app/test-pmd/txonly.c

   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_pci.h>
  32 #include <rte_ether.h>
  33 #include <rte_ethdev.h>
  34 #include <rte_ip.h>
  35 #include <rte_tcp.h>
  36 #include <rte_udp.h>
  37 #include <rte_string_fns.h>
  38 #include <rte_flow.h>
  39
  40 #include "testpmd.h"
  41
  42 struct tx_timestamp {
  43         rte_be32_t signature;
  44         rte_be16_t pkt_idx;
  45         rte_be16_t queue_idx;
  46         rte_be64_t ts;
  47 };
  48
  49 /* use RFC863 Discard Protocol */
  50 uint16_t tx_udp_src_port = 9;
  51 uint16_t tx_udp_dst_port = 9;
  52
  53 /* use RFC5735 / RFC2544 reserved network test addresses */
  54 uint32_t tx_ip_src_addr = (198U << 24) | (18 << 16) | (0 << 8) | 1;
  55 uint32_t tx_ip_dst_addr = (198U << 24) | (18 << 16) | (0 << 8) | 2;
  56
  57 #define IP_DEFTTL  64   /* from RFC 1340. */
  58
  59 static struct rte_ipv4_hdr pkt_ip_hdr; /**< IP header of transmitted packets. */
  60 RTE_DEFINE_PER_LCORE(uint8_t, _ip_var); /**< IP address variation */
  61 static struct rte_udp_hdr pkt_udp_hdr; /**< UDP header of tx packets. */
  62
  63 static uint64_t timestamp_mask; /**< Timestamp dynamic flag mask */
  64 static int32_t timestamp_off; /**< Timestamp dynamic field offset */
  65 static bool timestamp_enable; /**< Timestamp enable */
  66 static uint64_t timestamp_initial[RTE_MAX_ETHPORTS];
  67
  68 static void
  69 copy_buf_to_pkt_segs(void* buf, unsigned len, struct rte_mbuf *pkt,
  70                      unsigned offset)
  71 {
  72         struct rte_mbuf *seg;
  73         void *seg_buf;
  74         unsigned copy_len;
  75
  76         seg = pkt;
  77         while (offset >= seg->data_len) {
  78                 offset -= seg->data_len;
  79                 seg = seg->next;
  80         }
  81         copy_len = seg->data_len - offset;
  82         seg_buf = rte_pktmbuf_mtod_offset(seg, char *, offset);
  83         while (len > copy_len) {
  84                 rte_memcpy(seg_buf, buf, (size_t) copy_len);
  85                 len -= copy_len;
  86                 buf = ((char*) buf + copy_len);
  87                 seg = seg->next;
  88                 seg_buf = rte_pktmbuf_mtod(seg, char *);
  89                 copy_len = seg->data_len;
  90         }
  91         rte_memcpy(seg_buf, buf, (size_t) len);
  92 }
  93
  94 static inline void
  95 copy_buf_to_pkt(void* buf, unsigned len, struct rte_mbuf *pkt, unsigned offset)
  96 {
  97         if (offset + len <= pkt->data_len) {
  98                 rte_memcpy(rte_pktmbuf_mtod_offset(pkt, char *, offset),
  99                         buf, (size_t) len);
 100                 return;
 101         }
 102         copy_buf_to_pkt_segs(buf, len, pkt, offset);
 103 }
 104
 105 static void
 106 setup_pkt_udp_ip_headers(struct rte_ipv4_hdr *ip_hdr,
 107                          struct rte_udp_hdr *udp_hdr,
 108                          uint16_t pkt_data_len)
 109 {
 110         uint16_t *ptr16;
 111         uint32_t ip_cksum;
 112         uint16_t pkt_len;
 113
 114         /*
 115          * Initialize UDP header.
 116          */
 117         pkt_len = (uint16_t) (pkt_data_len + sizeof(struct rte_udp_hdr));
 118         udp_hdr->src_port = rte_cpu_to_be_16(tx_udp_src_port);
 119         udp_hdr->dst_port = rte_cpu_to_be_16(tx_udp_dst_port);
 120         udp_hdr->dgram_len      = RTE_CPU_TO_BE_16(pkt_len);
 121         udp_hdr->dgram_cksum    = 0; /* No UDP checksum. */
 122
 123         /*
 124          * Initialize IP header.
 125          */
 126         pkt_len = (uint16_t) (pkt_len + sizeof(struct rte_ipv4_hdr));
 127         ip_hdr->version_ihl   = RTE_IPV4_VHL_DEF;
 128         ip_hdr->type_of_service   = 0;
 129         ip_hdr->fragment_offset = 0;
 130         ip_hdr->time_to_live   = IP_DEFTTL;
 131         ip_hdr->next_proto_id = IPPROTO_UDP;
 132         ip_hdr->packet_id = 0;
 133         ip_hdr->total_length   = RTE_CPU_TO_BE_16(pkt_len);
 134         ip_hdr->src_addr = rte_cpu_to_be_32(tx_ip_src_addr);
 135         ip_hdr->dst_addr = rte_cpu_to_be_32(tx_ip_dst_addr);
 136
 137         /*
 138          * Compute IP header checksum.
 139          */
 140         ptr16 = (unaligned_uint16_t*) ip_hdr;
 141         ip_cksum = 0;
 142         ip_cksum += ptr16[0]; ip_cksum += ptr16[1];
 143         ip_cksum += ptr16[2]; ip_cksum += ptr16[3];
 144         ip_cksum += ptr16[4];
 145         ip_cksum += ptr16[6]; ip_cksum += ptr16[7];
 146         ip_cksum += ptr16[8]; ip_cksum += ptr16[9];
 147
 148         /*
 149          * Reduce 32 bit checksum to 16 bits and complement it.
 150          */
 151         ip_cksum = ((ip_cksum & 0xFFFF0000) >> 16) +
 152                 (ip_cksum & 0x0000FFFF);
 153         if (ip_cksum > 65535)
 154                 ip_cksum -= 65535;
 155         ip_cksum = (~ip_cksum) & 0x0000FFFF;
 156         if (ip_cksum == 0)
 157                 ip_cksum = 0xFFFF;
 158         ip_hdr->hdr_checksum = (uint16_t) ip_cksum;
 159 }
 160
 161 static inline void
 162 update_pkt_header(struct rte_mbuf *pkt, uint32_t total_pkt_len)
 163 {
 164         struct rte_ipv4_hdr *ip_hdr;
 165         struct rte_udp_hdr *udp_hdr;
 166         uint16_t pkt_data_len;
 167         uint16_t pkt_len;
 168
 169         pkt_data_len = (uint16_t) (total_pkt_len - (
 170                                         sizeof(struct rte_ether_hdr) +
 171                                         sizeof(struct rte_ipv4_hdr) +
 172                                         sizeof(struct rte_udp_hdr)));
 173         /* update UDP packet length */
 174         udp_hdr = rte_pktmbuf_mtod_offset(pkt, struct rte_udp_hdr *,
 175                                 sizeof(struct rte_ether_hdr) +
 176                                 sizeof(struct rte_ipv4_hdr));
 177         pkt_len = (uint16_t) (pkt_data_len + sizeof(struct rte_udp_hdr));
 178         udp_hdr->dgram_len = RTE_CPU_TO_BE_16(pkt_len);
 179
 180         /* update IP packet length and checksum */
 181         ip_hdr = rte_pktmbuf_mtod_offset(pkt, struct rte_ipv4_hdr *,
 182                                 sizeof(struct rte_ether_hdr));
 183         ip_hdr->hdr_checksum = 0;
 184         pkt_len = (uint16_t) (pkt_len + sizeof(struct rte_ipv4_hdr));
 185         ip_hdr->total_length = RTE_CPU_TO_BE_16(pkt_len);
 186         ip_hdr->hdr_checksum = rte_ipv4_cksum(ip_hdr);
 187 }
 188
 189 static inline bool
 190 pkt_burst_prepare(struct rte_mbuf *pkt, struct rte_mempool *mbp,
 191                 struct rte_ether_hdr *eth_hdr, const uint16_t vlan_tci,
 192                 const uint16_t vlan_tci_outer, const uint64_t ol_flags,
 193                 const uint16_t idx, struct fwd_stream *fs)
 194 {
 195         struct rte_mbuf *pkt_segs[RTE_MAX_SEGS_PER_PKT];
 196         struct rte_mbuf *pkt_seg;
 197         uint32_t nb_segs, pkt_len;
 198         uint8_t i;
 199
 200         if (unlikely(tx_pkt_split == TX_PKT_SPLIT_RND))
 201                 nb_segs = rte_rand() % tx_pkt_nb_segs + 1;
 202         else
 203                 nb_segs = tx_pkt_nb_segs;
 204
 205         if (nb_segs > 1) {
 206                 if (rte_mempool_get_bulk(mbp, (void **)pkt_segs, nb_segs - 1))
 207                         return false;
 208         }
 209
 210         rte_pktmbuf_reset_headroom(pkt);
 211         pkt->data_len = tx_pkt_seg_lengths[0];
 212         pkt->ol_flags &= RTE_MBUF_F_EXTERNAL;
 213         pkt->ol_flags |= ol_flags;
 214         pkt->vlan_tci = vlan_tci;
 215         pkt->vlan_tci_outer = vlan_tci_outer;
 216         pkt->l2_len = sizeof(struct rte_ether_hdr);
 217         pkt->l3_len = sizeof(struct rte_ipv4_hdr);
 218
 219         pkt_len = pkt->data_len;
 220         pkt_seg = pkt;
 221         for (i = 1; i < nb_segs; i++) {
 222                 pkt_seg->next = pkt_segs[i - 1];
 223                 pkt_seg = pkt_seg->next;
 224                 pkt_seg->data_len = tx_pkt_seg_lengths[i];
 225                 pkt_len += pkt_seg->data_len;
 226         }
 227         pkt_seg->next = NULL; /* Last segment of packet. */
 228         /*
 229          * Copy headers in first packet segment(s).
 230          */
 231         copy_buf_to_pkt(eth_hdr, sizeof(*eth_hdr), pkt, 0);
 232         copy_buf_to_pkt(&pkt_ip_hdr, sizeof(pkt_ip_hdr), pkt,
 233                         sizeof(struct rte_ether_hdr));
 234         if (txonly_multi_flow) {
 235                 uint8_t  ip_var = RTE_PER_LCORE(_ip_var);
 236                 struct rte_ipv4_hdr *ip_hdr;
 237                 uint32_t addr;
 238
 239                 ip_hdr = rte_pktmbuf_mtod_offset(pkt,
 240                                 struct rte_ipv4_hdr *,
 241                                 sizeof(struct rte_ether_hdr));
 242                 /*
 243                  * Generate multiple flows by varying IP src addr. This
 244                  * enables packets are well distributed by RSS in
 245                  * receiver side if any and txonly mode can be a decent
 246                  * packet generator for developer's quick performance
 247                  * regression test.
 248                  */
 249                 addr = (tx_ip_dst_addr | (ip_var++ << 8)) + rte_lcore_id();
 250                 ip_hdr->src_addr = rte_cpu_to_be_32(addr);
 251                 RTE_PER_LCORE(_ip_var) = ip_var;
 252         }
 253         copy_buf_to_pkt(&pkt_udp_hdr, sizeof(pkt_udp_hdr), pkt,
 254                         sizeof(struct rte_ether_hdr) +
 255                         sizeof(struct rte_ipv4_hdr));
 256
 257         if (unlikely(tx_pkt_split == TX_PKT_SPLIT_RND) || txonly_multi_flow)
 258                 update_pkt_header(pkt, pkt_len);
 259
 260         if (unlikely(timestamp_enable)) {
 261                 uint64_t skew = fs->ts_skew;
 262                 struct tx_timestamp timestamp_mark;
 263
 264                 if (unlikely(!skew)) {
 265                         struct rte_eth_dev_info dev_info;
 266                         unsigned int txqs_n;
 267                         uint64_t phase;
 268                         int ret;
 269
 270                         ret = eth_dev_info_get_print_err(fs->tx_port, &dev_info);
 271                         if (ret != 0) {
 272                                 TESTPMD_LOG(ERR,
 273                                         "Failed to get device info for port %d,"
 274                                         "could not finish timestamp init",
 275                                         fs->tx_port);
 276                                 return false;
 277                         }
 278                         txqs_n = dev_info.nb_tx_queues;
 279                         phase = tx_pkt_times_inter * fs->tx_queue /
 280                                          (txqs_n ? txqs_n : 1);
 281                         /*
 282                          * Initialize the scheduling time phase shift
 283                          * depending on queue index.
 284                          */
 285                         skew = timestamp_initial[fs->tx_port] +
 286                                tx_pkt_times_inter + phase;
 287                         fs->ts_skew = skew;
 288                 }
 289                 timestamp_mark.pkt_idx = rte_cpu_to_be_16(idx);
 290                 timestamp_mark.queue_idx = rte_cpu_to_be_16(fs->tx_queue);
 291                 timestamp_mark.signature = rte_cpu_to_be_32(0xBEEFC0DE);
 292                 if (unlikely(!idx)) {
 293                         skew += tx_pkt_times_inter;
 294                         pkt->ol_flags |= timestamp_mask;
 295                         *RTE_MBUF_DYNFIELD
 296                                 (pkt, timestamp_off, uint64_t *) = skew;
 297                         fs->ts_skew = skew;
 298                         timestamp_mark.ts = rte_cpu_to_be_64(skew);
 299                 } else if (tx_pkt_times_intra) {
 300                         skew += tx_pkt_times_intra;
 301                         pkt->ol_flags |= timestamp_mask;
 302                         *RTE_MBUF_DYNFIELD
 303                                 (pkt, timestamp_off, uint64_t *) = skew;
 304                         fs->ts_skew = skew;
 305                         timestamp_mark.ts = rte_cpu_to_be_64(skew);
 306                 } else {
 307                         timestamp_mark.ts = RTE_BE64(0);
 308                 }
 309                 copy_buf_to_pkt(&timestamp_mark, sizeof(timestamp_mark), pkt,
 310                         sizeof(struct rte_ether_hdr) +
 311                         sizeof(struct rte_ipv4_hdr) +
 312                         sizeof(pkt_udp_hdr));
 313         }
 314         /*
 315          * Complete first mbuf of packet and append it to the
 316          * burst of packets to be transmitted.
 317          */
 318         pkt->nb_segs = nb_segs;
 319         pkt->pkt_len = pkt_len;
 320
 321         return true;
 322 }
 323
 324 /*
 325  * Transmit a burst of multi-segments packets.
 326  */
 327 static void
 328 pkt_burst_transmit(struct fwd_stream *fs)
 329 {
 330         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 331         struct rte_port *txp;
 332         struct rte_mbuf *pkt;
 333         struct rte_mempool *mbp;
 334         struct rte_ether_hdr eth_hdr;
 335         uint16_t nb_tx;
 336         uint16_t nb_pkt;
 337         uint16_t vlan_tci, vlan_tci_outer;
 338         uint32_t retry;
 339         uint64_t ol_flags = 0;
 340         uint64_t tx_offloads;
 341         uint64_t start_tsc = 0;
 342
 343         get_start_cycles(&start_tsc);
 344
 345         mbp = current_fwd_lcore()->mbp;
 346         txp = &ports[fs->tx_port];
 347         tx_offloads = txp->dev_conf.txmode.offloads;
 348         vlan_tci = txp->tx_vlan_id;
 349         vlan_tci_outer = txp->tx_vlan_id_outer;
 350         if (tx_offloads & RTE_ETH_TX_OFFLOAD_VLAN_INSERT)
 351                 ol_flags = RTE_MBUF_F_TX_VLAN;
 352         if (tx_offloads & RTE_ETH_TX_OFFLOAD_QINQ_INSERT)
 353                 ol_flags |= RTE_MBUF_F_TX_QINQ;
 354         if (tx_offloads & RTE_ETH_TX_OFFLOAD_MACSEC_INSERT)
 355                 ol_flags |= RTE_MBUF_F_TX_MACSEC;
 356
 357         /*
 358          * Initialize Ethernet header.
 359          */
 360         rte_ether_addr_copy(&peer_eth_addrs[fs->peer_addr], &eth_hdr.dst_addr);
 361         rte_ether_addr_copy(&ports[fs->tx_port].eth_addr, &eth_hdr.src_addr);
 362         eth_hdr.ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
 363
 364         if (rte_mempool_get_bulk(mbp, (void **)pkts_burst,
 365                                 nb_pkt_per_burst) == 0) {
 366                 for (nb_pkt = 0; nb_pkt < nb_pkt_per_burst; nb_pkt++) {
 367                         if (unlikely(!pkt_burst_prepare(pkts_burst[nb_pkt], mbp,
 368                                                         &eth_hdr, vlan_tci,
 369                                                         vlan_tci_outer,
 370                                                         ol_flags,
 371                                                         nb_pkt, fs))) {
 372                                 rte_mempool_put_bulk(mbp,
 373                                                 (void **)&pkts_burst[nb_pkt],
 374                                                 nb_pkt_per_burst - nb_pkt);
 375                                 break;
 376                         }
 377                 }
 378         } else {
 379                 for (nb_pkt = 0; nb_pkt < nb_pkt_per_burst; nb_pkt++) {
 380                         pkt = rte_mbuf_raw_alloc(mbp);
 381                         if (pkt == NULL)
 382                                 break;
 383                         if (unlikely(!pkt_burst_prepare(pkt, mbp, &eth_hdr,
 384                                                         vlan_tci,
 385                                                         vlan_tci_outer,
 386                                                         ol_flags,
 387                                                         nb_pkt, fs))) {
 388                                 rte_pktmbuf_free(pkt);
 389                                 break;
 390                         }
 391                         pkts_burst[nb_pkt] = pkt;
 392                 }
 393         }
 394
 395         if (nb_pkt == 0)
 396                 return;
 397
 398         nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_pkt);
 399
 400         /*
 401          * Retry if necessary
 402          */
 403         if (unlikely(nb_tx < nb_pkt) && fs->retry_enabled) {
 404                 retry = 0;
 405                 while (nb_tx < nb_pkt && retry++ < burst_tx_retry_num) {
 406                         rte_delay_us(burst_tx_delay_time);
 407                         nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
 408                                         &pkts_burst[nb_tx], nb_pkt - nb_tx);
 409                 }
 410         }
 411         fs->tx_packets += nb_tx;
 412
 413         if (txonly_multi_flow)
 414                 RTE_PER_LCORE(_ip_var) -= nb_pkt - nb_tx;
 415
 416         inc_tx_burst_stats(fs, nb_tx);
 417         if (unlikely(nb_tx < nb_pkt)) {
 418                 if (verbose_level > 0 && fs->fwd_dropped == 0)
 419                         printf("port %d tx_queue %d - drop "
 420                                "(nb_pkt:%u - nb_tx:%u)=%u packets\n",
 421                                fs->tx_port, fs->tx_queue,
 422                                (unsigned) nb_pkt, (unsigned) nb_tx,
 423                                (unsigned) (nb_pkt - nb_tx));
 424                 fs->fwd_dropped += (nb_pkt - nb_tx);
 425                 do {
 426                         rte_pktmbuf_free(pkts_burst[nb_tx]);
 427                 } while (++nb_tx < nb_pkt);
 428         }
 429
 430         get_end_cycles(fs, start_tsc);
 431 }
 432
 433 static int
 434 tx_only_begin(portid_t pi)
 435 {
 436         uint16_t pkt_hdr_len, pkt_data_len;
 437         int dynf;
 438
 439         pkt_hdr_len = (uint16_t)(sizeof(struct rte_ether_hdr) +
 440                                  sizeof(struct rte_ipv4_hdr) +
 441                                  sizeof(struct rte_udp_hdr));
 442         pkt_data_len = tx_pkt_length - pkt_hdr_len;
 443
 444         if ((tx_pkt_split == TX_PKT_SPLIT_RND || txonly_multi_flow) &&
 445             tx_pkt_seg_lengths[0] < pkt_hdr_len) {
 446                 TESTPMD_LOG(ERR,
 447                             "Random segment number or multiple flow is enabled, "
 448                             "but tx_pkt_seg_lengths[0] %u < %u (needed)\n",
 449                             tx_pkt_seg_lengths[0], pkt_hdr_len);
 450                 return -EINVAL;
 451         }
 452
 453         setup_pkt_udp_ip_headers(&pkt_ip_hdr, &pkt_udp_hdr, pkt_data_len);
 454
 455         timestamp_enable = false;
 456         timestamp_mask = 0;
 457         timestamp_off = -1;
 458         dynf = rte_mbuf_dynflag_lookup
 459                                 (RTE_MBUF_DYNFLAG_TX_TIMESTAMP_NAME, NULL);
 460         if (dynf >= 0)
 461                 timestamp_mask = 1ULL << dynf;
 462         dynf = rte_mbuf_dynfield_lookup
 463                                 (RTE_MBUF_DYNFIELD_TIMESTAMP_NAME, NULL);
 464         if (dynf >= 0)
 465                 timestamp_off = dynf;
 466         timestamp_enable = tx_pkt_times_inter &&
 467                            timestamp_mask &&
 468                            timestamp_off >= 0 &&
 469                            !rte_eth_read_clock(pi, &timestamp_initial[pi]);
 470
 471         if (timestamp_enable) {
 472                 pkt_hdr_len += sizeof(struct tx_timestamp);
 473
 474                 if (tx_pkt_split == TX_PKT_SPLIT_RND) {
 475                         if (tx_pkt_seg_lengths[0] < pkt_hdr_len) {
 476                                 TESTPMD_LOG(ERR,
 477                                             "Time stamp and random segment number are enabled, "
 478                                             "but tx_pkt_seg_lengths[0] %u < %u (needed)\n",
 479                                             tx_pkt_seg_lengths[0], pkt_hdr_len);
 480                                 return -EINVAL;
 481                         }
 482                 } else {
 483                         uint16_t total = 0;
 484                         uint8_t i;
 485
 486                         for (i = 0; i < tx_pkt_nb_segs; i++) {
 487                                 total += tx_pkt_seg_lengths[i];
 488                                 if (total >= pkt_hdr_len)
 489                                         break;
 490                         }
 491
 492                         if (total < pkt_hdr_len) {
 493                                 TESTPMD_LOG(ERR,
 494                                             "Not enough Tx segment space for time stamp info, "
 495                                             "total %u < %u (needed)\n",
 496                                             total, pkt_hdr_len);
 497                                 return -EINVAL;
 498                         }
 499                 }
 500         }
 501
 502         /* Make sure all settings are visible on forwarding cores.*/
 503         rte_wmb();
 504         return 0;
 505 }
 506
 507 static void
 508 tx_only_stream_init(struct fwd_stream *fs)
 509 {
 510         fs->disabled = ports[fs->tx_port].txq[fs->tx_queue].state ==
 511                                                 RTE_ETH_QUEUE_STATE_STOPPED;
 512 }
 513
 514 struct fwd_engine tx_only_engine = {
 515         .fwd_mode_name  = "txonly",
 516         .port_fwd_begin = tx_only_begin,
 517         .port_fwd_end   = NULL,
 518         .stream_init    = tx_only_stream_init,
 519         .packet_fwd     = pkt_burst_transmit,
 520 };