Re: [PATCH] workqueue: fix selection of wake_cpu in kick_pool()

[Sven Schnelle]

Tejun Heo <tj@xxxxxxxxxx> writes:

> Hello,
>
> On Mon, Apr 15, 2024 at 07:35:49AM +0200, Sven Schnelle wrote:
>> diff --git a/kernel/workqueue.c b/kernel/workqueue.c
>> index 0066c8f6c154..d02b0c02c9e2 100644
>> --- a/kernel/workqueue.c
>> +++ b/kernel/workqueue.c
>> @@ -1277,7 +1277,8 @@ static bool kick_pool(struct worker_pool *pool)
>>  	    !cpumask_test_cpu(p->wake_cpu, pool->attrs->__pod_cpumask)) {
>>  		struct work_struct *work = list_first_entry(&pool->worklist,
>>  						struct work_struct, entry);
>> -		p->wake_cpu = cpumask_any_distribute(pool->attrs->__pod_cpumask);
>> +		p->wake_cpu = cpumask_any_and_distribute(pool->attrs->__pod_cpumask,
>> +							 cpu_online_mask);
>
> So, this wouldn't necessarily fix the problem completely but regardless of
> how that's plugged, this is still something we want to do to avoid picking
> offline CPUs. Can you please update the patch description accordingly and
> resend?

I'll just sent v2. I didn't mention the arch_vcpu_is_preempted() issue
in the commit description, as i'm not yet sure whether that's a wrong
assumption the s390 code or in the common code. Still waiting whether
Peter has some insight.