xref: /dpdk/app/test-pmd/csumonly.c (revision 4e30ead5e7ca886535e2b30632b2948d2aac1681)
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 <rte_flow.h>
74 #include "testpmd.h"
75 
76 #define IP_DEFTTL  64   /* from RFC 1340. */
77 #define IP_VERSION 0x40
78 #define IP_HDRLEN  0x05 /* default IP header length == five 32-bits words. */
79 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
80 
81 #define GRE_KEY_PRESENT 0x2000
82 #define GRE_KEY_LEN     4
83 #define GRE_SUPPORTED_FIELDS GRE_KEY_PRESENT
84 
85 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */
86 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
87 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
88 #else
89 #define _htons(x) (x)
90 #endif
91 
92 /* structure that caches offload info for the current packet */
93 struct testpmd_offload_info {
94 	uint16_t ethertype;
95 	uint16_t l2_len;
96 	uint16_t l3_len;
97 	uint16_t l4_len;
98 	uint8_t l4_proto;
99 	uint8_t is_tunnel;
100 	uint16_t outer_ethertype;
101 	uint16_t outer_l2_len;
102 	uint16_t outer_l3_len;
103 	uint8_t outer_l4_proto;
104 	uint16_t tso_segsz;
105 	uint16_t tunnel_tso_segsz;
106 	uint32_t pkt_len;
107 };
108 
109 /* simplified GRE header */
110 struct simple_gre_hdr {
111 	uint16_t flags;
112 	uint16_t proto;
113 } __attribute__((__packed__));
114 
115 static uint16_t
116 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
117 {
118 	if (ethertype == _htons(ETHER_TYPE_IPv4))
119 		return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
120 	else /* assume ethertype == ETHER_TYPE_IPv6 */
121 		return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
122 }
123 
124 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
125 static void
126 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
127 {
128 	struct tcp_hdr *tcp_hdr;
129 
130 	info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
131 	info->l4_proto = ipv4_hdr->next_proto_id;
132 
133 	/* only fill l4_len for TCP, it's useful for TSO */
134 	if (info->l4_proto == IPPROTO_TCP) {
135 		tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
136 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
137 	} else
138 		info->l4_len = 0;
139 }
140 
141 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
142 static void
143 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
144 {
145 	struct tcp_hdr *tcp_hdr;
146 
147 	info->l3_len = sizeof(struct ipv6_hdr);
148 	info->l4_proto = ipv6_hdr->proto;
149 
150 	/* only fill l4_len for TCP, it's useful for TSO */
151 	if (info->l4_proto == IPPROTO_TCP) {
152 		tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
153 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
154 	} else
155 		info->l4_len = 0;
156 }
157 
158 /*
159  * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
160  * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
161  * header. The l4_len argument is only set in case of TCP (useful for TSO).
162  */
163 static void
164 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info)
165 {
166 	struct ipv4_hdr *ipv4_hdr;
167 	struct ipv6_hdr *ipv6_hdr;
168 
169 	info->l2_len = sizeof(struct ether_hdr);
170 	info->ethertype = eth_hdr->ether_type;
171 
172 	if (info->ethertype == _htons(ETHER_TYPE_VLAN)) {
173 		struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
174 
175 		info->l2_len  += sizeof(struct vlan_hdr);
176 		info->ethertype = vlan_hdr->eth_proto;
177 	}
178 
179 	switch (info->ethertype) {
180 	case _htons(ETHER_TYPE_IPv4):
181 		ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
182 		parse_ipv4(ipv4_hdr, info);
183 		break;
184 	case _htons(ETHER_TYPE_IPv6):
185 		ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
186 		parse_ipv6(ipv6_hdr, info);
187 		break;
188 	default:
189 		info->l4_len = 0;
190 		info->l3_len = 0;
191 		info->l4_proto = 0;
192 		break;
193 	}
194 }
195 
196 /* Parse a vxlan header */
197 static void
198 parse_vxlan(struct udp_hdr *udp_hdr,
199 	    struct testpmd_offload_info *info,
200 	    uint32_t pkt_type)
201 {
202 	struct ether_hdr *eth_hdr;
203 
204 	/* check udp destination port, 4789 is the default vxlan port
205 	 * (rfc7348) or that the rx offload flag is set (i40e only
206 	 * currently) */
207 	if (udp_hdr->dst_port != _htons(4789) &&
208 		RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
209 		return;
210 
211 	info->is_tunnel = 1;
212 	info->outer_ethertype = info->ethertype;
213 	info->outer_l2_len = info->l2_len;
214 	info->outer_l3_len = info->l3_len;
215 	info->outer_l4_proto = info->l4_proto;
216 
217 	eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
218 		sizeof(struct udp_hdr) +
219 		sizeof(struct vxlan_hdr));
220 
221 	parse_ethernet(eth_hdr, info);
222 	info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */
223 }
224 
225 /* Parse a gre header */
226 static void
227 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
228 {
229 	struct ether_hdr *eth_hdr;
230 	struct ipv4_hdr *ipv4_hdr;
231 	struct ipv6_hdr *ipv6_hdr;
232 	uint8_t gre_len = 0;
233 
234 	/* check which fields are supported */
235 	if ((gre_hdr->flags & _htons(~GRE_SUPPORTED_FIELDS)) != 0)
236 		return;
237 
238 	gre_len += sizeof(struct simple_gre_hdr);
239 
240 	if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
241 		gre_len += GRE_KEY_LEN;
242 
243 	if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) {
244 		info->is_tunnel = 1;
245 		info->outer_ethertype = info->ethertype;
246 		info->outer_l2_len = info->l2_len;
247 		info->outer_l3_len = info->l3_len;
248 		info->outer_l4_proto = info->l4_proto;
249 
250 		ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len);
251 
252 		parse_ipv4(ipv4_hdr, info);
253 		info->ethertype = _htons(ETHER_TYPE_IPv4);
254 		info->l2_len = 0;
255 
256 	} else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) {
257 		info->is_tunnel = 1;
258 		info->outer_ethertype = info->ethertype;
259 		info->outer_l2_len = info->l2_len;
260 		info->outer_l3_len = info->l3_len;
261 		info->outer_l4_proto = info->l4_proto;
262 
263 		ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len);
264 
265 		info->ethertype = _htons(ETHER_TYPE_IPv6);
266 		parse_ipv6(ipv6_hdr, info);
267 		info->l2_len = 0;
268 
269 	} else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) {
270 		info->is_tunnel = 1;
271 		info->outer_ethertype = info->ethertype;
272 		info->outer_l2_len = info->l2_len;
273 		info->outer_l3_len = info->l3_len;
274 		info->outer_l4_proto = info->l4_proto;
275 
276 		eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len);
277 
278 		parse_ethernet(eth_hdr, info);
279 	} else
280 		return;
281 
282 	info->l2_len += gre_len;
283 }
284 
285 
286 /* Parse an encapsulated ip or ipv6 header */
287 static void
288 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
289 {
290 	struct ipv4_hdr *ipv4_hdr = encap_ip;
291 	struct ipv6_hdr *ipv6_hdr = encap_ip;
292 	uint8_t ip_version;
293 
294 	ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
295 
296 	if (ip_version != 4 && ip_version != 6)
297 		return;
298 
299 	info->is_tunnel = 1;
300 	info->outer_ethertype = info->ethertype;
301 	info->outer_l2_len = info->l2_len;
302 	info->outer_l3_len = info->l3_len;
303 
304 	if (ip_version == 4) {
305 		parse_ipv4(ipv4_hdr, info);
306 		info->ethertype = _htons(ETHER_TYPE_IPv4);
307 	} else {
308 		parse_ipv6(ipv6_hdr, info);
309 		info->ethertype = _htons(ETHER_TYPE_IPv6);
310 	}
311 	info->l2_len = 0;
312 }
313 
314 /* if possible, calculate the checksum of a packet in hw or sw,
315  * depending on the testpmd command line configuration */
316 static uint64_t
317 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
318 	uint16_t testpmd_ol_flags)
319 {
320 	struct ipv4_hdr *ipv4_hdr = l3_hdr;
321 	struct udp_hdr *udp_hdr;
322 	struct tcp_hdr *tcp_hdr;
323 	struct sctp_hdr *sctp_hdr;
324 	uint64_t ol_flags = 0;
325 	uint32_t max_pkt_len, tso_segsz = 0;
326 
327 	/* ensure packet is large enough to require tso */
328 	if (!info->is_tunnel) {
329 		max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
330 			info->tso_segsz;
331 		if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
332 			tso_segsz = info->tso_segsz;
333 	} else {
334 		max_pkt_len = info->outer_l2_len + info->outer_l3_len +
335 			info->l2_len + info->l3_len + info->l4_len +
336 			info->tunnel_tso_segsz;
337 		if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
338 			tso_segsz = info->tunnel_tso_segsz;
339 	}
340 
341 	if (info->ethertype == _htons(ETHER_TYPE_IPv4)) {
342 		ipv4_hdr = l3_hdr;
343 		ipv4_hdr->hdr_checksum = 0;
344 
345 		ol_flags |= PKT_TX_IPV4;
346 		if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
347 			ol_flags |= PKT_TX_IP_CKSUM;
348 		} else {
349 			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
350 				ol_flags |= PKT_TX_IP_CKSUM;
351 			else
352 				ipv4_hdr->hdr_checksum =
353 					rte_ipv4_cksum(ipv4_hdr);
354 		}
355 	} else if (info->ethertype == _htons(ETHER_TYPE_IPv6))
356 		ol_flags |= PKT_TX_IPV6;
357 	else
358 		return 0; /* packet type not supported, nothing to do */
359 
360 	if (info->l4_proto == IPPROTO_UDP) {
361 		udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
362 		/* do not recalculate udp cksum if it was 0 */
363 		if (udp_hdr->dgram_cksum != 0) {
364 			udp_hdr->dgram_cksum = 0;
365 			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM)
366 				ol_flags |= PKT_TX_UDP_CKSUM;
367 			else {
368 				udp_hdr->dgram_cksum =
369 					get_udptcp_checksum(l3_hdr, udp_hdr,
370 						info->ethertype);
371 			}
372 		}
373 	} else if (info->l4_proto == IPPROTO_TCP) {
374 		tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
375 		tcp_hdr->cksum = 0;
376 		if (tso_segsz)
377 			ol_flags |= PKT_TX_TCP_SEG;
378 		else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM)
379 			ol_flags |= PKT_TX_TCP_CKSUM;
380 		else {
381 			tcp_hdr->cksum =
382 				get_udptcp_checksum(l3_hdr, tcp_hdr,
383 					info->ethertype);
384 		}
385 	} else if (info->l4_proto == IPPROTO_SCTP) {
386 		sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
387 		sctp_hdr->cksum = 0;
388 		/* sctp payload must be a multiple of 4 to be
389 		 * offloaded */
390 		if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_SCTP_CKSUM) &&
391 			((ipv4_hdr->total_length & 0x3) == 0)) {
392 			ol_flags |= PKT_TX_SCTP_CKSUM;
393 		} else {
394 			/* XXX implement CRC32c, example available in
395 			 * RFC3309 */
396 		}
397 	}
398 
399 	return ol_flags;
400 }
401 
402 /* Calculate the checksum of outer header */
403 static uint64_t
404 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
405 	uint16_t testpmd_ol_flags, int tso_enabled)
406 {
407 	struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
408 	struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
409 	struct udp_hdr *udp_hdr;
410 	uint64_t ol_flags = 0;
411 
412 	if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
413 		ipv4_hdr->hdr_checksum = 0;
414 		ol_flags |= PKT_TX_OUTER_IPV4;
415 
416 		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
417 			ol_flags |= PKT_TX_OUTER_IP_CKSUM;
418 		else
419 			ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
420 	} else
421 		ol_flags |= PKT_TX_OUTER_IPV6;
422 
423 	if (info->outer_l4_proto != IPPROTO_UDP)
424 		return ol_flags;
425 
426 	udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
427 
428 	/* outer UDP checksum is done in software as we have no hardware
429 	 * supporting it today, and no API for it. In the other side, for
430 	 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
431 	 * set to zero.
432 	 *
433 	 * If a packet will be TSOed into small packets by NIC, we cannot
434 	 * set/calculate a non-zero checksum, because it will be a wrong
435 	 * value after the packet be split into several small packets.
436 	 */
437 	if (tso_enabled)
438 		udp_hdr->dgram_cksum = 0;
439 
440 	/* do not recalculate udp cksum if it was 0 */
441 	if (udp_hdr->dgram_cksum != 0) {
442 		udp_hdr->dgram_cksum = 0;
443 		if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
444 			udp_hdr->dgram_cksum =
445 				rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
446 		else
447 			udp_hdr->dgram_cksum =
448 				rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
449 	}
450 
451 	return ol_flags;
452 }
453 
454 /*
455  * Helper function.
456  * Performs actual copying.
457  * Returns number of segments in the destination mbuf on success,
458  * or negative error code on failure.
459  */
460 static int
461 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
462 	uint16_t seglen[], uint8_t nb_seg)
463 {
464 	uint32_t dlen, slen, tlen;
465 	uint32_t i, len;
466 	const struct rte_mbuf *m;
467 	const uint8_t *src;
468 	uint8_t *dst;
469 
470 	dlen = 0;
471 	slen = 0;
472 	tlen = 0;
473 
474 	dst = NULL;
475 	src = NULL;
476 
477 	m = ms;
478 	i = 0;
479 	while (ms != NULL && i != nb_seg) {
480 
481 		if (slen == 0) {
482 			slen = rte_pktmbuf_data_len(ms);
483 			src = rte_pktmbuf_mtod(ms, const uint8_t *);
484 		}
485 
486 		if (dlen == 0) {
487 			dlen = RTE_MIN(seglen[i], slen);
488 			md[i]->data_len = dlen;
489 			md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
490 			dst = rte_pktmbuf_mtod(md[i], uint8_t *);
491 		}
492 
493 		len = RTE_MIN(slen, dlen);
494 		memcpy(dst, src, len);
495 		tlen += len;
496 		slen -= len;
497 		dlen -= len;
498 		src += len;
499 		dst += len;
500 
501 		if (slen == 0)
502 			ms = ms->next;
503 		if (dlen == 0)
504 			i++;
505 	}
506 
507 	if (ms != NULL)
508 		return -ENOBUFS;
509 	else if (tlen != m->pkt_len)
510 		return -EINVAL;
511 
512 	md[0]->nb_segs = nb_seg;
513 	md[0]->pkt_len = tlen;
514 	md[0]->vlan_tci = m->vlan_tci;
515 	md[0]->vlan_tci_outer = m->vlan_tci_outer;
516 	md[0]->ol_flags = m->ol_flags;
517 	md[0]->tx_offload = m->tx_offload;
518 
519 	return nb_seg;
520 }
521 
522 /*
523  * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
524  * Copy packet contents and offload information into then new segmented mbuf.
525  */
526 static struct rte_mbuf *
527 pkt_copy_split(const struct rte_mbuf *pkt)
528 {
529 	int32_t n, rc;
530 	uint32_t i, len, nb_seg;
531 	struct rte_mempool *mp;
532 	uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
533 	struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
534 
535 	mp = current_fwd_lcore()->mbp;
536 
537 	if (tx_pkt_split == TX_PKT_SPLIT_RND)
538 		nb_seg = random() % tx_pkt_nb_segs + 1;
539 	else
540 		nb_seg = tx_pkt_nb_segs;
541 
542 	memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
543 
544 	/* calculate number of segments to use and their length. */
545 	len = 0;
546 	for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
547 		len += seglen[i];
548 		md[i] = NULL;
549 	}
550 
551 	n = pkt->pkt_len - len;
552 
553 	/* update size of the last segment to fit rest of the packet */
554 	if (n >= 0) {
555 		seglen[i - 1] += n;
556 		len += n;
557 	}
558 
559 	nb_seg = i;
560 	while (i != 0) {
561 		p = rte_pktmbuf_alloc(mp);
562 		if (p == NULL) {
563 			RTE_LOG(ERR, USER1,
564 				"failed to allocate %u-th of %u mbuf "
565 				"from mempool: %s\n",
566 				nb_seg - i, nb_seg, mp->name);
567 			break;
568 		}
569 
570 		md[--i] = p;
571 		if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
572 			RTE_LOG(ERR, USER1, "mempool %s, %u-th segment: "
573 				"expected seglen: %u, "
574 				"actual mbuf tailroom: %u\n",
575 				mp->name, i, seglen[i],
576 				rte_pktmbuf_tailroom(md[i]));
577 			break;
578 		}
579 	}
580 
581 	/* all mbufs successfully allocated, do copy */
582 	if (i == 0) {
583 		rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
584 		if (rc < 0)
585 			RTE_LOG(ERR, USER1,
586 				"mbuf_copy_split for %p(len=%u, nb_seg=%u) "
587 				"into %u segments failed with error code: %d\n",
588 				pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
589 
590 		/* figure out how many mbufs to free. */
591 		i = RTE_MAX(rc, 0);
592 	}
593 
594 	/* free unused mbufs */
595 	for (; i != nb_seg; i++) {
596 		rte_pktmbuf_free_seg(md[i]);
597 		md[i] = NULL;
598 	}
599 
600 	return md[0];
601 }
602 
603 /*
604  * Receive a burst of packets, and for each packet:
605  *  - parse packet, and try to recognize a supported packet type (1)
606  *  - if it's not a supported packet type, don't touch the packet, else:
607  *  - reprocess the checksum of all supported layers. This is done in SW
608  *    or HW, depending on testpmd command line configuration
609  *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
610  *    segmentation offload (this implies HW TCP checksum)
611  * Then transmit packets on the output port.
612  *
613  * (1) Supported packets are:
614  *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
615  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
616  *           UDP|TCP|SCTP
617  *   Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
618  *   Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
619  *   Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
620  *
621  * The testpmd command line for this forward engine sets the flags
622  * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
623  * wether a checksum must be calculated in software or in hardware. The
624  * IP, UDP, TCP and SCTP flags always concern the inner layer. The
625  * OUTER_IP is only useful for tunnel packets.
626  */
627 static void
628 pkt_burst_checksum_forward(struct fwd_stream *fs)
629 {
630 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
631 	struct rte_port *txp;
632 	struct rte_mbuf *m, *p;
633 	struct ether_hdr *eth_hdr;
634 	void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
635 	uint16_t nb_rx;
636 	uint16_t nb_tx;
637 	uint16_t nb_prep;
638 	uint16_t i;
639 	uint64_t rx_ol_flags, tx_ol_flags;
640 	uint16_t testpmd_ol_flags;
641 	uint32_t retry;
642 	uint32_t rx_bad_ip_csum;
643 	uint32_t rx_bad_l4_csum;
644 	struct testpmd_offload_info info;
645 
646 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
647 	uint64_t start_tsc;
648 	uint64_t end_tsc;
649 	uint64_t core_cycles;
650 #endif
651 
652 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
653 	start_tsc = rte_rdtsc();
654 #endif
655 
656 	/* receive a burst of packet */
657 	nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
658 				 nb_pkt_per_burst);
659 	if (unlikely(nb_rx == 0))
660 		return;
661 
662 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
663 	fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
664 #endif
665 	fs->rx_packets += nb_rx;
666 	rx_bad_ip_csum = 0;
667 	rx_bad_l4_csum = 0;
668 
669 	txp = &ports[fs->tx_port];
670 	testpmd_ol_flags = txp->tx_ol_flags;
671 	memset(&info, 0, sizeof(info));
672 	info.tso_segsz = txp->tso_segsz;
673 	info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
674 
675 	for (i = 0; i < nb_rx; i++) {
676 		if (likely(i < nb_rx - 1))
677 			rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
678 						       void *));
679 
680 		m = pkts_burst[i];
681 		info.is_tunnel = 0;
682 		info.pkt_len = rte_pktmbuf_pkt_len(m);
683 		tx_ol_flags = 0;
684 		rx_ol_flags = m->ol_flags;
685 
686 		/* Update the L3/L4 checksum error packet statistics */
687 		if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
688 			rx_bad_ip_csum += 1;
689 		if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
690 			rx_bad_l4_csum += 1;
691 
692 		/* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
693 		 * and inner headers */
694 
695 		eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
696 		ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
697 				&eth_hdr->d_addr);
698 		ether_addr_copy(&ports[fs->tx_port].eth_addr,
699 				&eth_hdr->s_addr);
700 		parse_ethernet(eth_hdr, &info);
701 		l3_hdr = (char *)eth_hdr + info.l2_len;
702 
703 		/* check if it's a supported tunnel */
704 		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_PARSE_TUNNEL) {
705 			if (info.l4_proto == IPPROTO_UDP) {
706 				struct udp_hdr *udp_hdr;
707 
708 				udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
709 					info.l3_len);
710 				parse_vxlan(udp_hdr, &info, m->packet_type);
711 				if (info.is_tunnel)
712 					tx_ol_flags |= PKT_TX_TUNNEL_VXLAN;
713 			} else if (info.l4_proto == IPPROTO_GRE) {
714 				struct simple_gre_hdr *gre_hdr;
715 
716 				gre_hdr = (struct simple_gre_hdr *)
717 					((char *)l3_hdr + info.l3_len);
718 				parse_gre(gre_hdr, &info);
719 				if (info.is_tunnel)
720 					tx_ol_flags |= PKT_TX_TUNNEL_GRE;
721 			} else if (info.l4_proto == IPPROTO_IPIP) {
722 				void *encap_ip_hdr;
723 
724 				encap_ip_hdr = (char *)l3_hdr + info.l3_len;
725 				parse_encap_ip(encap_ip_hdr, &info);
726 				if (info.is_tunnel)
727 					tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
728 			}
729 		}
730 
731 		/* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
732 		if (info.is_tunnel) {
733 			outer_l3_hdr = l3_hdr;
734 			l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
735 		}
736 
737 		/* step 2: depending on user command line configuration,
738 		 * recompute checksum either in software or flag the
739 		 * mbuf to offload the calculation to the NIC. If TSO
740 		 * is configured, prepare the mbuf for TCP segmentation. */
741 
742 		/* process checksums of inner headers first */
743 		tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
744 			testpmd_ol_flags);
745 
746 		/* Then process outer headers if any. Note that the software
747 		 * checksum will be wrong if one of the inner checksums is
748 		 * processed in hardware. */
749 		if (info.is_tunnel == 1) {
750 			tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
751 					testpmd_ol_flags,
752 					!!(tx_ol_flags & PKT_TX_TCP_SEG));
753 		}
754 
755 		/* step 3: fill the mbuf meta data (flags and header lengths) */
756 
757 		if (info.is_tunnel == 1) {
758 			if (info.tunnel_tso_segsz ||
759 			    (testpmd_ol_flags &
760 			    TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) ||
761 			    (tx_ol_flags & PKT_TX_OUTER_IPV6)) {
762 				m->outer_l2_len = info.outer_l2_len;
763 				m->outer_l3_len = info.outer_l3_len;
764 				m->l2_len = info.l2_len;
765 				m->l3_len = info.l3_len;
766 				m->l4_len = info.l4_len;
767 				m->tso_segsz = info.tunnel_tso_segsz;
768 			}
769 			else {
770 				/* if there is a outer UDP cksum
771 				   processed in sw and the inner in hw,
772 				   the outer checksum will be wrong as
773 				   the payload will be modified by the
774 				   hardware */
775 				m->l2_len = info.outer_l2_len +
776 					info.outer_l3_len + info.l2_len;
777 				m->l3_len = info.l3_len;
778 				m->l4_len = info.l4_len;
779 			}
780 		} else {
781 			/* this is only useful if an offload flag is
782 			 * set, but it does not hurt to fill it in any
783 			 * case */
784 			m->l2_len = info.l2_len;
785 			m->l3_len = info.l3_len;
786 			m->l4_len = info.l4_len;
787 			m->tso_segsz = info.tso_segsz;
788 		}
789 		m->ol_flags = tx_ol_flags;
790 
791 		/* Do split & copy for the packet. */
792 		if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
793 			p = pkt_copy_split(m);
794 			if (p != NULL) {
795 				rte_pktmbuf_free(m);
796 				m = p;
797 				pkts_burst[i] = m;
798 			}
799 		}
800 
801 		/* if verbose mode is enabled, dump debug info */
802 		if (verbose_level > 0) {
803 			char buf[256];
804 
805 			printf("-----------------\n");
806 			printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
807 				fs->rx_port, m, m->pkt_len, m->nb_segs);
808 			/* dump rx parsed packet info */
809 			rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
810 			printf("rx: l2_len=%d ethertype=%x l3_len=%d "
811 				"l4_proto=%d l4_len=%d flags=%s\n",
812 				info.l2_len, rte_be_to_cpu_16(info.ethertype),
813 				info.l3_len, info.l4_proto, info.l4_len, buf);
814 			if (rx_ol_flags & PKT_RX_LRO)
815 				printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
816 			if (info.is_tunnel == 1)
817 				printf("rx: outer_l2_len=%d outer_ethertype=%x "
818 					"outer_l3_len=%d\n", info.outer_l2_len,
819 					rte_be_to_cpu_16(info.outer_ethertype),
820 					info.outer_l3_len);
821 			/* dump tx packet info */
822 			if ((testpmd_ol_flags & (TESTPMD_TX_OFFLOAD_IP_CKSUM |
823 						TESTPMD_TX_OFFLOAD_UDP_CKSUM |
824 						TESTPMD_TX_OFFLOAD_TCP_CKSUM |
825 						TESTPMD_TX_OFFLOAD_SCTP_CKSUM)) ||
826 				info.tso_segsz != 0)
827 				printf("tx: m->l2_len=%d m->l3_len=%d "
828 					"m->l4_len=%d\n",
829 					m->l2_len, m->l3_len, m->l4_len);
830 			if (info.is_tunnel == 1) {
831 				if ((testpmd_ol_flags &
832 				    TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) ||
833 				    (tx_ol_flags & PKT_TX_OUTER_IPV6))
834 					printf("tx: m->outer_l2_len=%d "
835 						"m->outer_l3_len=%d\n",
836 						m->outer_l2_len,
837 						m->outer_l3_len);
838 				if (info.tunnel_tso_segsz != 0 &&
839 						(m->ol_flags & PKT_TX_TCP_SEG))
840 					printf("tx: m->tso_segsz=%d\n",
841 						m->tso_segsz);
842 			} else if (info.tso_segsz != 0 &&
843 					(m->ol_flags & PKT_TX_TCP_SEG))
844 				printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
845 			rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
846 			printf("tx: flags=%s", buf);
847 			printf("\n");
848 		}
849 	}
850 
851 	nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
852 			pkts_burst, nb_rx);
853 	if (nb_prep != nb_rx)
854 		printf("Preparing packet burst to transmit failed: %s\n",
855 				rte_strerror(rte_errno));
856 
857 	nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst,
858 			nb_prep);
859 
860 	/*
861 	 * Retry if necessary
862 	 */
863 	if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
864 		retry = 0;
865 		while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
866 			rte_delay_us(burst_tx_delay_time);
867 			nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
868 					&pkts_burst[nb_tx], nb_rx - nb_tx);
869 		}
870 	}
871 	fs->tx_packets += nb_tx;
872 	fs->rx_bad_ip_csum += rx_bad_ip_csum;
873 	fs->rx_bad_l4_csum += rx_bad_l4_csum;
874 
875 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
876 	fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
877 #endif
878 	if (unlikely(nb_tx < nb_rx)) {
879 		fs->fwd_dropped += (nb_rx - nb_tx);
880 		do {
881 			rte_pktmbuf_free(pkts_burst[nb_tx]);
882 		} while (++nb_tx < nb_rx);
883 	}
884 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
885 	end_tsc = rte_rdtsc();
886 	core_cycles = (end_tsc - start_tsc);
887 	fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
888 #endif
889 }
890 
891 struct fwd_engine csum_fwd_engine = {
892 	.fwd_mode_name  = "csum",
893 	.port_fwd_begin = NULL,
894 	.port_fwd_end   = NULL,
895 	.packet_fwd     = pkt_burst_checksum_forward,
896 };
897