Re: [PATCH next] cpuset: fix warning when attaching tasks with offline CPUs

[Peter Zijlstra <peterz@infradead.org>]

On Mon, Jul 14, 2025 at 07:30:39PM +0800, Chen Ridong wrote:

> >> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> >> index f74d04429a29..5401adbdffa6 100644
> >> --- a/kernel/cgroup/cpuset.c
> >> +++ b/kernel/cgroup/cpuset.c
> >> @@ -3121,7 +3121,7 @@ static void cpuset_attach_task(struct cpuset *cs, struct task_struct *task)
> >>  	if (cs != &top_cpuset)
> >>  		guarantee_active_cpus(task, cpus_attach);
> >>  	else
> >> -		cpumask_andnot(cpus_attach, task_cpu_possible_mask(task),
> >> +		cpumask_andnot(cpus_attach, cpu_active_mask,
> >>  			       subpartitions_cpus);
> > 
> > This breaks things. Any task mask must be a subset of
> > task_cpu_possible_mask() at all times. It might not be able to run
> > outside of that mask.
> 
> Hi Peter,
> 
> Thanks for your feedback. I'm afraid I don't fully understand what you
> mean by "breaks things".  Could you please explain in more detail?
> 
> To clarify my current understanding: this patch simply changes the
> cpus_attach initialization from task_cpu_possible_mask(task) to
> cpu_active_mask. The intention is that when CPUs are offlined and
> tasks get migrated to root cpuset, we shouldn't try to migrate tasks
> to offline CPUs. And since cpu_active_mask is a subset of
> cpu_possible_mask, I thought this would be safe. Did I miss anything?

task_cpu_possible_mask() is the mask a task *MUST* stay inside of.

Specifically, this was introduced for ARMv9 where some CPUs drop the
capability to run ARM32 instructions. Trying to schedule an ARM32 task
on a CPU that does not support that instruction set is an immediate and
fatal fail.

Your change results in in something akin to:

  set_cpus_allowed_task(task, cpu_active_mask & ~subpartition_cpus);

Which does not honor the task_cpu_possible_mask() constraint.