From f215e02bf85f68d3a6106c2a1f4f7f063f819064 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Thu, 11 Apr 2024 10:17:27 +0200 Subject: Adding upstream version 7.0.14-dfsg. Signed-off-by: Daniel Baumann --- src/VBox/VMM/VMMAll/PGMAllHandler.cpp | 2001 +++++++++++++++++++++++++++++++++ 1 file changed, 2001 insertions(+) create mode 100644 src/VBox/VMM/VMMAll/PGMAllHandler.cpp (limited to 'src/VBox/VMM/VMMAll/PGMAllHandler.cpp') diff --git a/src/VBox/VMM/VMMAll/PGMAllHandler.cpp b/src/VBox/VMM/VMMAll/PGMAllHandler.cpp new file mode 100644 index 00000000..c7d8a433 --- /dev/null +++ b/src/VBox/VMM/VMMAll/PGMAllHandler.cpp @@ -0,0 +1,2001 @@ +/* $Id: PGMAllHandler.cpp $ */ +/** @file + * PGM - Page Manager / Monitor, Access Handlers. + */ + +/* + * Copyright (C) 2006-2023 Oracle and/or its affiliates. + * + * This file is part of VirtualBox base platform packages, as + * available from https://www.virtualbox.org. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation, in version 3 of the + * License. + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, see . + * + * SPDX-License-Identifier: GPL-3.0-only + */ + + +/********************************************************************************************************************************* +* Header Files * +*********************************************************************************************************************************/ +#define LOG_GROUP LOG_GROUP_PGM +#define VBOX_WITHOUT_PAGING_BIT_FIELDS /* 64-bit bitfields are just asking for trouble. See @bugref{9841} and others. */ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#ifdef IN_RING0 +# include +#endif +#include "PGMInternal.h" +#include +#include "PGMInline.h" + +#include +#include +#if defined(RT_ARCH_AMD64) || defined(RT_ARCH_X86) +# include +#endif +#include +#include +#include +#include + + +/********************************************************************************************************************************* +* Global Variables * +*********************************************************************************************************************************/ +/** Dummy physical access handler type record. */ +CTX_SUFF(PGMPHYSHANDLERTYPEINT) const g_pgmHandlerPhysicalDummyType = +{ + /* .hType = */ UINT64_C(0x93b7557e1937aaff), + /* .enmKind = */ PGMPHYSHANDLERKIND_INVALID, + /* .uState = */ PGM_PAGE_HNDL_PHYS_STATE_ALL, + /* .fKeepPgmLock = */ true, + /* .fRing0DevInsIdx = */ false, +#ifdef IN_RING0 + /* .fNotInHm = */ false, + /* .pfnHandler = */ pgmR0HandlerPhysicalHandlerToRing3, + /* .pfnPfHandler = */ pgmR0HandlerPhysicalPfHandlerToRing3, +#elif defined(IN_RING3) + /* .fRing0Enabled = */ false, + /* .fNotInHm = */ false, + /* .pfnHandler = */ pgmR3HandlerPhysicalHandlerInvalid, +#else +# error "unsupported context" +#endif + /* .pszDesc = */ "dummy" +}; + + +/********************************************************************************************************************************* +* Internal Functions * +*********************************************************************************************************************************/ +static int pgmHandlerPhysicalSetRamFlagsAndFlushShadowPTs(PVMCC pVM, PPGMPHYSHANDLER pCur, PPGMRAMRANGE pRam, + void *pvBitmap, uint32_t offBitmap); +static void pgmHandlerPhysicalDeregisterNotifyNEM(PVMCC pVM, PPGMPHYSHANDLER pCur); +static void pgmHandlerPhysicalResetRamFlags(PVMCC pVM, PPGMPHYSHANDLER pCur); + + +#ifndef IN_RING3 + +/** + * @callback_method_impl{FNPGMPHYSHANDLER, + * Dummy for forcing ring-3 handling of the access.} + */ +DECLCALLBACK(VBOXSTRICTRC) +pgmR0HandlerPhysicalHandlerToRing3(PVMCC pVM, PVMCPUCC pVCpu, RTGCPHYS GCPhys, void *pvPhys, void *pvBuf, size_t cbBuf, + PGMACCESSTYPE enmAccessType, PGMACCESSORIGIN enmOrigin, uint64_t uUser) +{ + RT_NOREF(pVM, pVCpu, GCPhys, pvPhys, pvBuf, cbBuf, enmAccessType, enmOrigin, uUser); + return VINF_EM_RAW_EMULATE_INSTR; +} + + +/** + * @callback_method_impl{FNPGMRZPHYSPFHANDLER, + * Dummy for forcing ring-3 handling of the access.} + */ +DECLCALLBACK(VBOXSTRICTRC) +pgmR0HandlerPhysicalPfHandlerToRing3(PVMCC pVM, PVMCPUCC pVCpu, RTGCUINT uErrorCode, PCPUMCTX pCtx, + RTGCPTR pvFault, RTGCPHYS GCPhysFault, uint64_t uUser) +{ + RT_NOREF(pVM, pVCpu, uErrorCode, pCtx, pvFault, GCPhysFault, uUser); + return VINF_EM_RAW_EMULATE_INSTR; +} + +#endif /* !IN_RING3 */ + + +/** + * Creates a physical access handler, allocation part. + * + * @returns VBox status code. + * @retval VERR_OUT_OF_RESOURCES if no more handlers available. + * + * @param pVM The cross context VM structure. + * @param hType The handler type registration handle. + * @param uUser User argument to the handlers (not pointer). + * @param pszDesc Description of this handler. If NULL, the type + * description will be used instead. + * @param ppPhysHandler Where to return the access handler structure on + * success. + */ +int pgmHandlerPhysicalExCreate(PVMCC pVM, PGMPHYSHANDLERTYPE hType, uint64_t uUser, + R3PTRTYPE(const char *) pszDesc, PPGMPHYSHANDLER *ppPhysHandler) +{ + /* + * Validate input. + */ + PCPGMPHYSHANDLERTYPEINT const pType = pgmHandlerPhysicalTypeHandleToPtr(pVM, hType); + AssertReturn(pType, VERR_INVALID_HANDLE); + AssertReturn(pType->enmKind > PGMPHYSHANDLERKIND_INVALID && pType->enmKind < PGMPHYSHANDLERKIND_END, VERR_INVALID_HANDLE); + AssertPtr(ppPhysHandler); + + Log(("pgmHandlerPhysicalExCreate: uUser=%#RX64 hType=%#x (%d, %s) pszDesc=%RHv:%s\n", + uUser, hType, pType->enmKind, pType->pszDesc, pszDesc, R3STRING(pszDesc))); + + /* + * Allocate and initialize the new entry. + */ + int rc = PGM_LOCK(pVM); + AssertRCReturn(rc, rc); + + PPGMPHYSHANDLER pNew = pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator.allocateNode(); + if (pNew) + { + pNew->Key = NIL_RTGCPHYS; + pNew->KeyLast = NIL_RTGCPHYS; + pNew->cPages = 0; + pNew->cAliasedPages = 0; + pNew->cTmpOffPages = 0; + pNew->uUser = uUser; + pNew->hType = hType; + pNew->pszDesc = pszDesc != NIL_RTR3PTR ? pszDesc +#ifdef IN_RING3 + : pType->pszDesc; +#else + : pVM->pgm.s.aPhysHandlerTypes[hType & PGMPHYSHANDLERTYPE_IDX_MASK].pszDesc; +#endif + + PGM_UNLOCK(pVM); + *ppPhysHandler = pNew; + return VINF_SUCCESS; + } + + PGM_UNLOCK(pVM); + return VERR_OUT_OF_RESOURCES; +} + + +/** + * Duplicates a physical access handler. + * + * @returns VBox status code. + * @retval VINF_SUCCESS when successfully installed. + * + * @param pVM The cross context VM structure. + * @param pPhysHandlerSrc The source handler to duplicate + * @param ppPhysHandler Where to return the access handler structure on + * success. + */ +int pgmHandlerPhysicalExDup(PVMCC pVM, PPGMPHYSHANDLER pPhysHandlerSrc, PPGMPHYSHANDLER *ppPhysHandler) +{ + return pgmHandlerPhysicalExCreate(pVM, pPhysHandlerSrc->hType, pPhysHandlerSrc->uUser, + pPhysHandlerSrc->pszDesc, ppPhysHandler); +} + + +/** + * Register a access handler for a physical range. + * + * @returns VBox status code. + * @retval VINF_SUCCESS when successfully installed. + * @retval VINF_PGM_GCPHYS_ALIASED could be returned. + * + * @param pVM The cross context VM structure. + * @param pPhysHandler The physical handler. + * @param GCPhys Start physical address. + * @param GCPhysLast Last physical address. (inclusive) + */ +int pgmHandlerPhysicalExRegister(PVMCC pVM, PPGMPHYSHANDLER pPhysHandler, RTGCPHYS GCPhys, RTGCPHYS GCPhysLast) +{ + /* + * Validate input. + */ + AssertReturn(pPhysHandler, VERR_INVALID_POINTER); + PGMPHYSHANDLERTYPE const hType = pPhysHandler->hType; + PCPGMPHYSHANDLERTYPEINT const pType = pgmHandlerPhysicalTypeHandleToPtr(pVM, hType); + AssertReturn(pType, VERR_INVALID_HANDLE); + AssertReturn(pType->enmKind > PGMPHYSHANDLERKIND_INVALID && pType->enmKind < PGMPHYSHANDLERKIND_END, VERR_INVALID_HANDLE); + + AssertPtr(pPhysHandler); + + Log(("pgmHandlerPhysicalExRegister: GCPhys=%RGp GCPhysLast=%RGp hType=%#x (%d, %s) pszDesc=%RHv:%s\n", GCPhys, GCPhysLast, + hType, pType->enmKind, pType->pszDesc, pPhysHandler->pszDesc, R3STRING(pPhysHandler->pszDesc))); + AssertReturn(pPhysHandler->Key == NIL_RTGCPHYS, VERR_WRONG_ORDER); + + AssertMsgReturn(GCPhys < GCPhysLast, ("GCPhys >= GCPhysLast (%#x >= %#x)\n", GCPhys, GCPhysLast), VERR_INVALID_PARAMETER); + Assert(GCPhysLast - GCPhys < _4G); /* ASSUMPTION in PGMAllPhys.cpp */ + + switch (pType->enmKind) + { + case PGMPHYSHANDLERKIND_WRITE: + if (!pType->fNotInHm) + break; + RT_FALL_THRU(); /* Simplification: fNotInHm can only be used with full pages */ + case PGMPHYSHANDLERKIND_MMIO: + case PGMPHYSHANDLERKIND_ALL: + /* Simplification for PGMPhysRead, PGMR0Trap0eHandlerNPMisconfig and others: Full pages. */ + AssertMsgReturn(!(GCPhys & GUEST_PAGE_OFFSET_MASK), ("%RGp\n", GCPhys), VERR_INVALID_PARAMETER); + AssertMsgReturn((GCPhysLast & GUEST_PAGE_OFFSET_MASK) == GUEST_PAGE_OFFSET_MASK, ("%RGp\n", GCPhysLast), VERR_INVALID_PARAMETER); + break; + default: + AssertMsgFailed(("Invalid input enmKind=%d!\n", pType->enmKind)); + return VERR_INVALID_PARAMETER; + } + + /* + * We require the range to be within registered ram. + * There is no apparent need to support ranges which cover more than one ram range. + */ + PPGMRAMRANGE pRam = pgmPhysGetRange(pVM, GCPhys); + if ( !pRam + || GCPhysLast > pRam->GCPhysLast) + { +#ifdef IN_RING3 + DBGFR3Info(pVM->pUVM, "phys", NULL, NULL); +#endif + AssertMsgFailed(("No RAM range for %RGp-%RGp\n", GCPhys, GCPhysLast)); + return VERR_PGM_HANDLER_PHYSICAL_NO_RAM_RANGE; + } + Assert(GCPhys >= pRam->GCPhys && GCPhys < pRam->GCPhysLast); + Assert(GCPhysLast <= pRam->GCPhysLast && GCPhysLast >= pRam->GCPhys); + + /* + * Try insert into list. + */ + pPhysHandler->Key = GCPhys; + pPhysHandler->KeyLast = GCPhysLast; + pPhysHandler->cPages = (GCPhysLast - (GCPhys & X86_PTE_PAE_PG_MASK) + GUEST_PAGE_SIZE) >> GUEST_PAGE_SHIFT; + + int rc = PGM_LOCK(pVM); + if (RT_SUCCESS(rc)) + { + rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->insert(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, pPhysHandler); + if (RT_SUCCESS(rc)) + { + rc = pgmHandlerPhysicalSetRamFlagsAndFlushShadowPTs(pVM, pPhysHandler, pRam, NULL /*pvBitmap*/, 0 /*offBitmap*/); + if (rc == VINF_PGM_SYNC_CR3) + rc = VINF_PGM_GCPHYS_ALIASED; + +#if defined(IN_RING3) || defined(IN_RING0) + NEMHCNotifyHandlerPhysicalRegister(pVM, pType->enmKind, GCPhys, GCPhysLast - GCPhys + 1); +#endif + PGM_UNLOCK(pVM); + + if (rc != VINF_SUCCESS) + Log(("PGMHandlerPhysicalRegisterEx: returns %Rrc (%RGp-%RGp)\n", rc, GCPhys, GCPhysLast)); + return rc; + } + PGM_UNLOCK(pVM); + } + + pPhysHandler->Key = NIL_RTGCPHYS; + pPhysHandler->KeyLast = NIL_RTGCPHYS; + + AssertMsgReturn(rc == VERR_ALREADY_EXISTS, ("%Rrc GCPhys=%RGp GCPhysLast=%RGp\n", rc, GCPhys, GCPhysLast), rc); + +#if defined(IN_RING3) && defined(VBOX_STRICT) + DBGFR3Info(pVM->pUVM, "handlers", "phys nostats", NULL); +#endif + AssertMsgFailed(("Conflict! GCPhys=%RGp GCPhysLast=%RGp pszDesc=%s/%s\n", + GCPhys, GCPhysLast, R3STRING(pPhysHandler->pszDesc), R3STRING(pType->pszDesc))); + return VERR_PGM_HANDLER_PHYSICAL_CONFLICT; +} + + +/** + * Register a access handler for a physical range. + * + * @returns VBox status code. + * @retval VINF_SUCCESS when successfully installed. + * @retval VINF_PGM_GCPHYS_ALIASED when the shadow PTs could be updated because + * the guest page aliased or/and mapped by multiple PTs. A CR3 sync has been + * flagged together with a pool clearing. + * @retval VERR_PGM_HANDLER_PHYSICAL_CONFLICT if the range conflicts with an existing + * one. A debug assertion is raised. + * + * @param pVM The cross context VM structure. + * @param GCPhys Start physical address. + * @param GCPhysLast Last physical address. (inclusive) + * @param hType The handler type registration handle. + * @param uUser User argument to the handler. + * @param pszDesc Description of this handler. If NULL, the type + * description will be used instead. + */ +VMMDECL(int) PGMHandlerPhysicalRegister(PVMCC pVM, RTGCPHYS GCPhys, RTGCPHYS GCPhysLast, PGMPHYSHANDLERTYPE hType, + uint64_t uUser, R3PTRTYPE(const char *) pszDesc) +{ +#ifdef LOG_ENABLED + PCPGMPHYSHANDLERTYPEINT pType = pgmHandlerPhysicalTypeHandleToPtr(pVM, hType); + Log(("PGMHandlerPhysicalRegister: GCPhys=%RGp GCPhysLast=%RGp uUser=%#RX64 hType=%#x (%d, %s) pszDesc=%RHv:%s\n", + GCPhys, GCPhysLast, uUser, hType, pType->enmKind, R3STRING(pType->pszDesc), pszDesc, R3STRING(pszDesc))); +#endif + + PPGMPHYSHANDLER pNew; + int rc = pgmHandlerPhysicalExCreate(pVM, hType, uUser, pszDesc, &pNew); + if (RT_SUCCESS(rc)) + { + rc = pgmHandlerPhysicalExRegister(pVM, pNew, GCPhys, GCPhysLast); + if (RT_SUCCESS(rc)) + return rc; + pgmHandlerPhysicalExDestroy(pVM, pNew); + } + return rc; +} + + +/** + * Sets ram range flags and attempts updating shadow PTs. + * + * @returns VBox status code. + * @retval VINF_SUCCESS when shadow PTs was successfully updated. + * @retval VINF_PGM_SYNC_CR3 when the shadow PTs could be updated because + * the guest page aliased or/and mapped by multiple PTs. FFs set. + * @param pVM The cross context VM structure. + * @param pCur The physical handler. + * @param pRam The RAM range. + * @param pvBitmap Dirty bitmap. Optional. + * @param offBitmap Dirty bitmap offset. + */ +static int pgmHandlerPhysicalSetRamFlagsAndFlushShadowPTs(PVMCC pVM, PPGMPHYSHANDLER pCur, PPGMRAMRANGE pRam, + void *pvBitmap, uint32_t offBitmap) +{ + /* + * Iterate the guest ram pages updating the flags and flushing PT entries + * mapping the page. + */ + bool fFlushTLBs = false; + int rc = VINF_SUCCESS; + PCPGMPHYSHANDLERTYPEINT pCurType = PGMPHYSHANDLER_GET_TYPE_NO_NULL(pVM, pCur); + const unsigned uState = pCurType->uState; + uint32_t cPages = pCur->cPages; + uint32_t i = (pCur->Key - pRam->GCPhys) >> GUEST_PAGE_SHIFT; + for (;;) + { + PPGMPAGE pPage = &pRam->aPages[i]; + AssertMsg(pCurType->enmKind != PGMPHYSHANDLERKIND_MMIO || PGM_PAGE_IS_MMIO(pPage), + ("%RGp %R[pgmpage]\n", pRam->GCPhys + (i << GUEST_PAGE_SHIFT), pPage)); + + /* Only do upgrades. */ + if (PGM_PAGE_GET_HNDL_PHYS_STATE(pPage) < uState) + { + PGM_PAGE_SET_HNDL_PHYS_STATE(pPage, uState, pCurType->fNotInHm); + + const RTGCPHYS GCPhysPage = pRam->GCPhys + (i << GUEST_PAGE_SHIFT); + int rc2 = pgmPoolTrackUpdateGCPhys(pVM, GCPhysPage, pPage, + false /* allow updates of PTEs (instead of flushing) */, &fFlushTLBs); + if (rc2 != VINF_SUCCESS && rc == VINF_SUCCESS) + rc = rc2; + +#ifdef VBOX_WITH_NATIVE_NEM + /* Tell NEM about the protection update. */ + if (VM_IS_NEM_ENABLED(pVM)) + { + uint8_t u2State = PGM_PAGE_GET_NEM_STATE(pPage); + PGMPAGETYPE enmType = (PGMPAGETYPE)PGM_PAGE_GET_TYPE(pPage); + NEMHCNotifyPhysPageProtChanged(pVM, GCPhysPage, PGM_PAGE_GET_HCPHYS(pPage), + PGM_RAMRANGE_CALC_PAGE_R3PTR(pRam, GCPhysPage), + pgmPhysPageCalcNemProtection(pPage, enmType), enmType, &u2State); + PGM_PAGE_SET_NEM_STATE(pPage, u2State); + } +#endif + if (pvBitmap) + ASMBitSet(pvBitmap, offBitmap); + } + + /* next */ + if (--cPages == 0) + break; + i++; + offBitmap++; + } + + if (fFlushTLBs) + { + PGM_INVL_ALL_VCPU_TLBS(pVM); + Log(("pgmHandlerPhysicalSetRamFlagsAndFlushShadowPTs: flushing guest TLBs; rc=%d\n", rc)); + } + else + Log(("pgmHandlerPhysicalSetRamFlagsAndFlushShadowPTs: doesn't flush guest TLBs. rc=%Rrc; sync flags=%x VMCPU_FF_PGM_SYNC_CR3=%d\n", rc, VMMGetCpu(pVM)->pgm.s.fSyncFlags, VMCPU_FF_IS_SET(VMMGetCpu(pVM), VMCPU_FF_PGM_SYNC_CR3))); + + return rc; +} + + +/** + * Deregister a physical page access handler. + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param pPhysHandler The handler to deregister (but not free). + */ +int pgmHandlerPhysicalExDeregister(PVMCC pVM, PPGMPHYSHANDLER pPhysHandler) +{ + LogFlow(("pgmHandlerPhysicalExDeregister: Removing Range %RGp-%RGp %s\n", + pPhysHandler->Key, pPhysHandler->KeyLast, R3STRING(pPhysHandler->pszDesc))); + + int rc = PGM_LOCK(pVM); + AssertRCReturn(rc, rc); + + RTGCPHYS const GCPhys = pPhysHandler->Key; + AssertReturnStmt(GCPhys != NIL_RTGCPHYS, PGM_UNLOCK(pVM), VERR_PGM_HANDLER_NOT_FOUND); + + /* + * Remove the handler from the tree. + */ + + PPGMPHYSHANDLER pRemoved; + rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->remove(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, GCPhys, &pRemoved); + if (RT_SUCCESS(rc)) + { + if (pRemoved == pPhysHandler) + { + /* + * Clear the page bits, notify the REM about this change and clear + * the cache. + */ + pgmHandlerPhysicalResetRamFlags(pVM, pPhysHandler); + if (VM_IS_NEM_ENABLED(pVM)) + pgmHandlerPhysicalDeregisterNotifyNEM(pVM, pPhysHandler); + pVM->pgm.s.idxLastPhysHandler = 0; + + pPhysHandler->Key = NIL_RTGCPHYS; + pPhysHandler->KeyLast = NIL_RTGCPHYS; + + PGM_UNLOCK(pVM); + + return VINF_SUCCESS; + } + + /* + * Both of the failure conditions here are considered internal processing + * errors because they can only be caused by race conditions or corruption. + * If we ever need to handle concurrent deregistration, we have to move + * the NIL_RTGCPHYS check inside the PGM lock. + */ + pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->insert(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, pRemoved); + } + + PGM_UNLOCK(pVM); + + if (RT_FAILURE(rc)) + AssertMsgFailed(("Didn't find range starting at %RGp in the tree! %Rrc=rc\n", GCPhys, rc)); + else + AssertMsgFailed(("Found different handle at %RGp in the tree: got %p insteaded of %p\n", + GCPhys, pRemoved, pPhysHandler)); + return VERR_PGM_HANDLER_IPE_1; +} + + +/** + * Destroys (frees) a physical handler. + * + * The caller must deregister it before destroying it! + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param pHandler The handler to free. NULL if ignored. + */ +int pgmHandlerPhysicalExDestroy(PVMCC pVM, PPGMPHYSHANDLER pHandler) +{ + if (pHandler) + { + AssertPtr(pHandler); + AssertReturn(pHandler->Key == NIL_RTGCPHYS, VERR_WRONG_ORDER); + + int rc = PGM_LOCK(pVM); + if (RT_SUCCESS(rc)) + { + rc = pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator.freeNode(pHandler); + PGM_UNLOCK(pVM); + } + return rc; + } + return VINF_SUCCESS; +} + + +/** + * Deregister a physical page access handler. + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param GCPhys Start physical address. + */ +VMMDECL(int) PGMHandlerPhysicalDeregister(PVMCC pVM, RTGCPHYS GCPhys) +{ + AssertReturn(pVM->VMCC_CTX(pgm).s.pPhysHandlerTree, VERR_PGM_HANDLER_IPE_1); + + /* + * Find the handler. + */ + int rc = PGM_LOCK(pVM); + AssertRCReturn(rc, rc); + + PPGMPHYSHANDLER pRemoved; + rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->remove(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, GCPhys, &pRemoved); + if (RT_SUCCESS(rc)) + { + Assert(pRemoved->Key == GCPhys); + LogFlow(("PGMHandlerPhysicalDeregister: Removing Range %RGp-%RGp %s\n", + pRemoved->Key, pRemoved->KeyLast, R3STRING(pRemoved->pszDesc))); + + /* + * Clear the page bits, notify the REM about this change and clear + * the cache. + */ + pgmHandlerPhysicalResetRamFlags(pVM, pRemoved); + if (VM_IS_NEM_ENABLED(pVM)) + pgmHandlerPhysicalDeregisterNotifyNEM(pVM, pRemoved); + pVM->pgm.s.idxLastPhysHandler = 0; + + pRemoved->Key = NIL_RTGCPHYS; + rc = pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator.freeNode(pRemoved); + + PGM_UNLOCK(pVM); + return rc; + } + + PGM_UNLOCK(pVM); + + if (rc == VERR_NOT_FOUND) + { + AssertMsgFailed(("Didn't find range starting at %RGp\n", GCPhys)); + rc = VERR_PGM_HANDLER_NOT_FOUND; + } + return rc; +} + + +/** + * Shared code with modify. + */ +static void pgmHandlerPhysicalDeregisterNotifyNEM(PVMCC pVM, PPGMPHYSHANDLER pCur) +{ +#ifdef VBOX_WITH_NATIVE_NEM + PCPGMPHYSHANDLERTYPEINT pCurType = PGMPHYSHANDLER_GET_TYPE_NO_NULL(pVM, pCur); + RTGCPHYS GCPhysStart = pCur->Key; + RTGCPHYS GCPhysLast = pCur->KeyLast; + + /* + * Page align the range. + * + * Since we've reset (recalculated) the physical handler state of all pages + * we can make use of the page states to figure out whether a page should be + * included in the REM notification or not. + */ + if ( (pCur->Key & GUEST_PAGE_OFFSET_MASK) + || ((pCur->KeyLast + 1) & GUEST_PAGE_OFFSET_MASK)) + { + Assert(pCurType->enmKind != PGMPHYSHANDLERKIND_MMIO); + + if (GCPhysStart & GUEST_PAGE_OFFSET_MASK) + { + PPGMPAGE pPage = pgmPhysGetPage(pVM, GCPhysStart); + if ( pPage + && PGM_PAGE_GET_HNDL_PHYS_STATE(pPage) != PGM_PAGE_HNDL_PHYS_STATE_NONE) + { + RTGCPHYS GCPhys = (GCPhysStart + (GUEST_PAGE_SIZE - 1)) & X86_PTE_PAE_PG_MASK; + if ( GCPhys > GCPhysLast + || GCPhys < GCPhysStart) + return; + GCPhysStart = GCPhys; + } + else + GCPhysStart &= X86_PTE_PAE_PG_MASK; + Assert(!pPage || PGM_PAGE_GET_TYPE(pPage) != PGMPAGETYPE_MMIO); /* these are page aligned atm! */ + } + + if (GCPhysLast & GUEST_PAGE_OFFSET_MASK) + { + PPGMPAGE pPage = pgmPhysGetPage(pVM, GCPhysLast); + if ( pPage + && PGM_PAGE_GET_HNDL_PHYS_STATE(pPage) != PGM_PAGE_HNDL_PHYS_STATE_NONE) + { + RTGCPHYS GCPhys = (GCPhysLast & X86_PTE_PAE_PG_MASK) - 1; + if ( GCPhys < GCPhysStart + || GCPhys > GCPhysLast) + return; + GCPhysLast = GCPhys; + } + else + GCPhysLast |= GUEST_PAGE_OFFSET_MASK; + Assert(!pPage || PGM_PAGE_GET_TYPE(pPage) != PGMPAGETYPE_MMIO); /* these are page aligned atm! */ + } + } + + /* + * Tell NEM. + */ + PPGMRAMRANGE const pRam = pgmPhysGetRange(pVM, GCPhysStart); + RTGCPHYS const cb = GCPhysLast - GCPhysStart + 1; + uint8_t u2State = UINT8_MAX; + NEMHCNotifyHandlerPhysicalDeregister(pVM, pCurType->enmKind, GCPhysStart, cb, + pRam ? PGM_RAMRANGE_CALC_PAGE_R3PTR(pRam, GCPhysStart) : NULL, &u2State); + if (u2State != UINT8_MAX && pRam) + pgmPhysSetNemStateForPages(&pRam->aPages[(GCPhysStart - pRam->GCPhys) >> GUEST_PAGE_SHIFT], + cb >> GUEST_PAGE_SHIFT, u2State); +#else + RT_NOREF(pVM, pCur); +#endif +} + + +/** + * pgmHandlerPhysicalResetRamFlags helper that checks for other handlers on + * edge pages. + */ +DECLINLINE(void) pgmHandlerPhysicalRecalcPageState(PVMCC pVM, RTGCPHYS GCPhys, bool fAbove, PPGMRAMRANGE *ppRamHint) +{ + /* + * Look for other handlers. + */ + unsigned uState = PGM_PAGE_HNDL_PHYS_STATE_NONE; + for (;;) + { + PPGMPHYSHANDLER pCur; + int rc; + if (fAbove) + rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->lookupMatchingOrAbove(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, + GCPhys, &pCur); + else + rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->lookupMatchingOrBelow(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, + GCPhys, &pCur); + if (rc == VERR_NOT_FOUND) + break; + AssertRCBreak(rc); + if (((fAbove ? pCur->Key : pCur->KeyLast) >> GUEST_PAGE_SHIFT) != (GCPhys >> GUEST_PAGE_SHIFT)) + break; + PCPGMPHYSHANDLERTYPEINT pCurType = PGMPHYSHANDLER_GET_TYPE_NO_NULL(pVM, pCur); + uState = RT_MAX(uState, pCurType->uState); + + /* next? */ + RTGCPHYS GCPhysNext = fAbove + ? pCur->KeyLast + 1 + : pCur->Key - 1; + if ((GCPhysNext >> GUEST_PAGE_SHIFT) != (GCPhys >> GUEST_PAGE_SHIFT)) + break; + GCPhys = GCPhysNext; + } + + /* + * Update if we found something that is a higher priority state than the current. + * Note! The PGMPHYSHANDLER_F_NOT_IN_HM can be ignored here as it requires whole pages. + */ + if (uState != PGM_PAGE_HNDL_PHYS_STATE_NONE) + { + PPGMPAGE pPage; + int rc = pgmPhysGetPageWithHintEx(pVM, GCPhys, &pPage, ppRamHint); + if ( RT_SUCCESS(rc) + && PGM_PAGE_GET_HNDL_PHYS_STATE(pPage) < uState) + { + /* This should normally not be necessary. */ + PGM_PAGE_SET_HNDL_PHYS_STATE_ONLY(pPage, uState); + bool fFlushTLBs; + rc = pgmPoolTrackUpdateGCPhys(pVM, GCPhys, pPage, false /*fFlushPTEs*/, &fFlushTLBs); + if (RT_SUCCESS(rc) && fFlushTLBs) + PGM_INVL_ALL_VCPU_TLBS(pVM); + else + AssertRC(rc); + +#ifdef VBOX_WITH_NATIVE_NEM + /* Tell NEM about the protection update. */ + if (VM_IS_NEM_ENABLED(pVM)) + { + uint8_t u2State = PGM_PAGE_GET_NEM_STATE(pPage); + PGMPAGETYPE enmType = (PGMPAGETYPE)PGM_PAGE_GET_TYPE(pPage); + NEMHCNotifyPhysPageProtChanged(pVM, GCPhys, PGM_PAGE_GET_HCPHYS(pPage), + PGM_RAMRANGE_CALC_PAGE_R3PTR(*ppRamHint, GCPhys), + pgmPhysPageCalcNemProtection(pPage, enmType), enmType, &u2State); + PGM_PAGE_SET_NEM_STATE(pPage, u2State); + } +#endif + } + else + AssertRC(rc); + } +} + + +/** + * Resets an aliased page. + * + * @param pVM The cross context VM structure. + * @param pPage The page. + * @param GCPhysPage The page address in case it comes in handy. + * @param pRam The RAM range the page is associated with (for NEM + * notifications). + * @param fDoAccounting Whether to perform accounting. (Only set during + * reset where pgmR3PhysRamReset doesn't have the + * handler structure handy.) + * @param fFlushIemTlbs Whether to perform IEM TLB flushing or not. This + * can be cleared only if the caller does the flushing + * after calling this function. + */ +void pgmHandlerPhysicalResetAliasedPage(PVMCC pVM, PPGMPAGE pPage, RTGCPHYS GCPhysPage, PPGMRAMRANGE pRam, + bool fDoAccounting, bool fFlushIemTlbs) +{ + Assert( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_MMIO2_ALIAS_MMIO + || PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_SPECIAL_ALIAS_MMIO); + Assert(PGM_PAGE_GET_HNDL_PHYS_STATE(pPage) == PGM_PAGE_HNDL_PHYS_STATE_DISABLED); +#ifdef VBOX_WITH_NATIVE_NEM + RTHCPHYS const HCPhysPrev = PGM_PAGE_GET_HCPHYS(pPage); +#endif + + /* + * Flush any shadow page table references *first*. + */ + bool fFlushTLBs = false; + int rc = pgmPoolTrackUpdateGCPhys(pVM, GCPhysPage, pPage, true /*fFlushPTEs*/, &fFlushTLBs); + AssertLogRelRCReturnVoid(rc); + HMFlushTlbOnAllVCpus(pVM); + + /* + * Make it an MMIO/Zero page. + */ + PGM_PAGE_SET_HCPHYS(pVM, pPage, pVM->pgm.s.HCPhysZeroPg); + PGM_PAGE_SET_TYPE(pVM, pPage, PGMPAGETYPE_MMIO); + PGM_PAGE_SET_STATE(pVM, pPage, PGM_PAGE_STATE_ZERO); + PGM_PAGE_SET_PAGEID(pVM, pPage, NIL_GMM_PAGEID); + PGM_PAGE_SET_HNDL_PHYS_STATE_ONLY(pPage, PGM_PAGE_HNDL_PHYS_STATE_ALL); + + /* + * Flush its TLB entry. + */ + pgmPhysInvalidatePageMapTLBEntry(pVM, GCPhysPage); + if (fFlushIemTlbs) + IEMTlbInvalidateAllPhysicalAllCpus(pVM, NIL_VMCPUID); + + /* + * Do accounting for pgmR3PhysRamReset. + */ + if (fDoAccounting) + { + PPGMPHYSHANDLER pHandler; + rc = pgmHandlerPhysicalLookup(pVM, GCPhysPage, &pHandler); + if (RT_SUCCESS(rc)) + { + Assert(pHandler->cAliasedPages > 0); + pHandler->cAliasedPages--; + } + else + AssertMsgFailed(("rc=%Rrc GCPhysPage=%RGp\n", rc, GCPhysPage)); + } + +#ifdef VBOX_WITH_NATIVE_NEM + /* + * Tell NEM about the protection change. + */ + if (VM_IS_NEM_ENABLED(pVM)) + { + uint8_t u2State = PGM_PAGE_GET_NEM_STATE(pPage); + NEMHCNotifyPhysPageChanged(pVM, GCPhysPage, HCPhysPrev, pVM->pgm.s.HCPhysZeroPg, + PGM_RAMRANGE_CALC_PAGE_R3PTR(pRam, GCPhysPage), + NEM_PAGE_PROT_NONE, PGMPAGETYPE_MMIO, &u2State); + PGM_PAGE_SET_NEM_STATE(pPage, u2State); + } +#else + RT_NOREF(pRam); +#endif +} + + +/** + * Resets ram range flags. + * + * @param pVM The cross context VM structure. + * @param pCur The physical handler. + * + * @remark We don't start messing with the shadow page tables, as we've + * already got code in Trap0e which deals with out of sync handler + * flags (originally conceived for global pages). + */ +static void pgmHandlerPhysicalResetRamFlags(PVMCC pVM, PPGMPHYSHANDLER pCur) +{ + /* + * Iterate the guest ram pages updating the state. + */ + RTUINT cPages = pCur->cPages; + RTGCPHYS GCPhys = pCur->Key; + PPGMRAMRANGE pRamHint = NULL; + for (;;) + { + PPGMPAGE pPage; + int rc = pgmPhysGetPageWithHintEx(pVM, GCPhys, &pPage, &pRamHint); + if (RT_SUCCESS(rc)) + { + /* Reset aliased MMIO pages to MMIO, since this aliasing is our business. + (We don't flip MMIO to RAM though, that's PGMPhys.cpp's job.) */ + bool fNemNotifiedAlready = false; + if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_MMIO2_ALIAS_MMIO + || PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_SPECIAL_ALIAS_MMIO) + { + Assert(pCur->cAliasedPages > 0); + pgmHandlerPhysicalResetAliasedPage(pVM, pPage, GCPhys, pRamHint, false /*fDoAccounting*/, true /*fFlushIemTlbs*/); + pCur->cAliasedPages--; + fNemNotifiedAlready = true; + } +#ifdef VBOX_STRICT + PCPGMPHYSHANDLERTYPEINT const pCurType = PGMPHYSHANDLER_GET_TYPE(pVM, pCur); + AssertMsg(pCurType && (pCurType->enmKind != PGMPHYSHANDLERKIND_MMIO || PGM_PAGE_IS_MMIO(pPage)), + ("%RGp %R[pgmpage]\n", GCPhys, pPage)); +#endif + PGM_PAGE_SET_HNDL_PHYS_STATE(pPage, PGM_PAGE_HNDL_PHYS_STATE_NONE, false); + +#ifdef VBOX_WITH_NATIVE_NEM + /* Tell NEM about the protection change. */ + if (VM_IS_NEM_ENABLED(pVM) && !fNemNotifiedAlready) + { + uint8_t u2State = PGM_PAGE_GET_NEM_STATE(pPage); + PGMPAGETYPE enmType = (PGMPAGETYPE)PGM_PAGE_GET_TYPE(pPage); + NEMHCNotifyPhysPageProtChanged(pVM, GCPhys, PGM_PAGE_GET_HCPHYS(pPage), + PGM_RAMRANGE_CALC_PAGE_R3PTR(pRamHint, GCPhys), + pgmPhysPageCalcNemProtection(pPage, enmType), enmType, &u2State); + PGM_PAGE_SET_NEM_STATE(pPage, u2State); + } +#endif + RT_NOREF(fNemNotifiedAlready); + } + else + AssertRC(rc); + + /* next */ + if (--cPages == 0) + break; + GCPhys += GUEST_PAGE_SIZE; + } + + pCur->cAliasedPages = 0; + pCur->cTmpOffPages = 0; + + /* + * Check for partial start and end pages. + */ + if (pCur->Key & GUEST_PAGE_OFFSET_MASK) + pgmHandlerPhysicalRecalcPageState(pVM, pCur->Key - 1, false /* fAbove */, &pRamHint); + if ((pCur->KeyLast & GUEST_PAGE_OFFSET_MASK) != GUEST_PAGE_OFFSET_MASK) + pgmHandlerPhysicalRecalcPageState(pVM, pCur->KeyLast + 1, true /* fAbove */, &pRamHint); +} + + +#if 0 /* unused */ +/** + * Modify a physical page access handler. + * + * Modification can only be done to the range it self, not the type or anything else. + * + * @returns VBox status code. + * For all return codes other than VERR_PGM_HANDLER_NOT_FOUND and VINF_SUCCESS the range is deregistered + * and a new registration must be performed! + * @param pVM The cross context VM structure. + * @param GCPhysCurrent Current location. + * @param GCPhys New location. + * @param GCPhysLast New last location. + */ +VMMDECL(int) PGMHandlerPhysicalModify(PVMCC pVM, RTGCPHYS GCPhysCurrent, RTGCPHYS GCPhys, RTGCPHYS GCPhysLast) +{ + /* + * Remove it. + */ + int rc; + PGM_LOCK_VOID(pVM); + PPGMPHYSHANDLER pCur = (PPGMPHYSHANDLER)RTAvlroGCPhysRemove(&pVM->pgm.s.CTX_SUFF(pTrees)->PhysHandlers, GCPhysCurrent); + if (pCur) + { + /* + * Clear the ram flags. (We're gonna move or free it!) + */ + pgmHandlerPhysicalResetRamFlags(pVM, pCur); + PCPGMPHYSHANDLERTYPEINT const pCurType = PGMPHYSHANDLER_GET_TYPE_NO_NULL(pVM, pCur); + @todo pCurType validation + bool const fRestoreAsRAM = pCurType->pfnHandlerR3 /** @todo this isn't entirely correct. */ + && pCurType->enmKind != PGMPHYSHANDLERKIND_MMIO; + + /* + * Validate the new range, modify and reinsert. + */ + if (GCPhysLast >= GCPhys) + { + /* + * We require the range to be within registered ram. + * There is no apparent need to support ranges which cover more than one ram range. + */ + PPGMRAMRANGE pRam = pgmPhysGetRange(pVM, GCPhys); + if ( pRam + && GCPhys <= pRam->GCPhysLast + && GCPhysLast >= pRam->GCPhys) + { + pCur->Core.Key = GCPhys; + pCur->Core.KeyLast = GCPhysLast; + pCur->cPages = (GCPhysLast - (GCPhys & X86_PTE_PAE_PG_MASK) + 1) >> GUEST_PAGE_SHIFT; + + if (RTAvlroGCPhysInsert(&pVM->pgm.s.CTX_SUFF(pTrees)->PhysHandlers, &pCur->Core)) + { + RTGCPHYS const cb = GCPhysLast - GCPhys + 1; + PGMPHYSHANDLERKIND const enmKind = pCurType->enmKind; + + /* + * Set ram flags, flush shadow PT entries and finally tell REM about this. + */ + rc = pgmHandlerPhysicalSetRamFlagsAndFlushShadowPTs(pVM, pCur, pRam, NULL, 0); + + /** @todo NEM: not sure we need this notification... */ + NEMHCNotifyHandlerPhysicalModify(pVM, enmKind, GCPhysCurrent, GCPhys, cb, fRestoreAsRAM); + + PGM_UNLOCK(pVM); + + PGM_INVL_ALL_VCPU_TLBS(pVM); + Log(("PGMHandlerPhysicalModify: GCPhysCurrent=%RGp -> GCPhys=%RGp GCPhysLast=%RGp\n", + GCPhysCurrent, GCPhys, GCPhysLast)); + return VINF_SUCCESS; + } + + AssertMsgFailed(("Conflict! GCPhys=%RGp GCPhysLast=%RGp\n", GCPhys, GCPhysLast)); + rc = VERR_PGM_HANDLER_PHYSICAL_CONFLICT; + } + else + { + AssertMsgFailed(("No RAM range for %RGp-%RGp\n", GCPhys, GCPhysLast)); + rc = VERR_PGM_HANDLER_PHYSICAL_NO_RAM_RANGE; + } + } + else + { + AssertMsgFailed(("Invalid range %RGp-%RGp\n", GCPhys, GCPhysLast)); + rc = VERR_INVALID_PARAMETER; + } + + /* + * Invalid new location, flush the cache and free it. + * We've only gotta notify REM and free the memory. + */ + if (VM_IS_NEM_ENABLED(pVM)) + pgmHandlerPhysicalDeregisterNotifyNEM(pVM, pCur); + pVM->pgm.s.pLastPhysHandlerR0 = 0; + pVM->pgm.s.pLastPhysHandlerR3 = 0; + PGMHandlerPhysicalTypeRelease(pVM, pCur->hType); + MMHyperFree(pVM, pCur); + } + else + { + AssertMsgFailed(("Didn't find range starting at %RGp\n", GCPhysCurrent)); + rc = VERR_PGM_HANDLER_NOT_FOUND; + } + + PGM_UNLOCK(pVM); + return rc; +} +#endif /* unused */ + + +/** + * Changes the user callback arguments associated with a physical access handler. + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param GCPhys Start physical address of the handler. + * @param uUser User argument to the handlers. + */ +VMMDECL(int) PGMHandlerPhysicalChangeUserArg(PVMCC pVM, RTGCPHYS GCPhys, uint64_t uUser) +{ + /* + * Find the handler and make the change. + */ + int rc = PGM_LOCK(pVM); + AssertRCReturn(rc, rc); + + PPGMPHYSHANDLER pCur; + rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->lookup(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, GCPhys, &pCur); + if (RT_SUCCESS(rc)) + { + Assert(pCur->Key == GCPhys); + pCur->uUser = uUser; + } + else if (rc == VERR_NOT_FOUND) + { + AssertMsgFailed(("Didn't find range starting at %RGp\n", GCPhys)); + rc = VERR_PGM_HANDLER_NOT_FOUND; + } + + PGM_UNLOCK(pVM); + return rc; +} + +#if 0 /* unused */ + +/** + * Splits a physical access handler in two. + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param GCPhys Start physical address of the handler. + * @param GCPhysSplit The split address. + */ +VMMDECL(int) PGMHandlerPhysicalSplit(PVMCC pVM, RTGCPHYS GCPhys, RTGCPHYS GCPhysSplit) +{ + AssertReturn(GCPhys < GCPhysSplit, VERR_INVALID_PARAMETER); + + /* + * Do the allocation without owning the lock. + */ + PPGMPHYSHANDLER pNew; + int rc = MMHyperAlloc(pVM, sizeof(*pNew), 0, MM_TAG_PGM_HANDLERS, (void **)&pNew); + if (RT_FAILURE(rc)) + return rc; + + /* + * Get the handler. + */ + PGM_LOCK_VOID(pVM); + PPGMPHYSHANDLER pCur = (PPGMPHYSHANDLER)RTAvlroGCPhysGet(&pVM->pgm.s.CTX_SUFF(pTrees)->PhysHandlers, GCPhys); + if (RT_LIKELY(pCur)) + { + if (RT_LIKELY(GCPhysSplit <= pCur->Core.KeyLast)) + { + /* + * Create new handler node for the 2nd half. + */ + *pNew = *pCur; + pNew->Core.Key = GCPhysSplit; + pNew->cPages = (pNew->Core.KeyLast - (pNew->Core.Key & X86_PTE_PAE_PG_MASK) + GUEST_PAGE_SIZE) >> GUEST_PAGE_SHIFT; + + pCur->Core.KeyLast = GCPhysSplit - 1; + pCur->cPages = (pCur->Core.KeyLast - (pCur->Core.Key & X86_PTE_PAE_PG_MASK) + GUEST_PAGE_SIZE) >> GUEST_PAGE_SHIFT; + + if (RT_LIKELY(RTAvlroGCPhysInsert(&pVM->pgm.s.CTX_SUFF(pTrees)->PhysHandlers, &pNew->Core))) + { + LogFlow(("PGMHandlerPhysicalSplit: %RGp-%RGp and %RGp-%RGp\n", + pCur->Core.Key, pCur->Core.KeyLast, pNew->Core.Key, pNew->Core.KeyLast)); + PGM_UNLOCK(pVM); + return VINF_SUCCESS; + } + AssertMsgFailed(("whu?\n")); + rc = VERR_PGM_PHYS_HANDLER_IPE; + } + else + { + AssertMsgFailed(("outside range: %RGp-%RGp split %RGp\n", pCur->Core.Key, pCur->Core.KeyLast, GCPhysSplit)); + rc = VERR_INVALID_PARAMETER; + } + } + else + { + AssertMsgFailed(("Didn't find range starting at %RGp\n", GCPhys)); + rc = VERR_PGM_HANDLER_NOT_FOUND; + } + PGM_UNLOCK(pVM); + MMHyperFree(pVM, pNew); + return rc; +} + + +/** + * Joins up two adjacent physical access handlers which has the same callbacks. + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param GCPhys1 Start physical address of the first handler. + * @param GCPhys2 Start physical address of the second handler. + */ +VMMDECL(int) PGMHandlerPhysicalJoin(PVMCC pVM, RTGCPHYS GCPhys1, RTGCPHYS GCPhys2) +{ + /* + * Get the handlers. + */ + int rc; + PGM_LOCK_VOID(pVM); + PPGMPHYSHANDLER pCur1 = (PPGMPHYSHANDLER)RTAvlroGCPhysGet(&pVM->pgm.s.CTX_SUFF(pTrees)->PhysHandlers, GCPhys1); + if (RT_LIKELY(pCur1)) + { + PPGMPHYSHANDLER pCur2 = (PPGMPHYSHANDLER)RTAvlroGCPhysGet(&pVM->pgm.s.CTX_SUFF(pTrees)->PhysHandlers, GCPhys2); + if (RT_LIKELY(pCur2)) + { + /* + * Make sure that they are adjacent, and that they've got the same callbacks. + */ + if (RT_LIKELY(pCur1->Core.KeyLast + 1 == pCur2->Core.Key)) + { + if (RT_LIKELY(pCur1->hType == pCur2->hType)) + { + PPGMPHYSHANDLER pCur3 = (PPGMPHYSHANDLER)RTAvlroGCPhysRemove(&pVM->pgm.s.CTX_SUFF(pTrees)->PhysHandlers, GCPhys2); + if (RT_LIKELY(pCur3 == pCur2)) + { + pCur1->Core.KeyLast = pCur2->Core.KeyLast; + pCur1->cPages = (pCur1->Core.KeyLast - (pCur1->Core.Key & X86_PTE_PAE_PG_MASK) + GUEST_PAGE_SIZE) >> GUEST_PAGE_SHIFT; + LogFlow(("PGMHandlerPhysicalJoin: %RGp-%RGp %RGp-%RGp\n", + pCur1->Core.Key, pCur1->Core.KeyLast, pCur2->Core.Key, pCur2->Core.KeyLast)); + pVM->pgm.s.pLastPhysHandlerR0 = 0; + pVM->pgm.s.pLastPhysHandlerR3 = 0; + PGMHandlerPhysicalTypeRelease(pVM, pCur2->hType); + MMHyperFree(pVM, pCur2); + PGM_UNLOCK(pVM); + return VINF_SUCCESS; + } + + Assert(pCur3 == pCur2); + rc = VERR_PGM_PHYS_HANDLER_IPE; + } + else + { + AssertMsgFailed(("mismatching handlers\n")); + rc = VERR_ACCESS_DENIED; + } + } + else + { + AssertMsgFailed(("not adjacent: %RGp-%RGp %RGp-%RGp\n", + pCur1->Core.Key, pCur1->Core.KeyLast, pCur2->Core.Key, pCur2->Core.KeyLast)); + rc = VERR_INVALID_PARAMETER; + } + } + else + { + AssertMsgFailed(("Didn't find range starting at %RGp\n", GCPhys2)); + rc = VERR_PGM_HANDLER_NOT_FOUND; + } + } + else + { + AssertMsgFailed(("Didn't find range starting at %RGp\n", GCPhys1)); + rc = VERR_PGM_HANDLER_NOT_FOUND; + } + PGM_UNLOCK(pVM); + return rc; + +} + +#endif /* unused */ + +/** + * Resets any modifications to individual pages in a physical page access + * handler region. + * + * This is used in pair with PGMHandlerPhysicalPageTempOff(), + * PGMHandlerPhysicalPageAliasMmio2() or PGMHandlerPhysicalPageAliasHC(). + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param GCPhys The start address of the handler regions, i.e. what you + * passed to PGMR3HandlerPhysicalRegister(), + * PGMHandlerPhysicalRegisterEx() or + * PGMHandlerPhysicalModify(). + */ +VMMDECL(int) PGMHandlerPhysicalReset(PVMCC pVM, RTGCPHYS GCPhys) +{ + LogFlow(("PGMHandlerPhysicalReset GCPhys=%RGp\n", GCPhys)); + int rc = PGM_LOCK(pVM); + AssertRCReturn(rc, rc); + + /* + * Find the handler. + */ + PPGMPHYSHANDLER pCur; + rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->lookup(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, GCPhys, &pCur); + if (RT_SUCCESS(rc)) + { + Assert(pCur->Key == GCPhys); + + /* + * Validate kind. + */ + PCPGMPHYSHANDLERTYPEINT pCurType = PGMPHYSHANDLER_GET_TYPE_NO_NULL(pVM, pCur); + switch (pCurType->enmKind) + { + case PGMPHYSHANDLERKIND_WRITE: + case PGMPHYSHANDLERKIND_ALL: + case PGMPHYSHANDLERKIND_MMIO: /* NOTE: Only use when clearing MMIO ranges with aliased MMIO2 pages! */ + { + STAM_COUNTER_INC(&pVM->pgm.s.Stats.CTX_MID_Z(Stat,PhysHandlerReset)); /** @todo move out of switch */ + PPGMRAMRANGE pRam = pgmPhysGetRange(pVM, GCPhys); + Assert(pRam); + Assert(pRam->GCPhys <= pCur->Key); + Assert(pRam->GCPhysLast >= pCur->KeyLast); + + if (pCurType->enmKind == PGMPHYSHANDLERKIND_MMIO) + { + /* + * Reset all the PGMPAGETYPE_MMIO2_ALIAS_MMIO pages first and that's it. + * This could probably be optimized a bit wrt to flushing, but I'm too lazy + * to do that now... + */ + if (pCur->cAliasedPages) + { + PPGMPAGE pPage = &pRam->aPages[(pCur->Key - pRam->GCPhys) >> GUEST_PAGE_SHIFT]; + RTGCPHYS GCPhysPage = pCur->Key; + uint32_t cLeft = pCur->cPages; + bool fFlushIemTlb = false; + while (cLeft-- > 0) + { + if ( PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_MMIO2_ALIAS_MMIO + || PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_SPECIAL_ALIAS_MMIO) + { + fFlushIemTlb |= PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_MMIO2_ALIAS_MMIO; + Assert(pCur->cAliasedPages > 0); + pgmHandlerPhysicalResetAliasedPage(pVM, pPage, GCPhysPage, pRam, + false /*fDoAccounting*/, false /*fFlushIemTlbs*/); + --pCur->cAliasedPages; +#ifndef VBOX_STRICT + if (pCur->cAliasedPages == 0) + break; +#endif + } + Assert(PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_MMIO); + GCPhysPage += GUEST_PAGE_SIZE; + pPage++; + } + Assert(pCur->cAliasedPages == 0); + + /* + * Flush IEM TLBs in case they contain any references to aliased pages. + * This is only necessary for MMIO2 aliases. + */ + if (fFlushIemTlb) + IEMTlbInvalidateAllPhysicalAllCpus(pVM, NIL_VMCPUID); + } + } + else if (pCur->cTmpOffPages > 0) + { + /* + * Set the flags and flush shadow PT entries. + */ + rc = pgmHandlerPhysicalSetRamFlagsAndFlushShadowPTs(pVM, pCur, pRam, NULL /*pvBitmap*/, 0 /*offBitmap*/); + } + + pCur->cAliasedPages = 0; + pCur->cTmpOffPages = 0; + + rc = VINF_SUCCESS; + break; + } + + /* + * Invalid. + */ + default: + AssertMsgFailed(("Invalid type %d/%#x! Corruption!\n", pCurType->enmKind, pCur->hType)); + rc = VERR_PGM_PHYS_HANDLER_IPE; + break; + } + } + else if (rc == VERR_NOT_FOUND) + { + AssertMsgFailed(("Didn't find MMIO Range starting at %#x\n", GCPhys)); + rc = VERR_PGM_HANDLER_NOT_FOUND; + } + + PGM_UNLOCK(pVM); + return rc; +} + + +/** + * Special version of PGMHandlerPhysicalReset used by MMIO2 w/ dirty page + * tracking. + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param GCPhys The start address of the handler region. + * @param pvBitmap Dirty bitmap. Caller has cleared this already, only + * dirty bits will be set. Caller also made sure it's big + * enough. + * @param offBitmap Dirty bitmap offset. + * @remarks Caller must own the PGM critical section. + */ +DECLHIDDEN(int) pgmHandlerPhysicalResetMmio2WithBitmap(PVMCC pVM, RTGCPHYS GCPhys, void *pvBitmap, uint32_t offBitmap) +{ + LogFlow(("pgmHandlerPhysicalResetMmio2WithBitmap GCPhys=%RGp\n", GCPhys)); + PGM_LOCK_ASSERT_OWNER(pVM); + + /* + * Find the handler. + */ + PPGMPHYSHANDLER pCur; + int rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->lookup(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, GCPhys, &pCur); + if (RT_SUCCESS(rc)) + { + Assert(pCur->Key == GCPhys); + + /* + * Validate kind. + */ + PCPGMPHYSHANDLERTYPEINT pCurType = PGMPHYSHANDLER_GET_TYPE(pVM, pCur); + if ( pCurType + && pCurType->enmKind == PGMPHYSHANDLERKIND_WRITE) + { + STAM_COUNTER_INC(&pVM->pgm.s.Stats.CTX_MID_Z(Stat,PhysHandlerReset)); + + PPGMRAMRANGE pRam = pgmPhysGetRange(pVM, GCPhys); + Assert(pRam); + Assert(pRam->GCPhys <= pCur->Key); + Assert(pRam->GCPhysLast >= pCur->KeyLast); + + /* + * Set the flags and flush shadow PT entries. + */ + if (pCur->cTmpOffPages > 0) + { + rc = pgmHandlerPhysicalSetRamFlagsAndFlushShadowPTs(pVM, pCur, pRam, pvBitmap, offBitmap); + pCur->cTmpOffPages = 0; + } + else + rc = VINF_SUCCESS; + } + else + { + AssertFailed(); + rc = VERR_WRONG_TYPE; + } + } + else if (rc == VERR_NOT_FOUND) + { + AssertMsgFailed(("Didn't find MMIO Range starting at %#x\n", GCPhys)); + rc = VERR_PGM_HANDLER_NOT_FOUND; + } + + return rc; +} + + +/** + * Temporarily turns off the access monitoring of a page within a monitored + * physical write/all page access handler region. + * + * Use this when no further \#PFs are required for that page. Be aware that + * a page directory sync might reset the flags, and turn on access monitoring + * for the page. + * + * The caller must do required page table modifications. + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param GCPhys The start address of the access handler. This + * must be a fully page aligned range or we risk + * messing up other handlers installed for the + * start and end pages. + * @param GCPhysPage The physical address of the page to turn off + * access monitoring for. + */ +VMMDECL(int) PGMHandlerPhysicalPageTempOff(PVMCC pVM, RTGCPHYS GCPhys, RTGCPHYS GCPhysPage) +{ + LogFlow(("PGMHandlerPhysicalPageTempOff GCPhysPage=%RGp\n", GCPhysPage)); + int rc = PGM_LOCK(pVM); + AssertRCReturn(rc, rc); + + /* + * Validate the range. + */ + PPGMPHYSHANDLER pCur; + rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->lookup(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, GCPhys, &pCur); + if (RT_SUCCESS(rc)) + { + Assert(pCur->Key == GCPhys); + if (RT_LIKELY( GCPhysPage >= pCur->Key + && GCPhysPage <= pCur->KeyLast)) + { + Assert(!(pCur->Key & GUEST_PAGE_OFFSET_MASK)); + Assert((pCur->KeyLast & GUEST_PAGE_OFFSET_MASK) == GUEST_PAGE_OFFSET_MASK); + + PCPGMPHYSHANDLERTYPEINT const pCurType = PGMPHYSHANDLER_GET_TYPE(pVM, pCur); + AssertReturnStmt( pCurType + && ( pCurType->enmKind == PGMPHYSHANDLERKIND_WRITE + || pCurType->enmKind == PGMPHYSHANDLERKIND_ALL), + PGM_UNLOCK(pVM), VERR_ACCESS_DENIED); + + /* + * Change the page status. + */ + PPGMPAGE pPage; + PPGMRAMRANGE pRam; + rc = pgmPhysGetPageAndRangeEx(pVM, GCPhysPage, &pPage, &pRam); + AssertReturnStmt(RT_SUCCESS_NP(rc), PGM_UNLOCK(pVM), rc); + if (PGM_PAGE_GET_HNDL_PHYS_STATE(pPage) != PGM_PAGE_HNDL_PHYS_STATE_DISABLED) + { + PGM_PAGE_SET_HNDL_PHYS_STATE_ONLY(pPage, PGM_PAGE_HNDL_PHYS_STATE_DISABLED); + pCur->cTmpOffPages++; + +#ifdef VBOX_WITH_NATIVE_NEM + /* Tell NEM about the protection change (VGA is using this to track dirty pages). */ + if (VM_IS_NEM_ENABLED(pVM)) + { + uint8_t u2State = PGM_PAGE_GET_NEM_STATE(pPage); + PGMPAGETYPE enmType = (PGMPAGETYPE)PGM_PAGE_GET_TYPE(pPage); + NEMHCNotifyPhysPageProtChanged(pVM, GCPhysPage, PGM_PAGE_GET_HCPHYS(pPage), + PGM_RAMRANGE_CALC_PAGE_R3PTR(pRam, GCPhysPage), + pgmPhysPageCalcNemProtection(pPage, enmType), enmType, &u2State); + PGM_PAGE_SET_NEM_STATE(pPage, u2State); + } +#endif + } + PGM_UNLOCK(pVM); + return VINF_SUCCESS; + } + PGM_UNLOCK(pVM); + AssertMsgFailed(("The page %#x is outside the range %#x-%#x\n", GCPhysPage, pCur->Key, pCur->KeyLast)); + return VERR_INVALID_PARAMETER; + } + PGM_UNLOCK(pVM); + + if (rc == VERR_NOT_FOUND) + { + AssertMsgFailed(("Specified physical handler start address %#x is invalid.\n", GCPhys)); + return VERR_PGM_HANDLER_NOT_FOUND; + } + return rc; +} + + +/** + * Resolves an MMIO2 page. + * + * Caller as taken the PGM lock. + * + * @returns Pointer to the page if valid, NULL otherwise + * @param pVM The cross context VM structure. + * @param pDevIns The device owning it. + * @param hMmio2 The MMIO2 region. + * @param offMmio2Page The offset into the region. + */ +static PPGMPAGE pgmPhysResolveMmio2PageLocked(PVMCC pVM, PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2, RTGCPHYS offMmio2Page) +{ + /* Only works if the handle is in the handle table! */ + AssertReturn(hMmio2 != 0, NULL); + hMmio2--; + + /* Must check the first one for PGMREGMMIO2RANGE_F_FIRST_CHUNK. */ + AssertReturn(hMmio2 < RT_ELEMENTS(pVM->pgm.s.apMmio2RangesR3), NULL); + PPGMREGMMIO2RANGE pCur = pVM->pgm.s.CTX_SUFF(apMmio2Ranges)[hMmio2]; + AssertReturn(pCur, NULL); + AssertReturn(pCur->fFlags & PGMREGMMIO2RANGE_F_FIRST_CHUNK, NULL); + + /* Loop thru the sub-ranges till we find the one covering offMmio2. */ + for (;;) + { +#ifdef IN_RING3 + AssertReturn(pCur->pDevInsR3 == pDevIns, NULL); +#else + AssertReturn(pCur->pDevInsR3 == pDevIns->pDevInsForR3, NULL); +#endif + + /* Does it match the offset? */ + if (offMmio2Page < pCur->cbReal) + return &pCur->RamRange.aPages[offMmio2Page >> GUEST_PAGE_SHIFT]; + + /* Advance if we can. */ + AssertReturn(!(pCur->fFlags & PGMREGMMIO2RANGE_F_LAST_CHUNK), NULL); + offMmio2Page -= pCur->cbReal; + hMmio2++; + AssertReturn(hMmio2 < RT_ELEMENTS(pVM->pgm.s.apMmio2RangesR3), NULL); + pCur = pVM->pgm.s.CTX_SUFF(apMmio2Ranges)[hMmio2]; + AssertReturn(pCur, NULL); + } +} + + +/** + * Replaces an MMIO page with an MMIO2 page. + * + * This is a worker for IOMMMIOMapMMIO2Page that works in a similar way to + * PGMHandlerPhysicalPageTempOff but for an MMIO page. Since an MMIO page has no + * backing, the caller must provide a replacement page. For various reasons the + * replacement page must be an MMIO2 page. + * + * The caller must do required page table modifications. You can get away + * without making any modifications since it's an MMIO page, the cost is an extra + * \#PF which will the resync the page. + * + * Call PGMHandlerPhysicalReset() to restore the MMIO page. + * + * The caller may still get handler callback even after this call and must be + * able to deal correctly with such calls. The reason for these callbacks are + * either that we're executing in the recompiler (which doesn't know about this + * arrangement) or that we've been restored from saved state (where we won't + * save the change). + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param GCPhys The start address of the access handler. This + * must be a fully page aligned range or we risk + * messing up other handlers installed for the + * start and end pages. + * @param GCPhysPage The physical address of the page to turn off + * access monitoring for and replace with the MMIO2 + * page. + * @param pDevIns The device instance owning @a hMmio2. + * @param hMmio2 Handle to the MMIO2 region containing the page + * to remap in the the MMIO page at @a GCPhys. + * @param offMmio2PageRemap The offset into @a hMmio2 of the MMIO2 page that + * should serve as backing memory. + * + * @remark May cause a page pool flush if used on a page that is already + * aliased. + * + * @note This trick does only work reliably if the two pages are never ever + * mapped in the same page table. If they are the page pool code will + * be confused should either of them be flushed. See the special case + * of zero page aliasing mentioned in #3170. + * + */ +VMMDECL(int) PGMHandlerPhysicalPageAliasMmio2(PVMCC pVM, RTGCPHYS GCPhys, RTGCPHYS GCPhysPage, + PPDMDEVINS pDevIns, PGMMMIO2HANDLE hMmio2, RTGCPHYS offMmio2PageRemap) +{ +#ifdef VBOX_WITH_PGM_NEM_MODE + AssertReturn(!VM_IS_NEM_ENABLED(pVM) || !pVM->pgm.s.fNemMode, VERR_PGM_NOT_SUPPORTED_FOR_NEM_MODE); +#endif + int rc = PGM_LOCK(pVM); + AssertRCReturn(rc, rc); + + /* + * Resolve the MMIO2 reference. + */ + PPGMPAGE pPageRemap = pgmPhysResolveMmio2PageLocked(pVM, pDevIns, hMmio2, offMmio2PageRemap); + if (RT_LIKELY(pPageRemap)) + AssertMsgReturnStmt(PGM_PAGE_GET_TYPE(pPageRemap) == PGMPAGETYPE_MMIO2, + ("hMmio2=%RU64 offMmio2PageRemap=%RGp %R[pgmpage]\n", hMmio2, offMmio2PageRemap, pPageRemap), + PGM_UNLOCK(pVM), VERR_PGM_PHYS_NOT_MMIO2); + else + { + PGM_UNLOCK(pVM); + return VERR_OUT_OF_RANGE; + } + + /* + * Lookup and validate the range. + */ + PPGMPHYSHANDLER pCur; + rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->lookup(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, GCPhys, &pCur); + if (RT_SUCCESS(rc)) + { + Assert(pCur->Key == GCPhys); + if (RT_LIKELY( GCPhysPage >= pCur->Key + && GCPhysPage <= pCur->KeyLast)) + { + PCPGMPHYSHANDLERTYPEINT const pCurType = PGMPHYSHANDLER_GET_TYPE_NO_NULL(pVM, pCur); + AssertReturnStmt(pCurType->enmKind == PGMPHYSHANDLERKIND_MMIO, PGM_UNLOCK(pVM), VERR_ACCESS_DENIED); + AssertReturnStmt(!(pCur->Key & GUEST_PAGE_OFFSET_MASK), PGM_UNLOCK(pVM), VERR_INVALID_PARAMETER); + AssertReturnStmt((pCur->KeyLast & GUEST_PAGE_OFFSET_MASK) == GUEST_PAGE_OFFSET_MASK, + PGM_UNLOCK(pVM), VERR_INVALID_PARAMETER); + + /* + * Validate the page. + */ + PPGMPAGE pPage; + PPGMRAMRANGE pRam; + rc = pgmPhysGetPageAndRangeEx(pVM, GCPhysPage, &pPage, &pRam); + AssertReturnStmt(RT_SUCCESS_NP(rc), PGM_UNLOCK(pVM), rc); + if (PGM_PAGE_GET_TYPE(pPage) != PGMPAGETYPE_MMIO) + { + AssertMsgReturn(PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_MMIO2_ALIAS_MMIO, + ("GCPhysPage=%RGp %R[pgmpage]\n", GCPhysPage, pPage), + VERR_PGM_PHYS_NOT_MMIO2); + if (PGM_PAGE_GET_HCPHYS(pPage) == PGM_PAGE_GET_HCPHYS(pPageRemap)) + { + PGM_UNLOCK(pVM); + return VINF_PGM_HANDLER_ALREADY_ALIASED; + } + + /* + * The page is already mapped as some other page, reset it + * to an MMIO/ZERO page before doing the new mapping. + */ + Log(("PGMHandlerPhysicalPageAliasMmio2: GCPhysPage=%RGp (%R[pgmpage]; %RHp -> %RHp\n", + GCPhysPage, pPage, PGM_PAGE_GET_HCPHYS(pPage), PGM_PAGE_GET_HCPHYS(pPageRemap))); + pgmHandlerPhysicalResetAliasedPage(pVM, pPage, GCPhysPage, pRam, + false /*fDoAccounting*/, false /*fFlushIemTlbs*/); + pCur->cAliasedPages--; + + /* Since this may be present in the TLB and now be wrong, invalid + the guest physical address part of the IEM TLBs. Note, we do + this here as we will not invalid */ + IEMTlbInvalidateAllPhysicalAllCpus(pVM, NIL_VMCPUID); + } + Assert(PGM_PAGE_IS_ZERO(pPage)); + + /* + * Do the actual remapping here. + * This page now serves as an alias for the backing memory specified. + */ + LogFlow(("PGMHandlerPhysicalPageAliasMmio2: %RGp (%R[pgmpage]) alias for %RU64/%RGp (%R[pgmpage])\n", + GCPhysPage, pPage, hMmio2, offMmio2PageRemap, pPageRemap )); + PGM_PAGE_SET_HCPHYS(pVM, pPage, PGM_PAGE_GET_HCPHYS(pPageRemap)); + PGM_PAGE_SET_TYPE(pVM, pPage, PGMPAGETYPE_MMIO2_ALIAS_MMIO); + PGM_PAGE_SET_STATE(pVM, pPage, PGM_PAGE_STATE_ALLOCATED); + PGM_PAGE_SET_PAGEID(pVM, pPage, PGM_PAGE_GET_PAGEID(pPageRemap)); + PGM_PAGE_SET_HNDL_PHYS_STATE_ONLY(pPage, PGM_PAGE_HNDL_PHYS_STATE_DISABLED); + pCur->cAliasedPages++; + Assert(pCur->cAliasedPages <= pCur->cPages); + + /* + * Flush its TLB entry. + * + * Not calling IEMTlbInvalidateAllPhysicalAllCpus here to conserve + * all the other IEM TLB entires. When this one is kicked out and + * reloaded, it will be using the MMIO2 alias, but till then we'll + * continue doing MMIO. + */ + pgmPhysInvalidatePageMapTLBEntry(pVM, GCPhysPage); + /** @todo Do some preformance checks of calling + * IEMTlbInvalidateAllPhysicalAllCpus when in IEM mode, to see if it + * actually makes sense or not. Screen updates are typically massive + * and important when this kind of aliasing is used, so it may pay of... */ + +#ifdef VBOX_WITH_NATIVE_NEM + /* Tell NEM about the backing and protection change. */ + if (VM_IS_NEM_ENABLED(pVM)) + { + uint8_t u2State = PGM_PAGE_GET_NEM_STATE(pPage); + NEMHCNotifyPhysPageChanged(pVM, GCPhysPage, pVM->pgm.s.HCPhysZeroPg, PGM_PAGE_GET_HCPHYS(pPage), + PGM_RAMRANGE_CALC_PAGE_R3PTR(pRam, GCPhysPage), + pgmPhysPageCalcNemProtection(pPage, PGMPAGETYPE_MMIO2_ALIAS_MMIO), + PGMPAGETYPE_MMIO2_ALIAS_MMIO, &u2State); + PGM_PAGE_SET_NEM_STATE(pPage, u2State); + } +#endif + LogFlow(("PGMHandlerPhysicalPageAliasMmio2: => %R[pgmpage]\n", pPage)); + PGM_UNLOCK(pVM); + return VINF_SUCCESS; + } + + PGM_UNLOCK(pVM); + AssertMsgFailed(("The page %#x is outside the range %#x-%#x\n", GCPhysPage, pCur->Key, pCur->KeyLast)); + return VERR_INVALID_PARAMETER; + } + + PGM_UNLOCK(pVM); + if (rc == VERR_NOT_FOUND) + { + AssertMsgFailed(("Specified physical handler start address %#x is invalid.\n", GCPhys)); + return VERR_PGM_HANDLER_NOT_FOUND; + } + return rc; +} + + +/** + * Replaces an MMIO page with an arbitrary HC page in the shadow page tables. + * + * This differs from PGMHandlerPhysicalPageAliasMmio2 in that the page doesn't + * need to be a known MMIO2 page and that only shadow paging may access the + * page. The latter distinction is important because the only use for this + * feature is for mapping the special APIC access page that VT-x uses to detect + * APIC MMIO operations, the page is shared between all guest CPUs and actually + * not written to. At least at the moment. + * + * The caller must do required page table modifications. You can get away + * without making any modifications since it's an MMIO page, the cost is an extra + * \#PF which will the resync the page. + * + * Call PGMHandlerPhysicalReset() to restore the MMIO page. + * + * + * @returns VBox status code. + * @param pVM The cross context VM structure. + * @param GCPhys The start address of the access handler. This + * must be a fully page aligned range or we risk + * messing up other handlers installed for the + * start and end pages. + * @param GCPhysPage The physical address of the page to turn off + * access monitoring for. + * @param HCPhysPageRemap The physical address of the HC page that + * serves as backing memory. + * + * @remark May cause a page pool flush if used on a page that is already + * aliased. + */ +VMMDECL(int) PGMHandlerPhysicalPageAliasHC(PVMCC pVM, RTGCPHYS GCPhys, RTGCPHYS GCPhysPage, RTHCPHYS HCPhysPageRemap) +{ +/// Assert(!IOMIsLockOwner(pVM)); /* We mustn't own any other locks when calling this */ +#ifdef VBOX_WITH_PGM_NEM_MODE + AssertReturn(!VM_IS_NEM_ENABLED(pVM) || !pVM->pgm.s.fNemMode, VERR_PGM_NOT_SUPPORTED_FOR_NEM_MODE); +#endif + int rc = PGM_LOCK(pVM); + AssertRCReturn(rc, rc); + + /* + * Lookup and validate the range. + */ + PPGMPHYSHANDLER pCur; + rc = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree->lookup(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, GCPhys, &pCur); + if (RT_SUCCESS(rc)) + { + Assert(pCur->Key == GCPhys); + if (RT_LIKELY( GCPhysPage >= pCur->Key + && GCPhysPage <= pCur->KeyLast)) + { + PCPGMPHYSHANDLERTYPEINT const pCurType = PGMPHYSHANDLER_GET_TYPE_NO_NULL(pVM, pCur); + AssertReturnStmt(pCurType->enmKind == PGMPHYSHANDLERKIND_MMIO, PGM_UNLOCK(pVM), VERR_ACCESS_DENIED); + AssertReturnStmt(!(pCur->Key & GUEST_PAGE_OFFSET_MASK), PGM_UNLOCK(pVM), VERR_INVALID_PARAMETER); + AssertReturnStmt((pCur->KeyLast & GUEST_PAGE_OFFSET_MASK) == GUEST_PAGE_OFFSET_MASK, + PGM_UNLOCK(pVM), VERR_INVALID_PARAMETER); + + /* + * Get and validate the pages. + */ + PPGMPAGE pPage; + rc = pgmPhysGetPageEx(pVM, GCPhysPage, &pPage); + AssertReturnStmt(RT_SUCCESS_NP(rc), PGM_UNLOCK(pVM), rc); + if (PGM_PAGE_GET_TYPE(pPage) != PGMPAGETYPE_MMIO) + { + PGM_UNLOCK(pVM); + AssertMsgReturn(PGM_PAGE_GET_TYPE(pPage) == PGMPAGETYPE_SPECIAL_ALIAS_MMIO, + ("GCPhysPage=%RGp %R[pgmpage]\n", GCPhysPage, pPage), + VERR_PGM_PHYS_NOT_MMIO2); + return VINF_PGM_HANDLER_ALREADY_ALIASED; + } + Assert(PGM_PAGE_IS_ZERO(pPage)); + + /* + * Do the actual remapping here. + * This page now serves as an alias for the backing memory + * specified as far as shadow paging is concerned. + */ + LogFlow(("PGMHandlerPhysicalPageAliasHC: %RGp (%R[pgmpage]) alias for %RHp\n", + GCPhysPage, pPage, HCPhysPageRemap)); + PGM_PAGE_SET_HCPHYS(pVM, pPage, HCPhysPageRemap); + PGM_PAGE_SET_TYPE(pVM, pPage, PGMPAGETYPE_SPECIAL_ALIAS_MMIO); + PGM_PAGE_SET_STATE(pVM, pPage, PGM_PAGE_STATE_ALLOCATED); + PGM_PAGE_SET_PAGEID(pVM, pPage, NIL_GMM_PAGEID); + PGM_PAGE_SET_HNDL_PHYS_STATE_ONLY(pPage, PGM_PAGE_HNDL_PHYS_STATE_DISABLED); + pCur->cAliasedPages++; + Assert(pCur->cAliasedPages <= pCur->cPages); + + /* + * Flush its TLB entry. + * + * Not calling IEMTlbInvalidateAllPhysicalAllCpus here as special + * aliased MMIO pages are handled like MMIO by the IEM TLB. + */ + pgmPhysInvalidatePageMapTLBEntry(pVM, GCPhysPage); + +#ifdef VBOX_WITH_NATIVE_NEM + /* Tell NEM about the backing and protection change. */ + if (VM_IS_NEM_ENABLED(pVM)) + { + PPGMRAMRANGE pRam = pgmPhysGetRange(pVM, GCPhysPage); + uint8_t u2State = PGM_PAGE_GET_NEM_STATE(pPage); + NEMHCNotifyPhysPageChanged(pVM, GCPhysPage, pVM->pgm.s.HCPhysZeroPg, PGM_PAGE_GET_HCPHYS(pPage), + PGM_RAMRANGE_CALC_PAGE_R3PTR(pRam, GCPhysPage), + pgmPhysPageCalcNemProtection(pPage, PGMPAGETYPE_SPECIAL_ALIAS_MMIO), + PGMPAGETYPE_SPECIAL_ALIAS_MMIO, &u2State); + PGM_PAGE_SET_NEM_STATE(pPage, u2State); + } +#endif + LogFlow(("PGMHandlerPhysicalPageAliasHC: => %R[pgmpage]\n", pPage)); + PGM_UNLOCK(pVM); + return VINF_SUCCESS; + } + PGM_UNLOCK(pVM); + AssertMsgFailed(("The page %#x is outside the range %#x-%#x\n", GCPhysPage, pCur->Key, pCur->KeyLast)); + return VERR_INVALID_PARAMETER; + } + PGM_UNLOCK(pVM); + + if (rc == VERR_NOT_FOUND) + { + AssertMsgFailed(("Specified physical handler start address %#x is invalid.\n", GCPhys)); + return VERR_PGM_HANDLER_NOT_FOUND; + } + return rc; +} + + +/** + * Checks if a physical range is handled + * + * @returns boolean + * @param pVM The cross context VM structure. + * @param GCPhys Start physical address earlier passed to PGMR3HandlerPhysicalRegister(). + * @remarks Caller must take the PGM lock... + * @thread EMT. + */ +VMMDECL(bool) PGMHandlerPhysicalIsRegistered(PVMCC pVM, RTGCPHYS GCPhys) +{ + /* + * Find the handler. + */ + PGM_LOCK_VOID(pVM); + PPGMPHYSHANDLER pCur; + int rc = pgmHandlerPhysicalLookup(pVM, GCPhys, &pCur); + if (RT_SUCCESS(rc)) + { +#ifdef VBOX_STRICT + Assert(GCPhys >= pCur->Key && GCPhys <= pCur->KeyLast); + PCPGMPHYSHANDLERTYPEINT const pCurType = PGMPHYSHANDLER_GET_TYPE_NO_NULL(pVM, pCur); + Assert( pCurType->enmKind == PGMPHYSHANDLERKIND_WRITE + || pCurType->enmKind == PGMPHYSHANDLERKIND_ALL + || pCurType->enmKind == PGMPHYSHANDLERKIND_MMIO); +#endif + PGM_UNLOCK(pVM); + return true; + } + PGM_UNLOCK(pVM); + return false; +} + + +/** + * Checks if it's an disabled all access handler or write access handler at the + * given address. + * + * @returns true if it's an all access handler, false if it's a write access + * handler. + * @param pVM The cross context VM structure. + * @param GCPhys The address of the page with a disabled handler. + * + * @remarks The caller, PGMR3PhysTlbGCPhys2Ptr, must hold the PGM lock. + */ +bool pgmHandlerPhysicalIsAll(PVMCC pVM, RTGCPHYS GCPhys) +{ + PGM_LOCK_VOID(pVM); + PPGMPHYSHANDLER pCur; + int rc = pgmHandlerPhysicalLookup(pVM, GCPhys, &pCur); + AssertRCReturnStmt(rc, PGM_UNLOCK(pVM), true); + + /* Only whole pages can be disabled. */ + Assert( pCur->Key <= (GCPhys & ~(RTGCPHYS)GUEST_PAGE_OFFSET_MASK) + && pCur->KeyLast >= (GCPhys | GUEST_PAGE_OFFSET_MASK)); + + PCPGMPHYSHANDLERTYPEINT const pCurType = PGMPHYSHANDLER_GET_TYPE_NO_NULL(pVM, pCur); + Assert( pCurType->enmKind == PGMPHYSHANDLERKIND_WRITE + || pCurType->enmKind == PGMPHYSHANDLERKIND_ALL + || pCurType->enmKind == PGMPHYSHANDLERKIND_MMIO); /* sanity */ + bool const fRet = pCurType->enmKind != PGMPHYSHANDLERKIND_WRITE; + PGM_UNLOCK(pVM); + return fRet; +} + +#ifdef VBOX_STRICT + +/** + * State structure used by the PGMAssertHandlerAndFlagsInSync() function + * and its AVL enumerators. + */ +typedef struct PGMAHAFIS +{ + /** The current physical address. */ + RTGCPHYS GCPhys; + /** Number of errors. */ + unsigned cErrors; + /** Pointer to the VM. */ + PVM pVM; +} PGMAHAFIS, *PPGMAHAFIS; + + +/** + * Asserts that the handlers+guest-page-tables == ramrange-flags and + * that the physical addresses associated with virtual handlers are correct. + * + * @returns Number of mismatches. + * @param pVM The cross context VM structure. + */ +VMMDECL(unsigned) PGMAssertHandlerAndFlagsInSync(PVMCC pVM) +{ + PPGM pPGM = &pVM->pgm.s; + PGMAHAFIS State; + State.GCPhys = 0; + State.cErrors = 0; + State.pVM = pVM; + + PGM_LOCK_ASSERT_OWNER(pVM); + + /* + * Check the RAM flags against the handlers. + */ + PPGMPHYSHANDLERTREE const pPhysHandlerTree = pVM->VMCC_CTX(pgm).s.pPhysHandlerTree; + for (PPGMRAMRANGE pRam = pPGM->CTX_SUFF(pRamRangesX); pRam; pRam = pRam->CTX_SUFF(pNext)) + { + const uint32_t cPages = pRam->cb >> GUEST_PAGE_SHIFT; + for (uint32_t iPage = 0; iPage < cPages; iPage++) + { + PGMPAGE const *pPage = &pRam->aPages[iPage]; + if (PGM_PAGE_HAS_ANY_HANDLERS(pPage)) + { + State.GCPhys = pRam->GCPhys + (iPage << GUEST_PAGE_SHIFT); + + /* + * Physical first - calculate the state based on the handlers + * active on the page, then compare. + */ + if (PGM_PAGE_HAS_ANY_PHYSICAL_HANDLERS(pPage)) + { + /* the first */ + PPGMPHYSHANDLER pPhys; + int rc = pPhysHandlerTree->lookup(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, State.GCPhys, &pPhys); + if (rc == VERR_NOT_FOUND) + { + rc = pPhysHandlerTree->lookupMatchingOrAbove(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, + State.GCPhys, &pPhys); + if (RT_SUCCESS(rc)) + { + Assert(pPhys->Key >= State.GCPhys); + if (pPhys->Key > (State.GCPhys + GUEST_PAGE_SIZE - 1)) + pPhys = NULL; + } + else + AssertLogRelMsgReturn(rc == VERR_NOT_FOUND, ("rc=%Rrc GCPhys=%RGp\n", rc, State.GCPhys), 999); + } + else + AssertLogRelMsgReturn(RT_SUCCESS(rc), ("rc=%Rrc GCPhys=%RGp\n", rc, State.GCPhys), 999); + + if (pPhys) + { + PCPGMPHYSHANDLERTYPEINT pPhysType = pgmHandlerPhysicalTypeHandleToPtr(pVM, pPhys->hType); + unsigned uState = pPhysType->uState; + bool const fNotInHm = pPhysType->fNotInHm; /* whole pages, so no need to accumulate sub-page configs. */ + + /* more? */ + while (pPhys->KeyLast < (State.GCPhys | GUEST_PAGE_OFFSET_MASK)) + { + PPGMPHYSHANDLER pPhys2; + rc = pPhysHandlerTree->lookupMatchingOrAbove(&pVM->VMCC_CTX(pgm).s.PhysHandlerAllocator, + pPhys->KeyLast + 1, &pPhys2); + if (rc == VERR_NOT_FOUND) + break; + AssertLogRelMsgReturn(RT_SUCCESS(rc), ("rc=%Rrc KeyLast+1=%RGp\n", rc, pPhys->KeyLast + 1), 999); + if (pPhys2->Key > (State.GCPhys | GUEST_PAGE_OFFSET_MASK)) + break; + PCPGMPHYSHANDLERTYPEINT pPhysType2 = pgmHandlerPhysicalTypeHandleToPtr(pVM, pPhys2->hType); + uState = RT_MAX(uState, pPhysType2->uState); + pPhys = pPhys2; + } + + /* compare.*/ + if ( PGM_PAGE_GET_HNDL_PHYS_STATE(pPage) != uState + && PGM_PAGE_GET_HNDL_PHYS_STATE(pPage) != PGM_PAGE_HNDL_PHYS_STATE_DISABLED) + { + AssertMsgFailed(("ram range vs phys handler flags mismatch. GCPhys=%RGp state=%d expected=%d %s\n", + State.GCPhys, PGM_PAGE_GET_HNDL_PHYS_STATE(pPage), uState, pPhysType->pszDesc)); + State.cErrors++; + } + AssertMsgStmt(PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPage) == fNotInHm, + ("ram range vs phys handler flags mismatch. GCPhys=%RGp fNotInHm=%d, %d %s\n", + State.GCPhys, PGM_PAGE_IS_HNDL_PHYS_NOT_IN_HM(pPage), fNotInHm, pPhysType->pszDesc), + State.cErrors++); + } + else + { + AssertMsgFailed(("ram range vs phys handler mismatch. no handler for GCPhys=%RGp\n", State.GCPhys)); + State.cErrors++; + } + } + } + } /* foreach page in ram range. */ + } /* foreach ram range. */ + + /* + * Do the reverse check for physical handlers. + */ + /** @todo */ + + return State.cErrors; +} + +#endif /* VBOX_STRICT */ + -- cgit v1.2.3