[Intel-gfx] [PATCH 04/12] drm/i915/gt: Expose reset stop timeout via sysfs

[Chris Wilson]

Allow ourselves to sleep before a GPU reset, for a few milliseconds.
This allows for userspace to gracefully quiesce before being shot down.
By stopping submission and then waiting for the current requests to
complete, we are less likely to disrupt innocent contexts by performing
the reset while they are active. The longer we wait, the more likely we
are to spare innocents, but also the more disruptive to interactive, or
other low latency task, the reset itself will be.

The timeout can be adjusted using

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

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
Cc: Jon Bloomfield <jon.bloomfield at intel.com>
---
 drivers/gpu/drm/i915/Kconfig.profile         | 14 +++++++++
 drivers/gpu/drm/i915/gt/intel_engine_cs.c    | 19 +++++++++++-
 drivers/gpu/drm/i915/gt/intel_engine_sysfs.c | 31 ++++++++++++++++++++
 drivers/gpu/drm/i915/gt/intel_engine_types.h |  1 +
 4 files changed, 64 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile
index b8df80bc0b47..274bdd208e0d 100644
--- a/drivers/gpu/drm/i915/Kconfig.profile
+++ b/drivers/gpu/drm/i915/Kconfig.profile
@@ -26,6 +26,20 @@ config DRM_I915_SPIN_REQUEST
 	  the cost of enabling the interrupt (if currently disabled) to be
 	  a few microseconds.
 
+config DRM_I915_STOP_TIMEOUT
+	int "How long to wait for an engine to quiesce gracefully before reset (ms)"
+	default 100 # milliseconds
+	help
+	  By stopping submission and sleeping for a short time before resetting
+	  the GPU, we allow the innocent contexts also on the system to quiesce.
+	  It is then less likely for a hanging context to cause collateral
+	  damage as the system is reset in order to recover. The colorary is
+	  that the reset itself may take longer and so be more disruptive to
+	  interactive or low latency workloads.
+
+	  This is adjustable via
+	  /sys/class/drm/card?/engine/*/stop_timeout_ms
+
 config DRM_I915_TIMESLICE_DURATION
 	int "Scheduling quantum for userspace batches (ms, jiffy granularity)"
 	default 1 # milliseconds
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index 98dbaaaaf3db..5f8dad04915f 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -308,6 +308,8 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id)
 	engine->instance = info->instance;
 	__sprint_engine_name(engine);
 
+	engine->props.stop_timeout_ms =
+		CONFIG_DRM_I915_STOP_TIMEOUT;
 	engine->props.timeslice_duration_ms =
 		CONFIG_DRM_I915_TIMESLICE_DURATION;
 
@@ -878,6 +880,21 @@ u64 intel_engine_get_last_batch_head(const struct intel_engine_cs *engine)
 	return bbaddr;
 }
 
+static unsigned long stop_timeout(const struct intel_engine_cs *engine)
+{
+	if (in_atomic() || irqs_disabled()) /* inside atomic preempt-reset? */
+		return 0; /* Going too fast, can't stop now! */
+
+	/*
+	 * If we are doing a normal GPU reset, we can take our time and allow
+	 * the engine to quiesce. We've stopped submission to the engine, and
+	 * if we wait long enough an innocent context should complete and
+	 * leave the engine idle. So they should not be caught unaware by
+	 * the forthcoming GPU reset (which usually follows the stop_cs)!
+	 */
+	return READ_ONCE(engine->props.stop_timeout_ms);
+}
+
 int intel_engine_stop_cs(struct intel_engine_cs *engine)
 {
 	struct intel_uncore *uncore = engine->uncore;
@@ -895,7 +912,7 @@ int intel_engine_stop_cs(struct intel_engine_cs *engine)
 	err = 0;
 	if (__intel_wait_for_register_fw(uncore,
 					 mode, MODE_IDLE, MODE_IDLE,
-					 1000, 0,
+					 1000, stop_timeout(engine),
 					 NULL)) {
 		GEM_TRACE("%s: timed out on STOP_RING -> IDLE\n", engine->name);
 		err = -ETIMEDOUT;
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c b/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c
index 55ae81769a8e..73a755f44cce 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_sysfs.c
@@ -184,6 +184,36 @@ timeslice_store(struct kobject *kobj, struct kobj_attribute *attr,
 static struct kobj_attribute timeslice_duration_attr =
 __ATTR(timeslice_duration_ms, 0644, timeslice_show, timeslice_store);
 
+static ssize_t
+stop_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.stop_timeout_ms);
+}
+
+static ssize_t
+stop_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;
+
+	err = kstrtoull(buf, 0, &duration);
+	if (err)
+		return err;
+
+	if (duration > jiffies_to_msecs(MAX_SCHEDULE_TIMEOUT))
+		return -EINVAL;
+
+	WRITE_ONCE(engine->props.stop_timeout_ms, duration);
+	return count;
+}
+
+static struct kobj_attribute stop_timeout_attr =
+__ATTR(stop_timeout_ms, 0644, stop_show, stop_store);
+
 static void kobj_engine_release(struct kobject *kobj)
 {
 	kfree(kobj);
@@ -224,6 +254,7 @@ void intel_engines_add_sysfs(struct drm_i915_private *i915)
 		&mmio_attr.attr,
 		&caps_attr.attr,
 		&all_caps_attr.attr,
+		&stop_timeout_attr.attr,
 		NULL
 	};
 
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h
index 89a9616e8539..76ef8a63f921 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
@@ -544,6 +544,7 @@ struct intel_engine_cs {
 	} stats;
 
 	struct {
+		unsigned long stop_timeout_ms;
 		unsigned long timeslice_duration_ms;
 	} props;
 };
-- 
2.24.0.rc0