PGMPool.cpp@ 17562

Last change on this file since 17562 was 17489, checked in by vboxsync, 16 years ago
Don't use a special root for nested paging anymore.
Property svn:eol-style set to `native` Property svn:keywords set to `Id`
File size: 30.4 KB

Line
1	/* $Id: PGMPool.cpp 17489 2009-03-06 16:35:33Z 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	/*
131	* Query Pool config.
132	*/
133	PCFGMNODE pCfg = CFGMR3GetChild(CFGMR3GetRoot(pVM), "/PGM/Pool");
134
135	/** @cfgm{/PGM/Pool/MaxPages, uint16_t, #pages, 16, 0x3fff, 1024}
136	* The max size of the shadow page pool in pages. The pool will grow dynamically
137	* up to this limit.
138	*/
139	uint16_t cMaxPages;
140	int rc = CFGMR3QueryU16Def(pCfg, "MaxPages", &cMaxPages, 4*_1M >> PAGE_SHIFT);
141	AssertLogRelRCReturn(rc, rc);
142	AssertLogRelMsgReturn(cMaxPages <= PGMPOOL_IDX_LAST && cMaxPages >= RT_ALIGN(PGMPOOL_IDX_FIRST, 16),
143	("cMaxPages=%u (%#x)\n", cMaxPages, cMaxPages), VERR_INVALID_PARAMETER);
144	cMaxPages = RT_ALIGN(cMaxPages, 16);
145
146	/** @cfgm{/PGM/Pool/MaxUsers, uint16_t, #users, MaxUsers, 32K, MaxPages*2}
147	* The max number of shadow page user tracking records. Each shadow page has
148	* zero of other shadow pages (or CR3s) that references it, or uses it if you
149	* like. The structures describing these relationships are allocated from a
150	* fixed sized pool. This configuration variable defines the pool size.
151	*/
152	uint16_t cMaxUsers;
153	rc = CFGMR3QueryU16Def(pCfg, "MaxUsers", &cMaxUsers, cMaxPages * 2);
154	AssertLogRelRCReturn(rc, rc);
155	AssertLogRelMsgReturn(cMaxUsers >= cMaxPages && cMaxPages <= _32K,
156	("cMaxUsers=%u (%#x)\n", cMaxUsers, cMaxUsers), VERR_INVALID_PARAMETER);
157
158	/** @cfgm{/PGM/Pool/MaxPhysExts, uint16_t, #extents, 16, MaxPages * 2, MAX(MaxPages*2,0x3fff)}
159	* The max number of extents for tracking aliased guest pages.
160	*/
161	uint16_t cMaxPhysExts;
162	rc = CFGMR3QueryU16Def(pCfg, "MaxPhysExts", &cMaxPhysExts, RT_MAX(cMaxPages * 2, PGMPOOL_IDX_LAST));
163	AssertLogRelRCReturn(rc, rc);
164	AssertLogRelMsgReturn(cMaxPhysExts >= 16 && cMaxPages <= PGMPOOL_IDX_LAST,
165	("cMaxPhysExts=%u (%#x)\n", cMaxPhysExts, cMaxPhysExts), VERR_INVALID_PARAMETER);
166
167	/** @cfgm{/PGM/Pool/ChacheEnabled, bool, true}
168	* Enables or disabling caching of shadow pages. Chaching means that we will try
169	* reuse shadow pages instead of recreating them everything SyncCR3, SyncPT or
170	* SyncPage requests one. When reusing a shadow page, we can save time
171	* reconstructing it and it's children.
172	*/
173	bool fCacheEnabled;
174	rc = CFGMR3QueryBoolDef(pCfg, "CacheEnabled", &fCacheEnabled, true);
175	AssertLogRelRCReturn(rc, rc);
176
177	Log(("pgmR3PoolInit: cMaxPages=%#RX16 cMaxUsers=%#RX16 cMaxPhysExts=%#RX16 fCacheEnable=%RTbool\n",
178	cMaxPages, cMaxUsers, cMaxPhysExts, fCacheEnabled));
179
180	/*
181	* Allocate the data structures.
182	*/
183	uint32_t cb = RT_OFFSETOF(PGMPOOL, aPages[cMaxPages]);
184	#ifdef PGMPOOL_WITH_USER_TRACKING
185	cb += cMaxUsers * sizeof(PGMPOOLUSER);
186	#endif
187	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
188	cb += cMaxPhysExts * sizeof(PGMPOOLPHYSEXT);
189	#endif
190	PPGMPOOL pPool;
191	rc = MMR3HyperAllocOnceNoRel(pVM, cb, 0, MM_TAG_PGM_POOL, (void **)&pPool);
192	if (RT_FAILURE(rc))
193	return rc;
194	pVM->pgm.s.pPoolR3 = pPool;
195	pVM->pgm.s.pPoolR0 = MMHyperR3ToR0(pVM, pPool);
196	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pPool);
197
198	/*
199	* Initialize it.
200	*/
201	pPool->pVMR3 = pVM;
202	pPool->pVMR0 = pVM->pVMR0;
203	pPool->pVMRC = pVM->pVMRC;
204	pPool->cMaxPages = cMaxPages;
205	pPool->cCurPages = PGMPOOL_IDX_FIRST;
206	#ifdef PGMPOOL_WITH_USER_TRACKING
207	pPool->iUserFreeHead = 0;
208	pPool->cMaxUsers = cMaxUsers;
209	PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
210	pPool->paUsersR3 = paUsers;
211	pPool->paUsersR0 = MMHyperR3ToR0(pVM, paUsers);
212	pPool->paUsersRC = MMHyperR3ToRC(pVM, paUsers);
213	for (unsigned i = 0; i < cMaxUsers; i++)
214	{
215	paUsers[i].iNext = i + 1;
216	paUsers[i].iUser = NIL_PGMPOOL_IDX;
217	paUsers[i].iUserTable = 0xfffffffe;
218	}
219	paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
220	#endif
221	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
222	pPool->iPhysExtFreeHead = 0;
223	pPool->cMaxPhysExts = cMaxPhysExts;
224	PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
225	pPool->paPhysExtsR3 = paPhysExts;
226	pPool->paPhysExtsR0 = MMHyperR3ToR0(pVM, paPhysExts);
227	pPool->paPhysExtsRC = MMHyperR3ToRC(pVM, paPhysExts);
228	for (unsigned i = 0; i < cMaxPhysExts; i++)
229	{
230	paPhysExts[i].iNext = i + 1;
231	paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
232	paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
233	paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
234	}
235	paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
236	#endif
237	#ifdef PGMPOOL_WITH_CACHE
238	for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
239	pPool->aiHash[i] = NIL_PGMPOOL_IDX;
240	pPool->iAgeHead = NIL_PGMPOOL_IDX;
241	pPool->iAgeTail = NIL_PGMPOOL_IDX;
242	pPool->fCacheEnabled = fCacheEnabled;
243	#endif
244	#ifdef PGMPOOL_WITH_MONITORING
245	pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
246	pPool->pszAccessHandler = "Guest Paging Access Handler";
247	#endif
248	pPool->HCPhysTree = 0;
249
250	/* The NIL entry. */
251	Assert(NIL_PGMPOOL_IDX == 0);
252	pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
253
254	/* The Shadow 32-bit PD. (32 bits guest paging) */
255	pPool->aPages[PGMPOOL_IDX_PD].Core.Key = NIL_RTHCPHYS;
256	pPool->aPages[PGMPOOL_IDX_PD].GCPhys = NIL_RTGCPHYS;
257	#ifdef VBOX_WITH_PGMPOOL_PAGING_ONLY
258	pPool->aPages[PGMPOOL_IDX_PD].pvPageR3 = 0;
259	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_32BIT_PD;
260	#else
261	pPool->aPages[PGMPOOL_IDX_PD].pvPageR3 = pVM->pgm.s.pShw32BitPdR3;
262	pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_ROOT_32BIT_PD;
263	#endif
264	pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
265
266	#ifndef VBOX_WITH_PGMPOOL_PAGING_ONLY
267	/* The Shadow PAE PDs. This is actually 4 pages! (32 bits guest paging) */
268	pPool->aPages[PGMPOOL_IDX_PAE_PD].Core.Key = NIL_RTHCPHYS;
269	pPool->aPages[PGMPOOL_IDX_PAE_PD].GCPhys = NIL_RTGCPHYS;
270	pPool->aPages[PGMPOOL_IDX_PAE_PD].pvPageR3 = pVM->pgm.s.apShwPaePDsR3[0];
271	pPool->aPages[PGMPOOL_IDX_PAE_PD].enmKind = PGMPOOLKIND_ROOT_PAE_PD;
272	pPool->aPages[PGMPOOL_IDX_PAE_PD].idx = PGMPOOL_IDX_PAE_PD;
273
274	/* The Shadow PAE PDs for PAE guest mode. */
275	for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
276	{
277	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].Core.Key = NIL_RTHCPHYS;
278	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].GCPhys = NIL_RTGCPHYS;
279	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].pvPageR3 = pVM->pgm.s.apShwPaePDsR3[i];
280	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].enmKind = PGMPOOLKIND_PAE_PD_FOR_PAE_PD;
281	pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].idx = PGMPOOL_IDX_PAE_PD_0 + i;
282	}
283	#endif
284
285	/* The Shadow PDPT. */
286	pPool->aPages[PGMPOOL_IDX_PDPT].Core.Key = NIL_RTHCPHYS;
287	pPool->aPages[PGMPOOL_IDX_PDPT].GCPhys = NIL_RTGCPHYS;
288	#ifdef VBOX_WITH_PGMPOOL_PAGING_ONLY
289	pPool->aPages[PGMPOOL_IDX_PDPT].pvPageR3 = 0;
290	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_PAE_PDPT;
291	#else
292	pPool->aPages[PGMPOOL_IDX_PDPT].pvPageR3 = pVM->pgm.s.pShwPaePdptR3;
293	pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_ROOT_PDPT;
294	#endif
295	pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
296
297	/* The Shadow AMD64 CR3. */
298	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].Core.Key = NIL_RTHCPHYS;
299	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].GCPhys = NIL_RTGCPHYS;
300	#ifdef VBOX_WITH_PGMPOOL_PAGING_ONLY
301	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].pvPageR3 = 0;
302	#else
303	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].pvPageR3 = pVM->pgm.s.pShwPaePdptR3; /* not used - isn't it wrong as well? */
304	#endif
305	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].enmKind = PGMPOOLKIND_64BIT_PML4;
306	pPool->aPages[PGMPOOL_IDX_AMD64_CR3].idx = PGMPOOL_IDX_AMD64_CR3;
307
308	/* The Nested Paging CR3. */
309	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].Core.Key = NIL_RTHCPHYS;
310	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].GCPhys = NIL_RTGCPHYS;
311	#ifdef VBOX_WITH_PGMPOOL_PAGING_ONLY
312	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].pvPageR3 = 0;
313	#else
314	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].pvPageR3 = pVM->pgm.s.pShwNestedRootR3;
315	#endif
316	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].enmKind = PGMPOOLKIND_ROOT_NESTED;
317	pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].idx = PGMPOOL_IDX_NESTED_ROOT;
318
319	/*
320	* Set common stuff.
321	*/
322	for (unsigned iPage = 1; iPage < PGMPOOL_IDX_FIRST; iPage++)
323	{
324	pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
325	#ifdef PGMPOOL_WITH_USER_TRACKING
326	pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
327	#endif
328	#ifdef PGMPOOL_WITH_MONITORING
329	pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
330	pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
331	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
332	pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
333	#endif
334	#ifdef PGMPOOL_WITH_CACHE
335	pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
336	pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
337	#endif
338	#ifndef VBOX_WITH_PGMPOOL_PAGING_ONLY
339	Assert(VALID_PTR(pPool->aPages[iPage].pvPageR3));
340	#endif
341	Assert(pPool->aPages[iPage].idx == iPage);
342	Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
343	Assert(!pPool->aPages[iPage].fSeenNonGlobal);
344	Assert(!pPool->aPages[iPage].fMonitored);
345	Assert(!pPool->aPages[iPage].fCached);
346	Assert(!pPool->aPages[iPage].fZeroed);
347	Assert(!pPool->aPages[iPage].fReusedFlushPending);
348	}
349
350	#ifdef VBOX_WITH_STATISTICS
351	/*
352	* Register statistics.
353	*/
354	STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
355	STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
356	STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
357	STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
358	STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
359	STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolClearAll.");
360	STAM_REG(pVM, &pPool->StatFlushAllInt, STAMTYPE_PROFILE, "/PGM/Pool/FlushAllInt", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushAllInt.");
361	STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
362	STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
363	STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spent zeroing pages. Overlaps with Alloc.");
364	# ifdef PGMPOOL_WITH_USER_TRACKING
365	STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
366	STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
367	STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackDeref.");
368	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPT.");
369	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
370	STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
371	STAM_REG(pVM, &pPool->StatTrackFreeUpOneUser, STAMTYPE_COUNTER, "/PGM/Pool/Track/FreeUpOneUser", STAMUNIT_OCCURENCES, "The number of times we were out of user tracking records.");
372	# endif
373	# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
374	STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_OCCURENCES, "Profiling deref activity related tracking GC physical pages.");
375	STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
376	STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
377	# endif
378	# ifdef PGMPOOL_WITH_MONITORING
379	STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
380	STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
381	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.");
382	STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
383	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).");
384	STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
385	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.");
386	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.");
387	STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
388	STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
389	STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
390	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.");
391	STAM_REG(pVM, &pPool->StatMonitorR3Fork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
392	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).");
393	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.");
394	STAM_REG(pVM, &pPool->StatMonitorR3RepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
395	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.");
396	STAM_REG(pVM, &pPool->cModifiedPages, STAMTYPE_U16, "/PGM/Pool/Monitor/cModifiedPages", STAMUNIT_PAGES, "The current cModifiedPages value.");
397	STAM_REG(pVM, &pPool->cModifiedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/Monitor/cModifiedPagesHigh", STAMUNIT_PAGES, "The high watermark for cModifiedPages.");
398	# endif
399	# ifdef PGMPOOL_WITH_CACHE
400	STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
401	STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
402	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!)");
403	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.");
404	STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
405	STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
406	# endif
407	#endif /* VBOX_WITH_STATISTICS */
408
409	return VINF_SUCCESS;
410	}
411
412
413	/**
414	* Relocate the page pool data.
415	*
416	* @param pVM The VM handle.
417	*/
418	void pgmR3PoolRelocate(PVM pVM)
419	{
420	pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
421	pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
422	#ifdef PGMPOOL_WITH_USER_TRACKING
423	pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
424	#endif
425	#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
426	pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
427	#endif
428	#ifdef PGMPOOL_WITH_MONITORING
429	int rc = PDMR3LdrGetSymbolRC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerRC);
430	AssertReleaseRC(rc);
431	/* init order hack. */
432	if (!pVM->pgm.s.pPoolR3->pfnAccessHandlerR0)
433	{
434	rc = PDMR3LdrGetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerR0);
435	AssertReleaseRC(rc);
436	}
437	#endif
438	}
439
440
441	/**
442	* Reset notification.
443	*
444	* This will flush the pool.
445	* @param pVM The VM handle.
446	*/
447	void pgmR3PoolReset(PVM pVM)
448	{
449	pgmPoolFlushAll(pVM);
450	}
451
452
453	/**
454	* Grows the shadow page pool.
455	*
456	* I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
457	*
458	* @returns VBox status code.
459	* @param pVM The VM handle.
460	*/
461	VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
462	{
463	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
464	AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_INTERNAL_ERROR);
465
466	/*
467	* How much to grow it by?
468	*/
469	uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
470	cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
471	LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages.\n", cPages, cPages));
472
473	for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
474	{
475	PPGMPOOLPAGE pPage = &pPool->aPages[i];
476
477	#ifdef VBOX_WITH_PGMPOOL_PAGING_ONLY
478	/* Allocate all pages in low (below 4 GB) memory as 32 bits guests need a page table root in low memory. */
479	pPage->pvPageR3 = MMR3PageAllocLow(pVM);
480	#else
481	pPage->pvPageR3 = MMR3PageAlloc(pVM);
482	#endif
483	if (!pPage->pvPageR3)
484	{
485	Log(("We're out of memory!! i=%d\n", i));
486	return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
487	}
488	pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
489	pPage->GCPhys = NIL_RTGCPHYS;
490	pPage->enmKind = PGMPOOLKIND_FREE;
491	pPage->idx = pPage - &pPool->aPages[0];
492	LogFlow(("PGMR3PoolGrow: insert page #%#x - %RHp\n", pPage->idx, pPage->Core.Key));
493	pPage->iNext = pPool->iFreeHead;
494	#ifdef PGMPOOL_WITH_USER_TRACKING
495	pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
496	#endif
497	#ifdef PGMPOOL_WITH_MONITORING
498	pPage->iModifiedNext = NIL_PGMPOOL_IDX;
499	pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
500	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
501	pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
502	#endif
503	#ifdef PGMPOOL_WITH_CACHE
504	pPage->iAgeNext = NIL_PGMPOOL_IDX;
505	pPage->iAgePrev = NIL_PGMPOOL_IDX;
506	#endif
507	/* commit it */
508	bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
509	pPool->iFreeHead = i;
510	pPool->cCurPages = i + 1;
511	}
512
513	Assert(pPool->cCurPages <= pPool->cMaxPages);
514	return VINF_SUCCESS;
515	}
516
517
518	#ifdef PGMPOOL_WITH_MONITORING
519
520	/**
521	* Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
522	*
523	* @param pPool The pool.
524	* @param pPage The page.
525	*/
526	static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
527	{
528	/* for the present this should be safe enough I think... */
529	pgmLock(pPool->pVMR3);
530	if ( pPage->fReusedFlushPending
531	&& pPage->enmKind != PGMPOOLKIND_FREE)
532	pgmPoolFlushPage(pPool, pPage);
533	pgmUnlock(pPool->pVMR3);
534	}
535
536
537	/**
538	* \#PF Handler callback for PT write accesses.
539	*
540	* The handler can not raise any faults, it's mainly for monitoring write access
541	* to certain pages.
542	*
543	* @returns VINF_SUCCESS if the handler have carried out the operation.
544	* @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
545	* @param pVM VM Handle.
546	* @param GCPhys The physical address the guest is writing to.
547	* @param pvPhys The HC mapping of that address.
548	* @param pvBuf What the guest is reading/writing.
549	* @param cbBuf How much it's reading/writing.
550	* @param enmAccessType The access type.
551	* @param pvUser User argument.
552	*/
553	static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void pvPhys, void pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser)
554	{
555	STAM_PROFILE_START(&pVM->pgm.s.pPoolR3->StatMonitorR3, a);
556	PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
557	PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
558	LogFlow(("pgmR3PoolAccessHandler: GCPhys=%RGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
559	GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
560
561	/*
562	* We don't have to be very sophisiticated about this since there are relativly few calls here.
563	* However, we must try our best to detect any non-cpu accesses (disk / networking).
564	*
565	* Just to make life more interesting, we'll have to deal with the async threads too.
566	* We cannot flush a page if we're in an async thread because of REM notifications.
567	*/
568	if (!VM_IS_EMT(pVM))
569	{
570	Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
571	pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
572	STAM_COUNTER_INC(&pPool->StatMonitorR3Async);
573	if (!pPage->fReusedFlushPending)
574	{
575	int rc = VMR3ReqCallEx(pPool->pVMR3, VMREQDEST_ANY, NULL, 0, VMREQFLAGS_NO_WAIT \| VMREQFLAGS_VOID, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
576	AssertRCReturn(rc, rc);
577	pPage->fReusedFlushPending = true;
578	pPage->cModifications += 0x1000;
579	}
580	pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
581	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
582	while (cbBuf > 4)
583	{
584	cbBuf -= 4;
585	pvPhys = (uint8_t *)pvPhys + 4;
586	GCPhys += 4;
587	pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
588	}
589	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
590	}
591	else if ( ( pPage->cModifications < 96 /* it's cheaper here. */
592	#ifdef VBOX_WITH_PGMPOOL_PAGING_ONLY
593	\|\| pgmPoolIsPageLocked(&pVM->pgm.s, pPage)
594	#else
595	\|\| pPage->fCR3Mix
596	#endif
597	)
598	&& cbBuf <= 4)
599	{
600	/* Clear the shadow entry. */
601	if (!pPage->cModifications++)
602	pgmPoolMonitorModifiedInsert(pPool, pPage);
603	/** @todo r=bird: making unsafe assumption about not crossing entries here! */
604	pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
605	STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
606	}
607	else
608	{
609	pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
610	STAM_PROFILE_STOP_EX(&pPool->StatMonitorR3, &pPool->StatMonitorR3FlushPage, a);
611	}
612
613	return VINF_PGM_HANDLER_DO_DEFAULT;
614	}
615
616	#endif /* PGMPOOL_WITH_MONITORING */
617

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

source: vbox/trunk/src/VBox/VMM/PGMPool.cpp@ 17562

Download in other formats: