Re: [RFC PATCH 3/4] vhost: byteswap virtqueue attributes

[Cornelia Huck <cornelia.huck@de.ibm.com>]

On Wed, 29 Oct 2014 09:38:44 +0100
Cédric Le Goater <clg@fr.ibm.com> wrote:

> The virtqueue structure shares a few attributes with the guest OS
> which need to be byteswapped when the endian order of the host is
> different.
> 
> This patch uses the vq->byteswap attribute to decide whether to
> byteswap or not data being accessed in the guest memory.
> 
> Signed-off-by: Cédric Le Goater <clg@fr.ibm.com>
> ---
>  drivers/vhost/vhost.c |   52 +++++++++++++++++++++++++++++++++++++------------
>  1 file changed, 40 insertions(+), 12 deletions(-)
> 

> +static int __copyhead_to_user(struct vhost_virtqueue *vq,
> +			struct vring_used_elem *heads,
> +			struct vring_used_elem __user *used,
> +			unsigned count)

__copy_used_elems_to_user() ?

> +{
> +	int i;
> +
> +	for (i = 0; i < count; i++) {
> +		if (__vq_put_user(vq, heads[i].id, &used[i].id)) {
> +			vq_err(vq, "Failed to write used id");
> +			return -EFAULT;
> +		}
> +		if (__vq_put_user(vq, heads[i].len, &used[i].len)) {
> +			vq_err(vq, "Failed to write used len");
> +			return -EFAULT;
> +		}
> +	}

Is there a number of elements where it would be more efficient to
byteswap the used elements first and then do __copy_to_user() in one
go? Depends on the backend, I guess.

> +	return 0;
> +}
> +