c++ - ThreadSanitizer 说我的 spin_lock 有数据竞争，但是怎么办？

Question

我的自旋锁实现如下所示，我认为它不会导致任何数据竞争，但是当我使用 -fsanitize=thread 测试我的代码时，它报告 spin_unlock 有写数据竞争。怎么会这样？是误报吗？

#define barrier() asm volatile("": : :"memory")
#define cpu_relax() asm volatile("pause\n": : :"memory")

static inline void spin_lock(volatile int *lock) {
    while (1)
    {
        int i = 0;
        if (!atomic_swap(lock, EBUSY)) return;
        while (*lock) {
            i++;
            if (i == 4000) {
                i = 0;
                thread_yield();
            }
            cpu_relax();
        }
    }
}

static inline void spin_unlock(volatile int *lock) {
    barrier();
    *lock = 0;
}

atomic_swap 是一个函数:

static inline int atomic_swap(volatile void *lockword, int value) {
    unsigned long tmp;
    int result;
    __asm__ __volatile__ ("dmb" : : : "memory");
    __asm__ __volatile__("@ atomic_swap\n"
    "1: ldrex   %0, [%2]\n"
    "   strex   %1, %3, [%2]\n"
    "   teq %1, #0\n"
    "   bne 1b"
    : "=&r" (result), "=&r" (tmp)
    : "r" (lockword), "Ir" (value)
    : "cc");

    __asm__ __volatile__ ("dmb" : : : "memory");
    return result;
}

Answer 1

ThreadSanitizer 不知道您自己开发的 atomic_swap() 或 barrier() 函数的语义。来自ThreadSanitizer FAQ :

Q: What synchronization primitives are supported? TSan supports pthread synchronization primitives, built-in compiler atomic operations (sync/atomic), C++ operations are supported with llvm libc++ (not very throughly [sic] tested, though).

所以它不知道 atomic_swap() 应该做什么，除此之外 ThreadSanitizer currently doesn't support memory fences .