Re: [PATCH v3 13/24] thermal: intel: hfi: Store per-CPU IPCC scores

[Ricardo Neri]

On Mon, Mar 27, 2023 at 06:37:32PM +0200, Rafael J. Wysocki wrote:
> On Tue, Feb 7, 2023 at 6:02 AM Ricardo Neri
> <ricardo.neri-calderon@xxxxxxxxxxxxxxx> wrote:
> >
> > The scheduler reads the IPCC scores when balancing load. These reads can
> > be quite frequent. Hardware can also update the HFI table frequently.
> > Concurrent access may cause a lot of lock contention. It gets worse as the
> > number of CPUs increases.
> >
> > Instead, create separate per-CPU IPCC scores that the scheduler can read
> > without the HFI table lock.
> >
> > Cc: Ben Segall <bsegall@xxxxxxxxxx>
> > Cc: Daniel Bristot de Oliveira <bristot@xxxxxxxxxx>
> > Cc: Dietmar Eggemann <dietmar.eggemann@xxxxxxx>
> > Cc: Ionela Voinescu <ionela.voinescu@xxxxxxx>
> > Cc: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
> > Cc: Len Brown <len.brown@xxxxxxxxx>
> > Cc: Lukasz Luba <lukasz.luba@xxxxxxx>
> > Cc: Mel Gorman <mgorman@xxxxxxx>
> > Cc: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> > Cc: Srinivas Pandruvada <srinivas.pandruvada@xxxxxxxxxxxxxxx>
> > Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
> > Cc: Tim C. Chen <tim.c.chen@xxxxxxxxx>
> > Cc: Valentin Schneider <vschneid@xxxxxxxxxx>
> > Cc: x86@xxxxxxxxxx
> > Cc: linux-pm@xxxxxxxxxxxxxxx
> > Cc: linux-kernel@xxxxxxxxxxxxxxx
> > Suggested-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> > Signed-off-by: Ricardo Neri <ricardo.neri-calderon@xxxxxxxxxxxxxxx>
> > ---
> > Changes since v2:
> >  * Only create these per-CPU variables when Intel Thread Director is
> >    supported.
> >
> > Changes since v1:
> >  * Added this patch.
> > ---
> >  drivers/thermal/intel/intel_hfi.c | 46 +++++++++++++++++++++++++++++++
> >  1 file changed, 46 insertions(+)
> >
> > diff --git a/drivers/thermal/intel/intel_hfi.c b/drivers/thermal/intel/intel_hfi.c
> > index 2527ae3836c7..b06021828892 100644
> > --- a/drivers/thermal/intel/intel_hfi.c
> > +++ b/drivers/thermal/intel/intel_hfi.c
> > @@ -29,6 +29,7 @@
> >  #include <linux/kernel.h>
> >  #include <linux/math.h>
> >  #include <linux/mutex.h>
> > +#include <linux/percpu.h>
> >  #include <linux/percpu-defs.h>
> >  #include <linux/printk.h>
> >  #include <linux/processor.h>
> > @@ -170,6 +171,43 @@ static struct workqueue_struct *hfi_updates_wq;
> >  #define HFI_UPDATE_INTERVAL            HZ
> >  #define HFI_MAX_THERM_NOTIFY_COUNT     16
> >
> > +#ifdef CONFIG_IPC_CLASSES
> 
> It would be good to provide a (concise) description of this variable.
> 
> > +static int __percpu *hfi_ipcc_scores;

Do you mean hfi_ipcc_scores or CONFIG_IPC_CLASSES?

> > +
> > +static int alloc_hfi_ipcc_scores(void)
> > +{
> > +       if (!cpu_feature_enabled(X86_FEATURE_ITD))
> > +               return 0;
> > +
> > +       hfi_ipcc_scores = __alloc_percpu(sizeof(*hfi_ipcc_scores) *
> > +                                        hfi_features.nr_classes,
> > +                                        sizeof(*hfi_ipcc_scores));
> > +
> > +       return !hfi_ipcc_scores;
> 
> I would do
> 
> if (!hfi_ipcc_scores)
>         return -ENOMEM;
> 
> return 0;
> 
> Or make the function return bool.

Sure, I can make this function return -ENOMEM.

> 
> > +}
> > +
> > +static void set_hfi_ipcc_score(void *caps, int cpu)
> > +{
> > +       int i, *hfi_class;
> > +
> > +       if (!cpu_feature_enabled(X86_FEATURE_ITD))
> > +               return;
> > +
> > +       hfi_class = per_cpu_ptr(hfi_ipcc_scores, cpu);
> > +
> > +       for (i = 0;  i < hfi_features.nr_classes; i++) {
> > +               struct hfi_cpu_data *class_caps;
> > +
> > +               class_caps = caps + i * hfi_features.class_stride;
> > +               WRITE_ONCE(hfi_class[i], class_caps->perf_cap);
> 
> As it stands, it is unclear why WRITE_ONCE() is needed here.

The CPU handling the HFI interrupt will update all the per-CPU IPCC
scores. My intention is to ensure that a WRITE of a given IPCC score
is completed before another CPU READs an IPCC score. The corresponding
READ_ONCE happens in patch 15.