VM.cpp@ 28688

Last change on this file since 28688 was 28317, checked in by vboxsync, 15 years ago
RTMemPageFree + all users: Added size parameter to RTMemPageFree so we can avoid tracking structures when using mmap/munmap.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 144.6 KB

Line
1	/* $Id: VM.cpp 28317 2010-04-14 18:06:05Z vboxsync $ */
2	/** @file
3	* VM - Virtual Machine
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Sun Microsystems, Inc.
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	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
18	* Clara, CA 95054 USA or visit http://www.sun.com if you need
19	* additional information or have any questions.
20	*/
21
22	/** @page pg_vm VM API
23	*
24	* This is the encapsulating bit. It provides the APIs that Main and VBoxBFE
25	* use to create a VMM instance for running a guest in. It also provides
26	* facilities for queuing request for execution in EMT (serialization purposes
27	* mostly) and for reporting error back to the VMM user (Main/VBoxBFE).
28	*
29	*
30	* @section sec_vm_design Design Critique / Things To Do
31	*
32	* In hindsight this component is a big design mistake, all this stuff really
33	* belongs in the VMM component. It just seemed like a kind of ok idea at a
34	* time when the VMM bit was a kind of vague. 'VM' also happend to be the name
35	* of the per-VM instance structure (see vm.h), so it kind of made sense.
36	* However as it turned out, VMM(.cpp) is almost empty all it provides in ring-3
37	* is some minor functionally and some "routing" services.
38	*
39	* Fixing this is just a matter of some more or less straight forward
40	* refactoring, the question is just when someone will get to it. Moving the EMT
41	* would be a good start.
42	*
43	*/
44
45	/*******************************************************************************
46	* Header Files *
47	*******************************************************************************/
48	#define LOG_GROUP LOG_GROUP_VM
49	#include <VBox/cfgm.h>
50	#include <VBox/vmm.h>
51	#include <VBox/gvmm.h>
52	#include <VBox/mm.h>
53	#include <VBox/cpum.h>
54	#include <VBox/selm.h>
55	#include <VBox/trpm.h>
56	#include <VBox/dbgf.h>
57	#include <VBox/pgm.h>
58	#include <VBox/pdmapi.h>
59	#include <VBox/pdmcritsect.h>
60	#include <VBox/em.h>
61	#include <VBox/rem.h>
62	#include <VBox/tm.h>
63	#include <VBox/stam.h>
64	#include <VBox/patm.h>
65	#ifdef VBOX_WITH_VMI
66	# include <VBox/parav.h>
67	#endif
68	#include <VBox/csam.h>
69	#include <VBox/iom.h>
70	#include <VBox/ssm.h>
71	#include <VBox/hwaccm.h>
72	#include "VMInternal.h"
73	#include <VBox/vm.h>
74	#include <VBox/uvm.h>
75
76	#include <VBox/sup.h>
77	#include <VBox/dbg.h>
78	#include <VBox/err.h>
79	#include <VBox/param.h>
80	#include <VBox/log.h>
81	#include <iprt/assert.h>
82	#include <iprt/alloc.h>
83	#include <iprt/asm.h>
84	#include <iprt/env.h>
85	#include <iprt/string.h>
86	#include <iprt/time.h>
87	#include <iprt/semaphore.h>
88	#include <iprt/thread.h>
89
90
91	/*******************************************************************************
92	* Structures and Typedefs *
93	*******************************************************************************/
94	/**
95	* VM destruction callback registration record.
96	*/
97	typedef struct VMATDTOR
98	{
99	/** Pointer to the next record in the list. */
100	struct VMATDTOR *pNext;
101	/** Pointer to the callback function. */
102	PFNVMATDTOR pfnAtDtor;
103	/** The user argument. */
104	void *pvUser;
105	} VMATDTOR;
106	/** Pointer to a VM destruction callback registration record. */
107	typedef VMATDTOR *PVMATDTOR;
108
109
110	/*******************************************************************************
111	* Global Variables *
112	*******************************************************************************/
113	/** Pointer to the list of VMs. */
114	static PUVM g_pUVMsHead = NULL;
115
116	/** Pointer to the list of at VM destruction callbacks. */
117	static PVMATDTOR g_pVMAtDtorHead = NULL;
118	/** Lock the g_pVMAtDtorHead list. */
119	#define VM_ATDTOR_LOCK() do { } while (0)
120	/** Unlock the g_pVMAtDtorHead list. */
121	#define VM_ATDTOR_UNLOCK() do { } while (0)
122
123
124	/*******************************************************************************
125	* Internal Functions *
126	*******************************************************************************/
127	static int vmR3CreateUVM(uint32_t cCpus, PUVM *ppUVM);
128	static int vmR3CreateU(PUVM pUVM, uint32_t cCpus, PFNCFGMCONSTRUCTOR pfnCFGMConstructor, void *pvUserCFGM);
129	static int vmR3InitRing3(PVM pVM, PUVM pUVM);
130	static int vmR3InitVMCpu(PVM pVM);
131	static int vmR3InitRing0(PVM pVM);
132	static int vmR3InitGC(PVM pVM);
133	static int vmR3InitDoCompleted(PVM pVM, VMINITCOMPLETED enmWhat);
134	static DECLCALLBACK(size_t) vmR3LogPrefixCallback(PRTLOGGER pLogger, char pchBuf, size_t cchBuf, void pvUser);
135	static void vmR3DestroyUVM(PUVM pUVM, uint32_t cMilliesEMTWait);
136	static void vmR3AtDtor(PVM pVM);
137	static bool vmR3ValidateStateTransition(VMSTATE enmStateOld, VMSTATE enmStateNew);
138	static void vmR3DoAtState(PVM pVM, PUVM pUVM, VMSTATE enmStateNew, VMSTATE enmStateOld);
139	static int vmR3TrySetState(PVM pVM, const char *pszWho, unsigned cTransitions, ...);
140	static void vmR3SetStateLocked(PVM pVM, PUVM pUVM, VMSTATE enmStateNew, VMSTATE enmStateOld);
141	static void vmR3SetState(PVM pVM, VMSTATE enmStateNew, VMSTATE enmStateOld);
142	static int vmR3SetErrorU(PUVM pUVM, int rc, RT_SRC_POS_DECL, const char *pszFormat, ...);
143
144
145	/**
146	* Do global VMM init.
147	*
148	* @returns VBox status code.
149	*/
150	VMMR3DECL(int) VMR3GlobalInit(void)
151	{
152	/*
153	* Only once.
154	*/
155	static bool volatile s_fDone = false;
156	if (s_fDone)
157	return VINF_SUCCESS;
158
159	/*
160	* We're done.
161	*/
162	s_fDone = true;
163	return VINF_SUCCESS;
164	}
165
166
167
168	/**
169	* Creates a virtual machine by calling the supplied configuration constructor.
170	*
171	* On successful returned the VM is powered, i.e. VMR3PowerOn() should be
172	* called to start the execution.
173	*
174	* @returns 0 on success.
175	* @returns VBox error code on failure.
176	* @param cCpus Number of virtual CPUs for the new VM.
177	* @param pfnVMAtError Pointer to callback function for setting VM
178	* errors. This was added as an implicit call to
179	* VMR3AtErrorRegister() since there is no way the
180	* caller can get to the VM handle early enough to
181	* do this on its own.
182	* This is called in the context of an EMT.
183	* @param pvUserVM The user argument passed to pfnVMAtError.
184	* @param pfnCFGMConstructor Pointer to callback function for constructing the VM configuration tree.
185	* This is called in the context of an EMT0.
186	* @param pvUserCFGM The user argument passed to pfnCFGMConstructor.
187	* @param ppVM Where to store the 'handle' of the created VM.
188	*/
189	VMMR3DECL(int) VMR3Create(uint32_t cCpus, PFNVMATERROR pfnVMAtError, void pvUserVM, PFNCFGMCONSTRUCTOR pfnCFGMConstructor, void pvUserCFGM, PVM *ppVM)
190	{
191	LogFlow(("VMR3Create: cCpus=%RU32 pfnVMAtError=%p pvUserVM=%p pfnCFGMConstructor=%p pvUserCFGM=%p ppVM=%p\n",
192	cCpus, pfnVMAtError, pvUserVM, pfnCFGMConstructor, pvUserCFGM, ppVM));
193
194	/*
195	* Because of the current hackiness of the applications
196	* we'll have to initialize global stuff from here.
197	* Later the applications will take care of this in a proper way.
198	*/
199	static bool fGlobalInitDone = false;
200	if (!fGlobalInitDone)
201	{
202	int rc = VMR3GlobalInit();
203	if (RT_FAILURE(rc))
204	return rc;
205	fGlobalInitDone = true;
206	}
207
208	/*
209	* Validate input.
210	*/
211	AssertLogRelMsgReturn(cCpus > 0 && cCpus <= VMM_MAX_CPU_COUNT, ("%RU32\n", cCpus), VERR_TOO_MANY_CPUS);
212
213	/*
214	* Create the UVM so we can register the at-error callback
215	* and consoliate a bit of cleanup code.
216	*/
217	PUVM pUVM = NULL; /* shuts up gcc */
218	int rc = vmR3CreateUVM(cCpus, &pUVM);
219	if (RT_FAILURE(rc))
220	return rc;
221	if (pfnVMAtError)
222	rc = VMR3AtErrorRegisterU(pUVM, pfnVMAtError, pvUserVM);
223	if (RT_SUCCESS(rc))
224	{
225	/*
226	* Initialize the support library creating the session for this VM.
227	*/
228	rc = SUPR3Init(&pUVM->vm.s.pSession);
229	if (RT_SUCCESS(rc))
230	{
231	/*
232	* Call vmR3CreateU in the EMT thread and wait for it to finish.
233	*
234	* Note! VMCPUID_ANY is used here because VMR3ReqQueueU would have trouble
235	* submitting a request to a specific VCPU without a pVM. So, to make
236	* sure init is running on EMT(0), vmR3EmulationThreadWithId makes sure
237	* that only EMT(0) is servicing VMCPUID_ANY requests when pVM is NULL.
238	*/
239	PVMREQ pReq;
240	rc = VMR3ReqCallU(pUVM, VMCPUID_ANY, &pReq, RT_INDEFINITE_WAIT, VMREQFLAGS_VBOX_STATUS,
241	(PFNRT)vmR3CreateU, 4, pUVM, cCpus, pfnCFGMConstructor, pvUserCFGM);
242	if (RT_SUCCESS(rc))
243	{
244	rc = pReq->iStatus;
245	VMR3ReqFree(pReq);
246	if (RT_SUCCESS(rc))
247	{
248	/*
249	* Success!
250	*/
251	*ppVM = pUVM->pVM;
252	LogFlow(("VMR3Create: returns VINF_SUCCESS ppVM=%p\n", ppVM));
253	return VINF_SUCCESS;
254	}
255	}
256	else
257	AssertMsgFailed(("VMR3ReqCallU failed rc=%Rrc\n", rc));
258
259	/*
260	* An error occurred during VM creation. Set the error message directly
261	* using the initial callback, as the callback list doesn't exist yet.
262	*/
263	const char *pszError = NULL;
264	switch (rc)
265	{
266	case VERR_VMX_IN_VMX_ROOT_MODE:
267	#ifdef RT_OS_LINUX
268	pszError = N_("VirtualBox can't operate in VMX root mode. "
269	"Please disable the KVM kernel extension, recompile your kernel and reboot");
270	#else
271	pszError = N_("VirtualBox can't operate in VMX root mode. Please close all other virtualization programs.");
272	#endif
273	break;
274
275	case VERR_SVM_IN_USE:
276	#ifdef RT_OS_LINUX
277	pszError = N_("VirtualBox can't enable the AMD-V extension. "
278	"Please disable the KVM kernel extension, recompile your kernel and reboot");
279	#else
280	pszError = N_("VirtualBox can't enable the AMD-V extension. Please close all other virtualization programs.");
281	#endif
282	break;
283
284	case VERR_VERSION_MISMATCH:
285	pszError = N_("VMMR0 driver version mismatch. Please terminate all VMs, make sure that "
286	"VBoxNetDHCP is not running and try again. If you still get this error, "
287	"re-install VirtualBox");
288	break;
289
290	#ifdef RT_OS_LINUX
291	case VERR_SUPDRV_COMPONENT_NOT_FOUND:
292	pszError = N_("One of the kernel modules was not successfully loaded. Make sure "
293	"that no kernel modules from an older version of VirtualBox exist. "
294	"Then try to recompile and reload the kernel modules by executing "
295	"'/etc/init.d/vboxdrv setup' as root");
296	break;
297	#endif
298
299	case VERR_RAW_MODE_INVALID_SMP:
300	pszError = N_("VT-x/AMD-V is either not available on your host or disabled. "
301	"VirtualBox requires this hardware extension to emulate more than one "
302	"guest CPU");
303	break;
304
305	case VERR_SUPDRV_KERNEL_TOO_OLD_FOR_VTX:
306	#ifdef RT_OS_LINUX
307	pszError = N_("Because the host kernel is too old, VirtualBox cannot enable the VT-x "
308	"extension. Either upgrade your kernel to Linux 2.6.13 or later or disable "
309	"the VT-x extension in the VM settings. Note that without VT-x you have "
310	"to reduce the number of guest CPUs to one");
311	#else
312	pszError = N_("Because the host kernel is too old, VirtualBox cannot enable the VT-x "
313	"extension. Either upgrade your kernel or disable the VT-x extension in the "
314	"VM settings. Note that without VT-x you have to reduce the number of guest "
315	"CPUs to one");
316	#endif
317	break;
318
319	default:
320	pszError = N_("Unknown error creating VM");
321	break;
322	}
323	vmR3SetErrorU(pUVM, rc, RT_SRC_POS, pszError, rc);
324	}
325	else
326	{
327	/*
328	* An error occurred at support library initialization time (before the
329	* VM could be created). Set the error message directly using the
330	* initial callback, as the callback list doesn't exist yet.
331	*/
332	const char *pszError;
333	switch (rc)
334	{
335	case VERR_VM_DRIVER_LOAD_ERROR:
336	#ifdef RT_OS_LINUX
337	pszError = N_("VirtualBox kernel driver not loaded. The vboxdrv kernel module "
338	"was either not loaded or /dev/vboxdrv is not set up properly. "
339	"Re-setup the kernel module by executing "
340	"'/etc/init.d/vboxdrv setup' as root");
341	#else
342	pszError = N_("VirtualBox kernel driver not loaded");
343	#endif
344	break;
345	case VERR_VM_DRIVER_OPEN_ERROR:
346	pszError = N_("VirtualBox kernel driver cannot be opened");
347	break;
348	case VERR_VM_DRIVER_NOT_ACCESSIBLE:
349	#ifdef VBOX_WITH_HARDENING
350	/* This should only happen if the executable wasn't hardened - bad code/build. */
351	pszError = N_("VirtualBox kernel driver not accessible, permission problem. "
352	"Re-install VirtualBox. If you are building it yourself, you "
353	"should make sure it installed correctly and that the setuid "
354	"bit is set on the executables calling VMR3Create.");
355	#else
356	/* This should only happen when mixing builds or with the usual /dev/vboxdrv access issues. */
357	# if defined(RT_OS_DARWIN)
358	pszError = N_("VirtualBox KEXT is not accessible, permission problem. "
359	"If you have built VirtualBox yourself, make sure that you do not "
360	"have the vboxdrv KEXT from a different build or installation loaded.");
361	# elif defined(RT_OS_LINUX)
362	pszError = N_("VirtualBox kernel driver is not accessible, permission problem. "
363	"If you have built VirtualBox yourself, make sure that you do "
364	"not have the vboxdrv kernel module from a different build or "
365	"installation loaded. Also, make sure the vboxdrv udev rule gives "
366	"you the permission you need to access the device.");
367	# elif defined(RT_OS_WINDOWS)
368	pszError = N_("VirtualBox kernel driver is not accessible, permission problem.");
369	# else /* solaris, freebsd, ++. */
370	pszError = N_("VirtualBox kernel module is not accessible, permission problem. "
371	"If you have built VirtualBox yourself, make sure that you do "
372	"not have the vboxdrv kernel module from a different install loaded.");
373	# endif
374	#endif
375	break;
376	case VERR_INVALID_HANDLE: /** @todo track down and fix this error. */
377	case VERR_VM_DRIVER_NOT_INSTALLED:
378	#ifdef RT_OS_LINUX
379	pszError = N_("VirtualBox kernel driver not installed. The vboxdrv kernel module "
380	"was either not loaded or /dev/vboxdrv was not created for some "
381	"reason. Re-setup the kernel module by executing "
382	"'/etc/init.d/vboxdrv setup' as root");
383	#else
384	pszError = N_("VirtualBox kernel driver not installed");
385	#endif
386	break;
387	case VERR_NO_MEMORY:
388	pszError = N_("VirtualBox support library out of memory");
389	break;
390	case VERR_VERSION_MISMATCH:
391	case VERR_VM_DRIVER_VERSION_MISMATCH:
392	pszError = N_("The VirtualBox support driver which is running is from a different "
393	"version of VirtualBox. You can correct this by stopping all "
394	"running instances of VirtualBox and reinstalling the software.");
395	break;
396	default:
397	pszError = N_("Unknown error initializing kernel driver");
398	AssertMsgFailed(("Add error message for rc=%d (%Rrc)\n", rc, rc));
399	}
400	vmR3SetErrorU(pUVM, rc, RT_SRC_POS, pszError, rc);
401	}
402	}
403
404	/* cleanup */
405	vmR3DestroyUVM(pUVM, 2000);
406	LogFlow(("VMR3Create: returns %Rrc\n", rc));
407	return rc;
408	}
409
410
411	/**
412	* Creates the UVM.
413	*
414	* This will not initialize the support library even if vmR3DestroyUVM
415	* will terminate that.
416	*
417	* @returns VBox status code.
418	* @param cCpus Number of virtual CPUs
419	* @param ppUVM Where to store the UVM pointer.
420	*/
421	static int vmR3CreateUVM(uint32_t cCpus, PUVM *ppUVM)
422	{
423	uint32_t i;
424
425	/*
426	* Create and initialize the UVM.
427	*/
428	PUVM pUVM = (PUVM)RTMemPageAllocZ(RT_OFFSETOF(UVM, aCpus[cCpus]));
429	AssertReturn(pUVM, VERR_NO_MEMORY);
430	pUVM->u32Magic = UVM_MAGIC;
431	pUVM->cCpus = cCpus;
432
433	AssertCompile(sizeof(pUVM->vm.s) <= sizeof(pUVM->vm.padding));
434
435	pUVM->vm.s.ppAtStateNext = &pUVM->vm.s.pAtState;
436	pUVM->vm.s.ppAtErrorNext = &pUVM->vm.s.pAtError;
437	pUVM->vm.s.ppAtRuntimeErrorNext = &pUVM->vm.s.pAtRuntimeError;
438
439	pUVM->vm.s.enmHaltMethod = VMHALTMETHOD_BOOTSTRAP;
440
441	/* Initialize the VMCPU array in the UVM. */
442	for (i = 0; i < cCpus; i++)
443	{
444	pUVM->aCpus[i].pUVM = pUVM;
445	pUVM->aCpus[i].idCpu = i;
446	}
447
448	/* Allocate a TLS entry to store the VMINTUSERPERVMCPU pointer. */
449	int rc = RTTlsAllocEx(&pUVM->vm.s.idxTLS, NULL);
450	AssertRC(rc);
451	if (RT_SUCCESS(rc))
452	{
453	/* Allocate a halt method event semaphore for each VCPU. */
454	for (i = 0; i < cCpus; i++)
455	pUVM->aCpus[i].vm.s.EventSemWait = NIL_RTSEMEVENT;
456	for (i = 0; i < cCpus; i++)
457	{
458	rc = RTSemEventCreate(&pUVM->aCpus[i].vm.s.EventSemWait);
459	if (RT_FAILURE(rc))
460	break;
461	}
462	if (RT_SUCCESS(rc))
463	{
464	rc = RTCritSectInit(&pUVM->vm.s.AtStateCritSect);
465	if (RT_SUCCESS(rc))
466	{
467	rc = RTCritSectInit(&pUVM->vm.s.AtErrorCritSect);
468	if (RT_SUCCESS(rc))
469	{
470	/*
471	* Init fundamental (sub-)components - STAM, MMR3Heap and PDMLdr.
472	*/
473	rc = STAMR3InitUVM(pUVM);
474	if (RT_SUCCESS(rc))
475	{
476	rc = MMR3InitUVM(pUVM);
477	if (RT_SUCCESS(rc))
478	{
479	rc = PDMR3InitUVM(pUVM);
480	if (RT_SUCCESS(rc))
481	{
482	/*
483	* Start the emulation threads for all VMCPUs.
484	*/
485	for (i = 0; i < cCpus; i++)
486	{
487	rc = RTThreadCreateF(&pUVM->aCpus[i].vm.s.ThreadEMT, vmR3EmulationThread, &pUVM->aCpus[i], _1M,
488	RTTHREADTYPE_EMULATION, RTTHREADFLAGS_WAITABLE,
489	cCpus > 1 ? "EMT-%u" : "EMT", i);
490	if (RT_FAILURE(rc))
491	break;
492
493	pUVM->aCpus[i].vm.s.NativeThreadEMT = RTThreadGetNative(pUVM->aCpus[i].vm.s.ThreadEMT);
494	}
495
496	if (RT_SUCCESS(rc))
497	{
498	*ppUVM = pUVM;
499	return VINF_SUCCESS;
500	}
501
502	/* bail out. */
503	while (i-- > 0)
504	{
505	/** @todo rainy day: terminate the EMTs. */
506	}
507	PDMR3TermUVM(pUVM);
508	}
509	MMR3TermUVM(pUVM);
510	}
511	STAMR3TermUVM(pUVM);
512	}
513	RTCritSectDelete(&pUVM->vm.s.AtErrorCritSect);
514	}
515	RTCritSectDelete(&pUVM->vm.s.AtStateCritSect);
516	}
517	}
518	for (i = 0; i < cCpus; i++)
519	{
520	RTSemEventDestroy(pUVM->aCpus[i].vm.s.EventSemWait);
521	pUVM->aCpus[i].vm.s.EventSemWait = NIL_RTSEMEVENT;
522	}
523	RTTlsFree(pUVM->vm.s.idxTLS);
524	}
525	RTMemPageFree(pUVM, sizeof(*pUVM));
526	return rc;
527	}
528
529
530	/**
531	* Creates and initializes the VM.
532	*
533	* @thread EMT
534	*/
535	static int vmR3CreateU(PUVM pUVM, uint32_t cCpus, PFNCFGMCONSTRUCTOR pfnCFGMConstructor, void *pvUserCFGM)
536	{
537	int rc = VINF_SUCCESS;
538
539	/*
540	* Load the VMMR0.r0 module so that we can call GVMMR0CreateVM.
541	*/
542	rc = PDMR3LdrLoadVMMR0U(pUVM);
543	if (RT_FAILURE(rc))
544	{
545	/** @todo we need a cleaner solution for this (VERR_VMX_IN_VMX_ROOT_MODE).
546	* bird: what about moving the message down here? Main picks the first message, right? */
547	if (rc == VERR_VMX_IN_VMX_ROOT_MODE)
548	return rc; /* proper error message set later on */
549	return vmR3SetErrorU(pUVM, rc, RT_SRC_POS, N_("Failed to load VMMR0.r0"));
550	}
551
552	/*
553	* Request GVMM to create a new VM for us.
554	*/
555	GVMMCREATEVMREQ CreateVMReq;
556	CreateVMReq.Hdr.u32Magic = SUPVMMR0REQHDR_MAGIC;
557	CreateVMReq.Hdr.cbReq = sizeof(CreateVMReq);
558	CreateVMReq.pSession = pUVM->vm.s.pSession;
559	CreateVMReq.pVMR0 = NIL_RTR0PTR;
560	CreateVMReq.pVMR3 = NULL;
561	CreateVMReq.cCpus = cCpus;
562	rc = SUPR3CallVMMR0Ex(NIL_RTR0PTR, NIL_VMCPUID, VMMR0_DO_GVMM_CREATE_VM, 0, &CreateVMReq.Hdr);
563	if (RT_SUCCESS(rc))
564	{
565	PVM pVM = pUVM->pVM = CreateVMReq.pVMR3;
566	AssertRelease(VALID_PTR(pVM));
567	AssertRelease(pVM->pVMR0 == CreateVMReq.pVMR0);
568	AssertRelease(pVM->pSession == pUVM->vm.s.pSession);
569	AssertRelease(pVM->cCpus == cCpus);
570	AssertRelease(pVM->offVMCPU == RT_UOFFSETOF(VM, aCpus));
571
572	Log(("VMR3Create: Created pUVM=%p pVM=%p pVMR0=%p hSelf=%#x cCpus=%RU32\n",
573	pUVM, pVM, pVM->pVMR0, pVM->hSelf, pVM->cCpus));
574
575	/*
576	* Initialize the VM structure and our internal data (VMINT).
577	*/
578	pVM->pUVM = pUVM;
579
580	for (VMCPUID i = 0; i < pVM->cCpus; i++)
581	{
582	pVM->aCpus[i].pUVCpu = &pUVM->aCpus[i];
583	pVM->aCpus[i].idCpu = i;
584	pVM->aCpus[i].hNativeThread = pUVM->aCpus[i].vm.s.NativeThreadEMT;
585	Assert(pVM->aCpus[i].hNativeThread != NIL_RTNATIVETHREAD);
586
587	pUVM->aCpus[i].pVCpu = &pVM->aCpus[i];
588	pUVM->aCpus[i].pVM = pVM;
589	}
590
591
592	/*
593	* Init the configuration.
594	*/
595	rc = CFGMR3Init(pVM, pfnCFGMConstructor, pvUserCFGM);
596	if (RT_SUCCESS(rc))
597	{
598	rc = CFGMR3QueryBoolDef(CFGMR3GetRoot(pVM), "HwVirtExtForced", &pVM->fHwVirtExtForced, false);
599	if (RT_SUCCESS(rc) && pVM->fHwVirtExtForced)
600	pVM->fHWACCMEnabled = true;
601
602	/*
603	* If executing in fake suplib mode disable RR3 and RR0 in the config.
604	*/
605	const char *psz = RTEnvGet("VBOX_SUPLIB_FAKE");
606	if (psz && !strcmp(psz, "fake"))
607	{
608	CFGMR3RemoveValue(CFGMR3GetRoot(pVM), "RawR3Enabled");
609	CFGMR3InsertInteger(CFGMR3GetRoot(pVM), "RawR3Enabled", 0);
610	CFGMR3RemoveValue(CFGMR3GetRoot(pVM), "RawR0Enabled");
611	CFGMR3InsertInteger(CFGMR3GetRoot(pVM), "RawR0Enabled", 0);
612	}
613
614	/*
615	* Make sure the CPU count in the config data matches.
616	*/
617	if (RT_SUCCESS(rc))
618	{
619	uint32_t cCPUsCfg;
620	rc = CFGMR3QueryU32Def(CFGMR3GetRoot(pVM), "NumCPUs", &cCPUsCfg, 1);
621	AssertLogRelMsgRC(rc, ("Configuration error: Querying \"NumCPUs\" as integer failed, rc=%Rrc\n", rc));
622	if (RT_SUCCESS(rc) && cCPUsCfg != cCpus)
623	{
624	AssertLogRelMsgFailed(("Configuration error: \"NumCPUs\"=%RU32 and VMR3CreateVM::cCpus=%RU32 does not match!\n",
625	cCPUsCfg, cCpus));
626	rc = VERR_INVALID_PARAMETER;
627	}
628	}
629	if (RT_SUCCESS(rc))
630	{
631	/*
632	* Init the ring-3 components and ring-3 per cpu data, finishing it off
633	* by a relocation round (intermediate context finalization will do this).
634	*/
635	rc = vmR3InitRing3(pVM, pUVM);
636	if (RT_SUCCESS(rc))
637	{
638	rc = vmR3InitVMCpu(pVM);
639	if (RT_SUCCESS(rc))
640	rc = PGMR3FinalizeMappings(pVM);
641	if (RT_SUCCESS(rc))
642	{
643
644	LogFlow(("Ring-3 init succeeded\n"));
645
646	/*
647	* Init the Ring-0 components.
648	*/
649	rc = vmR3InitRing0(pVM);
650	if (RT_SUCCESS(rc))
651	{
652	/* Relocate again, because some switcher fixups depends on R0 init results. */
653	VMR3Relocate(pVM, 0);
654
655	#ifdef VBOX_WITH_DEBUGGER
656	/*
657	* Init the tcp debugger console if we're building
658	* with debugger support.
659	*/
660	void *pvUser = NULL;
661	rc = DBGCTcpCreate(pVM, &pvUser);
662	if ( RT_SUCCESS(rc)
663	\|\| rc == VERR_NET_ADDRESS_IN_USE)
664	{
665	pUVM->vm.s.pvDBGC = pvUser;
666	#endif
667	/*
668	* Init the Guest Context components.
669	*/
670	rc = vmR3InitGC(pVM);
671	if (RT_SUCCESS(rc))
672	{
673	/*
674	* Now we can safely set the VM halt method to default.
675	*/
676	rc = vmR3SetHaltMethodU(pUVM, VMHALTMETHOD_DEFAULT);
677	if (RT_SUCCESS(rc))
678	{
679	/*
680	* Set the state and link into the global list.
681	*/
682	vmR3SetState(pVM, VMSTATE_CREATED, VMSTATE_CREATING);
683	pUVM->pNext = g_pUVMsHead;
684	g_pUVMsHead = pUVM;
685
686	#ifdef LOG_ENABLED
687	RTLogSetCustomPrefixCallback(NULL, vmR3LogPrefixCallback, pUVM);
688	#endif
689	return VINF_SUCCESS;
690	}
691	}
692	#ifdef VBOX_WITH_DEBUGGER
693	DBGCTcpTerminate(pVM, pUVM->vm.s.pvDBGC);
694	pUVM->vm.s.pvDBGC = NULL;
695	}
696	#endif
697	//..
698	}
699	}
700	vmR3Destroy(pVM);
701	}
702	}
703	//..
704
705	/* Clean CFGM. */
706	int rc2 = CFGMR3Term(pVM);
707	AssertRC(rc2);
708	}
709
710	/*
711	* Do automatic cleanups while the VM structure is still alive and all
712	* references to it are still working.
713	*/
714	PDMR3CritSectTerm(pVM);
715
716	/*
717	* Drop all references to VM and the VMCPU structures, then
718	* tell GVMM to destroy the VM.
719	*/
720	pUVM->pVM = NULL;
721	for (VMCPUID i = 0; i < pUVM->cCpus; i++)
722	{
723	pUVM->aCpus[i].pVM = NULL;
724	pUVM->aCpus[i].pVCpu = NULL;
725	}
726	Assert(pUVM->vm.s.enmHaltMethod == VMHALTMETHOD_BOOTSTRAP);
727
728	if (pUVM->cCpus > 1)
729	{
730	/* Poke the other EMTs since they may have stale pVM and pVCpu references
731	on the stack (see VMR3WaitU for instance) if they've been awakened after
732	VM creation. */
733	for (VMCPUID i = 1; i < pUVM->cCpus; i++)
734	VMR3NotifyCpuFFU(&pUVM->aCpus[i], 0);
735	RTThreadSleep(RT_MIN(100 + 25 (pUVM->cCpus - 1), 500)); / very sophisticated */
736	}
737
738	int rc2 = SUPR3CallVMMR0Ex(CreateVMReq.pVMR0, 0 /idCpu/, VMMR0_DO_GVMM_DESTROY_VM, 0, NULL);
739	AssertRC(rc2);
740	}
741	else
742	vmR3SetErrorU(pUVM, rc, RT_SRC_POS, N_("VM creation failed (GVMM)"));
743
744	LogFlow(("vmR3CreateU: returns %Rrc\n", rc));
745	return rc;
746	}
747
748
749	/**
750	* Register the calling EMT with GVM.
751	*
752	* @returns VBox status code.
753	* @param pVM The VM handle.
754	* @param idCpu The Virtual CPU ID.
755	*/
756	static DECLCALLBACK(int) vmR3RegisterEMT(PVM pVM, VMCPUID idCpu)
757	{
758	Assert(VMMGetCpuId(pVM) == idCpu);
759	int rc = SUPR3CallVMMR0Ex(pVM->pVMR0, idCpu, VMMR0_DO_GVMM_REGISTER_VMCPU, 0, NULL);
760	if (RT_FAILURE(rc))
761	LogRel(("idCpu=%u rc=%Rrc\n", idCpu, rc));
762	return rc;
763	}
764
765
766	/**
767	* Initializes all R3 components of the VM
768	*/
769	static int vmR3InitRing3(PVM pVM, PUVM pUVM)
770	{
771	int rc;
772
773	/*
774	* Register the other EMTs with GVM.
775	*/
776	for (VMCPUID idCpu = 1; idCpu < pVM->cCpus; idCpu++)
777	{
778	rc = VMR3ReqCallWaitU(pUVM, idCpu, (PFNRT)vmR3RegisterEMT, 2, pVM, idCpu);
779	if (RT_FAILURE(rc))
780	return rc;
781	}
782
783	/*
784	* Init all R3 components, the order here might be important.
785	*/
786	rc = MMR3Init(pVM);
787	if (RT_SUCCESS(rc))
788	{
789	STAM_REG(pVM, &pVM->StatTotalInGC, STAMTYPE_PROFILE_ADV, "/PROF/VM/InGC", STAMUNIT_TICKS_PER_CALL, "Profiling the total time spent in GC.");
790	STAM_REG(pVM, &pVM->StatSwitcherToGC, STAMTYPE_PROFILE_ADV, "/PROF/VM/SwitchToGC", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
791	STAM_REG(pVM, &pVM->StatSwitcherToHC, STAMTYPE_PROFILE_ADV, "/PROF/VM/SwitchToHC", STAMUNIT_TICKS_PER_CALL, "Profiling switching to HC.");
792	STAM_REG(pVM, &pVM->StatSwitcherSaveRegs, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/SaveRegs", STAMUNIT_TICKS_PER_CALL,"Profiling switching to GC.");
793	STAM_REG(pVM, &pVM->StatSwitcherSysEnter, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/SysEnter", STAMUNIT_TICKS_PER_CALL,"Profiling switching to GC.");
794	STAM_REG(pVM, &pVM->StatSwitcherDebug, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/Debug", STAMUNIT_TICKS_PER_CALL,"Profiling switching to GC.");
795	STAM_REG(pVM, &pVM->StatSwitcherCR0, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/CR0", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
796	STAM_REG(pVM, &pVM->StatSwitcherCR4, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/CR4", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
797	STAM_REG(pVM, &pVM->StatSwitcherLgdt, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/Lgdt", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
798	STAM_REG(pVM, &pVM->StatSwitcherLidt, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/Lidt", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
799	STAM_REG(pVM, &pVM->StatSwitcherLldt, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/Lldt", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
800	STAM_REG(pVM, &pVM->StatSwitcherTSS, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/TSS", STAMUNIT_TICKS_PER_CALL, "Profiling switching to GC.");
801	STAM_REG(pVM, &pVM->StatSwitcherJmpCR3, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/JmpCR3", STAMUNIT_TICKS_PER_CALL,"Profiling switching to GC.");
802	STAM_REG(pVM, &pVM->StatSwitcherRstrRegs, STAMTYPE_PROFILE_ADV, "/VM/Switcher/ToGC/RstrRegs", STAMUNIT_TICKS_PER_CALL,"Profiling switching to GC.");
803
804	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
805	{
806	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);
807	AssertRC(rc);
808	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);
809	AssertRC(rc);
810	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);
811	AssertRC(rc);
812	}
813
814	STAM_REG(pVM, &pUVM->vm.s.StatReqAllocNew, STAMTYPE_COUNTER, "/VM/Req/AllocNew", STAMUNIT_OCCURENCES, "Number of VMR3ReqAlloc returning a new packet.");
815	STAM_REG(pVM, &pUVM->vm.s.StatReqAllocRaces, STAMTYPE_COUNTER, "/VM/Req/AllocRaces", STAMUNIT_OCCURENCES, "Number of VMR3ReqAlloc causing races.");
816	STAM_REG(pVM, &pUVM->vm.s.StatReqAllocRecycled, STAMTYPE_COUNTER, "/VM/Req/AllocRecycled", STAMUNIT_OCCURENCES, "Number of VMR3ReqAlloc returning a recycled packet.");
817	STAM_REG(pVM, &pUVM->vm.s.StatReqFree, STAMTYPE_COUNTER, "/VM/Req/Free", STAMUNIT_OCCURENCES, "Number of VMR3ReqFree calls.");
818	STAM_REG(pVM, &pUVM->vm.s.StatReqFreeOverflow, STAMTYPE_COUNTER, "/VM/Req/FreeOverflow", STAMUNIT_OCCURENCES, "Number of times the request was actually freed.");
819	STAM_REG(pVM, &pUVM->vm.s.StatReqProcessed, STAMTYPE_COUNTER, "/VM/Req/Processed", STAMUNIT_OCCURENCES, "Number of processed requests (any queue).");
820	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.");
821	STAM_REG(pVM, &pUVM->vm.s.StatReqPushBackRaces, STAMTYPE_COUNTER, "/VM/Req/PushBackRaces", STAMUNIT_OCCURENCES, "Number of push back races.");
822
823	rc = CPUMR3Init(pVM);
824	if (RT_SUCCESS(rc))
825	{
826	rc = HWACCMR3Init(pVM);
827	if (RT_SUCCESS(rc))
828	{
829	rc = PGMR3Init(pVM);
830	if (RT_SUCCESS(rc))
831	{
832	rc = REMR3Init(pVM);
833	if (RT_SUCCESS(rc))
834	{
835	rc = MMR3InitPaging(pVM);
836	if (RT_SUCCESS(rc))
837	rc = TMR3Init(pVM);
838	if (RT_SUCCESS(rc))
839	{
840	rc = VMMR3Init(pVM);
841	if (RT_SUCCESS(rc))
842	{
843	rc = SELMR3Init(pVM);
844	if (RT_SUCCESS(rc))
845	{
846	rc = TRPMR3Init(pVM);
847	if (RT_SUCCESS(rc))
848	{
849	rc = CSAMR3Init(pVM);
850	if (RT_SUCCESS(rc))
851	{
852	rc = PATMR3Init(pVM);
853	if (RT_SUCCESS(rc))
854	{
855	#ifdef VBOX_WITH_VMI
856	rc = PARAVR3Init(pVM);
857	if (RT_SUCCESS(rc))
858	{
859	#endif
860	rc = IOMR3Init(pVM);
861	if (RT_SUCCESS(rc))
862	{
863	rc = EMR3Init(pVM);
864	if (RT_SUCCESS(rc))
865	{
866	rc = DBGFR3Init(pVM);
867	if (RT_SUCCESS(rc))
868	{
869	rc = PDMR3Init(pVM);
870	if (RT_SUCCESS(rc))
871	{
872	rc = PGMR3InitDynMap(pVM);
873	if (RT_SUCCESS(rc))
874	rc = MMR3HyperInitFinalize(pVM);
875	if (RT_SUCCESS(rc))
876	rc = PATMR3InitFinalize(pVM);
877	if (RT_SUCCESS(rc))
878	rc = PGMR3InitFinalize(pVM);
879	if (RT_SUCCESS(rc))
880	rc = SELMR3InitFinalize(pVM);
881	if (RT_SUCCESS(rc))
882	rc = TMR3InitFinalize(pVM);
883	if (RT_SUCCESS(rc))
884	rc = VMMR3InitFinalize(pVM);
885	if (RT_SUCCESS(rc))
886	rc = REMR3InitFinalize(pVM);
887	if (RT_SUCCESS(rc))
888	rc = vmR3InitDoCompleted(pVM, VMINITCOMPLETED_RING3);
889	if (RT_SUCCESS(rc))
890	{
891	LogFlow(("vmR3InitRing3: returns %Rrc\n", VINF_SUCCESS));
892	return VINF_SUCCESS;
893	}
894	int rc2 = PDMR3Term(pVM);
895	AssertRC(rc2);
896	}
897	int rc2 = DBGFR3Term(pVM);
898	AssertRC(rc2);
899	}
900	int rc2 = EMR3Term(pVM);
901	AssertRC(rc2);
902	}
903	int rc2 = IOMR3Term(pVM);
904	AssertRC(rc2);
905	}
906	#ifdef VBOX_WITH_VMI
907	int rc2 = PARAVR3Term(pVM);
908	AssertRC(rc2);
909	}
910	#endif
911	int rc2 = PATMR3Term(pVM);
912	AssertRC(rc2);
913	}
914	int rc2 = CSAMR3Term(pVM);
915	AssertRC(rc2);
916	}
917	int rc2 = TRPMR3Term(pVM);
918	AssertRC(rc2);
919	}
920	int rc2 = SELMR3Term(pVM);
921	AssertRC(rc2);
922	}
923	int rc2 = VMMR3Term(pVM);
924	AssertRC(rc2);
925	}
926	int rc2 = TMR3Term(pVM);
927	AssertRC(rc2);
928	}
929	int rc2 = REMR3Term(pVM);
930	AssertRC(rc2);
931	}
932	int rc2 = PGMR3Term(pVM);
933	AssertRC(rc2);
934	}
935	int rc2 = HWACCMR3Term(pVM);
936	AssertRC(rc2);
937	}
938	//int rc2 = CPUMR3Term(pVM);
939	//AssertRC(rc2);
940	}
941	/* MMR3Term is not called here because it'll kill the heap. */
942	}
943
944	LogFlow(("vmR3InitRing3: returns %Rrc\n", rc));
945	return rc;
946	}
947
948
949	/**
950	* Initializes all VM CPU components of the VM
951	*/
952	static int vmR3InitVMCpu(PVM pVM)
953	{
954	int rc = VINF_SUCCESS;
955	int rc2;
956
957	rc = CPUMR3InitCPU(pVM);
958	if (RT_SUCCESS(rc))
959	{
960	rc = HWACCMR3InitCPU(pVM);
961	if (RT_SUCCESS(rc))
962	{
963	rc = PGMR3InitCPU(pVM);
964	if (RT_SUCCESS(rc))
965	{
966	rc = TMR3InitCPU(pVM);
967	if (RT_SUCCESS(rc))
968	{
969	rc = VMMR3InitCPU(pVM);
970	if (RT_SUCCESS(rc))
971	{
972	rc = EMR3InitCPU(pVM);
973	if (RT_SUCCESS(rc))
974	{
975	LogFlow(("vmR3InitVMCpu: returns %Rrc\n", VINF_SUCCESS));
976	return VINF_SUCCESS;
977	}
978
979	rc2 = VMMR3TermCPU(pVM);
980	AssertRC(rc2);
981	}
982	rc2 = TMR3TermCPU(pVM);
983	AssertRC(rc2);
984	}
985	rc2 = PGMR3TermCPU(pVM);
986	AssertRC(rc2);
987	}
988	rc2 = HWACCMR3TermCPU(pVM);
989	AssertRC(rc2);
990	}
991	rc2 = CPUMR3TermCPU(pVM);
992	AssertRC(rc2);
993	}
994	LogFlow(("vmR3InitVMCpu: returns %Rrc\n", rc));
995	return rc;
996	}
997
998
999	/**
1000	* Initializes all R0 components of the VM
1001	*/
1002	static int vmR3InitRing0(PVM pVM)
1003	{
1004	LogFlow(("vmR3InitRing0:\n"));
1005
1006	/*
1007	* Check for FAKE suplib mode.
1008	*/
1009	int rc = VINF_SUCCESS;
1010	const char *psz = RTEnvGet("VBOX_SUPLIB_FAKE");
1011	if (!psz \|\| strcmp(psz, "fake"))
1012	{
1013	/*
1014	* Call the VMMR0 component and let it do the init.
1015	*/
1016	rc = VMMR3InitR0(pVM);
1017	}
1018	else
1019	Log(("vmR3InitRing0: skipping because of VBOX_SUPLIB_FAKE=fake\n"));
1020
1021	/*
1022	* Do notifications and return.
1023	*/
1024	if (RT_SUCCESS(rc))
1025	rc = vmR3InitDoCompleted(pVM, VMINITCOMPLETED_RING0);
1026
1027	/** @todo Move this to the VMINITCOMPLETED_RING0 notification handler. */
1028	if (RT_SUCCESS(rc))
1029	{
1030	rc = HWACCMR3InitFinalizeR0(pVM);
1031	CPUMR3SetHWVirtEx(pVM, HWACCMIsEnabled(pVM));
1032	}
1033
1034	LogFlow(("vmR3InitRing0: returns %Rrc\n", rc));
1035	return rc;
1036	}
1037
1038
1039	/**
1040	* Initializes all GC components of the VM
1041	*/
1042	static int vmR3InitGC(PVM pVM)
1043	{
1044	LogFlow(("vmR3InitGC:\n"));
1045
1046	/*
1047	* Check for FAKE suplib mode.
1048	*/
1049	int rc = VINF_SUCCESS;
1050	const char *psz = RTEnvGet("VBOX_SUPLIB_FAKE");
1051	if (!psz \|\| strcmp(psz, "fake"))
1052	{
1053	/*
1054	* Call the VMMR0 component and let it do the init.
1055	*/
1056	rc = VMMR3InitRC(pVM);
1057	}
1058	else
1059	Log(("vmR3InitGC: skipping because of VBOX_SUPLIB_FAKE=fake\n"));
1060
1061	/*
1062	* Do notifications and return.
1063	*/
1064	if (RT_SUCCESS(rc))
1065	rc = vmR3InitDoCompleted(pVM, VMINITCOMPLETED_GC);
1066	LogFlow(("vmR3InitGC: returns %Rrc\n", rc));
1067	return rc;
1068	}
1069
1070
1071	/**
1072	* Do init completed notifications.
1073	* This notifications can fail.
1074	*
1075	* @param pVM The VM handle.
1076	* @param enmWhat What's completed.
1077	*/
1078	static int vmR3InitDoCompleted(PVM pVM, VMINITCOMPLETED enmWhat)
1079	{
1080	return VINF_SUCCESS;
1081	}
1082
1083
1084	/**
1085	* Logger callback for inserting a custom prefix.
1086	*
1087	* @returns Number of chars written.
1088	* @param pLogger The logger.
1089	* @param pchBuf The output buffer.
1090	* @param cchBuf The output buffer size.
1091	* @param pvUser Pointer to the UVM structure.
1092	*/
1093	static DECLCALLBACK(size_t) vmR3LogPrefixCallback(PRTLOGGER pLogger, char pchBuf, size_t cchBuf, void pvUser)
1094	{
1095	AssertReturn(cchBuf >= 2, 0);
1096	PUVM pUVM = (PUVM)pvUser;
1097	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pUVM->vm.s.idxTLS);
1098	if (pUVCpu)
1099	{
1100	static const char s_szHex[17] = "0123456789abcdef";
1101	VMCPUID const idCpu = pUVCpu->idCpu;
1102	pchBuf[1] = s_szHex[ idCpu & 15];
1103	pchBuf[0] = s_szHex[(idCpu >> 4) & 15];
1104	}
1105	else
1106	{
1107	pchBuf[0] = 'x';
1108	pchBuf[1] = 'y';
1109	}
1110
1111	return 2;
1112	}
1113
1114
1115	/**
1116	* Calls the relocation functions for all VMM components so they can update
1117	* any GC pointers. When this function is called all the basic VM members
1118	* have been updated and the actual memory relocation have been done
1119	* by the PGM/MM.
1120	*
1121	* This is used both on init and on runtime relocations.
1122	*
1123	* @param pVM VM handle.
1124	* @param offDelta Relocation delta relative to old location.
1125	*/
1126	VMMR3DECL(void) VMR3Relocate(PVM pVM, RTGCINTPTR offDelta)
1127	{
1128	LogFlow(("VMR3Relocate: offDelta=%RGv\n", offDelta));
1129
1130	/*
1131	* The order here is very important!
1132	*/
1133	PGMR3Relocate(pVM, offDelta);
1134	PDMR3LdrRelocateU(pVM->pUVM, offDelta);
1135	PGMR3Relocate(pVM, 0); /* Repeat after PDM relocation. */
1136	CPUMR3Relocate(pVM);
1137	HWACCMR3Relocate(pVM);
1138	SELMR3Relocate(pVM);
1139	VMMR3Relocate(pVM, offDelta);
1140	SELMR3Relocate(pVM); /* !hack! fix stack! */
1141	TRPMR3Relocate(pVM, offDelta);
1142	PATMR3Relocate(pVM);
1143	CSAMR3Relocate(pVM, offDelta);
1144	IOMR3Relocate(pVM, offDelta);
1145	EMR3Relocate(pVM);
1146	TMR3Relocate(pVM, offDelta);
1147	DBGFR3Relocate(pVM, offDelta);
1148	PDMR3Relocate(pVM, offDelta);
1149	}
1150
1151
1152	/**
1153	* EMT rendezvous worker for VMR3PowerOn.
1154	*
1155	* @returns VERR_VM_INVALID_VM_STATE or VINF_SUCCESS. (This is a strict return
1156	* code, see FNVMMEMTRENDEZVOUS.)
1157	*
1158	* @param pVM The VM handle.
1159	* @param pVCpu The VMCPU handle of the EMT.
1160	* @param pvUser Ignored.
1161	*/
1162	static DECLCALLBACK(VBOXSTRICTRC) vmR3PowerOn(PVM pVM, PVMCPU pVCpu, void *pvUser)
1163	{
1164	LogFlow(("vmR3PowerOn: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1165	Assert(!pvUser); NOREF(pvUser);
1166
1167	/*
1168	* The first thread thru here tries to change the state. We shouldn't be
1169	* called again if this fails.
1170	*/
1171	if (pVCpu->idCpu == pVM->cCpus - 1)
1172	{
1173	int rc = vmR3TrySetState(pVM, "VMR3PowerOn", 1, VMSTATE_POWERING_ON, VMSTATE_CREATED);
1174	if (RT_FAILURE(rc))
1175	return rc;
1176	}
1177
1178	VMSTATE enmVMState = VMR3GetState(pVM);
1179	AssertMsgReturn(enmVMState == VMSTATE_POWERING_ON,
1180	("%s\n", VMR3GetStateName(enmVMState)),
1181	VERR_INTERNAL_ERROR_4);
1182
1183	/*
1184	* All EMTs changes their state to started.
1185	*/
1186	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED);
1187
1188	/*
1189	* EMT(0) is last thru here and it will make the notification calls
1190	* and advance the state.
1191	*/
1192	if (pVCpu->idCpu == 0)
1193	{
1194	PDMR3PowerOn(pVM);
1195	vmR3SetState(pVM, VMSTATE_RUNNING, VMSTATE_POWERING_ON);
1196	}
1197
1198	return VINF_SUCCESS;
1199	}
1200
1201
1202	/**
1203	* Powers on the virtual machine.
1204	*
1205	* @returns VBox status code.
1206	*
1207	* @param pVM The VM to power on.
1208	*
1209	* @thread Any thread.
1210	* @vmstate Created
1211	* @vmstateto PoweringOn+Running
1212	*/
1213	VMMR3DECL(int) VMR3PowerOn(PVM pVM)
1214	{
1215	LogFlow(("VMR3PowerOn: pVM=%p\n", pVM));
1216	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1217
1218	/*
1219	* Gather all the EMTs to reduce the init TSC drift and keep
1220	* the state changing APIs a bit uniform.
1221	*/
1222	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
1223	vmR3PowerOn, NULL);
1224	LogFlow(("VMR3PowerOn: returns %Rrc\n", rc));
1225	return rc;
1226	}
1227
1228
1229	/**
1230	* Does the suspend notifications.
1231	*
1232	* @param pVM The VM handle.
1233	* @thread EMT(0)
1234	*/
1235	static void vmR3SuspendDoWork(PVM pVM)
1236	{
1237	PDMR3Suspend(pVM);
1238	}
1239
1240
1241	/**
1242	* EMT rendezvous worker for VMR3Suspend.
1243	*
1244	* @returns VERR_VM_INVALID_VM_STATE or VINF_EM_SUSPEND. (This is a strict
1245	* return code, see FNVMMEMTRENDEZVOUS.)
1246	*
1247	* @param pVM The VM handle.
1248	* @param pVCpu The VMCPU handle of the EMT.
1249	* @param pvUser Ignored.
1250	*/
1251	static DECLCALLBACK(VBOXSTRICTRC) vmR3Suspend(PVM pVM, PVMCPU pVCpu, void *pvUser)
1252	{
1253	LogFlow(("vmR3Suspend: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1254	Assert(!pvUser); NOREF(pvUser);
1255
1256	/*
1257	* The first EMT switches the state to suspending. If this fails because
1258	* something was racing us in one way or the other, there will be no more
1259	* calls and thus the state assertion below is not going to annoy anyone.
1260	*/
1261	if (pVCpu->idCpu == pVM->cCpus - 1)
1262	{
1263	int rc = vmR3TrySetState(pVM, "VMR3Suspend", 2,
1264	VMSTATE_SUSPENDING, VMSTATE_RUNNING,
1265	VMSTATE_SUSPENDING_EXT_LS, VMSTATE_RUNNING_LS);
1266	if (RT_FAILURE(rc))
1267	return rc;
1268	}
1269
1270	VMSTATE enmVMState = VMR3GetState(pVM);
1271	AssertMsgReturn( enmVMState == VMSTATE_SUSPENDING
1272	\|\| enmVMState == VMSTATE_SUSPENDING_EXT_LS,
1273	("%s\n", VMR3GetStateName(enmVMState)),
1274	VERR_INTERNAL_ERROR_4);
1275
1276	/*
1277	* EMT(0) does the actually suspending after all the other CPUs have
1278	* been thru here.
1279	*/
1280	if (pVCpu->idCpu == 0)
1281	{
1282	vmR3SuspendDoWork(pVM);
1283
1284	int rc = vmR3TrySetState(pVM, "VMR3Suspend", 2,
1285	VMSTATE_SUSPENDED, VMSTATE_SUSPENDING,
1286	VMSTATE_SUSPENDED_EXT_LS, VMSTATE_SUSPENDING_EXT_LS);
1287	if (RT_FAILURE(rc))
1288	return VERR_INTERNAL_ERROR_3;
1289	}
1290
1291	return VINF_EM_SUSPEND;
1292	}
1293
1294
1295	/**
1296	* Suspends a running VM.
1297	*
1298	* @returns VBox status code. When called on EMT, this will be a strict status
1299	* code that has to be propagated up the call stack.
1300	*
1301	* @param pVM The VM to suspend.
1302	*
1303	* @thread Any thread.
1304	* @vmstate Running or RunningLS
1305	* @vmstateto Suspending + Suspended or SuspendingExtLS + SuspendedExtLS
1306	*/
1307	VMMR3DECL(int) VMR3Suspend(PVM pVM)
1308	{
1309	LogFlow(("VMR3Suspend: pVM=%p\n", pVM));
1310	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1311
1312	/*
1313	* Gather all the EMTs to make sure there are no races before
1314	* changing the VM state.
1315	*/
1316	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
1317	vmR3Suspend, NULL);
1318	LogFlow(("VMR3Suspend: returns %Rrc\n", rc));
1319	return rc;
1320	}
1321
1322
1323	/**
1324	* EMT rendezvous worker for VMR3Resume.
1325	*
1326	* @returns VERR_VM_INVALID_VM_STATE or VINF_EM_RESUME. (This is a strict
1327	* return code, see FNVMMEMTRENDEZVOUS.)
1328	*
1329	* @param pVM The VM handle.
1330	* @param pVCpu The VMCPU handle of the EMT.
1331	* @param pvUser Ignored.
1332	*/
1333	static DECLCALLBACK(VBOXSTRICTRC) vmR3Resume(PVM pVM, PVMCPU pVCpu, void *pvUser)
1334	{
1335	LogFlow(("vmR3Resume: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1336	Assert(!pvUser); NOREF(pvUser);
1337
1338	/*
1339	* The first thread thru here tries to change the state. We shouldn't be
1340	* called again if this fails.
1341	*/
1342	if (pVCpu->idCpu == pVM->cCpus - 1)
1343	{
1344	int rc = vmR3TrySetState(pVM, "VMR3Resume", 1, VMSTATE_RESUMING, VMSTATE_SUSPENDED);
1345	if (RT_FAILURE(rc))
1346	return rc;
1347	}
1348
1349	VMSTATE enmVMState = VMR3GetState(pVM);
1350	AssertMsgReturn(enmVMState == VMSTATE_RESUMING,
1351	("%s\n", VMR3GetStateName(enmVMState)),
1352	VERR_INTERNAL_ERROR_4);
1353
1354	#if 0
1355	/*
1356	* All EMTs changes their state to started.
1357	*/
1358	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STARTED);
1359	#endif
1360
1361	/*
1362	* EMT(0) is last thru here and it will make the notification calls
1363	* and advance the state.
1364	*/
1365	if (pVCpu->idCpu == 0)
1366	{
1367	PDMR3Resume(pVM);
1368	vmR3SetState(pVM, VMSTATE_RUNNING, VMSTATE_RESUMING);
1369	pVM->vm.s.fTeleportedAndNotFullyResumedYet = false;
1370	}
1371
1372	return VINF_EM_RESUME;
1373	}
1374
1375
1376	/**
1377	* Resume VM execution.
1378	*
1379	* @returns VBox status code. When called on EMT, this will be a strict status
1380	* code that has to be propagated up the call stack.
1381	*
1382	* @param pVM The VM to resume.
1383	*
1384	* @thread Any thread.
1385	* @vmstate Suspended
1386	* @vmstateto Running
1387	*/
1388	VMMR3DECL(int) VMR3Resume(PVM pVM)
1389	{
1390	LogFlow(("VMR3Resume: pVM=%p\n", pVM));
1391	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1392
1393	/*
1394	* Gather all the EMTs to make sure there are no races before
1395	* changing the VM state.
1396	*/
1397	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
1398	vmR3Resume, NULL);
1399	LogFlow(("VMR3Resume: returns %Rrc\n", rc));
1400	return rc;
1401	}
1402
1403
1404	/**
1405	* EMT rendezvous worker for VMR3Save and VMR3Teleport that suspends the VM
1406	* after the live step has been completed.
1407	*
1408	* @returns VERR_VM_INVALID_VM_STATE or VINF_EM_RESUME. (This is a strict
1409	* return code, see FNVMMEMTRENDEZVOUS.)
1410	*
1411	* @param pVM The VM handle.
1412	* @param pVCpu The VMCPU handle of the EMT.
1413	* @param pvUser The pfSuspended argument of vmR3SaveTeleport.
1414	*/
1415	static DECLCALLBACK(VBOXSTRICTRC) vmR3LiveDoSuspend(PVM pVM, PVMCPU pVCpu, void *pvUser)
1416	{
1417	LogFlow(("vmR3LiveDoSuspend: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1418	bool pfSuspended = (bool )pvUser;
1419
1420	/*
1421	* The first thread thru here tries to change the state. We shouldn't be
1422	* called again if this fails.
1423	*/
1424	if (pVCpu->idCpu == pVM->cCpus - 1U)
1425	{
1426	PUVM pUVM = pVM->pUVM;
1427	int rc;
1428
1429	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
1430	VMSTATE enmVMState = pVM->enmVMState;
1431	switch (enmVMState)
1432	{
1433	case VMSTATE_RUNNING_LS:
1434	vmR3SetStateLocked(pVM, pUVM, VMSTATE_SUSPENDING_LS, VMSTATE_RUNNING_LS);
1435	rc = VINF_SUCCESS;
1436	break;
1437
1438	case VMSTATE_SUSPENDED_EXT_LS:
1439	case VMSTATE_SUSPENDED_LS: /* (via reset) */
1440	rc = VINF_SUCCESS;
1441	break;
1442
1443	case VMSTATE_DEBUGGING_LS:
1444	rc = VERR_TRY_AGAIN;
1445	break;
1446
1447	case VMSTATE_OFF_LS:
1448	vmR3SetStateLocked(pVM, pUVM, VMSTATE_OFF, VMSTATE_OFF_LS);
1449	rc = VERR_SSM_LIVE_POWERED_OFF;
1450	break;
1451
1452	case VMSTATE_FATAL_ERROR_LS:
1453	vmR3SetStateLocked(pVM, pUVM, VMSTATE_FATAL_ERROR, VMSTATE_FATAL_ERROR_LS);
1454	rc = VERR_SSM_LIVE_FATAL_ERROR;
1455	break;
1456
1457	case VMSTATE_GURU_MEDITATION_LS:
1458	vmR3SetStateLocked(pVM, pUVM, VMSTATE_GURU_MEDITATION, VMSTATE_GURU_MEDITATION_LS);
1459	rc = VERR_SSM_LIVE_GURU_MEDITATION;
1460	break;
1461
1462	case VMSTATE_POWERING_OFF_LS:
1463	case VMSTATE_SUSPENDING_EXT_LS:
1464	case VMSTATE_RESETTING_LS:
1465	default:
1466	AssertMsgFailed(("%s\n", VMR3GetStateName(enmVMState)));
1467	rc = VERR_INTERNAL_ERROR_3;
1468	break;
1469	}
1470	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
1471	if (RT_FAILURE(rc))
1472	{
1473	LogFlow(("vmR3LiveDoSuspend: returns %Rrc (state was %s)\n", rc, VMR3GetStateName(enmVMState)));
1474	return rc;
1475	}
1476	}
1477
1478	VMSTATE enmVMState = VMR3GetState(pVM);
1479	AssertMsgReturn(enmVMState == VMSTATE_SUSPENDING_LS,
1480	("%s\n", VMR3GetStateName(enmVMState)),
1481	VERR_INTERNAL_ERROR_4);
1482
1483	/*
1484	* Only EMT(0) have work to do since it's last thru here.
1485	*/
1486	if (pVCpu->idCpu == 0)
1487	{
1488	vmR3SuspendDoWork(pVM);
1489	int rc = vmR3TrySetState(pVM, "VMR3Suspend", 1,
1490	VMSTATE_SUSPENDED_LS, VMSTATE_SUSPENDING_LS);
1491	if (RT_FAILURE(rc))
1492	return VERR_INTERNAL_ERROR_3;
1493
1494	*pfSuspended = true;
1495	}
1496
1497	return VINF_EM_SUSPEND;
1498	}
1499
1500
1501	/**
1502	* EMT rendezvous worker that VMR3Save and VMR3Teleport uses to clean up a
1503	* SSMR3LiveDoStep1 failure.
1504	*
1505	* Doing this as a rendezvous operation avoids all annoying transition
1506	* states.
1507	*
1508	* @returns VERR_VM_INVALID_VM_STATE, VINF_SUCCESS or some specific VERR_SSM_*
1509	* status code. (This is a strict return code, see FNVMMEMTRENDEZVOUS.)
1510	*
1511	* @param pVM The VM handle.
1512	* @param pVCpu The VMCPU handle of the EMT.
1513	* @param pvUser The pfSuspended argument of vmR3SaveTeleport.
1514	*/
1515	static DECLCALLBACK(VBOXSTRICTRC) vmR3LiveDoStep1Cleanup(PVM pVM, PVMCPU pVCpu, void *pvUser)
1516	{
1517	LogFlow(("vmR3LiveDoStep1Cleanup: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1518	bool pfSuspended = (bool )pvUser;
1519	NOREF(pVCpu);
1520
1521	int rc = vmR3TrySetState(pVM, "vmR3LiveDoStep1Cleanup", 8,
1522	VMSTATE_OFF, VMSTATE_OFF_LS, /* 1 */
1523	VMSTATE_FATAL_ERROR, VMSTATE_FATAL_ERROR_LS, /* 2 */
1524	VMSTATE_GURU_MEDITATION, VMSTATE_GURU_MEDITATION_LS, /* 3 */
1525	VMSTATE_SUSPENDED, VMSTATE_SUSPENDED_LS, /* 4 */
1526	VMSTATE_SUSPENDED, VMSTATE_SAVING,
1527	VMSTATE_SUSPENDED, VMSTATE_SUSPENDED_EXT_LS,
1528	VMSTATE_RUNNING, VMSTATE_RUNNING_LS,
1529	VMSTATE_DEBUGGING, VMSTATE_DEBUGGING_LS);
1530	if (rc == 1)
1531	rc = VERR_SSM_LIVE_POWERED_OFF;
1532	else if (rc == 2)
1533	rc = VERR_SSM_LIVE_FATAL_ERROR;
1534	else if (rc == 3)
1535	rc = VERR_SSM_LIVE_GURU_MEDITATION;
1536	else if (rc == 4)
1537	{
1538	*pfSuspended = true;
1539	rc = VINF_SUCCESS;
1540	}
1541	else if (rc > 0)
1542	rc = VINF_SUCCESS;
1543	return rc;
1544	}
1545
1546
1547	/**
1548	* EMT(0) worker for VMR3Save and VMR3Teleport that completes the live save.
1549	*
1550	* @returns VBox status code.
1551	* @retval VINF_SSM_LIVE_SUSPENDED if VMR3Suspend was called.
1552	*
1553	* @param pVM The VM handle.
1554	* @param pSSM The handle of saved state operation.
1555	*
1556	* @thread EMT(0)
1557	*/
1558	static DECLCALLBACK(int) vmR3LiveDoStep2(PVM pVM, PSSMHANDLE pSSM)
1559	{
1560	LogFlow(("vmR3LiveDoStep2: pVM=%p pSSM=%p\n", pVM, pSSM));
1561	VM_ASSERT_EMT0(pVM);
1562
1563	/*
1564	* Advance the state and mark if VMR3Suspend was called.
1565	*/
1566	int rc = VINF_SUCCESS;
1567	VMSTATE enmVMState = VMR3GetState(pVM);
1568	if (enmVMState == VMSTATE_SUSPENDED_LS)
1569	vmR3SetState(pVM, VMSTATE_SAVING, VMSTATE_SUSPENDED_LS);
1570	else
1571	{
1572	if (enmVMState != VMSTATE_SAVING)
1573	vmR3SetState(pVM, VMSTATE_SAVING, VMSTATE_SUSPENDED_EXT_LS);
1574	rc = VINF_SSM_LIVE_SUSPENDED;
1575	}
1576
1577	/*
1578	* Finish up and release the handle. Careful with the status codes.
1579	*/
1580	int rc2 = SSMR3LiveDoStep2(pSSM);
1581	if (rc == VINF_SUCCESS \|\| (RT_FAILURE(rc2) && RT_SUCCESS(rc)))
1582	rc = rc2;
1583
1584	rc2 = SSMR3LiveDone(pSSM);
1585	if (rc == VINF_SUCCESS \|\| (RT_FAILURE(rc2) && RT_SUCCESS(rc)))
1586	rc = rc2;
1587
1588	/*
1589	* Advance to the final state and return.
1590	*/
1591	vmR3SetState(pVM, VMSTATE_SUSPENDED, VMSTATE_SAVING);
1592	Assert(rc > VINF_EM_LAST \|\| rc < VINF_EM_FIRST);
1593	return rc;
1594	}
1595
1596
1597	/**
1598	* Worker for vmR3SaveTeleport that validates the state and calls SSMR3Save or
1599	* SSMR3LiveSave.
1600	*
1601	* @returns VBox status code.
1602	*
1603	* @param pVM The VM handle.
1604	* @param cMsMaxDowntime The maximum downtime given as milliseconds.
1605	* @param pszFilename The name of the file. NULL if pStreamOps is used.
1606	* @param pStreamOps The stream methods. NULL if pszFilename is used.
1607	* @param pvStreamOpsUser The user argument to the stream methods.
1608	* @param enmAfter What to do afterwards.
1609	* @param pfnProgress Progress callback. Optional.
1610	* @param pvProgressUser User argument for the progress callback.
1611	* @param ppSSM Where to return the saved state handle in case of a
1612	* live snapshot scenario.
1613	* @thread EMT
1614	*/
1615	static DECLCALLBACK(int) vmR3Save(PVM pVM, uint32_t cMsMaxDowntime, const char pszFilename, PCSSMSTRMOPS pStreamOps, void pvStreamOpsUser,
1616	SSMAFTER enmAfter, PFNVMPROGRESS pfnProgress, void pvProgressUser, PSSMHANDLE ppSSM)
1617	{
1618	LogFlow(("vmR3Save: pVM=%p cMsMaxDowntime=%u pszFilename=%p:{%s} pStreamOps=%p pvStreamOpsUser=%p enmAfter=%d pfnProgress=%p pvProgressUser=%p ppSSM=%p\n",
1619	pVM, cMsMaxDowntime, pszFilename, pszFilename, pStreamOps, pvStreamOpsUser, enmAfter, pfnProgress, pvProgressUser, ppSSM));
1620
1621	/*
1622	* Validate input.
1623	*/
1624	AssertPtrNull(pszFilename);
1625	AssertPtrNull(pStreamOps);
1626	AssertPtr(pVM);
1627	Assert( enmAfter == SSMAFTER_DESTROY
1628	\|\| enmAfter == SSMAFTER_CONTINUE
1629	\|\| enmAfter == SSMAFTER_TELEPORT);
1630	AssertPtr(ppSSM);
1631	*ppSSM = NULL;
1632
1633	/*
1634	* Change the state and perform/start the saving.
1635	*/
1636	int rc = vmR3TrySetState(pVM, "VMR3Save", 2,
1637	VMSTATE_SAVING, VMSTATE_SUSPENDED,
1638	VMSTATE_RUNNING_LS, VMSTATE_RUNNING);
1639	if (rc == 1 && enmAfter != SSMAFTER_TELEPORT)
1640	{
1641	Assert(!pStreamOps);
1642	rc = SSMR3Save(pVM, pszFilename, enmAfter, pfnProgress, pvProgressUser);
1643	vmR3SetState(pVM, VMSTATE_SUSPENDED, VMSTATE_SAVING);
1644	}
1645	else if (rc == 2 \|\| enmAfter == SSMAFTER_TELEPORT)
1646	{
1647	if (enmAfter == SSMAFTER_TELEPORT)
1648	pVM->vm.s.fTeleportedAndNotFullyResumedYet = true;
1649	rc = SSMR3LiveSave(pVM, cMsMaxDowntime, pszFilename, pStreamOps, pvStreamOpsUser,
1650	enmAfter, pfnProgress, pvProgressUser, ppSSM);
1651	/* (We're not subject to cancellation just yet.) */
1652	}
1653	else
1654	Assert(RT_FAILURE(rc));
1655	return rc;
1656	}
1657
1658
1659	/**
1660	* Commmon worker for VMR3Save and VMR3Teleport.
1661	*
1662	* @returns VBox status code.
1663	*
1664	* @param pVM The VM handle.
1665	* @param cMsMaxDowntime The maximum downtime given as milliseconds.
1666	* @param pszFilename The name of the file. NULL if pStreamOps is used.
1667	* @param pStreamOps The stream methods. NULL if pszFilename is used.
1668	* @param pvStreamOpsUser The user argument to the stream methods.
1669	* @param enmAfter What to do afterwards.
1670	* @param pfnProgress Progress callback. Optional.
1671	* @param pvProgressUser User argument for the progress callback.
1672	* @param pfSuspended Set if we suspended the VM.
1673	*
1674	* @thread Non-EMT
1675	*/
1676	static int vmR3SaveTeleport(PVM pVM, uint32_t cMsMaxDowntime,
1677	const char pszFilename, PCSSMSTRMOPS pStreamOps, void pvStreamOpsUser,
1678	SSMAFTER enmAfter, PFNVMPROGRESS pfnProgress, void pvProgressUser, bool pfSuspended)
1679	{
1680	/*
1681	* Request the operation in EMT(0).
1682	*/
1683	PSSMHANDLE pSSM;
1684	int rc = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/,
1685	(PFNRT)vmR3Save, 9, pVM, cMsMaxDowntime, pszFilename, pStreamOps, pvStreamOpsUser,
1686	enmAfter, pfnProgress, pvProgressUser, &pSSM);
1687	if ( RT_SUCCESS(rc)
1688	&& pSSM)
1689	{
1690	/*
1691	* Live snapshot.
1692	*
1693	* The state handling here is kind of tricky, doing it on EMT(0) helps
1694	* a bit. See the VMSTATE diagram for details.
1695	*/
1696	rc = SSMR3LiveDoStep1(pSSM);
1697	if (RT_SUCCESS(rc))
1698	{
1699	if (VMR3GetState(pVM) != VMSTATE_SAVING)
1700	for (;;)
1701	{
1702	/* Try suspend the VM. */
1703	rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
1704	vmR3LiveDoSuspend, pfSuspended);
1705	if (rc != VERR_TRY_AGAIN)
1706	break;
1707
1708	/* Wait for the state to change. */
1709	RTThreadSleep(250); /** @todo Live Migration: fix this polling wait by some smart use of multiple release event semaphores.. */
1710	}
1711	if (RT_SUCCESS(rc))
1712	rc = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/, (PFNRT)vmR3LiveDoStep2, 2, pVM, pSSM);
1713	else
1714	{
1715	int rc2 = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/, (PFNRT)SSMR3LiveDone, 1, pSSM);
1716	AssertMsg(rc2 == rc, ("%Rrc != %Rrc\n", rc2, rc));
1717	}
1718	}
1719	else
1720	{
1721	int rc2 = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/, (PFNRT)SSMR3LiveDone, 1, pSSM);
1722	AssertMsg(rc2 == rc, ("%Rrc != %Rrc\n", rc2, rc));
1723
1724	rc2 = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, vmR3LiveDoStep1Cleanup, pfSuspended);
1725	if (RT_FAILURE(rc2) && rc == VERR_SSM_CANCELLED)
1726	rc = rc2;
1727	}
1728	}
1729
1730	return rc;
1731	}
1732
1733
1734	/**
1735	* Save current VM state.
1736	*
1737	* Can be used for both saving the state and creating snapshots.
1738	*
1739	* When called for a VM in the Running state, the saved state is created live
1740	* and the VM is only suspended when the final part of the saving is preformed.
1741	* The VM state will not be restored to Running in this case and it's up to the
1742	* caller to call VMR3Resume if this is desirable. (The rational is that the
1743	* caller probably wish to reconfigure the disks before resuming the VM.)
1744	*
1745	* @returns VBox status code.
1746	*
1747	* @param pVM The VM which state should be saved.
1748	* @param pszFilename The name of the save state file.
1749	* @param fContinueAfterwards Whether continue execution afterwards or not.
1750	* When in doubt, set this to true.
1751	* @param pfnProgress Progress callback. Optional.
1752	* @param pvUser User argument for the progress callback.
1753	* @param pfSuspended Set if we suspended the VM.
1754	*
1755	* @thread Non-EMT.
1756	* @vmstate Suspended or Running
1757	* @vmstateto Saving+Suspended or
1758	* RunningLS+SuspeningLS+SuspendedLS+Saving+Suspended.
1759	*/
1760	VMMR3DECL(int) VMR3Save(PVM pVM, const char *pszFilename, bool fContinueAfterwards,
1761	PFNVMPROGRESS pfnProgress, void pvUser, bool pfSuspended)
1762	{
1763	LogFlow(("VMR3Save: pVM=%p pszFilename=%p:{%s} fContinueAfterwards=%RTbool pfnProgress=%p pvUser=%p pfSuspended=%p\n",
1764	pVM, pszFilename, pszFilename, fContinueAfterwards, pfnProgress, pvUser, pfSuspended));
1765
1766	/*
1767	* Validate input.
1768	*/
1769	AssertPtr(pfSuspended);
1770	*pfSuspended = false;
1771	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1772	VM_ASSERT_OTHER_THREAD(pVM);
1773	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
1774	AssertReturn(*pszFilename, VERR_INVALID_PARAMETER);
1775	AssertPtrNullReturn(pfnProgress, VERR_INVALID_POINTER);
1776
1777	/*
1778	* Join paths with VMR3Teleport.
1779	*/
1780	SSMAFTER enmAfter = fContinueAfterwards ? SSMAFTER_CONTINUE : SSMAFTER_DESTROY;
1781	int rc = vmR3SaveTeleport(pVM, 250 /cMsMaxDowntime/,
1782	pszFilename, NULL /pStreamOps/, NULL /pvStreamOpsUser/,
1783	enmAfter, pfnProgress, pvUser, pfSuspended);
1784	LogFlow(("VMR3Save: returns %Rrc (pfSuspended=%RTbool)\n", rc, pfSuspended));
1785	return rc;
1786	}
1787
1788
1789	/**
1790	* Teleport the VM (aka live migration).
1791	*
1792	* @returns VBox status code.
1793	*
1794	* @param pVM The VM which state should be saved.
1795	* @param cMsMaxDowntime The maximum downtime given as milliseconds.
1796	* @param pStreamOps The stream methods.
1797	* @param pvStreamOpsUser The user argument to the stream methods.
1798	* @param pfnProgress Progress callback. Optional.
1799	* @param pvProgressUser User argument for the progress callback.
1800	* @param pfSuspended Set if we suspended the VM.
1801	*
1802	* @thread Non-EMT.
1803	* @vmstate Suspended or Running
1804	* @vmstateto Saving+Suspended or
1805	* RunningLS+SuspeningLS+SuspendedLS+Saving+Suspended.
1806	*/
1807	VMMR3DECL(int) VMR3Teleport(PVM pVM, uint32_t cMsMaxDowntime, PCSSMSTRMOPS pStreamOps, void *pvStreamOpsUser,
1808	PFNVMPROGRESS pfnProgress, void pvProgressUser, bool pfSuspended)
1809	{
1810	LogFlow(("VMR3Teleport: pVM=%p cMsMaxDowntime=%u pStreamOps=%p pvStreamOps=%p pfnProgress=%p pvProgressUser=%p\n",
1811	pVM, cMsMaxDowntime, pStreamOps, pvStreamOpsUser, pfnProgress, pvProgressUser));
1812
1813	/*
1814	* Validate input.
1815	*/
1816	AssertPtr(pfSuspended);
1817	*pfSuspended = false;
1818	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1819	VM_ASSERT_OTHER_THREAD(pVM);
1820	AssertPtrReturn(pStreamOps, VERR_INVALID_POINTER);
1821	AssertPtrNullReturn(pfnProgress, VERR_INVALID_POINTER);
1822
1823	/*
1824	* Join paths with VMR3Save.
1825	*/
1826	int rc = vmR3SaveTeleport(pVM, cMsMaxDowntime,
1827	NULL /pszFilename/, pStreamOps, pvStreamOpsUser,
1828	SSMAFTER_TELEPORT, pfnProgress, pvProgressUser, pfSuspended);
1829	LogFlow(("VMR3Teleport: returns %Rrc (pfSuspended=%RTbool)\n", rc, pfSuspended));
1830	return rc;
1831	}
1832
1833
1834
1835	/**
1836	* EMT(0) worker for VMR3LoadFromFile and VMR3LoadFromStream.
1837	*
1838	* @returns VBox status code.
1839	*
1840	* @param pVM The VM handle.
1841	* @param pszFilename The name of the file. NULL if pStreamOps is used.
1842	* @param pStreamOps The stream methods. NULL if pszFilename is used.
1843	* @param pvStreamOpsUser The user argument to the stream methods.
1844	* @param pfnProgress Progress callback. Optional.
1845	* @param pvUser User argument for the progress callback.
1846	* @param fTeleporting Indicates whether we're teleporting or not.
1847	*
1848	* @thread EMT.
1849	*/
1850	static DECLCALLBACK(int) vmR3Load(PVM pVM, const char pszFilename, PCSSMSTRMOPS pStreamOps, void pvStreamOpsUser,
1851	PFNVMPROGRESS pfnProgress, void *pvProgressUser, bool fTeleporting)
1852	{
1853	LogFlow(("vmR3Load: pVM=%p pszFilename=%p:{%s} pStreamOps=%p pvStreamOpsUser=%p pfnProgress=%p pvProgressUser=%p fTeleporting=%RTbool\n",
1854	pVM, pszFilename, pszFilename, pStreamOps, pvStreamOpsUser, pfnProgress, pvProgressUser, fTeleporting));
1855
1856	/*
1857	* Validate input (paranoia).
1858	*/
1859	AssertPtr(pVM);
1860	AssertPtrNull(pszFilename);
1861	AssertPtrNull(pStreamOps);
1862	AssertPtrNull(pfnProgress);
1863
1864	/*
1865	* Change the state and perform the load.
1866	*
1867	* Always perform a relocation round afterwards to make sure hypervisor
1868	* selectors and such are correct.
1869	*/
1870	int rc = vmR3TrySetState(pVM, "VMR3Load", 2,
1871	VMSTATE_LOADING, VMSTATE_CREATED,
1872	VMSTATE_LOADING, VMSTATE_SUSPENDED);
1873	if (RT_FAILURE(rc))
1874	return rc;
1875	pVM->vm.s.fTeleportedAndNotFullyResumedYet = fTeleporting;
1876
1877	uint32_t cErrorsPriorToSave = VMR3GetErrorCount(pVM);
1878	rc = SSMR3Load(pVM, pszFilename, pStreamOps, pvStreamOpsUser, SSMAFTER_RESUME, pfnProgress, pvProgressUser);
1879	if (RT_SUCCESS(rc))
1880	{
1881	VMR3Relocate(pVM, 0 /offDelta/);
1882	vmR3SetState(pVM, VMSTATE_SUSPENDED, VMSTATE_LOADING);
1883	}
1884	else
1885	{
1886	pVM->vm.s.fTeleportedAndNotFullyResumedYet = false;
1887	vmR3SetState(pVM, VMSTATE_LOAD_FAILURE, VMSTATE_LOADING);
1888	if (cErrorsPriorToSave == VMR3GetErrorCount(pVM))
1889	rc = VMSetError(pVM, rc, RT_SRC_POS,
1890	N_("Unable to restore the virtual machine's saved state from '%s'. "
1891	"It may be damaged or from an older version of VirtualBox. "
1892	"Please discard the saved state before starting the virtual machine"),
1893	pszFilename);
1894	}
1895
1896	return rc;
1897	}
1898
1899
1900	/**
1901	* Loads a VM state into a newly created VM or a one that is suspended.
1902	*
1903	* To restore a saved state on VM startup, call this function and then resume
1904	* the VM instead of powering it on.
1905	*
1906	* @returns VBox status code.
1907	*
1908	* @param pVM The VM handle.
1909	* @param pszFilename The name of the save state file.
1910	* @param pfnProgress Progress callback. Optional.
1911	* @param pvUser User argument for the progress callback.
1912	*
1913	* @thread Any thread.
1914	* @vmstate Created, Suspended
1915	* @vmstateto Loading+Suspended
1916	*/
1917	VMMR3DECL(int) VMR3LoadFromFile(PVM pVM, const char pszFilename, PFNVMPROGRESS pfnProgress, void pvUser)
1918	{
1919	LogFlow(("VMR3LoadFromFile: pVM=%p pszFilename=%p:{%s} pfnProgress=%p pvUser=%p\n",
1920	pVM, pszFilename, pszFilename, pfnProgress, pvUser));
1921
1922	/*
1923	* Validate input.
1924	*/
1925	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1926	AssertPtrReturn(pszFilename, VERR_INVALID_POINTER);
1927
1928	/*
1929	* Forward the request to EMT(0). No need to setup a rendezvous here
1930	* since there is no execution taking place when this call is allowed.
1931	*/
1932	int rc = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/, (PFNRT)vmR3Load, 7,
1933	pVM, pszFilename, (uintptr_t)NULL /pStreamOps/, (uintptr_t)NULL /pvStreamOpsUser/, pfnProgress, pvUser,
1934	false /fTeleporting/);
1935	LogFlow(("VMR3LoadFromFile: returns %Rrc\n", rc));
1936	return rc;
1937	}
1938
1939
1940	/**
1941	* VMR3LoadFromFile for arbritrary file streams.
1942	*
1943	* @returns VBox status code.
1944	*
1945	* @param pVM The VM handle.
1946	* @param pStreamOps The stream methods.
1947	* @param pvStreamOpsUser The user argument to the stream methods.
1948	* @param pfnProgress Progress callback. Optional.
1949	* @param pvProgressUser User argument for the progress callback.
1950	*
1951	* @thread Any thread.
1952	* @vmstate Created, Suspended
1953	* @vmstateto Loading+Suspended
1954	*/
1955	VMMR3DECL(int) VMR3LoadFromStream(PVM pVM, PCSSMSTRMOPS pStreamOps, void *pvStreamOpsUser,
1956	PFNVMPROGRESS pfnProgress, void *pvProgressUser)
1957	{
1958	LogFlow(("VMR3LoadFromStream: pVM=%p pStreamOps=%p pvStreamOpsUser=%p pfnProgress=%p pvProgressUser=%p\n",
1959	pVM, pStreamOps, pvStreamOpsUser, pfnProgress, pvProgressUser));
1960
1961	/*
1962	* Validate input.
1963	*/
1964	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
1965	AssertPtrReturn(pStreamOps, VERR_INVALID_POINTER);
1966
1967	/*
1968	* Forward the request to EMT(0). No need to setup a rendezvous here
1969	* since there is no execution taking place when this call is allowed.
1970	*/
1971	int rc = VMR3ReqCallWaitU(pVM->pUVM, 0 /idDstCpu/, (PFNRT)vmR3Load, 7,
1972	pVM, (uintptr_t)NULL /pszFilename/, pStreamOps, pvStreamOpsUser, pfnProgress, pvProgressUser,
1973	true /fTeleporting/);
1974	LogFlow(("VMR3LoadFromStream: returns %Rrc\n", rc));
1975	return rc;
1976	}
1977
1978
1979	/**
1980	* EMT rendezvous worker for VMR3PowerOff.
1981	*
1982	* @returns VERR_VM_INVALID_VM_STATE or VINF_EM_OFF. (This is a strict
1983	* return code, see FNVMMEMTRENDEZVOUS.)
1984	*
1985	* @param pVM The VM handle.
1986	* @param pVCpu The VMCPU handle of the EMT.
1987	* @param pvUser Ignored.
1988	*/
1989	static DECLCALLBACK(VBOXSTRICTRC) vmR3PowerOff(PVM pVM, PVMCPU pVCpu, void *pvUser)
1990	{
1991	LogFlow(("vmR3PowerOff: pVM=%p pVCpu=%p/#%u\n", pVM, pVCpu, pVCpu->idCpu));
1992	Assert(!pvUser); NOREF(pvUser);
1993
1994	/*
1995	* The first EMT thru here will change the state to PoweringOff.
1996	*/
1997	if (pVCpu->idCpu == pVM->cCpus - 1)
1998	{
1999	int rc = vmR3TrySetState(pVM, "VMR3PowerOff", 11,
2000	VMSTATE_POWERING_OFF, VMSTATE_RUNNING, /* 1 */
2001	VMSTATE_POWERING_OFF, VMSTATE_SUSPENDED, /* 2 */
2002	VMSTATE_POWERING_OFF, VMSTATE_DEBUGGING, /* 3 */
2003	VMSTATE_POWERING_OFF, VMSTATE_LOAD_FAILURE, /* 4 */
2004	VMSTATE_POWERING_OFF, VMSTATE_GURU_MEDITATION, /* 5 */
2005	VMSTATE_POWERING_OFF, VMSTATE_FATAL_ERROR, /* 6 */
2006	VMSTATE_POWERING_OFF, VMSTATE_CREATED, /* 7 / /* @todo update the diagram! */
2007	VMSTATE_POWERING_OFF_LS, VMSTATE_RUNNING_LS, /* 8 */
2008	VMSTATE_POWERING_OFF_LS, VMSTATE_DEBUGGING_LS, /* 9 */
2009	VMSTATE_POWERING_OFF_LS, VMSTATE_GURU_MEDITATION_LS,/* 10 */
2010	VMSTATE_POWERING_OFF_LS, VMSTATE_FATAL_ERROR_LS); /* 11 */
2011	if (RT_FAILURE(rc))
2012	return rc;
2013	if (rc >= 7)
2014	SSMR3Cancel(pVM);
2015	}
2016
2017	/*
2018	* Check the state.
2019	*/
2020	VMSTATE enmVMState = VMR3GetState(pVM);
2021	AssertMsgReturn( enmVMState == VMSTATE_POWERING_OFF
2022	\|\| enmVMState == VMSTATE_POWERING_OFF_LS,
2023	("%s\n", VMR3GetStateName(enmVMState)),
2024	VERR_VM_INVALID_VM_STATE);
2025
2026	/*
2027	* EMT(0) does the actual power off work here after all the other EMTs
2028	* have been thru and entered the STOPPED state.
2029	*/
2030	VMCPU_SET_STATE(pVCpu, VMCPUSTATE_STOPPED);
2031	if (pVCpu->idCpu == 0)
2032	{
2033	/*
2034	* For debugging purposes, we will log a summary of the guest state at this point.
2035	*/
2036	if (enmVMState != VMSTATE_GURU_MEDITATION)
2037	{
2038	/** @todo SMP support? */
2039	/** @todo make the state dumping at VMR3PowerOff optional. */
2040	RTLogRelPrintf("**************** Guest state at power off ****************\n");
2041	DBGFR3Info(pVM, "cpumguest", "verbose", DBGFR3InfoLogRelHlp());
2042	RTLogRelPrintf("***\n");
2043	DBGFR3Info(pVM, "mode", NULL, DBGFR3InfoLogRelHlp());
2044	RTLogRelPrintf("***\n");
2045	DBGFR3Info(pVM, "activetimers", NULL, DBGFR3InfoLogRelHlp());
2046	RTLogRelPrintf("***\n");
2047	DBGFR3Info(pVM, "gdt", NULL, DBGFR3InfoLogRelHlp());
2048	/** @todo dump guest call stack. */
2049	#if 1 // "temporary" while debugging #1589
2050	RTLogRelPrintf("***\n");
2051	uint32_t esp = CPUMGetGuestESP(pVCpu);
2052	if ( CPUMGetGuestSS(pVCpu) == 0
2053	&& esp < _64K)
2054	{
2055	uint8_t abBuf[PAGE_SIZE];
2056	RTLogRelPrintf("***\n"
2057	"ss:sp=0000:%04x ", esp);
2058	uint32_t Start = esp & ~(uint32_t)63;
2059	int rc = PGMPhysSimpleReadGCPhys(pVM, abBuf, Start, 0x100);
2060	if (RT_SUCCESS(rc))
2061	RTLogRelPrintf("0000:%04x TO 0000:%04x:\n"
2062	"%.*Rhxd\n",
2063	Start, Start + 0x100 - 1,
2064	0x100, abBuf);
2065	else
2066	RTLogRelPrintf("rc=%Rrc\n", rc);
2067
2068	/* grub ... */
2069	if (esp < 0x2000 && esp > 0x1fc0)
2070	{
2071	rc = PGMPhysSimpleReadGCPhys(pVM, abBuf, 0x8000, 0x800);
2072	if (RT_SUCCESS(rc))
2073	RTLogRelPrintf("0000:8000 TO 0000:87ff:\n"
2074	"%.*Rhxd\n",
2075	0x800, abBuf);
2076	}
2077	/* microsoft cdrom hang ... */
2078	if (true)
2079	{
2080	rc = PGMPhysSimpleReadGCPhys(pVM, abBuf, 0x8000, 0x200);
2081	if (RT_SUCCESS(rc))
2082	RTLogRelPrintf("2000:0000 TO 2000:01ff:\n"
2083	"%.*Rhxd\n",
2084	0x200, abBuf);
2085	}
2086	}
2087	#endif
2088	RTLogRelPrintf("************ End of Guest state at power off *************\n");
2089	}
2090
2091	/*
2092	* Perform the power off notifications and advance the state to
2093	* Off or OffLS.
2094	*/
2095	PDMR3PowerOff(pVM);
2096
2097	PUVM pUVM = pVM->pUVM;
2098	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
2099	enmVMState = pVM->enmVMState;
2100	if (enmVMState == VMSTATE_POWERING_OFF_LS)
2101	vmR3SetStateLocked(pVM, pUVM, VMSTATE_OFF_LS, VMSTATE_POWERING_OFF_LS);
2102	else
2103	vmR3SetStateLocked(pVM, pUVM, VMSTATE_OFF, VMSTATE_POWERING_OFF);
2104	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
2105	}
2106	return VINF_EM_OFF;
2107	}
2108
2109
2110	/**
2111	* Power off the VM.
2112	*
2113	* @returns VBox status code. When called on EMT, this will be a strict status
2114	* code that has to be propagated up the call stack.
2115	*
2116	* @param pVM The handle of the VM to be powered off.
2117	*
2118	* @thread Any thread.
2119	* @vmstate Suspended, Running, Guru Meditation, Load Failure
2120	* @vmstateto Off or OffLS
2121	*/
2122	VMMR3DECL(int) VMR3PowerOff(PVM pVM)
2123	{
2124	LogFlow(("VMR3PowerOff: pVM=%p\n", pVM));
2125	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
2126
2127	/*
2128	* Gather all the EMTs to make sure there are no races before
2129	* changing the VM state.
2130	*/
2131	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
2132	vmR3PowerOff, NULL);
2133	LogFlow(("VMR3PowerOff: returns %Rrc\n", rc));
2134	return rc;
2135	}
2136
2137
2138	/**
2139	* Destroys the VM.
2140	*
2141	* The VM must be powered off (or never really powered on) to call this
2142	* function. The VM handle is destroyed and can no longer be used up successful
2143	* return.
2144	*
2145	* @returns VBox status code.
2146	*
2147	* @param pVM The handle of the VM which should be destroyed.
2148	*
2149	* @thread EMT(0) or any none emulation thread.
2150	* @vmstate Off, Created
2151	* @vmstateto N/A
2152	*/
2153	VMMR3DECL(int) VMR3Destroy(PVM pVM)
2154	{
2155	LogFlow(("VMR3Destroy: pVM=%p\n", pVM));
2156
2157	/*
2158	* Validate input.
2159	*/
2160	if (!pVM)
2161	return VERR_INVALID_PARAMETER;
2162	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
2163	Assert(VMMGetCpuId(pVM) == 0 \|\| VMMGetCpuId(pVM) == NIL_VMCPUID);
2164
2165	/*
2166	* Change VM state to destroying and unlink the VM.
2167	*/
2168	int rc = vmR3TrySetState(pVM, "VMR3Destroy", 1, VMSTATE_DESTROYING, VMSTATE_OFF);
2169	if (RT_FAILURE(rc))
2170	return rc;
2171
2172	/** @todo lock this when we start having multiple machines in a process... */
2173	PUVM pUVM = pVM->pUVM; AssertPtr(pUVM);
2174	if (g_pUVMsHead == pUVM)
2175	g_pUVMsHead = pUVM->pNext;
2176	else
2177	{
2178	PUVM pPrev = g_pUVMsHead;
2179	while (pPrev && pPrev->pNext != pUVM)
2180	pPrev = pPrev->pNext;
2181	AssertMsgReturn(pPrev, ("pUVM=%p / pVM=%p is INVALID!\n", pUVM, pVM), VERR_INVALID_PARAMETER);
2182
2183	pPrev->pNext = pUVM->pNext;
2184	}
2185	pUVM->pNext = NULL;
2186
2187	/*
2188	* Notify registered at destruction listeners.
2189	*/
2190	vmR3AtDtor(pVM);
2191
2192	/*
2193	* EMT(0) does the final cleanup, so if we're it calling VMR3Destroy then
2194	* we'll have to postpone parts of it till later. Otherwise, call
2195	* vmR3Destroy on each of the EMTs in ending with EMT(0) doing the bulk
2196	* of the cleanup.
2197	*/
2198	if (VMMGetCpuId(pVM) == 0)
2199	{
2200	pUVM->vm.s.fEMTDoesTheCleanup = true;
2201	pUVM->vm.s.fTerminateEMT = true;
2202	VM_FF_SET(pVM, VM_FF_TERMINATE);
2203
2204	/* Terminate the other EMTs. */
2205	for (VMCPUID idCpu = 1; idCpu < pVM->cCpus; idCpu++)
2206	{
2207	rc = VMR3ReqCallWaitU(pUVM, idCpu, (PFNRT)vmR3Destroy, 1, pVM);
2208	AssertLogRelRC(rc);
2209	}
2210	}
2211	else
2212	{
2213	/* vmR3Destroy on all EMTs, ending with EMT(0). */
2214	rc = VMR3ReqCallWaitU(pUVM, VMCPUID_ALL_REVERSE, (PFNRT)vmR3Destroy, 1, pVM);
2215	AssertLogRelRC(rc);
2216
2217	/* Wait for EMTs and destroy the UVM. */
2218	vmR3DestroyUVM(pUVM, 30000);
2219	}
2220
2221	LogFlow(("VMR3Destroy: returns VINF_SUCCESS\n"));
2222	return VINF_SUCCESS;
2223	}
2224
2225
2226	/**
2227	* Internal destruction worker.
2228	*
2229	* This is either called from VMR3Destroy via VMR3ReqCallU or from
2230	* vmR3EmulationThreadWithId when EMT(0) terminates after having called
2231	* VMR3Destroy().
2232	*
2233	* When called on EMT(0), it will performed the great bulk of the destruction.
2234	* When called on the other EMTs, they will do nothing and the whole purpose is
2235	* to return VINF_EM_TERMINATE so they break out of their run loops.
2236	*
2237	* @returns VINF_EM_TERMINATE.
2238	* @param pVM The VM handle.
2239	*/
2240	DECLCALLBACK(int) vmR3Destroy(PVM pVM)
2241	{
2242	PUVM pUVM = pVM->pUVM;
2243	PVMCPU pVCpu = VMMGetCpu(pVM);
2244	Assert(pVCpu);
2245	LogFlow(("vmR3Destroy: pVM=%p pUVM=%p pVCpu=%p idCpu=%u\n", pVM, pUVM, pVCpu, pVCpu->idCpu));
2246
2247	/*
2248	* Only VCPU 0 does the full cleanup.
2249	*/
2250	if (pVCpu->idCpu == 0)
2251	{
2252
2253	/*
2254	* Dump statistics to the log.
2255	*/
2256	#if defined(VBOX_WITH_STATISTICS) \|\| defined(LOG_ENABLED)
2257	RTLogFlags(NULL, "nodisabled nobuffered");
2258	#endif
2259	#ifdef VBOX_WITH_STATISTICS
2260	STAMR3Dump(pVM, "*");
2261	#else
2262	LogRel(("*********************** Statistics ***********************\n"));
2263	STAMR3DumpToReleaseLog(pVM, "*");
2264	LogRel(("******************* End of statistics ********************\n"));
2265	#endif
2266
2267	/*
2268	* Destroy the VM components.
2269	*/
2270	int rc = TMR3Term(pVM);
2271	AssertRC(rc);
2272	#ifdef VBOX_WITH_DEBUGGER
2273	rc = DBGCTcpTerminate(pVM, pUVM->vm.s.pvDBGC);
2274	pUVM->vm.s.pvDBGC = NULL;
2275	#endif
2276	AssertRC(rc);
2277	rc = DBGFR3Term(pVM);
2278	AssertRC(rc);
2279	rc = PDMR3Term(pVM);
2280	AssertRC(rc);
2281	rc = EMR3Term(pVM);
2282	AssertRC(rc);
2283	rc = IOMR3Term(pVM);
2284	AssertRC(rc);
2285	rc = CSAMR3Term(pVM);
2286	AssertRC(rc);
2287	rc = PATMR3Term(pVM);
2288	AssertRC(rc);
2289	rc = TRPMR3Term(pVM);
2290	AssertRC(rc);
2291	rc = SELMR3Term(pVM);
2292	AssertRC(rc);
2293	rc = REMR3Term(pVM);
2294	AssertRC(rc);
2295	rc = HWACCMR3Term(pVM);
2296	AssertRC(rc);
2297	rc = PGMR3Term(pVM);
2298	AssertRC(rc);
2299	rc = VMMR3Term(pVM); /* Terminates the ring-0 code! */
2300	AssertRC(rc);
2301	rc = CPUMR3Term(pVM);
2302	AssertRC(rc);
2303	SSMR3Term(pVM);
2304	rc = PDMR3CritSectTerm(pVM);
2305	AssertRC(rc);
2306	rc = MMR3Term(pVM);
2307	AssertRC(rc);
2308
2309	/*
2310	* We're done in this thread (EMT).
2311	*/
2312	ASMAtomicUoWriteBool(&pUVM->vm.s.fTerminateEMT, true);
2313	ASMAtomicWriteU32(&pVM->fGlobalForcedActions, VM_FF_TERMINATE);
2314	LogFlow(("vmR3Destroy: returning %Rrc\n", VINF_EM_TERMINATE));
2315	}
2316	return VINF_EM_TERMINATE;
2317	}
2318
2319
2320	/**
2321	* Called at the end of the EMT procedure to take care of the final cleanup.
2322	*
2323	* Currently only EMT(0) will do work here. It will destroy the shared VM
2324	* structure if it is still around. If EMT(0) was the caller of VMR3Destroy it
2325	* will destroy UVM and nothing will be left behind upon exit. But if some
2326	* other thread is calling VMR3Destroy, they will clean up UVM after all EMTs
2327	* has exitted.
2328	*
2329	* @param pUVM The UVM handle.
2330	* @param idCpu The virtual CPU id.
2331	*/
2332	void vmR3DestroyFinalBitFromEMT(PUVM pUVM, VMCPUID idCpu)
2333	{
2334	/*
2335	* Only EMT(0) has work to do here.
2336	*/
2337	if (idCpu != 0)
2338	return;
2339	Assert( !pUVM->pVM
2340	\|\| VMMGetCpuId(pUVM->pVM) == 0);
2341
2342	/*
2343	* If we have a shared VM structure, change its state to Terminated and
2344	* tell GVMM to destroy it.
2345	*/
2346	if (pUVM->pVM)
2347	{
2348	vmR3SetState(pUVM->pVM, VMSTATE_TERMINATED, VMSTATE_DESTROYING);
2349	int rc = SUPR3CallVMMR0Ex(pUVM->pVM->pVMR0, 0 /idCpu/, VMMR0_DO_GVMM_DESTROY_VM, 0, NULL);
2350	AssertLogRelRC(rc);
2351	pUVM->pVM = NULL;
2352	}
2353
2354	/*
2355	* If EMT(0) called VMR3Destroy, then it will destroy UVM as well.
2356	*/
2357	if (pUVM->vm.s.fEMTDoesTheCleanup)
2358	vmR3DestroyUVM(pUVM, 30000);
2359	}
2360
2361
2362	/**
2363	* Destroys the UVM portion.
2364	*
2365	* This is called as the final step in the VM destruction or as the cleanup
2366	* in case of a creation failure. If EMT(0) called VMR3Destroy, meaning
2367	* VMINTUSERPERVM::fEMTDoesTheCleanup is true, it will call this as
2368	* vmR3DestroyFinalBitFromEMT completes.
2369	*
2370	* @param pVM VM Handle.
2371	* @param cMilliesEMTWait The number of milliseconds to wait for the emulation
2372	* threads.
2373	*/
2374	static void vmR3DestroyUVM(PUVM pUVM, uint32_t cMilliesEMTWait)
2375	{
2376	/*
2377	* Signal termination of each the emulation threads and
2378	* wait for them to complete.
2379	*/
2380	/* Signal them. */
2381	ASMAtomicUoWriteBool(&pUVM->vm.s.fTerminateEMT, true);
2382	for (VMCPUID i = 0; i < pUVM->cCpus; i++)
2383	{
2384	ASMAtomicUoWriteBool(&pUVM->aCpus[i].vm.s.fTerminateEMT, true);
2385	if (pUVM->pVM)
2386	VM_FF_SET(pUVM->pVM, VM_FF_TERMINATE);
2387	VMR3NotifyGlobalFFU(pUVM, VMNOTIFYFF_FLAGS_DONE_REM);
2388	RTSemEventSignal(pUVM->aCpus[i].vm.s.EventSemWait);
2389	}
2390
2391	/* Wait for them. */
2392	uint64_t NanoTS = RTTimeNanoTS();
2393	RTTHREAD hSelf = RTThreadSelf();
2394	ASMAtomicUoWriteBool(&pUVM->vm.s.fTerminateEMT, true);
2395	for (VMCPUID i = 0; i < pUVM->cCpus; i++)
2396	{
2397	RTTHREAD hThread = pUVM->aCpus[i].vm.s.ThreadEMT;
2398	if ( hThread != NIL_RTTHREAD
2399	&& hThread != hSelf)
2400	{
2401	uint64_t cMilliesElapsed = (RTTimeNanoTS() - NanoTS) / 1000000;
2402	int rc2 = RTThreadWait(hThread,
2403	cMilliesElapsed < cMilliesEMTWait
2404	? RT_MAX(cMilliesEMTWait - cMilliesElapsed, 2000)
2405	: 2000,
2406	NULL);
2407	if (rc2 == VERR_TIMEOUT) /* avoid the assertion when debugging. */
2408	rc2 = RTThreadWait(hThread, 1000, NULL);
2409	AssertLogRelMsgRC(rc2, ("i=%u rc=%Rrc\n", i, rc2));
2410	if (RT_SUCCESS(rc2))
2411	pUVM->aCpus[0].vm.s.ThreadEMT = NIL_RTTHREAD;
2412	}
2413	}
2414
2415	/* Cleanup the semaphores. */
2416	for (VMCPUID i = 0; i < pUVM->cCpus; i++)
2417	{
2418	RTSemEventDestroy(pUVM->aCpus[i].vm.s.EventSemWait);
2419	pUVM->aCpus[i].vm.s.EventSemWait = NIL_RTSEMEVENT;
2420	}
2421
2422	/*
2423	* Free the event semaphores associated with the request packets.
2424	*/
2425	unsigned cReqs = 0;
2426	for (unsigned i = 0; i < RT_ELEMENTS(pUVM->vm.s.apReqFree); i++)
2427	{
2428	PVMREQ pReq = pUVM->vm.s.apReqFree[i];
2429	pUVM->vm.s.apReqFree[i] = NULL;
2430	for (; pReq; pReq = pReq->pNext, cReqs++)
2431	{
2432	pReq->enmState = VMREQSTATE_INVALID;
2433	RTSemEventDestroy(pReq->EventSem);
2434	}
2435	}
2436	Assert(cReqs == pUVM->vm.s.cReqFree); NOREF(cReqs);
2437
2438	/*
2439	* Kill all queued requests. (There really shouldn't be any!)
2440	*/
2441	for (unsigned i = 0; i < 10; i++)
2442	{
2443	PVMREQ pReqHead = (PVMREQ)ASMAtomicXchgPtr((void * volatile *)&pUVM->vm.s.pReqs, NULL);
2444	AssertMsg(!pReqHead, ("This isn't supposed to happen! VMR3Destroy caller has to serialize this.\n"));
2445	if (!pReqHead)
2446	break;
2447	for (PVMREQ pReq = pReqHead; pReq; pReq = pReq->pNext)
2448	{
2449	ASMAtomicUoWriteSize(&pReq->iStatus, VERR_INTERNAL_ERROR);
2450	ASMAtomicWriteSize(&pReq->enmState, VMREQSTATE_INVALID);
2451	RTSemEventSignal(pReq->EventSem);
2452	RTThreadSleep(2);
2453	RTSemEventDestroy(pReq->EventSem);
2454	}
2455	/* give them a chance to respond before we free the request memory. */
2456	RTThreadSleep(32);
2457	}
2458
2459	/*
2460	* Now all queued VCPU requests (again, there shouldn't be any).
2461	*/
2462	for (VMCPUID idCpu = 0; idCpu < pUVM->cCpus; idCpu++)
2463	{
2464	PUVMCPU pUVCpu = &pUVM->aCpus[idCpu];
2465
2466	for (unsigned i = 0; i < 10; i++)
2467	{
2468	PVMREQ pReqHead = (PVMREQ)ASMAtomicXchgPtr((void * volatile *)&pUVCpu->vm.s.pReqs, NULL);
2469	AssertMsg(!pReqHead, ("This isn't supposed to happen! VMR3Destroy caller has to serialize this.\n"));
2470	if (!pReqHead)
2471	break;
2472	for (PVMREQ pReq = pReqHead; pReq; pReq = pReq->pNext)
2473	{
2474	ASMAtomicUoWriteSize(&pReq->iStatus, VERR_INTERNAL_ERROR);
2475	ASMAtomicWriteSize(&pReq->enmState, VMREQSTATE_INVALID);
2476	RTSemEventSignal(pReq->EventSem);
2477	RTThreadSleep(2);
2478	RTSemEventDestroy(pReq->EventSem);
2479	}
2480	/* give them a chance to respond before we free the request memory. */
2481	RTThreadSleep(32);
2482	}
2483	}
2484
2485	/*
2486	* Make sure the VMMR0.r0 module and whatever else is unloaded.
2487	*/
2488	PDMR3TermUVM(pUVM);
2489
2490	/*
2491	* Terminate the support library if initialized.
2492	*/
2493	if (pUVM->vm.s.pSession)
2494	{
2495	int rc = SUPR3Term(false /fForced/);
2496	AssertRC(rc);
2497	pUVM->vm.s.pSession = NIL_RTR0PTR;
2498	}
2499
2500	/*
2501	* Destroy the MM heap and free the UVM structure.
2502	*/
2503	MMR3TermUVM(pUVM);
2504	STAMR3TermUVM(pUVM);
2505
2506	#ifdef LOG_ENABLED
2507	RTLogSetCustomPrefixCallback(NULL, NULL, NULL);
2508	#endif
2509	RTTlsFree(pUVM->vm.s.idxTLS);
2510
2511	ASMAtomicUoWriteU32(&pUVM->u32Magic, UINT32_MAX);
2512	RTMemPageFree(pUVM, sizeof(*pUVM));
2513
2514	RTLogFlush(NULL);
2515	}
2516
2517
2518	/**
2519	* Enumerates the VMs in this process.
2520	*
2521	* @returns Pointer to the next VM.
2522	* @returns NULL when no more VMs.
2523	* @param pVMPrev The previous VM
2524	* Use NULL to start the enumeration.
2525	*/
2526	VMMR3DECL(PVM) VMR3EnumVMs(PVM pVMPrev)
2527	{
2528	/*
2529	* This is quick and dirty. It has issues with VM being
2530	* destroyed during the enumeration.
2531	*/
2532	PUVM pNext;
2533	if (pVMPrev)
2534	pNext = pVMPrev->pUVM->pNext;
2535	else
2536	pNext = g_pUVMsHead;
2537	return pNext ? pNext->pVM : NULL;
2538	}
2539
2540
2541	/**
2542	* Registers an at VM destruction callback.
2543	*
2544	* @returns VBox status code.
2545	* @param pfnAtDtor Pointer to callback.
2546	* @param pvUser User argument.
2547	*/
2548	VMMR3DECL(int) VMR3AtDtorRegister(PFNVMATDTOR pfnAtDtor, void *pvUser)
2549	{
2550	/*
2551	* Check if already registered.
2552	*/
2553	VM_ATDTOR_LOCK();
2554	PVMATDTOR pCur = g_pVMAtDtorHead;
2555	while (pCur)
2556	{
2557	if (pfnAtDtor == pCur->pfnAtDtor)
2558	{
2559	VM_ATDTOR_UNLOCK();
2560	AssertMsgFailed(("Already registered at destruction callback %p!\n", pfnAtDtor));
2561	return VERR_INVALID_PARAMETER;
2562	}
2563
2564	/* next */
2565	pCur = pCur->pNext;
2566	}
2567	VM_ATDTOR_UNLOCK();
2568
2569	/*
2570	* Allocate new entry.
2571	*/
2572	PVMATDTOR pVMAtDtor = (PVMATDTOR)RTMemAlloc(sizeof(*pVMAtDtor));
2573	if (!pVMAtDtor)
2574	return VERR_NO_MEMORY;
2575
2576	VM_ATDTOR_LOCK();
2577	pVMAtDtor->pfnAtDtor = pfnAtDtor;
2578	pVMAtDtor->pvUser = pvUser;
2579	pVMAtDtor->pNext = g_pVMAtDtorHead;
2580	g_pVMAtDtorHead = pVMAtDtor;
2581	VM_ATDTOR_UNLOCK();
2582
2583	return VINF_SUCCESS;
2584	}
2585
2586
2587	/**
2588	* Deregisters an at VM destruction callback.
2589	*
2590	* @returns VBox status code.
2591	* @param pfnAtDtor Pointer to callback.
2592	*/
2593	VMMR3DECL(int) VMR3AtDtorDeregister(PFNVMATDTOR pfnAtDtor)
2594	{
2595	/*
2596	* Find it, unlink it and free it.
2597	*/
2598	VM_ATDTOR_LOCK();
2599	PVMATDTOR pPrev = NULL;
2600	PVMATDTOR pCur = g_pVMAtDtorHead;
2601	while (pCur)
2602	{
2603	if (pfnAtDtor == pCur->pfnAtDtor)
2604	{
2605	if (pPrev)
2606	pPrev->pNext = pCur->pNext;
2607	else
2608	g_pVMAtDtorHead = pCur->pNext;
2609	pCur->pNext = NULL;
2610	VM_ATDTOR_UNLOCK();
2611
2612	RTMemFree(pCur);
2613	return VINF_SUCCESS;
2614	}
2615
2616	/* next */
2617	pPrev = pCur;
2618	pCur = pCur->pNext;
2619	}
2620	VM_ATDTOR_UNLOCK();
2621
2622	return VERR_INVALID_PARAMETER;
2623	}
2624
2625
2626	/**
2627	* Walks the list of at VM destructor callbacks.
2628	* @param pVM The VM which is about to be destroyed.
2629	*/
2630	static void vmR3AtDtor(PVM pVM)
2631	{
2632	/*
2633	* Find it, unlink it and free it.
2634	*/
2635	VM_ATDTOR_LOCK();
2636	for (PVMATDTOR pCur = g_pVMAtDtorHead; pCur; pCur = pCur->pNext)
2637	pCur->pfnAtDtor(pVM, pCur->pvUser);
2638	VM_ATDTOR_UNLOCK();
2639	}
2640
2641
2642	/**
2643	* Worker which checks integrity of some internal structures.
2644	* This is yet another attempt to track down that AVL tree crash.
2645	*/
2646	static void vmR3CheckIntegrity(PVM pVM)
2647	{
2648	#ifdef VBOX_STRICT
2649	int rc = PGMR3CheckIntegrity(pVM);
2650	AssertReleaseRC(rc);
2651	#endif
2652	}
2653
2654
2655	/**
2656	* EMT rendezvous worker for VMR3Reset.
2657	*
2658	* This is called by the emulation threads as a response to the reset request
2659	* issued by VMR3Reset().
2660	*
2661	* @returns VERR_VM_INVALID_VM_STATE, VINF_EM_RESET or VINF_EM_SUSPEND. (This
2662	* is a strict return code, see FNVMMEMTRENDEZVOUS.)
2663	*
2664	* @param pVM The VM handle.
2665	* @param pVCpu The VMCPU handle of the EMT.
2666	* @param pvUser Ignored.
2667	*/
2668	static DECLCALLBACK(VBOXSTRICTRC) vmR3Reset(PVM pVM, PVMCPU pVCpu, void *pvUser)
2669	{
2670	Assert(!pvUser); NOREF(pvUser);
2671
2672	/*
2673	* The first EMT will try change the state to resetting. If this fails,
2674	* we won't get called for the other EMTs.
2675	*/
2676	if (pVCpu->idCpu == pVM->cCpus - 1)
2677	{
2678	int rc = vmR3TrySetState(pVM, "VMR3Reset", 3,
2679	VMSTATE_RESETTING, VMSTATE_RUNNING,
2680	VMSTATE_RESETTING, VMSTATE_SUSPENDED,
2681	VMSTATE_RESETTING_LS, VMSTATE_RUNNING_LS);
2682	if (RT_FAILURE(rc))
2683	return rc;
2684	}
2685
2686	/*
2687	* Check the state.
2688	*/
2689	VMSTATE enmVMState = VMR3GetState(pVM);
2690	AssertLogRelMsgReturn( enmVMState == VMSTATE_RESETTING
2691	\|\| enmVMState == VMSTATE_RESETTING_LS,
2692	("%s\n", VMR3GetStateName(enmVMState)),
2693	VERR_INTERNAL_ERROR_4);
2694
2695	/*
2696	* EMT(0) does the full cleanup after all the other EMTs has been
2697	* thru here and been told to enter the EMSTATE_WAIT_SIPI state.
2698	*
2699	* Because there are per-cpu reset routines and order may/is important,
2700	* the following sequence looks a bit ugly...
2701	*/
2702	if (pVCpu->idCpu == 0)
2703	vmR3CheckIntegrity(pVM);
2704
2705	/* Reset the VCpu state. */
2706	VMCPU_ASSERT_STATE(pVCpu, VMCPUSTATE_STARTED);
2707
2708	/* Clear all pending forced actions. */
2709	VMCPU_FF_CLEAR(pVCpu, VMCPU_FF_ALL_MASK & ~VMCPU_FF_REQUEST);
2710
2711	/*
2712	* Reset the VM components.
2713	*/
2714	if (pVCpu->idCpu == 0)
2715	{
2716	PATMR3Reset(pVM);
2717	CSAMR3Reset(pVM);
2718	PGMR3Reset(pVM); /* We clear VM RAM in PGMR3Reset. It's vital PDMR3Reset is executed
2719	* _afterwards_. E.g. ACPI sets up RAM tables during init/reset. */
2720	/** @todo PGMR3Reset should be called after PDMR3Reset really, because we'll trash OS <-> hardware
2721	* communication structures residing in RAM when done in the other order. I.e. the device must be
2722	* quiesced first, then we clear the memory and plan tables. Probably have to make these things
2723	* explicit in some way, some memory setup pass or something.
2724	* (Example: DevAHCI may assert if memory is zeroed before it've read the FIS.)
2725	*
2726	* @bugref{4467}
2727	*/
2728	MMR3Reset(pVM);
2729	PDMR3Reset(pVM);
2730	SELMR3Reset(pVM);
2731	TRPMR3Reset(pVM);
2732	REMR3Reset(pVM);
2733	IOMR3Reset(pVM);
2734	CPUMR3Reset(pVM);
2735	}
2736	CPUMR3ResetCpu(pVCpu);
2737	if (pVCpu->idCpu == 0)
2738	{
2739	TMR3Reset(pVM);
2740	EMR3Reset(pVM);
2741	HWACCMR3Reset(pVM); /* This must come after PATM, CSAM, CPUM, SELM and TRPM. */
2742
2743	#ifdef LOG_ENABLED
2744	/*
2745	* Debug logging.
2746	*/
2747	RTLogPrintf("\n\nThe VM was reset:\n");
2748	DBGFR3Info(pVM, "cpum", "verbose", NULL);
2749	#endif
2750
2751	/*
2752	* Since EMT(0) is the last to go thru here, it will advance the state.
2753	* When a live save is active, we will move on to SuspendingLS but
2754	* leave it for VMR3Reset to do the actual suspending due to deadlock risks.
2755	*/
2756	PUVM pUVM = pVM->pUVM;
2757	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
2758	enmVMState = pVM->enmVMState;
2759	if (enmVMState == VMSTATE_RESETTING)
2760	{
2761	if (pUVM->vm.s.enmPrevVMState == VMSTATE_SUSPENDED)
2762	vmR3SetStateLocked(pVM, pUVM, VMSTATE_SUSPENDED, VMSTATE_RESETTING);
2763	else
2764	vmR3SetStateLocked(pVM, pUVM, VMSTATE_RUNNING, VMSTATE_RESETTING);
2765	}
2766	else
2767	vmR3SetStateLocked(pVM, pUVM, VMSTATE_SUSPENDING_LS, VMSTATE_RESETTING_LS);
2768	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
2769
2770	vmR3CheckIntegrity(pVM);
2771
2772	/*
2773	* Do the suspend bit as well.
2774	* It only requires some EMT(0) work at present.
2775	*/
2776	if (enmVMState != VMSTATE_RESETTING)
2777	{
2778	vmR3SuspendDoWork(pVM);
2779	vmR3SetState(pVM, VMSTATE_SUSPENDED_LS, VMSTATE_SUSPENDING_LS);
2780	}
2781	}
2782
2783	return enmVMState == VMSTATE_RESETTING
2784	? VINF_EM_RESET
2785	: 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. */
2786	}
2787
2788
2789	/**
2790	* Reset the current VM.
2791	*
2792	* @returns VBox status code.
2793	* @param pVM VM to reset.
2794	*/
2795	VMMR3DECL(int) VMR3Reset(PVM pVM)
2796	{
2797	LogFlow(("VMR3Reset:\n"));
2798	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
2799
2800	/*
2801	* Gather all the EMTs to make sure there are no races before
2802	* changing the VM state.
2803	*/
2804	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_DESCENDING \| VMMEMTRENDEZVOUS_FLAGS_STOP_ON_ERROR,
2805	vmR3Reset, NULL);
2806	LogFlow(("VMR3Reset: returns %Rrc\n", rc));
2807	return rc;
2808	}
2809
2810
2811	/**
2812	* Gets the current VM state.
2813	*
2814	* @returns The current VM state.
2815	* @param pVM VM handle.
2816	* @thread Any
2817	*/
2818	VMMR3DECL(VMSTATE) VMR3GetState(PVM pVM)
2819	{
2820	return pVM->enmVMState;
2821	}
2822
2823
2824	/**
2825	* Gets the state name string for a VM state.
2826	*
2827	* @returns Pointer to the state name. (readonly)
2828	* @param enmState The state.
2829	*/
2830	VMMR3DECL(const char *) VMR3GetStateName(VMSTATE enmState)
2831	{
2832	switch (enmState)
2833	{
2834	case VMSTATE_CREATING: return "CREATING";
2835	case VMSTATE_CREATED: return "CREATED";
2836	case VMSTATE_LOADING: return "LOADING";
2837	case VMSTATE_POWERING_ON: return "POWERING_ON";
2838	case VMSTATE_RESUMING: return "RESUMING";
2839	case VMSTATE_RUNNING: return "RUNNING";
2840	case VMSTATE_RUNNING_LS: return "RUNNING_LS";
2841	case VMSTATE_RESETTING: return "RESETTING";
2842	case VMSTATE_RESETTING_LS: return "RESETTING_LS";
2843	case VMSTATE_SUSPENDED: return "SUSPENDED";
2844	case VMSTATE_SUSPENDED_LS: return "SUSPENDED_LS";
2845	case VMSTATE_SUSPENDED_EXT_LS: return "SUSPENDED_EXT_LS";
2846	case VMSTATE_SUSPENDING: return "SUSPENDING";
2847	case VMSTATE_SUSPENDING_LS: return "SUSPENDING_LS";
2848	case VMSTATE_SUSPENDING_EXT_LS: return "SUSPENDING_EXT_LS";
2849	case VMSTATE_SAVING: return "SAVING";
2850	case VMSTATE_DEBUGGING: return "DEBUGGING";
2851	case VMSTATE_DEBUGGING_LS: return "DEBUGGING_LS";
2852	case VMSTATE_POWERING_OFF: return "POWERING_OFF";
2853	case VMSTATE_POWERING_OFF_LS: return "POWERING_OFF_LS";
2854	case VMSTATE_FATAL_ERROR: return "FATAL_ERROR";
2855	case VMSTATE_FATAL_ERROR_LS: return "FATAL_ERROR_LS";
2856	case VMSTATE_GURU_MEDITATION: return "GURU_MEDITATION";
2857	case VMSTATE_GURU_MEDITATION_LS:return "GURU_MEDITATION_LS";
2858	case VMSTATE_LOAD_FAILURE: return "LOAD_FAILURE";
2859	case VMSTATE_OFF: return "OFF";
2860	case VMSTATE_OFF_LS: return "OFF_LS";
2861	case VMSTATE_DESTROYING: return "DESTROYING";
2862	case VMSTATE_TERMINATED: return "TERMINATED";
2863
2864	default:
2865	AssertMsgFailed(("Unknown state %d\n", enmState));
2866	return "Unknown!\n";
2867	}
2868	}
2869
2870
2871	/**
2872	* Validates the state transition in strict builds.
2873	*
2874	* @returns true if valid, false if not.
2875	*
2876	* @param enmStateOld The old (current) state.
2877	* @param enmStateNew The proposed new state.
2878	*
2879	* @remarks The reference for this is found in doc/vp/VMM.vpp, the VMSTATE
2880	* diagram (under State Machine Diagram).
2881	*/
2882	static bool vmR3ValidateStateTransition(VMSTATE enmStateOld, VMSTATE enmStateNew)
2883	{
2884	#ifdef VBOX_STRICT
2885	switch (enmStateOld)
2886	{
2887	case VMSTATE_CREATING:
2888	AssertMsgReturn(enmStateNew == VMSTATE_CREATED, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2889	break;
2890
2891	case VMSTATE_CREATED:
2892	AssertMsgReturn( enmStateNew == VMSTATE_LOADING
2893	\|\| enmStateNew == VMSTATE_POWERING_ON
2894	\|\| enmStateNew == VMSTATE_POWERING_OFF
2895	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2896	break;
2897
2898	case VMSTATE_LOADING:
2899	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDED
2900	\|\| enmStateNew == VMSTATE_LOAD_FAILURE
2901	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2902	break;
2903
2904	case VMSTATE_POWERING_ON:
2905	AssertMsgReturn( enmStateNew == VMSTATE_RUNNING
2906	/\|\| enmStateNew == VMSTATE_FATAL_ERROR ?/
2907	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2908	break;
2909
2910	case VMSTATE_RESUMING:
2911	AssertMsgReturn( enmStateNew == VMSTATE_RUNNING
2912	/\|\| enmStateNew == VMSTATE_FATAL_ERROR ?/
2913	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2914	break;
2915
2916	case VMSTATE_RUNNING:
2917	AssertMsgReturn( enmStateNew == VMSTATE_POWERING_OFF
2918	\|\| enmStateNew == VMSTATE_SUSPENDING
2919	\|\| enmStateNew == VMSTATE_RESETTING
2920	\|\| enmStateNew == VMSTATE_RUNNING_LS
2921	\|\| enmStateNew == VMSTATE_DEBUGGING
2922	\|\| enmStateNew == VMSTATE_FATAL_ERROR
2923	\|\| enmStateNew == VMSTATE_GURU_MEDITATION
2924	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2925	break;
2926
2927	case VMSTATE_RUNNING_LS:
2928	AssertMsgReturn( enmStateNew == VMSTATE_POWERING_OFF_LS
2929	\|\| enmStateNew == VMSTATE_SUSPENDING_LS
2930	\|\| enmStateNew == VMSTATE_SUSPENDING_EXT_LS
2931	\|\| enmStateNew == VMSTATE_RESETTING_LS
2932	\|\| enmStateNew == VMSTATE_RUNNING
2933	\|\| enmStateNew == VMSTATE_DEBUGGING_LS
2934	\|\| enmStateNew == VMSTATE_FATAL_ERROR_LS
2935	\|\| enmStateNew == VMSTATE_GURU_MEDITATION_LS
2936	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2937	break;
2938
2939	case VMSTATE_RESETTING:
2940	AssertMsgReturn(enmStateNew == VMSTATE_RUNNING, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2941	break;
2942
2943	case VMSTATE_RESETTING_LS:
2944	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDING_LS
2945	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2946	break;
2947
2948	case VMSTATE_SUSPENDING:
2949	AssertMsgReturn(enmStateNew == VMSTATE_SUSPENDED, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2950	break;
2951
2952	case VMSTATE_SUSPENDING_LS:
2953	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDING
2954	\|\| enmStateNew == VMSTATE_SUSPENDED_LS
2955	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2956	break;
2957
2958	case VMSTATE_SUSPENDING_EXT_LS:
2959	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDING
2960	\|\| enmStateNew == VMSTATE_SUSPENDED_EXT_LS
2961	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2962	break;
2963
2964	case VMSTATE_SUSPENDED:
2965	AssertMsgReturn( enmStateNew == VMSTATE_POWERING_OFF
2966	\|\| enmStateNew == VMSTATE_SAVING
2967	\|\| enmStateNew == VMSTATE_RESETTING
2968	\|\| enmStateNew == VMSTATE_RESUMING
2969	\|\| enmStateNew == VMSTATE_LOADING
2970	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2971	break;
2972
2973	case VMSTATE_SUSPENDED_LS:
2974	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDED
2975	\|\| enmStateNew == VMSTATE_SAVING
2976	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2977	break;
2978
2979	case VMSTATE_SUSPENDED_EXT_LS:
2980	AssertMsgReturn( enmStateNew == VMSTATE_SUSPENDED
2981	\|\| enmStateNew == VMSTATE_SAVING
2982	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2983	break;
2984
2985	case VMSTATE_SAVING:
2986	AssertMsgReturn(enmStateNew == VMSTATE_SUSPENDED, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2987	break;
2988
2989	case VMSTATE_DEBUGGING:
2990	AssertMsgReturn( enmStateNew == VMSTATE_RUNNING
2991	\|\| enmStateNew == VMSTATE_POWERING_OFF
2992	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
2993	break;
2994
2995	case VMSTATE_DEBUGGING_LS:
2996	AssertMsgReturn( enmStateNew == VMSTATE_DEBUGGING
2997	\|\| enmStateNew == VMSTATE_RUNNING_LS
2998	\|\| enmStateNew == VMSTATE_POWERING_OFF_LS
2999	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3000	break;
3001
3002	case VMSTATE_POWERING_OFF:
3003	AssertMsgReturn(enmStateNew == VMSTATE_OFF, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3004	break;
3005
3006	case VMSTATE_POWERING_OFF_LS:
3007	AssertMsgReturn( enmStateNew == VMSTATE_POWERING_OFF
3008	\|\| enmStateNew == VMSTATE_OFF_LS
3009	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3010	break;
3011
3012	case VMSTATE_OFF:
3013	AssertMsgReturn(enmStateNew == VMSTATE_DESTROYING, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3014	break;
3015
3016	case VMSTATE_OFF_LS:
3017	AssertMsgReturn(enmStateNew == VMSTATE_OFF, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3018	break;
3019
3020	case VMSTATE_FATAL_ERROR:
3021	AssertMsgReturn(enmStateNew == VMSTATE_POWERING_OFF, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3022	break;
3023
3024	case VMSTATE_FATAL_ERROR_LS:
3025	AssertMsgReturn( enmStateNew == VMSTATE_FATAL_ERROR
3026	\|\| enmStateNew == VMSTATE_POWERING_OFF_LS
3027	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3028	break;
3029
3030	case VMSTATE_GURU_MEDITATION:
3031	AssertMsgReturn( enmStateNew == VMSTATE_DEBUGGING
3032	\|\| enmStateNew == VMSTATE_POWERING_OFF
3033	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3034	break;
3035
3036	case VMSTATE_GURU_MEDITATION_LS:
3037	AssertMsgReturn( enmStateNew == VMSTATE_GURU_MEDITATION
3038	\|\| enmStateNew == VMSTATE_DEBUGGING_LS
3039	\|\| enmStateNew == VMSTATE_POWERING_OFF_LS
3040	, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3041	break;
3042
3043	case VMSTATE_LOAD_FAILURE:
3044	AssertMsgReturn(enmStateNew == VMSTATE_POWERING_OFF, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3045	break;
3046
3047	case VMSTATE_DESTROYING:
3048	AssertMsgReturn(enmStateNew == VMSTATE_TERMINATED, ("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3049	break;
3050
3051	case VMSTATE_TERMINATED:
3052	default:
3053	AssertMsgFailedReturn(("%s -> %s\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)), false);
3054	break;
3055	}
3056	#endif /* VBOX_STRICT */
3057	return true;
3058	}
3059
3060
3061	/**
3062	* Does the state change callouts.
3063	*
3064	* The caller owns the AtStateCritSect.
3065	*
3066	* @param pVM The VM handle.
3067	* @param pUVM The UVM handle.
3068	* @param enmStateNew The New state.
3069	* @param enmStateOld The old state.
3070	*/
3071	static void vmR3DoAtState(PVM pVM, PUVM pUVM, VMSTATE enmStateNew, VMSTATE enmStateOld)
3072	{
3073	LogRel(("Changing the VM state from '%s' to '%s'.\n", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)));
3074
3075	for (PVMATSTATE pCur = pUVM->vm.s.pAtState; pCur; pCur = pCur->pNext)
3076	{
3077	pCur->pfnAtState(pVM, enmStateNew, enmStateOld, pCur->pvUser);
3078	if ( enmStateNew != VMSTATE_DESTROYING
3079	&& pVM->enmVMState == VMSTATE_DESTROYING)
3080	break;
3081	AssertMsg(pVM->enmVMState == enmStateNew,
3082	("You are not allowed to change the state while in the change callback, except "
3083	"from destroying the VM. There are restrictions in the way the state changes "
3084	"are propagated up to the EM execution loop and it makes the program flow very "
3085	"difficult to follow. (%s, expected %s, old %s)\n",
3086	VMR3GetStateName(pVM->enmVMState), VMR3GetStateName(enmStateNew),
3087	VMR3GetStateName(enmStateOld)));
3088	}
3089	}
3090
3091
3092	/**
3093	* Sets the current VM state, with the AtStatCritSect already entered.
3094	*
3095	* @param pVM The VM handle.
3096	* @param pUVM The UVM handle.
3097	* @param enmStateNew The new state.
3098	* @param enmStateOld The old state.
3099	*/
3100	static void vmR3SetStateLocked(PVM pVM, PUVM pUVM, VMSTATE enmStateNew, VMSTATE enmStateOld)
3101	{
3102	vmR3ValidateStateTransition(enmStateOld, enmStateNew);
3103
3104	AssertMsg(pVM->enmVMState == enmStateOld,
3105	("%s != %s\n", VMR3GetStateName(pVM->enmVMState), VMR3GetStateName(enmStateOld)));
3106	pUVM->vm.s.enmPrevVMState = enmStateOld;
3107	pVM->enmVMState = enmStateNew;
3108
3109	vmR3DoAtState(pVM, pUVM, enmStateNew, enmStateOld);
3110	}
3111
3112
3113	/**
3114	* Sets the current VM state.
3115	*
3116	* @param pVM VM handle.
3117	* @param enmStateNew The new state.
3118	* @param enmStateOld The old state (for asserting only).
3119	*/
3120	static void vmR3SetState(PVM pVM, VMSTATE enmStateNew, VMSTATE enmStateOld)
3121	{
3122	PUVM pUVM = pVM->pUVM;
3123	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
3124
3125	AssertMsg(pVM->enmVMState == enmStateOld,
3126	("%s != %s\n", VMR3GetStateName(pVM->enmVMState), VMR3GetStateName(enmStateOld)));
3127	vmR3SetStateLocked(pVM, pUVM, enmStateNew, pVM->enmVMState);
3128
3129	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3130	}
3131
3132
3133	/**
3134	* Tries to perform a state transition.
3135	*
3136	* @returns The 1-based ordinal of the succeeding transition.
3137	* VERR_VM_INVALID_VM_STATE and Assert+LogRel on failure.
3138	*
3139	* @param pVM The VM handle.
3140	* @param pszWho Who is trying to change it.
3141	* @param cTransitions The number of transitions in the ellipsis.
3142	* @param ... Transition pairs; new, old.
3143	*/
3144	static int vmR3TrySetState(PVM pVM, const char *pszWho, unsigned cTransitions, ...)
3145	{
3146	va_list va;
3147	VMSTATE enmStateNew = VMSTATE_CREATED;
3148	VMSTATE enmStateOld = VMSTATE_CREATED;
3149
3150	#ifdef VBOX_STRICT
3151	/*
3152	* Validate the input first.
3153	*/
3154	va_start(va, cTransitions);
3155	for (unsigned i = 0; i < cTransitions; i++)
3156	{
3157	enmStateNew = (VMSTATE)va_arg(va, /VMSTATE/int);
3158	enmStateOld = (VMSTATE)va_arg(va, /VMSTATE/int);
3159	vmR3ValidateStateTransition(enmStateOld, enmStateNew);
3160	}
3161	va_end(va);
3162	#endif
3163
3164	/*
3165	* Grab the lock and see if any of the proposed transisions works out.
3166	*/
3167	va_start(va, cTransitions);
3168	int rc = VERR_VM_INVALID_VM_STATE;
3169	PUVM pUVM = pVM->pUVM;
3170	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
3171
3172	VMSTATE enmStateCur = pVM->enmVMState;
3173
3174	for (unsigned i = 0; i < cTransitions; i++)
3175	{
3176	enmStateNew = (VMSTATE)va_arg(va, /VMSTATE/int);
3177	enmStateOld = (VMSTATE)va_arg(va, /VMSTATE/int);
3178	if (enmStateCur == enmStateOld)
3179	{
3180	vmR3SetStateLocked(pVM, pUVM, enmStateNew, enmStateOld);
3181	rc = i + 1;
3182	break;
3183	}
3184	}
3185
3186	if (RT_FAILURE(rc))
3187	{
3188	/*
3189	* Complain about it.
3190	*/
3191	if (cTransitions == 1)
3192	{
3193	LogRel(("%s: %s -> %s failed, because the VM state is actually %s\n",
3194	pszWho, VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew), VMR3GetStateName(enmStateCur)));
3195	VMSetError(pVM, VERR_VM_INVALID_VM_STATE, RT_SRC_POS,
3196	N_("%s failed because the VM state is %s instead of %s"),
3197	pszWho, VMR3GetStateName(enmStateCur), VMR3GetStateName(enmStateOld));
3198	AssertMsgFailed(("%s: %s -> %s failed, because the VM state is actually %s\n",
3199	pszWho, VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew), VMR3GetStateName(enmStateCur)));
3200	}
3201	else
3202	{
3203	va_end(va);
3204	va_start(va, cTransitions);
3205	LogRel(("%s:\n", pszWho));
3206	for (unsigned i = 0; i < cTransitions; i++)
3207	{
3208	enmStateNew = (VMSTATE)va_arg(va, /VMSTATE/int);
3209	enmStateOld = (VMSTATE)va_arg(va, /VMSTATE/int);
3210	LogRel(("%s%s -> %s",
3211	i ? ", " : " ", VMR3GetStateName(enmStateOld), VMR3GetStateName(enmStateNew)));
3212	}
3213	LogRel((" failed, because the VM state is actually %s\n", VMR3GetStateName(enmStateCur)));
3214	VMSetError(pVM, VERR_VM_INVALID_VM_STATE, RT_SRC_POS,
3215	N_("%s failed because the current VM state, %s, was not found in the state transition table"),
3216	pszWho, VMR3GetStateName(enmStateCur), VMR3GetStateName(enmStateOld));
3217	AssertMsgFailed(("%s - state=%s, see release log for full details. Check the cTransitions passed us.\n",
3218	pszWho, VMR3GetStateName(enmStateCur)));
3219	}
3220	}
3221
3222	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3223	va_end(va);
3224	Assert(rc > 0 \|\| rc < 0);
3225	return rc;
3226	}
3227
3228
3229	/**
3230	* Flag a guru meditation ... a hack.
3231	*
3232	* @param pVM The VM handle
3233	*
3234	* @todo Rewrite this part. The guru meditation should be flagged
3235	* immediately by the VMM and not by VMEmt.cpp when it's all over.
3236	*/
3237	void vmR3SetGuruMeditation(PVM pVM)
3238	{
3239	PUVM pUVM = pVM->pUVM;
3240	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
3241
3242	VMSTATE enmStateCur = pVM->enmVMState;
3243	if (enmStateCur == VMSTATE_RUNNING)
3244	vmR3SetStateLocked(pVM, pUVM, VMSTATE_GURU_MEDITATION, VMSTATE_RUNNING);
3245	else if (enmStateCur == VMSTATE_RUNNING_LS)
3246	{
3247	vmR3SetStateLocked(pVM, pUVM, VMSTATE_GURU_MEDITATION_LS, VMSTATE_RUNNING_LS);
3248	SSMR3Cancel(pVM);
3249	}
3250
3251	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3252	}
3253
3254
3255	/**
3256	* Checks if the VM was teleported and hasn't been fully resumed yet.
3257	*
3258	* This applies to both sides of the teleportation since we may leave a working
3259	* clone behind and the user is allowed to resume this...
3260	*
3261	* @returns true / false.
3262	* @param pVM The VM handle.
3263	* @thread Any thread.
3264	*/
3265	VMMR3DECL(bool) VMR3TeleportedAndNotFullyResumedYet(PVM pVM)
3266	{
3267	VM_ASSERT_VALID_EXT_RETURN(pVM, false);
3268	return pVM->vm.s.fTeleportedAndNotFullyResumedYet;
3269	}
3270
3271
3272	/**
3273	* Registers a VM state change callback.
3274	*
3275	* You are not allowed to call any function which changes the VM state from a
3276	* state callback, except VMR3Destroy().
3277	*
3278	* @returns VBox status code.
3279	* @param pVM VM handle.
3280	* @param pfnAtState Pointer to callback.
3281	* @param pvUser User argument.
3282	* @thread Any.
3283	*/
3284	VMMR3DECL(int) VMR3AtStateRegister(PVM pVM, PFNVMATSTATE pfnAtState, void *pvUser)
3285	{
3286	LogFlow(("VMR3AtStateRegister: pfnAtState=%p pvUser=%p\n", pfnAtState, pvUser));
3287
3288	/*
3289	* Validate input.
3290	*/
3291	AssertPtrReturn(pfnAtState, VERR_INVALID_PARAMETER);
3292	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
3293
3294	/*
3295	* Allocate a new record.
3296	*/
3297	PUVM pUVM = pVM->pUVM;
3298	PVMATSTATE pNew = (PVMATSTATE)MMR3HeapAllocU(pUVM, MM_TAG_VM, sizeof(*pNew));
3299	if (!pNew)
3300	return VERR_NO_MEMORY;
3301
3302	/* fill */
3303	pNew->pfnAtState = pfnAtState;
3304	pNew->pvUser = pvUser;
3305
3306	/* insert */
3307	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
3308	pNew->pNext = *pUVM->vm.s.ppAtStateNext;
3309	*pUVM->vm.s.ppAtStateNext = pNew;
3310	pUVM->vm.s.ppAtStateNext = &pNew->pNext;
3311	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3312
3313	return VINF_SUCCESS;
3314	}
3315
3316
3317	/**
3318	* Deregisters a VM state change callback.
3319	*
3320	* @returns VBox status code.
3321	* @param pVM VM handle.
3322	* @param pfnAtState Pointer to callback.
3323	* @param pvUser User argument.
3324	* @thread Any.
3325	*/
3326	VMMR3DECL(int) VMR3AtStateDeregister(PVM pVM, PFNVMATSTATE pfnAtState, void *pvUser)
3327	{
3328	LogFlow(("VMR3AtStateDeregister: pfnAtState=%p pvUser=%p\n", pfnAtState, pvUser));
3329
3330	/*
3331	* Validate input.
3332	*/
3333	AssertPtrReturn(pfnAtState, VERR_INVALID_PARAMETER);
3334	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
3335
3336	PUVM pUVM = pVM->pUVM;
3337	RTCritSectEnter(&pUVM->vm.s.AtStateCritSect);
3338
3339	/*
3340	* Search the list for the entry.
3341	*/
3342	PVMATSTATE pPrev = NULL;
3343	PVMATSTATE pCur = pUVM->vm.s.pAtState;
3344	while ( pCur
3345	&& ( pCur->pfnAtState != pfnAtState
3346	\|\| pCur->pvUser != pvUser))
3347	{
3348	pPrev = pCur;
3349	pCur = pCur->pNext;
3350	}
3351	if (!pCur)
3352	{
3353	AssertMsgFailed(("pfnAtState=%p was not found\n", pfnAtState));
3354	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3355	return VERR_FILE_NOT_FOUND;
3356	}
3357
3358	/*
3359	* Unlink it.
3360	*/
3361	if (pPrev)
3362	{
3363	pPrev->pNext = pCur->pNext;
3364	if (!pCur->pNext)
3365	pUVM->vm.s.ppAtStateNext = &pPrev->pNext;
3366	}
3367	else
3368	{
3369	pUVM->vm.s.pAtState = pCur->pNext;
3370	if (!pCur->pNext)
3371	pUVM->vm.s.ppAtStateNext = &pUVM->vm.s.pAtState;
3372	}
3373
3374	RTCritSectLeave(&pUVM->vm.s.AtStateCritSect);
3375
3376	/*
3377	* Free it.
3378	*/
3379	pCur->pfnAtState = NULL;
3380	pCur->pNext = NULL;
3381	MMR3HeapFree(pCur);
3382
3383	return VINF_SUCCESS;
3384	}
3385
3386
3387	/**
3388	* Registers a VM error callback.
3389	*
3390	* @returns VBox status code.
3391	* @param pVM The VM handle.
3392	* @param pfnAtError Pointer to callback.
3393	* @param pvUser User argument.
3394	* @thread Any.
3395	*/
3396	VMMR3DECL(int) VMR3AtErrorRegister(PVM pVM, PFNVMATERROR pfnAtError, void *pvUser)
3397	{
3398	return VMR3AtErrorRegisterU(pVM->pUVM, pfnAtError, pvUser);
3399	}
3400
3401
3402	/**
3403	* Registers a VM error callback.
3404	*
3405	* @returns VBox status code.
3406	* @param pUVM The VM handle.
3407	* @param pfnAtError Pointer to callback.
3408	* @param pvUser User argument.
3409	* @thread Any.
3410	*/
3411	VMMR3DECL(int) VMR3AtErrorRegisterU(PUVM pUVM, PFNVMATERROR pfnAtError, void *pvUser)
3412	{
3413	LogFlow(("VMR3AtErrorRegister: pfnAtError=%p pvUser=%p\n", pfnAtError, pvUser));
3414
3415	/*
3416	* Validate input.
3417	*/
3418	AssertPtrReturn(pfnAtError, VERR_INVALID_PARAMETER);
3419	UVM_ASSERT_VALID_EXT_RETURN(pUVM, VERR_INVALID_VM_HANDLE);
3420
3421	/*
3422	* Allocate a new record.
3423	*/
3424	PVMATERROR pNew = (PVMATERROR)MMR3HeapAllocU(pUVM, MM_TAG_VM, sizeof(*pNew));
3425	if (!pNew)
3426	return VERR_NO_MEMORY;
3427
3428	/* fill */
3429	pNew->pfnAtError = pfnAtError;
3430	pNew->pvUser = pvUser;
3431
3432	/* insert */
3433	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3434	pNew->pNext = *pUVM->vm.s.ppAtErrorNext;
3435	*pUVM->vm.s.ppAtErrorNext = pNew;
3436	pUVM->vm.s.ppAtErrorNext = &pNew->pNext;
3437	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3438
3439	return VINF_SUCCESS;
3440	}
3441
3442
3443	/**
3444	* Deregisters a VM error callback.
3445	*
3446	* @returns VBox status code.
3447	* @param pVM The VM handle.
3448	* @param pfnAtError Pointer to callback.
3449	* @param pvUser User argument.
3450	* @thread Any.
3451	*/
3452	VMMR3DECL(int) VMR3AtErrorDeregister(PVM pVM, PFNVMATERROR pfnAtError, void *pvUser)
3453	{
3454	LogFlow(("VMR3AtErrorDeregister: pfnAtError=%p pvUser=%p\n", pfnAtError, pvUser));
3455
3456	/*
3457	* Validate input.
3458	*/
3459	AssertPtrReturn(pfnAtError, VERR_INVALID_PARAMETER);
3460	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
3461
3462	PUVM pUVM = pVM->pUVM;
3463	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3464
3465	/*
3466	* Search the list for the entry.
3467	*/
3468	PVMATERROR pPrev = NULL;
3469	PVMATERROR pCur = pUVM->vm.s.pAtError;
3470	while ( pCur
3471	&& ( pCur->pfnAtError != pfnAtError
3472	\|\| pCur->pvUser != pvUser))
3473	{
3474	pPrev = pCur;
3475	pCur = pCur->pNext;
3476	}
3477	if (!pCur)
3478	{
3479	AssertMsgFailed(("pfnAtError=%p was not found\n", pfnAtError));
3480	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3481	return VERR_FILE_NOT_FOUND;
3482	}
3483
3484	/*
3485	* Unlink it.
3486	*/
3487	if (pPrev)
3488	{
3489	pPrev->pNext = pCur->pNext;
3490	if (!pCur->pNext)
3491	pUVM->vm.s.ppAtErrorNext = &pPrev->pNext;
3492	}
3493	else
3494	{
3495	pUVM->vm.s.pAtError = pCur->pNext;
3496	if (!pCur->pNext)
3497	pUVM->vm.s.ppAtErrorNext = &pUVM->vm.s.pAtError;
3498	}
3499
3500	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3501
3502	/*
3503	* Free it.
3504	*/
3505	pCur->pfnAtError = NULL;
3506	pCur->pNext = NULL;
3507	MMR3HeapFree(pCur);
3508
3509	return VINF_SUCCESS;
3510	}
3511
3512
3513	/**
3514	* Ellipsis to va_list wrapper for calling pfnAtError.
3515	*/
3516	static void vmR3SetErrorWorkerDoCall(PVM pVM, PVMATERROR pCur, int rc, RT_SRC_POS_DECL, const char *pszFormat, ...)
3517	{
3518	va_list va;
3519	va_start(va, pszFormat);
3520	pCur->pfnAtError(pVM, pCur->pvUser, rc, RT_SRC_POS_ARGS, pszFormat, va);
3521	va_end(va);
3522	}
3523
3524
3525	/**
3526	* This is a worker function for GC and Ring-0 calls to VMSetError and VMSetErrorV.
3527	* The message is found in VMINT.
3528	*
3529	* @param pVM The VM handle.
3530	* @thread EMT.
3531	*/
3532	VMMR3DECL(void) VMR3SetErrorWorker(PVM pVM)
3533	{
3534	VM_ASSERT_EMT(pVM);
3535	AssertReleaseMsgFailed(("And we have a winner! You get to implement Ring-0 and GC VMSetErrorV! Contrats!\n"));
3536
3537	/*
3538	* Unpack the error (if we managed to format one).
3539	*/
3540	PVMERROR pErr = pVM->vm.s.pErrorR3;
3541	const char *pszFile = NULL;
3542	const char *pszFunction = NULL;
3543	uint32_t iLine = 0;
3544	const char *pszMessage;
3545	int32_t rc = VERR_MM_HYPER_NO_MEMORY;
3546	if (pErr)
3547	{
3548	AssertCompile(sizeof(const char) == sizeof(uint8_t));
3549	if (pErr->offFile)
3550	pszFile = (const char *)pErr + pErr->offFile;
3551	iLine = pErr->iLine;
3552	if (pErr->offFunction)
3553	pszFunction = (const char *)pErr + pErr->offFunction;
3554	if (pErr->offMessage)
3555	pszMessage = (const char *)pErr + pErr->offMessage;
3556	else
3557	pszMessage = "No message!";
3558	}
3559	else
3560	pszMessage = "No message! (Failed to allocate memory to put the error message in!)";
3561
3562	/*
3563	* Call the at error callbacks.
3564	*/
3565	PUVM pUVM = pVM->pUVM;
3566	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3567	ASMAtomicIncU32(&pUVM->vm.s.cRuntimeErrors);
3568	for (PVMATERROR pCur = pUVM->vm.s.pAtError; pCur; pCur = pCur->pNext)
3569	vmR3SetErrorWorkerDoCall(pVM, pCur, rc, RT_SRC_POS_ARGS, "%s", pszMessage);
3570	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3571	}
3572
3573
3574	/**
3575	* Gets the number of errors raised via VMSetError.
3576	*
3577	* This can be used avoid double error messages.
3578	*
3579	* @returns The error count.
3580	* @param pVM The VM handle.
3581	*/
3582	VMMR3DECL(uint32_t) VMR3GetErrorCount(PVM pVM)
3583	{
3584	return pVM->pUVM->vm.s.cErrors;
3585	}
3586
3587
3588	/**
3589	* Creation time wrapper for vmR3SetErrorUV.
3590	*
3591	* @returns rc.
3592	* @param pUVM Pointer to the user mode VM structure.
3593	* @param rc The VBox status code.
3594	* @param RT_SRC_POS_DECL The source position of this error.
3595	* @param pszFormat Format string.
3596	* @param ... The arguments.
3597	* @thread Any thread.
3598	*/
3599	static int vmR3SetErrorU(PUVM pUVM, int rc, RT_SRC_POS_DECL, const char *pszFormat, ...)
3600	{
3601	va_list va;
3602	va_start(va, pszFormat);
3603	vmR3SetErrorUV(pUVM, rc, pszFile, iLine, pszFunction, pszFormat, &va);
3604	va_end(va);
3605	return rc;
3606	}
3607
3608
3609	/**
3610	* Worker which calls everyone listening to the VM error messages.
3611	*
3612	* @param pUVM Pointer to the user mode VM structure.
3613	* @param rc The VBox status code.
3614	* @param RT_SRC_POS_DECL The source position of this error.
3615	* @param pszFormat Format string.
3616	* @param pArgs Pointer to the format arguments.
3617	* @thread EMT
3618	*/
3619	DECLCALLBACK(void) vmR3SetErrorUV(PUVM pUVM, int rc, RT_SRC_POS_DECL, const char pszFormat, va_list pArgs)
3620	{
3621	/*
3622	* Log the error.
3623	*/
3624	va_list va3;
3625	va_copy(va3, *pArgs);
3626	RTLogRelPrintf("VMSetError: %s(%d) %s; rc=%Rrc\n"
3627	"VMSetError: %N\n",
3628	pszFile, iLine, pszFunction, rc,
3629	pszFormat, &va3);
3630	va_end(va3);
3631
3632	#ifdef LOG_ENABLED
3633	va_copy(va3, *pArgs);
3634	RTLogPrintf("VMSetError: %s(%d) %s; rc=%Rrc\n"
3635	"%N\n",
3636	pszFile, iLine, pszFunction, rc,
3637	pszFormat, &va3);
3638	va_end(va3);
3639	#endif
3640
3641	/*
3642	* Make a copy of the message.
3643	*/
3644	if (pUVM->pVM)
3645	vmSetErrorCopy(pUVM->pVM, rc, RT_SRC_POS_ARGS, pszFormat, *pArgs);
3646
3647	/*
3648	* Call the at error callbacks.
3649	*/
3650	bool fCalledSomeone = false;
3651	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3652	ASMAtomicIncU32(&pUVM->vm.s.cErrors);
3653	for (PVMATERROR pCur = pUVM->vm.s.pAtError; pCur; pCur = pCur->pNext)
3654	{
3655	va_list va2;
3656	va_copy(va2, *pArgs);
3657	pCur->pfnAtError(pUVM->pVM, pCur->pvUser, rc, RT_SRC_POS_ARGS, pszFormat, va2);
3658	va_end(va2);
3659	fCalledSomeone = true;
3660	}
3661	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3662	}
3663
3664
3665	/**
3666	* Registers a VM runtime error callback.
3667	*
3668	* @returns VBox status code.
3669	* @param pVM The VM handle.
3670	* @param pfnAtRuntimeError Pointer to callback.
3671	* @param pvUser User argument.
3672	* @thread Any.
3673	*/
3674	VMMR3DECL(int) VMR3AtRuntimeErrorRegister(PVM pVM, PFNVMATRUNTIMEERROR pfnAtRuntimeError, void *pvUser)
3675	{
3676	LogFlow(("VMR3AtRuntimeErrorRegister: pfnAtRuntimeError=%p pvUser=%p\n", pfnAtRuntimeError, pvUser));
3677
3678	/*
3679	* Validate input.
3680	*/
3681	AssertPtrReturn(pfnAtRuntimeError, VERR_INVALID_PARAMETER);
3682	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
3683
3684	/*
3685	* Allocate a new record.
3686	*/
3687	PUVM pUVM = pVM->pUVM;
3688	PVMATRUNTIMEERROR pNew = (PVMATRUNTIMEERROR)MMR3HeapAllocU(pUVM, MM_TAG_VM, sizeof(*pNew));
3689	if (!pNew)
3690	return VERR_NO_MEMORY;
3691
3692	/* fill */
3693	pNew->pfnAtRuntimeError = pfnAtRuntimeError;
3694	pNew->pvUser = pvUser;
3695
3696	/* insert */
3697	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3698	pNew->pNext = *pUVM->vm.s.ppAtRuntimeErrorNext;
3699	*pUVM->vm.s.ppAtRuntimeErrorNext = pNew;
3700	pUVM->vm.s.ppAtRuntimeErrorNext = &pNew->pNext;
3701	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3702
3703	return VINF_SUCCESS;
3704	}
3705
3706
3707	/**
3708	* Deregisters a VM runtime error callback.
3709	*
3710	* @returns VBox status code.
3711	* @param pVM The VM handle.
3712	* @param pfnAtRuntimeError Pointer to callback.
3713	* @param pvUser User argument.
3714	* @thread Any.
3715	*/
3716	VMMR3DECL(int) VMR3AtRuntimeErrorDeregister(PVM pVM, PFNVMATRUNTIMEERROR pfnAtRuntimeError, void *pvUser)
3717	{
3718	LogFlow(("VMR3AtRuntimeErrorDeregister: pfnAtRuntimeError=%p pvUser=%p\n", pfnAtRuntimeError, pvUser));
3719
3720	/*
3721	* Validate input.
3722	*/
3723	AssertPtrReturn(pfnAtRuntimeError, VERR_INVALID_PARAMETER);
3724	VM_ASSERT_VALID_EXT_RETURN(pVM, VERR_INVALID_VM_HANDLE);
3725
3726	PUVM pUVM = pVM->pUVM;
3727	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3728
3729	/*
3730	* Search the list for the entry.
3731	*/
3732	PVMATRUNTIMEERROR pPrev = NULL;
3733	PVMATRUNTIMEERROR pCur = pUVM->vm.s.pAtRuntimeError;
3734	while ( pCur
3735	&& ( pCur->pfnAtRuntimeError != pfnAtRuntimeError
3736	\|\| pCur->pvUser != pvUser))
3737	{
3738	pPrev = pCur;
3739	pCur = pCur->pNext;
3740	}
3741	if (!pCur)
3742	{
3743	AssertMsgFailed(("pfnAtRuntimeError=%p was not found\n", pfnAtRuntimeError));
3744	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3745	return VERR_FILE_NOT_FOUND;
3746	}
3747
3748	/*
3749	* Unlink it.
3750	*/
3751	if (pPrev)
3752	{
3753	pPrev->pNext = pCur->pNext;
3754	if (!pCur->pNext)
3755	pUVM->vm.s.ppAtRuntimeErrorNext = &pPrev->pNext;
3756	}
3757	else
3758	{
3759	pUVM->vm.s.pAtRuntimeError = pCur->pNext;
3760	if (!pCur->pNext)
3761	pUVM->vm.s.ppAtRuntimeErrorNext = &pUVM->vm.s.pAtRuntimeError;
3762	}
3763
3764	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3765
3766	/*
3767	* Free it.
3768	*/
3769	pCur->pfnAtRuntimeError = NULL;
3770	pCur->pNext = NULL;
3771	MMR3HeapFree(pCur);
3772
3773	return VINF_SUCCESS;
3774	}
3775
3776
3777	/**
3778	* EMT rendezvous worker that vmR3SetRuntimeErrorCommon uses to safely change
3779	* the state to FatalError(LS).
3780	*
3781	* @returns VERR_VM_INVALID_VM_STATE or VINF_SUCCESS. (This is a strict return
3782	* code, see FNVMMEMTRENDEZVOUS.)
3783	*
3784	* @param pVM The VM handle.
3785	* @param pVCpu The VMCPU handle of the EMT.
3786	* @param pvUser Ignored.
3787	*/
3788	static DECLCALLBACK(VBOXSTRICTRC) vmR3SetRuntimeErrorChangeState(PVM pVM, PVMCPU pVCpu, void *pvUser)
3789	{
3790	NOREF(pVCpu);
3791	Assert(!pvUser); NOREF(pvUser);
3792
3793	int rc = vmR3TrySetState(pVM, "VMSetRuntimeError", 2,
3794	VMSTATE_FATAL_ERROR, VMSTATE_RUNNING,
3795	VMSTATE_FATAL_ERROR_LS, VMSTATE_RUNNING_LS);
3796	if (rc == 2)
3797	SSMR3Cancel(pVM);
3798	return RT_SUCCESS(rc) ? VINF_SUCCESS : rc;
3799	}
3800
3801
3802	/**
3803	* Worker for VMR3SetRuntimeErrorWorker and vmR3SetRuntimeErrorV.
3804	*
3805	* This does the common parts after the error has been saved / retrieved.
3806	*
3807	* @returns VBox status code with modifications, see VMSetRuntimeErrorV.
3808	*
3809	* @param pVM The VM handle.
3810	* @param fFlags The error flags.
3811	* @param pszErrorId Error ID string.
3812	* @param pszFormat Format string.
3813	* @param pVa Pointer to the format arguments.
3814	*/
3815	static int vmR3SetRuntimeErrorCommon(PVM pVM, uint32_t fFlags, const char pszErrorId, const char pszFormat, va_list *pVa)
3816	{
3817	LogRel(("VM: Raising runtime error '%s' (fFlags=%#x)\n", pszErrorId, fFlags));
3818
3819	/*
3820	* Take actions before the call.
3821	*/
3822	int rc;
3823	if (fFlags & VMSETRTERR_FLAGS_FATAL)
3824	rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, vmR3SetRuntimeErrorChangeState, NULL);
3825	else if (fFlags & VMSETRTERR_FLAGS_SUSPEND)
3826	rc = VMR3Suspend(pVM);
3827	else
3828	rc = VINF_SUCCESS;
3829
3830	/*
3831	* Do the callback round.
3832	*/
3833	PUVM pUVM = pVM->pUVM;
3834	RTCritSectEnter(&pUVM->vm.s.AtErrorCritSect);
3835	ASMAtomicIncU32(&pUVM->vm.s.cRuntimeErrors);
3836	for (PVMATRUNTIMEERROR pCur = pUVM->vm.s.pAtRuntimeError; pCur; pCur = pCur->pNext)
3837	{
3838	va_list va;
3839	va_copy(va, *pVa);
3840	pCur->pfnAtRuntimeError(pVM, pCur->pvUser, fFlags, pszErrorId, pszFormat, va);
3841	va_end(va);
3842	}
3843	RTCritSectLeave(&pUVM->vm.s.AtErrorCritSect);
3844
3845	return rc;
3846	}
3847
3848
3849	/**
3850	* Ellipsis to va_list wrapper for calling vmR3SetRuntimeErrorCommon.
3851	*/
3852	static int vmR3SetRuntimeErrorCommonF(PVM pVM, uint32_t fFlags, const char pszErrorId, const char pszFormat, ...)
3853	{
3854	va_list va;
3855	va_start(va, pszFormat);
3856	int rc = vmR3SetRuntimeErrorCommon(pVM, fFlags, pszErrorId, pszFormat, &va);
3857	va_end(va);
3858	return rc;
3859	}
3860
3861
3862	/**
3863	* This is a worker function for RC and Ring-0 calls to VMSetError and
3864	* VMSetErrorV.
3865	*
3866	* The message is found in VMINT.
3867	*
3868	* @returns VBox status code, see VMSetRuntimeError.
3869	* @param pVM The VM handle.
3870	* @thread EMT.
3871	*/
3872	VMMR3DECL(int) VMR3SetRuntimeErrorWorker(PVM pVM)
3873	{
3874	VM_ASSERT_EMT(pVM);
3875	AssertReleaseMsgFailed(("And we have a winner! You get to implement Ring-0 and GC VMSetRuntimeErrorV! Congrats!\n"));
3876
3877	/*
3878	* Unpack the error (if we managed to format one).
3879	*/
3880	const char *pszErrorId = "SetRuntimeError";
3881	const char *pszMessage = "No message!";
3882	uint32_t fFlags = VMSETRTERR_FLAGS_FATAL;
3883	PVMRUNTIMEERROR pErr = pVM->vm.s.pRuntimeErrorR3;
3884	if (pErr)
3885	{
3886	AssertCompile(sizeof(const char) == sizeof(uint8_t));
3887	if (pErr->offErrorId)
3888	pszErrorId = (const char *)pErr + pErr->offErrorId;
3889	if (pErr->offMessage)
3890	pszMessage = (const char *)pErr + pErr->offMessage;
3891	fFlags = pErr->fFlags;
3892	}
3893
3894	/*
3895	* Join cause with vmR3SetRuntimeErrorV.
3896	*/
3897	return vmR3SetRuntimeErrorCommonF(pVM, fFlags, pszErrorId, "%s", pszMessage);
3898	}
3899
3900
3901	/**
3902	* Worker for VMSetRuntimeErrorV for doing the job on EMT in ring-3.
3903	*
3904	* @returns VBox status code with modifications, see VMSetRuntimeErrorV.
3905	*
3906	* @param pVM The VM handle.
3907	* @param fFlags The error flags.
3908	* @param pszErrorId Error ID string.
3909	* @param pszMessage The error message residing the MM heap.
3910	*
3911	* @thread EMT
3912	*/
3913	DECLCALLBACK(int) vmR3SetRuntimeError(PVM pVM, uint32_t fFlags, const char pszErrorId, char pszMessage)
3914	{
3915	#if 0 /** @todo make copy of the error msg. */
3916	/*
3917	* Make a copy of the message.
3918	*/
3919	va_list va2;
3920	va_copy(va2, *pVa);
3921	vmSetRuntimeErrorCopy(pVM, fFlags, pszErrorId, pszFormat, va2);
3922	va_end(va2);
3923	#endif
3924
3925	/*
3926	* Join paths with VMR3SetRuntimeErrorWorker.
3927	*/
3928	int rc = vmR3SetRuntimeErrorCommonF(pVM, fFlags, pszErrorId, "%s", pszMessage);
3929	MMR3HeapFree(pszMessage);
3930	return rc;
3931	}
3932
3933
3934	/**
3935	* Worker for VMSetRuntimeErrorV for doing the job on EMT in ring-3.
3936	*
3937	* @returns VBox status code with modifications, see VMSetRuntimeErrorV.
3938	*
3939	* @param pVM The VM handle.
3940	* @param fFlags The error flags.
3941	* @param pszErrorId Error ID string.
3942	* @param pszFormat Format string.
3943	* @param pVa Pointer to the format arguments.
3944	*
3945	* @thread EMT
3946	*/
3947	DECLCALLBACK(int) vmR3SetRuntimeErrorV(PVM pVM, uint32_t fFlags, const char pszErrorId, const char pszFormat, va_list *pVa)
3948	{
3949	/*
3950	* Make a copy of the message.
3951	*/
3952	va_list va2;
3953	va_copy(va2, *pVa);
3954	vmSetRuntimeErrorCopy(pVM, fFlags, pszErrorId, pszFormat, va2);
3955	va_end(va2);
3956
3957	/*
3958	* Join paths with VMR3SetRuntimeErrorWorker.
3959	*/
3960	return vmR3SetRuntimeErrorCommon(pVM, fFlags, pszErrorId, pszFormat, pVa);
3961	}
3962
3963
3964	/**
3965	* Gets the number of runtime errors raised via VMR3SetRuntimeError.
3966	*
3967	* This can be used avoid double error messages.
3968	*
3969	* @returns The runtime error count.
3970	* @param pVM The VM handle.
3971	*/
3972	VMMR3DECL(uint32_t) VMR3GetRuntimeErrorCount(PVM pVM)
3973	{
3974	return pVM->pUVM->vm.s.cRuntimeErrors;
3975	}
3976
3977
3978	/**
3979	* Gets the ID virtual of the virtual CPU assoicated with the calling thread.
3980	*
3981	* @returns The CPU ID. NIL_VMCPUID if the thread isn't an EMT.
3982	*
3983	* @param pVM The VM handle.
3984	*/
3985	VMMR3DECL(RTCPUID) VMR3GetVMCPUId(PVM pVM)
3986	{
3987	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pVM->pUVM->vm.s.idxTLS);
3988	return pUVCpu
3989	? pUVCpu->idCpu
3990	: NIL_VMCPUID;
3991	}
3992
3993
3994	/**
3995	* Returns the native handle of the current EMT VMCPU thread.
3996	*
3997	* @returns Handle if this is an EMT thread; NIL_RTNATIVETHREAD otherwise
3998	* @param pVM The VM handle.
3999	* @thread EMT
4000	*/
4001	VMMR3DECL(RTNATIVETHREAD) VMR3GetVMCPUNativeThread(PVM pVM)
4002	{
4003	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pVM->pUVM->vm.s.idxTLS);
4004
4005	if (!pUVCpu)
4006	return NIL_RTNATIVETHREAD;
4007
4008	return pUVCpu->vm.s.NativeThreadEMT;
4009	}
4010
4011
4012	/**
4013	* Returns the native handle of the current EMT VMCPU thread.
4014	*
4015	* @returns Handle if this is an EMT thread; NIL_RTNATIVETHREAD otherwise
4016	* @param pVM The VM handle.
4017	* @thread EMT
4018	*/
4019	VMMR3DECL(RTNATIVETHREAD) VMR3GetVMCPUNativeThreadU(PUVM pUVM)
4020	{
4021	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pUVM->vm.s.idxTLS);
4022
4023	if (!pUVCpu)
4024	return NIL_RTNATIVETHREAD;
4025
4026	return pUVCpu->vm.s.NativeThreadEMT;
4027	}
4028
4029
4030	/**
4031	* Returns the handle of the current EMT VMCPU thread.
4032	*
4033	* @returns Handle if this is an EMT thread; NIL_RTNATIVETHREAD otherwise
4034	* @param pVM The VM handle.
4035	* @thread EMT
4036	*/
4037	VMMR3DECL(RTTHREAD) VMR3GetVMCPUThread(PVM pVM)
4038	{
4039	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pVM->pUVM->vm.s.idxTLS);
4040
4041	if (!pUVCpu)
4042	return NIL_RTTHREAD;
4043
4044	return pUVCpu->vm.s.ThreadEMT;
4045	}
4046
4047
4048	/**
4049	* Returns the handle of the current EMT VMCPU thread.
4050	*
4051	* @returns Handle if this is an EMT thread; NIL_RTNATIVETHREAD otherwise
4052	* @param pVM The VM handle.
4053	* @thread EMT
4054	*/
4055	VMMR3DECL(RTTHREAD) VMR3GetVMCPUThreadU(PUVM pUVM)
4056	{
4057	PUVMCPU pUVCpu = (PUVMCPU)RTTlsGet(pUVM->vm.s.idxTLS);
4058
4059	if (!pUVCpu)
4060	return NIL_RTTHREAD;
4061
4062	return pUVCpu->vm.s.ThreadEMT;
4063	}
4064
4065
4066	/**
4067	* Return the package and core id of a CPU.
4068	*
4069	* @returns VBOX status code.
4070	* @param pVM The VM to operate on.
4071	* @param idCpu Virtual CPU to get the ID from.
4072	* @param pidCpuCore Where to store the core ID of the virtual CPU.
4073	* @param pidCpuPackage Where to store the package ID of the virtual CPU.
4074	*
4075	*/
4076	VMMR3DECL(int) VMR3GetCpuCoreAndPackageIdFromCpuId(PVM pVM, VMCPUID idCpu, uint32_t pidCpuCore, uint32_t pidCpuPackage)
4077	{
4078	if (idCpu >= pVM->cCpus)
4079	return VERR_INVALID_CPU_ID;
4080
4081	#ifdef VBOX_WITH_MULTI_CORE
4082	*pidCpuCore = idCpu;
4083	*pidCpuPackage = 0;
4084	#else
4085	*pidCpuCore = 0;
4086	*pidCpuPackage = idCpu;
4087	#endif
4088
4089	return VINF_SUCCESS;
4090	}
4091
4092
4093	/**
4094	* Worker for VMR3HotUnplugCpu.
4095	*
4096	* @returns VINF_EM_WAIT_SPIP (strict status code).
4097	* @param pVM The VM handle.
4098	* @param idCpu The current CPU.
4099	*/
4100	static DECLCALLBACK(int) vmR3HotUnplugCpu(PVM pVM, VMCPUID idCpu)
4101	{
4102	PVMCPU pVCpu = VMMGetCpuById(pVM, idCpu);
4103	VMCPU_ASSERT_EMT(pVCpu);
4104
4105	/*
4106	* Reset per CPU resources.
4107	*
4108	* Actually only needed for VT-x because the CPU seems to be still in some
4109	* paged mode and startup fails after a new hot plug event. SVM works fine
4110	* even without this.
4111	*/
4112	Log(("vmR3HotUnplugCpu for VCPU %u\n", idCpu));
4113	PGMR3ResetUnpluggedCpu(pVM, pVCpu);
4114	PDMR3ResetCpu(pVCpu);
4115	TRPMR3ResetCpu(pVCpu);
4116	CPUMR3ResetCpu(pVCpu);
4117	EMR3ResetCpu(pVCpu);
4118	HWACCMR3ResetCpu(pVCpu);
4119	return VINF_EM_WAIT_SIPI;
4120	}
4121
4122
4123	/**
4124	* Hot-unplugs a CPU from the guest.
4125	*
4126	* @returns VBox status code.
4127	* @param pVM The VM to operate on.
4128	* @param idCpu Virtual CPU to perform the hot unplugging operation on.
4129	*/
4130	VMMR3DECL(int) VMR3HotUnplugCpu(PVM pVM, VMCPUID idCpu)
4131	{
4132	AssertReturn(idCpu < pVM->cCpus, VERR_INVALID_CPU_ID);
4133
4134	/** @todo r=bird: Don't destroy the EMT, it'll break VMMR3EmtRendezvous and
4135	* broadcast requests. Just note down somewhere that the CPU is
4136	* offline and send it to SPIP wait. Maybe modify VMCPUSTATE and push
4137	* it out of the EM loops when offline. */
4138	return VMR3ReqCallNoWaitU(pVM->pUVM, idCpu, (PFNRT)vmR3HotUnplugCpu, 2, pVM, idCpu);
4139	}
4140
4141
4142	/**
4143	* Hot-plugs a CPU on the guest.
4144	*
4145	* @returns VBox status code.
4146	* @param pVM The VM to operate on.
4147	* @param idCpu Virtual CPU to perform the hot plugging operation on.
4148	*/
4149	VMMR3DECL(int) VMR3HotPlugCpu(PVM pVM, VMCPUID idCpu)
4150	{
4151	AssertReturn(idCpu < pVM->cCpus, VERR_INVALID_CPU_ID);
4152
4153	/** @todo r-bird: Just mark it online and make sure it waits on SPIP. */
4154	return VINF_SUCCESS;
4155	}
4156

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

source: vbox/trunk/src/VBox/VMM/VM.cpp@ 28688

Download in other formats: