xref: /dpdk/app/test-pmd/csumonly.c (revision 2ebacaa7d13fa33d91e65f6a2e060eba3b0e1e5c)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5  *   Copyright 2014 6WIND S.A.
6  *   All rights reserved.
7  *
8  *   Redistribution and use in source and binary forms, with or without
9  *   modification, are permitted provided that the following conditions
10  *   are met:
11  *
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
17  *       distribution.
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.
21  *
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.
33  */
34 
35 #include <stdarg.h>
36 #include <stdio.h>
37 #include <errno.h>
38 #include <stdint.h>
39 #include <unistd.h>
40 #include <inttypes.h>
41 
42 #include <sys/queue.h>
43 #include <sys/stat.h>
44 
45 #include <rte_common.h>
46 #include <rte_byteorder.h>
47 #include <rte_log.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_eal.h>
55 #include <rte_per_lcore.h>
56 #include <rte_lcore.h>
57 #include <rte_atomic.h>
58 #include <rte_branch_prediction.h>
59 #include <rte_memory.h>
60 #include <rte_mempool.h>
61 #include <rte_mbuf.h>
62 #include <rte_memcpy.h>
63 #include <rte_interrupts.h>
64 #include <rte_pci.h>
65 #include <rte_ether.h>
66 #include <rte_ethdev.h>
67 #include <rte_ip.h>
68 #include <rte_tcp.h>
69 #include <rte_udp.h>
70 #include <rte_sctp.h>
71 #include <rte_prefetch.h>
72 #include <rte_string_fns.h>
73 #include "testpmd.h"
74 
75 #define IP_DEFTTL  64   /* from RFC 1340. */
76 #define IP_VERSION 0x40
77 #define IP_HDRLEN  0x05 /* default IP header length == five 32-bits words. */
78 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
79 
80 #define GRE_KEY_PRESENT 0x2000
81 #define GRE_KEY_LEN     4
82 #define GRE_SUPPORTED_FIELDS GRE_KEY_PRESENT
83 
84 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */
85 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
86 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
87 #else
88 #define _htons(x) (x)
89 #endif
90 
91 /* structure that caches offload info for the current packet */
92 struct testpmd_offload_info {
93 	uint16_t ethertype;
94 	uint16_t l2_len;
95 	uint16_t l3_len;
96 	uint16_t l4_len;
97 	uint8_t l4_proto;
98 	uint8_t is_tunnel;
99 	uint16_t outer_ethertype;
100 	uint16_t outer_l2_len;
101 	uint16_t outer_l3_len;
102 	uint8_t outer_l4_proto;
103 	uint16_t tso_segsz;
104 	uint16_t tunnel_tso_segsz;
105 	uint32_t pkt_len;
106 };
107 
108 /* simplified GRE header */
109 struct simple_gre_hdr {
110 	uint16_t flags;
111 	uint16_t proto;
112 } __attribute__((__packed__));
113 
114 static uint16_t
115 get_psd_sum(void *l3_hdr, uint16_t ethertype, uint64_t ol_flags)
116 {
117 	if (ethertype == _htons(ETHER_TYPE_IPv4))
118 		return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
119 	else /* assume ethertype == ETHER_TYPE_IPv6 */
120 		return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
121 }
122 
123 static uint16_t
124 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
125 {
126 	if (ethertype == _htons(ETHER_TYPE_IPv4))
127 		return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
128 	else /* assume ethertype == ETHER_TYPE_IPv6 */
129 		return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
130 }
131 
132 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
133 static void
134 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
135 {
136 	struct tcp_hdr *tcp_hdr;
137 
138 	info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
139 	info->l4_proto = ipv4_hdr->next_proto_id;
140 
141 	/* only fill l4_len for TCP, it's useful for TSO */
142 	if (info->l4_proto == IPPROTO_TCP) {
143 		tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
144 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
145 	} else
146 		info->l4_len = 0;
147 }
148 
149 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
150 static void
151 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
152 {
153 	struct tcp_hdr *tcp_hdr;
154 
155 	info->l3_len = sizeof(struct ipv6_hdr);
156 	info->l4_proto = ipv6_hdr->proto;
157 
158 	/* only fill l4_len for TCP, it's useful for TSO */
159 	if (info->l4_proto == IPPROTO_TCP) {
160 		tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
161 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
162 	} else
163 		info->l4_len = 0;
164 }
165 
166 /*
167  * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
168  * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
169  * header. The l4_len argument is only set in case of TCP (useful for TSO).
170  */
171 static void
172 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info)
173 {
174 	struct ipv4_hdr *ipv4_hdr;
175 	struct ipv6_hdr *ipv6_hdr;
176 
177 	info->l2_len = sizeof(struct ether_hdr);
178 	info->ethertype = eth_hdr->ether_type;
179 
180 	if (info->ethertype == _htons(ETHER_TYPE_VLAN)) {
181 		struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
182 
183 		info->l2_len  += sizeof(struct vlan_hdr);
184 		info->ethertype = vlan_hdr->eth_proto;
185 	}
186 
187 	switch (info->ethertype) {
188 	case _htons(ETHER_TYPE_IPv4):
189 		ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
190 		parse_ipv4(ipv4_hdr, info);
191 		break;
192 	case _htons(ETHER_TYPE_IPv6):
193 		ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
194 		parse_ipv6(ipv6_hdr, info);
195 		break;
196 	default:
197 		info->l4_len = 0;
198 		info->l3_len = 0;
199 		info->l4_proto = 0;
200 		break;
201 	}
202 }
203 
204 /* Parse a vxlan header */
205 static void
206 parse_vxlan(struct udp_hdr *udp_hdr,
207 	    struct testpmd_offload_info *info,
208 	    uint32_t pkt_type)
209 {
210 	struct ether_hdr *eth_hdr;
211 
212 	/* check udp destination port, 4789 is the default vxlan port
213 	 * (rfc7348) or that the rx offload flag is set (i40e only
214 	 * currently) */
215 	if (udp_hdr->dst_port != _htons(4789) &&
216 		RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
217 		return;
218 
219 	info->is_tunnel = 1;
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;
224 
225 	eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
226 		sizeof(struct udp_hdr) +
227 		sizeof(struct vxlan_hdr));
228 
229 	parse_ethernet(eth_hdr, info);
230 	info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */
231 }
232 
233 /* Parse a gre header */
234 static void
235 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
236 {
237 	struct ether_hdr *eth_hdr;
238 	struct ipv4_hdr *ipv4_hdr;
239 	struct ipv6_hdr *ipv6_hdr;
240 	uint8_t gre_len = 0;
241 
242 	/* check which fields are supported */
243 	if ((gre_hdr->flags & _htons(~GRE_SUPPORTED_FIELDS)) != 0)
244 		return;
245 
246 	gre_len += sizeof(struct simple_gre_hdr);
247 
248 	if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
249 		gre_len += GRE_KEY_LEN;
250 
251 	if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) {
252 		info->is_tunnel = 1;
253 		info->outer_ethertype = info->ethertype;
254 		info->outer_l2_len = info->l2_len;
255 		info->outer_l3_len = info->l3_len;
256 		info->outer_l4_proto = info->l4_proto;
257 
258 		ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len);
259 
260 		parse_ipv4(ipv4_hdr, info);
261 		info->ethertype = _htons(ETHER_TYPE_IPv4);
262 		info->l2_len = 0;
263 
264 	} else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) {
265 		info->is_tunnel = 1;
266 		info->outer_ethertype = info->ethertype;
267 		info->outer_l2_len = info->l2_len;
268 		info->outer_l3_len = info->l3_len;
269 		info->outer_l4_proto = info->l4_proto;
270 
271 		ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len);
272 
273 		info->ethertype = _htons(ETHER_TYPE_IPv6);
274 		parse_ipv6(ipv6_hdr, info);
275 		info->l2_len = 0;
276 
277 	} else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) {
278 		info->is_tunnel = 1;
279 		info->outer_ethertype = info->ethertype;
280 		info->outer_l2_len = info->l2_len;
281 		info->outer_l3_len = info->l3_len;
282 		info->outer_l4_proto = info->l4_proto;
283 
284 		eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len);
285 
286 		parse_ethernet(eth_hdr, info);
287 	} else
288 		return;
289 
290 	info->l2_len += gre_len;
291 }
292 
293 
294 /* Parse an encapsulated ip or ipv6 header */
295 static void
296 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
297 {
298 	struct ipv4_hdr *ipv4_hdr = encap_ip;
299 	struct ipv6_hdr *ipv6_hdr = encap_ip;
300 	uint8_t ip_version;
301 
302 	ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
303 
304 	if (ip_version != 4 && ip_version != 6)
305 		return;
306 
307 	info->is_tunnel = 1;
308 	info->outer_ethertype = info->ethertype;
309 	info->outer_l2_len = info->l2_len;
310 	info->outer_l3_len = info->l3_len;
311 
312 	if (ip_version == 4) {
313 		parse_ipv4(ipv4_hdr, info);
314 		info->ethertype = _htons(ETHER_TYPE_IPv4);
315 	} else {
316 		parse_ipv6(ipv6_hdr, info);
317 		info->ethertype = _htons(ETHER_TYPE_IPv6);
318 	}
319 	info->l2_len = 0;
320 }
321 
322 /* if possible, calculate the checksum of a packet in hw or sw,
323  * depending on the testpmd command line configuration */
324 static uint64_t
325 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
326 	uint16_t testpmd_ol_flags)
327 {
328 	struct ipv4_hdr *ipv4_hdr = l3_hdr;
329 	struct udp_hdr *udp_hdr;
330 	struct tcp_hdr *tcp_hdr;
331 	struct sctp_hdr *sctp_hdr;
332 	uint64_t ol_flags = 0;
333 	uint32_t max_pkt_len, tso_segsz = 0;
334 
335 	/* ensure packet is large enough to require tso */
336 	if (!info->is_tunnel) {
337 		max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
338 			info->tso_segsz;
339 		if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
340 			tso_segsz = info->tso_segsz;
341 	} else {
342 		max_pkt_len = info->outer_l2_len + info->outer_l3_len +
343 			info->l2_len + info->l3_len + info->l4_len +
344 			info->tunnel_tso_segsz;
345 		if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
346 			tso_segsz = info->tunnel_tso_segsz;
347 	}
348 
349 	if (info->ethertype == _htons(ETHER_TYPE_IPv4)) {
350 		ipv4_hdr = l3_hdr;
351 		ipv4_hdr->hdr_checksum = 0;
352 
353 		ol_flags |= PKT_TX_IPV4;
354 		if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
355 			ol_flags |= PKT_TX_IP_CKSUM;
356 		} else {
357 			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
358 				ol_flags |= PKT_TX_IP_CKSUM;
359 			else
360 				ipv4_hdr->hdr_checksum =
361 					rte_ipv4_cksum(ipv4_hdr);
362 		}
363 	} else if (info->ethertype == _htons(ETHER_TYPE_IPv6))
364 		ol_flags |= PKT_TX_IPV6;
365 	else
366 		return 0; /* packet type not supported, nothing to do */
367 
368 	if (info->l4_proto == IPPROTO_UDP) {
369 		udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
370 		/* do not recalculate udp cksum if it was 0 */
371 		if (udp_hdr->dgram_cksum != 0) {
372 			udp_hdr->dgram_cksum = 0;
373 			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM) {
374 				ol_flags |= PKT_TX_UDP_CKSUM;
375 				udp_hdr->dgram_cksum = get_psd_sum(l3_hdr,
376 					info->ethertype, ol_flags);
377 			} else {
378 				udp_hdr->dgram_cksum =
379 					get_udptcp_checksum(l3_hdr, udp_hdr,
380 						info->ethertype);
381 			}
382 		}
383 	} else if (info->l4_proto == IPPROTO_TCP) {
384 		tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
385 		tcp_hdr->cksum = 0;
386 		if (tso_segsz) {
387 			ol_flags |= PKT_TX_TCP_SEG;
388 			tcp_hdr->cksum = get_psd_sum(l3_hdr, info->ethertype,
389 				ol_flags);
390 		} else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
391 			ol_flags |= PKT_TX_TCP_CKSUM;
392 			tcp_hdr->cksum = get_psd_sum(l3_hdr, info->ethertype,
393 				ol_flags);
394 		} else {
395 			tcp_hdr->cksum =
396 				get_udptcp_checksum(l3_hdr, tcp_hdr,
397 					info->ethertype);
398 		}
399 	} else if (info->l4_proto == IPPROTO_SCTP) {
400 		sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
401 		sctp_hdr->cksum = 0;
402 		/* sctp payload must be a multiple of 4 to be
403 		 * offloaded */
404 		if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_SCTP_CKSUM) &&
405 			((ipv4_hdr->total_length & 0x3) == 0)) {
406 			ol_flags |= PKT_TX_SCTP_CKSUM;
407 		} else {
408 			/* XXX implement CRC32c, example available in
409 			 * RFC3309 */
410 		}
411 	}
412 
413 	return ol_flags;
414 }
415 
416 /* Calculate the checksum of outer header */
417 static uint64_t
418 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
419 	uint16_t testpmd_ol_flags, int tso_enabled)
420 {
421 	struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
422 	struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
423 	struct udp_hdr *udp_hdr;
424 	uint64_t ol_flags = 0;
425 
426 	if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
427 		ipv4_hdr->hdr_checksum = 0;
428 		ol_flags |= PKT_TX_OUTER_IPV4;
429 
430 		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
431 			ol_flags |= PKT_TX_OUTER_IP_CKSUM;
432 		else
433 			ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
434 	} else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
435 		ol_flags |= PKT_TX_OUTER_IPV6;
436 
437 	if (info->outer_l4_proto != IPPROTO_UDP)
438 		return ol_flags;
439 
440 	udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
441 
442 	/* outer UDP checksum is done in software as we have no hardware
443 	 * supporting it today, and no API for it. In the other side, for
444 	 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
445 	 * set to zero.
446 	 *
447 	 * If a packet will be TSOed into small packets by NIC, we cannot
448 	 * set/calculate a non-zero checksum, because it will be a wrong
449 	 * value after the packet be split into several small packets.
450 	 */
451 	if (tso_enabled)
452 		udp_hdr->dgram_cksum = 0;
453 
454 	/* do not recalculate udp cksum if it was 0 */
455 	if (udp_hdr->dgram_cksum != 0) {
456 		udp_hdr->dgram_cksum = 0;
457 		if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
458 			udp_hdr->dgram_cksum =
459 				rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
460 		else
461 			udp_hdr->dgram_cksum =
462 				rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
463 	}
464 
465 	return ol_flags;
466 }
467 
468 /*
469  * Helper function.
470  * Performs actual copying.
471  * Returns number of segments in the destination mbuf on success,
472  * or negative error code on failure.
473  */
474 static int
475 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
476 	uint16_t seglen[], uint8_t nb_seg)
477 {
478 	uint32_t dlen, slen, tlen;
479 	uint32_t i, len;
480 	const struct rte_mbuf *m;
481 	const uint8_t *src;
482 	uint8_t *dst;
483 
484 	dlen = 0;
485 	slen = 0;
486 	tlen = 0;
487 
488 	dst = NULL;
489 	src = NULL;
490 
491 	m = ms;
492 	i = 0;
493 	while (ms != NULL && i != nb_seg) {
494 
495 		if (slen == 0) {
496 			slen = rte_pktmbuf_data_len(ms);
497 			src = rte_pktmbuf_mtod(ms, const uint8_t *);
498 		}
499 
500 		if (dlen == 0) {
501 			dlen = RTE_MIN(seglen[i], slen);
502 			md[i]->data_len = dlen;
503 			md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
504 			dst = rte_pktmbuf_mtod(md[i], uint8_t *);
505 		}
506 
507 		len = RTE_MIN(slen, dlen);
508 		memcpy(dst, src, len);
509 		tlen += len;
510 		slen -= len;
511 		dlen -= len;
512 		src += len;
513 		dst += len;
514 
515 		if (slen == 0)
516 			ms = ms->next;
517 		if (dlen == 0)
518 			i++;
519 	}
520 
521 	if (ms != NULL)
522 		return -ENOBUFS;
523 	else if (tlen != m->pkt_len)
524 		return -EINVAL;
525 
526 	md[0]->nb_segs = nb_seg;
527 	md[0]->pkt_len = tlen;
528 	md[0]->vlan_tci = m->vlan_tci;
529 	md[0]->vlan_tci_outer = m->vlan_tci_outer;
530 	md[0]->ol_flags = m->ol_flags;
531 	md[0]->tx_offload = m->tx_offload;
532 
533 	return nb_seg;
534 }
535 
536 /*
537  * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
538  * Copy packet contents and offload information into then new segmented mbuf.
539  */
540 static struct rte_mbuf *
541 pkt_copy_split(const struct rte_mbuf *pkt)
542 {
543 	int32_t n, rc;
544 	uint32_t i, len, nb_seg;
545 	struct rte_mempool *mp;
546 	uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
547 	struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
548 
549 	mp = current_fwd_lcore()->mbp;
550 
551 	if (tx_pkt_split == TX_PKT_SPLIT_RND)
552 		nb_seg = random() % tx_pkt_nb_segs + 1;
553 	else
554 		nb_seg = tx_pkt_nb_segs;
555 
556 	memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
557 
558 	/* calculate number of segments to use and their length. */
559 	len = 0;
560 	for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
561 		len += seglen[i];
562 		md[i] = NULL;
563 	}
564 
565 	n = pkt->pkt_len - len;
566 
567 	/* update size of the last segment to fit rest of the packet */
568 	if (n >= 0) {
569 		seglen[i - 1] += n;
570 		len += n;
571 	}
572 
573 	nb_seg = i;
574 	while (i != 0) {
575 		p = rte_pktmbuf_alloc(mp);
576 		if (p == NULL) {
577 			RTE_LOG(ERR, USER1,
578 				"failed to allocate %u-th of %u mbuf "
579 				"from mempool: %s\n",
580 				nb_seg - i, nb_seg, mp->name);
581 			break;
582 		}
583 
584 		md[--i] = p;
585 		if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
586 			RTE_LOG(ERR, USER1, "mempool %s, %u-th segment: "
587 				"expected seglen: %u, "
588 				"actual mbuf tailroom: %u\n",
589 				mp->name, i, seglen[i],
590 				rte_pktmbuf_tailroom(md[i]));
591 			break;
592 		}
593 	}
594 
595 	/* all mbufs successfully allocated, do copy */
596 	if (i == 0) {
597 		rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
598 		if (rc < 0)
599 			RTE_LOG(ERR, USER1,
600 				"mbuf_copy_split for %p(len=%u, nb_seg=%hhu) "
601 				"into %u segments failed with error code: %d\n",
602 				pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
603 
604 		/* figure out how many mbufs to free. */
605 		i = RTE_MAX(rc, 0);
606 	}
607 
608 	/* free unused mbufs */
609 	for (; i != nb_seg; i++) {
610 		rte_pktmbuf_free_seg(md[i]);
611 		md[i] = NULL;
612 	}
613 
614 	return md[0];
615 }
616 
617 /*
618  * Receive a burst of packets, and for each packet:
619  *  - parse packet, and try to recognize a supported packet type (1)
620  *  - if it's not a supported packet type, don't touch the packet, else:
621  *  - reprocess the checksum of all supported layers. This is done in SW
622  *    or HW, depending on testpmd command line configuration
623  *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
624  *    segmentation offload (this implies HW TCP checksum)
625  * Then transmit packets on the output port.
626  *
627  * (1) Supported packets are:
628  *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
629  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
630  *           UDP|TCP|SCTP
631  *   Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
632  *   Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
633  *   Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
634  *
635  * The testpmd command line for this forward engine sets the flags
636  * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
637  * wether a checksum must be calculated in software or in hardware. The
638  * IP, UDP, TCP and SCTP flags always concern the inner layer. The
639  * OUTER_IP is only useful for tunnel packets.
640  */
641 static void
642 pkt_burst_checksum_forward(struct fwd_stream *fs)
643 {
644 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
645 	struct rte_port *txp;
646 	struct rte_mbuf *m, *p;
647 	struct ether_hdr *eth_hdr;
648 	void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
649 	uint16_t nb_rx;
650 	uint16_t nb_tx;
651 	uint16_t i;
652 	uint64_t rx_ol_flags, tx_ol_flags;
653 	uint16_t testpmd_ol_flags;
654 	uint32_t retry;
655 	uint32_t rx_bad_ip_csum;
656 	uint32_t rx_bad_l4_csum;
657 	struct testpmd_offload_info info;
658 
659 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
660 	uint64_t start_tsc;
661 	uint64_t end_tsc;
662 	uint64_t core_cycles;
663 #endif
664 
665 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
666 	start_tsc = rte_rdtsc();
667 #endif
668 
669 	/* receive a burst of packet */
670 	nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
671 				 nb_pkt_per_burst);
672 	if (unlikely(nb_rx == 0))
673 		return;
674 
675 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
676 	fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
677 #endif
678 	fs->rx_packets += nb_rx;
679 	rx_bad_ip_csum = 0;
680 	rx_bad_l4_csum = 0;
681 
682 	txp = &ports[fs->tx_port];
683 	testpmd_ol_flags = txp->tx_ol_flags;
684 	memset(&info, 0, sizeof(info));
685 	info.tso_segsz = txp->tso_segsz;
686 	info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
687 
688 	for (i = 0; i < nb_rx; i++) {
689 		if (likely(i < nb_rx - 1))
690 			rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
691 						       void *));
692 
693 		m = pkts_burst[i];
694 		info.is_tunnel = 0;
695 		info.pkt_len = rte_pktmbuf_pkt_len(m);
696 		tx_ol_flags = 0;
697 		rx_ol_flags = m->ol_flags;
698 
699 		/* Update the L3/L4 checksum error packet statistics */
700 		rx_bad_ip_csum += ((rx_ol_flags & PKT_RX_IP_CKSUM_BAD) != 0);
701 		rx_bad_l4_csum += ((rx_ol_flags & PKT_RX_L4_CKSUM_BAD) != 0);
702 
703 		/* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
704 		 * and inner headers */
705 
706 		eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
707 		ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
708 				&eth_hdr->d_addr);
709 		ether_addr_copy(&ports[fs->tx_port].eth_addr,
710 				&eth_hdr->s_addr);
711 		parse_ethernet(eth_hdr, &info);
712 		l3_hdr = (char *)eth_hdr + info.l2_len;
713 
714 		/* check if it's a supported tunnel */
715 		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_PARSE_TUNNEL) {
716 			if (info.l4_proto == IPPROTO_UDP) {
717 				struct udp_hdr *udp_hdr;
718 
719 				udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
720 					info.l3_len);
721 				parse_vxlan(udp_hdr, &info, m->packet_type);
722 				if (info.is_tunnel)
723 					tx_ol_flags |= PKT_TX_TUNNEL_VXLAN;
724 			} else if (info.l4_proto == IPPROTO_GRE) {
725 				struct simple_gre_hdr *gre_hdr;
726 
727 				gre_hdr = (struct simple_gre_hdr *)
728 					((char *)l3_hdr + info.l3_len);
729 				parse_gre(gre_hdr, &info);
730 				if (info.is_tunnel)
731 					tx_ol_flags |= PKT_TX_TUNNEL_GRE;
732 			} else if (info.l4_proto == IPPROTO_IPIP) {
733 				void *encap_ip_hdr;
734 
735 				encap_ip_hdr = (char *)l3_hdr + info.l3_len;
736 				parse_encap_ip(encap_ip_hdr, &info);
737 				if (info.is_tunnel)
738 					tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
739 			}
740 		}
741 
742 		/* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
743 		if (info.is_tunnel) {
744 			outer_l3_hdr = l3_hdr;
745 			l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
746 		}
747 
748 		/* step 2: depending on user command line configuration,
749 		 * recompute checksum either in software or flag the
750 		 * mbuf to offload the calculation to the NIC. If TSO
751 		 * is configured, prepare the mbuf for TCP segmentation. */
752 
753 		/* process checksums of inner headers first */
754 		tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
755 			testpmd_ol_flags);
756 
757 		/* Then process outer headers if any. Note that the software
758 		 * checksum will be wrong if one of the inner checksums is
759 		 * processed in hardware. */
760 		if (info.is_tunnel == 1) {
761 			tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
762 					testpmd_ol_flags,
763 					!!(tx_ol_flags & PKT_TX_TCP_SEG));
764 		}
765 
766 		/* step 3: fill the mbuf meta data (flags and header lengths) */
767 
768 		if (info.is_tunnel == 1) {
769 			if (info.tunnel_tso_segsz ||
770 			    testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) {
771 				m->outer_l2_len = info.outer_l2_len;
772 				m->outer_l3_len = info.outer_l3_len;
773 				m->l2_len = info.l2_len;
774 				m->l3_len = info.l3_len;
775 				m->l4_len = info.l4_len;
776 				m->tso_segsz = info.tunnel_tso_segsz;
777 			}
778 			else {
779 				/* if there is a outer UDP cksum
780 				   processed in sw and the inner in hw,
781 				   the outer checksum will be wrong as
782 				   the payload will be modified by the
783 				   hardware */
784 				m->l2_len = info.outer_l2_len +
785 					info.outer_l3_len + info.l2_len;
786 				m->l3_len = info.l3_len;
787 				m->l4_len = info.l4_len;
788 			}
789 		} else {
790 			/* this is only useful if an offload flag is
791 			 * set, but it does not hurt to fill it in any
792 			 * case */
793 			m->l2_len = info.l2_len;
794 			m->l3_len = info.l3_len;
795 			m->l4_len = info.l4_len;
796 			m->tso_segsz = info.tso_segsz;
797 		}
798 		m->ol_flags = tx_ol_flags;
799 
800 		/* Do split & copy for the packet. */
801 		if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
802 			p = pkt_copy_split(m);
803 			if (p != NULL) {
804 				rte_pktmbuf_free(m);
805 				m = p;
806 				pkts_burst[i] = m;
807 			}
808 		}
809 
810 		/* if verbose mode is enabled, dump debug info */
811 		if (verbose_level > 0) {
812 			char buf[256];
813 
814 			printf("-----------------\n");
815 			printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%hhu:\n",
816 				fs->rx_port, m, m->pkt_len, m->nb_segs);
817 			/* dump rx parsed packet info */
818 			rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
819 			printf("rx: l2_len=%d ethertype=%x l3_len=%d "
820 				"l4_proto=%d l4_len=%d flags=%s\n",
821 				info.l2_len, rte_be_to_cpu_16(info.ethertype),
822 				info.l3_len, info.l4_proto, info.l4_len, buf);
823 			if (info.is_tunnel == 1)
824 				printf("rx: outer_l2_len=%d outer_ethertype=%x "
825 					"outer_l3_len=%d\n", info.outer_l2_len,
826 					rte_be_to_cpu_16(info.outer_ethertype),
827 					info.outer_l3_len);
828 			/* dump tx packet info */
829 			if ((testpmd_ol_flags & (TESTPMD_TX_OFFLOAD_IP_CKSUM |
830 						TESTPMD_TX_OFFLOAD_UDP_CKSUM |
831 						TESTPMD_TX_OFFLOAD_TCP_CKSUM |
832 						TESTPMD_TX_OFFLOAD_SCTP_CKSUM)) ||
833 				info.tso_segsz != 0)
834 				printf("tx: m->l2_len=%d m->l3_len=%d "
835 					"m->l4_len=%d\n",
836 					m->l2_len, m->l3_len, m->l4_len);
837 			if (info.is_tunnel == 1) {
838 				if (testpmd_ol_flags &
839 				    TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
840 					printf("tx: m->outer_l2_len=%d "
841 						"m->outer_l3_len=%d\n",
842 						m->outer_l2_len,
843 						m->outer_l3_len);
844 				if (info.tunnel_tso_segsz != 0 &&
845 						(m->ol_flags & PKT_TX_TCP_SEG))
846 					printf("tx: m->tso_segsz=%d\n",
847 						m->tso_segsz);
848 			} else if (info.tso_segsz != 0 &&
849 					(m->ol_flags & PKT_TX_TCP_SEG))
850 				printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
851 			rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
852 			printf("tx: flags=%s", buf);
853 			printf("\n");
854 		}
855 	}
856 	nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
857 	/*
858 	 * Retry if necessary
859 	 */
860 	if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
861 		retry = 0;
862 		while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
863 			rte_delay_us(burst_tx_delay_time);
864 			nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
865 					&pkts_burst[nb_tx], nb_rx - nb_tx);
866 		}
867 	}
868 	fs->tx_packets += nb_tx;
869 	fs->rx_bad_ip_csum += rx_bad_ip_csum;
870 	fs->rx_bad_l4_csum += rx_bad_l4_csum;
871 
872 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
873 	fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
874 #endif
875 	if (unlikely(nb_tx < nb_rx)) {
876 		fs->fwd_dropped += (nb_rx - nb_tx);
877 		do {
878 			rte_pktmbuf_free(pkts_burst[nb_tx]);
879 		} while (++nb_tx < nb_rx);
880 	}
881 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
882 	end_tsc = rte_rdtsc();
883 	core_cycles = (end_tsc - start_tsc);
884 	fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
885 #endif
886 }
887 
888 struct fwd_engine csum_fwd_engine = {
889 	.fwd_mode_name  = "csum",
890 	.port_fwd_begin = NULL,
891 	.port_fwd_end   = NULL,
892 	.packet_fwd     = pkt_burst_checksum_forward,
893 };
894