Re: [PATCH v6 0/4] Add a vhost RPMsg API

From: Vincent Whitchurch <vincent.whitchurch@axis.com>
To: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com>
Cc: Arnaud POULIQUEN <arnaud.pouliquen@st.com>,
	"kvm@vger.kernel.org" <kvm@vger.kernel.org>,
	"linux-remoteproc@vger.kernel.org"
	<linux-remoteproc@vger.kernel.org>,
	"virtualization@lists.linux-foundation.org" 
	<virtualization@lists.linux-foundation.org>,
	"sound-open-firmware@alsa-project.org" 
	<sound-open-firmware@alsa-project.org>,
	Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com>,
	Liam Girdwood <liam.r.girdwood@linux.intel.com>,
	"Michael S. Tsirkin" <mst@redhat.com>,
	Jason Wang <jasowang@redhat.com>,
	Ohad Ben-Cohen <ohad@wizery.com>,
	Bjorn Andersson <bjorn.andersson@linaro.org>,
	Mathieu Poirier <mathieu.poirier@linaro.org>
Subject: Re: [PATCH v6 0/4] Add a vhost RPMsg API
Date: Fri, 18 Sep 2020 12:39:07 +0200	[thread overview]
Message-ID: <20200918103907.2ts4l5xiwm4542rs@axis.com> (raw)
In-Reply-To: <20200918094719.GD19246@ubuntu>

On Fri, Sep 18, 2020 at 11:47:20AM +0200, Guennadi Liakhovetski wrote:
> On Fri, Sep 18, 2020 at 09:47:45AM +0200, Arnaud POULIQUEN wrote:
> > IMO, as this API is defined in the Linux documentation [5] we should respect it, to ensure
> > one generic implementation. The RPMsg sample client[4] uses this user API, so seems to me
> > a good candidate to verify this. 
> > 
> > That's said, shall we multiple the RPMsg implementations in Linux with several APIs,
> > With the risk to make the RPMsg clients devices dependent on these implementations?
> > That could lead to complex code or duplications...
> 
> So, no, in my understanding there aren't two competing alternative APIs, you'd never have 
> to choose between them. If you're writing a driver for Linux to communicate with remote 
> processors or to run on VMs, you use the existing API. If you're writing a driver for 
> Linux to communicate with those VMs, you use the vhost API and whatever help is available 
> for RPMsg processing.
> 
> However, I can in principle imagine a single driver, written to work on both sides. 
> Something like the rpmsg_char.c or maybe some networking driver. Is that what you're 
> referring to? I can see that as a fun exercise, but are there any real uses for that? 

I hinted at a real use case for this in the previous mail thread[0].
I'm exploring using rpmsg-char to allow communication between two chips,
both running Linux.  rpmsg-char can be used pretty much as-is for both
sides of the userspace-to-userspace communication and (the userspace
side of the) userspace-to-kernel communication between the two chips.

> You could do the same with VirtIO, however, it has been decided to go with two 
> distinct APIs: virtio for guests and vhost for the host, noone bothered to create a 
> single API for both and nobody seems to miss one. Why would we want one with RPMsg?

I think I answered this question in the previous mail thread as well[1]:
| virtio has distinct driver and device roles so the completely different
| APIs on each side are understandable.  But I don't see that distinction
| in the rpmsg API which is why it seems like a good idea to me to make it
| work from both sides of the link and allow the reuse of drivers like
| rpmsg-char, instead of imposing virtio's distinction on rpmsg.

[0] https://www.spinics.net/lists/linux-virtualization/msg43799.html
[1] https://www.spinics.net/lists/linux-virtualization/msg43802.html