Re: [PATCH v3 10/19] sched: Fix migrate_disable() vs set_cpus_allowed_ptr()

[Peter Zijlstra]

On Thu, Oct 15, 2020 at 02:54:53PM +0100, Valentin Schneider wrote:
> 
> On 15/10/20 12:05, Peter Zijlstra wrote:
> > +static int affine_move_task(struct rq *rq, struct rq_flags *rf,
> > +			    struct task_struct *p, int dest_cpu, unsigned int flags)
> > +{
> > +	struct set_affinity_pending my_pending = { }, *pending = NULL;
> > +	struct migration_arg arg = {
> > +		.task = p,
> > +		.dest_cpu = dest_cpu,
> > +	};
> > +	bool complete = false;
> > +
> > +	/* Can the task run on the task's current CPU? If so, we're done */
> > +	if (cpumask_test_cpu(task_cpu(p), &p->cpus_mask)) {
> > +		pending = p->migration_pending;
> > +		if (pending) {
> > +			p->migration_pending = NULL;
> > +			complete = true;
> 
> Deciphering my TLA+ deadlock traces leads me to think this needs
> 
>                         refcount_inc(&pending->refs);
> 
> because the 'goto do_complete' leads us to an unconditional decrement.

Hurmm. I think you're right. I've updated the patch.