4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5 * Copyright 2014 6WIND S.A.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * * Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * * Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * * Neither the name of Intel Corporation nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
25 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
26 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
28 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
29 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
30 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
31 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
32 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
42 #include <sys/queue.h>
45 #include <rte_common.h>
46 #include <rte_byteorder.h>
48 #include <rte_debug.h>
49 #include <rte_cycles.h>
50 #include <rte_memory.h>
51 #include <rte_memcpy.h>
52 #include <rte_memzone.h>
53 #include <rte_launch.h>
55 #include <rte_per_lcore.h>
56 #include <rte_lcore.h>
57 #include <rte_atomic.h>
58 #include <rte_branch_prediction.h>
59 #include <rte_mempool.h>
61 #include <rte_interrupts.h>
63 #include <rte_ether.h>
64 #include <rte_ethdev.h>
69 #include <rte_prefetch.h>
70 #include <rte_string_fns.h>
75 #define IP_DEFTTL 64 /* from RFC 1340. */
76 #define IP_VERSION 0x40
77 #define IP_HDRLEN 0x05 /* default IP header length == five 32-bits words. */
78 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
80 #define GRE_KEY_PRESENT 0x2000
82 #define GRE_SUPPORTED_FIELDS GRE_KEY_PRESENT
84 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */
85 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
86 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
91 /* structure that caches offload info for the current packet */
92 struct testpmd_offload_info {
99 uint16_t outer_ethertype;
100 uint16_t outer_l2_len;
101 uint16_t outer_l3_len;
102 uint8_t outer_l4_proto;
104 uint16_t tunnel_tso_segsz;
108 /* simplified GRE header */
109 struct simple_gre_hdr {
112 } __attribute__((__packed__));
115 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
117 if (ethertype == _htons(ETHER_TYPE_IPv4))
118 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
119 else /* assume ethertype == ETHER_TYPE_IPv6 */
120 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
123 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
125 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
127 struct tcp_hdr *tcp_hdr;
129 info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
130 info->l4_proto = ipv4_hdr->next_proto_id;
132 /* only fill l4_len for TCP, it's useful for TSO */
133 if (info->l4_proto == IPPROTO_TCP) {
134 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
135 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
140 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
142 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
144 struct tcp_hdr *tcp_hdr;
146 info->l3_len = sizeof(struct ipv6_hdr);
147 info->l4_proto = ipv6_hdr->proto;
149 /* only fill l4_len for TCP, it's useful for TSO */
150 if (info->l4_proto == IPPROTO_TCP) {
151 tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
152 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
158 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
159 * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
160 * header. The l4_len argument is only set in case of TCP (useful for TSO).
163 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info)
165 struct ipv4_hdr *ipv4_hdr;
166 struct ipv6_hdr *ipv6_hdr;
168 info->l2_len = sizeof(struct ether_hdr);
169 info->ethertype = eth_hdr->ether_type;
171 if (info->ethertype == _htons(ETHER_TYPE_VLAN)) {
172 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
174 info->l2_len += sizeof(struct vlan_hdr);
175 info->ethertype = vlan_hdr->eth_proto;
178 switch (info->ethertype) {
179 case _htons(ETHER_TYPE_IPv4):
180 ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
181 parse_ipv4(ipv4_hdr, info);
183 case _htons(ETHER_TYPE_IPv6):
184 ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
185 parse_ipv6(ipv6_hdr, info);
195 /* Parse a vxlan header */
197 parse_vxlan(struct udp_hdr *udp_hdr,
198 struct testpmd_offload_info *info,
201 struct ether_hdr *eth_hdr;
203 /* check udp destination port, 4789 is the default vxlan port
204 * (rfc7348) or that the rx offload flag is set (i40e only
206 if (udp_hdr->dst_port != _htons(4789) &&
207 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
211 info->outer_ethertype = info->ethertype;
212 info->outer_l2_len = info->l2_len;
213 info->outer_l3_len = info->l3_len;
214 info->outer_l4_proto = info->l4_proto;
216 eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
217 sizeof(struct udp_hdr) +
218 sizeof(struct vxlan_hdr));
220 parse_ethernet(eth_hdr, info);
221 info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */
224 /* Parse a gre header */
226 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
228 struct ether_hdr *eth_hdr;
229 struct ipv4_hdr *ipv4_hdr;
230 struct ipv6_hdr *ipv6_hdr;
233 /* check which fields are supported */
234 if ((gre_hdr->flags & _htons(~GRE_SUPPORTED_FIELDS)) != 0)
237 gre_len += sizeof(struct simple_gre_hdr);
239 if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
240 gre_len += GRE_KEY_LEN;
242 if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) {
244 info->outer_ethertype = info->ethertype;
245 info->outer_l2_len = info->l2_len;
246 info->outer_l3_len = info->l3_len;
247 info->outer_l4_proto = info->l4_proto;
249 ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len);
251 parse_ipv4(ipv4_hdr, info);
252 info->ethertype = _htons(ETHER_TYPE_IPv4);
255 } else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) {
257 info->outer_ethertype = info->ethertype;
258 info->outer_l2_len = info->l2_len;
259 info->outer_l3_len = info->l3_len;
260 info->outer_l4_proto = info->l4_proto;
262 ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len);
264 info->ethertype = _htons(ETHER_TYPE_IPv6);
265 parse_ipv6(ipv6_hdr, info);
268 } else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) {
270 info->outer_ethertype = info->ethertype;
271 info->outer_l2_len = info->l2_len;
272 info->outer_l3_len = info->l3_len;
273 info->outer_l4_proto = info->l4_proto;
275 eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len);
277 parse_ethernet(eth_hdr, info);
281 info->l2_len += gre_len;
285 /* Parse an encapsulated ip or ipv6 header */
287 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
289 struct ipv4_hdr *ipv4_hdr = encap_ip;
290 struct ipv6_hdr *ipv6_hdr = encap_ip;
293 ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
295 if (ip_version != 4 && ip_version != 6)
299 info->outer_ethertype = info->ethertype;
300 info->outer_l2_len = info->l2_len;
301 info->outer_l3_len = info->l3_len;
303 if (ip_version == 4) {
304 parse_ipv4(ipv4_hdr, info);
305 info->ethertype = _htons(ETHER_TYPE_IPv4);
307 parse_ipv6(ipv6_hdr, info);
308 info->ethertype = _htons(ETHER_TYPE_IPv6);
313 /* if possible, calculate the checksum of a packet in hw or sw,
314 * depending on the testpmd command line configuration */
316 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
317 uint16_t testpmd_ol_flags)
319 struct ipv4_hdr *ipv4_hdr = l3_hdr;
320 struct udp_hdr *udp_hdr;
321 struct tcp_hdr *tcp_hdr;
322 struct sctp_hdr *sctp_hdr;
323 uint64_t ol_flags = 0;
324 uint32_t max_pkt_len, tso_segsz = 0;
326 /* ensure packet is large enough to require tso */
327 if (!info->is_tunnel) {
328 max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
330 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
331 tso_segsz = info->tso_segsz;
333 max_pkt_len = info->outer_l2_len + info->outer_l3_len +
334 info->l2_len + info->l3_len + info->l4_len +
335 info->tunnel_tso_segsz;
336 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
337 tso_segsz = info->tunnel_tso_segsz;
340 if (info->ethertype == _htons(ETHER_TYPE_IPv4)) {
342 ipv4_hdr->hdr_checksum = 0;
344 ol_flags |= PKT_TX_IPV4;
345 if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
346 ol_flags |= PKT_TX_IP_CKSUM;
348 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
349 ol_flags |= PKT_TX_IP_CKSUM;
351 ipv4_hdr->hdr_checksum =
352 rte_ipv4_cksum(ipv4_hdr);
354 } else if (info->ethertype == _htons(ETHER_TYPE_IPv6))
355 ol_flags |= PKT_TX_IPV6;
357 return 0; /* packet type not supported, nothing to do */
359 if (info->l4_proto == IPPROTO_UDP) {
360 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
361 /* do not recalculate udp cksum if it was 0 */
362 if (udp_hdr->dgram_cksum != 0) {
363 udp_hdr->dgram_cksum = 0;
364 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM)
365 ol_flags |= PKT_TX_UDP_CKSUM;
367 udp_hdr->dgram_cksum =
368 get_udptcp_checksum(l3_hdr, udp_hdr,
372 } else if (info->l4_proto == IPPROTO_TCP) {
373 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
376 ol_flags |= PKT_TX_TCP_SEG;
377 else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM)
378 ol_flags |= PKT_TX_TCP_CKSUM;
381 get_udptcp_checksum(l3_hdr, tcp_hdr,
384 } else if (info->l4_proto == IPPROTO_SCTP) {
385 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
387 /* sctp payload must be a multiple of 4 to be
389 if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_SCTP_CKSUM) &&
390 ((ipv4_hdr->total_length & 0x3) == 0)) {
391 ol_flags |= PKT_TX_SCTP_CKSUM;
393 /* XXX implement CRC32c, example available in
401 /* Calculate the checksum of outer header */
403 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
404 uint16_t testpmd_ol_flags, int tso_enabled)
406 struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
407 struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
408 struct udp_hdr *udp_hdr;
409 uint64_t ol_flags = 0;
411 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
412 ipv4_hdr->hdr_checksum = 0;
413 ol_flags |= PKT_TX_OUTER_IPV4;
415 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
416 ol_flags |= PKT_TX_OUTER_IP_CKSUM;
418 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
420 ol_flags |= PKT_TX_OUTER_IPV6;
422 if (info->outer_l4_proto != IPPROTO_UDP)
425 udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
427 /* outer UDP checksum is done in software as we have no hardware
428 * supporting it today, and no API for it. In the other side, for
429 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
432 * If a packet will be TSOed into small packets by NIC, we cannot
433 * set/calculate a non-zero checksum, because it will be a wrong
434 * value after the packet be split into several small packets.
437 udp_hdr->dgram_cksum = 0;
439 /* do not recalculate udp cksum if it was 0 */
440 if (udp_hdr->dgram_cksum != 0) {
441 udp_hdr->dgram_cksum = 0;
442 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
443 udp_hdr->dgram_cksum =
444 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
446 udp_hdr->dgram_cksum =
447 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
455 * Performs actual copying.
456 * Returns number of segments in the destination mbuf on success,
457 * or negative error code on failure.
460 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
461 uint16_t seglen[], uint8_t nb_seg)
463 uint32_t dlen, slen, tlen;
465 const struct rte_mbuf *m;
478 while (ms != NULL && i != nb_seg) {
481 slen = rte_pktmbuf_data_len(ms);
482 src = rte_pktmbuf_mtod(ms, const uint8_t *);
486 dlen = RTE_MIN(seglen[i], slen);
487 md[i]->data_len = dlen;
488 md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
489 dst = rte_pktmbuf_mtod(md[i], uint8_t *);
492 len = RTE_MIN(slen, dlen);
493 memcpy(dst, src, len);
508 else if (tlen != m->pkt_len)
511 md[0]->nb_segs = nb_seg;
512 md[0]->pkt_len = tlen;
513 md[0]->vlan_tci = m->vlan_tci;
514 md[0]->vlan_tci_outer = m->vlan_tci_outer;
515 md[0]->ol_flags = m->ol_flags;
516 md[0]->tx_offload = m->tx_offload;
522 * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
523 * Copy packet contents and offload information into then new segmented mbuf.
525 static struct rte_mbuf *
526 pkt_copy_split(const struct rte_mbuf *pkt)
529 uint32_t i, len, nb_seg;
530 struct rte_mempool *mp;
531 uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
532 struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
534 mp = current_fwd_lcore()->mbp;
536 if (tx_pkt_split == TX_PKT_SPLIT_RND)
537 nb_seg = random() % tx_pkt_nb_segs + 1;
539 nb_seg = tx_pkt_nb_segs;
541 memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
543 /* calculate number of segments to use and their length. */
545 for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
550 n = pkt->pkt_len - len;
552 /* update size of the last segment to fit rest of the packet */
560 p = rte_pktmbuf_alloc(mp);
563 "failed to allocate %u-th of %u mbuf "
564 "from mempool: %s\n",
565 nb_seg - i, nb_seg, mp->name);
570 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
571 RTE_LOG(ERR, USER1, "mempool %s, %u-th segment: "
572 "expected seglen: %u, "
573 "actual mbuf tailroom: %u\n",
574 mp->name, i, seglen[i],
575 rte_pktmbuf_tailroom(md[i]));
580 /* all mbufs successfully allocated, do copy */
582 rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
585 "mbuf_copy_split for %p(len=%u, nb_seg=%u) "
586 "into %u segments failed with error code: %d\n",
587 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
589 /* figure out how many mbufs to free. */
593 /* free unused mbufs */
594 for (; i != nb_seg; i++) {
595 rte_pktmbuf_free_seg(md[i]);
603 * Receive a burst of packets, and for each packet:
604 * - parse packet, and try to recognize a supported packet type (1)
605 * - if it's not a supported packet type, don't touch the packet, else:
606 * - reprocess the checksum of all supported layers. This is done in SW
607 * or HW, depending on testpmd command line configuration
608 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
609 * segmentation offload (this implies HW TCP checksum)
610 * Then transmit packets on the output port.
612 * (1) Supported packets are:
613 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
614 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
616 * Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
617 * Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
618 * Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
620 * The testpmd command line for this forward engine sets the flags
621 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
622 * wether a checksum must be calculated in software or in hardware. The
623 * IP, UDP, TCP and SCTP flags always concern the inner layer. The
624 * OUTER_IP is only useful for tunnel packets.
627 pkt_burst_checksum_forward(struct fwd_stream *fs)
629 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
630 struct rte_port *txp;
631 struct rte_mbuf *m, *p;
632 struct ether_hdr *eth_hdr;
633 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
638 uint64_t rx_ol_flags, tx_ol_flags;
639 uint16_t testpmd_ol_flags;
641 uint32_t rx_bad_ip_csum;
642 uint32_t rx_bad_l4_csum;
643 struct testpmd_offload_info info;
645 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
648 uint64_t core_cycles;
651 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
652 start_tsc = rte_rdtsc();
655 /* receive a burst of packet */
656 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
658 if (unlikely(nb_rx == 0))
660 if (unlikely(gro_ports[fs->rx_port].enable))
661 nb_rx = rte_gro_reassemble_burst(pkts_burst,
663 &(gro_ports[fs->rx_port].param));
665 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
666 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
668 fs->rx_packets += nb_rx;
672 txp = &ports[fs->tx_port];
673 testpmd_ol_flags = txp->tx_ol_flags;
674 memset(&info, 0, sizeof(info));
675 info.tso_segsz = txp->tso_segsz;
676 info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
678 for (i = 0; i < nb_rx; i++) {
679 if (likely(i < nb_rx - 1))
680 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
685 info.pkt_len = rte_pktmbuf_pkt_len(m);
687 rx_ol_flags = m->ol_flags;
689 /* Update the L3/L4 checksum error packet statistics */
690 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
692 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
695 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
696 * and inner headers */
698 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
699 ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
701 ether_addr_copy(&ports[fs->tx_port].eth_addr,
703 parse_ethernet(eth_hdr, &info);
704 l3_hdr = (char *)eth_hdr + info.l2_len;
706 /* check if it's a supported tunnel */
707 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_PARSE_TUNNEL) {
708 if (info.l4_proto == IPPROTO_UDP) {
709 struct udp_hdr *udp_hdr;
711 udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
713 parse_vxlan(udp_hdr, &info, m->packet_type);
715 tx_ol_flags |= PKT_TX_TUNNEL_VXLAN;
716 } else if (info.l4_proto == IPPROTO_GRE) {
717 struct simple_gre_hdr *gre_hdr;
719 gre_hdr = (struct simple_gre_hdr *)
720 ((char *)l3_hdr + info.l3_len);
721 parse_gre(gre_hdr, &info);
723 tx_ol_flags |= PKT_TX_TUNNEL_GRE;
724 } else if (info.l4_proto == IPPROTO_IPIP) {
727 encap_ip_hdr = (char *)l3_hdr + info.l3_len;
728 parse_encap_ip(encap_ip_hdr, &info);
730 tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
734 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
735 if (info.is_tunnel) {
736 outer_l3_hdr = l3_hdr;
737 l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
740 /* step 2: depending on user command line configuration,
741 * recompute checksum either in software or flag the
742 * mbuf to offload the calculation to the NIC. If TSO
743 * is configured, prepare the mbuf for TCP segmentation. */
745 /* process checksums of inner headers first */
746 tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
749 /* Then process outer headers if any. Note that the software
750 * checksum will be wrong if one of the inner checksums is
751 * processed in hardware. */
752 if (info.is_tunnel == 1) {
753 tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
755 !!(tx_ol_flags & PKT_TX_TCP_SEG));
758 /* step 3: fill the mbuf meta data (flags and header lengths) */
760 if (info.is_tunnel == 1) {
761 if (info.tunnel_tso_segsz ||
763 TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) ||
764 (tx_ol_flags & PKT_TX_OUTER_IPV6)) {
765 m->outer_l2_len = info.outer_l2_len;
766 m->outer_l3_len = info.outer_l3_len;
767 m->l2_len = info.l2_len;
768 m->l3_len = info.l3_len;
769 m->l4_len = info.l4_len;
770 m->tso_segsz = info.tunnel_tso_segsz;
773 /* if there is a outer UDP cksum
774 processed in sw and the inner in hw,
775 the outer checksum will be wrong as
776 the payload will be modified by the
778 m->l2_len = info.outer_l2_len +
779 info.outer_l3_len + info.l2_len;
780 m->l3_len = info.l3_len;
781 m->l4_len = info.l4_len;
784 /* this is only useful if an offload flag is
785 * set, but it does not hurt to fill it in any
787 m->l2_len = info.l2_len;
788 m->l3_len = info.l3_len;
789 m->l4_len = info.l4_len;
790 m->tso_segsz = info.tso_segsz;
792 m->ol_flags = tx_ol_flags;
794 /* Do split & copy for the packet. */
795 if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
796 p = pkt_copy_split(m);
804 /* if verbose mode is enabled, dump debug info */
805 if (verbose_level > 0) {
808 printf("-----------------\n");
809 printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
810 fs->rx_port, m, m->pkt_len, m->nb_segs);
811 /* dump rx parsed packet info */
812 rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
813 printf("rx: l2_len=%d ethertype=%x l3_len=%d "
814 "l4_proto=%d l4_len=%d flags=%s\n",
815 info.l2_len, rte_be_to_cpu_16(info.ethertype),
816 info.l3_len, info.l4_proto, info.l4_len, buf);
817 if (rx_ol_flags & PKT_RX_LRO)
818 printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
819 if (info.is_tunnel == 1)
820 printf("rx: outer_l2_len=%d outer_ethertype=%x "
821 "outer_l3_len=%d\n", info.outer_l2_len,
822 rte_be_to_cpu_16(info.outer_ethertype),
824 /* dump tx packet info */
825 if ((testpmd_ol_flags & (TESTPMD_TX_OFFLOAD_IP_CKSUM |
826 TESTPMD_TX_OFFLOAD_UDP_CKSUM |
827 TESTPMD_TX_OFFLOAD_TCP_CKSUM |
828 TESTPMD_TX_OFFLOAD_SCTP_CKSUM)) ||
830 printf("tx: m->l2_len=%d m->l3_len=%d "
832 m->l2_len, m->l3_len, m->l4_len);
833 if (info.is_tunnel == 1) {
834 if ((testpmd_ol_flags &
835 TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) ||
836 (tx_ol_flags & PKT_TX_OUTER_IPV6))
837 printf("tx: m->outer_l2_len=%d "
838 "m->outer_l3_len=%d\n",
841 if (info.tunnel_tso_segsz != 0 &&
842 (m->ol_flags & PKT_TX_TCP_SEG))
843 printf("tx: m->tso_segsz=%d\n",
845 } else if (info.tso_segsz != 0 &&
846 (m->ol_flags & PKT_TX_TCP_SEG))
847 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
848 rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
849 printf("tx: flags=%s", buf);
854 nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
856 if (nb_prep != nb_rx)
857 printf("Preparing packet burst to transmit failed: %s\n",
858 rte_strerror(rte_errno));
860 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst,
866 if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
868 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
869 rte_delay_us(burst_tx_delay_time);
870 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
871 &pkts_burst[nb_tx], nb_rx - nb_tx);
874 fs->tx_packets += nb_tx;
875 fs->rx_bad_ip_csum += rx_bad_ip_csum;
876 fs->rx_bad_l4_csum += rx_bad_l4_csum;
878 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
879 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
881 if (unlikely(nb_tx < nb_rx)) {
882 fs->fwd_dropped += (nb_rx - nb_tx);
884 rte_pktmbuf_free(pkts_burst[nb_tx]);
885 } while (++nb_tx < nb_rx);
887 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
888 end_tsc = rte_rdtsc();
889 core_cycles = (end_tsc - start_tsc);
890 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
894 struct fwd_engine csum_fwd_engine = {
895 .fwd_mode_name = "csum",
896 .port_fwd_begin = NULL,
897 .port_fwd_end = NULL,
898 .packet_fwd = pkt_burst_checksum_forward,