xref: /dpdk/app/test-pmd/csumonly.c (revision 2f45703c17acb943aaded9f79676fd56a72542b2)
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 };
105 
106 /* simplified GRE header */
107 struct simple_gre_hdr {
108 	uint16_t flags;
109 	uint16_t proto;
110 } __attribute__((__packed__));
111 
112 static uint16_t
113 get_psd_sum(void *l3_hdr, uint16_t ethertype, uint64_t ol_flags)
114 {
115 	if (ethertype == _htons(ETHER_TYPE_IPv4))
116 		return rte_ipv4_phdr_cksum(l3_hdr, ol_flags);
117 	else /* assume ethertype == ETHER_TYPE_IPv6 */
118 		return rte_ipv6_phdr_cksum(l3_hdr, ol_flags);
119 }
120 
121 static uint16_t
122 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
123 {
124 	if (ethertype == _htons(ETHER_TYPE_IPv4))
125 		return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
126 	else /* assume ethertype == ETHER_TYPE_IPv6 */
127 		return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
128 }
129 
130 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
131 static void
132 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
133 {
134 	struct tcp_hdr *tcp_hdr;
135 
136 	info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
137 	info->l4_proto = ipv4_hdr->next_proto_id;
138 
139 	/* only fill l4_len for TCP, it's useful for TSO */
140 	if (info->l4_proto == IPPROTO_TCP) {
141 		tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
142 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
143 	} else
144 		info->l4_len = 0;
145 }
146 
147 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
148 static void
149 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
150 {
151 	struct tcp_hdr *tcp_hdr;
152 
153 	info->l3_len = sizeof(struct ipv6_hdr);
154 	info->l4_proto = ipv6_hdr->proto;
155 
156 	/* only fill l4_len for TCP, it's useful for TSO */
157 	if (info->l4_proto == IPPROTO_TCP) {
158 		tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
159 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
160 	} else
161 		info->l4_len = 0;
162 }
163 
164 /*
165  * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
166  * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
167  * header. The l4_len argument is only set in case of TCP (useful for TSO).
168  */
169 static void
170 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info)
171 {
172 	struct ipv4_hdr *ipv4_hdr;
173 	struct ipv6_hdr *ipv6_hdr;
174 
175 	info->l2_len = sizeof(struct ether_hdr);
176 	info->ethertype = eth_hdr->ether_type;
177 
178 	if (info->ethertype == _htons(ETHER_TYPE_VLAN)) {
179 		struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
180 
181 		info->l2_len  += sizeof(struct vlan_hdr);
182 		info->ethertype = vlan_hdr->eth_proto;
183 	}
184 
185 	switch (info->ethertype) {
186 	case _htons(ETHER_TYPE_IPv4):
187 		ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
188 		parse_ipv4(ipv4_hdr, info);
189 		break;
190 	case _htons(ETHER_TYPE_IPv6):
191 		ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
192 		parse_ipv6(ipv6_hdr, info);
193 		break;
194 	default:
195 		info->l4_len = 0;
196 		info->l3_len = 0;
197 		info->l4_proto = 0;
198 		break;
199 	}
200 }
201 
202 /* Parse a vxlan header */
203 static void
204 parse_vxlan(struct udp_hdr *udp_hdr,
205 	    struct testpmd_offload_info *info,
206 	    uint32_t pkt_type)
207 {
208 	struct ether_hdr *eth_hdr;
209 
210 	/* check udp destination port, 4789 is the default vxlan port
211 	 * (rfc7348) or that the rx offload flag is set (i40e only
212 	 * currently) */
213 	if (udp_hdr->dst_port != _htons(4789) &&
214 		RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
215 		return;
216 
217 	info->is_tunnel = 1;
218 	info->outer_ethertype = info->ethertype;
219 	info->outer_l2_len = info->l2_len;
220 	info->outer_l3_len = info->l3_len;
221 	info->outer_l4_proto = info->l4_proto;
222 
223 	eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
224 		sizeof(struct udp_hdr) +
225 		sizeof(struct vxlan_hdr));
226 
227 	parse_ethernet(eth_hdr, info);
228 	info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */
229 }
230 
231 /* Parse a gre header */
232 static void
233 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
234 {
235 	struct ether_hdr *eth_hdr;
236 	struct ipv4_hdr *ipv4_hdr;
237 	struct ipv6_hdr *ipv6_hdr;
238 	uint8_t gre_len = 0;
239 
240 	/* check which fields are supported */
241 	if ((gre_hdr->flags & _htons(~GRE_SUPPORTED_FIELDS)) != 0)
242 		return;
243 
244 	gre_len += sizeof(struct simple_gre_hdr);
245 
246 	if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
247 		gre_len += GRE_KEY_LEN;
248 
249 	if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) {
250 		info->is_tunnel = 1;
251 		info->outer_ethertype = info->ethertype;
252 		info->outer_l2_len = info->l2_len;
253 		info->outer_l3_len = info->l3_len;
254 		info->outer_l4_proto = info->l4_proto;
255 
256 		ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len);
257 
258 		parse_ipv4(ipv4_hdr, info);
259 		info->ethertype = _htons(ETHER_TYPE_IPv4);
260 		info->l2_len = 0;
261 
262 	} else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) {
263 		info->is_tunnel = 1;
264 		info->outer_ethertype = info->ethertype;
265 		info->outer_l2_len = info->l2_len;
266 		info->outer_l3_len = info->l3_len;
267 		info->outer_l4_proto = info->l4_proto;
268 
269 		ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len);
270 
271 		info->ethertype = _htons(ETHER_TYPE_IPv6);
272 		parse_ipv6(ipv6_hdr, info);
273 		info->l2_len = 0;
274 
275 	} else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) {
276 		info->is_tunnel = 1;
277 		info->outer_ethertype = info->ethertype;
278 		info->outer_l2_len = info->l2_len;
279 		info->outer_l3_len = info->l3_len;
280 		info->outer_l4_proto = info->l4_proto;
281 
282 		eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len);
283 
284 		parse_ethernet(eth_hdr, info);
285 	} else
286 		return;
287 
288 	info->l2_len += gre_len;
289 }
290 
291 
292 /* Parse an encapsulated ip or ipv6 header */
293 static void
294 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
295 {
296 	struct ipv4_hdr *ipv4_hdr = encap_ip;
297 	struct ipv6_hdr *ipv6_hdr = encap_ip;
298 	uint8_t ip_version;
299 
300 	ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
301 
302 	if (ip_version != 4 && ip_version != 6)
303 		return;
304 
305 	info->is_tunnel = 1;
306 	info->outer_ethertype = info->ethertype;
307 	info->outer_l2_len = info->l2_len;
308 	info->outer_l3_len = info->l3_len;
309 
310 	if (ip_version == 4) {
311 		parse_ipv4(ipv4_hdr, info);
312 		info->ethertype = _htons(ETHER_TYPE_IPv4);
313 	} else {
314 		parse_ipv6(ipv6_hdr, info);
315 		info->ethertype = _htons(ETHER_TYPE_IPv6);
316 	}
317 	info->l2_len = 0;
318 }
319 
320 /* modify the IPv4 or IPv4 source address of a packet */
321 static void
322 change_ip_addresses(void *l3_hdr, uint16_t ethertype)
323 {
324 	struct ipv4_hdr *ipv4_hdr = l3_hdr;
325 	struct ipv6_hdr *ipv6_hdr = l3_hdr;
326 
327 	if (ethertype == _htons(ETHER_TYPE_IPv4)) {
328 		ipv4_hdr->src_addr =
329 			rte_cpu_to_be_32(rte_be_to_cpu_32(ipv4_hdr->src_addr) + 1);
330 	} else if (ethertype == _htons(ETHER_TYPE_IPv6)) {
331 		ipv6_hdr->src_addr[15] = ipv6_hdr->src_addr[15] + 1;
332 	}
333 }
334 
335 /* if possible, calculate the checksum of a packet in hw or sw,
336  * depending on the testpmd command line configuration */
337 static uint64_t
338 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
339 	uint16_t testpmd_ol_flags)
340 {
341 	struct ipv4_hdr *ipv4_hdr = l3_hdr;
342 	struct udp_hdr *udp_hdr;
343 	struct tcp_hdr *tcp_hdr;
344 	struct sctp_hdr *sctp_hdr;
345 	uint64_t ol_flags = 0;
346 
347 	if (info->ethertype == _htons(ETHER_TYPE_IPv4)) {
348 		ipv4_hdr = l3_hdr;
349 		ipv4_hdr->hdr_checksum = 0;
350 
351 		ol_flags |= PKT_TX_IPV4;
352 		if (info->tso_segsz != 0 && info->l4_proto == IPPROTO_TCP) {
353 			ol_flags |= PKT_TX_IP_CKSUM;
354 		} else {
355 			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_IP_CKSUM)
356 				ol_flags |= PKT_TX_IP_CKSUM;
357 			else
358 				ipv4_hdr->hdr_checksum =
359 					rte_ipv4_cksum(ipv4_hdr);
360 		}
361 	} else if (info->ethertype == _htons(ETHER_TYPE_IPv6))
362 		ol_flags |= PKT_TX_IPV6;
363 	else
364 		return 0; /* packet type not supported, nothing to do */
365 
366 	if (info->l4_proto == IPPROTO_UDP) {
367 		udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
368 		/* do not recalculate udp cksum if it was 0 */
369 		if (udp_hdr->dgram_cksum != 0) {
370 			udp_hdr->dgram_cksum = 0;
371 			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_UDP_CKSUM) {
372 				ol_flags |= PKT_TX_UDP_CKSUM;
373 				udp_hdr->dgram_cksum = get_psd_sum(l3_hdr,
374 					info->ethertype, ol_flags);
375 			} else {
376 				udp_hdr->dgram_cksum =
377 					get_udptcp_checksum(l3_hdr, udp_hdr,
378 						info->ethertype);
379 			}
380 		}
381 	} else if (info->l4_proto == IPPROTO_TCP) {
382 		tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
383 		tcp_hdr->cksum = 0;
384 		if (info->tso_segsz != 0) {
385 			ol_flags |= PKT_TX_TCP_SEG;
386 			tcp_hdr->cksum = get_psd_sum(l3_hdr, info->ethertype,
387 				ol_flags);
388 		} else if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_TCP_CKSUM) {
389 			ol_flags |= PKT_TX_TCP_CKSUM;
390 			tcp_hdr->cksum = get_psd_sum(l3_hdr, info->ethertype,
391 				ol_flags);
392 		} else {
393 			tcp_hdr->cksum =
394 				get_udptcp_checksum(l3_hdr, tcp_hdr,
395 					info->ethertype);
396 		}
397 	} else if (info->l4_proto == IPPROTO_SCTP) {
398 		sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
399 		sctp_hdr->cksum = 0;
400 		/* sctp payload must be a multiple of 4 to be
401 		 * offloaded */
402 		if ((testpmd_ol_flags & TESTPMD_TX_OFFLOAD_SCTP_CKSUM) &&
403 			((ipv4_hdr->total_length & 0x3) == 0)) {
404 			ol_flags |= PKT_TX_SCTP_CKSUM;
405 		} else {
406 			/* XXX implement CRC32c, example available in
407 			 * RFC3309 */
408 		}
409 	}
410 
411 	return ol_flags;
412 }
413 
414 /* Calculate the checksum of outer header (only vxlan is supported,
415  * meaning IP + UDP). The caller already checked that it's a vxlan
416  * packet */
417 static uint64_t
418 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
419 	uint16_t testpmd_ol_flags)
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 	/* outer UDP checksum is always done in software as we have no
441 	 * hardware supporting it today, and no API for it. */
442 
443 	udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
444 	/* do not recalculate udp cksum if it was 0 */
445 	if (udp_hdr->dgram_cksum != 0) {
446 		udp_hdr->dgram_cksum = 0;
447 		if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
448 			udp_hdr->dgram_cksum =
449 				rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
450 		else
451 			udp_hdr->dgram_cksum =
452 				rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
453 	}
454 
455 	return ol_flags;
456 }
457 
458 /*
459  * Helper function.
460  * Performs actual copying.
461  * Returns number of segments in the destination mbuf on success,
462  * or negative error code on failure.
463  */
464 static int
465 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
466 	uint16_t seglen[], uint8_t nb_seg)
467 {
468 	uint32_t dlen, slen, tlen;
469 	uint32_t i, len;
470 	const struct rte_mbuf *m;
471 	const uint8_t *src;
472 	uint8_t *dst;
473 
474 	dlen = 0;
475 	slen = 0;
476 	tlen = 0;
477 
478 	dst = NULL;
479 	src = NULL;
480 
481 	m = ms;
482 	i = 0;
483 	while (ms != NULL && i != nb_seg) {
484 
485 		if (slen == 0) {
486 			slen = rte_pktmbuf_data_len(ms);
487 			src = rte_pktmbuf_mtod(ms, const uint8_t *);
488 		}
489 
490 		if (dlen == 0) {
491 			dlen = RTE_MIN(seglen[i], slen);
492 			md[i]->data_len = dlen;
493 			md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
494 			dst = rte_pktmbuf_mtod(md[i], uint8_t *);
495 		}
496 
497 		len = RTE_MIN(slen, dlen);
498 		memcpy(dst, src, len);
499 		tlen += len;
500 		slen -= len;
501 		dlen -= len;
502 		src += len;
503 		dst += len;
504 
505 		if (slen == 0)
506 			ms = ms->next;
507 		if (dlen == 0)
508 			i++;
509 	}
510 
511 	if (ms != NULL)
512 		return -ENOBUFS;
513 	else if (tlen != m->pkt_len)
514 		return -EINVAL;
515 
516 	md[0]->nb_segs = nb_seg;
517 	md[0]->pkt_len = tlen;
518 	md[0]->vlan_tci = m->vlan_tci;
519 	md[0]->vlan_tci_outer = m->vlan_tci_outer;
520 	md[0]->ol_flags = m->ol_flags;
521 	md[0]->tx_offload = m->tx_offload;
522 
523 	return nb_seg;
524 }
525 
526 /*
527  * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
528  * Copy packet contents and offload information into then new segmented mbuf.
529  */
530 static struct rte_mbuf *
531 pkt_copy_split(const struct rte_mbuf *pkt)
532 {
533 	int32_t n, rc;
534 	uint32_t i, len, nb_seg;
535 	struct rte_mempool *mp;
536 	uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
537 	struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
538 
539 	mp = current_fwd_lcore()->mbp;
540 
541 	if (tx_pkt_split == TX_PKT_SPLIT_RND)
542 		nb_seg = random() % tx_pkt_nb_segs + 1;
543 	else
544 		nb_seg = tx_pkt_nb_segs;
545 
546 	memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
547 
548 	/* calculate number of segments to use and their length. */
549 	len = 0;
550 	for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
551 		len += seglen[i];
552 		md[i] = NULL;
553 	}
554 
555 	n = pkt->pkt_len - len;
556 
557 	/* update size of the last segment to fit rest of the packet */
558 	if (n >= 0) {
559 		seglen[i - 1] += n;
560 		len += n;
561 	}
562 
563 	nb_seg = i;
564 	while (i != 0) {
565 		p = rte_pktmbuf_alloc(mp);
566 		if (p == NULL) {
567 			RTE_LOG(ERR, USER1,
568 				"failed to allocate %u-th of %u mbuf "
569 				"from mempool: %s\n",
570 				nb_seg - i, nb_seg, mp->name);
571 			break;
572 		}
573 
574 		md[--i] = p;
575 		if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
576 			RTE_LOG(ERR, USER1, "mempool %s, %u-th segment: "
577 				"expected seglen: %u, "
578 				"actual mbuf tailroom: %u\n",
579 				mp->name, i, seglen[i],
580 				rte_pktmbuf_tailroom(md[i]));
581 			break;
582 		}
583 	}
584 
585 	/* all mbufs successfully allocated, do copy */
586 	if (i == 0) {
587 		rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
588 		if (rc < 0)
589 			RTE_LOG(ERR, USER1,
590 				"mbuf_copy_split for %p(len=%u, nb_seg=%hhu) "
591 				"into %u segments failed with error code: %d\n",
592 				pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
593 
594 		/* figure out how many mbufs to free. */
595 		i = RTE_MAX(rc, 0);
596 	}
597 
598 	/* free unused mbufs */
599 	for (; i != nb_seg; i++) {
600 		rte_pktmbuf_free_seg(md[i]);
601 		md[i] = NULL;
602 	}
603 
604 	return md[0];
605 }
606 
607 /*
608  * Receive a burst of packets, and for each packet:
609  *  - parse packet, and try to recognize a supported packet type (1)
610  *  - if it's not a supported packet type, don't touch the packet, else:
611  *  - modify the IPs in inner headers and in outer headers if any
612  *  - reprocess the checksum of all supported layers. This is done in SW
613  *    or HW, depending on testpmd command line configuration
614  *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
615  *    segmentation offload (this implies HW TCP checksum)
616  * Then transmit packets on the output port.
617  *
618  * (1) Supported packets are:
619  *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
620  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
621  *           UDP|TCP|SCTP
622  *   Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
623  *   Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
624  *   Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
625  *
626  * The testpmd command line for this forward engine sets the flags
627  * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
628  * wether a checksum must be calculated in software or in hardware. The
629  * IP, UDP, TCP and SCTP flags always concern the inner layer. The
630  * OUTER_IP is only useful for tunnel packets.
631  */
632 static void
633 pkt_burst_checksum_forward(struct fwd_stream *fs)
634 {
635 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
636 	struct rte_port *txp;
637 	struct rte_mbuf *m, *p;
638 	struct ether_hdr *eth_hdr;
639 	void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
640 	uint16_t nb_rx;
641 	uint16_t nb_tx;
642 	uint16_t i;
643 	uint64_t ol_flags;
644 	uint16_t testpmd_ol_flags;
645 	uint32_t retry;
646 	uint32_t rx_bad_ip_csum;
647 	uint32_t rx_bad_l4_csum;
648 	struct testpmd_offload_info info;
649 
650 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
651 	uint64_t start_tsc;
652 	uint64_t end_tsc;
653 	uint64_t core_cycles;
654 #endif
655 
656 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
657 	start_tsc = rte_rdtsc();
658 #endif
659 
660 	/* receive a burst of packet */
661 	nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
662 				 nb_pkt_per_burst);
663 	if (unlikely(nb_rx == 0))
664 		return;
665 
666 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
667 	fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
668 #endif
669 	fs->rx_packets += nb_rx;
670 	rx_bad_ip_csum = 0;
671 	rx_bad_l4_csum = 0;
672 
673 	txp = &ports[fs->tx_port];
674 	testpmd_ol_flags = txp->tx_ol_flags;
675 	memset(&info, 0, sizeof(info));
676 	info.tso_segsz = txp->tso_segsz;
677 
678 	for (i = 0; i < nb_rx; i++) {
679 		if (likely(i < nb_rx - 1))
680 			rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
681 						       void *));
682 
683 		ol_flags = 0;
684 		info.is_tunnel = 0;
685 		m = pkts_burst[i];
686 
687 		/* Update the L3/L4 checksum error packet statistics */
688 		rx_bad_ip_csum += ((m->ol_flags & PKT_RX_IP_CKSUM_BAD) != 0);
689 		rx_bad_l4_csum += ((m->ol_flags & PKT_RX_L4_CKSUM_BAD) != 0);
690 
691 		/* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
692 		 * and inner headers */
693 
694 		eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
695 		ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
696 				&eth_hdr->d_addr);
697 		ether_addr_copy(&ports[fs->tx_port].eth_addr,
698 				&eth_hdr->s_addr);
699 		parse_ethernet(eth_hdr, &info);
700 		l3_hdr = (char *)eth_hdr + info.l2_len;
701 
702 		/* check if it's a supported tunnel */
703 		if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_PARSE_TUNNEL) {
704 			if (info.l4_proto == IPPROTO_UDP) {
705 				struct udp_hdr *udp_hdr;
706 				udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
707 					info.l3_len);
708 				parse_vxlan(udp_hdr, &info, m->packet_type);
709 			} else if (info.l4_proto == IPPROTO_GRE) {
710 				struct simple_gre_hdr *gre_hdr;
711 				gre_hdr = (struct simple_gre_hdr *)
712 					((char *)l3_hdr + info.l3_len);
713 				parse_gre(gre_hdr, &info);
714 			} else if (info.l4_proto == IPPROTO_IPIP) {
715 				void *encap_ip_hdr;
716 				encap_ip_hdr = (char *)l3_hdr + info.l3_len;
717 				parse_encap_ip(encap_ip_hdr, &info);
718 			}
719 		}
720 
721 		/* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
722 		if (info.is_tunnel) {
723 			outer_l3_hdr = l3_hdr;
724 			l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
725 		}
726 
727 		/* step 2: change all source IPs (v4 or v6) so we need
728 		 * to recompute the chksums even if they were correct */
729 
730 		change_ip_addresses(l3_hdr, info.ethertype);
731 		if (info.is_tunnel == 1)
732 			change_ip_addresses(outer_l3_hdr, info.outer_ethertype);
733 
734 		/* step 3: depending on user command line configuration,
735 		 * recompute checksum either in software or flag the
736 		 * mbuf to offload the calculation to the NIC. If TSO
737 		 * is configured, prepare the mbuf for TCP segmentation. */
738 
739 		/* process checksums of inner headers first */
740 		ol_flags |= process_inner_cksums(l3_hdr, &info, testpmd_ol_flags);
741 
742 		/* Then process outer headers if any. Note that the software
743 		 * checksum will be wrong if one of the inner checksums is
744 		 * processed in hardware. */
745 		if (info.is_tunnel == 1) {
746 			ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
747 				testpmd_ol_flags);
748 		}
749 
750 		/* step 4: fill the mbuf meta data (flags and header lengths) */
751 
752 		if (info.is_tunnel == 1) {
753 			if (testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM) {
754 				m->outer_l2_len = info.outer_l2_len;
755 				m->outer_l3_len = info.outer_l3_len;
756 				m->l2_len = info.l2_len;
757 				m->l3_len = info.l3_len;
758 				m->l4_len = info.l4_len;
759 			}
760 			else {
761 				/* if there is a outer UDP cksum
762 				   processed in sw and the inner in hw,
763 				   the outer checksum will be wrong as
764 				   the payload will be modified by the
765 				   hardware */
766 				m->l2_len = info.outer_l2_len +
767 					info.outer_l3_len + info.l2_len;
768 				m->l3_len = info.l3_len;
769 				m->l4_len = info.l4_len;
770 			}
771 		} else {
772 			/* this is only useful if an offload flag is
773 			 * set, but it does not hurt to fill it in any
774 			 * case */
775 			m->l2_len = info.l2_len;
776 			m->l3_len = info.l3_len;
777 			m->l4_len = info.l4_len;
778 		}
779 		m->tso_segsz = info.tso_segsz;
780 		m->ol_flags = ol_flags;
781 
782 		/* Do split & copy for the packet. */
783 		if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
784 			p = pkt_copy_split(m);
785 			if (p != NULL) {
786 				rte_pktmbuf_free(m);
787 				m = p;
788 				pkts_burst[i] = m;
789 			}
790 		}
791 
792 		/* if verbose mode is enabled, dump debug info */
793 		if (verbose_level > 0) {
794 			struct {
795 				uint64_t flag;
796 				uint64_t mask;
797 			} tx_flags[] = {
798 				{ PKT_TX_IP_CKSUM, PKT_TX_IP_CKSUM },
799 				{ PKT_TX_UDP_CKSUM, PKT_TX_L4_MASK },
800 				{ PKT_TX_TCP_CKSUM, PKT_TX_L4_MASK },
801 				{ PKT_TX_SCTP_CKSUM, PKT_TX_L4_MASK },
802 				{ PKT_TX_IPV4, PKT_TX_IPV4 },
803 				{ PKT_TX_IPV6, PKT_TX_IPV6 },
804 				{ PKT_TX_OUTER_IP_CKSUM, PKT_TX_OUTER_IP_CKSUM },
805 				{ PKT_TX_OUTER_IPV4, PKT_TX_OUTER_IPV4 },
806 				{ PKT_TX_OUTER_IPV6, PKT_TX_OUTER_IPV6 },
807 				{ PKT_TX_TCP_SEG, PKT_TX_TCP_SEG },
808 			};
809 			unsigned j;
810 			const char *name;
811 
812 			printf("-----------------\n");
813 			printf("mbuf=%p, pkt_len=%u, nb_segs=%hhu:\n",
814 				m, m->pkt_len, m->nb_segs);
815 			/* dump rx parsed packet info */
816 			printf("rx: l2_len=%d ethertype=%x l3_len=%d "
817 				"l4_proto=%d l4_len=%d\n",
818 				info.l2_len, rte_be_to_cpu_16(info.ethertype),
819 				info.l3_len, info.l4_proto, info.l4_len);
820 			if (info.is_tunnel == 1)
821 				printf("rx: outer_l2_len=%d outer_ethertype=%x "
822 					"outer_l3_len=%d\n", info.outer_l2_len,
823 					rte_be_to_cpu_16(info.outer_ethertype),
824 					info.outer_l3_len);
825 			/* dump tx packet info */
826 			if ((testpmd_ol_flags & (TESTPMD_TX_OFFLOAD_IP_CKSUM |
827 						TESTPMD_TX_OFFLOAD_UDP_CKSUM |
828 						TESTPMD_TX_OFFLOAD_TCP_CKSUM |
829 						TESTPMD_TX_OFFLOAD_SCTP_CKSUM)) ||
830 				info.tso_segsz != 0)
831 				printf("tx: m->l2_len=%d m->l3_len=%d "
832 					"m->l4_len=%d\n",
833 					m->l2_len, m->l3_len, m->l4_len);
834 			if ((info.is_tunnel == 1) &&
835 				(testpmd_ol_flags & TESTPMD_TX_OFFLOAD_OUTER_IP_CKSUM))
836 				printf("tx: m->outer_l2_len=%d m->outer_l3_len=%d\n",
837 					m->outer_l2_len, m->outer_l3_len);
838 			if (info.tso_segsz != 0)
839 				printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
840 			printf("tx: flags=");
841 			for (j = 0; j < sizeof(tx_flags)/sizeof(*tx_flags); j++) {
842 				name = rte_get_tx_ol_flag_name(tx_flags[j].flag);
843 				if ((m->ol_flags & tx_flags[j].mask) ==
844 					tx_flags[j].flag)
845 					printf("%s ", name);
846 			}
847 			printf("\n");
848 		}
849 	}
850 	nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
851 	/*
852 	 * Retry if necessary
853 	 */
854 	if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
855 		retry = 0;
856 		while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
857 			rte_delay_us(burst_tx_delay_time);
858 			nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
859 					&pkts_burst[nb_tx], nb_rx - nb_tx);
860 		}
861 	}
862 	fs->tx_packets += nb_tx;
863 	fs->rx_bad_ip_csum += rx_bad_ip_csum;
864 	fs->rx_bad_l4_csum += rx_bad_l4_csum;
865 
866 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
867 	fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
868 #endif
869 	if (unlikely(nb_tx < nb_rx)) {
870 		fs->fwd_dropped += (nb_rx - nb_tx);
871 		do {
872 			rte_pktmbuf_free(pkts_burst[nb_tx]);
873 		} while (++nb_tx < nb_rx);
874 	}
875 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
876 	end_tsc = rte_rdtsc();
877 	core_cycles = (end_tsc - start_tsc);
878 	fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
879 #endif
880 }
881 
882 struct fwd_engine csum_fwd_engine = {
883 	.fwd_mode_name  = "csum",
884 	.port_fwd_begin = NULL,
885 	.port_fwd_end   = NULL,
886 	.packet_fwd     = pkt_burst_checksum_forward,
887 };
888