Re: [PATCH v3 5/6] fs/resctrl: Do not invoke smp_processor_id() in preemptible context

[Chen, Yu C]

Hi Reinette,

On 6/8/2026 11:10 PM, Reinette Chatre wrote:
> Hi Chenyu,
>
> On 6/8/26 4:26 AM, Chen, Yu C wrote:
> > On 6/8/2026 4:36 PM, Thomas Gleixner wrote:
> > > On Sat, Jun 06 2026 at 10:38, Chen Yu wrote:
> > > > From: Tony Luck <tony.luck@xxxxxxxxx>
> > > >
> > > > __l3_mon_event_count() and __l3_mon_event_count_sum() call
> > > > smp_processor_id() to obtain the current CPU. However, some
> > > > monitor events can be read from any CPU in task context via
> > > > mon_event_count(); in that case the calling context is
> > > > preemptible and smp_processor_id() triggers a debug warning.
> > >
> > > Is this new with this MMIO stuff or is this an existing issue? If the
> > > latter then this patch should be in front of the series and get a fixes
> > > tag. If not the change log should explain it.
> >
> > It is an existing issue. I'll adjust the sequence and provide a fixes tag.
> > Thanks!
>
>
> Could you please elaborate how this is an existing issue?
>
> At this point in implementation mon_evt::any_cpu is and can only be false for
> all events associated with resource RDT_RESOURCE_L3. This is highlighted by
> the existing splat that will be triggered by the
> 		WARN_ON_ONCE(rr->evt->any_cpu);
> that is removed by this patch.
>
> It is this series that makes it possible to read L3 cache occupancy from any CPU.
> This is only done in the following patch (6/6) that makes mon_evt::any_cpu for
> L3 cache occupancy be conditional on MMIO support enabled by this series.
>
> This patch (5/6) replaces the above mentioned splat with a subtler check since it
> is about to become possible for L3 cache occupancy events to be read from any CPU.
>
> I thus do not see this as an existing issue needing fixing but instead a
> necessary part of this series that is specifically a preparatory patch for the
> final MMIO based event enabling done in patch 6/6.

You are right. For RDT_RESOURCE_L3 monitor data reading, before MMIO 
monitor read is
introduced, any_cpu is false, and MSRs are always read on the CPU of 
that domain in the
IPI context. Let me refine the commit log for [PATCH 5/6].

thanks,
Chenyu