[PATCH slof v2] virtio: Fix vring allocation

[Thomas Huth]

On Fri, 13 Mar 2015 20:42:03 +1100
Alexey Kardashevskiy <aik at ozlabs.ru> wrote:

> The value returned by virtio_vring_size() is used to allocate memory
> for vring. The used descriptor list (array of vring_used_elem) is
> counted by the header - vring_used struct - is not.
> 
> This fixes virtio_vring_size() to return the correct size.
> At the moment rings are quite small (256) and allocated with
> 4096 alignment, this is why we have not been having issues with
> this so far.
> 
> Signed-off-by: Alexey Kardashevskiy <aik at ozlabs.ru>
> ---
> Changes:
> v2:
> * remove magic numbers
> ---
>  lib/libvirtio/virtio.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/lib/libvirtio/virtio.c b/lib/libvirtio/virtio.c
> index b010796..f9c00a6 100644
> --- a/lib/libvirtio/virtio.c
> +++ b/lib/libvirtio/virtio.c
> @@ -32,8 +32,10 @@
>   */
>  unsigned long virtio_vring_size(unsigned int qsize)
>  {
> -	return VQ_ALIGN(sizeof(struct vring_desc) * qsize + 2 * (2 + qsize))
> -		+ VQ_ALIGN(sizeof(struct vring_used_elem) * qsize);
> +	return VQ_ALIGN(sizeof(struct vring_desc) * qsize +
> +                        sizeof(struct vring_avail) + sizeof(uint16_t) * qsize) +
> +               VQ_ALIGN(sizeof(struct vring_used) +
> +                        sizeof(struct vring_used_elem) * qsize);
>  }

Reviewed-by: Thomas Huth <thuth at linux.vnet.ibm.com>

That should be at least ok for the current status that we support in
SLOF. I think I might have written the original version of that
function with a very old version of the virtio spec (v0.8.10), when
ther was no "used_event" event field in the struct vring_avail defined
yet.
But if SLOF ever supports VIRTIO_F_EVENT_IDX, we've got to keep in mind
to add some additional bytes here, too.

 Thomas