PGMPool.cpp@ 7961

Last change on this file since 7961 was 7961, checked in by vboxsync, 17 years ago
Updates for PAE paging in raw mode
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 27.6 KB

Line
1	/* $Id: PGMPool.cpp 7961 2008-04-14 17:09:45Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 innotek GmbH
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 (currently it's kind of limitless IIRC).
25	* -# Allocate shadow pages from GC. Currently we're allocating at SyncCR3 time.
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. page
33	* tables, page directory, page directory pointer table and page map level-4). Each
34	* page in the pool has an unique identifier. This identifier is used to link a guest
35	* physical page to a shadow PT. The identifier is a non-zero value and has a
36	* relativly low max value - say 14 bits. This makes it possible to fit it into the
37	* upper bits of the of the aHCPhys entries in the ram range.
38	*
39	* By restricting host physical memory to the first 48 bits (which is the announced
40	* physical memory range of the K8L chip (scheduled for 2008)), we can safely use the
41	* upper 16 bits for shadow page ID and reference counting.
42	*
43	* Now, it's possible for a page to be aliased, i.e. mapped by more than one PT or
44	* PD. This is solved by creating a list of physical cross reference extents when
45	* ever this happens. Each node in the list (extent) is can contain 3 page pool
46	* indexes. The list it self is chained using indexes into the paPhysExt array.
47	*
48	*
49	* @section sec_pgm_pool_life Life Cycle of a Shadow Page
50	*
51	* -# The SyncPT function requests a page from the pool.
52	* The request includes the kind of page it is (PT/PD, PAE/legacy), the
53	* address of the page it's shadowing, and more.
54	* -# The pool responds to the request by allocating a new page.
55	* When the cache is enabled, it will first check if it's in the cache.
56	* Should the pool be exhausted, one of two things can be done:
57	* -# Flush the whole pool and current CR3.
58	* -# Use the cache to find a page which can be flushed (~age).
59	* -# The SyncPT function will sync one or more pages and insert it into the
60	* shadow PD.
61	* -# The SyncPage function may sync more pages on a later \#PFs.
62	* -# The page is freed / flushed in SyncCR3 (perhaps) and some other cases.
63	* When caching is enabled, the page isn't flush but remains in the cache.
64	*
65	*
66	* @section sec_pgm_pool_impl Monitoring
67	*
68	* We always monitor PAGE_SIZE chunks of memory. When we've got multiple shadow
69	* pages for the same PAGE_SIZE of guest memory (PAE and mixed PD/PT) the pages
70	* sharing the monitor get linked using the iMonitoredNext/Prev. The head page
71	* is the pvUser to the access handlers.
72	*
73	*
74	* @section sec_pgm_pool_impl Implementation
75	*
76	* The pool will take pages from the MM page pool. The tracking data (attributes,
77	* bitmaps and so on) are allocated from the hypervisor heap. The pool content can
78	* be accessed both by using the page id and the physical address (HC). The former
79	* is managed by means of an array, the latter by an offset based AVL tree.
80	*
81	* Flushing of a pool page means that we iterate the content (we know what kind
82	* it is) and updates the link information in the ram range.
83	*
84	* ...
85	*/
86
87
88	/*******************************************************************************
89	* Header Files *
90	*******************************************************************************/
91	#define LOG_GROUP LOG_GROUP_PGM_POOL
92	#include <VBox/pgm.h>
93	#include <VBox/mm.h>
94	#include "PGMInternal.h"
95	#include <VBox/vm.h>
96
97	#include <VBox/log.h>
98	#include <VBox/err.h>
99	#include <iprt/asm.h>
100	#include <iprt/string.h>
101
102
103	/*******************************************************************************
104	* Internal Functions *
105	*******************************************************************************/
106	#ifdef PGMPOOL_WITH_MONITORING
107	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser);
108	#endif /* PGMPOOL_WITH_MONITORING */
109
110
111	/**
112	* Initalizes the pool
113	*
114	* @returns VBox status code.
115	* @param pVM The VM handle.
116	*/
117	int pgmR3PoolInit(PVM pVM)
118	{
119	/*
120	* Query Pool config.
121	*/
122	PCFGMNODE pCfg = CFGMR3GetChild(CFGMR3GetRoot(pVM), "/PGM/Pool");
123	uint16_t cMaxPages;
124	int rc = CFGMR3QueryU16(pCfg, "MaxPages", &cMaxPages);
125	if (rc == VERR_CFGM_VALUE_NOT_FOUND \|\| rc == VERR_CFGM_NO_PARENT)
126	cMaxPages = 4*_1M >> PAGE_SHIFT;
127	else if (VBOX_FAILURE(rc))
128	AssertRCReturn(rc, rc);
129	else
130	AssertMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
131	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
132	cMaxPages = RT_ALIGN(cMaxPages, 16);
133
134	uint16_t cMaxUsers;
135	rc = CFGMR3QueryU16(pCfg, "MaxUsers", &cMaxUsers);
136	if (rc == VERR_CFGM_VALUE_NOT_FOUND \|\| rc == VERR_CFGM_NO_PARENT)
137	cMaxUsers = cMaxPages * 2;
138	else if (VBOX_FAILURE(rc))
139	AssertRCReturn(rc, rc);
140	else
141	AssertMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
142	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
143
144	uint16_t cMaxPhysExts;
145	rc = CFGMR3QueryU16(pCfg, "MaxPhysExts", &cMaxPhysExts);
146	if (rc == VERR_CFGM_VALUE_NOT_FOUND \|\| rc == VERR_CFGM_NO_PARENT)
147	cMaxPhysExts = RT_MAX(cMaxPages * 2, PGMPOOL_IDX_LAST);
148	else if (VBOX_FAILURE(rc))
149	AssertRCReturn(rc, rc);
150	else
151	AssertMsgReturn(cMaxPhysExts >= 16 && cMaxPages <= PGMPOOL_IDX_LAST,
152	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxUsers), VERR_INVALID_PARAMETER);
153
154	bool fCacheEnabled;
155	rc = CFGMR3QueryBool(pCfg, "CacheEnabled", &fCacheEnabled);
156	if (rc == VERR_CFGM_VALUE_NOT_FOUND \|\| rc == VERR_CFGM_NO_PARENT)
157	fCacheEnabled = true;
158	else if (VBOX_FAILURE(rc))
159	AssertRCReturn(rc, rc);
160
161	Log(("pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
162	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
163
164	/*
165	* Allocate the data structures.
166	*/
167	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
168	#ifdef PGMPOOL_WITH_USER_TRACKING
169	cb += cMaxUsers * sizeof(PGMPOOLUSER);
170	#endif
171	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
172	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
173	#endif
174	PPGMPOOL pPool;
175	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
176	if (VBOX_FAILURE(rc))
177	return rc;
178	pVM->pgm.s.pPoolHC = pPool;
179	pVM->pgm.s.pPoolGC = MMHyperHC2GC(pVM, pPool);
180
181	/*
182	* Initialize it.
183	*/
184	pPool->pVMHC = pVM;
185	pPool->pVMGC = pVM->pVMGC;
186	pPool->cMaxPages = cMaxPages;
187	pPool->cCurPages = PGMPOOL_IDX_FIRST;
188	#ifdef PGMPOOL_WITH_USER_TRACKING
189	pPool->iUserFreeHead = 0;
190	pPool->cMaxUsers = cMaxUsers;
191	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
192	pPool->paUsersHC = paUsers;
193	pPool->paUsersGC = MMHyperHC2GC(pVM, paUsers);
194	for (unsigned i = 0; i < cMaxUsers; i++)
195	{
196	paUsers[i].iNext = i + 1;
197	paUsers[i].iUser = NIL_PGMPOOL_IDX;
198	paUsers[i].iUserTable = 0xfffe;
199	}
200	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
201	#endif
202	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
203	pPool->iPhysExtFreeHead = 0;
204	pPool->cMaxPhysExts = cMaxPhysExts;
205	PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
206	pPool->paPhysExtsHC = paPhysExts;
207	pPool->paPhysExtsGC = MMHyperHC2GC(pVM, paPhysExts);
208	for (unsigned i = 0; i < cMaxPhysExts; i++)
209	{
210	paPhysExts[i].iNext = i + 1;
211	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
212	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
213	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
214	}
215	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
216	#endif
217	#ifdef PGMPOOL_WITH_CACHE
218	for (unsigned i = 0; i < ELEMENTS(pPool->aiHash); i++)
219	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
220	pPool->iAgeHead = NIL_PGMPOOL_IDX;
221	pPool->iAgeTail = NIL_PGMPOOL_IDX;
222	pPool->fCacheEnabled = fCacheEnabled;
223	#endif
224	#ifdef PGMPOOL_WITH_MONITORING
225	pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
226	pPool->pszAccessHandler = "Guest Paging Access Handler";
227	#endif
228	pPool->HCPhysTree = 0;
229
230	/* The NIL entry. */
231	Assert(NIL_PGMPOOL_IDX == 0);
232	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
233
234	/* The Shadow 32-bit PD. (32 bits guest paging) */
235	pPool->aPages[PGMPOOL_IDX_PD].Core.Key = NIL_RTHCPHYS;
236	pPool->aPages[PGMPOOL_IDX_PD].GCPhys = NIL_RTGCPHYS;
237	pPool->aPages[PGMPOOL_IDX_PD].pvPageHC = pVM->pgm.s.pHC32BitPD;
238	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_ROOT_32BIT_PD;
239	pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
240
241	/* The Shadow PAE PDs. This is actually 4 pages! (32 bits guest paging) */
242	pPool->aPages[PGMPOOL_IDX_PAE_PD].Core.Key = NIL_RTHCPHYS;
243	pPool->aPages[PGMPOOL_IDX_PAE_PD].GCPhys = NIL_RTGCPHYS;
244	pPool->aPages[PGMPOOL_IDX_PAE_PD].pvPageHC = pVM->pgm.s.apHCPaePDs[0];
245	pPool->aPages[PGMPOOL_IDX_PAE_PD].enmKind = PGMPOOLKIND_ROOT_PAE_PD;
246	pPool->aPages[PGMPOOL_IDX_PAE_PD].idx = PGMPOOL_IDX_PAE_PD;
247
248	/* The Shadow PAE PDs for PAE guest mode. */
249	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
250	{
251	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].Core.Key = NIL_RTHCPHYS;
252	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].GCPhys = NIL_RTGCPHYS;
253	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].pvPageHC = pVM->pgm.s.apHCPaePDs[i];
254	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].enmKind = PGMPOOLKIND_PAE_PD_FOR_PAE_PD;
255	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].idx = PGMPOOL_IDX_PAE_PD_0 + i;
256	}
257
258	/* The Shadow PDPT. */
259	pPool->aPages[PGMPOOL_IDX_PDPT].Core.Key = NIL_RTHCPHYS;
260	pPool->aPages[PGMPOOL_IDX_PDPT].GCPhys = NIL_RTGCPHYS;
261	pPool->aPages[PGMPOOL_IDX_PDPT].pvPageHC = pVM->pgm.s.pHCPaePDPT;
262	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_ROOT_PDPT;
263	pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
264
265	/* The Shadow Page Map Level-4. */
266	pPool->aPages[PGMPOOL_IDX_PML4].Core.Key = NIL_RTHCPHYS;
267	pPool->aPages[PGMPOOL_IDX_PML4].GCPhys = NIL_RTGCPHYS;
268	pPool->aPages[PGMPOOL_IDX_PML4].pvPageHC = pVM->pgm.s.pHCPaePML4;
269	pPool->aPages[PGMPOOL_IDX_PML4].enmKind = PGMPOOLKIND_ROOT_PML4;
270	pPool->aPages[PGMPOOL_IDX_PML4].idx = PGMPOOL_IDX_PML4;
271
272	/*
273	* Set common stuff.
274	*/
275	for (unsigned iPage = 1; iPage < PGMPOOL_IDX_FIRST; iPage++)
276	{
277	pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
278	#ifdef PGMPOOL_WITH_USER_TRACKING
279	pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
280	#endif
281	#ifdef PGMPOOL_WITH_MONITORING
282	pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
283	pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
284	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
285	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
286	#endif
287	#ifdef PGMPOOL_WITH_CACHE
288	pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
289	pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
290	#endif
291	Assert(VALID_PTR(pPool->aPages[iPage].pvPageHC));
292	Assert(pPool->aPages[iPage].idx == iPage);
293	Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
294	Assert(!pPool->aPages[iPage].fSeenNonGlobal);
295	Assert(!pPool->aPages[iPage].fMonitored);
296	Assert(!pPool->aPages[iPage].fCached);
297	Assert(!pPool->aPages[iPage].fZeroed);
298	Assert(!pPool->aPages[iPage].fReusedFlushPending);
299	}
300
301	#ifdef VBOX_WITH_STATISTICS
302	/*
303	* Register statistics.
304	*/
305	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
306	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
307	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
308	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
309	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
310	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolClearAll.");
311	STAM_REG(pVM, &pPool->StatFlushAllInt, STAMTYPE_PROFILE, "/PGM/Pool/FlushAllInt", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushAllInt.");
312	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
313	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
314	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spend zeroing pages. Overlaps with Alloc.");
315	# ifdef PGMPOOL_WITH_USER_TRACKING
316	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
317	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
318	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackDeref.");
319	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPT.");
320	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
321	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
322	STAM_REG(pVM, &pPool->StatTrackFreeUpOneUser, STAMTYPE_COUNTER, "/PGM/Pool/Track/FreeUpOneUser", STAMUNIT_OCCURENCES, "The number of times we were out of user tracking records.");
323	# endif
324	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
325	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_OCCURENCES, "Profiling deref activity related tracking GC physical pages.");
326	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
327	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
328	# endif
329	# ifdef PGMPOOL_WITH_MONITORING
330	STAM_REG(pVM, &pPool->StatMonitorGC, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/GC", STAMUNIT_TICKS_PER_CALL, "Profiling the GC PT access handler.");
331	STAM_REG(pVM, &pPool->StatMonitorGCEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/GCEmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
332	STAM_REG(pVM, &pPool->StatMonitorGCFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/GCFlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling the pgmPoolFlushPage calls made from the GC PT access handler.");
333	STAM_REG(pVM, &pPool->StatMonitorGCFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/GCFork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
334	STAM_REG(pVM, &pPool->StatMonitorGCHandled, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/GCHandled", STAMUNIT_TICKS_PER_CALL, "Profiling the GC access we've handled (except REP STOSD).");
335	STAM_REG(pVM, &pPool->StatMonitorGCIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/GCIntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
336	STAM_REG(pVM, &pPool->StatMonitorGCIntrFailPatch2, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/GCIntrFailPatch2", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction during flushing.");
337	STAM_REG(pVM, &pPool->StatMonitorGCRepPrefix, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/GCRepPrefix", STAMUNIT_OCCURENCES, "The number of times we've seen rep prefixes we can't handle.");
338	STAM_REG(pVM, &pPool->StatMonitorGCRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/GCRepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
339	STAM_REG(pVM, &pPool->StatMonitorHC, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/HC", STAMUNIT_TICKS_PER_CALL, "Profiling the HC PT access handler.");
340	STAM_REG(pVM, &pPool->StatMonitorHCEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/HCEmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
341	STAM_REG(pVM, &pPool->StatMonitorHCFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/HCFlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling the pgmPoolFlushPage calls made from the HC PT access handler.");
342	STAM_REG(pVM, &pPool->StatMonitorHCFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/HCFork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
343	STAM_REG(pVM, &pPool->StatMonitorHCHandled, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/HCHandled", STAMUNIT_TICKS_PER_CALL, "Profiling the HC access we've handled (except REP STOSD).");
344	STAM_REG(pVM, &pPool->StatMonitorHCRepPrefix, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/HCRepPrefix", STAMUNIT_OCCURENCES, "The number of times we've seen rep prefixes we can't handle.");
345	STAM_REG(pVM, &pPool->StatMonitorHCRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/HCRepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
346	STAM_REG(pVM, &pPool->StatMonitorHCAsync, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/HCAsync", STAMUNIT_OCCURENCES, "Times we're called in an async thread and need to flush.");
347	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
348	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
349	# endif
350	# ifdef PGMPOOL_WITH_CACHE
351	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
352	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
353	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!)");
354	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.");
355	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
356	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
357	# endif
358	#endif /* VBOX_WITH_STATISTICS */
359
360	return VINF_SUCCESS;
361	}
362
363
364	/**
365	* Relocate the page pool data.
366	*
367	* @param pVM The VM handle.
368	*/
369	void pgmR3PoolRelocate(PVM pVM)
370	{
371	pVM->pgm.s.pPoolGC = MMHyperHC2GC(pVM, pVM->pgm.s.pPoolHC);
372	pVM->pgm.s.pPoolHC->pVMGC = pVM->pVMGC;
373	#ifdef PGMPOOL_WITH_USER_TRACKING
374	pVM->pgm.s.pPoolHC->paUsersGC = MMHyperHC2GC(pVM, pVM->pgm.s.pPoolHC->paUsersHC);
375	#endif
376	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
377	pVM->pgm.s.pPoolHC->paPhysExtsGC = MMHyperHC2GC(pVM, pVM->pgm.s.pPoolHC->paPhysExtsHC);
378	#endif
379	#ifdef PGMPOOL_WITH_MONITORING
380	int rc = PDMR3GetSymbolGC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolHC->pfnAccessHandlerGC);
381	AssertReleaseRC(rc);
382	/* init order hack. */
383	if (!pVM->pgm.s.pPoolHC->pfnAccessHandlerR0)
384	{
385	rc = PDMR3GetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolHC->pfnAccessHandlerR0);
386	AssertReleaseRC(rc);
387	}
388	#endif
389	}
390
391
392	/**
393	* Reset notification.
394	*
395	* This will flush the pool.
396	* @param pVM The VM handle.
397	*/
398	void pgmR3PoolReset(PVM pVM)
399	{
400	pgmPoolFlushAll(pVM);
401	}
402
403
404	/**
405	* Grows the shadow page pool.
406	*
407	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
408	*
409	* @returns VBox status code.
410	* @param pVM The VM handle.
411	*/
412	PDMR3DECL(int) PGMR3PoolGrow(PVM pVM)
413	{
414	PPGMPOOL pPool = pVM->pgm.s.pPoolHC;
415	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_INTERNAL_ERROR);
416
417	/*
418	* How much to grow it by?
419	*/
420	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
421	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
422	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages.\n", cPages, cPages));
423
424	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
425	{
426	PPGMPOOLPAGE pPage = &pPool->aPages[i];
427
428	pPage->pvPageHC = MMR3PageAlloc(pVM);
429	if (!pPage->pvPageHC)
430	{
431	Log(("We're out of memory!! i=%d\n", i));
432	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
433	}
434	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageHC);
435	pPage->GCPhys = NIL_RTGCPHYS;
436	pPage->enmKind = PGMPOOLKIND_FREE;
437	pPage->idx = pPage - &pPool->aPages[0];
438	pPage->iNext = pPool->iFreeHead;
439	#ifdef PGMPOOL_WITH_USER_TRACKING
440	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
441	#endif
442	#ifdef PGMPOOL_WITH_MONITORING
443	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
444	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
445	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
446	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
447	#endif
448	#ifdef PGMPOOL_WITH_CACHE
449	pPage->iAgeNext = NIL_PGMPOOL_IDX;
450	pPage->iAgePrev = NIL_PGMPOOL_IDX;
451	#endif
452	/* commit it */
453	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
454	pPool->iFreeHead = i;
455	pPool->cCurPages = i + 1;
456	}
457
458	Assert(pPool->cCurPages <= pPool->cMaxPages);
459	return VINF_SUCCESS;
460	}
461
462
463	#ifdef PGMPOOL_WITH_MONITORING
464
465	/**
466	* Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
467	*
468	* @param pPool The pool.
469	* @param pPage The page.
470	*/
471	static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
472	{
473	/* for the present this should be safe enough I think... */
474	pgmLock(pPool->pVMHC);
475	if ( pPage->fReusedFlushPending
476	&& pPage->enmKind != PGMPOOLKIND_FREE)
477	pgmPoolFlushPage(pPool, pPage);
478	pgmUnlock(pPool->pVMHC);
479	}
480
481
482	/**
483	* \#PF Handler callback for PT write accesses.
484	*
485	* The handler can not raise any faults, it's mainly for monitoring write access
486	* to certain pages.
487	*
488	* @returns VINF_SUCCESS if the handler have carried out the operation.
489	* @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
490	* @param pVM VM Handle.
491	* @param GCPhys The physical address the guest is writing to.
492	* @param pvPhys The HC mapping of that address.
493	* @param pvBuf What the guest is reading/writing.
494	* @param cbBuf How much it's reading/writing.
495	* @param enmAccessType The access type.
496	* @param pvUser User argument.
497	*/
498	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser)
499	{
500	STAM_PROFILE_START(&pVM->pgm.s.pPoolHC->StatMonitorHC, a);
501	PPGMPOOL pPool = pVM->pgm.s.pPoolHC;
502	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
503	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%VGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
504	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
505
506	/*
507	* We don't have to be very sophisiticated about this since there are relativly few calls here.
508	* However, we must try our best to detect any non-cpu accesses (disk / networking).
509	*
510	* Just to make life more interesting, we'll have to deal with the async threads too.
511	* We cannot flush a page if we're in an async thread because of REM notifications.
512	*/
513	if (!VM_IS_EMT(pVM))
514	{
515	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
516	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
517	STAM_COUNTER_INC(&pPool->StatMonitorHCAsync);
518	if (!pPage->fReusedFlushPending)
519	{
520	int rc = VMR3ReqCallEx(pPool->pVMHC, NULL, 0, VMREQFLAGS_NO_WAIT \| VMREQFLAGS_VOID, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
521	AssertRCReturn(rc, rc);
522	pPage->fReusedFlushPending = true;
523	pPage->cModifications += 0x1000;
524	}
525	pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
526	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
527	while (cbBuf > 4)
528	{
529	cbBuf -= 4;
530	pvPhys = (uint8_t *)pvPhys + 4;
531	GCPhys += 4;
532	pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
533	}
534	STAM_PROFILE_STOP(&pPool->StatMonitorHC, a);
535	}
536	else if ( (pPage->fCR3Mix \|\| pPage->cModifications < 96) /* it's cheaper here. */
537	&& cbBuf <= 4)
538	{
539	/* Clear the shadow entry. */
540	if (!pPage->cModifications++)
541	pgmPoolMonitorModifiedInsert(pPool, pPage);
542	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
543	pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
544	STAM_PROFILE_STOP(&pPool->StatMonitorHC, a);
545	}
546	else
547	{
548	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
549	STAM_PROFILE_STOP_EX(&pPool->StatMonitorHC, &pPool->StatMonitorHCFlushPage, a);
550	}
551
552	return VINF_PGM_HANDLER_DO_DEFAULT;
553	}
554
555	#endif /* PGMPOOL_WITH_MONITORING */
556

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

source: vbox/trunk/src/VBox/VMM/PGMPool.cpp@ 7961

Download in other formats: