VM.cpp@ 30164

Last change on this file since 30164 was 30111, checked in by vboxsync, 14 years ago
iprt/asm.h,: Revised the ASMAtomicPtr functions and macros. The new saves lots of unsafe (void * volatile *) casts as well as adding some type safety when using GCC (typeof rulez).
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 143.7 KB

Line
1	/* $Id: VM.cpp 30111 2010-06-09 12:14:59Z vboxsync $ */
2	/** @file
3	* VM - Virtual Machine
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/** @page pg_vm VM API
19	*
20	* This is the encapsulating bit. It provides the APIs that Main and VBoxBFE
21	* use to create a VMM instance for running a guest in. It also provides
22	* facilities for queuing request for execution in EMT (serialization purposes
23	* mostly) and for reporting error back to the VMM user (Main/VBoxBFE).
24	*
25	*
26	* @section sec_vm_design Design Critique / Things To Do
27	*
28	* In hindsight this component is a big design mistake, all this stuff really
29	* belongs in the VMM component. It just seemed like a kind of ok idea at a
30	* time when the VMM bit was a kind of vague. 'VM' also happend to be the name
31	* of the per-VM instance structure (see vm.h), so it kind of made sense.
32	* However as it turned out, VMM(.cpp) is almost empty all it provides in ring-3
33	* is some minor functionally and some "routing" services.
34	*
35	* Fixing this is just a matter of some more or less straight forward
36	* refactoring, the question is just when someone will get to it. Moving the EMT
37	* would be a good start.
38	*
39	*/
40
41	/*******************************************************************************
42	* Header Files *
43	*******************************************************************************/
44	#define LOG_GROUP LOG_GROUP_VM
45	#include <VBox/cfgm.h>
46	#include <VBox/vmm.h>
47	#include <VBox/gvmm.h>
48	#include <VBox/mm.h>
49	#include <VBox/cpum.h>
50	#include <VBox/selm.h>
51	#include <VBox/trpm.h>
52	#include <VBox/dbgf.h>
53	#include <VBox/pgm.h>
54	#include <VBox/pdmapi.h>
55	#include <VBox/pdmcritsect.h>
56	#include <VBox/em.h>
57	#include <VBox/rem.h>
58	#include <VBox/tm.h>
59	#include <VBox/stam.h>
60	#include <VBox/patm.h>
61	#include <VBox/csam.h>
62	#include <VBox/iom.h>
63	#include <VBox/ssm.h>
64	#include <VBox/hwaccm.h>
65	#include "VMInternal.h"
66	#include <VBox/vm.h>
67	#include <VBox/uvm.h>
68
69	#include <VBox/sup.h>
70	#include <VBox/dbg.h>
71	#include <VBox/err.h>
72	#include <VBox/param.h>
73	#include <VBox/log.h>
74	#include <iprt/assert.h>
75	#include <iprt/alloc.h>
76	#include <iprt/asm.h>
77	#include <iprt/env.h>
78	#include <iprt/string.h>
79	#include <iprt/time.h>
80	#include <iprt/semaphore.h>
81	#include <iprt/thread.h>
82
83
84	/*******************************************************************************
85	* Structures and Typedefs *
86	*******************************************************************************/
87	/**
88	* VM destruction callback registration record.
89	*/
90	typedef struct VMATDTOR
91	{
92	/** Pointer to the next record in the list. */
93	struct VMATDTOR *pNext;
94	/** Pointer to the callback function. */
95	PFNVMATDTOR pfnAtDtor;
96	/** The user argument. */
97	void *pvUser;
98	} VMATDTOR;
99	/** Pointer to a VM destruction callback registration record. */
100	typedef VMATDTOR *PVMATDTOR;
101
102
103	/*******************************************************************************
104	* Global Variables *
105	*******************************************************************************/
106	/** Pointer to the list of VMs. */
107	static PUVM g_pUVMsHead = NULL;
108
109	/** Pointer to the list of at VM destruction callbacks. */
110	static PVMATDTOR g_pVMAtDtorHead = NULL;
111	/** Lock the g_pVMAtDtorHead list. */
112	#define VM_ATDTOR_LOCK() do { } while (0)
113	/** Unlock the g_pVMAtDtorHead list. */
114	#define VM_ATDTOR_UNLOCK() do { } while (0)
115
116
117	/*******************************************************************************
118	* Internal Functions *
119	*******************************************************************************/
120	static int vmR3CreateUVM(uint32_t cCpus, PUVM *ppUVM);
121	static int vmR3CreateU(PUVM pUVM, uint32_t cCpus, PFNCFGMCONSTRUCTOR pfnCFGMConstructor, void *pvUserCFGM);
122	static int vmR3InitRing3(PVM pVM, PUVM pUVM);
123	static int vmR3InitVMCpu(PVM pVM);
124	static int vmR3InitRing0(PVM pVM);
125	static int vmR3InitGC(PVM pVM);
126	static int vmR3InitDoCompleted(PVM pVM, VMINITCOMPLETED enmWhat);
127	static DECLCALLBACK(size_t) vmR3LogPrefixCallback(PRTLOGGER pLogger, char pchBuf, size_t cchBuf, void pvUser);
128	static void vmR3DestroyUVM(PUVM pUVM, uint32_t cMilliesEMTWait);
129	static void vmR3AtDtor(PVM pVM);
130	static bool vmR3ValidateStateTransition(VMSTATE enmStateOld, VMSTATE enmStateNew);
131	static void vmR3DoAtState(PVM pVM, PUVM pUVM, VMSTATE enmStateNew, VMSTATE enmStateOld);
132	static int vmR3TrySetState(PVM pVM, const char *pszWho, unsigned cTransitions, ...);
133	static void vmR3SetStateLocked(PVM pVM, PUVM pUVM, VMSTATE enmStateNew, VMSTATE enmStateOld);
134	static void vmR3SetState(PVM pVM, VMSTATE enmStateNew, VMSTATE enmStateOld);
135	static int vmR3SetErrorU(PUVM pUVM, int rc, RT_SRC_POS_DECL, const char *pszFormat, ...);
136
137
138	/**
139	* Do global VMM init.
140	*
141	* @returns VBox status code.
142	*/
143	VMMR3DECL(int) VMR3GlobalInit(void)
144	{
145	/*
146	* Only once.
147	*/
148	static bool volatile s_fDone = false;
149	if (s_fDone)
150	return VINF_SUCCESS;
151
152	/*
153	* We're done.
154	*/
155	s_fDone = true;
156	return VINF_SUCCESS;
157	}
158
159
160
161	/**
162	* Creates a virtual machine by calling the supplied configuration constructor.
163	*
164	* On successful returned the VM is powered, i.e. VMR3PowerOn() should be
165	* called to start the execution.
166	*
167	* @returns 0 on success.
168	* @returns VBox error code on failure.
169	* @param cCpus Number of virtual CPUs for the new VM.
170	* @param pfnVMAtError Pointer to callback function for setting VM
171	* errors. This was added as an implicit call to
172	* VMR3AtErrorRegister() since there is no way the
173	* caller can get to the VM handle early enough to
174	* do this on its own.
175	* This is called in the context of an EMT.
176	* @param pvUserVM The user argument passed to pfnVMAtError.
177	* @param pfnCFGMConstructor Pointer to callback function for constructing the VM configuration tree.
178	* This is called in the context of an EMT0.
179	* @param pvUserCFGM The user argument passed to pfnCFGMConstructor.
180	* @param ppVM Where to store the 'handle' of the created VM.
181	*/
182	VMMR3DECL(int) VMR3Create(uint32_t cCpus, PFNVMATERROR pfnVMAtError, void pvUserVM, PFNCFGMCONSTRUCTOR pfnCFGMConstructor, void pvUserCFGM, PVM *ppVM)
183	{
184	LogFlow(("VMR3Create: cCpus=%RU32 pfnVMAtError=%p pvUserVM=%p pfnCFGMConstructor=%p pvUserCFGM=%p ppVM=%p\n",
185	cCpus, pfnVMAtError, pvUserVM, pfnCFGMConstructor, pvUserCFGM, ppVM));
186
187	/*
188	* Because of the current hackiness of the applications
189	* we'll have to initialize global stuff from here.
190	* Later the applications will take care of this in a proper way.
191	*/
192	static bool fGlobalInitDone = false;
193	if (!fGlobalInitDone)
194	{
195	int rc = VMR3GlobalInit();
196	if (RT_FAILURE(rc))
197	return rc;
198	fGlobalInitDone = true;
199	}
200
201	/*
202	* Validate input.
203	*/
204	AssertLogRelMsgReturn(cCpus > 0 && cCpus <= VMM_MAX_CPU_COUNT, ("%RU32\n", cCpus), VERR_TOO_MANY_CPUS);
205
206	/*
207	* Create the UVM so we can register the at-error callback
208	* and consoliate a bit of cleanup code.
209	*/
210	PUVM pUVM = NULL; /* shuts up gcc */
211	int rc = vmR3CreateUVM(cCpus, &pUVM);
212	if (RT_FAILURE(rc))
213	return rc;
214	if (pfnVMAtError)
215	rc = VMR3AtErrorRegisterU(pUVM, pfnVMAtError, pvUserVM);
216	if (RT_SUCCESS(rc))
217	{
218	/*
219	* Initialize the support library creating the session for this VM.
220	*/
221	rc = SUPR3Init(&pUVM->vm.s.pSession);
222	if (RT_SUCCESS(rc))
223	{
224	/*
225	* Call vmR3CreateU in the EMT thread and wait for it to finish.
226	*
227	* Note! VMCPUID_ANY is used here because VMR3ReqQueueU would have trouble
228	* submitting a request to a specific VCPU without a pVM. So, to make
229	* sure init is running on EMT(0), vmR3EmulationThreadWithId makes sure
230	* that only EMT(0) is servicing VMCPUID_ANY requests when pVM is NULL.
231	*/
232	PVMREQ pReq;
233	rc = VMR3ReqCallU(pUVM, VMCPUID_ANY, &pReq, RT_INDEFINITE_WAIT, VMREQFLAGS_VBOX_STATUS,
234	(PFNRT)vmR3CreateU, 4, pUVM, cCpus, pfnCFGMConstructor, pvUserCFGM);
235	if (RT_SUCCESS(rc))
236	{
237	rc = pReq->iStatus;
238	VMR3ReqFree(pReq);
239	if (RT_SUCCESS(rc))
240	{
241	/*
242	* Success!
243	*/
244	*ppVM = pUVM->pVM;
245	LogFlow(("VMR3Create: returns VINF_SUCCESS ppVM=%p\n", ppVM));
246	return VINF_SUCCESS;
247	}
248	}
249	else
250	AssertMsgFailed(("VMR3ReqCallU failed rc=%Rrc\n", rc));
251
252	/*
253	* An error occurred during VM creation. Set the error message directly
254	* using the initial callback, as the callback list doesn't exist yet.
255	*/
256	const char *pszError = NULL;
257	switch (rc)
258	{
259	case VERR_VMX_IN_VMX_ROOT_MODE:
260	#ifdef RT_OS_LINUX
261	pszError = N_("VirtualBox can't operate in VMX root mode. "
262	"Please disable the KVM kernel extension, recompile your kernel and reboot");
263	#else
264	pszError = N_("VirtualBox can't operate in VMX root mode. Please close all other virtualization programs.");
265	#endif
266	break;
267
268	case VERR_SVM_IN_USE:
269	#ifdef RT_OS_LINUX
270	pszError = N_("VirtualBox can't enable the AMD-V extension. "
271	"Please disable the KVM kernel extension, recompile your kernel and reboot");
272	#else
273	pszError = N_("VirtualBox can't enable the AMD-V extension. Please close all other virtualization programs.");
274	#endif
275	break;
276
277	case VERR_VERSION_MISMATCH:
278	pszError = N_("VMMR0 driver version mismatch. Please terminate all VMs, make sure that "
279	"VBoxNetDHCP is not running and try again. If you still get this error, "
280	"re-install VirtualBox");
281	break;
282
283	#ifdef RT_OS_LINUX
284	case VERR_SUPDRV_COMPONENT_NOT_FOUND:
285	pszError = N_("One of the kernel modules was not successfully loaded. Make sure "
286	"that no kernel modules from an older version of VirtualBox exist. "
287	"Then try to recompile and reload the kernel modules by executing "
288	"'/etc/init.d/vboxdrv setup' as root");
289	break;
290	#endif
291
292	case VERR_RAW_MODE_INVALID_SMP:
293	pszError = N_("VT-x/AMD-V is either not available on your host or disabled. "
294	"VirtualBox requires this hardware extension to emulate more than one "
295	"guest CPU");
296	break;
297
298	case VERR_SUPDRV_KERNEL_TOO_OLD_FOR_VTX:
299	#ifdef RT_OS_LINUX
300	pszError = N_("Because the host kernel is too old, VirtualBox cannot enable the VT-x "
301	"extension. Either upgrade your kernel to Linux 2.6.13 or later or disable "
302	"the VT-x extension in the VM settings. Note that without VT-x you have "
303	"to reduce the number of guest CPUs to one");
304	#else
305	pszError = N_("Because the host kernel is too old, VirtualBox cannot enable the VT-x "
306	"extension. Either upgrade your kernel or disable the VT-x extension in the "
307	"VM settings. Note that without VT-x you have to reduce the number of guest "
308	"CPUs to one");
309	#endif
310	break;
311
312	default:
313	pszError = N_("Unknown error creating VM");
314	break;
315	}
316	vmR3SetErrorU(pUVM, rc, RT_SRC_POS, pszError, rc);
317	}
318	else
319	{
320	/*
321	* An error occurred at support library initialization time (before the
322	* VM could be created). Set the error message directly using the
323	* initial callback, as the callback list doesn't exist yet.
324	*/
325	const char *pszError;
326	switch (rc)
327	{
328	case VERR_VM_DRIVER_LOAD_ERROR:
329	#ifdef RT_OS_LINUX
330	pszError = N_("VirtualBox kernel driver not loaded. The vboxdrv kernel module "
331	"was either not loaded or /dev/vboxdrv is not set up properly. "
332	"Re-setup the kernel module by executing "
333	"'/etc/init.d/vboxdrv setup' as root");
334	#else
335	pszError = N_("VirtualBox kernel driver not loaded");
336	#endif
337	break;
338	case VERR_VM_DRIVER_OPEN_ERROR:
339	pszError = N_("VirtualBox kernel driver cannot be opened");
340	break;
341	case VERR_VM_DRIVER_NOT_ACCESSIBLE:
342	#ifdef VBOX_WITH_HARDENING
343	/* This should only happen if the executable wasn't hardened - bad code/build. */
344	pszError = N_("VirtualBox kernel driver not accessible, permission problem. "
345	"Re-install VirtualBox. If you are building it yourself, you "
346	"should make sure it installed correctly and that the setuid "
347	"bit is set on the executables calling VMR3Create.");
348	#else
349	/* This should only happen when mixing builds or with the usual /dev/vboxdrv access issues. */
350	# if defined(RT_OS_DARWIN)
351	pszError = N_("VirtualBox KEXT is not accessible, permission problem. "
352	"If you have built VirtualBox yourself, make sure that you do not "
353	"have the vboxdrv KEXT from a different build or installation loaded.");
354	# elif defined(RT_OS_LINUX)
355	pszError = N_("VirtualBox kernel driver is not accessible, permission problem. "
356	"If you have built VirtualBox yourself, make sure that you do "
357	"not have the vboxdrv kernel module from a different build or "
358	"installation loaded. Also, make sure the vboxdrv udev rule gives "
359	"you the permission you need to access the device.");
360	# elif defined(RT_OS_WINDOWS)
361	pszError = N_("VirtualBox kernel driver is not accessible, permission problem.");
362	# else /* solaris, freebsd, ++. */
363	pszError = N_("VirtualBox kernel module is not accessible, permission problem. "
364	"If you have built VirtualBox yourself, make sure that you do "
365	"not have the vboxdrv kernel module from a different install loaded.");
366	# endif
367	#endif
368	break;
369	case VERR_INVALID_HANDLE: /** @todo track down and fix this error. */
370	case VERR_VM_DRIVER_NOT_INSTALLED:
371	#ifdef RT_OS_LINUX
372	pszError = N_("VirtualBox kernel driver not installed. The vboxdrv kernel module "
373	"was either not loaded or /dev/vboxdrv was not created for some "
374	"reason. Re-setup the kernel module by executing "
375	"'/etc/init.d/vboxdrv setup' as root");
376	#else
377	pszError = N_("VirtualBox kernel driver not installed");
378	#endif
379	break;
380	case VERR_NO_MEMORY:
381	pszError = N_("VirtualBox support library out of memory");
382	break;
383	case VERR_VERSION_MISMATCH:
384	case VERR_VM_DRIVER_VERSION_MISMATCH:
385	pszError = N_("The VirtualBox support driver which is running is from a different "
386	"version of VirtualBox. You can correct this by stopping all "
387	"running instances of VirtualBox and reinstalling the software.");
388	break;
389	default:
390	pszError = N_("Unknown error initializing kernel driver");
391	AssertMsgFailed(("Add error message for rc=%d (%Rrc)\n", rc, rc));
392	}
393	vmR3SetErrorU(pUVM, rc, RT_SRC_POS, pszError, rc);
394	}
395	}
396
397	/* cleanup */
398	vmR3DestroyUVM(pUVM, 2000);
399	LogFlow(("VMR3Create: returns %Rrc\n", rc));
400	return rc;
401	}
402
403
404	/**
405	* Creates the UVM.
406	*
407	* This will not initialize the support library even if vmR3DestroyUVM
408	* will terminate that.
409	*
410	* @returns VBox status code.
411	* @param cCpus Number of virtual CPUs
412	* @param ppUVM Where to store the UVM pointer.
413	*/
414	static int vmR3CreateUVM(uint32_t cCpus, PUVM *ppUVM)
415	{
416	uint32_t i;
417
418	/*
419	* Create and initialize the UVM.
420	*/
421	PUVM pUVM = (PUVM)RTMemPageAllocZ(RT_OFFSETOF(UVM, aCpus[cCpus]));
422	AssertReturn(pUVM, VERR_NO_MEMORY);
423	pUVM->u32Magic = UVM_MAGIC;
424	pUVM->cCpus = cCpus;
425
426	AssertCompile(sizeof(pUVM->vm.s) <= sizeof(pUVM->vm.padding));
427
428	pUVM->vm.s.ppAtStateNext = &pUVM->vm.s.pAtState;
429	pUVM->vm.s.ppAtErrorNext = &pUVM->vm.s.pAtError;
430	pUVM->vm.s.ppAtRuntimeErrorNext = &pUVM->vm.s.pAtRuntimeError;
431
432	pUVM->vm.s.enmHaltMethod = VMHALTMETHOD_BOOTSTRAP;
433
434	/* Initialize the VMCPU array in the UVM. */
435	for (i = 0; i < cCpus; i++)
436	{
437	pUVM->aCpus[i].pUVM = pUVM;
438	pUVM->aCpus[i].idCpu = i;
439	}
440
441	/* Allocate a TLS entry to store the VMINTUSERPERVMCPU pointer. */
442	int rc = RTTlsAllocEx(&pUVM->vm.s.idxTLS, NULL);
443	AssertRC(rc);
444	if (RT_SUCCESS(rc))
445	{
446	/* Allocate a halt method event semaphore for each VCPU. */
447	for (i = 0; i < cCpus; i++)
448	pUVM->aCpus[i].vm.s.EventSemWait = NIL_RTSEMEVENT;
449	for (i = 0; i < cCpus; i++)
450	{
451	rc = RTSemEventCreate(&pUVM->aCpus[i].vm.s.EventSemWait);
452	if (RT_FAILURE(rc))
453	break;
454	}
455	if (RT_SUCCESS(rc))
456	{
457	rc = RTCritSectInit(&pUVM->vm.s.AtStateCritSect);
458	if (RT_SUCCESS(rc))
459	{
460	rc = RTCritSectInit(&pUVM->vm.s.AtErrorCritSect);
461	if (RT_SUCCESS(rc))
462	{
463	/*
464	* Init fundamental (sub-)components - STAM, MMR3Heap and PDMLdr.
465	*/
466	rc = STAMR3InitUVM(pUVM);
467	if (RT_SUCCESS(rc))
468	{
469	rc = MMR3InitUVM(pUVM);
470	if (RT_SUCCESS(rc))
471	{
472	rc = PDMR3InitUVM(pUVM);
473	if (RT_SUCCESS(rc))
474	{
475	/*
476	* Start the emulation threads for all VMCPUs.
477	*/
478	for (i = 0; i < cCpus; i++)
479	{
480	rc = RTThreadCreateF(&pUVM->aCpus[i].vm.s.ThreadEMT, vmR3EmulationThread, &pUVM->aCpus[i], _1M,
481	RTTHREADTYPE_EMULATION, RTTHREADFLAGS_WAITABLE,
482	cCpus > 1 ? "EMT-%u" : "EMT", i);
483	if (RT_FAILURE(rc))
484	break;
485
486	pUVM->aCpus[i].vm.s.NativeThreadEMT = RTThreadGetNative(pUVM->aCpus[i].vm.s.ThreadEMT);
487	}
488
489	if (RT_SUCCESS(rc))
490	{
491	*ppUVM = pUVM;
492	return VINF_SUCCESS;
493	}
494
495	/* bail out. */
496	while (i-- > 0)
497	{
498	/** @todo rainy day: terminate the EMTs. */
499	}
500	PDMR3TermUVM(pUVM);
501	}
502	MMR3TermUVM(pUVM);
503	}
504	STAMR3TermUVM(pUVM);
505	}
506	RTCritSectDelete(&pUVM->vm.s.AtErrorCritSect);
507	}
508	RTCritSectDelete(&pUVM->vm.s.AtStateCritSect);
509	}
510	}
511	for (i = 0; i < cCpus; i++)
512	{
513	RTSemEventDestroy(pUVM->aCpus[i].vm.s.EventSemWait);
514	pUVM->aCpus[i].vm.s.EventSemWait = NIL_RTSEMEVENT;
515	}
516	RTTlsFree(pUVM->vm.s.idxTLS);
517	}
518	RTMemPageFree(pUVM, sizeof(*pUVM));
519	return rc;
520	}
521
522
523	/**
524	* Creates and initializes the VM.
525	*
526	* @thread EMT
527	*/
528	static int vmR3CreateU(PUVM pUVM, uint32_t cCpus, PFNCFGMCONSTRUCTOR pfnCFGMConstructor, void *pvUserCFGM)
529	{
530	int rc = VINF_SUCCESS;
531
532	/*
533	* Load the VMMR0.r0 module so that we can call GVMMR0CreateVM.
534	*/
535	rc = PDMR3LdrLoadVMMR0U(pUVM);
536	if (RT_FAILURE(rc))
537	{
538	/** @todo we need a cleaner solution for this (VERR_VMX_IN_VMX_ROOT_MODE).
539	* bird: what about moving the message down here? Main picks the first message, right? */
540	if (rc == VERR_VMX_IN_VMX_ROOT_MODE)
541	return rc; /* proper error message set later on */
542	return vmR3SetErrorU(pUVM, rc, RT_SRC_POS, N_("Failed to load VMMR0.r0"));
543	}
544
545	/*
546	* Request GVMM to create a new VM for us.
547	*/
548	GVMMCREATEVMREQ CreateVMReq;
549	CreateVMReq.Hdr.u32Magic = SUPVMMR0REQHDR_MAGIC;
550	CreateVMReq.Hdr.cbReq = sizeof(CreateVMReq);
551	CreateVMReq.pSession = pUVM->vm.s.pSession;
552	CreateVMReq.pVMR0 = NIL_RTR0PTR;
553	CreateVMReq.pVMR3 = NULL;
554	CreateVMReq.cCpus = cCpus;
555	rc = SUPR3CallVMMR0Ex(NIL_RTR0PTR, NIL_VMCPUID, VMMR0_DO_GVMM_CREATE_VM, 0, &CreateVMReq.Hdr);
556	if (RT_SUCCESS(rc))
557	{
558	PVM pVM = pUVM->pVM = CreateVMReq.pVMR3;
559	AssertRelease(VALID_PTR(pVM));
560	AssertRelease(pVM->pVMR0 == CreateVMReq.pVMR0);
561	AssertRelease(pVM->pSession == pUVM->vm.s.pSession);
562	AssertRelease(pVM->cCpus == cCpus);
563	AssertRelease(pVM->offVMCPU == RT_UOFFSETOF(VM, aCpus));
564
565	Log(("VMR3Create: Created pUVM=%p pVM=%p pVMR0=%p hSelf=%#x cCpus=%RU32\n",
566	pUVM, pVM, pVM->pVMR0, pVM->hSelf, pVM->cCpus));
567
568	/*
569	* Initialize the VM structure and our internal data (VMINT).
570	*/
571	pVM->pUVM = pUVM;
572
573	for (VMCPUID i = 0; i < pVM->cCpus; i++)
574	{
575	pVM->aCpus[i].pUVCpu = &pUVM->aCpus[i];
576	pVM->aCpus[i].idCpu = i;
577	pVM->aCpus[i].hNativeThread = pUVM->aCpus[i].vm.s.NativeThreadEMT;
578	Assert(pVM->aCpus[i].hNativeThread != NIL_RTNATIVETHREAD);
579
580	pUVM->aCpus[i].pVCpu = &pVM->aCpus[i];
581	pUVM->aCpus[i].pVM = pVM;
582	}
583
584
585	/*
586	* Init the configuration.
587	*/
588	rc = CFGMR3Init(pVM, pfnCFGMConstructor, pvUserCFGM);
589	if (RT_SUCCESS(rc))
590	{
591	rc = CFGMR3QueryBoolDef(CFGMR3GetRoot(pVM), "HwVirtExtForced", &pVM->fHwVirtExtForced, false);
592	if (RT_SUCCESS(rc) && pVM->fHwVirtExtForced)
593	pVM->fHWACCMEnabled = true;
594
595	/*
596	* If executing in fake suplib mode disable RR3 and RR0 in the config.
597	*/
598	const char *psz = RTEnvGet("VBOX_SUPLIB_FAKE");
599	if (psz && !strcmp(psz, "fake"))
600	{
601	CFGMR3RemoveValue(CFGMR3GetRoot(pVM), "RawR3Enabled");
602	CFGMR3InsertInteger(CFGMR3GetRoot(pVM), "RawR3Enabled", 0);
603	CFGMR3RemoveValue(CFGMR3GetRoot(pVM), "RawR0Enabled");
604	CFGMR3InsertInteger(CFGMR3GetRoot(pVM), "RawR0Enabled", 0);
605	}
606
607	/*
608	* Make sure the CPU count in the config data matches.
609	*/
610	if (RT_SUCCESS(rc))
611	{
612	uint32_t cCPUsCfg;
613	rc = CFGMR3QueryU32Def(CFGMR3GetRoot(pVM), "NumCPUs", &cCPUsCfg, 1);
614	AssertLogRelMsgRC(rc, ("Configuration error: Querying \"NumCPUs\" as integer failed, rc=%Rrc\n", rc));
615	if (RT_SUCCESS(rc) && cCPUsCfg != cCpus)
616	{
617	AssertLogRelMsgFailed(("Configuration error: \"NumCPUs\"=%RU32 and VMR3CreateVM::cCpus=%RU32 does not match!\n",
618	cCPUsCfg, cCpus));
619	rc = VERR_INVALID_PARAMETER;
620	}
621	}
622	if (RT_SUCCESS(rc))
623	{
624	/*
625	* Init the ring-3 components and ring-3 per cpu data, finishing it off
626	* by a relocation round (intermediate context finalization will do this).
627	*/
628	rc = vmR3InitRing3(pVM, pUVM);
629	if (RT_SUCCESS(rc))
630	{
631	rc = vmR3InitVMCpu(pVM);
632	if (RT_SUCCESS(rc))
633	rc = PGMR3FinalizeMappings(pVM);
634	if (RT_SUCCESS(rc))
635	{
636
637	LogFlow(("Ring-3 init succeeded\n"));
638
639	/*
640	* Init the Ring-0 components.
641	*/
642	rc = vmR3InitRing0(pVM);
643	if (RT_SUCCESS(rc))
644	{
645	/* Relocate again, because some switcher fixups depends on R0 init results. */
646	VMR3Relocate(pVM, 0);
647
648	#ifdef VBOX_WITH_DEBUGGER
649	/*
650	* Init the tcp debugger console if we're building
651	* with debugger support.
652	*/
653	void *pvUser = NULL;
654	rc = DBGCTcpCreate(pVM, &pvUser);
655	if ( RT_SUCCESS(rc)
656	\|\| rc == VERR_NET_ADDRESS_IN_USE)
657	{
658	pUVM->vm.s.pvDBGC = pvUser;
659	#endif
660	/*
661	* Init the Guest Context components.
662	*/
663	rc = vmR3InitGC(pVM);
664	if (RT_SUCCESS(rc))
665	{
666	/*
667	* Now we can safely set the VM halt method to default.
668	*/
669	rc = vmR3SetHaltMethodU(pUVM, VMHALTMETHOD_DEFAULT);
670	if (RT_SUCCESS(rc))
671	{
672	/*
673	* Set the state and link into the global list.
674	*/
675	vmR3SetState(pVM, VMSTATE_CREATED, VMSTATE_CREATING);
676	pUVM->pNext = g_pUVMsHead;
677	g_pUVMsHead = pUVM;
678
679	#ifdef LOG_ENABLED
680	RTLogSetCustomPrefixCallback(NULL, vmR3LogPrefixCallback, pUVM);
681	#endif
682	return VINF_SUCCESS;
683	}
684	}
685	#ifdef VBOX_WITH_DEBUGGER
686	DBGCTcpTerminate(pVM, pUVM->vm.s.pvDBGC);
687	pUVM->vm.s.pvDBGC = NULL;
688	}
689	#endif
690	//..
691	}
692	}
693	vmR3Destroy(pVM);
694	}
695	}
696	//..
697
698	/* Clean CFGM. */
699	int rc2 = CFGMR3Term(pVM);
700	AssertRC(rc2);
701	}
702
703	/*
704	* Do automatic cleanups while the VM structure is still alive and all
705	* references to it are still working.
706	*/
707	PDMR3CritSectTerm(pVM);
708
709	/*
710	* Drop all references to VM and the VMCPU structures, then
711	* tell GVMM to destroy the VM.
712	*/
713	pUVM->pVM = NULL;
714	for (VMCPUID i = 0; i < pUVM->cCpus; i++)
715	{
716	pUVM->aCpus[i].pVM = NULL;
717	pUVM->aCpus[i].pVCpu = NULL;
718	}
719	Assert(pUVM->vm.s.enmHaltMethod == VMHALTMETHOD_BOOTSTRAP);
720
721	if (pUVM->cCpus > 1)
722	{
723	/* Poke the other EMTs since they may have stale pVM and pVCpu references
724	on the stack (see VMR3WaitU for instance) if they've been awakened after
725	VM creation. */
726	for (VMCPUID i = 1; i < pUVM->cCpus; i++)
727	VMR3NotifyCpuFFU(&pUVM->aCpus[i], 0);
728	RTThreadSleep(RT_MIN(100 + 25 (pUVM->cCpus - 1), 500)); / very sophisticated */
729	}
730
731	int rc2 = SUPR3CallVMMR0Ex(CreateVMReq.pVMR0, 0 /idCpu/, VMMR0_DO_GVMM_DESTROY_VM, 0, NULL);
732	AssertRC(rc2);
733	}
734	else
735	vmR3SetErrorU(pUVM, rc, RT_SRC_POS, N_("VM creation failed (GVMM)"));
736
737	LogFlow(("vmR3CreateU: returns %Rrc\n", rc));
738	return rc;
739	}
740
741
742	/**
743	* Register the calling EMT with GVM.
744	*
745	* @returns VBox status code.
746	* @param pVM The VM handle.
747	* @param idCpu The Virtual CPU ID.
748	*/
749	static DECLCALLBACK(int) vmR3RegisterEMT(PVM pVM, VMCPUID idCpu)
750	{
751	Assert(VMMGetCpuId(pVM) == idCpu);
752	int rc = SUPR3CallVMMR0Ex(pVM->pVMR0, idCpu, VMMR0_DO_GVMM_REGISTER_VMCPU, 0, NULL);
753	if (RT_FAILURE(rc))
754	LogRel(("idCpu=%u rc=%Rrc\n", idCpu, rc));
755	return rc;
756	}
757
758
759	/**
760	* Initializes all R3 components of the VM
761	*/
762	static int vmR3InitRing3(PVM pVM, PUVM pUVM)
763	{
764	int rc;
765
766	/*
767	* Register the other EMTs with GVM.
768	*/
769	for (VMCPUID idCpu = 1; idCpu < pVM->cCpus; idCpu++)
770	{
771	rc = VMR3ReqCallWaitU(pUVM, idCpu, (PFNRT)vmR3RegisterEMT, 2, pVM, idCpu);
772	if (RT_FAILURE(rc))
773	return rc;
774	}
775
776	/*
777	* Init all R3 components, the order here might be important.
778	*/
779	rc = MMR3Init(pVM);
780	if (RT_SUCCESS(rc))
781	{
782	STAM_REG(pVM, &pVM->StatTotalInGC, STAMTYPE_PROFILE_ADV, "/PROF/VM/InGC", STAMUNIT_TICKS_PER_CALL, "Profiling the total time spent in GC.");
783	STAM_REG(pVM, &pVM->StatSwitcherToGC, STAMTYPE_PROFILE_ADV, "/PROF/VM/SwitchToGC", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
784	STAM_REG(pVM, &pVM->StatSwitcherToHC, STAMTYPE_PROFILE_ADV, "/PROF/VM/SwitchToHC", STAMUNIT_TICKS_PER_CALL, "Profiling switching to HC.");
785	STAM_REG(pVM, &pVM->StatSwitcherSaveRegs, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/SaveRegs", STAMUNIT_TICKS_PER_CALL,"Profiling switching to GC.");
786	STAM_REG(pVM, &pVM->StatSwitcherSysEnter, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/SysEnter", STAMUNIT_TICKS_PER_CALL,"Profiling switching to GC.");
787	STAM_REG(pVM, &pVM->StatSwitcherDebug, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/Debug", STAMUNIT_TICKS_PER_CALL,"Profiling switching to GC.");
788	STAM_REG(pVM, &pVM->StatSwitcherCR0, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/CR0", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
789	STAM_REG(pVM, &pVM->StatSwitcherCR4, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/CR4", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
790	STAM_REG(pVM, &pVM->StatSwitcherLgdt, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/Lgdt", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
791	STAM_REG(pVM, &pVM->StatSwitcherLidt, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/Lidt", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
792	STAM_REG(pVM, &pVM->StatSwitcherLldt, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/Lldt", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
793	STAM_REG(pVM, &pVM->StatSwitcherTSS, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/TSS", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
794	STAM_REG(pVM, &pVM->StatSwitcherJmpCR3, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/JmpCR3", STAMUNIT_TICKS_PER_CALL,"Profiling switching to GC.");
795	STAM_REG(pVM, &pVM->StatSwitcherRstrRegs, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/RstrRegs", STAMUNIT_TICKS_PER_CALL,"Profiling switching to GC.");
796
797	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
798	{
799	rc = STAMR3RegisterF(pVM, &pUVM->aCpus[idCpu].vm.s.StatHaltYield, STAMTYPE_PROFILE, STAMVISIBILITY_ALWAYS, STAMUNIT_TICKS_PER_CALL, "Profiling halted state yielding.", "/PROF/VM/CPU%d/Halt/Yield", idCpu);
800	AssertRC(rc);
801	rc = STAMR3RegisterF(pVM, &pUVM->aCpus[idCpu].vm.s.StatHaltBlock, STAMTYPE_PROFILE, STAMVISIBILITY_ALWAYS, STAMUNIT_TICKS_PER_CALL, "Profiling halted state blocking.", "/PROF/VM/CPU%d/Halt/Block", idCpu);
802	AssertRC(rc);
803	rc = STAMR3RegisterF(pVM, &pUVM->aCpus[idCpu].vm.s.StatHaltTimers, STAMTYPE_PROFILE, STAMVISIBILITY_ALWAYS, STAMUNIT_TICKS_PER_CALL, "Profiling halted state timer tasks.", "/PROF/VM/CPU%d/Halt/Timers", idCpu);
804	AssertRC(rc);
805	}
806
807	STAM_REG(pVM, &pUVM->vm.s.StatReqAllocNew, STAMTYPE_COUNTER, "/VM/Req/AllocNew", STAMUNIT_OCCURENCES, "Number of VMR3ReqAlloc returning a new packet.");
808	STAM_REG(pVM, &pUVM->vm.s.StatReqAllocRaces, STAMTYPE_COUNTER, "/VM/Req/AllocRaces", STAMUNIT_OCCURENCES, "Number of VMR3ReqAlloc causing races.");
809	STAM_REG(pVM, &pUVM->vm.s.StatReqAllocRecycled, STAMTYPE_COUNTER, "/VM/Req/AllocRecycled", STAMUNIT_OCCURENCES, "Number of VMR3ReqAlloc returning a recycled packet.");
810	STAM_REG(pVM, &pUVM->vm.s.StatReqFree, STAMTYPE_COUNTER, "/VM/Req/Free", STAMUNIT_OCCURENCES, "Number of VMR3ReqFree calls.");
811	STAM_REG(pVM, &pUVM->vm.s.StatReqFreeOverflow, STAMTYPE_COUNTER, "/VM/Req/FreeOverflow", STAMUNIT_OCCURENCES, "Number of times the request was actually freed.");
812	STAM_REG(pVM, &pUVM->vm.s.StatReqProcessed, STAMTYPE_COUNTER, "/VM/Req/Processed", STAMUNIT_OCCURENCES, "Number of processed requests (any queue).");
813	STAM_REG(pVM, &pUVM->vm.s.StatReqMoreThan1, STAMTYPE_COUNTER, "/VM/Req/MoreThan1", STAMUNIT_OCCURENCES, "Number of times there are more than one request on the queue when processing it.");
814	STAM_REG(pVM, &pUVM->vm.s.StatReqPushBackRaces, STAMTYPE_COUNTER, "/VM/Req/PushBackRaces", STAMUNIT_OCCURENCES, "Number of push back races.");
815
816	rc = CPUMR3Init(pVM);
817	if (RT_SUCCESS(rc))
818	{
819	rc = HWACCMR3Init(pVM);
820	if (RT_SUCCESS(rc))
821	{
822	rc = PGMR3Init(pVM);
823	if (RT_SUCCESS(rc))
824	{
825	rc = REMR3Init(pVM);
826	if (RT_SUCCESS(rc))
827	{
828	rc = MMR3InitPaging(pVM);
829	if (RT_SUCCESS(rc))
830	rc = TMR3Init(pVM);
831	if (RT_SUCCESS(rc))
832	{
833	rc = VMMR3Init(pVM);
834	if (RT_SUCCESS(rc))
835	{
836	rc = SELMR3Init(pVM);
837	if (RT_SUCCESS(rc))
838	{
839	rc = TRPMR3Init(pVM);
840	if (RT_SUCCESS(rc))
841	{
842	rc = CSAMR3Init(pVM);
843	if (RT_SUCCESS(rc))
844	{
845	rc = PATMR3Init(pVM);
846	if (RT_SUCCESS(rc))
847	{
848	rc = IOMR3Init(pVM);
849	if (RT_SUCCESS(rc))
850	{
851	rc = EMR3Init(pVM);
852	if (RT_SUCCESS(rc))
853	{
854	rc = DBGFR3Init(pVM);
855	if (RT_SUCCESS(rc))
856	{
857	rc = PDMR3Init(pVM);
858	if (RT_SUCCESS(rc))
859	{
860	rc = PGMR3InitDynMap(pVM);
861	if (RT_SUCCESS(rc))
862	rc = MMR3HyperInitFinalize(pVM);
863	if (RT_SUCCESS(rc))
864	rc = PATMR3InitFinalize(pVM);
865	if (RT_SUCCESS(rc))
866	rc = PGMR3InitFinalize(pVM);
867	if (RT_SUCCESS(rc))
868	rc = SELMR3InitFinalize(pVM);
869	if (RT_SUCCESS(rc))
870	rc = TMR3InitFinalize(pVM);
871	if (RT_SUCCESS(rc))
872	rc = VMMR3InitFinalize(pVM);
873	if (RT_SUCCESS(rc))
874	rc = REMR3InitFinalize(pVM);
875	if (RT_SUCCESS(rc))
876	rc = vmR3InitDoCompleted(pVM, VMINITCOMPLETED_RING3);
877	if (RT_SUCCESS(rc))
878	{
879	LogFlow(("vmR3InitRing3: returns %Rrc\n", VINF_SUCCESS));
880	return VINF_SUCCESS;
881	}
882	int rc2 = PDMR3Term(pVM);
883	AssertRC(rc2);
884	}
885	int rc2 = DBGFR3Term(pVM);
886	AssertRC(rc2);
887	}
888	int rc2 = EMR3Term(pVM);
889	AssertRC(rc2);
890	}
891	int rc2 = IOMR3Term(pVM);
892	AssertRC(rc2);
893	}
894	int rc2 = PATMR3Term(pVM);
895	AssertRC(rc2);
896	}
897	int rc2 = CSAMR3Term(pVM);
898	AssertRC(rc2);
899	}
900	int rc2 = TRPMR3Term(pVM);
901	AssertRC(rc2);
902	}
903	int rc2 = SELMR3Term(pVM);
904	AssertRC(rc2);
905	}
906	int rc2 = VMMR3Term(pVM);
907	AssertRC(rc2);
908	}
909	int rc2 = TMR3Term(pVM);
910	AssertRC(rc2);
911	}
912	int rc2 = REMR3Term(pVM);
913	AssertRC(rc2);
914	}
915	int rc2 = PGMR3Term(pVM);
916	AssertRC(rc2);
917	}
918	int rc2 = HWACCMR3Term(pVM);
919	AssertRC(rc2);
920	}
921	//int rc2 = CPUMR3Term(pVM);
922	//AssertRC(rc2);
923	}
924	/* MMR3Term is not called here because it'll kill the heap. */
925	}
926
927	LogFlow(("vmR3InitRing3: returns %Rrc\n", rc));
928	return rc;
929	}
930
931
932	/**
933	* Initializes all VM CPU components of the VM
934	*/
935	static int vmR3InitVMCpu(PVM pVM)
936	{
937	int rc = VINF_SUCCESS;
938	int rc2;
939
940	rc = CPUMR3InitCPU(pVM);
941	if (RT_SUCCESS(rc))
942	{
943	rc = HWACCMR3InitCPU(pVM);
944	if (RT_SUCCESS(rc))
945	{
946	rc = PGMR3InitCPU(pVM);
947	if (RT_SUCCESS(rc))
948	{
949	rc = TMR3InitCPU(pVM);
950	if (RT_SUCCESS(rc))
951	{
952	rc = VMMR3InitCPU(pVM);
953	if (RT_SUCCESS(rc))
954	{
955	rc = EMR3InitCPU(pVM);
956	if (RT_SUCCESS(rc))
957	{
958	LogFlow(("vmR3InitVMCpu: returns %Rrc\n", VINF_SUCCESS));
959	return VINF_SUCCESS;
960	}
961
962	rc2 = VMMR3TermCPU(pVM);
963	AssertRC(rc2);
964	}
965	rc2 = TMR3TermCPU(pVM);
966	AssertRC(rc2);
967	}
968	rc2 = PGMR3TermCPU(pVM);
969	AssertRC(rc2);
970	}
971	rc2 = HWACCMR3TermCPU(pVM);
972	AssertRC(rc2);
973	}
974	rc2 = CPUMR3TermCPU(pVM);
975	AssertRC(rc2);
976	}
977	LogFlow(("vmR3InitVMCpu: returns %Rrc\n", rc));
978	return rc;
979	}
980
981
982	/**
983	* Initializes all R0 components of the VM
984	*/
985	static int vmR3InitRing0(PVM pVM)
986	{
987	LogFlow(("vmR3InitRing0:\n"));
988
989	/*
990	* Check for FAKE suplib mode.
991	*/
992	int rc = VINF_SUCCESS;
993	const char *psz = RTEnvGet("VBOX_SUPLIB_FAKE");
994	if (!psz \|\| strcmp(psz, "fake"))
995	{
996	/*
997	* Call the VMMR0 component and let it do the init.
998	*/
999	rc = VMMR3InitR0(pVM);
1000	}
1001	else
1002	Log(("vmR3InitRing0: skipping because of VBOX_SUPLIB_FAKE=fake\n"));
1003
1004	/*
1005	* Do notifications and return.
1006	*/
1007	if (RT_SUCCESS(rc))
1008	rc = vmR3InitDoCompleted(pVM, VMINITCOMPLETED_RING0);
1009
1010	/** @todo Move this to the VMINITCOMPLETED_RING0 notification handler. */
1011	if (RT_SUCCESS(rc))
1012	{
1013	rc = HWACCMR3InitFinalizeR0(pVM);
1014	CPUMR3SetHWVirtEx(pVM, HWACCMIsEnabled(pVM));
1015	}
1016
1017	LogFlow(("vmR3InitRing0: returns %Rrc\n", rc));
1018	return rc;
1019	}
1020
1021
1022	/**
1023	* Initializes all GC components of the VM
1024	*/
1025	static int vmR3InitGC(PVM pVM)
1026	{
1027	LogFlow(("vmR3InitGC:\n"));
1028
1029	/*
1030	* Check for FAKE suplib mode.
1031	*/
1032	int rc = VINF_SUCCESS;
1033	const char *psz = RTEnvGet("VBOX_SUPLIB_FAKE");
1034	if (!psz \|\| strcmp(psz, "fake"))
1035	{
1036	/*
1037	* Call the VMMR0 component and let it do the init.
1038	*/
1039	rc = VMMR3InitRC(pVM);
1040	}
1041	else
1042	Log(("vmR3InitGC: skipping because of VBOX_SUPLIB_FAKE=fake\n"));
1043
1044	/*
1045	* Do notifications and return.
1046	*/
1047	if (RT_SUCCESS(rc))
1048	rc = vmR3InitDoCompleted(pVM, VMINITCOMPLETED_GC);
1049	LogFlow(("vmR3InitGC: returns %Rrc\n", rc));
1050	return rc;
1051	}
1052
1053
1054	/**
1055	* Do init completed notifications.
1056	* This notifications can fail.
1057	*
1058	* @param pVM The VM handle.
1059	* @param enmWhat What's completed.
1060	*/
1061	static int vmR3InitDoCompleted(PVM pVM, VMINITCOMPLETED enmWhat)
1062	{
1063	return VINF_SUCCESS;
1064	}
1065
1066
1067	/**
1068	* Logger callback for inserting a custom prefix.
1069	*
1070	* @returns Number of chars written.
1071	* @param pLogger The logger.
1072	* @param pchBuf The output buffer.
1073	* @param cchBuf The output buffer size.
1074	* @param pvUser Pointer to the UVM structure.
1075	*/
1076	static DECLCALLBACK(size_t) vmR3LogPrefixCallback(PRTLOGGER pLogger, char pchBuf, size_t cchBuf, void pvUser)
1077	{
1078	AssertReturn(cchBuf >= 2, 0);
1079	PUVM pUVM = (PUVM)pvUser;
1080	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pUVM->vm.s.idxTLS);
1081	if (pUVCpu)
1082	{
1083	static const char s_szHex[17] = "0123456789abcdef";
1084	VMCPUID const idCpu = pUVCpu->idCpu;
1085	pchBuf[1] = s_szHex[ idCpu & 15];
1086	pchBuf[0] = s_szHex[(idCpu >> 4) & 15];
1087	}
1088	else
1089	{
1090	pchBuf[0] = 'x';
1091	pchBuf[1] = 'y';
1092	}
1093
1094	return 2;
1095	}
1096
1097
1098	/**
1099	* Calls the relocation functions for all VMM components so they can update
1100	* any GC pointers. When this function is called all the basic VM members
1101	* have been updated and the actual memory relocation have been done
1102	* by the PGM/MM.
1103	*
1104	* This is used both on init and on runtime relocations.
1105	*
1106	* @param pVM VM handle.
1107	* @param offDelta Relocation delta relative to old location.
1108	*/
1109	VMMR3DECL(void) VMR3Relocate(PVM pVM, RTGCINTPTR offDelta)
1110	{
1111	LogFlow(("VMR3Relocate: offDelta=%RGv\n", offDelta));
1112
1113	/*
1114	* The order here is very important!
1115	*/
1116	PGMR3Relocate(pVM, offDelta);
1117	PDMR3LdrRelocateU(pVM->pUVM, offDelta);
1118	PGMR3Relocate(pVM, 0); /* Repeat after PDM relocation. */
1119	CPUMR3Relocate(pVM);
1120	HWACCMR3Relocate(pVM);
1121	SELMR3Relocate(pVM);
1122	VMMR3Relocate(pVM, offDelta);
1123	SELMR3Relocate(pVM); /* !hack! fix stack! */
1124	TRPMR3Relocate(pVM, offDelta);
1125	PATMR3Relocate(pVM);
1126	CSAMR3Relocate(pVM, offDelta);
1127	IOMR3Relocate(pVM, offDelta);
1128	EMR3Relocate(pVM);
1129	TMR3Relocate(pVM, offDelta);
1130	DBGFR3Relocate(pVM, offDelta);
1131	PDMR3Relocate(pVM, offDelta);
1132	}
1133
1134
1135	/**
1136	* EMT rendezvous worker for VMR3PowerOn.
1137	*
1138	* @returns VERR_VM_INVALID_VM_STATE or VINF_SUCCESS. (This is a strict return
1139	* code, see FNVMMEMTRENDEZVOUS.)
1140	*
1141	* @param pVM The VM handle.
1142	* @param pVCpu The VMCPU handle of the EMT.
1143	* @param pvUser Ignored.
1144	*/
1145	static DECLCALLBACK(VBOXSTRICTRC) vmR3PowerOn(PVM pVM, PVMCPU pVCpu, void *pvUser)
1146	{
1147	LogFlow(("vmR3PowerOn: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1148	Assert(!pvUser); NOREF(pvUser);
1149
1150	/*
1151	* The first thread thru here tries to change the state. We shouldn't be
1152	* called again if this fails.
1153	*/
1154	if (pVCpu->idCpu == pVM->cCpus - 1)
1155	{
1156	int rc = vmR3TrySetState(pVM, "VMR3PowerOn", 1, VMSTATE_POWERING_ON, VMSTATE_CREATED);
1157	if (RT_FAILURE(rc))
1158	return rc;
1159	}
1160
1161	VMSTATE enmVMState = VMR3GetState(pVM);
1162	AssertMsgReturn(enmVMState == VMSTATE_POWERING_ON,
1163	("%s\n", VMR3GetStateName(enmVMState)),
1164	VERR_INTERNAL_ERROR_4);
1165
1166	/*
1167	* All EMTs changes their state to started.
1168	*/
1169	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED);
1170
1171	/*
1172	* EMT(0) is last thru here and it will make the notification calls
1173	* and advance the state.
1174	*/
1175	if (pVCpu->idCpu == 0)
1176	{
1177	PDMR3PowerOn(pVM);
1178	vmR3SetState(pVM, VMSTATE_RUNNING, VMSTATE_POWERING_ON);
1179	}
1180
1181	return VINF_SUCCESS;
1182	}
1183
1184
1185	/**
1186	* Powers on the virtual machine.
1187	*
1188	* @returns VBox status code.
1189	*
1190	* @param pVM The VM to power on.
1191	*
1192	* @thread Any thread.
1193	* @vmstate Created
1194	* @vmstateto PoweringOn+Running
1195	*/
1196	VMMR3DECL(int) VMR3PowerOn(PVM pVM)
1197	{
1198	LogFlow(("VMR3PowerOn: pVM=%p\n", pVM));
1199	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1200
1201	/*
1202	* Gather all the EMTs to reduce the init TSC drift and keep
1203	* the state changing APIs a bit uniform.
1204	*/
1205	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
1206	vmR3PowerOn, NULL);
1207	LogFlow(("VMR3PowerOn: returns %Rrc\n", rc));
1208	return rc;
1209	}
1210
1211
1212	/**
1213	* Does the suspend notifications.
1214	*
1215	* @param pVM The VM handle.
1216	* @thread EMT(0)
1217	*/
1218	static void vmR3SuspendDoWork(PVM pVM)
1219	{
1220	PDMR3Suspend(pVM);
1221	}
1222
1223
1224	/**
1225	* EMT rendezvous worker for VMR3Suspend.
1226	*
1227	* @returns VERR_VM_INVALID_VM_STATE or VINF_EM_SUSPEND. (This is a strict
1228	* return code, see FNVMMEMTRENDEZVOUS.)
1229	*
1230	* @param pVM The VM handle.
1231	* @param pVCpu The VMCPU handle of the EMT.
1232	* @param pvUser Ignored.
1233	*/
1234	static DECLCALLBACK(VBOXSTRICTRC) vmR3Suspend(PVM pVM, PVMCPU pVCpu, void *pvUser)
1235	{
1236	LogFlow(("vmR3Suspend: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1237	Assert(!pvUser); NOREF(pvUser);
1238
1239	/*
1240	* The first EMT switches the state to suspending. If this fails because
1241	* something was racing us in one way or the other, there will be no more
1242	* calls and thus the state assertion below is not going to annoy anyone.
1243	*/
1244	if (pVCpu->idCpu == pVM->cCpus - 1)
1245	{
1246	int rc = vmR3TrySetState(pVM, "VMR3Suspend", 2,
1247	VMSTATE_SUSPENDING, VMSTATE_RUNNING,
1248	VMSTATE_SUSPENDING_EXT_LS, VMSTATE_RUNNING_LS);
1249	if (RT_FAILURE(rc))
1250	return rc;
1251	}
1252
1253	VMSTATE enmVMState = VMR3GetState(pVM);
1254	AssertMsgReturn( enmVMState == VMSTATE_SUSPENDING
1255	\|\| enmVMState == VMSTATE_SUSPENDING_EXT_LS,
1256	("%s\n", VMR3GetStateName(enmVMState)),
1257	VERR_INTERNAL_ERROR_4);
1258
1259	/*
1260	* EMT(0) does the actually suspending after all the other CPUs have
1261	* been thru here.
1262	*/
1263	if (pVCpu->idCpu == 0)
1264	{
1265	vmR3SuspendDoWork(pVM);
1266
1267	int rc = vmR3TrySetState(pVM, "VMR3Suspend", 2,
1268	VMSTATE_SUSPENDED, VMSTATE_SUSPENDING,
1269	VMSTATE_SUSPENDED_EXT_LS, VMSTATE_SUSPENDING_EXT_LS);
1270	if (RT_FAILURE(rc))
1271	return VERR_INTERNAL_ERROR_3;
1272	}
1273
1274	return VINF_EM_SUSPEND;
1275	}
1276
1277
1278	/**
1279	* Suspends a running VM.
1280	*
1281	* @returns VBox status code. When called on EMT, this will be a strict status
1282	* code that has to be propagated up the call stack.
1283	*
1284	* @param pVM The VM to suspend.
1285	*
1286	* @thread Any thread.
1287	* @vmstate Running or RunningLS
1288	* @vmstateto Suspending + Suspended or SuspendingExtLS + SuspendedExtLS
1289	*/
1290	VMMR3DECL(int) VMR3Suspend(PVM pVM)
1291	{
1292	LogFlow(("VMR3Suspend: pVM=%p\n", pVM));
1293	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1294
1295	/*
1296	* Gather all the EMTs to make sure there are no races before
1297	* changing the VM state.
1298	*/
1299	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
1300	vmR3Suspend, NULL);
1301	LogFlow(("VMR3Suspend: returns %Rrc\n", rc));
1302	return rc;
1303	}
1304
1305
1306	/**
1307	* EMT rendezvous worker for VMR3Resume.
1308	*
1309	* @returns VERR_VM_INVALID_VM_STATE or VINF_EM_RESUME. (This is a strict
1310	* return code, see FNVMMEMTRENDEZVOUS.)
1311	*
1312	* @param pVM The VM handle.
1313	* @param pVCpu The VMCPU handle of the EMT.
1314	* @param pvUser Ignored.
1315	*/
1316	static DECLCALLBACK(VBOXSTRICTRC) vmR3Resume(PVM pVM, PVMCPU pVCpu, void *pvUser)
1317	{
1318	LogFlow(("vmR3Resume: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1319	Assert(!pvUser); NOREF(pvUser);
1320
1321	/*
1322	* The first thread thru here tries to change the state. We shouldn't be
1323	* called again if this fails.
1324	*/
1325	if (pVCpu->idCpu == pVM->cCpus - 1)
1326	{
1327	int rc = vmR3TrySetState(pVM, "VMR3Resume", 1, VMSTATE_RESUMING, VMSTATE_SUSPENDED);
1328	if (RT_FAILURE(rc))
1329	return rc;
1330	}
1331
1332	VMSTATE enmVMState = VMR3GetState(pVM);
1333	AssertMsgReturn(enmVMState == VMSTATE_RESUMING,
1334	("%s\n", VMR3GetStateName(enmVMState)),
1335	VERR_INTERNAL_ERROR_4);
1336
1337	#if 0
1338	/*
1339	* All EMTs changes their state to started.
1340	*/
1341	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED);
1342	#endif
1343
1344	/*
1345	* EMT(0) is last thru here and it will make the notification calls
1346	* and advance the state.
1347	*/
1348	if (pVCpu->idCpu == 0)
1349	{
1350	PDMR3Resume(pVM);
1351	vmR3SetState(pVM, VMSTATE_RUNNING, VMSTATE_RESUMING);
1352	pVM->vm.s.fTeleportedAndNotFullyResumedYet = false;
1353	}
1354
1355	return VINF_EM_RESUME;
1356	}
1357
1358
1359	/**
1360	* Resume VM execution.
1361	*
1362	* @returns VBox status code. When called on EMT, this will be a strict status
1363	* code that has to be propagated up the call stack.
1364	*
1365	* @param pVM The VM to resume.
1366	*
1367	* @thread Any thread.
1368	* @vmstate Suspended
1369	* @vmstateto Running
1370	*/
1371	VMMR3DECL(int) VMR3Resume(PVM pVM)
1372	{
1373	LogFlow(("VMR3Resume: pVM=%p\n", pVM));
1374	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1375
1376	/*
1377	* Gather all the EMTs to make sure there are no races before
1378	* changing the VM state.
1379	*/
1380	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
1381	vmR3Resume, NULL);
1382	LogFlow(("VMR3Resume: returns %Rrc\n", rc));
1383	return rc;
1384	}
1385
1386
1387	/**
1388	* EMT rendezvous worker for VMR3Save and VMR3Teleport that suspends the VM
1389	* after the live step has been completed.
1390	*
1391	* @returns VERR_VM_INVALID_VM_STATE or VINF_EM_RESUME. (This is a strict
1392	* return code, see FNVMMEMTRENDEZVOUS.)
1393	*
1394	* @param pVM The VM handle.
1395	* @param pVCpu The VMCPU handle of the EMT.
1396	* @param pvUser The pfSuspended argument of vmR3SaveTeleport.
1397	*/
1398	static DECLCALLBACK(VBOXSTRICTRC) vmR3LiveDoSuspend(PVM pVM, PVMCPU pVCpu, void *pvUser)
1399	{
1400	LogFlow(("vmR3LiveDoSuspend: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1401	bool pfSuspended = (bool )pvUser;
1402
1403	/*
1404	* The first thread thru here tries to change the state. We shouldn't be
1405	* called again if this fails.
1406	*/
1407	if (pVCpu->idCpu == pVM->cCpus - 1U)
1408	{
1409	PUVM pUVM = pVM->pUVM;
1410	int rc;
1411
1412	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
1413	VMSTATE enmVMState = pVM->enmVMState;
1414	switch (enmVMState)
1415	{
1416	case VMSTATE_RUNNING_LS:
1417	vmR3SetStateLocked(pVM, pUVM, VMSTATE_SUSPENDING_LS, VMSTATE_RUNNING_LS);
1418	rc = VINF_SUCCESS;
1419	break;
1420
1421	case VMSTATE_SUSPENDED_EXT_LS:
1422	case VMSTATE_SUSPENDED_LS: /* (via reset) */
1423	rc = VINF_SUCCESS;
1424	break;
1425
1426	case VMSTATE_DEBUGGING_LS:
1427	rc = VERR_TRY_AGAIN;
1428	break;
1429
1430	case VMSTATE_OFF_LS:
1431	vmR3SetStateLocked(pVM, pUVM, VMSTATE_OFF, VMSTATE_OFF_LS);
1432	rc = VERR_SSM_LIVE_POWERED_OFF;
1433	break;
1434
1435	case VMSTATE_FATAL_ERROR_LS:
1436	vmR3SetStateLocked(pVM, pUVM, VMSTATE_FATAL_ERROR, VMSTATE_FATAL_ERROR_LS);
1437	rc = VERR_SSM_LIVE_FATAL_ERROR;
1438	break;
1439
1440	case VMSTATE_GURU_MEDITATION_LS:
1441	vmR3SetStateLocked(pVM, pUVM, VMSTATE_GURU_MEDITATION, VMSTATE_GURU_MEDITATION_LS);
1442	rc = VERR_SSM_LIVE_GURU_MEDITATION;
1443	break;
1444
1445	case VMSTATE_POWERING_OFF_LS:
1446	case VMSTATE_SUSPENDING_EXT_LS:
1447	case VMSTATE_RESETTING_LS:
1448	default:
1449	AssertMsgFailed(("%s\n", VMR3GetStateName(enmVMState)));
1450	rc = VERR_INTERNAL_ERROR_3;
1451	break;
1452	}
1453	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
1454	if (RT_FAILURE(rc))
1455	{
1456	LogFlow(("vmR3LiveDoSuspend: returns %Rrc (state was %s)\n", rc, VMR3GetStateName(enmVMState)));
1457	return rc;
1458	}
1459	}
1460
1461	VMSTATE enmVMState = VMR3GetState(pVM);
1462	AssertMsgReturn(enmVMState == VMSTATE_SUSPENDING_LS,
1463	("%s\n", VMR3GetStateName(enmVMState)),
1464	VERR_INTERNAL_ERROR_4);
1465
1466	/*
1467	* Only EMT(0) have work to do since it's last thru here.
1468	*/
1469	if (pVCpu->idCpu == 0)
1470	{
1471	vmR3SuspendDoWork(pVM);
1472	int rc = vmR3TrySetState(pVM, "VMR3Suspend", 1,
1473	VMSTATE_SUSPENDED_LS, VMSTATE_SUSPENDING_LS);
1474	if (RT_FAILURE(rc))
1475	return VERR_INTERNAL_ERROR_3;
1476
1477	*pfSuspended = true;
1478	}
1479
1480	return VINF_EM_SUSPEND;
1481	}
1482
1483
1484	/**
1485	* EMT rendezvous worker that VMR3Save and VMR3Teleport uses to clean up a
1486	* SSMR3LiveDoStep1 failure.
1487	*
1488	* Doing this as a rendezvous operation avoids all annoying transition
1489	* states.
1490	*
1491	* @returns VERR_VM_INVALID_VM_STATE, VINF_SUCCESS or some specific VERR_SSM_*
1492	* status code. (This is a strict return code, see FNVMMEMTRENDEZVOUS.)
1493	*
1494	* @param pVM The VM handle.
1495	* @param pVCpu The VMCPU handle of the EMT.
1496	* @param pvUser The pfSuspended argument of vmR3SaveTeleport.
1497	*/
1498	static DECLCALLBACK(VBOXSTRICTRC) vmR3LiveDoStep1Cleanup(PVM pVM, PVMCPU pVCpu, void *pvUser)
1499	{
1500	LogFlow(("vmR3LiveDoStep1Cleanup: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1501	bool pfSuspended = (bool )pvUser;
1502	NOREF(pVCpu);
1503
1504	int rc = vmR3TrySetState(pVM, "vmR3LiveDoStep1Cleanup", 8,
1505	VMSTATE_OFF, VMSTATE_OFF_LS, /* 1 */
1506	VMSTATE_FATAL_ERROR, VMSTATE_FATAL_ERROR_LS, /* 2 */
1507	VMSTATE_GURU_MEDITATION, VMSTATE_GURU_MEDITATION_LS, /* 3 */
1508	VMSTATE_SUSPENDED, VMSTATE_SUSPENDED_LS, /* 4 */
1509	VMSTATE_SUSPENDED, VMSTATE_SAVING,
1510	VMSTATE_SUSPENDED, VMSTATE_SUSPENDED_EXT_LS,
1511	VMSTATE_RUNNING, VMSTATE_RUNNING_LS,
1512	VMSTATE_DEBUGGING, VMSTATE_DEBUGGING_LS);
1513	if (rc == 1)
1514	rc = VERR_SSM_LIVE_POWERED_OFF;
1515	else if (rc == 2)
1516	rc = VERR_SSM_LIVE_FATAL_ERROR;
1517	else if (rc == 3)
1518	rc = VERR_SSM_LIVE_GURU_MEDITATION;
1519	else if (rc == 4)
1520	{
1521	*pfSuspended = true;
1522	rc = VINF_SUCCESS;
1523	}
1524	else if (rc > 0)
1525	rc = VINF_SUCCESS;
1526	return rc;
1527	}
1528
1529
1530	/**
1531	* EMT(0) worker for VMR3Save and VMR3Teleport that completes the live save.
1532	*
1533	* @returns VBox status code.
1534	* @retval VINF_SSM_LIVE_SUSPENDED if VMR3Suspend was called.
1535	*
1536	* @param pVM The VM handle.
1537	* @param pSSM The handle of saved state operation.
1538	*
1539	* @thread EMT(0)
1540	*/
1541	static DECLCALLBACK(int) vmR3LiveDoStep2(PVM pVM, PSSMHANDLE pSSM)
1542	{
1543	LogFlow(("vmR3LiveDoStep2: pVM=%p pSSM=%p\n", pVM, pSSM));
1544	VM_ASSERT_EMT0(pVM);
1545
1546	/*
1547	* Advance the state and mark if VMR3Suspend was called.
1548	*/
1549	int rc = VINF_SUCCESS;
1550	VMSTATE enmVMState = VMR3GetState(pVM);
1551	if (enmVMState == VMSTATE_SUSPENDED_LS)
1552	vmR3SetState(pVM, VMSTATE_SAVING, VMSTATE_SUSPENDED_LS);
1553	else
1554	{
1555	if (enmVMState != VMSTATE_SAVING)
1556	vmR3SetState(pVM, VMSTATE_SAVING, VMSTATE_SUSPENDED_EXT_LS);
1557	rc = VINF_SSM_LIVE_SUSPENDED;
1558	}
1559
1560	/*
1561	* Finish up and release the handle. Careful with the status codes.
1562	*/
1563	int rc2 = SSMR3LiveDoStep2(pSSM);
1564	if (rc == VINF_SUCCESS \|\| (RT_FAILURE(rc2) && RT_SUCCESS(rc)))
1565	rc = rc2;
1566
1567	rc2 = SSMR3LiveDone(pSSM);
1568	if (rc == VINF_SUCCESS \|\| (RT_FAILURE(rc2) && RT_SUCCESS(rc)))
1569	rc = rc2;
1570
1571	/*
1572	* Advance to the final state and return.
1573	*/
1574	vmR3SetState(pVM, VMSTATE_SUSPENDED, VMSTATE_SAVING);
1575	Assert(rc > VINF_EM_LAST \|\| rc < VINF_EM_FIRST);
1576	return rc;
1577	}
1578
1579
1580	/**
1581	* Worker for vmR3SaveTeleport that validates the state and calls SSMR3Save or
1582	* SSMR3LiveSave.
1583	*
1584	* @returns VBox status code.
1585	*
1586	* @param pVM The VM handle.
1587	* @param cMsMaxDowntime The maximum downtime given as milliseconds.
1588	* @param pszFilename The name of the file. NULL if pStreamOps is used.
1589	* @param pStreamOps The stream methods. NULL if pszFilename is used.
1590	* @param pvStreamOpsUser The user argument to the stream methods.
1591	* @param enmAfter What to do afterwards.
1592	* @param pfnProgress Progress callback. Optional.
1593	* @param pvProgressUser User argument for the progress callback.
1594	* @param ppSSM Where to return the saved state handle in case of a
1595	* live snapshot scenario.
1596	* @thread EMT
1597	*/
1598	static DECLCALLBACK(int) vmR3Save(PVM pVM, uint32_t cMsMaxDowntime, const char pszFilename, PCSSMSTRMOPS pStreamOps, void pvStreamOpsUser,
1599	SSMAFTER enmAfter, PFNVMPROGRESS pfnProgress, void pvProgressUser, PSSMHANDLE ppSSM)
1600	{
1601	LogFlow(("vmR3Save: pVM=%p cMsMaxDowntime=%u pszFilename=%p:{%s} pStreamOps=%p pvStreamOpsUser=%p enmAfter=%d pfnProgress=%p pvProgressUser=%p ppSSM=%p\n",
1602	pVM, cMsMaxDowntime, pszFilename, pszFilename, pStreamOps, pvStreamOpsUser, enmAfter, pfnProgress, pvProgressUser, ppSSM));
1603
1604	/*
1605	* Validate input.
1606	*/
1607	AssertPtrNull(pszFilename);
1608	AssertPtrNull(pStreamOps);
1609	AssertPtr(pVM);
1610	Assert( enmAfter == SSMAFTER_DESTROY
1611	\|\| enmAfter == SSMAFTER_CONTINUE
1612	\|\| enmAfter == SSMAFTER_TELEPORT);
1613	AssertPtr(ppSSM);
1614	*ppSSM = NULL;
1615
1616	/*
1617	* Change the state and perform/start the saving.
1618	*/
1619	int rc = vmR3TrySetState(pVM, "VMR3Save", 2,
1620	VMSTATE_SAVING, VMSTATE_SUSPENDED,
1621	VMSTATE_RUNNING_LS, VMSTATE_RUNNING);
1622	if (rc == 1 && enmAfter != SSMAFTER_TELEPORT)
1623	{
1624	Assert(!pStreamOps);
1625	rc = SSMR3Save(pVM, pszFilename, enmAfter, pfnProgress, pvProgressUser);
1626	vmR3SetState(pVM, VMSTATE_SUSPENDED, VMSTATE_SAVING);
1627	}
1628	else if (rc == 2 \|\| enmAfter == SSMAFTER_TELEPORT)
1629	{
1630	if (enmAfter == SSMAFTER_TELEPORT)
1631	pVM->vm.s.fTeleportedAndNotFullyResumedYet = true;
1632	rc = SSMR3LiveSave(pVM, cMsMaxDowntime, pszFilename, pStreamOps, pvStreamOpsUser,
1633	enmAfter, pfnProgress, pvProgressUser, ppSSM);
1634	/* (We're not subject to cancellation just yet.) */
1635	}
1636	else
1637	Assert(RT_FAILURE(rc));
1638	return rc;
1639	}
1640
1641
1642	/**
1643	* Commmon worker for VMR3Save and VMR3Teleport.
1644	*
1645	* @returns VBox status code.
1646	*
1647	* @param pVM The VM handle.
1648	* @param cMsMaxDowntime The maximum downtime given as milliseconds.
1649	* @param pszFilename The name of the file. NULL if pStreamOps is used.
1650	* @param pStreamOps The stream methods. NULL if pszFilename is used.
1651	* @param pvStreamOpsUser The user argument to the stream methods.
1652	* @param enmAfter What to do afterwards.
1653	* @param pfnProgress Progress callback. Optional.
1654	* @param pvProgressUser User argument for the progress callback.
1655	* @param pfSuspended Set if we suspended the VM.
1656	*
1657	* @thread Non-EMT
1658	*/
1659	static int vmR3SaveTeleport(PVM pVM, uint32_t cMsMaxDowntime,
1660	const char pszFilename, PCSSMSTRMOPS pStreamOps, void pvStreamOpsUser,
1661	SSMAFTER enmAfter, PFNVMPROGRESS pfnProgress, void pvProgressUser, bool pfSuspended)
1662	{
1663	/*
1664	* Request the operation in EMT(0).
1665	*/
1666	PSSMHANDLE pSSM;
1667	int rc = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/,
1668	(PFNRT)vmR3Save, 9, pVM, cMsMaxDowntime, pszFilename, pStreamOps, pvStreamOpsUser,
1669	enmAfter, pfnProgress, pvProgressUser, &pSSM);
1670	if ( RT_SUCCESS(rc)
1671	&& pSSM)
1672	{
1673	/*
1674	* Live snapshot.
1675	*
1676	* The state handling here is kind of tricky, doing it on EMT(0) helps
1677	* a bit. See the VMSTATE diagram for details.
1678	*/
1679	rc = SSMR3LiveDoStep1(pSSM);
1680	if (RT_SUCCESS(rc))
1681	{
1682	if (VMR3GetState(pVM) != VMSTATE_SAVING)
1683	for (;;)
1684	{
1685	/* Try suspend the VM. */
1686	rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
1687	vmR3LiveDoSuspend, pfSuspended);
1688	if (rc != VERR_TRY_AGAIN)
1689	break;
1690
1691	/* Wait for the state to change. */
1692	RTThreadSleep(250); /** @todo Live Migration: fix this polling wait by some smart use of multiple release event semaphores.. */
1693	}
1694	if (RT_SUCCESS(rc))
1695	rc = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/, (PFNRT)vmR3LiveDoStep2, 2, pVM, pSSM);
1696	else
1697	{
1698	int rc2 = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/, (PFNRT)SSMR3LiveDone, 1, pSSM);
1699	AssertMsg(rc2 == rc, ("%Rrc != %Rrc\n", rc2, rc));
1700	}
1701	}
1702	else
1703	{
1704	int rc2 = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/, (PFNRT)SSMR3LiveDone, 1, pSSM);
1705	AssertMsg(rc2 == rc, ("%Rrc != %Rrc\n", rc2, rc));
1706
1707	rc2 = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, vmR3LiveDoStep1Cleanup, pfSuspended);
1708	if (RT_FAILURE(rc2) && rc == VERR_SSM_CANCELLED)
1709	rc = rc2;
1710	}
1711	}
1712
1713	return rc;
1714	}
1715
1716
1717	/**
1718	* Save current VM state.
1719	*
1720	* Can be used for both saving the state and creating snapshots.
1721	*
1722	* When called for a VM in the Running state, the saved state is created live
1723	* and the VM is only suspended when the final part of the saving is preformed.
1724	* The VM state will not be restored to Running in this case and it's up to the
1725	* caller to call VMR3Resume if this is desirable. (The rational is that the
1726	* caller probably wish to reconfigure the disks before resuming the VM.)
1727	*
1728	* @returns VBox status code.
1729	*
1730	* @param pVM The VM which state should be saved.
1731	* @param pszFilename The name of the save state file.
1732	* @param fContinueAfterwards Whether continue execution afterwards or not.
1733	* When in doubt, set this to true.
1734	* @param pfnProgress Progress callback. Optional.
1735	* @param pvUser User argument for the progress callback.
1736	* @param pfSuspended Set if we suspended the VM.
1737	*
1738	* @thread Non-EMT.
1739	* @vmstate Suspended or Running
1740	* @vmstateto Saving+Suspended or
1741	* RunningLS+SuspeningLS+SuspendedLS+Saving+Suspended.
1742	*/
1743	VMMR3DECL(int) VMR3Save(PVM pVM, const char *pszFilename, bool fContinueAfterwards,
1744	PFNVMPROGRESS pfnProgress, void pvUser, bool pfSuspended)
1745	{
1746	LogFlow(("VMR3Save: pVM=%p pszFilename=%p:{%s} fContinueAfterwards=%RTbool pfnProgress=%p pvUser=%p pfSuspended=%p\n",
1747	pVM, pszFilename, pszFilename, fContinueAfterwards, pfnProgress, pvUser, pfSuspended));
1748
1749	/*
1750	* Validate input.
1751	*/
1752	AssertPtr(pfSuspended);
1753	*pfSuspended = false;
1754	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1755	VM_ASSERT_OTHER_THREAD(pVM);
1756	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
1757	AssertReturn(*pszFilename, VERR_INVALID_PARAMETER);
1758	AssertPtrNullReturn(pfnProgress, VERR_INVALID_POINTER);
1759
1760	/*
1761	* Join paths with VMR3Teleport.
1762	*/
1763	SSMAFTER enmAfter = fContinueAfterwards ? SSMAFTER_CONTINUE : SSMAFTER_DESTROY;
1764	int rc = vmR3SaveTeleport(pVM, 250 /cMsMaxDowntime/,
1765	pszFilename, NULL /pStreamOps/, NULL /pvStreamOpsUser/,
1766	enmAfter, pfnProgress, pvUser, pfSuspended);
1767	LogFlow(("VMR3Save: returns %Rrc (pfSuspended=%RTbool)\n", rc, pfSuspended));
1768	return rc;
1769	}
1770
1771
1772	/**
1773	* Teleport the VM (aka live migration).
1774	*
1775	* @returns VBox status code.
1776	*
1777	* @param pVM The VM which state should be saved.
1778	* @param cMsMaxDowntime The maximum downtime given as milliseconds.
1779	* @param pStreamOps The stream methods.
1780	* @param pvStreamOpsUser The user argument to the stream methods.
1781	* @param pfnProgress Progress callback. Optional.
1782	* @param pvProgressUser User argument for the progress callback.
1783	* @param pfSuspended Set if we suspended the VM.
1784	*
1785	* @thread Non-EMT.
1786	* @vmstate Suspended or Running
1787	* @vmstateto Saving+Suspended or
1788	* RunningLS+SuspeningLS+SuspendedLS+Saving+Suspended.
1789	*/
1790	VMMR3DECL(int) VMR3Teleport(PVM pVM, uint32_t cMsMaxDowntime, PCSSMSTRMOPS pStreamOps, void *pvStreamOpsUser,
1791	PFNVMPROGRESS pfnProgress, void pvProgressUser, bool pfSuspended)
1792	{
1793	LogFlow(("VMR3Teleport: pVM=%p cMsMaxDowntime=%u pStreamOps=%p pvStreamOps=%p pfnProgress=%p pvProgressUser=%p\n",
1794	pVM, cMsMaxDowntime, pStreamOps, pvStreamOpsUser, pfnProgress, pvProgressUser));
1795
1796	/*
1797	* Validate input.
1798	*/
1799	AssertPtr(pfSuspended);
1800	*pfSuspended = false;
1801	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1802	VM_ASSERT_OTHER_THREAD(pVM);
1803	AssertPtrReturn(pStreamOps, VERR_INVALID_POINTER);
1804	AssertPtrNullReturn(pfnProgress, VERR_INVALID_POINTER);
1805
1806	/*
1807	* Join paths with VMR3Save.
1808	*/
1809	int rc = vmR3SaveTeleport(pVM, cMsMaxDowntime,
1810	NULL /pszFilename/, pStreamOps, pvStreamOpsUser,
1811	SSMAFTER_TELEPORT, pfnProgress, pvProgressUser, pfSuspended);
1812	LogFlow(("VMR3Teleport: returns %Rrc (pfSuspended=%RTbool)\n", rc, pfSuspended));
1813	return rc;
1814	}
1815
1816
1817
1818	/**
1819	* EMT(0) worker for VMR3LoadFromFile and VMR3LoadFromStream.
1820	*
1821	* @returns VBox status code.
1822	*
1823	* @param pVM The VM handle.
1824	* @param pszFilename The name of the file. NULL if pStreamOps is used.
1825	* @param pStreamOps The stream methods. NULL if pszFilename is used.
1826	* @param pvStreamOpsUser The user argument to the stream methods.
1827	* @param pfnProgress Progress callback. Optional.
1828	* @param pvUser User argument for the progress callback.
1829	* @param fTeleporting Indicates whether we're teleporting or not.
1830	*
1831	* @thread EMT.
1832	*/
1833	static DECLCALLBACK(int) vmR3Load(PVM pVM, const char pszFilename, PCSSMSTRMOPS pStreamOps, void pvStreamOpsUser,
1834	PFNVMPROGRESS pfnProgress, void *pvProgressUser, bool fTeleporting)
1835	{
1836	LogFlow(("vmR3Load: pVM=%p pszFilename=%p:{%s} pStreamOps=%p pvStreamOpsUser=%p pfnProgress=%p pvProgressUser=%p fTeleporting=%RTbool\n",
1837	pVM, pszFilename, pszFilename, pStreamOps, pvStreamOpsUser, pfnProgress, pvProgressUser, fTeleporting));
1838
1839	/*
1840	* Validate input (paranoia).
1841	*/
1842	AssertPtr(pVM);
1843	AssertPtrNull(pszFilename);
1844	AssertPtrNull(pStreamOps);
1845	AssertPtrNull(pfnProgress);
1846
1847	/*
1848	* Change the state and perform the load.
1849	*
1850	* Always perform a relocation round afterwards to make sure hypervisor
1851	* selectors and such are correct.
1852	*/
1853	int rc = vmR3TrySetState(pVM, "VMR3Load", 2,
1854	VMSTATE_LOADING, VMSTATE_CREATED,
1855	VMSTATE_LOADING, VMSTATE_SUSPENDED);
1856	if (RT_FAILURE(rc))
1857	return rc;
1858	pVM->vm.s.fTeleportedAndNotFullyResumedYet = fTeleporting;
1859
1860	uint32_t cErrorsPriorToSave = VMR3GetErrorCount(pVM);
1861	rc = SSMR3Load(pVM, pszFilename, pStreamOps, pvStreamOpsUser, SSMAFTER_RESUME, pfnProgress, pvProgressUser);
1862	if (RT_SUCCESS(rc))
1863	{
1864	VMR3Relocate(pVM, 0 /offDelta/);
1865	vmR3SetState(pVM, VMSTATE_SUSPENDED, VMSTATE_LOADING);
1866	}
1867	else
1868	{
1869	pVM->vm.s.fTeleportedAndNotFullyResumedYet = false;
1870	vmR3SetState(pVM, VMSTATE_LOAD_FAILURE, VMSTATE_LOADING);
1871	if (cErrorsPriorToSave == VMR3GetErrorCount(pVM))
1872	rc = VMSetError(pVM, rc, RT_SRC_POS,
1873	N_("Unable to restore the virtual machine's saved state from '%s'. "
1874	"It may be damaged or from an older version of VirtualBox. "
1875	"Please discard the saved state before starting the virtual machine"),
1876	pszFilename);
1877	}
1878
1879	return rc;
1880	}
1881
1882
1883	/**
1884	* Loads a VM state into a newly created VM or a one that is suspended.
1885	*
1886	* To restore a saved state on VM startup, call this function and then resume
1887	* the VM instead of powering it on.
1888	*
1889	* @returns VBox status code.
1890	*
1891	* @param pVM The VM handle.
1892	* @param pszFilename The name of the save state file.
1893	* @param pfnProgress Progress callback. Optional.
1894	* @param pvUser User argument for the progress callback.
1895	*
1896	* @thread Any thread.
1897	* @vmstate Created, Suspended
1898	* @vmstateto Loading+Suspended
1899	*/
1900	VMMR3DECL(int) VMR3LoadFromFile(PVM pVM, const char pszFilename, PFNVMPROGRESS pfnProgress, void pvUser)
1901	{
1902	LogFlow(("VMR3LoadFromFile: pVM=%p pszFilename=%p:{%s} pfnProgress=%p pvUser=%p\n",
1903	pVM, pszFilename, pszFilename, pfnProgress, pvUser));
1904
1905	/*
1906	* Validate input.
1907	*/
1908	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1909	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
1910
1911	/*
1912	* Forward the request to EMT(0). No need to setup a rendezvous here
1913	* since there is no execution taking place when this call is allowed.
1914	*/
1915	int rc = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/, (PFNRT)vmR3Load, 7,
1916	pVM, pszFilename, (uintptr_t)NULL /pStreamOps/, (uintptr_t)NULL /pvStreamOpsUser/, pfnProgress, pvUser,
1917	false /fTeleporting/);
1918	LogFlow(("VMR3LoadFromFile: returns %Rrc\n", rc));
1919	return rc;
1920	}
1921
1922
1923	/**
1924	* VMR3LoadFromFile for arbritrary file streams.
1925	*
1926	* @returns VBox status code.
1927	*
1928	* @param pVM The VM handle.
1929	* @param pStreamOps The stream methods.
1930	* @param pvStreamOpsUser The user argument to the stream methods.
1931	* @param pfnProgress Progress callback. Optional.
1932	* @param pvProgressUser User argument for the progress callback.
1933	*
1934	* @thread Any thread.
1935	* @vmstate Created, Suspended
1936	* @vmstateto Loading+Suspended
1937	*/
1938	VMMR3DECL(int) VMR3LoadFromStream(PVM pVM, PCSSMSTRMOPS pStreamOps, void *pvStreamOpsUser,
1939	PFNVMPROGRESS pfnProgress, void *pvProgressUser)
1940	{
1941	LogFlow(("VMR3LoadFromStream: pVM=%p pStreamOps=%p pvStreamOpsUser=%p pfnProgress=%p pvProgressUser=%p\n",
1942	pVM, pStreamOps, pvStreamOpsUser, pfnProgress, pvProgressUser));
1943
1944	/*
1945	* Validate input.
1946	*/
1947	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1948	AssertPtrReturn(pStreamOps, VERR_INVALID_POINTER);
1949
1950	/*
1951	* Forward the request to EMT(0). No need to setup a rendezvous here
1952	* since there is no execution taking place when this call is allowed.
1953	*/
1954	int rc = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/, (PFNRT)vmR3Load, 7,
1955	pVM, (uintptr_t)NULL /pszFilename/, pStreamOps, pvStreamOpsUser, pfnProgress, pvProgressUser,
1956	true /fTeleporting/);
1957	LogFlow(("VMR3LoadFromStream: returns %Rrc\n", rc));
1958	return rc;
1959	}
1960
1961
1962	/**
1963	* EMT rendezvous worker for VMR3PowerOff.
1964	*
1965	* @returns VERR_VM_INVALID_VM_STATE or VINF_EM_OFF. (This is a strict
1966	* return code, see FNVMMEMTRENDEZVOUS.)
1967	*
1968	* @param pVM The VM handle.
1969	* @param pVCpu The VMCPU handle of the EMT.
1970	* @param pvUser Ignored.
1971	*/
1972	static DECLCALLBACK(VBOXSTRICTRC) vmR3PowerOff(PVM pVM, PVMCPU pVCpu, void *pvUser)
1973	{
1974	LogFlow(("vmR3PowerOff: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1975	Assert(!pvUser); NOREF(pvUser);
1976
1977	/*
1978	* The first EMT thru here will change the state to PoweringOff.
1979	*/
1980	if (pVCpu->idCpu == pVM->cCpus - 1)
1981	{
1982	int rc = vmR3TrySetState(pVM, "VMR3PowerOff", 11,
1983	VMSTATE_POWERING_OFF, VMSTATE_RUNNING, /* 1 */
1984	VMSTATE_POWERING_OFF, VMSTATE_SUSPENDED, /* 2 */
1985	VMSTATE_POWERING_OFF, VMSTATE_DEBUGGING, /* 3 */
1986	VMSTATE_POWERING_OFF, VMSTATE_LOAD_FAILURE, /* 4 */
1987	VMSTATE_POWERING_OFF, VMSTATE_GURU_MEDITATION, /* 5 */
1988	VMSTATE_POWERING_OFF, VMSTATE_FATAL_ERROR, /* 6 */
1989	VMSTATE_POWERING_OFF, VMSTATE_CREATED, /* 7 / /* @todo update the diagram! */
1990	VMSTATE_POWERING_OFF_LS, VMSTATE_RUNNING_LS, /* 8 */
1991	VMSTATE_POWERING_OFF_LS, VMSTATE_DEBUGGING_LS, /* 9 */
1992	VMSTATE_POWERING_OFF_LS, VMSTATE_GURU_MEDITATION_LS,/* 10 */
1993	VMSTATE_POWERING_OFF_LS, VMSTATE_FATAL_ERROR_LS); /* 11 */
1994	if (RT_FAILURE(rc))
1995	return rc;
1996	if (rc >= 7)
1997	SSMR3Cancel(pVM);
1998	}
1999
2000	/*
2001	* Check the state.
2002	*/
2003	VMSTATE enmVMState = VMR3GetState(pVM);
2004	AssertMsgReturn( enmVMState == VMSTATE_POWERING_OFF
2005	\|\| enmVMState == VMSTATE_POWERING_OFF_LS,
2006	("%s\n", VMR3GetStateName(enmVMState)),
2007	VERR_VM_INVALID_VM_STATE);
2008
2009	/*
2010	* EMT(0) does the actual power off work here after all the other EMTs
2011	* have been thru and entered the STOPPED state.
2012	*/
2013	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STOPPED);
2014	if (pVCpu->idCpu == 0)
2015	{
2016	/*
2017	* For debugging purposes, we will log a summary of the guest state at this point.
2018	*/
2019	if (enmVMState != VMSTATE_GURU_MEDITATION)
2020	{
2021	/** @todo SMP support? */
2022	/** @todo make the state dumping at VMR3PowerOff optional. */
2023	RTLogRelPrintf("**************** Guest state at power off ****************\n");
2024	DBGFR3Info(pVM, "cpumguest", "verbose", DBGFR3InfoLogRelHlp());
2025	RTLogRelPrintf("***\n");
2026	DBGFR3Info(pVM, "mode", NULL, DBGFR3InfoLogRelHlp());
2027	RTLogRelPrintf("***\n");
2028	DBGFR3Info(pVM, "activetimers", NULL, DBGFR3InfoLogRelHlp());
2029	RTLogRelPrintf("***\n");
2030	DBGFR3Info(pVM, "gdt", NULL, DBGFR3InfoLogRelHlp());
2031	/** @todo dump guest call stack. */
2032	#if 1 // "temporary" while debugging #1589
2033	RTLogRelPrintf("***\n");
2034	uint32_t esp = CPUMGetGuestESP(pVCpu);
2035	if ( CPUMGetGuestSS(pVCpu) == 0
2036	&& esp < _64K)
2037	{
2038	uint8_t abBuf[PAGE_SIZE];
2039	RTLogRelPrintf("***\n"
2040	"ss:sp=0000:%04x ", esp);
2041	uint32_t Start = esp & ~(uint32_t)63;
2042	int rc = PGMPhysSimpleReadGCPhys(pVM, abBuf, Start, 0x100);
2043	if (RT_SUCCESS(rc))
2044	RTLogRelPrintf("0000:%04x TO 0000:%04x:\n"
2045	"%.*Rhxd\n",
2046	Start, Start + 0x100 - 1,
2047	0x100, abBuf);
2048	else
2049	RTLogRelPrintf("rc=%Rrc\n", rc);
2050
2051	/* grub ... */
2052	if (esp < 0x2000 && esp > 0x1fc0)
2053	{
2054	rc = PGMPhysSimpleReadGCPhys(pVM, abBuf, 0x8000, 0x800);
2055	if (RT_SUCCESS(rc))
2056	RTLogRelPrintf("0000:8000 TO 0000:87ff:\n"
2057	"%.*Rhxd\n",
2058	0x800, abBuf);
2059	}
2060	/* microsoft cdrom hang ... */
2061	if (true)
2062	{
2063	rc = PGMPhysSimpleReadGCPhys(pVM, abBuf, 0x8000, 0x200);
2064	if (RT_SUCCESS(rc))
2065	RTLogRelPrintf("2000:0000 TO 2000:01ff:\n"
2066	"%.*Rhxd\n",
2067	0x200, abBuf);
2068	}
2069	}
2070	#endif
2071	RTLogRelPrintf("************ End of Guest state at power off *************\n");
2072	}
2073
2074	/*
2075	* Perform the power off notifications and advance the state to
2076	* Off or OffLS.
2077	*/
2078	PDMR3PowerOff(pVM);
2079
2080	PUVM pUVM = pVM->pUVM;
2081	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
2082	enmVMState = pVM->enmVMState;
2083	if (enmVMState == VMSTATE_POWERING_OFF_LS)
2084	vmR3SetStateLocked(pVM, pUVM, VMSTATE_OFF_LS, VMSTATE_POWERING_OFF_LS);
2085	else
2086	vmR3SetStateLocked(pVM, pUVM, VMSTATE_OFF, VMSTATE_POWERING_OFF);
2087	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
2088	}
2089	return VINF_EM_OFF;
2090	}
2091
2092
2093	/**
2094	* Power off the VM.
2095	*
2096	* @returns VBox status code. When called on EMT, this will be a strict status
2097	* code that has to be propagated up the call stack.
2098	*
2099	* @param pVM The handle of the VM to be powered off.
2100	*
2101	* @thread Any thread.
2102	* @vmstate Suspended, Running, Guru Meditation, Load Failure
2103	* @vmstateto Off or OffLS
2104	*/
2105	VMMR3DECL(int) VMR3PowerOff(PVM pVM)
2106	{
2107	LogFlow(("VMR3PowerOff: pVM=%p\n", pVM));
2108	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
2109
2110	/*
2111	* Gather all the EMTs to make sure there are no races before
2112	* changing the VM state.
2113	*/
2114	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
2115	vmR3PowerOff, NULL);
2116	LogFlow(("VMR3PowerOff: returns %Rrc\n", rc));
2117	return rc;
2118	}
2119
2120
2121	/**
2122	* Destroys the VM.
2123	*
2124	* The VM must be powered off (or never really powered on) to call this
2125	* function. The VM handle is destroyed and can no longer be used up successful
2126	* return.
2127	*
2128	* @returns VBox status code.
2129	*
2130	* @param pVM The handle of the VM which should be destroyed.
2131	*
2132	* @thread EMT(0) or any none emulation thread.
2133	* @vmstate Off, Created
2134	* @vmstateto N/A
2135	*/
2136	VMMR3DECL(int) VMR3Destroy(PVM pVM)
2137	{
2138	LogFlow(("VMR3Destroy: pVM=%p\n", pVM));
2139
2140	/*
2141	* Validate input.
2142	*/
2143	if (!pVM)
2144	return VERR_INVALID_PARAMETER;
2145	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
2146	Assert(VMMGetCpuId(pVM) == 0 \|\| VMMGetCpuId(pVM) == NIL_VMCPUID);
2147
2148	/*
2149	* Change VM state to destroying and unlink the VM.
2150	*/
2151	int rc = vmR3TrySetState(pVM, "VMR3Destroy", 1, VMSTATE_DESTROYING, VMSTATE_OFF);
2152	if (RT_FAILURE(rc))
2153	return rc;
2154
2155	/** @todo lock this when we start having multiple machines in a process... */
2156	PUVM pUVM = pVM->pUVM; AssertPtr(pUVM);
2157	if (g_pUVMsHead == pUVM)
2158	g_pUVMsHead = pUVM->pNext;
2159	else
2160	{
2161	PUVM pPrev = g_pUVMsHead;
2162	while (pPrev && pPrev->pNext != pUVM)
2163	pPrev = pPrev->pNext;
2164	AssertMsgReturn(pPrev, ("pUVM=%p / pVM=%p is INVALID!\n", pUVM, pVM), VERR_INVALID_PARAMETER);
2165
2166	pPrev->pNext = pUVM->pNext;
2167	}
2168	pUVM->pNext = NULL;
2169
2170	/*
2171	* Notify registered at destruction listeners.
2172	*/
2173	vmR3AtDtor(pVM);
2174
2175	/*
2176	* EMT(0) does the final cleanup, so if we're it calling VMR3Destroy then
2177	* we'll have to postpone parts of it till later. Otherwise, call
2178	* vmR3Destroy on each of the EMTs in ending with EMT(0) doing the bulk
2179	* of the cleanup.
2180	*/
2181	if (VMMGetCpuId(pVM) == 0)
2182	{
2183	pUVM->vm.s.fEMTDoesTheCleanup = true;
2184	pUVM->vm.s.fTerminateEMT = true;
2185	VM_FF_SET(pVM, VM_FF_TERMINATE);
2186
2187	/* Terminate the other EMTs. */
2188	for (VMCPUID idCpu = 1; idCpu < pVM->cCpus; idCpu++)
2189	{
2190	rc = VMR3ReqCallWaitU(pUVM, idCpu, (PFNRT)vmR3Destroy, 1, pVM);
2191	AssertLogRelRC(rc);
2192	}
2193	}
2194	else
2195	{
2196	/* vmR3Destroy on all EMTs, ending with EMT(0). */
2197	rc = VMR3ReqCallWaitU(pUVM, VMCPUID_ALL_REVERSE, (PFNRT)vmR3Destroy, 1, pVM);
2198	AssertLogRelRC(rc);
2199
2200	/* Wait for EMTs and destroy the UVM. */
2201	vmR3DestroyUVM(pUVM, 30000);
2202	}
2203
2204	LogFlow(("VMR3Destroy: returns VINF_SUCCESS\n"));
2205	return VINF_SUCCESS;
2206	}
2207
2208
2209	/**
2210	* Internal destruction worker.
2211	*
2212	* This is either called from VMR3Destroy via VMR3ReqCallU or from
2213	* vmR3EmulationThreadWithId when EMT(0) terminates after having called
2214	* VMR3Destroy().
2215	*
2216	* When called on EMT(0), it will performed the great bulk of the destruction.
2217	* When called on the other EMTs, they will do nothing and the whole purpose is
2218	* to return VINF_EM_TERMINATE so they break out of their run loops.
2219	*
2220	* @returns VINF_EM_TERMINATE.
2221	* @param pVM The VM handle.
2222	*/
2223	DECLCALLBACK(int) vmR3Destroy(PVM pVM)
2224	{
2225	PUVM pUVM = pVM->pUVM;
2226	PVMCPU pVCpu = VMMGetCpu(pVM);
2227	Assert(pVCpu);
2228	LogFlow(("vmR3Destroy: pVM=%p pUVM=%p pVCpu=%p idCpu=%u\n", pVM, pUVM, pVCpu, pVCpu->idCpu));
2229
2230	/*
2231	* Only VCPU 0 does the full cleanup.
2232	*/
2233	if (pVCpu->idCpu == 0)
2234	{
2235
2236	/*
2237	* Dump statistics to the log.
2238	*/
2239	#if defined(VBOX_WITH_STATISTICS) \|\| defined(LOG_ENABLED)
2240	RTLogFlags(NULL, "nodisabled nobuffered");
2241	#endif
2242	#ifdef VBOX_WITH_STATISTICS
2243	STAMR3Dump(pVM, "*");
2244	#else
2245	LogRel(("*********************** Statistics ***********************\n"));
2246	STAMR3DumpToReleaseLog(pVM, "*");
2247	LogRel(("******************* End of statistics ********************\n"));
2248	#endif
2249
2250	/*
2251	* Destroy the VM components.
2252	*/
2253	int rc = TMR3Term(pVM);
2254	AssertRC(rc);
2255	#ifdef VBOX_WITH_DEBUGGER
2256	rc = DBGCTcpTerminate(pVM, pUVM->vm.s.pvDBGC);
2257	pUVM->vm.s.pvDBGC = NULL;
2258	#endif
2259	AssertRC(rc);
2260	rc = DBGFR3Term(pVM);
2261	AssertRC(rc);
2262	rc = PDMR3Term(pVM);
2263	AssertRC(rc);
2264	rc = EMR3Term(pVM);
2265	AssertRC(rc);
2266	rc = IOMR3Term(pVM);
2267	AssertRC(rc);
2268	rc = CSAMR3Term(pVM);
2269	AssertRC(rc);
2270	rc = PATMR3Term(pVM);
2271	AssertRC(rc);
2272	rc = TRPMR3Term(pVM);
2273	AssertRC(rc);
2274	rc = SELMR3Term(pVM);
2275	AssertRC(rc);
2276	rc = REMR3Term(pVM);
2277	AssertRC(rc);
2278	rc = HWACCMR3Term(pVM);
2279	AssertRC(rc);
2280	rc = PGMR3Term(pVM);
2281	AssertRC(rc);
2282	rc = VMMR3Term(pVM); /* Terminates the ring-0 code! */
2283	AssertRC(rc);
2284	rc = CPUMR3Term(pVM);
2285	AssertRC(rc);
2286	SSMR3Term(pVM);
2287	rc = PDMR3CritSectTerm(pVM);
2288	AssertRC(rc);
2289	rc = MMR3Term(pVM);
2290	AssertRC(rc);
2291
2292	/*
2293	* We're done in this thread (EMT).
2294	*/
2295	ASMAtomicUoWriteBool(&pUVM->vm.s.fTerminateEMT, true);
2296	ASMAtomicWriteU32(&pVM->fGlobalForcedActions, VM_FF_TERMINATE);
2297	LogFlow(("vmR3Destroy: returning %Rrc\n", VINF_EM_TERMINATE));
2298	}
2299	return VINF_EM_TERMINATE;
2300	}
2301
2302
2303	/**
2304	* Called at the end of the EMT procedure to take care of the final cleanup.
2305	*
2306	* Currently only EMT(0) will do work here. It will destroy the shared VM
2307	* structure if it is still around. If EMT(0) was the caller of VMR3Destroy it
2308	* will destroy UVM and nothing will be left behind upon exit. But if some
2309	* other thread is calling VMR3Destroy, they will clean up UVM after all EMTs
2310	* has exitted.
2311	*
2312	* @param pUVM The UVM handle.
2313	* @param idCpu The virtual CPU id.
2314	*/
2315	void vmR3DestroyFinalBitFromEMT(PUVM pUVM, VMCPUID idCpu)
2316	{
2317	/*
2318	* Only EMT(0) has work to do here.
2319	*/
2320	if (idCpu != 0)
2321	return;
2322	Assert( !pUVM->pVM
2323	\|\| VMMGetCpuId(pUVM->pVM) == 0);
2324
2325	/*
2326	* If we have a shared VM structure, change its state to Terminated and
2327	* tell GVMM to destroy it.
2328	*/
2329	if (pUVM->pVM)
2330	{
2331	vmR3SetState(pUVM->pVM, VMSTATE_TERMINATED, VMSTATE_DESTROYING);
2332	int rc = SUPR3CallVMMR0Ex(pUVM->pVM->pVMR0, 0 /idCpu/, VMMR0_DO_GVMM_DESTROY_VM, 0, NULL);
2333	AssertLogRelRC(rc);
2334	pUVM->pVM = NULL;
2335	}
2336
2337	/*
2338	* If EMT(0) called VMR3Destroy, then it will destroy UVM as well.
2339	*/
2340	if (pUVM->vm.s.fEMTDoesTheCleanup)
2341	vmR3DestroyUVM(pUVM, 30000);
2342	}
2343
2344
2345	/**
2346	* Destroys the UVM portion.
2347	*
2348	* This is called as the final step in the VM destruction or as the cleanup
2349	* in case of a creation failure. If EMT(0) called VMR3Destroy, meaning
2350	* VMINTUSERPERVM::fEMTDoesTheCleanup is true, it will call this as
2351	* vmR3DestroyFinalBitFromEMT completes.
2352	*
2353	* @param pVM VM Handle.
2354	* @param cMilliesEMTWait The number of milliseconds to wait for the emulation
2355	* threads.
2356	*/
2357	static void vmR3DestroyUVM(PUVM pUVM, uint32_t cMilliesEMTWait)
2358	{
2359	/*
2360	* Signal termination of each the emulation threads and
2361	* wait for them to complete.
2362	*/
2363	/* Signal them. */
2364	ASMAtomicUoWriteBool(&pUVM->vm.s.fTerminateEMT, true);
2365	for (VMCPUID i = 0; i < pUVM->cCpus; i++)
2366	{
2367	ASMAtomicUoWriteBool(&pUVM->aCpus[i].vm.s.fTerminateEMT, true);
2368	if (pUVM->pVM)
2369	VM_FF_SET(pUVM->pVM, VM_FF_TERMINATE);
2370	VMR3NotifyGlobalFFU(pUVM, VMNOTIFYFF_FLAGS_DONE_REM);
2371	RTSemEventSignal(pUVM->aCpus[i].vm.s.EventSemWait);
2372	}
2373
2374	/* Wait for them. */
2375	uint64_t NanoTS = RTTimeNanoTS();
2376	RTTHREAD hSelf = RTThreadSelf();
2377	ASMAtomicUoWriteBool(&pUVM->vm.s.fTerminateEMT, true);
2378	for (VMCPUID i = 0; i < pUVM->cCpus; i++)
2379	{
2380	RTTHREAD hThread = pUVM->aCpus[i].vm.s.ThreadEMT;
2381	if ( hThread != NIL_RTTHREAD
2382	&& hThread != hSelf)
2383	{
2384	uint64_t cMilliesElapsed = (RTTimeNanoTS() - NanoTS) / 1000000;
2385	int rc2 = RTThreadWait(hThread,
2386	cMilliesElapsed < cMilliesEMTWait
2387	? RT_MAX(cMilliesEMTWait - cMilliesElapsed, 2000)
2388	: 2000,
2389	NULL);
2390	if (rc2 == VERR_TIMEOUT) /* avoid the assertion when debugging. */
2391	rc2 = RTThreadWait(hThread, 1000, NULL);
2392	AssertLogRelMsgRC(rc2, ("i=%u rc=%Rrc\n", i, rc2));
2393	if (RT_SUCCESS(rc2))
2394	pUVM->aCpus[0].vm.s.ThreadEMT = NIL_RTTHREAD;
2395	}
2396	}
2397
2398	/* Cleanup the semaphores. */
2399	for (VMCPUID i = 0; i < pUVM->cCpus; i++)
2400	{
2401	RTSemEventDestroy(pUVM->aCpus[i].vm.s.EventSemWait);
2402	pUVM->aCpus[i].vm.s.EventSemWait = NIL_RTSEMEVENT;
2403	}
2404
2405	/*
2406	* Free the event semaphores associated with the request packets.
2407	*/
2408	unsigned cReqs = 0;
2409	for (unsigned i = 0; i < RT_ELEMENTS(pUVM->vm.s.apReqFree); i++)
2410	{
2411	PVMREQ pReq = pUVM->vm.s.apReqFree[i];
2412	pUVM->vm.s.apReqFree[i] = NULL;
2413	for (; pReq; pReq = pReq->pNext, cReqs++)
2414	{
2415	pReq->enmState = VMREQSTATE_INVALID;
2416	RTSemEventDestroy(pReq->EventSem);
2417	}
2418	}
2419	Assert(cReqs == pUVM->vm.s.cReqFree); NOREF(cReqs);
2420
2421	/*
2422	* Kill all queued requests. (There really shouldn't be any!)
2423	*/
2424	for (unsigned i = 0; i < 10; i++)
2425	{
2426	PVMREQ pReqHead = ASMAtomicXchgPtrT(&pUVM->vm.s.pReqs, NULL, PVMREQ);
2427	AssertMsg(!pReqHead, ("This isn't supposed to happen! VMR3Destroy caller has to serialize this.\n"));
2428	if (!pReqHead)
2429	break;
2430	for (PVMREQ pReq = pReqHead; pReq; pReq = pReq->pNext)
2431	{
2432	ASMAtomicUoWriteSize(&pReq->iStatus, VERR_INTERNAL_ERROR);
2433	ASMAtomicWriteSize(&pReq->enmState, VMREQSTATE_INVALID);
2434	RTSemEventSignal(pReq->EventSem);
2435	RTThreadSleep(2);
2436	RTSemEventDestroy(pReq->EventSem);
2437	}
2438	/* give them a chance to respond before we free the request memory. */
2439	RTThreadSleep(32);
2440	}
2441
2442	/*
2443	* Now all queued VCPU requests (again, there shouldn't be any).
2444	*/
2445	for (VMCPUID idCpu = 0; idCpu < pUVM->cCpus; idCpu++)
2446	{
2447	PUVMCPU pUVCpu = &pUVM->aCpus[idCpu];
2448
2449	for (unsigned i = 0; i < 10; i++)
2450	{
2451	PVMREQ pReqHead = ASMAtomicXchgPtrT(&pUVCpu->vm.s.pReqs, NULL, PVMREQ);
2452	AssertMsg(!pReqHead, ("This isn't supposed to happen! VMR3Destroy caller has to serialize this.\n"));
2453	if (!pReqHead)
2454	break;
2455	for (PVMREQ pReq = pReqHead; pReq; pReq = pReq->pNext)
2456	{
2457	ASMAtomicUoWriteSize(&pReq->iStatus, VERR_INTERNAL_ERROR);
2458	ASMAtomicWriteSize(&pReq->enmState, VMREQSTATE_INVALID);
2459	RTSemEventSignal(pReq->EventSem);
2460	RTThreadSleep(2);
2461	RTSemEventDestroy(pReq->EventSem);
2462	}
2463	/* give them a chance to respond before we free the request memory. */
2464	RTThreadSleep(32);
2465	}
2466	}
2467
2468	/*
2469	* Make sure the VMMR0.r0 module and whatever else is unloaded.
2470	*/
2471	PDMR3TermUVM(pUVM);
2472
2473	/*
2474	* Terminate the support library if initialized.
2475	*/
2476	if (pUVM->vm.s.pSession)
2477	{
2478	int rc = SUPR3Term(false /fForced/);
2479	AssertRC(rc);
2480	pUVM->vm.s.pSession = NIL_RTR0PTR;
2481	}
2482
2483	/*
2484	* Destroy the MM heap and free the UVM structure.
2485	*/
2486	MMR3TermUVM(pUVM);
2487	STAMR3TermUVM(pUVM);
2488
2489	#ifdef LOG_ENABLED
2490	RTLogSetCustomPrefixCallback(NULL, NULL, NULL);
2491	#endif
2492	RTTlsFree(pUVM->vm.s.idxTLS);
2493
2494	ASMAtomicUoWriteU32(&pUVM->u32Magic, UINT32_MAX);
2495	RTMemPageFree(pUVM, sizeof(*pUVM));
2496
2497	RTLogFlush(NULL);
2498	}
2499
2500
2501	/**
2502	* Enumerates the VMs in this process.
2503	*
2504	* @returns Pointer to the next VM.
2505	* @returns NULL when no more VMs.
2506	* @param pVMPrev The previous VM
2507	* Use NULL to start the enumeration.
2508	*/
2509	VMMR3DECL(PVM) VMR3EnumVMs(PVM pVMPrev)
2510	{
2511	/*
2512	* This is quick and dirty. It has issues with VM being
2513	* destroyed during the enumeration.
2514	*/
2515	PUVM pNext;
2516	if (pVMPrev)
2517	pNext = pVMPrev->pUVM->pNext;
2518	else
2519	pNext = g_pUVMsHead;
2520	return pNext ? pNext->pVM : NULL;
2521	}
2522
2523
2524	/**
2525	* Registers an at VM destruction callback.
2526	*
2527	* @returns VBox status code.
2528	* @param pfnAtDtor Pointer to callback.
2529	* @param pvUser User argument.
2530	*/
2531	VMMR3DECL(int) VMR3AtDtorRegister(PFNVMATDTOR pfnAtDtor, void *pvUser)
2532	{
2533	/*
2534	* Check if already registered.
2535	*/
2536	VM_ATDTOR_LOCK();
2537	PVMATDTOR pCur = g_pVMAtDtorHead;
2538	while (pCur)
2539	{
2540	if (pfnAtDtor == pCur->pfnAtDtor)
2541	{
2542	VM_ATDTOR_UNLOCK();
2543	AssertMsgFailed(("Already registered at destruction callback %p!\n", pfnAtDtor));
2544	return VERR_INVALID_PARAMETER;
2545	}
2546
2547	/* next */
2548	pCur = pCur->pNext;
2549	}
2550	VM_ATDTOR_UNLOCK();
2551
2552	/*
2553	* Allocate new entry.
2554	*/
2555	PVMATDTOR pVMAtDtor = (PVMATDTOR)RTMemAlloc(sizeof(*pVMAtDtor));
2556	if (!pVMAtDtor)
2557	return VERR_NO_MEMORY;
2558
2559	VM_ATDTOR_LOCK();
2560	pVMAtDtor->pfnAtDtor = pfnAtDtor;
2561	pVMAtDtor->pvUser = pvUser;
2562	pVMAtDtor->pNext = g_pVMAtDtorHead;
2563	g_pVMAtDtorHead = pVMAtDtor;
2564	VM_ATDTOR_UNLOCK();
2565
2566	return VINF_SUCCESS;
2567	}
2568
2569
2570	/**
2571	* Deregisters an at VM destruction callback.
2572	*
2573	* @returns VBox status code.
2574	* @param pfnAtDtor Pointer to callback.
2575	*/
2576	VMMR3DECL(int) VMR3AtDtorDeregister(PFNVMATDTOR pfnAtDtor)
2577	{
2578	/*
2579	* Find it, unlink it and free it.
2580	*/
2581	VM_ATDTOR_LOCK();
2582	PVMATDTOR pPrev = NULL;
2583	PVMATDTOR pCur = g_pVMAtDtorHead;
2584	while (pCur)
2585	{
2586	if (pfnAtDtor == pCur->pfnAtDtor)
2587	{
2588	if (pPrev)
2589	pPrev->pNext = pCur->pNext;
2590	else
2591	g_pVMAtDtorHead = pCur->pNext;
2592	pCur->pNext = NULL;
2593	VM_ATDTOR_UNLOCK();
2594
2595	RTMemFree(pCur);
2596	return VINF_SUCCESS;
2597	}
2598
2599	/* next */
2600	pPrev = pCur;
2601	pCur = pCur->pNext;
2602	}
2603	VM_ATDTOR_UNLOCK();
2604
2605	return VERR_INVALID_PARAMETER;
2606	}
2607
2608
2609	/**
2610	* Walks the list of at VM destructor callbacks.
2611	* @param pVM The VM which is about to be destroyed.
2612	*/
2613	static void vmR3AtDtor(PVM pVM)
2614	{
2615	/*
2616	* Find it, unlink it and free it.
2617	*/
2618	VM_ATDTOR_LOCK();
2619	for (PVMATDTOR pCur = g_pVMAtDtorHead; pCur; pCur = pCur->pNext)
2620	pCur->pfnAtDtor(pVM, pCur->pvUser);
2621	VM_ATDTOR_UNLOCK();
2622	}
2623
2624
2625	/**
2626	* Worker which checks integrity of some internal structures.
2627	* This is yet another attempt to track down that AVL tree crash.
2628	*/
2629	static void vmR3CheckIntegrity(PVM pVM)
2630	{
2631	#ifdef VBOX_STRICT
2632	int rc = PGMR3CheckIntegrity(pVM);
2633	AssertReleaseRC(rc);
2634	#endif
2635	}
2636
2637
2638	/**
2639	* EMT rendezvous worker for VMR3Reset.
2640	*
2641	* This is called by the emulation threads as a response to the reset request
2642	* issued by VMR3Reset().
2643	*
2644	* @returns VERR_VM_INVALID_VM_STATE, VINF_EM_RESET or VINF_EM_SUSPEND. (This
2645	* is a strict return code, see FNVMMEMTRENDEZVOUS.)
2646	*
2647	* @param pVM The VM handle.
2648	* @param pVCpu The VMCPU handle of the EMT.
2649	* @param pvUser Ignored.
2650	*/
2651	static DECLCALLBACK(VBOXSTRICTRC) vmR3Reset(PVM pVM, PVMCPU pVCpu, void *pvUser)
2652	{
2653	Assert(!pvUser); NOREF(pvUser);
2654
2655	/*
2656	* The first EMT will try change the state to resetting. If this fails,
2657	* we won't get called for the other EMTs.
2658	*/
2659	if (pVCpu->idCpu == pVM->cCpus - 1)
2660	{
2661	int rc = vmR3TrySetState(pVM, "VMR3Reset", 3,
2662	VMSTATE_RESETTING, VMSTATE_RUNNING,
2663	VMSTATE_RESETTING, VMSTATE_SUSPENDED,
2664	VMSTATE_RESETTING_LS, VMSTATE_RUNNING_LS);
2665	if (RT_FAILURE(rc))
2666	return rc;
2667	}
2668
2669	/*
2670	* Check the state.
2671	*/
2672	VMSTATE enmVMState = VMR3GetState(pVM);
2673	AssertLogRelMsgReturn( enmVMState == VMSTATE_RESETTING
2674	\|\| enmVMState == VMSTATE_RESETTING_LS,
2675	("%s\n", VMR3GetStateName(enmVMState)),
2676	VERR_INTERNAL_ERROR_4);
2677
2678	/*
2679	* EMT(0) does the full cleanup after all the other EMTs has been
2680	* thru here and been told to enter the EMSTATE_WAIT_SIPI state.
2681	*
2682	* Because there are per-cpu reset routines and order may/is important,
2683	* the following sequence looks a bit ugly...
2684	*/
2685	if (pVCpu->idCpu == 0)
2686	vmR3CheckIntegrity(pVM);
2687
2688	/* Reset the VCpu state. */
2689	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED);
2690
2691	/* Clear all pending forced actions. */
2692	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_ALL_MASK & ~VMCPU_FF_REQUEST);
2693
2694	/*
2695	* Reset the VM components.
2696	*/
2697	if (pVCpu->idCpu == 0)
2698	{
2699	PATMR3Reset(pVM);
2700	CSAMR3Reset(pVM);
2701	PGMR3Reset(pVM); /* We clear VM RAM in PGMR3Reset. It's vital PDMR3Reset is executed
2702	* _afterwards_. E.g. ACPI sets up RAM tables during init/reset. */
2703	/** @todo PGMR3Reset should be called after PDMR3Reset really, because we'll trash OS <-> hardware
2704	* communication structures residing in RAM when done in the other order. I.e. the device must be
2705	* quiesced first, then we clear the memory and plan tables. Probably have to make these things
2706	* explicit in some way, some memory setup pass or something.
2707	* (Example: DevAHCI may assert if memory is zeroed before it've read the FIS.)
2708	*
2709	* @bugref{4467}
2710	*/
2711	MMR3Reset(pVM);
2712	PDMR3Reset(pVM);
2713	SELMR3Reset(pVM);
2714	TRPMR3Reset(pVM);
2715	REMR3Reset(pVM);
2716	IOMR3Reset(pVM);
2717	CPUMR3Reset(pVM);
2718	}
2719	CPUMR3ResetCpu(pVCpu);
2720	if (pVCpu->idCpu == 0)
2721	{
2722	TMR3Reset(pVM);
2723	EMR3Reset(pVM);
2724	HWACCMR3Reset(pVM); /* This must come after PATM, CSAM, CPUM, SELM and TRPM. */
2725
2726	#ifdef LOG_ENABLED
2727	/*
2728	* Debug logging.
2729	*/
2730	RTLogPrintf("\n\nThe VM was reset:\n");
2731	DBGFR3Info(pVM, "cpum", "verbose", NULL);
2732	#endif
2733
2734	/*
2735	* Since EMT(0) is the last to go thru here, it will advance the state.
2736	* When a live save is active, we will move on to SuspendingLS but
2737	* leave it for VMR3Reset to do the actual suspending due to deadlock risks.
2738	*/
2739	PUVM pUVM = pVM->pUVM;
2740	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
2741	enmVMState = pVM->enmVMState;
2742	if (enmVMState == VMSTATE_RESETTING)
2743	{
2744	if (pUVM->vm.s.enmPrevVMState == VMSTATE_SUSPENDED)
2745	vmR3SetStateLocked(pVM, pUVM, VMSTATE_SUSPENDED, VMSTATE_RESETTING);
2746	else
2747	vmR3SetStateLocked(pVM, pUVM, VMSTATE_RUNNING, VMSTATE_RESETTING);
2748	}
2749	else
2750	vmR3SetStateLocked(pVM, pUVM, VMSTATE_SUSPENDING_LS, VMSTATE_RESETTING_LS);
2751	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
2752
2753	vmR3CheckIntegrity(pVM);
2754
2755	/*
2756	* Do the suspend bit as well.
2757	* It only requires some EMT(0) work at present.
2758	*/
2759	if (enmVMState != VMSTATE_RESETTING)
2760	{
2761	vmR3SuspendDoWork(pVM);
2762	vmR3SetState(pVM, VMSTATE_SUSPENDED_LS, VMSTATE_SUSPENDING_LS);
2763	}
2764	}
2765
2766	return enmVMState == VMSTATE_RESETTING
2767	? VINF_EM_RESET
2768	: VINF_EM_SUSPEND; /** @todo VINF_EM_SUSPEND has lower priority than VINF_EM_RESET, so fix races. Perhaps add a new code for this combined case. */
2769	}
2770
2771
2772	/**
2773	* Reset the current VM.
2774	*
2775	* @returns VBox status code.
2776	* @param pVM VM to reset.
2777	*/
2778	VMMR3DECL(int) VMR3Reset(PVM pVM)
2779	{
2780	LogFlow(("VMR3Reset:\n"));
2781	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
2782
2783	/*
2784	* Gather all the EMTs to make sure there are no races before
2785	* changing the VM state.
2786	*/
2787	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
2788	vmR3Reset, NULL);
2789	LogFlow(("VMR3Reset: returns %Rrc\n", rc));
2790	return rc;
2791	}
2792
2793
2794	/**
2795	* Gets the current VM state.
2796	*
2797	* @returns The current VM state.
2798	* @param pVM VM handle.
2799	* @thread Any
2800	*/
2801	VMMR3DECL(VMSTATE) VMR3GetState(PVM pVM)
2802	{
2803	return pVM->enmVMState;
2804	}
2805
2806
2807	/**
2808	* Gets the state name string for a VM state.
2809	*
2810	* @returns Pointer to the state name. (readonly)
2811	* @param enmState The state.
2812	*/
2813	VMMR3DECL(const char *) VMR3GetStateName(VMSTATE enmState)
2814	{
2815	switch (enmState)
2816	{
2817	case VMSTATE_CREATING: return "CREATING";
2818	case VMSTATE_CREATED: return "CREATED";
2819	case VMSTATE_LOADING: return "LOADING";
2820	case VMSTATE_POWERING_ON: return "POWERING_ON";
2821	case VMSTATE_RESUMING: return "RESUMING";
2822	case VMSTATE_RUNNING: return "RUNNING";
2823	case VMSTATE_RUNNING_LS: return "RUNNING_LS";
2824	case VMSTATE_RESETTING: return "RESETTING";
2825	case VMSTATE_RESETTING_LS: return "RESETTING_LS";
2826	case VMSTATE_SUSPENDED: return "SUSPENDED";
2827	case VMSTATE_SUSPENDED_LS: return "SUSPENDED_LS";
2828	case VMSTATE_SUSPENDED_EXT_LS: return "SUSPENDED_EXT_LS";
2829	case VMSTATE_SUSPENDING: return "SUSPENDING";
2830	case VMSTATE_SUSPENDING_LS: return "SUSPENDING_LS";
2831	case VMSTATE_SUSPENDING_EXT_LS: return "SUSPENDING_EXT_LS";
2832	case VMSTATE_SAVING: return "SAVING";
2833	case VMSTATE_DEBUGGING: return "DEBUGGING";
2834	case VMSTATE_DEBUGGING_LS: return "DEBUGGING_LS";
2835	case VMSTATE_POWERING_OFF: return "POWERING_OFF";
2836	case VMSTATE_POWERING_OFF_LS: return "POWERING_OFF_LS";
2837	case VMSTATE_FATAL_ERROR: return "FATAL_ERROR";
2838	case VMSTATE_FATAL_ERROR_LS: return "FATAL_ERROR_LS";
2839	case VMSTATE_GURU_MEDITATION: return "GURU_MEDITATION";
2840	case VMSTATE_GURU_MEDITATION_LS:return "GURU_MEDITATION_LS";
2841	case VMSTATE_LOAD_FAILURE: return "LOAD_FAILURE";
2842	case VMSTATE_OFF: return "OFF";
2843	case VMSTATE_OFF_LS: return "OFF_LS";
2844	case VMSTATE_DESTROYING: return "DESTROYING";
2845	case VMSTATE_TERMINATED: return "TERMINATED";
2846
2847	default:
2848	AssertMsgFailed(("Unknown state %d\n", enmState));
2849	return "Unknown!\n";
2850	}
2851	}
2852
2853
2854	/**
2855	* Validates the state transition in strict builds.
2856	*
2857	* @returns true if valid, false if not.
2858	*
2859	* @param enmStateOld The old (current) state.
2860	* @param enmStateNew The proposed new state.
2861	*
2862	* @remarks The reference for this is found in doc/vp/VMM.vpp, the VMSTATE
2863	* diagram (under State Machine Diagram).
2864	*/
2865	static bool vmR3ValidateStateTransition(VMSTATE enmStateOld, VMSTATE enmStateNew)
2866	{
2867	#ifdef VBOX_STRICT
2868	switch (enmStateOld)
2869	{
2870	case VMSTATE_CREATING:
2871	AssertMsgReturn(enmStateNew == VMSTATE_CREATED, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2872	break;
2873
2874	case VMSTATE_CREATED:
2875	AssertMsgReturn( enmStateNew == VMSTATE_LOADING
2876	\|\| enmStateNew == VMSTATE_POWERING_ON
2877	\|\| enmStateNew == VMSTATE_POWERING_OFF
2878	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2879	break;
2880
2881	case VMSTATE_LOADING:
2882	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDED
2883	\|\| enmStateNew == VMSTATE_LOAD_FAILURE
2884	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2885	break;
2886
2887	case VMSTATE_POWERING_ON:
2888	AssertMsgReturn( enmStateNew == VMSTATE_RUNNING
2889	/\|\| enmStateNew == VMSTATE_FATAL_ERROR ?/
2890	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2891	break;
2892
2893	case VMSTATE_RESUMING:
2894	AssertMsgReturn( enmStateNew == VMSTATE_RUNNING
2895	/\|\| enmStateNew == VMSTATE_FATAL_ERROR ?/
2896	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2897	break;
2898
2899	case VMSTATE_RUNNING:
2900	AssertMsgReturn( enmStateNew == VMSTATE_POWERING_OFF
2901	\|\| enmStateNew == VMSTATE_SUSPENDING
2902	\|\| enmStateNew == VMSTATE_RESETTING
2903	\|\| enmStateNew == VMSTATE_RUNNING_LS
2904	\|\| enmStateNew == VMSTATE_DEBUGGING
2905	\|\| enmStateNew == VMSTATE_FATAL_ERROR
2906	\|\| enmStateNew == VMSTATE_GURU_MEDITATION
2907	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2908	break;
2909
2910	case VMSTATE_RUNNING_LS:
2911	AssertMsgReturn( enmStateNew == VMSTATE_POWERING_OFF_LS
2912	\|\| enmStateNew == VMSTATE_SUSPENDING_LS
2913	\|\| enmStateNew == VMSTATE_SUSPENDING_EXT_LS
2914	\|\| enmStateNew == VMSTATE_RESETTING_LS
2915	\|\| enmStateNew == VMSTATE_RUNNING
2916	\|\| enmStateNew == VMSTATE_DEBUGGING_LS
2917	\|\| enmStateNew == VMSTATE_FATAL_ERROR_LS
2918	\|\| enmStateNew == VMSTATE_GURU_MEDITATION_LS
2919	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2920	break;
2921
2922	case VMSTATE_RESETTING:
2923	AssertMsgReturn(enmStateNew == VMSTATE_RUNNING, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2924	break;
2925
2926	case VMSTATE_RESETTING_LS:
2927	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDING_LS
2928	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2929	break;
2930
2931	case VMSTATE_SUSPENDING:
2932	AssertMsgReturn(enmStateNew == VMSTATE_SUSPENDED, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2933	break;
2934
2935	case VMSTATE_SUSPENDING_LS:
2936	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDING
2937	\|\| enmStateNew == VMSTATE_SUSPENDED_LS
2938	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2939	break;
2940
2941	case VMSTATE_SUSPENDING_EXT_LS:
2942	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDING
2943	\|\| enmStateNew == VMSTATE_SUSPENDED_EXT_LS
2944	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2945	break;
2946
2947	case VMSTATE_SUSPENDED:
2948	AssertMsgReturn( enmStateNew == VMSTATE_POWERING_OFF
2949	\|\| enmStateNew == VMSTATE_SAVING
2950	\|\| enmStateNew == VMSTATE_RESETTING
2951	\|\| enmStateNew == VMSTATE_RESUMING
2952	\|\| enmStateNew == VMSTATE_LOADING
2953	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2954	break;
2955
2956	case VMSTATE_SUSPENDED_LS:
2957	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDED
2958	\|\| enmStateNew == VMSTATE_SAVING
2959	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2960	break;
2961
2962	case VMSTATE_SUSPENDED_EXT_LS:
2963	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDED
2964	\|\| enmStateNew == VMSTATE_SAVING
2965	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2966	break;
2967
2968	case VMSTATE_SAVING:
2969	AssertMsgReturn(enmStateNew == VMSTATE_SUSPENDED, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2970	break;
2971
2972	case VMSTATE_DEBUGGING:
2973	AssertMsgReturn( enmStateNew == VMSTATE_RUNNING
2974	\|\| enmStateNew == VMSTATE_POWERING_OFF
2975	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2976	break;
2977
2978	case VMSTATE_DEBUGGING_LS:
2979	AssertMsgReturn( enmStateNew == VMSTATE_DEBUGGING
2980	\|\| enmStateNew == VMSTATE_RUNNING_LS
2981	\|\| enmStateNew == VMSTATE_POWERING_OFF_LS
2982	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2983	break;
2984
2985	case VMSTATE_POWERING_OFF:
2986	AssertMsgReturn(enmStateNew == VMSTATE_OFF, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2987	break;
2988
2989	case VMSTATE_POWERING_OFF_LS:
2990	AssertMsgReturn( enmStateNew == VMSTATE_POWERING_OFF
2991	\|\| enmStateNew == VMSTATE_OFF_LS
2992	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2993	break;
2994
2995	case VMSTATE_OFF:
2996	AssertMsgReturn(enmStateNew == VMSTATE_DESTROYING, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2997	break;
2998
2999	case VMSTATE_OFF_LS:
3000	AssertMsgReturn(enmStateNew == VMSTATE_OFF, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3001	break;
3002
3003	case VMSTATE_FATAL_ERROR:
3004	AssertMsgReturn(enmStateNew == VMSTATE_POWERING_OFF, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3005	break;
3006
3007	case VMSTATE_FATAL_ERROR_LS:
3008	AssertMsgReturn( enmStateNew == VMSTATE_FATAL_ERROR
3009	\|\| enmStateNew == VMSTATE_POWERING_OFF_LS
3010	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3011	break;
3012
3013	case VMSTATE_GURU_MEDITATION:
3014	AssertMsgReturn( enmStateNew == VMSTATE_DEBUGGING
3015	\|\| enmStateNew == VMSTATE_POWERING_OFF
3016	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3017	break;
3018
3019	case VMSTATE_GURU_MEDITATION_LS:
3020	AssertMsgReturn( enmStateNew == VMSTATE_GURU_MEDITATION
3021	\|\| enmStateNew == VMSTATE_DEBUGGING_LS
3022	\|\| enmStateNew == VMSTATE_POWERING_OFF_LS
3023	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3024	break;
3025
3026	case VMSTATE_LOAD_FAILURE:
3027	AssertMsgReturn(enmStateNew == VMSTATE_POWERING_OFF, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3028	break;
3029
3030	case VMSTATE_DESTROYING:
3031	AssertMsgReturn(enmStateNew == VMSTATE_TERMINATED, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3032	break;
3033
3034	case VMSTATE_TERMINATED:
3035	default:
3036	AssertMsgFailedReturn(("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3037	break;
3038	}
3039	#endif /* VBOX_STRICT */
3040	return true;
3041	}
3042
3043
3044	/**
3045	* Does the state change callouts.
3046	*
3047	* The caller owns the AtStateCritSect.
3048	*
3049	* @param pVM The VM handle.
3050	* @param pUVM The UVM handle.
3051	* @param enmStateNew The New state.
3052	* @param enmStateOld The old state.
3053	*/
3054	static void vmR3DoAtState(PVM pVM, PUVM pUVM, VMSTATE enmStateNew, VMSTATE enmStateOld)
3055	{
3056	LogRel(("Changing the VM state from '%s' to '%s'.\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)));
3057
3058	for (PVMATSTATE pCur = pUVM->vm.s.pAtState; pCur; pCur = pCur->pNext)
3059	{
3060	pCur->pfnAtState(pVM, enmStateNew, enmStateOld, pCur->pvUser);
3061	if ( enmStateNew != VMSTATE_DESTROYING
3062	&& pVM->enmVMState == VMSTATE_DESTROYING)
3063	break;
3064	AssertMsg(pVM->enmVMState == enmStateNew,
3065	("You are not allowed to change the state while in the change callback, except "
3066	"from destroying the VM. There are restrictions in the way the state changes "
3067	"are propagated up to the EM execution loop and it makes the program flow very "
3068	"difficult to follow. (%s, expected %s, old %s)\n",
3069	VMR3GetStateName(pVM->enmVMState), VMR3GetStateName(enmStateNew),
3070	VMR3GetStateName(enmStateOld)));
3071	}
3072	}
3073
3074
3075	/**
3076	* Sets the current VM state, with the AtStatCritSect already entered.
3077	*
3078	* @param pVM The VM handle.
3079	* @param pUVM The UVM handle.
3080	* @param enmStateNew The new state.
3081	* @param enmStateOld The old state.
3082	*/
3083	static void vmR3SetStateLocked(PVM pVM, PUVM pUVM, VMSTATE enmStateNew, VMSTATE enmStateOld)
3084	{
3085	vmR3ValidateStateTransition(enmStateOld, enmStateNew);
3086
3087	AssertMsg(pVM->enmVMState == enmStateOld,
3088	("%s != %s\n", VMR3GetStateName(pVM->enmVMState), VMR3GetStateName(enmStateOld)));
3089	pUVM->vm.s.enmPrevVMState = enmStateOld;
3090	pVM->enmVMState = enmStateNew;
3091
3092	vmR3DoAtState(pVM, pUVM, enmStateNew, enmStateOld);
3093	}
3094
3095
3096	/**
3097	* Sets the current VM state.
3098	*
3099	* @param pVM VM handle.
3100	* @param enmStateNew The new state.
3101	* @param enmStateOld The old state (for asserting only).
3102	*/
3103	static void vmR3SetState(PVM pVM, VMSTATE enmStateNew, VMSTATE enmStateOld)
3104	{
3105	PUVM pUVM = pVM->pUVM;
3106	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
3107
3108	AssertMsg(pVM->enmVMState == enmStateOld,
3109	("%s != %s\n", VMR3GetStateName(pVM->enmVMState), VMR3GetStateName(enmStateOld)));
3110	vmR3SetStateLocked(pVM, pUVM, enmStateNew, pVM->enmVMState);
3111
3112	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3113	}
3114
3115
3116	/**
3117	* Tries to perform a state transition.
3118	*
3119	* @returns The 1-based ordinal of the succeeding transition.
3120	* VERR_VM_INVALID_VM_STATE and Assert+LogRel on failure.
3121	*
3122	* @param pVM The VM handle.
3123	* @param pszWho Who is trying to change it.
3124	* @param cTransitions The number of transitions in the ellipsis.
3125	* @param ... Transition pairs; new, old.
3126	*/
3127	static int vmR3TrySetState(PVM pVM, const char *pszWho, unsigned cTransitions, ...)
3128	{
3129	va_list va;
3130	VMSTATE enmStateNew = VMSTATE_CREATED;
3131	VMSTATE enmStateOld = VMSTATE_CREATED;
3132
3133	#ifdef VBOX_STRICT
3134	/*
3135	* Validate the input first.
3136	*/
3137	va_start(va, cTransitions);
3138	for (unsigned i = 0; i < cTransitions; i++)
3139	{
3140	enmStateNew = (VMSTATE)va_arg(va, /VMSTATE/int);
3141	enmStateOld = (VMSTATE)va_arg(va, /VMSTATE/int);
3142	vmR3ValidateStateTransition(enmStateOld, enmStateNew);
3143	}
3144	va_end(va);
3145	#endif
3146
3147	/*
3148	* Grab the lock and see if any of the proposed transisions works out.
3149	*/
3150	va_start(va, cTransitions);
3151	int rc = VERR_VM_INVALID_VM_STATE;
3152	PUVM pUVM = pVM->pUVM;
3153	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
3154
3155	VMSTATE enmStateCur = pVM->enmVMState;
3156
3157	for (unsigned i = 0; i < cTransitions; i++)
3158	{
3159	enmStateNew = (VMSTATE)va_arg(va, /VMSTATE/int);
3160	enmStateOld = (VMSTATE)va_arg(va, /VMSTATE/int);
3161	if (enmStateCur == enmStateOld)
3162	{
3163	vmR3SetStateLocked(pVM, pUVM, enmStateNew, enmStateOld);
3164	rc = i + 1;
3165	break;
3166	}
3167	}
3168
3169	if (RT_FAILURE(rc))
3170	{
3171	/*
3172	* Complain about it.
3173	*/
3174	if (cTransitions == 1)
3175	{
3176	LogRel(("%s: %s -> %s failed, because the VM state is actually %s\n",
3177	pszWho, VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew), VMR3GetStateName(enmStateCur)));
3178	VMSetError(pVM, VERR_VM_INVALID_VM_STATE, RT_SRC_POS,
3179	N_("%s failed because the VM state is %s instead of %s"),
3180	pszWho, VMR3GetStateName(enmStateCur), VMR3GetStateName(enmStateOld));
3181	AssertMsgFailed(("%s: %s -> %s failed, because the VM state is actually %s\n",
3182	pszWho, VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew), VMR3GetStateName(enmStateCur)));
3183	}
3184	else
3185	{
3186	va_end(va);
3187	va_start(va, cTransitions);
3188	LogRel(("%s:\n", pszWho));
3189	for (unsigned i = 0; i < cTransitions; i++)
3190	{
3191	enmStateNew = (VMSTATE)va_arg(va, /VMSTATE/int);
3192	enmStateOld = (VMSTATE)va_arg(va, /VMSTATE/int);
3193	LogRel(("%s%s -> %s",
3194	i ? ", " : " ", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)));
3195	}
3196	LogRel((" failed, because the VM state is actually %s\n", VMR3GetStateName(enmStateCur)));
3197	VMSetError(pVM, VERR_VM_INVALID_VM_STATE, RT_SRC_POS,
3198	N_("%s failed because the current VM state, %s, was not found in the state transition table"),
3199	pszWho, VMR3GetStateName(enmStateCur), VMR3GetStateName(enmStateOld));
3200	AssertMsgFailed(("%s - state=%s, see release log for full details. Check the cTransitions passed us.\n",
3201	pszWho, VMR3GetStateName(enmStateCur)));
3202	}
3203	}
3204
3205	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3206	va_end(va);
3207	Assert(rc > 0 \|\| rc < 0);
3208	return rc;
3209	}
3210
3211
3212	/**
3213	* Flag a guru meditation ... a hack.
3214	*
3215	* @param pVM The VM handle
3216	*
3217	* @todo Rewrite this part. The guru meditation should be flagged
3218	* immediately by the VMM and not by VMEmt.cpp when it's all over.
3219	*/
3220	void vmR3SetGuruMeditation(PVM pVM)
3221	{
3222	PUVM pUVM = pVM->pUVM;
3223	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
3224
3225	VMSTATE enmStateCur = pVM->enmVMState;
3226	if (enmStateCur == VMSTATE_RUNNING)
3227	vmR3SetStateLocked(pVM, pUVM, VMSTATE_GURU_MEDITATION, VMSTATE_RUNNING);
3228	else if (enmStateCur == VMSTATE_RUNNING_LS)
3229	{
3230	vmR3SetStateLocked(pVM, pUVM, VMSTATE_GURU_MEDITATION_LS, VMSTATE_RUNNING_LS);
3231	SSMR3Cancel(pVM);
3232	}
3233
3234	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3235	}
3236
3237
3238	/**
3239	* Checks if the VM was teleported and hasn't been fully resumed yet.
3240	*
3241	* This applies to both sides of the teleportation since we may leave a working
3242	* clone behind and the user is allowed to resume this...
3243	*
3244	* @returns true / false.
3245	* @param pVM The VM handle.
3246	* @thread Any thread.
3247	*/
3248	VMMR3DECL(bool) VMR3TeleportedAndNotFullyResumedYet(PVM pVM)
3249	{
3250	VM_ASSERT_VALID_EXT_RETURN(pVM, false);
3251	return pVM->vm.s.fTeleportedAndNotFullyResumedYet;
3252	}
3253
3254
3255	/**
3256	* Registers a VM state change callback.
3257	*
3258	* You are not allowed to call any function which changes the VM state from a
3259	* state callback, except VMR3Destroy().
3260	*
3261	* @returns VBox status code.
3262	* @param pVM VM handle.
3263	* @param pfnAtState Pointer to callback.
3264	* @param pvUser User argument.
3265	* @thread Any.
3266	*/
3267	VMMR3DECL(int) VMR3AtStateRegister(PVM pVM, PFNVMATSTATE pfnAtState, void *pvUser)
3268	{
3269	LogFlow(("VMR3AtStateRegister: pfnAtState=%p pvUser=%p\n", pfnAtState, pvUser));
3270
3271	/*
3272	* Validate input.
3273	*/
3274	AssertPtrReturn(pfnAtState, VERR_INVALID_PARAMETER);
3275	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
3276
3277	/*
3278	* Allocate a new record.
3279	*/
3280	PUVM pUVM = pVM->pUVM;
3281	PVMATSTATE pNew = (PVMATSTATE)MMR3HeapAllocU(pUVM, MM_TAG_VM, sizeof(*pNew));
3282	if (!pNew)
3283	return VERR_NO_MEMORY;
3284
3285	/* fill */
3286	pNew->pfnAtState = pfnAtState;
3287	pNew->pvUser = pvUser;
3288
3289	/* insert */
3290	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
3291	pNew->pNext = *pUVM->vm.s.ppAtStateNext;
3292	*pUVM->vm.s.ppAtStateNext = pNew;
3293	pUVM->vm.s.ppAtStateNext = &pNew->pNext;
3294	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3295
3296	return VINF_SUCCESS;
3297	}
3298
3299
3300	/**
3301	* Deregisters a VM state change callback.
3302	*
3303	* @returns VBox status code.
3304	* @param pVM VM handle.
3305	* @param pfnAtState Pointer to callback.
3306	* @param pvUser User argument.
3307	* @thread Any.
3308	*/
3309	VMMR3DECL(int) VMR3AtStateDeregister(PVM pVM, PFNVMATSTATE pfnAtState, void *pvUser)
3310	{
3311	LogFlow(("VMR3AtStateDeregister: pfnAtState=%p pvUser=%p\n", pfnAtState, pvUser));
3312
3313	/*
3314	* Validate input.
3315	*/
3316	AssertPtrReturn(pfnAtState, VERR_INVALID_PARAMETER);
3317	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
3318
3319	PUVM pUVM = pVM->pUVM;
3320	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
3321
3322	/*
3323	* Search the list for the entry.
3324	*/
3325	PVMATSTATE pPrev = NULL;
3326	PVMATSTATE pCur = pUVM->vm.s.pAtState;
3327	while ( pCur
3328	&& ( pCur->pfnAtState != pfnAtState
3329	\|\| pCur->pvUser != pvUser))
3330	{
3331	pPrev = pCur;
3332	pCur = pCur->pNext;
3333	}
3334	if (!pCur)
3335	{
3336	AssertMsgFailed(("pfnAtState=%p was not found\n", pfnAtState));
3337	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3338	return VERR_FILE_NOT_FOUND;
3339	}
3340
3341	/*
3342	* Unlink it.
3343	*/
3344	if (pPrev)
3345	{
3346	pPrev->pNext = pCur->pNext;
3347	if (!pCur->pNext)
3348	pUVM->vm.s.ppAtStateNext = &pPrev->pNext;
3349	}
3350	else
3351	{
3352	pUVM->vm.s.pAtState = pCur->pNext;
3353	if (!pCur->pNext)
3354	pUVM->vm.s.ppAtStateNext = &pUVM->vm.s.pAtState;
3355	}
3356
3357	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3358
3359	/*
3360	* Free it.
3361	*/
3362	pCur->pfnAtState = NULL;
3363	pCur->pNext = NULL;
3364	MMR3HeapFree(pCur);
3365
3366	return VINF_SUCCESS;
3367	}
3368
3369
3370	/**
3371	* Registers a VM error callback.
3372	*
3373	* @returns VBox status code.
3374	* @param pVM The VM handle.
3375	* @param pfnAtError Pointer to callback.
3376	* @param pvUser User argument.
3377	* @thread Any.
3378	*/
3379	VMMR3DECL(int) VMR3AtErrorRegister(PVM pVM, PFNVMATERROR pfnAtError, void *pvUser)
3380	{
3381	return VMR3AtErrorRegisterU(pVM->pUVM, pfnAtError, pvUser);
3382	}
3383
3384
3385	/**
3386	* Registers a VM error callback.
3387	*
3388	* @returns VBox status code.
3389	* @param pUVM The VM handle.
3390	* @param pfnAtError Pointer to callback.
3391	* @param pvUser User argument.
3392	* @thread Any.
3393	*/
3394	VMMR3DECL(int) VMR3AtErrorRegisterU(PUVM pUVM, PFNVMATERROR pfnAtError, void *pvUser)
3395	{
3396	LogFlow(("VMR3AtErrorRegister: pfnAtError=%p pvUser=%p\n", pfnAtError, pvUser));
3397
3398	/*
3399	* Validate input.
3400	*/
3401	AssertPtrReturn(pfnAtError, VERR_INVALID_PARAMETER);
3402	UVM_ASSERT_VALID_EXT_RETURN(pUVM, VERR_INVALID_VM_HANDLE);
3403
3404	/*
3405	* Allocate a new record.
3406	*/
3407	PVMATERROR pNew = (PVMATERROR)MMR3HeapAllocU(pUVM, MM_TAG_VM, sizeof(*pNew));
3408	if (!pNew)
3409	return VERR_NO_MEMORY;
3410
3411	/* fill */
3412	pNew->pfnAtError = pfnAtError;
3413	pNew->pvUser = pvUser;
3414
3415	/* insert */
3416	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3417	pNew->pNext = *pUVM->vm.s.ppAtErrorNext;
3418	*pUVM->vm.s.ppAtErrorNext = pNew;
3419	pUVM->vm.s.ppAtErrorNext = &pNew->pNext;
3420	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3421
3422	return VINF_SUCCESS;
3423	}
3424
3425
3426	/**
3427	* Deregisters a VM error callback.
3428	*
3429	* @returns VBox status code.
3430	* @param pVM The VM handle.
3431	* @param pfnAtError Pointer to callback.
3432	* @param pvUser User argument.
3433	* @thread Any.
3434	*/
3435	VMMR3DECL(int) VMR3AtErrorDeregister(PVM pVM, PFNVMATERROR pfnAtError, void *pvUser)
3436	{
3437	LogFlow(("VMR3AtErrorDeregister: pfnAtError=%p pvUser=%p\n", pfnAtError, pvUser));
3438
3439	/*
3440	* Validate input.
3441	*/
3442	AssertPtrReturn(pfnAtError, VERR_INVALID_PARAMETER);
3443	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
3444
3445	PUVM pUVM = pVM->pUVM;
3446	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3447
3448	/*
3449	* Search the list for the entry.
3450	*/
3451	PVMATERROR pPrev = NULL;
3452	PVMATERROR pCur = pUVM->vm.s.pAtError;
3453	while ( pCur
3454	&& ( pCur->pfnAtError != pfnAtError
3455	\|\| pCur->pvUser != pvUser))
3456	{
3457	pPrev = pCur;
3458	pCur = pCur->pNext;
3459	}
3460	if (!pCur)
3461	{
3462	AssertMsgFailed(("pfnAtError=%p was not found\n", pfnAtError));
3463	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3464	return VERR_FILE_NOT_FOUND;
3465	}
3466
3467	/*
3468	* Unlink it.
3469	*/
3470	if (pPrev)
3471	{
3472	pPrev->pNext = pCur->pNext;
3473	if (!pCur->pNext)
3474	pUVM->vm.s.ppAtErrorNext = &pPrev->pNext;
3475	}
3476	else
3477	{
3478	pUVM->vm.s.pAtError = pCur->pNext;
3479	if (!pCur->pNext)
3480	pUVM->vm.s.ppAtErrorNext = &pUVM->vm.s.pAtError;
3481	}
3482
3483	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3484
3485	/*
3486	* Free it.
3487	*/
3488	pCur->pfnAtError = NULL;
3489	pCur->pNext = NULL;
3490	MMR3HeapFree(pCur);
3491
3492	return VINF_SUCCESS;
3493	}
3494
3495
3496	/**
3497	* Ellipsis to va_list wrapper for calling pfnAtError.
3498	*/
3499	static void vmR3SetErrorWorkerDoCall(PVM pVM, PVMATERROR pCur, int rc, RT_SRC_POS_DECL, const char *pszFormat, ...)
3500	{
3501	va_list va;
3502	va_start(va, pszFormat);
3503	pCur->pfnAtError(pVM, pCur->pvUser, rc, RT_SRC_POS_ARGS, pszFormat, va);
3504	va_end(va);
3505	}
3506
3507
3508	/**
3509	* This is a worker function for GC and Ring-0 calls to VMSetError and VMSetErrorV.
3510	* The message is found in VMINT.
3511	*
3512	* @param pVM The VM handle.
3513	* @thread EMT.
3514	*/
3515	VMMR3DECL(void) VMR3SetErrorWorker(PVM pVM)
3516	{
3517	VM_ASSERT_EMT(pVM);
3518	AssertReleaseMsgFailed(("And we have a winner! You get to implement Ring-0 and GC VMSetErrorV! Contrats!\n"));
3519
3520	/*
3521	* Unpack the error (if we managed to format one).
3522	*/
3523	PVMERROR pErr = pVM->vm.s.pErrorR3;
3524	const char *pszFile = NULL;
3525	const char *pszFunction = NULL;
3526	uint32_t iLine = 0;
3527	const char *pszMessage;
3528	int32_t rc = VERR_MM_HYPER_NO_MEMORY;
3529	if (pErr)
3530	{
3531	AssertCompile(sizeof(const char) == sizeof(uint8_t));
3532	if (pErr->offFile)
3533	pszFile = (const char *)pErr + pErr->offFile;
3534	iLine = pErr->iLine;
3535	if (pErr->offFunction)
3536	pszFunction = (const char *)pErr + pErr->offFunction;
3537	if (pErr->offMessage)
3538	pszMessage = (const char *)pErr + pErr->offMessage;
3539	else
3540	pszMessage = "No message!";
3541	}
3542	else
3543	pszMessage = "No message! (Failed to allocate memory to put the error message in!)";
3544
3545	/*
3546	* Call the at error callbacks.
3547	*/
3548	PUVM pUVM = pVM->pUVM;
3549	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3550	ASMAtomicIncU32(&pUVM->vm.s.cRuntimeErrors);
3551	for (PVMATERROR pCur = pUVM->vm.s.pAtError; pCur; pCur = pCur->pNext)
3552	vmR3SetErrorWorkerDoCall(pVM, pCur, rc, RT_SRC_POS_ARGS, "%s", pszMessage);
3553	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3554	}
3555
3556
3557	/**
3558	* Gets the number of errors raised via VMSetError.
3559	*
3560	* This can be used avoid double error messages.
3561	*
3562	* @returns The error count.
3563	* @param pVM The VM handle.
3564	*/
3565	VMMR3DECL(uint32_t) VMR3GetErrorCount(PVM pVM)
3566	{
3567	return pVM->pUVM->vm.s.cErrors;
3568	}
3569
3570
3571	/**
3572	* Creation time wrapper for vmR3SetErrorUV.
3573	*
3574	* @returns rc.
3575	* @param pUVM Pointer to the user mode VM structure.
3576	* @param rc The VBox status code.
3577	* @param RT_SRC_POS_DECL The source position of this error.
3578	* @param pszFormat Format string.
3579	* @param ... The arguments.
3580	* @thread Any thread.
3581	*/
3582	static int vmR3SetErrorU(PUVM pUVM, int rc, RT_SRC_POS_DECL, const char *pszFormat, ...)
3583	{
3584	va_list va;
3585	va_start(va, pszFormat);
3586	vmR3SetErrorUV(pUVM, rc, pszFile, iLine, pszFunction, pszFormat, &va);
3587	va_end(va);
3588	return rc;
3589	}
3590
3591
3592	/**
3593	* Worker which calls everyone listening to the VM error messages.
3594	*
3595	* @param pUVM Pointer to the user mode VM structure.
3596	* @param rc The VBox status code.
3597	* @param RT_SRC_POS_DECL The source position of this error.
3598	* @param pszFormat Format string.
3599	* @param pArgs Pointer to the format arguments.
3600	* @thread EMT
3601	*/
3602	DECLCALLBACK(void) vmR3SetErrorUV(PUVM pUVM, int rc, RT_SRC_POS_DECL, const char pszFormat, va_list pArgs)
3603	{
3604	/*
3605	* Log the error.
3606	*/
3607	va_list va3;
3608	va_copy(va3, *pArgs);
3609	RTLogRelPrintf("VMSetError: %s(%d) %s; rc=%Rrc\n"
3610	"VMSetError: %N\n",
3611	pszFile, iLine, pszFunction, rc,
3612	pszFormat, &va3);
3613	va_end(va3);
3614
3615	#ifdef LOG_ENABLED
3616	va_copy(va3, *pArgs);
3617	RTLogPrintf("VMSetError: %s(%d) %s; rc=%Rrc\n"
3618	"%N\n",
3619	pszFile, iLine, pszFunction, rc,
3620	pszFormat, &va3);
3621	va_end(va3);
3622	#endif
3623
3624	/*
3625	* Make a copy of the message.
3626	*/
3627	if (pUVM->pVM)
3628	vmSetErrorCopy(pUVM->pVM, rc, RT_SRC_POS_ARGS, pszFormat, *pArgs);
3629
3630	/*
3631	* Call the at error callbacks.
3632	*/
3633	bool fCalledSomeone = false;
3634	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3635	ASMAtomicIncU32(&pUVM->vm.s.cErrors);
3636	for (PVMATERROR pCur = pUVM->vm.s.pAtError; pCur; pCur = pCur->pNext)
3637	{
3638	va_list va2;
3639	va_copy(va2, *pArgs);
3640	pCur->pfnAtError(pUVM->pVM, pCur->pvUser, rc, RT_SRC_POS_ARGS, pszFormat, va2);
3641	va_end(va2);
3642	fCalledSomeone = true;
3643	}
3644	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3645	}
3646
3647
3648	/**
3649	* Registers a VM runtime error callback.
3650	*
3651	* @returns VBox status code.
3652	* @param pVM The VM handle.
3653	* @param pfnAtRuntimeError Pointer to callback.
3654	* @param pvUser User argument.
3655	* @thread Any.
3656	*/
3657	VMMR3DECL(int) VMR3AtRuntimeErrorRegister(PVM pVM, PFNVMATRUNTIMEERROR pfnAtRuntimeError, void *pvUser)
3658	{
3659	LogFlow(("VMR3AtRuntimeErrorRegister: pfnAtRuntimeError=%p pvUser=%p\n", pfnAtRuntimeError, pvUser));
3660
3661	/*
3662	* Validate input.
3663	*/
3664	AssertPtrReturn(pfnAtRuntimeError, VERR_INVALID_PARAMETER);
3665	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
3666
3667	/*
3668	* Allocate a new record.
3669	*/
3670	PUVM pUVM = pVM->pUVM;
3671	PVMATRUNTIMEERROR pNew = (PVMATRUNTIMEERROR)MMR3HeapAllocU(pUVM, MM_TAG_VM, sizeof(*pNew));
3672	if (!pNew)
3673	return VERR_NO_MEMORY;
3674
3675	/* fill */
3676	pNew->pfnAtRuntimeError = pfnAtRuntimeError;
3677	pNew->pvUser = pvUser;
3678
3679	/* insert */
3680	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3681	pNew->pNext = *pUVM->vm.s.ppAtRuntimeErrorNext;
3682	*pUVM->vm.s.ppAtRuntimeErrorNext = pNew;
3683	pUVM->vm.s.ppAtRuntimeErrorNext = &pNew->pNext;
3684	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3685
3686	return VINF_SUCCESS;
3687	}
3688
3689
3690	/**
3691	* Deregisters a VM runtime error callback.
3692	*
3693	* @returns VBox status code.
3694	* @param pVM The VM handle.
3695	* @param pfnAtRuntimeError Pointer to callback.
3696	* @param pvUser User argument.
3697	* @thread Any.
3698	*/
3699	VMMR3DECL(int) VMR3AtRuntimeErrorDeregister(PVM pVM, PFNVMATRUNTIMEERROR pfnAtRuntimeError, void *pvUser)
3700	{
3701	LogFlow(("VMR3AtRuntimeErrorDeregister: pfnAtRuntimeError=%p pvUser=%p\n", pfnAtRuntimeError, pvUser));
3702
3703	/*
3704	* Validate input.
3705	*/
3706	AssertPtrReturn(pfnAtRuntimeError, VERR_INVALID_PARAMETER);
3707	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
3708
3709	PUVM pUVM = pVM->pUVM;
3710	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3711
3712	/*
3713	* Search the list for the entry.
3714	*/
3715	PVMATRUNTIMEERROR pPrev = NULL;
3716	PVMATRUNTIMEERROR pCur = pUVM->vm.s.pAtRuntimeError;
3717	while ( pCur
3718	&& ( pCur->pfnAtRuntimeError != pfnAtRuntimeError
3719	\|\| pCur->pvUser != pvUser))
3720	{
3721	pPrev = pCur;
3722	pCur = pCur->pNext;
3723	}
3724	if (!pCur)
3725	{
3726	AssertMsgFailed(("pfnAtRuntimeError=%p was not found\n", pfnAtRuntimeError));
3727	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3728	return VERR_FILE_NOT_FOUND;
3729	}
3730
3731	/*
3732	* Unlink it.
3733	*/
3734	if (pPrev)
3735	{
3736	pPrev->pNext = pCur->pNext;
3737	if (!pCur->pNext)
3738	pUVM->vm.s.ppAtRuntimeErrorNext = &pPrev->pNext;
3739	}
3740	else
3741	{
3742	pUVM->vm.s.pAtRuntimeError = pCur->pNext;
3743	if (!pCur->pNext)
3744	pUVM->vm.s.ppAtRuntimeErrorNext = &pUVM->vm.s.pAtRuntimeError;
3745	}
3746
3747	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3748
3749	/*
3750	* Free it.
3751	*/
3752	pCur->pfnAtRuntimeError = NULL;
3753	pCur->pNext = NULL;
3754	MMR3HeapFree(pCur);
3755
3756	return VINF_SUCCESS;
3757	}
3758
3759
3760	/**
3761	* EMT rendezvous worker that vmR3SetRuntimeErrorCommon uses to safely change
3762	* the state to FatalError(LS).
3763	*
3764	* @returns VERR_VM_INVALID_VM_STATE or VINF_SUCCESS. (This is a strict return
3765	* code, see FNVMMEMTRENDEZVOUS.)
3766	*
3767	* @param pVM The VM handle.
3768	* @param pVCpu The VMCPU handle of the EMT.
3769	* @param pvUser Ignored.
3770	*/
3771	static DECLCALLBACK(VBOXSTRICTRC) vmR3SetRuntimeErrorChangeState(PVM pVM, PVMCPU pVCpu, void *pvUser)
3772	{
3773	NOREF(pVCpu);
3774	Assert(!pvUser); NOREF(pvUser);
3775
3776	int rc = vmR3TrySetState(pVM, "VMSetRuntimeError", 2,
3777	VMSTATE_FATAL_ERROR, VMSTATE_RUNNING,
3778	VMSTATE_FATAL_ERROR_LS, VMSTATE_RUNNING_LS);
3779	if (rc == 2)
3780	SSMR3Cancel(pVM);
3781	return RT_SUCCESS(rc) ? VINF_SUCCESS : rc;
3782	}
3783
3784
3785	/**
3786	* Worker for VMR3SetRuntimeErrorWorker and vmR3SetRuntimeErrorV.
3787	*
3788	* This does the common parts after the error has been saved / retrieved.
3789	*
3790	* @returns VBox status code with modifications, see VMSetRuntimeErrorV.
3791	*
3792	* @param pVM The VM handle.
3793	* @param fFlags The error flags.
3794	* @param pszErrorId Error ID string.
3795	* @param pszFormat Format string.
3796	* @param pVa Pointer to the format arguments.
3797	*/
3798	static int vmR3SetRuntimeErrorCommon(PVM pVM, uint32_t fFlags, const char pszErrorId, const char pszFormat, va_list *pVa)
3799	{
3800	LogRel(("VM: Raising runtime error '%s' (fFlags=%#x)\n", pszErrorId, fFlags));
3801
3802	/*
3803	* Take actions before the call.
3804	*/
3805	int rc;
3806	if (fFlags & VMSETRTERR_FLAGS_FATAL)
3807	rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, vmR3SetRuntimeErrorChangeState, NULL);
3808	else if (fFlags & VMSETRTERR_FLAGS_SUSPEND)
3809	rc = VMR3Suspend(pVM);
3810	else
3811	rc = VINF_SUCCESS;
3812
3813	/*
3814	* Do the callback round.
3815	*/
3816	PUVM pUVM = pVM->pUVM;
3817	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3818	ASMAtomicIncU32(&pUVM->vm.s.cRuntimeErrors);
3819	for (PVMATRUNTIMEERROR pCur = pUVM->vm.s.pAtRuntimeError; pCur; pCur = pCur->pNext)
3820	{
3821	va_list va;
3822	va_copy(va, *pVa);
3823	pCur->pfnAtRuntimeError(pVM, pCur->pvUser, fFlags, pszErrorId, pszFormat, va);
3824	va_end(va);
3825	}
3826	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3827
3828	return rc;
3829	}
3830
3831
3832	/**
3833	* Ellipsis to va_list wrapper for calling vmR3SetRuntimeErrorCommon.
3834	*/
3835	static int vmR3SetRuntimeErrorCommonF(PVM pVM, uint32_t fFlags, const char pszErrorId, const char pszFormat, ...)
3836	{
3837	va_list va;
3838	va_start(va, pszFormat);
3839	int rc = vmR3SetRuntimeErrorCommon(pVM, fFlags, pszErrorId, pszFormat, &va);
3840	va_end(va);
3841	return rc;
3842	}
3843
3844
3845	/**
3846	* This is a worker function for RC and Ring-0 calls to VMSetError and
3847	* VMSetErrorV.
3848	*
3849	* The message is found in VMINT.
3850	*
3851	* @returns VBox status code, see VMSetRuntimeError.
3852	* @param pVM The VM handle.
3853	* @thread EMT.
3854	*/
3855	VMMR3DECL(int) VMR3SetRuntimeErrorWorker(PVM pVM)
3856	{
3857	VM_ASSERT_EMT(pVM);
3858	AssertReleaseMsgFailed(("And we have a winner! You get to implement Ring-0 and GC VMSetRuntimeErrorV! Congrats!\n"));
3859
3860	/*
3861	* Unpack the error (if we managed to format one).
3862	*/
3863	const char *pszErrorId = "SetRuntimeError";
3864	const char *pszMessage = "No message!";
3865	uint32_t fFlags = VMSETRTERR_FLAGS_FATAL;
3866	PVMRUNTIMEERROR pErr = pVM->vm.s.pRuntimeErrorR3;
3867	if (pErr)
3868	{
3869	AssertCompile(sizeof(const char) == sizeof(uint8_t));
3870	if (pErr->offErrorId)
3871	pszErrorId = (const char *)pErr + pErr->offErrorId;
3872	if (pErr->offMessage)
3873	pszMessage = (const char *)pErr + pErr->offMessage;
3874	fFlags = pErr->fFlags;
3875	}
3876
3877	/*
3878	* Join cause with vmR3SetRuntimeErrorV.
3879	*/
3880	return vmR3SetRuntimeErrorCommonF(pVM, fFlags, pszErrorId, "%s", pszMessage);
3881	}
3882
3883
3884	/**
3885	* Worker for VMSetRuntimeErrorV for doing the job on EMT in ring-3.
3886	*
3887	* @returns VBox status code with modifications, see VMSetRuntimeErrorV.
3888	*
3889	* @param pVM The VM handle.
3890	* @param fFlags The error flags.
3891	* @param pszErrorId Error ID string.
3892	* @param pszMessage The error message residing the MM heap.
3893	*
3894	* @thread EMT
3895	*/
3896	DECLCALLBACK(int) vmR3SetRuntimeError(PVM pVM, uint32_t fFlags, const char pszErrorId, char pszMessage)
3897	{
3898	#if 0 /** @todo make copy of the error msg. */
3899	/*
3900	* Make a copy of the message.
3901	*/
3902	va_list va2;
3903	va_copy(va2, *pVa);
3904	vmSetRuntimeErrorCopy(pVM, fFlags, pszErrorId, pszFormat, va2);
3905	va_end(va2);
3906	#endif
3907
3908	/*
3909	* Join paths with VMR3SetRuntimeErrorWorker.
3910	*/
3911	int rc = vmR3SetRuntimeErrorCommonF(pVM, fFlags, pszErrorId, "%s", pszMessage);
3912	MMR3HeapFree(pszMessage);
3913	return rc;
3914	}
3915
3916
3917	/**
3918	* Worker for VMSetRuntimeErrorV for doing the job on EMT in ring-3.
3919	*
3920	* @returns VBox status code with modifications, see VMSetRuntimeErrorV.
3921	*
3922	* @param pVM The VM handle.
3923	* @param fFlags The error flags.
3924	* @param pszErrorId Error ID string.
3925	* @param pszFormat Format string.
3926	* @param pVa Pointer to the format arguments.
3927	*
3928	* @thread EMT
3929	*/
3930	DECLCALLBACK(int) vmR3SetRuntimeErrorV(PVM pVM, uint32_t fFlags, const char pszErrorId, const char pszFormat, va_list *pVa)
3931	{
3932	/*
3933	* Make a copy of the message.
3934	*/
3935	va_list va2;
3936	va_copy(va2, *pVa);
3937	vmSetRuntimeErrorCopy(pVM, fFlags, pszErrorId, pszFormat, va2);
3938	va_end(va2);
3939
3940	/*
3941	* Join paths with VMR3SetRuntimeErrorWorker.
3942	*/
3943	return vmR3SetRuntimeErrorCommon(pVM, fFlags, pszErrorId, pszFormat, pVa);
3944	}
3945
3946
3947	/**
3948	* Gets the number of runtime errors raised via VMR3SetRuntimeError.
3949	*
3950	* This can be used avoid double error messages.
3951	*
3952	* @returns The runtime error count.
3953	* @param pVM The VM handle.
3954	*/
3955	VMMR3DECL(uint32_t) VMR3GetRuntimeErrorCount(PVM pVM)
3956	{
3957	return pVM->pUVM->vm.s.cRuntimeErrors;
3958	}
3959
3960
3961	/**
3962	* Gets the ID virtual of the virtual CPU assoicated with the calling thread.
3963	*
3964	* @returns The CPU ID. NIL_VMCPUID if the thread isn't an EMT.
3965	*
3966	* @param pVM The VM handle.
3967	*/
3968	VMMR3DECL(RTCPUID) VMR3GetVMCPUId(PVM pVM)
3969	{
3970	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pVM->pUVM->vm.s.idxTLS);
3971	return pUVCpu
3972	? pUVCpu->idCpu
3973	: NIL_VMCPUID;
3974	}
3975
3976
3977	/**
3978	* Returns the native handle of the current EMT VMCPU thread.
3979	*
3980	* @returns Handle if this is an EMT thread; NIL_RTNATIVETHREAD otherwise
3981	* @param pVM The VM handle.
3982	* @thread EMT
3983	*/
3984	VMMR3DECL(RTNATIVETHREAD) VMR3GetVMCPUNativeThread(PVM pVM)
3985	{
3986	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pVM->pUVM->vm.s.idxTLS);
3987
3988	if (!pUVCpu)
3989	return NIL_RTNATIVETHREAD;
3990
3991	return pUVCpu->vm.s.NativeThreadEMT;
3992	}
3993
3994
3995	/**
3996	* Returns the native handle of the current EMT VMCPU thread.
3997	*
3998	* @returns Handle if this is an EMT thread; NIL_RTNATIVETHREAD otherwise
3999	* @param pVM The VM handle.
4000	* @thread EMT
4001	*/
4002	VMMR3DECL(RTNATIVETHREAD) VMR3GetVMCPUNativeThreadU(PUVM pUVM)
4003	{
4004	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pUVM->vm.s.idxTLS);
4005
4006	if (!pUVCpu)
4007	return NIL_RTNATIVETHREAD;
4008
4009	return pUVCpu->vm.s.NativeThreadEMT;
4010	}
4011
4012
4013	/**
4014	* Returns the handle of the current EMT VMCPU thread.
4015	*
4016	* @returns Handle if this is an EMT thread; NIL_RTNATIVETHREAD otherwise
4017	* @param pVM The VM handle.
4018	* @thread EMT
4019	*/
4020	VMMR3DECL(RTTHREAD) VMR3GetVMCPUThread(PVM pVM)
4021	{
4022	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pVM->pUVM->vm.s.idxTLS);
4023
4024	if (!pUVCpu)
4025	return NIL_RTTHREAD;
4026
4027	return pUVCpu->vm.s.ThreadEMT;
4028	}
4029
4030
4031	/**
4032	* Returns the handle of the current EMT VMCPU thread.
4033	*
4034	* @returns Handle if this is an EMT thread; NIL_RTNATIVETHREAD otherwise
4035	* @param pVM The VM handle.
4036	* @thread EMT
4037	*/
4038	VMMR3DECL(RTTHREAD) VMR3GetVMCPUThreadU(PUVM pUVM)
4039	{
4040	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pUVM->vm.s.idxTLS);
4041
4042	if (!pUVCpu)
4043	return NIL_RTTHREAD;
4044
4045	return pUVCpu->vm.s.ThreadEMT;
4046	}
4047
4048
4049	/**
4050	* Return the package and core id of a CPU.
4051	*
4052	* @returns VBOX status code.
4053	* @param pVM The VM to operate on.
4054	* @param idCpu Virtual CPU to get the ID from.
4055	* @param pidCpuCore Where to store the core ID of the virtual CPU.
4056	* @param pidCpuPackage Where to store the package ID of the virtual CPU.
4057	*
4058	*/
4059	VMMR3DECL(int) VMR3GetCpuCoreAndPackageIdFromCpuId(PVM pVM, VMCPUID idCpu, uint32_t pidCpuCore, uint32_t pidCpuPackage)
4060	{
4061	if (idCpu >= pVM->cCpus)
4062	return VERR_INVALID_CPU_ID;
4063
4064	#ifdef VBOX_WITH_MULTI_CORE
4065	*pidCpuCore = idCpu;
4066	*pidCpuPackage = 0;
4067	#else
4068	*pidCpuCore = 0;
4069	*pidCpuPackage = idCpu;
4070	#endif
4071
4072	return VINF_SUCCESS;
4073	}
4074
4075
4076	/**
4077	* Worker for VMR3HotUnplugCpu.
4078	*
4079	* @returns VINF_EM_WAIT_SPIP (strict status code).
4080	* @param pVM The VM handle.
4081	* @param idCpu The current CPU.
4082	*/
4083	static DECLCALLBACK(int) vmR3HotUnplugCpu(PVM pVM, VMCPUID idCpu)
4084	{
4085	PVMCPU pVCpu = VMMGetCpuById(pVM, idCpu);
4086	VMCPU_ASSERT_EMT(pVCpu);
4087
4088	/*
4089	* Reset per CPU resources.
4090	*
4091	* Actually only needed for VT-x because the CPU seems to be still in some
4092	* paged mode and startup fails after a new hot plug event. SVM works fine
4093	* even without this.
4094	*/
4095	Log(("vmR3HotUnplugCpu for VCPU %u\n", idCpu));
4096	PGMR3ResetUnpluggedCpu(pVM, pVCpu);
4097	PDMR3ResetCpu(pVCpu);
4098	TRPMR3ResetCpu(pVCpu);
4099	CPUMR3ResetCpu(pVCpu);
4100	EMR3ResetCpu(pVCpu);
4101	HWACCMR3ResetCpu(pVCpu);
4102	return VINF_EM_WAIT_SIPI;
4103	}
4104
4105
4106	/**
4107	* Hot-unplugs a CPU from the guest.
4108	*
4109	* @returns VBox status code.
4110	* @param pVM The VM to operate on.
4111	* @param idCpu Virtual CPU to perform the hot unplugging operation on.
4112	*/
4113	VMMR3DECL(int) VMR3HotUnplugCpu(PVM pVM, VMCPUID idCpu)
4114	{
4115	AssertReturn(idCpu < pVM->cCpus, VERR_INVALID_CPU_ID);
4116
4117	/** @todo r=bird: Don't destroy the EMT, it'll break VMMR3EmtRendezvous and
4118	* broadcast requests. Just note down somewhere that the CPU is
4119	* offline and send it to SPIP wait. Maybe modify VMCPUSTATE and push
4120	* it out of the EM loops when offline. */
4121	return VMR3ReqCallNoWaitU(pVM->pUVM, idCpu, (PFNRT)vmR3HotUnplugCpu, 2, pVM, idCpu);
4122	}
4123
4124
4125	/**
4126	* Hot-plugs a CPU on the guest.
4127	*
4128	* @returns VBox status code.
4129	* @param pVM The VM to operate on.
4130	* @param idCpu Virtual CPU to perform the hot plugging operation on.
4131	*/
4132	VMMR3DECL(int) VMR3HotPlugCpu(PVM pVM, VMCPUID idCpu)
4133	{
4134	AssertReturn(idCpu < pVM->cCpus, VERR_INVALID_CPU_ID);
4135
4136	/** @todo r-bird: Just mark it online and make sure it waits on SPIP. */
4137	return VINF_SUCCESS;
4138	}
4139

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

source: vbox/trunk/src/VBox/VMM/VM.cpp@ 30164

Download in other formats: