Re: [RFC PATCH 2/2] rpmsg: virtio_rpmsg_bus: get buffer size from config space

[Mathieu Poirier <mathieu.poirier@linaro.org>]

On Thu, Dec 04, 2025 at 10:55:10AM -0600, Tanmay Shah wrote:
> 
> 
> On 12/3/25 1:35 PM, Michael S. Tsirkin wrote:
> > On Wed, Dec 03, 2025 at 12:12:46PM -0600, Tanmay Shah wrote:
> > > Hello,
> > > 
> > > Thanks for your reviews. Please find the response below.
> > > 
> > > On 11/22/25 6:05 AM, Zhongqiu Han wrote:
> > > > On 11/15/2025 2:46 AM, Tanmay Shah wrote:
> > > > > From: Xiang Xiao <xiaoxiang781216@gmail.com>
> > > > > 
> 
> [...]
> 
> > > 
> > > Thanks, that's a good question. I keept id 2 unmodified from the original
> > > series. I don't know why ID 2 was chosen in the original series. I will have
> > > to discuss this with the linux remoteproc/rpmsg maintainers and choose the
> > > correct ID.
> > > 
> > > I don't see any problem choosing ID 1, but for some reason if ID 1 was
> > > assigned and deprecated (I don't think that is the case) then only we should
> > > use ID 2.
> > > 
> > > 
> > > Arnaud, Mathieu, Bjorn any input here?
> > > 
> > > > > +
> > > > > +struct virtio_rpmsg_config {
> > > > > +    /* The tx/rx individual buffer size(if VIRTIO_RPMSG_F_BUFSZ) */
> > > > > +    __u32 txbuf_size;
> > > > > +    __u32 rxbuf_size;
> > > > > +    __u32 reserved[14]; /* Reserve for the future use */
> > > > 
> > > > Should we use __virtio32 instead of __u32 to avoid endianness issues?
> > > > 
> > > > 
> > > 
> > > Sure, if that is the standard in other virtio headers I will modify it.
> > > 
> > > Thanks,
> > > Tanmay
> > 
> > rpmsg is still not standardized, sadly. It's really time it was.
> > 
> > 
> > 
> > Modern virtio devices use __le32.
> > Accordingly, accessed with virtio_cread_le
> > 
> > 
> > __virtioXX and virtio_cread are for legacy compatible parts of config space.
> > 
> 
> As of now, I am using virtio_cread so I think it is legacy compatible.
> 
> > 
> > Does rpmsg want to be modern or keep using legacy? I donnu.
> > 
> 
> I don't know either. This we have to discuss with the maintainers and other
> vendors.
>

I would certainly like to see a modernization effort, but it needs to be
backward compatible.  Probably outside the scope of this set though.
 
> 
> > Ideally it should finally be documented and at that point we
> > definitely will want to switch to __le32.
> > 
> > 
> > For now, run sparse to make sure you don't introduce new endian-ness
> > issues.
> 
> Ack.
> 
> > 
> > 
> 
> 
> 
> > > > > +    /* Put the customize config here */
> > > > > +} __attribute__((packed));
> > > > > +
> > > > > +#endif /* _LINUX_VIRTIO_RPMSG_H */
> > > > 
> > > > 
> > 
>