1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright(c) 2010-2014 Intel Corporation.
3 * Copyright 2014 6WIND S.A.
13 #include <sys/queue.h>
16 #include <rte_common.h>
17 #include <rte_byteorder.h>
19 #include <rte_debug.h>
20 #include <rte_cycles.h>
21 #include <rte_memory.h>
22 #include <rte_memcpy.h>
23 #include <rte_launch.h>
25 #include <rte_per_lcore.h>
26 #include <rte_lcore.h>
27 #include <rte_atomic.h>
28 #include <rte_branch_prediction.h>
29 #include <rte_mempool.h>
31 #include <rte_interrupts.h>
33 #include <rte_ether.h>
34 #include <rte_ethdev.h>
39 #include <rte_prefetch.h>
40 #include <rte_string_fns.h>
47 #define IP_DEFTTL 64 /* from RFC 1340. */
48 #define IP_VERSION 0x40
49 #define IP_HDRLEN 0x05 /* default IP header length == five 32-bits words. */
50 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
52 #define GRE_KEY_PRESENT 0x2000
54 #define GRE_SUPPORTED_FIELDS GRE_KEY_PRESENT
56 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */
57 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
58 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
63 uint16_t vxlan_gpe_udp_port = 4790;
65 /* structure that caches offload info for the current packet */
66 struct testpmd_offload_info {
74 uint16_t outer_ethertype;
75 uint16_t outer_l2_len;
76 uint16_t outer_l3_len;
77 uint8_t outer_l4_proto;
79 uint16_t tunnel_tso_segsz;
83 /* simplified GRE header */
84 struct simple_gre_hdr {
87 } __attribute__((__packed__));
90 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
92 if (ethertype == _htons(ETHER_TYPE_IPv4))
93 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
94 else /* assume ethertype == ETHER_TYPE_IPv6 */
95 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
98 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
100 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
102 struct tcp_hdr *tcp_hdr;
104 info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
105 info->l4_proto = ipv4_hdr->next_proto_id;
107 /* only fill l4_len for TCP, it's useful for TSO */
108 if (info->l4_proto == IPPROTO_TCP) {
109 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
110 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
115 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
117 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
119 struct tcp_hdr *tcp_hdr;
121 info->l3_len = sizeof(struct ipv6_hdr);
122 info->l4_proto = ipv6_hdr->proto;
124 /* only fill l4_len for TCP, it's useful for TSO */
125 if (info->l4_proto == IPPROTO_TCP) {
126 tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
127 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
133 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
134 * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
135 * header. The l4_len argument is only set in case of TCP (useful for TSO).
138 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info)
140 struct ipv4_hdr *ipv4_hdr;
141 struct ipv6_hdr *ipv6_hdr;
143 info->l2_len = sizeof(struct ether_hdr);
144 info->ethertype = eth_hdr->ether_type;
146 if (info->ethertype == _htons(ETHER_TYPE_VLAN)) {
147 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
149 info->l2_len += sizeof(struct vlan_hdr);
150 info->ethertype = vlan_hdr->eth_proto;
153 switch (info->ethertype) {
154 case _htons(ETHER_TYPE_IPv4):
155 ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
156 parse_ipv4(ipv4_hdr, info);
158 case _htons(ETHER_TYPE_IPv6):
159 ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
160 parse_ipv6(ipv6_hdr, info);
170 /* Parse a vxlan header */
172 parse_vxlan(struct udp_hdr *udp_hdr,
173 struct testpmd_offload_info *info,
176 struct ether_hdr *eth_hdr;
178 /* check udp destination port, 4789 is the default vxlan port
179 * (rfc7348) or that the rx offload flag is set (i40e only
181 if (udp_hdr->dst_port != _htons(4789) &&
182 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
186 info->outer_ethertype = info->ethertype;
187 info->outer_l2_len = info->l2_len;
188 info->outer_l3_len = info->l3_len;
189 info->outer_l4_proto = info->l4_proto;
191 eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
192 sizeof(struct udp_hdr) +
193 sizeof(struct vxlan_hdr));
195 parse_ethernet(eth_hdr, info);
196 info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */
199 /* Parse a vxlan-gpe header */
201 parse_vxlan_gpe(struct udp_hdr *udp_hdr,
202 struct testpmd_offload_info *info)
204 struct ether_hdr *eth_hdr;
205 struct ipv4_hdr *ipv4_hdr;
206 struct ipv6_hdr *ipv6_hdr;
207 struct vxlan_gpe_hdr *vxlan_gpe_hdr;
208 uint8_t vxlan_gpe_len = sizeof(*vxlan_gpe_hdr);
210 /* Check udp destination port. */
211 if (udp_hdr->dst_port != _htons(vxlan_gpe_udp_port))
214 vxlan_gpe_hdr = (struct vxlan_gpe_hdr *)((char *)udp_hdr +
215 sizeof(struct udp_hdr));
217 if (!vxlan_gpe_hdr->proto || vxlan_gpe_hdr->proto ==
218 VXLAN_GPE_TYPE_IPV4) {
220 info->outer_ethertype = info->ethertype;
221 info->outer_l2_len = info->l2_len;
222 info->outer_l3_len = info->l3_len;
223 info->outer_l4_proto = info->l4_proto;
225 ipv4_hdr = (struct ipv4_hdr *)((char *)vxlan_gpe_hdr +
228 parse_ipv4(ipv4_hdr, info);
229 info->ethertype = _htons(ETHER_TYPE_IPv4);
232 } else if (vxlan_gpe_hdr->proto == VXLAN_GPE_TYPE_IPV6) {
234 info->outer_ethertype = info->ethertype;
235 info->outer_l2_len = info->l2_len;
236 info->outer_l3_len = info->l3_len;
237 info->outer_l4_proto = info->l4_proto;
239 ipv6_hdr = (struct ipv6_hdr *)((char *)vxlan_gpe_hdr +
242 info->ethertype = _htons(ETHER_TYPE_IPv6);
243 parse_ipv6(ipv6_hdr, info);
246 } else if (vxlan_gpe_hdr->proto == VXLAN_GPE_TYPE_ETH) {
248 info->outer_ethertype = info->ethertype;
249 info->outer_l2_len = info->l2_len;
250 info->outer_l3_len = info->l3_len;
251 info->outer_l4_proto = info->l4_proto;
253 eth_hdr = (struct ether_hdr *)((char *)vxlan_gpe_hdr +
256 parse_ethernet(eth_hdr, info);
260 info->l2_len += ETHER_VXLAN_GPE_HLEN;
263 /* Parse a gre header */
265 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
267 struct ether_hdr *eth_hdr;
268 struct ipv4_hdr *ipv4_hdr;
269 struct ipv6_hdr *ipv6_hdr;
272 /* check which fields are supported */
273 if ((gre_hdr->flags & _htons(~GRE_SUPPORTED_FIELDS)) != 0)
276 gre_len += sizeof(struct simple_gre_hdr);
278 if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
279 gre_len += GRE_KEY_LEN;
281 if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) {
283 info->outer_ethertype = info->ethertype;
284 info->outer_l2_len = info->l2_len;
285 info->outer_l3_len = info->l3_len;
286 info->outer_l4_proto = info->l4_proto;
288 ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len);
290 parse_ipv4(ipv4_hdr, info);
291 info->ethertype = _htons(ETHER_TYPE_IPv4);
294 } else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) {
296 info->outer_ethertype = info->ethertype;
297 info->outer_l2_len = info->l2_len;
298 info->outer_l3_len = info->l3_len;
299 info->outer_l4_proto = info->l4_proto;
301 ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len);
303 info->ethertype = _htons(ETHER_TYPE_IPv6);
304 parse_ipv6(ipv6_hdr, info);
307 } else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) {
309 info->outer_ethertype = info->ethertype;
310 info->outer_l2_len = info->l2_len;
311 info->outer_l3_len = info->l3_len;
312 info->outer_l4_proto = info->l4_proto;
314 eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len);
316 parse_ethernet(eth_hdr, info);
320 info->l2_len += gre_len;
324 /* Parse an encapsulated ip or ipv6 header */
326 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
328 struct ipv4_hdr *ipv4_hdr = encap_ip;
329 struct ipv6_hdr *ipv6_hdr = encap_ip;
332 ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
334 if (ip_version != 4 && ip_version != 6)
338 info->outer_ethertype = info->ethertype;
339 info->outer_l2_len = info->l2_len;
340 info->outer_l3_len = info->l3_len;
342 if (ip_version == 4) {
343 parse_ipv4(ipv4_hdr, info);
344 info->ethertype = _htons(ETHER_TYPE_IPv4);
346 parse_ipv6(ipv6_hdr, info);
347 info->ethertype = _htons(ETHER_TYPE_IPv6);
352 /* if possible, calculate the checksum of a packet in hw or sw,
353 * depending on the testpmd command line configuration */
355 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
356 uint64_t tx_offloads)
358 struct ipv4_hdr *ipv4_hdr = l3_hdr;
359 struct udp_hdr *udp_hdr;
360 struct tcp_hdr *tcp_hdr;
361 struct sctp_hdr *sctp_hdr;
362 uint64_t ol_flags = 0;
363 uint32_t max_pkt_len, tso_segsz = 0;
365 /* ensure packet is large enough to require tso */
366 if (!info->is_tunnel) {
367 max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
369 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
370 tso_segsz = info->tso_segsz;
372 max_pkt_len = info->outer_l2_len + info->outer_l3_len +
373 info->l2_len + info->l3_len + info->l4_len +
374 info->tunnel_tso_segsz;
375 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
376 tso_segsz = info->tunnel_tso_segsz;
379 if (info->ethertype == _htons(ETHER_TYPE_IPv4)) {
381 ipv4_hdr->hdr_checksum = 0;
383 ol_flags |= PKT_TX_IPV4;
384 if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
385 ol_flags |= PKT_TX_IP_CKSUM;
387 if (tx_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
388 ol_flags |= PKT_TX_IP_CKSUM;
390 ipv4_hdr->hdr_checksum =
391 rte_ipv4_cksum(ipv4_hdr);
393 } else if (info->ethertype == _htons(ETHER_TYPE_IPv6))
394 ol_flags |= PKT_TX_IPV6;
396 return 0; /* packet type not supported, nothing to do */
398 if (info->l4_proto == IPPROTO_UDP) {
399 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
400 /* do not recalculate udp cksum if it was 0 */
401 if (udp_hdr->dgram_cksum != 0) {
402 udp_hdr->dgram_cksum = 0;
403 if (tx_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)
404 ol_flags |= PKT_TX_UDP_CKSUM;
406 udp_hdr->dgram_cksum =
407 get_udptcp_checksum(l3_hdr, udp_hdr,
411 } else if (info->l4_proto == IPPROTO_TCP) {
412 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
415 ol_flags |= PKT_TX_TCP_SEG;
416 else if (tx_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)
417 ol_flags |= PKT_TX_TCP_CKSUM;
420 get_udptcp_checksum(l3_hdr, tcp_hdr,
423 if (info->gso_enable)
424 ol_flags |= PKT_TX_TCP_SEG;
425 } else if (info->l4_proto == IPPROTO_SCTP) {
426 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
428 /* sctp payload must be a multiple of 4 to be
430 if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) &&
431 ((ipv4_hdr->total_length & 0x3) == 0)) {
432 ol_flags |= PKT_TX_SCTP_CKSUM;
434 /* XXX implement CRC32c, example available in
442 /* Calculate the checksum of outer header */
444 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
445 uint64_t tx_offloads, int tso_enabled)
447 struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
448 struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
449 struct udp_hdr *udp_hdr;
450 uint64_t ol_flags = 0;
452 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
453 ipv4_hdr->hdr_checksum = 0;
454 ol_flags |= PKT_TX_OUTER_IPV4;
456 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)
457 ol_flags |= PKT_TX_OUTER_IP_CKSUM;
459 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
461 ol_flags |= PKT_TX_OUTER_IPV6;
463 if (info->outer_l4_proto != IPPROTO_UDP)
466 udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
468 /* outer UDP checksum is done in software as we have no hardware
469 * supporting it today, and no API for it. In the other side, for
470 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
473 * If a packet will be TSOed into small packets by NIC, we cannot
474 * set/calculate a non-zero checksum, because it will be a wrong
475 * value after the packet be split into several small packets.
478 udp_hdr->dgram_cksum = 0;
480 /* do not recalculate udp cksum if it was 0 */
481 if (udp_hdr->dgram_cksum != 0) {
482 udp_hdr->dgram_cksum = 0;
483 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
484 udp_hdr->dgram_cksum =
485 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
487 udp_hdr->dgram_cksum =
488 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
496 * Performs actual copying.
497 * Returns number of segments in the destination mbuf on success,
498 * or negative error code on failure.
501 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
502 uint16_t seglen[], uint8_t nb_seg)
504 uint32_t dlen, slen, tlen;
506 const struct rte_mbuf *m;
519 while (ms != NULL && i != nb_seg) {
522 slen = rte_pktmbuf_data_len(ms);
523 src = rte_pktmbuf_mtod(ms, const uint8_t *);
527 dlen = RTE_MIN(seglen[i], slen);
528 md[i]->data_len = dlen;
529 md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
530 dst = rte_pktmbuf_mtod(md[i], uint8_t *);
533 len = RTE_MIN(slen, dlen);
534 memcpy(dst, src, len);
549 else if (tlen != m->pkt_len)
552 md[0]->nb_segs = nb_seg;
553 md[0]->pkt_len = tlen;
554 md[0]->vlan_tci = m->vlan_tci;
555 md[0]->vlan_tci_outer = m->vlan_tci_outer;
556 md[0]->ol_flags = m->ol_flags;
557 md[0]->tx_offload = m->tx_offload;
563 * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
564 * Copy packet contents and offload information into then new segmented mbuf.
566 static struct rte_mbuf *
567 pkt_copy_split(const struct rte_mbuf *pkt)
570 uint32_t i, len, nb_seg;
571 struct rte_mempool *mp;
572 uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
573 struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
575 mp = current_fwd_lcore()->mbp;
577 if (tx_pkt_split == TX_PKT_SPLIT_RND)
578 nb_seg = random() % tx_pkt_nb_segs + 1;
580 nb_seg = tx_pkt_nb_segs;
582 memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
584 /* calculate number of segments to use and their length. */
586 for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
591 n = pkt->pkt_len - len;
593 /* update size of the last segment to fit rest of the packet */
601 p = rte_pktmbuf_alloc(mp);
604 "failed to allocate %u-th of %u mbuf "
605 "from mempool: %s\n",
606 nb_seg - i, nb_seg, mp->name);
611 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
612 TESTPMD_LOG(ERR, "mempool %s, %u-th segment: "
613 "expected seglen: %u, "
614 "actual mbuf tailroom: %u\n",
615 mp->name, i, seglen[i],
616 rte_pktmbuf_tailroom(md[i]));
621 /* all mbufs successfully allocated, do copy */
623 rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
626 "mbuf_copy_split for %p(len=%u, nb_seg=%u) "
627 "into %u segments failed with error code: %d\n",
628 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
630 /* figure out how many mbufs to free. */
634 /* free unused mbufs */
635 for (; i != nb_seg; i++) {
636 rte_pktmbuf_free_seg(md[i]);
644 * Receive a burst of packets, and for each packet:
645 * - parse packet, and try to recognize a supported packet type (1)
646 * - if it's not a supported packet type, don't touch the packet, else:
647 * - reprocess the checksum of all supported layers. This is done in SW
648 * or HW, depending on testpmd command line configuration
649 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
650 * segmentation offload (this implies HW TCP checksum)
651 * Then transmit packets on the output port.
653 * (1) Supported packets are:
654 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
655 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
657 * Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 /
659 * Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 /
661 * Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
662 * Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
663 * Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
665 * The testpmd command line for this forward engine sets the flags
666 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
667 * wether a checksum must be calculated in software or in hardware. The
668 * IP, UDP, TCP and SCTP flags always concern the inner layer. The
669 * OUTER_IP is only useful for tunnel packets.
672 pkt_burst_checksum_forward(struct fwd_stream *fs)
674 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
675 struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST];
676 struct rte_gso_ctx *gso_ctx;
677 struct rte_mbuf **tx_pkts_burst;
678 struct rte_port *txp;
679 struct rte_mbuf *m, *p;
680 struct ether_hdr *eth_hdr;
681 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
683 uint16_t gro_pkts_num;
689 uint64_t rx_ol_flags, tx_ol_flags;
690 uint64_t tx_offloads;
692 uint32_t rx_bad_ip_csum;
693 uint32_t rx_bad_l4_csum;
694 struct testpmd_offload_info info;
695 uint16_t nb_segments = 0;
698 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
701 uint64_t core_cycles;
704 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
705 start_tsc = rte_rdtsc();
708 /* receive a burst of packet */
709 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
711 if (unlikely(nb_rx == 0))
713 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
714 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
716 fs->rx_packets += nb_rx;
719 gro_enable = gro_ports[fs->rx_port].enable;
721 txp = &ports[fs->tx_port];
722 tx_offloads = txp->dev_conf.txmode.offloads;
723 memset(&info, 0, sizeof(info));
724 info.tso_segsz = txp->tso_segsz;
725 info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
726 if (gso_ports[fs->tx_port].enable)
729 for (i = 0; i < nb_rx; i++) {
730 if (likely(i < nb_rx - 1))
731 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
736 info.pkt_len = rte_pktmbuf_pkt_len(m);
738 rx_ol_flags = m->ol_flags;
740 /* Update the L3/L4 checksum error packet statistics */
741 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
743 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
746 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
747 * and inner headers */
749 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
750 ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
752 ether_addr_copy(&ports[fs->tx_port].eth_addr,
754 parse_ethernet(eth_hdr, &info);
755 l3_hdr = (char *)eth_hdr + info.l2_len;
757 /* check if it's a supported tunnel */
758 if (txp->parse_tunnel) {
759 if (info.l4_proto == IPPROTO_UDP) {
760 struct udp_hdr *udp_hdr;
762 udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
764 parse_vxlan_gpe(udp_hdr, &info);
765 if (info.is_tunnel) {
766 tx_ol_flags |= PKT_TX_TUNNEL_VXLAN_GPE;
768 parse_vxlan(udp_hdr, &info,
774 } else if (info.l4_proto == IPPROTO_GRE) {
775 struct simple_gre_hdr *gre_hdr;
777 gre_hdr = (struct simple_gre_hdr *)
778 ((char *)l3_hdr + info.l3_len);
779 parse_gre(gre_hdr, &info);
781 tx_ol_flags |= PKT_TX_TUNNEL_GRE;
782 } else if (info.l4_proto == IPPROTO_IPIP) {
785 encap_ip_hdr = (char *)l3_hdr + info.l3_len;
786 parse_encap_ip(encap_ip_hdr, &info);
788 tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
792 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
793 if (info.is_tunnel) {
794 outer_l3_hdr = l3_hdr;
795 l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
798 /* step 2: depending on user command line configuration,
799 * recompute checksum either in software or flag the
800 * mbuf to offload the calculation to the NIC. If TSO
801 * is configured, prepare the mbuf for TCP segmentation. */
803 /* process checksums of inner headers first */
804 tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
807 /* Then process outer headers if any. Note that the software
808 * checksum will be wrong if one of the inner checksums is
809 * processed in hardware. */
810 if (info.is_tunnel == 1) {
811 tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
813 !!(tx_ol_flags & PKT_TX_TCP_SEG));
816 /* step 3: fill the mbuf meta data (flags and header lengths) */
818 if (info.is_tunnel == 1) {
819 if (info.tunnel_tso_segsz ||
821 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
822 (tx_ol_flags & PKT_TX_OUTER_IPV6)) {
823 m->outer_l2_len = info.outer_l2_len;
824 m->outer_l3_len = info.outer_l3_len;
825 m->l2_len = info.l2_len;
826 m->l3_len = info.l3_len;
827 m->l4_len = info.l4_len;
828 m->tso_segsz = info.tunnel_tso_segsz;
831 /* if there is a outer UDP cksum
832 processed in sw and the inner in hw,
833 the outer checksum will be wrong as
834 the payload will be modified by the
836 m->l2_len = info.outer_l2_len +
837 info.outer_l3_len + info.l2_len;
838 m->l3_len = info.l3_len;
839 m->l4_len = info.l4_len;
842 /* this is only useful if an offload flag is
843 * set, but it does not hurt to fill it in any
845 m->l2_len = info.l2_len;
846 m->l3_len = info.l3_len;
847 m->l4_len = info.l4_len;
848 m->tso_segsz = info.tso_segsz;
850 m->ol_flags = tx_ol_flags;
852 /* Do split & copy for the packet. */
853 if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
854 p = pkt_copy_split(m);
862 /* if verbose mode is enabled, dump debug info */
863 if (verbose_level > 0) {
866 printf("-----------------\n");
867 printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
868 fs->rx_port, m, m->pkt_len, m->nb_segs);
869 /* dump rx parsed packet info */
870 rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
871 printf("rx: l2_len=%d ethertype=%x l3_len=%d "
872 "l4_proto=%d l4_len=%d flags=%s\n",
873 info.l2_len, rte_be_to_cpu_16(info.ethertype),
874 info.l3_len, info.l4_proto, info.l4_len, buf);
875 if (rx_ol_flags & PKT_RX_LRO)
876 printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
877 if (info.is_tunnel == 1)
878 printf("rx: outer_l2_len=%d outer_ethertype=%x "
879 "outer_l3_len=%d\n", info.outer_l2_len,
880 rte_be_to_cpu_16(info.outer_ethertype),
882 /* dump tx packet info */
883 if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
884 DEV_TX_OFFLOAD_UDP_CKSUM |
885 DEV_TX_OFFLOAD_TCP_CKSUM |
886 DEV_TX_OFFLOAD_SCTP_CKSUM)) ||
888 printf("tx: m->l2_len=%d m->l3_len=%d "
890 m->l2_len, m->l3_len, m->l4_len);
891 if (info.is_tunnel == 1) {
893 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
894 (tx_ol_flags & PKT_TX_OUTER_IPV6))
895 printf("tx: m->outer_l2_len=%d "
896 "m->outer_l3_len=%d\n",
899 if (info.tunnel_tso_segsz != 0 &&
900 (m->ol_flags & PKT_TX_TCP_SEG))
901 printf("tx: m->tso_segsz=%d\n",
903 } else if (info.tso_segsz != 0 &&
904 (m->ol_flags & PKT_TX_TCP_SEG))
905 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
906 rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
907 printf("tx: flags=%s", buf);
912 if (unlikely(gro_enable)) {
913 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
914 nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx,
915 &(gro_ports[fs->rx_port].param));
917 gro_ctx = current_fwd_lcore()->gro_ctx;
918 nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx);
920 if (++fs->gro_times >= gro_flush_cycles) {
921 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx);
922 if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
923 gro_pkts_num = MAX_PKT_BURST - nb_rx;
925 nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
934 if (gso_ports[fs->tx_port].enable == 0)
935 tx_pkts_burst = pkts_burst;
937 gso_ctx = &(current_fwd_lcore()->gso_ctx);
938 gso_ctx->gso_size = gso_max_segment_size;
939 for (i = 0; i < nb_rx; i++) {
940 ret = rte_gso_segment(pkts_burst[i], gso_ctx,
941 &gso_segments[nb_segments],
942 GSO_MAX_PKT_BURST - nb_segments);
946 TESTPMD_LOG(DEBUG, "Unable to segment packet");
947 rte_pktmbuf_free(pkts_burst[i]);
951 tx_pkts_burst = gso_segments;
955 nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
956 tx_pkts_burst, nb_rx);
957 if (nb_prep != nb_rx)
958 printf("Preparing packet burst to transmit failed: %s\n",
959 rte_strerror(rte_errno));
961 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
967 if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
969 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
970 rte_delay_us(burst_tx_delay_time);
971 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
972 &tx_pkts_burst[nb_tx], nb_rx - nb_tx);
975 fs->tx_packets += nb_tx;
976 fs->rx_bad_ip_csum += rx_bad_ip_csum;
977 fs->rx_bad_l4_csum += rx_bad_l4_csum;
979 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
980 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
982 if (unlikely(nb_tx < nb_rx)) {
983 fs->fwd_dropped += (nb_rx - nb_tx);
985 rte_pktmbuf_free(tx_pkts_burst[nb_tx]);
986 } while (++nb_tx < nb_rx);
989 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
990 end_tsc = rte_rdtsc();
991 core_cycles = (end_tsc - start_tsc);
992 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
996 struct fwd_engine csum_fwd_engine = {
997 .fwd_mode_name = "csum",
998 .port_fwd_begin = NULL,
999 .port_fwd_end = NULL,
1000 .packet_fwd = pkt_burst_checksum_forward,