Re: [PATCH] perf/smmuv3: Enable HiSilicon Erratum 162001900 quirk for HIP08/09

[Yicong Yang]

On 2023/8/11 19:17, Will Deacon wrote:
> On Wed, Aug 09, 2023 at 06:06:54PM +0800, Yicong Yang wrote:
>> diff --git a/drivers/perf/arm_smmuv3_pmu.c b/drivers/perf/arm_smmuv3_pmu.c
>> index 25a269d431e4..b854b67b81fc 100644
>> --- a/drivers/perf/arm_smmuv3_pmu.c
>> +++ b/drivers/perf/arm_smmuv3_pmu.c
>> @@ -115,6 +115,7 @@
>>  #define SMMU_PMCG_PA_SHIFT              12
>>  
>>  #define SMMU_PMCG_EVCNTR_RDONLY         BIT(0)
>> +#define SMMU_PMCG_HARDEN_DISABLE        BIT(1)
>>  
>>  static int cpuhp_state_num;
>>  
>> @@ -150,6 +151,22 @@ SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_stream_id, config1, 0, 31);
>>  SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_span, config1, 32, 32);
>>  SMMU_PMU_EVENT_ATTR_EXTRACTOR(filter_enable, config1, 33, 33);
>>  
>> +static int smmu_pmu_apply_event_filter(struct smmu_pmu *smmu_pmu,
>> +				       struct perf_event *event, int idx);
>> +
>> +static inline void smmu_pmu_enable_quirk_hip08_09(struct pmu *pmu)
>> +{
>> +	struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu);
>> +	unsigned int idx;
>> +
>> +	for_each_set_bit(idx, smmu_pmu->used_counters, smmu_pmu->num_counters)
>> +		smmu_pmu_apply_event_filter(smmu_pmu, smmu_pmu->events[idx], idx);
>> +
>> +	writel(SMMU_PMCG_IRQ_CTRL_IRQEN,
>> +	       smmu_pmu->reg_base + SMMU_PMCG_IRQ_CTRL);
>> +	writel(SMMU_PMCG_CR_ENABLE, smmu_pmu->reg_base + SMMU_PMCG_CR);
> 
> Can you tail-call smmu_pmu_enable() instead of duplicating it here?
> 
>> +static inline void smmu_pmu_disable_quirk_hip08_09(struct pmu *pmu)
>> +{
>> +	struct smmu_pmu *smmu_pmu = to_smmu_pmu(pmu);
>> +	unsigned int idx;
>> +
>> +	/*
>> +	 * The global disable of PMU sometimes fail to stop the counting.
>> +	 * Harden this by writing an invalid event type to each used counter
>> +	 * to forcibly stop counting.
>> +	 */
>> +	for_each_set_bit(idx, smmu_pmu->used_counters, smmu_pmu->num_counters)
>> +		writel(0xffff, smmu_pmu->reg_base + SMMU_PMCG_EVTYPER(idx));
>> +
>> +	writel(0, smmu_pmu->reg_base + SMMU_PMCG_CR);
>> +	writel(0, smmu_pmu->reg_base + SMMU_PMCG_IRQ_CTRL);
> 
> Same things here, but with smmu_pmu_disable()
> 

Sure. Will tail call smmu_pmu_{enable, disable} in both case to avoid duplication.
Thanks for the comments.

Thanks.