xref: /dpdk/app/test-pmd/csumonly.c (revision 8f1d23ece06adff5eae9f1b4365bdbbd3abee2b2)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2014 Intel Corporation.
3  * Copyright 2014 6WIND S.A.
4  */
5 
6 #include <stdarg.h>
7 #include <stdio.h>
8 #include <errno.h>
9 #include <stdint.h>
10 #include <unistd.h>
11 #include <inttypes.h>
12 
13 #include <sys/queue.h>
14 #include <sys/stat.h>
15 
16 #include <rte_common.h>
17 #include <rte_byteorder.h>
18 #include <rte_log.h>
19 #include <rte_debug.h>
20 #include <rte_cycles.h>
21 #include <rte_memory.h>
22 #include <rte_memcpy.h>
23 #include <rte_launch.h>
24 #include <rte_eal.h>
25 #include <rte_per_lcore.h>
26 #include <rte_lcore.h>
27 #include <rte_branch_prediction.h>
28 #include <rte_mempool.h>
29 #include <rte_mbuf.h>
30 #include <rte_interrupts.h>
31 #include <rte_ether.h>
32 #include <rte_ethdev.h>
33 #include <rte_ip.h>
34 #include <rte_tcp.h>
35 #include <rte_udp.h>
36 #include <rte_vxlan.h>
37 #include <rte_sctp.h>
38 #include <rte_gtp.h>
39 #include <rte_prefetch.h>
40 #include <rte_string_fns.h>
41 #include <rte_flow.h>
42 #ifdef RTE_LIB_GRO
43 #include <rte_gro.h>
44 #endif
45 #ifdef RTE_LIB_GSO
46 #include <rte_gso.h>
47 #endif
48 #include <rte_geneve.h>
49 
50 #include "testpmd.h"
51 
52 #define IP_DEFTTL  64   /* from RFC 1340. */
53 
54 #define GRE_CHECKSUM_PRESENT	0x8000
55 #define GRE_KEY_PRESENT		0x2000
56 #define GRE_SEQUENCE_PRESENT	0x1000
57 #define GRE_EXT_LEN		4
58 #define GRE_SUPPORTED_FIELDS	(GRE_CHECKSUM_PRESENT | GRE_KEY_PRESENT |\
59 				 GRE_SEQUENCE_PRESENT)
60 
61 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */
62 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
63 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
64 #else
65 #define _htons(x) (x)
66 #endif
67 
68 uint16_t vxlan_gpe_udp_port = RTE_VXLAN_GPE_DEFAULT_PORT;
69 uint16_t geneve_udp_port = RTE_GENEVE_DEFAULT_PORT;
70 
71 /* structure that caches offload info for the current packet */
72 struct testpmd_offload_info {
73 	uint16_t ethertype;
74 #ifdef RTE_LIB_GSO
75 	uint8_t gso_enable;
76 #endif
77 	uint16_t l2_len;
78 	uint16_t l3_len;
79 	uint16_t l4_len;
80 	uint8_t l4_proto;
81 	uint8_t is_tunnel;
82 	uint16_t outer_ethertype;
83 	uint16_t outer_l2_len;
84 	uint16_t outer_l3_len;
85 	uint8_t outer_l4_proto;
86 	uint16_t tso_segsz;
87 	uint16_t tunnel_tso_segsz;
88 	uint32_t pkt_len;
89 };
90 
91 /* simplified GRE header */
92 struct simple_gre_hdr {
93 	uint16_t flags;
94 	uint16_t proto;
95 } __rte_packed;
96 
97 static uint16_t
98 get_udptcp_checksum(struct rte_mbuf *m, void *l3_hdr, uint16_t l4_off,
99 		    uint16_t ethertype)
100 {
101 	if (ethertype == _htons(RTE_ETHER_TYPE_IPV4))
102 		return rte_ipv4_udptcp_cksum_mbuf(m, l3_hdr, l4_off);
103 	else /* assume ethertype == RTE_ETHER_TYPE_IPV6 */
104 		return rte_ipv6_udptcp_cksum_mbuf(m, l3_hdr, l4_off);
105 }
106 
107 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
108 static void
109 parse_ipv4(struct rte_ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
110 {
111 	struct rte_tcp_hdr *tcp_hdr;
112 
113 	info->l3_len = rte_ipv4_hdr_len(ipv4_hdr);
114 	info->l4_proto = ipv4_hdr->next_proto_id;
115 
116 	/* only fill l4_len for TCP, it's useful for TSO */
117 	if (info->l4_proto == IPPROTO_TCP) {
118 		tcp_hdr = (struct rte_tcp_hdr *)
119 			((char *)ipv4_hdr + info->l3_len);
120 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
121 	} else if (info->l4_proto == IPPROTO_UDP)
122 		info->l4_len = sizeof(struct rte_udp_hdr);
123 	else
124 		info->l4_len = 0;
125 }
126 
127 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
128 static void
129 parse_ipv6(struct rte_ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
130 {
131 	struct rte_tcp_hdr *tcp_hdr;
132 
133 	info->l3_len = sizeof(struct rte_ipv6_hdr);
134 	info->l4_proto = ipv6_hdr->proto;
135 
136 	/* only fill l4_len for TCP, it's useful for TSO */
137 	if (info->l4_proto == IPPROTO_TCP) {
138 		tcp_hdr = (struct rte_tcp_hdr *)
139 			((char *)ipv6_hdr + info->l3_len);
140 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
141 	} else if (info->l4_proto == IPPROTO_UDP)
142 		info->l4_len = sizeof(struct rte_udp_hdr);
143 	else
144 		info->l4_len = 0;
145 }
146 
147 /*
148  * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
149  * ipproto. This function is able to recognize IPv4/IPv6 with optional VLAN
150  * headers. The l4_len argument is only set in case of TCP (useful for TSO).
151  */
152 static void
153 parse_ethernet(struct rte_ether_hdr *eth_hdr, struct testpmd_offload_info *info)
154 {
155 	struct rte_ipv4_hdr *ipv4_hdr;
156 	struct rte_ipv6_hdr *ipv6_hdr;
157 	struct rte_vlan_hdr *vlan_hdr;
158 
159 	info->l2_len = sizeof(struct rte_ether_hdr);
160 	info->ethertype = eth_hdr->ether_type;
161 
162 	while (info->ethertype == _htons(RTE_ETHER_TYPE_VLAN) ||
163 	       info->ethertype == _htons(RTE_ETHER_TYPE_QINQ)) {
164 		vlan_hdr = (struct rte_vlan_hdr *)
165 			((char *)eth_hdr + info->l2_len);
166 		info->l2_len  += sizeof(struct rte_vlan_hdr);
167 		info->ethertype = vlan_hdr->eth_proto;
168 	}
169 
170 	switch (info->ethertype) {
171 	case _htons(RTE_ETHER_TYPE_IPV4):
172 		ipv4_hdr = (struct rte_ipv4_hdr *)
173 			((char *)eth_hdr + info->l2_len);
174 		parse_ipv4(ipv4_hdr, info);
175 		break;
176 	case _htons(RTE_ETHER_TYPE_IPV6):
177 		ipv6_hdr = (struct rte_ipv6_hdr *)
178 			((char *)eth_hdr + info->l2_len);
179 		parse_ipv6(ipv6_hdr, info);
180 		break;
181 	default:
182 		info->l4_len = 0;
183 		info->l3_len = 0;
184 		info->l4_proto = 0;
185 		break;
186 	}
187 }
188 
189 /* Fill in outer layers length */
190 static void
191 update_tunnel_outer(struct testpmd_offload_info *info)
192 {
193 	info->is_tunnel = 1;
194 	info->outer_ethertype = info->ethertype;
195 	info->outer_l2_len = info->l2_len;
196 	info->outer_l3_len = info->l3_len;
197 	info->outer_l4_proto = info->l4_proto;
198 }
199 
200 /*
201  * Parse a GTP protocol header.
202  * No optional fields and next extension header type.
203  */
204 static void
205 parse_gtp(struct rte_udp_hdr *udp_hdr,
206 	  struct testpmd_offload_info *info)
207 {
208 	struct rte_ipv4_hdr *ipv4_hdr;
209 	struct rte_ipv6_hdr *ipv6_hdr;
210 	struct rte_gtp_hdr *gtp_hdr;
211 	uint8_t gtp_len = sizeof(*gtp_hdr);
212 	uint8_t ip_ver;
213 
214 	/* Check udp destination port. */
215 	if (udp_hdr->dst_port != _htons(RTE_GTPC_UDP_PORT) &&
216 	    udp_hdr->src_port != _htons(RTE_GTPC_UDP_PORT) &&
217 	    udp_hdr->dst_port != _htons(RTE_GTPU_UDP_PORT))
218 		return;
219 
220 	update_tunnel_outer(info);
221 	info->l2_len = 0;
222 
223 	gtp_hdr = (struct rte_gtp_hdr *)((char *)udp_hdr +
224 		  sizeof(struct rte_udp_hdr));
225 	if (gtp_hdr->e || gtp_hdr->s || gtp_hdr->pn)
226 		gtp_len += sizeof(struct rte_gtp_hdr_ext_word);
227 	/*
228 	 * Check message type. If message type is 0xff, it is
229 	 * a GTP data packet. If not, it is a GTP control packet
230 	 */
231 	if (gtp_hdr->msg_type == 0xff) {
232 		ip_ver = *(uint8_t *)((char *)gtp_hdr + gtp_len);
233 		ip_ver = (ip_ver) & 0xf0;
234 
235 		if (ip_ver == RTE_GTP_TYPE_IPV4) {
236 			ipv4_hdr = (struct rte_ipv4_hdr *)((char *)gtp_hdr +
237 				   gtp_len);
238 			info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
239 			parse_ipv4(ipv4_hdr, info);
240 		} else if (ip_ver == RTE_GTP_TYPE_IPV6) {
241 			ipv6_hdr = (struct rte_ipv6_hdr *)((char *)gtp_hdr +
242 				   gtp_len);
243 			info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
244 			parse_ipv6(ipv6_hdr, info);
245 		}
246 	} else {
247 		info->ethertype = 0;
248 		info->l4_len = 0;
249 		info->l3_len = 0;
250 		info->l4_proto = 0;
251 	}
252 
253 	info->l2_len += RTE_ETHER_GTP_HLEN;
254 }
255 
256 /* Parse a vxlan header */
257 static void
258 parse_vxlan(struct rte_udp_hdr *udp_hdr,
259 	    struct testpmd_offload_info *info)
260 {
261 	struct rte_ether_hdr *eth_hdr;
262 
263 	/* check udp destination port, RTE_VXLAN_DEFAULT_PORT (4789) is the
264 	 * default vxlan port (rfc7348) or that the rx offload flag is set
265 	 * (i40e only currently)
266 	 */
267 	if (udp_hdr->dst_port != _htons(RTE_VXLAN_DEFAULT_PORT))
268 		return;
269 
270 	update_tunnel_outer(info);
271 
272 	eth_hdr = (struct rte_ether_hdr *)((char *)udp_hdr +
273 		sizeof(struct rte_udp_hdr) +
274 		sizeof(struct rte_vxlan_hdr));
275 
276 	parse_ethernet(eth_hdr, info);
277 	info->l2_len += RTE_ETHER_VXLAN_HLEN; /* add udp + vxlan */
278 }
279 
280 /* Parse a vxlan-gpe header */
281 static void
282 parse_vxlan_gpe(struct rte_udp_hdr *udp_hdr,
283 	    struct testpmd_offload_info *info)
284 {
285 	struct rte_ether_hdr *eth_hdr;
286 	struct rte_ipv4_hdr *ipv4_hdr;
287 	struct rte_ipv6_hdr *ipv6_hdr;
288 	struct rte_vxlan_gpe_hdr *vxlan_gpe_hdr;
289 	uint8_t vxlan_gpe_len = sizeof(*vxlan_gpe_hdr);
290 
291 	/* Check udp destination port. */
292 	if (udp_hdr->dst_port != _htons(vxlan_gpe_udp_port))
293 		return;
294 
295 	vxlan_gpe_hdr = (struct rte_vxlan_gpe_hdr *)((char *)udp_hdr +
296 				sizeof(struct rte_udp_hdr));
297 
298 	if (!vxlan_gpe_hdr->proto || vxlan_gpe_hdr->proto ==
299 	    RTE_VXLAN_GPE_TYPE_IPV4) {
300 		update_tunnel_outer(info);
301 
302 		ipv4_hdr = (struct rte_ipv4_hdr *)((char *)vxlan_gpe_hdr +
303 			   vxlan_gpe_len);
304 
305 		parse_ipv4(ipv4_hdr, info);
306 		info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
307 		info->l2_len = 0;
308 
309 	} else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_IPV6) {
310 		update_tunnel_outer(info);
311 
312 		ipv6_hdr = (struct rte_ipv6_hdr *)((char *)vxlan_gpe_hdr +
313 			   vxlan_gpe_len);
314 
315 		info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
316 		parse_ipv6(ipv6_hdr, info);
317 		info->l2_len = 0;
318 
319 	} else if (vxlan_gpe_hdr->proto == RTE_VXLAN_GPE_TYPE_ETH) {
320 		update_tunnel_outer(info);
321 
322 		eth_hdr = (struct rte_ether_hdr *)((char *)vxlan_gpe_hdr +
323 			  vxlan_gpe_len);
324 
325 		parse_ethernet(eth_hdr, info);
326 	} else
327 		return;
328 
329 	info->l2_len += RTE_ETHER_VXLAN_GPE_HLEN;
330 }
331 
332 /* Parse a geneve header */
333 static void
334 parse_geneve(struct rte_udp_hdr *udp_hdr,
335 	    struct testpmd_offload_info *info)
336 {
337 	struct rte_ether_hdr *eth_hdr;
338 	struct rte_ipv4_hdr *ipv4_hdr;
339 	struct rte_ipv6_hdr *ipv6_hdr;
340 	struct rte_geneve_hdr *geneve_hdr;
341 	uint16_t geneve_len;
342 
343 	/* Check udp destination port. */
344 	if (udp_hdr->dst_port != _htons(geneve_udp_port))
345 		return;
346 
347 	geneve_hdr = (struct rte_geneve_hdr *)((char *)udp_hdr +
348 				sizeof(struct rte_udp_hdr));
349 	geneve_len = sizeof(struct rte_geneve_hdr) + geneve_hdr->opt_len * 4;
350 	if (!geneve_hdr->proto || geneve_hdr->proto ==
351 	    _htons(RTE_ETHER_TYPE_IPV4)) {
352 		update_tunnel_outer(info);
353 		ipv4_hdr = (struct rte_ipv4_hdr *)((char *)geneve_hdr +
354 			   geneve_len);
355 		parse_ipv4(ipv4_hdr, info);
356 		info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
357 		info->l2_len = 0;
358 	} else if (geneve_hdr->proto == _htons(RTE_ETHER_TYPE_IPV6)) {
359 		update_tunnel_outer(info);
360 		ipv6_hdr = (struct rte_ipv6_hdr *)((char *)geneve_hdr +
361 			   geneve_len);
362 		info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
363 		parse_ipv6(ipv6_hdr, info);
364 		info->l2_len = 0;
365 
366 	} else if (geneve_hdr->proto == _htons(RTE_GENEVE_TYPE_ETH)) {
367 		update_tunnel_outer(info);
368 		eth_hdr = (struct rte_ether_hdr *)((char *)geneve_hdr +
369 			  geneve_len);
370 		parse_ethernet(eth_hdr, info);
371 	} else
372 		return;
373 
374 	info->l2_len +=
375 		(sizeof(struct rte_udp_hdr) + sizeof(struct rte_geneve_hdr) +
376 		((struct rte_geneve_hdr *)geneve_hdr)->opt_len * 4);
377 }
378 
379 /* Parse a gre header */
380 static void
381 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
382 {
383 	struct rte_ether_hdr *eth_hdr;
384 	struct rte_ipv4_hdr *ipv4_hdr;
385 	struct rte_ipv6_hdr *ipv6_hdr;
386 	uint8_t gre_len = 0;
387 
388 	gre_len += sizeof(struct simple_gre_hdr);
389 
390 	if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
391 		gre_len += GRE_EXT_LEN;
392 	if (gre_hdr->flags & _htons(GRE_SEQUENCE_PRESENT))
393 		gre_len += GRE_EXT_LEN;
394 	if (gre_hdr->flags & _htons(GRE_CHECKSUM_PRESENT))
395 		gre_len += GRE_EXT_LEN;
396 
397 	if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPV4)) {
398 		update_tunnel_outer(info);
399 
400 		ipv4_hdr = (struct rte_ipv4_hdr *)((char *)gre_hdr + gre_len);
401 
402 		parse_ipv4(ipv4_hdr, info);
403 		info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
404 		info->l2_len = 0;
405 
406 	} else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_IPV6)) {
407 		update_tunnel_outer(info);
408 
409 		ipv6_hdr = (struct rte_ipv6_hdr *)((char *)gre_hdr + gre_len);
410 
411 		info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
412 		parse_ipv6(ipv6_hdr, info);
413 		info->l2_len = 0;
414 
415 	} else if (gre_hdr->proto == _htons(RTE_ETHER_TYPE_TEB)) {
416 		update_tunnel_outer(info);
417 
418 		eth_hdr = (struct rte_ether_hdr *)((char *)gre_hdr + gre_len);
419 
420 		parse_ethernet(eth_hdr, info);
421 	} else
422 		return;
423 
424 	info->l2_len += gre_len;
425 }
426 
427 
428 /* Parse an encapsulated ip or ipv6 header */
429 static void
430 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
431 {
432 	struct rte_ipv4_hdr *ipv4_hdr = encap_ip;
433 	struct rte_ipv6_hdr *ipv6_hdr = encap_ip;
434 	uint8_t ip_version;
435 
436 	ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
437 
438 	if (ip_version != 4 && ip_version != 6)
439 		return;
440 
441 	info->is_tunnel = 1;
442 	info->outer_ethertype = info->ethertype;
443 	info->outer_l2_len = info->l2_len;
444 	info->outer_l3_len = info->l3_len;
445 
446 	if (ip_version == 4) {
447 		parse_ipv4(ipv4_hdr, info);
448 		info->ethertype = _htons(RTE_ETHER_TYPE_IPV4);
449 	} else {
450 		parse_ipv6(ipv6_hdr, info);
451 		info->ethertype = _htons(RTE_ETHER_TYPE_IPV6);
452 	}
453 	info->l2_len = 0;
454 }
455 
456 /* if possible, calculate the checksum of a packet in hw or sw,
457  * depending on the testpmd command line configuration */
458 static uint64_t
459 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
460 	uint64_t tx_offloads, struct rte_mbuf *m)
461 {
462 	struct rte_ipv4_hdr *ipv4_hdr = l3_hdr;
463 	struct rte_udp_hdr *udp_hdr;
464 	struct rte_tcp_hdr *tcp_hdr;
465 	struct rte_sctp_hdr *sctp_hdr;
466 	uint64_t ol_flags = 0;
467 	uint32_t max_pkt_len, tso_segsz = 0;
468 	uint16_t l4_off;
469 
470 	/* ensure packet is large enough to require tso */
471 	if (!info->is_tunnel) {
472 		max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
473 			info->tso_segsz;
474 		if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
475 			tso_segsz = info->tso_segsz;
476 	} else {
477 		max_pkt_len = info->outer_l2_len + info->outer_l3_len +
478 			info->l2_len + info->l3_len + info->l4_len +
479 			info->tunnel_tso_segsz;
480 		if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
481 			tso_segsz = info->tunnel_tso_segsz;
482 	}
483 
484 	if (info->ethertype == _htons(RTE_ETHER_TYPE_IPV4)) {
485 		ipv4_hdr = l3_hdr;
486 
487 		ol_flags |= RTE_MBUF_F_TX_IPV4;
488 		if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
489 			ol_flags |= RTE_MBUF_F_TX_IP_CKSUM;
490 		} else {
491 			if (tx_offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM) {
492 				ol_flags |= RTE_MBUF_F_TX_IP_CKSUM;
493 			} else {
494 				ipv4_hdr->hdr_checksum = 0;
495 				ipv4_hdr->hdr_checksum =
496 					rte_ipv4_cksum(ipv4_hdr);
497 			}
498 		}
499 	} else if (info->ethertype == _htons(RTE_ETHER_TYPE_IPV6))
500 		ol_flags |= RTE_MBUF_F_TX_IPV6;
501 	else
502 		return 0; /* packet type not supported, nothing to do */
503 
504 	if (info->l4_proto == IPPROTO_UDP) {
505 		udp_hdr = (struct rte_udp_hdr *)((char *)l3_hdr + info->l3_len);
506 		/* do not recalculate udp cksum if it was 0 */
507 		if (udp_hdr->dgram_cksum != 0) {
508 			if (tx_offloads & RTE_ETH_TX_OFFLOAD_UDP_CKSUM) {
509 				ol_flags |= RTE_MBUF_F_TX_UDP_CKSUM;
510 			} else {
511 				if (info->is_tunnel)
512 					l4_off = info->outer_l2_len +
513 						 info->outer_l3_len +
514 						 info->l2_len + info->l3_len;
515 				else
516 					l4_off = info->l2_len +	info->l3_len;
517 				udp_hdr->dgram_cksum = 0;
518 				udp_hdr->dgram_cksum =
519 					get_udptcp_checksum(m, l3_hdr, l4_off,
520 						info->ethertype);
521 			}
522 		}
523 #ifdef RTE_LIB_GSO
524 		if (info->gso_enable)
525 			ol_flags |= RTE_MBUF_F_TX_UDP_SEG;
526 #endif
527 	} else if (info->l4_proto == IPPROTO_TCP) {
528 		tcp_hdr = (struct rte_tcp_hdr *)((char *)l3_hdr + info->l3_len);
529 		if (tso_segsz)
530 			ol_flags |= RTE_MBUF_F_TX_TCP_SEG;
531 		else if (tx_offloads & RTE_ETH_TX_OFFLOAD_TCP_CKSUM) {
532 			ol_flags |= RTE_MBUF_F_TX_TCP_CKSUM;
533 		} else {
534 			if (info->is_tunnel)
535 				l4_off = info->outer_l2_len + info->outer_l3_len +
536 					 info->l2_len + info->l3_len;
537 			else
538 				l4_off = info->l2_len + info->l3_len;
539 			tcp_hdr->cksum = 0;
540 			tcp_hdr->cksum =
541 				get_udptcp_checksum(m, l3_hdr, l4_off,
542 					info->ethertype);
543 		}
544 #ifdef RTE_LIB_GSO
545 		if (info->gso_enable)
546 			ol_flags |= RTE_MBUF_F_TX_TCP_SEG;
547 #endif
548 	} else if (info->l4_proto == IPPROTO_SCTP) {
549 		sctp_hdr = (struct rte_sctp_hdr *)
550 			((char *)l3_hdr + info->l3_len);
551 		/* sctp payload must be a multiple of 4 to be
552 		 * offloaded */
553 		if ((tx_offloads & RTE_ETH_TX_OFFLOAD_SCTP_CKSUM) &&
554 			((ipv4_hdr->total_length & 0x3) == 0)) {
555 			ol_flags |= RTE_MBUF_F_TX_SCTP_CKSUM;
556 		} else {
557 			sctp_hdr->cksum = 0;
558 			/* XXX implement CRC32c, example available in
559 			 * RFC3309 */
560 		}
561 	}
562 
563 	return ol_flags;
564 }
565 
566 /* Calculate the checksum of outer header */
567 static uint64_t
568 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
569 	uint64_t tx_offloads, int tso_enabled, struct rte_mbuf *m)
570 {
571 	struct rte_ipv4_hdr *ipv4_hdr = outer_l3_hdr;
572 	struct rte_ipv6_hdr *ipv6_hdr = outer_l3_hdr;
573 	struct rte_udp_hdr *udp_hdr;
574 	uint64_t ol_flags = 0;
575 
576 	if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4)) {
577 		ipv4_hdr->hdr_checksum = 0;
578 		ol_flags |= RTE_MBUF_F_TX_OUTER_IPV4;
579 
580 		if (tx_offloads	& RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM)
581 			ol_flags |= RTE_MBUF_F_TX_OUTER_IP_CKSUM;
582 		else
583 			ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
584 	} else
585 		ol_flags |= RTE_MBUF_F_TX_OUTER_IPV6;
586 
587 	if (info->outer_l4_proto != IPPROTO_UDP)
588 		return ol_flags;
589 
590 	udp_hdr = (struct rte_udp_hdr *)
591 		((char *)outer_l3_hdr + info->outer_l3_len);
592 
593 	if (tso_enabled)
594 		ol_flags |= RTE_MBUF_F_TX_TCP_SEG;
595 
596 	/* Skip SW outer UDP checksum generation if HW supports it */
597 	if (tx_offloads & RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM) {
598 		if (info->outer_ethertype == _htons(RTE_ETHER_TYPE_IPV4))
599 			udp_hdr->dgram_cksum
600 				= rte_ipv4_phdr_cksum(ipv4_hdr, ol_flags);
601 		else
602 			udp_hdr->dgram_cksum
603 				= rte_ipv6_phdr_cksum(ipv6_hdr, ol_flags);
604 
605 		ol_flags |= RTE_MBUF_F_TX_OUTER_UDP_CKSUM;
606 		return ol_flags;
607 	}
608 
609 	/* outer UDP checksum is done in software. In the other side, for
610 	 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
611 	 * set to zero.
612 	 *
613 	 * If a packet will be TSOed into small packets by NIC, we cannot
614 	 * set/calculate a non-zero checksum, because it will be a wrong
615 	 * value after the packet be split into several small packets.
616 	 */
617 	if (tso_enabled)
618 		udp_hdr->dgram_cksum = 0;
619 
620 	/* do not recalculate udp cksum if it was 0 */
621 	if (udp_hdr->dgram_cksum != 0) {
622 		udp_hdr->dgram_cksum = 0;
623 		udp_hdr->dgram_cksum = get_udptcp_checksum(m, outer_l3_hdr,
624 					info->outer_l2_len + info->outer_l3_len,
625 					info->outer_ethertype);
626 	}
627 
628 	return ol_flags;
629 }
630 
631 /*
632  * Helper function.
633  * Performs actual copying.
634  * Returns number of segments in the destination mbuf on success,
635  * or negative error code on failure.
636  */
637 static int
638 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
639 	uint16_t seglen[], uint8_t nb_seg)
640 {
641 	uint32_t dlen, slen, tlen;
642 	uint32_t i, len;
643 	const struct rte_mbuf *m;
644 	const uint8_t *src;
645 	uint8_t *dst;
646 
647 	dlen = 0;
648 	slen = 0;
649 	tlen = 0;
650 
651 	dst = NULL;
652 	src = NULL;
653 
654 	m = ms;
655 	i = 0;
656 	while (ms != NULL && i != nb_seg) {
657 
658 		if (slen == 0) {
659 			slen = rte_pktmbuf_data_len(ms);
660 			src = rte_pktmbuf_mtod(ms, const uint8_t *);
661 		}
662 
663 		if (dlen == 0) {
664 			dlen = RTE_MIN(seglen[i], slen);
665 			md[i]->data_len = dlen;
666 			md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
667 			dst = rte_pktmbuf_mtod(md[i], uint8_t *);
668 		}
669 
670 		len = RTE_MIN(slen, dlen);
671 		memcpy(dst, src, len);
672 		tlen += len;
673 		slen -= len;
674 		dlen -= len;
675 		src += len;
676 		dst += len;
677 
678 		if (slen == 0)
679 			ms = ms->next;
680 		if (dlen == 0)
681 			i++;
682 	}
683 
684 	if (ms != NULL)
685 		return -ENOBUFS;
686 	else if (tlen != m->pkt_len)
687 		return -EINVAL;
688 
689 	md[0]->nb_segs = nb_seg;
690 	md[0]->pkt_len = tlen;
691 	md[0]->vlan_tci = m->vlan_tci;
692 	md[0]->vlan_tci_outer = m->vlan_tci_outer;
693 	md[0]->ol_flags = m->ol_flags;
694 	md[0]->tx_offload = m->tx_offload;
695 
696 	return nb_seg;
697 }
698 
699 /*
700  * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
701  * Copy packet contents and offload information into the new segmented mbuf.
702  */
703 static struct rte_mbuf *
704 pkt_copy_split(const struct rte_mbuf *pkt)
705 {
706 	int32_t n, rc;
707 	uint32_t i, len, nb_seg;
708 	struct rte_mempool *mp;
709 	uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
710 	struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
711 
712 	mp = current_fwd_lcore()->mbp;
713 
714 	if (tx_pkt_split == TX_PKT_SPLIT_RND)
715 		nb_seg = rte_rand() % tx_pkt_nb_segs + 1;
716 	else
717 		nb_seg = tx_pkt_nb_segs;
718 
719 	memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
720 
721 	/* calculate number of segments to use and their length. */
722 	len = 0;
723 	for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
724 		len += seglen[i];
725 		md[i] = NULL;
726 	}
727 
728 	n = pkt->pkt_len - len;
729 
730 	/* update size of the last segment to fit rest of the packet */
731 	if (n >= 0) {
732 		seglen[i - 1] += n;
733 		len += n;
734 	}
735 
736 	nb_seg = i;
737 	while (i != 0) {
738 		p = rte_pktmbuf_alloc(mp);
739 		if (p == NULL) {
740 			TESTPMD_LOG(ERR,
741 				"failed to allocate %u-th of %u mbuf "
742 				"from mempool: %s\n",
743 				nb_seg - i, nb_seg, mp->name);
744 			break;
745 		}
746 
747 		md[--i] = p;
748 		if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
749 			TESTPMD_LOG(ERR, "mempool %s, %u-th segment: "
750 				"expected seglen: %u, "
751 				"actual mbuf tailroom: %u\n",
752 				mp->name, i, seglen[i],
753 				rte_pktmbuf_tailroom(md[i]));
754 			break;
755 		}
756 	}
757 
758 	/* all mbufs successfully allocated, do copy */
759 	if (i == 0) {
760 		rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
761 		if (rc < 0)
762 			TESTPMD_LOG(ERR,
763 				"mbuf_copy_split for %p(len=%u, nb_seg=%u) "
764 				"into %u segments failed with error code: %d\n",
765 				pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
766 
767 		/* figure out how many mbufs to free. */
768 		i = RTE_MAX(rc, 0);
769 	}
770 
771 	/* free unused mbufs */
772 	for (; i != nb_seg; i++) {
773 		rte_pktmbuf_free_seg(md[i]);
774 		md[i] = NULL;
775 	}
776 
777 	return md[0];
778 }
779 
780 #if defined(RTE_LIB_GRO) || defined(RTE_LIB_GSO)
781 /*
782  * Re-calculate IP checksum for merged/fragmented packets.
783  */
784 static void
785 pkts_ip_csum_recalc(struct rte_mbuf **pkts_burst, const uint16_t nb_pkts, uint64_t tx_offloads)
786 {
787 	int i;
788 	struct rte_ipv4_hdr *ipv4_hdr;
789 	for (i = 0; i < nb_pkts; i++) {
790 		if ((pkts_burst[i]->ol_flags & RTE_MBUF_F_TX_IPV4) &&
791 			(tx_offloads & RTE_ETH_TX_OFFLOAD_IPV4_CKSUM) == 0) {
792 			ipv4_hdr = rte_pktmbuf_mtod_offset(pkts_burst[i],
793 						struct rte_ipv4_hdr *,
794 						pkts_burst[i]->l2_len);
795 			ipv4_hdr->hdr_checksum = 0;
796 			ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
797 		}
798 	}
799 }
800 #endif
801 
802 /*
803  * Receive a burst of packets, and for each packet:
804  *  - parse packet, and try to recognize a supported packet type (1)
805  *  - if it's not a supported packet type, don't touch the packet, else:
806  *  - reprocess the checksum of all supported layers. This is done in SW
807  *    or HW, depending on testpmd command line configuration
808  *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
809  *    segmentation offload (this implies HW TCP checksum)
810  * Then transmit packets on the output port.
811  *
812  * (1) Supported packets are:
813  *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
814  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
815  *           UDP|TCP|SCTP
816  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 /
817  *           UDP|TCP|SCTP
818  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 /
819  *           UDP|TCP|SCTP
820  *   Ether / (vlan) / outer IP / outer UDP / GTP / IP|IP6 / UDP|TCP|SCTP
821  *   Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
822  *   Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
823  *   Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
824  *
825  * The testpmd command line for this forward engine sets the flags
826  * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
827  * whether a checksum must be calculated in software or in hardware. The
828  * IP, UDP, TCP and SCTP flags always concern the inner layer. The
829  * OUTER_IP is only useful for tunnel packets.
830  */
831 static void
832 pkt_burst_checksum_forward(struct fwd_stream *fs)
833 {
834 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
835 #ifdef RTE_LIB_GSO
836 	struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST];
837 	struct rte_gso_ctx *gso_ctx;
838 #endif
839 	struct rte_mbuf **tx_pkts_burst;
840 	struct rte_port *txp;
841 	struct rte_mbuf *m, *p;
842 	struct rte_ether_hdr *eth_hdr;
843 	void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
844 #ifdef RTE_LIB_GRO
845 	void **gro_ctx;
846 	uint16_t gro_pkts_num;
847 	uint8_t gro_enable;
848 #endif
849 	uint16_t nb_rx;
850 	uint16_t nb_tx;
851 	uint16_t nb_prep;
852 	uint16_t i;
853 	uint64_t rx_ol_flags, tx_ol_flags;
854 	uint64_t tx_offloads;
855 	uint32_t retry;
856 	uint32_t rx_bad_ip_csum;
857 	uint32_t rx_bad_l4_csum;
858 	uint32_t rx_bad_outer_l4_csum;
859 	uint32_t rx_bad_outer_ip_csum;
860 	struct testpmd_offload_info info;
861 
862 	uint64_t start_tsc = 0;
863 
864 	get_start_cycles(&start_tsc);
865 
866 	/* receive a burst of packet */
867 	nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
868 				 nb_pkt_per_burst);
869 	inc_rx_burst_stats(fs, nb_rx);
870 	if (unlikely(nb_rx == 0))
871 		return;
872 
873 	fs->rx_packets += nb_rx;
874 	rx_bad_ip_csum = 0;
875 	rx_bad_l4_csum = 0;
876 	rx_bad_outer_l4_csum = 0;
877 	rx_bad_outer_ip_csum = 0;
878 #ifdef RTE_LIB_GRO
879 	gro_enable = gro_ports[fs->rx_port].enable;
880 #endif
881 
882 	txp = &ports[fs->tx_port];
883 	tx_offloads = txp->dev_conf.txmode.offloads;
884 	memset(&info, 0, sizeof(info));
885 	info.tso_segsz = txp->tso_segsz;
886 	info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
887 #ifdef RTE_LIB_GSO
888 	if (gso_ports[fs->tx_port].enable)
889 		info.gso_enable = 1;
890 #endif
891 
892 	for (i = 0; i < nb_rx; i++) {
893 		if (likely(i < nb_rx - 1))
894 			rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
895 						       void *));
896 
897 		m = pkts_burst[i];
898 		info.is_tunnel = 0;
899 		info.pkt_len = rte_pktmbuf_pkt_len(m);
900 		tx_ol_flags = m->ol_flags &
901 			      (RTE_MBUF_F_INDIRECT | RTE_MBUF_F_EXTERNAL);
902 		rx_ol_flags = m->ol_flags;
903 
904 		/* Update the L3/L4 checksum error packet statistics */
905 		if ((rx_ol_flags & RTE_MBUF_F_RX_IP_CKSUM_MASK) == RTE_MBUF_F_RX_IP_CKSUM_BAD)
906 			rx_bad_ip_csum += 1;
907 		if ((rx_ol_flags & RTE_MBUF_F_RX_L4_CKSUM_MASK) == RTE_MBUF_F_RX_L4_CKSUM_BAD)
908 			rx_bad_l4_csum += 1;
909 		if (rx_ol_flags & RTE_MBUF_F_RX_OUTER_L4_CKSUM_BAD)
910 			rx_bad_outer_l4_csum += 1;
911 		if (rx_ol_flags & RTE_MBUF_F_RX_OUTER_IP_CKSUM_BAD)
912 			rx_bad_outer_ip_csum += 1;
913 
914 		/* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
915 		 * and inner headers */
916 
917 		eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
918 		parse_ethernet(eth_hdr, &info);
919 		l3_hdr = (char *)eth_hdr + info.l2_len;
920 
921 		/* check if it's a supported tunnel */
922 		if (txp->parse_tunnel) {
923 			if (info.l4_proto == IPPROTO_UDP) {
924 				struct rte_udp_hdr *udp_hdr;
925 
926 				udp_hdr = (struct rte_udp_hdr *)
927 					((char *)l3_hdr + info.l3_len);
928 				parse_gtp(udp_hdr, &info);
929 				if (info.is_tunnel) {
930 					tx_ol_flags |= RTE_MBUF_F_TX_TUNNEL_GTP;
931 					goto tunnel_update;
932 				}
933 				parse_vxlan_gpe(udp_hdr, &info);
934 				if (info.is_tunnel) {
935 					tx_ol_flags |=
936 						RTE_MBUF_F_TX_TUNNEL_VXLAN_GPE;
937 					goto tunnel_update;
938 				}
939 				parse_vxlan(udp_hdr, &info);
940 				if (info.is_tunnel) {
941 					tx_ol_flags |=
942 						RTE_MBUF_F_TX_TUNNEL_VXLAN;
943 					goto tunnel_update;
944 				}
945 				parse_geneve(udp_hdr, &info);
946 				if (info.is_tunnel) {
947 					tx_ol_flags |=
948 						RTE_MBUF_F_TX_TUNNEL_GENEVE;
949 					goto tunnel_update;
950 				}
951 				/* Always keep last. */
952 				if (unlikely(RTE_ETH_IS_TUNNEL_PKT(
953 							m->packet_type) != 0)) {
954 					TESTPMD_LOG(DEBUG, "Unknown tunnel packet. UDP dst port: %hu",
955 						udp_hdr->dst_port);
956 				}
957 			} else if (info.l4_proto == IPPROTO_GRE) {
958 				struct simple_gre_hdr *gre_hdr;
959 
960 				gre_hdr = (struct simple_gre_hdr *)
961 					((char *)l3_hdr + info.l3_len);
962 				parse_gre(gre_hdr, &info);
963 				if (info.is_tunnel)
964 					tx_ol_flags |= RTE_MBUF_F_TX_TUNNEL_GRE;
965 			} else if (info.l4_proto == IPPROTO_IPIP) {
966 				void *encap_ip_hdr;
967 
968 				encap_ip_hdr = (char *)l3_hdr + info.l3_len;
969 				parse_encap_ip(encap_ip_hdr, &info);
970 				if (info.is_tunnel)
971 					tx_ol_flags |= RTE_MBUF_F_TX_TUNNEL_IPIP;
972 			}
973 		}
974 
975 tunnel_update:
976 		/* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
977 		if (info.is_tunnel) {
978 			outer_l3_hdr = l3_hdr;
979 			l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
980 		}
981 
982 		/* step 2: depending on user command line configuration,
983 		 * recompute checksum either in software or flag the
984 		 * mbuf to offload the calculation to the NIC. If TSO
985 		 * is configured, prepare the mbuf for TCP segmentation. */
986 
987 		/* process checksums of inner headers first */
988 		tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
989 			tx_offloads, m);
990 
991 		/* Then process outer headers if any. Note that the software
992 		 * checksum will be wrong if one of the inner checksums is
993 		 * processed in hardware. */
994 		if (info.is_tunnel == 1) {
995 			tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
996 					tx_offloads,
997 					!!(tx_ol_flags & RTE_MBUF_F_TX_TCP_SEG),
998 					m);
999 		}
1000 
1001 		/* step 3: fill the mbuf meta data (flags and header lengths) */
1002 
1003 		m->tx_offload = 0;
1004 		if (info.is_tunnel == 1) {
1005 			if (info.tunnel_tso_segsz ||
1006 			    (tx_offloads &
1007 			     RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
1008 			    (tx_offloads &
1009 			     RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM)) {
1010 				m->outer_l2_len = info.outer_l2_len;
1011 				m->outer_l3_len = info.outer_l3_len;
1012 				m->l2_len = info.l2_len;
1013 				m->l3_len = info.l3_len;
1014 				m->l4_len = info.l4_len;
1015 				m->tso_segsz = info.tunnel_tso_segsz;
1016 			}
1017 			else {
1018 				/* if there is a outer UDP cksum
1019 				   processed in sw and the inner in hw,
1020 				   the outer checksum will be wrong as
1021 				   the payload will be modified by the
1022 				   hardware */
1023 				m->l2_len = info.outer_l2_len +
1024 					info.outer_l3_len + info.l2_len;
1025 				m->l3_len = info.l3_len;
1026 				m->l4_len = info.l4_len;
1027 			}
1028 		} else {
1029 			/* this is only useful if an offload flag is
1030 			 * set, but it does not hurt to fill it in any
1031 			 * case */
1032 			m->l2_len = info.l2_len;
1033 			m->l3_len = info.l3_len;
1034 			m->l4_len = info.l4_len;
1035 			m->tso_segsz = info.tso_segsz;
1036 		}
1037 		m->ol_flags = tx_ol_flags;
1038 
1039 		/* Do split & copy for the packet. */
1040 		if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
1041 			p = pkt_copy_split(m);
1042 			if (p != NULL) {
1043 				rte_pktmbuf_free(m);
1044 				m = p;
1045 				pkts_burst[i] = m;
1046 			}
1047 		}
1048 
1049 		/* if verbose mode is enabled, dump debug info */
1050 		if (verbose_level > 0) {
1051 			char buf[256];
1052 
1053 			printf("-----------------\n");
1054 			printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
1055 				fs->rx_port, m, m->pkt_len, m->nb_segs);
1056 			/* dump rx parsed packet info */
1057 			rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
1058 			printf("rx: l2_len=%d ethertype=%x l3_len=%d "
1059 				"l4_proto=%d l4_len=%d flags=%s\n",
1060 				info.l2_len, rte_be_to_cpu_16(info.ethertype),
1061 				info.l3_len, info.l4_proto, info.l4_len, buf);
1062 			if (rx_ol_flags & RTE_MBUF_F_RX_LRO)
1063 				printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
1064 			if (info.is_tunnel == 1)
1065 				printf("rx: outer_l2_len=%d outer_ethertype=%x "
1066 					"outer_l3_len=%d\n", info.outer_l2_len,
1067 					rte_be_to_cpu_16(info.outer_ethertype),
1068 					info.outer_l3_len);
1069 			/* dump tx packet info */
1070 			if ((tx_offloads & (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
1071 					    RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
1072 					    RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
1073 					    RTE_ETH_TX_OFFLOAD_SCTP_CKSUM)) ||
1074 				info.tso_segsz != 0)
1075 				printf("tx: m->l2_len=%d m->l3_len=%d "
1076 					"m->l4_len=%d\n",
1077 					m->l2_len, m->l3_len, m->l4_len);
1078 			if (info.is_tunnel == 1) {
1079 				if ((tx_offloads &
1080 				    RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
1081 				    (tx_offloads &
1082 				    RTE_ETH_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
1083 				    (tx_ol_flags & RTE_MBUF_F_TX_OUTER_IPV6))
1084 					printf("tx: m->outer_l2_len=%d "
1085 						"m->outer_l3_len=%d\n",
1086 						m->outer_l2_len,
1087 						m->outer_l3_len);
1088 				if (info.tunnel_tso_segsz != 0 &&
1089 						(m->ol_flags & RTE_MBUF_F_TX_TCP_SEG))
1090 					printf("tx: m->tso_segsz=%d\n",
1091 						m->tso_segsz);
1092 			} else if (info.tso_segsz != 0 &&
1093 					(m->ol_flags & RTE_MBUF_F_TX_TCP_SEG))
1094 				printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
1095 			rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
1096 			printf("tx: flags=%s", buf);
1097 			printf("\n");
1098 		}
1099 	}
1100 
1101 #ifdef RTE_LIB_GRO
1102 	if (unlikely(gro_enable)) {
1103 		if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
1104 			nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx,
1105 					&(gro_ports[fs->rx_port].param));
1106 		} else {
1107 			gro_ctx = current_fwd_lcore()->gro_ctx;
1108 			nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx);
1109 
1110 			if (++fs->gro_times >= gro_flush_cycles) {
1111 				gro_pkts_num = rte_gro_get_pkt_count(gro_ctx);
1112 				if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
1113 					gro_pkts_num = MAX_PKT_BURST - nb_rx;
1114 
1115 				nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
1116 						RTE_GRO_TCP_IPV4,
1117 						&pkts_burst[nb_rx],
1118 						gro_pkts_num);
1119 				fs->gro_times = 0;
1120 			}
1121 		}
1122 
1123 		pkts_ip_csum_recalc(pkts_burst, nb_rx, tx_offloads);
1124 	}
1125 #endif
1126 
1127 #ifdef RTE_LIB_GSO
1128 	if (gso_ports[fs->tx_port].enable != 0) {
1129 		uint16_t nb_segments = 0;
1130 
1131 		gso_ctx = &(current_fwd_lcore()->gso_ctx);
1132 		gso_ctx->gso_size = gso_max_segment_size;
1133 		for (i = 0; i < nb_rx; i++) {
1134 			int ret;
1135 
1136 			ret = rte_gso_segment(pkts_burst[i], gso_ctx,
1137 					&gso_segments[nb_segments],
1138 					GSO_MAX_PKT_BURST - nb_segments);
1139 			if (ret >= 1) {
1140 				/* pkts_burst[i] can be freed safely here. */
1141 				rte_pktmbuf_free(pkts_burst[i]);
1142 				nb_segments += ret;
1143 			} else if (ret == 0) {
1144 				/* 0 means it can be transmitted directly
1145 				 * without gso.
1146 				 */
1147 				gso_segments[nb_segments] = pkts_burst[i];
1148 				nb_segments += 1;
1149 			} else {
1150 				TESTPMD_LOG(DEBUG, "Unable to segment packet");
1151 				rte_pktmbuf_free(pkts_burst[i]);
1152 			}
1153 		}
1154 
1155 		tx_pkts_burst = gso_segments;
1156 		nb_rx = nb_segments;
1157 
1158 		pkts_ip_csum_recalc(tx_pkts_burst, nb_rx, tx_offloads);
1159 	} else
1160 #endif
1161 		tx_pkts_burst = pkts_burst;
1162 
1163 	nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
1164 			tx_pkts_burst, nb_rx);
1165 	if (nb_prep != nb_rx)
1166 		fprintf(stderr,
1167 			"Preparing packet burst to transmit failed: %s\n",
1168 			rte_strerror(rte_errno));
1169 
1170 	nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
1171 			nb_prep);
1172 
1173 	/*
1174 	 * Retry if necessary
1175 	 */
1176 	if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
1177 		retry = 0;
1178 		while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
1179 			rte_delay_us(burst_tx_delay_time);
1180 			nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
1181 					&tx_pkts_burst[nb_tx], nb_rx - nb_tx);
1182 		}
1183 	}
1184 	fs->tx_packets += nb_tx;
1185 	fs->rx_bad_ip_csum += rx_bad_ip_csum;
1186 	fs->rx_bad_l4_csum += rx_bad_l4_csum;
1187 	fs->rx_bad_outer_l4_csum += rx_bad_outer_l4_csum;
1188 	fs->rx_bad_outer_ip_csum += rx_bad_outer_ip_csum;
1189 
1190 	inc_tx_burst_stats(fs, nb_tx);
1191 	if (unlikely(nb_tx < nb_rx)) {
1192 		fs->fwd_dropped += (nb_rx - nb_tx);
1193 		do {
1194 			rte_pktmbuf_free(tx_pkts_burst[nb_tx]);
1195 		} while (++nb_tx < nb_rx);
1196 	}
1197 
1198 	get_end_cycles(fs, start_tsc);
1199 }
1200 
1201 static void
1202 stream_init_checksum_forward(struct fwd_stream *fs)
1203 {
1204 	bool rx_stopped, tx_stopped;
1205 
1206 	rx_stopped = ports[fs->rx_port].rxq[fs->rx_queue].state ==
1207 						RTE_ETH_QUEUE_STATE_STOPPED;
1208 	tx_stopped = ports[fs->tx_port].txq[fs->tx_queue].state ==
1209 						RTE_ETH_QUEUE_STATE_STOPPED;
1210 	fs->disabled = rx_stopped || tx_stopped;
1211 }
1212 
1213 struct fwd_engine csum_fwd_engine = {
1214 	.fwd_mode_name  = "csum",
1215 	.port_fwd_begin = NULL,
1216 	.port_fwd_end   = NULL,
1217 	.stream_init    = stream_init_checksum_forward,
1218 	.packet_fwd     = pkt_burst_checksum_forward,
1219 };
1220