PGMR0.cpp@ 29287

Last change on this file since 29287 was 29217, checked in by vboxsync, 15 years ago
Shared paging updates
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 16.5 KB

Line
1	/* $Id: PGMR0.cpp 29217 2010-05-07 14:38:51Z vboxsync $ */
2	/** @file
3	* PGM - Page Manager and Monitor, Ring-0.
4	*/
5
6	/*
7	* Copyright (C) 2007 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/*******************************************************************************
19	* Header Files *
20	*******************************************************************************/
21	#define LOG_GROUP LOG_GROUP_PGM
22	#include <VBox/pgm.h>
23	#include "../PGMInternal.h"
24	#include <VBox/vm.h>
25	#include "../PGMInline.h"
26	#include <VBox/log.h>
27	#include <VBox/err.h>
28	#include <iprt/assert.h>
29	#include <iprt/mem.h>
30
31	RT_C_DECLS_BEGIN
32	#define PGM_BTH_NAME(name) PGM_BTH_NAME_32BIT_PROT(name)
33	#include "PGMR0Bth.h"
34	#undef PGM_BTH_NAME
35
36	#define PGM_BTH_NAME(name) PGM_BTH_NAME_PAE_PROT(name)
37	#include "PGMR0Bth.h"
38	#undef PGM_BTH_NAME
39
40	#define PGM_BTH_NAME(name) PGM_BTH_NAME_AMD64_PROT(name)
41	#include "PGMR0Bth.h"
42	#undef PGM_BTH_NAME
43
44	#define PGM_BTH_NAME(name) PGM_BTH_NAME_EPT_PROT(name)
45	#include "PGMR0Bth.h"
46	#undef PGM_BTH_NAME
47
48	RT_C_DECLS_END
49
50
51	/**
52	* Worker function for PGMR3PhysAllocateHandyPages and pgmPhysEnsureHandyPage.
53	*
54	* @returns The following VBox status codes.
55	* @retval VINF_SUCCESS on success. FF cleared.
56	* @retval VINF_EM_NO_MEMORY if we're out of memory. The FF is set in this case.
57	*
58	* @param pVM The VM handle.
59	* @param pVCpu The VMCPU handle.
60	*
61	* @remarks Must be called from within the PGM critical section. The caller
62	* must clear the new pages.
63	*/
64	VMMR0DECL(int) PGMR0PhysAllocateHandyPages(PVM pVM, PVMCPU pVCpu)
65	{
66	Assert(PDMCritSectIsOwnerEx(&pVM->pgm.s.CritSect, pVCpu->idCpu));
67
68	/*
69	* Check for error injection.
70	*/
71	if (RT_UNLIKELY(pVM->pgm.s.fErrInjHandyPages))
72	return VERR_NO_MEMORY;
73
74	/*
75	* Try allocate a full set of handy pages.
76	*/
77	uint32_t iFirst = pVM->pgm.s.cHandyPages;
78	AssertReturn(iFirst <= RT_ELEMENTS(pVM->pgm.s.aHandyPages), VERR_INTERNAL_ERROR);
79	uint32_t cPages = RT_ELEMENTS(pVM->pgm.s.aHandyPages) - iFirst;
80	if (!cPages)
81	return VINF_SUCCESS;
82	int rc = GMMR0AllocateHandyPages(pVM, pVCpu->idCpu, cPages, cPages, &pVM->pgm.s.aHandyPages[iFirst]);
83	if (RT_SUCCESS(rc))
84	{
85	for (uint32_t i = 0; i < RT_ELEMENTS(pVM->pgm.s.aHandyPages); i++)
86	{
87	Assert(pVM->pgm.s.aHandyPages[i].idPage != NIL_GMM_PAGEID);
88	Assert(pVM->pgm.s.aHandyPages[i].idPage <= GMM_PAGEID_LAST);
89	Assert(pVM->pgm.s.aHandyPages[i].idSharedPage == NIL_GMM_PAGEID);
90	Assert(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys != NIL_RTHCPHYS);
91	Assert(!(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys & ~X86_PTE_PAE_PG_MASK));
92	}
93
94	pVM->pgm.s.cHandyPages = RT_ELEMENTS(pVM->pgm.s.aHandyPages);
95	}
96	else if (rc != VERR_GMM_SEED_ME)
97	{
98	if ( ( rc == VERR_GMM_HIT_GLOBAL_LIMIT
99	\|\| rc == VERR_GMM_HIT_VM_ACCOUNT_LIMIT)
100	&& iFirst < PGM_HANDY_PAGES_MIN)
101	{
102
103	#ifdef VBOX_STRICT
104	/* We're ASSUMING that GMM has updated all the entires before failing us. */
105	uint32_t i;
106	for (i = iFirst; i < RT_ELEMENTS(pVM->pgm.s.aHandyPages); i++)
107	{
108	Assert(pVM->pgm.s.aHandyPages[i].idPage == NIL_GMM_PAGEID);
109	Assert(pVM->pgm.s.aHandyPages[i].idSharedPage == NIL_GMM_PAGEID);
110	Assert(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys == NIL_RTHCPHYS);
111	}
112	#endif
113
114	/*
115	* Reduce the number of pages until we hit the minimum limit.
116	*/
117	do
118	{
119	cPages >>= 2;
120	if (cPages + iFirst < PGM_HANDY_PAGES_MIN)
121	cPages = PGM_HANDY_PAGES_MIN - iFirst;
122	rc = GMMR0AllocateHandyPages(pVM, pVCpu->idCpu, cPages, cPages, &pVM->pgm.s.aHandyPages[iFirst]);
123	} while ( ( rc == VERR_GMM_HIT_GLOBAL_LIMIT
124	\|\| rc == VERR_GMM_HIT_VM_ACCOUNT_LIMIT)
125	&& cPages + iFirst > PGM_HANDY_PAGES_MIN);
126	if (RT_SUCCESS(rc))
127	{
128	#ifdef VBOX_STRICT
129	i = iFirst + cPages;
130	while (i-- > 0)
131	{
132	Assert(pVM->pgm.s.aHandyPages[i].idPage != NIL_GMM_PAGEID);
133	Assert(pVM->pgm.s.aHandyPages[i].idPage <= GMM_PAGEID_LAST);
134	Assert(pVM->pgm.s.aHandyPages[i].idSharedPage == NIL_GMM_PAGEID);
135	Assert(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys != NIL_RTHCPHYS);
136	Assert(!(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys & ~X86_PTE_PAE_PG_MASK));
137	}
138
139	for (i = cPages + iFirst; i < RT_ELEMENTS(pVM->pgm.s.aHandyPages); i++)
140	{
141	Assert(pVM->pgm.s.aHandyPages[i].idPage == NIL_GMM_PAGEID);
142	Assert(pVM->pgm.s.aHandyPages[i].idSharedPage == NIL_GMM_PAGEID);
143	Assert(pVM->pgm.s.aHandyPages[i].HCPhysGCPhys == NIL_RTHCPHYS);
144	}
145	#endif
146
147	pVM->pgm.s.cHandyPages = iFirst + cPages;
148	}
149	}
150
151	if (RT_FAILURE(rc) && rc != VERR_GMM_SEED_ME)
152	{
153	LogRel(("PGMR0PhysAllocateHandyPages: rc=%Rrc iFirst=%d cPages=%d\n", rc, iFirst, cPages));
154	VM_FF_SET(pVM, VM_FF_PGM_NO_MEMORY);
155	}
156	}
157
158
159	LogFlow(("PGMR0PhysAllocateHandyPages: cPages=%d rc=%Rrc\n", cPages, rc));
160	return rc;
161	}
162
163	/**
164	* Worker function for PGMR3PhysAllocateLargeHandyPage
165	*
166	* @returns The following VBox status codes.
167	* @retval VINF_SUCCESS on success.
168	* @retval VINF_EM_NO_MEMORY if we're out of memory.
169	*
170	* @param pVM The VM handle.
171	* @param pVCpu The VMCPU handle.
172	*
173	* @remarks Must be called from within the PGM critical section. The caller
174	* must clear the new pages.
175	*/
176	VMMR0DECL(int) PGMR0PhysAllocateLargeHandyPage(PVM pVM, PVMCPU pVCpu)
177	{
178	Assert(PDMCritSectIsOwnerEx(&pVM->pgm.s.CritSect, pVCpu->idCpu));
179
180	Assert(!pVM->pgm.s.cLargeHandyPages);
181	int rc = GMMR0AllocateLargePage(pVM, pVCpu->idCpu, _2M, &pVM->pgm.s.aLargeHandyPage[0].idPage, &pVM->pgm.s.aLargeHandyPage[0].HCPhysGCPhys);
182	if (RT_SUCCESS(rc))
183	pVM->pgm.s.cLargeHandyPages = 1;
184
185	return rc;
186	}
187
188	/**
189	* #PF Handler for nested paging.
190	*
191	* @returns VBox status code (appropriate for trap handling and GC return).
192	* @param pVM VM Handle.
193	* @param pVCpu VMCPU Handle.
194	* @param enmShwPagingMode Paging mode for the nested page tables
195	* @param uErr The trap error code.
196	* @param pRegFrame Trap register frame.
197	* @param pvFault The fault address.
198	*/
199	VMMR0DECL(int) PGMR0Trap0eHandlerNestedPaging(PVM pVM, PVMCPU pVCpu, PGMMODE enmShwPagingMode, RTGCUINT uErr, PCPUMCTXCORE pRegFrame, RTGCPHYS pvFault)
200	{
201	int rc;
202
203	LogFlow(("PGMTrap0eHandler: uErr=%RGx pvFault=%RGp eip=%RGv\n", uErr, pvFault, (RTGCPTR)pRegFrame->rip));
204	STAM_PROFILE_START(&pVCpu->pgm.s.StatRZTrap0e, a);
205	STAM_STATS({ pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = NULL; } );
206
207	/* AMD uses the host's paging mode; Intel has a single mode (EPT). */
208	AssertMsg(enmShwPagingMode == PGMMODE_32_BIT \|\| enmShwPagingMode == PGMMODE_PAE \|\| enmShwPagingMode == PGMMODE_PAE_NX \|\| enmShwPagingMode == PGMMODE_AMD64 \|\| enmShwPagingMode == PGMMODE_AMD64_NX \|\| enmShwPagingMode == PGMMODE_EPT, ("enmShwPagingMode=%d\n", enmShwPagingMode));
209
210	#ifdef VBOX_WITH_STATISTICS
211	/*
212	* Error code stats.
213	*/
214	if (uErr & X86_TRAP_PF_US)
215	{
216	if (!(uErr & X86_TRAP_PF_P))
217	{
218	if (uErr & X86_TRAP_PF_RW)
219	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eUSNotPresentWrite);
220	else
221	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eUSNotPresentRead);
222	}
223	else if (uErr & X86_TRAP_PF_RW)
224	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eUSWrite);
225	else if (uErr & X86_TRAP_PF_RSVD)
226	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eUSReserved);
227	else if (uErr & X86_TRAP_PF_ID)
228	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eUSNXE);
229	else
230	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eUSRead);
231	}
232	else
233	{ /* Supervisor */
234	if (!(uErr & X86_TRAP_PF_P))
235	{
236	if (uErr & X86_TRAP_PF_RW)
237	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eSVNotPresentWrite);
238	else
239	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eSVNotPresentRead);
240	}
241	else if (uErr & X86_TRAP_PF_RW)
242	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eSVWrite);
243	else if (uErr & X86_TRAP_PF_ID)
244	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eSNXE);
245	else if (uErr & X86_TRAP_PF_RSVD)
246	STAM_COUNTER_INC(&pVCpu->pgm.s.StatRZTrap0eSVReserved);
247	}
248	#endif
249
250	/*
251	* Call the worker.
252	*
253	* We pretend the guest is in protected mode without paging, so we can use existing code to build the
254	* nested page tables.
255	*/
256	bool fLockTaken = false;
257	switch(enmShwPagingMode)
258	{
259	case PGMMODE_32_BIT:
260	rc = PGM_BTH_NAME_32BIT_PROT(Trap0eHandler)(pVCpu, uErr, pRegFrame, pvFault, &fLockTaken);
261	break;
262	case PGMMODE_PAE:
263	case PGMMODE_PAE_NX:
264	rc = PGM_BTH_NAME_PAE_PROT(Trap0eHandler)(pVCpu, uErr, pRegFrame, pvFault, &fLockTaken);
265	break;
266	case PGMMODE_AMD64:
267	case PGMMODE_AMD64_NX:
268	rc = PGM_BTH_NAME_AMD64_PROT(Trap0eHandler)(pVCpu, uErr, pRegFrame, pvFault, &fLockTaken);
269	break;
270	case PGMMODE_EPT:
271	rc = PGM_BTH_NAME_EPT_PROT(Trap0eHandler)(pVCpu, uErr, pRegFrame, pvFault, &fLockTaken);
272	break;
273	default:
274	AssertFailed();
275	rc = VERR_INVALID_PARAMETER;
276	break;
277	}
278	if (fLockTaken)
279	{
280	Assert(PGMIsLockOwner(pVM));
281	pgmUnlock(pVM);
282	}
283	if (rc == VINF_PGM_SYNCPAGE_MODIFIED_PDE)
284	rc = VINF_SUCCESS;
285	else
286	/* Note: hack alert for difficult to reproduce problem. */
287	if ( rc == VERR_PAGE_NOT_PRESENT /* SMP only ; disassembly might fail. */
288	\|\| rc == VERR_PAGE_TABLE_NOT_PRESENT /* seen with UNI & SMP */
289	\|\| rc == VERR_PAGE_DIRECTORY_PTR_NOT_PRESENT /* seen with SMP */
290	\|\| rc == VERR_PAGE_MAP_LEVEL4_NOT_PRESENT) /* precaution */
291	{
292	Log(("WARNING: Unexpected VERR_PAGE_TABLE_NOT_PRESENT (%d) for page fault at %RGp error code %x (rip=%RGv)\n", rc, pvFault, uErr, pRegFrame->rip));
293	/* Some kind of inconsistency in the SMP case; it's safe to just execute the instruction again; not sure about single VCPU VMs though. */
294	rc = VINF_SUCCESS;
295	}
296
297	STAM_STATS({ if (!pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution))
298	pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution) = &pVCpu->pgm.s.StatRZTrap0eTime2Misc; });
299	STAM_PROFILE_STOP_EX(&pVCpu->pgm.s.StatRZTrap0e, pVCpu->pgm.s.CTX_SUFF(pStatTrap0eAttribution), a);
300	return rc;
301	}
302
303	#ifdef VBOX_WITH_PAGE_SHARING
304	/**
305	* Check a registered module for shared page changes
306	*
307	* @returns The following VBox status codes.
308	*
309	* @param pVM The VM handle.
310	* @param pVCpu The VMCPU handle.
311	* @param pReq Module request packet
312	*/
313	VMMR0DECL(int) PGMR0SharedModuleCheck(PVM pVM, PVMCPU pVCpu, PGMMREGISTERSHAREDMODULEREQ pReq)
314	{
315	int rc = VINF_SUCCESS;
316	PGMMSHAREDPAGEDESC paPageDesc = NULL;
317	uint32_t cbPreviousRegion = 0;
318	bool fFlushTLBs = false;
319
320	/*
321	* Validate input.
322	*/
323	AssertPtrReturn(pReq, VERR_INVALID_POINTER);
324	AssertMsgReturn(pReq->Hdr.cbReq >= sizeof(pReq) && pReq->Hdr.cbReq == RT_UOFFSETOF(GMMREGISTERSHAREDMODULEREQ, aRegions[pReq->cRegions]), ("%#x != %#x\n", pReq->Hdr.cbReq, sizeof(pReq)), VERR_INVALID_PARAMETER);
325
326	pgmLock(pVM);
327
328	/* Check every region of the shared module. */
329	for (unsigned i = 0; i < pReq->cRegions; i++)
330	{
331	Assert((pReq->aRegions[i].cbRegion & 0xfff) == 0);
332	Assert((pReq->aRegions[i].GCRegionAddr & 0xfff) == 0);
333
334	RTGCPTR GCRegion = pReq->aRegions[i].GCRegionAddr;
335	unsigned cbRegion = pReq->aRegions[i].cbRegion & ~0xfff;
336	unsigned idxPage = 0;
337	bool fValidChanges = false;
338
339	if (cbPreviousRegion < cbRegion)
340	{
341	if (paPageDesc)
342	RTMemFree(paPageDesc);
343
344	paPageDesc = (PGMMSHAREDPAGEDESC)RTMemAlloc((cbRegion >> PAGE_SHIFT) * sizeof(*paPageDesc));
345	if (!paPageDesc)
346	{
347	AssertFailed();
348	rc = VERR_NO_MEMORY;
349	goto end;
350	}
351	cbPreviousRegion = cbRegion;
352	}
353
354	while (cbRegion)
355	{
356	RTGCPHYS GCPhys;
357	uint64_t fFlags;
358
359	rc = PGMGstGetPage(pVCpu, GCRegion, &GCPhys, &fFlags);
360	if ( rc == VINF_SUCCESS
361	&& !(fFlags & X86_PTE_RW)) /* important as we make assumptions about this below! */
362	{
363	PPGMPAGE pPage = pgmPhysGetPage(&pVM->pgm.s, GCPhys);
364	if ( pPage
365	&& !PGM_PAGE_IS_SHARED(pPage))
366	{
367	fValidChanges = true;
368	paPageDesc[idxPage].uHCPhysPageId = PGM_PAGE_GET_PAGEID(pPage);
369	paPageDesc[idxPage].HCPhys = PGM_PAGE_GET_HCPHYS(pPage);
370	paPageDesc[idxPage].GCPhys = GCPhys;
371	}
372	else
373	paPageDesc[idxPage].uHCPhysPageId = NIL_GMM_PAGEID;
374	}
375	else
376	paPageDesc[idxPage].uHCPhysPageId = NIL_GMM_PAGEID;
377
378	idxPage++;
379	GCRegion += PAGE_SIZE;
380	cbRegion -= PAGE_SIZE;
381	}
382
383	if (fValidChanges)
384	{
385	rc = GMMR0SharedModuleCheckRange(pVM, pVCpu->idCpu, pReq, i, idxPage, paPageDesc);
386	AssertRC(rc);
387	if (RT_FAILURE(rc))
388	break;
389
390	for (unsigned i = 0; i < idxPage; i++)
391	{
392	/* Any change for this page? */
393	if (paPageDesc[i].uHCPhysPageId != NIL_GMM_PAGEID)
394	{
395	/** todo: maybe cache these to prevent the nth lookup. */
396	PPGMPAGE pPage = pgmPhysGetPage(&pVM->pgm.s, paPageDesc[i].GCPhys);
397	if (!pPage)
398	{
399	/* Should never happen. */
400	AssertFailed();
401	rc = VERR_PGM_PHYS_INVALID_PAGE_ID;
402	goto end;
403	}
404	Assert(!PGM_PAGE_IS_SHARED(pPage));
405
406	if (paPageDesc[i].HCPhys != PGM_PAGE_GET_HCPHYS(pPage))
407	{
408	bool fFlush = false;
409
410	/* Page was replaced by an existing shared version of it; clear all references first. */
411	rc = pgmPoolTrackUpdateGCPhys(pVM, paPageDesc[i].GCPhys, pPage, true /* clear the entries */, &fFlush);
412	if (RT_FAILURE(rc))
413	{
414	AssertRC(rc);
415	goto end;
416	}
417	Assert(rc == VINF_SUCCESS \|\| (VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3) && (pVCpu->pgm.s.fSyncFlags & PGM_SYNC_CLEAR_PGM_POOL)));
418	if (rc = VINF_SUCCESS)
419	fFlushTLBs \|= fFlush;
420
421	/* Update the physical address and page id now. */
422	PGM_PAGE_SET_HCPHYS(pPage, paPageDesc[i].HCPhys);
423	PGM_PAGE_SET_PAGEID(pPage, paPageDesc[i].uHCPhysPageId);
424
425	/* Invalidate page map TLB entry for this page too. */
426	PGMPhysInvalidatePageMapTLBEntry(pVM, paPageDesc[i].GCPhys);
427	}
428	/* else nothing changed (== this page is now a shared page), so no need to flush anything. */
429
430	PGM_PAGE_SET_STATE(pPage, PGM_PAGE_STATE_SHARED);
431	}
432	}
433	}
434	}
435
436	end:
437	pgmUnlock(pVM);
438	if (fFlushTLBs)
439	PGM_INVL_ALL_VCPU_TLBS(pVM);
440
441	if (paPageDesc)
442	RTMemFree(paPageDesc);
443
444	return rc;
445	}
446	#endif

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

source: vbox/trunk/src/VBox/VMM/VMMR0/PGMR0.cpp@ 29287

Download in other formats: