On 03/09/2018 11:46 AM, Peter Zijlstra wrote:
On Fri, Mar 09, 2018 at 12:04:18PM +0100, Sebastian Andrzej Siewior wrote:
+void swake_add_all_wq(struct swait_queue_head *q, struct wake_q_head *wq)
{
struct swait_queue *curr;
while (!list_empty(&q->task_list)) {
curr = list_first_entry(&q->task_list, typeof(*curr),
task_list);
list_del_init(&curr->task_list);
+ wake_q_add(wq, curr->task);
}
}
+EXPORT_SYMBOL(swake_add_all_wq);
void swake_up(struct swait_queue_head *q)
{
@@ -66,25 +62,14 @@ EXPORT_SYMBOL(swake_up);
*/
void swake_up_all(struct swait_queue_head *q)
{
+ unsigned long flags;
+ DEFINE_WAKE_Q(wq);
+ raw_spin_lock_irqsave(&q->lock, flags);
+ swake_add_all_wq(q, &wq);
+ raw_spin_unlock_irqrestore(&q->lock, flags);
+ wake_up_q(&wq);
}
EXPORT_SYMBOL(swake_up_all);
This is fundamentally wrong. The whole point of wake_up_all() is that
_all_ is unbounded and should not ever land in a single critical
section, be it IRQ or PREEMPT disabled. The above does both.
It seems to me to be better than what was there, certainly more efficient.
And if I understand this correctly it is unbounded when !RT, but it is
bounded
on RT.
And I'm biased, because it should fix my problem :).
Yes, wake_up_all() is crap, it is also fundamentally incompatible with
in-*irq usage. Nothing to be done about that.
So NAK on this.
So what would you suggest? At this point getting rid of all the users of
wake_up_all() from interrupt context is not really an option, though as
an eventual goal it would be good.
-corey
