diff --git a/core/os/spin_lock.h b/core/os/spin_lock.h
index d386cd589052..fa345ff3418f 100644
--- a/core/os/spin_lock.h
+++ b/core/os/spin_lock.h
@@ -52,19 +52,35 @@ class SpinLock {
 
 #else
 
+#include "core/os/thread.h"
+
 #include <atomic>
+#include <thread>
 
-class SpinLock {
-	mutable std::atomic_flag locked = ATOMIC_FLAG_INIT;
+static_assert(std::atomic_bool::is_always_lock_free);
+
+class alignas(Thread::CACHE_LINE_BYTES) SpinLock {
+	mutable std::atomic_bool locked = ATOMIC_VAR_INIT(false);
 
 public:
 	_ALWAYS_INLINE_ void lock() const {
-		while (locked.test_and_set(std::memory_order_acquire)) {
-			// Continue.
+		while (true) {
+			bool expected = false;
+			if (locked.compare_exchange_weak(expected, true, std::memory_order_acq_rel, std::memory_order_relaxed)) {
+				break;
+			}
+			do {
+				std::this_thread::yield();
+			} while (locked.load(std::memory_order_relaxed));
 		}
 	}
+
 	_ALWAYS_INLINE_ void unlock() const {
-		locked.clear(std::memory_order_release);
+		locked.store(false, std::memory_order_release);
+	}
+
+	_ALWAYS_INLINE_ void acquire() const {
+		(void)locked.load(std::memory_order_acquire);
 	}
 };
 
diff --git a/core/os/thread.h b/core/os/thread.h
index a0ecc24c9165..d83b00aafcb6 100644
--- a/core/os/thread.h
+++ b/core/os/thread.h
@@ -85,6 +85,12 @@ class Thread {
 		void (*term)() = nullptr;
 	};
 
+#ifdef __cpp_lib_hardware_interference_size
+	static const size_t CACHE_LINE_BYTES = std::hardware_constructive_interference_size;
+#else
+	static const size_t CACHE_LINE_BYTES = 64;
+#endif
+
 private:
 	friend class Main;