All of lore.kernel.org
 help / color / mirror / Atom feed
From: j.glisse@gmail.com
To: dri-devel@lists.freedesktop.org
Cc: "Christian König" <deathsimple@vodafone.de>
Subject: [PATCH 05/27] drm/radeon: rework gpu lockup detection and processing
Date: Tue,  1 May 2012 13:19:15 -0400	[thread overview]
Message-ID: <1335892777-7357-6-git-send-email-j.glisse@gmail.com> (raw)
In-Reply-To: <1335892777-7357-1-git-send-email-j.glisse@gmail.com>

From: Christian König <deathsimple@vodafone.de>

Previusly multiple rings could trigger multiple GPU
resets at the same time.

Signed-off-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
---
 drivers/gpu/drm/radeon/radeon.h       |    3 +-
 drivers/gpu/drm/radeon/radeon_fence.c |  150 +++++++++++++++++----------------
 2 files changed, 77 insertions(+), 76 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 8801657..85a3aa9 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -255,8 +255,7 @@ struct radeon_fence_driver {
 	volatile uint32_t		*cpu_addr;
 	atomic_t			seq;
 	uint32_t			last_seq;
-	unsigned long			last_jiffies;
-	unsigned long			last_timeout;
+	unsigned long			last_activity;
 	wait_queue_head_t		queue;
 	struct list_head		created;
 	struct list_head		emitted;
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c b/drivers/gpu/drm/radeon/radeon_fence.c
index 36c411f..1a9765a 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -74,6 +74,10 @@ int radeon_fence_emit(struct radeon_device *rdev, struct radeon_fence *fence)
 	radeon_fence_ring_emit(rdev, fence->ring, fence);
 	trace_radeon_fence_emit(rdev->ddev, fence->seq);
 	fence->emitted = true;
+	/* are we the first fence on a previusly idle ring? */
+	if (list_empty(&rdev->fence_drv[fence->ring].emitted)) {
+		rdev->fence_drv[fence->ring].last_activity = jiffies;
+	}
 	list_move_tail(&fence->list, &rdev->fence_drv[fence->ring].emitted);
 	write_unlock_irqrestore(&rdev->fence_lock, irq_flags);
 	return 0;
@@ -85,34 +89,14 @@ static bool radeon_fence_poll_locked(struct radeon_device *rdev, int ring)
 	struct list_head *i, *n;
 	uint32_t seq;
 	bool wake = false;
-	unsigned long cjiffies;
 
 	seq = radeon_fence_read(rdev, ring);
-	if (seq != rdev->fence_drv[ring].last_seq) {
-		rdev->fence_drv[ring].last_seq = seq;
-		rdev->fence_drv[ring].last_jiffies = jiffies;
-		rdev->fence_drv[ring].last_timeout = RADEON_FENCE_JIFFIES_TIMEOUT;
-	} else {
-		cjiffies = jiffies;
-		if (time_after(cjiffies, rdev->fence_drv[ring].last_jiffies)) {
-			cjiffies -= rdev->fence_drv[ring].last_jiffies;
-			if (time_after(rdev->fence_drv[ring].last_timeout, cjiffies)) {
-				/* update the timeout */
-				rdev->fence_drv[ring].last_timeout -= cjiffies;
-			} else {
-				/* the 500ms timeout is elapsed we should test
-				 * for GPU lockup
-				 */
-				rdev->fence_drv[ring].last_timeout = 1;
-			}
-		} else {
-			/* wrap around update last jiffies, we will just wait
-			 * a little longer
-			 */
-			rdev->fence_drv[ring].last_jiffies = cjiffies;
-		}
+	if (seq == rdev->fence_drv[ring].last_seq)
 		return false;
-	}
+
+	rdev->fence_drv[ring].last_seq = seq;
+	rdev->fence_drv[ring].last_activity = jiffies;
+
 	n = NULL;
 	list_for_each(i, &rdev->fence_drv[ring].emitted) {
 		fence = list_entry(i, struct radeon_fence, list);
@@ -207,66 +191,84 @@ int radeon_fence_wait(struct radeon_fence *fence, bool intr)
 	struct radeon_device *rdev;
 	unsigned long irq_flags, timeout;
 	u32 seq;
-	int r;
+	int i, r;
+	bool signaled;
 
 	if (fence == NULL) {
 		WARN(1, "Querying an invalid fence : %p !\n", fence);
-		return 0;
+		return -EINVAL;
 	}
+
 	rdev = fence->rdev;
-	if (radeon_fence_signaled(fence)) {
-		return 0;
-	}
-	timeout = rdev->fence_drv[fence->ring].last_timeout;
-retry:
-	/* save current sequence used to check for GPU lockup */
-	seq = rdev->fence_drv[fence->ring].last_seq;
-	trace_radeon_fence_wait_begin(rdev->ddev, seq);
-	if (intr) {
+	signaled = radeon_fence_signaled(fence);
+	while (!signaled) {
+		read_lock_irqsave(&rdev->fence_lock, irq_flags);
+		timeout = jiffies - RADEON_FENCE_JIFFIES_TIMEOUT;
+		if (time_after(rdev->fence_drv[fence->ring].last_activity, timeout)) {
+			/* the normal case, timeout is somewhere before last_activity */
+			timeout = rdev->fence_drv[fence->ring].last_activity - timeout;
+		} else {
+			/* either jiffies wrapped around, or no fence was signaled in the last 500ms
+			 * anyway we will just wait for the minimum amount and then check for a lockup */
+			timeout = 1;
+		}
+		/* save current sequence value used to check for GPU lockups */
+		seq = rdev->fence_drv[fence->ring].last_seq;
+		read_unlock_irqrestore(&rdev->fence_lock, irq_flags);
+
+		trace_radeon_fence_wait_begin(rdev->ddev, seq);
 		radeon_irq_kms_sw_irq_get(rdev, fence->ring);
-		r = wait_event_interruptible_timeout(rdev->fence_drv[fence->ring].queue,
-				radeon_fence_signaled(fence), timeout);
+		if (intr) {
+			r = wait_event_interruptible_timeout(
+				rdev->fence_drv[fence->ring].queue,
+				(signaled = radeon_fence_signaled(fence)), timeout);
+		} else {
+			r = wait_event_timeout(
+				rdev->fence_drv[fence->ring].queue,
+				(signaled = radeon_fence_signaled(fence)), timeout);
+		}
 		radeon_irq_kms_sw_irq_put(rdev, fence->ring);
 		if (unlikely(r < 0)) {
 			return r;
 		}
-	} else {
-		radeon_irq_kms_sw_irq_get(rdev, fence->ring);
-		r = wait_event_timeout(rdev->fence_drv[fence->ring].queue,
-			 radeon_fence_signaled(fence), timeout);
-		radeon_irq_kms_sw_irq_put(rdev, fence->ring);
-	}
-	trace_radeon_fence_wait_end(rdev->ddev, seq);
-	if (unlikely(!radeon_fence_signaled(fence))) {
-		/* we were interrupted for some reason and fence isn't
-		 * isn't signaled yet, resume wait
-		 */
-		if (r) {
-			timeout = r;
-			goto retry;
-		}
-		/* don't protect read access to rdev->fence_drv[t].last_seq
-		 * if we experiencing a lockup the value doesn't change
-		 */
-		if (seq == rdev->fence_drv[fence->ring].last_seq &&
-		    radeon_ring_is_lockup(rdev, fence->ring, &rdev->ring[fence->ring])) {
-
-			/* good news we believe it's a lockup */
-			printk(KERN_WARNING "GPU lockup (waiting for 0x%08X last fence id 0x%08X)\n",
-			     fence->seq, seq);
-
-			/* mark the ring as not ready any more */
-			rdev->ring[fence->ring].ready = false;
-			r = radeon_gpu_reset(rdev);
-			if (r)
-				return r;
+		trace_radeon_fence_wait_end(rdev->ddev, seq);
+
+		if (unlikely(!signaled)) {
+			/* we were interrupted for some reason and fence
+			 * isn't signaled yet, resume waiting */
+			if (r) {
+				continue;
+			}
+
+			write_lock_irqsave(&rdev->fence_lock, irq_flags);
+			/* check if sequence value has changed since last_activity */
+			if (seq != rdev->fence_drv[fence->ring].last_seq) {
+				write_unlock_irqrestore(&rdev->fence_lock, irq_flags);
+				continue;
+			}
+
+			/* change sequence value on all rings, so nobody else things there is a lockup */
+			for (i = 0; i < RADEON_NUM_RINGS; ++i)
+				rdev->fence_drv[i].last_seq -= 0x10000;
+			write_unlock_irqrestore(&rdev->fence_lock, irq_flags);
+
+			if (radeon_ring_is_lockup(rdev, fence->ring, &rdev->ring[fence->ring])) {
+
+				/* good news we believe it's a lockup */
+				printk(KERN_WARNING "GPU lockup (waiting for 0x%08X last fence id 0x%08X)\n",
+				     fence->seq, seq);
+
+				/* mark the ring as not ready any more */
+				rdev->ring[fence->ring].ready = false;
+				r = radeon_gpu_reset(rdev);
+				if (r)
+					return r;
+
+				write_lock_irqsave(&rdev->fence_lock, irq_flags);
+				rdev->fence_drv[fence->ring].last_activity = jiffies;
+				write_unlock_irqrestore(&rdev->fence_lock, irq_flags);
+			}
 		}
-		timeout = RADEON_FENCE_JIFFIES_TIMEOUT;
-		write_lock_irqsave(&rdev->fence_lock, irq_flags);
-		rdev->fence_drv[fence->ring].last_timeout = RADEON_FENCE_JIFFIES_TIMEOUT;
-		rdev->fence_drv[fence->ring].last_jiffies = jiffies;
-		write_unlock_irqrestore(&rdev->fence_lock, irq_flags);
-		goto retry;
 	}
 	return 0;
 }
-- 
1.7.7.6

_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/dri-devel

  parent reply	other threads:[~2012-05-01 17:20 UTC|newest]

Thread overview: 30+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <·>
2012-05-01 17:19 ` Include request for reset-rework branch v2 j.glisse
2012-05-01 17:19   ` [PATCH 01/27] drm/radeon: make radeon_gpu_is_lockup a per ring function j.glisse
2012-05-01 17:19   ` [PATCH 02/27] drm/radeon: replace gpu_lockup with ring->ready flag j.glisse
2012-05-01 17:19   ` [PATCH 03/27] drm/radeon: register ring debugfs handlers on init j.glisse
2012-05-01 17:19   ` [PATCH 04/27] drm/radeon: use central function for IB testing j.glisse
2012-05-01 17:19   ` j.glisse [this message]
2012-05-01 17:19   ` [PATCH 06/27] drm/radeon: fix a bug in the SA code j.glisse
2012-05-01 17:19   ` [PATCH 07/27] drm/radeon: add proper locking to the SA v3 j.glisse
2012-05-02  8:21     ` Christian König
2012-05-01 17:19   ` [PATCH 08/27] drm/radeon: add sub allocator debugfs file v2 j.glisse
2012-05-01 17:19   ` [PATCH 09/27] drm/radeon: improve sa allocator j.glisse
2012-05-01 17:19   ` [PATCH 10/27] drm/radeon: sa allocator add wakequeue j.glisse
2012-05-01 17:19   ` [PATCH 11/27] drm/radeon: use inline functions to calc sa_bo addr j.glisse
2012-05-01 17:19   ` [PATCH 12/27] drm/radeon: simplify semaphore handling j.glisse
2012-05-01 17:19   ` [PATCH 13/27] drm/radeon: return -ENOENT in fence_wait_next v2 j.glisse
2012-05-01 17:19   ` [PATCH 14/27] drm/radeon: rename fence_wait_last to fence_wait_empty j.glisse
2012-05-01 17:19   ` [PATCH 15/27] drm/radeon: add general purpose fence signaled callback j.glisse
2012-05-01 17:19   ` [PATCH 16/27] drm/radeon: don't keep list of created fences j.glisse
2012-05-01 17:19   ` [PATCH 17/27] drm/radeon: add try_free callback to the SA v2 j.glisse
2012-05-01 17:19   ` [PATCH 18/27] drm/radeon: rip out the ib pool v3 j.glisse
2012-05-01 17:19   ` [PATCH 19/27] drm/radeon: fix a bug with the ring syncing code j.glisse
2012-05-01 17:19   ` [PATCH 20/27] drm/radeon: rework recursive gpu reset handling j.glisse
2012-05-01 17:19   ` [PATCH 21/27] drm/radeon: remove recursive mutex implementation j.glisse
2012-05-01 17:19   ` [PATCH 22/27] drm/radeon: move lockup detection code into radeon_ring.c j.glisse
2012-05-01 17:19   ` [PATCH 23/27] drm/radeon: make lockup timeout a module param j.glisse
2012-05-01 17:19   ` [PATCH 24/27] drm/radeon: unlock the ring mutex while waiting for the next fence j.glisse
2012-05-01 17:19   ` [PATCH 25/27] drm/radeon: make forcing ring activity a common function j.glisse
2012-05-01 17:19   ` [PATCH 26/27] drm/radeon: remove r300_gpu_is_lockup j.glisse
2012-05-01 17:19   ` [PATCH 27/27] drm/radeon: remove cayman_gpu_is_lockup j.glisse
2012-05-01 17:28   ` Include request for reset-rework branch v2 Jerome Glisse

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1335892777-7357-6-git-send-email-j.glisse@gmail.com \
    --to=j.glisse@gmail.com \
    --cc=deathsimple@vodafone.de \
    --cc=dri-devel@lists.freedesktop.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body. 
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.