Skip to contentSkip to conversation

sched: fix warning in sched_setaffinity

PATCH In review Josh Don 4 participants 8 messages Aug 29, 2024 – Oct 25, 2024 Scheduler

lore
Share & tools
lore

Josh Don

author
#0 lore raw

PATCH 1/1 sched: fix warning in sched_setaffinity

Commit 8f9ea86fdf99b added some logic to sched_setaffinity that included a WARN when a per-task affinity assignment races with a cpuset update.

Specifically, we can have a race where a cpuset update results in the task affinity no longer being a subset of the cpuset. That's fine; we have a fallback to instead use the cpuset mask. However, we have a WARN set up that will trigger if the cpuset mask has no overlap at all with the requested task affinity. This shouldn't be a warning condition; its trivial to create this condition.

Reproduced the warning by the following setup:

- $PID inside a cpuset cgroup
- another thread repeatedly switching the cpuset cpus from 1-2 to just 1
- another thread repeatedly setting the $PID affinity (via taskset) to 2

Fixes 8f9ea86fdf99b ("sched: Always preserve the user requested cpumask")
---
 kernel/sched/syscalls.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/sched/syscalls.c +1 -1
index 4fae3cf25a3a..3a88f7c0cb69 100644
@@ -1321,7 +1321,7 @@ int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx)
 			bool empty = !cpumask_and(new_mask, new_mask,
 						  ctx->user_mask);
 
-			if (WARN_ON_ONCE(empty))
+			if (empty)
 				cpumask_copy(new_mask, cpus_allowed);
 		}
 		__set_cpus_allowed_ptr(p, ctx);
Signature · 1 line
2.46.0.469.g59c65b2a67-goog
Acked-by on v1 1/1
#1 lore raw
Quoted 27 lines · Josh Don

Commit 8f9ea86fdf99b added some logic to sched_setaffinity that included a WARN when a per-task affinity assignment races with a cpuset update.

Specifically, we can have a race where a cpuset update results in the task affinity no longer being a subset of the cpuset. That's fine; we have a fallback to instead use the cpuset mask. However, we have a WARN set up that will trigger if the cpuset mask has no overlap at all with the requested task affinity. This shouldn't be a warning condition; its trivial to create this condition.

Reproduced the warning by the following setup:

- $PID inside a cpuset cgroup
- another thread repeatedly switching the cpuset cpus from 1-2 to just 1
- another thread repeatedly setting the $PID affinity (via taskset) to 2

Fixes 8f9ea86fdf99b ("sched: Always preserve the user requested cpumask")
---
  kernel/sched/syscalls.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
kernel/sched/syscalls.c +0 -0
index 4fae3cf25a3a..3a88f7c0cb69 100644

kernel/sched/syscalls.c int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx) · line 1327

Quoted 8 lines
kernel/sched/syscalls.c +1 -1
@@ -1321,7 +1321,7 @@ int __sched_setaffinity(struct task_struct *p, struct affinity_context *ctx)
  			bool empty = !cpumask_and(new_mask, new_mask,
  						  ctx->user_mask);
 
-			if (WARN_ON_ONCE(empty))
+			if (empty)
  				cpumask_copy(new_mask, cpus_allowed);
  		}
  		__set_cpus_allowed_ptr(p, ctx);

Taking out the WARN_ON_ONCE() should be fine.

#2 lore raw
In reply to Josh Don

Hi Josh Don,

Quoted 10 lines · Josh Don

Commit 8f9ea86fdf99b added some logic to sched_setaffinity that included a WARN when a per-task affinity assignment races with a cpuset update.

Specifically, we can have a race where a cpuset update results in the task affinity no longer being a subset of the cpuset. That's fine; we have a fallback to instead use the cpuset mask. However, we have a WARN set up that will trigger if the cpuset mask has no overlap at all with the requested task affinity. This shouldn't be a warning condition; its trivial to create this condition.

Quoted 6 lines

Reproduced the warning by the following setup:

- $PID inside a cpuset cgroup
- another thread repeatedly switching the cpuset cpus from 1-2 to just 1
- another thread repeatedly setting the $PID affinity (via taskset) to 2

I was testing the patch using the following two scripts run concurrently:

Script 1:
while true; do

    echo 1 > /sys/fs/cgroup/test_group/cpuset.cpus;
    echo 1-2 > /sys/fs/cgroup/test_group/cpuset.cpus;

done

Script 2:
while true; do

    sudo taskset -p 0x2 $$;

done

However, I am unable to trigger the warning in dmesg on the unpatched kernel. I was expecting to see the warning as described, but it doesn't seem to appear.

Additionally, I also tried the following script to increase the chances of triggering the race condition:

while true; do

    echo 1 > /sys/fs/cgroup/test_group/cpuset.cpus;
    sudo taskset -p 0x2 $$;
    sleep 0.1;
    echo 1-2 > /sys/fs/cgroup/test_group/cpuset.cpus;

done

Despite this, the warning still does not appear in dmesg.

Am I missing something in my testing approach, or is there a different setup required to reproduce the issue?

Thanks and Regards
Madadi Vineeth Reddy

Quoted 2 lines
Fixes 8f9ea86fdf99b ("sched: Always preserve the user requested cpumask")
author
#3 lore raw

Hi Madadi,

Quoted 45 lines · Madadi Vineeth Reddy

Hi Josh Don,

Quoted 10 lines · Josh Don

Commit 8f9ea86fdf99b added some logic to sched_setaffinity that included a WARN when a per-task affinity assignment races with a cpuset update.

Specifically, we can have a race where a cpuset update results in the task affinity no longer being a subset of the cpuset. That's fine; we have a fallback to instead use the cpuset mask. However, we have a WARN set up that will trigger if the cpuset mask has no overlap at all with the requested task affinity. This shouldn't be a warning condition; its trivial to create this condition.

Quoted 6 lines

Reproduced the warning by the following setup:

- $PID inside a cpuset cgroup
- another thread repeatedly switching the cpuset cpus from 1-2 to just 1
- another thread repeatedly setting the $PID affinity (via taskset) to 2

I was testing the patch using the following two scripts run concurrently:

Script 1:
while true; do

    echo 1 > /sys/fs/cgroup/test_group/cpuset.cpus;
    echo 1-2 > /sys/fs/cgroup/test_group/cpuset.cpus;

done

Script 2:
while true; do

    sudo taskset -p 0x2 $$;

done

However, I am unable to trigger the warning in dmesg on the unpatched kernel. I was expecting to see the warning as described, but it doesn't seem to appear.

Additionally, I also tried the following script to increase the chances of triggering the race condition:

while true; do

    echo 1 > /sys/fs/cgroup/test_group/cpuset.cpus;
    sudo taskset -p 0x2 $$;
    sleep 0.1;
    echo 1-2 > /sys/fs/cgroup/test_group/cpuset.cpus;
Quoted 6 lines

done

Despite this, the warning still does not appear in dmesg.

Am I missing something in my testing approach, or is there a different setup required to reproduce the issue?

taskset -p 0x2 $$ will affine to cpu 1 :)

I'd recommend using the '-c' arg to specify the mask as a cpulist, as it is easier to validate.

taskset -c -p 2 $$

Quoted 7 lines

Thanks and Regards
Madadi Vineeth Reddy

Quoted 2 lines
Fixes 8f9ea86fdf99b ("sched: Always preserve the user requested cpumask")

Best,
Josh

#4 lore raw

Hi Josh Don,

Quoted 56 lines · Josh Don

Hi Madadi,

Quoted 45 lines · Madadi Vineeth Reddy

Hi Josh Don,

Quoted 10 lines · Josh Don

Commit 8f9ea86fdf99b added some logic to sched_setaffinity that included a WARN when a per-task affinity assignment races with a cpuset update.

Specifically, we can have a race where a cpuset update results in the task affinity no longer being a subset of the cpuset. That's fine; we have a fallback to instead use the cpuset mask. However, we have a WARN set up that will trigger if the cpuset mask has no overlap at all with the requested task affinity. This shouldn't be a warning condition; its trivial to create this condition.

Quoted 6 lines

Reproduced the warning by the following setup:

- $PID inside a cpuset cgroup
- another thread repeatedly switching the cpuset cpus from 1-2 to just 1
- another thread repeatedly setting the $PID affinity (via taskset) to 2

I was testing the patch using the following two scripts run concurrently:

Script 1:
while true; do

    echo 1 > /sys/fs/cgroup/test_group/cpuset.cpus;
    echo 1-2 > /sys/fs/cgroup/test_group/cpuset.cpus;

done

Script 2:
while true; do

    sudo taskset -p 0x2 $$;

done

However, I am unable to trigger the warning in dmesg on the unpatched kernel. I was expecting to see the warning as described, but it doesn't seem to appear.

Additionally, I also tried the following script to increase the chances of triggering the race condition:

while true; do

    echo 1 > /sys/fs/cgroup/test_group/cpuset.cpus;
    sudo taskset -p 0x2 $$;
    sleep 0.1;
    echo 1-2 > /sys/fs/cgroup/test_group/cpuset.cpus;
Quoted 6 lines

done

Despite this, the warning still does not appear in dmesg.

Am I missing something in my testing approach, or is there a different setup required to reproduce the issue?

Quoted 7 lines

taskset -p 0x2 $$ will affine to cpu 1 :)

I'd recommend using the '-c' arg to specify the mask as a cpulist, as it is easier to validate.

taskset -c -p 2 $$

Thanks for the clarification.

Thanks and Regards
Madadi Vineeth Reddy

Quoted 10 lines
Quoted 7 lines

Thanks and Regards
Madadi Vineeth Reddy

Quoted 2 lines
Fixes 8f9ea86fdf99b ("sched: Always preserve the user requested cpumask")

Best,
Josh

author Acked-by Waiman Long Tested-by Madadi Vineeth Reddy on v1 1/1
#5 lore raw
In reply to Josh Don
Quoted 11 lines · Josh Don

Commit 8f9ea86fdf99b added some logic to sched_setaffinity that included a WARN when a per-task affinity assignment races with a cpuset update.

Specifically, we can have a race where a cpuset update results in the task affinity no longer being a subset of the cpuset. That's fine; we have a fallback to instead use the cpuset mask. However, we have a WARN set up that will trigger if the cpuset mask has no overlap at all with the requested task affinity. This shouldn't be a warning condition; its trivial to create this condition.

Quoted 8 lines

Reproduced the warning by the following setup:

- $PID inside a cpuset cgroup
- another thread repeatedly switching the cpuset cpus from 1-2 to just 1
- another thread repeatedly setting the $PID affinity (via taskset) to 2

Fixes 8f9ea86fdf99b ("sched: Always preserve the user requested cpumask")

Gentle ping to bump this in case it got lost.

I've also collected the following:

Best,
Josh

maintainer
#6 lore raw
Quoted 15 lines · Josh Don
Quoted 13 lines · Josh Don

Commit 8f9ea86fdf99b added some logic to sched_setaffinity that included a WARN when a per-task affinity assignment races with a cpuset update.

Specifically, we can have a race where a cpuset update results in the task affinity no longer being a subset of the cpuset. That's fine; we have a fallback to instead use the cpuset mask. However, we have a WARN set up that will trigger if the cpuset mask has no overlap at all with the requested task affinity. This shouldn't be a warning condition; its trivial to create this condition.

Reproduced the warning by the following setup:

Quoted 8 lines
Quoted 6 lines

- $PID inside a cpuset cgroup
- another thread repeatedly switching the cpuset cpus from 1-2 to just 1
- another thread repeatedly setting the $PID affinity (via taskset) to 2

Fixes 8f9ea86fdf99b ("sched: Always preserve the user requested cpumask")

Gentle ping to bump this in case it got lost.

I have also been able to reproduce the race and the WARN with the steps described in the commit description

Quoted 8 lines

I've also collected the following:

Best,
Josh