VMMR0.cpp@ 92368

Last change on this file since 92368 was 92368, checked in by vboxsync, 3 years ago
VMM/PGM,GMM: Baked PGMR3PhysAllocateLargePage into PGMR0PhysAllocateLargePage eliminating VMMCALLRING3_PGM_ALLOCATE_LARGE_HANDY_PAGE; adjusted GMMR0AllocateLargePage to be ring-0 callable. Changed the large page allocation backoff logic a bit. Some more release stats. bugref:10093
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 134.2 KB

Line
1	/* $Id: VMMR0.cpp 92368 2021-11-11 13:31:21Z vboxsync $ */
2	/** @file
3	* VMM - Host Context Ring 0.
4	*/
5
6	/*
7	* Copyright (C) 2006-2020 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18
19	/*********************************************************************************************************************************
20	* Header Files *
21	*********************************************************************************************************************************/
22	#define LOG_GROUP LOG_GROUP_VMM
23	#include <VBox/vmm/vmm.h>
24	#include <VBox/sup.h>
25	#include <VBox/vmm/iom.h>
26	#include <VBox/vmm/trpm.h>
27	#include <VBox/vmm/cpum.h>
28	#include <VBox/vmm/pdmapi.h>
29	#include <VBox/vmm/pgm.h>
30	#ifdef VBOX_WITH_NEM_R0
31	# include <VBox/vmm/nem.h>
32	#endif
33	#include <VBox/vmm/em.h>
34	#include <VBox/vmm/stam.h>
35	#include <VBox/vmm/tm.h>
36	#include "VMMInternal.h"
37	#include <VBox/vmm/vmcc.h>
38	#include <VBox/vmm/gvm.h>
39	#ifdef VBOX_WITH_PCI_PASSTHROUGH
40	# include <VBox/vmm/pdmpci.h>
41	#endif
42	#include <VBox/vmm/apic.h>
43
44	#include <VBox/vmm/gvmm.h>
45	#include <VBox/vmm/gmm.h>
46	#include <VBox/vmm/gim.h>
47	#include <VBox/intnet.h>
48	#include <VBox/vmm/hm.h>
49	#include <VBox/param.h>
50	#include <VBox/err.h>
51	#include <VBox/version.h>
52	#include <VBox/log.h>
53
54	#include <iprt/asm-amd64-x86.h>
55	#include <iprt/assert.h>
56	#include <iprt/crc.h>
57	#include <iprt/mem.h>
58	#include <iprt/memobj.h>
59	#include <iprt/mp.h>
60	#include <iprt/once.h>
61	#include <iprt/semaphore.h>
62	#include <iprt/spinlock.h>
63	#include <iprt/stdarg.h>
64	#include <iprt/string.h>
65	#include <iprt/thread.h>
66	#include <iprt/timer.h>
67	#include <iprt/time.h>
68
69	#include "dtrace/VBoxVMM.h"
70
71
72	#if defined(_MSC_VER) && defined(RT_ARCH_AMD64) /** @todo check this with with VC7! */
73	# pragma intrinsic(_AddressOfReturnAddress)
74	#endif
75
76	#if defined(RT_OS_DARWIN) && ARCH_BITS == 32
77	# error "32-bit darwin is no longer supported. Go back to 4.3 or earlier!"
78	#endif
79
80
81	/*********************************************************************************************************************************
82	* Internal Functions *
83	*********************************************************************************************************************************/
84	RT_C_DECLS_BEGIN
85	#if defined(RT_ARCH_X86) && (defined(RT_OS_SOLARIS) \|\| defined(RT_OS_FREEBSD))
86	extern uint64_t __udivdi3(uint64_t, uint64_t);
87	extern uint64_t __umoddi3(uint64_t, uint64_t);
88	#endif
89	RT_C_DECLS_END
90	static int vmmR0UpdateLoggers(PGVM pGVM, VMCPUID idCpu, PVMMR0UPDATELOGGERSREQ pReq, size_t idxLogger);
91	static int vmmR0LogFlusher(PGVM pGVM);
92	static int vmmR0LogWaitFlushed(PGVM pGVM, VMCPUID idCpu, size_t idxLogger);
93	static int vmmR0InitLoggers(PGVM pGVM);
94	static void vmmR0CleanupLoggers(PGVM pGVM);
95
96
97	/*********************************************************************************************************************************
98	* Global Variables *
99	*********************************************************************************************************************************/
100	/** Drag in necessary library bits.
101	* The runtime lives here (in VMMR0.r0) and VBoxDDR0.r0 links against us. /
102	struct CLANG11WEIRDNOTHROW { PFNRT pfn; } g_VMMR0Deps[] =
103	{
104	{ (PFNRT)RTCrc32 },
105	{ (PFNRT)RTOnce },
106	#if defined(RT_ARCH_X86) && (defined(RT_OS_SOLARIS) \|\| defined(RT_OS_FREEBSD))
107	{ (PFNRT)__udivdi3 },
108	{ (PFNRT)__umoddi3 },
109	#endif
110	{ NULL }
111	};
112
113	#ifdef RT_OS_SOLARIS
114	/* Dependency information for the native solaris loader. */
115	extern "C" { char _depends_on[] = "vboxdrv"; }
116	#endif
117
118
119	/**
120	* Initialize the module.
121	* This is called when we're first loaded.
122	*
123	* @returns 0 on success.
124	* @returns VBox status on failure.
125	* @param hMod Image handle for use in APIs.
126	*/
127	DECLEXPORT(int) ModuleInit(void *hMod)
128	{
129	#ifdef VBOX_WITH_DTRACE_R0
130	/*
131	* The first thing to do is register the static tracepoints.
132	* (Deregistration is automatic.)
133	*/
134	int rc2 = SUPR0TracerRegisterModule(hMod, &g_VTGObjHeader);
135	if (RT_FAILURE(rc2))
136	return rc2;
137	#endif
138	LogFlow(("ModuleInit:\n"));
139
140	#ifdef VBOX_WITH_64ON32_CMOS_DEBUG
141	/*
142	* Display the CMOS debug code.
143	*/
144	ASMOutU8(0x72, 0x03);
145	uint8_t bDebugCode = ASMInU8(0x73);
146	LogRel(("CMOS Debug Code: %#x (%d)\n", bDebugCode, bDebugCode));
147	RTLogComPrintf("CMOS Debug Code: %#x (%d)\n", bDebugCode, bDebugCode);
148	#endif
149
150	/*
151	* Initialize the VMM, GVMM, GMM, HM, PGM (Darwin) and INTNET.
152	*/
153	int rc = vmmInitFormatTypes();
154	if (RT_SUCCESS(rc))
155	{
156	rc = GVMMR0Init();
157	if (RT_SUCCESS(rc))
158	{
159	rc = GMMR0Init();
160	if (RT_SUCCESS(rc))
161	{
162	rc = HMR0Init();
163	if (RT_SUCCESS(rc))
164	{
165	PDMR0Init(hMod);
166
167	rc = PGMRegisterStringFormatTypes();
168	if (RT_SUCCESS(rc))
169	{
170	rc = IntNetR0Init();
171	if (RT_SUCCESS(rc))
172	{
173	#ifdef VBOX_WITH_PCI_PASSTHROUGH
174	rc = PciRawR0Init();
175	#endif
176	if (RT_SUCCESS(rc))
177	{
178	rc = CPUMR0ModuleInit();
179	if (RT_SUCCESS(rc))
180	{
181	#ifdef VBOX_WITH_TRIPLE_FAULT_HACK
182	rc = vmmR0TripleFaultHackInit();
183	if (RT_SUCCESS(rc))
184	#endif
185	{
186	#ifdef VBOX_WITH_NEM_R0
187	rc = NEMR0Init();
188	if (RT_SUCCESS(rc))
189	#endif
190	{
191	LogFlow(("ModuleInit: returns success\n"));
192	return VINF_SUCCESS;
193	}
194	}
195
196	/*
197	* Bail out.
198	*/
199	#ifdef VBOX_WITH_TRIPLE_FAULT_HACK
200	vmmR0TripleFaultHackTerm();
201	#endif
202	}
203	else
204	LogRel(("ModuleInit: CPUMR0ModuleInit -> %Rrc\n", rc));
205	#ifdef VBOX_WITH_PCI_PASSTHROUGH
206	PciRawR0Term();
207	#endif
208	}
209	else
210	LogRel(("ModuleInit: PciRawR0Init -> %Rrc\n", rc));
211	IntNetR0Term();
212	}
213	else
214	LogRel(("ModuleInit: IntNetR0Init -> %Rrc\n", rc));
215	PGMDeregisterStringFormatTypes();
216	}
217	else
218	LogRel(("ModuleInit: PGMRegisterStringFormatTypes -> %Rrc\n", rc));
219	HMR0Term();
220	}
221	else
222	LogRel(("ModuleInit: HMR0Init -> %Rrc\n", rc));
223	GMMR0Term();
224	}
225	else
226	LogRel(("ModuleInit: GMMR0Init -> %Rrc\n", rc));
227	GVMMR0Term();
228	}
229	else
230	LogRel(("ModuleInit: GVMMR0Init -> %Rrc\n", rc));
231	vmmTermFormatTypes();
232	}
233	else
234	LogRel(("ModuleInit: vmmInitFormatTypes -> %Rrc\n", rc));
235
236	LogFlow(("ModuleInit: failed %Rrc\n", rc));
237	return rc;
238	}
239
240
241	/**
242	* Terminate the module.
243	* This is called when we're finally unloaded.
244	*
245	* @param hMod Image handle for use in APIs.
246	*/
247	DECLEXPORT(void) ModuleTerm(void *hMod)
248	{
249	NOREF(hMod);
250	LogFlow(("ModuleTerm:\n"));
251
252	/*
253	* Terminate the CPUM module (Local APIC cleanup).
254	*/
255	CPUMR0ModuleTerm();
256
257	/*
258	* Terminate the internal network service.
259	*/
260	IntNetR0Term();
261
262	/*
263	* PGM (Darwin), HM and PciRaw global cleanup.
264	*/
265	#ifdef VBOX_WITH_PCI_PASSTHROUGH
266	PciRawR0Term();
267	#endif
268	PGMDeregisterStringFormatTypes();
269	HMR0Term();
270	#ifdef VBOX_WITH_TRIPLE_FAULT_HACK
271	vmmR0TripleFaultHackTerm();
272	#endif
273	#ifdef VBOX_WITH_NEM_R0
274	NEMR0Term();
275	#endif
276
277	/*
278	* Destroy the GMM and GVMM instances.
279	*/
280	GMMR0Term();
281	GVMMR0Term();
282
283	vmmTermFormatTypes();
284
285	LogFlow(("ModuleTerm: returns\n"));
286	}
287
288
289	/**
290	* Initializes VMM specific members when the GVM structure is created,
291	* allocating loggers and stuff.
292	*
293	* The loggers are allocated here so that we can update their settings before
294	* doing VMMR0_DO_VMMR0_INIT and have correct logging at that time.
295	*
296	* @returns VBox status code.
297	* @param pGVM The global (ring-0) VM structure.
298	*/
299	VMMR0_INT_DECL(int) VMMR0InitPerVMData(PGVM pGVM)
300	{
301	AssertCompile(sizeof(pGVM->vmmr0.s) <= sizeof(pGVM->vmmr0.padding));
302
303	/*
304	* Initialize all members first.
305	*/
306	pGVM->vmmr0.s.fCalledInitVm = false;
307	pGVM->vmmr0.s.hMemObjLogger = NIL_RTR0MEMOBJ;
308	pGVM->vmmr0.s.hMapObjLogger = NIL_RTR0MEMOBJ;
309	pGVM->vmmr0.s.hMemObjReleaseLogger = NIL_RTR0MEMOBJ;
310	pGVM->vmmr0.s.hMapObjReleaseLogger = NIL_RTR0MEMOBJ;
311	pGVM->vmmr0.s.LogFlusher.hSpinlock = NIL_RTSPINLOCK;
312	pGVM->vmmr0.s.LogFlusher.hThread = NIL_RTNATIVETHREAD;
313	pGVM->vmmr0.s.LogFlusher.hEvent = NIL_RTSEMEVENT;
314	pGVM->vmmr0.s.LogFlusher.idxRingHead = 0;
315	pGVM->vmmr0.s.LogFlusher.idxRingTail = 0;
316	pGVM->vmmr0.s.LogFlusher.fThreadWaiting = false;
317
318	for (VMCPUID idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
319	{
320	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
321	Assert(pGVCpu->idHostCpu == NIL_RTCPUID);
322	Assert(pGVCpu->iHostCpuSet == UINT32_MAX);
323	pGVCpu->vmmr0.s.pPreemptState = NULL;
324	pGVCpu->vmmr0.s.hCtxHook = NIL_RTTHREADCTXHOOK;
325	for (size_t iLogger = 0; iLogger < RT_ELEMENTS(pGVCpu->vmmr0.s.u.aLoggers); iLogger++)
326	pGVCpu->vmmr0.s.u.aLoggers[iLogger].hEventFlushWait = NIL_RTSEMEVENT;
327	}
328
329	/*
330	* Create the loggers.
331	*/
332	return vmmR0InitLoggers(pGVM);
333	}
334
335
336	/**
337	* Initiates the R0 driver for a particular VM instance.
338	*
339	* @returns VBox status code.
340	*
341	* @param pGVM The global (ring-0) VM structure.
342	* @param uSvnRev The SVN revision of the ring-3 part.
343	* @param uBuildType Build type indicator.
344	* @thread EMT(0)
345	*/
346	static int vmmR0InitVM(PGVM pGVM, uint32_t uSvnRev, uint32_t uBuildType)
347	{
348	/*
349	* Match the SVN revisions and build type.
350	*/
351	if (uSvnRev != VMMGetSvnRev())
352	{
353	LogRel(("VMMR0InitVM: Revision mismatch, r3=%d r0=%d\n", uSvnRev, VMMGetSvnRev()));
354	SUPR0Printf("VMMR0InitVM: Revision mismatch, r3=%d r0=%d\n", uSvnRev, VMMGetSvnRev());
355	return VERR_VMM_R0_VERSION_MISMATCH;
356	}
357	if (uBuildType != vmmGetBuildType())
358	{
359	LogRel(("VMMR0InitVM: Build type mismatch, r3=%#x r0=%#x\n", uBuildType, vmmGetBuildType()));
360	SUPR0Printf("VMMR0InitVM: Build type mismatch, r3=%#x r0=%#x\n", uBuildType, vmmGetBuildType());
361	return VERR_VMM_R0_VERSION_MISMATCH;
362	}
363
364	int rc = GVMMR0ValidateGVMandEMT(pGVM, 0 /idCpu/);
365	if (RT_FAILURE(rc))
366	return rc;
367
368	/* Don't allow this to be called more than once. */
369	if (!pGVM->vmmr0.s.fCalledInitVm)
370	pGVM->vmmr0.s.fCalledInitVm = true;
371	else
372	return VERR_ALREADY_INITIALIZED;
373
374	#ifdef LOG_ENABLED
375
376	/*
377	* Register the EMT R0 logger instance for VCPU 0.
378	*/
379	PVMCPUCC pVCpu = VMCC_GET_CPU_0(pGVM);
380	if (pVCpu->vmmr0.s.u.s.Logger.pLogger)
381	{
382	# if 0 /* testing of the logger. */
383	LogCom(("vmmR0InitVM: before %p\n", RTLogDefaultInstance()));
384	LogCom(("vmmR0InitVM: pfnFlush=%p actual=%p\n", pR0Logger->Logger.pfnFlush, vmmR0LoggerFlush));
385	LogCom(("vmmR0InitVM: pfnLogger=%p actual=%p\n", pR0Logger->Logger.pfnLogger, vmmR0LoggerWrapper));
386	LogCom(("vmmR0InitVM: offScratch=%d fFlags=%#x fDestFlags=%#x\n", pR0Logger->Logger.offScratch, pR0Logger->Logger.fFlags, pR0Logger->Logger.fDestFlags));
387
388	RTLogSetDefaultInstanceThread(&pR0Logger->Logger, (uintptr_t)pGVM->pSession);
389	LogCom(("vmmR0InitVM: after %p reg\n", RTLogDefaultInstance()));
390	RTLogSetDefaultInstanceThread(NULL, pGVM->pSession);
391	LogCom(("vmmR0InitVM: after %p dereg\n", RTLogDefaultInstance()));
392
393	pR0Logger->Logger.pfnLogger("hello ring-0 logger\n");
394	LogCom(("vmmR0InitVM: returned successfully from direct logger call.\n"));
395	pR0Logger->Logger.pfnFlush(&pR0Logger->Logger);
396	LogCom(("vmmR0InitVM: returned successfully from direct flush call.\n"));
397
398	RTLogSetDefaultInstanceThread(&pR0Logger->Logger, (uintptr_t)pGVM->pSession);
399	LogCom(("vmmR0InitVM: after %p reg2\n", RTLogDefaultInstance()));
400	pR0Logger->Logger.pfnLogger("hello ring-0 logger\n");
401	LogCom(("vmmR0InitVM: returned successfully from direct logger call (2). offScratch=%d\n", pR0Logger->Logger.offScratch));
402	RTLogSetDefaultInstanceThread(NULL, pGVM->pSession);
403	LogCom(("vmmR0InitVM: after %p dereg2\n", RTLogDefaultInstance()));
404
405	RTLogLoggerEx(&pR0Logger->Logger, 0, ~0U, "hello ring-0 logger (RTLogLoggerEx)\n");
406	LogCom(("vmmR0InitVM: RTLogLoggerEx returned fine offScratch=%d\n", pR0Logger->Logger.offScratch));
407
408	RTLogSetDefaultInstanceThread(&pR0Logger->Logger, (uintptr_t)pGVM->pSession);
409	RTLogPrintf("hello ring-0 logger (RTLogPrintf)\n");
410	LogCom(("vmmR0InitVM: RTLogPrintf returned fine offScratch=%d\n", pR0Logger->Logger.offScratch));
411	# endif
412	# ifdef VBOX_WITH_R0_LOGGING
413	Log(("Switching to per-thread logging instance %p (key=%p)\n", pVCpu->vmmr0.s.u.s.Logger.pLogger, pGVM->pSession));
414	RTLogSetDefaultInstanceThread(pVCpu->vmmr0.s.u.s.Logger.pLogger, (uintptr_t)pGVM->pSession);
415	pVCpu->vmmr0.s.u.s.Logger.fRegistered = true;
416	# endif
417	}
418	#endif /* LOG_ENABLED */
419
420	/*
421	* Check if the host supports high resolution timers or not.
422	*/
423	if ( pGVM->vmm.s.fUsePeriodicPreemptionTimers
424	&& !RTTimerCanDoHighResolution())
425	pGVM->vmm.s.fUsePeriodicPreemptionTimers = false;
426
427	/*
428	* Initialize the per VM data for GVMM and GMM.
429	*/
430	rc = GVMMR0InitVM(pGVM);
431	if (RT_SUCCESS(rc))
432	{
433	/*
434	* Init HM, CPUM and PGM (Darwin only).
435	*/
436	rc = HMR0InitVM(pGVM);
437	if (RT_SUCCESS(rc))
438	{
439	rc = CPUMR0InitVM(pGVM);
440	if (RT_SUCCESS(rc))
441	{
442	rc = PGMR0InitVM(pGVM);
443	if (RT_SUCCESS(rc))
444	{
445	rc = EMR0InitVM(pGVM);
446	if (RT_SUCCESS(rc))
447	{
448	#ifdef VBOX_WITH_PCI_PASSTHROUGH
449	rc = PciRawR0InitVM(pGVM);
450	#endif
451	if (RT_SUCCESS(rc))
452	{
453	rc = GIMR0InitVM(pGVM);
454	if (RT_SUCCESS(rc))
455	{
456	GVMMR0DoneInitVM(pGVM);
457
458	/*
459	* Collect a bit of info for the VM release log.
460	*/
461	pGVM->vmm.s.fIsPreemptPendingApiTrusty = RTThreadPreemptIsPendingTrusty();
462	pGVM->vmm.s.fIsPreemptPossible = RTThreadPreemptIsPossible();;
463	return rc;
464
465	/* bail out*/
466	//GIMR0TermVM(pGVM);
467	}
468	#ifdef VBOX_WITH_PCI_PASSTHROUGH
469	PciRawR0TermVM(pGVM);
470	#endif
471	}
472	}
473	}
474	}
475	HMR0TermVM(pGVM);
476	}
477	}
478
479	RTLogSetDefaultInstanceThread(NULL, (uintptr_t)pGVM->pSession);
480	return rc;
481	}
482
483
484	/**
485	* Does EMT specific VM initialization.
486	*
487	* @returns VBox status code.
488	* @param pGVM The ring-0 VM structure.
489	* @param idCpu The EMT that's calling.
490	*/
491	static int vmmR0InitVMEmt(PGVM pGVM, VMCPUID idCpu)
492	{
493	/* Paranoia (caller checked these already). */
494	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID);
495	AssertReturn(pGVM->aCpus[idCpu].hEMT == RTThreadNativeSelf(), VERR_INVALID_CPU_ID);
496
497	#if defined(LOG_ENABLED) && defined(VBOX_WITH_R0_LOGGING)
498	/*
499	* Registration of ring 0 loggers.
500	*/
501	PVMCPUCC pVCpu = &pGVM->aCpus[idCpu];
502	if ( pVCpu->vmmr0.s.u.s.Logger.pLogger
503	&& !pVCpu->vmmr0.s.u.s.Logger.fRegistered)
504	{
505	RTLogSetDefaultInstanceThread(pVCpu->vmmr0.s.u.s.Logger.pLogger, (uintptr_t)pGVM->pSession);
506	pVCpu->vmmr0.s.u.s.Logger.fRegistered = true;
507	}
508	#endif
509
510	return VINF_SUCCESS;
511	}
512
513
514
515	/**
516	* Terminates the R0 bits for a particular VM instance.
517	*
518	* This is normally called by ring-3 as part of the VM termination process, but
519	* may alternatively be called during the support driver session cleanup when
520	* the VM object is destroyed (see GVMM).
521	*
522	* @returns VBox status code.
523	*
524	* @param pGVM The global (ring-0) VM structure.
525	* @param idCpu Set to 0 if EMT(0) or NIL_VMCPUID if session cleanup
526	* thread.
527	* @thread EMT(0) or session clean up thread.
528	*/
529	VMMR0_INT_DECL(int) VMMR0TermVM(PGVM pGVM, VMCPUID idCpu)
530	{
531	/*
532	* Check EMT(0) claim if we're called from userland.
533	*/
534	if (idCpu != NIL_VMCPUID)
535	{
536	AssertReturn(idCpu == 0, VERR_INVALID_CPU_ID);
537	int rc = GVMMR0ValidateGVMandEMT(pGVM, idCpu);
538	if (RT_FAILURE(rc))
539	return rc;
540	}
541
542	#ifdef VBOX_WITH_PCI_PASSTHROUGH
543	PciRawR0TermVM(pGVM);
544	#endif
545
546	/*
547	* Tell GVMM what we're up to and check that we only do this once.
548	*/
549	if (GVMMR0DoingTermVM(pGVM))
550	{
551	GIMR0TermVM(pGVM);
552
553	/** @todo I wish to call PGMR0PhysFlushHandyPages(pGVM, &pGVM->aCpus[idCpu])
554	* here to make sure we don't leak any shared pages if we crash... */
555	HMR0TermVM(pGVM);
556	}
557
558	/*
559	* Deregister the logger for this EMT.
560	*/
561	RTLogSetDefaultInstanceThread(NULL, (uintptr_t)pGVM->pSession);
562
563	/*
564	* Start log flusher thread termination.
565	*/
566	ASMAtomicWriteBool(&pGVM->vmmr0.s.LogFlusher.fThreadShutdown, true);
567	if (pGVM->vmmr0.s.LogFlusher.hEvent != NIL_RTSEMEVENT)
568	RTSemEventSignal(pGVM->vmmr0.s.LogFlusher.hEvent);
569
570	return VINF_SUCCESS;
571	}
572
573
574	/**
575	* This is called at the end of gvmmR0CleanupVM().
576	*
577	* @param pGVM The global (ring-0) VM structure.
578	*/
579	VMMR0_INT_DECL(void) VMMR0CleanupVM(PGVM pGVM)
580	{
581	AssertCompile(NIL_RTTHREADCTXHOOK == (RTTHREADCTXHOOK)0); /* Depends on zero initialized memory working for NIL at the moment. */
582	for (VMCPUID idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
583	{
584	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
585
586	/** @todo Can we busy wait here for all thread-context hooks to be
587	* deregistered before releasing (destroying) it? Only until we find a
588	* solution for not deregistering hooks everytime we're leaving HMR0
589	* context. */
590	VMMR0ThreadCtxHookDestroyForEmt(pGVCpu);
591	}
592
593	vmmR0CleanupLoggers(pGVM);
594	}
595
596
597	/**
598	* An interrupt or unhalt force flag is set, deal with it.
599	*
600	* @returns VINF_SUCCESS (or VINF_EM_HALT).
601	* @param pVCpu The cross context virtual CPU structure.
602	* @param uMWait Result from EMMonitorWaitIsActive().
603	* @param enmInterruptibility Guest CPU interruptbility level.
604	*/
605	static int vmmR0DoHaltInterrupt(PVMCPUCC pVCpu, unsigned uMWait, CPUMINTERRUPTIBILITY enmInterruptibility)
606	{
607	Assert(!TRPMHasTrap(pVCpu));
608	Assert( enmInterruptibility > CPUMINTERRUPTIBILITY_INVALID
609	&& enmInterruptibility < CPUMINTERRUPTIBILITY_END);
610
611	/*
612	* Pending interrupts w/o any SMIs or NMIs? That the usual case.
613	*/
614	if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
615	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_SMI \| VMCPU_FF_INTERRUPT_NMI))
616	{
617	if (enmInterruptibility <= CPUMINTERRUPTIBILITY_UNRESTRAINED)
618	{
619	uint8_t u8Interrupt = 0;
620	int rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
621	Log(("vmmR0DoHaltInterrupt: CPU%d u8Interrupt=%d (%#x) rc=%Rrc\n", pVCpu->idCpu, u8Interrupt, u8Interrupt, rc));
622	if (RT_SUCCESS(rc))
623	{
624	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_UNHALT);
625
626	rc = TRPMAssertTrap(pVCpu, u8Interrupt, TRPM_HARDWARE_INT);
627	AssertRCSuccess(rc);
628	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltExec);
629	return rc;
630	}
631	}
632	}
633	/*
634	* SMI is not implemented yet, at least not here.
635	*/
636	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_SMI))
637	{
638	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #3\n", pVCpu->idCpu));
639	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
640	return VINF_EM_HALT;
641	}
642	/*
643	* NMI.
644	*/
645	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
646	{
647	if (enmInterruptibility < CPUMINTERRUPTIBILITY_NMI_INHIBIT)
648	{
649	/** @todo later. */
650	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #2 (uMWait=%u enmInt=%d)\n", pVCpu->idCpu, uMWait, enmInterruptibility));
651	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
652	return VINF_EM_HALT;
653	}
654	}
655	/*
656	* Nested-guest virtual interrupt.
657	*/
658	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NESTED_GUEST))
659	{
660	if (enmInterruptibility < CPUMINTERRUPTIBILITY_VIRT_INT_DISABLED)
661	{
662	/** @todo NSTVMX: NSTSVM: Remember, we might have to check and perform VM-exits
663	* here before injecting the virtual interrupt. See emR3ForcedActions
664	* for details. */
665	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #1 (uMWait=%u enmInt=%d)\n", pVCpu->idCpu, uMWait, enmInterruptibility));
666	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
667	return VINF_EM_HALT;
668	}
669	}
670
671	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UNHALT))
672	{
673	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltExec);
674	Log11(("vmmR0DoHaltInterrupt: CPU%d success VINF_SUCCESS (UNHALT)\n", pVCpu->idCpu));
675	return VINF_SUCCESS;
676	}
677	if (uMWait > 1)
678	{
679	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltExec);
680	Log11(("vmmR0DoHaltInterrupt: CPU%d success VINF_SUCCESS (uMWait=%u > 1)\n", pVCpu->idCpu, uMWait));
681	return VINF_SUCCESS;
682	}
683
684	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #0 (uMWait=%u enmInt=%d)\n", pVCpu->idCpu, uMWait, enmInterruptibility));
685	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
686	return VINF_EM_HALT;
687	}
688
689
690	/**
691	* This does one round of vmR3HaltGlobal1Halt().
692	*
693	* The rational here is that we'll reduce latency in interrupt situations if we
694	* don't go to ring-3 immediately on a VINF_EM_HALT (guest executed HLT or
695	* MWAIT), but do one round of blocking here instead and hope the interrupt is
696	* raised in the meanwhile.
697	*
698	* If we go to ring-3 we'll quit the inner HM/NEM loop in EM and end up in the
699	* outer loop, which will then call VMR3WaitHalted() and that in turn will do a
700	* ring-0 call (unless we're too close to a timer event). When the interrupt
701	* wakes us up, we'll return from ring-0 and EM will by instinct do a
702	* rescheduling (because of raw-mode) before it resumes the HM/NEM loop and gets
703	* back to VMMR0EntryFast().
704	*
705	* @returns VINF_SUCCESS or VINF_EM_HALT.
706	* @param pGVM The ring-0 VM structure.
707	* @param pGVCpu The ring-0 virtual CPU structure.
708	*
709	* @todo r=bird: All the blocking/waiting and EMT managment should move out of
710	* the VM module, probably to VMM. Then this would be more weird wrt
711	* parameters and statistics.
712	*/
713	static int vmmR0DoHalt(PGVM pGVM, PGVMCPU pGVCpu)
714	{
715	/*
716	* Do spin stat historization.
717	*/
718	if (++pGVCpu->vmm.s.cR0Halts & 0xff)
719	{ /* likely */ }
720	else if (pGVCpu->vmm.s.cR0HaltsSucceeded > pGVCpu->vmm.s.cR0HaltsToRing3)
721	{
722	pGVCpu->vmm.s.cR0HaltsSucceeded = 2;
723	pGVCpu->vmm.s.cR0HaltsToRing3 = 0;
724	}
725	else
726	{
727	pGVCpu->vmm.s.cR0HaltsSucceeded = 0;
728	pGVCpu->vmm.s.cR0HaltsToRing3 = 2;
729	}
730
731	/*
732	* Flags that makes us go to ring-3.
733	*/
734	uint32_t const fVmFFs = VM_FF_TM_VIRTUAL_SYNC \| VM_FF_PDM_QUEUES \| VM_FF_PDM_DMA
735	\| VM_FF_DBGF \| VM_FF_REQUEST \| VM_FF_CHECK_VM_STATE
736	\| VM_FF_RESET \| VM_FF_EMT_RENDEZVOUS \| VM_FF_PGM_NEED_HANDY_PAGES
737	\| VM_FF_PGM_NO_MEMORY \| VM_FF_DEBUG_SUSPEND;
738	uint64_t const fCpuFFs = VMCPU_FF_TIMER \| VMCPU_FF_PDM_CRITSECT \| VMCPU_FF_IEM
739	\| VMCPU_FF_REQUEST \| VMCPU_FF_DBGF \| VMCPU_FF_HM_UPDATE_CR3
740	\| VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL
741	\| VMCPU_FF_TO_R3 \| VMCPU_FF_IOM;
742
743	/*
744	* Check preconditions.
745	*/
746	unsigned const uMWait = EMMonitorWaitIsActive(pGVCpu);
747	CPUMINTERRUPTIBILITY const enmInterruptibility = CPUMGetGuestInterruptibility(pGVCpu);
748	if ( pGVCpu->vmm.s.fMayHaltInRing0
749	&& !TRPMHasTrap(pGVCpu)
750	&& ( enmInterruptibility == CPUMINTERRUPTIBILITY_UNRESTRAINED
751	\|\| uMWait > 1))
752	{
753	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
754	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
755	{
756	/*
757	* Interrupts pending already?
758	*/
759	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
760	APICUpdatePendingInterrupts(pGVCpu);
761
762	/*
763	* Flags that wake up from the halted state.
764	*/
765	uint64_t const fIntMask = VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC \| VMCPU_FF_INTERRUPT_NESTED_GUEST
766	\| VMCPU_FF_INTERRUPT_NMI \| VMCPU_FF_INTERRUPT_SMI \| VMCPU_FF_UNHALT;
767
768	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
769	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
770	ASMNopPause();
771
772	/*
773	* Check out how long till the next timer event.
774	*/
775	uint64_t u64Delta;
776	uint64_t u64GipTime = TMTimerPollGIP(pGVM, pGVCpu, &u64Delta);
777
778	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
779	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
780	{
781	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
782	APICUpdatePendingInterrupts(pGVCpu);
783
784	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
785	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
786
787	/*
788	* Wait if there is enough time to the next timer event.
789	*/
790	if (u64Delta >= pGVCpu->vmm.s.cNsSpinBlockThreshold)
791	{
792	/* If there are few other CPU cores around, we will procrastinate a
793	little before going to sleep, hoping for some device raising an
794	interrupt or similar. Though, the best thing here would be to
795	dynamically adjust the spin count according to its usfulness or
796	something... */
797	if ( pGVCpu->vmm.s.cR0HaltsSucceeded > pGVCpu->vmm.s.cR0HaltsToRing3
798	&& RTMpGetOnlineCount() >= 4)
799	{
800	/** @todo Figure out how we can skip this if it hasn't help recently...
801	* @bugref{9172#c12} */
802	uint32_t cSpinLoops = 42;
803	while (cSpinLoops-- > 0)
804	{
805	ASMNopPause();
806	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
807	APICUpdatePendingInterrupts(pGVCpu);
808	ASMNopPause();
809	if (VM_FF_IS_ANY_SET(pGVM, fVmFFs))
810	{
811	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3FromSpin);
812	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3);
813	return VINF_EM_HALT;
814	}
815	ASMNopPause();
816	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
817	{
818	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3FromSpin);
819	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3);
820	return VINF_EM_HALT;
821	}
822	ASMNopPause();
823	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
824	{
825	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltExecFromSpin);
826	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
827	}
828	ASMNopPause();
829	}
830	}
831
832	/*
833	* We have to set the state to VMCPUSTATE_STARTED_HALTED here so ring-3
834	* knows when to notify us (cannot access VMINTUSERPERVMCPU::fWait from here).
835	* After changing the state we must recheck the force flags of course.
836	*/
837	if (VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED_HALTED, VMCPUSTATE_STARTED))
838	{
839	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
840	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
841	{
842	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
843	APICUpdatePendingInterrupts(pGVCpu);
844
845	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
846	{
847	VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED, VMCPUSTATE_STARTED_HALTED);
848	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
849	}
850
851	/* Okay, block! */
852	uint64_t const u64StartSchedHalt = RTTimeNanoTS();
853	int rc = GVMMR0SchedHalt(pGVM, pGVCpu, u64GipTime);
854	uint64_t const u64EndSchedHalt = RTTimeNanoTS();
855	uint64_t const cNsElapsedSchedHalt = u64EndSchedHalt - u64StartSchedHalt;
856	Log10(("vmmR0DoHalt: CPU%d: halted %llu ns\n", pGVCpu->idCpu, cNsElapsedSchedHalt));
857
858	VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED, VMCPUSTATE_STARTED_HALTED);
859	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlock, cNsElapsedSchedHalt);
860	if ( rc == VINF_SUCCESS
861	\|\| rc == VERR_INTERRUPTED)
862	{
863	/* Keep some stats like ring-3 does. */
864	int64_t const cNsOverslept = u64EndSchedHalt - u64GipTime;
865	if (cNsOverslept > 50000)
866	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlockOverslept, cNsOverslept);
867	else if (cNsOverslept < -50000)
868	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlockInsomnia, cNsElapsedSchedHalt);
869	else
870	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlockOnTime, cNsElapsedSchedHalt);
871
872	/*
873	* Recheck whether we can resume execution or have to go to ring-3.
874	*/
875	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
876	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
877	{
878	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
879	APICUpdatePendingInterrupts(pGVCpu);
880	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
881	{
882	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltExecFromBlock);
883	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
884	}
885	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PostNoInt);
886	Log12(("vmmR0DoHalt: CPU%d post #2 - No pending interrupt\n", pGVCpu->idCpu));
887	}
888	else
889	{
890	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PostPendingFF);
891	Log12(("vmmR0DoHalt: CPU%d post #1 - Pending FF\n", pGVCpu->idCpu));
892	}
893	}
894	else
895	{
896	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3Other);
897	Log12(("vmmR0DoHalt: CPU%d GVMMR0SchedHalt failed: %Rrc\n", pGVCpu->idCpu, rc));
898	}
899	}
900	else
901	{
902	VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED, VMCPUSTATE_STARTED_HALTED);
903	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PendingFF);
904	Log12(("vmmR0DoHalt: CPU%d failed #5 - Pending FF\n", pGVCpu->idCpu));
905	}
906	}
907	else
908	{
909	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3Other);
910	Log12(("vmmR0DoHalt: CPU%d failed #4 - enmState=%d\n", pGVCpu->idCpu, VMCPU_GET_STATE(pGVCpu)));
911	}
912	}
913	else
914	{
915	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3SmallDelta);
916	Log12(("vmmR0DoHalt: CPU%d failed #3 - delta too small: %RU64\n", pGVCpu->idCpu, u64Delta));
917	}
918	}
919	else
920	{
921	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PendingFF);
922	Log12(("vmmR0DoHalt: CPU%d failed #2 - Pending FF\n", pGVCpu->idCpu));
923	}
924	}
925	else
926	{
927	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PendingFF);
928	Log12(("vmmR0DoHalt: CPU%d failed #1 - Pending FF\n", pGVCpu->idCpu));
929	}
930	}
931	else
932	{
933	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3Other);
934	Log12(("vmmR0DoHalt: CPU%d failed #0 - fMayHaltInRing0=%d TRPMHasTrap=%d enmInt=%d uMWait=%u\n",
935	pGVCpu->idCpu, pGVCpu->vmm.s.fMayHaltInRing0, TRPMHasTrap(pGVCpu), enmInterruptibility, uMWait));
936	}
937
938	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3);
939	return VINF_EM_HALT;
940	}
941
942
943	/**
944	* VMM ring-0 thread-context callback.
945	*
946	* This does common HM state updating and calls the HM-specific thread-context
947	* callback.
948	*
949	* This is used together with RTThreadCtxHookCreate() on platforms which
950	* supports it, and directly from VMMR0EmtPrepareForBlocking() and
951	* VMMR0EmtResumeAfterBlocking() on platforms which don't.
952	*
953	* @param enmEvent The thread-context event.
954	* @param pvUser Opaque pointer to the VMCPU.
955	*
956	* @thread EMT(pvUser)
957	*/
958	static DECLCALLBACK(void) vmmR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, void *pvUser)
959	{
960	PVMCPUCC pVCpu = (PVMCPUCC)pvUser;
961
962	switch (enmEvent)
963	{
964	case RTTHREADCTXEVENT_IN:
965	{
966	/*
967	* Linux may call us with preemption enabled (really!) but technically we
968	* cannot get preempted here, otherwise we end up in an infinite recursion
969	* scenario (i.e. preempted in resume hook -> preempt hook -> resume hook...
970	* ad infinitum). Let's just disable preemption for now...
971	*/
972	/** @todo r=bird: I don't believe the above. The linux code is clearly enabling
973	* preemption after doing the callout (one or two functions up the
974	* call chain). */
975	/** @todo r=ramshankar: See @bugref{5313#c30}. */
976	RTTHREADPREEMPTSTATE ParanoidPreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
977	RTThreadPreemptDisable(&ParanoidPreemptState);
978
979	/* We need to update the VCPU <-> host CPU mapping. */
980	RTCPUID idHostCpu;
981	uint32_t iHostCpuSet = RTMpCurSetIndexAndId(&idHostCpu);
982	pVCpu->iHostCpuSet = iHostCpuSet;
983	ASMAtomicWriteU32(&pVCpu->idHostCpu, idHostCpu);
984
985	/* In the very unlikely event that the GIP delta for the CPU we're
986	rescheduled needs calculating, try force a return to ring-3.
987	We unfortunately cannot do the measurements right here. */
988	if (RT_LIKELY(!SUPIsTscDeltaAvailableForCpuSetIndex(iHostCpuSet)))
989	{ /* likely */ }
990	else
991	VMCPU_FF_SET(pVCpu, VMCPU_FF_TO_R3);
992
993	/* Invoke the HM-specific thread-context callback. */
994	HMR0ThreadCtxCallback(enmEvent, pvUser);
995
996	/* Restore preemption. */
997	RTThreadPreemptRestore(&ParanoidPreemptState);
998	break;
999	}
1000
1001	case RTTHREADCTXEVENT_OUT:
1002	{
1003	/* Invoke the HM-specific thread-context callback. */
1004	HMR0ThreadCtxCallback(enmEvent, pvUser);
1005
1006	/*
1007	* Sigh. See VMMGetCpu() used by VMCPU_ASSERT_EMT(). We cannot let several VCPUs
1008	* have the same host CPU associated with it.
1009	*/
1010	pVCpu->iHostCpuSet = UINT32_MAX;
1011	ASMAtomicWriteU32(&pVCpu->idHostCpu, NIL_RTCPUID);
1012	break;
1013	}
1014
1015	default:
1016	/* Invoke the HM-specific thread-context callback. */
1017	HMR0ThreadCtxCallback(enmEvent, pvUser);
1018	break;
1019	}
1020	}
1021
1022
1023	/**
1024	* Creates thread switching hook for the current EMT thread.
1025	*
1026	* This is called by GVMMR0CreateVM and GVMMR0RegisterVCpu. If the host
1027	* platform does not implement switcher hooks, no hooks will be create and the
1028	* member set to NIL_RTTHREADCTXHOOK.
1029	*
1030	* @returns VBox status code.
1031	* @param pVCpu The cross context virtual CPU structure.
1032	* @thread EMT(pVCpu)
1033	*/
1034	VMMR0_INT_DECL(int) VMMR0ThreadCtxHookCreateForEmt(PVMCPUCC pVCpu)
1035	{
1036	VMCPU_ASSERT_EMT(pVCpu);
1037	Assert(pVCpu->vmmr0.s.hCtxHook == NIL_RTTHREADCTXHOOK);
1038
1039	#if 1 /* To disable this stuff change to zero. */
1040	int rc = RTThreadCtxHookCreate(&pVCpu->vmmr0.s.hCtxHook, 0, vmmR0ThreadCtxCallback, pVCpu);
1041	if (RT_SUCCESS(rc))
1042	{
1043	pVCpu->pGVM->vmm.s.fIsUsingContextHooks = true;
1044	return rc;
1045	}
1046	#else
1047	RT_NOREF(vmmR0ThreadCtxCallback);
1048	int rc = VERR_NOT_SUPPORTED;
1049	#endif
1050
1051	pVCpu->vmmr0.s.hCtxHook = NIL_RTTHREADCTXHOOK;
1052	pVCpu->pGVM->vmm.s.fIsUsingContextHooks = false;
1053	if (rc == VERR_NOT_SUPPORTED)
1054	return VINF_SUCCESS;
1055
1056	LogRelMax(32, ("RTThreadCtxHookCreate failed! rc=%Rrc pVCpu=%p idCpu=%RU32\n", rc, pVCpu, pVCpu->idCpu));
1057	return VINF_SUCCESS; /* Just ignore it, we can live without context hooks. */
1058	}
1059
1060
1061	/**
1062	* Destroys the thread switching hook for the specified VCPU.
1063	*
1064	* @param pVCpu The cross context virtual CPU structure.
1065	* @remarks Can be called from any thread.
1066	*/
1067	VMMR0_INT_DECL(void) VMMR0ThreadCtxHookDestroyForEmt(PVMCPUCC pVCpu)
1068	{
1069	int rc = RTThreadCtxHookDestroy(pVCpu->vmmr0.s.hCtxHook);
1070	AssertRC(rc);
1071	pVCpu->vmmr0.s.hCtxHook = NIL_RTTHREADCTXHOOK;
1072	}
1073
1074
1075	/**
1076	* Disables the thread switching hook for this VCPU (if we got one).
1077	*
1078	* @param pVCpu The cross context virtual CPU structure.
1079	* @thread EMT(pVCpu)
1080	*
1081	* @remarks This also clears GVMCPU::idHostCpu, so the mapping is invalid after
1082	* this call. This means you have to be careful with what you do!
1083	*/
1084	VMMR0_INT_DECL(void) VMMR0ThreadCtxHookDisable(PVMCPUCC pVCpu)
1085	{
1086	/*
1087	* Clear the VCPU <-> host CPU mapping as we've left HM context.
1088	* @bugref{7726#c19} explains the need for this trick:
1089	*
1090	* VMXR0CallRing3Callback/SVMR0CallRing3Callback &
1091	* hmR0VmxLeaveSession/hmR0SvmLeaveSession disables context hooks during
1092	* longjmp & normal return to ring-3, which opens a window where we may be
1093	* rescheduled without changing GVMCPUID::idHostCpu and cause confusion if
1094	* the CPU starts executing a different EMT. Both functions first disables
1095	* preemption and then calls HMR0LeaveCpu which invalids idHostCpu, leaving
1096	* an opening for getting preempted.
1097	*/
1098	/** @todo Make HM not need this API! Then we could leave the hooks enabled
1099	* all the time. */
1100
1101	/*
1102	* Disable the context hook, if we got one.
1103	*/
1104	if (pVCpu->vmmr0.s.hCtxHook != NIL_RTTHREADCTXHOOK)
1105	{
1106	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1107	ASMAtomicWriteU32(&pVCpu->idHostCpu, NIL_RTCPUID);
1108	int rc = RTThreadCtxHookDisable(pVCpu->vmmr0.s.hCtxHook);
1109	AssertRC(rc);
1110	}
1111	}
1112
1113
1114	/**
1115	* Internal version of VMMR0ThreadCtxHooksAreRegistered.
1116	*
1117	* @returns true if registered, false otherwise.
1118	* @param pVCpu The cross context virtual CPU structure.
1119	*/
1120	DECLINLINE(bool) vmmR0ThreadCtxHookIsEnabled(PVMCPUCC pVCpu)
1121	{
1122	return RTThreadCtxHookIsEnabled(pVCpu->vmmr0.s.hCtxHook);
1123	}
1124
1125
1126	/**
1127	* Whether thread-context hooks are registered for this VCPU.
1128	*
1129	* @returns true if registered, false otherwise.
1130	* @param pVCpu The cross context virtual CPU structure.
1131	*/
1132	VMMR0_INT_DECL(bool) VMMR0ThreadCtxHookIsEnabled(PVMCPUCC pVCpu)
1133	{
1134	return vmmR0ThreadCtxHookIsEnabled(pVCpu);
1135	}
1136
1137
1138	/**
1139	* Returns the ring-0 release logger instance.
1140	*
1141	* @returns Pointer to release logger, NULL if not configured.
1142	* @param pVCpu The cross context virtual CPU structure of the caller.
1143	* @thread EMT(pVCpu)
1144	*/
1145	VMMR0_INT_DECL(PRTLOGGER) VMMR0GetReleaseLogger(PVMCPUCC pVCpu)
1146	{
1147	return pVCpu->vmmr0.s.u.s.RelLogger.pLogger;
1148	}
1149
1150
1151	#ifdef VBOX_WITH_STATISTICS
1152	/**
1153	* Record return code statistics
1154	* @param pVM The cross context VM structure.
1155	* @param pVCpu The cross context virtual CPU structure.
1156	* @param rc The status code.
1157	*/
1158	static void vmmR0RecordRC(PVMCC pVM, PVMCPUCC pVCpu, int rc)
1159	{
1160	/*
1161	* Collect statistics.
1162	*/
1163	switch (rc)
1164	{
1165	case VINF_SUCCESS:
1166	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetNormal);
1167	break;
1168	case VINF_EM_RAW_INTERRUPT:
1169	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetInterrupt);
1170	break;
1171	case VINF_EM_RAW_INTERRUPT_HYPER:
1172	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetInterruptHyper);
1173	break;
1174	case VINF_EM_RAW_GUEST_TRAP:
1175	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetGuestTrap);
1176	break;
1177	case VINF_EM_RAW_RING_SWITCH:
1178	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetRingSwitch);
1179	break;
1180	case VINF_EM_RAW_RING_SWITCH_INT:
1181	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetRingSwitchInt);
1182	break;
1183	case VINF_EM_RAW_STALE_SELECTOR:
1184	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetStaleSelector);
1185	break;
1186	case VINF_EM_RAW_IRET_TRAP:
1187	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIRETTrap);
1188	break;
1189	case VINF_IOM_R3_IOPORT_READ:
1190	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIORead);
1191	break;
1192	case VINF_IOM_R3_IOPORT_WRITE:
1193	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIOWrite);
1194	break;
1195	case VINF_IOM_R3_IOPORT_COMMIT_WRITE:
1196	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIOCommitWrite);
1197	break;
1198	case VINF_IOM_R3_MMIO_READ:
1199	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIORead);
1200	break;
1201	case VINF_IOM_R3_MMIO_WRITE:
1202	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOWrite);
1203	break;
1204	case VINF_IOM_R3_MMIO_COMMIT_WRITE:
1205	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOCommitWrite);
1206	break;
1207	case VINF_IOM_R3_MMIO_READ_WRITE:
1208	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOReadWrite);
1209	break;
1210	case VINF_PATM_HC_MMIO_PATCH_READ:
1211	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOPatchRead);
1212	break;
1213	case VINF_PATM_HC_MMIO_PATCH_WRITE:
1214	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOPatchWrite);
1215	break;
1216	case VINF_CPUM_R3_MSR_READ:
1217	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMSRRead);
1218	break;
1219	case VINF_CPUM_R3_MSR_WRITE:
1220	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMSRWrite);
1221	break;
1222	case VINF_EM_RAW_EMULATE_INSTR:
1223	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetEmulate);
1224	break;
1225	case VINF_PATCH_EMULATE_INSTR:
1226	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchEmulate);
1227	break;
1228	case VINF_EM_RAW_EMULATE_INSTR_LDT_FAULT:
1229	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetLDTFault);
1230	break;
1231	case VINF_EM_RAW_EMULATE_INSTR_GDT_FAULT:
1232	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetGDTFault);
1233	break;
1234	case VINF_EM_RAW_EMULATE_INSTR_IDT_FAULT:
1235	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIDTFault);
1236	break;
1237	case VINF_EM_RAW_EMULATE_INSTR_TSS_FAULT:
1238	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetTSSFault);
1239	break;
1240	case VINF_CSAM_PENDING_ACTION:
1241	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetCSAMTask);
1242	break;
1243	case VINF_PGM_SYNC_CR3:
1244	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetSyncCR3);
1245	break;
1246	case VINF_PATM_PATCH_INT3:
1247	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchInt3);
1248	break;
1249	case VINF_PATM_PATCH_TRAP_PF:
1250	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchPF);
1251	break;
1252	case VINF_PATM_PATCH_TRAP_GP:
1253	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchGP);
1254	break;
1255	case VINF_PATM_PENDING_IRQ_AFTER_IRET:
1256	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchIretIRQ);
1257	break;
1258	case VINF_EM_RESCHEDULE_REM:
1259	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetRescheduleREM);
1260	break;
1261	case VINF_EM_RAW_TO_R3:
1262	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Total);
1263	if (VM_FF_IS_SET(pVM, VM_FF_TM_VIRTUAL_SYNC))
1264	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3TMVirt);
1265	else if (VM_FF_IS_SET(pVM, VM_FF_PGM_NEED_HANDY_PAGES))
1266	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3HandyPages);
1267	else if (VM_FF_IS_SET(pVM, VM_FF_PDM_QUEUES))
1268	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3PDMQueues);
1269	else if (VM_FF_IS_SET(pVM, VM_FF_EMT_RENDEZVOUS))
1270	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Rendezvous);
1271	else if (VM_FF_IS_SET(pVM, VM_FF_PDM_DMA))
1272	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3DMA);
1273	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TIMER))
1274	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Timer);
1275	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PDM_CRITSECT))
1276	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3CritSect);
1277	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TO_R3))
1278	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3FF);
1279	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_IEM))
1280	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Iem);
1281	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_IOM))
1282	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Iom);
1283	else
1284	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Unknown);
1285	break;
1286
1287	case VINF_EM_RAW_TIMER_PENDING:
1288	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetTimerPending);
1289	break;
1290	case VINF_EM_RAW_INTERRUPT_PENDING:
1291	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetInterruptPending);
1292	break;
1293	case VINF_VMM_CALL_HOST:
1294	switch (pVCpu->vmm.s.enmCallRing3Operation)
1295	{
1296	case VMMCALLRING3_PGM_ALLOCATE_HANDY_PAGES:
1297	STAM_COUNTER_INC(&pVM->vmm.s.StatRZCallPGMAllocHandy);
1298	break;
1299	case VMMCALLRING3_VM_R0_ASSERTION:
1300	default:
1301	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetCallRing3);
1302	break;
1303	}
1304	break;
1305	case VINF_PATM_DUPLICATE_FUNCTION:
1306	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPATMDuplicateFn);
1307	break;
1308	case VINF_PGM_CHANGE_MODE:
1309	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPGMChangeMode);
1310	break;
1311	case VINF_PGM_POOL_FLUSH_PENDING:
1312	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPGMFlushPending);
1313	break;
1314	case VINF_EM_PENDING_REQUEST:
1315	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPendingRequest);
1316	break;
1317	case VINF_EM_HM_PATCH_TPR_INSTR:
1318	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchTPR);
1319	break;
1320	default:
1321	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMisc);
1322	break;
1323	}
1324	}
1325	#endif /* VBOX_WITH_STATISTICS */
1326
1327
1328	/**
1329	* The Ring 0 entry point, called by the fast-ioctl path.
1330	*
1331	* @param pGVM The global (ring-0) VM structure.
1332	* @param pVMIgnored The cross context VM structure. The return code is
1333	* stored in pVM->vmm.s.iLastGZRc.
1334	* @param idCpu The Virtual CPU ID of the calling EMT.
1335	* @param enmOperation Which operation to execute.
1336	* @remarks Assume called with interrupts _enabled_.
1337	*/
1338	VMMR0DECL(void) VMMR0EntryFast(PGVM pGVM, PVMCC pVMIgnored, VMCPUID idCpu, VMMR0OPERATION enmOperation)
1339	{
1340	RT_NOREF(pVMIgnored);
1341
1342	/*
1343	* Validation.
1344	*/
1345	if ( idCpu < pGVM->cCpus
1346	&& pGVM->cCpus == pGVM->cCpusUnsafe)
1347	{ /likely/ }
1348	else
1349	{
1350	SUPR0Printf("VMMR0EntryFast: Bad idCpu=%#x cCpus=%#x cCpusUnsafe=%#x\n", idCpu, pGVM->cCpus, pGVM->cCpusUnsafe);
1351	return;
1352	}
1353
1354	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
1355	RTNATIVETHREAD const hNativeThread = RTThreadNativeSelf();
1356	if (RT_LIKELY( pGVCpu->hEMT == hNativeThread
1357	&& pGVCpu->hNativeThreadR0 == hNativeThread))
1358	{ /* likely */ }
1359	else
1360	{
1361	SUPR0Printf("VMMR0EntryFast: Bad thread idCpu=%#x hNativeSelf=%p pGVCpu->hEmt=%p pGVCpu->hNativeThreadR0=%p\n",
1362	idCpu, hNativeThread, pGVCpu->hEMT, pGVCpu->hNativeThreadR0);
1363	return;
1364	}
1365
1366	/*
1367	* Perform requested operation.
1368	*/
1369	switch (enmOperation)
1370	{
1371	/*
1372	* Run guest code using the available hardware acceleration technology.
1373	*/
1374	case VMMR0_DO_HM_RUN:
1375	{
1376	for (;;) /* hlt loop */
1377	{
1378	/*
1379	* Disable ring-3 calls & blocking till we've successfully entered HM.
1380	* Otherwise we sometimes end up blocking at the finall Log4 statement
1381	* in VMXR0Enter, while still in a somewhat inbetween state.
1382	*/
1383	VMMRZCallRing3Disable(pGVCpu);
1384
1385	/*
1386	* Disable preemption.
1387	*/
1388	Assert(!vmmR0ThreadCtxHookIsEnabled(pGVCpu));
1389	RTTHREADPREEMPTSTATE PreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
1390	RTThreadPreemptDisable(&PreemptState);
1391	pGVCpu->vmmr0.s.pPreemptState = &PreemptState;
1392
1393	/*
1394	* Get the host CPU identifiers, make sure they are valid and that
1395	* we've got a TSC delta for the CPU.
1396	*/
1397	RTCPUID idHostCpu;
1398	uint32_t iHostCpuSet = RTMpCurSetIndexAndId(&idHostCpu);
1399	if (RT_LIKELY( iHostCpuSet < RTCPUSET_MAX_CPUS
1400	&& SUPIsTscDeltaAvailableForCpuSetIndex(iHostCpuSet)))
1401	{
1402	pGVCpu->iHostCpuSet = iHostCpuSet;
1403	ASMAtomicWriteU32(&pGVCpu->idHostCpu, idHostCpu);
1404
1405	/*
1406	* Update the periodic preemption timer if it's active.
1407	*/
1408	if (pGVM->vmm.s.fUsePeriodicPreemptionTimers)
1409	GVMMR0SchedUpdatePeriodicPreemptionTimer(pGVM, pGVCpu->idHostCpu, TMCalcHostTimerFrequency(pGVM, pGVCpu));
1410
1411	#ifdef VMM_R0_TOUCH_FPU
1412	/*
1413	* Make sure we've got the FPU state loaded so and we don't need to clear
1414	* CR0.TS and get out of sync with the host kernel when loading the guest
1415	* FPU state. @ref sec_cpum_fpu (CPUM.cpp) and @bugref{4053}.
1416	*/
1417	CPUMR0TouchHostFpu();
1418	#endif
1419	int rc;
1420	bool fPreemptRestored = false;
1421	if (!HMR0SuspendPending())
1422	{
1423	/*
1424	* Enable the context switching hook.
1425	*/
1426	if (pGVCpu->vmmr0.s.hCtxHook != NIL_RTTHREADCTXHOOK)
1427	{
1428	Assert(!RTThreadCtxHookIsEnabled(pGVCpu->vmmr0.s.hCtxHook));
1429	int rc2 = RTThreadCtxHookEnable(pGVCpu->vmmr0.s.hCtxHook); AssertRC(rc2);
1430	}
1431
1432	/*
1433	* Enter HM context.
1434	*/
1435	rc = HMR0Enter(pGVCpu);
1436	if (RT_SUCCESS(rc))
1437	{
1438	VMCPU_SET_STATE(pGVCpu, VMCPUSTATE_STARTED_HM);
1439
1440	/*
1441	* When preemption hooks are in place, enable preemption now that
1442	* we're in HM context.
1443	*/
1444	if (vmmR0ThreadCtxHookIsEnabled(pGVCpu))
1445	{
1446	fPreemptRestored = true;
1447	pGVCpu->vmmr0.s.pPreemptState = NULL;
1448	RTThreadPreemptRestore(&PreemptState);
1449	}
1450	VMMRZCallRing3Enable(pGVCpu);
1451
1452	/*
1453	* Setup the longjmp machinery and execute guest code (calls HMR0RunGuestCode).
1454	*/
1455	rc = vmmR0CallRing3SetJmp(&pGVCpu->vmm.s.CallRing3JmpBufR0, HMR0RunGuestCode, pGVM, pGVCpu);
1456
1457	/*
1458	* Assert sanity on the way out. Using manual assertions code here as normal
1459	* assertions are going to panic the host since we're outside the setjmp/longjmp zone.
1460	*/
1461	if (RT_UNLIKELY( VMCPU_GET_STATE(pGVCpu) != VMCPUSTATE_STARTED_HM
1462	&& RT_SUCCESS_NP(rc)
1463	&& rc != VINF_VMM_CALL_HOST ))
1464	{
1465	pGVM->vmm.s.szRing0AssertMsg1[0] = '\0';
1466	RTStrPrintf(pGVM->vmm.s.szRing0AssertMsg2, sizeof(pGVM->vmm.s.szRing0AssertMsg2),
1467	"Got VMCPU state %d expected %d.\n", VMCPU_GET_STATE(pGVCpu), VMCPUSTATE_STARTED_HM);
1468	rc = VERR_VMM_WRONG_HM_VMCPU_STATE;
1469	}
1470	#if 0
1471	/** @todo Get rid of this. HM shouldn't disable the context hook. */
1472	else if (RT_UNLIKELY(vmmR0ThreadCtxHookIsEnabled(pGVCpu)))
1473	{
1474	pGVM->vmm.s.szRing0AssertMsg1[0] = '\0';
1475	RTStrPrintf(pGVM->vmm.s.szRing0AssertMsg2, sizeof(pGVM->vmm.s.szRing0AssertMsg2),
1476	"Thread-context hooks still enabled! VCPU=%p Id=%u rc=%d.\n", pGVCpu, pGVCpu->idCpu, rc);
1477	rc = VERR_VMM_CONTEXT_HOOK_STILL_ENABLED;
1478	}
1479	#endif
1480
1481	VMMRZCallRing3Disable(pGVCpu); /* Lazy bird: Simpler just disabling it again... */
1482	VMCPU_SET_STATE(pGVCpu, VMCPUSTATE_STARTED);
1483	}
1484	STAM_COUNTER_INC(&pGVM->vmm.s.StatRunGC);
1485
1486	/*
1487	* Invalidate the host CPU identifiers before we disable the context
1488	* hook / restore preemption.
1489	*/
1490	pGVCpu->iHostCpuSet = UINT32_MAX;
1491	ASMAtomicWriteU32(&pGVCpu->idHostCpu, NIL_RTCPUID);
1492
1493	/*
1494	* Disable context hooks. Due to unresolved cleanup issues, we
1495	* cannot leave the hooks enabled when we return to ring-3.
1496	*
1497	* Note! At the moment HM may also have disabled the hook
1498	* when we get here, but the IPRT API handles that.
1499	*/
1500	if (pGVCpu->vmmr0.s.hCtxHook != NIL_RTTHREADCTXHOOK)
1501	RTThreadCtxHookDisable(pGVCpu->vmmr0.s.hCtxHook);
1502	}
1503	/*
1504	* The system is about to go into suspend mode; go back to ring 3.
1505	*/
1506	else
1507	{
1508	pGVCpu->iHostCpuSet = UINT32_MAX;
1509	ASMAtomicWriteU32(&pGVCpu->idHostCpu, NIL_RTCPUID);
1510	rc = VINF_EM_RAW_INTERRUPT;
1511	}
1512
1513	/** @todo When HM stops messing with the context hook state, we'll disable
1514	* preemption again before the RTThreadCtxHookDisable call. */
1515	if (!fPreemptRestored)
1516	{
1517	pGVCpu->vmmr0.s.pPreemptState = NULL;
1518	RTThreadPreemptRestore(&PreemptState);
1519	}
1520
1521	pGVCpu->vmm.s.iLastGZRc = rc;
1522
1523	/* Fire dtrace probe and collect statistics. */
1524	VBOXVMM_R0_VMM_RETURN_TO_RING3_HM(pGVCpu, CPUMQueryGuestCtxPtr(pGVCpu), rc);
1525	#ifdef VBOX_WITH_STATISTICS
1526	vmmR0RecordRC(pGVM, pGVCpu, rc);
1527	#endif
1528	VMMRZCallRing3Enable(pGVCpu);
1529
1530	/*
1531	* If this is a halt.
1532	*/
1533	if (rc != VINF_EM_HALT)
1534	{ /* we're not in a hurry for a HLT, so prefer this path */ }
1535	else
1536	{
1537	pGVCpu->vmm.s.iLastGZRc = rc = vmmR0DoHalt(pGVM, pGVCpu);
1538	if (rc == VINF_SUCCESS)
1539	{
1540	pGVCpu->vmm.s.cR0HaltsSucceeded++;
1541	continue;
1542	}
1543	pGVCpu->vmm.s.cR0HaltsToRing3++;
1544	}
1545	}
1546	/*
1547	* Invalid CPU set index or TSC delta in need of measuring.
1548	*/
1549	else
1550	{
1551	pGVCpu->vmmr0.s.pPreemptState = NULL;
1552	pGVCpu->iHostCpuSet = UINT32_MAX;
1553	ASMAtomicWriteU32(&pGVCpu->idHostCpu, NIL_RTCPUID);
1554	RTThreadPreemptRestore(&PreemptState);
1555
1556	VMMRZCallRing3Enable(pGVCpu);
1557
1558	if (iHostCpuSet < RTCPUSET_MAX_CPUS)
1559	{
1560	int rc = SUPR0TscDeltaMeasureBySetIndex(pGVM->pSession, iHostCpuSet, 0 /fFlags/,
1561	2 /cMsWaitRetry/, 5RT_MS_1SEC /cMsWaitThread*/,
1562	0 /default cTries/);
1563	if (RT_SUCCESS(rc) \|\| rc == VERR_CPU_OFFLINE)
1564	pGVCpu->vmm.s.iLastGZRc = VINF_EM_RAW_TO_R3;
1565	else
1566	pGVCpu->vmm.s.iLastGZRc = rc;
1567	}
1568	else
1569	pGVCpu->vmm.s.iLastGZRc = VERR_INVALID_CPU_INDEX;
1570	}
1571	break;
1572	} /* halt loop. */
1573	break;
1574	}
1575
1576	#ifdef VBOX_WITH_NEM_R0
1577	# if defined(RT_ARCH_AMD64) && defined(RT_OS_WINDOWS)
1578	case VMMR0_DO_NEM_RUN:
1579	{
1580	/*
1581	* Setup the longjmp machinery and execute guest code (calls NEMR0RunGuestCode).
1582	*/
1583	# ifdef VBOXSTRICTRC_STRICT_ENABLED
1584	int rc = vmmR0CallRing3SetJmp2(&pGVCpu->vmm.s.CallRing3JmpBufR0, (PFNVMMR0SETJMP2)NEMR0RunGuestCode, pGVM, idCpu);
1585	# else
1586	int rc = vmmR0CallRing3SetJmp2(&pGVCpu->vmm.s.CallRing3JmpBufR0, NEMR0RunGuestCode, pGVM, idCpu);
1587	# endif
1588	STAM_COUNTER_INC(&pGVM->vmm.s.StatRunGC);
1589
1590	pGVCpu->vmm.s.iLastGZRc = rc;
1591
1592	/*
1593	* Fire dtrace probe and collect statistics.
1594	*/
1595	VBOXVMM_R0_VMM_RETURN_TO_RING3_NEM(pGVCpu, CPUMQueryGuestCtxPtr(pGVCpu), rc);
1596	# ifdef VBOX_WITH_STATISTICS
1597	vmmR0RecordRC(pGVM, pGVCpu, rc);
1598	# endif
1599	break;
1600	}
1601	# endif
1602	#endif
1603
1604	/*
1605	* For profiling.
1606	*/
1607	case VMMR0_DO_NOP:
1608	pGVCpu->vmm.s.iLastGZRc = VINF_SUCCESS;
1609	break;
1610
1611	/*
1612	* Shouldn't happen.
1613	*/
1614	default:
1615	AssertMsgFailed(("%#x\n", enmOperation));
1616	pGVCpu->vmm.s.iLastGZRc = VERR_NOT_SUPPORTED;
1617	break;
1618	}
1619	}
1620
1621
1622	/**
1623	* Validates a session or VM session argument.
1624	*
1625	* @returns true / false accordingly.
1626	* @param pGVM The global (ring-0) VM structure.
1627	* @param pClaimedSession The session claim to validate.
1628	* @param pSession The session argument.
1629	*/
1630	DECLINLINE(bool) vmmR0IsValidSession(PGVM pGVM, PSUPDRVSESSION pClaimedSession, PSUPDRVSESSION pSession)
1631	{
1632	/* This must be set! */
1633	if (!pSession)
1634	return false;
1635
1636	/* Only one out of the two. */
1637	if (pGVM && pClaimedSession)
1638	return false;
1639	if (pGVM)
1640	pClaimedSession = pGVM->pSession;
1641	return pClaimedSession == pSession;
1642	}
1643
1644
1645	/**
1646	* VMMR0EntryEx worker function, either called directly or when ever possible
1647	* called thru a longjmp so we can exit safely on failure.
1648	*
1649	* @returns VBox status code.
1650	* @param pGVM The global (ring-0) VM structure.
1651	* @param idCpu Virtual CPU ID argument. Must be NIL_VMCPUID if pVM
1652	* is NIL_RTR0PTR, and may be NIL_VMCPUID if it isn't
1653	* @param enmOperation Which operation to execute.
1654	* @param pReqHdr This points to a SUPVMMR0REQHDR packet. Optional.
1655	* The support driver validates this if it's present.
1656	* @param u64Arg Some simple constant argument.
1657	* @param pSession The session of the caller.
1658	*
1659	* @remarks Assume called with interrupts _enabled_.
1660	*/
1661	DECL_NO_INLINE(static, int) vmmR0EntryExWorker(PGVM pGVM, VMCPUID idCpu, VMMR0OPERATION enmOperation,
1662	PSUPVMMR0REQHDR pReqHdr, uint64_t u64Arg, PSUPDRVSESSION pSession)
1663	{
1664	/*
1665	* Validate pGVM and idCpu for consistency and validity.
1666	*/
1667	if (pGVM != NULL)
1668	{
1669	if (RT_LIKELY(((uintptr_t)pGVM & PAGE_OFFSET_MASK) == 0))
1670	{ /* likely */ }
1671	else
1672	{
1673	SUPR0Printf("vmmR0EntryExWorker: Invalid pGVM=%p! (op=%d)\n", pGVM, enmOperation);
1674	return VERR_INVALID_POINTER;
1675	}
1676
1677	if (RT_LIKELY(idCpu == NIL_VMCPUID \|\| idCpu < pGVM->cCpus))
1678	{ /* likely */ }
1679	else
1680	{
1681	SUPR0Printf("vmmR0EntryExWorker: Invalid idCpu %#x (cCpus=%#x)\n", idCpu, pGVM->cCpus);
1682	return VERR_INVALID_PARAMETER;
1683	}
1684
1685	if (RT_LIKELY( pGVM->enmVMState >= VMSTATE_CREATING
1686	&& pGVM->enmVMState <= VMSTATE_TERMINATED
1687	&& pGVM->pSession == pSession
1688	&& pGVM->pSelf == pGVM))
1689	{ /* likely */ }
1690	else
1691	{
1692	SUPR0Printf("vmmR0EntryExWorker: Invalid pGVM=%p:{.enmVMState=%d, .cCpus=%#x, .pSession=%p(==%p), .pSelf=%p(==%p)}! (op=%d)\n",
1693	pGVM, pGVM->enmVMState, pGVM->cCpus, pGVM->pSession, pSession, pGVM->pSelf, pGVM, enmOperation);
1694	return VERR_INVALID_POINTER;
1695	}
1696	}
1697	else if (RT_LIKELY(idCpu == NIL_VMCPUID))
1698	{ /* likely */ }
1699	else
1700	{
1701	SUPR0Printf("vmmR0EntryExWorker: Invalid idCpu=%u\n", idCpu);
1702	return VERR_INVALID_PARAMETER;
1703	}
1704
1705	/*
1706	* Process the request.
1707	*/
1708	int rc;
1709	switch (enmOperation)
1710	{
1711	/*
1712	* GVM requests
1713	*/
1714	case VMMR0_DO_GVMM_CREATE_VM:
1715	if (pGVM == NULL && u64Arg == 0 && idCpu == NIL_VMCPUID)
1716	rc = GVMMR0CreateVMReq((PGVMMCREATEVMREQ)pReqHdr, pSession);
1717	else
1718	rc = VERR_INVALID_PARAMETER;
1719	break;
1720
1721	case VMMR0_DO_GVMM_DESTROY_VM:
1722	if (pReqHdr == NULL && u64Arg == 0)
1723	rc = GVMMR0DestroyVM(pGVM);
1724	else
1725	rc = VERR_INVALID_PARAMETER;
1726	break;
1727
1728	case VMMR0_DO_GVMM_REGISTER_VMCPU:
1729	if (pGVM != NULL)
1730	rc = GVMMR0RegisterVCpu(pGVM, idCpu);
1731	else
1732	rc = VERR_INVALID_PARAMETER;
1733	break;
1734
1735	case VMMR0_DO_GVMM_DEREGISTER_VMCPU:
1736	if (pGVM != NULL)
1737	rc = GVMMR0DeregisterVCpu(pGVM, idCpu);
1738	else
1739	rc = VERR_INVALID_PARAMETER;
1740	break;
1741
1742	case VMMR0_DO_GVMM_REGISTER_WORKER_THREAD:
1743	if (pGVM != NULL && pReqHdr && pReqHdr->cbReq == sizeof(GVMMREGISTERWORKERTHREADREQ))
1744	rc = GVMMR0RegisterWorkerThread(pGVM, (GVMMWORKERTHREAD)(unsigned)u64Arg,
1745	((PGVMMREGISTERWORKERTHREADREQ)(pReqHdr))->hNativeThreadR3);
1746	else
1747	rc = VERR_INVALID_PARAMETER;
1748	break;
1749
1750	case VMMR0_DO_GVMM_DEREGISTER_WORKER_THREAD:
1751	if (pGVM != NULL)
1752	rc = GVMMR0DeregisterWorkerThread(pGVM, (GVMMWORKERTHREAD)(unsigned)u64Arg);
1753	else
1754	rc = VERR_INVALID_PARAMETER;
1755	break;
1756
1757	case VMMR0_DO_GVMM_SCHED_HALT:
1758	if (pReqHdr)
1759	return VERR_INVALID_PARAMETER;
1760	rc = GVMMR0SchedHaltReq(pGVM, idCpu, u64Arg);
1761	break;
1762
1763	case VMMR0_DO_GVMM_SCHED_WAKE_UP:
1764	if (pReqHdr \|\| u64Arg)
1765	return VERR_INVALID_PARAMETER;
1766	rc = GVMMR0SchedWakeUp(pGVM, idCpu);
1767	break;
1768
1769	case VMMR0_DO_GVMM_SCHED_POKE:
1770	if (pReqHdr \|\| u64Arg)
1771	return VERR_INVALID_PARAMETER;
1772	rc = GVMMR0SchedPoke(pGVM, idCpu);
1773	break;
1774
1775	case VMMR0_DO_GVMM_SCHED_WAKE_UP_AND_POKE_CPUS:
1776	if (u64Arg)
1777	return VERR_INVALID_PARAMETER;
1778	rc = GVMMR0SchedWakeUpAndPokeCpusReq(pGVM, (PGVMMSCHEDWAKEUPANDPOKECPUSREQ)pReqHdr);
1779	break;
1780
1781	case VMMR0_DO_GVMM_SCHED_POLL:
1782	if (pReqHdr \|\| u64Arg > 1)
1783	return VERR_INVALID_PARAMETER;
1784	rc = GVMMR0SchedPoll(pGVM, idCpu, !!u64Arg);
1785	break;
1786
1787	case VMMR0_DO_GVMM_QUERY_STATISTICS:
1788	if (u64Arg)
1789	return VERR_INVALID_PARAMETER;
1790	rc = GVMMR0QueryStatisticsReq(pGVM, (PGVMMQUERYSTATISTICSSREQ)pReqHdr, pSession);
1791	break;
1792
1793	case VMMR0_DO_GVMM_RESET_STATISTICS:
1794	if (u64Arg)
1795	return VERR_INVALID_PARAMETER;
1796	rc = GVMMR0ResetStatisticsReq(pGVM, (PGVMMRESETSTATISTICSSREQ)pReqHdr, pSession);
1797	break;
1798
1799	/*
1800	* Initialize the R0 part of a VM instance.
1801	*/
1802	case VMMR0_DO_VMMR0_INIT:
1803	rc = vmmR0InitVM(pGVM, RT_LODWORD(u64Arg), RT_HIDWORD(u64Arg));
1804	break;
1805
1806	/*
1807	* Does EMT specific ring-0 init.
1808	*/
1809	case VMMR0_DO_VMMR0_INIT_EMT:
1810	rc = vmmR0InitVMEmt(pGVM, idCpu);
1811	break;
1812
1813	/*
1814	* Terminate the R0 part of a VM instance.
1815	*/
1816	case VMMR0_DO_VMMR0_TERM:
1817	rc = VMMR0TermVM(pGVM, 0 /idCpu/);
1818	break;
1819
1820	/*
1821	* Update release or debug logger instances.
1822	*/
1823	case VMMR0_DO_VMMR0_UPDATE_LOGGERS:
1824	if (idCpu == NIL_VMCPUID)
1825	return VERR_INVALID_CPU_ID;
1826	if (u64Arg < VMMLOGGER_IDX_MAX && pReqHdr != NULL)
1827	rc = vmmR0UpdateLoggers(pGVM, idCpu /idCpu/, (PVMMR0UPDATELOGGERSREQ)pReqHdr, (size_t)u64Arg);
1828	else
1829	return VERR_INVALID_PARAMETER;
1830	break;
1831
1832	/*
1833	* Log flusher thread.
1834	*/
1835	case VMMR0_DO_VMMR0_LOG_FLUSHER:
1836	if (idCpu != NIL_VMCPUID)
1837	return VERR_INVALID_CPU_ID;
1838	if (pReqHdr == NULL)
1839	rc = vmmR0LogFlusher(pGVM);
1840	else
1841	return VERR_INVALID_PARAMETER;
1842	break;
1843
1844	/*
1845	* Wait for the flush to finish with all the buffers for the given logger.
1846	*/
1847	case VMMR0_DO_VMMR0_LOG_WAIT_FLUSHED:
1848	if (idCpu == NIL_VMCPUID)
1849	return VERR_INVALID_CPU_ID;
1850	if (u64Arg < VMMLOGGER_IDX_MAX && pReqHdr == NULL)
1851	rc = vmmR0LogWaitFlushed(pGVM, idCpu /idCpu/, (size_t)u64Arg);
1852	else
1853	return VERR_INVALID_PARAMETER;
1854	break;
1855
1856	/*
1857	* Attempt to enable hm mode and check the current setting.
1858	*/
1859	case VMMR0_DO_HM_ENABLE:
1860	rc = HMR0EnableAllCpus(pGVM);
1861	break;
1862
1863	/*
1864	* Setup the hardware accelerated session.
1865	*/
1866	case VMMR0_DO_HM_SETUP_VM:
1867	rc = HMR0SetupVM(pGVM);
1868	break;
1869
1870	/*
1871	* PGM wrappers.
1872	*/
1873	case VMMR0_DO_PGM_ALLOCATE_HANDY_PAGES:
1874	if (idCpu == NIL_VMCPUID)
1875	return VERR_INVALID_CPU_ID;
1876	rc = PGMR0PhysAllocateHandyPages(pGVM, idCpu);
1877	break;
1878
1879	case VMMR0_DO_PGM_FLUSH_HANDY_PAGES:
1880	if (idCpu == NIL_VMCPUID)
1881	return VERR_INVALID_CPU_ID;
1882	rc = PGMR0PhysFlushHandyPages(pGVM, idCpu);
1883	break;
1884
1885	case VMMR0_DO_PGM_ALLOCATE_LARGE_PAGE:
1886	if (idCpu == NIL_VMCPUID)
1887	return VERR_INVALID_CPU_ID;
1888	rc = PGMR0PhysAllocateLargePage(pGVM, idCpu, u64Arg);
1889	break;
1890
1891	case VMMR0_DO_PGM_PHYS_SETUP_IOMMU:
1892	if (idCpu != 0)
1893	return VERR_INVALID_CPU_ID;
1894	rc = PGMR0PhysSetupIoMmu(pGVM);
1895	break;
1896
1897	case VMMR0_DO_PGM_POOL_GROW:
1898	if (idCpu == NIL_VMCPUID)
1899	return VERR_INVALID_CPU_ID;
1900	rc = PGMR0PoolGrow(pGVM, idCpu);
1901	break;
1902
1903	/*
1904	* GMM wrappers.
1905	*/
1906	case VMMR0_DO_GMM_INITIAL_RESERVATION:
1907	if (u64Arg)
1908	return VERR_INVALID_PARAMETER;
1909	rc = GMMR0InitialReservationReq(pGVM, idCpu, (PGMMINITIALRESERVATIONREQ)pReqHdr);
1910	break;
1911
1912	case VMMR0_DO_GMM_UPDATE_RESERVATION:
1913	if (u64Arg)
1914	return VERR_INVALID_PARAMETER;
1915	rc = GMMR0UpdateReservationReq(pGVM, idCpu, (PGMMUPDATERESERVATIONREQ)pReqHdr);
1916	break;
1917
1918	case VMMR0_DO_GMM_ALLOCATE_PAGES:
1919	if (u64Arg)
1920	return VERR_INVALID_PARAMETER;
1921	rc = GMMR0AllocatePagesReq(pGVM, idCpu, (PGMMALLOCATEPAGESREQ)pReqHdr);
1922	break;
1923
1924	case VMMR0_DO_GMM_FREE_PAGES:
1925	if (u64Arg)
1926	return VERR_INVALID_PARAMETER;
1927	rc = GMMR0FreePagesReq(pGVM, idCpu, (PGMMFREEPAGESREQ)pReqHdr);
1928	break;
1929
1930	case VMMR0_DO_GMM_FREE_LARGE_PAGE:
1931	if (u64Arg)
1932	return VERR_INVALID_PARAMETER;
1933	rc = GMMR0FreeLargePageReq(pGVM, idCpu, (PGMMFREELARGEPAGEREQ)pReqHdr);
1934	break;
1935
1936	case VMMR0_DO_GMM_QUERY_HYPERVISOR_MEM_STATS:
1937	if (u64Arg)
1938	return VERR_INVALID_PARAMETER;
1939	rc = GMMR0QueryHypervisorMemoryStatsReq((PGMMMEMSTATSREQ)pReqHdr);
1940	break;
1941
1942	case VMMR0_DO_GMM_QUERY_MEM_STATS:
1943	if (idCpu == NIL_VMCPUID)
1944	return VERR_INVALID_CPU_ID;
1945	if (u64Arg)
1946	return VERR_INVALID_PARAMETER;
1947	rc = GMMR0QueryMemoryStatsReq(pGVM, idCpu, (PGMMMEMSTATSREQ)pReqHdr);
1948	break;
1949
1950	case VMMR0_DO_GMM_BALLOONED_PAGES:
1951	if (u64Arg)
1952	return VERR_INVALID_PARAMETER;
1953	rc = GMMR0BalloonedPagesReq(pGVM, idCpu, (PGMMBALLOONEDPAGESREQ)pReqHdr);
1954	break;
1955
1956	case VMMR0_DO_GMM_MAP_UNMAP_CHUNK:
1957	if (u64Arg)
1958	return VERR_INVALID_PARAMETER;
1959	rc = GMMR0MapUnmapChunkReq(pGVM, (PGMMMAPUNMAPCHUNKREQ)pReqHdr);
1960	break;
1961
1962	case VMMR0_DO_GMM_REGISTER_SHARED_MODULE:
1963	if (idCpu == NIL_VMCPUID)
1964	return VERR_INVALID_CPU_ID;
1965	if (u64Arg)
1966	return VERR_INVALID_PARAMETER;
1967	rc = GMMR0RegisterSharedModuleReq(pGVM, idCpu, (PGMMREGISTERSHAREDMODULEREQ)pReqHdr);
1968	break;
1969
1970	case VMMR0_DO_GMM_UNREGISTER_SHARED_MODULE:
1971	if (idCpu == NIL_VMCPUID)
1972	return VERR_INVALID_CPU_ID;
1973	if (u64Arg)
1974	return VERR_INVALID_PARAMETER;
1975	rc = GMMR0UnregisterSharedModuleReq(pGVM, idCpu, (PGMMUNREGISTERSHAREDMODULEREQ)pReqHdr);
1976	break;
1977
1978	case VMMR0_DO_GMM_RESET_SHARED_MODULES:
1979	if (idCpu == NIL_VMCPUID)
1980	return VERR_INVALID_CPU_ID;
1981	if ( u64Arg
1982	\|\| pReqHdr)
1983	return VERR_INVALID_PARAMETER;
1984	rc = GMMR0ResetSharedModules(pGVM, idCpu);
1985	break;
1986
1987	#ifdef VBOX_WITH_PAGE_SHARING
1988	case VMMR0_DO_GMM_CHECK_SHARED_MODULES:
1989	{
1990	if (idCpu == NIL_VMCPUID)
1991	return VERR_INVALID_CPU_ID;
1992	if ( u64Arg
1993	\|\| pReqHdr)
1994	return VERR_INVALID_PARAMETER;
1995	rc = GMMR0CheckSharedModules(pGVM, idCpu);
1996	break;
1997	}
1998	#endif
1999
2000	#if defined(VBOX_STRICT) && HC_ARCH_BITS == 64
2001	case VMMR0_DO_GMM_FIND_DUPLICATE_PAGE:
2002	if (u64Arg)
2003	return VERR_INVALID_PARAMETER;
2004	rc = GMMR0FindDuplicatePageReq(pGVM, (PGMMFINDDUPLICATEPAGEREQ)pReqHdr);
2005	break;
2006	#endif
2007
2008	case VMMR0_DO_GMM_QUERY_STATISTICS:
2009	if (u64Arg)
2010	return VERR_INVALID_PARAMETER;
2011	rc = GMMR0QueryStatisticsReq(pGVM, (PGMMQUERYSTATISTICSSREQ)pReqHdr);
2012	break;
2013
2014	case VMMR0_DO_GMM_RESET_STATISTICS:
2015	if (u64Arg)
2016	return VERR_INVALID_PARAMETER;
2017	rc = GMMR0ResetStatisticsReq(pGVM, (PGMMRESETSTATISTICSSREQ)pReqHdr);
2018	break;
2019
2020	/*
2021	* A quick GCFGM mock-up.
2022	*/
2023	/** @todo GCFGM with proper access control, ring-3 management interface and all that. */
2024	case VMMR0_DO_GCFGM_SET_VALUE:
2025	case VMMR0_DO_GCFGM_QUERY_VALUE:
2026	{
2027	if (pGVM \|\| !pReqHdr \|\| u64Arg \|\| idCpu != NIL_VMCPUID)
2028	return VERR_INVALID_PARAMETER;
2029	PGCFGMVALUEREQ pReq = (PGCFGMVALUEREQ)pReqHdr;
2030	if (pReq->Hdr.cbReq != sizeof(*pReq))
2031	return VERR_INVALID_PARAMETER;
2032	if (enmOperation == VMMR0_DO_GCFGM_SET_VALUE)
2033	{
2034	rc = GVMMR0SetConfig(pReq->pSession, &pReq->szName[0], pReq->u64Value);
2035	//if (rc == VERR_CFGM_VALUE_NOT_FOUND)
2036	// rc = GMMR0SetConfig(pReq->pSession, &pReq->szName[0], pReq->u64Value);
2037	}
2038	else
2039	{
2040	rc = GVMMR0QueryConfig(pReq->pSession, &pReq->szName[0], &pReq->u64Value);
2041	//if (rc == VERR_CFGM_VALUE_NOT_FOUND)
2042	// rc = GMMR0QueryConfig(pReq->pSession, &pReq->szName[0], &pReq->u64Value);
2043	}
2044	break;
2045	}
2046
2047	/*
2048	* PDM Wrappers.
2049	*/
2050	case VMMR0_DO_PDM_DRIVER_CALL_REQ_HANDLER:
2051	{
2052	if (!pReqHdr \|\| u64Arg \|\| idCpu != NIL_VMCPUID)
2053	return VERR_INVALID_PARAMETER;
2054	rc = PDMR0DriverCallReqHandler(pGVM, (PPDMDRIVERCALLREQHANDLERREQ)pReqHdr);
2055	break;
2056	}
2057
2058	case VMMR0_DO_PDM_DEVICE_CREATE:
2059	{
2060	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2061	return VERR_INVALID_PARAMETER;
2062	rc = PDMR0DeviceCreateReqHandler(pGVM, (PPDMDEVICECREATEREQ)pReqHdr);
2063	break;
2064	}
2065
2066	case VMMR0_DO_PDM_DEVICE_GEN_CALL:
2067	{
2068	if (!pReqHdr \|\| u64Arg)
2069	return VERR_INVALID_PARAMETER;
2070	rc = PDMR0DeviceGenCallReqHandler(pGVM, (PPDMDEVICEGENCALLREQ)pReqHdr, idCpu);
2071	break;
2072	}
2073
2074	/** @todo Remove the once all devices has been converted to new style! @bugref{9218} */
2075	case VMMR0_DO_PDM_DEVICE_COMPAT_SET_CRITSECT:
2076	{
2077	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2078	return VERR_INVALID_PARAMETER;
2079	rc = PDMR0DeviceCompatSetCritSectReqHandler(pGVM, (PPDMDEVICECOMPATSETCRITSECTREQ)pReqHdr);
2080	break;
2081	}
2082
2083	/*
2084	* Requests to the internal networking service.
2085	*/
2086	case VMMR0_DO_INTNET_OPEN:
2087	{
2088	PINTNETOPENREQ pReq = (PINTNETOPENREQ)pReqHdr;
2089	if (u64Arg \|\| !pReq \|\| !vmmR0IsValidSession(pGVM, pReq->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2090	return VERR_INVALID_PARAMETER;
2091	rc = IntNetR0OpenReq(pSession, pReq);
2092	break;
2093	}
2094
2095	case VMMR0_DO_INTNET_IF_CLOSE:
2096	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFCLOSEREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2097	return VERR_INVALID_PARAMETER;
2098	rc = IntNetR0IfCloseReq(pSession, (PINTNETIFCLOSEREQ)pReqHdr);
2099	break;
2100
2101
2102	case VMMR0_DO_INTNET_IF_GET_BUFFER_PTRS:
2103	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFGETBUFFERPTRSREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2104	return VERR_INVALID_PARAMETER;
2105	rc = IntNetR0IfGetBufferPtrsReq(pSession, (PINTNETIFGETBUFFERPTRSREQ)pReqHdr);
2106	break;
2107
2108	case VMMR0_DO_INTNET_IF_SET_PROMISCUOUS_MODE:
2109	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSETPROMISCUOUSMODEREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2110	return VERR_INVALID_PARAMETER;
2111	rc = IntNetR0IfSetPromiscuousModeReq(pSession, (PINTNETIFSETPROMISCUOUSMODEREQ)pReqHdr);
2112	break;
2113
2114	case VMMR0_DO_INTNET_IF_SET_MAC_ADDRESS:
2115	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSETMACADDRESSREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2116	return VERR_INVALID_PARAMETER;
2117	rc = IntNetR0IfSetMacAddressReq(pSession, (PINTNETIFSETMACADDRESSREQ)pReqHdr);
2118	break;
2119
2120	case VMMR0_DO_INTNET_IF_SET_ACTIVE:
2121	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSETACTIVEREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2122	return VERR_INVALID_PARAMETER;
2123	rc = IntNetR0IfSetActiveReq(pSession, (PINTNETIFSETACTIVEREQ)pReqHdr);
2124	break;
2125
2126	case VMMR0_DO_INTNET_IF_SEND:
2127	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSENDREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2128	return VERR_INVALID_PARAMETER;
2129	rc = IntNetR0IfSendReq(pSession, (PINTNETIFSENDREQ)pReqHdr);
2130	break;
2131
2132	case VMMR0_DO_INTNET_IF_WAIT:
2133	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFWAITREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2134	return VERR_INVALID_PARAMETER;
2135	rc = IntNetR0IfWaitReq(pSession, (PINTNETIFWAITREQ)pReqHdr);
2136	break;
2137
2138	case VMMR0_DO_INTNET_IF_ABORT_WAIT:
2139	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFWAITREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2140	return VERR_INVALID_PARAMETER;
2141	rc = IntNetR0IfAbortWaitReq(pSession, (PINTNETIFABORTWAITREQ)pReqHdr);
2142	break;
2143
2144	#if 0 //def VBOX_WITH_PCI_PASSTHROUGH
2145	/*
2146	* Requests to host PCI driver service.
2147	*/
2148	case VMMR0_DO_PCIRAW_REQ:
2149	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PPCIRAWSENDREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2150	return VERR_INVALID_PARAMETER;
2151	rc = PciRawR0ProcessReq(pGVM, pSession, (PPCIRAWSENDREQ)pReqHdr);
2152	break;
2153	#endif
2154
2155	/*
2156	* NEM requests.
2157	*/
2158	#ifdef VBOX_WITH_NEM_R0
2159	# if defined(RT_ARCH_AMD64) && defined(RT_OS_WINDOWS)
2160	case VMMR0_DO_NEM_INIT_VM:
2161	if (u64Arg \|\| pReqHdr \|\| idCpu != 0)
2162	return VERR_INVALID_PARAMETER;
2163	rc = NEMR0InitVM(pGVM);
2164	break;
2165
2166	case VMMR0_DO_NEM_INIT_VM_PART_2:
2167	if (u64Arg \|\| pReqHdr \|\| idCpu != 0)
2168	return VERR_INVALID_PARAMETER;
2169	rc = NEMR0InitVMPart2(pGVM);
2170	break;
2171
2172	case VMMR0_DO_NEM_MAP_PAGES:
2173	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2174	return VERR_INVALID_PARAMETER;
2175	rc = NEMR0MapPages(pGVM, idCpu);
2176	break;
2177
2178	case VMMR0_DO_NEM_UNMAP_PAGES:
2179	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2180	return VERR_INVALID_PARAMETER;
2181	rc = NEMR0UnmapPages(pGVM, idCpu);
2182	break;
2183
2184	case VMMR0_DO_NEM_EXPORT_STATE:
2185	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2186	return VERR_INVALID_PARAMETER;
2187	rc = NEMR0ExportState(pGVM, idCpu);
2188	break;
2189
2190	case VMMR0_DO_NEM_IMPORT_STATE:
2191	if (pReqHdr \|\| idCpu == NIL_VMCPUID)
2192	return VERR_INVALID_PARAMETER;
2193	rc = NEMR0ImportState(pGVM, idCpu, u64Arg);
2194	break;
2195
2196	case VMMR0_DO_NEM_QUERY_CPU_TICK:
2197	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2198	return VERR_INVALID_PARAMETER;
2199	rc = NEMR0QueryCpuTick(pGVM, idCpu);
2200	break;
2201
2202	case VMMR0_DO_NEM_RESUME_CPU_TICK_ON_ALL:
2203	if (pReqHdr \|\| idCpu == NIL_VMCPUID)
2204	return VERR_INVALID_PARAMETER;
2205	rc = NEMR0ResumeCpuTickOnAll(pGVM, idCpu, u64Arg);
2206	break;
2207
2208	case VMMR0_DO_NEM_UPDATE_STATISTICS:
2209	if (u64Arg \|\| pReqHdr)
2210	return VERR_INVALID_PARAMETER;
2211	rc = NEMR0UpdateStatistics(pGVM, idCpu);
2212	break;
2213
2214	# if 1 && defined(DEBUG_bird)
2215	case VMMR0_DO_NEM_EXPERIMENT:
2216	if (pReqHdr)
2217	return VERR_INVALID_PARAMETER;
2218	rc = NEMR0DoExperiment(pGVM, idCpu, u64Arg);
2219	break;
2220	# endif
2221	# endif
2222	#endif
2223
2224	/*
2225	* IOM requests.
2226	*/
2227	case VMMR0_DO_IOM_GROW_IO_PORTS:
2228	{
2229	if (pReqHdr \|\| idCpu != 0)
2230	return VERR_INVALID_PARAMETER;
2231	rc = IOMR0IoPortGrowRegistrationTables(pGVM, u64Arg);
2232	break;
2233	}
2234
2235	case VMMR0_DO_IOM_GROW_IO_PORT_STATS:
2236	{
2237	if (pReqHdr \|\| idCpu != 0)
2238	return VERR_INVALID_PARAMETER;
2239	rc = IOMR0IoPortGrowStatisticsTable(pGVM, u64Arg);
2240	break;
2241	}
2242
2243	case VMMR0_DO_IOM_GROW_MMIO_REGS:
2244	{
2245	if (pReqHdr \|\| idCpu != 0)
2246	return VERR_INVALID_PARAMETER;
2247	rc = IOMR0MmioGrowRegistrationTables(pGVM, u64Arg);
2248	break;
2249	}
2250
2251	case VMMR0_DO_IOM_GROW_MMIO_STATS:
2252	{
2253	if (pReqHdr \|\| idCpu != 0)
2254	return VERR_INVALID_PARAMETER;
2255	rc = IOMR0MmioGrowStatisticsTable(pGVM, u64Arg);
2256	break;
2257	}
2258
2259	case VMMR0_DO_IOM_SYNC_STATS_INDICES:
2260	{
2261	if (pReqHdr \|\| idCpu != 0)
2262	return VERR_INVALID_PARAMETER;
2263	rc = IOMR0IoPortSyncStatisticsIndices(pGVM);
2264	if (RT_SUCCESS(rc))
2265	rc = IOMR0MmioSyncStatisticsIndices(pGVM);
2266	break;
2267	}
2268
2269	/*
2270	* DBGF requests.
2271	*/
2272	#ifdef VBOX_WITH_DBGF_TRACING
2273	case VMMR0_DO_DBGF_TRACER_CREATE:
2274	{
2275	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2276	return VERR_INVALID_PARAMETER;
2277	rc = DBGFR0TracerCreateReqHandler(pGVM, (PDBGFTRACERCREATEREQ)pReqHdr);
2278	break;
2279	}
2280
2281	case VMMR0_DO_DBGF_TRACER_CALL_REQ_HANDLER:
2282	{
2283	if (!pReqHdr \|\| u64Arg)
2284	return VERR_INVALID_PARAMETER;
2285	# if 0 /** @todo */
2286	rc = DBGFR0TracerGenCallReqHandler(pGVM, (PDBGFTRACERGENCALLREQ)pReqHdr, idCpu);
2287	# else
2288	rc = VERR_NOT_IMPLEMENTED;
2289	# endif
2290	break;
2291	}
2292	#endif
2293
2294	case VMMR0_DO_DBGF_BP_INIT:
2295	{
2296	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2297	return VERR_INVALID_PARAMETER;
2298	rc = DBGFR0BpInitReqHandler(pGVM, (PDBGFBPINITREQ)pReqHdr);
2299	break;
2300	}
2301
2302	case VMMR0_DO_DBGF_BP_CHUNK_ALLOC:
2303	{
2304	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2305	return VERR_INVALID_PARAMETER;
2306	rc = DBGFR0BpChunkAllocReqHandler(pGVM, (PDBGFBPCHUNKALLOCREQ)pReqHdr);
2307	break;
2308	}
2309
2310	case VMMR0_DO_DBGF_BP_L2_TBL_CHUNK_ALLOC:
2311	{
2312	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2313	return VERR_INVALID_PARAMETER;
2314	rc = DBGFR0BpL2TblChunkAllocReqHandler(pGVM, (PDBGFBPL2TBLCHUNKALLOCREQ)pReqHdr);
2315	break;
2316	}
2317
2318	case VMMR0_DO_DBGF_BP_OWNER_INIT:
2319	{
2320	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2321	return VERR_INVALID_PARAMETER;
2322	rc = DBGFR0BpOwnerInitReqHandler(pGVM, (PDBGFBPOWNERINITREQ)pReqHdr);
2323	break;
2324	}
2325
2326	case VMMR0_DO_DBGF_BP_PORTIO_INIT:
2327	{
2328	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2329	return VERR_INVALID_PARAMETER;
2330	rc = DBGFR0BpPortIoInitReqHandler(pGVM, (PDBGFBPINITREQ)pReqHdr);
2331	break;
2332	}
2333
2334
2335	/*
2336	* TM requests.
2337	*/
2338	case VMMR0_DO_TM_GROW_TIMER_QUEUE:
2339	{
2340	if (pReqHdr \|\| idCpu == NIL_VMCPUID)
2341	return VERR_INVALID_PARAMETER;
2342	rc = TMR0TimerQueueGrow(pGVM, RT_HI_U32(u64Arg), RT_LO_U32(u64Arg));
2343	break;
2344	}
2345
2346	/*
2347	* For profiling.
2348	*/
2349	case VMMR0_DO_NOP:
2350	case VMMR0_DO_SLOW_NOP:
2351	return VINF_SUCCESS;
2352
2353	/*
2354	* For testing Ring-0 APIs invoked in this environment.
2355	*/
2356	case VMMR0_DO_TESTS:
2357	/** @todo make new test */
2358	return VINF_SUCCESS;
2359
2360	default:
2361	/*
2362	* We're returning VERR_NOT_SUPPORT here so we've got something else
2363	* than -1 which the interrupt gate glue code might return.
2364	*/
2365	Log(("operation %#x is not supported\n", enmOperation));
2366	return VERR_NOT_SUPPORTED;
2367	}
2368	return rc;
2369	}
2370
2371	#ifndef VMM_R0_SWITCH_STACK /* Not safe unless we disable preemption first. */
2372	/**
2373	* This is just a longjmp wrapper function for VMMR0EntryEx calls.
2374	*
2375	* @returns VBox status code.
2376	* @param pvArgs The argument package
2377	*/
2378	static DECLCALLBACK(int) vmmR0EntryExWrapper(void *pvArgs)
2379	{
2380	PGVMCPU pGVCpu = (PGVMCPU)pvArgs;
2381	return vmmR0EntryExWorker(pGVCpu->vmmr0.s.pGVM,
2382	pGVCpu->vmmr0.s.idCpu,
2383	pGVCpu->vmmr0.s.enmOperation,
2384	pGVCpu->vmmr0.s.pReq,
2385	pGVCpu->vmmr0.s.u64Arg,
2386	pGVCpu->vmmr0.s.pSession);
2387	}
2388	#endif
2389
2390
2391	/**
2392	* The Ring 0 entry point, called by the support library (SUP).
2393	*
2394	* @returns VBox status code.
2395	* @param pGVM The global (ring-0) VM structure.
2396	* @param pVM The cross context VM structure.
2397	* @param idCpu Virtual CPU ID argument. Must be NIL_VMCPUID if pVM
2398	* is NIL_RTR0PTR, and may be NIL_VMCPUID if it isn't
2399	* @param enmOperation Which operation to execute.
2400	* @param pReq Pointer to the SUPVMMR0REQHDR packet. Optional.
2401	* @param u64Arg Some simple constant argument.
2402	* @param pSession The session of the caller.
2403	* @remarks Assume called with interrupts _enabled_.
2404	*/
2405	VMMR0DECL(int) VMMR0EntryEx(PGVM pGVM, PVMCC pVM, VMCPUID idCpu, VMMR0OPERATION enmOperation,
2406	PSUPVMMR0REQHDR pReq, uint64_t u64Arg, PSUPDRVSESSION pSession)
2407	{
2408	#ifndef VMM_R0_SWITCH_STACK /* Not safe unless we disable preemption first. */
2409	/*
2410	* Requests that should only happen on the EMT thread will be
2411	* wrapped in a setjmp so we can assert without causing trouble.
2412	*/
2413	if ( pVM != NULL
2414	&& pGVM != NULL
2415	&& pVM == pGVM /** @todo drop pVM or pGVM */
2416	&& idCpu < pGVM->cCpus
2417	&& pGVM->pSession == pSession
2418	&& pGVM->pSelf == pVM)
2419	{
2420	switch (enmOperation)
2421	{
2422	/* These might/will be called before VMMR3Init. */
2423	case VMMR0_DO_GMM_INITIAL_RESERVATION:
2424	case VMMR0_DO_GMM_UPDATE_RESERVATION:
2425	case VMMR0_DO_GMM_ALLOCATE_PAGES:
2426	case VMMR0_DO_GMM_FREE_PAGES:
2427	case VMMR0_DO_GMM_BALLOONED_PAGES:
2428	/* On the mac we might not have a valid jmp buf, so check these as well. */
2429	case VMMR0_DO_VMMR0_INIT:
2430	case VMMR0_DO_VMMR0_TERM:
2431
2432	case VMMR0_DO_PDM_DEVICE_CREATE:
2433	case VMMR0_DO_PDM_DEVICE_GEN_CALL:
2434	case VMMR0_DO_IOM_GROW_IO_PORTS:
2435	case VMMR0_DO_IOM_GROW_IO_PORT_STATS:
2436	case VMMR0_DO_DBGF_BP_INIT:
2437	case VMMR0_DO_DBGF_BP_CHUNK_ALLOC:
2438	case VMMR0_DO_DBGF_BP_L2_TBL_CHUNK_ALLOC:
2439	{
2440	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2441	RTNATIVETHREAD hNativeThread = RTThreadNativeSelf();
2442	if (RT_LIKELY( pGVCpu->hEMT == hNativeThread
2443	&& pGVCpu->hNativeThreadR0 == hNativeThread))
2444	{
2445	if (!pGVCpu->vmm.s.CallRing3JmpBufR0.pvSavedStack)
2446	break;
2447
2448	pGVCpu->vmmr0.s.pGVM = pGVM;
2449	pGVCpu->vmmr0.s.idCpu = idCpu;
2450	pGVCpu->vmmr0.s.enmOperation = enmOperation;
2451	pGVCpu->vmmr0.s.pReq = pReq;
2452	pGVCpu->vmmr0.s.u64Arg = u64Arg;
2453	pGVCpu->vmmr0.s.pSession = pSession;
2454	return vmmR0CallRing3SetJmpEx(&pGVCpu->vmm.s.CallRing3JmpBufR0, vmmR0EntryExWrapper, pGVCpu,
2455	((uintptr_t)u64Arg << 16) \| (uintptr_t)enmOperation);
2456	}
2457	return VERR_VM_THREAD_NOT_EMT;
2458	}
2459
2460	default:
2461	case VMMR0_DO_PGM_POOL_GROW:
2462	break;
2463	}
2464	}
2465	#else
2466	RT_NOREF(pVM);
2467	#endif
2468	return vmmR0EntryExWorker(pGVM, idCpu, enmOperation, pReq, u64Arg, pSession);
2469	}
2470
2471
2472	/*********************************************************************************************************************************
2473	* EMT Blocking *
2474	*********************************************************************************************************************************/
2475
2476	/**
2477	* Checks whether we've armed the ring-0 long jump machinery.
2478	*
2479	* @returns @c true / @c false
2480	* @param pVCpu The cross context virtual CPU structure.
2481	* @thread EMT
2482	* @sa VMMIsLongJumpArmed
2483	*/
2484	VMMR0_INT_DECL(bool) VMMR0IsLongJumpArmed(PVMCPUCC pVCpu)
2485	{
2486	#ifdef RT_ARCH_X86
2487	return pVCpu->vmm.s.CallRing3JmpBufR0.eip
2488	&& !pVCpu->vmm.s.CallRing3JmpBufR0.fInRing3Call;
2489	#else
2490	return pVCpu->vmm.s.CallRing3JmpBufR0.rip
2491	&& !pVCpu->vmm.s.CallRing3JmpBufR0.fInRing3Call;
2492	#endif
2493	}
2494
2495
2496	/**
2497	* Checks whether we've done a ring-3 long jump.
2498	*
2499	* @returns @c true / @c false
2500	* @param pVCpu The cross context virtual CPU structure.
2501	* @thread EMT
2502	*/
2503	VMMR0_INT_DECL(bool) VMMR0IsInRing3LongJump(PVMCPUCC pVCpu)
2504	{
2505	return pVCpu->vmm.s.CallRing3JmpBufR0.fInRing3Call;
2506	}
2507
2508
2509	/**
2510	* Locking helper that deals with HM context and checks if the thread can block.
2511	*
2512	* @returns VINF_SUCCESS if we can block. Returns @a rcBusy or
2513	* VERR_VMM_CANNOT_BLOCK if not able to block.
2514	* @param pVCpu The cross context virtual CPU structure of the calling
2515	* thread.
2516	* @param rcBusy What to return in case of a blocking problem. Will IPE
2517	* if VINF_SUCCESS and we cannot block.
2518	* @param pszCaller The caller (for logging problems).
2519	* @param pvLock The lock address (for logging problems).
2520	* @param pCtx Where to return context info for the resume call.
2521	* @thread EMT(pVCpu)
2522	*/
2523	VMMR0_INT_DECL(int) VMMR0EmtPrepareToBlock(PVMCPUCC pVCpu, int rcBusy, const char pszCaller, void pvLock,
2524	PVMMR0EMTBLOCKCTX pCtx)
2525	{
2526	const char *pszMsg;
2527
2528	/*
2529	* Check that we are allowed to block.
2530	*/
2531	if (RT_LIKELY(VMMRZCallRing3IsEnabled(pVCpu)))
2532	{
2533	/*
2534	* Are we in HM context and w/o a context hook? If so work the context hook.
2535	*/
2536	if (pVCpu->idHostCpu != NIL_RTCPUID)
2537	{
2538	Assert(pVCpu->iHostCpuSet != UINT32_MAX);
2539
2540	if (pVCpu->vmmr0.s.hCtxHook == NIL_RTTHREADCTXHOOK)
2541	{
2542	vmmR0ThreadCtxCallback(RTTHREADCTXEVENT_OUT, pVCpu);
2543	if (pVCpu->vmmr0.s.pPreemptState)
2544	RTThreadPreemptRestore(pVCpu->vmmr0.s.pPreemptState);
2545
2546	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC;
2547	pCtx->fWasInHmContext = true;
2548	return VINF_SUCCESS;
2549	}
2550	}
2551
2552	if (RT_LIKELY(!pVCpu->vmmr0.s.pPreemptState))
2553	{
2554	/*
2555	* Not in HM context or we've got hooks, so just check that preemption
2556	* is enabled.
2557	*/
2558	if (RT_LIKELY(RTThreadPreemptIsEnabled(NIL_RTTHREAD)))
2559	{
2560	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC;
2561	pCtx->fWasInHmContext = false;
2562	return VINF_SUCCESS;
2563	}
2564	pszMsg = "Preemption is disabled!";
2565	}
2566	else
2567	pszMsg = "Preemption state w/o HM state!";
2568	}
2569	else
2570	pszMsg = "Ring-3 calls are disabled!";
2571
2572	static uint32_t volatile s_cWarnings = 0;
2573	if (++s_cWarnings < 50)
2574	SUPR0Printf("VMMR0EmtPrepareToBlock: %s pvLock=%p pszCaller=%s rcBusy=%p\n", pszMsg, pvLock, pszCaller, rcBusy);
2575	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC_DEAD;
2576	pCtx->fWasInHmContext = false;
2577	return rcBusy != VINF_SUCCESS ? rcBusy : VERR_VMM_CANNOT_BLOCK;
2578	}
2579
2580
2581	/**
2582	* Counterpart to VMMR0EmtPrepareToBlock.
2583	*
2584	* @param pVCpu The cross context virtual CPU structure of the calling
2585	* thread.
2586	* @param pCtx The context structure used with VMMR0EmtPrepareToBlock.
2587	* @thread EMT(pVCpu)
2588	*/
2589	VMMR0_INT_DECL(void) VMMR0EmtResumeAfterBlocking(PVMCPUCC pVCpu, PVMMR0EMTBLOCKCTX pCtx)
2590	{
2591	AssertReturnVoid(pCtx->uMagic == VMMR0EMTBLOCKCTX_MAGIC);
2592	if (pCtx->fWasInHmContext)
2593	{
2594	if (pVCpu->vmmr0.s.pPreemptState)
2595	RTThreadPreemptDisable(pVCpu->vmmr0.s.pPreemptState);
2596
2597	pCtx->fWasInHmContext = false;
2598	vmmR0ThreadCtxCallback(RTTHREADCTXEVENT_IN, pVCpu);
2599	}
2600	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC_DEAD;
2601	}
2602
2603
2604	/**
2605	* Helper for waiting on an RTSEMEVENT, caller did VMMR0EmtPrepareToBlock.
2606	*
2607	* @returns
2608	* @retval VERR_THREAD_IS_TERMINATING
2609	* @retval VERR_TIMEOUT if we ended up waiting too long, either according to
2610	* @a cMsTimeout or to maximum wait values.
2611	*
2612	* @param pGVCpu The ring-0 virtual CPU structure.
2613	* @param fFlags VMMR0EMTWAIT_F_XXX.
2614	* @param hEvent The event to wait on.
2615	* @param cMsTimeout The timeout or RT_INDEFINITE_WAIT.
2616	*/
2617	VMMR0_INT_DECL(int) VMMR0EmtWaitEventInner(PGVMCPU pGVCpu, uint32_t fFlags, RTSEMEVENT hEvent, RTMSINTERVAL cMsTimeout)
2618	{
2619	AssertReturn(pGVCpu->hEMT == RTThreadNativeSelf(), VERR_VM_THREAD_NOT_EMT);
2620
2621	/*
2622	* Note! Similar code is found in the PDM critical sections too.
2623	*/
2624	uint64_t const nsStart = RTTimeNanoTS();
2625	uint64_t cNsMaxTotal = cMsTimeout == RT_INDEFINITE_WAIT
2626	? RT_NS_5MIN : RT_MIN(RT_NS_5MIN, RT_NS_1MS_64 * cMsTimeout);
2627	uint32_t cMsMaxOne = RT_MS_5SEC;
2628	bool fNonInterruptible = false;
2629	for (;;)
2630	{
2631	/* Wait. */
2632	int rcWait = !fNonInterruptible
2633	? RTSemEventWaitNoResume(hEvent, cMsMaxOne)
2634	: RTSemEventWait(hEvent, cMsMaxOne);
2635	if (RT_SUCCESS(rcWait))
2636	return rcWait;
2637
2638	if (rcWait == VERR_TIMEOUT \|\| rcWait == VERR_INTERRUPTED)
2639	{
2640	uint64_t const cNsElapsed = RTTimeNanoTS() - nsStart;
2641
2642	/*
2643	* Check the thread termination status.
2644	*/
2645	int const rcTerm = RTThreadQueryTerminationStatus(NIL_RTTHREAD);
2646	AssertMsg(rcTerm == VINF_SUCCESS \|\| rcTerm == VERR_NOT_SUPPORTED \|\| rcTerm == VINF_THREAD_IS_TERMINATING,
2647	("rcTerm=%Rrc\n", rcTerm));
2648	if ( rcTerm == VERR_NOT_SUPPORTED
2649	&& !fNonInterruptible
2650	&& cNsMaxTotal > RT_NS_1MIN)
2651	cNsMaxTotal = RT_NS_1MIN;
2652
2653	/* We return immediately if it looks like the thread is terminating. */
2654	if (rcTerm == VINF_THREAD_IS_TERMINATING)
2655	return VERR_THREAD_IS_TERMINATING;
2656
2657	/* We may suppress VERR_INTERRUPTED if VMMR0EMTWAIT_F_TRY_SUPPRESS_INTERRUPTED was
2658	specified, otherwise we'll just return it. */
2659	if (rcWait == VERR_INTERRUPTED)
2660	{
2661	if (!(fFlags & VMMR0EMTWAIT_F_TRY_SUPPRESS_INTERRUPTED))
2662	return VERR_INTERRUPTED;
2663	if (!fNonInterruptible)
2664	{
2665	/* First time: Adjust down the wait parameters and make sure we get at least
2666	one non-interruptible wait before timing out. */
2667	fNonInterruptible = true;
2668	cMsMaxOne = 32;
2669	uint64_t const cNsLeft = cNsMaxTotal - cNsElapsed;
2670	if (cNsLeft > RT_NS_10SEC)
2671	cNsMaxTotal = cNsElapsed + RT_NS_10SEC;
2672	continue;
2673	}
2674	}
2675
2676	/* Check for timeout. */
2677	if (cNsElapsed > cNsMaxTotal)
2678	return VERR_TIMEOUT;
2679	}
2680	else
2681	return rcWait;
2682	}
2683	/* not reached */
2684	}
2685
2686
2687	/**
2688	* Helper for signalling an SUPSEMEVENT.
2689	*
2690	* This may temporarily leave the HM context if the host requires that for
2691	* signalling SUPSEMEVENT objects.
2692	*
2693	* @returns VBox status code (see VMMR0EmtPrepareToBlock)
2694	* @param pGVM The ring-0 VM structure.
2695	* @param pGVCpu The ring-0 virtual CPU structure.
2696	* @param hEvent The event to signal.
2697	*/
2698	VMMR0_INT_DECL(int) VMMR0EmtSignalSupEvent(PGVM pGVM, PGVMCPU pGVCpu, SUPSEMEVENT hEvent)
2699	{
2700	AssertReturn(pGVCpu->hEMT == RTThreadNativeSelf(), VERR_VM_THREAD_NOT_EMT);
2701	if (RTSemEventIsSignalSafe())
2702	return SUPSemEventSignal(pGVM->pSession, hEvent);
2703
2704	VMMR0EMTBLOCKCTX Ctx;
2705	int rc = VMMR0EmtPrepareToBlock(pGVCpu, VINF_SUCCESS, __FUNCTION__, (void *)(uintptr_t)hEvent, &Ctx);
2706	if (RT_SUCCESS(rc))
2707	{
2708	rc = SUPSemEventSignal(pGVM->pSession, hEvent);
2709	VMMR0EmtResumeAfterBlocking(pGVCpu, &Ctx);
2710	}
2711	return rc;
2712	}
2713
2714
2715	/**
2716	* Helper for signalling an SUPSEMEVENT, variant supporting non-EMTs.
2717	*
2718	* This may temporarily leave the HM context if the host requires that for
2719	* signalling SUPSEMEVENT objects.
2720	*
2721	* @returns VBox status code (see VMMR0EmtPrepareToBlock)
2722	* @param pGVM The ring-0 VM structure.
2723	* @param hEvent The event to signal.
2724	*/
2725	VMMR0_INT_DECL(int) VMMR0EmtSignalSupEventByGVM(PGVM pGVM, SUPSEMEVENT hEvent)
2726	{
2727	if (!RTSemEventIsSignalSafe())
2728	{
2729	PGVMCPU pGVCpu = GVMMR0GetGVCpuByGVMandEMT(pGVM, NIL_RTNATIVETHREAD);
2730	if (pGVCpu)
2731	{
2732	VMMR0EMTBLOCKCTX Ctx;
2733	int rc = VMMR0EmtPrepareToBlock(pGVCpu, VINF_SUCCESS, __FUNCTION__, (void *)(uintptr_t)hEvent, &Ctx);
2734	if (RT_SUCCESS(rc))
2735	{
2736	rc = SUPSemEventSignal(pGVM->pSession, hEvent);
2737	VMMR0EmtResumeAfterBlocking(pGVCpu, &Ctx);
2738	}
2739	return rc;
2740	}
2741	}
2742	return SUPSemEventSignal(pGVM->pSession, hEvent);
2743	}
2744
2745
2746	/*********************************************************************************************************************************
2747	* Logging. *
2748	*********************************************************************************************************************************/
2749
2750	/**
2751	* VMMR0_DO_VMMR0_UPDATE_LOGGERS: Updates the EMT loggers for the VM.
2752	*
2753	* @returns VBox status code.
2754	* @param pGVM The global (ring-0) VM structure.
2755	* @param idCpu The ID of the calling EMT.
2756	* @param pReq The request data.
2757	* @param idxLogger Which logger set to update.
2758	* @thread EMT(idCpu)
2759	*/
2760	static int vmmR0UpdateLoggers(PGVM pGVM, VMCPUID idCpu, PVMMR0UPDATELOGGERSREQ pReq, size_t idxLogger)
2761	{
2762	/*
2763	* Check sanity. First we require EMT to be calling us.
2764	*/
2765	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID);
2766	AssertReturn(pGVM->aCpus[idCpu].hEMT == RTThreadNativeSelf(), VERR_INVALID_CPU_ID);
2767
2768	AssertReturn(pReq->Hdr.cbReq >= RT_UOFFSETOF_DYN(VMMR0UPDATELOGGERSREQ, afGroups[0]), VERR_INVALID_PARAMETER);
2769	AssertReturn(pReq->cGroups < _8K, VERR_INVALID_PARAMETER);
2770	AssertReturn(pReq->Hdr.cbReq == RT_UOFFSETOF_DYN(VMMR0UPDATELOGGERSREQ, afGroups[pReq->cGroups]), VERR_INVALID_PARAMETER);
2771
2772	AssertReturn(idxLogger < VMMLOGGER_IDX_MAX, VERR_OUT_OF_RANGE);
2773
2774	/*
2775	* Adjust flags.
2776	*/
2777	/* Always buffered: */
2778	pReq->fFlags \|= RTLOGFLAGS_BUFFERED;
2779	/* These doesn't make sense at present: */
2780	pReq->fFlags &= ~(RTLOGFLAGS_FLUSH \| RTLOGFLAGS_WRITE_THROUGH);
2781	/* We've traditionally skipped the group restrictions. */
2782	pReq->fFlags &= ~RTLOGFLAGS_RESTRICT_GROUPS;
2783
2784	/*
2785	* Do the updating.
2786	*/
2787	int rc = VINF_SUCCESS;
2788	for (idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
2789	{
2790	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2791	PRTLOGGER pLogger = pGVCpu->vmmr0.s.u.aLoggers[idxLogger].pLogger;
2792	if (pLogger)
2793	{
2794	RTLogSetR0ProgramStart(pLogger, pGVM->vmm.s.nsProgramStart);
2795	rc = RTLogBulkUpdate(pLogger, pReq->fFlags, pReq->uGroupCrc32, pReq->cGroups, pReq->afGroups);
2796	}
2797	}
2798
2799	return rc;
2800	}
2801
2802
2803	/**
2804	* VMMR0_DO_VMMR0_LOG_FLUSHER: Get the next log flushing job.
2805	*
2806	* The job info is copied into VMM::LogFlusherItem.
2807	*
2808	* @returns VBox status code.
2809	* @retval VERR_OBJECT_DESTROYED if we're shutting down.
2810	* @retval VERR_NOT_OWNER if the calling thread is not the flusher thread.
2811	* @param pGVM The global (ring-0) VM structure.
2812	* @thread The log flusher thread (first caller automatically becomes the log
2813	* flusher).
2814	*/
2815	static int vmmR0LogFlusher(PGVM pGVM)
2816	{
2817	/*
2818	* Check that this really is the flusher thread.
2819	*/
2820	RTNATIVETHREAD const hNativeSelf = RTThreadNativeSelf();
2821	AssertReturn(hNativeSelf != NIL_RTNATIVETHREAD, VERR_INTERNAL_ERROR_3);
2822	if (RT_LIKELY(pGVM->vmmr0.s.LogFlusher.hThread == hNativeSelf))
2823	{ /* likely */ }
2824	else
2825	{
2826	/* The first caller becomes the flusher thread. */
2827	bool fOk;
2828	ASMAtomicCmpXchgHandle(&pGVM->vmmr0.s.LogFlusher.hThread, hNativeSelf, NIL_RTNATIVETHREAD, fOk);
2829	if (!fOk)
2830	return VERR_NOT_OWNER;
2831	pGVM->vmmr0.s.LogFlusher.fThreadRunning = true;
2832	}
2833
2834	/*
2835	* Acknowledge flush, waking up waiting EMT.
2836	*/
2837	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2838
2839	uint32_t idxTail = pGVM->vmmr0.s.LogFlusher.idxRingTail % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2840	uint32_t idxHead = pGVM->vmmr0.s.LogFlusher.idxRingHead % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2841	if ( idxTail != idxHead
2842	&& pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.fProcessing)
2843	{
2844	/* Pop the head off the ring buffer. */
2845	uint32_t const idCpu = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.idCpu;
2846	uint32_t const idxLogger = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.idxLogger;
2847	uint32_t const idxBuffer = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.idxBuffer;
2848
2849	pGVM->vmmr0.s.LogFlusher.aRing[idxHead].u32 = UINT32_MAX >> 1; /* invalidate the entry */
2850	pGVM->vmmr0.s.LogFlusher.idxRingHead = (idxHead + 1) % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2851
2852	/* Validate content. */
2853	if ( idCpu < pGVM->cCpus
2854	&& idxLogger < VMMLOGGER_IDX_MAX
2855	&& idxBuffer < VMMLOGGER_BUFFER_COUNT)
2856	{
2857	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2858	PVMMR0PERVCPULOGGER pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
2859	PVMMR3CPULOGGER pShared = &pGVCpu->vmm.s.u.aLoggers[idxLogger];
2860
2861	/*
2862	* Accounting.
2863	*/
2864	uint32_t cFlushing = pR0Log->cFlushing - 1;
2865	if (RT_LIKELY(cFlushing < VMMLOGGER_BUFFER_COUNT))
2866	{ /likely/ }
2867	else
2868	cFlushing = 0;
2869	pR0Log->cFlushing = cFlushing;
2870	ASMAtomicWriteU32(&pShared->cFlushing, cFlushing);
2871
2872	/*
2873	* Wake up the EMT if it's waiting.
2874	*/
2875	if (!pR0Log->fEmtWaiting)
2876	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2877	else
2878	{
2879	pR0Log->fEmtWaiting = false;
2880	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2881
2882	int rc = RTSemEventSignal(pR0Log->hEventFlushWait);
2883	if (RT_FAILURE(rc))
2884	LogRelMax(64, ("vmmR0LogFlusher: RTSemEventSignal failed ACKing entry #%u (%u/%u/%u): %Rrc!\n",
2885	idxHead, idCpu, idxLogger, idxBuffer, rc));
2886	}
2887	}
2888	else
2889	{
2890	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2891	LogRelMax(64, ("vmmR0LogFlusher: Bad ACK entry #%u: %u/%u/%u!\n", idxHead, idCpu, idxLogger, idxBuffer));
2892	}
2893
2894	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2895	}
2896
2897	/*
2898	* The wait loop.
2899	*/
2900	int rc;
2901	for (;;)
2902	{
2903	/*
2904	* Work pending?
2905	*/
2906	idxTail = pGVM->vmmr0.s.LogFlusher.idxRingTail % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2907	idxHead = pGVM->vmmr0.s.LogFlusher.idxRingHead % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2908	if (idxTail != idxHead)
2909	{
2910	pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.fProcessing = true;
2911	pGVM->vmm.s.LogFlusherItem.u32 = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].u32;
2912
2913	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2914	return VINF_SUCCESS;
2915	}
2916
2917	/*
2918	* Nothing to do, so, check for termination and go to sleep.
2919	*/
2920	if (!pGVM->vmmr0.s.LogFlusher.fThreadShutdown)
2921	{ /* likely */ }
2922	else
2923	{
2924	rc = VERR_OBJECT_DESTROYED;
2925	break;
2926	}
2927
2928	pGVM->vmmr0.s.LogFlusher.fThreadWaiting = true;
2929	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2930
2931	rc = RTSemEventWaitNoResume(pGVM->vmmr0.s.LogFlusher.hEvent, RT_MS_5MIN);
2932
2933	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2934	pGVM->vmmr0.s.LogFlusher.fThreadWaiting = false;
2935
2936	if (RT_SUCCESS(rc) \|\| rc == VERR_TIMEOUT)
2937	{ /* likely */ }
2938	else if (rc == VERR_INTERRUPTED)
2939	{
2940	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2941	return rc;
2942	}
2943	else if (rc == VERR_SEM_DESTROYED \|\| rc == VERR_INVALID_HANDLE)
2944	break;
2945	else
2946	{
2947	LogRel(("vmmR0LogFlusher: RTSemEventWaitNoResume returned unexpected status %Rrc\n", rc));
2948	break;
2949	}
2950	}
2951
2952	/*
2953	* Terminating - prevent further calls and indicate to the EMTs that we're no longer around.
2954	*/
2955	pGVM->vmmr0.s.LogFlusher.hThread = ~pGVM->vmmr0.s.LogFlusher.hThread; /* (should be reasonably safe) */
2956	pGVM->vmmr0.s.LogFlusher.fThreadRunning = false;
2957
2958	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2959	return rc;
2960	}
2961
2962
2963	/**
2964	* VMMR0_DO_VMMR0_LOG_WAIT_FLUSHED: Waits for the flusher thread to finish all
2965	* buffers for logger @a idxLogger.
2966	*
2967	* @returns VBox status code.
2968	* @param pGVM The global (ring-0) VM structure.
2969	* @param idCpu The ID of the calling EMT.
2970	* @param idxLogger Which logger to wait on.
2971	* @thread EMT(idCpu)
2972	*/
2973	static int vmmR0LogWaitFlushed(PGVM pGVM, VMCPUID idCpu, size_t idxLogger)
2974	{
2975	/*
2976	* Check sanity. First we require EMT to be calling us.
2977	*/
2978	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID);
2979	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2980	AssertReturn(pGVCpu->hEMT == RTThreadNativeSelf(), VERR_INVALID_CPU_ID);
2981	AssertReturn(idxLogger < VMMLOGGER_IDX_MAX, VERR_OUT_OF_RANGE);
2982	PVMMR0PERVCPULOGGER const pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
2983
2984	/*
2985	* Do the waiting.
2986	*/
2987	int rc = VINF_SUCCESS;
2988	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2989	uint32_t cFlushing = pR0Log->cFlushing;
2990	while (cFlushing > 0)
2991	{
2992	pR0Log->fEmtWaiting = true;
2993	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2994
2995	rc = RTSemEventWaitNoResume(pR0Log->hEventFlushWait, RT_MS_5MIN);
2996
2997	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2998	pR0Log->fEmtWaiting = false;
2999	if (RT_SUCCESS(rc))
3000	{
3001	/* Read the new count, make sure it decreased before looping. That
3002	way we can guarentee that we will only wait more than 5 min * buffers. */
3003	uint32_t const cPrevFlushing = cFlushing;
3004	cFlushing = pR0Log->cFlushing;
3005	if (cFlushing < cPrevFlushing)
3006	continue;
3007	rc = VERR_INTERNAL_ERROR_3;
3008	}
3009	break;
3010	}
3011	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3012	return rc;
3013	}
3014
3015
3016	/**
3017	* Inner worker for vmmR0LoggerFlushCommon.
3018	*/
3019	#ifndef VMM_R0_SWITCH_STACK
3020	static bool vmmR0LoggerFlushInner(PGVM pGVM, PGVMCPU pGVCpu, uint32_t idxLogger, size_t idxBuffer, uint32_t cbToFlush)
3021	#else
3022	DECLASM(bool) vmmR0LoggerFlushInner(PGVM pGVM, PGVMCPU pGVCpu, uint32_t idxLogger, size_t idxBuffer, uint32_t cbToFlush);
3023	DECLASM(bool) StkBack_vmmR0LoggerFlushInner(PGVM pGVM, PGVMCPU pGVCpu, uint32_t idxLogger, size_t idxBuffer, uint32_t cbToFlush)
3024	#endif
3025	{
3026	PVMMR0PERVCPULOGGER const pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
3027	PVMMR3CPULOGGER const pShared = &pGVCpu->vmm.s.u.aLoggers[idxLogger];
3028
3029	/*
3030	* Figure out what we need to do and whether we can.
3031	*/
3032	enum { kJustSignal, kPrepAndSignal, kPrepSignalAndWait } enmAction;
3033	#if VMMLOGGER_BUFFER_COUNT >= 2
3034	if (pR0Log->cFlushing < VMMLOGGER_BUFFER_COUNT - 1)
3035	{
3036	if (RTSemEventIsSignalSafe())
3037	enmAction = kJustSignal;
3038	else if (VMMRZCallRing3IsEnabled(pGVCpu))
3039	enmAction = kPrepAndSignal;
3040	else
3041	{
3042	/** @todo This is a bit simplistic. We could introduce a FF to signal the
3043	* thread or similar. */
3044	STAM_REL_COUNTER_INC(&pShared->StatCannotBlock);
3045	# if defined(RT_OS_LINUX)
3046	SUP_DPRINTF(("vmmR0LoggerFlush: Signalling not safe and EMT blocking disabled! (%u bytes)\n", cbToFlush));
3047	# endif
3048	pShared->cbDropped += cbToFlush;
3049	return true;
3050	}
3051	}
3052	else
3053	#endif
3054	if (VMMRZCallRing3IsEnabled(pGVCpu))
3055	enmAction = kPrepSignalAndWait;
3056	else
3057	{
3058	STAM_REL_COUNTER_INC(&pShared->StatCannotBlock);
3059	# if defined(RT_OS_LINUX)
3060	SUP_DPRINTF(("vmmR0LoggerFlush: EMT blocking disabled! (%u bytes)\n", cbToFlush));
3061	# endif
3062	pShared->cbDropped += cbToFlush;
3063	return true;
3064	}
3065
3066	/*
3067	* Prepare for blocking if necessary.
3068	*/
3069	VMMR0EMTBLOCKCTX Ctx;
3070	if (enmAction != kJustSignal)
3071	{
3072	int rc = VMMR0EmtPrepareToBlock(pGVCpu, VINF_SUCCESS, "vmmR0LoggerFlushInner", pR0Log->hEventFlushWait, &Ctx);
3073	if (RT_SUCCESS(rc))
3074	{ /* likely */ }
3075	else
3076	{
3077	STAM_REL_COUNTER_INC(&pShared->StatCannotBlock);
3078	SUP_DPRINTF(("vmmR0LoggerFlush: VMMR0EmtPrepareToBlock failed! rc=%d\n", rc));
3079	return false;
3080	}
3081	}
3082
3083	/*
3084	* Queue the flush job.
3085	*/
3086	bool fFlushedBuffer;
3087	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3088	if (pGVM->vmmr0.s.LogFlusher.fThreadRunning)
3089	{
3090	uint32_t const idxHead = pGVM->vmmr0.s.LogFlusher.idxRingHead % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
3091	uint32_t const idxTail = pGVM->vmmr0.s.LogFlusher.idxRingTail % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
3092	uint32_t const idxNewTail = (idxTail + 1) % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
3093	if (idxNewTail != idxHead)
3094	{
3095	/* Queue it. */
3096	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.idCpu = pGVCpu->idCpu;
3097	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.idxLogger = idxLogger;
3098	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.idxBuffer = (uint32_t)idxBuffer;
3099	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.fProcessing = 0;
3100	pGVM->vmmr0.s.LogFlusher.idxRingTail = idxNewTail;
3101
3102	/* Update the number of buffers currently being flushed. */
3103	uint32_t cFlushing = pR0Log->cFlushing;
3104	cFlushing = RT_MIN(cFlushing + 1, VMMLOGGER_BUFFER_COUNT);
3105	pShared->cFlushing = pR0Log->cFlushing = cFlushing;
3106
3107	/* We must wait if all buffers are currently being flushed. */
3108	bool const fEmtWaiting = cFlushing >= VMMLOGGER_BUFFER_COUNT && enmAction != kJustSignal /* paranoia */;
3109	pR0Log->fEmtWaiting = fEmtWaiting;
3110
3111	/* Stats. */
3112	STAM_REL_COUNTER_INC(&pShared->StatFlushes);
3113	STAM_REL_COUNTER_INC(&pGVM->vmm.s.StatLogFlusherFlushes);
3114
3115	/* Signal the worker thread. */
3116	if (pGVM->vmmr0.s.LogFlusher.fThreadWaiting)
3117	{
3118	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3119	RTSemEventSignal(pGVM->vmmr0.s.LogFlusher.hEvent);
3120	}
3121	else
3122	{
3123	STAM_REL_COUNTER_INC(&pGVM->vmm.s.StatLogFlusherNoWakeUp);
3124	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3125	}
3126
3127	/*
3128	* Wait for a buffer to finish flushing.
3129	*
3130	* Note! Lazy bird is ignoring the status code here. The result is
3131	* that we might end up with an extra even signalling and the
3132	* next time we need to wait we won't and end up with some log
3133	* corruption. However, it's too much hazzle right now for
3134	* a scenario which would most likely end the process rather
3135	* than causing log corruption.
3136	*/
3137	if (fEmtWaiting)
3138	{
3139	STAM_REL_PROFILE_START(&pShared->StatWait, a);
3140	VMMR0EmtWaitEventInner(pGVCpu, VMMR0EMTWAIT_F_TRY_SUPPRESS_INTERRUPTED,
3141	pR0Log->hEventFlushWait, RT_INDEFINITE_WAIT);
3142	STAM_REL_PROFILE_STOP(&pShared->StatWait, a);
3143	}
3144
3145	/*
3146	* We always switch buffer if we have more than one.
3147	*/
3148	#if VMMLOGGER_BUFFER_COUNT == 1
3149	fFlushedBuffer = true;
3150	#else
3151	AssertCompile(VMMLOGGER_BUFFER_COUNT >= 1);
3152	pShared->idxBuf = (idxBuffer + 1) % VMMLOGGER_BUFFER_COUNT;
3153	fFlushedBuffer = false;
3154	#endif
3155	}
3156	else
3157	{
3158	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3159	SUP_DPRINTF(("vmmR0LoggerFlush: ring buffer is full!\n"));
3160	fFlushedBuffer = true;
3161	}
3162	}
3163	else
3164	{
3165	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3166	SUP_DPRINTF(("vmmR0LoggerFlush: flusher not active - dropping %u bytes\n", cbToFlush));
3167	fFlushedBuffer = true;
3168	}
3169
3170	/*
3171	* Restore the HM context.
3172	*/
3173	if (enmAction != kJustSignal)
3174	VMMR0EmtResumeAfterBlocking(pGVCpu, &Ctx);
3175
3176	return fFlushedBuffer;
3177	}
3178
3179
3180	/**
3181	* Common worker for vmmR0LogFlush and vmmR0LogRelFlush.
3182	*/
3183	static bool vmmR0LoggerFlushCommon(PRTLOGGER pLogger, PRTLOGBUFFERDESC pBufDesc, uint32_t idxLogger)
3184	{
3185	/*
3186	* Convert the pLogger into a GVMCPU handle and 'call' back to Ring-3.
3187	* (This is a bit paranoid code.)
3188	*/
3189	if (RT_VALID_PTR(pLogger))
3190	{
3191	if ( pLogger->u32Magic == RTLOGGER_MAGIC
3192	&& (pLogger->u32UserValue1 & VMMR0_LOGGER_FLAGS_MAGIC_MASK) == VMMR0_LOGGER_FLAGS_MAGIC_VALUE
3193	&& pLogger->u64UserValue2 == pLogger->u64UserValue3)
3194	{
3195	PGVMCPU const pGVCpu = (PGVMCPU)(uintptr_t)pLogger->u64UserValue2;
3196	if ( RT_VALID_PTR(pGVCpu)
3197	&& ((uintptr_t)pGVCpu & PAGE_OFFSET_MASK) == 0)
3198	{
3199	RTNATIVETHREAD const hNativeSelf = RTThreadNativeSelf();
3200	PGVM const pGVM = pGVCpu->pGVM;
3201	if ( hNativeSelf == pGVCpu->hEMT
3202	&& RT_VALID_PTR(pGVM))
3203	{
3204	PVMMR0PERVCPULOGGER const pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
3205	size_t const idxBuffer = pBufDesc - &pR0Log->aBufDescs[0];
3206	if (idxBuffer < VMMLOGGER_BUFFER_COUNT)
3207	{
3208	/*
3209	* Make sure we don't recurse forever here should something in the
3210	* following code trigger logging or an assertion. Do the rest in
3211	* an inner work to avoid hitting the right margin too hard.
3212	*/
3213	if (!pR0Log->fFlushing)
3214	{
3215	pR0Log->fFlushing = true;
3216	bool fFlushed = vmmR0LoggerFlushInner(pGVM, pGVCpu, idxLogger, idxBuffer, pBufDesc->offBuf);
3217	pR0Log->fFlushing = false;
3218	return fFlushed;
3219	}
3220
3221	SUP_DPRINTF(("vmmR0LoggerFlush: Recursive flushing!\n"));
3222	}
3223	else
3224	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p pGVCpu=%p: idxBuffer=%#zx\n", pLogger, pGVCpu, idxBuffer));
3225	}
3226	else
3227	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p pGVCpu=%p hEMT=%p hNativeSelf=%p!\n",
3228	pLogger, pGVCpu, pGVCpu->hEMT, hNativeSelf));
3229	}
3230	else
3231	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p pGVCpu=%p!\n", pLogger, pGVCpu));
3232	}
3233	else
3234	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p u32Magic=%#x u32UserValue1=%#x u64UserValue2=%#RX64 u64UserValue3=%#RX64!\n",
3235	pLogger, pLogger->u32Magic, pLogger->u32UserValue1, pLogger->u64UserValue2, pLogger->u64UserValue3));
3236	}
3237	else
3238	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p!\n", pLogger));
3239	return true;
3240	}
3241
3242
3243	/**
3244	* @callback_method_impl{FNRTLOGFLUSH, Release logger buffer flush callback.}
3245	*/
3246	static DECLCALLBACK(bool) vmmR0LogRelFlush(PRTLOGGER pLogger, PRTLOGBUFFERDESC pBufDesc)
3247	{
3248	return vmmR0LoggerFlushCommon(pLogger, pBufDesc, VMMLOGGER_IDX_RELEASE);
3249	}
3250
3251
3252	/**
3253	* @callback_method_impl{FNRTLOGFLUSH, Logger (debug) buffer flush callback.}
3254	*/
3255	static DECLCALLBACK(bool) vmmR0LogFlush(PRTLOGGER pLogger, PRTLOGBUFFERDESC pBufDesc)
3256	{
3257	#ifdef LOG_ENABLED
3258	return vmmR0LoggerFlushCommon(pLogger, pBufDesc, VMMLOGGER_IDX_REGULAR);
3259	#else
3260	RT_NOREF(pLogger, pBufDesc);
3261	return true;
3262	#endif
3263	}
3264
3265
3266	/*
3267	* Override RTLogDefaultInstanceEx so we can do logging from EMTs in ring-0.
3268	*/
3269	DECLEXPORT(PRTLOGGER) RTLogDefaultInstanceEx(uint32_t fFlagsAndGroup)
3270	{
3271	#ifdef LOG_ENABLED
3272	PGVMCPU pGVCpu = GVMMR0GetGVCpuByEMT(NIL_RTNATIVETHREAD);
3273	if (pGVCpu)
3274	{
3275	PRTLOGGER pLogger = pGVCpu->vmmr0.s.u.s.Logger.pLogger;
3276	if (RT_VALID_PTR(pLogger))
3277	{
3278	if ( pLogger->u64UserValue2 == (uintptr_t)pGVCpu
3279	&& pLogger->u64UserValue3 == (uintptr_t)pGVCpu)
3280	{
3281	if (!pGVCpu->vmmr0.s.u.s.Logger.fFlushing)
3282	{
3283	if (!(pGVCpu->vmmr0.s.fLogFlushingDisabled))
3284	return RTLogCheckGroupFlags(pLogger, fFlagsAndGroup);
3285	return NULL;
3286	}
3287
3288	/*
3289	* When we're flushing we _must_ return NULL here to suppress any
3290	* attempts at using the logger while in vmmR0LoggerFlushCommon.
3291	* The VMMR0EmtPrepareToBlock code may trigger logging in HM,
3292	* which will reset the buffer content before we even get to queue
3293	* the flush request. (Only an issue when VBOX_WITH_R0_LOGGING
3294	* is enabled.)
3295	*/
3296	return NULL;
3297	}
3298	}
3299	}
3300	#endif
3301	return SUPR0DefaultLogInstanceEx(fFlagsAndGroup);
3302	}
3303
3304
3305	/*
3306	* Override RTLogRelGetDefaultInstanceEx so we can do LogRel to VBox.log from EMTs in ring-0.
3307	*/
3308	DECLEXPORT(PRTLOGGER) RTLogRelGetDefaultInstanceEx(uint32_t fFlagsAndGroup)
3309	{
3310	PGVMCPU pGVCpu = GVMMR0GetGVCpuByEMT(NIL_RTNATIVETHREAD);
3311	if (pGVCpu)
3312	{
3313	PRTLOGGER pLogger = pGVCpu->vmmr0.s.u.s.RelLogger.pLogger;
3314	if (RT_VALID_PTR(pLogger))
3315	{
3316	if ( pLogger->u64UserValue2 == (uintptr_t)pGVCpu
3317	&& pLogger->u64UserValue3 == (uintptr_t)pGVCpu)
3318	{
3319	if (!pGVCpu->vmmr0.s.u.s.RelLogger.fFlushing)
3320	{
3321	if (!(pGVCpu->vmmr0.s.fLogFlushingDisabled))
3322	return RTLogCheckGroupFlags(pLogger, fFlagsAndGroup);
3323	return NULL;
3324	}
3325	}
3326	}
3327	}
3328	return SUPR0GetDefaultLogRelInstanceEx(fFlagsAndGroup);
3329	}
3330
3331
3332	/**
3333	* Helper for vmmR0InitLoggerSet
3334	*/
3335	static int vmmR0InitLoggerOne(PGVMCPU pGVCpu, bool fRelease, PVMMR0PERVCPULOGGER pR0Log, PVMMR3CPULOGGER pShared,
3336	uint32_t cbBuf, char *pchBuf, RTR3PTR pchBufR3)
3337	{
3338	/*
3339	* Create and configure the logger.
3340	*/
3341	for (size_t i = 0; i < VMMLOGGER_BUFFER_COUNT; i++)
3342	{
3343	pR0Log->aBufDescs[i].u32Magic = RTLOGBUFFERDESC_MAGIC;
3344	pR0Log->aBufDescs[i].uReserved = 0;
3345	pR0Log->aBufDescs[i].cbBuf = cbBuf;
3346	pR0Log->aBufDescs[i].offBuf = 0;
3347	pR0Log->aBufDescs[i].pchBuf = pchBuf + i * cbBuf;
3348	pR0Log->aBufDescs[i].pAux = &pShared->aBufs[i].AuxDesc;
3349
3350	pShared->aBufs[i].AuxDesc.fFlushedIndicator = false;
3351	pShared->aBufs[i].AuxDesc.afPadding[0] = 0;
3352	pShared->aBufs[i].AuxDesc.afPadding[1] = 0;
3353	pShared->aBufs[i].AuxDesc.afPadding[2] = 0;
3354	pShared->aBufs[i].AuxDesc.offBuf = 0;
3355	pShared->aBufs[i].pchBufR3 = pchBufR3 + i * cbBuf;
3356	}
3357	pShared->cbBuf = cbBuf;
3358
3359	static const char * const s_apszGroups[] = VBOX_LOGGROUP_NAMES;
3360	int rc = RTLogCreateEx(&pR0Log->pLogger, fRelease ? "VBOX_RELEASE_LOG" : "VBOX_LOG", RTLOG_F_NO_LOCKING \| RTLOGFLAGS_BUFFERED,
3361	"all", RT_ELEMENTS(s_apszGroups), s_apszGroups, UINT32_MAX,
3362	VMMLOGGER_BUFFER_COUNT, pR0Log->aBufDescs, RTLOGDEST_DUMMY,
3363	NULL /pfnPhase/, 0 /cHistory/, 0 /cbHistoryFileMax/, 0 /cSecsHistoryTimeSlot/,
3364	NULL /pErrInfo/, NULL /pszFilenameFmt/);
3365	if (RT_SUCCESS(rc))
3366	{
3367	PRTLOGGER pLogger = pR0Log->pLogger;
3368	pLogger->u32UserValue1 = VMMR0_LOGGER_FLAGS_MAGIC_VALUE;
3369	pLogger->u64UserValue2 = (uintptr_t)pGVCpu;
3370	pLogger->u64UserValue3 = (uintptr_t)pGVCpu;
3371
3372	rc = RTLogSetFlushCallback(pLogger, fRelease ? vmmR0LogRelFlush : vmmR0LogFlush);
3373	if (RT_SUCCESS(rc))
3374	{
3375	RTLogSetR0ThreadNameF(pLogger, "EMT-%u-R0", pGVCpu->idCpu);
3376
3377	/*
3378	* Create the event sem the EMT waits on while flushing is happening.
3379	*/
3380	rc = RTSemEventCreate(&pR0Log->hEventFlushWait);
3381	if (RT_SUCCESS(rc))
3382	return VINF_SUCCESS;
3383	pR0Log->hEventFlushWait = NIL_RTSEMEVENT;
3384	}
3385	RTLogDestroy(pLogger);
3386	}
3387	pR0Log->pLogger = NULL;
3388	return rc;
3389	}
3390
3391
3392	/**
3393	* Worker for VMMR0CleanupVM and vmmR0InitLoggerSet that destroys one logger.
3394	*/
3395	static void vmmR0TermLoggerOne(PVMMR0PERVCPULOGGER pR0Log, PVMMR3CPULOGGER pShared)
3396	{
3397	RTLogDestroy(pR0Log->pLogger);
3398	pR0Log->pLogger = NULL;
3399
3400	for (size_t i = 0; i < VMMLOGGER_BUFFER_COUNT; i++)
3401	pShared->aBufs[i].pchBufR3 = NIL_RTR3PTR;
3402
3403	RTSemEventDestroy(pR0Log->hEventFlushWait);
3404	pR0Log->hEventFlushWait = NIL_RTSEMEVENT;
3405	}
3406
3407
3408	/**
3409	* Initializes one type of loggers for each EMT.
3410	*/
3411	static int vmmR0InitLoggerSet(PGVM pGVM, uint8_t idxLogger, uint32_t cbBuf, PRTR0MEMOBJ phMemObj, PRTR0MEMOBJ phMapObj)
3412	{
3413	/* Allocate buffers first. */
3414	int rc = RTR0MemObjAllocPage(phMemObj, cbBuf * pGVM->cCpus * VMMLOGGER_BUFFER_COUNT, false /fExecutable/);
3415	if (RT_SUCCESS(rc))
3416	{
3417	rc = RTR0MemObjMapUser(phMapObj, phMemObj, (RTR3PTR)-1, 0 /uAlignment*/, RTMEM_PROT_READ, NIL_RTR0PROCESS);
3418	if (RT_SUCCESS(rc))
3419	{
3420	char * const pchBuf = (char )RTR0MemObjAddress(phMemObj);
3421	AssertPtrReturn(pchBuf, VERR_INTERNAL_ERROR_2);
3422
3423	RTR3PTR const pchBufR3 = RTR0MemObjAddressR3(*phMapObj);
3424	AssertReturn(pchBufR3 != NIL_RTR3PTR, VERR_INTERNAL_ERROR_3);
3425
3426	/* Initialize the per-CPU loggers. */
3427	for (uint32_t i = 0; i < pGVM->cCpus; i++)
3428	{
3429	PGVMCPU pGVCpu = &pGVM->aCpus[i];
3430	PVMMR0PERVCPULOGGER pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
3431	PVMMR3CPULOGGER pShared = &pGVCpu->vmm.s.u.aLoggers[idxLogger];
3432	rc = vmmR0InitLoggerOne(pGVCpu, idxLogger == VMMLOGGER_IDX_RELEASE, pR0Log, pShared, cbBuf,
3433	pchBuf + i * cbBuf * VMMLOGGER_BUFFER_COUNT,
3434	pchBufR3 + i * cbBuf * VMMLOGGER_BUFFER_COUNT);
3435	if (RT_FAILURE(rc))
3436	{
3437	vmmR0TermLoggerOne(pR0Log, pShared);
3438	while (i-- > 0)
3439	{
3440	pGVCpu = &pGVM->aCpus[i];
3441	vmmR0TermLoggerOne(&pGVCpu->vmmr0.s.u.aLoggers[idxLogger], &pGVCpu->vmm.s.u.aLoggers[idxLogger]);
3442	}
3443	break;
3444	}
3445	}
3446	if (RT_SUCCESS(rc))
3447	return VINF_SUCCESS;
3448
3449	/* Bail out. */
3450	RTR0MemObjFree(phMapObj, false /fFreeMappings*/);
3451	*phMapObj = NIL_RTR0MEMOBJ;
3452	}
3453	RTR0MemObjFree(phMemObj, true /fFreeMappings*/);
3454	*phMemObj = NIL_RTR0MEMOBJ;
3455	}
3456	return rc;
3457	}
3458
3459
3460	/**
3461	* Worker for VMMR0InitPerVMData that initializes all the logging related stuff.
3462	*
3463	* @returns VBox status code.
3464	* @param pGVM The global (ring-0) VM structure.
3465	*/
3466	static int vmmR0InitLoggers(PGVM pGVM)
3467	{
3468	/*
3469	* Invalidate the ring buffer (not really necessary).
3470	*/
3471	for (size_t idx = 0; idx < RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing); idx++)
3472	pGVM->vmmr0.s.LogFlusher.aRing[idx].u32 = UINT32_MAX >> 1; /* (all bits except fProcessing set) */
3473
3474	/*
3475	* Create the spinlock and flusher event semaphore.
3476	*/
3477	int rc = RTSpinlockCreate(&pGVM->vmmr0.s.LogFlusher.hSpinlock, RTSPINLOCK_FLAGS_INTERRUPT_SAFE, "VM-Log-Flusher");
3478	if (RT_SUCCESS(rc))
3479	{
3480	rc = RTSemEventCreate(&pGVM->vmmr0.s.LogFlusher.hEvent);
3481	if (RT_SUCCESS(rc))
3482	{
3483	/*
3484	* Create the ring-0 release loggers.
3485	*/
3486	rc = vmmR0InitLoggerSet(pGVM, VMMLOGGER_IDX_RELEASE, _4K,
3487	&pGVM->vmmr0.s.hMemObjReleaseLogger, &pGVM->vmmr0.s.hMapObjReleaseLogger);
3488	#ifdef LOG_ENABLED
3489	if (RT_SUCCESS(rc))
3490	{
3491	/*
3492	* Create debug loggers.
3493	*/
3494	rc = vmmR0InitLoggerSet(pGVM, VMMLOGGER_IDX_REGULAR, _64K,
3495	&pGVM->vmmr0.s.hMemObjLogger, &pGVM->vmmr0.s.hMapObjLogger);
3496	}
3497	#endif
3498	}
3499	}
3500	return rc;
3501	}
3502
3503
3504	/**
3505	* Worker for VMMR0InitPerVMData that initializes all the logging related stuff.
3506	*
3507	* @param pGVM The global (ring-0) VM structure.
3508	*/
3509	static void vmmR0CleanupLoggers(PGVM pGVM)
3510	{
3511	for (VMCPUID idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
3512	{
3513	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
3514	for (size_t iLogger = 0; iLogger < RT_ELEMENTS(pGVCpu->vmmr0.s.u.aLoggers); iLogger++)
3515	vmmR0TermLoggerOne(&pGVCpu->vmmr0.s.u.aLoggers[iLogger], &pGVCpu->vmm.s.u.aLoggers[iLogger]);
3516	}
3517
3518	/*
3519	* Free logger buffer memory.
3520	*/
3521	RTR0MemObjFree(pGVM->vmmr0.s.hMapObjReleaseLogger, false /fFreeMappings/);
3522	pGVM->vmmr0.s.hMapObjReleaseLogger = NIL_RTR0MEMOBJ;
3523	RTR0MemObjFree(pGVM->vmmr0.s.hMemObjReleaseLogger, true /fFreeMappings/);
3524	pGVM->vmmr0.s.hMemObjReleaseLogger = NIL_RTR0MEMOBJ;
3525
3526	RTR0MemObjFree(pGVM->vmmr0.s.hMapObjLogger, false /fFreeMappings/);
3527	pGVM->vmmr0.s.hMapObjLogger = NIL_RTR0MEMOBJ;
3528	RTR0MemObjFree(pGVM->vmmr0.s.hMemObjLogger, true /fFreeMappings/);
3529	pGVM->vmmr0.s.hMemObjLogger = NIL_RTR0MEMOBJ;
3530
3531	/*
3532	* Free log flusher related stuff.
3533	*/
3534	RTSpinlockDestroy(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3535	pGVM->vmmr0.s.LogFlusher.hSpinlock = NIL_RTSPINLOCK;
3536	RTSemEventDestroy(pGVM->vmmr0.s.LogFlusher.hEvent);
3537	pGVM->vmmr0.s.LogFlusher.hEvent = NIL_RTSEMEVENT;
3538	}
3539
3540
3541	/*********************************************************************************************************************************
3542	* Assertions *
3543	*********************************************************************************************************************************/
3544
3545	/*
3546	* Jump back to ring-3 if we're the EMT and the longjmp is armed.
3547	*
3548	* @returns true if the breakpoint should be hit, false if it should be ignored.
3549	*/
3550	DECLEXPORT(bool) RTCALL RTAssertShouldPanic(void)
3551	{
3552	#if 0
3553	return true;
3554	#else
3555	PVMCC pVM = GVMMR0GetVMByEMT(NIL_RTNATIVETHREAD);
3556	if (pVM)
3557	{
3558	PVMCPUCC pVCpu = VMMGetCpu(pVM);
3559
3560	if (pVCpu)
3561	{
3562	# ifdef RT_ARCH_X86
3563	if ( pVCpu->vmm.s.CallRing3JmpBufR0.eip
3564	&& !pVCpu->vmm.s.CallRing3JmpBufR0.fInRing3Call)
3565	# else
3566	if ( pVCpu->vmm.s.CallRing3JmpBufR0.rip
3567	&& !pVCpu->vmm.s.CallRing3JmpBufR0.fInRing3Call)
3568	# endif
3569	{
3570	int rc = VMMRZCallRing3(pVM, pVCpu, VMMCALLRING3_VM_R0_ASSERTION, 0);
3571	return RT_FAILURE_NP(rc);
3572	}
3573	}
3574	}
3575	# ifdef RT_OS_LINUX
3576	return true;
3577	# else
3578	return false;
3579	# endif
3580	#endif
3581	}
3582
3583
3584	/*
3585	* Override this so we can push it up to ring-3.
3586	*/
3587	DECLEXPORT(void) RTCALL RTAssertMsg1Weak(const char pszExpr, unsigned uLine, const char pszFile, const char *pszFunction)
3588	{
3589	/*
3590	* To host kernel log/whatever.
3591	*/
3592	SUPR0Printf("!!R0-Assertion Failed!!\n"
3593	"Expression: %s\n"
3594	"Location : %s(%d) %s\n",
3595	pszExpr, pszFile, uLine, pszFunction);
3596
3597	/*
3598	* To the log.
3599	*/
3600	LogAlways(("\n!!R0-Assertion Failed!!\n"
3601	"Expression: %s\n"
3602	"Location : %s(%d) %s\n",
3603	pszExpr, pszFile, uLine, pszFunction));
3604
3605	/*
3606	* To the global VMM buffer.
3607	*/
3608	PVMCC pVM = GVMMR0GetVMByEMT(NIL_RTNATIVETHREAD);
3609	if (pVM)
3610	RTStrPrintf(pVM->vmm.s.szRing0AssertMsg1, sizeof(pVM->vmm.s.szRing0AssertMsg1),
3611	"\n!!R0-Assertion Failed!!\n"
3612	"Expression: %.*s\n"
3613	"Location : %s(%d) %s\n",
3614	sizeof(pVM->vmm.s.szRing0AssertMsg1) / 4 * 3, pszExpr,
3615	pszFile, uLine, pszFunction);
3616
3617	/*
3618	* Continue the normal way.
3619	*/
3620	RTAssertMsg1(pszExpr, uLine, pszFile, pszFunction);
3621	}
3622
3623
3624	/**
3625	* Callback for RTLogFormatV which writes to the ring-3 log port.
3626	* See PFNLOGOUTPUT() for details.
3627	*/
3628	static DECLCALLBACK(size_t) rtLogOutput(void pv, const char pachChars, size_t cbChars)
3629	{
3630	for (size_t i = 0; i < cbChars; i++)
3631	{
3632	LogAlways(("%c", pachChars[i])); NOREF(pachChars);
3633	}
3634
3635	NOREF(pv);
3636	return cbChars;
3637	}
3638
3639
3640	/*
3641	* Override this so we can push it up to ring-3.
3642	*/
3643	DECLEXPORT(void) RTCALL RTAssertMsg2WeakV(const char *pszFormat, va_list va)
3644	{
3645	va_list vaCopy;
3646
3647	/*
3648	* Push the message to the loggers.
3649	*/
3650	PRTLOGGER pLog = RTLogRelGetDefaultInstance();
3651	if (pLog)
3652	{
3653	va_copy(vaCopy, va);
3654	RTLogFormatV(rtLogOutput, pLog, pszFormat, vaCopy);
3655	va_end(vaCopy);
3656	}
3657	pLog = RTLogGetDefaultInstance(); /* Don't initialize it here... */
3658	if (pLog)
3659	{
3660	va_copy(vaCopy, va);
3661	RTLogFormatV(rtLogOutput, pLog, pszFormat, vaCopy);
3662	va_end(vaCopy);
3663	}
3664
3665	/*
3666	* Push it to the global VMM buffer.
3667	*/
3668	PVMCC pVM = GVMMR0GetVMByEMT(NIL_RTNATIVETHREAD);
3669	if (pVM)
3670	{
3671	va_copy(vaCopy, va);
3672	RTStrPrintfV(pVM->vmm.s.szRing0AssertMsg2, sizeof(pVM->vmm.s.szRing0AssertMsg2), pszFormat, vaCopy);
3673	va_end(vaCopy);
3674	}
3675
3676	/*
3677	* Continue the normal way.
3678	*/
3679	RTAssertMsg2V(pszFormat, va);
3680	}
3681

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

source: vbox/trunk/src/VBox/VMM/VMMR0/VMMR0.cpp@ 92368

Download in other formats: