Merge pull request #9215 from liamwhite/swordfight

Ensure correctness of atomic store ordering

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 b1cabbca0..d6676904b 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 73314b45e..129d60472 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);