VMMR0.cpp@ 92679

Last change on this file since 92679 was 92493, checked in by vboxsync, 3 years ago
VMM: Nested VMX: bugref:10092 Purge VINF_PGM_CHANGE_MODE, no longer used.
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 133.6 KB

Line
1	/* $Id: VMMR0.cpp 92493 2021-11-18 14:01:56Z 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	pGVCpu->vmmr0.s.AssertJmpBuf.pMirrorBuf = &pGVCpu->vmm.s.AssertJmpBuf;
326	pGVCpu->vmmr0.s.AssertJmpBuf.pvStackBuf = &pGVCpu->vmm.s.abAssertStack[0];
327	pGVCpu->vmmr0.s.AssertJmpBuf.cbStackBuf = sizeof(pGVCpu->vmm.s.abAssertStack);
328
329	for (size_t iLogger = 0; iLogger < RT_ELEMENTS(pGVCpu->vmmr0.s.u.aLoggers); iLogger++)
330	pGVCpu->vmmr0.s.u.aLoggers[iLogger].hEventFlushWait = NIL_RTSEMEVENT;
331	}
332
333	/*
334	* Create the loggers.
335	*/
336	return vmmR0InitLoggers(pGVM);
337	}
338
339
340	/**
341	* Initiates the R0 driver for a particular VM instance.
342	*
343	* @returns VBox status code.
344	*
345	* @param pGVM The global (ring-0) VM structure.
346	* @param uSvnRev The SVN revision of the ring-3 part.
347	* @param uBuildType Build type indicator.
348	* @thread EMT(0)
349	*/
350	static int vmmR0InitVM(PGVM pGVM, uint32_t uSvnRev, uint32_t uBuildType)
351	{
352	/*
353	* Match the SVN revisions and build type.
354	*/
355	if (uSvnRev != VMMGetSvnRev())
356	{
357	LogRel(("VMMR0InitVM: Revision mismatch, r3=%d r0=%d\n", uSvnRev, VMMGetSvnRev()));
358	SUPR0Printf("VMMR0InitVM: Revision mismatch, r3=%d r0=%d\n", uSvnRev, VMMGetSvnRev());
359	return VERR_VMM_R0_VERSION_MISMATCH;
360	}
361	if (uBuildType != vmmGetBuildType())
362	{
363	LogRel(("VMMR0InitVM: Build type mismatch, r3=%#x r0=%#x\n", uBuildType, vmmGetBuildType()));
364	SUPR0Printf("VMMR0InitVM: Build type mismatch, r3=%#x r0=%#x\n", uBuildType, vmmGetBuildType());
365	return VERR_VMM_R0_VERSION_MISMATCH;
366	}
367
368	int rc = GVMMR0ValidateGVMandEMT(pGVM, 0 /idCpu/);
369	if (RT_FAILURE(rc))
370	return rc;
371
372	/* Don't allow this to be called more than once. */
373	if (!pGVM->vmmr0.s.fCalledInitVm)
374	pGVM->vmmr0.s.fCalledInitVm = true;
375	else
376	return VERR_ALREADY_INITIALIZED;
377
378	#ifdef LOG_ENABLED
379
380	/*
381	* Register the EMT R0 logger instance for VCPU 0.
382	*/
383	PVMCPUCC pVCpu = VMCC_GET_CPU_0(pGVM);
384	if (pVCpu->vmmr0.s.u.s.Logger.pLogger)
385	{
386	# if 0 /* testing of the logger. */
387	LogCom(("vmmR0InitVM: before %p\n", RTLogDefaultInstance()));
388	LogCom(("vmmR0InitVM: pfnFlush=%p actual=%p\n", pR0Logger->Logger.pfnFlush, vmmR0LoggerFlush));
389	LogCom(("vmmR0InitVM: pfnLogger=%p actual=%p\n", pR0Logger->Logger.pfnLogger, vmmR0LoggerWrapper));
390	LogCom(("vmmR0InitVM: offScratch=%d fFlags=%#x fDestFlags=%#x\n", pR0Logger->Logger.offScratch, pR0Logger->Logger.fFlags, pR0Logger->Logger.fDestFlags));
391
392	RTLogSetDefaultInstanceThread(&pR0Logger->Logger, (uintptr_t)pGVM->pSession);
393	LogCom(("vmmR0InitVM: after %p reg\n", RTLogDefaultInstance()));
394	RTLogSetDefaultInstanceThread(NULL, pGVM->pSession);
395	LogCom(("vmmR0InitVM: after %p dereg\n", RTLogDefaultInstance()));
396
397	pR0Logger->Logger.pfnLogger("hello ring-0 logger\n");
398	LogCom(("vmmR0InitVM: returned successfully from direct logger call.\n"));
399	pR0Logger->Logger.pfnFlush(&pR0Logger->Logger);
400	LogCom(("vmmR0InitVM: returned successfully from direct flush call.\n"));
401
402	RTLogSetDefaultInstanceThread(&pR0Logger->Logger, (uintptr_t)pGVM->pSession);
403	LogCom(("vmmR0InitVM: after %p reg2\n", RTLogDefaultInstance()));
404	pR0Logger->Logger.pfnLogger("hello ring-0 logger\n");
405	LogCom(("vmmR0InitVM: returned successfully from direct logger call (2). offScratch=%d\n", pR0Logger->Logger.offScratch));
406	RTLogSetDefaultInstanceThread(NULL, pGVM->pSession);
407	LogCom(("vmmR0InitVM: after %p dereg2\n", RTLogDefaultInstance()));
408
409	RTLogLoggerEx(&pR0Logger->Logger, 0, ~0U, "hello ring-0 logger (RTLogLoggerEx)\n");
410	LogCom(("vmmR0InitVM: RTLogLoggerEx returned fine offScratch=%d\n", pR0Logger->Logger.offScratch));
411
412	RTLogSetDefaultInstanceThread(&pR0Logger->Logger, (uintptr_t)pGVM->pSession);
413	RTLogPrintf("hello ring-0 logger (RTLogPrintf)\n");
414	LogCom(("vmmR0InitVM: RTLogPrintf returned fine offScratch=%d\n", pR0Logger->Logger.offScratch));
415	# endif
416	# ifdef VBOX_WITH_R0_LOGGING
417	Log(("Switching to per-thread logging instance %p (key=%p)\n", pVCpu->vmmr0.s.u.s.Logger.pLogger, pGVM->pSession));
418	RTLogSetDefaultInstanceThread(pVCpu->vmmr0.s.u.s.Logger.pLogger, (uintptr_t)pGVM->pSession);
419	pVCpu->vmmr0.s.u.s.Logger.fRegistered = true;
420	# endif
421	}
422	#endif /* LOG_ENABLED */
423
424	/*
425	* Check if the host supports high resolution timers or not.
426	*/
427	if ( pGVM->vmm.s.fUsePeriodicPreemptionTimers
428	&& !RTTimerCanDoHighResolution())
429	pGVM->vmm.s.fUsePeriodicPreemptionTimers = false;
430
431	/*
432	* Initialize the per VM data for GVMM and GMM.
433	*/
434	rc = GVMMR0InitVM(pGVM);
435	if (RT_SUCCESS(rc))
436	{
437	/*
438	* Init HM, CPUM and PGM (Darwin only).
439	*/
440	rc = HMR0InitVM(pGVM);
441	if (RT_SUCCESS(rc))
442	{
443	rc = CPUMR0InitVM(pGVM);
444	if (RT_SUCCESS(rc))
445	{
446	rc = PGMR0InitVM(pGVM);
447	if (RT_SUCCESS(rc))
448	{
449	rc = EMR0InitVM(pGVM);
450	if (RT_SUCCESS(rc))
451	{
452	#ifdef VBOX_WITH_PCI_PASSTHROUGH
453	rc = PciRawR0InitVM(pGVM);
454	#endif
455	if (RT_SUCCESS(rc))
456	{
457	rc = GIMR0InitVM(pGVM);
458	if (RT_SUCCESS(rc))
459	{
460	GVMMR0DoneInitVM(pGVM);
461
462	/*
463	* Collect a bit of info for the VM release log.
464	*/
465	pGVM->vmm.s.fIsPreemptPendingApiTrusty = RTThreadPreemptIsPendingTrusty();
466	pGVM->vmm.s.fIsPreemptPossible = RTThreadPreemptIsPossible();;
467	return rc;
468
469	/* bail out*/
470	//GIMR0TermVM(pGVM);
471	}
472	#ifdef VBOX_WITH_PCI_PASSTHROUGH
473	PciRawR0TermVM(pGVM);
474	#endif
475	}
476	}
477	}
478	}
479	HMR0TermVM(pGVM);
480	}
481	}
482
483	RTLogSetDefaultInstanceThread(NULL, (uintptr_t)pGVM->pSession);
484	return rc;
485	}
486
487
488	/**
489	* Does EMT specific VM initialization.
490	*
491	* @returns VBox status code.
492	* @param pGVM The ring-0 VM structure.
493	* @param idCpu The EMT that's calling.
494	*/
495	static int vmmR0InitVMEmt(PGVM pGVM, VMCPUID idCpu)
496	{
497	/* Paranoia (caller checked these already). */
498	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID);
499	AssertReturn(pGVM->aCpus[idCpu].hEMT == RTThreadNativeSelf(), VERR_INVALID_CPU_ID);
500
501	#if defined(LOG_ENABLED) && defined(VBOX_WITH_R0_LOGGING)
502	/*
503	* Registration of ring 0 loggers.
504	*/
505	PVMCPUCC pVCpu = &pGVM->aCpus[idCpu];
506	if ( pVCpu->vmmr0.s.u.s.Logger.pLogger
507	&& !pVCpu->vmmr0.s.u.s.Logger.fRegistered)
508	{
509	RTLogSetDefaultInstanceThread(pVCpu->vmmr0.s.u.s.Logger.pLogger, (uintptr_t)pGVM->pSession);
510	pVCpu->vmmr0.s.u.s.Logger.fRegistered = true;
511	}
512	#endif
513
514	return VINF_SUCCESS;
515	}
516
517
518
519	/**
520	* Terminates the R0 bits for a particular VM instance.
521	*
522	* This is normally called by ring-3 as part of the VM termination process, but
523	* may alternatively be called during the support driver session cleanup when
524	* the VM object is destroyed (see GVMM).
525	*
526	* @returns VBox status code.
527	*
528	* @param pGVM The global (ring-0) VM structure.
529	* @param idCpu Set to 0 if EMT(0) or NIL_VMCPUID if session cleanup
530	* thread.
531	* @thread EMT(0) or session clean up thread.
532	*/
533	VMMR0_INT_DECL(int) VMMR0TermVM(PGVM pGVM, VMCPUID idCpu)
534	{
535	/*
536	* Check EMT(0) claim if we're called from userland.
537	*/
538	if (idCpu != NIL_VMCPUID)
539	{
540	AssertReturn(idCpu == 0, VERR_INVALID_CPU_ID);
541	int rc = GVMMR0ValidateGVMandEMT(pGVM, idCpu);
542	if (RT_FAILURE(rc))
543	return rc;
544	}
545
546	#ifdef VBOX_WITH_PCI_PASSTHROUGH
547	PciRawR0TermVM(pGVM);
548	#endif
549
550	/*
551	* Tell GVMM what we're up to and check that we only do this once.
552	*/
553	if (GVMMR0DoingTermVM(pGVM))
554	{
555	GIMR0TermVM(pGVM);
556
557	/** @todo I wish to call PGMR0PhysFlushHandyPages(pGVM, &pGVM->aCpus[idCpu])
558	* here to make sure we don't leak any shared pages if we crash... */
559	HMR0TermVM(pGVM);
560	}
561
562	/*
563	* Deregister the logger for this EMT.
564	*/
565	RTLogSetDefaultInstanceThread(NULL, (uintptr_t)pGVM->pSession);
566
567	/*
568	* Start log flusher thread termination.
569	*/
570	ASMAtomicWriteBool(&pGVM->vmmr0.s.LogFlusher.fThreadShutdown, true);
571	if (pGVM->vmmr0.s.LogFlusher.hEvent != NIL_RTSEMEVENT)
572	RTSemEventSignal(pGVM->vmmr0.s.LogFlusher.hEvent);
573
574	return VINF_SUCCESS;
575	}
576
577
578	/**
579	* This is called at the end of gvmmR0CleanupVM().
580	*
581	* @param pGVM The global (ring-0) VM structure.
582	*/
583	VMMR0_INT_DECL(void) VMMR0CleanupVM(PGVM pGVM)
584	{
585	AssertCompile(NIL_RTTHREADCTXHOOK == (RTTHREADCTXHOOK)0); /* Depends on zero initialized memory working for NIL at the moment. */
586	for (VMCPUID idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
587	{
588	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
589
590	/** @todo Can we busy wait here for all thread-context hooks to be
591	* deregistered before releasing (destroying) it? Only until we find a
592	* solution for not deregistering hooks everytime we're leaving HMR0
593	* context. */
594	VMMR0ThreadCtxHookDestroyForEmt(pGVCpu);
595	}
596
597	vmmR0CleanupLoggers(pGVM);
598	}
599
600
601	/**
602	* An interrupt or unhalt force flag is set, deal with it.
603	*
604	* @returns VINF_SUCCESS (or VINF_EM_HALT).
605	* @param pVCpu The cross context virtual CPU structure.
606	* @param uMWait Result from EMMonitorWaitIsActive().
607	* @param enmInterruptibility Guest CPU interruptbility level.
608	*/
609	static int vmmR0DoHaltInterrupt(PVMCPUCC pVCpu, unsigned uMWait, CPUMINTERRUPTIBILITY enmInterruptibility)
610	{
611	Assert(!TRPMHasTrap(pVCpu));
612	Assert( enmInterruptibility > CPUMINTERRUPTIBILITY_INVALID
613	&& enmInterruptibility < CPUMINTERRUPTIBILITY_END);
614
615	/*
616	* Pending interrupts w/o any SMIs or NMIs? That the usual case.
617	*/
618	if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
619	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_SMI \| VMCPU_FF_INTERRUPT_NMI))
620	{
621	if (enmInterruptibility <= CPUMINTERRUPTIBILITY_UNRESTRAINED)
622	{
623	uint8_t u8Interrupt = 0;
624	int rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
625	Log(("vmmR0DoHaltInterrupt: CPU%d u8Interrupt=%d (%#x) rc=%Rrc\n", pVCpu->idCpu, u8Interrupt, u8Interrupt, rc));
626	if (RT_SUCCESS(rc))
627	{
628	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_UNHALT);
629
630	rc = TRPMAssertTrap(pVCpu, u8Interrupt, TRPM_HARDWARE_INT);
631	AssertRCSuccess(rc);
632	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltExec);
633	return rc;
634	}
635	}
636	}
637	/*
638	* SMI is not implemented yet, at least not here.
639	*/
640	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_SMI))
641	{
642	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #3\n", pVCpu->idCpu));
643	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
644	return VINF_EM_HALT;
645	}
646	/*
647	* NMI.
648	*/
649	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
650	{
651	if (enmInterruptibility < CPUMINTERRUPTIBILITY_NMI_INHIBIT)
652	{
653	/** @todo later. */
654	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #2 (uMWait=%u enmInt=%d)\n", pVCpu->idCpu, uMWait, enmInterruptibility));
655	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
656	return VINF_EM_HALT;
657	}
658	}
659	/*
660	* Nested-guest virtual interrupt.
661	*/
662	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NESTED_GUEST))
663	{
664	if (enmInterruptibility < CPUMINTERRUPTIBILITY_VIRT_INT_DISABLED)
665	{
666	/** @todo NSTVMX: NSTSVM: Remember, we might have to check and perform VM-exits
667	* here before injecting the virtual interrupt. See emR3ForcedActions
668	* for details. */
669	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #1 (uMWait=%u enmInt=%d)\n", pVCpu->idCpu, uMWait, enmInterruptibility));
670	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
671	return VINF_EM_HALT;
672	}
673	}
674
675	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UNHALT))
676	{
677	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltExec);
678	Log11(("vmmR0DoHaltInterrupt: CPU%d success VINF_SUCCESS (UNHALT)\n", pVCpu->idCpu));
679	return VINF_SUCCESS;
680	}
681	if (uMWait > 1)
682	{
683	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltExec);
684	Log11(("vmmR0DoHaltInterrupt: CPU%d success VINF_SUCCESS (uMWait=%u > 1)\n", pVCpu->idCpu, uMWait));
685	return VINF_SUCCESS;
686	}
687
688	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #0 (uMWait=%u enmInt=%d)\n", pVCpu->idCpu, uMWait, enmInterruptibility));
689	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
690	return VINF_EM_HALT;
691	}
692
693
694	/**
695	* This does one round of vmR3HaltGlobal1Halt().
696	*
697	* The rational here is that we'll reduce latency in interrupt situations if we
698	* don't go to ring-3 immediately on a VINF_EM_HALT (guest executed HLT or
699	* MWAIT), but do one round of blocking here instead and hope the interrupt is
700	* raised in the meanwhile.
701	*
702	* If we go to ring-3 we'll quit the inner HM/NEM loop in EM and end up in the
703	* outer loop, which will then call VMR3WaitHalted() and that in turn will do a
704	* ring-0 call (unless we're too close to a timer event). When the interrupt
705	* wakes us up, we'll return from ring-0 and EM will by instinct do a
706	* rescheduling (because of raw-mode) before it resumes the HM/NEM loop and gets
707	* back to VMMR0EntryFast().
708	*
709	* @returns VINF_SUCCESS or VINF_EM_HALT.
710	* @param pGVM The ring-0 VM structure.
711	* @param pGVCpu The ring-0 virtual CPU structure.
712	*
713	* @todo r=bird: All the blocking/waiting and EMT managment should move out of
714	* the VM module, probably to VMM. Then this would be more weird wrt
715	* parameters and statistics.
716	*/
717	static int vmmR0DoHalt(PGVM pGVM, PGVMCPU pGVCpu)
718	{
719	/*
720	* Do spin stat historization.
721	*/
722	if (++pGVCpu->vmm.s.cR0Halts & 0xff)
723	{ /* likely */ }
724	else if (pGVCpu->vmm.s.cR0HaltsSucceeded > pGVCpu->vmm.s.cR0HaltsToRing3)
725	{
726	pGVCpu->vmm.s.cR0HaltsSucceeded = 2;
727	pGVCpu->vmm.s.cR0HaltsToRing3 = 0;
728	}
729	else
730	{
731	pGVCpu->vmm.s.cR0HaltsSucceeded = 0;
732	pGVCpu->vmm.s.cR0HaltsToRing3 = 2;
733	}
734
735	/*
736	* Flags that makes us go to ring-3.
737	*/
738	uint32_t const fVmFFs = VM_FF_TM_VIRTUAL_SYNC \| VM_FF_PDM_QUEUES \| VM_FF_PDM_DMA
739	\| VM_FF_DBGF \| VM_FF_REQUEST \| VM_FF_CHECK_VM_STATE
740	\| VM_FF_RESET \| VM_FF_EMT_RENDEZVOUS \| VM_FF_PGM_NEED_HANDY_PAGES
741	\| VM_FF_PGM_NO_MEMORY \| VM_FF_DEBUG_SUSPEND;
742	uint64_t const fCpuFFs = VMCPU_FF_TIMER \| VMCPU_FF_PDM_CRITSECT \| VMCPU_FF_IEM
743	\| VMCPU_FF_REQUEST \| VMCPU_FF_DBGF \| VMCPU_FF_HM_UPDATE_CR3
744	\| VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL
745	\| VMCPU_FF_TO_R3 \| VMCPU_FF_IOM;
746
747	/*
748	* Check preconditions.
749	*/
750	unsigned const uMWait = EMMonitorWaitIsActive(pGVCpu);
751	CPUMINTERRUPTIBILITY const enmInterruptibility = CPUMGetGuestInterruptibility(pGVCpu);
752	if ( pGVCpu->vmm.s.fMayHaltInRing0
753	&& !TRPMHasTrap(pGVCpu)
754	&& ( enmInterruptibility == CPUMINTERRUPTIBILITY_UNRESTRAINED
755	\|\| uMWait > 1))
756	{
757	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
758	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
759	{
760	/*
761	* Interrupts pending already?
762	*/
763	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
764	APICUpdatePendingInterrupts(pGVCpu);
765
766	/*
767	* Flags that wake up from the halted state.
768	*/
769	uint64_t const fIntMask = VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC \| VMCPU_FF_INTERRUPT_NESTED_GUEST
770	\| VMCPU_FF_INTERRUPT_NMI \| VMCPU_FF_INTERRUPT_SMI \| VMCPU_FF_UNHALT;
771
772	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
773	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
774	ASMNopPause();
775
776	/*
777	* Check out how long till the next timer event.
778	*/
779	uint64_t u64Delta;
780	uint64_t u64GipTime = TMTimerPollGIP(pGVM, pGVCpu, &u64Delta);
781
782	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
783	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
784	{
785	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
786	APICUpdatePendingInterrupts(pGVCpu);
787
788	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
789	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
790
791	/*
792	* Wait if there is enough time to the next timer event.
793	*/
794	if (u64Delta >= pGVCpu->vmm.s.cNsSpinBlockThreshold)
795	{
796	/* If there are few other CPU cores around, we will procrastinate a
797	little before going to sleep, hoping for some device raising an
798	interrupt or similar. Though, the best thing here would be to
799	dynamically adjust the spin count according to its usfulness or
800	something... */
801	if ( pGVCpu->vmm.s.cR0HaltsSucceeded > pGVCpu->vmm.s.cR0HaltsToRing3
802	&& RTMpGetOnlineCount() >= 4)
803	{
804	/** @todo Figure out how we can skip this if it hasn't help recently...
805	* @bugref{9172#c12} */
806	uint32_t cSpinLoops = 42;
807	while (cSpinLoops-- > 0)
808	{
809	ASMNopPause();
810	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
811	APICUpdatePendingInterrupts(pGVCpu);
812	ASMNopPause();
813	if (VM_FF_IS_ANY_SET(pGVM, fVmFFs))
814	{
815	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3FromSpin);
816	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3);
817	return VINF_EM_HALT;
818	}
819	ASMNopPause();
820	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
821	{
822	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3FromSpin);
823	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3);
824	return VINF_EM_HALT;
825	}
826	ASMNopPause();
827	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
828	{
829	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltExecFromSpin);
830	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
831	}
832	ASMNopPause();
833	}
834	}
835
836	/*
837	* We have to set the state to VMCPUSTATE_STARTED_HALTED here so ring-3
838	* knows when to notify us (cannot access VMINTUSERPERVMCPU::fWait from here).
839	* After changing the state we must recheck the force flags of course.
840	*/
841	if (VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED_HALTED, VMCPUSTATE_STARTED))
842	{
843	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
844	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
845	{
846	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
847	APICUpdatePendingInterrupts(pGVCpu);
848
849	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
850	{
851	VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED, VMCPUSTATE_STARTED_HALTED);
852	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
853	}
854
855	/* Okay, block! */
856	uint64_t const u64StartSchedHalt = RTTimeNanoTS();
857	int rc = GVMMR0SchedHalt(pGVM, pGVCpu, u64GipTime);
858	uint64_t const u64EndSchedHalt = RTTimeNanoTS();
859	uint64_t const cNsElapsedSchedHalt = u64EndSchedHalt - u64StartSchedHalt;
860	Log10(("vmmR0DoHalt: CPU%d: halted %llu ns\n", pGVCpu->idCpu, cNsElapsedSchedHalt));
861
862	VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED, VMCPUSTATE_STARTED_HALTED);
863	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlock, cNsElapsedSchedHalt);
864	if ( rc == VINF_SUCCESS
865	\|\| rc == VERR_INTERRUPTED)
866	{
867	/* Keep some stats like ring-3 does. */
868	int64_t const cNsOverslept = u64EndSchedHalt - u64GipTime;
869	if (cNsOverslept > 50000)
870	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlockOverslept, cNsOverslept);
871	else if (cNsOverslept < -50000)
872	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlockInsomnia, cNsElapsedSchedHalt);
873	else
874	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlockOnTime, cNsElapsedSchedHalt);
875
876	/*
877	* Recheck whether we can resume execution or have to go to ring-3.
878	*/
879	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
880	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
881	{
882	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
883	APICUpdatePendingInterrupts(pGVCpu);
884	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
885	{
886	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltExecFromBlock);
887	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
888	}
889	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PostNoInt);
890	Log12(("vmmR0DoHalt: CPU%d post #2 - No pending interrupt\n", pGVCpu->idCpu));
891	}
892	else
893	{
894	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PostPendingFF);
895	Log12(("vmmR0DoHalt: CPU%d post #1 - Pending FF\n", pGVCpu->idCpu));
896	}
897	}
898	else
899	{
900	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3Other);
901	Log12(("vmmR0DoHalt: CPU%d GVMMR0SchedHalt failed: %Rrc\n", pGVCpu->idCpu, rc));
902	}
903	}
904	else
905	{
906	VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED, VMCPUSTATE_STARTED_HALTED);
907	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PendingFF);
908	Log12(("vmmR0DoHalt: CPU%d failed #5 - Pending FF\n", pGVCpu->idCpu));
909	}
910	}
911	else
912	{
913	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3Other);
914	Log12(("vmmR0DoHalt: CPU%d failed #4 - enmState=%d\n", pGVCpu->idCpu, VMCPU_GET_STATE(pGVCpu)));
915	}
916	}
917	else
918	{
919	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3SmallDelta);
920	Log12(("vmmR0DoHalt: CPU%d failed #3 - delta too small: %RU64\n", pGVCpu->idCpu, u64Delta));
921	}
922	}
923	else
924	{
925	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PendingFF);
926	Log12(("vmmR0DoHalt: CPU%d failed #2 - Pending FF\n", pGVCpu->idCpu));
927	}
928	}
929	else
930	{
931	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PendingFF);
932	Log12(("vmmR0DoHalt: CPU%d failed #1 - Pending FF\n", pGVCpu->idCpu));
933	}
934	}
935	else
936	{
937	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3Other);
938	Log12(("vmmR0DoHalt: CPU%d failed #0 - fMayHaltInRing0=%d TRPMHasTrap=%d enmInt=%d uMWait=%u\n",
939	pGVCpu->idCpu, pGVCpu->vmm.s.fMayHaltInRing0, TRPMHasTrap(pGVCpu), enmInterruptibility, uMWait));
940	}
941
942	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3);
943	return VINF_EM_HALT;
944	}
945
946
947	/**
948	* VMM ring-0 thread-context callback.
949	*
950	* This does common HM state updating and calls the HM-specific thread-context
951	* callback.
952	*
953	* This is used together with RTThreadCtxHookCreate() on platforms which
954	* supports it, and directly from VMMR0EmtPrepareForBlocking() and
955	* VMMR0EmtResumeAfterBlocking() on platforms which don't.
956	*
957	* @param enmEvent The thread-context event.
958	* @param pvUser Opaque pointer to the VMCPU.
959	*
960	* @thread EMT(pvUser)
961	*/
962	static DECLCALLBACK(void) vmmR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, void *pvUser)
963	{
964	PVMCPUCC pVCpu = (PVMCPUCC)pvUser;
965
966	switch (enmEvent)
967	{
968	case RTTHREADCTXEVENT_IN:
969	{
970	/*
971	* Linux may call us with preemption enabled (really!) but technically we
972	* cannot get preempted here, otherwise we end up in an infinite recursion
973	* scenario (i.e. preempted in resume hook -> preempt hook -> resume hook...
974	* ad infinitum). Let's just disable preemption for now...
975	*/
976	/** @todo r=bird: I don't believe the above. The linux code is clearly enabling
977	* preemption after doing the callout (one or two functions up the
978	* call chain). */
979	/** @todo r=ramshankar: See @bugref{5313#c30}. */
980	RTTHREADPREEMPTSTATE ParanoidPreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
981	RTThreadPreemptDisable(&ParanoidPreemptState);
982
983	/* We need to update the VCPU <-> host CPU mapping. */
984	RTCPUID idHostCpu;
985	uint32_t iHostCpuSet = RTMpCurSetIndexAndId(&idHostCpu);
986	pVCpu->iHostCpuSet = iHostCpuSet;
987	ASMAtomicWriteU32(&pVCpu->idHostCpu, idHostCpu);
988
989	/* In the very unlikely event that the GIP delta for the CPU we're
990	rescheduled needs calculating, try force a return to ring-3.
991	We unfortunately cannot do the measurements right here. */
992	if (RT_LIKELY(!SUPIsTscDeltaAvailableForCpuSetIndex(iHostCpuSet)))
993	{ /* likely */ }
994	else
995	VMCPU_FF_SET(pVCpu, VMCPU_FF_TO_R3);
996
997	/* Invoke the HM-specific thread-context callback. */
998	HMR0ThreadCtxCallback(enmEvent, pvUser);
999
1000	/* Restore preemption. */
1001	RTThreadPreemptRestore(&ParanoidPreemptState);
1002	break;
1003	}
1004
1005	case RTTHREADCTXEVENT_OUT:
1006	{
1007	/* Invoke the HM-specific thread-context callback. */
1008	HMR0ThreadCtxCallback(enmEvent, pvUser);
1009
1010	/*
1011	* Sigh. See VMMGetCpu() used by VMCPU_ASSERT_EMT(). We cannot let several VCPUs
1012	* have the same host CPU associated with it.
1013	*/
1014	pVCpu->iHostCpuSet = UINT32_MAX;
1015	ASMAtomicWriteU32(&pVCpu->idHostCpu, NIL_RTCPUID);
1016	break;
1017	}
1018
1019	default:
1020	/* Invoke the HM-specific thread-context callback. */
1021	HMR0ThreadCtxCallback(enmEvent, pvUser);
1022	break;
1023	}
1024	}
1025
1026
1027	/**
1028	* Creates thread switching hook for the current EMT thread.
1029	*
1030	* This is called by GVMMR0CreateVM and GVMMR0RegisterVCpu. If the host
1031	* platform does not implement switcher hooks, no hooks will be create and the
1032	* member set to NIL_RTTHREADCTXHOOK.
1033	*
1034	* @returns VBox status code.
1035	* @param pVCpu The cross context virtual CPU structure.
1036	* @thread EMT(pVCpu)
1037	*/
1038	VMMR0_INT_DECL(int) VMMR0ThreadCtxHookCreateForEmt(PVMCPUCC pVCpu)
1039	{
1040	VMCPU_ASSERT_EMT(pVCpu);
1041	Assert(pVCpu->vmmr0.s.hCtxHook == NIL_RTTHREADCTXHOOK);
1042
1043	#if 1 /* To disable this stuff change to zero. */
1044	int rc = RTThreadCtxHookCreate(&pVCpu->vmmr0.s.hCtxHook, 0, vmmR0ThreadCtxCallback, pVCpu);
1045	if (RT_SUCCESS(rc))
1046	{
1047	pVCpu->pGVM->vmm.s.fIsUsingContextHooks = true;
1048	return rc;
1049	}
1050	#else
1051	RT_NOREF(vmmR0ThreadCtxCallback);
1052	int rc = VERR_NOT_SUPPORTED;
1053	#endif
1054
1055	pVCpu->vmmr0.s.hCtxHook = NIL_RTTHREADCTXHOOK;
1056	pVCpu->pGVM->vmm.s.fIsUsingContextHooks = false;
1057	if (rc == VERR_NOT_SUPPORTED)
1058	return VINF_SUCCESS;
1059
1060	LogRelMax(32, ("RTThreadCtxHookCreate failed! rc=%Rrc pVCpu=%p idCpu=%RU32\n", rc, pVCpu, pVCpu->idCpu));
1061	return VINF_SUCCESS; /* Just ignore it, we can live without context hooks. */
1062	}
1063
1064
1065	/**
1066	* Destroys the thread switching hook for the specified VCPU.
1067	*
1068	* @param pVCpu The cross context virtual CPU structure.
1069	* @remarks Can be called from any thread.
1070	*/
1071	VMMR0_INT_DECL(void) VMMR0ThreadCtxHookDestroyForEmt(PVMCPUCC pVCpu)
1072	{
1073	int rc = RTThreadCtxHookDestroy(pVCpu->vmmr0.s.hCtxHook);
1074	AssertRC(rc);
1075	pVCpu->vmmr0.s.hCtxHook = NIL_RTTHREADCTXHOOK;
1076	}
1077
1078
1079	/**
1080	* Disables the thread switching hook for this VCPU (if we got one).
1081	*
1082	* @param pVCpu The cross context virtual CPU structure.
1083	* @thread EMT(pVCpu)
1084	*
1085	* @remarks This also clears GVMCPU::idHostCpu, so the mapping is invalid after
1086	* this call. This means you have to be careful with what you do!
1087	*/
1088	VMMR0_INT_DECL(void) VMMR0ThreadCtxHookDisable(PVMCPUCC pVCpu)
1089	{
1090	/*
1091	* Clear the VCPU <-> host CPU mapping as we've left HM context.
1092	* @bugref{7726#c19} explains the need for this trick:
1093	*
1094	* VMXR0CallRing3Callback/SVMR0CallRing3Callback &
1095	* hmR0VmxLeaveSession/hmR0SvmLeaveSession disables context hooks during
1096	* longjmp & normal return to ring-3, which opens a window where we may be
1097	* rescheduled without changing GVMCPUID::idHostCpu and cause confusion if
1098	* the CPU starts executing a different EMT. Both functions first disables
1099	* preemption and then calls HMR0LeaveCpu which invalids idHostCpu, leaving
1100	* an opening for getting preempted.
1101	*/
1102	/** @todo Make HM not need this API! Then we could leave the hooks enabled
1103	* all the time. */
1104
1105	/*
1106	* Disable the context hook, if we got one.
1107	*/
1108	if (pVCpu->vmmr0.s.hCtxHook != NIL_RTTHREADCTXHOOK)
1109	{
1110	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1111	ASMAtomicWriteU32(&pVCpu->idHostCpu, NIL_RTCPUID);
1112	int rc = RTThreadCtxHookDisable(pVCpu->vmmr0.s.hCtxHook);
1113	AssertRC(rc);
1114	}
1115	}
1116
1117
1118	/**
1119	* Internal version of VMMR0ThreadCtxHooksAreRegistered.
1120	*
1121	* @returns true if registered, false otherwise.
1122	* @param pVCpu The cross context virtual CPU structure.
1123	*/
1124	DECLINLINE(bool) vmmR0ThreadCtxHookIsEnabled(PVMCPUCC pVCpu)
1125	{
1126	return RTThreadCtxHookIsEnabled(pVCpu->vmmr0.s.hCtxHook);
1127	}
1128
1129
1130	/**
1131	* Whether thread-context hooks are registered for this VCPU.
1132	*
1133	* @returns true if registered, false otherwise.
1134	* @param pVCpu The cross context virtual CPU structure.
1135	*/
1136	VMMR0_INT_DECL(bool) VMMR0ThreadCtxHookIsEnabled(PVMCPUCC pVCpu)
1137	{
1138	return vmmR0ThreadCtxHookIsEnabled(pVCpu);
1139	}
1140
1141
1142	/**
1143	* Returns the ring-0 release logger instance.
1144	*
1145	* @returns Pointer to release logger, NULL if not configured.
1146	* @param pVCpu The cross context virtual CPU structure of the caller.
1147	* @thread EMT(pVCpu)
1148	*/
1149	VMMR0_INT_DECL(PRTLOGGER) VMMR0GetReleaseLogger(PVMCPUCC pVCpu)
1150	{
1151	return pVCpu->vmmr0.s.u.s.RelLogger.pLogger;
1152	}
1153
1154
1155	#ifdef VBOX_WITH_STATISTICS
1156	/**
1157	* Record return code statistics
1158	* @param pVM The cross context VM structure.
1159	* @param pVCpu The cross context virtual CPU structure.
1160	* @param rc The status code.
1161	*/
1162	static void vmmR0RecordRC(PVMCC pVM, PVMCPUCC pVCpu, int rc)
1163	{
1164	/*
1165	* Collect statistics.
1166	*/
1167	switch (rc)
1168	{
1169	case VINF_SUCCESS:
1170	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetNormal);
1171	break;
1172	case VINF_EM_RAW_INTERRUPT:
1173	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetInterrupt);
1174	break;
1175	case VINF_EM_RAW_INTERRUPT_HYPER:
1176	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetInterruptHyper);
1177	break;
1178	case VINF_EM_RAW_GUEST_TRAP:
1179	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetGuestTrap);
1180	break;
1181	case VINF_EM_RAW_RING_SWITCH:
1182	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetRingSwitch);
1183	break;
1184	case VINF_EM_RAW_RING_SWITCH_INT:
1185	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetRingSwitchInt);
1186	break;
1187	case VINF_EM_RAW_STALE_SELECTOR:
1188	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetStaleSelector);
1189	break;
1190	case VINF_EM_RAW_IRET_TRAP:
1191	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIRETTrap);
1192	break;
1193	case VINF_IOM_R3_IOPORT_READ:
1194	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIORead);
1195	break;
1196	case VINF_IOM_R3_IOPORT_WRITE:
1197	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIOWrite);
1198	break;
1199	case VINF_IOM_R3_IOPORT_COMMIT_WRITE:
1200	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIOCommitWrite);
1201	break;
1202	case VINF_IOM_R3_MMIO_READ:
1203	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIORead);
1204	break;
1205	case VINF_IOM_R3_MMIO_WRITE:
1206	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOWrite);
1207	break;
1208	case VINF_IOM_R3_MMIO_COMMIT_WRITE:
1209	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOCommitWrite);
1210	break;
1211	case VINF_IOM_R3_MMIO_READ_WRITE:
1212	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOReadWrite);
1213	break;
1214	case VINF_PATM_HC_MMIO_PATCH_READ:
1215	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOPatchRead);
1216	break;
1217	case VINF_PATM_HC_MMIO_PATCH_WRITE:
1218	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOPatchWrite);
1219	break;
1220	case VINF_CPUM_R3_MSR_READ:
1221	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMSRRead);
1222	break;
1223	case VINF_CPUM_R3_MSR_WRITE:
1224	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMSRWrite);
1225	break;
1226	case VINF_EM_RAW_EMULATE_INSTR:
1227	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetEmulate);
1228	break;
1229	case VINF_PATCH_EMULATE_INSTR:
1230	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchEmulate);
1231	break;
1232	case VINF_EM_RAW_EMULATE_INSTR_LDT_FAULT:
1233	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetLDTFault);
1234	break;
1235	case VINF_EM_RAW_EMULATE_INSTR_GDT_FAULT:
1236	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetGDTFault);
1237	break;
1238	case VINF_EM_RAW_EMULATE_INSTR_IDT_FAULT:
1239	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIDTFault);
1240	break;
1241	case VINF_EM_RAW_EMULATE_INSTR_TSS_FAULT:
1242	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetTSSFault);
1243	break;
1244	case VINF_CSAM_PENDING_ACTION:
1245	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetCSAMTask);
1246	break;
1247	case VINF_PGM_SYNC_CR3:
1248	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetSyncCR3);
1249	break;
1250	case VINF_PATM_PATCH_INT3:
1251	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchInt3);
1252	break;
1253	case VINF_PATM_PATCH_TRAP_PF:
1254	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchPF);
1255	break;
1256	case VINF_PATM_PATCH_TRAP_GP:
1257	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchGP);
1258	break;
1259	case VINF_PATM_PENDING_IRQ_AFTER_IRET:
1260	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchIretIRQ);
1261	break;
1262	case VINF_EM_RESCHEDULE_REM:
1263	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetRescheduleREM);
1264	break;
1265	case VINF_EM_RAW_TO_R3:
1266	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Total);
1267	if (VM_FF_IS_SET(pVM, VM_FF_TM_VIRTUAL_SYNC))
1268	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3TMVirt);
1269	else if (VM_FF_IS_SET(pVM, VM_FF_PGM_NEED_HANDY_PAGES))
1270	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3HandyPages);
1271	else if (VM_FF_IS_SET(pVM, VM_FF_PDM_QUEUES))
1272	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3PDMQueues);
1273	else if (VM_FF_IS_SET(pVM, VM_FF_EMT_RENDEZVOUS))
1274	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Rendezvous);
1275	else if (VM_FF_IS_SET(pVM, VM_FF_PDM_DMA))
1276	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3DMA);
1277	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TIMER))
1278	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Timer);
1279	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PDM_CRITSECT))
1280	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3CritSect);
1281	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TO_R3))
1282	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3FF);
1283	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_IEM))
1284	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Iem);
1285	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_IOM))
1286	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Iom);
1287	else
1288	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Unknown);
1289	break;
1290
1291	case VINF_EM_RAW_TIMER_PENDING:
1292	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetTimerPending);
1293	break;
1294	case VINF_EM_RAW_INTERRUPT_PENDING:
1295	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetInterruptPending);
1296	break;
1297	case VINF_PATM_DUPLICATE_FUNCTION:
1298	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPATMDuplicateFn);
1299	break;
1300	case VINF_PGM_POOL_FLUSH_PENDING:
1301	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPGMFlushPending);
1302	break;
1303	case VINF_EM_PENDING_REQUEST:
1304	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPendingRequest);
1305	break;
1306	case VINF_EM_HM_PATCH_TPR_INSTR:
1307	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchTPR);
1308	break;
1309	default:
1310	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMisc);
1311	break;
1312	}
1313	}
1314	#endif /* VBOX_WITH_STATISTICS */
1315
1316
1317	/**
1318	* The Ring 0 entry point, called by the fast-ioctl path.
1319	*
1320	* @param pGVM The global (ring-0) VM structure.
1321	* @param pVMIgnored The cross context VM structure. The return code is
1322	* stored in pVM->vmm.s.iLastGZRc.
1323	* @param idCpu The Virtual CPU ID of the calling EMT.
1324	* @param enmOperation Which operation to execute.
1325	* @remarks Assume called with interrupts _enabled_.
1326	*/
1327	VMMR0DECL(void) VMMR0EntryFast(PGVM pGVM, PVMCC pVMIgnored, VMCPUID idCpu, VMMR0OPERATION enmOperation)
1328	{
1329	RT_NOREF(pVMIgnored);
1330
1331	/*
1332	* Validation.
1333	*/
1334	if ( idCpu < pGVM->cCpus
1335	&& pGVM->cCpus == pGVM->cCpusUnsafe)
1336	{ /likely/ }
1337	else
1338	{
1339	SUPR0Printf("VMMR0EntryFast: Bad idCpu=%#x cCpus=%#x cCpusUnsafe=%#x\n", idCpu, pGVM->cCpus, pGVM->cCpusUnsafe);
1340	return;
1341	}
1342
1343	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
1344	RTNATIVETHREAD const hNativeThread = RTThreadNativeSelf();
1345	if (RT_LIKELY( pGVCpu->hEMT == hNativeThread
1346	&& pGVCpu->hNativeThreadR0 == hNativeThread))
1347	{ /* likely */ }
1348	else
1349	{
1350	SUPR0Printf("VMMR0EntryFast: Bad thread idCpu=%#x hNativeSelf=%p pGVCpu->hEmt=%p pGVCpu->hNativeThreadR0=%p\n",
1351	idCpu, hNativeThread, pGVCpu->hEMT, pGVCpu->hNativeThreadR0);
1352	return;
1353	}
1354
1355	/*
1356	* Perform requested operation.
1357	*/
1358	switch (enmOperation)
1359	{
1360	/*
1361	* Run guest code using the available hardware acceleration technology.
1362	*/
1363	case VMMR0_DO_HM_RUN:
1364	{
1365	for (;;) /* hlt loop */
1366	{
1367	/*
1368	* Disable ring-3 calls & blocking till we've successfully entered HM.
1369	* Otherwise we sometimes end up blocking at the finall Log4 statement
1370	* in VMXR0Enter, while still in a somewhat inbetween state.
1371	*/
1372	VMMRZCallRing3Disable(pGVCpu);
1373
1374	/*
1375	* Disable preemption.
1376	*/
1377	Assert(!vmmR0ThreadCtxHookIsEnabled(pGVCpu));
1378	RTTHREADPREEMPTSTATE PreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
1379	RTThreadPreemptDisable(&PreemptState);
1380	pGVCpu->vmmr0.s.pPreemptState = &PreemptState;
1381
1382	/*
1383	* Get the host CPU identifiers, make sure they are valid and that
1384	* we've got a TSC delta for the CPU.
1385	*/
1386	RTCPUID idHostCpu;
1387	uint32_t iHostCpuSet = RTMpCurSetIndexAndId(&idHostCpu);
1388	if (RT_LIKELY( iHostCpuSet < RTCPUSET_MAX_CPUS
1389	&& SUPIsTscDeltaAvailableForCpuSetIndex(iHostCpuSet)))
1390	{
1391	pGVCpu->iHostCpuSet = iHostCpuSet;
1392	ASMAtomicWriteU32(&pGVCpu->idHostCpu, idHostCpu);
1393
1394	/*
1395	* Update the periodic preemption timer if it's active.
1396	*/
1397	if (pGVM->vmm.s.fUsePeriodicPreemptionTimers)
1398	GVMMR0SchedUpdatePeriodicPreemptionTimer(pGVM, pGVCpu->idHostCpu, TMCalcHostTimerFrequency(pGVM, pGVCpu));
1399
1400	#ifdef VMM_R0_TOUCH_FPU
1401	/*
1402	* Make sure we've got the FPU state loaded so and we don't need to clear
1403	* CR0.TS and get out of sync with the host kernel when loading the guest
1404	* FPU state. @ref sec_cpum_fpu (CPUM.cpp) and @bugref{4053}.
1405	*/
1406	CPUMR0TouchHostFpu();
1407	#endif
1408	int rc;
1409	bool fPreemptRestored = false;
1410	if (!HMR0SuspendPending())
1411	{
1412	/*
1413	* Enable the context switching hook.
1414	*/
1415	if (pGVCpu->vmmr0.s.hCtxHook != NIL_RTTHREADCTXHOOK)
1416	{
1417	Assert(!RTThreadCtxHookIsEnabled(pGVCpu->vmmr0.s.hCtxHook));
1418	int rc2 = RTThreadCtxHookEnable(pGVCpu->vmmr0.s.hCtxHook); AssertRC(rc2);
1419	}
1420
1421	/*
1422	* Enter HM context.
1423	*/
1424	rc = HMR0Enter(pGVCpu);
1425	if (RT_SUCCESS(rc))
1426	{
1427	VMCPU_SET_STATE(pGVCpu, VMCPUSTATE_STARTED_HM);
1428
1429	/*
1430	* When preemption hooks are in place, enable preemption now that
1431	* we're in HM context.
1432	*/
1433	if (vmmR0ThreadCtxHookIsEnabled(pGVCpu))
1434	{
1435	fPreemptRestored = true;
1436	pGVCpu->vmmr0.s.pPreemptState = NULL;
1437	RTThreadPreemptRestore(&PreemptState);
1438	}
1439	VMMRZCallRing3Enable(pGVCpu);
1440
1441	/*
1442	* Setup the longjmp machinery and execute guest code (calls HMR0RunGuestCode).
1443	*/
1444	rc = vmmR0CallRing3SetJmp(&pGVCpu->vmmr0.s.AssertJmpBuf, HMR0RunGuestCode, pGVM, pGVCpu);
1445
1446	/*
1447	* Assert sanity on the way out. Using manual assertions code here as normal
1448	* assertions are going to panic the host since we're outside the setjmp/longjmp zone.
1449	*/
1450	if (RT_UNLIKELY( VMCPU_GET_STATE(pGVCpu) != VMCPUSTATE_STARTED_HM
1451	&& RT_SUCCESS_NP(rc)
1452	&& rc != VERR_VMM_RING0_ASSERTION ))
1453	{
1454	pGVM->vmm.s.szRing0AssertMsg1[0] = '\0';
1455	RTStrPrintf(pGVM->vmm.s.szRing0AssertMsg2, sizeof(pGVM->vmm.s.szRing0AssertMsg2),
1456	"Got VMCPU state %d expected %d.\n", VMCPU_GET_STATE(pGVCpu), VMCPUSTATE_STARTED_HM);
1457	rc = VERR_VMM_WRONG_HM_VMCPU_STATE;
1458	}
1459	#if 0
1460	/** @todo Get rid of this. HM shouldn't disable the context hook. */
1461	else if (RT_UNLIKELY(vmmR0ThreadCtxHookIsEnabled(pGVCpu)))
1462	{
1463	pGVM->vmm.s.szRing0AssertMsg1[0] = '\0';
1464	RTStrPrintf(pGVM->vmm.s.szRing0AssertMsg2, sizeof(pGVM->vmm.s.szRing0AssertMsg2),
1465	"Thread-context hooks still enabled! VCPU=%p Id=%u rc=%d.\n", pGVCpu, pGVCpu->idCpu, rc);
1466	rc = VERR_VMM_CONTEXT_HOOK_STILL_ENABLED;
1467	}
1468	#endif
1469
1470	VMMRZCallRing3Disable(pGVCpu); /* Lazy bird: Simpler just disabling it again... */
1471	VMCPU_SET_STATE(pGVCpu, VMCPUSTATE_STARTED);
1472	}
1473	STAM_COUNTER_INC(&pGVM->vmm.s.StatRunGC);
1474
1475	/*
1476	* Invalidate the host CPU identifiers before we disable the context
1477	* hook / restore preemption.
1478	*/
1479	pGVCpu->iHostCpuSet = UINT32_MAX;
1480	ASMAtomicWriteU32(&pGVCpu->idHostCpu, NIL_RTCPUID);
1481
1482	/*
1483	* Disable context hooks. Due to unresolved cleanup issues, we
1484	* cannot leave the hooks enabled when we return to ring-3.
1485	*
1486	* Note! At the moment HM may also have disabled the hook
1487	* when we get here, but the IPRT API handles that.
1488	*/
1489	if (pGVCpu->vmmr0.s.hCtxHook != NIL_RTTHREADCTXHOOK)
1490	RTThreadCtxHookDisable(pGVCpu->vmmr0.s.hCtxHook);
1491	}
1492	/*
1493	* The system is about to go into suspend mode; go back to ring 3.
1494	*/
1495	else
1496	{
1497	pGVCpu->iHostCpuSet = UINT32_MAX;
1498	ASMAtomicWriteU32(&pGVCpu->idHostCpu, NIL_RTCPUID);
1499	rc = VINF_EM_RAW_INTERRUPT;
1500	}
1501
1502	/** @todo When HM stops messing with the context hook state, we'll disable
1503	* preemption again before the RTThreadCtxHookDisable call. */
1504	if (!fPreemptRestored)
1505	{
1506	pGVCpu->vmmr0.s.pPreemptState = NULL;
1507	RTThreadPreemptRestore(&PreemptState);
1508	}
1509
1510	pGVCpu->vmm.s.iLastGZRc = rc;
1511
1512	/* Fire dtrace probe and collect statistics. */
1513	VBOXVMM_R0_VMM_RETURN_TO_RING3_HM(pGVCpu, CPUMQueryGuestCtxPtr(pGVCpu), rc);
1514	#ifdef VBOX_WITH_STATISTICS
1515	vmmR0RecordRC(pGVM, pGVCpu, rc);
1516	#endif
1517	VMMRZCallRing3Enable(pGVCpu);
1518
1519	/*
1520	* If this is a halt.
1521	*/
1522	if (rc != VINF_EM_HALT)
1523	{ /* we're not in a hurry for a HLT, so prefer this path */ }
1524	else
1525	{
1526	pGVCpu->vmm.s.iLastGZRc = rc = vmmR0DoHalt(pGVM, pGVCpu);
1527	if (rc == VINF_SUCCESS)
1528	{
1529	pGVCpu->vmm.s.cR0HaltsSucceeded++;
1530	continue;
1531	}
1532	pGVCpu->vmm.s.cR0HaltsToRing3++;
1533	}
1534	}
1535	/*
1536	* Invalid CPU set index or TSC delta in need of measuring.
1537	*/
1538	else
1539	{
1540	pGVCpu->vmmr0.s.pPreemptState = NULL;
1541	pGVCpu->iHostCpuSet = UINT32_MAX;
1542	ASMAtomicWriteU32(&pGVCpu->idHostCpu, NIL_RTCPUID);
1543	RTThreadPreemptRestore(&PreemptState);
1544
1545	VMMRZCallRing3Enable(pGVCpu);
1546
1547	if (iHostCpuSet < RTCPUSET_MAX_CPUS)
1548	{
1549	int rc = SUPR0TscDeltaMeasureBySetIndex(pGVM->pSession, iHostCpuSet, 0 /fFlags/,
1550	2 /cMsWaitRetry/, 5RT_MS_1SEC /cMsWaitThread*/,
1551	0 /default cTries/);
1552	if (RT_SUCCESS(rc) \|\| rc == VERR_CPU_OFFLINE)
1553	pGVCpu->vmm.s.iLastGZRc = VINF_EM_RAW_TO_R3;
1554	else
1555	pGVCpu->vmm.s.iLastGZRc = rc;
1556	}
1557	else
1558	pGVCpu->vmm.s.iLastGZRc = VERR_INVALID_CPU_INDEX;
1559	}
1560	break;
1561	} /* halt loop. */
1562	break;
1563	}
1564
1565	#ifdef VBOX_WITH_NEM_R0
1566	# if defined(RT_ARCH_AMD64) && defined(RT_OS_WINDOWS)
1567	case VMMR0_DO_NEM_RUN:
1568	{
1569	/*
1570	* Setup the longjmp machinery and execute guest code (calls NEMR0RunGuestCode).
1571	*/
1572	# ifdef VBOXSTRICTRC_STRICT_ENABLED
1573	int rc = vmmR0CallRing3SetJmp2(&pGVCpu->vmmr0.s.AssertJmpBuf, (PFNVMMR0SETJMP2)NEMR0RunGuestCode, pGVM, idCpu);
1574	# else
1575	int rc = vmmR0CallRing3SetJmp2(&pGVCpu->vmmr0.s.AssertJmpBuf, NEMR0RunGuestCode, pGVM, idCpu);
1576	# endif
1577	STAM_COUNTER_INC(&pGVM->vmm.s.StatRunGC);
1578
1579	pGVCpu->vmm.s.iLastGZRc = rc;
1580
1581	/*
1582	* Fire dtrace probe and collect statistics.
1583	*/
1584	VBOXVMM_R0_VMM_RETURN_TO_RING3_NEM(pGVCpu, CPUMQueryGuestCtxPtr(pGVCpu), rc);
1585	# ifdef VBOX_WITH_STATISTICS
1586	vmmR0RecordRC(pGVM, pGVCpu, rc);
1587	# endif
1588	break;
1589	}
1590	# endif
1591	#endif
1592
1593	/*
1594	* For profiling.
1595	*/
1596	case VMMR0_DO_NOP:
1597	pGVCpu->vmm.s.iLastGZRc = VINF_SUCCESS;
1598	break;
1599
1600	/*
1601	* Shouldn't happen.
1602	*/
1603	default:
1604	AssertMsgFailed(("%#x\n", enmOperation));
1605	pGVCpu->vmm.s.iLastGZRc = VERR_NOT_SUPPORTED;
1606	break;
1607	}
1608	}
1609
1610
1611	/**
1612	* Validates a session or VM session argument.
1613	*
1614	* @returns true / false accordingly.
1615	* @param pGVM The global (ring-0) VM structure.
1616	* @param pClaimedSession The session claim to validate.
1617	* @param pSession The session argument.
1618	*/
1619	DECLINLINE(bool) vmmR0IsValidSession(PGVM pGVM, PSUPDRVSESSION pClaimedSession, PSUPDRVSESSION pSession)
1620	{
1621	/* This must be set! */
1622	if (!pSession)
1623	return false;
1624
1625	/* Only one out of the two. */
1626	if (pGVM && pClaimedSession)
1627	return false;
1628	if (pGVM)
1629	pClaimedSession = pGVM->pSession;
1630	return pClaimedSession == pSession;
1631	}
1632
1633
1634	/**
1635	* VMMR0EntryEx worker function, either called directly or when ever possible
1636	* called thru a longjmp so we can exit safely on failure.
1637	*
1638	* @returns VBox status code.
1639	* @param pGVM The global (ring-0) VM structure.
1640	* @param idCpu Virtual CPU ID argument. Must be NIL_VMCPUID if pVM
1641	* is NIL_RTR0PTR, and may be NIL_VMCPUID if it isn't
1642	* @param enmOperation Which operation to execute.
1643	* @param pReqHdr This points to a SUPVMMR0REQHDR packet. Optional.
1644	* The support driver validates this if it's present.
1645	* @param u64Arg Some simple constant argument.
1646	* @param pSession The session of the caller.
1647	*
1648	* @remarks Assume called with interrupts _enabled_.
1649	*/
1650	DECL_NO_INLINE(static, int) vmmR0EntryExWorker(PGVM pGVM, VMCPUID idCpu, VMMR0OPERATION enmOperation,
1651	PSUPVMMR0REQHDR pReqHdr, uint64_t u64Arg, PSUPDRVSESSION pSession)
1652	{
1653	/*
1654	* Validate pGVM and idCpu for consistency and validity.
1655	*/
1656	if (pGVM != NULL)
1657	{
1658	if (RT_LIKELY(((uintptr_t)pGVM & PAGE_OFFSET_MASK) == 0))
1659	{ /* likely */ }
1660	else
1661	{
1662	SUPR0Printf("vmmR0EntryExWorker: Invalid pGVM=%p! (op=%d)\n", pGVM, enmOperation);
1663	return VERR_INVALID_POINTER;
1664	}
1665
1666	if (RT_LIKELY(idCpu == NIL_VMCPUID \|\| idCpu < pGVM->cCpus))
1667	{ /* likely */ }
1668	else
1669	{
1670	SUPR0Printf("vmmR0EntryExWorker: Invalid idCpu %#x (cCpus=%#x)\n", idCpu, pGVM->cCpus);
1671	return VERR_INVALID_PARAMETER;
1672	}
1673
1674	if (RT_LIKELY( pGVM->enmVMState >= VMSTATE_CREATING
1675	&& pGVM->enmVMState <= VMSTATE_TERMINATED
1676	&& pGVM->pSession == pSession
1677	&& pGVM->pSelf == pGVM))
1678	{ /* likely */ }
1679	else
1680	{
1681	SUPR0Printf("vmmR0EntryExWorker: Invalid pGVM=%p:{.enmVMState=%d, .cCpus=%#x, .pSession=%p(==%p), .pSelf=%p(==%p)}! (op=%d)\n",
1682	pGVM, pGVM->enmVMState, pGVM->cCpus, pGVM->pSession, pSession, pGVM->pSelf, pGVM, enmOperation);
1683	return VERR_INVALID_POINTER;
1684	}
1685	}
1686	else if (RT_LIKELY(idCpu == NIL_VMCPUID))
1687	{ /* likely */ }
1688	else
1689	{
1690	SUPR0Printf("vmmR0EntryExWorker: Invalid idCpu=%u\n", idCpu);
1691	return VERR_INVALID_PARAMETER;
1692	}
1693
1694	/*
1695	* Process the request.
1696	*/
1697	int rc;
1698	switch (enmOperation)
1699	{
1700	/*
1701	* GVM requests
1702	*/
1703	case VMMR0_DO_GVMM_CREATE_VM:
1704	if (pGVM == NULL && u64Arg == 0 && idCpu == NIL_VMCPUID)
1705	rc = GVMMR0CreateVMReq((PGVMMCREATEVMREQ)pReqHdr, pSession);
1706	else
1707	rc = VERR_INVALID_PARAMETER;
1708	break;
1709
1710	case VMMR0_DO_GVMM_DESTROY_VM:
1711	if (pReqHdr == NULL && u64Arg == 0)
1712	rc = GVMMR0DestroyVM(pGVM);
1713	else
1714	rc = VERR_INVALID_PARAMETER;
1715	break;
1716
1717	case VMMR0_DO_GVMM_REGISTER_VMCPU:
1718	if (pGVM != NULL)
1719	rc = GVMMR0RegisterVCpu(pGVM, idCpu);
1720	else
1721	rc = VERR_INVALID_PARAMETER;
1722	break;
1723
1724	case VMMR0_DO_GVMM_DEREGISTER_VMCPU:
1725	if (pGVM != NULL)
1726	rc = GVMMR0DeregisterVCpu(pGVM, idCpu);
1727	else
1728	rc = VERR_INVALID_PARAMETER;
1729	break;
1730
1731	case VMMR0_DO_GVMM_REGISTER_WORKER_THREAD:
1732	if (pGVM != NULL && pReqHdr && pReqHdr->cbReq == sizeof(GVMMREGISTERWORKERTHREADREQ))
1733	rc = GVMMR0RegisterWorkerThread(pGVM, (GVMMWORKERTHREAD)(unsigned)u64Arg,
1734	((PGVMMREGISTERWORKERTHREADREQ)(pReqHdr))->hNativeThreadR3);
1735	else
1736	rc = VERR_INVALID_PARAMETER;
1737	break;
1738
1739	case VMMR0_DO_GVMM_DEREGISTER_WORKER_THREAD:
1740	if (pGVM != NULL)
1741	rc = GVMMR0DeregisterWorkerThread(pGVM, (GVMMWORKERTHREAD)(unsigned)u64Arg);
1742	else
1743	rc = VERR_INVALID_PARAMETER;
1744	break;
1745
1746	case VMMR0_DO_GVMM_SCHED_HALT:
1747	if (pReqHdr)
1748	return VERR_INVALID_PARAMETER;
1749	rc = GVMMR0SchedHaltReq(pGVM, idCpu, u64Arg);
1750	break;
1751
1752	case VMMR0_DO_GVMM_SCHED_WAKE_UP:
1753	if (pReqHdr \|\| u64Arg)
1754	return VERR_INVALID_PARAMETER;
1755	rc = GVMMR0SchedWakeUp(pGVM, idCpu);
1756	break;
1757
1758	case VMMR0_DO_GVMM_SCHED_POKE:
1759	if (pReqHdr \|\| u64Arg)
1760	return VERR_INVALID_PARAMETER;
1761	rc = GVMMR0SchedPoke(pGVM, idCpu);
1762	break;
1763
1764	case VMMR0_DO_GVMM_SCHED_WAKE_UP_AND_POKE_CPUS:
1765	if (u64Arg)
1766	return VERR_INVALID_PARAMETER;
1767	rc = GVMMR0SchedWakeUpAndPokeCpusReq(pGVM, (PGVMMSCHEDWAKEUPANDPOKECPUSREQ)pReqHdr);
1768	break;
1769
1770	case VMMR0_DO_GVMM_SCHED_POLL:
1771	if (pReqHdr \|\| u64Arg > 1)
1772	return VERR_INVALID_PARAMETER;
1773	rc = GVMMR0SchedPoll(pGVM, idCpu, !!u64Arg);
1774	break;
1775
1776	case VMMR0_DO_GVMM_QUERY_STATISTICS:
1777	if (u64Arg)
1778	return VERR_INVALID_PARAMETER;
1779	rc = GVMMR0QueryStatisticsReq(pGVM, (PGVMMQUERYSTATISTICSSREQ)pReqHdr, pSession);
1780	break;
1781
1782	case VMMR0_DO_GVMM_RESET_STATISTICS:
1783	if (u64Arg)
1784	return VERR_INVALID_PARAMETER;
1785	rc = GVMMR0ResetStatisticsReq(pGVM, (PGVMMRESETSTATISTICSSREQ)pReqHdr, pSession);
1786	break;
1787
1788	/*
1789	* Initialize the R0 part of a VM instance.
1790	*/
1791	case VMMR0_DO_VMMR0_INIT:
1792	rc = vmmR0InitVM(pGVM, RT_LODWORD(u64Arg), RT_HIDWORD(u64Arg));
1793	break;
1794
1795	/*
1796	* Does EMT specific ring-0 init.
1797	*/
1798	case VMMR0_DO_VMMR0_INIT_EMT:
1799	rc = vmmR0InitVMEmt(pGVM, idCpu);
1800	break;
1801
1802	/*
1803	* Terminate the R0 part of a VM instance.
1804	*/
1805	case VMMR0_DO_VMMR0_TERM:
1806	rc = VMMR0TermVM(pGVM, 0 /idCpu/);
1807	break;
1808
1809	/*
1810	* Update release or debug logger instances.
1811	*/
1812	case VMMR0_DO_VMMR0_UPDATE_LOGGERS:
1813	if (idCpu == NIL_VMCPUID)
1814	return VERR_INVALID_CPU_ID;
1815	if (u64Arg < VMMLOGGER_IDX_MAX && pReqHdr != NULL)
1816	rc = vmmR0UpdateLoggers(pGVM, idCpu /idCpu/, (PVMMR0UPDATELOGGERSREQ)pReqHdr, (size_t)u64Arg);
1817	else
1818	return VERR_INVALID_PARAMETER;
1819	break;
1820
1821	/*
1822	* Log flusher thread.
1823	*/
1824	case VMMR0_DO_VMMR0_LOG_FLUSHER:
1825	if (idCpu != NIL_VMCPUID)
1826	return VERR_INVALID_CPU_ID;
1827	if (pReqHdr == NULL)
1828	rc = vmmR0LogFlusher(pGVM);
1829	else
1830	return VERR_INVALID_PARAMETER;
1831	break;
1832
1833	/*
1834	* Wait for the flush to finish with all the buffers for the given logger.
1835	*/
1836	case VMMR0_DO_VMMR0_LOG_WAIT_FLUSHED:
1837	if (idCpu == NIL_VMCPUID)
1838	return VERR_INVALID_CPU_ID;
1839	if (u64Arg < VMMLOGGER_IDX_MAX && pReqHdr == NULL)
1840	rc = vmmR0LogWaitFlushed(pGVM, idCpu /idCpu/, (size_t)u64Arg);
1841	else
1842	return VERR_INVALID_PARAMETER;
1843	break;
1844
1845	/*
1846	* Attempt to enable hm mode and check the current setting.
1847	*/
1848	case VMMR0_DO_HM_ENABLE:
1849	rc = HMR0EnableAllCpus(pGVM);
1850	break;
1851
1852	/*
1853	* Setup the hardware accelerated session.
1854	*/
1855	case VMMR0_DO_HM_SETUP_VM:
1856	rc = HMR0SetupVM(pGVM);
1857	break;
1858
1859	/*
1860	* PGM wrappers.
1861	*/
1862	case VMMR0_DO_PGM_ALLOCATE_HANDY_PAGES:
1863	if (idCpu == NIL_VMCPUID)
1864	return VERR_INVALID_CPU_ID;
1865	rc = PGMR0PhysAllocateHandyPages(pGVM, idCpu);
1866	break;
1867
1868	case VMMR0_DO_PGM_FLUSH_HANDY_PAGES:
1869	if (idCpu == NIL_VMCPUID)
1870	return VERR_INVALID_CPU_ID;
1871	rc = PGMR0PhysFlushHandyPages(pGVM, idCpu);
1872	break;
1873
1874	case VMMR0_DO_PGM_ALLOCATE_LARGE_PAGE:
1875	if (idCpu == NIL_VMCPUID)
1876	return VERR_INVALID_CPU_ID;
1877	rc = PGMR0PhysAllocateLargePage(pGVM, idCpu, u64Arg);
1878	break;
1879
1880	case VMMR0_DO_PGM_PHYS_SETUP_IOMMU:
1881	if (idCpu != 0)
1882	return VERR_INVALID_CPU_ID;
1883	rc = PGMR0PhysSetupIoMmu(pGVM);
1884	break;
1885
1886	case VMMR0_DO_PGM_POOL_GROW:
1887	if (idCpu == NIL_VMCPUID)
1888	return VERR_INVALID_CPU_ID;
1889	rc = PGMR0PoolGrow(pGVM, idCpu);
1890	break;
1891
1892	/*
1893	* GMM wrappers.
1894	*/
1895	case VMMR0_DO_GMM_INITIAL_RESERVATION:
1896	if (u64Arg)
1897	return VERR_INVALID_PARAMETER;
1898	rc = GMMR0InitialReservationReq(pGVM, idCpu, (PGMMINITIALRESERVATIONREQ)pReqHdr);
1899	break;
1900
1901	case VMMR0_DO_GMM_UPDATE_RESERVATION:
1902	if (u64Arg)
1903	return VERR_INVALID_PARAMETER;
1904	rc = GMMR0UpdateReservationReq(pGVM, idCpu, (PGMMUPDATERESERVATIONREQ)pReqHdr);
1905	break;
1906
1907	case VMMR0_DO_GMM_ALLOCATE_PAGES:
1908	if (u64Arg)
1909	return VERR_INVALID_PARAMETER;
1910	rc = GMMR0AllocatePagesReq(pGVM, idCpu, (PGMMALLOCATEPAGESREQ)pReqHdr);
1911	break;
1912
1913	case VMMR0_DO_GMM_FREE_PAGES:
1914	if (u64Arg)
1915	return VERR_INVALID_PARAMETER;
1916	rc = GMMR0FreePagesReq(pGVM, idCpu, (PGMMFREEPAGESREQ)pReqHdr);
1917	break;
1918
1919	case VMMR0_DO_GMM_FREE_LARGE_PAGE:
1920	if (u64Arg)
1921	return VERR_INVALID_PARAMETER;
1922	rc = GMMR0FreeLargePageReq(pGVM, idCpu, (PGMMFREELARGEPAGEREQ)pReqHdr);
1923	break;
1924
1925	case VMMR0_DO_GMM_QUERY_HYPERVISOR_MEM_STATS:
1926	if (u64Arg)
1927	return VERR_INVALID_PARAMETER;
1928	rc = GMMR0QueryHypervisorMemoryStatsReq((PGMMMEMSTATSREQ)pReqHdr);
1929	break;
1930
1931	case VMMR0_DO_GMM_QUERY_MEM_STATS:
1932	if (idCpu == NIL_VMCPUID)
1933	return VERR_INVALID_CPU_ID;
1934	if (u64Arg)
1935	return VERR_INVALID_PARAMETER;
1936	rc = GMMR0QueryMemoryStatsReq(pGVM, idCpu, (PGMMMEMSTATSREQ)pReqHdr);
1937	break;
1938
1939	case VMMR0_DO_GMM_BALLOONED_PAGES:
1940	if (u64Arg)
1941	return VERR_INVALID_PARAMETER;
1942	rc = GMMR0BalloonedPagesReq(pGVM, idCpu, (PGMMBALLOONEDPAGESREQ)pReqHdr);
1943	break;
1944
1945	case VMMR0_DO_GMM_MAP_UNMAP_CHUNK:
1946	if (u64Arg)
1947	return VERR_INVALID_PARAMETER;
1948	rc = GMMR0MapUnmapChunkReq(pGVM, (PGMMMAPUNMAPCHUNKREQ)pReqHdr);
1949	break;
1950
1951	case VMMR0_DO_GMM_REGISTER_SHARED_MODULE:
1952	if (idCpu == NIL_VMCPUID)
1953	return VERR_INVALID_CPU_ID;
1954	if (u64Arg)
1955	return VERR_INVALID_PARAMETER;
1956	rc = GMMR0RegisterSharedModuleReq(pGVM, idCpu, (PGMMREGISTERSHAREDMODULEREQ)pReqHdr);
1957	break;
1958
1959	case VMMR0_DO_GMM_UNREGISTER_SHARED_MODULE:
1960	if (idCpu == NIL_VMCPUID)
1961	return VERR_INVALID_CPU_ID;
1962	if (u64Arg)
1963	return VERR_INVALID_PARAMETER;
1964	rc = GMMR0UnregisterSharedModuleReq(pGVM, idCpu, (PGMMUNREGISTERSHAREDMODULEREQ)pReqHdr);
1965	break;
1966
1967	case VMMR0_DO_GMM_RESET_SHARED_MODULES:
1968	if (idCpu == NIL_VMCPUID)
1969	return VERR_INVALID_CPU_ID;
1970	if ( u64Arg
1971	\|\| pReqHdr)
1972	return VERR_INVALID_PARAMETER;
1973	rc = GMMR0ResetSharedModules(pGVM, idCpu);
1974	break;
1975
1976	#ifdef VBOX_WITH_PAGE_SHARING
1977	case VMMR0_DO_GMM_CHECK_SHARED_MODULES:
1978	{
1979	if (idCpu == NIL_VMCPUID)
1980	return VERR_INVALID_CPU_ID;
1981	if ( u64Arg
1982	\|\| pReqHdr)
1983	return VERR_INVALID_PARAMETER;
1984	rc = GMMR0CheckSharedModules(pGVM, idCpu);
1985	break;
1986	}
1987	#endif
1988
1989	#if defined(VBOX_STRICT) && HC_ARCH_BITS == 64
1990	case VMMR0_DO_GMM_FIND_DUPLICATE_PAGE:
1991	if (u64Arg)
1992	return VERR_INVALID_PARAMETER;
1993	rc = GMMR0FindDuplicatePageReq(pGVM, (PGMMFINDDUPLICATEPAGEREQ)pReqHdr);
1994	break;
1995	#endif
1996
1997	case VMMR0_DO_GMM_QUERY_STATISTICS:
1998	if (u64Arg)
1999	return VERR_INVALID_PARAMETER;
2000	rc = GMMR0QueryStatisticsReq(pGVM, (PGMMQUERYSTATISTICSSREQ)pReqHdr);
2001	break;
2002
2003	case VMMR0_DO_GMM_RESET_STATISTICS:
2004	if (u64Arg)
2005	return VERR_INVALID_PARAMETER;
2006	rc = GMMR0ResetStatisticsReq(pGVM, (PGMMRESETSTATISTICSSREQ)pReqHdr);
2007	break;
2008
2009	/*
2010	* A quick GCFGM mock-up.
2011	*/
2012	/** @todo GCFGM with proper access control, ring-3 management interface and all that. */
2013	case VMMR0_DO_GCFGM_SET_VALUE:
2014	case VMMR0_DO_GCFGM_QUERY_VALUE:
2015	{
2016	if (pGVM \|\| !pReqHdr \|\| u64Arg \|\| idCpu != NIL_VMCPUID)
2017	return VERR_INVALID_PARAMETER;
2018	PGCFGMVALUEREQ pReq = (PGCFGMVALUEREQ)pReqHdr;
2019	if (pReq->Hdr.cbReq != sizeof(*pReq))
2020	return VERR_INVALID_PARAMETER;
2021	if (enmOperation == VMMR0_DO_GCFGM_SET_VALUE)
2022	{
2023	rc = GVMMR0SetConfig(pReq->pSession, &pReq->szName[0], pReq->u64Value);
2024	//if (rc == VERR_CFGM_VALUE_NOT_FOUND)
2025	// rc = GMMR0SetConfig(pReq->pSession, &pReq->szName[0], pReq->u64Value);
2026	}
2027	else
2028	{
2029	rc = GVMMR0QueryConfig(pReq->pSession, &pReq->szName[0], &pReq->u64Value);
2030	//if (rc == VERR_CFGM_VALUE_NOT_FOUND)
2031	// rc = GMMR0QueryConfig(pReq->pSession, &pReq->szName[0], &pReq->u64Value);
2032	}
2033	break;
2034	}
2035
2036	/*
2037	* PDM Wrappers.
2038	*/
2039	case VMMR0_DO_PDM_DRIVER_CALL_REQ_HANDLER:
2040	{
2041	if (!pReqHdr \|\| u64Arg \|\| idCpu != NIL_VMCPUID)
2042	return VERR_INVALID_PARAMETER;
2043	rc = PDMR0DriverCallReqHandler(pGVM, (PPDMDRIVERCALLREQHANDLERREQ)pReqHdr);
2044	break;
2045	}
2046
2047	case VMMR0_DO_PDM_DEVICE_CREATE:
2048	{
2049	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2050	return VERR_INVALID_PARAMETER;
2051	rc = PDMR0DeviceCreateReqHandler(pGVM, (PPDMDEVICECREATEREQ)pReqHdr);
2052	break;
2053	}
2054
2055	case VMMR0_DO_PDM_DEVICE_GEN_CALL:
2056	{
2057	if (!pReqHdr \|\| u64Arg)
2058	return VERR_INVALID_PARAMETER;
2059	rc = PDMR0DeviceGenCallReqHandler(pGVM, (PPDMDEVICEGENCALLREQ)pReqHdr, idCpu);
2060	break;
2061	}
2062
2063	/** @todo Remove the once all devices has been converted to new style! @bugref{9218} */
2064	case VMMR0_DO_PDM_DEVICE_COMPAT_SET_CRITSECT:
2065	{
2066	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2067	return VERR_INVALID_PARAMETER;
2068	rc = PDMR0DeviceCompatSetCritSectReqHandler(pGVM, (PPDMDEVICECOMPATSETCRITSECTREQ)pReqHdr);
2069	break;
2070	}
2071
2072	/*
2073	* Requests to the internal networking service.
2074	*/
2075	case VMMR0_DO_INTNET_OPEN:
2076	{
2077	PINTNETOPENREQ pReq = (PINTNETOPENREQ)pReqHdr;
2078	if (u64Arg \|\| !pReq \|\| !vmmR0IsValidSession(pGVM, pReq->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2079	return VERR_INVALID_PARAMETER;
2080	rc = IntNetR0OpenReq(pSession, pReq);
2081	break;
2082	}
2083
2084	case VMMR0_DO_INTNET_IF_CLOSE:
2085	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFCLOSEREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2086	return VERR_INVALID_PARAMETER;
2087	rc = IntNetR0IfCloseReq(pSession, (PINTNETIFCLOSEREQ)pReqHdr);
2088	break;
2089
2090
2091	case VMMR0_DO_INTNET_IF_GET_BUFFER_PTRS:
2092	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFGETBUFFERPTRSREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2093	return VERR_INVALID_PARAMETER;
2094	rc = IntNetR0IfGetBufferPtrsReq(pSession, (PINTNETIFGETBUFFERPTRSREQ)pReqHdr);
2095	break;
2096
2097	case VMMR0_DO_INTNET_IF_SET_PROMISCUOUS_MODE:
2098	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSETPROMISCUOUSMODEREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2099	return VERR_INVALID_PARAMETER;
2100	rc = IntNetR0IfSetPromiscuousModeReq(pSession, (PINTNETIFSETPROMISCUOUSMODEREQ)pReqHdr);
2101	break;
2102
2103	case VMMR0_DO_INTNET_IF_SET_MAC_ADDRESS:
2104	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSETMACADDRESSREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2105	return VERR_INVALID_PARAMETER;
2106	rc = IntNetR0IfSetMacAddressReq(pSession, (PINTNETIFSETMACADDRESSREQ)pReqHdr);
2107	break;
2108
2109	case VMMR0_DO_INTNET_IF_SET_ACTIVE:
2110	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSETACTIVEREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2111	return VERR_INVALID_PARAMETER;
2112	rc = IntNetR0IfSetActiveReq(pSession, (PINTNETIFSETACTIVEREQ)pReqHdr);
2113	break;
2114
2115	case VMMR0_DO_INTNET_IF_SEND:
2116	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSENDREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2117	return VERR_INVALID_PARAMETER;
2118	rc = IntNetR0IfSendReq(pSession, (PINTNETIFSENDREQ)pReqHdr);
2119	break;
2120
2121	case VMMR0_DO_INTNET_IF_WAIT:
2122	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFWAITREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2123	return VERR_INVALID_PARAMETER;
2124	rc = IntNetR0IfWaitReq(pSession, (PINTNETIFWAITREQ)pReqHdr);
2125	break;
2126
2127	case VMMR0_DO_INTNET_IF_ABORT_WAIT:
2128	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFWAITREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2129	return VERR_INVALID_PARAMETER;
2130	rc = IntNetR0IfAbortWaitReq(pSession, (PINTNETIFABORTWAITREQ)pReqHdr);
2131	break;
2132
2133	#if 0 //def VBOX_WITH_PCI_PASSTHROUGH
2134	/*
2135	* Requests to host PCI driver service.
2136	*/
2137	case VMMR0_DO_PCIRAW_REQ:
2138	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PPCIRAWSENDREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2139	return VERR_INVALID_PARAMETER;
2140	rc = PciRawR0ProcessReq(pGVM, pSession, (PPCIRAWSENDREQ)pReqHdr);
2141	break;
2142	#endif
2143
2144	/*
2145	* NEM requests.
2146	*/
2147	#ifdef VBOX_WITH_NEM_R0
2148	# if defined(RT_ARCH_AMD64) && defined(RT_OS_WINDOWS)
2149	case VMMR0_DO_NEM_INIT_VM:
2150	if (u64Arg \|\| pReqHdr \|\| idCpu != 0)
2151	return VERR_INVALID_PARAMETER;
2152	rc = NEMR0InitVM(pGVM);
2153	break;
2154
2155	case VMMR0_DO_NEM_INIT_VM_PART_2:
2156	if (u64Arg \|\| pReqHdr \|\| idCpu != 0)
2157	return VERR_INVALID_PARAMETER;
2158	rc = NEMR0InitVMPart2(pGVM);
2159	break;
2160
2161	case VMMR0_DO_NEM_MAP_PAGES:
2162	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2163	return VERR_INVALID_PARAMETER;
2164	rc = NEMR0MapPages(pGVM, idCpu);
2165	break;
2166
2167	case VMMR0_DO_NEM_UNMAP_PAGES:
2168	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2169	return VERR_INVALID_PARAMETER;
2170	rc = NEMR0UnmapPages(pGVM, idCpu);
2171	break;
2172
2173	case VMMR0_DO_NEM_EXPORT_STATE:
2174	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2175	return VERR_INVALID_PARAMETER;
2176	rc = NEMR0ExportState(pGVM, idCpu);
2177	break;
2178
2179	case VMMR0_DO_NEM_IMPORT_STATE:
2180	if (pReqHdr \|\| idCpu == NIL_VMCPUID)
2181	return VERR_INVALID_PARAMETER;
2182	rc = NEMR0ImportState(pGVM, idCpu, u64Arg);
2183	break;
2184
2185	case VMMR0_DO_NEM_QUERY_CPU_TICK:
2186	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2187	return VERR_INVALID_PARAMETER;
2188	rc = NEMR0QueryCpuTick(pGVM, idCpu);
2189	break;
2190
2191	case VMMR0_DO_NEM_RESUME_CPU_TICK_ON_ALL:
2192	if (pReqHdr \|\| idCpu == NIL_VMCPUID)
2193	return VERR_INVALID_PARAMETER;
2194	rc = NEMR0ResumeCpuTickOnAll(pGVM, idCpu, u64Arg);
2195	break;
2196
2197	case VMMR0_DO_NEM_UPDATE_STATISTICS:
2198	if (u64Arg \|\| pReqHdr)
2199	return VERR_INVALID_PARAMETER;
2200	rc = NEMR0UpdateStatistics(pGVM, idCpu);
2201	break;
2202
2203	# if 1 && defined(DEBUG_bird)
2204	case VMMR0_DO_NEM_EXPERIMENT:
2205	if (pReqHdr)
2206	return VERR_INVALID_PARAMETER;
2207	rc = NEMR0DoExperiment(pGVM, idCpu, u64Arg);
2208	break;
2209	# endif
2210	# endif
2211	#endif
2212
2213	/*
2214	* IOM requests.
2215	*/
2216	case VMMR0_DO_IOM_GROW_IO_PORTS:
2217	{
2218	if (pReqHdr \|\| idCpu != 0)
2219	return VERR_INVALID_PARAMETER;
2220	rc = IOMR0IoPortGrowRegistrationTables(pGVM, u64Arg);
2221	break;
2222	}
2223
2224	case VMMR0_DO_IOM_GROW_IO_PORT_STATS:
2225	{
2226	if (pReqHdr \|\| idCpu != 0)
2227	return VERR_INVALID_PARAMETER;
2228	rc = IOMR0IoPortGrowStatisticsTable(pGVM, u64Arg);
2229	break;
2230	}
2231
2232	case VMMR0_DO_IOM_GROW_MMIO_REGS:
2233	{
2234	if (pReqHdr \|\| idCpu != 0)
2235	return VERR_INVALID_PARAMETER;
2236	rc = IOMR0MmioGrowRegistrationTables(pGVM, u64Arg);
2237	break;
2238	}
2239
2240	case VMMR0_DO_IOM_GROW_MMIO_STATS:
2241	{
2242	if (pReqHdr \|\| idCpu != 0)
2243	return VERR_INVALID_PARAMETER;
2244	rc = IOMR0MmioGrowStatisticsTable(pGVM, u64Arg);
2245	break;
2246	}
2247
2248	case VMMR0_DO_IOM_SYNC_STATS_INDICES:
2249	{
2250	if (pReqHdr \|\| idCpu != 0)
2251	return VERR_INVALID_PARAMETER;
2252	rc = IOMR0IoPortSyncStatisticsIndices(pGVM);
2253	if (RT_SUCCESS(rc))
2254	rc = IOMR0MmioSyncStatisticsIndices(pGVM);
2255	break;
2256	}
2257
2258	/*
2259	* DBGF requests.
2260	*/
2261	#ifdef VBOX_WITH_DBGF_TRACING
2262	case VMMR0_DO_DBGF_TRACER_CREATE:
2263	{
2264	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2265	return VERR_INVALID_PARAMETER;
2266	rc = DBGFR0TracerCreateReqHandler(pGVM, (PDBGFTRACERCREATEREQ)pReqHdr);
2267	break;
2268	}
2269
2270	case VMMR0_DO_DBGF_TRACER_CALL_REQ_HANDLER:
2271	{
2272	if (!pReqHdr \|\| u64Arg)
2273	return VERR_INVALID_PARAMETER;
2274	# if 0 /** @todo */
2275	rc = DBGFR0TracerGenCallReqHandler(pGVM, (PDBGFTRACERGENCALLREQ)pReqHdr, idCpu);
2276	# else
2277	rc = VERR_NOT_IMPLEMENTED;
2278	# endif
2279	break;
2280	}
2281	#endif
2282
2283	case VMMR0_DO_DBGF_BP_INIT:
2284	{
2285	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2286	return VERR_INVALID_PARAMETER;
2287	rc = DBGFR0BpInitReqHandler(pGVM, (PDBGFBPINITREQ)pReqHdr);
2288	break;
2289	}
2290
2291	case VMMR0_DO_DBGF_BP_CHUNK_ALLOC:
2292	{
2293	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2294	return VERR_INVALID_PARAMETER;
2295	rc = DBGFR0BpChunkAllocReqHandler(pGVM, (PDBGFBPCHUNKALLOCREQ)pReqHdr);
2296	break;
2297	}
2298
2299	case VMMR0_DO_DBGF_BP_L2_TBL_CHUNK_ALLOC:
2300	{
2301	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2302	return VERR_INVALID_PARAMETER;
2303	rc = DBGFR0BpL2TblChunkAllocReqHandler(pGVM, (PDBGFBPL2TBLCHUNKALLOCREQ)pReqHdr);
2304	break;
2305	}
2306
2307	case VMMR0_DO_DBGF_BP_OWNER_INIT:
2308	{
2309	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2310	return VERR_INVALID_PARAMETER;
2311	rc = DBGFR0BpOwnerInitReqHandler(pGVM, (PDBGFBPOWNERINITREQ)pReqHdr);
2312	break;
2313	}
2314
2315	case VMMR0_DO_DBGF_BP_PORTIO_INIT:
2316	{
2317	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2318	return VERR_INVALID_PARAMETER;
2319	rc = DBGFR0BpPortIoInitReqHandler(pGVM, (PDBGFBPINITREQ)pReqHdr);
2320	break;
2321	}
2322
2323
2324	/*
2325	* TM requests.
2326	*/
2327	case VMMR0_DO_TM_GROW_TIMER_QUEUE:
2328	{
2329	if (pReqHdr \|\| idCpu == NIL_VMCPUID)
2330	return VERR_INVALID_PARAMETER;
2331	rc = TMR0TimerQueueGrow(pGVM, RT_HI_U32(u64Arg), RT_LO_U32(u64Arg));
2332	break;
2333	}
2334
2335	/*
2336	* For profiling.
2337	*/
2338	case VMMR0_DO_NOP:
2339	case VMMR0_DO_SLOW_NOP:
2340	return VINF_SUCCESS;
2341
2342	/*
2343	* For testing Ring-0 APIs invoked in this environment.
2344	*/
2345	case VMMR0_DO_TESTS:
2346	/** @todo make new test */
2347	return VINF_SUCCESS;
2348
2349	default:
2350	/*
2351	* We're returning VERR_NOT_SUPPORT here so we've got something else
2352	* than -1 which the interrupt gate glue code might return.
2353	*/
2354	Log(("operation %#x is not supported\n", enmOperation));
2355	return VERR_NOT_SUPPORTED;
2356	}
2357	return rc;
2358	}
2359
2360
2361	/**
2362	* This is just a longjmp wrapper function for VMMR0EntryEx calls.
2363	*
2364	* @returns VBox status code.
2365	* @param pvArgs The argument package
2366	*/
2367	static DECLCALLBACK(int) vmmR0EntryExWrapper(void *pvArgs)
2368	{
2369	PGVMCPU pGVCpu = (PGVMCPU)pvArgs;
2370	return vmmR0EntryExWorker(pGVCpu->vmmr0.s.pGVM,
2371	pGVCpu->vmmr0.s.idCpu,
2372	pGVCpu->vmmr0.s.enmOperation,
2373	pGVCpu->vmmr0.s.pReq,
2374	pGVCpu->vmmr0.s.u64Arg,
2375	pGVCpu->vmmr0.s.pSession);
2376	}
2377
2378
2379	/**
2380	* The Ring 0 entry point, called by the support library (SUP).
2381	*
2382	* @returns VBox status code.
2383	* @param pGVM The global (ring-0) VM structure.
2384	* @param pVM The cross context VM structure.
2385	* @param idCpu Virtual CPU ID argument. Must be NIL_VMCPUID if pVM
2386	* is NIL_RTR0PTR, and may be NIL_VMCPUID if it isn't
2387	* @param enmOperation Which operation to execute.
2388	* @param pReq Pointer to the SUPVMMR0REQHDR packet. Optional.
2389	* @param u64Arg Some simple constant argument.
2390	* @param pSession The session of the caller.
2391	* @remarks Assume called with interrupts _enabled_.
2392	*/
2393	VMMR0DECL(int) VMMR0EntryEx(PGVM pGVM, PVMCC pVM, VMCPUID idCpu, VMMR0OPERATION enmOperation,
2394	PSUPVMMR0REQHDR pReq, uint64_t u64Arg, PSUPDRVSESSION pSession)
2395	{
2396	/*
2397	* Requests that should only happen on the EMT thread will be
2398	* wrapped in a setjmp so we can assert without causing too much trouble.
2399	*/
2400	if ( pVM != NULL
2401	&& pGVM != NULL
2402	&& pVM == pGVM /** @todo drop pVM or pGVM */
2403	&& idCpu < pGVM->cCpus
2404	&& pGVM->pSession == pSession
2405	&& pGVM->pSelf == pGVM
2406	&& enmOperation != VMMR0_DO_GVMM_DESTROY_VM
2407	&& enmOperation != VMMR0_DO_GVMM_REGISTER_VMCPU
2408	&& enmOperation != VMMR0_DO_GVMM_SCHED_WAKE_UP /* idCpu is not caller but target. Sigh. / /* @todo fix*/
2409	&& enmOperation != VMMR0_DO_GVMM_SCHED_POKE /* idCpu is not caller but target. Sigh. / /* @todo fix*/
2410	)
2411	{
2412	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2413	RTNATIVETHREAD hNativeThread = RTThreadNativeSelf();
2414	if (RT_LIKELY( pGVCpu->hEMT == hNativeThread
2415	&& pGVCpu->hNativeThreadR0 == hNativeThread))
2416	{
2417	pGVCpu->vmmr0.s.pGVM = pGVM;
2418	pGVCpu->vmmr0.s.idCpu = idCpu;
2419	pGVCpu->vmmr0.s.enmOperation = enmOperation;
2420	pGVCpu->vmmr0.s.pReq = pReq;
2421	pGVCpu->vmmr0.s.u64Arg = u64Arg;
2422	pGVCpu->vmmr0.s.pSession = pSession;
2423	return vmmR0CallRing3SetJmpEx(&pGVCpu->vmmr0.s.AssertJmpBuf, vmmR0EntryExWrapper, pGVCpu,
2424	((uintptr_t)u64Arg << 16) \| (uintptr_t)enmOperation);
2425	}
2426	return VERR_VM_THREAD_NOT_EMT;
2427	}
2428	return vmmR0EntryExWorker(pGVM, idCpu, enmOperation, pReq, u64Arg, pSession);
2429	}
2430
2431
2432	/*********************************************************************************************************************************
2433	* EMT Blocking *
2434	*********************************************************************************************************************************/
2435
2436	/**
2437	* Checks whether we've armed the ring-0 long jump machinery.
2438	*
2439	* @returns @c true / @c false
2440	* @param pVCpu The cross context virtual CPU structure.
2441	* @thread EMT
2442	* @sa VMMIsLongJumpArmed
2443	*/
2444	VMMR0_INT_DECL(bool) VMMR0IsLongJumpArmed(PVMCPUCC pVCpu)
2445	{
2446	#ifdef RT_ARCH_X86
2447	return pVCpu->vmmr0.s.AssertJmpBuf.eip != 0;
2448	#else
2449	return pVCpu->vmmr0.s.AssertJmpBuf.rip != 0;
2450	#endif
2451	}
2452
2453
2454	/**
2455	* Locking helper that deals with HM context and checks if the thread can block.
2456	*
2457	* @returns VINF_SUCCESS if we can block. Returns @a rcBusy or
2458	* VERR_VMM_CANNOT_BLOCK if not able to block.
2459	* @param pVCpu The cross context virtual CPU structure of the calling
2460	* thread.
2461	* @param rcBusy What to return in case of a blocking problem. Will IPE
2462	* if VINF_SUCCESS and we cannot block.
2463	* @param pszCaller The caller (for logging problems).
2464	* @param pvLock The lock address (for logging problems).
2465	* @param pCtx Where to return context info for the resume call.
2466	* @thread EMT(pVCpu)
2467	*/
2468	VMMR0_INT_DECL(int) VMMR0EmtPrepareToBlock(PVMCPUCC pVCpu, int rcBusy, const char pszCaller, void pvLock,
2469	PVMMR0EMTBLOCKCTX pCtx)
2470	{
2471	const char *pszMsg;
2472
2473	/*
2474	* Check that we are allowed to block.
2475	*/
2476	if (RT_LIKELY(VMMRZCallRing3IsEnabled(pVCpu)))
2477	{
2478	/*
2479	* Are we in HM context and w/o a context hook? If so work the context hook.
2480	*/
2481	if (pVCpu->idHostCpu != NIL_RTCPUID)
2482	{
2483	Assert(pVCpu->iHostCpuSet != UINT32_MAX);
2484
2485	if (pVCpu->vmmr0.s.hCtxHook == NIL_RTTHREADCTXHOOK)
2486	{
2487	vmmR0ThreadCtxCallback(RTTHREADCTXEVENT_OUT, pVCpu);
2488	if (pVCpu->vmmr0.s.pPreemptState)
2489	RTThreadPreemptRestore(pVCpu->vmmr0.s.pPreemptState);
2490
2491	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC;
2492	pCtx->fWasInHmContext = true;
2493	return VINF_SUCCESS;
2494	}
2495	}
2496
2497	if (RT_LIKELY(!pVCpu->vmmr0.s.pPreemptState))
2498	{
2499	/*
2500	* Not in HM context or we've got hooks, so just check that preemption
2501	* is enabled.
2502	*/
2503	if (RT_LIKELY(RTThreadPreemptIsEnabled(NIL_RTTHREAD)))
2504	{
2505	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC;
2506	pCtx->fWasInHmContext = false;
2507	return VINF_SUCCESS;
2508	}
2509	pszMsg = "Preemption is disabled!";
2510	}
2511	else
2512	pszMsg = "Preemption state w/o HM state!";
2513	}
2514	else
2515	pszMsg = "Ring-3 calls are disabled!";
2516
2517	static uint32_t volatile s_cWarnings = 0;
2518	if (++s_cWarnings < 50)
2519	SUPR0Printf("VMMR0EmtPrepareToBlock: %s pvLock=%p pszCaller=%s rcBusy=%p\n", pszMsg, pvLock, pszCaller, rcBusy);
2520	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC_DEAD;
2521	pCtx->fWasInHmContext = false;
2522	return rcBusy != VINF_SUCCESS ? rcBusy : VERR_VMM_CANNOT_BLOCK;
2523	}
2524
2525
2526	/**
2527	* Counterpart to VMMR0EmtPrepareToBlock.
2528	*
2529	* @param pVCpu The cross context virtual CPU structure of the calling
2530	* thread.
2531	* @param pCtx The context structure used with VMMR0EmtPrepareToBlock.
2532	* @thread EMT(pVCpu)
2533	*/
2534	VMMR0_INT_DECL(void) VMMR0EmtResumeAfterBlocking(PVMCPUCC pVCpu, PVMMR0EMTBLOCKCTX pCtx)
2535	{
2536	AssertReturnVoid(pCtx->uMagic == VMMR0EMTBLOCKCTX_MAGIC);
2537	if (pCtx->fWasInHmContext)
2538	{
2539	if (pVCpu->vmmr0.s.pPreemptState)
2540	RTThreadPreemptDisable(pVCpu->vmmr0.s.pPreemptState);
2541
2542	pCtx->fWasInHmContext = false;
2543	vmmR0ThreadCtxCallback(RTTHREADCTXEVENT_IN, pVCpu);
2544	}
2545	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC_DEAD;
2546	}
2547
2548
2549	/**
2550	* Helper for waiting on an RTSEMEVENT, caller did VMMR0EmtPrepareToBlock.
2551	*
2552	* @returns
2553	* @retval VERR_THREAD_IS_TERMINATING
2554	* @retval VERR_TIMEOUT if we ended up waiting too long, either according to
2555	* @a cMsTimeout or to maximum wait values.
2556	*
2557	* @param pGVCpu The ring-0 virtual CPU structure.
2558	* @param fFlags VMMR0EMTWAIT_F_XXX.
2559	* @param hEvent The event to wait on.
2560	* @param cMsTimeout The timeout or RT_INDEFINITE_WAIT.
2561	*/
2562	VMMR0_INT_DECL(int) VMMR0EmtWaitEventInner(PGVMCPU pGVCpu, uint32_t fFlags, RTSEMEVENT hEvent, RTMSINTERVAL cMsTimeout)
2563	{
2564	AssertReturn(pGVCpu->hEMT == RTThreadNativeSelf(), VERR_VM_THREAD_NOT_EMT);
2565
2566	/*
2567	* Note! Similar code is found in the PDM critical sections too.
2568	*/
2569	uint64_t const nsStart = RTTimeNanoTS();
2570	uint64_t cNsMaxTotal = cMsTimeout == RT_INDEFINITE_WAIT
2571	? RT_NS_5MIN : RT_MIN(RT_NS_5MIN, RT_NS_1MS_64 * cMsTimeout);
2572	uint32_t cMsMaxOne = RT_MS_5SEC;
2573	bool fNonInterruptible = false;
2574	for (;;)
2575	{
2576	/* Wait. */
2577	int rcWait = !fNonInterruptible
2578	? RTSemEventWaitNoResume(hEvent, cMsMaxOne)
2579	: RTSemEventWait(hEvent, cMsMaxOne);
2580	if (RT_SUCCESS(rcWait))
2581	return rcWait;
2582
2583	if (rcWait == VERR_TIMEOUT \|\| rcWait == VERR_INTERRUPTED)
2584	{
2585	uint64_t const cNsElapsed = RTTimeNanoTS() - nsStart;
2586
2587	/*
2588	* Check the thread termination status.
2589	*/
2590	int const rcTerm = RTThreadQueryTerminationStatus(NIL_RTTHREAD);
2591	AssertMsg(rcTerm == VINF_SUCCESS \|\| rcTerm == VERR_NOT_SUPPORTED \|\| rcTerm == VINF_THREAD_IS_TERMINATING,
2592	("rcTerm=%Rrc\n", rcTerm));
2593	if ( rcTerm == VERR_NOT_SUPPORTED
2594	&& !fNonInterruptible
2595	&& cNsMaxTotal > RT_NS_1MIN)
2596	cNsMaxTotal = RT_NS_1MIN;
2597
2598	/* We return immediately if it looks like the thread is terminating. */
2599	if (rcTerm == VINF_THREAD_IS_TERMINATING)
2600	return VERR_THREAD_IS_TERMINATING;
2601
2602	/* We may suppress VERR_INTERRUPTED if VMMR0EMTWAIT_F_TRY_SUPPRESS_INTERRUPTED was
2603	specified, otherwise we'll just return it. */
2604	if (rcWait == VERR_INTERRUPTED)
2605	{
2606	if (!(fFlags & VMMR0EMTWAIT_F_TRY_SUPPRESS_INTERRUPTED))
2607	return VERR_INTERRUPTED;
2608	if (!fNonInterruptible)
2609	{
2610	/* First time: Adjust down the wait parameters and make sure we get at least
2611	one non-interruptible wait before timing out. */
2612	fNonInterruptible = true;
2613	cMsMaxOne = 32;
2614	uint64_t const cNsLeft = cNsMaxTotal - cNsElapsed;
2615	if (cNsLeft > RT_NS_10SEC)
2616	cNsMaxTotal = cNsElapsed + RT_NS_10SEC;
2617	continue;
2618	}
2619	}
2620
2621	/* Check for timeout. */
2622	if (cNsElapsed > cNsMaxTotal)
2623	return VERR_TIMEOUT;
2624	}
2625	else
2626	return rcWait;
2627	}
2628	/* not reached */
2629	}
2630
2631
2632	/**
2633	* Helper for signalling an SUPSEMEVENT.
2634	*
2635	* This may temporarily leave the HM context if the host requires that for
2636	* signalling SUPSEMEVENT objects.
2637	*
2638	* @returns VBox status code (see VMMR0EmtPrepareToBlock)
2639	* @param pGVM The ring-0 VM structure.
2640	* @param pGVCpu The ring-0 virtual CPU structure.
2641	* @param hEvent The event to signal.
2642	*/
2643	VMMR0_INT_DECL(int) VMMR0EmtSignalSupEvent(PGVM pGVM, PGVMCPU pGVCpu, SUPSEMEVENT hEvent)
2644	{
2645	AssertReturn(pGVCpu->hEMT == RTThreadNativeSelf(), VERR_VM_THREAD_NOT_EMT);
2646	if (RTSemEventIsSignalSafe())
2647	return SUPSemEventSignal(pGVM->pSession, hEvent);
2648
2649	VMMR0EMTBLOCKCTX Ctx;
2650	int rc = VMMR0EmtPrepareToBlock(pGVCpu, VINF_SUCCESS, __FUNCTION__, (void *)(uintptr_t)hEvent, &Ctx);
2651	if (RT_SUCCESS(rc))
2652	{
2653	rc = SUPSemEventSignal(pGVM->pSession, hEvent);
2654	VMMR0EmtResumeAfterBlocking(pGVCpu, &Ctx);
2655	}
2656	return rc;
2657	}
2658
2659
2660	/**
2661	* Helper for signalling an SUPSEMEVENT, variant supporting non-EMTs.
2662	*
2663	* This may temporarily leave the HM context if the host requires that for
2664	* signalling SUPSEMEVENT objects.
2665	*
2666	* @returns VBox status code (see VMMR0EmtPrepareToBlock)
2667	* @param pGVM The ring-0 VM structure.
2668	* @param hEvent The event to signal.
2669	*/
2670	VMMR0_INT_DECL(int) VMMR0EmtSignalSupEventByGVM(PGVM pGVM, SUPSEMEVENT hEvent)
2671	{
2672	if (!RTSemEventIsSignalSafe())
2673	{
2674	PGVMCPU pGVCpu = GVMMR0GetGVCpuByGVMandEMT(pGVM, NIL_RTNATIVETHREAD);
2675	if (pGVCpu)
2676	{
2677	VMMR0EMTBLOCKCTX Ctx;
2678	int rc = VMMR0EmtPrepareToBlock(pGVCpu, VINF_SUCCESS, __FUNCTION__, (void *)(uintptr_t)hEvent, &Ctx);
2679	if (RT_SUCCESS(rc))
2680	{
2681	rc = SUPSemEventSignal(pGVM->pSession, hEvent);
2682	VMMR0EmtResumeAfterBlocking(pGVCpu, &Ctx);
2683	}
2684	return rc;
2685	}
2686	}
2687	return SUPSemEventSignal(pGVM->pSession, hEvent);
2688	}
2689
2690
2691	/*********************************************************************************************************************************
2692	* Logging. *
2693	*********************************************************************************************************************************/
2694
2695	/**
2696	* VMMR0_DO_VMMR0_UPDATE_LOGGERS: Updates the EMT loggers for the VM.
2697	*
2698	* @returns VBox status code.
2699	* @param pGVM The global (ring-0) VM structure.
2700	* @param idCpu The ID of the calling EMT.
2701	* @param pReq The request data.
2702	* @param idxLogger Which logger set to update.
2703	* @thread EMT(idCpu)
2704	*/
2705	static int vmmR0UpdateLoggers(PGVM pGVM, VMCPUID idCpu, PVMMR0UPDATELOGGERSREQ pReq, size_t idxLogger)
2706	{
2707	/*
2708	* Check sanity. First we require EMT to be calling us.
2709	*/
2710	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID);
2711	AssertReturn(pGVM->aCpus[idCpu].hEMT == RTThreadNativeSelf(), VERR_INVALID_CPU_ID);
2712
2713	AssertReturn(pReq->Hdr.cbReq >= RT_UOFFSETOF_DYN(VMMR0UPDATELOGGERSREQ, afGroups[0]), VERR_INVALID_PARAMETER);
2714	AssertReturn(pReq->cGroups < _8K, VERR_INVALID_PARAMETER);
2715	AssertReturn(pReq->Hdr.cbReq == RT_UOFFSETOF_DYN(VMMR0UPDATELOGGERSREQ, afGroups[pReq->cGroups]), VERR_INVALID_PARAMETER);
2716
2717	AssertReturn(idxLogger < VMMLOGGER_IDX_MAX, VERR_OUT_OF_RANGE);
2718
2719	/*
2720	* Adjust flags.
2721	*/
2722	/* Always buffered: */
2723	pReq->fFlags \|= RTLOGFLAGS_BUFFERED;
2724	/* These doesn't make sense at present: */
2725	pReq->fFlags &= ~(RTLOGFLAGS_FLUSH \| RTLOGFLAGS_WRITE_THROUGH);
2726	/* We've traditionally skipped the group restrictions. */
2727	pReq->fFlags &= ~RTLOGFLAGS_RESTRICT_GROUPS;
2728
2729	/*
2730	* Do the updating.
2731	*/
2732	int rc = VINF_SUCCESS;
2733	for (idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
2734	{
2735	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2736	PRTLOGGER pLogger = pGVCpu->vmmr0.s.u.aLoggers[idxLogger].pLogger;
2737	if (pLogger)
2738	{
2739	RTLogSetR0ProgramStart(pLogger, pGVM->vmm.s.nsProgramStart);
2740	rc = RTLogBulkUpdate(pLogger, pReq->fFlags, pReq->uGroupCrc32, pReq->cGroups, pReq->afGroups);
2741	}
2742	}
2743
2744	return rc;
2745	}
2746
2747
2748	/**
2749	* VMMR0_DO_VMMR0_LOG_FLUSHER: Get the next log flushing job.
2750	*
2751	* The job info is copied into VMM::LogFlusherItem.
2752	*
2753	* @returns VBox status code.
2754	* @retval VERR_OBJECT_DESTROYED if we're shutting down.
2755	* @retval VERR_NOT_OWNER if the calling thread is not the flusher thread.
2756	* @param pGVM The global (ring-0) VM structure.
2757	* @thread The log flusher thread (first caller automatically becomes the log
2758	* flusher).
2759	*/
2760	static int vmmR0LogFlusher(PGVM pGVM)
2761	{
2762	/*
2763	* Check that this really is the flusher thread.
2764	*/
2765	RTNATIVETHREAD const hNativeSelf = RTThreadNativeSelf();
2766	AssertReturn(hNativeSelf != NIL_RTNATIVETHREAD, VERR_INTERNAL_ERROR_3);
2767	if (RT_LIKELY(pGVM->vmmr0.s.LogFlusher.hThread == hNativeSelf))
2768	{ /* likely */ }
2769	else
2770	{
2771	/* The first caller becomes the flusher thread. */
2772	bool fOk;
2773	ASMAtomicCmpXchgHandle(&pGVM->vmmr0.s.LogFlusher.hThread, hNativeSelf, NIL_RTNATIVETHREAD, fOk);
2774	if (!fOk)
2775	return VERR_NOT_OWNER;
2776	pGVM->vmmr0.s.LogFlusher.fThreadRunning = true;
2777	}
2778
2779	/*
2780	* Acknowledge flush, waking up waiting EMT.
2781	*/
2782	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2783
2784	uint32_t idxTail = pGVM->vmmr0.s.LogFlusher.idxRingTail % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2785	uint32_t idxHead = pGVM->vmmr0.s.LogFlusher.idxRingHead % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2786	if ( idxTail != idxHead
2787	&& pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.fProcessing)
2788	{
2789	/* Pop the head off the ring buffer. */
2790	uint32_t const idCpu = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.idCpu;
2791	uint32_t const idxLogger = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.idxLogger;
2792	uint32_t const idxBuffer = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.idxBuffer;
2793
2794	pGVM->vmmr0.s.LogFlusher.aRing[idxHead].u32 = UINT32_MAX >> 1; /* invalidate the entry */
2795	pGVM->vmmr0.s.LogFlusher.idxRingHead = (idxHead + 1) % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2796
2797	/* Validate content. */
2798	if ( idCpu < pGVM->cCpus
2799	&& idxLogger < VMMLOGGER_IDX_MAX
2800	&& idxBuffer < VMMLOGGER_BUFFER_COUNT)
2801	{
2802	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2803	PVMMR0PERVCPULOGGER pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
2804	PVMMR3CPULOGGER pShared = &pGVCpu->vmm.s.u.aLoggers[idxLogger];
2805
2806	/*
2807	* Accounting.
2808	*/
2809	uint32_t cFlushing = pR0Log->cFlushing - 1;
2810	if (RT_LIKELY(cFlushing < VMMLOGGER_BUFFER_COUNT))
2811	{ /likely/ }
2812	else
2813	cFlushing = 0;
2814	pR0Log->cFlushing = cFlushing;
2815	ASMAtomicWriteU32(&pShared->cFlushing, cFlushing);
2816
2817	/*
2818	* Wake up the EMT if it's waiting.
2819	*/
2820	if (!pR0Log->fEmtWaiting)
2821	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2822	else
2823	{
2824	pR0Log->fEmtWaiting = false;
2825	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2826
2827	int rc = RTSemEventSignal(pR0Log->hEventFlushWait);
2828	if (RT_FAILURE(rc))
2829	LogRelMax(64, ("vmmR0LogFlusher: RTSemEventSignal failed ACKing entry #%u (%u/%u/%u): %Rrc!\n",
2830	idxHead, idCpu, idxLogger, idxBuffer, rc));
2831	}
2832	}
2833	else
2834	{
2835	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2836	LogRelMax(64, ("vmmR0LogFlusher: Bad ACK entry #%u: %u/%u/%u!\n", idxHead, idCpu, idxLogger, idxBuffer));
2837	}
2838
2839	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2840	}
2841
2842	/*
2843	* The wait loop.
2844	*/
2845	int rc;
2846	for (;;)
2847	{
2848	/*
2849	* Work pending?
2850	*/
2851	idxTail = pGVM->vmmr0.s.LogFlusher.idxRingTail % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2852	idxHead = pGVM->vmmr0.s.LogFlusher.idxRingHead % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2853	if (idxTail != idxHead)
2854	{
2855	pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.fProcessing = true;
2856	pGVM->vmm.s.LogFlusherItem.u32 = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].u32;
2857
2858	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2859	return VINF_SUCCESS;
2860	}
2861
2862	/*
2863	* Nothing to do, so, check for termination and go to sleep.
2864	*/
2865	if (!pGVM->vmmr0.s.LogFlusher.fThreadShutdown)
2866	{ /* likely */ }
2867	else
2868	{
2869	rc = VERR_OBJECT_DESTROYED;
2870	break;
2871	}
2872
2873	pGVM->vmmr0.s.LogFlusher.fThreadWaiting = true;
2874	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2875
2876	rc = RTSemEventWaitNoResume(pGVM->vmmr0.s.LogFlusher.hEvent, RT_MS_5MIN);
2877
2878	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2879	pGVM->vmmr0.s.LogFlusher.fThreadWaiting = false;
2880
2881	if (RT_SUCCESS(rc) \|\| rc == VERR_TIMEOUT)
2882	{ /* likely */ }
2883	else if (rc == VERR_INTERRUPTED)
2884	{
2885	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2886	return rc;
2887	}
2888	else if (rc == VERR_SEM_DESTROYED \|\| rc == VERR_INVALID_HANDLE)
2889	break;
2890	else
2891	{
2892	LogRel(("vmmR0LogFlusher: RTSemEventWaitNoResume returned unexpected status %Rrc\n", rc));
2893	break;
2894	}
2895	}
2896
2897	/*
2898	* Terminating - prevent further calls and indicate to the EMTs that we're no longer around.
2899	*/
2900	pGVM->vmmr0.s.LogFlusher.hThread = ~pGVM->vmmr0.s.LogFlusher.hThread; /* (should be reasonably safe) */
2901	pGVM->vmmr0.s.LogFlusher.fThreadRunning = false;
2902
2903	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2904	return rc;
2905	}
2906
2907
2908	/**
2909	* VMMR0_DO_VMMR0_LOG_WAIT_FLUSHED: Waits for the flusher thread to finish all
2910	* buffers for logger @a idxLogger.
2911	*
2912	* @returns VBox status code.
2913	* @param pGVM The global (ring-0) VM structure.
2914	* @param idCpu The ID of the calling EMT.
2915	* @param idxLogger Which logger to wait on.
2916	* @thread EMT(idCpu)
2917	*/
2918	static int vmmR0LogWaitFlushed(PGVM pGVM, VMCPUID idCpu, size_t idxLogger)
2919	{
2920	/*
2921	* Check sanity. First we require EMT to be calling us.
2922	*/
2923	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID);
2924	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2925	AssertReturn(pGVCpu->hEMT == RTThreadNativeSelf(), VERR_INVALID_CPU_ID);
2926	AssertReturn(idxLogger < VMMLOGGER_IDX_MAX, VERR_OUT_OF_RANGE);
2927	PVMMR0PERVCPULOGGER const pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
2928
2929	/*
2930	* Do the waiting.
2931	*/
2932	int rc = VINF_SUCCESS;
2933	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2934	uint32_t cFlushing = pR0Log->cFlushing;
2935	while (cFlushing > 0)
2936	{
2937	pR0Log->fEmtWaiting = true;
2938	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2939
2940	rc = RTSemEventWaitNoResume(pR0Log->hEventFlushWait, RT_MS_5MIN);
2941
2942	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2943	pR0Log->fEmtWaiting = false;
2944	if (RT_SUCCESS(rc))
2945	{
2946	/* Read the new count, make sure it decreased before looping. That
2947	way we can guarentee that we will only wait more than 5 min * buffers. */
2948	uint32_t const cPrevFlushing = cFlushing;
2949	cFlushing = pR0Log->cFlushing;
2950	if (cFlushing < cPrevFlushing)
2951	continue;
2952	rc = VERR_INTERNAL_ERROR_3;
2953	}
2954	break;
2955	}
2956	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2957	return rc;
2958	}
2959
2960
2961	/**
2962	* Inner worker for vmmR0LoggerFlushCommon.
2963	*/
2964	static bool vmmR0LoggerFlushInner(PGVM pGVM, PGVMCPU pGVCpu, uint32_t idxLogger, size_t idxBuffer, uint32_t cbToFlush)
2965	{
2966	PVMMR0PERVCPULOGGER const pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
2967	PVMMR3CPULOGGER const pShared = &pGVCpu->vmm.s.u.aLoggers[idxLogger];
2968
2969	/*
2970	* Figure out what we need to do and whether we can.
2971	*/
2972	enum { kJustSignal, kPrepAndSignal, kPrepSignalAndWait } enmAction;
2973	#if VMMLOGGER_BUFFER_COUNT >= 2
2974	if (pR0Log->cFlushing < VMMLOGGER_BUFFER_COUNT - 1)
2975	{
2976	if (RTSemEventIsSignalSafe())
2977	enmAction = kJustSignal;
2978	else if (VMMRZCallRing3IsEnabled(pGVCpu))
2979	enmAction = kPrepAndSignal;
2980	else
2981	{
2982	/** @todo This is a bit simplistic. We could introduce a FF to signal the
2983	* thread or similar. */
2984	STAM_REL_COUNTER_INC(&pShared->StatCannotBlock);
2985	# if defined(RT_OS_LINUX)
2986	SUP_DPRINTF(("vmmR0LoggerFlush: Signalling not safe and EMT blocking disabled! (%u bytes)\n", cbToFlush));
2987	# endif
2988	pShared->cbDropped += cbToFlush;
2989	return true;
2990	}
2991	}
2992	else
2993	#endif
2994	if (VMMRZCallRing3IsEnabled(pGVCpu))
2995	enmAction = kPrepSignalAndWait;
2996	else
2997	{
2998	STAM_REL_COUNTER_INC(&pShared->StatCannotBlock);
2999	# if defined(RT_OS_LINUX)
3000	SUP_DPRINTF(("vmmR0LoggerFlush: EMT blocking disabled! (%u bytes)\n", cbToFlush));
3001	# endif
3002	pShared->cbDropped += cbToFlush;
3003	return true;
3004	}
3005
3006	/*
3007	* Prepare for blocking if necessary.
3008	*/
3009	VMMR0EMTBLOCKCTX Ctx;
3010	if (enmAction != kJustSignal)
3011	{
3012	int rc = VMMR0EmtPrepareToBlock(pGVCpu, VINF_SUCCESS, "vmmR0LoggerFlushInner", pR0Log->hEventFlushWait, &Ctx);
3013	if (RT_SUCCESS(rc))
3014	{ /* likely */ }
3015	else
3016	{
3017	STAM_REL_COUNTER_INC(&pShared->StatCannotBlock);
3018	SUP_DPRINTF(("vmmR0LoggerFlush: VMMR0EmtPrepareToBlock failed! rc=%d\n", rc));
3019	return false;
3020	}
3021	}
3022
3023	/*
3024	* Queue the flush job.
3025	*/
3026	bool fFlushedBuffer;
3027	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3028	if (pGVM->vmmr0.s.LogFlusher.fThreadRunning)
3029	{
3030	uint32_t const idxHead = pGVM->vmmr0.s.LogFlusher.idxRingHead % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
3031	uint32_t const idxTail = pGVM->vmmr0.s.LogFlusher.idxRingTail % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
3032	uint32_t const idxNewTail = (idxTail + 1) % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
3033	if (idxNewTail != idxHead)
3034	{
3035	/* Queue it. */
3036	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.idCpu = pGVCpu->idCpu;
3037	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.idxLogger = idxLogger;
3038	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.idxBuffer = (uint32_t)idxBuffer;
3039	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.fProcessing = 0;
3040	pGVM->vmmr0.s.LogFlusher.idxRingTail = idxNewTail;
3041
3042	/* Update the number of buffers currently being flushed. */
3043	uint32_t cFlushing = pR0Log->cFlushing;
3044	cFlushing = RT_MIN(cFlushing + 1, VMMLOGGER_BUFFER_COUNT);
3045	pShared->cFlushing = pR0Log->cFlushing = cFlushing;
3046
3047	/* We must wait if all buffers are currently being flushed. */
3048	bool const fEmtWaiting = cFlushing >= VMMLOGGER_BUFFER_COUNT && enmAction != kJustSignal /* paranoia */;
3049	pR0Log->fEmtWaiting = fEmtWaiting;
3050
3051	/* Stats. */
3052	STAM_REL_COUNTER_INC(&pShared->StatFlushes);
3053	STAM_REL_COUNTER_INC(&pGVM->vmm.s.StatLogFlusherFlushes);
3054
3055	/* Signal the worker thread. */
3056	if (pGVM->vmmr0.s.LogFlusher.fThreadWaiting)
3057	{
3058	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3059	RTSemEventSignal(pGVM->vmmr0.s.LogFlusher.hEvent);
3060	}
3061	else
3062	{
3063	STAM_REL_COUNTER_INC(&pGVM->vmm.s.StatLogFlusherNoWakeUp);
3064	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3065	}
3066
3067	/*
3068	* Wait for a buffer to finish flushing.
3069	*
3070	* Note! Lazy bird is ignoring the status code here. The result is
3071	* that we might end up with an extra even signalling and the
3072	* next time we need to wait we won't and end up with some log
3073	* corruption. However, it's too much hazzle right now for
3074	* a scenario which would most likely end the process rather
3075	* than causing log corruption.
3076	*/
3077	if (fEmtWaiting)
3078	{
3079	STAM_REL_PROFILE_START(&pShared->StatWait, a);
3080	VMMR0EmtWaitEventInner(pGVCpu, VMMR0EMTWAIT_F_TRY_SUPPRESS_INTERRUPTED,
3081	pR0Log->hEventFlushWait, RT_INDEFINITE_WAIT);
3082	STAM_REL_PROFILE_STOP(&pShared->StatWait, a);
3083	}
3084
3085	/*
3086	* We always switch buffer if we have more than one.
3087	*/
3088	#if VMMLOGGER_BUFFER_COUNT == 1
3089	fFlushedBuffer = true;
3090	#else
3091	AssertCompile(VMMLOGGER_BUFFER_COUNT >= 1);
3092	pShared->idxBuf = (idxBuffer + 1) % VMMLOGGER_BUFFER_COUNT;
3093	fFlushedBuffer = false;
3094	#endif
3095	}
3096	else
3097	{
3098	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3099	SUP_DPRINTF(("vmmR0LoggerFlush: ring buffer is full!\n"));
3100	fFlushedBuffer = true;
3101	}
3102	}
3103	else
3104	{
3105	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3106	SUP_DPRINTF(("vmmR0LoggerFlush: flusher not active - dropping %u bytes\n", cbToFlush));
3107	fFlushedBuffer = true;
3108	}
3109
3110	/*
3111	* Restore the HM context.
3112	*/
3113	if (enmAction != kJustSignal)
3114	VMMR0EmtResumeAfterBlocking(pGVCpu, &Ctx);
3115
3116	return fFlushedBuffer;
3117	}
3118
3119
3120	/**
3121	* Common worker for vmmR0LogFlush and vmmR0LogRelFlush.
3122	*/
3123	static bool vmmR0LoggerFlushCommon(PRTLOGGER pLogger, PRTLOGBUFFERDESC pBufDesc, uint32_t idxLogger)
3124	{
3125	/*
3126	* Convert the pLogger into a GVMCPU handle and 'call' back to Ring-3.
3127	* (This is a bit paranoid code.)
3128	*/
3129	if (RT_VALID_PTR(pLogger))
3130	{
3131	if ( pLogger->u32Magic == RTLOGGER_MAGIC
3132	&& (pLogger->u32UserValue1 & VMMR0_LOGGER_FLAGS_MAGIC_MASK) == VMMR0_LOGGER_FLAGS_MAGIC_VALUE
3133	&& pLogger->u64UserValue2 == pLogger->u64UserValue3)
3134	{
3135	PGVMCPU const pGVCpu = (PGVMCPU)(uintptr_t)pLogger->u64UserValue2;
3136	if ( RT_VALID_PTR(pGVCpu)
3137	&& ((uintptr_t)pGVCpu & PAGE_OFFSET_MASK) == 0)
3138	{
3139	RTNATIVETHREAD const hNativeSelf = RTThreadNativeSelf();
3140	PGVM const pGVM = pGVCpu->pGVM;
3141	if ( hNativeSelf == pGVCpu->hEMT
3142	&& RT_VALID_PTR(pGVM))
3143	{
3144	PVMMR0PERVCPULOGGER const pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
3145	size_t const idxBuffer = pBufDesc - &pR0Log->aBufDescs[0];
3146	if (idxBuffer < VMMLOGGER_BUFFER_COUNT)
3147	{
3148	/*
3149	* Make sure we don't recurse forever here should something in the
3150	* following code trigger logging or an assertion. Do the rest in
3151	* an inner work to avoid hitting the right margin too hard.
3152	*/
3153	if (!pR0Log->fFlushing)
3154	{
3155	pR0Log->fFlushing = true;
3156	bool fFlushed = vmmR0LoggerFlushInner(pGVM, pGVCpu, idxLogger, idxBuffer, pBufDesc->offBuf);
3157	pR0Log->fFlushing = false;
3158	return fFlushed;
3159	}
3160
3161	SUP_DPRINTF(("vmmR0LoggerFlush: Recursive flushing!\n"));
3162	}
3163	else
3164	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p pGVCpu=%p: idxBuffer=%#zx\n", pLogger, pGVCpu, idxBuffer));
3165	}
3166	else
3167	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p pGVCpu=%p hEMT=%p hNativeSelf=%p!\n",
3168	pLogger, pGVCpu, pGVCpu->hEMT, hNativeSelf));
3169	}
3170	else
3171	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p pGVCpu=%p!\n", pLogger, pGVCpu));
3172	}
3173	else
3174	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p u32Magic=%#x u32UserValue1=%#x u64UserValue2=%#RX64 u64UserValue3=%#RX64!\n",
3175	pLogger, pLogger->u32Magic, pLogger->u32UserValue1, pLogger->u64UserValue2, pLogger->u64UserValue3));
3176	}
3177	else
3178	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p!\n", pLogger));
3179	return true;
3180	}
3181
3182
3183	/**
3184	* @callback_method_impl{FNRTLOGFLUSH, Release logger buffer flush callback.}
3185	*/
3186	static DECLCALLBACK(bool) vmmR0LogRelFlush(PRTLOGGER pLogger, PRTLOGBUFFERDESC pBufDesc)
3187	{
3188	return vmmR0LoggerFlushCommon(pLogger, pBufDesc, VMMLOGGER_IDX_RELEASE);
3189	}
3190
3191
3192	/**
3193	* @callback_method_impl{FNRTLOGFLUSH, Logger (debug) buffer flush callback.}
3194	*/
3195	static DECLCALLBACK(bool) vmmR0LogFlush(PRTLOGGER pLogger, PRTLOGBUFFERDESC pBufDesc)
3196	{
3197	#ifdef LOG_ENABLED
3198	return vmmR0LoggerFlushCommon(pLogger, pBufDesc, VMMLOGGER_IDX_REGULAR);
3199	#else
3200	RT_NOREF(pLogger, pBufDesc);
3201	return true;
3202	#endif
3203	}
3204
3205
3206	/*
3207	* Override RTLogDefaultInstanceEx so we can do logging from EMTs in ring-0.
3208	*/
3209	DECLEXPORT(PRTLOGGER) RTLogDefaultInstanceEx(uint32_t fFlagsAndGroup)
3210	{
3211	#ifdef LOG_ENABLED
3212	PGVMCPU pGVCpu = GVMMR0GetGVCpuByEMT(NIL_RTNATIVETHREAD);
3213	if (pGVCpu)
3214	{
3215	PRTLOGGER pLogger = pGVCpu->vmmr0.s.u.s.Logger.pLogger;
3216	if (RT_VALID_PTR(pLogger))
3217	{
3218	if ( pLogger->u64UserValue2 == (uintptr_t)pGVCpu
3219	&& pLogger->u64UserValue3 == (uintptr_t)pGVCpu)
3220	{
3221	if (!pGVCpu->vmmr0.s.u.s.Logger.fFlushing)
3222	{
3223	if (!(pGVCpu->vmmr0.s.fLogFlushingDisabled))
3224	return RTLogCheckGroupFlags(pLogger, fFlagsAndGroup);
3225	return NULL;
3226	}
3227
3228	/*
3229	* When we're flushing we _must_ return NULL here to suppress any
3230	* attempts at using the logger while in vmmR0LoggerFlushCommon.
3231	* The VMMR0EmtPrepareToBlock code may trigger logging in HM,
3232	* which will reset the buffer content before we even get to queue
3233	* the flush request. (Only an issue when VBOX_WITH_R0_LOGGING
3234	* is enabled.)
3235	*/
3236	return NULL;
3237	}
3238	}
3239	}
3240	#endif
3241	return SUPR0DefaultLogInstanceEx(fFlagsAndGroup);
3242	}
3243
3244
3245	/*
3246	* Override RTLogRelGetDefaultInstanceEx so we can do LogRel to VBox.log from EMTs in ring-0.
3247	*/
3248	DECLEXPORT(PRTLOGGER) RTLogRelGetDefaultInstanceEx(uint32_t fFlagsAndGroup)
3249	{
3250	PGVMCPU pGVCpu = GVMMR0GetGVCpuByEMT(NIL_RTNATIVETHREAD);
3251	if (pGVCpu)
3252	{
3253	PRTLOGGER pLogger = pGVCpu->vmmr0.s.u.s.RelLogger.pLogger;
3254	if (RT_VALID_PTR(pLogger))
3255	{
3256	if ( pLogger->u64UserValue2 == (uintptr_t)pGVCpu
3257	&& pLogger->u64UserValue3 == (uintptr_t)pGVCpu)
3258	{
3259	if (!pGVCpu->vmmr0.s.u.s.RelLogger.fFlushing)
3260	{
3261	if (!(pGVCpu->vmmr0.s.fLogFlushingDisabled))
3262	return RTLogCheckGroupFlags(pLogger, fFlagsAndGroup);
3263	return NULL;
3264	}
3265	}
3266	}
3267	}
3268	return SUPR0GetDefaultLogRelInstanceEx(fFlagsAndGroup);
3269	}
3270
3271
3272	/**
3273	* Helper for vmmR0InitLoggerSet
3274	*/
3275	static int vmmR0InitLoggerOne(PGVMCPU pGVCpu, bool fRelease, PVMMR0PERVCPULOGGER pR0Log, PVMMR3CPULOGGER pShared,
3276	uint32_t cbBuf, char *pchBuf, RTR3PTR pchBufR3)
3277	{
3278	/*
3279	* Create and configure the logger.
3280	*/
3281	for (size_t i = 0; i < VMMLOGGER_BUFFER_COUNT; i++)
3282	{
3283	pR0Log->aBufDescs[i].u32Magic = RTLOGBUFFERDESC_MAGIC;
3284	pR0Log->aBufDescs[i].uReserved = 0;
3285	pR0Log->aBufDescs[i].cbBuf = cbBuf;
3286	pR0Log->aBufDescs[i].offBuf = 0;
3287	pR0Log->aBufDescs[i].pchBuf = pchBuf + i * cbBuf;
3288	pR0Log->aBufDescs[i].pAux = &pShared->aBufs[i].AuxDesc;
3289
3290	pShared->aBufs[i].AuxDesc.fFlushedIndicator = false;
3291	pShared->aBufs[i].AuxDesc.afPadding[0] = 0;
3292	pShared->aBufs[i].AuxDesc.afPadding[1] = 0;
3293	pShared->aBufs[i].AuxDesc.afPadding[2] = 0;
3294	pShared->aBufs[i].AuxDesc.offBuf = 0;
3295	pShared->aBufs[i].pchBufR3 = pchBufR3 + i * cbBuf;
3296	}
3297	pShared->cbBuf = cbBuf;
3298
3299	static const char * const s_apszGroups[] = VBOX_LOGGROUP_NAMES;
3300	int rc = RTLogCreateEx(&pR0Log->pLogger, fRelease ? "VBOX_RELEASE_LOG" : "VBOX_LOG", RTLOG_F_NO_LOCKING \| RTLOGFLAGS_BUFFERED,
3301	"all", RT_ELEMENTS(s_apszGroups), s_apszGroups, UINT32_MAX,
3302	VMMLOGGER_BUFFER_COUNT, pR0Log->aBufDescs, RTLOGDEST_DUMMY,
3303	NULL /pfnPhase/, 0 /cHistory/, 0 /cbHistoryFileMax/, 0 /cSecsHistoryTimeSlot/,
3304	NULL /pErrInfo/, NULL /pszFilenameFmt/);
3305	if (RT_SUCCESS(rc))
3306	{
3307	PRTLOGGER pLogger = pR0Log->pLogger;
3308	pLogger->u32UserValue1 = VMMR0_LOGGER_FLAGS_MAGIC_VALUE;
3309	pLogger->u64UserValue2 = (uintptr_t)pGVCpu;
3310	pLogger->u64UserValue3 = (uintptr_t)pGVCpu;
3311
3312	rc = RTLogSetFlushCallback(pLogger, fRelease ? vmmR0LogRelFlush : vmmR0LogFlush);
3313	if (RT_SUCCESS(rc))
3314	{
3315	RTLogSetR0ThreadNameF(pLogger, "EMT-%u-R0", pGVCpu->idCpu);
3316
3317	/*
3318	* Create the event sem the EMT waits on while flushing is happening.
3319	*/
3320	rc = RTSemEventCreate(&pR0Log->hEventFlushWait);
3321	if (RT_SUCCESS(rc))
3322	return VINF_SUCCESS;
3323	pR0Log->hEventFlushWait = NIL_RTSEMEVENT;
3324	}
3325	RTLogDestroy(pLogger);
3326	}
3327	pR0Log->pLogger = NULL;
3328	return rc;
3329	}
3330
3331
3332	/**
3333	* Worker for VMMR0CleanupVM and vmmR0InitLoggerSet that destroys one logger.
3334	*/
3335	static void vmmR0TermLoggerOne(PVMMR0PERVCPULOGGER pR0Log, PVMMR3CPULOGGER pShared)
3336	{
3337	RTLogDestroy(pR0Log->pLogger);
3338	pR0Log->pLogger = NULL;
3339
3340	for (size_t i = 0; i < VMMLOGGER_BUFFER_COUNT; i++)
3341	pShared->aBufs[i].pchBufR3 = NIL_RTR3PTR;
3342
3343	RTSemEventDestroy(pR0Log->hEventFlushWait);
3344	pR0Log->hEventFlushWait = NIL_RTSEMEVENT;
3345	}
3346
3347
3348	/**
3349	* Initializes one type of loggers for each EMT.
3350	*/
3351	static int vmmR0InitLoggerSet(PGVM pGVM, uint8_t idxLogger, uint32_t cbBuf, PRTR0MEMOBJ phMemObj, PRTR0MEMOBJ phMapObj)
3352	{
3353	/* Allocate buffers first. */
3354	int rc = RTR0MemObjAllocPage(phMemObj, cbBuf * pGVM->cCpus * VMMLOGGER_BUFFER_COUNT, false /fExecutable/);
3355	if (RT_SUCCESS(rc))
3356	{
3357	rc = RTR0MemObjMapUser(phMapObj, phMemObj, (RTR3PTR)-1, 0 /uAlignment*/, RTMEM_PROT_READ, NIL_RTR0PROCESS);
3358	if (RT_SUCCESS(rc))
3359	{
3360	char * const pchBuf = (char )RTR0MemObjAddress(phMemObj);
3361	AssertPtrReturn(pchBuf, VERR_INTERNAL_ERROR_2);
3362
3363	RTR3PTR const pchBufR3 = RTR0MemObjAddressR3(*phMapObj);
3364	AssertReturn(pchBufR3 != NIL_RTR3PTR, VERR_INTERNAL_ERROR_3);
3365
3366	/* Initialize the per-CPU loggers. */
3367	for (uint32_t i = 0; i < pGVM->cCpus; i++)
3368	{
3369	PGVMCPU pGVCpu = &pGVM->aCpus[i];
3370	PVMMR0PERVCPULOGGER pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
3371	PVMMR3CPULOGGER pShared = &pGVCpu->vmm.s.u.aLoggers[idxLogger];
3372	rc = vmmR0InitLoggerOne(pGVCpu, idxLogger == VMMLOGGER_IDX_RELEASE, pR0Log, pShared, cbBuf,
3373	pchBuf + i * cbBuf * VMMLOGGER_BUFFER_COUNT,
3374	pchBufR3 + i * cbBuf * VMMLOGGER_BUFFER_COUNT);
3375	if (RT_FAILURE(rc))
3376	{
3377	vmmR0TermLoggerOne(pR0Log, pShared);
3378	while (i-- > 0)
3379	{
3380	pGVCpu = &pGVM->aCpus[i];
3381	vmmR0TermLoggerOne(&pGVCpu->vmmr0.s.u.aLoggers[idxLogger], &pGVCpu->vmm.s.u.aLoggers[idxLogger]);
3382	}
3383	break;
3384	}
3385	}
3386	if (RT_SUCCESS(rc))
3387	return VINF_SUCCESS;
3388
3389	/* Bail out. */
3390	RTR0MemObjFree(phMapObj, false /fFreeMappings*/);
3391	*phMapObj = NIL_RTR0MEMOBJ;
3392	}
3393	RTR0MemObjFree(phMemObj, true /fFreeMappings*/);
3394	*phMemObj = NIL_RTR0MEMOBJ;
3395	}
3396	return rc;
3397	}
3398
3399
3400	/**
3401	* Worker for VMMR0InitPerVMData that initializes all the logging related stuff.
3402	*
3403	* @returns VBox status code.
3404	* @param pGVM The global (ring-0) VM structure.
3405	*/
3406	static int vmmR0InitLoggers(PGVM pGVM)
3407	{
3408	/*
3409	* Invalidate the ring buffer (not really necessary).
3410	*/
3411	for (size_t idx = 0; idx < RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing); idx++)
3412	pGVM->vmmr0.s.LogFlusher.aRing[idx].u32 = UINT32_MAX >> 1; /* (all bits except fProcessing set) */
3413
3414	/*
3415	* Create the spinlock and flusher event semaphore.
3416	*/
3417	int rc = RTSpinlockCreate(&pGVM->vmmr0.s.LogFlusher.hSpinlock, RTSPINLOCK_FLAGS_INTERRUPT_SAFE, "VM-Log-Flusher");
3418	if (RT_SUCCESS(rc))
3419	{
3420	rc = RTSemEventCreate(&pGVM->vmmr0.s.LogFlusher.hEvent);
3421	if (RT_SUCCESS(rc))
3422	{
3423	/*
3424	* Create the ring-0 release loggers.
3425	*/
3426	rc = vmmR0InitLoggerSet(pGVM, VMMLOGGER_IDX_RELEASE, _4K,
3427	&pGVM->vmmr0.s.hMemObjReleaseLogger, &pGVM->vmmr0.s.hMapObjReleaseLogger);
3428	#ifdef LOG_ENABLED
3429	if (RT_SUCCESS(rc))
3430	{
3431	/*
3432	* Create debug loggers.
3433	*/
3434	rc = vmmR0InitLoggerSet(pGVM, VMMLOGGER_IDX_REGULAR, _64K,
3435	&pGVM->vmmr0.s.hMemObjLogger, &pGVM->vmmr0.s.hMapObjLogger);
3436	}
3437	#endif
3438	}
3439	}
3440	return rc;
3441	}
3442
3443
3444	/**
3445	* Worker for VMMR0InitPerVMData that initializes all the logging related stuff.
3446	*
3447	* @param pGVM The global (ring-0) VM structure.
3448	*/
3449	static void vmmR0CleanupLoggers(PGVM pGVM)
3450	{
3451	for (VMCPUID idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
3452	{
3453	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
3454	for (size_t iLogger = 0; iLogger < RT_ELEMENTS(pGVCpu->vmmr0.s.u.aLoggers); iLogger++)
3455	vmmR0TermLoggerOne(&pGVCpu->vmmr0.s.u.aLoggers[iLogger], &pGVCpu->vmm.s.u.aLoggers[iLogger]);
3456	}
3457
3458	/*
3459	* Free logger buffer memory.
3460	*/
3461	RTR0MemObjFree(pGVM->vmmr0.s.hMapObjReleaseLogger, false /fFreeMappings/);
3462	pGVM->vmmr0.s.hMapObjReleaseLogger = NIL_RTR0MEMOBJ;
3463	RTR0MemObjFree(pGVM->vmmr0.s.hMemObjReleaseLogger, true /fFreeMappings/);
3464	pGVM->vmmr0.s.hMemObjReleaseLogger = NIL_RTR0MEMOBJ;
3465
3466	RTR0MemObjFree(pGVM->vmmr0.s.hMapObjLogger, false /fFreeMappings/);
3467	pGVM->vmmr0.s.hMapObjLogger = NIL_RTR0MEMOBJ;
3468	RTR0MemObjFree(pGVM->vmmr0.s.hMemObjLogger, true /fFreeMappings/);
3469	pGVM->vmmr0.s.hMemObjLogger = NIL_RTR0MEMOBJ;
3470
3471	/*
3472	* Free log flusher related stuff.
3473	*/
3474	RTSpinlockDestroy(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3475	pGVM->vmmr0.s.LogFlusher.hSpinlock = NIL_RTSPINLOCK;
3476	RTSemEventDestroy(pGVM->vmmr0.s.LogFlusher.hEvent);
3477	pGVM->vmmr0.s.LogFlusher.hEvent = NIL_RTSEMEVENT;
3478	}
3479
3480
3481	/*********************************************************************************************************************************
3482	* Assertions *
3483	*********************************************************************************************************************************/
3484
3485	/**
3486	* Installs a notification callback for ring-0 assertions.
3487	*
3488	* @param pVCpu The cross context virtual CPU structure.
3489	* @param pfnCallback Pointer to the callback.
3490	* @param pvUser The user argument.
3491	*
3492	* @return VBox status code.
3493	*/
3494	VMMR0_INT_DECL(int) VMMR0AssertionSetNotification(PVMCPUCC pVCpu, PFNVMMR0ASSERTIONNOTIFICATION pfnCallback, RTR0PTR pvUser)
3495	{
3496	AssertPtrReturn(pVCpu, VERR_INVALID_POINTER);
3497	AssertPtrReturn(pfnCallback, VERR_INVALID_POINTER);
3498
3499	if (!pVCpu->vmmr0.s.pfnAssertCallback)
3500	{
3501	pVCpu->vmmr0.s.pfnAssertCallback = pfnCallback;
3502	pVCpu->vmmr0.s.pvAssertCallbackUser = pvUser;
3503	return VINF_SUCCESS;
3504	}
3505	return VERR_ALREADY_EXISTS;
3506	}
3507
3508
3509	/**
3510	* Removes the ring-0 callback.
3511	*
3512	* @param pVCpu The cross context virtual CPU structure.
3513	*/
3514	VMMR0_INT_DECL(void) VMMR0AssertionRemoveNotification(PVMCPUCC pVCpu)
3515	{
3516	pVCpu->vmmr0.s.pfnAssertCallback = NULL;
3517	pVCpu->vmmr0.s.pvAssertCallbackUser = NULL;
3518	}
3519
3520
3521	/**
3522	* Checks whether there is a ring-0 callback notification active.
3523	*
3524	* @param pVCpu The cross context virtual CPU structure.
3525	* @returns true if there the notification is active, false otherwise.
3526	*/
3527	VMMR0_INT_DECL(bool) VMMR0AssertionIsNotificationSet(PVMCPUCC pVCpu)
3528	{
3529	return pVCpu->vmmr0.s.pfnAssertCallback != NULL;
3530	}
3531
3532
3533	/*
3534	* Jump back to ring-3 if we're the EMT and the longjmp is armed.
3535	*
3536	* @returns true if the breakpoint should be hit, false if it should be ignored.
3537	*/
3538	DECLEXPORT(bool) RTCALL RTAssertShouldPanic(void)
3539	{
3540	#if 0
3541	return true;
3542	#else
3543	PVMCC pVM = GVMMR0GetVMByEMT(NIL_RTNATIVETHREAD);
3544	if (pVM)
3545	{
3546	PVMCPUCC pVCpu = VMMGetCpu(pVM);
3547
3548	if (pVCpu)
3549	{
3550	# ifdef RT_ARCH_X86
3551	if (pVCpu->vmmr0.s.AssertJmpBuf.eip)
3552	# else
3553	if (pVCpu->vmmr0.s.AssertJmpBuf.rip)
3554	# endif
3555	{
3556	if (pVCpu->vmmr0.s.pfnAssertCallback)
3557	pVCpu->vmmr0.s.pfnAssertCallback(pVCpu, pVCpu->vmmr0.s.pvAssertCallbackUser);
3558	int rc = vmmR0CallRing3LongJmp(&pVCpu->vmmr0.s.AssertJmpBuf, VERR_VMM_RING0_ASSERTION);
3559	return RT_FAILURE_NP(rc);
3560	}
3561	}
3562	}
3563	# ifdef RT_OS_LINUX
3564	return true;
3565	# else
3566	return false;
3567	# endif
3568	#endif
3569	}
3570
3571
3572	/*
3573	* Override this so we can push it up to ring-3.
3574	*/
3575	DECLEXPORT(void) RTCALL RTAssertMsg1Weak(const char pszExpr, unsigned uLine, const char pszFile, const char *pszFunction)
3576	{
3577	/*
3578	* To host kernel log/whatever.
3579	*/
3580	SUPR0Printf("!!R0-Assertion Failed!!\n"
3581	"Expression: %s\n"
3582	"Location : %s(%d) %s\n",
3583	pszExpr, pszFile, uLine, pszFunction);
3584
3585	/*
3586	* To the log.
3587	*/
3588	LogAlways(("\n!!R0-Assertion Failed!!\n"
3589	"Expression: %s\n"
3590	"Location : %s(%d) %s\n",
3591	pszExpr, pszFile, uLine, pszFunction));
3592
3593	/*
3594	* To the global VMM buffer.
3595	*/
3596	PVMCC pVM = GVMMR0GetVMByEMT(NIL_RTNATIVETHREAD);
3597	if (pVM)
3598	RTStrPrintf(pVM->vmm.s.szRing0AssertMsg1, sizeof(pVM->vmm.s.szRing0AssertMsg1),
3599	"\n!!R0-Assertion Failed!!\n"
3600	"Expression: %.*s\n"
3601	"Location : %s(%d) %s\n",
3602	sizeof(pVM->vmm.s.szRing0AssertMsg1) / 4 * 3, pszExpr,
3603	pszFile, uLine, pszFunction);
3604
3605	/*
3606	* Continue the normal way.
3607	*/
3608	RTAssertMsg1(pszExpr, uLine, pszFile, pszFunction);
3609	}
3610
3611
3612	/**
3613	* Callback for RTLogFormatV which writes to the ring-3 log port.
3614	* See PFNLOGOUTPUT() for details.
3615	*/
3616	static DECLCALLBACK(size_t) rtLogOutput(void pv, const char pachChars, size_t cbChars)
3617	{
3618	for (size_t i = 0; i < cbChars; i++)
3619	{
3620	LogAlways(("%c", pachChars[i])); NOREF(pachChars);
3621	}
3622
3623	NOREF(pv);
3624	return cbChars;
3625	}
3626
3627
3628	/*
3629	* Override this so we can push it up to ring-3.
3630	*/
3631	DECLEXPORT(void) RTCALL RTAssertMsg2WeakV(const char *pszFormat, va_list va)
3632	{
3633	va_list vaCopy;
3634
3635	/*
3636	* Push the message to the loggers.
3637	*/
3638	PRTLOGGER pLog = RTLogRelGetDefaultInstance();
3639	if (pLog)
3640	{
3641	va_copy(vaCopy, va);
3642	RTLogFormatV(rtLogOutput, pLog, pszFormat, vaCopy);
3643	va_end(vaCopy);
3644	}
3645	pLog = RTLogGetDefaultInstance(); /* Don't initialize it here... */
3646	if (pLog)
3647	{
3648	va_copy(vaCopy, va);
3649	RTLogFormatV(rtLogOutput, pLog, pszFormat, vaCopy);
3650	va_end(vaCopy);
3651	}
3652
3653	/*
3654	* Push it to the global VMM buffer.
3655	*/
3656	PVMCC pVM = GVMMR0GetVMByEMT(NIL_RTNATIVETHREAD);
3657	if (pVM)
3658	{
3659	va_copy(vaCopy, va);
3660	RTStrPrintfV(pVM->vmm.s.szRing0AssertMsg2, sizeof(pVM->vmm.s.szRing0AssertMsg2), pszFormat, vaCopy);
3661	va_end(vaCopy);
3662	}
3663
3664	/*
3665	* Continue the normal way.
3666	*/
3667	RTAssertMsg2V(pszFormat, va);
3668	}
3669

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

source: vbox/trunk/src/VBox/VMM/VMMR0/VMMR0.cpp@ 92679

Download in other formats: