMX_ALWAYS_TRAP_ALL_XCPTS
49# define HMVMX_ALWAYS_TRAP_PF
50# define HMVMX_ALWAYS_SWAP_FPU_STATE
51# define HMVMX_ALWAYS_FLUSH_TLB
52# define HMVMX_ALWAYS_SWAP_EFER
53#endif
54
55
56/*********************************************************************************************************************************
57* Defined Constants And Macros *
58*********************************************************************************************************************************/
59/** Use the function table. */
60#define HMVMX_USE_FUNCTION_TABLE
61
62/** Determine which tagged-TLB flush handler to use. */
63#define HMVMX_FLUSH_TAGGED_TLB_EPT_VPID 0
64#define HMVMX_FLUSH_TAGGED_TLB_EPT 1
65#define HMVMX_FLUSH_TAGGED_TLB_VPID 2
66#define HMVMX_FLUSH_TAGGED_TLB_NONE 3
67
68/** @name Updated-guest-state flags.
69 * @{ */
70#define HMVMX_UPDATED_GUEST_RIP RT_BIT(0)
71#define HMVMX_UPDATED_GUEST_RSP RT_BIT(1)
72#define HMVMX_UPDATED_GUEST_RFLAGS RT_BIT(2)
73#define HMVMX_UPDATED_GUEST_CR0 RT_BIT(3)
74#define HMVMX_UPDATED_GUEST_CR3 RT_BIT(4)
75#define HMVMX_UPDATED_GUEST_CR4 RT_BIT(5)
76#define HMVMX_UPDATED_GUEST_GDTR RT_BIT(6)
77#define HMVMX_UPDATED_GUEST_IDTR RT_BIT(7)
78#define HMVMX_UPDATED_GUEST_LDTR RT_BIT(8)
79#define HMVMX_UPDATED_GUEST_TR RT_BIT(9)
80#define HMVMX_UPDATED_GUEST_SEGMENT_REGS RT_BIT(10)
81#define HMVMX_UPDATED_GUEST_DEBUG RT_BIT(11)
82#define HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR RT_BIT(12)
83#define HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR RT_BIT(13)
84#define HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR RT_BIT(14)
85#define HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS RT_BIT(15)
86#define HMVMX_UPDATED_GUEST_LAZY_MSRS RT_BIT(16)
87#define HMVMX_UPDATED_GUEST_ACTIVITY_STATE RT_BIT(17)
88#define HMVMX_UPDATED_GUEST_INTR_STATE RT_BIT(18)
89#define HMVMX_UPDATED_GUEST_APIC_STATE RT_BIT(19)
90#define HMVMX_UPDATED_GUEST_ALL ( HMVMX_UPDATED_GUEST_RIP \
91 | HMVMX_UPDATED_GUEST_RSP \
92 | HMVMX_UPDATED_GUEST_RFLAGS \
93 | HMVMX_UPDATED_GUEST_CR0 \
94 | HMVMX_UPDATED_GUEST_CR3 \
95 | HMVMX_UPDATED_GUEST_CR4 \
96 | HMVMX_UPDATED_GUEST_GDTR \
97 | HMVMX_UPDATED_GUEST_IDTR \
98 | HMVMX_UPDATED_GUEST_LDTR \
99 | HMVMX_UPDATED_GUEST_TR \
100 | HMVMX_UPDATED_GUEST_SEGMENT_REGS \
101 | HMVMX_UPDATED_GUEST_DEBUG \
102 | HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR \
103 | HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR \
104 | HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR \
105 | HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS \
106 | HMVMX_UPDATED_GUEST_LAZY_MSRS \
107 | HMVMX_UPDATED_GUEST_ACTIVITY_STATE \
108 | HMVMX_UPDATED_GUEST_INTR_STATE \
109 | HMVMX_UPDATED_GUEST_APIC_STATE)
110/** @} */
111
112/** @name
113 * Flags to skip redundant reads of some common VMCS fields that are not part of
114 * the guest-CPU state but are in the transient structure.
115 */
116#define HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO RT_BIT(0)
117#define HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE RT_BIT(1)
118#define HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION RT_BIT(2)
119#define HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN RT_BIT(3)
120#define HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO RT_BIT(4)
121#define HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE RT_BIT(5)
122#define HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO RT_BIT(6)
123/** @} */
124
125/** @name
126 * States of the VMCS.
127 *
128 * This does not reflect all possible VMCS states but currently only those
129 * needed for maintaining the VMCS consistently even when thread-context hooks
130 * are used. Maybe later this can be extended (i.e. Nested Virtualization).
131 */
132#define HMVMX_VMCS_STATE_CLEAR RT_BIT(0)
133#define HMVMX_VMCS_STATE_ACTIVE RT_BIT(1)
134#define HMVMX_VMCS_STATE_LAUNCHED RT_BIT(2)
135/** @} */
136
137/**
138 * Exception bitmap mask for real-mode guests (real-on-v86).
139 *
140 * We need to intercept all exceptions manually except:
141 * - \#NM, \#MF handled in hmR0VmxLoadSharedCR0().
142 * - \#AC and \#DB are always intercepted to prevent the CPU from deadlocking
143 * due to bugs in Intel CPUs.
144 * - \#PF need not be intercepted even in real-mode if we have Nested Paging
145 * support.
146 */
147#define HMVMX_REAL_MODE_XCPT_MASK ( RT_BIT(X86_XCPT_DE) /* always: | RT_BIT(X86_XCPT_DB) */ | RT_BIT(X86_XCPT_NMI) \
148 | RT_BIT(X86_XCPT_BP) | RT_BIT(X86_XCPT_OF) | RT_BIT(X86_XCPT_BR) \
149 | RT_BIT(X86_XCPT_UD) /* RT_BIT(X86_XCPT_NM) */ | RT_BIT(X86_XCPT_DF) \
150 | RT_BIT(X86_XCPT_CO_SEG_OVERRUN) | RT_BIT(X86_XCPT_TS) | RT_BIT(X86_XCPT_NP) \
151 | RT_BIT(X86_XCPT_SS) | RT_BIT(X86_XCPT_GP) /* RT_BIT(X86_XCPT_PF) */ \
152 /* RT_BIT(X86_XCPT_MF) always: | RT_BIT(X86_XCPT_AC) */ | RT_BIT(X86_XCPT_MC) \
153 | RT_BIT(X86_XCPT_XF))
154
155/**
156 * Exception bitmap mask for all contributory exceptions.
157 *
158 * Page fault is deliberately excluded here as it's conditional as to whether
159 * it's contributory or benign. Page faults are handled separately.
160 */
161#define HMVMX_CONTRIBUTORY_XCPT_MASK ( RT_BIT(X86_XCPT_GP) | RT_BIT(X86_XCPT_NP) | RT_BIT(X86_XCPT_SS) | RT_BIT(X86_XCPT_TS) \
162 | RT_BIT(X86_XCPT_DE))
163
164/** Maximum VM-instruction error number. */
165#define HMVMX_INSTR_ERROR_MAX 28
166
167/** Profiling macro. */
168#ifdef HM_PROFILE_EXIT_DISPATCH
169# define HMVMX_START_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitDispatch, ed)
170# define HMVMX_STOP_EXIT_DISPATCH_PROF() STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitDispatch, ed)
171#else
172# define HMVMX_START_EXIT_DISPATCH_PROF() do { } while (0)
173# define HMVMX_STOP_EXIT_DISPATCH_PROF() do { } while (0)
174#endif
175
176/** Assert that preemption is disabled or covered by thread-context hooks. */
177#define HMVMX_ASSERT_PREEMPT_SAFE() Assert( VMMR0ThreadCtxHookIsEnabled(pVCpu) \
178 || !RTThreadPreemptIsEnabled(NIL_RTTHREAD));
179
180/** Assert that we haven't migrated CPUs when thread-context hooks are not
181 * used. */
182#define HMVMX_ASSERT_CPU_SAFE() AssertMsg( VMMR0ThreadCtxHookIsEnabled(pVCpu) \
183 || pVCpu->hm.s.idEnteredCpu == RTMpCpuId(), \
184 ("Illegal migration! Entered on CPU %u Current %u\n", \
185 pVCpu->hm.s.idEnteredCpu, RTMpCpuId())); \
186
187/** Helper macro for VM-exit handlers called unexpectedly. */
188#define HMVMX_RETURN_UNEXPECTED_EXIT() \
189 do { \
190 pVCpu->hm.s.u32HMError = pVmxTransient->uExitReason; \
191 return VERR_VMX_UNEXPECTED_EXIT; \
192 } while (0)
193
194
195/*********************************************************************************************************************************
196* Structures and Typedefs *
197*********************************************************************************************************************************/
198/**
199 * VMX transient state.
200 *
201 * A state structure for holding miscellaneous information across
202 * VMX non-root operation and restored after the transition.
203 */
204typedef struct VMXTRANSIENT
205{
206 /** The host's rflags/eflags. */
207 RTCCUINTREG fEFlags;
208#if HC_ARCH_BITS == 32
209 uint32_t u32Alignment0;
210#endif
211 /** The guest's TPR value used for TPR shadowing. */
212 uint8_t u8GuestTpr;
213 /** Alignment. */
214 uint8_t abAlignment0[7];
215
216 /** The basic VM-exit reason. */
217 uint16_t uExitReason;
218 /** Alignment. */
219 uint16_t u16Alignment0;
220 /** The VM-exit interruption error code. */
221 uint32_t uExitIntErrorCode;
222 /** The VM-exit exit code qualification. */
223 uint64_t uExitQualification;
224
225 /** The VM-exit interruption-information field. */
226 uint32_t uExitIntInfo;
227 /** The VM-exit instruction-length field. */
228 uint32_t cbInstr;
229 /** The VM-exit instruction-information field. */
230 union
231 {
232 /** Plain unsigned int representation. */
233 uint32_t u;
234 /** INS and OUTS information. */
235 struct
236 {
237 uint32_t u7Reserved0 : 7;
238 /** The address size; 0=16-bit, 1=32-bit, 2=64-bit, rest undefined. */
239 uint32_t u3AddrSize : 3;
240 uint32_t u5Reserved1 : 5;
241 /** The segment register (X86_SREG_XXX). */
242 uint32_t iSegReg : 3;
243 uint32_t uReserved2 : 14;
244 } StrIo;
245 } ExitInstrInfo;
246 /** Whether the VM-entry failed or not. */
247 bool fVMEntryFailed;
248 /** Alignment. */
249 uint8_t abAlignment1[3];
250
251 /** The VM-entry interruption-information field. */
252 uint32_t uEntryIntInfo;
253 /** The VM-entry exception error code field. */
254 uint32_t uEntryXcptErrorCode;
255 /** The VM-entry instruction length field. */
256 uint32_t cbEntryInstr;
257
258 /** IDT-vectoring information field. */
259 uint32_t uIdtVectoringInfo;
260 /** IDT-vectoring error code. */
261 uint32_t uIdtVectoringErrorCode;
262
263 /** Mask of currently read VMCS fields; HMVMX_UPDATED_TRANSIENT_*. */
264 uint32_t fVmcsFieldsRead;
265
266 /** Whether the guest FPU was active at the time of VM-exit. */
267 bool fWasGuestFPUStateActive;
268 /** Whether the guest debug state was active at the time of VM-exit. */
269 bool fWasGuestDebugStateActive;
270 /** Whether the hyper debug state was active at the time of VM-exit. */
271 bool fWasHyperDebugStateActive;
272 /** Whether TSC-offsetting should be setup before VM-entry. */
273 bool fUpdateTscOffsettingAndPreemptTimer;
274 /** Whether the VM-exit was caused by a page-fault during delivery of a
275 * contributory exception or a page-fault. */
276 bool fVectoringDoublePF;
277 /** Whether the VM-exit was caused by a page-fault during delivery of an
278 * external interrupt or NMI. */
279 bool fVectoringPF;
280} VMXTRANSIENT;
281AssertCompileMemberAlignment(VMXTRANSIENT, uExitReason, sizeof(uint64_t));
282AssertCompileMemberAlignment(VMXTRANSIENT, uExitIntInfo, sizeof(uint64_t));
283AssertCompileMemberAlignment(VMXTRANSIENT, uEntryIntInfo, sizeof(uint64_t));
284AssertCompileMemberAlignment(VMXTRANSIENT, fWasGuestFPUStateActive, sizeof(uint64_t));
285AssertCompileMemberSize(VMXTRANSIENT, ExitInstrInfo, sizeof(uint32_t));
286/** Pointer to VMX transient state. */
287typedef VMXTRANSIENT *PVMXTRANSIENT;
288
289
290/**
291 * MSR-bitmap read permissions.
292 */
293typedef enum VMXMSREXITREAD
294{
295 /** Reading this MSR causes a VM-exit. */
296 VMXMSREXIT_INTERCEPT_READ = 0xb,
297 /** Reading this MSR does not cause a VM-exit. */
298 VMXMSREXIT_PASSTHRU_READ
299} VMXMSREXITREAD;
300/** Pointer to MSR-bitmap read permissions. */
301typedef VMXMSREXITREAD* PVMXMSREXITREAD;
302
303/**
304 * MSR-bitmap write permissions.
305 */
306typedef enum VMXMSREXITWRITE
307{
308 /** Writing to this MSR causes a VM-exit. */
309 VMXMSREXIT_INTERCEPT_WRITE = 0xd,
310 /** Writing to this MSR does not cause a VM-exit. */
311 VMXMSREXIT_PASSTHRU_WRITE
312} VMXMSREXITWRITE;
313/** Pointer to MSR-bitmap write permissions. */
314typedef VMXMSREXITWRITE* PVMXMSREXITWRITE;
315
316
317/**
318 * VMX VM-exit handler.
319 *
320 * @returns Strict VBox status code (i.e. informational status codes too).
321 * @param pVCpu The cross context virtual CPU structure.
322 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
323 * out-of-sync. Make sure to update the required
324 * fields before using them.
325 * @param pVmxTransient Pointer to the VMX-transient structure.
326 */
327#ifndef HMVMX_USE_FUNCTION_TABLE
328typedef VBOXSTRICTRC FNVMXEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
329#else
330typedef DECLCALLBACK(VBOXSTRICTRC) FNVMXEXITHANDLER(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
331/** Pointer to VM-exit handler. */
332typedef FNVMXEXITHANDLER *PFNVMXEXITHANDLER;
333#endif
334
335/**
336 * VMX VM-exit handler, non-strict status code.
337 *
338 * This is generally the same as FNVMXEXITHANDLER, the NSRC bit is just FYI.
339 *
340 * @returns VBox status code, no informational status code returned.
341 * @param pVCpu The cross context virtual CPU structure.
342 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
343 * out-of-sync. Make sure to update the required
344 * fields before using them.
345 * @param pVmxTransient Pointer to the VMX-transient structure.
346 *
347 * @remarks This is not used on anything returning VERR_EM_INTERPRETER as the
348 * use of that status code will be replaced with VINF_EM_SOMETHING
349 * later when switching over to IEM.
350 */
351#ifndef HMVMX_USE_FUNCTION_TABLE
352typedef int FNVMXEXITHANDLERNSRC(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
353#else
354typedef FNVMXEXITHANDLER FNVMXEXITHANDLERNSRC;
355#endif
356
357
358/*********************************************************************************************************************************
359* Internal Functions *
360*********************************************************************************************************************************/
361static void hmR0VmxFlushEpt(PVMCPU pVCpu, VMXFLUSHEPT enmFlush);
362static void hmR0VmxFlushVpid(PVM pVM, PVMCPU pVCpu, VMXFLUSHVPID enmFlush, RTGCPTR GCPtr);
363static void hmR0VmxClearIntNmiWindowsVmcs(PVMCPU pVCpu);
364static VBOXSTRICTRC hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint64_t u64IntInfo, uint32_t cbInstr,
365 uint32_t u32ErrCode, RTGCUINTREG GCPtrFaultAddress,
366 bool fStepping, uint32_t *puIntState);
367#if HC_ARCH_BITS == 32
368static int hmR0VmxInitVmcsReadCache(PVM pVM, PVMCPU pVCpu);
369#endif
370#ifndef HMVMX_USE_FUNCTION_TABLE
371DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, uint32_t rcReason);
372# define HMVMX_EXIT_DECL DECLINLINE(VBOXSTRICTRC)
373# define HMVMX_EXIT_NSRC_DECL DECLINLINE(int)
374#else
375# define HMVMX_EXIT_DECL static DECLCALLBACK(VBOXSTRICTRC)
376# define HMVMX_EXIT_NSRC_DECL HMVMX_EXIT_DECL
377#endif
378
379
380/** @name VM-exit handlers.
381 * @{
382 */
383static FNVMXEXITHANDLER hmR0VmxExitXcptOrNmi;
384static FNVMXEXITHANDLER hmR0VmxExitExtInt;
385static FNVMXEXITHANDLER hmR0VmxExitTripleFault;
386static FNVMXEXITHANDLERNSRC hmR0VmxExitInitSignal;
387static FNVMXEXITHANDLERNSRC hmR0VmxExitSipi;
388static FNVMXEXITHANDLERNSRC hmR0VmxExitIoSmi;
389static FNVMXEXITHANDLERNSRC hmR0VmxExitSmi;
390static FNVMXEXITHANDLERNSRC hmR0VmxExitIntWindow;
391static FNVMXEXITHANDLERNSRC hmR0VmxExitNmiWindow;
392static FNVMXEXITHANDLER hmR0VmxExitTaskSwitch;
393static FNVMXEXITHANDLER hmR0VmxExitCpuid;
394static FNVMXEXITHANDLER hmR0VmxExitGetsec;
395static FNVMXEXITHANDLER hmR0VmxExitHlt;
396static FNVMXEXITHANDLERNSRC hmR0VmxExitInvd;
397static FNVMXEXITHANDLER hmR0VmxExitInvlpg;
398static FNVMXEXITHANDLER hmR0VmxExitRdpmc;
399static FNVMXEXITHANDLER hmR0VmxExitVmcall;
400static FNVMXEXITHANDLER hmR0VmxExitRdtsc;
401static FNVMXEXITHANDLERNSRC hmR0VmxExitRsm;
402static FNVMXEXITHANDLERNSRC hmR0VmxExitSetPendingXcptUD;
403static FNVMXEXITHANDLER hmR0VmxExitMovCRx;
404static FNVMXEXITHANDLER hmR0VmxExitMovDRx;
405static FNVMXEXITHANDLER hmR0VmxExitIoInstr;
406static FNVMXEXITHANDLER hmR0VmxExitRdmsr;
407static FNVMXEXITHANDLER hmR0VmxExitWrmsr;
408static FNVMXEXITHANDLERNSRC hmR0VmxExitErrInvalidGuestState;
409static FNVMXEXITHANDLERNSRC hmR0VmxExitErrMsrLoad;
410static FNVMXEXITHANDLERNSRC hmR0VmxExitErrUndefined;
411static FNVMXEXITHANDLER hmR0VmxExitMwait;
412static FNVMXEXITHANDLER hmR0VmxExitMtf;
413static FNVMXEXITHANDLER hmR0VmxExitMonitor;
414static FNVMXEXITHANDLER hmR0VmxExitPause;
415static FNVMXEXITHANDLERNSRC hmR0VmxExitErrMachineCheck;
416static FNVMXEXITHANDLERNSRC hmR0VmxExitTprBelowThreshold;
417static FNVMXEXITHANDLER hmR0VmxExitApicAccess;
418static FNVMXEXITHANDLER hmR0VmxExitXdtrAccess;
419static FNVMXEXITHANDLER hmR0VmxExitXdtrAccess;
420static FNVMXEXITHANDLER hmR0VmxExitEptViolation;
421static FNVMXEXITHANDLER hmR0VmxExitEptMisconfig;
422static FNVMXEXITHANDLER hmR0VmxExitRdtscp;
423static FNVMXEXITHANDLER hmR0VmxExitPreemptTimer;
424static FNVMXEXITHANDLERNSRC hmR0VmxExitWbinvd;
425static FNVMXEXITHANDLER hmR0VmxExitXsetbv;
426static FNVMXEXITHANDLER hmR0VmxExitRdrand;
427static FNVMXEXITHANDLER hmR0VmxExitInvpcid;
428/** @} */
429
430static int hmR0VmxExitXcptNM(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
431static int hmR0VmxExitXcptPF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
432static int hmR0VmxExitXcptMF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
433static int hmR0VmxExitXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
434static int hmR0VmxExitXcptBP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
435static int hmR0VmxExitXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
436static int hmR0VmxExitXcptAC(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
437static int hmR0VmxExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient);
438static uint32_t hmR0VmxCheckGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx);
439
440
441/*********************************************************************************************************************************
442* Global Variables *
443*********************************************************************************************************************************/
444#ifdef HMVMX_USE_FUNCTION_TABLE
445
446/**
447 * VMX_EXIT dispatch table.
448 */
449static const PFNVMXEXITHANDLER g_apfnVMExitHandlers[VMX_EXIT_MAX + 1] =
450{
451 /* 00 VMX_EXIT_XCPT_OR_NMI */ hmR0VmxExitXcptOrNmi,
452 /* 01 VMX_EXIT_EXT_INT */ hmR0VmxExitExtInt,
453 /* 02 VMX_EXIT_TRIPLE_FAULT */ hmR0VmxExitTripleFault,
454 /* 03 VMX_EXIT_INIT_SIGNAL */ hmR0VmxExitInitSignal,
455 /* 04 VMX_EXIT_SIPI */ hmR0VmxExitSipi,
456 /* 05 VMX_EXIT_IO_SMI */ hmR0VmxExitIoSmi,
457 /* 06 VMX_EXIT_SMI */ hmR0VmxExitSmi,
458 /* 07 VMX_EXIT_INT_WINDOW */ hmR0VmxExitIntWindow,
459 /* 08 VMX_EXIT_NMI_WINDOW */ hmR0VmxExitNmiWindow,
460 /* 09 VMX_EXIT_TASK_SWITCH */ hmR0VmxExitTaskSwitch,
461 /* 10 VMX_EXIT_CPUID */ hmR0VmxExitCpuid,
462 /* 11 VMX_EXIT_GETSEC */ hmR0VmxExitGetsec,
463 /* 12 VMX_EXIT_HLT */ hmR0VmxExitHlt,
464 /* 13 VMX_EXIT_INVD */ hmR0VmxExitInvd,
465 /* 14 VMX_EXIT_INVLPG */ hmR0VmxExitInvlpg,
466 /* 15 VMX_EXIT_RDPMC */ hmR0VmxExitRdpmc,
467 /* 16 VMX_EXIT_RDTSC */ hmR0VmxExitRdtsc,
468 /* 17 VMX_EXIT_RSM */ hmR0VmxExitRsm,
469 /* 18 VMX_EXIT_VMCALL */ hmR0VmxExitVmcall,
470 /* 19 VMX_EXIT_VMCLEAR */ hmR0VmxExitSetPendingXcptUD,
471 /* 20 VMX_EXIT_VMLAUNCH */ hmR0VmxExitSetPendingXcptUD,
472 /* 21 VMX_EXIT_VMPTRLD */ hmR0VmxExitSetPendingXcptUD,
473 /* 22 VMX_EXIT_VMPTRST */ hmR0VmxExitSetPendingXcptUD,
474 /* 23 VMX_EXIT_VMREAD */ hmR0VmxExitSetPendingXcptUD,
475 /* 24 VMX_EXIT_VMRESUME */ hmR0VmxExitSetPendingXcptUD,
476 /* 25 VMX_EXIT_VMWRITE */ hmR0VmxExitSetPendingXcptUD,
477 /* 26 VMX_EXIT_VMXOFF */ hmR0VmxExitSetPendingXcptUD,
478 /* 27 VMX_EXIT_VMXON */ hmR0VmxExitSetPendingXcptUD,
479 /* 28 VMX_EXIT_MOV_CRX */ hmR0VmxExitMovCRx,
480 /* 29 VMX_EXIT_MOV_DRX */ hmR0VmxExitMovDRx,
481 /* 30 VMX_EXIT_IO_INSTR */ hmR0VmxExitIoInstr,
482 /* 31 VMX_EXIT_RDMSR */ hmR0VmxExitRdmsr,
483 /* 32 VMX_EXIT_WRMSR */ hmR0VmxExitWrmsr,
484 /* 33 VMX_EXIT_ERR_INVALID_GUEST_STATE */ hmR0VmxExitErrInvalidGuestState,
485 /* 34 VMX_EXIT_ERR_MSR_LOAD */ hmR0VmxExitErrMsrLoad,
486 /* 35 UNDEFINED */ hmR0VmxExitErrUndefined,
487 /* 36 VMX_EXIT_MWAIT */ hmR0VmxExitMwait,
488 /* 37 VMX_EXIT_MTF */ hmR0VmxExitMtf,
489 /* 38 UNDEFINED */ hmR0VmxExitErrUndefined,
490 /* 39 VMX_EXIT_MONITOR */ hmR0VmxExitMonitor,
491 /* 40 UNDEFINED */ hmR0VmxExitPause,
492 /* 41 VMX_EXIT_PAUSE */ hmR0VmxExitErrMachineCheck,
493 /* 42 VMX_EXIT_ERR_MACHINE_CHECK */ hmR0VmxExitErrUndefined,
494 /* 43 VMX_EXIT_TPR_BELOW_THRESHOLD */ hmR0VmxExitTprBelowThreshold,
495 /* 44 VMX_EXIT_APIC_ACCESS */ hmR0VmxExitApicAccess,
496 /* 45 UNDEFINED */ hmR0VmxExitErrUndefined,
497 /* 46 VMX_EXIT_XDTR_ACCESS */ hmR0VmxExitXdtrAccess,
498 /* 47 VMX_EXIT_TR_ACCESS */ hmR0VmxExitXdtrAccess,
499 /* 48 VMX_EXIT_EPT_VIOLATION */ hmR0VmxExitEptViolation,
500 /* 49 VMX_EXIT_EPT_MISCONFIG */ hmR0VmxExitEptMisconfig,
501 /* 50 VMX_EXIT_INVEPT */ hmR0VmxExitSetPendingXcptUD,
502 /* 51 VMX_EXIT_RDTSCP */ hmR0VmxExitRdtscp,
503 /* 52 VMX_EXIT_PREEMPT_TIMER */ hmR0VmxExitPreemptTimer,
504 /* 53 VMX_EXIT_INVVPID */ hmR0VmxExitSetPendingXcptUD,
505 /* 54 VMX_EXIT_WBINVD */ hmR0VmxExitWbinvd,
506 /* 55 VMX_EXIT_XSETBV */ hmR0VmxExitXsetbv,
507 /* 56 VMX_EXIT_APIC_WRITE */ hmR0VmxExitErrUndefined,
508 /* 57 VMX_EXIT_RDRAND */ hmR0VmxExitRdrand,
509 /* 58 VMX_EXIT_INVPCID */ hmR0VmxExitInvpcid,
510 /* 59 VMX_EXIT_VMFUNC */ hmR0VmxExitSetPendingXcptUD,
511 /* 60 VMX_EXIT_ENCLS */ hmR0VmxExitErrUndefined,
512 /* 61 VMX_EXIT_RDSEED */ hmR0VmxExitErrUndefined, /* only spurious exits, so undefined */
513 /* 62 VMX_EXIT_PML_FULL */ hmR0VmxExitErrUndefined,
514 /* 63 VMX_EXIT_XSAVES */ hmR0VmxExitSetPendingXcptUD,
515 /* 64 VMX_EXIT_XRSTORS */ hmR0VmxExitSetPendingXcptUD,
516};
517#endif /* HMVMX_USE_FUNCTION_TABLE */
518
519#ifdef VBOX_STRICT
520static const char * const g_apszVmxInstrErrors[HMVMX_INSTR_ERROR_MAX + 1] =
521{
522 /* 0 */ "(Not Used)",
523 /* 1 */ "VMCALL executed in VMX root operation.",
524 /* 2 */ "VMCLEAR with invalid physical address.",
525 /* 3 */ "VMCLEAR with VMXON pointer.",
526 /* 4 */ "VMLAUNCH with non-clear VMCS.",
527 /* 5 */ "VMRESUME with non-launched VMCS.",
528 /* 6 */ "VMRESUME after VMXOFF",
529 /* 7 */ "VM-entry with invalid control fields.",
530 /* 8 */ "VM-entry with invalid host state fields.",
531 /* 9 */ "VMPTRLD with invalid physical address.",
532 /* 10 */ "VMPTRLD with VMXON pointer.",
533 /* 11 */ "VMPTRLD with incorrect revision identifier.",
534 /* 12 */ "VMREAD/VMWRITE from/to unsupported VMCS component.",
535 /* 13 */ "VMWRITE to read-only VMCS component.",
536 /* 14 */ "(Not Used)",
537 /* 15 */ "VMXON executed in VMX root operation.",
538 /* 16 */ "VM-entry with invalid executive-VMCS pointer.",
539 /* 17 */ "VM-entry with non-launched executing VMCS.",
540 /* 18 */ "VM-entry with executive-VMCS pointer not VMXON pointer.",
541 /* 19 */ "VMCALL with non-clear VMCS.",
542 /* 20 */ "VMCALL with invalid VM-exit control fields.",
543 /* 21 */ "(Not Used)",
544 /* 22 */ "VMCALL with incorrect MSEG revision identifier.",
545 /* 23 */ "VMXOFF under dual monitor treatment of SMIs and SMM.",
546 /* 24 */ "VMCALL with invalid SMM-monitor features.",
547 /* 25 */ "VM-entry with invalid VM-execution control fields in executive VMCS.",
548 /* 26 */ "VM-entry with events blocked by MOV SS.",
549 /* 27 */ "(Not Used)",
550 /* 28 */ "Invalid operand to INVEPT/INVVPID."
551};
552#endif /* VBOX_STRICT */
553
554
555
556/**
557 * Updates the VM's last error record.
558 *
559 * If there was a VMX instruction error, reads the error data from the VMCS and
560 * updates VCPU's last error record as well.
561 *
562 * @param pVM The cross context VM structure.
563 * @param pVCpu The cross context virtual CPU structure of the calling EMT.
564 * Can be NULL if @a rc is not VERR_VMX_UNABLE_TO_START_VM or
565 * VERR_VMX_INVALID_VMCS_FIELD.
566 * @param rc The error code.
567 */
568static void hmR0VmxUpdateErrorRecord(PVM pVM, PVMCPU pVCpu, int rc)
569{
570 AssertPtr(pVM);
571 if ( rc == VERR_VMX_INVALID_VMCS_FIELD
572 || rc == VERR_VMX_UNABLE_TO_START_VM)
573 {
574 AssertPtrReturnVoid(pVCpu);
575 VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
576 }
577 pVM->hm.s.lLastError = rc;
578}
579
580
581/**
582 * Reads the VM-entry interruption-information field from the VMCS into the VMX
583 * transient structure.
584 *
585 * @returns VBox status code.
586 * @param pVmxTransient Pointer to the VMX transient structure.
587 *
588 * @remarks No-long-jump zone!!!
589 */
590DECLINLINE(int) hmR0VmxReadEntryIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
591{
592 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &pVmxTransient->uEntryIntInfo);
593 AssertRCReturn(rc, rc);
594 return VINF_SUCCESS;
595}
596
597
598#ifdef VBOX_STRICT
599/**
600 * Reads the VM-entry exception error code field from the VMCS into
601 * the VMX transient structure.
602 *
603 * @returns VBox status code.
604 * @param pVmxTransient Pointer to the VMX transient structure.
605 *
606 * @remarks No-long-jump zone!!!
607 */
608DECLINLINE(int) hmR0VmxReadEntryXcptErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
609{
610 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &pVmxTransient->uEntryXcptErrorCode);
611 AssertRCReturn(rc, rc);
612 return VINF_SUCCESS;
613}
614#endif /* VBOX_STRICT */
615
616
617#ifdef VBOX_STRICT
618/**
619 * Reads the VM-entry exception error code field from the VMCS into
620 * the VMX transient structure.
621 *
622 * @returns VBox status code.
623 * @param pVmxTransient Pointer to the VMX transient structure.
624 *
625 * @remarks No-long-jump zone!!!
626 */
627DECLINLINE(int) hmR0VmxReadEntryInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
628{
629 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &pVmxTransient->cbEntryInstr);
630 AssertRCReturn(rc, rc);
631 return VINF_SUCCESS;
632}
633#endif /* VBOX_STRICT */
634
635
636/**
637 * Reads the VM-exit interruption-information field from the VMCS into the VMX
638 * transient structure.
639 *
640 * @returns VBox status code.
641 * @param pVmxTransient Pointer to the VMX transient structure.
642 */
643DECLINLINE(int) hmR0VmxReadExitIntInfoVmcs(PVMXTRANSIENT pVmxTransient)
644{
645 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO))
646 {
647 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_INFO, &pVmxTransient->uExitIntInfo);
648 AssertRCReturn(rc, rc);
649 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO;
650 }
651 return VINF_SUCCESS;
652}
653
654
655/**
656 * Reads the VM-exit interruption error code from the VMCS into the VMX
657 * transient structure.
658 *
659 * @returns VBox status code.
660 * @param pVmxTransient Pointer to the VMX transient structure.
661 */
662DECLINLINE(int) hmR0VmxReadExitIntErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
663{
664 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE))
665 {
666 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INTERRUPTION_ERROR_CODE, &pVmxTransient->uExitIntErrorCode);
667 AssertRCReturn(rc, rc);
668 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_ERROR_CODE;
669 }
670 return VINF_SUCCESS;
671}
672
673
674/**
675 * Reads the VM-exit instruction length field from the VMCS into the VMX
676 * transient structure.
677 *
678 * @returns VBox status code.
679 * @param pVmxTransient Pointer to the VMX transient structure.
680 */
681DECLINLINE(int) hmR0VmxReadExitInstrLenVmcs(PVMXTRANSIENT pVmxTransient)
682{
683 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN))
684 {
685 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_LENGTH, &pVmxTransient->cbInstr);
686 AssertRCReturn(rc, rc);
687 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_LEN;
688 }
689 return VINF_SUCCESS;
690}
691
692
693/**
694 * Reads the VM-exit instruction-information field from the VMCS into
695 * the VMX transient structure.
696 *
697 * @returns VBox status code.
698 * @param pVmxTransient Pointer to the VMX transient structure.
699 */
700DECLINLINE(int) hmR0VmxReadExitInstrInfoVmcs(PVMXTRANSIENT pVmxTransient)
701{
702 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO))
703 {
704 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_INSTR_INFO, &pVmxTransient->ExitInstrInfo.u);
705 AssertRCReturn(rc, rc);
706 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_INSTR_INFO;
707 }
708 return VINF_SUCCESS;
709}
710
711
712/**
713 * Reads the exit code qualification from the VMCS into the VMX transient
714 * structure.
715 *
716 * @returns VBox status code.
717 * @param pVCpu The cross context virtual CPU structure of the
718 * calling EMT. (Required for the VMCS cache case.)
719 * @param pVmxTransient Pointer to the VMX transient structure.
720 */
721DECLINLINE(int) hmR0VmxReadExitQualificationVmcs(PVMCPU pVCpu, PVMXTRANSIENT pVmxTransient)
722{
723 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION))
724 {
725 int rc = VMXReadVmcsGstN(VMX_VMCS_RO_EXIT_QUALIFICATION, &pVmxTransient->uExitQualification); NOREF(pVCpu);
726 AssertRCReturn(rc, rc);
727 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_EXIT_QUALIFICATION;
728 }
729 return VINF_SUCCESS;
730}
731
732
733/**
734 * Reads the IDT-vectoring information field from the VMCS into the VMX
735 * transient structure.
736 *
737 * @returns VBox status code.
738 * @param pVmxTransient Pointer to the VMX transient structure.
739 *
740 * @remarks No-long-jump zone!!!
741 */
742DECLINLINE(int) hmR0VmxReadIdtVectoringInfoVmcs(PVMXTRANSIENT pVmxTransient)
743{
744 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO))
745 {
746 int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_INFO, &pVmxTransient->uIdtVectoringInfo);
747 AssertRCReturn(rc, rc);
748 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_INFO;
749 }
750 return VINF_SUCCESS;
751}
752
753
754/**
755 * Reads the IDT-vectoring error code from the VMCS into the VMX
756 * transient structure.
757 *
758 * @returns VBox status code.
759 * @param pVmxTransient Pointer to the VMX transient structure.
760 */
761DECLINLINE(int) hmR0VmxReadIdtVectoringErrorCodeVmcs(PVMXTRANSIENT pVmxTransient)
762{
763 if (!(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE))
764 {
765 int rc = VMXReadVmcs32(VMX_VMCS32_RO_IDT_ERROR_CODE, &pVmxTransient->uIdtVectoringErrorCode);
766 AssertRCReturn(rc, rc);
767 pVmxTransient->fVmcsFieldsRead |= HMVMX_UPDATED_TRANSIENT_IDT_VECTORING_ERROR_CODE;
768 }
769 return VINF_SUCCESS;
770}
771
772
773/**
774 * Enters VMX root mode operation on the current CPU.
775 *
776 * @returns VBox status code.
777 * @param pVM The cross context VM structure. Can be
778 * NULL, after a resume.
779 * @param HCPhysCpuPage Physical address of the VMXON region.
780 * @param pvCpuPage Pointer to the VMXON region.
781 */
782static int hmR0VmxEnterRootMode(PVM pVM, RTHCPHYS HCPhysCpuPage, void *pvCpuPage)
783{
784 Assert(HCPhysCpuPage && HCPhysCpuPage != NIL_RTHCPHYS);
785 Assert(RT_ALIGN_T(HCPhysCpuPage, _4K, RTHCPHYS) == HCPhysCpuPage);
786 Assert(pvCpuPage);
787 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
788
789 if (pVM)
790 {
791 /* Write the VMCS revision dword to the VMXON region. */
792 *(uint32_t *)pvCpuPage = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo);
793 }
794
795 /* Paranoid: Disable interrupts as, in theory, interrupt handlers might mess with CR4. */
796 RTCCUINTREG fEFlags = ASMIntDisableFlags();
797
798 /* Enable the VMX bit in CR4 if necessary. */
799 RTCCUINTREG uOldCr4 = SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
800
801 /* Enter VMX root mode. */
802 int rc = VMXEnable(HCPhysCpuPage);
803 if (RT_FAILURE(rc))
804 {
805 if (!(uOldCr4 & X86_CR4_VMXE))
806 SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
807
808 if (pVM)
809 pVM->hm.s.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
810 }
811
812 /* Restore interrupts. */
813 ASMSetFlags(fEFlags);
814 return rc;
815}
816
817
818/**
819 * Exits VMX root mode operation on the current CPU.
820 *
821 * @returns VBox status code.
822 */
823static int hmR0VmxLeaveRootMode(void)
824{
825 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
826
827 /* Paranoid: Disable interrupts as, in theory, interrupts handlers might mess with CR4. */
828 RTCCUINTREG fEFlags = ASMIntDisableFlags();
829
830 /* If we're for some reason not in VMX root mode, then don't leave it. */
831 RTCCUINTREG uHostCR4 = ASMGetCR4();
832
833 int rc;
834 if (uHostCR4 & X86_CR4_VMXE)
835 {
836 /* Exit VMX root mode and clear the VMX bit in CR4. */
837 VMXDisable();
838 SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
839 rc = VINF_SUCCESS;
840 }
841 else
842 rc = VERR_VMX_NOT_IN_VMX_ROOT_MODE;
843
844 /* Restore interrupts. */
845 ASMSetFlags(fEFlags);
846 return rc;
847}
848
849
850/**
851 * Allocates and maps one physically contiguous page. The allocated page is
852 * zero'd out. (Used by various VT-x structures).
853 *
854 * @returns IPRT status code.
855 * @param pMemObj Pointer to the ring-0 memory object.
856 * @param ppVirt Where to store the virtual address of the
857 * allocation.
858 * @param pHCPhys Where to store the physical address of the
859 * allocation.
860 */
861DECLINLINE(int) hmR0VmxPageAllocZ(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
862{
863 AssertPtrReturn(pMemObj, VERR_INVALID_PARAMETER);
864 AssertPtrReturn(ppVirt, VERR_INVALID_PARAMETER);
865 AssertPtrReturn(pHCPhys, VERR_INVALID_PARAMETER);
866
867 int rc = RTR0MemObjAllocCont(pMemObj, PAGE_SIZE, false /* fExecutable */);
868 if (RT_FAILURE(rc))
869 return rc;
870 *ppVirt = RTR0MemObjAddress(*pMemObj);
871 *pHCPhys = RTR0MemObjGetPagePhysAddr(*pMemObj, 0 /* iPage */);
872 ASMMemZero32(*ppVirt, PAGE_SIZE);
873 return VINF_SUCCESS;
874}
875
876
877/**
878 * Frees and unmaps an allocated physical page.
879 *
880 * @param pMemObj Pointer to the ring-0 memory object.
881 * @param ppVirt Where to re-initialize the virtual address of
882 * allocation as 0.
883 * @param pHCPhys Where to re-initialize the physical address of the
884 * allocation as 0.
885 */
886DECLINLINE(void) hmR0VmxPageFree(PRTR0MEMOBJ pMemObj, PRTR0PTR ppVirt, PRTHCPHYS pHCPhys)
887{
888 AssertPtr(pMemObj);
889 AssertPtr(ppVirt);
890 AssertPtr(pHCPhys);
891 if (*pMemObj != NIL_RTR0MEMOBJ)
892 {
893 int rc = RTR0MemObjFree(*pMemObj, true /* fFreeMappings */);
894 AssertRC(rc);
895 *pMemObj = NIL_RTR0MEMOBJ;
896 *ppVirt = 0;
897 *pHCPhys = 0;
898 }
899}
900
901
902/**
903 * Worker function to free VT-x related structures.
904 *
905 * @returns IPRT status code.
906 * @param pVM The cross context VM structure.
907 */
908static void hmR0VmxStructsFree(PVM pVM)
909{
910 for (VMCPUID i = 0; i < pVM->cCpus; i++)
911 {
912 PVMCPU pVCpu = &pVM->aCpus[i];
913 AssertPtr(pVCpu);
914
915 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjHostMsr, &pVCpu->hm.s.vmx.pvHostMsr, &pVCpu->hm.s.vmx.HCPhysHostMsr);
916 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjGuestMsr, &pVCpu->hm.s.vmx.pvGuestMsr, &pVCpu->hm.s.vmx.HCPhysGuestMsr);
917
918 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
919 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjMsrBitmap, &pVCpu->hm.s.vmx.pvMsrBitmap, &pVCpu->hm.s.vmx.HCPhysMsrBitmap);
920
921 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjVirtApic, (PRTR0PTR)&pVCpu->hm.s.vmx.pbVirtApic, &pVCpu->hm.s.vmx.HCPhysVirtApic);
922 hmR0VmxPageFree(&pVCpu->hm.s.vmx.hMemObjVmcs, &pVCpu->hm.s.vmx.pvVmcs, &pVCpu->hm.s.vmx.HCPhysVmcs);
923 }
924
925 hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess, &pVM->hm.s.vmx.HCPhysApicAccess);
926#ifdef VBOX_WITH_CRASHDUMP_MAGIC
927 hmR0VmxPageFree(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
928#endif
929}
930
931
932/**
933 * Worker function to allocate VT-x related VM structures.
934 *
935 * @returns IPRT status code.
936 * @param pVM The cross context VM structure.
937 */
938static int hmR0VmxStructsAlloc(PVM pVM)
939{
940 /*
941 * Initialize members up-front so we can cleanup properly on allocation failure.
942 */
943#define VMXLOCAL_INIT_VM_MEMOBJ(a_Name, a_VirtPrefix) \
944 pVM->hm.s.vmx.hMemObj##a_Name = NIL_RTR0MEMOBJ; \
945 pVM->hm.s.vmx.a_VirtPrefix##a_Name = 0; \
946 pVM->hm.s.vmx.HCPhys##a_Name = 0;
947
948#define VMXLOCAL_INIT_VMCPU_MEMOBJ(a_Name, a_VirtPrefix) \
949 pVCpu->hm.s.vmx.hMemObj##a_Name = NIL_RTR0MEMOBJ; \
950 pVCpu->hm.s.vmx.a_VirtPrefix##a_Name = 0; \
951 pVCpu->hm.s.vmx.HCPhys##a_Name = 0;
952
953#ifdef VBOX_WITH_CRASHDUMP_MAGIC
954 VMXLOCAL_INIT_VM_MEMOBJ(Scratch, pv);
955#endif
956 VMXLOCAL_INIT_VM_MEMOBJ(ApicAccess, pb);
957
958 AssertCompile(sizeof(VMCPUID) == sizeof(pVM->cCpus));
959 for (VMCPUID i = 0; i < pVM->cCpus; i++)
960 {
961 PVMCPU pVCpu = &pVM->aCpus[i];
962 VMXLOCAL_INIT_VMCPU_MEMOBJ(Vmcs, pv);
963 VMXLOCAL_INIT_VMCPU_MEMOBJ(VirtApic, pb);
964 VMXLOCAL_INIT_VMCPU_MEMOBJ(MsrBitmap, pv);
965 VMXLOCAL_INIT_VMCPU_MEMOBJ(GuestMsr, pv);
966 VMXLOCAL_INIT_VMCPU_MEMOBJ(HostMsr, pv);
967 }
968#undef VMXLOCAL_INIT_VMCPU_MEMOBJ
969#undef VMXLOCAL_INIT_VM_MEMOBJ
970
971 /* The VMCS size cannot be more than 4096 bytes. See Intel spec. Appendix A.1 "Basic VMX Information". */
972 AssertReturnStmt(MSR_IA32_VMX_BASIC_INFO_VMCS_SIZE(pVM->hm.s.vmx.Msrs.u64BasicInfo) <= PAGE_SIZE,
973 (&pVM->aCpus[0])->hm.s.u32HMError = VMX_UFC_INVALID_VMCS_SIZE,
974 VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO);
975
976 /*
977 * Allocate all the VT-x structures.
978 */
979 int rc = VINF_SUCCESS;
980#ifdef VBOX_WITH_CRASHDUMP_MAGIC
981 rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjScratch, &pVM->hm.s.vmx.pbScratch, &pVM->hm.s.vmx.HCPhysScratch);
982 if (RT_FAILURE(rc))
983 goto cleanup;
984 strcpy((char *)pVM->hm.s.vmx.pbScratch, "SCRATCH Magic");
985 *(uint64_t *)(pVM->hm.s.vmx.pbScratch + 16) = UINT64_C(0xdeadbeefdeadbeef);
986#endif
987
988 /* Allocate the APIC-access page for trapping APIC accesses from the guest. */
989 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)
990 {
991 rc = hmR0VmxPageAllocZ(&pVM->hm.s.vmx.hMemObjApicAccess, (PRTR0PTR)&pVM->hm.s.vmx.pbApicAccess,
992 &pVM->hm.s.vmx.HCPhysApicAccess);
993 if (RT_FAILURE(rc))
994 goto cleanup;
995 }
996
997 /*
998 * Initialize per-VCPU VT-x structures.
999 */
1000 for (VMCPUID i = 0; i < pVM->cCpus; i++)
1001 {
1002 PVMCPU pVCpu = &pVM->aCpus[i];
1003 AssertPtr(pVCpu);
1004
1005 /* Allocate the VM control structure (VMCS). */
1006 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjVmcs, &pVCpu->hm.s.vmx.pvVmcs, &pVCpu->hm.s.vmx.HCPhysVmcs);
1007 if (RT_FAILURE(rc))
1008 goto cleanup;
1009
1010 /* Allocate the Virtual-APIC page for transparent TPR accesses. */
1011 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
1012 {
1013 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjVirtApic, (PRTR0PTR)&pVCpu->hm.s.vmx.pbVirtApic,
1014 &pVCpu->hm.s.vmx.HCPhysVirtApic);
1015 if (RT_FAILURE(rc))
1016 goto cleanup;
1017 }
1018
1019 /*
1020 * Allocate the MSR-bitmap if supported by the CPU. The MSR-bitmap is for
1021 * transparent accesses of specific MSRs.
1022 *
1023 * If the condition for enabling MSR bitmaps changes here, don't forget to
1024 * update HMAreMsrBitmapsAvailable().
1025 */
1026 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1027 {
1028 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjMsrBitmap, &pVCpu->hm.s.vmx.pvMsrBitmap,
1029 &pVCpu->hm.s.vmx.HCPhysMsrBitmap);
1030 if (RT_FAILURE(rc))
1031 goto cleanup;
1032 ASMMemFill32(pVCpu->hm.s.vmx.pvMsrBitmap, PAGE_SIZE, UINT32_C(0xffffffff));
1033 }
1034
1035 /* Allocate the VM-entry MSR-load and VM-exit MSR-store page for the guest MSRs. */
1036 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjGuestMsr, &pVCpu->hm.s.vmx.pvGuestMsr, &pVCpu->hm.s.vmx.HCPhysGuestMsr);
1037 if (RT_FAILURE(rc))
1038 goto cleanup;
1039
1040 /* Allocate the VM-exit MSR-load page for the host MSRs. */
1041 rc = hmR0VmxPageAllocZ(&pVCpu->hm.s.vmx.hMemObjHostMsr, &pVCpu->hm.s.vmx.pvHostMsr, &pVCpu->hm.s.vmx.HCPhysHostMsr);
1042 if (RT_FAILURE(rc))
1043 goto cleanup;
1044 }
1045
1046 return VINF_SUCCESS;
1047
1048cleanup:
1049 hmR0VmxStructsFree(pVM);
1050 return rc;
1051}
1052
1053
1054/**
1055 * Does global VT-x initialization (called during module initialization).
1056 *
1057 * @returns VBox status code.
1058 */
1059VMMR0DECL(int) VMXR0GlobalInit(void)
1060{
1061#ifdef HMVMX_USE_FUNCTION_TABLE
1062 AssertCompile(VMX_EXIT_MAX + 1 == RT_ELEMENTS(g_apfnVMExitHandlers));
1063# ifdef VBOX_STRICT
1064 for (unsigned i = 0; i < RT_ELEMENTS(g_apfnVMExitHandlers); i++)
1065 Assert(g_apfnVMExitHandlers[i]);
1066# endif
1067#endif
1068 return VINF_SUCCESS;
1069}
1070
1071
1072/**
1073 * Does global VT-x termination (called during module termination).
1074 */
1075VMMR0DECL(void) VMXR0GlobalTerm()
1076{
1077 /* Nothing to do currently. */
1078}
1079
1080
1081/**
1082 * Sets up and activates VT-x on the current CPU.
1083 *
1084 * @returns VBox status code.
1085 * @param pCpu Pointer to the global CPU info struct.
1086 * @param pVM The cross context VM structure. Can be
1087 * NULL after a host resume operation.
1088 * @param pvCpuPage Pointer to the VMXON region (can be NULL if @a
1089 * fEnabledByHost is @c true).
1090 * @param HCPhysCpuPage Physical address of the VMXON region (can be 0 if
1091 * @a fEnabledByHost is @c true).
1092 * @param fEnabledByHost Set if SUPR0EnableVTx() or similar was used to
1093 * enable VT-x on the host.
1094 * @param pvMsrs Opaque pointer to VMXMSRS struct.
1095 */
1096VMMR0DECL(int) VMXR0EnableCpu(PHMGLOBALCPUINFO pCpu, PVM pVM, void *pvCpuPage, RTHCPHYS HCPhysCpuPage, bool fEnabledByHost,
1097 void *pvMsrs)
1098{
1099 Assert(pCpu);
1100 Assert(pvMsrs);
1101 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1102
1103 /* Enable VT-x if it's not already enabled by the host. */
1104 if (!fEnabledByHost)
1105 {
1106 int rc = hmR0VmxEnterRootMode(pVM, HCPhysCpuPage, pvCpuPage);
1107 if (RT_FAILURE(rc))
1108 return rc;
1109 }
1110
1111 /*
1112 * Flush all EPT tagged-TLB entries (in case VirtualBox or any other hypervisor have been using EPTPs) so
1113 * we don't retain any stale guest-physical mappings which won't get invalidated when flushing by VPID.
1114 */
1115 PVMXMSRS pMsrs = (PVMXMSRS)pvMsrs;
1116 if (pMsrs->u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
1117 {
1118 hmR0VmxFlushEpt(NULL /* pVCpu */, VMXFLUSHEPT_ALL_CONTEXTS);
1119 pCpu->fFlushAsidBeforeUse = false;
1120 }
1121 else
1122 pCpu->fFlushAsidBeforeUse = true;
1123
1124 /* Ensure each VCPU scheduled on this CPU gets a new VPID on resume. See @bugref{6255}. */
1125 ++pCpu->cTlbFlushes;
1126
1127 return VINF_SUCCESS;
1128}
1129
1130
1131/**
1132 * Deactivates VT-x on the current CPU.
1133 *
1134 * @returns VBox status code.
1135 * @param pCpu Pointer to the global CPU info struct.
1136 * @param pvCpuPage Pointer to the VMXON region.
1137 * @param HCPhysCpuPage Physical address of the VMXON region.
1138 *
1139 * @remarks This function should never be called when SUPR0EnableVTx() or
1140 * similar was used to enable VT-x on the host.
1141 */
1142VMMR0DECL(int) VMXR0DisableCpu(PHMGLOBALCPUINFO pCpu, void *pvCpuPage, RTHCPHYS HCPhysCpuPage)
1143{
1144 NOREF(pCpu);
1145 NOREF(pvCpuPage);
1146 NOREF(HCPhysCpuPage);
1147
1148 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1149 return hmR0VmxLeaveRootMode();
1150}
1151
1152
1153/**
1154 * Sets the permission bits for the specified MSR in the MSR bitmap.
1155 *
1156 * @param pVCpu The cross context virtual CPU structure.
1157 * @param uMsr The MSR value.
1158 * @param enmRead Whether reading this MSR causes a VM-exit.
1159 * @param enmWrite Whether writing this MSR causes a VM-exit.
1160 */
1161static void hmR0VmxSetMsrPermission(PVMCPU pVCpu, uint32_t uMsr, VMXMSREXITREAD enmRead, VMXMSREXITWRITE enmWrite)
1162{
1163 int32_t iBit;
1164 uint8_t *pbMsrBitmap = (uint8_t *)pVCpu->hm.s.vmx.pvMsrBitmap;
1165
1166 /*
1167 * Layout:
1168 * 0x000 - 0x3ff - Low MSR read bits
1169 * 0x400 - 0x7ff - High MSR read bits
1170 * 0x800 - 0xbff - Low MSR write bits
1171 * 0xc00 - 0xfff - High MSR write bits
1172 */
1173 if (uMsr <= 0x00001FFF)
1174 iBit = uMsr;
1175 else if (uMsr - UINT32_C(0xC0000000) <= UINT32_C(0x00001FFF))
1176 {
1177 iBit = uMsr - UINT32_C(0xC0000000);
1178 pbMsrBitmap += 0x400;
1179 }
1180 else
1181 AssertMsgFailedReturnVoid(("hmR0VmxSetMsrPermission: Invalid MSR %#RX32\n", uMsr));
1182
1183 Assert(iBit <= 0x1fff);
1184 if (enmRead == VMXMSREXIT_INTERCEPT_READ)
1185 ASMBitSet(pbMsrBitmap, iBit);
1186 else
1187 ASMBitClear(pbMsrBitmap, iBit);
1188
1189 if (enmWrite == VMXMSREXIT_INTERCEPT_WRITE)
1190 ASMBitSet(pbMsrBitmap + 0x800, iBit);
1191 else
1192 ASMBitClear(pbMsrBitmap + 0x800, iBit);
1193}
1194
1195
1196#ifdef VBOX_STRICT
1197/**
1198 * Gets the permission bits for the specified MSR in the MSR bitmap.
1199 *
1200 * @returns VBox status code.
1201 * @retval VINF_SUCCESS if the specified MSR is found.
1202 * @retval VERR_NOT_FOUND if the specified MSR is not found.
1203 * @retval VERR_NOT_SUPPORTED if VT-x doesn't allow the MSR.
1204 *
1205 * @param pVCpu The cross context virtual CPU structure.
1206 * @param uMsr The MSR.
1207 * @param penmRead Where to store the read permissions.
1208 * @param penmWrite Where to store the write permissions.
1209 */
1210static int hmR0VmxGetMsrPermission(PVMCPU pVCpu, uint32_t uMsr, PVMXMSREXITREAD penmRead, PVMXMSREXITWRITE penmWrite)
1211{
1212 AssertPtrReturn(penmRead, VERR_INVALID_PARAMETER);
1213 AssertPtrReturn(penmWrite, VERR_INVALID_PARAMETER);
1214 int32_t iBit;
1215 uint8_t *pbMsrBitmap = (uint8_t *)pVCpu->hm.s.vmx.pvMsrBitmap;
1216
1217 /* See hmR0VmxSetMsrPermission() for the layout. */
1218 if (uMsr <= 0x00001FFF)
1219 iBit = uMsr;
1220 else if ( uMsr >= 0xC0000000
1221 && uMsr <= 0xC0001FFF)
1222 {
1223 iBit = (uMsr - 0xC0000000);
1224 pbMsrBitmap += 0x400;
1225 }
1226 else
1227 AssertMsgFailedReturn(("hmR0VmxGetMsrPermission: Invalid MSR %#RX32\n", uMsr), VERR_NOT_SUPPORTED);
1228
1229 Assert(iBit <= 0x1fff);
1230 if (ASMBitTest(pbMsrBitmap, iBit))
1231 *penmRead = VMXMSREXIT_INTERCEPT_READ;
1232 else
1233 *penmRead = VMXMSREXIT_PASSTHRU_READ;
1234
1235 if (ASMBitTest(pbMsrBitmap + 0x800, iBit))
1236 *penmWrite = VMXMSREXIT_INTERCEPT_WRITE;
1237 else
1238 *penmWrite = VMXMSREXIT_PASSTHRU_WRITE;
1239 return VINF_SUCCESS;
1240}
1241#endif /* VBOX_STRICT */
1242
1243
1244/**
1245 * Updates the VMCS with the number of effective MSRs in the auto-load/store MSR
1246 * area.
1247 *
1248 * @returns VBox status code.
1249 * @param pVCpu The cross context virtual CPU structure.
1250 * @param cMsrs The number of MSRs.
1251 */
1252DECLINLINE(int) hmR0VmxSetAutoLoadStoreMsrCount(PVMCPU pVCpu, uint32_t cMsrs)
1253{
1254 /* Shouldn't ever happen but there -is- a number. We're well within the recommended 512. */
1255 uint32_t const cMaxSupportedMsrs = MSR_IA32_VMX_MISC_MAX_MSR(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.u64Misc);
1256 if (RT_UNLIKELY(cMsrs > cMaxSupportedMsrs))
1257 {
1258 LogRel(("CPU auto-load/store MSR count in VMCS exceeded cMsrs=%u Supported=%u.\n", cMsrs, cMaxSupportedMsrs));
1259 pVCpu->hm.s.u32HMError = VMX_UFC_INSUFFICIENT_GUEST_MSR_STORAGE;
1260 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
1261 }
1262
1263 /* Update number of guest MSRs to load/store across the world-switch. */
1264 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, cMsrs);
1265 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, cMsrs);
1266
1267 /* Update number of host MSRs to load after the world-switch. Identical to guest-MSR count as it's always paired. */
1268 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, cMsrs);
1269 AssertRCReturn(rc, rc);
1270
1271 /* Update the VCPU's copy of the MSR count. */
1272 pVCpu->hm.s.vmx.cMsrs = cMsrs;
1273
1274 return VINF_SUCCESS;
1275}
1276
1277
1278/**
1279 * Adds a new (or updates the value of an existing) guest/host MSR
1280 * pair to be swapped during the world-switch as part of the
1281 * auto-load/store MSR area in the VMCS.
1282 *
1283 * @returns VBox status code.
1284 * @param pVCpu The cross context virtual CPU structure.
1285 * @param uMsr The MSR.
1286 * @param uGuestMsrValue Value of the guest MSR.
1287 * @param fUpdateHostMsr Whether to update the value of the host MSR if
1288 * necessary.
1289 * @param pfAddedAndUpdated Where to store whether the MSR was added -and-
1290 * its value was updated. Optional, can be NULL.
1291 */
1292static int hmR0VmxAddAutoLoadStoreMsr(PVMCPU pVCpu, uint32_t uMsr, uint64_t uGuestMsrValue, bool fUpdateHostMsr,
1293 bool *pfAddedAndUpdated)
1294{
1295 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1296 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1297 uint32_t i;
1298 for (i = 0; i < cMsrs; i++)
1299 {
1300 if (pGuestMsr->u32Msr == uMsr)
1301 break;
1302 pGuestMsr++;
1303 }
1304
1305 bool fAdded = false;
1306 if (i == cMsrs)
1307 {
1308 ++cMsrs;
1309 int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, cMsrs);
1310 AssertMsgRCReturn(rc, ("hmR0VmxAddAutoLoadStoreMsr: Insufficient space to add MSR %u\n", uMsr), rc);
1311
1312 /* Now that we're swapping MSRs during the world-switch, allow the guest to read/write them without causing VM-exits. */
1313 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1314 hmR0VmxSetMsrPermission(pVCpu, uMsr, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
1315
1316 fAdded = true;
1317 }
1318
1319 /* Update the MSR values in the auto-load/store MSR area. */
1320 pGuestMsr->u32Msr = uMsr;
1321 pGuestMsr->u64Value = uGuestMsrValue;
1322
1323 /* Create/update the MSR slot in the host MSR area. */
1324 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1325 pHostMsr += i;
1326 pHostMsr->u32Msr = uMsr;
1327
1328 /*
1329 * Update the host MSR only when requested by the caller AND when we're
1330 * adding it to the auto-load/store area. Otherwise, it would have been
1331 * updated by hmR0VmxSaveHostMsrs(). We do this for performance reasons.
1332 */
1333 bool fUpdatedMsrValue = false;
1334 if ( fAdded
1335 && fUpdateHostMsr)
1336 {
1337 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1338 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1339 pHostMsr->u64Value = ASMRdMsr(pHostMsr->u32Msr);
1340 fUpdatedMsrValue = true;
1341 }
1342
1343 if (pfAddedAndUpdated)
1344 *pfAddedAndUpdated = fUpdatedMsrValue;
1345 return VINF_SUCCESS;
1346}
1347
1348
1349/**
1350 * Removes a guest/host MSR pair to be swapped during the world-switch from the
1351 * auto-load/store MSR area in the VMCS.
1352 *
1353 * @returns VBox status code.
1354 * @param pVCpu The cross context virtual CPU structure.
1355 * @param uMsr The MSR.
1356 */
1357static int hmR0VmxRemoveAutoLoadStoreMsr(PVMCPU pVCpu, uint32_t uMsr)
1358{
1359 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1360 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1361 for (uint32_t i = 0; i < cMsrs; i++)
1362 {
1363 /* Find the MSR. */
1364 if (pGuestMsr->u32Msr == uMsr)
1365 {
1366 /* If it's the last MSR, simply reduce the count. */
1367 if (i == cMsrs - 1)
1368 {
1369 --cMsrs;
1370 break;
1371 }
1372
1373 /* Remove it by swapping the last MSR in place of it, and reducing the count. */
1374 PVMXAUTOMSR pLastGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1375 pLastGuestMsr += cMsrs - 1;
1376 pGuestMsr->u32Msr = pLastGuestMsr->u32Msr;
1377 pGuestMsr->u64Value = pLastGuestMsr->u64Value;
1378
1379 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1380 PVMXAUTOMSR pLastHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1381 pLastHostMsr += cMsrs - 1;
1382 pHostMsr->u32Msr = pLastHostMsr->u32Msr;
1383 pHostMsr->u64Value = pLastHostMsr->u64Value;
1384 --cMsrs;
1385 break;
1386 }
1387 pGuestMsr++;
1388 }
1389
1390 /* Update the VMCS if the count changed (meaning the MSR was found). */
1391 if (cMsrs != pVCpu->hm.s.vmx.cMsrs)
1392 {
1393 int rc = hmR0VmxSetAutoLoadStoreMsrCount(pVCpu, cMsrs);
1394 AssertRCReturn(rc, rc);
1395
1396 /* We're no longer swapping MSRs during the world-switch, intercept guest read/writes to them. */
1397 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1398 hmR0VmxSetMsrPermission(pVCpu, uMsr, VMXMSREXIT_INTERCEPT_READ, VMXMSREXIT_INTERCEPT_WRITE);
1399
1400 Log4(("Removed MSR %#RX32 new cMsrs=%u\n", uMsr, pVCpu->hm.s.vmx.cMsrs));
1401 return VINF_SUCCESS;
1402 }
1403
1404 return VERR_NOT_FOUND;
1405}
1406
1407
1408/**
1409 * Checks if the specified guest MSR is part of the auto-load/store area in
1410 * the VMCS.
1411 *
1412 * @returns true if found, false otherwise.
1413 * @param pVCpu The cross context virtual CPU structure.
1414 * @param uMsr The MSR to find.
1415 */
1416static bool hmR0VmxIsAutoLoadStoreGuestMsr(PVMCPU pVCpu, uint32_t uMsr)
1417{
1418 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1419 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1420
1421 for (uint32_t i = 0; i < cMsrs; i++, pGuestMsr++)
1422 {
1423 if (pGuestMsr->u32Msr == uMsr)
1424 return true;
1425 }
1426 return false;
1427}
1428
1429
1430/**
1431 * Updates the value of all host MSRs in the auto-load/store area in the VMCS.
1432 *
1433 * @param pVCpu The cross context virtual CPU structure.
1434 *
1435 * @remarks No-long-jump zone!!!
1436 */
1437static void hmR0VmxUpdateAutoLoadStoreHostMsrs(PVMCPU pVCpu)
1438{
1439 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1440 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1441 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1442 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
1443
1444 for (uint32_t i = 0; i < cMsrs; i++, pHostMsr++, pGuestMsr++)
1445 {
1446 AssertReturnVoid(pHostMsr->u32Msr == pGuestMsr->u32Msr);
1447
1448 /*
1449 * Performance hack for the host EFER MSR. We use the cached value rather than re-read it.
1450 * Strict builds will catch mismatches in hmR0VmxCheckAutoLoadStoreMsrs(). See @bugref{7368}.
1451 */
1452 if (pHostMsr->u32Msr == MSR_K6_EFER)
1453 pHostMsr->u64Value = pVCpu->CTX_SUFF(pVM)->hm.s.vmx.u64HostEfer;
1454 else
1455 pHostMsr->u64Value = ASMRdMsr(pHostMsr->u32Msr);
1456 }
1457
1458 pVCpu->hm.s.vmx.fUpdatedHostMsrs = true;
1459}
1460
1461
1462/**
1463 * Saves a set of host MSRs to allow read/write passthru access to the guest and
1464 * perform lazy restoration of the host MSRs while leaving VT-x.
1465 *
1466 * @param pVCpu The cross context virtual CPU structure.
1467 *
1468 * @remarks No-long-jump zone!!!
1469 */
1470static void hmR0VmxLazySaveHostMsrs(PVMCPU pVCpu)
1471{
1472 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1473
1474 /*
1475 * Note: If you're adding MSRs here, make sure to update the MSR-bitmap permissions in hmR0VmxSetupProcCtls().
1476 */
1477 if (!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST))
1478 {
1479 Assert(!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)); /* Guest MSRs better not be loaded now. */
1480#if HC_ARCH_BITS == 64
1481 if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
1482 {
1483 pVCpu->hm.s.vmx.u64HostLStarMsr = ASMRdMsr(MSR_K8_LSTAR);
1484 pVCpu->hm.s.vmx.u64HostStarMsr = ASMRdMsr(MSR_K6_STAR);
1485 pVCpu->hm.s.vmx.u64HostSFMaskMsr = ASMRdMsr(MSR_K8_SF_MASK);
1486 pVCpu->hm.s.vmx.u64HostKernelGSBaseMsr = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
1487 }
1488#endif
1489 pVCpu->hm.s.vmx.fLazyMsrs |= VMX_LAZY_MSRS_SAVED_HOST;
1490 }
1491}
1492
1493
1494/**
1495 * Checks whether the MSR belongs to the set of guest MSRs that we restore
1496 * lazily while leaving VT-x.
1497 *
1498 * @returns true if it does, false otherwise.
1499 * @param pVCpu The cross context virtual CPU structure.
1500 * @param uMsr The MSR to check.
1501 */
1502static bool hmR0VmxIsLazyGuestMsr(PVMCPU pVCpu, uint32_t uMsr)
1503{
1504 NOREF(pVCpu);
1505#if HC_ARCH_BITS == 64
1506 if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
1507 {
1508 switch (uMsr)
1509 {
1510 case MSR_K8_LSTAR:
1511 case MSR_K6_STAR:
1512 case MSR_K8_SF_MASK:
1513 case MSR_K8_KERNEL_GS_BASE:
1514 return true;
1515 }
1516 }
1517#else
1518 RT_NOREF(pVCpu, uMsr);
1519#endif
1520 return false;
1521}
1522
1523
1524/**
1525 * Saves a set of guest MSRs back into the guest-CPU context.
1526 *
1527 * @param pVCpu The cross context virtual CPU structure.
1528 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
1529 * out-of-sync. Make sure to update the required fields
1530 * before using them.
1531 *
1532 * @remarks No-long-jump zone!!!
1533 */
1534static void hmR0VmxLazySaveGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
1535{
1536 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1537 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1538
1539 if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
1540 {
1541 Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1542#if HC_ARCH_BITS == 64
1543 if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
1544 {
1545 pMixedCtx->msrLSTAR = ASMRdMsr(MSR_K8_LSTAR);
1546 pMixedCtx->msrSTAR = ASMRdMsr(MSR_K6_STAR);
1547 pMixedCtx->msrSFMASK = ASMRdMsr(MSR_K8_SF_MASK);
1548 pMixedCtx->msrKERNELGSBASE = ASMRdMsr(MSR_K8_KERNEL_GS_BASE);
1549 }
1550#else
1551 NOREF(pMixedCtx);
1552#endif
1553 }
1554}
1555
1556
1557/**
1558 * Loads a set of guests MSRs to allow read/passthru to the guest.
1559 *
1560 * The name of this function is slightly confusing. This function does NOT
1561 * postpone loading, but loads the MSR right now. "hmR0VmxLazy" is simply a
1562 * common prefix for functions dealing with "lazy restoration" of the shared
1563 * MSRs.
1564 *
1565 * @param pVCpu The cross context virtual CPU structure.
1566 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
1567 * out-of-sync. Make sure to update the required fields
1568 * before using them.
1569 *
1570 * @remarks No-long-jump zone!!!
1571 */
1572static void hmR0VmxLazyLoadGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
1573{
1574 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1575 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1576
1577#define VMXLOCAL_LAZY_LOAD_GUEST_MSR(uMsr, a_GuestMsr, a_HostMsr) \
1578 do { \
1579 if (pMixedCtx->msr##a_GuestMsr != pVCpu->hm.s.vmx.u64Host##a_HostMsr##Msr) \
1580 ASMWrMsr(uMsr, pMixedCtx->msr##a_GuestMsr); \
1581 else \
1582 Assert(ASMRdMsr(uMsr) == pVCpu->hm.s.vmx.u64Host##a_HostMsr##Msr); \
1583 } while (0)
1584
1585 Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1586 if (!(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST))
1587 {
1588#if HC_ARCH_BITS == 64
1589 if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
1590 {
1591 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K8_LSTAR, LSTAR, LStar);
1592 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K6_STAR, STAR, Star);
1593 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K8_SF_MASK, SFMASK, SFMask);
1594 VMXLOCAL_LAZY_LOAD_GUEST_MSR(MSR_K8_KERNEL_GS_BASE, KERNELGSBASE, KernelGSBase);
1595 }
1596#else
1597 RT_NOREF(pMixedCtx);
1598#endif
1599 pVCpu->hm.s.vmx.fLazyMsrs |= VMX_LAZY_MSRS_LOADED_GUEST;
1600 }
1601
1602#undef VMXLOCAL_LAZY_LOAD_GUEST_MSR
1603}
1604
1605
1606/**
1607 * Performs lazy restoration of the set of host MSRs if they were previously
1608 * loaded with guest MSR values.
1609 *
1610 * @param pVCpu The cross context virtual CPU structure.
1611 *
1612 * @remarks No-long-jump zone!!!
1613 * @remarks The guest MSRs should have been saved back into the guest-CPU
1614 * context by hmR0VmxSaveGuestLazyMsrs()!!!
1615 */
1616static void hmR0VmxLazyRestoreHostMsrs(PVMCPU pVCpu)
1617{
1618 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1619 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
1620
1621 if (pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_LOADED_GUEST)
1622 {
1623 Assert(pVCpu->hm.s.vmx.fLazyMsrs & VMX_LAZY_MSRS_SAVED_HOST);
1624#if HC_ARCH_BITS == 64
1625 if (pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests)
1626 {
1627 ASMWrMsr(MSR_K8_LSTAR, pVCpu->hm.s.vmx.u64HostLStarMsr);
1628 ASMWrMsr(MSR_K6_STAR, pVCpu->hm.s.vmx.u64HostStarMsr);
1629 ASMWrMsr(MSR_K8_SF_MASK, pVCpu->hm.s.vmx.u64HostSFMaskMsr);
1630 ASMWrMsr(MSR_K8_KERNEL_GS_BASE, pVCpu->hm.s.vmx.u64HostKernelGSBaseMsr);
1631 }
1632#endif
1633 }
1634 pVCpu->hm.s.vmx.fLazyMsrs &= ~(VMX_LAZY_MSRS_LOADED_GUEST | VMX_LAZY_MSRS_SAVED_HOST);
1635}
1636
1637
1638/**
1639 * Verifies that our cached values of the VMCS controls are all
1640 * consistent with what's actually present in the VMCS.
1641 *
1642 * @returns VBox status code.
1643 * @param pVCpu The cross context virtual CPU structure.
1644 */
1645static int hmR0VmxCheckVmcsCtls(PVMCPU pVCpu)
1646{
1647 uint32_t u32Val;
1648 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
1649 AssertRCReturn(rc, rc);
1650 AssertMsgReturn(pVCpu->hm.s.vmx.u32EntryCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32EntryCtls, u32Val),
1651 VERR_VMX_ENTRY_CTLS_CACHE_INVALID);
1652
1653 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val);
1654 AssertRCReturn(rc, rc);
1655 AssertMsgReturn(pVCpu->hm.s.vmx.u32ExitCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ExitCtls, u32Val),
1656 VERR_VMX_EXIT_CTLS_CACHE_INVALID);
1657
1658 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val);
1659 AssertRCReturn(rc, rc);
1660 AssertMsgReturn(pVCpu->hm.s.vmx.u32PinCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32PinCtls, u32Val),
1661 VERR_VMX_PIN_EXEC_CTLS_CACHE_INVALID);
1662
1663 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val);
1664 AssertRCReturn(rc, rc);
1665 AssertMsgReturn(pVCpu->hm.s.vmx.u32ProcCtls == u32Val, ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ProcCtls, u32Val),
1666 VERR_VMX_PROC_EXEC_CTLS_CACHE_INVALID);
1667
1668 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
1669 {
1670 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val);
1671 AssertRCReturn(rc, rc);
1672 AssertMsgReturn(pVCpu->hm.s.vmx.u32ProcCtls2 == u32Val,
1673 ("Cache=%#RX32 VMCS=%#RX32", pVCpu->hm.s.vmx.u32ProcCtls2, u32Val),
1674 VERR_VMX_PROC_EXEC2_CTLS_CACHE_INVALID);
1675 }
1676
1677 return VINF_SUCCESS;
1678}
1679
1680
1681#ifdef VBOX_STRICT
1682/**
1683 * Verifies that our cached host EFER value has not changed
1684 * since we cached it.
1685 *
1686 * @param pVCpu The cross context virtual CPU structure.
1687 */
1688static void hmR0VmxCheckHostEferMsr(PVMCPU pVCpu)
1689{
1690 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1691
1692 if (pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR)
1693 {
1694 uint64_t u64Val;
1695 int rc = VMXReadVmcs64(VMX_VMCS64_HOST_EFER_FULL, &u64Val);
1696 AssertRC(rc);
1697
1698 uint64_t u64HostEferMsr = ASMRdMsr(MSR_K6_EFER);
1699 AssertMsgReturnVoid(u64HostEferMsr == u64Val, ("u64HostEferMsr=%#RX64 u64Val=%#RX64\n", u64HostEferMsr, u64Val));
1700 }
1701}
1702
1703
1704/**
1705 * Verifies whether the guest/host MSR pairs in the auto-load/store area in the
1706 * VMCS are correct.
1707 *
1708 * @param pVCpu The cross context virtual CPU structure.
1709 */
1710static void hmR0VmxCheckAutoLoadStoreMsrs(PVMCPU pVCpu)
1711{
1712 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1713
1714 /* Verify MSR counts in the VMCS are what we think it should be. */
1715 uint32_t cMsrs;
1716 int rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &cMsrs); AssertRC(rc);
1717 Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1718
1719 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &cMsrs); AssertRC(rc);
1720 Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1721
1722 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &cMsrs); AssertRC(rc);
1723 Assert(cMsrs == pVCpu->hm.s.vmx.cMsrs);
1724
1725 PVMXAUTOMSR pHostMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvHostMsr;
1726 PVMXAUTOMSR pGuestMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
1727 for (uint32_t i = 0; i < cMsrs; i++, pHostMsr++, pGuestMsr++)
1728 {
1729 /* Verify that the MSRs are paired properly and that the host MSR has the correct value. */
1730 AssertMsgReturnVoid(pHostMsr->u32Msr == pGuestMsr->u32Msr, ("HostMsr=%#RX32 GuestMsr=%#RX32 cMsrs=%u\n", pHostMsr->u32Msr,
1731 pGuestMsr->u32Msr, cMsrs));
1732
1733 uint64_t u64Msr = ASMRdMsr(pHostMsr->u32Msr);
1734 AssertMsgReturnVoid(pHostMsr->u64Value == u64Msr, ("u32Msr=%#RX32 VMCS Value=%#RX64 ASMRdMsr=%#RX64 cMsrs=%u\n",
1735 pHostMsr->u32Msr, pHostMsr->u64Value, u64Msr, cMsrs));
1736
1737 /* Verify that the permissions are as expected in the MSR bitmap. */
1738 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
1739 {
1740 VMXMSREXITREAD enmRead;
1741 VMXMSREXITWRITE enmWrite;
1742 rc = hmR0VmxGetMsrPermission(pVCpu, pGuestMsr->u32Msr, &enmRead, &enmWrite);
1743 AssertMsgReturnVoid(rc == VINF_SUCCESS, ("hmR0VmxGetMsrPermission! failed. rc=%Rrc\n", rc));
1744 if (pGuestMsr->u32Msr == MSR_K6_EFER)
1745 {
1746 AssertMsgReturnVoid(enmRead == VMXMSREXIT_INTERCEPT_READ, ("Passthru read for EFER!?\n"));
1747 AssertMsgReturnVoid(enmWrite == VMXMSREXIT_INTERCEPT_WRITE, ("Passthru write for EFER!?\n"));
1748 }
1749 else
1750 {
1751 AssertMsgReturnVoid(enmRead == VMXMSREXIT_PASSTHRU_READ, ("u32Msr=%#RX32 cMsrs=%u No passthru read!\n",
1752 pGuestMsr->u32Msr, cMsrs));
1753 AssertMsgReturnVoid(enmWrite == VMXMSREXIT_PASSTHRU_WRITE, ("u32Msr=%#RX32 cMsrs=%u No passthru write!\n",
1754 pGuestMsr->u32Msr, cMsrs));
1755 }
1756 }
1757 }
1758}
1759#endif /* VBOX_STRICT */
1760
1761
1762/**
1763 * Flushes the TLB using EPT.
1764 *
1765 * @returns VBox status code.
1766 * @param pVCpu The cross context virtual CPU structure of the calling
1767 * EMT. Can be NULL depending on @a enmFlush.
1768 * @param enmFlush Type of flush.
1769 *
1770 * @remarks Caller is responsible for making sure this function is called only
1771 * when NestedPaging is supported and providing @a enmFlush that is
1772 * supported by the CPU.
1773 * @remarks Can be called with interrupts disabled.
1774 */
1775static void hmR0VmxFlushEpt(PVMCPU pVCpu, VMXFLUSHEPT enmFlush)
1776{
1777 uint64_t au64Descriptor[2];
1778 if (enmFlush == VMXFLUSHEPT_ALL_CONTEXTS)
1779 au64Descriptor[0] = 0;
1780 else
1781 {
1782 Assert(pVCpu);
1783 au64Descriptor[0] = pVCpu->hm.s.vmx.HCPhysEPTP;
1784 }
1785 au64Descriptor[1] = 0; /* MBZ. Intel spec. 33.3 "VMX Instructions" */
1786
1787 int rc = VMXR0InvEPT(enmFlush, &au64Descriptor[0]);
1788 AssertMsg(rc == VINF_SUCCESS, ("VMXR0InvEPT %#x %RGv failed with %Rrc\n", enmFlush, pVCpu ? pVCpu->hm.s.vmx.HCPhysEPTP : 0,
1789 rc));
1790 if ( RT_SUCCESS(rc)
1791 && pVCpu)
1792 {
1793 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushNestedPaging);
1794 }
1795}
1796
1797
1798/**
1799 * Flushes the TLB using VPID.
1800 *
1801 * @returns VBox status code.
1802 * @param pVM The cross context VM structure.
1803 * @param pVCpu The cross context virtual CPU structure of the calling
1804 * EMT. Can be NULL depending on @a enmFlush.
1805 * @param enmFlush Type of flush.
1806 * @param GCPtr Virtual address of the page to flush (can be 0 depending
1807 * on @a enmFlush).
1808 *
1809 * @remarks Can be called with interrupts disabled.
1810 */
1811static void hmR0VmxFlushVpid(PVM pVM, PVMCPU pVCpu, VMXFLUSHVPID enmFlush, RTGCPTR GCPtr)
1812{
1813 NOREF(pVM);
1814 AssertPtr(pVM);
1815 Assert(pVM->hm.s.vmx.fVpid);
1816
1817 uint64_t au64Descriptor[2];
1818 if (enmFlush == VMXFLUSHVPID_ALL_CONTEXTS)
1819 {
1820 au64Descriptor[0] = 0;
1821 au64Descriptor[1] = 0;
1822 }
1823 else
1824 {
1825 AssertPtr(pVCpu);
1826 AssertMsg(pVCpu->hm.s.uCurrentAsid != 0, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
1827 AssertMsg(pVCpu->hm.s.uCurrentAsid <= UINT16_MAX, ("VMXR0InvVPID: invalid ASID %lu\n", pVCpu->hm.s.uCurrentAsid));
1828 au64Descriptor[0] = pVCpu->hm.s.uCurrentAsid;
1829 au64Descriptor[1] = GCPtr;
1830 }
1831
1832 int rc = VMXR0InvVPID(enmFlush, &au64Descriptor[0]); NOREF(rc);
1833 AssertMsg(rc == VINF_SUCCESS,
1834 ("VMXR0InvVPID %#x %u %RGv failed with %d\n", enmFlush, pVCpu ? pVCpu->hm.s.uCurrentAsid : 0, GCPtr, rc));
1835 if ( RT_SUCCESS(rc)
1836 && pVCpu)
1837 {
1838 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushAsid);
1839 }
1840}
1841
1842
1843/**
1844 * Invalidates a guest page by guest virtual address. Only relevant for
1845 * EPT/VPID, otherwise there is nothing really to invalidate.
1846 *
1847 * @returns VBox status code.
1848 * @param pVM The cross context VM structure.
1849 * @param pVCpu The cross context virtual CPU structure.
1850 * @param GCVirt Guest virtual address of the page to invalidate.
1851 */
1852VMMR0DECL(int) VMXR0InvalidatePage(PVM pVM, PVMCPU pVCpu, RTGCPTR GCVirt)
1853{
1854 AssertPtr(pVM);
1855 AssertPtr(pVCpu);
1856 LogFlowFunc(("pVM=%p pVCpu=%p GCVirt=%RGv\n", pVM, pVCpu, GCVirt));
1857
1858 bool fFlushPending = VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_TLB_FLUSH);
1859 if (!fFlushPending)
1860 {
1861 /*
1862 * We must invalidate the guest TLB entry in either case, we cannot ignore it even for the EPT case
1863 * See @bugref{6043} and @bugref{6177}.
1864 *
1865 * Set the VMCPU_FF_TLB_FLUSH force flag and flush before VM-entry in hmR0VmxFlushTLB*() as this
1866 * function maybe called in a loop with individual addresses.
1867 */
1868 if (pVM->hm.s.vmx.fVpid)
1869 {
1870 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
1871 {
1872 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_INDIV_ADDR, GCVirt);
1873 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgVirt);
1874 }
1875 else
1876 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1877 }
1878 else if (pVM->hm.s.fNestedPaging)
1879 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1880 }
1881
1882 return VINF_SUCCESS;
1883}
1884
1885
1886/**
1887 * Invalidates a guest page by physical address. Only relevant for EPT/VPID,
1888 * otherwise there is nothing really to invalidate.
1889 *
1890 * @returns VBox status code.
1891 * @param pVM The cross context VM structure.
1892 * @param pVCpu The cross context virtual CPU structure.
1893 * @param GCPhys Guest physical address of the page to invalidate.
1894 */
1895VMMR0DECL(int) VMXR0InvalidatePhysPage(PVM pVM, PVMCPU pVCpu, RTGCPHYS GCPhys)
1896{
1897 NOREF(pVM); NOREF(GCPhys);
1898 LogFlowFunc(("%RGp\n", GCPhys));
1899
1900 /*
1901 * We cannot flush a page by guest-physical address. invvpid takes only a linear address while invept only flushes
1902 * by EPT not individual addresses. We update the force flag here and flush before the next VM-entry in hmR0VmxFlushTLB*().
1903 * This function might be called in a loop. This should cause a flush-by-EPT if EPT is in use. See @bugref{6568}.
1904 */
1905 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
1906 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbInvlpgPhys);
1907 return VINF_SUCCESS;
1908}
1909
1910
1911/**
1912 * Dummy placeholder for tagged-TLB flush handling before VM-entry. Used in the
1913 * case where neither EPT nor VPID is supported by the CPU.
1914 *
1915 * @param pVM The cross context VM structure.
1916 * @param pVCpu The cross context virtual CPU structure.
1917 * @param pCpu Pointer to the global HM struct.
1918 *
1919 * @remarks Called with interrupts disabled.
1920 */
1921static void hmR0VmxFlushTaggedTlbNone(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
1922{
1923 AssertPtr(pVCpu);
1924 AssertPtr(pCpu);
1925 NOREF(pVM);
1926
1927 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH);
1928
1929 Assert(pCpu->idCpu != NIL_RTCPUID);
1930 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
1931 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
1932 pVCpu->hm.s.fForceTLBFlush = false;
1933 return;
1934}
1935
1936
1937/**
1938 * Flushes the tagged-TLB entries for EPT+VPID CPUs as necessary.
1939 *
1940 * @param pVM The cross context VM structure.
1941 * @param pVCpu The cross context virtual CPU structure.
1942 * @param pCpu Pointer to the global HM CPU struct.
1943 * @remarks All references to "ASID" in this function pertains to "VPID" in
1944 * Intel's nomenclature. The reason is, to avoid confusion in compare
1945 * statements since the host-CPU copies are named "ASID".
1946 *
1947 * @remarks Called with interrupts disabled.
1948 */
1949static void hmR0VmxFlushTaggedTlbBoth(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
1950{
1951#ifdef VBOX_WITH_STATISTICS
1952 bool fTlbFlushed = false;
1953# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { fTlbFlushed = true; } while (0)
1954# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { \
1955 if (!fTlbFlushed) \
1956 STAM_COUNTER_INC(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch); \
1957 } while (0)
1958#else
1959# define HMVMX_SET_TAGGED_TLB_FLUSHED() do { } while (0)
1960# define HMVMX_UPDATE_FLUSH_SKIPPED_STAT() do { } while (0)
1961#endif
1962
1963 AssertPtr(pVM);
1964 AssertPtr(pCpu);
1965 AssertPtr(pVCpu);
1966 Assert(pCpu->idCpu != NIL_RTCPUID);
1967
1968 AssertMsg(pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid,
1969 ("hmR0VmxFlushTaggedTlbBoth cannot be invoked unless NestedPaging & VPID are enabled."
1970 "fNestedPaging=%RTbool fVpid=%RTbool", pVM->hm.s.fNestedPaging, pVM->hm.s.vmx.fVpid));
1971
1972 /*
1973 * Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
1974 * If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB
1975 * or the host CPU is online after a suspend/resume, so we cannot reuse the current ASID anymore.
1976 */
1977 if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
1978 || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
1979 {
1980 ++pCpu->uCurrentAsid;
1981 if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
1982 {
1983 pCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0. */
1984 pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
1985 pCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
1986 }
1987
1988 pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid;
1989 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
1990 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
1991
1992 /*
1993 * Flush by EPT when we get rescheduled to a new host CPU to ensure EPT-only tagged mappings are also
1994 * invalidated. We don't need to flush-by-VPID here as flushing by EPT covers it. See @bugref{6568}.
1995 */
1996 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
1997 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
1998 HMVMX_SET_TAGGED_TLB_FLUSHED();
1999 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH); /* Already flushed-by-EPT, skip doing it again below. */
2000 }
2001
2002 /* Check for explicit TLB flushes. */
2003 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
2004 {
2005 /*
2006 * Changes to the EPT paging structure by VMM requires flushing by EPT as the CPU creates
2007 * guest-physical (only EPT-tagged) mappings while traversing the EPT tables when EPT is in use.
2008 * Flushing by VPID will only flush linear (only VPID-tagged) and combined (EPT+VPID tagged) mappings
2009 * but not guest-physical mappings.
2010 * See Intel spec. 28.3.2 "Creating and Using Cached Translation Information". See @bugref{6568}.
2011 */
2012 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
2013 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2014 HMVMX_SET_TAGGED_TLB_FLUSHED();
2015 }
2016
2017 pVCpu->hm.s.fForceTLBFlush = false;
2018 HMVMX_UPDATE_FLUSH_SKIPPED_STAT();
2019
2020 Assert(pVCpu->hm.s.idLastCpu == pCpu->idCpu);
2021 Assert(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes);
2022 AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes,
2023 ("Flush count mismatch for cpu %d (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
2024 AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
2025 ("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pCpu->idCpu,
2026 pCpu->uCurrentAsid, pCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
2027 AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
2028 ("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
2029
2030 /* Update VMCS with the VPID. */
2031 int rc = VMXWriteVmcs32(VMX_VMCS16_VPID, pVCpu->hm.s.uCurrentAsid);
2032 AssertRC(rc);
2033
2034#undef HMVMX_SET_TAGGED_TLB_FLUSHED
2035}
2036
2037
2038/**
2039 * Flushes the tagged-TLB entries for EPT CPUs as necessary.
2040 *
2041 * @returns VBox status code.
2042 * @param pVM The cross context VM structure.
2043 * @param pVCpu The cross context virtual CPU structure.
2044 * @param pCpu Pointer to the global HM CPU struct.
2045 *
2046 * @remarks Called with interrupts disabled.
2047 */
2048static void hmR0VmxFlushTaggedTlbEpt(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2049{
2050 AssertPtr(pVM);
2051 AssertPtr(pVCpu);
2052 AssertPtr(pCpu);
2053 Assert(pCpu->idCpu != NIL_RTCPUID);
2054 AssertMsg(pVM->hm.s.fNestedPaging, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with NestedPaging disabled."));
2055 AssertMsg(!pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTaggedTlbEpt cannot be invoked with VPID enabled."));
2056
2057 /*
2058 * Force a TLB flush for the first world-switch if the current CPU differs from the one we ran on last.
2059 * A change in the TLB flush count implies the host CPU is online after a suspend/resume.
2060 */
2061 if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
2062 || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
2063 {
2064 pVCpu->hm.s.fForceTLBFlush = true;
2065 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
2066 }
2067
2068 /* Check for explicit TLB flushes. */
2069 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
2070 {
2071 pVCpu->hm.s.fForceTLBFlush = true;
2072 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2073 }
2074
2075 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
2076 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
2077
2078 if (pVCpu->hm.s.fForceTLBFlush)
2079 {
2080 hmR0VmxFlushEpt(pVCpu, pVM->hm.s.vmx.enmFlushEpt);
2081 pVCpu->hm.s.fForceTLBFlush = false;
2082 }
2083}
2084
2085
2086/**
2087 * Flushes the tagged-TLB entries for VPID CPUs as necessary.
2088 *
2089 * @returns VBox status code.
2090 * @param pVM The cross context VM structure.
2091 * @param pVCpu The cross context virtual CPU structure.
2092 * @param pCpu Pointer to the global HM CPU struct.
2093 *
2094 * @remarks Called with interrupts disabled.
2095 */
2096static void hmR0VmxFlushTaggedTlbVpid(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2097{
2098 AssertPtr(pVM);
2099 AssertPtr(pVCpu);
2100 AssertPtr(pCpu);
2101 Assert(pCpu->idCpu != NIL_RTCPUID);
2102 AssertMsg(pVM->hm.s.vmx.fVpid, ("hmR0VmxFlushTlbVpid cannot be invoked with VPID disabled."));
2103 AssertMsg(!pVM->hm.s.fNestedPaging, ("hmR0VmxFlushTlbVpid cannot be invoked with NestedPaging enabled"));
2104
2105 /*
2106 * Force a TLB flush for the first world switch if the current CPU differs from the one we ran on last.
2107 * If the TLB flush count changed, another VM (VCPU rather) has hit the ASID limit while flushing the TLB
2108 * or the host CPU is online after a suspend/resume, so we cannot reuse the current ASID anymore.
2109 */
2110 if ( pVCpu->hm.s.idLastCpu != pCpu->idCpu
2111 || pVCpu->hm.s.cTlbFlushes != pCpu->cTlbFlushes)
2112 {
2113 pVCpu->hm.s.fForceTLBFlush = true;
2114 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlbWorldSwitch);
2115 }
2116
2117 /* Check for explicit TLB flushes. */
2118 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_TLB_FLUSH))
2119 {
2120 /*
2121 * If we ever support VPID flush combinations other than ALL or SINGLE-context (see hmR0VmxSetupTaggedTlb())
2122 * we would need to explicitly flush in this case (add an fExplicitFlush = true here and change the
2123 * pCpu->fFlushAsidBeforeUse check below to include fExplicitFlush's too) - an obscure corner case.
2124 */
2125 pVCpu->hm.s.fForceTLBFlush = true;
2126 STAM_COUNTER_INC(&pVCpu->hm.s.StatFlushTlb);
2127 }
2128
2129 pVCpu->hm.s.idLastCpu = pCpu->idCpu;
2130 if (pVCpu->hm.s.fForceTLBFlush)
2131 {
2132 ++pCpu->uCurrentAsid;
2133 if (pCpu->uCurrentAsid >= pVM->hm.s.uMaxAsid)
2134 {
2135 pCpu->uCurrentAsid = 1; /* Wraparound to 1; host uses 0 */
2136 pCpu->cTlbFlushes++; /* All VCPUs that run on this host CPU must use a new VPID. */
2137 pCpu->fFlushAsidBeforeUse = true; /* All VCPUs that run on this host CPU must flush their new VPID before use. */
2138 }
2139
2140 pVCpu->hm.s.fForceTLBFlush = false;
2141 pVCpu->hm.s.cTlbFlushes = pCpu->cTlbFlushes;
2142 pVCpu->hm.s.uCurrentAsid = pCpu->uCurrentAsid;
2143 if (pCpu->fFlushAsidBeforeUse)
2144 {
2145 if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_SINGLE_CONTEXT)
2146 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_SINGLE_CONTEXT, 0 /* GCPtr */);
2147 else if (pVM->hm.s.vmx.enmFlushVpid == VMXFLUSHVPID_ALL_CONTEXTS)
2148 {
2149 hmR0VmxFlushVpid(pVM, pVCpu, VMXFLUSHVPID_ALL_CONTEXTS, 0 /* GCPtr */);
2150 pCpu->fFlushAsidBeforeUse = false;
2151 }
2152 else
2153 {
2154 /* hmR0VmxSetupTaggedTlb() ensures we never get here. Paranoia. */
2155 AssertMsgFailed(("Unsupported VPID-flush context type.\n"));
2156 }
2157 }
2158 }
2159
2160 AssertMsg(pVCpu->hm.s.cTlbFlushes == pCpu->cTlbFlushes,
2161 ("Flush count mismatch for cpu %d (%u vs %u)\n", pCpu->idCpu, pVCpu->hm.s.cTlbFlushes, pCpu->cTlbFlushes));
2162 AssertMsg(pCpu->uCurrentAsid >= 1 && pCpu->uCurrentAsid < pVM->hm.s.uMaxAsid,
2163 ("Cpu[%u] uCurrentAsid=%u cTlbFlushes=%u pVCpu->idLastCpu=%u pVCpu->cTlbFlushes=%u\n", pCpu->idCpu,
2164 pCpu->uCurrentAsid, pCpu->cTlbFlushes, pVCpu->hm.s.idLastCpu, pVCpu->hm.s.cTlbFlushes));
2165 AssertMsg(pVCpu->hm.s.uCurrentAsid >= 1 && pVCpu->hm.s.uCurrentAsid < pVM->hm.s.uMaxAsid,
2166 ("Cpu[%u] pVCpu->uCurrentAsid=%u\n", pCpu->idCpu, pVCpu->hm.s.uCurrentAsid));
2167
2168 int rc = VMXWriteVmcs32(VMX_VMCS16_VPID, pVCpu->hm.s.uCurrentAsid);
2169 AssertRC(rc);
2170}
2171
2172
2173/**
2174 * Flushes the guest TLB entry based on CPU capabilities.
2175 *
2176 * @param pVCpu The cross context virtual CPU structure.
2177 * @param pCpu Pointer to the global HM CPU struct.
2178 */
2179DECLINLINE(void) hmR0VmxFlushTaggedTlb(PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
2180{
2181#ifdef HMVMX_ALWAYS_FLUSH_TLB
2182 VMCPU_FF_SET(pVCpu, VMCPU_FF_TLB_FLUSH);
2183#endif
2184 PVM pVM = pVCpu->CTX_SUFF(pVM);
2185 switch (pVM->hm.s.vmx.uFlushTaggedTlb)
2186 {
2187 case HMVMX_FLUSH_TAGGED_TLB_EPT_VPID: hmR0VmxFlushTaggedTlbBoth(pVM, pVCpu, pCpu); break;
2188 case HMVMX_FLUSH_TAGGED_TLB_EPT: hmR0VmxFlushTaggedTlbEpt(pVM, pVCpu, pCpu); break;
2189 case HMVMX_FLUSH_TAGGED_TLB_VPID: hmR0VmxFlushTaggedTlbVpid(pVM, pVCpu, pCpu); break;
2190 case HMVMX_FLUSH_TAGGED_TLB_NONE: hmR0VmxFlushTaggedTlbNone(pVM, pVCpu, pCpu); break;
2191 default:
2192 AssertMsgFailed(("Invalid flush-tag function identifier\n"));
2193 break;
2194 }
2195
2196 /* Don't assert that VMCPU_FF_TLB_FLUSH should no longer be pending. It can be set by other EMTs. */
2197}
2198
2199
2200/**
2201 * Sets up the appropriate tagged TLB-flush level and handler for flushing guest
2202 * TLB entries from the host TLB before VM-entry.
2203 *
2204 * @returns VBox status code.
2205 * @param pVM The cross context VM structure.
2206 */
2207static int hmR0VmxSetupTaggedTlb(PVM pVM)
2208{
2209 /*
2210 * Determine optimal flush type for Nested Paging.
2211 * We cannot ignore EPT if no suitable flush-types is supported by the CPU as we've already setup unrestricted
2212 * guest execution (see hmR3InitFinalizeR0()).
2213 */
2214 if (pVM->hm.s.fNestedPaging)
2215 {
2216 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT)
2217 {
2218 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT)
2219 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_SINGLE_CONTEXT;
2220 else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS)
2221 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_ALL_CONTEXTS;
2222 else
2223 {
2224 /* Shouldn't happen. EPT is supported but no suitable flush-types supported. */
2225 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2226 pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_FLUSH_TYPE_UNSUPPORTED;
2227 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2228 }
2229
2230 /* Make sure the write-back cacheable memory type for EPT is supported. */
2231 if (RT_UNLIKELY(!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB)))
2232 {
2233 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2234 pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_MEM_TYPE_NOT_WB;
2235 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2236 }
2237
2238 /* EPT requires a page-walk length of 4. */
2239 if (RT_UNLIKELY(!(pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_PAGE_WALK_LENGTH_4)))
2240 {
2241 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2242 pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_PAGE_WALK_LENGTH_UNSUPPORTED;
2243 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2244 }
2245 }
2246 else
2247 {
2248 /* Shouldn't happen. EPT is supported but INVEPT instruction is not supported. */
2249 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NOT_SUPPORTED;
2250 pVM->aCpus[0].hm.s.u32HMError = VMX_UFC_EPT_INVEPT_UNAVAILABLE;
2251 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2252 }
2253 }
2254
2255 /*
2256 * Determine optimal flush type for VPID.
2257 */
2258 if (pVM->hm.s.vmx.fVpid)
2259 {
2260 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID)
2261 {
2262 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT)
2263 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_SINGLE_CONTEXT;
2264 else if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS)
2265 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_ALL_CONTEXTS;
2266 else
2267 {
2268 /* Neither SINGLE nor ALL-context flush types for VPID is supported by the CPU. Ignore VPID capability. */
2269 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR)
2270 LogRel(("hmR0VmxSetupTaggedTlb: Only INDIV_ADDR supported. Ignoring VPID.\n"));
2271 if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
2272 LogRel(("hmR0VmxSetupTaggedTlb: Only SINGLE_CONTEXT_RETAIN_GLOBALS supported. Ignoring VPID.\n"));
2273 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NOT_SUPPORTED;
2274 pVM->hm.s.vmx.fVpid = false;
2275 }
2276 }
2277 else
2278 {
2279 /* Shouldn't happen. VPID is supported but INVVPID is not supported by the CPU. Ignore VPID capability. */
2280 Log4(("hmR0VmxSetupTaggedTlb: VPID supported without INVEPT support. Ignoring VPID.\n"));
2281 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NOT_SUPPORTED;
2282 pVM->hm.s.vmx.fVpid = false;
2283 }
2284 }
2285
2286 /*
2287 * Setup the handler for flushing tagged-TLBs.
2288 */
2289 if (pVM->hm.s.fNestedPaging && pVM->hm.s.vmx.fVpid)
2290 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_EPT_VPID;
2291 else if (pVM->hm.s.fNestedPaging)
2292 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_EPT;
2293 else if (pVM->hm.s.vmx.fVpid)
2294 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_VPID;
2295 else
2296 pVM->hm.s.vmx.uFlushTaggedTlb = HMVMX_FLUSH_TAGGED_TLB_NONE;
2297 return VINF_SUCCESS;
2298}
2299
2300
2301/**
2302 * Sets up pin-based VM-execution controls in the VMCS.
2303 *
2304 * @returns VBox status code.
2305 * @param pVM The cross context VM structure.
2306 * @param pVCpu The cross context virtual CPU structure.
2307 */
2308static int hmR0VmxSetupPinCtls(PVM pVM, PVMCPU pVCpu)
2309{
2310 AssertPtr(pVM);
2311 AssertPtr(pVCpu);
2312
2313 uint32_t val = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0; /* Bits set here must always be set. */
2314 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1; /* Bits cleared here must always be cleared. */
2315
2316 val |= VMX_VMCS_CTRL_PIN_EXEC_EXT_INT_EXIT /* External interrupts cause a VM-exit. */
2317 | VMX_VMCS_CTRL_PIN_EXEC_NMI_EXIT; /* Non-maskable interrupts (NMIs) cause a VM-exit. */
2318
2319 if (pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
2320 val |= VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI; /* Use virtual NMIs and virtual-NMI blocking features. */
2321
2322 /* Enable the VMX preemption timer. */
2323 if (pVM->hm.s.vmx.fUsePreemptTimer)
2324 {
2325 Assert(pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER);
2326 val |= VMX_VMCS_CTRL_PIN_EXEC_PREEMPT_TIMER;
2327 }
2328
2329#if 0
2330 /* Enable posted-interrupt processing. */
2331 if (pVM->hm.s.fPostedIntrs)
2332 {
2333 Assert(pVM->hm.s.vmx.Msrs.VmxPinCtls.n.allowed1 & VMX_VMCS_CTRL_PIN_EXEC_POSTED_INTR);
2334 Assert(pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT);
2335 val |= VMX_VMCS_CTRL_PIN_EXEC_POSTED_INTR;
2336 }
2337#endif
2338
2339 if ((val & zap) != val)
2340 {
2341 LogRel(("hmR0VmxSetupPinCtls: Invalid pin-based VM-execution controls combo! cpu=%#RX64 val=%#RX64 zap=%#RX64\n",
2342 pVM->hm.s.vmx.Msrs.VmxPinCtls.n.disallowed0, val, zap));
2343 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PIN_EXEC;
2344 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2345 }
2346
2347 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, val);
2348 AssertRCReturn(rc, rc);
2349
2350 pVCpu->hm.s.vmx.u32PinCtls = val;
2351 return rc;
2352}
2353
2354
2355/**
2356 * Sets up processor-based VM-execution controls in the VMCS.
2357 *
2358 * @returns VBox status code.
2359 * @param pVM The cross context VM structure.
2360 * @param pVCpu The cross context virtual CPU structure.
2361 */
2362static int hmR0VmxSetupProcCtls(PVM pVM, PVMCPU pVCpu)
2363{
2364 AssertPtr(pVM);
2365 AssertPtr(pVCpu);
2366
2367 int rc = VERR_INTERNAL_ERROR_5;
2368 uint32_t val = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0; /* Bits set here must be set in the VMCS. */
2369 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
2370
2371 val |= VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT /* HLT causes a VM-exit. */
2372 | VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING /* Use TSC-offsetting. */
2373 | VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT /* MOV DRx causes a VM-exit. */
2374 | VMX_VMCS_CTRL_PROC_EXEC_UNCOND_IO_EXIT /* All IO instructions cause a VM-exit. */
2375 | VMX_VMCS_CTRL_PROC_EXEC_RDPMC_EXIT /* RDPMC causes a VM-exit. */
2376 | VMX_VMCS_CTRL_PROC_EXEC_MONITOR_EXIT /* MONITOR causes a VM-exit. */
2377 | VMX_VMCS_CTRL_PROC_EXEC_MWAIT_EXIT; /* MWAIT causes a VM-exit. */
2378
2379 /* We toggle VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT later, check if it's not -always- needed to be set or clear. */
2380 if ( !(pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT)
2381 || (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0 & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT))
2382 {
2383 LogRel(("hmR0VmxSetupProcCtls: Unsupported VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT combo!"));
2384 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_MOV_DRX_EXIT;
2385 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2386 }
2387
2388 /* Without Nested Paging, INVLPG (also affects INVPCID) and MOV CR3 instructions should cause VM-exits. */
2389 if (!pVM->hm.s.fNestedPaging)
2390 {
2391 Assert(!pVM->hm.s.vmx.fUnrestrictedGuest); /* Paranoia. */
2392 val |= VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT
2393 | VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
2394 | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
2395 }
2396
2397 /* Use TPR shadowing if supported by the CPU. */
2398 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
2399 {
2400 Assert(pVCpu->hm.s.vmx.HCPhysVirtApic);
2401 Assert(!(pVCpu->hm.s.vmx.HCPhysVirtApic & 0xfff)); /* Bits 11:0 MBZ. */
2402 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, 0);
2403 rc |= VMXWriteVmcs64(VMX_VMCS64_CTRL_VAPIC_PAGEADDR_FULL, pVCpu->hm.s.vmx.HCPhysVirtApic);
2404 AssertRCReturn(rc, rc);
2405
2406 val |= VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW; /* CR8 reads from the Virtual-APIC page. */
2407 /* CR8 writes cause a VM-exit based on TPR threshold. */
2408 Assert(!(val & VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT));
2409 Assert(!(val & VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT));
2410 }
2411 else
2412 {
2413 /*
2414 * Some 32-bit CPUs do not support CR8 load/store exiting as MOV CR8 is invalid on 32-bit Intel CPUs.
2415 * Set this control only for 64-bit guests.
2416 */
2417 if (pVM->hm.s.fAllow64BitGuests)
2418 {
2419 val |= VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT /* CR8 reads cause a VM-exit. */
2420 | VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT; /* CR8 writes cause a VM-exit. */
2421 }
2422 }
2423
2424 /* Use MSR-bitmaps if supported by the CPU. */
2425 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
2426 {
2427 val |= VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS;
2428
2429 Assert(pVCpu->hm.s.vmx.HCPhysMsrBitmap);
2430 Assert(!(pVCpu->hm.s.vmx.HCPhysMsrBitmap & 0xfff)); /* Bits 11:0 MBZ. */
2431 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_MSR_BITMAP_FULL, pVCpu->hm.s.vmx.HCPhysMsrBitmap);
2432 AssertRCReturn(rc, rc);
2433
2434 /*
2435 * The guest can access the following MSRs (read, write) without causing VM-exits; they are loaded/stored
2436 * automatically using dedicated fields in the VMCS.
2437 */
2438 hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_CS, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2439 hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_ESP, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2440 hmR0VmxSetMsrPermission(pVCpu, MSR_IA32_SYSENTER_EIP, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2441 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_GS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2442 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_FS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2443
2444#if HC_ARCH_BITS == 64
2445 /*
2446 * Set passthru permissions for the following MSRs (mandatory for VT-x) required for 64-bit guests.
2447 */
2448 if (pVM->hm.s.fAllow64BitGuests)
2449 {
2450 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_LSTAR, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2451 hmR0VmxSetMsrPermission(pVCpu, MSR_K6_STAR, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2452 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_SF_MASK, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2453 hmR0VmxSetMsrPermission(pVCpu, MSR_K8_KERNEL_GS_BASE, VMXMSREXIT_PASSTHRU_READ, VMXMSREXIT_PASSTHRU_WRITE);
2454 }
2455#endif
2456 /* Though MSR_IA32_PERF_GLOBAL_CTRL is saved/restored lazily, we want intercept reads/write to it for now. */
2457 }
2458
2459 /* Use the secondary processor-based VM-execution controls if supported by the CPU. */
2460 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
2461 val |= VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL;
2462
2463 if ((val & zap) != val)
2464 {
2465 LogRel(("hmR0VmxSetupProcCtls: Invalid processor-based VM-execution controls combo! cpu=%#RX64 val=%#RX64 zap=%#RX64\n",
2466 pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0, val, zap));
2467 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC;
2468 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2469 }
2470
2471 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, val);
2472 AssertRCReturn(rc, rc);
2473
2474 pVCpu->hm.s.vmx.u32ProcCtls = val;
2475
2476 /*
2477 * Secondary processor-based VM-execution controls.
2478 */
2479 if (RT_LIKELY(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL))
2480 {
2481 val = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0; /* Bits set here must be set in the VMCS. */
2482 zap = pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
2483
2484 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT)
2485 val |= VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT; /* WBINVD causes a VM-exit. */
2486
2487 if (pVM->hm.s.fNestedPaging)
2488 val |= VMX_VMCS_CTRL_PROC_EXEC2_EPT; /* Enable EPT. */
2489 else
2490 {
2491 /*
2492 * Without Nested Paging, INVPCID should cause a VM-exit. Enabling this bit causes the CPU to refer to
2493 * VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT when INVPCID is executed by the guest.
2494 * See Intel spec. 25.4 "Changes to instruction behaviour in VMX non-root operation".
2495 */
2496 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_INVPCID)
2497 val |= VMX_VMCS_CTRL_PROC_EXEC2_INVPCID;
2498 }
2499
2500 if (pVM->hm.s.vmx.fVpid)
2501 val |= VMX_VMCS_CTRL_PROC_EXEC2_VPID; /* Enable VPID. */
2502
2503 if (pVM->hm.s.vmx.fUnrestrictedGuest)
2504 val |= VMX_VMCS_CTRL_PROC_EXEC2_UNRESTRICTED_GUEST; /* Enable Unrestricted Execution. */
2505
2506#if 0
2507 if (pVM->hm.s.fVirtApicRegs)
2508 {
2509 Assert(pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_APIC_REG_VIRT);
2510 val |= VMX_VMCS_CTRL_PROC_EXEC2_APIC_REG_VIRT; /* Enable APIC-register virtualization. */
2511
2512 Assert(pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_INTR_DELIVERY);
2513 val |= VMX_VMCS_CTRL_PROC_EXEC2_VIRT_INTR_DELIVERY; /* Enable virtual-interrupt delivery. */
2514 }
2515#endif
2516
2517 /* Enable Virtual-APIC page accesses if supported by the CPU. This is essentially where the TPR shadow resides. */
2518 /** @todo VIRT_X2APIC support, it's mutually exclusive with this. So must be
2519 * done dynamically. */
2520 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)
2521 {
2522 Assert(pVM->hm.s.vmx.HCPhysApicAccess);
2523 Assert(!(pVM->hm.s.vmx.HCPhysApicAccess & 0xfff)); /* Bits 11:0 MBZ. */
2524 val |= VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC; /* Virtualize APIC accesses. */
2525 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL, pVM->hm.s.vmx.HCPhysApicAccess);
2526 AssertRCReturn(rc, rc);
2527 }
2528
2529 if (pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP)
2530 val |= VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP; /* Enable RDTSCP support. */
2531
2532 if ( pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC2_PAUSE_LOOP_EXIT
2533 && pVM->hm.s.vmx.cPleGapTicks
2534 && pVM->hm.s.vmx.cPleWindowTicks)
2535 {
2536 val |= VMX_VMCS_CTRL_PROC_EXEC2_PAUSE_LOOP_EXIT; /* Enable pause-loop exiting. */
2537
2538 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_GAP, pVM->hm.s.vmx.cPleGapTicks);
2539 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_PLE_WINDOW, pVM->hm.s.vmx.cPleWindowTicks);
2540 AssertRCReturn(rc, rc);
2541 }
2542
2543 if ((val & zap) != val)
2544 {
2545 LogRel(("hmR0VmxSetupProcCtls: Invalid secondary processor-based VM-execution controls combo! "
2546 "cpu=%#RX64 val=%#RX64 zap=%#RX64\n", pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.disallowed0, val, zap));
2547 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_PROC_EXEC2;
2548 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2549 }
2550
2551 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, val);
2552 AssertRCReturn(rc, rc);
2553
2554 pVCpu->hm.s.vmx.u32ProcCtls2 = val;
2555 }
2556 else if (RT_UNLIKELY(pVM->hm.s.vmx.fUnrestrictedGuest))
2557 {
2558 LogRel(("hmR0VmxSetupProcCtls: Unrestricted Guest set as true when secondary processor-based VM-execution controls not "
2559 "available\n"));
2560 pVCpu->hm.s.u32HMError = VMX_UFC_INVALID_UX_COMBO;
2561 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
2562 }
2563
2564 return VINF_SUCCESS;
2565}
2566
2567
2568/**
2569 * Sets up miscellaneous (everything other than Pin & Processor-based
2570 * VM-execution) control fields in the VMCS.
2571 *
2572 * @returns VBox status code.
2573 * @param pVM The cross context VM structure.
2574 * @param pVCpu The cross context virtual CPU structure.
2575 */
2576static int hmR0VmxSetupMiscCtls(PVM pVM, PVMCPU pVCpu)
2577{
2578 NOREF(pVM);
2579 AssertPtr(pVM);
2580 AssertPtr(pVCpu);
2581
2582 int rc = VERR_GENERAL_FAILURE;
2583
2584 /* All fields are zero-initialized during allocation; but don't remove the commented block below. */
2585#if 0
2586 /* All CR3 accesses cause VM-exits. Later we optimize CR3 accesses (see hmR0VmxLoadGuestCR3AndCR4())*/
2587 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, 0);
2588 rc |= VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, 0);
2589
2590 /*
2591 * Set MASK & MATCH to 0. VMX checks if GuestPFErrCode & MASK == MATCH. If equal (in our case it always is)
2592 * 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.
2593 * We thus use the exception bitmap to control it rather than use both.
2594 */
2595 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, 0);
2596 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, 0);
2597
2598 /** @todo Explore possibility of using IO-bitmaps. */
2599 /* All IO & IOIO instructions cause VM-exits. */
2600 rc |= VMXWriteVmcs64(VMX_VMCS64_CTRL_IO_BITMAP_A_FULL, 0);
2601 rc |= VMXWriteVmcs64(VMX_VMCS64_CTRL_IO_BITMAP_B_FULL, 0);
2602
2603 /* Initialize the MSR-bitmap area. */
2604 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, 0);
2605 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, 0);
2606 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, 0);
2607 AssertRCReturn(rc, rc);
2608#endif
2609
2610 /* Setup MSR auto-load/store area. */
2611 Assert(pVCpu->hm.s.vmx.HCPhysGuestMsr);
2612 Assert(!(pVCpu->hm.s.vmx.HCPhysGuestMsr & 0xf)); /* Lower 4 bits MBZ. */
2613 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL, pVCpu->hm.s.vmx.HCPhysGuestMsr);
2614 rc |= VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL, pVCpu->hm.s.vmx.HCPhysGuestMsr);
2615 AssertRCReturn(rc, rc);
2616
2617 Assert(pVCpu->hm.s.vmx.HCPhysHostMsr);
2618 Assert(!(pVCpu->hm.s.vmx.HCPhysHostMsr & 0xf)); /* Lower 4 bits MBZ. */
2619 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL, pVCpu->hm.s.vmx.HCPhysHostMsr);
2620 AssertRCReturn(rc, rc);
2621
2622 /* Set VMCS link pointer. Reserved for future use, must be -1. Intel spec. 24.4 "Guest-State Area". */
2623 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, UINT64_C(0xffffffffffffffff));
2624 AssertRCReturn(rc, rc);
2625
2626 /* All fields are zero-initialized during allocation; but don't remove the commented block below. */
2627#if 0
2628 /* Setup debug controls */
2629 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, 0); /** @todo We don't support IA32_DEBUGCTL MSR. Should we? */
2630 rc |= VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, 0);
2631 AssertRCReturn(rc, rc);
2632#endif
2633
2634 return rc;
2635}
2636
2637
2638/**
2639 * Sets up the initial exception bitmap in the VMCS based on static conditions.
2640 *
2641 * @returns VBox status code.
2642 * @param pVM The cross context VM structure.
2643 * @param pVCpu The cross context virtual CPU structure.
2644 */
2645static int hmR0VmxInitXcptBitmap(PVM pVM, PVMCPU pVCpu)
2646{
2647 AssertPtr(pVM);
2648 AssertPtr(pVCpu);
2649
2650 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
2651
2652 /** @todo r=ramshankar: Shouldn't setting up \#UD intercepts be handled by
2653 * hmR0VmxLoadGuestXcptIntercepts()? Why do we check it here? */
2654 uint32_t u32XcptBitmap = pVCpu->hm.s.fGIMTrapXcptUD ? RT_BIT(X86_XCPT_UD) : 0;
2655
2656 /* Must always intercept #AC to prevent the guest from hanging the CPU. */
2657 u32XcptBitmap |= RT_BIT_32(X86_XCPT_AC);
2658
2659 /* Because we need to maintain the DR6 state even when intercepting DRx reads
2660 and writes, and because recursive #DBs can cause the CPU hang, we must always
2661 intercept #DB. */
2662 u32XcptBitmap |= RT_BIT_32(X86_XCPT_DB);
2663
2664 /* Without Nested Paging, #PF must cause a VM-exit so we can sync our shadow page tables. */
2665 if (!pVM->hm.s.fNestedPaging)
2666 u32XcptBitmap |= RT_BIT(X86_XCPT_PF);
2667
2668 pVCpu->hm.s.vmx.u32XcptBitmap = u32XcptBitmap;
2669 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, u32XcptBitmap);
2670 AssertRCReturn(rc, rc);
2671 return rc;
2672}
2673
2674
2675/**
2676 * Sets up the initial guest-state mask. The guest-state mask is consulted
2677 * before reading guest-state fields from the VMCS as VMREADs can be expensive
2678 * for the nested virtualization case (as it would cause a VM-exit).
2679 *
2680 * @param pVCpu The cross context virtual CPU structure.
2681 */
2682static int hmR0VmxInitUpdatedGuestStateMask(PVMCPU pVCpu)
2683{
2684 /* Initially the guest-state is up-to-date as there is nothing in the VMCS. */
2685 HMVMXCPU_GST_RESET_TO(pVCpu, HMVMX_UPDATED_GUEST_ALL);
2686 return VINF_SUCCESS;
2687}
2688
2689
2690/**
2691 * Does per-VM VT-x initialization.
2692 *
2693 * @returns VBox status code.
2694 * @param pVM The cross context VM structure.
2695 */
2696VMMR0DECL(int) VMXR0InitVM(PVM pVM)
2697{
2698 LogFlowFunc(("pVM=%p\n", pVM));
2699
2700 int rc = hmR0VmxStructsAlloc(pVM);
2701 if (RT_FAILURE(rc))
2702 {
2703 LogRel(("VMXR0InitVM: hmR0VmxStructsAlloc failed! rc=%Rrc\n", rc));
2704 return rc;
2705 }
2706
2707 return VINF_SUCCESS;
2708}
2709
2710
2711/**
2712 * Does per-VM VT-x termination.
2713 *
2714 * @returns VBox status code.
2715 * @param pVM The cross context VM structure.
2716 */
2717VMMR0DECL(int) VMXR0TermVM(PVM pVM)
2718{
2719 LogFlowFunc(("pVM=%p\n", pVM));
2720
2721#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2722 if (pVM->hm.s.vmx.hMemObjScratch != NIL_RTR0MEMOBJ)
2723 ASMMemZero32(pVM->hm.s.vmx.pvScratch, PAGE_SIZE);
2724#endif
2725 hmR0VmxStructsFree(pVM);
2726 return VINF_SUCCESS;
2727}
2728
2729
2730/**
2731 * Sets up the VM for execution under VT-x.
2732 * This function is only called once per-VM during initialization.
2733 *
2734 * @returns VBox status code.
2735 * @param pVM The cross context VM structure.
2736 */
2737VMMR0DECL(int) VMXR0SetupVM(PVM pVM)
2738{
2739 AssertPtrReturn(pVM, VERR_INVALID_PARAMETER);
2740 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
2741
2742 LogFlowFunc(("pVM=%p\n", pVM));
2743
2744 /*
2745 * Without UnrestrictedGuest, pRealModeTSS and pNonPagingModeEPTPageTable *must* always be allocated.
2746 * We no longer support the highly unlikely case of UnrestrictedGuest without pRealModeTSS. See hmR3InitFinalizeR0Intel().
2747 */
2748 if ( !pVM->hm.s.vmx.fUnrestrictedGuest
2749 && ( !pVM->hm.s.vmx.pNonPagingModeEPTPageTable
2750 || !pVM->hm.s.vmx.pRealModeTSS))
2751 {
2752 LogRel(("VMXR0SetupVM: Invalid real-on-v86 state.\n"));
2753 return VERR_INTERNAL_ERROR;
2754 }
2755
2756 /* Initialize these always, see hmR3InitFinalizeR0().*/
2757 pVM->hm.s.vmx.enmFlushEpt = VMXFLUSHEPT_NONE;
2758 pVM->hm.s.vmx.enmFlushVpid = VMXFLUSHVPID_NONE;
2759
2760 /* Setup the tagged-TLB flush handlers. */
2761 int rc = hmR0VmxSetupTaggedTlb(pVM);
2762 if (RT_FAILURE(rc))
2763 {
2764 LogRel(("VMXR0SetupVM: hmR0VmxSetupTaggedTlb failed! rc=%Rrc\n", rc));
2765 return rc;
2766 }
2767
2768 /* Check if we can use the VMCS controls for swapping the EFER MSR. */
2769 Assert(!pVM->hm.s.vmx.fSupportsVmcsEfer);
2770#if HC_ARCH_BITS == 64
2771 if ( (pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1 & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR)
2772 && (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR)
2773 && (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR))
2774 {
2775 pVM->hm.s.vmx.fSupportsVmcsEfer = true;
2776 }
2777#endif
2778
2779 /* At least verify VMX is enabled, since we can't check if we're in VMX root mode without #GP'ing. */
2780 RTCCUINTREG uHostCR4 = ASMGetCR4();
2781 if (RT_UNLIKELY(!(uHostCR4 & X86_CR4_VMXE)))
2782 return VERR_VMX_NOT_IN_VMX_ROOT_MODE;
2783
2784 for (VMCPUID i = 0; i < pVM->cCpus; i++)
2785 {
2786 PVMCPU pVCpu = &pVM->aCpus[i];
2787 AssertPtr(pVCpu);
2788 AssertPtr(pVCpu->hm.s.vmx.pvVmcs);
2789
2790 /* Log the VCPU pointers, useful for debugging SMP VMs. */
2791 Log4(("VMXR0SetupVM: pVCpu=%p idCpu=%RU32\n", pVCpu, pVCpu->idCpu));
2792
2793 /* Initialize the VM-exit history array with end-of-array markers (UINT16_MAX). */
2794 Assert(!pVCpu->hm.s.idxExitHistoryFree);
2795 HMCPU_EXIT_HISTORY_RESET(pVCpu);
2796
2797 /* Set revision dword at the beginning of the VMCS structure. */
2798 *(uint32_t *)pVCpu->hm.s.vmx.pvVmcs = MSR_IA32_VMX_BASIC_INFO_VMCS_ID(pVM->hm.s.vmx.Msrs.u64BasicInfo);
2799
2800 /* Initialize our VMCS region in memory, set the VMCS launch state to "clear". */
2801 rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2802 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXClearVmcs failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2803 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2804
2805 /* Load this VMCS as the current VMCS. */
2806 rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2807 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXActivateVmcs failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2808 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2809
2810 rc = hmR0VmxSetupPinCtls(pVM, pVCpu);
2811 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupPinCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2812 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2813
2814 rc = hmR0VmxSetupProcCtls(pVM, pVCpu);
2815 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupProcCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2816 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2817
2818 rc = hmR0VmxSetupMiscCtls(pVM, pVCpu);
2819 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxSetupMiscCtls failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2820 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2821
2822 rc = hmR0VmxInitXcptBitmap(pVM, pVCpu);
2823 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitXcptBitmap failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2824 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2825
2826 rc = hmR0VmxInitUpdatedGuestStateMask(pVCpu);
2827 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitUpdatedGuestStateMask failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2828 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2829
2830#if HC_ARCH_BITS == 32
2831 rc = hmR0VmxInitVmcsReadCache(pVM, pVCpu);
2832 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: hmR0VmxInitVmcsReadCache failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2833 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2834#endif
2835
2836 /* Re-sync the CPU's internal data into our VMCS memory region & reset the launch state to "clear". */
2837 rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
2838 AssertLogRelMsgRCReturnStmt(rc, ("VMXR0SetupVM: VMXClearVmcs(2) failed! rc=%Rrc (pVM=%p)\n", rc, pVM),
2839 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc), rc);
2840
2841 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
2842
2843 hmR0VmxUpdateErrorRecord(pVM, pVCpu, rc);
2844 }
2845
2846 return VINF_SUCCESS;
2847}
2848
2849
2850/**
2851 * Saves the host control registers (CR0, CR3, CR4) into the host-state area in
2852 * the VMCS.
2853 *
2854 * @returns VBox status code.
2855 * @param pVM The cross context VM structure.
2856 * @param pVCpu The cross context virtual CPU structure.
2857 */
2858DECLINLINE(int) hmR0VmxSaveHostControlRegs(PVM pVM, PVMCPU pVCpu)
2859{
2860 NOREF(pVM); NOREF(pVCpu);
2861
2862 RTCCUINTREG uReg = ASMGetCR0();
2863 int rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR0, uReg);
2864 AssertRCReturn(rc, rc);
2865
2866 uReg = ASMGetCR3();
2867 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR3, uReg);
2868 AssertRCReturn(rc, rc);
2869
2870 uReg = ASMGetCR4();
2871 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_CR4, uReg);
2872 AssertRCReturn(rc, rc);
2873 return rc;
2874}
2875
2876
2877#if HC_ARCH_BITS == 64
2878/**
2879 * Macro for adjusting host segment selectors to satisfy VT-x's VM-entry
2880 * requirements. See hmR0VmxSaveHostSegmentRegs().
2881 */
2882# define VMXLOCAL_ADJUST_HOST_SEG(seg, selValue) \
2883 if ((selValue) & (X86_SEL_RPL | X86_SEL_LDT)) \
2884 { \
2885 bool fValidSelector = true; \
2886 if ((selValue) & X86_SEL_LDT) \
2887 { \
2888 uint32_t uAttr = ASMGetSegAttr((selValue)); \
2889 fValidSelector = RT_BOOL(uAttr != UINT32_MAX && (uAttr & X86_DESC_P)); \
2890 } \
2891 if (fValidSelector) \
2892 { \
2893 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_SEL_##seg; \
2894 pVCpu->hm.s.vmx.RestoreHost.uHostSel##seg = (selValue); \
2895 } \
2896 (selValue) = 0; \
2897 }
2898#endif
2899
2900
2901/**
2902 * Saves the host segment registers and GDTR, IDTR, (TR, GS and FS bases) into
2903 * the host-state area in the VMCS.
2904 *
2905 * @returns VBox status code.
2906 * @param pVM The cross context VM structure.
2907 * @param pVCpu The cross context virtual CPU structure.
2908 */
2909DECLINLINE(int) hmR0VmxSaveHostSegmentRegs(PVM pVM, PVMCPU pVCpu)
2910{
2911 int rc = VERR_INTERNAL_ERROR_5;
2912
2913#if HC_ARCH_BITS == 64
2914 /*
2915 * If we've executed guest code using VT-x, the host-state bits will be messed up. We
2916 * should -not- save the messed up state without restoring the original host-state. See @bugref{7240}.
2917 *
2918 * This apparently can happen (most likely the FPU changes), deal with it rather than asserting.
2919 * Was observed booting Solaris10u10 32-bit guest.
2920 */
2921 if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
2922 && (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
2923 {
2924 Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags,
2925 pVCpu->idCpu));
2926 VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
2927 }
2928 pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
2929#else
2930 RT_NOREF(pVCpu);
2931#endif
2932
2933 /*
2934 * Host DS, ES, FS and GS segment registers.
2935 */
2936#if HC_ARCH_BITS == 64
2937 RTSEL uSelDS = ASMGetDS();
2938 RTSEL uSelES = ASMGetES();
2939 RTSEL uSelFS = ASMGetFS();
2940 RTSEL uSelGS = ASMGetGS();
2941#else
2942 RTSEL uSelDS = 0;
2943 RTSEL uSelES = 0;
2944 RTSEL uSelFS = 0;
2945 RTSEL uSelGS = 0;
2946#endif
2947
2948 /*
2949 * Host CS and SS segment registers.
2950 */
2951 RTSEL uSelCS = ASMGetCS();
2952 RTSEL uSelSS = ASMGetSS();
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_CS_SEL, uSelCS);
2990 rc |= VMXWriteVmcs32(VMX_VMCS16_HOST_SS_SEL, uSelSS);
2991#if HC_ARCH_BITS == 64
2992 rc |= VMXWriteVmcs32(VMX_VMCS16_HOST_DS_SEL, uSelDS);
2993 rc |= VMXWriteVmcs32(VMX_VMCS16_HOST_ES_SEL, uSelES);
2994 rc |= VMXWriteVmcs32(VMX_VMCS16_HOST_FS_SEL, uSelFS);
2995 rc |= VMXWriteVmcs32(VMX_VMCS16_HOST_GS_SEL, uSelGS);
2996#else
2997 NOREF(uSelDS);
2998 NOREF(uSelES);
2999 NOREF(uSelFS);
3000 NOREF(uSelGS);
3001#endif
3002 rc |= VMXWriteVmcs32(VMX_VMCS16_HOST_TR_SEL, uSelTR);
3003 AssertRCReturn(rc, rc);
3004
3005 /*
3006 * Host GDTR and IDTR.
3007 */
3008 RTGDTR Gdtr;
3009 RTIDTR Idtr;
3010 RT_ZERO(Gdtr);
3011 RT_ZERO(Idtr);
3012 ASMGetGDTR(&Gdtr);
3013 ASMGetIDTR(&Idtr);
3014 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, Gdtr.pGdt);
3015 rc |= VMXWriteVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, Idtr.pIdt);
3016 AssertRCReturn(rc, rc);
3017
3018#if HC_ARCH_BITS == 64
3019 /*
3020 * Determine if we need to manually need to restore the GDTR and IDTR limits as VT-x zaps them to the
3021 * maximum limit (0xffff) on every VM-exit.
3022 */
3023 if (Gdtr.cbGdt != 0xffff)
3024 {
3025 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_GDTR;
3026 AssertCompile(sizeof(Gdtr) == sizeof(X86XDTR64));
3027 memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
3028 }
3029
3030 /*
3031 * IDT limit is effectively capped at 0xfff. (See Intel spec. 6.14.1 "64-Bit Mode IDT"
3032 * and Intel spec. 6.2 "Exception and Interrupt Vectors".) Therefore if the host has the limit as 0xfff, VT-x
3033 * bloating the limit to 0xffff shouldn't cause any different CPU behavior. However, several hosts either insists
3034 * on 0xfff being the limit (Windows Patch Guard) or uses the limit for other purposes (darwin puts the CPU ID in there
3035 * but botches sidt alignment in at least one consumer). So, we're only allowing IDTR.LIMIT to be left at 0xffff on
3036 * hosts where we are pretty sure it won't cause trouble.
3037 */
3038# if defined(RT_OS_LINUX) || defined(RT_OS_SOLARIS)
3039 if (Idtr.cbIdt < 0x0fff)
3040# else
3041 if (Idtr.cbIdt != 0xffff)
3042# endif
3043 {
3044 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_IDTR;
3045 AssertCompile(sizeof(Idtr) == sizeof(X86XDTR64));
3046 memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostIdtr, &Idtr, sizeof(X86XDTR64));
3047 }
3048#endif
3049
3050 /*
3051 * Host TR base. Verify that TR selector doesn't point past the GDT. Masking off the TI and RPL bits
3052 * is effectively what the CPU does for "scaling by 8". TI is always 0 and RPL should be too in most cases.
3053 */
3054 AssertMsgReturn((uSelTR | X86_SEL_RPL_LDT) <= Gdtr.cbGdt,
3055 ("hmR0VmxSaveHostSegmentRegs: TR selector exceeds limit. TR=%RTsel cbGdt=%#x\n", uSelTR, Gdtr.cbGdt),
3056 VERR_VMX_INVALID_HOST_STATE);
3057
3058 PCX86DESCHC pDesc = (PCX86DESCHC)(Gdtr.pGdt + (uSelTR & X86_SEL_MASK));
3059#if HC_ARCH_BITS == 64
3060 uintptr_t uTRBase = X86DESC64_BASE(pDesc);
3061
3062 /*
3063 * VT-x unconditionally restores the TR limit to 0x67 and type to 11 (32-bit busy TSS) on all VM-exits.
3064 * The type is the same for 64-bit busy TSS[1]. The limit needs manual restoration if the host has something else.
3065 * Task switching is not supported in 64-bit mode[2], but the limit still matters as IOPM is supported in 64-bit mode.
3066 * Restoring the limit lazily while returning to ring-3 is safe because IOPM is not applicable in ring-0.
3067 *
3068 * [1] See Intel spec. 3.5 "System Descriptor Types".
3069 * [2] See Intel spec. 7.2.3 "TSS Descriptor in 64-bit mode".
3070 */
3071 Assert(pDesc->System.u4Type == 11);
3072 if ( pDesc->System.u16LimitLow != 0x67
3073 || pDesc->System.u4LimitHigh)
3074 {
3075 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_SEL_TR;
3076 /* If the host has made GDT read-only, we would need to temporarily toggle CR0.WP before writing the GDT. */
3077 if (pVM->hm.s.fHostKernelFeatures & SUPKERNELFEATURES_GDT_READ_ONLY)
3078 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_GDT_READ_ONLY;
3079 pVCpu->hm.s.vmx.RestoreHost.uHostSelTR = uSelTR;
3080
3081 /* Store the GDTR here as we need it while restoring TR. */
3082 memcpy(&pVCpu->hm.s.vmx.RestoreHost.HostGdtr, &Gdtr, sizeof(X86XDTR64));
3083 }
3084#else
3085 NOREF(pVM);
3086 uintptr_t uTRBase = X86DESC_BASE(pDesc);
3087#endif
3088 rc = VMXWriteVmcsHstN(VMX_VMCS_HOST_TR_BASE, uTRBase);
3089 AssertRCReturn(rc, rc);
3090
3091 /*
3092 * Host FS base and GS base.
3093 */
3094#if HC_ARCH_BITS == 64
3095 uint64_t u64FSBase = ASMRdMsr(MSR_K8_FS_BASE);
3096 uint64_t u64GSBase = ASMRdMsr(MSR_K8_GS_BASE);
3097 rc = VMXWriteVmcs64(VMX_VMCS_HOST_FS_BASE, u64FSBase);
3098 rc |= VMXWriteVmcs64(VMX_VMCS_HOST_GS_BASE, u64GSBase);
3099 AssertRCReturn(rc, rc);
3100
3101 /* Store the base if we have to restore FS or GS manually as we need to restore the base as well. */
3102 if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_FS)
3103 pVCpu->hm.s.vmx.RestoreHost.uHostFSBase = u64FSBase;
3104 if (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_SEL_GS)
3105 pVCpu->hm.s.vmx.RestoreHost.uHostGSBase = u64GSBase;
3106#endif
3107 return rc;
3108}
3109
3110
3111/**
3112 * Saves certain host MSRs in the VM-exit MSR-load area and some in the
3113 * host-state area of the VMCS. Theses MSRs will be automatically restored on
3114 * the host after every successful VM-exit.
3115 *
3116 * @returns VBox status code.
3117 * @param pVM The cross context VM structure.
3118 * @param pVCpu The cross context virtual CPU structure.
3119 *
3120 * @remarks No-long-jump zone!!!
3121 */
3122DECLINLINE(int) hmR0VmxSaveHostMsrs(PVM pVM, PVMCPU pVCpu)
3123{
3124 NOREF(pVM);
3125
3126 AssertPtr(pVCpu);
3127 AssertPtr(pVCpu->hm.s.vmx.pvHostMsr);
3128
3129 /*
3130 * Save MSRs that we restore lazily (due to preemption or transition to ring-3)
3131 * rather than swapping them on every VM-entry.
3132 */
3133 hmR0VmxLazySaveHostMsrs(pVCpu);
3134
3135 /*
3136 * Host Sysenter MSRs.
3137 */
3138 int rc = VMXWriteVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, ASMRdMsr_Low(MSR_IA32_SYSENTER_CS));
3139#if HC_ARCH_BITS == 32
3140 rc |= VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr_Low(MSR_IA32_SYSENTER_ESP));
3141 rc |= VMXWriteVmcs32(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr_Low(MSR_IA32_SYSENTER_EIP));
3142#else
3143 rc |= VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, ASMRdMsr(MSR_IA32_SYSENTER_ESP));
3144 rc |= VMXWriteVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, ASMRdMsr(MSR_IA32_SYSENTER_EIP));
3145#endif
3146 AssertRCReturn(rc, rc);
3147
3148 /*
3149 * Host EFER MSR.
3150 * If the CPU supports the newer VMCS controls for managing EFER, use it.
3151 * Otherwise it's done as part of auto-load/store MSR area in the VMCS, see hmR0VmxLoadGuestMsrs().
3152 */
3153 if (pVM->hm.s.vmx.fSupportsVmcsEfer)
3154 {
3155 rc = VMXWriteVmcs64(VMX_VMCS64_HOST_EFER_FULL, pVM->hm.s.vmx.u64HostEfer);
3156 AssertRCReturn(rc, rc);
3157 }
3158
3159 /** @todo IA32_PERF_GLOBALCTRL, IA32_PAT also see
3160 * hmR0VmxLoadGuestExitCtls() !! */
3161
3162 return rc;
3163}
3164
3165
3166/**
3167 * Figures out if we need to swap the EFER MSR which is particularly expensive.
3168 *
3169 * We check all relevant bits. For now, that's everything besides LMA/LME, as
3170 * these two bits are handled by VM-entry, see hmR0VmxLoadGuestExitCtls() and
3171 * hmR0VMxLoadGuestEntryCtls().
3172 *
3173 * @returns true if we need to load guest EFER, false otherwise.
3174 * @param pVCpu The cross context virtual CPU structure.
3175 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3176 * out-of-sync. Make sure to update the required fields
3177 * before using them.
3178 *
3179 * @remarks Requires EFER, CR4.
3180 * @remarks No-long-jump zone!!!
3181 */
3182static bool hmR0VmxShouldSwapEferMsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3183{
3184#ifdef HMVMX_ALWAYS_SWAP_EFER
3185 return true;
3186#endif
3187
3188#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
3189 /* For 32-bit hosts running 64-bit guests, we always swap EFER in the world-switcher. Nothing to do here. */
3190 if (CPUMIsGuestInLongMode(pVCpu))
3191 return false;
3192#endif
3193
3194 PVM pVM = pVCpu->CTX_SUFF(pVM);
3195 uint64_t u64HostEfer = pVM->hm.s.vmx.u64HostEfer;
3196 uint64_t u64GuestEfer = pMixedCtx->msrEFER;
3197
3198 /*
3199 * For 64-bit guests, if EFER.SCE bit differs, we need to swap to ensure that the
3200 * guest's SYSCALL behaviour isn't screwed. See @bugref{7386}.
3201 */
3202 if ( CPUMIsGuestInLongMode(pVCpu)
3203 && (u64GuestEfer & MSR_K6_EFER_SCE) != (u64HostEfer & MSR_K6_EFER_SCE))
3204 {
3205 return true;
3206 }
3207
3208 /*
3209 * If the guest uses PAE and EFER.NXE bit differs, we need to swap EFER as it
3210 * affects guest paging. 64-bit paging implies CR4.PAE as well.
3211 * See Intel spec. 4.5 "IA-32e Paging" and Intel spec. 4.1.1 "Three Paging Modes".
3212 */
3213 if ( (pMixedCtx->cr4 & X86_CR4_PAE)
3214 && (pMixedCtx->cr0 & X86_CR0_PG)
3215 && (u64GuestEfer & MSR_K6_EFER_NXE) != (u64HostEfer & MSR_K6_EFER_NXE))
3216 {
3217 /* Assert that host is PAE capable. */
3218 Assert(pVM->hm.s.cpuid.u32AMDFeatureEDX & X86_CPUID_EXT_FEATURE_EDX_NX);
3219 return true;
3220 }
3221
3222 /** @todo Check the latest Intel spec. for any other bits,
3223 * like SMEP/SMAP? */
3224 return false;
3225}
3226
3227
3228/**
3229 * Sets up VM-entry controls in the VMCS. These controls can affect things done
3230 * on VM-exit; e.g. "load debug controls", see Intel spec. 24.8.1 "VM-entry
3231 * controls".
3232 *
3233 * @returns VBox status code.
3234 * @param pVCpu The cross context virtual CPU structure.
3235 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3236 * out-of-sync. Make sure to update the required fields
3237 * before using them.
3238 *
3239 * @remarks Requires EFER.
3240 * @remarks No-long-jump zone!!!
3241 */
3242DECLINLINE(int) hmR0VmxLoadGuestEntryCtls(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3243{
3244 int rc = VINF_SUCCESS;
3245 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_ENTRY_CTLS))
3246 {
3247 PVM pVM = pVCpu->CTX_SUFF(pVM);
3248 uint32_t val = pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0; /* Bits set here must be set in the VMCS. */
3249 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxEntry.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3250
3251 /* Load debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x capable CPUs only supports the 1-setting of this bit. */
3252 val |= VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG;
3253
3254 /* Set if the guest is in long mode. This will set/clear the EFER.LMA bit on VM-entry. */
3255 if (CPUMIsGuestInLongModeEx(pMixedCtx))
3256 {
3257 val |= VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST;
3258 Log4(("Load[%RU32]: VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST\n", pVCpu->idCpu));
3259 }
3260 else
3261 Assert(!(val & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST));
3262
3263 /* If the CPU supports the newer VMCS controls for managing guest/host EFER, use it. */
3264 if ( pVM->hm.s.vmx.fSupportsVmcsEfer
3265 && hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
3266 {
3267 val |= VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR;
3268 Log4(("Load[%RU32]: VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR\n", pVCpu->idCpu));
3269 }
3270
3271 /*
3272 * The following should -not- be set (since we're not in SMM mode):
3273 * - VMX_VMCS_CTRL_ENTRY_ENTRY_SMM
3274 * - VMX_VMCS_CTRL_ENTRY_DEACTIVATE_DUALMON
3275 */
3276
3277 /** @todo VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR,
3278 * VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR. */
3279
3280 if ((val & zap) != val)
3281 {
3282 LogRel(("hmR0VmxLoadGuestEntryCtls: Invalid VM-entry controls combo! cpu=%RX64 val=%RX64 zap=%RX64\n",
3283 pVM->hm.s.vmx.Msrs.VmxEntry.n.disallowed0, val, zap));
3284 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_ENTRY;
3285 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3286 }
3287
3288 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY, val);
3289 AssertRCReturn(rc, rc);
3290
3291 pVCpu->hm.s.vmx.u32EntryCtls = val;
3292 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_ENTRY_CTLS);
3293 }
3294 return rc;
3295}
3296
3297
3298/**
3299 * Sets up the VM-exit controls in the VMCS.
3300 *
3301 * @returns VBox status code.
3302 * @param pVCpu The cross context virtual CPU structure.
3303 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3304 * out-of-sync. Make sure to update the required fields
3305 * before using them.
3306 *
3307 * @remarks Requires EFER.
3308 */
3309DECLINLINE(int) hmR0VmxLoadGuestExitCtls(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3310{
3311 NOREF(pMixedCtx);
3312
3313 int rc = VINF_SUCCESS;
3314 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_EXIT_CTLS))
3315 {
3316 PVM pVM = pVCpu->CTX_SUFF(pVM);
3317 uint32_t val = pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0; /* Bits set here must be set in the VMCS. */
3318 uint32_t zap = pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1; /* Bits cleared here must be cleared in the VMCS. */
3319
3320 /* Save debug controls (DR7 & IA32_DEBUGCTL_MSR). The first VT-x CPUs only supported the 1-setting of this bit. */
3321 val |= VMX_VMCS_CTRL_EXIT_SAVE_DEBUG;
3322
3323 /*
3324 * Set the host long mode active (EFER.LMA) bit (which Intel calls "Host address-space size") if necessary.
3325 * On VM-exit, VT-x sets both the host EFER.LMA and EFER.LME bit to this value. See assertion in hmR0VmxSaveHostMsrs().
3326 */
3327#if HC_ARCH_BITS == 64
3328 val |= VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE;
3329 Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE\n", pVCpu->idCpu));
3330#else
3331 Assert( pVCpu->hm.s.vmx.pfnStartVM == VMXR0SwitcherStartVM64
3332 || pVCpu->hm.s.vmx.pfnStartVM == VMXR0StartVM32);
3333 /* Set the host address-space size based on the switcher, not guest state. See @bugref{8432}. */
3334 if (pVCpu->hm.s.vmx.pfnStartVM == VMXR0SwitcherStartVM64)
3335 {
3336 /* The switcher returns to long mode, EFER is managed by the switcher. */
3337 val |= VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE;
3338 Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE\n", pVCpu->idCpu));
3339 }
3340 else
3341 Assert(!(val & VMX_VMCS_CTRL_EXIT_HOST_ADDR_SPACE_SIZE));
3342#endif
3343
3344 /* If the newer VMCS fields for managing EFER exists, use it. */
3345 if ( pVM->hm.s.vmx.fSupportsVmcsEfer
3346 && hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
3347 {
3348 val |= VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR
3349 | VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR;
3350 Log4(("Load[%RU32]: VMX_VMCS_CTRL_EXIT_SAVE_GUEST_EFER_MSR, VMX_VMCS_CTRL_EXIT_LOAD_HOST_EFER_MSR\n", pVCpu->idCpu));
3351 }
3352
3353 /* Don't acknowledge external interrupts on VM-exit. We want to let the host do that. */
3354 Assert(!(val & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT));
3355
3356 /** @todo VMX_VMCS_CTRL_EXIT_LOAD_PERF_MSR,
3357 * VMX_VMCS_CTRL_EXIT_SAVE_GUEST_PAT_MSR,
3358 * VMX_VMCS_CTRL_EXIT_LOAD_HOST_PAT_MSR. */
3359
3360 if ( pVM->hm.s.vmx.fUsePreemptTimer
3361 && (pVM->hm.s.vmx.Msrs.VmxExit.n.allowed1 & VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER))
3362 val |= VMX_VMCS_CTRL_EXIT_SAVE_VMX_PREEMPT_TIMER;
3363
3364 if ((val & zap) != val)
3365 {
3366 LogRel(("hmR0VmxSetupProcCtls: Invalid VM-exit controls combo! cpu=%RX64 val=%RX64 zap=%RX64\n",
3367 pVM->hm.s.vmx.Msrs.VmxExit.n.disallowed0, val, zap));
3368 pVCpu->hm.s.u32HMError = VMX_UFC_CTRL_EXIT;
3369 return VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO;
3370 }
3371
3372 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXIT, val);
3373 AssertRCReturn(rc, rc);
3374
3375 pVCpu->hm.s.vmx.u32ExitCtls = val;
3376 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_EXIT_CTLS);
3377 }
3378 return rc;
3379}
3380
3381
3382/**
3383 * Sets the TPR threshold in the VMCS.
3384 *
3385 * @returns VBox status code.
3386 * @param pVCpu The cross context virtual CPU structure.
3387 * @param u32TprThreshold The TPR threshold (task-priority class only).
3388 */
3389DECLINLINE(int) hmR0VmxApicSetTprThreshold(PVMCPU pVCpu, uint32_t u32TprThreshold)
3390{
3391 Assert(!(u32TprThreshold & 0xfffffff0)); /* Bits 31:4 MBZ. */
3392 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW); RT_NOREF_PV(pVCpu);
3393 return VMXWriteVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, u32TprThreshold);
3394}
3395
3396
3397/**
3398 * Loads the guest APIC and related state.
3399 *
3400 * @returns VBox status code.
3401 * @param pVCpu The cross context virtual CPU structure.
3402 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3403 * out-of-sync. Make sure to update the required fields
3404 * before using them.
3405 */
3406DECLINLINE(int) hmR0VmxLoadGuestApicState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3407{
3408 NOREF(pMixedCtx);
3409
3410 int rc = VINF_SUCCESS;
3411 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE))
3412 {
3413 if ( PDMHasApic(pVCpu->CTX_SUFF(pVM))
3414 && APICIsEnabled(pVCpu))
3415 {
3416 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
3417 {
3418 Assert(pVCpu->hm.s.vmx.HCPhysVirtApic);
3419
3420 bool fPendingIntr = false;
3421 uint8_t u8Tpr = 0;
3422 uint8_t u8PendingIntr = 0;
3423 rc = APICGetTpr(pVCpu, &u8Tpr, &fPendingIntr, &u8PendingIntr);
3424 AssertRCReturn(rc, rc);
3425
3426 /*
3427 * If there are interrupts pending but masked by the TPR, instruct VT-x to cause a TPR-below-threshold VM-exit
3428 * when the guest lowers its TPR below the priority of the pending interrupt so we can deliver the interrupt.
3429 * If there are no interrupts pending, set threshold to 0 to not cause any TPR-below-threshold VM-exits.
3430 */
3431 pVCpu->hm.s.vmx.pbVirtApic[XAPIC_OFF_TPR] = u8Tpr;
3432 uint32_t u32TprThreshold = 0;
3433 if (fPendingIntr)
3434 {
3435 /* Bits 3:0 of the TPR threshold field correspond to bits 7:4 of the TPR (which is the Task-Priority Class). */
3436 const uint8_t u8PendingPriority = u8PendingIntr >> 4;
3437 const uint8_t u8TprPriority = u8Tpr >> 4;
3438 if (u8PendingPriority <= u8TprPriority)
3439 u32TprThreshold = u8PendingPriority;
3440 }
3441
3442 rc = hmR0VmxApicSetTprThreshold(pVCpu, u32TprThreshold);
3443 AssertRCReturn(rc, rc);
3444 }
3445
3446#ifndef IEM_VERIFICATION_MODE_FULL
3447 /* Setup the Virtualized APIC accesses. */
3448 if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC)
3449 {
3450 uint64_t u64MsrApicBase = APICGetBaseMsrNoCheck(pVCpu);
3451 if (u64MsrApicBase != pVCpu->hm.s.vmx.u64MsrApicBase)
3452 {
3453 PVM pVM = pVCpu->CTX_SUFF(pVM);
3454 Assert(pVM->hm.s.vmx.HCPhysApicAccess);
3455 RTGCPHYS GCPhysApicBase;
3456 GCPhysApicBase = u64MsrApicBase;
3457 GCPhysApicBase &= PAGE_BASE_GC_MASK;
3458
3459 /* Unalias any existing mapping. */
3460 rc = PGMHandlerPhysicalReset(pVM, GCPhysApicBase);
3461 AssertRCReturn(rc, rc);
3462
3463 /* Map the HC APIC-access page into the GC space, this also updates the shadow page tables if necessary. */
3464 Log4(("Mapped HC APIC-access page into GC: GCPhysApicBase=%#RGp\n", GCPhysApicBase));
3465 rc = IOMMMIOMapMMIOHCPage(pVM, pVCpu, GCPhysApicBase, pVM->hm.s.vmx.HCPhysApicAccess, X86_PTE_RW | X86_PTE_P);
3466 AssertRCReturn(rc, rc);
3467
3468 /* Update VMX's cache of the APIC base. */
3469 pVCpu->hm.s.vmx.u64MsrApicBase = u64MsrApicBase;
3470 }
3471 }
3472#endif /* !IEM_VERIFICATION_MODE_FULL */
3473 }
3474 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
3475 }
3476
3477 return rc;
3478}
3479
3480
3481/**
3482 * Gets the guest's interruptibility-state ("interrupt shadow" as AMD calls it).
3483 *
3484 * @returns Guest's interruptibility-state.
3485 * @param pVCpu The cross context virtual CPU structure.
3486 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3487 * out-of-sync. Make sure to update the required fields
3488 * before using them.
3489 *
3490 * @remarks No-long-jump zone!!!
3491 */
3492DECLINLINE(uint32_t) hmR0VmxGetGuestIntrState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3493{
3494 /*
3495 * Check if we should inhibit interrupt delivery due to instructions like STI and MOV SS.
3496 */
3497 uint32_t uIntrState = 0;
3498 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
3499 {
3500 /* If inhibition is active, RIP & RFLAGS should've been accessed (i.e. read previously from the VMCS or from ring-3). */
3501 AssertMsg(HMVMXCPU_GST_IS_SET(pVCpu, HMVMX_UPDATED_GUEST_RIP | HMVMX_UPDATED_GUEST_RFLAGS),
3502 ("%#x\n", HMVMXCPU_GST_VALUE(pVCpu)));
3503 if (pMixedCtx->rip == EMGetInhibitInterruptsPC(pVCpu))
3504 {
3505 if (pMixedCtx->eflags.Bits.u1IF)
3506 uIntrState = VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI;
3507 else
3508 uIntrState = VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS;
3509 }
3510 else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
3511 {
3512 /*
3513 * We can clear the inhibit force flag as even if we go back to the recompiler without executing guest code in
3514 * VT-x, the flag's condition to be cleared is met and thus the cleared state is correct.
3515 */
3516 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
3517 }
3518 }
3519
3520 /*
3521 * NMIs to the guest are blocked after an NMI is injected until the guest executes an IRET. We only
3522 * bother with virtual-NMI blocking when we have support for virtual NMIs in the CPU, otherwise
3523 * setting this would block host-NMIs and IRET will not clear the blocking.
3524 *
3525 * See Intel spec. 26.6.1 "Interruptibility state". See @bugref{7445}.
3526 */
3527 if ( VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS)
3528 && (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI))
3529 {
3530 uIntrState |= VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI;
3531 }
3532
3533 return uIntrState;
3534}
3535
3536
3537/**
3538 * Loads the guest's interruptibility-state into the guest-state area in the
3539 * VMCS.
3540 *
3541 * @returns VBox status code.
3542 * @param pVCpu The cross context virtual CPU structure.
3543 * @param uIntrState The interruptibility-state to set.
3544 */
3545static int hmR0VmxLoadGuestIntrState(PVMCPU pVCpu, uint32_t uIntrState)
3546{
3547 NOREF(pVCpu);
3548 AssertMsg(!(uIntrState & 0xfffffff0), ("%#x\n", uIntrState)); /* Bits 31:4 MBZ. */
3549 Assert((uIntrState & 0x3) != 0x3); /* Block-by-STI and MOV SS cannot be simultaneously set. */
3550 int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, uIntrState);
3551 AssertRC(rc);
3552 return rc;
3553}
3554
3555
3556/**
3557 * Loads the exception intercepts required for guest execution in the VMCS.
3558 *
3559 * @returns VBox status code.
3560 * @param pVCpu The cross context virtual CPU structure.
3561 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3562 * out-of-sync. Make sure to update the required fields
3563 * before using them.
3564 */
3565static int hmR0VmxLoadGuestXcptIntercepts(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3566{
3567 NOREF(pMixedCtx);
3568 int rc = VINF_SUCCESS;
3569 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS))
3570 {
3571 /* The remaining exception intercepts are handled elsewhere, e.g. in hmR0VmxLoadSharedCR0(). */
3572 if (pVCpu->hm.s.fGIMTrapXcptUD)
3573 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_UD);
3574#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
3575 else
3576 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_UD);
3577#endif
3578
3579 Assert(pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT_32(X86_XCPT_AC));
3580 Assert(pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT_32(X86_XCPT_DB));
3581
3582 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
3583 AssertRCReturn(rc, rc);
3584
3585 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
3586 Log4(("Load[%RU32]: VMX_VMCS32_CTRL_EXCEPTION_BITMAP=%#RX64 fContextUseFlags=%#RX32\n", pVCpu->idCpu,
3587 pVCpu->hm.s.vmx.u32XcptBitmap, HMCPU_CF_VALUE(pVCpu)));
3588 }
3589 return rc;
3590}
3591
3592
3593/**
3594 * Loads the guest's RIP into the guest-state area in the VMCS.
3595 *
3596 * @returns VBox status code.
3597 * @param pVCpu The cross context virtual CPU structure.
3598 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3599 * out-of-sync. Make sure to update the required fields
3600 * before using them.
3601 *
3602 * @remarks No-long-jump zone!!!
3603 */
3604static int hmR0VmxLoadGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3605{
3606 int rc = VINF_SUCCESS;
3607 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RIP))
3608 {
3609 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RIP, pMixedCtx->rip);
3610 AssertRCReturn(rc, rc);
3611
3612 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RIP);
3613 Log4(("Load[%RU32]: VMX_VMCS_GUEST_RIP=%#RX64 fContextUseFlags=%#RX32\n", pVCpu->idCpu, pMixedCtx->rip,
3614 HMCPU_CF_VALUE(pVCpu)));
3615 }
3616 return rc;
3617}
3618
3619
3620/**
3621 * Loads the guest's RSP into the guest-state area in the VMCS.
3622 *
3623 * @returns VBox status code.
3624 * @param pVCpu The cross context virtual CPU structure.
3625 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3626 * out-of-sync. Make sure to update the required fields
3627 * before using them.
3628 *
3629 * @remarks No-long-jump zone!!!
3630 */
3631static int hmR0VmxLoadGuestRsp(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3632{
3633 int rc = VINF_SUCCESS;
3634 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RSP))
3635 {
3636 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_RSP, pMixedCtx->rsp);
3637 AssertRCReturn(rc, rc);
3638
3639 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RSP);
3640 Log4(("Load[%RU32]: VMX_VMCS_GUEST_RSP=%#RX64\n", pVCpu->idCpu, pMixedCtx->rsp));
3641 }
3642 return rc;
3643}
3644
3645
3646/**
3647 * Loads the guest's RFLAGS into the guest-state area in the VMCS.
3648 *
3649 * @returns VBox status code.
3650 * @param pVCpu The cross context virtual CPU structure.
3651 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3652 * out-of-sync. Make sure to update the required fields
3653 * before using them.
3654 *
3655 * @remarks No-long-jump zone!!!
3656 */
3657static int hmR0VmxLoadGuestRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3658{
3659 int rc = VINF_SUCCESS;
3660 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS))
3661 {
3662 /* Intel spec. 2.3.1 "System Flags and Fields in IA-32e Mode" claims the upper 32-bits of RFLAGS are reserved (MBZ).
3663 Let us assert it as such and use 32-bit VMWRITE. */
3664 Assert(!(pMixedCtx->rflags.u64 >> 32));
3665 X86EFLAGS Eflags = pMixedCtx->eflags;
3666 /** @todo r=bird: There shall be no need to OR in X86_EFL_1 here, nor
3667 * shall there be any reason for clearing bits 63:22, 15, 5 and 3.
3668 * These will never be cleared/set, unless some other part of the VMM
3669 * code is buggy - in which case we're better of finding and fixing
3670 * those bugs than hiding them. */
3671 Assert(Eflags.u32 & X86_EFL_RA1_MASK);
3672 Assert(!(Eflags.u32 & ~(X86_EFL_1 | X86_EFL_LIVE_MASK)));
3673 Eflags.u32 &= VMX_EFLAGS_RESERVED_0; /* Bits 22-31, 15, 5 & 3 MBZ. */
3674 Eflags.u32 |= VMX_EFLAGS_RESERVED_1; /* Bit 1 MB1. */
3675
3676 /*
3677 * If we're emulating real-mode using Virtual 8086 mode, save the real-mode eflags so we can restore them on VM-exit.
3678 * Modify the real-mode guest's eflags so that VT-x can run the real-mode guest code under Virtual 8086 mode.
3679 */
3680 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
3681 {
3682 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
3683 Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
3684 pVCpu->hm.s.vmx.RealMode.Eflags.u32 = Eflags.u32; /* Save the original eflags of the real-mode guest. */
3685 Eflags.Bits.u1VM = 1; /* Set the Virtual 8086 mode bit. */
3686 Eflags.Bits.u2IOPL = 0; /* Change IOPL to 0, otherwise certain instructions won't fault. */
3687 }
3688
3689 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_RFLAGS, Eflags.u32);
3690 AssertRCReturn(rc, rc);
3691
3692 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_RFLAGS);
3693 Log4(("Load[%RU32]: VMX_VMCS_GUEST_RFLAGS=%#RX32\n", pVCpu->idCpu, Eflags.u32));
3694 }
3695 return rc;
3696}
3697
3698
3699/**
3700 * Loads the guest RIP, RSP and RFLAGS into the guest-state area in the VMCS.
3701 *
3702 * @returns VBox status code.
3703 * @param pVCpu The cross context virtual CPU structure.
3704 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3705 * out-of-sync. Make sure to update the required fields
3706 * before using them.
3707 *
3708 * @remarks No-long-jump zone!!!
3709 */
3710DECLINLINE(int) hmR0VmxLoadGuestRipRspRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3711{
3712 int rc = hmR0VmxLoadGuestRip(pVCpu, pMixedCtx);
3713 rc |= hmR0VmxLoadGuestRsp(pVCpu, pMixedCtx);
3714 rc |= hmR0VmxLoadGuestRflags(pVCpu, pMixedCtx);
3715 AssertRCReturn(rc, rc);
3716 return rc;
3717}
3718
3719
3720/**
3721 * Loads the guest CR0 control register into the guest-state area in the VMCS.
3722 * CR0 is partially shared with the host and we have to consider the FPU bits.
3723 *
3724 * @returns VBox status code.
3725 * @param pVCpu The cross context virtual CPU structure.
3726 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3727 * out-of-sync. Make sure to update the required fields
3728 * before using them.
3729 *
3730 * @remarks No-long-jump zone!!!
3731 */
3732static int hmR0VmxLoadSharedCR0(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3733{
3734 /*
3735 * Guest CR0.
3736 * Guest FPU.
3737 */
3738 int rc = VINF_SUCCESS;
3739 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0))
3740 {
3741 Assert(!(pMixedCtx->cr0 >> 32));
3742 uint32_t u32GuestCR0 = pMixedCtx->cr0;
3743 PVM pVM = pVCpu->CTX_SUFF(pVM);
3744
3745 /* The guest's view (read access) of its CR0 is unblemished. */
3746 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, u32GuestCR0);
3747 AssertRCReturn(rc, rc);
3748 Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR0_READ_SHADOW=%#RX32\n", pVCpu->idCpu, u32GuestCR0));
3749
3750 /* Setup VT-x's view of the guest CR0. */
3751 /* Minimize VM-exits due to CR3 changes when we have NestedPaging. */
3752 if (pVM->hm.s.fNestedPaging)
3753 {
3754 if (CPUMIsGuestPagingEnabledEx(pMixedCtx))
3755 {
3756 /* The guest has paging enabled, let it access CR3 without causing a VM-exit if supported. */
3757 pVCpu->hm.s.vmx.u32ProcCtls &= ~( VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
3758 | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT);
3759 }
3760 else
3761 {
3762 /* The guest doesn't have paging enabled, make CR3 access cause a VM-exit to update our shadow. */
3763 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT
3764 | VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
3765 }
3766
3767 /* If we have unrestricted guest execution, we never have to intercept CR3 reads. */
3768 if (pVM->hm.s.vmx.fUnrestrictedGuest)
3769 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT;
3770
3771 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
3772 AssertRCReturn(rc, rc);
3773 }
3774 else
3775 u32GuestCR0 |= X86_CR0_WP; /* Guest CPL 0 writes to its read-only pages should cause a #PF VM-exit. */
3776
3777 /*
3778 * Guest FPU bits.
3779 * Intel spec. 23.8 "Restrictions on VMX operation" mentions that CR0.NE bit must always be set on the first
3780 * CPUs to support VT-x and no mention of with regards to UX in VM-entry checks.
3781 */
3782 u32GuestCR0 |= X86_CR0_NE;
3783 bool fInterceptNM = false;
3784 if (CPUMIsGuestFPUStateActive(pVCpu))
3785 {
3786 fInterceptNM = false; /* Guest FPU active, no need to VM-exit on #NM. */
3787 /* The guest should still get #NM exceptions when it expects it to, so we should not clear TS & MP bits here.
3788 We're only concerned about -us- not intercepting #NMs when the guest-FPU is active. Not the guest itself! */
3789 }
3790 else
3791 {
3792 fInterceptNM = true; /* Guest FPU inactive, VM-exit on #NM for lazy FPU loading. */
3793 u32GuestCR0 |= X86_CR0_TS /* Guest can task switch quickly and do lazy FPU syncing. */
3794 | X86_CR0_MP; /* FWAIT/WAIT should not ignore CR0.TS and should generate #NM. */
3795 }
3796
3797 /* Catch floating point exceptions if we need to report them to the guest in a different way. */
3798 bool fInterceptMF = false;
3799 if (!(pMixedCtx->cr0 & X86_CR0_NE))
3800 fInterceptMF = true;
3801
3802 /* Finally, intercept all exceptions as we cannot directly inject them in real-mode, see hmR0VmxInjectEventVmcs(). */
3803 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
3804 {
3805 Assert(PDMVmmDevHeapIsEnabled(pVM));
3806 Assert(pVM->hm.s.vmx.pRealModeTSS);
3807 pVCpu->hm.s.vmx.u32XcptBitmap |= HMVMX_REAL_MODE_XCPT_MASK;
3808 fInterceptNM = true;
3809 fInterceptMF = true;
3810 }
3811 else
3812 {
3813 /* For now, cleared here as mode-switches can happen outside HM/VT-x. See @bugref{7626#c11}. */
3814 pVCpu->hm.s.vmx.u32XcptBitmap &= ~HMVMX_REAL_MODE_XCPT_MASK;
3815 }
3816 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
3817
3818 if (fInterceptNM)
3819 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_NM);
3820 else
3821 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_NM);
3822
3823 if (fInterceptMF)
3824 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_MF);
3825 else
3826 pVCpu->hm.s.vmx.u32XcptBitmap &= ~RT_BIT(X86_XCPT_MF);
3827
3828 /* Additional intercepts for debugging, define these yourself explicitly. */
3829#ifdef HMVMX_ALWAYS_TRAP_ALL_XCPTS
3830 pVCpu->hm.s.vmx.u32XcptBitmap |= 0
3831 | RT_BIT(X86_XCPT_BP)
3832 | RT_BIT(X86_XCPT_DE)
3833 | RT_BIT(X86_XCPT_NM)
3834 | RT_BIT(X86_XCPT_TS)
3835 | RT_BIT(X86_XCPT_UD)
3836 | RT_BIT(X86_XCPT_NP)
3837 | RT_BIT(X86_XCPT_SS)
3838 | RT_BIT(X86_XCPT_GP)
3839 | RT_BIT(X86_XCPT_PF)
3840 | RT_BIT(X86_XCPT_MF)
3841 ;
3842#elif defined(HMVMX_ALWAYS_TRAP_PF)
3843 pVCpu->hm.s.vmx.u32XcptBitmap |= RT_BIT(X86_XCPT_PF);
3844#endif
3845
3846 Assert(pVM->hm.s.fNestedPaging || (pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT(X86_XCPT_PF)));
3847
3848 /* Set/clear the CR0 specific bits along with their exceptions (PE, PG, CD, NW). */
3849 uint32_t uSetCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
3850 uint32_t uZapCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
3851 if (pVM->hm.s.vmx.fUnrestrictedGuest) /* Exceptions for unrestricted-guests for fixed CR0 bits (PE, PG). */
3852 uSetCR0 &= ~(X86_CR0_PE | X86_CR0_PG);
3853 else
3854 Assert((uSetCR0 & (X86_CR0_PE | X86_CR0_PG)) == (X86_CR0_PE | X86_CR0_PG));
3855
3856 u32GuestCR0 |= uSetCR0;
3857 u32GuestCR0 &= uZapCR0;
3858 u32GuestCR0 &= ~(X86_CR0_CD | X86_CR0_NW); /* Always enable caching. */
3859
3860 /* Write VT-x's view of the guest CR0 into the VMCS. */
3861 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR0, u32GuestCR0);
3862 AssertRCReturn(rc, rc);
3863 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR0=%#RX32 (uSetCR0=%#RX32 uZapCR0=%#RX32)\n", pVCpu->idCpu, u32GuestCR0, uSetCR0,
3864 uZapCR0));
3865
3866 /*
3867 * CR0 is shared between host and guest along with a CR0 read shadow. Therefore, certain bits must not be changed
3868 * by the guest because VT-x ignores saving/restoring them (namely CD, ET, NW) and for certain other bits
3869 * we want to be notified immediately of guest CR0 changes (e.g. PG to update our shadow page tables).
3870 */
3871 uint32_t u32CR0Mask = 0;
3872 u32CR0Mask = X86_CR0_PE
3873 | X86_CR0_NE
3874 | X86_CR0_WP
3875 | X86_CR0_PG
3876 | X86_CR0_ET /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.ET */
3877 | X86_CR0_CD /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.CD */
3878 | X86_CR0_NW; /* Bit ignored on VM-entry and VM-exit. Don't let the guest modify the host CR0.NW */
3879
3880 /** @todo Avoid intercepting CR0.PE with unrestricted guests. Fix PGM
3881 * enmGuestMode to be in-sync with the current mode. See @bugref{6398}
3882 * and @bugref{6944}. */
3883#if 0
3884 if (pVM->hm.s.vmx.fUnrestrictedGuest)
3885 u32CR0Mask &= ~X86_CR0_PE;
3886#endif
3887 if (pVM->hm.s.fNestedPaging)
3888 u32CR0Mask &= ~X86_CR0_WP;
3889
3890 /* If the guest FPU state is active, don't need to VM-exit on writes to FPU related bits in CR0. */
3891 if (fInterceptNM)
3892 {
3893 u32CR0Mask |= X86_CR0_TS
3894 | X86_CR0_MP;
3895 }
3896
3897 /* Write the CR0 mask into the VMCS and update the VCPU's copy of the current CR0 mask. */
3898 pVCpu->hm.s.vmx.u32CR0Mask = u32CR0Mask;
3899 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_MASK, u32CR0Mask);
3900 AssertRCReturn(rc, rc);
3901 Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR0_MASK=%#RX32\n", pVCpu->idCpu, u32CR0Mask));
3902
3903 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR0);
3904 }
3905 return rc;
3906}
3907
3908
3909/**
3910 * Loads the guest control registers (CR3, CR4) into the guest-state area
3911 * in the VMCS.
3912 *
3913 * @returns VBox strict status code.
3914 * @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
3915 * without unrestricted guest access and the VMMDev is not presently
3916 * mapped (e.g. EFI32).
3917 *
3918 * @param pVCpu The cross context virtual CPU structure.
3919 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
3920 * out-of-sync. Make sure to update the required fields
3921 * before using them.
3922 *
3923 * @remarks No-long-jump zone!!!
3924 */
3925static VBOXSTRICTRC hmR0VmxLoadGuestCR3AndCR4(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
3926{
3927 int rc = VINF_SUCCESS;
3928 PVM pVM = pVCpu->CTX_SUFF(pVM);
3929
3930 /*
3931 * Guest CR2.
3932 * It's always loaded in the assembler code. Nothing to do here.
3933 */
3934
3935 /*
3936 * Guest CR3.
3937 */
3938 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR3))
3939 {
3940 RTGCPHYS GCPhysGuestCR3 = NIL_RTGCPHYS;
3941 if (pVM->hm.s.fNestedPaging)
3942 {
3943 pVCpu->hm.s.vmx.HCPhysEPTP = PGMGetHyperCR3(pVCpu);
3944
3945 /* Validate. See Intel spec. 28.2.2 "EPT Translation Mechanism" and 24.6.11 "Extended-Page-Table Pointer (EPTP)" */
3946 Assert(pVCpu->hm.s.vmx.HCPhysEPTP);
3947 Assert(!(pVCpu->hm.s.vmx.HCPhysEPTP & UINT64_C(0xfff0000000000000)));
3948 Assert(!(pVCpu->hm.s.vmx.HCPhysEPTP & 0xfff));
3949
3950 /* VMX_EPT_MEMTYPE_WB support is already checked in hmR0VmxSetupTaggedTlb(). */
3951 pVCpu->hm.s.vmx.HCPhysEPTP |= VMX_EPT_MEMTYPE_WB
3952 | (VMX_EPT_PAGE_WALK_LENGTH_DEFAULT << VMX_EPT_PAGE_WALK_LENGTH_SHIFT);
3953
3954 /* Validate. See Intel spec. 26.2.1 "Checks on VMX Controls" */
3955 AssertMsg( ((pVCpu->hm.s.vmx.HCPhysEPTP >> 3) & 0x07) == 3 /* Bits 3:5 (EPT page walk length - 1) must be 3. */
3956 && ((pVCpu->hm.s.vmx.HCPhysEPTP >> 7) & 0x1f) == 0, /* Bits 7:11 MBZ. */
3957 ("EPTP %#RX64\n", pVCpu->hm.s.vmx.HCPhysEPTP));
3958 AssertMsg( !((pVCpu->hm.s.vmx.HCPhysEPTP >> 6) & 0x01) /* Bit 6 (EPT accessed & dirty bit). */
3959 || (pVM->hm.s.vmx.Msrs.u64EptVpidCaps & MSR_IA32_VMX_EPT_VPID_CAP_EPT_ACCESS_DIRTY),
3960 ("EPTP accessed/dirty bit not supported by CPU but set %#RX64\n", pVCpu->hm.s.vmx.HCPhysEPTP));
3961
3962 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, pVCpu->hm.s.vmx.HCPhysEPTP);
3963 AssertRCReturn(rc, rc);
3964 Log4(("Load[%RU32]: VMX_VMCS64_CTRL_EPTP_FULL=%#RX64\n", pVCpu->idCpu, pVCpu->hm.s.vmx.HCPhysEPTP));
3965
3966 if ( pVM->hm.s.vmx.fUnrestrictedGuest
3967 || CPUMIsGuestPagingEnabledEx(pMixedCtx))
3968 {
3969 /* If the guest is in PAE mode, pass the PDPEs to VT-x using the VMCS fields. */
3970 if (CPUMIsGuestInPAEModeEx(pMixedCtx))
3971 {
3972 rc = PGMGstGetPaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
3973 AssertRCReturn(rc, rc);
3974 rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, pVCpu->hm.s.aPdpes[0].u);
3975 rc |= VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, pVCpu->hm.s.aPdpes[1].u);
3976 rc |= VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, pVCpu->hm.s.aPdpes[2].u);
3977 rc |= VMXWriteVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, pVCpu->hm.s.aPdpes[3].u);
3978 AssertRCReturn(rc, rc);
3979 }
3980
3981 /* The guest's view of its CR3 is unblemished with Nested Paging when the guest is using paging or we
3982 have Unrestricted Execution to handle the guest when it's not using paging. */
3983 GCPhysGuestCR3 = pMixedCtx->cr3;
3984 }
3985 else
3986 {
3987 /*
3988 * The guest is not using paging, but the CPU (VT-x) has to. While the guest thinks it accesses physical memory
3989 * directly, we use our identity-mapped page table to map guest-linear to guest-physical addresses.
3990 * EPT takes care of translating it to host-physical addresses.
3991 */
3992 RTGCPHYS GCPhys;
3993 Assert(pVM->hm.s.vmx.pNonPagingModeEPTPageTable);
3994
3995 /* We obtain it here every time as the guest could have relocated this PCI region. */
3996 rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
3997 if (RT_SUCCESS(rc))
3998 { /* likely */ }
3999 else if (rc == VERR_PDM_DEV_HEAP_R3_TO_GCPHYS)
4000 {
4001 Log4(("Load[%RU32]: VERR_PDM_DEV_HEAP_R3_TO_GCPHYS -> VINF_EM_RESCHEDULE_REM\n", pVCpu->idCpu));
4002 return VINF_EM_RESCHEDULE_REM; /* We cannot execute now, switch to REM/IEM till the guest maps in VMMDev. */
4003 }
4004 else
4005 AssertMsgFailedReturn(("%Rrc\n", rc), rc);
4006
4007 GCPhysGuestCR3 = GCPhys;
4008 }
4009
4010 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR3=%#RGp (GstN)\n", pVCpu->idCpu, GCPhysGuestCR3));
4011 rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_CR3, GCPhysGuestCR3);
4012 }
4013 else
4014 {
4015 /* Non-nested paging case, just use the hypervisor's CR3. */
4016 RTHCPHYS HCPhysGuestCR3 = PGMGetHyperCR3(pVCpu);
4017
4018 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR3=%#RHv (HstN)\n", pVCpu->idCpu, HCPhysGuestCR3));
4019 rc = VMXWriteVmcsHstN(VMX_VMCS_GUEST_CR3, HCPhysGuestCR3);
4020 }
4021 AssertRCReturn(rc, rc);
4022
4023 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR3);
4024 }
4025
4026 /*
4027 * Guest CR4.
4028 * ASSUMES this is done everytime we get in from ring-3! (XCR0)
4029 */
4030 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR4))
4031 {
4032 Assert(!(pMixedCtx->cr4 >> 32));
4033 uint32_t u32GuestCR4 = pMixedCtx->cr4;
4034
4035 /* The guest's view of its CR4 is unblemished. */
4036 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, u32GuestCR4);
4037 AssertRCReturn(rc, rc);
4038 Log4(("Load[%RU32]: VMX_VMCS_CTRL_CR4_READ_SHADOW=%#RX32\n", pVCpu->idCpu, u32GuestCR4));
4039
4040 /* Setup VT-x's view of the guest CR4. */
4041 /*
4042 * If we're emulating real-mode using virtual-8086 mode, we want to redirect software interrupts to the 8086 program
4043 * interrupt handler. Clear the VME bit (the interrupt redirection bitmap is already all 0, see hmR3InitFinalizeR0())
4044 * See Intel spec. 20.2 "Software Interrupt Handling Methods While in Virtual-8086 Mode".
4045 */
4046 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4047 {
4048 Assert(pVM->hm.s.vmx.pRealModeTSS);
4049 Assert(PDMVmmDevHeapIsEnabled(pVM));
4050 u32GuestCR4 &= ~X86_CR4_VME;
4051 }
4052
4053 if (pVM->hm.s.fNestedPaging)
4054 {
4055 if ( !CPUMIsGuestPagingEnabledEx(pMixedCtx)
4056 && !pVM->hm.s.vmx.fUnrestrictedGuest)
4057 {
4058 /* We use 4 MB pages in our identity mapping page table when the guest doesn't have paging. */
4059 u32GuestCR4 |= X86_CR4_PSE;
4060 /* Our identity mapping is a 32-bit page directory. */
4061 u32GuestCR4 &= ~X86_CR4_PAE;
4062 }
4063 /* else use guest CR4.*/
4064 }
4065 else
4066 {
4067 /*
4068 * The shadow paging modes and guest paging modes are different, the shadow is in accordance with the host
4069 * paging mode and thus we need to adjust VT-x's view of CR4 depending on our shadow page tables.
4070 */
4071 switch (pVCpu->hm.s.enmShadowMode)
4072 {
4073 case PGMMODE_REAL: /* Real-mode. */
4074 case PGMMODE_PROTECTED: /* Protected mode without paging. */
4075 case PGMMODE_32_BIT: /* 32-bit paging. */
4076 {
4077 u32GuestCR4 &= ~X86_CR4_PAE;
4078 break;
4079 }
4080
4081 case PGMMODE_PAE: /* PAE paging. */
4082 case PGMMODE_PAE_NX: /* PAE paging with NX. */
4083 {
4084 u32GuestCR4 |= X86_CR4_PAE;
4085 break;
4086 }
4087
4088 case PGMMODE_AMD64: /* 64-bit AMD paging (long mode). */
4089 case PGMMODE_AMD64_NX: /* 64-bit AMD paging (long mode) with NX enabled. */
4090#ifdef VBOX_ENABLE_64_BITS_GUESTS
4091 break;
4092#endif
4093 default:
4094 AssertFailed();
4095 return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
4096 }
4097 }
4098
4099 /* We need to set and clear the CR4 specific bits here (mainly the X86_CR4_VMXE bit). */
4100 uint64_t uSetCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
4101 uint64_t uZapCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
4102 u32GuestCR4 |= uSetCR4;
4103 u32GuestCR4 &= uZapCR4;
4104
4105 /* Write VT-x's view of the guest CR4 into the VMCS. */
4106 Log4(("Load[%RU32]: VMX_VMCS_GUEST_CR4=%#RX32 (Set=%#RX32 Zap=%#RX32)\n", pVCpu->idCpu, u32GuestCR4, uSetCR4, uZapCR4));
4107 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_CR4, u32GuestCR4);
4108 AssertRCReturn(rc, rc);
4109
4110 /* Setup CR4 mask. CR4 flags owned by the host, if the guest attempts to change them, that would cause a VM-exit. */
4111 uint32_t u32CR4Mask = X86_CR4_VME
4112 | X86_CR4_PAE
4113 | X86_CR4_PGE
4114 | X86_CR4_PSE
4115 | X86_CR4_VMXE;
4116 if (pVM->cpum.ro.HostFeatures.fXSaveRstor)
4117 u32CR4Mask |= X86_CR4_OSXSAVE;
4118 pVCpu->hm.s.vmx.u32CR4Mask = u32CR4Mask;
4119 rc = VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_MASK, u32CR4Mask);
4120 AssertRCReturn(rc, rc);
4121
4122 /* Whether to save/load/restore XCR0 during world switch depends on CR4.OSXSAVE and host+guest XCR0. */
4123 pVCpu->hm.s.fLoadSaveGuestXcr0 = (pMixedCtx->cr4 & X86_CR4_OSXSAVE) && pMixedCtx->aXcr[0] != ASMGetXcr0();
4124
4125 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR4);
4126 }
4127 return rc;
4128}
4129
4130
4131/**
4132 * Loads the guest debug registers into the guest-state area in the VMCS.
4133 *
4134 * This also sets up whether \#DB and MOV DRx accesses cause VM-exits.
4135 *
4136 * The guest debug bits are partially shared with the host (e.g. DR6, DR0-3).
4137 *
4138 * @returns VBox status code.
4139 * @param pVCpu The cross context virtual CPU structure.
4140 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4141 * out-of-sync. Make sure to update the required fields
4142 * before using them.
4143 *
4144 * @remarks No-long-jump zone!!!
4145 */
4146static int hmR0VmxLoadSharedDebugState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4147{
4148 if (!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG))
4149 return VINF_SUCCESS;
4150
4151#ifdef VBOX_STRICT
4152 /* Validate. Intel spec. 26.3.1.1 "Checks on Guest Controls Registers, Debug Registers, MSRs" */
4153 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
4154 {
4155 /* Validate. Intel spec. 17.2 "Debug Registers", recompiler paranoia checks. */
4156 Assert((pMixedCtx->dr[7] & (X86_DR7_MBZ_MASK | X86_DR7_RAZ_MASK)) == 0); /* Bits 63:32, 15, 14, 12, 11 are reserved. */
4157 Assert((pMixedCtx->dr[7] & X86_DR7_RA1_MASK) == X86_DR7_RA1_MASK); /* Bit 10 is reserved (RA1). */
4158 }
4159#endif
4160
4161 int rc;
4162 PVM pVM = pVCpu->CTX_SUFF(pVM);
4163 bool fSteppingDB = false;
4164 bool fInterceptMovDRx = false;
4165 if (pVCpu->hm.s.fSingleInstruction)
4166 {
4167 /* If the CPU supports the monitor trap flag, use it for single stepping in DBGF and avoid intercepting #DB. */
4168 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG)
4169 {
4170 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG;
4171 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
4172 AssertRCReturn(rc, rc);
4173 Assert(fSteppingDB == false);
4174 }
4175 else
4176 {
4177 pMixedCtx->eflags.u32 |= X86_EFL_TF;
4178 pVCpu->hm.s.fClearTrapFlag = true;
4179 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
4180 fSteppingDB = true;
4181 }
4182 }
4183
4184 if ( fSteppingDB
4185 || (CPUMGetHyperDR7(pVCpu) & X86_DR7_ENABLED_MASK))
4186 {
4187 /*
4188 * Use the combined guest and host DRx values found in the hypervisor
4189 * register set because the debugger has breakpoints active or someone
4190 * is single stepping on the host side without a monitor trap flag.
4191 *
4192 * Note! DBGF expects a clean DR6 state before executing guest code.
4193 */
4194#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
4195 if ( CPUMIsGuestInLongModeEx(pMixedCtx)
4196 && !CPUMIsHyperDebugStateActivePending(pVCpu))
4197 {
4198 CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
4199 Assert(CPUMIsHyperDebugStateActivePending(pVCpu));
4200 Assert(!CPUMIsGuestDebugStateActivePending(pVCpu));
4201 }
4202 else
4203#endif
4204 if (!CPUMIsHyperDebugStateActive(pVCpu))
4205 {
4206 CPUMR0LoadHyperDebugState(pVCpu, true /* include DR6 */);
4207 Assert(CPUMIsHyperDebugStateActive(pVCpu));
4208 Assert(!CPUMIsGuestDebugStateActive(pVCpu));
4209 }
4210
4211 /* Update DR7. (The other DRx values are handled by CPUM one way or the other.) */
4212 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)CPUMGetHyperDR7(pVCpu));
4213 AssertRCReturn(rc, rc);
4214
4215 pVCpu->hm.s.fUsingHyperDR7 = true;
4216 fInterceptMovDRx = true;
4217 }
4218 else
4219 {
4220 /*
4221 * If the guest has enabled debug registers, we need to load them prior to
4222 * executing guest code so they'll trigger at the right time.
4223 */
4224 if (pMixedCtx->dr[7] & (X86_DR7_ENABLED_MASK | X86_DR7_GD)) /** @todo Why GD? */
4225 {
4226#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
4227 if ( CPUMIsGuestInLongModeEx(pMixedCtx)
4228 && !CPUMIsGuestDebugStateActivePending(pVCpu))
4229 {
4230 CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
4231 Assert(CPUMIsGuestDebugStateActivePending(pVCpu));
4232 Assert(!CPUMIsHyperDebugStateActivePending(pVCpu));
4233 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
4234 }
4235 else
4236#endif
4237 if (!CPUMIsGuestDebugStateActive(pVCpu))
4238 {
4239 CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
4240 Assert(CPUMIsGuestDebugStateActive(pVCpu));
4241 Assert(!CPUMIsHyperDebugStateActive(pVCpu));
4242 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxArmed);
4243 }
4244 Assert(!fInterceptMovDRx);
4245 }
4246 /*
4247 * If no debugging enabled, we'll lazy load DR0-3. Unlike on AMD-V, we
4248 * must intercept #DB in order to maintain a correct DR6 guest value, and
4249 * because we need to intercept it to prevent nested #DBs from hanging the
4250 * CPU, we end up always having to intercept it. See hmR0VmxInitXcptBitmap.
4251 */
4252#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
4253 else if ( !CPUMIsGuestDebugStateActivePending(pVCpu)
4254 && !CPUMIsGuestDebugStateActive(pVCpu))
4255#else
4256 else if (!CPUMIsGuestDebugStateActive(pVCpu))
4257#endif
4258 {
4259 fInterceptMovDRx = true;
4260 }
4261
4262 /* Update guest DR7. */
4263 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, pMixedCtx->dr[7]);
4264 AssertRCReturn(rc, rc);
4265
4266 pVCpu->hm.s.fUsingHyperDR7 = false;
4267 }
4268
4269 /*
4270 * Update the processor-based VM-execution controls regarding intercepting MOV DRx instructions.
4271 */
4272 if (fInterceptMovDRx)
4273 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
4274 else
4275 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
4276 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
4277 AssertRCReturn(rc, rc);
4278
4279 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_DEBUG);
4280 return VINF_SUCCESS;
4281}
4282
4283
4284#ifdef VBOX_STRICT
4285/**
4286 * Strict function to validate segment registers.
4287 *
4288 * @remarks ASSUMES CR0 is up to date.
4289 */
4290static void hmR0VmxValidateSegmentRegs(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
4291{
4292 /* Validate segment registers. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers". */
4293 /* NOTE: The reason we check for attribute value 0 and not just the unusable bit here is because hmR0VmxWriteSegmentReg()
4294 * only updates the VMCS' copy of the value with the unusable bit and doesn't change the guest-context value. */
4295 if ( !pVM->hm.s.vmx.fUnrestrictedGuest
4296 && ( !CPUMIsGuestInRealModeEx(pCtx)
4297 && !CPUMIsGuestInV86ModeEx(pCtx)))
4298 {
4299 /* Protected mode checks */
4300 /* CS */
4301 Assert(pCtx->cs.Attr.n.u1Present);
4302 Assert(!(pCtx->cs.Attr.u & 0xf00));
4303 Assert(!(pCtx->cs.Attr.u & 0xfffe0000));
4304 Assert( (pCtx->cs.u32Limit & 0xfff) == 0xfff
4305 || !(pCtx->cs.Attr.n.u1Granularity));
4306 Assert( !(pCtx->cs.u32Limit & 0xfff00000)
4307 || (pCtx->cs.Attr.n.u1Granularity));
4308 /* CS cannot be loaded with NULL in protected mode. */
4309 Assert(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE)); /** @todo is this really true even for 64-bit CS? */
4310 if (pCtx->cs.Attr.n.u4Type == 9 || pCtx->cs.Attr.n.u4Type == 11)
4311 Assert(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl);
4312 else if (pCtx->cs.Attr.n.u4Type == 13 || pCtx->cs.Attr.n.u4Type == 15)
4313 Assert(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl);
4314 else
4315 AssertMsgFailed(("Invalid CS Type %#x\n", pCtx->cs.Attr.n.u2Dpl));
4316 /* SS */
4317 Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
4318 Assert(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL));
4319 if ( !(pCtx->cr0 & X86_CR0_PE)
4320 || pCtx->cs.Attr.n.u4Type == 3)
4321 {
4322 Assert(!pCtx->ss.Attr.n.u2Dpl);
4323 }
4324 if (pCtx->ss.Attr.u && !(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
4325 {
4326 Assert((pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL));
4327 Assert(pCtx->ss.Attr.n.u4Type == 3 || pCtx->ss.Attr.n.u4Type == 7);
4328 Assert(pCtx->ss.Attr.n.u1Present);
4329 Assert(!(pCtx->ss.Attr.u & 0xf00));
4330 Assert(!(pCtx->ss.Attr.u & 0xfffe0000));
4331 Assert( (pCtx->ss.u32Limit & 0xfff) == 0xfff
4332 || !(pCtx->ss.Attr.n.u1Granularity));
4333 Assert( !(pCtx->ss.u32Limit & 0xfff00000)
4334 || (pCtx->ss.Attr.n.u1Granularity));
4335 }
4336 /* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxWriteSegmentReg(). */
4337 if (pCtx->ds.Attr.u && !(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
4338 {
4339 Assert(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4340 Assert(pCtx->ds.Attr.n.u1Present);
4341 Assert(pCtx->ds.Attr.n.u4Type > 11 || pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL));
4342 Assert(!(pCtx->ds.Attr.u & 0xf00));
4343 Assert(!(pCtx->ds.Attr.u & 0xfffe0000));
4344 Assert( (pCtx->ds.u32Limit & 0xfff) == 0xfff
4345 || !(pCtx->ds.Attr.n.u1Granularity));
4346 Assert( !(pCtx->ds.u32Limit & 0xfff00000)
4347 || (pCtx->ds.Attr.n.u1Granularity));
4348 Assert( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4349 || (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ));
4350 }
4351 if (pCtx->es.Attr.u && !(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
4352 {
4353 Assert(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4354 Assert(pCtx->es.Attr.n.u1Present);
4355 Assert(pCtx->es.Attr.n.u4Type > 11 || pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL));
4356 Assert(!(pCtx->es.Attr.u & 0xf00));
4357 Assert(!(pCtx->es.Attr.u & 0xfffe0000));
4358 Assert( (pCtx->es.u32Limit & 0xfff) == 0xfff
4359 || !(pCtx->es.Attr.n.u1Granularity));
4360 Assert( !(pCtx->es.u32Limit & 0xfff00000)
4361 || (pCtx->es.Attr.n.u1Granularity));
4362 Assert( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4363 || (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ));
4364 }
4365 if (pCtx->fs.Attr.u && !(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
4366 {
4367 Assert(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4368 Assert(pCtx->fs.Attr.n.u1Present);
4369 Assert(pCtx->fs.Attr.n.u4Type > 11 || pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL));
4370 Assert(!(pCtx->fs.Attr.u & 0xf00));
4371 Assert(!(pCtx->fs.Attr.u & 0xfffe0000));
4372 Assert( (pCtx->fs.u32Limit & 0xfff) == 0xfff
4373 || !(pCtx->fs.Attr.n.u1Granularity));
4374 Assert( !(pCtx->fs.u32Limit & 0xfff00000)
4375 || (pCtx->fs.Attr.n.u1Granularity));
4376 Assert( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4377 || (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ));
4378 }
4379 if (pCtx->gs.Attr.u && !(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
4380 {
4381 Assert(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED);
4382 Assert(pCtx->gs.Attr.n.u1Present);
4383 Assert(pCtx->gs.Attr.n.u4Type > 11 || pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL));
4384 Assert(!(pCtx->gs.Attr.u & 0xf00));
4385 Assert(!(pCtx->gs.Attr.u & 0xfffe0000));
4386 Assert( (pCtx->gs.u32Limit & 0xfff) == 0xfff
4387 || !(pCtx->gs.Attr.n.u1Granularity));
4388 Assert( !(pCtx->gs.u32Limit & 0xfff00000)
4389 || (pCtx->gs.Attr.n.u1Granularity));
4390 Assert( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
4391 || (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ));
4392 }
4393 /* 64-bit capable CPUs. */
4394# if HC_ARCH_BITS == 64
4395 Assert(!(pCtx->cs.u64Base >> 32));
4396 Assert(!pCtx->ss.Attr.u || !(pCtx->ss.u64Base >> 32));
4397 Assert(!pCtx->ds.Attr.u || !(pCtx->ds.u64Base >> 32));
4398 Assert(!pCtx->es.Attr.u || !(pCtx->es.u64Base >> 32));
4399# endif
4400 }
4401 else if ( CPUMIsGuestInV86ModeEx(pCtx)
4402 || ( CPUMIsGuestInRealModeEx(pCtx)
4403 && !pVM->hm.s.vmx.fUnrestrictedGuest))
4404 {
4405 /* Real and v86 mode checks. */
4406 /* hmR0VmxWriteSegmentReg() writes the modified in VMCS. We want what we're feeding to VT-x. */
4407 uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
4408 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4409 {
4410 u32CSAttr = 0xf3; u32SSAttr = 0xf3; u32DSAttr = 0xf3; u32ESAttr = 0xf3; u32FSAttr = 0xf3; u32GSAttr = 0xf3;
4411 }
4412 else
4413 {
4414 u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u; u32DSAttr = pCtx->ds.Attr.u;
4415 u32ESAttr = pCtx->es.Attr.u; u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
4416 }
4417
4418 /* CS */
4419 AssertMsg((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), ("CS base %#x %#x\n", pCtx->cs.u64Base, pCtx->cs.Sel));
4420 Assert(pCtx->cs.u32Limit == 0xffff);
4421 Assert(u32CSAttr == 0xf3);
4422 /* SS */
4423 Assert(pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4);
4424 Assert(pCtx->ss.u32Limit == 0xffff);
4425 Assert(u32SSAttr == 0xf3);
4426 /* DS */
4427 Assert(pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4);
4428 Assert(pCtx->ds.u32Limit == 0xffff);
4429 Assert(u32DSAttr == 0xf3);
4430 /* ES */
4431 Assert(pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4);
4432 Assert(pCtx->es.u32Limit == 0xffff);
4433 Assert(u32ESAttr == 0xf3);
4434 /* FS */
4435 Assert(pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4);
4436 Assert(pCtx->fs.u32Limit == 0xffff);
4437 Assert(u32FSAttr == 0xf3);
4438 /* GS */
4439 Assert(pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4);
4440 Assert(pCtx->gs.u32Limit == 0xffff);
4441 Assert(u32GSAttr == 0xf3);
4442 /* 64-bit capable CPUs. */
4443# if HC_ARCH_BITS == 64
4444 Assert(!(pCtx->cs.u64Base >> 32));
4445 Assert(!u32SSAttr || !(pCtx->ss.u64Base >> 32));
4446 Assert(!u32DSAttr || !(pCtx->ds.u64Base >> 32));
4447 Assert(!u32ESAttr || !(pCtx->es.u64Base >> 32));
4448# endif
4449 }
4450}
4451#endif /* VBOX_STRICT */
4452
4453
4454/**
4455 * Writes a guest segment register into the guest-state area in the VMCS.
4456 *
4457 * @returns VBox status code.
4458 * @param pVCpu The cross context virtual CPU structure.
4459 * @param idxSel Index of the selector in the VMCS.
4460 * @param idxLimit Index of the segment limit in the VMCS.
4461 * @param idxBase Index of the segment base in the VMCS.
4462 * @param idxAccess Index of the access rights of the segment in the VMCS.
4463 * @param pSelReg Pointer to the segment selector.
4464 *
4465 * @remarks No-long-jump zone!!!
4466 */
4467static int hmR0VmxWriteSegmentReg(PVMCPU pVCpu, uint32_t idxSel, uint32_t idxLimit, uint32_t idxBase,
4468 uint32_t idxAccess, PCPUMSELREG pSelReg)
4469{
4470 int rc = VMXWriteVmcs32(idxSel, pSelReg->Sel); /* 16-bit guest selector field. */
4471 rc |= VMXWriteVmcs32(idxLimit, pSelReg->u32Limit); /* 32-bit guest segment limit field. */
4472 rc |= VMXWriteVmcsGstN(idxBase, pSelReg->u64Base); /* Natural width guest segment base field.*/
4473 AssertRCReturn(rc, rc);
4474
4475 uint32_t u32Access = pSelReg->Attr.u;
4476 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4477 {
4478 /* VT-x requires our real-using-v86 mode hack to override the segment access-right bits. */
4479 u32Access = 0xf3;
4480 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
4481 Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
4482 }
4483 else
4484 {
4485 /*
4486 * The way to differentiate between whether this is really a null selector or was just a selector loaded with 0 in
4487 * real-mode is using the segment attributes. A selector loaded in real-mode with the value 0 is valid and usable in
4488 * protected-mode and we should -not- mark it as an unusable segment. Both the recompiler & VT-x ensures NULL selectors
4489 * loaded in protected-mode have their attribute as 0.
4490 */
4491 if (!u32Access)
4492 u32Access = X86DESCATTR_UNUSABLE;
4493 }
4494
4495 /* Validate segment access rights. Refer to Intel spec. "26.3.1.2 Checks on Guest Segment Registers". */
4496 AssertMsg((u32Access & X86DESCATTR_UNUSABLE) || (u32Access & X86_SEL_TYPE_ACCESSED),
4497 ("Access bit not set for usable segment. idx=%#x sel=%#x attr %#x\n", idxBase, pSelReg, pSelReg->Attr.u));
4498
4499 rc = VMXWriteVmcs32(idxAccess, u32Access); /* 32-bit guest segment access-rights field. */
4500 AssertRCReturn(rc, rc);
4501 return rc;
4502}
4503
4504
4505/**
4506 * Loads the guest segment registers, GDTR, IDTR, LDTR, (TR, FS and GS bases)
4507 * into the guest-state area in the VMCS.
4508 *
4509 * @returns VBox status code.
4510 * @param pVCpu The cross context virtual CPU structure.
4511 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4512 * out-of-sync. Make sure to update the required fields
4513 * before using them.
4514 *
4515 * @remarks ASSUMES pMixedCtx->cr0 is up to date (strict builds validation).
4516 * @remarks No-long-jump zone!!!
4517 */
4518static int hmR0VmxLoadGuestSegmentRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4519{
4520 int rc = VERR_INTERNAL_ERROR_5;
4521 PVM pVM = pVCpu->CTX_SUFF(pVM);
4522
4523 /*
4524 * Guest Segment registers: CS, SS, DS, ES, FS, GS.
4525 */
4526 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS))
4527 {
4528 /* Save the segment attributes for real-on-v86 mode hack, so we can restore them on VM-exit. */
4529 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4530 {
4531 pVCpu->hm.s.vmx.RealMode.AttrCS.u = pMixedCtx->cs.Attr.u;
4532 pVCpu->hm.s.vmx.RealMode.AttrSS.u = pMixedCtx->ss.Attr.u;
4533 pVCpu->hm.s.vmx.RealMode.AttrDS.u = pMixedCtx->ds.Attr.u;
4534 pVCpu->hm.s.vmx.RealMode.AttrES.u = pMixedCtx->es.Attr.u;
4535 pVCpu->hm.s.vmx.RealMode.AttrFS.u = pMixedCtx->fs.Attr.u;
4536 pVCpu->hm.s.vmx.RealMode.AttrGS.u = pMixedCtx->gs.Attr.u;
4537 }
4538
4539#ifdef VBOX_WITH_REM
4540 if (!pVM->hm.s.vmx.fUnrestrictedGuest)
4541 {
4542 Assert(pVM->hm.s.vmx.pRealModeTSS);
4543 AssertCompile(PGMMODE_REAL < PGMMODE_PROTECTED);
4544 if ( pVCpu->hm.s.vmx.fWasInRealMode
4545 && PGMGetGuestMode(pVCpu) >= PGMMODE_PROTECTED)
4546 {
4547 /* Signal that the recompiler must flush its code-cache as the guest -may- rewrite code it will later execute
4548 in real-mode (e.g. OpenBSD 4.0) */
4549 REMFlushTBs(pVM);
4550 Log4(("Load[%RU32]: Switch to protected mode detected!\n", pVCpu->idCpu));
4551 pVCpu->hm.s.vmx.fWasInRealMode = false;
4552 }
4553 }
4554#endif
4555 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_CS_SEL, VMX_VMCS32_GUEST_CS_LIMIT, VMX_VMCS_GUEST_CS_BASE,
4556 VMX_VMCS32_GUEST_CS_ACCESS_RIGHTS, &pMixedCtx->cs);
4557 AssertRCReturn(rc, rc);
4558 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_SS_SEL, VMX_VMCS32_GUEST_SS_LIMIT, VMX_VMCS_GUEST_SS_BASE,
4559 VMX_VMCS32_GUEST_SS_ACCESS_RIGHTS, &pMixedCtx->ss);
4560 AssertRCReturn(rc, rc);
4561 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_DS_SEL, VMX_VMCS32_GUEST_DS_LIMIT, VMX_VMCS_GUEST_DS_BASE,
4562 VMX_VMCS32_GUEST_DS_ACCESS_RIGHTS, &pMixedCtx->ds);
4563 AssertRCReturn(rc, rc);
4564 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_ES_SEL, VMX_VMCS32_GUEST_ES_LIMIT, VMX_VMCS_GUEST_ES_BASE,
4565 VMX_VMCS32_GUEST_ES_ACCESS_RIGHTS, &pMixedCtx->es);
4566 AssertRCReturn(rc, rc);
4567 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_FS_SEL, VMX_VMCS32_GUEST_FS_LIMIT, VMX_VMCS_GUEST_FS_BASE,
4568 VMX_VMCS32_GUEST_FS_ACCESS_RIGHTS, &pMixedCtx->fs);
4569 AssertRCReturn(rc, rc);
4570 rc = hmR0VmxWriteSegmentReg(pVCpu, VMX_VMCS16_GUEST_GS_SEL, VMX_VMCS32_GUEST_GS_LIMIT, VMX_VMCS_GUEST_GS_BASE,
4571 VMX_VMCS32_GUEST_GS_ACCESS_RIGHTS, &pMixedCtx->gs);
4572 AssertRCReturn(rc, rc);
4573
4574#ifdef VBOX_STRICT
4575 /* Validate. */
4576 hmR0VmxValidateSegmentRegs(pVM, pVCpu, pMixedCtx);
4577#endif
4578
4579 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS);
4580 Log4(("Load[%RU32]: CS=%#RX16 Base=%#RX64 Limit=%#RX32 Attr=%#RX32\n", pVCpu->idCpu, pMixedCtx->cs.Sel,
4581 pMixedCtx->cs.u64Base, pMixedCtx->cs.u32Limit, pMixedCtx->cs.Attr.u));
4582 }
4583
4584 /*
4585 * Guest TR.
4586 */
4587 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_TR))
4588 {
4589 /*
4590 * Real-mode emulation using virtual-8086 mode with CR4.VME. Interrupt redirection is achieved
4591 * using the interrupt redirection bitmap (all bits cleared to let the guest handle INT-n's) in the TSS.
4592 * See hmR3InitFinalizeR0() to see how pRealModeTSS is setup.
4593 */
4594 uint16_t u16Sel = 0;
4595 uint32_t u32Limit = 0;
4596 uint64_t u64Base = 0;
4597 uint32_t u32AccessRights = 0;
4598
4599 if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
4600 {
4601 u16Sel = pMixedCtx->tr.Sel;
4602 u32Limit = pMixedCtx->tr.u32Limit;
4603 u64Base = pMixedCtx->tr.u64Base;
4604 u32AccessRights = pMixedCtx->tr.Attr.u;
4605 }
4606 else
4607 {
4608 Assert(pVM->hm.s.vmx.pRealModeTSS);
4609 Assert(PDMVmmDevHeapIsEnabled(pVM)); /* Guaranteed by HMR3CanExecuteGuest() -XXX- what about inner loop changes? */
4610
4611 /* We obtain it here every time as PCI regions could be reconfigured in the guest, changing the VMMDev base. */
4612 RTGCPHYS GCPhys;
4613 rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
4614 AssertRCReturn(rc, rc);
4615
4616 X86DESCATTR DescAttr;
4617 DescAttr.u = 0;
4618 DescAttr.n.u1Present = 1;
4619 DescAttr.n.u4Type = X86_SEL_TYPE_SYS_386_TSS_BUSY;
4620
4621 u16Sel = 0;
4622 u32Limit = HM_VTX_TSS_SIZE;
4623 u64Base = GCPhys; /* in real-mode phys = virt. */
4624 u32AccessRights = DescAttr.u;
4625 }
4626
4627 /* Validate. */
4628 Assert(!(u16Sel & RT_BIT(2)));
4629 AssertMsg( (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_386_TSS_BUSY
4630 || (u32AccessRights & 0xf) == X86_SEL_TYPE_SYS_286_TSS_BUSY, ("TSS is not busy!? %#x\n", u32AccessRights));
4631 AssertMsg(!(u32AccessRights & X86DESCATTR_UNUSABLE), ("TR unusable bit is not clear!? %#x\n", u32AccessRights));
4632 Assert(!(u32AccessRights & RT_BIT(4))); /* System MBZ.*/
4633 Assert(u32AccessRights & RT_BIT(7)); /* Present MB1.*/
4634 Assert(!(u32AccessRights & 0xf00)); /* 11:8 MBZ. */
4635 Assert(!(u32AccessRights & 0xfffe0000)); /* 31:17 MBZ. */
4636 Assert( (u32Limit & 0xfff) == 0xfff
4637 || !(u32AccessRights & RT_BIT(15))); /* Granularity MBZ. */
4638 Assert( !(pMixedCtx->tr.u32Limit & 0xfff00000)
4639 || (u32AccessRights & RT_BIT(15))); /* Granularity MB1. */
4640
4641 rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_TR_SEL, u16Sel);
4642 rc |= VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_LIMIT, u32Limit);
4643 rc |= VMXWriteVmcsGstN(VMX_VMCS_GUEST_TR_BASE, u64Base);
4644 rc |= VMXWriteVmcs32(VMX_VMCS32_GUEST_TR_ACCESS_RIGHTS, u32AccessRights);
4645 AssertRCReturn(rc, rc);
4646
4647 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_TR);
4648 Log4(("Load[%RU32]: VMX_VMCS_GUEST_TR_BASE=%#RX64\n", pVCpu->idCpu, u64Base));
4649 }
4650
4651 /*
4652 * Guest GDTR.
4653 */
4654 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_GDTR))
4655 {
4656 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, pMixedCtx->gdtr.cbGdt);
4657 rc |= VMXWriteVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, pMixedCtx->gdtr.pGdt);
4658 AssertRCReturn(rc, rc);
4659
4660 /* Validate. */
4661 Assert(!(pMixedCtx->gdtr.cbGdt & 0xffff0000)); /* Bits 31:16 MBZ. */
4662
4663 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_GDTR);
4664 Log4(("Load[%RU32]: VMX_VMCS_GUEST_GDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->gdtr.pGdt));
4665 }
4666
4667 /*
4668 * Guest LDTR.
4669 */
4670 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LDTR))
4671 {
4672 /* The unusable bit is specific to VT-x, if it's a null selector mark it as an unusable segment. */
4673 uint32_t u32Access = 0;
4674 if (!pMixedCtx->ldtr.Attr.u)
4675 u32Access = X86DESCATTR_UNUSABLE;
4676 else
4677 u32Access = pMixedCtx->ldtr.Attr.u;
4678
4679 rc = VMXWriteVmcs32(VMX_VMCS16_GUEST_LDTR_SEL, pMixedCtx->ldtr.Sel);
4680 rc |= VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_LIMIT, pMixedCtx->ldtr.u32Limit);
4681 rc |= VMXWriteVmcsGstN(VMX_VMCS_GUEST_LDTR_BASE, pMixedCtx->ldtr.u64Base);
4682 rc |= VMXWriteVmcs32(VMX_VMCS32_GUEST_LDTR_ACCESS_RIGHTS, u32Access);
4683 AssertRCReturn(rc, rc);
4684
4685 /* Validate. */
4686 if (!(u32Access & X86DESCATTR_UNUSABLE))
4687 {
4688 Assert(!(pMixedCtx->ldtr.Sel & RT_BIT(2))); /* TI MBZ. */
4689 Assert(pMixedCtx->ldtr.Attr.n.u4Type == 2); /* Type MB2 (LDT). */
4690 Assert(!pMixedCtx->ldtr.Attr.n.u1DescType); /* System MBZ. */
4691 Assert(pMixedCtx->ldtr.Attr.n.u1Present == 1); /* Present MB1. */
4692 Assert(!pMixedCtx->ldtr.Attr.n.u4LimitHigh); /* 11:8 MBZ. */
4693 Assert(!(pMixedCtx->ldtr.Attr.u & 0xfffe0000)); /* 31:17 MBZ. */
4694 Assert( (pMixedCtx->ldtr.u32Limit & 0xfff) == 0xfff
4695 || !pMixedCtx->ldtr.Attr.n.u1Granularity); /* Granularity MBZ. */
4696 Assert( !(pMixedCtx->ldtr.u32Limit & 0xfff00000)
4697 || pMixedCtx->ldtr.Attr.n.u1Granularity); /* Granularity MB1. */
4698 }
4699
4700 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_LDTR);
4701 Log4(("Load[%RU32]: VMX_VMCS_GUEST_LDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->ldtr.u64Base));
4702 }
4703
4704 /*
4705 * Guest IDTR.
4706 */
4707 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_IDTR))
4708 {
4709 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, pMixedCtx->idtr.cbIdt);
4710 rc |= VMXWriteVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, pMixedCtx->idtr.pIdt);
4711 AssertRCReturn(rc, rc);
4712
4713 /* Validate. */
4714 Assert(!(pMixedCtx->idtr.cbIdt & 0xffff0000)); /* Bits 31:16 MBZ. */
4715
4716 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_IDTR);
4717 Log4(("Load[%RU32]: VMX_VMCS_GUEST_IDTR_BASE=%#RX64\n", pVCpu->idCpu, pMixedCtx->idtr.pIdt));
4718 }
4719
4720 return VINF_SUCCESS;
4721}
4722
4723
4724/**
4725 * Loads certain guest MSRs into the VM-entry MSR-load and VM-exit MSR-store
4726 * areas.
4727 *
4728 * These MSRs will automatically be loaded to the host CPU on every successful
4729 * VM-entry and stored from the host CPU on every successful VM-exit. This also
4730 * creates/updates MSR slots for the host MSRs. The actual host MSR values are
4731 * -not- updated here for performance reasons. See hmR0VmxSaveHostMsrs().
4732 *
4733 * Also loads the sysenter MSRs into the guest-state area in the VMCS.
4734 *
4735 * @returns VBox status code.
4736 * @param pVCpu The cross context virtual CPU structure.
4737 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4738 * out-of-sync. Make sure to update the required fields
4739 * before using them.
4740 *
4741 * @remarks No-long-jump zone!!!
4742 */
4743static int hmR0VmxLoadGuestMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4744{
4745 AssertPtr(pVCpu);
4746 AssertPtr(pVCpu->hm.s.vmx.pvGuestMsr);
4747
4748 /*
4749 * MSRs that we use the auto-load/store MSR area in the VMCS.
4750 */
4751 PVM pVM = pVCpu->CTX_SUFF(pVM);
4752 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS))
4753 {
4754 /* For 64-bit hosts, we load/restore them lazily, see hmR0VmxLazyLoadGuestMsrs(). */
4755#if HC_ARCH_BITS == 32
4756 if (pVM->hm.s.fAllow64BitGuests)
4757 {
4758 int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_LSTAR, pMixedCtx->msrLSTAR, false, NULL);
4759 rc |= hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K6_STAR, pMixedCtx->msrSTAR, false, NULL);
4760 rc |= hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_SF_MASK, pMixedCtx->msrSFMASK, false, NULL);
4761 rc |= hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_KERNEL_GS_BASE, pMixedCtx->msrKERNELGSBASE, false, NULL);
4762 AssertRCReturn(rc, rc);
4763# ifdef LOG_ENABLED
4764 PVMXAUTOMSR pMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
4765 for (uint32_t i = 0; i < pVCpu->hm.s.vmx.cMsrs; i++, pMsr++)
4766 {
4767 Log4(("Load[%RU32]: MSR[%RU32]: u32Msr=%#RX32 u64Value=%#RX64\n", pVCpu->idCpu, i, pMsr->u32Msr,
4768 pMsr->u64Value));
4769 }
4770# endif
4771 }
4772#endif
4773 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
4774 }
4775
4776 /*
4777 * Guest Sysenter MSRs.
4778 * These flags are only set when MSR-bitmaps are not supported by the CPU and we cause
4779 * VM-exits on WRMSRs for these MSRs.
4780 */
4781 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR))
4782 {
4783 int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, pMixedCtx->SysEnter.cs); AssertRCReturn(rc, rc);
4784 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR);
4785 }
4786
4787 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR))
4788 {
4789 int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, pMixedCtx->SysEnter.eip); AssertRCReturn(rc, rc);
4790 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR);
4791 }
4792
4793 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR))
4794 {
4795 int rc = VMXWriteVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, pMixedCtx->SysEnter.esp); AssertRCReturn(rc, rc);
4796 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR);
4797 }
4798
4799 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_EFER_MSR))
4800 {
4801 if (hmR0VmxShouldSwapEferMsr(pVCpu, pMixedCtx))
4802 {
4803 /*
4804 * If the CPU supports VMCS controls for swapping EFER, use it. Otherwise, we have no option
4805 * but to use the auto-load store MSR area in the VMCS for swapping EFER. See @bugref{7368}.
4806 */
4807 if (pVM->hm.s.vmx.fSupportsVmcsEfer)
4808 {
4809 int rc = VMXWriteVmcs64(VMX_VMCS64_GUEST_EFER_FULL, pMixedCtx->msrEFER);
4810 AssertRCReturn(rc,rc);
4811 Log4(("Load[%RU32]: VMX_VMCS64_GUEST_EFER_FULL=%#RX64\n", pVCpu->idCpu, pMixedCtx->msrEFER));
4812 }
4813 else
4814 {
4815 int rc = hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K6_EFER, pMixedCtx->msrEFER, false /* fUpdateHostMsr */,
4816 NULL /* pfAddedAndUpdated */);
4817 AssertRCReturn(rc, rc);
4818
4819 /* We need to intercept reads too, see @bugref{7386#c16}. */
4820 if (pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
4821 hmR0VmxSetMsrPermission(pVCpu, MSR_K6_EFER, VMXMSREXIT_INTERCEPT_READ, VMXMSREXIT_INTERCEPT_WRITE);
4822 Log4(("Load[%RU32]: MSR[--]: u32Msr=%#RX32 u64Value=%#RX64 cMsrs=%u\n", pVCpu->idCpu, MSR_K6_EFER,
4823 pMixedCtx->msrEFER, pVCpu->hm.s.vmx.cMsrs));
4824 }
4825 }
4826 else if (!pVM->hm.s.vmx.fSupportsVmcsEfer)
4827 hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, MSR_K6_EFER);
4828 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_EFER_MSR);
4829 }
4830
4831 return VINF_SUCCESS;
4832}
4833
4834
4835/**
4836 * Loads the guest activity state into the guest-state area in the VMCS.
4837 *
4838 * @returns VBox status code.
4839 * @param pVCpu The cross context virtual CPU structure.
4840 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4841 * out-of-sync. Make sure to update the required fields
4842 * before using them.
4843 *
4844 * @remarks No-long-jump zone!!!
4845 */
4846static int hmR0VmxLoadGuestActivityState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4847{
4848 NOREF(pMixedCtx);
4849 /** @todo See if we can make use of other states, e.g.
4850 * VMX_VMCS_GUEST_ACTIVITY_SHUTDOWN or HLT. */
4851 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_VMX_GUEST_ACTIVITY_STATE))
4852 {
4853 int rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, VMX_VMCS_GUEST_ACTIVITY_ACTIVE);
4854 AssertRCReturn(rc, rc);
4855
4856 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_VMX_GUEST_ACTIVITY_STATE);
4857 }
4858 return VINF_SUCCESS;
4859}
4860
4861
4862#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
4863/**
4864 * Check if guest state allows safe use of 32-bit switcher again.
4865 *
4866 * Segment bases and protected mode structures must be 32-bit addressable
4867 * because the 32-bit switcher will ignore high dword when writing these VMCS
4868 * fields. See @bugref{8432} for details.
4869 *
4870 * @returns true if safe, false if must continue to use the 64-bit switcher.
4871 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4872 * out-of-sync. Make sure to update the required fields
4873 * before using them.
4874 *
4875 * @remarks No-long-jump zone!!!
4876 */
4877static bool hmR0VmxIs32BitSwitcherSafe(PCPUMCTX pMixedCtx)
4878{
4879 if (pMixedCtx->gdtr.pGdt & UINT64_C(0xffffffff00000000))
4880 return false;
4881 if (pMixedCtx->idtr.pIdt & UINT64_C(0xffffffff00000000))
4882 return false;
4883 if (pMixedCtx->ldtr.u64Base & UINT64_C(0xffffffff00000000))
4884 return false;
4885 if (pMixedCtx->tr.u64Base & UINT64_C(0xffffffff00000000))
4886 return false;
4887 if (pMixedCtx->es.u64Base & UINT64_C(0xffffffff00000000))
4888 return false;
4889 if (pMixedCtx->cs.u64Base & UINT64_C(0xffffffff00000000))
4890 return false;
4891 if (pMixedCtx->ss.u64Base & UINT64_C(0xffffffff00000000))
4892 return false;
4893 if (pMixedCtx->ds.u64Base & UINT64_C(0xffffffff00000000))
4894 return false;
4895 if (pMixedCtx->fs.u64Base & UINT64_C(0xffffffff00000000))
4896 return false;
4897 if (pMixedCtx->gs.u64Base & UINT64_C(0xffffffff00000000))
4898 return false;
4899 /* All good, bases are 32-bit. */
4900 return true;
4901}
4902#endif
4903
4904
4905/**
4906 * Sets up the appropriate function to run guest code.
4907 *
4908 * @returns VBox status code.
4909 * @param pVCpu The cross context virtual CPU structure.
4910 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
4911 * out-of-sync. Make sure to update the required fields
4912 * before using them.
4913 *
4914 * @remarks No-long-jump zone!!!
4915 */
4916static int hmR0VmxSetupVMRunHandler(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
4917{
4918 if (CPUMIsGuestInLongModeEx(pMixedCtx))
4919 {
4920#ifndef VBOX_ENABLE_64_BITS_GUESTS
4921 return VERR_PGM_UNSUPPORTED_SHADOW_PAGING_MODE;
4922#endif
4923 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests); /* Guaranteed by hmR3InitFinalizeR0(). */
4924#if HC_ARCH_BITS == 32
4925 /* 32-bit host. We need to switch to 64-bit before running the 64-bit guest. */
4926 if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0SwitcherStartVM64)
4927 {
4928 if (pVCpu->hm.s.vmx.pfnStartVM != NULL) /* Very first entry would have saved host-state already, ignore it. */
4929 {
4930 /* Currently, all mode changes sends us back to ring-3, so these should be set. See @bugref{6944}. */
4931 AssertMsg(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_VMX_EXIT_CTLS
4932 | HM_CHANGED_VMX_ENTRY_CTLS
4933 | HM_CHANGED_GUEST_EFER_MSR), ("flags=%#x\n", HMCPU_CF_VALUE(pVCpu)));
4934 }
4935 pVCpu->hm.s.vmx.pfnStartVM = VMXR0SwitcherStartVM64;
4936
4937 /* Mark that we've switched to 64-bit handler, we can't safely switch back to 32-bit for
4938 the rest of the VM run (until VM reset). See @bugref{8432#c7}. */
4939 pVCpu->hm.s.vmx.fSwitchedTo64on32 = true;
4940 Log4(("Load[%RU32]: hmR0VmxSetupVMRunHandler: selected 64-bit switcher\n", pVCpu->idCpu));
4941 }
4942#else
4943 /* 64-bit host. */
4944 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM64;
4945#endif
4946 }
4947 else
4948 {
4949 /* Guest is not in long mode, use the 32-bit handler. */
4950#if HC_ARCH_BITS == 32
4951 if ( pVCpu->hm.s.vmx.pfnStartVM != VMXR0StartVM32
4952 && !pVCpu->hm.s.vmx.fSwitchedTo64on32 /* If set, guest mode change does not imply switcher change. */
4953 && pVCpu->hm.s.vmx.pfnStartVM != NULL) /* Very first entry would have saved host-state already, ignore it. */
4954 {
4955 /* Currently, all mode changes sends us back to ring-3, so these should be set. See @bugref{6944}. */
4956 AssertMsg(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_VMX_EXIT_CTLS
4957 | HM_CHANGED_VMX_ENTRY_CTLS
4958 | HM_CHANGED_GUEST_EFER_MSR), ("flags=%#x\n", HMCPU_CF_VALUE(pVCpu)));
4959 }
4960# ifdef VBOX_ENABLE_64_BITS_GUESTS
4961 /*
4962 * Keep using the 64-bit switcher even though we're in 32-bit because of bad Intel design, see @bugref{8432#c7}.
4963 * If real-on-v86 mode is active, clear the 64-bit switcher flag because now we know the guest is in a sane
4964 * state where it's safe to use the 32-bit switcher. Otherwise check the guest state if it's safe to use
4965 * the much faster 32-bit switcher again.
4966 */
4967 if (!pVCpu->hm.s.vmx.fSwitchedTo64on32)
4968 {
4969 if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0StartVM32)
4970 Log4(("Load[%RU32]: hmR0VmxSetupVMRunHandler: selected 32-bit switcher\n", pVCpu->idCpu));
4971 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4972 }
4973 else
4974 {
4975 Assert(pVCpu->hm.s.vmx.pfnStartVM == VMXR0SwitcherStartVM64);
4976 if ( pVCpu->hm.s.vmx.RealMode.fRealOnV86Active
4977 || hmR0VmxIs32BitSwitcherSafe(pMixedCtx))
4978 {
4979 pVCpu->hm.s.vmx.fSwitchedTo64on32 = false;
4980 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4981 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_EFER_MSR
4982 | HM_CHANGED_VMX_ENTRY_CTLS
4983 | HM_CHANGED_VMX_EXIT_CTLS
4984 | HM_CHANGED_HOST_CONTEXT);
4985 Log4(("Load[%RU32]: hmR0VmxSetupVMRunHandler: selected 32-bit switcher (safe)\n", pVCpu->idCpu));
4986 }
4987 }
4988# else
4989 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4990# endif
4991#else
4992 pVCpu->hm.s.vmx.pfnStartVM = VMXR0StartVM32;
4993#endif
4994 }
4995 Assert(pVCpu->hm.s.vmx.pfnStartVM);
4996 return VINF_SUCCESS;
4997}
4998
4999
5000/**
5001 * Wrapper for running the guest code in VT-x.
5002 *
5003 * @returns VBox status code, no informational status codes.
5004 * @param pVM The cross context VM structure.
5005 * @param pVCpu The cross context virtual CPU structure.
5006 * @param pCtx Pointer to the guest-CPU context.
5007 *
5008 * @remarks No-long-jump zone!!!
5009 */
5010DECLINLINE(int) hmR0VmxRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
5011{
5012 /*
5013 * 64-bit Windows uses XMM registers in the kernel as the Microsoft compiler expresses floating-point operations
5014 * using SSE instructions. Some XMM registers (XMM6-XMM15) are callee-saved and thus the need for this XMM wrapper.
5015 * Refer MSDN docs. "Configuring Programs for 64-bit / x64 Software Conventions / Register Usage" for details.
5016 */
5017 bool const fResumeVM = RT_BOOL(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_LAUNCHED);
5018 /** @todo Add stats for resume vs launch. */
5019#ifdef VBOX_WITH_KERNEL_USING_XMM
5020 int rc = HMR0VMXStartVMWrapXMM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu, pVCpu->hm.s.vmx.pfnStartVM);
5021#else
5022 int rc = pVCpu->hm.s.vmx.pfnStartVM(fResumeVM, pCtx, &pVCpu->hm.s.vmx.VMCSCache, pVM, pVCpu);
5023#endif
5024 AssertMsg(rc <= VINF_SUCCESS, ("%Rrc\n", rc));
5025 return rc;
5026}
5027
5028
5029/**
5030 * Reports world-switch error and dumps some useful debug info.
5031 *
5032 * @param pVM The cross context VM structure.
5033 * @param pVCpu The cross context virtual CPU structure.
5034 * @param rcVMRun The return code from VMLAUNCH/VMRESUME.
5035 * @param pCtx Pointer to the guest-CPU context.
5036 * @param pVmxTransient Pointer to the VMX transient structure (only
5037 * exitReason updated).
5038 */
5039static void hmR0VmxReportWorldSwitchError(PVM pVM, PVMCPU pVCpu, int rcVMRun, PCPUMCTX pCtx, PVMXTRANSIENT pVmxTransient)
5040{
5041 Assert(pVM);
5042 Assert(pVCpu);
5043 Assert(pCtx);
5044 Assert(pVmxTransient);
5045 HMVMX_ASSERT_PREEMPT_SAFE();
5046
5047 Log4(("VM-entry failure: %Rrc\n", rcVMRun));
5048 switch (rcVMRun)
5049 {
5050 case VERR_VMX_INVALID_VMXON_PTR:
5051 AssertFailed();
5052 break;
5053 case VINF_SUCCESS: /* VMLAUNCH/VMRESUME succeeded but VM-entry failed... yeah, true story. */
5054 case VERR_VMX_UNABLE_TO_START_VM: /* VMLAUNCH/VMRESUME itself failed. */
5055 {
5056 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &pVCpu->hm.s.vmx.LastError.u32ExitReason);
5057 rc |= VMXReadVmcs32(VMX_VMCS32_RO_VM_INSTR_ERROR, &pVCpu->hm.s.vmx.LastError.u32InstrError);
5058 rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
5059 AssertRC(rc);
5060
5061 pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
5062 /* LastError.idCurrentCpu was already updated in hmR0VmxPreRunGuestCommitted().
5063 Cannot do it here as we may have been long preempted. */
5064
5065#ifdef VBOX_STRICT
5066 Log4(("uExitReason %#RX32 (VmxTransient %#RX16)\n", pVCpu->hm.s.vmx.LastError.u32ExitReason,
5067 pVmxTransient->uExitReason));
5068 Log4(("Exit Qualification %#RX64\n", pVmxTransient->uExitQualification));
5069 Log4(("InstrError %#RX32\n", pVCpu->hm.s.vmx.LastError.u32InstrError));
5070 if (pVCpu->hm.s.vmx.LastError.u32InstrError <= HMVMX_INSTR_ERROR_MAX)
5071 Log4(("InstrError Desc. \"%s\"\n", g_apszVmxInstrErrors[pVCpu->hm.s.vmx.LastError.u32InstrError]));
5072 else
5073 Log4(("InstrError Desc. Range exceeded %u\n", HMVMX_INSTR_ERROR_MAX));
5074 Log4(("Entered host CPU %u\n", pVCpu->hm.s.vmx.LastError.idEnteredCpu));
5075 Log4(("Current host CPU %u\n", pVCpu->hm.s.vmx.LastError.idCurrentCpu));
5076
5077 /* VMX control bits. */
5078 uint32_t u32Val;
5079 uint64_t u64Val;
5080 RTHCUINTREG uHCReg;
5081 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PIN_EXEC, &u32Val); AssertRC(rc);
5082 Log4(("VMX_VMCS32_CTRL_PIN_EXEC %#RX32\n", u32Val));
5083 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, &u32Val); AssertRC(rc);
5084 Log4(("VMX_VMCS32_CTRL_PROC_EXEC %#RX32\n", u32Val));
5085 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL)
5086 {
5087 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, &u32Val); AssertRC(rc);
5088 Log4(("VMX_VMCS32_CTRL_PROC_EXEC2 %#RX32\n", u32Val));
5089 }
5090 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val); AssertRC(rc);
5091 Log4(("VMX_VMCS32_CTRL_ENTRY %#RX32\n", u32Val));
5092 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT, &u32Val); AssertRC(rc);
5093 Log4(("VMX_VMCS32_CTRL_EXIT %#RX32\n", u32Val));
5094 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_CR3_TARGET_COUNT, &u32Val); AssertRC(rc);
5095 Log4(("VMX_VMCS32_CTRL_CR3_TARGET_COUNT %#RX32\n", u32Val));
5096 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32Val); AssertRC(rc);
5097 Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", u32Val));
5098 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, &u32Val); AssertRC(rc);
5099 Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", u32Val));
5100 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, &u32Val); AssertRC(rc);
5101 Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %u\n", u32Val));
5102 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_TPR_THRESHOLD, &u32Val); AssertRC(rc);
5103 Log4(("VMX_VMCS32_CTRL_TPR_THRESHOLD %u\n", u32Val));
5104 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT, &u32Val); AssertRC(rc);
5105 Log4(("VMX_VMCS32_CTRL_EXIT_MSR_STORE_COUNT %u (guest MSRs)\n", u32Val));
5106 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT, &u32Val); AssertRC(rc);
5107 Log4(("VMX_VMCS32_CTRL_EXIT_MSR_LOAD_COUNT %u (host MSRs)\n", u32Val));
5108 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT, &u32Val); AssertRC(rc);
5109 Log4(("VMX_VMCS32_CTRL_ENTRY_MSR_LOAD_COUNT %u (guest MSRs)\n", u32Val));
5110 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, &u32Val); AssertRC(rc);
5111 Log4(("VMX_VMCS32_CTRL_EXCEPTION_BITMAP %#RX32\n", u32Val));
5112 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK, &u32Val); AssertRC(rc);
5113 Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MASK %#RX32\n", u32Val));
5114 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH, &u32Val); AssertRC(rc);
5115 Log4(("VMX_VMCS32_CTRL_PAGEFAULT_ERROR_MATCH %#RX32\n", u32Val));
5116 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc);
5117 Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg));
5118 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc);
5119 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
5120 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc);
5121 Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg));
5122 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc);
5123 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
5124 if (pVM->hm.s.fNestedPaging)
5125 {
5126 rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
5127 Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
5128 }
5129
5130 /* Guest bits. */
5131 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val); AssertRC(rc);
5132 Log4(("Old Guest Rip %#RX64 New %#RX64\n", pCtx->rip, u64Val));
5133 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val); AssertRC(rc);
5134 Log4(("Old Guest Rsp %#RX64 New %#RX64\n", pCtx->rsp, u64Val));
5135 rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Val); AssertRC(rc);
5136 Log4(("Old Guest Rflags %#RX32 New %#RX32\n", pCtx->eflags.u32, u32Val));
5137 if (pVM->hm.s.vmx.fVpid)
5138 {
5139 rc = VMXReadVmcs32(VMX_VMCS16_VPID, &u32Val); AssertRC(rc);
5140 Log4(("VMX_VMCS16_VPID %u\n", u32Val));
5141 }
5142
5143 /* Host bits. */
5144 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR0, &uHCReg); AssertRC(rc);
5145 Log4(("Host CR0 %#RHr\n", uHCReg));
5146 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR3, &uHCReg); AssertRC(rc);
5147 Log4(("Host CR3 %#RHr\n", uHCReg));
5148 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_CR4, &uHCReg); AssertRC(rc);
5149 Log4(("Host CR4 %#RHr\n", uHCReg));
5150
5151 RTGDTR HostGdtr;
5152 PCX86DESCHC pDesc;
5153 ASMGetGDTR(&HostGdtr);
5154 rc = VMXReadVmcs32(VMX_VMCS16_HOST_CS_SEL, &u32Val); AssertRC(rc);
5155 Log4(("Host CS %#08x\n", u32Val));
5156 if (u32Val < HostGdtr.cbGdt)
5157 {
5158 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5159 HMR0DumpDescriptor(pDesc, u32Val, "CS: ");
5160 }
5161
5162 rc = VMXReadVmcs32(VMX_VMCS16_HOST_DS_SEL, &u32Val); AssertRC(rc);
5163 Log4(("Host DS %#08x\n", u32Val));
5164 if (u32Val < HostGdtr.cbGdt)
5165 {
5166 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5167 HMR0DumpDescriptor(pDesc, u32Val, "DS: ");
5168 }
5169
5170 rc = VMXReadVmcs32(VMX_VMCS16_HOST_ES_SEL, &u32Val); AssertRC(rc);
5171 Log4(("Host ES %#08x\n", u32Val));
5172 if (u32Val < HostGdtr.cbGdt)
5173 {
5174 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5175 HMR0DumpDescriptor(pDesc, u32Val, "ES: ");
5176 }
5177
5178 rc = VMXReadVmcs32(VMX_VMCS16_HOST_FS_SEL, &u32Val); AssertRC(rc);
5179 Log4(("Host FS %#08x\n", u32Val));
5180 if (u32Val < HostGdtr.cbGdt)
5181 {
5182 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5183 HMR0DumpDescriptor(pDesc, u32Val, "FS: ");
5184 }
5185
5186 rc = VMXReadVmcs32(VMX_VMCS16_HOST_GS_SEL, &u32Val); AssertRC(rc);
5187 Log4(("Host GS %#08x\n", u32Val));
5188 if (u32Val < HostGdtr.cbGdt)
5189 {
5190 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5191 HMR0DumpDescriptor(pDesc, u32Val, "GS: ");
5192 }
5193
5194 rc = VMXReadVmcs32(VMX_VMCS16_HOST_SS_SEL, &u32Val); AssertRC(rc);
5195 Log4(("Host SS %#08x\n", u32Val));
5196 if (u32Val < HostGdtr.cbGdt)
5197 {
5198 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5199 HMR0DumpDescriptor(pDesc, u32Val, "SS: ");
5200 }
5201
5202 rc = VMXReadVmcs32(VMX_VMCS16_HOST_TR_SEL, &u32Val); AssertRC(rc);
5203 Log4(("Host TR %#08x\n", u32Val));
5204 if (u32Val < HostGdtr.cbGdt)
5205 {
5206 pDesc = (PCX86DESCHC)(HostGdtr.pGdt + (u32Val & X86_SEL_MASK));
5207 HMR0DumpDescriptor(pDesc, u32Val, "TR: ");
5208 }
5209
5210 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_TR_BASE, &uHCReg); AssertRC(rc);
5211 Log4(("Host TR Base %#RHv\n", uHCReg));
5212 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_GDTR_BASE, &uHCReg); AssertRC(rc);
5213 Log4(("Host GDTR Base %#RHv\n", uHCReg));
5214 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_IDTR_BASE, &uHCReg); AssertRC(rc);
5215 Log4(("Host IDTR Base %#RHv\n", uHCReg));
5216 rc = VMXReadVmcs32(VMX_VMCS32_HOST_SYSENTER_CS, &u32Val); AssertRC(rc);
5217 Log4(("Host SYSENTER CS %#08x\n", u32Val));
5218 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_EIP, &uHCReg); AssertRC(rc);
5219 Log4(("Host SYSENTER EIP %#RHv\n", uHCReg));
5220 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_SYSENTER_ESP, &uHCReg); AssertRC(rc);
5221 Log4(("Host SYSENTER ESP %#RHv\n", uHCReg));
5222 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RSP, &uHCReg); AssertRC(rc);
5223 Log4(("Host RSP %#RHv\n", uHCReg));
5224 rc = VMXReadVmcsHstN(VMX_VMCS_HOST_RIP, &uHCReg); AssertRC(rc);
5225 Log4(("Host RIP %#RHv\n", uHCReg));
5226# if HC_ARCH_BITS == 64
5227 Log4(("MSR_K6_EFER = %#RX64\n", ASMRdMsr(MSR_K6_EFER)));
5228 Log4(("MSR_K8_CSTAR = %#RX64\n", ASMRdMsr(MSR_K8_CSTAR)));
5229 Log4(("MSR_K8_LSTAR = %#RX64\n", ASMRdMsr(MSR_K8_LSTAR)));
5230 Log4(("MSR_K6_STAR = %#RX64\n", ASMRdMsr(MSR_K6_STAR)));
5231 Log4(("MSR_K8_SF_MASK = %#RX64\n", ASMRdMsr(MSR_K8_SF_MASK)));
5232 Log4(("MSR_K8_KERNEL_GS_BASE = %#RX64\n", ASMRdMsr(MSR_K8_KERNEL_GS_BASE)));
5233# endif
5234#endif /* VBOX_STRICT */
5235 break;
5236 }
5237
5238 default:
5239 /* Impossible */
5240 AssertMsgFailed(("hmR0VmxReportWorldSwitchError %Rrc (%#x)\n", rcVMRun, rcVMRun));
5241 break;
5242 }
5243 NOREF(pVM); NOREF(pCtx);
5244}
5245
5246
5247#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
5248#ifndef VMX_USE_CACHED_VMCS_ACCESSES
5249# error "VMX_USE_CACHED_VMCS_ACCESSES not defined when it should be!"
5250#endif
5251#ifdef VBOX_STRICT
5252static bool hmR0VmxIsValidWriteField(uint32_t idxField)
5253{
5254 switch (idxField)
5255 {
5256 case VMX_VMCS_GUEST_RIP:
5257 case VMX_VMCS_GUEST_RSP:
5258 case VMX_VMCS_GUEST_SYSENTER_EIP:
5259 case VMX_VMCS_GUEST_SYSENTER_ESP:
5260 case VMX_VMCS_GUEST_GDTR_BASE:
5261 case VMX_VMCS_GUEST_IDTR_BASE:
5262 case VMX_VMCS_GUEST_CS_BASE:
5263 case VMX_VMCS_GUEST_DS_BASE:
5264 case VMX_VMCS_GUEST_ES_BASE:
5265 case VMX_VMCS_GUEST_FS_BASE:
5266 case VMX_VMCS_GUEST_GS_BASE:
5267 case VMX_VMCS_GUEST_SS_BASE:
5268 case VMX_VMCS_GUEST_LDTR_BASE:
5269 case VMX_VMCS_GUEST_TR_BASE:
5270 case VMX_VMCS_GUEST_CR3:
5271 return true;
5272 }
5273 return false;
5274}
5275
5276static bool hmR0VmxIsValidReadField(uint32_t idxField)
5277{
5278 switch (idxField)
5279 {
5280 /* Read-only fields. */
5281 case VMX_VMCS_RO_EXIT_QUALIFICATION:
5282 return true;
5283 }
5284 /* Remaining readable fields should also be writable. */
5285 return hmR0VmxIsValidWriteField(idxField);
5286}
5287#endif /* VBOX_STRICT */
5288
5289
5290/**
5291 * Executes the specified handler in 64-bit mode.
5292 *
5293 * @returns VBox status code (no informational status codes).
5294 * @param pVM The cross context VM structure.
5295 * @param pVCpu The cross context virtual CPU structure.
5296 * @param pCtx Pointer to the guest CPU context.
5297 * @param enmOp The operation to perform.
5298 * @param cParams Number of parameters.
5299 * @param paParam Array of 32-bit parameters.
5300 */
5301VMMR0DECL(int) VMXR0Execute64BitsHandler(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx, HM64ON32OP enmOp,
5302 uint32_t cParams, uint32_t *paParam)
5303{
5304 NOREF(pCtx);
5305
5306 AssertReturn(pVM->hm.s.pfnHost32ToGuest64R0, VERR_HM_NO_32_TO_64_SWITCHER);
5307 Assert(enmOp > HM64ON32OP_INVALID && enmOp < HM64ON32OP_END);
5308 Assert(pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Write.aField));
5309 Assert(pVCpu->hm.s.vmx.VMCSCache.Read.cValidEntries <= RT_ELEMENTS(pVCpu->hm.s.vmx.VMCSCache.Read.aField));
5310
5311#ifdef VBOX_STRICT
5312 for (uint32_t i = 0; i < pVCpu->hm.s.vmx.VMCSCache.Write.cValidEntries; i++)
5313 Assert(hmR0VmxIsValidWriteField(pVCpu->hm.s.vmx.VMCSCache.Write.aField[i]));
5314
5315 for (uint32_t i = 0; i <pVCpu->hm.s.vmx.VMCSCache.Read.cValidEntries; i++)
5316 Assert(hmR0VmxIsValidReadField(pVCpu->hm.s.vmx.VMCSCache.Read.aField[i]));
5317#endif
5318
5319 /* Disable interrupts. */
5320 RTCCUINTREG fOldEFlags = ASMIntDisableFlags();
5321
5322#ifdef VBOX_WITH_VMMR0_DISABLE_LAPIC_NMI
5323 RTCPUID idHostCpu = RTMpCpuId();
5324 CPUMR0SetLApic(pVCpu, idHostCpu);
5325#endif
5326
5327 PHMGLOBALCPUINFO pCpu = HMR0GetCurrentCpu();
5328 RTHCPHYS HCPhysCpuPage = pCpu->HCPhysMemObj;
5329
5330 /* Clear VMCS. Marking it inactive, clearing implementation-specific data and writing VMCS data back to memory. */
5331 VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
5332 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
5333
5334 /* Leave VMX Root Mode. */
5335 VMXDisable();
5336
5337 SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
5338
5339 CPUMSetHyperESP(pVCpu, VMMGetStackRC(pVCpu));
5340 CPUMSetHyperEIP(pVCpu, enmOp);
5341 for (int i = (int)cParams - 1; i >= 0; i--)
5342 CPUMPushHyper(pVCpu, paParam[i]);
5343
5344 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatWorldSwitch3264, z);
5345
5346 /* Call the switcher. */
5347 int rc = pVM->hm.s.pfnHost32ToGuest64R0(pVM, RT_OFFSETOF(VM, aCpus[pVCpu->idCpu].cpum) - RT_OFFSETOF(VM, cpum));
5348 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatWorldSwitch3264, z);
5349
5350 /** @todo replace with hmR0VmxEnterRootMode() and hmR0VmxLeaveRootMode(). */
5351 /* Make sure the VMX instructions don't cause #UD faults. */
5352 SUPR0ChangeCR4(X86_CR4_VMXE, RTCCUINTREG_MAX);
5353
5354 /* Re-enter VMX Root Mode */
5355 int rc2 = VMXEnable(HCPhysCpuPage);
5356 if (RT_FAILURE(rc2))
5357 {
5358 SUPR0ChangeCR4(0, ~X86_CR4_VMXE);
5359 ASMSetFlags(fOldEFlags);
5360 pVM->hm.s.vmx.HCPhysVmxEnableError = HCPhysCpuPage;
5361 return rc2;
5362 }
5363
5364 rc2 = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
5365 AssertRC(rc2);
5366 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE;
5367 Assert(!(ASMGetFlags() & X86_EFL_IF));
5368 ASMSetFlags(fOldEFlags);
5369 return rc;
5370}
5371
5372
5373/**
5374 * Prepares for and executes VMLAUNCH (64-bit guests) for 32-bit hosts
5375 * supporting 64-bit guests.
5376 *
5377 * @returns VBox status code.
5378 * @param fResume Whether to VMLAUNCH or VMRESUME.
5379 * @param pCtx Pointer to the guest-CPU context.
5380 * @param pCache Pointer to the VMCS cache.
5381 * @param pVM The cross context VM structure.
5382 * @param pVCpu The cross context virtual CPU structure.
5383 */
5384DECLASM(int) VMXR0SwitcherStartVM64(RTHCUINT fResume, PCPUMCTX pCtx, PVMCSCACHE pCache, PVM pVM, PVMCPU pVCpu)
5385{
5386 NOREF(fResume);
5387
5388 PHMGLOBALCPUINFO pCpu = HMR0GetCurrentCpu();
5389 RTHCPHYS HCPhysCpuPage = pCpu->HCPhysMemObj;
5390
5391#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5392 pCache->uPos = 1;
5393 pCache->interPD = PGMGetInterPaeCR3(pVM);
5394 pCache->pSwitcher = (uint64_t)pVM->hm.s.pfnHost32ToGuest64R0;
5395#endif
5396
5397#if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES)
5398 pCache->TestIn.HCPhysCpuPage = 0;
5399 pCache->TestIn.HCPhysVmcs = 0;
5400 pCache->TestIn.pCache = 0;
5401 pCache->TestOut.HCPhysVmcs = 0;
5402 pCache->TestOut.pCache = 0;
5403 pCache->TestOut.pCtx = 0;
5404 pCache->TestOut.eflags = 0;
5405#else
5406 NOREF(pCache);
5407#endif
5408
5409 uint32_t aParam[10];
5410 aParam[0] = (uint32_t)(HCPhysCpuPage); /* Param 1: VMXON physical address - Lo. */
5411 aParam[1] = (uint32_t)(HCPhysCpuPage >> 32); /* Param 1: VMXON physical address - Hi. */
5412 aParam[2] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs); /* Param 2: VMCS physical address - Lo. */
5413 aParam[3] = (uint32_t)(pVCpu->hm.s.vmx.HCPhysVmcs >> 32); /* Param 2: VMCS physical address - Hi. */
5414 aParam[4] = VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache);
5415 aParam[5] = 0;
5416 aParam[6] = VM_RC_ADDR(pVM, pVM);
5417 aParam[7] = 0;
5418 aParam[8] = VM_RC_ADDR(pVM, pVCpu);
5419 aParam[9] = 0;
5420
5421#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5422 pCtx->dr[4] = pVM->hm.s.vmx.pScratchPhys + 16 + 8;
5423 *(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) = 1;
5424#endif
5425 int rc = VMXR0Execute64BitsHandler(pVM, pVCpu, pCtx, HM64ON32OP_VMXRCStartVM64, RT_ELEMENTS(aParam), &aParam[0]);
5426
5427#ifdef VBOX_WITH_CRASHDUMP_MAGIC
5428 Assert(*(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) == 5);
5429 Assert(pCtx->dr[4] == 10);
5430 *(uint32_t *)(pVM->hm.s.vmx.pScratch + 16 + 8) = 0xff;
5431#endif
5432
5433#if defined(DEBUG) && defined(VMX_USE_CACHED_VMCS_ACCESSES)
5434 AssertMsg(pCache->TestIn.HCPhysCpuPage == HCPhysCpuPage, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysCpuPage, HCPhysCpuPage));
5435 AssertMsg(pCache->TestIn.HCPhysVmcs == pVCpu->hm.s.vmx.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
5436 pVCpu->hm.s.vmx.HCPhysVmcs));
5437 AssertMsg(pCache->TestIn.HCPhysVmcs == pCache->TestOut.HCPhysVmcs, ("%RHp vs %RHp\n", pCache->TestIn.HCPhysVmcs,
5438 pCache->TestOut.HCPhysVmcs));
5439 AssertMsg(pCache->TestIn.pCache == pCache->TestOut.pCache, ("%RGv vs %RGv\n", pCache->TestIn.pCache,
5440 pCache->TestOut.pCache));
5441 AssertMsg(pCache->TestIn.pCache == VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache),
5442 ("%RGv vs %RGv\n", pCache->TestIn.pCache, VM_RC_ADDR(pVM, &pVM->aCpus[pVCpu->idCpu].hm.s.vmx.VMCSCache)));
5443 AssertMsg(pCache->TestIn.pCtx == pCache->TestOut.pCtx, ("%RGv vs %RGv\n", pCache->TestIn.pCtx,
5444 pCache->TestOut.pCtx));
5445 Assert(!(pCache->TestOut.eflags & X86_EFL_IF));
5446#endif
5447 return rc;
5448}
5449
5450
5451/**
5452 * Initialize the VMCS-Read cache.
5453 *
5454 * The VMCS cache is used for 32-bit hosts running 64-bit guests (except 32-bit
5455 * Darwin which runs with 64-bit paging in 32-bit mode) for 64-bit fields that
5456 * cannot be accessed in 32-bit mode. Some 64-bit fields -can- be accessed
5457 * (those that have a 32-bit FULL & HIGH part).
5458 *
5459 * @returns VBox status code.
5460 * @param pVM The cross context VM structure.
5461 * @param pVCpu The cross context virtual CPU structure.
5462 */
5463static int hmR0VmxInitVmcsReadCache(PVM pVM, PVMCPU pVCpu)
5464{
5465#define VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, idxField) \
5466{ \
5467 Assert(pCache->Read.aField[idxField##_CACHE_IDX] == 0); \
5468 pCache->Read.aField[idxField##_CACHE_IDX] = idxField; \
5469 pCache->Read.aFieldVal[idxField##_CACHE_IDX] = 0; \
5470 ++cReadFields; \
5471}
5472
5473 AssertPtr(pVM);
5474 AssertPtr(pVCpu);
5475 PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache;
5476 uint32_t cReadFields = 0;
5477
5478 /*
5479 * Don't remove the #if 0'd fields in this code. They're listed here for consistency
5480 * and serve to indicate exceptions to the rules.
5481 */
5482
5483 /* Guest-natural selector base fields. */
5484#if 0
5485 /* These are 32-bit in practice. See Intel spec. 2.5 "Control Registers". */
5486 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR0);
5487 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR4);
5488#endif
5489 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_ES_BASE);
5490 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CS_BASE);
5491 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SS_BASE);
5492 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_DS_BASE);
5493 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_FS_BASE);
5494 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GS_BASE);
5495 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_LDTR_BASE);
5496 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_TR_BASE);
5497 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_GDTR_BASE);
5498 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_IDTR_BASE);
5499 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RSP);
5500 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_RIP);
5501#if 0
5502 /* Unused natural width guest-state fields. */
5503 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS);
5504 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3); /* Handled in Nested Paging case */
5505#endif
5506 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_ESP);
5507 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_SYSENTER_EIP);
5508
5509 /* 64-bit guest-state fields; unused as we use two 32-bit VMREADs for these 64-bit fields (using "FULL" and "HIGH" fields). */
5510#if 0
5511 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL);
5512 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_DEBUGCTL_FULL);
5513 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PAT_FULL);
5514 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_EFER_FULL);
5515 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL);
5516 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE0_FULL);
5517 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE1_FULL);
5518 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE2_FULL);
5519 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS64_GUEST_PDPTE3_FULL);
5520#endif
5521
5522 /* Natural width guest-state fields. */
5523 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_QUALIFICATION);
5524#if 0
5525 /* Currently unused field. */
5526 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_RO_EXIT_GUEST_LINEAR_ADDR);
5527#endif
5528
5529 if (pVM->hm.s.fNestedPaging)
5530 {
5531 VMXLOCAL_INIT_READ_CACHE_FIELD(pCache, VMX_VMCS_GUEST_CR3);
5532 AssertMsg(cReadFields == VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields,
5533 VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX));
5534 pCache->Read.cValidEntries = VMX_VMCS_MAX_NESTED_PAGING_CACHE_IDX;
5535 }
5536 else
5537 {
5538 AssertMsg(cReadFields == VMX_VMCS_MAX_CACHE_IDX, ("cReadFields=%u expected %u\n", cReadFields, VMX_VMCS_MAX_CACHE_IDX));
5539 pCache->Read.cValidEntries = VMX_VMCS_MAX_CACHE_IDX;
5540 }
5541
5542#undef VMXLOCAL_INIT_READ_CACHE_FIELD
5543 return VINF_SUCCESS;
5544}
5545
5546
5547/**
5548 * Writes a field into the VMCS. This can either directly invoke a VMWRITE or
5549 * queue up the VMWRITE by using the VMCS write cache (on 32-bit hosts, except
5550 * darwin, running 64-bit guests).
5551 *
5552 * @returns VBox status code.
5553 * @param pVCpu The cross context virtual CPU structure.
5554 * @param idxField The VMCS field encoding.
5555 * @param u64Val 16, 32 or 64-bit value.
5556 */
5557VMMR0DECL(int) VMXWriteVmcs64Ex(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val)
5558{
5559 int rc;
5560 switch (idxField)
5561 {
5562 /*
5563 * These fields consists of a "FULL" and a "HIGH" part which can be written to individually.
5564 */
5565 /* 64-bit Control fields. */
5566 case VMX_VMCS64_CTRL_IO_BITMAP_A_FULL:
5567 case VMX_VMCS64_CTRL_IO_BITMAP_B_FULL:
5568 case VMX_VMCS64_CTRL_MSR_BITMAP_FULL:
5569 case VMX_VMCS64_CTRL_EXIT_MSR_STORE_FULL:
5570 case VMX_VMCS64_CTRL_EXIT_MSR_LOAD_FULL:
5571 case VMX_VMCS64_CTRL_ENTRY_MSR_LOAD_FULL:
5572 case VMX_VMCS64_CTRL_EXEC_VMCS_PTR_FULL:
5573 case VMX_VMCS64_CTRL_TSC_OFFSET_FULL:
5574 case VMX_VMCS64_CTRL_VAPIC_PAGEADDR_FULL:
5575 case VMX_VMCS64_CTRL_APIC_ACCESSADDR_FULL:
5576 case VMX_VMCS64_CTRL_VMFUNC_CTRLS_FULL:
5577 case VMX_VMCS64_CTRL_EPTP_FULL:
5578 case VMX_VMCS64_CTRL_EPTP_LIST_FULL:
5579 /* 64-bit Guest-state fields. */
5580 case VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL:
5581 case VMX_VMCS64_GUEST_DEBUGCTL_FULL:
5582 case VMX_VMCS64_GUEST_PAT_FULL:
5583 case VMX_VMCS64_GUEST_EFER_FULL:
5584 case VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL:
5585 case VMX_VMCS64_GUEST_PDPTE0_FULL:
5586 case VMX_VMCS64_GUEST_PDPTE1_FULL:
5587 case VMX_VMCS64_GUEST_PDPTE2_FULL:
5588 case VMX_VMCS64_GUEST_PDPTE3_FULL:
5589 /* 64-bit Host-state fields. */
5590 case VMX_VMCS64_HOST_PAT_FULL:
5591 case VMX_VMCS64_HOST_EFER_FULL:
5592 case VMX_VMCS64_HOST_PERF_GLOBAL_CTRL_FULL:
5593 {
5594 rc = VMXWriteVmcs32(idxField, u64Val);
5595 rc |= VMXWriteVmcs32(idxField + 1, (uint32_t)(u64Val >> 32));
5596 break;
5597 }
5598
5599 /*
5600 * These fields do not have high and low parts. Queue up the VMWRITE by using the VMCS write-cache (for 64-bit
5601 * values). When we switch the host to 64-bit mode for running 64-bit guests, these VMWRITEs get executed then.
5602 */
5603 /* Natural-width Guest-state fields. */
5604 case VMX_VMCS_GUEST_CR3:
5605 case VMX_VMCS_GUEST_ES_BASE:
5606 case VMX_VMCS_GUEST_CS_BASE:
5607 case VMX_VMCS_GUEST_SS_BASE:
5608 case VMX_VMCS_GUEST_DS_BASE:
5609 case VMX_VMCS_GUEST_FS_BASE:
5610 case VMX_VMCS_GUEST_GS_BASE:
5611 case VMX_VMCS_GUEST_LDTR_BASE:
5612 case VMX_VMCS_GUEST_TR_BASE:
5613 case VMX_VMCS_GUEST_GDTR_BASE:
5614 case VMX_VMCS_GUEST_IDTR_BASE:
5615 case VMX_VMCS_GUEST_RSP:
5616 case VMX_VMCS_GUEST_RIP:
5617 case VMX_VMCS_GUEST_SYSENTER_ESP:
5618 case VMX_VMCS_GUEST_SYSENTER_EIP:
5619 {
5620 if (!(u64Val >> 32))
5621 {
5622 /* If this field is 64-bit, VT-x will zero out the top bits. */
5623 rc = VMXWriteVmcs32(idxField, (uint32_t)u64Val);
5624 }
5625 else
5626 {
5627 /* Assert that only the 32->64 switcher case should ever come here. */
5628 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fAllow64BitGuests);
5629 rc = VMXWriteCachedVmcsEx(pVCpu, idxField, u64Val);
5630 }
5631 break;
5632 }
5633
5634 default:
5635 {
5636 AssertMsgFailed(("VMXWriteVmcs64Ex: Invalid field %#RX32 (pVCpu=%p u64Val=%#RX64)\n", idxField, pVCpu, u64Val));
5637 rc = VERR_INVALID_PARAMETER;
5638 break;
5639 }
5640 }
5641 AssertRCReturn(rc, rc);
5642 return rc;
5643}
5644
5645
5646/**
5647 * Queue up a VMWRITE by using the VMCS write cache.
5648 * This is only used on 32-bit hosts (except darwin) for 64-bit guests.
5649 *
5650 * @param pVCpu The cross context virtual CPU structure.
5651 * @param idxField The VMCS field encoding.
5652 * @param u64Val 16, 32 or 64-bit value.
5653 */
5654VMMR0DECL(int) VMXWriteCachedVmcsEx(PVMCPU pVCpu, uint32_t idxField, uint64_t u64Val)
5655{
5656 AssertPtr(pVCpu);
5657 PVMCSCACHE pCache = &pVCpu->hm.s.vmx.VMCSCache;
5658
5659 AssertMsgReturn(pCache->Write.cValidEntries < VMCSCACHE_MAX_ENTRY - 1,
5660 ("entries=%u\n", pCache->Write.cValidEntries), VERR_ACCESS_DENIED);
5661
5662 /* Make sure there are no duplicates. */
5663 for (uint32_t i = 0; i < pCache->Write.cValidEntries; i++)
5664 {
5665 if (pCache->Write.aField[i] == idxField)
5666 {
5667 pCache->Write.aFieldVal[i] = u64Val;
5668 return VINF_SUCCESS;
5669 }
5670 }
5671
5672 pCache->Write.aField[pCache->Write.cValidEntries] = idxField;
5673 pCache->Write.aFieldVal[pCache->Write.cValidEntries] = u64Val;
5674 pCache->Write.cValidEntries++;
5675 return VINF_SUCCESS;
5676}
5677#endif /* HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS) */
5678
5679
5680/**
5681 * Sets up the usage of TSC-offsetting and updates the VMCS.
5682 *
5683 * If offsetting is not possible, cause VM-exits on RDTSC(P)s. Also sets up the
5684 * VMX preemption timer.
5685 *
5686 * @returns VBox status code.
5687 * @param pVM The cross context VM structure.
5688 * @param pVCpu The cross context virtual CPU structure.
5689 *
5690 * @remarks No-long-jump zone!!!
5691 */
5692static void hmR0VmxUpdateTscOffsettingAndPreemptTimer(PVM pVM, PVMCPU pVCpu)
5693{
5694 int rc;
5695 bool fOffsettedTsc;
5696 bool fParavirtTsc;
5697 if (pVM->hm.s.vmx.fUsePreemptTimer)
5698 {
5699 uint64_t cTicksToDeadline = TMCpuTickGetDeadlineAndTscOffset(pVM, pVCpu, &pVCpu->hm.s.vmx.u64TSCOffset,
5700 &fOffsettedTsc, &fParavirtTsc);
5701
5702 /* Make sure the returned values have sane upper and lower boundaries. */
5703 uint64_t u64CpuHz = SUPGetCpuHzFromGipBySetIndex(g_pSUPGlobalInfoPage, pVCpu->iHostCpuSet);
5704 cTicksToDeadline = RT_MIN(cTicksToDeadline, u64CpuHz / 64); /* 1/64th of a second */
5705 cTicksToDeadline = RT_MAX(cTicksToDeadline, u64CpuHz / 2048); /* 1/2048th of a second */
5706 cTicksToDeadline >>= pVM->hm.s.vmx.cPreemptTimerShift;
5707
5708 uint32_t cPreemptionTickCount = (uint32_t)RT_MIN(cTicksToDeadline, UINT32_MAX - 16);
5709 rc = VMXWriteVmcs32(VMX_VMCS32_GUEST_PREEMPT_TIMER_VALUE, cPreemptionTickCount); AssertRC(rc);
5710 }
5711 else
5712 fOffsettedTsc = TMCpuTickCanUseRealTSC(pVM, pVCpu, &pVCpu->hm.s.vmx.u64TSCOffset, &fParavirtTsc);
5713
5714 /** @todo later optimize this to be done elsewhere and not before every
5715 * VM-entry. */
5716 if (fParavirtTsc)
5717 {
5718 /* Currently neither Hyper-V nor KVM need to update their paravirt. TSC
5719 information before every VM-entry, hence disable it for performance sake. */
5720#if 0
5721 rc = GIMR0UpdateParavirtTsc(pVM, 0 /* u64Offset */);
5722 AssertRC(rc);
5723#endif
5724 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscParavirt);
5725 }
5726
5727 if (fOffsettedTsc && RT_LIKELY(!pVCpu->hm.s.fDebugWantRdTscExit))
5728 {
5729 /* Note: VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT takes precedence over TSC_OFFSET, applies to RDTSCP too. */
5730 rc = VMXWriteVmcs64(VMX_VMCS64_CTRL_TSC_OFFSET_FULL, pVCpu->hm.s.vmx.u64TSCOffset); AssertRC(rc);
5731
5732 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT;
5733 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc);
5734 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscOffset);
5735 }
5736 else
5737 {
5738 /* We can't use TSC-offsetting (non-fixed TSC, warp drive active etc.), VM-exit on RDTSC(P). */
5739 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT;
5740 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls); AssertRC(rc);
5741 STAM_COUNTER_INC(&pVCpu->hm.s.StatTscIntercept);
5742 }
5743}
5744
5745
5746/**
5747 * Determines if an exception is a contributory exception.
5748 *
5749 * Contributory exceptions are ones which can cause double-faults unless the
5750 * original exception was a benign exception. Page-fault is intentionally not
5751 * included here as it's a conditional contributory exception.
5752 *
5753 * @returns true if the exception is contributory, false otherwise.
5754 * @param uVector The exception vector.
5755 */
5756DECLINLINE(bool) hmR0VmxIsContributoryXcpt(const uint32_t uVector)
5757{
5758 switch (uVector)
5759 {
5760 case X86_XCPT_GP:
5761 case X86_XCPT_SS:
5762 case X86_XCPT_NP:
5763 case X86_XCPT_TS:
5764 case X86_XCPT_DE:
5765 return true;
5766 default:
5767 break;
5768 }
5769 return false;
5770}
5771
5772
5773/**
5774 * Sets an event as a pending event to be injected into the guest.
5775 *
5776 * @param pVCpu The cross context virtual CPU structure.
5777 * @param u32IntInfo The VM-entry interruption-information field.
5778 * @param cbInstr The VM-entry instruction length in bytes (for software
5779 * interrupts, exceptions and privileged software
5780 * exceptions).
5781 * @param u32ErrCode The VM-entry exception error code.
5782 * @param GCPtrFaultAddress The fault-address (CR2) in case it's a
5783 * page-fault.
5784 *
5785 * @remarks Statistics counter assumes this is a guest event being injected or
5786 * re-injected into the guest, i.e. 'StatInjectPendingReflect' is
5787 * always incremented.
5788 */
5789DECLINLINE(void) hmR0VmxSetPendingEvent(PVMCPU pVCpu, uint32_t u32IntInfo, uint32_t cbInstr, uint32_t u32ErrCode,
5790 RTGCUINTPTR GCPtrFaultAddress)
5791{
5792 Assert(!pVCpu->hm.s.Event.fPending);
5793 pVCpu->hm.s.Event.fPending = true;
5794 pVCpu->hm.s.Event.u64IntInfo = u32IntInfo;
5795 pVCpu->hm.s.Event.u32ErrCode = u32ErrCode;
5796 pVCpu->hm.s.Event.cbInstr = cbInstr;
5797 pVCpu->hm.s.Event.GCPtrFaultAddress = GCPtrFaultAddress;
5798}
5799
5800
5801/**
5802 * Sets a double-fault (\#DF) exception as pending-for-injection into the VM.
5803 *
5804 * @param pVCpu The cross context virtual CPU structure.
5805 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
5806 * out-of-sync. Make sure to update the required fields
5807 * before using them.
5808 */
5809DECLINLINE(void) hmR0VmxSetPendingXcptDF(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
5810{
5811 NOREF(pMixedCtx);
5812 uint32_t u32IntInfo = X86_XCPT_DF | VMX_EXIT_INTERRUPTION_INFO_VALID;
5813 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
5814 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
5815 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
5816}
5817
5818
5819/**
5820 * Handle a condition that occurred while delivering an event through the guest
5821 * IDT.
5822 *
5823 * @returns Strict VBox status code (i.e. informational status codes too).
5824 * @retval VINF_SUCCESS if we should continue handling the VM-exit.
5825 * @retval VINF_HM_DOUBLE_FAULT if a \#DF condition was detected and we ought
5826 * to continue execution of the guest which will delivery the \#DF.
5827 * @retval VINF_EM_RESET if we detected a triple-fault condition.
5828 * @retval VERR_EM_GUEST_CPU_HANG if we detected a guest CPU hang.
5829 *
5830 * @param pVCpu The cross context virtual CPU structure.
5831 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
5832 * out-of-sync. Make sure to update the required fields
5833 * before using them.
5834 * @param pVmxTransient Pointer to the VMX transient structure.
5835 *
5836 * @remarks No-long-jump zone!!!
5837 */
5838static VBOXSTRICTRC hmR0VmxCheckExitDueToEventDelivery(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
5839{
5840 uint32_t uExitVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo);
5841
5842 int rc2 = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient); AssertRCReturn(rc2, rc2);
5843 rc2 = hmR0VmxReadExitIntInfoVmcs(pVmxTransient); AssertRCReturn(rc2, rc2);
5844
5845 VBOXSTRICTRC rcStrict = VINF_SUCCESS;
5846 if (VMX_IDT_VECTORING_INFO_VALID(pVmxTransient->uIdtVectoringInfo))
5847 {
5848 uint32_t uIdtVectorType = VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo);
5849 uint32_t uIdtVector = VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo);
5850
5851 typedef enum
5852 {
5853 VMXREFLECTXCPT_XCPT, /* Reflect the exception to the guest or for further evaluation by VMM. */
5854 VMXREFLECTXCPT_DF, /* Reflect the exception as a double-fault to the guest. */
5855 VMXREFLECTXCPT_TF, /* Indicate a triple faulted state to the VMM. */
5856 VMXREFLECTXCPT_HANG, /* Indicate bad VM trying to deadlock the CPU. */
5857 VMXREFLECTXCPT_NONE /* Nothing to reflect. */
5858 } VMXREFLECTXCPT;
5859
5860 /* See Intel spec. 30.7.1.1 "Reflecting Exceptions to Guest Software". */
5861 VMXREFLECTXCPT enmReflect = VMXREFLECTXCPT_NONE;
5862 if (VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo))
5863 {
5864 if (uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT)
5865 {
5866 enmReflect = VMXREFLECTXCPT_XCPT;
5867#ifdef VBOX_STRICT
5868 if ( hmR0VmxIsContributoryXcpt(uIdtVector)
5869 && uExitVector == X86_XCPT_PF)
5870 {
5871 Log4(("IDT: vcpu[%RU32] Contributory #PF uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5872 }
5873#endif
5874 if ( uExitVector == X86_XCPT_PF
5875 && uIdtVector == X86_XCPT_PF)
5876 {
5877 pVmxTransient->fVectoringDoublePF = true;
5878 Log4(("IDT: vcpu[%RU32] Vectoring Double #PF uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5879 }
5880 else if ( uExitVector == X86_XCPT_AC
5881 && uIdtVector == X86_XCPT_AC)
5882 {
5883 enmReflect = VMXREFLECTXCPT_HANG;
5884 Log4(("IDT: Nested #AC - Bad guest\n"));
5885 }
5886 else if ( (pVCpu->hm.s.vmx.u32XcptBitmap & HMVMX_CONTRIBUTORY_XCPT_MASK)
5887 && hmR0VmxIsContributoryXcpt(uExitVector)
5888 && ( hmR0VmxIsContributoryXcpt(uIdtVector)
5889 || uIdtVector == X86_XCPT_PF))
5890 {
5891 enmReflect = VMXREFLECTXCPT_DF;
5892 }
5893 else if (uIdtVector == X86_XCPT_DF)
5894 enmReflect = VMXREFLECTXCPT_TF;
5895 }
5896 else if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT
5897 || uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI)
5898 {
5899 /*
5900 * Ignore software interrupts (INT n), software exceptions (#BP, #OF) and
5901 * privileged software exception (#DB from ICEBP) as they reoccur when restarting the instruction.
5902 */
5903 enmReflect = VMXREFLECTXCPT_XCPT;
5904
5905 if (uExitVector == X86_XCPT_PF)
5906 {
5907 pVmxTransient->fVectoringPF = true;
5908 Log4(("IDT: vcpu[%RU32] Vectoring #PF due to Ext-Int/NMI. uCR2=%#RX64\n", pVCpu->idCpu, pMixedCtx->cr2));
5909 }
5910 }
5911 }
5912 else if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
5913 || uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_EXT_INT
5914 || uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI)
5915 {
5916 /*
5917 * If event delivery caused an EPT violation/misconfig or APIC access VM-exit, then the VM-exit
5918 * interruption-information will not be valid as it's not an exception and we end up here. In such cases,
5919 * it is sufficient to reflect the original exception to the guest after handling the VM-exit.
5920 */
5921 enmReflect = VMXREFLECTXCPT_XCPT;
5922 }
5923
5924 /*
5925 * On CPUs that support Virtual NMIs, if this VM-exit (be it an exception or EPT violation/misconfig etc.) occurred
5926 * while delivering the NMI, we need to clear the block-by-NMI field in the guest interruptibility-state before
5927 * re-delivering the NMI after handling the VM-exit. Otherwise the subsequent VM-entry would fail.
5928 *
5929 * See Intel spec. 30.7.1.2 "Resuming Guest Software after Handling an Exception". See @bugref{7445}.
5930 */
5931 if ( uIdtVectorType == VMX_IDT_VECTORING_INFO_TYPE_NMI
5932 && enmReflect == VMXREFLECTXCPT_XCPT
5933 && (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
5934 && VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
5935 {
5936 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
5937 }
5938
5939 switch (enmReflect)
5940 {
5941 case VMXREFLECTXCPT_XCPT:
5942 {
5943 Assert( uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_SW_INT
5944 && uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
5945 && uIdtVectorType != VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT);
5946
5947 uint32_t u32ErrCode = 0;
5948 if (VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uIdtVectoringInfo))
5949 {
5950 rc2 = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
5951 AssertRCReturn(rc2, rc2);
5952 u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
5953 }
5954
5955 /* If uExitVector is #PF, CR2 value will be updated from the VMCS if it's a guest #PF. See hmR0VmxExitXcptPF(). */
5956 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect);
5957 hmR0VmxSetPendingEvent(pVCpu, VMX_ENTRY_INT_INFO_FROM_EXIT_IDT_INFO(pVmxTransient->uIdtVectoringInfo),
5958 0 /* cbInstr */, u32ErrCode, pMixedCtx->cr2);
5959 rcStrict = VINF_SUCCESS;
5960 Log4(("IDT: vcpu[%RU32] Pending vectoring event %#RX64 Err=%#RX32\n", pVCpu->idCpu,
5961 pVCpu->hm.s.Event.u64IntInfo, pVCpu->hm.s.Event.u32ErrCode));
5962
5963 break;
5964 }
5965
5966 case VMXREFLECTXCPT_DF:
5967 {
5968 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingReflect);
5969 hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
5970 rcStrict = VINF_HM_DOUBLE_FAULT;
5971 Log4(("IDT: vcpu[%RU32] Pending vectoring #DF %#RX64 uIdtVector=%#x uExitVector=%#x\n", pVCpu->idCpu,
5972 pVCpu->hm.s.Event.u64IntInfo, uIdtVector, uExitVector));
5973
5974 break;
5975 }
5976
5977 case VMXREFLECTXCPT_TF:
5978 {
5979 rcStrict = VINF_EM_RESET;
5980 Log4(("IDT: vcpu[%RU32] Pending vectoring triple-fault uIdt=%#x uExit=%#x\n", pVCpu->idCpu, uIdtVector,
5981 uExitVector));
5982 break;
5983 }
5984
5985 case VMXREFLECTXCPT_HANG:
5986 {
5987 rcStrict = VERR_EM_GUEST_CPU_HANG;
5988 break;
5989 }
5990
5991 default:
5992 Assert(rcStrict == VINF_SUCCESS);
5993 break;
5994 }
5995 }
5996 else if ( VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo)
5997 && VMX_EXIT_INTERRUPTION_INFO_NMI_UNBLOCK_IRET(pVmxTransient->uExitIntInfo)
5998 && uExitVector != X86_XCPT_DF
5999 && (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI))
6000 {
6001 /*
6002 * Execution of IRET caused this fault when NMI blocking was in effect (i.e we're in the guest NMI handler).
6003 * We need to set the block-by-NMI field so that NMIs remain blocked until the IRET execution is restarted.
6004 * See Intel spec. 30.7.1.2 "Resuming guest software after handling an exception".
6005 */
6006 if (!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6007 {
6008 Log4(("hmR0VmxCheckExitDueToEventDelivery: vcpu[%RU32] Setting VMCPU_FF_BLOCK_NMIS. Valid=%RTbool uExitReason=%u\n",
6009 pVCpu->idCpu, VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo), pVmxTransient->uExitReason));
6010 VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
6011 }
6012 }
6013
6014 Assert( rcStrict == VINF_SUCCESS || rcStrict == VINF_HM_DOUBLE_FAULT
6015 || rcStrict == VINF_EM_RESET || rcStrict == VERR_EM_GUEST_CPU_HANG);
6016 return rcStrict;
6017}
6018
6019
6020/**
6021 * Saves the guest's CR0 register from the VMCS into the guest-CPU context.
6022 *
6023 * @returns VBox status code.
6024 * @param pVCpu The cross context virtual CPU structure.
6025 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6026 * out-of-sync. Make sure to update the required fields
6027 * before using them.
6028 *
6029 * @remarks No-long-jump zone!!!
6030 */
6031static int hmR0VmxSaveGuestCR0(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6032{
6033 NOREF(pMixedCtx);
6034
6035 /*
6036 * While in the middle of saving guest-CR0, we could get preempted and re-invoked from the preemption hook,
6037 * see hmR0VmxLeave(). Safer to just make this code non-preemptible.
6038 */
6039 VMMRZCallRing3Disable(pVCpu);
6040 HM_DISABLE_PREEMPT();
6041
6042 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0))
6043 {
6044 uint32_t uVal = 0;
6045 uint32_t uShadow = 0;
6046 int rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &uVal);
6047 rc |= VMXReadVmcs32(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uShadow);
6048 AssertRCReturn(rc, rc);
6049
6050 uVal = (uShadow & pVCpu->hm.s.vmx.u32CR0Mask) | (uVal & ~pVCpu->hm.s.vmx.u32CR0Mask);
6051 CPUMSetGuestCR0(pVCpu, uVal);
6052 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0);
6053 }
6054
6055 HM_RESTORE_PREEMPT();
6056 VMMRZCallRing3Enable(pVCpu);
6057 return VINF_SUCCESS;
6058}
6059
6060
6061/**
6062 * Saves the guest's CR4 register from the VMCS into the guest-CPU context.
6063 *
6064 * @returns VBox status code.
6065 * @param pVCpu The cross context virtual CPU structure.
6066 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6067 * out-of-sync. Make sure to update the required fields
6068 * before using them.
6069 *
6070 * @remarks No-long-jump zone!!!
6071 */
6072static int hmR0VmxSaveGuestCR4(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6073{
6074 NOREF(pMixedCtx);
6075
6076 int rc = VINF_SUCCESS;
6077 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4))
6078 {
6079 uint32_t uVal = 0;
6080 uint32_t uShadow = 0;
6081 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &uVal);
6082 rc |= VMXReadVmcs32(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uShadow);
6083 AssertRCReturn(rc, rc);
6084
6085 uVal = (uShadow & pVCpu->hm.s.vmx.u32CR4Mask) | (uVal & ~pVCpu->hm.s.vmx.u32CR4Mask);
6086 CPUMSetGuestCR4(pVCpu, uVal);
6087 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4);
6088 }
6089 return rc;
6090}
6091
6092
6093/**
6094 * Saves the guest's RIP register from the VMCS into the guest-CPU context.
6095 *
6096 * @returns VBox status code.
6097 * @param pVCpu The cross context virtual CPU structure.
6098 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6099 * out-of-sync. Make sure to update the required fields
6100 * before using them.
6101 *
6102 * @remarks No-long-jump zone!!!
6103 */
6104static int hmR0VmxSaveGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6105{
6106 int rc = VINF_SUCCESS;
6107 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP))
6108 {
6109 uint64_t u64Val = 0;
6110 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RIP, &u64Val);
6111 AssertRCReturn(rc, rc);
6112
6113 pMixedCtx->rip = u64Val;
6114 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP);
6115 }
6116 return rc;
6117}
6118
6119
6120/**
6121 * Saves the guest's RSP register from the VMCS into the guest-CPU context.
6122 *
6123 * @returns VBox status code.
6124 * @param pVCpu The cross context virtual CPU structure.
6125 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6126 * out-of-sync. Make sure to update the required fields
6127 * before using them.
6128 *
6129 * @remarks No-long-jump zone!!!
6130 */
6131static int hmR0VmxSaveGuestRsp(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6132{
6133 int rc = VINF_SUCCESS;
6134 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RSP))
6135 {
6136 uint64_t u64Val = 0;
6137 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_RSP, &u64Val);
6138 AssertRCReturn(rc, rc);
6139
6140 pMixedCtx->rsp = u64Val;
6141 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RSP);
6142 }
6143 return rc;
6144}
6145
6146
6147/**
6148 * Saves the guest's RFLAGS from the VMCS into the guest-CPU context.
6149 *
6150 * @returns VBox status code.
6151 * @param pVCpu The cross context virtual CPU structure.
6152 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6153 * out-of-sync. Make sure to update the required fields
6154 * before using them.
6155 *
6156 * @remarks No-long-jump zone!!!
6157 */
6158static int hmR0VmxSaveGuestRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6159{
6160 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS))
6161 {
6162 uint32_t uVal = 0;
6163 int rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &uVal);
6164 AssertRCReturn(rc, rc);
6165
6166 pMixedCtx->eflags.u32 = uVal;
6167 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active) /* Undo our real-on-v86-mode changes to eflags if necessary. */
6168 {
6169 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
6170 Log4(("Saving real-mode EFLAGS VT-x view=%#RX32\n", pMixedCtx->eflags.u32));
6171
6172 pMixedCtx->eflags.Bits.u1VM = 0;
6173 pMixedCtx->eflags.Bits.u2IOPL = pVCpu->hm.s.vmx.RealMode.Eflags.Bits.u2IOPL;
6174 }
6175
6176 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS);
6177 }
6178 return VINF_SUCCESS;
6179}
6180
6181
6182/**
6183 * Wrapper for saving the guest's RIP, RSP and RFLAGS from the VMCS into the
6184 * guest-CPU context.
6185 */
6186DECLINLINE(int) hmR0VmxSaveGuestRipRspRflags(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6187{
6188 int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6189 rc |= hmR0VmxSaveGuestRsp(pVCpu, pMixedCtx);
6190 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
6191 return rc;
6192}
6193
6194
6195/**
6196 * Saves the guest's interruptibility-state ("interrupt shadow" as AMD calls it)
6197 * from the guest-state area in the VMCS.
6198 *
6199 * @param pVCpu The cross context virtual CPU structure.
6200 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6201 * out-of-sync. Make sure to update the required fields
6202 * before using them.
6203 *
6204 * @remarks No-long-jump zone!!!
6205 */
6206static void hmR0VmxSaveGuestIntrState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6207{
6208 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_INTR_STATE))
6209 {
6210 uint32_t uIntrState = 0;
6211 int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
6212 AssertRC(rc);
6213
6214 if (!uIntrState)
6215 {
6216 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
6217 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
6218
6219 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6220 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
6221 }
6222 else
6223 {
6224 if (uIntrState & ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS
6225 | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI))
6226 {
6227 rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6228 AssertRC(rc);
6229 rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); /* for hmR0VmxGetGuestIntrState(). */
6230 AssertRC(rc);
6231
6232 EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip);
6233 Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS));
6234 }
6235 else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
6236 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
6237
6238 if (uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI)
6239 {
6240 if (!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6241 VMCPU_FF_SET(pVCpu, VMCPU_FF_BLOCK_NMIS);
6242 }
6243 else if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS))
6244 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_BLOCK_NMIS);
6245 }
6246
6247 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_INTR_STATE);
6248 }
6249}
6250
6251
6252/**
6253 * Saves the guest's activity state.
6254 *
6255 * @returns VBox status code.
6256 * @param pVCpu The cross context virtual CPU structure.
6257 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6258 * out-of-sync. Make sure to update the required fields
6259 * before using them.
6260 *
6261 * @remarks No-long-jump zone!!!
6262 */
6263static int hmR0VmxSaveGuestActivityState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6264{
6265 NOREF(pMixedCtx);
6266 /* Nothing to do for now until we make use of different guest-CPU activity state. Just update the flag. */
6267 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_ACTIVITY_STATE);
6268 return VINF_SUCCESS;
6269}
6270
6271
6272/**
6273 * Saves the guest SYSENTER MSRs (SYSENTER_CS, SYSENTER_EIP, SYSENTER_ESP) from
6274 * the current VMCS into the guest-CPU context.
6275 *
6276 * @returns VBox status code.
6277 * @param pVCpu The cross context virtual CPU structure.
6278 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6279 * out-of-sync. Make sure to update the required fields
6280 * before using them.
6281 *
6282 * @remarks No-long-jump zone!!!
6283 */
6284static int hmR0VmxSaveGuestSysenterMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6285{
6286 int rc = VINF_SUCCESS;
6287 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR))
6288 {
6289 uint32_t u32Val = 0;
6290 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_SYSENTER_CS, &u32Val); AssertRCReturn(rc, rc);
6291 pMixedCtx->SysEnter.cs = u32Val;
6292 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_CS_MSR);
6293 }
6294
6295 uint64_t u64Val = 0;
6296 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR))
6297 {
6298 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_EIP, &u64Val); AssertRCReturn(rc, rc);
6299 pMixedCtx->SysEnter.eip = u64Val;
6300 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_EIP_MSR);
6301 }
6302 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR))
6303 {
6304 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_SYSENTER_ESP, &u64Val); AssertRCReturn(rc, rc);
6305 pMixedCtx->SysEnter.esp = u64Val;
6306 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SYSENTER_ESP_MSR);
6307 }
6308 return rc;
6309}
6310
6311
6312/**
6313 * Saves the set of guest MSRs (that we restore lazily while leaving VT-x) from
6314 * the CPU back into the guest-CPU context.
6315 *
6316 * @returns VBox status code.
6317 * @param pVCpu The cross context virtual CPU structure.
6318 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6319 * out-of-sync. Make sure to update the required fields
6320 * before using them.
6321 *
6322 * @remarks No-long-jump zone!!!
6323 */
6324static int hmR0VmxSaveGuestLazyMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6325{
6326 /* Since this can be called from our preemption hook it's safer to make the guest-MSRs update non-preemptible. */
6327 VMMRZCallRing3Disable(pVCpu);
6328 HM_DISABLE_PREEMPT();
6329
6330 /* Doing the check here ensures we don't overwrite already-saved guest MSRs from a preemption hook. */
6331 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS))
6332 {
6333 hmR0VmxLazySaveGuestMsrs(pVCpu, pMixedCtx);
6334 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS);
6335 }
6336
6337 HM_RESTORE_PREEMPT();
6338 VMMRZCallRing3Enable(pVCpu);
6339
6340 return VINF_SUCCESS;
6341}
6342
6343
6344/**
6345 * Saves the auto load/store'd guest MSRs from the current VMCS into
6346 * the guest-CPU context.
6347 *
6348 * @returns VBox status code.
6349 * @param pVCpu The cross context virtual CPU structure.
6350 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6351 * out-of-sync. Make sure to update the required fields
6352 * before using them.
6353 *
6354 * @remarks No-long-jump zone!!!
6355 */
6356static int hmR0VmxSaveGuestAutoLoadStoreMsrs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6357{
6358 if (HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS))
6359 return VINF_SUCCESS;
6360
6361 PVMXAUTOMSR pMsr = (PVMXAUTOMSR)pVCpu->hm.s.vmx.pvGuestMsr;
6362 uint32_t cMsrs = pVCpu->hm.s.vmx.cMsrs;
6363 Log4(("hmR0VmxSaveGuestAutoLoadStoreMsrs: cMsrs=%u\n", cMsrs));
6364 for (uint32_t i = 0; i < cMsrs; i++, pMsr++)
6365 {
6366 switch (pMsr->u32Msr)
6367 {
6368 case MSR_K8_TSC_AUX: CPUMR0SetGuestTscAux(pVCpu, pMsr->u64Value); break;
6369 case MSR_K8_LSTAR: pMixedCtx->msrLSTAR = pMsr->u64Value; break;
6370 case MSR_K6_STAR: pMixedCtx->msrSTAR = pMsr->u64Value; break;
6371 case MSR_K8_SF_MASK: pMixedCtx->msrSFMASK = pMsr->u64Value; break;
6372 case MSR_K8_KERNEL_GS_BASE: pMixedCtx->msrKERNELGSBASE = pMsr->u64Value; break;
6373 case MSR_K6_EFER: /* Nothing to do here since we intercept writes, see hmR0VmxLoadGuestMsrs(). */
6374 break;
6375
6376 default:
6377 {
6378 AssertMsgFailed(("Unexpected MSR in auto-load/store area. uMsr=%#RX32 cMsrs=%u\n", pMsr->u32Msr, cMsrs));
6379 pVCpu->hm.s.u32HMError = pMsr->u32Msr;
6380 return VERR_HM_UNEXPECTED_LD_ST_MSR;
6381 }
6382 }
6383 }
6384
6385 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS);
6386 return VINF_SUCCESS;
6387}
6388
6389
6390/**
6391 * Saves the guest control registers from the current VMCS into the guest-CPU
6392 * context.
6393 *
6394 * @returns VBox status code.
6395 * @param pVCpu The cross context virtual CPU structure.
6396 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6397 * out-of-sync. Make sure to update the required fields
6398 * before using them.
6399 *
6400 * @remarks No-long-jump zone!!!
6401 */
6402static int hmR0VmxSaveGuestControlRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6403{
6404 /* Guest CR0. Guest FPU. */
6405 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6406 AssertRCReturn(rc, rc);
6407
6408 /* Guest CR4. */
6409 rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
6410 AssertRCReturn(rc, rc);
6411
6412 /* Guest CR2 - updated always during the world-switch or in #PF. */
6413 /* Guest CR3. Only changes with Nested Paging. This must be done -after- saving CR0 and CR4 from the guest! */
6414 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR3))
6415 {
6416 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
6417 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR4));
6418
6419 PVM pVM = pVCpu->CTX_SUFF(pVM);
6420 if ( pVM->hm.s.vmx.fUnrestrictedGuest
6421 || ( pVM->hm.s.fNestedPaging
6422 && CPUMIsGuestPagingEnabledEx(pMixedCtx)))
6423 {
6424 uint64_t u64Val = 0;
6425 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_CR3, &u64Val);
6426 if (pMixedCtx->cr3 != u64Val)
6427 {
6428 CPUMSetGuestCR3(pVCpu, u64Val);
6429 if (VMMRZCallRing3IsEnabled(pVCpu))
6430 {
6431 PGMUpdateCR3(pVCpu, u64Val);
6432 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6433 }
6434 else
6435 {
6436 /* Set the force flag to inform PGM about it when necessary. It is cleared by PGMUpdateCR3().*/
6437 VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_CR3);
6438 }
6439 }
6440
6441 /* If the guest is in PAE mode, sync back the PDPE's into the guest state. */
6442 if (CPUMIsGuestInPAEModeEx(pMixedCtx)) /* Reads CR0, CR4 and EFER MSR (EFER is always up-to-date). */
6443 {
6444 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &pVCpu->hm.s.aPdpes[0].u);
6445 rc |= VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &pVCpu->hm.s.aPdpes[1].u);
6446 rc |= VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &pVCpu->hm.s.aPdpes[2].u);
6447 rc |= VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &pVCpu->hm.s.aPdpes[3].u);
6448 AssertRCReturn(rc, rc);
6449
6450 if (VMMRZCallRing3IsEnabled(pVCpu))
6451 {
6452 PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6453 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6454 }
6455 else
6456 {
6457 /* Set the force flag to inform PGM about it when necessary. It is cleared by PGMGstUpdatePaePdpes(). */
6458 VMCPU_FF_SET(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES);
6459 }
6460 }
6461 }
6462
6463 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR3);
6464 }
6465
6466 /*
6467 * Consider this scenario: VM-exit -> VMMRZCallRing3Enable() -> do stuff that causes a longjmp -> hmR0VmxCallRing3Callback()
6468 * -> VMMRZCallRing3Disable() -> hmR0VmxSaveGuestState() -> Set VMCPU_FF_HM_UPDATE_CR3 pending -> return from the longjmp
6469 * -> continue with VM-exit handling -> hmR0VmxSaveGuestControlRegs() and here we are.
6470 *
6471 * The reason for such complicated handling is because VM-exits that call into PGM expect CR3 to be up-to-date and thus
6472 * if any CR3-saves -before- the VM-exit (longjmp) postponed the CR3 update via the force-flag, any VM-exit handler that
6473 * calls into PGM when it re-saves CR3 will end up here and we call PGMUpdateCR3(). This is why the code below should
6474 * -NOT- check if HMVMX_UPDATED_GUEST_CR3 is already set or not!
6475 *
6476 * The longjmp exit path can't check these CR3 force-flags and call code that takes a lock again. We cover for it here.
6477 */
6478 if (VMMRZCallRing3IsEnabled(pVCpu))
6479 {
6480 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
6481 PGMUpdateCR3(pVCpu, CPUMGetGuestCR3(pVCpu));
6482
6483 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
6484 PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6485
6486 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6487 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6488 }
6489
6490 return rc;
6491}
6492
6493
6494/**
6495 * Reads a guest segment register from the current VMCS into the guest-CPU
6496 * context.
6497 *
6498 * @returns VBox status code.
6499 * @param pVCpu The cross context virtual CPU structure.
6500 * @param idxSel Index of the selector in the VMCS.
6501 * @param idxLimit Index of the segment limit in the VMCS.
6502 * @param idxBase Index of the segment base in the VMCS.
6503 * @param idxAccess Index of the access rights of the segment in the VMCS.
6504 * @param pSelReg Pointer to the segment selector.
6505 *
6506 * @remarks No-long-jump zone!!!
6507 * @remarks Never call this function directly!!! Use the VMXLOCAL_READ_SEG()
6508 * macro as that takes care of whether to read from the VMCS cache or
6509 * not.
6510 */
6511DECLINLINE(int) hmR0VmxReadSegmentReg(PVMCPU pVCpu, uint32_t idxSel, uint32_t idxLimit, uint32_t idxBase, uint32_t idxAccess,
6512 PCPUMSELREG pSelReg)
6513{
6514 NOREF(pVCpu);
6515
6516 uint32_t u32Val = 0;
6517 int rc = VMXReadVmcs32(idxSel, &u32Val);
6518 AssertRCReturn(rc, rc);
6519 pSelReg->Sel = (uint16_t)u32Val;
6520 pSelReg->ValidSel = (uint16_t)u32Val;
6521 pSelReg->fFlags = CPUMSELREG_FLAGS_VALID;
6522
6523 rc = VMXReadVmcs32(idxLimit, &u32Val);
6524 AssertRCReturn(rc, rc);
6525 pSelReg->u32Limit = u32Val;
6526
6527 uint64_t u64Val = 0;
6528 rc = VMXReadVmcsGstNByIdxVal(idxBase, &u64Val);
6529 AssertRCReturn(rc, rc);
6530 pSelReg->u64Base = u64Val;
6531
6532 rc = VMXReadVmcs32(idxAccess, &u32Val);
6533 AssertRCReturn(rc, rc);
6534 pSelReg->Attr.u = u32Val;
6535
6536 /*
6537 * If VT-x marks the segment as unusable, most other bits remain undefined:
6538 * - For CS the L, D and G bits have meaning.
6539 * - For SS the DPL has meaning (it -is- the CPL for Intel and VBox).
6540 * - For the remaining data segments no bits are defined.
6541 *
6542 * The present bit and the unusable bit has been observed to be set at the
6543 * same time (the selector was supposed to be invalid as we started executing
6544 * a V8086 interrupt in ring-0).
6545 *
6546 * What should be important for the rest of the VBox code, is that the P bit is
6547 * cleared. Some of the other VBox code recognizes the unusable bit, but
6548 * AMD-V certainly don't, and REM doesn't really either. So, to be on the
6549 * safe side here, we'll strip off P and other bits we don't care about. If
6550 * any code breaks because Attr.u != 0 when Sel < 4, it should be fixed.
6551 *
6552 * See Intel spec. 27.3.2 "Saving Segment Registers and Descriptor-Table Registers".
6553 */
6554 if (pSelReg->Attr.u & X86DESCATTR_UNUSABLE)
6555 {
6556 Assert(idxSel != VMX_VMCS16_GUEST_TR_SEL); /* TR is the only selector that can never be unusable. */
6557
6558 /* Masking off: X86DESCATTR_P, X86DESCATTR_LIMIT_HIGH, and X86DESCATTR_AVL. The latter two are really irrelevant. */
6559 pSelReg->Attr.u &= X86DESCATTR_UNUSABLE | X86DESCATTR_L | X86DESCATTR_D | X86DESCATTR_G
6560 | X86DESCATTR_DPL | X86DESCATTR_TYPE | X86DESCATTR_DT;
6561
6562 Log4(("hmR0VmxReadSegmentReg: Unusable idxSel=%#x attr=%#x -> %#x\n", idxSel, u32Val, pSelReg->Attr.u));
6563#ifdef DEBUG_bird
6564 AssertMsg((u32Val & ~X86DESCATTR_P) == pSelReg->Attr.u,
6565 ("%#x: %#x != %#x (sel=%#x base=%#llx limit=%#x)\n",
6566 idxSel, u32Val, pSelReg->Attr.u, pSelReg->Sel, pSelReg->u64Base, pSelReg->u32Limit));
6567#endif
6568 }
6569 return VINF_SUCCESS;
6570}
6571
6572
6573#ifdef VMX_USE_CACHED_VMCS_ACCESSES
6574# define VMXLOCAL_READ_SEG(Sel, CtxSel) \
6575 hmR0VmxReadSegmentReg(pVCpu, VMX_VMCS16_GUEST_##Sel##_SEL, VMX_VMCS32_GUEST_##Sel##_LIMIT, \
6576 VMX_VMCS_GUEST_##Sel##_BASE_CACHE_IDX, VMX_VMCS32_GUEST_##Sel##_ACCESS_RIGHTS, &pMixedCtx->CtxSel)
6577#else
6578# define VMXLOCAL_READ_SEG(Sel, CtxSel) \
6579 hmR0VmxReadSegmentReg(pVCpu, VMX_VMCS16_GUEST_##Sel##_SEL, VMX_VMCS32_GUEST_##Sel##_LIMIT, \
6580 VMX_VMCS_GUEST_##Sel##_BASE, VMX_VMCS32_GUEST_##Sel##_ACCESS_RIGHTS, &pMixedCtx->CtxSel)
6581#endif
6582
6583
6584/**
6585 * Saves the guest segment registers from the current VMCS into the guest-CPU
6586 * context.
6587 *
6588 * @returns VBox status code.
6589 * @param pVCpu The cross context virtual CPU structure.
6590 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6591 * out-of-sync. Make sure to update the required fields
6592 * before using them.
6593 *
6594 * @remarks No-long-jump zone!!!
6595 */
6596static int hmR0VmxSaveGuestSegmentRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6597{
6598 /* Guest segment registers. */
6599 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SEGMENT_REGS))
6600 {
6601 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6602 AssertRCReturn(rc, rc);
6603
6604 rc = VMXLOCAL_READ_SEG(CS, cs);
6605 rc |= VMXLOCAL_READ_SEG(SS, ss);
6606 rc |= VMXLOCAL_READ_SEG(DS, ds);
6607 rc |= VMXLOCAL_READ_SEG(ES, es);
6608 rc |= VMXLOCAL_READ_SEG(FS, fs);
6609 rc |= VMXLOCAL_READ_SEG(GS, gs);
6610 AssertRCReturn(rc, rc);
6611
6612 /* Restore segment attributes for real-on-v86 mode hack. */
6613 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
6614 {
6615 pMixedCtx->cs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrCS.u;
6616 pMixedCtx->ss.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrSS.u;
6617 pMixedCtx->ds.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrDS.u;
6618 pMixedCtx->es.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrES.u;
6619 pMixedCtx->fs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrFS.u;
6620 pMixedCtx->gs.Attr.u = pVCpu->hm.s.vmx.RealMode.AttrGS.u;
6621 }
6622 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_SEGMENT_REGS);
6623 }
6624
6625 return VINF_SUCCESS;
6626}
6627
6628
6629/**
6630 * Saves the guest descriptor table registers and task register from the current
6631 * VMCS into the guest-CPU context.
6632 *
6633 * @returns VBox status code.
6634 * @param pVCpu The cross context virtual CPU structure.
6635 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6636 * out-of-sync. Make sure to update the required fields
6637 * before using them.
6638 *
6639 * @remarks No-long-jump zone!!!
6640 */
6641static int hmR0VmxSaveGuestTableRegs(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6642{
6643 int rc = VINF_SUCCESS;
6644
6645 /* Guest LDTR. */
6646 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LDTR))
6647 {
6648 rc = VMXLOCAL_READ_SEG(LDTR, ldtr);
6649 AssertRCReturn(rc, rc);
6650 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LDTR);
6651 }
6652
6653 /* Guest GDTR. */
6654 uint64_t u64Val = 0;
6655 uint32_t u32Val = 0;
6656 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_GDTR))
6657 {
6658 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
6659 rc |= VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val); AssertRCReturn(rc, rc);
6660 pMixedCtx->gdtr.pGdt = u64Val;
6661 pMixedCtx->gdtr.cbGdt = u32Val;
6662 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_GDTR);
6663 }
6664
6665 /* Guest IDTR. */
6666 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_IDTR))
6667 {
6668 rc = VMXReadVmcsGstN(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
6669 rc |= VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val); AssertRCReturn(rc, rc);
6670 pMixedCtx->idtr.pIdt = u64Val;
6671 pMixedCtx->idtr.cbIdt = u32Val;
6672 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_IDTR);
6673 }
6674
6675 /* Guest TR. */
6676 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_TR))
6677 {
6678 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6679 AssertRCReturn(rc, rc);
6680
6681 /* For real-mode emulation using virtual-8086 mode we have the fake TSS (pRealModeTSS) in TR, don't save the fake one. */
6682 if (!pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
6683 {
6684 rc = VMXLOCAL_READ_SEG(TR, tr);
6685 AssertRCReturn(rc, rc);
6686 }
6687 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_TR);
6688 }
6689 return rc;
6690}
6691
6692#undef VMXLOCAL_READ_SEG
6693
6694
6695/**
6696 * Saves the guest debug-register DR7 from the current VMCS into the guest-CPU
6697 * context.
6698 *
6699 * @returns VBox status code.
6700 * @param pVCpu The cross context virtual CPU structure.
6701 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6702 * out-of-sync. Make sure to update the required fields
6703 * before using them.
6704 *
6705 * @remarks No-long-jump zone!!!
6706 */
6707static int hmR0VmxSaveGuestDR7(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6708{
6709 if (!HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_DEBUG))
6710 {
6711 if (!pVCpu->hm.s.fUsingHyperDR7)
6712 {
6713 /* Upper 32-bits are always zero. See Intel spec. 2.7.3 "Loading and Storing Debug Registers". */
6714 uint32_t u32Val;
6715 int rc = VMXReadVmcs32(VMX_VMCS_GUEST_DR7, &u32Val); AssertRCReturn(rc, rc);
6716 pMixedCtx->dr[7] = u32Val;
6717 }
6718
6719 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_DEBUG);
6720 }
6721 return VINF_SUCCESS;
6722}
6723
6724
6725/**
6726 * Saves the guest APIC state from the current VMCS into the guest-CPU context.
6727 *
6728 * @returns VBox status code.
6729 * @param pVCpu The cross context virtual CPU structure.
6730 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
6731 * out-of-sync. Make sure to update the required fields
6732 * before using them.
6733 *
6734 * @remarks No-long-jump zone!!!
6735 */
6736static int hmR0VmxSaveGuestApicState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6737{
6738 NOREF(pMixedCtx);
6739
6740 /* Updating TPR is already done in hmR0VmxPostRunGuest(). Just update the flag. */
6741 HMVMXCPU_GST_SET_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_APIC_STATE);
6742 return VINF_SUCCESS;
6743}
6744
6745
6746/**
6747 * Saves the entire guest state from the currently active VMCS into the
6748 * guest-CPU context.
6749 *
6750 * This essentially VMREADs all guest-data.
6751 *
6752 * @returns VBox status code.
6753 * @param pVCpu The cross context virtual CPU structure.
6754 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
6755 * out-of-sync. Make sure to update the required fields
6756 * before using them.
6757 */
6758static int hmR0VmxSaveGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6759{
6760 Assert(pVCpu);
6761 Assert(pMixedCtx);
6762
6763 if (HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL)
6764 return VINF_SUCCESS;
6765
6766 /* Though we can longjmp to ring-3 due to log-flushes here and get recalled
6767 again on the ring-3 callback path, there is no real need to. */
6768 if (VMMRZCallRing3IsEnabled(pVCpu))
6769 VMMR0LogFlushDisable(pVCpu);
6770 else
6771 Assert(VMMR0IsLogFlushDisabled(pVCpu));
6772 Log4Func(("vcpu[%RU32]\n", pVCpu->idCpu));
6773
6774 int rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
6775 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestRipRspRflags failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6776
6777 rc = hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6778 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestControlRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6779
6780 rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
6781 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestSegmentRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6782
6783 rc = hmR0VmxSaveGuestTableRegs(pVCpu, pMixedCtx);
6784 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestTableRegs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6785
6786 rc = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
6787 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestDR7 failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6788
6789 rc = hmR0VmxSaveGuestSysenterMsrs(pVCpu, pMixedCtx);
6790 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestSysenterMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6791
6792 rc = hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
6793 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestLazyMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6794
6795 rc = hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
6796 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestAutoLoadStoreMsrs failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6797
6798 rc = hmR0VmxSaveGuestActivityState(pVCpu, pMixedCtx);
6799 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestActivityState failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6800
6801 rc = hmR0VmxSaveGuestApicState(pVCpu, pMixedCtx);
6802 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveGuestApicState failed! rc=%Rrc (pVCpu=%p)\n", rc, pVCpu), rc);
6803
6804 AssertMsg(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL,
6805 ("Missed guest state bits while saving state; missing %RX32 (got %RX32, want %RX32) - check log for any previous errors!\n",
6806 HMVMX_UPDATED_GUEST_ALL ^ HMVMXCPU_GST_VALUE(pVCpu), HMVMXCPU_GST_VALUE(pVCpu), HMVMX_UPDATED_GUEST_ALL));
6807
6808 if (VMMRZCallRing3IsEnabled(pVCpu))
6809 VMMR0LogFlushEnable(pVCpu);
6810
6811 return VINF_SUCCESS;
6812}
6813
6814
6815/**
6816 * Saves basic guest registers needed for IEM instruction execution.
6817 *
6818 * @returns VBox status code (OR-able).
6819 * @param pVCpu The cross context virtual CPU structure of the calling EMT.
6820 * @param pMixedCtx Pointer to the CPU context of the guest.
6821 * @param fMemory Whether the instruction being executed operates on
6822 * memory or not. Only CR0 is synced up if clear.
6823 * @param fNeedRsp Need RSP (any instruction working on GPRs or stack).
6824 */
6825static int hmR0VmxSaveGuestRegsForIemExec(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fMemory, bool fNeedRsp)
6826{
6827 /*
6828 * We assume all general purpose registers other than RSP are available.
6829 *
6830 * RIP is a must, as it will be incremented or otherwise changed.
6831 *
6832 * RFLAGS are always required to figure the CPL.
6833 *
6834 * RSP isn't always required, however it's a GPR, so frequently required.
6835 *
6836 * SS and CS are the only segment register needed if IEM doesn't do memory
6837 * access (CPL + 16/32/64-bit mode), but we can only get all segment registers.
6838 *
6839 * CR0 is always required by IEM for the CPL, while CR3 and CR4 will only
6840 * be required for memory accesses.
6841 *
6842 * Note! Before IEM dispatches an exception, it will call us to sync in everything.
6843 */
6844 int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
6845 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
6846 if (fNeedRsp)
6847 rc |= hmR0VmxSaveGuestRsp(pVCpu, pMixedCtx);
6848 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
6849 if (!fMemory)
6850 rc |= hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
6851 else
6852 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6853 AssertRCReturn(rc, rc);
6854 return rc;
6855}
6856
6857
6858/**
6859 * Ensures that we've got a complete basic guest-context.
6860 *
6861 * This excludes the FPU, SSE, AVX, and similar extended state. The interface
6862 * is for the interpreter.
6863 *
6864 * @returns VBox status code.
6865 * @param pVCpu The cross context virtual CPU structure of the calling EMT.
6866 * @param pMixedCtx Pointer to the guest-CPU context which may have data
6867 * needing to be synced in.
6868 * @thread EMT(pVCpu)
6869 */
6870VMMR0_INT_DECL(int) HMR0EnsureCompleteBasicContext(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
6871{
6872 /* Note! Since this is only applicable to VT-x, the implementation is placed
6873 in the VT-x part of the sources instead of the generic stuff. */
6874 if (pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fSupported)
6875 {
6876 /* For now, imply that the caller might change everything too. */
6877 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
6878 return hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
6879 }
6880 return VINF_SUCCESS;
6881}
6882
6883
6884/**
6885 * Check per-VM and per-VCPU force flag actions that require us to go back to
6886 * ring-3 for one reason or another.
6887 *
6888 * @returns Strict VBox status code (i.e. informational status codes too)
6889 * @retval VINF_SUCCESS if we don't have any actions that require going back to
6890 * ring-3.
6891 * @retval VINF_PGM_SYNC_CR3 if we have pending PGM CR3 sync.
6892 * @retval VINF_EM_PENDING_REQUEST if we have pending requests (like hardware
6893 * interrupts)
6894 * @retval VINF_PGM_POOL_FLUSH_PENDING if PGM is doing a pool flush and requires
6895 * all EMTs to be in ring-3.
6896 * @retval VINF_EM_RAW_TO_R3 if there is pending DMA requests.
6897 * @retval VINF_EM_NO_MEMORY PGM is out of memory, we need to return
6898 * to the EM loop.
6899 *
6900 * @param pVM The cross context VM structure.
6901 * @param pVCpu The cross context virtual CPU structure.
6902 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
6903 * out-of-sync. Make sure to update the required fields
6904 * before using them.
6905 * @param fStepping Running in hmR0VmxRunGuestCodeStep().
6906 */
6907static VBOXSTRICTRC hmR0VmxCheckForceFlags(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fStepping)
6908{
6909 Assert(VMMRZCallRing3IsEnabled(pVCpu));
6910
6911 /*
6912 * Anything pending? Should be more likely than not if we're doing a good job.
6913 */
6914 if ( !fStepping
6915 ? !VM_FF_IS_PENDING(pVM, VM_FF_HP_R0_PRE_HM_MASK)
6916 && !VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HP_R0_PRE_HM_MASK)
6917 : !VM_FF_IS_PENDING(pVM, VM_FF_HP_R0_PRE_HM_STEP_MASK)
6918 && !VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HP_R0_PRE_HM_STEP_MASK) )
6919 return VINF_SUCCESS;
6920
6921 /* We need the control registers now, make sure the guest-CPU context is updated. */
6922 int rc3 = hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
6923 AssertRCReturn(rc3, rc3);
6924
6925 /* Pending HM CR3 sync. */
6926 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3))
6927 {
6928 int rc2 = PGMUpdateCR3(pVCpu, pMixedCtx->cr3);
6929 AssertMsgReturn(rc2 == VINF_SUCCESS || rc2 == VINF_PGM_SYNC_CR3,
6930 ("%Rrc\n", rc2), RT_FAILURE_NP(rc2) ? rc2 : VERR_IPE_UNEXPECTED_INFO_STATUS);
6931 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_CR3));
6932 }
6933
6934 /* Pending HM PAE PDPEs. */
6935 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES))
6936 {
6937 PGMGstUpdatePaePdpes(pVCpu, &pVCpu->hm.s.aPdpes[0]);
6938 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_UPDATE_PAE_PDPES));
6939 }
6940
6941 /* Pending PGM C3 sync. */
6942 if (VMCPU_FF_IS_PENDING(pVCpu,VMCPU_FF_PGM_SYNC_CR3 | VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL))
6943 {
6944 VBOXSTRICTRC rcStrict2 = PGMSyncCR3(pVCpu, pMixedCtx->cr0, pMixedCtx->cr3, pMixedCtx->cr4,
6945 VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_PGM_SYNC_CR3));
6946 if (rcStrict2 != VINF_SUCCESS)
6947 {
6948 AssertRC(VBOXSTRICTRC_VAL(rcStrict2));
6949 Log4(("hmR0VmxCheckForceFlags: PGMSyncCR3 forcing us back to ring-3. rc2=%d\n", VBOXSTRICTRC_VAL(rcStrict2)));
6950 return rcStrict2;
6951 }
6952 }
6953
6954 /* Pending HM-to-R3 operations (critsects, timers, EMT rendezvous etc.) */
6955 if ( VM_FF_IS_PENDING(pVM, VM_FF_HM_TO_R3_MASK)
6956 || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
6957 {
6958 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
6959 int rc2 = RT_UNLIKELY(VM_FF_IS_PENDING(pVM, VM_FF_PGM_NO_MEMORY)) ? VINF_EM_NO_MEMORY : VINF_EM_RAW_TO_R3;
6960 Log4(("hmR0VmxCheckForceFlags: HM_TO_R3 forcing us back to ring-3. rc=%d\n", rc2));
6961 return rc2;
6962 }
6963
6964 /* Pending VM request packets, such as hardware interrupts. */
6965 if ( VM_FF_IS_PENDING(pVM, VM_FF_REQUEST)
6966 || VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_REQUEST))
6967 {
6968 Log4(("hmR0VmxCheckForceFlags: Pending VM request forcing us back to ring-3\n"));
6969 return VINF_EM_PENDING_REQUEST;
6970 }
6971
6972 /* Pending PGM pool flushes. */
6973 if (VM_FF_IS_PENDING(pVM, VM_FF_PGM_POOL_FLUSH_PENDING))
6974 {
6975 Log4(("hmR0VmxCheckForceFlags: PGM pool flush pending forcing us back to ring-3\n"));
6976 return VINF_PGM_POOL_FLUSH_PENDING;
6977 }
6978
6979 /* Pending DMA requests. */
6980 if (VM_FF_IS_PENDING(pVM, VM_FF_PDM_DMA))
6981 {
6982 Log4(("hmR0VmxCheckForceFlags: Pending DMA request forcing us back to ring-3\n"));
6983 return VINF_EM_RAW_TO_R3;
6984 }
6985
6986 return VINF_SUCCESS;
6987}
6988
6989
6990/**
6991 * Converts any TRPM trap into a pending HM event. This is typically used when
6992 * entering from ring-3 (not longjmp returns).
6993 *
6994 * @param pVCpu The cross context virtual CPU structure.
6995 */
6996static void hmR0VmxTrpmTrapToPendingEvent(PVMCPU pVCpu)
6997{
6998 Assert(TRPMHasTrap(pVCpu));
6999 Assert(!pVCpu->hm.s.Event.fPending);
7000
7001 uint8_t uVector;
7002 TRPMEVENT enmTrpmEvent;
7003 RTGCUINT uErrCode;
7004 RTGCUINTPTR GCPtrFaultAddress;
7005 uint8_t cbInstr;
7006
7007 int rc = TRPMQueryTrapAll(pVCpu, &uVector, &enmTrpmEvent, &uErrCode, &GCPtrFaultAddress, &cbInstr);
7008 AssertRC(rc);
7009
7010 /* Refer Intel spec. 24.8.3 "VM-entry Controls for Event Injection" for the format of u32IntInfo. */
7011 uint32_t u32IntInfo = uVector | VMX_EXIT_INTERRUPTION_INFO_VALID;
7012 if (enmTrpmEvent == TRPM_TRAP)
7013 {
7014 switch (uVector)
7015 {
7016 case X86_XCPT_NMI:
7017 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7018 break;
7019
7020 case X86_XCPT_BP:
7021 case X86_XCPT_OF:
7022 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7023 break;
7024
7025 case X86_XCPT_PF:
7026 case X86_XCPT_DF:
7027 case X86_XCPT_TS:
7028 case X86_XCPT_NP:
7029 case X86_XCPT_SS:
7030 case X86_XCPT_GP:
7031 case X86_XCPT_AC:
7032 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7033 /* no break! */
7034 default:
7035 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7036 break;
7037 }
7038 }
7039 else if (enmTrpmEvent == TRPM_HARDWARE_INT)
7040 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7041 else if (enmTrpmEvent == TRPM_SOFTWARE_INT)
7042 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7043 else
7044 AssertMsgFailed(("Invalid TRPM event type %d\n", enmTrpmEvent));
7045
7046 rc = TRPMResetTrap(pVCpu);
7047 AssertRC(rc);
7048 Log4(("TRPM->HM event: u32IntInfo=%#RX32 enmTrpmEvent=%d cbInstr=%u uErrCode=%#RX32 GCPtrFaultAddress=%#RGv\n",
7049 u32IntInfo, enmTrpmEvent, cbInstr, uErrCode, GCPtrFaultAddress));
7050
7051 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, uErrCode, GCPtrFaultAddress);
7052}
7053
7054
7055/**
7056 * Converts the pending HM event into a TRPM trap.
7057 *
7058 * @param pVCpu The cross context virtual CPU structure.
7059 */
7060static void hmR0VmxPendingEventToTrpmTrap(PVMCPU pVCpu)
7061{
7062 Assert(pVCpu->hm.s.Event.fPending);
7063
7064 uint32_t uVectorType = VMX_IDT_VECTORING_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
7065 uint32_t uVector = VMX_IDT_VECTORING_INFO_VECTOR(pVCpu->hm.s.Event.u64IntInfo);
7066 bool fErrorCodeValid = VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVCpu->hm.s.Event.u64IntInfo);
7067 uint32_t uErrorCode = pVCpu->hm.s.Event.u32ErrCode;
7068
7069 /* If a trap was already pending, we did something wrong! */
7070 Assert(TRPMQueryTrap(pVCpu, NULL /* pu8TrapNo */, NULL /* pEnmType */) == VERR_TRPM_NO_ACTIVE_TRAP);
7071
7072 TRPMEVENT enmTrapType;
7073 switch (uVectorType)
7074 {
7075 case VMX_IDT_VECTORING_INFO_TYPE_EXT_INT:
7076 enmTrapType = TRPM_HARDWARE_INT;
7077 break;
7078
7079 case VMX_IDT_VECTORING_INFO_TYPE_SW_INT:
7080 enmTrapType = TRPM_SOFTWARE_INT;
7081 break;
7082
7083 case VMX_IDT_VECTORING_INFO_TYPE_NMI:
7084 case VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT:
7085 case VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT: /* #BP and #OF */
7086 case VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT:
7087 enmTrapType = TRPM_TRAP;
7088 break;
7089
7090 default:
7091 AssertMsgFailed(("Invalid trap type %#x\n", uVectorType));
7092 enmTrapType = TRPM_32BIT_HACK;
7093 break;
7094 }
7095
7096 Log4(("HM event->TRPM: uVector=%#x enmTrapType=%d\n", uVector, enmTrapType));
7097
7098 int rc = TRPMAssertTrap(pVCpu, uVector, enmTrapType);
7099 AssertRC(rc);
7100
7101 if (fErrorCodeValid)
7102 TRPMSetErrorCode(pVCpu, uErrorCode);
7103
7104 if ( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
7105 && uVector == X86_XCPT_PF)
7106 {
7107 TRPMSetFaultAddress(pVCpu, pVCpu->hm.s.Event.GCPtrFaultAddress);
7108 }
7109 else if ( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
7110 || uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_XCPT
7111 || uVectorType == VMX_IDT_VECTORING_INFO_TYPE_PRIV_SW_XCPT)
7112 {
7113 AssertMsg( uVectorType == VMX_IDT_VECTORING_INFO_TYPE_SW_INT
7114 || (uVector == X86_XCPT_BP || uVector == X86_XCPT_OF),
7115 ("Invalid vector: uVector=%#x uVectorType=%#x\n", uVector, uVectorType));
7116 TRPMSetInstrLength(pVCpu, pVCpu->hm.s.Event.cbInstr);
7117 }
7118
7119 /* Clear any pending events from the VMCS. */
7120 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, 0); AssertRC(rc);
7121 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, 0); AssertRC(rc);
7122
7123 /* We're now done converting the pending event. */
7124 pVCpu->hm.s.Event.fPending = false;
7125}
7126
7127
7128/**
7129 * Does the necessary state syncing before returning to ring-3 for any reason
7130 * (longjmp, preemption, voluntary exits to ring-3) from VT-x.
7131 *
7132 * @returns VBox status code.
7133 * @param pVCpu The cross context virtual CPU structure.
7134 * @param pMixedCtx Pointer to the guest-CPU context. The data may
7135 * be out-of-sync. Make sure to update the required
7136 * fields before using them.
7137 * @param fSaveGuestState Whether to save the guest state or not.
7138 *
7139 * @remarks No-long-jmp zone!!!
7140 */
7141static int hmR0VmxLeave(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fSaveGuestState)
7142{
7143 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
7144 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
7145
7146 RTCPUID idCpu = RTMpCpuId();
7147 Log4Func(("HostCpuId=%u\n", idCpu));
7148
7149 /*
7150 * !!! IMPORTANT !!!
7151 * If you modify code here, check whether hmR0VmxCallRing3Callback() needs to be updated too.
7152 */
7153
7154 /* Save the guest state if necessary. */
7155 if ( fSaveGuestState
7156 && HMVMXCPU_GST_VALUE(pVCpu) != HMVMX_UPDATED_GUEST_ALL)
7157 {
7158 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
7159 AssertRCReturn(rc, rc);
7160 Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
7161 }
7162
7163 /* Restore host FPU state if necessary and resync on next R0 reentry .*/
7164 if (CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu))
7165 {
7166 if (fSaveGuestState)
7167 {
7168 /* We shouldn't reload CR0 without saving it first. */
7169 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
7170 AssertRCReturn(rc, rc);
7171 }
7172 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
7173 }
7174
7175 /* Restore host debug registers if necessary and resync on next R0 reentry. */
7176#ifdef VBOX_STRICT
7177 if (CPUMIsHyperDebugStateActive(pVCpu))
7178 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT);
7179#endif
7180 if (CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */))
7181 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
7182 Assert(!CPUMIsGuestDebugStateActive(pVCpu) && !CPUMIsGuestDebugStateActivePending(pVCpu));
7183 Assert(!CPUMIsHyperDebugStateActive(pVCpu) && !CPUMIsHyperDebugStateActivePending(pVCpu));
7184
7185#if HC_ARCH_BITS == 64
7186 /* Restore host-state bits that VT-x only restores partially. */
7187 if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
7188 && (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
7189 {
7190 Log4Func(("Restoring Host State: fRestoreHostFlags=%#RX32 HostCpuId=%u\n", pVCpu->hm.s.vmx.fRestoreHostFlags, idCpu));
7191 VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
7192 }
7193 pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
7194#endif
7195
7196 /* Restore the lazy host MSRs as we're leaving VT-x context. */
7197 if (pVCpu->hm.s.vmx.fLazyMsrs)
7198 {
7199 /* We shouldn't reload the guest MSRs without saving it first. */
7200 if (!fSaveGuestState)
7201 {
7202 int rc = hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
7203 AssertRCReturn(rc, rc);
7204 }
7205 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_LAZY_MSRS));
7206 hmR0VmxLazyRestoreHostMsrs(pVCpu);
7207 Assert(!pVCpu->hm.s.vmx.fLazyMsrs);
7208 }
7209
7210 /* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
7211 pVCpu->hm.s.vmx.fUpdatedHostMsrs = false;
7212
7213 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatEntry);
7214 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatLoadGuestState);
7215 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit1);
7216 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExit2);
7217 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitIO);
7218 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitMovCRx);
7219 STAM_PROFILE_ADV_SET_STOPPED(&pVCpu->hm.s.StatExitXcptNmi);
7220 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
7221
7222 VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
7223
7224 /** @todo This partially defeats the purpose of having preemption hooks.
7225 * The problem is, deregistering the hooks should be moved to a place that
7226 * lasts until the EMT is about to be destroyed not everytime while leaving HM
7227 * context.
7228 */
7229 if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_ACTIVE)
7230 {
7231 int rc = VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
7232 AssertRCReturn(rc, rc);
7233
7234 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
7235 Log4Func(("Cleared Vmcs. HostCpuId=%u\n", idCpu));
7236 }
7237 Assert(!(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_LAUNCHED));
7238 NOREF(idCpu);
7239
7240 return VINF_SUCCESS;
7241}
7242
7243
7244/**
7245 * Leaves the VT-x session.
7246 *
7247 * @returns VBox status code.
7248 * @param pVCpu The cross context virtual CPU structure.
7249 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7250 * out-of-sync. Make sure to update the required fields
7251 * before using them.
7252 *
7253 * @remarks No-long-jmp zone!!!
7254 */
7255DECLINLINE(int) hmR0VmxLeaveSession(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7256{
7257 HM_DISABLE_PREEMPT();
7258 HMVMX_ASSERT_CPU_SAFE();
7259 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
7260 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
7261
7262 /* When thread-context hooks are used, we can avoid doing the leave again if we had been preempted before
7263 and done this from the VMXR0ThreadCtxCallback(). */
7264 if (!pVCpu->hm.s.fLeaveDone)
7265 {
7266 int rc2 = hmR0VmxLeave(pVCpu, pMixedCtx, true /* fSaveGuestState */);
7267 AssertRCReturnStmt(rc2, HM_RESTORE_PREEMPT(), rc2);
7268 pVCpu->hm.s.fLeaveDone = true;
7269 }
7270 Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
7271
7272 /*
7273 * !!! IMPORTANT !!!
7274 * If you modify code here, make sure to check whether hmR0VmxCallRing3Callback() needs to be updated too.
7275 */
7276
7277 /* Deregister hook now that we've left HM context before re-enabling preemption. */
7278 /** @todo Deregistering here means we need to VMCLEAR always
7279 * (longjmp/exit-to-r3) in VT-x which is not efficient. */
7280 /** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
7281 VMMR0ThreadCtxHookDisable(pVCpu);
7282
7283 /* Leave HM context. This takes care of local init (term). */
7284 int rc = HMR0LeaveCpu(pVCpu);
7285
7286 HM_RESTORE_PREEMPT();
7287 return rc;
7288}
7289
7290
7291/**
7292 * Does the necessary state syncing before doing a longjmp to ring-3.
7293 *
7294 * @returns VBox status code.
7295 * @param pVCpu The cross context virtual CPU structure.
7296 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7297 * out-of-sync. Make sure to update the required fields
7298 * before using them.
7299 *
7300 * @remarks No-long-jmp zone!!!
7301 */
7302DECLINLINE(int) hmR0VmxLongJmpToRing3(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7303{
7304 return hmR0VmxLeaveSession(pVCpu, pMixedCtx);
7305}
7306
7307
7308/**
7309 * Take necessary actions before going back to ring-3.
7310 *
7311 * An action requires us to go back to ring-3. This function does the necessary
7312 * steps before we can safely return to ring-3. This is not the same as longjmps
7313 * to ring-3, this is voluntary and prepares the guest so it may continue
7314 * executing outside HM (recompiler/IEM).
7315 *
7316 * @returns VBox status code.
7317 * @param pVM The cross context VM structure.
7318 * @param pVCpu The cross context virtual CPU structure.
7319 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7320 * out-of-sync. Make sure to update the required fields
7321 * before using them.
7322 * @param rcExit The reason for exiting to ring-3. Can be
7323 * VINF_VMM_UNKNOWN_RING3_CALL.
7324 */
7325static int hmR0VmxExitToRing3(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, VBOXSTRICTRC rcExit)
7326{
7327 Assert(pVM);
7328 Assert(pVCpu);
7329 Assert(pMixedCtx);
7330 HMVMX_ASSERT_PREEMPT_SAFE();
7331
7332 if (RT_UNLIKELY(rcExit == VERR_VMX_INVALID_VMCS_PTR))
7333 {
7334 VMXGetActivatedVmcs(&pVCpu->hm.s.vmx.LastError.u64VMCSPhys);
7335 pVCpu->hm.s.vmx.LastError.u32VMCSRevision = *(uint32_t *)pVCpu->hm.s.vmx.pvVmcs;
7336 pVCpu->hm.s.vmx.LastError.idEnteredCpu = pVCpu->hm.s.idEnteredCpu;
7337 /* LastError.idCurrentCpu was updated in hmR0VmxPreRunGuestCommitted(). */
7338 }
7339
7340 /* Please, no longjumps here (any logging shouldn't flush jump back to ring-3). NO LOGGING BEFORE THIS POINT! */
7341 VMMRZCallRing3Disable(pVCpu);
7342 Log4(("hmR0VmxExitToRing3: pVCpu=%p idCpu=%RU32 rcExit=%d\n", pVCpu, pVCpu->idCpu, VBOXSTRICTRC_VAL(rcExit)));
7343
7344 /* We need to do this only while truly exiting the "inner loop" back to ring-3 and -not- for any longjmp to ring3. */
7345 if (pVCpu->hm.s.Event.fPending)
7346 {
7347 hmR0VmxPendingEventToTrpmTrap(pVCpu);
7348 Assert(!pVCpu->hm.s.Event.fPending);
7349 }
7350
7351 /* Clear interrupt-window and NMI-window controls as we re-evaluate it when we return from ring-3. */
7352 hmR0VmxClearIntNmiWindowsVmcs(pVCpu);
7353
7354 /* If we're emulating an instruction, we shouldn't have any TRPM traps pending
7355 and if we're injecting an event we should have a TRPM trap pending. */
7356 AssertMsg(rcExit != VINF_EM_RAW_INJECT_TRPM_EVENT || TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
7357#ifndef DEBUG_bird /* Triggered after firing an NMI against NT4SP1, possibly a tripple fault in progress. */
7358 AssertMsg(rcExit != VINF_EM_RAW_EMULATE_INSTR || !TRPMHasTrap(pVCpu), ("%Rrc\n", VBOXSTRICTRC_VAL(rcExit)));
7359#endif
7360
7361 /* Save guest state and restore host state bits. */
7362 int rc = hmR0VmxLeaveSession(pVCpu, pMixedCtx);
7363 AssertRCReturn(rc, rc);
7364 STAM_COUNTER_DEC(&pVCpu->hm.s.StatSwitchLongJmpToR3);
7365 /* Thread-context hooks are unregistered at this point!!! */
7366
7367 /* Sync recompiler state. */
7368 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_TO_R3);
7369 CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_SYSENTER_MSR
7370 | CPUM_CHANGED_LDTR
7371 | CPUM_CHANGED_GDTR
7372 | CPUM_CHANGED_IDTR
7373 | CPUM_CHANGED_TR
7374 | CPUM_CHANGED_HIDDEN_SEL_REGS);
7375 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
7376 if ( pVM->hm.s.fNestedPaging
7377 && CPUMIsGuestPagingEnabledEx(pMixedCtx))
7378 {
7379 CPUMSetChangedFlags(pVCpu, CPUM_CHANGED_GLOBAL_TLB_FLUSH);
7380 }
7381
7382 Assert(!pVCpu->hm.s.fClearTrapFlag);
7383
7384 /* On our way back from ring-3 reload the guest state if there is a possibility of it being changed. */
7385 if (rcExit != VINF_EM_RAW_INTERRUPT)
7386 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
7387
7388 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchExitToR3);
7389
7390 /* We do -not- want any longjmp notifications after this! We must return to ring-3 ASAP. */
7391 VMMRZCallRing3RemoveNotification(pVCpu);
7392 VMMRZCallRing3Enable(pVCpu);
7393
7394 return rc;
7395}
7396
7397
7398/**
7399 * VMMRZCallRing3() callback wrapper which saves the guest state before we
7400 * longjump to ring-3 and possibly get preempted.
7401 *
7402 * @returns VBox status code.
7403 * @param pVCpu The cross context virtual CPU structure.
7404 * @param enmOperation The operation causing the ring-3 longjump.
7405 * @param pvUser Opaque pointer to the guest-CPU context. The data
7406 * may be out-of-sync. Make sure to update the required
7407 * fields before using them.
7408 */
7409static DECLCALLBACK(int) hmR0VmxCallRing3Callback(PVMCPU pVCpu, VMMCALLRING3 enmOperation, void *pvUser)
7410{
7411 if (enmOperation == VMMCALLRING3_VM_R0_ASSERTION)
7412 {
7413 /*
7414 * !!! IMPORTANT !!!
7415 * If you modify code here, check whether hmR0VmxLeave() and hmR0VmxLeaveSession() needs to be updated too.
7416 * This is a stripped down version which gets out ASAP, trying to not trigger any further assertions.
7417 */
7418 VMMRZCallRing3RemoveNotification(pVCpu);
7419 VMMRZCallRing3Disable(pVCpu);
7420 RTTHREADPREEMPTSTATE PreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
7421 RTThreadPreemptDisable(&PreemptState);
7422
7423 CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVCpu);
7424 CPUMR0DebugStateMaybeSaveGuestAndRestoreHost(pVCpu, true /* save DR6 */);
7425
7426#if HC_ARCH_BITS == 64
7427 /* Restore host-state bits that VT-x only restores partially. */
7428 if ( (pVCpu->hm.s.vmx.fRestoreHostFlags & VMX_RESTORE_HOST_REQUIRED)
7429 && (pVCpu->hm.s.vmx.fRestoreHostFlags & ~VMX_RESTORE_HOST_REQUIRED))
7430 VMXRestoreHostState(pVCpu->hm.s.vmx.fRestoreHostFlags, &pVCpu->hm.s.vmx.RestoreHost);
7431 pVCpu->hm.s.vmx.fRestoreHostFlags = 0;
7432#endif
7433 /* Restore the lazy host MSRs as we're leaving VT-x context. */
7434 if (pVCpu->hm.s.vmx.fLazyMsrs)
7435 hmR0VmxLazyRestoreHostMsrs(pVCpu);
7436
7437 /* Update auto-load/store host MSRs values when we re-enter VT-x (as we could be on a different CPU). */
7438 pVCpu->hm.s.vmx.fUpdatedHostMsrs = false;
7439 VMCPU_CMPXCHG_STATE(pVCpu, VMCPUSTATE_STARTED_HM, VMCPUSTATE_STARTED_EXEC);
7440 if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_ACTIVE)
7441 {
7442 VMXClearVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
7443 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_CLEAR;
7444 }
7445
7446 /** @todo eliminate the need for calling VMMR0ThreadCtxHookDisable here! */
7447 VMMR0ThreadCtxHookDisable(pVCpu);
7448 HMR0LeaveCpu(pVCpu);
7449 RTThreadPreemptRestore(&PreemptState);
7450 return VINF_SUCCESS;
7451 }
7452
7453 Assert(pVCpu);
7454 Assert(pvUser);
7455 Assert(VMMRZCallRing3IsEnabled(pVCpu));
7456 HMVMX_ASSERT_PREEMPT_SAFE();
7457
7458 VMMRZCallRing3Disable(pVCpu);
7459 Assert(VMMR0IsLogFlushDisabled(pVCpu));
7460
7461 Log4(("hmR0VmxCallRing3Callback->hmR0VmxLongJmpToRing3 pVCpu=%p idCpu=%RU32 enmOperation=%d\n", pVCpu, pVCpu->idCpu,
7462 enmOperation));
7463
7464 int rc = hmR0VmxLongJmpToRing3(pVCpu, (PCPUMCTX)pvUser);
7465 AssertRCReturn(rc, rc);
7466
7467 VMMRZCallRing3Enable(pVCpu);
7468 return VINF_SUCCESS;
7469}
7470
7471
7472/**
7473 * Sets the interrupt-window exiting control in the VMCS which instructs VT-x to
7474 * cause a VM-exit as soon as the guest is in a state to receive interrupts.
7475 *
7476 * @param pVCpu The cross context virtual CPU structure.
7477 */
7478DECLINLINE(void) hmR0VmxSetIntWindowExitVmcs(PVMCPU pVCpu)
7479{
7480 if (RT_LIKELY(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT))
7481 {
7482 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT))
7483 {
7484 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT;
7485 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7486 AssertRC(rc);
7487 Log4(("Setup interrupt-window exiting\n"));
7488 }
7489 } /* else we will deliver interrupts whenever the guest exits next and is in a state to receive events. */
7490}
7491
7492
7493/**
7494 * Clears the interrupt-window exiting control in the VMCS.
7495 *
7496 * @param pVCpu The cross context virtual CPU structure.
7497 */
7498DECLINLINE(void) hmR0VmxClearIntWindowExitVmcs(PVMCPU pVCpu)
7499{
7500 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT);
7501 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT;
7502 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7503 AssertRC(rc);
7504 Log4(("Cleared interrupt-window exiting\n"));
7505}
7506
7507
7508/**
7509 * Sets the NMI-window exiting control in the VMCS which instructs VT-x to
7510 * cause a VM-exit as soon as the guest is in a state to receive NMIs.
7511 *
7512 * @param pVCpu The cross context virtual CPU structure.
7513 */
7514DECLINLINE(void) hmR0VmxSetNmiWindowExitVmcs(PVMCPU pVCpu)
7515{
7516 if (RT_LIKELY(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT))
7517 {
7518 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT))
7519 {
7520 pVCpu->hm.s.vmx.u32ProcCtls |= VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT;
7521 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7522 AssertRC(rc);
7523 Log4(("Setup NMI-window exiting\n"));
7524 }
7525 } /* else we will deliver NMIs whenever we VM-exit next, even possibly nesting NMIs. Can't be helped on ancient CPUs. */
7526}
7527
7528
7529/**
7530 * Clears the NMI-window exiting control in the VMCS.
7531 *
7532 * @param pVCpu The cross context virtual CPU structure.
7533 */
7534DECLINLINE(void) hmR0VmxClearNmiWindowExitVmcs(PVMCPU pVCpu)
7535{
7536 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT);
7537 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT;
7538 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
7539 AssertRC(rc);
7540 Log4(("Cleared NMI-window exiting\n"));
7541}
7542
7543
7544/**
7545 * Evaluates the event to be delivered to the guest and sets it as the pending
7546 * event.
7547 *
7548 * @returns The VT-x guest-interruptibility state.
7549 * @param pVCpu The cross context virtual CPU structure.
7550 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7551 * out-of-sync. Make sure to update the required fields
7552 * before using them.
7553 */
7554static uint32_t hmR0VmxEvaluatePendingEvent(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7555{
7556 /* Get the current interruptibility-state of the guest and then figure out what can be injected. */
7557 uint32_t const uIntrState = hmR0VmxGetGuestIntrState(pVCpu, pMixedCtx);
7558 bool const fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7559 bool const fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7560 bool const fBlockNmi = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI);
7561
7562 Assert(!fBlockSti || HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS));
7563 Assert(!(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
7564 Assert(!fBlockSti || pMixedCtx->eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
7565 Assert(!TRPMHasTrap(pVCpu));
7566
7567 if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UPDATE_APIC))
7568 APICUpdatePendingInterrupts(pVCpu);
7569
7570 /*
7571 * Toggling of interrupt force-flags here is safe since we update TRPM on premature exits
7572 * to ring-3 before executing guest code, see hmR0VmxExitToRing3(). We must NOT restore these force-flags.
7573 */
7574 /** @todo SMI. SMIs take priority over NMIs. */
7575 if (VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INTERRUPT_NMI)) /* NMI. NMIs take priority over regular interrupts. */
7576 {
7577 /* On some CPUs block-by-STI also blocks NMIs. See Intel spec. 26.3.1.5 "Checks On Guest Non-Register State". */
7578 if ( !pVCpu->hm.s.Event.fPending
7579 && !fBlockNmi
7580 && !fBlockSti
7581 && !fBlockMovSS)
7582 {
7583 Log4(("Pending NMI vcpu[%RU32]\n", pVCpu->idCpu));
7584 uint32_t u32IntInfo = X86_XCPT_NMI | VMX_EXIT_INTERRUPTION_INFO_VALID;
7585 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7586
7587 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7588 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INTERRUPT_NMI);
7589 }
7590 else
7591 hmR0VmxSetNmiWindowExitVmcs(pVCpu);
7592 }
7593 /*
7594 * Check if the guest can receive external interrupts (PIC/APIC). Once PDMGetInterrupt() returns
7595 * a valid interrupt we must- deliver the interrupt. We can no longer re-request it from the APIC.
7596 */
7597 else if ( VMCPU_FF_IS_PENDING(pVCpu, (VMCPU_FF_INTERRUPT_APIC | VMCPU_FF_INTERRUPT_PIC))
7598 && !pVCpu->hm.s.fSingleInstruction)
7599 {
7600 Assert(!DBGFIsStepping(pVCpu));
7601 int rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
7602 AssertRC(rc);
7603 bool const fBlockInt = !(pMixedCtx->eflags.u32 & X86_EFL_IF);
7604 if ( !pVCpu->hm.s.Event.fPending
7605 && !fBlockInt
7606 && !fBlockSti
7607 && !fBlockMovSS)
7608 {
7609 uint8_t u8Interrupt;
7610 rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
7611 if (RT_SUCCESS(rc))
7612 {
7613 Log4(("Pending interrupt vcpu[%RU32] u8Interrupt=%#x \n", pVCpu->idCpu, u8Interrupt));
7614 uint32_t u32IntInfo = u8Interrupt | VMX_EXIT_INTERRUPTION_INFO_VALID;
7615 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7616
7617 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrfaultAddress */);
7618 }
7619 else if (rc == VERR_APIC_INTR_MASKED_BY_TPR)
7620 {
7621 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
7622 hmR0VmxApicSetTprThreshold(pVCpu, u8Interrupt >> 4);
7623 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchTprMaskedIrq);
7624
7625 /*
7626 * If the CPU doesn't have TPR shadowing, we will always get a VM-exit on TPR changes and
7627 * APICSetTpr() will end up setting the VMCPU_FF_INTERRUPT_APIC if required, so there is no
7628 * need to re-set this force-flag here.
7629 */
7630 }
7631 else
7632 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchGuestIrq);
7633 }
7634 else
7635 hmR0VmxSetIntWindowExitVmcs(pVCpu);
7636 }
7637
7638 return uIntrState;
7639}
7640
7641
7642/**
7643 * Sets a pending-debug exception to be delivered to the guest if the guest is
7644 * single-stepping in the VMCS.
7645 *
7646 * @param pVCpu The cross context virtual CPU structure.
7647 */
7648DECLINLINE(void) hmR0VmxSetPendingDebugXcptVmcs(PVMCPU pVCpu)
7649{
7650 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS)); NOREF(pVCpu);
7651 int rc = VMXWriteVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, VMX_VMCS_GUEST_DEBUG_EXCEPTIONS_BS);
7652 AssertRC(rc);
7653}
7654
7655
7656/**
7657 * Injects any pending events into the guest if the guest is in a state to
7658 * receive them.
7659 *
7660 * @returns Strict VBox status code (i.e. informational status codes too).
7661 * @param pVCpu The cross context virtual CPU structure.
7662 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7663 * out-of-sync. Make sure to update the required fields
7664 * before using them.
7665 * @param uIntrState The VT-x guest-interruptibility state.
7666 * @param fStepping Running in hmR0VmxRunGuestCodeStep() and we should
7667 * return VINF_EM_DBG_STEPPED if the event was
7668 * dispatched directly.
7669 */
7670static VBOXSTRICTRC hmR0VmxInjectPendingEvent(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t uIntrState, bool fStepping)
7671{
7672 HMVMX_ASSERT_PREEMPT_SAFE();
7673 Assert(VMMRZCallRing3IsEnabled(pVCpu));
7674
7675 bool fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7676 bool fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7677
7678 Assert(!fBlockSti || HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RFLAGS));
7679 Assert(!(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI)); /* We don't support block-by-SMI yet.*/
7680 Assert(!fBlockSti || pMixedCtx->eflags.Bits.u1IF); /* Cannot set block-by-STI when interrupts are disabled. */
7681 Assert(!TRPMHasTrap(pVCpu));
7682
7683 VBOXSTRICTRC rcStrict = VINF_SUCCESS;
7684 if (pVCpu->hm.s.Event.fPending)
7685 {
7686 /*
7687 * Do -not- clear any interrupt-window exiting control here. We might have an interrupt
7688 * pending even while injecting an event and in this case, we want a VM-exit as soon as
7689 * the guest is ready for the next interrupt, see @bugref{6208#c45}.
7690 *
7691 * See Intel spec. 26.6.5 "Interrupt-Window Exiting and Virtual-Interrupt Delivery".
7692 */
7693 uint32_t const uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVCpu->hm.s.Event.u64IntInfo);
7694#ifdef VBOX_STRICT
7695 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
7696 {
7697 bool const fBlockInt = !(pMixedCtx->eflags.u32 & X86_EFL_IF);
7698 Assert(!fBlockInt);
7699 Assert(!fBlockSti);
7700 Assert(!fBlockMovSS);
7701 }
7702 else if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
7703 {
7704 bool const fBlockNmi = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI);
7705 Assert(!fBlockSti);
7706 Assert(!fBlockMovSS);
7707 Assert(!fBlockNmi);
7708 }
7709#endif
7710 Log4(("Injecting pending event vcpu[%RU32] u64IntInfo=%#RX64 Type=%#x\n", pVCpu->idCpu, pVCpu->hm.s.Event.u64IntInfo,
7711 (uint8_t)uIntType));
7712 rcStrict = hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, pVCpu->hm.s.Event.u64IntInfo, pVCpu->hm.s.Event.cbInstr,
7713 pVCpu->hm.s.Event.u32ErrCode, pVCpu->hm.s.Event.GCPtrFaultAddress,
7714 fStepping, &uIntrState);
7715 AssertRCReturn(VBOXSTRICTRC_VAL(rcStrict), rcStrict);
7716
7717 /* Update the interruptibility-state as it could have been changed by
7718 hmR0VmxInjectEventVmcs() (e.g. real-on-v86 guest injecting software interrupts) */
7719 fBlockMovSS = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS);
7720 fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
7721
7722 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
7723 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectInterrupt);
7724 else
7725 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectXcpt);
7726 }
7727
7728 /* Deliver pending debug exception if the guest is single-stepping. Evaluate and set the BS bit. */
7729 if ( fBlockSti
7730 || fBlockMovSS)
7731 {
7732 if (!pVCpu->hm.s.fSingleInstruction)
7733 {
7734 /*
7735 * The pending-debug exceptions field is cleared on all VM-exits except VMX_EXIT_TPR_BELOW_THRESHOLD,
7736 * VMX_EXIT_MTF, VMX_EXIT_APIC_WRITE and VMX_EXIT_VIRTUALIZED_EOI.
7737 * See Intel spec. 27.3.4 "Saving Non-Register State".
7738 */
7739 Assert(!DBGFIsStepping(pVCpu));
7740 int rc2 = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
7741 AssertRCReturn(rc2, rc2);
7742 if (pMixedCtx->eflags.Bits.u1TF)
7743 hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
7744 }
7745 else if (pMixedCtx->eflags.Bits.u1TF)
7746 {
7747 /*
7748 * We are single-stepping in the hypervisor debugger using EFLAGS.TF. Clear interrupt inhibition as setting the
7749 * BS bit would mean delivering a #DB to the guest upon VM-entry when it shouldn't be.
7750 */
7751 Assert(!(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1 & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG));
7752 uIntrState = 0;
7753 }
7754 }
7755
7756 /*
7757 * There's no need to clear the VM-entry interruption-information field here if we're not injecting anything.
7758 * VT-x clears the valid bit on every VM-exit. See Intel spec. 24.8.3 "VM-Entry Controls for Event Injection".
7759 */
7760 int rc2 = hmR0VmxLoadGuestIntrState(pVCpu, uIntrState);
7761 AssertRC(rc2);
7762
7763 Assert(rcStrict == VINF_SUCCESS || rcStrict == VINF_EM_RESET || (rcStrict == VINF_EM_DBG_STEPPED && fStepping));
7764 NOREF(fBlockMovSS); NOREF(fBlockSti);
7765 return rcStrict;
7766}
7767
7768
7769/**
7770 * Sets an invalid-opcode (\#UD) exception as pending-for-injection into the VM.
7771 *
7772 * @param pVCpu The cross context virtual CPU structure.
7773 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7774 * out-of-sync. Make sure to update the required fields
7775 * before using them.
7776 */
7777DECLINLINE(void) hmR0VmxSetPendingXcptUD(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7778{
7779 NOREF(pMixedCtx);
7780 uint32_t u32IntInfo = X86_XCPT_UD | VMX_EXIT_INTERRUPTION_INFO_VALID;
7781 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7782}
7783
7784
7785/**
7786 * Injects a double-fault (\#DF) exception into the VM.
7787 *
7788 * @returns Strict VBox status code (i.e. informational status codes too).
7789 * @param pVCpu The cross context virtual CPU structure.
7790 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7791 * out-of-sync. Make sure to update the required fields
7792 * before using them.
7793 * @param fStepping Whether we're running in hmR0VmxRunGuestCodeStep()
7794 * and should return VINF_EM_DBG_STEPPED if the event
7795 * is injected directly (register modified by us, not
7796 * by hardware on VM-entry).
7797 * @param puIntrState Pointer to the current guest interruptibility-state.
7798 * This interruptibility-state will be updated if
7799 * necessary. This cannot not be NULL.
7800 */
7801DECLINLINE(VBOXSTRICTRC) hmR0VmxInjectXcptDF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fStepping, uint32_t *puIntrState)
7802{
7803 uint32_t u32IntInfo = X86_XCPT_DF | VMX_EXIT_INTERRUPTION_INFO_VALID;
7804 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7805 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7806 return hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */,
7807 fStepping, puIntrState);
7808}
7809
7810
7811/**
7812 * Sets a debug (\#DB) exception as pending-for-injection into the VM.
7813 *
7814 * @param pVCpu The cross context virtual CPU structure.
7815 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7816 * out-of-sync. Make sure to update the required fields
7817 * before using them.
7818 */
7819DECLINLINE(void) hmR0VmxSetPendingXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx)
7820{
7821 NOREF(pMixedCtx);
7822 uint32_t u32IntInfo = X86_XCPT_DB | VMX_EXIT_INTERRUPTION_INFO_VALID;
7823 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7824 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7825}
7826
7827
7828/**
7829 * Sets an overflow (\#OF) exception as pending-for-injection into the VM.
7830 *
7831 * @param pVCpu The cross context virtual CPU structure.
7832 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7833 * out-of-sync. Make sure to update the required fields
7834 * before using them.
7835 * @param cbInstr The value of RIP that is to be pushed on the guest
7836 * stack.
7837 */
7838DECLINLINE(void) hmR0VmxSetPendingXcptOF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t cbInstr)
7839{
7840 NOREF(pMixedCtx);
7841 uint32_t u32IntInfo = X86_XCPT_OF | VMX_EXIT_INTERRUPTION_INFO_VALID;
7842 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7843 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7844}
7845
7846
7847/**
7848 * Injects a general-protection (\#GP) fault into the VM.
7849 *
7850 * @returns Strict VBox status code (i.e. informational status codes too).
7851 * @param pVCpu The cross context virtual CPU structure.
7852 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7853 * out-of-sync. Make sure to update the required fields
7854 * before using them.
7855 * @param fErrorCodeValid Whether the error code is valid (depends on the CPU
7856 * mode, i.e. in real-mode it's not valid).
7857 * @param u32ErrorCode The error code associated with the \#GP.
7858 * @param fStepping Whether we're running in
7859 * hmR0VmxRunGuestCodeStep() and should return
7860 * VINF_EM_DBG_STEPPED if the event is injected
7861 * directly (register modified by us, not by
7862 * hardware on VM-entry).
7863 * @param puIntrState Pointer to the current guest interruptibility-state.
7864 * This interruptibility-state will be updated if
7865 * necessary. This cannot not be NULL.
7866 */
7867DECLINLINE(VBOXSTRICTRC) hmR0VmxInjectXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, bool fErrorCodeValid, uint32_t u32ErrorCode,
7868 bool fStepping, uint32_t *puIntrState)
7869{
7870 uint32_t u32IntInfo = X86_XCPT_GP | VMX_EXIT_INTERRUPTION_INFO_VALID;
7871 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7872 if (fErrorCodeValid)
7873 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7874 return hmR0VmxInjectEventVmcs(pVCpu, pMixedCtx, u32IntInfo, 0 /* cbInstr */, u32ErrorCode, 0 /* GCPtrFaultAddress */,
7875 fStepping, puIntrState);
7876}
7877
7878
7879#if 0 /* unused */
7880/**
7881 * Sets a general-protection (\#GP) exception as pending-for-injection into the
7882 * VM.
7883 *
7884 * @param pVCpu The cross context virtual CPU structure.
7885 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7886 * out-of-sync. Make sure to update the required fields
7887 * before using them.
7888 * @param u32ErrorCode The error code associated with the \#GP.
7889 */
7890DECLINLINE(void) hmR0VmxSetPendingXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t u32ErrorCode)
7891{
7892 NOREF(pMixedCtx);
7893 uint32_t u32IntInfo = X86_XCPT_GP | VMX_EXIT_INTERRUPTION_INFO_VALID;
7894 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7895 u32IntInfo |= VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
7896 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, 0 /* cbInstr */, u32ErrorCode, 0 /* GCPtrFaultAddress */);
7897}
7898#endif /* unused */
7899
7900
7901/**
7902 * Sets a software interrupt (INTn) as pending-for-injection into the VM.
7903 *
7904 * @param pVCpu The cross context virtual CPU structure.
7905 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
7906 * out-of-sync. Make sure to update the required fields
7907 * before using them.
7908 * @param uVector The software interrupt vector number.
7909 * @param cbInstr The value of RIP that is to be pushed on the guest
7910 * stack.
7911 */
7912DECLINLINE(void) hmR0VmxSetPendingIntN(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint16_t uVector, uint32_t cbInstr)
7913{
7914 NOREF(pMixedCtx);
7915 uint32_t u32IntInfo = uVector | VMX_EXIT_INTERRUPTION_INFO_VALID;
7916 if ( uVector == X86_XCPT_BP
7917 || uVector == X86_XCPT_OF)
7918 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7919 else
7920 u32IntInfo |= (VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT << VMX_EXIT_INTERRUPTION_INFO_TYPE_SHIFT);
7921 hmR0VmxSetPendingEvent(pVCpu, u32IntInfo, cbInstr, 0 /* u32ErrCode */, 0 /* GCPtrFaultAddress */);
7922}
7923
7924
7925/**
7926 * Pushes a 2-byte value onto the real-mode (in virtual-8086 mode) guest's
7927 * stack.
7928 *
7929 * @returns Strict VBox status code (i.e. informational status codes too).
7930 * @retval VINF_EM_RESET if pushing a value to the stack caused a triple-fault.
7931 * @param pVM The cross context VM structure.
7932 * @param pMixedCtx Pointer to the guest-CPU context.
7933 * @param uValue The value to push to the guest stack.
7934 */
7935DECLINLINE(VBOXSTRICTRC) hmR0VmxRealModeGuestStackPush(PVM pVM, PCPUMCTX pMixedCtx, uint16_t uValue)
7936{
7937 /*
7938 * The stack limit is 0xffff in real-on-virtual 8086 mode. Real-mode with weird stack limits cannot be run in
7939 * virtual 8086 mode in VT-x. See Intel spec. 26.3.1.2 "Checks on Guest Segment Registers".
7940 * See Intel Instruction reference for PUSH and Intel spec. 22.33.1 "Segment Wraparound".
7941 */
7942 if (pMixedCtx->sp == 1)
7943 return VINF_EM_RESET;
7944 pMixedCtx->sp -= sizeof(uint16_t); /* May wrap around which is expected behaviour. */
7945 int rc = PGMPhysSimpleWriteGCPhys(pVM, pMixedCtx->ss.u64Base + pMixedCtx->sp, &uValue, sizeof(uint16_t));
7946 AssertRC(rc);
7947 return rc;
7948}
7949
7950
7951/**
7952 * Injects an event into the guest upon VM-entry by updating the relevant fields
7953 * in the VM-entry area in the VMCS.
7954 *
7955 * @returns Strict VBox status code (i.e. informational status codes too).
7956 * @retval VINF_SUCCESS if the event is successfully injected into the VMCS.
7957 * @retval VINF_EM_RESET if event injection resulted in a triple-fault.
7958 *
7959 * @param pVCpu The cross context virtual CPU structure.
7960 * @param pMixedCtx Pointer to the guest-CPU context. The data may
7961 * be out-of-sync. Make sure to update the required
7962 * fields before using them.
7963 * @param u64IntInfo The VM-entry interruption-information field.
7964 * @param cbInstr The VM-entry instruction length in bytes (for
7965 * software interrupts, exceptions and privileged
7966 * software exceptions).
7967 * @param u32ErrCode The VM-entry exception error code.
7968 * @param GCPtrFaultAddress The page-fault address for \#PF exceptions.
7969 * @param puIntrState Pointer to the current guest interruptibility-state.
7970 * This interruptibility-state will be updated if
7971 * necessary. This cannot not be NULL.
7972 * @param fStepping Whether we're running in
7973 * hmR0VmxRunGuestCodeStep() and should return
7974 * VINF_EM_DBG_STEPPED if the event is injected
7975 * directly (register modified by us, not by
7976 * hardware on VM-entry).
7977 *
7978 * @remarks Requires CR0!
7979 * @remarks No-long-jump zone!!!
7980 */
7981static VBOXSTRICTRC hmR0VmxInjectEventVmcs(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint64_t u64IntInfo, uint32_t cbInstr,
7982 uint32_t u32ErrCode, RTGCUINTREG GCPtrFaultAddress, bool fStepping,
7983 uint32_t *puIntrState)
7984{
7985 /* Intel spec. 24.8.3 "VM-Entry Controls for Event Injection" specifies the interruption-information field to be 32-bits. */
7986 AssertMsg(u64IntInfo >> 32 == 0, ("%#RX64\n", u64IntInfo));
7987 Assert(puIntrState);
7988 uint32_t u32IntInfo = (uint32_t)u64IntInfo;
7989
7990 uint32_t const uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(u32IntInfo);
7991 uint32_t const uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(u32IntInfo);
7992
7993#ifdef VBOX_STRICT
7994 /* Validate the error-code-valid bit for hardware exceptions. */
7995 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT)
7996 {
7997 switch (uVector)
7998 {
7999 case X86_XCPT_PF:
8000 case X86_XCPT_DF:
8001 case X86_XCPT_TS:
8002 case X86_XCPT_NP:
8003 case X86_XCPT_SS:
8004 case X86_XCPT_GP:
8005 case X86_XCPT_AC:
8006 AssertMsg(VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(u32IntInfo),
8007 ("Error-code-valid bit not set for exception that has an error code uVector=%#x\n", uVector));
8008 /* fallthru */
8009 default:
8010 break;
8011 }
8012 }
8013#endif
8014
8015 /* Cannot inject an NMI when block-by-MOV SS is in effect. */
8016 Assert( uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI
8017 || !(*puIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS));
8018
8019 STAM_COUNTER_INC(&pVCpu->hm.s.paStatInjectedIrqsR0[uVector & MASK_INJECT_IRQ_STAT]);
8020
8021 /* We require CR0 to check if the guest is in real-mode. */
8022 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
8023 AssertRCReturn(rc, rc);
8024
8025 /*
8026 * Hardware interrupts & exceptions cannot be delivered through the software interrupt redirection bitmap to the real
8027 * mode task in virtual-8086 mode. We must jump to the interrupt handler in the (real-mode) guest.
8028 * See Intel spec. 20.3 "Interrupt and Exception handling in Virtual-8086 Mode" for interrupt & exception classes.
8029 * See Intel spec. 20.1.4 "Interrupt and Exception Handling" for real-mode interrupt handling.
8030 */
8031 if (CPUMIsGuestInRealModeEx(pMixedCtx))
8032 {
8033 PVM pVM = pVCpu->CTX_SUFF(pVM);
8034 if (!pVM->hm.s.vmx.fUnrestrictedGuest)
8035 {
8036 Assert(PDMVmmDevHeapIsEnabled(pVM));
8037 Assert(pVM->hm.s.vmx.pRealModeTSS);
8038
8039 /* We require RIP, RSP, RFLAGS, CS, IDTR. Save the required ones from the VMCS. */
8040 rc = hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
8041 rc |= hmR0VmxSaveGuestTableRegs(pVCpu, pMixedCtx);
8042 rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
8043 AssertRCReturn(rc, rc);
8044 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_RIP));
8045
8046 /* Check if the interrupt handler is present in the IVT (real-mode IDT). IDT limit is (4N - 1). */
8047 size_t const cbIdtEntry = sizeof(X86IDTR16);
8048 if (uVector * cbIdtEntry + (cbIdtEntry - 1) > pMixedCtx->idtr.cbIdt)
8049 {
8050 /* If we are trying to inject a #DF with no valid IDT entry, return a triple-fault. */
8051 if (uVector == X86_XCPT_DF)
8052 return VINF_EM_RESET;
8053
8054 /* If we're injecting a #GP with no valid IDT entry, inject a double-fault. */
8055 if (uVector == X86_XCPT_GP)
8056 return hmR0VmxInjectXcptDF(pVCpu, pMixedCtx, fStepping, puIntrState);
8057
8058 /* If we're injecting an interrupt/exception with no valid IDT entry, inject a general-protection fault. */
8059 /* No error codes for exceptions in real-mode. See Intel spec. 20.1.4 "Interrupt and Exception Handling" */
8060 return hmR0VmxInjectXcptGP(pVCpu, pMixedCtx, false /* fErrCodeValid */, 0 /* u32ErrCode */,
8061 fStepping, puIntrState);
8062 }
8063
8064 /* Software exceptions (#BP and #OF exceptions thrown as a result of INT3 or INTO) */
8065 uint16_t uGuestIp = pMixedCtx->ip;
8066 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT)
8067 {
8068 Assert(uVector == X86_XCPT_BP || uVector == X86_XCPT_OF);
8069 /* #BP and #OF are both benign traps, we need to resume the next instruction. */
8070 uGuestIp = pMixedCtx->ip + (uint16_t)cbInstr;
8071 }
8072 else if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT)
8073 uGuestIp = pMixedCtx->ip + (uint16_t)cbInstr;
8074
8075 /* Get the code segment selector and offset from the IDT entry for the interrupt handler. */
8076 X86IDTR16 IdtEntry;
8077 RTGCPHYS GCPhysIdtEntry = (RTGCPHYS)pMixedCtx->idtr.pIdt + uVector * cbIdtEntry;
8078 rc = PGMPhysSimpleReadGCPhys(pVM, &IdtEntry, GCPhysIdtEntry, cbIdtEntry);
8079 AssertRCReturn(rc, rc);
8080
8081 /* Construct the stack frame for the interrupt/exception handler. */
8082 VBOXSTRICTRC rcStrict;
8083 rcStrict = hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, pMixedCtx->eflags.u32);
8084 if (rcStrict == VINF_SUCCESS)
8085 rcStrict = hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, pMixedCtx->cs.Sel);
8086 if (rcStrict == VINF_SUCCESS)
8087 rcStrict = hmR0VmxRealModeGuestStackPush(pVM, pMixedCtx, uGuestIp);
8088
8089 /* Clear the required eflag bits and jump to the interrupt/exception handler. */
8090 if (rcStrict == VINF_SUCCESS)
8091 {
8092 pMixedCtx->eflags.u32 &= ~(X86_EFL_IF | X86_EFL_TF | X86_EFL_RF | X86_EFL_AC);
8093 pMixedCtx->rip = IdtEntry.offSel;
8094 pMixedCtx->cs.Sel = IdtEntry.uSel;
8095 pMixedCtx->cs.ValidSel = IdtEntry.uSel;
8096 pMixedCtx->cs.u64Base = IdtEntry.uSel << cbIdtEntry;
8097 if ( uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT
8098 && uVector == X86_XCPT_PF)
8099 pMixedCtx->cr2 = GCPtrFaultAddress;
8100
8101 /* If any other guest-state bits are changed here, make sure to update
8102 hmR0VmxPreRunGuestCommitted() when thread-context hooks are used. */
8103 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS
8104 | HM_CHANGED_GUEST_RIP
8105 | HM_CHANGED_GUEST_RFLAGS
8106 | HM_CHANGED_GUEST_RSP);
8107
8108 /* We're clearing interrupts, which means no block-by-STI interrupt-inhibition. */
8109 if (*puIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
8110 {
8111 Assert( uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI
8112 && uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT);
8113 Log4(("Clearing inhibition due to STI.\n"));
8114 *puIntrState &= ~VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI;
8115 }
8116 Log4(("Injecting real-mode: u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x Eflags=%#x CS:EIP=%04x:%04x\n",
8117 u32IntInfo, u32ErrCode, cbInstr, pMixedCtx->eflags.u, pMixedCtx->cs.Sel, pMixedCtx->eip));
8118
8119 /* The event has been truly dispatched. Mark it as no longer pending so we don't attempt to 'undo'
8120 it, if we are returning to ring-3 before executing guest code. */
8121 pVCpu->hm.s.Event.fPending = false;
8122
8123 /* Make hmR0VmxPreRunGuest return if we're stepping since we've changed cs:rip. */
8124 if (fStepping)
8125 rcStrict = VINF_EM_DBG_STEPPED;
8126 }
8127 AssertMsg(rcStrict == VINF_SUCCESS || rcStrict == VINF_EM_RESET || (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
8128 ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8129 return rcStrict;
8130 }
8131
8132 /*
8133 * For unrestricted execution enabled CPUs running real-mode guests, we must not set the deliver-error-code bit.
8134 * See Intel spec. 26.2.1.3 "VM-Entry Control Fields".
8135 */
8136 u32IntInfo &= ~VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_VALID;
8137 }
8138
8139 /* Validate. */
8140 Assert(VMX_EXIT_INTERRUPTION_INFO_IS_VALID(u32IntInfo)); /* Bit 31 (Valid bit) must be set by caller. */
8141 Assert(!VMX_EXIT_INTERRUPTION_INFO_NMI_UNBLOCK_IRET(u32IntInfo)); /* Bit 12 MBZ. */
8142 Assert(!(u32IntInfo & 0x7ffff000)); /* Bits 30:12 MBZ. */
8143
8144 /* Inject. */
8145 rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, u32IntInfo);
8146 if (VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(u32IntInfo))
8147 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE, u32ErrCode);
8148 rc |= VMXWriteVmcs32(VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH, cbInstr);
8149
8150 if ( VMX_EXIT_INTERRUPTION_INFO_TYPE(u32IntInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT
8151 && uVector == X86_XCPT_PF)
8152 pMixedCtx->cr2 = GCPtrFaultAddress;
8153
8154 Log4(("Injecting vcpu[%RU32] u32IntInfo=%#x u32ErrCode=%#x cbInstr=%#x pMixedCtx->uCR2=%#RX64\n", pVCpu->idCpu,
8155 u32IntInfo, u32ErrCode, cbInstr, pMixedCtx->cr2));
8156
8157 AssertRCReturn(rc, rc);
8158 return VINF_SUCCESS;
8159}
8160
8161
8162/**
8163 * Clears the interrupt-window exiting control in the VMCS and if necessary
8164 * clears the current event in the VMCS as well.
8165 *
8166 * @returns VBox status code.
8167 * @param pVCpu The cross context virtual CPU structure.
8168 *
8169 * @remarks Use this function only to clear events that have not yet been
8170 * delivered to the guest but are injected in the VMCS!
8171 * @remarks No-long-jump zone!!!
8172 */
8173static void hmR0VmxClearIntNmiWindowsVmcs(PVMCPU pVCpu)
8174{
8175 Log4Func(("vcpu[%d]\n", pVCpu->idCpu));
8176
8177 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_INT_WINDOW_EXIT)
8178 hmR0VmxClearIntWindowExitVmcs(pVCpu);
8179
8180 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT)
8181 hmR0VmxClearNmiWindowExitVmcs(pVCpu);
8182}
8183
8184
8185/**
8186 * Enters the VT-x session.
8187 *
8188 * @returns VBox status code.
8189 * @param pVM The cross context VM structure.
8190 * @param pVCpu The cross context virtual CPU structure.
8191 * @param pCpu Pointer to the CPU info struct.
8192 */
8193VMMR0DECL(int) VMXR0Enter(PVM pVM, PVMCPU pVCpu, PHMGLOBALCPUINFO pCpu)
8194{
8195 AssertPtr(pVM);
8196 AssertPtr(pVCpu);
8197 Assert(pVM->hm.s.vmx.fSupported);
8198 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8199 NOREF(pCpu); NOREF(pVM);
8200
8201 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8202 Assert(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE));
8203
8204#ifdef VBOX_STRICT
8205 /* At least verify VMX is enabled, since we can't check if we're in VMX root mode without #GP'ing. */
8206 RTCCUINTREG uHostCR4 = ASMGetCR4();
8207 if (!(uHostCR4 & X86_CR4_VMXE))
8208 {
8209 LogRel(("VMXR0Enter: X86_CR4_VMXE bit in CR4 is not set!\n"));
8210 return VERR_VMX_X86_CR4_VMXE_CLEARED;
8211 }
8212#endif
8213
8214 /*
8215 * Load the VCPU's VMCS as the current (and active) one.
8216 */
8217 Assert(pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_CLEAR);
8218 int rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
8219 if (RT_FAILURE(rc))
8220 return rc;
8221
8222 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE;
8223 pVCpu->hm.s.fLeaveDone = false;
8224 Log4Func(("Activated Vmcs. HostCpuId=%u\n", RTMpCpuId()));
8225
8226 return VINF_SUCCESS;
8227}
8228
8229
8230/**
8231 * The thread-context callback (only on platforms which support it).
8232 *
8233 * @param enmEvent The thread-context event.
8234 * @param pVCpu The cross context virtual CPU structure.
8235 * @param fGlobalInit Whether global VT-x/AMD-V init. was used.
8236 * @thread EMT(pVCpu)
8237 */
8238VMMR0DECL(void) VMXR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, PVMCPU pVCpu, bool fGlobalInit)
8239{
8240 NOREF(fGlobalInit);
8241
8242 switch (enmEvent)
8243 {
8244 case RTTHREADCTXEVENT_OUT:
8245 {
8246 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8247 Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
8248 VMCPU_ASSERT_EMT(pVCpu);
8249
8250 PCPUMCTX pMixedCtx = CPUMQueryGuestCtxPtr(pVCpu);
8251
8252 /* No longjmps (logger flushes, locks) in this fragile context. */
8253 VMMRZCallRing3Disable(pVCpu);
8254 Log4Func(("Preempting: HostCpuId=%u\n", RTMpCpuId()));
8255
8256 /*
8257 * Restore host-state (FPU, debug etc.)
8258 */
8259 if (!pVCpu->hm.s.fLeaveDone)
8260 {
8261 /* Do -not- save guest-state here as we might already be in the middle of saving it (esp. bad if we are
8262 holding the PGM lock while saving the guest state (see hmR0VmxSaveGuestControlRegs()). */
8263 hmR0VmxLeave(pVCpu, pMixedCtx, false /* fSaveGuestState */);
8264 pVCpu->hm.s.fLeaveDone = true;
8265 }
8266
8267 /* Leave HM context, takes care of local init (term). */
8268 int rc = HMR0LeaveCpu(pVCpu);
8269 AssertRC(rc); NOREF(rc);
8270
8271 /* Restore longjmp state. */
8272 VMMRZCallRing3Enable(pVCpu);
8273 STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreempt);
8274 break;
8275 }
8276
8277 case RTTHREADCTXEVENT_IN:
8278 {
8279 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8280 Assert(VMMR0ThreadCtxHookIsEnabled(pVCpu));
8281 VMCPU_ASSERT_EMT(pVCpu);
8282
8283 /* No longjmps here, as we don't want to trigger preemption (& its hook) while resuming. */
8284 VMMRZCallRing3Disable(pVCpu);
8285 Log4Func(("Resumed: HostCpuId=%u\n", RTMpCpuId()));
8286
8287 /* Initialize the bare minimum state required for HM. This takes care of
8288 initializing VT-x if necessary (onlined CPUs, local init etc.) */
8289 int rc = HMR0EnterCpu(pVCpu);
8290 AssertRC(rc);
8291 Assert(HMCPU_CF_IS_SET(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE));
8292
8293 /* Load the active VMCS as the current one. */
8294 if (pVCpu->hm.s.vmx.uVmcsState & HMVMX_VMCS_STATE_CLEAR)
8295 {
8296 rc = VMXActivateVmcs(pVCpu->hm.s.vmx.HCPhysVmcs);
8297 AssertRC(rc); NOREF(rc);
8298 pVCpu->hm.s.vmx.uVmcsState = HMVMX_VMCS_STATE_ACTIVE;
8299 Log4Func(("Resumed: Activated Vmcs. HostCpuId=%u\n", RTMpCpuId()));
8300 }
8301 pVCpu->hm.s.fLeaveDone = false;
8302
8303 /* Restore longjmp state. */
8304 VMMRZCallRing3Enable(pVCpu);
8305 break;
8306 }
8307
8308 default:
8309 break;
8310 }
8311}
8312
8313
8314/**
8315 * Saves the host state in the VMCS host-state.
8316 * Sets up the VM-exit MSR-load area.
8317 *
8318 * The CPU state will be loaded from these fields on every successful VM-exit.
8319 *
8320 * @returns VBox status code.
8321 * @param pVM The cross context VM structure.
8322 * @param pVCpu The cross context virtual CPU structure.
8323 *
8324 * @remarks No-long-jump zone!!!
8325 */
8326static int hmR0VmxSaveHostState(PVM pVM, PVMCPU pVCpu)
8327{
8328 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8329
8330 int rc = VINF_SUCCESS;
8331 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT))
8332 {
8333 rc = hmR0VmxSaveHostControlRegs(pVM, pVCpu);
8334 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostControlRegisters failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8335
8336 rc = hmR0VmxSaveHostSegmentRegs(pVM, pVCpu);
8337 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostSegmentRegisters failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8338
8339 rc = hmR0VmxSaveHostMsrs(pVM, pVCpu);
8340 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSaveHostMsrs failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8341
8342 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_HOST_CONTEXT);
8343 }
8344 return rc;
8345}
8346
8347
8348/**
8349 * Saves the host state in the VMCS host-state.
8350 *
8351 * @returns VBox status code.
8352 * @param pVM The cross context VM structure.
8353 * @param pVCpu The cross context virtual CPU structure.
8354 *
8355 * @remarks No-long-jump zone!!!
8356 */
8357VMMR0DECL(int) VMXR0SaveHostState(PVM pVM, PVMCPU pVCpu)
8358{
8359 AssertPtr(pVM);
8360 AssertPtr(pVCpu);
8361
8362 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8363
8364 /* Save the host state here while entering HM context. When thread-context hooks are used, we might get preempted
8365 and have to resave the host state but most of the time we won't be, so do it here before we disable interrupts. */
8366 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8367 return hmR0VmxSaveHostState(pVM, pVCpu);
8368}
8369
8370
8371/**
8372 * Loads the guest state into the VMCS guest-state area.
8373 *
8374 * The will typically be done before VM-entry when the guest-CPU state and the
8375 * VMCS state may potentially be out of sync.
8376 *
8377 * Sets up the VM-entry MSR-load and VM-exit MSR-store areas. Sets up the
8378 * VM-entry controls.
8379 * Sets up the appropriate VMX non-root function to execute guest code based on
8380 * the guest CPU mode.
8381 *
8382 * @returns VBox strict status code.
8383 * @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
8384 * without unrestricted guest access and the VMMDev is not presently
8385 * mapped (e.g. EFI32).
8386 *
8387 * @param pVM The cross context VM structure.
8388 * @param pVCpu The cross context virtual CPU structure.
8389 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8390 * out-of-sync. Make sure to update the required fields
8391 * before using them.
8392 *
8393 * @remarks No-long-jump zone!!! (Disables and enables long jmps for itself,
8394 * caller disables then again on successfull return. Confusing.)
8395 */
8396static VBOXSTRICTRC hmR0VmxLoadGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
8397{
8398 AssertPtr(pVM);
8399 AssertPtr(pVCpu);
8400 AssertPtr(pMixedCtx);
8401 HMVMX_ASSERT_PREEMPT_SAFE();
8402
8403 VMMRZCallRing3Disable(pVCpu);
8404 Assert(VMMR0IsLogFlushDisabled(pVCpu));
8405
8406 LogFlowFunc(("pVM=%p pVCpu=%p\n", pVM, pVCpu));
8407
8408 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatLoadGuestState, x);
8409
8410 /* Determine real-on-v86 mode. */
8411 pVCpu->hm.s.vmx.RealMode.fRealOnV86Active = false;
8412 if ( !pVM->hm.s.vmx.fUnrestrictedGuest
8413 && CPUMIsGuestInRealModeEx(pMixedCtx))
8414 {
8415 pVCpu->hm.s.vmx.RealMode.fRealOnV86Active = true;
8416 }
8417
8418 /*
8419 * Load the guest-state into the VMCS.
8420 * Any ordering dependency among the sub-functions below must be explicitly stated using comments.
8421 * Ideally, assert that the cross-dependent bits are up-to-date at the point of using it.
8422 */
8423 int rc = hmR0VmxSetupVMRunHandler(pVCpu, pMixedCtx);
8424 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSetupVMRunHandler! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8425
8426 /* This needs to be done after hmR0VmxSetupVMRunHandler() as changing pfnStartVM may require VM-entry control updates. */
8427 rc = hmR0VmxLoadGuestEntryCtls(pVCpu, pMixedCtx);
8428 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestEntryCtls! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8429
8430 /* This needs to be done after hmR0VmxSetupVMRunHandler() as changing pfnStartVM may require VM-exit control updates. */
8431 rc = hmR0VmxLoadGuestExitCtls(pVCpu, pMixedCtx);
8432 AssertLogRelMsgRCReturn(rc, ("hmR0VmxSetupExitCtls failed! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8433
8434 rc = hmR0VmxLoadGuestActivityState(pVCpu, pMixedCtx);
8435 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestActivityState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8436
8437 VBOXSTRICTRC rcStrict = hmR0VmxLoadGuestCR3AndCR4(pVCpu, pMixedCtx);
8438 if (rcStrict == VINF_SUCCESS)
8439 { /* likely */ }
8440 else
8441 {
8442 VMMRZCallRing3Enable(pVCpu);
8443 Assert(rcStrict == VINF_EM_RESCHEDULE_REM || RT_FAILURE_NP(rcStrict));
8444 return rcStrict;
8445 }
8446
8447 /* Assumes pMixedCtx->cr0 is up-to-date (strict builds require CR0 for segment register validation checks). */
8448 rc = hmR0VmxLoadGuestSegmentRegs(pVCpu, pMixedCtx);
8449 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestSegmentRegs: rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8450
8451 /* This needs to be done after hmR0VmxLoadGuestEntryCtls() and hmR0VmxLoadGuestExitCtls() as it may alter controls if we
8452 determine we don't have to swap EFER after all. */
8453 rc = hmR0VmxLoadGuestMsrs(pVCpu, pMixedCtx);
8454 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadSharedMsrs! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8455
8456 rc = hmR0VmxLoadGuestApicState(pVCpu, pMixedCtx);
8457 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestApicState! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8458
8459 rc = hmR0VmxLoadGuestXcptIntercepts(pVCpu, pMixedCtx);
8460 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestXcptIntercepts! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8461
8462 /*
8463 * Loading Rflags here is fine, even though Rflags.TF might depend on guest debug state (which is not loaded here).
8464 * It is re-evaluated and updated if necessary in hmR0VmxLoadSharedState().
8465 */
8466 rc = hmR0VmxLoadGuestRipRspRflags(pVCpu, pMixedCtx);
8467 AssertLogRelMsgRCReturn(rc, ("hmR0VmxLoadGuestRipRspRflags! rc=%Rrc (pVM=%p pVCpu=%p)\n", rc, pVM, pVCpu), rc);
8468
8469 /* Clear any unused and reserved bits. */
8470 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_CR2);
8471
8472 VMMRZCallRing3Enable(pVCpu);
8473
8474 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatLoadGuestState, x);
8475 return rc;
8476}
8477
8478
8479/**
8480 * Loads the state shared between the host and guest into the VMCS.
8481 *
8482 * @param pVM The cross context VM structure.
8483 * @param pVCpu The cross context virtual CPU structure.
8484 * @param pCtx Pointer to the guest-CPU context.
8485 *
8486 * @remarks No-long-jump zone!!!
8487 */
8488static void hmR0VmxLoadSharedState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
8489{
8490 NOREF(pVM);
8491
8492 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8493 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8494
8495 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0))
8496 {
8497 int rc = hmR0VmxLoadSharedCR0(pVCpu, pCtx);
8498 AssertRC(rc);
8499 }
8500
8501 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_DEBUG))
8502 {
8503 int rc = hmR0VmxLoadSharedDebugState(pVCpu, pCtx);
8504 AssertRC(rc);
8505
8506 /* Loading shared debug bits might have changed eflags.TF bit for debugging purposes. */
8507 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_RFLAGS))
8508 {
8509 rc = hmR0VmxLoadGuestRflags(pVCpu, pCtx);
8510 AssertRC(rc);
8511 }
8512 }
8513
8514 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS))
8515 {
8516 hmR0VmxLazyLoadGuestMsrs(pVCpu, pCtx);
8517 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS);
8518 }
8519
8520 /* Loading CR0, debug state might have changed intercepts, update VMCS. */
8521 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS))
8522 {
8523 Assert(pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT_32(X86_XCPT_AC));
8524 Assert(pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT_32(X86_XCPT_DB));
8525 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
8526 AssertRC(rc);
8527 HMCPU_CF_CLEAR(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS);
8528 }
8529
8530 AssertMsg(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE),
8531 ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8532}
8533
8534
8535/**
8536 * Worker for loading the guest-state bits in the inner VT-x execution loop.
8537 *
8538 * @returns Strict VBox status code (i.e. informational status codes too).
8539 * @retval VINF_EM_RESCHEDULE_REM if we try to emulate non-paged guest code
8540 * without unrestricted guest access and the VMMDev is not presently
8541 * mapped (e.g. EFI32).
8542 *
8543 * @param pVM The cross context VM structure.
8544 * @param pVCpu The cross context virtual CPU structure.
8545 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8546 * out-of-sync. Make sure to update the required fields
8547 * before using them.
8548 */
8549static VBOXSTRICTRC hmR0VmxLoadGuestStateOptimal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx)
8550{
8551 HMVMX_ASSERT_PREEMPT_SAFE();
8552
8553 Log5(("LoadFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8554#ifdef HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE
8555 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
8556#endif
8557
8558 VBOXSTRICTRC rcStrict = VINF_SUCCESS;
8559 if (HMCPU_CF_IS_SET_ONLY(pVCpu, HM_CHANGED_GUEST_RIP))
8560 {
8561 rcStrict = hmR0VmxLoadGuestRip(pVCpu, pMixedCtx);
8562 if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8563 { /* likely */}
8564 else
8565 {
8566 AssertMsgFailedReturn(("hmR0VmxLoadGuestStateOptimal: hmR0VmxLoadGuestRip failed! rc=%Rrc\n",
8567 VBOXSTRICTRC_VAL(rcStrict)), rcStrict);
8568 }
8569 STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadMinimal);
8570 }
8571 else if (HMCPU_CF_VALUE(pVCpu))
8572 {
8573 rcStrict = hmR0VmxLoadGuestState(pVM, pVCpu, pMixedCtx);
8574 if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8575 { /* likely */}
8576 else
8577 {
8578 AssertMsg(rcStrict == VINF_EM_RESCHEDULE_REM,
8579 ("hmR0VmxLoadGuestStateOptimal: hmR0VmxLoadGuestState failed! rc=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8580 return rcStrict;
8581 }
8582 STAM_COUNTER_INC(&pVCpu->hm.s.StatLoadFull);
8583 }
8584
8585 /* All the guest state bits should be loaded except maybe the host context and/or the shared host/guest bits. */
8586 AssertMsg( !HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_ALL_GUEST)
8587 || HMCPU_CF_IS_PENDING_ONLY(pVCpu, HM_CHANGED_HOST_CONTEXT | HM_CHANGED_HOST_GUEST_SHARED_STATE),
8588 ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8589 return rcStrict;
8590}
8591
8592
8593/**
8594 * Does the preparations before executing guest code in VT-x.
8595 *
8596 * This may cause longjmps to ring-3 and may even result in rescheduling to the
8597 * recompiler/IEM. We must be cautious what we do here regarding committing
8598 * guest-state information into the VMCS assuming we assuredly execute the
8599 * guest in VT-x mode.
8600 *
8601 * If we fall back to the recompiler/IEM after updating the VMCS and clearing
8602 * the common-state (TRPM/forceflags), we must undo those changes so that the
8603 * recompiler/IEM can (and should) use them when it resumes guest execution.
8604 * Otherwise such operations must be done when we can no longer exit to ring-3.
8605 *
8606 * @returns Strict VBox status code (i.e. informational status codes too).
8607 * @retval VINF_SUCCESS if we can proceed with running the guest, interrupts
8608 * have been disabled.
8609 * @retval VINF_EM_RESET if a triple-fault occurs while injecting a
8610 * double-fault into the guest.
8611 * @retval VINF_EM_DBG_STEPPED if @a fStepping is true and an event was
8612 * dispatched directly.
8613 * @retval VINF_* scheduling changes, we have to go back to ring-3.
8614 *
8615 * @param pVM The cross context VM structure.
8616 * @param pVCpu The cross context virtual CPU structure.
8617 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8618 * out-of-sync. Make sure to update the required fields
8619 * before using them.
8620 * @param pVmxTransient Pointer to the VMX transient structure.
8621 * @param fStepping Set if called from hmR0VmxRunGuestCodeStep(). Makes
8622 * us ignore some of the reasons for returning to
8623 * ring-3, and return VINF_EM_DBG_STEPPED if event
8624 * dispatching took place.
8625 */
8626static VBOXSTRICTRC hmR0VmxPreRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, bool fStepping)
8627{
8628 Assert(VMMRZCallRing3IsEnabled(pVCpu));
8629
8630#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
8631 PGMRZDynMapFlushAutoSet(pVCpu);
8632#endif
8633
8634 /* Check force flag actions that might require us to go back to ring-3. */
8635 VBOXSTRICTRC rcStrict = hmR0VmxCheckForceFlags(pVM, pVCpu, pMixedCtx, fStepping);
8636 if (rcStrict == VINF_SUCCESS)
8637 { /* FFs doesn't get set all the time. */ }
8638 else
8639 return rcStrict;
8640
8641 if (TRPMHasTrap(pVCpu))
8642 hmR0VmxTrpmTrapToPendingEvent(pVCpu);
8643 uint32_t uIntrState = hmR0VmxEvaluatePendingEvent(pVCpu, pMixedCtx);
8644
8645 /*
8646 * Event injection may take locks (currently the PGM lock for real-on-v86 case) and thus needs to be done with
8647 * longjmps or interrupts + preemption enabled. Event injection might also result in triple-faulting the VM.
8648 */
8649 rcStrict = hmR0VmxInjectPendingEvent(pVCpu, pMixedCtx, uIntrState, fStepping);
8650 if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8651 { /* likely */ }
8652 else
8653 {
8654 AssertMsg(rcStrict == VINF_EM_RESET || (rcStrict == VINF_EM_DBG_STEPPED && fStepping),
8655 ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
8656 return rcStrict;
8657 }
8658
8659 /*
8660 * Load the guest state bits, we can handle longjmps/getting preempted here.
8661 *
8662 * If we are injecting events to a real-on-v86 mode guest, we will have to update
8663 * RIP and some segment registers, i.e. hmR0VmxInjectPendingEvent()->hmR0VmxInjectEventVmcs().
8664 * Hence, this needs to be done -after- injection of events.
8665 */
8666 rcStrict = hmR0VmxLoadGuestStateOptimal(pVM, pVCpu, pMixedCtx);
8667 if (RT_LIKELY(rcStrict == VINF_SUCCESS))
8668 { /* likely */ }
8669 else
8670 return rcStrict;
8671
8672 /*
8673 * No longjmps to ring-3 from this point on!!!
8674 * Asserts() will still longjmp to ring-3 (but won't return), which is intentional, better than a kernel panic.
8675 * This also disables flushing of the R0-logger instance (if any).
8676 */
8677 VMMRZCallRing3Disable(pVCpu);
8678
8679 /*
8680 * We disable interrupts so that we don't miss any interrupts that would flag preemption (IPI/timers etc.)
8681 * when thread-context hooks aren't used and we've been running with preemption disabled for a while.
8682 *
8683 * We need to check for force-flags that could've possible been altered since we last checked them (e.g.
8684 * by PDMGetInterrupt() leaving the PDM critical section, see @bugref{6398}).
8685 *
8686 * We also check a couple of other force-flags as a last opportunity to get the EMT back to ring-3 before
8687 * executing guest code.
8688 */
8689 pVmxTransient->fEFlags = ASMIntDisableFlags();
8690
8691 if ( ( !VM_FF_IS_PENDING(pVM, VM_FF_EMT_RENDEZVOUS | VM_FF_TM_VIRTUAL_SYNC)
8692 && !VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK))
8693 || ( fStepping /* Optimized for the non-stepping case, so a bit of unnecessary work when stepping. */
8694 && !VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_HM_TO_R3_MASK & ~(VMCPU_FF_TIMER | VMCPU_FF_PDM_CRITSECT))) )
8695 {
8696 if (!RTThreadPreemptIsPending(NIL_RTTHREAD))
8697 {
8698 /* We've injected any pending events. This is really the point of no return (to ring-3). */
8699 pVCpu->hm.s.Event.fPending = false;
8700
8701 return VINF_SUCCESS;
8702 }
8703
8704 STAM_COUNTER_INC(&pVCpu->hm.s.StatPendingHostIrq);
8705 rcStrict = VINF_EM_RAW_INTERRUPT;
8706 }
8707 else
8708 {
8709 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHmToR3FF);
8710 rcStrict = VINF_EM_RAW_TO_R3;
8711 }
8712
8713 ASMSetFlags(pVmxTransient->fEFlags);
8714 VMMRZCallRing3Enable(pVCpu);
8715
8716 return rcStrict;
8717}
8718
8719
8720/**
8721 * Prepares to run guest code in VT-x and we've committed to doing so. This
8722 * means there is no backing out to ring-3 or anywhere else at this
8723 * point.
8724 *
8725 * @param pVM The cross context VM structure.
8726 * @param pVCpu The cross context virtual CPU structure.
8727 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
8728 * out-of-sync. Make sure to update the required fields
8729 * before using them.
8730 * @param pVmxTransient Pointer to the VMX transient structure.
8731 *
8732 * @remarks Called with preemption disabled.
8733 * @remarks No-long-jump zone!!!
8734 */
8735static void hmR0VmxPreRunGuestCommitted(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
8736{
8737 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8738 Assert(VMMR0IsLogFlushDisabled(pVCpu));
8739 Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
8740
8741 /*
8742 * Indicate start of guest execution and where poking EMT out of guest-context is recognized.
8743 */
8744 VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
8745 VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_EXEC);
8746
8747#ifdef HMVMX_ALWAYS_SWAP_FPU_STATE
8748 if (!CPUMIsGuestFPUStateActive(pVCpu))
8749 if (CPUMR0LoadGuestFPU(pVM, pVCpu) == VINF_CPUM_HOST_CR0_MODIFIED)
8750 HMCPU_CF_SET(pVCpu, HM_CHANGED_HOST_CONTEXT);
8751 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8752#endif
8753
8754 if ( pVCpu->hm.s.fPreloadGuestFpu
8755 && !CPUMIsGuestFPUStateActive(pVCpu))
8756 {
8757 if (CPUMR0LoadGuestFPU(pVM, pVCpu) == VINF_CPUM_HOST_CR0_MODIFIED)
8758 HMCPU_CF_SET(pVCpu, HM_CHANGED_HOST_CONTEXT);
8759 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_CR0));
8760 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8761 }
8762
8763 /*
8764 * Lazy-update of the host MSRs values in the auto-load/store MSR area.
8765 */
8766 if ( !pVCpu->hm.s.vmx.fUpdatedHostMsrs
8767 && pVCpu->hm.s.vmx.cMsrs > 0)
8768 {
8769 hmR0VmxUpdateAutoLoadStoreHostMsrs(pVCpu);
8770 }
8771
8772 /*
8773 * Load the host state bits as we may've been preempted (only happens when
8774 * thread-context hooks are used or when hmR0VmxSetupVMRunHandler() changes pfnStartVM).
8775 * Note that the 64-on-32 switcher saves the (64-bit) host state into the VMCS and
8776 * if we change the switcher back to 32-bit, we *must* save the 32-bit host state here.
8777 * See @bugref{8432}.
8778 */
8779 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT))
8780 {
8781 int rc = hmR0VmxSaveHostState(pVM, pVCpu);
8782 AssertRC(rc);
8783 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchPreemptSaveHostState);
8784 }
8785 Assert(!HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_CONTEXT));
8786
8787 /*
8788 * Load the state shared between host and guest (FPU, debug, lazy MSRs).
8789 */
8790 if (HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_HOST_GUEST_SHARED_STATE))
8791 hmR0VmxLoadSharedState(pVM, pVCpu, pMixedCtx);
8792 AssertMsg(!HMCPU_CF_VALUE(pVCpu), ("fContextUseFlags=%#RX32\n", HMCPU_CF_VALUE(pVCpu)));
8793
8794 /* Store status of the shared guest-host state at the time of VM-entry. */
8795#if HC_ARCH_BITS == 32 && defined(VBOX_WITH_64_BITS_GUESTS)
8796 if (CPUMIsGuestInLongModeEx(pMixedCtx))
8797 {
8798 pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActivePending(pVCpu);
8799 pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActivePending(pVCpu);
8800 }
8801 else
8802#endif
8803 {
8804 pVmxTransient->fWasGuestDebugStateActive = CPUMIsGuestDebugStateActive(pVCpu);
8805 pVmxTransient->fWasHyperDebugStateActive = CPUMIsHyperDebugStateActive(pVCpu);
8806 }
8807 pVmxTransient->fWasGuestFPUStateActive = CPUMIsGuestFPUStateActive(pVCpu);
8808
8809 /*
8810 * Cache the TPR-shadow for checking on every VM-exit if it might have changed.
8811 */
8812 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
8813 pVmxTransient->u8GuestTpr = pVCpu->hm.s.vmx.pbVirtApic[XAPIC_OFF_TPR];
8814
8815 PHMGLOBALCPUINFO pCpu = HMR0GetCurrentCpu();
8816 RTCPUID idCurrentCpu = pCpu->idCpu;
8817 if ( pVmxTransient->fUpdateTscOffsettingAndPreemptTimer
8818 || idCurrentCpu != pVCpu->hm.s.idLastCpu)
8819 {
8820 hmR0VmxUpdateTscOffsettingAndPreemptTimer(pVM, pVCpu);
8821 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = false;
8822 }
8823
8824 ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, true); /* Used for TLB flushing, set this across the world switch. */
8825 hmR0VmxFlushTaggedTlb(pVCpu, pCpu); /* Invalidate the appropriate guest entries from the TLB. */
8826 Assert(idCurrentCpu == pVCpu->hm.s.idLastCpu);
8827 pVCpu->hm.s.vmx.LastError.idCurrentCpu = idCurrentCpu; /* Update the error reporting info. with the current host CPU. */
8828
8829 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatEntry, &pVCpu->hm.s.StatInGC, x);
8830
8831 TMNotifyStartOfExecution(pVCpu); /* Finally, notify TM to resume its clocks as we're about
8832 to start executing. */
8833
8834 /*
8835 * Load the TSC_AUX MSR when we are not intercepting RDTSCP.
8836 */
8837 if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_RDTSCP)
8838 {
8839 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT))
8840 {
8841 bool fMsrUpdated;
8842 int rc2 = hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
8843 AssertRC(rc2);
8844 Assert(HMVMXCPU_GST_IS_UPDATED(pVCpu, HMVMX_UPDATED_GUEST_AUTO_LOAD_STORE_MSRS));
8845
8846 rc2 = hmR0VmxAddAutoLoadStoreMsr(pVCpu, MSR_K8_TSC_AUX, CPUMR0GetGuestTscAux(pVCpu), true /* fUpdateHostMsr */,
8847 &fMsrUpdated);
8848 AssertRC(rc2);
8849 Assert(fMsrUpdated || pVCpu->hm.s.vmx.fUpdatedHostMsrs);
8850
8851 /* Finally, mark that all host MSR values are updated so we don't redo it without leaving VT-x. See @bugref{6956}. */
8852 pVCpu->hm.s.vmx.fUpdatedHostMsrs = true;
8853 }
8854 else
8855 {
8856 hmR0VmxRemoveAutoLoadStoreMsr(pVCpu, MSR_K8_TSC_AUX);
8857 Assert(!pVCpu->hm.s.vmx.cMsrs || pVCpu->hm.s.vmx.fUpdatedHostMsrs);
8858 }
8859 }
8860
8861#ifdef VBOX_STRICT
8862 hmR0VmxCheckAutoLoadStoreMsrs(pVCpu);
8863 hmR0VmxCheckHostEferMsr(pVCpu);
8864 AssertRC(hmR0VmxCheckVmcsCtls(pVCpu));
8865#endif
8866#ifdef HMVMX_ALWAYS_CHECK_GUEST_STATE
8867 uint32_t uInvalidReason = hmR0VmxCheckGuestState(pVM, pVCpu, pMixedCtx);
8868 if (uInvalidReason != VMX_IGS_REASON_NOT_FOUND)
8869 Log4(("hmR0VmxCheckGuestState returned %#x\n", uInvalidReason));
8870#endif
8871}
8872
8873
8874/**
8875 * Performs some essential restoration of state after running guest code in
8876 * VT-x.
8877 *
8878 * @param pVM The cross context VM structure.
8879 * @param pVCpu The cross context virtual CPU structure.
8880 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
8881 * out-of-sync. Make sure to update the required fields
8882 * before using them.
8883 * @param pVmxTransient Pointer to the VMX transient structure.
8884 * @param rcVMRun Return code of VMLAUNCH/VMRESUME.
8885 *
8886 * @remarks Called with interrupts disabled, and returns with interrupts enabled!
8887 *
8888 * @remarks No-long-jump zone!!! This function will however re-enable longjmps
8889 * unconditionally when it is safe to do so.
8890 */
8891static void hmR0VmxPostRunGuest(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, int rcVMRun)
8892{
8893 NOREF(pVM);
8894
8895 Assert(!VMMRZCallRing3IsEnabled(pVCpu));
8896
8897 ASMAtomicWriteBool(&pVCpu->hm.s.fCheckedTLBFlush, false); /* See HMInvalidatePageOnAllVCpus(): used for TLB flushing. */
8898 ASMAtomicIncU32(&pVCpu->hm.s.cWorldSwitchExits); /* Initialized in vmR3CreateUVM(): used for EMT poking. */
8899 HMVMXCPU_GST_RESET_TO(pVCpu, 0); /* Exits/longjmps to ring-3 requires saving the guest state. */
8900 pVmxTransient->fVmcsFieldsRead = 0; /* Transient fields need to be read from the VMCS. */
8901 pVmxTransient->fVectoringPF = false; /* Vectoring page-fault needs to be determined later. */
8902 pVmxTransient->fVectoringDoublePF = false; /* Vectoring double page-fault needs to be determined later. */
8903
8904 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT))
8905 TMCpuTickSetLastSeen(pVCpu, ASMReadTSC() + pVCpu->hm.s.vmx.u64TSCOffset);
8906
8907 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatInGC, &pVCpu->hm.s.StatExit1, x);
8908 TMNotifyEndOfExecution(pVCpu); /* Notify TM that the guest is no longer running. */
8909 Assert(!ASMIntAreEnabled());
8910 VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED_HM);
8911
8912#ifdef HMVMX_ALWAYS_SWAP_FPU_STATE
8913 if (CPUMR0FpuStateMaybeSaveGuestAndRestoreHost(pVM, pVCpu))
8914 {
8915 hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
8916 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
8917 }
8918#endif
8919
8920#if HC_ARCH_BITS == 64
8921 pVCpu->hm.s.vmx.fRestoreHostFlags |= VMX_RESTORE_HOST_REQUIRED; /* Host state messed up by VT-x, we must restore. */
8922#endif
8923#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
8924 /* The 64-on-32 switcher maintains uVmcsState on its own and we need to leave it alone here. */
8925 if (pVCpu->hm.s.vmx.pfnStartVM != VMXR0SwitcherStartVM64)
8926 pVCpu->hm.s.vmx.uVmcsState |= HMVMX_VMCS_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
8927#else
8928 pVCpu->hm.s.vmx.uVmcsState |= HMVMX_VMCS_STATE_LAUNCHED; /* Use VMRESUME instead of VMLAUNCH in the next run. */
8929#endif
8930#ifdef VBOX_STRICT
8931 hmR0VmxCheckHostEferMsr(pVCpu); /* Verify that VMRUN/VMLAUNCH didn't modify host EFER. */
8932#endif
8933 ASMSetFlags(pVmxTransient->fEFlags); /* Enable interrupts. */
8934 VMMRZCallRing3Enable(pVCpu); /* It is now safe to do longjmps to ring-3!!! */
8935
8936 /* Save the basic VM-exit reason. Refer Intel spec. 24.9.1 "Basic VM-exit Information". */
8937 uint32_t uExitReason;
8938 int rc = VMXReadVmcs32(VMX_VMCS32_RO_EXIT_REASON, &uExitReason);
8939 rc |= hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
8940 AssertRC(rc);
8941 pVmxTransient->uExitReason = (uint16_t)VMX_EXIT_REASON_BASIC(uExitReason);
8942 pVmxTransient->fVMEntryFailed = VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uEntryIntInfo);
8943
8944 /* If the VMLAUNCH/VMRESUME failed, we can bail out early. This does -not- cover VMX_EXIT_ERR_*. */
8945 if (RT_UNLIKELY(rcVMRun != VINF_SUCCESS))
8946 {
8947 Log4(("VM-entry failure: pVCpu=%p idCpu=%RU32 rcVMRun=%Rrc fVMEntryFailed=%RTbool\n", pVCpu, pVCpu->idCpu, rcVMRun,
8948 pVmxTransient->fVMEntryFailed));
8949 return;
8950 }
8951
8952 /*
8953 * Update the VM-exit history array here even if the VM-entry failed due to:
8954 * - Invalid guest state.
8955 * - MSR loading.
8956 * - Machine-check event.
8957 *
8958 * In any of the above cases we will still have a "valid" VM-exit reason
8959 * despite @a fVMEntryFailed being false.
8960 *
8961 * See Intel spec. 26.7 "VM-Entry failures during or after loading guest state".
8962 */
8963 HMCPU_EXIT_HISTORY_ADD(pVCpu, pVmxTransient->uExitReason);
8964
8965 if (RT_LIKELY(!pVmxTransient->fVMEntryFailed))
8966 {
8967 /** @todo We can optimize this by only syncing with our force-flags when
8968 * really needed and keeping the VMCS state as it is for most
8969 * VM-exits. */
8970 /* Update the guest interruptibility-state from the VMCS. */
8971 hmR0VmxSaveGuestIntrState(pVCpu, pMixedCtx);
8972
8973#if defined(HMVMX_ALWAYS_SYNC_FULL_GUEST_STATE) || defined(HMVMX_ALWAYS_SAVE_FULL_GUEST_STATE)
8974 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
8975 AssertRC(rc);
8976#elif defined(HMVMX_ALWAYS_SAVE_GUEST_RFLAGS)
8977 rc = hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
8978 AssertRC(rc);
8979#endif
8980
8981 /*
8982 * Sync the TPR shadow with our APIC state.
8983 */
8984 if ( (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
8985 && pVmxTransient->u8GuestTpr != pVCpu->hm.s.vmx.pbVirtApic[XAPIC_OFF_TPR])
8986 {
8987 rc = APICSetTpr(pVCpu, pVCpu->hm.s.vmx.pbVirtApic[XAPIC_OFF_TPR]);
8988 AssertRC(rc);
8989 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
8990 }
8991 }
8992}
8993
8994
8995/**
8996 * Runs the guest code using VT-x the normal way.
8997 *
8998 * @returns VBox status code.
8999 * @param pVM The cross context VM structure.
9000 * @param pVCpu The cross context virtual CPU structure.
9001 * @param pCtx Pointer to the guest-CPU context.
9002 *
9003 * @note Mostly the same as hmR0VmxRunGuestCodeStep().
9004 */
9005static VBOXSTRICTRC hmR0VmxRunGuestCodeNormal(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
9006{
9007 VMXTRANSIENT VmxTransient;
9008 VmxTransient.fUpdateTscOffsettingAndPreemptTimer = true;
9009 VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
9010 uint32_t cLoops = 0;
9011
9012 for (;; cLoops++)
9013 {
9014 Assert(!HMR0SuspendPending());
9015 HMVMX_ASSERT_CPU_SAFE();
9016
9017 /* Preparatory work for running guest code, this may force us to return
9018 to ring-3. This bugger disables interrupts on VINF_SUCCESS! */
9019 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
9020 rcStrict = hmR0VmxPreRunGuest(pVM, pVCpu, pCtx, &VmxTransient, false /* fStepping */);
9021 if (rcStrict != VINF_SUCCESS)
9022 break;
9023
9024 hmR0VmxPreRunGuestCommitted(pVM, pVCpu, pCtx, &VmxTransient);
9025 int rcRun = hmR0VmxRunGuest(pVM, pVCpu, pCtx);
9026 /* The guest-CPU context is now outdated, 'pCtx' is to be treated as 'pMixedCtx' from this point on!!! */
9027
9028 /* Restore any residual host-state and save any bits shared between host
9029 and guest into the guest-CPU state. Re-enables interrupts! */
9030 hmR0VmxPostRunGuest(pVM, pVCpu, pCtx, &VmxTransient, rcRun);
9031
9032 /* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
9033 if (RT_SUCCESS(rcRun))
9034 { /* very likely */ }
9035 else
9036 {
9037 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x);
9038 hmR0VmxReportWorldSwitchError(pVM, pVCpu, rcRun, pCtx, &VmxTransient);
9039 return rcRun;
9040 }
9041
9042 /* Profile the VM-exit. */
9043 AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
9044 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
9045 STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
9046 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
9047 HMVMX_START_EXIT_DISPATCH_PROF();
9048
9049 VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, pCtx, VmxTransient.uExitReason);
9050
9051 /* Handle the VM-exit. */
9052#ifdef HMVMX_USE_FUNCTION_TABLE
9053 rcStrict = g_apfnVMExitHandlers[VmxTransient.uExitReason](pVCpu, pCtx, &VmxTransient);
9054#else
9055 rcStrict = hmR0VmxHandleExit(pVCpu, pCtx, &VmxTransient, VmxTransient.uExitReason);
9056#endif
9057 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
9058 if (rcStrict == VINF_SUCCESS)
9059 {
9060 if (cLoops <= pVM->hm.s.cMaxResumeLoops)
9061 continue; /* likely */
9062 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
9063 rcStrict = VINF_EM_RAW_INTERRUPT;
9064 }
9065 break;
9066 }
9067
9068 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
9069 return rcStrict;
9070}
9071
9072
9073
9074/** @name Execution loop for single stepping, DBGF events and expensive Dtrace
9075 * probes.
9076 *
9077 * The following few functions and associated structure contains the bloat
9078 * necessary for providing detailed debug events and dtrace probes as well as
9079 * reliable host side single stepping. This works on the principle of
9080 * "subclassing" the normal execution loop and workers. We replace the loop
9081 * method completely and override selected helpers to add necessary adjustments
9082 * to their core operation.
9083 *
9084 * The goal is to keep the "parent" code lean and mean, so as not to sacrifice
9085 * any performance for debug and analysis features.
9086 *
9087 * @{
9088 */
9089
9090/**
9091 * Transient per-VCPU debug state of VMCS and related info. we save/restore in
9092 * the debug run loop.
9093 */
9094typedef struct VMXRUNDBGSTATE
9095{
9096 /** The RIP we started executing at. This is for detecting that we stepped. */
9097 uint64_t uRipStart;
9098 /** The CS we started executing with. */
9099 uint16_t uCsStart;
9100
9101 /** Whether we've actually modified the 1st execution control field. */
9102 bool fModifiedProcCtls : 1;
9103 /** Whether we've actually modified the 2nd execution control field. */
9104 bool fModifiedProcCtls2 : 1;
9105 /** Whether we've actually modified the exception bitmap. */
9106 bool fModifiedXcptBitmap : 1;
9107
9108 /** We desire the modified the CR0 mask to be cleared. */
9109 bool fClearCr0Mask : 1;
9110 /** We desire the modified the CR4 mask to be cleared. */
9111 bool fClearCr4Mask : 1;
9112 /** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC. */
9113 uint32_t fCpe1Extra;
9114 /** Stuff we do not want in VMX_VMCS32_CTRL_PROC_EXEC. */
9115 uint32_t fCpe1Unwanted;
9116 /** Stuff we need in VMX_VMCS32_CTRL_PROC_EXEC2. */
9117 uint32_t fCpe2Extra;
9118 /** Extra stuff we need in VMX_VMCS32_CTRL_EXCEPTION_BITMAP. */
9119 uint32_t bmXcptExtra;
9120 /** The sequence number of the Dtrace provider settings the state was
9121 * configured against. */
9122 uint32_t uDtraceSettingsSeqNo;
9123 /** VM-exits to check (one bit per VM-exit). */
9124 uint32_t bmExitsToCheck[3];
9125
9126 /** The initial VMX_VMCS32_CTRL_PROC_EXEC value (helps with restore). */
9127 uint32_t fProcCtlsInitial;
9128 /** The initial VMX_VMCS32_CTRL_PROC_EXEC2 value (helps with restore). */
9129 uint32_t fProcCtls2Initial;
9130 /** The initial VMX_VMCS32_CTRL_EXCEPTION_BITMAP value (helps with restore). */
9131 uint32_t bmXcptInitial;
9132} VMXRUNDBGSTATE;
9133AssertCompileMemberSize(VMXRUNDBGSTATE, bmExitsToCheck, (VMX_EXIT_MAX + 1 + 31) / 32 * 4);
9134typedef VMXRUNDBGSTATE *PVMXRUNDBGSTATE;
9135
9136
9137/**
9138 * Initializes the VMXRUNDBGSTATE structure.
9139 *
9140 * @param pVCpu The cross context virtual CPU structure of the
9141 * calling EMT.
9142 * @param pCtx The CPU register context to go with @a pVCpu.
9143 * @param pDbgState The structure to initialize.
9144 */
9145DECLINLINE(void) hmR0VmxRunDebugStateInit(PVMCPU pVCpu, PCCPUMCTX pCtx, PVMXRUNDBGSTATE pDbgState)
9146{
9147 pDbgState->uRipStart = pCtx->rip;
9148 pDbgState->uCsStart = pCtx->cs.Sel;
9149
9150 pDbgState->fModifiedProcCtls = false;
9151 pDbgState->fModifiedProcCtls2 = false;
9152 pDbgState->fModifiedXcptBitmap = false;
9153 pDbgState->fClearCr0Mask = false;
9154 pDbgState->fClearCr4Mask = false;
9155 pDbgState->fCpe1Extra = 0;
9156 pDbgState->fCpe1Unwanted = 0;
9157 pDbgState->fCpe2Extra = 0;
9158 pDbgState->bmXcptExtra = 0;
9159 pDbgState->fProcCtlsInitial = pVCpu->hm.s.vmx.u32ProcCtls;
9160 pDbgState->fProcCtls2Initial = pVCpu->hm.s.vmx.u32ProcCtls2;
9161 pDbgState->bmXcptInitial = pVCpu->hm.s.vmx.u32XcptBitmap;
9162}
9163
9164
9165/**
9166 * Updates the VMSC fields with changes requested by @a pDbgState.
9167 *
9168 * This is performed after hmR0VmxPreRunGuestDebugStateUpdate as well
9169 * immediately before executing guest code, i.e. when interrupts are disabled.
9170 * We don't check status codes here as we cannot easily assert or return in the
9171 * latter case.
9172 *
9173 * @param pVCpu The cross context virtual CPU structure.
9174 * @param pDbgState The debug state.
9175 */
9176DECLINLINE(void) hmR0VmxPreRunGuestDebugStateApply(PVMCPU pVCpu, PVMXRUNDBGSTATE pDbgState)
9177{
9178 /*
9179 * Ensure desired flags in VMCS control fields are set.
9180 * (Ignoring write failure here, as we're committed and it's just debug extras.)
9181 *
9182 * Note! We load the shadow CR0 & CR4 bits when we flag the clearing, so
9183 * there should be no stale data in pCtx at this point.
9184 */
9185 if ( (pVCpu->hm.s.vmx.u32ProcCtls & pDbgState->fCpe1Extra) != pDbgState->fCpe1Extra
9186 || (pVCpu->hm.s.vmx.u32ProcCtls & pDbgState->fCpe1Unwanted))
9187 {
9188 pVCpu->hm.s.vmx.u32ProcCtls |= pDbgState->fCpe1Extra;
9189 pVCpu->hm.s.vmx.u32ProcCtls &= ~pDbgState->fCpe1Unwanted;
9190 VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
9191 Log6(("hmR0VmxRunDebugStateRevert: VMX_VMCS32_CTRL_PROC_EXEC: %#RX32\n", pVCpu->hm.s.vmx.u32ProcCtls));
9192 pDbgState->fModifiedProcCtls = true;
9193 }
9194
9195 if ((pVCpu->hm.s.vmx.u32ProcCtls2 & pDbgState->fCpe2Extra) != pDbgState->fCpe2Extra)
9196 {
9197 pVCpu->hm.s.vmx.u32ProcCtls2 |= pDbgState->fCpe2Extra;
9198 VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pVCpu->hm.s.vmx.u32ProcCtls2);
9199 Log6(("hmR0VmxRunDebugStateRevert: VMX_VMCS32_CTRL_PROC_EXEC2: %#RX32\n", pVCpu->hm.s.vmx.u32ProcCtls2));
9200 pDbgState->fModifiedProcCtls2 = true;
9201 }
9202
9203 if ((pVCpu->hm.s.vmx.u32XcptBitmap & pDbgState->bmXcptExtra) != pDbgState->bmXcptExtra)
9204 {
9205 pVCpu->hm.s.vmx.u32XcptBitmap |= pDbgState->bmXcptExtra;
9206 VMXWriteVmcs32(VMX_VMCS32_CTRL_EXCEPTION_BITMAP, pVCpu->hm.s.vmx.u32XcptBitmap);
9207 Log6(("hmR0VmxRunDebugStateRevert: VMX_VMCS32_CTRL_EXCEPTION_BITMAP: %#RX32\n", pVCpu->hm.s.vmx.u32XcptBitmap));
9208 pDbgState->fModifiedXcptBitmap = true;
9209 }
9210
9211 if (pDbgState->fClearCr0Mask && pVCpu->hm.s.vmx.u32CR0Mask != 0)
9212 {
9213 pVCpu->hm.s.vmx.u32CR0Mask = 0;
9214 VMXWriteVmcs32(VMX_VMCS_CTRL_CR0_MASK, 0);
9215 Log6(("hmR0VmxRunDebugStateRevert: VMX_VMCS_CTRL_CR0_MASK: 0\n"));
9216 }
9217
9218 if (pDbgState->fClearCr4Mask && pVCpu->hm.s.vmx.u32CR4Mask != 0)
9219 {
9220 pVCpu->hm.s.vmx.u32CR4Mask = 0;
9221 VMXWriteVmcs32(VMX_VMCS_CTRL_CR4_MASK, 0);
9222 Log6(("hmR0VmxRunDebugStateRevert: VMX_VMCS_CTRL_CR4_MASK: 0\n"));
9223 }
9224}
9225
9226
9227DECLINLINE(VBOXSTRICTRC) hmR0VmxRunDebugStateRevert(PVMCPU pVCpu, PVMXRUNDBGSTATE pDbgState, VBOXSTRICTRC rcStrict)
9228{
9229 /*
9230 * Restore VM-exit control settings as we may not reenter this function the
9231 * next time around.
9232 */
9233 /* We reload the initial value, trigger what we can of recalculations the
9234 next time around. From the looks of things, that's all that's required atm. */
9235 if (pDbgState->fModifiedProcCtls)
9236 {
9237 if (!(pDbgState->fProcCtlsInitial & VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT) && CPUMIsHyperDebugStateActive(pVCpu))
9238 pDbgState->fProcCtlsInitial |= VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT; /* Avoid assertion in hmR0VmxLeave */
9239 int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pDbgState->fProcCtlsInitial);
9240 AssertRCReturn(rc2, rc2);
9241 pVCpu->hm.s.vmx.u32ProcCtls = pDbgState->fProcCtlsInitial;
9242 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0 | HM_CHANGED_GUEST_DEBUG);
9243 }
9244
9245 /* We're currently the only ones messing with this one, so just restore the
9246 cached value and reload the field. */
9247 if ( pDbgState->fModifiedProcCtls2
9248 && pVCpu->hm.s.vmx.u32ProcCtls2 != pDbgState->fProcCtls2Initial)
9249 {
9250 int rc2 = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC2, pDbgState->fProcCtls2Initial);
9251 AssertRCReturn(rc2, rc2);
9252 pVCpu->hm.s.vmx.u32ProcCtls2 = pDbgState->fProcCtls2Initial;
9253 }
9254
9255 /* If we've modified the exception bitmap, we restore it and trigger
9256 reloading and partial recalculation the next time around. */
9257 if (pDbgState->fModifiedXcptBitmap)
9258 {
9259 pVCpu->hm.s.vmx.u32XcptBitmap = pDbgState->bmXcptInitial;
9260 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_XCPT_INTERCEPTS | HM_CHANGED_GUEST_CR0);
9261 }
9262
9263 /* We assume hmR0VmxLoadSharedCR0 will recalculate and load the CR0 mask. */
9264 if (pDbgState->fClearCr0Mask)
9265 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
9266
9267 /* We assume hmR0VmxLoadGuestCR3AndCR4 will recalculate and load the CR4 mask. */
9268 if (pDbgState->fClearCr4Mask)
9269 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR4);
9270
9271 return rcStrict;
9272}
9273
9274
9275/**
9276 * Configures VM-exit controls for current DBGF and DTrace settings.
9277 *
9278 * This updates @a pDbgState and the VMCS execution control fields to reflect
9279 * the necessary VM-exits demanded by DBGF and DTrace.
9280 *
9281 * @param pVM The cross context VM structure.
9282 * @param pVCpu The cross context virtual CPU structure.
9283 * @param pCtx Pointer to the guest-CPU context.
9284 * @param pDbgState The debug state.
9285 * @param pVmxTransient Pointer to the VMX transient structure. May update
9286 * fUpdateTscOffsettingAndPreemptTimer.
9287 */
9288static void hmR0VmxPreRunGuestDebugStateUpdate(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx,
9289 PVMXRUNDBGSTATE pDbgState, PVMXTRANSIENT pVmxTransient)
9290{
9291 /*
9292 * Take down the dtrace serial number so we can spot changes.
9293 */
9294 pDbgState->uDtraceSettingsSeqNo = VBOXVMM_GET_SETTINGS_SEQ_NO();
9295 ASMCompilerBarrier();
9296
9297 /*
9298 * We'll rebuild most of the middle block of data members (holding the
9299 * current settings) as we go along here, so start by clearing it all.
9300 */
9301 pDbgState->bmXcptExtra = 0;
9302 pDbgState->fCpe1Extra = 0;
9303 pDbgState->fCpe1Unwanted = 0;
9304 pDbgState->fCpe2Extra = 0;
9305 for (unsigned i = 0; i < RT_ELEMENTS(pDbgState->bmExitsToCheck); i++)
9306 pDbgState->bmExitsToCheck[i] = 0;
9307
9308 /*
9309 * Software interrupts (INT XXh) - no idea how to trigger these...
9310 */
9311 if ( DBGF_IS_EVENT_ENABLED(pVM, DBGFEVENT_INTERRUPT_SOFTWARE)
9312 || VBOXVMM_INT_SOFTWARE_ENABLED())
9313 {
9314 ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
9315 }
9316
9317 /*
9318 * INT3 breakpoints - triggered by #BP exceptions.
9319 */
9320 if (pVM->dbgf.ro.cEnabledInt3Breakpoints > 0)
9321 pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_BP);
9322
9323 /*
9324 * Exception bitmap and XCPT events+probes.
9325 */
9326 for (int iXcpt = 0; iXcpt < (DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST + 1); iXcpt++)
9327 if (DBGF_IS_EVENT_ENABLED(pVM, (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + iXcpt)))
9328 pDbgState->bmXcptExtra |= RT_BIT_32(iXcpt);
9329
9330 if (VBOXVMM_XCPT_DE_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_DE);
9331 if (VBOXVMM_XCPT_DB_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_DB);
9332 if (VBOXVMM_XCPT_BP_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_BP);
9333 if (VBOXVMM_XCPT_OF_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_OF);
9334 if (VBOXVMM_XCPT_BR_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_BR);
9335 if (VBOXVMM_XCPT_UD_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_UD);
9336 if (VBOXVMM_XCPT_NM_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_NM);
9337 if (VBOXVMM_XCPT_DF_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_DF);
9338 if (VBOXVMM_XCPT_TS_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_TS);
9339 if (VBOXVMM_XCPT_NP_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_NP);
9340 if (VBOXVMM_XCPT_SS_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_SS);
9341 if (VBOXVMM_XCPT_GP_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_GP);
9342 if (VBOXVMM_XCPT_PF_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_PF);
9343 if (VBOXVMM_XCPT_MF_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_MF);
9344 if (VBOXVMM_XCPT_AC_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_AC);
9345 if (VBOXVMM_XCPT_XF_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_XF);
9346 if (VBOXVMM_XCPT_VE_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_VE);
9347 if (VBOXVMM_XCPT_SX_ENABLED()) pDbgState->bmXcptExtra |= RT_BIT_32(X86_XCPT_SX);
9348
9349 if (pDbgState->bmXcptExtra)
9350 ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XCPT_OR_NMI);
9351
9352 /*
9353 * Process events and probes for VM-exits, making sure we get the wanted VM-exits.
9354 *
9355 * Note! This is the reverse of waft hmR0VmxHandleExitDtraceEvents does.
9356 * So, when adding/changing/removing please don't forget to update it.
9357 *
9358 * Some of the macros are picking up local variables to save horizontal space,
9359 * (being able to see it in a table is the lesser evil here).
9360 */
9361#define IS_EITHER_ENABLED(a_pVM, a_EventSubName) \
9362 ( DBGF_IS_EVENT_ENABLED(a_pVM, RT_CONCAT(DBGFEVENT_, a_EventSubName)) \
9363 || RT_CONCAT3(VBOXVMM_, a_EventSubName, _ENABLED)() )
9364#define SET_ONLY_XBM_IF_EITHER_EN(a_EventSubName, a_uExit) \
9365 if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
9366 { AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
9367 ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
9368 } else do { } while (0)
9369#define SET_CPE1_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec) \
9370 if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
9371 { \
9372 (pDbgState)->fCpe1Extra |= (a_fCtrlProcExec); \
9373 AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
9374 ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
9375 } else do { } while (0)
9376#define SET_CPEU_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fUnwantedCtrlProcExec) \
9377 if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
9378 { \
9379 (pDbgState)->fCpe1Unwanted |= (a_fUnwantedCtrlProcExec); \
9380 AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
9381 ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
9382 } else do { } while (0)
9383#define SET_CPE2_XBM_IF_EITHER_EN(a_EventSubName, a_uExit, a_fCtrlProcExec2) \
9384 if (IS_EITHER_ENABLED(pVM, a_EventSubName)) \
9385 { \
9386 (pDbgState)->fCpe2Extra |= (a_fCtrlProcExec2); \
9387 AssertCompile((unsigned)(a_uExit) < sizeof(pDbgState->bmExitsToCheck) * 8); \
9388 ASMBitSet((pDbgState)->bmExitsToCheck, a_uExit); \
9389 } else do { } while (0)
9390
9391 SET_ONLY_XBM_IF_EITHER_EN(EXIT_TASK_SWITCH, VMX_EXIT_TASK_SWITCH); /* unconditional */
9392 SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_VIOLATION, VMX_EXIT_EPT_VIOLATION); /* unconditional */
9393 SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_EPT_MISCONFIG, VMX_EXIT_EPT_MISCONFIG); /* unconditional (unless #VE) */
9394 SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_ACCESS, VMX_EXIT_APIC_ACCESS); /* feature dependent, nothing to enable here */
9395 SET_ONLY_XBM_IF_EITHER_EN(EXIT_VMX_VAPIC_WRITE, VMX_EXIT_APIC_WRITE); /* feature dependent, nothing to enable here */
9396
9397 SET_ONLY_XBM_IF_EITHER_EN(INSTR_CPUID, VMX_EXIT_CPUID); /* unconditional */
9398 SET_ONLY_XBM_IF_EITHER_EN( EXIT_CPUID, VMX_EXIT_CPUID);
9399 SET_ONLY_XBM_IF_EITHER_EN(INSTR_GETSEC, VMX_EXIT_GETSEC); /* unconditional */
9400 SET_ONLY_XBM_IF_EITHER_EN( EXIT_GETSEC, VMX_EXIT_GETSEC);
9401 SET_CPE1_XBM_IF_EITHER_EN(INSTR_HALT, VMX_EXIT_HLT, VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT); /* paranoia */
9402 SET_ONLY_XBM_IF_EITHER_EN( EXIT_HALT, VMX_EXIT_HLT);
9403 SET_ONLY_XBM_IF_EITHER_EN(INSTR_INVD, VMX_EXIT_INVD); /* unconditional */
9404 SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVD, VMX_EXIT_INVD);
9405 SET_CPE1_XBM_IF_EITHER_EN(INSTR_INVLPG, VMX_EXIT_INVLPG, VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT);
9406 SET_ONLY_XBM_IF_EITHER_EN( EXIT_INVLPG, VMX_EXIT_INVLPG);
9407 SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDPMC, VMX_EXIT_RDPMC, VMX_VMCS_CTRL_PROC_EXEC_RDPMC_EXIT);
9408 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDPMC, VMX_EXIT_RDPMC);
9409 SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSC, VMX_EXIT_RDTSC, VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT);
9410 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSC, VMX_EXIT_RDTSC);
9411 SET_ONLY_XBM_IF_EITHER_EN(INSTR_RSM, VMX_EXIT_RSM); /* unconditional */
9412 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RSM, VMX_EXIT_RSM);
9413 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMM_CALL, VMX_EXIT_VMCALL); /* unconditional */
9414 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMM_CALL, VMX_EXIT_VMCALL);
9415 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMCLEAR, VMX_EXIT_VMCLEAR); /* unconditional */
9416 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMCLEAR, VMX_EXIT_VMCLEAR);
9417 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH); /* unconditional */
9418 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMLAUNCH, VMX_EXIT_VMLAUNCH);
9419 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRLD, VMX_EXIT_VMPTRLD); /* unconditional */
9420 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRLD, VMX_EXIT_VMPTRLD);
9421 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMPTRST, VMX_EXIT_VMPTRST); /* unconditional */
9422 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMPTRST, VMX_EXIT_VMPTRST);
9423 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMREAD, VMX_EXIT_VMREAD); /* unconditional */
9424 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMREAD, VMX_EXIT_VMREAD);
9425 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMRESUME, VMX_EXIT_VMRESUME); /* unconditional */
9426 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMRESUME, VMX_EXIT_VMRESUME);
9427 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMWRITE, VMX_EXIT_VMWRITE); /* unconditional */
9428 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMWRITE, VMX_EXIT_VMWRITE);
9429 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXOFF, VMX_EXIT_VMXOFF); /* unconditional */
9430 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXOFF, VMX_EXIT_VMXOFF);
9431 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMXON, VMX_EXIT_VMXON); /* unconditional */
9432 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMXON, VMX_EXIT_VMXON);
9433
9434 if ( IS_EITHER_ENABLED(pVM, INSTR_CRX_READ)
9435 || IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
9436 {
9437 int rc2 = hmR0VmxSaveGuestCR0(pVCpu, pCtx);
9438 rc2 |= hmR0VmxSaveGuestCR4(pVCpu, pCtx);
9439 rc2 |= hmR0VmxSaveGuestApicState(pVCpu, pCtx);
9440 AssertRC(rc2);
9441
9442#if 0 /** @todo fix me */
9443 pDbgState->fClearCr0Mask = true;
9444 pDbgState->fClearCr4Mask = true;
9445#endif
9446 if (IS_EITHER_ENABLED(pVM, INSTR_CRX_READ))
9447 pDbgState->fCpe1Extra |= VMX_VMCS_CTRL_PROC_EXEC_CR3_STORE_EXIT | VMX_VMCS_CTRL_PROC_EXEC_CR8_STORE_EXIT;
9448 if (IS_EITHER_ENABLED(pVM, INSTR_CRX_WRITE))
9449 pDbgState->fCpe1Extra |= VMX_VMCS_CTRL_PROC_EXEC_CR3_LOAD_EXIT | VMX_VMCS_CTRL_PROC_EXEC_CR8_LOAD_EXIT;
9450 pDbgState->fCpe1Unwanted |= VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW; /* risky? */
9451 /* Note! We currently don't use VMX_VMCS32_CTRL_CR3_TARGET_COUNT. It would
9452 require clearing here and in the loop if we start using it. */
9453 ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_CRX);
9454 }
9455 else
9456 {
9457 if (pDbgState->fClearCr0Mask)
9458 {
9459 pDbgState->fClearCr0Mask = false;
9460 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
9461 }
9462 if (pDbgState->fClearCr4Mask)
9463 {
9464 pDbgState->fClearCr4Mask = false;
9465 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR4);
9466 }
9467 }
9468 SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_READ, VMX_EXIT_MOV_CRX);
9469 SET_ONLY_XBM_IF_EITHER_EN( EXIT_CRX_WRITE, VMX_EXIT_MOV_CRX);
9470
9471 if ( IS_EITHER_ENABLED(pVM, INSTR_DRX_READ)
9472 || IS_EITHER_ENABLED(pVM, INSTR_DRX_WRITE))
9473 {
9474 /** @todo later, need to fix handler as it assumes this won't usually happen. */
9475 ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_MOV_DRX);
9476 }
9477 SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_READ, VMX_EXIT_MOV_DRX);
9478 SET_ONLY_XBM_IF_EITHER_EN( EXIT_DRX_WRITE, VMX_EXIT_MOV_DRX);
9479
9480 SET_CPEU_XBM_IF_EITHER_EN(INSTR_RDMSR, VMX_EXIT_RDMSR, VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS); /* risky clearing this? */
9481 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDMSR, VMX_EXIT_RDMSR);
9482 SET_CPEU_XBM_IF_EITHER_EN(INSTR_WRMSR, VMX_EXIT_WRMSR, VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS);
9483 SET_ONLY_XBM_IF_EITHER_EN( EXIT_WRMSR, VMX_EXIT_WRMSR);
9484 SET_CPE1_XBM_IF_EITHER_EN(INSTR_MWAIT, VMX_EXIT_MWAIT, VMX_VMCS_CTRL_PROC_EXEC_MWAIT_EXIT); /* paranoia */
9485 SET_ONLY_XBM_IF_EITHER_EN( EXIT_MWAIT, VMX_EXIT_MWAIT);
9486 SET_CPE1_XBM_IF_EITHER_EN(INSTR_MONITOR, VMX_EXIT_MONITOR, VMX_VMCS_CTRL_PROC_EXEC_MONITOR_EXIT); /* paranoia */
9487 SET_ONLY_XBM_IF_EITHER_EN( EXIT_MONITOR, VMX_EXIT_MONITOR);
9488#if 0 /** @todo too slow, fix handler. */
9489 SET_CPE1_XBM_IF_EITHER_EN(INSTR_PAUSE, VMX_EXIT_PAUSE, VMX_VMCS_CTRL_PROC_EXEC_PAUSE_EXIT);
9490#endif
9491 SET_ONLY_XBM_IF_EITHER_EN( EXIT_PAUSE, VMX_EXIT_PAUSE);
9492
9493 if ( IS_EITHER_ENABLED(pVM, INSTR_SGDT)
9494 || IS_EITHER_ENABLED(pVM, INSTR_SIDT)
9495 || IS_EITHER_ENABLED(pVM, INSTR_LGDT)
9496 || IS_EITHER_ENABLED(pVM, INSTR_LIDT))
9497 {
9498 pDbgState->fCpe2Extra |= VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT;
9499 ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_XDTR_ACCESS);
9500 }
9501 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SGDT, VMX_EXIT_XDTR_ACCESS);
9502 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SIDT, VMX_EXIT_XDTR_ACCESS);
9503 SET_ONLY_XBM_IF_EITHER_EN( EXIT_LGDT, VMX_EXIT_XDTR_ACCESS);
9504 SET_ONLY_XBM_IF_EITHER_EN( EXIT_LIDT, VMX_EXIT_XDTR_ACCESS);
9505
9506 if ( IS_EITHER_ENABLED(pVM, INSTR_SLDT)
9507 || IS_EITHER_ENABLED(pVM, INSTR_STR)
9508 || IS_EITHER_ENABLED(pVM, INSTR_LLDT)
9509 || IS_EITHER_ENABLED(pVM, INSTR_LTR))
9510 {
9511 pDbgState->fCpe2Extra |= VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT;
9512 ASMBitSet(pDbgState->bmExitsToCheck, VMX_EXIT_TR_ACCESS);
9513 }
9514 SET_ONLY_XBM_IF_EITHER_EN( EXIT_SLDT, VMX_EXIT_TR_ACCESS);
9515 SET_ONLY_XBM_IF_EITHER_EN( EXIT_STR, VMX_EXIT_TR_ACCESS);
9516 SET_ONLY_XBM_IF_EITHER_EN( EXIT_LLDT, VMX_EXIT_TR_ACCESS);
9517 SET_ONLY_XBM_IF_EITHER_EN( EXIT_LTR, VMX_EXIT_TR_ACCESS);
9518
9519 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVEPT, VMX_EXIT_INVEPT); /* unconditional */
9520 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVEPT, VMX_EXIT_INVEPT);
9521 SET_CPE1_XBM_IF_EITHER_EN(INSTR_RDTSCP, VMX_EXIT_RDTSCP, VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT);
9522 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDTSCP, VMX_EXIT_RDTSCP);
9523 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_INVVPID, VMX_EXIT_INVVPID); /* unconditional */
9524 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVVPID, VMX_EXIT_INVVPID);
9525 SET_CPE2_XBM_IF_EITHER_EN(INSTR_WBINVD, VMX_EXIT_WBINVD, VMX_VMCS_CTRL_PROC_EXEC2_WBINVD_EXIT);
9526 SET_ONLY_XBM_IF_EITHER_EN( EXIT_WBINVD, VMX_EXIT_WBINVD);
9527 SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSETBV, VMX_EXIT_XSETBV); /* unconditional */
9528 SET_ONLY_XBM_IF_EITHER_EN( EXIT_XSETBV, VMX_EXIT_XSETBV);
9529 SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDRAND, VMX_EXIT_RDRAND, VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT);
9530 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDRAND, VMX_EXIT_RDRAND);
9531 SET_CPE1_XBM_IF_EITHER_EN(INSTR_VMX_INVPCID, VMX_EXIT_INVPCID, VMX_VMCS_CTRL_PROC_EXEC_INVLPG_EXIT);
9532 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_INVPCID, VMX_EXIT_INVPCID);
9533 SET_ONLY_XBM_IF_EITHER_EN(INSTR_VMX_VMFUNC, VMX_EXIT_VMFUNC); /* unconditional for the current setup */
9534 SET_ONLY_XBM_IF_EITHER_EN( EXIT_VMX_VMFUNC, VMX_EXIT_VMFUNC);
9535 SET_CPE2_XBM_IF_EITHER_EN(INSTR_RDSEED, VMX_EXIT_RDSEED, VMX_VMCS_CTRL_PROC_EXEC2_RDSEED_EXIT);
9536 SET_ONLY_XBM_IF_EITHER_EN( EXIT_RDSEED, VMX_EXIT_RDSEED);
9537 SET_ONLY_XBM_IF_EITHER_EN(INSTR_XSAVES, VMX_EXIT_XSAVES); /* unconditional (enabled by host, guest cfg) */
9538 SET_ONLY_XBM_IF_EITHER_EN(EXIT_XSAVES, VMX_EXIT_XSAVES);
9539 SET_ONLY_XBM_IF_EITHER_EN(INSTR_XRSTORS, VMX_EXIT_XRSTORS); /* unconditional (enabled by host, guest cfg) */
9540 SET_ONLY_XBM_IF_EITHER_EN( EXIT_XRSTORS, VMX_EXIT_XRSTORS);
9541
9542#undef IS_EITHER_ENABLED
9543#undef SET_ONLY_XBM_IF_EITHER_EN
9544#undef SET_CPE1_XBM_IF_EITHER_EN
9545#undef SET_CPEU_XBM_IF_EITHER_EN
9546#undef SET_CPE2_XBM_IF_EITHER_EN
9547
9548 /*
9549 * Sanitize the control stuff.
9550 */
9551 pDbgState->fCpe2Extra &= pVM->hm.s.vmx.Msrs.VmxProcCtls2.n.allowed1;
9552 if (pDbgState->fCpe2Extra)
9553 pDbgState->fCpe1Extra |= VMX_VMCS_CTRL_PROC_EXEC_USE_SECONDARY_EXEC_CTRL;
9554 pDbgState->fCpe1Extra &= pVM->hm.s.vmx.Msrs.VmxProcCtls.n.allowed1;
9555 pDbgState->fCpe1Unwanted &= ~pVM->hm.s.vmx.Msrs.VmxProcCtls.n.disallowed0;
9556 if (pVCpu->hm.s.fDebugWantRdTscExit != RT_BOOL(pDbgState->fCpe1Extra & VMX_VMCS_CTRL_PROC_EXEC_RDTSC_EXIT))
9557 {
9558 pVCpu->hm.s.fDebugWantRdTscExit ^= true;
9559 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
9560 }
9561
9562 Log6(("HM: debug state: cpe1=%#RX32 cpeu=%#RX32 cpe2=%#RX32%s%s\n",
9563 pDbgState->fCpe1Extra, pDbgState->fCpe1Unwanted, pDbgState->fCpe2Extra,
9564 pDbgState->fClearCr0Mask ? " clr-cr0" : "",
9565 pDbgState->fClearCr4Mask ? " clr-cr4" : ""));
9566}
9567
9568
9569/**
9570 * Fires off DBGF events and dtrace probes for a VM-exit, when it's
9571 * appropriate.
9572 *
9573 * The caller has checked the VM-exit against the
9574 * VMXRUNDBGSTATE::bmExitsToCheck bitmap. The caller has checked for NMIs
9575 * already, so we don't have to do that either.
9576 *
9577 * @returns Strict VBox status code (i.e. informational status codes too).
9578 * @param pVM The cross context VM structure.
9579 * @param pVCpu The cross context virtual CPU structure.
9580 * @param pMixedCtx Pointer to the guest-CPU context.
9581 * @param pVmxTransient Pointer to the VMX-transient structure.
9582 * @param uExitReason The VM-exit reason.
9583 *
9584 * @remarks The name of this function is displayed by dtrace, so keep it short
9585 * and to the point. No longer than 33 chars long, please.
9586 */
9587static VBOXSTRICTRC hmR0VmxHandleExitDtraceEvents(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx,
9588 PVMXTRANSIENT pVmxTransient, uint32_t uExitReason)
9589{
9590 /*
9591 * Translate the event into a DBGF event (enmEvent + uEventArg) and at the
9592 * same time check whether any corresponding Dtrace event is enabled (fDtrace).
9593 *
9594 * Note! This is the reverse operation of what hmR0VmxPreRunGuestDebugStateUpdate
9595 * does. Must add/change/remove both places. Same ordering, please.
9596 *
9597 * Added/removed events must also be reflected in the next section
9598 * where we dispatch dtrace events.
9599 */
9600 bool fDtrace1 = false;
9601 bool fDtrace2 = false;
9602 DBGFEVENTTYPE enmEvent1 = DBGFEVENT_END;
9603 DBGFEVENTTYPE enmEvent2 = DBGFEVENT_END;
9604 uint32_t uEventArg = 0;
9605#define SET_EXIT(a_EventSubName) \
9606 do { \
9607 enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
9608 fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
9609 } while (0)
9610#define SET_BOTH(a_EventSubName) \
9611 do { \
9612 enmEvent1 = RT_CONCAT(DBGFEVENT_INSTR_, a_EventSubName); \
9613 enmEvent2 = RT_CONCAT(DBGFEVENT_EXIT_, a_EventSubName); \
9614 fDtrace1 = RT_CONCAT3(VBOXVMM_INSTR_, a_EventSubName, _ENABLED)(); \
9615 fDtrace2 = RT_CONCAT3(VBOXVMM_EXIT_, a_EventSubName, _ENABLED)(); \
9616 } while (0)
9617 switch (uExitReason)
9618 {
9619 case VMX_EXIT_MTF:
9620 return hmR0VmxExitMtf(pVCpu, pMixedCtx, pVmxTransient);
9621
9622 case VMX_EXIT_XCPT_OR_NMI:
9623 {
9624 uint8_t const idxVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo);
9625 switch (VMX_EXIT_INTERRUPTION_INFO_TYPE(pVmxTransient->uExitIntInfo))
9626 {
9627 case VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT:
9628 case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT:
9629 case VMX_EXIT_INTERRUPTION_INFO_TYPE_PRIV_SW_XCPT:
9630 if (idxVector <= (unsigned)(DBGFEVENT_XCPT_LAST - DBGFEVENT_XCPT_FIRST))
9631 {
9632 if (VMX_EXIT_INTERRUPTION_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uExitIntInfo))
9633 {
9634 hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
9635 uEventArg = pVmxTransient->uExitIntErrorCode;
9636 }
9637 enmEvent1 = (DBGFEVENTTYPE)(DBGFEVENT_XCPT_FIRST + idxVector);
9638 switch (enmEvent1)
9639 {
9640 case DBGFEVENT_XCPT_DE: fDtrace1 = VBOXVMM_XCPT_DE_ENABLED(); break;
9641 case DBGFEVENT_XCPT_DB: fDtrace1 = VBOXVMM_XCPT_DB_ENABLED(); break;
9642 case DBGFEVENT_XCPT_BP: fDtrace1 = VBOXVMM_XCPT_BP_ENABLED(); break;
9643 case DBGFEVENT_XCPT_OF: fDtrace1 = VBOXVMM_XCPT_OF_ENABLED(); break;
9644 case DBGFEVENT_XCPT_BR: fDtrace1 = VBOXVMM_XCPT_BR_ENABLED(); break;
9645 case DBGFEVENT_XCPT_UD: fDtrace1 = VBOXVMM_XCPT_UD_ENABLED(); break;
9646 case DBGFEVENT_XCPT_NM: fDtrace1 = VBOXVMM_XCPT_NM_ENABLED(); break;
9647 case DBGFEVENT_XCPT_DF: fDtrace1 = VBOXVMM_XCPT_DF_ENABLED(); break;
9648 case DBGFEVENT_XCPT_TS: fDtrace1 = VBOXVMM_XCPT_TS_ENABLED(); break;
9649 case DBGFEVENT_XCPT_NP: fDtrace1 = VBOXVMM_XCPT_NP_ENABLED(); break;
9650 case DBGFEVENT_XCPT_SS: fDtrace1 = VBOXVMM_XCPT_SS_ENABLED(); break;
9651 case DBGFEVENT_XCPT_GP: fDtrace1 = VBOXVMM_XCPT_GP_ENABLED(); break;
9652 case DBGFEVENT_XCPT_PF: fDtrace1 = VBOXVMM_XCPT_PF_ENABLED(); break;
9653 case DBGFEVENT_XCPT_MF: fDtrace1 = VBOXVMM_XCPT_MF_ENABLED(); break;
9654 case DBGFEVENT_XCPT_AC: fDtrace1 = VBOXVMM_XCPT_AC_ENABLED(); break;
9655 case DBGFEVENT_XCPT_XF: fDtrace1 = VBOXVMM_XCPT_XF_ENABLED(); break;
9656 case DBGFEVENT_XCPT_VE: fDtrace1 = VBOXVMM_XCPT_VE_ENABLED(); break;
9657 case DBGFEVENT_XCPT_SX: fDtrace1 = VBOXVMM_XCPT_SX_ENABLED(); break;
9658 default: break;
9659 }
9660 }
9661 else
9662 AssertFailed();
9663 break;
9664
9665 case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_INT:
9666 uEventArg = idxVector;
9667 enmEvent1 = DBGFEVENT_INTERRUPT_SOFTWARE;
9668 fDtrace1 = VBOXVMM_INT_SOFTWARE_ENABLED();
9669 break;
9670 }
9671 break;
9672 }
9673
9674 case VMX_EXIT_TRIPLE_FAULT:
9675 enmEvent1 = DBGFEVENT_TRIPLE_FAULT;
9676 //fDtrace1 = VBOXVMM_EXIT_TRIPLE_FAULT_ENABLED();
9677 break;
9678 case VMX_EXIT_TASK_SWITCH: SET_EXIT(TASK_SWITCH); break;
9679 case VMX_EXIT_EPT_VIOLATION: SET_EXIT(VMX_EPT_VIOLATION); break;
9680 case VMX_EXIT_EPT_MISCONFIG: SET_EXIT(VMX_EPT_MISCONFIG); break;
9681 case VMX_EXIT_APIC_ACCESS: SET_EXIT(VMX_VAPIC_ACCESS); break;
9682 case VMX_EXIT_APIC_WRITE: SET_EXIT(VMX_VAPIC_WRITE); break;
9683
9684 /* Instruction specific VM-exits: */
9685 case VMX_EXIT_CPUID: SET_BOTH(CPUID); break;
9686 case VMX_EXIT_GETSEC: SET_BOTH(GETSEC); break;
9687 case VMX_EXIT_HLT: SET_BOTH(HALT); break;
9688 case VMX_EXIT_INVD: SET_BOTH(INVD); break;
9689 case VMX_EXIT_INVLPG: SET_BOTH(INVLPG); break;
9690 case VMX_EXIT_RDPMC: SET_BOTH(RDPMC); break;
9691 case VMX_EXIT_RDTSC: SET_BOTH(RDTSC); break;
9692 case VMX_EXIT_RSM: SET_BOTH(RSM); break;
9693 case VMX_EXIT_VMCALL: SET_BOTH(VMM_CALL); break;
9694 case VMX_EXIT_VMCLEAR: SET_BOTH(VMX_VMCLEAR); break;
9695 case VMX_EXIT_VMLAUNCH: SET_BOTH(VMX_VMLAUNCH); break;
9696 case VMX_EXIT_VMPTRLD: SET_BOTH(VMX_VMPTRLD); break;
9697 case VMX_EXIT_VMPTRST: SET_BOTH(VMX_VMPTRST); break;
9698 case VMX_EXIT_VMREAD: SET_BOTH(VMX_VMREAD); break;
9699 case VMX_EXIT_VMRESUME: SET_BOTH(VMX_VMRESUME); break;
9700 case VMX_EXIT_VMWRITE: SET_BOTH(VMX_VMWRITE); break;
9701 case VMX_EXIT_VMXOFF: SET_BOTH(VMX_VMXOFF); break;
9702 case VMX_EXIT_VMXON: SET_BOTH(VMX_VMXON); break;
9703 case VMX_EXIT_MOV_CRX:
9704 hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
9705/** @todo r=bird: I feel these macros aren't very descriptive and needs to be at least 30 chars longer! ;-)
9706* Sensible abbreviations strongly recommended here because even with 130 columns this stuff get too wide! */
9707 if ( VMX_EXIT_QUALIFICATION_CRX_ACCESS(pVmxTransient->uExitQualification)
9708 == VMX_EXIT_QUALIFICATION_CRX_ACCESS_READ)
9709 SET_BOTH(CRX_READ);
9710 else
9711 SET_BOTH(CRX_WRITE);
9712 uEventArg = VMX_EXIT_QUALIFICATION_CRX_REGISTER(pVmxTransient->uExitQualification);
9713 break;
9714 case VMX_EXIT_MOV_DRX:
9715 hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
9716 if ( VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification)
9717 == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_READ)
9718 SET_BOTH(DRX_READ);
9719 else
9720 SET_BOTH(DRX_WRITE);
9721 uEventArg = VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification);
9722 break;
9723 case VMX_EXIT_RDMSR: SET_BOTH(RDMSR); break;
9724 case VMX_EXIT_WRMSR: SET_BOTH(WRMSR); break;
9725 case VMX_EXIT_MWAIT: SET_BOTH(MWAIT); break;
9726 case VMX_EXIT_MONITOR: SET_BOTH(MONITOR); break;
9727 case VMX_EXIT_PAUSE: SET_BOTH(PAUSE); break;
9728 case VMX_EXIT_XDTR_ACCESS:
9729 hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
9730 switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_XDTR_INSINFO_INSTR_ID))
9731 {
9732 case VMX_XDTR_INSINFO_II_SGDT: SET_BOTH(SGDT); break;
9733 case VMX_XDTR_INSINFO_II_SIDT: SET_BOTH(SIDT); break;
9734 case VMX_XDTR_INSINFO_II_LGDT: SET_BOTH(LGDT); break;
9735 case VMX_XDTR_INSINFO_II_LIDT: SET_BOTH(LIDT); break;
9736 }
9737 break;
9738
9739 case VMX_EXIT_TR_ACCESS:
9740 hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
9741 switch (RT_BF_GET(pVmxTransient->ExitInstrInfo.u, VMX_YYTR_INSINFO_INSTR_ID))
9742 {
9743 case VMX_YYTR_INSINFO_II_SLDT: SET_BOTH(SLDT); break;
9744 case VMX_YYTR_INSINFO_II_STR: SET_BOTH(STR); break;
9745 case VMX_YYTR_INSINFO_II_LLDT: SET_BOTH(LLDT); break;
9746 case VMX_YYTR_INSINFO_II_LTR: SET_BOTH(LTR); break;
9747 }
9748 break;
9749
9750 case VMX_EXIT_INVEPT: SET_BOTH(VMX_INVEPT); break;
9751 case VMX_EXIT_RDTSCP: SET_BOTH(RDTSCP); break;
9752 case VMX_EXIT_INVVPID: SET_BOTH(VMX_INVVPID); break;
9753 case VMX_EXIT_WBINVD: SET_BOTH(WBINVD); break;
9754 case VMX_EXIT_XSETBV: SET_BOTH(XSETBV); break;
9755 case VMX_EXIT_RDRAND: SET_BOTH(RDRAND); break;
9756 case VMX_EXIT_INVPCID: SET_BOTH(VMX_INVPCID); break;
9757 case VMX_EXIT_VMFUNC: SET_BOTH(VMX_VMFUNC); break;
9758 case VMX_EXIT_RDSEED: SET_BOTH(RDSEED); break;
9759 case VMX_EXIT_XSAVES: SET_BOTH(XSAVES); break;
9760 case VMX_EXIT_XRSTORS: SET_BOTH(XRSTORS); break;
9761
9762 /* Events that aren't relevant at this point. */
9763 case VMX_EXIT_EXT_INT:
9764 case VMX_EXIT_INT_WINDOW:
9765 case VMX_EXIT_NMI_WINDOW:
9766 case VMX_EXIT_TPR_BELOW_THRESHOLD:
9767 case VMX_EXIT_PREEMPT_TIMER:
9768 case VMX_EXIT_IO_INSTR:
9769 break;
9770
9771 /* Errors and unexpected events. */
9772 case VMX_EXIT_INIT_SIGNAL:
9773 case VMX_EXIT_SIPI:
9774 case VMX_EXIT_IO_SMI:
9775 case VMX_EXIT_SMI:
9776 case VMX_EXIT_ERR_INVALID_GUEST_STATE:
9777 case VMX_EXIT_ERR_MSR_LOAD:
9778 case VMX_EXIT_ERR_MACHINE_CHECK:
9779 break;
9780
9781 default:
9782 AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
9783 break;
9784 }
9785#undef SET_BOTH
9786#undef SET_EXIT
9787
9788 /*
9789 * Dtrace tracepoints go first. We do them here at once so we don't
9790 * have to copy the guest state saving and stuff a few dozen times.
9791 * Down side is that we've got to repeat the switch, though this time
9792 * we use enmEvent since the probes are a subset of what DBGF does.
9793 */
9794 if (fDtrace1 || fDtrace2)
9795 {
9796 hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
9797 hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
9798 switch (enmEvent1)
9799 {
9800 /** @todo consider which extra parameters would be helpful for each probe. */
9801 case DBGFEVENT_END: break;
9802 case DBGFEVENT_XCPT_DE: VBOXVMM_XCPT_DE(pVCpu, pMixedCtx); break;
9803 case DBGFEVENT_XCPT_DB: VBOXVMM_XCPT_DB(pVCpu, pMixedCtx, pMixedCtx->dr[6]); break;
9804 case DBGFEVENT_XCPT_BP: VBOXVMM_XCPT_BP(pVCpu, pMixedCtx); break;
9805 case DBGFEVENT_XCPT_OF: VBOXVMM_XCPT_OF(pVCpu, pMixedCtx); break;
9806 case DBGFEVENT_XCPT_BR: VBOXVMM_XCPT_BR(pVCpu, pMixedCtx); break;
9807 case DBGFEVENT_XCPT_UD: VBOXVMM_XCPT_UD(pVCpu, pMixedCtx); break;
9808 case DBGFEVENT_XCPT_NM: VBOXVMM_XCPT_NM(pVCpu, pMixedCtx); break;
9809 case DBGFEVENT_XCPT_DF: VBOXVMM_XCPT_DF(pVCpu, pMixedCtx); break;
9810 case DBGFEVENT_XCPT_TS: VBOXVMM_XCPT_TS(pVCpu, pMixedCtx, uEventArg); break;
9811 case DBGFEVENT_XCPT_NP: VBOXVMM_XCPT_NP(pVCpu, pMixedCtx, uEventArg); break;
9812 case DBGFEVENT_XCPT_SS: VBOXVMM_XCPT_SS(pVCpu, pMixedCtx, uEventArg); break;
9813 case DBGFEVENT_XCPT_GP: VBOXVMM_XCPT_GP(pVCpu, pMixedCtx, uEventArg); break;
9814 case DBGFEVENT_XCPT_PF: VBOXVMM_XCPT_PF(pVCpu, pMixedCtx, uEventArg, pMixedCtx->cr2); break;
9815 case DBGFEVENT_XCPT_MF: VBOXVMM_XCPT_MF(pVCpu, pMixedCtx); break;
9816 case DBGFEVENT_XCPT_AC: VBOXVMM_XCPT_AC(pVCpu, pMixedCtx); break;
9817 case DBGFEVENT_XCPT_XF: VBOXVMM_XCPT_XF(pVCpu, pMixedCtx); break;
9818 case DBGFEVENT_XCPT_VE: VBOXVMM_XCPT_VE(pVCpu, pMixedCtx); break;
9819 case DBGFEVENT_XCPT_SX: VBOXVMM_XCPT_SX(pVCpu, pMixedCtx, uEventArg); break;
9820 case DBGFEVENT_INTERRUPT_SOFTWARE: VBOXVMM_INT_SOFTWARE(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9821 case DBGFEVENT_INSTR_CPUID: VBOXVMM_INSTR_CPUID(pVCpu, pMixedCtx, pMixedCtx->eax, pMixedCtx->ecx); break;
9822 case DBGFEVENT_INSTR_GETSEC: VBOXVMM_INSTR_GETSEC(pVCpu, pMixedCtx); break;
9823 case DBGFEVENT_INSTR_HALT: VBOXVMM_INSTR_HALT(pVCpu, pMixedCtx); break;
9824 case DBGFEVENT_INSTR_INVD: VBOXVMM_INSTR_INVD(pVCpu, pMixedCtx); break;
9825 case DBGFEVENT_INSTR_INVLPG: VBOXVMM_INSTR_INVLPG(pVCpu, pMixedCtx); break;
9826 case DBGFEVENT_INSTR_RDPMC: VBOXVMM_INSTR_RDPMC(pVCpu, pMixedCtx); break;
9827 case DBGFEVENT_INSTR_RDTSC: VBOXVMM_INSTR_RDTSC(pVCpu, pMixedCtx); break;
9828 case DBGFEVENT_INSTR_RSM: VBOXVMM_INSTR_RSM(pVCpu, pMixedCtx); break;
9829 case DBGFEVENT_INSTR_CRX_READ: VBOXVMM_INSTR_CRX_READ(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9830 case DBGFEVENT_INSTR_CRX_WRITE: VBOXVMM_INSTR_CRX_WRITE(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9831 case DBGFEVENT_INSTR_DRX_READ: VBOXVMM_INSTR_DRX_READ(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9832 case DBGFEVENT_INSTR_DRX_WRITE: VBOXVMM_INSTR_DRX_WRITE(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9833 case DBGFEVENT_INSTR_RDMSR: VBOXVMM_INSTR_RDMSR(pVCpu, pMixedCtx, pMixedCtx->ecx); break;
9834 case DBGFEVENT_INSTR_WRMSR: VBOXVMM_INSTR_WRMSR(pVCpu, pMixedCtx, pMixedCtx->ecx,
9835 RT_MAKE_U64(pMixedCtx->eax, pMixedCtx->edx)); break;
9836 case DBGFEVENT_INSTR_MWAIT: VBOXVMM_INSTR_MWAIT(pVCpu, pMixedCtx); break;
9837 case DBGFEVENT_INSTR_MONITOR: VBOXVMM_INSTR_MONITOR(pVCpu, pMixedCtx); break;
9838 case DBGFEVENT_INSTR_PAUSE: VBOXVMM_INSTR_PAUSE(pVCpu, pMixedCtx); break;
9839 case DBGFEVENT_INSTR_SGDT: VBOXVMM_INSTR_SGDT(pVCpu, pMixedCtx); break;
9840 case DBGFEVENT_INSTR_SIDT: VBOXVMM_INSTR_SIDT(pVCpu, pMixedCtx); break;
9841 case DBGFEVENT_INSTR_LGDT: VBOXVMM_INSTR_LGDT(pVCpu, pMixedCtx); break;
9842 case DBGFEVENT_INSTR_LIDT: VBOXVMM_INSTR_LIDT(pVCpu, pMixedCtx); break;
9843 case DBGFEVENT_INSTR_SLDT: VBOXVMM_INSTR_SLDT(pVCpu, pMixedCtx); break;
9844 case DBGFEVENT_INSTR_STR: VBOXVMM_INSTR_STR(pVCpu, pMixedCtx); break;
9845 case DBGFEVENT_INSTR_LLDT: VBOXVMM_INSTR_LLDT(pVCpu, pMixedCtx); break;
9846 case DBGFEVENT_INSTR_LTR: VBOXVMM_INSTR_LTR(pVCpu, pMixedCtx); break;
9847 case DBGFEVENT_INSTR_RDTSCP: VBOXVMM_INSTR_RDTSCP(pVCpu, pMixedCtx); break;
9848 case DBGFEVENT_INSTR_WBINVD: VBOXVMM_INSTR_WBINVD(pVCpu, pMixedCtx); break;
9849 case DBGFEVENT_INSTR_XSETBV: VBOXVMM_INSTR_XSETBV(pVCpu, pMixedCtx); break;
9850 case DBGFEVENT_INSTR_RDRAND: VBOXVMM_INSTR_RDRAND(pVCpu, pMixedCtx); break;
9851 case DBGFEVENT_INSTR_RDSEED: VBOXVMM_INSTR_RDSEED(pVCpu, pMixedCtx); break;
9852 case DBGFEVENT_INSTR_XSAVES: VBOXVMM_INSTR_XSAVES(pVCpu, pMixedCtx); break;
9853 case DBGFEVENT_INSTR_XRSTORS: VBOXVMM_INSTR_XRSTORS(pVCpu, pMixedCtx); break;
9854 case DBGFEVENT_INSTR_VMM_CALL: VBOXVMM_INSTR_VMM_CALL(pVCpu, pMixedCtx); break;
9855 case DBGFEVENT_INSTR_VMX_VMCLEAR: VBOXVMM_INSTR_VMX_VMCLEAR(pVCpu, pMixedCtx); break;
9856 case DBGFEVENT_INSTR_VMX_VMLAUNCH: VBOXVMM_INSTR_VMX_VMLAUNCH(pVCpu, pMixedCtx); break;
9857 case DBGFEVENT_INSTR_VMX_VMPTRLD: VBOXVMM_INSTR_VMX_VMPTRLD(pVCpu, pMixedCtx); break;
9858 case DBGFEVENT_INSTR_VMX_VMPTRST: VBOXVMM_INSTR_VMX_VMPTRST(pVCpu, pMixedCtx); break;
9859 case DBGFEVENT_INSTR_VMX_VMREAD: VBOXVMM_INSTR_VMX_VMREAD(pVCpu, pMixedCtx); break;
9860 case DBGFEVENT_INSTR_VMX_VMRESUME: VBOXVMM_INSTR_VMX_VMRESUME(pVCpu, pMixedCtx); break;
9861 case DBGFEVENT_INSTR_VMX_VMWRITE: VBOXVMM_INSTR_VMX_VMWRITE(pVCpu, pMixedCtx); break;
9862 case DBGFEVENT_INSTR_VMX_VMXOFF: VBOXVMM_INSTR_VMX_VMXOFF(pVCpu, pMixedCtx); break;
9863 case DBGFEVENT_INSTR_VMX_VMXON: VBOXVMM_INSTR_VMX_VMXON(pVCpu, pMixedCtx); break;
9864 case DBGFEVENT_INSTR_VMX_INVEPT: VBOXVMM_INSTR_VMX_INVEPT(pVCpu, pMixedCtx); break;
9865 case DBGFEVENT_INSTR_VMX_INVVPID: VBOXVMM_INSTR_VMX_INVVPID(pVCpu, pMixedCtx); break;
9866 case DBGFEVENT_INSTR_VMX_INVPCID: VBOXVMM_INSTR_VMX_INVPCID(pVCpu, pMixedCtx); break;
9867 case DBGFEVENT_INSTR_VMX_VMFUNC: VBOXVMM_INSTR_VMX_VMFUNC(pVCpu, pMixedCtx); break;
9868 default: AssertMsgFailed(("enmEvent1=%d uExitReason=%d\n", enmEvent1, uExitReason)); break;
9869 }
9870 switch (enmEvent2)
9871 {
9872 /** @todo consider which extra parameters would be helpful for each probe. */
9873 case DBGFEVENT_END: break;
9874 case DBGFEVENT_EXIT_TASK_SWITCH: VBOXVMM_EXIT_TASK_SWITCH(pVCpu, pMixedCtx); break;
9875 case DBGFEVENT_EXIT_CPUID: VBOXVMM_EXIT_CPUID(pVCpu, pMixedCtx, pMixedCtx->eax, pMixedCtx->ecx); break;
9876 case DBGFEVENT_EXIT_GETSEC: VBOXVMM_EXIT_GETSEC(pVCpu, pMixedCtx); break;
9877 case DBGFEVENT_EXIT_HALT: VBOXVMM_EXIT_HALT(pVCpu, pMixedCtx); break;
9878 case DBGFEVENT_EXIT_INVD: VBOXVMM_EXIT_INVD(pVCpu, pMixedCtx); break;
9879 case DBGFEVENT_EXIT_INVLPG: VBOXVMM_EXIT_INVLPG(pVCpu, pMixedCtx); break;
9880 case DBGFEVENT_EXIT_RDPMC: VBOXVMM_EXIT_RDPMC(pVCpu, pMixedCtx); break;
9881 case DBGFEVENT_EXIT_RDTSC: VBOXVMM_EXIT_RDTSC(pVCpu, pMixedCtx); break;
9882 case DBGFEVENT_EXIT_RSM: VBOXVMM_EXIT_RSM(pVCpu, pMixedCtx); break;
9883 case DBGFEVENT_EXIT_CRX_READ: VBOXVMM_EXIT_CRX_READ(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9884 case DBGFEVENT_EXIT_CRX_WRITE: VBOXVMM_EXIT_CRX_WRITE(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9885 case DBGFEVENT_EXIT_DRX_READ: VBOXVMM_EXIT_DRX_READ(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9886 case DBGFEVENT_EXIT_DRX_WRITE: VBOXVMM_EXIT_DRX_WRITE(pVCpu, pMixedCtx, (uint8_t)uEventArg); break;
9887 case DBGFEVENT_EXIT_RDMSR: VBOXVMM_EXIT_RDMSR(pVCpu, pMixedCtx, pMixedCtx->ecx); break;
9888 case DBGFEVENT_EXIT_WRMSR: VBOXVMM_EXIT_WRMSR(pVCpu, pMixedCtx, pMixedCtx->ecx,
9889 RT_MAKE_U64(pMixedCtx->eax, pMixedCtx->edx)); break;
9890 case DBGFEVENT_EXIT_MWAIT: VBOXVMM_EXIT_MWAIT(pVCpu, pMixedCtx); break;
9891 case DBGFEVENT_EXIT_MONITOR: VBOXVMM_EXIT_MONITOR(pVCpu, pMixedCtx); break;
9892 case DBGFEVENT_EXIT_PAUSE: VBOXVMM_EXIT_PAUSE(pVCpu, pMixedCtx); break;
9893 case DBGFEVENT_EXIT_SGDT: VBOXVMM_EXIT_SGDT(pVCpu, pMixedCtx); break;
9894 case DBGFEVENT_EXIT_SIDT: VBOXVMM_EXIT_SIDT(pVCpu, pMixedCtx); break;
9895 case DBGFEVENT_EXIT_LGDT: VBOXVMM_EXIT_LGDT(pVCpu, pMixedCtx); break;
9896 case DBGFEVENT_EXIT_LIDT: VBOXVMM_EXIT_LIDT(pVCpu, pMixedCtx); break;
9897 case DBGFEVENT_EXIT_SLDT: VBOXVMM_EXIT_SLDT(pVCpu, pMixedCtx); break;
9898 case DBGFEVENT_EXIT_STR: VBOXVMM_EXIT_STR(pVCpu, pMixedCtx); break;
9899 case DBGFEVENT_EXIT_LLDT: VBOXVMM_EXIT_LLDT(pVCpu, pMixedCtx); break;
9900 case DBGFEVENT_EXIT_LTR: VBOXVMM_EXIT_LTR(pVCpu, pMixedCtx); break;
9901 case DBGFEVENT_EXIT_RDTSCP: VBOXVMM_EXIT_RDTSCP(pVCpu, pMixedCtx); break;
9902 case DBGFEVENT_EXIT_WBINVD: VBOXVMM_EXIT_WBINVD(pVCpu, pMixedCtx); break;
9903 case DBGFEVENT_EXIT_XSETBV: VBOXVMM_EXIT_XSETBV(pVCpu, pMixedCtx); break;
9904 case DBGFEVENT_EXIT_RDRAND: VBOXVMM_EXIT_RDRAND(pVCpu, pMixedCtx); break;
9905 case DBGFEVENT_EXIT_RDSEED: VBOXVMM_EXIT_RDSEED(pVCpu, pMixedCtx); break;
9906 case DBGFEVENT_EXIT_XSAVES: VBOXVMM_EXIT_XSAVES(pVCpu, pMixedCtx); break;
9907 case DBGFEVENT_EXIT_XRSTORS: VBOXVMM_EXIT_XRSTORS(pVCpu, pMixedCtx); break;
9908 case DBGFEVENT_EXIT_VMM_CALL: VBOXVMM_EXIT_VMM_CALL(pVCpu, pMixedCtx); break;
9909 case DBGFEVENT_EXIT_VMX_VMCLEAR: VBOXVMM_EXIT_VMX_VMCLEAR(pVCpu, pMixedCtx); break;
9910 case DBGFEVENT_EXIT_VMX_VMLAUNCH: VBOXVMM_EXIT_VMX_VMLAUNCH(pVCpu, pMixedCtx); break;
9911 case DBGFEVENT_EXIT_VMX_VMPTRLD: VBOXVMM_EXIT_VMX_VMPTRLD(pVCpu, pMixedCtx); break;
9912 case DBGFEVENT_EXIT_VMX_VMPTRST: VBOXVMM_EXIT_VMX_VMPTRST(pVCpu, pMixedCtx); break;
9913 case DBGFEVENT_EXIT_VMX_VMREAD: VBOXVMM_EXIT_VMX_VMREAD(pVCpu, pMixedCtx); break;
9914 case DBGFEVENT_EXIT_VMX_VMRESUME: VBOXVMM_EXIT_VMX_VMRESUME(pVCpu, pMixedCtx); break;
9915 case DBGFEVENT_EXIT_VMX_VMWRITE: VBOXVMM_EXIT_VMX_VMWRITE(pVCpu, pMixedCtx); break;
9916 case DBGFEVENT_EXIT_VMX_VMXOFF: VBOXVMM_EXIT_VMX_VMXOFF(pVCpu, pMixedCtx); break;
9917 case DBGFEVENT_EXIT_VMX_VMXON: VBOXVMM_EXIT_VMX_VMXON(pVCpu, pMixedCtx); break;
9918 case DBGFEVENT_EXIT_VMX_INVEPT: VBOXVMM_EXIT_VMX_INVEPT(pVCpu, pMixedCtx); break;
9919 case DBGFEVENT_EXIT_VMX_INVVPID: VBOXVMM_EXIT_VMX_INVVPID(pVCpu, pMixedCtx); break;
9920 case DBGFEVENT_EXIT_VMX_INVPCID: VBOXVMM_EXIT_VMX_INVPCID(pVCpu, pMixedCtx); break;
9921 case DBGFEVENT_EXIT_VMX_VMFUNC: VBOXVMM_EXIT_VMX_VMFUNC(pVCpu, pMixedCtx); break;
9922 case DBGFEVENT_EXIT_VMX_EPT_MISCONFIG: VBOXVMM_EXIT_VMX_EPT_MISCONFIG(pVCpu, pMixedCtx); break;
9923 case DBGFEVENT_EXIT_VMX_EPT_VIOLATION: VBOXVMM_EXIT_VMX_EPT_VIOLATION(pVCpu, pMixedCtx); break;
9924 case DBGFEVENT_EXIT_VMX_VAPIC_ACCESS: VBOXVMM_EXIT_VMX_VAPIC_ACCESS(pVCpu, pMixedCtx); break;
9925 case DBGFEVENT_EXIT_VMX_VAPIC_WRITE: VBOXVMM_EXIT_VMX_VAPIC_WRITE(pVCpu, pMixedCtx); break;
9926 default: AssertMsgFailed(("enmEvent2=%d uExitReason=%d\n", enmEvent2, uExitReason)); break;
9927 }
9928 }
9929
9930 /*
9931 * Fire of the DBGF event, if enabled (our check here is just a quick one,
9932 * the DBGF call will do a full check).
9933 *
9934 * Note! DBGF sets DBGFEVENT_INTERRUPT_SOFTWARE in the bitmap.
9935 * Note! If we have to events, we prioritize the first, i.e. the instruction
9936 * one, in order to avoid event nesting.
9937 */
9938 if ( enmEvent1 != DBGFEVENT_END
9939 && DBGF_IS_EVENT_ENABLED(pVM, enmEvent1))
9940 {
9941 VBOXSTRICTRC rcStrict = DBGFEventGenericWithArg(pVM, pVCpu, enmEvent1, uEventArg, DBGFEVENTCTX_HM);
9942 if (rcStrict != VINF_SUCCESS)
9943 return rcStrict;
9944 }
9945 else if ( enmEvent2 != DBGFEVENT_END
9946 && DBGF_IS_EVENT_ENABLED(pVM, enmEvent2))
9947 {
9948 VBOXSTRICTRC rcStrict = DBGFEventGenericWithArg(pVM, pVCpu, enmEvent2, uEventArg, DBGFEVENTCTX_HM);
9949 if (rcStrict != VINF_SUCCESS)
9950 return rcStrict;
9951 }
9952
9953 return VINF_SUCCESS;
9954}
9955
9956
9957/**
9958 * Single-stepping VM-exit filtering.
9959 *
9960 * This is preprocessing the VM-exits and deciding whether we've gotten far
9961 * enough to return VINF_EM_DBG_STEPPED already. If not, normal VM-exit
9962 * handling is performed.
9963 *
9964 * @returns Strict VBox status code (i.e. informational status codes too).
9965 * @param pVM The cross context VM structure.
9966 * @param pVCpu The cross context virtual CPU structure of the calling EMT.
9967 * @param pMixedCtx Pointer to the guest-CPU context. The data may be
9968 * out-of-sync. Make sure to update the required
9969 * fields before using them.
9970 * @param pVmxTransient Pointer to the VMX-transient structure.
9971 * @param uExitReason The VM-exit reason.
9972 * @param pDbgState The debug state.
9973 */
9974DECLINLINE(VBOXSTRICTRC) hmR0VmxRunDebugHandleExit(PVM pVM, PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient,
9975 uint32_t uExitReason, PVMXRUNDBGSTATE pDbgState)
9976{
9977 /*
9978 * Expensive (saves context) generic dtrace VM-exit probe.
9979 */
9980 if (!VBOXVMM_R0_HMVMX_VMEXIT_ENABLED())
9981 { /* more likely */ }
9982 else
9983 {
9984 hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
9985 hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
9986 VBOXVMM_R0_HMVMX_VMEXIT(pVCpu, pMixedCtx, pVmxTransient->uExitReason, pVmxTransient->uExitQualification);
9987 }
9988
9989 /*
9990 * Check for host NMI, just to get that out of the way.
9991 */
9992 if (uExitReason != VMX_EXIT_XCPT_OR_NMI)
9993 { /* normally likely */ }
9994 else
9995 {
9996 int rc2 = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
9997 AssertRCReturn(rc2, rc2);
9998 uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVmxTransient->uExitIntInfo);
9999 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
10000 return hmR0VmxExitXcptOrNmi(pVCpu, pMixedCtx, pVmxTransient);
10001 }
10002
10003 /*
10004 * Check for single stepping event if we're stepping.
10005 */
10006 if (pVCpu->hm.s.fSingleInstruction)
10007 {
10008 switch (uExitReason)
10009 {
10010 case VMX_EXIT_MTF:
10011 return hmR0VmxExitMtf(pVCpu, pMixedCtx, pVmxTransient);
10012
10013 /* Various events: */
10014 case VMX_EXIT_XCPT_OR_NMI:
10015 case VMX_EXIT_EXT_INT:
10016 case VMX_EXIT_TRIPLE_FAULT:
10017 case VMX_EXIT_INT_WINDOW:
10018 case VMX_EXIT_NMI_WINDOW:
10019 case VMX_EXIT_TASK_SWITCH:
10020 case VMX_EXIT_TPR_BELOW_THRESHOLD:
10021 case VMX_EXIT_APIC_ACCESS:
10022 case VMX_EXIT_EPT_VIOLATION:
10023 case VMX_EXIT_EPT_MISCONFIG:
10024 case VMX_EXIT_PREEMPT_TIMER:
10025
10026 /* Instruction specific VM-exits: */
10027 case VMX_EXIT_CPUID:
10028 case VMX_EXIT_GETSEC:
10029 case VMX_EXIT_HLT:
10030 case VMX_EXIT_INVD:
10031 case VMX_EXIT_INVLPG:
10032 case VMX_EXIT_RDPMC:
10033 case VMX_EXIT_RDTSC:
10034 case VMX_EXIT_RSM:
10035 case VMX_EXIT_VMCALL:
10036 case VMX_EXIT_VMCLEAR:
10037 case VMX_EXIT_VMLAUNCH:
10038 case VMX_EXIT_VMPTRLD:
10039 case VMX_EXIT_VMPTRST:
10040 case VMX_EXIT_VMREAD:
10041 case VMX_EXIT_VMRESUME:
10042 case VMX_EXIT_VMWRITE:
10043 case VMX_EXIT_VMXOFF:
10044 case VMX_EXIT_VMXON:
10045 case VMX_EXIT_MOV_CRX:
10046 case VMX_EXIT_MOV_DRX:
10047 case VMX_EXIT_IO_INSTR:
10048 case VMX_EXIT_RDMSR:
10049 case VMX_EXIT_WRMSR:
10050 case VMX_EXIT_MWAIT:
10051 case VMX_EXIT_MONITOR:
10052 case VMX_EXIT_PAUSE:
10053 case VMX_EXIT_XDTR_ACCESS:
10054 case VMX_EXIT_TR_ACCESS:
10055 case VMX_EXIT_INVEPT:
10056 case VMX_EXIT_RDTSCP:
10057 case VMX_EXIT_INVVPID:
10058 case VMX_EXIT_WBINVD:
10059 case VMX_EXIT_XSETBV:
10060 case VMX_EXIT_RDRAND:
10061 case VMX_EXIT_INVPCID:
10062 case VMX_EXIT_VMFUNC:
10063 case VMX_EXIT_RDSEED:
10064 case VMX_EXIT_XSAVES:
10065 case VMX_EXIT_XRSTORS:
10066 {
10067 int rc2 = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
10068 rc2 |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
10069 AssertRCReturn(rc2, rc2);
10070 if ( pMixedCtx->rip != pDbgState->uRipStart
10071 || pMixedCtx->cs.Sel != pDbgState->uCsStart)
10072 return VINF_EM_DBG_STEPPED;
10073 break;
10074 }
10075
10076 /* Errors and unexpected events: */
10077 case VMX_EXIT_INIT_SIGNAL:
10078 case VMX_EXIT_SIPI:
10079 case VMX_EXIT_IO_SMI:
10080 case VMX_EXIT_SMI:
10081 case VMX_EXIT_ERR_INVALID_GUEST_STATE:
10082 case VMX_EXIT_ERR_MSR_LOAD:
10083 case VMX_EXIT_ERR_MACHINE_CHECK:
10084 case VMX_EXIT_APIC_WRITE: /* Some talk about this being fault like, so I guess we must process it? */
10085 break;
10086
10087 default:
10088 AssertMsgFailed(("Unexpected VM-exit=%#x\n", uExitReason));
10089 break;
10090 }
10091 }
10092
10093 /*
10094 * Check for debugger event breakpoints and dtrace probes.
10095 */
10096 if ( uExitReason < RT_ELEMENTS(pDbgState->bmExitsToCheck) * 32U
10097 && ASMBitTest(pDbgState->bmExitsToCheck, uExitReason) )
10098 {
10099 VBOXSTRICTRC rcStrict = hmR0VmxHandleExitDtraceEvents(pVM, pVCpu, pMixedCtx, pVmxTransient, uExitReason);
10100 if (rcStrict != VINF_SUCCESS)
10101 return rcStrict;
10102 }
10103
10104 /*
10105 * Normal processing.
10106 */
10107#ifdef HMVMX_USE_FUNCTION_TABLE
10108 return g_apfnVMExitHandlers[uExitReason](pVCpu, pMixedCtx, pVmxTransient);
10109#else
10110 return hmR0VmxHandleExit(pVCpu, pMixedCtx, pVmxTransient, uExitReason);
10111#endif
10112}
10113
10114
10115/**
10116 * Single steps guest code using VT-x.
10117 *
10118 * @returns Strict VBox status code (i.e. informational status codes too).
10119 * @param pVM The cross context VM structure.
10120 * @param pVCpu The cross context virtual CPU structure.
10121 * @param pCtx Pointer to the guest-CPU context.
10122 *
10123 * @note Mostly the same as hmR0VmxRunGuestCodeNormal().
10124 */
10125static VBOXSTRICTRC hmR0VmxRunGuestCodeDebug(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
10126{
10127 VMXTRANSIENT VmxTransient;
10128 VmxTransient.fUpdateTscOffsettingAndPreemptTimer = true;
10129
10130 /* Set HMCPU indicators. */
10131 bool const fSavedSingleInstruction = pVCpu->hm.s.fSingleInstruction;
10132 pVCpu->hm.s.fSingleInstruction = pVCpu->hm.s.fSingleInstruction || DBGFIsStepping(pVCpu);
10133 pVCpu->hm.s.fDebugWantRdTscExit = false;
10134 pVCpu->hm.s.fUsingDebugLoop = true;
10135
10136 /* State we keep to help modify and later restore the VMCS fields we alter, and for detecting steps. */
10137 VMXRUNDBGSTATE DbgState;
10138 hmR0VmxRunDebugStateInit(pVCpu, pCtx, &DbgState);
10139 hmR0VmxPreRunGuestDebugStateUpdate(pVM, pVCpu, pCtx, &DbgState, &VmxTransient);
10140
10141 /*
10142 * The loop.
10143 */
10144 VBOXSTRICTRC rcStrict = VERR_INTERNAL_ERROR_5;
10145 for (uint32_t cLoops = 0; ; cLoops++)
10146 {
10147 Assert(!HMR0SuspendPending());
10148 HMVMX_ASSERT_CPU_SAFE();
10149 bool fStepping = pVCpu->hm.s.fSingleInstruction;
10150
10151 /*
10152 * Preparatory work for running guest code, this may force us to return
10153 * to ring-3. This bugger disables interrupts on VINF_SUCCESS!
10154 */
10155 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatEntry, x);
10156 hmR0VmxPreRunGuestDebugStateApply(pVCpu, &DbgState); /* Set up execute controls the next to can respond to. */
10157 rcStrict = hmR0VmxPreRunGuest(pVM, pVCpu, pCtx, &VmxTransient, fStepping);
10158 if (rcStrict != VINF_SUCCESS)
10159 break;
10160
10161 hmR0VmxPreRunGuestCommitted(pVM, pVCpu, pCtx, &VmxTransient);
10162 hmR0VmxPreRunGuestDebugStateApply(pVCpu, &DbgState); /* Override any obnoxious code in the above two calls. */
10163
10164 /*
10165 * Now we can run the guest code.
10166 */
10167 int rcRun = hmR0VmxRunGuest(pVM, pVCpu, pCtx);
10168
10169 /* The guest-CPU context is now outdated, 'pCtx' is to be treated as 'pMixedCtx' from this point on!!! */
10170
10171 /*
10172 * Restore any residual host-state and save any bits shared between host
10173 * and guest into the guest-CPU state. Re-enables interrupts!
10174 */
10175 hmR0VmxPostRunGuest(pVM, pVCpu, pCtx, &VmxTransient, rcRun);
10176
10177 /* Check for errors with running the VM (VMLAUNCH/VMRESUME). */
10178 if (RT_SUCCESS(rcRun))
10179 { /* very likely */ }
10180 else
10181 {
10182 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit1, x);
10183 hmR0VmxReportWorldSwitchError(pVM, pVCpu, rcRun, pCtx, &VmxTransient);
10184 return rcRun;
10185 }
10186
10187 /* Profile the VM-exit. */
10188 AssertMsg(VmxTransient.uExitReason <= VMX_EXIT_MAX, ("%#x\n", VmxTransient.uExitReason));
10189 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitAll);
10190 STAM_COUNTER_INC(&pVCpu->hm.s.paStatExitReasonR0[VmxTransient.uExitReason & MASK_EXITREASON_STAT]);
10191 STAM_PROFILE_ADV_STOP_START(&pVCpu->hm.s.StatExit1, &pVCpu->hm.s.StatExit2, x);
10192 HMVMX_START_EXIT_DISPATCH_PROF();
10193
10194 VBOXVMM_R0_HMVMX_VMEXIT_NOCTX(pVCpu, pCtx, VmxTransient.uExitReason);
10195
10196 /*
10197 * Handle the VM-exit - we quit earlier on certain VM-exits, see hmR0VmxHandleExitDebug().
10198 */
10199 rcStrict = hmR0VmxRunDebugHandleExit(pVM, pVCpu, pCtx, &VmxTransient, VmxTransient.uExitReason, &DbgState);
10200 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExit2, x);
10201 if (rcStrict != VINF_SUCCESS)
10202 break;
10203 if (cLoops > pVM->hm.s.cMaxResumeLoops)
10204 {
10205 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchMaxResumeLoops);
10206 rcStrict = VINF_EM_RAW_INTERRUPT;
10207 break;
10208 }
10209
10210 /*
10211 * Stepping: Did the RIP change, if so, consider it a single step.
10212 * Otherwise, make sure one of the TFs gets set.
10213 */
10214 if (fStepping)
10215 {
10216 int rc2 = hmR0VmxSaveGuestRip(pVCpu, pCtx);
10217 rc2 |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pCtx);
10218 AssertRCReturn(rc2, rc2);
10219 if ( pCtx->rip != DbgState.uRipStart
10220 || pCtx->cs.Sel != DbgState.uCsStart)
10221 {
10222 rcStrict = VINF_EM_DBG_STEPPED;
10223 break;
10224 }
10225 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
10226 }
10227
10228 /*
10229 * Update when dtrace settings changes (DBGF kicks us, so no need to check).
10230 */
10231 if (VBOXVMM_GET_SETTINGS_SEQ_NO() != DbgState.uDtraceSettingsSeqNo)
10232 hmR0VmxPreRunGuestDebugStateUpdate(pVM, pVCpu, pCtx, &DbgState, &VmxTransient);
10233 }
10234
10235 /*
10236 * Clear the X86_EFL_TF if necessary.
10237 */
10238 if (pVCpu->hm.s.fClearTrapFlag)
10239 {
10240 int rc2 = hmR0VmxSaveGuestRflags(pVCpu, pCtx);
10241 AssertRCReturn(rc2, rc2);
10242 pVCpu->hm.s.fClearTrapFlag = false;
10243 pCtx->eflags.Bits.u1TF = 0;
10244 }
10245 /** @todo there seems to be issues with the resume flag when the monitor trap
10246 * flag is pending without being used. Seen early in bios init when
10247 * accessing APIC page in protected mode. */
10248
10249 /*
10250 * Restore VM-exit control settings as we may not reenter this function the
10251 * next time around.
10252 */
10253 rcStrict = hmR0VmxRunDebugStateRevert(pVCpu, &DbgState, rcStrict);
10254
10255 /* Restore HMCPU indicators. */
10256 pVCpu->hm.s.fUsingDebugLoop = false;
10257 pVCpu->hm.s.fDebugWantRdTscExit = false;
10258 pVCpu->hm.s.fSingleInstruction = fSavedSingleInstruction;
10259
10260 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatEntry, x);
10261 return rcStrict;
10262}
10263
10264
10265/** @} */
10266
10267
10268/**
10269 * Checks if any expensive dtrace probes are enabled and we should go to the
10270 * debug loop.
10271 *
10272 * @returns true if we should use debug loop, false if not.
10273 */
10274static bool hmR0VmxAnyExpensiveProbesEnabled(void)
10275{
10276 /* It's probably faster to OR the raw 32-bit counter variables together.
10277 Since the variables are in an array and the probes are next to one
10278 another (more or less), we have good locality. So, better read
10279 eight-nine cache lines ever time and only have one conditional, than
10280 128+ conditionals, right? */
10281 return ( VBOXVMM_R0_HMVMX_VMEXIT_ENABLED_RAW() /* expensive too due to context */
10282 | VBOXVMM_XCPT_DE_ENABLED_RAW()
10283 | VBOXVMM_XCPT_DB_ENABLED_RAW()
10284 | VBOXVMM_XCPT_BP_ENABLED_RAW()
10285 | VBOXVMM_XCPT_OF_ENABLED_RAW()
10286 | VBOXVMM_XCPT_BR_ENABLED_RAW()
10287 | VBOXVMM_XCPT_UD_ENABLED_RAW()
10288 | VBOXVMM_XCPT_NM_ENABLED_RAW()
10289 | VBOXVMM_XCPT_DF_ENABLED_RAW()
10290 | VBOXVMM_XCPT_TS_ENABLED_RAW()
10291 | VBOXVMM_XCPT_NP_ENABLED_RAW()
10292 | VBOXVMM_XCPT_SS_ENABLED_RAW()
10293 | VBOXVMM_XCPT_GP_ENABLED_RAW()
10294 | VBOXVMM_XCPT_PF_ENABLED_RAW()
10295 | VBOXVMM_XCPT_MF_ENABLED_RAW()
10296 | VBOXVMM_XCPT_AC_ENABLED_RAW()
10297 | VBOXVMM_XCPT_XF_ENABLED_RAW()
10298 | VBOXVMM_XCPT_VE_ENABLED_RAW()
10299 | VBOXVMM_XCPT_SX_ENABLED_RAW()
10300 | VBOXVMM_INT_SOFTWARE_ENABLED_RAW()
10301 | VBOXVMM_INT_HARDWARE_ENABLED_RAW()
10302 ) != 0
10303 || ( VBOXVMM_INSTR_HALT_ENABLED_RAW()
10304 | VBOXVMM_INSTR_MWAIT_ENABLED_RAW()
10305 | VBOXVMM_INSTR_MONITOR_ENABLED_RAW()
10306 | VBOXVMM_INSTR_CPUID_ENABLED_RAW()
10307 | VBOXVMM_INSTR_INVD_ENABLED_RAW()
10308 | VBOXVMM_INSTR_WBINVD_ENABLED_RAW()
10309 | VBOXVMM_INSTR_INVLPG_ENABLED_RAW()
10310 | VBOXVMM_INSTR_RDTSC_ENABLED_RAW()
10311 | VBOXVMM_INSTR_RDTSCP_ENABLED_RAW()
10312 | VBOXVMM_INSTR_RDPMC_ENABLED_RAW()
10313 | VBOXVMM_INSTR_RDMSR_ENABLED_RAW()
10314 | VBOXVMM_INSTR_WRMSR_ENABLED_RAW()
10315 | VBOXVMM_INSTR_CRX_READ_ENABLED_RAW()
10316 | VBOXVMM_INSTR_CRX_WRITE_ENABLED_RAW()
10317 | VBOXVMM_INSTR_DRX_READ_ENABLED_RAW()
10318 | VBOXVMM_INSTR_DRX_WRITE_ENABLED_RAW()
10319 | VBOXVMM_INSTR_PAUSE_ENABLED_RAW()
10320 | VBOXVMM_INSTR_XSETBV_ENABLED_RAW()
10321 | VBOXVMM_INSTR_SIDT_ENABLED_RAW()
10322 | VBOXVMM_INSTR_LIDT_ENABLED_RAW()
10323 | VBOXVMM_INSTR_SGDT_ENABLED_RAW()
10324 | VBOXVMM_INSTR_LGDT_ENABLED_RAW()
10325 | VBOXVMM_INSTR_SLDT_ENABLED_RAW()
10326 | VBOXVMM_INSTR_LLDT_ENABLED_RAW()
10327 | VBOXVMM_INSTR_STR_ENABLED_RAW()
10328 | VBOXVMM_INSTR_LTR_ENABLED_RAW()
10329 | VBOXVMM_INSTR_GETSEC_ENABLED_RAW()
10330 | VBOXVMM_INSTR_RSM_ENABLED_RAW()
10331 | VBOXVMM_INSTR_RDRAND_ENABLED_RAW()
10332 | VBOXVMM_INSTR_RDSEED_ENABLED_RAW()
10333 | VBOXVMM_INSTR_XSAVES_ENABLED_RAW()
10334 | VBOXVMM_INSTR_XRSTORS_ENABLED_RAW()
10335 | VBOXVMM_INSTR_VMM_CALL_ENABLED_RAW()
10336 | VBOXVMM_INSTR_VMX_VMCLEAR_ENABLED_RAW()
10337 | VBOXVMM_INSTR_VMX_VMLAUNCH_ENABLED_RAW()
10338 | VBOXVMM_INSTR_VMX_VMPTRLD_ENABLED_RAW()
10339 | VBOXVMM_INSTR_VMX_VMPTRST_ENABLED_RAW()
10340 | VBOXVMM_INSTR_VMX_VMREAD_ENABLED_RAW()
10341 | VBOXVMM_INSTR_VMX_VMRESUME_ENABLED_RAW()
10342 | VBOXVMM_INSTR_VMX_VMWRITE_ENABLED_RAW()
10343 | VBOXVMM_INSTR_VMX_VMXOFF_ENABLED_RAW()
10344 | VBOXVMM_INSTR_VMX_VMXON_ENABLED_RAW()
10345 | VBOXVMM_INSTR_VMX_VMFUNC_ENABLED_RAW()
10346 | VBOXVMM_INSTR_VMX_INVEPT_ENABLED_RAW()
10347 | VBOXVMM_INSTR_VMX_INVVPID_ENABLED_RAW()
10348 | VBOXVMM_INSTR_VMX_INVPCID_ENABLED_RAW()
10349 ) != 0
10350 || ( VBOXVMM_EXIT_TASK_SWITCH_ENABLED_RAW()
10351 | VBOXVMM_EXIT_HALT_ENABLED_RAW()
10352 | VBOXVMM_EXIT_MWAIT_ENABLED_RAW()
10353 | VBOXVMM_EXIT_MONITOR_ENABLED_RAW()
10354 | VBOXVMM_EXIT_CPUID_ENABLED_RAW()
10355 | VBOXVMM_EXIT_INVD_ENABLED_RAW()
10356 | VBOXVMM_EXIT_WBINVD_ENABLED_RAW()
10357 | VBOXVMM_EXIT_INVLPG_ENABLED_RAW()
10358 | VBOXVMM_EXIT_RDTSC_ENABLED_RAW()
10359 | VBOXVMM_EXIT_RDTSCP_ENABLED_RAW()
10360 | VBOXVMM_EXIT_RDPMC_ENABLED_RAW()
10361 | VBOXVMM_EXIT_RDMSR_ENABLED_RAW()
10362 | VBOXVMM_EXIT_WRMSR_ENABLED_RAW()
10363 | VBOXVMM_EXIT_CRX_READ_ENABLED_RAW()
10364 | VBOXVMM_EXIT_CRX_WRITE_ENABLED_RAW()
10365 | VBOXVMM_EXIT_DRX_READ_ENABLED_RAW()
10366 | VBOXVMM_EXIT_DRX_WRITE_ENABLED_RAW()
10367 | VBOXVMM_EXIT_PAUSE_ENABLED_RAW()
10368 | VBOXVMM_EXIT_XSETBV_ENABLED_RAW()
10369 | VBOXVMM_EXIT_SIDT_ENABLED_RAW()
10370 | VBOXVMM_EXIT_LIDT_ENABLED_RAW()
10371 | VBOXVMM_EXIT_SGDT_ENABLED_RAW()
10372 | VBOXVMM_EXIT_LGDT_ENABLED_RAW()
10373 | VBOXVMM_EXIT_SLDT_ENABLED_RAW()
10374 | VBOXVMM_EXIT_LLDT_ENABLED_RAW()
10375 | VBOXVMM_EXIT_STR_ENABLED_RAW()
10376 | VBOXVMM_EXIT_LTR_ENABLED_RAW()
10377 | VBOXVMM_EXIT_GETSEC_ENABLED_RAW()
10378 | VBOXVMM_EXIT_RSM_ENABLED_RAW()
10379 | VBOXVMM_EXIT_RDRAND_ENABLED_RAW()
10380 | VBOXVMM_EXIT_RDSEED_ENABLED_RAW()
10381 | VBOXVMM_EXIT_XSAVES_ENABLED_RAW()
10382 | VBOXVMM_EXIT_XRSTORS_ENABLED_RAW()
10383 | VBOXVMM_EXIT_VMM_CALL_ENABLED_RAW()
10384 | VBOXVMM_EXIT_VMX_VMCLEAR_ENABLED_RAW()
10385 | VBOXVMM_EXIT_VMX_VMLAUNCH_ENABLED_RAW()
10386 | VBOXVMM_EXIT_VMX_VMPTRLD_ENABLED_RAW()
10387 | VBOXVMM_EXIT_VMX_VMPTRST_ENABLED_RAW()
10388 | VBOXVMM_EXIT_VMX_VMREAD_ENABLED_RAW()
10389 | VBOXVMM_EXIT_VMX_VMRESUME_ENABLED_RAW()
10390 | VBOXVMM_EXIT_VMX_VMWRITE_ENABLED_RAW()
10391 | VBOXVMM_EXIT_VMX_VMXOFF_ENABLED_RAW()
10392 | VBOXVMM_EXIT_VMX_VMXON_ENABLED_RAW()
10393 | VBOXVMM_EXIT_VMX_VMFUNC_ENABLED_RAW()
10394 | VBOXVMM_EXIT_VMX_INVEPT_ENABLED_RAW()
10395 | VBOXVMM_EXIT_VMX_INVVPID_ENABLED_RAW()
10396 | VBOXVMM_EXIT_VMX_INVPCID_ENABLED_RAW()
10397 | VBOXVMM_EXIT_VMX_EPT_VIOLATION_ENABLED_RAW()
10398 | VBOXVMM_EXIT_VMX_EPT_MISCONFIG_ENABLED_RAW()
10399 | VBOXVMM_EXIT_VMX_VAPIC_ACCESS_ENABLED_RAW()
10400 | VBOXVMM_EXIT_VMX_VAPIC_WRITE_ENABLED_RAW()
10401 ) != 0;
10402}
10403
10404
10405/**
10406 * Runs the guest code using VT-x.
10407 *
10408 * @returns Strict VBox status code (i.e. informational status codes too).
10409 * @param pVM The cross context VM structure.
10410 * @param pVCpu The cross context virtual CPU structure.
10411 * @param pCtx Pointer to the guest-CPU context.
10412 */
10413VMMR0DECL(VBOXSTRICTRC) VMXR0RunGuestCode(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
10414{
10415 Assert(VMMRZCallRing3IsEnabled(pVCpu));
10416 Assert(HMVMXCPU_GST_VALUE(pVCpu) == HMVMX_UPDATED_GUEST_ALL);
10417 HMVMX_ASSERT_PREEMPT_SAFE();
10418
10419 VMMRZCallRing3SetNotification(pVCpu, hmR0VmxCallRing3Callback, pCtx);
10420
10421 VBOXSTRICTRC rcStrict;
10422 if ( !pVCpu->hm.s.fUseDebugLoop
10423 && (!VBOXVMM_ANY_PROBES_ENABLED() || !hmR0VmxAnyExpensiveProbesEnabled())
10424 && !DBGFIsStepping(pVCpu)
10425 && !pVM->dbgf.ro.cEnabledInt3Breakpoints)
10426 rcStrict = hmR0VmxRunGuestCodeNormal(pVM, pVCpu, pCtx);
10427 else
10428 rcStrict = hmR0VmxRunGuestCodeDebug(pVM, pVCpu, pCtx);
10429
10430 if (rcStrict == VERR_EM_INTERPRETER)
10431 rcStrict = VINF_EM_RAW_EMULATE_INSTR;
10432 else if (rcStrict == VINF_EM_RESET)
10433 rcStrict = VINF_EM_TRIPLE_FAULT;
10434
10435 int rc2 = hmR0VmxExitToRing3(pVM, pVCpu, pCtx, rcStrict);
10436 if (RT_FAILURE(rc2))
10437 {
10438 pVCpu->hm.s.u32HMError = (uint32_t)VBOXSTRICTRC_VAL(rcStrict);
10439 rcStrict = rc2;
10440 }
10441 Assert(!VMMRZCallRing3IsNotificationSet(pVCpu));
10442 return rcStrict;
10443}
10444
10445
10446#ifndef HMVMX_USE_FUNCTION_TABLE
10447DECLINLINE(VBOXSTRICTRC) hmR0VmxHandleExit(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient, uint32_t rcReason)
10448{
10449# ifdef DEBUG_ramshankar
10450# define RETURN_EXIT_CALL(a_CallExpr) \
10451 do { \
10452 int rc2 = hmR0VmxSaveGuestState(pVCpu, pMixedCtx); AssertRC(rc2); \
10453 VBOXSTRICTRC rcStrict = a_CallExpr; \
10454 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST); \
10455 return rcStrict; \
10456 } while (0)
10457# else
10458# define RETURN_EXIT_CALL(a_CallExpr) return a_CallExpr
10459# endif
10460 switch (rcReason)
10461 {
10462 case VMX_EXIT_EPT_MISCONFIG: RETURN_EXIT_CALL(hmR0VmxExitEptMisconfig(pVCpu, pMixedCtx, pVmxTransient));
10463 case VMX_EXIT_EPT_VIOLATION: RETURN_EXIT_CALL(hmR0VmxExitEptViolation(pVCpu, pMixedCtx, pVmxTransient));
10464 case VMX_EXIT_IO_INSTR: RETURN_EXIT_CALL(hmR0VmxExitIoInstr(pVCpu, pMixedCtx, pVmxTransient));
10465 case VMX_EXIT_CPUID: RETURN_EXIT_CALL(hmR0VmxExitCpuid(pVCpu, pMixedCtx, pVmxTransient));
10466 case VMX_EXIT_RDTSC: RETURN_EXIT_CALL(hmR0VmxExitRdtsc(pVCpu, pMixedCtx, pVmxTransient));
10467 case VMX_EXIT_RDTSCP: RETURN_EXIT_CALL(hmR0VmxExitRdtscp(pVCpu, pMixedCtx, pVmxTransient));
10468 case VMX_EXIT_APIC_ACCESS: RETURN_EXIT_CALL(hmR0VmxExitApicAccess(pVCpu, pMixedCtx, pVmxTransient));
10469 case VMX_EXIT_XCPT_OR_NMI: RETURN_EXIT_CALL(hmR0VmxExitXcptOrNmi(pVCpu, pMixedCtx, pVmxTransient));
10470 case VMX_EXIT_MOV_CRX: RETURN_EXIT_CALL(hmR0VmxExitMovCRx(pVCpu, pMixedCtx, pVmxTransient));
10471 case VMX_EXIT_EXT_INT: RETURN_EXIT_CALL(hmR0VmxExitExtInt(pVCpu, pMixedCtx, pVmxTransient));
10472 case VMX_EXIT_INT_WINDOW: RETURN_EXIT_CALL(hmR0VmxExitIntWindow(pVCpu, pMixedCtx, pVmxTransient));
10473 case VMX_EXIT_MWAIT: RETURN_EXIT_CALL(hmR0VmxExitMwait(pVCpu, pMixedCtx, pVmxTransient));
10474 case VMX_EXIT_MONITOR: RETURN_EXIT_CALL(hmR0VmxExitMonitor(pVCpu, pMixedCtx, pVmxTransient));
10475 case VMX_EXIT_TASK_SWITCH: RETURN_EXIT_CALL(hmR0VmxExitTaskSwitch(pVCpu, pMixedCtx, pVmxTransient));
10476 case VMX_EXIT_PREEMPT_TIMER: RETURN_EXIT_CALL(hmR0VmxExitPreemptTimer(pVCpu, pMixedCtx, pVmxTransient));
10477 case VMX_EXIT_RDMSR: RETURN_EXIT_CALL(hmR0VmxExitRdmsr(pVCpu, pMixedCtx, pVmxTransient));
10478 case VMX_EXIT_WRMSR: RETURN_EXIT_CALL(hmR0VmxExitWrmsr(pVCpu, pMixedCtx, pVmxTransient));
10479 case VMX_EXIT_MOV_DRX: RETURN_EXIT_CALL(hmR0VmxExitMovDRx(pVCpu, pMixedCtx, pVmxTransient));
10480 case VMX_EXIT_TPR_BELOW_THRESHOLD: RETURN_EXIT_CALL(hmR0VmxExitTprBelowThreshold(pVCpu, pMixedCtx, pVmxTransient));
10481 case VMX_EXIT_HLT: RETURN_EXIT_CALL(hmR0VmxExitHlt(pVCpu, pMixedCtx, pVmxTransient));
10482 case VMX_EXIT_INVD: RETURN_EXIT_CALL(hmR0VmxExitInvd(pVCpu, pMixedCtx, pVmxTransient));
10483 case VMX_EXIT_INVLPG: RETURN_EXIT_CALL(hmR0VmxExitInvlpg(pVCpu, pMixedCtx, pVmxTransient));
10484 case VMX_EXIT_RSM: RETURN_EXIT_CALL(hmR0VmxExitRsm(pVCpu, pMixedCtx, pVmxTransient));
10485 case VMX_EXIT_MTF: RETURN_EXIT_CALL(hmR0VmxExitMtf(pVCpu, pMixedCtx, pVmxTransient));
10486 case VMX_EXIT_PAUSE: RETURN_EXIT_CALL(hmR0VmxExitPause(pVCpu, pMixedCtx, pVmxTransient));
10487 case VMX_EXIT_XDTR_ACCESS: RETURN_EXIT_CALL(hmR0VmxExitXdtrAccess(pVCpu, pMixedCtx, pVmxTransient));
10488 case VMX_EXIT_TR_ACCESS: RETURN_EXIT_CALL(hmR0VmxExitXdtrAccess(pVCpu, pMixedCtx, pVmxTransient));
10489 case VMX_EXIT_WBINVD: RETURN_EXIT_CALL(hmR0VmxExitWbinvd(pVCpu, pMixedCtx, pVmxTransient));
10490 case VMX_EXIT_XSETBV: RETURN_EXIT_CALL(hmR0VmxExitXsetbv(pVCpu, pMixedCtx, pVmxTransient));
10491 case VMX_EXIT_RDRAND: RETURN_EXIT_CALL(hmR0VmxExitRdrand(pVCpu, pMixedCtx, pVmxTransient));
10492 case VMX_EXIT_INVPCID: RETURN_EXIT_CALL(hmR0VmxExitInvpcid(pVCpu, pMixedCtx, pVmxTransient));
10493 case VMX_EXIT_GETSEC: RETURN_EXIT_CALL(hmR0VmxExitGetsec(pVCpu, pMixedCtx, pVmxTransient));
10494 case VMX_EXIT_RDPMC: RETURN_EXIT_CALL(hmR0VmxExitRdpmc(pVCpu, pMixedCtx, pVmxTransient));
10495 case VMX_EXIT_VMCALL: RETURN_EXIT_CALL(hmR0VmxExitVmcall(pVCpu, pMixedCtx, pVmxTransient));
10496
10497 case VMX_EXIT_TRIPLE_FAULT: return hmR0VmxExitTripleFault(pVCpu, pMixedCtx, pVmxTransient);
10498 case VMX_EXIT_NMI_WINDOW: return hmR0VmxExitNmiWindow(pVCpu, pMixedCtx, pVmxTransient);
10499 case VMX_EXIT_INIT_SIGNAL: return hmR0VmxExitInitSignal(pVCpu, pMixedCtx, pVmxTransient);
10500 case VMX_EXIT_SIPI: return hmR0VmxExitSipi(pVCpu, pMixedCtx, pVmxTransient);
10501 case VMX_EXIT_IO_SMI: return hmR0VmxExitIoSmi(pVCpu, pMixedCtx, pVmxTransient);
10502 case VMX_EXIT_SMI: return hmR0VmxExitSmi(pVCpu, pMixedCtx, pVmxTransient);
10503 case VMX_EXIT_ERR_MSR_LOAD: return hmR0VmxExitErrMsrLoad(pVCpu, pMixedCtx, pVmxTransient);
10504 case VMX_EXIT_ERR_INVALID_GUEST_STATE: return hmR0VmxExitErrInvalidGuestState(pVCpu, pMixedCtx, pVmxTransient);
10505 case VMX_EXIT_ERR_MACHINE_CHECK: return hmR0VmxExitErrMachineCheck(pVCpu, pMixedCtx, pVmxTransient);
10506
10507 case VMX_EXIT_VMCLEAR:
10508 case VMX_EXIT_VMLAUNCH:
10509 case VMX_EXIT_VMPTRLD:
10510 case VMX_EXIT_VMPTRST:
10511 case VMX_EXIT_VMREAD:
10512 case VMX_EXIT_VMRESUME:
10513 case VMX_EXIT_VMWRITE:
10514 case VMX_EXIT_VMXOFF:
10515 case VMX_EXIT_VMXON:
10516 case VMX_EXIT_INVEPT:
10517 case VMX_EXIT_INVVPID:
10518 case VMX_EXIT_VMFUNC:
10519 case VMX_EXIT_XSAVES:
10520 case VMX_EXIT_XRSTORS:
10521 return hmR0VmxExitSetPendingXcptUD(pVCpu, pMixedCtx, pVmxTransient);
10522 case VMX_EXIT_ENCLS:
10523 case VMX_EXIT_RDSEED: /* only spurious VM-exits, so undefined */
10524 case VMX_EXIT_PML_FULL:
10525 default:
10526 return hmR0VmxExitErrUndefined(pVCpu, pMixedCtx, pVmxTransient);
10527 }
10528#undef RETURN_EXIT_CALL
10529}
10530#endif /* !HMVMX_USE_FUNCTION_TABLE */
10531
10532
10533#ifdef VBOX_STRICT
10534/* Is there some generic IPRT define for this that are not in Runtime/internal/\* ?? */
10535# define HMVMX_ASSERT_PREEMPT_CPUID_VAR() \
10536 RTCPUID const idAssertCpu = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId()
10537
10538# define HMVMX_ASSERT_PREEMPT_CPUID() \
10539 do { \
10540 RTCPUID const idAssertCpuNow = RTThreadPreemptIsEnabled(NIL_RTTHREAD) ? NIL_RTCPUID : RTMpCpuId(); \
10541 AssertMsg(idAssertCpu == idAssertCpuNow, ("VMX %#x, %#x\n", idAssertCpu, idAssertCpuNow)); \
10542 } while (0)
10543
10544# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS() \
10545 do { \
10546 AssertPtr(pVCpu); \
10547 AssertPtr(pMixedCtx); \
10548 AssertPtr(pVmxTransient); \
10549 Assert(pVmxTransient->fVMEntryFailed == false); \
10550 Assert(ASMIntAreEnabled()); \
10551 HMVMX_ASSERT_PREEMPT_SAFE(); \
10552 HMVMX_ASSERT_PREEMPT_CPUID_VAR(); \
10553 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)); \
10554 HMVMX_ASSERT_PREEMPT_SAFE(); \
10555 if (VMMR0IsLogFlushDisabled(pVCpu)) \
10556 HMVMX_ASSERT_PREEMPT_CPUID(); \
10557 HMVMX_STOP_EXIT_DISPATCH_PROF(); \
10558 } while (0)
10559
10560# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS() \
10561 do { \
10562 Log4Func(("\n")); \
10563 } while (0)
10564#else /* nonstrict builds: */
10565# define HMVMX_VALIDATE_EXIT_HANDLER_PARAMS() \
10566 do { \
10567 HMVMX_STOP_EXIT_DISPATCH_PROF(); \
10568 NOREF(pVCpu); NOREF(pMixedCtx); NOREF(pVmxTransient); \
10569 } while (0)
10570# define HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS() do { } while (0)
10571#endif
10572
10573
10574/**
10575 * Advances the guest RIP by the specified number of bytes.
10576 *
10577 * @param pVCpu The cross context virtual CPU structure.
10578 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
10579 * out-of-sync. Make sure to update the required fields
10580 * before using them.
10581 * @param cbInstr Number of bytes to advance the RIP by.
10582 *
10583 * @remarks No-long-jump zone!!!
10584 */
10585DECLINLINE(void) hmR0VmxAdvanceGuestRipBy(PVMCPU pVCpu, PCPUMCTX pMixedCtx, uint32_t cbInstr)
10586{
10587 /* Advance the RIP. */
10588 pMixedCtx->rip += cbInstr;
10589 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
10590
10591 /* Update interrupt inhibition. */
10592 if ( VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS)
10593 && pMixedCtx->rip != EMGetInhibitInterruptsPC(pVCpu))
10594 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
10595}
10596
10597
10598/**
10599 * Advances the guest RIP after reading it from the VMCS.
10600 *
10601 * @returns VBox status code, no informational status codes.
10602 * @param pVCpu The cross context virtual CPU structure.
10603 * @param pMixedCtx Pointer to the guest-CPU context. The data maybe
10604 * out-of-sync. Make sure to update the required fields
10605 * before using them.
10606 * @param pVmxTransient Pointer to the VMX transient structure.
10607 *
10608 * @remarks No-long-jump zone!!!
10609 */
10610static int hmR0VmxAdvanceGuestRip(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
10611{
10612 int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
10613 rc |= hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
10614 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
10615 AssertRCReturn(rc, rc);
10616
10617 hmR0VmxAdvanceGuestRipBy(pVCpu, pMixedCtx, pVmxTransient->cbInstr);
10618
10619 /*
10620 * Deliver a debug exception to the guest if it is single-stepping. Don't directly inject a #DB but use the
10621 * pending debug exception field as it takes care of priority of events.
10622 *
10623 * See Intel spec. 32.2.1 "Debug Exceptions".
10624 */
10625 if ( !pVCpu->hm.s.fSingleInstruction
10626 && pMixedCtx->eflags.Bits.u1TF)
10627 hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
10628
10629 return VINF_SUCCESS;
10630}
10631
10632
10633/**
10634 * Tries to determine what part of the guest-state VT-x has deemed as invalid
10635 * and update error record fields accordingly.
10636 *
10637 * @return VMX_IGS_* return codes.
10638 * @retval VMX_IGS_REASON_NOT_FOUND if this function could not find anything
10639 * wrong with the guest state.
10640 *
10641 * @param pVM The cross context VM structure.
10642 * @param pVCpu The cross context virtual CPU structure.
10643 * @param pCtx Pointer to the guest-CPU state.
10644 *
10645 * @remarks This function assumes our cache of the VMCS controls
10646 * are valid, i.e. hmR0VmxCheckVmcsCtls() succeeded.
10647 */
10648static uint32_t hmR0VmxCheckGuestState(PVM pVM, PVMCPU pVCpu, PCPUMCTX pCtx)
10649{
10650#define HMVMX_ERROR_BREAK(err) { uError = (err); break; }
10651#define HMVMX_CHECK_BREAK(expr, err) if (!(expr)) { \
10652 uError = (err); \
10653 break; \
10654 } else do { } while (0)
10655
10656 int rc;
10657 uint32_t uError = VMX_IGS_ERROR;
10658 uint32_t u32Val;
10659 bool fUnrestrictedGuest = pVM->hm.s.vmx.fUnrestrictedGuest;
10660
10661 do
10662 {
10663 /*
10664 * CR0.
10665 */
10666 uint32_t uSetCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
10667 uint32_t uZapCR0 = (uint32_t)(pVM->hm.s.vmx.Msrs.u64Cr0Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr0Fixed1);
10668 /* Exceptions for unrestricted-guests for fixed CR0 bits (PE, PG).
10669 See Intel spec. 26.3.1 "Checks on Guest Control Registers, Debug Registers and MSRs." */
10670 if (fUnrestrictedGuest)
10671 uSetCR0 &= ~(X86_CR0_PE | X86_CR0_PG);
10672
10673 uint32_t u32GuestCR0;
10674 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32GuestCR0);
10675 AssertRCBreak(rc);
10676 HMVMX_CHECK_BREAK((u32GuestCR0 & uSetCR0) == uSetCR0, VMX_IGS_CR0_FIXED1);
10677 HMVMX_CHECK_BREAK(!(u32GuestCR0 & ~uZapCR0), VMX_IGS_CR0_FIXED0);
10678 if ( !fUnrestrictedGuest
10679 && (u32GuestCR0 & X86_CR0_PG)
10680 && !(u32GuestCR0 & X86_CR0_PE))
10681 {
10682 HMVMX_ERROR_BREAK(VMX_IGS_CR0_PG_PE_COMBO);
10683 }
10684
10685 /*
10686 * CR4.
10687 */
10688 uint64_t uSetCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 & pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
10689 uint64_t uZapCR4 = (pVM->hm.s.vmx.Msrs.u64Cr4Fixed0 | pVM->hm.s.vmx.Msrs.u64Cr4Fixed1);
10690
10691 uint32_t u32GuestCR4;
10692 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR4, &u32GuestCR4);
10693 AssertRCBreak(rc);
10694 HMVMX_CHECK_BREAK((u32GuestCR4 & uSetCR4) == uSetCR4, VMX_IGS_CR4_FIXED1);
10695 HMVMX_CHECK_BREAK(!(u32GuestCR4 & ~uZapCR4), VMX_IGS_CR4_FIXED0);
10696
10697 /*
10698 * IA32_DEBUGCTL MSR.
10699 */
10700 uint64_t u64Val;
10701 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_DEBUGCTL_FULL, &u64Val);
10702 AssertRCBreak(rc);
10703 if ( (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
10704 && (u64Val & 0xfffffe3c)) /* Bits 31:9, bits 5:2 MBZ. */
10705 {
10706 HMVMX_ERROR_BREAK(VMX_IGS_DEBUGCTL_MSR_RESERVED);
10707 }
10708 uint64_t u64DebugCtlMsr = u64Val;
10709
10710#ifdef VBOX_STRICT
10711 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY, &u32Val);
10712 AssertRCBreak(rc);
10713 Assert(u32Val == pVCpu->hm.s.vmx.u32EntryCtls);
10714#endif
10715 bool const fLongModeGuest = RT_BOOL(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST);
10716
10717 /*
10718 * RIP and RFLAGS.
10719 */
10720 uint32_t u32Eflags;
10721#if HC_ARCH_BITS == 64
10722 rc = VMXReadVmcs64(VMX_VMCS_GUEST_RIP, &u64Val);
10723 AssertRCBreak(rc);
10724 /* pCtx->rip can be different than the one in the VMCS (e.g. run guest code and VM-exits that don't update it). */
10725 if ( !fLongModeGuest
10726 || !pCtx->cs.Attr.n.u1Long)
10727 {
10728 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffff00000000)), VMX_IGS_LONGMODE_RIP_INVALID);
10729 }
10730 /** @todo If the processor supports N < 64 linear-address bits, bits 63:N
10731 * must be identical if the "IA-32e mode guest" VM-entry
10732 * control is 1 and CS.L is 1. No check applies if the
10733 * CPU supports 64 linear-address bits. */
10734
10735 /* Flags in pCtx can be different (real-on-v86 for instance). We are only concerned about the VMCS contents here. */
10736 rc = VMXReadVmcs64(VMX_VMCS_GUEST_RFLAGS, &u64Val);
10737 AssertRCBreak(rc);
10738 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffc08028)), /* Bit 63:22, Bit 15, 5, 3 MBZ. */
10739 VMX_IGS_RFLAGS_RESERVED);
10740 HMVMX_CHECK_BREAK((u64Val & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
10741 u32Eflags = u64Val;
10742#else
10743 rc = VMXReadVmcs32(VMX_VMCS_GUEST_RFLAGS, &u32Eflags);
10744 AssertRCBreak(rc);
10745 HMVMX_CHECK_BREAK(!(u32Eflags & 0xffc08028), VMX_IGS_RFLAGS_RESERVED); /* Bit 31:22, Bit 15, 5, 3 MBZ. */
10746 HMVMX_CHECK_BREAK((u32Eflags & X86_EFL_RA1_MASK), VMX_IGS_RFLAGS_RESERVED1); /* Bit 1 MB1. */
10747#endif
10748
10749 if ( fLongModeGuest
10750 || ( fUnrestrictedGuest
10751 && !(u32GuestCR0 & X86_CR0_PE)))
10752 {
10753 HMVMX_CHECK_BREAK(!(u32Eflags & X86_EFL_VM), VMX_IGS_RFLAGS_VM_INVALID);
10754 }
10755
10756 uint32_t u32EntryInfo;
10757 rc = VMXReadVmcs32(VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO, &u32EntryInfo);
10758 AssertRCBreak(rc);
10759 if ( VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo)
10760 && VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
10761 {
10762 HMVMX_CHECK_BREAK(u32Eflags & X86_EFL_IF, VMX_IGS_RFLAGS_IF_INVALID);
10763 }
10764
10765 /*
10766 * 64-bit checks.
10767 */
10768#if HC_ARCH_BITS == 64
10769 if (fLongModeGuest)
10770 {
10771 HMVMX_CHECK_BREAK(u32GuestCR0 & X86_CR0_PG, VMX_IGS_CR0_PG_LONGMODE);
10772 HMVMX_CHECK_BREAK(u32GuestCR4 & X86_CR4_PAE, VMX_IGS_CR4_PAE_LONGMODE);
10773 }
10774
10775 if ( !fLongModeGuest
10776 && (u32GuestCR4 & X86_CR4_PCIDE))
10777 {
10778 HMVMX_ERROR_BREAK(VMX_IGS_CR4_PCIDE);
10779 }
10780
10781 /** @todo CR3 field must be such that bits 63:52 and bits in the range
10782 * 51:32 beyond the processor's physical-address width are 0. */
10783
10784 if ( (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_DEBUG)
10785 && (pCtx->dr[7] & X86_DR7_MBZ_MASK))
10786 {
10787 HMVMX_ERROR_BREAK(VMX_IGS_DR7_RESERVED);
10788 }
10789
10790 rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_ESP, &u64Val);
10791 AssertRCBreak(rc);
10792 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_ESP_NOT_CANONICAL);
10793
10794 rc = VMXReadVmcs64(VMX_VMCS_HOST_SYSENTER_EIP, &u64Val);
10795 AssertRCBreak(rc);
10796 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_SYSENTER_EIP_NOT_CANONICAL);
10797#endif
10798
10799 /*
10800 * PERF_GLOBAL MSR.
10801 */
10802 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PERF_MSR)
10803 {
10804 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PERF_GLOBAL_CTRL_FULL, &u64Val);
10805 AssertRCBreak(rc);
10806 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffff8fffffffc)),
10807 VMX_IGS_PERF_GLOBAL_MSR_RESERVED); /* Bits 63:35, bits 31:2 MBZ. */
10808 }
10809
10810 /*
10811 * PAT MSR.
10812 */
10813 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_PAT_MSR)
10814 {
10815 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PAT_FULL, &u64Val);
10816 AssertRCBreak(rc);
10817 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0x707070707070707)), VMX_IGS_PAT_MSR_RESERVED);
10818 for (unsigned i = 0; i < 8; i++)
10819 {
10820 uint8_t u8Val = (u64Val & 0xff);
10821 if ( u8Val != 0 /* UC */
10822 && u8Val != 1 /* WC */
10823 && u8Val != 4 /* WT */
10824 && u8Val != 5 /* WP */
10825 && u8Val != 6 /* WB */
10826 && u8Val != 7 /* UC- */)
10827 {
10828 HMVMX_ERROR_BREAK(VMX_IGS_PAT_MSR_INVALID);
10829 }
10830 u64Val >>= 8;
10831 }
10832 }
10833
10834 /*
10835 * EFER MSR.
10836 */
10837 if (pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_LOAD_GUEST_EFER_MSR)
10838 {
10839 Assert(pVM->hm.s.vmx.fSupportsVmcsEfer);
10840 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_EFER_FULL, &u64Val);
10841 AssertRCBreak(rc);
10842 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xfffffffffffff2fe)),
10843 VMX_IGS_EFER_MSR_RESERVED); /* Bits 63:12, bit 9, bits 7:1 MBZ. */
10844 HMVMX_CHECK_BREAK(RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL( pVCpu->hm.s.vmx.u32EntryCtls
10845 & VMX_VMCS_CTRL_ENTRY_IA32E_MODE_GUEST),
10846 VMX_IGS_EFER_LMA_GUEST_MODE_MISMATCH);
10847 HMVMX_CHECK_BREAK( fUnrestrictedGuest
10848 || !(u32GuestCR0 & X86_CR0_PG)
10849 || RT_BOOL(u64Val & MSR_K6_EFER_LMA) == RT_BOOL(u64Val & MSR_K6_EFER_LME),
10850 VMX_IGS_EFER_LMA_LME_MISMATCH);
10851 }
10852
10853 /*
10854 * Segment registers.
10855 */
10856 HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10857 || !(pCtx->ldtr.Sel & X86_SEL_LDT), VMX_IGS_LDTR_TI_INVALID);
10858 if (!(u32Eflags & X86_EFL_VM))
10859 {
10860 /* CS */
10861 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1Present, VMX_IGS_CS_ATTR_P_INVALID);
10862 HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xf00), VMX_IGS_CS_ATTR_RESERVED);
10863 HMVMX_CHECK_BREAK(!(pCtx->cs.Attr.u & 0xfffe0000), VMX_IGS_CS_ATTR_RESERVED);
10864 HMVMX_CHECK_BREAK( (pCtx->cs.u32Limit & 0xfff) == 0xfff
10865 || !(pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
10866 HMVMX_CHECK_BREAK( !(pCtx->cs.u32Limit & 0xfff00000)
10867 || (pCtx->cs.Attr.n.u1Granularity), VMX_IGS_CS_ATTR_G_INVALID);
10868 /* CS cannot be loaded with NULL in protected mode. */
10869 HMVMX_CHECK_BREAK(pCtx->cs.Attr.u && !(pCtx->cs.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_CS_ATTR_UNUSABLE);
10870 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u1DescType, VMX_IGS_CS_ATTR_S_INVALID);
10871 if (pCtx->cs.Attr.n.u4Type == 9 || pCtx->cs.Attr.n.u4Type == 11)
10872 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_UNEQUAL);
10873 else if (pCtx->cs.Attr.n.u4Type == 13 || pCtx->cs.Attr.n.u4Type == 15)
10874 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl <= pCtx->ss.Attr.n.u2Dpl, VMX_IGS_CS_SS_ATTR_DPL_MISMATCH);
10875 else if (pVM->hm.s.vmx.fUnrestrictedGuest && pCtx->cs.Attr.n.u4Type == 3)
10876 HMVMX_CHECK_BREAK(pCtx->cs.Attr.n.u2Dpl == 0, VMX_IGS_CS_ATTR_DPL_INVALID);
10877 else
10878 HMVMX_ERROR_BREAK(VMX_IGS_CS_ATTR_TYPE_INVALID);
10879
10880 /* SS */
10881 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10882 || (pCtx->ss.Sel & X86_SEL_RPL) == (pCtx->cs.Sel & X86_SEL_RPL), VMX_IGS_SS_CS_RPL_UNEQUAL);
10883 HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u2Dpl == (pCtx->ss.Sel & X86_SEL_RPL), VMX_IGS_SS_ATTR_DPL_RPL_UNEQUAL);
10884 if ( !(pCtx->cr0 & X86_CR0_PE)
10885 || pCtx->cs.Attr.n.u4Type == 3)
10886 {
10887 HMVMX_CHECK_BREAK(!pCtx->ss.Attr.n.u2Dpl, VMX_IGS_SS_ATTR_DPL_INVALID);
10888 }
10889 if (!(pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE))
10890 {
10891 HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u4Type == 3 || pCtx->ss.Attr.n.u4Type == 7, VMX_IGS_SS_ATTR_TYPE_INVALID);
10892 HMVMX_CHECK_BREAK(pCtx->ss.Attr.n.u1Present, VMX_IGS_SS_ATTR_P_INVALID);
10893 HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xf00), VMX_IGS_SS_ATTR_RESERVED);
10894 HMVMX_CHECK_BREAK(!(pCtx->ss.Attr.u & 0xfffe0000), VMX_IGS_SS_ATTR_RESERVED);
10895 HMVMX_CHECK_BREAK( (pCtx->ss.u32Limit & 0xfff) == 0xfff
10896 || !(pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
10897 HMVMX_CHECK_BREAK( !(pCtx->ss.u32Limit & 0xfff00000)
10898 || (pCtx->ss.Attr.n.u1Granularity), VMX_IGS_SS_ATTR_G_INVALID);
10899 }
10900
10901 /* DS, ES, FS, GS - only check for usable selectors, see hmR0VmxWriteSegmentReg(). */
10902 if (!(pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE))
10903 {
10904 HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_DS_ATTR_A_INVALID);
10905 HMVMX_CHECK_BREAK(pCtx->ds.Attr.n.u1Present, VMX_IGS_DS_ATTR_P_INVALID);
10906 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10907 || pCtx->ds.Attr.n.u4Type > 11
10908 || pCtx->ds.Attr.n.u2Dpl >= (pCtx->ds.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
10909 HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xf00), VMX_IGS_DS_ATTR_RESERVED);
10910 HMVMX_CHECK_BREAK(!(pCtx->ds.Attr.u & 0xfffe0000), VMX_IGS_DS_ATTR_RESERVED);
10911 HMVMX_CHECK_BREAK( (pCtx->ds.u32Limit & 0xfff) == 0xfff
10912 || !(pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
10913 HMVMX_CHECK_BREAK( !(pCtx->ds.u32Limit & 0xfff00000)
10914 || (pCtx->ds.Attr.n.u1Granularity), VMX_IGS_DS_ATTR_G_INVALID);
10915 HMVMX_CHECK_BREAK( !(pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10916 || (pCtx->ds.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_DS_ATTR_TYPE_INVALID);
10917 }
10918 if (!(pCtx->es.Attr.u & X86DESCATTR_UNUSABLE))
10919 {
10920 HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_ES_ATTR_A_INVALID);
10921 HMVMX_CHECK_BREAK(pCtx->es.Attr.n.u1Present, VMX_IGS_ES_ATTR_P_INVALID);
10922 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10923 || pCtx->es.Attr.n.u4Type > 11
10924 || pCtx->es.Attr.n.u2Dpl >= (pCtx->es.Sel & X86_SEL_RPL), VMX_IGS_DS_ATTR_DPL_RPL_UNEQUAL);
10925 HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xf00), VMX_IGS_ES_ATTR_RESERVED);
10926 HMVMX_CHECK_BREAK(!(pCtx->es.Attr.u & 0xfffe0000), VMX_IGS_ES_ATTR_RESERVED);
10927 HMVMX_CHECK_BREAK( (pCtx->es.u32Limit & 0xfff) == 0xfff
10928 || !(pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
10929 HMVMX_CHECK_BREAK( !(pCtx->es.u32Limit & 0xfff00000)
10930 || (pCtx->es.Attr.n.u1Granularity), VMX_IGS_ES_ATTR_G_INVALID);
10931 HMVMX_CHECK_BREAK( !(pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10932 || (pCtx->es.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_ES_ATTR_TYPE_INVALID);
10933 }
10934 if (!(pCtx->fs.Attr.u & X86DESCATTR_UNUSABLE))
10935 {
10936 HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_FS_ATTR_A_INVALID);
10937 HMVMX_CHECK_BREAK(pCtx->fs.Attr.n.u1Present, VMX_IGS_FS_ATTR_P_INVALID);
10938 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10939 || pCtx->fs.Attr.n.u4Type > 11
10940 || pCtx->fs.Attr.n.u2Dpl >= (pCtx->fs.Sel & X86_SEL_RPL), VMX_IGS_FS_ATTR_DPL_RPL_UNEQUAL);
10941 HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xf00), VMX_IGS_FS_ATTR_RESERVED);
10942 HMVMX_CHECK_BREAK(!(pCtx->fs.Attr.u & 0xfffe0000), VMX_IGS_FS_ATTR_RESERVED);
10943 HMVMX_CHECK_BREAK( (pCtx->fs.u32Limit & 0xfff) == 0xfff
10944 || !(pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
10945 HMVMX_CHECK_BREAK( !(pCtx->fs.u32Limit & 0xfff00000)
10946 || (pCtx->fs.Attr.n.u1Granularity), VMX_IGS_FS_ATTR_G_INVALID);
10947 HMVMX_CHECK_BREAK( !(pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10948 || (pCtx->fs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_FS_ATTR_TYPE_INVALID);
10949 }
10950 if (!(pCtx->gs.Attr.u & X86DESCATTR_UNUSABLE))
10951 {
10952 HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_ACCESSED, VMX_IGS_GS_ATTR_A_INVALID);
10953 HMVMX_CHECK_BREAK(pCtx->gs.Attr.n.u1Present, VMX_IGS_GS_ATTR_P_INVALID);
10954 HMVMX_CHECK_BREAK( pVM->hm.s.vmx.fUnrestrictedGuest
10955 || pCtx->gs.Attr.n.u4Type > 11
10956 || pCtx->gs.Attr.n.u2Dpl >= (pCtx->gs.Sel & X86_SEL_RPL), VMX_IGS_GS_ATTR_DPL_RPL_UNEQUAL);
10957 HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xf00), VMX_IGS_GS_ATTR_RESERVED);
10958 HMVMX_CHECK_BREAK(!(pCtx->gs.Attr.u & 0xfffe0000), VMX_IGS_GS_ATTR_RESERVED);
10959 HMVMX_CHECK_BREAK( (pCtx->gs.u32Limit & 0xfff) == 0xfff
10960 || !(pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
10961 HMVMX_CHECK_BREAK( !(pCtx->gs.u32Limit & 0xfff00000)
10962 || (pCtx->gs.Attr.n.u1Granularity), VMX_IGS_GS_ATTR_G_INVALID);
10963 HMVMX_CHECK_BREAK( !(pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_CODE)
10964 || (pCtx->gs.Attr.n.u4Type & X86_SEL_TYPE_READ), VMX_IGS_GS_ATTR_TYPE_INVALID);
10965 }
10966 /* 64-bit capable CPUs. */
10967#if HC_ARCH_BITS == 64
10968 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
10969 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
10970 HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
10971 || X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
10972 HMVMX_CHECK_BREAK(!(pCtx->cs.u64Base >> 32), VMX_IGS_LONGMODE_CS_BASE_INVALID);
10973 HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ss.u64Base >> 32),
10974 VMX_IGS_LONGMODE_SS_BASE_INVALID);
10975 HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ds.u64Base >> 32),
10976 VMX_IGS_LONGMODE_DS_BASE_INVALID);
10977 HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->es.u64Base >> 32),
10978 VMX_IGS_LONGMODE_ES_BASE_INVALID);
10979#endif
10980 }
10981 else
10982 {
10983 /* V86 mode checks. */
10984 uint32_t u32CSAttr, u32SSAttr, u32DSAttr, u32ESAttr, u32FSAttr, u32GSAttr;
10985 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
10986 {
10987 u32CSAttr = 0xf3; u32SSAttr = 0xf3;
10988 u32DSAttr = 0xf3; u32ESAttr = 0xf3;
10989 u32FSAttr = 0xf3; u32GSAttr = 0xf3;
10990 }
10991 else
10992 {
10993 u32CSAttr = pCtx->cs.Attr.u; u32SSAttr = pCtx->ss.Attr.u;
10994 u32DSAttr = pCtx->ds.Attr.u; u32ESAttr = pCtx->es.Attr.u;
10995 u32FSAttr = pCtx->fs.Attr.u; u32GSAttr = pCtx->gs.Attr.u;
10996 }
10997
10998 /* CS */
10999 HMVMX_CHECK_BREAK((pCtx->cs.u64Base == (uint64_t)pCtx->cs.Sel << 4), VMX_IGS_V86_CS_BASE_INVALID);
11000 HMVMX_CHECK_BREAK(pCtx->cs.u32Limit == 0xffff, VMX_IGS_V86_CS_LIMIT_INVALID);
11001 HMVMX_CHECK_BREAK(u32CSAttr == 0xf3, VMX_IGS_V86_CS_ATTR_INVALID);
11002 /* SS */
11003 HMVMX_CHECK_BREAK((pCtx->ss.u64Base == (uint64_t)pCtx->ss.Sel << 4), VMX_IGS_V86_SS_BASE_INVALID);
11004 HMVMX_CHECK_BREAK(pCtx->ss.u32Limit == 0xffff, VMX_IGS_V86_SS_LIMIT_INVALID);
11005 HMVMX_CHECK_BREAK(u32SSAttr == 0xf3, VMX_IGS_V86_SS_ATTR_INVALID);
11006 /* DS */
11007 HMVMX_CHECK_BREAK((pCtx->ds.u64Base == (uint64_t)pCtx->ds.Sel << 4), VMX_IGS_V86_DS_BASE_INVALID);
11008 HMVMX_CHECK_BREAK(pCtx->ds.u32Limit == 0xffff, VMX_IGS_V86_DS_LIMIT_INVALID);
11009 HMVMX_CHECK_BREAK(u32DSAttr == 0xf3, VMX_IGS_V86_DS_ATTR_INVALID);
11010 /* ES */
11011 HMVMX_CHECK_BREAK((pCtx->es.u64Base == (uint64_t)pCtx->es.Sel << 4), VMX_IGS_V86_ES_BASE_INVALID);
11012 HMVMX_CHECK_BREAK(pCtx->es.u32Limit == 0xffff, VMX_IGS_V86_ES_LIMIT_INVALID);
11013 HMVMX_CHECK_BREAK(u32ESAttr == 0xf3, VMX_IGS_V86_ES_ATTR_INVALID);
11014 /* FS */
11015 HMVMX_CHECK_BREAK((pCtx->fs.u64Base == (uint64_t)pCtx->fs.Sel << 4), VMX_IGS_V86_FS_BASE_INVALID);
11016 HMVMX_CHECK_BREAK(pCtx->fs.u32Limit == 0xffff, VMX_IGS_V86_FS_LIMIT_INVALID);
11017 HMVMX_CHECK_BREAK(u32FSAttr == 0xf3, VMX_IGS_V86_FS_ATTR_INVALID);
11018 /* GS */
11019 HMVMX_CHECK_BREAK((pCtx->gs.u64Base == (uint64_t)pCtx->gs.Sel << 4), VMX_IGS_V86_GS_BASE_INVALID);
11020 HMVMX_CHECK_BREAK(pCtx->gs.u32Limit == 0xffff, VMX_IGS_V86_GS_LIMIT_INVALID);
11021 HMVMX_CHECK_BREAK(u32GSAttr == 0xf3, VMX_IGS_V86_GS_ATTR_INVALID);
11022 /* 64-bit capable CPUs. */
11023#if HC_ARCH_BITS == 64
11024 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->fs.u64Base), VMX_IGS_FS_BASE_NOT_CANONICAL);
11025 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->gs.u64Base), VMX_IGS_GS_BASE_NOT_CANONICAL);
11026 HMVMX_CHECK_BREAK( (pCtx->ldtr.Attr.u & X86DESCATTR_UNUSABLE)
11027 || X86_IS_CANONICAL(pCtx->ldtr.u64Base), VMX_IGS_LDTR_BASE_NOT_CANONICAL);
11028 HMVMX_CHECK_BREAK(!(pCtx->cs.u64Base >> 32), VMX_IGS_LONGMODE_CS_BASE_INVALID);
11029 HMVMX_CHECK_BREAK((pCtx->ss.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ss.u64Base >> 32),
11030 VMX_IGS_LONGMODE_SS_BASE_INVALID);
11031 HMVMX_CHECK_BREAK((pCtx->ds.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->ds.u64Base >> 32),
11032 VMX_IGS_LONGMODE_DS_BASE_INVALID);
11033 HMVMX_CHECK_BREAK((pCtx->es.Attr.u & X86DESCATTR_UNUSABLE) || !(pCtx->es.u64Base >> 32),
11034 VMX_IGS_LONGMODE_ES_BASE_INVALID);
11035#endif
11036 }
11037
11038 /*
11039 * TR.
11040 */
11041 HMVMX_CHECK_BREAK(!(pCtx->tr.Sel & X86_SEL_LDT), VMX_IGS_TR_TI_INVALID);
11042 /* 64-bit capable CPUs. */
11043#if HC_ARCH_BITS == 64
11044 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(pCtx->tr.u64Base), VMX_IGS_TR_BASE_NOT_CANONICAL);
11045#endif
11046 if (fLongModeGuest)
11047 {
11048 HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u4Type == 11, /* 64-bit busy TSS. */
11049 VMX_IGS_LONGMODE_TR_ATTR_TYPE_INVALID);
11050 }
11051 else
11052 {
11053 HMVMX_CHECK_BREAK( pCtx->tr.Attr.n.u4Type == 3 /* 16-bit busy TSS. */
11054 || pCtx->tr.Attr.n.u4Type == 11, /* 32-bit busy TSS.*/
11055 VMX_IGS_TR_ATTR_TYPE_INVALID);
11056 }
11057 HMVMX_CHECK_BREAK(!pCtx->tr.Attr.n.u1DescType, VMX_IGS_TR_ATTR_S_INVALID);
11058 HMVMX_CHECK_BREAK(pCtx->tr.Attr.n.u1Present, VMX_IGS_TR_ATTR_P_INVALID);
11059 HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & 0xf00), VMX_IGS_TR_ATTR_RESERVED); /* Bits 11:8 MBZ. */
11060 HMVMX_CHECK_BREAK( (pCtx->tr.u32Limit & 0xfff) == 0xfff
11061 || !(pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
11062 HMVMX_CHECK_BREAK( !(pCtx->tr.u32Limit & 0xfff00000)
11063 || (pCtx->tr.Attr.n.u1Granularity), VMX_IGS_TR_ATTR_G_INVALID);
11064 HMVMX_CHECK_BREAK(!(pCtx->tr.Attr.u & X86DESCATTR_UNUSABLE), VMX_IGS_TR_ATTR_UNUSABLE);
11065
11066 /*
11067 * GDTR and IDTR.
11068 */
11069#if HC_ARCH_BITS == 64
11070 rc = VMXReadVmcs64(VMX_VMCS_GUEST_GDTR_BASE, &u64Val);
11071 AssertRCBreak(rc);
11072 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_GDTR_BASE_NOT_CANONICAL);
11073
11074 rc = VMXReadVmcs64(VMX_VMCS_GUEST_IDTR_BASE, &u64Val);
11075 AssertRCBreak(rc);
11076 HMVMX_CHECK_BREAK(X86_IS_CANONICAL(u64Val), VMX_IGS_IDTR_BASE_NOT_CANONICAL);
11077#endif
11078
11079 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_GDTR_LIMIT, &u32Val);
11080 AssertRCBreak(rc);
11081 HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_GDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
11082
11083 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_IDTR_LIMIT, &u32Val);
11084 AssertRCBreak(rc);
11085 HMVMX_CHECK_BREAK(!(u32Val & 0xffff0000), VMX_IGS_IDTR_LIMIT_INVALID); /* Bits 31:16 MBZ. */
11086
11087 /*
11088 * Guest Non-Register State.
11089 */
11090 /* Activity State. */
11091 uint32_t u32ActivityState;
11092 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_ACTIVITY_STATE, &u32ActivityState);
11093 AssertRCBreak(rc);
11094 HMVMX_CHECK_BREAK( !u32ActivityState
11095 || (u32ActivityState & MSR_IA32_VMX_MISC_ACTIVITY_STATES(pVM->hm.s.vmx.Msrs.u64Misc)),
11096 VMX_IGS_ACTIVITY_STATE_INVALID);
11097 HMVMX_CHECK_BREAK( !(pCtx->ss.Attr.n.u2Dpl)
11098 || u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_HLT, VMX_IGS_ACTIVITY_STATE_HLT_INVALID);
11099 uint32_t u32IntrState;
11100 rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &u32IntrState);
11101 AssertRCBreak(rc);
11102 if ( u32IntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS
11103 || u32IntrState == VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
11104 {
11105 HMVMX_CHECK_BREAK(u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_ACTIVE, VMX_IGS_ACTIVITY_STATE_ACTIVE_INVALID);
11106 }
11107
11108 /** @todo Activity state and injecting interrupts. Left as a todo since we
11109 * currently don't use activity states but ACTIVE. */
11110
11111 HMVMX_CHECK_BREAK( !(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_ENTRY_SMM)
11112 || u32ActivityState != VMX_VMCS_GUEST_ACTIVITY_SIPI_WAIT, VMX_IGS_ACTIVITY_STATE_SIPI_WAIT_INVALID);
11113
11114 /* Guest interruptibility-state. */
11115 HMVMX_CHECK_BREAK(!(u32IntrState & 0xfffffff0), VMX_IGS_INTERRUPTIBILITY_STATE_RESERVED);
11116 HMVMX_CHECK_BREAK((u32IntrState & ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI
11117 | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS))
11118 != ( VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI
11119 | VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
11120 VMX_IGS_INTERRUPTIBILITY_STATE_STI_MOVSS_INVALID);
11121 HMVMX_CHECK_BREAK( (u32Eflags & X86_EFL_IF)
11122 || !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI),
11123 VMX_IGS_INTERRUPTIBILITY_STATE_STI_EFL_INVALID);
11124 if (VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo))
11125 {
11126 if (VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT)
11127 {
11128 HMVMX_CHECK_BREAK( !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
11129 && !(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
11130 VMX_IGS_INTERRUPTIBILITY_STATE_EXT_INT_INVALID);
11131 }
11132 else if (VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
11133 {
11134 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS),
11135 VMX_IGS_INTERRUPTIBILITY_STATE_MOVSS_INVALID);
11136 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI),
11137 VMX_IGS_INTERRUPTIBILITY_STATE_STI_INVALID);
11138 }
11139 }
11140 /** @todo Assumes the processor is not in SMM. */
11141 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI),
11142 VMX_IGS_INTERRUPTIBILITY_STATE_SMI_INVALID);
11143 HMVMX_CHECK_BREAK( !(pVCpu->hm.s.vmx.u32EntryCtls & VMX_VMCS_CTRL_ENTRY_ENTRY_SMM)
11144 || (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_SMI),
11145 VMX_IGS_INTERRUPTIBILITY_STATE_SMI_SMM_INVALID);
11146 if ( (pVCpu->hm.s.vmx.u32PinCtls & VMX_VMCS_CTRL_PIN_EXEC_VIRTUAL_NMI)
11147 && VMX_ENTRY_INTERRUPTION_INFO_IS_VALID(u32EntryInfo)
11148 && VMX_ENTRY_INTERRUPTION_INFO_TYPE(u32EntryInfo) == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
11149 {
11150 HMVMX_CHECK_BREAK(!(u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_NMI),
11151 VMX_IGS_INTERRUPTIBILITY_STATE_NMI_INVALID);
11152 }
11153
11154 /* Pending debug exceptions. */
11155#if HC_ARCH_BITS == 64
11156 rc = VMXReadVmcs64(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, &u64Val);
11157 AssertRCBreak(rc);
11158 /* Bits 63:15, Bit 13, Bits 11:4 MBZ. */
11159 HMVMX_CHECK_BREAK(!(u64Val & UINT64_C(0xffffffffffffaff0)), VMX_IGS_LONGMODE_PENDING_DEBUG_RESERVED);
11160 u32Val = u64Val; /* For pending debug exceptions checks below. */
11161#else
11162 rc = VMXReadVmcs32(VMX_VMCS_GUEST_PENDING_DEBUG_EXCEPTIONS, &u32Val);
11163 AssertRCBreak(rc);
11164 /* Bits 31:15, Bit 13, Bits 11:4 MBZ. */
11165 HMVMX_CHECK_BREAK(!(u32Val & 0xffffaff0), VMX_IGS_PENDING_DEBUG_RESERVED);
11166#endif
11167
11168 if ( (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI)
11169 || (u32IntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_MOVSS)
11170 || u32ActivityState == VMX_VMCS_GUEST_ACTIVITY_HLT)
11171 {
11172 if ( (u32Eflags & X86_EFL_TF)
11173 && !(u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
11174 {
11175 /* Bit 14 is PendingDebug.BS. */
11176 HMVMX_CHECK_BREAK(u32Val & RT_BIT(14), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_SET);
11177 }
11178 if ( !(u32Eflags & X86_EFL_TF)
11179 || (u64DebugCtlMsr & RT_BIT_64(1))) /* Bit 1 is IA32_DEBUGCTL.BTF. */
11180 {
11181 /* Bit 14 is PendingDebug.BS. */
11182 HMVMX_CHECK_BREAK(!(u32Val & RT_BIT(14)), VMX_IGS_PENDING_DEBUG_XCPT_BS_NOT_CLEAR);
11183 }
11184 }
11185
11186 /* VMCS link pointer. */
11187 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_VMCS_LINK_PTR_FULL, &u64Val);
11188 AssertRCBreak(rc);
11189 if (u64Val != UINT64_C(0xffffffffffffffff))
11190 {
11191 HMVMX_CHECK_BREAK(!(u64Val & 0xfff), VMX_IGS_VMCS_LINK_PTR_RESERVED);
11192 /** @todo Bits beyond the processor's physical-address width MBZ. */
11193 /** @todo 32-bit located in memory referenced by value of this field (as a
11194 * physical address) must contain the processor's VMCS revision ID. */
11195 /** @todo SMM checks. */
11196 }
11197
11198 /** @todo Checks on Guest Page-Directory-Pointer-Table Entries when guest is
11199 * not using Nested Paging? */
11200 if ( pVM->hm.s.fNestedPaging
11201 && !fLongModeGuest
11202 && CPUMIsGuestInPAEModeEx(pCtx))
11203 {
11204 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE0_FULL, &u64Val);
11205 AssertRCBreak(rc);
11206 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
11207
11208 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE1_FULL, &u64Val);
11209 AssertRCBreak(rc);
11210 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
11211
11212 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE2_FULL, &u64Val);
11213 AssertRCBreak(rc);
11214 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
11215
11216 rc = VMXReadVmcs64(VMX_VMCS64_GUEST_PDPTE3_FULL, &u64Val);
11217 AssertRCBreak(rc);
11218 HMVMX_CHECK_BREAK(!(u64Val & X86_PDPE_PAE_MBZ_MASK), VMX_IGS_PAE_PDPTE_RESERVED);
11219 }
11220
11221 /* Shouldn't happen but distinguish it from AssertRCBreak() errors. */
11222 if (uError == VMX_IGS_ERROR)
11223 uError = VMX_IGS_REASON_NOT_FOUND;
11224 } while (0);
11225
11226 pVCpu->hm.s.u32HMError = uError;
11227 return uError;
11228
11229#undef HMVMX_ERROR_BREAK
11230#undef HMVMX_CHECK_BREAK
11231}
11232
11233/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
11234/* -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- VM-exit handlers -=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=- */
11235/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= */
11236
11237/** @name VM-exit handlers.
11238 * @{
11239 */
11240
11241/**
11242 * VM-exit handler for external interrupts (VMX_EXIT_EXT_INT).
11243 */
11244HMVMX_EXIT_DECL hmR0VmxExitExtInt(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11245{
11246 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11247 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitExtInt);
11248 /* Windows hosts (32-bit and 64-bit) have DPC latency issues. See @bugref{6853}. */
11249 if (VMMR0ThreadCtxHookIsEnabled(pVCpu))
11250 return VINF_SUCCESS;
11251 return VINF_EM_RAW_INTERRUPT;
11252}
11253
11254
11255/**
11256 * VM-exit handler for exceptions or NMIs (VMX_EXIT_XCPT_OR_NMI).
11257 */
11258HMVMX_EXIT_DECL hmR0VmxExitXcptOrNmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11259{
11260 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11261 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitXcptNmi, y3);
11262
11263 int rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
11264 AssertRCReturn(rc, rc);
11265
11266 uint32_t uIntType = VMX_EXIT_INTERRUPTION_INFO_TYPE(pVmxTransient->uExitIntInfo);
11267 Assert( !(pVCpu->hm.s.vmx.u32ExitCtls & VMX_VMCS_CTRL_EXIT_ACK_EXT_INT)
11268 && uIntType != VMX_EXIT_INTERRUPTION_INFO_TYPE_EXT_INT);
11269 Assert(VMX_EXIT_INTERRUPTION_INFO_IS_VALID(pVmxTransient->uExitIntInfo));
11270
11271 if (uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_NMI)
11272 {
11273 /*
11274 * This cannot be a guest NMI as the only way for the guest to receive an NMI is if we injected it ourselves and
11275 * anything we inject is not going to cause a VM-exit directly for the event being injected.
11276 * See Intel spec. 27.2.3 "Information for VM Exits During Event Delivery".
11277 *
11278 * Dispatch the NMI to the host. See Intel spec. 27.5.5 "Updating Non-Register State".
11279 */
11280 VMXDispatchHostNmi();
11281 STAM_REL_COUNTER_INC(&pVCpu->hm.s.StatExitHostNmiInGC);
11282 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
11283 return VINF_SUCCESS;
11284 }
11285
11286 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
11287 VBOXSTRICTRC rcStrictRc1 = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
11288 if (RT_UNLIKELY(rcStrictRc1 == VINF_SUCCESS))
11289 { /* likely */ }
11290 else
11291 {
11292 if (rcStrictRc1 == VINF_HM_DOUBLE_FAULT)
11293 rcStrictRc1 = VINF_SUCCESS;
11294 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
11295 return rcStrictRc1;
11296 }
11297
11298 uint32_t uExitIntInfo = pVmxTransient->uExitIntInfo;
11299 uint32_t uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(uExitIntInfo);
11300 switch (uIntType)
11301 {
11302 case VMX_EXIT_INTERRUPTION_INFO_TYPE_PRIV_SW_XCPT: /* Privileged software exception. (#DB from ICEBP) */
11303 Assert(uVector == X86_XCPT_DB);
11304 /* no break */
11305 case VMX_EXIT_INTERRUPTION_INFO_TYPE_SW_XCPT: /* Software exception. (#BP or #OF) */
11306 Assert(uVector == X86_XCPT_BP || uVector == X86_XCPT_OF || uIntType == VMX_EXIT_INTERRUPTION_INFO_TYPE_PRIV_SW_XCPT);
11307 /* no break */
11308 case VMX_EXIT_INTERRUPTION_INFO_TYPE_HW_XCPT:
11309 {
11310 /*
11311 * If there's any exception caused as a result of event injection, go back to
11312 * the interpreter. The page-fault case is complicated and we manually handle
11313 * any currently pending event in hmR0VmxExitXcptPF. Nested #ACs are already
11314 * handled in hmR0VmxCheckExitDueToEventDelivery.
11315 */
11316 if (!pVCpu->hm.s.Event.fPending)
11317 { /* likely */ }
11318 else if ( uVector != X86_XCPT_PF
11319 && uVector != X86_XCPT_AC)
11320 {
11321 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingInterpret);
11322 rc = VERR_EM_INTERPRETER;
11323 break;
11324 }
11325
11326 switch (uVector)
11327 {
11328 case X86_XCPT_PF: rc = hmR0VmxExitXcptPF(pVCpu, pMixedCtx, pVmxTransient); break;
11329 case X86_XCPT_GP: rc = hmR0VmxExitXcptGP(pVCpu, pMixedCtx, pVmxTransient); break;
11330 case X86_XCPT_NM: rc = hmR0VmxExitXcptNM(pVCpu, pMixedCtx, pVmxTransient); break;
11331 case X86_XCPT_MF: rc = hmR0VmxExitXcptMF(pVCpu, pMixedCtx, pVmxTransient); break;
11332 case X86_XCPT_DB: rc = hmR0VmxExitXcptDB(pVCpu, pMixedCtx, pVmxTransient); break;
11333 case X86_XCPT_BP: rc = hmR0VmxExitXcptBP(pVCpu, pMixedCtx, pVmxTransient); break;
11334 case X86_XCPT_AC: rc = hmR0VmxExitXcptAC(pVCpu, pMixedCtx, pVmxTransient); break;
11335
11336 case X86_XCPT_XF: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXF);
11337 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11338 case X86_XCPT_DE: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDE);
11339 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11340 case X86_XCPT_UD: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestUD);
11341 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11342 case X86_XCPT_SS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestSS);
11343 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11344 case X86_XCPT_NP: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNP);
11345 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11346 case X86_XCPT_TS: STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestTS);
11347 rc = hmR0VmxExitXcptGeneric(pVCpu, pMixedCtx, pVmxTransient); break;
11348 default:
11349 {
11350 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11351 AssertRCReturn(rc, rc);
11352
11353 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestXcpUnk);
11354 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
11355 {
11356 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.vmx.pRealModeTSS);
11357 Assert(PDMVmmDevHeapIsEnabled(pVCpu->CTX_SUFF(pVM)));
11358 Assert(CPUMIsGuestInRealModeEx(pMixedCtx));
11359
11360 rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11361 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
11362 AssertRCReturn(rc, rc);
11363 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(uExitIntInfo),
11364 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode,
11365 0 /* GCPtrFaultAddress */);
11366 AssertRCReturn(rc, rc);
11367 }
11368 else
11369 {
11370 AssertMsgFailed(("Unexpected VM-exit caused by exception %#x\n", uVector));
11371 pVCpu->hm.s.u32HMError = uVector;
11372 rc = VERR_VMX_UNEXPECTED_EXCEPTION;
11373 }
11374 break;
11375 }
11376 }
11377 break;
11378 }
11379
11380 default:
11381 {
11382 pVCpu->hm.s.u32HMError = uExitIntInfo;
11383 rc = VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE;
11384 AssertMsgFailed(("Unexpected interruption info %#x\n", VMX_EXIT_INTERRUPTION_INFO_TYPE(uExitIntInfo)));
11385 break;
11386 }
11387 }
11388 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitXcptNmi, y3);
11389 return rc;
11390}
11391
11392
11393/**
11394 * VM-exit handler for interrupt-window exiting (VMX_EXIT_INT_WINDOW).
11395 */
11396HMVMX_EXIT_NSRC_DECL hmR0VmxExitIntWindow(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11397{
11398 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11399
11400 /* Indicate that we no longer need to VM-exit when the guest is ready to receive interrupts, it is now ready. */
11401 hmR0VmxClearIntWindowExitVmcs(pVCpu);
11402
11403 /* Deliver the pending interrupts via hmR0VmxEvaluatePendingEvent() and resume guest execution. */
11404 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIntWindow);
11405 return VINF_SUCCESS;
11406}
11407
11408
11409/**
11410 * VM-exit handler for NMI-window exiting (VMX_EXIT_NMI_WINDOW).
11411 */
11412HMVMX_EXIT_NSRC_DECL hmR0VmxExitNmiWindow(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11413{
11414 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11415 if (RT_UNLIKELY(!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_NMI_WINDOW_EXIT)))
11416 {
11417 AssertMsgFailed(("Unexpected NMI-window exit.\n"));
11418 HMVMX_RETURN_UNEXPECTED_EXIT();
11419 }
11420
11421 Assert(!VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_BLOCK_NMIS));
11422
11423 /*
11424 * If block-by-STI is set when we get this VM-exit, it means the CPU doesn't block NMIs following STI.
11425 * It is therefore safe to unblock STI and deliver the NMI ourselves. See @bugref{7445}.
11426 */
11427 uint32_t uIntrState = 0;
11428 int rc = VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
11429 AssertRCReturn(rc, rc);
11430
11431 bool const fBlockSti = RT_BOOL(uIntrState & VMX_VMCS_GUEST_INTERRUPTIBILITY_STATE_BLOCK_STI);
11432 if ( fBlockSti
11433 && VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS))
11434 {
11435 VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS);
11436 }
11437
11438 /* Indicate that we no longer need to VM-exit when the guest is ready to receive NMIs, it is now ready */
11439 hmR0VmxClearNmiWindowExitVmcs(pVCpu);
11440
11441 /* Deliver the pending NMI via hmR0VmxEvaluatePendingEvent() and resume guest execution. */
11442 return VINF_SUCCESS;
11443}
11444
11445
11446/**
11447 * VM-exit handler for WBINVD (VMX_EXIT_WBINVD). Conditional VM-exit.
11448 */
11449HMVMX_EXIT_NSRC_DECL hmR0VmxExitWbinvd(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11450{
11451 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11452 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWbinvd);
11453 return hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11454}
11455
11456
11457/**
11458 * VM-exit handler for INVD (VMX_EXIT_INVD). Unconditional VM-exit.
11459 */
11460HMVMX_EXIT_NSRC_DECL hmR0VmxExitInvd(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11461{
11462 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11463 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvd);
11464 return hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11465}
11466
11467
11468/**
11469 * VM-exit handler for CPUID (VMX_EXIT_CPUID). Unconditional VM-exit.
11470 */
11471HMVMX_EXIT_DECL hmR0VmxExitCpuid(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11472{
11473 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11474 PVM pVM = pVCpu->CTX_SUFF(pVM);
11475 int rc = EMInterpretCpuId(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11476 if (RT_LIKELY(rc == VINF_SUCCESS))
11477 {
11478 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11479 Assert(pVmxTransient->cbInstr == 2);
11480 }
11481 else
11482 {
11483 AssertMsgFailed(("hmR0VmxExitCpuid: EMInterpretCpuId failed with %Rrc\n", rc));
11484 rc = VERR_EM_INTERPRETER;
11485 }
11486 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCpuid);
11487 return rc;
11488}
11489
11490
11491/**
11492 * VM-exit handler for GETSEC (VMX_EXIT_GETSEC). Unconditional VM-exit.
11493 */
11494HMVMX_EXIT_DECL hmR0VmxExitGetsec(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11495{
11496 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11497 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
11498 AssertRCReturn(rc, rc);
11499
11500 if (pMixedCtx->cr4 & X86_CR4_SMXE)
11501 return VINF_EM_RAW_EMULATE_INSTR;
11502
11503 AssertMsgFailed(("hmR0VmxExitGetsec: unexpected VM-exit when CR4.SMXE is 0.\n"));
11504 HMVMX_RETURN_UNEXPECTED_EXIT();
11505}
11506
11507
11508/**
11509 * VM-exit handler for RDTSC (VMX_EXIT_RDTSC). Conditional VM-exit.
11510 */
11511HMVMX_EXIT_DECL hmR0VmxExitRdtsc(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11512{
11513 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11514 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
11515 AssertRCReturn(rc, rc);
11516
11517 PVM pVM = pVCpu->CTX_SUFF(pVM);
11518 rc = EMInterpretRdtsc(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11519 if (RT_LIKELY(rc == VINF_SUCCESS))
11520 {
11521 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11522 Assert(pVmxTransient->cbInstr == 2);
11523 /* If we get a spurious VM-exit when offsetting is enabled, we must reset offsetting on VM-reentry. See @bugref{6634}. */
11524 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING)
11525 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
11526 }
11527 else
11528 rc = VERR_EM_INTERPRETER;
11529 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc);
11530 return rc;
11531}
11532
11533
11534/**
11535 * VM-exit handler for RDTSCP (VMX_EXIT_RDTSCP). Conditional VM-exit.
11536 */
11537HMVMX_EXIT_DECL hmR0VmxExitRdtscp(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11538{
11539 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11540 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
11541 rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx); /* For MSR_K8_TSC_AUX */
11542 AssertRCReturn(rc, rc);
11543
11544 PVM pVM = pVCpu->CTX_SUFF(pVM);
11545 rc = EMInterpretRdtscp(pVM, pVCpu, pMixedCtx);
11546 if (RT_SUCCESS(rc))
11547 {
11548 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11549 Assert(pVmxTransient->cbInstr == 3);
11550 /* If we get a spurious VM-exit when offsetting is enabled, we must reset offsetting on VM-reentry. See @bugref{6634}. */
11551 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TSC_OFFSETTING)
11552 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
11553 }
11554 else
11555 {
11556 AssertMsgFailed(("hmR0VmxExitRdtscp: EMInterpretRdtscp failed with %Rrc\n", rc));
11557 rc = VERR_EM_INTERPRETER;
11558 }
11559 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdtsc);
11560 return rc;
11561}
11562
11563
11564/**
11565 * VM-exit handler for RDPMC (VMX_EXIT_RDPMC). Conditional VM-exit.
11566 */
11567HMVMX_EXIT_DECL hmR0VmxExitRdpmc(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11568{
11569 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11570 int rc = hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
11571 rc |= hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11572 AssertRCReturn(rc, rc);
11573
11574 PVM pVM = pVCpu->CTX_SUFF(pVM);
11575 rc = EMInterpretRdpmc(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11576 if (RT_LIKELY(rc == VINF_SUCCESS))
11577 {
11578 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11579 Assert(pVmxTransient->cbInstr == 2);
11580 }
11581 else
11582 {
11583 AssertMsgFailed(("hmR0VmxExitRdpmc: EMInterpretRdpmc failed with %Rrc\n", rc));
11584 rc = VERR_EM_INTERPRETER;
11585 }
11586 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdpmc);
11587 return rc;
11588}
11589
11590
11591/**
11592 * VM-exit handler for VMCALL (VMX_EXIT_VMCALL). Unconditional VM-exit.
11593 */
11594HMVMX_EXIT_DECL hmR0VmxExitVmcall(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11595{
11596 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11597 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitVmcall);
11598
11599 VBOXSTRICTRC rcStrict = VERR_VMX_IPE_3;
11600 if (pVCpu->hm.s.fHypercallsEnabled)
11601 {
11602#if 0
11603 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11604#else
11605 /* Aggressive state sync. for now. */
11606 int rc = hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
11607 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); /* For long-mode checks in gimKvmHypercall(). */
11608 AssertRCReturn(rc, rc);
11609#endif
11610
11611 /* Perform the hypercall. */
11612 rcStrict = GIMHypercall(pVCpu, pMixedCtx);
11613 if (rcStrict == VINF_SUCCESS)
11614 {
11615 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11616 AssertRCReturn(rc, rc);
11617 }
11618 else
11619 Assert( rcStrict == VINF_GIM_R3_HYPERCALL
11620 || rcStrict == VINF_GIM_HYPERCALL_CONTINUING
11621 || RT_FAILURE(VBOXSTRICTRC_VAL(rcStrict)));
11622
11623 /* If the hypercall changes anything other than guest's general-purpose registers,
11624 we would need to reload the guest changed bits here before VM-entry. */
11625 }
11626 else
11627 Log4(("hmR0VmxExitVmcall: Hypercalls not enabled\n"));
11628
11629 /* If hypercalls are disabled or the hypercall failed for some reason, raise #UD and continue. */
11630 if (RT_FAILURE(VBOXSTRICTRC_VAL(rcStrict)))
11631 {
11632 hmR0VmxSetPendingXcptUD(pVCpu, pMixedCtx);
11633 rcStrict = VINF_SUCCESS;
11634 }
11635
11636 return rcStrict;
11637}
11638
11639
11640/**
11641 * VM-exit handler for INVLPG (VMX_EXIT_INVLPG). Conditional VM-exit.
11642 */
11643HMVMX_EXIT_DECL hmR0VmxExitInvlpg(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11644{
11645 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11646 PVM pVM = pVCpu->CTX_SUFF(pVM);
11647 Assert(!pVM->hm.s.fNestedPaging || pVCpu->hm.s.fUsingDebugLoop);
11648
11649 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
11650 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
11651 AssertRCReturn(rc, rc);
11652
11653 VBOXSTRICTRC rcStrict = EMInterpretInvlpg(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), pVmxTransient->uExitQualification);
11654 if (RT_LIKELY(rcStrict == VINF_SUCCESS))
11655 rcStrict = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11656 else
11657 AssertMsg(rcStrict == VERR_EM_INTERPRETER, ("hmR0VmxExitInvlpg: EMInterpretInvlpg %#RX64 failed with %Rrc\n",
11658 pVmxTransient->uExitQualification, VBOXSTRICTRC_VAL(rcStrict)));
11659 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInvlpg);
11660 return rcStrict;
11661}
11662
11663
11664/**
11665 * VM-exit handler for MONITOR (VMX_EXIT_MONITOR). Conditional VM-exit.
11666 */
11667HMVMX_EXIT_DECL hmR0VmxExitMonitor(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11668{
11669 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11670 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11671 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
11672 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11673 AssertRCReturn(rc, rc);
11674
11675 PVM pVM = pVCpu->CTX_SUFF(pVM);
11676 rc = EMInterpretMonitor(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11677 if (RT_LIKELY(rc == VINF_SUCCESS))
11678 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11679 else
11680 {
11681 AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMonitor: EMInterpretMonitor failed with %Rrc\n", rc));
11682 rc = VERR_EM_INTERPRETER;
11683 }
11684 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMonitor);
11685 return rc;
11686}
11687
11688
11689/**
11690 * VM-exit handler for MWAIT (VMX_EXIT_MWAIT). Conditional VM-exit.
11691 */
11692HMVMX_EXIT_DECL hmR0VmxExitMwait(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11693{
11694 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11695 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
11696 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
11697 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
11698 AssertRCReturn(rc, rc);
11699
11700 PVM pVM = pVCpu->CTX_SUFF(pVM);
11701 VBOXSTRICTRC rc2 = EMInterpretMWait(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
11702 rc = VBOXSTRICTRC_VAL(rc2);
11703 if (RT_LIKELY( rc == VINF_SUCCESS
11704 || rc == VINF_EM_HALT))
11705 {
11706 int rc3 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11707 AssertRCReturn(rc3, rc3);
11708
11709 if ( rc == VINF_EM_HALT
11710 && EMMonitorWaitShouldContinue(pVCpu, pMixedCtx))
11711 {
11712 rc = VINF_SUCCESS;
11713 }
11714 }
11715 else
11716 {
11717 AssertMsg(rc == VERR_EM_INTERPRETER, ("hmR0VmxExitMwait: EMInterpretMWait failed with %Rrc\n", rc));
11718 rc = VERR_EM_INTERPRETER;
11719 }
11720 AssertMsg(rc == VINF_SUCCESS || rc == VINF_EM_HALT || rc == VERR_EM_INTERPRETER,
11721 ("hmR0VmxExitMwait: failed, invalid error code %Rrc\n", rc));
11722 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMwait);
11723 return rc;
11724}
11725
11726
11727/**
11728 * VM-exit handler for RSM (VMX_EXIT_RSM). Unconditional VM-exit.
11729 */
11730HMVMX_EXIT_NSRC_DECL hmR0VmxExitRsm(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11731{
11732 /*
11733 * Execution of RSM outside of SMM mode causes #UD regardless of VMX root or VMX non-root mode. In theory, we should never
11734 * get this VM-exit. This can happen only if dual-monitor treatment of SMI and VMX is enabled, which can (only?) be done by
11735 * executing VMCALL in VMX root operation. If we get here, something funny is going on.
11736 * See Intel spec. "33.15.5 Enabling the Dual-Monitor Treatment".
11737 */
11738 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11739 AssertMsgFailed(("Unexpected RSM VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
11740 HMVMX_RETURN_UNEXPECTED_EXIT();
11741}
11742
11743
11744/**
11745 * VM-exit handler for SMI (VMX_EXIT_SMI). Unconditional VM-exit.
11746 */
11747HMVMX_EXIT_NSRC_DECL hmR0VmxExitSmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11748{
11749 /*
11750 * This can only happen if we support dual-monitor treatment of SMI, which can be activated by executing VMCALL in VMX
11751 * root operation. Only an STM (SMM transfer monitor) would get this VM-exit when we (the executive monitor) execute a VMCALL
11752 * in VMX root mode or receive an SMI. If we get here, something funny is going on.
11753 * See Intel spec. "33.15.6 Activating the Dual-Monitor Treatment" and Intel spec. 25.3 "Other Causes of VM-Exits"
11754 */
11755 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11756 AssertMsgFailed(("Unexpected SMI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
11757 HMVMX_RETURN_UNEXPECTED_EXIT();
11758}
11759
11760
11761/**
11762 * VM-exit handler for IO SMI (VMX_EXIT_IO_SMI). Unconditional VM-exit.
11763 */
11764HMVMX_EXIT_NSRC_DECL hmR0VmxExitIoSmi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11765{
11766 /* Same treatment as VMX_EXIT_SMI. See comment in hmR0VmxExitSmi(). */
11767 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11768 AssertMsgFailed(("Unexpected IO SMI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
11769 HMVMX_RETURN_UNEXPECTED_EXIT();
11770}
11771
11772
11773/**
11774 * VM-exit handler for SIPI (VMX_EXIT_SIPI). Conditional VM-exit.
11775 */
11776HMVMX_EXIT_NSRC_DECL hmR0VmxExitSipi(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11777{
11778 /*
11779 * SIPI exits can only occur in VMX non-root operation when the "wait-for-SIPI" guest activity state is used. We currently
11780 * don't make use of it (see hmR0VmxLoadGuestActivityState()) as our guests don't have direct access to the host LAPIC.
11781 * See Intel spec. 25.3 "Other Causes of VM-exits".
11782 */
11783 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11784 AssertMsgFailed(("Unexpected SIPI VM-exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
11785 HMVMX_RETURN_UNEXPECTED_EXIT();
11786}
11787
11788
11789/**
11790 * VM-exit handler for INIT signal (VMX_EXIT_INIT_SIGNAL). Unconditional
11791 * VM-exit.
11792 */
11793HMVMX_EXIT_NSRC_DECL hmR0VmxExitInitSignal(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11794{
11795 /*
11796 * INIT signals are blocked in VMX root operation by VMXON and by SMI in SMM.
11797 * See Intel spec. 33.14.1 Default Treatment of SMI Delivery" and Intel spec. 29.3 "VMX Instructions" for "VMXON".
11798 *
11799 * It is -NOT- blocked in VMX non-root operation so we can, in theory, still get these VM-exits.
11800 * See Intel spec. "23.8 Restrictions on VMX operation".
11801 */
11802 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11803 return VINF_SUCCESS;
11804}
11805
11806
11807/**
11808 * VM-exit handler for triple faults (VMX_EXIT_TRIPLE_FAULT). Unconditional
11809 * VM-exit.
11810 */
11811HMVMX_EXIT_DECL hmR0VmxExitTripleFault(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11812{
11813 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11814 return VINF_EM_RESET;
11815}
11816
11817
11818/**
11819 * VM-exit handler for HLT (VMX_EXIT_HLT). Conditional VM-exit.
11820 */
11821HMVMX_EXIT_DECL hmR0VmxExitHlt(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11822{
11823 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11824 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_HLT_EXIT);
11825
11826 int rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
11827 AssertRCReturn(rc, rc);
11828
11829 if (EMShouldContinueAfterHalt(pVCpu, pMixedCtx)) /* Requires eflags. */
11830 rc = VINF_SUCCESS;
11831 else
11832 rc = VINF_EM_HALT;
11833
11834 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt);
11835 if (rc != VINF_SUCCESS)
11836 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchHltToR3);
11837 return rc;
11838}
11839
11840
11841/**
11842 * VM-exit handler for instructions that result in a \#UD exception delivered to
11843 * the guest.
11844 */
11845HMVMX_EXIT_NSRC_DECL hmR0VmxExitSetPendingXcptUD(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11846{
11847 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11848 hmR0VmxSetPendingXcptUD(pVCpu, pMixedCtx);
11849 return VINF_SUCCESS;
11850}
11851
11852
11853/**
11854 * VM-exit handler for expiry of the VMX preemption timer.
11855 */
11856HMVMX_EXIT_DECL hmR0VmxExitPreemptTimer(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11857{
11858 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11859
11860 /* If the preemption-timer has expired, reinitialize the preemption timer on next VM-entry. */
11861 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
11862
11863 /* If there are any timer events pending, fall back to ring-3, otherwise resume guest execution. */
11864 PVM pVM = pVCpu->CTX_SUFF(pVM);
11865 bool fTimersPending = TMTimerPollBool(pVM, pVCpu);
11866 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPreemptTimer);
11867 return fTimersPending ? VINF_EM_RAW_TIMER_PENDING : VINF_SUCCESS;
11868}
11869
11870
11871/**
11872 * VM-exit handler for XSETBV (VMX_EXIT_XSETBV). Unconditional VM-exit.
11873 */
11874HMVMX_EXIT_DECL hmR0VmxExitXsetbv(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11875{
11876 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11877
11878 int rc = hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
11879 rc |= hmR0VmxSaveGuestRegsForIemExec(pVCpu, pMixedCtx, false /*fMemory*/, false /*fNeedRsp*/);
11880 rc |= hmR0VmxSaveGuestCR4(pVCpu, pMixedCtx);
11881 AssertRCReturn(rc, rc);
11882
11883 VBOXSTRICTRC rcStrict = IEMExecDecodedXsetbv(pVCpu, pVmxTransient->cbInstr);
11884 HMCPU_CF_SET(pVCpu, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS : HM_CHANGED_ALL_GUEST);
11885
11886 pVCpu->hm.s.fLoadSaveGuestXcr0 = (pMixedCtx->cr4 & X86_CR4_OSXSAVE) && pMixedCtx->aXcr[0] != ASMGetXcr0();
11887
11888 return rcStrict;
11889}
11890
11891
11892/**
11893 * VM-exit handler for INVPCID (VMX_EXIT_INVPCID). Conditional VM-exit.
11894 */
11895HMVMX_EXIT_DECL hmR0VmxExitInvpcid(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11896{
11897 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
11898
11899 /* The guest should not invalidate the host CPU's TLBs, fallback to interpreter. */
11900 /** @todo implement EMInterpretInvpcid() */
11901 return VERR_EM_INTERPRETER;
11902}
11903
11904
11905/**
11906 * VM-exit handler for invalid-guest-state (VMX_EXIT_ERR_INVALID_GUEST_STATE).
11907 * Error VM-exit.
11908 */
11909HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrInvalidGuestState(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11910{
11911 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
11912 AssertRCReturn(rc, rc);
11913
11914 rc = hmR0VmxCheckVmcsCtls(pVCpu);
11915 AssertRCReturn(rc, rc);
11916
11917 uint32_t uInvalidReason = hmR0VmxCheckGuestState(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx);
11918 NOREF(uInvalidReason);
11919
11920#ifdef VBOX_STRICT
11921 uint32_t uIntrState;
11922 RTHCUINTREG uHCReg;
11923 uint64_t u64Val;
11924 uint32_t u32Val;
11925
11926 rc = hmR0VmxReadEntryIntInfoVmcs(pVmxTransient);
11927 rc |= hmR0VmxReadEntryXcptErrorCodeVmcs(pVmxTransient);
11928 rc |= hmR0VmxReadEntryInstrLenVmcs(pVmxTransient);
11929 rc |= VMXReadVmcs32(VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE, &uIntrState);
11930 AssertRCReturn(rc, rc);
11931
11932 Log4(("uInvalidReason %u\n", uInvalidReason));
11933 Log4(("VMX_VMCS32_CTRL_ENTRY_INTERRUPTION_INFO %#RX32\n", pVmxTransient->uEntryIntInfo));
11934 Log4(("VMX_VMCS32_CTRL_ENTRY_EXCEPTION_ERRCODE %#RX32\n", pVmxTransient->uEntryXcptErrorCode));
11935 Log4(("VMX_VMCS32_CTRL_ENTRY_INSTR_LENGTH %#RX32\n", pVmxTransient->cbEntryInstr));
11936 Log4(("VMX_VMCS32_GUEST_INTERRUPTIBILITY_STATE %#RX32\n", uIntrState));
11937
11938 rc = VMXReadVmcs32(VMX_VMCS_GUEST_CR0, &u32Val); AssertRC(rc);
11939 Log4(("VMX_VMCS_GUEST_CR0 %#RX32\n", u32Val));
11940 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_MASK, &uHCReg); AssertRC(rc);
11941 Log4(("VMX_VMCS_CTRL_CR0_MASK %#RHr\n", uHCReg));
11942 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR0_READ_SHADOW, &uHCReg); AssertRC(rc);
11943 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
11944 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_MASK, &uHCReg); AssertRC(rc);
11945 Log4(("VMX_VMCS_CTRL_CR4_MASK %#RHr\n", uHCReg));
11946 rc = VMXReadVmcsHstN(VMX_VMCS_CTRL_CR4_READ_SHADOW, &uHCReg); AssertRC(rc);
11947 Log4(("VMX_VMCS_CTRL_CR4_READ_SHADOW %#RHr\n", uHCReg));
11948 rc = VMXReadVmcs64(VMX_VMCS64_CTRL_EPTP_FULL, &u64Val); AssertRC(rc);
11949 Log4(("VMX_VMCS64_CTRL_EPTP_FULL %#RX64\n", u64Val));
11950#else
11951 NOREF(pVmxTransient);
11952#endif
11953
11954 HMDumpRegs(pVCpu->CTX_SUFF(pVM), pVCpu, pMixedCtx);
11955 return VERR_VMX_INVALID_GUEST_STATE;
11956}
11957
11958
11959/**
11960 * VM-exit handler for VM-entry failure due to an MSR-load
11961 * (VMX_EXIT_ERR_MSR_LOAD). Error VM-exit.
11962 */
11963HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrMsrLoad(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11964{
11965 NOREF(pVmxTransient);
11966 AssertMsgFailed(("Unexpected MSR-load exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); NOREF(pMixedCtx);
11967 HMVMX_RETURN_UNEXPECTED_EXIT();
11968}
11969
11970
11971/**
11972 * VM-exit handler for VM-entry failure due to a machine-check event
11973 * (VMX_EXIT_ERR_MACHINE_CHECK). Error VM-exit.
11974 */
11975HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrMachineCheck(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11976{
11977 NOREF(pVmxTransient);
11978 AssertMsgFailed(("Unexpected machine-check event exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx)); NOREF(pMixedCtx);
11979 HMVMX_RETURN_UNEXPECTED_EXIT();
11980}
11981
11982
11983/**
11984 * VM-exit handler for all undefined reasons. Should never ever happen.. in
11985 * theory.
11986 */
11987HMVMX_EXIT_NSRC_DECL hmR0VmxExitErrUndefined(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
11988{
11989 AssertMsgFailed(("Huh!? Undefined VM-exit reason %d. pVCpu=%p pMixedCtx=%p\n", pVmxTransient->uExitReason, pVCpu, pMixedCtx));
11990 NOREF(pVCpu); NOREF(pMixedCtx); NOREF(pVmxTransient);
11991 return VERR_VMX_UNDEFINED_EXIT_CODE;
11992}
11993
11994
11995/**
11996 * VM-exit handler for XDTR (LGDT, SGDT, LIDT, SIDT) accesses
11997 * (VMX_EXIT_XDTR_ACCESS) and LDT and TR access (LLDT, LTR, SLDT, STR).
11998 * Conditional VM-exit.
11999 */
12000HMVMX_EXIT_DECL hmR0VmxExitXdtrAccess(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12001{
12002 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12003
12004 /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT. */
12005 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitXdtrAccess);
12006 if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_DESCRIPTOR_TABLE_EXIT)
12007 return VERR_EM_INTERPRETER;
12008 AssertMsgFailed(("Unexpected XDTR access. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
12009 HMVMX_RETURN_UNEXPECTED_EXIT();
12010}
12011
12012
12013/**
12014 * VM-exit handler for RDRAND (VMX_EXIT_RDRAND). Conditional VM-exit.
12015 */
12016HMVMX_EXIT_DECL hmR0VmxExitRdrand(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12017{
12018 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12019
12020 /* By default, we don't enable VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT. */
12021 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdrand);
12022 if (pVCpu->hm.s.vmx.u32ProcCtls2 & VMX_VMCS_CTRL_PROC_EXEC2_RDRAND_EXIT)
12023 return VERR_EM_INTERPRETER;
12024 AssertMsgFailed(("Unexpected RDRAND exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
12025 HMVMX_RETURN_UNEXPECTED_EXIT();
12026}
12027
12028
12029/**
12030 * VM-exit handler for RDMSR (VMX_EXIT_RDMSR).
12031 */
12032HMVMX_EXIT_DECL hmR0VmxExitRdmsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12033{
12034 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12035
12036 /* EMInterpretRdmsr() requires CR0, Eflags and SS segment register. */
12037 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
12038 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
12039 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12040 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
12041 {
12042 rc |= hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
12043 rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
12044 }
12045 AssertRCReturn(rc, rc);
12046 Log4(("ecx=%#RX32\n", pMixedCtx->ecx));
12047
12048#ifdef VBOX_STRICT
12049 if (pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS)
12050 {
12051 if ( hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx)
12052 && pMixedCtx->ecx != MSR_K6_EFER)
12053 {
12054 AssertMsgFailed(("Unexpected RDMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
12055 pMixedCtx->ecx));
12056 HMVMX_RETURN_UNEXPECTED_EXIT();
12057 }
12058 if (hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
12059 {
12060 VMXMSREXITREAD enmRead;
12061 VMXMSREXITWRITE enmWrite;
12062 int rc2 = hmR0VmxGetMsrPermission(pVCpu, pMixedCtx->ecx, &enmRead, &enmWrite);
12063 AssertRCReturn(rc2, rc2);
12064 if (enmRead == VMXMSREXIT_PASSTHRU_READ)
12065 {
12066 AssertMsgFailed(("Unexpected RDMSR for a passthru lazy-restore MSR. ecx=%#RX32\n", pMixedCtx->ecx));
12067 HMVMX_RETURN_UNEXPECTED_EXIT();
12068 }
12069 }
12070 }
12071#endif
12072
12073 PVM pVM = pVCpu->CTX_SUFF(pVM);
12074 rc = EMInterpretRdmsr(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
12075 AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER,
12076 ("hmR0VmxExitRdmsr: failed, invalid error code %Rrc\n", rc));
12077 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitRdmsr);
12078 if (RT_SUCCESS(rc))
12079 {
12080 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
12081 Assert(pVmxTransient->cbInstr == 2);
12082 }
12083 return rc;
12084}
12085
12086
12087/**
12088 * VM-exit handler for WRMSR (VMX_EXIT_WRMSR).
12089 */
12090HMVMX_EXIT_DECL hmR0VmxExitWrmsr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12091{
12092 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12093 PVM pVM = pVCpu->CTX_SUFF(pVM);
12094 int rc = VINF_SUCCESS;
12095
12096 /* EMInterpretWrmsr() requires CR0, EFLAGS and SS segment register. */
12097 rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
12098 rc |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx);
12099 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12100 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
12101 {
12102 rc |= hmR0VmxSaveGuestLazyMsrs(pVCpu, pMixedCtx);
12103 rc |= hmR0VmxSaveGuestAutoLoadStoreMsrs(pVCpu, pMixedCtx);
12104 }
12105 AssertRCReturn(rc, rc);
12106 Log4(("ecx=%#RX32 edx:eax=%#RX32:%#RX32\n", pMixedCtx->ecx, pMixedCtx->edx, pMixedCtx->eax));
12107
12108 rc = EMInterpretWrmsr(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
12109 AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER, ("hmR0VmxExitWrmsr: failed, invalid error code %Rrc\n", rc));
12110 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitWrmsr);
12111
12112 if (RT_SUCCESS(rc))
12113 {
12114 rc = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
12115
12116 /* If this is an X2APIC WRMSR access, update the APIC state as well. */
12117 if ( pMixedCtx->ecx == MSR_IA32_APICBASE
12118 || ( pMixedCtx->ecx >= MSR_IA32_X2APIC_START
12119 && pMixedCtx->ecx <= MSR_IA32_X2APIC_END))
12120 {
12121 /* We've already saved the APIC related guest-state (TPR) in hmR0VmxPostRunGuest(). When full APIC register
12122 * virtualization is implemented we'll have to make sure APIC state is saved from the VMCS before
12123 EMInterpretWrmsr() changes it. */
12124 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
12125 }
12126 else if (pMixedCtx->ecx == MSR_IA32_TSC) /* Windows 7 does this during bootup. See @bugref{6398}. */
12127 pVmxTransient->fUpdateTscOffsettingAndPreemptTimer = true;
12128 else if (pMixedCtx->ecx == MSR_K6_EFER)
12129 {
12130 /*
12131 * If the guest touches EFER we need to update the VM-Entry and VM-Exit controls as well,
12132 * even if it is -not- touching bits that cause paging mode changes (LMA/LME). We care about
12133 * the other bits as well, SCE and NXE. See @bugref{7368}.
12134 */
12135 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_EFER_MSR | HM_CHANGED_VMX_ENTRY_CTLS | HM_CHANGED_VMX_EXIT_CTLS);
12136 }
12137
12138 /* Update MSRs that are part of the VMCS and auto-load/store area when MSR-bitmaps are not supported. */
12139 if (!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_MSR_BITMAPS))
12140 {
12141 switch (pMixedCtx->ecx)
12142 {
12143 case MSR_IA32_SYSENTER_CS: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_CS_MSR); break;
12144 case MSR_IA32_SYSENTER_EIP: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_EIP_MSR); break;
12145 case MSR_IA32_SYSENTER_ESP: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SYSENTER_ESP_MSR); break;
12146 case MSR_K8_FS_BASE: /* no break */
12147 case MSR_K8_GS_BASE: HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_SEGMENT_REGS); break;
12148 case MSR_K6_EFER: /* already handled above */ break;
12149 default:
12150 {
12151 if (hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx))
12152 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_AUTO_MSRS);
12153 else if (hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
12154 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_LAZY_MSRS);
12155 break;
12156 }
12157 }
12158 }
12159#ifdef VBOX_STRICT
12160 else
12161 {
12162 /* Paranoia. Validate that MSRs in the MSR-bitmaps with write-passthru are not intercepted. */
12163 switch (pMixedCtx->ecx)
12164 {
12165 case MSR_IA32_SYSENTER_CS:
12166 case MSR_IA32_SYSENTER_EIP:
12167 case MSR_IA32_SYSENTER_ESP:
12168 case MSR_K8_FS_BASE:
12169 case MSR_K8_GS_BASE:
12170 {
12171 AssertMsgFailed(("Unexpected WRMSR for an MSR in the VMCS. ecx=%#RX32\n", pMixedCtx->ecx));
12172 HMVMX_RETURN_UNEXPECTED_EXIT();
12173 }
12174
12175 /* Writes to MSRs in auto-load/store area/swapped MSRs, shouldn't cause VM-exits with MSR-bitmaps. */
12176 default:
12177 {
12178 if (hmR0VmxIsAutoLoadStoreGuestMsr(pVCpu, pMixedCtx->ecx))
12179 {
12180 /* EFER writes are always intercepted, see hmR0VmxLoadGuestMsrs(). */
12181 if (pMixedCtx->ecx != MSR_K6_EFER)
12182 {
12183 AssertMsgFailed(("Unexpected WRMSR for an MSR in the auto-load/store area in the VMCS. ecx=%#RX32\n",
12184 pMixedCtx->ecx));
12185 HMVMX_RETURN_UNEXPECTED_EXIT();
12186 }
12187 }
12188
12189 if (hmR0VmxIsLazyGuestMsr(pVCpu, pMixedCtx->ecx))
12190 {
12191 VMXMSREXITREAD enmRead;
12192 VMXMSREXITWRITE enmWrite;
12193 int rc2 = hmR0VmxGetMsrPermission(pVCpu, pMixedCtx->ecx, &enmRead, &enmWrite);
12194 AssertRCReturn(rc2, rc2);
12195 if (enmWrite == VMXMSREXIT_PASSTHRU_WRITE)
12196 {
12197 AssertMsgFailed(("Unexpected WRMSR for passthru, lazy-restore MSR. ecx=%#RX32\n", pMixedCtx->ecx));
12198 HMVMX_RETURN_UNEXPECTED_EXIT();
12199 }
12200 }
12201 break;
12202 }
12203 }
12204 }
12205#endif /* VBOX_STRICT */
12206 }
12207 return rc;
12208}
12209
12210
12211/**
12212 * VM-exit handler for PAUSE (VMX_EXIT_PAUSE). Conditional VM-exit.
12213 */
12214HMVMX_EXIT_DECL hmR0VmxExitPause(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12215{
12216 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12217
12218 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPause);
12219 return VINF_EM_RAW_INTERRUPT;
12220}
12221
12222
12223/**
12224 * VM-exit handler for when the TPR value is lowered below the specified
12225 * threshold (VMX_EXIT_TPR_BELOW_THRESHOLD). Conditional VM-exit.
12226 */
12227HMVMX_EXIT_NSRC_DECL hmR0VmxExitTprBelowThreshold(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12228{
12229 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12230 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW);
12231
12232 /*
12233 * The TPR shadow would've been synced with the APIC TPR in hmR0VmxPostRunGuest(). We'll re-evaluate
12234 * pending interrupts and inject them before the next VM-entry so we can just continue execution here.
12235 */
12236 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTprBelowThreshold);
12237 return VINF_SUCCESS;
12238}
12239
12240
12241/**
12242 * VM-exit handler for control-register accesses (VMX_EXIT_MOV_CRX). Conditional
12243 * VM-exit.
12244 *
12245 * @retval VINF_SUCCESS when guest execution can continue.
12246 * @retval VINF_PGM_CHANGE_MODE when shadow paging mode changed, back to ring-3.
12247 * @retval VINF_PGM_SYNC_CR3 CR3 sync is required, back to ring-3.
12248 * @retval VERR_EM_INTERPRETER when something unexpected happened, fallback to
12249 * interpreter.
12250 */
12251HMVMX_EXIT_DECL hmR0VmxExitMovCRx(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12252{
12253 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12254 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitMovCRx, y2);
12255 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12256 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
12257 AssertRCReturn(rc, rc);
12258
12259 RTGCUINTPTR const uExitQualification = pVmxTransient->uExitQualification;
12260 uint32_t const uAccessType = VMX_EXIT_QUALIFICATION_CRX_ACCESS(uExitQualification);
12261 PVM pVM = pVCpu->CTX_SUFF(pVM);
12262 VBOXSTRICTRC rcStrict;
12263 rc = hmR0VmxSaveGuestRegsForIemExec(pVCpu, pMixedCtx, false /*fMemory*/, true /*fNeedRsp*/);
12264 switch (uAccessType)
12265 {
12266 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_WRITE: /* MOV to CRx */
12267 {
12268 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
12269 AssertRCReturn(rc, rc);
12270
12271 rcStrict = IEMExecDecodedMovCRxWrite(pVCpu, pVmxTransient->cbInstr,
12272 VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification),
12273 VMX_EXIT_QUALIFICATION_CRX_GENREG(uExitQualification));
12274 AssertMsg( rcStrict == VINF_SUCCESS || rcStrict == VINF_IEM_RAISED_XCPT || rcStrict == VINF_PGM_CHANGE_MODE
12275 || rcStrict == VINF_PGM_SYNC_CR3, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
12276 switch (VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification))
12277 {
12278 case 0: /* CR0 */
12279 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
12280 Log4(("CRX CR0 write rcStrict=%Rrc CR0=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->cr0));
12281 break;
12282 case 2: /* CR2 */
12283 /* Nothing to do here, CR2 it's not part of the VMCS. */
12284 break;
12285 case 3: /* CR3 */
12286 Assert(!pVM->hm.s.fNestedPaging || !CPUMIsGuestPagingEnabledEx(pMixedCtx) || pVCpu->hm.s.fUsingDebugLoop);
12287 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR3);
12288 Log4(("CRX CR3 write rcStrict=%Rrc CR3=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->cr3));
12289 break;
12290 case 4: /* CR4 */
12291 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR4);
12292 Log4(("CRX CR4 write rc=%Rrc CR4=%#RX64 fLoadSaveGuestXcr0=%u\n",
12293 VBOXSTRICTRC_VAL(rcStrict), pMixedCtx->cr4, pVCpu->hm.s.fLoadSaveGuestXcr0));
12294 break;
12295 case 8: /* CR8 */
12296 Assert(!(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW));
12297 /* CR8 contains the APIC TPR. Was updated by IEMExecDecodedMovCRxWrite(). */
12298 HMCPU_CF_SET(pVCpu, HM_CHANGED_VMX_GUEST_APIC_STATE);
12299 break;
12300 default:
12301 AssertMsgFailed(("Invalid CRx register %#x\n", VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)));
12302 break;
12303 }
12304
12305 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxWrite[VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)]);
12306 break;
12307 }
12308
12309 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_READ: /* MOV from CRx */
12310 {
12311 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
12312 AssertRCReturn(rc, rc);
12313
12314 Assert( !pVM->hm.s.fNestedPaging
12315 || !CPUMIsGuestPagingEnabledEx(pMixedCtx)
12316 || pVCpu->hm.s.fUsingDebugLoop
12317 || VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification) != 3);
12318
12319 /* CR8 reads only cause a VM-exit when the TPR shadow feature isn't enabled. */
12320 Assert( VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification) != 8
12321 || !(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW));
12322
12323 rcStrict = IEMExecDecodedMovCRxRead(pVCpu, pVmxTransient->cbInstr,
12324 VMX_EXIT_QUALIFICATION_CRX_GENREG(uExitQualification),
12325 VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification));
12326 AssertMsg(rcStrict == VINF_SUCCESS || rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
12327 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCRxRead[VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification)]);
12328 Log4(("CRX CR%d Read access rcStrict=%Rrc\n", VMX_EXIT_QUALIFICATION_CRX_REGISTER(uExitQualification),
12329 VBOXSTRICTRC_VAL(rcStrict)));
12330 break;
12331 }
12332
12333 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_CLTS: /* CLTS (Clear Task-Switch Flag in CR0) */
12334 {
12335 AssertRCReturn(rc, rc);
12336 rcStrict = IEMExecDecodedClts(pVCpu, pVmxTransient->cbInstr);
12337 AssertMsg(rcStrict == VINF_SUCCESS || rcStrict == VINF_IEM_RAISED_XCPT, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
12338 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
12339 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitClts);
12340 Log4(("CRX CLTS rcStrict=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
12341 break;
12342 }
12343
12344 case VMX_EXIT_QUALIFICATION_CRX_ACCESS_LMSW: /* LMSW (Load Machine-Status Word into CR0) */
12345 {
12346 AssertRCReturn(rc, rc);
12347 rcStrict = IEMExecDecodedLmsw(pVCpu, pVmxTransient->cbInstr,
12348 VMX_EXIT_QUALIFICATION_CRX_LMSW_DATA(uExitQualification));
12349 AssertMsg(rcStrict == VINF_SUCCESS || rcStrict == VINF_IEM_RAISED_XCPT || rcStrict == VINF_PGM_CHANGE_MODE,
12350 ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
12351 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitLmsw);
12352 Log4(("CRX LMSW rcStrict=%d\n", VBOXSTRICTRC_VAL(rcStrict)));
12353 break;
12354 }
12355
12356 default:
12357 AssertMsgFailedReturn(("Invalid access-type in Mov CRx VM-exit qualification %#x\n", uAccessType),
12358 VERR_VMX_UNEXPECTED_EXCEPTION);
12359 }
12360
12361 HMCPU_CF_SET(pVCpu, rcStrict != VINF_IEM_RAISED_XCPT ? HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS : HM_CHANGED_ALL_GUEST);
12362 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitMovCRx, y2);
12363 NOREF(pVM);
12364 return rcStrict;
12365}
12366
12367
12368/**
12369 * VM-exit handler for I/O instructions (VMX_EXIT_IO_INSTR). Conditional
12370 * VM-exit.
12371 */
12372HMVMX_EXIT_DECL hmR0VmxExitIoInstr(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12373{
12374 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12375 STAM_PROFILE_ADV_START(&pVCpu->hm.s.StatExitIO, y1);
12376
12377 int rc2 = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12378 rc2 |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
12379 rc2 |= hmR0VmxSaveGuestRip(pVCpu, pMixedCtx);
12380 rc2 |= hmR0VmxSaveGuestRflags(pVCpu, pMixedCtx); /* Eflag checks in EMInterpretDisasCurrent(). */
12381 rc2 |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx); /* CR0 checks & PGM* in EMInterpretDisasCurrent(). */
12382 rc2 |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx); /* SELM checks in EMInterpretDisasCurrent(). */
12383 /* EFER also required for longmode checks in EMInterpretDisasCurrent(), but it's always up-to-date. */
12384 AssertRCReturn(rc2, rc2);
12385
12386 /* Refer Intel spec. 27-5. "Exit Qualifications for I/O Instructions" for the format. */
12387 uint32_t uIOPort = VMX_EXIT_QUALIFICATION_IO_PORT(pVmxTransient->uExitQualification);
12388 uint8_t uIOWidth = VMX_EXIT_QUALIFICATION_IO_WIDTH(pVmxTransient->uExitQualification);
12389 bool fIOWrite = ( VMX_EXIT_QUALIFICATION_IO_DIRECTION(pVmxTransient->uExitQualification)
12390 == VMX_EXIT_QUALIFICATION_IO_DIRECTION_OUT);
12391 bool fIOString = VMX_EXIT_QUALIFICATION_IO_IS_STRING(pVmxTransient->uExitQualification);
12392 bool fGstStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
12393 bool fDbgStepping = pVCpu->hm.s.fSingleInstruction;
12394 AssertReturn(uIOWidth <= 3 && uIOWidth != 2, VERR_VMX_IPE_1);
12395
12396 /* I/O operation lookup arrays. */
12397 static uint32_t const s_aIOSizes[4] = { 1, 2, 0, 4 }; /* Size of the I/O accesses. */
12398 static uint32_t const s_aIOOpAnd[4] = { 0xff, 0xffff, 0, 0xffffffff }; /* AND masks for saving the result (in AL/AX/EAX). */
12399
12400 VBOXSTRICTRC rcStrict;
12401 uint32_t const cbValue = s_aIOSizes[uIOWidth];
12402 uint32_t const cbInstr = pVmxTransient->cbInstr;
12403 bool fUpdateRipAlready = false; /* ugly hack, should be temporary. */
12404 PVM pVM = pVCpu->CTX_SUFF(pVM);
12405 if (fIOString)
12406 {
12407#ifdef VBOX_WITH_2ND_IEM_STEP /* This used to gurus with debian 32-bit guest without NP (on ATA reads).
12408 See @bugref{5752#c158}. Should work now. */
12409 /*
12410 * INS/OUTS - I/O String instruction.
12411 *
12412 * Use instruction-information if available, otherwise fall back on
12413 * interpreting the instruction.
12414 */
12415 Log4(("CS:RIP=%04x:%08RX64 %#06x/%u %c str\n", pMixedCtx->cs.Sel, pMixedCtx->rip, uIOPort, cbValue,
12416 fIOWrite ? 'w' : 'r'));
12417 AssertReturn(pMixedCtx->dx == uIOPort, VERR_VMX_IPE_2);
12418 if (MSR_IA32_VMX_BASIC_INFO_VMCS_INS_OUTS(pVM->hm.s.vmx.Msrs.u64BasicInfo))
12419 {
12420 rc2 = hmR0VmxReadExitInstrInfoVmcs(pVmxTransient);
12421 /** @todo optimize this, IEM should request the additional state if it needs it (GP, PF, ++). */
12422 rc2 |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
12423 AssertRCReturn(rc2, rc2);
12424 AssertReturn(pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize <= 2, VERR_VMX_IPE_3);
12425 AssertCompile(IEMMODE_16BIT == 0 && IEMMODE_32BIT == 1 && IEMMODE_64BIT == 2);
12426 IEMMODE enmAddrMode = (IEMMODE)pVmxTransient->ExitInstrInfo.StrIo.u3AddrSize;
12427 bool fRep = VMX_EXIT_QUALIFICATION_IO_IS_REP(pVmxTransient->uExitQualification);
12428 if (fIOWrite)
12429 {
12430 rcStrict = IEMExecStringIoWrite(pVCpu, cbValue, enmAddrMode, fRep, cbInstr,
12431 pVmxTransient->ExitInstrInfo.StrIo.iSegReg, true /*fIoChecked*/);
12432 }
12433 else
12434 {
12435 /*
12436 * The segment prefix for INS cannot be overridden and is always ES. We can safely assume X86_SREG_ES.
12437 * Hence "iSegReg" field is undefined in the instruction-information field in VT-x for INS.
12438 * See Intel Instruction spec. for "INS".
12439 * See Intel spec. Table 27-8 "Format of the VM-Exit Instruction-Information Field as Used for INS and OUTS".
12440 */
12441 rcStrict = IEMExecStringIoRead(pVCpu, cbValue, enmAddrMode, fRep, cbInstr, true /*fIoChecked*/);
12442 }
12443 }
12444 else
12445 {
12446 /** @todo optimize this, IEM should request the additional state if it needs it (GP, PF, ++). */
12447 rc2 = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
12448 AssertRCReturn(rc2, rc2);
12449 rcStrict = IEMExecOne(pVCpu);
12450 }
12451 /** @todo IEM needs to be setting these flags somehow. */
12452 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
12453 fUpdateRipAlready = true;
12454#else
12455 PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
12456 rcStrict = EMInterpretDisasCurrent(pVM, pVCpu, pDis, NULL /* pcbInstr */);
12457 if (RT_SUCCESS(rcStrict))
12458 {
12459 if (fIOWrite)
12460 {
12461 rcStrict = IOMInterpretOUTSEx(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), uIOPort, pDis->fPrefix,
12462 (DISCPUMODE)pDis->uAddrMode, cbValue);
12463 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringWrite);
12464 }
12465 else
12466 {
12467 rcStrict = IOMInterpretINSEx(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx), uIOPort, pDis->fPrefix,
12468 (DISCPUMODE)pDis->uAddrMode, cbValue);
12469 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOStringRead);
12470 }
12471 }
12472 else
12473 {
12474 AssertMsg(rcStrict == VERR_EM_INTERPRETER, ("rcStrict=%Rrc RIP=%#RX64\n", VBOXSTRICTRC_VAL(rcStrict),
12475 pMixedCtx->rip));
12476 rcStrict = VINF_EM_RAW_EMULATE_INSTR;
12477 }
12478#endif
12479 }
12480 else
12481 {
12482 /*
12483 * IN/OUT - I/O instruction.
12484 */
12485 Log4(("CS:RIP=%04x:%08RX64 %#06x/%u %c\n", pMixedCtx->cs.Sel, pMixedCtx->rip, uIOPort, cbValue, fIOWrite ? 'w' : 'r'));
12486 uint32_t const uAndVal = s_aIOOpAnd[uIOWidth];
12487 Assert(!VMX_EXIT_QUALIFICATION_IO_IS_REP(pVmxTransient->uExitQualification));
12488 if (fIOWrite)
12489 {
12490 rcStrict = IOMIOPortWrite(pVM, pVCpu, uIOPort, pMixedCtx->eax & uAndVal, cbValue);
12491 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIOWrite);
12492 }
12493 else
12494 {
12495 uint32_t u32Result = 0;
12496 rcStrict = IOMIOPortRead(pVM, pVCpu, uIOPort, &u32Result, cbValue);
12497 if (IOM_SUCCESS(rcStrict))
12498 {
12499 /* Save result of I/O IN instr. in AL/AX/EAX. */
12500 pMixedCtx->eax = (pMixedCtx->eax & ~uAndVal) | (u32Result & uAndVal);
12501 }
12502 else if (rcStrict == VINF_IOM_R3_IOPORT_READ)
12503 HMR0SavePendingIOPortRead(pVCpu, pMixedCtx->rip, pMixedCtx->rip + cbInstr, uIOPort, uAndVal, cbValue);
12504 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIORead);
12505 }
12506 }
12507
12508 if (IOM_SUCCESS(rcStrict))
12509 {
12510 if (!fUpdateRipAlready)
12511 {
12512 hmR0VmxAdvanceGuestRipBy(pVCpu, pMixedCtx, cbInstr);
12513 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP);
12514 }
12515
12516 /*
12517 * INS/OUTS with REP prefix updates RFLAGS, can be observed with triple-fault guru while booting Fedora 17 64-bit guest.
12518 * See Intel Instruction reference for REP/REPE/REPZ/REPNE/REPNZ.
12519 */
12520 if (fIOString)
12521 {
12522 /** @todo Single-step for INS/OUTS with REP prefix? */
12523 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
12524 }
12525 else if ( !fDbgStepping
12526 && fGstStepping)
12527 {
12528 hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
12529 }
12530
12531 /*
12532 * If any I/O breakpoints are armed, we need to check if one triggered
12533 * and take appropriate action.
12534 * Note that the I/O breakpoint type is undefined if CR4.DE is 0.
12535 */
12536 rc2 = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
12537 AssertRCReturn(rc2, rc2);
12538
12539 /** @todo Optimize away the DBGFBpIsHwIoArmed call by having DBGF tell the
12540 * execution engines about whether hyper BPs and such are pending. */
12541 uint32_t const uDr7 = pMixedCtx->dr[7];
12542 if (RT_UNLIKELY( ( (uDr7 & X86_DR7_ENABLED_MASK)
12543 && X86_DR7_ANY_RW_IO(uDr7)
12544 && (pMixedCtx->cr4 & X86_CR4_DE))
12545 || DBGFBpIsHwIoArmed(pVM)))
12546 {
12547 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxIoCheck);
12548
12549 /* We're playing with the host CPU state here, make sure we don't preempt or longjmp. */
12550 VMMRZCallRing3Disable(pVCpu);
12551 HM_DISABLE_PREEMPT();
12552
12553 bool fIsGuestDbgActive = CPUMR0DebugStateMaybeSaveGuest(pVCpu, true /* fDr6 */);
12554
12555 VBOXSTRICTRC rcStrict2 = DBGFBpCheckIo(pVM, pVCpu, pMixedCtx, uIOPort, cbValue);
12556 if (rcStrict2 == VINF_EM_RAW_GUEST_TRAP)
12557 {
12558 /* Raise #DB. */
12559 if (fIsGuestDbgActive)
12560 ASMSetDR6(pMixedCtx->dr[6]);
12561 if (pMixedCtx->dr[7] != uDr7)
12562 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
12563
12564 hmR0VmxSetPendingXcptDB(pVCpu, pMixedCtx);
12565 }
12566 /* rcStrict is VINF_SUCCESS, VINF_IOM_R3_IOPORT_COMMIT_WRITE, or in [VINF_EM_FIRST..VINF_EM_LAST],
12567 however we can ditch VINF_IOM_R3_IOPORT_COMMIT_WRITE as it has VMCPU_FF_IOM as backup. */
12568 else if ( rcStrict2 != VINF_SUCCESS
12569 && (rcStrict == VINF_SUCCESS || rcStrict2 < rcStrict))
12570 rcStrict = rcStrict2;
12571 AssertCompile(VINF_EM_LAST < VINF_IOM_R3_IOPORT_COMMIT_WRITE);
12572
12573 HM_RESTORE_PREEMPT();
12574 VMMRZCallRing3Enable(pVCpu);
12575 }
12576 }
12577
12578#ifdef VBOX_STRICT
12579 if (rcStrict == VINF_IOM_R3_IOPORT_READ)
12580 Assert(!fIOWrite);
12581 else if (rcStrict == VINF_IOM_R3_IOPORT_WRITE || rcStrict == VINF_IOM_R3_IOPORT_COMMIT_WRITE)
12582 Assert(fIOWrite);
12583 else
12584 {
12585#if 0 /** @todo r=bird: This is missing a bunch of VINF_EM_FIRST..VINF_EM_LAST
12586 * statuses, that the VMM device and some others may return. See
12587 * IOM_SUCCESS() for guidance. */
12588 AssertMsg( RT_FAILURE(rcStrict)
12589 || rcStrict == VINF_SUCCESS
12590 || rcStrict == VINF_EM_RAW_EMULATE_INSTR
12591 || rcStrict == VINF_EM_DBG_BREAKPOINT
12592 || rcStrict == VINF_EM_RAW_GUEST_TRAP
12593 || rcStrict == VINF_EM_RAW_TO_R3
12594 || rcStrict == VINF_TRPM_XCPT_DISPATCHED, ("%Rrc\n", VBOXSTRICTRC_VAL(rcStrict)));
12595#endif
12596 }
12597#endif
12598
12599 STAM_PROFILE_ADV_STOP(&pVCpu->hm.s.StatExitIO, y1);
12600 return rcStrict;
12601}
12602
12603
12604/**
12605 * VM-exit handler for task switches (VMX_EXIT_TASK_SWITCH). Unconditional
12606 * VM-exit.
12607 */
12608HMVMX_EXIT_DECL hmR0VmxExitTaskSwitch(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12609{
12610 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12611
12612 /* Check if this task-switch occurred while delivery an event through the guest IDT. */
12613 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12614 AssertRCReturn(rc, rc);
12615 if (VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_TASK_SWITCH_TYPE_IDT)
12616 {
12617 rc = hmR0VmxReadIdtVectoringInfoVmcs(pVmxTransient);
12618 AssertRCReturn(rc, rc);
12619 if (VMX_IDT_VECTORING_INFO_VALID(pVmxTransient->uIdtVectoringInfo))
12620 {
12621 uint32_t uIntType = VMX_IDT_VECTORING_INFO_TYPE(pVmxTransient->uIdtVectoringInfo);
12622
12623 uint32_t uVector = VMX_IDT_VECTORING_INFO_VECTOR(pVmxTransient->uIdtVectoringInfo);
12624 bool fErrorCodeValid = VMX_IDT_VECTORING_INFO_ERROR_CODE_IS_VALID(pVmxTransient->uIdtVectoringInfo);
12625
12626 /* Save it as a pending event and it'll be converted to a TRPM event on the way out to ring-3. */
12627 Assert(!pVCpu->hm.s.Event.fPending);
12628 pVCpu->hm.s.Event.fPending = true;
12629 pVCpu->hm.s.Event.u64IntInfo = pVmxTransient->uIdtVectoringInfo;
12630 rc = hmR0VmxReadIdtVectoringErrorCodeVmcs(pVmxTransient);
12631 AssertRCReturn(rc, rc);
12632 if (fErrorCodeValid)
12633 pVCpu->hm.s.Event.u32ErrCode = pVmxTransient->uIdtVectoringErrorCode;
12634 else
12635 pVCpu->hm.s.Event.u32ErrCode = 0;
12636 if ( uIntType == VMX_IDT_VECTORING_INFO_TYPE_HW_XCPT
12637 && uVector == X86_XCPT_PF)
12638 {
12639 pVCpu->hm.s.Event.GCPtrFaultAddress = pMixedCtx->cr2;
12640 }
12641
12642 Log4(("Pending event on TaskSwitch uIntType=%#x uVector=%#x\n", uIntType, uVector));
12643 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
12644 return VINF_EM_RAW_INJECT_TRPM_EVENT;
12645 }
12646 }
12647
12648 /* Fall back to the interpreter to emulate the task-switch. */
12649 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitTaskSwitch);
12650 return VERR_EM_INTERPRETER;
12651}
12652
12653
12654/**
12655 * VM-exit handler for monitor-trap-flag (VMX_EXIT_MTF). Conditional VM-exit.
12656 */
12657HMVMX_EXIT_DECL hmR0VmxExitMtf(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12658{
12659 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12660 Assert(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG);
12661 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MONITOR_TRAP_FLAG;
12662 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
12663 AssertRCReturn(rc, rc);
12664 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitMtf);
12665 return VINF_EM_DBG_STEPPED;
12666}
12667
12668
12669/**
12670 * VM-exit handler for APIC access (VMX_EXIT_APIC_ACCESS). Conditional VM-exit.
12671 */
12672HMVMX_EXIT_DECL hmR0VmxExitApicAccess(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12673{
12674 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12675
12676 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitApicAccess);
12677
12678 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
12679 VBOXSTRICTRC rcStrict1 = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
12680 if (RT_LIKELY(rcStrict1 == VINF_SUCCESS))
12681 {
12682 /* For some crazy guest, if an event delivery causes an APIC-access VM-exit, go to instruction emulation. */
12683 if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
12684 {
12685 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingInterpret);
12686 return VERR_EM_INTERPRETER;
12687 }
12688 }
12689 else
12690 {
12691 if (rcStrict1 == VINF_HM_DOUBLE_FAULT)
12692 rcStrict1 = VINF_SUCCESS;
12693 return rcStrict1;
12694 }
12695
12696#if 0
12697 /** @todo Investigate if IOMMMIOPhysHandler() requires a lot of state, for now
12698 * just sync the whole thing. */
12699 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
12700#else
12701 /* Aggressive state sync. for now. */
12702 int rc = hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
12703 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
12704 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12705#endif
12706 rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12707 AssertRCReturn(rc, rc);
12708
12709 /* See Intel spec. 27-6 "Exit Qualifications for APIC-access VM-exits from Linear Accesses & Guest-Phyiscal Addresses" */
12710 uint32_t uAccessType = VMX_EXIT_QUALIFICATION_APIC_ACCESS_TYPE(pVmxTransient->uExitQualification);
12711 VBOXSTRICTRC rcStrict2;
12712 switch (uAccessType)
12713 {
12714 case VMX_APIC_ACCESS_TYPE_LINEAR_WRITE:
12715 case VMX_APIC_ACCESS_TYPE_LINEAR_READ:
12716 {
12717 AssertMsg( !(pVCpu->hm.s.vmx.u32ProcCtls & VMX_VMCS_CTRL_PROC_EXEC_USE_TPR_SHADOW)
12718 || VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification) != XAPIC_OFF_TPR,
12719 ("hmR0VmxExitApicAccess: can't access TPR offset while using TPR shadowing.\n"));
12720
12721 RTGCPHYS GCPhys = pVCpu->hm.s.vmx.u64MsrApicBase; /* Always up-to-date, u64MsrApicBase is not part of the VMCS. */
12722 GCPhys &= PAGE_BASE_GC_MASK;
12723 GCPhys += VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification);
12724 PVM pVM = pVCpu->CTX_SUFF(pVM);
12725 Log4(("ApicAccess uAccessType=%#x GCPhys=%#RGp Off=%#x\n", uAccessType, GCPhys,
12726 VMX_EXIT_QUALIFICATION_APIC_ACCESS_OFFSET(pVmxTransient->uExitQualification)));
12727
12728 rcStrict2 = IOMMMIOPhysHandler(pVM, pVCpu,
12729 uAccessType == VMX_APIC_ACCESS_TYPE_LINEAR_READ ? 0 : X86_TRAP_PF_RW,
12730 CPUMCTX2CORE(pMixedCtx), GCPhys);
12731 Log4(("ApicAccess rcStrict2=%d\n", VBOXSTRICTRC_VAL(rcStrict2)));
12732 if ( rcStrict2 == VINF_SUCCESS
12733 || rcStrict2 == VERR_PAGE_TABLE_NOT_PRESENT
12734 || rcStrict2 == VERR_PAGE_NOT_PRESENT)
12735 {
12736 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
12737 | HM_CHANGED_GUEST_RSP
12738 | HM_CHANGED_GUEST_RFLAGS
12739 | HM_CHANGED_VMX_GUEST_APIC_STATE);
12740 rcStrict2 = VINF_SUCCESS;
12741 }
12742 break;
12743 }
12744
12745 default:
12746 Log4(("ApicAccess uAccessType=%#x\n", uAccessType));
12747 rcStrict2 = VINF_EM_RAW_EMULATE_INSTR;
12748 break;
12749 }
12750
12751 if (rcStrict2 != VINF_SUCCESS)
12752 STAM_COUNTER_INC(&pVCpu->hm.s.StatSwitchApicAccessToR3);
12753 return rcStrict2;
12754}
12755
12756
12757/**
12758 * VM-exit handler for debug-register accesses (VMX_EXIT_MOV_DRX). Conditional
12759 * VM-exit.
12760 */
12761HMVMX_EXIT_DECL hmR0VmxExitMovDRx(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12762{
12763 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12764
12765 /* We should -not- get this VM-exit if the guest's debug registers were active. */
12766 if (pVmxTransient->fWasGuestDebugStateActive)
12767 {
12768 AssertMsgFailed(("Unexpected MOV DRx exit. pVCpu=%p pMixedCtx=%p\n", pVCpu, pMixedCtx));
12769 HMVMX_RETURN_UNEXPECTED_EXIT();
12770 }
12771
12772 if ( !pVCpu->hm.s.fSingleInstruction
12773 && !pVmxTransient->fWasHyperDebugStateActive)
12774 {
12775 Assert(!DBGFIsStepping(pVCpu));
12776 Assert(pVCpu->hm.s.vmx.u32XcptBitmap & RT_BIT_32(X86_XCPT_DB));
12777
12778 /* Don't intercept MOV DRx any more. */
12779 pVCpu->hm.s.vmx.u32ProcCtls &= ~VMX_VMCS_CTRL_PROC_EXEC_MOV_DR_EXIT;
12780 int rc = VMXWriteVmcs32(VMX_VMCS32_CTRL_PROC_EXEC, pVCpu->hm.s.vmx.u32ProcCtls);
12781 AssertRCReturn(rc, rc);
12782
12783 /* We're playing with the host CPU state here, make sure we can't preempt or longjmp. */
12784 VMMRZCallRing3Disable(pVCpu);
12785 HM_DISABLE_PREEMPT();
12786
12787 /* Save the host & load the guest debug state, restart execution of the MOV DRx instruction. */
12788 CPUMR0LoadGuestDebugState(pVCpu, true /* include DR6 */);
12789 Assert(CPUMIsGuestDebugStateActive(pVCpu) || HC_ARCH_BITS == 32);
12790
12791 HM_RESTORE_PREEMPT();
12792 VMMRZCallRing3Enable(pVCpu);
12793
12794#ifdef VBOX_WITH_STATISTICS
12795 rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12796 AssertRCReturn(rc, rc);
12797 if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
12798 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
12799 else
12800 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
12801#endif
12802 STAM_COUNTER_INC(&pVCpu->hm.s.StatDRxContextSwitch);
12803 return VINF_SUCCESS;
12804 }
12805
12806 /*
12807 * EMInterpretDRx[Write|Read]() calls CPUMIsGuestIn64BitCode() which requires EFER, CS. EFER is always up-to-date.
12808 * Update the segment registers and DR7 from the CPU.
12809 */
12810 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12811 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12812 rc |= hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
12813 AssertRCReturn(rc, rc);
12814 Log4(("CS:RIP=%04x:%08RX64\n", pMixedCtx->cs.Sel, pMixedCtx->rip));
12815
12816 PVM pVM = pVCpu->CTX_SUFF(pVM);
12817 if (VMX_EXIT_QUALIFICATION_DRX_DIRECTION(pVmxTransient->uExitQualification) == VMX_EXIT_QUALIFICATION_DRX_DIRECTION_WRITE)
12818 {
12819 rc = EMInterpretDRxWrite(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx),
12820 VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification),
12821 VMX_EXIT_QUALIFICATION_DRX_GENREG(pVmxTransient->uExitQualification));
12822 if (RT_SUCCESS(rc))
12823 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_DEBUG);
12824 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxWrite);
12825 }
12826 else
12827 {
12828 rc = EMInterpretDRxRead(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx),
12829 VMX_EXIT_QUALIFICATION_DRX_GENREG(pVmxTransient->uExitQualification),
12830 VMX_EXIT_QUALIFICATION_DRX_REGISTER(pVmxTransient->uExitQualification));
12831 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitDRxRead);
12832 }
12833
12834 Assert(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER);
12835 if (RT_SUCCESS(rc))
12836 {
12837 int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
12838 AssertRCReturn(rc2, rc2);
12839 return VINF_SUCCESS;
12840 }
12841 return rc;
12842}
12843
12844
12845/**
12846 * VM-exit handler for EPT misconfiguration (VMX_EXIT_EPT_MISCONFIG).
12847 * Conditional VM-exit.
12848 */
12849HMVMX_EXIT_DECL hmR0VmxExitEptMisconfig(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12850{
12851 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12852 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
12853
12854 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
12855 VBOXSTRICTRC rcStrict1 = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
12856 if (RT_LIKELY(rcStrict1 == VINF_SUCCESS))
12857 {
12858 /* If event delivery causes an EPT misconfig (MMIO), go back to instruction emulation as otherwise
12859 injecting the original pending event would most likely cause the same EPT misconfig VM-exit. */
12860 if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
12861 {
12862 STAM_COUNTER_INC(&pVCpu->hm.s.StatInjectPendingInterpret);
12863 return VERR_EM_INTERPRETER;
12864 }
12865 }
12866 else
12867 {
12868 if (rcStrict1 == VINF_HM_DOUBLE_FAULT)
12869 rcStrict1 = VINF_SUCCESS;
12870 return rcStrict1;
12871 }
12872
12873 RTGCPHYS GCPhys = 0;
12874 int rc = VMXReadVmcs64(VMX_VMCS64_EXIT_GUEST_PHYS_ADDR_FULL, &GCPhys);
12875
12876#if 0
12877 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); /** @todo Can we do better? */
12878#else
12879 /* Aggressive state sync. for now. */
12880 rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
12881 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
12882 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12883#endif
12884 AssertRCReturn(rc, rc);
12885
12886 /*
12887 * If we succeed, resume guest execution.
12888 * If we fail in interpreting the instruction because we couldn't get the guest physical address
12889 * of the page containing the instruction via the guest's page tables (we would invalidate the guest page
12890 * in the host TLB), resume execution which would cause a guest page fault to let the guest handle this
12891 * weird case. See @bugref{6043}.
12892 */
12893 PVM pVM = pVCpu->CTX_SUFF(pVM);
12894 VBOXSTRICTRC rcStrict2 = PGMR0Trap0eHandlerNPMisconfig(pVM, pVCpu, PGMMODE_EPT, CPUMCTX2CORE(pMixedCtx), GCPhys, UINT32_MAX);
12895 Log4(("EPT misconfig at %#RGp RIP=%#RX64 rc=%Rrc\n", GCPhys, pMixedCtx->rip, VBOXSTRICTRC_VAL(rcStrict2)));
12896 if ( rcStrict2 == VINF_SUCCESS
12897 || rcStrict2 == VERR_PAGE_TABLE_NOT_PRESENT
12898 || rcStrict2 == VERR_PAGE_NOT_PRESENT)
12899 {
12900 /* Successfully handled MMIO operation. */
12901 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
12902 | HM_CHANGED_GUEST_RSP
12903 | HM_CHANGED_GUEST_RFLAGS
12904 | HM_CHANGED_VMX_GUEST_APIC_STATE);
12905 return VINF_SUCCESS;
12906 }
12907 return rcStrict2;
12908}
12909
12910
12911/**
12912 * VM-exit handler for EPT violation (VMX_EXIT_EPT_VIOLATION). Conditional
12913 * VM-exit.
12914 */
12915HMVMX_EXIT_DECL hmR0VmxExitEptViolation(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
12916{
12917 HMVMX_VALIDATE_EXIT_HANDLER_PARAMS();
12918 Assert(pVCpu->CTX_SUFF(pVM)->hm.s.fNestedPaging);
12919
12920 /* If this VM-exit occurred while delivering an event through the guest IDT, handle it accordingly. */
12921 VBOXSTRICTRC rcStrict1 = hmR0VmxCheckExitDueToEventDelivery(pVCpu, pMixedCtx, pVmxTransient);
12922 if (RT_LIKELY(rcStrict1 == VINF_SUCCESS))
12923 {
12924 /* In the unlikely case that the EPT violation happened as a result of delivering an event, log it. */
12925 if (RT_UNLIKELY(pVCpu->hm.s.Event.fPending))
12926 Log4(("EPT violation with an event pending u64IntInfo=%#RX64\n", pVCpu->hm.s.Event.u64IntInfo));
12927 }
12928 else
12929 {
12930 if (rcStrict1 == VINF_HM_DOUBLE_FAULT)
12931 rcStrict1 = VINF_SUCCESS;
12932 return rcStrict1;
12933 }
12934
12935 RTGCPHYS GCPhys = 0;
12936 int rc = VMXReadVmcs64(VMX_VMCS64_EXIT_GUEST_PHYS_ADDR_FULL, &GCPhys);
12937 rc |= hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
12938#if 0
12939 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx); /** @todo Can we do better? */
12940#else
12941 /* Aggressive state sync. for now. */
12942 rc |= hmR0VmxSaveGuestRipRspRflags(pVCpu, pMixedCtx);
12943 rc |= hmR0VmxSaveGuestControlRegs(pVCpu, pMixedCtx);
12944 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
12945#endif
12946 AssertRCReturn(rc, rc);
12947
12948 /* Intel spec. Table 27-7 "Exit Qualifications for EPT violations". */
12949 AssertMsg(((pVmxTransient->uExitQualification >> 7) & 3) != 2, ("%#RX64", pVmxTransient->uExitQualification));
12950
12951 RTGCUINT uErrorCode = 0;
12952 if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_INSTR_FETCH)
12953 uErrorCode |= X86_TRAP_PF_ID;
12954 if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_DATA_WRITE)
12955 uErrorCode |= X86_TRAP_PF_RW;
12956 if (pVmxTransient->uExitQualification & VMX_EXIT_QUALIFICATION_EPT_ENTRY_PRESENT)
12957 uErrorCode |= X86_TRAP_PF_P;
12958
12959 TRPMAssertXcptPF(pVCpu, GCPhys, uErrorCode);
12960
12961 Log4(("EPT violation %#x at %#RX64 ErrorCode %#x CS:RIP=%04x:%08RX64\n", pVmxTransient->uExitQualification, GCPhys,
12962 uErrorCode, pMixedCtx->cs.Sel, pMixedCtx->rip));
12963
12964 /* Handle the pagefault trap for the nested shadow table. */
12965 PVM pVM = pVCpu->CTX_SUFF(pVM);
12966 VBOXSTRICTRC rcStrict2 = PGMR0Trap0eHandlerNestedPaging(pVM, pVCpu, PGMMODE_EPT, uErrorCode, CPUMCTX2CORE(pMixedCtx), GCPhys);
12967 TRPMResetTrap(pVCpu);
12968
12969 /* Same case as PGMR0Trap0eHandlerNPMisconfig(). See comment above, @bugref{6043}. */
12970 if ( rcStrict2 == VINF_SUCCESS
12971 || rcStrict2 == VERR_PAGE_TABLE_NOT_PRESENT
12972 || rcStrict2 == VERR_PAGE_NOT_PRESENT)
12973 {
12974 /* Successfully synced our nested page tables. */
12975 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitReasonNpf);
12976 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
12977 | HM_CHANGED_GUEST_RSP
12978 | HM_CHANGED_GUEST_RFLAGS);
12979 return VINF_SUCCESS;
12980 }
12981
12982 Log4(("EPT return to ring-3 rcStrict2=%Rrc\n", VBOXSTRICTRC_VAL(rcStrict2)));
12983 return rcStrict2;
12984}
12985
12986/** @} */
12987
12988/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-= */
12989/* -=-=-=-=-=-=-=-=-=- VM-exit Exception Handlers -=-=-=-=-=-=-=-=-=-=- */
12990/* -=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=--=-=-=-=-=-=-=-=-=-=-=-=-=-= */
12991
12992/** @name VM-exit exception handlers.
12993 * @{
12994 */
12995
12996/**
12997 * VM-exit exception handler for \#MF (Math Fault: floating point exception).
12998 */
12999static int hmR0VmxExitXcptMF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13000{
13001 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13002 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestMF);
13003
13004 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
13005 AssertRCReturn(rc, rc);
13006
13007 if (!(pMixedCtx->cr0 & X86_CR0_NE))
13008 {
13009 /* Convert a #MF into a FERR -> IRQ 13. See @bugref{6117}. */
13010 rc = PDMIsaSetIrq(pVCpu->CTX_SUFF(pVM), 13, 1, 0 /* uTagSrc */);
13011
13012 /** @todo r=ramshankar: The Intel spec. does -not- specify that this VM-exit
13013 * provides VM-exit instruction length. If this causes problem later,
13014 * disassemble the instruction like it's done on AMD-V. */
13015 int rc2 = hmR0VmxAdvanceGuestRip(pVCpu, pMixedCtx, pVmxTransient);
13016 AssertRCReturn(rc2, rc2);
13017 return rc;
13018 }
13019
13020 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13021 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13022 return rc;
13023}
13024
13025
13026/**
13027 * VM-exit exception handler for \#BP (Breakpoint exception).
13028 */
13029static int hmR0VmxExitXcptBP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13030{
13031 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13032 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestBP);
13033
13034 /** @todo Try optimize this by not saving the entire guest state unless
13035 * really needed. */
13036 int rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
13037 AssertRCReturn(rc, rc);
13038
13039 PVM pVM = pVCpu->CTX_SUFF(pVM);
13040 rc = DBGFRZTrap03Handler(pVM, pVCpu, CPUMCTX2CORE(pMixedCtx));
13041 if (rc == VINF_EM_RAW_GUEST_TRAP)
13042 {
13043 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
13044 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13045 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13046 AssertRCReturn(rc, rc);
13047
13048 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13049 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13050 }
13051
13052 Assert(rc == VINF_SUCCESS || rc == VINF_EM_RAW_GUEST_TRAP || rc == VINF_EM_DBG_BREAKPOINT);
13053 return rc;
13054}
13055
13056
13057/**
13058 * VM-exit exception handler for \#AC (alignment check exception).
13059 */
13060static int hmR0VmxExitXcptAC(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13061{
13062 RT_NOREF_PV(pMixedCtx);
13063 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13064
13065 /*
13066 * Re-inject it. We'll detect any nesting before getting here.
13067 */
13068 int rc = hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13069 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13070 AssertRCReturn(rc, rc);
13071 Assert(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO);
13072
13073 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13074 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13075 return VINF_SUCCESS;
13076}
13077
13078
13079/**
13080 * VM-exit exception handler for \#DB (Debug exception).
13081 */
13082static int hmR0VmxExitXcptDB(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13083{
13084 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13085 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestDB);
13086 Log6(("XcptDB\n"));
13087
13088 /*
13089 * Get the DR6-like values from the VM-exit qualification and pass it to DBGF
13090 * for processing.
13091 */
13092 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
13093 AssertRCReturn(rc, rc);
13094
13095 /* Refer Intel spec. Table 27-1. "Exit Qualifications for debug exceptions" for the format. */
13096 uint64_t uDR6 = X86_DR6_INIT_VAL;
13097 uDR6 |= ( pVmxTransient->uExitQualification
13098 & (X86_DR6_B0 | X86_DR6_B1 | X86_DR6_B2 | X86_DR6_B3 | X86_DR6_BD | X86_DR6_BS));
13099
13100 rc = DBGFRZTrap01Handler(pVCpu->CTX_SUFF(pVM), pVCpu, CPUMCTX2CORE(pMixedCtx), uDR6, pVCpu->hm.s.fSingleInstruction);
13101 if (rc == VINF_EM_RAW_GUEST_TRAP)
13102 {
13103 /*
13104 * The exception was for the guest. Update DR6, DR7.GD and
13105 * IA32_DEBUGCTL.LBR before forwarding it.
13106 * (See Intel spec. 27.1 "Architectural State before a VM-Exit".)
13107 */
13108 VMMRZCallRing3Disable(pVCpu);
13109 HM_DISABLE_PREEMPT();
13110
13111 pMixedCtx->dr[6] &= ~X86_DR6_B_MASK;
13112 pMixedCtx->dr[6] |= uDR6;
13113 if (CPUMIsGuestDebugStateActive(pVCpu))
13114 ASMSetDR6(pMixedCtx->dr[6]);
13115
13116 HM_RESTORE_PREEMPT();
13117 VMMRZCallRing3Enable(pVCpu);
13118
13119 rc = hmR0VmxSaveGuestDR7(pVCpu, pMixedCtx);
13120 AssertRCReturn(rc, rc);
13121
13122 /* X86_DR7_GD will be cleared if DRx accesses should be trapped inside the guest. */
13123 pMixedCtx->dr[7] &= ~X86_DR7_GD;
13124
13125 /* Paranoia. */
13126 pMixedCtx->dr[7] &= ~X86_DR7_RAZ_MASK;
13127 pMixedCtx->dr[7] |= X86_DR7_RA1_MASK;
13128
13129 rc = VMXWriteVmcs32(VMX_VMCS_GUEST_DR7, (uint32_t)pMixedCtx->dr[7]);
13130 AssertRCReturn(rc, rc);
13131
13132 /*
13133 * Raise #DB in the guest.
13134 *
13135 * It is important to reflect what the VM-exit gave us (preserving the interruption-type) rather than use
13136 * hmR0VmxSetPendingXcptDB() as the #DB could've been raised while executing ICEBP and not the 'normal' #DB.
13137 * Thus it -may- trigger different handling in the CPU (like skipped DPL checks). See @bugref{6398}.
13138 *
13139 * Since ICEBP isn't documented on Intel, see AMD spec. 15.20 "Event Injection".
13140 */
13141 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
13142 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13143 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13144 AssertRCReturn(rc, rc);
13145 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13146 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13147 return VINF_SUCCESS;
13148 }
13149
13150 /*
13151 * Not a guest trap, must be a hypervisor related debug event then.
13152 * Update DR6 in case someone is interested in it.
13153 */
13154 AssertMsg(rc == VINF_EM_DBG_STEPPED || rc == VINF_EM_DBG_BREAKPOINT, ("%Rrc\n", rc));
13155 AssertReturn(pVmxTransient->fWasHyperDebugStateActive, VERR_HM_IPE_5);
13156 CPUMSetHyperDR6(pVCpu, uDR6);
13157
13158 return rc;
13159}
13160
13161
13162/**
13163 * VM-exit exception handler for \#NM (Device-not-available exception: floating
13164 * point exception).
13165 */
13166static int hmR0VmxExitXcptNM(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13167{
13168 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13169
13170 /* We require CR0 and EFER. EFER is always up-to-date. */
13171 int rc = hmR0VmxSaveGuestCR0(pVCpu, pMixedCtx);
13172 AssertRCReturn(rc, rc);
13173
13174 /* We're playing with the host CPU state here, have to disable preemption or longjmp. */
13175 VMMRZCallRing3Disable(pVCpu);
13176 HM_DISABLE_PREEMPT();
13177
13178 /* If the guest FPU was active at the time of the #NM VM-exit, then it's a guest fault. */
13179 if (pVmxTransient->fWasGuestFPUStateActive)
13180 {
13181 rc = VINF_EM_RAW_GUEST_TRAP;
13182 Assert(CPUMIsGuestFPUStateActive(pVCpu) || HMCPU_CF_IS_PENDING(pVCpu, HM_CHANGED_GUEST_CR0));
13183 }
13184 else
13185 {
13186#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
13187 Assert(!pVmxTransient->fWasGuestFPUStateActive || pVCpu->hm.s.fUsingDebugLoop);
13188#endif
13189 rc = CPUMR0Trap07Handler(pVCpu->CTX_SUFF(pVM), pVCpu);
13190 Assert( rc == VINF_EM_RAW_GUEST_TRAP
13191 || ((rc == VINF_SUCCESS || rc == VINF_CPUM_HOST_CR0_MODIFIED) && CPUMIsGuestFPUStateActive(pVCpu)));
13192 if (rc == VINF_CPUM_HOST_CR0_MODIFIED)
13193 HMCPU_CF_SET(pVCpu, HM_CHANGED_HOST_CONTEXT);
13194 }
13195
13196 HM_RESTORE_PREEMPT();
13197 VMMRZCallRing3Enable(pVCpu);
13198
13199 if (rc == VINF_SUCCESS || rc == VINF_CPUM_HOST_CR0_MODIFIED)
13200 {
13201 /* Guest FPU state was activated, we'll want to change CR0 FPU intercepts before the next VM-reentry. */
13202 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_CR0);
13203 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowNM);
13204 pVCpu->hm.s.fPreloadGuestFpu = true;
13205 }
13206 else
13207 {
13208 /* Forward #NM to the guest. */
13209 Assert(rc == VINF_EM_RAW_GUEST_TRAP);
13210 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
13211 AssertRCReturn(rc, rc);
13212 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13213 pVmxTransient->cbInstr, 0 /* error code */, 0 /* GCPtrFaultAddress */);
13214 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestNM);
13215 }
13216
13217 return VINF_SUCCESS;
13218}
13219
13220
13221/**
13222 * VM-exit exception handler for \#GP (General-protection exception).
13223 *
13224 * @remarks Requires pVmxTransient->uExitIntInfo to be up-to-date.
13225 */
13226static int hmR0VmxExitXcptGP(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13227{
13228 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13229 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestGP);
13230
13231 int rc;
13232 if (pVCpu->hm.s.vmx.RealMode.fRealOnV86Active)
13233 { /* likely */ }
13234 else
13235 {
13236#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
13237 Assert(pVCpu->hm.s.fUsingDebugLoop);
13238#endif
13239 /* If the guest is not in real-mode or we have unrestricted execution support, reflect #GP to the guest. */
13240 rc = hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
13241 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13242 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13243 rc |= hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
13244 AssertRCReturn(rc, rc);
13245 Log4(("#GP Gst: CS:RIP %04x:%08RX64 ErrorCode=%#x CR0=%#RX64 CPL=%u TR=%#04x\n", pMixedCtx->cs.Sel, pMixedCtx->rip,
13246 pVmxTransient->uExitIntErrorCode, pMixedCtx->cr0, CPUMGetGuestCPL(pVCpu), pMixedCtx->tr.Sel));
13247 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13248 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13249 return rc;
13250 }
13251
13252 Assert(CPUMIsGuestInRealModeEx(pMixedCtx));
13253 Assert(!pVCpu->CTX_SUFF(pVM)->hm.s.vmx.fUnrestrictedGuest);
13254
13255 /* EMInterpretDisasCurrent() requires a lot of the state, save the entire state. */
13256 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
13257 AssertRCReturn(rc, rc);
13258
13259 PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
13260 uint32_t cbOp = 0;
13261 PVM pVM = pVCpu->CTX_SUFF(pVM);
13262 bool fDbgStepping = pVCpu->hm.s.fSingleInstruction;
13263 rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
13264 if (RT_SUCCESS(rc))
13265 {
13266 rc = VINF_SUCCESS;
13267 Assert(cbOp == pDis->cbInstr);
13268 Log4(("#GP Disas OpCode=%u CS:EIP %04x:%04RX64\n", pDis->pCurInstr->uOpcode, pMixedCtx->cs.Sel, pMixedCtx->rip));
13269 switch (pDis->pCurInstr->uOpcode)
13270 {
13271 case OP_CLI:
13272 {
13273 pMixedCtx->eflags.Bits.u1IF = 0;
13274 pMixedCtx->eflags.Bits.u1RF = 0;
13275 pMixedCtx->rip += pDis->cbInstr;
13276 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS);
13277 if ( !fDbgStepping
13278 && pMixedCtx->eflags.Bits.u1TF)
13279 hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
13280 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitCli);
13281 break;
13282 }
13283
13284 case OP_STI:
13285 {
13286 bool fOldIF = pMixedCtx->eflags.Bits.u1IF;
13287 pMixedCtx->eflags.Bits.u1IF = 1;
13288 pMixedCtx->eflags.Bits.u1RF = 0;
13289 pMixedCtx->rip += pDis->cbInstr;
13290 if (!fOldIF)
13291 {
13292 EMSetInhibitInterruptsPC(pVCpu, pMixedCtx->rip);
13293 Assert(VMCPU_FF_IS_PENDING(pVCpu, VMCPU_FF_INHIBIT_INTERRUPTS));
13294 }
13295 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS);
13296 if ( !fDbgStepping
13297 && pMixedCtx->eflags.Bits.u1TF)
13298 hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
13299 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitSti);
13300 break;
13301 }
13302
13303 case OP_HLT:
13304 {
13305 rc = VINF_EM_HALT;
13306 pMixedCtx->rip += pDis->cbInstr;
13307 pMixedCtx->eflags.Bits.u1RF = 0;
13308 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP | HM_CHANGED_GUEST_RFLAGS);
13309 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitHlt);
13310 break;
13311 }
13312
13313 case OP_POPF:
13314 {
13315 Log4(("POPF CS:EIP %04x:%04RX64\n", pMixedCtx->cs.Sel, pMixedCtx->rip));
13316 uint32_t cbParm;
13317 uint32_t uMask;
13318 bool fGstStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
13319 if (pDis->fPrefix & DISPREFIX_OPSIZE)
13320 {
13321 cbParm = 4;
13322 uMask = 0xffffffff;
13323 }
13324 else
13325 {
13326 cbParm = 2;
13327 uMask = 0xffff;
13328 }
13329
13330 /* Get the stack pointer & pop the contents of the stack onto Eflags. */
13331 RTGCPTR GCPtrStack = 0;
13332 X86EFLAGS Eflags;
13333 Eflags.u32 = 0;
13334 rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), pMixedCtx->esp & uMask, SELMTOFLAT_FLAGS_CPL0,
13335 &GCPtrStack);
13336 if (RT_SUCCESS(rc))
13337 {
13338 Assert(sizeof(Eflags.u32) >= cbParm);
13339 rc = VBOXSTRICTRC_TODO(PGMPhysRead(pVM, (RTGCPHYS)GCPtrStack, &Eflags.u32, cbParm, PGMACCESSORIGIN_HM));
13340 AssertMsg(rc == VINF_SUCCESS, ("%Rrc\n", rc)); /** @todo allow strict return codes here */
13341 }
13342 if (RT_FAILURE(rc))
13343 {
13344 rc = VERR_EM_INTERPRETER;
13345 break;
13346 }
13347 Log4(("POPF %#x -> %#RX64 mask=%#x RIP=%#RX64\n", Eflags.u, pMixedCtx->rsp, uMask, pMixedCtx->rip));
13348 pMixedCtx->eflags.u32 = (pMixedCtx->eflags.u32 & ~((X86_EFL_POPF_BITS & uMask) | X86_EFL_RF))
13349 | (Eflags.u32 & X86_EFL_POPF_BITS & uMask);
13350 pMixedCtx->esp += cbParm;
13351 pMixedCtx->esp &= uMask;
13352 pMixedCtx->rip += pDis->cbInstr;
13353 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
13354 | HM_CHANGED_GUEST_RSP
13355 | HM_CHANGED_GUEST_RFLAGS);
13356 /* Generate a pending-debug exception when the guest stepping over POPF regardless of how
13357 POPF restores EFLAGS.TF. */
13358 if ( !fDbgStepping
13359 && fGstStepping)
13360 hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
13361 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPopf);
13362 break;
13363 }
13364
13365 case OP_PUSHF:
13366 {
13367 uint32_t cbParm;
13368 uint32_t uMask;
13369 if (pDis->fPrefix & DISPREFIX_OPSIZE)
13370 {
13371 cbParm = 4;
13372 uMask = 0xffffffff;
13373 }
13374 else
13375 {
13376 cbParm = 2;
13377 uMask = 0xffff;
13378 }
13379
13380 /* Get the stack pointer & push the contents of eflags onto the stack. */
13381 RTGCPTR GCPtrStack = 0;
13382 rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), (pMixedCtx->esp - cbParm) & uMask,
13383 SELMTOFLAT_FLAGS_CPL0, &GCPtrStack);
13384 if (RT_FAILURE(rc))
13385 {
13386 rc = VERR_EM_INTERPRETER;
13387 break;
13388 }
13389 X86EFLAGS Eflags = pMixedCtx->eflags;
13390 /* The RF & VM bits are cleared on image stored on stack; see Intel Instruction reference for PUSHF. */
13391 Eflags.Bits.u1RF = 0;
13392 Eflags.Bits.u1VM = 0;
13393
13394 rc = VBOXSTRICTRC_TODO(PGMPhysWrite(pVM, (RTGCPHYS)GCPtrStack, &Eflags.u, cbParm, PGMACCESSORIGIN_HM));
13395 if (RT_UNLIKELY(rc != VINF_SUCCESS))
13396 {
13397 AssertMsgFailed(("%Rrc\n", rc)); /** @todo allow strict return codes here */
13398 rc = VERR_EM_INTERPRETER;
13399 break;
13400 }
13401 Log4(("PUSHF %#x -> %#RGv\n", Eflags.u, GCPtrStack));
13402 pMixedCtx->esp -= cbParm;
13403 pMixedCtx->esp &= uMask;
13404 pMixedCtx->rip += pDis->cbInstr;
13405 pMixedCtx->eflags.Bits.u1RF = 0;
13406 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
13407 | HM_CHANGED_GUEST_RSP
13408 | HM_CHANGED_GUEST_RFLAGS);
13409 if ( !fDbgStepping
13410 && pMixedCtx->eflags.Bits.u1TF)
13411 hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
13412 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitPushf);
13413 break;
13414 }
13415
13416 case OP_IRET:
13417 {
13418 /** @todo Handle 32-bit operand sizes and check stack limits. See Intel
13419 * instruction reference. */
13420 RTGCPTR GCPtrStack = 0;
13421 uint32_t uMask = 0xffff;
13422 bool fGstStepping = RT_BOOL(pMixedCtx->eflags.Bits.u1TF);
13423 uint16_t aIretFrame[3];
13424 if (pDis->fPrefix & (DISPREFIX_OPSIZE | DISPREFIX_ADDRSIZE))
13425 {
13426 rc = VERR_EM_INTERPRETER;
13427 break;
13428 }
13429 rc = SELMToFlatEx(pVCpu, DISSELREG_SS, CPUMCTX2CORE(pMixedCtx), pMixedCtx->esp & uMask, SELMTOFLAT_FLAGS_CPL0,
13430 &GCPtrStack);
13431 if (RT_SUCCESS(rc))
13432 {
13433 rc = VBOXSTRICTRC_TODO(PGMPhysRead(pVM, (RTGCPHYS)GCPtrStack, &aIretFrame[0], sizeof(aIretFrame),
13434 PGMACCESSORIGIN_HM));
13435 AssertMsg(rc == VINF_SUCCESS, ("%Rrc\n", rc)); /** @todo allow strict return codes here */
13436 }
13437 if (RT_FAILURE(rc))
13438 {
13439 rc = VERR_EM_INTERPRETER;
13440 break;
13441 }
13442 pMixedCtx->eip = 0;
13443 pMixedCtx->ip = aIretFrame[0];
13444 pMixedCtx->cs.Sel = aIretFrame[1];
13445 pMixedCtx->cs.ValidSel = aIretFrame[1];
13446 pMixedCtx->cs.u64Base = (uint64_t)pMixedCtx->cs.Sel << 4;
13447 pMixedCtx->eflags.u32 = (pMixedCtx->eflags.u32 & ((UINT32_C(0xffff0000) | X86_EFL_1) & ~X86_EFL_RF))
13448 | (aIretFrame[2] & X86_EFL_POPF_BITS & uMask);
13449 pMixedCtx->sp += sizeof(aIretFrame);
13450 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
13451 | HM_CHANGED_GUEST_SEGMENT_REGS
13452 | HM_CHANGED_GUEST_RSP
13453 | HM_CHANGED_GUEST_RFLAGS);
13454 /* Generate a pending-debug exception when stepping over IRET regardless of how IRET modifies EFLAGS.TF. */
13455 if ( !fDbgStepping
13456 && fGstStepping)
13457 hmR0VmxSetPendingDebugXcptVmcs(pVCpu);
13458 Log4(("IRET %#RX32 to %04x:%04x\n", GCPtrStack, pMixedCtx->cs.Sel, pMixedCtx->ip));
13459 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitIret);
13460 break;
13461 }
13462
13463 case OP_INT:
13464 {
13465 uint16_t uVector = pDis->Param1.uValue & 0xff;
13466 hmR0VmxSetPendingIntN(pVCpu, pMixedCtx, uVector, pDis->cbInstr);
13467 /* INT clears EFLAGS.TF, we must not set any pending debug exceptions here. */
13468 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInt);
13469 break;
13470 }
13471
13472 case OP_INTO:
13473 {
13474 if (pMixedCtx->eflags.Bits.u1OF)
13475 {
13476 hmR0VmxSetPendingXcptOF(pVCpu, pMixedCtx, pDis->cbInstr);
13477 /* INTO clears EFLAGS.TF, we must not set any pending debug exceptions here. */
13478 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitInt);
13479 }
13480 else
13481 {
13482 pMixedCtx->eflags.Bits.u1RF = 0;
13483 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RFLAGS);
13484 }
13485 break;
13486 }
13487
13488 default:
13489 {
13490 pMixedCtx->eflags.Bits.u1RF = 0; /* This is correct most of the time... */
13491 VBOXSTRICTRC rc2 = EMInterpretInstructionDisasState(pVCpu, pDis, CPUMCTX2CORE(pMixedCtx), 0 /* pvFault */,
13492 EMCODETYPE_SUPERVISOR);
13493 rc = VBOXSTRICTRC_VAL(rc2);
13494 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
13495 /** @todo We have to set pending-debug exceptions here when the guest is
13496 * single-stepping depending on the instruction that was interpreted. */
13497 Log4(("#GP rc=%Rrc\n", rc));
13498 break;
13499 }
13500 }
13501 }
13502 else
13503 rc = VERR_EM_INTERPRETER;
13504
13505 AssertMsg(rc == VINF_SUCCESS || rc == VERR_EM_INTERPRETER || rc == VINF_PGM_CHANGE_MODE || rc == VINF_EM_HALT,
13506 ("#GP Unexpected rc=%Rrc\n", rc));
13507 return rc;
13508}
13509
13510
13511/**
13512 * VM-exit exception handler wrapper for generic exceptions. Simply re-injects
13513 * the exception reported in the VMX transient structure back into the VM.
13514 *
13515 * @remarks Requires uExitIntInfo in the VMX transient structure to be
13516 * up-to-date.
13517 */
13518static int hmR0VmxExitXcptGeneric(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13519{
13520 RT_NOREF_PV(pMixedCtx);
13521 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13522#ifndef HMVMX_ALWAYS_TRAP_ALL_XCPTS
13523 Assert(pVCpu->hm.s.fUsingDebugLoop);
13524#endif
13525
13526 /* Re-inject the exception into the guest. This cannot be a double-fault condition which would have been handled in
13527 hmR0VmxCheckExitDueToEventDelivery(). */
13528 int rc = hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13529 rc |= hmR0VmxReadExitInstrLenVmcs(pVmxTransient);
13530 AssertRCReturn(rc, rc);
13531 Assert(pVmxTransient->fVmcsFieldsRead & HMVMX_UPDATED_TRANSIENT_EXIT_INTERRUPTION_INFO);
13532
13533#ifdef DEBUG_ramshankar
13534 rc |= hmR0VmxSaveGuestSegmentRegs(pVCpu, pMixedCtx);
13535 uint8_t uVector = VMX_EXIT_INTERRUPTION_INFO_VECTOR(pVmxTransient->uExitIntInfo);
13536 Log(("hmR0VmxExitXcptGeneric: Reinjecting Xcpt. uVector=%#x cs:rip=%#04x:%#RX64\n", uVector, pCtx->cs.Sel, pCtx->rip));
13537#endif
13538
13539 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13540 pVmxTransient->cbInstr, pVmxTransient->uExitIntErrorCode, 0 /* GCPtrFaultAddress */);
13541 return VINF_SUCCESS;
13542}
13543
13544
13545/**
13546 * VM-exit exception handler for \#PF (Page-fault exception).
13547 */
13548static int hmR0VmxExitXcptPF(PVMCPU pVCpu, PCPUMCTX pMixedCtx, PVMXTRANSIENT pVmxTransient)
13549{
13550 HMVMX_VALIDATE_EXIT_XCPT_HANDLER_PARAMS();
13551 PVM pVM = pVCpu->CTX_SUFF(pVM);
13552 int rc = hmR0VmxReadExitQualificationVmcs(pVCpu, pVmxTransient);
13553 rc |= hmR0VmxReadExitIntInfoVmcs(pVmxTransient);
13554 rc |= hmR0VmxReadExitIntErrorCodeVmcs(pVmxTransient);
13555 AssertRCReturn(rc, rc);
13556
13557 if (!pVM->hm.s.fNestedPaging)
13558 { /* likely */ }
13559 else
13560 {
13561#if !defined(HMVMX_ALWAYS_TRAP_ALL_XCPTS) && !defined(HMVMX_ALWAYS_TRAP_PF)
13562 Assert(pVCpu->hm.s.fUsingDebugLoop);
13563#endif
13564 pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory or vectoring #PF. */
13565 if (RT_LIKELY(!pVmxTransient->fVectoringDoublePF))
13566 {
13567 pMixedCtx->cr2 = pVmxTransient->uExitQualification; /* Update here in case we go back to ring-3 before injection. */
13568 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13569 0 /* cbInstr */, pVmxTransient->uExitIntErrorCode, pVmxTransient->uExitQualification);
13570 }
13571 else
13572 {
13573 /* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
13574 hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
13575 Log4(("Pending #DF due to vectoring #PF. NP\n"));
13576 }
13577 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
13578 return rc;
13579 }
13580
13581 /* If it's a vectoring #PF, emulate injecting the original event injection as PGMTrap0eHandler() is incapable
13582 of differentiating between instruction emulation and event injection that caused a #PF. See @bugref{6607}. */
13583 if (pVmxTransient->fVectoringPF)
13584 {
13585 Assert(pVCpu->hm.s.Event.fPending);
13586 return VINF_EM_RAW_INJECT_TRPM_EVENT;
13587 }
13588
13589 rc = hmR0VmxSaveGuestState(pVCpu, pMixedCtx);
13590 AssertRCReturn(rc, rc);
13591
13592 Log4(("#PF: cr2=%#RX64 cs:rip=%#04x:%#RX64 uErrCode %#RX32 cr3=%#RX64\n", pVmxTransient->uExitQualification,
13593 pMixedCtx->cs.Sel, pMixedCtx->rip, pVmxTransient->uExitIntErrorCode, pMixedCtx->cr3));
13594
13595 TRPMAssertXcptPF(pVCpu, pVmxTransient->uExitQualification, (RTGCUINT)pVmxTransient->uExitIntErrorCode);
13596 rc = PGMTrap0eHandler(pVCpu, pVmxTransient->uExitIntErrorCode, CPUMCTX2CORE(pMixedCtx),
13597 (RTGCPTR)pVmxTransient->uExitQualification);
13598
13599 Log4(("#PF: rc=%Rrc\n", rc));
13600 if (rc == VINF_SUCCESS)
13601 {
13602#if 0
13603 /* Successfully synced shadow pages tables or emulated an MMIO instruction. */
13604 /** @todo this isn't quite right, what if guest does lgdt with some MMIO
13605 * memory? We don't update the whole state here... */
13606 HMCPU_CF_SET(pVCpu, HM_CHANGED_GUEST_RIP
13607 | HM_CHANGED_GUEST_RSP
13608 | HM_CHANGED_GUEST_RFLAGS
13609 | HM_CHANGED_VMX_GUEST_APIC_STATE);
13610#else
13611 /*
13612 * This is typically a shadow page table sync or a MMIO instruction. But we may have
13613 * emulated something like LTR or a far jump. Any part of the CPU context may have changed.
13614 */
13615 /** @todo take advantage of CPUM changed flags instead of brute forcing. */
13616 HMCPU_CF_SET(pVCpu, HM_CHANGED_ALL_GUEST);
13617#endif
13618 TRPMResetTrap(pVCpu);
13619 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPF);
13620 return rc;
13621 }
13622
13623 if (rc == VINF_EM_RAW_GUEST_TRAP)
13624 {
13625 if (!pVmxTransient->fVectoringDoublePF)
13626 {
13627 /* It's a guest page fault and needs to be reflected to the guest. */
13628 uint32_t uGstErrorCode = TRPMGetErrorCode(pVCpu);
13629 TRPMResetTrap(pVCpu);
13630 pVCpu->hm.s.Event.fPending = false; /* In case it's a contributory #PF. */
13631 pMixedCtx->cr2 = pVmxTransient->uExitQualification; /* Update here in case we go back to ring-3 before injection. */
13632 hmR0VmxSetPendingEvent(pVCpu, VMX_VMCS_CTRL_ENTRY_IRQ_INFO_FROM_EXIT_INT_INFO(pVmxTransient->uExitIntInfo),
13633 0 /* cbInstr */, uGstErrorCode, pVmxTransient->uExitQualification);
13634 }
13635 else
13636 {
13637 /* A guest page-fault occurred during delivery of a page-fault. Inject #DF. */
13638 TRPMResetTrap(pVCpu);
13639 pVCpu->hm.s.Event.fPending = false; /* Clear pending #PF to replace it with #DF. */
13640 hmR0VmxSetPendingXcptDF(pVCpu, pMixedCtx);
13641 Log4(("#PF: Pending #DF due to vectoring #PF\n"));
13642 }
13643
13644 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitGuestPF);
13645 return VINF_SUCCESS;
13646 }
13647
13648 TRPMResetTrap(pVCpu);
13649 STAM_COUNTER_INC(&pVCpu->hm.s.StatExitShadowPFEM);
13650 return rc;
13651}
13652
13653/** @} */
13654
Note: See TracBrowser for help on using the repository browser.

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