1 /* SPDX-License-Identifier: BSD-3-Clause 2 * Copyright(c) 2010-2014 Intel Corporation 3 */ 4 5 #include <stdarg.h> 6 #include <string.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_string_fns.h> 37 #include <rte_flow.h> 38 39 #include "testpmd.h" 40 41 struct tx_timestamp { 42 rte_be32_t signature; 43 rte_be16_t pkt_idx; 44 rte_be16_t queue_idx; 45 rte_be64_t ts; 46 }; 47 48 /* use RFC863 Discard Protocol */ 49 uint16_t tx_udp_src_port = 9; 50 uint16_t tx_udp_dst_port = 9; 51 52 /* use RFC5735 / RFC2544 reserved network test addresses */ 53 uint32_t tx_ip_src_addr = (198U << 24) | (18 << 16) | (0 << 8) | 1; 54 uint32_t tx_ip_dst_addr = (198U << 24) | (18 << 16) | (0 << 8) | 2; 55 56 #define IP_DEFTTL 64 /* from RFC 1340. */ 57 58 static struct rte_ipv4_hdr pkt_ip_hdr; /**< IP header of transmitted packets. */ 59 RTE_DEFINE_PER_LCORE(uint8_t, _src_port_var); /**< Source port variation */ 60 static struct rte_udp_hdr pkt_udp_hdr; /**< UDP header of tx packets. */ 61 62 static uint64_t timestamp_mask; /**< Timestamp dynamic flag mask */ 63 static int32_t timestamp_off; /**< Timestamp dynamic field offset */ 64 static bool timestamp_enable; /**< Timestamp enable */ 65 static uint64_t timestamp_initial[RTE_MAX_ETHPORTS]; 66 67 static void 68 copy_buf_to_pkt_segs(void* buf, unsigned len, struct rte_mbuf *pkt, 69 unsigned offset) 70 { 71 struct rte_mbuf *seg; 72 void *seg_buf; 73 unsigned copy_len; 74 75 seg = pkt; 76 while (offset >= seg->data_len) { 77 offset -= seg->data_len; 78 seg = seg->next; 79 } 80 copy_len = seg->data_len - offset; 81 seg_buf = rte_pktmbuf_mtod_offset(seg, char *, offset); 82 while (len > copy_len) { 83 rte_memcpy(seg_buf, buf, (size_t) copy_len); 84 len -= copy_len; 85 buf = ((char*) buf + copy_len); 86 seg = seg->next; 87 seg_buf = rte_pktmbuf_mtod(seg, char *); 88 copy_len = seg->data_len; 89 } 90 rte_memcpy(seg_buf, buf, (size_t) len); 91 } 92 93 static inline void 94 copy_buf_to_pkt(void* buf, unsigned len, struct rte_mbuf *pkt, unsigned offset) 95 { 96 if (offset + len <= pkt->data_len) { 97 rte_memcpy(rte_pktmbuf_mtod_offset(pkt, char *, offset), 98 buf, (size_t) len); 99 return; 100 } 101 copy_buf_to_pkt_segs(buf, len, pkt, offset); 102 } 103 104 static void 105 setup_pkt_udp_ip_headers(struct rte_ipv4_hdr *ip_hdr, 106 struct rte_udp_hdr *udp_hdr, 107 uint16_t pkt_data_len) 108 { 109 uint16_t pkt_len; 110 111 /* 112 * Initialize UDP header. 113 */ 114 pkt_len = (uint16_t) (pkt_data_len + sizeof(struct rte_udp_hdr)); 115 udp_hdr->src_port = rte_cpu_to_be_16(tx_udp_src_port); 116 udp_hdr->dst_port = rte_cpu_to_be_16(tx_udp_dst_port); 117 udp_hdr->dgram_len = RTE_CPU_TO_BE_16(pkt_len); 118 udp_hdr->dgram_cksum = 0; /* No UDP checksum. */ 119 120 /* 121 * Initialize IP header. 122 */ 123 pkt_len = (uint16_t) (pkt_len + sizeof(struct rte_ipv4_hdr)); 124 ip_hdr->version_ihl = RTE_IPV4_VHL_DEF; 125 ip_hdr->type_of_service = 0; 126 ip_hdr->fragment_offset = 0; 127 ip_hdr->time_to_live = IP_DEFTTL; 128 ip_hdr->next_proto_id = IPPROTO_UDP; 129 ip_hdr->packet_id = 0; 130 ip_hdr->total_length = RTE_CPU_TO_BE_16(pkt_len); 131 ip_hdr->src_addr = rte_cpu_to_be_32(tx_ip_src_addr); 132 ip_hdr->dst_addr = rte_cpu_to_be_32(tx_ip_dst_addr); 133 134 /* 135 * Compute IP header checksum. 136 */ 137 ip_hdr->hdr_checksum = rte_ipv4_cksum_simple(ip_hdr); 138 } 139 140 static inline void 141 update_pkt_header(struct rte_mbuf *pkt, uint32_t total_pkt_len) 142 { 143 struct rte_ipv4_hdr *ip_hdr; 144 struct rte_udp_hdr *udp_hdr; 145 uint16_t pkt_data_len; 146 uint16_t pkt_len; 147 148 pkt_data_len = (uint16_t) (total_pkt_len - ( 149 sizeof(struct rte_ether_hdr) + 150 sizeof(struct rte_ipv4_hdr) + 151 sizeof(struct rte_udp_hdr))); 152 /* update UDP packet length */ 153 udp_hdr = rte_pktmbuf_mtod_offset(pkt, struct rte_udp_hdr *, 154 sizeof(struct rte_ether_hdr) + 155 sizeof(struct rte_ipv4_hdr)); 156 pkt_len = (uint16_t) (pkt_data_len + sizeof(struct rte_udp_hdr)); 157 udp_hdr->dgram_len = RTE_CPU_TO_BE_16(pkt_len); 158 159 /* update IP packet length and checksum */ 160 ip_hdr = rte_pktmbuf_mtod_offset(pkt, struct rte_ipv4_hdr *, 161 sizeof(struct rte_ether_hdr)); 162 ip_hdr->hdr_checksum = 0; 163 pkt_len = (uint16_t) (pkt_len + sizeof(struct rte_ipv4_hdr)); 164 ip_hdr->total_length = RTE_CPU_TO_BE_16(pkt_len); 165 ip_hdr->hdr_checksum = rte_ipv4_cksum(ip_hdr); 166 } 167 168 static inline bool 169 pkt_burst_prepare(struct rte_mbuf *pkt, struct rte_mempool *mbp, 170 struct rte_ether_hdr *eth_hdr, const uint16_t vlan_tci, 171 const uint16_t vlan_tci_outer, const uint64_t ol_flags, 172 const uint16_t idx, struct fwd_stream *fs) 173 { 174 struct rte_mbuf *pkt_segs[RTE_MAX_SEGS_PER_PKT]; 175 struct rte_mbuf *pkt_seg; 176 uint32_t nb_segs, pkt_len; 177 uint8_t i; 178 179 if (unlikely(tx_pkt_split == TX_PKT_SPLIT_RND)) 180 nb_segs = rte_rand() % tx_pkt_nb_segs + 1; 181 else 182 nb_segs = tx_pkt_nb_segs; 183 184 if (nb_segs > 1) { 185 if (rte_mempool_get_bulk(mbp, (void **)pkt_segs, nb_segs - 1)) 186 return false; 187 } 188 189 rte_pktmbuf_reset_headroom(pkt); 190 pkt->data_len = tx_pkt_seg_lengths[0]; 191 pkt->ol_flags &= RTE_MBUF_F_EXTERNAL; 192 pkt->ol_flags |= ol_flags; 193 pkt->vlan_tci = vlan_tci; 194 pkt->vlan_tci_outer = vlan_tci_outer; 195 pkt->l2_len = sizeof(struct rte_ether_hdr); 196 pkt->l3_len = sizeof(struct rte_ipv4_hdr); 197 198 pkt_len = pkt->data_len; 199 pkt_seg = pkt; 200 for (i = 1; i < nb_segs; i++) { 201 pkt_seg->next = pkt_segs[i - 1]; 202 pkt_seg = pkt_seg->next; 203 pkt_seg->data_len = tx_pkt_seg_lengths[i]; 204 pkt_len += pkt_seg->data_len; 205 } 206 pkt_seg->next = NULL; /* Last segment of packet. */ 207 /* 208 * Copy headers in first packet segment(s). 209 */ 210 copy_buf_to_pkt(eth_hdr, sizeof(*eth_hdr), pkt, 0); 211 copy_buf_to_pkt(&pkt_ip_hdr, sizeof(pkt_ip_hdr), pkt, 212 sizeof(struct rte_ether_hdr)); 213 copy_buf_to_pkt(&pkt_udp_hdr, sizeof(pkt_udp_hdr), pkt, 214 sizeof(struct rte_ether_hdr) + 215 sizeof(struct rte_ipv4_hdr)); 216 if (txonly_multi_flow) { 217 uint16_t src_var = RTE_PER_LCORE(_src_port_var); 218 struct rte_udp_hdr *udp_hdr; 219 uint16_t src_port; 220 221 udp_hdr = rte_pktmbuf_mtod_offset(pkt, 222 struct rte_udp_hdr *, 223 sizeof(struct rte_ether_hdr) + 224 sizeof(struct rte_ipv4_hdr)); 225 /* 226 * Generate multiple flows by varying UDP source port. 227 * This enables packets are well distributed by RSS in 228 * receiver side if any and txonly mode can be a decent 229 * packet generator for developer's quick performance 230 * regression test. 231 * 232 * Only ports in the range 49152 (0xC000) and 65535 (0xFFFF) 233 * will be used, with the least significant byte representing 234 * the lcore ID. As such, the most significant byte will cycle 235 * through 0xC0 and 0xFF. 236 */ 237 src_port = ((src_var++ | 0xC0) << 8) + rte_lcore_id(); 238 udp_hdr->src_port = rte_cpu_to_be_16(src_port); 239 RTE_PER_LCORE(_src_port_var) = src_var; 240 } 241 242 if (unlikely(tx_pkt_split == TX_PKT_SPLIT_RND) || txonly_multi_flow) 243 update_pkt_header(pkt, pkt_len); 244 245 if (unlikely(timestamp_enable)) { 246 uint64_t skew = fs->ts_skew; 247 struct tx_timestamp timestamp_mark; 248 249 if (unlikely(!skew)) { 250 struct rte_eth_dev_info dev_info; 251 unsigned int txqs_n; 252 uint64_t phase; 253 int ret; 254 255 ret = eth_dev_info_get_print_err(fs->tx_port, &dev_info); 256 if (ret != 0) { 257 TESTPMD_LOG(ERR, 258 "Failed to get device info for port %d," 259 "could not finish timestamp init", 260 fs->tx_port); 261 return false; 262 } 263 txqs_n = dev_info.nb_tx_queues; 264 phase = tx_pkt_times_inter * fs->tx_queue / 265 (txqs_n ? txqs_n : 1); 266 /* 267 * Initialize the scheduling time phase shift 268 * depending on queue index. 269 */ 270 skew = timestamp_initial[fs->tx_port] + 271 tx_pkt_times_inter + phase; 272 fs->ts_skew = skew; 273 } 274 timestamp_mark.pkt_idx = rte_cpu_to_be_16(idx); 275 timestamp_mark.queue_idx = rte_cpu_to_be_16(fs->tx_queue); 276 timestamp_mark.signature = rte_cpu_to_be_32(0xBEEFC0DE); 277 if (unlikely(!idx)) { 278 skew += tx_pkt_times_inter; 279 pkt->ol_flags |= timestamp_mask; 280 *RTE_MBUF_DYNFIELD 281 (pkt, timestamp_off, uint64_t *) = skew; 282 fs->ts_skew = skew; 283 timestamp_mark.ts = rte_cpu_to_be_64(skew); 284 } else if (tx_pkt_times_intra) { 285 skew += tx_pkt_times_intra; 286 pkt->ol_flags |= timestamp_mask; 287 *RTE_MBUF_DYNFIELD 288 (pkt, timestamp_off, uint64_t *) = skew; 289 fs->ts_skew = skew; 290 timestamp_mark.ts = rte_cpu_to_be_64(skew); 291 } else { 292 timestamp_mark.ts = RTE_BE64(0); 293 } 294 copy_buf_to_pkt(×tamp_mark, sizeof(timestamp_mark), pkt, 295 sizeof(struct rte_ether_hdr) + 296 sizeof(struct rte_ipv4_hdr) + 297 sizeof(pkt_udp_hdr)); 298 } 299 /* 300 * Complete first mbuf of packet and append it to the 301 * burst of packets to be transmitted. 302 */ 303 pkt->nb_segs = nb_segs; 304 pkt->pkt_len = pkt_len; 305 306 return true; 307 } 308 309 /* 310 * Transmit a burst of multi-segments packets. 311 */ 312 static bool 313 pkt_burst_transmit(struct fwd_stream *fs) 314 { 315 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 316 struct rte_port *txp; 317 struct rte_mbuf *pkt; 318 struct rte_mempool *mbp; 319 struct rte_ether_hdr eth_hdr; 320 uint16_t nb_tx; 321 uint16_t nb_pkt; 322 uint16_t vlan_tci, vlan_tci_outer; 323 uint64_t ol_flags = 0; 324 uint64_t tx_offloads; 325 326 mbp = current_fwd_lcore()->mbp; 327 txp = &ports[fs->tx_port]; 328 tx_offloads = txp->dev_conf.txmode.offloads; 329 vlan_tci = txp->tx_vlan_id; 330 vlan_tci_outer = txp->tx_vlan_id_outer; 331 if (tx_offloads & RTE_ETH_TX_OFFLOAD_VLAN_INSERT) 332 ol_flags = RTE_MBUF_F_TX_VLAN; 333 if (tx_offloads & RTE_ETH_TX_OFFLOAD_QINQ_INSERT) 334 ol_flags |= RTE_MBUF_F_TX_QINQ; 335 if (tx_offloads & RTE_ETH_TX_OFFLOAD_MACSEC_INSERT) 336 ol_flags |= RTE_MBUF_F_TX_MACSEC; 337 338 /* 339 * Initialize Ethernet header. 340 */ 341 rte_ether_addr_copy(&peer_eth_addrs[fs->peer_addr], ð_hdr.dst_addr); 342 rte_ether_addr_copy(&ports[fs->tx_port].eth_addr, ð_hdr.src_addr); 343 eth_hdr.ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4); 344 345 if (rte_mempool_get_bulk(mbp, (void **)pkts_burst, 346 nb_pkt_per_burst) == 0) { 347 for (nb_pkt = 0; nb_pkt < nb_pkt_per_burst; nb_pkt++) { 348 if (unlikely(!pkt_burst_prepare(pkts_burst[nb_pkt], mbp, 349 ð_hdr, vlan_tci, 350 vlan_tci_outer, 351 ol_flags, 352 nb_pkt, fs))) { 353 rte_mempool_put_bulk(mbp, 354 (void **)&pkts_burst[nb_pkt], 355 nb_pkt_per_burst - nb_pkt); 356 break; 357 } 358 } 359 } else { 360 for (nb_pkt = 0; nb_pkt < nb_pkt_per_burst; nb_pkt++) { 361 pkt = rte_mbuf_raw_alloc(mbp); 362 if (pkt == NULL) 363 break; 364 if (unlikely(!pkt_burst_prepare(pkt, mbp, ð_hdr, 365 vlan_tci, 366 vlan_tci_outer, 367 ol_flags, 368 nb_pkt, fs))) { 369 rte_pktmbuf_free(pkt); 370 break; 371 } 372 pkts_burst[nb_pkt] = pkt; 373 } 374 } 375 376 if (nb_pkt == 0) 377 return false; 378 379 nb_tx = common_fwd_stream_transmit(fs, pkts_burst, nb_pkt); 380 381 if (txonly_multi_flow) 382 RTE_PER_LCORE(_src_port_var) -= nb_pkt - nb_tx; 383 384 if (unlikely(nb_tx < nb_pkt)) { 385 if (verbose_level > 0 && fs->fwd_dropped == 0) 386 printf("port %d tx_queue %d - drop " 387 "(nb_pkt:%u - nb_tx:%u)=%u packets\n", 388 fs->tx_port, fs->tx_queue, 389 (unsigned) nb_pkt, (unsigned) nb_tx, 390 (unsigned) (nb_pkt - nb_tx)); 391 } 392 393 return true; 394 } 395 396 static int 397 tx_only_begin(portid_t pi) 398 { 399 uint16_t pkt_hdr_len, pkt_data_len; 400 int dynf; 401 402 pkt_hdr_len = (uint16_t)(sizeof(struct rte_ether_hdr) + 403 sizeof(struct rte_ipv4_hdr) + 404 sizeof(struct rte_udp_hdr)); 405 pkt_data_len = tx_pkt_length - pkt_hdr_len; 406 407 if ((tx_pkt_split == TX_PKT_SPLIT_RND || txonly_multi_flow) && 408 tx_pkt_seg_lengths[0] < pkt_hdr_len) { 409 TESTPMD_LOG(ERR, 410 "Random segment number or multiple flow is enabled, " 411 "but tx_pkt_seg_lengths[0] %u < %u (needed)\n", 412 tx_pkt_seg_lengths[0], pkt_hdr_len); 413 return -EINVAL; 414 } 415 416 setup_pkt_udp_ip_headers(&pkt_ip_hdr, &pkt_udp_hdr, pkt_data_len); 417 418 timestamp_enable = false; 419 timestamp_mask = 0; 420 timestamp_off = -1; 421 dynf = rte_mbuf_dynflag_lookup 422 (RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME, NULL); 423 if (dynf >= 0) 424 timestamp_mask = 1ULL << dynf; 425 dynf = rte_mbuf_dynfield_lookup 426 (RTE_MBUF_DYNFIELD_TIMESTAMP_NAME, NULL); 427 if (dynf >= 0) 428 timestamp_off = dynf; 429 timestamp_enable = tx_pkt_times_inter && 430 timestamp_mask && 431 timestamp_off >= 0 && 432 !rte_eth_read_clock(pi, ×tamp_initial[pi]); 433 434 if (timestamp_enable) { 435 pkt_hdr_len += sizeof(struct tx_timestamp); 436 437 if (tx_pkt_split == TX_PKT_SPLIT_RND) { 438 if (tx_pkt_seg_lengths[0] < pkt_hdr_len) { 439 TESTPMD_LOG(ERR, 440 "Time stamp and random segment number are enabled, " 441 "but tx_pkt_seg_lengths[0] %u < %u (needed)\n", 442 tx_pkt_seg_lengths[0], pkt_hdr_len); 443 return -EINVAL; 444 } 445 } else { 446 uint16_t total = 0; 447 uint8_t i; 448 449 for (i = 0; i < tx_pkt_nb_segs; i++) { 450 total += tx_pkt_seg_lengths[i]; 451 if (total >= pkt_hdr_len) 452 break; 453 } 454 455 if (total < pkt_hdr_len) { 456 TESTPMD_LOG(ERR, 457 "Not enough Tx segment space for time stamp info, " 458 "total %u < %u (needed)\n", 459 total, pkt_hdr_len); 460 return -EINVAL; 461 } 462 } 463 } 464 465 /* Make sure all settings are visible on forwarding cores.*/ 466 rte_wmb(); 467 return 0; 468 } 469 470 static void 471 tx_only_stream_init(struct fwd_stream *fs) 472 { 473 fs->disabled = ports[fs->tx_port].txq[fs->tx_queue].state == 474 RTE_ETH_QUEUE_STATE_STOPPED; 475 } 476 477 struct fwd_engine tx_only_engine = { 478 .fwd_mode_name = "txonly", 479 .port_fwd_begin = tx_only_begin, 480 .stream_init = tx_only_stream_init, 481 .packet_fwd = pkt_burst_transmit, 482 }; 483