xref: /dpdk/app/test-pmd/csumonly.c (revision 1edccebcccdbe600dc0a3a418fae68336648a87e)
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_atomic.h>
28 #include <rte_branch_prediction.h>
29 #include <rte_mempool.h>
30 #include <rte_mbuf.h>
31 #include <rte_interrupts.h>
32 #include <rte_pci.h>
33 #include <rte_ether.h>
34 #include <rte_ethdev.h>
35 #include <rte_ip.h>
36 #include <rte_tcp.h>
37 #include <rte_udp.h>
38 #include <rte_sctp.h>
39 #include <rte_prefetch.h>
40 #include <rte_string_fns.h>
41 #include <rte_flow.h>
42 #include <rte_gro.h>
43 #include <rte_gso.h>
44 
45 #include "testpmd.h"
46 
47 #define IP_DEFTTL  64   /* from RFC 1340. */
48 #define IP_VERSION 0x40
49 #define IP_HDRLEN  0x05 /* default IP header length == five 32-bits words. */
50 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN)
51 
52 #define GRE_CHECKSUM_PRESENT	0x8000
53 #define GRE_KEY_PRESENT		0x2000
54 #define GRE_SEQUENCE_PRESENT	0x1000
55 #define GRE_EXT_LEN		4
56 #define GRE_SUPPORTED_FIELDS	(GRE_CHECKSUM_PRESENT | GRE_KEY_PRESENT |\
57 				 GRE_SEQUENCE_PRESENT)
58 
59 /* We cannot use rte_cpu_to_be_16() on a constant in a switch/case */
60 #if RTE_BYTE_ORDER == RTE_LITTLE_ENDIAN
61 #define _htons(x) ((uint16_t)((((x) & 0x00ffU) << 8) | (((x) & 0xff00U) >> 8)))
62 #else
63 #define _htons(x) (x)
64 #endif
65 
66 uint16_t vxlan_gpe_udp_port = 4790;
67 
68 /* structure that caches offload info for the current packet */
69 struct testpmd_offload_info {
70 	uint16_t ethertype;
71 	uint8_t gso_enable;
72 	uint16_t l2_len;
73 	uint16_t l3_len;
74 	uint16_t l4_len;
75 	uint8_t l4_proto;
76 	uint8_t is_tunnel;
77 	uint16_t outer_ethertype;
78 	uint16_t outer_l2_len;
79 	uint16_t outer_l3_len;
80 	uint8_t outer_l4_proto;
81 	uint16_t tso_segsz;
82 	uint16_t tunnel_tso_segsz;
83 	uint32_t pkt_len;
84 };
85 
86 /* simplified GRE header */
87 struct simple_gre_hdr {
88 	uint16_t flags;
89 	uint16_t proto;
90 } __attribute__((__packed__));
91 
92 static uint16_t
93 get_udptcp_checksum(void *l3_hdr, void *l4_hdr, uint16_t ethertype)
94 {
95 	if (ethertype == _htons(ETHER_TYPE_IPv4))
96 		return rte_ipv4_udptcp_cksum(l3_hdr, l4_hdr);
97 	else /* assume ethertype == ETHER_TYPE_IPv6 */
98 		return rte_ipv6_udptcp_cksum(l3_hdr, l4_hdr);
99 }
100 
101 /* Parse an IPv4 header to fill l3_len, l4_len, and l4_proto */
102 static void
103 parse_ipv4(struct ipv4_hdr *ipv4_hdr, struct testpmd_offload_info *info)
104 {
105 	struct tcp_hdr *tcp_hdr;
106 
107 	info->l3_len = (ipv4_hdr->version_ihl & 0x0f) * 4;
108 	info->l4_proto = ipv4_hdr->next_proto_id;
109 
110 	/* only fill l4_len for TCP, it's useful for TSO */
111 	if (info->l4_proto == IPPROTO_TCP) {
112 		tcp_hdr = (struct tcp_hdr *)((char *)ipv4_hdr + info->l3_len);
113 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
114 	} else
115 		info->l4_len = 0;
116 }
117 
118 /* Parse an IPv6 header to fill l3_len, l4_len, and l4_proto */
119 static void
120 parse_ipv6(struct ipv6_hdr *ipv6_hdr, struct testpmd_offload_info *info)
121 {
122 	struct tcp_hdr *tcp_hdr;
123 
124 	info->l3_len = sizeof(struct ipv6_hdr);
125 	info->l4_proto = ipv6_hdr->proto;
126 
127 	/* only fill l4_len for TCP, it's useful for TSO */
128 	if (info->l4_proto == IPPROTO_TCP) {
129 		tcp_hdr = (struct tcp_hdr *)((char *)ipv6_hdr + info->l3_len);
130 		info->l4_len = (tcp_hdr->data_off & 0xf0) >> 2;
131 	} else
132 		info->l4_len = 0;
133 }
134 
135 /*
136  * Parse an ethernet header to fill the ethertype, l2_len, l3_len and
137  * ipproto. This function is able to recognize IPv4/IPv6 with one optional vlan
138  * header. The l4_len argument is only set in case of TCP (useful for TSO).
139  */
140 static void
141 parse_ethernet(struct ether_hdr *eth_hdr, struct testpmd_offload_info *info)
142 {
143 	struct ipv4_hdr *ipv4_hdr;
144 	struct ipv6_hdr *ipv6_hdr;
145 
146 	info->l2_len = sizeof(struct ether_hdr);
147 	info->ethertype = eth_hdr->ether_type;
148 
149 	if (info->ethertype == _htons(ETHER_TYPE_VLAN)) {
150 		struct vlan_hdr *vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
151 
152 		info->l2_len  += sizeof(struct vlan_hdr);
153 		info->ethertype = vlan_hdr->eth_proto;
154 	}
155 
156 	switch (info->ethertype) {
157 	case _htons(ETHER_TYPE_IPv4):
158 		ipv4_hdr = (struct ipv4_hdr *) ((char *)eth_hdr + info->l2_len);
159 		parse_ipv4(ipv4_hdr, info);
160 		break;
161 	case _htons(ETHER_TYPE_IPv6):
162 		ipv6_hdr = (struct ipv6_hdr *) ((char *)eth_hdr + info->l2_len);
163 		parse_ipv6(ipv6_hdr, info);
164 		break;
165 	default:
166 		info->l4_len = 0;
167 		info->l3_len = 0;
168 		info->l4_proto = 0;
169 		break;
170 	}
171 }
172 
173 /* Parse a vxlan header */
174 static void
175 parse_vxlan(struct udp_hdr *udp_hdr,
176 	    struct testpmd_offload_info *info,
177 	    uint32_t pkt_type)
178 {
179 	struct ether_hdr *eth_hdr;
180 
181 	/* check udp destination port, 4789 is the default vxlan port
182 	 * (rfc7348) or that the rx offload flag is set (i40e only
183 	 * currently) */
184 	if (udp_hdr->dst_port != _htons(4789) &&
185 		RTE_ETH_IS_TUNNEL_PKT(pkt_type) == 0)
186 		return;
187 
188 	info->is_tunnel = 1;
189 	info->outer_ethertype = info->ethertype;
190 	info->outer_l2_len = info->l2_len;
191 	info->outer_l3_len = info->l3_len;
192 	info->outer_l4_proto = info->l4_proto;
193 
194 	eth_hdr = (struct ether_hdr *)((char *)udp_hdr +
195 		sizeof(struct udp_hdr) +
196 		sizeof(struct vxlan_hdr));
197 
198 	parse_ethernet(eth_hdr, info);
199 	info->l2_len += ETHER_VXLAN_HLEN; /* add udp + vxlan */
200 }
201 
202 /* Parse a vxlan-gpe header */
203 static void
204 parse_vxlan_gpe(struct udp_hdr *udp_hdr,
205 	    struct testpmd_offload_info *info)
206 {
207 	struct ether_hdr *eth_hdr;
208 	struct ipv4_hdr *ipv4_hdr;
209 	struct ipv6_hdr *ipv6_hdr;
210 	struct vxlan_gpe_hdr *vxlan_gpe_hdr;
211 	uint8_t vxlan_gpe_len = sizeof(*vxlan_gpe_hdr);
212 
213 	/* Check udp destination port. */
214 	if (udp_hdr->dst_port != _htons(vxlan_gpe_udp_port))
215 		return;
216 
217 	vxlan_gpe_hdr = (struct vxlan_gpe_hdr *)((char *)udp_hdr +
218 				sizeof(struct udp_hdr));
219 
220 	if (!vxlan_gpe_hdr->proto || vxlan_gpe_hdr->proto ==
221 	    VXLAN_GPE_TYPE_IPV4) {
222 		info->is_tunnel = 1;
223 		info->outer_ethertype = info->ethertype;
224 		info->outer_l2_len = info->l2_len;
225 		info->outer_l3_len = info->l3_len;
226 		info->outer_l4_proto = info->l4_proto;
227 
228 		ipv4_hdr = (struct ipv4_hdr *)((char *)vxlan_gpe_hdr +
229 			   vxlan_gpe_len);
230 
231 		parse_ipv4(ipv4_hdr, info);
232 		info->ethertype = _htons(ETHER_TYPE_IPv4);
233 		info->l2_len = 0;
234 
235 	} else if (vxlan_gpe_hdr->proto == VXLAN_GPE_TYPE_IPV6) {
236 		info->is_tunnel = 1;
237 		info->outer_ethertype = info->ethertype;
238 		info->outer_l2_len = info->l2_len;
239 		info->outer_l3_len = info->l3_len;
240 		info->outer_l4_proto = info->l4_proto;
241 
242 		ipv6_hdr = (struct ipv6_hdr *)((char *)vxlan_gpe_hdr +
243 			   vxlan_gpe_len);
244 
245 		info->ethertype = _htons(ETHER_TYPE_IPv6);
246 		parse_ipv6(ipv6_hdr, info);
247 		info->l2_len = 0;
248 
249 	} else if (vxlan_gpe_hdr->proto == VXLAN_GPE_TYPE_ETH) {
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 		eth_hdr = (struct ether_hdr *)((char *)vxlan_gpe_hdr +
257 			  vxlan_gpe_len);
258 
259 		parse_ethernet(eth_hdr, info);
260 	} else
261 		return;
262 
263 	info->l2_len += ETHER_VXLAN_GPE_HLEN;
264 }
265 
266 /* Parse a gre header */
267 static void
268 parse_gre(struct simple_gre_hdr *gre_hdr, struct testpmd_offload_info *info)
269 {
270 	struct ether_hdr *eth_hdr;
271 	struct ipv4_hdr *ipv4_hdr;
272 	struct ipv6_hdr *ipv6_hdr;
273 	uint8_t gre_len = 0;
274 
275 	gre_len += sizeof(struct simple_gre_hdr);
276 
277 	if (gre_hdr->flags & _htons(GRE_KEY_PRESENT))
278 		gre_len += GRE_EXT_LEN;
279 	if (gre_hdr->flags & _htons(GRE_SEQUENCE_PRESENT))
280 		gre_len += GRE_EXT_LEN;
281 	if (gre_hdr->flags & _htons(GRE_CHECKSUM_PRESENT))
282 		gre_len += GRE_EXT_LEN;
283 
284 	if (gre_hdr->proto == _htons(ETHER_TYPE_IPv4)) {
285 		info->is_tunnel = 1;
286 		info->outer_ethertype = info->ethertype;
287 		info->outer_l2_len = info->l2_len;
288 		info->outer_l3_len = info->l3_len;
289 		info->outer_l4_proto = info->l4_proto;
290 
291 		ipv4_hdr = (struct ipv4_hdr *)((char *)gre_hdr + gre_len);
292 
293 		parse_ipv4(ipv4_hdr, info);
294 		info->ethertype = _htons(ETHER_TYPE_IPv4);
295 		info->l2_len = 0;
296 
297 	} else if (gre_hdr->proto == _htons(ETHER_TYPE_IPv6)) {
298 		info->is_tunnel = 1;
299 		info->outer_ethertype = info->ethertype;
300 		info->outer_l2_len = info->l2_len;
301 		info->outer_l3_len = info->l3_len;
302 		info->outer_l4_proto = info->l4_proto;
303 
304 		ipv6_hdr = (struct ipv6_hdr *)((char *)gre_hdr + gre_len);
305 
306 		info->ethertype = _htons(ETHER_TYPE_IPv6);
307 		parse_ipv6(ipv6_hdr, info);
308 		info->l2_len = 0;
309 
310 	} else if (gre_hdr->proto == _htons(ETHER_TYPE_TEB)) {
311 		info->is_tunnel = 1;
312 		info->outer_ethertype = info->ethertype;
313 		info->outer_l2_len = info->l2_len;
314 		info->outer_l3_len = info->l3_len;
315 		info->outer_l4_proto = info->l4_proto;
316 
317 		eth_hdr = (struct ether_hdr *)((char *)gre_hdr + gre_len);
318 
319 		parse_ethernet(eth_hdr, info);
320 	} else
321 		return;
322 
323 	info->l2_len += gre_len;
324 }
325 
326 
327 /* Parse an encapsulated ip or ipv6 header */
328 static void
329 parse_encap_ip(void *encap_ip, struct testpmd_offload_info *info)
330 {
331 	struct ipv4_hdr *ipv4_hdr = encap_ip;
332 	struct ipv6_hdr *ipv6_hdr = encap_ip;
333 	uint8_t ip_version;
334 
335 	ip_version = (ipv4_hdr->version_ihl & 0xf0) >> 4;
336 
337 	if (ip_version != 4 && ip_version != 6)
338 		return;
339 
340 	info->is_tunnel = 1;
341 	info->outer_ethertype = info->ethertype;
342 	info->outer_l2_len = info->l2_len;
343 	info->outer_l3_len = info->l3_len;
344 
345 	if (ip_version == 4) {
346 		parse_ipv4(ipv4_hdr, info);
347 		info->ethertype = _htons(ETHER_TYPE_IPv4);
348 	} else {
349 		parse_ipv6(ipv6_hdr, info);
350 		info->ethertype = _htons(ETHER_TYPE_IPv6);
351 	}
352 	info->l2_len = 0;
353 }
354 
355 /* if possible, calculate the checksum of a packet in hw or sw,
356  * depending on the testpmd command line configuration */
357 static uint64_t
358 process_inner_cksums(void *l3_hdr, const struct testpmd_offload_info *info,
359 	uint64_t tx_offloads)
360 {
361 	struct ipv4_hdr *ipv4_hdr = l3_hdr;
362 	struct udp_hdr *udp_hdr;
363 	struct tcp_hdr *tcp_hdr;
364 	struct sctp_hdr *sctp_hdr;
365 	uint64_t ol_flags = 0;
366 	uint32_t max_pkt_len, tso_segsz = 0;
367 
368 	/* ensure packet is large enough to require tso */
369 	if (!info->is_tunnel) {
370 		max_pkt_len = info->l2_len + info->l3_len + info->l4_len +
371 			info->tso_segsz;
372 		if (info->tso_segsz != 0 && info->pkt_len > max_pkt_len)
373 			tso_segsz = info->tso_segsz;
374 	} else {
375 		max_pkt_len = info->outer_l2_len + info->outer_l3_len +
376 			info->l2_len + info->l3_len + info->l4_len +
377 			info->tunnel_tso_segsz;
378 		if (info->tunnel_tso_segsz != 0 && info->pkt_len > max_pkt_len)
379 			tso_segsz = info->tunnel_tso_segsz;
380 	}
381 
382 	if (info->ethertype == _htons(ETHER_TYPE_IPv4)) {
383 		ipv4_hdr = l3_hdr;
384 		ipv4_hdr->hdr_checksum = 0;
385 
386 		ol_flags |= PKT_TX_IPV4;
387 		if (info->l4_proto == IPPROTO_TCP && tso_segsz) {
388 			ol_flags |= PKT_TX_IP_CKSUM;
389 		} else {
390 			if (tx_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
391 				ol_flags |= PKT_TX_IP_CKSUM;
392 			else
393 				ipv4_hdr->hdr_checksum =
394 					rte_ipv4_cksum(ipv4_hdr);
395 		}
396 	} else if (info->ethertype == _htons(ETHER_TYPE_IPv6))
397 		ol_flags |= PKT_TX_IPV6;
398 	else
399 		return 0; /* packet type not supported, nothing to do */
400 
401 	if (info->l4_proto == IPPROTO_UDP) {
402 		udp_hdr = (struct udp_hdr *)((char *)l3_hdr + info->l3_len);
403 		/* do not recalculate udp cksum if it was 0 */
404 		if (udp_hdr->dgram_cksum != 0) {
405 			udp_hdr->dgram_cksum = 0;
406 			if (tx_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)
407 				ol_flags |= PKT_TX_UDP_CKSUM;
408 			else {
409 				udp_hdr->dgram_cksum =
410 					get_udptcp_checksum(l3_hdr, udp_hdr,
411 						info->ethertype);
412 			}
413 		}
414 		if (info->gso_enable)
415 			ol_flags |= PKT_TX_UDP_SEG;
416 	} else if (info->l4_proto == IPPROTO_TCP) {
417 		tcp_hdr = (struct tcp_hdr *)((char *)l3_hdr + info->l3_len);
418 		tcp_hdr->cksum = 0;
419 		if (tso_segsz)
420 			ol_flags |= PKT_TX_TCP_SEG;
421 		else if (tx_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)
422 			ol_flags |= PKT_TX_TCP_CKSUM;
423 		else {
424 			tcp_hdr->cksum =
425 				get_udptcp_checksum(l3_hdr, tcp_hdr,
426 					info->ethertype);
427 		}
428 		if (info->gso_enable)
429 			ol_flags |= PKT_TX_TCP_SEG;
430 	} else if (info->l4_proto == IPPROTO_SCTP) {
431 		sctp_hdr = (struct sctp_hdr *)((char *)l3_hdr + info->l3_len);
432 		sctp_hdr->cksum = 0;
433 		/* sctp payload must be a multiple of 4 to be
434 		 * offloaded */
435 		if ((tx_offloads & DEV_TX_OFFLOAD_SCTP_CKSUM) &&
436 			((ipv4_hdr->total_length & 0x3) == 0)) {
437 			ol_flags |= PKT_TX_SCTP_CKSUM;
438 		} else {
439 			/* XXX implement CRC32c, example available in
440 			 * RFC3309 */
441 		}
442 	}
443 
444 	return ol_flags;
445 }
446 
447 /* Calculate the checksum of outer header */
448 static uint64_t
449 process_outer_cksums(void *outer_l3_hdr, struct testpmd_offload_info *info,
450 	uint64_t tx_offloads, int tso_enabled)
451 {
452 	struct ipv4_hdr *ipv4_hdr = outer_l3_hdr;
453 	struct ipv6_hdr *ipv6_hdr = outer_l3_hdr;
454 	struct udp_hdr *udp_hdr;
455 	uint64_t ol_flags = 0;
456 
457 	if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4)) {
458 		ipv4_hdr->hdr_checksum = 0;
459 		ol_flags |= PKT_TX_OUTER_IPV4;
460 
461 		if (tx_offloads	& DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)
462 			ol_flags |= PKT_TX_OUTER_IP_CKSUM;
463 		else
464 			ipv4_hdr->hdr_checksum = rte_ipv4_cksum(ipv4_hdr);
465 	} else
466 		ol_flags |= PKT_TX_OUTER_IPV6;
467 
468 	if (info->outer_l4_proto != IPPROTO_UDP)
469 		return ol_flags;
470 
471 	/* Skip SW outer UDP checksum generation if HW supports it */
472 	if (tx_offloads & DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) {
473 		ol_flags |= PKT_TX_OUTER_UDP_CKSUM;
474 		return ol_flags;
475 	}
476 
477 	udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
478 
479 	/* outer UDP checksum is done in software. In the other side, for
480 	 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
481 	 * set to zero.
482 	 *
483 	 * If a packet will be TSOed into small packets by NIC, we cannot
484 	 * set/calculate a non-zero checksum, because it will be a wrong
485 	 * value after the packet be split into several small packets.
486 	 */
487 	if (tso_enabled)
488 		udp_hdr->dgram_cksum = 0;
489 
490 	/* do not recalculate udp cksum if it was 0 */
491 	if (udp_hdr->dgram_cksum != 0) {
492 		udp_hdr->dgram_cksum = 0;
493 		if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
494 			udp_hdr->dgram_cksum =
495 				rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
496 		else
497 			udp_hdr->dgram_cksum =
498 				rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
499 	}
500 
501 	return ol_flags;
502 }
503 
504 /*
505  * Helper function.
506  * Performs actual copying.
507  * Returns number of segments in the destination mbuf on success,
508  * or negative error code on failure.
509  */
510 static int
511 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
512 	uint16_t seglen[], uint8_t nb_seg)
513 {
514 	uint32_t dlen, slen, tlen;
515 	uint32_t i, len;
516 	const struct rte_mbuf *m;
517 	const uint8_t *src;
518 	uint8_t *dst;
519 
520 	dlen = 0;
521 	slen = 0;
522 	tlen = 0;
523 
524 	dst = NULL;
525 	src = NULL;
526 
527 	m = ms;
528 	i = 0;
529 	while (ms != NULL && i != nb_seg) {
530 
531 		if (slen == 0) {
532 			slen = rte_pktmbuf_data_len(ms);
533 			src = rte_pktmbuf_mtod(ms, const uint8_t *);
534 		}
535 
536 		if (dlen == 0) {
537 			dlen = RTE_MIN(seglen[i], slen);
538 			md[i]->data_len = dlen;
539 			md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
540 			dst = rte_pktmbuf_mtod(md[i], uint8_t *);
541 		}
542 
543 		len = RTE_MIN(slen, dlen);
544 		memcpy(dst, src, len);
545 		tlen += len;
546 		slen -= len;
547 		dlen -= len;
548 		src += len;
549 		dst += len;
550 
551 		if (slen == 0)
552 			ms = ms->next;
553 		if (dlen == 0)
554 			i++;
555 	}
556 
557 	if (ms != NULL)
558 		return -ENOBUFS;
559 	else if (tlen != m->pkt_len)
560 		return -EINVAL;
561 
562 	md[0]->nb_segs = nb_seg;
563 	md[0]->pkt_len = tlen;
564 	md[0]->vlan_tci = m->vlan_tci;
565 	md[0]->vlan_tci_outer = m->vlan_tci_outer;
566 	md[0]->ol_flags = m->ol_flags;
567 	md[0]->tx_offload = m->tx_offload;
568 
569 	return nb_seg;
570 }
571 
572 /*
573  * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
574  * Copy packet contents and offload information into then new segmented mbuf.
575  */
576 static struct rte_mbuf *
577 pkt_copy_split(const struct rte_mbuf *pkt)
578 {
579 	int32_t n, rc;
580 	uint32_t i, len, nb_seg;
581 	struct rte_mempool *mp;
582 	uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
583 	struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
584 
585 	mp = current_fwd_lcore()->mbp;
586 
587 	if (tx_pkt_split == TX_PKT_SPLIT_RND)
588 		nb_seg = random() % tx_pkt_nb_segs + 1;
589 	else
590 		nb_seg = tx_pkt_nb_segs;
591 
592 	memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
593 
594 	/* calculate number of segments to use and their length. */
595 	len = 0;
596 	for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
597 		len += seglen[i];
598 		md[i] = NULL;
599 	}
600 
601 	n = pkt->pkt_len - len;
602 
603 	/* update size of the last segment to fit rest of the packet */
604 	if (n >= 0) {
605 		seglen[i - 1] += n;
606 		len += n;
607 	}
608 
609 	nb_seg = i;
610 	while (i != 0) {
611 		p = rte_pktmbuf_alloc(mp);
612 		if (p == NULL) {
613 			TESTPMD_LOG(ERR,
614 				"failed to allocate %u-th of %u mbuf "
615 				"from mempool: %s\n",
616 				nb_seg - i, nb_seg, mp->name);
617 			break;
618 		}
619 
620 		md[--i] = p;
621 		if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
622 			TESTPMD_LOG(ERR, "mempool %s, %u-th segment: "
623 				"expected seglen: %u, "
624 				"actual mbuf tailroom: %u\n",
625 				mp->name, i, seglen[i],
626 				rte_pktmbuf_tailroom(md[i]));
627 			break;
628 		}
629 	}
630 
631 	/* all mbufs successfully allocated, do copy */
632 	if (i == 0) {
633 		rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
634 		if (rc < 0)
635 			TESTPMD_LOG(ERR,
636 				"mbuf_copy_split for %p(len=%u, nb_seg=%u) "
637 				"into %u segments failed with error code: %d\n",
638 				pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
639 
640 		/* figure out how many mbufs to free. */
641 		i = RTE_MAX(rc, 0);
642 	}
643 
644 	/* free unused mbufs */
645 	for (; i != nb_seg; i++) {
646 		rte_pktmbuf_free_seg(md[i]);
647 		md[i] = NULL;
648 	}
649 
650 	return md[0];
651 }
652 
653 /*
654  * Receive a burst of packets, and for each packet:
655  *  - parse packet, and try to recognize a supported packet type (1)
656  *  - if it's not a supported packet type, don't touch the packet, else:
657  *  - reprocess the checksum of all supported layers. This is done in SW
658  *    or HW, depending on testpmd command line configuration
659  *  - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
660  *    segmentation offload (this implies HW TCP checksum)
661  * Then transmit packets on the output port.
662  *
663  * (1) Supported packets are:
664  *   Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
665  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
666  *           UDP|TCP|SCTP
667  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 /
668  *           UDP|TCP|SCTP
669  *   Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 /
670  *           UDP|TCP|SCTP
671  *   Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
672  *   Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
673  *   Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
674  *
675  * The testpmd command line for this forward engine sets the flags
676  * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
677  * wether a checksum must be calculated in software or in hardware. The
678  * IP, UDP, TCP and SCTP flags always concern the inner layer. The
679  * OUTER_IP is only useful for tunnel packets.
680  */
681 static void
682 pkt_burst_checksum_forward(struct fwd_stream *fs)
683 {
684 	struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
685 	struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST];
686 	struct rte_gso_ctx *gso_ctx;
687 	struct rte_mbuf **tx_pkts_burst;
688 	struct rte_port *txp;
689 	struct rte_mbuf *m, *p;
690 	struct ether_hdr *eth_hdr;
691 	void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
692 	void **gro_ctx;
693 	uint16_t gro_pkts_num;
694 	uint8_t gro_enable;
695 	uint16_t nb_rx;
696 	uint16_t nb_tx;
697 	uint16_t nb_prep;
698 	uint16_t i;
699 	uint64_t rx_ol_flags, tx_ol_flags;
700 	uint64_t tx_offloads;
701 	uint32_t retry;
702 	uint32_t rx_bad_ip_csum;
703 	uint32_t rx_bad_l4_csum;
704 	uint32_t rx_bad_outer_l4_csum;
705 	struct testpmd_offload_info info;
706 	uint16_t nb_segments = 0;
707 	int ret;
708 
709 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
710 	uint64_t start_tsc;
711 	uint64_t end_tsc;
712 	uint64_t core_cycles;
713 #endif
714 
715 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
716 	start_tsc = rte_rdtsc();
717 #endif
718 
719 	/* receive a burst of packet */
720 	nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
721 				 nb_pkt_per_burst);
722 	if (unlikely(nb_rx == 0))
723 		return;
724 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
725 	fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
726 #endif
727 	fs->rx_packets += nb_rx;
728 	rx_bad_ip_csum = 0;
729 	rx_bad_l4_csum = 0;
730 	rx_bad_outer_l4_csum = 0;
731 	gro_enable = gro_ports[fs->rx_port].enable;
732 
733 	txp = &ports[fs->tx_port];
734 	tx_offloads = txp->dev_conf.txmode.offloads;
735 	memset(&info, 0, sizeof(info));
736 	info.tso_segsz = txp->tso_segsz;
737 	info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
738 	if (gso_ports[fs->tx_port].enable)
739 		info.gso_enable = 1;
740 
741 	for (i = 0; i < nb_rx; i++) {
742 		if (likely(i < nb_rx - 1))
743 			rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
744 						       void *));
745 
746 		m = pkts_burst[i];
747 		info.is_tunnel = 0;
748 		info.pkt_len = rte_pktmbuf_pkt_len(m);
749 		tx_ol_flags = m->ol_flags &
750 			      (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF);
751 		rx_ol_flags = m->ol_flags;
752 
753 		/* Update the L3/L4 checksum error packet statistics */
754 		if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
755 			rx_bad_ip_csum += 1;
756 		if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
757 			rx_bad_l4_csum += 1;
758 		if (rx_ol_flags & PKT_RX_OUTER_L4_CKSUM_BAD)
759 			rx_bad_outer_l4_csum += 1;
760 
761 		/* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
762 		 * and inner headers */
763 
764 		eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
765 		ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
766 				&eth_hdr->d_addr);
767 		ether_addr_copy(&ports[fs->tx_port].eth_addr,
768 				&eth_hdr->s_addr);
769 		parse_ethernet(eth_hdr, &info);
770 		l3_hdr = (char *)eth_hdr + info.l2_len;
771 
772 		/* check if it's a supported tunnel */
773 		if (txp->parse_tunnel) {
774 			if (info.l4_proto == IPPROTO_UDP) {
775 				struct udp_hdr *udp_hdr;
776 
777 				udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
778 					info.l3_len);
779 				parse_vxlan_gpe(udp_hdr, &info);
780 				if (info.is_tunnel) {
781 					tx_ol_flags |= PKT_TX_TUNNEL_VXLAN_GPE;
782 				} else {
783 					parse_vxlan(udp_hdr, &info,
784 						    m->packet_type);
785 					if (info.is_tunnel)
786 						tx_ol_flags |=
787 							PKT_TX_TUNNEL_VXLAN;
788 				}
789 			} else if (info.l4_proto == IPPROTO_GRE) {
790 				struct simple_gre_hdr *gre_hdr;
791 
792 				gre_hdr = (struct simple_gre_hdr *)
793 					((char *)l3_hdr + info.l3_len);
794 				parse_gre(gre_hdr, &info);
795 				if (info.is_tunnel)
796 					tx_ol_flags |= PKT_TX_TUNNEL_GRE;
797 			} else if (info.l4_proto == IPPROTO_IPIP) {
798 				void *encap_ip_hdr;
799 
800 				encap_ip_hdr = (char *)l3_hdr + info.l3_len;
801 				parse_encap_ip(encap_ip_hdr, &info);
802 				if (info.is_tunnel)
803 					tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
804 			}
805 		}
806 
807 		/* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
808 		if (info.is_tunnel) {
809 			outer_l3_hdr = l3_hdr;
810 			l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
811 		}
812 
813 		/* step 2: depending on user command line configuration,
814 		 * recompute checksum either in software or flag the
815 		 * mbuf to offload the calculation to the NIC. If TSO
816 		 * is configured, prepare the mbuf for TCP segmentation. */
817 
818 		/* process checksums of inner headers first */
819 		tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
820 			tx_offloads);
821 
822 		/* Then process outer headers if any. Note that the software
823 		 * checksum will be wrong if one of the inner checksums is
824 		 * processed in hardware. */
825 		if (info.is_tunnel == 1) {
826 			tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
827 					tx_offloads,
828 					!!(tx_ol_flags & PKT_TX_TCP_SEG));
829 		}
830 
831 		/* step 3: fill the mbuf meta data (flags and header lengths) */
832 
833 		m->tx_offload = 0;
834 		if (info.is_tunnel == 1) {
835 			if (info.tunnel_tso_segsz ||
836 			    (tx_offloads &
837 			     DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
838 			    (tx_offloads &
839 			     DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
840 			    (tx_ol_flags & PKT_TX_OUTER_IPV6)) {
841 				m->outer_l2_len = info.outer_l2_len;
842 				m->outer_l3_len = info.outer_l3_len;
843 				m->l2_len = info.l2_len;
844 				m->l3_len = info.l3_len;
845 				m->l4_len = info.l4_len;
846 				m->tso_segsz = info.tunnel_tso_segsz;
847 			}
848 			else {
849 				/* if there is a outer UDP cksum
850 				   processed in sw and the inner in hw,
851 				   the outer checksum will be wrong as
852 				   the payload will be modified by the
853 				   hardware */
854 				m->l2_len = info.outer_l2_len +
855 					info.outer_l3_len + info.l2_len;
856 				m->l3_len = info.l3_len;
857 				m->l4_len = info.l4_len;
858 			}
859 		} else {
860 			/* this is only useful if an offload flag is
861 			 * set, but it does not hurt to fill it in any
862 			 * case */
863 			m->l2_len = info.l2_len;
864 			m->l3_len = info.l3_len;
865 			m->l4_len = info.l4_len;
866 			m->tso_segsz = info.tso_segsz;
867 		}
868 		m->ol_flags = tx_ol_flags;
869 
870 		/* Do split & copy for the packet. */
871 		if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
872 			p = pkt_copy_split(m);
873 			if (p != NULL) {
874 				rte_pktmbuf_free(m);
875 				m = p;
876 				pkts_burst[i] = m;
877 			}
878 		}
879 
880 		/* if verbose mode is enabled, dump debug info */
881 		if (verbose_level > 0) {
882 			char buf[256];
883 
884 			printf("-----------------\n");
885 			printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
886 				fs->rx_port, m, m->pkt_len, m->nb_segs);
887 			/* dump rx parsed packet info */
888 			rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
889 			printf("rx: l2_len=%d ethertype=%x l3_len=%d "
890 				"l4_proto=%d l4_len=%d flags=%s\n",
891 				info.l2_len, rte_be_to_cpu_16(info.ethertype),
892 				info.l3_len, info.l4_proto, info.l4_len, buf);
893 			if (rx_ol_flags & PKT_RX_LRO)
894 				printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
895 			if (info.is_tunnel == 1)
896 				printf("rx: outer_l2_len=%d outer_ethertype=%x "
897 					"outer_l3_len=%d\n", info.outer_l2_len,
898 					rte_be_to_cpu_16(info.outer_ethertype),
899 					info.outer_l3_len);
900 			/* dump tx packet info */
901 			if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
902 					    DEV_TX_OFFLOAD_UDP_CKSUM |
903 					    DEV_TX_OFFLOAD_TCP_CKSUM |
904 					    DEV_TX_OFFLOAD_SCTP_CKSUM)) ||
905 				info.tso_segsz != 0)
906 				printf("tx: m->l2_len=%d m->l3_len=%d "
907 					"m->l4_len=%d\n",
908 					m->l2_len, m->l3_len, m->l4_len);
909 			if (info.is_tunnel == 1) {
910 				if ((tx_offloads &
911 				    DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
912 				    (tx_offloads &
913 				    DEV_TX_OFFLOAD_OUTER_UDP_CKSUM) ||
914 				    (tx_ol_flags & PKT_TX_OUTER_IPV6))
915 					printf("tx: m->outer_l2_len=%d "
916 						"m->outer_l3_len=%d\n",
917 						m->outer_l2_len,
918 						m->outer_l3_len);
919 				if (info.tunnel_tso_segsz != 0 &&
920 						(m->ol_flags & PKT_TX_TCP_SEG))
921 					printf("tx: m->tso_segsz=%d\n",
922 						m->tso_segsz);
923 			} else if (info.tso_segsz != 0 &&
924 					(m->ol_flags & PKT_TX_TCP_SEG))
925 				printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
926 			rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
927 			printf("tx: flags=%s", buf);
928 			printf("\n");
929 		}
930 	}
931 
932 	if (unlikely(gro_enable)) {
933 		if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
934 			nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx,
935 					&(gro_ports[fs->rx_port].param));
936 		} else {
937 			gro_ctx = current_fwd_lcore()->gro_ctx;
938 			nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx);
939 
940 			if (++fs->gro_times >= gro_flush_cycles) {
941 				gro_pkts_num = rte_gro_get_pkt_count(gro_ctx);
942 				if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
943 					gro_pkts_num = MAX_PKT_BURST - nb_rx;
944 
945 				nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
946 						RTE_GRO_TCP_IPV4,
947 						&pkts_burst[nb_rx],
948 						gro_pkts_num);
949 				fs->gro_times = 0;
950 			}
951 		}
952 	}
953 
954 	if (gso_ports[fs->tx_port].enable == 0)
955 		tx_pkts_burst = pkts_burst;
956 	else {
957 		gso_ctx = &(current_fwd_lcore()->gso_ctx);
958 		gso_ctx->gso_size = gso_max_segment_size;
959 		for (i = 0; i < nb_rx; i++) {
960 			ret = rte_gso_segment(pkts_burst[i], gso_ctx,
961 					&gso_segments[nb_segments],
962 					GSO_MAX_PKT_BURST - nb_segments);
963 			if (ret >= 0)
964 				nb_segments += ret;
965 			else {
966 				TESTPMD_LOG(DEBUG, "Unable to segment packet");
967 				rte_pktmbuf_free(pkts_burst[i]);
968 			}
969 		}
970 
971 		tx_pkts_burst = gso_segments;
972 		nb_rx = nb_segments;
973 	}
974 
975 	nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
976 			tx_pkts_burst, nb_rx);
977 	if (nb_prep != nb_rx)
978 		printf("Preparing packet burst to transmit failed: %s\n",
979 				rte_strerror(rte_errno));
980 
981 	nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
982 			nb_prep);
983 
984 	/*
985 	 * Retry if necessary
986 	 */
987 	if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
988 		retry = 0;
989 		while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
990 			rte_delay_us(burst_tx_delay_time);
991 			nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
992 					&tx_pkts_burst[nb_tx], nb_rx - nb_tx);
993 		}
994 	}
995 	fs->tx_packets += nb_tx;
996 	fs->rx_bad_ip_csum += rx_bad_ip_csum;
997 	fs->rx_bad_l4_csum += rx_bad_l4_csum;
998 	fs->rx_bad_outer_l4_csum += rx_bad_outer_l4_csum;
999 
1000 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
1001 	fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
1002 #endif
1003 	if (unlikely(nb_tx < nb_rx)) {
1004 		fs->fwd_dropped += (nb_rx - nb_tx);
1005 		do {
1006 			rte_pktmbuf_free(tx_pkts_burst[nb_tx]);
1007 		} while (++nb_tx < nb_rx);
1008 	}
1009 
1010 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
1011 	end_tsc = rte_rdtsc();
1012 	core_cycles = (end_tsc - start_tsc);
1013 	fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
1014 #endif
1015 }
1016 
1017 struct fwd_engine csum_fwd_engine = {
1018 	.fwd_mode_name  = "csum",
1019 	.port_fwd_begin = NULL,
1020 	.port_fwd_end   = NULL,
1021 	.packet_fwd     = pkt_burst_checksum_forward,
1022 };
1023