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)))
90 get_psd_sum(void *l3_hdr, uint16_t ethertype, uint64_t ol_flags)
92 if (ethertype == _htons(ETHER_TYPE_IPv4))
93 return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
94 else /* assume ethertype == ETHER_TYPE_IPv6 */
95 return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
99 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
101 if (ethertype == _htons(ETHER_TYPE_IPv4))
102 return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
103 else /* assume ethertype == ETHER_TYPE_IPv6 */
104 return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
108 * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
109 * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
110 * header. The l4_len argument is only set in case of TCP (useful for TSO).
113 parse_ethernet(struct ether_hdr *eth_hdr, uint16_t *ethertype, uint16_t *l2_len,
114 uint16_t *l3_len, uint8_t *l4_proto, uint16_t *l4_len)
116 struct ipv4_hdr *ipv4_hdr;
117 struct ipv6_hdr *ipv6_hdr;
118 struct tcp_hdr *tcp_hdr;
120 *l2_len = sizeof(struct ether_hdr);
121 *ethertype = eth_hdr->ether_type;
123 if (*ethertype == _htons(ETHER_TYPE_VLAN)) {
124 struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
126 *l2_len += sizeof(struct vlan_hdr);
127 *ethertype = vlan_hdr->eth_proto;
130 switch (*ethertype) {
131 case _htons(ETHER_TYPE_IPv4):
132 ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + *l2_len);
133 *l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
134 *l4_proto = ipv4_hdr->next_proto_id;
136 case _htons(ETHER_TYPE_IPv6):
137 ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + *l2_len);
138 *l3_len = sizeof(struct ipv6_hdr);
139 *l4_proto = ipv6_hdr->proto;
147 if (*l4_proto == IPPROTO_TCP) {
148 tcp_hdr = (struct tcp_hdr *)((char *)eth_hdr +
150 *l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
155 /* modify the IPv4 or IPv4 source address of a packet */
157 change_ip_addresses(void *l3_hdr, uint16_t ethertype)
159 struct ipv4_hdr *ipv4_hdr = l3_hdr;
160 struct ipv6_hdr *ipv6_hdr = l3_hdr;
162 if (ethertype == _htons(ETHER_TYPE_IPv4)) {
164 rte_cpu_to_be_32(rte_be_to_cpu_32(ipv4_hdr->src_addr) + 1);
165 } else if (ethertype == _htons(ETHER_TYPE_IPv6)) {
166 ipv6_hdr->src_addr[15] = ipv6_hdr->src_addr[15] + 1;
170 /* if possible, calculate the checksum of a packet in hw or sw,
171 * depending on the testpmd command line configuration */
173 process_inner_cksums(void *l3_hdr, uint16_t ethertype, uint16_t l3_len,
174 uint8_t l4_proto, uint16_t tso_segsz, uint16_t testpmd_ol_flags)
176 struct ipv4_hdr *ipv4_hdr = l3_hdr;
177 struct udp_hdr *udp_hdr;
178 struct tcp_hdr *tcp_hdr;
179 struct sctp_hdr *sctp_hdr;
180 uint64_t ol_flags = 0;
182 if (ethertype == _htons(ETHER_TYPE_IPv4)) {
184 ipv4_hdr->hdr_checksum = 0;
186 if (tso_segsz != 0 && l4_proto == IPPROTO_TCP) {
187 ol_flags |= PKT_TX_IP_CKSUM;
189 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
190 ol_flags |= PKT_TX_IP_CKSUM;
192 ipv4_hdr->hdr_checksum =
193 rte_ipv4_cksum(ipv4_hdr);
194 ol_flags |= PKT_TX_IPV4;
197 } else if (ethertype == _htons(ETHER_TYPE_IPv6))
198 ol_flags |= PKT_TX_IPV6;
200 return 0; /* packet type not supported, nothing to do */
202 if (l4_proto == IPPROTO_UDP) {
203 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + l3_len);
204 /* do not recalculate udp cksum if it was 0 */
205 if (udp_hdr->dgram_cksum != 0) {
206 udp_hdr->dgram_cksum = 0;
207 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM) {
208 ol_flags |= PKT_TX_UDP_CKSUM;
209 udp_hdr->dgram_cksum = get_psd_sum(l3_hdr,
210 ethertype, ol_flags);
212 udp_hdr->dgram_cksum =
213 get_udptcp_checksum(l3_hdr, udp_hdr,
217 } else if (l4_proto == IPPROTO_TCP) {
218 tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + l3_len);
220 if (tso_segsz != 0) {
221 ol_flags |= PKT_TX_TCP_SEG;
222 tcp_hdr->cksum = get_psd_sum(l3_hdr, ethertype, ol_flags);
223 } else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
224 ol_flags |= PKT_TX_TCP_CKSUM;
225 tcp_hdr->cksum = get_psd_sum(l3_hdr, ethertype, ol_flags);
228 get_udptcp_checksum(l3_hdr, tcp_hdr, ethertype);
230 } else if (l4_proto == IPPROTO_SCTP) {
231 sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + l3_len);
233 /* sctp payload must be a multiple of 4 to be
235 if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_SCTP_CKSUM) &&
236 ((ipv4_hdr->total_length & 0x3) == 0)) {
237 ol_flags |= PKT_TX_SCTP_CKSUM;
239 /* XXX implement CRC32c, example available in
247 /* Calculate the checksum of outer header (only vxlan is supported,
248 * meaning IP + UDP). The caller already checked that it's a vxlan
251 process_outer_cksums(void *outer_l3_hdr, uint16_t outer_ethertype,
252 uint16_t outer_l3_len, uint16_t testpmd_ol_flags)
254 struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
255 struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
256 struct udp_hdr *udp_hdr;
257 uint64_t ol_flags = 0;
259 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM)
260 ol_flags |= PKT_TX_UDP_TUNNEL_PKT;
262 if (outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
263 ipv4_hdr->hdr_checksum = 0;
265 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM)
266 ol_flags |= PKT_TX_OUTER_IP_CKSUM;
268 ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
269 } else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM)
270 ol_flags |= PKT_TX_OUTER_IPV6;
272 udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + outer_l3_len);
273 /* do not recalculate udp cksum if it was 0 */
274 if (udp_hdr->dgram_cksum != 0) {
275 udp_hdr->dgram_cksum = 0;
276 if (outer_ethertype == _htons(ETHER_TYPE_IPv4))
277 udp_hdr->dgram_cksum =
278 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
280 udp_hdr->dgram_cksum =
281 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
288 * Receive a burst of packets, and for each packet:
289 * - parse packet, and try to recognize a supported packet type (1)
290 * - if it's not a supported packet type, don't touch the packet, else:
291 * - modify the IPs in inner headers and in outer headers if any
292 * - reprocess the checksum of all supported layers. This is done in SW
293 * or HW, depending on testpmd command line configuration
294 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
295 * segmentation offload (this implies HW TCP checksum)
296 * Then transmit packets on the output port.
298 * (1) Supported packets are:
299 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
300 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
303 * The testpmd command line for this forward engine sets the flags
304 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
305 * wether a checksum must be calculated in software or in hardware. The
306 * IP, UDP, TCP and SCTP flags always concern the inner layer. The
307 * VxLAN flag concerns the outer IP (if packet is recognized as a vxlan packet).
310 pkt_burst_checksum_forward(struct fwd_stream *fs)
312 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
313 struct rte_port *txp;
315 struct ether_hdr *eth_hdr;
316 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
317 struct udp_hdr *udp_hdr;
322 uint16_t testpmd_ol_flags;
323 uint8_t l4_proto, l4_tun_len = 0;
324 uint16_t ethertype = 0, outer_ethertype = 0;
325 uint16_t l2_len = 0, l3_len = 0, l4_len = 0;
326 uint16_t outer_l2_len = 0, outer_l3_len = 0;
329 uint32_t rx_bad_ip_csum;
330 uint32_t rx_bad_l4_csum;
332 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
335 uint64_t core_cycles;
338 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
339 start_tsc = rte_rdtsc();
342 /* receive a burst of packet */
343 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
345 if (unlikely(nb_rx == 0))
348 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
349 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
351 fs->rx_packets += nb_rx;
355 txp = &ports[fs->tx_port];
356 testpmd_ol_flags = txp->tx_ol_flags;
357 tso_segsz = txp->tso_segsz;
359 for (i = 0; i < nb_rx; i++) {
366 /* Update the L3/L4 checksum error packet statistics */
367 rx_bad_ip_csum += ((m->ol_flags & PKT_RX_IP_CKSUM_BAD) != 0);
368 rx_bad_l4_csum += ((m->ol_flags & PKT_RX_L4_CKSUM_BAD) != 0);
370 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
371 * and inner headers */
373 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
374 parse_ethernet(eth_hdr, ðertype, &l2_len, &l3_len,
376 l3_hdr = (char *)eth_hdr + l2_len;
378 /* check if it's a supported tunnel (only vxlan for now) */
379 if (l4_proto == IPPROTO_UDP) {
380 udp_hdr = (struct udp_hdr *)((char *)l3_hdr + l3_len);
382 /* check udp destination port, 4789 is the default
383 * vxlan port (rfc7348) */
384 if (udp_hdr->dst_port == _htons(4789)) {
385 l4_tun_len = ETHER_VXLAN_HLEN;
388 /* currently, this flag is set by i40e only if the
390 } else if (m->ol_flags & (PKT_RX_TUNNEL_IPV4_HDR |
391 PKT_RX_TUNNEL_IPV6_HDR))
395 outer_ethertype = ethertype;
396 outer_l2_len = l2_len;
397 outer_l3_len = l3_len;
398 outer_l3_hdr = l3_hdr;
400 eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
401 sizeof(struct udp_hdr) +
402 sizeof(struct vxlan_hdr));
404 parse_ethernet(eth_hdr, ðertype, &l2_len,
405 &l3_len, &l4_proto, &l4_len);
406 l3_hdr = (char *)eth_hdr + l2_len;
410 /* step 2: change all source IPs (v4 or v6) so we need
411 * to recompute the chksums even if they were correct */
413 change_ip_addresses(l3_hdr, ethertype);
415 change_ip_addresses(outer_l3_hdr, outer_ethertype);
417 /* step 3: depending on user command line configuration,
418 * recompute checksum either in software or flag the
419 * mbuf to offload the calculation to the NIC. If TSO
420 * is configured, prepare the mbuf for TCP segmentation. */
422 /* process checksums of inner headers first */
423 ol_flags |= process_inner_cksums(l3_hdr, ethertype,
424 l3_len, l4_proto, tso_segsz, testpmd_ol_flags);
426 /* Then process outer headers if any. Note that the software
427 * checksum will be wrong if one of the inner checksums is
428 * processed in hardware. */
430 ol_flags |= process_outer_cksums(outer_l3_hdr,
431 outer_ethertype, outer_l3_len, testpmd_ol_flags);
434 /* step 4: fill the mbuf meta data (flags and header lengths) */
437 if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM) {
438 m->outer_l2_len = outer_l2_len;
439 m->outer_l3_len = outer_l3_len;
440 m->l2_len = l4_tun_len + l2_len;
444 /* if we don't do vxlan cksum in hw,
445 outer checksum will be wrong because
446 we changed the ip, but it shows that
447 we can process the inner header cksum
449 m->l2_len = outer_l2_len + outer_l3_len +
450 sizeof(struct udp_hdr) +
451 sizeof(struct vxlan_hdr) + l2_len;
456 /* this is only useful if an offload flag is
457 * set, but it does not hurt to fill it in any
463 m->tso_segsz = tso_segsz;
464 m->ol_flags = ol_flags;
466 /* if verbose mode is enabled, dump debug info */
467 if (verbose_level > 0) {
472 { PKT_TX_IP_CKSUM, PKT_TX_IP_CKSUM },
473 { PKT_TX_UDP_CKSUM, PKT_TX_L4_MASK },
474 { PKT_TX_TCP_CKSUM, PKT_TX_L4_MASK },
475 { PKT_TX_SCTP_CKSUM, PKT_TX_L4_MASK },
476 { PKT_TX_UDP_TUNNEL_PKT, PKT_TX_UDP_TUNNEL_PKT },
477 { PKT_TX_IPV4, PKT_TX_IPV4 },
478 { PKT_TX_IPV6, PKT_TX_IPV6 },
479 { PKT_TX_OUTER_IP_CKSUM, PKT_TX_OUTER_IP_CKSUM },
480 { PKT_TX_OUTER_IPV4, PKT_TX_OUTER_IPV4 },
481 { PKT_TX_OUTER_IPV6, PKT_TX_OUTER_IPV6 },
482 { PKT_TX_TCP_SEG, PKT_TX_TCP_SEG },
487 printf("-----------------\n");
488 /* dump rx parsed packet info */
489 printf("rx: l2_len=%d ethertype=%x l3_len=%d "
490 "l4_proto=%d l4_len=%d\n",
491 l2_len, rte_be_to_cpu_16(ethertype),
492 l3_len, l4_proto, l4_len);
494 printf("rx: outer_l2_len=%d outer_ethertype=%x "
495 "outer_l3_len=%d\n", outer_l2_len,
496 rte_be_to_cpu_16(outer_ethertype),
498 /* dump tx packet info */
499 if ((testpmd_ol_flags & (TESTPMD_TX_OFFLOAD_IP_CKSUM |
500 TESTPMD_TX_OFFLOAD_UDP_CKSUM |
501 TESTPMD_TX_OFFLOAD_TCP_CKSUM |
502 TESTPMD_TX_OFFLOAD_SCTP_CKSUM)) ||
504 printf("tx: m->l2_len=%d m->l3_len=%d "
506 m->l2_len, m->l3_len, m->l4_len);
508 (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_VXLAN_CKSUM))
509 printf("tx: m->outer_l2_len=%d m->outer_l3_len=%d\n",
510 m->outer_l2_len, m->outer_l3_len);
512 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
513 printf("tx: flags=");
514 for (j = 0; j < sizeof(tx_flags)/sizeof(*tx_flags); j++) {
515 name = rte_get_tx_ol_flag_name(tx_flags[j].flag);
516 if ((m->ol_flags & tx_flags[j].mask) ==
523 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
524 fs->tx_packets += nb_tx;
525 fs->rx_bad_ip_csum += rx_bad_ip_csum;
526 fs->rx_bad_l4_csum += rx_bad_l4_csum;
528 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
529 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
531 if (unlikely(nb_tx < nb_rx)) {
532 fs->fwd_dropped += (nb_rx - nb_tx);
534 rte_pktmbuf_free(pkts_burst[nb_tx]);
535 } while (++nb_tx < nb_rx);
537 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
538 end_tsc = rte_rdtsc();
539 core_cycles = (end_tsc - start_tsc);
540 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
544 struct fwd_engine csum_fwd_engine = {
545 .fwd_mode_name = "csum",
546 .port_fwd_begin = NULL,
547 .port_fwd_end = NULL,
548 .packet_fwd = pkt_burst_checksum_forward,