All of lore.kernel.org
 help / color / mirror / Atom feed
* [Intel-gfx] [PATCH] drm/i915/pmu: Fix synchronization of PMU callback with reset
@ 2021-11-03 22:47 Umesh Nerlige Ramappa
  2021-11-03 23:47 ` [Intel-gfx] ✓ Fi.CI.BAT: success for " Patchwork
                   ` (3 more replies)
  0 siblings, 4 replies; 13+ messages in thread
From: Umesh Nerlige Ramappa @ 2021-11-03 22:47 UTC (permalink / raw)
  To: intel-gfx, Tvrtko Ursulin

Since the PMU callback runs in irq context, it synchronizes with gt
reset using the reset count. We could run into a case where the PMU
callback could read the reset count before it is updated. This has a
potential of corrupting the busyness stats.

In addition to the reset count, check if the reset bit is set before
capturing busyness.

In addition save the previous stats only if you intend to update them.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
---
 drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

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 5cc49c0b3889..d83ade77ca07 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
@@ -1183,6 +1183,7 @@ static ktime_t guc_engine_busyness(struct intel_engine_cs *engine, ktime_t *now)
 	u64 total, gt_stamp_saved;
 	unsigned long flags;
 	u32 reset_count;
+	bool in_reset;
 
 	spin_lock_irqsave(&guc->timestamp.lock, flags);
 
@@ -1191,7 +1192,9 @@ static ktime_t guc_engine_busyness(struct intel_engine_cs *engine, ktime_t *now)
 	 * engine busyness from GuC, so we just use the driver stored
 	 * copy of busyness. Synchronize with gt reset using reset_count.
 	 */
-	reset_count = i915_reset_count(gpu_error);
+	rcu_read_lock();
+	in_reset = test_bit(I915_RESET_BACKOFF, &gt->reset.flags);
+	rcu_read_unlock();
 
 	*now = ktime_get();
 
@@ -1201,9 +1204,10 @@ static ktime_t guc_engine_busyness(struct intel_engine_cs *engine, ktime_t *now)
 	 * start_gt_clk is derived from GuC state. To get a consistent
 	 * view of activity, we query the GuC state only if gt is awake.
 	 */
-	stats_saved = *stats;
-	gt_stamp_saved = guc->timestamp.gt_stamp;
-	if (intel_gt_pm_get_if_awake(gt)) {
+	if (intel_gt_pm_get_if_awake(gt) && !in_reset) {
+		stats_saved = *stats;
+		gt_stamp_saved = guc->timestamp.gt_stamp;
+		reset_count = i915_reset_count(gpu_error);
 		guc_update_engine_gt_clks(engine);
 		guc_update_pm_timestamp(guc, engine, now);
 		intel_gt_pm_put_async(gt);
-- 
2.20.1


^ permalink raw reply related	[flat|nested] 13+ messages in thread
* [Intel-gfx] [PATCH] drm/i915/pmu: Fix synchronization of PMU callback with reset
@ 2021-11-08 21:10 Umesh Nerlige Ramappa
  0 siblings, 0 replies; 13+ messages in thread
From: Umesh Nerlige Ramappa @ 2021-11-08 21:10 UTC (permalink / raw)
  To: intel-gfx, dri-devel

Since the PMU callback runs in irq context, it synchronizes with gt
reset using the reset count. We could run into a case where the PMU
callback could read the reset count before it is updated. This has a
potential of corrupting the busyness stats.

In addition to the reset count, check if the reset bit is set before
capturing busyness.

In addition save the previous stats only if you intend to update them.

v2:
- The 2 reset counts captured in the PMU callback can end up being the
  same if they were captured right after the count is incremented in the
  reset flow. This can lead to a bad busyness state. Ensure that reset
  is not in progress when the initial reset count is captured.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
---
 .../gpu/drm/i915/gt/uc/intel_guc_submission.c   | 17 +++++++++++------
 1 file changed, 11 insertions(+), 6 deletions(-)

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 5cc49c0b3889..0dfc6032cd6b 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
@@ -1183,15 +1183,20 @@ static ktime_t guc_engine_busyness(struct intel_engine_cs *engine, ktime_t *now)
 	u64 total, gt_stamp_saved;
 	unsigned long flags;
 	u32 reset_count;
+	bool in_reset;
 
 	spin_lock_irqsave(&guc->timestamp.lock, flags);
 
 	/*
-	 * If a reset happened, we risk reading partially updated
-	 * engine busyness from GuC, so we just use the driver stored
-	 * copy of busyness. Synchronize with gt reset using reset_count.
+	 * If a reset happened, we risk reading partially updated engine
+	 * busyness from GuC, so we just use the driver stored copy of busyness.
+	 * Synchronize with gt reset using reset_count and the
+	 * I915_RESET_BACKOFF flag. Note that reset flow updates the reset_count
+	 * after I915_RESET_BACKOFF flag, so ensure that the reset_count is
+	 * usable by checking the flag afterwards.
 	 */
 	reset_count = i915_reset_count(gpu_error);
+	in_reset = test_bit(I915_RESET_BACKOFF, &gt->reset.flags);
 
 	*now = ktime_get();
 
@@ -1201,9 +1206,9 @@ static ktime_t guc_engine_busyness(struct intel_engine_cs *engine, ktime_t *now)
 	 * start_gt_clk is derived from GuC state. To get a consistent
 	 * view of activity, we query the GuC state only if gt is awake.
 	 */
-	stats_saved = *stats;
-	gt_stamp_saved = guc->timestamp.gt_stamp;
-	if (intel_gt_pm_get_if_awake(gt)) {
+	if (intel_gt_pm_get_if_awake(gt) && !in_reset) {
+		stats_saved = *stats;
+		gt_stamp_saved = guc->timestamp.gt_stamp;
 		guc_update_engine_gt_clks(engine);
 		guc_update_pm_timestamp(guc, engine, now);
 		intel_gt_pm_put_async(gt);
-- 
2.20.1


^ permalink raw reply related	[flat|nested] 13+ messages in thread
end of thread, other threads:[~2021-11-23  9:15 UTC | newest]

Thread overview: 13+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-11-03 22:47 [Intel-gfx] [PATCH] drm/i915/pmu: Fix synchronization of PMU callback with reset Umesh Nerlige Ramappa
2021-11-03 23:47 ` [Intel-gfx] ✓ Fi.CI.BAT: success for " Patchwork
2021-11-04  0:55 ` [Intel-gfx] ✗ Fi.CI.IGT: failure " Patchwork
2021-11-04 15:57 ` [Intel-gfx] [PATCH] " Matthew Brost
2021-11-04 17:37 ` Tvrtko Ursulin
2021-11-04 22:04   ` Umesh Nerlige Ramappa
2021-11-11 14:37     ` Tvrtko Ursulin
2021-11-11 16:48       ` Umesh Nerlige Ramappa
2021-11-20  0:25         ` Umesh Nerlige Ramappa
2021-11-22 15:44         ` Tvrtko Ursulin
2021-11-22 23:39           ` Umesh Nerlige Ramappa
2021-11-23  9:15             ` Tvrtko Ursulin
2021-11-08 21:10 Umesh Nerlige Ramappa

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.