From 71c0e20f95861a3766c04bc92afe677205848407 Mon Sep 17 00:00:00 2001 From: Liam Date: Tue, 8 Nov 2022 22:26:14 -0500 Subject: [PATCH] Ensure correctness of atomic store ordering --- src/core/hle/kernel/k_scheduler.cpp | 9 ++++++--- src/core/hle/kernel/k_scheduler_lock.h | 3 +++ 2 files changed, 9 insertions(+), 3 deletions(-) diff --git a/src/core/hle/kernel/k_scheduler.cpp b/src/core/hle/kernel/k_scheduler.cpp index b1cabbca06..d6676904ba 100644 --- a/src/core/hle/kernel/k_scheduler.cpp +++ b/src/core/hle/kernel/k_scheduler.cpp @@ -384,7 +384,8 @@ void KScheduler::SwitchThread(KThread* next_thread) { void KScheduler::ScheduleImpl() { // First, clear the needs scheduling bool. - m_state.needs_scheduling.store(false, std::memory_order_seq_cst); + m_state.needs_scheduling.store(false, std::memory_order_relaxed); + std::atomic_thread_fence(std::memory_order_seq_cst); // Load the appropriate thread pointers for scheduling. KThread* const cur_thread{GetCurrentThreadPointer(kernel)}; @@ -400,7 +401,8 @@ void KScheduler::ScheduleImpl() { // If there aren't, we want to check if the highest priority thread is the same as the current // thread. if (highest_priority_thread == cur_thread) { - // If they're the same, then we can just return. + // If they're the same, then we can just issue a memory barrier and return. + std::atomic_thread_fence(std::memory_order_seq_cst); return; } @@ -476,7 +478,8 @@ void KScheduler::ScheduleImplFiber() { // We failed to successfully do the context switch, and need to retry. // Clear needs_scheduling. - m_state.needs_scheduling.store(false, std::memory_order_seq_cst); + m_state.needs_scheduling.store(false, std::memory_order_relaxed); + std::atomic_thread_fence(std::memory_order_seq_cst); // Refresh the highest priority thread. highest_priority_thread = m_state.highest_priority_thread; diff --git a/src/core/hle/kernel/k_scheduler_lock.h b/src/core/hle/kernel/k_scheduler_lock.h index 73314b45e9..129d604722 100644 --- a/src/core/hle/kernel/k_scheduler_lock.h +++ b/src/core/hle/kernel/k_scheduler_lock.h @@ -60,6 +60,9 @@ public: // Release an instance of the lock. if ((--lock_count) == 0) { + // Perform a memory barrier here. + std::atomic_thread_fence(std::memory_order_seq_cst); + // We're no longer going to hold the lock. Take note of what cores need scheduling. const u64 cores_needing_scheduling = SchedulerType::UpdateHighestPriorityThreads(kernel);