drm/i915/gt: Expose busywait duration to sysfs

We busywait on an inflight request (one that is currently executing on
HW, and so might complete quickly) prior to setting up an interrupt and
sleeping. The trade off is that we keep an expensive CPU core busy in
order to avoid wake up latency: where that trade off should lie is best
left to the sysadmin.

The busywait mechanism can be compiled out with

	./scripts/config --set-val DRM_I915_SPIN_REQUEST 0

The maximum busywait duration can be adjusted per-engine using,

	/sys/class/drm/card?/engine/*/ms_busywait_duration_ns

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Steve Carbonari <steven.carbonari@intel.com>
Tested-by: Steve Carbonari <steven.carbonari@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-4-chris@chris-wilson.co.uk

3 files changed, 52 insertions, 0 deletions

diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index 119c9cb24fd4..53ac3f00909a 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -313,6 +313,8 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id)
 
 	engine->props.heartbeat_interval_ms =
 		CONFIG_DRM_I915_HEARTBEAT_INTERVAL;
+	engine->props.max_busywait_duration_ns =
+		CONFIG_DRM_I915_MAX_REQUEST_BUSYWAIT;
 	engine->props.preempt_timeout_ms =
 		CONFIG_DRM_I915_PREEMPT_TIMEOUT;
 	engine->props.stop_timeout_ms =
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h
index b23366a81048..80cdde712842 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
@@ -547,6 +547,7 @@ struct intel_engine_cs {
 
 	struct {
 		unsigned long heartbeat_interval_ms;
+		unsigned long max_busywait_duration_ns;
 		unsigned long preempt_timeout_ms;
 		unsigned long stop_timeout_ms;
 		unsigned long timeslice_duration_ms;
diff --git a/drivers/gpu/drm/i915/gt/sysfs_engines.c b/drivers/gpu/drm/i915/gt/sysfs_engines.c
index 53402edbf98a..619c43d1172d 100644
--- a/drivers/gpu/drm/i915/gt/sysfs_engines.c
+++ b/drivers/gpu/drm/i915/gt/sysfs_engines.c
@@ -143,6 +143,54 @@ static struct kobj_attribute all_caps_attr =
 __ATTR(known_capabilities, 0444, all_caps_show, NULL);
 
 static ssize_t
+max_spin_store(struct kobject *kobj, struct kobj_attribute *attr,
+	       const char *buf, size_t count)
+{
+	struct intel_engine_cs *engine = kobj_to_engine(kobj);
+	unsigned long long duration;
+	int err;
+
+	/*
+	 * When waiting for a request, if is it currently being executed
+	 * on the GPU, we busywait for a short while before sleeping. The
+	 * premise is that most requests are short, and if it is already
+	 * executing then there is a good chance that it will complete
+	 * before we can setup the interrupt handler and go to sleep.
+	 * We try to offset the cost of going to sleep, by first spinning
+	 * on the request -- if it completed in less time than it would take
+	 * to go sleep, process the interrupt and return back to the client,
+	 * then we have saved the client some latency, albeit at the cost
+	 * of spinning on an expensive CPU core.
+	 *
+	 * While we try to avoid waiting at all for a request that is unlikely
+	 * to complete, deciding how long it is worth spinning is for is an
+	 * arbitrary decision: trading off power vs latency.
+	 */
+
+	err = kstrtoull(buf, 0, &duration);
+	if (err)
+		return err;
+
+	if (duration > jiffies_to_nsecs(2))
+		return -EINVAL;
+
+	WRITE_ONCE(engine->props.max_busywait_duration_ns, duration);
+
+	return count;
+}
+
+static ssize_t
+max_spin_show(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
+{
+	struct intel_engine_cs *engine = kobj_to_engine(kobj);
+
+	return sprintf(buf, "%lu\n", engine->props.max_busywait_duration_ns);
+}
+
+static struct kobj_attribute max_spin_attr =
+__ATTR(max_busywait_duration_ns, 0644, max_spin_show, max_spin_store);
+
+static ssize_t
 timeslice_store(struct kobject *kobj, struct kobj_attribute *attr,
 		const char *buf, size_t count)
 {
@@ -224,6 +272,7 @@ void intel_engines_add_sysfs(struct drm_i915_private *i915)
 		&mmio_attr.attr,
 		&caps_attr.attr,
 		&all_caps_attr.attr,
+		&max_spin_attr.attr,
 		NULL
 	};