Re: [RFC PATCH 2/4] arm64: use IPIs to pause/resume remote CPUs

[David Hildenbrand <david@redhat.com>]

[...]

> +
> +void resume_remote_cpus(void)
> +{
> +	cpumask_t cpus_to_resume;
> +
> +	lockdep_assert_cpus_held();
> +	lockdep_assert_preemption_disabled();
> +
> +	cpumask_copy(&cpus_to_resume, cpu_online_mask);
> +	cpumask_clear_cpu(smp_processor_id(), &cpus_to_resume);
> +
> +	spin_lock(&cpu_pause_lock);
> +
> +	cpumask_setall(&resumed_cpus);
> +	/* A typical example for sleep and wake-up functions. */
> +	smp_mb();
> +	while (cpumask_intersects(&cpus_to_resume, &paused_cpus)) {
> +		sev();
> +		cpu_relax();
> +		barrier();
> +	}
>

I'm curious, is there a fundamental reason why we wait for paused CPUs 
to actually start running, or is it simply easier to get the 
implementation race-free, in particular when we have two 
pause_remote_cpus() calls shortly after each other and another remote 
CPU might still be on its way out of pause_local_cpu() from the first pause.

-- 
Cheers,

David / dhildenb