Re: [PATCH] virtio: acknowledge all features before access

[Michael S. Tsirkin]

On Mon, Jan 17, 2022 at 01:38:42PM +0100, Cornelia Huck wrote:
> On Mon, Jan 17 2022, "Michael S. Tsirkin" <mst@xxxxxxxxxx> wrote:
> 
> > On Mon, Jan 17, 2022 at 02:31:49PM +0800, Jason Wang wrote:
> >> 
> >> 在 2022/1/15 上午4:09, Michael S. Tsirkin 写道:
> >> > @@ -495,6 +494,10 @@ int virtio_device_restore(struct virtio_device *dev)
> >> >   	/* We have a driver! */
> >> >   	virtio_add_status(dev, VIRTIO_CONFIG_S_DRIVER);
> >> > +	ret = dev->config->finalize_features(dev);
> >> > +	if (ret)
> >> > +		goto err;
> >> 
> >> 
> >> Is this part of code related?
> >> 
> >> Thanks
> >> 
> >
> > Yes. virtio_finalize_features no longer calls dev->config->finalize_features.
> >
> > I think the dev->config->finalize_features callback is actually
> > a misnomer now, it just sends the features to device,
> > finalize is FEATURES_OK. Renaming that is a bigger
> > patch though, and I'd like this one to be cherry-pickable
> > to stable.
> 
> Do we want to add a comment before the calls to ->finalize_features()
> (/* write features to device */) and adapt the comment in virtio_ring.h?
> Should still be stable-friendly, and giving the callback a better name
> can be a follow-up patch.


Sounds like a good idea. I can also document that near
virtio_finalize_features.

> >
> >> > +
> >> >   	ret = virtio_finalize_features(dev);
> >> >   	if (ret)
> >> >   		goto err;