> -----Original Message-----
> From: Olivier MATZ [mailto:olivier.matz at 6wind.com]
> Sent: Thursday, December 11, 2014 6:53 PM
> To: Liu, Jijiang; dev at dpdk.org
> Subject: Re: [dpdk-dev] [PATCH v3 3/3] app/testpmd:change tx_checksum
> command and csum forwarding engine
>
> Hi Jijiang,
>
> Some more comments, in addition to the one I've made in the cover letter.
> Reference link for patchwork readers:
> http://dpdk.org/ml/archives/dev/2014-December/009886.html
>
> On 12/10/2014 02:03 AM, Jijiang Liu wrote:
> > --- a/app/test-pmd/cmdline.c
> > +++ b/app/test-pmd/cmdline.c
> > @@ -316,16 +316,30 @@ static void cmd_help_long_parsed(void
> *parsed_result,
> > " Disable hardware insertion of a VLAN header in"
> > " packets sent on a port.\n\n"
> >
> > - "tx_cksum set (ip|udp|tcp|sctp|vxlan) (hw|sw)
> (port_id)\n"
> > + "tx_cksum set (ip|udp|tcp|sctp) (hw|sw) (port_id)\n"
> > " Select hardware or software calculation of the"
> > " checksum with when transmitting a packet using the"
> > " csum forward engine.\n"
> > - " ip|udp|tcp|sctp always concern the inner layer.\n"
> > - " vxlan concerns the outer IP and UDP layer (in"
> > - " case the packet is recognized as a vxlan packet by"
> > - " the forward engine)\n"
> > + " In the case of tunneling packet, ip|udp|tcp|sctp"
> > + " always concern the inner layer.\n\n"
> > +
> > + "tx_cksum set tunnel (hw|sw|none) (port_id)\n"
> > + " Select hardware or software calculation of the"
> > + " checksum with when transmitting a tunneling packet"
> > + " using the csum forward engine.\n"
> > + " The none option means treat tunneling packet as
> ordinary"
> > + " packet when using the csum forward engine\n."
> > + " Tunneling packet concerns the outer IP, inner IP"
> > + " and inner L4\n"
> > " Please check the NIC datasheet for HW limits.\n\n"
> >
> > + "tx_cksum set (outer-ip) (hw|sw) (port_id)\n"
> > + " Select hardware or software calculation of the"
> > + " checksum with when transmitting a packet using the"
> > + " csum forward engine.\n"
> > + " outer-ip always concern the outer layer of"
> > + " tunneling packet.\n\n"
> > +
> > "tx_checksum show (port_id)\n"
> > " Display tx checksum offload configuration\n\n"
> >
>
> not sure we need 2 different commands for tx_cksum set (outer-ip) and tx_cksum
> set (ip|udp|tcp|sctp). As the syntax is exactly the same, it may result in
> less code
> to have only one command.
Why do we have a separate command for outer layer, I have explained this in
other mail.
Do you agree on this?
>
> > --- a/app/test-pmd/csumonly.c
> > +++ b/app/test-pmd/csumonly.c
> > @@ -256,17 +256,16 @@ process_outer_cksums(void *outer_l3_hdr, uint16_t
> outer_ethertype,
> > struct udp_hdr *udp_hdr;
> > uint64_t ol_flags = 0;
> >
> > - if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM)
> > - ol_flags |= PKT_TX_UDP_TUNNEL_PKT;
> > -
> > if (outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
> > ipv4_hdr->hdr_checksum = 0;
> >
> > - if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM)
> > + if (testpmd_ol_flags &
> TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
> > ol_flags |= PKT_TX_OUTER_IP_CKSUM;
> > - else
> > + else {
> > ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
> > - } else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM)
> > + ol_flags |= PKT_TX_OUTER_IPV4;
> > + }
> > + } else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TUNNEL_CKSUM)
> > ol_flags |= PKT_TX_OUTER_IPV6;
> >
> > udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + outer_l3_len);
> > @@ -300,11 +299,14 @@ process_outer_cksums(void *outer_l3_hdr, uint16_t
> outer_ethertype,
> > * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
> > * UDP|TCP|SCTP
> > *
> > - * The testpmd command line for this forward engine sets the flags
> > + * These testpmd command lines for this forward engine sets the flags
> > * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
> > - * wether a checksum must be calculated in software or in hardware.
> > The
> > - * IP, UDP, TCP and SCTP flags always concern the inner layer. The
> > - * VxLAN flag concerns the outer IP (if packet is recognized as a vxlan
> > packet).
> > + * wether a checksum must be calculated in software or in hardware.
> > + * In the case of tunneling packet, the IP, UDP, TCP and SCTP flags
> > + * always concern the inner layer; the outer IP flag always concern
> > + * the outer layer; the tunnel flag is used to tell the NIC that it
> > + * is a tunneing packet, want hardware offload for outer layer,
> > + * or inner layer, or both.
>
> tunneing -> tunneling
>
> "the tunnel flag is used to tell the NIC that it is a tunneing packet, want
> hardware
> offload for outer layer, or inner layer, or both."
>
> what does that mean?
Ok,
Will replace the above the description with the following:
The tunnel flag is used to set/clear the flag of enabling TX tunneling packet
checksum hardware offload in application.
>
>
> > */
> > static void
> > pkt_burst_checksum_forward(struct fwd_stream *fs) @@ -376,7 +378,9
> > @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
> > l3_hdr = (char *)eth_hdr + l2_len;
> >
> > /* check if it's a supported tunnel (only vxlan for now) */
> > - if (l4_proto == IPPROTO_UDP) {
> > + if (((testpmd_ol_flags &
> > + TESTPMD_TX_OFFLOAD_NON_TUNNEL_CKSUM) == 0)
> > + && (l4_proto == IPPROTO_UDP)) {
> > udp_hdr = (struct udp_hdr *)((char *)l3_hdr + l3_len);
> >
> > /* check udp destination port, 4789 is the default @@ -
> 386,17
> > +390,23 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
> > tunnel = 1;
> >
> > /* currently, this flag is set by i40e only if the
> > - * packet is vxlan */
> > + * packet is a tunneling packet */
> > } else if (m->ol_flags & (PKT_RX_TUNNEL_IPV4_HDR |
> > PKT_RX_TUNNEL_IPV6_HDR))
> > tunnel = 1;
> >
> > if (tunnel == 1) {
> > +
> > + if (testpmd_ol_flags
> > + &
> TESTPMD_TX_OFFLOAD_TUNNEL_CKSUM)
> > + ol_flags |= PKT_TX_UDP_TUNNEL_PKT;
> > +
> > outer_ethertype = ethertype;
> > outer_l2_len = l2_len;
> > outer_l3_len = l3_len;
> > outer_l3_hdr = l3_hdr;
> >
> > + /* currently, only VXLAN packet is supported */
> > eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
> > sizeof(struct udp_hdr) +
> > sizeof(struct vxlan_hdr));
> > @@ -434,7 +444,7 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)
> > /* step 4: fill the mbuf meta data (flags and header lengths) */
> >
> > if (tunnel == 1) {
> > - if (testpmd_ol_flags &
> TESTPMD_TX_OFFLOAD_VXLAN_CKSUM) {
> > + if (testpmd_ol_flags &
> TESTPMD_TX_OFFLOAD_TUNNEL_CKSUM) {
> > m->outer_l2_len = outer_l2_len;
> > m->outer_l3_len = outer_l3_len;
> > m->l2_len = l4_tun_len + l2_len; @@ -505,7
> +515,7 @@
> > pkt_burst_checksum_forward(struct fwd_stream *fs)
> > "m->l4_len=%d\n",
> > m->l2_len, m->l3_len, m->l4_len);
> > if ((tunnel == 1) &&
> > - (testpmd_ol_flags &
> TESTPMD_TX_OFFLOAD_VXLAN_CKSUM))
> > + (testpmd_ol_flags &
> TESTPMD_TX_OFFLOAD_TUNNEL_CKSUM))
> > printf("tx: m->outer_l2_len=%d m-
> >outer_l3_len=%d\n",
> > m->outer_l2_len, m->outer_l3_len);
> > if (tso_segsz != 0)
> > diff --git a/app/test-pmd/testpmd.h b/app/test-pmd/testpmd.h index
> > f8b0740..09caa6a 100644
> > --- a/app/test-pmd/testpmd.h
> > +++ b/app/test-pmd/testpmd.h
> > @@ -125,10 +125,20 @@ struct fwd_stream {
> > #define TESTPMD_TX_OFFLOAD_TCP_CKSUM 0x0004
> > /** Offload SCTP checksum in csum forward engine */
> > #define TESTPMD_TX_OFFLOAD_SCTP_CKSUM 0x0008
> > -/** Offload VxLAN checksum in csum forward engine */
> > -#define TESTPMD_TX_OFFLOAD_VXLAN_CKSUM 0x0010
> > +/** Offload tunneling packet checksum in csum forward engine */
> > +#define TESTPMD_TX_OFFLOAD_TUNNEL_CKSUM 0x0010
> > /** Insert VLAN header in forward engine */
> > #define TESTPMD_TX_OFFLOAD_INSERT_VLAN 0x0020
> > +/**
> > + * Offload outer-IP checksum in csum forward engine
> > + * for tunneling packet
> > + */
> > +#define TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM 0x0040
> > +/**
> > + * For a tunneling packet, user requests HW offload for its outer
> > + * layer checksum, and don't care is it a tunneled packet or not.
> > + */
> > +#define TESTPMD_TX_OFFLOAD_NON_TUNNEL_CKSUM 0x0080
> >
> > /**
> > * The data structure associated with each port.
> >
>
> For now, I did not check the implementation of the patch. I'm not sure I
> understand the specifications, so I cannot check if the code conforms to it.
>
> How this patch is tested?
Yes, it is tested for i40e and ixgbe.
> I think a report similar to
> http://dpdk.org/ml/archives/dev/2014-November/007991.html would help to
> verify that it works (at least on one driver), and to understand the test-pmd
> API
> and the different use cases.
Ok, will send test report.
> Regards,
> Olivier
