Skip to contentSkip to conversation

Only set epoll timeout if it's in the future

RFC Superseded Joe Damato 3 participants 4 messages Apr 15, 2025 – Apr 16, 2025 VFS

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Joe Damato

author
#0 lore raw

cover Only set epoll timeout if it's in the future

Greetings:

Sending as an RFC against vfs/for-next because I am not sure which branch to send this against.

It's possible this should be against vfs.fixes with a Fixes bf3b9f6372c4 ("epoll: Add busy poll support to epoll with socket fds."), but I am not sure.

The commit message explains the issue in detail, please see the commit message for details.

If this patch is appropriate, please let me know:
- which branch to generate the patch against (I suspect vfs/for-next is wrong),
- whether to include the fixes tag mentioned above or if this is considered "new code" instead of a bug fix and I'll omit the fixes.

I'll re-send as instructed (without this cover letter)... or not if this patch is incorrect/not desirable :)

Thanks,
Joe

Joe Damato (1):
eventpoll: Set epoll timeout if it's in the future

 fs/eventpoll.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

base-commit: 2e72b1e0aac24a12f3bf3eec620efaca7ab7d4de

Signature · 1 line
2.43.0
author
#1 lore raw

RFC 1/1 eventpoll: Set epoll timeout if it's in the future

Avoid an edge case where epoll_wait arms a timer and calls schedule() even if the timer will expire immediately.

For example: if the user has specified an epoll busy poll usecs which is equal or larger than the epoll_wait/epoll_pwait2 timeout, it is unnecessary to call schedule_hrtimeout_range; the busy poll usecs have consumed the entire timeout duration so it is unnecessary to induce scheduling latency by calling schedule() (via schedule_hrtimeout_range).

This can be measured using a simple bpftrace script:

tracepoint:sched:sched_switch
/ args->prev_pid == $1 /
{
print(kstack());
print(ustack());
}

Before this patch is applied:

Testing an epoll_wait app with busy poll usecs set to 1000, and epoll_wait timeout set to 1ms using the script above shows:

     __traceiter_sched_switch+69
     __schedule+1495
     schedule+32
     schedule_hrtimeout_range+159
     do_epoll_wait+1424
     __x64_sys_epoll_wait+97
     do_syscall_64+95
     entry_SYSCALL_64_after_hwframe+118

     epoll_wait+82

Which is unexpected; the busy poll usecs should have consumed the entire timeout and there should be no reason to arm a timer.

After this patch is applied: the same test scenario does not generate a call to schedule() in the above edge case. If the busy poll usecs are reduced (for example usecs: 100, epoll_wait timeout 1ms) the timer is armed as expected.

---
 fs/eventpoll.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
fs/eventpoll.c +9 -1
index f9898e60dd8b..ca0c7e843da7 100644
@@ -1980,6 +1980,14 @@ static int ep_autoremove_wake_function(struct wait_queue_entry *wq_entry,
 	return ret;
 }
 
+static int ep_schedule_timeout(ktime_t *to)
+{
+	if (to)
+		return ktime_after(*to, ktime_get());
+	else
+		return 1;
+}
+
 /**
  * ep_poll - Retrieves ready events, and delivers them to the caller-supplied
  *           event buffer.
@@ -2095,7 +2103,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
 
 		write_unlock_irq(&ep->lock);
 
-		if (!eavail)
+		if (!eavail && ep_schedule_timeout(to))
 			timed_out = !schedule_hrtimeout_range(to, slack,
 							      HRTIMER_MODE_ABS);
 		__set_current_state(TASK_RUNNING);
Signature · 1 line
2.43.0
reviewer Reviewed-by on v1 1/1
#2 lore raw

Re: [RFC vfs/for-next 1/1] eventpoll: Set epoll timeout if it's in the future

Quoted 37 lines · Joe Damato

Avoid an edge case where epoll_wait arms a timer and calls schedule() even if the timer will expire immediately.

For example: if the user has specified an epoll busy poll usecs which is equal or larger than the epoll_wait/epoll_pwait2 timeout, it is unnecessary to call schedule_hrtimeout_range; the busy poll usecs have consumed the entire timeout duration so it is unnecessary to induce scheduling latency by calling schedule() (via schedule_hrtimeout_range).

This can be measured using a simple bpftrace script:

tracepoint:sched:sched_switch
/ args->prev_pid == $1 /
{
print(kstack());
print(ustack());
}

Before this patch is applied:

Testing an epoll_wait app with busy poll usecs set to 1000, and epoll_wait timeout set to 1ms using the script above shows:

     __traceiter_sched_switch+69
     __schedule+1495
     schedule+32
     schedule_hrtimeout_range+159
     do_epoll_wait+1424
     __x64_sys_epoll_wait+97
     do_syscall_64+95
     entry_SYSCALL_64_after_hwframe+118

     epoll_wait+82

Which is unexpected; the busy poll usecs should have consumed the entire timeout and there should be no reason to arm a timer.

Quoted 6 lines

After this patch is applied: the same test scenario does not generate a call to schedule() in the above edge case. If the busy poll usecs are reduced (for example usecs: 100, epoll_wait timeout 1ms) the timer is armed as expected.

Hum, I guess this is about the interpretation of the 'timeout' value of epoll_pwait2() and friends. Does the runtime of the system call itself (including possible polling) count into the timeout or does timeout mean how long we should sleep after we've done all our work? The manpage says "The timeout argument specifies the number of milliseconds that epoll_wait() will block." which I guess can be understood both ways. Seeing the epoll code it seems the author's intention was indeed rather the former. So I guess feel free to add:

Honza

Quoted 35 lines
---
 fs/eventpoll.c | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)
fs/eventpoll.c +9 -1
index f9898e60dd8b..ca0c7e843da7 100644
@@ -1980,6 +1980,14 @@ static int ep_autoremove_wake_function(struct wait_queue_entry *wq_entry,
 	return ret;
 }
 
+static int ep_schedule_timeout(ktime_t *to)
+{
+	if (to)
+		return ktime_after(*to, ktime_get());
+	else
+		return 1;
+}
+
 /**
  * ep_poll - Retrieves ready events, and delivers them to the caller-supplied
  *           event buffer.
@@ -2095,7 +2103,7 @@ static int ep_poll(struct eventpoll *ep, struct epoll_event __user *events,
 
 		write_unlock_irq(&ep->lock);
 
-		if (!eavail)
+		if (!eavail && ep_schedule_timeout(to))
 			timed_out = !schedule_hrtimeout_range(to, slack,
 							      HRTIMER_MODE_ABS);
 		__set_current_state(TASK_RUNNING);
Signature · 1 line
2.43.0
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#3 lore raw
In reply to Joe Damato
Quoted 6 lines · Joe Damato

Greetings:

Sending as an RFC against vfs/for-next because I am not sure which branch to send this against.

It's possible this should be against vfs.fixes with a Fixes

Yes, can you resend against vfs.fixes, please? :)