On 07/15, Josef Bacik wrote:
>
> Oleg noticed that our checking of data.got_token is unsafe in the
> cleanup case, and should really use a memory barrier. Use the
> READ_ONCE/WRITE_ONCE helpers on got_token so we can be sure we're always
> safe.
READ/WRITE_ONCE can't help, both are compiler barriers. You need smp_wmb/rmb.
Alternatively,
> @@ -246,7 +246,7 @@ void rq_qos_wait(struct rq_wait *rqw, void *private_data,
> prepare_to_wait_exclusive(&rqw->wait, &data.wq, TASK_UNINTERRUPTIBLE);
> has_sleeper = !wq_has_single_sleeper(&rqw->wait);
> do {
> - if (data.got_token)
> + if (READ_ONCE(data.got_token))
> break;
> if (!has_sleeper && acquire_inflight_cb(rqw, private_data)) {
> finish_wait(&rqw->wait, &data.wq);
You can use remove_wait_queue() which takes rqw->wait->lock unconditonally,
but then you will need to do __set_current_state(TASK_RUNNING) and use
"return" instead of break.
Oleg.
