Re: [PATCH-next v9 03/11] cgroup/cpuset: Prevent race between task attach and cpuset state change

[Waiman Long]

On 6/30/26 9:41 PM, Ridong Chen wrote:
> > @@ -3246,7 +3268,12 @@ ssize_t cpuset_write_resmask(struct 
> > kernfs_open_file *of,
> >           return -EACCES;
> >         buf = strstrip(buf);
> > -    cpuset_full_lock();
> > +
> > +    /* cpuset_mutex acquired in wait_attach_done_lock() */
> > +    mutex_lock(&cpuset_top_mutex);
> > +    cpus_read_lock();
> > +    wait_attach_done_lock();
> > +
>
> Would it be cleaner to just pass this into cpuset_full_lock() as a flag?
>
> void cpuset_full_lock(bool wait_attach)
> {
>       mutex_lock(&cpuset_top_mutex);
>       cpus_read_lock();
>       if (wait_attach)
>               wait_attach_done_lock();
>       else
>               mutex_lock(&cpuset_mutex);
> }
> Then the two write paths become a single cpuset_full_lock(true). The
> downside is the other 6 callers would need cpuset_full_lock(false). 
> Not sure it's worth it — what do you think?

I would like to put wait_attach_done_lock() at places where it will wait 
for the completion of the attach operation. Putting it inside 
cpuset_full_lock() will make it harder to find out which operations will 
need to wait for task attach. So I will keep the code as is.

Cheers,
Longman