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 #define UDP_SRC_PORT 1024 44 #define UDP_DST_PORT 1024 45 46 #define IP_SRC_ADDR ((192U << 24) | (168 << 16) | (0 << 8) | 1) 47 #define IP_DST_ADDR ((192U << 24) | (168 << 16) | (0 << 8) | 2) 48 49 #define IP_DEFTTL 64 /* from RFC 1340. */ 50 #define IP_VERSION 0x40 51 #define IP_HDRLEN 0x05 /* default IP header length == five 32-bits words. */ 52 #define IP_VHL_DEF (IP_VERSION | IP_HDRLEN) 53 54 static struct ipv4_hdr pkt_ip_hdr; /**< IP header of transmitted packets. */ 55 RTE_DEFINE_PER_LCORE(uint8_t, _ip_var); /**< IP address variation */ 56 static struct udp_hdr pkt_udp_hdr; /**< UDP header of transmitted packets. */ 57 58 static void 59 copy_buf_to_pkt_segs(void* buf, unsigned len, struct rte_mbuf *pkt, 60 unsigned offset) 61 { 62 struct rte_mbuf *seg; 63 void *seg_buf; 64 unsigned copy_len; 65 66 seg = pkt; 67 while (offset >= seg->data_len) { 68 offset -= seg->data_len; 69 seg = seg->next; 70 } 71 copy_len = seg->data_len - offset; 72 seg_buf = rte_pktmbuf_mtod_offset(seg, char *, offset); 73 while (len > copy_len) { 74 rte_memcpy(seg_buf, buf, (size_t) copy_len); 75 len -= copy_len; 76 buf = ((char*) buf + copy_len); 77 seg = seg->next; 78 seg_buf = rte_pktmbuf_mtod(seg, char *); 79 copy_len = seg->data_len; 80 } 81 rte_memcpy(seg_buf, buf, (size_t) len); 82 } 83 84 static inline void 85 copy_buf_to_pkt(void* buf, unsigned len, struct rte_mbuf *pkt, unsigned offset) 86 { 87 if (offset + len <= pkt->data_len) { 88 rte_memcpy(rte_pktmbuf_mtod_offset(pkt, char *, offset), 89 buf, (size_t) len); 90 return; 91 } 92 copy_buf_to_pkt_segs(buf, len, pkt, offset); 93 } 94 95 static void 96 setup_pkt_udp_ip_headers(struct ipv4_hdr *ip_hdr, 97 struct udp_hdr *udp_hdr, 98 uint16_t pkt_data_len) 99 { 100 uint16_t *ptr16; 101 uint32_t ip_cksum; 102 uint16_t pkt_len; 103 104 /* 105 * Initialize UDP header. 106 */ 107 pkt_len = (uint16_t) (pkt_data_len + sizeof(struct udp_hdr)); 108 udp_hdr->src_port = rte_cpu_to_be_16(UDP_SRC_PORT); 109 udp_hdr->dst_port = rte_cpu_to_be_16(UDP_DST_PORT); 110 udp_hdr->dgram_len = RTE_CPU_TO_BE_16(pkt_len); 111 udp_hdr->dgram_cksum = 0; /* No UDP checksum. */ 112 113 /* 114 * Initialize IP header. 115 */ 116 pkt_len = (uint16_t) (pkt_len + sizeof(struct ipv4_hdr)); 117 ip_hdr->version_ihl = IP_VHL_DEF; 118 ip_hdr->type_of_service = 0; 119 ip_hdr->fragment_offset = 0; 120 ip_hdr->time_to_live = IP_DEFTTL; 121 ip_hdr->next_proto_id = IPPROTO_UDP; 122 ip_hdr->packet_id = 0; 123 ip_hdr->total_length = RTE_CPU_TO_BE_16(pkt_len); 124 ip_hdr->src_addr = rte_cpu_to_be_32(IP_SRC_ADDR); 125 ip_hdr->dst_addr = rte_cpu_to_be_32(IP_DST_ADDR); 126 127 /* 128 * Compute IP header checksum. 129 */ 130 ptr16 = (unaligned_uint16_t*) ip_hdr; 131 ip_cksum = 0; 132 ip_cksum += ptr16[0]; ip_cksum += ptr16[1]; 133 ip_cksum += ptr16[2]; ip_cksum += ptr16[3]; 134 ip_cksum += ptr16[4]; 135 ip_cksum += ptr16[6]; ip_cksum += ptr16[7]; 136 ip_cksum += ptr16[8]; ip_cksum += ptr16[9]; 137 138 /* 139 * Reduce 32 bit checksum to 16 bits and complement it. 140 */ 141 ip_cksum = ((ip_cksum & 0xFFFF0000) >> 16) + 142 (ip_cksum & 0x0000FFFF); 143 if (ip_cksum > 65535) 144 ip_cksum -= 65535; 145 ip_cksum = (~ip_cksum) & 0x0000FFFF; 146 if (ip_cksum == 0) 147 ip_cksum = 0xFFFF; 148 ip_hdr->hdr_checksum = (uint16_t) ip_cksum; 149 } 150 151 /* 152 * Transmit a burst of multi-segments packets. 153 */ 154 static void 155 pkt_burst_transmit(struct fwd_stream *fs) 156 { 157 struct rte_mbuf *pkts_burst[MAX_PKT_BURST]; 158 struct rte_port *txp; 159 struct rte_mbuf *pkt; 160 struct rte_mbuf *pkt_seg; 161 struct rte_mempool *mbp; 162 struct ether_hdr eth_hdr; 163 uint16_t nb_tx; 164 uint16_t nb_pkt; 165 uint16_t vlan_tci, vlan_tci_outer; 166 uint32_t retry; 167 uint64_t ol_flags = 0; 168 uint8_t ip_var = RTE_PER_LCORE(_ip_var); 169 uint8_t i; 170 uint64_t tx_offloads; 171 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 172 uint64_t start_tsc; 173 uint64_t end_tsc; 174 uint64_t core_cycles; 175 #endif 176 uint32_t nb_segs, pkt_len; 177 178 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 179 start_tsc = rte_rdtsc(); 180 #endif 181 182 mbp = current_fwd_lcore()->mbp; 183 txp = &ports[fs->tx_port]; 184 tx_offloads = txp->dev_conf.txmode.offloads; 185 vlan_tci = txp->tx_vlan_id; 186 vlan_tci_outer = txp->tx_vlan_id_outer; 187 if (tx_offloads & DEV_TX_OFFLOAD_VLAN_INSERT) 188 ol_flags = PKT_TX_VLAN_PKT; 189 if (tx_offloads & DEV_TX_OFFLOAD_QINQ_INSERT) 190 ol_flags |= PKT_TX_QINQ_PKT; 191 if (tx_offloads & DEV_TX_OFFLOAD_MACSEC_INSERT) 192 ol_flags |= PKT_TX_MACSEC; 193 for (nb_pkt = 0; nb_pkt < nb_pkt_per_burst; nb_pkt++) { 194 pkt = rte_mbuf_raw_alloc(mbp); 195 if (pkt == NULL) { 196 nomore_mbuf: 197 if (nb_pkt == 0) 198 return; 199 break; 200 } 201 202 /* 203 * Using raw alloc is good to improve performance, 204 * but some consumers may use the headroom and so 205 * decrement data_off. We need to make sure it is 206 * reset to default value. 207 */ 208 rte_pktmbuf_reset_headroom(pkt); 209 pkt->data_len = tx_pkt_seg_lengths[0]; 210 pkt_seg = pkt; 211 if (tx_pkt_split == TX_PKT_SPLIT_RND) 212 nb_segs = random() % tx_pkt_nb_segs + 1; 213 else 214 nb_segs = tx_pkt_nb_segs; 215 pkt_len = pkt->data_len; 216 for (i = 1; i < nb_segs; i++) { 217 pkt_seg->next = rte_mbuf_raw_alloc(mbp); 218 if (pkt_seg->next == NULL) { 219 pkt->nb_segs = i; 220 rte_pktmbuf_free(pkt); 221 goto nomore_mbuf; 222 } 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 /* 230 * Initialize Ethernet header. 231 */ 232 ether_addr_copy(&peer_eth_addrs[fs->peer_addr],ð_hdr.d_addr); 233 ether_addr_copy(&ports[fs->tx_port].eth_addr, ð_hdr.s_addr); 234 eth_hdr.ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4); 235 236 /* 237 * Copy headers in first packet segment(s). 238 */ 239 copy_buf_to_pkt(ð_hdr, sizeof(eth_hdr), pkt, 0); 240 copy_buf_to_pkt(&pkt_ip_hdr, sizeof(pkt_ip_hdr), pkt, 241 sizeof(struct ether_hdr)); 242 if (txonly_multi_flow) { 243 struct ipv4_hdr *ip_hdr; 244 uint32_t addr; 245 246 ip_hdr = rte_pktmbuf_mtod_offset(pkt, 247 struct ipv4_hdr *, 248 sizeof(struct ether_hdr)); 249 /* 250 * Generate multiple flows by varying IP src addr. This 251 * enables packets are well distributed by RSS in 252 * receiver side if any and txonly mode can be a decent 253 * packet generator for developer's quick performance 254 * regression test. 255 */ 256 addr = (IP_DST_ADDR | (ip_var++ << 8)) + rte_lcore_id(); 257 ip_hdr->src_addr = rte_cpu_to_be_32(addr); 258 } 259 copy_buf_to_pkt(&pkt_udp_hdr, sizeof(pkt_udp_hdr), pkt, 260 sizeof(struct ether_hdr) + 261 sizeof(struct ipv4_hdr)); 262 263 /* 264 * Complete first mbuf of packet and append it to the 265 * burst of packets to be transmitted. 266 */ 267 pkt->nb_segs = nb_segs; 268 pkt->pkt_len = pkt_len; 269 pkt->ol_flags = ol_flags; 270 pkt->vlan_tci = vlan_tci; 271 pkt->vlan_tci_outer = vlan_tci_outer; 272 pkt->l2_len = sizeof(struct ether_hdr); 273 pkt->l3_len = sizeof(struct ipv4_hdr); 274 pkts_burst[nb_pkt] = pkt; 275 } 276 nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_pkt); 277 /* 278 * Retry if necessary 279 */ 280 if (unlikely(nb_tx < nb_pkt) && fs->retry_enabled) { 281 retry = 0; 282 while (nb_tx < nb_pkt && retry++ < burst_tx_retry_num) { 283 rte_delay_us(burst_tx_delay_time); 284 nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue, 285 &pkts_burst[nb_tx], nb_pkt - nb_tx); 286 } 287 } 288 fs->tx_packets += nb_tx; 289 290 if (txonly_multi_flow) 291 RTE_PER_LCORE(_ip_var) += nb_tx; 292 293 #ifdef RTE_TEST_PMD_RECORD_BURST_STATS 294 fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; 295 #endif 296 if (unlikely(nb_tx < nb_pkt)) { 297 if (verbose_level > 0 && fs->fwd_dropped == 0) 298 printf("port %d tx_queue %d - drop " 299 "(nb_pkt:%u - nb_tx:%u)=%u packets\n", 300 fs->tx_port, fs->tx_queue, 301 (unsigned) nb_pkt, (unsigned) nb_tx, 302 (unsigned) (nb_pkt - nb_tx)); 303 fs->fwd_dropped += (nb_pkt - nb_tx); 304 do { 305 rte_pktmbuf_free(pkts_burst[nb_tx]); 306 } while (++nb_tx < nb_pkt); 307 } 308 309 #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES 310 end_tsc = rte_rdtsc(); 311 core_cycles = (end_tsc - start_tsc); 312 fs->core_cycles = (uint64_t) (fs->core_cycles + core_cycles); 313 #endif 314 } 315 316 static void 317 tx_only_begin(__attribute__((unused)) portid_t pi) 318 { 319 uint16_t pkt_data_len; 320 321 pkt_data_len = (uint16_t) (tx_pkt_length - (sizeof(struct ether_hdr) + 322 sizeof(struct ipv4_hdr) + 323 sizeof(struct udp_hdr))); 324 setup_pkt_udp_ip_headers(&pkt_ip_hdr, &pkt_udp_hdr, pkt_data_len); 325 } 326 327 struct fwd_engine tx_only_engine = { 328 .fwd_mode_name = "txonly", 329 .port_fwd_begin = tx_only_begin, 330 .port_fwd_end = NULL, 331 .packet_fwd = pkt_burst_transmit, 332 }; 333