[Intel-gfx] [PATCH 12/21] drm/i915: Replace wait-on-mutex with wait-on-bit in reset worker

[Chris Wilson]

Since we have a cooperative mode now with a direct reset, we can avoid
the contention on struct_mutex and instead try then sleep on the
I915_RESET_IN_PROGRESS bit. If the mutex is held and that bit is
cleared, all is fine. Otherwise, we sleep for a bit and try again. In
the worst case we sleep for an extra second waiting for the mutex to be
released (no one touching the GPU is allowed the struct_mutex whilst the
I915_RESET_IN_PROGRESS bit is set). But when we have a direct reset,
this allows us to clean up the reset worker faster.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_irq.c | 30 ++++++++++++++++++------------
 1 file changed, 18 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 2c7cb5041511..699ee2c7a3e4 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2497,7 +2497,7 @@ static void i915_reset_and_wakeup(struct drm_i915_private *dev_priv)
 	char *error_event[] = { I915_ERROR_UEVENT "=1", NULL };
 	char *reset_event[] = { I915_RESET_UEVENT "=1", NULL };
 	char *reset_done_event[] = { I915_ERROR_UEVENT "=0", NULL };
-	int ret;
+	int ret = -EAGAIN;
 
 	kobject_uevent_env(kobj, KOBJ_CHANGE, error_event);
 
@@ -2512,21 +2512,27 @@ static void i915_reset_and_wakeup(struct drm_i915_private *dev_priv)
 	 * simulated reset via debugs, so get an RPM reference.
 	 */
 	intel_runtime_pm_get(dev_priv);
-
 	intel_prepare_reset(dev_priv);
 
-	/*
-	 * All state reset _must_ be completed before we update the
-	 * reset counter, for otherwise waiters might miss the reset
-	 * pending state and not properly drop locks, resulting in
-	 * deadlocks with the reset work.
-	 */
-	mutex_lock(&dev_priv->drm.struct_mutex);
-	ret = i915_reset(dev_priv);
-	mutex_unlock(&dev_priv->drm.struct_mutex);
+	do {
+		/*
+		 * All state reset _must_ be completed before we update the
+		 * reset counter, for otherwise waiters might miss the reset
+		 * pending state and not properly drop locks, resulting in
+		 * deadlocks with the reset work.
+		 */
+		if (mutex_trylock(&dev_priv->drm.struct_mutex)) {
+			ret = i915_reset(dev_priv);
+			mutex_unlock(&dev_priv->drm.struct_mutex);
+		}
 
-	intel_finish_reset(dev_priv);
+		/* We need to wait for anyone holding the lock to wakeup */
+	} while (wait_on_bit_timeout(&dev_priv->gpu_error.flags,
+				     I915_RESET_IN_PROGRESS,
+				     TASK_UNINTERRUPTIBLE,
+				     HZ));
 
+	intel_finish_reset(dev_priv);
 	intel_runtime_pm_put(dev_priv);
 
 	if (ret == 0)
-- 
2.9.3