[PATCH v5] drm/xe/pmu: Add GT frequency events
Belgaumkar, Vinay
vinay.belgaumkar at intel.com
Mon Mar 31 17:47:06 UTC 2025
On 3/31/2025 10:44 AM, Rodrigo Vivi wrote:
> On Fri, Mar 28, 2025 at 03:53:30PM -0700, Vinay Belgaumkar wrote:
>> Define PMU events for GT frequency (actual and requested). The
>> instantaneous values for these frequencies will be displayed.
>>
>> Following PMU events are being added:
>> xe_0000_00_02.0/gt-actual-frequency/ [Kernel PMU event]
>> xe_0000_00_02.0/gt-requested-frequency/ [Kernel PMU event]
>>
>> Standard perf commands can be used to monitor GT frequency:
>> $ perf stat -e xe_0000_00_02.0/gt-requested-frequency,gt=0/ -I1000
>>
>> 1.001229762 1483 Mhz xe_0000_00_02.0/gt-requested-frequency,gt=0/
>> 2.006175406 1483 Mhz xe_0000_00_02.0/gt-requested-frequency,gt=0/
>>
>> v2: Use locks while storing/reading samples, keep track of multiple
>> clients (Lucas) and other general cleanup.
>> v3: Review comments (Lucas) and use event counts instead of mask for
>> active events.
>> v4: Add freq events to event_param_valid method (Riana)
>> v5: Use instantaneous values instead of aggregating (Lucas)
>> v6: Obtain fwake at init for freq events as well and use non fwake
>> variant method for reading requested freq to avoid lockdep issues (Lucas)
>>
>> Cc: Riana Tauro <riana.tauro at intel.com>
>> Cc: Lucas De Marchi <lucas.demarchi at intel.com>
>> Cc: Rodrigo Vivi <rodrigo.vivi at intel.com>
>> Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar at intel.com>
>> ---
>> drivers/gpu/drm/xe/xe_guc_pc.c | 15 +++++++++++++
>> drivers/gpu/drm/xe/xe_guc_pc.h | 1 +
>> drivers/gpu/drm/xe/xe_pmu.c | 40 +++++++++++++++++++++++++++++++---
>> 3 files changed, 53 insertions(+), 3 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/xe/xe_guc_pc.c b/drivers/gpu/drm/xe/xe_guc_pc.c
>> index 85215313976c..f70762917aed 100644
>> --- a/drivers/gpu/drm/xe/xe_guc_pc.c
>> +++ b/drivers/gpu/drm/xe/xe_guc_pc.c
>> @@ -461,6 +461,21 @@ static u32 get_cur_freq(struct xe_gt *gt)
>> return decode_freq(freq);
>> }
>>
>> +/**
>> + * xe_guc_pc_get_cur_freq_fw - With fw held, get requested frequency
>> + * @pc: The GuC PC
>> + *
>> + * Returns: the requested frequency for that GT instance
>> + */
>> +u32 xe_guc_pc_get_cur_freq_fw(struct xe_guc_pc *pc)
>> +{
>> + struct xe_gt *gt = pc_to_gt(pc);
>> +
>> + xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GT);
>> +
>> + return get_cur_freq(gt);
>> +}
>> +
>> /**
>> * xe_guc_pc_get_cur_freq - Get Current requested frequency
>> * @pc: The GuC PC
>> diff --git a/drivers/gpu/drm/xe/xe_guc_pc.h b/drivers/gpu/drm/xe/xe_guc_pc.h
>> index 39102b79602f..0a2664d5c811 100644
>> --- a/drivers/gpu/drm/xe/xe_guc_pc.h
>> +++ b/drivers/gpu/drm/xe/xe_guc_pc.h
>> @@ -22,6 +22,7 @@ void xe_guc_pc_print(struct xe_guc_pc *pc, struct drm_printer *p);
>>
>> u32 xe_guc_pc_get_act_freq(struct xe_guc_pc *pc);
>> int xe_guc_pc_get_cur_freq(struct xe_guc_pc *pc, u32 *freq);
>> +u32 xe_guc_pc_get_cur_freq_fw(struct xe_guc_pc *pc);
>> u32 xe_guc_pc_get_rp0_freq(struct xe_guc_pc *pc);
>> u32 xe_guc_pc_get_rpa_freq(struct xe_guc_pc *pc);
>> u32 xe_guc_pc_get_rpe_freq(struct xe_guc_pc *pc);
>> diff --git a/drivers/gpu/drm/xe/xe_pmu.c b/drivers/gpu/drm/xe/xe_pmu.c
>> index fde4222675fd..835d21429b58 100644
>> --- a/drivers/gpu/drm/xe/xe_pmu.c
>> +++ b/drivers/gpu/drm/xe/xe_pmu.c
>> @@ -10,6 +10,7 @@
>> #include "xe_force_wake.h"
>> #include "xe_gt_idle.h"
>> #include "xe_guc_engine_activity.h"
>> +#include "xe_guc_pc.h"
>> #include "xe_hw_engine.h"
>> #include "xe_pm.h"
>> #include "xe_pmu.h"
>> @@ -84,6 +85,8 @@ static unsigned int config_to_gt_id(u64 config)
>> #define XE_PMU_EVENT_GT_C6_RESIDENCY 0x01
>> #define XE_PMU_EVENT_ENGINE_ACTIVE_TICKS 0x02
>> #define XE_PMU_EVENT_ENGINE_TOTAL_TICKS 0x03
>> +#define XE_PMU_EVENT_GT_ACTUAL_FREQUENCY 0x04
>> +#define XE_PMU_EVENT_GT_REQUESTED_FREQUENCY 0x05
>>
>> static struct xe_gt *event_to_gt(struct perf_event *event)
>> {
>> @@ -119,6 +122,14 @@ static bool is_engine_event(u64 config)
>> event_id == XE_PMU_EVENT_ENGINE_ACTIVE_TICKS);
>> }
>>
>> +static bool is_gt_frequency_event(struct perf_event *event)
>> +{
>> + uint32_t id = config_to_event_id(event->attr.config);
>> +
>> + return (id == XE_PMU_EVENT_GT_ACTUAL_FREQUENCY) ||
>> + (id == XE_PMU_EVENT_GT_REQUESTED_FREQUENCY);
>> +}
>> +
>> static bool event_gt_forcewake(struct perf_event *event)
>> {
>> struct xe_device *xe = container_of(event->pmu, typeof(*xe), pmu.base);
>> @@ -126,7 +137,7 @@ static bool event_gt_forcewake(struct perf_event *event)
>> struct xe_gt *gt;
>> unsigned int *fw_ref;
>>
>> - if (!is_engine_event(config))
>> + if (!is_engine_event(config) && !is_gt_frequency_event(event))
>> return true;
>>
>> gt = xe_device_get_gt(xe, config_to_gt_id(config));
>> @@ -173,6 +184,8 @@ static bool event_param_valid(struct perf_event *event)
>>
>> switch (config_to_event_id(config)) {
>> case XE_PMU_EVENT_GT_C6_RESIDENCY:
>> + case XE_PMU_EVENT_GT_ACTUAL_FREQUENCY:
>> + case XE_PMU_EVENT_GT_REQUESTED_FREQUENCY:
>> if (engine_class || engine_instance || function_id)
>> return false;
>> break;
>> @@ -190,6 +203,8 @@ static bool event_param_valid(struct perf_event *event)
>> }
>>
>> break;
>> + default:
>> + return false;
> why do we now need this here?
> Doesn't it deserve a separate patch with explanation?
Sure. I had initially missed out adding the freq cases, and the switch
just fell through since there was no default case.
Thanks,
Vinay.
>
>> }
>>
>> return true;
>> @@ -288,6 +303,10 @@ static u64 __xe_pmu_event_read(struct perf_event *event)
>> case XE_PMU_EVENT_ENGINE_ACTIVE_TICKS:
>> case XE_PMU_EVENT_ENGINE_TOTAL_TICKS:
>> return read_engine_events(gt, event);
>> + case XE_PMU_EVENT_GT_ACTUAL_FREQUENCY:
>> + return xe_guc_pc_get_act_freq(>->uc.guc.pc);
>> + case XE_PMU_EVENT_GT_REQUESTED_FREQUENCY:
>> + return xe_guc_pc_get_cur_freq_fw(>->uc.guc.pc);
>> }
>>
>> return 0;
>> @@ -303,7 +322,13 @@ static void xe_pmu_event_update(struct perf_event *event)
>> new = __xe_pmu_event_read(event);
>> } while (!local64_try_cmpxchg(&hwc->prev_count, &prev, new));
>>
>> - local64_add(new - prev, &event->count);
>> + /* GT frequency is not a monotonically increasing counter, so record the
>> + * instantaneous value instead.
>> + */
>> + if (is_gt_frequency_event(event))
>> + local64_add(new, &event->count);
>> + else
>> + local64_add(new - prev, &event->count);
>> }
>>
>> static void xe_pmu_event_read(struct perf_event *event)
>> @@ -443,6 +468,10 @@ static ssize_t event_attr_show(struct device *dev,
>> XE_EVENT_ATTR_SIMPLE(gt-c6-residency, gt_c6_residency, XE_PMU_EVENT_GT_C6_RESIDENCY, "ms");
>> XE_EVENT_ATTR_NOUNIT(engine-active-ticks, engine_active_ticks, XE_PMU_EVENT_ENGINE_ACTIVE_TICKS);
>> XE_EVENT_ATTR_NOUNIT(engine-total-ticks, engine_total_ticks, XE_PMU_EVENT_ENGINE_TOTAL_TICKS);
>> +XE_EVENT_ATTR_SIMPLE(gt-actual-frequency, gt_actual_frequency,
>> + XE_PMU_EVENT_GT_ACTUAL_FREQUENCY, "Mhz");
>> +XE_EVENT_ATTR_SIMPLE(gt-requested-frequency, gt_requested_frequency,
>> + XE_PMU_EVENT_GT_REQUESTED_FREQUENCY, "Mhz");
>>
>> static struct attribute *pmu_empty_event_attrs[] = {
>> /* Empty - all events are added as groups with .attr_update() */
>> @@ -458,6 +487,8 @@ static const struct attribute_group *pmu_events_attr_update[] = {
>> &pmu_group_gt_c6_residency,
>> &pmu_group_engine_active_ticks,
>> &pmu_group_engine_total_ticks,
>> + &pmu_group_gt_actual_frequency,
>> + &pmu_group_gt_requested_frequency,
>> NULL,
>> };
>>
>> @@ -466,8 +497,11 @@ static void set_supported_events(struct xe_pmu *pmu)
>> struct xe_device *xe = container_of(pmu, typeof(*xe), pmu);
>> struct xe_gt *gt = xe_device_get_gt(xe, 0);
>>
>> - if (!xe->info.skip_guc_pc)
>> + if (!xe->info.skip_guc_pc) {
>> pmu->supported_events |= BIT_ULL(XE_PMU_EVENT_GT_C6_RESIDENCY);
>> + pmu->supported_events |= BIT_ULL(XE_PMU_EVENT_GT_ACTUAL_FREQUENCY);
>> + pmu->supported_events |= BIT_ULL(XE_PMU_EVENT_GT_REQUESTED_FREQUENCY);
>> + }
>>
>> if (xe_guc_engine_activity_supported(>->uc.guc)) {
>> pmu->supported_events |= BIT_ULL(XE_PMU_EVENT_ENGINE_ACTIVE_TICKS);
>> --
>> 2.38.1
>>
More information about the Intel-xe
mailing list