On Wed, 13 Jan 2016 10:53:11 +0100 Laurent Vivier <lviv...@redhat.com> wrote:
> On 13/01/2016 10:06, Greg Kurz wrote: > > On Tue, 12 Jan 2016 18:24:28 +0100 > > Laurent Vivier <lviv...@redhat.com> wrote: > > > >> On 11/01/2016 17:12, Greg Kurz wrote: > >>> When running a fully emulated device in cross-endian conditions, including > >>> a virtio 1.0 device offered to a big endian guest, we need to fix the vnet > >>> headers. This is currently handled by the virtio_net_hdr_swap() function > >>> in the core virtio-net code but it should actually be handled by the net > >>> backend. > >>> > >>> With this patch, virtio-net now tries to configure the backend to do the > >>> endian fixing when the device starts (i.e. drivers sets the CONFIG_OK > >>> bit). > >>> If the backend cannot support the requested endiannes, we have to fallback > >>> onto virtio_net_hdr_swap(): this is recorded in the needs_vnet_hdr_swap > >>> flag, > >>> to be used in the TX and RX paths. > >>> > >>> Note that we reset the backend to the default behaviour (guest native > >>> endianness) when the device stops (i.e. device status had CONFIG_OK bit > >>> and > >>> driver unsets it). This is needed, with the linux tap backend at least, > >>> otherwise the guest may loose network connectivity if rebooted into a > >>> different endianness. > >>> > >>> The current vhost-net code also tries to configure net backends. This will > >>> be no more needed and will be reverted in a subsequent patch. > >>> > >>> Signed-off-by: Greg Kurz <gk...@linux.vnet.ibm.com> > >>> --- > >>> v2: > >>> - dropped useless check in the 'else if' branch in > >>> virtio_net_vnet_status() > >>> - merged virtio_net_vhost_status() change from patch 2 > >>> - use semicolon in "backend does no support..." error message > >>> - merged patch 3 (drop the virtio_needs_swap() helper) > >>> - provided some more details in changelog and comments > >>> --- > >>> hw/net/virtio-net.c | 49 > >>> +++++++++++++++++++++++++++++++++++-- > >>> include/hw/virtio/virtio-access.h | 9 ------- > >>> include/hw/virtio/virtio-net.h | 1 + > >>> 3 files changed, 48 insertions(+), 11 deletions(-) > >>> > >>> diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c > >>> index a877614e3e7a..497fb7119a08 100644 > >>> --- a/hw/net/virtio-net.c > >>> +++ b/hw/net/virtio-net.c > >>> @@ -128,6 +128,13 @@ static void virtio_net_vhost_status(VirtIONet *n, > >>> uint8_t status) > >>> if (!n->vhost_started) { > >>> int r, i; > >>> > >>> + if (n->needs_vnet_hdr_swap) { > >>> + error_report("backend does not support %s vnet headers; " > >>> + "falling back on userspace virtio", > >>> + virtio_is_big_endian(vdev) ? "BE" : "LE"); > >>> + return; > >>> + } > >>> + > >>> /* Any packets outstanding? Purge them to avoid touching rings > >>> * when vhost is running. > >>> */ > >>> @@ -152,6 +159,40 @@ static void virtio_net_vhost_status(VirtIONet *n, > >>> uint8_t status) > >>> } > >>> } > >>> > >>> +static void virtio_net_vnet_status(VirtIONet *n, uint8_t status) > >>> +{ > >>> + VirtIODevice *vdev = VIRTIO_DEVICE(n); > >>> + NetClientState *peer = qemu_get_queue(n->nic)->peer; > >>> + > >>> + if (virtio_net_started(n, status)) { > >>> + int r; > >>> + > >>> + /* Before using the device, we tell the network backend about the > >>> + * endianness to use when parsing vnet headers. If the backend > >>> can't > >>> + * do it, we fallback onto fixing the headers in the core > >>> virtio-net > >>> + * code. > >>> + */ > >>> + if (virtio_is_big_endian(vdev)) { > >>> + r = qemu_set_vnet_be(peer, true); > >>> + } else { > >>> + r = qemu_set_vnet_le(peer, true); > >>> + } > >> > >> If endianess of the guest and the virtio device is the same, but r is < > >> 0 (-ENOSYS or -EINVAL) you will badly swap header (and disable vhost). > >> > > > > This can only happen if the endianness of the host is not the same as the > > OK, you're right, I was studying sources without commit: > > 052bd52 net: don't set native endianness > > Sorry for the noise... > No problem. Thank you again for your time ! > > endianness of the device. In this case (we usually call cross-endian) the > > vnet headers must be byteswapped but the backend cannot handle it. This > > has two consequences: > > - vhost cannot be used since it requires the backend to support cross-endian > > vnet headers, so we fallback onto full emulation in QEMU > > - the emulation code must byteswap vnet headers > > > >> I think you need something like this to fall back to the old method: > >> > >> if (r < 0) { > >> #ifdef HOST_WORDS_BIGENDIAN > >> r = virtio_access_is_big_endian(vdev) ? false : true; > >> #else > >> r = virtio_access_is_big_endian(vdev) ? true : false; > >> #endif > >> } > >> > >> > >> But... > >> > >>> + n->needs_vnet_hdr_swap = !!r; > >>> + } else if (virtio_net_started(n, vdev->status)) { > >>> + /* After using the device, we need to reset the network backend > >>> to > >>> + * the default (guest native endianness), otherwise the guest may > >>> + * loose network connectivity if it is rebooted into a different > >>> + * endianness. > >>> + */ > >>> + if (virtio_is_big_endian(vdev)) { > >>> + qemu_set_vnet_be(peer, false); > >>> + } else { > >>> + qemu_set_vnet_le(peer, false); > >>> + } > >>> + } > >>> +} > >>> + > >>> static void virtio_net_set_status(struct VirtIODevice *vdev, uint8_t > >>> status) > >>> { > >>> VirtIONet *n = VIRTIO_NET(vdev); > >>> @@ -159,6 +200,7 @@ static void virtio_net_set_status(struct VirtIODevice > >>> *vdev, uint8_t status) > >>> int i; > >>> uint8_t queue_status; > >>> > >>> + virtio_net_vnet_status(n, status); > >>> virtio_net_vhost_status(n, status); > >>> > >>> for (i = 0; i < n->max_queues; i++) { > >>> @@ -957,7 +999,10 @@ static void receive_header(VirtIONet *n, const > >>> struct iovec *iov, int iov_cnt, > >>> void *wbuf = (void *)buf; > >>> work_around_broken_dhclient(wbuf, wbuf + n->host_hdr_len, > >>> size - n->host_hdr_len); > >>> - virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf); > >>> + > >>> + if (n->needs_vnet_hdr_swap) { > >>> + virtio_net_hdr_swap(VIRTIO_DEVICE(n), wbuf); > >>> + } > >> > >> ... this will change the behavior here, as before it was not > >> conditional. Why ? > >> > > > > This is what this patch is all about as described above. > > > >>> iov_from_buf(iov, iov_cnt, 0, buf, sizeof(struct > >>> virtio_net_hdr)); > >>> } else { > >>> struct virtio_net_hdr hdr = { > >>> @@ -1167,7 +1212,7 @@ static int32_t virtio_net_flush_tx(VirtIONetQueue > >>> *q) > >>> error_report("virtio-net header incorrect"); > >>> exit(1); > >>> } > >>> - if (virtio_needs_swap(vdev)) { > >>> + if (n->needs_vnet_hdr_swap) { > >>> virtio_net_hdr_swap(vdev, (void *) &mhdr); > >>> sg2[0].iov_base = &mhdr; > >>> sg2[0].iov_len = n->guest_hdr_len; > >>> diff --git a/include/hw/virtio/virtio-access.h > >>> b/include/hw/virtio/virtio-access.h > >>> index 8aec843c8ff3..a01fff2e51d7 100644 > >>> --- a/include/hw/virtio/virtio-access.h > >>> +++ b/include/hw/virtio/virtio-access.h > >>> @@ -143,15 +143,6 @@ static inline uint64_t virtio_ldq_p(VirtIODevice > >>> *vdev, const void *ptr) > >>> } > >>> } > >>> > >>> -static inline bool virtio_needs_swap(VirtIODevice *vdev) > >>> -{ > >>> -#ifdef HOST_WORDS_BIGENDIAN > >>> - return virtio_access_is_big_endian(vdev) ? false : true; > >>> -#else > >>> - return virtio_access_is_big_endian(vdev) ? true : false; > >>> -#endif > >>> -} > >>> - > >>> static inline uint16_t virtio_tswap16(VirtIODevice *vdev, uint16_t s) > >>> { > >>> #ifdef HOST_WORDS_BIGENDIAN > >>> diff --git a/include/hw/virtio/virtio-net.h > >>> b/include/hw/virtio/virtio-net.h > >>> index f3cc25feca2b..27bc868fbc7d 100644 > >>> --- a/include/hw/virtio/virtio-net.h > >>> +++ b/include/hw/virtio/virtio-net.h > >>> @@ -94,6 +94,7 @@ typedef struct VirtIONet { > >>> uint64_t curr_guest_offloads; > >>> QEMUTimer *announce_timer; > >>> int announce_counter; > >>> + bool needs_vnet_hdr_swap; > >>> } VirtIONet; > >>> > >>> void virtio_net_set_netclient_name(VirtIONet *n, const char *name, > >>> > >>> > >> > > >