[Intel-gfx] [PATCH] drm/i915: Skip repeated calls to i915_gem_set_wedged()

[Chris Wilson]

If we already wedged, i915_gem_set_wedged() becomes a complicated no-op.

v2: Make sure the double set-wedged is synchronous, a parallel call
should not return before the driver is indeed wedged.

References: https://bugs.freedesktop.org/show_bug.cgi?id=107343
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c       | 26 ++++++++++++++++++++++----
 drivers/gpu/drm/i915/i915_gpu_error.h |  3 ++-
 2 files changed, 24 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 8b52cb768a67..99c91c7ad46c 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3298,12 +3298,24 @@ static void nop_complete_submit_request(struct i915_request *request)
 	spin_unlock_irqrestore(&request->engine->timeline.lock, flags);
 }
 
+static void wait_for_wedged(struct drm_i915_private *i915)
+{
+	struct i915_gpu_error *error = &i915->gpu_error;
+	DEFINE_WAIT_BIT(wq_entry, &error->flags, I915_WEDGED);
+
+	__wait_on_bit(&error->reset_queue,
+		      &wq_entry, bit_wait, TASK_UNINTERRUPTIBLE);
+}
+
 void i915_gem_set_wedged(struct drm_i915_private *i915)
 {
 	struct intel_engine_cs *engine;
 	enum intel_engine_id id;
 
-	GEM_TRACE("start\n");
+	if (test_and_set_bit(I915_WEDGE_IN_PROGRESS, &i915->gpu_error.flags)) {
+		wait_for_wedged(i915);
+		return;
+	}
 
 	if (GEM_SHOW_DEBUG()) {
 		struct drm_printer p = drm_debug_printer(__func__);
@@ -3312,8 +3324,7 @@ void i915_gem_set_wedged(struct drm_i915_private *i915)
 			intel_engine_dump(engine, &p, "%s\n", engine->name);
 	}
 
-	set_bit(I915_WEDGED, &i915->gpu_error.flags);
-	smp_mb__after_atomic();
+	GEM_TRACE("start\n");
 
 	/*
 	 * First, stop submission to hw, but do not yet complete requests by
@@ -3372,6 +3383,9 @@ void i915_gem_set_wedged(struct drm_i915_private *i915)
 		i915_gem_reset_finish_engine(engine);
 	}
 
+	smp_mb__before_atomic();
+	set_bit(I915_WEDGED, &i915->gpu_error.flags);
+
 	GEM_TRACE("end\n");
 
 	wake_up_all(&i915->gpu_error.reset_queue);
@@ -3379,10 +3393,14 @@ void i915_gem_set_wedged(struct drm_i915_private *i915)
 
 bool i915_gem_unset_wedged(struct drm_i915_private *i915)
 {
+	struct i915_gpu_error *error = &i915->gpu_error;
 	struct i915_timeline *tl;
 
 	lockdep_assert_held(&i915->drm.struct_mutex);
-	if (!test_bit(I915_WEDGED, &i915->gpu_error.flags))
+
+	while (test_and_clear_bit(I915_WEDGE_IN_PROGRESS, &error->flags))
+		wait_for_wedged(i915);
+	if (!test_bit(I915_WEDGE_IN_PROGRESS, &error->flags))
 		return true;
 
 	GEM_TRACE("start\n");
diff --git a/drivers/gpu/drm/i915/i915_gpu_error.h b/drivers/gpu/drm/i915/i915_gpu_error.h
index f893a4e8b783..1a78a8f330f2 100644
--- a/drivers/gpu/drm/i915/i915_gpu_error.h
+++ b/drivers/gpu/drm/i915/i915_gpu_error.h
@@ -267,8 +267,9 @@ struct i915_gpu_error {
 #define I915_RESET_BACKOFF	0
 #define I915_RESET_HANDOFF	1
 #define I915_RESET_MODESET	2
+#define I915_RESET_ENGINE	3
 #define I915_WEDGED		(BITS_PER_LONG - 1)
-#define I915_RESET_ENGINE	(I915_WEDGED - I915_NUM_ENGINES)
+#define I915_WEDGE_IN_PROGRESS	(I915_WEDGED - 1)
 
 	/** Number of times an engine has been reset */
 	u32 reset_engine_count[I915_NUM_ENGINES];
-- 
2.18.0