Intel-XE Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: Matthew Brost <matthew.brost@intel.com>
To: intel-xe@lists.freedesktop.org
Cc: francois.dugast@intel.com, thomas.hellstrom@linux.intel.com,
	michal.mrozek@intel.com
Subject: [PATCH v2 7/7] drm/xe: Add more GT stats around pagefault mode switch flows
Date: Fri, 12 Dec 2025 10:28:47 -0800	[thread overview]
Message-ID: <20251212182847.1683222-8-matthew.brost@intel.com> (raw)
In-Reply-To: <20251212182847.1683222-1-matthew.brost@intel.com>

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>
---
 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


  parent reply	other threads:[~2025-12-12 18:28 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 ` Matthew Brost [this message]
2025-12-15 11:00   ` [PATCH v2 7/7] drm/xe: Add more GT stats around pagefault mode switch flows Thomas Hellström
2025-12-15 13:05   ` Francois Dugast
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=20251212182847.1683222-8-matthew.brost@intel.com \
    --to=matthew.brost@intel.com \
    --cc=francois.dugast@intel.com \
    --cc=intel-xe@lists.freedesktop.org \
    --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