The GRO library doesn't re-calculate checksums for merged packets. If users want the merged packets to have correct IP checksums, they should select HW IP checksum calculation for the port which the merged packets are transmitted to. But if the port doesn't support HW IP checksum, users may perform a SW IP checksum. This patch add the code about it. Signed-off-by: Wenwu Ma <wenwux.ma@intel.com> --- app/test-pmd/csumonly.c | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/app/test-pmd/csumonly.c b/app/test-pmd/csumonly.c index 8b6665d6f3..76ce58f6d0 100644 --- a/app/test-pmd/csumonly.c +++ b/app/test-pmd/csumonly.c @@ -824,6 +824,7 @@ pkt_burst_checksum_forward(struct fwd_stream *fs) void **gro_ctx; uint16_t gro_pkts_num; uint8_t gro_enable; + struct rte_ipv4_hdr *ipv4_hdr; #endif uint16_t nb_rx; uint16_t nb_tx; @@ -1098,6 +1099,17 @@ pkt_burst_checksum_forward(struct fwd_stream *fs) fs->gro_times = 0; } } + + for (i = 0; i < nb_rx; i++) { + if ((pkts_burst[i]->ol_flags & RTE_MBUF_F_TX_IPV4) && + (tx_offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM) == 0) { + ipv4_hdr = rte_pktmbuf_mtod_offset(pkts_burst[i], + struct rte_ipv4_hdr *, + pkts_burst[i]->l2_len); + ipv4_hdr->hdr_checksum = 0; + ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr); + } + } } #endif -- 2.25.1
On Wed, May 11, 2022 at 3:40 AM Wenwu Ma <wenwux.ma@intel.com> wrote: > > The GRO library doesn't re-calculate checksums for > merged packets. If users want the merged packets to > have correct IP checksums, they should select HW IP > checksum calculation for the port which the merged > packets are transmitted to. But if the port doesn't > support HW IP checksum, users may perform a SW IP > checksum. This patch add the code about it. It looks like a fix, please add a Fixes: line. Out of curiosity, would there be an issue too when using the gso library? > > Signed-off-by: Wenwu Ma <wenwux.ma@intel.com> -- David Marchand
> -----Original Message----- > From: David Marchand <david.marchand@redhat.com> > Sent: 2022年5月11日 15:28 > To: Ma, WenwuX <wenwux.ma@intel.com> > Cc: Li, Xiaoyun <xiaoyun.li@intel.com>; Singh, Aman Deep > <aman.deep.singh@intel.com>; Zhang, Yuying <yuying.zhang@intel.com>; > dev <dev@dpdk.org>; Hu, Jiayu <jiayu.hu@intel.com>; Wang, Yinan > <yinan.wang@intel.com>; He, Xingguang <xingguang.he@intel.com> > Subject: Re: [PATCH] app/testpmd: perform SW IP checksum for GRO > packets > > On Wed, May 11, 2022 at 3:40 AM Wenwu Ma <wenwux.ma@intel.com> > wrote: > > > > The GRO library doesn't re-calculate checksums for merged packets. If > > users want the merged packets to have correct IP checksums, they > > should select HW IP checksum calculation for the port which the merged > > packets are transmitted to. But if the port doesn't support HW IP > > checksum, users may perform a SW IP checksum. This patch add the code > > about it. > > It looks like a fix, please add a Fixes: line. ok > Out of curiosity, would there be an issue too when using the gso library? > yes, gso library has the same issue. IP checksum need to be re-calculated for fragmented packets > > > > Signed-off-by: Wenwu Ma <wenwux.ma@intel.com> > > > -- > David Marchand
The GRO/GSO library doesn't re-calculate checksums for merged/fragmented packets. If users want the packets to have correct IP checksums, they should select HW IP checksum calculation for the port which the packets are transmitted to. But if the port doesn't support HW IP checksum, users may perform a SW IP checksum. This patch add the code about it. Fixes: b7091f1dcfbc ("app/testpmd: enable the heavyweight mode TCP/IPv4 GRO") Fixes: 52f38a2055ed ("app/testpmd: enable TCP/IPv4 VxLAN and GRE GSO") Cc: stable@dpdk.org Signed-off-by: Wenwu Ma <wenwux.ma@intel.com> --- app/test-pmd/csumonly.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) diff --git a/app/test-pmd/csumonly.c b/app/test-pmd/csumonly.c index 8b6665d6f3..e69f31d92c 100644 --- a/app/test-pmd/csumonly.c +++ b/app/test-pmd/csumonly.c @@ -778,6 +778,26 @@ pkt_copy_split(const struct rte_mbuf *pkt) return md[0]; } +/* + * Re-calculate IP checksum for merged/fragmented packets. + */ +static void +pkts_ip_csum_recalc(struct rte_mbuf **pkts_burst, const uint16_t nb_pkts, uint64_t tx_offloads) +{ + int i; + struct rte_ipv4_hdr *ipv4_hdr; + for (i = 0; i < nb_pkts; i++) { + if ((pkts_burst[i]->ol_flags & RTE_MBUF_F_TX_IPV4) && + (tx_offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM) == 0) { + ipv4_hdr = rte_pktmbuf_mtod_offset(pkts_burst[i], + struct rte_ipv4_hdr *, + pkts_burst[i]->l2_len); + ipv4_hdr->hdr_checksum = 0; + ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr); + } + } +} + /* * Receive a burst of packets, and for each packet: * - parse packet, and try to recognize a supported packet type (1) @@ -1098,6 +1118,8 @@ pkt_burst_checksum_forward(struct fwd_stream *fs) fs->gro_times = 0; } } + + pkts_ip_csum_recalc(pkts_burst, nb_rx, tx_offloads); } #endif @@ -1131,6 +1153,8 @@ pkt_burst_checksum_forward(struct fwd_stream *fs) tx_pkts_burst = gso_segments; nb_rx = nb_segments; + + pkts_ip_csum_recalc(tx_pkts_burst, nb_rx, tx_offloads); } else #endif tx_pkts_burst = pkts_burst; -- 2.25.1
The GRO/GSO library doesn't re-calculate checksums for merged/fragmented packets. If users want the packets to have correct IP checksums, they should select HW IP checksum calculation for the port which the packets are transmitted to. But if the port doesn't support HW IP checksum, users may perform a SW IP checksum. This patch add the code about it. Fixes: b7091f1dcfbc ("app/testpmd: enable the heavyweight mode TCP/IPv4 GRO") Fixes: 52f38a2055ed ("app/testpmd: enable TCP/IPv4 VxLAN and GRE GSO") Cc: stable@dpdk.org Signed-off-by: Wenwu Ma <wenwux.ma@intel.com> --- app/test-pmd/csumonly.c | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) diff --git a/app/test-pmd/csumonly.c b/app/test-pmd/csumonly.c index 8b6665d6f3..2498ee88a2 100644 --- a/app/test-pmd/csumonly.c +++ b/app/test-pmd/csumonly.c @@ -778,6 +778,28 @@ pkt_copy_split(const struct rte_mbuf *pkt) return md[0]; } +#if defined(RTE_LIB_GRO) || defined(RTE_LIB_GSO) +/* + * Re-calculate IP checksum for merged/fragmented packets. + */ +static void +pkts_ip_csum_recalc(struct rte_mbuf **pkts_burst, const uint16_t nb_pkts, uint64_t tx_offloads) +{ + int i; + struct rte_ipv4_hdr *ipv4_hdr; + for (i = 0; i < nb_pkts; i++) { + if ((pkts_burst[i]->ol_flags & RTE_MBUF_F_TX_IPV4) && + (tx_offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM) == 0) { + ipv4_hdr = rte_pktmbuf_mtod_offset(pkts_burst[i], + struct rte_ipv4_hdr *, + pkts_burst[i]->l2_len); + ipv4_hdr->hdr_checksum = 0; + ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr); + } + } +} +#endif + /* * Receive a burst of packets, and for each packet: * - parse packet, and try to recognize a supported packet type (1) @@ -1098,6 +1120,8 @@ pkt_burst_checksum_forward(struct fwd_stream *fs) fs->gro_times = 0; } } + + pkts_ip_csum_recalc(pkts_burst, nb_rx, tx_offloads); } #endif @@ -1131,6 +1155,8 @@ pkt_burst_checksum_forward(struct fwd_stream *fs) tx_pkts_burst = gso_segments; nb_rx = nb_segments; + + pkts_ip_csum_recalc(tx_pkts_burst, nb_rx, tx_offloads); } else #endif tx_pkts_burst = pkts_burst; -- 2.25.1
> -----Original Message-----
> From: Ma, WenwuX <wenwux.ma@intel.com>
> Sent: Thursday, May 12, 2022 9:08 AM
> To: Li, Xiaoyun <xiaoyun.li@intel.com>; Singh, Aman Deep
> <aman.deep.singh@intel.com>; Zhang, Yuying <yuying.zhang@intel.com>;
> dev@dpdk.org
> Cc: Hu, Jiayu <jiayu.hu@intel.com>; Wang, Yinan <yinan.wang@intel.com>;
> He, Xingguang <xingguang.he@intel.com>; Ma, WenwuX
> <wenwux.ma@intel.com>; stable@dpdk.org
> Subject: [PATCH v3] app/testpmd: perform SW IP checksum for GRO/GSO
> packets
>
> The GRO/GSO library doesn't re-calculate checksums for merged/fragmented
> packets. If users want the packets to have correct IP checksums, they should
> select HW IP checksum calculation for the port which the packets are
> transmitted to. But if the port doesn't support HW IP checksum, users may
> perform a SW IP checksum. This patch add the code about it.
>
> Fixes: b7091f1dcfbc ("app/testpmd: enable the heavyweight mode TCP/IPv4
> GRO")
> Fixes: 52f38a2055ed ("app/testpmd: enable TCP/IPv4 VxLAN and GRE GSO")
> Cc: stable@dpdk.org
>
> Signed-off-by: Wenwu Ma <wenwux.ma@intel.com>
> ---
> app/test-pmd/csumonly.c | 26 ++++++++++++++++++++++++++
> 1 file changed, 26 insertions(+)
>
> diff --git a/app/test-pmd/csumonly.c b/app/test-pmd/csumonly.c index
> 8b6665d6f3..2498ee88a2 100644
> --- a/app/test-pmd/csumonly.c
> +++ b/app/test-pmd/csumonly.c
> @@ -778,6 +778,28 @@ pkt_copy_split(const struct rte_mbuf *pkt)
> return md[0];
> }
>
> +#if defined(RTE_LIB_GRO) || defined(RTE_LIB_GSO)
> +/*
> + * Re-calculate IP checksum for merged/fragmented packets.
> + */
> +static void
> +pkts_ip_csum_recalc(struct rte_mbuf **pkts_burst, const uint16_t
> +nb_pkts, uint64_t tx_offloads) {
> + int i;
> + struct rte_ipv4_hdr *ipv4_hdr;
> + for (i = 0; i < nb_pkts; i++) {
> + if ((pkts_burst[i]->ol_flags & RTE_MBUF_F_TX_IPV4) &&
> + (tx_offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM)
> == 0) {
> + ipv4_hdr = rte_pktmbuf_mtod_offset(pkts_burst[i],
> + struct rte_ipv4_hdr *,
> + pkts_burst[i]->l2_len);
> + ipv4_hdr->hdr_checksum = 0;
> + ipv4_hdr->hdr_checksum =
> rte_ipv4_cksum(ipv4_hdr);
> + }
> + }
> +}
> +#endif
> +
> /*
> * Receive a burst of packets, and for each packet:
> * - parse packet, and try to recognize a supported packet type (1) @@ -
> 1098,6 +1120,8 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
> fs->gro_times = 0;
> }
> }
> +
> + pkts_ip_csum_recalc(pkts_burst, nb_rx, tx_offloads);
> }
> #endif
>
> @@ -1131,6 +1155,8 @@ pkt_burst_checksum_forward(struct fwd_stream
> *fs)
>
> tx_pkts_burst = gso_segments;
> nb_rx = nb_segments;
> +
> + pkts_ip_csum_recalc(tx_pkts_burst, nb_rx, tx_offloads);
> } else
> #endif
> tx_pkts_burst = pkts_burst;
> --
> 2.25.1
Reviewed-by: Jiayu Hu <jiayu.hu@intel.com>
Thanks,
Jiayu
> -----Original Message-----
> From: Wenwu Ma <wenwux.ma@intel.com>
> Sent: Thursday, May 12, 2022 9:08 AM
> To: Li, Xiaoyun <xiaoyun.li@intel.com>; Singh, Aman Deep
> <aman.deep.singh@intel.com>; Zhang, Yuying <yuying.zhang@intel.com>;
> dev@dpdk.org
> Cc: Hu, Jiayu <jiayu.hu@intel.com>; Wang, Yinan <yinan.wang@intel.com>;
> He, Xingguang <xingguang.he@intel.com>; Ma, WenwuX
> <wenwux.ma@intel.com>; stable@dpdk.org
> Subject: [PATCH v3] app/testpmd: perform SW IP checksum for GRO/GSO
> packets
>
> The GRO/GSO library doesn't re-calculate checksums for
> merged/fragmented packets. If users want the packets to have correct IP
> checksums, they should select HW IP checksum calculation for the port which
> the packets are transmitted to. But if the port doesn't support HW IP
> checksum, users may perform a SW IP checksum. This patch add the code
> about it.
>
> Fixes: b7091f1dcfbc ("app/testpmd: enable the heavyweight mode TCP/IPv4
> GRO")
> Fixes: 52f38a2055ed ("app/testpmd: enable TCP/IPv4 VxLAN and GRE GSO")
> Cc: stable@dpdk.org
>
> Signed-off-by: Wenwu Ma <wenwux.ma@intel.com>
> ---
Tested-by: Wei Ling <weix.ling@intel.com>
Hi Wenwu, LGTM. > -----Original Message----- > From: Ma, WenwuX <wenwux.ma@intel.com> > Sent: Thursday, May 12, 2022 9:08 AM > To: Li, Xiaoyun <xiaoyun.li@intel.com>; Singh, Aman Deep > <aman.deep.singh@intel.com>; Zhang, Yuying <yuying.zhang@intel.com>; > dev@dpdk.org > Cc: Hu, Jiayu <jiayu.hu@intel.com>; Wang, Yinan <yinan.wang@intel.com>; > He, Xingguang <xingguang.he@intel.com>; Ma, WenwuX > <wenwux.ma@intel.com>; stable@dpdk.org > Subject: [PATCH v3] app/testpmd: perform SW IP checksum for GRO/GSO > packets > > The GRO/GSO library doesn't re-calculate checksums for merged/fragmented > packets. If users want the packets to have correct IP checksums, they should > select HW IP checksum calculation for the port which the packets are > transmitted to. But if the port doesn't support HW IP checksum, users may > perform a SW IP checksum. This patch add the code about it. > > Fixes: b7091f1dcfbc ("app/testpmd: enable the heavyweight mode TCP/IPv4 > GRO") > Fixes: 52f38a2055ed ("app/testpmd: enable TCP/IPv4 VxLAN and GRE GSO") > Cc: stable@dpdk.org > > Signed-off-by: Wenwu Ma <wenwux.ma@intel.com> Acked-by: Yuying Zhang <yuying.zhang@intel.com> > --- > app/test-pmd/csumonly.c | 26 ++++++++++++++++++++++++++ > 1 file changed, 26 insertions(+) > > diff --git a/app/test-pmd/csumonly.c b/app/test-pmd/csumonly.c index > 8b6665d6f3..2498ee88a2 100644 > --- a/app/test-pmd/csumonly.c > +++ b/app/test-pmd/csumonly.c > @@ -778,6 +778,28 @@ pkt_copy_split(const struct rte_mbuf *pkt) > return md[0]; > } > > +#if defined(RTE_LIB_GRO) || defined(RTE_LIB_GSO) > +/* > + * Re-calculate IP checksum for merged/fragmented packets. > + */ > +static void > +pkts_ip_csum_recalc(struct rte_mbuf **pkts_burst, const uint16_t > +nb_pkts, uint64_t tx_offloads) { > + int i; > + struct rte_ipv4_hdr *ipv4_hdr; > + for (i = 0; i < nb_pkts; i++) { > + if ((pkts_burst[i]->ol_flags & RTE_MBUF_F_TX_IPV4) && > + (tx_offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM) > == 0) { > + ipv4_hdr = rte_pktmbuf_mtod_offset(pkts_burst[i], > + struct rte_ipv4_hdr *, > + pkts_burst[i]->l2_len); > + ipv4_hdr->hdr_checksum = 0; > + ipv4_hdr->hdr_checksum = > rte_ipv4_cksum(ipv4_hdr); > + } > + } > +} > +#endif > + > /* > * Receive a burst of packets, and for each packet: > * - parse packet, and try to recognize a supported packet type (1) @@ - > 1098,6 +1120,8 @@ pkt_burst_checksum_forward(struct fwd_stream *fs) > fs->gro_times = 0; > } > } > + > + pkts_ip_csum_recalc(pkts_burst, nb_rx, tx_offloads); > } > #endif > > @@ -1131,6 +1155,8 @@ pkt_burst_checksum_forward(struct fwd_stream > *fs) > > tx_pkts_burst = gso_segments; > nb_rx = nb_segments; > + > + pkts_ip_csum_recalc(tx_pkts_burst, nb_rx, tx_offloads); > } else > #endif > tx_pkts_burst = pkts_burst; > -- > 2.25.1
On 5/17/22 19:02, Zhang, Yuying wrote: > Hi Wenwu, > LGTM. > >> -----Original Message----- >> From: Ma, WenwuX <wenwux.ma@intel.com> >> Sent: Thursday, May 12, 2022 9:08 AM >> To: Li, Xiaoyun <xiaoyun.li@intel.com>; Singh, Aman Deep >> <aman.deep.singh@intel.com>; Zhang, Yuying <yuying.zhang@intel.com>; >> dev@dpdk.org >> Cc: Hu, Jiayu <jiayu.hu@intel.com>; Wang, Yinan <yinan.wang@intel.com>; >> He, Xingguang <xingguang.he@intel.com>; Ma, WenwuX >> <wenwux.ma@intel.com>; stable@dpdk.org >> Subject: [PATCH v3] app/testpmd: perform SW IP checksum for GRO/GSO >> packets >> >> The GRO/GSO library doesn't re-calculate checksums for merged/fragmented >> packets. If users want the packets to have correct IP checksums, they should >> select HW IP checksum calculation for the port which the packets are >> transmitted to. But if the port doesn't support HW IP checksum, users may >> perform a SW IP checksum. This patch add the code about it. >> >> Fixes: b7091f1dcfbc ("app/testpmd: enable the heavyweight mode TCP/IPv4 >> GRO") >> Fixes: 52f38a2055ed ("app/testpmd: enable TCP/IPv4 VxLAN and GRE GSO") >> Cc: stable@dpdk.org >> >> Signed-off-by: Wenwu Ma <wenwux.ma@intel.com> > > Acked-by: Yuying Zhang <yuying.zhang@intel.com> Since CI is happy with the patch now [1]. [1] https://lab.dpdk.org/results/dashboard/patchsets/22087/ Applied, thanks.