Re: [PATCH -fixes] drivers: perf: Fix smp_processor_id() use in preemptible code

[Alexandre Ghiti]

Hi Will,

On Tue, Aug 27, 2024 at 2:53 PM Will Deacon <will@xxxxxxxxxx> wrote:
>
> On Mon, Aug 26, 2024 at 06:52:10PM +0200, Alexandre Ghiti wrote:
> > As reported in [1], the use of smp_processor_id() in
> > pmu_sbi_device_probe() must be protected by disabling the preemption, so
> > simple use get_cpu()/put_cpu() instead.
> >
> > Reported-by: Nam Cao <namcao@xxxxxxxxxxxxx>
> > Closes: https://lore.kernel.org/linux-riscv/20240820074925.ReMKUPP3@xxxxxxxxxxxxx/ [1]
> > Signed-off-by: Alexandre Ghiti <alexghiti@xxxxxxxxxxxx>
> > ---
> >  drivers/perf/riscv_pmu_sbi.c | 7 ++++++-
> >  1 file changed, 6 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/perf/riscv_pmu_sbi.c b/drivers/perf/riscv_pmu_sbi.c
> > index 31a17a56eb3b..25b1b699b3e2 100644
> > --- a/drivers/perf/riscv_pmu_sbi.c
> > +++ b/drivers/perf/riscv_pmu_sbi.c
> > @@ -1373,11 +1373,15 @@ static int pmu_sbi_device_probe(struct platform_device *pdev)
> >
> >       /* SBI PMU Snapsphot is only available in SBI v2.0 */
> >       if (sbi_v2_available) {
> > +             int cpu;
> > +
> >               ret = pmu_sbi_snapshot_alloc(pmu);
> >               if (ret)
> >                       goto out_unregister;
> >
> > -             ret = pmu_sbi_snapshot_setup(pmu, smp_processor_id());
> > +             cpu = get_cpu();
> > +
> > +             ret = pmu_sbi_snapshot_setup(pmu, cpu);
> >               if (ret) {
> >                       /* Snapshot is an optional feature. Continue if not available */
> >                       pmu_sbi_snapshot_free(pmu);
> > @@ -1391,6 +1395,7 @@ static int pmu_sbi_device_probe(struct platform_device *pdev)
> >                        */
> >                       static_branch_enable(&sbi_pmu_snapshot_available);
> >               }
> > +             put_cpu();
>
> Are you sure it's safe to enable the static key with preemption disabled?
> I thought that could block on a mutex.

Yep, it seems you're right, thanks for jumping in.

I'm discussing with Atish how to fix that differently, I'll be back
with another version very soon.

Thanks again,

Alex

>
> Will