PGMPool.cpp@ 10140

Last change on this file since 10140 was 10032, checked in by vboxsync, 16 years ago
Nested paging updates
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 28.4 KB

Line
1	/* $Id: PGMPool.cpp 10032 2008-06-30 17:03:54Z vboxsync $ */
2	/** @file
3	* PGM Shadow Page Pool.
4	*/
5
6	/*
7	* Copyright (C) 2006-2007 Sun Microsystems, Inc.
8	*
9	* This file is part of VirtualBox Open Source Edition (OSE), as
10	* available from http://www.virtualbox.org. This file is free software;
11	* you can redistribute it and/or modify it under the terms of the GNU
12	* General Public License (GPL) as published by the Free Software
13	* Foundation, in version 2 as it comes in the "COPYING" file of the
14	* VirtualBox OSE distribution. VirtualBox OSE is distributed in the
15	* hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
16	*
17	* Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
18	* Clara, CA 95054 USA or visit http://www.sun.com if you need
19	* additional information or have any questions.
20	*/
21
22	/** @page pg_pgm_pool PGM Shadow Page Pool
23	*
24	* Motivations:
25	* -# Relationship between shadow page tables and physical guest pages. This
26	* should allow us to skip most of the global flushes now following access
27	* handler changes. The main expense is flushing shadow pages.
28	* -# Limit the pool size (currently it's kind of limitless IIRC).
29	* -# Allocate shadow pages from GC. Currently we're allocating at SyncCR3 time.
30	* -# Required for 64-bit guests.
31	* -# Combining the PD cache and page pool in order to simplify caching.
32	*
33	*
34	* @section sec_pgm_pool_outline Design Outline
35	*
36	* The shadow page pool tracks pages used for shadowing paging structures (i.e. page
37	* tables, page directory, page directory pointer table and page map level-4). Each
38	* page in the pool has an unique identifier. This identifier is used to link a guest
39	* physical page to a shadow PT. The identifier is a non-zero value and has a
40	* relativly low max value - say 14 bits. This makes it possible to fit it into the
41	* upper bits of the of the aHCPhys entries in the ram range.
42	*
43	* By restricting host physical memory to the first 48 bits (which is the announced
44	* physical memory range of the K8L chip (scheduled for 2008)), we can safely use the
45	* upper 16 bits for shadow page ID and reference counting.
46	*
47	* Now, it's possible for a page to be aliased, i.e. mapped by more than one PT or
48	* PD. This is solved by creating a list of physical cross reference extents when
49	* ever this happens. Each node in the list (extent) is can contain 3 page pool
50	* indexes. The list it self is chained using indexes into the paPhysExt array.
51	*
52	*
53	* @section sec_pgm_pool_life Life Cycle of a Shadow Page
54	*
55	* -# The SyncPT function requests a page from the pool.
56	* The request includes the kind of page it is (PT/PD, PAE/legacy), the
57	* address of the page it's shadowing, and more.
58	* -# The pool responds to the request by allocating a new page.
59	* When the cache is enabled, it will first check if it's in the cache.
60	* Should the pool be exhausted, one of two things can be done:
61	* -# Flush the whole pool and current CR3.
62	* -# Use the cache to find a page which can be flushed (~age).
63	* -# The SyncPT function will sync one or more pages and insert it into the
64	* shadow PD.
65	* -# The SyncPage function may sync more pages on a later \#PFs.
66	* -# The page is freed / flushed in SyncCR3 (perhaps) and some other cases.
67	* When caching is enabled, the page isn't flush but remains in the cache.
68	*
69	*
70	* @section sec_pgm_pool_impl Monitoring
71	*
72	* We always monitor PAGE_SIZE chunks of memory. When we've got multiple shadow
73	* pages for the same PAGE_SIZE of guest memory (PAE and mixed PD/PT) the pages
74	* sharing the monitor get linked using the iMonitoredNext/Prev. The head page
75	* is the pvUser to the access handlers.
76	*
77	*
78	* @section sec_pgm_pool_impl Implementation
79	*
80	* The pool will take pages from the MM page pool. The tracking data (attributes,
81	* bitmaps and so on) are allocated from the hypervisor heap. The pool content can
82	* be accessed both by using the page id and the physical address (HC). The former
83	* is managed by means of an array, the latter by an offset based AVL tree.
84	*
85	* Flushing of a pool page means that we iterate the content (we know what kind
86	* it is) and updates the link information in the ram range.
87	*
88	* ...
89	*/
90
91
92	/*******************************************************************************
93	* Header Files *
94	*******************************************************************************/
95	#define LOG_GROUP LOG_GROUP_PGM_POOL
96	#include <VBox/pgm.h>
97	#include <VBox/mm.h>
98	#include "PGMInternal.h"
99	#include <VBox/vm.h>
100
101	#include <VBox/log.h>
102	#include <VBox/err.h>
103	#include <iprt/asm.h>
104	#include <iprt/string.h>
105
106
107	/*******************************************************************************
108	* Internal Functions *
109	*******************************************************************************/
110	#ifdef PGMPOOL_WITH_MONITORING
111	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser);
112	#endif /* PGMPOOL_WITH_MONITORING */
113
114
115	/**
116	* Initalizes the pool
117	*
118	* @returns VBox status code.
119	* @param pVM The VM handle.
120	*/
121	int pgmR3PoolInit(PVM pVM)
122	{
123	/*
124	* Query Pool config.
125	*/
126	PCFGMNODE pCfg = CFGMR3GetChild(CFGMR3GetRoot(pVM), "/PGM/Pool");
127	uint16_t cMaxPages;
128	int rc = CFGMR3QueryU16(pCfg, "MaxPages", &cMaxPages);
129	if (rc == VERR_CFGM_VALUE_NOT_FOUND \|\| rc == VERR_CFGM_NO_PARENT)
130	cMaxPages = 4*_1M >> PAGE_SHIFT;
131	else if (VBOX_FAILURE(rc))
132	AssertRCReturn(rc, rc);
133	else
134	AssertMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
135	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
136	cMaxPages = RT_ALIGN(cMaxPages, 16);
137
138	uint16_t cMaxUsers;
139	rc = CFGMR3QueryU16(pCfg, "MaxUsers", &cMaxUsers);
140	if (rc == VERR_CFGM_VALUE_NOT_FOUND \|\| rc == VERR_CFGM_NO_PARENT)
141	cMaxUsers = cMaxPages * 2;
142	else if (VBOX_FAILURE(rc))
143	AssertRCReturn(rc, rc);
144	else
145	AssertMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
146	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
147
148	uint16_t cMaxPhysExts;
149	rc = CFGMR3QueryU16(pCfg, "MaxPhysExts", &cMaxPhysExts);
150	if (rc == VERR_CFGM_VALUE_NOT_FOUND \|\| rc == VERR_CFGM_NO_PARENT)
151	cMaxPhysExts = RT_MAX(cMaxPages * 2, PGMPOOL_IDX_LAST);
152	else if (VBOX_FAILURE(rc))
153	AssertRCReturn(rc, rc);
154	else
155	AssertMsgReturn(cMaxPhysExts >= 16 && cMaxPages <= PGMPOOL_IDX_LAST,
156	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxUsers), VERR_INVALID_PARAMETER);
157
158	bool fCacheEnabled;
159	rc = CFGMR3QueryBool(pCfg, "CacheEnabled", &fCacheEnabled);
160	if (rc == VERR_CFGM_VALUE_NOT_FOUND \|\| rc == VERR_CFGM_NO_PARENT)
161	fCacheEnabled = true;
162	else if (VBOX_FAILURE(rc))
163	AssertRCReturn(rc, rc);
164
165	Log(("pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
166	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
167
168	/*
169	* Allocate the data structures.
170	*/
171	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
172	#ifdef PGMPOOL_WITH_USER_TRACKING
173	cb += cMaxUsers * sizeof(PGMPOOLUSER);
174	#endif
175	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
176	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
177	#endif
178	PPGMPOOL pPool;
179	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
180	if (VBOX_FAILURE(rc))
181	return rc;
182	pVM->pgm.s.pPoolHC = pPool;
183	pVM->pgm.s.pPoolGC = MMHyperHC2GC(pVM, pPool);
184
185	/*
186	* Initialize it.
187	*/
188	pPool->pVMHC = pVM;
189	pPool->pVMGC = pVM->pVMGC;
190	pPool->cMaxPages = cMaxPages;
191	pPool->cCurPages = PGMPOOL_IDX_FIRST;
192	#ifdef PGMPOOL_WITH_USER_TRACKING
193	pPool->iUserFreeHead = 0;
194	pPool->cMaxUsers = cMaxUsers;
195	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
196	pPool->paUsersHC = paUsers;
197	pPool->paUsersGC = MMHyperHC2GC(pVM, paUsers);
198	for (unsigned i = 0; i < cMaxUsers; i++)
199	{
200	paUsers[i].iNext = i + 1;
201	paUsers[i].iUser = NIL_PGMPOOL_IDX;
202	paUsers[i].iUserTable = 0xfffffffe;
203	}
204	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
205	#endif
206	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
207	pPool->iPhysExtFreeHead = 0;
208	pPool->cMaxPhysExts = cMaxPhysExts;
209	PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
210	pPool->paPhysExtsHC = paPhysExts;
211	pPool->paPhysExtsGC = MMHyperHC2GC(pVM, paPhysExts);
212	for (unsigned i = 0; i < cMaxPhysExts; i++)
213	{
214	paPhysExts[i].iNext = i + 1;
215	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
216	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
217	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
218	}
219	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
220	#endif
221	#ifdef PGMPOOL_WITH_CACHE
222	for (unsigned i = 0; i < ELEMENTS(pPool->aiHash); i++)
223	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
224	pPool->iAgeHead = NIL_PGMPOOL_IDX;
225	pPool->iAgeTail = NIL_PGMPOOL_IDX;
226	pPool->fCacheEnabled = fCacheEnabled;
227	#endif
228	#ifdef PGMPOOL_WITH_MONITORING
229	pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
230	pPool->pszAccessHandler = "Guest Paging Access Handler";
231	#endif
232	pPool->HCPhysTree = 0;
233
234	/* The NIL entry. */
235	Assert(NIL_PGMPOOL_IDX == 0);
236	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
237
238	/* The Shadow 32-bit PD. (32 bits guest paging) */
239	pPool->aPages[PGMPOOL_IDX_PD].Core.Key = NIL_RTHCPHYS;
240	pPool->aPages[PGMPOOL_IDX_PD].GCPhys = NIL_RTGCPHYS;
241	pPool->aPages[PGMPOOL_IDX_PD].pvPageHC = pVM->pgm.s.pHC32BitPD;
242	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_ROOT_32BIT_PD;
243	pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
244
245	/* The Shadow PAE PDs. This is actually 4 pages! (32 bits guest paging) */
246	pPool->aPages[PGMPOOL_IDX_PAE_PD].Core.Key = NIL_RTHCPHYS;
247	pPool->aPages[PGMPOOL_IDX_PAE_PD].GCPhys = NIL_RTGCPHYS;
248	pPool->aPages[PGMPOOL_IDX_PAE_PD].pvPageHC = pVM->pgm.s.apHCPaePDs[0];
249	pPool->aPages[PGMPOOL_IDX_PAE_PD].enmKind = PGMPOOLKIND_ROOT_PAE_PD;
250	pPool->aPages[PGMPOOL_IDX_PAE_PD].idx = PGMPOOL_IDX_PAE_PD;
251
252	/* The Shadow PAE PDs for PAE guest mode. */
253	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
254	{
255	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].Core.Key = NIL_RTHCPHYS;
256	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].GCPhys = NIL_RTGCPHYS;
257	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].pvPageHC = pVM->pgm.s.apHCPaePDs[i];
258	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].enmKind = PGMPOOLKIND_PAE_PD_FOR_PAE_PD;
259	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].idx = PGMPOOL_IDX_PAE_PD_0 + i;
260	}
261
262	/* The Shadow PDPT. */
263	pPool->aPages[PGMPOOL_IDX_PDPT].Core.Key = NIL_RTHCPHYS;
264	pPool->aPages[PGMPOOL_IDX_PDPT].GCPhys = NIL_RTGCPHYS;
265	pPool->aPages[PGMPOOL_IDX_PDPT].pvPageHC = pVM->pgm.s.pHCPaePDPT;
266	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_ROOT_PDPT;
267	pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
268
269	/* The Shadow AMD64 CR3. */
270	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].Core.Key = NIL_RTHCPHYS;
271	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].GCPhys = NIL_RTGCPHYS;
272	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].pvPageHC = pVM->pgm.s.pHCPaePDPT; /* not used */
273	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].enmKind = PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4;
274	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].idx = PGMPOOL_IDX_AMD64_CR3;
275
276	/* The Shadow AMD64 CR3. */
277	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].Core.Key = NIL_RTHCPHYS;
278	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].GCPhys = NIL_RTGCPHYS;
279	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].pvPageHC = pVM->pgm.s.pHCNestedRoot;
280	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].enmKind = PGMPOOLKIND_ROOT_NESTED;
281	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].idx = PGMPOOL_IDX_NESTED_ROOT;
282
283	/*
284	* Set common stuff.
285	*/
286	for (unsigned iPage = 1; iPage < PGMPOOL_IDX_FIRST; iPage++)
287	{
288	pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
289	#ifdef PGMPOOL_WITH_USER_TRACKING
290	pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
291	#endif
292	#ifdef PGMPOOL_WITH_MONITORING
293	pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
294	pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
295	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
296	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
297	#endif
298	#ifdef PGMPOOL_WITH_CACHE
299	pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
300	pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
301	#endif
302	Assert(VALID_PTR(pPool->aPages[iPage].pvPageHC));
303	Assert(pPool->aPages[iPage].idx == iPage);
304	Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
305	Assert(!pPool->aPages[iPage].fSeenNonGlobal);
306	Assert(!pPool->aPages[iPage].fMonitored);
307	Assert(!pPool->aPages[iPage].fCached);
308	Assert(!pPool->aPages[iPage].fZeroed);
309	Assert(!pPool->aPages[iPage].fReusedFlushPending);
310	}
311
312	#ifdef VBOX_WITH_STATISTICS
313	/*
314	* Register statistics.
315	*/
316	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
317	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
318	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
319	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
320	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
321	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolClearAll.");
322	STAM_REG(pVM, &pPool->StatFlushAllInt, STAMTYPE_PROFILE, "/PGM/Pool/FlushAllInt", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushAllInt.");
323	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
324	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
325	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spend zeroing pages. Overlaps with Alloc.");
326	# ifdef PGMPOOL_WITH_USER_TRACKING
327	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
328	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
329	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackDeref.");
330	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPT.");
331	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
332	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
333	STAM_REG(pVM, &pPool->StatTrackFreeUpOneUser, STAMTYPE_COUNTER, "/PGM/Pool/Track/FreeUpOneUser", STAMUNIT_OCCURENCES, "The number of times we were out of user tracking records.");
334	# endif
335	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
336	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_OCCURENCES, "Profiling deref activity related tracking GC physical pages.");
337	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
338	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
339	# endif
340	# ifdef PGMPOOL_WITH_MONITORING
341	STAM_REG(pVM, &pPool->StatMonitorGC, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/GC", STAMUNIT_TICKS_PER_CALL, "Profiling the GC PT access handler.");
342	STAM_REG(pVM, &pPool->StatMonitorGCEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/GCEmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
343	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.");
344	STAM_REG(pVM, &pPool->StatMonitorGCFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/GCFork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
345	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).");
346	STAM_REG(pVM, &pPool->StatMonitorGCIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/GCIntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
347	STAM_REG(pVM, &pPool->StatMonitorGCIntrFailPatch2, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/GCIntrFailPatch2", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction during flushing.");
348	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.");
349	STAM_REG(pVM, &pPool->StatMonitorGCRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/GCRepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
350	STAM_REG(pVM, &pPool->StatMonitorHC, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/HC", STAMUNIT_TICKS_PER_CALL, "Profiling the HC PT access handler.");
351	STAM_REG(pVM, &pPool->StatMonitorHCEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/HCEmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
352	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.");
353	STAM_REG(pVM, &pPool->StatMonitorHCFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/HCFork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
354	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).");
355	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.");
356	STAM_REG(pVM, &pPool->StatMonitorHCRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/HCRepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
357	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.");
358	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
359	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
360	# endif
361	# ifdef PGMPOOL_WITH_CACHE
362	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
363	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
364	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!)");
365	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.");
366	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
367	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
368	# endif
369	#endif /* VBOX_WITH_STATISTICS */
370
371	return VINF_SUCCESS;
372	}
373
374
375	/**
376	* Relocate the page pool data.
377	*
378	* @param pVM The VM handle.
379	*/
380	void pgmR3PoolRelocate(PVM pVM)
381	{
382	pVM->pgm.s.pPoolGC = MMHyperHC2GC(pVM, pVM->pgm.s.pPoolHC);
383	pVM->pgm.s.pPoolHC->pVMGC = pVM->pVMGC;
384	#ifdef PGMPOOL_WITH_USER_TRACKING
385	pVM->pgm.s.pPoolHC->paUsersGC = MMHyperHC2GC(pVM, pVM->pgm.s.pPoolHC->paUsersHC);
386	#endif
387	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
388	pVM->pgm.s.pPoolHC->paPhysExtsGC = MMHyperHC2GC(pVM, pVM->pgm.s.pPoolHC->paPhysExtsHC);
389	#endif
390	#ifdef PGMPOOL_WITH_MONITORING
391	int rc = PDMR3GetSymbolGC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolHC->pfnAccessHandlerGC);
392	AssertReleaseRC(rc);
393	/* init order hack. */
394	if (!pVM->pgm.s.pPoolHC->pfnAccessHandlerR0)
395	{
396	rc = PDMR3GetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolHC->pfnAccessHandlerR0);
397	AssertReleaseRC(rc);
398	}
399	#endif
400	}
401
402
403	/**
404	* Reset notification.
405	*
406	* This will flush the pool.
407	* @param pVM The VM handle.
408	*/
409	void pgmR3PoolReset(PVM pVM)
410	{
411	pgmPoolFlushAll(pVM);
412	}
413
414
415	/**
416	* Grows the shadow page pool.
417	*
418	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
419	*
420	* @returns VBox status code.
421	* @param pVM The VM handle.
422	*/
423	PDMR3DECL(int) PGMR3PoolGrow(PVM pVM)
424	{
425	PPGMPOOL pPool = pVM->pgm.s.pPoolHC;
426	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_INTERNAL_ERROR);
427
428	/*
429	* How much to grow it by?
430	*/
431	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
432	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
433	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages.\n", cPages, cPages));
434
435	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
436	{
437	PPGMPOOLPAGE pPage = &pPool->aPages[i];
438
439	pPage->pvPageHC = MMR3PageAlloc(pVM);
440	if (!pPage->pvPageHC)
441	{
442	Log(("We're out of memory!! i=%d\n", i));
443	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
444	}
445	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageHC);
446	LogFlow(("PGMR3PoolGrow: insert page %VHp\n", pPage->Core.Key));
447	pPage->GCPhys = NIL_RTGCPHYS;
448	pPage->enmKind = PGMPOOLKIND_FREE;
449	pPage->idx = pPage - &pPool->aPages[0];
450	pPage->iNext = pPool->iFreeHead;
451	#ifdef PGMPOOL_WITH_USER_TRACKING
452	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
453	#endif
454	#ifdef PGMPOOL_WITH_MONITORING
455	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
456	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
457	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
458	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
459	#endif
460	#ifdef PGMPOOL_WITH_CACHE
461	pPage->iAgeNext = NIL_PGMPOOL_IDX;
462	pPage->iAgePrev = NIL_PGMPOOL_IDX;
463	#endif
464	/* commit it */
465	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
466	pPool->iFreeHead = i;
467	pPool->cCurPages = i + 1;
468	}
469
470	Assert(pPool->cCurPages <= pPool->cMaxPages);
471	return VINF_SUCCESS;
472	}
473
474
475	#ifdef PGMPOOL_WITH_MONITORING
476
477	/**
478	* Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
479	*
480	* @param pPool The pool.
481	* @param pPage The page.
482	*/
483	static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
484	{
485	/* for the present this should be safe enough I think... */
486	pgmLock(pPool->pVMHC);
487	if ( pPage->fReusedFlushPending
488	&& pPage->enmKind != PGMPOOLKIND_FREE)
489	pgmPoolFlushPage(pPool, pPage);
490	pgmUnlock(pPool->pVMHC);
491	}
492
493
494	/**
495	* \#PF Handler callback for PT write accesses.
496	*
497	* The handler can not raise any faults, it's mainly for monitoring write access
498	* to certain pages.
499	*
500	* @returns VINF_SUCCESS if the handler have carried out the operation.
501	* @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
502	* @param pVM VM Handle.
503	* @param GCPhys The physical address the guest is writing to.
504	* @param pvPhys The HC mapping of that address.
505	* @param pvBuf What the guest is reading/writing.
506	* @param cbBuf How much it's reading/writing.
507	* @param enmAccessType The access type.
508	* @param pvUser User argument.
509	*/
510	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser)
511	{
512	STAM_PROFILE_START(&pVM->pgm.s.pPoolHC->StatMonitorHC, a);
513	PPGMPOOL pPool = pVM->pgm.s.pPoolHC;
514	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
515	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%VGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
516	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
517
518	/*
519	* We don't have to be very sophisiticated about this since there are relativly few calls here.
520	* However, we must try our best to detect any non-cpu accesses (disk / networking).
521	*
522	* Just to make life more interesting, we'll have to deal with the async threads too.
523	* We cannot flush a page if we're in an async thread because of REM notifications.
524	*/
525	if (!VM_IS_EMT(pVM))
526	{
527	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
528	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
529	STAM_COUNTER_INC(&pPool->StatMonitorHCAsync);
530	if (!pPage->fReusedFlushPending)
531	{
532	int rc = VMR3ReqCallEx(pPool->pVMHC, NULL, 0, VMREQFLAGS_NO_WAIT \| VMREQFLAGS_VOID, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
533	AssertRCReturn(rc, rc);
534	pPage->fReusedFlushPending = true;
535	pPage->cModifications += 0x1000;
536	}
537	pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
538	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
539	while (cbBuf > 4)
540	{
541	cbBuf -= 4;
542	pvPhys = (uint8_t *)pvPhys + 4;
543	GCPhys += 4;
544	pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
545	}
546	STAM_PROFILE_STOP(&pPool->StatMonitorHC, a);
547	}
548	else if ( (pPage->fCR3Mix \|\| pPage->cModifications < 96) /* it's cheaper here. */
549	&& cbBuf <= 4)
550	{
551	/* Clear the shadow entry. */
552	if (!pPage->cModifications++)
553	pgmPoolMonitorModifiedInsert(pPool, pPage);
554	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
555	pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
556	STAM_PROFILE_STOP(&pPool->StatMonitorHC, a);
557	}
558	else
559	{
560	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
561	STAM_PROFILE_STOP_EX(&pPool->StatMonitorHC, &pPool->StatMonitorHCFlushPage, a);
562	}
563
564	return VINF_PGM_HANDLER_DO_DEFAULT;
565	}
566
567	#endif /* PGMPOOL_WITH_MONITORING */
568

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

source: vbox/trunk/src/VBox/VMM/PGMPool.cpp@ 10140

Download in other formats: