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>
38 #include <rte_vxlan.h>
41 #include <rte_prefetch.h>
42 #include <rte_string_fns.h>
46 #include <rte_geneve.h>
50 #define IP_DEFTTL 64 /* from RFC 1340. */
52 #define GRE_CHECKSUM_PRESENT 0x8000
53 #define GRE_KEY_PRESENT 0x2000
54 #define GRE_SEQUENCE_PRESENT 0x1000
56 #define GRE_SUPPORTED_FIELDS (GRE_CHECKSUM_PRESENT | GRE_KEY_PRESENT |\
59 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */
60 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
61 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
66 uint16_t vxlan_gpe_udp_port = RTE_VXLAN_GPE_DEFAULT_PORT;
67 uint16_t geneve_udp_port = RTE_GENEVE_DEFAULT_PORT;
69 /* structure that caches offload info for the current packet */
70 struct testpmd_offload_info {
78 uint16_t outer_ethertype;
79 uint16_t outer_l2_len;
80 uint16_t outer_l3_len;
81 uint8_t outer_l4_proto;
83 uint16_t tunnel_tso_segsz;
87 /* simplified GRE header */
88 struct simple_gre_hdr {
94 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
96 if (ethertype == _htons(RTE_ETHER_TYPE_IPV4))
97 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
98 else /* assume ethertype == RTE_ETHER_TYPE_IPV6 */
99 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
102 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
104 parse_ipv4(struct rte_ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
106 struct rte_tcp_hdr *tcp_hdr;
108 info->l3_len = rte_ipv4_hdr_len(ipv4_hdr);
109 info->l4_proto = ipv4_hdr->next_proto_id;
111 /* only fill l4_len for TCP, it's useful for TSO */
112 if (info->l4_proto == IPPROTO_TCP) {
113 tcp_hdr = (struct rte_tcp_hdr *)
114 ((char *)ipv4_hdr + info->l3_len);
115 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
116 } else if (info->l4_proto == IPPROTO_UDP)
117 info->l4_len = sizeof(struct rte_udp_hdr);
122 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
124 parse_ipv6(struct rte_ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
126 struct rte_tcp_hdr *tcp_hdr;
128 info->l3_len = sizeof(struct rte_ipv6_hdr);
129 info->l4_proto = ipv6_hdr->proto;
131 /* only fill l4_len for TCP, it's useful for TSO */
132 if (info->l4_proto == IPPROTO_TCP) {
133 tcp_hdr = (struct rte_tcp_hdr *)
134 ((char *)ipv6_hdr + info->l3_len);
135 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
136 } else if (info->l4_proto == IPPROTO_UDP)
137 info->l4_len = sizeof(struct rte_udp_hdr);
143 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
144 * ipproto. This function is able to recognize IPv4/IPv6 with optional VLAN
145 * headers. The l4_len argument is only set in case of TCP (useful for TSO).
148 parse_ethernet(struct rte_ether_hdr *eth_hdr, struct testpmd_offload_info *info)
150 struct rte_ipv4_hdr *ipv4_hdr;
151 struct rte_ipv6_hdr *ipv6_hdr;
152 struct rte_vlan_hdr *vlan_hdr;
154 info->l2_len = sizeof(struct rte_ether_hdr);
155 info->ethertype = eth_hdr->ether_type;
157 while (info->ethertype == _htons(RTE_ETHER_TYPE_VLAN) ||
158 info->ethertype == _htons(RTE_ETHER_TYPE_QINQ)) {
159 vlan_hdr = (struct rte_vlan_hdr *)
160 ((char *)eth_hdr + info->l2_len);
161 info->l2_len += sizeof(struct rte_vlan_hdr);
162 info->ethertype = vlan_hdr->eth_proto;
165 switch (info->ethertype) {
166 case _htons(RTE_ETHER_TYPE_IPV4):
167 ipv4_hdr = (struct rte_ipv4_hdr *)
168 ((char *)eth_hdr + info->l2_len);
169 parse_ipv4(ipv4_hdr, info);
171 case _htons(RTE_ETHER_TYPE_IPV6):
172 ipv6_hdr = (struct rte_ipv6_hdr *)
173 ((char *)eth_hdr + info->l2_len);
174 parse_ipv6(ipv6_hdr, info);
184 /* Fill in outer layers length */
186 update_tunnel_outer(struct testpmd_offload_info *info)
189 info->outer_ethertype = info->ethertype;
190 info->outer_l2_len = info->l2_len;
191 info->outer_l3_len = info->l3_len;
192 info->outer_l4_proto = info->l4_proto;
196 * Parse a GTP protocol header.
197 * No optional fields and next extension header type.
200 parse_gtp(struct rte_udp_hdr *udp_hdr,
201 struct testpmd_offload_info *info)
203 struct rte_ipv4_hdr *ipv4_hdr;
204 struct rte_ipv6_hdr *ipv6_hdr;
205 struct rte_gtp_hdr *gtp_hdr;
206 uint8_t gtp_len = sizeof(*gtp_hdr);
209 /* Check udp destination port. */
210 if (udp_hdr->dst_port != _htons(RTE_GTPC_UDP_PORT) &&
211 udp_hdr->src_port != _htons(RTE_GTPC_UDP_PORT) &&
212 udp_hdr->dst_port != _htons(RTE_GTPU_UDP_PORT))
215 update_tunnel_outer(info);
218 gtp_hdr = (struct rte_gtp_hdr *)((char *)udp_hdr +
219 sizeof(struct rte_udp_hdr));
222 * Check message type. If message type is 0xff, it is
223 * a GTP data packet. If not, it is a GTP control packet
225 if (gtp_hdr->msg_type == 0xff) {
226 ip_ver = *(uint8_t *)((char *)udp_hdr +
227 sizeof(struct rte_udp_hdr) +
228 sizeof(struct rte_gtp_hdr));
229 ip_ver = (ip_ver) & 0xf0;
231 if (ip_ver == RTE_GTP_TYPE_IPV4) {
232 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)gtp_hdr +
234 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
235 parse_ipv4(ipv4_hdr, info);
236 } else if (ip_ver == RTE_GTP_TYPE_IPV6) {
237 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)gtp_hdr +
239 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
240 parse_ipv6(ipv6_hdr, info);
249 info->l2_len += RTE_ETHER_GTP_HLEN;
252 /* Parse a vxlan header */
254 parse_vxlan(struct rte_udp_hdr *udp_hdr,
255 struct testpmd_offload_info *info,
258 struct rte_ether_hdr *eth_hdr;
260 /* check udp destination port, RTE_VXLAN_DEFAULT_PORT (4789) is the
261 * default vxlan port (rfc7348) or that the rx offload flag is set
262 * (i40e only currently)
264 if (udp_hdr->dst_port != _htons(RTE_VXLAN_DEFAULT_PORT) &&
265 RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
268 update_tunnel_outer(info);
270 eth_hdr = (struct rte_ether_hdr *)((char *)udp_hdr +
271 sizeof(struct rte_udp_hdr) +
272 sizeof(struct rte_vxlan_hdr));
274 parse_ethernet(eth_hdr, info);
275 info->l2_len += RTE_ETHER_VXLAN_HLEN; /* add udp + vxlan */
278 /* Parse a vxlan-gpe header */
280 parse_vxlan_gpe(struct rte_udp_hdr *udp_hdr,
281 struct testpmd_offload_info *info)
283 struct rte_ether_hdr *eth_hdr;
284 struct rte_ipv4_hdr *ipv4_hdr;
285 struct rte_ipv6_hdr *ipv6_hdr;
286 struct rte_vxlan_gpe_hdr *vxlan_gpe_hdr;
287 uint8_t vxlan_gpe_len = sizeof(*vxlan_gpe_hdr);
289 /* Check udp destination port. */
290 if (udp_hdr->dst_port != _htons(vxlan_gpe_udp_port))
293 vxlan_gpe_hdr = (struct rte_vxlan_gpe_hdr *)((char *)udp_hdr +
294 sizeof(struct rte_udp_hdr));
296 if (!vxlan_gpe_hdr->proto || vxlan_gpe_hdr->proto ==
297 RTE_VXLAN_GPE_TYPE_IPV4) {
298 update_tunnel_outer(info);
300 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)vxlan_gpe_hdr +
303 parse_ipv4(ipv4_hdr, info);
304 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
307 } else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_IPV6) {
308 update_tunnel_outer(info);
310 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)vxlan_gpe_hdr +
313 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
314 parse_ipv6(ipv6_hdr, info);
317 } else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_ETH) {
318 update_tunnel_outer(info);
320 eth_hdr = (struct rte_ether_hdr *)((char *)vxlan_gpe_hdr +
323 parse_ethernet(eth_hdr, info);
327 info->l2_len += RTE_ETHER_VXLAN_GPE_HLEN;
330 /* Parse a geneve header */
332 parse_geneve(struct rte_udp_hdr *udp_hdr,
333 struct testpmd_offload_info *info)
335 struct rte_ether_hdr *eth_hdr;
336 struct rte_ipv4_hdr *ipv4_hdr;
337 struct rte_ipv6_hdr *ipv6_hdr;
338 struct rte_geneve_hdr *geneve_hdr;
341 /* Check udp destination port. */
342 if (udp_hdr->dst_port != _htons(geneve_udp_port))
345 geneve_hdr = (struct rte_geneve_hdr *)((char *)udp_hdr +
346 sizeof(struct rte_udp_hdr));
347 geneve_len = sizeof(struct rte_geneve_hdr) + geneve_hdr->opt_len * 4;
348 if (!geneve_hdr->proto || geneve_hdr->proto ==
349 _htons(RTE_ETHER_TYPE_IPV4)) {
350 update_tunnel_outer(info);
351 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)geneve_hdr +
353 parse_ipv4(ipv4_hdr, info);
354 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
356 } else if (geneve_hdr->proto == _htons(RTE_ETHER_TYPE_IPV6)) {
357 update_tunnel_outer(info);
358 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)geneve_hdr +
360 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
361 parse_ipv6(ipv6_hdr, info);
364 } else if (geneve_hdr->proto == _htons(RTE_GENEVE_TYPE_ETH)) {
365 update_tunnel_outer(info);
366 eth_hdr = (struct rte_ether_hdr *)((char *)geneve_hdr +
368 parse_ethernet(eth_hdr, info);
373 (sizeof(struct rte_udp_hdr) + sizeof(struct rte_geneve_hdr) +
374 ((struct rte_geneve_hdr *)geneve_hdr)->opt_len * 4);
377 /* Parse a gre header */
379 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
381 struct rte_ether_hdr *eth_hdr;
382 struct rte_ipv4_hdr *ipv4_hdr;
383 struct rte_ipv6_hdr *ipv6_hdr;
386 gre_len += sizeof(struct simple_gre_hdr);
388 if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
389 gre_len += GRE_EXT_LEN;
390 if (gre_hdr->flags & _htons(GRE_SEQUENCE_PRESENT))
391 gre_len += GRE_EXT_LEN;
392 if (gre_hdr->flags & _htons(GRE_CHECKSUM_PRESENT))
393 gre_len += GRE_EXT_LEN;
395 if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPV4)) {
396 update_tunnel_outer(info);
398 ipv4_hdr = (struct rte_ipv4_hdr *)((char *)gre_hdr + gre_len);
400 parse_ipv4(ipv4_hdr, info);
401 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
404 } else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPV6)) {
405 update_tunnel_outer(info);
407 ipv6_hdr = (struct rte_ipv6_hdr *)((char *)gre_hdr + gre_len);
409 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
410 parse_ipv6(ipv6_hdr, info);
413 } else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_TEB)) {
414 update_tunnel_outer(info);
416 eth_hdr = (struct rte_ether_hdr *)((char *)gre_hdr + gre_len);
418 parse_ethernet(eth_hdr, info);
422 info->l2_len += gre_len;
426 /* Parse an encapsulated ip or ipv6 header */
428 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
430 struct rte_ipv4_hdr *ipv4_hdr = encap_ip;
431 struct rte_ipv6_hdr *ipv6_hdr = encap_ip;
434 ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
436 if (ip_version != 4 && ip_version != 6)
440 info->outer_ethertype = info->ethertype;
441 info->outer_l2_len = info->l2_len;
442 info->outer_l3_len = info->l3_len;
444 if (ip_version == 4) {
445 parse_ipv4(ipv4_hdr, info);
446 info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
448 parse_ipv6(ipv6_hdr, info);
449 info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
454 /* if possible, calculate the checksum of a packet in hw or sw,
455 * depending on the testpmd command line configuration */
457 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
458 uint64_t tx_offloads)
460 struct rte_ipv4_hdr *ipv4_hdr = l3_hdr;
461 struct rte_udp_hdr *udp_hdr;
462 struct rte_tcp_hdr *tcp_hdr;
463 struct rte_sctp_hdr *sctp_hdr;
464 uint64_t ol_flags = 0;
465 uint32_t max_pkt_len, tso_segsz = 0;
467 /* ensure packet is large enough to require tso */
468 if (!info->is_tunnel) {
469 max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
471 if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
472 tso_segsz = info->tso_segsz;
474 max_pkt_len = info->outer_l2_len + info->outer_l3_len +
475 info->l2_len + info->l3_len + info->l4_len +
476 info->tunnel_tso_segsz;
477 if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
478 tso_segsz = info->tunnel_tso_segsz;
481 if (info->ethertype == _htons(RTE_ETHER_TYPE_IPV4)) {
484 ol_flags |= PKT_TX_IPV4;
485 if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
486 ol_flags |= PKT_TX_IP_CKSUM;
488 if (tx_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM) {
489 ol_flags |= PKT_TX_IP_CKSUM;
491 ipv4_hdr->hdr_checksum = 0;
492 ipv4_hdr->hdr_checksum =
493 rte_ipv4_cksum(ipv4_hdr);
496 } else if (info->ethertype == _htons(RTE_ETHER_TYPE_IPV6))
497 ol_flags |= PKT_TX_IPV6;
499 return 0; /* packet type not supported, nothing to do */
501 if (info->l4_proto == IPPROTO_UDP) {
502 udp_hdr = (struct rte_udp_hdr *)((char *)l3_hdr + info->l3_len);
503 /* do not recalculate udp cksum if it was 0 */
504 if (udp_hdr->dgram_cksum != 0) {
505 if (tx_offloads & DEV_TX_OFFLOAD_UDP_CKSUM) {
506 ol_flags |= PKT_TX_UDP_CKSUM;
508 udp_hdr->dgram_cksum = 0;
509 udp_hdr->dgram_cksum =
510 get_udptcp_checksum(l3_hdr, udp_hdr,
514 if (info->gso_enable)
515 ol_flags |= PKT_TX_UDP_SEG;
516 } else if (info->l4_proto == IPPROTO_TCP) {
517 tcp_hdr = (struct rte_tcp_hdr *)((char *)l3_hdr + info->l3_len);
519 ol_flags |= PKT_TX_TCP_SEG;
520 else if (tx_offloads & DEV_TX_OFFLOAD_TCP_CKSUM) {
521 ol_flags |= PKT_TX_TCP_CKSUM;
525 get_udptcp_checksum(l3_hdr, tcp_hdr,
528 if (info->gso_enable)
529 ol_flags |= PKT_TX_TCP_SEG;
530 } else if (info->l4_proto == IPPROTO_SCTP) {
531 sctp_hdr = (struct rte_sctp_hdr *)
532 ((char *)l3_hdr + info->l3_len);
533 /* sctp payload must be a multiple of 4 to be
535 if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) &&
536 ((ipv4_hdr->total_length & 0x3) == 0)) {
537 ol_flags |= PKT_TX_SCTP_CKSUM;
540 /* XXX implement CRC32c, example available in
548 /* Calculate the checksum of outer header */
550 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
551 uint64_t tx_offloads, int tso_enabled)
553 struct rte_ipv4_hdr *ipv4_hdr = outer_l3_hdr;
554 struct rte_ipv6_hdr *ipv6_hdr = outer_l3_hdr;
555 struct rte_udp_hdr *udp_hdr;
556 uint64_t ol_flags = 0;
558 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4)) {
559 ipv4_hdr->hdr_checksum = 0;
560 ol_flags |= PKT_TX_OUTER_IPV4;
562 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)
563 ol_flags |= PKT_TX_OUTER_IP_CKSUM;
565 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
567 ol_flags |= PKT_TX_OUTER_IPV6;
569 if (info->outer_l4_proto != IPPROTO_UDP)
572 udp_hdr = (struct rte_udp_hdr *)
573 ((char *)outer_l3_hdr + info->outer_l3_len);
576 ol_flags |= PKT_TX_TCP_SEG;
578 /* Skip SW outer UDP checksum generation if HW supports it */
579 if (tx_offloads & DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) {
580 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4))
582 = rte_ipv4_phdr_cksum(ipv4_hdr, ol_flags);
585 = rte_ipv6_phdr_cksum(ipv6_hdr, ol_flags);
587 ol_flags |= PKT_TX_OUTER_UDP_CKSUM;
591 /* outer UDP checksum is done in software. In the other side, for
592 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
595 * If a packet will be TSOed into small packets by NIC, we cannot
596 * set/calculate a non-zero checksum, because it will be a wrong
597 * value after the packet be split into several small packets.
600 udp_hdr->dgram_cksum = 0;
602 /* do not recalculate udp cksum if it was 0 */
603 if (udp_hdr->dgram_cksum != 0) {
604 udp_hdr->dgram_cksum = 0;
605 if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4))
606 udp_hdr->dgram_cksum =
607 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
609 udp_hdr->dgram_cksum =
610 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
618 * Performs actual copying.
619 * Returns number of segments in the destination mbuf on success,
620 * or negative error code on failure.
623 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
624 uint16_t seglen[], uint8_t nb_seg)
626 uint32_t dlen, slen, tlen;
628 const struct rte_mbuf *m;
641 while (ms != NULL && i != nb_seg) {
644 slen = rte_pktmbuf_data_len(ms);
645 src = rte_pktmbuf_mtod(ms, const uint8_t *);
649 dlen = RTE_MIN(seglen[i], slen);
650 md[i]->data_len = dlen;
651 md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
652 dst = rte_pktmbuf_mtod(md[i], uint8_t *);
655 len = RTE_MIN(slen, dlen);
656 memcpy(dst, src, len);
671 else if (tlen != m->pkt_len)
674 md[0]->nb_segs = nb_seg;
675 md[0]->pkt_len = tlen;
676 md[0]->vlan_tci = m->vlan_tci;
677 md[0]->vlan_tci_outer = m->vlan_tci_outer;
678 md[0]->ol_flags = m->ol_flags;
679 md[0]->tx_offload = m->tx_offload;
685 * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
686 * Copy packet contents and offload information into the new segmented mbuf.
688 static struct rte_mbuf *
689 pkt_copy_split(const struct rte_mbuf *pkt)
692 uint32_t i, len, nb_seg;
693 struct rte_mempool *mp;
694 uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
695 struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
697 mp = current_fwd_lcore()->mbp;
699 if (tx_pkt_split == TX_PKT_SPLIT_RND)
700 nb_seg = rte_rand() % tx_pkt_nb_segs + 1;
702 nb_seg = tx_pkt_nb_segs;
704 memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
706 /* calculate number of segments to use and their length. */
708 for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
713 n = pkt->pkt_len - len;
715 /* update size of the last segment to fit rest of the packet */
723 p = rte_pktmbuf_alloc(mp);
726 "failed to allocate %u-th of %u mbuf "
727 "from mempool: %s\n",
728 nb_seg - i, nb_seg, mp->name);
733 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
734 TESTPMD_LOG(ERR, "mempool %s, %u-th segment: "
735 "expected seglen: %u, "
736 "actual mbuf tailroom: %u\n",
737 mp->name, i, seglen[i],
738 rte_pktmbuf_tailroom(md[i]));
743 /* all mbufs successfully allocated, do copy */
745 rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
748 "mbuf_copy_split for %p(len=%u, nb_seg=%u) "
749 "into %u segments failed with error code: %d\n",
750 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
752 /* figure out how many mbufs to free. */
756 /* free unused mbufs */
757 for (; i != nb_seg; i++) {
758 rte_pktmbuf_free_seg(md[i]);
766 * Receive a burst of packets, and for each packet:
767 * - parse packet, and try to recognize a supported packet type (1)
768 * - if it's not a supported packet type, don't touch the packet, else:
769 * - reprocess the checksum of all supported layers. This is done in SW
770 * or HW, depending on testpmd command line configuration
771 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
772 * segmentation offload (this implies HW TCP checksum)
773 * Then transmit packets on the output port.
775 * (1) Supported packets are:
776 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
777 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
779 * Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 /
781 * Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 /
783 * Ether / (vlan) / outer IP / outer UDP / GTP / IP|IP6 / UDP|TCP|SCTP
784 * Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
785 * Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
786 * Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
788 * The testpmd command line for this forward engine sets the flags
789 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
790 * wether a checksum must be calculated in software or in hardware. The
791 * IP, UDP, TCP and SCTP flags always concern the inner layer. The
792 * OUTER_IP is only useful for tunnel packets.
795 pkt_burst_checksum_forward(struct fwd_stream *fs)
797 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
798 struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST];
799 struct rte_gso_ctx *gso_ctx;
800 struct rte_mbuf **tx_pkts_burst;
801 struct rte_port *txp;
802 struct rte_mbuf *m, *p;
803 struct rte_ether_hdr *eth_hdr;
804 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
806 uint16_t gro_pkts_num;
812 uint64_t rx_ol_flags, tx_ol_flags;
813 uint64_t tx_offloads;
815 uint32_t rx_bad_ip_csum;
816 uint32_t rx_bad_l4_csum;
817 uint32_t rx_bad_outer_l4_csum;
818 uint32_t rx_bad_outer_ip_csum;
819 struct testpmd_offload_info info;
820 uint16_t nb_segments = 0;
823 uint64_t start_tsc = 0;
825 get_start_cycles(&start_tsc);
827 /* receive a burst of packet */
828 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
830 inc_rx_burst_stats(fs, nb_rx);
831 if (unlikely(nb_rx == 0))
834 fs->rx_packets += nb_rx;
837 rx_bad_outer_l4_csum = 0;
838 rx_bad_outer_ip_csum = 0;
839 gro_enable = gro_ports[fs->rx_port].enable;
841 txp = &ports[fs->tx_port];
842 tx_offloads = txp->dev_conf.txmode.offloads;
843 memset(&info, 0, sizeof(info));
844 info.tso_segsz = txp->tso_segsz;
845 info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
846 if (gso_ports[fs->tx_port].enable)
849 for (i = 0; i < nb_rx; i++) {
850 if (likely(i < nb_rx - 1))
851 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
856 info.pkt_len = rte_pktmbuf_pkt_len(m);
857 tx_ol_flags = m->ol_flags &
858 (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF);
859 rx_ol_flags = m->ol_flags;
861 /* Update the L3/L4 checksum error packet statistics */
862 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
864 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
866 if (rx_ol_flags & PKT_RX_OUTER_L4_CKSUM_BAD)
867 rx_bad_outer_l4_csum += 1;
868 if (rx_ol_flags & PKT_RX_OUTER_IP_CKSUM_BAD)
869 rx_bad_outer_ip_csum += 1;
871 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
872 * and inner headers */
874 eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
875 rte_ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
877 rte_ether_addr_copy(&ports[fs->tx_port].eth_addr,
879 parse_ethernet(eth_hdr, &info);
880 l3_hdr = (char *)eth_hdr + info.l2_len;
882 /* check if it's a supported tunnel */
883 if (txp->parse_tunnel) {
884 if (info.l4_proto == IPPROTO_UDP) {
885 struct rte_udp_hdr *udp_hdr;
887 udp_hdr = (struct rte_udp_hdr *)
888 ((char *)l3_hdr + info.l3_len);
889 parse_gtp(udp_hdr, &info);
890 if (info.is_tunnel) {
891 tx_ol_flags |= PKT_TX_TUNNEL_GTP;
894 parse_vxlan_gpe(udp_hdr, &info);
895 if (info.is_tunnel) {
897 PKT_TX_TUNNEL_VXLAN_GPE;
900 parse_vxlan(udp_hdr, &info,
902 if (info.is_tunnel) {
907 parse_geneve(udp_hdr, &info);
908 if (info.is_tunnel) {
910 PKT_TX_TUNNEL_GENEVE;
913 } else if (info.l4_proto == IPPROTO_GRE) {
914 struct simple_gre_hdr *gre_hdr;
916 gre_hdr = (struct simple_gre_hdr *)
917 ((char *)l3_hdr + info.l3_len);
918 parse_gre(gre_hdr, &info);
920 tx_ol_flags |= PKT_TX_TUNNEL_GRE;
921 } else if (info.l4_proto == IPPROTO_IPIP) {
924 encap_ip_hdr = (char *)l3_hdr + info.l3_len;
925 parse_encap_ip(encap_ip_hdr, &info);
927 tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
932 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
933 if (info.is_tunnel) {
934 outer_l3_hdr = l3_hdr;
935 l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
938 /* step 2: depending on user command line configuration,
939 * recompute checksum either in software or flag the
940 * mbuf to offload the calculation to the NIC. If TSO
941 * is configured, prepare the mbuf for TCP segmentation. */
943 /* process checksums of inner headers first */
944 tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
947 /* Then process outer headers if any. Note that the software
948 * checksum will be wrong if one of the inner checksums is
949 * processed in hardware. */
950 if (info.is_tunnel == 1) {
951 tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
953 !!(tx_ol_flags & PKT_TX_TCP_SEG));
956 /* step 3: fill the mbuf meta data (flags and header lengths) */
959 if (info.is_tunnel == 1) {
960 if (info.tunnel_tso_segsz ||
962 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
964 DEV_TX_OFFLOAD_OUTER_UDP_CKSUM)) {
965 m->outer_l2_len = info.outer_l2_len;
966 m->outer_l3_len = info.outer_l3_len;
967 m->l2_len = info.l2_len;
968 m->l3_len = info.l3_len;
969 m->l4_len = info.l4_len;
970 m->tso_segsz = info.tunnel_tso_segsz;
973 /* if there is a outer UDP cksum
974 processed in sw and the inner in hw,
975 the outer checksum will be wrong as
976 the payload will be modified by the
978 m->l2_len = info.outer_l2_len +
979 info.outer_l3_len + info.l2_len;
980 m->l3_len = info.l3_len;
981 m->l4_len = info.l4_len;
984 /* this is only useful if an offload flag is
985 * set, but it does not hurt to fill it in any
987 m->l2_len = info.l2_len;
988 m->l3_len = info.l3_len;
989 m->l4_len = info.l4_len;
990 m->tso_segsz = info.tso_segsz;
992 m->ol_flags = tx_ol_flags;
994 /* Do split & copy for the packet. */
995 if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
996 p = pkt_copy_split(m);
1004 /* if verbose mode is enabled, dump debug info */
1005 if (verbose_level > 0) {
1008 printf("-----------------\n");
1009 printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
1010 fs->rx_port, m, m->pkt_len, m->nb_segs);
1011 /* dump rx parsed packet info */
1012 rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
1013 printf("rx: l2_len=%d ethertype=%x l3_len=%d "
1014 "l4_proto=%d l4_len=%d flags=%s\n",
1015 info.l2_len, rte_be_to_cpu_16(info.ethertype),
1016 info.l3_len, info.l4_proto, info.l4_len, buf);
1017 if (rx_ol_flags & PKT_RX_LRO)
1018 printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
1019 if (info.is_tunnel == 1)
1020 printf("rx: outer_l2_len=%d outer_ethertype=%x "
1021 "outer_l3_len=%d\n", info.outer_l2_len,
1022 rte_be_to_cpu_16(info.outer_ethertype),
1024 /* dump tx packet info */
1025 if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
1026 DEV_TX_OFFLOAD_UDP_CKSUM |
1027 DEV_TX_OFFLOAD_TCP_CKSUM |
1028 DEV_TX_OFFLOAD_SCTP_CKSUM)) ||
1029 info.tso_segsz != 0)
1030 printf("tx: m->l2_len=%d m->l3_len=%d "
1032 m->l2_len, m->l3_len, m->l4_len);
1033 if (info.is_tunnel == 1) {
1035 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
1037 DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
1038 (tx_ol_flags & PKT_TX_OUTER_IPV6))
1039 printf("tx: m->outer_l2_len=%d "
1040 "m->outer_l3_len=%d\n",
1043 if (info.tunnel_tso_segsz != 0 &&
1044 (m->ol_flags & PKT_TX_TCP_SEG))
1045 printf("tx: m->tso_segsz=%d\n",
1047 } else if (info.tso_segsz != 0 &&
1048 (m->ol_flags & PKT_TX_TCP_SEG))
1049 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
1050 rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
1051 printf("tx: flags=%s", buf);
1056 if (unlikely(gro_enable)) {
1057 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
1058 nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx,
1059 &(gro_ports[fs->rx_port].param));
1061 gro_ctx = current_fwd_lcore()->gro_ctx;
1062 nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx);
1064 if (++fs->gro_times >= gro_flush_cycles) {
1065 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx);
1066 if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
1067 gro_pkts_num = MAX_PKT_BURST - nb_rx;
1069 nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
1078 if (gso_ports[fs->tx_port].enable == 0)
1079 tx_pkts_burst = pkts_burst;
1081 gso_ctx = &(current_fwd_lcore()->gso_ctx);
1082 gso_ctx->gso_size = gso_max_segment_size;
1083 for (i = 0; i < nb_rx; i++) {
1084 ret = rte_gso_segment(pkts_burst[i], gso_ctx,
1085 &gso_segments[nb_segments],
1086 GSO_MAX_PKT_BURST - nb_segments);
1088 /* pkts_burst[i] can be freed safely here. */
1089 rte_pktmbuf_free(pkts_burst[i]);
1091 } else if (ret == 0) {
1092 /* 0 means it can be transmitted directly
1095 gso_segments[nb_segments] = pkts_burst[i];
1098 TESTPMD_LOG(DEBUG, "Unable to segment packet");
1099 rte_pktmbuf_free(pkts_burst[i]);
1103 tx_pkts_burst = gso_segments;
1104 nb_rx = nb_segments;
1107 nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
1108 tx_pkts_burst, nb_rx);
1109 if (nb_prep != nb_rx)
1111 "Preparing packet burst to transmit failed: %s\n",
1112 rte_strerror(rte_errno));
1114 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
1118 * Retry if necessary
1120 if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
1122 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
1123 rte_delay_us(burst_tx_delay_time);
1124 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
1125 &tx_pkts_burst[nb_tx], nb_rx - nb_tx);
1128 fs->tx_packets += nb_tx;
1129 fs->rx_bad_ip_csum += rx_bad_ip_csum;
1130 fs->rx_bad_l4_csum += rx_bad_l4_csum;
1131 fs->rx_bad_outer_l4_csum += rx_bad_outer_l4_csum;
1132 fs->rx_bad_outer_ip_csum += rx_bad_outer_ip_csum;
1134 inc_tx_burst_stats(fs, nb_tx);
1135 if (unlikely(nb_tx < nb_rx)) {
1136 fs->fwd_dropped += (nb_rx - nb_tx);
1138 rte_pktmbuf_free(tx_pkts_burst[nb_tx]);
1139 } while (++nb_tx < nb_rx);
1142 get_end_cycles(fs, start_tsc);
1145 struct fwd_engine csum_fwd_engine = {
1146 .fwd_mode_name = "csum",
1147 .port_fwd_begin = NULL,
1148 .port_fwd_end = NULL,
1149 .packet_fwd = pkt_burst_checksum_forward,