From: Francois Dugast <francois.dugast@intel.com>
To: Matthew Brost <matthew.brost@intel.com>
Cc: <intel-xe@lists.freedesktop.org>,
<thomas.hellstrom@linux.intel.com>, <michal.mrozek@intel.com>
Subject: Re: [PATCH v2 7/7] drm/xe: Add more GT stats around pagefault mode switch flows
Date: Mon, 15 Dec 2025 14:05:16 +0100 [thread overview]
Message-ID: <aUAHjCZdTZxECohh@fdugast-desk> (raw)
In-Reply-To: <20251212182847.1683222-8-matthew.brost@intel.com>
On Fri, Dec 12, 2025 at 10:28:47AM -0800, Matthew Brost wrote:
> Add GT stats to measure the time spent switching between pagefault mode
> and dma-fence mode. Also add a GT stat to indicate when pagefault
> suspend is skipped because the system is idle. These metrics will help
> profile pagefault workloads while 3D and display are enabled.
>
> v2:
> - Use GT stats helper functions (Francois)
>
> Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Francois Dugast <francois.dugast@intel.com>
> ---
> drivers/gpu/drm/xe/xe_gt_stats.c | 6 ++++++
> drivers/gpu/drm/xe/xe_gt_stats_types.h | 3 +++
> drivers/gpu/drm/xe/xe_hw_engine_group.c | 22 +++++++++++++++++++++-
> 3 files changed, 30 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/xe/xe_gt_stats.c b/drivers/gpu/drm/xe/xe_gt_stats.c
> index 714045ad9354..fb2904bd0abd 100644
> --- a/drivers/gpu/drm/xe/xe_gt_stats.c
> +++ b/drivers/gpu/drm/xe/xe_gt_stats.c
> @@ -68,8 +68,14 @@ static const char *const stat_description[__XE_GT_STATS_NUM_IDS] = {
> DEF_STAT_STR(SVM_2M_BIND_US, "svm_2M_bind_us"),
> DEF_STAT_STR(HW_ENGINE_GROUP_SUSPEND_LR_QUEUE_COUNT,
> "hw_engine_group_suspend_lr_queue_count"),
> + DEF_STAT_STR(HW_ENGINE_GROUP_SKIP_LR_QUEUE_COUNT,
> + "hw_engine_group_skip_lr_queue_count"),
> DEF_STAT_STR(HW_ENGINE_GROUP_WAIT_DMA_QUEUE_COUNT,
> "hw_engine_group_wait_dma_queue_count"),
> + DEF_STAT_STR(HW_ENGINE_GROUP_SUSPEND_LR_QUEUE_US,
> + "hw_engine_group_suspend_lr_queue_us"),
> + DEF_STAT_STR(HW_ENGINE_GROUP_WAIT_DMA_QUEUE_US,
> + "hw_engine_group_wait_dma_queue_us"),
> };
>
> /**
> diff --git a/drivers/gpu/drm/xe/xe_gt_stats_types.h b/drivers/gpu/drm/xe/xe_gt_stats_types.h
> index aada5df421e5..b92d013091d5 100644
> --- a/drivers/gpu/drm/xe/xe_gt_stats_types.h
> +++ b/drivers/gpu/drm/xe/xe_gt_stats_types.h
> @@ -45,7 +45,10 @@ enum xe_gt_stats_id {
> XE_GT_STATS_ID_SVM_64K_BIND_US,
> XE_GT_STATS_ID_SVM_2M_BIND_US,
> XE_GT_STATS_ID_HW_ENGINE_GROUP_SUSPEND_LR_QUEUE_COUNT,
> + XE_GT_STATS_ID_HW_ENGINE_GROUP_SKIP_LR_QUEUE_COUNT,
> XE_GT_STATS_ID_HW_ENGINE_GROUP_WAIT_DMA_QUEUE_COUNT,
> + XE_GT_STATS_ID_HW_ENGINE_GROUP_SUSPEND_LR_QUEUE_US,
> + XE_GT_STATS_ID_HW_ENGINE_GROUP_WAIT_DMA_QUEUE_US,
> /* must be the last entry */
> __XE_GT_STATS_NUM_IDS,
> };
> diff --git a/drivers/gpu/drm/xe/xe_hw_engine_group.c b/drivers/gpu/drm/xe/xe_hw_engine_group.c
> index 022fc0c30d38..9a53021bbfa7 100644
> --- a/drivers/gpu/drm/xe/xe_hw_engine_group.c
> +++ b/drivers/gpu/drm/xe/xe_hw_engine_group.c
> @@ -200,7 +200,9 @@ static int xe_hw_engine_group_suspend_faulting_lr_jobs(struct xe_hw_engine_group
> {
> int err;
> struct xe_exec_queue *q;
> + struct xe_gt *gt = NULL;
> bool need_resume = false;
> + ktime_t start = xe_gt_stats_ktime_get();
>
> lockdep_assert_held_write(&group->mode_sem);
>
> @@ -215,10 +217,13 @@ static int xe_hw_engine_group_suspend_faulting_lr_jobs(struct xe_hw_engine_group
> return -EAGAIN;
>
> xe_gt_stats_incr(q->gt, XE_GT_STATS_ID_HW_ENGINE_GROUP_SUSPEND_LR_QUEUE_COUNT, 1);
> -
> + if (idle_skip_suspend)
> + xe_gt_stats_incr(q->gt,
> + XE_GT_STATS_ID_HW_ENGINE_GROUP_SKIP_LR_QUEUE_COUNT, 1);
>
> need_resume |= !idle_skip_suspend;
> q->ops->suspend(q);
> + gt = q->gt;
> }
>
> list_for_each_entry(q, &group->exec_queue_list, hw_engine_group_link) {
> @@ -230,6 +235,12 @@ static int xe_hw_engine_group_suspend_faulting_lr_jobs(struct xe_hw_engine_group
> return err;
> }
>
> + if (gt) {
> + xe_gt_stats_incr(gt,
> + XE_GT_STATS_ID_HW_ENGINE_GROUP_SUSPEND_LR_QUEUE_US,
> + xe_gt_stats_ktime_us_delta(start));
> + }
> +
> if (need_resume)
> xe_hw_engine_group_resume_faulting_lr_jobs(group);
>
> @@ -250,7 +261,9 @@ static int xe_hw_engine_group_wait_for_dma_fence_jobs(struct xe_hw_engine_group
> {
> long timeout;
> struct xe_exec_queue *q;
> + struct xe_gt *gt = NULL;
> struct dma_fence *fence;
> + ktime_t start = xe_gt_stats_ktime_get();
>
> lockdep_assert_held_write(&group->mode_sem);
>
> @@ -262,11 +275,18 @@ static int xe_hw_engine_group_wait_for_dma_fence_jobs(struct xe_hw_engine_group
> fence = xe_exec_queue_last_fence_get_for_resume(q, q->vm);
> timeout = dma_fence_wait(fence, false);
> dma_fence_put(fence);
> + gt = q->gt;
>
> if (timeout < 0)
> return -ETIME;
> }
>
> + if (gt) {
> + xe_gt_stats_incr(gt,
> + XE_GT_STATS_ID_HW_ENGINE_GROUP_WAIT_DMA_QUEUE_US,
> + xe_gt_stats_ktime_us_delta(start));
> + }
> +
> return 0;
> }
>
> --
> 2.34.1
>
next prev parent reply other threads:[~2025-12-15 13:05 UTC|newest]
Thread overview: 24+ messages / expand[flat|nested] mbox.gz Atom feed top
2025-12-12 18:28 [PATCH v2 0/7] Fix performance when pagefaults and 3d/display share resources Matthew Brost
2025-12-12 18:28 ` [PATCH v2 1/7] drm/xe: Adjust long-running workload timeslices to reasonable values Matthew Brost
2025-12-15 10:08 ` Thomas Hellström
2025-12-15 21:48 ` Matthew Brost
2025-12-12 18:28 ` [PATCH v2 2/7] drm/xe: Use usleep_range for accurate long-running workload timeslicing Matthew Brost
2025-12-15 10:10 ` Thomas Hellström
2025-12-12 18:28 ` [PATCH v2 3/7] drm/xe: Add debugfs knobs to control long running " Matthew Brost
2025-12-15 10:11 ` Thomas Hellström
2025-12-12 18:28 ` [PATCH v2 4/7] drm/xe: Skip exec queue schedule toggle if queue is idle during suspend Matthew Brost
2025-12-15 12:08 ` Thomas Hellström
2025-12-12 18:28 ` [PATCH v2 5/7] drm/xe: Wait on in-syncs when swicthing to dma-fence mode Matthew Brost
2025-12-15 10:32 ` Thomas Hellström
2025-12-15 21:46 ` Matthew Brost
2025-12-15 21:48 ` Thomas Hellström
2025-12-16 1:12 ` Matthew Brost
2025-12-12 18:28 ` [PATCH v2 6/7] drm/xe: Add GT stats ktime helpers Matthew Brost
2025-12-15 10:17 ` Thomas Hellström
2025-12-12 18:28 ` [PATCH v2 7/7] drm/xe: Add more GT stats around pagefault mode switch flows Matthew Brost
2025-12-15 11:00 ` Thomas Hellström
2025-12-15 13:05 ` Francois Dugast [this message]
2025-12-12 22:37 ` ✗ CI.checkpatch: warning for Fix performance when pagefaults and 3d/display share resources (rev2) Patchwork
2025-12-12 22:38 ` ✓ CI.KUnit: success " Patchwork
2025-12-12 23:33 ` ✓ Xe.CI.BAT: " Patchwork
2025-12-13 19:27 ` ✗ Xe.CI.Full: failure " Patchwork
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=aUAHjCZdTZxECohh@fdugast-desk \
--to=francois.dugast@intel.com \
--cc=intel-xe@lists.freedesktop.org \
--cc=matthew.brost@intel.com \
--cc=michal.mrozek@intel.com \
--cc=thomas.hellstrom@linux.intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox