On Wed, Oct 14, 2015 at 03:54:32PM +0200, Greg Kurz wrote:
> On Wed, 14 Oct 2015 12:26:58 +0300
> "Michael S. Tsirkin" <m...@redhat.com> wrote:
>
> > commit 5be7d9f1b1452613b95c6ba70b8d7ad3d0797991
> > vhost-net: tell tap backend about the vnet endianness
> > makes vhost net always try to set LE - even if that matches the
> > native endian-ness.
> >
> > This makes it fail on older kernels on x86 without TUNSETVNETLE support.
> >
>
> Since qemu_set_vnet_le() is only called from vhost_net_set_vnet_endian():
>
>
> if (virtio_vdev_has_feature(dev, VIRTIO_F_VERSION_1) ||
> (virtio_legacy_is_cross_endian(dev) && !virtio_is_big_endian(dev))) {
> r = qemu_set_vnet_le(peer, set);
> if (r) {
> error_report("backend does not support LE vnet headers");
> }
>
> and virtio_legacy_is_cross_endian() is { return false; } on x86, I guess this
> is about virtio 1.0, right ?
>
> > To fix, make qemu_set_vnet_le/qemu_set_vnet_be skip the
> > ioctl if it matches the host endian-ness.
> >
>
> The qemu_set_vnet_le() change indeed makes sense since it fixes a bug.
>
> I am not so sure about qemu_set_vnet_be() since it is only called in the
> case we have a ppc64le host and a legacy ppc64 guest, in which case
> HOST_WORDS_BIGENDIAN is not defined...
You are right, but it's there for symmetry.
> IMHO it is better to rework the logic so that we only call qemu_set_vnet_*
> when it is needed. There are only 3 cases actually:
> - BE host with virtio 1 needs vnet_le
> - BE host with legacy LE needs vnet_le
> - LE host with legacy BE needs vnet_be
>
> Something like:
>
> static inline bool vhost_net_needs_vnet_le(VirtIODevice *dev)
> {
> #if defined(HOST_WORDS_BIGENDIAN)
> return
> virtio_vdev_has_feature(dev, VIRTIO_F_VERSION_1) ||
> !virtio_is_big_endian(dev);
> #else
> return false;
> #endif
> }
>
> static inline bool vhost_net_needs_vnet_be(VirtIODevice *dev)
> {
> #if defined(HOST_WORDS_BIGENDIAN)
> return false;
> #else
> return virtio_is_big_endian(dev);
> #endif
> }
>
> static int vhost_net_set_vnet_endian(VirtIODevice *dev, NetClientState *peer,
> bool set)
> {
> int r = 0;
>
> if (vhost_net_needs_vnet_le(dev)) {
> r = qemu_set_vnet_le(peer, set);
> if (r) {
> error_report("backend does not support LE vnet headers");
> }
> } else if (vhost_net_needs_vnet_be(dev)) {
> r = qemu_set_vnet_be(peer, set);
> if (r) {
> error_report("backend does not support BE vnet headers");
> }
> }
>
> return r;
> }
>
> Cheers.
To me this looks wrong: why not call it unconditionally?
The reason is that qemu_set_vnet_be/qemu_set_vnet_le fails on
some backends and that is a low-level detail that should be
hidden.
> --
> Greg
> > Reported-by: Marcel Apfelbaum <mar...@redhat.com>
> > Cc: Greg Kurz <gk...@linux.vnet.ibm.com>
> > Cc: qemu-sta...@nongnu.org
> > Signed-off-by: Michael S. Tsirkin <m...@redhat.com>
> > ---
> > net/net.c | 8 ++++++++
> > 1 file changed, 8 insertions(+)
> >
> > diff --git a/net/net.c b/net/net.c
> > index 28a5597..8e96011 100644
> > --- a/net/net.c
> > +++ b/net/net.c
> > @@ -517,20 +517,28 @@ void qemu_set_vnet_hdr_len(NetClientState *nc, int
> > len)
> >
> > int qemu_set_vnet_le(NetClientState *nc, bool is_le)
> > {
> > +#ifdef HOST_WORDS_BIGENDIAN
> > if (!nc || !nc->info->set_vnet_le) {
> > return -ENOSYS;
> > }
> >
> > return nc->info->set_vnet_le(nc, is_le);
> > +#else
> > + return 0;
> > +#endif
> > }
> >
> > int qemu_set_vnet_be(NetClientState *nc, bool is_be)
> > {
> > +#ifdef HOST_WORDS_BIGENDIAN
> > + return 0;
> > +#else
> > if (!nc || !nc->info->set_vnet_be) {
> > return -ENOSYS;
> > }
> >
> > return nc->info->set_vnet_be(nc, is_be);
> > +#endif
> > }
> >
> > int qemu_can_send_packet(NetClientState *sender)
