arm_pmu: Invert fallback armpmu_filter() return value

[PATCH] arm_pmu: Invert fallback armpmu_filter() return value

Posted by Janne Grunau 2 years, 6 months ago

Fixes a regression in "perf: Rewrite core context handling" for arm_pmu
based drivers without a filter function pointer as the Apple M1 PMU.
The event is ignored if filter() returns 'true' opposite to
filter_match(). The refactoring failed to invert the return value if
arm_pmu has no filter() function pointer of its own.

Fixes: bd2756811766 ("perf: Rewrite core context handling")
Signed-off-by: Janne Grunau <j@jannau.net>
---
 drivers/perf/arm_pmu.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
index 9b593f985805..9cb22f36cf66 100644
--- a/drivers/perf/arm_pmu.c
+++ b/drivers/perf/arm_pmu.c
@@ -556,7 +556,7 @@ static bool armpmu_filter(struct pmu *pmu, int cpu)
 	if (ret && armpmu->filter)
 		return armpmu->filter(pmu, cpu);
 
-	return ret;
+	return !ret;
 }
 
 static ssize_t cpus_show(struct device *dev,

---
base-commit: ceaa837f96adb69c0df0397937cd74991d5d821a
change-id: 20230215-arm_pmu_m1_regression-b6dd48d0c792

Best regards,
-- 
Janne Grunau <j@jannau.net>

Re: [PATCH] arm_pmu: Invert fallback armpmu_filter() return value

Posted by Mark Rutland 2 years, 6 months ago

Hi,

On Wed, Feb 15, 2023 at 12:27:13PM +0100, Janne Grunau wrote:
> Fixes a regression in "perf: Rewrite core context handling" for arm_pmu
> based drivers without a filter function pointer as the Apple M1 PMU.
> The event is ignored if filter() returns 'true' opposite to
> filter_match(). The refactoring failed to invert the return value if
> arm_pmu has no filter() function pointer of its own.
> 
> Fixes: bd2756811766 ("perf: Rewrite core context handling")
> Signed-off-by: Janne Grunau <j@jannau.net>

Thanks for the report and the patch, and apologies for this happening in the
first place; this has clearly slipped by our testing...

I think this fix is correct, but I spot that bd2756811766 also broken CHAIN
event filtering, and made armv8pmu_filter() trivial, so we can get rid of
arm_pmu::filter() entirely and make this simpler and clearer.

I'm going to send a couple of patches to addres both those issues in a moment,
with everyone here on Cc.

Thanks,
Mark.

> ---
>  drivers/perf/arm_pmu.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
> index 9b593f985805..9cb22f36cf66 100644
> --- a/drivers/perf/arm_pmu.c
> +++ b/drivers/perf/arm_pmu.c
> @@ -556,7 +556,7 @@ static bool armpmu_filter(struct pmu *pmu, int cpu)
>  	if (ret && armpmu->filter)
>  		return armpmu->filter(pmu, cpu);
>  
> -	return ret;
> +	return !ret;
>  }
>  
>  static ssize_t cpus_show(struct device *dev,
> 
> ---
> base-commit: ceaa837f96adb69c0df0397937cd74991d5d821a
> change-id: 20230215-arm_pmu_m1_regression-b6dd48d0c792
> 
> Best regards,
> -- 
> Janne Grunau <j@jannau.net>
>

Re: [PATCH] arm_pmu: Invert fallback armpmu_filter() return value (*6.2 regression*)

Posted by Asahi Lina 2 years, 6 months ago

On 15/02/2023 20.27, Janne Grunau wrote:
> Fixes a regression in "perf: Rewrite core context handling" for arm_pmu
> based drivers without a filter function pointer as the Apple M1 PMU.
> The event is ignored if filter() returns 'true' opposite to
> filter_match(). The refactoring failed to invert the return value if
> arm_pmu has no filter() function pointer of its own.
> 
> Fixes: bd2756811766 ("perf: Rewrite core context handling")
> Signed-off-by: Janne Grunau <j@jannau.net>
> ---
>  drivers/perf/arm_pmu.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
> index 9b593f985805..9cb22f36cf66 100644
> --- a/drivers/perf/arm_pmu.c
> +++ b/drivers/perf/arm_pmu.c
> @@ -556,7 +556,7 @@ static bool armpmu_filter(struct pmu *pmu, int cpu)
>  	if (ret && armpmu->filter)
>  		return armpmu->filter(pmu, cpu);
>  
> -	return ret;
> +	return !ret;
>  }
>  
>  static ssize_t cpus_show(struct device *dev,
> 
> ---
> base-commit: ceaa837f96adb69c0df0397937cd74991d5d821a
> change-id: 20230215-arm_pmu_m1_regression-b6dd48d0c792
> 
> Best regards,

Tested-by: Asahi Lina <lina@asahilina.net>

This fixes `perf` being completely broken on 6.2. Adding a note to the
subject, I hope this can get picked up before release...

~~ Lina

Re: [PATCH] arm_pmu: Invert fallback armpmu_filter() return value

Posted by Eric Curtin 2 years, 6 months ago

On Wed, 15 Feb 2023 at 11:36, Janne Grunau <j@jannau.net> wrote:
>
> Fixes a regression in "perf: Rewrite core context handling" for arm_pmu
> based drivers without a filter function pointer as the Apple M1 PMU.
> The event is ignored if filter() returns 'true' opposite to
> filter_match(). The refactoring failed to invert the return value if
> arm_pmu has no filter() function pointer of its own.
>
> Fixes: bd2756811766 ("perf: Rewrite core context handling")
> Signed-off-by: Janne Grunau <j@jannau.net>

Makes sense we return the inverse of armpmu_filter everywhere else in the code.

Reviewed-by: Eric Curtin <ecurtin@redhat.com>

Is mise le meas/Regards,

Eric Curtin

> ---
>  drivers/perf/arm_pmu.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
> index 9b593f985805..9cb22f36cf66 100644
> --- a/drivers/perf/arm_pmu.c
> +++ b/drivers/perf/arm_pmu.c
> @@ -556,7 +556,7 @@ static bool armpmu_filter(struct pmu *pmu, int cpu)
>         if (ret && armpmu->filter)
>                 return armpmu->filter(pmu, cpu);
>
> -       return ret;
> +       return !ret;
>  }
>
>  static ssize_t cpus_show(struct device *dev,
>
> ---
> base-commit: ceaa837f96adb69c0df0397937cd74991d5d821a
> change-id: 20230215-arm_pmu_m1_regression-b6dd48d0c792
>
> Best regards,
> --
> Janne Grunau <j@jannau.net>
>
>

Re: [PATCH] arm_pmu: Invert fallback armpmu_filter() return value

Posted by Eric Curtin 2 years, 6 months ago

On Wed, 15 Feb 2023 at 11:40, Eric Curtin <ecurtin@redhat.com> wrote:
>
> On Wed, 15 Feb 2023 at 11:36, Janne Grunau <j@jannau.net> wrote:
> >
> > Fixes a regression in "perf: Rewrite core context handling" for arm_pmu
> > based drivers without a filter function pointer as the Apple M1 PMU.
> > The event is ignored if filter() returns 'true' opposite to
> > filter_match(). The refactoring failed to invert the return value if
> > arm_pmu has no filter() function pointer of its own.
> >
> > Fixes: bd2756811766 ("perf: Rewrite core context handling")
> > Signed-off-by: Janne Grunau <j@jannau.net>
>
> Makes sense we return the inverse of armpmu_filter everywhere else in the code.

or we return the inverse of cpumask_test_cpu rather... copy/paste
error on my part.

>
> Reviewed-by: Eric Curtin <ecurtin@redhat.com>
>
> Is mise le meas/Regards,
>
> Eric Curtin
>
> > ---
> >  drivers/perf/arm_pmu.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
> > index 9b593f985805..9cb22f36cf66 100644
> > --- a/drivers/perf/arm_pmu.c
> > +++ b/drivers/perf/arm_pmu.c
> > @@ -556,7 +556,7 @@ static bool armpmu_filter(struct pmu *pmu, int cpu)
> >         if (ret && armpmu->filter)
> >                 return armpmu->filter(pmu, cpu);
> >
> > -       return ret;
> > +       return !ret;
> >  }
> >
> >  static ssize_t cpus_show(struct device *dev,
> >
> > ---
> > base-commit: ceaa837f96adb69c0df0397937cd74991d5d821a
> > change-id: 20230215-arm_pmu_m1_regression-b6dd48d0c792
> >
> > Best regards,
> > --
> > Janne Grunau <j@jannau.net>
> >
> >