[PATCH v2] drivers/perf: arm-pmu: Handle per-interrupt affinity mask

[Marc Zyngier]

Hi Geert,

On 19/07/16 14:25, Geert Uytterhoeven wrote:
> Hi Marc, Catalin, Will,
> 
> On Wed, Jul 6, 2016 at 4:33 PM, Marc Zyngier <marc.zyngier at arm.com> wrote:
>> On a big-little system, PMUs can be wired to CPUs using per CPU
>> interrups (PPI). In this case, it is important to make sure that
>> the enable/disable do happen on the right set of CPUs.
>>
>> So instead of relying on the interrupt-affinity property, we can
>> use the actual percpu affinity that DT exposes as part of the
>> interrupt specifier. The DT binding is also updated to reflect
>> the fact that the interrupt-affinity property shouldn't be used
>> in that case.
>>
>> Signed-off-by: Marc Zyngier <marc.zyngier at arm.com>
>> ---
>> * From v1:
>>   - propagate the error if irq_get_percpu_devid_partition fails
> 
> This patch, which is commit 19a469a58720ea96 in arm64/for-next/core, broke
> the PMU on r8a7740/armadillo800eva:
> 
>     -hw perfevents: enabled with armv7_cortex_a9 PMU driver, 7
> counters available
>     +hw perfevents: /pmu: failed to probe PMU!
>     +hw perfevents: /pmu: failed to register PMU devices!
>     +armv7-pmu: probe of pmu failed with error -22
> 
> This is a single-core Cortex A9.
> 
>>  Documentation/devicetree/bindings/arm/pmu.txt |  4 +++-
>>  drivers/perf/arm_pmu.c                        | 27 ++++++++++++++++++++++-----
>>  2 files changed, 25 insertions(+), 6 deletions(-)
>>
>> diff --git a/Documentation/devicetree/bindings/arm/pmu.txt b/Documentation/devicetree/bindings/arm/pmu.txt
>> index 74d5417..61c8b46 100644
>> --- a/Documentation/devicetree/bindings/arm/pmu.txt
>> +++ b/Documentation/devicetree/bindings/arm/pmu.txt
>> @@ -39,7 +39,9 @@ Optional properties:
>>                         When using a PPI, specifies a list of phandles to CPU
>>                        nodes corresponding to the set of CPUs which have
>>                        a PMU of this type signalling the PPI listed in the
>> -                      interrupts property.
>> +                      interrupts property, unless this is already specified
>> +                      by the PPI interrupt specifier itself (in which case
>> +                      the interrupt-affinity property shouldn't be present).
>>
>>                         This property should be present when there is more than
>>                        a single SPI.
> 
> On a single core, there's only a single SPI, hence there's no need for an
> "interrupt-affinity" property.
> 
>> diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
>> index 140436a..8e4d7f5 100644
>> --- a/drivers/perf/arm_pmu.c
>> +++ b/drivers/perf/arm_pmu.c
>> @@ -961,9 +964,23 @@ static int of_pmu_irq_cfg(struct arm_pmu *pmu)
>>                 i++;
>>         } while (1);
>>
>> -       /* If we didn't manage to parse anything, claim to support all CPUs */
>> -       if (cpumask_weight(&pmu->supported_cpus) == 0)
>> -               cpumask_setall(&pmu->supported_cpus);
>> +       /* If we didn't manage to parse anything, try the interrupt affinity */
>> +       if (cpumask_weight(&pmu->supported_cpus) == 0) {
>> +               if (!using_spi) {
> 
> However, using_spi is never set to true in the absence of that property,
> causing the wrong branch to be taken...
> 
>> +                       /* If using PPIs, check the affinity of the partition */
>> +                       int ret, irq;
>> +
>> +                       irq = platform_get_irq(pdev, 0);
>> +                       ret = irq_get_percpu_devid_partition(irq, &pmu->supported_cpus);
> 
> ... and ret to become -22 here.

Thanks for the thorough analysis. Could you please give the following
patchlet a go:

diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
index 2513365..9275e08 100644
--- a/drivers/perf/arm_pmu.c
+++ b/drivers/perf/arm_pmu.c
@@ -958,11 +958,12 @@ static int of_pmu_irq_cfg(struct arm_pmu *pmu)
 
 	/* If we didn't manage to parse anything, try the interrupt affinity */
 	if (cpumask_weight(&pmu->supported_cpus) == 0) {
-		if (!using_spi) {
+		int irq = platform_get_irq(pdev, 0);
+
+		if (irq_is_percpu(irq)) {
 			/* If using PPIs, check the affinity of the partition */
-			int ret, irq;
+			int ret;
 
-			irq = platform_get_irq(pdev, 0);
 			ret = irq_get_percpu_devid_partition(irq, &pmu->supported_cpus);
 			if (ret) {
 				kfree(irqs);


and let me know if that helps?

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...