/* $Id: PDMAllCritSect.cpp 93554 2022-02-02 22:57:02Z vboxsync $ */ /** @file * PDM - Write-Only Critical Section, All Contexts. */ /* * Copyright (C) 2006-2022 Oracle Corporation * * This file is part of VirtualBox Open Source Edition (OSE), as * available from http://www.virtualbox.org. This file is free software; * you can redistribute it and/or modify it under the terms of the GNU * General Public License (GPL) as published by the Free Software * Foundation, in version 2 as it comes in the "COPYING" file of the * VirtualBox OSE distribution. VirtualBox OSE is distributed in the * hope that it will be useful, but WITHOUT ANY WARRANTY of any kind. */ /********************************************************************************************************************************* * Header Files * *********************************************************************************************************************************/ #define LOG_GROUP LOG_GROUP_PDM_CRITSECT #include "PDMInternal.h" #include #include #include #include #include #include #include #include #include #include #ifdef IN_RING3 # include #endif #if defined(IN_RING3) || defined(IN_RING0) # include #endif #ifdef IN_RING0 # include #endif #if defined(IN_RING3) || defined(IN_RING0) # include #endif /********************************************************************************************************************************* * Defined Constants And Macros * *********************************************************************************************************************************/ /** The number loops to spin for in ring-3. */ #define PDMCRITSECT_SPIN_COUNT_R3 20 /** The number loops to spin for in ring-0. */ #define PDMCRITSECT_SPIN_COUNT_R0 256 /** The number loops to spin for in the raw-mode context. */ #define PDMCRITSECT_SPIN_COUNT_RC 256 /** Skips some of the overly paranoid atomic updates. * Makes some assumptions about cache coherence, though not brave enough not to * always end with an atomic update. */ #define PDMCRITSECT_WITH_LESS_ATOMIC_STUFF /* Undefine the automatic VBOX_STRICT API mappings. */ #undef PDMCritSectEnter #undef PDMCritSectTryEnter /** * Gets the ring-3 native thread handle of the calling thread. * * @returns native thread handle (ring-3). * @param pVM The cross context VM structure. * @param pCritSect The critical section. This is used in R0 and RC. */ DECL_FORCE_INLINE(RTNATIVETHREAD) pdmCritSectGetNativeSelf(PVMCC pVM, PCPDMCRITSECT pCritSect) { #ifdef IN_RING3 RT_NOREF(pVM, pCritSect); RTNATIVETHREAD hNativeSelf = RTThreadNativeSelf(); #elif defined(IN_RING0) AssertMsgReturn(pCritSect->s.Core.u32Magic == RTCRITSECT_MAGIC, ("%RX32\n", pCritSect->s.Core.u32Magic), NIL_RTNATIVETHREAD); RTNATIVETHREAD hNativeSelf = GVMMR0GetRing3ThreadForSelf(pVM); Assert(hNativeSelf != NIL_RTNATIVETHREAD); #else # error "Invalid context" #endif return hNativeSelf; } #ifdef IN_RING0 /** * Marks the critical section as corrupted. */ DECL_NO_INLINE(static, int) pdmCritSectCorrupted(PPDMCRITSECT pCritSect, const char *pszMsg) { ASMAtomicWriteU32(&pCritSect->s.Core.u32Magic, PDMCRITSECT_MAGIC_CORRUPTED); LogRel(("PDMCritSect: %s pCritSect=%p\n", pszMsg, pCritSect)); return VERR_PDM_CRITSECT_IPE; } #endif /** * Tail code called when we've won the battle for the lock. * * @returns VINF_SUCCESS. * * @param pCritSect The critical section. * @param hNativeSelf The native handle of this thread. * @param pSrcPos The source position of the lock operation. */ DECL_FORCE_INLINE(int) pdmCritSectEnterFirst(PPDMCRITSECT pCritSect, RTNATIVETHREAD hNativeSelf, PCRTLOCKVALSRCPOS pSrcPos) { Assert(hNativeSelf != NIL_RTNATIVETHREAD); AssertMsg(pCritSect->s.Core.NativeThreadOwner == NIL_RTNATIVETHREAD, ("NativeThreadOwner=%p\n", pCritSect->s.Core.NativeThreadOwner)); Assert(!(pCritSect->s.Core.fFlags & PDMCRITSECT_FLAGS_PENDING_UNLOCK)); # ifdef PDMCRITSECT_WITH_LESS_ATOMIC_STUFF pCritSect->s.Core.cNestings = 1; # else ASMAtomicWriteS32(&pCritSect->s.Core.cNestings, 1); # endif Assert(pCritSect->s.Core.cNestings == 1); ASMAtomicWriteHandle(&pCritSect->s.Core.NativeThreadOwner, hNativeSelf); # ifdef PDMCRITSECT_STRICT RTLockValidatorRecExclSetOwner(pCritSect->s.Core.pValidatorRec, NIL_RTTHREAD, pSrcPos, true); # else NOREF(pSrcPos); # endif if (pSrcPos) Log12Func(("%p: uId=%p ln=%u fn=%s\n", pCritSect, pSrcPos->uId, pSrcPos->uLine, pSrcPos->pszFunction)); else Log12Func(("%p\n", pCritSect)); STAM_PROFILE_ADV_START(&pCritSect->s.StatLocked, l); return VINF_SUCCESS; } #if defined(IN_RING3) || defined(IN_RING0) /** * Deals with the contended case in ring-3 and ring-0. * * @retval VINF_SUCCESS on success. * @retval VERR_SEM_DESTROYED if destroyed. * * @param pVM The cross context VM structure. * @param pVCpu The cross context virtual CPU structure if ring-0 and on * an EMT, otherwise NULL. * @param pCritSect The critsect. * @param hNativeSelf The native thread handle. * @param pSrcPos The source position of the lock operation. * @param rcBusy The status code to return when we're in RC or R0 */ static int pdmR3R0CritSectEnterContended(PVMCC pVM, PVMCPU pVCpu, PPDMCRITSECT pCritSect, RTNATIVETHREAD hNativeSelf, PCRTLOCKVALSRCPOS pSrcPos, int rcBusy) { # ifdef IN_RING0 /* * If we've got queued critical section leave operations and rcBusy isn't * VINF_SUCCESS, return to ring-3 immediately to avoid deadlocks. */ if ( !pVCpu || !VMCPU_FF_IS_SET(pVCpu, VMCPU_FF_PDM_CRITSECT) || rcBusy == VINF_SUCCESS ) { /* likely */ } else { /** @todo statistics. */ STAM_REL_COUNTER_INC(&pCritSect->s.StatContentionRZLock); return rcBusy; } # endif /* * Start waiting. */ if (ASMAtomicIncS32(&pCritSect->s.Core.cLockers) == 0) return pdmCritSectEnterFirst(pCritSect, hNativeSelf, pSrcPos); # ifdef IN_RING3 STAM_REL_COUNTER_INC(&pCritSect->s.StatContentionR3); # else STAM_REL_COUNTER_INC(&pCritSect->s.StatContentionRZLock); # endif /* * The wait loop. * * This handles VERR_TIMEOUT and VERR_INTERRUPTED. */ STAM_REL_PROFILE_START(&pCritSect->s.CTX_MID_Z(StatContention,Wait), a); PSUPDRVSESSION const pSession = pVM->pSession; SUPSEMEVENT const hEvent = (SUPSEMEVENT)pCritSect->s.Core.EventSem; # ifdef IN_RING3 # ifdef PDMCRITSECT_STRICT RTTHREAD const hThreadSelf = RTThreadSelfAutoAdopt(); int rc2 = RTLockValidatorRecExclCheckOrder(pCritSect->s.Core.pValidatorRec, hThreadSelf, pSrcPos, RT_INDEFINITE_WAIT); if (RT_FAILURE(rc2)) return rc2; # else RTTHREAD const hThreadSelf = RTThreadSelf(); # endif # else /* IN_RING0 */ uint64_t const tsStart = RTTimeNanoTS(); uint64_t const cNsMaxTotalDef = RT_NS_5MIN; uint64_t cNsMaxTotal = cNsMaxTotalDef; uint64_t const cNsMaxRetry = RT_NS_15SEC; uint32_t cMsMaxOne = RT_MS_5SEC; bool fNonInterruptible = false; # endif for (;;) { /* * Do the wait. * * In ring-3 this gets cluttered by lock validation and thread state * maintainence. * * In ring-0 we have to deal with the possibility that the thread has * been signalled and the interruptible wait function returning * immediately. In that case we do normal R0/RC rcBusy handling. * * We always do a timed wait here, so the event handle is revalidated * regularly and we won't end up stuck waiting for a destroyed critsect. */ /** @todo Make SUPSemEventClose wake up all waiters. */ # ifdef IN_RING3 # ifdef PDMCRITSECT_STRICT int rc9 = RTLockValidatorRecExclCheckBlocking(pCritSect->s.Core.pValidatorRec, hThreadSelf, pSrcPos, !(pCritSect->s.Core.fFlags & RTCRITSECT_FLAGS_NO_NESTING), RT_INDEFINITE_WAIT, RTTHREADSTATE_CRITSECT, true); if (RT_FAILURE(rc9)) return rc9; # else RTThreadBlocking(hThreadSelf, RTTHREADSTATE_CRITSECT, true); # endif int const rc = SUPSemEventWaitNoResume(pSession, hEvent, RT_MS_5SEC); RTThreadUnblocked(hThreadSelf, RTTHREADSTATE_CRITSECT); # else /* IN_RING0 */ int const rc = !fNonInterruptible ? SUPSemEventWaitNoResume(pSession, hEvent, cMsMaxOne) : SUPSemEventWait(pSession, hEvent, cMsMaxOne); Log11Func(("%p: rc=%Rrc %'RU64 ns (cMsMaxOne=%RU64 hOwner=%p)\n", pCritSect, rc, RTTimeNanoTS() - tsStart, cMsMaxOne, pCritSect->s.Core.NativeThreadOwner)); # endif /* IN_RING0 */ /* * Make sure the critical section hasn't been delete before continuing. */ if (RT_LIKELY(pCritSect->s.Core.u32Magic == RTCRITSECT_MAGIC)) { /* likely */ } else { LogRel(("PDMCritSectEnter: Destroyed while waiting; pCritSect=%p rc=%Rrc\n", pCritSect, rc)); return VERR_SEM_DESTROYED; } /* * Most likely we're here because we got signalled. */ if (rc == VINF_SUCCESS) { STAM_REL_PROFILE_STOP(&pCritSect->s.CTX_MID_Z(StatContention,Wait), a); return pdmCritSectEnterFirst(pCritSect, hNativeSelf, pSrcPos); } /* * Timeout and interrupted waits needs careful handling in ring-0 * because we're cooperating with ring-3 on this critical section * and thus need to make absolutely sure we won't get stuck here. * * The r0 interrupted case means something is pending (termination, * signal, APC, debugger, whatever), so we must try our best to * return to the caller and to ring-3 so it can be dealt with. */ if (RT_LIKELY(rc == VERR_TIMEOUT || rc == VERR_INTERRUPTED)) { # ifdef IN_RING0 uint64_t const cNsElapsed = RTTimeNanoTS() - tsStart; int const rcTerm = RTThreadQueryTerminationStatus(NIL_RTTHREAD); AssertMsg(rcTerm == VINF_SUCCESS || rcTerm == VERR_NOT_SUPPORTED || rcTerm == VINF_THREAD_IS_TERMINATING, ("rcTerm=%Rrc\n", rcTerm)); if (rcTerm == VERR_NOT_SUPPORTED && cNsMaxTotal == cNsMaxTotalDef) cNsMaxTotal = RT_NS_1MIN; if (rc == VERR_TIMEOUT) { /* Try return get out of here with a non-VINF_SUCCESS status if the thread is terminating or if the timeout has been exceeded. */ STAM_REL_COUNTER_INC(&pVM->pdm.s.StatCritSectVerrTimeout); if ( rcTerm != VINF_THREAD_IS_TERMINATING && cNsElapsed <= cNsMaxTotal) continue; } else { /* For interrupt cases, we must return if we can. If rcBusy is VINF_SUCCESS, we will try non-interruptible sleep for a while to help resolve the issue w/o guru'ing. */ STAM_REL_COUNTER_INC(&pVM->pdm.s.StatCritSectVerrInterrupted); if ( rcTerm != VINF_THREAD_IS_TERMINATING && rcBusy == VINF_SUCCESS && pVCpu != NULL && cNsElapsed <= cNsMaxTotal) { if (!fNonInterruptible) { STAM_REL_COUNTER_INC(&pVM->pdm.s.StatCritSectNonInterruptibleWaits); fNonInterruptible = true; cMsMaxOne = 32; uint64_t cNsLeft = cNsMaxTotal - cNsElapsed; if (cNsLeft > RT_NS_10SEC) cNsMaxTotal = cNsElapsed + RT_NS_10SEC; } continue; } } /* * Let try get out of here. We must very carefully undo the * cLockers increment we did using compare-and-exchange so that * we don't race the semaphore signalling in PDMCritSectLeave * and end up with spurious wakeups and two owners at once. */ uint32_t cNoIntWaits = 0; uint32_t cCmpXchgs = 0; int32_t cLockers = ASMAtomicReadS32(&pCritSect->s.Core.cLockers); for (;;) { if (pCritSect->s.Core.u32Magic == RTCRITSECT_MAGIC) { if (cLockers > 0 && cCmpXchgs < _64M) { bool fRc = ASMAtomicCmpXchgExS32(&pCritSect->s.Core.cLockers, cLockers - 1, cLockers, &cLockers); if (fRc) { LogFunc(("Aborting wait on %p (rc=%Rrc rcTerm=%Rrc cNsElapsed=%'RU64) -> %Rrc\n", pCritSect, rc, rcTerm, cNsElapsed, rcBusy != VINF_SUCCESS ? rcBusy : rc)); STAM_REL_COUNTER_INC(&pVM->pdm.s.StatAbortedCritSectEnters); return rcBusy != VINF_SUCCESS ? rcBusy : rc; } cCmpXchgs++; if ((cCmpXchgs & 0xffff) == 0) Log11Func(("%p: cLockers=%d cCmpXchgs=%u (hOwner=%p)\n", pCritSect, cLockers, cCmpXchgs, pCritSect->s.Core.NativeThreadOwner)); ASMNopPause(); continue; } if (cLockers == 0) { /* * We are racing someone in PDMCritSectLeave. * * For the VERR_TIMEOUT case we'll just retry taking it the normal * way for a while. For VERR_INTERRUPTED we're in for more fun as * the previous owner might not have signalled the semaphore yet, * so we'll do a short non-interruptible wait instead and then guru. */ if ( rc == VERR_TIMEOUT && RTTimeNanoTS() - tsStart <= cNsMaxTotal + cNsMaxRetry) break; if ( rc == VERR_INTERRUPTED && ( cNoIntWaits == 0 || RTTimeNanoTS() - (tsStart + cNsElapsed) < RT_NS_100MS)) { int const rc2 = SUPSemEventWait(pSession, hEvent, 1 /*ms*/); if (rc2 == VINF_SUCCESS) { STAM_REL_COUNTER_INC(&pVM->pdm.s.StatCritSectEntersWhileAborting); STAM_REL_PROFILE_STOP(&pCritSect->s.CTX_MID_Z(StatContention,Wait), a); return pdmCritSectEnterFirst(pCritSect, hNativeSelf, pSrcPos); } cNoIntWaits++; cLockers = ASMAtomicReadS32(&pCritSect->s.Core.cLockers); continue; } } else LogFunc(("Critical section %p has a broken cLockers count. Aborting.\n", pCritSect)); /* Sabotage the critical section and return error to caller. */ ASMAtomicWriteU32(&pCritSect->s.Core.u32Magic, PDMCRITSECT_MAGIC_FAILED_ABORT); LogRel(("PDMCritSectEnter: Failed to abort wait on pCritSect=%p (rc=%Rrc rcTerm=%Rrc)\n", pCritSect, rc, rcTerm)); return VERR_PDM_CRITSECT_ABORT_FAILED; } LogRel(("PDMCritSectEnter: Destroyed while aborting wait; pCritSect=%p/%#x rc=%Rrc rcTerm=%Rrc\n", pCritSect, pCritSect->s.Core.u32Magic, rc, rcTerm)); return VERR_SEM_DESTROYED; } /* We get here if we timed out. Just retry now that it appears someone left already. */ Assert(rc == VERR_TIMEOUT); cMsMaxOne = 10 /*ms*/; # else /* IN_RING3 */ RT_NOREF(pVM, pVCpu, rcBusy); # endif /* IN_RING3 */ } /* * Any other return code is fatal. */ else { AssertMsgFailed(("rc=%Rrc\n", rc)); return RT_FAILURE_NP(rc) ? rc : -rc; } } /* won't get here */ } #endif /* IN_RING3 || IN_RING0 */ /** * Common worker for the debug and normal APIs. * * @returns VINF_SUCCESS if entered successfully. * @returns rcBusy when encountering a busy critical section in RC/R0. * @retval VERR_SEM_DESTROYED if the critical section is delete before or * during the operation. * * @param pVM The cross context VM structure. * @param pCritSect The PDM critical section to enter. * @param rcBusy The status code to return when we're in RC or R0 * @param pSrcPos The source position of the lock operation. */ DECL_FORCE_INLINE(int) pdmCritSectEnter(PVMCC pVM, PPDMCRITSECT pCritSect, int rcBusy, PCRTLOCKVALSRCPOS pSrcPos) { Assert(pCritSect->s.Core.cNestings < 8); /* useful to catch incorrect locking */ Assert(pCritSect->s.Core.cNestings >= 0); #if defined(VBOX_STRICT) && defined(IN_RING0) /* Hope we're not messing with critical sections while in the no-block zone, that would complicate things a lot. */ PVMCPUCC pVCpuAssert = VMMGetCpu(pVM); Assert(pVCpuAssert && VMMRZCallRing3IsEnabled(pVCpuAssert)); #endif /* * If the critical section has already been destroyed, then inform the caller. */ AssertMsgReturn(pCritSect->s.Core.u32Magic == RTCRITSECT_MAGIC, ("%p %RX32\n", pCritSect, pCritSect->s.Core.u32Magic), VERR_SEM_DESTROYED); /* * See if we're lucky. */ /* NOP ... */ if (!(pCritSect->s.Core.fFlags & RTCRITSECT_FLAGS_NOP)) { /* We're more likely to end up here with real critsects than a NOP one. */ } else return VINF_SUCCESS; RTNATIVETHREAD hNativeSelf = pdmCritSectGetNativeSelf(pVM, pCritSect); AssertReturn(hNativeSelf != NIL_RTNATIVETHREAD, VERR_VM_THREAD_NOT_EMT); /* ... not owned ... */ if (ASMAtomicCmpXchgS32(&pCritSect->s.Core.cLockers, 0, -1)) return pdmCritSectEnterFirst(pCritSect, hNativeSelf, pSrcPos); /* ... or nested. */ if (pCritSect->s.Core.NativeThreadOwner == hNativeSelf) { Assert(pCritSect->s.Core.cNestings >= 1); # ifdef PDMCRITSECT_WITH_LESS_ATOMIC_STUFF pCritSect->s.Core.cNestings += 1; # else ASMAtomicIncS32(&pCritSect->s.Core.cNestings); # endif ASMAtomicIncS32(&pCritSect->s.Core.cLockers); Log12Func(("%p: cNestings=%d cLockers=%d\n", pCritSect, pCritSect->s.Core.cNestings, pCritSect->s.Core.cLockers)); return VINF_SUCCESS; } /* * Spin for a bit without incrementing the counter. */ /** @todo Move this to cfgm variables since it doesn't make sense to spin on UNI * cpu systems. */ int32_t cSpinsLeft = CTX_SUFF(PDMCRITSECT_SPIN_COUNT_); while (cSpinsLeft-- > 0) { if (ASMAtomicCmpXchgS32(&pCritSect->s.Core.cLockers, 0, -1)) return pdmCritSectEnterFirst(pCritSect, hNativeSelf, pSrcPos); ASMNopPause(); /** @todo Should use monitor/mwait on e.g. &cLockers here, possibly with a cli'ed pendingpreemption check up front using sti w/ instruction fusing for avoiding races. Hmm ... This is assuming the other party is actually executing code on another CPU ... which we could keep track of if we wanted. */ } #ifdef IN_RING3 /* * Take the slow path. */ NOREF(rcBusy); return pdmR3R0CritSectEnterContended(pVM, NULL, pCritSect, hNativeSelf, pSrcPos, rcBusy); #elif defined(IN_RING0) # if 1 /* new code */ /* * In ring-0 context we have to take the special VT-x/AMD-V HM context into * account when waiting on contended locks. * * While we usually (it can be VINF_SUCCESS) have the option of returning * rcBusy and force the caller to go back to ring-3 and to re-start the work * there, it's almost always more efficient to try wait for the lock here. * The rcBusy will be used if we encounter an VERR_INTERRUPTED situation * though. */ PVMCPUCC pVCpu = VMMGetCpu(pVM); if (pVCpu) { VMMR0EMTBLOCKCTX Ctx; int rc = VMMR0EmtPrepareToBlock(pVCpu, rcBusy, __FUNCTION__, pCritSect, &Ctx); if (rc == VINF_SUCCESS) { Assert(RTThreadPreemptIsEnabled(NIL_RTTHREAD)); rc = pdmR3R0CritSectEnterContended(pVM, pVCpu, pCritSect, hNativeSelf, pSrcPos, rcBusy); VMMR0EmtResumeAfterBlocking(pVCpu, &Ctx); } else STAM_REL_COUNTER_INC(&pCritSect->s.StatContentionRZLockBusy); return rc; } /* Non-EMT. */ Assert(RTThreadPreemptIsEnabled(NIL_RTTHREAD)); return pdmR3R0CritSectEnterContended(pVM, NULL, pCritSect, hNativeSelf, pSrcPos, rcBusy); # else /* old code: */ /* * We preemption hasn't been disabled, we can block here in ring-0. */ if ( RTThreadPreemptIsEnabled(NIL_RTTHREAD) && ASMIntAreEnabled()) return pdmR3R0CritSectEnterContended(pVM, VMMGetCpu(pVM), pCritSect, hNativeSelf, pSrcPos, rcBusy); STAM_REL_COUNTER_INC(&pCritSect->s.StatContentionRZLock); /* * Call ring-3 to acquire the critical section? */ if (rcBusy == VINF_SUCCESS) { PVMCPUCC pVCpu = VMMGetCpu(pVM); AssertReturn(pVCpu, VERR_PDM_CRITSECT_IPE); return VMMRZCallRing3(pVM, pVCpu, VMMCALLRING3_PDM_CRIT_SECT_ENTER, MMHyperCCToR3(pVM, pCritSect)); } /* * Return busy. */ LogFlow(("PDMCritSectEnter: locked => R3 (%Rrc)\n", rcBusy)); return rcBusy; # endif /* old code */ #else # error "Unsupported context" #endif } /** * Enters a PDM critical section. * * @returns VINF_SUCCESS if entered successfully. * @returns rcBusy when encountering a busy critical section in RC/R0. * @retval VERR_SEM_DESTROYED if the critical section is delete before or * during the operation. * * @param pVM The cross context VM structure. * @param pCritSect The PDM critical section to enter. * @param rcBusy The status code to return when we're in RC or R0 * and the section is busy. Pass VINF_SUCCESS to * acquired the critical section thru a ring-3 * call if necessary. * * @note Even callers setting @a rcBusy to VINF_SUCCESS must either handle * possible failures in ring-0 or apply * PDM_CRITSECT_RELEASE_ASSERT_RC(), * PDM_CRITSECT_RELEASE_ASSERT_RC_DEV(), * PDM_CRITSECT_RELEASE_ASSERT_RC_DRV() or * PDM_CRITSECT_RELEASE_ASSERT_RC_USB() to the return value of this * function. */ VMMDECL(DECL_CHECK_RETURN_NOT_R3(int)) PDMCritSectEnter(PVMCC pVM, PPDMCRITSECT pCritSect, int rcBusy) { #ifndef PDMCRITSECT_STRICT return pdmCritSectEnter(pVM, pCritSect, rcBusy, NULL); #else RTLOCKVALSRCPOS SrcPos = RTLOCKVALSRCPOS_INIT_NORMAL_API(); return pdmCritSectEnter(pVM, pCritSect, rcBusy, &SrcPos); #endif } /** * Enters a PDM critical section, with location information for debugging. * * @returns VINF_SUCCESS if entered successfully. * @returns rcBusy when encountering a busy critical section in RC/R0. * @retval VERR_SEM_DESTROYED if the critical section is delete before or * during the operation. * * @param pVM The cross context VM structure. * @param pCritSect The PDM critical section to enter. * @param rcBusy The status code to return when we're in RC or R0 * and the section is busy. Pass VINF_SUCCESS to * acquired the critical section thru a ring-3 * call if necessary. * @param uId Some kind of locking location ID. Typically a * return address up the stack. Optional (0). * @param SRC_POS The source position where to lock is being * acquired from. Optional. */ VMMDECL(DECL_CHECK_RETURN_NOT_R3(int)) PDMCritSectEnterDebug(PVMCC pVM, PPDMCRITSECT pCritSect, int rcBusy, RTHCUINTPTR uId, RT_SRC_POS_DECL) { #ifdef PDMCRITSECT_STRICT RTLOCKVALSRCPOS SrcPos = RTLOCKVALSRCPOS_INIT_DEBUG_API(); return pdmCritSectEnter(pVM, pCritSect, rcBusy, &SrcPos); #else NOREF(uId); RT_SRC_POS_NOREF(); return pdmCritSectEnter(pVM, pCritSect, rcBusy, NULL); #endif } /** * Common worker for the debug and normal APIs. * * @retval VINF_SUCCESS on success. * @retval VERR_SEM_BUSY if the critsect was owned. * @retval VERR_SEM_NESTED if nested enter on a no nesting section. (Asserted.) * @retval VERR_SEM_DESTROYED if the critical section is delete before or * during the operation. * * @param pVM The cross context VM structure. * @param pCritSect The critical section. * @param pSrcPos The source position of the lock operation. */ static int pdmCritSectTryEnter(PVMCC pVM, PPDMCRITSECT pCritSect, PCRTLOCKVALSRCPOS pSrcPos) { /* * If the critical section has already been destroyed, then inform the caller. */ AssertMsgReturn(pCritSect->s.Core.u32Magic == RTCRITSECT_MAGIC, ("%p %RX32\n", pCritSect, pCritSect->s.Core.u32Magic), VERR_SEM_DESTROYED); /* * See if we're lucky. */ /* NOP ... */ if (!(pCritSect->s.Core.fFlags & RTCRITSECT_FLAGS_NOP)) { /* We're more likely to end up here with real critsects than a NOP one. */ } else return VINF_SUCCESS; RTNATIVETHREAD hNativeSelf = pdmCritSectGetNativeSelf(pVM, pCritSect); AssertReturn(hNativeSelf != NIL_RTNATIVETHREAD, VERR_VM_THREAD_NOT_EMT); /* ... not owned ... */ if (ASMAtomicCmpXchgS32(&pCritSect->s.Core.cLockers, 0, -1)) return pdmCritSectEnterFirst(pCritSect, hNativeSelf, pSrcPos); /* ... or nested. */ if (pCritSect->s.Core.NativeThreadOwner == hNativeSelf) { Assert(pCritSect->s.Core.cNestings >= 1); # ifdef PDMCRITSECT_WITH_LESS_ATOMIC_STUFF pCritSect->s.Core.cNestings += 1; # else ASMAtomicIncS32(&pCritSect->s.Core.cNestings); # endif ASMAtomicIncS32(&pCritSect->s.Core.cLockers); Log12Func(("%p: cNestings=%d cLockers=%d\n", pCritSect, pCritSect->s.Core.cNestings, pCritSect->s.Core.cLockers)); return VINF_SUCCESS; } /* no spinning */ /* * Return busy. */ #ifdef IN_RING3 STAM_REL_COUNTER_INC(&pCritSect->s.StatContentionR3); #else STAM_REL_COUNTER_INC(&pCritSect->s.StatContentionRZLockBusy); #endif LogFlow(("PDMCritSectTryEnter: locked\n")); return VERR_SEM_BUSY; } /** * Try enter a critical section. * * @retval VINF_SUCCESS on success. * @retval VERR_SEM_BUSY if the critsect was owned. * @retval VERR_SEM_NESTED if nested enter on a no nesting section. (Asserted.) * @retval VERR_SEM_DESTROYED if the critical section is delete before or * during the operation. * * @param pVM The cross context VM structure. * @param pCritSect The critical section. */ VMMDECL(DECL_CHECK_RETURN(int)) PDMCritSectTryEnter(PVMCC pVM, PPDMCRITSECT pCritSect) { #ifndef PDMCRITSECT_STRICT return pdmCritSectTryEnter(pVM, pCritSect, NULL); #else RTLOCKVALSRCPOS SrcPos = RTLOCKVALSRCPOS_INIT_NORMAL_API(); return pdmCritSectTryEnter(pVM, pCritSect, &SrcPos); #endif } /** * Try enter a critical section, with location information for debugging. * * @retval VINF_SUCCESS on success. * @retval VERR_SEM_BUSY if the critsect was owned. * @retval VERR_SEM_NESTED if nested enter on a no nesting section. (Asserted.) * @retval VERR_SEM_DESTROYED if the critical section is delete before or * during the operation. * * @param pVM The cross context VM structure. * @param pCritSect The critical section. * @param uId Some kind of locking location ID. Typically a * return address up the stack. Optional (0). * @param SRC_POS The source position where to lock is being * acquired from. Optional. */ VMMDECL(DECL_CHECK_RETURN(int)) PDMCritSectTryEnterDebug(PVMCC pVM, PPDMCRITSECT pCritSect, RTHCUINTPTR uId, RT_SRC_POS_DECL) { #ifdef PDMCRITSECT_STRICT RTLOCKVALSRCPOS SrcPos = RTLOCKVALSRCPOS_INIT_DEBUG_API(); return pdmCritSectTryEnter(pVM, pCritSect, &SrcPos); #else NOREF(uId); RT_SRC_POS_NOREF(); return pdmCritSectTryEnter(pVM, pCritSect, NULL); #endif } #ifdef IN_RING3 /** * Enters a PDM critical section. * * @returns VINF_SUCCESS if entered successfully. * @returns rcBusy when encountering a busy critical section in GC/R0. * @retval VERR_SEM_DESTROYED if the critical section is delete before or * during the operation. * * @param pVM The cross context VM structure. * @param pCritSect The PDM critical section to enter. * @param fCallRing3 Whether this is a VMMRZCallRing3()request. */ VMMR3DECL(int) PDMR3CritSectEnterEx(PVM pVM, PPDMCRITSECT pCritSect, bool fCallRing3) { int rc = PDMCritSectEnter(pVM, pCritSect, VERR_IGNORED); if ( rc == VINF_SUCCESS && fCallRing3 && pCritSect->s.Core.pValidatorRec && pCritSect->s.Core.pValidatorRec->hThread != NIL_RTTHREAD) RTLockValidatorRecExclReleaseOwnerUnchecked(pCritSect->s.Core.pValidatorRec); return rc; } #endif /* IN_RING3 */ /** * Leaves a critical section entered with PDMCritSectEnter(). * * @returns Indication whether we really exited the critical section. * @retval VINF_SUCCESS if we really exited. * @retval VINF_SEM_NESTED if we only reduced the nesting count. * @retval VERR_NOT_OWNER if you somehow ignore release assertions. * * @param pVM The cross context VM structure. * @param pCritSect The PDM critical section to leave. * * @remarks Can be called from no-ring-3-call context in ring-0 (TM/VirtualSync) * where we'll queue leaving operation for ring-3 processing. */ VMMDECL(int) PDMCritSectLeave(PVMCC pVM, PPDMCRITSECT pCritSect) { AssertMsg(pCritSect->s.Core.u32Magic == RTCRITSECT_MAGIC, ("%p %RX32\n", pCritSect, pCritSect->s.Core.u32Magic)); Assert(pCritSect->s.Core.u32Magic == RTCRITSECT_MAGIC); /* * Check for NOP sections before asserting ownership. */ if (!(pCritSect->s.Core.fFlags & RTCRITSECT_FLAGS_NOP)) { /* We're more likely to end up here with real critsects than a NOP one. */ } else return VINF_SUCCESS; /* * Always check that the caller is the owner (screw performance). */ RTNATIVETHREAD const hNativeSelf = pdmCritSectGetNativeSelf(pVM, pCritSect); VMM_ASSERT_RELEASE_MSG_RETURN(pVM, pCritSect->s.Core.NativeThreadOwner == hNativeSelf && hNativeSelf != NIL_RTNATIVETHREAD, ("%p %s: %p != %p; cLockers=%d cNestings=%d\n", pCritSect, R3STRING(pCritSect->s.pszName), pCritSect->s.Core.NativeThreadOwner, hNativeSelf, pCritSect->s.Core.cLockers, pCritSect->s.Core.cNestings), VERR_NOT_OWNER); /* * Nested leave. */ int32_t const cNestings = pCritSect->s.Core.cNestings; Assert(cNestings >= 1); if (cNestings > 1) { #ifdef PDMCRITSECT_WITH_LESS_ATOMIC_STUFF pCritSect->s.Core.cNestings = cNestings - 1; #else ASMAtomicWriteS32(&pCritSect->s.Core.cNestings, cNestings - 1); #endif int32_t const cLockers = ASMAtomicDecS32(&pCritSect->s.Core.cLockers); Assert(cLockers >= 0); RT_NOREF(cLockers); Log12Func(("%p: cNestings=%d cLockers=%d\n", pCritSect, cNestings - 1, cLockers)); return VINF_SEM_NESTED; } Log12Func(("%p: cNestings=%d cLockers=%d hOwner=%p - leave for real\n", pCritSect, cNestings, pCritSect->s.Core.cLockers, pCritSect->s.Core.NativeThreadOwner)); #ifdef IN_RING3 /* * Ring-3: Leave for real. */ SUPSEMEVENT const hEventToSignal = pCritSect->s.hEventToSignal; pCritSect->s.hEventToSignal = NIL_SUPSEMEVENT; # if defined(PDMCRITSECT_STRICT) if (pCritSect->s.Core.pValidatorRec->hThread != NIL_RTTHREAD) RTLockValidatorRecExclReleaseOwnerUnchecked(pCritSect->s.Core.pValidatorRec); # endif Assert(!pCritSect->s.Core.pValidatorRec || pCritSect->s.Core.pValidatorRec->hThread == NIL_RTTHREAD); # ifdef PDMCRITSECT_WITH_LESS_ATOMIC_STUFF //pCritSect->s.Core.cNestings = 0; /* not really needed */ pCritSect->s.Core.NativeThreadOwner = NIL_RTNATIVETHREAD; # else ASMAtomicWriteS32(&pCritSect->s.Core.cNestings, 0); ASMAtomicWriteHandle(&pCritSect->s.Core.NativeThreadOwner, NIL_RTNATIVETHREAD); # endif ASMAtomicAndU32(&pCritSect->s.Core.fFlags, ~PDMCRITSECT_FLAGS_PENDING_UNLOCK); /* Stop profiling and decrement lockers. */ STAM_PROFILE_ADV_STOP(&pCritSect->s.StatLocked, l); ASMCompilerBarrier(); int32_t const cLockers = ASMAtomicDecS32(&pCritSect->s.Core.cLockers); if (cLockers < 0) AssertMsg(cLockers == -1, ("cLockers=%d\n", cLockers)); else { /* Someone is waiting, wake up one of them. */ Assert(cLockers < _8K); Log8(("PDMCritSectLeave: Waking up %p (cLockers=%u)\n", pCritSect, cLockers)); SUPSEMEVENT hEvent = (SUPSEMEVENT)pCritSect->s.Core.EventSem; int rc = SUPSemEventSignal(pVM->pSession, hEvent); AssertRC(rc); } /* Signal exit event. */ if (RT_LIKELY(hEventToSignal == NIL_SUPSEMEVENT)) { /* likely */ } else { Log8(("PDMCritSectLeave: Signalling %#p (%p)\n", hEventToSignal, pCritSect)); int rc = SUPSemEventSignal(pVM->pSession, hEventToSignal); AssertRC(rc); } return VINF_SUCCESS; #elif defined(IN_RING0) /* * Ring-0: Try leave for real, depends on host and context. */ SUPSEMEVENT const hEventToSignal = pCritSect->s.hEventToSignal; pCritSect->s.hEventToSignal = NIL_SUPSEMEVENT; PVMCPUCC pVCpu = VMMGetCpu(pVM); bool fQueueOnTrouble = false; /* Set this to true to test queueing. */ if ( pVCpu == NULL /* non-EMT access, if we implement it must be able to block */ || VMMRZCallRing3IsEnabled(pVCpu) || RTSemEventIsSignalSafe() || ( VMMR0ThreadCtxHookIsEnabled(pVCpu) /* Doesn't matter if Signal() blocks if we have hooks, ... */ && RTThreadPreemptIsEnabled(NIL_RTTHREAD) /* ... and preemption is still enabled, */ && ASMIntAreEnabled()) /* ... and interrupts hasn't yet been disabled. Special pre-GC HM env. */ || (fQueueOnTrouble = ( hEventToSignal == NIL_SUPSEMEVENT && ASMAtomicUoReadS32(&pCritSect->s.Core.cLockers) == 0)) ) { pCritSect->s.hEventToSignal = NIL_SUPSEMEVENT; # ifdef PDMCRITSECT_WITH_LESS_ATOMIC_STUFF //pCritSect->s.Core.cNestings = 0; /* not really needed */ pCritSect->s.Core.NativeThreadOwner = NIL_RTNATIVETHREAD; # else ASMAtomicWriteS32(&pCritSect->s.Core.cNestings, 0); ASMAtomicWriteHandle(&pCritSect->s.Core.NativeThreadOwner, NIL_RTNATIVETHREAD); # endif ASMAtomicAndU32(&pCritSect->s.Core.fFlags, ~PDMCRITSECT_FLAGS_PENDING_UNLOCK); /* * Stop profiling and decrement lockers. */ STAM_PROFILE_ADV_STOP(&pCritSect->s.StatLocked, l); ASMCompilerBarrier(); bool fQueueIt = false; int32_t cLockers; if (!fQueueOnTrouble) cLockers = ASMAtomicDecS32(&pCritSect->s.Core.cLockers); else { cLockers = -1; if (!ASMAtomicCmpXchgS32(&pCritSect->s.Core.cLockers, -1, 0)) fQueueIt = true; } if (!fQueueIt) { VMMR0EMTBLOCKCTX Ctx; bool fLeaveCtx = false; if (cLockers < 0) AssertMsg(cLockers == -1, ("cLockers=%d\n", cLockers)); else { /* Someone is waiting, wake up one of them. */ Assert(cLockers < _8K); SUPSEMEVENT hEvent = (SUPSEMEVENT)pCritSect->s.Core.EventSem; if (!RTSemEventIsSignalSafe() && (pVCpu = VMMGetCpu(pVM)) != NULL) { int rc = VMMR0EmtPrepareToBlock(pVCpu, VINF_SUCCESS, __FUNCTION__, pCritSect, &Ctx); VMM_ASSERT_RELEASE_MSG_RETURN(pVM, RT_SUCCESS(rc), ("rc=%Rrc\n", rc), rc); fLeaveCtx = true; } int rc = SUPSemEventSignal(pVM->pSession, hEvent); AssertRC(rc); } /* * Signal exit event. */ if (RT_LIKELY(hEventToSignal == NIL_SUPSEMEVENT)) { /* likely */ } else { if (!fLeaveCtx && pVCpu != NULL && !RTSemEventIsSignalSafe() && (pVCpu = VMMGetCpu(pVM)) != NULL) { int rc = VMMR0EmtPrepareToBlock(pVCpu, VINF_SUCCESS, __FUNCTION__, pCritSect, &Ctx); VMM_ASSERT_RELEASE_MSG_RETURN(pVM, RT_SUCCESS(rc), ("rc=%Rrc\n", rc), rc); fLeaveCtx = true; } Log8(("Signalling %#p\n", hEventToSignal)); int rc = SUPSemEventSignal(pVM->pSession, hEventToSignal); AssertRC(rc); } /* * Restore HM context if needed. */ if (!fLeaveCtx) { /* contention should be unlikely */ } else VMMR0EmtResumeAfterBlocking(pVCpu, &Ctx); # ifdef DEBUG_bird VMMTrashVolatileXMMRegs(); # endif return VINF_SUCCESS; } /* * Darn, someone raced in on us. Restore the state (this works only * because the semaphore is effectively controlling ownership). */ bool fRc; RTNATIVETHREAD hMessOwner = NIL_RTNATIVETHREAD; ASMAtomicCmpXchgExHandle(&pCritSect->s.Core.NativeThreadOwner, hNativeSelf, NIL_RTNATIVETHREAD, fRc, &hMessOwner); AssertLogRelMsgReturn(fRc, ("pCritSect=%p hMessOwner=%p\n", pCritSect, hMessOwner), pdmCritSectCorrupted(pCritSect, "owner race")); STAM_PROFILE_ADV_START(&pCritSect->s.StatLocked, l); # ifdef PDMCRITSECT_WITH_LESS_ATOMIC_STUFF //pCritSect->s.Core.cNestings = 1; Assert(pCritSect->s.Core.cNestings == 1); # else //Assert(pCritSect->s.Core.cNestings == 0); ASMAtomicWriteS32(&pCritSect->s.Core.cNestings, 1); # endif Assert(hEventToSignal == NIL_SUPSEMEVENT); } #else /* IN_RC */ /* * Raw-mode: Try leave it. */ # error "This context is not use..." if (pCritSect->s.Core.cLockers == 0) { # ifdef PDMCRITSECT_WITH_LESS_ATOMIC_STUFF //pCritSect->s.Core.cNestings = 0; /* not really needed */ # else ASMAtomicWriteS32(&pCritSect->s.Core.cNestings, 0); # endif ASMAtomicAndU32(&pCritSect->s.Core.fFlags, ~PDMCRITSECT_FLAGS_PENDING_UNLOCK); STAM_PROFILE_ADV_STOP(&pCritSect->s.StatLocked, l); ASMAtomicWriteHandle(&pCritSect->s.Core.NativeThreadOwner, NIL_RTNATIVETHREAD); if (ASMAtomicCmpXchgS32(&pCritSect->s.Core.cLockers, -1, 0)) return VINF_SUCCESS; /* * Darn, someone raced in on us. Restore the state (this works only * because the semaphore is effectively controlling ownership). */ bool fRc; RTNATIVETHREAD hMessOwner = NIL_RTNATIVETHREAD; ASMAtomicCmpXchgExHandle(&pCritSect->s.Core.NativeThreadOwner, hNativeSelf, NIL_RTNATIVETHREAD, fRc, &hMessOwner); AssertLogRelMsgReturn(fRc, ("pCritSect=%p hMessOwner=%p\n", pCritSect, hMessOwner), pdmCritSectCorrupted(pCritSect, "owner race")); STAM_PROFILE_ADV_START(&pCritSect->s.StatLocked, l); # ifdef PDMCRITSECT_WITH_LESS_ATOMIC_STUFF //pCritSect->s.Core.cNestings = 1; Assert(pCritSect->s.Core.cNestings == 1); # else //Assert(pCritSect->s.Core.cNestings == 0); ASMAtomicWriteS32(&pCritSect->s.Core.cNestings, 1); # endif } #endif /* IN_RC */ #ifndef IN_RING3 /* * Ring-0/raw-mode: Unable to leave. Queue the leave for ring-3. */ ASMAtomicOrU32(&pCritSect->s.Core.fFlags, PDMCRITSECT_FLAGS_PENDING_UNLOCK); # ifndef IN_RING0 PVMCPUCC pVCpu = VMMGetCpu(pVM); # endif uint32_t i = pVCpu->pdm.s.cQueuedCritSectLeaves++; LogFlow(("PDMCritSectLeave: [%d]=%p => R3\n", i, pCritSect)); VMM_ASSERT_RELEASE_MSG_RETURN(pVM, i < RT_ELEMENTS(pVCpu->pdm.s.apQueuedCritSectLeaves), ("%d\n", i), VERR_PDM_CRITSECT_IPE); pVCpu->pdm.s.apQueuedCritSectLeaves[i] = pCritSect->s.pSelfR3; VMM_ASSERT_RELEASE_MSG_RETURN(pVM, RT_VALID_PTR(pVCpu->pdm.s.apQueuedCritSectLeaves[i]) && ((uintptr_t)pVCpu->pdm.s.apQueuedCritSectLeaves[i] & HOST_PAGE_OFFSET_MASK) == ((uintptr_t)pCritSect & HOST_PAGE_OFFSET_MASK), ("%p vs %p\n", pVCpu->pdm.s.apQueuedCritSectLeaves[i], pCritSect), pdmCritSectCorrupted(pCritSect, "Invalid pSelfR3 value")); VMCPU_FF_SET(pVCpu, VMCPU_FF_PDM_CRITSECT); /** @todo handle VMCPU_FF_PDM_CRITSECT in ring-0 outside the no-call-ring-3 part. */ VMCPU_FF_SET(pVCpu, VMCPU_FF_TO_R3); /* unnecessary paranoia */ STAM_REL_COUNTER_INC(&pVM->pdm.s.StatQueuedCritSectLeaves); STAM_REL_COUNTER_INC(&pCritSect->s.StatContentionRZUnlock); return VINF_SUCCESS; #endif /* IN_RING3 */ } #if defined(IN_RING0) || defined(IN_RING3) /** * Schedule a event semaphore for signalling upon critsect exit. * * @returns VINF_SUCCESS on success. * @returns VERR_TOO_MANY_SEMAPHORES if an event was already scheduled. * @returns VERR_NOT_OWNER if we're not the critsect owner (ring-3 only). * @returns VERR_SEM_DESTROYED if RTCritSectDelete was called while waiting. * * @param pCritSect The critical section. * @param hEventToSignal The support driver event semaphore that should be * signalled. */ VMMDECL(int) PDMHCCritSectScheduleExitEvent(PPDMCRITSECT pCritSect, SUPSEMEVENT hEventToSignal) { AssertPtr(pCritSect); Assert(!(pCritSect->s.Core.fFlags & RTCRITSECT_FLAGS_NOP)); Assert(hEventToSignal != NIL_SUPSEMEVENT); # ifdef IN_RING3 if (RT_UNLIKELY(!RTCritSectIsOwner(&pCritSect->s.Core))) return VERR_NOT_OWNER; # endif if (RT_LIKELY( pCritSect->s.hEventToSignal == NIL_RTSEMEVENT || pCritSect->s.hEventToSignal == hEventToSignal)) { pCritSect->s.hEventToSignal = hEventToSignal; return VINF_SUCCESS; } return VERR_TOO_MANY_SEMAPHORES; } #endif /* IN_RING0 || IN_RING3 */ /** * Checks the caller is the owner of the critical section. * * @returns true if owner. * @returns false if not owner. * @param pVM The cross context VM structure. * @param pCritSect The critical section. */ VMMDECL(bool) PDMCritSectIsOwner(PVMCC pVM, PCPDMCRITSECT pCritSect) { #ifdef IN_RING3 RT_NOREF(pVM); return RTCritSectIsOwner(&pCritSect->s.Core); #else PVMCPUCC pVCpu = VMMGetCpu(pVM); if ( !pVCpu || pCritSect->s.Core.NativeThreadOwner != pVCpu->hNativeThread) return false; return (pCritSect->s.Core.fFlags & PDMCRITSECT_FLAGS_PENDING_UNLOCK) == 0 || pCritSect->s.Core.cNestings > 1; #endif } /** * Checks the specified VCPU is the owner of the critical section. * * @returns true if owner. * @returns false if not owner. * @param pVCpu The cross context virtual CPU structure. * @param pCritSect The critical section. */ VMMDECL(bool) PDMCritSectIsOwnerEx(PVMCPUCC pVCpu, PCPDMCRITSECT pCritSect) { #ifdef IN_RING3 NOREF(pVCpu); return RTCritSectIsOwner(&pCritSect->s.Core); #else Assert(VMCC_GET_CPU(pVCpu->CTX_SUFF(pVM), pVCpu->idCpu) == pVCpu); if (pCritSect->s.Core.NativeThreadOwner != pVCpu->hNativeThread) return false; return (pCritSect->s.Core.fFlags & PDMCRITSECT_FLAGS_PENDING_UNLOCK) == 0 || pCritSect->s.Core.cNestings > 1; #endif } /** * Checks if anyone is waiting on the critical section we own. * * @returns true if someone is waiting. * @returns false if no one is waiting. * @param pVM The cross context VM structure. * @param pCritSect The critical section. */ VMMDECL(bool) PDMCritSectHasWaiters(PVMCC pVM, PCPDMCRITSECT pCritSect) { AssertReturn(pCritSect->s.Core.u32Magic == RTCRITSECT_MAGIC, false); Assert(pCritSect->s.Core.NativeThreadOwner == pdmCritSectGetNativeSelf(pVM, pCritSect)); RT_NOREF(pVM); return pCritSect->s.Core.cLockers >= pCritSect->s.Core.cNestings; } /** * Checks if a critical section is initialized or not. * * @returns true if initialized. * @returns false if not initialized. * @param pCritSect The critical section. */ VMMDECL(bool) PDMCritSectIsInitialized(PCPDMCRITSECT pCritSect) { return RTCritSectIsInitialized(&pCritSect->s.Core); } /** * Gets the recursion depth. * * @returns The recursion depth. * @param pCritSect The critical section. */ VMMDECL(uint32_t) PDMCritSectGetRecursion(PCPDMCRITSECT pCritSect) { return RTCritSectGetRecursion(&pCritSect->s.Core); }