Re: [PATCH] virtio_rpmsg: fix config space alignment
From: Francesco Valla
Date: Fri Nov 20 2026 - 17:29:50 EST
Hello Arnaud,
On Fri, Nov 20, 2026 at 06:50:47PM +0100, Arnaud POULIQUEN wrote:
> Hello,
>
> Sorry it goes out of my radar.
>
> On 11/19/26 16:29, Mathieu Poirier wrote:
> > [Adding openAMP mailing list]
> >
> > Heads up OpenAMP people, I will merge this patch in a week.
> >
> > Thanks,
> > Mathieu
> >
> > On Fri, Nov 06, 2026 at 11:37:24AM -0600, Mathieu Poirier wrote:
> > > On Sun, Nov 01, 2026 at 02:54:03PM +0100, Francesco Valla wrote:
> > > > The comment for virtio_rpmsg_config says that this structure is
> > > > naturally aligned to 32 bits, and this is desirable by some
> > > > architectures that don't get along with unaligned accesses.
> > > > The content of the structure itself however isn't aligned, and the
> > > > __packed attribute prevent the compiler to add this alignment.
> > > >
> > > > Fix this by adding a padding byte.
> > > >
> > > > Fixes: ca96ea25e51a ("rpmsg: virtio_rpmsg_bus: Get buffer size from config space")
> > > > Signed-off-by: Francesco Valla <francesco@xxxxxxxx>
> > > > ---
> > > > Hello,
> > > >
> > > > unfortunately I missed this when the original patch set was sent, but I
> > > > think it's worth fixing (or at least discussing) this now before it
> > > > enters a stable kernel.
> > > >
> > > > The same support in open-amp has already been merged [1], but maybe it
> > > > can be fixed before the next release there too?
> > > >
> > > > Another, lower-impact solution would be to fix the comment, but that can
> > > > then led to issues on architectures that does not support unaligned
> > > > accesses (e.g.: AMR Cortex-M0).
> > > >
> > >
> > > Thanks for bringing this to my attention. I agree with this patch. I will give
> > > Arnaud a couple of weeks to reply to this email though.
> > > > Thank you
> > > >
> > > > Francesco
> > > >
> > > > [1] https://github.com/OpenAMP/open-amp/commit/744376f566c3ec7e06a7d093d785295b78aa486f
> > > > ---
> > > > include/linux/rpmsg/virtio_rpmsg.h | 2 ++
> > > > 1 file changed, 2 insertions(+)
> > > >
> > > > diff --git a/include/linux/rpmsg/virtio_rpmsg.h b/include/linux/rpmsg/virtio_rpmsg.h
> > > > index 0872cbca4593..d24fedb63023 100644
> > > > --- a/include/linux/rpmsg/virtio_rpmsg.h
> > > > +++ b/include/linux/rpmsg/virtio_rpmsg.h
> > > > @@ -23,6 +23,7 @@
> > > > *
> > > > * @version: version of this structure, currently %RPMSG_VDEV_CONFIG_V1.
> > > > * @size: size of this structure in bytes.
> > > > + * @reserved: padding, reserved for future usage.
> > > > * @txbuf_size: Tx buf size from remote's view. For Linux this is rx buf size.
> > > > * @rxbuf_size: Rx buf size from remote's view. For Linux this is tx buf size.
> > > > *
> > > > @@ -33,6 +34,7 @@
> > > > struct virtio_rpmsg_config {
> > > > u8 version;
> > > > __virtio16 size;
> > > > + u8 reserved;
> > > > /* The tx/rx individual buffer size (if VIRTIO_RPMSG_F_BUFSZ) */
> > > > __virtio32 txbuf_size;
> > > > __virtio32 rxbuf_size;
>
>
> What about using a u16 for the version instead?
> Even if we never use the MSB of the field, it seems cleaner to me than
> adding a reserved field.
> Indeed, we can consider the config space header as containing the
> version and the size. The rest is version-dependent, including the
> reserved part.
>
Between the two options, I prefer the u16 version, to keep the
version-dependent "payload" aligned to 32 bits. I'll send a V2 with this
modification.
Shall I send also the corresponding modification for open-amp, or you'll
take care of that?
> Regards,
> Arnaud
>
Regards,
Francesco