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>
54 #include <rte_tailq.h>
56 #include <rte_per_lcore.h>
57 #include <rte_lcore.h>
58 #include <rte_atomic.h>
59 #include <rte_branch_prediction.h>
61 #include <rte_memory.h>
62 #include <rte_mempool.h>
64 #include <rte_memcpy.h>
65 #include <rte_interrupts.h>
67 #include <rte_ether.h>
68 #include <rte_ethdev.h>
73 #include <rte_prefetch.h>
74 #include <rte_string_fns.h>
77 #define IP_DEFTTL 64 /* from RFC 1340. */
78 #define IP_VERSION 0x40
79 #define IP_HDRLEN 0x05 /* default IP header length == five 32-bits words. */
80 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
82 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */
83 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
84 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
89 /* structure that caches offload info for the current packet */
90 struct testpmd_offload_info {
98 uint16_t outer_ethertype;
99 uint16_t outer_l2_len;
100 uint16_t outer_l3_len;
105 get_psd_sum(void *l3_hdr, uint16_t ethertype, uint64_t ol_flags)
107 if (ethertype == _htons(ETHER_TYPE_IPv4))
108 return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
109 else /* assume ethertype == ETHER_TYPE_IPv6 */
110 return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
114 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
116 if (ethertype == _htons(ETHER_TYPE_IPv4))
117 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
118 else /* assume ethertype == ETHER_TYPE_IPv6 */
119 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
122 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
124 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
126 struct tcp_hdr *tcp_hdr;
128 info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
129 info->l4_proto = ipv4_hdr->next_proto_id;
131 /* only fill l4_len for TCP, it's useful for TSO */
132 if (info->l4_proto == IPPROTO_TCP) {
133 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
134 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
139 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
141 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
143 struct tcp_hdr *tcp_hdr;
145 info->l3_len = sizeof(struct ipv6_hdr);
146 info->l4_proto = ipv6_hdr->proto;
148 /* only fill l4_len for TCP, it's useful for TSO */
149 if (info->l4_proto == IPPROTO_TCP) {
150 tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
151 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
157 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
158 * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
159 * header. The l4_len argument is only set in case of TCP (useful for TSO).
162 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info)
164 struct ipv4_hdr *ipv4_hdr;
165 struct ipv6_hdr *ipv6_hdr;
167 info->l2_len = sizeof(struct ether_hdr);
168 info->ethertype = eth_hdr->ether_type;
170 if (info->ethertype == _htons(ETHER_TYPE_VLAN)) {
171 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
173 info->l2_len += sizeof(struct vlan_hdr);
174 info->ethertype = vlan_hdr->eth_proto;
177 switch (info->ethertype) {
178 case _htons(ETHER_TYPE_IPv4):
179 ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
180 parse_ipv4(ipv4_hdr, info);
182 case _htons(ETHER_TYPE_IPv6):
183 ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
184 parse_ipv6(ipv6_hdr, info);
194 /* modify the IPv4 or IPv4 source address of a packet */
196 change_ip_addresses(void *l3_hdr, uint16_t ethertype)
198 struct ipv4_hdr *ipv4_hdr = l3_hdr;
199 struct ipv6_hdr *ipv6_hdr = l3_hdr;
201 if (ethertype == _htons(ETHER_TYPE_IPv4)) {
203 rte_cpu_to_be_32(rte_be_to_cpu_32(ipv4_hdr->src_addr) + 1);
204 } else if (ethertype == _htons(ETHER_TYPE_IPv6)) {
205 ipv6_hdr->src_addr[15] = ipv6_hdr->src_addr[15] + 1;
209 /* if possible, calculate the checksum of a packet in hw or sw,
210 * depending on the testpmd command line configuration */
212 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
213 uint16_t testpmd_ol_flags)
215 struct ipv4_hdr *ipv4_hdr = l3_hdr;
216 struct udp_hdr *udp_hdr;
217 struct tcp_hdr *tcp_hdr;
218 struct sctp_hdr *sctp_hdr;
219 uint64_t ol_flags = 0;
221 if (info->ethertype == _htons(ETHER_TYPE_IPv4)) {
223 ipv4_hdr->hdr_checksum = 0;
225 ol_flags |= PKT_TX_IPV4;
226 if (info->tso_segsz != 0 && info->l4_proto == IPPROTO_TCP) {
227 ol_flags |= PKT_TX_IP_CKSUM;
229 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
230 ol_flags |= PKT_TX_IP_CKSUM;
232 ipv4_hdr->hdr_checksum =
233 rte_ipv4_cksum(ipv4_hdr);
235 } else if (info->ethertype == _htons(ETHER_TYPE_IPv6))
236 ol_flags |= PKT_TX_IPV6;
238 return 0; /* packet type not supported, nothing to do */
240 if (info->l4_proto == IPPROTO_UDP) {
241 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
242 /* do not recalculate udp cksum if it was 0 */
243 if (udp_hdr->dgram_cksum != 0) {
244 udp_hdr->dgram_cksum = 0;
245 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM) {
246 ol_flags |= PKT_TX_UDP_CKSUM;
247 udp_hdr->dgram_cksum = get_psd_sum(l3_hdr,
248 info->ethertype, ol_flags);
250 udp_hdr->dgram_cksum =
251 get_udptcp_checksum(l3_hdr, udp_hdr,
255 } else if (info->l4_proto == IPPROTO_TCP) {
256 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
258 if (info->tso_segsz != 0) {
259 ol_flags |= PKT_TX_TCP_SEG;
260 tcp_hdr->cksum = get_psd_sum(l3_hdr, info->ethertype,
262 } else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
263 ol_flags |= PKT_TX_TCP_CKSUM;
264 tcp_hdr->cksum = get_psd_sum(l3_hdr, info->ethertype,
268 get_udptcp_checksum(l3_hdr, tcp_hdr,
271 } else if (info->l4_proto == IPPROTO_SCTP) {
272 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
274 /* sctp payload must be a multiple of 4 to be
276 if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_SCTP_CKSUM) &&
277 ((ipv4_hdr->total_length & 0x3) == 0)) {
278 ol_flags |= PKT_TX_SCTP_CKSUM;
280 /* XXX implement CRC32c, example available in
288 /* Calculate the checksum of outer header (only vxlan is supported,
289 * meaning IP + UDP). The caller already checked that it's a vxlan
292 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
293 uint16_t testpmd_ol_flags)
295 struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
296 struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
297 struct udp_hdr *udp_hdr;
298 uint64_t ol_flags = 0;
300 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
301 ipv4_hdr->hdr_checksum = 0;
302 ol_flags |= PKT_TX_OUTER_IPV4;
304 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
305 ol_flags |= PKT_TX_OUTER_IP_CKSUM;
307 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
308 } else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
309 ol_flags |= PKT_TX_OUTER_IPV6;
311 /* outer UDP checksum is always done in software as we have no
312 * hardware supporting it today, and no API for it. */
314 udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
315 /* do not recalculate udp cksum if it was 0 */
316 if (udp_hdr->dgram_cksum != 0) {
317 udp_hdr->dgram_cksum = 0;
318 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
319 udp_hdr->dgram_cksum =
320 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
322 udp_hdr->dgram_cksum =
323 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
330 * Receive a burst of packets, and for each packet:
331 * - parse packet, and try to recognize a supported packet type (1)
332 * - if it's not a supported packet type, don't touch the packet, else:
333 * - modify the IPs in inner headers and in outer headers if any
334 * - reprocess the checksum of all supported layers. This is done in SW
335 * or HW, depending on testpmd command line configuration
336 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
337 * segmentation offload (this implies HW TCP checksum)
338 * Then transmit packets on the output port.
340 * (1) Supported packets are:
341 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
342 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
345 * The testpmd command line for this forward engine sets the flags
346 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
347 * wether a checksum must be calculated in software or in hardware. The
348 * IP, UDP, TCP and SCTP flags always concern the inner layer. The
349 * OUTER_IP is only useful for tunnel packets.
352 pkt_burst_checksum_forward(struct fwd_stream *fs)
354 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
355 struct rte_port *txp;
357 struct ether_hdr *eth_hdr;
358 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
359 struct udp_hdr *udp_hdr;
364 uint16_t testpmd_ol_flags;
365 uint32_t rx_bad_ip_csum;
366 uint32_t rx_bad_l4_csum;
367 struct testpmd_offload_info info;
369 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
372 uint64_t core_cycles;
375 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
376 start_tsc = rte_rdtsc();
379 /* receive a burst of packet */
380 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
382 if (unlikely(nb_rx == 0))
385 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
386 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
388 fs->rx_packets += nb_rx;
392 txp = &ports[fs->tx_port];
393 testpmd_ol_flags = txp->tx_ol_flags;
394 memset(&info, 0, sizeof(info));
395 info.tso_segsz = txp->tso_segsz;
397 for (i = 0; i < nb_rx; i++) {
403 /* Update the L3/L4 checksum error packet statistics */
404 rx_bad_ip_csum += ((m->ol_flags & PKT_RX_IP_CKSUM_BAD) != 0);
405 rx_bad_l4_csum += ((m->ol_flags & PKT_RX_L4_CKSUM_BAD) != 0);
407 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
408 * and inner headers */
410 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
411 parse_ethernet(eth_hdr, &info);
412 l3_hdr = (char *)eth_hdr + info.l2_len;
414 /* check if it's a supported tunnel (only vxlan for now) */
415 if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_PARSE_TUNNEL) &&
416 info.l4_proto == IPPROTO_UDP) {
417 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info.l3_len);
419 /* check udp destination port, 4789 is the default
420 * vxlan port (rfc7348) */
421 if (udp_hdr->dst_port == _htons(4789)) {
422 info.l4_tun_len = ETHER_VXLAN_HLEN;
425 /* currently, this flag is set by i40e only if the
427 } else if (m->ol_flags & (PKT_RX_TUNNEL_IPV4_HDR |
428 PKT_RX_TUNNEL_IPV6_HDR))
431 if (info.is_tunnel == 1) {
432 info.outer_ethertype = info.ethertype;
433 info.outer_l2_len = info.l2_len;
434 info.outer_l3_len = info.l3_len;
435 outer_l3_hdr = l3_hdr;
437 eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
438 sizeof(struct udp_hdr) +
439 sizeof(struct vxlan_hdr));
441 parse_ethernet(eth_hdr, &info);
442 l3_hdr = (char *)eth_hdr + info.l2_len;
446 /* step 2: change all source IPs (v4 or v6) so we need
447 * to recompute the chksums even if they were correct */
449 change_ip_addresses(l3_hdr, info.ethertype);
450 if (info.is_tunnel == 1)
451 change_ip_addresses(outer_l3_hdr, info.outer_ethertype);
453 /* step 3: depending on user command line configuration,
454 * recompute checksum either in software or flag the
455 * mbuf to offload the calculation to the NIC. If TSO
456 * is configured, prepare the mbuf for TCP segmentation. */
458 /* process checksums of inner headers first */
459 ol_flags |= process_inner_cksums(l3_hdr, &info, testpmd_ol_flags);
461 /* Then process outer headers if any. Note that the software
462 * checksum will be wrong if one of the inner checksums is
463 * processed in hardware. */
464 if (info.is_tunnel == 1) {
465 ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
469 /* step 4: fill the mbuf meta data (flags and header lengths) */
471 if (info.is_tunnel == 1) {
472 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) {
473 m->outer_l2_len = info.outer_l2_len;
474 m->outer_l3_len = info.outer_l3_len;
475 m->l2_len = info.l4_tun_len + info.l2_len;
476 m->l3_len = info.l3_len;
479 /* if there is a outer UDP cksum
480 processed in sw and the inner in hw,
481 the outer checksum will be wrong as
482 the payload will be modified by the
484 m->l2_len = info.outer_l2_len +
486 sizeof(struct udp_hdr) +
487 sizeof(struct vxlan_hdr) + info.l2_len;
488 m->l3_len = info.l3_len;
489 m->l4_len = info.l4_len;
492 /* this is only useful if an offload flag is
493 * set, but it does not hurt to fill it in any
495 m->l2_len = info.l2_len;
496 m->l3_len = info.l3_len;
497 m->l4_len = info.l4_len;
499 m->tso_segsz = info.tso_segsz;
500 m->ol_flags = ol_flags;
502 /* if verbose mode is enabled, dump debug info */
503 if (verbose_level > 0) {
508 { PKT_TX_IP_CKSUM, PKT_TX_IP_CKSUM },
509 { PKT_TX_UDP_CKSUM, PKT_TX_L4_MASK },
510 { PKT_TX_TCP_CKSUM, PKT_TX_L4_MASK },
511 { PKT_TX_SCTP_CKSUM, PKT_TX_L4_MASK },
512 { PKT_TX_IPV4, PKT_TX_IPV4 },
513 { PKT_TX_IPV6, PKT_TX_IPV6 },
514 { PKT_TX_OUTER_IP_CKSUM, PKT_TX_OUTER_IP_CKSUM },
515 { PKT_TX_OUTER_IPV4, PKT_TX_OUTER_IPV4 },
516 { PKT_TX_OUTER_IPV6, PKT_TX_OUTER_IPV6 },
517 { PKT_TX_TCP_SEG, PKT_TX_TCP_SEG },
522 printf("-----------------\n");
523 /* dump rx parsed packet info */
524 printf("rx: l2_len=%d ethertype=%x l3_len=%d "
525 "l4_proto=%d l4_len=%d\n",
526 info.l2_len, rte_be_to_cpu_16(info.ethertype),
527 info.l3_len, info.l4_proto, info.l4_len);
528 if (info.is_tunnel == 1)
529 printf("rx: outer_l2_len=%d outer_ethertype=%x "
530 "outer_l3_len=%d\n", info.outer_l2_len,
531 rte_be_to_cpu_16(info.outer_ethertype),
533 /* dump tx packet info */
534 if ((testpmd_ol_flags & (TESTPMD_TX_OFFLOAD_IP_CKSUM |
535 TESTPMD_TX_OFFLOAD_UDP_CKSUM |
536 TESTPMD_TX_OFFLOAD_TCP_CKSUM |
537 TESTPMD_TX_OFFLOAD_SCTP_CKSUM)) ||
539 printf("tx: m->l2_len=%d m->l3_len=%d "
541 m->l2_len, m->l3_len, m->l4_len);
542 if ((info.is_tunnel == 1) &&
543 (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM))
544 printf("tx: m->outer_l2_len=%d m->outer_l3_len=%d\n",
545 m->outer_l2_len, m->outer_l3_len);
546 if (info.tso_segsz != 0)
547 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
548 printf("tx: flags=");
549 for (j = 0; j < sizeof(tx_flags)/sizeof(*tx_flags); j++) {
550 name = rte_get_tx_ol_flag_name(tx_flags[j].flag);
551 if ((m->ol_flags & tx_flags[j].mask) ==
558 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
559 fs->tx_packets += nb_tx;
560 fs->rx_bad_ip_csum += rx_bad_ip_csum;
561 fs->rx_bad_l4_csum += rx_bad_l4_csum;
563 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
564 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
566 if (unlikely(nb_tx < nb_rx)) {
567 fs->fwd_dropped += (nb_rx - nb_tx);
569 rte_pktmbuf_free(pkts_burst[nb_tx]);
570 } while (++nb_tx < nb_rx);
572 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
573 end_tsc = rte_rdtsc();
574 core_cycles = (end_tsc - start_tsc);
575 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
579 struct fwd_engine csum_fwd_engine = {
580 .fwd_mode_name = "csum",
581 .port_fwd_begin = NULL,
582 .port_fwd_end = NULL,
583 .packet_fwd = pkt_burst_checksum_forward,