[PATCH v6 2/3] arm64: perf: Add support for event counting threshold
Suzuki K Poulose
suzuki.poulose at arm.com
Tue Dec 5 08:34:52 PST 2023
On 05/12/2023 15:05, James Clark wrote:
>
>
> On 05/12/2023 13:14, Will Deacon wrote:
>> On Fri, Nov 24, 2023 at 10:28:56AM +0000, James Clark wrote:
>>> diff --git a/drivers/perf/arm_pmuv3.c b/drivers/perf/arm_pmuv3.c
>>> index 1d40d794f5e4..eb1ef84e1dbb 100644
>>> --- a/drivers/perf/arm_pmuv3.c
>>> +++ b/drivers/perf/arm_pmuv3.c
>>> @@ -15,6 +15,7 @@
>>> #include <clocksource/arm_arch_timer.h>
>>>
>>> #include <linux/acpi.h>
>>> +#include <linux/bitfield.h>
>>> #include <linux/clocksource.h>
>>> #include <linux/of.h>
>>> #include <linux/perf/arm_pmu.h>
>>> @@ -294,9 +295,20 @@ static const struct attribute_group armv8_pmuv3_events_attr_group = {
>>> .is_visible = armv8pmu_event_attr_is_visible,
>>> };
>>>
>>> +#define THRESHOLD_LOW 2
>>> +#define THRESHOLD_HIGH 13
>>> +#define THRESHOLD_CNT 14
>>> +#define THRESHOLD_CMP_LO 15
>>> +#define THRESHOLD_CMP_HI 16
>>
>> Do you think THWIDTH might extend beyond 12 bits in future? If so, it might
>> be worth juggling these bits around a bit so it's not sandwiched between
>> 'rdpmc' and 'threshold_compare'. I defer to your judgement, however.
>>
>
> It's possible, both PMMIR.THWIDTH and PMEVTYPER.TH currently have unused
> bits above them. I can easily move threshold to the end, I suppose that
> covers us at least until someone adds a new field above that.
>
>>> PMU_FORMAT_ATTR(event, "config:0-15");
>>> PMU_FORMAT_ATTR(long, "config1:0");
>>> PMU_FORMAT_ATTR(rdpmc, "config1:1");
>>> +PMU_FORMAT_ATTR(threshold, "config1:" __stringify(THRESHOLD_LOW) "-"
>>> + __stringify(THRESHOLD_HIGH));
>>> +PMU_FORMAT_ATTR(threshold_compare, "config1:" __stringify(THRESHOLD_CMP_LO) "-"
>>> + __stringify(THRESHOLD_CMP_HI));
>>> +PMU_FORMAT_ATTR(threshold_count, "config1:" __stringify(THRESHOLD_CNT));
>>>
>>> static int sysctl_perf_user_access __read_mostly;
>>>
>>> @@ -310,10 +322,33 @@ static inline bool armv8pmu_event_want_user_access(struct perf_event *event)
>>> return event->attr.config1 & 0x2;
>>> }
>>>
>>> +static inline u32 armv8pmu_event_threshold(struct perf_event_attr *attr)
>>> +{
>>> + return FIELD_GET(GENMASK(THRESHOLD_HIGH, THRESHOLD_LOW), attr->config1);
>>> +}
>>> +
>>> +static inline u8 armv8pmu_event_threshold_control(struct perf_event_attr *attr)
>>
>> You can drop the 'inline's for these functions (and, in fact, this whole
>> file could do with that cleanup :)
>>
>
> Will do.
>
>>> +{
>>> + u8 th_compare = FIELD_GET(GENMASK(THRESHOLD_CMP_HI, THRESHOLD_CMP_LO),
>>> + attr->config1);
>>> + u8 th_count = FIELD_GET(BIT(THRESHOLD_CNT), attr->config1);
>>
>> I think this is correct, but you might want to look at how we handle this
>> in the SPE driver as I think it ends up looking cleaner and makes it pretty
>> obvious which bits correspond to the user ABI (i.e. config fields) and which
>> bits are part of architectural registers. I'm not saying you have to do it
>> that way, but please take a look if you haven't already.
>>
>
> Yeah I could take the GEN_PMU_FORMAT_ATTR() etc macros out of there and
> re-use them here too. And also for the other existing configs in
> arm_pmuv3.c.
>
>>> + /*
>>> + * The count bit is always the bottom bit of the full control field, and
>>> + * the comparison is the upper two bits, but it's not explicitly
>>> + * labelled in the Arm ARM. For the Perf interface we split it into two
>>> + * fields, so reconstruct it here.
>>> + */
>>> + return (th_compare << 1) | th_count;
>>> +}
>>> +
>>> static struct attribute *armv8_pmuv3_format_attrs[] = {
>>> &format_attr_event.attr,
>>> &format_attr_long.attr,
>>> &format_attr_rdpmc.attr,
>>> + &format_attr_threshold.attr,
>>> + &format_attr_threshold_compare.attr,
>>> + &format_attr_threshold_count.attr,
>>> NULL,
>>> };
>>>
>>> @@ -365,10 +400,38 @@ static ssize_t bus_width_show(struct device *dev, struct device_attribute *attr,
>>>
>>> static DEVICE_ATTR_RO(bus_width);
>>>
>>> +static u32 threshold_max(struct arm_pmu *cpu_pmu)
>>> +{
>>> + /*
>>> + * PMMIR.THWIDTH is readable and non-zero on aarch32, but it would be
>>> + * impossible to write the threshold in the upper 32 bits of PMEVTYPER.
>>> + */
>>> + if (IS_ENABLED(CONFIG_ARM))
>>> + return 0;
>>> +
>>> + /*
>>> + * The largest value that can be written to PMEVTYPER<n>_EL0.TH is
>>> + * (2 ^ PMMIR.THWIDTH) - 1.
>>> + */
>>> + return (1 << FIELD_GET(ARMV8_PMU_THWIDTH, cpu_pmu->reg_pmmir)) - 1;
>>> +}
>>> +
>>> +static ssize_t threshold_max_show(struct device *dev,
>>> + struct device_attribute *attr, char *page)
>>> +{
>>> + struct pmu *pmu = dev_get_drvdata(dev);
>>> + struct arm_pmu *cpu_pmu = container_of(pmu, struct arm_pmu, pmu);
>>> +
>>> + return sysfs_emit(page, "0x%08x\n", threshold_max(cpu_pmu));
>>> +}
>>> +
>>> +static DEVICE_ATTR_RO(threshold_max);
>>> +
>>> static struct attribute *armv8_pmuv3_caps_attrs[] = {
>>> &dev_attr_slots.attr,
>>> &dev_attr_bus_slots.attr,
>>> &dev_attr_bus_width.attr,
>>> + &dev_attr_threshold_max.attr,
>>> NULL,
>>> };
>>>
>>> @@ -552,7 +615,7 @@ static void armv8pmu_write_counter(struct perf_event *event, u64 value)
>>> armv8pmu_write_hw_counter(event, value);
>>> }
>>>
>>> -static inline void armv8pmu_write_evtype(int idx, u32 val)
>>> +static inline void armv8pmu_write_evtype(int idx, unsigned long val)
>>> {
>>> u32 counter = ARMV8_IDX_TO_COUNTER(idx);
>>> unsigned long mask = ARMV8_PMU_EVTYPE_EVENT |
>>> @@ -921,6 +984,10 @@ static int armv8pmu_set_event_filter(struct hw_perf_event *event,
>>> struct perf_event_attr *attr)
>>> {
>>> unsigned long config_base = 0;
>>> + struct perf_event *perf_event = container_of(attr, struct perf_event,
>>> + attr);
>>> + struct arm_pmu *cpu_pmu = to_arm_pmu(perf_event->pmu);
>>> + u32 th, th_max;
>>>
>>> if (attr->exclude_idle)
>>> return -EPERM;
>>> @@ -952,6 +1019,21 @@ static int armv8pmu_set_event_filter(struct hw_perf_event *event,
>>> if (attr->exclude_user)
>>> config_base |= ARMV8_PMU_EXCLUDE_EL0;
>>>
>>> + /*
>>> + * Insert event counting threshold (FEAT_PMUv3_TH) values. If
>>> + * FEAT_PMUv3_TH isn't implemented, then THWIDTH (threshold_max) will be
>>> + * 0 and no values will be written.
>>> + */
>>> + th_max = threshold_max(cpu_pmu);
>>> + if (IS_ENABLED(CONFIG_ARM64) && th_max) {
>>
>> Why is the IS_ENABLED() check needed here?
>>
>
> The FIELD_PREP() below would cause a compilation error on arm32 because
> TH and TC are above 32 bits. I can add a comment.
>
>>> + th = min(armv8pmu_event_threshold(attr), th_max);
>>> + if (th) {
>>
>> Why is it useful to take the minimum here? If userspace asks for a value
>> bigger than the maximum support threshold, shouldn't we return an error
>> rather than silently clamp it?
>>
>
> I think it probably was just so I didn't have to think about what would
> happen when the value varied between cores.
>
> But you're right, it looks like I can add a validation function to
> struct arm_pmu and call it from armpmu_event_init(). armpmu->map_event()
> and armpmu->set_event_filter() are already called from there so I think
> the validation could technically be added to one of those, but that's
> probably a bit hacky. I don't know if you have any preference for where
> the threshold validation should happen?
I think the other reason was to make sure we never set those fields for
something that doesn't support the feature ? (Given th_max is set to 0
in that case). But I agree that validation is a better approach than
silently masking it.
Suzuki
>
>>> + config_base |= FIELD_PREP(ARMV8_PMU_EVTYPE_TH, th);
>>> + config_base |= FIELD_PREP(ARMV8_PMU_EVTYPE_TC,
>>> + armv8pmu_event_threshold_control(attr));
>>> + }
>>> + }
>>> +
>>> /*
>>> * Install the filter into config_base as this is used to
>>> * construct the event type.
>>> diff --git a/include/linux/perf/arm_pmuv3.h b/include/linux/perf/arm_pmuv3.h
>>> index ddd1fec86739..ccbc0f9a74d8 100644
>>> --- a/include/linux/perf/arm_pmuv3.h
>>> +++ b/include/linux/perf/arm_pmuv3.h
>>> @@ -258,6 +258,7 @@
>>> #define ARMV8_PMU_BUS_SLOTS_MASK 0xff
>>> #define ARMV8_PMU_BUS_WIDTH_SHIFT 16
>>> #define ARMV8_PMU_BUS_WIDTH_MASK 0xf
>>> +#define ARMV8_PMU_THWIDTH GENMASK(23, 20)
>>
>> It's a bit messy having a mixture of GENMASK and MASK/SHIFT pairs. Please
>> can you either update what's there to use GENMASK, or use SHIFT/MASK for the
>> new addition?
>>
>> Will
>>
>
> Yep will do. Thanks for the review.
>
> James
More information about the linux-arm-kernel
mailing list