Re: [RFC 3/3] virtio_ring: use new vring flags

[Michael S. Tsirkin]

On Sat, Dec 08, 2018 at 09:47:29PM +0800, Tiwei Bie wrote:
> On Fri, Dec 07, 2018 at 01:10:48PM -0500, Michael S. Tsirkin wrote:
> > On Fri, Dec 07, 2018 at 04:48:42PM +0800, Tiwei Bie wrote:
> > > Switch to using the _SPLIT_ and _PACKED_ variants of vring flags
> > > in split ring and packed ring respectively.
> > > 
> > > Signed-off-by: Tiwei Bie <tiwei.bie@xxxxxxxxx>
> > > ---
> > > @@ -502,7 +505,8 @@ static inline int virtqueue_add_split(struct virtqueue *_vq,
> > >  		}
> > >  	}
> > >  	/* Last one doesn't continue. */
> > > -	desc[prev].flags &= cpu_to_virtio16(_vq->vdev, ~VRING_DESC_F_NEXT);
> > > +	desc[prev].flags &= cpu_to_virtio16(_vq->vdev,
> > > +				(u16)~BIT(VRING_SPLIT_DESC_F_NEXT));
> > >  
> > >  	if (indirect) {
> > >  		/* Now that the indirect table is filled in, map it. */
> > 
> > I kind of dislike it that this forces use of a cast here.
> > Kind of makes it more fragile. Let's use a temporary instead?
> 
> I tried something like this:
> 
> u16 mask = ~BIT(VRING_SPLIT_DESC_F_NEXT);
> 
> And it will still cause the warning:
> 
> warning: large integer implicitly truncated to unsigned type [-Woverflow]
>   u16 mask = ~BIT(VRING_SPLIT_DESC_F_NEXT);
> 
> If the cast isn't wanted, maybe use ~(1 << VRING_SPLIT_DESC_F_NEXT)
> directly?

What I'd like to see is a macro that warns/errors out if the shift is
too large. I'll think it over on the weekend.


> 
> > 
> > > -- 
> > > 2.17.1