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

[Sven Schnelle <svens@linux.ibm.com>]

Tejun Heo <tj@kernel.org> writes:

> On Mon, Apr 15, 2024 at 07:35:49AM +0200, Sven Schnelle wrote:
>> @@ -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);
>
> I think this can still race with the last CPU in the pod going down and
> return nr_cpu_ids. Maybe something like the following would be better?
>
> 	int wake_cpu;
>
> 	wake_cpu = cpumask_any_distribute_and(...);
> 	if (wake_cpu < nr_cpus_ids) {
> 		p->wake_cpu = wake_cpu;
> 		// update stat;
> 	}
>
> This generally seems like a good idea but isn't this still racy? The CPU may
> go down between setting p->wake_cpu and wake_up_process().

Don't know without reading the source, but how does this code normally
protect against that?

Thanks
Sven