On Fri, Jul 17, 2020 at 11:29:29AM +0200, Marc Hartmayer wrote:
> libvhost-user has no support for legacy virtio devices therefore
> let's fence them.
> 
> Signed-off-by: Marc Hartmayer <mhartmay@linux.ibm.com>
> ---
>  contrib/libvhost-user/libvhost-access.h | 10 ++++++++++
>  contrib/libvhost-user/libvhost-user.c   |  6 ++++++
>  2 files changed, 16 insertions(+)
> 
> diff --git a/contrib/libvhost-user/libvhost-access.h b/contrib/libvhost-user/libvhost-access.h
> index 868ba3e7bfb8..aa505ea1ec02 100644
> --- a/contrib/libvhost-user/libvhost-access.h
> +++ b/contrib/libvhost-user/libvhost-access.h
> @@ -1,11 +1,21 @@
>  #ifndef LIBVHOST_ACCESS_H
>  
> +#include <assert.h>
> +
>  #include "qemu/bswap.h"
>  
>  #include "libvhost-user.h"
>  
> +static inline bool vu_has_feature(VuDev *dev, unsigned int fbit);
> +
>  static inline bool vu_access_is_big_endian(VuDev *dev)
>  {
> +    /*
> +     * TODO: can probably be removed as the fencing is already done in
> +     * `vu_set_features_exec`
> +     */
> +    assert(vu_has_feature(dev, VIRTIO_F_VERSION_1));

Great, please drop it since the memory accesses are called from
performance-critical code paths.