HM.cpp@ 79352

Last change on this file since 79352 was 79352, checked in by vboxsync, 6 years ago
VMM/HM: Nested VMX: bugref:9180 Comment on where to tweak host MSRs for testing VMX/SVM R0 code. Added 'full' or 'partial' VMCS shadowing support to the release log.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 151.1 KB

Line
1	/* $Id: HM.cpp 79352 2019-06-26 09:42:43Z vboxsync $ */
2	/** @file
3	* HM - Intel/AMD VM Hardware Support Manager.
4	*/
5
6	/*
7	* Copyright (C) 2006-2019 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/** @page pg_hm HM - Hardware Assisted Virtualization Manager
19	*
20	* The HM manages guest execution using the VT-x and AMD-V CPU hardware
21	* extensions.
22	*
23	* {summary of what HM does}
24	*
25	* Hardware assisted virtualization manager was originally abbreviated HWACCM,
26	* however that was cumbersome to write and parse for such a central component,
27	* so it was shortened to HM when refactoring the code in the 4.3 development
28	* cycle.
29	*
30	* {add sections with more details}
31	*
32	* @sa @ref grp_hm
33	*/
34
35
36	/*********************************************************************************************************************************
37	* Header Files *
38	*********************************************************************************************************************************/
39	#define LOG_GROUP LOG_GROUP_HM
40	#define VMCPU_INCL_CPUM_GST_CTX
41	#include <VBox/vmm/cpum.h>
42	#include <VBox/vmm/stam.h>
43	#include <VBox/vmm/mm.h>
44	#include <VBox/vmm/em.h>
45	#include <VBox/vmm/pdmapi.h>
46	#include <VBox/vmm/pgm.h>
47	#include <VBox/vmm/ssm.h>
48	#include <VBox/vmm/gim.h>
49	#include <VBox/vmm/trpm.h>
50	#include <VBox/vmm/dbgf.h>
51	#include <VBox/vmm/iom.h>
52	#include <VBox/vmm/iem.h>
53	#include <VBox/vmm/patm.h>
54	#include <VBox/vmm/csam.h>
55	#include <VBox/vmm/selm.h>
56	#include <VBox/vmm/nem.h>
57	#ifdef VBOX_WITH_REM
58	# include <VBox/vmm/rem.h>
59	#endif
60	#include <VBox/vmm/hm_vmx.h>
61	#include <VBox/vmm/hm_svm.h>
62	#include "HMInternal.h"
63	#include <VBox/vmm/vm.h>
64	#include <VBox/vmm/uvm.h>
65	#include <VBox/err.h>
66	#include <VBox/param.h>
67
68	#include <iprt/assert.h>
69	#include <VBox/log.h>
70	#include <iprt/asm.h>
71	#include <iprt/asm-amd64-x86.h>
72	#include <iprt/env.h>
73	#include <iprt/thread.h>
74
75
76	/*********************************************************************************************************************************
77	* Defined Constants And Macros *
78	*********************************************************************************************************************************/
79	/** @def HMVMX_REPORT_FEAT
80	* Reports VT-x feature to the release log.
81	*
82	* @param a_uAllowed1 Mask of allowed-1 feature bits.
83	* @param a_uAllowed0 Mask of allowed-0 feature bits.
84	* @param a_StrDesc The description string to report.
85	* @param a_Featflag Mask of the feature to report.
86	*/
87	#define HMVMX_REPORT_FEAT(a_uAllowed1, a_uAllowed0, a_StrDesc, a_Featflag) \
88	do { \
89	if ((a_uAllowed1) & (a_Featflag)) \
90	{ \
91	if ((a_uAllowed0) & (a_Featflag)) \
92	LogRel(("HM: " a_StrDesc " (must be set)\n")); \
93	else \
94	LogRel(("HM: " a_StrDesc "\n")); \
95	} \
96	else \
97	LogRel(("HM: " a_StrDesc " (must be cleared)\n")); \
98	} while (0)
99
100	/** @def HMVMX_REPORT_ALLOWED_FEAT
101	* Reports an allowed VT-x feature to the release log.
102	*
103	* @param a_uAllowed1 Mask of allowed-1 feature bits.
104	* @param a_StrDesc The description string to report.
105	* @param a_FeatFlag Mask of the feature to report.
106	*/
107	#define HMVMX_REPORT_ALLOWED_FEAT(a_uAllowed1, a_StrDesc, a_FeatFlag) \
108	do { \
109	if ((a_uAllowed1) & (a_FeatFlag)) \
110	LogRel(("HM: " a_StrDesc "\n")); \
111	else \
112	LogRel(("HM: " a_StrDesc " not supported\n")); \
113	} while (0)
114
115	/** @def HMVMX_REPORT_MSR_CAP
116	* Reports MSR feature capability.
117	*
118	* @param a_MsrCaps Mask of MSR feature bits.
119	* @param a_StrDesc The description string to report.
120	* @param a_fCap Mask of the feature to report.
121	*/
122	#define HMVMX_REPORT_MSR_CAP(a_MsrCaps, a_StrDesc, a_fCap) \
123	do { \
124	if ((a_MsrCaps) & (a_fCap)) \
125	LogRel(("HM: " a_StrDesc "\n")); \
126	} while (0)
127
128	/** @def HMVMX_LOGREL_FEAT
129	* Dumps a feature flag from a bitmap of features to the release log.
130	*
131	* @param a_fVal The value of all the features.
132	* @param a_fMask The specific bitmask of the feature.
133	*/
134	#define HMVMX_LOGREL_FEAT(a_fVal, a_fMask) \
135	do { \
136	if ((a_fVal) & (a_fMask)) \
137	LogRel(("HM: %s\n", #a_fMask)); \
138	} while (0)
139
140
141	/*********************************************************************************************************************************
142	* Internal Functions *
143	*********************************************************************************************************************************/
144	static DECLCALLBACK(int) hmR3Save(PVM pVM, PSSMHANDLE pSSM);
145	static DECLCALLBACK(int) hmR3Load(PVM pVM, PSSMHANDLE pSSM, uint32_t uVersion, uint32_t uPass);
146	static DECLCALLBACK(void) hmR3InfoSvmNstGstVmcbCache(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs);
147	static DECLCALLBACK(void) hmR3Info(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs);
148	static DECLCALLBACK(void) hmR3InfoEventPending(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs);
149	static int hmR3InitFinalizeR3(PVM pVM);
150	static int hmR3InitFinalizeR0(PVM pVM);
151	static int hmR3InitFinalizeR0Intel(PVM pVM);
152	static int hmR3InitFinalizeR0Amd(PVM pVM);
153	static int hmR3TermCPU(PVM pVM);
154
155
156
157	/**
158	* Initializes the HM.
159	*
160	* This is the very first component to really do init after CFGM so that we can
161	* establish the predominant execution engine for the VM prior to initializing
162	* other modules. It takes care of NEM initialization if needed (HM disabled or
163	* not available in HW).
164	*
165	* If VT-x or AMD-V hardware isn't available, HM will try fall back on a native
166	* hypervisor API via NEM, and then back on raw-mode if that isn't available
167	* either. The fallback to raw-mode will not happen if /HM/HMForced is set
168	* (like for guest using SMP or 64-bit as well as for complicated guest like OS
169	* X, OS/2 and others).
170	*
171	* Note that a lot of the set up work is done in ring-0 and thus postponed till
172	* the ring-3 and ring-0 callback to HMR3InitCompleted.
173	*
174	* @returns VBox status code.
175	* @param pVM The cross context VM structure.
176	*
177	* @remarks Be careful with what we call here, since most of the VMM components
178	* are uninitialized.
179	*/
180	VMMR3_INT_DECL(int) HMR3Init(PVM pVM)
181	{
182	LogFlow(("HMR3Init\n"));
183
184	/*
185	* Assert alignment and sizes.
186	*/
187	AssertCompileMemberAlignment(VM, hm.s, 32);
188	AssertCompile(sizeof(pVM->hm.s) <= sizeof(pVM->hm.padding));
189
190	/*
191	* Register the saved state data unit.
192	*/
193	int rc = SSMR3RegisterInternal(pVM, "HWACCM", 0, HM_SAVED_STATE_VERSION, sizeof(HM),
194	NULL, NULL, NULL,
195	NULL, hmR3Save, NULL,
196	NULL, hmR3Load, NULL);
197	if (RT_FAILURE(rc))
198	return rc;
199
200	/*
201	* Register info handlers.
202	*/
203	rc = DBGFR3InfoRegisterInternalEx(pVM, "hm", "Dumps HM info.", hmR3Info, DBGFINFO_FLAGS_ALL_EMTS);
204	AssertRCReturn(rc, rc);
205
206	rc = DBGFR3InfoRegisterInternalEx(pVM, "hmeventpending", "Dumps the pending HM event.", hmR3InfoEventPending,
207	DBGFINFO_FLAGS_ALL_EMTS);
208	AssertRCReturn(rc, rc);
209
210	rc = DBGFR3InfoRegisterInternalEx(pVM, "svmvmcbcache", "Dumps the HM SVM nested-guest VMCB cache.",
211	hmR3InfoSvmNstGstVmcbCache, DBGFINFO_FLAGS_ALL_EMTS);
212	AssertRCReturn(rc, rc);
213
214	/*
215	* Read configuration.
216	*/
217	PCFGMNODE pCfgHm = CFGMR3GetChild(CFGMR3GetRoot(pVM), "HM/");
218
219	/*
220	* Validate the HM settings.
221	*/
222	rc = CFGMR3ValidateConfig(pCfgHm, "/HM/",
223	"HMForced"
224	"\|UseNEMInstead"
225	"\|FallbackToNEM"
226	"\|EnableNestedPaging"
227	"\|EnableUX"
228	"\|EnableLargePages"
229	"\|EnableVPID"
230	"\|IBPBOnVMExit"
231	"\|IBPBOnVMEntry"
232	"\|SpecCtrlByHost"
233	"\|L1DFlushOnSched"
234	"\|L1DFlushOnVMEntry"
235	"\|MDSClearOnSched"
236	"\|MDSClearOnVMEntry"
237	"\|TPRPatchingEnabled"
238	"\|64bitEnabled"
239	"\|Exclusive"
240	"\|MaxResumeLoops"
241	"\|VmxPleGap"
242	"\|VmxPleWindow"
243	"\|UseVmxPreemptTimer"
244	"\|SvmPauseFilter"
245	"\|SvmPauseFilterThreshold"
246	"\|SvmVirtVmsaveVmload"
247	"\|SvmVGif"
248	"\|LovelyMesaDrvWorkaround",
249	"" /* pszValidNodes /, "HM" / pszWho /, 0 / uInstance */);
250	if (RT_FAILURE(rc))
251	return rc;
252
253	/** @cfgm{/HM/HMForced, bool, false}
254	* Forces hardware virtualization, no falling back on raw-mode. HM must be
255	* enabled, i.e. /HMEnabled must be true. */
256	bool fHMForced;
257	#ifdef VBOX_WITH_RAW_MODE
258	rc = CFGMR3QueryBoolDef(pCfgHm, "HMForced", &fHMForced, false);
259	AssertRCReturn(rc, rc);
260	AssertLogRelMsgReturn(!fHMForced \|\| pVM->fHMEnabled, ("Configuration error: HM forced but not enabled!\n"),
261	VERR_INVALID_PARAMETER);
262	# if defined(RT_OS_DARWIN)
263	if (pVM->fHMEnabled)
264	fHMForced = true;
265	# endif
266	AssertLogRelMsgReturn(pVM->cCpus == 1 \|\| pVM->fHMEnabled, ("Configuration error: SMP requires HM to be enabled!\n"),
267	VERR_INVALID_PARAMETER);
268	if (pVM->cCpus > 1)
269	fHMForced = true;
270	#else /* !VBOX_WITH_RAW_MODE */
271	AssertRelease(pVM->fHMEnabled);
272	fHMForced = true;
273	#endif /* !VBOX_WITH_RAW_MODE */
274
275	/** @cfgm{/HM/UseNEMInstead, bool, true}
276	* Don't use HM, use NEM instead. */
277	bool fUseNEMInstead = false;
278	rc = CFGMR3QueryBoolDef(pCfgHm, "UseNEMInstead", &fUseNEMInstead, false);
279	AssertRCReturn(rc, rc);
280	if (fUseNEMInstead && pVM->fHMEnabled)
281	{
282	LogRel(("HM: Setting fHMEnabled to false because fUseNEMInstead is set.\n"));
283	pVM->fHMEnabled = false;
284	}
285
286	/** @cfgm{/HM/FallbackToNEM, bool, true}
287	* Enables fallback on NEM. */
288	bool fFallbackToNEM = true;
289	rc = CFGMR3QueryBoolDef(pCfgHm, "FallbackToNEM", &fFallbackToNEM, true);
290	AssertRCReturn(rc, rc);
291
292	/** @cfgm{/HM/EnableNestedPaging, bool, false}
293	* Enables nested paging (aka extended page tables). */
294	rc = CFGMR3QueryBoolDef(pCfgHm, "EnableNestedPaging", &pVM->hm.s.fAllowNestedPaging, false);
295	AssertRCReturn(rc, rc);
296
297	/** @cfgm{/HM/EnableUX, bool, true}
298	* Enables the VT-x unrestricted execution feature. */
299	rc = CFGMR3QueryBoolDef(pCfgHm, "EnableUX", &pVM->hm.s.vmx.fAllowUnrestricted, true);
300	AssertRCReturn(rc, rc);
301
302	/** @cfgm{/HM/EnableLargePages, bool, false}
303	* Enables using large pages (2 MB) for guest memory, thus saving on (nested)
304	* page table walking and maybe better TLB hit rate in some cases. */
305	rc = CFGMR3QueryBoolDef(pCfgHm, "EnableLargePages", &pVM->hm.s.fLargePages, false);
306	AssertRCReturn(rc, rc);
307
308	/** @cfgm{/HM/EnableVPID, bool, false}
309	* Enables the VT-x VPID feature. */
310	rc = CFGMR3QueryBoolDef(pCfgHm, "EnableVPID", &pVM->hm.s.vmx.fAllowVpid, false);
311	AssertRCReturn(rc, rc);
312
313	/** @cfgm{/HM/TPRPatchingEnabled, bool, false}
314	* Enables TPR patching for 32-bit windows guests with IO-APIC. */
315	rc = CFGMR3QueryBoolDef(pCfgHm, "TPRPatchingEnabled", &pVM->hm.s.fTprPatchingAllowed, false);
316	AssertRCReturn(rc, rc);
317
318	/** @cfgm{/HM/64bitEnabled, bool, 32-bit:false, 64-bit:true}
319	* Enables AMD64 cpu features.
320	* On 32-bit hosts this isn't default and require host CPU support. 64-bit hosts
321	* already have the support. */
322	#ifdef VBOX_ENABLE_64_BITS_GUESTS
323	rc = CFGMR3QueryBoolDef(pCfgHm, "64bitEnabled", &pVM->hm.s.fAllow64BitGuests, HC_ARCH_BITS == 64);
324	AssertLogRelRCReturn(rc, rc);
325	#else
326	pVM->hm.s.fAllow64BitGuests = false;
327	#endif
328
329	/** @cfgm{/HM/VmxPleGap, uint32_t, 0}
330	* The pause-filter exiting gap in TSC ticks. When the number of ticks between
331	* two successive PAUSE instructions exceeds VmxPleGap, the CPU considers the
332	* latest PAUSE instruction to be start of a new PAUSE loop.
333	*/
334	rc = CFGMR3QueryU32Def(pCfgHm, "VmxPleGap", &pVM->hm.s.vmx.cPleGapTicks, 0);
335	AssertRCReturn(rc, rc);
336
337	/** @cfgm{/HM/VmxPleWindow, uint32_t, 0}
338	* The pause-filter exiting window in TSC ticks. When the number of ticks
339	* between the current PAUSE instruction and first PAUSE of a loop exceeds
340	* VmxPleWindow, a VM-exit is triggered.
341	*
342	* Setting VmxPleGap and VmxPleGap to 0 disables pause-filter exiting.
343	*/
344	rc = CFGMR3QueryU32Def(pCfgHm, "VmxPleWindow", &pVM->hm.s.vmx.cPleWindowTicks, 0);
345	AssertRCReturn(rc, rc);
346
347	/** @cfgm{/HM/SvmPauseFilterCount, uint16_t, 0}
348	* A counter that is decrement each time a PAUSE instruction is executed by the
349	* guest. When the counter is 0, a \#VMEXIT is triggered.
350	*
351	* Setting SvmPauseFilterCount to 0 disables pause-filter exiting.
352	*/
353	rc = CFGMR3QueryU16Def(pCfgHm, "SvmPauseFilter", &pVM->hm.s.svm.cPauseFilter, 0);
354	AssertRCReturn(rc, rc);
355
356	/** @cfgm{/HM/SvmPauseFilterThreshold, uint16_t, 0}
357	* The pause filter threshold in ticks. When the elapsed time (in ticks) between
358	* two successive PAUSE instructions exceeds SvmPauseFilterThreshold, the
359	* PauseFilter count is reset to its initial value. However, if PAUSE is
360	* executed PauseFilter times within PauseFilterThreshold ticks, a VM-exit will
361	* be triggered.
362	*
363	* Requires SvmPauseFilterCount to be non-zero for pause-filter threshold to be
364	* activated.
365	*/
366	rc = CFGMR3QueryU16Def(pCfgHm, "SvmPauseFilterThreshold", &pVM->hm.s.svm.cPauseFilterThresholdTicks, 0);
367	AssertRCReturn(rc, rc);
368
369	/** @cfgm{/HM/SvmVirtVmsaveVmload, bool, true}
370	* Whether to make use of virtualized VMSAVE/VMLOAD feature of the CPU if it's
371	* available. */
372	rc = CFGMR3QueryBoolDef(pCfgHm, "SvmVirtVmsaveVmload", &pVM->hm.s.svm.fVirtVmsaveVmload, true);
373	AssertRCReturn(rc, rc);
374
375	/** @cfgm{/HM/SvmVGif, bool, true}
376	* Whether to make use of Virtual GIF (Global Interrupt Flag) feature of the CPU
377	* if it's available. */
378	rc = CFGMR3QueryBoolDef(pCfgHm, "SvmVGif", &pVM->hm.s.svm.fVGif, true);
379	AssertRCReturn(rc, rc);
380
381	/** @cfgm{/HM/Exclusive, bool}
382	* Determines the init method for AMD-V and VT-x. If set to true, HM will do a
383	* global init for each host CPU. If false, we do local init each time we wish
384	* to execute guest code.
385	*
386	* On Windows, default is false due to the higher risk of conflicts with other
387	* hypervisors.
388	*
389	* On Mac OS X, this setting is ignored since the code does not handle local
390	* init when it utilizes the OS provided VT-x function, SUPR0EnableVTx().
391	*/
392	#if defined(RT_OS_DARWIN)
393	pVM->hm.s.fGlobalInit = true;
394	#else
395	rc = CFGMR3QueryBoolDef(pCfgHm, "Exclusive", &pVM->hm.s.fGlobalInit,
396	# if defined(RT_OS_WINDOWS)
397	false
398	# else
399	true
400	# endif
401	);
402	AssertLogRelRCReturn(rc, rc);
403	#endif
404
405	/** @cfgm{/HM/MaxResumeLoops, uint32_t}
406	* The number of times to resume guest execution before we forcibly return to
407	* ring-3. The return value of RTThreadPreemptIsPendingTrusty in ring-0
408	* determines the default value. */
409	rc = CFGMR3QueryU32Def(pCfgHm, "MaxResumeLoops", &pVM->hm.s.cMaxResumeLoops, 0 /* set by R0 later */);
410	AssertLogRelRCReturn(rc, rc);
411
412	/** @cfgm{/HM/UseVmxPreemptTimer, bool}
413	* Whether to make use of the VMX-preemption timer feature of the CPU if it's
414	* available. */
415	rc = CFGMR3QueryBoolDef(pCfgHm, "UseVmxPreemptTimer", &pVM->hm.s.vmx.fUsePreemptTimer, true);
416	AssertLogRelRCReturn(rc, rc);
417
418	/** @cfgm{/HM/IBPBOnVMExit, bool}
419	* Costly paranoia setting. */
420	rc = CFGMR3QueryBoolDef(pCfgHm, "IBPBOnVMExit", &pVM->hm.s.fIbpbOnVmExit, false);
421	AssertLogRelRCReturn(rc, rc);
422
423	/** @cfgm{/HM/IBPBOnVMEntry, bool}
424	* Costly paranoia setting. */
425	rc = CFGMR3QueryBoolDef(pCfgHm, "IBPBOnVMEntry", &pVM->hm.s.fIbpbOnVmEntry, false);
426	AssertLogRelRCReturn(rc, rc);
427
428	/** @cfgm{/HM/L1DFlushOnSched, bool, true}
429	* CVE-2018-3646 workaround, ignored on CPUs that aren't affected. */
430	rc = CFGMR3QueryBoolDef(pCfgHm, "L1DFlushOnSched", &pVM->hm.s.fL1dFlushOnSched, true);
431	AssertLogRelRCReturn(rc, rc);
432
433	/** @cfgm{/HM/L1DFlushOnVMEntry, bool}
434	* CVE-2018-3646 workaround, ignored on CPUs that aren't affected. */
435	rc = CFGMR3QueryBoolDef(pCfgHm, "L1DFlushOnVMEntry", &pVM->hm.s.fL1dFlushOnVmEntry, false);
436	AssertLogRelRCReturn(rc, rc);
437
438	/* Disable L1DFlushOnSched if L1DFlushOnVMEntry is enabled. */
439	if (pVM->hm.s.fL1dFlushOnVmEntry)
440	pVM->hm.s.fL1dFlushOnSched = false;
441
442	/** @cfgm{/HM/SpecCtrlByHost, bool}
443	* Another expensive paranoia setting. */
444	rc = CFGMR3QueryBoolDef(pCfgHm, "SpecCtrlByHost", &pVM->hm.s.fSpecCtrlByHost, false);
445	AssertLogRelRCReturn(rc, rc);
446
447	/** @cfgm{/HM/MDSClearOnSched, bool, true}
448	* CVE-2018-12126, CVE-2018-12130, CVE-2018-12127, CVE-2019-11091 workaround,
449	* ignored on CPUs that aren't affected. */
450	rc = CFGMR3QueryBoolDef(pCfgHm, "MDSClearOnSched", &pVM->hm.s.fMdsClearOnSched, true);
451	AssertLogRelRCReturn(rc, rc);
452
453	/** @cfgm{/HM/MDSClearOnVmEntry, bool, false}
454	* CVE-2018-12126, CVE-2018-12130, CVE-2018-12127, CVE-2019-11091 workaround,
455	* ignored on CPUs that aren't affected. */
456	rc = CFGMR3QueryBoolDef(pCfgHm, "MDSClearOnVmEntry", &pVM->hm.s.fMdsClearOnVmEntry, false);
457	AssertLogRelRCReturn(rc, rc);
458
459	/* Disable MDSClearOnSched if MDSClearOnVmEntry is enabled. */
460	if (pVM->hm.s.fMdsClearOnVmEntry)
461	pVM->hm.s.fMdsClearOnSched = false;
462
463	/** @cfgm{/HM/LovelyMesaDrvWorkaround,bool}
464	* Workaround for mesa vmsvga 3d driver making incorrect assumptions about
465	* the hypervisor it is running under. */
466	bool f;
467	rc = CFGMR3QueryBoolDef(pCfgHm, "LovelyMesaDrvWorkaround", &f, false);
468	AssertLogRelRCReturn(rc, rc);
469	for (VMCPUID i = 0; i < pVM->cCpus; i++)
470	pVM->aCpus[i].hm.s.fTrapXcptGpForLovelyMesaDrv = f;
471
472	/*
473	* Check if VT-x or AMD-v support according to the users wishes.
474	*/
475	/** @todo SUPR3QueryVTCaps won't catch VERR_VMX_IN_VMX_ROOT_MODE or
476	* VERR_SVM_IN_USE. */
477	if (pVM->fHMEnabled)
478	{
479	uint32_t fCaps;
480	rc = SUPR3QueryVTCaps(&fCaps);
481	if (RT_SUCCESS(rc))
482	{
483	if (fCaps & SUPVTCAPS_AMD_V)
484	{
485	pVM->hm.s.svm.fSupported = true;
486	LogRel(("HM: HMR3Init: AMD-V%s\n", fCaps & SUPVTCAPS_NESTED_PAGING ? " w/ nested paging" : ""));
487	VM_SET_MAIN_EXECUTION_ENGINE(pVM, VM_EXEC_ENGINE_HW_VIRT);
488	}
489	else if (fCaps & SUPVTCAPS_VT_X)
490	{
491	const char *pszWhy;
492	rc = SUPR3QueryVTxSupported(&pszWhy);
493	if (RT_SUCCESS(rc))
494	{
495	pVM->hm.s.vmx.fSupported = true;
496	LogRel(("HM: HMR3Init: VT-x%s%s%s\n",
497	fCaps & SUPVTCAPS_NESTED_PAGING ? " w/ nested paging" : "",
498	fCaps & SUPVTCAPS_VTX_UNRESTRICTED_GUEST ? " and unrestricted guest execution" : "",
499	(fCaps & (SUPVTCAPS_NESTED_PAGING \| SUPVTCAPS_VTX_UNRESTRICTED_GUEST)) ? " hw support" : ""));
500	VM_SET_MAIN_EXECUTION_ENGINE(pVM, VM_EXEC_ENGINE_HW_VIRT);
501	}
502	else
503	{
504	/*
505	* Before failing, try fallback to NEM if we're allowed to do that.
506	*/
507	pVM->fHMEnabled = false;
508	Assert(pVM->bMainExecutionEngine == VM_EXEC_ENGINE_NOT_SET);
509	if (fFallbackToNEM)
510	{
511	LogRel(("HM: HMR3Init: Attempting fall back to NEM: The host kernel does not support VT-x - %s\n", pszWhy));
512	int rc2 = NEMR3Init(pVM, true /fFallback/, fHMForced);
513
514	ASMCompilerBarrier(); /* NEMR3Init may have changed bMainExecutionEngine. */
515	if ( RT_SUCCESS(rc2)
516	&& pVM->bMainExecutionEngine != VM_EXEC_ENGINE_NOT_SET)
517	rc = VINF_SUCCESS;
518	}
519	if (RT_FAILURE(rc))
520	{
521	if (fHMForced)
522	return VMSetError(pVM, rc, RT_SRC_POS, "The host kernel does not support VT-x: %s\n", pszWhy);
523
524	/* Fall back to raw-mode. */
525	LogRel(("HM: HMR3Init: Falling back to raw-mode: The host kernel does not support VT-x - %s\n", pszWhy));
526	VM_SET_MAIN_EXECUTION_ENGINE(pVM, VM_EXEC_ENGINE_RAW_MODE);
527	}
528	}
529	}
530	else
531	AssertLogRelMsgFailedReturn(("SUPR3QueryVTCaps didn't return either AMD-V or VT-x flag set (%#x)!\n", fCaps),
532	VERR_INTERNAL_ERROR_5);
533
534	/*
535	* Do we require a little bit or raw-mode for 64-bit guest execution?
536	*/
537	pVM->fHMNeedRawModeCtx = HC_ARCH_BITS == 32
538	&& pVM->fHMEnabled
539	&& pVM->hm.s.fAllow64BitGuests;
540
541	/*
542	* Disable nested paging and unrestricted guest execution now if they're
543	* configured so that CPUM can make decisions based on our configuration.
544	*/
545	Assert(!pVM->hm.s.fNestedPaging);
546	if (pVM->hm.s.fAllowNestedPaging)
547	{
548	if (fCaps & SUPVTCAPS_NESTED_PAGING)
549	pVM->hm.s.fNestedPaging = true;
550	else
551	pVM->hm.s.fAllowNestedPaging = false;
552	}
553
554	if (fCaps & SUPVTCAPS_VT_X)
555	{
556	Assert(!pVM->hm.s.vmx.fUnrestrictedGuest);
557	if (pVM->hm.s.vmx.fAllowUnrestricted)
558	{
559	if ( (fCaps & SUPVTCAPS_VTX_UNRESTRICTED_GUEST)
560	&& pVM->hm.s.fNestedPaging)
561	pVM->hm.s.vmx.fUnrestrictedGuest = true;
562	else
563	pVM->hm.s.vmx.fAllowUnrestricted = false;
564	}
565	}
566	}
567	else
568	{
569	const char *pszMsg;
570	switch (rc)
571	{
572	case VERR_UNSUPPORTED_CPU: pszMsg = "Unknown CPU, VT-x or AMD-v features cannot be ascertained"; break;
573	case VERR_VMX_NO_VMX: pszMsg = "VT-x is not available"; break;
574	case VERR_VMX_MSR_VMX_DISABLED: pszMsg = "VT-x is disabled in the BIOS"; break;
575	case VERR_VMX_MSR_ALL_VMX_DISABLED: pszMsg = "VT-x is disabled in the BIOS for all CPU modes"; break;
576	case VERR_VMX_MSR_LOCKING_FAILED: pszMsg = "Failed to enable and lock VT-x features"; break;
577	case VERR_SVM_NO_SVM: pszMsg = "AMD-V is not available"; break;
578	case VERR_SVM_DISABLED: pszMsg = "AMD-V is disabled in the BIOS (or by the host OS)"; break;
579	default:
580	return VMSetError(pVM, rc, RT_SRC_POS, "SUPR3QueryVTCaps failed with %Rrc", rc);
581	}
582
583	/*
584	* Before failing, try fallback to NEM if we're allowed to do that.
585	*/
586	pVM->fHMEnabled = false;
587	if (fFallbackToNEM)
588	{
589	LogRel(("HM: HMR3Init: Attempting fall back to NEM: %s\n", pszMsg));
590	int rc2 = NEMR3Init(pVM, true /fFallback/, fHMForced);
591	ASMCompilerBarrier(); /* NEMR3Init may have changed bMainExecutionEngine. */
592	if ( RT_SUCCESS(rc2)
593	&& pVM->bMainExecutionEngine != VM_EXEC_ENGINE_NOT_SET)
594	rc = VINF_SUCCESS;
595	}
596	if (RT_FAILURE(rc))
597	{
598	if (fHMForced)
599	return VM_SET_ERROR(pVM, rc, pszMsg);
600
601	LogRel(("HM: HMR3Init: Falling back to raw-mode: %s\n", pszMsg));
602	VM_SET_MAIN_EXECUTION_ENGINE(pVM, VM_EXEC_ENGINE_RAW_MODE);
603	}
604	}
605	}
606	else
607	{
608	/*
609	* Disabled HM mean raw-mode, unless NEM is supposed to be used.
610	*/
611	if (!fUseNEMInstead)
612	VM_SET_MAIN_EXECUTION_ENGINE(pVM, VM_EXEC_ENGINE_RAW_MODE);
613	else
614	{
615	rc = NEMR3Init(pVM, false /fFallback/, true);
616	ASMCompilerBarrier(); /* NEMR3Init may have changed bMainExecutionEngine. */
617	if (RT_FAILURE(rc))
618	return rc;
619	}
620	}
621
622	return VINF_SUCCESS;
623	}
624
625
626	/**
627	* Initializes HM components after ring-3 phase has been fully initialized.
628	*
629	* @returns VBox status code.
630	* @param pVM The cross context VM structure.
631	*/
632	static int hmR3InitFinalizeR3(PVM pVM)
633	{
634	LogFlow(("HMR3InitCPU\n"));
635
636	if (!HMIsEnabled(pVM))
637	return VINF_SUCCESS;
638
639	for (VMCPUID i = 0; i < pVM->cCpus; i++)
640	{
641	PVMCPU pVCpu = &pVM->aCpus[i];
642	pVCpu->hm.s.fActive = false;
643	pVCpu->hm.s.fGIMTrapXcptUD = GIMShouldTrapXcptUD(pVCpu); /* Is safe to call now since GIMR3Init() has completed. */
644	}
645
646	#ifdef VBOX_WITH_STATISTICS
647	STAM_REG(pVM, &pVM->hm.s.StatTprPatchSuccess, STAMTYPE_COUNTER, "/HM/TPR/Patch/Success", STAMUNIT_OCCURENCES, "Number of times an instruction was successfully patched.");
648	STAM_REG(pVM, &pVM->hm.s.StatTprPatchFailure, STAMTYPE_COUNTER, "/HM/TPR/Patch/Failed", STAMUNIT_OCCURENCES, "Number of unsuccessful patch attempts.");
649	STAM_REG(pVM, &pVM->hm.s.StatTprReplaceSuccessCr8, STAMTYPE_COUNTER, "/HM/TPR/Replace/SuccessCR8", STAMUNIT_OCCURENCES, "Number of instruction replacements by MOV CR8.");
650	STAM_REG(pVM, &pVM->hm.s.StatTprReplaceSuccessVmc, STAMTYPE_COUNTER, "/HM/TPR/Replace/SuccessVMC", STAMUNIT_OCCURENCES, "Number of instruction replacements by VMMCALL.");
651	STAM_REG(pVM, &pVM->hm.s.StatTprReplaceFailure, STAMTYPE_COUNTER, "/HM/TPR/Replace/Failed", STAMUNIT_OCCURENCES, "Number of unsuccessful replace attempts.");
652	#endif
653
654	/*
655	* Statistics.
656	*/
657	for (VMCPUID i = 0; i < pVM->cCpus; i++)
658	{
659	PVMCPU pVCpu = &pVM->aCpus[i];
660	int rc;
661
662	#ifdef VBOX_WITH_STATISTICS
663	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatPoke, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
664	"Profiling of RTMpPokeCpu.",
665	"/PROF/CPU%d/HM/Poke", i);
666	AssertRC(rc);
667	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatSpinPoke, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
668	"Profiling of poke wait.",
669	"/PROF/CPU%d/HM/PokeWait", i);
670	AssertRC(rc);
671	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatSpinPokeFailed, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
672	"Profiling of poke wait when RTMpPokeCpu fails.",
673	"/PROF/CPU%d/HM/PokeWaitFailed", i);
674	AssertRC(rc);
675	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatEntry, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
676	"Profiling of entry until entering GC.",
677	"/PROF/CPU%d/HM/Entry", i);
678	AssertRC(rc);
679	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatPreExit, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
680	"Profiling of pre-exit processing after returning from GC.",
681	"/PROF/CPU%d/HM/SwitchFromGC_1", i);
682	AssertRC(rc);
683	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitHandling, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
684	"Profiling of exit handling (longjmps not included!)",
685	"/PROF/CPU%d/HM/SwitchFromGC_2", i);
686	AssertRC(rc);
687
688	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitIO, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
689	"I/O.",
690	"/PROF/CPU%d/HM/SwitchFromGC_2/IO", i);
691	AssertRC(rc);
692	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitMovCRx, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
693	"MOV CRx.",
694	"/PROF/CPU%d/HM/SwitchFromGC_2/MovCRx", i);
695	AssertRC(rc);
696	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitXcptNmi, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
697	"Exceptions, NMIs.",
698	"/PROF/CPU%d/HM/SwitchFromGC_2/XcptNmi", i);
699	AssertRC(rc);
700
701	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatImportGuestState, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
702	"Profiling of importing guest state from hardware after VM-exit.",
703	"/PROF/CPU%d/HM/ImportGuestState", i);
704	AssertRC(rc);
705	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExportGuestState, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
706	"Profiling of exporting guest state to hardware before VM-entry.",
707	"/PROF/CPU%d/HM/ExportGuestState", i);
708	AssertRC(rc);
709	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatLoadGuestFpuState, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
710	"Profiling of CPUMR0LoadGuestFPU.",
711	"/PROF/CPU%d/HM/LoadGuestFpuState", i);
712	AssertRC(rc);
713	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatInGC, STAMTYPE_PROFILE, STAMVISIBILITY_USED, STAMUNIT_TICKS_PER_CALL,
714	"Profiling of execution of guest-code in hardware.",
715	"/PROF/CPU%d/HM/InGC", i);
716	AssertRC(rc);
717
718	# if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
719	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatWorldSwitch3264, STAMTYPE_PROFILE, STAMVISIBILITY_USED,
720	STAMUNIT_TICKS_PER_CALL, "Profiling of the 32/64 switcher.",
721	"/PROF/CPU%d/HM/Switcher3264", i);
722	AssertRC(rc);
723	# endif
724
725	# ifdef HM_PROFILE_EXIT_DISPATCH
726	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitDispatch, STAMTYPE_PROFILE_ADV, STAMVISIBILITY_USED,
727	STAMUNIT_TICKS_PER_CALL, "Profiling the dispatching of exit handlers.",
728	"/PROF/CPU%d/HM/ExitDispatch", i);
729	AssertRC(rc);
730	# endif
731
732	#endif
733	# define HM_REG_COUNTER(a, b, desc) \
734	rc = STAMR3RegisterF(pVM, a, STAMTYPE_COUNTER, STAMVISIBILITY_ALWAYS, STAMUNIT_OCCURENCES, desc, b, i); \
735	AssertRC(rc);
736
737	#ifdef VBOX_WITH_STATISTICS
738	HM_REG_COUNTER(&pVCpu->hm.s.StatExitAll, "/HM/CPU%d/Exit/All", "Exits (total).");
739	HM_REG_COUNTER(&pVCpu->hm.s.StatExitShadowNM, "/HM/CPU%d/Exit/Trap/Shw/#NM", "Shadow #NM (device not available, no math co-processor) exception.");
740	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestNM, "/HM/CPU%d/Exit/Trap/Gst/#NM", "Guest #NM (device not available, no math co-processor) exception.");
741	HM_REG_COUNTER(&pVCpu->hm.s.StatExitShadowPF, "/HM/CPU%d/Exit/Trap/Shw/#PF", "Shadow #PF (page fault) exception.");
742	HM_REG_COUNTER(&pVCpu->hm.s.StatExitShadowPFEM, "/HM/CPU%d/Exit/Trap/Shw/#PF-EM", "#PF (page fault) exception going back to ring-3 for emulating the instruction.");
743	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestPF, "/HM/CPU%d/Exit/Trap/Gst/#PF", "Guest #PF (page fault) exception.");
744	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestUD, "/HM/CPU%d/Exit/Trap/Gst/#UD", "Guest #UD (undefined opcode) exception.");
745	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestSS, "/HM/CPU%d/Exit/Trap/Gst/#SS", "Guest #SS (stack-segment fault) exception.");
746	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestNP, "/HM/CPU%d/Exit/Trap/Gst/#NP", "Guest #NP (segment not present) exception.");
747	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestGP, "/HM/CPU%d/Exit/Trap/Gst/#GP", "Guest #GP (general protection) exception.");
748	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestMF, "/HM/CPU%d/Exit/Trap/Gst/#MF", "Guest #MF (x87 FPU error, math fault) exception.");
749	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestDE, "/HM/CPU%d/Exit/Trap/Gst/#DE", "Guest #DE (divide error) exception.");
750	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestDB, "/HM/CPU%d/Exit/Trap/Gst/#DB", "Guest #DB (debug) exception.");
751	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestBP, "/HM/CPU%d/Exit/Trap/Gst/#BP", "Guest #BP (breakpoint) exception.");
752	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestXF, "/HM/CPU%d/Exit/Trap/Gst/#XF", "Guest #XF (extended math fault, SIMD FPU) exception.");
753	HM_REG_COUNTER(&pVCpu->hm.s.StatExitGuestXcpUnk, "/HM/CPU%d/Exit/Trap/Gst/Other", "Other guest exceptions.");
754	HM_REG_COUNTER(&pVCpu->hm.s.StatExitRdmsr, "/HM/CPU%d/Exit/Instr/Rdmsr", "MSR read.");
755	HM_REG_COUNTER(&pVCpu->hm.s.StatExitWrmsr, "/HM/CPU%d/Exit/Instr/Wrmsr", "MSR write.");
756	HM_REG_COUNTER(&pVCpu->hm.s.StatExitDRxWrite, "/HM/CPU%d/Exit/Instr/DR-Write", "Debug register write.");
757	HM_REG_COUNTER(&pVCpu->hm.s.StatExitDRxRead, "/HM/CPU%d/Exit/Instr/DR-Read", "Debug register read.");
758	HM_REG_COUNTER(&pVCpu->hm.s.StatExitCR0Read, "/HM/CPU%d/Exit/Instr/CR-Read/CR0", "CR0 read.");
759	HM_REG_COUNTER(&pVCpu->hm.s.StatExitCR2Read, "/HM/CPU%d/Exit/Instr/CR-Read/CR2", "CR2 read.");
760	HM_REG_COUNTER(&pVCpu->hm.s.StatExitCR3Read, "/HM/CPU%d/Exit/Instr/CR-Read/CR3", "CR3 read.");
761	HM_REG_COUNTER(&pVCpu->hm.s.StatExitCR4Read, "/HM/CPU%d/Exit/Instr/CR-Read/CR4", "CR4 read.");
762	HM_REG_COUNTER(&pVCpu->hm.s.StatExitCR8Read, "/HM/CPU%d/Exit/Instr/CR-Read/CR8", "CR8 read.");
763	HM_REG_COUNTER(&pVCpu->hm.s.StatExitCR0Write, "/HM/CPU%d/Exit/Instr/CR-Write/CR0", "CR0 write.");
764	HM_REG_COUNTER(&pVCpu->hm.s.StatExitCR2Write, "/HM/CPU%d/Exit/Instr/CR-Write/CR2", "CR2 write.");
765	HM_REG_COUNTER(&pVCpu->hm.s.StatExitCR3Write, "/HM/CPU%d/Exit/Instr/CR-Write/CR3", "CR3 write.");
766	HM_REG_COUNTER(&pVCpu->hm.s.StatExitCR4Write, "/HM/CPU%d/Exit/Instr/CR-Write/CR4", "CR4 write.");
767	HM_REG_COUNTER(&pVCpu->hm.s.StatExitCR8Write, "/HM/CPU%d/Exit/Instr/CR-Write/CR8", "CR8 write.");
768	HM_REG_COUNTER(&pVCpu->hm.s.StatExitClts, "/HM/CPU%d/Exit/Instr/CLTS", "CLTS instruction.");
769	HM_REG_COUNTER(&pVCpu->hm.s.StatExitLmsw, "/HM/CPU%d/Exit/Instr/LMSW", "LMSW instruction.");
770	HM_REG_COUNTER(&pVCpu->hm.s.StatExitXdtrAccess, "/HM/CPU%d/Exit/Instr/XdtrAccess", "GDTR, IDTR, LDTR access.");
771	HM_REG_COUNTER(&pVCpu->hm.s.StatExitIOWrite, "/HM/CPU%d/Exit/Instr/IO/Write", "I/O write.");
772	HM_REG_COUNTER(&pVCpu->hm.s.StatExitIORead, "/HM/CPU%d/Exit/Instr/IO/Read", "I/O read.");
773	HM_REG_COUNTER(&pVCpu->hm.s.StatExitIOStringWrite, "/HM/CPU%d/Exit/Instr/IO/WriteString", "String I/O write.");
774	HM_REG_COUNTER(&pVCpu->hm.s.StatExitIOStringRead, "/HM/CPU%d/Exit/Instr/IO/ReadString", "String I/O read.");
775	HM_REG_COUNTER(&pVCpu->hm.s.StatExitIntWindow, "/HM/CPU%d/Exit/IntWindow", "Interrupt-window exit. Guest is ready to receive interrupts.");
776	HM_REG_COUNTER(&pVCpu->hm.s.StatExitExtInt, "/HM/CPU%d/Exit/ExtInt", "Physical maskable interrupt (host).");
777	#endif
778	HM_REG_COUNTER(&pVCpu->hm.s.StatExitHostNmiInGC, "/HM/CPU%d/Exit/HostNmiInGC", "Host NMI received while in guest context.");
779	#ifdef VBOX_WITH_STATISTICS
780	HM_REG_COUNTER(&pVCpu->hm.s.StatExitPreemptTimer, "/HM/CPU%d/Exit/PreemptTimer", "VMX-preemption timer expired.");
781	HM_REG_COUNTER(&pVCpu->hm.s.StatExitTprBelowThreshold, "/HM/CPU%d/Exit/TprBelowThreshold", "TPR lowered below threshold by the guest.");
782	HM_REG_COUNTER(&pVCpu->hm.s.StatExitTaskSwitch, "/HM/CPU%d/Exit/TaskSwitch", "Task switch.");
783	HM_REG_COUNTER(&pVCpu->hm.s.StatExitApicAccess, "/HM/CPU%d/Exit/ApicAccess", "APIC access. Guest attempted to access memory at a physical address on the APIC-access page.");
784
785	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchTprMaskedIrq, "/HM/CPU%d/Switch/TprMaskedIrq", "PDMGetInterrupt() signals TPR masks pending Irq.");
786	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchGuestIrq, "/HM/CPU%d/Switch/IrqPending", "PDMGetInterrupt() cleared behind our back!?!.");
787	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchPendingHostIrq, "/HM/CPU%d/Switch/PendingHostIrq", "Exit to ring-3 due to pending host interrupt before executing guest code.");
788	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchHmToR3FF, "/HM/CPU%d/Switch/HmToR3FF", "Exit to ring-3 due to pending timers, EMT rendezvous, critical section etc.");
789	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchExitToR3, "/HM/CPU%d/Switch/ExitToR3", "Exit to ring-3 (total).");
790	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchLongJmpToR3, "/HM/CPU%d/Switch/LongJmpToR3", "Longjump to ring-3.");
791	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchMaxResumeLoops, "/HM/CPU%d/Switch/MaxResumeToR3", "Maximum VMRESUME inner-loop counter reached.");
792	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchHltToR3, "/HM/CPU%d/Switch/HltToR3", "HLT causing us to go to ring-3.");
793	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchApicAccessToR3, "/HM/CPU%d/Switch/ApicAccessToR3", "APIC access causing us to go to ring-3.");
794	#endif
795	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchPreempt, "/HM/CPU%d/Switch/Preempting", "EMT has been preempted while in HM context.");
796	#ifdef VBOX_WITH_STATISTICS
797	HM_REG_COUNTER(&pVCpu->hm.s.StatSwitchPreemptExportHostState, "/HM/CPU%d/Switch/ExportHostState", "Preemption caused us to re-export the host state.");
798
799	HM_REG_COUNTER(&pVCpu->hm.s.StatInjectInterrupt, "/HM/CPU%d/EventInject/Interrupt", "Injected an external interrupt into the guest.");
800	HM_REG_COUNTER(&pVCpu->hm.s.StatInjectXcpt, "/HM/CPU%d/EventInject/Trap", "Injected an exception into the guest.");
801	HM_REG_COUNTER(&pVCpu->hm.s.StatInjectPendingReflect, "/HM/CPU%d/EventInject/PendingReflect", "Reflecting an exception (or #DF) caused due to event injection.");
802	HM_REG_COUNTER(&pVCpu->hm.s.StatInjectPendingInterpret, "/HM/CPU%d/EventInject/PendingInterpret", "Falling to interpreter for handling exception caused due to event injection.");
803
804	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushPage, "/HM/CPU%d/Flush/Page", "Invalidating a guest page on all guest CPUs.");
805	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushPageManual, "/HM/CPU%d/Flush/Page/Virt", "Invalidating a guest page using guest-virtual address.");
806	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushPhysPageManual, "/HM/CPU%d/Flush/Page/Phys", "Invalidating a guest page using guest-physical address.");
807	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushTlb, "/HM/CPU%d/Flush/TLB", "Forcing a full guest-TLB flush (ring-0).");
808	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushTlbManual, "/HM/CPU%d/Flush/TLB/Manual", "Request a full guest-TLB flush.");
809	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushTlbWorldSwitch, "/HM/CPU%d/Flush/TLB/CpuSwitch", "Forcing a full guest-TLB flush due to host-CPU reschedule or ASID-limit hit by another guest-VCPU.");
810	HM_REG_COUNTER(&pVCpu->hm.s.StatNoFlushTlbWorldSwitch, "/HM/CPU%d/Flush/TLB/Skipped", "No TLB flushing required.");
811	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushEntire, "/HM/CPU%d/Flush/TLB/Entire", "Flush the entire TLB (host + guest).");
812	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushAsid, "/HM/CPU%d/Flush/TLB/ASID", "Flushed guest-TLB entries for the current VPID.");
813	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushNestedPaging, "/HM/CPU%d/Flush/TLB/NestedPaging", "Flushed guest-TLB entries for the current EPT.");
814	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushTlbInvlpgVirt, "/HM/CPU%d/Flush/TLB/InvlpgVirt", "Invalidated a guest-TLB entry for a guest-virtual address.");
815	HM_REG_COUNTER(&pVCpu->hm.s.StatFlushTlbInvlpgPhys, "/HM/CPU%d/Flush/TLB/InvlpgPhys", "Currently not possible, flushes entire guest-TLB.");
816	HM_REG_COUNTER(&pVCpu->hm.s.StatTlbShootdown, "/HM/CPU%d/Flush/Shootdown/Page", "Inter-VCPU request to flush queued guest page.");
817	HM_REG_COUNTER(&pVCpu->hm.s.StatTlbShootdownFlush, "/HM/CPU%d/Flush/Shootdown/TLB", "Inter-VCPU request to flush entire guest-TLB.");
818
819	HM_REG_COUNTER(&pVCpu->hm.s.StatTscParavirt, "/HM/CPU%d/TSC/Paravirt", "Paravirtualized TSC in effect.");
820	HM_REG_COUNTER(&pVCpu->hm.s.StatTscOffset, "/HM/CPU%d/TSC/Offset", "TSC offsetting is in effect.");
821	HM_REG_COUNTER(&pVCpu->hm.s.StatTscIntercept, "/HM/CPU%d/TSC/Intercept", "Intercept TSC accesses.");
822
823	HM_REG_COUNTER(&pVCpu->hm.s.StatDRxArmed, "/HM/CPU%d/Debug/Armed", "Loaded guest-debug state while loading guest-state.");
824	HM_REG_COUNTER(&pVCpu->hm.s.StatDRxContextSwitch, "/HM/CPU%d/Debug/ContextSwitch", "Loaded guest-debug state on MOV DRx.");
825	HM_REG_COUNTER(&pVCpu->hm.s.StatDRxIoCheck, "/HM/CPU%d/Debug/IOCheck", "Checking for I/O breakpoint.");
826
827	HM_REG_COUNTER(&pVCpu->hm.s.StatExportMinimal, "/HM/CPU%d/Export/Minimal", "VM-entry exporting minimal guest-state.");
828	HM_REG_COUNTER(&pVCpu->hm.s.StatExportFull, "/HM/CPU%d/Export/Full", "VM-entry exporting the full guest-state.");
829	HM_REG_COUNTER(&pVCpu->hm.s.StatLoadGuestFpu, "/HM/CPU%d/Export/GuestFpu", "VM-entry loading the guest-FPU state.");
830
831	HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadRmSelBase, "/HM/CPU%d/VMXCheck/RMSelBase", "Could not use VMX due to unsuitable real-mode selector base.");
832	HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadRmSelLimit, "/HM/CPU%d/VMXCheck/RMSelLimit", "Could not use VMX due to unsuitable real-mode selector limit.");
833	HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadRmSelAttr, "/HM/CPU%d/VMXCheck/RMSelAttrs", "Could not use VMX due to unsuitable real-mode selector attributes.");
834
835	HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadV86SelBase, "/HM/CPU%d/VMXCheck/V86SelBase", "Could not use VMX due to unsuitable v8086-mode selector base.");
836	HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadV86SelLimit, "/HM/CPU%d/VMXCheck/V86SelLimit", "Could not use VMX due to unsuitable v8086-mode selector limit.");
837	HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadV86SelAttr, "/HM/CPU%d/VMXCheck/V86SelAttrs", "Could not use VMX due to unsuitable v8086-mode selector attributes.");
838
839	HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckRmOk, "/HM/CPU%d/VMXCheck/VMX_RM", "VMX execution in real (V86) mode OK.");
840	HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadSel, "/HM/CPU%d/VMXCheck/Selector", "Could not use VMX due to unsuitable selector.");
841	HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckBadRpl, "/HM/CPU%d/VMXCheck/RPL", "Could not use VMX due to unsuitable RPL.");
842	HM_REG_COUNTER(&pVCpu->hm.s.StatVmxCheckPmOk, "/HM/CPU%d/VMXCheck/VMX_PM", "VMX execution in protected mode OK.");
843
844	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
845	HM_REG_COUNTER(&pVCpu->hm.s.StatFpu64SwitchBack, "/HM/CPU%d/Switch64/Fpu", "Saving guest FPU/XMM state.");
846	HM_REG_COUNTER(&pVCpu->hm.s.StatDebug64SwitchBack, "/HM/CPU%d/Switch64/Debug", "Saving guest debug state.");
847	#endif
848
849	#undef HM_REG_COUNTER
850
851	bool const fCpuSupportsVmx = ASMIsIntelCpu() \|\| ASMIsViaCentaurCpu() \|\| ASMIsShanghaiCpu();
852
853	/*
854	* Guest Exit reason stats.
855	*/
856	pVCpu->hm.s.paStatExitReason = NULL;
857	rc = MMHyperAlloc(pVM, MAX_EXITREASON_STAT * sizeof(pVCpu->hm.s.paStatExitReason), 0 / uAlignment */, MM_TAG_HM,
858	(void **)&pVCpu->hm.s.paStatExitReason);
859	AssertRCReturn(rc, rc);
860
861	if (fCpuSupportsVmx)
862	{
863	for (int j = 0; j < MAX_EXITREASON_STAT; j++)
864	{
865	const char *pszExitName = HMGetVmxExitName(j);
866	if (pszExitName)
867	{
868	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.paStatExitReason[j], STAMTYPE_COUNTER, STAMVISIBILITY_USED,
869	STAMUNIT_OCCURENCES, pszExitName, "/HM/CPU%d/Exit/Reason/%02x", i, j);
870	AssertRCReturn(rc, rc);
871	}
872	}
873	}
874	else
875	{
876	for (int j = 0; j < MAX_EXITREASON_STAT; j++)
877	{
878	const char *pszExitName = HMGetSvmExitName(j);
879	if (pszExitName)
880	{
881	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.paStatExitReason[j], STAMTYPE_COUNTER, STAMVISIBILITY_USED,
882	STAMUNIT_OCCURENCES, pszExitName, "/HM/CPU%d/Exit/Reason/%02x", i, j);
883	AssertRCReturn(rc, rc);
884	}
885	}
886	}
887	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatExitReasonNpf, STAMTYPE_COUNTER, STAMVISIBILITY_USED, STAMUNIT_OCCURENCES,
888	"Nested page fault", "/HM/CPU%d/Exit/Reason/#NPF", i);
889	AssertRCReturn(rc, rc);
890	pVCpu->hm.s.paStatExitReasonR0 = MMHyperR3ToR0(pVM, pVCpu->hm.s.paStatExitReason);
891	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE
892	Assert(pVCpu->hm.s.paStatExitReasonR0 != NIL_RTR0PTR \|\| !HMIsEnabled(pVM));
893	# else
894	Assert(pVCpu->hm.s.paStatExitReasonR0 != NIL_RTR0PTR);
895	# endif
896
897	#if defined(VBOX_WITH_NESTED_HWVIRT_SVM) \|\| defined(VBOX_WITH_NESTED_HWVIRT_VMX)
898	/*
899	* Nested-guest VM-exit reason stats.
900	*/
901	pVCpu->hm.s.paStatNestedExitReason = NULL;
902	rc = MMHyperAlloc(pVM, MAX_EXITREASON_STAT * sizeof(pVCpu->hm.s.paStatNestedExitReason), 0 / uAlignment */, MM_TAG_HM,
903	(void **)&pVCpu->hm.s.paStatNestedExitReason);
904	AssertRCReturn(rc, rc);
905	if (fCpuSupportsVmx)
906	{
907	for (int j = 0; j < MAX_EXITREASON_STAT; j++)
908	{
909	const char *pszExitName = HMGetVmxExitName(j);
910	if (pszExitName)
911	{
912	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.paStatNestedExitReason[j], STAMTYPE_COUNTER, STAMVISIBILITY_USED,
913	STAMUNIT_OCCURENCES, pszExitName, "/HM/CPU%d/NestedExit/Reason/%02x", i, j);
914	AssertRC(rc);
915	}
916	}
917	}
918	else
919	{
920	for (int j = 0; j < MAX_EXITREASON_STAT; j++)
921	{
922	const char *pszExitName = HMGetSvmExitName(j);
923	if (pszExitName)
924	{
925	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.paStatNestedExitReason[j], STAMTYPE_COUNTER, STAMVISIBILITY_USED,
926	STAMUNIT_OCCURENCES, pszExitName, "/HM/CPU%d/NestedExit/Reason/%02x", i, j);
927	AssertRC(rc);
928	}
929	}
930	}
931	rc = STAMR3RegisterF(pVM, &pVCpu->hm.s.StatNestedExitReasonNpf, STAMTYPE_COUNTER, STAMVISIBILITY_USED,
932	STAMUNIT_OCCURENCES, "Nested page fault", "/HM/CPU%d/NestedExit/Reason/#NPF", i);
933	AssertRCReturn(rc, rc);
934	pVCpu->hm.s.paStatNestedExitReasonR0 = MMHyperR3ToR0(pVM, pVCpu->hm.s.paStatNestedExitReason);
935	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE
936	Assert(pVCpu->hm.s.paStatNestedExitReasonR0 != NIL_RTR0PTR \|\| !HMIsEnabled(pVM));
937	# else
938	Assert(pVCpu->hm.s.paStatNestedExitReasonR0 != NIL_RTR0PTR);
939	# endif
940	#endif
941
942	/*
943	* Injected events stats.
944	*/
945	rc = MMHyperAlloc(pVM, sizeof(STAMCOUNTER) * 256, 8, MM_TAG_HM, (void **)&pVCpu->hm.s.paStatInjectedIrqs);
946	AssertRCReturn(rc, rc);
947	pVCpu->hm.s.paStatInjectedIrqsR0 = MMHyperR3ToR0(pVM, pVCpu->hm.s.paStatInjectedIrqs);
948	# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE
949	Assert(pVCpu->hm.s.paStatInjectedIrqsR0 != NIL_RTR0PTR \|\| !HMIsEnabled(pVM));
950	# else
951	Assert(pVCpu->hm.s.paStatInjectedIrqsR0 != NIL_RTR0PTR);
952	# endif
953	for (unsigned j = 0; j < 255; j++)
954	{
955	STAMR3RegisterF(pVM, &pVCpu->hm.s.paStatInjectedIrqs[j], STAMTYPE_COUNTER, STAMVISIBILITY_USED, STAMUNIT_OCCURENCES,
956	"Injected event.",
957	(j < 0x20) ? "/HM/CPU%d/EventInject/InjectTrap/%02X" : "/HM/CPU%d/EventInject/InjectIRQ/%02X", i, j);
958	}
959
960	#endif /* VBOX_WITH_STATISTICS */
961	}
962
963	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
964	/*
965	* Magic marker for searching in crash dumps.
966	*/
967	for (VMCPUID i = 0; i < pVM->cCpus; i++)
968	{
969	PVMCPU pVCpu = &pVM->aCpus[i];
970
971	PVMXVMCSCACHE pVmcsCache = &pVCpu->hm.s.vmx.VmcsCache;
972	strcpy((char *)pVmcsCache->aMagic, "VMCSCACHE Magic");
973	pVmcsCache->uMagic = UINT64_C(0xdeadbeefdeadbeef);
974	}
975	#endif
976
977	return VINF_SUCCESS;
978	}
979
980
981	/**
982	* Called when a init phase has completed.
983	*
984	* @returns VBox status code.
985	* @param pVM The cross context VM structure.
986	* @param enmWhat The phase that completed.
987	*/
988	VMMR3_INT_DECL(int) HMR3InitCompleted(PVM pVM, VMINITCOMPLETED enmWhat)
989	{
990	switch (enmWhat)
991	{
992	case VMINITCOMPLETED_RING3:
993	return hmR3InitFinalizeR3(pVM);
994	case VMINITCOMPLETED_RING0:
995	return hmR3InitFinalizeR0(pVM);
996	default:
997	return VINF_SUCCESS;
998	}
999	}
1000
1001
1002	/**
1003	* Turns off normal raw mode features.
1004	*
1005	* @param pVM The cross context VM structure.
1006	*/
1007	static void hmR3DisableRawMode(PVM pVM)
1008	{
1009	/** @todo r=bird: HM shouldn't be doing this crap. */
1010	/* Reinit the paging mode to force the new shadow mode. */
1011	for (VMCPUID i = 0; i < pVM->cCpus; i++)
1012	{
1013	PVMCPU pVCpu = &pVM->aCpus[i];
1014	PGMHCChangeMode(pVM, pVCpu, PGMMODE_REAL);
1015	}
1016	}
1017
1018
1019	/**
1020	* Initialize VT-x or AMD-V.
1021	*
1022	* @returns VBox status code.
1023	* @param pVM The cross context VM structure.
1024	*/
1025	static int hmR3InitFinalizeR0(PVM pVM)
1026	{
1027	int rc;
1028
1029	if (!HMIsEnabled(pVM))
1030	return VINF_SUCCESS;
1031
1032	/*
1033	* Hack to allow users to work around broken BIOSes that incorrectly set
1034	* EFER.SVME, which makes us believe somebody else is already using AMD-V.
1035	*/
1036	if ( !pVM->hm.s.vmx.fSupported
1037	&& !pVM->hm.s.svm.fSupported
1038	&& pVM->hm.s.rcInit == VERR_SVM_IN_USE /* implies functional AMD-V */
1039	&& RTEnvExist("VBOX_HWVIRTEX_IGNORE_SVM_IN_USE"))
1040	{
1041	LogRel(("HM: VBOX_HWVIRTEX_IGNORE_SVM_IN_USE active!\n"));
1042	pVM->hm.s.svm.fSupported = true;
1043	pVM->hm.s.svm.fIgnoreInUseError = true;
1044	pVM->hm.s.rcInit = VINF_SUCCESS;
1045	}
1046
1047	/*
1048	* Report ring-0 init errors.
1049	*/
1050	if ( !pVM->hm.s.vmx.fSupported
1051	&& !pVM->hm.s.svm.fSupported)
1052	{
1053	LogRel(("HM: Failed to initialize VT-x / AMD-V: %Rrc\n", pVM->hm.s.rcInit));
1054	LogRel(("HM: VMX MSR_IA32_FEATURE_CONTROL=%RX64\n", pVM->hm.s.vmx.Msrs.u64FeatCtrl));
1055	switch (pVM->hm.s.rcInit)
1056	{
1057	case VERR_VMX_IN_VMX_ROOT_MODE:
1058	return VM_SET_ERROR(pVM, VERR_VMX_IN_VMX_ROOT_MODE, "VT-x is being used by another hypervisor");
1059	case VERR_VMX_NO_VMX:
1060	return VM_SET_ERROR(pVM, VERR_VMX_NO_VMX, "VT-x is not available");
1061	case VERR_VMX_MSR_VMX_DISABLED:
1062	return VM_SET_ERROR(pVM, VERR_VMX_MSR_VMX_DISABLED, "VT-x is disabled in the BIOS");
1063	case VERR_VMX_MSR_ALL_VMX_DISABLED:
1064	return VM_SET_ERROR(pVM, VERR_VMX_MSR_ALL_VMX_DISABLED, "VT-x is disabled in the BIOS for all CPU modes");
1065	case VERR_VMX_MSR_LOCKING_FAILED:
1066	return VM_SET_ERROR(pVM, VERR_VMX_MSR_LOCKING_FAILED, "Failed to lock VT-x features while trying to enable VT-x");
1067	case VERR_VMX_MSR_VMX_ENABLE_FAILED:
1068	return VM_SET_ERROR(pVM, VERR_VMX_MSR_VMX_ENABLE_FAILED, "Failed to enable VT-x features");
1069	case VERR_VMX_MSR_SMX_VMX_ENABLE_FAILED:
1070	return VM_SET_ERROR(pVM, VERR_VMX_MSR_SMX_VMX_ENABLE_FAILED, "Failed to enable VT-x features in SMX mode");
1071
1072	case VERR_SVM_IN_USE:
1073	return VM_SET_ERROR(pVM, VERR_SVM_IN_USE, "AMD-V is being used by another hypervisor");
1074	case VERR_SVM_NO_SVM:
1075	return VM_SET_ERROR(pVM, VERR_SVM_NO_SVM, "AMD-V is not available");
1076	case VERR_SVM_DISABLED:
1077	return VM_SET_ERROR(pVM, VERR_SVM_DISABLED, "AMD-V is disabled in the BIOS");
1078	}
1079	return VMSetError(pVM, pVM->hm.s.rcInit, RT_SRC_POS, "HM ring-0 init failed: %Rrc", pVM->hm.s.rcInit);
1080	}
1081
1082	/*
1083	* Enable VT-x or AMD-V on all host CPUs.
1084	*/
1085	rc = SUPR3CallVMMR0Ex(pVM->pVMR0, 0 /idCpu/, VMMR0_DO_HM_ENABLE, 0, NULL);
1086	if (RT_FAILURE(rc))
1087	{
1088	LogRel(("HM: Failed to enable, error %Rrc\n", rc));
1089	HMR3CheckError(pVM, rc);
1090	return rc;
1091	}
1092
1093	/*
1094	* No TPR patching is required when the IO-APIC is not enabled for this VM.
1095	* (Main should have taken care of this already)
1096	*/
1097	if (!PDMHasIoApic(pVM))
1098	{
1099	Assert(!pVM->hm.s.fTprPatchingAllowed); /* paranoia */
1100	pVM->hm.s.fTprPatchingAllowed = false;
1101	}
1102
1103	/*
1104	* Check if L1D flush is needed/possible.
1105	*/
1106	if ( !pVM->cpum.ro.HostFeatures.fFlushCmd
1107	\|\| pVM->cpum.ro.HostFeatures.enmMicroarch < kCpumMicroarch_Intel_Core7_Nehalem
1108	\|\| pVM->cpum.ro.HostFeatures.enmMicroarch >= kCpumMicroarch_Intel_Core7_End
1109	\|\| pVM->cpum.ro.HostFeatures.fArchVmmNeedNotFlushL1d
1110	\|\| pVM->cpum.ro.HostFeatures.fArchRdclNo)
1111	pVM->hm.s.fL1dFlushOnSched = pVM->hm.s.fL1dFlushOnVmEntry = false;
1112
1113	/*
1114	* Check if MDS flush is needed/possible.
1115	* On atoms and knight family CPUs, we will only allow clearing on scheduling.
1116	*/
1117	if ( !pVM->cpum.ro.HostFeatures.fMdsClear
1118	\|\| pVM->cpum.ro.HostFeatures.fArchMdsNo)
1119	pVM->hm.s.fMdsClearOnSched = pVM->hm.s.fMdsClearOnVmEntry = false;
1120	else if ( ( pVM->cpum.ro.HostFeatures.enmMicroarch >= kCpumMicroarch_Intel_Atom_Airmount
1121	&& pVM->cpum.ro.HostFeatures.enmMicroarch < kCpumMicroarch_Intel_Atom_End)
1122	\|\| ( pVM->cpum.ro.HostFeatures.enmMicroarch >= kCpumMicroarch_Intel_Phi_KnightsLanding
1123	&& pVM->cpum.ro.HostFeatures.enmMicroarch < kCpumMicroarch_Intel_Phi_End))
1124	{
1125	if (!pVM->hm.s.fMdsClearOnSched)
1126	pVM->hm.s.fMdsClearOnSched = pVM->hm.s.fMdsClearOnVmEntry;
1127	pVM->hm.s.fMdsClearOnVmEntry = false;
1128	}
1129	else if ( pVM->cpum.ro.HostFeatures.enmMicroarch < kCpumMicroarch_Intel_Core7_Nehalem
1130	\|\| pVM->cpum.ro.HostFeatures.enmMicroarch >= kCpumMicroarch_Intel_Core7_End)
1131	pVM->hm.s.fMdsClearOnSched = pVM->hm.s.fMdsClearOnVmEntry = false;
1132
1133	/*
1134	* Sync options.
1135	*/
1136	/** @todo Move this out of of CPUMCTX and into some ring-0 only HM structure.
1137	* That will require a little bit of work, of course. */
1138	for (VMCPUID iCpu = 0; iCpu < pVM->cCpus; iCpu++)
1139	{
1140	PVMCPU pVCpu = &pVM->aCpus[iCpu];
1141	PCPUMCTX pCpuCtx = &pVCpu->cpum.GstCtx;
1142	pCpuCtx->fWorldSwitcher &= ~(CPUMCTX_WSF_IBPB_EXIT \| CPUMCTX_WSF_IBPB_ENTRY);
1143	if (pVM->cpum.ro.HostFeatures.fIbpb)
1144	{
1145	if (pVM->hm.s.fIbpbOnVmExit)
1146	pCpuCtx->fWorldSwitcher \|= CPUMCTX_WSF_IBPB_EXIT;
1147	if (pVM->hm.s.fIbpbOnVmEntry)
1148	pCpuCtx->fWorldSwitcher \|= CPUMCTX_WSF_IBPB_ENTRY;
1149	}
1150	if (pVM->cpum.ro.HostFeatures.fFlushCmd && pVM->hm.s.fL1dFlushOnVmEntry)
1151	pCpuCtx->fWorldSwitcher \|= CPUMCTX_WSF_L1D_ENTRY;
1152	if (pVM->cpum.ro.HostFeatures.fMdsClear && pVM->hm.s.fMdsClearOnVmEntry)
1153	pCpuCtx->fWorldSwitcher \|= CPUMCTX_WSF_MDS_ENTRY;
1154	if (iCpu == 0)
1155	LogRel(("HM: fWorldSwitcher=%#x (fIbpbOnVmExit=%RTbool fIbpbOnVmEntry=%RTbool fL1dFlushOnVmEntry=%RTbool); fL1dFlushOnSched=%RTbool fMdsClearOnVmEntry=%RTbool\n",
1156	pCpuCtx->fWorldSwitcher, pVM->hm.s.fIbpbOnVmExit, pVM->hm.s.fIbpbOnVmEntry, pVM->hm.s.fL1dFlushOnVmEntry,
1157	pVM->hm.s.fL1dFlushOnSched, pVM->hm.s.fMdsClearOnVmEntry));
1158	}
1159
1160	/*
1161	* Do the vendor specific initialization
1162	*
1163	* Note! We disable release log buffering here since we're doing relatively
1164	* lot of logging and doesn't want to hit the disk with each LogRel
1165	* statement.
1166	*/
1167	AssertLogRelReturn(!pVM->hm.s.fInitialized, VERR_HM_IPE_5);
1168	bool fOldBuffered = RTLogRelSetBuffering(true /fBuffered/);
1169	if (pVM->hm.s.vmx.fSupported)
1170	rc = hmR3InitFinalizeR0Intel(pVM);
1171	else
1172	rc = hmR3InitFinalizeR0Amd(pVM);
1173	LogRel(("HM: VT-x/AMD-V init method: %s\n", (pVM->hm.s.fGlobalInit) ? "GLOBAL" : "LOCAL"));
1174	RTLogRelSetBuffering(fOldBuffered);
1175	pVM->hm.s.fInitialized = true;
1176
1177	return rc;
1178	}
1179
1180
1181	/**
1182	* @callback_method_impl{FNPDMVMMDEVHEAPNOTIFY}
1183	*/
1184	static DECLCALLBACK(void) hmR3VmmDevHeapNotify(PVM pVM, void *pvAllocation, RTGCPHYS GCPhysAllocation)
1185	{
1186	NOREF(pVM);
1187	NOREF(pvAllocation);
1188	NOREF(GCPhysAllocation);
1189	}
1190
1191
1192	/**
1193	* Returns a description of the VMCS (and associated regions') memory type given the
1194	* IA32_VMX_BASIC MSR.
1195	*
1196	* @returns The descriptive memory type.
1197	* @param uMsrVmxBasic IA32_VMX_BASIC MSR value.
1198	*/
1199	static const char *hmR3VmxGetMemTypeDesc(uint64_t uMsrVmxBasic)
1200	{
1201	uint8_t const uMemType = RT_BF_GET(uMsrVmxBasic, VMX_BF_BASIC_VMCS_MEM_TYPE);
1202	switch (uMemType)
1203	{
1204	case VMX_BASIC_MEM_TYPE_WB: return "Write Back (WB)";
1205	case VMX_BASIC_MEM_TYPE_UC: return "Uncacheable (UC)";
1206	}
1207	return "Unknown";
1208	}
1209
1210
1211	/**
1212	* Returns a single-line description of all the activity-states supported by the CPU
1213	* given the IA32_VMX_MISC MSR.
1214	*
1215	* @returns All supported activity states.
1216	* @param uMsrMisc IA32_VMX_MISC MSR value.
1217	*/
1218	static const char *hmR3VmxGetActivityStateAllDesc(uint64_t uMsrMisc)
1219	{
1220	static const char * const s_apszActStates[] =
1221	{
1222	"",
1223	" ( HLT )",
1224	" ( SHUTDOWN )",
1225	" ( HLT SHUTDOWN )",
1226	" ( SIPI_WAIT )",
1227	" ( HLT SIPI_WAIT )",
1228	" ( SHUTDOWN SIPI_WAIT )",
1229	" ( HLT SHUTDOWN SIPI_WAIT )"
1230	};
1231	uint8_t const idxActStates = RT_BF_GET(uMsrMisc, VMX_BF_MISC_ACTIVITY_STATES);
1232	Assert(idxActStates < RT_ELEMENTS(s_apszActStates));
1233	return s_apszActStates[idxActStates];
1234	}
1235
1236
1237	/**
1238	* Reports MSR_IA32_FEATURE_CONTROL MSR to the log.
1239	*
1240	* @param fFeatMsr The feature control MSR value.
1241	*/
1242	static void hmR3VmxReportFeatCtlMsr(uint64_t fFeatMsr)
1243	{
1244	uint64_t const val = fFeatMsr;
1245	LogRel(("HM: MSR_IA32_FEATURE_CONTROL = %#RX64\n", val));
1246	HMVMX_REPORT_MSR_CAP(val, "LOCK", MSR_IA32_FEATURE_CONTROL_LOCK);
1247	HMVMX_REPORT_MSR_CAP(val, "SMX_VMXON", MSR_IA32_FEATURE_CONTROL_SMX_VMXON);
1248	HMVMX_REPORT_MSR_CAP(val, "VMXON", MSR_IA32_FEATURE_CONTROL_VMXON);
1249	HMVMX_REPORT_MSR_CAP(val, "SENTER_LOCAL_FN0", MSR_IA32_FEATURE_CONTROL_SENTER_LOCAL_FN_0);
1250	HMVMX_REPORT_MSR_CAP(val, "SENTER_LOCAL_FN1", MSR_IA32_FEATURE_CONTROL_SENTER_LOCAL_FN_1);
1251	HMVMX_REPORT_MSR_CAP(val, "SENTER_LOCAL_FN2", MSR_IA32_FEATURE_CONTROL_SENTER_LOCAL_FN_2);
1252	HMVMX_REPORT_MSR_CAP(val, "SENTER_LOCAL_FN3", MSR_IA32_FEATURE_CONTROL_SENTER_LOCAL_FN_3);
1253	HMVMX_REPORT_MSR_CAP(val, "SENTER_LOCAL_FN4", MSR_IA32_FEATURE_CONTROL_SENTER_LOCAL_FN_4);
1254	HMVMX_REPORT_MSR_CAP(val, "SENTER_LOCAL_FN5", MSR_IA32_FEATURE_CONTROL_SENTER_LOCAL_FN_5);
1255	HMVMX_REPORT_MSR_CAP(val, "SENTER_LOCAL_FN6", MSR_IA32_FEATURE_CONTROL_SENTER_LOCAL_FN_6);
1256	HMVMX_REPORT_MSR_CAP(val, "SENTER_GLOBAL_EN", MSR_IA32_FEATURE_CONTROL_SENTER_GLOBAL_EN);
1257	HMVMX_REPORT_MSR_CAP(val, "SGX_LAUNCH_EN", MSR_IA32_FEATURE_CONTROL_SGX_LAUNCH_EN);
1258	HMVMX_REPORT_MSR_CAP(val, "SGX_GLOBAL_EN", MSR_IA32_FEATURE_CONTROL_SGX_GLOBAL_EN);
1259	HMVMX_REPORT_MSR_CAP(val, "LMCE", MSR_IA32_FEATURE_CONTROL_LMCE);
1260	if (!(val & MSR_IA32_FEATURE_CONTROL_LOCK))
1261	LogRel(("HM: MSR_IA32_FEATURE_CONTROL lock bit not set, possibly bad hardware!\n"));
1262	}
1263
1264
1265	/**
1266	* Reports MSR_IA32_VMX_BASIC MSR to the log.
1267	*
1268	* @param uBasicMsr The VMX basic MSR value.
1269	*/
1270	static void hmR3VmxReportBasicMsr(uint64_t uBasicMsr)
1271	{
1272	LogRel(("HM: MSR_IA32_VMX_BASIC = %#RX64\n", uBasicMsr));
1273	LogRel(("HM: VMCS id = %#x\n", RT_BF_GET(uBasicMsr, VMX_BF_BASIC_VMCS_ID)));
1274	LogRel(("HM: VMCS size = %u bytes\n", RT_BF_GET(uBasicMsr, VMX_BF_BASIC_VMCS_SIZE)));
1275	LogRel(("HM: VMCS physical address limit = %s\n", RT_BF_GET(uBasicMsr, VMX_BF_BASIC_PHYSADDR_WIDTH) ?
1276	"< 4 GB" : "None"));
1277	LogRel(("HM: VMCS memory type = %s\n", hmR3VmxGetMemTypeDesc(uBasicMsr)));
1278	LogRel(("HM: Dual-monitor treatment support = %RTbool\n", RT_BF_GET(uBasicMsr, VMX_BF_BASIC_DUAL_MON)));
1279	LogRel(("HM: OUTS & INS instruction-info = %RTbool\n", RT_BF_GET(uBasicMsr, VMX_BF_BASIC_VMCS_INS_OUTS)));
1280	LogRel(("HM: Supports true capability MSRs = %RTbool\n", RT_BF_GET(uBasicMsr, VMX_BF_BASIC_TRUE_CTLS)));
1281	}
1282
1283
1284	/**
1285	* Reports MSR_IA32_PINBASED_CTLS to the log.
1286	*
1287	* @param pVmxMsr Pointer to the VMX MSR.
1288	*/
1289	static void hmR3VmxReportPinBasedCtlsMsr(PCVMXCTLSMSR pVmxMsr)
1290	{
1291	uint64_t const fAllowed1 = pVmxMsr->n.allowed1;
1292	uint64_t const fAllowed0 = pVmxMsr->n.allowed0;
1293	LogRel(("HM: MSR_IA32_VMX_PINBASED_CTLS = %#RX64\n", pVmxMsr->u));
1294	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "EXT_INT_EXIT", VMX_PIN_CTLS_EXT_INT_EXIT);
1295	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "NMI_EXIT", VMX_PIN_CTLS_NMI_EXIT);
1296	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "VIRTUAL_NMI", VMX_PIN_CTLS_VIRT_NMI);
1297	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "PREEMPT_TIMER", VMX_PIN_CTLS_PREEMPT_TIMER);
1298	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "POSTED_INT", VMX_PIN_CTLS_POSTED_INT);
1299	}
1300
1301
1302	/**
1303	* Reports MSR_IA32_VMX_PROCBASED_CTLS MSR to the log.
1304	*
1305	* @param pVmxMsr Pointer to the VMX MSR.
1306	*/
1307	static void hmR3VmxReportProcBasedCtlsMsr(PCVMXCTLSMSR pVmxMsr)
1308	{
1309	uint64_t const fAllowed1 = pVmxMsr->n.allowed1;
1310	uint64_t const fAllowed0 = pVmxMsr->n.allowed0;
1311	LogRel(("HM: MSR_IA32_VMX_PROCBASED_CTLS = %#RX64\n", pVmxMsr->u));
1312	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "INT_WINDOW_EXIT", VMX_PROC_CTLS_INT_WINDOW_EXIT);
1313	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "USE_TSC_OFFSETTING", VMX_PROC_CTLS_USE_TSC_OFFSETTING);
1314	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "HLT_EXIT", VMX_PROC_CTLS_HLT_EXIT);
1315	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "INVLPG_EXIT", VMX_PROC_CTLS_INVLPG_EXIT);
1316	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "MWAIT_EXIT", VMX_PROC_CTLS_MWAIT_EXIT);
1317	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "RDPMC_EXIT", VMX_PROC_CTLS_RDPMC_EXIT);
1318	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "RDTSC_EXIT", VMX_PROC_CTLS_RDTSC_EXIT);
1319	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "CR3_LOAD_EXIT", VMX_PROC_CTLS_CR3_LOAD_EXIT);
1320	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "CR3_STORE_EXIT", VMX_PROC_CTLS_CR3_STORE_EXIT);
1321	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "CR8_LOAD_EXIT", VMX_PROC_CTLS_CR8_LOAD_EXIT);
1322	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "CR8_STORE_EXIT", VMX_PROC_CTLS_CR8_STORE_EXIT);
1323	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "USE_TPR_SHADOW", VMX_PROC_CTLS_USE_TPR_SHADOW);
1324	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "NMI_WINDOW_EXIT", VMX_PROC_CTLS_NMI_WINDOW_EXIT);
1325	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "MOV_DR_EXIT", VMX_PROC_CTLS_MOV_DR_EXIT);
1326	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "UNCOND_IO_EXIT", VMX_PROC_CTLS_UNCOND_IO_EXIT);
1327	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "USE_IO_BITMAPS", VMX_PROC_CTLS_USE_IO_BITMAPS);
1328	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "MONITOR_TRAP_FLAG", VMX_PROC_CTLS_MONITOR_TRAP_FLAG);
1329	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "USE_MSR_BITMAPS", VMX_PROC_CTLS_USE_MSR_BITMAPS);
1330	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "MONITOR_EXIT", VMX_PROC_CTLS_MONITOR_EXIT);
1331	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "PAUSE_EXIT", VMX_PROC_CTLS_PAUSE_EXIT);
1332	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "USE_SECONDARY_CTLS", VMX_PROC_CTLS_USE_SECONDARY_CTLS);
1333	}
1334
1335
1336	/**
1337	* Reports MSR_IA32_VMX_PROCBASED_CTLS2 MSR to the log.
1338	*
1339	* @param pVmxMsr Pointer to the VMX MSR.
1340	*/
1341	static void hmR3VmxReportProcBasedCtls2Msr(PCVMXCTLSMSR pVmxMsr)
1342	{
1343	uint64_t const fAllowed1 = pVmxMsr->n.allowed1;
1344	uint64_t const fAllowed0 = pVmxMsr->n.allowed0;
1345	LogRel(("HM: MSR_IA32_VMX_PROCBASED_CTLS2 = %#RX64\n", pVmxMsr->u));
1346	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "VIRT_APIC_ACCESS", VMX_PROC_CTLS2_VIRT_APIC_ACCESS);
1347	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "EPT", VMX_PROC_CTLS2_EPT);
1348	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "DESC_TABLE_EXIT", VMX_PROC_CTLS2_DESC_TABLE_EXIT);
1349	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "RDTSCP", VMX_PROC_CTLS2_RDTSCP);
1350	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "VIRT_X2APIC_MODE", VMX_PROC_CTLS2_VIRT_X2APIC_MODE);
1351	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "VPID", VMX_PROC_CTLS2_VPID);
1352	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "WBINVD_EXIT", VMX_PROC_CTLS2_WBINVD_EXIT);
1353	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "UNRESTRICTED_GUEST", VMX_PROC_CTLS2_UNRESTRICTED_GUEST);
1354	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "APIC_REG_VIRT", VMX_PROC_CTLS2_APIC_REG_VIRT);
1355	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "VIRT_INT_DELIVERY", VMX_PROC_CTLS2_VIRT_INT_DELIVERY);
1356	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "PAUSE_LOOP_EXIT", VMX_PROC_CTLS2_PAUSE_LOOP_EXIT);
1357	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "RDRAND_EXIT", VMX_PROC_CTLS2_RDRAND_EXIT);
1358	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "INVPCID", VMX_PROC_CTLS2_INVPCID);
1359	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "VMFUNC", VMX_PROC_CTLS2_VMFUNC);
1360	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "VMCS_SHADOWING", VMX_PROC_CTLS2_VMCS_SHADOWING);
1361	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "ENCLS_EXIT", VMX_PROC_CTLS2_ENCLS_EXIT);
1362	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "RDSEED_EXIT", VMX_PROC_CTLS2_RDSEED_EXIT);
1363	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "PML", VMX_PROC_CTLS2_PML);
1364	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "EPT_VE", VMX_PROC_CTLS2_EPT_VE);
1365	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "CONCEAL_VMX_FROM_PT", VMX_PROC_CTLS2_CONCEAL_VMX_FROM_PT);
1366	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "XSAVES_XRSTORS", VMX_PROC_CTLS2_XSAVES_XRSTORS);
1367	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "MODE_BASED_EPT_PERM", VMX_PROC_CTLS2_MODE_BASED_EPT_PERM);
1368	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "SPPTP_EPT", VMX_PROC_CTLS2_SPPTP_EPT);
1369	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "PT_EPT", VMX_PROC_CTLS2_PT_EPT);
1370	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "TSC_SCALING", VMX_PROC_CTLS2_TSC_SCALING);
1371	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "USER_WAIT_PAUSE", VMX_PROC_CTLS2_USER_WAIT_PAUSE);
1372	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "ENCLV_EXIT", VMX_PROC_CTLS2_ENCLV_EXIT);
1373	}
1374
1375
1376	/**
1377	* Reports MSR_IA32_VMX_ENTRY_CTLS to the log.
1378	*
1379	* @param pVmxMsr Pointer to the VMX MSR.
1380	*/
1381	static void hmR3VmxReportEntryCtlsMsr(PCVMXCTLSMSR pVmxMsr)
1382	{
1383	uint64_t const fAllowed1 = pVmxMsr->n.allowed1;
1384	uint64_t const fAllowed0 = pVmxMsr->n.allowed0;
1385	LogRel(("HM: MSR_IA32_VMX_ENTRY_CTLS = %#RX64\n", pVmxMsr->u));
1386	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "LOAD_DEBUG", VMX_ENTRY_CTLS_LOAD_DEBUG);
1387	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "IA32E_MODE_GUEST", VMX_ENTRY_CTLS_IA32E_MODE_GUEST);
1388	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "ENTRY_TO_SMM", VMX_ENTRY_CTLS_ENTRY_TO_SMM);
1389	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "DEACTIVATE_DUAL_MON", VMX_ENTRY_CTLS_DEACTIVATE_DUAL_MON);
1390	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "LOAD_PERF_MSR", VMX_ENTRY_CTLS_LOAD_PERF_MSR);
1391	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "LOAD_PAT_MSR", VMX_ENTRY_CTLS_LOAD_PAT_MSR);
1392	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "LOAD_EFER_MSR", VMX_ENTRY_CTLS_LOAD_EFER_MSR);
1393	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "LOAD_BNDCFGS_MSR", VMX_ENTRY_CTLS_LOAD_BNDCFGS_MSR);
1394	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "CONCEAL_VMX_FROM_PT", VMX_ENTRY_CTLS_CONCEAL_VMX_FROM_PT);
1395	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "LOAD_RTIT_CTL_MSR", VMX_ENTRY_CTLS_LOAD_RTIT_CTL_MSR);
1396	}
1397
1398
1399	/**
1400	* Reports MSR_IA32_VMX_EXIT_CTLS to the log.
1401	*
1402	* @param pVmxMsr Pointer to the VMX MSR.
1403	*/
1404	static void hmR3VmxReportExitCtlsMsr(PCVMXCTLSMSR pVmxMsr)
1405	{
1406	uint64_t const fAllowed1 = pVmxMsr->n.allowed1;
1407	uint64_t const fAllowed0 = pVmxMsr->n.allowed0;
1408	LogRel(("HM: MSR_IA32_VMX_EXIT_CTLS = %#RX64\n", pVmxMsr->u));
1409	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "SAVE_DEBUG", VMX_EXIT_CTLS_SAVE_DEBUG);
1410	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "HOST_ADDR_SPACE_SIZE", VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE);
1411	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "LOAD_PERF_MSR", VMX_EXIT_CTLS_LOAD_PERF_MSR);
1412	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "ACK_EXT_INT", VMX_EXIT_CTLS_ACK_EXT_INT);
1413	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "SAVE_PAT_MSR", VMX_EXIT_CTLS_SAVE_PAT_MSR);
1414	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "LOAD_PAT_MSR", VMX_EXIT_CTLS_LOAD_PAT_MSR);
1415	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "SAVE_EFER_MSR", VMX_EXIT_CTLS_SAVE_EFER_MSR);
1416	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "LOAD_EFER_MSR", VMX_EXIT_CTLS_LOAD_EFER_MSR);
1417	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "SAVE_PREEMPT_TIMER", VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER);
1418	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "CLEAR_BNDCFGS_MSR", VMX_EXIT_CTLS_CLEAR_BNDCFGS_MSR);
1419	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "CONCEAL_VMX_FROM_PT", VMX_EXIT_CTLS_CONCEAL_VMX_FROM_PT);
1420	HMVMX_REPORT_FEAT(fAllowed1, fAllowed0, "CLEAR_RTIT_CTL_MSR", VMX_EXIT_CTLS_CLEAR_RTIT_CTL_MSR);
1421	}
1422
1423
1424	/**
1425	* Reports MSR_IA32_VMX_EPT_VPID_CAP MSR to the log.
1426	*
1427	* @param fCaps The VMX EPT/VPID capability MSR value.
1428	*/
1429	static void hmR3VmxReportEptVpidCapsMsr(uint64_t fCaps)
1430	{
1431	LogRel(("HM: MSR_IA32_VMX_EPT_VPID_CAP = %#RX64\n", fCaps));
1432	HMVMX_REPORT_MSR_CAP(fCaps, "RWX_X_ONLY", MSR_IA32_VMX_EPT_VPID_CAP_RWX_X_ONLY);
1433	HMVMX_REPORT_MSR_CAP(fCaps, "PAGE_WALK_LENGTH_4", MSR_IA32_VMX_EPT_VPID_CAP_PAGE_WALK_LENGTH_4);
1434	HMVMX_REPORT_MSR_CAP(fCaps, "EMT_UC", MSR_IA32_VMX_EPT_VPID_CAP_EMT_UC);
1435	HMVMX_REPORT_MSR_CAP(fCaps, "EMT_WB", MSR_IA32_VMX_EPT_VPID_CAP_EMT_WB);
1436	HMVMX_REPORT_MSR_CAP(fCaps, "PDE_2M", MSR_IA32_VMX_EPT_VPID_CAP_PDE_2M);
1437	HMVMX_REPORT_MSR_CAP(fCaps, "PDPTE_1G", MSR_IA32_VMX_EPT_VPID_CAP_PDPTE_1G);
1438	HMVMX_REPORT_MSR_CAP(fCaps, "INVEPT", MSR_IA32_VMX_EPT_VPID_CAP_INVEPT);
1439	HMVMX_REPORT_MSR_CAP(fCaps, "EPT_ACCESS_DIRTY", MSR_IA32_VMX_EPT_VPID_CAP_EPT_ACCESS_DIRTY);
1440	HMVMX_REPORT_MSR_CAP(fCaps, "INVEPT_SINGLE_CONTEXT", MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_SINGLE_CONTEXT);
1441	HMVMX_REPORT_MSR_CAP(fCaps, "INVEPT_ALL_CONTEXTS", MSR_IA32_VMX_EPT_VPID_CAP_INVEPT_ALL_CONTEXTS);
1442	HMVMX_REPORT_MSR_CAP(fCaps, "INVVPID", MSR_IA32_VMX_EPT_VPID_CAP_INVVPID);
1443	HMVMX_REPORT_MSR_CAP(fCaps, "INVVPID_INDIV_ADDR", MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_INDIV_ADDR);
1444	HMVMX_REPORT_MSR_CAP(fCaps, "INVVPID_SINGLE_CONTEXT", MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT);
1445	HMVMX_REPORT_MSR_CAP(fCaps, "INVVPID_ALL_CONTEXTS", MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_ALL_CONTEXTS);
1446	HMVMX_REPORT_MSR_CAP(fCaps, "INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS", MSR_IA32_VMX_EPT_VPID_CAP_INVVPID_SINGLE_CONTEXT_RETAIN_GLOBALS);
1447	}
1448
1449
1450	/**
1451	* Reports MSR_IA32_VMX_MISC MSR to the log.
1452	*
1453	* @param pVM Pointer to the VM.
1454	* @param fMisc The VMX misc. MSR value.
1455	*/
1456	static void hmR3VmxReportMiscMsr(PVM pVM, uint64_t fMisc)
1457	{
1458	LogRel(("HM: MSR_IA32_VMX_MISC = %#RX64\n", fMisc));
1459	uint8_t const cPreemptTimerShift = RT_BF_GET(fMisc, VMX_BF_MISC_PREEMPT_TIMER_TSC);
1460	if (cPreemptTimerShift == pVM->hm.s.vmx.cPreemptTimerShift)
1461	LogRel(("HM: PREEMPT_TIMER_TSC = %#x\n", cPreemptTimerShift));
1462	else
1463	{
1464	LogRel(("HM: PREEMPT_TIMER_TSC = %#x - erratum detected, using %#x instead\n", cPreemptTimerShift,
1465	pVM->hm.s.vmx.cPreemptTimerShift));
1466	}
1467	LogRel(("HM: EXIT_SAVE_EFER_LMA = %RTbool\n", RT_BF_GET(fMisc, VMX_BF_MISC_EXIT_SAVE_EFER_LMA)));
1468	LogRel(("HM: ACTIVITY_STATES = %#x%s\n", RT_BF_GET(fMisc, VMX_BF_MISC_ACTIVITY_STATES),
1469	hmR3VmxGetActivityStateAllDesc(fMisc)));
1470	LogRel(("HM: INTEL_PT = %RTbool\n", RT_BF_GET(fMisc, VMX_BF_MISC_INTEL_PT)));
1471	LogRel(("HM: SMM_READ_SMBASE_MSR = %RTbool\n", RT_BF_GET(fMisc, VMX_BF_MISC_SMM_READ_SMBASE_MSR)));
1472	LogRel(("HM: CR3_TARGET = %#x\n", RT_BF_GET(fMisc, VMX_BF_MISC_CR3_TARGET)));
1473	LogRel(("HM: MAX_MSR = %#x ( %u )\n", RT_BF_GET(fMisc, VMX_BF_MISC_MAX_MSRS),
1474	VMX_MISC_MAX_MSRS(fMisc)));
1475	LogRel(("HM: VMXOFF_BLOCK_SMI = %RTbool\n", RT_BF_GET(fMisc, VMX_BF_MISC_VMXOFF_BLOCK_SMI)));
1476	LogRel(("HM: VMWRITE_ALL = %RTbool\n", RT_BF_GET(fMisc, VMX_BF_MISC_VMWRITE_ALL)));
1477	LogRel(("HM: ENTRY_INJECT_SOFT_INT = %#x\n", RT_BF_GET(fMisc, VMX_BF_MISC_ENTRY_INJECT_SOFT_INT)));
1478	LogRel(("HM: MSEG_ID = %#x\n", RT_BF_GET(fMisc, VMX_BF_MISC_MSEG_ID)));
1479	}
1480
1481
1482	/**
1483	* Reports MSR_IA32_VMX_VMCS_ENUM MSR to the log.
1484	*
1485	* @param uVmcsEnum The VMX VMCS enum MSR value.
1486	*/
1487	static void hmR3VmxReportVmcsEnumMsr(uint64_t uVmcsEnum)
1488	{
1489	LogRel(("HM: MSR_IA32_VMX_VMCS_ENUM = %#RX64\n", uVmcsEnum));
1490	LogRel(("HM: HIGHEST_IDX = %#x\n", RT_BF_GET(uVmcsEnum, VMX_BF_VMCS_ENUM_HIGHEST_IDX)));
1491	}
1492
1493
1494	/**
1495	* Reports MSR_IA32_VMX_VMFUNC MSR to the log.
1496	*
1497	* @param uVmFunc The VMX VMFUNC MSR value.
1498	*/
1499	static void hmR3VmxReportVmFuncMsr(uint64_t uVmFunc)
1500	{
1501	LogRel(("HM: MSR_IA32_VMX_VMFUNC = %#RX64\n", uVmFunc));
1502	HMVMX_REPORT_ALLOWED_FEAT(uVmFunc, "EPTP_SWITCHING", RT_BF_GET(uVmFunc, VMX_BF_VMFUNC_EPTP_SWITCHING));
1503	}
1504
1505
1506	/**
1507	* Reports VMX CR0, CR4 fixed MSRs.
1508	*
1509	* @param pMsrs Pointer to the VMX MSRs.
1510	*/
1511	static void hmR3VmxReportCrFixedMsrs(PVMXMSRS pMsrs)
1512	{
1513	LogRel(("HM: MSR_IA32_VMX_CR0_FIXED0 = %#RX64\n", pMsrs->u64Cr0Fixed0));
1514	LogRel(("HM: MSR_IA32_VMX_CR0_FIXED1 = %#RX64\n", pMsrs->u64Cr0Fixed1));
1515	LogRel(("HM: MSR_IA32_VMX_CR4_FIXED0 = %#RX64\n", pMsrs->u64Cr4Fixed0));
1516	LogRel(("HM: MSR_IA32_VMX_CR4_FIXED1 = %#RX64\n", pMsrs->u64Cr4Fixed1));
1517	}
1518
1519
1520	/**
1521	* Finish VT-x initialization (after ring-0 init).
1522	*
1523	* @returns VBox status code.
1524	* @param pVM The cross context VM structure.
1525	*/
1526	static int hmR3InitFinalizeR0Intel(PVM pVM)
1527	{
1528	int rc;
1529
1530	Log(("pVM->hm.s.vmx.fSupported = %d\n", pVM->hm.s.vmx.fSupported));
1531	AssertLogRelReturn(pVM->hm.s.vmx.Msrs.u64FeatCtrl != 0, VERR_HM_IPE_4);
1532
1533	LogRel(("HM: Using VT-x implementation 2.0\n"));
1534	LogRel(("HM: Max resume loops = %u\n", pVM->hm.s.cMaxResumeLoops));
1535	LogRel(("HM: Host CR4 = %#RX64\n", pVM->hm.s.vmx.u64HostCr4));
1536	LogRel(("HM: Host EFER = %#RX64\n", pVM->hm.s.vmx.u64HostMsrEfer));
1537	LogRel(("HM: MSR_IA32_SMM_MONITOR_CTL = %#RX64\n", pVM->hm.s.vmx.u64HostSmmMonitorCtl));
1538
1539	hmR3VmxReportFeatCtlMsr(pVM->hm.s.vmx.Msrs.u64FeatCtrl);
1540	hmR3VmxReportBasicMsr(pVM->hm.s.vmx.Msrs.u64Basic);
1541
1542	hmR3VmxReportPinBasedCtlsMsr(&pVM->hm.s.vmx.Msrs.PinCtls);
1543	hmR3VmxReportProcBasedCtlsMsr(&pVM->hm.s.vmx.Msrs.ProcCtls);
1544	if (pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
1545	hmR3VmxReportProcBasedCtls2Msr(&pVM->hm.s.vmx.Msrs.ProcCtls2);
1546
1547	hmR3VmxReportEntryCtlsMsr(&pVM->hm.s.vmx.Msrs.EntryCtls);
1548	hmR3VmxReportExitCtlsMsr(&pVM->hm.s.vmx.Msrs.ExitCtls);
1549
1550	if (RT_BF_GET(pVM->hm.s.vmx.Msrs.u64Basic, VMX_BF_BASIC_TRUE_CTLS))
1551	{
1552	/* We don't extensively dump the true capability MSRs as we don't use them, see @bugref{9180#c5}. */
1553	LogRel(("HM: MSR_IA32_VMX_TRUE_PINBASED_CTLS = %#RX64\n", pVM->hm.s.vmx.Msrs.TruePinCtls));
1554	LogRel(("HM: MSR_IA32_VMX_TRUE_PROCBASED_CTLS = %#RX64\n", pVM->hm.s.vmx.Msrs.TrueProcCtls));
1555	LogRel(("HM: MSR_IA32_VMX_TRUE_ENTRY_CTLS = %#RX64\n", pVM->hm.s.vmx.Msrs.TrueEntryCtls));
1556	LogRel(("HM: MSR_IA32_VMX_TRUE_EXIT_CTLS = %#RX64\n", pVM->hm.s.vmx.Msrs.TrueExitCtls));
1557	}
1558
1559	hmR3VmxReportMiscMsr(pVM, pVM->hm.s.vmx.Msrs.u64Misc);
1560	hmR3VmxReportVmcsEnumMsr(pVM->hm.s.vmx.Msrs.u64VmcsEnum);
1561	if (pVM->hm.s.vmx.Msrs.u64EptVpidCaps)
1562	hmR3VmxReportEptVpidCapsMsr(pVM->hm.s.vmx.Msrs.u64EptVpidCaps);
1563	if (pVM->hm.s.vmx.Msrs.u64VmFunc)
1564	hmR3VmxReportVmFuncMsr(pVM->hm.s.vmx.Msrs.u64VmFunc);
1565	hmR3VmxReportCrFixedMsrs(&pVM->hm.s.vmx.Msrs);
1566
1567	LogRel(("HM: APIC-access page physaddr = %#RHp\n", pVM->hm.s.vmx.HCPhysApicAccess));
1568	for (VMCPUID i = 0; i < pVM->cCpus; i++)
1569	{
1570	PCVMXVMCSINFO pVmcsInfo = &pVM->aCpus[i].hm.s.vmx.VmcsInfo;
1571	LogRel(("HM: VCPU%3d: MSR bitmap physaddr = %#RHp\n", i, pVmcsInfo->HCPhysMsrBitmap));
1572	LogRel(("HM: VCPU%3d: VMCS physaddr = %#RHp\n", i, pVmcsInfo->HCPhysVmcs));
1573	}
1574	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
1575	if (pVM->cpum.ro.GuestFeatures.fVmx)
1576	{
1577	LogRel(("HM: Nested-guest:\n"));
1578	for (VMCPUID i = 0; i < pVM->cCpus; i++)
1579	{
1580	PCVMXVMCSINFO pVmcsInfoNstGst = &pVM->aCpus[i].hm.s.vmx.VmcsInfoNstGst;
1581	LogRel(("HM: VCPU%3d: MSR bitmap physaddr = %#RHp\n", i, pVmcsInfoNstGst->HCPhysMsrBitmap));
1582	LogRel(("HM: VCPU%3d: VMCS physaddr = %#RHp\n", i, pVmcsInfoNstGst->HCPhysVmcs));
1583	}
1584	}
1585	#endif
1586
1587	/*
1588	* EPT and unrestricted guest execution are determined in HMR3Init, verify the sanity of that.
1589	*/
1590	AssertLogRelReturn( !pVM->hm.s.fNestedPaging
1591	\|\| (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_EPT),
1592	VERR_HM_IPE_1);
1593	AssertLogRelReturn( !pVM->hm.s.vmx.fUnrestrictedGuest
1594	\|\| ( (pVM->hm.s.vmx.Msrs.ProcCtls2.n.allowed1 & VMX_PROC_CTLS2_UNRESTRICTED_GUEST)
1595	&& pVM->hm.s.fNestedPaging),
1596	VERR_HM_IPE_1);
1597
1598	/*
1599	* Disallow RDTSCP in the guest if there is no secondary process-based VM execution controls as otherwise
1600	* RDTSCP would cause a #UD. There might be no CPUs out there where this happens, as RDTSCP was introduced
1601	* in Nehalems and secondary VM exec. controls should be supported in all of them, but nonetheless it's Intel...
1602	*/
1603	if ( !(pVM->hm.s.vmx.Msrs.ProcCtls.n.allowed1 & VMX_PROC_CTLS_USE_SECONDARY_CTLS)
1604	&& CPUMR3GetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_RDTSCP))
1605	{
1606	CPUMR3ClearGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_RDTSCP);
1607	LogRel(("HM: Disabled RDTSCP\n"));
1608	}
1609
1610	if (!pVM->hm.s.vmx.fUnrestrictedGuest)
1611	{
1612	/* Allocate three pages for the TSS we need for real mode emulation. (2 pages for the IO bitmap) */
1613	rc = PDMR3VmmDevHeapAlloc(pVM, HM_VTX_TOTAL_DEVHEAP_MEM, hmR3VmmDevHeapNotify, (RTR3PTR *)&pVM->hm.s.vmx.pRealModeTSS);
1614	if (RT_SUCCESS(rc))
1615	{
1616	/* The IO bitmap starts right after the virtual interrupt redirection bitmap.
1617	Refer Intel spec. 20.3.3 "Software Interrupt Handling in Virtual-8086 mode"
1618	esp. Figure 20-5.*/
1619	ASMMemZero32(pVM->hm.s.vmx.pRealModeTSS, sizeof(*pVM->hm.s.vmx.pRealModeTSS));
1620	pVM->hm.s.vmx.pRealModeTSS->offIoBitmap = sizeof(*pVM->hm.s.vmx.pRealModeTSS);
1621
1622	/* Bit set to 0 means software interrupts are redirected to the
1623	8086 program interrupt handler rather than switching to
1624	protected-mode handler. */
1625	memset(pVM->hm.s.vmx.pRealModeTSS->IntRedirBitmap, 0, sizeof(pVM->hm.s.vmx.pRealModeTSS->IntRedirBitmap));
1626
1627	/* Allow all port IO, so that port IO instructions do not cause
1628	exceptions and would instead cause a VM-exit (based on VT-x's
1629	IO bitmap which we currently configure to always cause an exit). */
1630	memset(pVM->hm.s.vmx.pRealModeTSS + 1, 0, PAGE_SIZE * 2);
1631	((unsigned char )pVM->hm.s.vmx.pRealModeTSS + HM_VTX_TSS_SIZE - 2) = 0xff;
1632
1633	/*
1634	* Construct a 1024 element page directory with 4 MB pages for the identity mapped
1635	* page table used in real and protected mode without paging with EPT.
1636	*/
1637	pVM->hm.s.vmx.pNonPagingModeEPTPageTable = (PX86PD)((char )pVM->hm.s.vmx.pRealModeTSS + PAGE_SIZE 3);
1638	for (uint32_t i = 0; i < X86_PG_ENTRIES; i++)
1639	{
1640	pVM->hm.s.vmx.pNonPagingModeEPTPageTable->a[i].u = _4M * i;
1641	pVM->hm.s.vmx.pNonPagingModeEPTPageTable->a[i].u \|= X86_PDE4M_P \| X86_PDE4M_RW \| X86_PDE4M_US
1642	\| X86_PDE4M_A \| X86_PDE4M_D \| X86_PDE4M_PS
1643	\| X86_PDE4M_G;
1644	}
1645
1646	/* We convert it here every time as PCI regions could be reconfigured. */
1647	if (PDMVmmDevHeapIsEnabled(pVM))
1648	{
1649	RTGCPHYS GCPhys;
1650	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pRealModeTSS, &GCPhys);
1651	AssertRCReturn(rc, rc);
1652	LogRel(("HM: Real Mode TSS guest physaddr = %#RGp\n", GCPhys));
1653
1654	rc = PDMVmmDevHeapR3ToGCPhys(pVM, pVM->hm.s.vmx.pNonPagingModeEPTPageTable, &GCPhys);
1655	AssertRCReturn(rc, rc);
1656	LogRel(("HM: Non-Paging Mode EPT CR3 = %#RGp\n", GCPhys));
1657	}
1658	}
1659	else
1660	{
1661	LogRel(("HM: No real mode VT-x support (PDMR3VMMDevHeapAlloc returned %Rrc)\n", rc));
1662	pVM->hm.s.vmx.pRealModeTSS = NULL;
1663	pVM->hm.s.vmx.pNonPagingModeEPTPageTable = NULL;
1664	return VMSetError(pVM, rc, RT_SRC_POS,
1665	"HM failure: No real mode VT-x support (PDMR3VMMDevHeapAlloc returned %Rrc)", rc);
1666	}
1667	}
1668
1669	LogRel((pVM->hm.s.fAllow64BitGuests
1670	? "HM: Guest support: 32-bit and 64-bit\n"
1671	: "HM: Guest support: 32-bit only\n"));
1672
1673	/*
1674	* Call ring-0 to set up the VM.
1675	*/
1676	rc = SUPR3CallVMMR0Ex(pVM->pVMR0, 0 /* idCpu /, VMMR0_DO_HM_SETUP_VM, 0 / u64Arg /, NULL / pReqHdr */);
1677	if (rc != VINF_SUCCESS)
1678	{
1679	LogRel(("HM: VMX setup failed with rc=%Rrc!\n", rc));
1680	for (VMCPUID i = 0; i < pVM->cCpus; i++)
1681	{
1682	PVMCPU pVCpu = &pVM->aCpus[i];
1683	LogRel(("HM: CPU[%u] Last instruction error %#x\n", i, pVCpu->hm.s.vmx.LastError.u32InstrError));
1684	LogRel(("HM: CPU[%u] HM error %#x (%u)\n", i, pVCpu->hm.s.u32HMError, pVCpu->hm.s.u32HMError));
1685	}
1686	HMR3CheckError(pVM, rc);
1687	return VMSetError(pVM, rc, RT_SRC_POS, "VT-x setup failed: %Rrc", rc);
1688	}
1689
1690	LogRel(("HM: Supports VMCS EFER fields = %RTbool\n", pVM->hm.s.vmx.fSupportsVmcsEfer));
1691	LogRel(("HM: Enabled VMX\n"));
1692	pVM->hm.s.vmx.fEnabled = true;
1693
1694	hmR3DisableRawMode(pVM); /** @todo make this go away! */
1695
1696	/*
1697	* Change the CPU features.
1698	*/
1699	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_SEP);
1700	if (pVM->hm.s.fAllow64BitGuests)
1701	{
1702	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_PAE);
1703	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_LONG_MODE);
1704	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_SYSCALL); /* 64 bits only on Intel CPUs */
1705	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_LAHF);
1706	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_NX);
1707	}
1708	/* Turn on NXE if PAE has been enabled and the host has turned on NXE
1709	(we reuse the host EFER in the switcher). */
1710	/** @todo this needs to be fixed properly!! */
1711	else if (CPUMR3GetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_PAE))
1712	{
1713	if (pVM->hm.s.vmx.u64HostMsrEfer & MSR_K6_EFER_NXE)
1714	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_NX);
1715	else
1716	LogRel(("HM: NX not enabled on the host, unavailable to PAE guest\n"));
1717	}
1718
1719	/*
1720	* Log configuration details.
1721	*/
1722	if (pVM->hm.s.fNestedPaging)
1723	{
1724	LogRel(("HM: Enabled nested paging\n"));
1725	if (pVM->hm.s.vmx.enmTlbFlushEpt == VMXTLBFLUSHEPT_SINGLE_CONTEXT)
1726	LogRel(("HM: EPT flush type = Single context\n"));
1727	else if (pVM->hm.s.vmx.enmTlbFlushEpt == VMXTLBFLUSHEPT_ALL_CONTEXTS)
1728	LogRel(("HM: EPT flush type = All contexts\n"));
1729	else if (pVM->hm.s.vmx.enmTlbFlushEpt == VMXTLBFLUSHEPT_NOT_SUPPORTED)
1730	LogRel(("HM: EPT flush type = Not supported\n"));
1731	else
1732	LogRel(("HM: EPT flush type = %#x\n", pVM->hm.s.vmx.enmTlbFlushEpt));
1733
1734	if (pVM->hm.s.vmx.fUnrestrictedGuest)
1735	LogRel(("HM: Enabled unrestricted guest execution\n"));
1736
1737	#if HC_ARCH_BITS == 64
1738	if (pVM->hm.s.fLargePages)
1739	{
1740	/* Use large (2 MB) pages for our EPT PDEs where possible. */
1741	PGMSetLargePageUsage(pVM, true);
1742	LogRel(("HM: Enabled large page support\n"));
1743	}
1744	#endif
1745	}
1746	else
1747	Assert(!pVM->hm.s.vmx.fUnrestrictedGuest);
1748
1749	if (pVM->hm.s.vmx.fVpid)
1750	{
1751	LogRel(("HM: Enabled VPID\n"));
1752	if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_INDIV_ADDR)
1753	LogRel(("HM: VPID flush type = Individual addresses\n"));
1754	else if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_SINGLE_CONTEXT)
1755	LogRel(("HM: VPID flush type = Single context\n"));
1756	else if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_ALL_CONTEXTS)
1757	LogRel(("HM: VPID flush type = All contexts\n"));
1758	else if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_SINGLE_CONTEXT_RETAIN_GLOBALS)
1759	LogRel(("HM: VPID flush type = Single context retain globals\n"));
1760	else
1761	LogRel(("HM: VPID flush type = %#x\n", pVM->hm.s.vmx.enmTlbFlushVpid));
1762	}
1763	else if (pVM->hm.s.vmx.enmTlbFlushVpid == VMXTLBFLUSHVPID_NOT_SUPPORTED)
1764	LogRel(("HM: Ignoring VPID capabilities of CPU\n"));
1765
1766	if (pVM->hm.s.vmx.fUsePreemptTimer)
1767	LogRel(("HM: Enabled VMX-preemption timer (cPreemptTimerShift=%u)\n", pVM->hm.s.vmx.cPreemptTimerShift));
1768	else
1769	LogRel(("HM: Disabled VMX-preemption timer\n"));
1770
1771	if (pVM->hm.s.fVirtApicRegs)
1772	LogRel(("HM: Enabled APIC-register virtualization support\n"));
1773
1774	if (pVM->hm.s.fPostedIntrs)
1775	LogRel(("HM: Enabled posted-interrupt processing support\n"));
1776
1777	if (pVM->hm.s.vmx.fUseVmcsShadowing)
1778	{
1779	bool const fFullVmcsShadow = RT_BOOL(pVM->hm.s.vmx.Msrs.u64Misc & VMX_MISC_VMWRITE_ALL);
1780	LogRel(("HM: Enabled %s VMCS shadowing\n", fFullVmcsShadow ? "full" : "partial"));
1781	}
1782
1783	return VINF_SUCCESS;
1784	}
1785
1786
1787	/**
1788	* Finish AMD-V initialization (after ring-0 init).
1789	*
1790	* @returns VBox status code.
1791	* @param pVM The cross context VM structure.
1792	*/
1793	static int hmR3InitFinalizeR0Amd(PVM pVM)
1794	{
1795	Log(("pVM->hm.s.svm.fSupported = %d\n", pVM->hm.s.svm.fSupported));
1796
1797	LogRel(("HM: Using AMD-V implementation 2.0\n"));
1798
1799	uint32_t u32Family;
1800	uint32_t u32Model;
1801	uint32_t u32Stepping;
1802	if (HMIsSubjectToSvmErratum170(&u32Family, &u32Model, &u32Stepping))
1803	LogRel(("HM: AMD Cpu with erratum 170 family %#x model %#x stepping %#x\n", u32Family, u32Model, u32Stepping));
1804	LogRel(("HM: Max resume loops = %u\n", pVM->hm.s.cMaxResumeLoops));
1805	LogRel(("HM: AMD HWCR MSR = %#RX64\n", pVM->hm.s.svm.u64MsrHwcr));
1806	LogRel(("HM: AMD-V revision = %#x\n", pVM->hm.s.svm.u32Rev));
1807	LogRel(("HM: AMD-V max ASID = %RU32\n", pVM->hm.s.uMaxAsid));
1808	LogRel(("HM: AMD-V features = %#x\n", pVM->hm.s.svm.u32Features));
1809
1810	/*
1811	* Enumerate AMD-V features.
1812	*/
1813	static const struct { uint32_t fFlag; const char *pszName; } s_aSvmFeatures[] =
1814	{
1815	#define HMSVM_REPORT_FEATURE(a_StrDesc, a_Define) { a_Define, a_StrDesc }
1816	HMSVM_REPORT_FEATURE("NESTED_PAGING", X86_CPUID_SVM_FEATURE_EDX_NESTED_PAGING),
1817	HMSVM_REPORT_FEATURE("LBR_VIRT", X86_CPUID_SVM_FEATURE_EDX_LBR_VIRT),
1818	HMSVM_REPORT_FEATURE("SVM_LOCK", X86_CPUID_SVM_FEATURE_EDX_SVM_LOCK),
1819	HMSVM_REPORT_FEATURE("NRIP_SAVE", X86_CPUID_SVM_FEATURE_EDX_NRIP_SAVE),
1820	HMSVM_REPORT_FEATURE("TSC_RATE_MSR", X86_CPUID_SVM_FEATURE_EDX_TSC_RATE_MSR),
1821	HMSVM_REPORT_FEATURE("VMCB_CLEAN", X86_CPUID_SVM_FEATURE_EDX_VMCB_CLEAN),
1822	HMSVM_REPORT_FEATURE("FLUSH_BY_ASID", X86_CPUID_SVM_FEATURE_EDX_FLUSH_BY_ASID),
1823	HMSVM_REPORT_FEATURE("DECODE_ASSISTS", X86_CPUID_SVM_FEATURE_EDX_DECODE_ASSISTS),
1824	HMSVM_REPORT_FEATURE("PAUSE_FILTER", X86_CPUID_SVM_FEATURE_EDX_PAUSE_FILTER),
1825	HMSVM_REPORT_FEATURE("PAUSE_FILTER_THRESHOLD", X86_CPUID_SVM_FEATURE_EDX_PAUSE_FILTER_THRESHOLD),
1826	HMSVM_REPORT_FEATURE("AVIC", X86_CPUID_SVM_FEATURE_EDX_AVIC),
1827	HMSVM_REPORT_FEATURE("VIRT_VMSAVE_VMLOAD", X86_CPUID_SVM_FEATURE_EDX_VIRT_VMSAVE_VMLOAD),
1828	HMSVM_REPORT_FEATURE("VGIF", X86_CPUID_SVM_FEATURE_EDX_VGIF),
1829	#undef HMSVM_REPORT_FEATURE
1830	};
1831
1832	uint32_t fSvmFeatures = pVM->hm.s.svm.u32Features;
1833	for (unsigned i = 0; i < RT_ELEMENTS(s_aSvmFeatures); i++)
1834	if (fSvmFeatures & s_aSvmFeatures[i].fFlag)
1835	{
1836	LogRel(("HM: %s\n", s_aSvmFeatures[i].pszName));
1837	fSvmFeatures &= ~s_aSvmFeatures[i].fFlag;
1838	}
1839	if (fSvmFeatures)
1840	for (unsigned iBit = 0; iBit < 32; iBit++)
1841	if (RT_BIT_32(iBit) & fSvmFeatures)
1842	LogRel(("HM: Reserved bit %u\n", iBit));
1843
1844	/*
1845	* Nested paging is determined in HMR3Init, verify the sanity of that.
1846	*/
1847	AssertLogRelReturn( !pVM->hm.s.fNestedPaging
1848	\|\| (pVM->hm.s.svm.u32Features & X86_CPUID_SVM_FEATURE_EDX_NESTED_PAGING),
1849	VERR_HM_IPE_1);
1850
1851	#if 0
1852	/** @todo Add and query IPRT API for host OS support for posted-interrupt IPI
1853	* here. */
1854	if (RTR0IsPostIpiSupport())
1855	pVM->hm.s.fPostedIntrs = true;
1856	#endif
1857
1858	/*
1859	* Call ring-0 to set up the VM.
1860	*/
1861	int rc = SUPR3CallVMMR0Ex(pVM->pVMR0, 0 /idCpu/, VMMR0_DO_HM_SETUP_VM, 0, NULL);
1862	if (rc != VINF_SUCCESS)
1863	{
1864	AssertMsgFailed(("%Rrc\n", rc));
1865	LogRel(("HM: AMD-V setup failed with rc=%Rrc!\n", rc));
1866	return VMSetError(pVM, rc, RT_SRC_POS, "AMD-V setup failed: %Rrc", rc);
1867	}
1868
1869	LogRel(("HM: Enabled SVM\n"));
1870	pVM->hm.s.svm.fEnabled = true;
1871
1872	if (pVM->hm.s.fNestedPaging)
1873	{
1874	LogRel(("HM: Enabled nested paging\n"));
1875
1876	/*
1877	* Enable large pages (2 MB) if applicable.
1878	*/
1879	#if HC_ARCH_BITS == 64
1880	if (pVM->hm.s.fLargePages)
1881	{
1882	PGMSetLargePageUsage(pVM, true);
1883	LogRel(("HM: Enabled large page support\n"));
1884	}
1885	#endif
1886	}
1887
1888	if (pVM->hm.s.fVirtApicRegs)
1889	LogRel(("HM: Enabled APIC-register virtualization support\n"));
1890
1891	if (pVM->hm.s.fPostedIntrs)
1892	LogRel(("HM: Enabled posted-interrupt processing support\n"));
1893
1894	hmR3DisableRawMode(pVM);
1895
1896	/*
1897	* Change the CPU features.
1898	*/
1899	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_SEP);
1900	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_SYSCALL);
1901	if (pVM->hm.s.fAllow64BitGuests)
1902	{
1903	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_PAE);
1904	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_LONG_MODE);
1905	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_NX);
1906	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_LAHF);
1907	}
1908	/* Turn on NXE if PAE has been enabled. */
1909	else if (CPUMR3GetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_PAE))
1910	CPUMR3SetGuestCpuIdFeature(pVM, CPUMCPUIDFEATURE_NX);
1911
1912	LogRel(("HM: %s TPR patching\n", (pVM->hm.s.fTprPatchingAllowed) ? "Enabled" : "Disabled"));
1913
1914	LogRel((pVM->hm.s.fAllow64BitGuests
1915	? "HM: Guest support: 32-bit and 64-bit\n"
1916	: "HM: Guest support: 32-bit only\n"));
1917
1918	return VINF_SUCCESS;
1919	}
1920
1921
1922	/**
1923	* Applies relocations to data and code managed by this
1924	* component. This function will be called at init and
1925	* whenever the VMM need to relocate it self inside the GC.
1926	*
1927	* @param pVM The cross context VM structure.
1928	*/
1929	VMMR3_INT_DECL(void) HMR3Relocate(PVM pVM)
1930	{
1931	Log(("HMR3Relocate to %RGv\n", MMHyperGetArea(pVM, 0)));
1932
1933	/* Fetch the current paging mode during the relocate callback during state loading. */
1934	if (VMR3GetState(pVM) == VMSTATE_LOADING)
1935	{
1936	for (VMCPUID i = 0; i < pVM->cCpus; i++)
1937	{
1938	PVMCPU pVCpu = &pVM->aCpus[i];
1939	pVCpu->hm.s.enmShadowMode = PGMGetShadowMode(pVCpu);
1940	}
1941	}
1942	#if HC_ARCH_BITS == 32 && defined(VBOX_ENABLE_64_BITS_GUESTS)
1943	if (HMIsEnabled(pVM))
1944	{
1945	switch (PGMGetHostMode(pVM))
1946	{
1947	case PGMMODE_32_BIT:
1948	pVM->hm.s.pfnHost32ToGuest64R0 = VMMR3GetHostToGuestSwitcher(pVM, VMMSWITCHER_32_TO_AMD64);
1949	break;
1950
1951	case PGMMODE_PAE:
1952	case PGMMODE_PAE_NX:
1953	pVM->hm.s.pfnHost32ToGuest64R0 = VMMR3GetHostToGuestSwitcher(pVM, VMMSWITCHER_PAE_TO_AMD64);
1954	break;
1955
1956	default:
1957	AssertFailed();
1958	break;
1959	}
1960	}
1961	#endif
1962	return;
1963	}
1964
1965
1966	/**
1967	* Terminates the HM.
1968	*
1969	* Termination means cleaning up and freeing all resources,
1970	* the VM itself is, at this point, powered off or suspended.
1971	*
1972	* @returns VBox status code.
1973	* @param pVM The cross context VM structure.
1974	*/
1975	VMMR3_INT_DECL(int) HMR3Term(PVM pVM)
1976	{
1977	if (pVM->hm.s.vmx.pRealModeTSS)
1978	{
1979	PDMR3VmmDevHeapFree(pVM, pVM->hm.s.vmx.pRealModeTSS);
1980	pVM->hm.s.vmx.pRealModeTSS = 0;
1981	}
1982	hmR3TermCPU(pVM);
1983	return 0;
1984	}
1985
1986
1987	/**
1988	* Terminates the per-VCPU HM.
1989	*
1990	* @returns VBox status code.
1991	* @param pVM The cross context VM structure.
1992	*/
1993	static int hmR3TermCPU(PVM pVM)
1994	{
1995	for (VMCPUID i = 0; i < pVM->cCpus; i++)
1996	{
1997	PVMCPU pVCpu = &pVM->aCpus[i]; NOREF(pVCpu);
1998
1999	#ifdef VBOX_WITH_STATISTICS
2000	if (pVCpu->hm.s.paStatExitReason)
2001	{
2002	MMHyperFree(pVM, pVCpu->hm.s.paStatExitReason);
2003	pVCpu->hm.s.paStatExitReason = NULL;
2004	pVCpu->hm.s.paStatExitReasonR0 = NIL_RTR0PTR;
2005	}
2006	if (pVCpu->hm.s.paStatInjectedIrqs)
2007	{
2008	MMHyperFree(pVM, pVCpu->hm.s.paStatInjectedIrqs);
2009	pVCpu->hm.s.paStatInjectedIrqs = NULL;
2010	pVCpu->hm.s.paStatInjectedIrqsR0 = NIL_RTR0PTR;
2011	}
2012	# if defined(VBOX_WITH_NESTED_HWVIRT_SVM) \|\| defined(VBOX_WITH_NESTED_HWVIRT_VMX)
2013	if (pVCpu->hm.s.paStatNestedExitReason)
2014	{
2015	MMHyperFree(pVM, pVCpu->hm.s.paStatNestedExitReason);
2016	pVCpu->hm.s.paStatNestedExitReason = NULL;
2017	pVCpu->hm.s.paStatNestedExitReasonR0 = NIL_RTR0PTR;
2018	}
2019	# endif
2020	#endif
2021
2022	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2023	memset(pVCpu->hm.s.vmx.VmcsCache.aMagic, 0, sizeof(pVCpu->hm.s.vmx.VmcsCache.aMagic));
2024	pVCpu->hm.s.vmx.VmcsCache.uMagic = 0;
2025	pVCpu->hm.s.vmx.VmcsCache.uPos = 0xffffffff;
2026	#endif
2027	}
2028	return 0;
2029	}
2030
2031
2032	/**
2033	* Resets a virtual CPU.
2034	*
2035	* Used by HMR3Reset and CPU hot plugging.
2036	*
2037	* @param pVCpu The cross context virtual CPU structure to reset.
2038	*/
2039	VMMR3_INT_DECL(void) HMR3ResetCpu(PVMCPU pVCpu)
2040	{
2041	/* Sync. entire state on VM reset ring-0 re-entry. It's safe to reset
2042	the HM flags here, all other EMTs are in ring-3. See VMR3Reset(). */
2043	pVCpu->hm.s.fCtxChanged \|= HM_CHANGED_HOST_CONTEXT \| HM_CHANGED_ALL_GUEST;
2044
2045	pVCpu->hm.s.fActive = false;
2046	pVCpu->hm.s.Event.fPending = false;
2047	pVCpu->hm.s.vmx.u64GstMsrApicBase = 0;
2048	pVCpu->hm.s.vmx.VmcsInfo.fSwitchedTo64on32 = false;
2049	pVCpu->hm.s.vmx.VmcsInfo.fWasInRealMode = true;
2050	#ifdef VBOX_WITH_NESTED_HWVIRT_VMX
2051	if (pVCpu->CTX_SUFF(pVM)->cpum.ro.GuestFeatures.fVmx)
2052	{
2053	pVCpu->hm.s.vmx.VmcsInfoNstGst.fSwitchedTo64on32 = false;
2054	pVCpu->hm.s.vmx.VmcsInfoNstGst.fWasInRealMode = true;
2055	}
2056	#endif
2057
2058	/* Reset the contents of the read cache. */
2059	PVMXVMCSCACHE pVmcsCache = &pVCpu->hm.s.vmx.VmcsCache;
2060	for (unsigned j = 0; j < pVmcsCache->Read.cValidEntries; j++)
2061	pVmcsCache->Read.aFieldVal[j] = 0;
2062
2063	#ifdef VBOX_WITH_CRASHDUMP_MAGIC
2064	/* Magic marker for searching in crash dumps. */
2065	strcpy((char *)pVmcsCache->aMagic, "VMCSCACHE Magic");
2066	pVmcsCache->uMagic = UINT64_C(0xdeadbeefdeadbeef);
2067	#endif
2068	}
2069
2070
2071	/**
2072	* The VM is being reset.
2073	*
2074	* For the HM component this means that any GDT/LDT/TSS monitors
2075	* needs to be removed.
2076	*
2077	* @param pVM The cross context VM structure.
2078	*/
2079	VMMR3_INT_DECL(void) HMR3Reset(PVM pVM)
2080	{
2081	LogFlow(("HMR3Reset:\n"));
2082
2083	if (HMIsEnabled(pVM))
2084	hmR3DisableRawMode(pVM);
2085
2086	for (VMCPUID i = 0; i < pVM->cCpus; i++)
2087	{
2088	PVMCPU pVCpu = &pVM->aCpus[i];
2089
2090	HMR3ResetCpu(pVCpu);
2091	}
2092
2093	/* Clear all patch information. */
2094	pVM->hm.s.pGuestPatchMem = 0;
2095	pVM->hm.s.pFreeGuestPatchMem = 0;
2096	pVM->hm.s.cbGuestPatchMem = 0;
2097	pVM->hm.s.cPatches = 0;
2098	pVM->hm.s.PatchTree = 0;
2099	pVM->hm.s.fTPRPatchingActive = false;
2100	ASMMemZero32(pVM->hm.s.aPatches, sizeof(pVM->hm.s.aPatches));
2101	}
2102
2103
2104	/**
2105	* Callback to patch a TPR instruction (vmmcall or mov cr8).
2106	*
2107	* @returns VBox strict status code.
2108	* @param pVM The cross context VM structure.
2109	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
2110	* @param pvUser Unused.
2111	*/
2112	static DECLCALLBACK(VBOXSTRICTRC) hmR3RemovePatches(PVM pVM, PVMCPU pVCpu, void *pvUser)
2113	{
2114	VMCPUID idCpu = (VMCPUID)(uintptr_t)pvUser;
2115
2116	/* Only execute the handler on the VCPU the original patch request was issued. */
2117	if (pVCpu->idCpu != idCpu)
2118	return VINF_SUCCESS;
2119
2120	Log(("hmR3RemovePatches\n"));
2121	for (unsigned i = 0; i < pVM->hm.s.cPatches; i++)
2122	{
2123	uint8_t abInstr[15];
2124	PHMTPRPATCH pPatch = &pVM->hm.s.aPatches[i];
2125	RTGCPTR pInstrGC = (RTGCPTR)pPatch->Core.Key;
2126	int rc;
2127
2128	#ifdef LOG_ENABLED
2129	char szOutput[256];
2130	rc = DBGFR3DisasInstrEx(pVM->pUVM, pVCpu->idCpu, CPUMGetGuestCS(pVCpu), pInstrGC, DBGF_DISAS_FLAGS_DEFAULT_MODE,
2131	szOutput, sizeof(szOutput), NULL);
2132	if (RT_SUCCESS(rc))
2133	Log(("Patched instr: %s\n", szOutput));
2134	#endif
2135
2136	/* Check if the instruction is still the same. */
2137	rc = PGMPhysSimpleReadGCPtr(pVCpu, abInstr, pInstrGC, pPatch->cbNewOp);
2138	if (rc != VINF_SUCCESS)
2139	{
2140	Log(("Patched code removed? (rc=%Rrc0\n", rc));
2141	continue; /* swapped out or otherwise removed; skip it. */
2142	}
2143
2144	if (memcmp(abInstr, pPatch->aNewOpcode, pPatch->cbNewOp))
2145	{
2146	Log(("Patched instruction was changed! (rc=%Rrc0\n", rc));
2147	continue; /* skip it. */
2148	}
2149
2150	rc = PGMPhysSimpleWriteGCPtr(pVCpu, pInstrGC, pPatch->aOpcode, pPatch->cbOp);
2151	AssertRC(rc);
2152
2153	#ifdef LOG_ENABLED
2154	rc = DBGFR3DisasInstrEx(pVM->pUVM, pVCpu->idCpu, CPUMGetGuestCS(pVCpu), pInstrGC, DBGF_DISAS_FLAGS_DEFAULT_MODE,
2155	szOutput, sizeof(szOutput), NULL);
2156	if (RT_SUCCESS(rc))
2157	Log(("Original instr: %s\n", szOutput));
2158	#endif
2159	}
2160	pVM->hm.s.cPatches = 0;
2161	pVM->hm.s.PatchTree = 0;
2162	pVM->hm.s.pFreeGuestPatchMem = pVM->hm.s.pGuestPatchMem;
2163	pVM->hm.s.fTPRPatchingActive = false;
2164	return VINF_SUCCESS;
2165	}
2166
2167
2168	/**
2169	* Worker for enabling patching in a VT-x/AMD-V guest.
2170	*
2171	* @returns VBox status code.
2172	* @param pVM The cross context VM structure.
2173	* @param idCpu VCPU to execute hmR3RemovePatches on.
2174	* @param pPatchMem Patch memory range.
2175	* @param cbPatchMem Size of the memory range.
2176	*/
2177	static int hmR3EnablePatching(PVM pVM, VMCPUID idCpu, RTRCPTR pPatchMem, unsigned cbPatchMem)
2178	{
2179	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONE_BY_ONE, hmR3RemovePatches, (void *)(uintptr_t)idCpu);
2180	AssertRC(rc);
2181
2182	pVM->hm.s.pGuestPatchMem = pPatchMem;
2183	pVM->hm.s.pFreeGuestPatchMem = pPatchMem;
2184	pVM->hm.s.cbGuestPatchMem = cbPatchMem;
2185	return VINF_SUCCESS;
2186	}
2187
2188
2189	/**
2190	* Enable patching in a VT-x/AMD-V guest
2191	*
2192	* @returns VBox status code.
2193	* @param pVM The cross context VM structure.
2194	* @param pPatchMem Patch memory range.
2195	* @param cbPatchMem Size of the memory range.
2196	*/
2197	VMMR3_INT_DECL(int) HMR3EnablePatching(PVM pVM, RTGCPTR pPatchMem, unsigned cbPatchMem)
2198	{
2199	VM_ASSERT_EMT(pVM);
2200	Log(("HMR3EnablePatching %RGv size %x\n", pPatchMem, cbPatchMem));
2201	if (pVM->cCpus > 1)
2202	{
2203	/* We own the IOM lock here and could cause a deadlock by waiting for a VCPU that is blocking on the IOM lock. */
2204	int rc = VMR3ReqCallNoWait(pVM, VMCPUID_ANY_QUEUE,
2205	(PFNRT)hmR3EnablePatching, 4, pVM, VMMGetCpuId(pVM), (RTRCPTR)pPatchMem, cbPatchMem);
2206	AssertRC(rc);
2207	return rc;
2208	}
2209	return hmR3EnablePatching(pVM, VMMGetCpuId(pVM), (RTRCPTR)pPatchMem, cbPatchMem);
2210	}
2211
2212
2213	/**
2214	* Disable patching in a VT-x/AMD-V guest.
2215	*
2216	* @returns VBox status code.
2217	* @param pVM The cross context VM structure.
2218	* @param pPatchMem Patch memory range.
2219	* @param cbPatchMem Size of the memory range.
2220	*/
2221	VMMR3_INT_DECL(int) HMR3DisablePatching(PVM pVM, RTGCPTR pPatchMem, unsigned cbPatchMem)
2222	{
2223	Log(("HMR3DisablePatching %RGv size %x\n", pPatchMem, cbPatchMem));
2224	RT_NOREF2(pPatchMem, cbPatchMem);
2225
2226	Assert(pVM->hm.s.pGuestPatchMem == pPatchMem);
2227	Assert(pVM->hm.s.cbGuestPatchMem == cbPatchMem);
2228
2229	/** @todo Potential deadlock when other VCPUs are waiting on the IOM lock (we own it)!! */
2230	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONE_BY_ONE, hmR3RemovePatches,
2231	(void *)(uintptr_t)VMMGetCpuId(pVM));
2232	AssertRC(rc);
2233
2234	pVM->hm.s.pGuestPatchMem = 0;
2235	pVM->hm.s.pFreeGuestPatchMem = 0;
2236	pVM->hm.s.cbGuestPatchMem = 0;
2237	pVM->hm.s.fTPRPatchingActive = false;
2238	return VINF_SUCCESS;
2239	}
2240
2241
2242	/**
2243	* Callback to patch a TPR instruction (vmmcall or mov cr8).
2244	*
2245	* @returns VBox strict status code.
2246	* @param pVM The cross context VM structure.
2247	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
2248	* @param pvUser User specified CPU context.
2249	*
2250	*/
2251	static DECLCALLBACK(VBOXSTRICTRC) hmR3ReplaceTprInstr(PVM pVM, PVMCPU pVCpu, void *pvUser)
2252	{
2253	/*
2254	* Only execute the handler on the VCPU the original patch request was
2255	* issued. (The other CPU(s) might not yet have switched to protected
2256	* mode, nor have the correct memory context.)
2257	*/
2258	VMCPUID idCpu = (VMCPUID)(uintptr_t)pvUser;
2259	if (pVCpu->idCpu != idCpu)
2260	return VINF_SUCCESS;
2261
2262	/*
2263	* We're racing other VCPUs here, so don't try patch the instruction twice
2264	* and make sure there is still room for our patch record.
2265	*/
2266	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2267	PHMTPRPATCH pPatch = (PHMTPRPATCH)RTAvloU32Get(&pVM->hm.s.PatchTree, (AVLOU32KEY)pCtx->eip);
2268	if (pPatch)
2269	{
2270	Log(("hmR3ReplaceTprInstr: already patched %RGv\n", pCtx->rip));
2271	return VINF_SUCCESS;
2272	}
2273	uint32_t const idx = pVM->hm.s.cPatches;
2274	if (idx >= RT_ELEMENTS(pVM->hm.s.aPatches))
2275	{
2276	Log(("hmR3ReplaceTprInstr: no available patch slots (%RGv)\n", pCtx->rip));
2277	return VINF_SUCCESS;
2278	}
2279	pPatch = &pVM->hm.s.aPatches[idx];
2280
2281	Log(("hmR3ReplaceTprInstr: rip=%RGv idxPatch=%u\n", pCtx->rip, idx));
2282
2283	/*
2284	* Disassembler the instruction and get cracking.
2285	*/
2286	DBGFR3_DISAS_INSTR_CUR_LOG(pVCpu, "hmR3ReplaceTprInstr");
2287	PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
2288	uint32_t cbOp;
2289	int rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
2290	AssertRC(rc);
2291	if ( rc == VINF_SUCCESS
2292	&& pDis->pCurInstr->uOpcode == OP_MOV
2293	&& cbOp >= 3)
2294	{
2295	static uint8_t const s_abVMMCall[3] = { 0x0f, 0x01, 0xd9 };
2296
2297	rc = PGMPhysSimpleReadGCPtr(pVCpu, pPatch->aOpcode, pCtx->rip, cbOp);
2298	AssertRC(rc);
2299
2300	pPatch->cbOp = cbOp;
2301
2302	if (pDis->Param1.fUse == DISUSE_DISPLACEMENT32)
2303	{
2304	/* write. */
2305	if (pDis->Param2.fUse == DISUSE_REG_GEN32)
2306	{
2307	pPatch->enmType = HMTPRINSTR_WRITE_REG;
2308	pPatch->uSrcOperand = pDis->Param2.Base.idxGenReg;
2309	Log(("hmR3ReplaceTprInstr: HMTPRINSTR_WRITE_REG %u\n", pDis->Param2.Base.idxGenReg));
2310	}
2311	else
2312	{
2313	Assert(pDis->Param2.fUse == DISUSE_IMMEDIATE32);
2314	pPatch->enmType = HMTPRINSTR_WRITE_IMM;
2315	pPatch->uSrcOperand = pDis->Param2.uValue;
2316	Log(("hmR3ReplaceTprInstr: HMTPRINSTR_WRITE_IMM %#llx\n", pDis->Param2.uValue));
2317	}
2318	rc = PGMPhysSimpleWriteGCPtr(pVCpu, pCtx->rip, s_abVMMCall, sizeof(s_abVMMCall));
2319	AssertRC(rc);
2320
2321	memcpy(pPatch->aNewOpcode, s_abVMMCall, sizeof(s_abVMMCall));
2322	pPatch->cbNewOp = sizeof(s_abVMMCall);
2323	STAM_COUNTER_INC(&pVM->hm.s.StatTprReplaceSuccessVmc);
2324	}
2325	else
2326	{
2327	/*
2328	* TPR Read.
2329	*
2330	* Found:
2331	* mov eax, dword [fffe0080] (5 bytes)
2332	* Check if next instruction is:
2333	* shr eax, 4
2334	*/
2335	Assert(pDis->Param1.fUse == DISUSE_REG_GEN32);
2336
2337	uint8_t const idxMmioReg = pDis->Param1.Base.idxGenReg;
2338	uint8_t const cbOpMmio = cbOp;
2339	uint64_t const uSavedRip = pCtx->rip;
2340
2341	pCtx->rip += cbOp;
2342	rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
2343	DBGFR3_DISAS_INSTR_CUR_LOG(pVCpu, "Following read");
2344	pCtx->rip = uSavedRip;
2345
2346	if ( rc == VINF_SUCCESS
2347	&& pDis->pCurInstr->uOpcode == OP_SHR
2348	&& pDis->Param1.fUse == DISUSE_REG_GEN32
2349	&& pDis->Param1.Base.idxGenReg == idxMmioReg
2350	&& pDis->Param2.fUse == DISUSE_IMMEDIATE8
2351	&& pDis->Param2.uValue == 4
2352	&& cbOpMmio + cbOp < sizeof(pVM->hm.s.aPatches[idx].aOpcode))
2353	{
2354	uint8_t abInstr[15];
2355
2356	/* Replacing the two instructions above with an AMD-V specific lock-prefixed 32-bit MOV CR8 instruction so as to
2357	access CR8 in 32-bit mode and not cause a #VMEXIT. */
2358	rc = PGMPhysSimpleReadGCPtr(pVCpu, &pPatch->aOpcode, pCtx->rip, cbOpMmio + cbOp);
2359	AssertRC(rc);
2360
2361	pPatch->cbOp = cbOpMmio + cbOp;
2362
2363	/* 0xf0, 0x0f, 0x20, 0xc0 = mov eax, cr8 */
2364	abInstr[0] = 0xf0;
2365	abInstr[1] = 0x0f;
2366	abInstr[2] = 0x20;
2367	abInstr[3] = 0xc0 \| pDis->Param1.Base.idxGenReg;
2368	for (unsigned i = 4; i < pPatch->cbOp; i++)
2369	abInstr[i] = 0x90; /* nop */
2370
2371	rc = PGMPhysSimpleWriteGCPtr(pVCpu, pCtx->rip, abInstr, pPatch->cbOp);
2372	AssertRC(rc);
2373
2374	memcpy(pPatch->aNewOpcode, abInstr, pPatch->cbOp);
2375	pPatch->cbNewOp = pPatch->cbOp;
2376	STAM_COUNTER_INC(&pVM->hm.s.StatTprReplaceSuccessCr8);
2377
2378	Log(("Acceptable read/shr candidate!\n"));
2379	pPatch->enmType = HMTPRINSTR_READ_SHR4;
2380	}
2381	else
2382	{
2383	pPatch->enmType = HMTPRINSTR_READ;
2384	pPatch->uDstOperand = idxMmioReg;
2385
2386	rc = PGMPhysSimpleWriteGCPtr(pVCpu, pCtx->rip, s_abVMMCall, sizeof(s_abVMMCall));
2387	AssertRC(rc);
2388
2389	memcpy(pPatch->aNewOpcode, s_abVMMCall, sizeof(s_abVMMCall));
2390	pPatch->cbNewOp = sizeof(s_abVMMCall);
2391	STAM_COUNTER_INC(&pVM->hm.s.StatTprReplaceSuccessVmc);
2392	Log(("hmR3ReplaceTprInstr: HMTPRINSTR_READ %u\n", pPatch->uDstOperand));
2393	}
2394	}
2395
2396	pPatch->Core.Key = pCtx->eip;
2397	rc = RTAvloU32Insert(&pVM->hm.s.PatchTree, &pPatch->Core);
2398	AssertRC(rc);
2399
2400	pVM->hm.s.cPatches++;
2401	return VINF_SUCCESS;
2402	}
2403
2404	/*
2405	* Save invalid patch, so we will not try again.
2406	*/
2407	Log(("hmR3ReplaceTprInstr: Failed to patch instr!\n"));
2408	pPatch->Core.Key = pCtx->eip;
2409	pPatch->enmType = HMTPRINSTR_INVALID;
2410	rc = RTAvloU32Insert(&pVM->hm.s.PatchTree, &pPatch->Core);
2411	AssertRC(rc);
2412	pVM->hm.s.cPatches++;
2413	STAM_COUNTER_INC(&pVM->hm.s.StatTprReplaceFailure);
2414	return VINF_SUCCESS;
2415	}
2416
2417
2418	/**
2419	* Callback to patch a TPR instruction (jump to generated code).
2420	*
2421	* @returns VBox strict status code.
2422	* @param pVM The cross context VM structure.
2423	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
2424	* @param pvUser User specified CPU context.
2425	*
2426	*/
2427	static DECLCALLBACK(VBOXSTRICTRC) hmR3PatchTprInstr(PVM pVM, PVMCPU pVCpu, void *pvUser)
2428	{
2429	/*
2430	* Only execute the handler on the VCPU the original patch request was
2431	* issued. (The other CPU(s) might not yet have switched to protected
2432	* mode, nor have the correct memory context.)
2433	*/
2434	VMCPUID idCpu = (VMCPUID)(uintptr_t)pvUser;
2435	if (pVCpu->idCpu != idCpu)
2436	return VINF_SUCCESS;
2437
2438	/*
2439	* We're racing other VCPUs here, so don't try patch the instruction twice
2440	* and make sure there is still room for our patch record.
2441	*/
2442	PCPUMCTX pCtx = &pVCpu->cpum.GstCtx;
2443	PHMTPRPATCH pPatch = (PHMTPRPATCH)RTAvloU32Get(&pVM->hm.s.PatchTree, (AVLOU32KEY)pCtx->eip);
2444	if (pPatch)
2445	{
2446	Log(("hmR3PatchTprInstr: already patched %RGv\n", pCtx->rip));
2447	return VINF_SUCCESS;
2448	}
2449	uint32_t const idx = pVM->hm.s.cPatches;
2450	if (idx >= RT_ELEMENTS(pVM->hm.s.aPatches))
2451	{
2452	Log(("hmR3PatchTprInstr: no available patch slots (%RGv)\n", pCtx->rip));
2453	return VINF_SUCCESS;
2454	}
2455	pPatch = &pVM->hm.s.aPatches[idx];
2456
2457	Log(("hmR3PatchTprInstr: rip=%RGv idxPatch=%u\n", pCtx->rip, idx));
2458	DBGFR3_DISAS_INSTR_CUR_LOG(pVCpu, "hmR3PatchTprInstr");
2459
2460	/*
2461	* Disassemble the instruction and get cracking.
2462	*/
2463	PDISCPUSTATE pDis = &pVCpu->hm.s.DisState;
2464	uint32_t cbOp;
2465	int rc = EMInterpretDisasCurrent(pVM, pVCpu, pDis, &cbOp);
2466	AssertRC(rc);
2467	if ( rc == VINF_SUCCESS
2468	&& pDis->pCurInstr->uOpcode == OP_MOV
2469	&& cbOp >= 5)
2470	{
2471	uint8_t aPatch[64];
2472	uint32_t off = 0;
2473
2474	rc = PGMPhysSimpleReadGCPtr(pVCpu, pPatch->aOpcode, pCtx->rip, cbOp);
2475	AssertRC(rc);
2476
2477	pPatch->cbOp = cbOp;
2478	pPatch->enmType = HMTPRINSTR_JUMP_REPLACEMENT;
2479
2480	if (pDis->Param1.fUse == DISUSE_DISPLACEMENT32)
2481	{
2482	/*
2483	* TPR write:
2484	*
2485	* push ECX [51]
2486	* push EDX [52]
2487	* push EAX [50]
2488	* xor EDX,EDX [31 D2]
2489	* mov EAX,EAX [89 C0]
2490	* or
2491	* mov EAX,0000000CCh [B8 CC 00 00 00]
2492	* mov ECX,0C0000082h [B9 82 00 00 C0]
2493	* wrmsr [0F 30]
2494	* pop EAX [58]
2495	* pop EDX [5A]
2496	* pop ECX [59]
2497	* jmp return_address [E9 return_address]
2498	*/
2499	bool fUsesEax = (pDis->Param2.fUse == DISUSE_REG_GEN32 && pDis->Param2.Base.idxGenReg == DISGREG_EAX);
2500
2501	aPatch[off++] = 0x51; /* push ecx */
2502	aPatch[off++] = 0x52; /* push edx */
2503	if (!fUsesEax)
2504	aPatch[off++] = 0x50; /* push eax */
2505	aPatch[off++] = 0x31; /* xor edx, edx */
2506	aPatch[off++] = 0xd2;
2507	if (pDis->Param2.fUse == DISUSE_REG_GEN32)
2508	{
2509	if (!fUsesEax)
2510	{
2511	aPatch[off++] = 0x89; /* mov eax, src_reg */
2512	aPatch[off++] = MAKE_MODRM(3, pDis->Param2.Base.idxGenReg, DISGREG_EAX);
2513	}
2514	}
2515	else
2516	{
2517	Assert(pDis->Param2.fUse == DISUSE_IMMEDIATE32);
2518	aPatch[off++] = 0xb8; /* mov eax, immediate */
2519	(uint32_t )&aPatch[off] = pDis->Param2.uValue;
2520	off += sizeof(uint32_t);
2521	}
2522	aPatch[off++] = 0xb9; /* mov ecx, 0xc0000082 */
2523	(uint32_t )&aPatch[off] = MSR_K8_LSTAR;
2524	off += sizeof(uint32_t);
2525
2526	aPatch[off++] = 0x0f; /* wrmsr */
2527	aPatch[off++] = 0x30;
2528	if (!fUsesEax)
2529	aPatch[off++] = 0x58; /* pop eax */
2530	aPatch[off++] = 0x5a; /* pop edx */
2531	aPatch[off++] = 0x59; /* pop ecx */
2532	}
2533	else
2534	{
2535	/*
2536	* TPR read:
2537	*
2538	* push ECX [51]
2539	* push EDX [52]
2540	* push EAX [50]
2541	* mov ECX,0C0000082h [B9 82 00 00 C0]
2542	* rdmsr [0F 32]
2543	* mov EAX,EAX [89 C0]
2544	* pop EAX [58]
2545	* pop EDX [5A]
2546	* pop ECX [59]
2547	* jmp return_address [E9 return_address]
2548	*/
2549	Assert(pDis->Param1.fUse == DISUSE_REG_GEN32);
2550
2551	if (pDis->Param1.Base.idxGenReg != DISGREG_ECX)
2552	aPatch[off++] = 0x51; /* push ecx */
2553	if (pDis->Param1.Base.idxGenReg != DISGREG_EDX )
2554	aPatch[off++] = 0x52; /* push edx */
2555	if (pDis->Param1.Base.idxGenReg != DISGREG_EAX)
2556	aPatch[off++] = 0x50; /* push eax */
2557
2558	aPatch[off++] = 0x31; /* xor edx, edx */
2559	aPatch[off++] = 0xd2;
2560
2561	aPatch[off++] = 0xb9; /* mov ecx, 0xc0000082 */
2562	(uint32_t )&aPatch[off] = MSR_K8_LSTAR;
2563	off += sizeof(uint32_t);
2564
2565	aPatch[off++] = 0x0f; /* rdmsr */
2566	aPatch[off++] = 0x32;
2567
2568	if (pDis->Param1.Base.idxGenReg != DISGREG_EAX)
2569	{
2570	aPatch[off++] = 0x89; /* mov dst_reg, eax */
2571	aPatch[off++] = MAKE_MODRM(3, DISGREG_EAX, pDis->Param1.Base.idxGenReg);
2572	}
2573
2574	if (pDis->Param1.Base.idxGenReg != DISGREG_EAX)
2575	aPatch[off++] = 0x58; /* pop eax */
2576	if (pDis->Param1.Base.idxGenReg != DISGREG_EDX )
2577	aPatch[off++] = 0x5a; /* pop edx */
2578	if (pDis->Param1.Base.idxGenReg != DISGREG_ECX)
2579	aPatch[off++] = 0x59; /* pop ecx */
2580	}
2581	aPatch[off++] = 0xe9; /* jmp return_address */
2582	(RTRCUINTPTR )&aPatch[off] = ((RTRCUINTPTR)pCtx->eip + cbOp) - ((RTRCUINTPTR)pVM->hm.s.pFreeGuestPatchMem + off + 4);
2583	off += sizeof(RTRCUINTPTR);
2584
2585	if (pVM->hm.s.pFreeGuestPatchMem + off <= pVM->hm.s.pGuestPatchMem + pVM->hm.s.cbGuestPatchMem)
2586	{
2587	/* Write new code to the patch buffer. */
2588	rc = PGMPhysSimpleWriteGCPtr(pVCpu, pVM->hm.s.pFreeGuestPatchMem, aPatch, off);
2589	AssertRC(rc);
2590
2591	#ifdef LOG_ENABLED
2592	uint32_t cbCurInstr;
2593	for (RTGCPTR GCPtrInstr = pVM->hm.s.pFreeGuestPatchMem;
2594	GCPtrInstr < pVM->hm.s.pFreeGuestPatchMem + off;
2595	GCPtrInstr += RT_MAX(cbCurInstr, 1))
2596	{
2597	char szOutput[256];
2598	rc = DBGFR3DisasInstrEx(pVM->pUVM, pVCpu->idCpu, pCtx->cs.Sel, GCPtrInstr, DBGF_DISAS_FLAGS_DEFAULT_MODE,
2599	szOutput, sizeof(szOutput), &cbCurInstr);
2600	if (RT_SUCCESS(rc))
2601	Log(("Patch instr %s\n", szOutput));
2602	else
2603	Log(("%RGv: rc=%Rrc\n", GCPtrInstr, rc));
2604	}
2605	#endif
2606
2607	pPatch->aNewOpcode[0] = 0xE9;
2608	(RTRCUINTPTR )&pPatch->aNewOpcode[1] = ((RTRCUINTPTR)pVM->hm.s.pFreeGuestPatchMem) - ((RTRCUINTPTR)pCtx->eip + 5);
2609
2610	/* Overwrite the TPR instruction with a jump. */
2611	rc = PGMPhysSimpleWriteGCPtr(pVCpu, pCtx->eip, pPatch->aNewOpcode, 5);
2612	AssertRC(rc);
2613
2614	DBGFR3_DISAS_INSTR_CUR_LOG(pVCpu, "Jump");
2615
2616	pVM->hm.s.pFreeGuestPatchMem += off;
2617	pPatch->cbNewOp = 5;
2618
2619	pPatch->Core.Key = pCtx->eip;
2620	rc = RTAvloU32Insert(&pVM->hm.s.PatchTree, &pPatch->Core);
2621	AssertRC(rc);
2622
2623	pVM->hm.s.cPatches++;
2624	pVM->hm.s.fTPRPatchingActive = true;
2625	STAM_COUNTER_INC(&pVM->hm.s.StatTprPatchSuccess);
2626	return VINF_SUCCESS;
2627	}
2628
2629	Log(("Ran out of space in our patch buffer!\n"));
2630	}
2631	else
2632	Log(("hmR3PatchTprInstr: Failed to patch instr!\n"));
2633
2634
2635	/*
2636	* Save invalid patch, so we will not try again.
2637	*/
2638	pPatch = &pVM->hm.s.aPatches[idx];
2639	pPatch->Core.Key = pCtx->eip;
2640	pPatch->enmType = HMTPRINSTR_INVALID;
2641	rc = RTAvloU32Insert(&pVM->hm.s.PatchTree, &pPatch->Core);
2642	AssertRC(rc);
2643	pVM->hm.s.cPatches++;
2644	STAM_COUNTER_INC(&pVM->hm.s.StatTprPatchFailure);
2645	return VINF_SUCCESS;
2646	}
2647
2648
2649	/**
2650	* Attempt to patch TPR mmio instructions.
2651	*
2652	* @returns VBox status code.
2653	* @param pVM The cross context VM structure.
2654	* @param pVCpu The cross context virtual CPU structure.
2655	*/
2656	VMMR3_INT_DECL(int) HMR3PatchTprInstr(PVM pVM, PVMCPU pVCpu)
2657	{
2658	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONE_BY_ONE,
2659	pVM->hm.s.pGuestPatchMem ? hmR3PatchTprInstr : hmR3ReplaceTprInstr,
2660	(void *)(uintptr_t)pVCpu->idCpu);
2661	AssertRC(rc);
2662	return rc;
2663	}
2664
2665
2666	/**
2667	* Checks if we need to reschedule due to VMM device heap changes.
2668	*
2669	* @returns true if a reschedule is required, otherwise false.
2670	* @param pVM The cross context VM structure.
2671	* @param pCtx VM execution context.
2672	*/
2673	VMMR3_INT_DECL(bool) HMR3IsRescheduleRequired(PVM pVM, PCCPUMCTX pCtx)
2674	{
2675	/*
2676	* The VMM device heap is a requirement for emulating real-mode or protected-mode without paging
2677	* when the unrestricted guest execution feature is missing (VT-x only).
2678	*/
2679	if ( pVM->hm.s.vmx.fEnabled
2680	&& !pVM->hm.s.vmx.fUnrestrictedGuest
2681	&& CPUMIsGuestInRealModeEx(pCtx)
2682	&& !PDMVmmDevHeapIsEnabled(pVM))
2683	return true;
2684
2685	return false;
2686	}
2687
2688
2689	/**
2690	* Noticiation callback from DBGF when interrupt breakpoints or generic debug
2691	* event settings changes.
2692	*
2693	* DBGF will call HMR3NotifyDebugEventChangedPerCpu on each CPU afterwards, this
2694	* function is just updating the VM globals.
2695	*
2696	* @param pVM The VM cross context VM structure.
2697	* @thread EMT(0)
2698	*/
2699	VMMR3_INT_DECL(void) HMR3NotifyDebugEventChanged(PVM pVM)
2700	{
2701	/* Interrupts. */
2702	bool fUseDebugLoop = pVM->dbgf.ro.cSoftIntBreakpoints > 0
2703	\|\| pVM->dbgf.ro.cHardIntBreakpoints > 0;
2704
2705	/* CPU Exceptions. */
2706	for (DBGFEVENTTYPE enmEvent = DBGFEVENT_XCPT_FIRST;
2707	!fUseDebugLoop && enmEvent <= DBGFEVENT_XCPT_LAST;
2708	enmEvent = (DBGFEVENTTYPE)(enmEvent + 1))
2709	fUseDebugLoop = DBGF_IS_EVENT_ENABLED(pVM, enmEvent);
2710
2711	/* Common VM exits. */
2712	for (DBGFEVENTTYPE enmEvent = DBGFEVENT_EXIT_FIRST;
2713	!fUseDebugLoop && enmEvent <= DBGFEVENT_EXIT_LAST_COMMON;
2714	enmEvent = (DBGFEVENTTYPE)(enmEvent + 1))
2715	fUseDebugLoop = DBGF_IS_EVENT_ENABLED(pVM, enmEvent);
2716
2717	/* Vendor specific VM exits. */
2718	if (HMR3IsVmxEnabled(pVM->pUVM))
2719	for (DBGFEVENTTYPE enmEvent = DBGFEVENT_EXIT_VMX_FIRST;
2720	!fUseDebugLoop && enmEvent <= DBGFEVENT_EXIT_VMX_LAST;
2721	enmEvent = (DBGFEVENTTYPE)(enmEvent + 1))
2722	fUseDebugLoop = DBGF_IS_EVENT_ENABLED(pVM, enmEvent);
2723	else
2724	for (DBGFEVENTTYPE enmEvent = DBGFEVENT_EXIT_SVM_FIRST;
2725	!fUseDebugLoop && enmEvent <= DBGFEVENT_EXIT_SVM_LAST;
2726	enmEvent = (DBGFEVENTTYPE)(enmEvent + 1))
2727	fUseDebugLoop = DBGF_IS_EVENT_ENABLED(pVM, enmEvent);
2728
2729	/* Done. */
2730	pVM->hm.s.fUseDebugLoop = fUseDebugLoop;
2731	}
2732
2733
2734	/**
2735	* Follow up notification callback to HMR3NotifyDebugEventChanged for each CPU.
2736	*
2737	* HM uses this to combine the decision made by HMR3NotifyDebugEventChanged with
2738	* per CPU settings.
2739	*
2740	* @param pVM The VM cross context VM structure.
2741	* @param pVCpu The cross context virtual CPU structure of the calling EMT.
2742	*/
2743	VMMR3_INT_DECL(void) HMR3NotifyDebugEventChangedPerCpu(PVM pVM, PVMCPU pVCpu)
2744	{
2745	pVCpu->hm.s.fUseDebugLoop = pVCpu->hm.s.fSingleInstruction \| pVM->hm.s.fUseDebugLoop;
2746	}
2747
2748
2749	/**
2750	* Checks if we are currently using hardware acceleration.
2751	*
2752	* @returns true if hardware acceleration is being used, otherwise false.
2753	* @param pVCpu The cross context virtual CPU structure.
2754	*/
2755	VMMR3_INT_DECL(bool) HMR3IsActive(PCVMCPU pVCpu)
2756	{
2757	return pVCpu->hm.s.fActive;
2758	}
2759
2760
2761	/**
2762	* External interface for querying whether hardware acceleration is enabled.
2763	*
2764	* @returns true if VT-x or AMD-V is being used, otherwise false.
2765	* @param pUVM The user mode VM handle.
2766	* @sa HMIsEnabled, HMIsEnabledNotMacro.
2767	*/
2768	VMMR3DECL(bool) HMR3IsEnabled(PUVM pUVM)
2769	{
2770	UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
2771	PVM pVM = pUVM->pVM;
2772	VM_ASSERT_VALID_EXT_RETURN(pVM, false);
2773	return pVM->fHMEnabled; /* Don't use the macro as the GUI may query us very very early. */
2774	}
2775
2776
2777	/**
2778	* External interface for querying whether VT-x is being used.
2779	*
2780	* @returns true if VT-x is being used, otherwise false.
2781	* @param pUVM The user mode VM handle.
2782	* @sa HMR3IsSvmEnabled, HMIsEnabled
2783	*/
2784	VMMR3DECL(bool) HMR3IsVmxEnabled(PUVM pUVM)
2785	{
2786	UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
2787	PVM pVM = pUVM->pVM;
2788	VM_ASSERT_VALID_EXT_RETURN(pVM, false);
2789	return pVM->hm.s.vmx.fEnabled
2790	&& pVM->hm.s.vmx.fSupported
2791	&& pVM->fHMEnabled;
2792	}
2793
2794
2795	/**
2796	* External interface for querying whether AMD-V is being used.
2797	*
2798	* @returns true if VT-x is being used, otherwise false.
2799	* @param pUVM The user mode VM handle.
2800	* @sa HMR3IsVmxEnabled, HMIsEnabled
2801	*/
2802	VMMR3DECL(bool) HMR3IsSvmEnabled(PUVM pUVM)
2803	{
2804	UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
2805	PVM pVM = pUVM->pVM;
2806	VM_ASSERT_VALID_EXT_RETURN(pVM, false);
2807	return pVM->hm.s.svm.fEnabled
2808	&& pVM->hm.s.svm.fSupported
2809	&& pVM->fHMEnabled;
2810	}
2811
2812
2813	/**
2814	* Checks if we are currently using nested paging.
2815	*
2816	* @returns true if nested paging is being used, otherwise false.
2817	* @param pUVM The user mode VM handle.
2818	*/
2819	VMMR3DECL(bool) HMR3IsNestedPagingActive(PUVM pUVM)
2820	{
2821	UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
2822	PVM pVM = pUVM->pVM;
2823	VM_ASSERT_VALID_EXT_RETURN(pVM, false);
2824	return pVM->hm.s.fNestedPaging;
2825	}
2826
2827
2828	/**
2829	* Checks if virtualized APIC registers is enabled.
2830	*
2831	* When enabled this feature allows the hardware to access most of the
2832	* APIC registers in the virtual-APIC page without causing VM-exits. See
2833	* Intel spec. 29.1.1 "Virtualized APIC Registers".
2834	*
2835	* @returns true if virtualized APIC registers is enabled, otherwise
2836	* false.
2837	* @param pUVM The user mode VM handle.
2838	*/
2839	VMMR3DECL(bool) HMR3IsVirtApicRegsEnabled(PUVM pUVM)
2840	{
2841	UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
2842	PVM pVM = pUVM->pVM;
2843	VM_ASSERT_VALID_EXT_RETURN(pVM, false);
2844	return pVM->hm.s.fVirtApicRegs;
2845	}
2846
2847
2848	/**
2849	* Checks if APIC posted-interrupt processing is enabled.
2850	*
2851	* This returns whether we can deliver interrupts to the guest without
2852	* leaving guest-context by updating APIC state from host-context.
2853	*
2854	* @returns true if APIC posted-interrupt processing is enabled,
2855	* otherwise false.
2856	* @param pUVM The user mode VM handle.
2857	*/
2858	VMMR3DECL(bool) HMR3IsPostedIntrsEnabled(PUVM pUVM)
2859	{
2860	UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
2861	PVM pVM = pUVM->pVM;
2862	VM_ASSERT_VALID_EXT_RETURN(pVM, false);
2863	return pVM->hm.s.fPostedIntrs;
2864	}
2865
2866
2867	/**
2868	* Checks if we are currently using VPID in VT-x mode.
2869	*
2870	* @returns true if VPID is being used, otherwise false.
2871	* @param pUVM The user mode VM handle.
2872	*/
2873	VMMR3DECL(bool) HMR3IsVpidActive(PUVM pUVM)
2874	{
2875	UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
2876	PVM pVM = pUVM->pVM;
2877	VM_ASSERT_VALID_EXT_RETURN(pVM, false);
2878	return pVM->hm.s.vmx.fVpid;
2879	}
2880
2881
2882	/**
2883	* Checks if we are currently using VT-x unrestricted execution,
2884	* aka UX.
2885	*
2886	* @returns true if UX is being used, otherwise false.
2887	* @param pUVM The user mode VM handle.
2888	*/
2889	VMMR3DECL(bool) HMR3IsUXActive(PUVM pUVM)
2890	{
2891	UVM_ASSERT_VALID_EXT_RETURN(pUVM, false);
2892	PVM pVM = pUVM->pVM;
2893	VM_ASSERT_VALID_EXT_RETURN(pVM, false);
2894	return pVM->hm.s.vmx.fUnrestrictedGuest
2895	\|\| pVM->hm.s.svm.fSupported;
2896	}
2897
2898
2899	/**
2900	* Checks if the VMX-preemption timer is being used.
2901	*
2902	* @returns true if the VMX-preemption timer is being used, otherwise false.
2903	* @param pVM The cross context VM structure.
2904	*/
2905	VMMR3_INT_DECL(bool) HMR3IsVmxPreemptionTimerUsed(PVM pVM)
2906	{
2907	return HMIsEnabled(pVM)
2908	&& pVM->hm.s.vmx.fEnabled
2909	&& pVM->hm.s.vmx.fUsePreemptTimer;
2910	}
2911
2912
2913	/**
2914	* Helper for HMR3CheckError to log VMCS controls to the release log.
2915	*
2916	* @param idCpu The Virtual CPU ID.
2917	* @param pVmcsInfo The VMCS info. object.
2918	*/
2919	static void hmR3CheckErrorLogVmcsCtls(VMCPUID idCpu, PCVMXVMCSINFO pVmcsInfo)
2920	{
2921	LogRel(("HM: CPU[%u] PinCtls %#RX32\n", idCpu, pVmcsInfo->u32PinCtls));
2922	{
2923	uint32_t const u32Val = pVmcsInfo->u32PinCtls;
2924	HMVMX_LOGREL_FEAT(u32Val, VMX_PIN_CTLS_EXT_INT_EXIT );
2925	HMVMX_LOGREL_FEAT(u32Val, VMX_PIN_CTLS_NMI_EXIT );
2926	HMVMX_LOGREL_FEAT(u32Val, VMX_PIN_CTLS_VIRT_NMI );
2927	HMVMX_LOGREL_FEAT(u32Val, VMX_PIN_CTLS_PREEMPT_TIMER);
2928	HMVMX_LOGREL_FEAT(u32Val, VMX_PIN_CTLS_POSTED_INT );
2929	}
2930	LogRel(("HM: CPU[%u] ProcCtls %#RX32\n", idCpu, pVmcsInfo->u32ProcCtls));
2931	{
2932	uint32_t const u32Val = pVmcsInfo->u32ProcCtls;
2933	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_INT_WINDOW_EXIT );
2934	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_USE_TSC_OFFSETTING);
2935	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_HLT_EXIT );
2936	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_INVLPG_EXIT );
2937	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_MWAIT_EXIT );
2938	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_RDPMC_EXIT );
2939	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_RDTSC_EXIT );
2940	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_CR3_LOAD_EXIT );
2941	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_CR3_STORE_EXIT );
2942	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_CR8_LOAD_EXIT );
2943	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_CR8_STORE_EXIT );
2944	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_USE_TPR_SHADOW );
2945	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_NMI_WINDOW_EXIT );
2946	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_MOV_DR_EXIT );
2947	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_UNCOND_IO_EXIT );
2948	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_USE_IO_BITMAPS );
2949	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_MONITOR_TRAP_FLAG );
2950	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_USE_MSR_BITMAPS );
2951	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_MONITOR_EXIT );
2952	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_PAUSE_EXIT );
2953	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS_USE_SECONDARY_CTLS);
2954	}
2955	LogRel(("HM: CPU[%u] ProcCtls2 %#RX32\n", idCpu, pVmcsInfo->u32ProcCtls2));
2956	{
2957	uint32_t const u32Val = pVmcsInfo->u32ProcCtls2;
2958	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_VIRT_APIC_ACCESS );
2959	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_EPT );
2960	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_DESC_TABLE_EXIT );
2961	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_RDTSCP );
2962	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_VIRT_X2APIC_MODE );
2963	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_VPID );
2964	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_WBINVD_EXIT );
2965	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_UNRESTRICTED_GUEST );
2966	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_APIC_REG_VIRT );
2967	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_VIRT_INT_DELIVERY );
2968	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_PAUSE_LOOP_EXIT );
2969	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_RDRAND_EXIT );
2970	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_INVPCID );
2971	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_VMFUNC );
2972	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_VMCS_SHADOWING );
2973	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_ENCLS_EXIT );
2974	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_RDSEED_EXIT );
2975	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_PML );
2976	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_EPT_VE );
2977	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_CONCEAL_VMX_FROM_PT);
2978	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_XSAVES_XRSTORS );
2979	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_MODE_BASED_EPT_PERM);
2980	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_SPPTP_EPT );
2981	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_PT_EPT );
2982	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_TSC_SCALING );
2983	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_USER_WAIT_PAUSE );
2984	HMVMX_LOGREL_FEAT(u32Val, VMX_PROC_CTLS2_ENCLV_EXIT );
2985	}
2986	LogRel(("HM: CPU[%u] EntryCtls %#RX32\n", idCpu, pVmcsInfo->u32EntryCtls));
2987	{
2988	uint32_t const u32Val = pVmcsInfo->u32EntryCtls;
2989	HMVMX_LOGREL_FEAT(u32Val, VMX_ENTRY_CTLS_LOAD_DEBUG );
2990	HMVMX_LOGREL_FEAT(u32Val, VMX_ENTRY_CTLS_IA32E_MODE_GUEST );
2991	HMVMX_LOGREL_FEAT(u32Val, VMX_ENTRY_CTLS_ENTRY_TO_SMM );
2992	HMVMX_LOGREL_FEAT(u32Val, VMX_ENTRY_CTLS_DEACTIVATE_DUAL_MON);
2993	HMVMX_LOGREL_FEAT(u32Val, VMX_ENTRY_CTLS_LOAD_PERF_MSR );
2994	HMVMX_LOGREL_FEAT(u32Val, VMX_ENTRY_CTLS_LOAD_PAT_MSR );
2995	HMVMX_LOGREL_FEAT(u32Val, VMX_ENTRY_CTLS_LOAD_EFER_MSR );
2996	HMVMX_LOGREL_FEAT(u32Val, VMX_ENTRY_CTLS_LOAD_BNDCFGS_MSR );
2997	HMVMX_LOGREL_FEAT(u32Val, VMX_ENTRY_CTLS_CONCEAL_VMX_FROM_PT);
2998	HMVMX_LOGREL_FEAT(u32Val, VMX_ENTRY_CTLS_LOAD_RTIT_CTL_MSR );
2999	}
3000	LogRel(("HM: CPU[%u] ExitCtls %#RX32\n", idCpu, pVmcsInfo->u32ExitCtls));
3001	{
3002	uint32_t const u32Val = pVmcsInfo->u32ExitCtls;
3003	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_SAVE_DEBUG );
3004	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_HOST_ADDR_SPACE_SIZE );
3005	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_LOAD_PERF_MSR );
3006	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_ACK_EXT_INT );
3007	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_SAVE_PAT_MSR );
3008	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_LOAD_PAT_MSR );
3009	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_SAVE_EFER_MSR );
3010	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_LOAD_EFER_MSR );
3011	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_SAVE_PREEMPT_TIMER );
3012	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_CLEAR_BNDCFGS_MSR );
3013	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_CONCEAL_VMX_FROM_PT );
3014	HMVMX_LOGREL_FEAT(u32Val, VMX_EXIT_CTLS_CLEAR_RTIT_CTL_MSR );
3015	}
3016	}
3017
3018
3019	/**
3020	* Check fatal VT-x/AMD-V error and produce some meaningful
3021	* log release message.
3022	*
3023	* @param pVM The cross context VM structure.
3024	* @param iStatusCode VBox status code.
3025	*/
3026	VMMR3_INT_DECL(void) HMR3CheckError(PVM pVM, int iStatusCode)
3027	{
3028	for (VMCPUID i = 0; i < pVM->cCpus; i++)
3029	{
3030	/** @todo r=ramshankar: Are all EMTs out of ring-0 at this point!? If not, we
3031	* might be getting inaccurate values for non-guru'ing EMTs. */
3032	PVMCPU pVCpu = &pVM->aCpus[i];
3033	PCVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
3034	bool const fNstGstVmcsActive = pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs;
3035	switch (iStatusCode)
3036	{
3037	case VERR_VMX_INVALID_VMCS_PTR:
3038	{
3039	LogRel(("HM: VERR_VMX_INVALID_VMCS_PTR:\n"));
3040	LogRel(("HM: CPU[%u] %s VMCS active\n", i, fNstGstVmcsActive ? "Nested-guest" : "Guest"));
3041	LogRel(("HM: CPU[%u] Current pointer %#RHp vs %#RHp\n", i, pVCpu->hm.s.vmx.LastError.HCPhysCurrentVmcs,
3042	pVmcsInfo->HCPhysVmcs));
3043	LogRel(("HM: CPU[%u] Current VMCS version %#x\n", i, pVCpu->hm.s.vmx.LastError.u32VmcsRev));
3044	LogRel(("HM: CPU[%u] Entered Host Cpu %u\n", i, pVCpu->hm.s.vmx.LastError.idEnteredCpu));
3045	LogRel(("HM: CPU[%u] Current Host Cpu %u\n", i, pVCpu->hm.s.vmx.LastError.idCurrentCpu));
3046	break;
3047	}
3048
3049	case VERR_VMX_UNABLE_TO_START_VM:
3050	{
3051	LogRel(("HM: VERR_VMX_UNABLE_TO_START_VM:\n"));
3052	LogRel(("HM: CPU[%u] %s VMCS active\n", i, fNstGstVmcsActive ? "Nested-guest" : "Guest"));
3053	LogRel(("HM: CPU[%u] Instruction error %#x\n", i, pVCpu->hm.s.vmx.LastError.u32InstrError));
3054	LogRel(("HM: CPU[%u] Exit reason %#x\n", i, pVCpu->hm.s.vmx.LastError.u32ExitReason));
3055
3056	if ( pVCpu->hm.s.vmx.LastError.u32InstrError == VMXINSTRERR_VMLAUNCH_NON_CLEAR_VMCS
3057	\|\| pVCpu->hm.s.vmx.LastError.u32InstrError == VMXINSTRERR_VMRESUME_NON_LAUNCHED_VMCS)
3058	{
3059	LogRel(("HM: CPU[%u] Entered Host Cpu %u\n", i, pVCpu->hm.s.vmx.LastError.idEnteredCpu));
3060	LogRel(("HM: CPU[%u] Current Host Cpu %u\n", i, pVCpu->hm.s.vmx.LastError.idCurrentCpu));
3061	}
3062	else if (pVCpu->hm.s.vmx.LastError.u32InstrError == VMXINSTRERR_VMENTRY_INVALID_CTLS)
3063	{
3064	hmR3CheckErrorLogVmcsCtls(i, pVmcsInfo);
3065	LogRel(("HM: CPU[%u] HCPhysMsrBitmap %#RHp\n", i, pVmcsInfo->HCPhysMsrBitmap));
3066	LogRel(("HM: CPU[%u] HCPhysGuestMsrLoad %#RHp\n", i, pVmcsInfo->HCPhysGuestMsrLoad));
3067	LogRel(("HM: CPU[%u] HCPhysGuestMsrStore %#RHp\n", i, pVmcsInfo->HCPhysGuestMsrStore));
3068	LogRel(("HM: CPU[%u] HCPhysHostMsrLoad %#RHp\n", i, pVmcsInfo->HCPhysHostMsrLoad));
3069	LogRel(("HM: CPU[%u] cEntryMsrLoad %u\n", i, pVmcsInfo->cEntryMsrLoad));
3070	LogRel(("HM: CPU[%u] cExitMsrStore %u\n", i, pVmcsInfo->cExitMsrStore));
3071	LogRel(("HM: CPU[%u] cExitMsrLoad %u\n", i, pVmcsInfo->cExitMsrLoad));
3072	}
3073	/** @todo Log VM-entry event injection control fields
3074	* VMX_VMCS_CTRL_ENTRY_IRQ_INFO, VMX_VMCS_CTRL_ENTRY_EXCEPTION_ERRCODE
3075	* and VMX_VMCS_CTRL_ENTRY_INSTR_LENGTH from the VMCS. */
3076	break;
3077	}
3078
3079	case VERR_VMX_INVALID_GUEST_STATE:
3080	{
3081	LogRel(("HM: VERR_VMX_INVALID_GUEST_STATE:\n"));
3082	hmR3CheckErrorLogVmcsCtls(i, pVmcsInfo);
3083	break;
3084	}
3085
3086	/* The guru will dump the HM error and exit history. Nothing extra to report for these errors. */
3087	case VERR_HM_UNSUPPORTED_CPU_FEATURE_COMBO:
3088	case VERR_VMX_INVALID_VMXON_PTR:
3089	case VERR_VMX_UNEXPECTED_EXIT:
3090	case VERR_VMX_INVALID_VMCS_FIELD:
3091	case VERR_SVM_UNKNOWN_EXIT:
3092	case VERR_SVM_UNEXPECTED_EXIT:
3093	case VERR_SVM_UNEXPECTED_PATCH_TYPE:
3094	case VERR_SVM_UNEXPECTED_XCPT_EXIT:
3095	case VERR_VMX_UNEXPECTED_INTERRUPTION_EXIT_TYPE:
3096	break;
3097	}
3098	}
3099
3100	if (iStatusCode == VERR_VMX_UNABLE_TO_START_VM)
3101	{
3102	LogRel(("HM: VERR_VMX_UNABLE_TO_START_VM: VM-entry allowed-1 %#RX32\n", pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed1));
3103	LogRel(("HM: VERR_VMX_UNABLE_TO_START_VM: VM-entry allowed-0 %#RX32\n", pVM->hm.s.vmx.Msrs.EntryCtls.n.allowed0));
3104	}
3105	else if (iStatusCode == VERR_VMX_INVALID_VMXON_PTR)
3106	LogRel(("HM: HCPhysVmxEnableError = %#RHp\n", pVM->hm.s.vmx.HCPhysVmxEnableError));
3107	}
3108
3109
3110	/**
3111	* Execute state save operation.
3112	*
3113	* Save only data that cannot be re-loaded while entering HM ring-0 code. This
3114	* is because we always save the VM state from ring-3 and thus most HM state
3115	* will be re-synced dynamically at runtime and don't need to be part of the VM
3116	* saved state.
3117	*
3118	* @returns VBox status code.
3119	* @param pVM The cross context VM structure.
3120	* @param pSSM SSM operation handle.
3121	*/
3122	static DECLCALLBACK(int) hmR3Save(PVM pVM, PSSMHANDLE pSSM)
3123	{
3124	int rc;
3125
3126	Log(("hmR3Save:\n"));
3127
3128	for (VMCPUID i = 0; i < pVM->cCpus; i++)
3129	{
3130	Assert(!pVM->aCpus[i].hm.s.Event.fPending);
3131	if (pVM->cpum.ro.GuestFeatures.fSvm)
3132	{
3133	PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVM->aCpus[i].hm.s.svm.NstGstVmcbCache;
3134	rc = SSMR3PutBool(pSSM, pVmcbNstGstCache->fCacheValid);
3135	rc \|= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16InterceptRdCRx);
3136	rc \|= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16InterceptWrCRx);
3137	rc \|= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16InterceptRdDRx);
3138	rc \|= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16InterceptWrDRx);
3139	rc \|= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16PauseFilterThreshold);
3140	rc \|= SSMR3PutU16(pSSM, pVmcbNstGstCache->u16PauseFilterCount);
3141	rc \|= SSMR3PutU32(pSSM, pVmcbNstGstCache->u32InterceptXcpt);
3142	rc \|= SSMR3PutU64(pSSM, pVmcbNstGstCache->u64InterceptCtrl);
3143	rc \|= SSMR3PutU64(pSSM, pVmcbNstGstCache->u64TSCOffset);
3144	rc \|= SSMR3PutBool(pSSM, pVmcbNstGstCache->fVIntrMasking);
3145	rc \|= SSMR3PutBool(pSSM, pVmcbNstGstCache->fNestedPaging);
3146	rc \|= SSMR3PutBool(pSSM, pVmcbNstGstCache->fLbrVirt);
3147	AssertRCReturn(rc, rc);
3148	}
3149	}
3150
3151	/* Save the guest patch data. */
3152	rc = SSMR3PutGCPtr(pSSM, pVM->hm.s.pGuestPatchMem);
3153	rc \|= SSMR3PutGCPtr(pSSM, pVM->hm.s.pFreeGuestPatchMem);
3154	rc \|= SSMR3PutU32(pSSM, pVM->hm.s.cbGuestPatchMem);
3155
3156	/* Store all the guest patch records too. */
3157	rc \|= SSMR3PutU32(pSSM, pVM->hm.s.cPatches);
3158	AssertRCReturn(rc, rc);
3159
3160	for (uint32_t i = 0; i < pVM->hm.s.cPatches; i++)
3161	{
3162	AssertCompileSize(HMTPRINSTR, 4);
3163	PCHMTPRPATCH pPatch = &pVM->hm.s.aPatches[i];
3164	rc = SSMR3PutU32(pSSM, pPatch->Core.Key);
3165	rc \|= SSMR3PutMem(pSSM, pPatch->aOpcode, sizeof(pPatch->aOpcode));
3166	rc \|= SSMR3PutU32(pSSM, pPatch->cbOp);
3167	rc \|= SSMR3PutMem(pSSM, pPatch->aNewOpcode, sizeof(pPatch->aNewOpcode));
3168	rc \|= SSMR3PutU32(pSSM, pPatch->cbNewOp);
3169	rc \|= SSMR3PutU32(pSSM, (uint32_t)pPatch->enmType);
3170	rc \|= SSMR3PutU32(pSSM, pPatch->uSrcOperand);
3171	rc \|= SSMR3PutU32(pSSM, pPatch->uDstOperand);
3172	rc \|= SSMR3PutU32(pSSM, pPatch->pJumpTarget);
3173	rc \|= SSMR3PutU32(pSSM, pPatch->cFaults);
3174	AssertRCReturn(rc, rc);
3175	}
3176
3177	return VINF_SUCCESS;
3178	}
3179
3180
3181	/**
3182	* Execute state load operation.
3183	*
3184	* @returns VBox status code.
3185	* @param pVM The cross context VM structure.
3186	* @param pSSM SSM operation handle.
3187	* @param uVersion Data layout version.
3188	* @param uPass The data pass.
3189	*/
3190	static DECLCALLBACK(int) hmR3Load(PVM pVM, PSSMHANDLE pSSM, uint32_t uVersion, uint32_t uPass)
3191	{
3192	int rc;
3193
3194	LogFlowFunc(("uVersion=%u\n", uVersion));
3195	Assert(uPass == SSM_PASS_FINAL); NOREF(uPass);
3196
3197	/*
3198	* Validate version.
3199	*/
3200	if ( uVersion != HM_SAVED_STATE_VERSION_SVM_NESTED_HWVIRT
3201	&& uVersion != HM_SAVED_STATE_VERSION_TPR_PATCHING
3202	&& uVersion != HM_SAVED_STATE_VERSION_NO_TPR_PATCHING
3203	&& uVersion != HM_SAVED_STATE_VERSION_2_0_X)
3204	{
3205	AssertMsgFailed(("hmR3Load: Invalid version uVersion=%d!\n", uVersion));
3206	return VERR_SSM_UNSUPPORTED_DATA_UNIT_VERSION;
3207	}
3208
3209	/*
3210	* Load per-VCPU state.
3211	*/
3212	for (VMCPUID i = 0; i < pVM->cCpus; i++)
3213	{
3214	if (uVersion >= HM_SAVED_STATE_VERSION_SVM_NESTED_HWVIRT)
3215	{
3216	/* Load the SVM nested hw.virt state if the VM is configured for it. */
3217	if (pVM->cpum.ro.GuestFeatures.fSvm)
3218	{
3219	PSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVM->aCpus[i].hm.s.svm.NstGstVmcbCache;
3220	rc = SSMR3GetBool(pSSM, &pVmcbNstGstCache->fCacheValid);
3221	rc \|= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16InterceptRdCRx);
3222	rc \|= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16InterceptWrCRx);
3223	rc \|= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16InterceptRdDRx);
3224	rc \|= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16InterceptWrDRx);
3225	rc \|= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16PauseFilterThreshold);
3226	rc \|= SSMR3GetU16(pSSM, &pVmcbNstGstCache->u16PauseFilterCount);
3227	rc \|= SSMR3GetU32(pSSM, &pVmcbNstGstCache->u32InterceptXcpt);
3228	rc \|= SSMR3GetU64(pSSM, &pVmcbNstGstCache->u64InterceptCtrl);
3229	rc \|= SSMR3GetU64(pSSM, &pVmcbNstGstCache->u64TSCOffset);
3230	rc \|= SSMR3GetBool(pSSM, &pVmcbNstGstCache->fVIntrMasking);
3231	rc \|= SSMR3GetBool(pSSM, &pVmcbNstGstCache->fNestedPaging);
3232	rc \|= SSMR3GetBool(pSSM, &pVmcbNstGstCache->fLbrVirt);
3233	AssertRCReturn(rc, rc);
3234	}
3235	}
3236	else
3237	{
3238	/* Pending HM event (obsolete for a long time since TPRM holds the info.) */
3239	rc = SSMR3GetU32(pSSM, &pVM->aCpus[i].hm.s.Event.fPending);
3240	rc \|= SSMR3GetU32(pSSM, &pVM->aCpus[i].hm.s.Event.u32ErrCode);
3241	rc \|= SSMR3GetU64(pSSM, &pVM->aCpus[i].hm.s.Event.u64IntInfo);
3242
3243	/* VMX fWasInRealMode related data. */
3244	uint32_t uDummy;
3245	rc \|= SSMR3GetU32(pSSM, &uDummy); AssertRCReturn(rc, rc);
3246	rc \|= SSMR3GetU32(pSSM, &uDummy); AssertRCReturn(rc, rc);
3247	rc \|= SSMR3GetU32(pSSM, &uDummy); AssertRCReturn(rc, rc);
3248	AssertRCReturn(rc, rc);
3249	}
3250	}
3251
3252	/*
3253	* Load TPR patching data.
3254	*/
3255	if (uVersion >= HM_SAVED_STATE_VERSION_TPR_PATCHING)
3256	{
3257	rc = SSMR3GetGCPtr(pSSM, &pVM->hm.s.pGuestPatchMem);
3258	rc \|= SSMR3GetGCPtr(pSSM, &pVM->hm.s.pFreeGuestPatchMem);
3259	rc \|= SSMR3GetU32(pSSM, &pVM->hm.s.cbGuestPatchMem);
3260
3261	/* Fetch all TPR patch records. */
3262	rc \|= SSMR3GetU32(pSSM, &pVM->hm.s.cPatches);
3263	AssertRCReturn(rc, rc);
3264	for (uint32_t i = 0; i < pVM->hm.s.cPatches; i++)
3265	{
3266	PHMTPRPATCH pPatch = &pVM->hm.s.aPatches[i];
3267	rc = SSMR3GetU32(pSSM, &pPatch->Core.Key);
3268	rc \|= SSMR3GetMem(pSSM, pPatch->aOpcode, sizeof(pPatch->aOpcode));
3269	rc \|= SSMR3GetU32(pSSM, &pPatch->cbOp);
3270	rc \|= SSMR3GetMem(pSSM, pPatch->aNewOpcode, sizeof(pPatch->aNewOpcode));
3271	rc \|= SSMR3GetU32(pSSM, &pPatch->cbNewOp);
3272	rc \|= SSMR3GetU32(pSSM, (uint32_t *)&pPatch->enmType);
3273
3274	if (pPatch->enmType == HMTPRINSTR_JUMP_REPLACEMENT)
3275	pVM->hm.s.fTPRPatchingActive = true;
3276	Assert(pPatch->enmType == HMTPRINSTR_JUMP_REPLACEMENT \|\| pVM->hm.s.fTPRPatchingActive == false);
3277
3278	rc \|= SSMR3GetU32(pSSM, &pPatch->uSrcOperand);
3279	rc \|= SSMR3GetU32(pSSM, &pPatch->uDstOperand);
3280	rc \|= SSMR3GetU32(pSSM, &pPatch->cFaults);
3281	rc \|= SSMR3GetU32(pSSM, &pPatch->pJumpTarget);
3282	AssertRCReturn(rc, rc);
3283
3284	LogFlow(("hmR3Load: patch %d\n", i));
3285	LogFlow(("Key = %x\n", pPatch->Core.Key));
3286	LogFlow(("cbOp = %d\n", pPatch->cbOp));
3287	LogFlow(("cbNewOp = %d\n", pPatch->cbNewOp));
3288	LogFlow(("type = %d\n", pPatch->enmType));
3289	LogFlow(("srcop = %d\n", pPatch->uSrcOperand));
3290	LogFlow(("dstop = %d\n", pPatch->uDstOperand));
3291	LogFlow(("cFaults = %d\n", pPatch->cFaults));
3292	LogFlow(("target = %x\n", pPatch->pJumpTarget));
3293
3294	rc = RTAvloU32Insert(&pVM->hm.s.PatchTree, &pPatch->Core);
3295	AssertRCReturn(rc, rc);
3296	}
3297	}
3298
3299	return VINF_SUCCESS;
3300	}
3301
3302
3303	/**
3304	* Displays HM info.
3305	*
3306	* @param pVM The cross context VM structure.
3307	* @param pHlp The info helper functions.
3308	* @param pszArgs Arguments, ignored.
3309	*/
3310	static DECLCALLBACK(void) hmR3Info(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs)
3311	{
3312	NOREF(pszArgs);
3313	PVMCPU pVCpu = VMMGetCpu(pVM);
3314	if (!pVCpu)
3315	pVCpu = &pVM->aCpus[0];
3316
3317	if (HMIsEnabled(pVM))
3318	{
3319	if (pVM->hm.s.vmx.fSupported)
3320	pHlp->pfnPrintf(pHlp, "CPU[%u]: VT-x info:\n", pVCpu->idCpu);
3321	else
3322	pHlp->pfnPrintf(pHlp, "CPU[%u]: AMD-V info:\n", pVCpu->idCpu);
3323	pHlp->pfnPrintf(pHlp, " HM error = %#x (%u)\n", pVCpu->hm.s.u32HMError, pVCpu->hm.s.u32HMError);
3324	pHlp->pfnPrintf(pHlp, " rcLastExitToR3 = %Rrc\n", pVCpu->hm.s.rcLastExitToR3);
3325	if (pVM->hm.s.vmx.fSupported)
3326	{
3327	PCVMXVMCSINFO pVmcsInfo = hmGetVmxActiveVmcsInfo(pVCpu);
3328	bool const fRealOnV86Active = pVmcsInfo->RealMode.fRealOnV86Active;
3329	bool const fNstGstVmcsActive = pVCpu->hm.s.vmx.fSwitchedToNstGstVmcs;
3330
3331	pHlp->pfnPrintf(pHlp, " %s VMCS active\n", fNstGstVmcsActive ? "Nested-guest" : "Guest");
3332	pHlp->pfnPrintf(pHlp, " Real-on-v86 active = %RTbool\n", fRealOnV86Active);
3333	if (fRealOnV86Active)
3334	{
3335	pHlp->pfnPrintf(pHlp, " EFlags = %#x\n", pVmcsInfo->RealMode.Eflags.u32);
3336	pHlp->pfnPrintf(pHlp, " Attr CS = %#x\n", pVmcsInfo->RealMode.AttrCS.u);
3337	pHlp->pfnPrintf(pHlp, " Attr SS = %#x\n", pVmcsInfo->RealMode.AttrSS.u);
3338	pHlp->pfnPrintf(pHlp, " Attr DS = %#x\n", pVmcsInfo->RealMode.AttrDS.u);
3339	pHlp->pfnPrintf(pHlp, " Attr ES = %#x\n", pVmcsInfo->RealMode.AttrES.u);
3340	pHlp->pfnPrintf(pHlp, " Attr FS = %#x\n", pVmcsInfo->RealMode.AttrFS.u);
3341	pHlp->pfnPrintf(pHlp, " Attr GS = %#x\n", pVmcsInfo->RealMode.AttrGS.u);
3342	}
3343	}
3344	}
3345	else
3346	pHlp->pfnPrintf(pHlp, "HM is not enabled for this VM!\n");
3347	}
3348
3349
3350	/**
3351	* Displays the HM pending event.
3352	*
3353	* @param pVM The cross context VM structure.
3354	* @param pHlp The info helper functions.
3355	* @param pszArgs Arguments, ignored.
3356	*/
3357	static DECLCALLBACK(void) hmR3InfoEventPending(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs)
3358	{
3359	NOREF(pszArgs);
3360	PVMCPU pVCpu = VMMGetCpu(pVM);
3361	if (!pVCpu)
3362	pVCpu = &pVM->aCpus[0];
3363
3364	if (HMIsEnabled(pVM))
3365	{
3366	pHlp->pfnPrintf(pHlp, "CPU[%u]: HM event (fPending=%RTbool)\n", pVCpu->idCpu, pVCpu->hm.s.Event.fPending);
3367	if (pVCpu->hm.s.Event.fPending)
3368	{
3369	pHlp->pfnPrintf(pHlp, " u64IntInfo = %#RX64\n", pVCpu->hm.s.Event.u64IntInfo);
3370	pHlp->pfnPrintf(pHlp, " u32ErrCode = %#RX64\n", pVCpu->hm.s.Event.u32ErrCode);
3371	pHlp->pfnPrintf(pHlp, " cbInstr = %u bytes\n", pVCpu->hm.s.Event.cbInstr);
3372	pHlp->pfnPrintf(pHlp, " GCPtrFaultAddress = %#RGp\n", pVCpu->hm.s.Event.GCPtrFaultAddress);
3373	}
3374	}
3375	else
3376	pHlp->pfnPrintf(pHlp, "HM is not enabled for this VM!\n");
3377	}
3378
3379
3380	/**
3381	* Displays the SVM nested-guest VMCB cache.
3382	*
3383	* @param pVM The cross context VM structure.
3384	* @param pHlp The info helper functions.
3385	* @param pszArgs Arguments, ignored.
3386	*/
3387	static DECLCALLBACK(void) hmR3InfoSvmNstGstVmcbCache(PVM pVM, PCDBGFINFOHLP pHlp, const char *pszArgs)
3388	{
3389	NOREF(pszArgs);
3390	PVMCPU pVCpu = VMMGetCpu(pVM);
3391	if (!pVCpu)
3392	pVCpu = &pVM->aCpus[0];
3393
3394	bool const fSvmEnabled = HMR3IsSvmEnabled(pVM->pUVM);
3395	if ( fSvmEnabled
3396	&& pVM->cpum.ro.GuestFeatures.fSvm)
3397	{
3398	PCSVMNESTEDVMCBCACHE pVmcbNstGstCache = &pVCpu->hm.s.svm.NstGstVmcbCache;
3399	pHlp->pfnPrintf(pHlp, "CPU[%u]: HM SVM nested-guest VMCB cache\n", pVCpu->idCpu);
3400	pHlp->pfnPrintf(pHlp, " fCacheValid = %#RTbool\n", pVmcbNstGstCache->fCacheValid);
3401	pHlp->pfnPrintf(pHlp, " u16InterceptRdCRx = %#RX16\n", pVmcbNstGstCache->u16InterceptRdCRx);
3402	pHlp->pfnPrintf(pHlp, " u16InterceptWrCRx = %#RX16\n", pVmcbNstGstCache->u16InterceptWrCRx);
3403	pHlp->pfnPrintf(pHlp, " u16InterceptRdDRx = %#RX16\n", pVmcbNstGstCache->u16InterceptRdDRx);
3404	pHlp->pfnPrintf(pHlp, " u16InterceptWrDRx = %#RX16\n", pVmcbNstGstCache->u16InterceptWrDRx);
3405	pHlp->pfnPrintf(pHlp, " u16PauseFilterThreshold = %#RX16\n", pVmcbNstGstCache->u16PauseFilterThreshold);
3406	pHlp->pfnPrintf(pHlp, " u16PauseFilterCount = %#RX16\n", pVmcbNstGstCache->u16PauseFilterCount);
3407	pHlp->pfnPrintf(pHlp, " u32InterceptXcpt = %#RX32\n", pVmcbNstGstCache->u32InterceptXcpt);
3408	pHlp->pfnPrintf(pHlp, " u64InterceptCtrl = %#RX64\n", pVmcbNstGstCache->u64InterceptCtrl);
3409	pHlp->pfnPrintf(pHlp, " u64TSCOffset = %#RX64\n", pVmcbNstGstCache->u64TSCOffset);
3410	pHlp->pfnPrintf(pHlp, " fVIntrMasking = %RTbool\n", pVmcbNstGstCache->fVIntrMasking);
3411	pHlp->pfnPrintf(pHlp, " fNestedPaging = %RTbool\n", pVmcbNstGstCache->fNestedPaging);
3412	pHlp->pfnPrintf(pHlp, " fLbrVirt = %RTbool\n", pVmcbNstGstCache->fLbrVirt);
3413	}
3414	else
3415	{
3416	if (!fSvmEnabled)
3417	pHlp->pfnPrintf(pHlp, "HM SVM is not enabled for this VM!\n");
3418	else
3419	pHlp->pfnPrintf(pHlp, "SVM feature is not exposed to the guest!\n");
3420	}
3421	}
3422

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

source: vbox/trunk/src/VBox/VMM/VMMR3/HM.cpp@ 79352

Download in other formats: