PGMR0.cpp@ 71963

Last change on this file since 71963 was 69111, checked in by vboxsync, 7 years ago
(C) year
Property svn:eol-style set to `native` Property svn:keywords set to `Id Revision`
File size: 24.5 KB

Line
1	/* $Id: PGMR0.cpp 69111 2017-10-17 14:26:02Z vboxsync $ */
2	/** @file
3	* PGM - Page Manager and Monitor, Ring-0.
4	*/
5
6	/*
7	* Copyright (C) 2007-2017 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18
19	/*********************************************************************************************************************************
20	* Header Files *
21	*********************************************************************************************************************************/
22	#define LOG_GROUP LOG_GROUP_PGM
23	#include <VBox/rawpci.h>
24	#include <VBox/vmm/pgm.h>
25	#include <VBox/vmm/gmm.h>
26	#include <VBox/vmm/gvm.h>
27	#include "PGMInternal.h"
28	#include <VBox/vmm/vm.h>
29	#include "PGMInline.h"
30	#include <VBox/log.h>
31	#include <VBox/err.h>
32	#include <iprt/assert.h>
33	#include <iprt/mem.h>
34
35
36	/*
37	* Instantiate the ring-0 header/code templates.
38	*/
39	#define PGM_BTH_NAME(name) PGM_BTH_NAME_32BIT_PROT(name)
40	#include "PGMR0Bth.h"
41	#undef PGM_BTH_NAME
42
43	#define PGM_BTH_NAME(name) PGM_BTH_NAME_PAE_PROT(name)
44	#include "PGMR0Bth.h"
45	#undef PGM_BTH_NAME
46
47	#define PGM_BTH_NAME(name) PGM_BTH_NAME_AMD64_PROT(name)
48	#include "PGMR0Bth.h"
49	#undef PGM_BTH_NAME
50
51	#define PGM_BTH_NAME(name) PGM_BTH_NAME_EPT_PROT(name)
52	#include "PGMR0Bth.h"
53	#undef PGM_BTH_NAME
54
55
56	/**
57	* Worker function for PGMR3PhysAllocateHandyPages and pgmPhysEnsureHandyPage.
58	*
59	* @returns The following VBox status codes.
60	* @retval VINF_SUCCESS on success. FF cleared.
61	* @retval VINF_EM_NO_MEMORY if we're out of memory. The FF is set in this case.
62	*
63	* @param pGVM The global (ring-0) VM structure.
64	* @param pVM The cross context VM structure.
65	* @param idCpu The ID of the calling EMT.
66	*
67	* @thread EMT(idCpu)
68	*
69	* @remarks Must be called from within the PGM critical section. The caller
70	* must clear the new pages.
71	*/
72	VMMR0_INT_DECL(int) PGMR0PhysAllocateHandyPages(PGVM pGVM, PVM pVM, VMCPUID idCpu)
73	{
74	/*
75	* Validate inputs.
76	*/
77	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID); /* caller already checked this, but just to be sure. */
78	AssertReturn(pGVM->aCpus[idCpu].hEMT == RTThreadNativeSelf(), VERR_NOT_OWNER);
79	PGM_LOCK_ASSERT_OWNER_EX(pVM, &pVM->aCpus[idCpu]);
80
81	/*
82	* Check for error injection.
83	*/
84	if (RT_UNLIKELY(pVM->pgm.s.fErrInjHandyPages))
85	return VERR_NO_MEMORY;
86
87	/*
88	* Try allocate a full set of handy pages.
89	*/
90	uint32_t iFirst = pVM->pgm.s.cHandyPages;
91	AssertReturn(iFirst <= RT_ELEMENTS(pVM->pgm.s.aHandyPages), VERR_PGM_HANDY_PAGE_IPE);
92	uint32_t cPages = RT_ELEMENTS(pVM->pgm.s.aHandyPages) - iFirst;
93	if (!cPages)
94	return VINF_SUCCESS;
95	int rc = GMMR0AllocateHandyPages(pGVM, pVM, idCpu, cPages, cPages, &pVM->pgm.s.aHandyPages[iFirst]);
96	if (RT_SUCCESS(rc))
97	{
98	#ifdef VBOX_STRICT
99	for (uint32_t i = 0; i < RT_ELEMENTS(pVM->pgm.s.aHandyPages); i++)
100	{
101	Assert(pVM->pgm.s.aHandyPages[i].idPage != NIL_GMM_PAGEID);
102	Assert(pVM->pgm.s.aHandyPages[i].idPage <= GMM_PAGEID_LAST);
103	Assert(pVM->pgm.s.aHandyPages[i].idSharedPage == NIL_GMM_PAGEID);
104	Assert(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys != NIL_RTHCPHYS);
105	Assert(!(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys & ~X86_PTE_PAE_PG_MASK));
106	}
107	#endif
108
109	pVM->pgm.s.cHandyPages = RT_ELEMENTS(pVM->pgm.s.aHandyPages);
110	}
111	else if (rc != VERR_GMM_SEED_ME)
112	{
113	if ( ( rc == VERR_GMM_HIT_GLOBAL_LIMIT
114	\|\| rc == VERR_GMM_HIT_VM_ACCOUNT_LIMIT)
115	&& iFirst < PGM_HANDY_PAGES_MIN)
116	{
117
118	#ifdef VBOX_STRICT
119	/* We're ASSUMING that GMM has updated all the entires before failing us. */
120	uint32_t i;
121	for (i = iFirst; i < RT_ELEMENTS(pVM->pgm.s.aHandyPages); i++)
122	{
123	Assert(pVM->pgm.s.aHandyPages[i].idPage == NIL_GMM_PAGEID);
124	Assert(pVM->pgm.s.aHandyPages[i].idSharedPage == NIL_GMM_PAGEID);
125	Assert(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys == NIL_RTHCPHYS);
126	}
127	#endif
128
129	/*
130	* Reduce the number of pages until we hit the minimum limit.
131	*/
132	do
133	{
134	cPages >>= 1;
135	if (cPages + iFirst < PGM_HANDY_PAGES_MIN)
136	cPages = PGM_HANDY_PAGES_MIN - iFirst;
137	rc = GMMR0AllocateHandyPages(pGVM, pVM, idCpu, 0, cPages, &pVM->pgm.s.aHandyPages[iFirst]);
138	} while ( ( rc == VERR_GMM_HIT_GLOBAL_LIMIT
139	\|\| rc == VERR_GMM_HIT_VM_ACCOUNT_LIMIT)
140	&& cPages + iFirst > PGM_HANDY_PAGES_MIN);
141	if (RT_SUCCESS(rc))
142	{
143	#ifdef VBOX_STRICT
144	i = iFirst + cPages;
145	while (i-- > 0)
146	{
147	Assert(pVM->pgm.s.aHandyPages[i].idPage != NIL_GMM_PAGEID);
148	Assert(pVM->pgm.s.aHandyPages[i].idPage <= GMM_PAGEID_LAST);
149	Assert(pVM->pgm.s.aHandyPages[i].idSharedPage == NIL_GMM_PAGEID);
150	Assert(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys != NIL_RTHCPHYS);
151	Assert(!(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys & ~X86_PTE_PAE_PG_MASK));
152	}
153
154	for (i = cPages + iFirst; i < RT_ELEMENTS(pVM->pgm.s.aHandyPages); i++)
155	{
156	Assert(pVM->pgm.s.aHandyPages[i].idPage == NIL_GMM_PAGEID);
157	Assert(pVM->pgm.s.aHandyPages[i].idSharedPage == NIL_GMM_PAGEID);
158	Assert(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys == NIL_RTHCPHYS);
159	}
160	#endif
161
162	pVM->pgm.s.cHandyPages = iFirst + cPages;
163	}
164	}
165
166	if (RT_FAILURE(rc) && rc != VERR_GMM_SEED_ME)
167	{
168	LogRel(("PGMR0PhysAllocateHandyPages: rc=%Rrc iFirst=%d cPages=%d\n", rc, iFirst, cPages));
169	VM_FF_SET(pVM, VM_FF_PGM_NO_MEMORY);
170	}
171	}
172
173
174	LogFlow(("PGMR0PhysAllocateHandyPages: cPages=%d rc=%Rrc\n", cPages, rc));
175	return rc;
176	}
177
178
179	/**
180	* Flushes any changes pending in the handy page array.
181	*
182	* It is very important that this gets done when page sharing is enabled.
183	*
184	* @returns The following VBox status codes.
185	* @retval VINF_SUCCESS on success. FF cleared.
186	*
187	* @param pGVM The global (ring-0) VM structure.
188	* @param pVM The cross context VM structure.
189	* @param idCpu The ID of the calling EMT.
190	*
191	* @thread EMT(idCpu)
192	*
193	* @remarks Must be called from within the PGM critical section.
194	*/
195	VMMR0_INT_DECL(int) PGMR0PhysFlushHandyPages(PGVM pGVM, PVM pVM, VMCPUID idCpu)
196	{
197	/*
198	* Validate inputs.
199	*/
200	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID); /* caller already checked this, but just to be sure. */
201	AssertReturn(pGVM->aCpus[idCpu].hEMT == RTThreadNativeSelf(), VERR_NOT_OWNER);
202	PGM_LOCK_ASSERT_OWNER_EX(pVM, &pVM->aCpus[idCpu]);
203
204	/*
205	* Try allocate a full set of handy pages.
206	*/
207	uint32_t iFirst = pVM->pgm.s.cHandyPages;
208	AssertReturn(iFirst <= RT_ELEMENTS(pVM->pgm.s.aHandyPages), VERR_PGM_HANDY_PAGE_IPE);
209	uint32_t cPages = RT_ELEMENTS(pVM->pgm.s.aHandyPages) - iFirst;
210	if (!cPages)
211	return VINF_SUCCESS;
212	int rc = GMMR0AllocateHandyPages(pGVM, pVM, idCpu, cPages, 0, &pVM->pgm.s.aHandyPages[iFirst]);
213
214	LogFlow(("PGMR0PhysFlushHandyPages: cPages=%d rc=%Rrc\n", cPages, rc));
215	return rc;
216	}
217
218
219	/**
220	* Worker function for PGMR3PhysAllocateLargeHandyPage
221	*
222	* @returns The following VBox status codes.
223	* @retval VINF_SUCCESS on success.
224	* @retval VINF_EM_NO_MEMORY if we're out of memory.
225	*
226	* @param pGVM The global (ring-0) VM structure.
227	* @param pVM The cross context VM structure.
228	* @param idCpu The ID of the calling EMT.
229	*
230	* @thread EMT(idCpu)
231	*
232	* @remarks Must be called from within the PGM critical section. The caller
233	* must clear the new pages.
234	*/
235	VMMR0_INT_DECL(int) PGMR0PhysAllocateLargeHandyPage(PGVM pGVM, PVM pVM, VMCPUID idCpu)
236	{
237	/*
238	* Validate inputs.
239	*/
240	AssertReturn(idCpu < pGVM->cCpus, VERR_INVALID_CPU_ID); /* caller already checked this, but just to be sure. */
241	AssertReturn(pGVM->aCpus[idCpu].hEMT == RTThreadNativeSelf(), VERR_NOT_OWNER);
242	PGM_LOCK_ASSERT_OWNER_EX(pVM, &pVM->aCpus[idCpu]);
243	Assert(!pVM->pgm.s.cLargeHandyPages);
244
245	/*
246	* Do the job.
247	*/
248	int rc = GMMR0AllocateLargePage(pGVM, pVM, idCpu, _2M,
249	&pVM->pgm.s.aLargeHandyPage[0].idPage,
250	&pVM->pgm.s.aLargeHandyPage[0].HCPhysGCPhys);
251	if (RT_SUCCESS(rc))
252	pVM->pgm.s.cLargeHandyPages = 1;
253
254	return rc;
255	}
256
257
258	#ifdef VBOX_WITH_PCI_PASSTHROUGH
259	/* Interface sketch. The interface belongs to a global PCI pass-through
260	manager. It shall use the global VM handle, not the user VM handle to
261	store the per-VM info (domain) since that is all ring-0 stuff, thus
262	passing pGVM here. I've tentitively prefixed the functions 'GPciRawR0',
263	we can discuss the PciRaw code re-organtization when I'm back from
264	vacation.
265
266	I've implemented the initial IOMMU set up below. For things to work
267	reliably, we will probably need add a whole bunch of checks and
268	GPciRawR0GuestPageUpdate call to the PGM code. For the present,
269	assuming nested paging (enforced) and prealloc (enforced), no
270	ballooning (check missing), page sharing (check missing) or live
271	migration (check missing), it might work fine. At least if some
272	VM power-off hook is present and can tear down the IOMMU page tables. */
273
274	/**
275	* Tells the global PCI pass-through manager that we are about to set up the
276	* guest page to host page mappings for the specfied VM.
277	*
278	* @returns VBox status code.
279	*
280	* @param pGVM The ring-0 VM structure.
281	*/
282	VMMR0_INT_DECL(int) GPciRawR0GuestPageBeginAssignments(PGVM pGVM)
283	{
284	NOREF(pGVM);
285	return VINF_SUCCESS;
286	}
287
288
289	/**
290	* Assigns a host page mapping for a guest page.
291	*
292	* This is only used when setting up the mappings, i.e. between
293	* GPciRawR0GuestPageBeginAssignments and GPciRawR0GuestPageEndAssignments.
294	*
295	* @returns VBox status code.
296	* @param pGVM The ring-0 VM structure.
297	* @param GCPhys The address of the guest page (page aligned).
298	* @param HCPhys The address of the host page (page aligned).
299	*/
300	VMMR0_INT_DECL(int) GPciRawR0GuestPageAssign(PGVM pGVM, RTGCPHYS GCPhys, RTHCPHYS HCPhys)
301	{
302	AssertReturn(!(GCPhys & PAGE_OFFSET_MASK), VERR_INTERNAL_ERROR_3);
303	AssertReturn(!(HCPhys & PAGE_OFFSET_MASK), VERR_INTERNAL_ERROR_3);
304
305	if (pGVM->rawpci.s.pfnContigMemInfo)
306	/** @todo what do we do on failure? */
307	pGVM->rawpci.s.pfnContigMemInfo(&pGVM->rawpci.s, HCPhys, GCPhys, PAGE_SIZE, PCIRAW_MEMINFO_MAP);
308
309	return VINF_SUCCESS;
310	}
311
312
313	/**
314	* Indicates that the specified guest page doesn't exists but doesn't have host
315	* page mapping we trust PCI pass-through with.
316	*
317	* This is only used when setting up the mappings, i.e. between
318	* GPciRawR0GuestPageBeginAssignments and GPciRawR0GuestPageEndAssignments.
319	*
320	* @returns VBox status code.
321	* @param pGVM The ring-0 VM structure.
322	* @param GCPhys The address of the guest page (page aligned).
323	* @param HCPhys The address of the host page (page aligned).
324	*/
325	VMMR0_INT_DECL(int) GPciRawR0GuestPageUnassign(PGVM pGVM, RTGCPHYS GCPhys)
326	{
327	AssertReturn(!(GCPhys & PAGE_OFFSET_MASK), VERR_INTERNAL_ERROR_3);
328
329	if (pGVM->rawpci.s.pfnContigMemInfo)
330	/** @todo what do we do on failure? */
331	pGVM->rawpci.s.pfnContigMemInfo(&pGVM->rawpci.s, 0, GCPhys, PAGE_SIZE, PCIRAW_MEMINFO_UNMAP);
332
333	return VINF_SUCCESS;
334	}
335
336
337	/**
338	* Tells the global PCI pass-through manager that we have completed setting up
339	* the guest page to host page mappings for the specfied VM.
340	*
341	* This complements GPciRawR0GuestPageBeginAssignments and will be called even
342	* if some page assignment failed.
343	*
344	* @returns VBox status code.
345	*
346	* @param pGVM The ring-0 VM structure.
347	*/
348	VMMR0_INT_DECL(int) GPciRawR0GuestPageEndAssignments(PGVM pGVM)
349	{
350	NOREF(pGVM);
351	return VINF_SUCCESS;
352	}
353
354
355	/**
356	* Tells the global PCI pass-through manager that a guest page mapping has
357	* changed after the initial setup.
358	*
359	* @returns VBox status code.
360	* @param pGVM The ring-0 VM structure.
361	* @param GCPhys The address of the guest page (page aligned).
362	* @param HCPhys The new host page address or NIL_RTHCPHYS if
363	* now unassigned.
364	*/
365	VMMR0_INT_DECL(int) GPciRawR0GuestPageUpdate(PGVM pGVM, RTGCPHYS GCPhys, RTHCPHYS HCPhys)
366	{
367	AssertReturn(!(GCPhys & PAGE_OFFSET_MASK), VERR_INTERNAL_ERROR_4);
368	AssertReturn(!(HCPhys & PAGE_OFFSET_MASK) \|\| HCPhys == NIL_RTHCPHYS, VERR_INTERNAL_ERROR_4);
369	NOREF(pGVM);
370	return VINF_SUCCESS;
371	}
372
373	#endif /* VBOX_WITH_PCI_PASSTHROUGH */
374
375
376	/**
377	* Sets up the IOMMU when raw PCI device is enabled.
378	*
379	* @note This is a hack that will probably be remodelled and refined later!
380	*
381	* @returns VBox status code.
382	*
383	* @param pGVM The global (ring-0) VM structure.
384	* @param pVM The cross context VM structure.
385	*/
386	VMMR0_INT_DECL(int) PGMR0PhysSetupIoMmu(PGVM pGVM, PVM pVM)
387	{
388	int rc = GVMMR0ValidateGVMandVM(pGVM, pVM);
389	if (RT_FAILURE(rc))
390	return rc;
391
392	#ifdef VBOX_WITH_PCI_PASSTHROUGH
393	if (pVM->pgm.s.fPciPassthrough)
394	{
395	/*
396	* The Simplistic Approach - Enumerate all the pages and call tell the
397	* IOMMU about each of them.
398	*/
399	pgmLock(pVM);
400	rc = GPciRawR0GuestPageBeginAssignments(pGVM);
401	if (RT_SUCCESS(rc))
402	{
403	for (PPGMRAMRANGE pRam = pVM->pgm.s.pRamRangesXR0; RT_SUCCESS(rc) && pRam; pRam = pRam->pNextR0)
404	{
405	PPGMPAGE pPage = &pRam->aPages[0];
406	RTGCPHYS GCPhys = pRam->GCPhys;
407	uint32_t cLeft = pRam->cb >> PAGE_SHIFT;
408	while (cLeft-- > 0)
409	{
410	/* Only expose pages that are 100% safe for now. */
411	if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_RAM
412	&& PGM_PAGE_GET_STATE(pPage) == PGM_PAGE_STATE_ALLOCATED
413	&& !PGM_PAGE_HAS_ANY_HANDLERS(pPage))
414	rc = GPciRawR0GuestPageAssign(pGVM, GCPhys, PGM_PAGE_GET_HCPHYS(pPage));
415	else
416	rc = GPciRawR0GuestPageUnassign(pGVM, GCPhys);
417
418	/* next */
419	pPage++;
420	GCPhys += PAGE_SIZE;
421	}
422	}
423
424	int rc2 = GPciRawR0GuestPageEndAssignments(pGVM);
425	if (RT_FAILURE(rc2) && RT_SUCCESS(rc))
426	rc = rc2;
427	}
428	pgmUnlock(pVM);
429	}
430	else
431	#endif
432	rc = VERR_NOT_SUPPORTED;
433	return rc;
434	}
435
436
437	/**
438	* \#PF Handler for nested paging.
439	*
440	* @returns VBox status code (appropriate for trap handling and GC return).
441	* @param pVM The cross context VM structure.
442	* @param pVCpu The cross context virtual CPU structure.
443	* @param enmShwPagingMode Paging mode for the nested page tables.
444	* @param uErr The trap error code.
445	* @param pRegFrame Trap register frame.
446	* @param GCPhysFault The fault address.
447	*/
448	VMMR0DECL(int) PGMR0Trap0eHandlerNestedPaging(PVM pVM, PVMCPU pVCpu, PGMMODE enmShwPagingMode, RTGCUINT uErr,
449	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault)
450	{
451	int rc;
452
453	LogFlow(("PGMTrap0eHandler: uErr=%RGx GCPhysFault=%RGp eip=%RGv\n", uErr, GCPhysFault, (RTGCPTR)pRegFrame->rip));
454	STAM_PROFILE_START(&pVCpu->pgm.s.StatRZTrap0e, a);
455	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = NULL; } );
456
457	/* AMD uses the host's paging mode; Intel has a single mode (EPT). */
458	AssertMsg( enmShwPagingMode == PGMMODE_32_BIT \|\| enmShwPagingMode == PGMMODE_PAE \|\| enmShwPagingMode == PGMMODE_PAE_NX
459	\|\| enmShwPagingMode == PGMMODE_AMD64 \|\| enmShwPagingMode == PGMMODE_AMD64_NX \|\| enmShwPagingMode == PGMMODE_EPT,
460	("enmShwPagingMode=%d\n", enmShwPagingMode));
461
462	/* Reserved shouldn't end up here. */
463	Assert(!(uErr & X86_TRAP_PF_RSVD));
464
465	#ifdef VBOX_WITH_STATISTICS
466	/*
467	* Error code stats.
468	*/
469	if (uErr & X86_TRAP_PF_US)
470	{
471	if (!(uErr & X86_TRAP_PF_P))
472	{
473	if (uErr & X86_TRAP_PF_RW)
474	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSNotPresentWrite);
475	else
476	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSNotPresentRead);
477	}
478	else if (uErr & X86_TRAP_PF_RW)
479	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSWrite);
480	else if (uErr & X86_TRAP_PF_RSVD)
481	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSReserved);
482	else if (uErr & X86_TRAP_PF_ID)
483	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSNXE);
484	else
485	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eUSRead);
486	}
487	else
488	{ /* Supervisor */
489	if (!(uErr & X86_TRAP_PF_P))
490	{
491	if (uErr & X86_TRAP_PF_RW)
492	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eSVNotPresentWrite);
493	else
494	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eSVNotPresentRead);
495	}
496	else if (uErr & X86_TRAP_PF_RW)
497	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eSVWrite);
498	else if (uErr & X86_TRAP_PF_ID)
499	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eSNXE);
500	else if (uErr & X86_TRAP_PF_RSVD)
501	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eSVReserved);
502	}
503	#endif
504
505	/*
506	* Call the worker.
507	*
508	* Note! We pretend the guest is in protected mode without paging, so we
509	* can use existing code to build the nested page tables.
510	*/
511	bool fLockTaken = false;
512	switch (enmShwPagingMode)
513	{
514	case PGMMODE_32_BIT:
515	rc = PGM_BTH_NAME_32BIT_PROT(Trap0eHandler)(pVCpu, uErr, pRegFrame, GCPhysFault, &fLockTaken);
516	break;
517	case PGMMODE_PAE:
518	case PGMMODE_PAE_NX:
519	rc = PGM_BTH_NAME_PAE_PROT(Trap0eHandler)(pVCpu, uErr, pRegFrame, GCPhysFault, &fLockTaken);
520	break;
521	case PGMMODE_AMD64:
522	case PGMMODE_AMD64_NX:
523	rc = PGM_BTH_NAME_AMD64_PROT(Trap0eHandler)(pVCpu, uErr, pRegFrame, GCPhysFault, &fLockTaken);
524	break;
525	case PGMMODE_EPT:
526	rc = PGM_BTH_NAME_EPT_PROT(Trap0eHandler)(pVCpu, uErr, pRegFrame, GCPhysFault, &fLockTaken);
527	break;
528	default:
529	AssertFailed();
530	rc = VERR_INVALID_PARAMETER;
531	break;
532	}
533	if (fLockTaken)
534	{
535	PGM_LOCK_ASSERT_OWNER(pVM);
536	pgmUnlock(pVM);
537	}
538
539	if (rc == VINF_PGM_SYNCPAGE_MODIFIED_PDE)
540	rc = VINF_SUCCESS;
541	/*
542	* Handle the case where we cannot interpret the instruction because we cannot get the guest physical address
543	* via its page tables, see @bugref{6043}.
544	*/
545	else if ( rc == VERR_PAGE_NOT_PRESENT /* SMP only ; disassembly might fail. */
546	\|\| rc == VERR_PAGE_TABLE_NOT_PRESENT /* seen with UNI & SMP */
547	\|\| rc == VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT /* seen with SMP */
548	\|\| rc == VERR_PAGE_MAP_LEVEL4_NOT_PRESENT) /* precaution */
549	{
550	Log(("WARNING: Unexpected VERR_PAGE_TABLE_NOT_PRESENT (%d) for page fault at %RGp error code %x (rip=%RGv)\n", rc, GCPhysFault, uErr, pRegFrame->rip));
551	/* Some kind of inconsistency in the SMP case; it's safe to just execute the instruction again; not sure about
552	single VCPU VMs though. */
553	rc = VINF_SUCCESS;
554	}
555
556	STAM_STATS({ if (!pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution))
557	pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0eTime2Misc; });
558	STAM_PROFILE_STOP_EX(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatRZTrap0e, pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution), a);
559	return rc;
560	}
561
562
563	/**
564	* \#PF Handler for deliberate nested paging misconfiguration (/reserved bit)
565	* employed for MMIO pages.
566	*
567	* @returns VBox status code (appropriate for trap handling and GC return).
568	* @param pVM The cross context VM structure.
569	* @param pVCpu The cross context virtual CPU structure.
570	* @param enmShwPagingMode Paging mode for the nested page tables.
571	* @param pRegFrame Trap register frame.
572	* @param GCPhysFault The fault address.
573	* @param uErr The error code, UINT32_MAX if not available
574	* (VT-x).
575	*/
576	VMMR0DECL(VBOXSTRICTRC) PGMR0Trap0eHandlerNPMisconfig(PVM pVM, PVMCPU pVCpu, PGMMODE enmShwPagingMode,
577	PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, uint32_t uErr)
578	{
579	#ifdef PGM_WITH_MMIO_OPTIMIZATIONS
580	STAM_PROFILE_START(&pVCpu->CTX_SUFF(pStats)->StatR0NpMiscfg, a);
581	VBOXSTRICTRC rc;
582
583	/*
584	* Try lookup the all access physical handler for the address.
585	*/
586	pgmLock(pVM);
587	PPGMPHYSHANDLER pHandler = pgmHandlerPhysicalLookup(pVM, GCPhysFault);
588	PPGMPHYSHANDLERTYPEINT pHandlerType = RT_LIKELY(pHandler) ? PGMPHYSHANDLER_GET_TYPE(pVM, pHandler) : NULL;
589	if (RT_LIKELY(pHandler && pHandlerType->enmKind != PGMPHYSHANDLERKIND_WRITE))
590	{
591	/*
592	* If the handle has aliases page or pages that have been temporarily
593	* disabled, we'll have to take a detour to make sure we resync them
594	* to avoid lots of unnecessary exits.
595	*/
596	PPGMPAGE pPage;
597	if ( ( pHandler->cAliasedPages
598	\|\| pHandler->cTmpOffPages)
599	&& ( (pPage = pgmPhysGetPage(pVM, GCPhysFault)) == NULL
600	\|\| PGM_PAGE_GET_HNDL_PHYS_STATE(pPage) == PGM_PAGE_HNDL_PHYS_STATE_DISABLED)
601	)
602	{
603	Log(("PGMR0Trap0eHandlerNPMisconfig: Resyncing aliases / tmp-off page at %RGp (uErr=%#x) %R[pgmpage]\n", GCPhysFault, uErr, pPage));
604	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatR0NpMiscfgSyncPage);
605	rc = pgmShwSyncNestedPageLocked(pVCpu, GCPhysFault, 1 /cPages/, enmShwPagingMode);
606	pgmUnlock(pVM);
607	}
608	else
609	{
610	if (pHandlerType->CTX_SUFF(pfnPfHandler))
611	{
612	void *pvUser = pHandler->CTX_SUFF(pvUser);
613	STAM_PROFILE_START(&pHandler->Stat, h);
614	pgmUnlock(pVM);
615
616	Log6(("PGMR0Trap0eHandlerNPMisconfig: calling %p(,%#x,,%RGp,%p)\n", pHandlerType->CTX_SUFF(pfnPfHandler), uErr, GCPhysFault, pvUser));
617	rc = pHandlerType->CTX_SUFF(pfnPfHandler)(pVM, pVCpu, uErr == UINT32_MAX ? RTGCPTR_MAX : uErr, pRegFrame,
618	GCPhysFault, GCPhysFault, pvUser);
619
620	#ifdef VBOX_WITH_STATISTICS
621	pgmLock(pVM);
622	pHandler = pgmHandlerPhysicalLookup(pVM, GCPhysFault);
623	if (pHandler)
624	STAM_PROFILE_STOP(&pHandler->Stat, h);
625	pgmUnlock(pVM);
626	#endif
627	}
628	else
629	{
630	pgmUnlock(pVM);
631	Log(("PGMR0Trap0eHandlerNPMisconfig: %RGp (uErr=%#x) -> R3\n", GCPhysFault, uErr));
632	rc = VINF_EM_RAW_EMULATE_INSTR;
633	}
634	}
635	}
636	else
637	{
638	/*
639	* Must be out of sync, so do a SyncPage and restart the instruction.
640	*
641	* ASSUMES that ALL handlers are page aligned and covers whole pages
642	* (assumption asserted in PGMHandlerPhysicalRegisterEx).
643	*/
644	Log(("PGMR0Trap0eHandlerNPMisconfig: Out of sync page at %RGp (uErr=%#x)\n", GCPhysFault, uErr));
645	STAM_COUNTER_INC(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatR0NpMiscfgSyncPage);
646	rc = pgmShwSyncNestedPageLocked(pVCpu, GCPhysFault, 1 /cPages/, enmShwPagingMode);
647	pgmUnlock(pVM);
648	}
649
650	STAM_PROFILE_STOP(&pVCpu->pgm.s.CTX_SUFF(pStats)->StatR0NpMiscfg, a);
651	return rc;
652
653	#else
654	AssertLogRelFailed();
655	return VERR_PGM_NOT_USED_IN_MODE;
656	#endif
657	}
658

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

source: vbox/trunk/src/VBox/VMM/VMMR0/PGMR0.cpp@ 71963

Download in other formats: