lk中的mutex_t机制

59 阅读 0 评论 39 点赞

我是靠谱客的博主明亮菠萝，最近开发中收集的这篇文章主要介绍lk中的mutex_t机制，觉得挺不错的，现在分享给大家，希望可以做个参考。

概述

lk中还实现thread对全局资源保护的机制：mutex_t。主要用于对全局变量资源保护.而event 主要用于多thread之间

实现这套机制总共就5个函数，我们一个一个看一下

mutex_x的初始化函数和event一样，也是每个mutex_t 需要时wait queue
void mutex_init(mutex_t *m)
{
#if MUTEX_CHECK
// ASSERT(m->magic != MUTEX_MAGIC);
#endif

m->magic = MUTEX_MAGIC;
m->count = 0;
m->holder = 0;
wait_queue_init(&m->wait);
}

mutex_x的获取函数如下，可见如果只是两个thread之间协调的话m->count ==1，并不会使用wait_queue，而且会记住这个mutex_x 是哪个thread持有
status_t mutex_acquire(mutex_t *m)
{
status_t ret = NO_ERROR;

if (current_thread == m->holder)
panic("mutex_acquire: thread %p (%s) tried to acquire mutex %p it already owns.n",
current_thread, current_thread->name, m);

enter_critical_section();

m->count++;
if (unlikely(m->count > 1)) {
ret = wait_queue_block(&m->wait, INFINITE_TIME);
if (ret < 0)
goto err;
}
m->holder = current_thread;

err:
exit_critical_section();

return ret;
}

如果用在一个thread之内保护资源的话mutex_acquire_timeout 和 mutex_acquire 是一样
status_t mutex_release(mutex_t *m)
{
if (current_thread != m->holder)
panic("mutex_release: thread %p (%s) tried to release mutex %p it doesn't own. owned by %p (%s)n",
current_thread, current_thread->name, m, m->holder, m->holder ? m->holder->name : "none");

enter_critical_section();

m->holder = 0;
m->count--;
if (unlikely(m->count >= 1)) {
/* release a thread */
// dprintf("releasing threadn");
wait_queue_wake_one(&m->wait, true, NO_ERROR);
}

exit_critical_section();

return NO_ERROR;
}
mutex_release的实现也很简单，将m->count--，一般情况下等于0.也就是if (unlikely(m->count >= 1))
一般不成立。

void mutex_destroy(mutex_t *m)
{
enter_critical_section();

#if MUTEX_CHECK
ASSERT(m->magic == MUTEX_MAGIC);
#endif

// if (m->holder != 0 && current_thread != m->holder)
// panic("mutex_destroy: thread %p (%s) tried to release mutex %p it doesn't own. owned by %p (%s)n",
// current_thread, current_thread->name, m, m->holder, m->holder ? m->holder->name : "none");

m->magic = 0;
m->count = 0;
wait_queue_destroy(&m->wait, true);
exit_critical_section();
}

mutex_destroy用于销毁mutex_x。个人感觉这个函数应该要先判断一下m->count-- 是否为0，如果为0的话，就没必要调用wait_queue_destroy 这个函数了

void wait_queue_destroy(wait_queue_t *wait, bool reschedule)
{
#if THREAD_CHECKS
ASSERT(wait->magic == WAIT_QUEUE_MAGIC);
ASSERT(in_critical_section());
#endif
wait_queue_wake_all(wait, reschedule, ERR_OBJECT_DESTROYED);
wait->magic = 0;
}

继续调用wait_queue_wake_all 来wakeup在等待这个mutex_x的所有thread

int wait_queue_wake_all(wait_queue_t *wait, bool reschedule, status_t wait_queue_error)
{
thread_t *t;
int ret = 0;

if (reschedule && wait->count > 0) {
current_thread->state = THREAD_READY;
insert_in_run_queue_head(current_thread);
}

/* pop all the threads off the wait queue into the run queue */
while ((t = list_remove_head_type(&wait->list, thread_t, queue_node))) {
wait->count--;

t->state = THREAD_READY;
t->wait_queue_block_ret = wait_queue_error;
t->blocking_wait_queue = NULL;

insert_in_run_queue_head(t);
ret++;
}

if (reschedule && ret > 0)
thread_resched();

return ret;
}

如果wait->count > 0的话，由于reschedule ==1.就说明多个thread在使用这个mutex_x.就先将当前thread 放到run_queue中，然后设置current_thread->state = THREAD_READY;，再将其放到run_queue中，饭后将wait list中的每一个thread 都取出来放到run_queue中，单后调用thread_resched来调用
这样run_queue就先运行wait list中的thread.