AuroraRuntime/Source/Threading/AuWakeOnAddress.cpp

1089 lines
32 KiB
C++

/***
Copyright (C) 2023 J Reece Wilson (a/k/a "Reece"). All rights reserved.
File: AuWakeOnAddress.cpp
Date: 2023-3-10
Author: Reece
***/
#include <Source/RuntimeInternal.hpp>
#include "AuWakeOnAddress.hpp"
#include "Primitives/SMTYield.hpp"
#include <Time/Time.hpp>
#define HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD
namespace Aurora::Threading
{
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
static thread_local AuSPtr<WaitEntry> tlsWaitEntry = AuMakeSharedPanic<WaitEntry>();
#else
static thread_local WaitEntry tlsWaitEntry;
#endif
static ProcessWaitContainer gProcessWaitables;
static int gShouldSpinOnlyInCPU = 1; // TODO: havent decided
template<typename T>
static void DoSpinLockOnVar(T *uPointer)
{
if (gShouldSpinOnlyInCPU == 0)
{
while (!Primitives::DoTryIf([&]()
{
return AuAtomicTestAndSet(uPointer, 0) == 0;
}))
{
}
}
else if (gShouldSpinOnlyInCPU == 1)
{
while (!Primitives::DoTryIf([&]()
{
return AuAtomicTestAndSet(uPointer, 0) == 0;
}))
{
ContextYield();
}
}
else if (gShouldSpinOnlyInCPU == 2)
{
while (AuAtomicTestAndSet(uPointer, 0))
{
while (*uPointer)
{
ContextYield();
}
}
}
else
{
SysUnreachable();
}
}
bool WaitEntry::TryAcquire(const void *pAddress, AuUInt8 uSize)
{
DoSpinLockOnVar(&this->uAtomic);
//AU_LOCK_GUARD(this->mutex);
if (this->pAddress)
{
this->uAtomic = 0;
return false;
}
this->pAddress = pAddress;
this->uSize = uSize;
this->uAtomic = 0;
return true;
}
void WaitEntry::Release()
{
AuResetMember(this->uSize);
AuResetMember(this->pAddress);
}
WaitEntry::WaitEntry()
{
}
WaitEntry::~WaitEntry()
{
this->Release();
}
bool WaitEntry::SleepOn(WaitState &state)
{
AU_LOCK_GUARD(this->mutex);
if (state.qwNanosecondsAbs)
{
if (!WaitBuffer::Compare(this->pAddress, this->uSize, state))
{
return true;
}
auto uNow = AuTime::SteadyClockNS();
auto uEndTime = state.qwNanosecondsAbs.value();
while (uNow < uEndTime)
{
if (!WaitBuffer::Compare(this->pAddress, this->uSize, state))
{
return true;
}
auto uTimeRemNS = uEndTime - uNow;
#if defined(AURORA_PLATFORM_WIN32)
Win32DropSchedulerResolution();
#endif
this->variable.WaitForSignalNsEx(&this->mutex, uTimeRemNS);
uNow = AuTime::SteadyClockNS();
}
return !WaitBuffer::Compare(this->pAddress, this->uSize, state);
}
else
{
while (WaitBuffer::Compare(this->pAddress, this->uSize, state))
{
this->variable.WaitForSignalNsEx(&this->mutex, 0);
}
return true;
}
return false;
}
bool WaitEntry::TryWake(const void *pAddress)
{
DoSpinLockOnVar(&this->uAtomic);
auto bRet = TryWakeNoLock(pAddress);
if (!bRet)
{
this->uAtomic = 0;
}
return bRet;
}
bool WaitEntry::TryWakeNoLockNoReallyNoLock(const void *pAddress)
{
if (this->pAddress != pAddress)
{
return false;
}
this->uAtomic = 0;
this->variable.Signal();
return true;
}
bool WaitEntry::TryWakeNoLock(const void *pAddress)
{
if (this->pAddress != pAddress)
{
return false;
}
AU_LOCK_GUARD(this->mutex);
this->uAtomic = 0;
this->variable.Signal();
return true;
}
WaitBuffer WaitBuffer::From(const void *pBuf, AuUInt8 uSize)
{
WaitBuffer wait;
AuMemcpy(wait.buffer, pBuf, uSize);
wait.uSize = uSize;
return AuMove(wait);
}
bool WaitBuffer::Compare(const void *pBuf, AuUInt8 uSize, WaitState &state)
{
if (!state.uDownsizeMask)
{
return AuMemcmp(pBuf, state.compare.buffer, AuMin(uSize, state.compare.uSize)) == 0;
}
else
{
auto uMask = state.uDownsizeMask.value();
auto &uSrcWord = *AuReinterpretCast<const AuUInt32 *>(pBuf);
auto &uCmpWord = *AuReinterpretCast<const AuUInt32 *>(state.compare.buffer);
return (uSrcWord & uMask) == (uCmpWord & uMask);
}
}
bool WaitBuffer::Compare(const void *pBuf, AuUInt8 uSize, const void *pBuf2)
{
switch (uSize)
{
case 1:
return AuReadU8(pBuf, 0) == AuReadU8(pBuf2, 0);
case 2:
return AuReadU16(pBuf, 0) == AuReadU16(pBuf2, 0);
case 4:
return AuReadU32(pBuf, 0) == AuReadU32(pBuf2, 0);
case 8:
return AuReadU64(pBuf, 0) == AuReadU64(pBuf2, 0);
default:
return AuMemcmp(pBuf, pBuf2, uSize) == 0;
}
}
bool WaitBuffer::Compare(const void *pBuf)
{
return WaitBuffer::Compare(this->buffer, this->uSize, pBuf);
}
bool WaitBuffer::Compare(WaitState &state)
{
if (!state.uDownsizeMask)
{
return WaitBuffer::Compare(this->buffer, AuMin(this->uSize, state.compare.uSize), state.compare.buffer);
}
else
{
auto uMask = state.uDownsizeMask.value();
auto &uSrcWord = *AuReinterpretCast<const AuUInt32 *>(this->buffer);
auto &uCmpWord = *AuReinterpretCast<const AuUInt32 *>(state.compare.buffer);
return (uSrcWord & uMask) == (uCmpWord & uMask);
}
}
WaitEntry *ProcessWaitNodeContainer::WaitBufferFrom(const void *pAddress, AuUInt8 uSize)
{
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
auto pReturn = tlsWaitEntry.get();
#else
auto pReturn = &tlsWaitEntry;
#endif
pReturn->bReleaseOnWake = true;
pReturn->pAddress = pAddress;
pReturn->uSize = uSize;
pReturn->uAtomic = 0;
{
Lock();
if (auto pLoadFromMemory = this->waitList.pHead)
{
pReturn->pNext = pLoadFromMemory;
pLoadFromMemory->pBefore = pReturn;
}
else
{
this->waitList.pTail = pReturn;
}
this->waitList.pHead = pReturn;
Unlock();
}
return pReturn;
}
template <typename T>
bool ProcessWaitNodeContainer::IterateWake(T callback)
{
bool bRetStatus { true };
if (AuAtomicLoad((AuUInt *)&this->waitList.pTail) == 0)
{
return true;
}
Lock();
{
// FIFO
auto pCurrentHead = this->waitList.pTail;
decltype(pCurrentHead) pLast {};
while (pCurrentHead)
{
AU_LOCK_GUARD(pCurrentHead->mutex);
auto [bCont, bRemove] = callback(*pCurrentHead);
if (bRemove)
{
if (pLast)
{
pLast->pNext = pCurrentHead->pNext;
}
if (this->waitList.pHead == pCurrentHead)
{
this->waitList.pHead = pCurrentHead->pNext;
}
if (pCurrentHead->pNext)
{
pCurrentHead->pNext->pBefore = pCurrentHead->pBefore;
}
if (this->waitList.pTail == pCurrentHead)
{
this->waitList.pTail = pCurrentHead->pBefore;
}
}
if (!bCont)
{
bRetStatus = false;
break;
}
pLast = pCurrentHead;
if (pCurrentHead->pBefore == pCurrentHead)
{
break;
}
pCurrentHead = pCurrentHead->pBefore;
}
}
Unlock();
return bRetStatus;
}
void ProcessWaitNodeContainer::RemoveSelf(WaitEntry *pSelf)
{
Lock();
{
auto pCurrentHead = this->waitList.pTail;
decltype(pCurrentHead) pLast {};
while (pCurrentHead)
{
if (pCurrentHead == pSelf)
{
if (pLast)
{
pLast->pNext = pCurrentHead->pNext;
}
if (this->waitList.pHead == pCurrentHead)
{
this->waitList.pHead = pCurrentHead->pNext;
}
if (pCurrentHead->pNext)
{
pCurrentHead->pNext->pBefore = pCurrentHead->pBefore;
}
if (this->waitList.pTail == pCurrentHead)
{
this->waitList.pTail = pCurrentHead->pBefore;
}
break;
}
pLast = pCurrentHead;
if (pCurrentHead->pBefore == pCurrentHead)
{
break;
}
pCurrentHead = pCurrentHead->pBefore;
}
}
Unlock();
}
void ProcessWaitNodeContainer::Lock()
{
DoSpinLockOnVar(&this->uAtomic);
}
void ProcessWaitNodeContainer::Unlock()
{
this->uAtomic = 0;
}
#define AddressToIndex AuHashCode(pAddress) & (AuArraySize(this->list) - 1)
WaitEntry *ProcessWaitContainer::WaitBufferFrom(const void *pAddress, AuUInt8 uSize)
{
return this->list[AddressToIndex].WaitBufferFrom(pAddress, uSize);
}
template <typename T>
bool ProcessWaitContainer::IterateWake(const void *pAddress, T callback)
{
return this->list[AddressToIndex].IterateWake(callback);
}
void ProcessWaitContainer::RemoveSelf(const void *pAddress, WaitEntry *pSelf)
{
return this->list[AddressToIndex].RemoveSelf(pSelf);
}
bool IsNativeWaitOnSupported()
{
#if defined(AURORA_IS_MODERNNT_DERIVED)
return pWaitOnAddress &&
AuSwInfo::IsWindows8Point1OrGreater();
#elif defined(AURORA_PLATFORM_LINUX)
return true;
#else
return false;
#endif
}
AUKN_SYM bool IsWaitOnRecommended()
{
static AuOptionalEx<bool> gIsWaitOnRecommendedCache {};
if (gIsWaitOnRecommendedCache)
{
return gIsWaitOnRecommendedCache.value();
}
if (Primitives::ThrdCfg::gPreferEmulatedWakeOnAddress)
{
return false;
}
bool bState = IsNativeWaitOnSupported();
gIsWaitOnRecommendedCache = bState;
return bState;
}
/// @deprecated
AUKN_SYM const AuList<AuUInt8> &GetValidWordSizes()
{
static const AuList<AuUInt8> kArray =
#if defined(AURORA_IS_MODERNNT_DERIVED)
{ 1, 2, 4, 8 };
#else
{ 4 };
#endif
return kArray;
}
bool WaitOnAddressWide(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuOptional<AuUInt64> qwNanoseconds,
AuOptional<AuUInt64> qwNanosecondsAbs,
bool bOSSupportsWait
)
{
WaitState state;
SysAssertDbg(uWordSize <= 32);
auto pWaitEntry = gProcessWaitables.WaitBufferFrom(pTargetAddress, uWordSize);
state.compare = WaitBuffer::From(pCompareAddress, uWordSize);
if (qwNanoseconds)
{
state.qwNanosecondsAbs = AuTime::SteadyClockNS() + qwNanoseconds.value();
}
else if (qwNanosecondsAbs)
{
state.qwNanosecondsAbs = qwNanosecondsAbs.value();
}
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
auto pTempHoldMe = tlsWaitEntry;
#endif
auto bResult = pWaitEntry->SleepOn(state);
#if defined(HACK_NO_INVALID_ACCESS_LEAK_SHARED_REF_ON_DESTROYED_THREAD)
pTempHoldMe.reset();
#endif
if (!bResult)
{
gProcessWaitables.RemoveSelf(pTargetAddress, pWaitEntry);
}
return bResult;
}
AuTuple<const void *, AuUInt8, AuOptionalEx<AuUInt32>> DecodeAddress(const void *pAddress,
AuUInt32 uWordSize)
{
#if defined(AURORA_IS_MODERNNT_DERIVED)
return AuMakeTuple(pAddress, 0, AuOptionalEx<AuUInt32> {});
#endif
auto pRounded = AuPageRound(AuUInt(pAddress), AuUInt(4));
auto uDelta = (AuUInt)pAddress - pRounded;
if (uWordSize == 8)
{
return AuMakeTuple((const void *)pRounded, uDelta, 0xFFFFFFFF);
}
AuUInt32 uSizeMask = (1ull << (uWordSize * 8)) - 1ull;
switch (uDelta)
{
case 0:
return AuMakeTuple(pAddress, 0, 0xFFFFFFFF & (uSizeMask << 0));
case 1:
return AuMakeTuple(pAddress, 1, 0xFFFFFF00 & (uSizeMask << 8));
case 2:
return AuMakeTuple(pAddress, 2, 0xFFFF0000 & (uSizeMask << 16));
case 3:
return AuMakeTuple(pAddress, 3, 0xFF000000 & (uSizeMask << 24));
default:
SysPanic("Invalid Branch");
}
}
static bool RunOSWaitOnAddressNoTimed(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize)
{
#if defined(AURORA_IS_MODERNNT_DERIVED)
if (pRtlWaitOnAddress)
{
return pRtlWaitOnAddress((void *)pTargetAddress, (void *)pCompareAddress, uWordSize, nullptr);
}
else
{
return pWaitOnAddress((void *)pTargetAddress, (void *)pCompareAddress, uWordSize, INFINITE);
}
#endif
#if defined(AURORA_IS_LINUX_DERIVED)
int ret {};
#if defined(AU_CPU_ENDIAN_BIG)
if (uWordSize == 8)
{
pTargetAddress = AuReinterpretCast<const char *>(pTargetAddress) + 4;
pCompareAddress = AuReinterpretCast<const char *>(pCompareAddress) + 4;
}
#endif
auto uCurrent = *(AuUInt32 *)pCompareAddress;
auto expect = WaitBuffer::From(pCompareAddress, uWordSize);
do
{
ret = futex_wait((AuUInt32 *)pTargetAddress, uCurrent, nullptr);
if (ret == 0)
{
continue;
}
if (ret == -EAGAIN)
{
continue;
}
if (ret == -ETIMEDOUT)
{
return false;
}
}
while (ret == -EINTR);
return !expect.Compare(pTargetAddress);
#endif
return false;
}
static bool RunOSWaitOnAddressTimed(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuUInt64 uAbsTimeSteadyClock,
AuUInt64 uRelativeNanoseconds,
AuOptional<AuUInt64> uAbsTimeAltClock /* hint */,
bool bSpun = false)
{
#if defined(AURORA_IS_MODERNNT_DERIVED)
if (pRtlWaitOnAddress)
{
auto expect = WaitBuffer::From(pCompareAddress, uWordSize);
AuUInt64 uNow {};
while (uAbsTimeSteadyClock ?
(uAbsTimeSteadyClock > (uNow = AuTime::SteadyClockNS())) :
true)
{
LARGE_INTEGER word {};
if (uAbsTimeAltClock)
{
word.QuadPart = AuTime::ConvertTimestampNs(uAbsTimeAltClock.value());
}
else if (uAbsTimeSteadyClock)
{
if (uAbsTimeSteadyClock <= uNow)
{
return !expect.Compare(pTargetAddress);
}
word.QuadPart = -(AuInt64(uAbsTimeSteadyClock - uNow) / 100ull);
if (!word.QuadPart)
{
word.QuadPart = 1;
}
}
if (expect.Compare(pTargetAddress))
{
pRtlWaitOnAddress(pTargetAddress, pCompareAddress, uWordSize, &word);
if (!expect.Compare(pTargetAddress))
{
return true;
}
else if (!uAbsTimeSteadyClock)
{
return false;
}
}
else
{
return true;
}
}
return false;
}
else
{
// ~~some paths might miss the uRelativeNanoseconds, like cas loops.~~
// most paths will now skimp on the relative values
if (uAbsTimeSteadyClock && !uRelativeNanoseconds)
{
AuInt64 iDelta = uAbsTimeSteadyClock;
iDelta -= AuTime::SteadyClockNS();
if (iDelta <= 0)
{
return !WaitBuffer::Compare(pCompareAddress, uWordSize, pTargetAddress);
}
uRelativeNanoseconds = iDelta;
}
auto uMaxSwitches = gRuntimeConfig.threadingConfig.uUWPNanosecondEmulationMaxYields;
auto bUWPNanosecondEmulationCheckFirst = Primitives::ThrdCfg::gUWPNanosecondEmulationCheckFirst;
// LockN(<1MS) on a platform without that resolution of yielding... damn
auto uMS = AuNSToMS<AuUInt32>(uRelativeNanoseconds);
if (!uMS)
{
// take a copy
auto expect = WaitBuffer::From(pCompareAddress, uWordSize);
// first: cpu spin to avoid the kernel all together
if (!bSpun)
{
if (TryWaitOnAddress((void *)pTargetAddress, (void *)pCompareAddress, uWordSize))
{
return true;
}
}
// second: yield
unsigned uLimit {};
do
{
if (!expect.Compare(pTargetAddress))
{
break;
}
AuThreading::ContextYield();
if (bUWPNanosecondEmulationCheckFirst)
{
if (uLimit++ > uMaxSwitches)
{
break;
}
}
}
while (uAbsTimeSteadyClock > AuTime::SteadyClockNS()); // ...until times up
}
else // high level lock function was called with ms scale resolution
{
// first: wait on the address with an ms scale timeout
(void)pWaitOnAddress((void *)pTargetAddress, (void *)pCompareAddress, uWordSize, uMS);
// take a copy
auto expect = WaitBuffer::From(pCompareAddress, uWordSize);
// never trust the error value/status provided by wait addresses - instead, do a quick compare
if (!expect.Compare(pTargetAddress))
{
// best case: we woke up during the ms-res waitonaddress
return true;
}
// attempt to yield again, potentially context switching a few times to hit any NS remainder
AuUInt64 uNow {};
unsigned uLimit {};
while (uAbsTimeSteadyClock > (uNow = AuTime::SteadyClockNS()))
{
uMS = AuNSToMS<AuUInt32>(uAbsTimeSteadyClock - uNow);
if (Primitives::DoTryIf([&]()
{
return !expect.Compare(pTargetAddress);
}))
{
// hit it within the span of 1 << SpinLoopPowerA SMT stalls
return true;
}
if (!uMS)
{
// burn off any remainder cycles by switching contexts (this isnt a very long time usually)
if (uLimit++ < uMaxSwitches)
{
AuThreading::ContextYield();
}
else
{
// do not burn the cpu to meet the timeout. we'll just undershoot.
return false;
}
}
else
{
(void)pWaitOnAddress((void *)pTargetAddress, (void *)pCompareAddress, uWordSize, uMS);
}
}
}
}
#endif
#if defined(AURORA_IS_LINUX_DERIVED)
int ret {};
#if defined(AU_CPU_ENDIAN_BIG)
if (uWordSize == 8)
{
pTargetAddress = AuReinterpretCast<const char *>(pTargetAddress) + 4;
pCompareAddress = AuReinterpretCast<const char *>(pCompareAddress) + 4;
}
#endif
auto uCurrent = *(AuUInt32 *)pCompareAddress;
struct timespec tspec;
Time::monoabsns2ts(&tspec, uAbsTimeAltClock ? uAbsTimeAltClock.value() : uAbsTimeSteadyClock);
do
{
ret = futex_wait((AuUInt32 *)pTargetAddress, uCurrent, &tspec);
if (ret == 0)
{
continue;
}
if (ret == -EAGAIN)
{
continue;
}
if (ret == -ETIMEDOUT)
{
return false;
}
}
while (ret == -EINTR);
#endif
return !WaitBuffer::Compare(pCompareAddress, uWordSize, pTargetAddress);
}
static void RunOSWaitOnAddressNoTimedNoErrors(const void *pTargetAddress,
const void *pCompareAddress,
WaitState &state)
{
while (WaitBuffer::Compare(pTargetAddress, state.uWordSize, state))
{
if (!RunOSWaitOnAddressNoTimed(pTargetAddress, pCompareAddress, state.uWordSize))
{
AuThreading::ContextYield();
}
}
}
static bool RunOSWaitOnAddressTimedSteady(const void *pTargetAddress,
const void *pCompareAddress,
WaitState &state,
bool bSpun = false)
{
if (!WaitBuffer::Compare(pTargetAddress, state.uWordSize, state))
{
return true;
}
(void)RunOSWaitOnAddressTimed(pTargetAddress, pCompareAddress, state.uWordSize, state.qwNanosecondsAbs.value(), { }, { }, bSpun);
return !WaitBuffer::Compare(pTargetAddress, state.uWordSize, state);
}
static void RunOSWakeNOnAddress(const void *pAddress,
AuUInt32 dwCount)
{
#if defined(AURORA_IS_LINUX_DERIVED)
futex_wake((AuUInt32 *)pAddress, dwCount);
#endif
#if defined(AURORA_IS_MODERNNT_DERIVED)
if (pRtlWakeAddressSingle)
{
if (dwCount < 6)
{
for (AuUInt i = 0; i < dwCount; i++)
{
pRtlWakeAddressSingle((void *)pAddress);
}
}
else
{
pRtlWakeByAddressAll((void *)pAddress);
}
}
else
{
for (AuUInt i = 0; i < dwCount; i++)
{
pWakeByAddressSingle((void *)pAddress);
}
}
#endif
}
static void RunOSWakeAllOnAddress(const void *pAddress)
{
#if defined(AURORA_IS_LINUX_DERIVED)
futex_wake((AuUInt32 *)pAddress, INT_MAX);
#endif
#if defined(AURORA_IS_MODERNNT_DERIVED)
if (pRtlWakeByAddressAll)
{
pRtlWakeByAddressAll((void *)pAddress);
}
else
{
pWakeByAddressAll((void *)pAddress);
}
#endif
}
// Windows 8+ thread primitives might use me instead of the public API
// it does work on Linux and Windows 8+
// it does not, however, work on emulated platforms
// this is intentional
bool InternalLTSWaitOnAddressHighRes(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuUInt64 qwNanosecondsAbs)
{
auto [pWaitAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, uWordSize);
auto pCompareAddress2 = AuReinterpretCast<const char *>(pCompareAddress) - uDelta;
WaitState state;
state.uDownsizeMask = uMask;
state.compare = uMask ?
WaitBuffer::From(pCompareAddress2, 4) :
WaitBuffer::From(pCompareAddress2, uWordSize);
state.uWordSize = uMask ? 4 : uWordSize;
if (!qwNanosecondsAbs)
{
RunOSWaitOnAddressNoTimedNoErrors(pWaitAddress, pCompareAddress2, state);
return true;
}
else
{
state.qwNanosecondsAbs = qwNanosecondsAbs;
return RunOSWaitOnAddressTimedSteady(pWaitAddress, pCompareAddress2, state, true);
}
}
void InternalLTSWakeAll(const void *pTargetAddress)
{
auto [pWakeAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, 1);
RunOSWakeAllOnAddress(pWakeAddress);
}
void InternalLTSWakeOne(const void *pTargetAddress)
{
auto [pWakeAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, 1);
RunOSWakeNOnAddress(pWakeAddress, uDelta ? INT_MAX : 1);
}
void InternalLTSWakeCount(const void *pTargetAddress, AuUInt32 uCount)
{
auto [pWakeAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, 1);
RunOSWakeNOnAddress(pWakeAddress, uDelta ? INT_MAX : uCount);
}
AUKN_SYM bool WaitOnAddress(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuUInt64 qwNanoseconds,
AuOptional<bool> optAlreadySpun)
{
// Avoid SteadyTime syscall in the event of HAL retardation (missing KUSER QPC, Linux vDSO, etc)
if (!WaitBuffer::Compare(pTargetAddress, uWordSize, pCompareAddress))
{
return true;
}
return WaitOnAddressSteady(pTargetAddress,
pCompareAddress,
uWordSize,
qwNanoseconds ? qwNanoseconds + AuTime::SteadyClockNS() : 0,
optAlreadySpun);
}
AUKN_SYM bool TryWaitOnAddress(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize)
{
return Primitives::DoTryIf([&]()
{
return !WaitBuffer::Compare(pCompareAddress, uWordSize, pTargetAddress);
});
}
AUKN_SYM bool TryWaitOnAddressEx(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
const AuFunction<bool(const void *, const void *, AuUInt8)> &check)
{
if (!check)
{
return TryWaitOnAddress(pTargetAddress, pCompareAddress, uWordSize);
}
return Primitives::DoTryIf([&]()
{
if (WaitBuffer::Compare(pCompareAddress, uWordSize, pTargetAddress))
{
return false;
}
return check(pTargetAddress, pCompareAddress, uWordSize);
});
}
AUKN_SYM void WakeNOnAddress(const void *pTargetAddress,
AuUInt8 uNMaximumThreads)
{
if (IsWaitOnRecommended())
{
auto [pWakeAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, 1);
RunOSWakeNOnAddress(pWakeAddress, uDelta ? INT_MAX : uNMaximumThreads);
}
else
{
(void)gProcessWaitables.IterateWake(pTargetAddress, [&](WaitEntry &entry) -> AuPair<bool, bool>
{
if (!uNMaximumThreads)
{
return AuMakePair(false, false);
}
bool bWake {};
if (entry.TryWakeNoLockNoReallyNoLock(pTargetAddress))
{
bWake = true;
uNMaximumThreads--;
}
bool bCont = uNMaximumThreads != 0;
return AuMakePair(bCont, bWake);
});
}
}
AUKN_SYM void WakeOnAddress(const void *pTargetAddress)
{
WakeNOnAddress(pTargetAddress, 1);
}
AUKN_SYM void WakeAllOnAddress(const void *pTargetAddress)
{
if (IsWaitOnRecommended())
{
auto [pWakeAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, 1);
RunOSWakeAllOnAddress(pWakeAddress);
}
else
{
(void)gProcessWaitables.IterateWake(pTargetAddress, [&](WaitEntry &entry) -> AuPair<bool, bool>
{
return AuMakePair(true, entry.TryWakeNoLockNoReallyNoLock(pTargetAddress));
});
}
}
AUKN_SYM bool WaitOnAddressSteady(const void *pTargetAddress,
const void *pCompareAddress,
AuUInt8 uWordSize,
AuUInt64 qwNanoseconds,
AuOptional<bool> optAlreadySpun)
{
bool bWaitOnAddress = IsWaitOnRecommended();
if (bWaitOnAddress)
{
auto [pWaitAddress, uDelta, uMask] = DecodeAddress(pTargetAddress, uWordSize);
auto pCompareAddress2 = AuReinterpretCast<const char *>(pCompareAddress) - uDelta;
WaitState state;
state.uDownsizeMask = uMask;
state.compare = uMask ?
WaitBuffer::From(pCompareAddress2, 4) :
WaitBuffer::From(pCompareAddress2, uWordSize);
state.uWordSize = uMask ? 4 : uWordSize;
bool bSpun {};
if (Primitives::ThrdCfg::gPreferWaitOnAddressAlwaysSpinNative &&
optAlreadySpun.value_or(false))
{
if (TryWaitOnAddress(pTargetAddress, pCompareAddress, uWordSize))
{
return true;
}
bSpun = true;
}
if (!qwNanoseconds)
{
RunOSWaitOnAddressNoTimedNoErrors(pWaitAddress, pCompareAddress2, state);
return true;
}
else
{
state.qwNanosecondsAbs = qwNanoseconds;
return RunOSWaitOnAddressTimedSteady(pWaitAddress, pCompareAddress2, state, bSpun);
}
}
else
{
if (Primitives::ThrdCfg::gPreferWaitOnAddressAlwaysSpin &&
optAlreadySpun.value_or(false))
{
if (TryWaitOnAddress(pTargetAddress, pCompareAddress, uWordSize))
{
return true;
}
}
return WaitOnAddressWide(pTargetAddress, pCompareAddress, uWordSize, {}, qwNanoseconds, false);
}
return false;
}
}