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 {
97 uint16_t outer_ethertype;
98 uint16_t outer_l2_len;
99 uint16_t outer_l3_len;
104 get_psd_sum(void *l3_hdr, uint16_t ethertype, uint64_t ol_flags)
106 if (ethertype == _htons(ETHER_TYPE_IPv4))
107 return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
108 else /* assume ethertype == ETHER_TYPE_IPv6 */
109 return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
113 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
115 if (ethertype == _htons(ETHER_TYPE_IPv4))
116 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
117 else /* assume ethertype == ETHER_TYPE_IPv6 */
118 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
121 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
123 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
125 struct tcp_hdr *tcp_hdr;
127 info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
128 info->l4_proto = ipv4_hdr->next_proto_id;
130 /* only fill l4_len for TCP, it's useful for TSO */
131 if (info->l4_proto == IPPROTO_TCP) {
132 tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
133 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
138 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
140 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
142 struct tcp_hdr *tcp_hdr;
144 info->l3_len = sizeof(struct ipv6_hdr);
145 info->l4_proto = ipv6_hdr->proto;
147 /* only fill l4_len for TCP, it's useful for TSO */
148 if (info->l4_proto == IPPROTO_TCP) {
149 tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
150 info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
156 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
157 * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
158 * header. The l4_len argument is only set in case of TCP (useful for TSO).
161 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info)
163 struct ipv4_hdr *ipv4_hdr;
164 struct ipv6_hdr *ipv6_hdr;
166 info->l2_len = sizeof(struct ether_hdr);
167 info->ethertype = eth_hdr->ether_type;
169 if (info->ethertype == _htons(ETHER_TYPE_VLAN)) {
170 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
172 info->l2_len += sizeof(struct vlan_hdr);
173 info->ethertype = vlan_hdr->eth_proto;
176 switch (info->ethertype) {
177 case _htons(ETHER_TYPE_IPv4):
178 ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
179 parse_ipv4(ipv4_hdr, info);
181 case _htons(ETHER_TYPE_IPv6):
182 ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
183 parse_ipv6(ipv6_hdr, info);
193 /* Parse a vxlan header */
195 parse_vxlan(struct udp_hdr *udp_hdr, struct testpmd_offload_info *info,
196 uint64_t mbuf_olflags)
198 struct ether_hdr *eth_hdr;
200 /* check udp destination port, 4789 is the default vxlan port
201 * (rfc7348) or that the rx offload flag is set (i40e only
203 if (udp_hdr->dst_port != _htons(4789) &&
204 (mbuf_olflags & (PKT_RX_TUNNEL_IPV4_HDR |
205 PKT_RX_TUNNEL_IPV6_HDR)) == 0)
209 info->outer_ethertype = info->ethertype;
210 info->outer_l2_len = info->l2_len;
211 info->outer_l3_len = info->l3_len;
213 eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
214 sizeof(struct udp_hdr) +
215 sizeof(struct vxlan_hdr));
217 parse_ethernet(eth_hdr, info);
218 info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */
221 /* modify the IPv4 or IPv4 source address of a packet */
223 change_ip_addresses(void *l3_hdr, uint16_t ethertype)
225 struct ipv4_hdr *ipv4_hdr = l3_hdr;
226 struct ipv6_hdr *ipv6_hdr = l3_hdr;
228 if (ethertype == _htons(ETHER_TYPE_IPv4)) {
230 rte_cpu_to_be_32(rte_be_to_cpu_32(ipv4_hdr->src_addr) + 1);
231 } else if (ethertype == _htons(ETHER_TYPE_IPv6)) {
232 ipv6_hdr->src_addr[15] = ipv6_hdr->src_addr[15] + 1;
236 /* if possible, calculate the checksum of a packet in hw or sw,
237 * depending on the testpmd command line configuration */
239 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
240 uint16_t testpmd_ol_flags)
242 struct ipv4_hdr *ipv4_hdr = l3_hdr;
243 struct udp_hdr *udp_hdr;
244 struct tcp_hdr *tcp_hdr;
245 struct sctp_hdr *sctp_hdr;
246 uint64_t ol_flags = 0;
248 if (info->ethertype == _htons(ETHER_TYPE_IPv4)) {
250 ipv4_hdr->hdr_checksum = 0;
252 ol_flags |= PKT_TX_IPV4;
253 if (info->tso_segsz != 0 && info->l4_proto == IPPROTO_TCP) {
254 ol_flags |= PKT_TX_IP_CKSUM;
256 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
257 ol_flags |= PKT_TX_IP_CKSUM;
259 ipv4_hdr->hdr_checksum =
260 rte_ipv4_cksum(ipv4_hdr);
262 } else if (info->ethertype == _htons(ETHER_TYPE_IPv6))
263 ol_flags |= PKT_TX_IPV6;
265 return 0; /* packet type not supported, nothing to do */
267 if (info->l4_proto == IPPROTO_UDP) {
268 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
269 /* do not recalculate udp cksum if it was 0 */
270 if (udp_hdr->dgram_cksum != 0) {
271 udp_hdr->dgram_cksum = 0;
272 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM) {
273 ol_flags |= PKT_TX_UDP_CKSUM;
274 udp_hdr->dgram_cksum = get_psd_sum(l3_hdr,
275 info->ethertype, ol_flags);
277 udp_hdr->dgram_cksum =
278 get_udptcp_checksum(l3_hdr, udp_hdr,
282 } else if (info->l4_proto == IPPROTO_TCP) {
283 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
285 if (info->tso_segsz != 0) {
286 ol_flags |= PKT_TX_TCP_SEG;
287 tcp_hdr->cksum = get_psd_sum(l3_hdr, info->ethertype,
289 } else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
290 ol_flags |= PKT_TX_TCP_CKSUM;
291 tcp_hdr->cksum = get_psd_sum(l3_hdr, info->ethertype,
295 get_udptcp_checksum(l3_hdr, tcp_hdr,
298 } else if (info->l4_proto == IPPROTO_SCTP) {
299 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
301 /* sctp payload must be a multiple of 4 to be
303 if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_SCTP_CKSUM) &&
304 ((ipv4_hdr->total_length & 0x3) == 0)) {
305 ol_flags |= PKT_TX_SCTP_CKSUM;
307 /* XXX implement CRC32c, example available in
315 /* Calculate the checksum of outer header (only vxlan is supported,
316 * meaning IP + UDP). The caller already checked that it's a vxlan
319 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
320 uint16_t testpmd_ol_flags)
322 struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
323 struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
324 struct udp_hdr *udp_hdr;
325 uint64_t ol_flags = 0;
327 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
328 ipv4_hdr->hdr_checksum = 0;
329 ol_flags |= PKT_TX_OUTER_IPV4;
331 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
332 ol_flags |= PKT_TX_OUTER_IP_CKSUM;
334 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
335 } else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
336 ol_flags |= PKT_TX_OUTER_IPV6;
338 /* outer UDP checksum is always done in software as we have no
339 * hardware supporting it today, and no API for it. */
341 udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
342 /* do not recalculate udp cksum if it was 0 */
343 if (udp_hdr->dgram_cksum != 0) {
344 udp_hdr->dgram_cksum = 0;
345 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
346 udp_hdr->dgram_cksum =
347 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
349 udp_hdr->dgram_cksum =
350 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
357 * Receive a burst of packets, and for each packet:
358 * - parse packet, and try to recognize a supported packet type (1)
359 * - if it's not a supported packet type, don't touch the packet, else:
360 * - modify the IPs in inner headers and in outer headers if any
361 * - reprocess the checksum of all supported layers. This is done in SW
362 * or HW, depending on testpmd command line configuration
363 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
364 * segmentation offload (this implies HW TCP checksum)
365 * Then transmit packets on the output port.
367 * (1) Supported packets are:
368 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
369 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
372 * The testpmd command line for this forward engine sets the flags
373 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
374 * wether a checksum must be calculated in software or in hardware. The
375 * IP, UDP, TCP and SCTP flags always concern the inner layer. The
376 * OUTER_IP is only useful for tunnel packets.
379 pkt_burst_checksum_forward(struct fwd_stream *fs)
381 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
382 struct rte_port *txp;
384 struct ether_hdr *eth_hdr;
385 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
390 uint16_t testpmd_ol_flags;
391 uint32_t rx_bad_ip_csum;
392 uint32_t rx_bad_l4_csum;
393 struct testpmd_offload_info info;
395 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
398 uint64_t core_cycles;
401 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
402 start_tsc = rte_rdtsc();
405 /* receive a burst of packet */
406 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
408 if (unlikely(nb_rx == 0))
411 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
412 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
414 fs->rx_packets += nb_rx;
418 txp = &ports[fs->tx_port];
419 testpmd_ol_flags = txp->tx_ol_flags;
420 memset(&info, 0, sizeof(info));
421 info.tso_segsz = txp->tso_segsz;
423 for (i = 0; i < nb_rx; i++) {
429 /* Update the L3/L4 checksum error packet statistics */
430 rx_bad_ip_csum += ((m->ol_flags & PKT_RX_IP_CKSUM_BAD) != 0);
431 rx_bad_l4_csum += ((m->ol_flags & PKT_RX_L4_CKSUM_BAD) != 0);
433 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
434 * and inner headers */
436 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
437 parse_ethernet(eth_hdr, &info);
438 l3_hdr = (char *)eth_hdr + info.l2_len;
440 /* check if it's a supported tunnel (only vxlan for now) */
441 if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_PARSE_TUNNEL) &&
442 info.l4_proto == IPPROTO_UDP) {
443 struct udp_hdr *udp_hdr;
444 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info.l3_len);
445 parse_vxlan(udp_hdr, &info, m->ol_flags);
448 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
449 if (info.is_tunnel) {
450 outer_l3_hdr = l3_hdr;
451 l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
454 /* step 2: change all source IPs (v4 or v6) so we need
455 * to recompute the chksums even if they were correct */
457 change_ip_addresses(l3_hdr, info.ethertype);
458 if (info.is_tunnel == 1)
459 change_ip_addresses(outer_l3_hdr, info.outer_ethertype);
461 /* step 3: depending on user command line configuration,
462 * recompute checksum either in software or flag the
463 * mbuf to offload the calculation to the NIC. If TSO
464 * is configured, prepare the mbuf for TCP segmentation. */
466 /* process checksums of inner headers first */
467 ol_flags |= process_inner_cksums(l3_hdr, &info, testpmd_ol_flags);
469 /* Then process outer headers if any. Note that the software
470 * checksum will be wrong if one of the inner checksums is
471 * processed in hardware. */
472 if (info.is_tunnel == 1) {
473 ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
477 /* step 4: fill the mbuf meta data (flags and header lengths) */
479 if (info.is_tunnel == 1) {
480 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) {
481 m->outer_l2_len = info.outer_l2_len;
482 m->outer_l3_len = info.outer_l3_len;
483 m->l2_len = info.l2_len;
484 m->l3_len = info.l3_len;
487 /* if there is a outer UDP cksum
488 processed in sw and the inner in hw,
489 the outer checksum will be wrong as
490 the payload will be modified by the
492 m->l2_len = info.outer_l2_len +
493 info.outer_l3_len + info.l2_len;
494 m->l3_len = info.l3_len;
495 m->l4_len = info.l4_len;
498 /* this is only useful if an offload flag is
499 * set, but it does not hurt to fill it in any
501 m->l2_len = info.l2_len;
502 m->l3_len = info.l3_len;
503 m->l4_len = info.l4_len;
505 m->tso_segsz = info.tso_segsz;
506 m->ol_flags = ol_flags;
508 /* if verbose mode is enabled, dump debug info */
509 if (verbose_level > 0) {
514 { PKT_TX_IP_CKSUM, PKT_TX_IP_CKSUM },
515 { PKT_TX_UDP_CKSUM, PKT_TX_L4_MASK },
516 { PKT_TX_TCP_CKSUM, PKT_TX_L4_MASK },
517 { PKT_TX_SCTP_CKSUM, PKT_TX_L4_MASK },
518 { PKT_TX_IPV4, PKT_TX_IPV4 },
519 { PKT_TX_IPV6, PKT_TX_IPV6 },
520 { PKT_TX_OUTER_IP_CKSUM, PKT_TX_OUTER_IP_CKSUM },
521 { PKT_TX_OUTER_IPV4, PKT_TX_OUTER_IPV4 },
522 { PKT_TX_OUTER_IPV6, PKT_TX_OUTER_IPV6 },
523 { PKT_TX_TCP_SEG, PKT_TX_TCP_SEG },
528 printf("-----------------\n");
529 /* dump rx parsed packet info */
530 printf("rx: l2_len=%d ethertype=%x l3_len=%d "
531 "l4_proto=%d l4_len=%d\n",
532 info.l2_len, rte_be_to_cpu_16(info.ethertype),
533 info.l3_len, info.l4_proto, info.l4_len);
534 if (info.is_tunnel == 1)
535 printf("rx: outer_l2_len=%d outer_ethertype=%x "
536 "outer_l3_len=%d\n", info.outer_l2_len,
537 rte_be_to_cpu_16(info.outer_ethertype),
539 /* dump tx packet info */
540 if ((testpmd_ol_flags & (TESTPMD_TX_OFFLOAD_IP_CKSUM |
541 TESTPMD_TX_OFFLOAD_UDP_CKSUM |
542 TESTPMD_TX_OFFLOAD_TCP_CKSUM |
543 TESTPMD_TX_OFFLOAD_SCTP_CKSUM)) ||
545 printf("tx: m->l2_len=%d m->l3_len=%d "
547 m->l2_len, m->l3_len, m->l4_len);
548 if ((info.is_tunnel == 1) &&
549 (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM))
550 printf("tx: m->outer_l2_len=%d m->outer_l3_len=%d\n",
551 m->outer_l2_len, m->outer_l3_len);
552 if (info.tso_segsz != 0)
553 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
554 printf("tx: flags=");
555 for (j = 0; j < sizeof(tx_flags)/sizeof(*tx_flags); j++) {
556 name = rte_get_tx_ol_flag_name(tx_flags[j].flag);
557 if ((m->ol_flags & tx_flags[j].mask) ==
564 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
565 fs->tx_packets += nb_tx;
566 fs->rx_bad_ip_csum += rx_bad_ip_csum;
567 fs->rx_bad_l4_csum += rx_bad_l4_csum;
569 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
570 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
572 if (unlikely(nb_tx < nb_rx)) {
573 fs->fwd_dropped += (nb_rx - nb_tx);
575 rte_pktmbuf_free(pkts_burst[nb_tx]);
576 } while (++nb_tx < nb_rx);
578 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
579 end_tsc = rte_rdtsc();
580 core_cycles = (end_tsc - start_tsc);
581 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
585 struct fwd_engine csum_fwd_engine = {
586 .fwd_mode_name = "csum",
587 .port_fwd_begin = NULL,
588 .port_fwd_end = NULL,
589 .packet_fwd = pkt_burst_checksum_forward,