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_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_string_fns.h> 39 #include <rte_flow.h> 40 41 #include "testpmd.h" 42 43 /* use RFC863 Discard Protocol */ 44 uint16_t tx_udp_src_port = 9; 45 uint16_t tx_udp_dst_port = 9; 46 47 /* use RFC5735 / RFC2544 reserved network test addresses */ 48 uint32_t tx_ip_src_addr = (198U << 24) | (18 << 16) | (0 << 8) | 1; 49 uint32_t tx_ip_dst_addr = (198U << 24) | (18 << 16) | (0 << 8) | 2; 50 51 #define IP_DEFTTL 64 /* from RFC 1340. */ 52 53 static struct rte_ipv4_hdr pkt_ip_hdr; /**< IP header of transmitted packets. */ 54 RTE_DEFINE_PER_LCORE(uint8_t, _ip_var); /**< IP address variation */ 55 static struct rte_udp_hdr pkt_udp_hdr; /**< UDP header of tx packets. */ 56 RTE_DEFINE_PER_LCORE(uint64_t, timestamp_qskew); 57 /**< Timestamp offset per queue */ 58 RTE_DEFINE_PER_LCORE(uint32_t, timestamp_idone); /**< Timestamp init done. */ 59 60 static uint64_t timestamp_mask; /**< Timestamp dynamic flag mask */ 61 static int32_t timestamp_off; /**< Timestamp dynamic field offset */ 62 static bool timestamp_enable; /**< Timestamp enable */ 63 static uint32_t timestamp_init_req; /**< Timestamp initialization request. */ 64 static uint64_t timestamp_initial[RTE_MAX_ETHPORTS]; 65 66 static void 67 copy_buf_to_pkt_segs(void* buf, unsigned len, struct rte_mbuf *pkt, 68 unsigned offset) 69 { 70 struct rte_mbuf *seg; 71 void *seg_buf; 72 unsigned copy_len; 73 74 seg = pkt; 75 while (offset >= seg->data_len) { 76 offset -= seg->data_len; 77 seg = seg->next; 78 } 79 copy_len = seg->data_len - offset; 80 seg_buf = rte_pktmbuf_mtod_offset(seg, char *, offset); 81 while (len > copy_len) { 82 rte_memcpy(seg_buf, buf, (size_t) copy_len); 83 len -= copy_len; 84 buf = ((char*) buf + copy_len); 85 seg = seg->next; 86 seg_buf = rte_pktmbuf_mtod(seg, char *); 87 copy_len = seg->data_len; 88 } 89 rte_memcpy(seg_buf, buf, (size_t) len); 90 } 91 92 static inline void 93 copy_buf_to_pkt(void* buf, unsigned len, struct rte_mbuf *pkt, unsigned offset) 94 { 95 if (offset + len <= pkt->data_len) { 96 rte_memcpy(rte_pktmbuf_mtod_offset(pkt, char *, offset), 97 buf, (size_t) len); 98 return; 99 } 100 copy_buf_to_pkt_segs(buf, len, pkt, offset); 101 } 102 103 static void 104 setup_pkt_udp_ip_headers(struct rte_ipv4_hdr *ip_hdr, 105 struct rte_udp_hdr *udp_hdr, 106 uint16_t pkt_data_len) 107 { 108 uint16_t *ptr16; 109 uint32_t ip_cksum; 110 uint16_t pkt_len; 111 112 /* 113 * Initialize UDP header. 114 */ 115 pkt_len = (uint16_t) (pkt_data_len + sizeof(struct rte_udp_hdr)); 116 udp_hdr->src_port = rte_cpu_to_be_16(tx_udp_src_port); 117 udp_hdr->dst_port = rte_cpu_to_be_16(tx_udp_dst_port); 118 udp_hdr->dgram_len = RTE_CPU_TO_BE_16(pkt_len); 119 udp_hdr->dgram_cksum = 0; /* No UDP checksum. */ 120 121 /* 122 * Initialize IP header. 123 */ 124 pkt_len = (uint16_t) (pkt_len + sizeof(struct rte_ipv4_hdr)); 125 ip_hdr->version_ihl = RTE_IPV4_VHL_DEF; 126 ip_hdr->type_of_service = 0; 127 ip_hdr->fragment_offset = 0; 128 ip_hdr->time_to_live = IP_DEFTTL; 129 ip_hdr->next_proto_id = IPPROTO_UDP; 130 ip_hdr->packet_id = 0; 131 ip_hdr->total_length = RTE_CPU_TO_BE_16(pkt_len); 132 ip_hdr->src_addr = rte_cpu_to_be_32(tx_ip_src_addr); 133 ip_hdr->dst_addr = rte_cpu_to_be_32(tx_ip_dst_addr); 134 135 /* 136 * Compute IP header checksum. 137 */ 138 ptr16 = (unaligned_uint16_t*) ip_hdr; 139 ip_cksum = 0; 140 ip_cksum += ptr16[0]; ip_cksum += ptr16[1]; 141 ip_cksum += ptr16[2]; ip_cksum += ptr16[3]; 142 ip_cksum += ptr16[4]; 143 ip_cksum += ptr16[6]; ip_cksum += ptr16[7]; 144 ip_cksum += ptr16[8]; ip_cksum += ptr16[9]; 145 146 /* 147 * Reduce 32 bit checksum to 16 bits and complement it. 148 */ 149 ip_cksum = ((ip_cksum & 0xFFFF0000) >> 16) + 150 (ip_cksum & 0x0000FFFF); 151 if (ip_cksum > 65535) 152 ip_cksum -= 65535; 153 ip_cksum = (~ip_cksum) & 0x0000FFFF; 154 if (ip_cksum == 0) 155 ip_cksum = 0xFFFF; 156 ip_hdr->hdr_checksum = (uint16_t) ip_cksum; 157 } 158 159 static inline void 160 update_pkt_header(struct rte_mbuf *pkt, uint32_t total_pkt_len) 161 { 162 struct rte_ipv4_hdr *ip_hdr; 163 struct rte_udp_hdr *udp_hdr; 164 uint16_t pkt_data_len; 165 uint16_t pkt_len; 166 167 pkt_data_len = (uint16_t) (total_pkt_len - ( 168 sizeof(struct rte_ether_hdr) + 169 sizeof(struct rte_ipv4_hdr) + 170 sizeof(struct rte_udp_hdr))); 171 /* updata udp pkt length */ 172 udp_hdr = rte_pktmbuf_mtod_offset(pkt, struct rte_udp_hdr *, 173 sizeof(struct rte_ether_hdr) + 174 sizeof(struct rte_ipv4_hdr)); 175 pkt_len = (uint16_t) (pkt_data_len + sizeof(struct rte_udp_hdr)); 176 udp_hdr->dgram_len = RTE_CPU_TO_BE_16(pkt_len); 177 178 /* updata ip pkt length and csum */ 179 ip_hdr = rte_pktmbuf_mtod_offset(pkt, struct rte_ipv4_hdr *, 180 sizeof(struct rte_ether_hdr)); 181 ip_hdr->hdr_checksum = 0; 182 pkt_len = (uint16_t) (pkt_len + sizeof(struct rte_ipv4_hdr)); 183 ip_hdr->total_length = RTE_CPU_TO_BE_16(pkt_len); 184 ip_hdr->hdr_checksum = rte_ipv4_cksum(ip_hdr); 185 } 186 187 static inline bool 188 pkt_burst_prepare(struct rte_mbuf *pkt, struct rte_mempool *mbp, 189 struct rte_ether_hdr *eth_hdr, const uint16_t vlan_tci, 190 const uint16_t vlan_tci_outer, const uint64_t ol_flags, 191 const uint16_t idx, const struct fwd_stream *fs) 192 { 193 struct rte_mbuf *pkt_segs[RTE_MAX_SEGS_PER_PKT]; 194 struct rte_mbuf *pkt_seg; 195 uint32_t nb_segs, pkt_len; 196 uint8_t i; 197 198 if (unlikely(tx_pkt_split == TX_PKT_SPLIT_RND)) 199 nb_segs = rte_rand() % tx_pkt_nb_segs + 1; 200 else 201 nb_segs = tx_pkt_nb_segs; 202 203 if (nb_segs > 1) { 204 if (rte_mempool_get_bulk(mbp, (void **)pkt_segs, nb_segs - 1)) 205 return false; 206 } 207 208 rte_pktmbuf_reset_headroom(pkt); 209 pkt->data_len = tx_pkt_seg_lengths[0]; 210 pkt->ol_flags &= EXT_ATTACHED_MBUF; 211 pkt->ol_flags |= ol_flags; 212 pkt->vlan_tci = vlan_tci; 213 pkt->vlan_tci_outer = vlan_tci_outer; 214 pkt->l2_len = sizeof(struct rte_ether_hdr); 215 pkt->l3_len = sizeof(struct rte_ipv4_hdr); 216 217 pkt_len = pkt->data_len; 218 pkt_seg = pkt; 219 for (i = 1; i < nb_segs; i++) { 220 pkt_seg->next = pkt_segs[i - 1]; 221 pkt_seg = pkt_seg->next; 222 pkt_seg->data_len = tx_pkt_seg_lengths[i]; 223 pkt_len += pkt_seg->data_len; 224 } 225 pkt_seg->next = NULL; /* Last segment of packet. */ 226 /* 227 * Copy headers in first packet segment(s). 228 */ 229 copy_buf_to_pkt(eth_hdr, sizeof(*eth_hdr), pkt, 0); 230 copy_buf_to_pkt(&pkt_ip_hdr, sizeof(pkt_ip_hdr), pkt, 231 sizeof(struct rte_ether_hdr)); 232 if (txonly_multi_flow) { 233 uint8_t ip_var = RTE_PER_LCORE(_ip_var); 234 struct rte_ipv4_hdr *ip_hdr; 235 uint32_t addr; 236 237 ip_hdr = rte_pktmbuf_mtod_offset(pkt, 238 struct rte_ipv4_hdr *, 239 sizeof(struct rte_ether_hdr)); 240 /* 241 * Generate multiple flows by varying IP src addr. This 242 * enables packets are well distributed by RSS in 243 * receiver side if any and txonly mode can be a decent 244 * packet generator for developer's quick performance 245 * regression test. 246 */ 247 addr = (tx_ip_dst_addr | (ip_var++ << 8)) + rte_lcore_id(); 248 ip_hdr->src_addr = rte_cpu_to_be_32(addr); 249 RTE_PER_LCORE(_ip_var) = ip_var; 250 } 251 copy_buf_to_pkt(&pkt_udp_hdr, sizeof(pkt_udp_hdr), pkt, 252 sizeof(struct rte_ether_hdr) + 253 sizeof(struct rte_ipv4_hdr)); 254 255 if (unlikely(tx_pkt_split == TX_PKT_SPLIT_RND) || txonly_multi_flow) 256 update_pkt_header(pkt, pkt_len); 257 258 if (unlikely(timestamp_enable)) { 259 uint64_t skew = RTE_PER_LCORE(timestamp_qskew); 260 struct { 261 rte_be32_t signature; 262 rte_be16_t pkt_idx; 263 rte_be16_t queue_idx; 264 rte_be64_t ts; 265 } timestamp_mark; 266 267 if (unlikely(timestamp_init_req != 268 RTE_PER_LCORE(timestamp_idone))) { 269 struct rte_eth_dev *dev = &rte_eth_devices[fs->tx_port]; 270 unsigned int txqs_n = dev->data->nb_tx_queues; 271 uint64_t phase = tx_pkt_times_inter * fs->tx_queue / 272 (txqs_n ? txqs_n : 1); 273 /* 274 * Initialize the scheduling time phase shift 275 * depending on queue index. 276 */ 277 skew = timestamp_initial[fs->tx_port] + 278 tx_pkt_times_inter + phase; 279 RTE_PER_LCORE(timestamp_qskew) = skew; 280 RTE_PER_LCORE(timestamp_idone) = timestamp_init_req; 281 } 282 timestamp_mark.pkt_idx = rte_cpu_to_be_16(idx); 283 timestamp_mark.queue_idx = rte_cpu_to_be_16(fs->tx_queue); 284 timestamp_mark.signature = rte_cpu_to_be_32(0xBEEFC0DE); 285 if (unlikely(!idx)) { 286 skew += tx_pkt_times_inter; 287 pkt->ol_flags |= timestamp_mask; 288 *RTE_MBUF_DYNFIELD 289 (pkt, timestamp_off, uint64_t *) = skew; 290 RTE_PER_LCORE(timestamp_qskew) = skew; 291 timestamp_mark.ts = rte_cpu_to_be_64(skew); 292 } else if (tx_pkt_times_intra) { 293 skew += tx_pkt_times_intra; 294 pkt->ol_flags |= timestamp_mask; 295 *RTE_MBUF_DYNFIELD 296 (pkt, timestamp_off, uint64_t *) = skew; 297 RTE_PER_LCORE(timestamp_qskew) = skew; 298 timestamp_mark.ts = rte_cpu_to_be_64(skew); 299 } else { 300 timestamp_mark.ts = RTE_BE64(0); 301 } 302 copy_buf_to_pkt(×tamp_mark, sizeof(timestamp_mark), pkt, 303 sizeof(struct rte_ether_hdr) + 304 sizeof(struct rte_ipv4_hdr) + 305 sizeof(pkt_udp_hdr)); 306 } 307 /* 308 * Complete first mbuf of packet and append it to the 309 * burst of packets to be transmitted. 310 */ 311 pkt->nb_segs = nb_segs; 312 pkt->pkt_len = pkt_len; 313 314 return true; 315 } 316 317 /* 318 * Transmit a burst of multi-segments packets. 319 */ 320 static void 321 pkt_burst_transmit(struct fwd_stream *fs) 322 { 323 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 324 struct rte_port *txp; 325 struct rte_mbuf *pkt; 326 struct rte_mempool *mbp; 327 struct rte_ether_hdr eth_hdr; 328 uint16_t nb_tx; 329 uint16_t nb_pkt; 330 uint16_t vlan_tci, vlan_tci_outer; 331 uint32_t retry; 332 uint64_t ol_flags = 0; 333 uint64_t tx_offloads; 334 uint64_t start_tsc = 0; 335 336 get_start_cycles(&start_tsc); 337 338 mbp = current_fwd_lcore()->mbp; 339 txp = &ports[fs->tx_port]; 340 tx_offloads = txp->dev_conf.txmode.offloads; 341 vlan_tci = txp->tx_vlan_id; 342 vlan_tci_outer = txp->tx_vlan_id_outer; 343 if (tx_offloads & DEV_TX_OFFLOAD_VLAN_INSERT) 344 ol_flags = PKT_TX_VLAN_PKT; 345 if (tx_offloads & DEV_TX_OFFLOAD_QINQ_INSERT) 346 ol_flags |= PKT_TX_QINQ_PKT; 347 if (tx_offloads & DEV_TX_OFFLOAD_MACSEC_INSERT) 348 ol_flags |= PKT_TX_MACSEC; 349 350 /* 351 * Initialize Ethernet header. 352 */ 353 rte_ether_addr_copy(&peer_eth_addrs[fs->peer_addr], ð_hdr.d_addr); 354 rte_ether_addr_copy(&ports[fs->tx_port].eth_addr, ð_hdr.s_addr); 355 eth_hdr.ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4); 356 357 if (rte_mempool_get_bulk(mbp, (void **)pkts_burst, 358 nb_pkt_per_burst) == 0) { 359 for (nb_pkt = 0; nb_pkt < nb_pkt_per_burst; nb_pkt++) { 360 if (unlikely(!pkt_burst_prepare(pkts_burst[nb_pkt], mbp, 361 ð_hdr, vlan_tci, 362 vlan_tci_outer, 363 ol_flags, 364 nb_pkt, fs))) { 365 rte_mempool_put_bulk(mbp, 366 (void **)&pkts_burst[nb_pkt], 367 nb_pkt_per_burst - nb_pkt); 368 break; 369 } 370 } 371 } else { 372 for (nb_pkt = 0; nb_pkt < nb_pkt_per_burst; nb_pkt++) { 373 pkt = rte_mbuf_raw_alloc(mbp); 374 if (pkt == NULL) 375 break; 376 if (unlikely(!pkt_burst_prepare(pkt, mbp, ð_hdr, 377 vlan_tci, 378 vlan_tci_outer, 379 ol_flags, 380 nb_pkt, fs))) { 381 rte_pktmbuf_free(pkt); 382 break; 383 } 384 pkts_burst[nb_pkt] = pkt; 385 } 386 } 387 388 if (nb_pkt == 0) 389 return; 390 391 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_pkt); 392 393 /* 394 * Retry if necessary 395 */ 396 if (unlikely(nb_tx < nb_pkt) && fs->retry_enabled) { 397 retry = 0; 398 while (nb_tx < nb_pkt && retry++ < burst_tx_retry_num) { 399 rte_delay_us(burst_tx_delay_time); 400 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue, 401 &pkts_burst[nb_tx], nb_pkt - nb_tx); 402 } 403 } 404 fs->tx_packets += nb_tx; 405 406 if (txonly_multi_flow) 407 RTE_PER_LCORE(_ip_var) -= nb_pkt - nb_tx; 408 409 inc_tx_burst_stats(fs, nb_tx); 410 if (unlikely(nb_tx < nb_pkt)) { 411 if (verbose_level > 0 && fs->fwd_dropped == 0) 412 printf("port %d tx_queue %d - drop " 413 "(nb_pkt:%u - nb_tx:%u)=%u packets\n", 414 fs->tx_port, fs->tx_queue, 415 (unsigned) nb_pkt, (unsigned) nb_tx, 416 (unsigned) (nb_pkt - nb_tx)); 417 fs->fwd_dropped += (nb_pkt - nb_tx); 418 do { 419 rte_pktmbuf_free(pkts_burst[nb_tx]); 420 } while (++nb_tx < nb_pkt); 421 } 422 423 get_end_cycles(fs, start_tsc); 424 } 425 426 static void 427 tx_only_begin(portid_t pi) 428 { 429 uint16_t pkt_data_len; 430 int dynf; 431 432 pkt_data_len = (uint16_t) (tx_pkt_length - ( 433 sizeof(struct rte_ether_hdr) + 434 sizeof(struct rte_ipv4_hdr) + 435 sizeof(struct rte_udp_hdr))); 436 setup_pkt_udp_ip_headers(&pkt_ip_hdr, &pkt_udp_hdr, pkt_data_len); 437 438 timestamp_enable = false; 439 timestamp_mask = 0; 440 timestamp_off = -1; 441 RTE_PER_LCORE(timestamp_qskew) = 0; 442 dynf = rte_mbuf_dynflag_lookup 443 (RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME, NULL); 444 if (dynf >= 0) 445 timestamp_mask = 1ULL << dynf; 446 dynf = rte_mbuf_dynfield_lookup 447 (RTE_MBUF_DYNFIELD_TIMESTAMP_NAME, NULL); 448 if (dynf >= 0) 449 timestamp_off = dynf; 450 timestamp_enable = tx_pkt_times_inter && 451 timestamp_mask && 452 timestamp_off >= 0 && 453 !rte_eth_read_clock(pi, ×tamp_initial[pi]); 454 if (timestamp_enable) 455 timestamp_init_req++; 456 /* Make sure all settings are visible on forwarding cores.*/ 457 rte_wmb(); 458 } 459 460 struct fwd_engine tx_only_engine = { 461 .fwd_mode_name = "txonly", 462 .port_fwd_begin = tx_only_begin, 463 .port_fwd_end = NULL, 464 .packet_fwd = pkt_burst_transmit, 465 }; 466