Re: [PATCH] sched/wait: Fix the signal handling fix

[Oleg Nesterov <oleg@redhat.com>]

Peter, sorry. I didn't actually read this patch yet (and a lot of previous
emails). Will try tomorrow, I am not even sure I understand the problem(s)
correctly. But let me ask one question anyway,

On 12/13, Peter Zijlstra wrote:
>
> --- a/kernel/sched/wait.c
> +++ b/kernel/sched/wait.c
> @@ -392,7 +392,7 @@ __wait_on_bit(wait_queue_head_t *wq, str
>  	do {
>  		prepare_to_wait(wq, &q->wait, mode);
>  		if (test_bit(q->key.bit_nr, q->key.flags))
> -			ret = (*action)(&q->key);
> +			ret = (*action)(&q->key, mode);

And every action() should check signal_pending_state()...

So why we can't change __wait_on_bit/etc instead and remove all the signal-
pending checks from the callbacks? It seems that we can just check
signal_pending_state() before prepare_to_wait(). Or perhaps we can add
another helper which acts like prepare_to_wait_event().

Yes, some callers want -EINTR, some -ERESTARTSYS, but this shouldn't be a
problem.

And sorry if this was already discussed, another case when I am trying to
return to lkml with a lot of unread emails.

Oleg.