virtio-rng: Avoid repeated init of completion

Commit Message

David Tolnay Jan. 7, 2019, 11:44 p.m. UTC

From: David Tolnay <[e-mail]>
Date: Mon, 7 Jan 2019 14:36:11 -0800
Subject: [PATCH] virtio-rng: Avoid repeated init of completion

The virtio-rng driver uses a completion called have_data to wait for a
virtio read to be fulfilled by the hypervisor. The completion is reset
before placing a buffer on the virtio queue and completed by the virtio
callback once data has been written into the buffer.

Prior to this commit, the driver called init_completion on this
completion both during probe as well as when registering virtio buffers
as part of a hwrng read operation. The second of these init_completion
calls should instead be reinit_completion because the have_data
completion has already been inited by probe. As described in
Documentation/scheduler/completion.txt, "Calling init_completion() twice
on the same completion object is most likely a bug".

This bug was present in the initial implementation of virtio-rng in
f7f510ec1957 ("virtio: An entropy device, as suggested by hpa"). Back
then the have_data completion was a single static completion rather than
a member of one of potentially multiple virtrng_info structs as
implemented later by 08e53fbdb85c ("virtio-rng: support multiple
virtio-rng devices"). The original driver incorrectly used
init_completion rather than INIT_COMPLETION to reset have_data during
read.

Tested by running `head -c48 /dev/random | hexdump` within crosvm, the
Chrome OS virtual machine monitor, and confirming that the virtio-rng
driver successfully produces random bytes from the host.

Signed-off-by: David Tolnay <[e-mail]>
Tested-by: David Tolnay <[e-mail]>
---
 drivers/char/hw_random/virtio-rng.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Herbert Xu Jan. 18, 2019, 10:43 a.m. UTC | #1

On Mon, Jan 07, 2019 at 03:44:38PM -0800, David Tolnay wrote:
> From: David Tolnay <[e-mail]>
> Date: Mon, 7 Jan 2019 14:36:11 -0800
> Subject: [PATCH] virtio-rng: Avoid repeated init of completion
> 
> The virtio-rng driver uses a completion called have_data to wait for a
> virtio read to be fulfilled by the hypervisor. The completion is reset
> before placing a buffer on the virtio queue and completed by the virtio
> callback once data has been written into the buffer.
> 
> Prior to this commit, the driver called init_completion on this
> completion both during probe as well as when registering virtio buffers
> as part of a hwrng read operation. The second of these init_completion
> calls should instead be reinit_completion because the have_data
> completion has already been inited by probe. As described in
> Documentation/scheduler/completion.txt, "Calling init_completion() twice
> on the same completion object is most likely a bug".
> 
> This bug was present in the initial implementation of virtio-rng in
> f7f510ec1957 ("virtio: An entropy device, as suggested by hpa"). Back
> then the have_data completion was a single static completion rather than
> a member of one of potentially multiple virtrng_info structs as
> implemented later by 08e53fbdb85c ("virtio-rng: support multiple
> virtio-rng devices"). The original driver incorrectly used
> init_completion rather than INIT_COMPLETION to reset have_data during
> read.
> 
> Tested by running `head -c48 /dev/random | hexdump` within crosvm, the
> Chrome OS virtual machine monitor, and confirming that the virtio-rng
> driver successfully produces random bytes from the host.
> 
> Signed-off-by: David Tolnay <[e-mail]>
> Tested-by: David Tolnay <[e-mail]>
> ---
>  drivers/char/hw_random/virtio-rng.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)

Thanks, but your patch doesn't apply because it is white-space damaged.

Please resubmit.

Patch

diff --git a/drivers/char/hw_random/virtio-rng.c
b/drivers/char/hw_random/virtio-rng.c
index b89df66ea1ae..7abd604e938c 100644
--- a/drivers/char/hw_random/virtio-rng.c
+++ b/drivers/char/hw_random/virtio-rng.c
@@ -73,7 +73,7 @@  static int virtio_read(struct hwrng *rng, void *buf,
size_t size, bool wait)

     if (!vi->busy) {
         vi->busy = true;
-        init_completion(&vi->have_data);
+        reinit_completion(&vi->have_data);
         register_buffer(vi, buf, size);
     }