[4/4] drm/virtio: Add memory barriers for capset cache.

Submitted by David Riley on June 5, 2019, 11:44 p.m.

Details

Message ID 20190605234423.11348-4-davidriley@chromium.org
State New
Headers show
Series "Series without cover letter" ( rev: 2 1 ) in DRI devel

Not browsing as part of any series.

Commit Message

David Riley June 5, 2019, 11:44 p.m.
From: David Riley <davidriley@chromium.org>

After data is copied to the cache entry, atomic_set is used indicate
that the data is the entry is valid without appropriate memory barriers.
Similarly the read side was missing the same memory barries.

Signed-off-by: David Riley <davidriley@chromium.org>
---
 drivers/gpu/drm/virtio/virtgpu_ioctl.c | 3 +++
 drivers/gpu/drm/virtio/virtgpu_vq.c    | 2 ++
 2 files changed, 5 insertions(+)

Patch hide | download patch | download mbox

diff --git a/drivers/gpu/drm/virtio/virtgpu_ioctl.c b/drivers/gpu/drm/virtio/virtgpu_ioctl.c
index 88c1ed57a3c5..502f5f7c2298 100644
--- a/drivers/gpu/drm/virtio/virtgpu_ioctl.c
+++ b/drivers/gpu/drm/virtio/virtgpu_ioctl.c
@@ -542,6 +542,9 @@  static int virtio_gpu_get_caps_ioctl(struct drm_device *dev,
 	if (!ret)
 		return -EBUSY;
 
+	/* is_valid check must proceed before copy of the cache entry. */
+	virt_rmb();
+
 	ptr = cache_ent->caps_cache;
 
 	if (copy_to_user((void __user *)(unsigned long)args->addr, ptr, size))
diff --git a/drivers/gpu/drm/virtio/virtgpu_vq.c b/drivers/gpu/drm/virtio/virtgpu_vq.c
index dd5ead2541c2..b974eba4fe7d 100644
--- a/drivers/gpu/drm/virtio/virtgpu_vq.c
+++ b/drivers/gpu/drm/virtio/virtgpu_vq.c
@@ -583,6 +583,8 @@  static void virtio_gpu_cmd_capset_cb(struct virtio_gpu_device *vgdev,
 		    cache_ent->id == le32_to_cpu(cmd->capset_id)) {
 			memcpy(cache_ent->caps_cache, resp->capset_data,
 			       cache_ent->size);
+			/* Copy must occur before is_valid is signalled. */
+			virt_wmb();
 			atomic_set(&cache_ent->is_valid, 1);
 			break;
 		}

Comments

On Wed, Jun 05, 2019 at 04:44:23PM -0700, davidriley@chromium.org wrote:
> From: David Riley <davidriley@chromium.org>
> 
> After data is copied to the cache entry, atomic_set is used indicate
> that the data is the entry is valid without appropriate memory barriers.
> Similarly the read side was missing the same memory barries.
> 
> Signed-off-by: David Riley <davidriley@chromium.org>
> ---
>  drivers/gpu/drm/virtio/virtgpu_ioctl.c | 3 +++
>  drivers/gpu/drm/virtio/virtgpu_vq.c    | 2 ++
>  2 files changed, 5 insertions(+)
> 
> diff --git a/drivers/gpu/drm/virtio/virtgpu_ioctl.c b/drivers/gpu/drm/virtio/virtgpu_ioctl.c
> index 88c1ed57a3c5..502f5f7c2298 100644
> --- a/drivers/gpu/drm/virtio/virtgpu_ioctl.c
> +++ b/drivers/gpu/drm/virtio/virtgpu_ioctl.c
> @@ -542,6 +542,9 @@ static int virtio_gpu_get_caps_ioctl(struct drm_device *dev,
>  	if (!ret)
>  		return -EBUSY;
>  
> +	/* is_valid check must proceed before copy of the cache entry. */
> +	virt_rmb();

I don't think you need virt_rmb() here.  This isn't guest <=> host
communication, so a normal barrier should do.

The other three fixes are queued up for drm-misc-next.

cheers,
  Gerd