diff mbox series

[4/4] drm/i915: Improve long running OCL w/a for GuC submission

Message ID 20211211065859.2248188-5-John.C.Harrison@Intel.com (mailing list archive)
State New, archived
Headers show
Series More fixes/tweaks to GuC support | expand

Commit Message

John Harrison Dec. 11, 2021, 6:58 a.m. UTC
From: John Harrison <John.C.Harrison@Intel.com>

A workaround was added to the driver to allow OpenCL workloads to run
'forever' by disabling pre-emption on the RCS engine for Gen12.
It is not totally unbound as the heartbeat will kick in eventually
and cause a reset of the hung engine.

However, this does not work well in GuC submission mode. In GuC mode,
the pre-emption timeout is how GuC detects hung contexts and triggers
a per engine reset. Thus, disabling the timeout means also losing all
per engine reset ability. A full GT reset will still occur when the
heartbeat finally expires, but that is a much more destructive and
undesirable mechanism.

The purpose of the workaround is actually to give OpenCL tasks longer
to reach a pre-emption point after a pre-emption request has been
issued. This is necessary because Gen12 does not support mid-thread
pre-emption and OpenCL can have long running threads.

So, rather than disabling the timeout completely, just set it to a
'long' value. Likewise, bump the heartbeat interval. That gives the
OpenCL thread sufficient time to reach a pre-emption point without
being killed off either by the GuC or by the heartbeat.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
---
 drivers/gpu/drm/i915/gt/intel_engine_cs.c | 42 +++++++++++++++++++++--
 1 file changed, 39 insertions(+), 3 deletions(-)

Comments

Daniele Ceraolo Spurio Dec. 16, 2021, 1:11 a.m. UTC | #1
On 12/10/2021 10:58 PM, John.C.Harrison@Intel.com wrote:
> From: John Harrison <John.C.Harrison@Intel.com>
>
> A workaround was added to the driver to allow OpenCL workloads to run
> 'forever' by disabling pre-emption on the RCS engine for Gen12.
> It is not totally unbound as the heartbeat will kick in eventually
> and cause a reset of the hung engine.
>
> However, this does not work well in GuC submission mode. In GuC mode,
> the pre-emption timeout is how GuC detects hung contexts and triggers
> a per engine reset. Thus, disabling the timeout means also losing all
> per engine reset ability. A full GT reset will still occur when the
> heartbeat finally expires, but that is a much more destructive and
> undesirable mechanism.
>
> The purpose of the workaround is actually to give OpenCL tasks longer
> to reach a pre-emption point after a pre-emption request has been
> issued. This is necessary because Gen12 does not support mid-thread
> pre-emption and OpenCL can have long running threads.
>
> So, rather than disabling the timeout completely, just set it to a
> 'long' value. Likewise, bump the heartbeat interval. That gives the
> OpenCL thread sufficient time to reach a pre-emption point without
> being killed off either by the GuC or by the heartbeat.
>
> Signed-off-by: John Harrison <John.C.Harrison@Intel.com>

On the approach and the code:

Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>

Please get an ack from the interested parties on the actual numbers used 
for the timeouts (they look big enough to me, but I'm not familiar with 
the compute use-case).

Daniele

> ---
>   drivers/gpu/drm/i915/gt/intel_engine_cs.c | 42 +++++++++++++++++++++--
>   1 file changed, 39 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> index 352254e001b4..26af8d60fe2b 100644
> --- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> +++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
> @@ -382,9 +382,45 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id,
>   	engine->props.timeslice_duration_ms =
>   		CONFIG_DRM_I915_TIMESLICE_DURATION;
>   
> -	/* Override to uninterruptible for OpenCL workloads. */
> -	if (GRAPHICS_VER(i915) == 12 && engine->class == RENDER_CLASS)
> -		engine->props.preempt_timeout_ms = 0;
> +	/*
> +	 * Mid-thread pre-emption is not available in Gen12. Unfortunately,
> +	 * some OpenCL workloads run quite long threads. That means they get
> +	 * reset due to not pre-empting in a timely manner.
> +	 * The execlist solution was to disable pre-emption completely.
> +	 * However, pre-emption timeouts are the way GuC detects hung contexts
> +	 * and triggers engine resets. Thus, without pre-emption, there is no
> +	 * per engine reset. And full GT reset is much more intrusive. So keep
> +	 * the timeout for GuC submission platforms and just bump it to be
> +	 * much larger. Also bump the heartbeat timeout to match, otherwise
> +	 * the heartbeat can expire before the pre-emption can timeout and
> +	 * thus trigger a full GT reset anyway.
> +	 */
> +	if (GRAPHICS_VER(i915) == 12 && engine->class == RENDER_CLASS) {
> +		if (intel_uc_wants_guc_submission(&gt->uc)) {
> +			const unsigned long min_preempt = 7500;
> +			const unsigned long min_beat = 5000;
> +
> +			if (engine->props.preempt_timeout_ms &&
> +			    engine->props.preempt_timeout_ms < min_preempt) {
> +				drm_info(&gt->i915->drm, "Bumping pre-emption timeout from %ld to %ld on %s to allow slow compute pre-emption\n",
> +					 engine->props.preempt_timeout_ms,
> +					 min_preempt, engine->name);
> +
> +				engine->props.preempt_timeout_ms = min_preempt;
> +			}
> +
> +			if (engine->props.heartbeat_interval_ms &&
> +			    engine->props.heartbeat_interval_ms < min_beat) {
> +				drm_info(&gt->i915->drm, "Bumping heartbeat interval from %ld to %ld on %s to allow slow compute pre-emption\n",
> +					 engine->props.heartbeat_interval_ms,
> +					 min_beat, engine->name);
> +
> +				engine->props.heartbeat_interval_ms = min_beat;
> +			}
> +		} else {
> +			engine->props.preempt_timeout_ms = 0;
> +		}
> +	}
>   
>   	engine->defaults = engine->props; /* never to change again */
>
diff mbox series

Patch

diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index 352254e001b4..26af8d60fe2b 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -382,9 +382,45 @@  static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id,
 	engine->props.timeslice_duration_ms =
 		CONFIG_DRM_I915_TIMESLICE_DURATION;
 
-	/* Override to uninterruptible for OpenCL workloads. */
-	if (GRAPHICS_VER(i915) == 12 && engine->class == RENDER_CLASS)
-		engine->props.preempt_timeout_ms = 0;
+	/*
+	 * Mid-thread pre-emption is not available in Gen12. Unfortunately,
+	 * some OpenCL workloads run quite long threads. That means they get
+	 * reset due to not pre-empting in a timely manner.
+	 * The execlist solution was to disable pre-emption completely.
+	 * However, pre-emption timeouts are the way GuC detects hung contexts
+	 * and triggers engine resets. Thus, without pre-emption, there is no
+	 * per engine reset. And full GT reset is much more intrusive. So keep
+	 * the timeout for GuC submission platforms and just bump it to be
+	 * much larger. Also bump the heartbeat timeout to match, otherwise
+	 * the heartbeat can expire before the pre-emption can timeout and
+	 * thus trigger a full GT reset anyway.
+	 */
+	if (GRAPHICS_VER(i915) == 12 && engine->class == RENDER_CLASS) {
+		if (intel_uc_wants_guc_submission(&gt->uc)) {
+			const unsigned long min_preempt = 7500;
+			const unsigned long min_beat = 5000;
+
+			if (engine->props.preempt_timeout_ms &&
+			    engine->props.preempt_timeout_ms < min_preempt) {
+				drm_info(&gt->i915->drm, "Bumping pre-emption timeout from %ld to %ld on %s to allow slow compute pre-emption\n",
+					 engine->props.preempt_timeout_ms,
+					 min_preempt, engine->name);
+
+				engine->props.preempt_timeout_ms = min_preempt;
+			}
+
+			if (engine->props.heartbeat_interval_ms &&
+			    engine->props.heartbeat_interval_ms < min_beat) {
+				drm_info(&gt->i915->drm, "Bumping heartbeat interval from %ld to %ld on %s to allow slow compute pre-emption\n",
+					 engine->props.heartbeat_interval_ms,
+					 min_beat, engine->name);
+
+				engine->props.heartbeat_interval_ms = min_beat;
+			}
+		} else {
+			engine->props.preempt_timeout_ms = 0;
+		}
+	}
 
 	engine->defaults = engine->props; /* never to change again */