Re: [PATCH 2/2] sched_ext/idle: Make scx_bpf_select_cpu_and() usable from any context

[Andrea Righi <arighi@nvidia.com>]

Hi Tejun,

On Mon, May 12, 2025 at 07:19:53AM -1000, Tejun Heo wrote:
> Hello,
> 
> On Mon, May 12, 2025 at 05:14:56PM +0200, Andrea Righi wrote:
> >  #ifdef CONFIG_SMP
> > +	/*
> > +	 * If called from an unlocked context, try to acquire
> > +	 * cpus_read_lock() to avoid races with CPU hotplug.
> > +	 */
> > +	if (scx_kf_allowed_if_unlocked())
> > +		if (!cpus_read_trylock())
> > +			return -EBUSY;
> 
> Is this meaningful? The idle CPU selection is already racy against CPU
> hotplugs and we depend on the scheduler core to fix it up afterwards. Even
> if scx_bpf_select_cpu_and() is not racy, it will drop the cpus lock before
> returning and becomes racy again right there. ie. This doesn't add any
> meaningful protection.

I was concerned that accessing llc_span() / llc_weight() from
scx_select_cpu_dfl() might be problematic if a CPU goes offline underneath,
but we are accessing them with rcu_read_lock() held, we probably don't need
cpus_read_lock() protection.

And about the scheduler picking a bad CPU, that can be fixed by the
sched_ext core, so there's no problem for that.

I'll think more about the CPU hotplugging scenario.

Thanks,
-Andrea

> 
> Thanks.
> 
> -- 
> tejun