[Intel-gfx] [PATCH] drm/i915/gt: Try to more gracefully quiesce the system before resets

[Chris Wilson]

If we are doing a normal GPU reset triggered after detecting a long
period of stalled work, we can take our time and allow the engines to
quiesce. Since we've stopped submission to the engine, and if we wait
long enough an innocent context should complete, leaving the engine idle.
So by waiting a short amount of time, we should prevent clobbering other
users when resetting a stuck context.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
---
 drivers/gpu/drm/i915/Kconfig.profile      | 11 +++++++++++
 drivers/gpu/drm/i915/gt/intel_engine_cs.c | 17 ++++++++++++++++-
 2 files changed, 27 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/Kconfig.profile b/drivers/gpu/drm/i915/Kconfig.profile
index 48df8889a88a..97f01bfeda41 100644
--- a/drivers/gpu/drm/i915/Kconfig.profile
+++ b/drivers/gpu/drm/i915/Kconfig.profile
@@ -25,3 +25,14 @@ config DRM_I915_SPIN_REQUEST
 	  May be 0 to disable the initial spin. In practice, we estimate
 	  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.
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index 0e20713603ec..1b41613b59a5 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -875,6 +875,21 @@ 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 CONFIG_DRM_I915_STOP_TIMEOUT;
+}
+
 int intel_engine_stop_cs(struct intel_engine_cs *engine)
 {
 	struct intel_uncore *uncore = engine->uncore;
@@ -892,7 +907,7 @@ 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;
-- 
2.24.0.rc0