PGMPool.cpp@ 32963

Last change on this file since 32963 was 32550, checked in by vboxsync, 14 years ago
Correction
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 49.9 KB

Line
1	/* $Id: PGMPool.cpp 32550 2010-09-16 11:45:45Z 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	* Initalizes 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	/** @cfgm{/PGM/Pool/MaxPages, uint16_t, #pages, 16, 0x3fff, 1024}
147	* The max size of the shadow page pool in pages. The pool will grow dynamically
148	* up to this limit.
149	*/
150	uint16_t cMaxPages;
151	int rc = CFGMR3QueryU16Def(pCfg, "MaxPages", &cMaxPages, 4*_1M >> PAGE_SHIFT);
152	AssertLogRelRCReturn(rc, rc);
153	AssertLogRelMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
154	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
155	cMaxPages = RT_ALIGN(cMaxPages, 16);
156
157	/** @cfgm{/PGM/Pool/MaxUsers, uint16_t, #users, MaxUsers, 32K, MaxPages*2}
158	* The max number of shadow page user tracking records. Each shadow page has
159	* zero of other shadow pages (or CR3s) that references it, or uses it if you
160	* like. The structures describing these relationships are allocated from a
161	* fixed sized pool. This configuration variable defines the pool size.
162	*/
163	uint16_t cMaxUsers;
164	rc = CFGMR3QueryU16Def(pCfg, "MaxUsers", &cMaxUsers, cMaxPages * 2);
165	AssertLogRelRCReturn(rc, rc);
166	AssertLogRelMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
167	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
168
169	/** @cfgm{/PGM/Pool/MaxPhysExts, uint16_t, #extents, 16, MaxPages * 2, MAX(MaxPages*2,0x3fff)}
170	* The max number of extents for tracking aliased guest pages.
171	*/
172	uint16_t cMaxPhysExts;
173	rc = CFGMR3QueryU16Def(pCfg, "MaxPhysExts", &cMaxPhysExts, RT_MAX(cMaxPages * 2, PGMPOOL_IDX_LAST));
174	AssertLogRelRCReturn(rc, rc);
175	AssertLogRelMsgReturn(cMaxPhysExts >= 16 && cMaxPages <= PGMPOOL_IDX_LAST,
176	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxPhysExts), VERR_INVALID_PARAMETER);
177
178	/** @cfgm{/PGM/Pool/ChacheEnabled, bool, true}
179	* Enables or disabling caching of shadow pages. Chaching means that we will try
180	* reuse shadow pages instead of recreating them everything SyncCR3, SyncPT or
181	* SyncPage requests one. When reusing a shadow page, we can save time
182	* reconstructing it and it's children.
183	*/
184	bool fCacheEnabled;
185	rc = CFGMR3QueryBoolDef(pCfg, "CacheEnabled", &fCacheEnabled, true);
186	AssertLogRelRCReturn(rc, rc);
187
188	Log(("pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
189	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
190
191	/*
192	* Allocate the data structures.
193	*/
194	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
195	cb += cMaxUsers * sizeof(PGMPOOLUSER);
196	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
197	PPGMPOOL pPool;
198	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
199	if (RT_FAILURE(rc))
200	return rc;
201	pVM->pgm.s.pPoolR3 = pPool;
202	pVM->pgm.s.pPoolR0 = MMHyperR3ToR0(pVM, pPool);
203	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pPool);
204
205	/*
206	* Initialize it.
207	*/
208	pPool->pVMR3 = pVM;
209	pPool->pVMR0 = pVM->pVMR0;
210	pPool->pVMRC = pVM->pVMRC;
211	pPool->cMaxPages = cMaxPages;
212	pPool->cCurPages = PGMPOOL_IDX_FIRST;
213	pPool->iUserFreeHead = 0;
214	pPool->cMaxUsers = cMaxUsers;
215	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
216	pPool->paUsersR3 = paUsers;
217	pPool->paUsersR0 = MMHyperR3ToR0(pVM, paUsers);
218	pPool->paUsersRC = MMHyperR3ToRC(pVM, paUsers);
219	for (unsigned i = 0; i < cMaxUsers; i++)
220	{
221	paUsers[i].iNext = i + 1;
222	paUsers[i].iUser = NIL_PGMPOOL_IDX;
223	paUsers[i].iUserTable = 0xfffffffe;
224	}
225	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
226	pPool->iPhysExtFreeHead = 0;
227	pPool->cMaxPhysExts = cMaxPhysExts;
228	PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
229	pPool->paPhysExtsR3 = paPhysExts;
230	pPool->paPhysExtsR0 = MMHyperR3ToR0(pVM, paPhysExts);
231	pPool->paPhysExtsRC = MMHyperR3ToRC(pVM, paPhysExts);
232	for (unsigned i = 0; i < cMaxPhysExts; i++)
233	{
234	paPhysExts[i].iNext = i + 1;
235	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
236	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
237	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
238	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
239	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
240	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
241	}
242	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
243	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
244	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
245	pPool->iAgeHead = NIL_PGMPOOL_IDX;
246	pPool->iAgeTail = NIL_PGMPOOL_IDX;
247	pPool->fCacheEnabled = fCacheEnabled;
248	pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
249	pPool->pszAccessHandler = "Guest Paging Access Handler";
250	pPool->HCPhysTree = 0;
251
252	/* The NIL entry. */
253	Assert(NIL_PGMPOOL_IDX == 0);
254	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
255
256	/* The Shadow 32-bit PD. (32 bits guest paging) */
257	pPool->aPages[PGMPOOL_IDX_PD].Core.Key = NIL_RTHCPHYS;
258	pPool->aPages[PGMPOOL_IDX_PD].GCPhys = NIL_RTGCPHYS;
259	pPool->aPages[PGMPOOL_IDX_PD].pvPageR3 = 0;
260	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_32BIT_PD;
261	pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
262
263	/* The Shadow PDPT. */
264	pPool->aPages[PGMPOOL_IDX_PDPT].Core.Key = NIL_RTHCPHYS;
265	pPool->aPages[PGMPOOL_IDX_PDPT].GCPhys = NIL_RTGCPHYS;
266	pPool->aPages[PGMPOOL_IDX_PDPT].pvPageR3 = 0;
267	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_PAE_PDPT;
268	pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
269
270	/* The Shadow AMD64 CR3. */
271	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].Core.Key = NIL_RTHCPHYS;
272	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].GCPhys = NIL_RTGCPHYS;
273	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].pvPageR3 = 0;
274	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].enmKind = PGMPOOLKIND_64BIT_PML4;
275	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].idx = PGMPOOL_IDX_AMD64_CR3;
276
277	/* The Nested Paging CR3. */
278	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].Core.Key = NIL_RTHCPHYS;
279	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].GCPhys = NIL_RTGCPHYS;
280	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].pvPageR3 = 0;
281	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].enmKind = PGMPOOLKIND_ROOT_NESTED;
282	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].idx = PGMPOOL_IDX_NESTED_ROOT;
283
284	/*
285	* Set common stuff.
286	*/
287	for (unsigned iPage = 1; iPage < PGMPOOL_IDX_FIRST; iPage++)
288	{
289	pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
290	pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
291	pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
292	pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
293	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
294	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
295	pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
296	pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
297	Assert(pPool->aPages[iPage].idx == iPage);
298	Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
299	Assert(!pPool->aPages[iPage].fSeenNonGlobal);
300	Assert(!pPool->aPages[iPage].fMonitored);
301	Assert(!pPool->aPages[iPage].fCached);
302	Assert(!pPool->aPages[iPage].fZeroed);
303	Assert(!pPool->aPages[iPage].fReusedFlushPending);
304	}
305
306	#ifdef VBOX_WITH_STATISTICS
307	/*
308	* Register statistics.
309	*/
310	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
311	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
312	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
313	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
314	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
315	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolClearAll.");
316	STAM_REG(pVM, &pPool->StatR3Reset, STAMTYPE_PROFILE, "/PGM/Pool/R3Reset", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmR3PoolReset.");
317	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
318	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
319	STAM_REG(pVM, &pPool->StatForceFlushPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForce", STAMUNIT_OCCURENCES, "Counting explicit flushes by PGMPoolFlushPage().");
320	STAM_REG(pVM, &pPool->StatForceFlushDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/FlushForceDirty", STAMUNIT_OCCURENCES, "Counting explicit flushes of dirty pages by PGMPoolFlushPage().");
321	STAM_REG(pVM, &pPool->StatForceFlushReused, STAMTYPE_COUNTER, "/PGM/Pool/FlushReused", STAMUNIT_OCCURENCES, "Counting flushes for reused pages.");
322	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spent zeroing pages. Overlaps with Alloc.");
323	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
324	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
325	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackDeref.");
326	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPT.");
327	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
328	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
329	STAM_REG(pVM, &pPool->StatTrackFlushEntry, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Flush", STAMUNIT_COUNT, "Nr of flushed entries.");
330	STAM_REG(pVM, &pPool->StatTrackFlushEntryKeep, STAMTYPE_COUNTER, "/PGM/Pool/Track/Entry/Update", STAMUNIT_COUNT, "Nr of updated entries.");
331	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.");
332	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_TICKS_PER_CALL, "Profiling deref activity related tracking GC physical pages.");
333	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
334	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
335	STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
336	STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
337	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.");
338	STAM_REG(pVM, &pPool->StatMonitorRZFlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
339	STAM_REG(pVM, &pPool->StatMonitorRZFlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
340	STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
341	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).");
342	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
343	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.");
344	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.");
345	STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
346	STAM_REG(pVM, &pPool->StatMonitorRZFaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
347	STAM_REG(pVM, &pPool->StatMonitorRZFaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
348	STAM_REG(pVM, &pPool->StatMonitorRZFaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
349	STAM_REG(pVM, &pPool->StatMonitorRZFaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
350	STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
351	STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
352	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.");
353	STAM_REG(pVM, &pPool->StatMonitorR3FlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
354	STAM_REG(pVM, &pPool->StatMonitorR3FlushModOverflow,STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushOverflow", STAMUNIT_OCCURENCES, "Counting flushes for pages that are modified too often.");
355	STAM_REG(pVM, &pPool->StatMonitorR3Fork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
356	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).");
357	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.");
358	STAM_REG(pVM, &pPool->StatMonitorR3RepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
359	STAM_REG(pVM, &pPool->StatMonitorR3FaultPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PT", STAMUNIT_OCCURENCES, "Nr of handled PT faults.");
360	STAM_REG(pVM, &pPool->StatMonitorR3FaultPD, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PD", STAMUNIT_OCCURENCES, "Nr of handled PD faults.");
361	STAM_REG(pVM, &pPool->StatMonitorR3FaultPDPT, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PDPT", STAMUNIT_OCCURENCES, "Nr of handled PDPT faults.");
362	STAM_REG(pVM, &pPool->StatMonitorR3FaultPML4, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fault/PML4", STAMUNIT_OCCURENCES, "Nr of handled PML4 faults.");
363	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.");
364	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
365	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
366	STAM_REG(pVM, &pPool->StatResetDirtyPages, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Resets", STAMUNIT_OCCURENCES, "Times we've called pgmPoolResetDirtyPages (and there were dirty page).");
367	STAM_REG(pVM, &pPool->StatDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Pages", STAMUNIT_OCCURENCES, "Times we've called pgmPoolAddDirtyPage.");
368	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.");
369	STAM_REG(pVM, &pPool->StatDirtyPageOverFlowFlush, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/FlushOverflow",STAMUNIT_OCCURENCES, "Times we've had to flush because of overflow.");
370	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
371	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
372	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!)");
373	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.");
374	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
375	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
376	#endif /* VBOX_WITH_STATISTICS */
377
378	#ifdef VBOX_WITH_DEBUGGER
379	/*
380	* Debugger commands.
381	*/
382	static bool s_fRegisteredCmds = false;
383	if (!s_fRegisteredCmds)
384	{
385	rc = DBGCRegisterCommands(&g_aCmds[0], RT_ELEMENTS(g_aCmds));
386	if (RT_SUCCESS(rc))
387	s_fRegisteredCmds = true;
388	}
389	#endif
390
391	return VINF_SUCCESS;
392	}
393
394
395	/**
396	* Relocate the page pool data.
397	*
398	* @param pVM The VM handle.
399	*/
400	void pgmR3PoolRelocate(PVM pVM)
401	{
402	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
403	pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
404	pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
405	pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
406	int rc = PDMR3LdrGetSymbolRC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerRC);
407	AssertReleaseRC(rc);
408	/* init order hack. */
409	if (!pVM->pgm.s.pPoolR3->pfnAccessHandlerR0)
410	{
411	rc = PDMR3LdrGetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerR0);
412	AssertReleaseRC(rc);
413	}
414	}
415
416
417	/**
418	* Grows the shadow page pool.
419	*
420	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
421	*
422	* @returns VBox status code.
423	* @param pVM The VM handle.
424	*/
425	VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
426	{
427	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
428	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_INTERNAL_ERROR);
429
430	pgmLock(pVM);
431
432	/*
433	* How much to grow it by?
434	*/
435	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
436	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
437	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages.\n", cPages, cPages));
438
439	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
440	{
441	PPGMPOOLPAGE pPage = &pPool->aPages[i];
442
443	/* Allocate all pages in low (below 4 GB) memory as 32 bits guests need a page table root in low memory. */
444	pPage->pvPageR3 = MMR3PageAllocLow(pVM);
445	if (!pPage->pvPageR3)
446	{
447	Log(("We're out of memory!! i=%d\n", i));
448	pgmUnlock(pVM);
449	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
450	}
451	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
452	AssertFatal(pPage->Core.Key < _4G);
453	pPage->GCPhys = NIL_RTGCPHYS;
454	pPage->enmKind = PGMPOOLKIND_FREE;
455	pPage->idx = pPage - &pPool->aPages[0];
456	LogFlow(("PGMR3PoolGrow: insert page #%#x - %RHp\n", pPage->idx, pPage->Core.Key));
457	pPage->iNext = pPool->iFreeHead;
458	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
459	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
460	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
461	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
462	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
463	pPage->iAgeNext = NIL_PGMPOOL_IDX;
464	pPage->iAgePrev = NIL_PGMPOOL_IDX;
465	/* commit it */
466	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
467	pPool->iFreeHead = i;
468	pPool->cCurPages = i + 1;
469	}
470
471	pgmUnlock(pVM);
472	Assert(pPool->cCurPages <= pPool->cMaxPages);
473	return VINF_SUCCESS;
474	}
475
476
477
478	/**
479	* Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
480	*
481	* @param pPool The pool.
482	* @param pPage The page.
483	*/
484	static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
485	{
486	/* for the present this should be safe enough I think... */
487	pgmLock(pPool->pVMR3);
488	if ( pPage->fReusedFlushPending
489	&& pPage->enmKind != PGMPOOLKIND_FREE)
490	pgmPoolFlushPage(pPool, pPage);
491	pgmUnlock(pPool->pVMR3);
492	}
493
494
495	/**
496	* \#PF Handler callback for PT write accesses.
497	*
498	* The handler can not raise any faults, it's mainly for monitoring write access
499	* to certain pages.
500	*
501	* @returns VINF_SUCCESS if the handler has carried out the operation.
502	* @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
503	* @param pVM VM Handle.
504	* @param GCPhys The physical address the guest is writing to.
505	* @param pvPhys The HC mapping of that address.
506	* @param pvBuf What the guest is reading/writing.
507	* @param cbBuf How much it's reading/writing.
508	* @param enmAccessType The access type.
509	* @param pvUser User argument.
510	*/
511	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser)
512	{
513	STAM_PROFILE_START(&pVM->pgm.s.pPoolR3->StatMonitorR3, a);
514	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
515	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
516	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%RGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
517	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
518
519	PVMCPU pVCpu = VMMGetCpu(pVM);
520
521	/*
522	* We don't have to be very sophisticated about this since there are relativly few calls here.
523	* However, we must try our best to detect any non-cpu accesses (disk / networking).
524	*
525	* Just to make life more interesting, we'll have to deal with the async threads too.
526	* We cannot flush a page if we're in an async thread because of REM notifications.
527	*/
528	pgmLock(pVM);
529	if (PHYS_PAGE_ADDRESS(GCPhys) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
530	{
531	/* Pool page changed while we were waiting for the lock; ignore. */
532	Log(("CPU%d: pgmR3PoolAccessHandler pgm pool page for %RGp changed (to %RGp) while waiting!\n", pVCpu->idCpu, PHYS_PAGE_ADDRESS(GCPhys), PHYS_PAGE_ADDRESS(pPage->GCPhys)));
533	pgmUnlock(pVM);
534	return VINF_PGM_HANDLER_DO_DEFAULT;
535	}
536
537	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
538
539	/* @todo this code doesn't make any sense. remove the if (!pVCpu) block */
540	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} */
541	{
542	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
543	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
544	STAM_COUNTER_INC(&pPool->StatMonitorR3Async);
545	if (!pPage->fReusedFlushPending)
546	{
547	pgmUnlock(pVM);
548	int rc = VMR3ReqCallVoidNoWait(pPool->pVMR3, VMCPUID_ANY, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
549	AssertRCReturn(rc, rc);
550	pgmLock(pVM);
551	pPage->fReusedFlushPending = true;
552	pPage->cModifications += 0x1000;
553	}
554
555	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
556	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
557	while (cbBuf > 4)
558	{
559	cbBuf -= 4;
560	pvPhys = (uint8_t *)pvPhys + 4;
561	GCPhys += 4;
562	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
563	}
564	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
565	}
566	else if ( ( pPage->cModifications < 96 /* it's cheaper here. */
567	\|\| pgmPoolIsPageLocked(&pVM->pgm.s, pPage)
568	)
569	&& cbBuf <= 4)
570	{
571	/* Clear the shadow entry. */
572	if (!pPage->cModifications++)
573	pgmPoolMonitorModifiedInsert(pPool, pPage);
574	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
575	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, 0 /* unknown write size */);
576	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
577	}
578	else
579	{
580	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
581	STAM_PROFILE_STOP_EX(&pPool->StatMonitorR3, &pPool->StatMonitorR3FlushPage, a);
582	}
583	pgmUnlock(pVM);
584	return VINF_PGM_HANDLER_DO_DEFAULT;
585	}
586
587
588	/**
589	* Rendezvous callback used by pgmR3PoolClearAll that clears all shadow pages
590	* and all modification counters.
591	*
592	* This is only called on one of the EMTs while the other ones are waiting for
593	* it to complete this function.
594	*
595	* @returns VINF_SUCCESS (VBox strict status code).
596	* @param pVM The VM handle.
597	* @param pVCpu The VMCPU for the EMT we're being called on. Unused.
598	* @param fpvFlushRemTlb When not NULL, we'll flush the REM TLB as well.
599	* (This is the pvUser, so it has to be void *.)
600	*
601	*/
602	DECLCALLBACK(VBOXSTRICTRC) pgmR3PoolClearAllRendezvous(PVM pVM, PVMCPU pVCpu, void *fpvFlushRemTbl)
603	{
604	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
605	STAM_PROFILE_START(&pPool->StatClearAll, c);
606
607	pgmLock(pVM);
608	Log(("pgmR3PoolClearAllRendezvous: cUsedPages=%d fpvFlushRemTbl=%RTbool\n", pPool->cUsedPages, !!fpvFlushRemTbl));
609
610	/*
611	* Iterate all the pages until we've encountered all that are in use.
612	* This is a simple but not quite optimal solution.
613	*/
614	unsigned cModifiedPages = 0; NOREF(cModifiedPages);
615	unsigned cLeft = pPool->cUsedPages;
616	unsigned iPage = pPool->cCurPages;
617	while (--iPage >= PGMPOOL_IDX_FIRST)
618	{
619	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
620	if (pPage->GCPhys != NIL_RTGCPHYS)
621	{
622	switch (pPage->enmKind)
623	{
624	/*
625	* We only care about shadow page tables.
626	*/
627	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
628	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
629	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
630	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
631	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
632	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
633	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
634	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
635	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
636	{
637	if (pPage->cPresent)
638	{
639	void *pvShw = PGMPOOL_PAGE_2_PTR_V2(pPool->CTX_SUFF(pVM), pVCpu, pPage);
640	STAM_PROFILE_START(&pPool->StatZeroPage, z);
641	#if 0
642	/* Useful check for leaking references; very expensive though. */
643	switch (pPage->enmKind)
644	{
645	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
646	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
647	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
648	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
649	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
650	{
651	bool fFoundFirst = false;
652	PPGMSHWPTPAE pPT = (PPGMSHWPTPAE)pvShw;
653	for (unsigned ptIndex = 0; ptIndex < RT_ELEMENTS(pPT->a); ptIndex++)
654	{
655	if (pPT->a[ptIndex].u)
656	{
657	if (!fFoundFirst)
658	{
659	AssertFatalMsg(pPage->iFirstPresent <= ptIndex, ("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
660	if (pPage->iFirstPresent != ptIndex)
661	Log(("ptIndex = %d first present = %d\n", ptIndex, pPage->iFirstPresent));
662	fFoundFirst = true;
663	}
664	if (PGMSHWPTEPAE_IS_P(pPT->a[ptIndex]))
665	{
666	pgmPoolTracDerefGCPhysHint(pPool, pPage, PGMSHWPTEPAE_GET_HCPHYS(pPT->a[ptIndex]), NIL_RTGCPHYS);
667	if (pPage->iFirstPresent == ptIndex)
668	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
669	}
670	}
671	}
672	AssertFatalMsg(pPage->cPresent == 0, ("cPresent = %d pPage = %RGv\n", pPage->cPresent, pPage->GCPhys));
673	break;
674	}
675	default:
676	break;
677	}
678	#endif
679	ASMMemZeroPage(pvShw);
680	STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
681	pPage->cPresent = 0;
682	pPage->iFirstPresent = NIL_PGMPOOL_PRESENT_INDEX;
683	}
684	}
685	/* fall thru */
686
687	default:
688	Assert(!pPage->cModifications \|\| ++cModifiedPages);
689	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
690	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
691	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
692	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
693	pPage->cModifications = 0;
694	break;
695
696	}
697	if (!--cLeft)
698	break;
699	}
700	}
701
702	/* swipe the special pages too. */
703	for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
704	{
705	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
706	if (pPage->GCPhys != NIL_RTGCPHYS)
707	{
708	Assert(!pPage->cModifications \|\| ++cModifiedPages);
709	Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
710	Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX \|\| pPage->cModifications);
711	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
712	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
713	pPage->cModifications = 0;
714	}
715	}
716
717	#ifndef DEBUG_michael
718	AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
719	#endif
720	pPool->iModifiedHead = NIL_PGMPOOL_IDX;
721	pPool->cModifiedPages = 0;
722
723	/*
724	* Clear all the GCPhys links and rebuild the phys ext free list.
725	*/
726	for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
727	pRam;
728	pRam = pRam->CTX_SUFF(pNext))
729	{
730	iPage = pRam->cb >> PAGE_SHIFT;
731	while (iPage-- > 0)
732	PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
733	}
734
735	pPool->iPhysExtFreeHead = 0;
736	PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
737	const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
738	for (unsigned i = 0; i < cMaxPhysExts; i++)
739	{
740	paPhysExts[i].iNext = i + 1;
741	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
742	paPhysExts[i].apte[0] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
743	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
744	paPhysExts[i].apte[1] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
745	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
746	paPhysExts[i].apte[2] = NIL_PGMPOOL_PHYSEXT_IDX_PTE;
747	}
748	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
749
750
751	#ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
752	/* Reset all dirty pages to reactivate the page monitoring. */
753	/* Note: we must do this after clearing all page references and shadow page tables as there might be stale references to
754	* recently removed MMIO ranges around that might otherwise end up asserting in pgmPoolTracDerefGCPhysHint
755	*/
756	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aDirtyPages); i++)
757	{
758	PPGMPOOLPAGE pPage;
759	unsigned idxPage;
760
761	if (pPool->aDirtyPages[i].uIdx == NIL_PGMPOOL_IDX)
762	continue;
763
764	idxPage = pPool->aDirtyPages[i].uIdx;
765	AssertRelease(idxPage != NIL_PGMPOOL_IDX);
766	pPage = &pPool->aPages[idxPage];
767	Assert(pPage->idx == idxPage);
768	Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX && pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
769
770	AssertMsg(pPage->fDirty, ("Page %RGp (slot=%d) not marked dirty!", pPage->GCPhys, i));
771
772	Log(("Reactivate dirty page %RGp\n", pPage->GCPhys));
773
774	/* First write protect the page again to catch all write accesses. (before checking for changes -> SMP) */
775	int rc = PGMHandlerPhysicalReset(pVM, pPage->GCPhys & PAGE_BASE_GC_MASK);
776	Assert(rc == VINF_SUCCESS);
777	pPage->fDirty = false;
778
779	pPool->aDirtyPages[i].uIdx = NIL_PGMPOOL_IDX;
780	}
781
782	/* Clear all dirty pages. */
783	pPool->idxFreeDirtyPage = 0;
784	pPool->cDirtyPages = 0;
785	#endif
786
787	/* Clear the PGM_SYNC_CLEAR_PGM_POOL flag on all VCPUs to prevent redundant flushes. */
788	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
789	pVM->aCpus[idCpu].pgm.s.fSyncFlags &= ~PGM_SYNC_CLEAR_PGM_POOL;
790
791	/* Flush job finished. */
792	VM_FF_CLEAR(pVM, VM_FF_PGM_POOL_FLUSH_PENDING);
793	pPool->cPresent = 0;
794	pgmUnlock(pVM);
795
796	PGM_INVL_ALL_VCPU_TLBS(pVM);
797
798	if (fpvFlushRemTbl)
799	for (VMCPUID idCpu = 0; idCpu < pVM->cCpus; idCpu++)
800	CPUMSetChangedFlags(&pVM->aCpus[idCpu], CPUM_CHANGED_GLOBAL_TLB_FLUSH);
801
802	STAM_PROFILE_STOP(&pPool->StatClearAll, c);
803	return VINF_SUCCESS;
804	}
805
806
807	/**
808	* Clears the shadow page pool.
809	*
810	* @param pVM The VM handle.
811	* @param fFlushRemTlb When set, the REM TLB is scheduled for flushing as
812	* well.
813	*/
814	void pgmR3PoolClearAll(PVM pVM, bool fFlushRemTlb)
815	{
816	int rc = VMMR3EmtRendezvous(pVM, VMMEMTRENDEZVOUS_FLAGS_TYPE_ONCE, pgmR3PoolClearAllRendezvous, &fFlushRemTlb);
817	AssertRC(rc);
818	}
819
820	/**
821	* Protect all pgm pool page table entries to monitor writes
822	*
823	* @param pVM The VM handle.
824	*
825	* Remark: assumes the caller will flush all TLBs (!!)
826	*/
827	void pgmR3PoolWriteProtectPages(PVM pVM)
828	{
829	Assert(PGMIsLockOwner(pVM));
830	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
831	unsigned cLeft = pPool->cUsedPages;
832	unsigned iPage = pPool->cCurPages;
833	while (--iPage >= PGMPOOL_IDX_FIRST)
834	{
835	PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
836	if ( pPage->GCPhys != NIL_RTGCPHYS
837	&& pPage->cPresent)
838	{
839	union
840	{
841	void *pv;
842	PX86PT pPT;
843	PPGMSHWPTPAE pPTPae;
844	PEPTPT pPTEpt;
845	} uShw;
846	uShw.pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
847
848	switch (pPage->enmKind)
849	{
850	/*
851	* We only care about shadow page tables.
852	*/
853	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
854	case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
855	case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
856	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPT->a); iShw++)
857	{
858	if (uShw.pPT->a[iShw].n.u1Present)
859	uShw.pPT->a[iShw].n.u1Write = 0;
860	}
861	break;
862
863	case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
864	case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
865	case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
866	case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
867	case PGMPOOLKIND_PAE_PT_FOR_PHYS:
868	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTPae->a); iShw++)
869	{
870	if (PGMSHWPTEPAE_IS_P(uShw.pPTPae->a[iShw]))
871	PGMSHWPTEPAE_SET_RO(uShw.pPTPae->a[iShw]);
872	}
873	break;
874
875	case PGMPOOLKIND_EPT_PT_FOR_PHYS:
876	for (unsigned iShw = 0; iShw < RT_ELEMENTS(uShw.pPTEpt->a); iShw++)
877	{
878	if (uShw.pPTEpt->a[iShw].n.u1Present)
879	uShw.pPTEpt->a[iShw].n.u1Write = 0;
880	}
881	break;
882
883	default:
884	break;
885	}
886	if (!--cLeft)
887	break;
888	}
889	}
890	}
891
892	#ifdef VBOX_WITH_DEBUGGER
893	/**
894	* The '.pgmpoolcheck' command.
895	*
896	* @returns VBox status.
897	* @param pCmd Pointer to the command descriptor (as registered).
898	* @param pCmdHlp Pointer to command helper functions.
899	* @param pVM Pointer to the current VM (if any).
900	* @param paArgs Pointer to (readonly) array of arguments.
901	* @param cArgs Number of arguments in the array.
902	*/
903	static DECLCALLBACK(int) pgmR3PoolCmdCheck(PCDBGCCMD pCmd, PDBGCCMDHLP pCmdHlp, PVM pVM, PCDBGCVAR paArgs, unsigned cArgs, PDBGCVAR pResult)
904	{
905	/*
906	* Validate input.
907	*/
908	if (!pVM)
909	return pCmdHlp->pfnPrintf(pCmdHlp, NULL, "error: The command requires a VM to be selected.\n");
910
911	PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
912
913	for (unsigned i = 0; i < pPool->cCurPages; i++)
914	{
915	PPGMPOOLPAGE pPage = &pPool->aPages[i];
916	bool fFirstMsg = true;
917
918	/* Todo: cover other paging modes too. */
919	if (pPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
920	{
921	PPGMSHWPTPAE pShwPT = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
922	{
923	PX86PTPAE pGstPT;
924	PGMPAGEMAPLOCK LockPage;
925	int rc = PGMPhysGCPhys2CCPtrReadOnly(pVM, pPage->GCPhys, (const void **)&pGstPT, &LockPage); AssertReleaseRC(rc);
926
927	/* Check if any PTEs are out of sync. */
928	for (unsigned j = 0; j < RT_ELEMENTS(pShwPT->a); j++)
929	{
930	if (PGMSHWPTEPAE_IS_P(pShwPT->a[j]))
931	{
932	RTHCPHYS HCPhys = NIL_RTHCPHYS;
933	rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pGstPT->a[j].u & X86_PTE_PAE_PG_MASK, &HCPhys);
934	if ( rc != VINF_SUCCESS
935	\|\| PGMSHWPTEPAE_GET_HCPHYS(pShwPT->a[j]) != HCPhys)
936	{
937	if (fFirstMsg)
938	{
939	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Check pool page %RGp\n", pPage->GCPhys);
940	fFirstMsg = false;
941	}
942	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);
943	}
944	else if ( PGMSHWPTEPAE_IS_RW(pShwPT->a[j])
945	&& !pGstPT->a[j].n.u1Write)
946	{
947	if (fFirstMsg)
948	{
949	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Check pool page %RGp\n", pPage->GCPhys);
950	fFirstMsg = false;
951	}
952	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);
953	}
954	}
955	}
956	PGMPhysReleasePageMappingLock(pVM, &LockPage);
957	}
958
959	/* Make sure this page table can't be written to from any shadow mapping. */
960	RTHCPHYS HCPhysPT = NIL_RTHCPHYS;
961	int rc = PGMPhysGCPhys2HCPhys(pPool->CTX_SUFF(pVM), pPage->GCPhys, &HCPhysPT);
962	AssertMsgRC(rc, ("PGMPhysGCPhys2HCPhys failed with rc=%d for %RGp\n", rc, pPage->GCPhys));
963	if (rc == VINF_SUCCESS)
964	{
965	for (unsigned j = 0; j < pPool->cCurPages; j++)
966	{
967	PPGMPOOLPAGE pTempPage = &pPool->aPages[j];
968
969	if (pTempPage->enmKind == PGMPOOLKIND_PAE_PT_FOR_PAE_PT)
970	{
971	PPGMSHWPTPAE pShwPT2 = (PPGMSHWPTPAE)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pTempPage);
972
973	for (unsigned k = 0; k < RT_ELEMENTS(pShwPT->a); k++)
974	{
975	if ( PGMSHWPTEPAE_IS_P_RW(pShwPT2->a[k])
976	# ifdef PGMPOOL_WITH_OPTIMIZED_DIRTY_PT
977	&& !pPage->fDirty
978	# endif
979	&& PGMSHWPTEPAE_GET_HCPHYS(pShwPT2->a[k]) == HCPhysPT)
980	{
981	if (fFirstMsg)
982	{
983	pCmdHlp->pfnPrintf(pCmdHlp, NULL, "Check pool page %RGp\n", pPage->GCPhys);
984	fFirstMsg = false;
985	}
986	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]));
987	}
988	}
989	}
990	}
991	}
992	}
993	}
994	return VINF_SUCCESS;
995	}
996	#endif /* VBOX_WITH_DEBUGGER */

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

source: vbox/trunk/src/VBox/VMM/PGMPool.cpp@ 32963

Download in other formats: