[PATCH] drm/i915: Fix context runtime accounting

* [PATCH] drm/i915: Fix context runtime accounting
@ 2023-03-20 15:14 ` Tvrtko Ursulin
  0 siblings, 0 replies; 14+ messages in thread
From: Tvrtko Ursulin @ 2023-03-20 15:14 UTC (permalink / raw)
  To: Intel-gfx, dri-devel; +Cc: stable, Tvrtko Ursulin

From: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

When considering whether to mark one context as stopped and another as
started we need to look at whether the previous and new _contexts_ are
different and not just requests. Otherwise the software tracked context
start time was incorrectly updated to the most recent lite-restore time-
stamp, which was in some cases resulting in active time going backward,
until the context switch (typically the hearbeat pulse) would synchronise
with the hardware tracked context runtime. Easiest use case to observe
this behaviour was with a full screen clients with close to 100% engine
load.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Fixes: bb6287cb1886 ("drm/i915: Track context current active time")
Cc: <stable@vger.kernel.org> # v5.19+
---
 drivers/gpu/drm/i915/gt/intel_execlists_submission.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
index 1bbe6708d0a7..750326434677 100644
--- a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
@@ -2018,6 +2018,8 @@ process_csb(struct intel_engine_cs *engine, struct i915_request **inactive)
 	 * inspecting the queue to see if we need to resumbit.
 	 */
 	if (*prev != *execlists->active) { /* elide lite-restores */
+		struct intel_context *prev_ce = NULL, *active_ce = NULL;
+
 		/*
 		 * Note the inherent discrepancy between the HW runtime,
 		 * recorded as part of the context switch, and the CPU
@@ -2029,9 +2031,15 @@ process_csb(struct intel_engine_cs *engine, struct i915_request **inactive)
 		 * and correct overselves later when updating from HW.
 		 */
 		if (*prev)
-			lrc_runtime_stop((*prev)->context);
+			prev_ce = (*prev)->context;
 		if (*execlists->active)
-			lrc_runtime_start((*execlists->active)->context);
+			active_ce = (*execlists->active)->context;
+		if (prev_ce != active_ce) {
+			if (prev_ce)
+				lrc_runtime_stop(prev_ce);
+			if (active_ce)
+				lrc_runtime_start(active_ce);
+		}
 		new_timeslice(execlists);
 	}
 
-- 
2.37.2


^ permalink raw reply related	[flat|nested] 14+ messages in thread