PGMPool.cpp@ 26214

Last change on this file since 26214 was 26150, checked in by vboxsync, 15 years ago
PGM: Split out the inlined code from PGMInternal.h and into PGMInline.h so we can drop all the &pVM->pgm.s and &pVCpu->pgm.s stuff.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 43.3 KB

Line
1	/* $Id: PGMPool.cpp 26150 2010-02-02 15:52:54Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
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_pgm_pool PGM Shadow Page Pool
23	*
24	* Motivations:
25	* -# Relationship between shadow page tables and physical guest pages. This
26	* should allow us to skip most of the global flushes now following access
27	* handler changes. The main expense is flushing shadow pages.
28	* -# Limit the pool size if necessary (default is kind of limitless).
29	* -# Allocate shadow pages from RC. We use to only do this in SyncCR3.
30	* -# Required for 64-bit guests.
31	* -# Combining the PD cache and page pool in order to simplify caching.
32	*
33	*
34	* @section sec_pgm_pool_outline Design Outline
35	*
36	* The shadow page pool tracks pages used for shadowing paging structures (i.e.
37	* page tables, page directory, page directory pointer table and page map
38	* level-4). Each page in the pool has an unique identifier. This identifier is
39	* used to link a guest physical page to a shadow PT. The identifier is a
40	* non-zero value and has a relativly low max value - say 14 bits. This makes it
41	* possible to fit it into the upper bits of the of the aHCPhys entries in the
42	* ram range.
43	*
44	* By restricting host physical memory to the first 48 bits (which is the
45	* announced physical memory range of the K8L chip (scheduled for 2008)), we
46	* can safely use the upper 16 bits for shadow page ID and reference counting.
47	*
48	* Update: The 48 bit assumption will be lifted with the new physical memory
49	* management (PGMPAGE), so we won't have any trouble when someone stuffs 2TB
50	* into a box in some years.
51	*
52	* Now, it's possible for a page to be aliased, i.e. mapped by more than one PT
53	* or PD. This is solved by creating a list of physical cross reference extents
54	* when ever this happens. Each node in the list (extent) is can contain 3 page
55	* pool indexes. The list it self is chained using indexes into the paPhysExt
56	* array.
57	*
58	*
59	* @section sec_pgm_pool_life Life Cycle of a Shadow Page
60	*
61	* -# The SyncPT function requests a page from the pool.
62	* The request includes the kind of page it is (PT/PD, PAE/legacy), the
63	* address of the page it's shadowing, and more.
64	* -# The pool responds to the request by allocating a new page.
65	* When the cache is enabled, it will first check if it's in the cache.
66	* Should the pool be exhausted, one of two things can be done:
67	* -# Flush the whole pool and current CR3.
68	* -# Use the cache to find a page which can be flushed (~age).
69	* -# The SyncPT function will sync one or more pages and insert it into the
70	* shadow PD.
71	* -# The SyncPage function may sync more pages on a later \#PFs.
72	* -# The page is freed / flushed in SyncCR3 (perhaps) and some other cases.
73	* When caching is enabled, the page isn't flush but remains in the cache.
74	*
75	*
76	* @section sec_pgm_pool_impl Monitoring
77	*
78	* We always monitor PAGE_SIZE chunks of memory. When we've got multiple shadow
79	* pages for the same PAGE_SIZE of guest memory (PAE and mixed PD/PT) the pages
80	* sharing the monitor get linked using the iMonitoredNext/Prev. The head page
81	* is the pvUser to the access handlers.
82	*
83	*
84	* @section sec_pgm_pool_impl Implementation
85	*
86	* The pool will take pages from the MM page pool. The tracking data
87	* (attributes, bitmaps and so on) are allocated from the hypervisor heap. The
88	* pool content can be accessed both by using the page id and the physical
89	* address (HC). The former is managed by means of an array, the latter by an
90	* offset based AVL tree.
91	*
92	* Flushing of a pool page means that we iterate the content (we know what kind
93	* it is) and updates the link information in the ram range.
94	*
95	* ...
96	*/
97
98
99	/*******************************************************************************
100	* Header Files *
101	*******************************************************************************/
102	#define LOG_GROUP LOG_GROUP_PGM_POOL
103	#include <VBox/pgm.h>
104	#include <VBox/mm.h>
105	#include "PGMInternal.h"
106	#include <VBox/vm.h>
107	#include "PGMInline.h"
108
109	#include <VBox/log.h>
110	#include <VBox/err.h>
111	#include <iprt/asm.h>
112	#include <iprt/string.h>
113	#include <VBox/dbg.h>
114
115
116	/*******************************************************************************
117	* Internal Functions *
118	*******************************************************************************/
119	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser);
120	#ifdef VBOX_WITH_DEBUGGER
121	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PVM pVM, PCDBGCVAR paArgs, unsigned cArgs, PDBGCVAR pResult);
122	#endif
123
124	#ifdef VBOX_WITH_DEBUGGER
125	/** Command descriptors. */
126	static const DBGCCMD g_aCmds[] =
127	{
128	/* pszCmd, cArgsMin, cArgsMax, paArgDesc, cArgDescs, pResultDesc, fFlags, pfnHandler pszSyntax, ....pszDescription */
129	{ "pgmpoolcheck", 0, 0, NULL, 0, NULL, 0, pgmR3PoolCmdCheck, "", "Check the pgm pool pages." },
130	};
131	#endif
132
133	/**
134	* Initalizes the pool
135	*
136	* @returns VBox status code.
137	* @param pVM The VM handle.
138	*/
139	int pgmR3PoolInit(PVM pVM)
140	{
141	AssertCompile(NIL_PGMPOOL_IDX == 0);
142	/* pPage->cLocked is an unsigned byte. */
143	AssertCompile(VMM_MAX_CPU_COUNT <= 255);
144
145	/*
146	* Query Pool config.
147	*/
148	PCFGMNODE pCfg = CFGMR3GetChild(CFGMR3GetRoot(pVM), "/PGM/Pool");
149
150	/** @cfgm{/PGM/Pool/MaxPages, uint16_t, #pages, 16, 0x3fff, 1024}
151	* The max size of the shadow page pool in pages. The pool will grow dynamically
152	* up to this limit.
153	*/
154	uint16_t cMaxPages;
155	int rc = CFGMR3QueryU16Def(pCfg, "MaxPages", &cMaxPages, 4*_1M >> PAGE_SHIFT);
156	AssertLogRelRCReturn(rc, rc);
157	AssertLogRelMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
158	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
159	cMaxPages = RT_ALIGN(cMaxPages, 16);
160
161	/** @cfgm{/PGM/Pool/MaxUsers, uint16_t, #users, MaxUsers, 32K, MaxPages*2}
162	* The max number of shadow page user tracking records. Each shadow page has
163	* zero of other shadow pages (or CR3s) that references it, or uses it if you
164	* like. The structures describing these relationships are allocated from a
165	* fixed sized pool. This configuration variable defines the pool size.
166	*/
167	uint16_t cMaxUsers;
168	rc = CFGMR3QueryU16Def(pCfg, "MaxUsers", &cMaxUsers, cMaxPages * 2);
169	AssertLogRelRCReturn(rc, rc);
170	AssertLogRelMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
171	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
172
173	/** @cfgm{/PGM/Pool/MaxPhysExts, uint16_t, #extents, 16, MaxPages * 2, MAX(MaxPages*2,0x3fff)}
174	* The max number of extents for tracking aliased guest pages.
175	*/
176	uint16_t cMaxPhysExts;
177	rc = CFGMR3QueryU16Def(pCfg, "MaxPhysExts", &cMaxPhysExts, RT_MAX(cMaxPages * 2, PGMPOOL_IDX_LAST));
178	AssertLogRelRCReturn(rc, rc);
179	AssertLogRelMsgReturn(cMaxPhysExts >= 16 && cMaxPages <= PGMPOOL_IDX_LAST,
180	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxPhysExts), VERR_INVALID_PARAMETER);
181
182	/** @cfgm{/PGM/Pool/ChacheEnabled, bool, true}
183	* Enables or disabling caching of shadow pages. Chaching means that we will try
184	* reuse shadow pages instead of recreating them everything SyncCR3, SyncPT or
185	* SyncPage requests one. When reusing a shadow page, we can save time
186	* reconstructing it and it's children.
187	*/
188	bool fCacheEnabled;
189	rc = CFGMR3QueryBoolDef(pCfg, "CacheEnabled", &fCacheEnabled, true);
190	AssertLogRelRCReturn(rc, rc);
191
192	Log(("pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
193	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
194
195	/*
196	* Allocate the data structures.
197	*/
198	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
199	cb += cMaxUsers * sizeof(PGMPOOLUSER);
200	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
201	PPGMPOOL pPool;
202	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
203	if (RT_FAILURE(rc))
204	return rc;
205	pVM->pgm.s.pPoolR3 = pPool;
206	pVM->pgm.s.pPoolR0 = MMHyperR3ToR0(pVM, pPool);
207	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pPool);
208
209	/*
210	* Initialize it.
211	*/
212	pPool->pVMR3 = pVM;
213	pPool->pVMR0 = pVM->pVMR0;
214	pPool->pVMRC = pVM->pVMRC;
215	pPool->cMaxPages = cMaxPages;
216	pPool->cCurPages = PGMPOOL_IDX_FIRST;
217	pPool->iUserFreeHead = 0;
218	pPool->cMaxUsers = cMaxUsers;
219	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
220	pPool->paUsersR3 = paUsers;
221	pPool->paUsersR0 = MMHyperR3ToR0(pVM, paUsers);
222	pPool->paUsersRC = MMHyperR3ToRC(pVM, paUsers);
223	for (unsigned i = 0; i < cMaxUsers; i++)
224	{
225	paUsers[i].iNext = i + 1;
226	paUsers[i].iUser = NIL_PGMPOOL_IDX;
227	paUsers[i].iUserTable = 0xfffffffe;
228	}
229	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
230	pPool->iPhysExtFreeHead = 0;
231	pPool->cMaxPhysExts = cMaxPhysExts;
232	PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
233	pPool->paPhysExtsR3 = paPhysExts;
234	pPool->paPhysExtsR0 = MMHyperR3ToR0(pVM, paPhysExts);
235	pPool->paPhysExtsRC = MMHyperR3ToRC(pVM, paPhysExts);
236	for (unsigned i = 0; i < cMaxPhysExts; i++)
237	{
238	paPhysExts[i].iNext = i + 1;
239	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
240	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
241	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
242	}
243	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
244	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
245	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
246	pPool->iAgeHead = NIL_PGMPOOL_IDX;
247	pPool->iAgeTail = NIL_PGMPOOL_IDX;
248	pPool->fCacheEnabled = fCacheEnabled;
249	pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
250	pPool->pszAccessHandler = "Guest Paging Access Handler";
251	pPool->HCPhysTree = 0;
252
253	/* The NIL entry. */
254	Assert(NIL_PGMPOOL_IDX == 0);
255	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
256
257	/* The Shadow 32-bit PD. (32 bits guest paging) */
258	pPool->aPages[PGMPOOL_IDX_PD].Core.Key = NIL_RTHCPHYS;
259	pPool->aPages[PGMPOOL_IDX_PD].GCPhys = NIL_RTGCPHYS;
260	pPool->aPages[PGMPOOL_IDX_PD].pvPageR3 = 0;
261	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_32BIT_PD;
262	pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
263
264	/* The Shadow PDPT. */
265	pPool->aPages[PGMPOOL_IDX_PDPT].Core.Key = NIL_RTHCPHYS;
266	pPool->aPages[PGMPOOL_IDX_PDPT].GCPhys = NIL_RTGCPHYS;
267	pPool->aPages[PGMPOOL_IDX_PDPT].pvPageR3 = 0;
268	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_PAE_PDPT;
269	pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
270
271	/* The Shadow AMD64 CR3. */
272	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].Core.Key = NIL_RTHCPHYS;
273	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].GCPhys = NIL_RTGCPHYS;
274	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].pvPageR3 = 0;
275	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].enmKind = PGMPOOLKIND_64BIT_PML4;
276	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].idx = PGMPOOL_IDX_AMD64_CR3;
277
278	/* The Nested Paging CR3. */
279	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].Core.Key = NIL_RTHCPHYS;
280	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].GCPhys = NIL_RTGCPHYS;
281	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].pvPageR3 = 0;
282	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].enmKind = PGMPOOLKIND_ROOT_NESTED;
283	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].idx = PGMPOOL_IDX_NESTED_ROOT;
284
285	/*
286	* Set common stuff.
287	*/
288	for (unsigned iPage = 1; iPage < PGMPOOL_IDX_FIRST; iPage++)
289	{
290	pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
291	pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
292	pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
293	pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
294	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
295	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
296	pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
297	pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
298	Assert(pPool->aPages[iPage].idx == iPage);
299	Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
300	Assert(!pPool->aPages[iPage].fSeenNonGlobal);
301	Assert(!pPool->aPages[iPage].fMonitored);
302	Assert(!pPool->aPages[iPage].fCached);
303	Assert(!pPool->aPages[iPage].fZeroed);
304	Assert(!pPool->aPages[iPage].fReusedFlushPending);
305	}
306
307	#ifdef VBOX_WITH_STATISTICS
308	/*
309	* Register statistics.
310	*/
311	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
312	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
313	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
314	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
315	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
316	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolClearAll.");
317	STAM_REG(pVM, &pPool->StatR3Reset, STAMTYPE_PROFILE, "/PGM/Pool/R3Reset", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolReset.");
318	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
319	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
320	STAM_REG(pVM, &pPool->StatForceFlushPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForce", STAMUNIT_OCCURENCES, "Counting explicit flushes by PGMPoolFlushPage().");
321	STAM_REG(pVM, &pPool->StatForceFlushDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForceDirty", STAMUNIT_OCCURENCES, "Counting explicit flushes of dirty pages by PGMPoolFlushPage().");
322	STAM_REG(pVM, &pPool->StatForceFlushReused, STAMTYPE_COUNTER, "/PGM/Pool/FlushReused", STAMUNIT_OCCURENCES, "Counting flushes for reused pages.");
323	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spent zeroing pages. Overlaps with Alloc.");
324	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
325	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
326	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackDeref.");
327	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPT.");
328	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
329	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
330	STAM_REG(pVM, &pPool->StatTrackFlushEntry, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Flush", STAMUNIT_COUNT, "Nr of flushed entries.");
331	STAM_REG(pVM, &pPool->StatTrackFlushEntryKeep, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Update", STAMUNIT_COUNT, "Nr of updated entries.");
332	STAM_REG(pVM, &pPool->StatTrackFreeUpOneUser, STAMTYPE_COUNTER, "/PGM/Pool/Track/FreeUpOneUser", STAMUNIT_TICKS_PER_CALL, "The number of times we were out of user tracking records.");
333	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_TICKS_PER_CALL, "Profiling deref activity related tracking GC physical pages.");
334	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
335	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
336	STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
337	STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
338	STAM_REG(pVM, &pPool->StatMonitorRZFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling the pgmPoolFlushPage calls made from the RC/R0 access handler.");
339	STAM_REG(pVM, &pPool->StatMonitorRZFlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
340	STAM_REG(pVM, &pPool->StatMonitorRZFlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
341	STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
342	STAM_REG(pVM, &pPool->StatMonitorRZHandled, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/Handled", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access we've handled (except REP STOSD).");
343	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
344	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch2, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch2", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction during flushing.");
345	STAM_REG(pVM, &pPool->StatMonitorRZRepPrefix, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/RepPrefix", STAMUNIT_OCCURENCES, "The number of times we've seen rep prefixes we can't handle.");
346	STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
347	STAM_REG(pVM, &pPool->StatMonitorRZFaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
348	STAM_REG(pVM, &pPool->StatMonitorRZFaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
349	STAM_REG(pVM, &pPool->StatMonitorRZFaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
350	STAM_REG(pVM, &pPool->StatMonitorRZFaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
351	STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
352	STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
353	STAM_REG(pVM, &pPool->StatMonitorR3FlushPage, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling the pgmPoolFlushPage calls made from the R3 access handler.");
354	STAM_REG(pVM, &pPool->StatMonitorR3FlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
355	STAM_REG(pVM, &pPool->StatMonitorR3FlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
356	STAM_REG(pVM, &pPool->StatMonitorR3Fork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
357	STAM_REG(pVM, &pPool->StatMonitorR3Handled, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/Handled", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access we've handled (except REP STOSD).");
358	STAM_REG(pVM, &pPool->StatMonitorR3RepPrefix, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/RepPrefix", STAMUNIT_OCCURENCES, "The number of times we've seen rep prefixes we can't handle.");
359	STAM_REG(pVM, &pPool->StatMonitorR3RepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
360	STAM_REG(pVM, &pPool->StatMonitorR3FaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
361	STAM_REG(pVM, &pPool->StatMonitorR3FaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
362	STAM_REG(pVM, &pPool->StatMonitorR3FaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
363	STAM_REG(pVM, &pPool->StatMonitorR3FaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
364	STAM_REG(pVM, &pPool->StatMonitorR3Async, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Async", STAMUNIT_OCCURENCES, "Times we're called in an async thread and need to flush.");
365	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
366	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
367	STAM_REG(pVM, &pPool->StatResetDirtyPages, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Resets", STAMUNIT_OCCURENCES, "Times we've called pgmPoolResetDirtyPages (and there were dirty page).");
368	STAM_REG(pVM, &pPool->StatDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Pages", STAMUNIT_OCCURENCES, "Times we've called pgmPoolAddDirtyPage.");
369	STAM_REG(pVM, &pPool->StatDirtyPageDupFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushDup", STAMUNIT_OCCURENCES, "Times we've had to flush duplicates for dirty page management.");
370	STAM_REG(pVM, &pPool->StatDirtyPageOverFlowFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushOverflow",STAMUNIT_OCCURENCES, "Times we've had to flush because of overflow.");
371	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
372	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
373	STAM_REG(pVM, &pPool->StatCacheKindMismatches, STAMTYPE_COUNTER, "/PGM/Pool/Cache/KindMismatches", STAMUNIT_OCCURENCES, "The number of shadow page kind mismatches. (Better be low, preferably 0!)");
374	STAM_REG(pVM, &pPool->StatCacheFreeUpOne, STAMTYPE_COUNTER, "/PGM/Pool/Cache/FreeUpOne", STAMUNIT_OCCURENCES, "The number of times the cache was asked to free up a page.");
375	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
376	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
377	#endif /* VBOX_WITH_STATISTICS */
378
379	#ifdef VBOX_WITH_DEBUGGER
380	/*
381	* Debugger commands.
382	*/
383	static bool s_fRegisteredCmds = false;
384	if (!s_fRegisteredCmds)
385	{
386	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
387	if (RT_SUCCESS(rc))
388	s_fRegisteredCmds = true;
389	}
390	#endif
391
392	return VINF_SUCCESS;
393	}
394
395
396	/**
397	* Relocate the page pool data.
398	*
399	* @param pVM The VM handle.
400	*/
401	void pgmR3PoolRelocate(PVM pVM)
402	{
403	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
404	pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
405	pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
406	pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
407	int rc = PDMR3LdrGetSymbolRC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerRC);
408	AssertReleaseRC(rc);
409	/* init order hack. */
410	if (!pVM->pgm.s.pPoolR3->pfnAccessHandlerR0)
411	{
412	rc = PDMR3LdrGetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerR0);
413	AssertReleaseRC(rc);
414	}
415	}
416
417
418	/**
419	* Grows the shadow page pool.
420	*
421	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
422	*
423	* @returns VBox status code.
424	* @param pVM The VM handle.
425	*/
426	VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
427	{
428	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
429	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_INTERNAL_ERROR);
430
431	pgmLock(pVM);
432
433	/*
434	* How much to grow it by?
435	*/
436	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
437	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
438	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages.\n", cPages, cPages));
439
440	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
441	{
442	PPGMPOOLPAGE pPage = &pPool->aPages[i];
443
444	/* Allocate all pages in low (below 4 GB) memory as 32 bits guests need a page table root in low memory. */
445	pPage->pvPageR3 = MMR3PageAllocLow(pVM);
446	if (!pPage->pvPageR3)
447	{
448	Log(("We're out of memory!! i=%d\n", i));
449	pgmUnlock(pVM);
450	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
451	}
452	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
453	pPage->GCPhys = NIL_RTGCPHYS;
454	pPage->enmKind = PGMPOOLKIND_FREE;
455	pPage->idx = pPage - &pPool->aPages[0];
456	LogFlow(("PGMR3PoolGrow: insert page #%#x - %RHp\n", pPage->idx, pPage->Core.Key));
457	pPage->iNext = pPool->iFreeHead;
458	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
459	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
460	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
461	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
462	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
463	pPage->iAgeNext = NIL_PGMPOOL_IDX;
464	pPage->iAgePrev = NIL_PGMPOOL_IDX;
465	/* commit it */
466	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
467	pPool->iFreeHead = i;
468	pPool->cCurPages = i + 1;
469	}
470
471	pgmUnlock(pVM);
472	Assert(pPool->cCurPages <= pPool->cMaxPages);
473	return VINF_SUCCESS;
474	}
475
476
477
478	/**
479	* Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
480	*
481	* @param pPool The pool.
482	* @param pPage The page.
483	*/
484	static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
485	{
486	/* for the present this should be safe enough I think... */
487	pgmLock(pPool->pVMR3);
488	if ( pPage->fReusedFlushPending
489	&& pPage->enmKind != PGMPOOLKIND_FREE)
490	pgmPoolFlushPage(pPool, pPage);
491	pgmUnlock(pPool->pVMR3);
492	}
493
494
495	/**
496	* \#PF Handler callback for PT write accesses.
497	*
498	* The handler can not raise any faults, it's mainly for monitoring write access
499	* to certain pages.
500	*
501	* @returns VINF_SUCCESS if the handler has carried out the operation.
502	* @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
503	* @param pVM VM Handle.
504	* @param GCPhys The physical address the guest is writing to.
505	* @param pvPhys The HC mapping of that address.
506	* @param pvBuf What the guest is reading/writing.
507	* @param cbBuf How much it's reading/writing.
508	* @param enmAccessType The access type.
509	* @param pvUser User argument.
510	*/
511	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser)
512	{
513	STAM_PROFILE_START(&pVM->pgm.s.pPoolR3->StatMonitorR3, a);
514	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
515	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
516	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%RGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
517	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
518
519	PVMCPU pVCpu = VMMGetCpu(pVM);
520
521	/*
522	* We don't have to be very sophisticated about this since there are relativly few calls here.
523	* However, we must try our best to detect any non-cpu accesses (disk / networking).
524	*
525	* Just to make life more interesting, we'll have to deal with the async threads too.
526	* We cannot flush a page if we're in an async thread because of REM notifications.
527	*/
528	pgmLock(pVM);
529	if (PHYS_PAGE_ADDRESS(GCPhys) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
530	{
531	/* Pool page changed while we were waiting for the lock; ignore. */
532	Log(("CPU%d: pgmR3PoolAccessHandler pgm pool page for %RGp changed (to %RGp) while waiting!\n", pVCpu->idCpu, PHYS_PAGE_ADDRESS(GCPhys), PHYS_PAGE_ADDRESS(pPage->GCPhys)));
533	pgmUnlock(pVM);
534	return VINF_PGM_HANDLER_DO_DEFAULT;
535	}
536
537	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
538
539	if (!pVCpu) /** @todo This shouldn't happen any longer, all access handlers will be called on an EMT. All ring-3 handlers, except MMIO, already own the PGM lock. @bugref{3170} */
540	{
541	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
542	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
543	STAM_COUNTER_INC(&pPool->StatMonitorR3Async);
544	if (!pPage->fReusedFlushPending)
545	{
546	pgmUnlock(pVM);
547	int rc = VMR3ReqCallVoidNoWait(pPool->pVMR3, VMCPUID_ANY, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
548	AssertRCReturn(rc, rc);
549	pgmLock(pVM);
550	pPage->fReusedFlushPending = true;
551	pPage->cModifications += 0x1000;
552	}
553
554	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
555	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
556	while (cbBuf > 4)
557	{
558	cbBuf -= 4;
559	pvPhys = (uint8_t *)pvPhys + 4;
560	GCPhys += 4;
561	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
562	}
563	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
564	}
565	else if ( ( pPage->cModifications < 96 /* it's cheaper here. */
566	\|\| pgmPoolIsPageLocked(&pVM->pgm.s, pPage)
567	)
568	&& cbBuf <= 4)
569	{
570	/* Clear the shadow entry. */
571	if (!pPage->cModifications++)
572	pgmPoolMonitorModifiedInsert(pPool, pPage);
573	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
574	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
575	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
576	}
577	else
578	{
579	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
580	STAM_PROFILE_STOP_EX(&pPool->StatMonitorR3, &pPool->StatMonitorR3FlushPage, a);
581	}
582	pgmUnlock(pVM);
583	return VINF_PGM_HANDLER_DO_DEFAULT;
584	}
585
586
587	/**
588	* Rendezvous callback used by pgmR3PoolClearAll that clears all shadow pages
589	* and all modification counters.
590	*
591	* This is only called on one of the EMTs while the other ones are waiting for
592	* it to complete this function.
593	*
594	* @returns VINF_SUCCESS (VBox strict status code).
595	* @param pVM The VM handle.
596	* @param pVCpu The VMCPU for the EMT we're being called on. Unused.
597	* @param pvUser Unused parameter.
598	*
599	*/
600	static DECLCALLBACK(VBOXSTRICTRC) pgmR3PoolClearAllRendezvous(PVM pVM, PVMCPU pVCpu, void *pvUser)
601	{
602	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
603	STAM_PROFILE_START(&pPool->StatClearAll, c);
604	LogFlow(("pgmPoolClearAllDoIt: cUsedPages=%d\n", pPool->cUsedPages));
605
606	pgmLock(pVM);
607
608	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
609	pgmPoolResetDirtyPages(pVM);
610	#endif
611
612	/*
613	* Iterate all the pages until we've encountered all that are in use.
614	* This is simple but not quite optimal solution.
615	*/
616	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
617	unsigned cLeft = pPool->cUsedPages;
618	unsigned iPage = pPool->cCurPages;
619	while (--iPage >= PGMPOOL_IDX_FIRST)
620	{
621	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
622	if (pPage->GCPhys != NIL_RTGCPHYS)
623	{
624	switch (pPage->enmKind)
625	{
626	/*
627	* We only care about shadow page tables.
628	*/
629	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
630	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
631	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
632	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
633	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
634	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
635	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
636	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
637	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
638	{
639	if (pPage->cPresent)
640	{
641	void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
642	STAM_PROFILE_START(&pPool->StatZeroPage, z);
643	ASMMemZeroPage(pvShw);
644	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
645	pPage->cPresent = 0;
646	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
647	}
648	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
649	else
650	Assert(!pPage->fDirty);
651	#endif
652	}
653	/* fall thru */
654
655	default:
656	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
657	Assert(!pPage->fDirty);
658	#endif
659	Assert(!pPage->cModifications \|\| ++cModifiedPages);
660	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
661	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
662	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
663	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
664	pPage->cModifications = 0;
665	break;
666
667	}
668	if (!--cLeft)
669	break;
670	}
671	}
672
673	/* swipe the special pages too. */
674	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
675	{
676	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
677	if (pPage->GCPhys != NIL_RTGCPHYS)
678	{
679	Assert(!pPage->cModifications \|\| ++cModifiedPages);
680	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
681	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
682	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
683	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
684	pPage->cModifications = 0;
685	}
686	}
687
688	#ifndef DEBUG_michael
689	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
690	#endif
691	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
692	pPool->cModifiedPages = 0;
693
694	/*
695	* Clear all the GCPhys links and rebuild the phys ext free list.
696	*/
697	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
698	pRam;
699	pRam = pRam->CTX_SUFF(pNext))
700	{
701	iPage = pRam->cb >> PAGE_SHIFT;
702	while (iPage-- > 0)
703	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
704	}
705
706	pPool->iPhysExtFreeHead = 0;
707	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
708	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
709	for (unsigned i = 0; i < cMaxPhysExts; i++)
710	{
711	paPhysExts[i].iNext = i + 1;
712	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
713	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
714	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
715	}
716	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
717
718	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
719	/* Clear all dirty pages. */
720	pPool->idxFreeDirtyPage = 0;
721	pPool->cDirtyPages = 0;
722	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aIdxDirtyPages); i++)
723	pPool->aIdxDirtyPages[i] = NIL_PGMPOOL_IDX;
724	#endif
725
726	/* Clear the PGM_SYNC_CLEAR_PGM_POOL flag on all VCPUs to prevent redundant flushes. */
727	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
728	pVM->aCpus[idCpu].pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
729
730	/* Flush job finished. */
731	VM_FF_CLEAR(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
732
733	pPool->cPresent = 0;
734	pgmUnlock(pVM);
735	PGM_INVL_ALL_VCPU_TLBS(pVM);
736	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
737	return VINF_SUCCESS;
738	}
739
740
741	/**
742	* Clears the shadow page pool.
743	*
744	* @param pVM The VM handle.
745	*/
746	void pgmR3PoolClearAll(PVM pVM)
747	{
748	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PoolClearAllRendezvous, NULL);
749	AssertRC(rc);
750	}
751
752
753	#ifdef VBOX_WITH_DEBUGGER
754	/**
755	* The '.pgmpoolcheck' command.
756	*
757	* @returns VBox status.
758	* @param pCmd Pointer to the command descriptor (as registered).
759	* @param pCmdHlp Pointer to command helper functions.
760	* @param pVM Pointer to the current VM (if any).
761	* @param paArgs Pointer to (readonly) array of arguments.
762	* @param cArgs Number of arguments in the array.
763	*/
764	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PVM pVM, PCDBGCVAR paArgs, unsigned cArgs, PDBGCVAR pResult)
765	{
766	/*
767	* Validate input.
768	*/
769	if (!pVM)
770	return pCmdHlp->pfnPrintf(pCmdHlp, NULL, "error: The command requires a VM to be selected.\n");
771
772	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
773
774	for (unsigned i = 0; i < pPool->cCurPages; i++)
775	{
776	PPGMPOOLPAGE pPage = &pPool->aPages[i];
777	bool fFirstMsg = true;
778
779	/* Todo: cover other paging modes too. */
780	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
781	{
782	PX86PTPAE pShwPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
783	PX86PTPAE pGstPT;
784	int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pGstPT); AssertReleaseRC(rc);
785
786	/* Check if any PTEs are out of sync. */
787	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
788	{
789	if (pShwPT->a[j].n.u1Present)
790	{
791	RTHCPHYS HCPhys = -1;
792	rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[j].u & X86_PTE_PAE_PG_MASK, &HCPhys);
793	if ( rc != VINF_SUCCESS
794	\|\| (pShwPT->a[j].u & X86_PTE_PAE_PG_MASK) != HCPhys)
795	{
796	if (fFirstMsg)
797	{
798	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Check pool page %RGp\n", pPage->GCPhys);
799	fFirstMsg = false;
800	}
801	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Mismatch HCPhys: rc=%d idx=%d guest %RX64 shw=%RX64 vs %RHp\n", rc, j, pGstPT->a[j].u, pShwPT->a[j].u, HCPhys);
802	}
803	else
804	if ( pShwPT->a[j].n.u1Write
805	&& !pGstPT->a[j].n.u1Write)
806	{
807	if (fFirstMsg)
808	{
809	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Check pool page %RGp\n", pPage->GCPhys);
810	fFirstMsg = false;
811	}
812	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Mismatch r/w gst/shw: idx=%d guest %RX64 shw=%RX64 vs %RHp\n", j, pGstPT->a[j].u, pShwPT->a[j].u, HCPhys);
813	}
814	}
815	}
816
817	/* Make sure this page table can't be written to from any shadow mapping. */
818	RTHCPHYS HCPhysPT = -1;
819	rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pPage->GCPhys, &HCPhysPT);
820	AssertMsgRC(rc, ("PGMPhysGCPhys2HCPhys failed with rc=%d for %RGp\n", rc, pPage->GCPhys));
821	if (rc == VINF_SUCCESS)
822	{
823	for (unsigned j = 0; j < pPool->cCurPages; j++)
824	{
825	PPGMPOOLPAGE pTempPage = &pPool->aPages[j];
826
827	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
828	{
829	PX86PTPAE pShwPT2 = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pTempPage);
830
831	for (unsigned k = 0; k < RT_ELEMENTS(pShwPT->a); k++)
832	{
833	if ( pShwPT2->a[k].n.u1Present
834	&& pShwPT2->a[k].n.u1Write
835	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
836	&& !pPage->fDirty
837	# endif
838	&& ((pShwPT2->a[k].u & X86_PTE_PAE_PG_MASK) == HCPhysPT))
839	{
840	if (fFirstMsg)
841	{
842	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Check pool page %RGp\n", pPage->GCPhys);
843	fFirstMsg = false;
844	}
845	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Mismatch: r/w: GCPhys=%RGp idx=%d shw %RX64 %RX64\n", pTempPage->GCPhys, k, pShwPT->a[k].u, pShwPT2->a[k].u);
846	}
847	}
848	}
849	}
850	}
851	}
852	}
853	return VINF_SUCCESS;
854	}
855	#endif /* VBOX_WITH_DEBUGGER */

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

source: vbox/trunk/src/VBox/VMM/PGMPool.cpp@ 26214

Download in other formats: