PGMPool.cpp@ 37324

Last change on this file since 37324 was 36960, checked in by vboxsync, 14 years ago
PGM,MM: Heap and shadow pool size adjustments for large memory configs.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 53.6 KB

Line
1	/* $Id: PGMPool.cpp 36960 2011-05-04 16:08:29Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Oracle Corporation
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*/
17
18	/** @page pg_pgm_pool PGM Shadow Page Pool
19	*
20	* Motivations:
21	* -# Relationship between shadow page tables and physical guest pages. This
22	* should allow us to skip most of the global flushes now following access
23	* handler changes. The main expense is flushing shadow pages.
24	* -# Limit the pool size if necessary (default is kind of limitless).
25	* -# Allocate shadow pages from RC. We use to only do this in SyncCR3.
26	* -# Required for 64-bit guests.
27	* -# Combining the PD cache and page pool in order to simplify caching.
28	*
29	*
30	* @section sec_pgm_pool_outline Design Outline
31	*
32	* The shadow page pool tracks pages used for shadowing paging structures (i.e.
33	* page tables, page directory, page directory pointer table and page map
34	* level-4). Each page in the pool has an unique identifier. This identifier is
35	* used to link a guest physical page to a shadow PT. The identifier is a
36	* non-zero value and has a relativly low max value - say 14 bits. This makes it
37	* possible to fit it into the upper bits of the of the aHCPhys entries in the
38	* ram range.
39	*
40	* By restricting host physical memory to the first 48 bits (which is the
41	* announced physical memory range of the K8L chip (scheduled for 2008)), we
42	* can safely use the upper 16 bits for shadow page ID and reference counting.
43	*
44	* Update: The 48 bit assumption will be lifted with the new physical memory
45	* management (PGMPAGE), so we won't have any trouble when someone stuffs 2TB
46	* into a box in some years.
47	*
48	* Now, it's possible for a page to be aliased, i.e. mapped by more than one PT
49	* or PD. This is solved by creating a list of physical cross reference extents
50	* when ever this happens. Each node in the list (extent) is can contain 3 page
51	* pool indexes. The list it self is chained using indexes into the paPhysExt
52	* array.
53	*
54	*
55	* @section sec_pgm_pool_life Life Cycle of a Shadow Page
56	*
57	* -# The SyncPT function requests a page from the pool.
58	* The request includes the kind of page it is (PT/PD, PAE/legacy), the
59	* address of the page it's shadowing, and more.
60	* -# The pool responds to the request by allocating a new page.
61	* When the cache is enabled, it will first check if it's in the cache.
62	* Should the pool be exhausted, one of two things can be done:
63	* -# Flush the whole pool and current CR3.
64	* -# Use the cache to find a page which can be flushed (~age).
65	* -# The SyncPT function will sync one or more pages and insert it into the
66	* shadow PD.
67	* -# The SyncPage function may sync more pages on a later \#PFs.
68	* -# The page is freed / flushed in SyncCR3 (perhaps) and some other cases.
69	* When caching is enabled, the page isn't flush but remains in the cache.
70	*
71	*
72	* @section sec_pgm_pool_impl Monitoring
73	*
74	* We always monitor PAGE_SIZE chunks of memory. When we've got multiple shadow
75	* pages for the same PAGE_SIZE of guest memory (PAE and mixed PD/PT) the pages
76	* sharing the monitor get linked using the iMonitoredNext/Prev. The head page
77	* is the pvUser to the access handlers.
78	*
79	*
80	* @section sec_pgm_pool_impl Implementation
81	*
82	* The pool will take pages from the MM page pool. The tracking data
83	* (attributes, bitmaps and so on) are allocated from the hypervisor heap. The
84	* pool content can be accessed both by using the page id and the physical
85	* address (HC). The former is managed by means of an array, the latter by an
86	* offset based AVL tree.
87	*
88	* Flushing of a pool page means that we iterate the content (we know what kind
89	* it is) and updates the link information in the ram range.
90	*
91	* ...
92	*/
93
94
95	/*******************************************************************************
96	* Header Files *
97	*******************************************************************************/
98	#define LOG_GROUP LOG_GROUP_PGM_POOL
99	#include <VBox/vmm/pgm.h>
100	#include <VBox/vmm/mm.h>
101	#include "PGMInternal.h"
102	#include <VBox/vmm/vm.h>
103	#include "PGMInline.h"
104
105	#include <VBox/log.h>
106	#include <VBox/err.h>
107	#include <iprt/asm.h>
108	#include <iprt/string.h>
109	#include <VBox/dbg.h>
110
111
112	/*******************************************************************************
113	* Internal Functions *
114	*******************************************************************************/
115	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser);
116	#ifdef VBOX_WITH_DEBUGGER
117	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PVM pVM, PCDBGCVAR paArgs, unsigned cArgs);
118	#endif
119
120	#ifdef VBOX_WITH_DEBUGGER
121	/** Command descriptors. */
122	static const DBGCCMD g_aCmds[] =
123	{
124	/* pszCmd, cArgsMin, cArgsMax, paArgDesc, cArgDescs, fFlags, pfnHandler pszSyntax, ....pszDescription */
125	{ "pgmpoolcheck", 0, 0, NULL, 0, 0, pgmR3PoolCmdCheck, "", "Check the pgm pool pages." },
126	};
127	#endif
128
129	/**
130	* Initializes the pool
131	*
132	* @returns VBox status code.
133	* @param pVM The VM handle.
134	*/
135	int pgmR3PoolInit(PVM pVM)
136	{
137	int rc;
138
139	AssertCompile(NIL_PGMPOOL_IDX == 0);
140	/* pPage->cLocked is an unsigned byte. */
141	AssertCompile(VMM_MAX_CPU_COUNT <= 255);
142
143	/*
144	* Query Pool config.
145	*/
146	PCFGMNODE pCfg = CFGMR3GetChild(CFGMR3GetRoot(pVM), "/PGM/Pool");
147
148	/* Default pgm pool size is 1024 pages (4MB). */
149	uint16_t cMaxPages = 1024;
150
151	/* Adjust it up relative to the RAM size, using the nested paging formula. */
152	uint64_t cbRam;
153	rc = CFGMR3QueryU64Def(CFGMR3GetRoot(pVM), "RamSize", &cbRam, 0); AssertRCReturn(rc, rc);
154	uint64_t u64MaxPages = (cbRam >> 9)
155	+ (cbRam >> 18)
156	+ (cbRam >> 27)
157	+ 32 * PAGE_SIZE;
158	u64MaxPages >>= PAGE_SHIFT;
159	if (u64MaxPages > PGMPOOL_IDX_LAST)
160	cMaxPages = PGMPOOL_IDX_LAST;
161	else
162	cMaxPages = (uint16_t)u64MaxPages;
163
164	/** @cfgm{/PGM/Pool/MaxPages, uint16_t, #pages, 16, 0x3fff, F(ram-size)}
165	* The max size of the shadow page pool in pages. The pool will grow dynamically
166	* up to this limit.
167	*/
168	rc = CFGMR3QueryU16Def(pCfg, "MaxPages", &cMaxPages, cMaxPages);
169	AssertLogRelRCReturn(rc, rc);
170	AssertLogRelMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
171	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
172	cMaxPages = RT_ALIGN(cMaxPages, 16);
173	if (cMaxPages > PGMPOOL_IDX_LAST)
174	cMaxPages = PGMPOOL_IDX_LAST;
175	LogRel(("PGMPool: cMaxPages=%u (u64MaxPages=%llu)\n", cMaxPages, u64MaxPages));
176
177	/** todo:
178	* We need to be much more careful with our allocation strategy here.
179	* For nested paging we don't need pool user info nor extents at all, but
180	* we can't check for nested paging here (too early during init to get a
181	* confirmation it can be used). The default for large memory configs is a
182	* bit large for shadow paging, so I've restricted the extent maximum to 8k
183	* (8k * 16 = 128k of hyper heap).
184	*
185	* Also when large page support is enabled, we typically don't need so much,
186	* although that depends on the availability of 2 MB chunks on the host.
187	*/
188
189	/** @cfgm{/PGM/Pool/MaxUsers, uint16_t, #users, MaxUsers, 32K, MaxPages*2}
190	* The max number of shadow page user tracking records. Each shadow page has
191	* zero of other shadow pages (or CR3s) that references it, or uses it if you
192	* like. The structures describing these relationships are allocated from a
193	* fixed sized pool. This configuration variable defines the pool size.
194	*/
195	uint16_t cMaxUsers;
196	rc = CFGMR3QueryU16Def(pCfg, "MaxUsers", &cMaxUsers, cMaxPages * 2);
197	AssertLogRelRCReturn(rc, rc);
198	AssertLogRelMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
199	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
200
201	/** @cfgm{/PGM/Pool/MaxPhysExts, uint16_t, #extents, 16, MaxPages * 2, MIN(MaxPages*2,8192)}
202	* The max number of extents for tracking aliased guest pages.
203	*/
204	uint16_t cMaxPhysExts;
205	rc = CFGMR3QueryU16Def(pCfg, "MaxPhysExts", &cMaxPhysExts,
206	RT_MIN(cMaxPages * 2, 8192 /* 8Ki max as this eat too much hyper heap */));
207	AssertLogRelRCReturn(rc, rc);
208	AssertLogRelMsgReturn(cMaxPhysExts >= 16 && cMaxPhysExts <= PGMPOOL_IDX_LAST,
209	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxPhysExts), VERR_INVALID_PARAMETER);
210
211	/** @cfgm{/PGM/Pool/ChacheEnabled, bool, true}
212	* Enables or disabling caching of shadow pages. Caching means that we will try
213	* reuse shadow pages instead of recreating them everything SyncCR3, SyncPT or
214	* SyncPage requests one. When reusing a shadow page, we can save time
215	* reconstructing it and it's children.
216	*/
217	bool fCacheEnabled;
218	rc = CFGMR3QueryBoolDef(pCfg, "CacheEnabled", &fCacheEnabled, true);
219	AssertLogRelRCReturn(rc, rc);
220
221	LogRel(("pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
222	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
223
224	/*
225	* Allocate the data structures.
226	*/
227	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
228	cb += cMaxUsers * sizeof(PGMPOOLUSER);
229	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
230	PPGMPOOL pPool;
231	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
232	if (RT_FAILURE(rc))
233	return rc;
234	pVM->pgm.s.pPoolR3 = pPool;
235	pVM->pgm.s.pPoolR0 = MMHyperR3ToR0(pVM, pPool);
236	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pPool);
237
238	/*
239	* Initialize it.
240	*/
241	pPool->pVMR3 = pVM;
242	pPool->pVMR0 = pVM->pVMR0;
243	pPool->pVMRC = pVM->pVMRC;
244	pPool->cMaxPages = cMaxPages;
245	pPool->cCurPages = PGMPOOL_IDX_FIRST;
246	pPool->iUserFreeHead = 0;
247	pPool->cMaxUsers = cMaxUsers;
248	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
249	pPool->paUsersR3 = paUsers;
250	pPool->paUsersR0 = MMHyperR3ToR0(pVM, paUsers);
251	pPool->paUsersRC = MMHyperR3ToRC(pVM, paUsers);
252	for (unsigned i = 0; i < cMaxUsers; i++)
253	{
254	paUsers[i].iNext = i + 1;
255	paUsers[i].iUser = NIL_PGMPOOL_IDX;
256	paUsers[i].iUserTable = 0xfffffffe;
257	}
258	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
259	pPool->iPhysExtFreeHead = 0;
260	pPool->cMaxPhysExts = cMaxPhysExts;
261	PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
262	pPool->paPhysExtsR3 = paPhysExts;
263	pPool->paPhysExtsR0 = MMHyperR3ToR0(pVM, paPhysExts);
264	pPool->paPhysExtsRC = MMHyperR3ToRC(pVM, paPhysExts);
265	for (unsigned i = 0; i < cMaxPhysExts; i++)
266	{
267	paPhysExts[i].iNext = i + 1;
268	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
269	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
270	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
271	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
272	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
273	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
274	}
275	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
276	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
277	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
278	pPool->iAgeHead = NIL_PGMPOOL_IDX;
279	pPool->iAgeTail = NIL_PGMPOOL_IDX;
280	pPool->fCacheEnabled = fCacheEnabled;
281	pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
282	pPool->pszAccessHandler = "Guest Paging Access Handler";
283	pPool->HCPhysTree = 0;
284
285	/* The NIL entry. */
286	Assert(NIL_PGMPOOL_IDX == 0);
287	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
288
289	/* The Shadow 32-bit PD. (32 bits guest paging) */
290	pPool->aPages[PGMPOOL_IDX_PD].Core.Key = NIL_RTHCPHYS;
291	pPool->aPages[PGMPOOL_IDX_PD].GCPhys = NIL_RTGCPHYS;
292	pPool->aPages[PGMPOOL_IDX_PD].pvPageR3 = 0;
293	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_32BIT_PD;
294	pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
295
296	/* The Shadow PDPT. */
297	pPool->aPages[PGMPOOL_IDX_PDPT].Core.Key = NIL_RTHCPHYS;
298	pPool->aPages[PGMPOOL_IDX_PDPT].GCPhys = NIL_RTGCPHYS;
299	pPool->aPages[PGMPOOL_IDX_PDPT].pvPageR3 = 0;
300	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_PAE_PDPT;
301	pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
302
303	/* The Shadow AMD64 CR3. */
304	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].Core.Key = NIL_RTHCPHYS;
305	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].GCPhys = NIL_RTGCPHYS;
306	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].pvPageR3 = 0;
307	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].enmKind = PGMPOOLKIND_64BIT_PML4;
308	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].idx = PGMPOOL_IDX_AMD64_CR3;
309
310	/* The Nested Paging CR3. */
311	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].Core.Key = NIL_RTHCPHYS;
312	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].GCPhys = NIL_RTGCPHYS;
313	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].pvPageR3 = 0;
314	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].enmKind = PGMPOOLKIND_ROOT_NESTED;
315	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].idx = PGMPOOL_IDX_NESTED_ROOT;
316
317	/*
318	* Set common stuff.
319	*/
320	for (unsigned iPage = 1; iPage < PGMPOOL_IDX_FIRST; iPage++)
321	{
322	pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
323	pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
324	pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
325	pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
326	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
327	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
328	pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
329	pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
330	Assert(pPool->aPages[iPage].idx == iPage);
331	Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
332	Assert(!pPool->aPages[iPage].fSeenNonGlobal);
333	Assert(!pPool->aPages[iPage].fMonitored);
334	Assert(!pPool->aPages[iPage].fCached);
335	Assert(!pPool->aPages[iPage].fZeroed);
336	Assert(!pPool->aPages[iPage].fReusedFlushPending);
337	}
338
339	#ifdef VBOX_WITH_STATISTICS
340	/*
341	* Register statistics.
342	*/
343	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
344	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
345	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
346	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
347	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
348	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolClearAll.");
349	STAM_REG(pVM, &pPool->StatR3Reset, STAMTYPE_PROFILE, "/PGM/Pool/R3Reset", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolReset.");
350	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
351	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
352	STAM_REG(pVM, &pPool->StatForceFlushPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForce", STAMUNIT_OCCURENCES, "Counting explicit flushes by PGMPoolFlushPage().");
353	STAM_REG(pVM, &pPool->StatForceFlushDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForceDirty", STAMUNIT_OCCURENCES, "Counting explicit flushes of dirty pages by PGMPoolFlushPage().");
354	STAM_REG(pVM, &pPool->StatForceFlushReused, STAMTYPE_COUNTER, "/PGM/Pool/FlushReused", STAMUNIT_OCCURENCES, "Counting flushes for reused pages.");
355	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spent zeroing pages. Overlaps with Alloc.");
356	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
357	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
358	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackDeref.");
359	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPT.");
360	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
361	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
362	STAM_REG(pVM, &pPool->StatTrackFlushEntry, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Flush", STAMUNIT_COUNT, "Nr of flushed entries.");
363	STAM_REG(pVM, &pPool->StatTrackFlushEntryKeep, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Update", STAMUNIT_COUNT, "Nr of updated entries.");
364	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.");
365	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_TICKS_PER_CALL, "Profiling deref activity related tracking GC physical pages.");
366	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
367	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
368	STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
369	STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
370	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.");
371	STAM_REG(pVM, &pPool->StatMonitorRZFlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
372	STAM_REG(pVM, &pPool->StatMonitorRZFlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
373	STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
374	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).");
375	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
376	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.");
377	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.");
378	STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
379	STAM_REG(pVM, &pPool->StatMonitorRZFaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
380	STAM_REG(pVM, &pPool->StatMonitorRZFaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
381	STAM_REG(pVM, &pPool->StatMonitorRZFaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
382	STAM_REG(pVM, &pPool->StatMonitorRZFaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
383	STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
384	STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
385	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.");
386	STAM_REG(pVM, &pPool->StatMonitorR3FlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
387	STAM_REG(pVM, &pPool->StatMonitorR3FlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
388	STAM_REG(pVM, &pPool->StatMonitorR3Fork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
389	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).");
390	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.");
391	STAM_REG(pVM, &pPool->StatMonitorR3RepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
392	STAM_REG(pVM, &pPool->StatMonitorR3FaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
393	STAM_REG(pVM, &pPool->StatMonitorR3FaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
394	STAM_REG(pVM, &pPool->StatMonitorR3FaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
395	STAM_REG(pVM, &pPool->StatMonitorR3FaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
396	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.");
397	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
398	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
399	STAM_REG(pVM, &pPool->StatResetDirtyPages, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Resets", STAMUNIT_OCCURENCES, "Times we've called pgmPoolResetDirtyPages (and there were dirty page).");
400	STAM_REG(pVM, &pPool->StatDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Pages", STAMUNIT_OCCURENCES, "Times we've called pgmPoolAddDirtyPage.");
401	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.");
402	STAM_REG(pVM, &pPool->StatDirtyPageOverFlowFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushOverflow",STAMUNIT_OCCURENCES, "Times we've had to flush because of overflow.");
403	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
404	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
405	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!)");
406	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.");
407	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
408	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
409	#endif /* VBOX_WITH_STATISTICS */
410
411	#ifdef VBOX_WITH_DEBUGGER
412	/*
413	* Debugger commands.
414	*/
415	static bool s_fRegisteredCmds = false;
416	if (!s_fRegisteredCmds)
417	{
418	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
419	if (RT_SUCCESS(rc))
420	s_fRegisteredCmds = true;
421	}
422	#endif
423
424	return VINF_SUCCESS;
425	}
426
427
428	/**
429	* Relocate the page pool data.
430	*
431	* @param pVM The VM handle.
432	*/
433	void pgmR3PoolRelocate(PVM pVM)
434	{
435	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
436	pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
437	pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
438	pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
439	int rc = PDMR3LdrGetSymbolRC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerRC);
440	AssertReleaseRC(rc);
441	/* init order hack. */
442	if (!pVM->pgm.s.pPoolR3->pfnAccessHandlerR0)
443	{
444	rc = PDMR3LdrGetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerR0);
445	AssertReleaseRC(rc);
446	}
447	}
448
449
450	/**
451	* Grows the shadow page pool.
452	*
453	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
454	*
455	* @returns VBox status code.
456	* @param pVM The VM handle.
457	*/
458	VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
459	{
460	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
461	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_INTERNAL_ERROR);
462
463	pgmLock(pVM);
464
465	/*
466	* How much to grow it by?
467	*/
468	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
469	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
470	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages.\n", cPages, cPages));
471
472	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
473	{
474	PPGMPOOLPAGE pPage = &pPool->aPages[i];
475
476	/* Allocate all pages in low (below 4 GB) memory as 32 bits guests need a page table root in low memory. */
477	pPage->pvPageR3 = MMR3PageAllocLow(pVM);
478	if (!pPage->pvPageR3)
479	{
480	Log(("We're out of memory!! i=%d\n", i));
481	pgmUnlock(pVM);
482	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
483	}
484	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
485	AssertFatal(pPage->Core.Key < _4G);
486	pPage->GCPhys = NIL_RTGCPHYS;
487	pPage->enmKind = PGMPOOLKIND_FREE;
488	pPage->idx = pPage - &pPool->aPages[0];
489	LogFlow(("PGMR3PoolGrow: insert page #%#x - %RHp\n", pPage->idx, pPage->Core.Key));
490	pPage->iNext = pPool->iFreeHead;
491	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
492	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
493	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
494	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
495	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
496	pPage->iAgeNext = NIL_PGMPOOL_IDX;
497	pPage->iAgePrev = NIL_PGMPOOL_IDX;
498	/* commit it */
499	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
500	pPool->iFreeHead = i;
501	pPool->cCurPages = i + 1;
502	}
503
504	pgmUnlock(pVM);
505	Assert(pPool->cCurPages <= pPool->cMaxPages);
506	return VINF_SUCCESS;
507	}
508
509
510
511	/**
512	* Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
513	*
514	* @param pPool The pool.
515	* @param pPage The page.
516	*/
517	static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
518	{
519	/* for the present this should be safe enough I think... */
520	pgmLock(pPool->pVMR3);
521	if ( pPage->fReusedFlushPending
522	&& pPage->enmKind != PGMPOOLKIND_FREE)
523	pgmPoolFlushPage(pPool, pPage);
524	pgmUnlock(pPool->pVMR3);
525	}
526
527
528	/**
529	* \#PF Handler callback for PT write accesses.
530	*
531	* The handler can not raise any faults, it's mainly for monitoring write access
532	* to certain pages.
533	*
534	* @returns VINF_SUCCESS if the handler has carried out the operation.
535	* @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
536	* @param pVM VM Handle.
537	* @param GCPhys The physical address the guest is writing to.
538	* @param pvPhys The HC mapping of that address.
539	* @param pvBuf What the guest is reading/writing.
540	* @param cbBuf How much it's reading/writing.
541	* @param enmAccessType The access type.
542	* @param pvUser User argument.
543	*/
544	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser)
545	{
546	STAM_PROFILE_START(&pVM->pgm.s.pPoolR3->StatMonitorR3, a);
547	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
548	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
549	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%RGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
550	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
551
552	PVMCPU pVCpu = VMMGetCpu(pVM);
553
554	/*
555	* We don't have to be very sophisticated about this since there are relativly few calls here.
556	* However, we must try our best to detect any non-cpu accesses (disk / networking).
557	*
558	* Just to make life more interesting, we'll have to deal with the async threads too.
559	* We cannot flush a page if we're in an async thread because of REM notifications.
560	*/
561	pgmLock(pVM);
562	if (PHYS_PAGE_ADDRESS(GCPhys) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
563	{
564	/* Pool page changed while we were waiting for the lock; ignore. */
565	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)));
566	pgmUnlock(pVM);
567	return VINF_PGM_HANDLER_DO_DEFAULT;
568	}
569
570	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
571
572	/* @todo this code doesn't make any sense. remove the if (!pVCpu) block */
573	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} */
574	{
575	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
576	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
577	STAM_COUNTER_INC(&pPool->StatMonitorR3Async);
578	if (!pPage->fReusedFlushPending)
579	{
580	pgmUnlock(pVM);
581	int rc = VMR3ReqCallVoidNoWait(pPool->pVMR3, VMCPUID_ANY, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
582	AssertRCReturn(rc, rc);
583	pgmLock(pVM);
584	pPage->fReusedFlushPending = true;
585	pPage->cModifications += 0x1000;
586	}
587
588	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
589	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
590	while (cbBuf > 4)
591	{
592	cbBuf -= 4;
593	pvPhys = (uint8_t *)pvPhys + 4;
594	GCPhys += 4;
595	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
596	}
597	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
598	}
599	else if ( ( pPage->cModifications < 96 /* it's cheaper here. */
600	\|\| pgmPoolIsPageLocked(pPage)
601	)
602	&& cbBuf <= 4)
603	{
604	/* Clear the shadow entry. */
605	if (!pPage->cModifications++)
606	pgmPoolMonitorModifiedInsert(pPool, pPage);
607	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
608	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
609	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
610	}
611	else
612	{
613	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
614	STAM_PROFILE_STOP_EX(&pPool->StatMonitorR3, &pPool->StatMonitorR3FlushPage, a);
615	}
616	pgmUnlock(pVM);
617	return VINF_PGM_HANDLER_DO_DEFAULT;
618	}
619
620
621	/**
622	* Rendezvous callback used by pgmR3PoolClearAll that clears all shadow pages
623	* and all modification counters.
624	*
625	* This is only called on one of the EMTs while the other ones are waiting for
626	* it to complete this function.
627	*
628	* @returns VINF_SUCCESS (VBox strict status code).
629	* @param pVM The VM handle.
630	* @param pVCpu The VMCPU for the EMT we're being called on. Unused.
631	* @param fpvFlushRemTlb When not NULL, we'll flush the REM TLB as well.
632	* (This is the pvUser, so it has to be void *.)
633	*
634	*/
635	DECLCALLBACK(VBOXSTRICTRC) pgmR3PoolClearAllRendezvous(PVM pVM, PVMCPU pVCpu, void *fpvFlushRemTbl)
636	{
637	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
638	STAM_PROFILE_START(&pPool->StatClearAll, c);
639
640	pgmLock(pVM);
641	Log(("pgmR3PoolClearAllRendezvous: cUsedPages=%d fpvFlushRemTbl=%RTbool\n", pPool->cUsedPages, !!fpvFlushRemTbl));
642
643	/*
644	* Iterate all the pages until we've encountered all that are in use.
645	* This is a simple but not quite optimal solution.
646	*/
647	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
648	unsigned cLeft = pPool->cUsedPages;
649	uint32_t iPage = pPool->cCurPages;
650	while (--iPage >= PGMPOOL_IDX_FIRST)
651	{
652	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
653	if (pPage->GCPhys != NIL_RTGCPHYS)
654	{
655	switch (pPage->enmKind)
656	{
657	/*
658	* We only care about shadow page tables that reference physical memory
659	*/
660	#ifdef PGM_WITH_LARGE_PAGES
661	case PGMPOOLKIND_EPT_PD_FOR_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
662	if (pPage->cPresent)
663	{
664	PX86PDPAE pShwPD = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
665	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
666	{
667	if ( pShwPD->a[i].n.u1Present
668	&& pShwPD->a[i].b.u1Size)
669	{
670	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
671	pShwPD->a[i].u = 0;
672	Assert(pPage->cPresent);
673	pPage->cPresent--;
674	}
675	}
676	if (pPage->cPresent == 0)
677	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
678	}
679	goto default_case;
680
681	case PGMPOOLKIND_PAE_PD_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
682	if (pPage->cPresent)
683	{
684	PEPTPD pShwPD = (PEPTPD)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
685	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
686	{
687	Assert((pShwPD->a[i].u & UINT64_C(0xfff0000000000f80)) == 0);
688	if ( pShwPD->a[i].n.u1Present
689	&& pShwPD->a[i].b.u1Size)
690	{
691	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
692	pShwPD->a[i].u = 0;
693	Assert(pPage->cPresent);
694	pPage->cPresent--;
695	}
696	}
697	if (pPage->cPresent == 0)
698	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
699	}
700	goto default_case;
701	#endif /* PGM_WITH_LARGE_PAGES */
702
703	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
704	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
705	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
706	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
707	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
708	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
709	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
710	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
711	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
712	{
713	if (pPage->cPresent)
714	{
715	void *pvShw = PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
716	STAM_PROFILE_START(&pPool->StatZeroPage, z);
717	#if 0
718	/* Useful check for leaking references; very expensive though. */
719	switch (pPage->enmKind)
720	{
721	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
722	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
723	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
724	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
725	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
726	{
727	bool fFoundFirst = false;
728	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)pvShw;
729	for (unsigned ptIndex = 0; ptIndex < RT_ELEMENTS(pPT->a); ptIndex++)
730	{
731	if (pPT->a[ptIndex].u)
732	{
733	if (!fFoundFirst)
734	{
735	AssertFatalMsg(pPage->iFirstPresent <= ptIndex, ("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
736	if (pPage->iFirstPresent != ptIndex)
737	Log(("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
738	fFoundFirst = true;
739	}
740	if (PGMSHWPTEPAE_IS_P(pPT->a[ptIndex]))
741	{
742	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pPT->a[ptIndex]), NIL_RTGCPHYS);
743	if (pPage->iFirstPresent == ptIndex)
744	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
745	}
746	}
747	}
748	AssertFatalMsg(pPage->cPresent == 0, ("cPresent = %d pPage = %RGv\n", pPage->cPresent, pPage->GCPhys));
749	break;
750	}
751	default:
752	break;
753	}
754	#endif
755	ASMMemZeroPage(pvShw);
756	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
757	pPage->cPresent = 0;
758	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
759	}
760	}
761	/* fall thru */
762
763	#ifdef PGM_WITH_LARGE_PAGES
764	default_case:
765	#endif
766	default:
767	Assert(!pPage->cModifications \|\| ++cModifiedPages);
768	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
769	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
770	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
771	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
772	pPage->cModifications = 0;
773	break;
774
775	}
776	if (!--cLeft)
777	break;
778	}
779	}
780
781	/* swipe the special pages too. */
782	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
783	{
784	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
785	if (pPage->GCPhys != NIL_RTGCPHYS)
786	{
787	Assert(!pPage->cModifications \|\| ++cModifiedPages);
788	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
789	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
790	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
791	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
792	pPage->cModifications = 0;
793	}
794	}
795
796	#ifndef DEBUG_michael
797	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
798	#endif
799	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
800	pPool->cModifiedPages = 0;
801
802	/*
803	* Clear all the GCPhys links and rebuild the phys ext free list.
804	*/
805	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRangesX);
806	pRam;
807	pRam = pRam->CTX_SUFF(pNext))
808	{
809	iPage = pRam->cb >> PAGE_SHIFT;
810	while (iPage-- > 0)
811	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
812	}
813
814	pPool->iPhysExtFreeHead = 0;
815	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
816	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
817	for (unsigned i = 0; i < cMaxPhysExts; i++)
818	{
819	paPhysExts[i].iNext = i + 1;
820	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
821	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
822	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
823	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
824	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
825	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
826	}
827	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
828
829
830	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
831	/* Reset all dirty pages to reactivate the page monitoring. */
832	/* Note: we must do this after clearing all page references and shadow page tables as there might be stale references to
833	* recently removed MMIO ranges around that might otherwise end up asserting in pgmPoolTracDerefGCPhysHint
834	*/
835	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
836	{
837	PPGMPOOLPAGE pPage;
838	unsigned idxPage;
839
840	if (pPool->aDirtyPages[i].uIdx == NIL_PGMPOOL_IDX)
841	continue;
842
843	idxPage = pPool->aDirtyPages[i].uIdx;
844	AssertRelease(idxPage != NIL_PGMPOOL_IDX);
845	pPage = &pPool->aPages[idxPage];
846	Assert(pPage->idx == idxPage);
847	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
848
849	AssertMsg(pPage->fDirty, ("Page %RGp (slot=%d) not marked dirty!", pPage->GCPhys, i));
850
851	Log(("Reactivate dirty page %RGp\n", pPage->GCPhys));
852
853	/* First write protect the page again to catch all write accesses. (before checking for changes -> SMP) */
854	int rc = PGMHandlerPhysicalReset(pVM, pPage->GCPhys & PAGE_BASE_GC_MASK);
855	Assert(rc == VINF_SUCCESS);
856	pPage->fDirty = false;
857
858	pPool->aDirtyPages[i].uIdx = NIL_PGMPOOL_IDX;
859	}
860
861	/* Clear all dirty pages. */
862	pPool->idxFreeDirtyPage = 0;
863	pPool->cDirtyPages = 0;
864	#endif
865
866	/* Clear the PGM_SYNC_CLEAR_PGM_POOL flag on all VCPUs to prevent redundant flushes. */
867	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
868	pVM->aCpus[idCpu].pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
869
870	/* Flush job finished. */
871	VM_FF_CLEAR(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
872	pPool->cPresent = 0;
873	pgmUnlock(pVM);
874
875	PGM_INVL_ALL_VCPU_TLBS(pVM);
876
877	if (fpvFlushRemTbl)
878	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
879	CPUMSetChangedFlags(&pVM->aCpus[idCpu], CPUM_CHANGED_GLOBAL_TLB_FLUSH);
880
881	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
882	return VINF_SUCCESS;
883	}
884
885
886	/**
887	* Clears the shadow page pool.
888	*
889	* @param pVM The VM handle.
890	* @param fFlushRemTlb When set, the REM TLB is scheduled for flushing as
891	* well.
892	*/
893	void pgmR3PoolClearAll(PVM pVM, bool fFlushRemTlb)
894	{
895	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PoolClearAllRendezvous, &fFlushRemTlb);
896	AssertRC(rc);
897	}
898
899	/**
900	* Protect all pgm pool page table entries to monitor writes
901	*
902	* @param pVM The VM handle.
903	*
904	* Remark: assumes the caller will flush all TLBs (!!)
905	*/
906	void pgmR3PoolWriteProtectPages(PVM pVM)
907	{
908	Assert(PGMIsLockOwner(pVM));
909	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
910	unsigned cLeft = pPool->cUsedPages;
911	unsigned iPage = pPool->cCurPages;
912	while (--iPage >= PGMPOOL_IDX_FIRST)
913	{
914	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
915	if ( pPage->GCPhys != NIL_RTGCPHYS
916	&& pPage->cPresent)
917	{
918	union
919	{
920	void *pv;
921	PX86PT pPT;
922	PPGMSHWPTPAE pPTPae;
923	PEPTPT pPTEpt;
924	} uShw;
925	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
926
927	switch (pPage->enmKind)
928	{
929	/*
930	* We only care about shadow page tables.
931	*/
932	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
933	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
934	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
935	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPT->a); iShw++)
936	{
937	if (uShw.pPT->a[iShw].n.u1Present)
938	uShw.pPT->a[iShw].n.u1Write = 0;
939	}
940	break;
941
942	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
943	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
944	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
945	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
946	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
947	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTPae->a); iShw++)
948	{
949	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
950	PGMSHWPTEPAE_SET_RO(uShw.pPTPae->a[iShw]);
951	}
952	break;
953
954	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
955	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTEpt->a); iShw++)
956	{
957	if (uShw.pPTEpt->a[iShw].n.u1Present)
958	uShw.pPTEpt->a[iShw].n.u1Write = 0;
959	}
960	break;
961
962	default:
963	break;
964	}
965	if (!--cLeft)
966	break;
967	}
968	}
969	}
970
971	#ifdef VBOX_WITH_DEBUGGER
972	/**
973	* The '.pgmpoolcheck' command.
974	*
975	* @returns VBox status.
976	* @param pCmd Pointer to the command descriptor (as registered).
977	* @param pCmdHlp Pointer to command helper functions.
978	* @param pVM Pointer to the current VM (if any).
979	* @param paArgs Pointer to (readonly) array of arguments.
980	* @param cArgs Number of arguments in the array.
981	*/
982	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PVM pVM, PCDBGCVAR paArgs, unsigned cArgs)
983	{
984	DBGC_CMDHLP_REQ_VM_RET(pCmdHlp, pCmd, pVM);
985	DBGC_CMDHLP_ASSERT_PARSER_RET(pCmdHlp, pCmd, -1, cArgs == 0);
986	uint32_t cErrors = 0;
987
988	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
989	for (unsigned i = 0; i < pPool->cCurPages; i++)
990	{
991	PPGMPOOLPAGE pPage = &pPool->aPages[i];
992	bool fFirstMsg = true;
993
994	/* Todo: cover other paging modes too. */
995	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
996	{
997	PPGMSHWPTPAE pShwPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
998	{
999	PX86PTPAE pGstPT;
1000	PGMPAGEMAPLOCK LockPage;
1001	int rc = PGMPhysGCPhys2CCPtrReadOnly(pVM, pPage->GCPhys, (const void **)&pGstPT, &LockPage); AssertReleaseRC(rc);
1002
1003	/* Check if any PTEs are out of sync. */
1004	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
1005	{
1006	if (PGMSHWPTEPAE_IS_P(pShwPT->a[j]))
1007	{
1008	RTHCPHYS HCPhys = NIL_RTHCPHYS;
1009	rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[j].u & X86_PTE_PAE_PG_MASK, &HCPhys);
1010	if ( rc != VINF_SUCCESS
1011	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[j]) != HCPhys)
1012	{
1013	if (fFirstMsg)
1014	{
1015	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
1016	fFirstMsg = false;
1017	}
1018	DBGCCmdHlpPrintf(pCmdHlp, "Mismatch HCPhys: rc=%Rrc idx=%d guest %RX64 shw=%RX64 vs %RHp\n", rc, j, pGstPT->a[j].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[j]), HCPhys);
1019	cErrors++;
1020	}
1021	else if ( PGMSHWPTEPAE_IS_RW(pShwPT->a[j])
1022	&& !pGstPT->a[j].n.u1Write)
1023	{
1024	if (fFirstMsg)
1025	{
1026	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
1027	fFirstMsg = false;
1028	}
1029	DBGCCmdHlpPrintf(pCmdHlp, "Mismatch r/w gst/shw: idx=%d guest %RX64 shw=%RX64 vs %RHp\n", j, pGstPT->a[j].u, PGMSHWPTEPAE_GET_LOG(pShwPT->a[j]), HCPhys);
1030	cErrors++;
1031	}
1032	}
1033	}
1034	PGMPhysReleasePageMappingLock(pVM, &LockPage);
1035	}
1036
1037	/* Make sure this page table can't be written to from any shadow mapping. */
1038	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
1039	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pPage->GCPhys, &HCPhysPT);
1040	AssertMsgRC(rc, ("PGMPhysGCPhys2HCPhys failed with rc=%d for %RGp\n", rc, pPage->GCPhys));
1041	if (rc == VINF_SUCCESS)
1042	{
1043	for (unsigned j = 0; j < pPool->cCurPages; j++)
1044	{
1045	PPGMPOOLPAGE pTempPage = &pPool->aPages[j];
1046
1047	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1048	{
1049	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pTempPage);
1050
1051	for (unsigned k = 0; k < RT_ELEMENTS(pShwPT->a); k++)
1052	{
1053	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[k])
1054	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1055	&& !pPage->fDirty
1056	# endif
1057	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[k]) == HCPhysPT)
1058	{
1059	if (fFirstMsg)
1060	{
1061	DBGCCmdHlpPrintf(pCmdHlp, "Check pool page %RGp\n", pPage->GCPhys);
1062	fFirstMsg = false;
1063	}
1064	DBGCCmdHlpPrintf(pCmdHlp, "Mismatch: r/w: GCPhys=%RGp idx=%d shw %RX64 %RX64\n", pTempPage->GCPhys, k, PGMSHWPTEPAE_GET_LOG(pShwPT->a[k]), PGMSHWPTEPAE_GET_LOG(pShwPT2->a[k]));
1065	cErrors++;
1066	}
1067	}
1068	}
1069	}
1070	}
1071	}
1072	}
1073	if (cErrors > 0)
1074	return DBGCCmdHlpFail(pCmdHlp, pCmd, "Found %#x errors", cErrors);
1075	return VINF_SUCCESS;
1076	}
1077	#endif /* VBOX_WITH_DEBUGGER */

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

source: vbox/trunk/src/VBox/VMM/VMMR3/PGMPool.cpp@ 37324

Download in other formats: