[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH V2] virtio-net: unbreak any layout
From: |
Michael S. Tsirkin |
Subject: |
Re: [Qemu-devel] [PATCH V2] virtio-net: unbreak any layout |
Date: |
Wed, 15 Jul 2015 14:23:27 +0300 |
On Wed, Jul 15, 2015 at 03:56:07PM +0800, Jason Wang wrote:
> Commit 032a74a1c0fcdd5fd1c69e56126b4c857ee36611
> ("virtio-net: byteswap virtio-net header") breaks any layout by
> requiring out_sg[0].iov_len >= n->guest_hdr_len. Fixing this by
> copying header to temporary buffer if swap is needed, and then use
> this buffer as part of out_sg.
>
> Fixes 032a74a1c0fcdd5fd1c69e56126b4c857ee36611
> ("virtio-net: byteswap virtio-net header")
> Cc: address@hidden
> Cc: address@hidden
> Signed-off-by: Jason Wang <address@hidden>
> ---
> Changes from V1:
> - avoid header copying if there's no need to do header swap
> - don't write the header back
> ---
> hw/net/virtio-net.c | 17 ++++++++++++++---
> include/hw/virtio/virtio-access.h | 9 +++++++++
> 2 files changed, 23 insertions(+), 3 deletions(-)
>
> diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
> index e3c2db3..12322bd 100644
> --- a/hw/net/virtio-net.c
> +++ b/hw/net/virtio-net.c
> @@ -1142,7 +1142,8 @@ static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
> ssize_t ret, len;
> unsigned int out_num = elem.out_num;
> struct iovec *out_sg = &elem.out_sg[0];
> - struct iovec sg[VIRTQUEUE_MAX_SIZE];
> + struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE];
> + struct virtio_net_hdr hdr;
>
> if (out_num < 1) {
> error_report("virtio-net header not in first element");
> @@ -1150,11 +1151,21 @@ static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
> }
>
> if (n->has_vnet_hdr) {
> - if (out_sg[0].iov_len < n->guest_hdr_len) {
> + if (iov_size(out_sg, out_num) < n->guest_hdr_len) {
> error_report("virtio-net header incorrect");
> exit(1);
> }
> - virtio_net_hdr_swap(vdev, (void *) out_sg[0].iov_base);
> + if (virtio_needs_swap(vdev)) {
> + iov_to_buf(out_sg, out_num, 0, &hdr, sizeof(hdr));
> + virtio_net_hdr_swap(vdev, (void *) &hdr);
> + sg2[0].iov_base = &hdr;
> + sg2[0].iov_len = sizeof(hdr);
> + out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
> + out_sg, out_num,
> + sizeof(hdr), -1);
> + out_num += 1;
> + out_sg = sg2;
> + }
> }
>
> /*
I guess this works but iov cpy is pretty expensive.
- When using the tun backend, we can just set VNETBE/VNETLE
correctly and avoid swaps in userspace.
It's easy to probe for - why don't we do this?
- OTOH when not using the tun backend, offloads are
generally disabled, so the swap isn't needed
since header is going to be all 0s.
- what's left is old kernels when using tun.
how about probing for that explicitly?
In other words, virtio_net_needs_swap as opposed to
the generic virtio_needs_swap.
> diff --git a/include/hw/virtio/virtio-access.h
> b/include/hw/virtio/virtio-access.h
> index cee5dd7..1ec1dfd 100644
> --- a/include/hw/virtio/virtio-access.h
> +++ b/include/hw/virtio/virtio-access.h
> @@ -143,6 +143,15 @@ 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
> --
> 2.1.4