VMMR0.cpp@ 93705

Last change on this file since 93705 was 93650, checked in by vboxsync, 3 years ago
VMM/PGM,*: Split the physical access handler type registration into separate ring-0 and ring-3 steps, expanding the type to 64-bit. bugref:10094
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 134.5 KB

Line
1	/* $Id: VMMR0.cpp 93650 2022-02-08 10:43:53Z vboxsync $ */
2	/** @file
3	* VMM - Host Context Ring 0.
4	*/
5
6	/*
7	* Copyright (C) 2006-2022 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/iem.h>
26	#include <VBox/vmm/iom.h>
27	#include <VBox/vmm/trpm.h>
28	#include <VBox/vmm/cpum.h>
29	#include <VBox/vmm/pdmapi.h>
30	#include <VBox/vmm/pgm.h>
31	#ifdef VBOX_WITH_NEM_R0
32	# include <VBox/vmm/nem.h>
33	#endif
34	#include <VBox/vmm/em.h>
35	#include <VBox/vmm/stam.h>
36	#include <VBox/vmm/tm.h>
37	#include "VMMInternal.h"
38	#include <VBox/vmm/vmcc.h>
39	#include <VBox/vmm/gvm.h>
40	#ifdef VBOX_WITH_PCI_PASSTHROUGH
41	# include <VBox/vmm/pdmpci.h>
42	#endif
43	#include <VBox/vmm/apic.h>
44
45	#include <VBox/vmm/gvmm.h>
46	#include <VBox/vmm/gmm.h>
47	#include <VBox/vmm/gim.h>
48	#include <VBox/intnet.h>
49	#include <VBox/vmm/hm.h>
50	#include <VBox/param.h>
51	#include <VBox/err.h>
52	#include <VBox/version.h>
53	#include <VBox/log.h>
54
55	#include <iprt/asm-amd64-x86.h>
56	#include <iprt/assert.h>
57	#include <iprt/crc.h>
58	#include <iprt/initterm.h>
59	#include <iprt/mem.h>
60	#include <iprt/memobj.h>
61	#include <iprt/mp.h>
62	#include <iprt/once.h>
63	#include <iprt/semaphore.h>
64	#include <iprt/spinlock.h>
65	#include <iprt/stdarg.h>
66	#include <iprt/string.h>
67	#include <iprt/thread.h>
68	#include <iprt/timer.h>
69	#include <iprt/time.h>
70
71	#include "dtrace/VBoxVMM.h"
72
73
74	#if defined(_MSC_VER) && defined(RT_ARCH_AMD64) /** @todo check this with with VC7! */
75	# pragma intrinsic(_AddressOfReturnAddress)
76	#endif
77
78	#if defined(RT_OS_DARWIN) && ARCH_BITS == 32
79	# error "32-bit darwin is no longer supported. Go back to 4.3 or earlier!"
80	#endif
81
82
83	/*********************************************************************************************************************************
84	* Internal Functions *
85	*********************************************************************************************************************************/
86	RT_C_DECLS_BEGIN
87	#if defined(RT_ARCH_X86) && (defined(RT_OS_SOLARIS) \|\| defined(RT_OS_FREEBSD))
88	extern uint64_t __udivdi3(uint64_t, uint64_t);
89	extern uint64_t __umoddi3(uint64_t, uint64_t);
90	#endif
91	RT_C_DECLS_END
92	static int vmmR0UpdateLoggers(PGVM pGVM, VMCPUID idCpu, PVMMR0UPDATELOGGERSREQ pReq, size_t idxLogger);
93	static int vmmR0LogFlusher(PGVM pGVM);
94	static int vmmR0LogWaitFlushed(PGVM pGVM, VMCPUID idCpu, size_t idxLogger);
95	static int vmmR0InitLoggers(PGVM pGVM);
96	static void vmmR0CleanupLoggers(PGVM pGVM);
97
98
99	/*********************************************************************************************************************************
100	* Global Variables *
101	*********************************************************************************************************************************/
102	/** Drag in necessary library bits.
103	* The runtime lives here (in VMMR0.r0) and VBoxDDR0.r0 links against us. /
104	struct CLANG11WEIRDNOTHROW { PFNRT pfn; } g_VMMR0Deps[] =
105	{
106	{ (PFNRT)RTCrc32 },
107	{ (PFNRT)RTOnce },
108	#if defined(RT_ARCH_X86) && (defined(RT_OS_SOLARIS) \|\| defined(RT_OS_FREEBSD))
109	{ (PFNRT)__udivdi3 },
110	{ (PFNRT)__umoddi3 },
111	#endif
112	{ NULL }
113	};
114
115	#ifdef RT_OS_SOLARIS
116	/* Dependency information for the native solaris loader. */
117	extern "C" { char _depends_on[] = "vboxdrv"; }
118	#endif
119
120
121	/**
122	* Initialize the module.
123	* This is called when we're first loaded.
124	*
125	* @returns 0 on success.
126	* @returns VBox status on failure.
127	* @param hMod Image handle for use in APIs.
128	*/
129	DECLEXPORT(int) ModuleInit(void *hMod)
130	{
131	#ifdef VBOX_WITH_DTRACE_R0
132	/*
133	* The first thing to do is register the static tracepoints.
134	* (Deregistration is automatic.)
135	*/
136	int rc2 = SUPR0TracerRegisterModule(hMod, &g_VTGObjHeader);
137	if (RT_FAILURE(rc2))
138	return rc2;
139	#endif
140	LogFlow(("ModuleInit:\n"));
141
142	#ifdef VBOX_WITH_64ON32_CMOS_DEBUG
143	/*
144	* Display the CMOS debug code.
145	*/
146	ASMOutU8(0x72, 0x03);
147	uint8_t bDebugCode = ASMInU8(0x73);
148	LogRel(("CMOS Debug Code: %#x (%d)\n", bDebugCode, bDebugCode));
149	RTLogComPrintf("CMOS Debug Code: %#x (%d)\n", bDebugCode, bDebugCode);
150	#endif
151
152	/*
153	* Initialize the VMM, GVMM, GMM, HM, PGM (Darwin) and INTNET.
154	*/
155	int rc = vmmInitFormatTypes();
156	if (RT_SUCCESS(rc))
157	{
158	rc = GVMMR0Init();
159	if (RT_SUCCESS(rc))
160	{
161	rc = GMMR0Init();
162	if (RT_SUCCESS(rc))
163	{
164	rc = HMR0Init();
165	if (RT_SUCCESS(rc))
166	{
167	PDMR0Init(hMod);
168
169	rc = PGMRegisterStringFormatTypes();
170	if (RT_SUCCESS(rc))
171	{
172	rc = IntNetR0Init();
173	if (RT_SUCCESS(rc))
174	{
175	#ifdef VBOX_WITH_PCI_PASSTHROUGH
176	rc = PciRawR0Init();
177	#endif
178	if (RT_SUCCESS(rc))
179	{
180	rc = CPUMR0ModuleInit();
181	if (RT_SUCCESS(rc))
182	{
183	#ifdef VBOX_WITH_TRIPLE_FAULT_HACK
184	rc = vmmR0TripleFaultHackInit();
185	if (RT_SUCCESS(rc))
186	#endif
187	{
188	#ifdef VBOX_WITH_NEM_R0
189	rc = NEMR0Init();
190	if (RT_SUCCESS(rc))
191	#endif
192	{
193	LogFlow(("ModuleInit: returns success\n"));
194	return VINF_SUCCESS;
195	}
196	}
197
198	/*
199	* Bail out.
200	*/
201	#ifdef VBOX_WITH_TRIPLE_FAULT_HACK
202	vmmR0TripleFaultHackTerm();
203	#endif
204	}
205	else
206	LogRel(("ModuleInit: CPUMR0ModuleInit -> %Rrc\n", rc));
207	#ifdef VBOX_WITH_PCI_PASSTHROUGH
208	PciRawR0Term();
209	#endif
210	}
211	else
212	LogRel(("ModuleInit: PciRawR0Init -> %Rrc\n", rc));
213	IntNetR0Term();
214	}
215	else
216	LogRel(("ModuleInit: IntNetR0Init -> %Rrc\n", rc));
217	PGMDeregisterStringFormatTypes();
218	}
219	else
220	LogRel(("ModuleInit: PGMRegisterStringFormatTypes -> %Rrc\n", rc));
221	HMR0Term();
222	}
223	else
224	LogRel(("ModuleInit: HMR0Init -> %Rrc\n", rc));
225	GMMR0Term();
226	}
227	else
228	LogRel(("ModuleInit: GMMR0Init -> %Rrc\n", rc));
229	GVMMR0Term();
230	}
231	else
232	LogRel(("ModuleInit: GVMMR0Init -> %Rrc\n", rc));
233	vmmTermFormatTypes();
234	}
235	else
236	LogRel(("ModuleInit: vmmInitFormatTypes -> %Rrc\n", rc));
237
238	LogFlow(("ModuleInit: failed %Rrc\n", rc));
239	return rc;
240	}
241
242
243	/**
244	* Terminate the module.
245	* This is called when we're finally unloaded.
246	*
247	* @param hMod Image handle for use in APIs.
248	*/
249	DECLEXPORT(void) ModuleTerm(void *hMod)
250	{
251	NOREF(hMod);
252	LogFlow(("ModuleTerm:\n"));
253
254	/*
255	* Terminate the CPUM module (Local APIC cleanup).
256	*/
257	CPUMR0ModuleTerm();
258
259	/*
260	* Terminate the internal network service.
261	*/
262	IntNetR0Term();
263
264	/*
265	* PGM (Darwin), HM and PciRaw global cleanup.
266	*/
267	#ifdef VBOX_WITH_PCI_PASSTHROUGH
268	PciRawR0Term();
269	#endif
270	PGMDeregisterStringFormatTypes();
271	HMR0Term();
272	#ifdef VBOX_WITH_TRIPLE_FAULT_HACK
273	vmmR0TripleFaultHackTerm();
274	#endif
275	#ifdef VBOX_WITH_NEM_R0
276	NEMR0Term();
277	#endif
278
279	/*
280	* Destroy the GMM and GVMM instances.
281	*/
282	GMMR0Term();
283	GVMMR0Term();
284
285	vmmTermFormatTypes();
286	RTTermRunCallbacks(RTTERMREASON_UNLOAD, 0);
287
288	LogFlow(("ModuleTerm: returns\n"));
289	}
290
291
292	/**
293	* Initializes VMM specific members when the GVM structure is created,
294	* allocating loggers and stuff.
295	*
296	* The loggers are allocated here so that we can update their settings before
297	* doing VMMR0_DO_VMMR0_INIT and have correct logging at that time.
298	*
299	* @returns VBox status code.
300	* @param pGVM The global (ring-0) VM structure.
301	*/
302	VMMR0_INT_DECL(int) VMMR0InitPerVMData(PGVM pGVM)
303	{
304	AssertCompile(sizeof(pGVM->vmmr0.s) <= sizeof(pGVM->vmmr0.padding));
305
306	/*
307	* Initialize all members first.
308	*/
309	pGVM->vmmr0.s.fCalledInitVm = false;
310	pGVM->vmmr0.s.hMemObjLogger = NIL_RTR0MEMOBJ;
311	pGVM->vmmr0.s.hMapObjLogger = NIL_RTR0MEMOBJ;
312	pGVM->vmmr0.s.hMemObjReleaseLogger = NIL_RTR0MEMOBJ;
313	pGVM->vmmr0.s.hMapObjReleaseLogger = NIL_RTR0MEMOBJ;
314	pGVM->vmmr0.s.LogFlusher.hSpinlock = NIL_RTSPINLOCK;
315	pGVM->vmmr0.s.LogFlusher.hThread = NIL_RTNATIVETHREAD;
316	pGVM->vmmr0.s.LogFlusher.hEvent = NIL_RTSEMEVENT;
317	pGVM->vmmr0.s.LogFlusher.idxRingHead = 0;
318	pGVM->vmmr0.s.LogFlusher.idxRingTail = 0;
319	pGVM->vmmr0.s.LogFlusher.fThreadWaiting = false;
320
321	for (VMCPUID idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
322	{
323	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
324	Assert(pGVCpu->idHostCpu == NIL_RTCPUID);
325	Assert(pGVCpu->iHostCpuSet == UINT32_MAX);
326	pGVCpu->vmmr0.s.pPreemptState = NULL;
327	pGVCpu->vmmr0.s.hCtxHook = NIL_RTTHREADCTXHOOK;
328	pGVCpu->vmmr0.s.AssertJmpBuf.pMirrorBuf = &pGVCpu->vmm.s.AssertJmpBuf;
329	pGVCpu->vmmr0.s.AssertJmpBuf.pvStackBuf = &pGVCpu->vmm.s.abAssertStack[0];
330	pGVCpu->vmmr0.s.AssertJmpBuf.cbStackBuf = sizeof(pGVCpu->vmm.s.abAssertStack);
331
332	for (size_t iLogger = 0; iLogger < RT_ELEMENTS(pGVCpu->vmmr0.s.u.aLoggers); iLogger++)
333	pGVCpu->vmmr0.s.u.aLoggers[iLogger].hEventFlushWait = NIL_RTSEMEVENT;
334	}
335
336	/*
337	* Create the loggers.
338	*/
339	return vmmR0InitLoggers(pGVM);
340	}
341
342
343	/**
344	* Initiates the R0 driver for a particular VM instance.
345	*
346	* @returns VBox status code.
347	*
348	* @param pGVM The global (ring-0) VM structure.
349	* @param uSvnRev The SVN revision of the ring-3 part.
350	* @param uBuildType Build type indicator.
351	* @thread EMT(0)
352	*/
353	static int vmmR0InitVM(PGVM pGVM, uint32_t uSvnRev, uint32_t uBuildType)
354	{
355	/*
356	* Match the SVN revisions and build type.
357	*/
358	if (uSvnRev != VMMGetSvnRev())
359	{
360	LogRel(("VMMR0InitVM: Revision mismatch, r3=%d r0=%d\n", uSvnRev, VMMGetSvnRev()));
361	SUPR0Printf("VMMR0InitVM: Revision mismatch, r3=%d r0=%d\n", uSvnRev, VMMGetSvnRev());
362	return VERR_VMM_R0_VERSION_MISMATCH;
363	}
364	if (uBuildType != vmmGetBuildType())
365	{
366	LogRel(("VMMR0InitVM: Build type mismatch, r3=%#x r0=%#x\n", uBuildType, vmmGetBuildType()));
367	SUPR0Printf("VMMR0InitVM: Build type mismatch, r3=%#x r0=%#x\n", uBuildType, vmmGetBuildType());
368	return VERR_VMM_R0_VERSION_MISMATCH;
369	}
370
371	int rc = GVMMR0ValidateGVMandEMT(pGVM, 0 /idCpu/);
372	if (RT_FAILURE(rc))
373	return rc;
374
375	/* Don't allow this to be called more than once. */
376	if (!pGVM->vmmr0.s.fCalledInitVm)
377	pGVM->vmmr0.s.fCalledInitVm = true;
378	else
379	return VERR_ALREADY_INITIALIZED;
380
381	#ifdef LOG_ENABLED
382
383	/*
384	* Register the EMT R0 logger instance for VCPU 0.
385	*/
386	PVMCPUCC pVCpu = VMCC_GET_CPU_0(pGVM);
387	if (pVCpu->vmmr0.s.u.s.Logger.pLogger)
388	{
389	# if 0 /* testing of the logger. */
390	LogCom(("vmmR0InitVM: before %p\n", RTLogDefaultInstance()));
391	LogCom(("vmmR0InitVM: pfnFlush=%p actual=%p\n", pR0Logger->Logger.pfnFlush, vmmR0LoggerFlush));
392	LogCom(("vmmR0InitVM: pfnLogger=%p actual=%p\n", pR0Logger->Logger.pfnLogger, vmmR0LoggerWrapper));
393	LogCom(("vmmR0InitVM: offScratch=%d fFlags=%#x fDestFlags=%#x\n", pR0Logger->Logger.offScratch, pR0Logger->Logger.fFlags, pR0Logger->Logger.fDestFlags));
394
395	RTLogSetDefaultInstanceThread(&pR0Logger->Logger, (uintptr_t)pGVM->pSession);
396	LogCom(("vmmR0InitVM: after %p reg\n", RTLogDefaultInstance()));
397	RTLogSetDefaultInstanceThread(NULL, pGVM->pSession);
398	LogCom(("vmmR0InitVM: after %p dereg\n", RTLogDefaultInstance()));
399
400	pR0Logger->Logger.pfnLogger("hello ring-0 logger\n");
401	LogCom(("vmmR0InitVM: returned successfully from direct logger call.\n"));
402	pR0Logger->Logger.pfnFlush(&pR0Logger->Logger);
403	LogCom(("vmmR0InitVM: returned successfully from direct flush call.\n"));
404
405	RTLogSetDefaultInstanceThread(&pR0Logger->Logger, (uintptr_t)pGVM->pSession);
406	LogCom(("vmmR0InitVM: after %p reg2\n", RTLogDefaultInstance()));
407	pR0Logger->Logger.pfnLogger("hello ring-0 logger\n");
408	LogCom(("vmmR0InitVM: returned successfully from direct logger call (2). offScratch=%d\n", pR0Logger->Logger.offScratch));
409	RTLogSetDefaultInstanceThread(NULL, pGVM->pSession);
410	LogCom(("vmmR0InitVM: after %p dereg2\n", RTLogDefaultInstance()));
411
412	RTLogLoggerEx(&pR0Logger->Logger, 0, ~0U, "hello ring-0 logger (RTLogLoggerEx)\n");
413	LogCom(("vmmR0InitVM: RTLogLoggerEx returned fine offScratch=%d\n", pR0Logger->Logger.offScratch));
414
415	RTLogSetDefaultInstanceThread(&pR0Logger->Logger, (uintptr_t)pGVM->pSession);
416	RTLogPrintf("hello ring-0 logger (RTLogPrintf)\n");
417	LogCom(("vmmR0InitVM: RTLogPrintf returned fine offScratch=%d\n", pR0Logger->Logger.offScratch));
418	# endif
419	# ifdef VBOX_WITH_R0_LOGGING
420	Log(("Switching to per-thread logging instance %p (key=%p)\n", pVCpu->vmmr0.s.u.s.Logger.pLogger, pGVM->pSession));
421	RTLogSetDefaultInstanceThread(pVCpu->vmmr0.s.u.s.Logger.pLogger, (uintptr_t)pGVM->pSession);
422	pVCpu->vmmr0.s.u.s.Logger.fRegistered = true;
423	# endif
424	}
425	#endif /* LOG_ENABLED */
426
427	/*
428	* Check if the host supports high resolution timers or not.
429	*/
430	if ( pGVM->vmm.s.fUsePeriodicPreemptionTimers
431	&& !RTTimerCanDoHighResolution())
432	pGVM->vmm.s.fUsePeriodicPreemptionTimers = false;
433
434	/*
435	* Initialize the per VM data for GVMM and GMM.
436	*/
437	rc = GVMMR0InitVM(pGVM);
438	if (RT_SUCCESS(rc))
439	{
440	/*
441	* Init HM, CPUM and PGM.
442	*/
443	rc = HMR0InitVM(pGVM);
444	if (RT_SUCCESS(rc))
445	{
446	rc = CPUMR0InitVM(pGVM);
447	if (RT_SUCCESS(rc))
448	{
449	rc = PGMR0InitVM(pGVM);
450	if (RT_SUCCESS(rc))
451	{
452	rc = EMR0InitVM(pGVM);
453	if (RT_SUCCESS(rc))
454	{
455	rc = IEMR0InitVM(pGVM);
456	if (RT_SUCCESS(rc))
457	{
458	rc = IOMR0InitVM(pGVM);
459	if (RT_SUCCESS(rc))
460	{
461	#ifdef VBOX_WITH_PCI_PASSTHROUGH
462	rc = PciRawR0InitVM(pGVM);
463	#endif
464	if (RT_SUCCESS(rc))
465	{
466	rc = GIMR0InitVM(pGVM);
467	if (RT_SUCCESS(rc))
468	{
469	GVMMR0DoneInitVM(pGVM);
470	PGMR0DoneInitVM(pGVM);
471
472	/*
473	* Collect a bit of info for the VM release log.
474	*/
475	pGVM->vmm.s.fIsPreemptPendingApiTrusty = RTThreadPreemptIsPendingTrusty();
476	pGVM->vmm.s.fIsPreemptPossible = RTThreadPreemptIsPossible();;
477	return rc;
478
479	/* bail out*/
480	//GIMR0TermVM(pGVM);
481	}
482	#ifdef VBOX_WITH_PCI_PASSTHROUGH
483	PciRawR0TermVM(pGVM);
484	#endif
485	}
486	}
487	}
488	}
489	}
490	}
491	HMR0TermVM(pGVM);
492	}
493	}
494
495	RTLogSetDefaultInstanceThread(NULL, (uintptr_t)pGVM->pSession);
496	return rc;
497	}
498
499
500	/**
501	* Does EMT specific VM initialization.
502	*
503	* @returns VBox status code.
504	* @param pGVM The ring-0 VM structure.
505	* @param idCpu The EMT that's calling.
506	*/
507	static int vmmR0InitVMEmt(PGVM pGVM, VMCPUID idCpu)
508	{
509	/* Paranoia (caller checked these already). */
510	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID);
511	AssertReturn(pGVM->aCpus[idCpu].hEMT == RTThreadNativeSelf(), VERR_INVALID_CPU_ID);
512
513	#if defined(LOG_ENABLED) && defined(VBOX_WITH_R0_LOGGING)
514	/*
515	* Registration of ring 0 loggers.
516	*/
517	PVMCPUCC pVCpu = &pGVM->aCpus[idCpu];
518	if ( pVCpu->vmmr0.s.u.s.Logger.pLogger
519	&& !pVCpu->vmmr0.s.u.s.Logger.fRegistered)
520	{
521	RTLogSetDefaultInstanceThread(pVCpu->vmmr0.s.u.s.Logger.pLogger, (uintptr_t)pGVM->pSession);
522	pVCpu->vmmr0.s.u.s.Logger.fRegistered = true;
523	}
524	#endif
525
526	return VINF_SUCCESS;
527	}
528
529
530
531	/**
532	* Terminates the R0 bits for a particular VM instance.
533	*
534	* This is normally called by ring-3 as part of the VM termination process, but
535	* may alternatively be called during the support driver session cleanup when
536	* the VM object is destroyed (see GVMM).
537	*
538	* @returns VBox status code.
539	*
540	* @param pGVM The global (ring-0) VM structure.
541	* @param idCpu Set to 0 if EMT(0) or NIL_VMCPUID if session cleanup
542	* thread.
543	* @thread EMT(0) or session clean up thread.
544	*/
545	VMMR0_INT_DECL(int) VMMR0TermVM(PGVM pGVM, VMCPUID idCpu)
546	{
547	/*
548	* Check EMT(0) claim if we're called from userland.
549	*/
550	if (idCpu != NIL_VMCPUID)
551	{
552	AssertReturn(idCpu == 0, VERR_INVALID_CPU_ID);
553	int rc = GVMMR0ValidateGVMandEMT(pGVM, idCpu);
554	if (RT_FAILURE(rc))
555	return rc;
556	}
557
558	#ifdef VBOX_WITH_PCI_PASSTHROUGH
559	PciRawR0TermVM(pGVM);
560	#endif
561
562	/*
563	* Tell GVMM what we're up to and check that we only do this once.
564	*/
565	if (GVMMR0DoingTermVM(pGVM))
566	{
567	GIMR0TermVM(pGVM);
568
569	/** @todo I wish to call PGMR0PhysFlushHandyPages(pGVM, &pGVM->aCpus[idCpu])
570	* here to make sure we don't leak any shared pages if we crash... */
571	HMR0TermVM(pGVM);
572	}
573
574	/*
575	* Deregister the logger for this EMT.
576	*/
577	RTLogSetDefaultInstanceThread(NULL, (uintptr_t)pGVM->pSession);
578
579	/*
580	* Start log flusher thread termination.
581	*/
582	ASMAtomicWriteBool(&pGVM->vmmr0.s.LogFlusher.fThreadShutdown, true);
583	if (pGVM->vmmr0.s.LogFlusher.hEvent != NIL_RTSEMEVENT)
584	RTSemEventSignal(pGVM->vmmr0.s.LogFlusher.hEvent);
585
586	return VINF_SUCCESS;
587	}
588
589
590	/**
591	* This is called at the end of gvmmR0CleanupVM().
592	*
593	* @param pGVM The global (ring-0) VM structure.
594	*/
595	VMMR0_INT_DECL(void) VMMR0CleanupVM(PGVM pGVM)
596	{
597	AssertCompile(NIL_RTTHREADCTXHOOK == (RTTHREADCTXHOOK)0); /* Depends on zero initialized memory working for NIL at the moment. */
598	for (VMCPUID idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
599	{
600	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
601
602	/** @todo Can we busy wait here for all thread-context hooks to be
603	* deregistered before releasing (destroying) it? Only until we find a
604	* solution for not deregistering hooks everytime we're leaving HMR0
605	* context. */
606	VMMR0ThreadCtxHookDestroyForEmt(pGVCpu);
607	}
608
609	vmmR0CleanupLoggers(pGVM);
610	}
611
612
613	/**
614	* An interrupt or unhalt force flag is set, deal with it.
615	*
616	* @returns VINF_SUCCESS (or VINF_EM_HALT).
617	* @param pVCpu The cross context virtual CPU structure.
618	* @param uMWait Result from EMMonitorWaitIsActive().
619	* @param enmInterruptibility Guest CPU interruptbility level.
620	*/
621	static int vmmR0DoHaltInterrupt(PVMCPUCC pVCpu, unsigned uMWait, CPUMINTERRUPTIBILITY enmInterruptibility)
622	{
623	Assert(!TRPMHasTrap(pVCpu));
624	Assert( enmInterruptibility > CPUMINTERRUPTIBILITY_INVALID
625	&& enmInterruptibility < CPUMINTERRUPTIBILITY_END);
626
627	/*
628	* Pending interrupts w/o any SMIs or NMIs? That the usual case.
629	*/
630	if ( VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC)
631	&& !VMCPU_FF_IS_ANY_SET(pVCpu, VMCPU_FF_INTERRUPT_SMI \| VMCPU_FF_INTERRUPT_NMI))
632	{
633	if (enmInterruptibility <= CPUMINTERRUPTIBILITY_UNRESTRAINED)
634	{
635	uint8_t u8Interrupt = 0;
636	int rc = PDMGetInterrupt(pVCpu, &u8Interrupt);
637	Log(("vmmR0DoHaltInterrupt: CPU%d u8Interrupt=%d (%#x) rc=%Rrc\n", pVCpu->idCpu, u8Interrupt, u8Interrupt, rc));
638	if (RT_SUCCESS(rc))
639	{
640	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_UNHALT);
641
642	rc = TRPMAssertTrap(pVCpu, u8Interrupt, TRPM_HARDWARE_INT);
643	AssertRCSuccess(rc);
644	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltExec);
645	return rc;
646	}
647	}
648	}
649	/*
650	* SMI is not implemented yet, at least not here.
651	*/
652	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_SMI))
653	{
654	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #3\n", pVCpu->idCpu));
655	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
656	return VINF_EM_HALT;
657	}
658	/*
659	* NMI.
660	*/
661	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NMI))
662	{
663	if (enmInterruptibility < CPUMINTERRUPTIBILITY_NMI_INHIBIT)
664	{
665	/** @todo later. */
666	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #2 (uMWait=%u enmInt=%d)\n", pVCpu->idCpu, uMWait, enmInterruptibility));
667	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
668	return VINF_EM_HALT;
669	}
670	}
671	/*
672	* Nested-guest virtual interrupt.
673	*/
674	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_INTERRUPT_NESTED_GUEST))
675	{
676	if (enmInterruptibility < CPUMINTERRUPTIBILITY_VIRT_INT_DISABLED)
677	{
678	/** @todo NSTVMX: NSTSVM: Remember, we might have to check and perform VM-exits
679	* here before injecting the virtual interrupt. See emR3ForcedActions
680	* for details. */
681	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #1 (uMWait=%u enmInt=%d)\n", pVCpu->idCpu, uMWait, enmInterruptibility));
682	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
683	return VINF_EM_HALT;
684	}
685	}
686
687	if (VMCPU_FF_TEST_AND_CLEAR(pVCpu, VMCPU_FF_UNHALT))
688	{
689	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltExec);
690	Log11(("vmmR0DoHaltInterrupt: CPU%d success VINF_SUCCESS (UNHALT)\n", pVCpu->idCpu));
691	return VINF_SUCCESS;
692	}
693	if (uMWait > 1)
694	{
695	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltExec);
696	Log11(("vmmR0DoHaltInterrupt: CPU%d success VINF_SUCCESS (uMWait=%u > 1)\n", pVCpu->idCpu, uMWait));
697	return VINF_SUCCESS;
698	}
699
700	Log12(("vmmR0DoHaltInterrupt: CPU%d failed #0 (uMWait=%u enmInt=%d)\n", pVCpu->idCpu, uMWait, enmInterruptibility));
701	STAM_REL_COUNTER_INC(&pVCpu->vmm.s.StatR0HaltToR3);
702	return VINF_EM_HALT;
703	}
704
705
706	/**
707	* This does one round of vmR3HaltGlobal1Halt().
708	*
709	* The rational here is that we'll reduce latency in interrupt situations if we
710	* don't go to ring-3 immediately on a VINF_EM_HALT (guest executed HLT or
711	* MWAIT), but do one round of blocking here instead and hope the interrupt is
712	* raised in the meanwhile.
713	*
714	* If we go to ring-3 we'll quit the inner HM/NEM loop in EM and end up in the
715	* outer loop, which will then call VMR3WaitHalted() and that in turn will do a
716	* ring-0 call (unless we're too close to a timer event). When the interrupt
717	* wakes us up, we'll return from ring-0 and EM will by instinct do a
718	* rescheduling (because of raw-mode) before it resumes the HM/NEM loop and gets
719	* back to VMMR0EntryFast().
720	*
721	* @returns VINF_SUCCESS or VINF_EM_HALT.
722	* @param pGVM The ring-0 VM structure.
723	* @param pGVCpu The ring-0 virtual CPU structure.
724	*
725	* @todo r=bird: All the blocking/waiting and EMT managment should move out of
726	* the VM module, probably to VMM. Then this would be more weird wrt
727	* parameters and statistics.
728	*/
729	static int vmmR0DoHalt(PGVM pGVM, PGVMCPU pGVCpu)
730	{
731	/*
732	* Do spin stat historization.
733	*/
734	if (++pGVCpu->vmm.s.cR0Halts & 0xff)
735	{ /* likely */ }
736	else if (pGVCpu->vmm.s.cR0HaltsSucceeded > pGVCpu->vmm.s.cR0HaltsToRing3)
737	{
738	pGVCpu->vmm.s.cR0HaltsSucceeded = 2;
739	pGVCpu->vmm.s.cR0HaltsToRing3 = 0;
740	}
741	else
742	{
743	pGVCpu->vmm.s.cR0HaltsSucceeded = 0;
744	pGVCpu->vmm.s.cR0HaltsToRing3 = 2;
745	}
746
747	/*
748	* Flags that makes us go to ring-3.
749	*/
750	uint32_t const fVmFFs = VM_FF_TM_VIRTUAL_SYNC \| VM_FF_PDM_QUEUES \| VM_FF_PDM_DMA
751	\| VM_FF_DBGF \| VM_FF_REQUEST \| VM_FF_CHECK_VM_STATE
752	\| VM_FF_RESET \| VM_FF_EMT_RENDEZVOUS \| VM_FF_PGM_NEED_HANDY_PAGES
753	\| VM_FF_PGM_NO_MEMORY \| VM_FF_DEBUG_SUSPEND;
754	uint64_t const fCpuFFs = VMCPU_FF_TIMER \| VMCPU_FF_PDM_CRITSECT \| VMCPU_FF_IEM
755	\| VMCPU_FF_REQUEST \| VMCPU_FF_DBGF \| VMCPU_FF_HM_UPDATE_CR3
756	\| VMCPU_FF_PGM_SYNC_CR3 \| VMCPU_FF_PGM_SYNC_CR3_NON_GLOBAL
757	\| VMCPU_FF_TO_R3 \| VMCPU_FF_IOM;
758
759	/*
760	* Check preconditions.
761	*/
762	unsigned const uMWait = EMMonitorWaitIsActive(pGVCpu);
763	CPUMINTERRUPTIBILITY const enmInterruptibility = CPUMGetGuestInterruptibility(pGVCpu);
764	if ( pGVCpu->vmm.s.fMayHaltInRing0
765	&& !TRPMHasTrap(pGVCpu)
766	&& ( enmInterruptibility == CPUMINTERRUPTIBILITY_UNRESTRAINED
767	\|\| uMWait > 1))
768	{
769	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
770	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
771	{
772	/*
773	* Interrupts pending already?
774	*/
775	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
776	APICUpdatePendingInterrupts(pGVCpu);
777
778	/*
779	* Flags that wake up from the halted state.
780	*/
781	uint64_t const fIntMask = VMCPU_FF_INTERRUPT_APIC \| VMCPU_FF_INTERRUPT_PIC \| VMCPU_FF_INTERRUPT_NESTED_GUEST
782	\| VMCPU_FF_INTERRUPT_NMI \| VMCPU_FF_INTERRUPT_SMI \| VMCPU_FF_UNHALT;
783
784	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
785	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
786	ASMNopPause();
787
788	/*
789	* Check out how long till the next timer event.
790	*/
791	uint64_t u64Delta;
792	uint64_t u64GipTime = TMTimerPollGIP(pGVM, pGVCpu, &u64Delta);
793
794	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
795	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
796	{
797	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
798	APICUpdatePendingInterrupts(pGVCpu);
799
800	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
801	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
802
803	/*
804	* Wait if there is enough time to the next timer event.
805	*/
806	if (u64Delta >= pGVCpu->vmm.s.cNsSpinBlockThreshold)
807	{
808	/* If there are few other CPU cores around, we will procrastinate a
809	little before going to sleep, hoping for some device raising an
810	interrupt or similar. Though, the best thing here would be to
811	dynamically adjust the spin count according to its usfulness or
812	something... */
813	if ( pGVCpu->vmm.s.cR0HaltsSucceeded > pGVCpu->vmm.s.cR0HaltsToRing3
814	&& RTMpGetOnlineCount() >= 4)
815	{
816	/** @todo Figure out how we can skip this if it hasn't help recently...
817	* @bugref{9172#c12} */
818	uint32_t cSpinLoops = 42;
819	while (cSpinLoops-- > 0)
820	{
821	ASMNopPause();
822	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
823	APICUpdatePendingInterrupts(pGVCpu);
824	ASMNopPause();
825	if (VM_FF_IS_ANY_SET(pGVM, fVmFFs))
826	{
827	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3FromSpin);
828	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3);
829	return VINF_EM_HALT;
830	}
831	ASMNopPause();
832	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
833	{
834	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3FromSpin);
835	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3);
836	return VINF_EM_HALT;
837	}
838	ASMNopPause();
839	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
840	{
841	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltExecFromSpin);
842	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
843	}
844	ASMNopPause();
845	}
846	}
847
848	/*
849	* We have to set the state to VMCPUSTATE_STARTED_HALTED here so ring-3
850	* knows when to notify us (cannot access VMINTUSERPERVMCPU::fWait from here).
851	* After changing the state we must recheck the force flags of course.
852	*/
853	if (VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED_HALTED, VMCPUSTATE_STARTED))
854	{
855	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
856	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
857	{
858	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
859	APICUpdatePendingInterrupts(pGVCpu);
860
861	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
862	{
863	VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED, VMCPUSTATE_STARTED_HALTED);
864	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
865	}
866
867	/* Okay, block! */
868	uint64_t const u64StartSchedHalt = RTTimeNanoTS();
869	int rc = GVMMR0SchedHalt(pGVM, pGVCpu, u64GipTime);
870	uint64_t const u64EndSchedHalt = RTTimeNanoTS();
871	uint64_t const cNsElapsedSchedHalt = u64EndSchedHalt - u64StartSchedHalt;
872	Log10(("vmmR0DoHalt: CPU%d: halted %llu ns\n", pGVCpu->idCpu, cNsElapsedSchedHalt));
873
874	VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED, VMCPUSTATE_STARTED_HALTED);
875	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlock, cNsElapsedSchedHalt);
876	if ( rc == VINF_SUCCESS
877	\|\| rc == VERR_INTERRUPTED)
878	{
879	/* Keep some stats like ring-3 does. */
880	int64_t const cNsOverslept = u64EndSchedHalt - u64GipTime;
881	if (cNsOverslept > 50000)
882	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlockOverslept, cNsOverslept);
883	else if (cNsOverslept < -50000)
884	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlockInsomnia, cNsElapsedSchedHalt);
885	else
886	STAM_REL_PROFILE_ADD_PERIOD(&pGVCpu->vmm.s.StatR0HaltBlockOnTime, cNsElapsedSchedHalt);
887
888	/*
889	* Recheck whether we can resume execution or have to go to ring-3.
890	*/
891	if ( !VM_FF_IS_ANY_SET(pGVM, fVmFFs)
892	&& !VMCPU_FF_IS_ANY_SET(pGVCpu, fCpuFFs))
893	{
894	if (VMCPU_FF_TEST_AND_CLEAR(pGVCpu, VMCPU_FF_UPDATE_APIC))
895	APICUpdatePendingInterrupts(pGVCpu);
896	if (VMCPU_FF_IS_ANY_SET(pGVCpu, fIntMask))
897	{
898	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltExecFromBlock);
899	return vmmR0DoHaltInterrupt(pGVCpu, uMWait, enmInterruptibility);
900	}
901	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PostNoInt);
902	Log12(("vmmR0DoHalt: CPU%d post #2 - No pending interrupt\n", pGVCpu->idCpu));
903	}
904	else
905	{
906	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PostPendingFF);
907	Log12(("vmmR0DoHalt: CPU%d post #1 - Pending FF\n", pGVCpu->idCpu));
908	}
909	}
910	else
911	{
912	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3Other);
913	Log12(("vmmR0DoHalt: CPU%d GVMMR0SchedHalt failed: %Rrc\n", pGVCpu->idCpu, rc));
914	}
915	}
916	else
917	{
918	VMCPU_CMPXCHG_STATE(pGVCpu, VMCPUSTATE_STARTED, VMCPUSTATE_STARTED_HALTED);
919	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PendingFF);
920	Log12(("vmmR0DoHalt: CPU%d failed #5 - Pending FF\n", pGVCpu->idCpu));
921	}
922	}
923	else
924	{
925	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3Other);
926	Log12(("vmmR0DoHalt: CPU%d failed #4 - enmState=%d\n", pGVCpu->idCpu, VMCPU_GET_STATE(pGVCpu)));
927	}
928	}
929	else
930	{
931	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3SmallDelta);
932	Log12(("vmmR0DoHalt: CPU%d failed #3 - delta too small: %RU64\n", pGVCpu->idCpu, u64Delta));
933	}
934	}
935	else
936	{
937	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PendingFF);
938	Log12(("vmmR0DoHalt: CPU%d failed #2 - Pending FF\n", pGVCpu->idCpu));
939	}
940	}
941	else
942	{
943	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3PendingFF);
944	Log12(("vmmR0DoHalt: CPU%d failed #1 - Pending FF\n", pGVCpu->idCpu));
945	}
946	}
947	else
948	{
949	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3Other);
950	Log12(("vmmR0DoHalt: CPU%d failed #0 - fMayHaltInRing0=%d TRPMHasTrap=%d enmInt=%d uMWait=%u\n",
951	pGVCpu->idCpu, pGVCpu->vmm.s.fMayHaltInRing0, TRPMHasTrap(pGVCpu), enmInterruptibility, uMWait));
952	}
953
954	STAM_REL_COUNTER_INC(&pGVCpu->vmm.s.StatR0HaltToR3);
955	return VINF_EM_HALT;
956	}
957
958
959	/**
960	* VMM ring-0 thread-context callback.
961	*
962	* This does common HM state updating and calls the HM-specific thread-context
963	* callback.
964	*
965	* This is used together with RTThreadCtxHookCreate() on platforms which
966	* supports it, and directly from VMMR0EmtPrepareForBlocking() and
967	* VMMR0EmtResumeAfterBlocking() on platforms which don't.
968	*
969	* @param enmEvent The thread-context event.
970	* @param pvUser Opaque pointer to the VMCPU.
971	*
972	* @thread EMT(pvUser)
973	*/
974	static DECLCALLBACK(void) vmmR0ThreadCtxCallback(RTTHREADCTXEVENT enmEvent, void *pvUser)
975	{
976	PVMCPUCC pVCpu = (PVMCPUCC)pvUser;
977
978	switch (enmEvent)
979	{
980	case RTTHREADCTXEVENT_IN:
981	{
982	/*
983	* Linux may call us with preemption enabled (really!) but technically we
984	* cannot get preempted here, otherwise we end up in an infinite recursion
985	* scenario (i.e. preempted in resume hook -> preempt hook -> resume hook...
986	* ad infinitum). Let's just disable preemption for now...
987	*/
988	/** @todo r=bird: I don't believe the above. The linux code is clearly enabling
989	* preemption after doing the callout (one or two functions up the
990	* call chain). */
991	/** @todo r=ramshankar: See @bugref{5313#c30}. */
992	RTTHREADPREEMPTSTATE ParanoidPreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
993	RTThreadPreemptDisable(&ParanoidPreemptState);
994
995	/* We need to update the VCPU <-> host CPU mapping. */
996	RTCPUID idHostCpu;
997	uint32_t iHostCpuSet = RTMpCurSetIndexAndId(&idHostCpu);
998	pVCpu->iHostCpuSet = iHostCpuSet;
999	ASMAtomicWriteU32(&pVCpu->idHostCpu, idHostCpu);
1000
1001	/* In the very unlikely event that the GIP delta for the CPU we're
1002	rescheduled needs calculating, try force a return to ring-3.
1003	We unfortunately cannot do the measurements right here. */
1004	if (RT_LIKELY(!SUPIsTscDeltaAvailableForCpuSetIndex(iHostCpuSet)))
1005	{ /* likely */ }
1006	else
1007	VMCPU_FF_SET(pVCpu, VMCPU_FF_TO_R3);
1008
1009	/* Invoke the HM-specific thread-context callback. */
1010	HMR0ThreadCtxCallback(enmEvent, pvUser);
1011
1012	/* Restore preemption. */
1013	RTThreadPreemptRestore(&ParanoidPreemptState);
1014	break;
1015	}
1016
1017	case RTTHREADCTXEVENT_OUT:
1018	{
1019	/* Invoke the HM-specific thread-context callback. */
1020	HMR0ThreadCtxCallback(enmEvent, pvUser);
1021
1022	/*
1023	* Sigh. See VMMGetCpu() used by VMCPU_ASSERT_EMT(). We cannot let several VCPUs
1024	* have the same host CPU associated with it.
1025	*/
1026	pVCpu->iHostCpuSet = UINT32_MAX;
1027	ASMAtomicWriteU32(&pVCpu->idHostCpu, NIL_RTCPUID);
1028	break;
1029	}
1030
1031	default:
1032	/* Invoke the HM-specific thread-context callback. */
1033	HMR0ThreadCtxCallback(enmEvent, pvUser);
1034	break;
1035	}
1036	}
1037
1038
1039	/**
1040	* Creates thread switching hook for the current EMT thread.
1041	*
1042	* This is called by GVMMR0CreateVM and GVMMR0RegisterVCpu. If the host
1043	* platform does not implement switcher hooks, no hooks will be create and the
1044	* member set to NIL_RTTHREADCTXHOOK.
1045	*
1046	* @returns VBox status code.
1047	* @param pVCpu The cross context virtual CPU structure.
1048	* @thread EMT(pVCpu)
1049	*/
1050	VMMR0_INT_DECL(int) VMMR0ThreadCtxHookCreateForEmt(PVMCPUCC pVCpu)
1051	{
1052	VMCPU_ASSERT_EMT(pVCpu);
1053	Assert(pVCpu->vmmr0.s.hCtxHook == NIL_RTTHREADCTXHOOK);
1054
1055	#if 1 /* To disable this stuff change to zero. */
1056	int rc = RTThreadCtxHookCreate(&pVCpu->vmmr0.s.hCtxHook, 0, vmmR0ThreadCtxCallback, pVCpu);
1057	if (RT_SUCCESS(rc))
1058	{
1059	pVCpu->pGVM->vmm.s.fIsUsingContextHooks = true;
1060	return rc;
1061	}
1062	#else
1063	RT_NOREF(vmmR0ThreadCtxCallback);
1064	int rc = VERR_NOT_SUPPORTED;
1065	#endif
1066
1067	pVCpu->vmmr0.s.hCtxHook = NIL_RTTHREADCTXHOOK;
1068	pVCpu->pGVM->vmm.s.fIsUsingContextHooks = false;
1069	if (rc == VERR_NOT_SUPPORTED)
1070	return VINF_SUCCESS;
1071
1072	LogRelMax(32, ("RTThreadCtxHookCreate failed! rc=%Rrc pVCpu=%p idCpu=%RU32\n", rc, pVCpu, pVCpu->idCpu));
1073	return VINF_SUCCESS; /* Just ignore it, we can live without context hooks. */
1074	}
1075
1076
1077	/**
1078	* Destroys the thread switching hook for the specified VCPU.
1079	*
1080	* @param pVCpu The cross context virtual CPU structure.
1081	* @remarks Can be called from any thread.
1082	*/
1083	VMMR0_INT_DECL(void) VMMR0ThreadCtxHookDestroyForEmt(PVMCPUCC pVCpu)
1084	{
1085	int rc = RTThreadCtxHookDestroy(pVCpu->vmmr0.s.hCtxHook);
1086	AssertRC(rc);
1087	pVCpu->vmmr0.s.hCtxHook = NIL_RTTHREADCTXHOOK;
1088	}
1089
1090
1091	/**
1092	* Disables the thread switching hook for this VCPU (if we got one).
1093	*
1094	* @param pVCpu The cross context virtual CPU structure.
1095	* @thread EMT(pVCpu)
1096	*
1097	* @remarks This also clears GVMCPU::idHostCpu, so the mapping is invalid after
1098	* this call. This means you have to be careful with what you do!
1099	*/
1100	VMMR0_INT_DECL(void) VMMR0ThreadCtxHookDisable(PVMCPUCC pVCpu)
1101	{
1102	/*
1103	* Clear the VCPU <-> host CPU mapping as we've left HM context.
1104	* @bugref{7726#c19} explains the need for this trick:
1105	*
1106	* VMXR0CallRing3Callback/SVMR0CallRing3Callback &
1107	* hmR0VmxLeaveSession/hmR0SvmLeaveSession disables context hooks during
1108	* longjmp & normal return to ring-3, which opens a window where we may be
1109	* rescheduled without changing GVMCPUID::idHostCpu and cause confusion if
1110	* the CPU starts executing a different EMT. Both functions first disables
1111	* preemption and then calls HMR0LeaveCpu which invalids idHostCpu, leaving
1112	* an opening for getting preempted.
1113	*/
1114	/** @todo Make HM not need this API! Then we could leave the hooks enabled
1115	* all the time. */
1116
1117	/*
1118	* Disable the context hook, if we got one.
1119	*/
1120	if (pVCpu->vmmr0.s.hCtxHook != NIL_RTTHREADCTXHOOK)
1121	{
1122	Assert(!RTThreadPreemptIsEnabled(NIL_RTTHREAD));
1123	ASMAtomicWriteU32(&pVCpu->idHostCpu, NIL_RTCPUID);
1124	int rc = RTThreadCtxHookDisable(pVCpu->vmmr0.s.hCtxHook);
1125	AssertRC(rc);
1126	}
1127	}
1128
1129
1130	/**
1131	* Internal version of VMMR0ThreadCtxHooksAreRegistered.
1132	*
1133	* @returns true if registered, false otherwise.
1134	* @param pVCpu The cross context virtual CPU structure.
1135	*/
1136	DECLINLINE(bool) vmmR0ThreadCtxHookIsEnabled(PVMCPUCC pVCpu)
1137	{
1138	return RTThreadCtxHookIsEnabled(pVCpu->vmmr0.s.hCtxHook);
1139	}
1140
1141
1142	/**
1143	* Whether thread-context hooks are registered for this VCPU.
1144	*
1145	* @returns true if registered, false otherwise.
1146	* @param pVCpu The cross context virtual CPU structure.
1147	*/
1148	VMMR0_INT_DECL(bool) VMMR0ThreadCtxHookIsEnabled(PVMCPUCC pVCpu)
1149	{
1150	return vmmR0ThreadCtxHookIsEnabled(pVCpu);
1151	}
1152
1153
1154	/**
1155	* Returns the ring-0 release logger instance.
1156	*
1157	* @returns Pointer to release logger, NULL if not configured.
1158	* @param pVCpu The cross context virtual CPU structure of the caller.
1159	* @thread EMT(pVCpu)
1160	*/
1161	VMMR0_INT_DECL(PRTLOGGER) VMMR0GetReleaseLogger(PVMCPUCC pVCpu)
1162	{
1163	return pVCpu->vmmr0.s.u.s.RelLogger.pLogger;
1164	}
1165
1166
1167	#ifdef VBOX_WITH_STATISTICS
1168	/**
1169	* Record return code statistics
1170	* @param pVM The cross context VM structure.
1171	* @param pVCpu The cross context virtual CPU structure.
1172	* @param rc The status code.
1173	*/
1174	static void vmmR0RecordRC(PVMCC pVM, PVMCPUCC pVCpu, int rc)
1175	{
1176	/*
1177	* Collect statistics.
1178	*/
1179	switch (rc)
1180	{
1181	case VINF_SUCCESS:
1182	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetNormal);
1183	break;
1184	case VINF_EM_RAW_INTERRUPT:
1185	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetInterrupt);
1186	break;
1187	case VINF_EM_RAW_INTERRUPT_HYPER:
1188	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetInterruptHyper);
1189	break;
1190	case VINF_EM_RAW_GUEST_TRAP:
1191	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetGuestTrap);
1192	break;
1193	case VINF_EM_RAW_RING_SWITCH:
1194	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetRingSwitch);
1195	break;
1196	case VINF_EM_RAW_RING_SWITCH_INT:
1197	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetRingSwitchInt);
1198	break;
1199	case VINF_EM_RAW_STALE_SELECTOR:
1200	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetStaleSelector);
1201	break;
1202	case VINF_EM_RAW_IRET_TRAP:
1203	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIRETTrap);
1204	break;
1205	case VINF_IOM_R3_IOPORT_READ:
1206	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIORead);
1207	break;
1208	case VINF_IOM_R3_IOPORT_WRITE:
1209	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIOWrite);
1210	break;
1211	case VINF_IOM_R3_IOPORT_COMMIT_WRITE:
1212	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIOCommitWrite);
1213	break;
1214	case VINF_IOM_R3_MMIO_READ:
1215	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIORead);
1216	break;
1217	case VINF_IOM_R3_MMIO_WRITE:
1218	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOWrite);
1219	break;
1220	case VINF_IOM_R3_MMIO_COMMIT_WRITE:
1221	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOCommitWrite);
1222	break;
1223	case VINF_IOM_R3_MMIO_READ_WRITE:
1224	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOReadWrite);
1225	break;
1226	case VINF_PATM_HC_MMIO_PATCH_READ:
1227	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOPatchRead);
1228	break;
1229	case VINF_PATM_HC_MMIO_PATCH_WRITE:
1230	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMMIOPatchWrite);
1231	break;
1232	case VINF_CPUM_R3_MSR_READ:
1233	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMSRRead);
1234	break;
1235	case VINF_CPUM_R3_MSR_WRITE:
1236	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMSRWrite);
1237	break;
1238	case VINF_EM_RAW_EMULATE_INSTR:
1239	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetEmulate);
1240	break;
1241	case VINF_PATCH_EMULATE_INSTR:
1242	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchEmulate);
1243	break;
1244	case VINF_EM_RAW_EMULATE_INSTR_LDT_FAULT:
1245	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetLDTFault);
1246	break;
1247	case VINF_EM_RAW_EMULATE_INSTR_GDT_FAULT:
1248	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetGDTFault);
1249	break;
1250	case VINF_EM_RAW_EMULATE_INSTR_IDT_FAULT:
1251	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetIDTFault);
1252	break;
1253	case VINF_EM_RAW_EMULATE_INSTR_TSS_FAULT:
1254	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetTSSFault);
1255	break;
1256	case VINF_CSAM_PENDING_ACTION:
1257	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetCSAMTask);
1258	break;
1259	case VINF_PGM_SYNC_CR3:
1260	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetSyncCR3);
1261	break;
1262	case VINF_PATM_PATCH_INT3:
1263	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchInt3);
1264	break;
1265	case VINF_PATM_PATCH_TRAP_PF:
1266	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchPF);
1267	break;
1268	case VINF_PATM_PATCH_TRAP_GP:
1269	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchGP);
1270	break;
1271	case VINF_PATM_PENDING_IRQ_AFTER_IRET:
1272	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchIretIRQ);
1273	break;
1274	case VINF_EM_RESCHEDULE_REM:
1275	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetRescheduleREM);
1276	break;
1277	case VINF_EM_RAW_TO_R3:
1278	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Total);
1279	if (VM_FF_IS_SET(pVM, VM_FF_TM_VIRTUAL_SYNC))
1280	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3TMVirt);
1281	else if (VM_FF_IS_SET(pVM, VM_FF_PGM_NEED_HANDY_PAGES))
1282	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3HandyPages);
1283	else if (VM_FF_IS_SET(pVM, VM_FF_PDM_QUEUES))
1284	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3PDMQueues);
1285	else if (VM_FF_IS_SET(pVM, VM_FF_EMT_RENDEZVOUS))
1286	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Rendezvous);
1287	else if (VM_FF_IS_SET(pVM, VM_FF_PDM_DMA))
1288	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3DMA);
1289	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TIMER))
1290	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Timer);
1291	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PDM_CRITSECT))
1292	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3CritSect);
1293	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_TO_R3))
1294	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3FF);
1295	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_IEM))
1296	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Iem);
1297	else if (VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_IOM))
1298	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Iom);
1299	else
1300	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetToR3Unknown);
1301	break;
1302
1303	case VINF_EM_RAW_TIMER_PENDING:
1304	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetTimerPending);
1305	break;
1306	case VINF_EM_RAW_INTERRUPT_PENDING:
1307	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetInterruptPending);
1308	break;
1309	case VINF_PATM_DUPLICATE_FUNCTION:
1310	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPATMDuplicateFn);
1311	break;
1312	case VINF_PGM_POOL_FLUSH_PENDING:
1313	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPGMFlushPending);
1314	break;
1315	case VINF_EM_PENDING_REQUEST:
1316	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPendingRequest);
1317	break;
1318	case VINF_EM_HM_PATCH_TPR_INSTR:
1319	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetPatchTPR);
1320	break;
1321	default:
1322	STAM_COUNTER_INC(&pVM->vmm.s.StatRZRetMisc);
1323	break;
1324	}
1325	}
1326	#endif /* VBOX_WITH_STATISTICS */
1327
1328
1329	/**
1330	* The Ring 0 entry point, called by the fast-ioctl path.
1331	*
1332	* @param pGVM The global (ring-0) VM structure.
1333	* @param pVMIgnored The cross context VM structure. The return code is
1334	* stored in pVM->vmm.s.iLastGZRc.
1335	* @param idCpu The Virtual CPU ID of the calling EMT.
1336	* @param enmOperation Which operation to execute.
1337	* @remarks Assume called with interrupts _enabled_.
1338	*/
1339	VMMR0DECL(void) VMMR0EntryFast(PGVM pGVM, PVMCC pVMIgnored, VMCPUID idCpu, VMMR0OPERATION enmOperation)
1340	{
1341	RT_NOREF(pVMIgnored);
1342
1343	/*
1344	* Validation.
1345	*/
1346	if ( idCpu < pGVM->cCpus
1347	&& pGVM->cCpus == pGVM->cCpusUnsafe)
1348	{ /likely/ }
1349	else
1350	{
1351	SUPR0Printf("VMMR0EntryFast: Bad idCpu=%#x cCpus=%#x cCpusUnsafe=%#x\n", idCpu, pGVM->cCpus, pGVM->cCpusUnsafe);
1352	return;
1353	}
1354
1355	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
1356	RTNATIVETHREAD const hNativeThread = RTThreadNativeSelf();
1357	if (RT_LIKELY( pGVCpu->hEMT == hNativeThread
1358	&& pGVCpu->hNativeThreadR0 == hNativeThread))
1359	{ /* likely */ }
1360	else
1361	{
1362	SUPR0Printf("VMMR0EntryFast: Bad thread idCpu=%#x hNativeSelf=%p pGVCpu->hEmt=%p pGVCpu->hNativeThreadR0=%p\n",
1363	idCpu, hNativeThread, pGVCpu->hEMT, pGVCpu->hNativeThreadR0);
1364	return;
1365	}
1366
1367	/*
1368	* Perform requested operation.
1369	*/
1370	switch (enmOperation)
1371	{
1372	/*
1373	* Run guest code using the available hardware acceleration technology.
1374	*/
1375	case VMMR0_DO_HM_RUN:
1376	{
1377	for (;;) /* hlt loop */
1378	{
1379	/*
1380	* Disable ring-3 calls & blocking till we've successfully entered HM.
1381	* Otherwise we sometimes end up blocking at the finall Log4 statement
1382	* in VMXR0Enter, while still in a somewhat inbetween state.
1383	*/
1384	VMMRZCallRing3Disable(pGVCpu);
1385
1386	/*
1387	* Disable preemption.
1388	*/
1389	Assert(!vmmR0ThreadCtxHookIsEnabled(pGVCpu));
1390	RTTHREADPREEMPTSTATE PreemptState = RTTHREADPREEMPTSTATE_INITIALIZER;
1391	RTThreadPreemptDisable(&PreemptState);
1392	pGVCpu->vmmr0.s.pPreemptState = &PreemptState;
1393
1394	/*
1395	* Get the host CPU identifiers, make sure they are valid and that
1396	* we've got a TSC delta for the CPU.
1397	*/
1398	RTCPUID idHostCpu;
1399	uint32_t iHostCpuSet = RTMpCurSetIndexAndId(&idHostCpu);
1400	if (RT_LIKELY( iHostCpuSet < RTCPUSET_MAX_CPUS
1401	&& SUPIsTscDeltaAvailableForCpuSetIndex(iHostCpuSet)))
1402	{
1403	pGVCpu->iHostCpuSet = iHostCpuSet;
1404	ASMAtomicWriteU32(&pGVCpu->idHostCpu, idHostCpu);
1405
1406	/*
1407	* Update the periodic preemption timer if it's active.
1408	*/
1409	if (pGVM->vmm.s.fUsePeriodicPreemptionTimers)
1410	GVMMR0SchedUpdatePeriodicPreemptionTimer(pGVM, pGVCpu->idHostCpu, TMCalcHostTimerFrequency(pGVM, pGVCpu));
1411
1412	#ifdef VMM_R0_TOUCH_FPU
1413	/*
1414	* Make sure we've got the FPU state loaded so and we don't need to clear
1415	* CR0.TS and get out of sync with the host kernel when loading the guest
1416	* FPU state. @ref sec_cpum_fpu (CPUM.cpp) and @bugref{4053}.
1417	*/
1418	CPUMR0TouchHostFpu();
1419	#endif
1420	int rc;
1421	bool fPreemptRestored = false;
1422	if (!HMR0SuspendPending())
1423	{
1424	/*
1425	* Enable the context switching hook.
1426	*/
1427	if (pGVCpu->vmmr0.s.hCtxHook != NIL_RTTHREADCTXHOOK)
1428	{
1429	Assert(!RTThreadCtxHookIsEnabled(pGVCpu->vmmr0.s.hCtxHook));
1430	int rc2 = RTThreadCtxHookEnable(pGVCpu->vmmr0.s.hCtxHook); AssertRC(rc2);
1431	}
1432
1433	/*
1434	* Enter HM context.
1435	*/
1436	rc = HMR0Enter(pGVCpu);
1437	if (RT_SUCCESS(rc))
1438	{
1439	VMCPU_SET_STATE(pGVCpu, VMCPUSTATE_STARTED_HM);
1440
1441	/*
1442	* When preemption hooks are in place, enable preemption now that
1443	* we're in HM context.
1444	*/
1445	if (vmmR0ThreadCtxHookIsEnabled(pGVCpu))
1446	{
1447	fPreemptRestored = true;
1448	pGVCpu->vmmr0.s.pPreemptState = NULL;
1449	RTThreadPreemptRestore(&PreemptState);
1450	}
1451	VMMRZCallRing3Enable(pGVCpu);
1452
1453	/*
1454	* Setup the longjmp machinery and execute guest code (calls HMR0RunGuestCode).
1455	*/
1456	rc = vmmR0CallRing3SetJmp(&pGVCpu->vmmr0.s.AssertJmpBuf, HMR0RunGuestCode, pGVM, pGVCpu);
1457
1458	/*
1459	* Assert sanity on the way out. Using manual assertions code here as normal
1460	* assertions are going to panic the host since we're outside the setjmp/longjmp zone.
1461	*/
1462	if (RT_UNLIKELY( VMCPU_GET_STATE(pGVCpu) != VMCPUSTATE_STARTED_HM
1463	&& RT_SUCCESS_NP(rc)
1464	&& rc != VERR_VMM_RING0_ASSERTION ))
1465	{
1466	pGVM->vmm.s.szRing0AssertMsg1[0] = '\0';
1467	RTStrPrintf(pGVM->vmm.s.szRing0AssertMsg2, sizeof(pGVM->vmm.s.szRing0AssertMsg2),
1468	"Got VMCPU state %d expected %d.\n", VMCPU_GET_STATE(pGVCpu), VMCPUSTATE_STARTED_HM);
1469	rc = VERR_VMM_WRONG_HM_VMCPU_STATE;
1470	}
1471	#if 0
1472	/** @todo Get rid of this. HM shouldn't disable the context hook. */
1473	else if (RT_UNLIKELY(vmmR0ThreadCtxHookIsEnabled(pGVCpu)))
1474	{
1475	pGVM->vmm.s.szRing0AssertMsg1[0] = '\0';
1476	RTStrPrintf(pGVM->vmm.s.szRing0AssertMsg2, sizeof(pGVM->vmm.s.szRing0AssertMsg2),
1477	"Thread-context hooks still enabled! VCPU=%p Id=%u rc=%d.\n", pGVCpu, pGVCpu->idCpu, rc);
1478	rc = VERR_VMM_CONTEXT_HOOK_STILL_ENABLED;
1479	}
1480	#endif
1481
1482	VMMRZCallRing3Disable(pGVCpu); /* Lazy bird: Simpler just disabling it again... */
1483	VMCPU_SET_STATE(pGVCpu, VMCPUSTATE_STARTED);
1484	}
1485	STAM_COUNTER_INC(&pGVM->vmm.s.StatRunGC);
1486
1487	/*
1488	* Invalidate the host CPU identifiers before we disable the context
1489	* hook / restore preemption.
1490	*/
1491	pGVCpu->iHostCpuSet = UINT32_MAX;
1492	ASMAtomicWriteU32(&pGVCpu->idHostCpu, NIL_RTCPUID);
1493
1494	/*
1495	* Disable context hooks. Due to unresolved cleanup issues, we
1496	* cannot leave the hooks enabled when we return to ring-3.
1497	*
1498	* Note! At the moment HM may also have disabled the hook
1499	* when we get here, but the IPRT API handles that.
1500	*/
1501	if (pGVCpu->vmmr0.s.hCtxHook != NIL_RTTHREADCTXHOOK)
1502	RTThreadCtxHookDisable(pGVCpu->vmmr0.s.hCtxHook);
1503	}
1504	/*
1505	* The system is about to go into suspend mode; go back to ring 3.
1506	*/
1507	else
1508	{
1509	pGVCpu->iHostCpuSet = UINT32_MAX;
1510	ASMAtomicWriteU32(&pGVCpu->idHostCpu, NIL_RTCPUID);
1511	rc = VINF_EM_RAW_INTERRUPT;
1512	}
1513
1514	/** @todo When HM stops messing with the context hook state, we'll disable
1515	* preemption again before the RTThreadCtxHookDisable call. */
1516	if (!fPreemptRestored)
1517	{
1518	pGVCpu->vmmr0.s.pPreemptState = NULL;
1519	RTThreadPreemptRestore(&PreemptState);
1520	}
1521
1522	pGVCpu->vmm.s.iLastGZRc = rc;
1523
1524	/* Fire dtrace probe and collect statistics. */
1525	VBOXVMM_R0_VMM_RETURN_TO_RING3_HM(pGVCpu, CPUMQueryGuestCtxPtr(pGVCpu), rc);
1526	#ifdef VBOX_WITH_STATISTICS
1527	vmmR0RecordRC(pGVM, pGVCpu, rc);
1528	#endif
1529	VMMRZCallRing3Enable(pGVCpu);
1530
1531	/*
1532	* If this is a halt.
1533	*/
1534	if (rc != VINF_EM_HALT)
1535	{ /* we're not in a hurry for a HLT, so prefer this path */ }
1536	else
1537	{
1538	pGVCpu->vmm.s.iLastGZRc = rc = vmmR0DoHalt(pGVM, pGVCpu);
1539	if (rc == VINF_SUCCESS)
1540	{
1541	pGVCpu->vmm.s.cR0HaltsSucceeded++;
1542	continue;
1543	}
1544	pGVCpu->vmm.s.cR0HaltsToRing3++;
1545	}
1546	}
1547	/*
1548	* Invalid CPU set index or TSC delta in need of measuring.
1549	*/
1550	else
1551	{
1552	pGVCpu->vmmr0.s.pPreemptState = NULL;
1553	pGVCpu->iHostCpuSet = UINT32_MAX;
1554	ASMAtomicWriteU32(&pGVCpu->idHostCpu, NIL_RTCPUID);
1555	RTThreadPreemptRestore(&PreemptState);
1556
1557	VMMRZCallRing3Enable(pGVCpu);
1558
1559	if (iHostCpuSet < RTCPUSET_MAX_CPUS)
1560	{
1561	int rc = SUPR0TscDeltaMeasureBySetIndex(pGVM->pSession, iHostCpuSet, 0 /fFlags/,
1562	2 /cMsWaitRetry/, 5RT_MS_1SEC /cMsWaitThread*/,
1563	0 /default cTries/);
1564	if (RT_SUCCESS(rc) \|\| rc == VERR_CPU_OFFLINE)
1565	pGVCpu->vmm.s.iLastGZRc = VINF_EM_RAW_TO_R3;
1566	else
1567	pGVCpu->vmm.s.iLastGZRc = rc;
1568	}
1569	else
1570	pGVCpu->vmm.s.iLastGZRc = VERR_INVALID_CPU_INDEX;
1571	}
1572	break;
1573	} /* halt loop. */
1574	break;
1575	}
1576
1577	#ifdef VBOX_WITH_NEM_R0
1578	# if defined(RT_ARCH_AMD64) && defined(RT_OS_WINDOWS)
1579	case VMMR0_DO_NEM_RUN:
1580	{
1581	/*
1582	* Setup the longjmp machinery and execute guest code (calls NEMR0RunGuestCode).
1583	*/
1584	# ifdef VBOXSTRICTRC_STRICT_ENABLED
1585	int rc = vmmR0CallRing3SetJmp2(&pGVCpu->vmmr0.s.AssertJmpBuf, (PFNVMMR0SETJMP2)NEMR0RunGuestCode, pGVM, idCpu);
1586	# else
1587	int rc = vmmR0CallRing3SetJmp2(&pGVCpu->vmmr0.s.AssertJmpBuf, NEMR0RunGuestCode, pGVM, idCpu);
1588	# endif
1589	STAM_COUNTER_INC(&pGVM->vmm.s.StatRunGC);
1590
1591	pGVCpu->vmm.s.iLastGZRc = rc;
1592
1593	/*
1594	* Fire dtrace probe and collect statistics.
1595	*/
1596	VBOXVMM_R0_VMM_RETURN_TO_RING3_NEM(pGVCpu, CPUMQueryGuestCtxPtr(pGVCpu), rc);
1597	# ifdef VBOX_WITH_STATISTICS
1598	vmmR0RecordRC(pGVM, pGVCpu, rc);
1599	# endif
1600	break;
1601	}
1602	# endif
1603	#endif
1604
1605	/*
1606	* For profiling.
1607	*/
1608	case VMMR0_DO_NOP:
1609	pGVCpu->vmm.s.iLastGZRc = VINF_SUCCESS;
1610	break;
1611
1612	/*
1613	* Shouldn't happen.
1614	*/
1615	default:
1616	AssertMsgFailed(("%#x\n", enmOperation));
1617	pGVCpu->vmm.s.iLastGZRc = VERR_NOT_SUPPORTED;
1618	break;
1619	}
1620	}
1621
1622
1623	/**
1624	* Validates a session or VM session argument.
1625	*
1626	* @returns true / false accordingly.
1627	* @param pGVM The global (ring-0) VM structure.
1628	* @param pClaimedSession The session claim to validate.
1629	* @param pSession The session argument.
1630	*/
1631	DECLINLINE(bool) vmmR0IsValidSession(PGVM pGVM, PSUPDRVSESSION pClaimedSession, PSUPDRVSESSION pSession)
1632	{
1633	/* This must be set! */
1634	if (!pSession)
1635	return false;
1636
1637	/* Only one out of the two. */
1638	if (pGVM && pClaimedSession)
1639	return false;
1640	if (pGVM)
1641	pClaimedSession = pGVM->pSession;
1642	return pClaimedSession == pSession;
1643	}
1644
1645
1646	/**
1647	* VMMR0EntryEx worker function, either called directly or when ever possible
1648	* called thru a longjmp so we can exit safely on failure.
1649	*
1650	* @returns VBox status code.
1651	* @param pGVM The global (ring-0) VM structure.
1652	* @param idCpu Virtual CPU ID argument. Must be NIL_VMCPUID if pVM
1653	* is NIL_RTR0PTR, and may be NIL_VMCPUID if it isn't
1654	* @param enmOperation Which operation to execute.
1655	* @param pReqHdr This points to a SUPVMMR0REQHDR packet. Optional.
1656	* The support driver validates this if it's present.
1657	* @param u64Arg Some simple constant argument.
1658	* @param pSession The session of the caller.
1659	*
1660	* @remarks Assume called with interrupts _enabled_.
1661	*/
1662	DECL_NO_INLINE(static, int) vmmR0EntryExWorker(PGVM pGVM, VMCPUID idCpu, VMMR0OPERATION enmOperation,
1663	PSUPVMMR0REQHDR pReqHdr, uint64_t u64Arg, PSUPDRVSESSION pSession)
1664	{
1665	/*
1666	* Validate pGVM and idCpu for consistency and validity.
1667	*/
1668	if (pGVM != NULL)
1669	{
1670	if (RT_LIKELY(((uintptr_t)pGVM & HOST_PAGE_OFFSET_MASK) == 0))
1671	{ /* likely */ }
1672	else
1673	{
1674	SUPR0Printf("vmmR0EntryExWorker: Invalid pGVM=%p! (op=%d)\n", pGVM, enmOperation);
1675	return VERR_INVALID_POINTER;
1676	}
1677
1678	if (RT_LIKELY(idCpu == NIL_VMCPUID \|\| idCpu < pGVM->cCpus))
1679	{ /* likely */ }
1680	else
1681	{
1682	SUPR0Printf("vmmR0EntryExWorker: Invalid idCpu %#x (cCpus=%#x)\n", idCpu, pGVM->cCpus);
1683	return VERR_INVALID_PARAMETER;
1684	}
1685
1686	if (RT_LIKELY( pGVM->enmVMState >= VMSTATE_CREATING
1687	&& pGVM->enmVMState <= VMSTATE_TERMINATED
1688	&& pGVM->pSession == pSession
1689	&& pGVM->pSelf == pGVM))
1690	{ /* likely */ }
1691	else
1692	{
1693	SUPR0Printf("vmmR0EntryExWorker: Invalid pGVM=%p:{.enmVMState=%d, .cCpus=%#x, .pSession=%p(==%p), .pSelf=%p(==%p)}! (op=%d)\n",
1694	pGVM, pGVM->enmVMState, pGVM->cCpus, pGVM->pSession, pSession, pGVM->pSelf, pGVM, enmOperation);
1695	return VERR_INVALID_POINTER;
1696	}
1697	}
1698	else if (RT_LIKELY(idCpu == NIL_VMCPUID))
1699	{ /* likely */ }
1700	else
1701	{
1702	SUPR0Printf("vmmR0EntryExWorker: Invalid idCpu=%u\n", idCpu);
1703	return VERR_INVALID_PARAMETER;
1704	}
1705
1706	/*
1707	* Process the request.
1708	*/
1709	int rc;
1710	switch (enmOperation)
1711	{
1712	/*
1713	* GVM requests
1714	*/
1715	case VMMR0_DO_GVMM_CREATE_VM:
1716	if (pGVM == NULL && u64Arg == 0 && idCpu == NIL_VMCPUID)
1717	rc = GVMMR0CreateVMReq((PGVMMCREATEVMREQ)pReqHdr, pSession);
1718	else
1719	rc = VERR_INVALID_PARAMETER;
1720	break;
1721
1722	case VMMR0_DO_GVMM_DESTROY_VM:
1723	if (pReqHdr == NULL && u64Arg == 0)
1724	rc = GVMMR0DestroyVM(pGVM);
1725	else
1726	rc = VERR_INVALID_PARAMETER;
1727	break;
1728
1729	case VMMR0_DO_GVMM_REGISTER_VMCPU:
1730	if (pGVM != NULL)
1731	rc = GVMMR0RegisterVCpu(pGVM, idCpu);
1732	else
1733	rc = VERR_INVALID_PARAMETER;
1734	break;
1735
1736	case VMMR0_DO_GVMM_DEREGISTER_VMCPU:
1737	if (pGVM != NULL)
1738	rc = GVMMR0DeregisterVCpu(pGVM, idCpu);
1739	else
1740	rc = VERR_INVALID_PARAMETER;
1741	break;
1742
1743	case VMMR0_DO_GVMM_REGISTER_WORKER_THREAD:
1744	if (pGVM != NULL && pReqHdr && pReqHdr->cbReq == sizeof(GVMMREGISTERWORKERTHREADREQ))
1745	rc = GVMMR0RegisterWorkerThread(pGVM, (GVMMWORKERTHREAD)(unsigned)u64Arg,
1746	((PGVMMREGISTERWORKERTHREADREQ)(pReqHdr))->hNativeThreadR3);
1747	else
1748	rc = VERR_INVALID_PARAMETER;
1749	break;
1750
1751	case VMMR0_DO_GVMM_DEREGISTER_WORKER_THREAD:
1752	if (pGVM != NULL)
1753	rc = GVMMR0DeregisterWorkerThread(pGVM, (GVMMWORKERTHREAD)(unsigned)u64Arg);
1754	else
1755	rc = VERR_INVALID_PARAMETER;
1756	break;
1757
1758	case VMMR0_DO_GVMM_SCHED_HALT:
1759	if (pReqHdr)
1760	return VERR_INVALID_PARAMETER;
1761	rc = GVMMR0SchedHaltReq(pGVM, idCpu, u64Arg);
1762	break;
1763
1764	case VMMR0_DO_GVMM_SCHED_WAKE_UP:
1765	if (pReqHdr \|\| u64Arg)
1766	return VERR_INVALID_PARAMETER;
1767	rc = GVMMR0SchedWakeUp(pGVM, idCpu);
1768	break;
1769
1770	case VMMR0_DO_GVMM_SCHED_POKE:
1771	if (pReqHdr \|\| u64Arg)
1772	return VERR_INVALID_PARAMETER;
1773	rc = GVMMR0SchedPoke(pGVM, idCpu);
1774	break;
1775
1776	case VMMR0_DO_GVMM_SCHED_WAKE_UP_AND_POKE_CPUS:
1777	if (u64Arg)
1778	return VERR_INVALID_PARAMETER;
1779	rc = GVMMR0SchedWakeUpAndPokeCpusReq(pGVM, (PGVMMSCHEDWAKEUPANDPOKECPUSREQ)pReqHdr);
1780	break;
1781
1782	case VMMR0_DO_GVMM_SCHED_POLL:
1783	if (pReqHdr \|\| u64Arg > 1)
1784	return VERR_INVALID_PARAMETER;
1785	rc = GVMMR0SchedPoll(pGVM, idCpu, !!u64Arg);
1786	break;
1787
1788	case VMMR0_DO_GVMM_QUERY_STATISTICS:
1789	if (u64Arg)
1790	return VERR_INVALID_PARAMETER;
1791	rc = GVMMR0QueryStatisticsReq(pGVM, (PGVMMQUERYSTATISTICSSREQ)pReqHdr, pSession);
1792	break;
1793
1794	case VMMR0_DO_GVMM_RESET_STATISTICS:
1795	if (u64Arg)
1796	return VERR_INVALID_PARAMETER;
1797	rc = GVMMR0ResetStatisticsReq(pGVM, (PGVMMRESETSTATISTICSSREQ)pReqHdr, pSession);
1798	break;
1799
1800	/*
1801	* Initialize the R0 part of a VM instance.
1802	*/
1803	case VMMR0_DO_VMMR0_INIT:
1804	rc = vmmR0InitVM(pGVM, RT_LODWORD(u64Arg), RT_HIDWORD(u64Arg));
1805	break;
1806
1807	/*
1808	* Does EMT specific ring-0 init.
1809	*/
1810	case VMMR0_DO_VMMR0_INIT_EMT:
1811	if (idCpu == NIL_VMCPUID)
1812	return VERR_INVALID_CPU_ID;
1813	rc = vmmR0InitVMEmt(pGVM, idCpu);
1814	break;
1815
1816	/*
1817	* Terminate the R0 part of a VM instance.
1818	*/
1819	case VMMR0_DO_VMMR0_TERM:
1820	rc = VMMR0TermVM(pGVM, 0 /idCpu/);
1821	break;
1822
1823	/*
1824	* Update release or debug logger instances.
1825	*/
1826	case VMMR0_DO_VMMR0_UPDATE_LOGGERS:
1827	if (idCpu == NIL_VMCPUID)
1828	return VERR_INVALID_CPU_ID;
1829	if (u64Arg < VMMLOGGER_IDX_MAX && pReqHdr != NULL)
1830	rc = vmmR0UpdateLoggers(pGVM, idCpu /idCpu/, (PVMMR0UPDATELOGGERSREQ)pReqHdr, (size_t)u64Arg);
1831	else
1832	return VERR_INVALID_PARAMETER;
1833	break;
1834
1835	/*
1836	* Log flusher thread.
1837	*/
1838	case VMMR0_DO_VMMR0_LOG_FLUSHER:
1839	if (idCpu != NIL_VMCPUID)
1840	return VERR_INVALID_CPU_ID;
1841	if (pReqHdr == NULL && pGVM != NULL)
1842	rc = vmmR0LogFlusher(pGVM);
1843	else
1844	return VERR_INVALID_PARAMETER;
1845	break;
1846
1847	/*
1848	* Wait for the flush to finish with all the buffers for the given logger.
1849	*/
1850	case VMMR0_DO_VMMR0_LOG_WAIT_FLUSHED:
1851	if (idCpu == NIL_VMCPUID)
1852	return VERR_INVALID_CPU_ID;
1853	if (u64Arg < VMMLOGGER_IDX_MAX && pReqHdr == NULL)
1854	rc = vmmR0LogWaitFlushed(pGVM, idCpu /idCpu/, (size_t)u64Arg);
1855	else
1856	return VERR_INVALID_PARAMETER;
1857	break;
1858
1859	/*
1860	* Attempt to enable hm mode and check the current setting.
1861	*/
1862	case VMMR0_DO_HM_ENABLE:
1863	rc = HMR0EnableAllCpus(pGVM);
1864	break;
1865
1866	/*
1867	* Setup the hardware accelerated session.
1868	*/
1869	case VMMR0_DO_HM_SETUP_VM:
1870	rc = HMR0SetupVM(pGVM);
1871	break;
1872
1873	/*
1874	* PGM wrappers.
1875	*/
1876	case VMMR0_DO_PGM_ALLOCATE_HANDY_PAGES:
1877	if (idCpu == NIL_VMCPUID)
1878	return VERR_INVALID_CPU_ID;
1879	rc = PGMR0PhysAllocateHandyPages(pGVM, idCpu);
1880	break;
1881
1882	case VMMR0_DO_PGM_FLUSH_HANDY_PAGES:
1883	if (idCpu == NIL_VMCPUID)
1884	return VERR_INVALID_CPU_ID;
1885	rc = PGMR0PhysFlushHandyPages(pGVM, idCpu);
1886	break;
1887
1888	case VMMR0_DO_PGM_ALLOCATE_LARGE_PAGE:
1889	if (idCpu == NIL_VMCPUID)
1890	return VERR_INVALID_CPU_ID;
1891	rc = PGMR0PhysAllocateLargePage(pGVM, idCpu, u64Arg);
1892	break;
1893
1894	case VMMR0_DO_PGM_PHYS_SETUP_IOMMU:
1895	if (idCpu != 0)
1896	return VERR_INVALID_CPU_ID;
1897	rc = PGMR0PhysSetupIoMmu(pGVM);
1898	break;
1899
1900	case VMMR0_DO_PGM_POOL_GROW:
1901	if (idCpu == NIL_VMCPUID)
1902	return VERR_INVALID_CPU_ID;
1903	rc = PGMR0PoolGrow(pGVM, idCpu);
1904	break;
1905
1906	/*
1907	* GMM wrappers.
1908	*/
1909	case VMMR0_DO_GMM_INITIAL_RESERVATION:
1910	if (u64Arg)
1911	return VERR_INVALID_PARAMETER;
1912	rc = GMMR0InitialReservationReq(pGVM, idCpu, (PGMMINITIALRESERVATIONREQ)pReqHdr);
1913	break;
1914
1915	case VMMR0_DO_GMM_UPDATE_RESERVATION:
1916	if (u64Arg)
1917	return VERR_INVALID_PARAMETER;
1918	rc = GMMR0UpdateReservationReq(pGVM, idCpu, (PGMMUPDATERESERVATIONREQ)pReqHdr);
1919	break;
1920
1921	case VMMR0_DO_GMM_ALLOCATE_PAGES:
1922	if (u64Arg)
1923	return VERR_INVALID_PARAMETER;
1924	rc = GMMR0AllocatePagesReq(pGVM, idCpu, (PGMMALLOCATEPAGESREQ)pReqHdr);
1925	break;
1926
1927	case VMMR0_DO_GMM_FREE_PAGES:
1928	if (u64Arg)
1929	return VERR_INVALID_PARAMETER;
1930	rc = GMMR0FreePagesReq(pGVM, idCpu, (PGMMFREEPAGESREQ)pReqHdr);
1931	break;
1932
1933	case VMMR0_DO_GMM_FREE_LARGE_PAGE:
1934	if (u64Arg)
1935	return VERR_INVALID_PARAMETER;
1936	rc = GMMR0FreeLargePageReq(pGVM, idCpu, (PGMMFREELARGEPAGEREQ)pReqHdr);
1937	break;
1938
1939	case VMMR0_DO_GMM_QUERY_HYPERVISOR_MEM_STATS:
1940	if (u64Arg)
1941	return VERR_INVALID_PARAMETER;
1942	rc = GMMR0QueryHypervisorMemoryStatsReq((PGMMMEMSTATSREQ)pReqHdr);
1943	break;
1944
1945	case VMMR0_DO_GMM_QUERY_MEM_STATS:
1946	if (idCpu == NIL_VMCPUID)
1947	return VERR_INVALID_CPU_ID;
1948	if (u64Arg)
1949	return VERR_INVALID_PARAMETER;
1950	rc = GMMR0QueryMemoryStatsReq(pGVM, idCpu, (PGMMMEMSTATSREQ)pReqHdr);
1951	break;
1952
1953	case VMMR0_DO_GMM_BALLOONED_PAGES:
1954	if (u64Arg)
1955	return VERR_INVALID_PARAMETER;
1956	rc = GMMR0BalloonedPagesReq(pGVM, idCpu, (PGMMBALLOONEDPAGESREQ)pReqHdr);
1957	break;
1958
1959	case VMMR0_DO_GMM_MAP_UNMAP_CHUNK:
1960	if (u64Arg)
1961	return VERR_INVALID_PARAMETER;
1962	rc = GMMR0MapUnmapChunkReq(pGVM, (PGMMMAPUNMAPCHUNKREQ)pReqHdr);
1963	break;
1964
1965	case VMMR0_DO_GMM_REGISTER_SHARED_MODULE:
1966	if (idCpu == NIL_VMCPUID)
1967	return VERR_INVALID_CPU_ID;
1968	if (u64Arg)
1969	return VERR_INVALID_PARAMETER;
1970	rc = GMMR0RegisterSharedModuleReq(pGVM, idCpu, (PGMMREGISTERSHAREDMODULEREQ)pReqHdr);
1971	break;
1972
1973	case VMMR0_DO_GMM_UNREGISTER_SHARED_MODULE:
1974	if (idCpu == NIL_VMCPUID)
1975	return VERR_INVALID_CPU_ID;
1976	if (u64Arg)
1977	return VERR_INVALID_PARAMETER;
1978	rc = GMMR0UnregisterSharedModuleReq(pGVM, idCpu, (PGMMUNREGISTERSHAREDMODULEREQ)pReqHdr);
1979	break;
1980
1981	case VMMR0_DO_GMM_RESET_SHARED_MODULES:
1982	if (idCpu == NIL_VMCPUID)
1983	return VERR_INVALID_CPU_ID;
1984	if ( u64Arg
1985	\|\| pReqHdr)
1986	return VERR_INVALID_PARAMETER;
1987	rc = GMMR0ResetSharedModules(pGVM, idCpu);
1988	break;
1989
1990	#ifdef VBOX_WITH_PAGE_SHARING
1991	case VMMR0_DO_GMM_CHECK_SHARED_MODULES:
1992	{
1993	if (idCpu == NIL_VMCPUID)
1994	return VERR_INVALID_CPU_ID;
1995	if ( u64Arg
1996	\|\| pReqHdr)
1997	return VERR_INVALID_PARAMETER;
1998	rc = GMMR0CheckSharedModules(pGVM, idCpu);
1999	break;
2000	}
2001	#endif
2002
2003	#if defined(VBOX_STRICT) && HC_ARCH_BITS == 64
2004	case VMMR0_DO_GMM_FIND_DUPLICATE_PAGE:
2005	if (u64Arg)
2006	return VERR_INVALID_PARAMETER;
2007	rc = GMMR0FindDuplicatePageReq(pGVM, (PGMMFINDDUPLICATEPAGEREQ)pReqHdr);
2008	break;
2009	#endif
2010
2011	case VMMR0_DO_GMM_QUERY_STATISTICS:
2012	if (u64Arg)
2013	return VERR_INVALID_PARAMETER;
2014	rc = GMMR0QueryStatisticsReq(pGVM, (PGMMQUERYSTATISTICSSREQ)pReqHdr);
2015	break;
2016
2017	case VMMR0_DO_GMM_RESET_STATISTICS:
2018	if (u64Arg)
2019	return VERR_INVALID_PARAMETER;
2020	rc = GMMR0ResetStatisticsReq(pGVM, (PGMMRESETSTATISTICSSREQ)pReqHdr);
2021	break;
2022
2023	/*
2024	* A quick GCFGM mock-up.
2025	*/
2026	/** @todo GCFGM with proper access control, ring-3 management interface and all that. */
2027	case VMMR0_DO_GCFGM_SET_VALUE:
2028	case VMMR0_DO_GCFGM_QUERY_VALUE:
2029	{
2030	if (pGVM \|\| !pReqHdr \|\| u64Arg \|\| idCpu != NIL_VMCPUID)
2031	return VERR_INVALID_PARAMETER;
2032	PGCFGMVALUEREQ pReq = (PGCFGMVALUEREQ)pReqHdr;
2033	if (pReq->Hdr.cbReq != sizeof(*pReq))
2034	return VERR_INVALID_PARAMETER;
2035	if (enmOperation == VMMR0_DO_GCFGM_SET_VALUE)
2036	{
2037	rc = GVMMR0SetConfig(pReq->pSession, &pReq->szName[0], pReq->u64Value);
2038	//if (rc == VERR_CFGM_VALUE_NOT_FOUND)
2039	// rc = GMMR0SetConfig(pReq->pSession, &pReq->szName[0], pReq->u64Value);
2040	}
2041	else
2042	{
2043	rc = GVMMR0QueryConfig(pReq->pSession, &pReq->szName[0], &pReq->u64Value);
2044	//if (rc == VERR_CFGM_VALUE_NOT_FOUND)
2045	// rc = GMMR0QueryConfig(pReq->pSession, &pReq->szName[0], &pReq->u64Value);
2046	}
2047	break;
2048	}
2049
2050	/*
2051	* PDM Wrappers.
2052	*/
2053	case VMMR0_DO_PDM_DRIVER_CALL_REQ_HANDLER:
2054	{
2055	if (!pReqHdr \|\| u64Arg \|\| idCpu != NIL_VMCPUID)
2056	return VERR_INVALID_PARAMETER;
2057	rc = PDMR0DriverCallReqHandler(pGVM, (PPDMDRIVERCALLREQHANDLERREQ)pReqHdr);
2058	break;
2059	}
2060
2061	case VMMR0_DO_PDM_DEVICE_CREATE:
2062	{
2063	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2064	return VERR_INVALID_PARAMETER;
2065	rc = PDMR0DeviceCreateReqHandler(pGVM, (PPDMDEVICECREATEREQ)pReqHdr);
2066	break;
2067	}
2068
2069	case VMMR0_DO_PDM_DEVICE_GEN_CALL:
2070	{
2071	if (!pReqHdr \|\| u64Arg)
2072	return VERR_INVALID_PARAMETER;
2073	rc = PDMR0DeviceGenCallReqHandler(pGVM, (PPDMDEVICEGENCALLREQ)pReqHdr, idCpu);
2074	break;
2075	}
2076
2077	/** @todo Remove the once all devices has been converted to new style! @bugref{9218} */
2078	case VMMR0_DO_PDM_DEVICE_COMPAT_SET_CRITSECT:
2079	{
2080	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2081	return VERR_INVALID_PARAMETER;
2082	rc = PDMR0DeviceCompatSetCritSectReqHandler(pGVM, (PPDMDEVICECOMPATSETCRITSECTREQ)pReqHdr);
2083	break;
2084	}
2085
2086	case VMMR0_DO_PDM_QUEUE_CREATE:
2087	{
2088	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2089	return VERR_INVALID_PARAMETER;
2090	rc = PDMR0QueueCreateReqHandler(pGVM, (PPDMQUEUECREATEREQ)pReqHdr);
2091	break;
2092	}
2093
2094	/*
2095	* Requests to the internal networking service.
2096	*/
2097	case VMMR0_DO_INTNET_OPEN:
2098	{
2099	PINTNETOPENREQ pReq = (PINTNETOPENREQ)pReqHdr;
2100	if (u64Arg \|\| !pReq \|\| !vmmR0IsValidSession(pGVM, pReq->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2101	return VERR_INVALID_PARAMETER;
2102	rc = IntNetR0OpenReq(pSession, pReq);
2103	break;
2104	}
2105
2106	case VMMR0_DO_INTNET_IF_CLOSE:
2107	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFCLOSEREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2108	return VERR_INVALID_PARAMETER;
2109	rc = IntNetR0IfCloseReq(pSession, (PINTNETIFCLOSEREQ)pReqHdr);
2110	break;
2111
2112
2113	case VMMR0_DO_INTNET_IF_GET_BUFFER_PTRS:
2114	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFGETBUFFERPTRSREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2115	return VERR_INVALID_PARAMETER;
2116	rc = IntNetR0IfGetBufferPtrsReq(pSession, (PINTNETIFGETBUFFERPTRSREQ)pReqHdr);
2117	break;
2118
2119	case VMMR0_DO_INTNET_IF_SET_PROMISCUOUS_MODE:
2120	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSETPROMISCUOUSMODEREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2121	return VERR_INVALID_PARAMETER;
2122	rc = IntNetR0IfSetPromiscuousModeReq(pSession, (PINTNETIFSETPROMISCUOUSMODEREQ)pReqHdr);
2123	break;
2124
2125	case VMMR0_DO_INTNET_IF_SET_MAC_ADDRESS:
2126	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSETMACADDRESSREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2127	return VERR_INVALID_PARAMETER;
2128	rc = IntNetR0IfSetMacAddressReq(pSession, (PINTNETIFSETMACADDRESSREQ)pReqHdr);
2129	break;
2130
2131	case VMMR0_DO_INTNET_IF_SET_ACTIVE:
2132	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSETACTIVEREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2133	return VERR_INVALID_PARAMETER;
2134	rc = IntNetR0IfSetActiveReq(pSession, (PINTNETIFSETACTIVEREQ)pReqHdr);
2135	break;
2136
2137	case VMMR0_DO_INTNET_IF_SEND:
2138	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFSENDREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2139	return VERR_INVALID_PARAMETER;
2140	rc = IntNetR0IfSendReq(pSession, (PINTNETIFSENDREQ)pReqHdr);
2141	break;
2142
2143	case VMMR0_DO_INTNET_IF_WAIT:
2144	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFWAITREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2145	return VERR_INVALID_PARAMETER;
2146	rc = IntNetR0IfWaitReq(pSession, (PINTNETIFWAITREQ)pReqHdr);
2147	break;
2148
2149	case VMMR0_DO_INTNET_IF_ABORT_WAIT:
2150	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PINTNETIFWAITREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2151	return VERR_INVALID_PARAMETER;
2152	rc = IntNetR0IfAbortWaitReq(pSession, (PINTNETIFABORTWAITREQ)pReqHdr);
2153	break;
2154
2155	#if 0 //def VBOX_WITH_PCI_PASSTHROUGH
2156	/*
2157	* Requests to host PCI driver service.
2158	*/
2159	case VMMR0_DO_PCIRAW_REQ:
2160	if (u64Arg \|\| !pReqHdr \|\| !vmmR0IsValidSession(pGVM, ((PPCIRAWSENDREQ)pReqHdr)->pSession, pSession) \|\| idCpu != NIL_VMCPUID)
2161	return VERR_INVALID_PARAMETER;
2162	rc = PciRawR0ProcessReq(pGVM, pSession, (PPCIRAWSENDREQ)pReqHdr);
2163	break;
2164	#endif
2165
2166	/*
2167	* NEM requests.
2168	*/
2169	#ifdef VBOX_WITH_NEM_R0
2170	# if defined(RT_ARCH_AMD64) && defined(RT_OS_WINDOWS)
2171	case VMMR0_DO_NEM_INIT_VM:
2172	if (u64Arg \|\| pReqHdr \|\| idCpu != 0)
2173	return VERR_INVALID_PARAMETER;
2174	rc = NEMR0InitVM(pGVM);
2175	break;
2176
2177	case VMMR0_DO_NEM_INIT_VM_PART_2:
2178	if (u64Arg \|\| pReqHdr \|\| idCpu != 0)
2179	return VERR_INVALID_PARAMETER;
2180	rc = NEMR0InitVMPart2(pGVM);
2181	break;
2182
2183	case VMMR0_DO_NEM_MAP_PAGES:
2184	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2185	return VERR_INVALID_PARAMETER;
2186	rc = NEMR0MapPages(pGVM, idCpu);
2187	break;
2188
2189	case VMMR0_DO_NEM_UNMAP_PAGES:
2190	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2191	return VERR_INVALID_PARAMETER;
2192	rc = NEMR0UnmapPages(pGVM, idCpu);
2193	break;
2194
2195	case VMMR0_DO_NEM_EXPORT_STATE:
2196	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2197	return VERR_INVALID_PARAMETER;
2198	rc = NEMR0ExportState(pGVM, idCpu);
2199	break;
2200
2201	case VMMR0_DO_NEM_IMPORT_STATE:
2202	if (pReqHdr \|\| idCpu == NIL_VMCPUID)
2203	return VERR_INVALID_PARAMETER;
2204	rc = NEMR0ImportState(pGVM, idCpu, u64Arg);
2205	break;
2206
2207	case VMMR0_DO_NEM_QUERY_CPU_TICK:
2208	if (u64Arg \|\| pReqHdr \|\| idCpu == NIL_VMCPUID)
2209	return VERR_INVALID_PARAMETER;
2210	rc = NEMR0QueryCpuTick(pGVM, idCpu);
2211	break;
2212
2213	case VMMR0_DO_NEM_RESUME_CPU_TICK_ON_ALL:
2214	if (pReqHdr \|\| idCpu == NIL_VMCPUID)
2215	return VERR_INVALID_PARAMETER;
2216	rc = NEMR0ResumeCpuTickOnAll(pGVM, idCpu, u64Arg);
2217	break;
2218
2219	case VMMR0_DO_NEM_UPDATE_STATISTICS:
2220	if (u64Arg \|\| pReqHdr)
2221	return VERR_INVALID_PARAMETER;
2222	rc = NEMR0UpdateStatistics(pGVM, idCpu);
2223	break;
2224
2225	# if 1 && defined(DEBUG_bird)
2226	case VMMR0_DO_NEM_EXPERIMENT:
2227	if (pReqHdr)
2228	return VERR_INVALID_PARAMETER;
2229	rc = NEMR0DoExperiment(pGVM, idCpu, u64Arg);
2230	break;
2231	# endif
2232	# endif
2233	#endif
2234
2235	/*
2236	* IOM requests.
2237	*/
2238	case VMMR0_DO_IOM_GROW_IO_PORTS:
2239	{
2240	if (pReqHdr \|\| idCpu != 0)
2241	return VERR_INVALID_PARAMETER;
2242	rc = IOMR0IoPortGrowRegistrationTables(pGVM, u64Arg);
2243	break;
2244	}
2245
2246	case VMMR0_DO_IOM_GROW_IO_PORT_STATS:
2247	{
2248	if (pReqHdr \|\| idCpu != 0)
2249	return VERR_INVALID_PARAMETER;
2250	rc = IOMR0IoPortGrowStatisticsTable(pGVM, u64Arg);
2251	break;
2252	}
2253
2254	case VMMR0_DO_IOM_GROW_MMIO_REGS:
2255	{
2256	if (pReqHdr \|\| idCpu != 0)
2257	return VERR_INVALID_PARAMETER;
2258	rc = IOMR0MmioGrowRegistrationTables(pGVM, u64Arg);
2259	break;
2260	}
2261
2262	case VMMR0_DO_IOM_GROW_MMIO_STATS:
2263	{
2264	if (pReqHdr \|\| idCpu != 0)
2265	return VERR_INVALID_PARAMETER;
2266	rc = IOMR0MmioGrowStatisticsTable(pGVM, u64Arg);
2267	break;
2268	}
2269
2270	case VMMR0_DO_IOM_SYNC_STATS_INDICES:
2271	{
2272	if (pReqHdr \|\| idCpu != 0)
2273	return VERR_INVALID_PARAMETER;
2274	rc = IOMR0IoPortSyncStatisticsIndices(pGVM);
2275	if (RT_SUCCESS(rc))
2276	rc = IOMR0MmioSyncStatisticsIndices(pGVM);
2277	break;
2278	}
2279
2280	/*
2281	* DBGF requests.
2282	*/
2283	#ifdef VBOX_WITH_DBGF_TRACING
2284	case VMMR0_DO_DBGF_TRACER_CREATE:
2285	{
2286	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2287	return VERR_INVALID_PARAMETER;
2288	rc = DBGFR0TracerCreateReqHandler(pGVM, (PDBGFTRACERCREATEREQ)pReqHdr);
2289	break;
2290	}
2291
2292	case VMMR0_DO_DBGF_TRACER_CALL_REQ_HANDLER:
2293	{
2294	if (!pReqHdr \|\| u64Arg)
2295	return VERR_INVALID_PARAMETER;
2296	# if 0 /** @todo */
2297	rc = DBGFR0TracerGenCallReqHandler(pGVM, (PDBGFTRACERGENCALLREQ)pReqHdr, idCpu);
2298	# else
2299	rc = VERR_NOT_IMPLEMENTED;
2300	# endif
2301	break;
2302	}
2303	#endif
2304
2305	case VMMR0_DO_DBGF_BP_INIT:
2306	{
2307	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2308	return VERR_INVALID_PARAMETER;
2309	rc = DBGFR0BpInitReqHandler(pGVM, (PDBGFBPINITREQ)pReqHdr);
2310	break;
2311	}
2312
2313	case VMMR0_DO_DBGF_BP_CHUNK_ALLOC:
2314	{
2315	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2316	return VERR_INVALID_PARAMETER;
2317	rc = DBGFR0BpChunkAllocReqHandler(pGVM, (PDBGFBPCHUNKALLOCREQ)pReqHdr);
2318	break;
2319	}
2320
2321	case VMMR0_DO_DBGF_BP_L2_TBL_CHUNK_ALLOC:
2322	{
2323	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2324	return VERR_INVALID_PARAMETER;
2325	rc = DBGFR0BpL2TblChunkAllocReqHandler(pGVM, (PDBGFBPL2TBLCHUNKALLOCREQ)pReqHdr);
2326	break;
2327	}
2328
2329	case VMMR0_DO_DBGF_BP_OWNER_INIT:
2330	{
2331	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2332	return VERR_INVALID_PARAMETER;
2333	rc = DBGFR0BpOwnerInitReqHandler(pGVM, (PDBGFBPOWNERINITREQ)pReqHdr);
2334	break;
2335	}
2336
2337	case VMMR0_DO_DBGF_BP_PORTIO_INIT:
2338	{
2339	if (!pReqHdr \|\| u64Arg \|\| idCpu != 0)
2340	return VERR_INVALID_PARAMETER;
2341	rc = DBGFR0BpPortIoInitReqHandler(pGVM, (PDBGFBPINITREQ)pReqHdr);
2342	break;
2343	}
2344
2345
2346	/*
2347	* TM requests.
2348	*/
2349	case VMMR0_DO_TM_GROW_TIMER_QUEUE:
2350	{
2351	if (pReqHdr \|\| idCpu == NIL_VMCPUID)
2352	return VERR_INVALID_PARAMETER;
2353	rc = TMR0TimerQueueGrow(pGVM, RT_HI_U32(u64Arg), RT_LO_U32(u64Arg));
2354	break;
2355	}
2356
2357	/*
2358	* For profiling.
2359	*/
2360	case VMMR0_DO_NOP:
2361	case VMMR0_DO_SLOW_NOP:
2362	return VINF_SUCCESS;
2363
2364	/*
2365	* For testing Ring-0 APIs invoked in this environment.
2366	*/
2367	case VMMR0_DO_TESTS:
2368	/** @todo make new test */
2369	return VINF_SUCCESS;
2370
2371	default:
2372	/*
2373	* We're returning VERR_NOT_SUPPORT here so we've got something else
2374	* than -1 which the interrupt gate glue code might return.
2375	*/
2376	Log(("operation %#x is not supported\n", enmOperation));
2377	return VERR_NOT_SUPPORTED;
2378	}
2379	return rc;
2380	}
2381
2382
2383	/**
2384	* This is just a longjmp wrapper function for VMMR0EntryEx calls.
2385	*
2386	* @returns VBox status code.
2387	* @param pvArgs The argument package
2388	*/
2389	static DECLCALLBACK(int) vmmR0EntryExWrapper(void *pvArgs)
2390	{
2391	PGVMCPU pGVCpu = (PGVMCPU)pvArgs;
2392	return vmmR0EntryExWorker(pGVCpu->vmmr0.s.pGVM,
2393	pGVCpu->vmmr0.s.idCpu,
2394	pGVCpu->vmmr0.s.enmOperation,
2395	pGVCpu->vmmr0.s.pReq,
2396	pGVCpu->vmmr0.s.u64Arg,
2397	pGVCpu->vmmr0.s.pSession);
2398	}
2399
2400
2401	/**
2402	* The Ring 0 entry point, called by the support library (SUP).
2403	*
2404	* @returns VBox status code.
2405	* @param pGVM The global (ring-0) VM structure.
2406	* @param pVM The cross context VM structure.
2407	* @param idCpu Virtual CPU ID argument. Must be NIL_VMCPUID if pVM
2408	* is NIL_RTR0PTR, and may be NIL_VMCPUID if it isn't
2409	* @param enmOperation Which operation to execute.
2410	* @param pReq Pointer to the SUPVMMR0REQHDR packet. Optional.
2411	* @param u64Arg Some simple constant argument.
2412	* @param pSession The session of the caller.
2413	* @remarks Assume called with interrupts _enabled_.
2414	*/
2415	VMMR0DECL(int) VMMR0EntryEx(PGVM pGVM, PVMCC pVM, VMCPUID idCpu, VMMR0OPERATION enmOperation,
2416	PSUPVMMR0REQHDR pReq, uint64_t u64Arg, PSUPDRVSESSION pSession)
2417	{
2418	/*
2419	* Requests that should only happen on the EMT thread will be
2420	* wrapped in a setjmp so we can assert without causing too much trouble.
2421	*/
2422	if ( pVM != NULL
2423	&& pGVM != NULL
2424	&& pVM == pGVM /** @todo drop pVM or pGVM */
2425	&& idCpu < pGVM->cCpus
2426	&& pGVM->pSession == pSession
2427	&& pGVM->pSelf == pGVM
2428	&& enmOperation != VMMR0_DO_GVMM_DESTROY_VM
2429	&& enmOperation != VMMR0_DO_GVMM_REGISTER_VMCPU
2430	&& enmOperation != VMMR0_DO_GVMM_SCHED_WAKE_UP /* idCpu is not caller but target. Sigh. / /* @todo fix*/
2431	&& enmOperation != VMMR0_DO_GVMM_SCHED_POKE /* idCpu is not caller but target. Sigh. / /* @todo fix*/
2432	)
2433	{
2434	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2435	RTNATIVETHREAD hNativeThread = RTThreadNativeSelf();
2436	if (RT_LIKELY( pGVCpu->hEMT == hNativeThread
2437	&& pGVCpu->hNativeThreadR0 == hNativeThread))
2438	{
2439	pGVCpu->vmmr0.s.pGVM = pGVM;
2440	pGVCpu->vmmr0.s.idCpu = idCpu;
2441	pGVCpu->vmmr0.s.enmOperation = enmOperation;
2442	pGVCpu->vmmr0.s.pReq = pReq;
2443	pGVCpu->vmmr0.s.u64Arg = u64Arg;
2444	pGVCpu->vmmr0.s.pSession = pSession;
2445	return vmmR0CallRing3SetJmpEx(&pGVCpu->vmmr0.s.AssertJmpBuf, vmmR0EntryExWrapper, pGVCpu,
2446	((uintptr_t)u64Arg << 16) \| (uintptr_t)enmOperation);
2447	}
2448	return VERR_VM_THREAD_NOT_EMT;
2449	}
2450	return vmmR0EntryExWorker(pGVM, idCpu, enmOperation, pReq, u64Arg, pSession);
2451	}
2452
2453
2454	/*********************************************************************************************************************************
2455	* EMT Blocking *
2456	*********************************************************************************************************************************/
2457
2458	/**
2459	* Checks whether we've armed the ring-0 long jump machinery.
2460	*
2461	* @returns @c true / @c false
2462	* @param pVCpu The cross context virtual CPU structure.
2463	* @thread EMT
2464	* @sa VMMIsLongJumpArmed
2465	*/
2466	VMMR0_INT_DECL(bool) VMMR0IsLongJumpArmed(PVMCPUCC pVCpu)
2467	{
2468	#ifdef RT_ARCH_X86
2469	return pVCpu->vmmr0.s.AssertJmpBuf.eip != 0;
2470	#else
2471	return pVCpu->vmmr0.s.AssertJmpBuf.rip != 0;
2472	#endif
2473	}
2474
2475
2476	/**
2477	* Locking helper that deals with HM context and checks if the thread can block.
2478	*
2479	* @returns VINF_SUCCESS if we can block. Returns @a rcBusy or
2480	* VERR_VMM_CANNOT_BLOCK if not able to block.
2481	* @param pVCpu The cross context virtual CPU structure of the calling
2482	* thread.
2483	* @param rcBusy What to return in case of a blocking problem. Will IPE
2484	* if VINF_SUCCESS and we cannot block.
2485	* @param pszCaller The caller (for logging problems).
2486	* @param pvLock The lock address (for logging problems).
2487	* @param pCtx Where to return context info for the resume call.
2488	* @thread EMT(pVCpu)
2489	*/
2490	VMMR0_INT_DECL(int) VMMR0EmtPrepareToBlock(PVMCPUCC pVCpu, int rcBusy, const char pszCaller, void pvLock,
2491	PVMMR0EMTBLOCKCTX pCtx)
2492	{
2493	const char *pszMsg;
2494
2495	/*
2496	* Check that we are allowed to block.
2497	*/
2498	if (RT_LIKELY(VMMRZCallRing3IsEnabled(pVCpu)))
2499	{
2500	/*
2501	* Are we in HM context and w/o a context hook? If so work the context hook.
2502	*/
2503	if (pVCpu->idHostCpu != NIL_RTCPUID)
2504	{
2505	Assert(pVCpu->iHostCpuSet != UINT32_MAX);
2506
2507	if (pVCpu->vmmr0.s.hCtxHook == NIL_RTTHREADCTXHOOK)
2508	{
2509	vmmR0ThreadCtxCallback(RTTHREADCTXEVENT_OUT, pVCpu);
2510	if (pVCpu->vmmr0.s.pPreemptState)
2511	RTThreadPreemptRestore(pVCpu->vmmr0.s.pPreemptState);
2512
2513	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC;
2514	pCtx->fWasInHmContext = true;
2515	return VINF_SUCCESS;
2516	}
2517	}
2518
2519	if (RT_LIKELY(!pVCpu->vmmr0.s.pPreemptState))
2520	{
2521	/*
2522	* Not in HM context or we've got hooks, so just check that preemption
2523	* is enabled.
2524	*/
2525	if (RT_LIKELY(RTThreadPreemptIsEnabled(NIL_RTTHREAD)))
2526	{
2527	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC;
2528	pCtx->fWasInHmContext = false;
2529	return VINF_SUCCESS;
2530	}
2531	pszMsg = "Preemption is disabled!";
2532	}
2533	else
2534	pszMsg = "Preemption state w/o HM state!";
2535	}
2536	else
2537	pszMsg = "Ring-3 calls are disabled!";
2538
2539	static uint32_t volatile s_cWarnings = 0;
2540	if (++s_cWarnings < 50)
2541	SUPR0Printf("VMMR0EmtPrepareToBlock: %s pvLock=%p pszCaller=%s rcBusy=%p\n", pszMsg, pvLock, pszCaller, rcBusy);
2542	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC_DEAD;
2543	pCtx->fWasInHmContext = false;
2544	return rcBusy != VINF_SUCCESS ? rcBusy : VERR_VMM_CANNOT_BLOCK;
2545	}
2546
2547
2548	/**
2549	* Counterpart to VMMR0EmtPrepareToBlock.
2550	*
2551	* @param pVCpu The cross context virtual CPU structure of the calling
2552	* thread.
2553	* @param pCtx The context structure used with VMMR0EmtPrepareToBlock.
2554	* @thread EMT(pVCpu)
2555	*/
2556	VMMR0_INT_DECL(void) VMMR0EmtResumeAfterBlocking(PVMCPUCC pVCpu, PVMMR0EMTBLOCKCTX pCtx)
2557	{
2558	AssertReturnVoid(pCtx->uMagic == VMMR0EMTBLOCKCTX_MAGIC);
2559	if (pCtx->fWasInHmContext)
2560	{
2561	if (pVCpu->vmmr0.s.pPreemptState)
2562	RTThreadPreemptDisable(pVCpu->vmmr0.s.pPreemptState);
2563
2564	pCtx->fWasInHmContext = false;
2565	vmmR0ThreadCtxCallback(RTTHREADCTXEVENT_IN, pVCpu);
2566	}
2567	pCtx->uMagic = VMMR0EMTBLOCKCTX_MAGIC_DEAD;
2568	}
2569
2570
2571	/**
2572	* Helper for waiting on an RTSEMEVENT, caller did VMMR0EmtPrepareToBlock.
2573	*
2574	* @returns
2575	* @retval VERR_THREAD_IS_TERMINATING
2576	* @retval VERR_TIMEOUT if we ended up waiting too long, either according to
2577	* @a cMsTimeout or to maximum wait values.
2578	*
2579	* @param pGVCpu The ring-0 virtual CPU structure.
2580	* @param fFlags VMMR0EMTWAIT_F_XXX.
2581	* @param hEvent The event to wait on.
2582	* @param cMsTimeout The timeout or RT_INDEFINITE_WAIT.
2583	*/
2584	VMMR0_INT_DECL(int) VMMR0EmtWaitEventInner(PGVMCPU pGVCpu, uint32_t fFlags, RTSEMEVENT hEvent, RTMSINTERVAL cMsTimeout)
2585	{
2586	AssertReturn(pGVCpu->hEMT == RTThreadNativeSelf(), VERR_VM_THREAD_NOT_EMT);
2587
2588	/*
2589	* Note! Similar code is found in the PDM critical sections too.
2590	*/
2591	uint64_t const nsStart = RTTimeNanoTS();
2592	uint64_t cNsMaxTotal = cMsTimeout == RT_INDEFINITE_WAIT
2593	? RT_NS_5MIN : RT_MIN(RT_NS_5MIN, RT_NS_1MS_64 * cMsTimeout);
2594	uint32_t cMsMaxOne = RT_MS_5SEC;
2595	bool fNonInterruptible = false;
2596	for (;;)
2597	{
2598	/* Wait. */
2599	int rcWait = !fNonInterruptible
2600	? RTSemEventWaitNoResume(hEvent, cMsMaxOne)
2601	: RTSemEventWait(hEvent, cMsMaxOne);
2602	if (RT_SUCCESS(rcWait))
2603	return rcWait;
2604
2605	if (rcWait == VERR_TIMEOUT \|\| rcWait == VERR_INTERRUPTED)
2606	{
2607	uint64_t const cNsElapsed = RTTimeNanoTS() - nsStart;
2608
2609	/*
2610	* Check the thread termination status.
2611	*/
2612	int const rcTerm = RTThreadQueryTerminationStatus(NIL_RTTHREAD);
2613	AssertMsg(rcTerm == VINF_SUCCESS \|\| rcTerm == VERR_NOT_SUPPORTED \|\| rcTerm == VINF_THREAD_IS_TERMINATING,
2614	("rcTerm=%Rrc\n", rcTerm));
2615	if ( rcTerm == VERR_NOT_SUPPORTED
2616	&& !fNonInterruptible
2617	&& cNsMaxTotal > RT_NS_1MIN)
2618	cNsMaxTotal = RT_NS_1MIN;
2619
2620	/* We return immediately if it looks like the thread is terminating. */
2621	if (rcTerm == VINF_THREAD_IS_TERMINATING)
2622	return VERR_THREAD_IS_TERMINATING;
2623
2624	/* We may suppress VERR_INTERRUPTED if VMMR0EMTWAIT_F_TRY_SUPPRESS_INTERRUPTED was
2625	specified, otherwise we'll just return it. */
2626	if (rcWait == VERR_INTERRUPTED)
2627	{
2628	if (!(fFlags & VMMR0EMTWAIT_F_TRY_SUPPRESS_INTERRUPTED))
2629	return VERR_INTERRUPTED;
2630	if (!fNonInterruptible)
2631	{
2632	/* First time: Adjust down the wait parameters and make sure we get at least
2633	one non-interruptible wait before timing out. */
2634	fNonInterruptible = true;
2635	cMsMaxOne = 32;
2636	uint64_t const cNsLeft = cNsMaxTotal - cNsElapsed;
2637	if (cNsLeft > RT_NS_10SEC)
2638	cNsMaxTotal = cNsElapsed + RT_NS_10SEC;
2639	continue;
2640	}
2641	}
2642
2643	/* Check for timeout. */
2644	if (cNsElapsed > cNsMaxTotal)
2645	return VERR_TIMEOUT;
2646	}
2647	else
2648	return rcWait;
2649	}
2650	/* not reached */
2651	}
2652
2653
2654	/**
2655	* Helper for signalling an SUPSEMEVENT.
2656	*
2657	* This may temporarily leave the HM context if the host requires that for
2658	* signalling SUPSEMEVENT objects.
2659	*
2660	* @returns VBox status code (see VMMR0EmtPrepareToBlock)
2661	* @param pGVM The ring-0 VM structure.
2662	* @param pGVCpu The ring-0 virtual CPU structure.
2663	* @param hEvent The event to signal.
2664	*/
2665	VMMR0_INT_DECL(int) VMMR0EmtSignalSupEvent(PGVM pGVM, PGVMCPU pGVCpu, SUPSEMEVENT hEvent)
2666	{
2667	AssertReturn(pGVCpu->hEMT == RTThreadNativeSelf(), VERR_VM_THREAD_NOT_EMT);
2668	if (RTSemEventIsSignalSafe())
2669	return SUPSemEventSignal(pGVM->pSession, hEvent);
2670
2671	VMMR0EMTBLOCKCTX Ctx;
2672	int rc = VMMR0EmtPrepareToBlock(pGVCpu, VINF_SUCCESS, __FUNCTION__, (void *)(uintptr_t)hEvent, &Ctx);
2673	if (RT_SUCCESS(rc))
2674	{
2675	rc = SUPSemEventSignal(pGVM->pSession, hEvent);
2676	VMMR0EmtResumeAfterBlocking(pGVCpu, &Ctx);
2677	}
2678	return rc;
2679	}
2680
2681
2682	/**
2683	* Helper for signalling an SUPSEMEVENT, variant supporting non-EMTs.
2684	*
2685	* This may temporarily leave the HM context if the host requires that for
2686	* signalling SUPSEMEVENT objects.
2687	*
2688	* @returns VBox status code (see VMMR0EmtPrepareToBlock)
2689	* @param pGVM The ring-0 VM structure.
2690	* @param hEvent The event to signal.
2691	*/
2692	VMMR0_INT_DECL(int) VMMR0EmtSignalSupEventByGVM(PGVM pGVM, SUPSEMEVENT hEvent)
2693	{
2694	if (!RTSemEventIsSignalSafe())
2695	{
2696	PGVMCPU pGVCpu = GVMMR0GetGVCpuByGVMandEMT(pGVM, NIL_RTNATIVETHREAD);
2697	if (pGVCpu)
2698	{
2699	VMMR0EMTBLOCKCTX Ctx;
2700	int rc = VMMR0EmtPrepareToBlock(pGVCpu, VINF_SUCCESS, __FUNCTION__, (void *)(uintptr_t)hEvent, &Ctx);
2701	if (RT_SUCCESS(rc))
2702	{
2703	rc = SUPSemEventSignal(pGVM->pSession, hEvent);
2704	VMMR0EmtResumeAfterBlocking(pGVCpu, &Ctx);
2705	}
2706	return rc;
2707	}
2708	}
2709	return SUPSemEventSignal(pGVM->pSession, hEvent);
2710	}
2711
2712
2713	/*********************************************************************************************************************************
2714	* Logging. *
2715	*********************************************************************************************************************************/
2716
2717	/**
2718	* VMMR0_DO_VMMR0_UPDATE_LOGGERS: Updates the EMT loggers for the VM.
2719	*
2720	* @returns VBox status code.
2721	* @param pGVM The global (ring-0) VM structure.
2722	* @param idCpu The ID of the calling EMT.
2723	* @param pReq The request data.
2724	* @param idxLogger Which logger set to update.
2725	* @thread EMT(idCpu)
2726	*/
2727	static int vmmR0UpdateLoggers(PGVM pGVM, VMCPUID idCpu, PVMMR0UPDATELOGGERSREQ pReq, size_t idxLogger)
2728	{
2729	/*
2730	* Check sanity. First we require EMT to be calling us.
2731	*/
2732	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID);
2733	AssertReturn(pGVM->aCpus[idCpu].hEMT == RTThreadNativeSelf(), VERR_INVALID_CPU_ID);
2734
2735	AssertReturn(pReq->Hdr.cbReq >= RT_UOFFSETOF_DYN(VMMR0UPDATELOGGERSREQ, afGroups[0]), VERR_INVALID_PARAMETER);
2736	AssertReturn(pReq->cGroups < _8K, VERR_INVALID_PARAMETER);
2737	AssertReturn(pReq->Hdr.cbReq == RT_UOFFSETOF_DYN(VMMR0UPDATELOGGERSREQ, afGroups[pReq->cGroups]), VERR_INVALID_PARAMETER);
2738
2739	AssertReturn(idxLogger < VMMLOGGER_IDX_MAX, VERR_OUT_OF_RANGE);
2740
2741	/*
2742	* Adjust flags.
2743	*/
2744	/* Always buffered: */
2745	pReq->fFlags \|= RTLOGFLAGS_BUFFERED;
2746	/* These doesn't make sense at present: */
2747	pReq->fFlags &= ~(RTLOGFLAGS_FLUSH \| RTLOGFLAGS_WRITE_THROUGH);
2748	/* We've traditionally skipped the group restrictions. */
2749	pReq->fFlags &= ~RTLOGFLAGS_RESTRICT_GROUPS;
2750
2751	/*
2752	* Do the updating.
2753	*/
2754	int rc = VINF_SUCCESS;
2755	for (idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
2756	{
2757	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2758	PRTLOGGER pLogger = pGVCpu->vmmr0.s.u.aLoggers[idxLogger].pLogger;
2759	if (pLogger)
2760	{
2761	RTLogSetR0ProgramStart(pLogger, pGVM->vmm.s.nsProgramStart);
2762	rc = RTLogBulkUpdate(pLogger, pReq->fFlags, pReq->uGroupCrc32, pReq->cGroups, pReq->afGroups);
2763	}
2764	}
2765
2766	return rc;
2767	}
2768
2769
2770	/**
2771	* VMMR0_DO_VMMR0_LOG_FLUSHER: Get the next log flushing job.
2772	*
2773	* The job info is copied into VMM::LogFlusherItem.
2774	*
2775	* @returns VBox status code.
2776	* @retval VERR_OBJECT_DESTROYED if we're shutting down.
2777	* @retval VERR_NOT_OWNER if the calling thread is not the flusher thread.
2778	* @param pGVM The global (ring-0) VM structure.
2779	* @thread The log flusher thread (first caller automatically becomes the log
2780	* flusher).
2781	*/
2782	static int vmmR0LogFlusher(PGVM pGVM)
2783	{
2784	/*
2785	* Check that this really is the flusher thread.
2786	*/
2787	RTNATIVETHREAD const hNativeSelf = RTThreadNativeSelf();
2788	AssertReturn(hNativeSelf != NIL_RTNATIVETHREAD, VERR_INTERNAL_ERROR_3);
2789	if (RT_LIKELY(pGVM->vmmr0.s.LogFlusher.hThread == hNativeSelf))
2790	{ /* likely */ }
2791	else
2792	{
2793	/* The first caller becomes the flusher thread. */
2794	bool fOk;
2795	ASMAtomicCmpXchgHandle(&pGVM->vmmr0.s.LogFlusher.hThread, hNativeSelf, NIL_RTNATIVETHREAD, fOk);
2796	if (!fOk)
2797	return VERR_NOT_OWNER;
2798	pGVM->vmmr0.s.LogFlusher.fThreadRunning = true;
2799	}
2800
2801	/*
2802	* Acknowledge flush, waking up waiting EMT.
2803	*/
2804	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2805
2806	uint32_t idxTail = pGVM->vmmr0.s.LogFlusher.idxRingTail % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2807	uint32_t idxHead = pGVM->vmmr0.s.LogFlusher.idxRingHead % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2808	if ( idxTail != idxHead
2809	&& pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.fProcessing)
2810	{
2811	/* Pop the head off the ring buffer. */
2812	uint32_t const idCpu = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.idCpu;
2813	uint32_t const idxLogger = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.idxLogger;
2814	uint32_t const idxBuffer = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.idxBuffer;
2815
2816	pGVM->vmmr0.s.LogFlusher.aRing[idxHead].u32 = UINT32_MAX >> 1; /* invalidate the entry */
2817	pGVM->vmmr0.s.LogFlusher.idxRingHead = (idxHead + 1) % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2818
2819	/* Validate content. */
2820	if ( idCpu < pGVM->cCpus
2821	&& idxLogger < VMMLOGGER_IDX_MAX
2822	&& idxBuffer < VMMLOGGER_BUFFER_COUNT)
2823	{
2824	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2825	PVMMR0PERVCPULOGGER pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
2826	PVMMR3CPULOGGER pShared = &pGVCpu->vmm.s.u.aLoggers[idxLogger];
2827
2828	/*
2829	* Accounting.
2830	*/
2831	uint32_t cFlushing = pR0Log->cFlushing - 1;
2832	if (RT_LIKELY(cFlushing < VMMLOGGER_BUFFER_COUNT))
2833	{ /likely/ }
2834	else
2835	cFlushing = 0;
2836	pR0Log->cFlushing = cFlushing;
2837	ASMAtomicWriteU32(&pShared->cFlushing, cFlushing);
2838
2839	/*
2840	* Wake up the EMT if it's waiting.
2841	*/
2842	if (!pR0Log->fEmtWaiting)
2843	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2844	else
2845	{
2846	pR0Log->fEmtWaiting = false;
2847	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2848
2849	int rc = RTSemEventSignal(pR0Log->hEventFlushWait);
2850	if (RT_FAILURE(rc))
2851	LogRelMax(64, ("vmmR0LogFlusher: RTSemEventSignal failed ACKing entry #%u (%u/%u/%u): %Rrc!\n",
2852	idxHead, idCpu, idxLogger, idxBuffer, rc));
2853	}
2854	}
2855	else
2856	{
2857	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2858	LogRelMax(64, ("vmmR0LogFlusher: Bad ACK entry #%u: %u/%u/%u!\n", idxHead, idCpu, idxLogger, idxBuffer));
2859	}
2860
2861	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2862	}
2863
2864	/*
2865	* The wait loop.
2866	*/
2867	int rc;
2868	for (;;)
2869	{
2870	/*
2871	* Work pending?
2872	*/
2873	idxTail = pGVM->vmmr0.s.LogFlusher.idxRingTail % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2874	idxHead = pGVM->vmmr0.s.LogFlusher.idxRingHead % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
2875	if (idxTail != idxHead)
2876	{
2877	pGVM->vmmr0.s.LogFlusher.aRing[idxHead].s.fProcessing = true;
2878	pGVM->vmm.s.LogFlusherItem.u32 = pGVM->vmmr0.s.LogFlusher.aRing[idxHead].u32;
2879
2880	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2881	return VINF_SUCCESS;
2882	}
2883
2884	/*
2885	* Nothing to do, so, check for termination and go to sleep.
2886	*/
2887	if (!pGVM->vmmr0.s.LogFlusher.fThreadShutdown)
2888	{ /* likely */ }
2889	else
2890	{
2891	rc = VERR_OBJECT_DESTROYED;
2892	break;
2893	}
2894
2895	pGVM->vmmr0.s.LogFlusher.fThreadWaiting = true;
2896	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2897
2898	rc = RTSemEventWaitNoResume(pGVM->vmmr0.s.LogFlusher.hEvent, RT_MS_5MIN);
2899
2900	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2901	pGVM->vmmr0.s.LogFlusher.fThreadWaiting = false;
2902
2903	if (RT_SUCCESS(rc) \|\| rc == VERR_TIMEOUT)
2904	{ /* likely */ }
2905	else if (rc == VERR_INTERRUPTED)
2906	{
2907	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2908	return rc;
2909	}
2910	else if (rc == VERR_SEM_DESTROYED \|\| rc == VERR_INVALID_HANDLE)
2911	break;
2912	else
2913	{
2914	LogRel(("vmmR0LogFlusher: RTSemEventWaitNoResume returned unexpected status %Rrc\n", rc));
2915	break;
2916	}
2917	}
2918
2919	/*
2920	* Terminating - prevent further calls and indicate to the EMTs that we're no longer around.
2921	*/
2922	pGVM->vmmr0.s.LogFlusher.hThread = ~pGVM->vmmr0.s.LogFlusher.hThread; /* (should be reasonably safe) */
2923	pGVM->vmmr0.s.LogFlusher.fThreadRunning = false;
2924
2925	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2926	return rc;
2927	}
2928
2929
2930	/**
2931	* VMMR0_DO_VMMR0_LOG_WAIT_FLUSHED: Waits for the flusher thread to finish all
2932	* buffers for logger @a idxLogger.
2933	*
2934	* @returns VBox status code.
2935	* @param pGVM The global (ring-0) VM structure.
2936	* @param idCpu The ID of the calling EMT.
2937	* @param idxLogger Which logger to wait on.
2938	* @thread EMT(idCpu)
2939	*/
2940	static int vmmR0LogWaitFlushed(PGVM pGVM, VMCPUID idCpu, size_t idxLogger)
2941	{
2942	/*
2943	* Check sanity. First we require EMT to be calling us.
2944	*/
2945	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID);
2946	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
2947	AssertReturn(pGVCpu->hEMT == RTThreadNativeSelf(), VERR_INVALID_CPU_ID);
2948	AssertReturn(idxLogger < VMMLOGGER_IDX_MAX, VERR_OUT_OF_RANGE);
2949	PVMMR0PERVCPULOGGER const pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
2950
2951	/*
2952	* Do the waiting.
2953	*/
2954	int rc = VINF_SUCCESS;
2955	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2956	uint32_t cFlushing = pR0Log->cFlushing;
2957	while (cFlushing > 0)
2958	{
2959	pR0Log->fEmtWaiting = true;
2960	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2961
2962	rc = RTSemEventWaitNoResume(pR0Log->hEventFlushWait, RT_MS_5MIN);
2963
2964	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2965	pR0Log->fEmtWaiting = false;
2966	if (RT_SUCCESS(rc))
2967	{
2968	/* Read the new count, make sure it decreased before looping. That
2969	way we can guarentee that we will only wait more than 5 min * buffers. */
2970	uint32_t const cPrevFlushing = cFlushing;
2971	cFlushing = pR0Log->cFlushing;
2972	if (cFlushing < cPrevFlushing)
2973	continue;
2974	rc = VERR_INTERNAL_ERROR_3;
2975	}
2976	break;
2977	}
2978	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
2979	return rc;
2980	}
2981
2982
2983	/**
2984	* Inner worker for vmmR0LoggerFlushCommon.
2985	*/
2986	static bool vmmR0LoggerFlushInner(PGVM pGVM, PGVMCPU pGVCpu, uint32_t idxLogger, size_t idxBuffer, uint32_t cbToFlush)
2987	{
2988	PVMMR0PERVCPULOGGER const pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
2989	PVMMR3CPULOGGER const pShared = &pGVCpu->vmm.s.u.aLoggers[idxLogger];
2990
2991	/*
2992	* Figure out what we need to do and whether we can.
2993	*/
2994	enum { kJustSignal, kPrepAndSignal, kPrepSignalAndWait } enmAction;
2995	#if VMMLOGGER_BUFFER_COUNT >= 2
2996	if (pR0Log->cFlushing < VMMLOGGER_BUFFER_COUNT - 1)
2997	{
2998	if (RTSemEventIsSignalSafe())
2999	enmAction = kJustSignal;
3000	else if (VMMRZCallRing3IsEnabled(pGVCpu))
3001	enmAction = kPrepAndSignal;
3002	else
3003	{
3004	/** @todo This is a bit simplistic. We could introduce a FF to signal the
3005	* thread or similar. */
3006	STAM_REL_COUNTER_INC(&pShared->StatCannotBlock);
3007	# if defined(RT_OS_LINUX)
3008	SUP_DPRINTF(("vmmR0LoggerFlush: Signalling not safe and EMT blocking disabled! (%u bytes)\n", cbToFlush));
3009	# endif
3010	pShared->cbDropped += cbToFlush;
3011	return true;
3012	}
3013	}
3014	else
3015	#endif
3016	if (VMMRZCallRing3IsEnabled(pGVCpu))
3017	enmAction = kPrepSignalAndWait;
3018	else
3019	{
3020	STAM_REL_COUNTER_INC(&pShared->StatCannotBlock);
3021	# if defined(RT_OS_LINUX)
3022	SUP_DPRINTF(("vmmR0LoggerFlush: EMT blocking disabled! (%u bytes)\n", cbToFlush));
3023	# endif
3024	pShared->cbDropped += cbToFlush;
3025	return true;
3026	}
3027
3028	/*
3029	* Prepare for blocking if necessary.
3030	*/
3031	VMMR0EMTBLOCKCTX Ctx;
3032	if (enmAction != kJustSignal)
3033	{
3034	int rc = VMMR0EmtPrepareToBlock(pGVCpu, VINF_SUCCESS, "vmmR0LoggerFlushInner", pR0Log->hEventFlushWait, &Ctx);
3035	if (RT_SUCCESS(rc))
3036	{ /* likely */ }
3037	else
3038	{
3039	STAM_REL_COUNTER_INC(&pShared->StatCannotBlock);
3040	SUP_DPRINTF(("vmmR0LoggerFlush: VMMR0EmtPrepareToBlock failed! rc=%d\n", rc));
3041	return false;
3042	}
3043	}
3044
3045	/*
3046	* Queue the flush job.
3047	*/
3048	bool fFlushedBuffer;
3049	RTSpinlockAcquire(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3050	if (pGVM->vmmr0.s.LogFlusher.fThreadRunning)
3051	{
3052	uint32_t const idxHead = pGVM->vmmr0.s.LogFlusher.idxRingHead % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
3053	uint32_t const idxTail = pGVM->vmmr0.s.LogFlusher.idxRingTail % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
3054	uint32_t const idxNewTail = (idxTail + 1) % RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing);
3055	if (idxNewTail != idxHead)
3056	{
3057	/* Queue it. */
3058	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.idCpu = pGVCpu->idCpu;
3059	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.idxLogger = idxLogger;
3060	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.idxBuffer = (uint32_t)idxBuffer;
3061	pGVM->vmmr0.s.LogFlusher.aRing[idxTail].s.fProcessing = 0;
3062	pGVM->vmmr0.s.LogFlusher.idxRingTail = idxNewTail;
3063
3064	/* Update the number of buffers currently being flushed. */
3065	uint32_t cFlushing = pR0Log->cFlushing;
3066	cFlushing = RT_MIN(cFlushing + 1, VMMLOGGER_BUFFER_COUNT);
3067	pShared->cFlushing = pR0Log->cFlushing = cFlushing;
3068
3069	/* We must wait if all buffers are currently being flushed. */
3070	bool const fEmtWaiting = cFlushing >= VMMLOGGER_BUFFER_COUNT && enmAction != kJustSignal /* paranoia */;
3071	pR0Log->fEmtWaiting = fEmtWaiting;
3072
3073	/* Stats. */
3074	STAM_REL_COUNTER_INC(&pShared->StatFlushes);
3075	STAM_REL_COUNTER_INC(&pGVM->vmm.s.StatLogFlusherFlushes);
3076
3077	/* Signal the worker thread. */
3078	if (pGVM->vmmr0.s.LogFlusher.fThreadWaiting)
3079	{
3080	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3081	RTSemEventSignal(pGVM->vmmr0.s.LogFlusher.hEvent);
3082	}
3083	else
3084	{
3085	STAM_REL_COUNTER_INC(&pGVM->vmm.s.StatLogFlusherNoWakeUp);
3086	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3087	}
3088
3089	/*
3090	* Wait for a buffer to finish flushing.
3091	*
3092	* Note! Lazy bird is ignoring the status code here. The result is
3093	* that we might end up with an extra even signalling and the
3094	* next time we need to wait we won't and end up with some log
3095	* corruption. However, it's too much hazzle right now for
3096	* a scenario which would most likely end the process rather
3097	* than causing log corruption.
3098	*/
3099	if (fEmtWaiting)
3100	{
3101	STAM_REL_PROFILE_START(&pShared->StatWait, a);
3102	VMMR0EmtWaitEventInner(pGVCpu, VMMR0EMTWAIT_F_TRY_SUPPRESS_INTERRUPTED,
3103	pR0Log->hEventFlushWait, RT_INDEFINITE_WAIT);
3104	STAM_REL_PROFILE_STOP(&pShared->StatWait, a);
3105	}
3106
3107	/*
3108	* We always switch buffer if we have more than one.
3109	*/
3110	#if VMMLOGGER_BUFFER_COUNT == 1
3111	fFlushedBuffer = true;
3112	#else
3113	AssertCompile(VMMLOGGER_BUFFER_COUNT >= 1);
3114	pShared->idxBuf = (idxBuffer + 1) % VMMLOGGER_BUFFER_COUNT;
3115	fFlushedBuffer = false;
3116	#endif
3117	}
3118	else
3119	{
3120	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3121	SUP_DPRINTF(("vmmR0LoggerFlush: ring buffer is full!\n"));
3122	fFlushedBuffer = true;
3123	}
3124	}
3125	else
3126	{
3127	RTSpinlockRelease(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3128	SUP_DPRINTF(("vmmR0LoggerFlush: flusher not active - dropping %u bytes\n", cbToFlush));
3129	fFlushedBuffer = true;
3130	}
3131
3132	/*
3133	* Restore the HM context.
3134	*/
3135	if (enmAction != kJustSignal)
3136	VMMR0EmtResumeAfterBlocking(pGVCpu, &Ctx);
3137
3138	return fFlushedBuffer;
3139	}
3140
3141
3142	/**
3143	* Common worker for vmmR0LogFlush and vmmR0LogRelFlush.
3144	*/
3145	static bool vmmR0LoggerFlushCommon(PRTLOGGER pLogger, PRTLOGBUFFERDESC pBufDesc, uint32_t idxLogger)
3146	{
3147	/*
3148	* Convert the pLogger into a GVMCPU handle and 'call' back to Ring-3.
3149	* (This is a bit paranoid code.)
3150	*/
3151	if (RT_VALID_PTR(pLogger))
3152	{
3153	if ( pLogger->u32Magic == RTLOGGER_MAGIC
3154	&& (pLogger->u32UserValue1 & VMMR0_LOGGER_FLAGS_MAGIC_MASK) == VMMR0_LOGGER_FLAGS_MAGIC_VALUE
3155	&& pLogger->u64UserValue2 == pLogger->u64UserValue3)
3156	{
3157	PGVMCPU const pGVCpu = (PGVMCPU)(uintptr_t)pLogger->u64UserValue2;
3158	if ( RT_VALID_PTR(pGVCpu)
3159	&& ((uintptr_t)pGVCpu & HOST_PAGE_OFFSET_MASK) == 0)
3160	{
3161	RTNATIVETHREAD const hNativeSelf = RTThreadNativeSelf();
3162	PGVM const pGVM = pGVCpu->pGVM;
3163	if ( hNativeSelf == pGVCpu->hEMT
3164	&& RT_VALID_PTR(pGVM))
3165	{
3166	PVMMR0PERVCPULOGGER const pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
3167	size_t const idxBuffer = pBufDesc - &pR0Log->aBufDescs[0];
3168	if (idxBuffer < VMMLOGGER_BUFFER_COUNT)
3169	{
3170	/*
3171	* Make sure we don't recurse forever here should something in the
3172	* following code trigger logging or an assertion. Do the rest in
3173	* an inner work to avoid hitting the right margin too hard.
3174	*/
3175	if (!pR0Log->fFlushing)
3176	{
3177	pR0Log->fFlushing = true;
3178	bool fFlushed = vmmR0LoggerFlushInner(pGVM, pGVCpu, idxLogger, idxBuffer, pBufDesc->offBuf);
3179	pR0Log->fFlushing = false;
3180	return fFlushed;
3181	}
3182
3183	SUP_DPRINTF(("vmmR0LoggerFlush: Recursive flushing!\n"));
3184	}
3185	else
3186	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p pGVCpu=%p: idxBuffer=%#zx\n", pLogger, pGVCpu, idxBuffer));
3187	}
3188	else
3189	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p pGVCpu=%p hEMT=%p hNativeSelf=%p!\n",
3190	pLogger, pGVCpu, pGVCpu->hEMT, hNativeSelf));
3191	}
3192	else
3193	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p pGVCpu=%p!\n", pLogger, pGVCpu));
3194	}
3195	else
3196	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p u32Magic=%#x u32UserValue1=%#x u64UserValue2=%#RX64 u64UserValue3=%#RX64!\n",
3197	pLogger, pLogger->u32Magic, pLogger->u32UserValue1, pLogger->u64UserValue2, pLogger->u64UserValue3));
3198	}
3199	else
3200	SUP_DPRINTF(("vmmR0LoggerFlush: pLogger=%p!\n", pLogger));
3201	return true;
3202	}
3203
3204
3205	/**
3206	* @callback_method_impl{FNRTLOGFLUSH, Release logger buffer flush callback.}
3207	*/
3208	static DECLCALLBACK(bool) vmmR0LogRelFlush(PRTLOGGER pLogger, PRTLOGBUFFERDESC pBufDesc)
3209	{
3210	return vmmR0LoggerFlushCommon(pLogger, pBufDesc, VMMLOGGER_IDX_RELEASE);
3211	}
3212
3213
3214	/**
3215	* @callback_method_impl{FNRTLOGFLUSH, Logger (debug) buffer flush callback.}
3216	*/
3217	static DECLCALLBACK(bool) vmmR0LogFlush(PRTLOGGER pLogger, PRTLOGBUFFERDESC pBufDesc)
3218	{
3219	#ifdef LOG_ENABLED
3220	return vmmR0LoggerFlushCommon(pLogger, pBufDesc, VMMLOGGER_IDX_REGULAR);
3221	#else
3222	RT_NOREF(pLogger, pBufDesc);
3223	return true;
3224	#endif
3225	}
3226
3227
3228	/*
3229	* Override RTLogDefaultInstanceEx so we can do logging from EMTs in ring-0.
3230	*/
3231	DECLEXPORT(PRTLOGGER) RTLogDefaultInstanceEx(uint32_t fFlagsAndGroup)
3232	{
3233	#ifdef LOG_ENABLED
3234	PGVMCPU pGVCpu = GVMMR0GetGVCpuByEMT(NIL_RTNATIVETHREAD);
3235	if (pGVCpu)
3236	{
3237	PRTLOGGER pLogger = pGVCpu->vmmr0.s.u.s.Logger.pLogger;
3238	if (RT_VALID_PTR(pLogger))
3239	{
3240	if ( pLogger->u64UserValue2 == (uintptr_t)pGVCpu
3241	&& pLogger->u64UserValue3 == (uintptr_t)pGVCpu)
3242	{
3243	if (!pGVCpu->vmmr0.s.u.s.Logger.fFlushing)
3244	{
3245	if (!(pGVCpu->vmmr0.s.fLogFlushingDisabled))
3246	return RTLogCheckGroupFlags(pLogger, fFlagsAndGroup);
3247	return NULL;
3248	}
3249
3250	/*
3251	* When we're flushing we _must_ return NULL here to suppress any
3252	* attempts at using the logger while in vmmR0LoggerFlushCommon.
3253	* The VMMR0EmtPrepareToBlock code may trigger logging in HM,
3254	* which will reset the buffer content before we even get to queue
3255	* the flush request. (Only an issue when VBOX_WITH_R0_LOGGING
3256	* is enabled.)
3257	*/
3258	return NULL;
3259	}
3260	}
3261	}
3262	#endif
3263	return SUPR0DefaultLogInstanceEx(fFlagsAndGroup);
3264	}
3265
3266
3267	/*
3268	* Override RTLogRelGetDefaultInstanceEx so we can do LogRel to VBox.log from EMTs in ring-0.
3269	*/
3270	DECLEXPORT(PRTLOGGER) RTLogRelGetDefaultInstanceEx(uint32_t fFlagsAndGroup)
3271	{
3272	PGVMCPU pGVCpu = GVMMR0GetGVCpuByEMT(NIL_RTNATIVETHREAD);
3273	if (pGVCpu)
3274	{
3275	PRTLOGGER pLogger = pGVCpu->vmmr0.s.u.s.RelLogger.pLogger;
3276	if (RT_VALID_PTR(pLogger))
3277	{
3278	if ( pLogger->u64UserValue2 == (uintptr_t)pGVCpu
3279	&& pLogger->u64UserValue3 == (uintptr_t)pGVCpu)
3280	{
3281	if (!pGVCpu->vmmr0.s.u.s.RelLogger.fFlushing)
3282	{
3283	if (!(pGVCpu->vmmr0.s.fLogFlushingDisabled))
3284	return RTLogCheckGroupFlags(pLogger, fFlagsAndGroup);
3285	return NULL;
3286	}
3287	}
3288	}
3289	}
3290	return SUPR0GetDefaultLogRelInstanceEx(fFlagsAndGroup);
3291	}
3292
3293
3294	/**
3295	* Helper for vmmR0InitLoggerSet
3296	*/
3297	static int vmmR0InitLoggerOne(PGVMCPU pGVCpu, bool fRelease, PVMMR0PERVCPULOGGER pR0Log, PVMMR3CPULOGGER pShared,
3298	uint32_t cbBuf, char *pchBuf, RTR3PTR pchBufR3)
3299	{
3300	/*
3301	* Create and configure the logger.
3302	*/
3303	for (size_t i = 0; i < VMMLOGGER_BUFFER_COUNT; i++)
3304	{
3305	pR0Log->aBufDescs[i].u32Magic = RTLOGBUFFERDESC_MAGIC;
3306	pR0Log->aBufDescs[i].uReserved = 0;
3307	pR0Log->aBufDescs[i].cbBuf = cbBuf;
3308	pR0Log->aBufDescs[i].offBuf = 0;
3309	pR0Log->aBufDescs[i].pchBuf = pchBuf + i * cbBuf;
3310	pR0Log->aBufDescs[i].pAux = &pShared->aBufs[i].AuxDesc;
3311
3312	pShared->aBufs[i].AuxDesc.fFlushedIndicator = false;
3313	pShared->aBufs[i].AuxDesc.afPadding[0] = 0;
3314	pShared->aBufs[i].AuxDesc.afPadding[1] = 0;
3315	pShared->aBufs[i].AuxDesc.afPadding[2] = 0;
3316	pShared->aBufs[i].AuxDesc.offBuf = 0;
3317	pShared->aBufs[i].pchBufR3 = pchBufR3 + i * cbBuf;
3318	}
3319	pShared->cbBuf = cbBuf;
3320
3321	static const char * const s_apszGroups[] = VBOX_LOGGROUP_NAMES;
3322	int rc = RTLogCreateEx(&pR0Log->pLogger, fRelease ? "VBOX_RELEASE_LOG" : "VBOX_LOG", RTLOG_F_NO_LOCKING \| RTLOGFLAGS_BUFFERED,
3323	"all", RT_ELEMENTS(s_apszGroups), s_apszGroups, UINT32_MAX,
3324	VMMLOGGER_BUFFER_COUNT, pR0Log->aBufDescs, RTLOGDEST_DUMMY,
3325	NULL /pfnPhase/, 0 /cHistory/, 0 /cbHistoryFileMax/, 0 /cSecsHistoryTimeSlot/,
3326	NULL /pErrInfo/, NULL /pszFilenameFmt/);
3327	if (RT_SUCCESS(rc))
3328	{
3329	PRTLOGGER pLogger = pR0Log->pLogger;
3330	pLogger->u32UserValue1 = VMMR0_LOGGER_FLAGS_MAGIC_VALUE;
3331	pLogger->u64UserValue2 = (uintptr_t)pGVCpu;
3332	pLogger->u64UserValue3 = (uintptr_t)pGVCpu;
3333
3334	rc = RTLogSetFlushCallback(pLogger, fRelease ? vmmR0LogRelFlush : vmmR0LogFlush);
3335	if (RT_SUCCESS(rc))
3336	{
3337	RTLogSetR0ThreadNameF(pLogger, "EMT-%u-R0", pGVCpu->idCpu);
3338
3339	/*
3340	* Create the event sem the EMT waits on while flushing is happening.
3341	*/
3342	rc = RTSemEventCreate(&pR0Log->hEventFlushWait);
3343	if (RT_SUCCESS(rc))
3344	return VINF_SUCCESS;
3345	pR0Log->hEventFlushWait = NIL_RTSEMEVENT;
3346	}
3347	RTLogDestroy(pLogger);
3348	}
3349	pR0Log->pLogger = NULL;
3350	return rc;
3351	}
3352
3353
3354	/**
3355	* Worker for VMMR0CleanupVM and vmmR0InitLoggerSet that destroys one logger.
3356	*/
3357	static void vmmR0TermLoggerOne(PVMMR0PERVCPULOGGER pR0Log, PVMMR3CPULOGGER pShared)
3358	{
3359	RTLogDestroy(pR0Log->pLogger);
3360	pR0Log->pLogger = NULL;
3361
3362	for (size_t i = 0; i < VMMLOGGER_BUFFER_COUNT; i++)
3363	pShared->aBufs[i].pchBufR3 = NIL_RTR3PTR;
3364
3365	RTSemEventDestroy(pR0Log->hEventFlushWait);
3366	pR0Log->hEventFlushWait = NIL_RTSEMEVENT;
3367	}
3368
3369
3370	/**
3371	* Initializes one type of loggers for each EMT.
3372	*/
3373	static int vmmR0InitLoggerSet(PGVM pGVM, uint8_t idxLogger, uint32_t cbBuf, PRTR0MEMOBJ phMemObj, PRTR0MEMOBJ phMapObj)
3374	{
3375	/* Allocate buffers first. */
3376	int rc = RTR0MemObjAllocPage(phMemObj, cbBuf * pGVM->cCpus * VMMLOGGER_BUFFER_COUNT, false /fExecutable/);
3377	if (RT_SUCCESS(rc))
3378	{
3379	rc = RTR0MemObjMapUser(phMapObj, phMemObj, (RTR3PTR)-1, 0 /uAlignment*/, RTMEM_PROT_READ, NIL_RTR0PROCESS);
3380	if (RT_SUCCESS(rc))
3381	{
3382	char * const pchBuf = (char )RTR0MemObjAddress(phMemObj);
3383	AssertPtrReturn(pchBuf, VERR_INTERNAL_ERROR_2);
3384
3385	RTR3PTR const pchBufR3 = RTR0MemObjAddressR3(*phMapObj);
3386	AssertReturn(pchBufR3 != NIL_RTR3PTR, VERR_INTERNAL_ERROR_3);
3387
3388	/* Initialize the per-CPU loggers. */
3389	for (uint32_t i = 0; i < pGVM->cCpus; i++)
3390	{
3391	PGVMCPU pGVCpu = &pGVM->aCpus[i];
3392	PVMMR0PERVCPULOGGER pR0Log = &pGVCpu->vmmr0.s.u.aLoggers[idxLogger];
3393	PVMMR3CPULOGGER pShared = &pGVCpu->vmm.s.u.aLoggers[idxLogger];
3394	rc = vmmR0InitLoggerOne(pGVCpu, idxLogger == VMMLOGGER_IDX_RELEASE, pR0Log, pShared, cbBuf,
3395	pchBuf + i * cbBuf * VMMLOGGER_BUFFER_COUNT,
3396	pchBufR3 + i * cbBuf * VMMLOGGER_BUFFER_COUNT);
3397	if (RT_FAILURE(rc))
3398	{
3399	vmmR0TermLoggerOne(pR0Log, pShared);
3400	while (i-- > 0)
3401	{
3402	pGVCpu = &pGVM->aCpus[i];
3403	vmmR0TermLoggerOne(&pGVCpu->vmmr0.s.u.aLoggers[idxLogger], &pGVCpu->vmm.s.u.aLoggers[idxLogger]);
3404	}
3405	break;
3406	}
3407	}
3408	if (RT_SUCCESS(rc))
3409	return VINF_SUCCESS;
3410
3411	/* Bail out. */
3412	RTR0MemObjFree(phMapObj, false /fFreeMappings*/);
3413	*phMapObj = NIL_RTR0MEMOBJ;
3414	}
3415	RTR0MemObjFree(phMemObj, true /fFreeMappings*/);
3416	*phMemObj = NIL_RTR0MEMOBJ;
3417	}
3418	return rc;
3419	}
3420
3421
3422	/**
3423	* Worker for VMMR0InitPerVMData that initializes all the logging related stuff.
3424	*
3425	* @returns VBox status code.
3426	* @param pGVM The global (ring-0) VM structure.
3427	*/
3428	static int vmmR0InitLoggers(PGVM pGVM)
3429	{
3430	/*
3431	* Invalidate the ring buffer (not really necessary).
3432	*/
3433	for (size_t idx = 0; idx < RT_ELEMENTS(pGVM->vmmr0.s.LogFlusher.aRing); idx++)
3434	pGVM->vmmr0.s.LogFlusher.aRing[idx].u32 = UINT32_MAX >> 1; /* (all bits except fProcessing set) */
3435
3436	/*
3437	* Create the spinlock and flusher event semaphore.
3438	*/
3439	int rc = RTSpinlockCreate(&pGVM->vmmr0.s.LogFlusher.hSpinlock, RTSPINLOCK_FLAGS_INTERRUPT_SAFE, "VM-Log-Flusher");
3440	if (RT_SUCCESS(rc))
3441	{
3442	rc = RTSemEventCreate(&pGVM->vmmr0.s.LogFlusher.hEvent);
3443	if (RT_SUCCESS(rc))
3444	{
3445	/*
3446	* Create the ring-0 release loggers.
3447	*/
3448	rc = vmmR0InitLoggerSet(pGVM, VMMLOGGER_IDX_RELEASE, _4K,
3449	&pGVM->vmmr0.s.hMemObjReleaseLogger, &pGVM->vmmr0.s.hMapObjReleaseLogger);
3450	#ifdef LOG_ENABLED
3451	if (RT_SUCCESS(rc))
3452	{
3453	/*
3454	* Create debug loggers.
3455	*/
3456	rc = vmmR0InitLoggerSet(pGVM, VMMLOGGER_IDX_REGULAR, _64K,
3457	&pGVM->vmmr0.s.hMemObjLogger, &pGVM->vmmr0.s.hMapObjLogger);
3458	}
3459	#endif
3460	}
3461	}
3462	return rc;
3463	}
3464
3465
3466	/**
3467	* Worker for VMMR0InitPerVMData that initializes all the logging related stuff.
3468	*
3469	* @param pGVM The global (ring-0) VM structure.
3470	*/
3471	static void vmmR0CleanupLoggers(PGVM pGVM)
3472	{
3473	for (VMCPUID idCpu = 0; idCpu < pGVM->cCpus; idCpu++)
3474	{
3475	PGVMCPU pGVCpu = &pGVM->aCpus[idCpu];
3476	for (size_t iLogger = 0; iLogger < RT_ELEMENTS(pGVCpu->vmmr0.s.u.aLoggers); iLogger++)
3477	vmmR0TermLoggerOne(&pGVCpu->vmmr0.s.u.aLoggers[iLogger], &pGVCpu->vmm.s.u.aLoggers[iLogger]);
3478	}
3479
3480	/*
3481	* Free logger buffer memory.
3482	*/
3483	RTR0MemObjFree(pGVM->vmmr0.s.hMapObjReleaseLogger, false /fFreeMappings/);
3484	pGVM->vmmr0.s.hMapObjReleaseLogger = NIL_RTR0MEMOBJ;
3485	RTR0MemObjFree(pGVM->vmmr0.s.hMemObjReleaseLogger, true /fFreeMappings/);
3486	pGVM->vmmr0.s.hMemObjReleaseLogger = NIL_RTR0MEMOBJ;
3487
3488	RTR0MemObjFree(pGVM->vmmr0.s.hMapObjLogger, false /fFreeMappings/);
3489	pGVM->vmmr0.s.hMapObjLogger = NIL_RTR0MEMOBJ;
3490	RTR0MemObjFree(pGVM->vmmr0.s.hMemObjLogger, true /fFreeMappings/);
3491	pGVM->vmmr0.s.hMemObjLogger = NIL_RTR0MEMOBJ;
3492
3493	/*
3494	* Free log flusher related stuff.
3495	*/
3496	RTSpinlockDestroy(pGVM->vmmr0.s.LogFlusher.hSpinlock);
3497	pGVM->vmmr0.s.LogFlusher.hSpinlock = NIL_RTSPINLOCK;
3498	RTSemEventDestroy(pGVM->vmmr0.s.LogFlusher.hEvent);
3499	pGVM->vmmr0.s.LogFlusher.hEvent = NIL_RTSEMEVENT;
3500	}
3501
3502
3503	/*********************************************************************************************************************************
3504	* Assertions *
3505	*********************************************************************************************************************************/
3506
3507	/**
3508	* Installs a notification callback for ring-0 assertions.
3509	*
3510	* @param pVCpu The cross context virtual CPU structure.
3511	* @param pfnCallback Pointer to the callback.
3512	* @param pvUser The user argument.
3513	*
3514	* @return VBox status code.
3515	*/
3516	VMMR0_INT_DECL(int) VMMR0AssertionSetNotification(PVMCPUCC pVCpu, PFNVMMR0ASSERTIONNOTIFICATION pfnCallback, RTR0PTR pvUser)
3517	{
3518	AssertPtrReturn(pVCpu, VERR_INVALID_POINTER);
3519	AssertPtrReturn(pfnCallback, VERR_INVALID_POINTER);
3520
3521	if (!pVCpu->vmmr0.s.pfnAssertCallback)
3522	{
3523	pVCpu->vmmr0.s.pfnAssertCallback = pfnCallback;
3524	pVCpu->vmmr0.s.pvAssertCallbackUser = pvUser;
3525	return VINF_SUCCESS;
3526	}
3527	return VERR_ALREADY_EXISTS;
3528	}
3529
3530
3531	/**
3532	* Removes the ring-0 callback.
3533	*
3534	* @param pVCpu The cross context virtual CPU structure.
3535	*/
3536	VMMR0_INT_DECL(void) VMMR0AssertionRemoveNotification(PVMCPUCC pVCpu)
3537	{
3538	pVCpu->vmmr0.s.pfnAssertCallback = NULL;
3539	pVCpu->vmmr0.s.pvAssertCallbackUser = NULL;
3540	}
3541
3542
3543	/**
3544	* Checks whether there is a ring-0 callback notification active.
3545	*
3546	* @param pVCpu The cross context virtual CPU structure.
3547	* @returns true if there the notification is active, false otherwise.
3548	*/
3549	VMMR0_INT_DECL(bool) VMMR0AssertionIsNotificationSet(PVMCPUCC pVCpu)
3550	{
3551	return pVCpu->vmmr0.s.pfnAssertCallback != NULL;
3552	}
3553
3554
3555	/*
3556	* Jump back to ring-3 if we're the EMT and the longjmp is armed.
3557	*
3558	* @returns true if the breakpoint should be hit, false if it should be ignored.
3559	*/
3560	DECLEXPORT(bool) RTCALL RTAssertShouldPanic(void)
3561	{
3562	#if 0
3563	return true;
3564	#else
3565	PVMCC pVM = GVMMR0GetVMByEMT(NIL_RTNATIVETHREAD);
3566	if (pVM)
3567	{
3568	PVMCPUCC pVCpu = VMMGetCpu(pVM);
3569
3570	if (pVCpu)
3571	{
3572	# ifdef RT_ARCH_X86
3573	if (pVCpu->vmmr0.s.AssertJmpBuf.eip)
3574	# else
3575	if (pVCpu->vmmr0.s.AssertJmpBuf.rip)
3576	# endif
3577	{
3578	if (pVCpu->vmmr0.s.pfnAssertCallback)
3579	pVCpu->vmmr0.s.pfnAssertCallback(pVCpu, pVCpu->vmmr0.s.pvAssertCallbackUser);
3580	int rc = vmmR0CallRing3LongJmp(&pVCpu->vmmr0.s.AssertJmpBuf, VERR_VMM_RING0_ASSERTION);
3581	return RT_FAILURE_NP(rc);
3582	}
3583	}
3584	}
3585	# ifdef RT_OS_LINUX
3586	return true;
3587	# else
3588	return false;
3589	# endif
3590	#endif
3591	}
3592
3593
3594	/*
3595	* Override this so we can push it up to ring-3.
3596	*/
3597	DECLEXPORT(void) RTCALL RTAssertMsg1Weak(const char pszExpr, unsigned uLine, const char pszFile, const char *pszFunction)
3598	{
3599	/*
3600	* To host kernel log/whatever.
3601	*/
3602	SUPR0Printf("!!R0-Assertion Failed!!\n"
3603	"Expression: %s\n"
3604	"Location : %s(%d) %s\n",
3605	pszExpr, pszFile, uLine, pszFunction);
3606
3607	/*
3608	* To the log.
3609	*/
3610	LogAlways(("\n!!R0-Assertion Failed!!\n"
3611	"Expression: %s\n"
3612	"Location : %s(%d) %s\n",
3613	pszExpr, pszFile, uLine, pszFunction));
3614
3615	/*
3616	* To the global VMM buffer.
3617	*/
3618	PVMCC pVM = GVMMR0GetVMByEMT(NIL_RTNATIVETHREAD);
3619	if (pVM)
3620	RTStrPrintf(pVM->vmm.s.szRing0AssertMsg1, sizeof(pVM->vmm.s.szRing0AssertMsg1),
3621	"\n!!R0-Assertion Failed!!\n"
3622	"Expression: %.*s\n"
3623	"Location : %s(%d) %s\n",
3624	sizeof(pVM->vmm.s.szRing0AssertMsg1) / 4 * 3, pszExpr,
3625	pszFile, uLine, pszFunction);
3626
3627	/*
3628	* Continue the normal way.
3629	*/
3630	RTAssertMsg1(pszExpr, uLine, pszFile, pszFunction);
3631	}
3632
3633
3634	/**
3635	* Callback for RTLogFormatV which writes to the ring-3 log port.
3636	* See PFNLOGOUTPUT() for details.
3637	*/
3638	static DECLCALLBACK(size_t) rtLogOutput(void pv, const char pachChars, size_t cbChars)
3639	{
3640	for (size_t i = 0; i < cbChars; i++)
3641	{
3642	LogAlways(("%c", pachChars[i])); NOREF(pachChars);
3643	}
3644
3645	NOREF(pv);
3646	return cbChars;
3647	}
3648
3649
3650	/*
3651	* Override this so we can push it up to ring-3.
3652	*/
3653	DECLEXPORT(void) RTCALL RTAssertMsg2WeakV(const char *pszFormat, va_list va)
3654	{
3655	va_list vaCopy;
3656
3657	/*
3658	* Push the message to the loggers.
3659	*/
3660	PRTLOGGER pLog = RTLogRelGetDefaultInstance();
3661	if (pLog)
3662	{
3663	va_copy(vaCopy, va);
3664	RTLogFormatV(rtLogOutput, pLog, pszFormat, vaCopy);
3665	va_end(vaCopy);
3666	}
3667	pLog = RTLogGetDefaultInstance(); /* Don't initialize it here... */
3668	if (pLog)
3669	{
3670	va_copy(vaCopy, va);
3671	RTLogFormatV(rtLogOutput, pLog, pszFormat, vaCopy);
3672	va_end(vaCopy);
3673	}
3674
3675	/*
3676	* Push it to the global VMM buffer.
3677	*/
3678	PVMCC pVM = GVMMR0GetVMByEMT(NIL_RTNATIVETHREAD);
3679	if (pVM)
3680	{
3681	va_copy(vaCopy, va);
3682	RTStrPrintfV(pVM->vmm.s.szRing0AssertMsg2, sizeof(pVM->vmm.s.szRing0AssertMsg2), pszFormat, vaCopy);
3683	va_end(vaCopy);
3684	}
3685
3686	/*
3687	* Continue the normal way.
3688	*/
3689	RTAssertMsg2V(pszFormat, va);
3690	}
3691

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

source: vbox/trunk/src/VBox/VMM/VMMR0/VMMR0.cpp@ 93705

Download in other formats: