PGMPool.cpp@ 34347

Last change on this file since 34347 was 33676, checked in by vboxsync, 14 years ago
scm cleanup run.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 53.5 KB

Line
1	/* $Id: PGMPool.cpp 33676 2010-11-02 09:48:24Z 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/pgm.h>
100	#include <VBox/mm.h>
101	#include "PGMInternal.h"
102	#include <VBox/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, PDBGCVAR pResult);
118	#endif
119
120	#ifdef VBOX_WITH_DEBUGGER
121	/** Command descriptors. */
122	static const DBGCCMD g_aCmds[] =
123	{
124	/* pszCmd, cArgsMin, cArgsMax, paArgDesc, cArgDescs, pResultDesc, fFlags, pfnHandler pszSyntax, ....pszDescription */
125	{ "pgmpoolcheck", 0, 0, NULL, 0, NULL, 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	AssertCompile(NIL_PGMPOOL_IDX == 0);
138	/* pPage->cLocked is an unsigned byte. */
139	AssertCompile(VMM_MAX_CPU_COUNT <= 255);
140
141	/*
142	* Query Pool config.
143	*/
144	PCFGMNODE pCfg = CFGMR3GetChild(CFGMR3GetRoot(pVM), "/PGM/Pool");
145
146	/* Default pgm pool size equals 1024 pages. */
147	uint16_t cMaxPages = 4*_1M >> PAGE_SHIFT;
148
149	#if HC_ARCH_BITS == 64
150	uint64_t cbRam = 0;
151	CFGMR3QueryU64Def(CFGMR3GetRoot(pVM), "RamSize", &cbRam, 0);
152
153	/* We should increase the pgm pool size for guests with more than 2 GB of ram */
154	if (cbRam >= UINT64_C(2) * _1G)
155	{
156	/* In the nested paging case we require 2 + 513 * (cbRam/1GB) pages to
157	* store the entire page table descriptors.
158	*/
159	uint64_t u64MaxPages = cbRam / (_1G / UINT64_C(512));
160	if (u64MaxPages > PGMPOOL_IDX_LAST)
161	cMaxPages = PGMPOOL_IDX_LAST;
162	else
163	cMaxPages = (uint16_t)u64MaxPages;
164	}
165	#endif
166
167	/** @cfgm{/PGM/Pool/MaxPages, uint16_t, #pages, 16, 0x3fff, 1024}
168	* The max size of the shadow page pool in pages. The pool will grow dynamically
169	* up to this limit.
170	*/
171	int rc = CFGMR3QueryU16Def(pCfg, "MaxPages", &cMaxPages, cMaxPages);
172	AssertLogRelRCReturn(rc, rc);
173	AssertLogRelMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
174	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
175	cMaxPages = RT_ALIGN(cMaxPages, 16);
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 we can't check for nested paging here (too early during init to get a confirmation it can be used)
180	* The default for large memory configs is a bit large for shadow paging, so I've restricted the extent maximum to 2k (2k * 16 = 32k of hyper heap)
181	*
182	* Also when large page support is enabled, we typically don't need so much,
183	* although that depends on the availability of 2 MB chunks on the host.
184	*/
185
186	/** @cfgm{/PGM/Pool/MaxUsers, uint16_t, #users, MaxUsers, 32K, MaxPages*2}
187	* The max number of shadow page user tracking records. Each shadow page has
188	* zero of other shadow pages (or CR3s) that references it, or uses it if you
189	* like. The structures describing these relationships are allocated from a
190	* fixed sized pool. This configuration variable defines the pool size.
191	*/
192	uint16_t cMaxUsers;
193	rc = CFGMR3QueryU16Def(pCfg, "MaxUsers", &cMaxUsers, cMaxPages * 2);
194	AssertLogRelRCReturn(rc, rc);
195	AssertLogRelMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
196	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
197
198	/** @cfgm{/PGM/Pool/MaxPhysExts, uint16_t, #extents, 16, MaxPages * 2, MAX(MaxPages*2,0x3fff)}
199	* The max number of extents for tracking aliased guest pages.
200	*/
201	uint16_t cMaxPhysExts;
202	rc = CFGMR3QueryU16Def(pCfg, "MaxPhysExts", &cMaxPhysExts, RT_MAX(cMaxPages * 2, 2048 /* 2k max as this eat too much hyper heap */));
203	AssertLogRelRCReturn(rc, rc);
204	AssertLogRelMsgReturn(cMaxPhysExts >= 16 && cMaxPages <= PGMPOOL_IDX_LAST,
205	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxPhysExts), VERR_INVALID_PARAMETER);
206
207	/** @cfgm{/PGM/Pool/ChacheEnabled, bool, true}
208	* Enables or disabling caching of shadow pages. Caching means that we will try
209	* reuse shadow pages instead of recreating them everything SyncCR3, SyncPT or
210	* SyncPage requests one. When reusing a shadow page, we can save time
211	* reconstructing it and it's children.
212	*/
213	bool fCacheEnabled;
214	rc = CFGMR3QueryBoolDef(pCfg, "CacheEnabled", &fCacheEnabled, true);
215	AssertLogRelRCReturn(rc, rc);
216
217	LogRel(("pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
218	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
219
220	/*
221	* Allocate the data structures.
222	*/
223	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
224	cb += cMaxUsers * sizeof(PGMPOOLUSER);
225	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
226	PPGMPOOL pPool;
227	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
228	if (RT_FAILURE(rc))
229	return rc;
230	pVM->pgm.s.pPoolR3 = pPool;
231	pVM->pgm.s.pPoolR0 = MMHyperR3ToR0(pVM, pPool);
232	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pPool);
233
234	/*
235	* Initialize it.
236	*/
237	pPool->pVMR3 = pVM;
238	pPool->pVMR0 = pVM->pVMR0;
239	pPool->pVMRC = pVM->pVMRC;
240	pPool->cMaxPages = cMaxPages;
241	pPool->cCurPages = PGMPOOL_IDX_FIRST;
242	pPool->iUserFreeHead = 0;
243	pPool->cMaxUsers = cMaxUsers;
244	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
245	pPool->paUsersR3 = paUsers;
246	pPool->paUsersR0 = MMHyperR3ToR0(pVM, paUsers);
247	pPool->paUsersRC = MMHyperR3ToRC(pVM, paUsers);
248	for (unsigned i = 0; i < cMaxUsers; i++)
249	{
250	paUsers[i].iNext = i + 1;
251	paUsers[i].iUser = NIL_PGMPOOL_IDX;
252	paUsers[i].iUserTable = 0xfffffffe;
253	}
254	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
255	pPool->iPhysExtFreeHead = 0;
256	pPool->cMaxPhysExts = cMaxPhysExts;
257	PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
258	pPool->paPhysExtsR3 = paPhysExts;
259	pPool->paPhysExtsR0 = MMHyperR3ToR0(pVM, paPhysExts);
260	pPool->paPhysExtsRC = MMHyperR3ToRC(pVM, paPhysExts);
261	for (unsigned i = 0; i < cMaxPhysExts; i++)
262	{
263	paPhysExts[i].iNext = i + 1;
264	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
265	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
266	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
267	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
268	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
269	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
270	}
271	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
272	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
273	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
274	pPool->iAgeHead = NIL_PGMPOOL_IDX;
275	pPool->iAgeTail = NIL_PGMPOOL_IDX;
276	pPool->fCacheEnabled = fCacheEnabled;
277	pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
278	pPool->pszAccessHandler = "Guest Paging Access Handler";
279	pPool->HCPhysTree = 0;
280
281	/* The NIL entry. */
282	Assert(NIL_PGMPOOL_IDX == 0);
283	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
284
285	/* The Shadow 32-bit PD. (32 bits guest paging) */
286	pPool->aPages[PGMPOOL_IDX_PD].Core.Key = NIL_RTHCPHYS;
287	pPool->aPages[PGMPOOL_IDX_PD].GCPhys = NIL_RTGCPHYS;
288	pPool->aPages[PGMPOOL_IDX_PD].pvPageR3 = 0;
289	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_32BIT_PD;
290	pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
291
292	/* The Shadow PDPT. */
293	pPool->aPages[PGMPOOL_IDX_PDPT].Core.Key = NIL_RTHCPHYS;
294	pPool->aPages[PGMPOOL_IDX_PDPT].GCPhys = NIL_RTGCPHYS;
295	pPool->aPages[PGMPOOL_IDX_PDPT].pvPageR3 = 0;
296	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_PAE_PDPT;
297	pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
298
299	/* The Shadow AMD64 CR3. */
300	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].Core.Key = NIL_RTHCPHYS;
301	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].GCPhys = NIL_RTGCPHYS;
302	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].pvPageR3 = 0;
303	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].enmKind = PGMPOOLKIND_64BIT_PML4;
304	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].idx = PGMPOOL_IDX_AMD64_CR3;
305
306	/* The Nested Paging CR3. */
307	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].Core.Key = NIL_RTHCPHYS;
308	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].GCPhys = NIL_RTGCPHYS;
309	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].pvPageR3 = 0;
310	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].enmKind = PGMPOOLKIND_ROOT_NESTED;
311	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].idx = PGMPOOL_IDX_NESTED_ROOT;
312
313	/*
314	* Set common stuff.
315	*/
316	for (unsigned iPage = 1; iPage < PGMPOOL_IDX_FIRST; iPage++)
317	{
318	pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
319	pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
320	pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
321	pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
322	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
323	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
324	pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
325	pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
326	Assert(pPool->aPages[iPage].idx == iPage);
327	Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
328	Assert(!pPool->aPages[iPage].fSeenNonGlobal);
329	Assert(!pPool->aPages[iPage].fMonitored);
330	Assert(!pPool->aPages[iPage].fCached);
331	Assert(!pPool->aPages[iPage].fZeroed);
332	Assert(!pPool->aPages[iPage].fReusedFlushPending);
333	}
334
335	#ifdef VBOX_WITH_STATISTICS
336	/*
337	* Register statistics.
338	*/
339	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
340	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
341	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
342	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
343	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
344	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolClearAll.");
345	STAM_REG(pVM, &pPool->StatR3Reset, STAMTYPE_PROFILE, "/PGM/Pool/R3Reset", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolReset.");
346	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
347	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
348	STAM_REG(pVM, &pPool->StatForceFlushPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForce", STAMUNIT_OCCURENCES, "Counting explicit flushes by PGMPoolFlushPage().");
349	STAM_REG(pVM, &pPool->StatForceFlushDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForceDirty", STAMUNIT_OCCURENCES, "Counting explicit flushes of dirty pages by PGMPoolFlushPage().");
350	STAM_REG(pVM, &pPool->StatForceFlushReused, STAMTYPE_COUNTER, "/PGM/Pool/FlushReused", STAMUNIT_OCCURENCES, "Counting flushes for reused pages.");
351	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spent zeroing pages. Overlaps with Alloc.");
352	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
353	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
354	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackDeref.");
355	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPT.");
356	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
357	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
358	STAM_REG(pVM, &pPool->StatTrackFlushEntry, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Flush", STAMUNIT_COUNT, "Nr of flushed entries.");
359	STAM_REG(pVM, &pPool->StatTrackFlushEntryKeep, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Update", STAMUNIT_COUNT, "Nr of updated entries.");
360	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.");
361	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_TICKS_PER_CALL, "Profiling deref activity related tracking GC physical pages.");
362	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
363	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
364	STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
365	STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
366	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.");
367	STAM_REG(pVM, &pPool->StatMonitorRZFlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
368	STAM_REG(pVM, &pPool->StatMonitorRZFlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
369	STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
370	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).");
371	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
372	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.");
373	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.");
374	STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
375	STAM_REG(pVM, &pPool->StatMonitorRZFaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
376	STAM_REG(pVM, &pPool->StatMonitorRZFaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
377	STAM_REG(pVM, &pPool->StatMonitorRZFaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
378	STAM_REG(pVM, &pPool->StatMonitorRZFaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
379	STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
380	STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
381	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.");
382	STAM_REG(pVM, &pPool->StatMonitorR3FlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
383	STAM_REG(pVM, &pPool->StatMonitorR3FlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
384	STAM_REG(pVM, &pPool->StatMonitorR3Fork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
385	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).");
386	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.");
387	STAM_REG(pVM, &pPool->StatMonitorR3RepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
388	STAM_REG(pVM, &pPool->StatMonitorR3FaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
389	STAM_REG(pVM, &pPool->StatMonitorR3FaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
390	STAM_REG(pVM, &pPool->StatMonitorR3FaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
391	STAM_REG(pVM, &pPool->StatMonitorR3FaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
392	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.");
393	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
394	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
395	STAM_REG(pVM, &pPool->StatResetDirtyPages, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Resets", STAMUNIT_OCCURENCES, "Times we've called pgmPoolResetDirtyPages (and there were dirty page).");
396	STAM_REG(pVM, &pPool->StatDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Pages", STAMUNIT_OCCURENCES, "Times we've called pgmPoolAddDirtyPage.");
397	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.");
398	STAM_REG(pVM, &pPool->StatDirtyPageOverFlowFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushOverflow",STAMUNIT_OCCURENCES, "Times we've had to flush because of overflow.");
399	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
400	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
401	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!)");
402	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.");
403	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
404	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
405	#endif /* VBOX_WITH_STATISTICS */
406
407	#ifdef VBOX_WITH_DEBUGGER
408	/*
409	* Debugger commands.
410	*/
411	static bool s_fRegisteredCmds = false;
412	if (!s_fRegisteredCmds)
413	{
414	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
415	if (RT_SUCCESS(rc))
416	s_fRegisteredCmds = true;
417	}
418	#endif
419
420	return VINF_SUCCESS;
421	}
422
423
424	/**
425	* Relocate the page pool data.
426	*
427	* @param pVM The VM handle.
428	*/
429	void pgmR3PoolRelocate(PVM pVM)
430	{
431	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
432	pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
433	pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
434	pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
435	int rc = PDMR3LdrGetSymbolRC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerRC);
436	AssertReleaseRC(rc);
437	/* init order hack. */
438	if (!pVM->pgm.s.pPoolR3->pfnAccessHandlerR0)
439	{
440	rc = PDMR3LdrGetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerR0);
441	AssertReleaseRC(rc);
442	}
443	}
444
445
446	/**
447	* Grows the shadow page pool.
448	*
449	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
450	*
451	* @returns VBox status code.
452	* @param pVM The VM handle.
453	*/
454	VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
455	{
456	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
457	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_INTERNAL_ERROR);
458
459	pgmLock(pVM);
460
461	/*
462	* How much to grow it by?
463	*/
464	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
465	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
466	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages.\n", cPages, cPages));
467
468	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
469	{
470	PPGMPOOLPAGE pPage = &pPool->aPages[i];
471
472	/* Allocate all pages in low (below 4 GB) memory as 32 bits guests need a page table root in low memory. */
473	pPage->pvPageR3 = MMR3PageAllocLow(pVM);
474	if (!pPage->pvPageR3)
475	{
476	Log(("We're out of memory!! i=%d\n", i));
477	pgmUnlock(pVM);
478	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
479	}
480	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
481	AssertFatal(pPage->Core.Key < _4G);
482	pPage->GCPhys = NIL_RTGCPHYS;
483	pPage->enmKind = PGMPOOLKIND_FREE;
484	pPage->idx = pPage - &pPool->aPages[0];
485	LogFlow(("PGMR3PoolGrow: insert page #%#x - %RHp\n", pPage->idx, pPage->Core.Key));
486	pPage->iNext = pPool->iFreeHead;
487	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
488	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
489	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
490	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
491	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
492	pPage->iAgeNext = NIL_PGMPOOL_IDX;
493	pPage->iAgePrev = NIL_PGMPOOL_IDX;
494	/* commit it */
495	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
496	pPool->iFreeHead = i;
497	pPool->cCurPages = i + 1;
498	}
499
500	pgmUnlock(pVM);
501	Assert(pPool->cCurPages <= pPool->cMaxPages);
502	return VINF_SUCCESS;
503	}
504
505
506
507	/**
508	* Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
509	*
510	* @param pPool The pool.
511	* @param pPage The page.
512	*/
513	static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
514	{
515	/* for the present this should be safe enough I think... */
516	pgmLock(pPool->pVMR3);
517	if ( pPage->fReusedFlushPending
518	&& pPage->enmKind != PGMPOOLKIND_FREE)
519	pgmPoolFlushPage(pPool, pPage);
520	pgmUnlock(pPool->pVMR3);
521	}
522
523
524	/**
525	* \#PF Handler callback for PT write accesses.
526	*
527	* The handler can not raise any faults, it's mainly for monitoring write access
528	* to certain pages.
529	*
530	* @returns VINF_SUCCESS if the handler has carried out the operation.
531	* @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
532	* @param pVM VM Handle.
533	* @param GCPhys The physical address the guest is writing to.
534	* @param pvPhys The HC mapping of that address.
535	* @param pvBuf What the guest is reading/writing.
536	* @param cbBuf How much it's reading/writing.
537	* @param enmAccessType The access type.
538	* @param pvUser User argument.
539	*/
540	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser)
541	{
542	STAM_PROFILE_START(&pVM->pgm.s.pPoolR3->StatMonitorR3, a);
543	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
544	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
545	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%RGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
546	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
547
548	PVMCPU pVCpu = VMMGetCpu(pVM);
549
550	/*
551	* We don't have to be very sophisticated about this since there are relativly few calls here.
552	* However, we must try our best to detect any non-cpu accesses (disk / networking).
553	*
554	* Just to make life more interesting, we'll have to deal with the async threads too.
555	* We cannot flush a page if we're in an async thread because of REM notifications.
556	*/
557	pgmLock(pVM);
558	if (PHYS_PAGE_ADDRESS(GCPhys) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
559	{
560	/* Pool page changed while we were waiting for the lock; ignore. */
561	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)));
562	pgmUnlock(pVM);
563	return VINF_PGM_HANDLER_DO_DEFAULT;
564	}
565
566	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
567
568	/* @todo this code doesn't make any sense. remove the if (!pVCpu) block */
569	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} */
570	{
571	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
572	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
573	STAM_COUNTER_INC(&pPool->StatMonitorR3Async);
574	if (!pPage->fReusedFlushPending)
575	{
576	pgmUnlock(pVM);
577	int rc = VMR3ReqCallVoidNoWait(pPool->pVMR3, VMCPUID_ANY, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
578	AssertRCReturn(rc, rc);
579	pgmLock(pVM);
580	pPage->fReusedFlushPending = true;
581	pPage->cModifications += 0x1000;
582	}
583
584	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
585	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
586	while (cbBuf > 4)
587	{
588	cbBuf -= 4;
589	pvPhys = (uint8_t *)pvPhys + 4;
590	GCPhys += 4;
591	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
592	}
593	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
594	}
595	else if ( ( pPage->cModifications < 96 /* it's cheaper here. */
596	\|\| pgmPoolIsPageLocked(&pVM->pgm.s, pPage)
597	)
598	&& cbBuf <= 4)
599	{
600	/* Clear the shadow entry. */
601	if (!pPage->cModifications++)
602	pgmPoolMonitorModifiedInsert(pPool, pPage);
603	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
604	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
605	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
606	}
607	else
608	{
609	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
610	STAM_PROFILE_STOP_EX(&pPool->StatMonitorR3, &pPool->StatMonitorR3FlushPage, a);
611	}
612	pgmUnlock(pVM);
613	return VINF_PGM_HANDLER_DO_DEFAULT;
614	}
615
616
617	/**
618	* Rendezvous callback used by pgmR3PoolClearAll that clears all shadow pages
619	* and all modification counters.
620	*
621	* This is only called on one of the EMTs while the other ones are waiting for
622	* it to complete this function.
623	*
624	* @returns VINF_SUCCESS (VBox strict status code).
625	* @param pVM The VM handle.
626	* @param pVCpu The VMCPU for the EMT we're being called on. Unused.
627	* @param fpvFlushRemTlb When not NULL, we'll flush the REM TLB as well.
628	* (This is the pvUser, so it has to be void *.)
629	*
630	*/
631	DECLCALLBACK(VBOXSTRICTRC) pgmR3PoolClearAllRendezvous(PVM pVM, PVMCPU pVCpu, void *fpvFlushRemTbl)
632	{
633	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
634	STAM_PROFILE_START(&pPool->StatClearAll, c);
635
636	pgmLock(pVM);
637	Log(("pgmR3PoolClearAllRendezvous: cUsedPages=%d fpvFlushRemTbl=%RTbool\n", pPool->cUsedPages, !!fpvFlushRemTbl));
638
639	/*
640	* Iterate all the pages until we've encountered all that are in use.
641	* This is a simple but not quite optimal solution.
642	*/
643	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
644	unsigned cLeft = pPool->cUsedPages;
645	unsigned iPage = pPool->cCurPages;
646	while (--iPage >= PGMPOOL_IDX_FIRST)
647	{
648	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
649	if (pPage->GCPhys != NIL_RTGCPHYS)
650	{
651	switch (pPage->enmKind)
652	{
653	/*
654	* We only care about shadow page tables that reference physical memory
655	*/
656	#ifdef PGM_WITH_LARGE_PAGES
657	case PGMPOOLKIND_EPT_PD_FOR_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
658	if (pPage->cPresent)
659	{
660	PX86PDPAE pShwPD = (PX86PDPAE)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
661	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
662	{
663	if ( pShwPD->a[i].n.u1Present
664	&& pShwPD->a[i].b.u1Size)
665	{
666	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
667	pShwPD->a[i].u = 0;
668	Assert(pPage->cPresent);
669	pPage->cPresent--;
670	}
671	}
672	if (pPage->cPresent == 0)
673	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
674	}
675	goto default_case;
676
677	case PGMPOOLKIND_PAE_PD_PHYS: /* Large pages reference 2 MB of physical memory, so we must clear them. */
678	if (pPage->cPresent)
679	{
680	PEPTPD pShwPD = (PEPTPD)PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
681	for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
682	{
683	Assert((pShwPD->a[i].u & UINT64_C(0xfff0000000000f80)) == 0);
684	if ( pShwPD->a[i].n.u1Present
685	&& pShwPD->a[i].b.u1Size)
686	{
687	Assert(!(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING));
688	pShwPD->a[i].u = 0;
689	Assert(pPage->cPresent);
690	pPage->cPresent--;
691	}
692	}
693	if (pPage->cPresent == 0)
694	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
695	}
696	goto default_case;
697	#endif /* PGM_WITH_LARGE_PAGES */
698
699	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
700	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
701	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
702	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
703	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
704	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
705	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
706	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
707	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
708	{
709	if (pPage->cPresent)
710	{
711	void *pvShw = PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
712	STAM_PROFILE_START(&pPool->StatZeroPage, z);
713	#if 0
714	/* Useful check for leaking references; very expensive though. */
715	switch (pPage->enmKind)
716	{
717	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
718	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
719	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
720	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
721	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
722	{
723	bool fFoundFirst = false;
724	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)pvShw;
725	for (unsigned ptIndex = 0; ptIndex < RT_ELEMENTS(pPT->a); ptIndex++)
726	{
727	if (pPT->a[ptIndex].u)
728	{
729	if (!fFoundFirst)
730	{
731	AssertFatalMsg(pPage->iFirstPresent <= ptIndex, ("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
732	if (pPage->iFirstPresent != ptIndex)
733	Log(("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
734	fFoundFirst = true;
735	}
736	if (PGMSHWPTEPAE_IS_P(pPT->a[ptIndex]))
737	{
738	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pPT->a[ptIndex]), NIL_RTGCPHYS);
739	if (pPage->iFirstPresent == ptIndex)
740	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
741	}
742	}
743	}
744	AssertFatalMsg(pPage->cPresent == 0, ("cPresent = %d pPage = %RGv\n", pPage->cPresent, pPage->GCPhys));
745	break;
746	}
747	default:
748	break;
749	}
750	#endif
751	ASMMemZeroPage(pvShw);
752	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
753	pPage->cPresent = 0;
754	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
755	}
756	}
757	/* fall thru */
758
759	#ifdef PGM_WITH_LARGE_PAGES
760	default_case:
761	#endif
762	default:
763	Assert(!pPage->cModifications \|\| ++cModifiedPages);
764	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
765	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
766	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
767	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
768	pPage->cModifications = 0;
769	break;
770
771	}
772	if (!--cLeft)
773	break;
774	}
775	}
776
777	/* swipe the special pages too. */
778	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
779	{
780	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
781	if (pPage->GCPhys != NIL_RTGCPHYS)
782	{
783	Assert(!pPage->cModifications \|\| ++cModifiedPages);
784	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
785	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
786	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
787	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
788	pPage->cModifications = 0;
789	}
790	}
791
792	#ifndef DEBUG_michael
793	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
794	#endif
795	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
796	pPool->cModifiedPages = 0;
797
798	/*
799	* Clear all the GCPhys links and rebuild the phys ext free list.
800	*/
801	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
802	pRam;
803	pRam = pRam->CTX_SUFF(pNext))
804	{
805	iPage = pRam->cb >> PAGE_SHIFT;
806	while (iPage-- > 0)
807	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
808	}
809
810	pPool->iPhysExtFreeHead = 0;
811	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
812	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
813	for (unsigned i = 0; i < cMaxPhysExts; i++)
814	{
815	paPhysExts[i].iNext = i + 1;
816	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
817	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
818	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
819	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
820	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
821	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
822	}
823	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
824
825
826	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
827	/* Reset all dirty pages to reactivate the page monitoring. */
828	/* Note: we must do this after clearing all page references and shadow page tables as there might be stale references to
829	* recently removed MMIO ranges around that might otherwise end up asserting in pgmPoolTracDerefGCPhysHint
830	*/
831	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
832	{
833	PPGMPOOLPAGE pPage;
834	unsigned idxPage;
835
836	if (pPool->aDirtyPages[i].uIdx == NIL_PGMPOOL_IDX)
837	continue;
838
839	idxPage = pPool->aDirtyPages[i].uIdx;
840	AssertRelease(idxPage != NIL_PGMPOOL_IDX);
841	pPage = &pPool->aPages[idxPage];
842	Assert(pPage->idx == idxPage);
843	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
844
845	AssertMsg(pPage->fDirty, ("Page %RGp (slot=%d) not marked dirty!", pPage->GCPhys, i));
846
847	Log(("Reactivate dirty page %RGp\n", pPage->GCPhys));
848
849	/* First write protect the page again to catch all write accesses. (before checking for changes -> SMP) */
850	int rc = PGMHandlerPhysicalReset(pVM, pPage->GCPhys & PAGE_BASE_GC_MASK);
851	Assert(rc == VINF_SUCCESS);
852	pPage->fDirty = false;
853
854	pPool->aDirtyPages[i].uIdx = NIL_PGMPOOL_IDX;
855	}
856
857	/* Clear all dirty pages. */
858	pPool->idxFreeDirtyPage = 0;
859	pPool->cDirtyPages = 0;
860	#endif
861
862	/* Clear the PGM_SYNC_CLEAR_PGM_POOL flag on all VCPUs to prevent redundant flushes. */
863	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
864	pVM->aCpus[idCpu].pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
865
866	/* Flush job finished. */
867	VM_FF_CLEAR(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
868	pPool->cPresent = 0;
869	pgmUnlock(pVM);
870
871	PGM_INVL_ALL_VCPU_TLBS(pVM);
872
873	if (fpvFlushRemTbl)
874	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
875	CPUMSetChangedFlags(&pVM->aCpus[idCpu], CPUM_CHANGED_GLOBAL_TLB_FLUSH);
876
877	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
878	return VINF_SUCCESS;
879	}
880
881
882	/**
883	* Clears the shadow page pool.
884	*
885	* @param pVM The VM handle.
886	* @param fFlushRemTlb When set, the REM TLB is scheduled for flushing as
887	* well.
888	*/
889	void pgmR3PoolClearAll(PVM pVM, bool fFlushRemTlb)
890	{
891	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PoolClearAllRendezvous, &fFlushRemTlb);
892	AssertRC(rc);
893	}
894
895	/**
896	* Protect all pgm pool page table entries to monitor writes
897	*
898	* @param pVM The VM handle.
899	*
900	* Remark: assumes the caller will flush all TLBs (!!)
901	*/
902	void pgmR3PoolWriteProtectPages(PVM pVM)
903	{
904	Assert(PGMIsLockOwner(pVM));
905	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
906	unsigned cLeft = pPool->cUsedPages;
907	unsigned iPage = pPool->cCurPages;
908	while (--iPage >= PGMPOOL_IDX_FIRST)
909	{
910	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
911	if ( pPage->GCPhys != NIL_RTGCPHYS
912	&& pPage->cPresent)
913	{
914	union
915	{
916	void *pv;
917	PX86PT pPT;
918	PPGMSHWPTPAE pPTPae;
919	PEPTPT pPTEpt;
920	} uShw;
921	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
922
923	switch (pPage->enmKind)
924	{
925	/*
926	* We only care about shadow page tables.
927	*/
928	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
929	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
930	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
931	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPT->a); iShw++)
932	{
933	if (uShw.pPT->a[iShw].n.u1Present)
934	uShw.pPT->a[iShw].n.u1Write = 0;
935	}
936	break;
937
938	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
939	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
940	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
941	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
942	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
943	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTPae->a); iShw++)
944	{
945	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
946	PGMSHWPTEPAE_SET_RO(uShw.pPTPae->a[iShw]);
947	}
948	break;
949
950	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
951	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTEpt->a); iShw++)
952	{
953	if (uShw.pPTEpt->a[iShw].n.u1Present)
954	uShw.pPTEpt->a[iShw].n.u1Write = 0;
955	}
956	break;
957
958	default:
959	break;
960	}
961	if (!--cLeft)
962	break;
963	}
964	}
965	}
966
967	#ifdef VBOX_WITH_DEBUGGER
968	/**
969	* The '.pgmpoolcheck' command.
970	*
971	* @returns VBox status.
972	* @param pCmd Pointer to the command descriptor (as registered).
973	* @param pCmdHlp Pointer to command helper functions.
974	* @param pVM Pointer to the current VM (if any).
975	* @param paArgs Pointer to (readonly) array of arguments.
976	* @param cArgs Number of arguments in the array.
977	*/
978	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PVM pVM, PCDBGCVAR paArgs, unsigned cArgs, PDBGCVAR pResult)
979	{
980	/*
981	* Validate input.
982	*/
983	if (!pVM)
984	return pCmdHlp->pfnPrintf(pCmdHlp, NULL, "error: The command requires a VM to be selected.\n");
985
986	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
987
988	for (unsigned i = 0; i < pPool->cCurPages; i++)
989	{
990	PPGMPOOLPAGE pPage = &pPool->aPages[i];
991	bool fFirstMsg = true;
992
993	/* Todo: cover other paging modes too. */
994	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
995	{
996	PPGMSHWPTPAE pShwPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
997	{
998	PX86PTPAE pGstPT;
999	PGMPAGEMAPLOCK LockPage;
1000	int rc = PGMPhysGCPhys2CCPtrReadOnly(pVM, pPage->GCPhys, (const void **)&pGstPT, &LockPage); AssertReleaseRC(rc);
1001
1002	/* Check if any PTEs are out of sync. */
1003	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
1004	{
1005	if (PGMSHWPTEPAE_IS_P(pShwPT->a[j]))
1006	{
1007	RTHCPHYS HCPhys = NIL_RTHCPHYS;
1008	rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[j].u & X86_PTE_PAE_PG_MASK, &HCPhys);
1009	if ( rc != VINF_SUCCESS
1010	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[j]) != HCPhys)
1011	{
1012	if (fFirstMsg)
1013	{
1014	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Check pool page %RGp\n", pPage->GCPhys);
1015	fFirstMsg = false;
1016	}
1017	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "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);
1018	}
1019	else if ( PGMSHWPTEPAE_IS_RW(pShwPT->a[j])
1020	&& !pGstPT->a[j].n.u1Write)
1021	{
1022	if (fFirstMsg)
1023	{
1024	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Check pool page %RGp\n", pPage->GCPhys);
1025	fFirstMsg = false;
1026	}
1027	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "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);
1028	}
1029	}
1030	}
1031	PGMPhysReleasePageMappingLock(pVM, &LockPage);
1032	}
1033
1034	/* Make sure this page table can't be written to from any shadow mapping. */
1035	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
1036	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pPage->GCPhys, &HCPhysPT);
1037	AssertMsgRC(rc, ("PGMPhysGCPhys2HCPhys failed with rc=%d for %RGp\n", rc, pPage->GCPhys));
1038	if (rc == VINF_SUCCESS)
1039	{
1040	for (unsigned j = 0; j < pPool->cCurPages; j++)
1041	{
1042	PPGMPOOLPAGE pTempPage = &pPool->aPages[j];
1043
1044	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
1045	{
1046	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pTempPage);
1047
1048	for (unsigned k = 0; k < RT_ELEMENTS(pShwPT->a); k++)
1049	{
1050	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[k])
1051	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
1052	&& !pPage->fDirty
1053	# endif
1054	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[k]) == HCPhysPT)
1055	{
1056	if (fFirstMsg)
1057	{
1058	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Check pool page %RGp\n", pPage->GCPhys);
1059	fFirstMsg = false;
1060	}
1061	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "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]));
1062	}
1063	}
1064	}
1065	}
1066	}
1067	}
1068	}
1069	return VINF_SUCCESS;
1070	}
1071	#endif /* VBOX_WITH_DEBUGGER */

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

source: vbox/trunk/src/VBox/VMM/PGMPool.cpp@ 34347

Download in other formats: