xref: /dpdk/app/test-pmd/csumonly.c (revision a997a33b2a0145ad3e6320ea1fc7df8d51a2fcdf)
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_psd_sum(void *l3_hdr, uint16_t ethertype, uint64_t ol_flags)
117 {
118 	if (ethertype == _htons(ETHER_TYPE_IPv4))
119 		return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
120 	else /* assume ethertype == ETHER_TYPE_IPv6 */
121 		return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
122 }
123 
124 static uint16_t
125 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
126 {
127 	if (ethertype == _htons(ETHER_TYPE_IPv4))
128 		return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
129 	else /* assume ethertype == ETHER_TYPE_IPv6 */
130 		return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
131 }
132 
133 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
134 static void
135 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
136 {
137 	struct tcp_hdr *tcp_hdr;
138 
139 	info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
140 	info->l4_proto = ipv4_hdr->next_proto_id;
141 
142 	/* only fill l4_len for TCP, it's useful for TSO */
143 	if (info->l4_proto == IPPROTO_TCP) {
144 		tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
145 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
146 	} else
147 		info->l4_len = 0;
148 }
149 
150 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
151 static void
152 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
153 {
154 	struct tcp_hdr *tcp_hdr;
155 
156 	info->l3_len = sizeof(struct ipv6_hdr);
157 	info->l4_proto = ipv6_hdr->proto;
158 
159 	/* only fill l4_len for TCP, it's useful for TSO */
160 	if (info->l4_proto == IPPROTO_TCP) {
161 		tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
162 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
163 	} else
164 		info->l4_len = 0;
165 }
166 
167 /*
168  * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
169  * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
170  * header. The l4_len argument is only set in case of TCP (useful for TSO).
171  */
172 static void
173 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info)
174 {
175 	struct ipv4_hdr *ipv4_hdr;
176 	struct ipv6_hdr *ipv6_hdr;
177 
178 	info->l2_len = sizeof(struct ether_hdr);
179 	info->ethertype = eth_hdr->ether_type;
180 
181 	if (info->ethertype == _htons(ETHER_TYPE_VLAN)) {
182 		struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
183 
184 		info->l2_len  += sizeof(struct vlan_hdr);
185 		info->ethertype = vlan_hdr->eth_proto;
186 	}
187 
188 	switch (info->ethertype) {
189 	case _htons(ETHER_TYPE_IPv4):
190 		ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
191 		parse_ipv4(ipv4_hdr, info);
192 		break;
193 	case _htons(ETHER_TYPE_IPv6):
194 		ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
195 		parse_ipv6(ipv6_hdr, info);
196 		break;
197 	default:
198 		info->l4_len = 0;
199 		info->l3_len = 0;
200 		info->l4_proto = 0;
201 		break;
202 	}
203 }
204 
205 /* Parse a vxlan header */
206 static void
207 parse_vxlan(struct udp_hdr *udp_hdr,
208 	    struct testpmd_offload_info *info,
209 	    uint32_t pkt_type)
210 {
211 	struct ether_hdr *eth_hdr;
212 
213 	/* check udp destination port, 4789 is the default vxlan port
214 	 * (rfc7348) or that the rx offload flag is set (i40e only
215 	 * currently) */
216 	if (udp_hdr->dst_port != _htons(4789) &&
217 		RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
218 		return;
219 
220 	info->is_tunnel = 1;
221 	info->outer_ethertype = info->ethertype;
222 	info->outer_l2_len = info->l2_len;
223 	info->outer_l3_len = info->l3_len;
224 	info->outer_l4_proto = info->l4_proto;
225 
226 	eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
227 		sizeof(struct udp_hdr) +
228 		sizeof(struct vxlan_hdr));
229 
230 	parse_ethernet(eth_hdr, info);
231 	info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */
232 }
233 
234 /* Parse a gre header */
235 static void
236 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
237 {
238 	struct ether_hdr *eth_hdr;
239 	struct ipv4_hdr *ipv4_hdr;
240 	struct ipv6_hdr *ipv6_hdr;
241 	uint8_t gre_len = 0;
242 
243 	/* check which fields are supported */
244 	if ((gre_hdr->flags & _htons(~GRE_SUPPORTED_FIELDS)) != 0)
245 		return;
246 
247 	gre_len += sizeof(struct simple_gre_hdr);
248 
249 	if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
250 		gre_len += GRE_KEY_LEN;
251 
252 	if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) {
253 		info->is_tunnel = 1;
254 		info->outer_ethertype = info->ethertype;
255 		info->outer_l2_len = info->l2_len;
256 		info->outer_l3_len = info->l3_len;
257 		info->outer_l4_proto = info->l4_proto;
258 
259 		ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len);
260 
261 		parse_ipv4(ipv4_hdr, info);
262 		info->ethertype = _htons(ETHER_TYPE_IPv4);
263 		info->l2_len = 0;
264 
265 	} else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) {
266 		info->is_tunnel = 1;
267 		info->outer_ethertype = info->ethertype;
268 		info->outer_l2_len = info->l2_len;
269 		info->outer_l3_len = info->l3_len;
270 		info->outer_l4_proto = info->l4_proto;
271 
272 		ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len);
273 
274 		info->ethertype = _htons(ETHER_TYPE_IPv6);
275 		parse_ipv6(ipv6_hdr, info);
276 		info->l2_len = 0;
277 
278 	} else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) {
279 		info->is_tunnel = 1;
280 		info->outer_ethertype = info->ethertype;
281 		info->outer_l2_len = info->l2_len;
282 		info->outer_l3_len = info->l3_len;
283 		info->outer_l4_proto = info->l4_proto;
284 
285 		eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len);
286 
287 		parse_ethernet(eth_hdr, info);
288 	} else
289 		return;
290 
291 	info->l2_len += gre_len;
292 }
293 
294 
295 /* Parse an encapsulated ip or ipv6 header */
296 static void
297 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
298 {
299 	struct ipv4_hdr *ipv4_hdr = encap_ip;
300 	struct ipv6_hdr *ipv6_hdr = encap_ip;
301 	uint8_t ip_version;
302 
303 	ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
304 
305 	if (ip_version != 4 && ip_version != 6)
306 		return;
307 
308 	info->is_tunnel = 1;
309 	info->outer_ethertype = info->ethertype;
310 	info->outer_l2_len = info->l2_len;
311 	info->outer_l3_len = info->l3_len;
312 
313 	if (ip_version == 4) {
314 		parse_ipv4(ipv4_hdr, info);
315 		info->ethertype = _htons(ETHER_TYPE_IPv4);
316 	} else {
317 		parse_ipv6(ipv6_hdr, info);
318 		info->ethertype = _htons(ETHER_TYPE_IPv6);
319 	}
320 	info->l2_len = 0;
321 }
322 
323 /* if possible, calculate the checksum of a packet in hw or sw,
324  * depending on the testpmd command line configuration */
325 static uint64_t
326 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
327 	uint16_t testpmd_ol_flags)
328 {
329 	struct ipv4_hdr *ipv4_hdr = l3_hdr;
330 	struct udp_hdr *udp_hdr;
331 	struct tcp_hdr *tcp_hdr;
332 	struct sctp_hdr *sctp_hdr;
333 	uint64_t ol_flags = 0;
334 	uint32_t max_pkt_len, tso_segsz = 0;
335 
336 	/* ensure packet is large enough to require tso */
337 	if (!info->is_tunnel) {
338 		max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
339 			info->tso_segsz;
340 		if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
341 			tso_segsz = info->tso_segsz;
342 	} else {
343 		max_pkt_len = info->outer_l2_len + info->outer_l3_len +
344 			info->l2_len + info->l3_len + info->l4_len +
345 			info->tunnel_tso_segsz;
346 		if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
347 			tso_segsz = info->tunnel_tso_segsz;
348 	}
349 
350 	if (info->ethertype == _htons(ETHER_TYPE_IPv4)) {
351 		ipv4_hdr = l3_hdr;
352 		ipv4_hdr->hdr_checksum = 0;
353 
354 		ol_flags |= PKT_TX_IPV4;
355 		if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
356 			ol_flags |= PKT_TX_IP_CKSUM;
357 		} else {
358 			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
359 				ol_flags |= PKT_TX_IP_CKSUM;
360 			else
361 				ipv4_hdr->hdr_checksum =
362 					rte_ipv4_cksum(ipv4_hdr);
363 		}
364 	} else if (info->ethertype == _htons(ETHER_TYPE_IPv6))
365 		ol_flags |= PKT_TX_IPV6;
366 	else
367 		return 0; /* packet type not supported, nothing to do */
368 
369 	if (info->l4_proto == IPPROTO_UDP) {
370 		udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
371 		/* do not recalculate udp cksum if it was 0 */
372 		if (udp_hdr->dgram_cksum != 0) {
373 			udp_hdr->dgram_cksum = 0;
374 			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM) {
375 				ol_flags |= PKT_TX_UDP_CKSUM;
376 				udp_hdr->dgram_cksum = get_psd_sum(l3_hdr,
377 					info->ethertype, ol_flags);
378 			} else {
379 				udp_hdr->dgram_cksum =
380 					get_udptcp_checksum(l3_hdr, udp_hdr,
381 						info->ethertype);
382 			}
383 		}
384 	} else if (info->l4_proto == IPPROTO_TCP) {
385 		tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
386 		tcp_hdr->cksum = 0;
387 		if (tso_segsz) {
388 			ol_flags |= PKT_TX_TCP_SEG;
389 			tcp_hdr->cksum = get_psd_sum(l3_hdr, info->ethertype,
390 				ol_flags);
391 		} else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
392 			ol_flags |= PKT_TX_TCP_CKSUM;
393 			tcp_hdr->cksum = get_psd_sum(l3_hdr, info->ethertype,
394 				ol_flags);
395 		} else {
396 			tcp_hdr->cksum =
397 				get_udptcp_checksum(l3_hdr, tcp_hdr,
398 					info->ethertype);
399 		}
400 	} else if (info->l4_proto == IPPROTO_SCTP) {
401 		sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
402 		sctp_hdr->cksum = 0;
403 		/* sctp payload must be a multiple of 4 to be
404 		 * offloaded */
405 		if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_SCTP_CKSUM) &&
406 			((ipv4_hdr->total_length & 0x3) == 0)) {
407 			ol_flags |= PKT_TX_SCTP_CKSUM;
408 		} else {
409 			/* XXX implement CRC32c, example available in
410 			 * RFC3309 */
411 		}
412 	}
413 
414 	return ol_flags;
415 }
416 
417 /* Calculate the checksum of outer header */
418 static uint64_t
419 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
420 	uint16_t testpmd_ol_flags, int tso_enabled)
421 {
422 	struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
423 	struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
424 	struct udp_hdr *udp_hdr;
425 	uint64_t ol_flags = 0;
426 
427 	if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
428 		ipv4_hdr->hdr_checksum = 0;
429 		ol_flags |= PKT_TX_OUTER_IPV4;
430 
431 		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
432 			ol_flags |= PKT_TX_OUTER_IP_CKSUM;
433 		else
434 			ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
435 	} else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
436 		ol_flags |= PKT_TX_OUTER_IPV6;
437 
438 	if (info->outer_l4_proto != IPPROTO_UDP)
439 		return ol_flags;
440 
441 	udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
442 
443 	/* outer UDP checksum is done in software as we have no hardware
444 	 * supporting it today, and no API for it. In the other side, for
445 	 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
446 	 * set to zero.
447 	 *
448 	 * If a packet will be TSOed into small packets by NIC, we cannot
449 	 * set/calculate a non-zero checksum, because it will be a wrong
450 	 * value after the packet be split into several small packets.
451 	 */
452 	if (tso_enabled)
453 		udp_hdr->dgram_cksum = 0;
454 
455 	/* do not recalculate udp cksum if it was 0 */
456 	if (udp_hdr->dgram_cksum != 0) {
457 		udp_hdr->dgram_cksum = 0;
458 		if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
459 			udp_hdr->dgram_cksum =
460 				rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
461 		else
462 			udp_hdr->dgram_cksum =
463 				rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
464 	}
465 
466 	return ol_flags;
467 }
468 
469 /*
470  * Helper function.
471  * Performs actual copying.
472  * Returns number of segments in the destination mbuf on success,
473  * or negative error code on failure.
474  */
475 static int
476 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
477 	uint16_t seglen[], uint8_t nb_seg)
478 {
479 	uint32_t dlen, slen, tlen;
480 	uint32_t i, len;
481 	const struct rte_mbuf *m;
482 	const uint8_t *src;
483 	uint8_t *dst;
484 
485 	dlen = 0;
486 	slen = 0;
487 	tlen = 0;
488 
489 	dst = NULL;
490 	src = NULL;
491 
492 	m = ms;
493 	i = 0;
494 	while (ms != NULL && i != nb_seg) {
495 
496 		if (slen == 0) {
497 			slen = rte_pktmbuf_data_len(ms);
498 			src = rte_pktmbuf_mtod(ms, const uint8_t *);
499 		}
500 
501 		if (dlen == 0) {
502 			dlen = RTE_MIN(seglen[i], slen);
503 			md[i]->data_len = dlen;
504 			md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
505 			dst = rte_pktmbuf_mtod(md[i], uint8_t *);
506 		}
507 
508 		len = RTE_MIN(slen, dlen);
509 		memcpy(dst, src, len);
510 		tlen += len;
511 		slen -= len;
512 		dlen -= len;
513 		src += len;
514 		dst += len;
515 
516 		if (slen == 0)
517 			ms = ms->next;
518 		if (dlen == 0)
519 			i++;
520 	}
521 
522 	if (ms != NULL)
523 		return -ENOBUFS;
524 	else if (tlen != m->pkt_len)
525 		return -EINVAL;
526 
527 	md[0]->nb_segs = nb_seg;
528 	md[0]->pkt_len = tlen;
529 	md[0]->vlan_tci = m->vlan_tci;
530 	md[0]->vlan_tci_outer = m->vlan_tci_outer;
531 	md[0]->ol_flags = m->ol_flags;
532 	md[0]->tx_offload = m->tx_offload;
533 
534 	return nb_seg;
535 }
536 
537 /*
538  * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
539  * Copy packet contents and offload information into then new segmented mbuf.
540  */
541 static struct rte_mbuf *
542 pkt_copy_split(const struct rte_mbuf *pkt)
543 {
544 	int32_t n, rc;
545 	uint32_t i, len, nb_seg;
546 	struct rte_mempool *mp;
547 	uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
548 	struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
549 
550 	mp = current_fwd_lcore()->mbp;
551 
552 	if (tx_pkt_split == TX_PKT_SPLIT_RND)
553 		nb_seg = random() % tx_pkt_nb_segs + 1;
554 	else
555 		nb_seg = tx_pkt_nb_segs;
556 
557 	memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
558 
559 	/* calculate number of segments to use and their length. */
560 	len = 0;
561 	for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
562 		len += seglen[i];
563 		md[i] = NULL;
564 	}
565 
566 	n = pkt->pkt_len - len;
567 
568 	/* update size of the last segment to fit rest of the packet */
569 	if (n >= 0) {
570 		seglen[i - 1] += n;
571 		len += n;
572 	}
573 
574 	nb_seg = i;
575 	while (i != 0) {
576 		p = rte_pktmbuf_alloc(mp);
577 		if (p == NULL) {
578 			RTE_LOG(ERR, USER1,
579 				"failed to allocate %u-th of %u mbuf "
580 				"from mempool: %s\n",
581 				nb_seg - i, nb_seg, mp->name);
582 			break;
583 		}
584 
585 		md[--i] = p;
586 		if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
587 			RTE_LOG(ERR, USER1, "mempool %s, %u-th segment: "
588 				"expected seglen: %u, "
589 				"actual mbuf tailroom: %u\n",
590 				mp->name, i, seglen[i],
591 				rte_pktmbuf_tailroom(md[i]));
592 			break;
593 		}
594 	}
595 
596 	/* all mbufs successfully allocated, do copy */
597 	if (i == 0) {
598 		rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
599 		if (rc < 0)
600 			RTE_LOG(ERR, USER1,
601 				"mbuf_copy_split for %p(len=%u, nb_seg=%hhu) "
602 				"into %u segments failed with error code: %d\n",
603 				pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
604 
605 		/* figure out how many mbufs to free. */
606 		i = RTE_MAX(rc, 0);
607 	}
608 
609 	/* free unused mbufs */
610 	for (; i != nb_seg; i++) {
611 		rte_pktmbuf_free_seg(md[i]);
612 		md[i] = NULL;
613 	}
614 
615 	return md[0];
616 }
617 
618 /*
619  * Receive a burst of packets, and for each packet:
620  *  - parse packet, and try to recognize a supported packet type (1)
621  *  - if it's not a supported packet type, don't touch the packet, else:
622  *  - reprocess the checksum of all supported layers. This is done in SW
623  *    or HW, depending on testpmd command line configuration
624  *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
625  *    segmentation offload (this implies HW TCP checksum)
626  * Then transmit packets on the output port.
627  *
628  * (1) Supported packets are:
629  *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
630  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
631  *           UDP|TCP|SCTP
632  *   Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
633  *   Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
634  *   Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
635  *
636  * The testpmd command line for this forward engine sets the flags
637  * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
638  * wether a checksum must be calculated in software or in hardware. The
639  * IP, UDP, TCP and SCTP flags always concern the inner layer. The
640  * OUTER_IP is only useful for tunnel packets.
641  */
642 static void
643 pkt_burst_checksum_forward(struct fwd_stream *fs)
644 {
645 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
646 	struct rte_port *txp;
647 	struct rte_mbuf *m, *p;
648 	struct ether_hdr *eth_hdr;
649 	void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
650 	uint16_t nb_rx;
651 	uint16_t nb_tx;
652 	uint16_t i;
653 	uint64_t rx_ol_flags, tx_ol_flags;
654 	uint16_t testpmd_ol_flags;
655 	uint32_t retry;
656 	uint32_t rx_bad_ip_csum;
657 	uint32_t rx_bad_l4_csum;
658 	struct testpmd_offload_info info;
659 
660 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
661 	uint64_t start_tsc;
662 	uint64_t end_tsc;
663 	uint64_t core_cycles;
664 #endif
665 
666 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
667 	start_tsc = rte_rdtsc();
668 #endif
669 
670 	/* receive a burst of packet */
671 	nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
672 				 nb_pkt_per_burst);
673 	if (unlikely(nb_rx == 0))
674 		return;
675 
676 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
677 	fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
678 #endif
679 	fs->rx_packets += nb_rx;
680 	rx_bad_ip_csum = 0;
681 	rx_bad_l4_csum = 0;
682 
683 	txp = &ports[fs->tx_port];
684 	testpmd_ol_flags = txp->tx_ol_flags;
685 	memset(&info, 0, sizeof(info));
686 	info.tso_segsz = txp->tso_segsz;
687 	info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
688 
689 	for (i = 0; i < nb_rx; i++) {
690 		if (likely(i < nb_rx - 1))
691 			rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
692 						       void *));
693 
694 		m = pkts_burst[i];
695 		info.is_tunnel = 0;
696 		info.pkt_len = rte_pktmbuf_pkt_len(m);
697 		tx_ol_flags = 0;
698 		rx_ol_flags = m->ol_flags;
699 
700 		/* Update the L3/L4 checksum error packet statistics */
701 		if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
702 			rx_bad_ip_csum += 1;
703 		if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
704 			rx_bad_l4_csum += 1;
705 
706 		/* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
707 		 * and inner headers */
708 
709 		eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
710 		ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
711 				&eth_hdr->d_addr);
712 		ether_addr_copy(&ports[fs->tx_port].eth_addr,
713 				&eth_hdr->s_addr);
714 		parse_ethernet(eth_hdr, &info);
715 		l3_hdr = (char *)eth_hdr + info.l2_len;
716 
717 		/* check if it's a supported tunnel */
718 		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_PARSE_TUNNEL) {
719 			if (info.l4_proto == IPPROTO_UDP) {
720 				struct udp_hdr *udp_hdr;
721 
722 				udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
723 					info.l3_len);
724 				parse_vxlan(udp_hdr, &info, m->packet_type);
725 				if (info.is_tunnel)
726 					tx_ol_flags |= PKT_TX_TUNNEL_VXLAN;
727 			} else if (info.l4_proto == IPPROTO_GRE) {
728 				struct simple_gre_hdr *gre_hdr;
729 
730 				gre_hdr = (struct simple_gre_hdr *)
731 					((char *)l3_hdr + info.l3_len);
732 				parse_gre(gre_hdr, &info);
733 				if (info.is_tunnel)
734 					tx_ol_flags |= PKT_TX_TUNNEL_GRE;
735 			} else if (info.l4_proto == IPPROTO_IPIP) {
736 				void *encap_ip_hdr;
737 
738 				encap_ip_hdr = (char *)l3_hdr + info.l3_len;
739 				parse_encap_ip(encap_ip_hdr, &info);
740 				if (info.is_tunnel)
741 					tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
742 			}
743 		}
744 
745 		/* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
746 		if (info.is_tunnel) {
747 			outer_l3_hdr = l3_hdr;
748 			l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
749 		}
750 
751 		/* step 2: depending on user command line configuration,
752 		 * recompute checksum either in software or flag the
753 		 * mbuf to offload the calculation to the NIC. If TSO
754 		 * is configured, prepare the mbuf for TCP segmentation. */
755 
756 		/* process checksums of inner headers first */
757 		tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
758 			testpmd_ol_flags);
759 
760 		/* Then process outer headers if any. Note that the software
761 		 * checksum will be wrong if one of the inner checksums is
762 		 * processed in hardware. */
763 		if (info.is_tunnel == 1) {
764 			tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
765 					testpmd_ol_flags,
766 					!!(tx_ol_flags & PKT_TX_TCP_SEG));
767 		}
768 
769 		/* step 3: fill the mbuf meta data (flags and header lengths) */
770 
771 		if (info.is_tunnel == 1) {
772 			if (info.tunnel_tso_segsz ||
773 			    testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) {
774 				m->outer_l2_len = info.outer_l2_len;
775 				m->outer_l3_len = info.outer_l3_len;
776 				m->l2_len = info.l2_len;
777 				m->l3_len = info.l3_len;
778 				m->l4_len = info.l4_len;
779 				m->tso_segsz = info.tunnel_tso_segsz;
780 			}
781 			else {
782 				/* if there is a outer UDP cksum
783 				   processed in sw and the inner in hw,
784 				   the outer checksum will be wrong as
785 				   the payload will be modified by the
786 				   hardware */
787 				m->l2_len = info.outer_l2_len +
788 					info.outer_l3_len + info.l2_len;
789 				m->l3_len = info.l3_len;
790 				m->l4_len = info.l4_len;
791 			}
792 		} else {
793 			/* this is only useful if an offload flag is
794 			 * set, but it does not hurt to fill it in any
795 			 * case */
796 			m->l2_len = info.l2_len;
797 			m->l3_len = info.l3_len;
798 			m->l4_len = info.l4_len;
799 			m->tso_segsz = info.tso_segsz;
800 		}
801 		m->ol_flags = tx_ol_flags;
802 
803 		/* Do split & copy for the packet. */
804 		if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
805 			p = pkt_copy_split(m);
806 			if (p != NULL) {
807 				rte_pktmbuf_free(m);
808 				m = p;
809 				pkts_burst[i] = m;
810 			}
811 		}
812 
813 		/* if verbose mode is enabled, dump debug info */
814 		if (verbose_level > 0) {
815 			char buf[256];
816 
817 			printf("-----------------\n");
818 			printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%hhu:\n",
819 				fs->rx_port, m, m->pkt_len, m->nb_segs);
820 			/* dump rx parsed packet info */
821 			rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
822 			printf("rx: l2_len=%d ethertype=%x l3_len=%d "
823 				"l4_proto=%d l4_len=%d flags=%s\n",
824 				info.l2_len, rte_be_to_cpu_16(info.ethertype),
825 				info.l3_len, info.l4_proto, info.l4_len, buf);
826 			if (rx_ol_flags & PKT_RX_LRO)
827 				printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
828 			if (info.is_tunnel == 1)
829 				printf("rx: outer_l2_len=%d outer_ethertype=%x "
830 					"outer_l3_len=%d\n", info.outer_l2_len,
831 					rte_be_to_cpu_16(info.outer_ethertype),
832 					info.outer_l3_len);
833 			/* dump tx packet info */
834 			if ((testpmd_ol_flags & (TESTPMD_TX_OFFLOAD_IP_CKSUM |
835 						TESTPMD_TX_OFFLOAD_UDP_CKSUM |
836 						TESTPMD_TX_OFFLOAD_TCP_CKSUM |
837 						TESTPMD_TX_OFFLOAD_SCTP_CKSUM)) ||
838 				info.tso_segsz != 0)
839 				printf("tx: m->l2_len=%d m->l3_len=%d "
840 					"m->l4_len=%d\n",
841 					m->l2_len, m->l3_len, m->l4_len);
842 			if (info.is_tunnel == 1) {
843 				if (testpmd_ol_flags &
844 				    TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM)
845 					printf("tx: m->outer_l2_len=%d "
846 						"m->outer_l3_len=%d\n",
847 						m->outer_l2_len,
848 						m->outer_l3_len);
849 				if (info.tunnel_tso_segsz != 0 &&
850 						(m->ol_flags & PKT_TX_TCP_SEG))
851 					printf("tx: m->tso_segsz=%d\n",
852 						m->tso_segsz);
853 			} else if (info.tso_segsz != 0 &&
854 					(m->ol_flags & PKT_TX_TCP_SEG))
855 				printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
856 			rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
857 			printf("tx: flags=%s", buf);
858 			printf("\n");
859 		}
860 	}
861 	nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
862 	/*
863 	 * Retry if necessary
864 	 */
865 	if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
866 		retry = 0;
867 		while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
868 			rte_delay_us(burst_tx_delay_time);
869 			nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
870 					&pkts_burst[nb_tx], nb_rx - nb_tx);
871 		}
872 	}
873 	fs->tx_packets += nb_tx;
874 	fs->rx_bad_ip_csum += rx_bad_ip_csum;
875 	fs->rx_bad_l4_csum += rx_bad_l4_csum;
876 
877 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
878 	fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
879 #endif
880 	if (unlikely(nb_tx < nb_rx)) {
881 		fs->fwd_dropped += (nb_rx - nb_tx);
882 		do {
883 			rte_pktmbuf_free(pkts_burst[nb_tx]);
884 		} while (++nb_tx < nb_rx);
885 	}
886 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
887 	end_tsc = rte_rdtsc();
888 	core_cycles = (end_tsc - start_tsc);
889 	fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
890 #endif
891 }
892 
893 struct fwd_engine csum_fwd_engine = {
894 	.fwd_mode_name  = "csum",
895 	.port_fwd_begin = NULL,
896 	.port_fwd_end   = NULL,
897 	.packet_fwd     = pkt_burst_checksum_forward,
898 };
899