VirtualBox

source: vbox/trunk/src/VBox/VMM/PGMPool.cpp@ 13538

Last change on this file since 13538 was 13135, checked in by vboxsync, 16 years ago

Wrong comment
  • Property svn:eol-style set to native
  • Property svn:keywords set to Id
File size: 28.6 KB
Line 
1/* $Id: PGMPool.cpp 13135 2008-10-09 14:36:48Z 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
111static 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 */
121int 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.pPoolR3 = pPool;
183 pVM->pgm.s.pPoolR0 = MMHyperR3ToR0(pVM, pPool);
184 pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pPool);
185
186 /*
187 * Initialize it.
188 */
189 pPool->pVMR3 = pVM;
190 pPool->pVMR0 = pVM->pVMR0;
191 pPool->pVMRC = pVM->pVMRC;
192 pPool->cMaxPages = cMaxPages;
193 pPool->cCurPages = PGMPOOL_IDX_FIRST;
194#ifdef PGMPOOL_WITH_USER_TRACKING
195 pPool->iUserFreeHead = 0;
196 pPool->cMaxUsers = cMaxUsers;
197 PPGMPOOLUSER paUsers = (PPGMPOOLUSER)&pPool->aPages[pPool->cMaxPages];
198 pPool->paUsersR3 = paUsers;
199 pPool->paUsersR0 = MMHyperR3ToR0(pVM, paUsers);
200 pPool->paUsersRC = MMHyperR3ToRC(pVM, paUsers);
201 for (unsigned i = 0; i < cMaxUsers; i++)
202 {
203 paUsers[i].iNext = i + 1;
204 paUsers[i].iUser = NIL_PGMPOOL_IDX;
205 paUsers[i].iUserTable = 0xfffffffe;
206 }
207 paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
208#endif
209#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
210 pPool->iPhysExtFreeHead = 0;
211 pPool->cMaxPhysExts = cMaxPhysExts;
212 PPGMPOOLPHYSEXT paPhysExts = (PPGMPOOLPHYSEXT)&paUsers[cMaxUsers];
213 pPool->paPhysExtsR3 = paPhysExts;
214 pPool->paPhysExtsR0 = MMHyperR3ToR0(pVM, paPhysExts);
215 pPool->paPhysExtsRC = MMHyperR3ToRC(pVM, paPhysExts);
216 for (unsigned i = 0; i < cMaxPhysExts; i++)
217 {
218 paPhysExts[i].iNext = i + 1;
219 paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
220 paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
221 paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
222 }
223 paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
224#endif
225#ifdef PGMPOOL_WITH_CACHE
226 for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
227 pPool->aiHash[i] = NIL_PGMPOOL_IDX;
228 pPool->iAgeHead = NIL_PGMPOOL_IDX;
229 pPool->iAgeTail = NIL_PGMPOOL_IDX;
230 pPool->fCacheEnabled = fCacheEnabled;
231#endif
232#ifdef PGMPOOL_WITH_MONITORING
233 pPool->pfnAccessHandlerR3 = pgmR3PoolAccessHandler;
234 pPool->pszAccessHandler = "Guest Paging Access Handler";
235#endif
236 pPool->HCPhysTree = 0;
237
238 /* The NIL entry. */
239 Assert(NIL_PGMPOOL_IDX == 0);
240 pPool->aPages[NIL_PGMPOOL_IDX].enmKind = PGMPOOLKIND_INVALID;
241
242 /* The Shadow 32-bit PD. (32 bits guest paging) */
243 pPool->aPages[PGMPOOL_IDX_PD].Core.Key = NIL_RTHCPHYS;
244 pPool->aPages[PGMPOOL_IDX_PD].GCPhys = NIL_RTGCPHYS;
245 pPool->aPages[PGMPOOL_IDX_PD].pvPageR3 = pVM->pgm.s.pHC32BitPD;
246 pPool->aPages[PGMPOOL_IDX_PD].enmKind = PGMPOOLKIND_ROOT_32BIT_PD;
247 pPool->aPages[PGMPOOL_IDX_PD].idx = PGMPOOL_IDX_PD;
248
249 /* The Shadow PAE PDs. This is actually 4 pages! (32 bits guest paging) */
250 pPool->aPages[PGMPOOL_IDX_PAE_PD].Core.Key = NIL_RTHCPHYS;
251 pPool->aPages[PGMPOOL_IDX_PAE_PD].GCPhys = NIL_RTGCPHYS;
252 pPool->aPages[PGMPOOL_IDX_PAE_PD].pvPageR3 = pVM->pgm.s.apHCPaePDs[0];
253 pPool->aPages[PGMPOOL_IDX_PAE_PD].enmKind = PGMPOOLKIND_ROOT_PAE_PD;
254 pPool->aPages[PGMPOOL_IDX_PAE_PD].idx = PGMPOOL_IDX_PAE_PD;
255
256 /* The Shadow PAE PDs for PAE guest mode. */
257 for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
258 {
259 pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].Core.Key = NIL_RTHCPHYS;
260 pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].GCPhys = NIL_RTGCPHYS;
261 pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].pvPageR3 = pVM->pgm.s.apHCPaePDs[i];
262 pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].enmKind = PGMPOOLKIND_PAE_PD_FOR_PAE_PD;
263 pPool->aPages[PGMPOOL_IDX_PAE_PD_0 + i].idx = PGMPOOL_IDX_PAE_PD_0 + i;
264 }
265
266 /* The Shadow PDPT. */
267 pPool->aPages[PGMPOOL_IDX_PDPT].Core.Key = NIL_RTHCPHYS;
268 pPool->aPages[PGMPOOL_IDX_PDPT].GCPhys = NIL_RTGCPHYS;
269 pPool->aPages[PGMPOOL_IDX_PDPT].pvPageR3 = pVM->pgm.s.pHCPaePDPT;
270 pPool->aPages[PGMPOOL_IDX_PDPT].enmKind = PGMPOOLKIND_ROOT_PDPT;
271 pPool->aPages[PGMPOOL_IDX_PDPT].idx = PGMPOOL_IDX_PDPT;
272
273 /* The Shadow AMD64 CR3. */
274 pPool->aPages[PGMPOOL_IDX_AMD64_CR3].Core.Key = NIL_RTHCPHYS;
275 pPool->aPages[PGMPOOL_IDX_AMD64_CR3].GCPhys = NIL_RTGCPHYS;
276 pPool->aPages[PGMPOOL_IDX_AMD64_CR3].pvPageR3 = pVM->pgm.s.pHCPaePDPT; /* not used */
277 pPool->aPages[PGMPOOL_IDX_AMD64_CR3].enmKind = PGMPOOLKIND_64BIT_PML4_FOR_64BIT_PML4;
278 pPool->aPages[PGMPOOL_IDX_AMD64_CR3].idx = PGMPOOL_IDX_AMD64_CR3;
279
280 /* The Nested Paging CR3. */
281 pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].Core.Key = NIL_RTHCPHYS;
282 pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].GCPhys = NIL_RTGCPHYS;
283 pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].pvPageR3 = pVM->pgm.s.pHCNestedRoot;
284 pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].enmKind = PGMPOOLKIND_ROOT_NESTED;
285 pPool->aPages[PGMPOOL_IDX_NESTED_ROOT].idx = PGMPOOL_IDX_NESTED_ROOT;
286
287 /*
288 * Set common stuff.
289 */
290 for (unsigned iPage = 1; iPage < PGMPOOL_IDX_FIRST; iPage++)
291 {
292 pPool->aPages[iPage].iNext = NIL_PGMPOOL_IDX;
293#ifdef PGMPOOL_WITH_USER_TRACKING
294 pPool->aPages[iPage].iUserHead = NIL_PGMPOOL_USER_INDEX;
295#endif
296#ifdef PGMPOOL_WITH_MONITORING
297 pPool->aPages[iPage].iModifiedNext = NIL_PGMPOOL_IDX;
298 pPool->aPages[iPage].iModifiedPrev = NIL_PGMPOOL_IDX;
299 pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
300 pPool->aPages[iPage].iMonitoredNext = NIL_PGMPOOL_IDX;
301#endif
302#ifdef PGMPOOL_WITH_CACHE
303 pPool->aPages[iPage].iAgeNext = NIL_PGMPOOL_IDX;
304 pPool->aPages[iPage].iAgePrev = NIL_PGMPOOL_IDX;
305#endif
306 Assert(VALID_PTR(pPool->aPages[iPage].pvPageR3));
307 Assert(pPool->aPages[iPage].idx == iPage);
308 Assert(pPool->aPages[iPage].GCPhys == NIL_RTGCPHYS);
309 Assert(!pPool->aPages[iPage].fSeenNonGlobal);
310 Assert(!pPool->aPages[iPage].fMonitored);
311 Assert(!pPool->aPages[iPage].fCached);
312 Assert(!pPool->aPages[iPage].fZeroed);
313 Assert(!pPool->aPages[iPage].fReusedFlushPending);
314 }
315
316#ifdef VBOX_WITH_STATISTICS
317 /*
318 * Register statistics.
319 */
320 STAM_REG(pVM, &pPool->cCurPages, STAMTYPE_U16, "/PGM/Pool/cCurPages", STAMUNIT_PAGES, "Current pool size.");
321 STAM_REG(pVM, &pPool->cMaxPages, STAMTYPE_U16, "/PGM/Pool/cMaxPages", STAMUNIT_PAGES, "Max pool size.");
322 STAM_REG(pVM, &pPool->cUsedPages, STAMTYPE_U16, "/PGM/Pool/cUsedPages", STAMUNIT_PAGES, "The number of pages currently in use.");
323 STAM_REG(pVM, &pPool->cUsedPagesHigh, STAMTYPE_U16_RESET, "/PGM/Pool/cUsedPagesHigh", STAMUNIT_PAGES, "The high watermark for cUsedPages.");
324 STAM_REG(pVM, &pPool->StatAlloc, STAMTYPE_PROFILE_ADV, "/PGM/Pool/Alloc", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolAlloc.");
325 STAM_REG(pVM, &pPool->StatClearAll, STAMTYPE_PROFILE, "/PGM/Pool/ClearAll", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolClearAll.");
326 STAM_REG(pVM, &pPool->StatFlushAllInt, STAMTYPE_PROFILE, "/PGM/Pool/FlushAllInt", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushAllInt.");
327 STAM_REG(pVM, &pPool->StatFlushPage, STAMTYPE_PROFILE, "/PGM/Pool/FlushPage", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFlushPage.");
328 STAM_REG(pVM, &pPool->StatFree, STAMTYPE_PROFILE, "/PGM/Pool/Free", STAMUNIT_TICKS_PER_CALL, "Profiling of pgmPoolFree.");
329 STAM_REG(pVM, &pPool->StatZeroPage, STAMTYPE_PROFILE, "/PGM/Pool/ZeroPage", STAMUNIT_TICKS_PER_CALL, "Profiling time spend zeroing pages. Overlaps with Alloc.");
330# ifdef PGMPOOL_WITH_USER_TRACKING
331 STAM_REG(pVM, &pPool->cMaxUsers, STAMTYPE_U16, "/PGM/Pool/Track/cMaxUsers", STAMUNIT_COUNT, "Max user tracking records.");
332 STAM_REG(pVM, &pPool->cPresent, STAMTYPE_U32, "/PGM/Pool/Track/cPresent", STAMUNIT_COUNT, "Number of present page table entries.");
333 STAM_REG(pVM, &pPool->StatTrackDeref, STAMTYPE_PROFILE, "/PGM/Pool/Track/Deref", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackDeref.");
334 STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPT, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPT", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPT.");
335 STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTs, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTs", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPTs.");
336 STAM_REG(pVM, &pPool->StatTrackFlushGCPhysPTsSlow, STAMTYPE_PROFILE, "/PGM/Pool/Track/FlushGCPhysPTsSlow", STAMUNIT_OCCURENCES, "Profiling of pgmPoolTrackFlushGCPhysPTsSlow.");
337 STAM_REG(pVM, &pPool->StatTrackFreeUpOneUser, STAMTYPE_COUNTER, "/PGM/Pool/Track/FreeUpOneUser", STAMUNIT_OCCURENCES, "The number of times we were out of user tracking records.");
338# endif
339# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
340 STAM_REG(pVM, &pPool->StatTrackDerefGCPhys, STAMTYPE_PROFILE, "/PGM/Pool/Track/DrefGCPhys", STAMUNIT_OCCURENCES, "Profiling deref activity related tracking GC physical pages.");
341 STAM_REG(pVM, &pPool->StatTrackLinearRamSearches, STAMTYPE_COUNTER, "/PGM/Pool/Track/LinearRamSearches", STAMUNIT_OCCURENCES, "The number of times we had to do linear ram searches.");
342 STAM_REG(pVM, &pPool->StamTrackPhysExtAllocFailures,STAMTYPE_COUNTER, "/PGM/Pool/Track/PhysExtAllocFailures", STAMUNIT_OCCURENCES, "The number of failing pgmPoolTrackPhysExtAlloc calls.");
343# endif
344# ifdef PGMPOOL_WITH_MONITORING
345 STAM_REG(pVM, &pPool->StatMonitorRZ, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ", STAMUNIT_TICKS_PER_CALL, "Profiling the RC/R0 access handler.");
346 STAM_REG(pVM, &pPool->StatMonitorRZEmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
347 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.");
348 STAM_REG(pVM, &pPool->StatMonitorRZFork, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/Fork", STAMUNIT_OCCURENCES, "Times we've detected fork().");
349 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).");
350 STAM_REG(pVM, &pPool->StatMonitorRZIntrFailPatch1, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/RZ/IntrFailPatch1", STAMUNIT_OCCURENCES, "Times we've failed interpreting a patch code instruction.");
351 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.");
352 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.");
353 STAM_REG(pVM, &pPool->StatMonitorRZRepStosd, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/RZ/RepStosd", STAMUNIT_TICKS_PER_CALL, "Profiling the REP STOSD cases we've handled.");
354 STAM_REG(pVM, &pPool->StatMonitorR3, STAMTYPE_PROFILE, "/PGM/Pool/Monitor/R3", STAMUNIT_TICKS_PER_CALL, "Profiling the R3 access handler.");
355 STAM_REG(pVM, &pPool->StatMonitorR3EmulateInstr, STAMTYPE_COUNTER, "/PGM/Pool/Monitor/R3/EmulateInstr", STAMUNIT_OCCURENCES, "Times we've failed interpreting the instruction.");
356 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.");
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# endif
365# ifdef PGMPOOL_WITH_CACHE
366 STAM_REG(pVM, &pPool->StatCacheHits, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Hits", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls satisfied by the cache.");
367 STAM_REG(pVM, &pPool->StatCacheMisses, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Misses", STAMUNIT_OCCURENCES, "The number of pgmPoolAlloc calls not statisfied by the cache.");
368 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!)");
369 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.");
370 STAM_REG(pVM, &pPool->StatCacheCacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Cacheable", STAMUNIT_OCCURENCES, "The number of cacheable allocations.");
371 STAM_REG(pVM, &pPool->StatCacheUncacheable, STAMTYPE_COUNTER, "/PGM/Pool/Cache/Uncacheable", STAMUNIT_OCCURENCES, "The number of uncacheable allocations.");
372# endif
373#endif /* VBOX_WITH_STATISTICS */
374
375 return VINF_SUCCESS;
376}
377
378
379/**
380 * Relocate the page pool data.
381 *
382 * @param pVM The VM handle.
383 */
384void pgmR3PoolRelocate(PVM pVM)
385{
386 pVM->pgm.s.pPoolRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3);
387 pVM->pgm.s.pPoolR3->pVMRC = pVM->pVMRC;
388#ifdef PGMPOOL_WITH_USER_TRACKING
389 pVM->pgm.s.pPoolR3->paUsersRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paUsersR3);
390#endif
391#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
392 pVM->pgm.s.pPoolR3->paPhysExtsRC = MMHyperR3ToRC(pVM, pVM->pgm.s.pPoolR3->paPhysExtsR3);
393#endif
394#ifdef PGMPOOL_WITH_MONITORING
395 int rc = PDMR3LdrGetSymbolRC(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerRC);
396 AssertReleaseRC(rc);
397 /* init order hack. */
398 if (!pVM->pgm.s.pPoolR3->pfnAccessHandlerR0)
399 {
400 rc = PDMR3LdrGetSymbolR0(pVM, NULL, "pgmPoolAccessHandler", &pVM->pgm.s.pPoolR3->pfnAccessHandlerR0);
401 AssertReleaseRC(rc);
402 }
403#endif
404}
405
406
407/**
408 * Reset notification.
409 *
410 * This will flush the pool.
411 * @param pVM The VM handle.
412 */
413void pgmR3PoolReset(PVM pVM)
414{
415 pgmPoolFlushAll(pVM);
416}
417
418
419/**
420 * Grows the shadow page pool.
421 *
422 * I.e. adds more pages to it, assuming that hasn't reached cMaxPages yet.
423 *
424 * @returns VBox status code.
425 * @param pVM The VM handle.
426 */
427VMMR3DECL(int) PGMR3PoolGrow(PVM pVM)
428{
429 PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
430 AssertReturn(pPool->cCurPages < pPool->cMaxPages, VERR_INTERNAL_ERROR);
431
432 /*
433 * How much to grow it by?
434 */
435 uint32_t cPages = pPool->cMaxPages - pPool->cCurPages;
436 cPages = RT_MIN(PGMPOOL_CFG_MAX_GROW, cPages);
437 LogFlow(("PGMR3PoolGrow: Growing the pool by %d (%#x) pages.\n", cPages, cPages));
438
439 for (unsigned i = pPool->cCurPages; cPages-- > 0; i++)
440 {
441 PPGMPOOLPAGE pPage = &pPool->aPages[i];
442
443 pPage->pvPageR3 = MMR3PageAlloc(pVM);
444 if (!pPage->pvPageR3)
445 {
446 Log(("We're out of memory!! i=%d\n", i));
447 return i ? VINF_SUCCESS : VERR_NO_PAGE_MEMORY;
448 }
449 pPage->Core.Key = MMPage2Phys(pVM, pPage->pvPageR3);
450 LogFlow(("PGMR3PoolGrow: insert page %VHp\n", pPage->Core.Key));
451 pPage->GCPhys = NIL_RTGCPHYS;
452 pPage->enmKind = PGMPOOLKIND_FREE;
453 pPage->idx = pPage - &pPool->aPages[0];
454 pPage->iNext = pPool->iFreeHead;
455#ifdef PGMPOOL_WITH_USER_TRACKING
456 pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
457#endif
458#ifdef PGMPOOL_WITH_MONITORING
459 pPage->iModifiedNext = NIL_PGMPOOL_IDX;
460 pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
461 pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
462 pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
463#endif
464#ifdef PGMPOOL_WITH_CACHE
465 pPage->iAgeNext = NIL_PGMPOOL_IDX;
466 pPage->iAgePrev = NIL_PGMPOOL_IDX;
467#endif
468 /* commit it */
469 bool fRc = RTAvloHCPhysInsert(&pPool->HCPhysTree, &pPage->Core); Assert(fRc); NOREF(fRc);
470 pPool->iFreeHead = i;
471 pPool->cCurPages = i + 1;
472 }
473
474 Assert(pPool->cCurPages <= pPool->cMaxPages);
475 return VINF_SUCCESS;
476}
477
478
479#ifdef PGMPOOL_WITH_MONITORING
480
481/**
482 * Worker used by pgmR3PoolAccessHandler when it's invoked by an async thread.
483 *
484 * @param pPool The pool.
485 * @param pPage The page.
486 */
487static DECLCALLBACK(void) pgmR3PoolFlushReusedPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
488{
489 /* for the present this should be safe enough I think... */
490 pgmLock(pPool->pVMR3);
491 if ( pPage->fReusedFlushPending
492 && pPage->enmKind != PGMPOOLKIND_FREE)
493 pgmPoolFlushPage(pPool, pPage);
494 pgmUnlock(pPool->pVMR3);
495}
496
497
498/**
499 * \#PF Handler callback for PT write accesses.
500 *
501 * The handler can not raise any faults, it's mainly for monitoring write access
502 * to certain pages.
503 *
504 * @returns VINF_SUCCESS if the handler have carried out the operation.
505 * @returns VINF_PGM_HANDLER_DO_DEFAULT if the caller should carry out the access operation.
506 * @param pVM VM Handle.
507 * @param GCPhys The physical address the guest is writing to.
508 * @param pvPhys The HC mapping of that address.
509 * @param pvBuf What the guest is reading/writing.
510 * @param cbBuf How much it's reading/writing.
511 * @param enmAccessType The access type.
512 * @param pvUser User argument.
513 */
514static DECLCALLBACK(int) pgmR3PoolAccessHandler(PVM pVM, RTGCPHYS GCPhys, void *pvPhys, void *pvBuf, size_t cbBuf, PGMACCESSTYPE enmAccessType, void *pvUser)
515{
516 STAM_PROFILE_START(&pVM->pgm.s.pPoolR3->StatMonitorR3, a);
517 PPGMPOOL pPool = pVM->pgm.s.pPoolR3;
518 PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
519 LogFlow(("pgmR3PoolAccessHandler: GCPhys=%VGp %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
520 GCPhys, pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
521
522 /*
523 * We don't have to be very sophisiticated about this since there are relativly few calls here.
524 * However, we must try our best to detect any non-cpu accesses (disk / networking).
525 *
526 * Just to make life more interesting, we'll have to deal with the async threads too.
527 * We cannot flush a page if we're in an async thread because of REM notifications.
528 */
529 if (!VM_IS_EMT(pVM))
530 {
531 Log(("pgmR3PoolAccessHandler: async thread, requesting EMT to flush the page: %p:{.Core=%RHp, .idx=%d, .GCPhys=%RGp, .enmType=%d}\n",
532 pPage, pPage->Core.Key, pPage->idx, pPage->GCPhys, pPage->enmKind));
533 STAM_COUNTER_INC(&pPool->StatMonitorR3Async);
534 if (!pPage->fReusedFlushPending)
535 {
536 int rc = VMR3ReqCallEx(pPool->pVMR3, NULL, 0, VMREQFLAGS_NO_WAIT | VMREQFLAGS_VOID, (PFNRT)pgmR3PoolFlushReusedPage, 2, pPool, pPage);
537 AssertRCReturn(rc, rc);
538 pPage->fReusedFlushPending = true;
539 pPage->cModifications += 0x1000;
540 }
541 pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
542 /** @todo r=bird: making unsafe assumption about not crossing entries here! */
543 while (cbBuf > 4)
544 {
545 cbBuf -= 4;
546 pvPhys = (uint8_t *)pvPhys + 4;
547 GCPhys += 4;
548 pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
549 }
550 STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
551 }
552 else if ( (pPage->fCR3Mix || pPage->cModifications < 96) /* it's cheaper here. */
553 && cbBuf <= 4)
554 {
555 /* Clear the shadow entry. */
556 if (!pPage->cModifications++)
557 pgmPoolMonitorModifiedInsert(pPool, pPage);
558 /** @todo r=bird: making unsafe assumption about not crossing entries here! */
559 pgmPoolMonitorChainChanging(pPool, pPage, GCPhys, pvPhys, NULL);
560 STAM_PROFILE_STOP(&pPool->StatMonitorR3, a);
561 }
562 else
563 {
564 pgmPoolMonitorChainFlush(pPool, pPage); /* ASSUME that VERR_PGM_POOL_CLEARED can be ignored here and that FFs will deal with it in due time. */
565 STAM_PROFILE_STOP_EX(&pPool->StatMonitorR3, &pPool->StatMonitorR3FlushPage, a);
566 }
567
568 return VINF_PGM_HANDLER_DO_DEFAULT;
569}
570
571#endif /* PGMPOOL_WITH_MONITORING */
572
Note: See TracBrowser for help on using the repository browser.

© 2023 Oracle
Contact – Privacy policy – Terms of Use