2020-07-20 14:46:01

by 何鑫

Subject: [PATCH v2] drm/virtio: fixed memory leak in virtio_gpu_execbuffer_ioctl()

commit 2cd7b6f08bc4 ("drm/virtio: add in/out fence support
for explicit synchronization") deleted dma_fence_put(),
which will cause the memory leak. To balance the reference
count initialized when allocating fence, dma_fence_put()
should not be deleted.

Fixes: 2cd7b6f08bc4 ("drm/virtio: add in/out fence support for explicit synchronization")
Signed-off-by: Liu Qi <[email protected]>
Signed-off-by: hexin <[email protected]>
---
drivers/gpu/drm/virtio/virtgpu_ioctl.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/drivers/gpu/drm/virtio/virtgpu_ioctl.c b/drivers/gpu/drm/virtio/virtgpu_ioctl.c
index 5df722072ba0..19c5bc01eb79 100644
--- a/drivers/gpu/drm/virtio/virtgpu_ioctl.c
+++ b/drivers/gpu/drm/virtio/virtgpu_ioctl.c
@@ -179,6 +179,7 @@ static int virtio_gpu_execbuffer_ioctl(struct drm_device *dev, void *data,

virtio_gpu_cmd_submit(vgdev, buf, exbuf->size,
vfpriv->ctx_id, buflist, out_fence);
+ dma_fence_put(&out_fence->f);
virtio_gpu_notify(vgdev);
return 0;

--
2.21.1 (Apple Git-122.3)

2020-07-20 15:29:48

by Markus Elfring

Subject: Re: [PATCH v2] drm/virtio: Fix memory leak in virtio_gpu_execbuffer_ioctl()

> … To balance the reference
> count initialized when allocating fence, dma_fence_put()
> should not be deleted.

* Would an imperative wording be more appropriate for the change description?

* Is the information “hexin” sufficient for a real name?

Regards,
Markus