]> git.proxmox.com Git - ceph.git/blob - ceph/src/spdk/dpdk/app/test-pmd/csumonly.c
projects / ceph.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
update sources to ceph Nautilus 14.2.1
[ceph.git] / ceph / src / spdk / dpdk / app / test-pmd / csumonly.c
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 udp_hdr = (struct udp_hdr *)((char *)outer_l3_hdr + info->outer_l3_len);
472
473 /* outer UDP checksum is done in software as we have no hardware
474 * supporting it today, and no API for it. In the other side, for
475 * UDP tunneling, like VXLAN or Geneve, outer UDP checksum can be
476 * set to zero.
477 *
478 * If a packet will be TSOed into small packets by NIC, we cannot
479 * set/calculate a non-zero checksum, because it will be a wrong
480 * value after the packet be split into several small packets.
481 */
482 if (tso_enabled)
483 udp_hdr->dgram_cksum = 0;
484
485 /* do not recalculate udp cksum if it was 0 */
486 if (udp_hdr->dgram_cksum != 0) {
487 udp_hdr->dgram_cksum = 0;
488 if (info->outer_ethertype == _htons(ETHER_TYPE_IPv4))
489 udp_hdr->dgram_cksum =
490 rte_ipv4_udptcp_cksum(ipv4_hdr, udp_hdr);
491 else
492 udp_hdr->dgram_cksum =
493 rte_ipv6_udptcp_cksum(ipv6_hdr, udp_hdr);
494 }
495
496 return ol_flags;
497 }
498
499 /*
500 * Helper function.
501 * Performs actual copying.
502 * Returns number of segments in the destination mbuf on success,
503 * or negative error code on failure.
504 */
505 static int
506 mbuf_copy_split(const struct rte_mbuf *ms, struct rte_mbuf *md[],
507 uint16_t seglen[], uint8_t nb_seg)
508 {
509 uint32_t dlen, slen, tlen;
510 uint32_t i, len;
511 const struct rte_mbuf *m;
512 const uint8_t *src;
513 uint8_t *dst;
514
515 dlen = 0;
516 slen = 0;
517 tlen = 0;
518
519 dst = NULL;
520 src = NULL;
521
522 m = ms;
523 i = 0;
524 while (ms != NULL && i != nb_seg) {
525
526 if (slen == 0) {
527 slen = rte_pktmbuf_data_len(ms);
528 src = rte_pktmbuf_mtod(ms, const uint8_t *);
529 }
530
531 if (dlen == 0) {
532 dlen = RTE_MIN(seglen[i], slen);
533 md[i]->data_len = dlen;
534 md[i]->next = (i + 1 == nb_seg) ? NULL : md[i + 1];
535 dst = rte_pktmbuf_mtod(md[i], uint8_t *);
536 }
537
538 len = RTE_MIN(slen, dlen);
539 memcpy(dst, src, len);
540 tlen += len;
541 slen -= len;
542 dlen -= len;
543 src += len;
544 dst += len;
545
546 if (slen == 0)
547 ms = ms->next;
548 if (dlen == 0)
549 i++;
550 }
551
552 if (ms != NULL)
553 return -ENOBUFS;
554 else if (tlen != m->pkt_len)
555 return -EINVAL;
556
557 md[0]->nb_segs = nb_seg;
558 md[0]->pkt_len = tlen;
559 md[0]->vlan_tci = m->vlan_tci;
560 md[0]->vlan_tci_outer = m->vlan_tci_outer;
561 md[0]->ol_flags = m->ol_flags;
562 md[0]->tx_offload = m->tx_offload;
563
564 return nb_seg;
565 }
566
567 /*
568 * Allocate a new mbuf with up to tx_pkt_nb_segs segments.
569 * Copy packet contents and offload information into then new segmented mbuf.
570 */
571 static struct rte_mbuf *
572 pkt_copy_split(const struct rte_mbuf *pkt)
573 {
574 int32_t n, rc;
575 uint32_t i, len, nb_seg;
576 struct rte_mempool *mp;
577 uint16_t seglen[RTE_MAX_SEGS_PER_PKT];
578 struct rte_mbuf *p, *md[RTE_MAX_SEGS_PER_PKT];
579
580 mp = current_fwd_lcore()->mbp;
581
582 if (tx_pkt_split == TX_PKT_SPLIT_RND)
583 nb_seg = random() % tx_pkt_nb_segs + 1;
584 else
585 nb_seg = tx_pkt_nb_segs;
586
587 memcpy(seglen, tx_pkt_seg_lengths, nb_seg * sizeof(seglen[0]));
588
589 /* calculate number of segments to use and their length. */
590 len = 0;
591 for (i = 0; i != nb_seg && len < pkt->pkt_len; i++) {
592 len += seglen[i];
593 md[i] = NULL;
594 }
595
596 n = pkt->pkt_len - len;
597
598 /* update size of the last segment to fit rest of the packet */
599 if (n >= 0) {
600 seglen[i - 1] += n;
601 len += n;
602 }
603
604 nb_seg = i;
605 while (i != 0) {
606 p = rte_pktmbuf_alloc(mp);
607 if (p == NULL) {
608 TESTPMD_LOG(ERR,
609 "failed to allocate %u-th of %u mbuf "
610 "from mempool: %s\n",
611 nb_seg - i, nb_seg, mp->name);
612 break;
613 }
614
615 md[--i] = p;
616 if (rte_pktmbuf_tailroom(md[i]) < seglen[i]) {
617 TESTPMD_LOG(ERR, "mempool %s, %u-th segment: "
618 "expected seglen: %u, "
619 "actual mbuf tailroom: %u\n",
620 mp->name, i, seglen[i],
621 rte_pktmbuf_tailroom(md[i]));
622 break;
623 }
624 }
625
626 /* all mbufs successfully allocated, do copy */
627 if (i == 0) {
628 rc = mbuf_copy_split(pkt, md, seglen, nb_seg);
629 if (rc < 0)
630 TESTPMD_LOG(ERR,
631 "mbuf_copy_split for %p(len=%u, nb_seg=%u) "
632 "into %u segments failed with error code: %d\n",
633 pkt, pkt->pkt_len, pkt->nb_segs, nb_seg, rc);
634
635 /* figure out how many mbufs to free. */
636 i = RTE_MAX(rc, 0);
637 }
638
639 /* free unused mbufs */
640 for (; i != nb_seg; i++) {
641 rte_pktmbuf_free_seg(md[i]);
642 md[i] = NULL;
643 }
644
645 return md[0];
646 }
647
648 /*
649 * Receive a burst of packets, and for each packet:
650 * - parse packet, and try to recognize a supported packet type (1)
651 * - if it's not a supported packet type, don't touch the packet, else:
652 * - reprocess the checksum of all supported layers. This is done in SW
653 * or HW, depending on testpmd command line configuration
654 * - if TSO is enabled in testpmd command line, also flag the mbuf for TCP
655 * segmentation offload (this implies HW TCP checksum)
656 * Then transmit packets on the output port.
657 *
658 * (1) Supported packets are:
659 * Ether / (vlan) / IP|IP6 / UDP|TCP|SCTP .
660 * Ether / (vlan) / outer IP|IP6 / outer UDP / VxLAN / Ether / IP|IP6 /
661 * UDP|TCP|SCTP
662 * Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / Ether / IP|IP6 /
663 * UDP|TCP|SCTP
664 * Ether / (vlan) / outer IP|IP6 / outer UDP / VXLAN-GPE / IP|IP6 /
665 * UDP|TCP|SCTP
666 * Ether / (vlan) / outer IP|IP6 / GRE / Ether / IP|IP6 / UDP|TCP|SCTP
667 * Ether / (vlan) / outer IP|IP6 / GRE / IP|IP6 / UDP|TCP|SCTP
668 * Ether / (vlan) / outer IP|IP6 / IP|IP6 / UDP|TCP|SCTP
669 *
670 * The testpmd command line for this forward engine sets the flags
671 * TESTPMD_TX_OFFLOAD_* in ports[tx_port].tx_ol_flags. They control
672 * wether a checksum must be calculated in software or in hardware. The
673 * IP, UDP, TCP and SCTP flags always concern the inner layer. The
674 * OUTER_IP is only useful for tunnel packets.
675 */
676 static void
677 pkt_burst_checksum_forward(struct fwd_stream *fs)
678 {
679 struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
680 struct rte_mbuf *gso_segments[GSO_MAX_PKT_BURST];
681 struct rte_gso_ctx *gso_ctx;
682 struct rte_mbuf **tx_pkts_burst;
683 struct rte_port *txp;
684 struct rte_mbuf *m, *p;
685 struct ether_hdr *eth_hdr;
686 void *l3_hdr = NULL, *outer_l3_hdr = NULL; /* can be IPv4 or IPv6 */
687 void **gro_ctx;
688 uint16_t gro_pkts_num;
689 uint8_t gro_enable;
690 uint16_t nb_rx;
691 uint16_t nb_tx;
692 uint16_t nb_prep;
693 uint16_t i;
694 uint64_t rx_ol_flags, tx_ol_flags;
695 uint64_t tx_offloads;
696 uint32_t retry;
697 uint32_t rx_bad_ip_csum;
698 uint32_t rx_bad_l4_csum;
699 struct testpmd_offload_info info;
700 uint16_t nb_segments = 0;
701 int ret;
702
703 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
704 uint64_t start_tsc;
705 uint64_t end_tsc;
706 uint64_t core_cycles;
707 #endif
708
709 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
710 start_tsc = rte_rdtsc();
711 #endif
712
713 /* receive a burst of packet */
714 nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
715 nb_pkt_per_burst);
716 if (unlikely(nb_rx == 0))
717 return;
718 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
719 fs->rx_burst_stats.pkt_burst_spread[nb_rx]++;
720 #endif
721 fs->rx_packets += nb_rx;
722 rx_bad_ip_csum = 0;
723 rx_bad_l4_csum = 0;
724 gro_enable = gro_ports[fs->rx_port].enable;
725
726 txp = &ports[fs->tx_port];
727 tx_offloads = txp->dev_conf.txmode.offloads;
728 memset(&info, 0, sizeof(info));
729 info.tso_segsz = txp->tso_segsz;
730 info.tunnel_tso_segsz = txp->tunnel_tso_segsz;
731 if (gso_ports[fs->tx_port].enable)
732 info.gso_enable = 1;
733
734 for (i = 0; i < nb_rx; i++) {
735 if (likely(i < nb_rx - 1))
736 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i + 1],
737 void *));
738
739 m = pkts_burst[i];
740 info.is_tunnel = 0;
741 info.pkt_len = rte_pktmbuf_pkt_len(m);
742 tx_ol_flags = m->ol_flags &
743 (IND_ATTACHED_MBUF | EXT_ATTACHED_MBUF);
744 rx_ol_flags = m->ol_flags;
745
746 /* Update the L3/L4 checksum error packet statistics */
747 if ((rx_ol_flags & PKT_RX_IP_CKSUM_MASK) == PKT_RX_IP_CKSUM_BAD)
748 rx_bad_ip_csum += 1;
749 if ((rx_ol_flags & PKT_RX_L4_CKSUM_MASK) == PKT_RX_L4_CKSUM_BAD)
750 rx_bad_l4_csum += 1;
751
752 /* step 1: dissect packet, parsing optional vlan, ip4/ip6, vxlan
753 * and inner headers */
754
755 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
756 ether_addr_copy(&peer_eth_addrs[fs->peer_addr],
757 &eth_hdr->d_addr);
758 ether_addr_copy(&ports[fs->tx_port].eth_addr,
759 &eth_hdr->s_addr);
760 parse_ethernet(eth_hdr, &info);
761 l3_hdr = (char *)eth_hdr + info.l2_len;
762
763 /* check if it's a supported tunnel */
764 if (txp->parse_tunnel) {
765 if (info.l4_proto == IPPROTO_UDP) {
766 struct udp_hdr *udp_hdr;
767
768 udp_hdr = (struct udp_hdr *)((char *)l3_hdr +
769 info.l3_len);
770 parse_vxlan_gpe(udp_hdr, &info);
771 if (info.is_tunnel) {
772 tx_ol_flags |= PKT_TX_TUNNEL_VXLAN_GPE;
773 } else {
774 parse_vxlan(udp_hdr, &info,
775 m->packet_type);
776 if (info.is_tunnel)
777 tx_ol_flags |=
778 PKT_TX_TUNNEL_VXLAN;
779 }
780 } else if (info.l4_proto == IPPROTO_GRE) {
781 struct simple_gre_hdr *gre_hdr;
782
783 gre_hdr = (struct simple_gre_hdr *)
784 ((char *)l3_hdr + info.l3_len);
785 parse_gre(gre_hdr, &info);
786 if (info.is_tunnel)
787 tx_ol_flags |= PKT_TX_TUNNEL_GRE;
788 } else if (info.l4_proto == IPPROTO_IPIP) {
789 void *encap_ip_hdr;
790
791 encap_ip_hdr = (char *)l3_hdr + info.l3_len;
792 parse_encap_ip(encap_ip_hdr, &info);
793 if (info.is_tunnel)
794 tx_ol_flags |= PKT_TX_TUNNEL_IPIP;
795 }
796 }
797
798 /* update l3_hdr and outer_l3_hdr if a tunnel was parsed */
799 if (info.is_tunnel) {
800 outer_l3_hdr = l3_hdr;
801 l3_hdr = (char *)l3_hdr + info.outer_l3_len + info.l2_len;
802 }
803
804 /* step 2: depending on user command line configuration,
805 * recompute checksum either in software or flag the
806 * mbuf to offload the calculation to the NIC. If TSO
807 * is configured, prepare the mbuf for TCP segmentation. */
808
809 /* process checksums of inner headers first */
810 tx_ol_flags |= process_inner_cksums(l3_hdr, &info,
811 tx_offloads);
812
813 /* Then process outer headers if any. Note that the software
814 * checksum will be wrong if one of the inner checksums is
815 * processed in hardware. */
816 if (info.is_tunnel == 1) {
817 tx_ol_flags |= process_outer_cksums(outer_l3_hdr, &info,
818 tx_offloads,
819 !!(tx_ol_flags & PKT_TX_TCP_SEG));
820 }
821
822 /* step 3: fill the mbuf meta data (flags and header lengths) */
823
824 m->tx_offload = 0;
825 if (info.is_tunnel == 1) {
826 if (info.tunnel_tso_segsz ||
827 (tx_offloads &
828 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
829 (tx_ol_flags & PKT_TX_OUTER_IPV6)) {
830 m->outer_l2_len = info.outer_l2_len;
831 m->outer_l3_len = info.outer_l3_len;
832 m->l2_len = info.l2_len;
833 m->l3_len = info.l3_len;
834 m->l4_len = info.l4_len;
835 m->tso_segsz = info.tunnel_tso_segsz;
836 }
837 else {
838 /* if there is a outer UDP cksum
839 processed in sw and the inner in hw,
840 the outer checksum will be wrong as
841 the payload will be modified by the
842 hardware */
843 m->l2_len = info.outer_l2_len +
844 info.outer_l3_len + info.l2_len;
845 m->l3_len = info.l3_len;
846 m->l4_len = info.l4_len;
847 }
848 } else {
849 /* this is only useful if an offload flag is
850 * set, but it does not hurt to fill it in any
851 * case */
852 m->l2_len = info.l2_len;
853 m->l3_len = info.l3_len;
854 m->l4_len = info.l4_len;
855 m->tso_segsz = info.tso_segsz;
856 }
857 m->ol_flags = tx_ol_flags;
858
859 /* Do split & copy for the packet. */
860 if (tx_pkt_split != TX_PKT_SPLIT_OFF) {
861 p = pkt_copy_split(m);
862 if (p != NULL) {
863 rte_pktmbuf_free(m);
864 m = p;
865 pkts_burst[i] = m;
866 }
867 }
868
869 /* if verbose mode is enabled, dump debug info */
870 if (verbose_level > 0) {
871 char buf[256];
872
873 printf("-----------------\n");
874 printf("port=%u, mbuf=%p, pkt_len=%u, nb_segs=%u:\n",
875 fs->rx_port, m, m->pkt_len, m->nb_segs);
876 /* dump rx parsed packet info */
877 rte_get_rx_ol_flag_list(rx_ol_flags, buf, sizeof(buf));
878 printf("rx: l2_len=%d ethertype=%x l3_len=%d "
879 "l4_proto=%d l4_len=%d flags=%s\n",
880 info.l2_len, rte_be_to_cpu_16(info.ethertype),
881 info.l3_len, info.l4_proto, info.l4_len, buf);
882 if (rx_ol_flags & PKT_RX_LRO)
883 printf("rx: m->lro_segsz=%u\n", m->tso_segsz);
884 if (info.is_tunnel == 1)
885 printf("rx: outer_l2_len=%d outer_ethertype=%x "
886 "outer_l3_len=%d\n", info.outer_l2_len,
887 rte_be_to_cpu_16(info.outer_ethertype),
888 info.outer_l3_len);
889 /* dump tx packet info */
890 if ((tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
891 DEV_TX_OFFLOAD_UDP_CKSUM |
892 DEV_TX_OFFLOAD_TCP_CKSUM |
893 DEV_TX_OFFLOAD_SCTP_CKSUM)) ||
894 info.tso_segsz != 0)
895 printf("tx: m->l2_len=%d m->l3_len=%d "
896 "m->l4_len=%d\n",
897 m->l2_len, m->l3_len, m->l4_len);
898 if (info.is_tunnel == 1) {
899 if ((tx_offloads &
900 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM) ||
901 (tx_ol_flags & PKT_TX_OUTER_IPV6))
902 printf("tx: m->outer_l2_len=%d "
903 "m->outer_l3_len=%d\n",
904 m->outer_l2_len,
905 m->outer_l3_len);
906 if (info.tunnel_tso_segsz != 0 &&
907 (m->ol_flags & PKT_TX_TCP_SEG))
908 printf("tx: m->tso_segsz=%d\n",
909 m->tso_segsz);
910 } else if (info.tso_segsz != 0 &&
911 (m->ol_flags & PKT_TX_TCP_SEG))
912 printf("tx: m->tso_segsz=%d\n", m->tso_segsz);
913 rte_get_tx_ol_flag_list(m->ol_flags, buf, sizeof(buf));
914 printf("tx: flags=%s", buf);
915 printf("\n");
916 }
917 }
918
919 if (unlikely(gro_enable)) {
920 if (gro_flush_cycles == GRO_DEFAULT_FLUSH_CYCLES) {
921 nb_rx = rte_gro_reassemble_burst(pkts_burst, nb_rx,
922 &(gro_ports[fs->rx_port].param));
923 } else {
924 gro_ctx = current_fwd_lcore()->gro_ctx;
925 nb_rx = rte_gro_reassemble(pkts_burst, nb_rx, gro_ctx);
926
927 if (++fs->gro_times >= gro_flush_cycles) {
928 gro_pkts_num = rte_gro_get_pkt_count(gro_ctx);
929 if (gro_pkts_num > MAX_PKT_BURST - nb_rx)
930 gro_pkts_num = MAX_PKT_BURST - nb_rx;
931
932 nb_rx += rte_gro_timeout_flush(gro_ctx, 0,
933 RTE_GRO_TCP_IPV4,
934 &pkts_burst[nb_rx],
935 gro_pkts_num);
936 fs->gro_times = 0;
937 }
938 }
939 }
940
941 if (gso_ports[fs->tx_port].enable == 0)
942 tx_pkts_burst = pkts_burst;
943 else {
944 gso_ctx = &(current_fwd_lcore()->gso_ctx);
945 gso_ctx->gso_size = gso_max_segment_size;
946 for (i = 0; i < nb_rx; i++) {
947 ret = rte_gso_segment(pkts_burst[i], gso_ctx,
948 &gso_segments[nb_segments],
949 GSO_MAX_PKT_BURST - nb_segments);
950 if (ret >= 0)
951 nb_segments += ret;
952 else {
953 TESTPMD_LOG(DEBUG, "Unable to segment packet");
954 rte_pktmbuf_free(pkts_burst[i]);
955 }
956 }
957
958 tx_pkts_burst = gso_segments;
959 nb_rx = nb_segments;
960 }
961
962 nb_prep = rte_eth_tx_prepare(fs->tx_port, fs->tx_queue,
963 tx_pkts_burst, nb_rx);
964 if (nb_prep != nb_rx)
965 printf("Preparing packet burst to transmit failed: %s\n",
966 rte_strerror(rte_errno));
967
968 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
969 nb_prep);
970
971 /*
972 * Retry if necessary
973 */
974 if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
975 retry = 0;
976 while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
977 rte_delay_us(burst_tx_delay_time);
978 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
979 &tx_pkts_burst[nb_tx], nb_rx - nb_tx);
980 }
981 }
982 fs->tx_packets += nb_tx;
983 fs->rx_bad_ip_csum += rx_bad_ip_csum;
984 fs->rx_bad_l4_csum += rx_bad_l4_csum;
985
986 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS
987 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++;
988 #endif
989 if (unlikely(nb_tx < nb_rx)) {
990 fs->fwd_dropped += (nb_rx - nb_tx);
991 do {
992 rte_pktmbuf_free(tx_pkts_burst[nb_tx]);
993 } while (++nb_tx < nb_rx);
994 }
995
996 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
997 end_tsc = rte_rdtsc();
998 core_cycles = (end_tsc - start_tsc);
999 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles);
1000 #endif
1001 }
1002
1003 struct fwd_engine csum_fwd_engine = {
1004 .fwd_mode_name = "csum",
1005 .port_fwd_begin = NULL,
1006 .port_fwd_end = NULL,
1007 .packet_fwd = pkt_burst_checksum_forward,
1008 };
Ceph