kern/util: use custom atomics wrapper to substantially improve codegen

This commit is contained in:
Michael Scire 2021-10-19 15:24:15 -07:00
parent 52332e8d75
commit d74f364107
26 changed files with 688 additions and 260 deletions

View file

@ -96,7 +96,7 @@ namespace ams::kern::arch::arm64::cpu {
KLightLock m_lock;
KLightLock m_cv_lock;
KLightConditionVariable m_cv;
std::atomic<u64> m_target_cores;
util::Atomic<u64> m_target_cores;
volatile Operation m_operation;
private:
static void ThreadFunction(uintptr_t _this) {
@ -109,7 +109,7 @@ namespace ams::kern::arch::arm64::cpu {
/* Wait for a request to come in. */
{
KScopedLightLock lk(m_cv_lock);
while ((m_target_cores.load() & (1ul << core_id)) == 0) {
while ((m_target_cores.Load() & (1ul << core_id)) == 0) {
m_cv.Wait(std::addressof(m_cv_lock));
}
}
@ -120,7 +120,7 @@ namespace ams::kern::arch::arm64::cpu {
/* Broadcast, if there's nothing pending. */
{
KScopedLightLock lk(m_cv_lock);
if (m_target_cores.load() == 0) {
if (m_target_cores.Load() == 0) {
m_cv.Broadcast();
}
}
@ -129,7 +129,7 @@ namespace ams::kern::arch::arm64::cpu {
void ProcessOperation();
public:
constexpr KCacheHelperInterruptHandler() : KInterruptHandler(), m_lock(), m_cv_lock(), m_cv(), m_target_cores(), m_operation(Operation::Idle) { /* ... */ }
constexpr KCacheHelperInterruptHandler() : KInterruptHandler(), m_lock(), m_cv_lock(), m_cv(), m_target_cores(0), m_operation(Operation::Idle) { /* ... */ }
void Initialize(s32 core_id) {
/* Reserve a thread from the system limit. */
@ -163,7 +163,7 @@ namespace ams::kern::arch::arm64::cpu {
if ((op == Operation::InstructionMemoryBarrier) || (Kernel::GetState() == Kernel::State::Initializing)) {
/* Check that there's no on-going operation. */
MESOSPHERE_ABORT_UNLESS(m_operation == Operation::Idle);
MESOSPHERE_ABORT_UNLESS(m_target_cores.load() == 0);
MESOSPHERE_ABORT_UNLESS(m_target_cores.Load() == 0);
/* Set operation. */
m_operation = op;
@ -171,13 +171,13 @@ namespace ams::kern::arch::arm64::cpu {
/* For certain operations, we want to send an interrupt. */
m_target_cores = other_cores_mask;
const u64 target_mask = m_target_cores.load();
const u64 target_mask = m_target_cores.Load();
DataSynchronizationBarrier();
Kernel::GetInterruptManager().SendInterProcessorInterrupt(KInterruptName_CacheOperation, target_mask);
this->ProcessOperation();
while (m_target_cores.load() != 0) {
while (m_target_cores.Load() != 0) {
cpu::Yield();
}
@ -189,7 +189,7 @@ namespace ams::kern::arch::arm64::cpu {
/* Check that there's no on-going operation. */
MESOSPHERE_ABORT_UNLESS(m_operation == Operation::Idle);
MESOSPHERE_ABORT_UNLESS(m_target_cores.load() == 0);
MESOSPHERE_ABORT_UNLESS(m_target_cores.Load() == 0);
/* Set operation. */
m_operation = op;
@ -199,7 +199,7 @@ namespace ams::kern::arch::arm64::cpu {
/* Use the condvar. */
m_cv.Broadcast();
while (m_target_cores.load() != 0) {
while (m_target_cores.Load() != 0) {
m_cv.Wait(std::addressof(m_cv_lock));
}
@ -287,7 +287,7 @@ namespace ams::kern::arch::arm64::cpu {
break;
}
m_target_cores &= ~(1ul << GetCurrentCoreId());
m_target_cores.FetchAnd(~(1ul << GetCurrentCoreId()));
}
ALWAYS_INLINE void SetEventLocally() {