Re: [PATCH] sched_ext: Provide a sysfs enable_seq counter

[Andrea Righi <andrea.righi@linux.dev>]

On Mon, Sep 23, 2024 at 12:45:48PM +0200, Phil Auld wrote:
> 
> Hi Tejun,
> 
> On Sun, Sep 22, 2024 at 11:21:02AM -1000 Tejun Heo wrote:
> > Hello, Andrea.
> > 
> > On Sat, Sep 21, 2024 at 09:39:21PM +0200, andrea.righi@linux.dev wrote:
> > >  static struct attribute *scx_global_attrs[] = {
> > >  	&scx_attr_state.attr,
> > >  	&scx_attr_switch_all.attr,
> > >  	&scx_attr_nr_rejected.attr,
> > >  	&scx_attr_hotplug_seq.attr,
> > > +	&scx_attr_enable_seq.attr,
> > >  	NULL,
> > >  };
> > 
> > Can you put this in scx_sched_attrs instead as it probably would make sense
> > to track this per-scheduler in the future when we support stacked
> > schedulers.
> 
> It's not a per scheduler counter, though. It's global. We want to know
> that a (any) scx scheduler has been loaded at some time in the past. It's
> really only interesting when 0 or > 0. The actual non-zero number and which
> scheduler(s) don't matter that much.
> 
> And it needs to persist when the scheduler is unloaded (I didn't look but
> I uspect the per scheduler attrs come and go?).

Correct, if we make the counter per-scheduler we would lose this
information once the running scheduler is unloaded.

Instead we want to maintain this information persistent, so that user
support can clearly see if any of the BPF scheduler has ever been used
since boot.

Thanks,
-Andrea