Re: [PATCH] memcg: decouple memcg_hotplug_cpu_dead from stock_lock

[Shakeel Butt <shakeel.butt@linux.dev>]

On Fri, Apr 11, 2025 at 02:12:46PM -0400, Shakeel Butt wrote:
> On Fri, Apr 11, 2025 at 2:06 PM Vlastimil Babka <vbabka@suse.cz> wrote:
> >
> > On 4/11/25 19:54, Shakeel Butt wrote:
> > > (my migadu/linux.dev stopped working and I have to send through gmail,
> > > sorry for any formatting issue)
> > >
> > > I don't see how local_irq_save() will break anything. We are working on
> > > a stock of a dead remote cpu. We actually don't even need to disable irq
> > > or need local cpu's local_lock. It is actually the calls to
> > > __mod_memcg_lruvec_state() and __mod_memcg_state() in
> > > __drain_obj_stock() which need irq-disabled on non-RT kernels and for
> > > RT-kernels they already have preempt_disable_nested().
> > >
> > > Disabling irq even on RT seems excessive but this is not a performance
> > > critical code, so I don't see an issue unless there is
> > > local_lock_irqsave() alternative which does not disables irqs on RT
> > > kernels.
> >
> > local_lock_irqsave() does not disable irqs on RT kernels :)
> 
> Sorry, I wanted to say local_irq_save() here instead of local_lock_irqsave().
> 
> > so keeping
> > local_lock as is would do the irq disable on !RT and be more RT-friendly on
> > RT. It's just wrong from the logical scope of the lock to perform it on a
> > different cpu than the stock we modify. If one day we have some runtime
> > checks for that, they would complain.
> 
> Basically I don't want to use stock_lock here. Maybe I should explore
> adding a new local_lock for __mod_memcg_lruvec_state and
> __mod_memcg_state.

Vlastimil & Sebastian, if you don't have a strong opinion/push-back on
this patch then I will keep it as is. However I am planning to rework
the memcg stats (& vmstats) to see if I can use dedicated local_lock for
them and able to modify them in any context.