On Tue, Aug 22, 2023 at 10:55 AM Olivier Matz <olivier.m...@6wind.com> wrote:
>
> Hi David,
>
> On Tue, Aug 22, 2023 at 09:32:44AM +0200, David Marchand wrote:
> > The L4 checksum offloading API does not require l4_len to be set.
> > Make the driver discover the L4 headers size by itself.
> >
> > Fixes: 6546e76056e3 ("net/tap: calculate checksums of multi segs packets")
> > Cc: sta...@dpdk.org
> >
> > Signed-off-by: David Marchand <david.march...@redhat.com>
> > Tested-by: Ales Musil <amu...@redhat.com>
> > ---
> > .mailmap | 1 +
> > drivers/net/tap/rte_eth_tap.c | 13 +++++++++++--
> > 2 files changed, 12 insertions(+), 2 deletions(-)
> >
> > diff --git a/.mailmap b/.mailmap
> > index 864d33ee46..b6a21b35cb 100644
> > --- a/.mailmap
> > +++ b/.mailmap
> > @@ -40,6 +40,7 @@ Aleksandr Loktionov <aleksandr.loktio...@intel.com>
> > Aleksandr Miloshenko <a.miloshe...@f5.com>
> > Aleksey Baulin <aleksey.bau...@gmail.com>
> > Aleksey Katargin <gure...@gmail.com>
> > +Ales Musil <amu...@redhat.com>
> > Alexander Bechikov <asb.t...@gmail.com>
> > Alexander Belyakov <abely...@gmail.com>
> > Alexander Chernavin <acherna...@netgate.com>
> > diff --git a/drivers/net/tap/rte_eth_tap.c b/drivers/net/tap/rte_eth_tap.c
> > index bf98f75559..0ab214847a 100644
> > --- a/drivers/net/tap/rte_eth_tap.c
> > +++ b/drivers/net/tap/rte_eth_tap.c
> > @@ -645,13 +645,22 @@ tap_write_mbufs(struct tx_queue *txq, uint16_t
> > num_mbufs,
> > ((mbuf->ol_flags & (RTE_MBUF_F_TX_IP_CKSUM |
> > RTE_MBUF_F_TX_IPV4) ||
> > (mbuf->ol_flags & RTE_MBUF_F_TX_L4_MASK) ==
> > RTE_MBUF_F_TX_UDP_CKSUM ||
> > (mbuf->ol_flags & RTE_MBUF_F_TX_L4_MASK) ==
> > RTE_MBUF_F_TX_TCP_CKSUM))) {
>
> While looking at the patch, I noticed this line:
>
> mbuf->ol_flags & (RTE_MBUF_F_TX_IP_CKSUM | RTE_MBUF_F_TX_IPV4)
>
> I think only RTE_MBUF_F_TX_IP_CKSUM should be checked.
And tap_tx_l3_cksum is wrong too:
if (ol_flags & (RTE_MBUF_F_TX_IP_CKSUM | RTE_MBUF_F_TX_IPV4)) {
This is a separate issue, I'll send another patch.
>
> > + unsigned int l4_len = 0;
> > +
> > is_cksum = 1;
> >
> > + if ((mbuf->ol_flags & RTE_MBUF_F_TX_L4_MASK) ==
> > + RTE_MBUF_F_TX_UDP_CKSUM)
> > + l4_len = sizeof(struct rte_udp_hdr);
> > + else if ((mbuf->ol_flags & RTE_MBUF_F_TX_L4_MASK) ==
> > + RTE_MBUF_F_TX_TCP_CKSUM)
> > + l4_len = sizeof(struct rte_tcp_hdr);
> > +
> > /* Support only packets with at least layer 4
> > * header included in the first segment
> > */
> > seg_len = rte_pktmbuf_data_len(mbuf);
> > - l234_hlen = mbuf->l2_len + mbuf->l3_len +
> > mbuf->l4_len;
> > + l234_hlen = mbuf->l2_len + mbuf->l3_len + l4_len;
> > if (seg_len < l234_hlen)
> > return -1;
> >
> > @@ -661,7 +670,7 @@ tap_write_mbufs(struct tx_queue *txq, uint16_t
> > num_mbufs,
> > rte_memcpy(m_copy, rte_pktmbuf_mtod(mbuf, void *),
> > l234_hlen);
> > tap_tx_l3_cksum(m_copy, mbuf->ol_flags,
> > - mbuf->l2_len, mbuf->l3_len,
> > mbuf->l4_len,
> > + mbuf->l2_len, mbuf->l3_len, l4_len,
> > &l4_cksum, &l4_phdr_cksum,
> > &l4_raw_cksum);
> > iovecs[k].iov_base = m_copy;
> > --
> > 2.41.0
> >
>
> Using rte_ipv4_udptcp_cksum() in this code would probably simplify it, and may
> solve other issues (for instance the 0 checksum for UDP which has a special
> meaning).
I agree such a rework would make the code easier to read, and may
solve other issues.
But I prefer to keep my original fix as is, and do what you propose as
a followup patch.
--
David Marchand
