Skip to contentSkip to conversation

iouring: Fix min_timeout behaviour

PATCH Applied Christian A. Ehrhardt 2 participants 4 messages Jun 6 – Jun 7 io_uring

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 iouring: Fix min_timeout behaviour

The wakeup condition if a min timeout is present and has expired is that at least _one_ CQE was posted. Thus set the cq_tail target to ->cq_min_tail + 1. Without this commit a spurious wakeup can result in a premature wakeup because io_should_wake() will return true even if _no_ CQE was posted at all.

Tested by running the liburing testsuite with no regressions.

Additionally, tested by turning all calls to schedule() in io_uring/wait.c into calls to schedule_timeout(1) to force the spurious wakeups. With these spurious wakeups the min-timeout.t test fails before and passes after this commit.

Cc Jens Axboe <[email protected]>
Cc Tip ten Brink <[email protected]>
Fixes e15cb2200b93 ("io_uring: fix min_wait wakeups for SQPOLL")
---
 io_uring/wait.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
io_uring/wait.c +1 -1
index ec01e78a216d..d005ea17b35f 100644
@@ -103,7 +103,7 @@ static enum hrtimer_restart io_cqring_min_timer_wakeup(struct hrtimer *timer)
 	}
 
 	/* any generated CQE posted past this time should wake us up */
-	iowq->cq_tail = iowq->cq_min_tail;
+	iowq->cq_tail = iowq->cq_min_tail + 1;
 
 	hrtimer_update_function(&iowq->t, io_cqring_timer_wakeup);
 	hrtimer_set_expires(timer, iowq->timeout);
Signature · 1 line
2.43.0
maintainer
#1 lore raw
Quoted 7 lines · Christian A. Ehrhardt

The wakeup condition if a min timeout is present and has expired is that at least _one_ CQE was posted. Thus set the cq_tail target to ->cq_min_tail + 1. Without this commit a spurious wakeup can result in a premature wakeup because io_should_wake() will return true even if _no_ CQE was posted at all.

Quoted 6 lines

Tested by running the liburing testsuite with no regressions.

Additionally, tested by turning all calls to schedule() in io_uring/wait.c into calls to schedule_timeout(1) to force the spurious wakeups. With these spurious wakeups the min-timeout.t test fails before and passes after this commit.

Either this or the test case is broken, with or without the change you sent for the test case. I'll take a look, but it's definitely not passing as-is.

Signature · 1 line
Jens Axboe
#2 lore raw
Quoted 10 lines · Jens Axboe
Quoted 9 lines · Christian A. Ehrhardt

The wakeup condition if a min timeout is present and has expired is that at least _one_ CQE was posted. Thus set the cq_tail target to ->cq_min_tail + 1. Without this commit a spurious wakeup can result in a premature wakeup because io_should_wake() will return true even if _no_ CQE was posted at all.

Tested by running the liburing testsuite with no regressions.

Quoted 8 lines
Quoted 4 lines

Additionally, tested by turning all calls to schedule() in io_uring/wait.c into calls to schedule_timeout(1) to force the spurious wakeups. With these spurious wakeups the min-timeout.t test fails before and passes after this commit.

Either this or the test case is broken, with or without the change you sent for the test case. I'll take a look, but it's definitely not passing as-is.

I also tested with the zig reproducer from

	https://github.com/axboe/liburing/issues/1477

and with the spurious wakeups the reproducer shows the premature wakeup without any CQE posted, too. It seems that the missing "+1" is an oversight that got introduced between v1 and v2 of the commit that fixed the above issue.

Best regards,
Christian

maintainer Applied
#3 lore raw
In reply to Christian A. Ehrhardt
Quoted 8 lines · Christian A. Ehrhardt

The wakeup condition if a min timeout is present and has expired is that at least _one_ CQE was posted. Thus set the cq_tail target to ->cq_min_tail + 1. Without this commit a spurious wakeup can result in a premature wakeup because io_should_wake() will return true even if _no_ CQE was posted at all.

[...]

Applied, thanks!

[1/1] iouring: Fix min_timeout behaviour

      commit: 29fe1bd01b99714f3136f922230a643c2742cda9

Best regards,

Signature · 1 line
Jens Axboe