[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH 04/14] virtio-net: avoid sg copy
From: |
Michael S. Tsirkin |
Subject: |
Re: [Qemu-devel] [PATCH 04/14] virtio-net: avoid sg copy |
Date: |
Tue, 25 Sep 2012 02:44:33 +0200 |
On Mon, Sep 24, 2012 at 07:37:40PM -0500, Anthony Liguori wrote:
> "Michael S. Tsirkin" <address@hidden> writes:
>
> > Avoid tweaking iovec during receive. This removes
> > the need to copy the vector.
> > Note: we already have an evil cast in work_around_broken_dhclient
> > and unfortunately this adds another one.
> > const on buf is ignored by this function anyway so arguably
> > this is not making things much worse.
> >
> > Signed-off-by: Michael S. Tsirkin <address@hidden>
> > ---
> > hw/virtio-net.c | 45 +++++++++++++++++++++------------------------
> > iov.h | 8 ++++----
> > 2 files changed, 25 insertions(+), 28 deletions(-)
> >
> > diff --git a/hw/virtio-net.c b/hw/virtio-net.c
> > index 0c5081e..6e53858 100644
> > --- a/hw/virtio-net.c
> > +++ b/hw/virtio-net.c
> > @@ -503,7 +503,7 @@ static int virtio_net_has_buffers(VirtIONet *n, int
> > bufsize)
> > * we should provide a mechanism to disable it to avoid polluting the host
> > * cache.
> > */
> > -static void work_around_broken_dhclient(struct virtio_net_hdr *hdr,
> > +static void work_around_broken_dhclient(const struct virtio_net_hdr *hdr,
> > const uint8_t *buf, size_t size)
> > {
> > if ((hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && /* missing csum */
> > @@ -513,31 +513,27 @@ static void work_around_broken_dhclient(struct
> > virtio_net_hdr *hdr,
> > (buf[34] == 0 && buf[35] == 67)) { /* udp.srcport == bootps */
> > /* FIXME this cast is evil */
> > net_checksum_calculate((uint8_t *)buf, size);
> > - hdr->flags &= ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
> > + ((struct virtio_net_hdr *)hdr)->flags &=
> > ~VIRTIO_NET_HDR_F_NEEDS_CSUM;
> > }
> > }
> >
> > -static int receive_header(VirtIONet *n, struct iovec *iov, int iovcnt,
> > - const void *buf, size_t size, size_t hdr_len)
> > +static int receive_header(VirtIONet *n, const struct iovec *iov, int
> > iov_cnt,
> > + const void *buf, size_t size)
> > {
> > - struct virtio_net_hdr *hdr = (struct virtio_net_hdr *)iov[0].iov_base;
> > int offset = 0;
> >
> > - hdr->flags = 0;
> > - hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
> > -
> > if (n->has_vnet_hdr) {
> > - memcpy(hdr, buf, sizeof(*hdr));
> > - offset = sizeof(*hdr);
> > - work_around_broken_dhclient(hdr, buf + offset, size - offset);
> > + work_around_broken_dhclient(buf, buf + offset, size -
> > offset);
>
> I think it would be less ugly to cast here.
I was thinking about this but could not decide either way.
Will do.
> > + offset = sizeof(struct virtio_net_hdr);
> > + iov_from_buf(iov, iov_cnt, 0, buf, offset);
> > + } else {
> > + struct virtio_net_hdr hdr = {
> > + .flags = 0,
> > + .gso_type = VIRTIO_NET_HDR_GSO_NONE
> > + };
> > + iov_from_buf(iov, iov_cnt, 0, &hdr, sizeof hdr);
> > }
> >
> > - /* We only ever receive a struct virtio_net_hdr from the tapfd,
> > - * but we may be passing along a larger header to the guest.
> > - */
> > - iov[0].iov_base += hdr_len;
> > - iov[0].iov_len -= hdr_len;
> > -
> > return offset;
> > }
> >
> > @@ -598,7 +594,8 @@ static ssize_t virtio_net_receive(NetClientState *nc,
> > const uint8_t *buf, size_t
> > {
> > VirtIONet *n = DO_UPCAST(NICState, nc, nc)->opaque;
> > struct virtio_net_hdr_mrg_rxbuf *mhdr = NULL;
> > - size_t offset, i;
> > + const struct iovec *sg = elem.in_sg;
> > + size_t offset, i, guest_offset;
> >
> > if (!virtio_net_can_receive(&n->nic->nc))
> > return -1;
> > @@ -615,7 +612,7 @@ static ssize_t virtio_net_receive(NetClientState *nc,
> > const uint8_t *buf, size_t
> > while (offset < size) {
> > VirtQueueElement elem;
> > int len, total;
> > - struct iovec sg[VIRTQUEUE_MAX_SIZE];
> > + const struct iovec *sg = elem.in_sg;
> >
> > total = 0;
> >
> > @@ -640,20 +637,20 @@ static ssize_t virtio_net_receive(NetClientState *nc,
> > const uint8_t *buf, size_t
> > exit(1);
> > }
> >
> > - memcpy(&sg, &elem.in_sg[0], sizeof(sg[0]) * elem.in_num);
> > -
> > if (i == 0) {
> > if (n->mergeable_rx_bufs)
> > mhdr = (struct virtio_net_hdr_mrg_rxbuf *)sg[0].iov_base;
> >
> > offset += receive_header(n, sg, elem.in_num,
> > - buf + offset, size - offset,
> > - n->guest_hdr_len);
> > + buf + offset, size - offset);
> > total += n->guest_hdr_len;
> > + guest_offset = n->guest_hdr_len;
> > + } else {
> > + guest_offset = 0;
> > }
> >
> > /* copy in packet. ugh */
> > - len = iov_from_buf(sg, elem.in_num, 0,
> > + len = iov_from_buf(sg, elem.in_num, guest_offset,
> > buf + offset, size - offset);
> > total += len;
> > offset += len;
> > diff --git a/iov.h b/iov.h
> > index f0daefa..6c9fdf1 100644
> > --- a/iov.h
> > +++ b/iov.h
> > @@ -90,8 +90,8 @@ void iov_hexdump(const struct iovec *iov, const unsigned
> > int iov_cnt,
> > /*
> > * Partial copy of vector from iov to dst_iov (data is not copied).
> > * dst_iov overlaps iov at a specified offset.
> > - * size of dst_iov is at most bytes. Actual size is returned.
> > + * size of dst_iov is at most bytes. dst vector count is returned.
> > */
> > -size_t iov_cpy(struct iovec *dst_iov, unsigned int dst_iov_cnt,
> > - const struct iovec *iov, unsigned int iov_cnt,
> > - size_t offset, size_t bytes);
> > +unsigned iov_cpy(struct iovec *dst_iov, unsigned int dst_iov_cnt,
> > + const struct iovec *iov, unsigned int iov_cnt,
> > + size_t offset, size_t bytes);
>
> I think this belongs in an earlier patch.
Good catch thanks.
> Regards,
>
> Anthony Liguori
>
> > --
> > MST
- [Qemu-devel] [PATCH 00/14] virtio-net: iovec handling cleanup, Michael S. Tsirkin, 2012/09/24
- [Qemu-devel] [PATCH 01/14] virtio-net: track host/guest header length, Michael S. Tsirkin, 2012/09/24
- [Qemu-devel] [PATCH 03/14] iov: add iov_cpy, Michael S. Tsirkin, 2012/09/24
- [Qemu-devel] [PATCH 04/14] virtio-net: avoid sg copy, Michael S. Tsirkin, 2012/09/24
- [Qemu-devel] [PATCH 05/14] virtio-net: use safe iov operations for rx, Michael S. Tsirkin, 2012/09/24
- [Qemu-devel] [PATCH 07/14] virtio-net: first s/g is always at start of buf, Michael S. Tsirkin, 2012/09/24
- [Qemu-devel] [PATCH 02/14] iov: add const annotation, Michael S. Tsirkin, 2012/09/24
- [Qemu-devel] [PATCH 08/14] virtio-net: switch tx to safe iov functions, Michael S. Tsirkin, 2012/09/24
- [Qemu-devel] [PATCH 09/14] virtio-net: simplify rx code, Michael S. Tsirkin, 2012/09/24
- [Qemu-devel] [PATCH 06/14] virtio-net: refactor receive_hdr, Michael S. Tsirkin, 2012/09/24
- [Qemu-devel] [PATCH 10/14] virtio: don't mark unaccessed memory as dirty, Michael S. Tsirkin, 2012/09/24