PGMPool.cpp@ 22493

Last change on this file since 22493 was 22473, checked in by vboxsync, 15 years ago
Disabled experiment with optimized dirty PTs.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 30.0 KB

Line
1	/* $Id: PGMPool.cpp 22473 2009-08-26 14:51:50Z 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 if necessary (default is kind of limitless).
29	* -# Allocate shadow pages from RC. We use to only do this in SyncCR3.
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.
37	* page tables, page directory, page directory pointer table and page map
38	* level-4). Each page in the pool has an unique identifier. This identifier is
39	* used to link a guest physical page to a shadow PT. The identifier is a
40	* non-zero value and has a relativly low max value - say 14 bits. This makes it
41	* possible to fit it into the upper bits of the of the aHCPhys entries in the
42	* ram range.
43	*
44	* By restricting host physical memory to the first 48 bits (which is the
45	* announced physical memory range of the K8L chip (scheduled for 2008)), we
46	* can safely use the upper 16 bits for shadow page ID and reference counting.
47	*
48	* Update: The 48 bit assumption will be lifted with the new physical memory
49	* management (PGMPAGE), so we won't have any trouble when someone stuffs 2TB
50	* into a box in some years.
51	*
52	* Now, it's possible for a page to be aliased, i.e. mapped by more than one PT
53	* or PD. This is solved by creating a list of physical cross reference extents
54	* when ever this happens. Each node in the list (extent) is can contain 3 page
55	* pool indexes. The list it self is chained using indexes into the paPhysExt
56	* array.
57	*
58	*
59	* @section sec_pgm_pool_life Life Cycle of a Shadow Page
60	*
61	* -# The SyncPT function requests a page from the pool.
62	* The request includes the kind of page it is (PT/PD, PAE/legacy), the
63	* address of the page it's shadowing, and more.
64	* -# The pool responds to the request by allocating a new page.
65	* When the cache is enabled, it will first check if it's in the cache.
66	* Should the pool be exhausted, one of two things can be done:
67	* -# Flush the whole pool and current CR3.
68	* -# Use the cache to find a page which can be flushed (~age).
69	* -# The SyncPT function will sync one or more pages and insert it into the
70	* shadow PD.
71	* -# The SyncPage function may sync more pages on a later \#PFs.
72	* -# The page is freed / flushed in SyncCR3 (perhaps) and some other cases.
73	* When caching is enabled, the page isn't flush but remains in the cache.
74	*
75	*
76	* @section sec_pgm_pool_impl Monitoring
77	*
78	* We always monitor PAGE_SIZE chunks of memory. When we've got multiple shadow
79	* pages for the same PAGE_SIZE of guest memory (PAE and mixed PD/PT) the pages
80	* sharing the monitor get linked using the iMonitoredNext/Prev. The head page
81	* is the pvUser to the access handlers.
82	*
83	*
84	* @section sec_pgm_pool_impl Implementation
85	*
86	* The pool will take pages from the MM page pool. The tracking data
87	* (attributes, bitmaps and so on) are allocated from the hypervisor heap. The
88	* pool content can be accessed both by using the page id and the physical
89	* address (HC). The former is managed by means of an array, the latter by an
90	* offset based AVL tree.
91	*
92	* Flushing of a pool page means that we iterate the content (we know what kind
93	* it is) and updates the link information in the ram range.
94	*
95	* ...
96	*/
97
98
99	/*******************************************************************************
100	* Header Files *
101	*******************************************************************************/
102	#define LOG_GROUP LOG_GROUP_PGM_POOL
103	#include <VBox/pgm.h>
104	#include <VBox/mm.h>
105	#include "PGMInternal.h"
106	#include <VBox/vm.h>
107
108	#include <VBox/log.h>
109	#include <VBox/err.h>
110	#include <iprt/asm.h>
111	#include <iprt/string.h>
112
113
114	/*******************************************************************************
115	* Internal Functions *
116	*******************************************************************************/
117	#ifdef PGMPOOL_WITH_MONITORING
118	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser);
119	#endif /* PGMPOOL_WITH_MONITORING */
120
121
122	/**
123	* Initalizes the pool
124	*
125	* @returns VBox status code.
126	* @param pVM The VM handle.
127	*/
128	int pgmR3PoolInit(PVM pVM)
129	{
130	AssertCompile(NIL_PGMPOOL_IDX == 0);
131	/* pPage->cLocked is an unsigned byte. */
132	AssertCompile(VMM_MAX_CPU_COUNT <= 255);
133
134	/*
135	* Query Pool config.
136	*/
137	PCFGMNODE pCfg = CFGMR3GetChild(CFGMR3GetRoot(pVM), "/PGM/Pool");
138
139	/** @cfgm{/PGM/Pool/MaxPages, uint16_t, #pages, 16, 0x3fff, 1024}
140	* The max size of the shadow page pool in pages. The pool will grow dynamically
141	* up to this limit.
142	*/
143	uint16_t cMaxPages;
144	int rc = CFGMR3QueryU16Def(pCfg, "MaxPages", &cMaxPages, 4*_1M >> PAGE_SHIFT);
145	AssertLogRelRCReturn(rc, rc);
146	AssertLogRelMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
147	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
148	cMaxPages = RT_ALIGN(cMaxPages, 16);
149
150	/** @cfgm{/PGM/Pool/MaxUsers, uint16_t, #users, MaxUsers, 32K, MaxPages*2}
151	* The max number of shadow page user tracking records. Each shadow page has
152	* zero of other shadow pages (or CR3s) that references it, or uses it if you
153	* like. The structures describing these relationships are allocated from a
154	* fixed sized pool. This configuration variable defines the pool size.
155	*/
156	uint16_t cMaxUsers;
157	rc = CFGMR3QueryU16Def(pCfg, "MaxUsers", &cMaxUsers, cMaxPages * 2);
158	AssertLogRelRCReturn(rc, rc);
159	AssertLogRelMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
160	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
161
162	/** @cfgm{/PGM/Pool/MaxPhysExts, uint16_t, #extents, 16, MaxPages * 2, MAX(MaxPages*2,0x3fff)}
163	* The max number of extents for tracking aliased guest pages.
164	*/
165	uint16_t cMaxPhysExts;
166	rc = CFGMR3QueryU16Def(pCfg, "MaxPhysExts", &cMaxPhysExts, RT_MAX(cMaxPages * 2, PGMPOOL_IDX_LAST));
167	AssertLogRelRCReturn(rc, rc);
168	AssertLogRelMsgReturn(cMaxPhysExts >= 16 && cMaxPages <= PGMPOOL_IDX_LAST,
169	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxPhysExts), VERR_INVALID_PARAMETER);
170
171	/** @cfgm{/PGM/Pool/ChacheEnabled, bool, true}
172	* Enables or disabling caching of shadow pages. Chaching means that we will try
173	* reuse shadow pages instead of recreating them everything SyncCR3, SyncPT or
174	* SyncPage requests one. When reusing a shadow page, we can save time
175	* reconstructing it and it's children.
176	*/
177	bool fCacheEnabled;
178	rc = CFGMR3QueryBoolDef(pCfg, "CacheEnabled", &fCacheEnabled, true);
179	AssertLogRelRCReturn(rc, rc);
180
181	Log(("pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
182	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
183
184	/*
185	* Allocate the data structures.
186	*/
187	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
188	#ifdef PGMPOOL_WITH_USER_TRACKING
189	cb += cMaxUsers * sizeof(PGMPOOLUSER);
190	#endif
191	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
192	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
193	#endif
194	PPGMPOOL pPool;
195	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
196	if (RT_FAILURE(rc))
197	return rc;
198	pVM->pgm.s.pPoolR3 = pPool;
199	pVM->pgm.s.pPoolR0 = MMHyperR3ToR0(pVM, pPool);
200	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pPool);
201
202	/*
203	* Initialize it.
204	*/
205	pPool->pVMR3 = pVM;
206	pPool->pVMR0 = pVM->pVMR0;
207	pPool->pVMRC = pVM->pVMRC;
208	pPool->cMaxPages = cMaxPages;
209	pPool->cCurPages = PGMPOOL_IDX_FIRST;
210	#ifdef PGMPOOL_WITH_USER_TRACKING
211	pPool->iUserFreeHead = 0;
212	pPool->cMaxUsers = cMaxUsers;
213	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
214	pPool->paUsersR3 = paUsers;
215	pPool->paUsersR0 = MMHyperR3ToR0(pVM, paUsers);
216	pPool->paUsersRC = MMHyperR3ToRC(pVM, paUsers);
217	for (unsigned i = 0; i < cMaxUsers; i++)
218	{
219	paUsers[i].iNext = i + 1;
220	paUsers[i].iUser = NIL_PGMPOOL_IDX;
221	paUsers[i].iUserTable = 0xfffffffe;
222	}
223	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
224	#endif
225	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
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].aidx[1] = NIL_PGMPOOL_IDX;
237	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
238	}
239	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
240	#endif
241	#ifdef PGMPOOL_WITH_CACHE
242	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
243	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
244	pPool->iAgeHead = NIL_PGMPOOL_IDX;
245	pPool->iAgeTail = NIL_PGMPOOL_IDX;
246	pPool->fCacheEnabled = fCacheEnabled;
247	#endif
248	#ifdef PGMPOOL_WITH_MONITORING
249	pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
250	pPool->pszAccessHandler = "Guest Paging Access Handler";
251	#endif
252	pPool->HCPhysTree = 0;
253
254	/* The NIL entry. */
255	Assert(NIL_PGMPOOL_IDX == 0);
256	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
257
258	/* The Shadow 32-bit PD. (32 bits guest paging) */
259	pPool->aPages[PGMPOOL_IDX_PD].Core.Key = NIL_RTHCPHYS;
260	pPool->aPages[PGMPOOL_IDX_PD].GCPhys = NIL_RTGCPHYS;
261	pPool->aPages[PGMPOOL_IDX_PD].pvPageR3 = 0;
262	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_32BIT_PD;
263	pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
264
265	/* The Shadow PDPT. */
266	pPool->aPages[PGMPOOL_IDX_PDPT].Core.Key = NIL_RTHCPHYS;
267	pPool->aPages[PGMPOOL_IDX_PDPT].GCPhys = NIL_RTGCPHYS;
268	pPool->aPages[PGMPOOL_IDX_PDPT].pvPageR3 = 0;
269	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_PAE_PDPT;
270	pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
271
272	/* The Shadow AMD64 CR3. */
273	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].Core.Key = NIL_RTHCPHYS;
274	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].GCPhys = NIL_RTGCPHYS;
275	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].pvPageR3 = 0;
276	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].enmKind = PGMPOOLKIND_64BIT_PML4;
277	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].idx = PGMPOOL_IDX_AMD64_CR3;
278
279	/* The Nested Paging CR3. */
280	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].Core.Key = NIL_RTHCPHYS;
281	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].GCPhys = NIL_RTGCPHYS;
282	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].pvPageR3 = 0;
283	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].enmKind = PGMPOOLKIND_ROOT_NESTED;
284	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].idx = PGMPOOL_IDX_NESTED_ROOT;
285
286	/*
287	* Set common stuff.
288	*/
289	for (unsigned iPage = 1; iPage < PGMPOOL_IDX_FIRST; iPage++)
290	{
291	pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
292	#ifdef PGMPOOL_WITH_USER_TRACKING
293	pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
294	#endif
295	#ifdef PGMPOOL_WITH_MONITORING
296	pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
297	pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
298	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
299	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
300	#endif
301	#ifdef PGMPOOL_WITH_CACHE
302	pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
303	pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
304	#endif
305	Assert(pPool->aPages[iPage].idx == iPage);
306	Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
307	Assert(!pPool->aPages[iPage].fSeenNonGlobal);
308	Assert(!pPool->aPages[iPage].fMonitored);
309	Assert(!pPool->aPages[iPage].fCached);
310	Assert(!pPool->aPages[iPage].fZeroed);
311	Assert(!pPool->aPages[iPage].fReusedFlushPending);
312	}
313
314	#ifdef VBOX_WITH_STATISTICS
315	/*
316	* Register statistics.
317	*/
318	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
319	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
320	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
321	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
322	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
323	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolClearAll.");
324	STAM_REG(pVM, &pPool->StatFlushAllInt, STAMTYPE_PROFILE, "/PGM/Pool/FlushAllInt", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushAllInt.");
325	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
326	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
327	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spent zeroing pages. Overlaps with Alloc.");
328	# ifdef PGMPOOL_WITH_USER_TRACKING
329	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
330	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
331	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackDeref.");
332	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPT.");
333	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
334	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
335	STAM_REG(pVM, &pPool->StatTrackFreeUpOneUser, STAMTYPE_COUNTER, "/PGM/Pool/Track/FreeUpOneUser", STAMUNIT_OCCURENCES, "The number of times we were out of user tracking records.");
336	# endif
337	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
338	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_OCCURENCES, "Profiling deref activity related tracking GC physical pages.");
339	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
340	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
341	# endif
342	# ifdef PGMPOOL_WITH_MONITORING
343	STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
344	STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
345	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.");
346	STAM_REG(pVM, &pPool->StatMonitorRZFlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
347	STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
348	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).");
349	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
350	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.");
351	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.");
352	STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
353	STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
354	STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
355	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.");
356	STAM_REG(pVM, &pPool->StatMonitorR3FlushReinit, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/FlushReinit", STAMUNIT_OCCURENCES, "Times we've detected a page table reinit.");
357	STAM_REG(pVM, &pPool->StatMonitorR3Fork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
358	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).");
359	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.");
360	STAM_REG(pVM, &pPool->StatMonitorR3RepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
361	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.");
362	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
363	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
364	STAM_REG(pVM, &pPool->StatResetDirtyPages, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Resets", STAMUNIT_OCCURENCES, "Times we've called pgmPoolResetDirtyPages (and there were dirty page).");
365	STAM_REG(pVM, &pPool->StatDirtyPage, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/Dirty/Pages", STAMUNIT_OCCURENCES, "Times we've called pgmPoolAddDirtyPage.");
366
367	# endif
368	# ifdef PGMPOOL_WITH_CACHE
369	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
370	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
371	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!)");
372	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.");
373	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
374	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
375	# endif
376	#endif /* VBOX_WITH_STATISTICS */
377
378	return VINF_SUCCESS;
379	}
380
381
382	/**
383	* Relocate the page pool data.
384	*
385	* @param pVM The VM handle.
386	*/
387	void pgmR3PoolRelocate(PVM pVM)
388	{
389	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
390	pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
391	#ifdef PGMPOOL_WITH_USER_TRACKING
392	pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
393	#endif
394	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
395	pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
396	#endif
397	#ifdef PGMPOOL_WITH_MONITORING
398	int rc = PDMR3LdrGetSymbolRC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerRC);
399	AssertReleaseRC(rc);
400	/* init order hack. */
401	if (!pVM->pgm.s.pPoolR3->pfnAccessHandlerR0)
402	{
403	rc = PDMR3LdrGetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerR0);
404	AssertReleaseRC(rc);
405	}
406	#endif
407	}
408
409
410	/**
411	* Grows the shadow page pool.
412	*
413	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
414	*
415	* @returns VBox status code.
416	* @param pVM The VM handle.
417	*/
418	VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
419	{
420	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
421	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_INTERNAL_ERROR);
422
423	pgmLock(pVM);
424
425	/*
426	* How much to grow it by?
427	*/
428	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
429	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
430	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages.\n", cPages, cPages));
431
432	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
433	{
434	PPGMPOOLPAGE pPage = &pPool->aPages[i];
435
436	/* Allocate all pages in low (below 4 GB) memory as 32 bits guests need a page table root in low memory. */
437	pPage->pvPageR3 = MMR3PageAllocLow(pVM);
438	if (!pPage->pvPageR3)
439	{
440	Log(("We're out of memory!! i=%d\n", i));
441	pgmUnlock(pVM);
442	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
443	}
444	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
445	pPage->GCPhys = NIL_RTGCPHYS;
446	pPage->enmKind = PGMPOOLKIND_FREE;
447	pPage->idx = pPage - &pPool->aPages[0];
448	LogFlow(("PGMR3PoolGrow: insert page #%#x - %RHp\n", pPage->idx, pPage->Core.Key));
449	pPage->iNext = pPool->iFreeHead;
450	#ifdef PGMPOOL_WITH_USER_TRACKING
451	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
452	#endif
453	#ifdef PGMPOOL_WITH_MONITORING
454	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
455	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
456	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
457	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
458	#endif
459	#ifdef PGMPOOL_WITH_CACHE
460	pPage->iAgeNext = NIL_PGMPOOL_IDX;
461	pPage->iAgePrev = NIL_PGMPOOL_IDX;
462	#endif
463	/* commit it */
464	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
465	pPool->iFreeHead = i;
466	pPool->cCurPages = i + 1;
467	}
468
469	pgmUnlock(pVM);
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->pVMR3);
487	if ( pPage->fReusedFlushPending
488	&& pPage->enmKind != PGMPOOLKIND_FREE)
489	pgmPoolFlushPage(pPool, pPage);
490	pgmUnlock(pPool->pVMR3);
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 has 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.pPoolR3->StatMonitorR3, a);
513	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
514	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
515	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%RGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
516	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
517
518	PVMCPU pVCpu = VMMGetCpu(pVM);
519
520	/*
521	* We don't have to be very sophisticated about this since there are relativly few calls here.
522	* However, we must try our best to detect any non-cpu accesses (disk / networking).
523	*
524	* Just to make life more interesting, we'll have to deal with the async threads too.
525	* We cannot flush a page if we're in an async thread because of REM notifications.
526	*/
527	pgmLock(pVM);
528	if (PHYS_PAGE_ADDRESS(GCPhys) != PHYS_PAGE_ADDRESS(pPage->GCPhys))
529	{
530	/* Pool page changed while we were waiting for the lock; ignore. */
531	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)));
532	pgmUnlock(pVM);
533	return VINF_PGM_HANDLER_DO_DEFAULT;
534	}
535
536	Assert(pPage->enmKind != PGMPOOLKIND_FREE);
537
538	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} */
539	{
540	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
541	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
542	STAM_COUNTER_INC(&pPool->StatMonitorR3Async);
543	if (!pPage->fReusedFlushPending)
544	{
545	pgmUnlock(pVM);
546	int rc = VMR3ReqCallEx(pPool->pVMR3, VMCPUID_ANY, NULL, 0, VMREQFLAGS_NO_WAIT \| VMREQFLAGS_VOID, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
547	AssertRCReturn(rc, rc);
548	pgmLock(pVM);
549	pPage->fReusedFlushPending = true;
550	pPage->cModifications += 0x1000;
551	}
552
553	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, NULL);
554	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
555	while (cbBuf > 4)
556	{
557	cbBuf -= 4;
558	pvPhys = (uint8_t *)pvPhys + 4;
559	GCPhys += 4;
560	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, NULL);
561	}
562	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
563	}
564	else if ( ( pPage->cModifications < 96 /* it's cheaper here. */
565	\|\| pgmPoolIsPageLocked(&pVM->pgm.s, pPage)
566	)
567	&& cbBuf <= 4)
568	{
569	/* Clear the shadow entry. */
570	if (!pPage->cModifications++)
571	pgmPoolMonitorModifiedInsert(pPool, pPage);
572	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
573	pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhys, pvPhys, NULL);
574	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
575	}
576	else
577	{
578	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
579	STAM_PROFILE_STOP_EX(&pPool->StatMonitorR3, &pPool->StatMonitorR3FlushPage, a);
580	}
581	pgmUnlock(pVM);
582	return VINF_PGM_HANDLER_DO_DEFAULT;
583	}
584
585	#endif /* PGMPOOL_WITH_MONITORING */
586

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

source: vbox/trunk/src/VBox/VMM/PGMPool.cpp@ 22493

Download in other formats: