diff mbox series

[v2] drm/imagination: avoid deadlock on fence release

Message ID 20250226-fence-release-deadlock-v2-1-6fed2fc1fe88@imgtec.com (mailing list archive)
State New, archived
Headers show
Series [v2] drm/imagination: avoid deadlock on fence release | expand

Commit Message

Brendan King via B4 Relay Feb. 26, 2025, 3:42 p.m. UTC 
From: Brendan King <Brendan.King@imgtec.com>

Do scheduler queue fence release processing on a workqueue, rather
than in the release function itself.

Fixes deadlock issues such as the following:

[  607.400437] ============================================
[  607.405755] WARNING: possible recursive locking detected
[  607.415500] --------------------------------------------
[  607.420817] weston:zfq0/24149 is trying to acquire lock:
[  607.426131] ffff000017d041a0 (reservation_ww_class_mutex){+.+.}-{3:3}, at: pvr_gem_object_vunmap+0x40/0xc0 [powervr]
[  607.436728]
               but task is already holding lock:
[  607.442554] ffff000017d105a0 (reservation_ww_class_mutex){+.+.}-{3:3}, at: dma_buf_ioctl+0x250/0x554
[  607.451727]
               other info that might help us debug this:
[  607.458245]  Possible unsafe locking scenario:

[  607.464155]        CPU0
[  607.466601]        ----
[  607.469044]   lock(reservation_ww_class_mutex);
[  607.473584]   lock(reservation_ww_class_mutex);
[  607.478114]
                *** DEADLOCK ***

Cc: stable@vger.kernel.org
Fixes: eaf01ee5ba28 ("drm/imagination: Implement job submission and scheduling")
Signed-off-by: Brendan King <brendan.king@imgtec.com>
---
Changes in v2:
- Added 'Cc:' and 'Fixes:' tags
- Link to v1: https://lore.kernel.org/r/20250203-fence-release-deadlock-v1-1-6e1de1260b38@imgtec.com
---
 drivers/gpu/drm/imagination/pvr_queue.c | 13 +++++++++++--
 drivers/gpu/drm/imagination/pvr_queue.h |  4 ++++
 2 files changed, 15 insertions(+), 2 deletions(-)


---
base-commit: 3ab334814dc7dff39075e055e12847d51878916e
change-id: 20250203-fence-release-deadlock-a0753c07bfdd

Best regards,

Comments

Matt Coster March 3, 2025, 12:14 p.m. UTC | #1 
On Wed, 26 Feb 2025 15:42:19 +0000, Brendan King wrote:
> Do scheduler queue fence release processing on a workqueue, rather
> than in the release function itself.
> 
> Fixes deadlock issues such as the following:
> 
> [  607.400437] ============================================
> [  607.405755] WARNING: possible recursive locking detected
> [  607.415500] --------------------------------------------
> [  607.420817] weston:zfq0/24149 is trying to acquire lock:
> [  607.426131] ffff000017d041a0 (reservation_ww_class_mutex){+.+.}-{3:3}, at: pvr_gem_object_vunmap+0x40/0xc0 [powervr]
> [  607.436728]
>                but task is already holding lock:
> [  607.442554] ffff000017d105a0 (reservation_ww_class_mutex){+.+.}-{3:3}, at: dma_buf_ioctl+0x250/0x554
> [  607.451727]
>                other info that might help us debug this:
> [  607.458245]  Possible unsafe locking scenario:
> 
> [...]

Applied, thanks!

[1/1] drm/imagination: avoid deadlock on fence release
      commit: df1a1ed5e1bdd9cc13148e0e5549f5ebcf76cf13

Best regards,
diff mbox series

Patch

diff --git a/drivers/gpu/drm/imagination/pvr_queue.c b/drivers/gpu/drm/imagination/pvr_queue.c
index c4f08432882b12f5cdfeb7fc991fd941f0946676..f3f1c5212df7432161919ddc510cececacbbe143 100644
--- a/drivers/gpu/drm/imagination/pvr_queue.c
+++ b/drivers/gpu/drm/imagination/pvr_queue.c
@@ -109,12 +109,20 @@  pvr_queue_fence_get_driver_name(struct dma_fence *f)
 	return PVR_DRIVER_NAME;
 }
 
+static void pvr_queue_fence_release_work(struct work_struct *w)
+{
+	struct pvr_queue_fence *fence = container_of(w, struct pvr_queue_fence, release_work);
+
+	pvr_context_put(fence->queue->ctx);
+	dma_fence_free(&fence->base);
+}
+
 static void pvr_queue_fence_release(struct dma_fence *f)
 {
 	struct pvr_queue_fence *fence = container_of(f, struct pvr_queue_fence, base);
+	struct pvr_device *pvr_dev = fence->queue->ctx->pvr_dev;
 
-	pvr_context_put(fence->queue->ctx);
-	dma_fence_free(f);
+	queue_work(pvr_dev->sched_wq, &fence->release_work);
 }
 
 static const char *
@@ -268,6 +276,7 @@  pvr_queue_fence_init(struct dma_fence *f,
 
 	pvr_context_get(queue->ctx);
 	fence->queue = queue;
+	INIT_WORK(&fence->release_work, pvr_queue_fence_release_work);
 	dma_fence_init(&fence->base, fence_ops,
 		       &fence_ctx->lock, fence_ctx->id,
 		       atomic_inc_return(&fence_ctx->seqno));
diff --git a/drivers/gpu/drm/imagination/pvr_queue.h b/drivers/gpu/drm/imagination/pvr_queue.h
index e06ced69302fca47fc26451dfb09ebbb24b57f52..93fe9ac9f58ccc020615485e86be438548dcee43 100644
--- a/drivers/gpu/drm/imagination/pvr_queue.h
+++ b/drivers/gpu/drm/imagination/pvr_queue.h
@@ -5,6 +5,7 @@ 
 #define PVR_QUEUE_H
 
 #include <drm/gpu_scheduler.h>
+#include <linux/workqueue.h>
 
 #include "pvr_cccb.h"
 #include "pvr_device.h"
@@ -63,6 +64,9 @@  struct pvr_queue_fence {
 
 	/** @queue: Queue that created this fence. */
 	struct pvr_queue *queue;
+
+	/** @release_work: Fence release work structure. */
+	struct work_struct release_work;
 };
 
 /**