Linux Virtualization - Aug 2014 - [3.16 stable PATCH 1/1] virtio-rng: fix multi-device startup

This is a 3.16-only patch.  The linux.git fix is
5c06273401f2eb7b290cadbae18ee00f8f65e893, which fixes this issue in a
different way.

Amos reported using multiple virtio-rng devices cause boot to freeze
even after the recent fixes.  We need the probe_done test per device
rather than just once for the driver.

Reported-by: Amos Kong <akong at redhat.com>
Signed-off-by: Amit Shah <amit.shah at redhat.com>
---
 drivers/char/hw_random/virtio-rng.c | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/char/hw_random/virtio-rng.c
b/drivers/char/hw_random/virtio-rng.c
index e9b15bc..124cac5 100644
--- a/drivers/char/hw_random/virtio-rng.c
+++ b/drivers/char/hw_random/virtio-rng.c
@@ -34,12 +34,11 @@ struct virtrng_info {
 	unsigned int data_avail;
 	struct completion have_data;
 	bool busy;
+	bool probe_done;
 	char name[25];
 	int index;
 };
 
-static bool probe_done;
-
 static void random_recv_done(struct virtqueue *vq)
 {
 	struct virtrng_info *vi = vq->vdev->priv;
@@ -73,7 +72,7 @@ static int virtio_read(struct hwrng *rng, void *buf, size_t
size, bool wait)
 	 * Don't ask host for data till we're setup.  This call can
 	 * happen during hwrng_register(), after commit d9e7972619.
 	 */
-	if (unlikely(!probe_done))
+	if (unlikely(!vi->probe_done))
 		return 0;
 
 	if (!vi->busy) {
@@ -146,7 +145,7 @@ static int probe_common(struct virtio_device *vdev)
 		return err;
 	}
 
-	probe_done = true;
+	vi->probe_done = true;
 	return 0;
 }
 
-- 
1.9.3

On Mon, Aug 11, 2014 at 06:11:47PM +0530, Amit Shah
wrote:
> This is a 3.16-only patch.  The linux.git fix is
> 5c06273401f2eb7b290cadbae18ee00f8f65e893, which fixes this issue in a
> different way.
Why "different"?  Why can't I take that original patch instead? 
What is
different in this patch, and why?

thanks,

greg k-h

On (Tue) 12 Aug 2014 [06:55:27], Greg KH wrote:
> On Mon, Aug 11, 2014 at 06:11:47PM +0530, Amit Shah wrote:
> > This is a 3.16-only patch.  The linux.git fix is
> > 5c06273401f2eb7b290cadbae18ee00f8f65e893, which fixes this issue in a
> > different way.
> 
> Why "different"?  Why can't I take that original patch
instead?  What is
> different in this patch, and why?
The commit referenced moves the hwrng_register() call to the ->scan()
callback instead of it being in probe().  This was done to ensure the
virtio-rng devices can contribute to the initial system entropy
introduced in commit d9e7972619334.

That patch is quite small too, but will need a slight conflict
resolution due to the previous two code-shuffling patches, and also
the following revert.

However, I decided against the backport of the ->scan() method, since
it wasn't designed to solve this regression, it happens to solve it,
and it actually introduces new functionality.  I would be happy to
provide a backport of the relevant patches, if you think that would be
alright.

Thanks,

		Amit