Re: [virtio-dev] [PATCH v3 1/7] Vhost-vsock device: Fixed endiannes issue.

[Greg Kurz <gkurz@linux.vnet.ibm.com>]

On Mon, 14 Mar 2016 18:48:46 +0100
Claudio Imbrenda <imbrenda@linux.vnet.ibm.com> wrote:

> On 14/03/16 18:36, Greg Kurz wrote:
> >> diff --git a/drivers/vhost/vsock.c b/drivers/vhost/vsock.c
> >> index 2c5963c..9378f23 100644
> >> --- a/drivers/vhost/vsock.c
> >> +++ b/drivers/vhost/vsock.c
> >> @@ -481,6 +481,8 @@ static int vhost_vsock_set_features(struct vhost_vsock *vsock, u64 features)
> >>   		vq = &vsock->vqs[i];
> >>   		mutex_lock(&vq->mutex);
> >>   		vq->acked_features = features;
> >> +		vq->is_le = virtio_legacy_is_little_endian();
> >> +		vhost_init_is_le(vq);  
> >
> > How can you call vhost_init_is_le() from here ?  
> 
> I can't, it's an error, sorry!
> Also, as I wrote already earlier, I noticed that that call is not needed 
> because that field is initialized already somewhere else. I already 
> removed it in my branch and it won't be there in the next version of the 
> patchset.
> 
> Claudio

vq->is_le is initialized to virtio_legacy_is_little_endian() in vhost_vq_reset(): this
is the default behavior for vhost devices to use rings with native endianness.

Then vq->is_le can be set again from vhost_init_used()->vhost_init_is_le() if the
device is virtio 1 or userspace called the VHOST_SET_VRING_ENDIAN ioctl. This
obviously won't work for vsock since it does not call vhost_init_used()... 

I don't know enough on vsock to see how to handle that, but basically we must
be able to:

1) negotiate virtio 1
2) optionally ask for cross-endian
3) use the device

1) and 2) can be done in any order but must be done before 3)

--
Greg