[PATCH 10/18] drm/i915: Perform a direct reset of the GPU from the waiter

[Chris Wilson <chris@chris-wilson.co.uk>]

If a waiter is holding the struct_mutex, then the reset worker cannot
reset the GPU until the waiter returns. We do not want to return -EAGAIN
form i915_wait_request as that breaks delicate operations like
i915_vma_unbind() which often cannot be restarted easily, and returning
-EIO is just as useless (and has in the past proven dangerous). The
remaining WARN_ON(i915_wait_request) serve as a valuable reminder that
handling errors from an indefinite wait are tricky.

We can keep the current semantic that knowing after a reset is complete,
so is the request, by performing the reset ourselves if we hold the
mutex.

uevent emission is still handled by the reset worker, so it may appear
slightly out of order with respect to the actual reset (and concurrent
use of the device).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_drv.c         | 11 ++++++-----
 drivers/gpu/drm/i915/i915_drv.h         | 15 +++------------
 drivers/gpu/drm/i915/i915_gem.c         |  3 ++-
 drivers/gpu/drm/i915/i915_gem_request.c | 29 +++++++++++++++++++++++++----
 drivers/gpu/drm/i915/i915_gem_request.h |  9 +++++++--
 drivers/gpu/drm/i915/i915_irq.c         |  2 ++
 drivers/gpu/drm/i915/intel_ringbuffer.c |  6 ++----
 7 files changed, 47 insertions(+), 28 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
index 01b518dcbd7a..ea7d3e87815c 100644
--- a/drivers/gpu/drm/i915/i915_drv.c
+++ b/drivers/gpu/drm/i915/i915_drv.c
@@ -1752,6 +1752,8 @@ int i915_resume_switcheroo(struct drm_device *dev)
  * Reset the chip.  Useful if a hang is detected. Returns zero on successful
  * reset or otherwise an error code.
  *
+ * Caller must hold the struct_mutex.
+ *
  * Procedure is fairly simple:
  *   - reset the chip using the reset reg
  *   - re-init context state
@@ -1766,7 +1768,10 @@ int i915_reset(struct drm_i915_private *dev_priv)
 	struct i915_gpu_error *error = &dev_priv->gpu_error;
 	int ret;
 
-	mutex_lock(&dev->struct_mutex);
+	lockdep_assert_held(&dev->struct_mutex);
+
+	if (!test_and_clear_bit(I915_RESET_IN_PROGRESS, &error->flags))
+		return 0;
 
 	/* Clear any previous failed attempts at recovery. Time to try again. */
 	__clear_bit(I915_WEDGED, &error->flags);
@@ -1807,9 +1812,6 @@ int i915_reset(struct drm_i915_private *dev_priv)
 		goto error;
 	}
 
-	clear_bit(I915_RESET_IN_PROGRESS, &error->flags);
-	mutex_unlock(&dev->struct_mutex);
-
 	/*
 	 * rps/rc6 re-init is necessary to restore state lost after the
 	 * reset and the re-install of gt irqs. Skip for ironlake per
@@ -1823,7 +1825,6 @@ int i915_reset(struct drm_i915_private *dev_priv)
 
 error:
 	set_bit(I915_WEDGED, &error->flags);
-	mutex_unlock(&dev->struct_mutex);
 	return ret;
 }
 
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index e574eaa65c4d..c6dbc6b5798a 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -3849,7 +3849,9 @@ wait_remaining_ms_from_jiffies(unsigned long timestamp_jiffies, int to_wait_ms)
 			    schedule_timeout_uninterruptible(remaining_jiffies);
 	}
 }
-static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
+
+static inline bool
+__i915_request_irq_complete(struct drm_i915_gem_request *req)
 {
 	struct intel_engine_cs *engine = req->engine;
 
@@ -3911,17 +3913,6 @@ static inline bool __i915_request_irq_complete(struct drm_i915_gem_request *req)
 			return true;
 	}
 
-	/* We need to check whether any gpu reset happened in between
-	 * the request being submitted and now. If a reset has occurred,
-	 * the seqno will have been advance past ours and our request
-	 * is complete. If we are in the process of handling a reset,
-	 * the request is effectively complete as the rendering will
-	 * be discarded, but we need to return in order to drop the
-	 * struct_mutex.
-	 */
-	if (i915_reset_in_progress(&req->i915->gpu_error))
-		return true;
-
 	return false;
 }
 
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 51bb05f62686..cc565f785888 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -2800,7 +2800,8 @@ __i915_gem_object_sync(struct drm_i915_gem_request *to,
 
 	if (!i915.semaphores) {
 		ret = i915_wait_request(from,
-					from->i915->mm.interruptible,
+					from->i915->mm.interruptible |
+					I915_WAIT_LOCKED,
 					NULL,
 					NO_WAITBOOST);
 		if (ret)
diff --git a/drivers/gpu/drm/i915/i915_gem_request.c b/drivers/gpu/drm/i915/i915_gem_request.c
index 3c600ff3e9d5..10a11b7d1114 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.c
+++ b/drivers/gpu/drm/i915/i915_gem_request.c
@@ -666,8 +666,8 @@ int i915_wait_request(struct drm_i915_gem_request *req,
 	if (i915_spin_request(req, state, 5))
 		goto complete;
 
-	set_current_state(state);
-	add_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
+	if (flags & I915_WAIT_LOCKED)
+		add_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
 
 	intel_wait_init(&wait, req->fence.seqno);
 	if (intel_engine_add_wait(req->engine, &wait))
@@ -692,9 +692,9 @@ int i915_wait_request(struct drm_i915_gem_request *req,
 		if (intel_wait_complete(&wait))
 			break;
 
+wakeup:
 		set_current_state(state);
 
-wakeup:
 		/* Carefully check if the request is complete, giving time
 		 * for the seqno to be visible following the interrupt.
 		 * We also have to check in case we are kicked by the GPU
@@ -703,11 +703,32 @@ wakeup:
 		if (__i915_request_irq_complete(req))
 			break;
 
+		/* If the GPU is hung, and we hold the lock, reset the GPU
+		 * and then check for completion. On a full reset, the engine's
+		 * HW seqno will be advanced passed us and we are complete.
+		 * If we do a partial reset, we have to wait for the GPU to
+		 * resume and update the breadcrumb.
+		 *
+		 * If we don't hold the mutex, we can just wait for the worker
+		 * to come along and update the breadcrumb (either directly
+		 * itself, or indirectly by recovering the GPU).
+		 */
+		if (flags & I915_WAIT_LOCKED &&
+		    i915_reset_in_progress(&req->i915->gpu_error)) {
+			__set_current_state(TASK_RUNNING);
+			if (!i915_reset(req->i915))
+				goto wakeup;
+
+			break;
+		}
+
 		/* Only spin if we know the GPU is processing this request */
 		if (i915_spin_request(req, state, 2))
 			break;
 	}
-	remove_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
+
+	if (flags & I915_WAIT_LOCKED)
+		remove_wait_queue(&req->i915->gpu_error.wait_queue, &reset);
 
 	intel_engine_remove_wait(req->engine, &wait);
 	__set_current_state(TASK_RUNNING);
diff --git a/drivers/gpu/drm/i915/i915_gem_request.h b/drivers/gpu/drm/i915/i915_gem_request.h
index 41de4d9aa3fe..49c3c006ec17 100644
--- a/drivers/gpu/drm/i915/i915_gem_request.h
+++ b/drivers/gpu/drm/i915/i915_gem_request.h
@@ -223,6 +223,7 @@ int i915_wait_request(struct drm_i915_gem_request *req,
 		      struct intel_rps_client *rps)
 	__attribute__((nonnull(1)));
 #define I915_WAIT_INTERRUPTIBLE BIT(0)
+#define I915_WAIT_LOCKED BIT(1) /* struct_mutex held, handle GPU reset */
 
 static inline u32 intel_engine_get_seqno(struct intel_engine_cs *engine);
 
@@ -576,7 +577,9 @@ i915_gem_active_wait(const struct i915_gem_active *active, struct mutex *mutex)
 	if (!request)
 		return 0;
 
-	return i915_wait_request(request, I915_WAIT_INTERRUPTIBLE, NULL, NULL);
+	return i915_wait_request(request,
+				 I915_WAIT_INTERRUPTIBLE | I915_WAIT_LOCKED,
+				 NULL, NULL);
 }
 
 /**
@@ -639,7 +642,9 @@ i915_gem_active_retire(struct i915_gem_active *active,
 	if (!request)
 		return 0;
 
-	ret = i915_wait_request(request, I915_WAIT_INTERRUPTIBLE, NULL, NULL);
+	ret = i915_wait_request(request,
+				I915_WAIT_INTERRUPTIBLE | I915_WAIT_LOCKED,
+				NULL, NULL);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 3fd9cbdf2adb..0d1fb47f86c5 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -2521,7 +2521,9 @@ static void i915_reset_and_wakeup(struct drm_i915_private *dev_priv)
 	 * 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);
 
 	intel_finish_reset(dev_priv);
 
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index a6e6f2c49299..fa4f3731be29 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -2223,14 +2223,12 @@ static int wait_for_space(struct drm_i915_gem_request *req, int bytes)
 	if (WARN_ON(&target->ring_link == &ring->request_list))
 		return -ENOSPC;
 
-	ret = i915_wait_request(target, I915_WAIT_INTERRUPTIBLE,
+	ret = i915_wait_request(target,
+				I915_WAIT_INTERRUPTIBLE | I915_WAIT_LOCKED,
 				NULL, NO_WAITBOOST);
 	if (ret)
 		return ret;
 
-	if (i915_reset_in_progress(&target->i915->gpu_error))
-		return -EAGAIN;
-
 	i915_gem_request_retire_upto(target);
 
 	intel_ring_update_space(ring);
-- 
2.9.3

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx