[PATCH 2/3] i915/guc: Ensure busyness counter increases monotonically
Umesh Nerlige Ramappa
umesh.nerlige.ramappa at intel.com
Mon Nov 25 20:10:46 UTC 2024
On Thu, Nov 21, 2024 at 04:34:14PM -0800, John Harrison wrote:
>On 11/21/2024 16:31, John Harrison wrote:
>>On 11/18/2024 15:22, Umesh Nerlige Ramappa wrote:
>>>Active busyness of an engine is calculated using gt timestamp and the
>>>context switch in time. While capturing the gt timestamp, it's possible
>>>that the context switches out. This race could result in an active
>>>busyness value that is greater than the actual context runtime
>>>value by a
>>>small amount. This leads to a negative delta and throws off busyness
>>>calculations for the user.
>>>
>>>If a subsequent count is smaller than the previous one, just return the
>>>previous one, since we expect the busyness to catch up.
>>>
>>>Fixes: 77cdd054dd2c ("drm/i915/pmu: Connect engine busyness stats
>>>from GuC to pmu")
>>>Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa at intel.com>
>>Reviewed-by: John Harrison <John.C.Harrison at Intel.com>
>Actually, does this need to be cleared in
>__reset_guc_busyness_stats()? If there is a reset and the live value
>is returned to zero then you won't get any more updates until it gets
>back to where it used to be.
We don't return the live value. We just accumulate the deltas from live
values into stats->total_gt_clks, compare it to stats->total and return
the one that is monotonically increasing.
The stats->total_gt_clks and stats->total are retained across resets, so
we should be good.
Thanks,
Umesh
>
>John.
>
>>
>>>---
>>> drivers/gpu/drm/i915/gt/intel_engine_types.h | 5 +++++
>>> drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 5 ++++-
>>> 2 files changed, 9 insertions(+), 1 deletion(-)
>>>
>>>diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h
>>>b/drivers/gpu/drm/i915/gt/intel_engine_types.h
>>>index ba55c059063d..fe1f85e5dda3 100644
>>>--- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
>>>+++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
>>>@@ -343,6 +343,11 @@ struct intel_engine_guc_stats {
>>> * @start_gt_clk: GT clock time of last idle to active transition.
>>> */
>>> u64 start_gt_clk;
>>>+
>>>+ /**
>>>+ * @total: The last value of total returned
>>>+ */
>>>+ u64 total;
>>> };
>>> union intel_engine_tlb_inv_reg {
>>>diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
>>>b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
>>>index c71aedcbce43..56be9f385270 100644
>>>--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
>>>+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
>>>@@ -1378,9 +1378,12 @@ static ktime_t guc_engine_busyness(struct
>>>intel_engine_cs *engine, ktime_t *now)
>>> total += intel_gt_clock_interval_to_ns(gt, clk);
>>> }
>>> + if (total > stats->total)
>>>+ stats->total = total;
>>>+
>>> spin_unlock_irqrestore(&guc->timestamp.lock, flags);
>>> - return ns_to_ktime(total);
>>>+ return ns_to_ktime(stats->total);
>>> }
>>> static void guc_enable_busyness_worker(struct intel_guc *guc)
>>
>
More information about the Intel-gfx
mailing list