PGMAllPool.cpp revision a4eb337e9a5a6a72bcfe50134dc42732f5ee65c0
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync/* $Id$ */
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync/** @file
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * PGM Shadow Page Pool.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync */
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync/*
e64031e20c39650a7bc902a3e1aba613b9415deevboxsync * Copyright (C) 2006-2007 Sun Microsystems, Inc.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync *
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * This file is part of VirtualBox Open Source Edition (OSE), as
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * available from http://www.virtualbox.org. This file is free software;
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * you can redistribute it and/or modify it under the terms of the GNU
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * General Public License (GPL) as published by the Free Software
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * Foundation, in version 2 as it comes in the "COPYING" file of the
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * VirtualBox OSE distribution. VirtualBox OSE is distributed in the
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync *
0d12c7f9423f2745f8e282523d0930f91bff03b3vboxsync * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa
0d12c7f9423f2745f8e282523d0930f91bff03b3vboxsync * Clara, CA 95054 USA or visit http://www.sun.com if you need
0d12c7f9423f2745f8e282523d0930f91bff03b3vboxsync * additional information or have any questions.
0d12c7f9423f2745f8e282523d0930f91bff03b3vboxsync */
0d12c7f9423f2745f8e282523d0930f91bff03b3vboxsync
0d12c7f9423f2745f8e282523d0930f91bff03b3vboxsync
0d12c7f9423f2745f8e282523d0930f91bff03b3vboxsync/*******************************************************************************
0d12c7f9423f2745f8e282523d0930f91bff03b3vboxsync* Header Files *
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync*******************************************************************************/
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#define LOG_GROUP LOG_GROUP_PGM_POOL
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <VBox/pgm.h>
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <VBox/mm.h>
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <VBox/em.h>
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <VBox/cpum.h>
032a52c5b2984e26e84c2961f8f7f98a3954c8f2vboxsync#ifdef IN_RC
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync# include <VBox/patm.h>
590bfe12ce22cd3716448fbb9f4dc51664bfe5e2vboxsync#endif
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include "PGMInternal.h"
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <VBox/vm.h>
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <VBox/disopcode.h>
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <VBox/hwacc_vmx.h>
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <VBox/log.h>
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <VBox/err.h>
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <iprt/asm.h>
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#include <iprt/string.h>
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync/*******************************************************************************
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync* Internal Functions *
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync*******************************************************************************/
c2046db2cc346cc299f0cd9b2d1e160179159cfcvboxsync__BEGIN_DECLS
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsyncstatic void pgmPoolFlushAllInt(PPGMPOOL pPool);
c2046db2cc346cc299f0cd9b2d1e160179159cfcvboxsync#ifdef PGMPOOL_WITH_USER_TRACKING
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsyncDECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind);
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsyncDECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind);
c2046db2cc346cc299f0cd9b2d1e160179159cfcvboxsyncstatic void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#endif
c2046db2cc346cc299f0cd9b2d1e160179159cfcvboxsync#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
c2046db2cc346cc299f0cd9b2d1e160179159cfcvboxsyncstatic void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint);
c2046db2cc346cc299f0cd9b2d1e160179159cfcvboxsync#endif
c2046db2cc346cc299f0cd9b2d1e160179159cfcvboxsync#ifdef PGMPOOL_WITH_CACHE
c2046db2cc346cc299f0cd9b2d1e160179159cfcvboxsyncstatic int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable);
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#endif
044af0d1e6474076366759db86f101778c5f20ccvboxsync#ifdef PGMPOOL_WITH_MONITORING
044af0d1e6474076366759db86f101778c5f20ccvboxsyncstatic void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage);
044af0d1e6474076366759db86f101778c5f20ccvboxsync#endif
044af0d1e6474076366759db86f101778c5f20ccvboxsync#ifndef IN_RING3
044af0d1e6474076366759db86f101778c5f20ccvboxsyncDECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser);
044af0d1e6474076366759db86f101778c5f20ccvboxsync#endif
044af0d1e6474076366759db86f101778c5f20ccvboxsync#ifdef LOG_ENABLED
044af0d1e6474076366759db86f101778c5f20ccvboxsyncstatic const char *pgmPoolPoolKindToStr(uint8_t enmKind);
044af0d1e6474076366759db86f101778c5f20ccvboxsync#endif
044af0d1e6474076366759db86f101778c5f20ccvboxsync
044af0d1e6474076366759db86f101778c5f20ccvboxsyncvoid pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs);
044af0d1e6474076366759db86f101778c5f20ccvboxsyncvoid pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt);
044af0d1e6474076366759db86f101778c5f20ccvboxsyncint pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage);
044af0d1e6474076366759db86f101778c5f20ccvboxsyncPPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt);
044af0d1e6474076366759db86f101778c5f20ccvboxsyncvoid pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt);
044af0d1e6474076366759db86f101778c5f20ccvboxsyncvoid pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt);
044af0d1e6474076366759db86f101778c5f20ccvboxsync
044af0d1e6474076366759db86f101778c5f20ccvboxsync__END_DECLS
044af0d1e6474076366759db86f101778c5f20ccvboxsync
044af0d1e6474076366759db86f101778c5f20ccvboxsync
044af0d1e6474076366759db86f101778c5f20ccvboxsync/**
044af0d1e6474076366759db86f101778c5f20ccvboxsync * Checks if the specified page pool kind is for a 4MB or 2MB guest page.
044af0d1e6474076366759db86f101778c5f20ccvboxsync *
044af0d1e6474076366759db86f101778c5f20ccvboxsync * @returns true if it's the shadow of a 4MB or 2MB guest page, otherwise false.
044af0d1e6474076366759db86f101778c5f20ccvboxsync * @param enmKind The page kind.
044af0d1e6474076366759db86f101778c5f20ccvboxsync */
044af0d1e6474076366759db86f101778c5f20ccvboxsyncDECLINLINE(bool) pgmPoolIsBigPage(PGMPOOLKIND enmKind)
044af0d1e6474076366759db86f101778c5f20ccvboxsync{
044af0d1e6474076366759db86f101778c5f20ccvboxsync switch (enmKind)
044af0d1e6474076366759db86f101778c5f20ccvboxsync {
044af0d1e6474076366759db86f101778c5f20ccvboxsync case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
044af0d1e6474076366759db86f101778c5f20ccvboxsync case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync return true;
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync default:
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync return false;
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync }
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync}
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync/** @def PGMPOOL_PAGE_2_LOCKED_PTR
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * Maps a pool page pool into the current context and lock it (RC only).
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync *
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * @returns VBox status code.
044af0d1e6474076366759db86f101778c5f20ccvboxsync * @param pVM The VM handle.
044af0d1e6474076366759db86f101778c5f20ccvboxsync * @param pPage The pool page.
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync *
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * small page window employeed by that function. Be careful.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * @remark There is no need to assert on the result.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync */
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#if defined(IN_RC)
044af0d1e6474076366759db86f101778c5f20ccvboxsyncDECLINLINE(void *) PGMPOOL_PAGE_2_LOCKED_PTR(PVM pVM, PPGMPOOLPAGE pPage)
044af0d1e6474076366759db86f101778c5f20ccvboxsync{
044af0d1e6474076366759db86f101778c5f20ccvboxsync void *pv = pgmPoolMapPageInlined(&pVM->pgm.s, pPage);
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync /* Make sure the dynamic mapping will not be reused. */
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync if (pv)
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync PGMDynLockHCPage(pVM, (uint8_t *)pv);
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync return pv;
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync}
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync#else
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync# define PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage) PGMPOOL_PAGE_2_PTR(pVM, pPage)
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync#endif
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync/** @def PGMPOOL_UNLOCK_PTR
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * Unlock a previously locked dynamic caching (RC only).
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync *
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * @returns VBox status code.
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * @param pVM The VM handle.
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * @param pPage The pool page.
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync *
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * @remark In RC this uses PGMGCDynMapHCPage(), so it will consume of the
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * small page window employeed by that function. Be careful.
a1df400bbe9d64aad400442e56eb637019300a5evboxsync * @remark There is no need to assert on the result.
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync */
a1df400bbe9d64aad400442e56eb637019300a5evboxsync#if defined(IN_RC)
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsyncDECLINLINE(void) PGMPOOL_UNLOCK_PTR(PVM pVM, void *pvPage)
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync{
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync if (pvPage)
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync PGMDynUnlockHCPage(pVM, (uint8_t *)pvPage);
a1df400bbe9d64aad400442e56eb637019300a5evboxsync}
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync#else
a1df400bbe9d64aad400442e56eb637019300a5evboxsync# define PGMPOOL_UNLOCK_PTR(pVM, pPage) do {} while (0)
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync#endif
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync
044af0d1e6474076366759db86f101778c5f20ccvboxsync#ifdef PGMPOOL_WITH_MONITORING
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync/**
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * Determin the size of a write instruction.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * @returns number of bytes written.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * @param pDis The disassembler state.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync */
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsyncstatic unsigned pgmPoolDisasWriteSize(PDISCPUSTATE pDis)
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync{
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync /*
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * This is very crude and possibly wrong for some opcodes,
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * but since it's not really supposed to be called we can
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * probably live with that.
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync */
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync return DISGetParamSize(pDis, &pDis->param1);
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync}
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync/**
044af0d1e6474076366759db86f101778c5f20ccvboxsync * Flushes a chain of pages sharing the same access monitor.
044af0d1e6474076366759db86f101778c5f20ccvboxsync *
044af0d1e6474076366759db86f101778c5f20ccvboxsync * @returns VBox status code suitable for scheduling.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * @param pPool The pool.
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * @param pPage A page in the chain.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync */
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsyncint pgmPoolMonitorChainFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync{
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync LogFlow(("pgmPoolMonitorChainFlush: Flush page %RGp type=%d\n", pPage->GCPhys, pPage->enmKind));
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync /*
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync * Find the list head.
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync */
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync uint16_t idx = pPage->idx;
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync if (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync {
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync {
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync idx = pPage->iMonitoredPrev;
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync Assert(idx != pPage->idx);
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync pPage = &pPool->aPages[idx];
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync }
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync }
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync /*
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync * Iterate the list flushing each shadow page.
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync */
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync int rc = VINF_SUCCESS;
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync for (;;)
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync {
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync idx = pPage->iMonitoredNext;
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync Assert(idx != pPage->idx);
7d80dfbe5d66fc4c6de6fe109ce96a081496dcd4vboxsync if (pPage->idx >= PGMPOOL_IDX_FIRST)
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync {
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync int rc2 = pgmPoolFlushPage(pPool, pPage);
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync AssertRC(rc2);
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync }
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync /* next */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (idx == NIL_PGMPOOL_IDX)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pPage = &pPool->aPages[idx];
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync return rc;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync}
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync/**
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * Wrapper for getting the current context pointer to the entry being modified.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync *
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @returns VBox status code suitable for scheduling.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param pVM VM Handle.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param pvDst Destination address
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param pvSrc Source guest virtual address.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param GCPhysSrc The source guest physical address.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param cb Size of data to read
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsyncDECLINLINE(int) pgmPoolPhysSimpleReadGCPhys(PVM pVM, void *pvDst, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvSrc, RTGCPHYS GCPhysSrc, size_t cb)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync{
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync#if defined(IN_RING3)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync memcpy(pvDst, (RTHCPTR)((uintptr_t)pvSrc & ~(RTHCUINTPTR)(cb - 1)), cb);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync return VINF_SUCCESS;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync#else
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync /* @todo in RC we could attempt to use the virtual address, although this can cause many faults (PAE Windows XP guest). */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync return PGMPhysSimpleReadGCPhys(pVM, pvDst, GCPhysSrc & ~(RTGCPHYS)(cb - 1), cb);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync#endif
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync}
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync/**
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * Process shadow entries before they are changed by the guest.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync *
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * For PT entries we will clear them. For PD entries, we'll simply check
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * for mapping conflicts and set the SyncCR3 FF if found.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync *
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param pVCpu VMCPU handle
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param pPool The pool.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param pPage The head page.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param GCPhysFault The guest physical fault address.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param uAddress In R0 and GC this is the guest context fault address (flat).
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync * In R3 this is the host context 'fault' address.
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * @param pCpu The disassembler state for figuring out the write size.
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync * This need not be specified if the caller knows we won't do cross entry accesses.
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync */
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsyncvoid pgmPoolMonitorChainChanging(PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhysFault, CTXTYPE(RTGCPTR, RTHCPTR, RTGCPTR) pvAddress, PDISCPUSTATE pCpu)
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync{
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync const unsigned off = GCPhysFault & PAGE_OFFSET_MASK;
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync const unsigned cbWrite = (pCpu) ? pgmPoolDisasWriteSize(pCpu) : 0;
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync PVM pVM = pPool->CTX_SUFF(pVM);
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync LogFlow(("pgmPoolMonitorChainChanging: %RGv phys=%RGp kind=%s cbWrite=%d\n", (RTGCPTR)pvAddress, GCPhysFault, pgmPoolPoolKindToStr(pPage->enmKind), cbWrite));
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync for (;;)
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync {
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync union
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync void *pv;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync PX86PT pPT;
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync PX86PTPAE pPTPae;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync PX86PD pPD;
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync PX86PDPAE pPDPae;
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync PX86PDPT pPDPT;
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync PX86PML4 pPML4;
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync } uShw;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync uShw.pv = NULL;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync switch (pPage->enmKind)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync const unsigned iShw = off / sizeof(X86PTE);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT iShw=%x\n", iShw));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (uShw.pPT->a[iShw].n.u1Present)
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync {
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync X86PTE GstPte;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync AssertRC(rc);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync Log4(("pgmPoolMonitorChainChanging 32_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PG_MASK));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pgmPoolTracDerefGCPhysHint(pPool, pPage,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync GstPte.u & X86_PTE_PG_MASK);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPT->a[iShw].u = 0;
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync }
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync /* page/2 sized */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (!((off ^ pPage->GCPhys) & (PAGE_SIZE / 2)))
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync const unsigned iShw = (off / sizeof(X86PTE)) & (X86_PG_PAE_ENTRIES - 1);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("PGMPOOLKIND_PAE_PT_FOR_32BIT_PT iShw=%x\n", iShw));
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync if (uShw.pPTPae->a[iShw].n.u1Present)
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync X86PTE GstPte;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync AssertRC(rc);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync Log4(("pgmPoolMonitorChainChanging pae_32: deref %016RX64 GCPhys %08RX32\n", uShw.pPT->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PG_MASK));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pgmPoolTracDerefGCPhysHint(pPool, pPage,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync GstPte.u & X86_PTE_PG_MASK);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPTPae->a[iShw].u = 0;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync }
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync {
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync unsigned iGst = off / sizeof(X86PDE);
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync unsigned iShwPdpt = iGst / 256;
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync unsigned iShw = (iGst % 256) * 2;
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync LogFlow(("pgmPoolMonitorChainChanging PAE for 32 bits: iGst=%x iShw=%x idx = %d page idx=%d\n", iGst, iShw, iShwPdpt, pPage->enmKind - PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (iShwPdpt == pPage->enmKind - (unsigned)PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync for (unsigned i = 0; i < 2; i++)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# ifndef IN_RING0
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if ((uShw.pPDPae->a[iShw + i].u & (PGM_PDFLAGS_MAPPING | X86_PDE_P)) == (PGM_PDFLAGS_MAPPING | X86_PDE_P))
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw=%#x!\n", iShwPdpt, iShw+i));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync else
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync# endif /* !IN_RING0 */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (uShw.pPDPae->a[iShw+i].n.u1Present)
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw+i, uShw.pPDPae->a[iShw+i].u));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pgmPoolFree(pVM,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPDPae->a[iShw+i].u & X86_PDE_PAE_PG_MASK,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pPage->idx,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync iShw + i);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPDPae->a[iShw+i].u = 0;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync /* paranoia / a bit assumptive. */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if ( pCpu
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync && (off & 3)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync && (off & 3) + cbWrite > 4)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync const unsigned iShw2 = iShw + 2 + i;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (iShw2 < RT_ELEMENTS(uShw.pPDPae->a))
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync# ifndef IN_RING0
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if ((uShw.pPDPae->a[iShw2].u & (PGM_PDFLAGS_MAPPING | X86_PDE_P)) == (PGM_PDFLAGS_MAPPING | X86_PDE_P))
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShwPdpt=%#x iShw2=%#x!\n", iShwPdpt, iShw2));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync else
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif /* !IN_RING0 */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (uShw.pPDPae->a[iShw2].n.u1Present)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pgmPoolFree(pVM,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pPage->idx,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync iShw2);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPDPae->a[iShw2].u = 0;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync const unsigned iShw = off / sizeof(X86PTEPAE);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (uShw.pPTPae->a[iShw].n.u1Present)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync X86PTEPAE GstPte;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress, GCPhysFault, sizeof(GstPte));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync AssertRC(rc);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PAE_PG_MASK));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pgmPoolTracDerefGCPhysHint(pPool, pPage,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPTPae->a[iShw].u & X86_PTE_PAE_PG_MASK,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync GstPte.u & X86_PTE_PAE_PG_MASK);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPTPae->a[iShw].u = 0;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync /* paranoia / a bit assumptive. */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if ( pCpu
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync && (off & 7)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync && (off & 7) + cbWrite > sizeof(X86PTEPAE))
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTEPAE);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPTPae->a));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (uShw.pPTPae->a[iShw2].n.u1Present)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# ifdef PGMPOOL_WITH_GCPHYS_TRACKING
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync X86PTEPAE GstPte;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# ifdef IN_RING3
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, (RTHCPTR)((RTHCUINTPTR)pvAddress + sizeof(GstPte)), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# else
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync int rc = pgmPoolPhysSimpleReadGCPhys(pVM, &GstPte, pvAddress + sizeof(GstPte), GCPhysFault + sizeof(GstPte), sizeof(GstPte));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync AssertRC(rc);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync Log4(("pgmPoolMonitorChainChanging pae: deref %016RX64 GCPhys %016RX64\n", uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK, GstPte.u & X86_PTE_PAE_PG_MASK));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pgmPoolTracDerefGCPhysHint(pPool, pPage,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPTPae->a[iShw2].u & X86_PTE_PAE_PG_MASK,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync GstPte.u & X86_PTE_PAE_PG_MASK);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPTPae->a[iShw2].u = 0;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
ad27e1d5e48ca41245120c331cc88b50464813cevboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync case PGMPOOLKIND_32BIT_PD:
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync const unsigned iShw = off / sizeof(X86PTE); // ASSUMING 32-bit guest paging!
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync LogFlow(("pgmPoolMonitorChainChanging: PGMPOOLKIND_32BIT_PD %x\n", iShw));
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync# ifndef IN_RING0
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync if (uShw.pPD->a[iShw].u & PGM_PDFLAGS_MAPPING)
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync {
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync break;
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync }
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync# endif /* !IN_RING0 */
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync# ifndef IN_RING0
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync else
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync# endif /* !IN_RING0 */
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync {
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync if (uShw.pPD->a[iShw].n.u1Present)
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync {
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync pgmPoolFree(pVM,
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync uShw.pPD->a[iShw].u & X86_PDE_PAE_PG_MASK,
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync pPage->idx,
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync iShw);
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync uShw.pPD->a[iShw].u = 0;
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync }
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync }
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync /* paranoia / a bit assumptive. */
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync if ( pCpu
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync && (off & 3)
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync && (off & 3) + cbWrite > sizeof(X86PTE))
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync {
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PTE);
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync if ( iShw2 != iShw
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync && iShw2 < RT_ELEMENTS(uShw.pPD->a))
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync {
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync# ifndef IN_RING0
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync if (uShw.pPD->a[iShw2].u & PGM_PDFLAGS_MAPPING)
9b19ad593b379ebfcc8273f85b90763b14b1da63vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync# endif /* !IN_RING0 */
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync# ifndef IN_RING0
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync else
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif /* !IN_RING0 */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (uShw.pPD->a[iShw2].n.u1Present)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync LogFlow(("pgmPoolMonitorChainChanging: 32 bit pd iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPD->a[iShw2].u));
044af0d1e6474076366759db86f101778c5f20ccvboxsync pgmPoolFree(pVM,
044af0d1e6474076366759db86f101778c5f20ccvboxsync uShw.pPD->a[iShw2].u & X86_PDE_PAE_PG_MASK,
044af0d1e6474076366759db86f101778c5f20ccvboxsync pPage->idx,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync iShw2);
044af0d1e6474076366759db86f101778c5f20ccvboxsync uShw.pPD->a[iShw2].u = 0;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync#if 0 /* useful when running PGMAssertCR3(), a bit too troublesome for general use (TLBs). */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if ( uShw.pPD->a[iShw].n.u1Present
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync && !VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("pgmPoolMonitorChainChanging: iShw=%#x: %RX32 -> freeing it!\n", iShw, uShw.pPD->a[iShw].u));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# ifdef IN_RC /* TLB load - we're pushing things a bit... */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync ASMProbeReadByte(pvAddress);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif
88acfa6629a7976c0583c1712d2b5b22a87a5121vboxsync pgmPoolFree(pVM, uShw.pPD->a[iShw].u & X86_PDE_PG_MASK, pPage->idx, iShw);
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync uShw.pPD->a[iShw].u = 0;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
88acfa6629a7976c0583c1712d2b5b22a87a5121vboxsync#endif
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync const unsigned iShw = off / sizeof(X86PDEPAE);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync#ifndef IN_RING0
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
a1df400bbe9d64aad400442e56eb637019300a5evboxsync STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
a1df400bbe9d64aad400442e56eb637019300a5evboxsync LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw=%#x!\n", iShw));
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync#endif /* !IN_RING0 */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync /*
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * Causes trouble when the guest uses a PDE to refer to the whole page table level
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * structure. (Invalidate here; faults later on when it tries to change the page
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync * table entries -> recheck; probably only applies to the RC case.)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# ifndef IN_RING0
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync else
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif /* !IN_RING0 */
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync {
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync if (uShw.pPDPae->a[iShw].n.u1Present)
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync {
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync pgmPoolFree(pVM,
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pPage->idx,
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync iShw);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPDPae->a[iShw].u = 0;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync /* paranoia / a bit assumptive. */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if ( pCpu
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync && (off & 7)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync && (off & 7) + cbWrite > sizeof(X86PDEPAE))
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync#ifndef IN_RING0
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if ( iShw2 != iShw
9d020a0622f95aec3aabaff436a495e88dbbd71avboxsync && uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync#endif /* !IN_RING0 */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# ifndef IN_RING0
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync else
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif /* !IN_RING0 */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (uShw.pPDPae->a[iShw2].n.u1Present)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pgmPoolFree(pVM,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
044af0d1e6474076366759db86f101778c5f20ccvboxsync pPage->idx,
044af0d1e6474076366759db86f101778c5f20ccvboxsync iShw2);
044af0d1e6474076366759db86f101778c5f20ccvboxsync uShw.pPDPae->a[iShw2].u = 0;
044af0d1e6474076366759db86f101778c5f20ccvboxsync }
044af0d1e6474076366759db86f101778c5f20ccvboxsync }
044af0d1e6474076366759db86f101778c5f20ccvboxsync break;
044af0d1e6474076366759db86f101778c5f20ccvboxsync }
044af0d1e6474076366759db86f101778c5f20ccvboxsync
044af0d1e6474076366759db86f101778c5f20ccvboxsync case PGMPOOLKIND_PAE_PDPT:
044af0d1e6474076366759db86f101778c5f20ccvboxsync {
044af0d1e6474076366759db86f101778c5f20ccvboxsync /*
044af0d1e6474076366759db86f101778c5f20ccvboxsync * Hopefully this doesn't happen very often:
044af0d1e6474076366759db86f101778c5f20ccvboxsync * - touching unused parts of the page
044af0d1e6474076366759db86f101778c5f20ccvboxsync * - messing with the bits of pd pointers without changing the physical address
044af0d1e6474076366759db86f101778c5f20ccvboxsync */
044af0d1e6474076366759db86f101778c5f20ccvboxsync /* PDPT roots are not page aligned; 32 byte only! */
044af0d1e6474076366759db86f101778c5f20ccvboxsync const unsigned offPdpt = GCPhysFault - pPage->GCPhys;
044af0d1e6474076366759db86f101778c5f20ccvboxsync
044af0d1e6474076366759db86f101778c5f20ccvboxsync uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync const unsigned iShw = offPdpt / sizeof(X86PDPE);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (iShw < X86_PG_PAE_PDPE_ENTRIES) /* don't use RT_ELEMENTS(uShw.pPDPT->a), because that's for long mode only */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# ifndef IN_RING0
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (uShw.pPDPT->a[iShw].u & PGM_PLXFLAGS_MAPPING)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("pgmPoolMonitorChainChanging: Detected pdpt conflict at iShw=%#x!\n", iShw));
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync break;
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif /* !IN_RING0 */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# ifndef IN_RING0
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync else
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync# endif /* !IN_RING0 */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (uShw.pPDPT->a[iShw].n.u1Present)
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync {
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync pgmPoolFree(pVM,
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK,
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync pPage->idx,
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync iShw);
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync uShw.pPDPT->a[iShw].u = 0;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync /* paranoia / a bit assumptive. */
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if ( pCpu
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync && (offPdpt & 7)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync && (offPdpt & 7) + cbWrite > sizeof(X86PDPE))
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync const unsigned iShw2 = (offPdpt + cbWrite - 1) / sizeof(X86PDPE);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if ( iShw2 != iShw
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync && iShw2 < X86_PG_PAE_PDPE_ENTRIES)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync# ifndef IN_RING0
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if (uShw.pPDPT->a[iShw2].u & PGM_PLXFLAGS_MAPPING)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync Assert(pgmMapAreMappingsEnabled(&pVM->pgm.s));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync STAM_COUNTER_INC(&(pVCpu->pgm.s.StatRZGuestCR3WriteConflict));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync LogFlow(("pgmPoolMonitorChainChanging: Detected conflict at iShw2=%#x!\n", iShw2));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync break;
611e5e148a74d4b54cf76c97e4d36acaa816d8c0vboxsync }
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync# endif /* !IN_RING0 */
9939c713bffcfc4305d99d994552aa2ad9bce097vboxsync# ifndef IN_RING0
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync else
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync# endif /* !IN_RING0 */
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if (uShw.pPDPT->a[iShw2].n.u1Present)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync LogFlow(("pgmPoolMonitorChainChanging: pae pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync pgmPoolFree(pVM,
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK,
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync pPage->idx,
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync iShw2);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync uShw.pPDPT->a[iShw2].u = 0;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync break;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync#ifndef IN_RC
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync const unsigned iShw = off / sizeof(X86PDEPAE);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync Assert(!(uShw.pPDPae->a[iShw].u & PGM_PDFLAGS_MAPPING));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if (uShw.pPDPae->a[iShw].n.u1Present)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPae->a[iShw].u));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync pgmPoolFree(pVM,
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync uShw.pPDPae->a[iShw].u & X86_PDE_PAE_PG_MASK,
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync pPage->idx,
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync iShw);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync uShw.pPDPae->a[iShw].u = 0;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync /* paranoia / a bit assumptive. */
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if ( pCpu
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync && (off & 7)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync && (off & 7) + cbWrite > sizeof(X86PDEPAE))
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDEPAE);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync AssertBreak(iShw2 < RT_ELEMENTS(uShw.pPDPae->a));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync Assert(!(uShw.pPDPae->a[iShw2].u & PGM_PDFLAGS_MAPPING));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if (uShw.pPDPae->a[iShw2].n.u1Present)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync LogFlow(("pgmPoolMonitorChainChanging: pae pd iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPae->a[iShw2].u));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync pgmPoolFree(pVM,
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync uShw.pPDPae->a[iShw2].u & X86_PDE_PAE_PG_MASK,
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync pPage->idx,
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync iShw2);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync uShw.pPDPae->a[iShw2].u = 0;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
ad27e1d5e48ca41245120c331cc88b50464813cevboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync break;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
a1df400bbe9d64aad400442e56eb637019300a5evboxsync /*
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * Hopefully this doesn't happen very often:
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * - messing with the bits of pd pointers without changing the physical address
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync if (!VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
044af0d1e6474076366759db86f101778c5f20ccvboxsync {
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync const unsigned iShw = off / sizeof(X86PDPE);
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync if (uShw.pPDPT->a[iShw].n.u1Present)
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync {
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPDPT->a[iShw].u));
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync pgmPoolFree(pVM, uShw.pPDPT->a[iShw].u & X86_PDPE_PG_MASK, pPage->idx, iShw);
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync uShw.pPDPT->a[iShw].u = 0;
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync }
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync /* paranoia / a bit assumptive. */
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync if ( pCpu
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync && (off & 7)
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync && (off & 7) + cbWrite > sizeof(X86PDPE))
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync {
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PDPE);
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync if (uShw.pPDPT->a[iShw2].n.u1Present)
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync {
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync LogFlow(("pgmPoolMonitorChainChanging: pdpt iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPDPT->a[iShw2].u));
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync pgmPoolFree(pVM, uShw.pPDPT->a[iShw2].u & X86_PDPE_PG_MASK, pPage->idx, iShw2);
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync uShw.pPDPT->a[iShw2].u = 0;
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync }
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync }
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync }
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync break;
bbb4c0bfd5ea55e99591d8811771257a437053eevboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync case PGMPOOLKIND_64BIT_PML4:
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync /*
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * Hopefully this doesn't happen very often:
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * - messing with the bits of pd pointers without changing the physical address
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync */
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if (!VMCPU_FF_ISSET(pVCpu, VMCPU_FF_PGM_SYNC_CR3))
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync uShw.pv = PGMPOOL_PAGE_2_LOCKED_PTR(pVM, pPage);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync const unsigned iShw = off / sizeof(X86PDPE);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if (uShw.pPML4->a[iShw].n.u1Present)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw=%#x: %RX64 -> freeing it!\n", iShw, uShw.pPML4->a[iShw].u));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync pgmPoolFree(pVM, uShw.pPML4->a[iShw].u & X86_PML4E_PG_MASK, pPage->idx, iShw);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync uShw.pPML4->a[iShw].u = 0;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync /* paranoia / a bit assumptive. */
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if ( pCpu
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync && (off & 7)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync && (off & 7) + cbWrite > sizeof(X86PDPE))
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync const unsigned iShw2 = (off + cbWrite - 1) / sizeof(X86PML4E);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if (uShw.pPML4->a[iShw2].n.u1Present)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync LogFlow(("pgmPoolMonitorChainChanging: pml4 iShw2=%#x: %RX64 -> freeing it!\n", iShw2, uShw.pPML4->a[iShw2].u));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync pgmPoolFree(pVM, uShw.pPML4->a[iShw2].u & X86_PML4E_PG_MASK, pPage->idx, iShw2);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync uShw.pPML4->a[iShw2].u = 0;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync break;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync#endif /* IN_RING0 */
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync default:
ad27e1d5e48ca41245120c331cc88b50464813cevboxsync AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync PGMPOOL_UNLOCK_PTR(pVM, uShw.pv);
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync /* next */
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if (pPage->iMonitoredNext == NIL_PGMPOOL_IDX)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync return;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync pPage = &pPool->aPages[pPage->iMonitoredNext];
044af0d1e6474076366759db86f101778c5f20ccvboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync}
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync# ifndef IN_RING3
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync/**
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * Checks if a access could be a fork operation in progress.
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync *
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * Meaning, that the guest is setuping up the parent process for Copy-On-Write.
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync *
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * @returns true if it's likly that we're forking, otherwise false.
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * @param pPool The pool.
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * @param pCpu The disassembled instruction.
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * @param offFault The access offset.
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync */
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsyncDECLINLINE(bool) pgmPoolMonitorIsForking(PPGMPOOL pPool, PDISCPUSTATE pCpu, unsigned offFault)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync{
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync /*
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * i386 linux is using btr to clear X86_PTE_RW.
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * The functions involved are (2.6.16 source inspection):
044af0d1e6474076366759db86f101778c5f20ccvboxsync * clear_bit
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * ptep_set_wrprotect
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * copy_one_pte
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * copy_pte_range
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * copy_pmd_range
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * copy_pud_range
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * copy_page_range
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * dup_mmap
a1df400bbe9d64aad400442e56eb637019300a5evboxsync * dup_mm
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * copy_mm
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * copy_process
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * do_fork
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync */
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync if ( pCpu->pCurInstr->opcode == OP_BTR
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync && !(offFault & 4)
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync /** @todo Validate that the bit index is X86_PTE_RW. */
4a1654dd5b9f0ae4e149d909843a3ab07b8bec33vboxsync )
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync {
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,Fork));
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync return true;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync }
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync return false;
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync}
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync
a1df400bbe9d64aad400442e56eb637019300a5evboxsync
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync/**
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * Determine whether the page is likely to have been reused.
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync *
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * @returns true if we consider the page as being reused for a different purpose.
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * @returns false if we consider it to still be a paging page.
c0e27f622f9bd6d9e77d2d959aab71d69dabf0d3vboxsync * @param pVM VM Handle.
ca3da10d05961c339b5180fbd40a54587d6bad35vboxsync * @param pPage The page in question.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * @param pRegFrame Trap register frame.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pCpu The disassembly info for the faulting instruction.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pvFault The fault address.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync *
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @remark The REP prefix check is left to the caller because of STOSD/W.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsyncDECLINLINE(bool) pgmPoolMonitorIsReused(PVM pVM, PPGMPOOLPAGE pPage, PCPUMCTXCORE pRegFrame, PDISCPUSTATE pCpu, RTGCPTR pvFault)
ad27e1d5e48ca41245120c331cc88b50464813cevboxsync{
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#ifndef IN_RC
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /** @todo could make this general, faulting close to rsp should be safe reuse heuristic. */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync if ( HWACCMHasPendingIrq(pVM)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync && (pRegFrame->rsp - pvFault) < 32)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync {
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /* Fault caused by stack writes while trying to inject an interrupt event. */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log(("pgmPoolMonitorIsReused: reused %RGv for interrupt stack (rsp=%RGv).\n", pvFault, pRegFrame->rsp));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return true;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync }
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#else
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync NOREF(pVM); NOREF(pvFault);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#endif
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync switch (pCpu->pCurInstr->opcode)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync {
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /* call implies the actual push of the return address faulted */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync case OP_CALL:
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log4(("pgmPoolMonitorIsReused: CALL\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return true;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync case OP_PUSH:
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log4(("pgmPoolMonitorIsReused: PUSH\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return true;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync case OP_PUSHF:
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log4(("pgmPoolMonitorIsReused: PUSHF\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return true;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync case OP_PUSHA:
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log4(("pgmPoolMonitorIsReused: PUSHA\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return true;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync case OP_FXSAVE:
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log4(("pgmPoolMonitorIsReused: FXSAVE\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return true;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync case OP_MOVNTI: /* solaris - block_zero_no_xmm */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log4(("pgmPoolMonitorIsReused: MOVNTI\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return true;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync case OP_MOVNTDQ: /* solaris - hwblkclr & hwblkpagecopy */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log4(("pgmPoolMonitorIsReused: MOVNTDQ\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return true;
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync case OP_MOVSWD:
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync case OP_STOSWD:
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync if ( pCpu->prefix == (PREFIX_REP|PREFIX_REX)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync && pRegFrame->rcx >= 0x40
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync )
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync {
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Assert(pCpu->mode == CPUMODE_64BIT);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log(("pgmPoolMonitorIsReused: OP_STOSQ\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return true;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync }
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return false;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync }
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync if ( (pCpu->param1.flags & USE_REG_GEN32)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync && (pCpu->param1.base.reg_gen == USE_REG_ESP))
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync {
a1df400bbe9d64aad400442e56eb637019300a5evboxsync Log4(("pgmPoolMonitorIsReused: ESP\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return true;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync }
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync //if (pPage->fCR3Mix)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync // return false;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return false;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync}
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync/**
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Flushes the page being accessed.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync *
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @returns VBox status code suitable for scheduling.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pVM The VM handle.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pVCpu The VMCPU handle.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pPool The pool.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pPage The pool page (head).
a1df400bbe9d64aad400442e56eb637019300a5evboxsync * @param pCpu The disassembly of the write instruction.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pRegFrame The trap register frame.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param GCPhysFault The fault address as guest physical address.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pvFault The fault address.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
ad27e1d5e48ca41245120c331cc88b50464813cevboxsyncstatic int pgmPoolAccessHandlerFlush(PVM pVM, PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync{
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * First, do the flushing.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync int rc = pgmPoolMonitorChainFlush(pPool, pPage);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Emulate the instruction (xp/w2k problem, requires pc/cr2/sp detection).
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync uint32_t cbWritten;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync int rc2 = EMInterpretInstructionCPU(pVM, pVCpu, pCpu, pRegFrame, pvFault, &cbWritten);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync if (RT_SUCCESS(rc2))
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pRegFrame->rip += pCpu->opsize;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync else if (rc2 == VERR_EM_INTERPRETER)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync {
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#ifdef IN_RC
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync if (PATMIsPatchGCAddr(pVM, (RTRCPTR)pRegFrame->eip))
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync {
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for patch code %04x:%RGv, ignoring.\n",
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pRegFrame->cs, (RTGCPTR)pRegFrame->eip));
a1df400bbe9d64aad400442e56eb637019300a5evboxsync rc = VINF_SUCCESS;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync STAM_COUNTER_INC(&pPool->StatMonitorRZIntrFailPatch2);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync }
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync else
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#endif
ad27e1d5e48ca41245120c331cc88b50464813cevboxsync {
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync rc = VINF_EM_RAW_EMULATE_INSTR;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync }
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync }
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync else
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync rc = rc2;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /* See use in pgmPoolAccessHandlerSimple(). */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PGM_INVL_GUEST_TLBS();
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync LogFlow(("pgmPoolAccessHandlerPT: returns %Rrc (flushed)\n", rc));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return rc;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync}
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync/**
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Handles the STOSD write accesses.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync *
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @returns VBox status code suitable for scheduling.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pVM The VM handle.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pPool The pool.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pPage The pool page (head).
a1df400bbe9d64aad400442e56eb637019300a5evboxsync * @param pCpu The disassembly of the write instruction.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pRegFrame The trap register frame.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param GCPhysFault The fault address as guest physical address.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pvFault The fault address.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsyncDECLINLINE(int) pgmPoolAccessHandlerSTOSD(PVM pVM, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync{
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Assert(pCpu->mode == CPUMODE_32BIT);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log3(("pgmPoolAccessHandlerSTOSD\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Increment the modification counter and insert it into the list
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * of modified pages the first time.
a1df400bbe9d64aad400442e56eb637019300a5evboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync if (!pPage->cModifications++)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pgmPoolMonitorModifiedInsert(pPool, pPage);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Execute REP STOSD.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync *
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * This ASSUMES that we're not invoked by Trap0e on in a out-of-sync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * write situation, meaning that it's safe to write here.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PVMCPU pVCpu = VMMGetCpu(pPool->CTX_SUFF(pVM));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync RTGCUINTPTR pu32 = (RTGCUINTPTR)pvFault;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync while (pRegFrame->ecx)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync {
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, (RTGCPTR)pu32, NULL);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
a1df400bbe9d64aad400442e56eb637019300a5evboxsync#else
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, (RTGCPTR)pu32, NULL);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#endif
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#ifdef IN_RC
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync *(uint32_t *)pu32 = pRegFrame->eax;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#else
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PGMPhysSimpleWriteGCPhys(pVM, GCPhysFault, &pRegFrame->eax, 4);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#endif
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pu32 += 4;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync GCPhysFault += 4;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pRegFrame->edi += 4;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pRegFrame->ecx--;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync }
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pRegFrame->rip += pCpu->opsize;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#ifdef IN_RC
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /* See use in pgmPoolAccessHandlerSimple(). */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PGM_INVL_GUEST_TLBS();
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#endif
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync LogFlow(("pgmPoolAccessHandlerSTOSD: returns\n"));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return VINF_SUCCESS;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync}
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync/**
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Handles the simple write accesses.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync *
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @returns VBox status code suitable for scheduling.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pVM The VM handle.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pVCpu The VMCPU handle.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pPool The pool.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pPage The pool page (head).
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pCpu The disassembly of the write instruction.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pRegFrame The trap register frame.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param GCPhysFault The fault address as guest physical address.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pvFault The fault address.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsyncDECLINLINE(int) pgmPoolAccessHandlerSimple(PVM pVM, PVMCPU pVCpu, PPGMPOOL pPool, PPGMPOOLPAGE pPage, PDISCPUSTATE pCpu,
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PCPUMCTXCORE pRegFrame, RTGCPHYS GCPhysFault, RTGCPTR pvFault)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync{
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Log3(("pgmPoolAccessHandlerSimple\n"));
a1df400bbe9d64aad400442e56eb637019300a5evboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Increment the modification counter and insert it into the list
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * of modified pages the first time.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync if (!pPage->cModifications++)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pgmPoolMonitorModifiedInsert(pPool, pPage);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Clear all the pages. ASSUMES that pvFault is readable.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, pCpu);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#else
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pgmPoolMonitorChainChanging(pVCpu, pPool, pPage, GCPhysFault, pvFault, pCpu);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#endif
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Interpret the instruction.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync uint32_t cb;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync int rc = EMInterpretInstructionCPU(pVM, pVCpu, pCpu, pRegFrame, pvFault, &cb);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync if (RT_SUCCESS(rc))
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pRegFrame->rip += pCpu->opsize;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync else if (rc == VERR_EM_INTERPRETER)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync {
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync LogFlow(("pgmPoolAccessHandlerPTWorker: Interpretation failed for %04x:%RGv - opcode=%d\n",
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pRegFrame->cs, (RTGCPTR)pRegFrame->rip, pCpu->pCurInstr->opcode));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync rc = VINF_EM_RAW_EMULATE_INSTR;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,EmulateInstr));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync }
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#ifdef IN_RC
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Quick hack, with logging enabled we're getting stale
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * code TLBs but no data TLB for EIP and crash in EMInterpretDisasOne.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Flushing here is BAD and expensive, I think EMInterpretDisasOne will
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * have to be fixed to support this. But that'll have to wait till next week.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync *
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * An alternative is to keep track of the changed PTEs together with the
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * GCPhys from the guest PT. This may proove expensive though.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync *
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * At the moment, it's VITAL that it's done AFTER the instruction interpreting
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * because we need the stale TLBs in some cases (XP boot). This MUST be fixed properly!
ad27e1d5e48ca41245120c331cc88b50464813cevboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PGM_INVL_GUEST_TLBS();
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync#endif
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync LogFlow(("pgmPoolAccessHandlerSimple: returns %Rrc cb=%d\n", rc, cb));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync return rc;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync}
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync/**
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * \#PF Handler callback for PT write accesses.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync *
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @returns VBox status code (appropriate for GC return).
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pVM VM Handle.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param uErrorCode CPU Error code.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pRegFrame Trap register frame.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * NULL on DMA and other non CPU access.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pvFault The fault address (cr2).
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param GCPhysFault The GC physical address corresponding to pvFault.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * @param pvUser User argument.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
88acfa6629a7976c0583c1712d2b5b22a87a5121vboxsyncDECLEXPORT(int) pgmPoolAccessHandler(PVM pVM, RTGCUINT uErrorCode, PCPUMCTXCORE pRegFrame, RTGCPTR pvFault, RTGCPHYS GCPhysFault, void *pvUser)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync{
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync STAM_PROFILE_START(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), a);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)pvUser;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync PVMCPU pVCpu = VMMGetCpu(pVM);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
ad27e1d5e48ca41245120c331cc88b50464813cevboxsync LogFlow(("pgmPoolAccessHandler: pvFault=%RGv pPage=%p:{.idx=%d} GCPhysFault=%RGp\n", pvFault, pPage, pPage->idx, GCPhysFault));
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * We should ALWAYS have the list head as user parameter. This
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * is because we use that page to record the changes.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Disassemble the faulting instruction.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync DISCPUSTATE Cpu;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync int rc = EMInterpretDisasOne(pVM, pVCpu, pRegFrame, &Cpu, NULL);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync AssertRCReturn(rc, rc);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync pgmLock(pVM);
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync /*
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync * Check if it's worth dealing with.
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync bool fReused = false;
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync if ( ( pPage->cModifications < 48 /** @todo #define */ /** @todo need to check that it's not mapping EIP. */ /** @todo adjust this! */
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync || pgmPoolIsPageLocked(&pVM->pgm.s, pPage)
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync )
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync && !(fReused = pgmPoolMonitorIsReused(pVM, pPage, pRegFrame, &Cpu, pvFault))
df8e6a449f00e1884fbf4a1fc67143614d7d528dvboxsync && !pgmPoolMonitorIsForking(pPool, &Cpu, GCPhysFault & PAGE_OFFSET_MASK))
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync {
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync /*
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync * Simple instructions, no REP prefix.
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync */
590bfe12ce22cd3716448fbb9f4dc51664bfe5e2vboxsync if (!(Cpu.prefix & (PREFIX_REP | PREFIX_REPNE)))
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync {
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync rc = pgmPoolAccessHandlerSimple(pVM, pVCpu, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
3a4a6501d0ccd629d9951b644d380c7bb2d46086vboxsync STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,Handled), a);
pgmUnlock(pVM);
return rc;
}
/*
* Windows is frequently doing small memset() operations (netio test 4k+).
* We have to deal with these or we'll kill the cache and performance.
*/
if ( Cpu.pCurInstr->opcode == OP_STOSWD
&& CPUMGetGuestCPL(pVCpu, pRegFrame) == 0
&& pRegFrame->ecx <= 0x20
&& pRegFrame->ecx * 4 <= PAGE_SIZE - ((uintptr_t)pvFault & PAGE_OFFSET_MASK)
&& !((uintptr_t)pvFault & 3)
&& (pRegFrame->eax == 0 || pRegFrame->eax == 0x80) /* the two values observed. */
&& Cpu.mode == CPUMODE_32BIT
&& Cpu.opmode == CPUMODE_32BIT
&& Cpu.addrmode == CPUMODE_32BIT
&& Cpu.prefix == PREFIX_REP
&& !pRegFrame->eflags.Bits.u1DF
)
{
rc = pgmPoolAccessHandlerSTOSD(pVM, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,RepStosd), a);
pgmUnlock(pVM);
return rc;
}
/* REP prefix, don't bother. */
STAM_COUNTER_INC(&pPool->CTX_MID_Z(StatMonitor,RepPrefix));
Log4(("pgmPoolAccessHandler: eax=%#x ecx=%#x edi=%#x esi=%#x rip=%RGv opcode=%d prefix=%#x\n",
pRegFrame->eax, pRegFrame->ecx, pRegFrame->edi, pRegFrame->esi, (RTGCPTR)pRegFrame->rip, Cpu.pCurInstr->opcode, Cpu.prefix));
}
/*
* Not worth it, so flush it.
*
* If we considered it to be reused, don't go back to ring-3
* to emulate failed instructions since we usually cannot
* interpret then. This may be a bit risky, in which case
* the reuse detection must be fixed.
*/
rc = pgmPoolAccessHandlerFlush(pVM, pVCpu, pPool, pPage, &Cpu, pRegFrame, GCPhysFault, pvFault);
if (rc == VINF_EM_RAW_EMULATE_INSTR && fReused)
rc = VINF_SUCCESS;
STAM_PROFILE_STOP_EX(&pVM->pgm.s.CTX_SUFF(pPool)->CTX_SUFF_Z(StatMonitor), &pPool->CTX_MID_Z(StatMonitor,FlushPage), a);
pgmUnlock(pVM);
return rc;
}
# endif /* !IN_RING3 */
#endif /* PGMPOOL_WITH_MONITORING */
#ifdef PGMPOOL_WITH_CACHE
/**
* Inserts a page into the GCPhys hash table.
*
* @param pPool The pool.
* @param pPage The page.
*/
DECLINLINE(void) pgmPoolHashInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
{
Log3(("pgmPoolHashInsert: %RGp\n", pPage->GCPhys));
Assert(pPage->GCPhys != NIL_RTGCPHYS); Assert(pPage->iNext == NIL_PGMPOOL_IDX);
uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
pPage->iNext = pPool->aiHash[iHash];
pPool->aiHash[iHash] = pPage->idx;
}
/**
* Removes a page from the GCPhys hash table.
*
* @param pPool The pool.
* @param pPage The page.
*/
DECLINLINE(void) pgmPoolHashRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
{
Log3(("pgmPoolHashRemove: %RGp\n", pPage->GCPhys));
uint16_t iHash = PGMPOOL_HASH(pPage->GCPhys);
if (pPool->aiHash[iHash] == pPage->idx)
pPool->aiHash[iHash] = pPage->iNext;
else
{
uint16_t iPrev = pPool->aiHash[iHash];
for (;;)
{
const int16_t i = pPool->aPages[iPrev].iNext;
if (i == pPage->idx)
{
pPool->aPages[iPrev].iNext = pPage->iNext;
break;
}
if (i == NIL_PGMPOOL_IDX)
{
AssertReleaseMsgFailed(("GCPhys=%RGp idx=%#x\n", pPage->GCPhys, pPage->idx));
break;
}
iPrev = i;
}
}
pPage->iNext = NIL_PGMPOOL_IDX;
}
/**
* Frees up one cache page.
*
* @returns VBox status code.
* @retval VINF_SUCCESS on success.
* @param pPool The pool.
* @param iUser The user index.
*/
static int pgmPoolCacheFreeOne(PPGMPOOL pPool, uint16_t iUser)
{
#ifndef IN_RC
const PVM pVM = pPool->CTX_SUFF(pVM);
#endif
Assert(pPool->iAgeHead != pPool->iAgeTail); /* We shouldn't be here if there < 2 cached entries! */
STAM_COUNTER_INC(&pPool->StatCacheFreeUpOne);
/*
* Select one page from the tail of the age list.
*/
PPGMPOOLPAGE pPage;
for (unsigned iLoop = 0; ; iLoop++)
{
uint16_t iToFree = pPool->iAgeTail;
if (iToFree == iUser)
iToFree = pPool->aPages[iToFree].iAgePrev;
/* This is the alternative to the SyncCR3 pgmPoolCacheUsed calls.
if (pPool->aPages[iToFree].iUserHead != NIL_PGMPOOL_USER_INDEX)
{
uint16_t i = pPool->aPages[iToFree].iAgePrev;
for (unsigned j = 0; j < 10 && i != NIL_PGMPOOL_USER_INDEX; j++, i = pPool->aPages[i].iAgePrev)
{
if (pPool->aPages[iToFree].iUserHead == NIL_PGMPOOL_USER_INDEX)
continue;
iToFree = i;
break;
}
}
*/
Assert(iToFree != iUser);
AssertRelease(iToFree != NIL_PGMPOOL_IDX);
pPage = &pPool->aPages[iToFree];
/*
* Reject any attempts at flushing the currently active shadow CR3 mapping.
* Call pgmPoolCacheUsed to move the page to the head of the age list.
*/
if (!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage))
break;
LogFlow(("pgmPoolCacheFreeOne: refuse CR3 mapping\n"));
pgmPoolCacheUsed(pPool, pPage);
AssertLogRelReturn(iLoop < 8192, VERR_INTERNAL_ERROR);
}
/*
* Found a usable page, flush it and return.
*/
int rc = pgmPoolFlushPage(pPool, pPage);
if (rc == VINF_SUCCESS)
PGM_INVL_GUEST_TLBS(); /* see PT handler. */
return rc;
}
/**
* Checks if a kind mismatch is really a page being reused
* or if it's just normal remappings.
*
* @returns true if reused and the cached page (enmKind1) should be flushed
* @returns false if not reused.
* @param enmKind1 The kind of the cached page.
* @param enmKind2 The kind of the requested page.
*/
static bool pgmPoolCacheReusedByKind(PGMPOOLKIND enmKind1, PGMPOOLKIND enmKind2)
{
switch (enmKind1)
{
/*
* Never reuse them. There is no remapping in non-paging mode.
*/
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
case PGMPOOLKIND_32BIT_PD_PHYS:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PD_PHYS:
case PGMPOOLKIND_PAE_PDPT_PHYS:
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
case PGMPOOLKIND_PAE_PDPT_FOR_32BIT: /* never reuse them for other types */
return false;
/*
* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
*/
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
case PGMPOOLKIND_32BIT_PD:
case PGMPOOLKIND_PAE_PDPT:
switch (enmKind2)
{
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
case PGMPOOLKIND_64BIT_PML4:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
return true;
default:
return false;
}
/*
* It's perfectly fine to reuse these, except for PAE and non-paging stuff.
*/
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
case PGMPOOLKIND_64BIT_PML4:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
switch (enmKind2)
{
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
return true;
default:
return false;
}
/*
* These cannot be flushed, and it's common to reuse the PDs as PTs.
*/
case PGMPOOLKIND_ROOT_NESTED:
return false;
default:
AssertFatalMsgFailed(("enmKind1=%d\n", enmKind1));
}
}
/**
* Attempts to satisfy a pgmPoolAlloc request from the cache.
*
* @returns VBox status code.
* @retval VINF_PGM_CACHED_PAGE on success.
* @retval VERR_FILE_NOT_FOUND if not found.
* @param pPool The pool.
* @param GCPhys The GC physical address of the page we're gonna shadow.
* @param enmKind The kind of mapping.
* @param iUser The shadow page pool index of the user table.
* @param iUserTable The index into the user table (shadowed).
* @param ppPage Where to store the pointer to the page.
*/
static int pgmPoolCacheAlloc(PPGMPOOL pPool, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
{
#ifndef IN_RC
const PVM pVM = pPool->CTX_SUFF(pVM);
#endif
/*
* Look up the GCPhys in the hash.
*/
unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
Log3(("pgmPoolCacheAlloc: %RGp kind %s iUser=%x iUserTable=%x SLOT=%d\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable, i));
if (i != NIL_PGMPOOL_IDX)
{
do
{
PPGMPOOLPAGE pPage = &pPool->aPages[i];
Log4(("pgmPoolCacheAlloc: slot %d found page %RGp\n", i, pPage->GCPhys));
if (pPage->GCPhys == GCPhys)
{
if ((PGMPOOLKIND)pPage->enmKind == enmKind)
{
/* Put it at the start of the use list to make sure pgmPoolTrackAddUser
* doesn't flush it in case there are no more free use records.
*/
pgmPoolCacheUsed(pPool, pPage);
int rc = pgmPoolTrackAddUser(pPool, pPage, iUser, iUserTable);
if (RT_SUCCESS(rc))
{
Assert((PGMPOOLKIND)pPage->enmKind == enmKind);
*ppPage = pPage;
STAM_COUNTER_INC(&pPool->StatCacheHits);
return VINF_PGM_CACHED_PAGE;
}
return rc;
}
/*
* The kind is different. In some cases we should now flush the page
* as it has been reused, but in most cases this is normal remapping
* of PDs as PT or big pages using the GCPhys field in a slightly
* different way than the other kinds.
*/
if (pgmPoolCacheReusedByKind((PGMPOOLKIND)pPage->enmKind, enmKind))
{
STAM_COUNTER_INC(&pPool->StatCacheKindMismatches);
pgmPoolFlushPage(pPool, pPage);
PGM_INVL_GUEST_TLBS(); /* see PT handler. */
break;
}
}
/* next */
i = pPage->iNext;
} while (i != NIL_PGMPOOL_IDX);
}
Log3(("pgmPoolCacheAlloc: Missed GCPhys=%RGp enmKind=%s\n", GCPhys, pgmPoolPoolKindToStr(enmKind)));
STAM_COUNTER_INC(&pPool->StatCacheMisses);
return VERR_FILE_NOT_FOUND;
}
/**
* Inserts a page into the cache.
*
* @param pPool The pool.
* @param pPage The cached page.
* @param fCanBeCached Set if the page is fit for caching from the caller's point of view.
*/
static void pgmPoolCacheInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, bool fCanBeCached)
{
/*
* Insert into the GCPhys hash if the page is fit for that.
*/
Assert(!pPage->fCached);
if (fCanBeCached)
{
pPage->fCached = true;
pgmPoolHashInsert(pPool, pPage);
Log3(("pgmPoolCacheInsert: Caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
STAM_COUNTER_INC(&pPool->StatCacheCacheable);
}
else
{
Log3(("pgmPoolCacheInsert: Not caching %p:{.Core=%RHp, .idx=%d, .enmKind=%s, GCPhys=%RGp}\n",
pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
STAM_COUNTER_INC(&pPool->StatCacheUncacheable);
}
/*
* Insert at the head of the age list.
*/
pPage->iAgePrev = NIL_PGMPOOL_IDX;
pPage->iAgeNext = pPool->iAgeHead;
if (pPool->iAgeHead != NIL_PGMPOOL_IDX)
pPool->aPages[pPool->iAgeHead].iAgePrev = pPage->idx;
else
pPool->iAgeTail = pPage->idx;
pPool->iAgeHead = pPage->idx;
}
/**
* Flushes a cached page.
*
* @param pPool The pool.
* @param pPage The cached page.
*/
static void pgmPoolCacheFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
{
Log3(("pgmPoolCacheFlushPage: %RGp\n", pPage->GCPhys));
/*
* Remove the page from the hash.
*/
if (pPage->fCached)
{
pPage->fCached = false;
pgmPoolHashRemove(pPool, pPage);
}
else
Assert(pPage->iNext == NIL_PGMPOOL_IDX);
/*
* Remove it from the age list.
*/
if (pPage->iAgeNext != NIL_PGMPOOL_IDX)
pPool->aPages[pPage->iAgeNext].iAgePrev = pPage->iAgePrev;
else
pPool->iAgeTail = pPage->iAgePrev;
if (pPage->iAgePrev != NIL_PGMPOOL_IDX)
pPool->aPages[pPage->iAgePrev].iAgeNext = pPage->iAgeNext;
else
pPool->iAgeHead = pPage->iAgeNext;
pPage->iAgeNext = NIL_PGMPOOL_IDX;
pPage->iAgePrev = NIL_PGMPOOL_IDX;
}
#endif /* PGMPOOL_WITH_CACHE */
#ifdef PGMPOOL_WITH_MONITORING
/**
* Looks for pages sharing the monitor.
*
* @returns Pointer to the head page.
* @returns NULL if not found.
* @param pPool The Pool
* @param pNewPage The page which is going to be monitored.
*/
static PPGMPOOLPAGE pgmPoolMonitorGetPageByGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pNewPage)
{
#ifdef PGMPOOL_WITH_CACHE
/*
* Look up the GCPhys in the hash.
*/
RTGCPHYS GCPhys = pNewPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
unsigned i = pPool->aiHash[PGMPOOL_HASH(GCPhys)];
if (i == NIL_PGMPOOL_IDX)
return NULL;
do
{
PPGMPOOLPAGE pPage = &pPool->aPages[i];
if ( pPage->GCPhys - GCPhys < PAGE_SIZE
&& pPage != pNewPage)
{
switch (pPage->enmKind)
{
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
case PGMPOOLKIND_64BIT_PML4:
case PGMPOOLKIND_32BIT_PD:
case PGMPOOLKIND_PAE_PDPT:
{
/* find the head */
while (pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
{
Assert(pPage->iMonitoredPrev != pPage->idx);
pPage = &pPool->aPages[pPage->iMonitoredPrev];
}
return pPage;
}
/* ignore, no monitoring. */
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
case PGMPOOLKIND_ROOT_NESTED:
case PGMPOOLKIND_PAE_PD_PHYS:
case PGMPOOLKIND_PAE_PDPT_PHYS:
case PGMPOOLKIND_32BIT_PD_PHYS:
case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
break;
default:
AssertFatalMsgFailed(("enmKind=%d idx=%d\n", pPage->enmKind, pPage->idx));
}
}
/* next */
i = pPage->iNext;
} while (i != NIL_PGMPOOL_IDX);
#endif
return NULL;
}
/**
* Enabled write monitoring of a guest page.
*
* @returns VBox status code.
* @retval VINF_SUCCESS on success.
* @param pPool The pool.
* @param pPage The cached page.
*/
static int pgmPoolMonitorInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
{
LogFlow(("pgmPoolMonitorInsert %RGp\n", pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1)));
/*
* Filter out the relevant kinds.
*/
switch (pPage->enmKind)
{
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
case PGMPOOLKIND_64BIT_PML4:
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
case PGMPOOLKIND_32BIT_PD:
case PGMPOOLKIND_PAE_PDPT:
break;
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
case PGMPOOLKIND_ROOT_NESTED:
/* Nothing to monitor here. */
return VINF_SUCCESS;
case PGMPOOLKIND_32BIT_PD_PHYS:
case PGMPOOLKIND_PAE_PDPT_PHYS:
case PGMPOOLKIND_PAE_PD_PHYS:
case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
/* Nothing to monitor here. */
return VINF_SUCCESS;
#ifdef PGMPOOL_WITH_MIXED_PT_CR3
break;
#else
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
#endif
default:
AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
}
/*
* Install handler.
*/
int rc;
PPGMPOOLPAGE pPageHead = pgmPoolMonitorGetPageByGCPhys(pPool, pPage);
if (pPageHead)
{
Assert(pPageHead != pPage); Assert(pPageHead->iMonitoredNext != pPage->idx);
Assert(pPageHead->iMonitoredPrev != pPage->idx);
pPage->iMonitoredPrev = pPageHead->idx;
pPage->iMonitoredNext = pPageHead->iMonitoredNext;
if (pPageHead->iMonitoredNext != NIL_PGMPOOL_IDX)
pPool->aPages[pPageHead->iMonitoredNext].iMonitoredPrev = pPage->idx;
pPageHead->iMonitoredNext = pPage->idx;
rc = VINF_SUCCESS;
}
else
{
Assert(pPage->iMonitoredNext == NIL_PGMPOOL_IDX); Assert(pPage->iMonitoredPrev == NIL_PGMPOOL_IDX);
PVM pVM = pPool->CTX_SUFF(pVM);
const RTGCPHYS GCPhysPage = pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1);
rc = PGMHandlerPhysicalRegisterEx(pVM, PGMPHYSHANDLERTYPE_PHYSICAL_WRITE,
GCPhysPage, GCPhysPage + (PAGE_SIZE - 1),
pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
pPool->pszAccessHandler);
/** @todo we should probably deal with out-of-memory conditions here, but for now increasing
* the heap size should suffice. */
AssertFatalMsgRC(rc, ("PGMHandlerPhysicalRegisterEx %RGp failed with %Rrc\n", GCPhysPage, rc));
Assert(!(pVM->pgm.s.fGlobalSyncFlags & PGM_GLOBAL_SYNC_CLEAR_PGM_POOL) || VMCPU_FF_ISSET(VMMGetCpu(pVM), VMCPU_FF_PGM_SYNC_CR3));
}
pPage->fMonitored = true;
return rc;
}
/**
* Disables write monitoring of a guest page.
*
* @returns VBox status code.
* @retval VINF_SUCCESS on success.
* @param pPool The pool.
* @param pPage The cached page.
*/
static int pgmPoolMonitorFlush(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
{
/*
* Filter out the relevant kinds.
*/
switch (pPage->enmKind)
{
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
case PGMPOOLKIND_64BIT_PML4:
case PGMPOOLKIND_32BIT_PD:
case PGMPOOLKIND_PAE_PDPT:
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
break;
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
case PGMPOOLKIND_ROOT_NESTED:
case PGMPOOLKIND_PAE_PD_PHYS:
case PGMPOOLKIND_PAE_PDPT_PHYS:
case PGMPOOLKIND_32BIT_PD_PHYS:
/* Nothing to monitor here. */
return VINF_SUCCESS;
#ifdef PGMPOOL_WITH_MIXED_PT_CR3
break;
#endif
default:
AssertFatalMsgFailed(("This can't happen! enmKind=%d\n", pPage->enmKind));
}
/*
* Remove the page from the monitored list or uninstall it if last.
*/
const PVM pVM = pPool->CTX_SUFF(pVM);
int rc;
if ( pPage->iMonitoredNext != NIL_PGMPOOL_IDX
|| pPage->iMonitoredPrev != NIL_PGMPOOL_IDX)
{
if (pPage->iMonitoredPrev == NIL_PGMPOOL_IDX)
{
PPGMPOOLPAGE pNewHead = &pPool->aPages[pPage->iMonitoredNext];
pNewHead->iMonitoredPrev = NIL_PGMPOOL_IDX;
rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pNewHead),
pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pNewHead),
pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pNewHead),
pPool->pszAccessHandler);
AssertFatalRCSuccess(rc);
pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
}
else
{
pPool->aPages[pPage->iMonitoredPrev].iMonitoredNext = pPage->iMonitoredNext;
if (pPage->iMonitoredNext != NIL_PGMPOOL_IDX)
{
pPool->aPages[pPage->iMonitoredNext].iMonitoredPrev = pPage->iMonitoredPrev;
pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
}
pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
rc = VINF_SUCCESS;
}
}
else
{
rc = PGMHandlerPhysicalDeregister(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1));
AssertFatalRC(rc);
AssertMsg(!(pVM->pgm.s.fGlobalSyncFlags & PGM_GLOBAL_SYNC_CLEAR_PGM_POOL) || VMCPU_FF_ISSET(VMMGetCpu(pVM), VMCPU_FF_PGM_SYNC_CR3),
("%#x %#x\n", pVM->pgm.s.fGlobalSyncFlags, pVM->fGlobalForcedActions));
}
pPage->fMonitored = false;
/*
* Remove it from the list of modified pages (if in it).
*/
pgmPoolMonitorModifiedRemove(pPool, pPage);
return rc;
}
/**
* Inserts the page into the list of modified pages.
*
* @param pPool The pool.
* @param pPage The page.
*/
void pgmPoolMonitorModifiedInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
{
Log3(("pgmPoolMonitorModifiedInsert: idx=%d\n", pPage->idx));
AssertMsg( pPage->iModifiedNext == NIL_PGMPOOL_IDX
&& pPage->iModifiedPrev == NIL_PGMPOOL_IDX
&& pPool->iModifiedHead != pPage->idx,
("Next=%d Prev=%d idx=%d cModifications=%d Head=%d cModifiedPages=%d\n",
pPage->iModifiedNext, pPage->iModifiedPrev, pPage->idx, pPage->cModifications,
pPool->iModifiedHead, pPool->cModifiedPages));
pPage->iModifiedNext = pPool->iModifiedHead;
if (pPool->iModifiedHead != NIL_PGMPOOL_IDX)
pPool->aPages[pPool->iModifiedHead].iModifiedPrev = pPage->idx;
pPool->iModifiedHead = pPage->idx;
pPool->cModifiedPages++;
#ifdef VBOX_WITH_STATISTICS
if (pPool->cModifiedPages > pPool->cModifiedPagesHigh)
pPool->cModifiedPagesHigh = pPool->cModifiedPages;
#endif
}
/**
* Removes the page from the list of modified pages and resets the
* moficiation counter.
*
* @param pPool The pool.
* @param pPage The page which is believed to be in the list of modified pages.
*/
static void pgmPoolMonitorModifiedRemove(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
{
Log3(("pgmPoolMonitorModifiedRemove: idx=%d cModifications=%d\n", pPage->idx, pPage->cModifications));
if (pPool->iModifiedHead == pPage->idx)
{
Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
pPool->iModifiedHead = pPage->iModifiedNext;
if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
{
pPool->aPages[pPage->iModifiedNext].iModifiedPrev = NIL_PGMPOOL_IDX;
pPage->iModifiedNext = NIL_PGMPOOL_IDX;
}
pPool->cModifiedPages--;
}
else if (pPage->iModifiedPrev != NIL_PGMPOOL_IDX)
{
pPool->aPages[pPage->iModifiedPrev].iModifiedNext = pPage->iModifiedNext;
if (pPage->iModifiedNext != NIL_PGMPOOL_IDX)
{
pPool->aPages[pPage->iModifiedNext].iModifiedPrev = pPage->iModifiedPrev;
pPage->iModifiedNext = NIL_PGMPOOL_IDX;
}
pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
pPool->cModifiedPages--;
}
else
Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX);
pPage->cModifications = 0;
}
/**
* Zaps the list of modified pages, resetting their modification counters in the process.
*
* @param pVM The VM handle.
*/
void pgmPoolMonitorModifiedClearAll(PVM pVM)
{
pgmLock(pVM);
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
LogFlow(("pgmPoolMonitorModifiedClearAll: cModifiedPages=%d\n", pPool->cModifiedPages));
unsigned cPages = 0; NOREF(cPages);
uint16_t idx = pPool->iModifiedHead;
pPool->iModifiedHead = NIL_PGMPOOL_IDX;
while (idx != NIL_PGMPOOL_IDX)
{
PPGMPOOLPAGE pPage = &pPool->aPages[idx];
idx = pPage->iModifiedNext;
pPage->iModifiedNext = NIL_PGMPOOL_IDX;
pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
pPage->cModifications = 0;
Assert(++cPages);
}
AssertMsg(cPages == pPool->cModifiedPages, ("%d != %d\n", cPages, pPool->cModifiedPages));
pPool->cModifiedPages = 0;
pgmUnlock(pVM);
}
#ifdef IN_RING3
/**
* Callback to clear all shadow pages and clear all modification counters.
*
* @returns VBox status code.
* @param pVM The VM handle.
* @param pvUser Unused parameter
* @remark Should only be used when monitoring is available, thus placed in
* the PGMPOOL_WITH_MONITORING #ifdef.
*/
DECLCALLBACK(int) pgmPoolClearAll(PVM pVM, void *pvUser)
{
NOREF(pvUser);
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
STAM_PROFILE_START(&pPool->StatClearAll, c);
LogFlow(("pgmPoolClearAll: cUsedPages=%d\n", pPool->cUsedPages));
/*
* Iterate all the pages until we've encountered all that in use.
* This is simple but not quite optimal solution.
*/
unsigned cModifiedPages = 0; NOREF(cModifiedPages);
unsigned cLeft = pPool->cUsedPages;
unsigned iPage = pPool->cCurPages;
while (--iPage >= PGMPOOL_IDX_FIRST)
{
PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
if (pPage->GCPhys != NIL_RTGCPHYS)
{
switch (pPage->enmKind)
{
/*
* We only care about shadow page tables.
*/
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
{
#ifdef PGMPOOL_WITH_USER_TRACKING
if (pPage->cPresent)
#endif
{
void *pvShw = PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pPage);
STAM_PROFILE_START(&pPool->StatZeroPage, z);
ASMMemZeroPage(pvShw);
STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
#ifdef PGMPOOL_WITH_USER_TRACKING
pPage->cPresent = 0;
pPage->iFirstPresent = ~0;
#endif
}
}
/* fall thru */
default:
Assert(!pPage->cModifications || ++cModifiedPages);
Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX || pPage->cModifications);
Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX || pPage->cModifications);
pPage->iModifiedNext = NIL_PGMPOOL_IDX;
pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
pPage->cModifications = 0;
break;
}
if (!--cLeft)
break;
}
}
/* swipe the special pages too. */
for (iPage = PGMPOOL_IDX_FIRST_SPECIAL; iPage < PGMPOOL_IDX_FIRST; iPage++)
{
PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
if (pPage->GCPhys != NIL_RTGCPHYS)
{
Assert(!pPage->cModifications || ++cModifiedPages);
Assert(pPage->iModifiedNext == NIL_PGMPOOL_IDX || pPage->cModifications);
Assert(pPage->iModifiedPrev == NIL_PGMPOOL_IDX || pPage->cModifications);
pPage->iModifiedNext = NIL_PGMPOOL_IDX;
pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
pPage->cModifications = 0;
}
}
#ifndef DEBUG_michael
AssertMsg(cModifiedPages == pPool->cModifiedPages, ("%d != %d\n", cModifiedPages, pPool->cModifiedPages));
#endif
pPool->iModifiedHead = NIL_PGMPOOL_IDX;
pPool->cModifiedPages = 0;
#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
/*
* Clear all the GCPhys links and rebuild the phys ext free list.
*/
for (PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
pRam;
pRam = pRam->CTX_SUFF(pNext))
{
unsigned iPage = pRam->cb >> PAGE_SHIFT;
while (iPage-- > 0)
PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
}
pPool->iPhysExtFreeHead = 0;
PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
for (unsigned i = 0; i < cMaxPhysExts; i++)
{
paPhysExts[i].iNext = i + 1;
paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
}
paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
#endif
pPool->cPresent = 0;
PGM_INVL_GUEST_TLBS();
STAM_PROFILE_STOP(&pPool->StatClearAll, c);
return VINF_SUCCESS;
}
#endif /* IN_RING3 */
/**
* Handle SyncCR3 pool tasks
*
* @returns VBox status code.
* @retval VINF_SUCCESS if successfully added.
* @retval VINF_PGM_SYNC_CR3 is it needs to be deferred to ring 3 (GC only)
* @param pVM The VM handle.
* @remark Should only be used when monitoring is available, thus placed in
* the PGMPOOL_WITH_MONITORING #ifdef.
*/
int pgmPoolSyncCR3(PVM pVM)
{
LogFlow(("pgmPoolSyncCR3\n"));
/*
* When monitoring shadowed pages, we reset the modification counters on CR3 sync.
* Occasionally we will have to clear all the shadow page tables because we wanted
* to monitor a page which was mapped by too many shadowed page tables. This operation
* sometimes refered to as a 'lightweight flush'.
*/
# ifdef IN_RING3 /* Don't flush in ring-0 or raw mode, it's taking too long. */
if (ASMBitTestAndClear(&pVM->pgm.s.fGlobalSyncFlags, PGM_GLOBAL_SYNC_CLEAR_PGM_POOL_BIT))
{
VMMR3AtomicExecuteHandler(pVM, pgmPoolClearAll, NULL);
# else /* !IN_RING3 */
if (pVM->pgm.s.fGlobalSyncFlags & PGM_GLOBAL_SYNC_CLEAR_PGM_POOL)
{
LogFlow(("SyncCR3: PGM_GLOBAL_SYNC_CLEAR_PGM_POOL is set -> VINF_PGM_SYNC_CR3\n"));
VMCPU_FF_SET(VMMGetCpu(pVM), VMCPU_FF_PGM_SYNC_CR3); /** @todo no need to do global sync, right? */
return VINF_PGM_SYNC_CR3;
# endif /* !IN_RING3 */
}
else
pgmPoolMonitorModifiedClearAll(pVM);
return VINF_SUCCESS;
}
#endif /* PGMPOOL_WITH_MONITORING */
#ifdef PGMPOOL_WITH_USER_TRACKING
/**
* Frees up at least one user entry.
*
* @returns VBox status code.
* @retval VINF_SUCCESS if successfully added.
* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
* @param pPool The pool.
* @param iUser The user index.
*/
static int pgmPoolTrackFreeOneUser(PPGMPOOL pPool, uint16_t iUser)
{
STAM_COUNTER_INC(&pPool->StatTrackFreeUpOneUser);
#ifdef PGMPOOL_WITH_CACHE
/*
* Just free cached pages in a braindead fashion.
*/
/** @todo walk the age list backwards and free the first with usage. */
int rc = VINF_SUCCESS;
do
{
int rc2 = pgmPoolCacheFreeOne(pPool, iUser);
if (RT_FAILURE(rc2) && rc == VINF_SUCCESS)
rc = rc2;
} while (pPool->iUserFreeHead == NIL_PGMPOOL_USER_INDEX);
return rc;
#else
/*
* Lazy approach.
*/
/* @todo This path no longer works (CR3 root pages will be flushed)!! */
AssertCompileFailed();
Assert(!CPUMIsGuestInLongMode(pVM));
pgmPoolFlushAllInt(pPool);
return VERR_PGM_POOL_FLUSHED;
#endif
}
/**
* Inserts a page into the cache.
*
* This will create user node for the page, insert it into the GCPhys
* hash, and insert it into the age list.
*
* @returns VBox status code.
* @retval VINF_SUCCESS if successfully added.
* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
* @param pPool The pool.
* @param pPage The cached page.
* @param GCPhys The GC physical address of the page we're gonna shadow.
* @param iUser The user index.
* @param iUserTable The user table index.
*/
DECLINLINE(int) pgmPoolTrackInsert(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTGCPHYS GCPhys, uint16_t iUser, uint32_t iUserTable)
{
int rc = VINF_SUCCESS;
PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
LogFlow(("pgmPoolTrackInsert GCPhys=%RGp iUser %x iUserTable %x\n", GCPhys, iUser, iUserTable));
#ifdef VBOX_STRICT
/*
* Check that the entry doesn't already exists.
*/
if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
{
uint16_t i = pPage->iUserHead;
do
{
Assert(i < pPool->cMaxUsers);
AssertMsg(paUsers[i].iUser != iUser || paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
i = paUsers[i].iNext;
} while (i != NIL_PGMPOOL_USER_INDEX);
}
#endif
/*
* Find free a user node.
*/
uint16_t i = pPool->iUserFreeHead;
if (i == NIL_PGMPOOL_USER_INDEX)
{
int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
if (RT_FAILURE(rc))
return rc;
i = pPool->iUserFreeHead;
}
/*
* Unlink the user node from the free list,
* initialize and insert it into the user list.
*/
pPool->iUserFreeHead = paUsers[i].iNext;
paUsers[i].iNext = NIL_PGMPOOL_USER_INDEX;
paUsers[i].iUser = iUser;
paUsers[i].iUserTable = iUserTable;
pPage->iUserHead = i;
/*
* Insert into cache and enable monitoring of the guest page if enabled.
*
* Until we implement caching of all levels, including the CR3 one, we'll
* have to make sure we don't try monitor & cache any recursive reuse of
* a monitored CR3 page. Because all windows versions are doing this we'll
* have to be able to do combined access monitoring, CR3 + PT and
* PD + PT (guest PAE).
*
* Update:
* We're now cooperating with the CR3 monitor if an uncachable page is found.
*/
#if defined(PGMPOOL_WITH_MONITORING) || defined(PGMPOOL_WITH_CACHE)
# ifdef PGMPOOL_WITH_MIXED_PT_CR3
const bool fCanBeMonitored = true;
# else
bool fCanBeMonitored = pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored == NIL_RTGCPHYS
|| (GCPhys & X86_PTE_PAE_PG_MASK) != (pPool->CTX_SUFF(pVM)->pgm.s.GCPhysGstCR3Monitored & X86_PTE_PAE_PG_MASK)
|| pgmPoolIsBigPage((PGMPOOLKIND)pPage->enmKind);
# endif
# ifdef PGMPOOL_WITH_CACHE
pgmPoolCacheInsert(pPool, pPage, fCanBeMonitored); /* This can be expanded. */
# endif
if (fCanBeMonitored)
{
# ifdef PGMPOOL_WITH_MONITORING
rc = pgmPoolMonitorInsert(pPool, pPage);
AssertRC(rc);
}
# endif
#endif /* PGMPOOL_WITH_MONITORING */
return rc;
}
# ifdef PGMPOOL_WITH_CACHE /* (only used when the cache is enabled.) */
/**
* Adds a user reference to a page.
*
* This will move the page to the head of the
*
* @returns VBox status code.
* @retval VINF_SUCCESS if successfully added.
* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
* @param pPool The pool.
* @param pPage The cached page.
* @param iUser The user index.
* @param iUserTable The user table.
*/
static int pgmPoolTrackAddUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
{
PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
Log3(("pgmPoolTrackAddUser GCPhys = %RGp iUser %x iUserTable %x\n", pPage->GCPhys, iUser, iUserTable));
# ifdef VBOX_STRICT
/*
* Check that the entry doesn't already exists. We only allow multiple users of top-level paging structures (SHW_POOL_ROOT_IDX).
*/
if (pPage->iUserHead != NIL_PGMPOOL_USER_INDEX)
{
uint16_t i = pPage->iUserHead;
do
{
Assert(i < pPool->cMaxUsers);
AssertMsg(iUser != PGMPOOL_IDX_PD || iUser != PGMPOOL_IDX_PDPT || iUser != PGMPOOL_IDX_NESTED_ROOT || iUser != PGMPOOL_IDX_AMD64_CR3 ||
paUsers[i].iUser != iUser || paUsers[i].iUserTable != iUserTable, ("%x %x vs new %x %x\n", paUsers[i].iUser, paUsers[i].iUserTable, iUser, iUserTable));
i = paUsers[i].iNext;
} while (i != NIL_PGMPOOL_USER_INDEX);
}
# endif
/*
* Allocate a user node.
*/
uint16_t i = pPool->iUserFreeHead;
if (i == NIL_PGMPOOL_USER_INDEX)
{
int rc = pgmPoolTrackFreeOneUser(pPool, iUser);
if (RT_FAILURE(rc))
return rc;
i = pPool->iUserFreeHead;
}
pPool->iUserFreeHead = paUsers[i].iNext;
/*
* Initialize the user node and insert it.
*/
paUsers[i].iNext = pPage->iUserHead;
paUsers[i].iUser = iUser;
paUsers[i].iUserTable = iUserTable;
pPage->iUserHead = i;
# ifdef PGMPOOL_WITH_CACHE
/*
* Tell the cache to update its replacement stats for this page.
*/
pgmPoolCacheUsed(pPool, pPage);
# endif
return VINF_SUCCESS;
}
# endif /* PGMPOOL_WITH_CACHE */
/**
* Frees a user record associated with a page.
*
* This does not clear the entry in the user table, it simply replaces the
* user record to the chain of free records.
*
* @param pPool The pool.
* @param HCPhys The HC physical address of the shadow page.
* @param iUser The shadow page pool index of the user table.
* @param iUserTable The index into the user table (shadowed).
*/
static void pgmPoolTrackFreeUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
{
/*
* Unlink and free the specified user entry.
*/
PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
Log3(("pgmPoolTrackFreeUser %RGp %x %x\n", pPage->GCPhys, iUser, iUserTable));
/* Special: For PAE and 32-bit paging, there is usually no more than one user. */
uint16_t i = pPage->iUserHead;
if ( i != NIL_PGMPOOL_USER_INDEX
&& paUsers[i].iUser == iUser
&& paUsers[i].iUserTable == iUserTable)
{
pPage->iUserHead = paUsers[i].iNext;
paUsers[i].iUser = NIL_PGMPOOL_IDX;
paUsers[i].iNext = pPool->iUserFreeHead;
pPool->iUserFreeHead = i;
return;
}
/* General: Linear search. */
uint16_t iPrev = NIL_PGMPOOL_USER_INDEX;
while (i != NIL_PGMPOOL_USER_INDEX)
{
if ( paUsers[i].iUser == iUser
&& paUsers[i].iUserTable == iUserTable)
{
if (iPrev != NIL_PGMPOOL_USER_INDEX)
paUsers[iPrev].iNext = paUsers[i].iNext;
else
pPage->iUserHead = paUsers[i].iNext;
paUsers[i].iUser = NIL_PGMPOOL_IDX;
paUsers[i].iNext = pPool->iUserFreeHead;
pPool->iUserFreeHead = i;
return;
}
iPrev = i;
i = paUsers[i].iNext;
}
/* Fatal: didn't find it */
AssertFatalMsgFailed(("Didn't find the user entry! iUser=%#x iUserTable=%#x GCPhys=%RGp\n",
iUser, iUserTable, pPage->GCPhys));
}
/**
* Gets the entry size of a shadow table.
*
* @param enmKind The kind of page.
*
* @returns The size of the entry in bytes. That is, 4 or 8.
* @returns If the kind is not for a table, an assertion is raised and 0 is
* returned.
*/
DECLINLINE(unsigned) pgmPoolTrackGetShadowEntrySize(PGMPOOLKIND enmKind)
{
switch (enmKind)
{
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_32BIT_PD:
case PGMPOOLKIND_32BIT_PD_PHYS:
return 4;
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
case PGMPOOLKIND_64BIT_PML4:
case PGMPOOLKIND_PAE_PDPT:
case PGMPOOLKIND_ROOT_NESTED:
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PD_PHYS:
case PGMPOOLKIND_PAE_PDPT_PHYS:
return 8;
default:
AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
}
}
/**
* Gets the entry size of a guest table.
*
* @param enmKind The kind of page.
*
* @returns The size of the entry in bytes. That is, 0, 4 or 8.
* @returns If the kind is not for a table, an assertion is raised and 0 is
* returned.
*/
DECLINLINE(unsigned) pgmPoolTrackGetGuestEntrySize(PGMPOOLKIND enmKind)
{
switch (enmKind)
{
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_32BIT_PD:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
return 4;
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
case PGMPOOLKIND_64BIT_PML4:
case PGMPOOLKIND_PAE_PDPT:
return 8;
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
case PGMPOOLKIND_ROOT_NESTED:
case PGMPOOLKIND_PAE_PD_PHYS:
case PGMPOOLKIND_PAE_PDPT_PHYS:
case PGMPOOLKIND_32BIT_PD_PHYS:
/** @todo can we return 0? (nobody is calling this...) */
AssertFailed();
return 0;
default:
AssertFatalMsgFailed(("enmKind=%d\n", enmKind));
}
}
#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
/**
* Scans one shadow page table for mappings of a physical page.
*
* @param pVM The VM handle.
* @param pPhysPage The guest page in question.
* @param iShw The shadow page table.
* @param cRefs The number of references made in that PT.
*/
static void pgmPoolTrackFlushGCPhysPTInt(PVM pVM, PCPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
{
LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
/*
* Assert sanity.
*/
Assert(cRefs == 1);
AssertFatalMsg(iShw < pPool->cCurPages && iShw != NIL_PGMPOOL_IDX, ("iShw=%d\n", iShw));
PPGMPOOLPAGE pPage = &pPool->aPages[iShw];
/*
* Then, clear the actual mappings to the page in the shadow PT.
*/
switch (pPage->enmKind)
{
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
{
const uint32_t u32 = PGM_PAGE_GET_HCPHYS(pPhysPage) | X86_PTE_P;
PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
if ((pPT->a[i].u & (X86_PTE_PG_MASK | X86_PTE_P)) == u32)
{
Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX32 cRefs=%#x\n", i, pPT->a[i], cRefs));
pPT->a[i].u = 0;
cRefs--;
if (!cRefs)
return;
}
#ifdef LOG_ENABLED
RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
if ((pPT->a[i].u & (X86_PTE_PG_MASK | X86_PTE_P)) == u32)
{
RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
pPT->a[i].u = 0;
}
#endif
AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
break;
}
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
{
const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) | X86_PTE_P;
PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK | X86_PTE_P)) == u64)
{
Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
pPT->a[i].u = 0;
cRefs--;
if (!cRefs)
return;
}
#ifdef LOG_ENABLED
RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK | X86_PTE_P)) == u64)
{
RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
pPT->a[i].u = 0;
}
#endif
AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d u64=%RX64\n", cRefs, pPage->iFirstPresent, pPage->cPresent, u64));
break;
}
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
{
const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) | X86_PTE_P;
PEPTPT pPT = (PEPTPT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
if ((pPT->a[i].u & (EPT_PTE_PG_MASK | X86_PTE_P)) == u64)
{
Log4(("pgmPoolTrackFlushGCPhysPTs: i=%d pte=%RX64 cRefs=%#x\n", i, pPT->a[i], cRefs));
pPT->a[i].u = 0;
cRefs--;
if (!cRefs)
return;
}
#ifdef LOG_ENABLED
RTLogPrintf("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent);
for (unsigned i = 0; i < RT_ELEMENTS(pPT->a); i++)
if ((pPT->a[i].u & (EPT_PTE_PG_MASK | X86_PTE_P)) == u64)
{
RTLogPrintf("i=%d cRefs=%d\n", i, cRefs--);
pPT->a[i].u = 0;
}
#endif
AssertFatalMsgFailed(("cRefs=%d iFirstPresent=%d cPresent=%d\n", cRefs, pPage->iFirstPresent, pPage->cPresent));
break;
}
default:
AssertFatalMsgFailed(("enmKind=%d iShw=%d\n", pPage->enmKind, iShw));
}
}
/**
* Scans one shadow page table for mappings of a physical page.
*
* @param pVM The VM handle.
* @param pPhysPage The guest page in question.
* @param iShw The shadow page table.
* @param cRefs The number of references made in that PT.
*/
void pgmPoolTrackFlushGCPhysPT(PVM pVM, PPGMPAGE pPhysPage, uint16_t iShw, uint16_t cRefs)
{
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool); NOREF(pPool);
LogFlow(("pgmPoolTrackFlushGCPhysPT: pPhysPage=%R[pgmpage] iShw=%d cRefs=%d\n", pPhysPage, iShw, cRefs));
STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPT, f);
pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, iShw, cRefs);
PGM_PAGE_SET_TRACKING(pPhysPage, 0);
STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPT, f);
}
/**
* Flushes a list of shadow page tables mapping the same physical page.
*
* @param pVM The VM handle.
* @param pPhysPage The guest page in question.
* @param iPhysExt The physical cross reference extent list to flush.
*/
void pgmPoolTrackFlushGCPhysPTs(PVM pVM, PPGMPAGE pPhysPage, uint16_t iPhysExt)
{
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTs, f);
LogFlow(("pgmPoolTrackFlushGCPhysPTs: pPhysPage=%R[pgmpage] iPhysExt\n", pPhysPage, iPhysExt));
const uint16_t iPhysExtStart = iPhysExt;
PPGMPOOLPHYSEXT pPhysExt;
do
{
Assert(iPhysExt < pPool->cMaxPhysExts);
pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
if (pPhysExt->aidx[i] != NIL_PGMPOOL_IDX)
{
pgmPoolTrackFlushGCPhysPTInt(pVM, pPhysPage, pPhysExt->aidx[i], 1);
pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
}
/* next */
iPhysExt = pPhysExt->iNext;
} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
/* insert the list into the free list and clear the ram range entry. */
pPhysExt->iNext = pPool->iPhysExtFreeHead;
pPool->iPhysExtFreeHead = iPhysExtStart;
PGM_PAGE_SET_TRACKING(pPhysPage, 0);
STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTs, f);
}
#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
/**
* Flushes all shadow page table mappings of the given guest page.
*
* This is typically called when the host page backing the guest one has been
* replaced or when the page protection was changed due to an access handler.
*
* @returns VBox status code.
* @retval VINF_SUCCESS if all references has been successfully cleared.
* @retval VINF_PGM_SYNC_CR3 if we're better off with a CR3 sync and a page
* pool cleaning. FF and sync flags are set.
*
* @param pVM The VM handle.
* @param pPhysPage The guest page in question.
* @param pfFlushTLBs This is set to @a true if the shadow TLBs should be
* flushed, it is NOT touched if this isn't necessary.
* The caller MUST initialized this to @a false.
*/
int pgmPoolTrackFlushGCPhys(PVM pVM, PPGMPAGE pPhysPage, bool *pfFlushTLBs)
{
pgmLock(pVM);
int rc = VINF_SUCCESS;
#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
const uint16_t u16 = PGM_PAGE_GET_TRACKING(pPhysPage);
if (u16)
{
/*
* The zero page is currently screwing up the tracking and we'll
* have to flush the whole shebang. Unless VBOX_WITH_NEW_LAZY_PAGE_ALLOC
* is defined, zero pages won't normally be mapped. Some kind of solution
* will be needed for this problem of course, but it will have to wait...
*/
if (PGM_PAGE_IS_ZERO(pPhysPage))
rc = VINF_PGM_GCPHYS_ALIASED;
else
{
# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow and
pgmPoolTrackFlushGCPhysPTs will/may kill the pool otherwise. */
PVMCPU pVCpu = VMMGetCpu(pVM);
uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
# endif
if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
pgmPoolTrackFlushGCPhysPT(pVM,
pPhysPage,
PGMPOOL_TD_GET_IDX(u16),
PGMPOOL_TD_GET_CREFS(u16));
else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
pgmPoolTrackFlushGCPhysPTs(pVM, pPhysPage, PGMPOOL_TD_GET_IDX(u16));
else
rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
*pfFlushTLBs = true;
# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
# endif
}
}
#elif defined(PGMPOOL_WITH_CACHE)
if (PGM_PAGE_IS_ZERO(pPhysPage))
rc = VINF_PGM_GCPHYS_ALIASED;
else
{
# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
/* Start a subset here because pgmPoolTrackFlushGCPhysPTsSlow kill the pool otherwise. */
PVMCPU pVCpu = VMMGetCpu(pVM);
uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
# endif
rc = pgmPoolTrackFlushGCPhysPTsSlow(pVM, pPhysPage);
if (rc == VINF_SUCCESS)
*pfFlushTLBs = true;
}
# ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
# endif
#else
rc = VINF_PGM_GCPHYS_ALIASED;
#endif
if (rc == VINF_PGM_GCPHYS_ALIASED)
{
pVM->pgm.s.fGlobalSyncFlags |= PGM_GLOBAL_SYNC_CLEAR_PGM_POOL;
for (unsigned i=0;i<pVM->cCPUs;i++)
{
PVMCPU pVCpu = &pVM->aCpus[i];
VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
}
rc = VINF_PGM_SYNC_CR3;
}
pgmUnlock(pVM);
return rc;
}
/**
* Scans all shadow page tables for mappings of a physical page.
*
* This may be slow, but it's most likely more efficient than cleaning
* out the entire page pool / cache.
*
* @returns VBox status code.
* @retval VINF_SUCCESS if all references has been successfully cleared.
* @retval VINF_PGM_GCPHYS_ALIASED if we're better off with a CR3 sync and
* a page pool cleaning.
*
* @param pVM The VM handle.
* @param pPhysPage The guest page in question.
*/
int pgmPoolTrackFlushGCPhysPTsSlow(PVM pVM, PPGMPAGE pPhysPage)
{
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
STAM_PROFILE_START(&pPool->StatTrackFlushGCPhysPTsSlow, s);
LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: cUsedPages=%d cPresent=%d pPhysPage=%R[pgmpage]\n",
pPool->cUsedPages, pPool->cPresent, pPhysPage));
#if 1
/*
* There is a limit to what makes sense.
*/
if (pPool->cPresent > 1024)
{
LogFlow(("pgmPoolTrackFlushGCPhysPTsSlow: giving up... (cPresent=%d)\n", pPool->cPresent));
STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
return VINF_PGM_GCPHYS_ALIASED;
}
#endif
/*
* Iterate all the pages until we've encountered all that in use.
* This is simple but not quite optimal solution.
*/
const uint64_t u64 = PGM_PAGE_GET_HCPHYS(pPhysPage) | X86_PTE_P;
const uint32_t u32 = u64;
unsigned cLeft = pPool->cUsedPages;
unsigned iPage = pPool->cCurPages;
while (--iPage >= PGMPOOL_IDX_FIRST)
{
PPGMPOOLPAGE pPage = &pPool->aPages[iPage];
if (pPage->GCPhys != NIL_RTGCPHYS)
{
switch (pPage->enmKind)
{
/*
* We only care about shadow page tables.
*/
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
{
unsigned cPresent = pPage->cPresent;
PX86PT pPT = (PX86PT)PGMPOOL_PAGE_2_PTR(pVM, pPage);
for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
if (pPT->a[i].n.u1Present)
{
if ((pPT->a[i].u & (X86_PTE_PG_MASK | X86_PTE_P)) == u32)
{
//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX32\n", iPage, i, pPT->a[i]));
pPT->a[i].u = 0;
}
if (!--cPresent)
break;
}
break;
}
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
{
unsigned cPresent = pPage->cPresent;
PX86PTPAE pPT = (PX86PTPAE)PGMPOOL_PAGE_2_PTR(pVM, pPage);
for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pPT->a); i++)
if (pPT->a[i].n.u1Present)
{
if ((pPT->a[i].u & (X86_PTE_PAE_PG_MASK | X86_PTE_P)) == u64)
{
//Log4(("pgmPoolTrackFlushGCPhysPTsSlow: idx=%d i=%d pte=%RX64\n", iPage, i, pPT->a[i]));
pPT->a[i].u = 0;
}
if (!--cPresent)
break;
}
break;
}
}
if (!--cLeft)
break;
}
}
PGM_PAGE_SET_TRACKING(pPhysPage, 0);
STAM_PROFILE_STOP(&pPool->StatTrackFlushGCPhysPTsSlow, s);
return VINF_SUCCESS;
}
/**
* Clears the user entry in a user table.
*
* This is used to remove all references to a page when flushing it.
*/
static void pgmPoolTrackClearPageUser(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PCPGMPOOLUSER pUser)
{
Assert(pUser->iUser != NIL_PGMPOOL_IDX);
Assert(pUser->iUser < pPool->cCurPages);
uint32_t iUserTable = pUser->iUserTable;
/*
* Map the user page.
*/
PPGMPOOLPAGE pUserPage = &pPool->aPages[pUser->iUser];
union
{
uint64_t *pau64;
uint32_t *pau32;
} u;
u.pau64 = (uint64_t *)PGMPOOL_PAGE_2_PTR(pPool->CTX_SUFF(pVM), pUserPage);
LogFlow(("pgmPoolTrackClearPageUser: clear %x in %s (%RGp) (flushing %s)\n", iUserTable, pgmPoolPoolKindToStr(pUserPage->enmKind), pUserPage->Core.Key, pgmPoolPoolKindToStr(pPage->enmKind)));
/* Safety precaution in case we change the paging for other modes too in the future. */
Assert(!pgmPoolIsPageLocked(&pPool->CTX_SUFF(pVM)->pgm.s, pPage));
#ifdef VBOX_STRICT
/*
* Some sanity checks.
*/
switch (pUserPage->enmKind)
{
case PGMPOOLKIND_32BIT_PD:
case PGMPOOLKIND_32BIT_PD_PHYS:
Assert(iUserTable < X86_PG_ENTRIES);
break;
case PGMPOOLKIND_PAE_PDPT:
case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
case PGMPOOLKIND_PAE_PDPT_PHYS:
Assert(iUserTable < 4);
Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
break;
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
case PGMPOOLKIND_PAE_PD_PHYS:
Assert(iUserTable < X86_PG_PAE_ENTRIES);
break;
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
Assert(iUserTable < X86_PG_PAE_ENTRIES);
Assert(!(u.pau64[iUserTable] & PGM_PDFLAGS_MAPPING));
break;
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
Assert(iUserTable < X86_PG_PAE_ENTRIES);
Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
break;
case PGMPOOLKIND_64BIT_PML4:
Assert(!(u.pau64[iUserTable] & PGM_PLXFLAGS_PERMANENT));
/* GCPhys >> PAGE_SHIFT is the index here */
break;
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
Assert(iUserTable < X86_PG_PAE_ENTRIES);
break;
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
Assert(iUserTable < X86_PG_PAE_ENTRIES);
break;
case PGMPOOLKIND_ROOT_NESTED:
Assert(iUserTable < X86_PG_PAE_ENTRIES);
break;
default:
AssertMsgFailed(("enmKind=%d\n", pUserPage->enmKind));
break;
}
#endif /* VBOX_STRICT */
/*
* Clear the entry in the user page.
*/
switch (pUserPage->enmKind)
{
/* 32-bit entries */
case PGMPOOLKIND_32BIT_PD:
case PGMPOOLKIND_32BIT_PD_PHYS:
u.pau32[iUserTable] = 0;
break;
/* 64-bit entries */
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
#if defined(IN_RC)
/* In 32 bits PAE mode we *must* invalidate the TLB when changing a PDPT entry; the CPU fetches them only during cr3 load, so any
* non-present PDPT will continue to cause page faults.
*/
ASMReloadCR3();
#endif
/* no break */
case PGMPOOLKIND_PAE_PD_PHYS:
case PGMPOOLKIND_PAE_PDPT_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
case PGMPOOLKIND_64BIT_PML4:
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
case PGMPOOLKIND_PAE_PDPT:
case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
case PGMPOOLKIND_ROOT_NESTED:
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
u.pau64[iUserTable] = 0;
break;
default:
AssertFatalMsgFailed(("enmKind=%d iUser=%#x iUserTable=%#x\n", pUserPage->enmKind, pUser->iUser, pUser->iUserTable));
}
}
/**
* Clears all users of a page.
*/
static void pgmPoolTrackClearPageUsers(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
{
/*
* Free all the user records.
*/
LogFlow(("pgmPoolTrackClearPageUsers %RGp\n", pPage->GCPhys));
PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
uint16_t i = pPage->iUserHead;
while (i != NIL_PGMPOOL_USER_INDEX)
{
/* Clear enter in user table. */
pgmPoolTrackClearPageUser(pPool, pPage, &paUsers[i]);
/* Free it. */
const uint16_t iNext = paUsers[i].iNext;
paUsers[i].iUser = NIL_PGMPOOL_IDX;
paUsers[i].iNext = pPool->iUserFreeHead;
pPool->iUserFreeHead = i;
/* Next. */
i = iNext;
}
pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
}
#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
/**
* Allocates a new physical cross reference extent.
*
* @returns Pointer to the allocated extent on success. NULL if we're out of them.
* @param pVM The VM handle.
* @param piPhysExt Where to store the phys ext index.
*/
PPGMPOOLPHYSEXT pgmPoolTrackPhysExtAlloc(PVM pVM, uint16_t *piPhysExt)
{
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
uint16_t iPhysExt = pPool->iPhysExtFreeHead;
if (iPhysExt == NIL_PGMPOOL_PHYSEXT_INDEX)
{
STAM_COUNTER_INC(&pPool->StamTrackPhysExtAllocFailures);
return NULL;
}
PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
pPool->iPhysExtFreeHead = pPhysExt->iNext;
pPhysExt->iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
*piPhysExt = iPhysExt;
return pPhysExt;
}
/**
* Frees a physical cross reference extent.
*
* @param pVM The VM handle.
* @param iPhysExt The extent to free.
*/
void pgmPoolTrackPhysExtFree(PVM pVM, uint16_t iPhysExt)
{
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
Assert(iPhysExt < pPool->cMaxPhysExts);
PPGMPOOLPHYSEXT pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
pPhysExt->iNext = pPool->iPhysExtFreeHead;
pPool->iPhysExtFreeHead = iPhysExt;
}
/**
* Frees a physical cross reference extent.
*
* @param pVM The VM handle.
* @param iPhysExt The extent to free.
*/
void pgmPoolTrackPhysExtFreeList(PVM pVM, uint16_t iPhysExt)
{
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
const uint16_t iPhysExtStart = iPhysExt;
PPGMPOOLPHYSEXT pPhysExt;
do
{
Assert(iPhysExt < pPool->cMaxPhysExts);
pPhysExt = &pPool->CTX_SUFF(paPhysExts)[iPhysExt];
for (unsigned i = 0; i < RT_ELEMENTS(pPhysExt->aidx); i++)
pPhysExt->aidx[i] = NIL_PGMPOOL_IDX;
/* next */
iPhysExt = pPhysExt->iNext;
} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
pPhysExt->iNext = pPool->iPhysExtFreeHead;
pPool->iPhysExtFreeHead = iPhysExtStart;
}
/**
* Insert a reference into a list of physical cross reference extents.
*
* @returns The new tracking data for PGMPAGE.
*
* @param pVM The VM handle.
* @param iPhysExt The physical extent index of the list head.
* @param iShwPT The shadow page table index.
*
*/
static uint16_t pgmPoolTrackPhysExtInsert(PVM pVM, uint16_t iPhysExt, uint16_t iShwPT)
{
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
/* special common case. */
if (paPhysExts[iPhysExt].aidx[2] == NIL_PGMPOOL_IDX)
{
paPhysExts[iPhysExt].aidx[2] = iShwPT;
STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
LogFlow(("pgmPoolTrackPhysExtAddref: %d:{,,%d}\n", iPhysExt, iShwPT));
return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
}
/* general treatment. */
const uint16_t iPhysExtStart = iPhysExt;
unsigned cMax = 15;
for (;;)
{
Assert(iPhysExt < pPool->cMaxPhysExts);
for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
if (paPhysExts[iPhysExt].aidx[i] == NIL_PGMPOOL_IDX)
{
paPhysExts[iPhysExt].aidx[i] = iShwPT;
STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedMany);
LogFlow(("pgmPoolTrackPhysExtAddref: %d:{%d} i=%d cMax=%d\n", iPhysExt, iShwPT, i, cMax));
return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtStart);
}
if (!--cMax)
{
STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
LogFlow(("pgmPoolTrackPhysExtAddref: overflow (1) iShwPT=%d\n", iShwPT));
return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
}
}
/* add another extent to the list. */
PPGMPOOLPHYSEXT pNew = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
if (!pNew)
{
STAM_COUNTER_INC(&pVM->pgm.s.StatTrackOverflows);
pgmPoolTrackPhysExtFreeList(pVM, iPhysExtStart);
return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
}
pNew->iNext = iPhysExtStart;
pNew->aidx[0] = iShwPT;
LogFlow(("pgmPoolTrackPhysExtAddref: added new extent %d:{%d}->%d\n", iPhysExt, iShwPT, iPhysExtStart));
return PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
}
/**
* Add a reference to guest physical page where extents are in use.
*
* @returns The new tracking data for PGMPAGE.
*
* @param pVM The VM handle.
* @param u16 The ram range flags (top 16-bits).
* @param iShwPT The shadow page table index.
*/
uint16_t pgmPoolTrackPhysExtAddref(PVM pVM, uint16_t u16, uint16_t iShwPT)
{
pgmLock(pVM);
if (PGMPOOL_TD_GET_CREFS(u16) != PGMPOOL_TD_CREFS_PHYSEXT)
{
/*
* Convert to extent list.
*/
Assert(PGMPOOL_TD_GET_CREFS(u16) == 1);
uint16_t iPhysExt;
PPGMPOOLPHYSEXT pPhysExt = pgmPoolTrackPhysExtAlloc(pVM, &iPhysExt);
if (pPhysExt)
{
LogFlow(("pgmPoolTrackPhysExtAddref: new extent: %d:{%d, %d}\n", iPhysExt, PGMPOOL_TD_GET_IDX(u16), iShwPT));
STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliased);
pPhysExt->aidx[0] = PGMPOOL_TD_GET_IDX(u16);
pPhysExt->aidx[1] = iShwPT;
u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExt);
}
else
u16 = PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED);
}
else if (u16 != PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, PGMPOOL_TD_IDX_OVERFLOWED))
{
/*
* Insert into the extent list.
*/
u16 = pgmPoolTrackPhysExtInsert(pVM, PGMPOOL_TD_GET_IDX(u16), iShwPT);
}
else
STAM_COUNTER_INC(&pVM->pgm.s.StatTrackAliasedLots);
pgmUnlock(pVM);
return u16;
}
/**
* Clear references to guest physical memory.
*
* @param pPool The pool.
* @param pPage The page.
* @param pPhysPage Pointer to the aPages entry in the ram range.
*/
void pgmPoolTrackPhysExtDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PPGMPAGE pPhysPage)
{
const unsigned cRefs = PGM_PAGE_GET_TD_CREFS(pPhysPage);
AssertFatalMsg(cRefs == PGMPOOL_TD_CREFS_PHYSEXT, ("cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
uint16_t iPhysExt = PGM_PAGE_GET_TD_IDX(pPhysPage);
if (iPhysExt != PGMPOOL_TD_IDX_OVERFLOWED)
{
PVM pVM = pPool->CTX_SUFF(pVM);
pgmLock(pVM);
uint16_t iPhysExtPrev = NIL_PGMPOOL_PHYSEXT_INDEX;
PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
do
{
Assert(iPhysExt < pPool->cMaxPhysExts);
/*
* Look for the shadow page and check if it's all freed.
*/
for (unsigned i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
{
if (paPhysExts[iPhysExt].aidx[i] == pPage->idx)
{
paPhysExts[iPhysExt].aidx[i] = NIL_PGMPOOL_IDX;
for (i = 0; i < RT_ELEMENTS(paPhysExts[iPhysExt].aidx); i++)
if (paPhysExts[iPhysExt].aidx[i] != NIL_PGMPOOL_IDX)
{
Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
pgmUnlock(pVM);
return;
}
/* we can free the node. */
const uint16_t iPhysExtNext = paPhysExts[iPhysExt].iNext;
if ( iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX
&& iPhysExtNext == NIL_PGMPOOL_PHYSEXT_INDEX)
{
/* lonely node */
pgmPoolTrackPhysExtFree(pVM, iPhysExt);
Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d lonely\n", pPhysPage, pPage->idx));
PGM_PAGE_SET_TRACKING(pPhysPage, 0);
}
else if (iPhysExtPrev == NIL_PGMPOOL_PHYSEXT_INDEX)
{
/* head */
Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d head\n", pPhysPage, pPage->idx));
PGM_PAGE_SET_TRACKING(pPhysPage, PGMPOOL_TD_MAKE(PGMPOOL_TD_CREFS_PHYSEXT, iPhysExtNext));
pgmPoolTrackPhysExtFree(pVM, iPhysExt);
}
else
{
/* in list */
Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage] idx=%d\n", pPhysPage, pPage->idx));
paPhysExts[iPhysExtPrev].iNext = iPhysExtNext;
pgmPoolTrackPhysExtFree(pVM, iPhysExt);
}
iPhysExt = iPhysExtNext;
pgmUnlock(pVM);
return;
}
}
/* next */
iPhysExtPrev = iPhysExt;
iPhysExt = paPhysExts[iPhysExt].iNext;
} while (iPhysExt != NIL_PGMPOOL_PHYSEXT_INDEX);
pgmUnlock(pVM);
AssertFatalMsgFailed(("not-found! cRefs=%d pPhysPage=%R[pgmpage] pPage=%p:{.idx=%d}\n", cRefs, pPhysPage, pPage, pPage->idx));
}
else /* nothing to do */
Log2(("pgmPoolTrackPhysExtDerefGCPhys: pPhysPage=%R[pgmpage]\n", pPhysPage));
}
/**
* Clear references to guest physical memory.
*
* This is the same as pgmPoolTracDerefGCPhys except that the guest physical address
* is assumed to be correct, so the linear search can be skipped and we can assert
* at an earlier point.
*
* @param pPool The pool.
* @param pPage The page.
* @param HCPhys The host physical address corresponding to the guest page.
* @param GCPhys The guest physical address corresponding to HCPhys.
*/
static void pgmPoolTracDerefGCPhys(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhys)
{
/*
* Walk range list.
*/
PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
while (pRam)
{
RTGCPHYS off = GCPhys - pRam->GCPhys;
if (off < pRam->cb)
{
/* does it match? */
const unsigned iPage = off >> PAGE_SHIFT;
Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
#ifdef LOG_ENABLED
RTHCPHYS HCPhysPage = PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]);
Log2(("pgmPoolTracDerefGCPhys %RHp vs %RHp\n", HCPhysPage, HCPhys));
#endif
if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
{
pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
return;
}
break;
}
pRam = pRam->CTX_SUFF(pNext);
}
AssertFatalMsgFailed(("HCPhys=%RHp GCPhys=%RGp\n", HCPhys, GCPhys));
}
/**
* Clear references to guest physical memory.
*
* @param pPool The pool.
* @param pPage The page.
* @param HCPhys The host physical address corresponding to the guest page.
* @param GCPhysHint The guest physical address which may corresponding to HCPhys.
*/
static void pgmPoolTracDerefGCPhysHint(PPGMPOOL pPool, PPGMPOOLPAGE pPage, RTHCPHYS HCPhys, RTGCPHYS GCPhysHint)
{
Log4(("pgmPoolTracDerefGCPhysHint %RHp %RGp\n", HCPhys, GCPhysHint));
/*
* Walk range list.
*/
PPGMRAMRANGE pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
while (pRam)
{
RTGCPHYS off = GCPhysHint - pRam->GCPhys;
if (off < pRam->cb)
{
/* does it match? */
const unsigned iPage = off >> PAGE_SHIFT;
Assert(PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]));
if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
{
pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
return;
}
break;
}
pRam = pRam->CTX_SUFF(pNext);
}
/*
* Damn, the hint didn't work. We'll have to do an expensive linear search.
*/
STAM_COUNTER_INC(&pPool->StatTrackLinearRamSearches);
pRam = pPool->CTX_SUFF(pVM)->pgm.s.CTX_SUFF(pRamRanges);
while (pRam)
{
unsigned iPage = pRam->cb >> PAGE_SHIFT;
while (iPage-- > 0)
{
if (PGM_PAGE_GET_HCPHYS(&pRam->aPages[iPage]) == HCPhys)
{
Log4(("pgmPoolTracDerefGCPhysHint: Linear HCPhys=%RHp GCPhysHint=%RGp GCPhysReal=%RGp\n",
HCPhys, GCPhysHint, pRam->GCPhys + (iPage << PAGE_SHIFT)));
pgmTrackDerefGCPhys(pPool, pPage, &pRam->aPages[iPage]);
return;
}
}
pRam = pRam->CTX_SUFF(pNext);
}
AssertFatalMsgFailed(("HCPhys=%RHp GCPhysHint=%RGp\n", HCPhys, GCPhysHint));
}
/**
* Clear references to guest physical memory in a 32-bit / 32-bit page table.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPT The shadow page table (mapping of the page).
* @param pGstPT The guest page table.
*/
DECLINLINE(void) pgmPoolTrackDerefPT32Bit32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT, PCX86PT pGstPT)
{
for (unsigned i = pPage->iFirstPresent; i < RT_ELEMENTS(pShwPT->a); i++)
if (pShwPT->a[i].n.u1Present)
{
Log4(("pgmPoolTrackDerefPT32Bit32Bit: i=%d pte=%RX32 hint=%RX32\n",
i, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
if (!--pPage->cPresent)
break;
}
}
/**
* Clear references to guest physical memory in a PAE / 32-bit page table.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPT The shadow page table (mapping of the page).
* @param pGstPT The guest page table (just a half one).
*/
DECLINLINE(void) pgmPoolTrackDerefPTPae32Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PT pGstPT)
{
for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
if (pShwPT->a[i].n.u1Present)
{
Log4(("pgmPoolTrackDerefPTPae32Bit: i=%d pte=%RX64 hint=%RX32\n",
i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK));
pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PG_MASK);
}
}
/**
* Clear references to guest physical memory in a PAE / PAE page table.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPT The shadow page table (mapping of the page).
* @param pGstPT The guest page table.
*/
DECLINLINE(void) pgmPoolTrackDerefPTPaePae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT, PCX86PTPAE pGstPT)
{
for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++)
if (pShwPT->a[i].n.u1Present)
{
Log4(("pgmPoolTrackDerefPTPaePae: i=%d pte=%RX32 hint=%RX32\n",
i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK));
pgmPoolTracDerefGCPhysHint(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, pGstPT->a[i].u & X86_PTE_PAE_PG_MASK);
}
}
/**
* Clear references to guest physical memory in a 32-bit / 4MB page table.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPT The shadow page table (mapping of the page).
*/
DECLINLINE(void) pgmPoolTrackDerefPT32Bit4MB(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PT pShwPT)
{
RTGCPHYS GCPhys = pPage->GCPhys;
for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
if (pShwPT->a[i].n.u1Present)
{
Log4(("pgmPoolTrackDerefPT32Bit4MB: i=%d pte=%RX32 GCPhys=%RGp\n",
i, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys));
pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PG_MASK, GCPhys);
}
}
/**
* Clear references to guest physical memory in a PAE / 2/4MB page table.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPT The shadow page table (mapping of the page).
*/
DECLINLINE(void) pgmPoolTrackDerefPTPaeBig(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PTPAE pShwPT)
{
RTGCPHYS GCPhys = pPage->GCPhys;
for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
if (pShwPT->a[i].n.u1Present)
{
Log4(("pgmPoolTrackDerefPTPaeBig: i=%d pte=%RX64 hint=%RGp\n",
i, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys));
pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & X86_PTE_PAE_PG_MASK, GCPhys);
}
}
#endif /* PGMPOOL_WITH_GCPHYS_TRACKING */
/**
* Clear references to shadowed pages in a 32 bits page directory.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPD The shadow page directory (mapping of the page).
*/
DECLINLINE(void) pgmPoolTrackDerefPD(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PD pShwPD)
{
for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
{
if ( pShwPD->a[i].n.u1Present
&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
)
{
PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PG_MASK);
if (pSubPage)
pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
else
AssertFatalMsgFailed(("%x\n", pShwPD->a[i].u & X86_PDE_PG_MASK));
}
}
}
/**
* Clear references to shadowed pages in a PAE (legacy or 64 bits) page directory.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPD The shadow page directory (mapping of the page).
*/
DECLINLINE(void) pgmPoolTrackDerefPDPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPAE pShwPD)
{
for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
{
if ( pShwPD->a[i].n.u1Present
&& !(pShwPD->a[i].u & PGM_PDFLAGS_MAPPING)
)
{
PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & X86_PDE_PAE_PG_MASK);
if (pSubPage)
pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
else
AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & X86_PDE_PAE_PG_MASK));
/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
}
}
}
/**
* Clear references to shadowed pages in a PAE page directory pointer table.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
*/
DECLINLINE(void) pgmPoolTrackDerefPDPTPae(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
{
for (unsigned i = 0; i < X86_PG_PAE_PDPE_ENTRIES; i++)
{
if ( pShwPDPT->a[i].n.u1Present
&& !(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING)
)
{
PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
if (pSubPage)
pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
else
AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
}
}
}
/**
* Clear references to shadowed pages in a 64-bit page directory pointer table.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
*/
DECLINLINE(void) pgmPoolTrackDerefPDPT64Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PDPT pShwPDPT)
{
for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
{
Assert(!(pShwPDPT->a[i].u & PGM_PLXFLAGS_MAPPING));
if (pShwPDPT->a[i].n.u1Present)
{
PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & X86_PDPE_PG_MASK);
if (pSubPage)
pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
else
AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & X86_PDPE_PG_MASK));
/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
}
}
}
/**
* Clear references to shadowed pages in a 64-bit level 4 page table.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
*/
DECLINLINE(void) pgmPoolTrackDerefPML464Bit(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PX86PML4 pShwPML4)
{
for (unsigned i = 0; i < RT_ELEMENTS(pShwPML4->a); i++)
{
if (pShwPML4->a[i].n.u1Present)
{
PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPML4->a[i].u & X86_PDPE_PG_MASK);
if (pSubPage)
pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
else
AssertFatalMsgFailed(("%RX64\n", pShwPML4->a[i].u & X86_PML4E_PG_MASK));
/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
}
}
}
/**
* Clear references to shadowed pages in an EPT page table.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPML4 The shadow page directory pointer table (mapping of the page).
*/
DECLINLINE(void) pgmPoolTrackDerefPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPT pShwPT)
{
RTGCPHYS GCPhys = pPage->GCPhys;
for (unsigned i = 0; i < RT_ELEMENTS(pShwPT->a); i++, GCPhys += PAGE_SIZE)
if (pShwPT->a[i].n.u1Present)
{
Log4(("pgmPoolTrackDerefPTEPT: i=%d pte=%RX64 GCPhys=%RX64\n",
i, pShwPT->a[i].u & EPT_PTE_PG_MASK, pPage->GCPhys));
pgmPoolTracDerefGCPhys(pPool, pPage, pShwPT->a[i].u & EPT_PTE_PG_MASK, GCPhys);
}
}
/**
* Clear references to shadowed pages in an EPT page directory.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPD The shadow page directory (mapping of the page).
*/
DECLINLINE(void) pgmPoolTrackDerefPDEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPD pShwPD)
{
for (unsigned i = 0; i < RT_ELEMENTS(pShwPD->a); i++)
{
if (pShwPD->a[i].n.u1Present)
{
PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPD->a[i].u & EPT_PDE_PG_MASK);
if (pSubPage)
pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
else
AssertFatalMsgFailed(("%RX64\n", pShwPD->a[i].u & EPT_PDE_PG_MASK));
/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
}
}
}
/**
* Clear references to shadowed pages in an EPT page directory pointer table.
*
* @param pPool The pool.
* @param pPage The page.
* @param pShwPDPT The shadow page directory pointer table (mapping of the page).
*/
DECLINLINE(void) pgmPoolTrackDerefPDPTEPT(PPGMPOOL pPool, PPGMPOOLPAGE pPage, PEPTPDPT pShwPDPT)
{
for (unsigned i = 0; i < RT_ELEMENTS(pShwPDPT->a); i++)
{
if (pShwPDPT->a[i].n.u1Present)
{
PPGMPOOLPAGE pSubPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK);
if (pSubPage)
pgmPoolTrackFreeUser(pPool, pSubPage, pPage->idx, i);
else
AssertFatalMsgFailed(("%RX64\n", pShwPDPT->a[i].u & EPT_PDPTE_PG_MASK));
/** @todo 64-bit guests: have to ensure that we're not exhausting the dynamic mappings! */
}
}
}
/**
* Clears all references made by this page.
*
* This includes other shadow pages and GC physical addresses.
*
* @param pPool The pool.
* @param pPage The page.
*/
static void pgmPoolTrackDeref(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
{
/*
* Map the shadow page and take action according to the page kind.
*/
void *pvShw = PGMPOOL_PAGE_2_LOCKED_PTR(pPool->CTX_SUFF(pVM), pPage);
switch (pPage->enmKind)
{
#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
{
STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
void *pvGst;
int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
pgmPoolTrackDerefPT32Bit32Bit(pPool, pPage, (PX86PT)pvShw, (PCX86PT)pvGst);
STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
break;
}
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
{
STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
void *pvGst;
int rc = PGM_GCPHYS_2_PTR_EX(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
pgmPoolTrackDerefPTPae32Bit(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PT)pvGst);
STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
break;
}
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
{
STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
void *pvGst;
int rc = PGM_GCPHYS_2_PTR(pPool->CTX_SUFF(pVM), pPage->GCPhys, &pvGst); AssertReleaseRC(rc);
pgmPoolTrackDerefPTPaePae(pPool, pPage, (PX86PTPAE)pvShw, (PCX86PTPAE)pvGst);
STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
break;
}
case PGMPOOLKIND_32BIT_PT_FOR_PHYS: /* treat it like a 4 MB page */
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
{
STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
pgmPoolTrackDerefPT32Bit4MB(pPool, pPage, (PX86PT)pvShw);
STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
break;
}
case PGMPOOLKIND_PAE_PT_FOR_PHYS: /* treat it like a 2 MB page */
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
{
STAM_PROFILE_START(&pPool->StatTrackDerefGCPhys, g);
pgmPoolTrackDerefPTPaeBig(pPool, pPage, (PX86PTPAE)pvShw);
STAM_PROFILE_STOP(&pPool->StatTrackDerefGCPhys, g);
break;
}
#else /* !PGMPOOL_WITH_GCPHYS_TRACKING */
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
break;
#endif /* !PGMPOOL_WITH_GCPHYS_TRACKING */
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
case PGMPOOLKIND_PAE_PD_PHYS:
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
pgmPoolTrackDerefPDPae(pPool, pPage, (PX86PDPAE)pvShw);
break;
case PGMPOOLKIND_32BIT_PD_PHYS:
case PGMPOOLKIND_32BIT_PD:
pgmPoolTrackDerefPD(pPool, pPage, (PX86PD)pvShw);
break;
case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
case PGMPOOLKIND_PAE_PDPT:
case PGMPOOLKIND_PAE_PDPT_PHYS:
pgmPoolTrackDerefPDPTPae(pPool, pPage, (PX86PDPT)pvShw);
break;
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
pgmPoolTrackDerefPDPT64Bit(pPool, pPage, (PX86PDPT)pvShw);
break;
case PGMPOOLKIND_64BIT_PML4:
pgmPoolTrackDerefPML464Bit(pPool, pPage, (PX86PML4)pvShw);
break;
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
pgmPoolTrackDerefPTEPT(pPool, pPage, (PEPTPT)pvShw);
break;
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
pgmPoolTrackDerefPDEPT(pPool, pPage, (PEPTPD)pvShw);
break;
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
pgmPoolTrackDerefPDPTEPT(pPool, pPage, (PEPTPDPT)pvShw);
break;
default:
AssertFatalMsgFailed(("enmKind=%d\n", pPage->enmKind));
}
/* paranoia, clear the shadow page. Remove this laser (i.e. let Alloc and ClearAll do it). */
STAM_PROFILE_START(&pPool->StatZeroPage, z);
ASMMemZeroPage(pvShw);
STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
pPage->fZeroed = true;
PGMPOOL_UNLOCK_PTR(pPool->CTX_SUFF(pVM), pvShw);
}
#endif /* PGMPOOL_WITH_USER_TRACKING */
#ifdef IN_RING3
/**
* Flushes the entire cache.
*
* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
* and execute this CR3 flush.
*
* @param pPool The pool.
*
* @remark Only used during reset now, we might want to rename and/or move it.
*/
static void pgmPoolFlushAllInt(PPGMPOOL pPool)
{
PVM pVM = pPool->CTX_SUFF(pVM);
STAM_PROFILE_START(&pPool->StatFlushAllInt, a);
LogFlow(("pgmPoolFlushAllInt:\n"));
/*
* If there are no pages in the pool, there is nothing to do.
*/
if (pPool->cCurPages <= PGMPOOL_IDX_FIRST)
{
STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
return;
}
/*
* Exit the shadow mode since we're going to clear everything,
* including the root page.
*/
/** @todo Need to synchronize this across all VCPUs! */
Assert(pVM->cCPUs == 1);
for (unsigned i=0;i<pVM->cCPUs;i++)
{
PVMCPU pVCpu = &pVM->aCpus[i];
pgmR3ExitShadowModeBeforePoolFlush(pVM, pVCpu);
}
/*
* Nuke the free list and reinsert all pages into it.
*/
for (unsigned i = pPool->cCurPages - 1; i >= PGMPOOL_IDX_FIRST; i--)
{
PPGMPOOLPAGE pPage = &pPool->aPages[i];
Assert(pPage->Core.Key == MMPage2Phys(pVM, pPage->pvPageR3));
#ifdef PGMPOOL_WITH_MONITORING
if (pPage->fMonitored)
pgmPoolMonitorFlush(pPool, pPage);
pPage->iModifiedNext = NIL_PGMPOOL_IDX;
pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
pPage->cModifications = 0;
#endif
pPage->GCPhys = NIL_RTGCPHYS;
pPage->enmKind = PGMPOOLKIND_FREE;
Assert(pPage->idx == i);
pPage->iNext = i + 1;
pPage->fZeroed = false; /* This could probably be optimized, but better safe than sorry. */
pPage->fSeenNonGlobal = false;
pPage->fMonitored= false;
pPage->fCached = false;
pPage->fReusedFlushPending = false;
#ifdef PGMPOOL_WITH_USER_TRACKING
pPage->iUserHead = NIL_PGMPOOL_USER_INDEX;
#else
pPage->fCR3Mix = false;
#endif
#ifdef PGMPOOL_WITH_CACHE
pPage->iAgeNext = NIL_PGMPOOL_IDX;
pPage->iAgePrev = NIL_PGMPOOL_IDX;
#endif
pPage->cLocked = 0;
}
pPool->aPages[pPool->cCurPages - 1].iNext = NIL_PGMPOOL_IDX;
pPool->iFreeHead = PGMPOOL_IDX_FIRST;
pPool->cUsedPages = 0;
#ifdef PGMPOOL_WITH_USER_TRACKING
/*
* Zap and reinitialize the user records.
*/
pPool->cPresent = 0;
pPool->iUserFreeHead = 0;
PPGMPOOLUSER paUsers = pPool->CTX_SUFF(paUsers);
const unsigned cMaxUsers = pPool->cMaxUsers;
for (unsigned i = 0; i < cMaxUsers; i++)
{
paUsers[i].iNext = i + 1;
paUsers[i].iUser = NIL_PGMPOOL_IDX;
paUsers[i].iUserTable = 0xfffffffe;
}
paUsers[cMaxUsers - 1].iNext = NIL_PGMPOOL_USER_INDEX;
#endif
#ifdef PGMPOOL_WITH_GCPHYS_TRACKING
/*
* Clear all the GCPhys links and rebuild the phys ext free list.
*/
for (PPGMRAMRANGE pRam = pVM->pgm.s.CTX_SUFF(pRamRanges);
pRam;
pRam = pRam->CTX_SUFF(pNext))
{
unsigned iPage = pRam->cb >> PAGE_SHIFT;
while (iPage-- > 0)
PGM_PAGE_SET_TRACKING(&pRam->aPages[iPage], 0);
}
pPool->iPhysExtFreeHead = 0;
PPGMPOOLPHYSEXT paPhysExts = pPool->CTX_SUFF(paPhysExts);
const unsigned cMaxPhysExts = pPool->cMaxPhysExts;
for (unsigned i = 0; i < cMaxPhysExts; i++)
{
paPhysExts[i].iNext = i + 1;
paPhysExts[i].aidx[0] = NIL_PGMPOOL_IDX;
paPhysExts[i].aidx[1] = NIL_PGMPOOL_IDX;
paPhysExts[i].aidx[2] = NIL_PGMPOOL_IDX;
}
paPhysExts[cMaxPhysExts - 1].iNext = NIL_PGMPOOL_PHYSEXT_INDEX;
#endif
#ifdef PGMPOOL_WITH_MONITORING
/*
* Just zap the modified list.
*/
pPool->cModifiedPages = 0;
pPool->iModifiedHead = NIL_PGMPOOL_IDX;
#endif
#ifdef PGMPOOL_WITH_CACHE
/*
* Clear the GCPhys hash and the age list.
*/
for (unsigned i = 0; i < RT_ELEMENTS(pPool->aiHash); i++)
pPool->aiHash[i] = NIL_PGMPOOL_IDX;
pPool->iAgeHead = NIL_PGMPOOL_IDX;
pPool->iAgeTail = NIL_PGMPOOL_IDX;
#endif
/*
* Reinsert active pages into the hash and ensure monitoring chains are correct.
*/
for (unsigned i = PGMPOOL_IDX_FIRST_SPECIAL; i < PGMPOOL_IDX_FIRST; i++)
{
PPGMPOOLPAGE pPage = &pPool->aPages[i];
pPage->iNext = NIL_PGMPOOL_IDX;
#ifdef PGMPOOL_WITH_MONITORING
pPage->iModifiedNext = NIL_PGMPOOL_IDX;
pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
pPage->cModifications = 0;
/* ASSUMES that we're not sharing with any of the other special pages (safe for now). */
pPage->iMonitoredNext = NIL_PGMPOOL_IDX;
pPage->iMonitoredPrev = NIL_PGMPOOL_IDX;
if (pPage->fMonitored)
{
int rc = PGMHandlerPhysicalChangeCallbacks(pVM, pPage->GCPhys & ~(RTGCPHYS)(PAGE_SIZE - 1),
pPool->pfnAccessHandlerR3, MMHyperCCToR3(pVM, pPage),
pPool->pfnAccessHandlerR0, MMHyperCCToR0(pVM, pPage),
pPool->pfnAccessHandlerRC, MMHyperCCToRC(pVM, pPage),
pPool->pszAccessHandler);
AssertFatalRCSuccess(rc);
# ifdef PGMPOOL_WITH_CACHE
pgmPoolHashInsert(pPool, pPage);
# endif
}
#endif
#ifdef PGMPOOL_WITH_USER_TRACKING
Assert(pPage->iUserHead == NIL_PGMPOOL_USER_INDEX); /* for now */
#endif
#ifdef PGMPOOL_WITH_CACHE
Assert(pPage->iAgeNext == NIL_PGMPOOL_IDX);
Assert(pPage->iAgePrev == NIL_PGMPOOL_IDX);
#endif
}
for (unsigned i=0;i<pVM->cCPUs;i++)
{
PVMCPU pVCpu = &pVM->aCpus[i];
/*
* Re-enter the shadowing mode and assert Sync CR3 FF.
*/
pgmR3ReEnterShadowModeAfterPoolFlush(pVM, pVCpu);
VMCPU_FF_SET(pVCpu, VMCPU_FF_PGM_SYNC_CR3);
}
STAM_PROFILE_STOP(&pPool->StatFlushAllInt, a);
}
#endif /* IN_RING3 */
/**
* Flushes a pool page.
*
* This moves the page to the free list after removing all user references to it.
* In GC this will cause a CR3 reload if the page is traced back to an active root page.
*
* @returns VBox status code.
* @retval VINF_SUCCESS on success.
* @param pPool The pool.
* @param HCPhys The HC physical address of the shadow page.
*/
int pgmPoolFlushPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage)
{
PVM pVM = pPool->CTX_SUFF(pVM);
int rc = VINF_SUCCESS;
STAM_PROFILE_START(&pPool->StatFlushPage, f);
LogFlow(("pgmPoolFlushPage: pPage=%p:{.Key=%RHp, .idx=%d, .enmKind=%s, .GCPhys=%RGp}\n",
pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), pPage->GCPhys));
/*
* Quietly reject any attempts at flushing any of the special root pages.
*/
if (pPage->idx < PGMPOOL_IDX_FIRST)
{
AssertFailed(); /* can no longer happen */
Log(("pgmPoolFlushPage: special root page, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
return VINF_SUCCESS;
}
pgmLock(pVM);
/*
* Quietly reject any attempts at flushing the currently active shadow CR3 mapping
*/
if (pgmPoolIsPageLocked(&pVM->pgm.s, pPage))
{
AssertMsg( pPage->enmKind == PGMPOOLKIND_64BIT_PML4
|| pPage->enmKind == PGMPOOLKIND_PAE_PDPT
|| pPage->enmKind == PGMPOOLKIND_PAE_PDPT_FOR_32BIT
|| pPage->enmKind == PGMPOOLKIND_32BIT_PD
|| pPage->enmKind == PGMPOOLKIND_PAE_PD_FOR_PAE_PD
|| pPage->enmKind == PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD
|| pPage->enmKind == PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD
|| pPage->enmKind == PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD
|| pPage->enmKind == PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD,
("Can't free the shadow CR3! (%RHp vs %RHp kind=%d\n", PGMGetHyperCR3(VMMGetCpu(pVM)), pPage->Core.Key, pPage->enmKind));
Log(("pgmPoolFlushPage: current active shadow CR3, rejected. enmKind=%s idx=%d\n", pgmPoolPoolKindToStr(pPage->enmKind), pPage->idx));
pgmUnlock(pVM);
return VINF_SUCCESS;
}
#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
/* Start a subset so we won't run out of mapping space. */
PVMCPU pVCpu = VMMGetCpu(pVM);
uint32_t iPrevSubset = PGMDynMapPushAutoSubset(pVCpu);
#endif
/*
* Mark the page as being in need of a ASMMemZeroPage().
*/
pPage->fZeroed = false;
#ifdef PGMPOOL_WITH_USER_TRACKING
/*
* Clear the page.
*/
pgmPoolTrackClearPageUsers(pPool, pPage);
STAM_PROFILE_START(&pPool->StatTrackDeref,a);
pgmPoolTrackDeref(pPool, pPage);
STAM_PROFILE_STOP(&pPool->StatTrackDeref,a);
#endif
#ifdef PGMPOOL_WITH_CACHE
/*
* Flush it from the cache.
*/
pgmPoolCacheFlushPage(pPool, pPage);
#endif /* PGMPOOL_WITH_CACHE */
#ifdef VBOX_WITH_2X_4GB_ADDR_SPACE_IN_R0
/* Heavy stuff done. */
PGMDynMapPopAutoSubset(pVCpu, iPrevSubset);
#endif
#ifdef PGMPOOL_WITH_MONITORING
/*
* Deregistering the monitoring.
*/
if (pPage->fMonitored)
rc = pgmPoolMonitorFlush(pPool, pPage);
#endif
/*
* Free the page.
*/
Assert(pPage->iNext == NIL_PGMPOOL_IDX);
pPage->iNext = pPool->iFreeHead;
pPool->iFreeHead = pPage->idx;
pPage->enmKind = PGMPOOLKIND_FREE;
pPage->GCPhys = NIL_RTGCPHYS;
pPage->fReusedFlushPending = false;
pPool->cUsedPages--;
pgmUnlock(pVM);
STAM_PROFILE_STOP(&pPool->StatFlushPage, f);
return rc;
}
/**
* Frees a usage of a pool page.
*
* The caller is responsible to updating the user table so that it no longer
* references the shadow page.
*
* @param pPool The pool.
* @param HCPhys The HC physical address of the shadow page.
* @param iUser The shadow page pool index of the user table.
* @param iUserTable The index into the user table (shadowed).
*/
void pgmPoolFreeByPage(PPGMPOOL pPool, PPGMPOOLPAGE pPage, uint16_t iUser, uint32_t iUserTable)
{
PVM pVM = pPool->CTX_SUFF(pVM);
STAM_PROFILE_START(&pPool->StatFree, a);
LogFlow(("pgmPoolFreeByPage: pPage=%p:{.Key=%RHp, .idx=%d, enmKind=%s} iUser=%#x iUserTable=%#x\n",
pPage, pPage->Core.Key, pPage->idx, pgmPoolPoolKindToStr(pPage->enmKind), iUser, iUserTable));
Assert(pPage->idx >= PGMPOOL_IDX_FIRST);
pgmLock(pVM);
#ifdef PGMPOOL_WITH_USER_TRACKING
pgmPoolTrackFreeUser(pPool, pPage, iUser, iUserTable);
#endif
#ifdef PGMPOOL_WITH_CACHE
if (!pPage->fCached)
#endif
pgmPoolFlushPage(pPool, pPage);
pgmUnlock(pVM);
STAM_PROFILE_STOP(&pPool->StatFree, a);
}
/**
* Makes one or more free page free.
*
* @returns VBox status code.
* @retval VINF_SUCCESS on success.
* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
*
* @param pPool The pool.
* @param enmKind Page table kind
* @param iUser The user of the page.
*/
static int pgmPoolMakeMoreFreePages(PPGMPOOL pPool, PGMPOOLKIND enmKind, uint16_t iUser)
{
LogFlow(("pgmPoolMakeMoreFreePages: iUser=%#x\n", iUser));
/*
* If the pool isn't full grown yet, expand it.
*/
if ( pPool->cCurPages < pPool->cMaxPages
#if defined(IN_RC)
/* Hack alert: we can't deal with jumps to ring 3 when called from MapCR3 and allocating pages for PAE PDs. */
&& enmKind != PGMPOOLKIND_PAE_PD_FOR_PAE_PD
&& (enmKind < PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD || enmKind > PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD)
#endif
)
{
STAM_PROFILE_ADV_SUSPEND(&pPool->StatAlloc, a);
#ifdef IN_RING3
int rc = PGMR3PoolGrow(pPool->pVMR3);
#else
int rc = CTXALLMID(VMM, CallHost)(pPool->CTX_SUFF(pVM), VMMCALLHOST_PGM_POOL_GROW, 0);
#endif
if (RT_FAILURE(rc))
return rc;
STAM_PROFILE_ADV_RESUME(&pPool->StatAlloc, a);
if (pPool->iFreeHead != NIL_PGMPOOL_IDX)
return VINF_SUCCESS;
}
#ifdef PGMPOOL_WITH_CACHE
/*
* Free one cached page.
*/
return pgmPoolCacheFreeOne(pPool, iUser);
#else
/*
* Flush the pool.
*
* If we have tracking enabled, it should be possible to come up with
* a cheap replacement strategy...
*/
/* @todo This path no longer works (CR3 root pages will be flushed)!! */
AssertCompileFailed();
Assert(!CPUMIsGuestInLongMode(pVM));
pgmPoolFlushAllInt(pPool);
return VERR_PGM_POOL_FLUSHED;
#endif
}
/**
* Allocates a page from the pool.
*
* This page may actually be a cached page and not in need of any processing
* on the callers part.
*
* @returns VBox status code.
* @retval VINF_SUCCESS if a NEW page was allocated.
* @retval VINF_PGM_CACHED_PAGE if a CACHED page was returned.
* @retval VERR_PGM_POOL_FLUSHED if the pool was flushed.
* @param pVM The VM handle.
* @param GCPhys The GC physical address of the page we're gonna shadow.
* For 4MB and 2MB PD entries, it's the first address the
* shadow PT is covering.
* @param enmKind The kind of mapping.
* @param iUser The shadow page pool index of the user table.
* @param iUserTable The index into the user table (shadowed).
* @param ppPage Where to store the pointer to the page. NULL is stored here on failure.
*/
int pgmPoolAlloc(PVM pVM, RTGCPHYS GCPhys, PGMPOOLKIND enmKind, uint16_t iUser, uint32_t iUserTable, PPPGMPOOLPAGE ppPage)
{
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
STAM_PROFILE_ADV_START(&pPool->StatAlloc, a);
LogFlow(("pgmPoolAlloc: GCPhys=%RGp enmKind=%s iUser=%#x iUserTable=%#x\n", GCPhys, pgmPoolPoolKindToStr(enmKind), iUser, iUserTable));
*ppPage = NULL;
/** @todo CSAM/PGMPrefetchPage messes up here during CSAMR3CheckGates
* (TRPMR3SyncIDT) because of FF priority. Try fix that?
* Assert(!(pVM->pgm.s.fGlobalSyncFlags & PGM_GLOBAL_SYNC_CLEAR_PGM_POOL)); */
pgmLock(pVM);
#ifdef PGMPOOL_WITH_CACHE
if (pPool->fCacheEnabled)
{
int rc2 = pgmPoolCacheAlloc(pPool, GCPhys, enmKind, iUser, iUserTable, ppPage);
if (RT_SUCCESS(rc2))
{
pgmUnlock(pVM);
STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
LogFlow(("pgmPoolAlloc: cached returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d}\n", rc2, *ppPage, (*ppPage)->Core.Key, (*ppPage)->idx));
return rc2;
}
}
#endif
/*
* Allocate a new one.
*/
int rc = VINF_SUCCESS;
uint16_t iNew = pPool->iFreeHead;
if (iNew == NIL_PGMPOOL_IDX)
{
rc = pgmPoolMakeMoreFreePages(pPool, enmKind, iUser);
if (RT_FAILURE(rc))
{
pgmUnlock(pVM);
Log(("pgmPoolAlloc: returns %Rrc (Free)\n", rc));
STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
return rc;
}
iNew = pPool->iFreeHead;
AssertReleaseReturn(iNew != NIL_PGMPOOL_IDX, VERR_INTERNAL_ERROR);
}
/* unlink the free head */
PPGMPOOLPAGE pPage = &pPool->aPages[iNew];
pPool->iFreeHead = pPage->iNext;
pPage->iNext = NIL_PGMPOOL_IDX;
/*
* Initialize it.
*/
pPool->cUsedPages++; /* physical handler registration / pgmPoolTrackFlushGCPhysPTsSlow requirement. */
pPage->enmKind = enmKind;
pPage->GCPhys = GCPhys;
pPage->fSeenNonGlobal = false; /* Set this to 'true' to disable this feature. */
pPage->fMonitored = false;
pPage->fCached = false;
pPage->fReusedFlushPending = false;
#ifdef PGMPOOL_WITH_MONITORING
pPage->cModifications = 0;
pPage->iModifiedNext = NIL_PGMPOOL_IDX;
pPage->iModifiedPrev = NIL_PGMPOOL_IDX;
#else
pPage->fCR3Mix = false;
#endif
#ifdef PGMPOOL_WITH_USER_TRACKING
pPage->cPresent = 0;
pPage->iFirstPresent = ~0;
/*
* Insert into the tracking and cache. If this fails, free the page.
*/
int rc3 = pgmPoolTrackInsert(pPool, pPage, GCPhys, iUser, iUserTable);
if (RT_FAILURE(rc3))
{
pPool->cUsedPages--;
pPage->enmKind = PGMPOOLKIND_FREE;
pPage->GCPhys = NIL_RTGCPHYS;
pPage->iNext = pPool->iFreeHead;
pPool->iFreeHead = pPage->idx;
pgmUnlock(pVM);
STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
Log(("pgmPoolAlloc: returns %Rrc (Insert)\n", rc3));
return rc3;
}
#endif /* PGMPOOL_WITH_USER_TRACKING */
/*
* Commit the allocation, clear the page and return.
*/
#ifdef VBOX_WITH_STATISTICS
if (pPool->cUsedPages > pPool->cUsedPagesHigh)
pPool->cUsedPagesHigh = pPool->cUsedPages;
#endif
if (!pPage->fZeroed)
{
STAM_PROFILE_START(&pPool->StatZeroPage, z);
void *pv = PGMPOOL_PAGE_2_PTR(pVM, pPage);
ASMMemZeroPage(pv);
STAM_PROFILE_STOP(&pPool->StatZeroPage, z);
}
*ppPage = pPage;
pgmUnlock(pVM);
LogFlow(("pgmPoolAlloc: returns %Rrc *ppPage=%p:{.Key=%RHp, .idx=%d, .fCached=%RTbool, .fMonitored=%RTbool}\n",
rc, pPage, pPage->Core.Key, pPage->idx, pPage->fCached, pPage->fMonitored));
STAM_PROFILE_ADV_STOP(&pPool->StatAlloc, a);
return rc;
}
/**
* Frees a usage of a pool page.
*
* @param pVM The VM handle.
* @param HCPhys The HC physical address of the shadow page.
* @param iUser The shadow page pool index of the user table.
* @param iUserTable The index into the user table (shadowed).
*/
void pgmPoolFree(PVM pVM, RTHCPHYS HCPhys, uint16_t iUser, uint32_t iUserTable)
{
LogFlow(("pgmPoolFree: HCPhys=%RHp iUser=%#x iUserTable=%#x\n", HCPhys, iUser, iUserTable));
PPGMPOOL pPool = pVM->pgm.s.CTX_SUFF(pPool);
pgmPoolFreeByPage(pPool, pgmPoolGetPage(pPool, HCPhys), iUser, iUserTable);
}
/**
* Internal worker for finding a 'in-use' shadow page give by it's physical address.
*
* @returns Pointer to the shadow page structure.
* @param pPool The pool.
* @param HCPhys The HC physical address of the shadow page.
*/
PPGMPOOLPAGE pgmPoolGetPage(PPGMPOOL pPool, RTHCPHYS HCPhys)
{
PVM pVM = pPool->CTX_SUFF(pVM);
/*
* Look up the page.
*/
pgmLock(pVM);
PPGMPOOLPAGE pPage = (PPGMPOOLPAGE)RTAvloHCPhysGet(&pPool->HCPhysTree, HCPhys & X86_PTE_PAE_PG_MASK);
pgmUnlock(pVM);
AssertFatalMsg(pPage && pPage->enmKind != PGMPOOLKIND_FREE, ("HCPhys=%RHp pPage=%p idx=%d\n", HCPhys, pPage, (pPage) ? pPage->idx : 0));
return pPage;
}
#ifdef IN_RING3
/**
* Flushes the entire cache.
*
* It will assert a global CR3 flush (FF) and assumes the caller is aware of this
* and execute this CR3 flush.
*
* @param pPool The pool.
*/
void pgmPoolFlushAll(PVM pVM)
{
LogFlow(("pgmPoolFlushAll:\n"));
pgmPoolFlushAllInt(pVM->pgm.s.CTX_SUFF(pPool));
}
#endif /* IN_RING3 */
#ifdef LOG_ENABLED
static const char *pgmPoolPoolKindToStr(uint8_t enmKind)
{
switch(enmKind)
{
case PGMPOOLKIND_INVALID:
return "PGMPOOLKIND_INVALID";
case PGMPOOLKIND_FREE:
return "PGMPOOLKIND_FREE";
case PGMPOOLKIND_32BIT_PT_FOR_PHYS:
return "PGMPOOLKIND_32BIT_PT_FOR_PHYS";
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT:
return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_PT";
case PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB:
return "PGMPOOLKIND_32BIT_PT_FOR_32BIT_4MB";
case PGMPOOLKIND_PAE_PT_FOR_PHYS:
return "PGMPOOLKIND_PAE_PT_FOR_PHYS";
case PGMPOOLKIND_PAE_PT_FOR_32BIT_PT:
return "PGMPOOLKIND_PAE_PT_FOR_32BIT_PT";
case PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB:
return "PGMPOOLKIND_PAE_PT_FOR_32BIT_4MB";
case PGMPOOLKIND_PAE_PT_FOR_PAE_PT:
return "PGMPOOLKIND_PAE_PT_FOR_PAE_PT";
case PGMPOOLKIND_PAE_PT_FOR_PAE_2MB:
return "PGMPOOLKIND_PAE_PT_FOR_PAE_2MB";
case PGMPOOLKIND_32BIT_PD:
return "PGMPOOLKIND_32BIT_PD";
case PGMPOOLKIND_32BIT_PD_PHYS:
return "PGMPOOLKIND_32BIT_PD_PHYS";
case PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD:
return "PGMPOOLKIND_PAE_PD0_FOR_32BIT_PD";
case PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD:
return "PGMPOOLKIND_PAE_PD1_FOR_32BIT_PD";
case PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD:
return "PGMPOOLKIND_PAE_PD2_FOR_32BIT_PD";
case PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD:
return "PGMPOOLKIND_PAE_PD3_FOR_32BIT_PD";
case PGMPOOLKIND_PAE_PD_FOR_PAE_PD:
return "PGMPOOLKIND_PAE_PD_FOR_PAE_PD";
case PGMPOOLKIND_PAE_PD_PHYS:
return "PGMPOOLKIND_PAE_PD_PHYS";
case PGMPOOLKIND_PAE_PDPT_FOR_32BIT:
return "PGMPOOLKIND_PAE_PDPT_FOR_32BIT";
case PGMPOOLKIND_PAE_PDPT:
return "PGMPOOLKIND_PAE_PDPT";
case PGMPOOLKIND_PAE_PDPT_PHYS:
return "PGMPOOLKIND_PAE_PDPT_PHYS";
case PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT:
return "PGMPOOLKIND_64BIT_PDPT_FOR_64BIT_PDPT";
case PGMPOOLKIND_64BIT_PDPT_FOR_PHYS:
return "PGMPOOLKIND_64BIT_PDPT_FOR_PHYS";
case PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD:
return "PGMPOOLKIND_64BIT_PD_FOR_64BIT_PD";
case PGMPOOLKIND_64BIT_PD_FOR_PHYS:
return "PGMPOOLKIND_64BIT_PD_FOR_PHYS";
case PGMPOOLKIND_64BIT_PML4:
return "PGMPOOLKIND_64BIT_PML4";
case PGMPOOLKIND_EPT_PDPT_FOR_PHYS:
return "PGMPOOLKIND_EPT_PDPT_FOR_PHYS";
case PGMPOOLKIND_EPT_PD_FOR_PHYS:
return "PGMPOOLKIND_EPT_PD_FOR_PHYS";
case PGMPOOLKIND_EPT_PT_FOR_PHYS:
return "PGMPOOLKIND_EPT_PT_FOR_PHYS";
case PGMPOOLKIND_ROOT_NESTED:
return "PGMPOOLKIND_ROOT_NESTED";
}
return "Unknown kind!";
}
#endif /* LOG_ENABLED*/