Re: [PATCH v1] vhost: cache avail index in vhost_enable_notify()

[Stefano Garzarella]

On Mon, Jan 24, 2022 at 11:31:49AM +0000, Stefan Hajnoczi wrote:
> On Fri, Jan 14, 2022 at 10:05:08AM +0100, Stefano Garzarella wrote:
> > In vhost_enable_notify() we enable the notifications and we read
> > the avail index to check if new buffers have become available in
> > the meantime.
> >
> > We are not caching the avail index, so when the device will call
> > vhost_get_vq_desc(), it will find the old value in the cache and
> > it will read the avail index again.
>
> I think this wording is clearer because we do keep a cached the avail
> index value, but the issue is we don't update it:
> s/We are not caching the avail index/We do not update the cached avail
> index value/

I'll fix in v3.
It seems I forgot to CC you on v2: 
https://lore.kernel.org/virtualization/20220121153108.187291-1-sgarzare@xxxxxxxxxx/

> > It would be better to refresh the cache every time we read avail
> > index, so let's change vhost_enable_notify() caching the value in
> > `avail_idx` and compare it with `last_avail_idx` to check if there
> > are new buffers available.
> >
> > Anyway, we don't expect a significant performance boost because
> > the above path is not very common, indeed vhost_enable_notify()
> > is often called with unlikely(), expecting that avail index has
> > not been updated.
> >
> > Signed-off-by: Stefano Garzarella <sgarzare@xxxxxxxxxx>
> > ---
> > v1:
> > - improved the commit description [MST, Jason]
> > ---
> >  drivers/vhost/vhost.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
> > index 59edb5a1ffe2..07363dff559e 100644
> > --- a/drivers/vhost/vhost.c
> > +++ b/drivers/vhost/vhost.c
> > @@ -2543,8 +2543,9 @@ bool vhost_enable_notify(struct vhost_dev *dev, 
> > struct vhost_virtqueue *vq)
> >  		       &vq->avail->idx, r);
> >  		return false;
> >  	}
> > +	vq->avail_idx = vhost16_to_cpu(vq, avail_idx);
> >
> > -	return vhost16_to_cpu(vq, avail_idx) != vq->avail_idx;
> > +	return vq->avail_idx != vq->last_avail_idx;
>
> vhost_vq_avail_empty() has a fast path that's missing in
> vhost_enable_notify():
>
>  if (vq->avail_idx != vq->last_avail_idx)
>      return false;

Yep, I thought about that, but devices usually call 
vhost_enable_notify() right when vq->avail_idx == vq->last_avail_idx, so 
I don't know if it's an extra check for a branch that will never be 
taken.

Do you think it is better to add that check? (maybe with unlikely())

Thanks,
Stefano