Re: [PATCH v4 4/4] vhost: add an RPMsg API

From: "Michael S. Tsirkin" <mst@redhat.com>
To: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Cc: kvm@vger.kernel.org, linux-remoteproc@vger.kernel.org,
	virtualization@lists.linux-foundation.org,
	sound-open-firmware@alsa-project.org,
	Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>,
	Liam Girdwood <liam.r.girdwood@linux.intel.com>,
	Jason Wang <jasowang@redhat.com>,
	Ohad Ben-Cohen <ohad@wizery.com>,
	Bjorn Andersson <bjorn.andersson@linaro.org>,
	Mathieu Poirier <mathieu.poirier@linaro.org>,
	Vincent Whitchurch <vincent.whitchurch@axis.com>
Subject: Re: [PATCH v4 4/4] vhost: add an RPMsg API
Date: Tue, 25 Aug 2020 09:53:24 -0400	[thread overview]
Message-ID: <20200825095148-mutt-send-email-mst@kernel.org> (raw)
In-Reply-To: <20200825134109.GA9822@ubuntu>

On Tue, Aug 25, 2020 at 03:41:10PM +0200, Guennadi Liakhovetski wrote:
> Hi Michael,
> 
> ...back from holidays and still unsure what your preferred solution 
> for the message layout would be:
> 
> On Wed, Aug 12, 2020 at 02:32:43PM +0200, Guennadi Liakhovetski wrote:
> > Hi Michael,
> > 
> > Thanks for a review.
> > 
> > On Mon, Aug 10, 2020 at 09:44:15AM -0400, Michael S. Tsirkin wrote:
> > > On Tue, Aug 04, 2020 at 05:19:17PM +0200, Guennadi Liakhovetski wrote:
> 
> [snip]
> 
> > > > > > +static int vhost_rpmsg_get_single(struct vhost_virtqueue *vq)
> > > > > > +{
> > > > > > +	struct vhost_rpmsg *vr = container_of(vq->dev, struct vhost_rpmsg, dev);
> > > > > > +	unsigned int out, in;
> > > > > > +	int head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), &out, &in,
> > > > > > +				     NULL, NULL);
> > > > > > +	if (head < 0) {
> > > > > > +		vq_err(vq, "%s(): error %d getting buffer\n",
> > > > > > +		       __func__, head);
> > > > > > +		return head;
> > > > > > +	}
> > > > > > +
> > > > > > +	/* Nothing new? */
> > > > > > +	if (head == vq->num)
> > > > > > +		return head;
> > > > > > +
> > > > > > +	if (vq == &vr->vq[VIRTIO_RPMSG_RESPONSE] && (out || in != 1)) {
> > > > > 
> > > > > This in != 1 looks like a dependency on a specific message layout.
> > > > > virtio spec says to avoid these. Using iov iters it's not too hard to do
> > > > > ...
> > > > 
> > > > This is an RPMsg VirtIO implementation, and it has to match the virtio_rpmsg_bus.c 
> > > > driver, and that one has specific VirtIO queue and message usage patterns.
> > > 
> > > That could be fine for legacy virtio, but now you are claiming support
> > > for virtio 1, so need to fix these assumptions in the device.
> > 
> > I can just deop these checks without changing anything else, that still would work. 
> > I could also make this work with "any" layout - either ignoring any left-over 
> > buffers or maybe even getting them one by one. But I wouldn't even be able to test 
> > those modes without modifying / breaking the current virtio-rpmsg driver. What's 
> > the preferred solution?
> 
> Could you elaborate a bit please?
> 
> Thanks
> Guennadi

I'd prefer it that we make it work with any layout.
For testing, there was a hack for virtio ring which
split up descriptors at a random boundary.
I'll try to locate it and post, sounds like something
we might want upstream for debugging.

-- 
MST