This series add ioctl VHOST_GET_VRING_ENDIAN and VHOST_SET_VRING_ENDIAN support for vDPA. Endian-ness is a device wide attribute, so QEMU can get/set endian-ness of the device by these ioctls, and QEMU can be aware of the endian-ness of the device. To support these ioctls, vendor driver needs to implement: vdpa_config_ops.get_vq_endian: set vq endian-ness vdpa_config_ops.set_vq_endian: get vq endian-ness This series also call vdpa_config_ops.get_config() instead of vdpa_get_config_unlocked() in vdpa_dev_net_config_fill(). Then do endian-ness covert properly through __virtio16_to_cpu by support of vdpa_config_ops.get_vq_endian. So there are no race on set_features() any more. Zhu Lingshan (4): vDPA/ifcvf: add get/set_vq_endian support for vDPA vDPA: support ioctl VHOST_GET/SET_VRING_ENDIAN vDPA: detect device endian in _net_config_fill() and _fill_stats_rec() vDPA: report device endian-ness to userspace through netlink drivers/vdpa/ifcvf/ifcvf_base.h | 1 + drivers/vdpa/ifcvf/ifcvf_main.c | 15 +++++++++++++++ drivers/vdpa/vdpa.c | 31 +++++++++++++++++++++++++++---- drivers/vhost/vdpa.c | 32 +++++++++++++++++++++++++++++++- include/linux/vdpa.h | 13 +++++++++++++ include/uapi/linux/vdpa.h | 1 + 6 files changed, 88 insertions(+), 5 deletions(-) -- 2.31.1
