drm/i915: Split I915_RESET_IN_PROGRESS into two flags

I915_RESET_IN_PROGRESS is being used for both signaling the requirement
to i915_mutex_lock_interruptible() to avoid taking the struct_mutex and
to instruct a waiter (already holding the struct_mutex) to perform the
reset. To allow for a little more coordination, split these two meaning
into a couple of distinct flags. I915_RESET_BACKOFF tells
i915_mutex_lock_interruptible() not to acquire the mutex and
I915_RESET_HANDOFF tells the waiter to call i915_reset().
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170316171305.12972-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/i915_debugfs.c

@@ -1305,16 +1305,18 @@ static int i915_hangcheck_info(struct seq_file *m, void *unused)
 	enum intel_engine_id id;
 
 	if (test_bit(I915_WEDGED, &dev_priv->gpu_error.flags))
-		seq_printf(m, "Wedged\n");
-	if (test_bit(I915_RESET_IN_PROGRESS, &dev_priv->gpu_error.flags))
-		seq_printf(m, "Reset in progress\n");
+		seq_puts(m, "Wedged\n");
+	if (test_bit(I915_RESET_BACKOFF, &dev_priv->gpu_error.flags))
+		seq_puts(m, "Reset in progress: struct_mutex backoff\n");
+	if (test_bit(I915_RESET_HANDOFF, &dev_priv->gpu_error.flags))
+		seq_puts(m, "Reset in progress: reset handoff to waiter\n");
 	if (waitqueue_active(&dev_priv->gpu_error.wait_queue))
-		seq_printf(m, "Waiter holding struct mutex\n");
+		seq_puts(m, "Waiter holding struct mutex\n");
 	if (waitqueue_active(&dev_priv->gpu_error.reset_queue))
-		seq_printf(m, "struct_mutex blocked for reset\n");
+		seq_puts(m, "struct_mutex blocked for reset\n");
 
 	if (!i915.enable_hangcheck) {
-		seq_printf(m, "Hangcheck disabled\n");
+		seq_puts(m, "Hangcheck disabled\n");
 		return 0;
 	}
 
@@ -4127,7 +4129,7 @@ i915_wedged_set(void *data, u64 val)
 	 * while it is writing to 'i915_wedged'
 	 */
 
-	if (i915_reset_in_progress(&dev_priv->gpu_error))
+	if (i915_reset_backoff(&dev_priv->gpu_error))
 		return -EAGAIN;
 
 	i915_handle_error(dev_priv, val,

drivers/gpu/drm/i915/i915_drv.c

@@ -1815,8 +1815,9 @@ void i915_reset(struct drm_i915_private *dev_priv)
 	int ret;
 
 	lockdep_assert_held(&dev_priv->drm.struct_mutex);
+	GEM_BUG_ON(!test_bit(I915_RESET_BACKOFF, &error->flags));
 
-	if (!test_and_clear_bit(I915_RESET_IN_PROGRESS, &error->flags))
+	if (!test_bit(I915_RESET_HANDOFF, &error->flags))
 		return;
 
 	/* Clear any previous failed attempts at recovery. Time to try again. */
@@ -1869,7 +1870,9 @@ void i915_reset(struct drm_i915_private *dev_priv)
 wakeup:
 	i915_gem_reset_finish(dev_priv);
 	enable_irq(dev_priv->drm.irq);
-	wake_up_bit(&error->flags, I915_RESET_IN_PROGRESS);
+
+	clear_bit(I915_RESET_HANDOFF, &error->flags);
+	wake_up_bit(&error->flags, I915_RESET_HANDOFF);
 	return;
 
 error:

drivers/gpu/drm/i915/i915_drv.h

@@ -1595,8 +1595,33 @@ struct i915_gpu_error {
 	 */
 	unsigned long reset_count;
 
+	/**
+	 * flags: Control various stages of the GPU reset
+	 *
+	 * #I915_RESET_BACKOFF - When we start a reset, we want to stop any
+	 * other users acquiring the struct_mutex. To do this we set the
+	 * #I915_RESET_BACKOFF bit in the error flags when we detect a reset
+	 * and then check for that bit before acquiring the struct_mutex (in
+	 * i915_mutex_lock_interruptible()?). I915_RESET_BACKOFF serves a
+	 * secondary role in preventing two concurrent global reset attempts.
+	 *
+	 * #I915_RESET_HANDOFF - To perform the actual GPU reset, we need the
+	 * struct_mutex. We try to acquire the struct_mutex in the reset worker,
+	 * but it may be held by some long running waiter (that we cannot
+	 * interrupt without causing trouble). Once we are ready to do the GPU
+	 * reset, we set the I915_RESET_HANDOFF bit and wakeup any waiters. If
+	 * they already hold the struct_mutex and want to participate they can
+	 * inspect the bit and do the reset directly, otherwise the worker
+	 * waits for the struct_mutex.
+	 *
+	 * #I915_WEDGED - If reset fails and we can no longer use the GPU,
+	 * we set the #I915_WEDGED bit. Prior to command submission, e.g.
+	 * i915_gem_request_alloc(), this bit is checked and the sequence
+	 * aborted (with -EIO reported to userspace) if set.
+	 */
 	unsigned long flags;
-#define I915_RESET_IN_PROGRESS	0
+#define I915_RESET_BACKOFF	0
+#define I915_RESET_HANDOFF	1
 #define I915_WEDGED		(BITS_PER_LONG - 1)
 
 	/**
@@ -3387,9 +3412,14 @@ i915_gem_find_active_request(struct intel_engine_cs *engine);
 
 void i915_gem_retire_requests(struct drm_i915_private *dev_priv);
 
-static inline bool i915_reset_in_progress(struct i915_gpu_error *error)
+static inline bool i915_reset_backoff(struct i915_gpu_error *error)
+{
+	return unlikely(test_bit(I915_RESET_BACKOFF, &error->flags));
+}
+
+static inline bool i915_reset_handoff(struct i915_gpu_error *error)
 {
-	return unlikely(test_bit(I915_RESET_IN_PROGRESS, &error->flags));
+	return unlikely(test_bit(I915_RESET_HANDOFF, &error->flags));
 }
 
 static inline bool i915_terminally_wedged(struct i915_gpu_error *error)
@@ -3397,9 +3427,9 @@ static inline bool i915_terminally_wedged(struct i915_gpu_error *error)
 	return unlikely(test_bit(I915_WEDGED, &error->flags));
 }
 
-static inline bool i915_reset_in_progress_or_wedged(struct i915_gpu_error *error)
+static inline bool i915_reset_backoff_or_wedged(struct i915_gpu_error *error)
 {
-	return i915_reset_in_progress(error) | i915_terminally_wedged(error);
+	return i915_reset_backoff(error) | i915_terminally_wedged(error);
 }
 
 static inline u32 i915_reset_count(struct i915_gpu_error *error)

drivers/gpu/drm/i915/i915_gem.c

@@ -103,16 +103,13 @@ i915_gem_wait_for_error(struct i915_gpu_error *error)
 
 	might_sleep();
 
-	if (!i915_reset_in_progress(error))
-		return 0;
-
 	/*
 	 * Only wait 10 seconds for the gpu reset to complete to avoid hanging
 	 * userspace. If it takes that long something really bad is going on and
 	 * we should simply try to bail out and fail as gracefully as possible.
 	 */
 	ret = wait_event_interruptible_timeout(error->reset_queue,
-					       !i915_reset_in_progress(error),
+					       !i915_reset_backoff(error),
 					       I915_RESET_TIMEOUT);
 	if (ret == 0) {
 		DRM_ERROR("Timed out waiting for the gpu reset to complete\n");

drivers/gpu/drm/i915/i915_gem_request.c

@@ -1012,7 +1012,7 @@ bool __i915_spin_request(const struct drm_i915_gem_request *req,
 
 static bool __i915_wait_request_check_and_reset(struct drm_i915_gem_request *request)
 {
-	if (likely(!i915_reset_in_progress(&request->i915->gpu_error)))
+	if (likely(!i915_reset_handoff(&request->i915->gpu_error)))
 		return false;
 
 	__set_current_state(TASK_RUNNING);

drivers/gpu/drm/i915/i915_irq.c

@@ -2631,22 +2631,6 @@ static irqreturn_t gen8_irq_handler(int irq, void *arg)
 	return ret;
 }
 
-static void i915_error_wake_up(struct drm_i915_private *dev_priv)
-{
-	/*
-	 * Notify all waiters for GPU completion events that reset state has
-	 * been changed, and that they need to restart their wait after
-	 * checking for potential errors (and bail out to drop locks if there is
-	 * a gpu reset pending so that i915_error_work_func can acquire them).
-	 */
-
-	/* Wake up __wait_seqno, potentially holding dev->struct_mutex. */
-	wake_up_all(&dev_priv->gpu_error.wait_queue);
-
-	/* Wake up intel_crtc_wait_for_pending_flips, holding crtc->mutex. */
-	wake_up_all(&dev_priv->pending_flip_queue);
-}
-
 /**
  * i915_reset_and_wakeup - do process context error handling work
  * @dev_priv: i915 device private
@@ -2668,6 +2652,9 @@ static void i915_reset_and_wakeup(struct drm_i915_private *dev_priv)
 
 	intel_prepare_reset(dev_priv);
 
+	set_bit(I915_RESET_HANDOFF, &dev_priv->gpu_error.flags);
+	wake_up_all(&dev_priv->gpu_error.wait_queue);
+
 	do {
 		/*
 		 * All state reset _must_ be completed before we update the
@@ -2682,7 +2669,7 @@ static void i915_reset_and_wakeup(struct drm_i915_private *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,
+				     I915_RESET_HANDOFF,
 				     TASK_UNINTERRUPTIBLE,
 				     HZ));
 
@@ -2696,6 +2683,7 @@ static void i915_reset_and_wakeup(struct drm_i915_private *dev_priv)
 	 * Note: The wake_up also serves as a memory barrier so that
 	 * waiters see the updated value of the dev_priv->gpu_error.
 	 */
+	clear_bit(I915_RESET_BACKOFF, &dev_priv->gpu_error.flags);
 	wake_up_all(&dev_priv->gpu_error.reset_queue);
 }
 
@@ -2788,24 +2776,10 @@ void i915_handle_error(struct drm_i915_private *dev_priv,
 	if (!engine_mask)
 		goto out;
 
-	if (test_and_set_bit(I915_RESET_IN_PROGRESS,
+	if (test_and_set_bit(I915_RESET_BACKOFF,
 			     &dev_priv->gpu_error.flags))
 		goto out;
 
-	/*
-	 * Wakeup waiting processes so that the reset function
-	 * i915_reset_and_wakeup doesn't deadlock trying to grab
-	 * various locks. By bumping the reset counter first, the woken
-	 * processes will see a reset in progress and back off,
-	 * releasing their locks and then wait for the reset completion.
-	 * We must do this for _all_ gpu waiters that might hold locks
-	 * that the reset work needs to acquire.
-	 *
-	 * Note: The wake_up also provides a memory barrier to ensure that the
-	 * waiters see the updated value of the reset flags.
-	 */
-	i915_error_wake_up(dev_priv);
-
 	i915_reset_and_wakeup(dev_priv);
 
 out:

drivers/gpu/drm/i915/intel_display.c

@@ -3639,7 +3639,7 @@ static bool abort_flip_on_reset(struct intel_crtc *crtc)
 {
 	struct i915_gpu_error *error = &to_i915(crtc->base.dev)->gpu_error;
 
-	if (i915_reset_in_progress(error))
+	if (i915_reset_backoff(error))
 		return true;
 
 	if (crtc->reset_count != i915_reset_count(error))
@@ -10595,7 +10595,7 @@ static int intel_crtc_page_flip(struct drm_crtc *crtc,
 		goto cleanup;
 
 	intel_crtc->reset_count = i915_reset_count(&dev_priv->gpu_error);
-	if (i915_reset_in_progress_or_wedged(&dev_priv->gpu_error)) {
+	if (i915_reset_backoff_or_wedged(&dev_priv->gpu_error)) {
 		ret = -EIO;
 		goto unlock;
 	}

drivers/gpu/drm/i915/selftests/intel_hangcheck.c

@@ -301,7 +301,8 @@ static int igt_global_reset(void *arg)
 
 	/* Check that we can issue a global GPU reset */
 
-	set_bit(I915_RESET_IN_PROGRESS, &i915->gpu_error.flags);
+	set_bit(I915_RESET_BACKOFF, &i915->gpu_error.flags);
+	set_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags);
 
 	mutex_lock(&i915->drm.struct_mutex);
 	reset_count = i915_reset_count(&i915->gpu_error);
@@ -314,7 +315,8 @@ static int igt_global_reset(void *arg)
 	}
 	mutex_unlock(&i915->drm.struct_mutex);
 
-	GEM_BUG_ON(test_bit(I915_RESET_IN_PROGRESS, &i915->gpu_error.flags));
+	GEM_BUG_ON(test_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags));
+	clear_bit(I915_RESET_BACKOFF, &i915->gpu_error.flags);
 	if (i915_terminally_wedged(&i915->gpu_error))
 		err = -EIO;
 
@@ -330,7 +332,7 @@ static u32 fake_hangcheck(struct drm_i915_gem_request *rq)
 
 	reset_count = i915_reset_count(&rq->i915->gpu_error);
 
-	set_bit(I915_RESET_IN_PROGRESS, &rq->i915->gpu_error.flags);
+	set_bit(I915_RESET_HANDOFF, &rq->i915->gpu_error.flags);
 	wake_up_all(&rq->i915->gpu_error.wait_queue);
 
 	return reset_count;
@@ -357,7 +359,7 @@ static int igt_wait_reset(void *arg)
 
 	/* Check that we detect a stuck waiter and issue a reset */
 
-	set_bit(I915_RESET_IN_PROGRESS, &i915->gpu_error.flags);
+	set_bit(I915_RESET_BACKOFF, &i915->gpu_error.flags);
 
 	mutex_lock(&i915->drm.struct_mutex);
 	err = hang_init(&h, i915);
@@ -388,8 +390,8 @@ static int igt_wait_reset(void *arg)
 		err = timeout;
 		goto out_rq;
 	}
-	GEM_BUG_ON(test_bit(I915_RESET_IN_PROGRESS, &i915->gpu_error.flags));
 
+	GEM_BUG_ON(test_bit(I915_RESET_HANDOFF, &i915->gpu_error.flags));
 	if (i915_reset_count(&i915->gpu_error) == reset_count) {
 		pr_err("No GPU reset recorded!\n");
 		err = -EINVAL;
@@ -402,6 +404,7 @@ static int igt_wait_reset(void *arg)
 	hang_fini(&h);
 unlock:
 	mutex_unlock(&i915->drm.struct_mutex);
+	clear_bit(I915_RESET_BACKOFF, &i915->gpu_error.flags);
 
 	if (i915_terminally_wedged(&i915->gpu_error))
 		return -EIO;
@@ -422,6 +425,7 @@ static int igt_reset_queue(void *arg)
 	if (!igt_can_mi_store_dword_imm(i915))
 		return 0;
 
+	set_bit(I915_RESET_BACKOFF, &i915->gpu_error.flags);
 	mutex_lock(&i915->drm.struct_mutex);
 	err = hang_init(&h, i915);
 	if (err)
@@ -470,8 +474,9 @@ static int igt_reset_queue(void *arg)
 
 			i915_reset(i915);
 
-			GEM_BUG_ON(test_bit(I915_RESET_IN_PROGRESS,
+			GEM_BUG_ON(test_bit(I915_RESET_HANDOFF,
 					    &i915->gpu_error.flags));
+
 			if (prev->fence.error != -EIO) {
 				pr_err("GPU reset not recorded on hanging request [fence.error=%d]!\n",
 				       prev->fence.error);
@@ -514,6 +519,7 @@ static int igt_reset_queue(void *arg)
 	hang_fini(&h);
 unlock:
 	mutex_unlock(&i915->drm.struct_mutex);
+	clear_bit(I915_RESET_BACKOFF, &i915->gpu_error.flags);
 
 	if (i915_terminally_wedged(&i915->gpu_error))
 		return -EIO;