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_launch.h>
54 #include <rte_per_lcore.h>
55 #include <rte_lcore.h>
56 #include <rte_atomic.h>
57 #include <rte_branch_prediction.h>
58 #include <rte_mempool.h>
60 #include <rte_interrupts.h>
62 #include <rte_ether.h>
63 #include <rte_ethdev.h>
68 #include <rte_prefetch.h>
69 #include <rte_string_fns.h>
76 #define IP_DEFTTL 64 /* from RFC 1340. */
77 #define IP_VERSION 0x40
78 #define IP_HDRLEN 0x05 /* default IP header length == five 32-bits words. */
79 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
81 #define GRE_KEY_PRESENT 0x2000
83 #define GRE_SUPPORTED_FIELDS GRE_KEY_PRESENT
85 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */
86 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
87 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
92 /* structure that caches offload info for the current packet */
93 struct testpmd_offload_info {
101 uint16_t outer_ethertype;
102 uint16_t outer_l2_len;
103 uint16_t outer_l3_len;
104 uint8_t outer_l4_proto;
106 uint16_t tunnel_tso_segsz;
110 /* simplified GRE header */
111 struct simple_gre_hdr {
114 } __attribute__((__packed__));
117 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
119 if (ethertype == _htons(ETHER_TYPE_IPv4))
120 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
121 else /* assume ethertype == ETHER_TYPE_IPv6 */
122 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
125 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
127 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
129 struct tcp_hdr *tcp_hdr;
131 info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
132 info->l4_proto = ipv4_hdr->next_proto_id;
134 /* only fill l4_len for TCP, it's useful for TSO */
135 if (info->l4_proto == IPPROTO_TCP) {
136 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
137 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
142 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
144 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
146 struct tcp_hdr *tcp_hdr;
148 info->l3_len = sizeof(struct ipv6_hdr);
149 info->l4_proto = ipv6_hdr->proto;
151 /* only fill l4_len for TCP, it's useful for TSO */
152 if (info->l4_proto == IPPROTO_TCP) {
153 tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
154 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
160 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
161 * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
162 * header. The l4_len argument is only set in case of TCP (useful for TSO).
165 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info)
167 struct ipv4_hdr *ipv4_hdr;
168 struct ipv6_hdr *ipv6_hdr;
170 info->l2_len = sizeof(struct ether_hdr);
171 info->ethertype = eth_hdr->ether_type;
173 if (info->ethertype == _htons(ETHER_TYPE_VLAN)) {
174 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
176 info->l2_len += sizeof(struct vlan_hdr);
177 info->ethertype = vlan_hdr->eth_proto;
180 switch (info->ethertype) {
181 case _htons(ETHER_TYPE_IPv4):
182 ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
183 parse_ipv4(ipv4_hdr, info);
185 case _htons(ETHER_TYPE_IPv6):
186 ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
187 parse_ipv6(ipv6_hdr, info);
197 /* Parse a vxlan header */
199 parse_vxlan(struct udp_hdr *udp_hdr,
200 struct testpmd_offload_info *info,
203 struct ether_hdr *eth_hdr;
205 /* check udp destination port, 4789 is the default vxlan port
206 * (rfc7348) or that the rx offload flag is set (i40e only
208 if (udp_hdr->dst_port != _htons(4789) &&
209 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
213 info->outer_ethertype = info->ethertype;
214 info->outer_l2_len = info->l2_len;
215 info->outer_l3_len = info->l3_len;
216 info->outer_l4_proto = info->l4_proto;
218 eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
219 sizeof(struct udp_hdr) +
220 sizeof(struct vxlan_hdr));
222 parse_ethernet(eth_hdr, info);
223 info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */
226 /* Parse a gre header */
228 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
230 struct ether_hdr *eth_hdr;
231 struct ipv4_hdr *ipv4_hdr;
232 struct ipv6_hdr *ipv6_hdr;
235 /* check which fields are supported */
236 if ((gre_hdr->flags & _htons(~GRE_SUPPORTED_FIELDS)) != 0)
239 gre_len += sizeof(struct simple_gre_hdr);
241 if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
242 gre_len += GRE_KEY_LEN;
244 if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) {
246 info->outer_ethertype = info->ethertype;
247 info->outer_l2_len = info->l2_len;
248 info->outer_l3_len = info->l3_len;
249 info->outer_l4_proto = info->l4_proto;
251 ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len);
253 parse_ipv4(ipv4_hdr, info);
254 info->ethertype = _htons(ETHER_TYPE_IPv4);
257 } else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) {
259 info->outer_ethertype = info->ethertype;
260 info->outer_l2_len = info->l2_len;
261 info->outer_l3_len = info->l3_len;
262 info->outer_l4_proto = info->l4_proto;
264 ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len);
266 info->ethertype = _htons(ETHER_TYPE_IPv6);
267 parse_ipv6(ipv6_hdr, info);
270 } else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) {
272 info->outer_ethertype = info->ethertype;
273 info->outer_l2_len = info->l2_len;
274 info->outer_l3_len = info->l3_len;
275 info->outer_l4_proto = info->l4_proto;
277 eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len);
279 parse_ethernet(eth_hdr, info);
283 info->l2_len += gre_len;
287 /* Parse an encapsulated ip or ipv6 header */
289 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
291 struct ipv4_hdr *ipv4_hdr = encap_ip;
292 struct ipv6_hdr *ipv6_hdr = encap_ip;
295 ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
297 if (ip_version != 4 && ip_version != 6)
301 info->outer_ethertype = info->ethertype;
302 info->outer_l2_len = info->l2_len;
303 info->outer_l3_len = info->l3_len;
305 if (ip_version == 4) {
306 parse_ipv4(ipv4_hdr, info);
307 info->ethertype = _htons(ETHER_TYPE_IPv4);
309 parse_ipv6(ipv6_hdr, info);
310 info->ethertype = _htons(ETHER_TYPE_IPv6);
315 /* if possible, calculate the checksum of a packet in hw or sw,
316 * depending on the testpmd command line configuration */
318 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
319 uint16_t testpmd_ol_flags)
321 struct ipv4_hdr *ipv4_hdr = l3_hdr;
322 struct udp_hdr *udp_hdr;
323 struct tcp_hdr *tcp_hdr;
324 struct sctp_hdr *sctp_hdr;
325 uint64_t ol_flags = 0;
326 uint32_t max_pkt_len, tso_segsz = 0;
328 /* ensure packet is large enough to require tso */
329 if (!info->is_tunnel) {
330 max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
332 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
333 tso_segsz = info->tso_segsz;
335 max_pkt_len = info->outer_l2_len + info->outer_l3_len +
336 info->l2_len + info->l3_len + info->l4_len +
337 info->tunnel_tso_segsz;
338 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
339 tso_segsz = info->tunnel_tso_segsz;
342 if (info->ethertype == _htons(ETHER_TYPE_IPv4)) {
344 ipv4_hdr->hdr_checksum = 0;
346 ol_flags |= PKT_TX_IPV4;
347 if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
348 ol_flags |= PKT_TX_IP_CKSUM;
350 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
351 ol_flags |= PKT_TX_IP_CKSUM;
353 ipv4_hdr->hdr_checksum =
354 rte_ipv4_cksum(ipv4_hdr);
356 } else if (info->ethertype == _htons(ETHER_TYPE_IPv6))
357 ol_flags |= PKT_TX_IPV6;
359 return 0; /* packet type not supported, nothing to do */
361 if (info->l4_proto == IPPROTO_UDP) {
362 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
363 /* do not recalculate udp cksum if it was 0 */
364 if (udp_hdr->dgram_cksum != 0) {
365 udp_hdr->dgram_cksum = 0;
366 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM)
367 ol_flags |= PKT_TX_UDP_CKSUM;
369 udp_hdr->dgram_cksum =
370 get_udptcp_checksum(l3_hdr, udp_hdr,
374 } else if (info->l4_proto == IPPROTO_TCP) {
375 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
378 ol_flags |= PKT_TX_TCP_SEG;
379 else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM)
380 ol_flags |= PKT_TX_TCP_CKSUM;
383 get_udptcp_checksum(l3_hdr, tcp_hdr,
386 if (info->gso_enable)
387 ol_flags |= PKT_TX_TCP_SEG;
388 } else if (info->l4_proto == IPPROTO_SCTP) {
389 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
391 /* sctp payload must be a multiple of 4 to be
393 if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_SCTP_CKSUM) &&
394 ((ipv4_hdr->total_length & 0x3) == 0)) {
395 ol_flags |= PKT_TX_SCTP_CKSUM;
397 /* XXX implement CRC32c, example available in
405 /* Calculate the checksum of outer header */
407 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
408 uint16_t testpmd_ol_flags, int tso_enabled)
410 struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
411 struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
412 struct udp_hdr *udp_hdr;
413 uint64_t ol_flags = 0;
415 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
416 ipv4_hdr->hdr_checksum = 0;
417 ol_flags |= PKT_TX_OUTER_IPV4;
419 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
420 ol_flags |= PKT_TX_OUTER_IP_CKSUM;
422 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
424 ol_flags |= PKT_TX_OUTER_IPV6;
426 if (info->outer_l4_proto != IPPROTO_UDP)
429 udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
431 /* outer UDP checksum is done in software as we have no hardware
432 * supporting it today, and no API for it. In the other side, for
433 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
436 * If a packet will be TSOed into small packets by NIC, we cannot
437 * set/calculate a non-zero checksum, because it will be a wrong
438 * value after the packet be split into several small packets.
441 udp_hdr->dgram_cksum = 0;
443 /* do not recalculate udp cksum if it was 0 */
444 if (udp_hdr->dgram_cksum != 0) {
445 udp_hdr->dgram_cksum = 0;
446 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
447 udp_hdr->dgram_cksum =
448 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
450 udp_hdr->dgram_cksum =
451 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
459 * Performs actual copying.
460 * Returns number of segments in the destination mbuf on success,
461 * or negative error code on failure.
464 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
465 uint16_t seglen[], uint8_t nb_seg)
467 uint32_t dlen, slen, tlen;
469 const struct rte_mbuf *m;
482 while (ms != NULL && i != nb_seg) {
485 slen = rte_pktmbuf_data_len(ms);
486 src = rte_pktmbuf_mtod(ms, const uint8_t *);
490 dlen = RTE_MIN(seglen[i], slen);
491 md[i]->data_len = dlen;
492 md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
493 dst = rte_pktmbuf_mtod(md[i], uint8_t *);
496 len = RTE_MIN(slen, dlen);
497 memcpy(dst, src, len);
512 else if (tlen != m->pkt_len)
515 md[0]->nb_segs = nb_seg;
516 md[0]->pkt_len = tlen;
517 md[0]->vlan_tci = m->vlan_tci;
518 md[0]->vlan_tci_outer = m->vlan_tci_outer;
519 md[0]->ol_flags = m->ol_flags;
520 md[0]->tx_offload = m->tx_offload;
526 * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
527 * Copy packet contents and offload information into then new segmented mbuf.
529 static struct rte_mbuf *
530 pkt_copy_split(const struct rte_mbuf *pkt)
533 uint32_t i, len, nb_seg;
534 struct rte_mempool *mp;
535 uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
536 struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
538 mp = current_fwd_lcore()->mbp;
540 if (tx_pkt_split == TX_PKT_SPLIT_RND)
541 nb_seg = random() % tx_pkt_nb_segs + 1;
543 nb_seg = tx_pkt_nb_segs;
545 memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
547 /* calculate number of segments to use and their length. */
549 for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
554 n = pkt->pkt_len - len;
556 /* update size of the last segment to fit rest of the packet */
564 p = rte_pktmbuf_alloc(mp);
567 "failed to allocate %u-th of %u mbuf "
568 "from mempool: %s\n",
569 nb_seg - i, nb_seg, mp->name);
574 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
575 RTE_LOG(ERR, USER1, "mempool %s, %u-th segment: "
576 "expected seglen: %u, "
577 "actual mbuf tailroom: %u\n",
578 mp->name, i, seglen[i],
579 rte_pktmbuf_tailroom(md[i]));
584 /* all mbufs successfully allocated, do copy */
586 rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
589 "mbuf_copy_split for %p(len=%u, nb_seg=%u) "
590 "into %u segments failed with error code: %d\n",
591 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
593 /* figure out how many mbufs to free. */
597 /* free unused mbufs */
598 for (; i != nb_seg; i++) {
599 rte_pktmbuf_free_seg(md[i]);
607 * Receive a burst of packets, and for each packet:
608 * - parse packet, and try to recognize a supported packet type (1)
609 * - if it's not a supported packet type, don't touch the packet, else:
610 * - reprocess the checksum of all supported layers. This is done in SW
611 * or HW, depending on testpmd command line configuration
612 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
613 * segmentation offload (this implies HW TCP checksum)
614 * Then transmit packets on the output port.
616 * (1) Supported packets are:
617 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
618 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
620 * Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
621 * Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
622 * Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
624 * The testpmd command line for this forward engine sets the flags
625 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
626 * wether a checksum must be calculated in software or in hardware. The
627 * IP, UDP, TCP and SCTP flags always concern the inner layer. The
628 * OUTER_IP is only useful for tunnel packets.
631 pkt_burst_checksum_forward(struct fwd_stream *fs)
633 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
634 struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST];
635 struct rte_gso_ctx *gso_ctx;
636 struct rte_mbuf **tx_pkts_burst;
637 struct rte_port *txp;
638 struct rte_mbuf *m, *p;
639 struct ether_hdr *eth_hdr;
640 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
642 uint16_t gro_pkts_num;
648 uint64_t rx_ol_flags, tx_ol_flags;
649 uint16_t testpmd_ol_flags;
651 uint32_t rx_bad_ip_csum;
652 uint32_t rx_bad_l4_csum;
653 struct testpmd_offload_info info;
654 uint16_t nb_segments = 0;
657 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
660 uint64_t core_cycles;
663 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
664 start_tsc = rte_rdtsc();
667 /* receive a burst of packet */
668 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
670 if (unlikely(nb_rx == 0))
672 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
673 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
675 fs->rx_packets += nb_rx;
678 gro_enable = gro_ports[fs->rx_port].enable;
680 txp = &ports[fs->tx_port];
681 testpmd_ol_flags = txp->tx_ol_flags;
682 memset(&info, 0, sizeof(info));
683 info.tso_segsz = txp->tso_segsz;
684 info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
685 if (gso_ports[fs->tx_port].enable)
688 for (i = 0; i < nb_rx; i++) {
689 if (likely(i < nb_rx - 1))
690 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
695 info.pkt_len = rte_pktmbuf_pkt_len(m);
697 rx_ol_flags = m->ol_flags;
699 /* Update the L3/L4 checksum error packet statistics */
700 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
702 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
705 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
706 * and inner headers */
708 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
709 ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
711 ether_addr_copy(&ports[fs->tx_port].eth_addr,
713 parse_ethernet(eth_hdr, &info);
714 l3_hdr = (char *)eth_hdr + info.l2_len;
716 /* check if it's a supported tunnel */
717 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_PARSE_TUNNEL) {
718 if (info.l4_proto == IPPROTO_UDP) {
719 struct udp_hdr *udp_hdr;
721 udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
723 parse_vxlan(udp_hdr, &info, m->packet_type);
725 tx_ol_flags |= PKT_TX_TUNNEL_VXLAN;
726 } else if (info.l4_proto == IPPROTO_GRE) {
727 struct simple_gre_hdr *gre_hdr;
729 gre_hdr = (struct simple_gre_hdr *)
730 ((char *)l3_hdr + info.l3_len);
731 parse_gre(gre_hdr, &info);
733 tx_ol_flags |= PKT_TX_TUNNEL_GRE;
734 } else if (info.l4_proto == IPPROTO_IPIP) {
737 encap_ip_hdr = (char *)l3_hdr + info.l3_len;
738 parse_encap_ip(encap_ip_hdr, &info);
740 tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
744 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
745 if (info.is_tunnel) {
746 outer_l3_hdr = l3_hdr;
747 l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
750 /* step 2: depending on user command line configuration,
751 * recompute checksum either in software or flag the
752 * mbuf to offload the calculation to the NIC. If TSO
753 * is configured, prepare the mbuf for TCP segmentation. */
755 /* process checksums of inner headers first */
756 tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
759 /* Then process outer headers if any. Note that the software
760 * checksum will be wrong if one of the inner checksums is
761 * processed in hardware. */
762 if (info.is_tunnel == 1) {
763 tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
765 !!(tx_ol_flags & PKT_TX_TCP_SEG));
768 /* step 3: fill the mbuf meta data (flags and header lengths) */
770 if (info.is_tunnel == 1) {
771 if (info.tunnel_tso_segsz ||
773 TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) ||
774 (tx_ol_flags & PKT_TX_OUTER_IPV6)) {
775 m->outer_l2_len = info.outer_l2_len;
776 m->outer_l3_len = info.outer_l3_len;
777 m->l2_len = info.l2_len;
778 m->l3_len = info.l3_len;
779 m->l4_len = info.l4_len;
780 m->tso_segsz = info.tunnel_tso_segsz;
783 /* if there is a outer UDP cksum
784 processed in sw and the inner in hw,
785 the outer checksum will be wrong as
786 the payload will be modified by the
788 m->l2_len = info.outer_l2_len +
789 info.outer_l3_len + info.l2_len;
790 m->l3_len = info.l3_len;
791 m->l4_len = info.l4_len;
794 /* this is only useful if an offload flag is
795 * set, but it does not hurt to fill it in any
797 m->l2_len = info.l2_len;
798 m->l3_len = info.l3_len;
799 m->l4_len = info.l4_len;
800 m->tso_segsz = info.tso_segsz;
802 m->ol_flags = tx_ol_flags;
804 /* Do split & copy for the packet. */
805 if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
806 p = pkt_copy_split(m);
814 /* if verbose mode is enabled, dump debug info */
815 if (verbose_level > 0) {
818 printf("-----------------\n");
819 printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
820 fs->rx_port, m, m->pkt_len, m->nb_segs);
821 /* dump rx parsed packet info */
822 rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
823 printf("rx: l2_len=%d ethertype=%x l3_len=%d "
824 "l4_proto=%d l4_len=%d flags=%s\n",
825 info.l2_len, rte_be_to_cpu_16(info.ethertype),
826 info.l3_len, info.l4_proto, info.l4_len, buf);
827 if (rx_ol_flags & PKT_RX_LRO)
828 printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
829 if (info.is_tunnel == 1)
830 printf("rx: outer_l2_len=%d outer_ethertype=%x "
831 "outer_l3_len=%d\n", info.outer_l2_len,
832 rte_be_to_cpu_16(info.outer_ethertype),
834 /* dump tx packet info */
835 if ((testpmd_ol_flags & (TESTPMD_TX_OFFLOAD_IP_CKSUM |
836 TESTPMD_TX_OFFLOAD_UDP_CKSUM |
837 TESTPMD_TX_OFFLOAD_TCP_CKSUM |
838 TESTPMD_TX_OFFLOAD_SCTP_CKSUM)) ||
840 printf("tx: m->l2_len=%d m->l3_len=%d "
842 m->l2_len, m->l3_len, m->l4_len);
843 if (info.is_tunnel == 1) {
844 if ((testpmd_ol_flags &
845 TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) ||
846 (tx_ol_flags & PKT_TX_OUTER_IPV6))
847 printf("tx: m->outer_l2_len=%d "
848 "m->outer_l3_len=%d\n",
851 if (info.tunnel_tso_segsz != 0 &&
852 (m->ol_flags & PKT_TX_TCP_SEG))
853 printf("tx: m->tso_segsz=%d\n",
855 } else if (info.tso_segsz != 0 &&
856 (m->ol_flags & PKT_TX_TCP_SEG))
857 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
858 rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
859 printf("tx: flags=%s", buf);
864 if (unlikely(gro_enable)) {
865 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
866 nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx,
867 &(gro_ports[fs->rx_port].param));
869 gro_ctx = current_fwd_lcore()->gro_ctx;
870 nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx);
872 if (++fs->gro_times >= gro_flush_cycles) {
873 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx);
874 if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
875 gro_pkts_num = MAX_PKT_BURST - nb_rx;
877 nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
886 if (gso_ports[fs->tx_port].enable == 0)
887 tx_pkts_burst = pkts_burst;
889 gso_ctx = &(current_fwd_lcore()->gso_ctx);
890 gso_ctx->gso_size = gso_max_segment_size;
891 for (i = 0; i < nb_rx; i++) {
892 ret = rte_gso_segment(pkts_burst[i], gso_ctx,
893 &gso_segments[nb_segments],
894 GSO_MAX_PKT_BURST - nb_segments);
898 RTE_LOG(DEBUG, USER1,
899 "Unable to segment packet");
900 rte_pktmbuf_free(pkts_burst[i]);
904 tx_pkts_burst = gso_segments;
908 nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
909 tx_pkts_burst, nb_rx);
910 if (nb_prep != nb_rx)
911 printf("Preparing packet burst to transmit failed: %s\n",
912 rte_strerror(rte_errno));
914 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
920 if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
922 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
923 rte_delay_us(burst_tx_delay_time);
924 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
925 &tx_pkts_burst[nb_tx], nb_rx - nb_tx);
928 fs->tx_packets += nb_tx;
929 fs->rx_bad_ip_csum += rx_bad_ip_csum;
930 fs->rx_bad_l4_csum += rx_bad_l4_csum;
932 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
933 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
935 if (unlikely(nb_tx < nb_rx)) {
936 fs->fwd_dropped += (nb_rx - nb_tx);
938 rte_pktmbuf_free(tx_pkts_burst[nb_tx]);
939 } while (++nb_tx < nb_rx);
942 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
943 end_tsc = rte_rdtsc();
944 core_cycles = (end_tsc - start_tsc);
945 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
949 struct fwd_engine csum_fwd_engine = {
950 .fwd_mode_name = "csum",
951 .port_fwd_begin = NULL,
952 .port_fwd_end = NULL,
953 .packet_fwd = pkt_burst_checksum_forward,