ceph/src/seastar/dpdk/examples/server_node_efd/node/node.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2017 Intel Corporation
   3  */
   4
   5 #include <stdint.h>
   6 #include <stdio.h>
   7 #include <inttypes.h>
   8 #include <stdarg.h>
   9 #include <errno.h>
  10 #include <sys/queue.h>
  11 #include <stdlib.h>
  12 #include <getopt.h>
  13 #include <string.h>
  14
  15 #include <rte_common.h>
  16 #include <rte_malloc.h>
  17 #include <rte_memory.h>
  18 #include <rte_memzone.h>
  19 #include <rte_eal.h>
  20 #include <rte_atomic.h>
  21 #include <rte_branch_prediction.h>
  22 #include <rte_log.h>
  23 #include <rte_per_lcore.h>
  24 #include <rte_launch.h>
  25 #include <rte_lcore.h>
  26 #include <rte_ring.h>
  27 #include <rte_debug.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_string_fns.h>
  34 #include <rte_ip.h>
  35
  36 #include "common.h"
  37
  38 /* Number of packets to attempt to read from queue */
  39 #define PKT_READ_SIZE  ((uint16_t)32)
  40
  41 /*
  42  * Our node id number - tells us which rx queue to read, and NIC TX
  43  * queue to write to.
  44  */
  45 static uint8_t node_id;
  46
  47 #define MBQ_CAPACITY 32
  48
  49 /* maps input ports to output ports for packets */
  50 static uint16_t output_ports[RTE_MAX_ETHPORTS];
  51
  52 /* buffers up a set of packet that are ready to send */
  53 struct rte_eth_dev_tx_buffer *tx_buffer[RTE_MAX_ETHPORTS];
  54
  55 /* shared data from server. We update statistics here */
  56 static struct tx_stats *tx_stats;
  57
  58 static struct filter_stats *filter_stats;
  59
  60 /*
  61  * print a usage message
  62  */
  63 static void
  64 usage(const char *progname)
  65 {
  66         printf("Usage: %s [EAL args] -- -n <node_id>\n\n", progname);
  67 }
  68
  69 /*
  70  * Convert the node id number from a string to an int.
  71  */
  72 static int
  73 parse_node_num(const char *node)
  74 {
  75         char *end = NULL;
  76         unsigned long temp;
  77
  78         if (node == NULL || *node == '\0')
  79                 return -1;
  80
  81         temp = strtoul(node, &end, 10);
  82         if (end == NULL || *end != '\0')
  83                 return -1;
  84
  85         node_id = (uint8_t)temp;
  86         return 0;
  87 }
  88
  89 /*
  90  * Parse the application arguments to the node app.
  91  */
  92 static int
  93 parse_app_args(int argc, char *argv[])
  94 {
  95         int option_index, opt;
  96         char **argvopt = argv;
  97         const char *progname = NULL;
  98         static struct option lgopts[] = { /* no long options */
  99                 {NULL, 0, 0, 0 }
 100         };
 101         progname = argv[0];
 102
 103         while ((opt = getopt_long(argc, argvopt, "n:", lgopts,
 104                 &option_index)) != EOF) {
 105                 switch (opt) {
 106                 case 'n':
 107                         if (parse_node_num(optarg) != 0) {
 108                                 usage(progname);
 109                                 return -1;
 110                         }
 111                         break;
 112                 default:
 113                         usage(progname);
 114                         return -1;
 115                 }
 116         }
 117         return 0;
 118 }
 119
 120 /*
 121  * Tx buffer error callback
 122  */
 123 static void
 124 flush_tx_error_callback(struct rte_mbuf **unsent, uint16_t count,
 125                 void *userdata) {
 126         int i;
 127         uint16_t port_id = (uintptr_t)userdata;
 128
 129         tx_stats->tx_drop[port_id] += count;
 130
 131         /* free the mbufs which failed from transmit */
 132         for (i = 0; i < count; i++)
 133                 rte_pktmbuf_free(unsent[i]);
 134
 135 }
 136
 137 static void
 138 configure_tx_buffer(uint16_t port_id, uint16_t size)
 139 {
 140         int ret;
 141
 142         /* Initialize TX buffers */
 143         tx_buffer[port_id] = rte_zmalloc_socket("tx_buffer",
 144                         RTE_ETH_TX_BUFFER_SIZE(size), 0,
 145                         rte_eth_dev_socket_id(port_id));
 146         if (tx_buffer[port_id] == NULL)
 147                 rte_exit(EXIT_FAILURE,
 148                         "Cannot allocate buffer for tx on port %u\n", port_id);
 149
 150         rte_eth_tx_buffer_init(tx_buffer[port_id], size);
 151
 152         ret = rte_eth_tx_buffer_set_err_callback(tx_buffer[port_id],
 153                         flush_tx_error_callback, (void *)(intptr_t)port_id);
 154         if (ret < 0)
 155                 rte_exit(EXIT_FAILURE,
 156                         "Cannot set error callback for tx buffer on port %u\n",
 157                         port_id);
 158 }
 159
 160 /*
 161  * set up output ports so that all traffic on port gets sent out
 162  * its paired port. Index using actual port numbers since that is
 163  * what comes in the mbuf structure.
 164  */
 165 static void
 166 configure_output_ports(const struct shared_info *info)
 167 {
 168         int i;
 169
 170         if (info->num_ports > RTE_MAX_ETHPORTS)
 171                 rte_exit(EXIT_FAILURE, "Too many ethernet ports. "
 172                                 "RTE_MAX_ETHPORTS = %u\n",
 173                                 (unsigned int)RTE_MAX_ETHPORTS);
 174         for (i = 0; i < info->num_ports - 1; i += 2) {
 175                 uint8_t p1 = info->id[i];
 176                 uint8_t p2 = info->id[i+1];
 177
 178                 output_ports[p1] = p2;
 179                 output_ports[p2] = p1;
 180
 181                 configure_tx_buffer(p1, MBQ_CAPACITY);
 182                 configure_tx_buffer(p2, MBQ_CAPACITY);
 183
 184         }
 185 }
 186
 187 /*
 188  * Create the hash table that will contain the flows that
 189  * the node will handle, which will be used to decide if packet
 190  * is transmitted or dropped.
 191  */
 192 static struct rte_hash *
 193 create_hash_table(const struct shared_info *info)
 194 {
 195         uint32_t num_flows_node = info->num_flows / info->num_nodes;
 196         char name[RTE_HASH_NAMESIZE];
 197         struct rte_hash *h;
 198
 199         /* create table */
 200         struct rte_hash_parameters hash_params = {
 201                 .entries = num_flows_node * 2, /* table load = 50% */
 202                 .key_len = sizeof(uint32_t), /* Store IPv4 dest IP address */
 203                 .socket_id = rte_socket_id(),
 204                 .hash_func_init_val = 0,
 205         };
 206
 207         snprintf(name, sizeof(name), "hash_table_%d", node_id);
 208         hash_params.name = name;
 209         h = rte_hash_create(&hash_params);
 210
 211         if (h == NULL)
 212                 rte_exit(EXIT_FAILURE,
 213                                 "Problem creating the hash table for node %d\n",
 214                                 node_id);
 215         return h;
 216 }
 217
 218 static void
 219 populate_hash_table(const struct rte_hash *h, const struct shared_info *info)
 220 {
 221         unsigned int i;
 222         int32_t ret;
 223         uint32_t ip_dst;
 224         uint32_t num_flows_node = 0;
 225         uint64_t target_node;
 226
 227         /* Add flows in table */
 228         for (i = 0; i < info->num_flows; i++) {
 229                 target_node = i % info->num_nodes;
 230                 if (target_node != node_id)
 231                         continue;
 232
 233                 ip_dst = rte_cpu_to_be_32(i);
 234
 235                 ret = rte_hash_add_key(h, (void *) &ip_dst);
 236                 if (ret < 0)
 237                         rte_exit(EXIT_FAILURE, "Unable to add entry %u "
 238                                         "in hash table\n", i);
 239                 else
 240                         num_flows_node++;
 241
 242         }
 243
 244         printf("Hash table: Adding 0x%x keys\n", num_flows_node);
 245 }
 246
 247 /*
 248  * This function performs routing of packets
 249  * Just sends each input packet out an output port based solely on the input
 250  * port it arrived on.
 251  */
 252 static inline void
 253 transmit_packet(struct rte_mbuf *buf)
 254 {
 255         int sent;
 256         const uint16_t in_port = buf->port;
 257         const uint16_t out_port = output_ports[in_port];
 258         struct rte_eth_dev_tx_buffer *buffer = tx_buffer[out_port];
 259
 260         sent = rte_eth_tx_buffer(out_port, node_id, buffer, buf);
 261         if (sent)
 262                 tx_stats->tx[out_port] += sent;
 263
 264 }
 265
 266 static inline void
 267 handle_packets(struct rte_hash *h, struct rte_mbuf **bufs, uint16_t num_packets)
 268 {
 269         struct ipv4_hdr *ipv4_hdr;
 270         uint32_t ipv4_dst_ip[PKT_READ_SIZE];
 271         const void *key_ptrs[PKT_READ_SIZE];
 272         unsigned int i;
 273         int32_t positions[PKT_READ_SIZE] = {0};
 274
 275         for (i = 0; i < num_packets; i++) {
 276                 /* Handle IPv4 header.*/
 277                 ipv4_hdr = rte_pktmbuf_mtod_offset(bufs[i], struct ipv4_hdr *,
 278                                 sizeof(struct ether_hdr));
 279                 ipv4_dst_ip[i] = ipv4_hdr->dst_addr;
 280                 key_ptrs[i] = &ipv4_dst_ip[i];
 281         }
 282         /* Check if packets belongs to any flows handled by this node */
 283         rte_hash_lookup_bulk(h, key_ptrs, num_packets, positions);
 284
 285         for (i = 0; i < num_packets; i++) {
 286                 if (likely(positions[i] >= 0)) {
 287                         filter_stats->passed++;
 288                         transmit_packet(bufs[i]);
 289                 } else {
 290                         filter_stats->drop++;
 291                         /* Drop packet, as flow is not handled by this node */
 292                         rte_pktmbuf_free(bufs[i]);
 293                 }
 294         }
 295 }
 296
 297 /*
 298  * Application main function - loops through
 299  * receiving and processing packets. Never returns
 300  */
 301 int
 302 main(int argc, char *argv[])
 303 {
 304         const struct rte_memzone *mz;
 305         struct rte_ring *rx_ring;
 306         struct rte_hash *h;
 307         struct rte_mempool *mp;
 308         struct shared_info *info;
 309         int need_flush = 0; /* indicates whether we have unsent packets */
 310         int retval;
 311         void *pkts[PKT_READ_SIZE];
 312         uint16_t sent;
 313
 314         retval = rte_eal_init(argc, argv);
 315         if (retval  < 0)
 316                 return -1;
 317         argc -= retval;
 318         argv += retval;
 319
 320         if (parse_app_args(argc, argv) < 0)
 321                 rte_exit(EXIT_FAILURE, "Invalid command-line arguments\n");
 322
 323         if (rte_eth_dev_count_avail() == 0)
 324                 rte_exit(EXIT_FAILURE, "No Ethernet ports - bye\n");
 325
 326         rx_ring = rte_ring_lookup(get_rx_queue_name(node_id));
 327         if (rx_ring == NULL)
 328                 rte_exit(EXIT_FAILURE, "Cannot get RX ring - "
 329                                 "is server process running?\n");
 330
 331         mp = rte_mempool_lookup(PKTMBUF_POOL_NAME);
 332         if (mp == NULL)
 333                 rte_exit(EXIT_FAILURE, "Cannot get mempool for mbufs\n");
 334
 335         mz = rte_memzone_lookup(MZ_SHARED_INFO);
 336         if (mz == NULL)
 337                 rte_exit(EXIT_FAILURE, "Cannot get port info structure\n");
 338         info = mz->addr;
 339         tx_stats = &(info->tx_stats[node_id]);
 340         filter_stats = &(info->filter_stats[node_id]);
 341
 342         configure_output_ports(info);
 343
 344         h = create_hash_table(info);
 345
 346         populate_hash_table(h, info);
 347
 348         RTE_LOG(INFO, APP, "Finished Process Init.\n");
 349
 350         printf("\nNode process %d handling packets\n", node_id);
 351         printf("[Press Ctrl-C to quit ...]\n");
 352
 353         for (;;) {
 354                 uint16_t  rx_pkts = PKT_READ_SIZE;
 355                 uint16_t port;
 356
 357                 /*
 358                  * Try dequeuing max possible packets first, if that fails,
 359                  * get the most we can. Loop body should only execute once,
 360                  * maximum
 361                  */
 362                 while (rx_pkts > 0 &&
 363                                 unlikely(rte_ring_dequeue_bulk(rx_ring, pkts,
 364                                         rx_pkts, NULL) == 0))
 365                         rx_pkts = (uint16_t)RTE_MIN(rte_ring_count(rx_ring),
 366                                         PKT_READ_SIZE);
 367
 368                 if (unlikely(rx_pkts == 0)) {
 369                         if (need_flush)
 370                                 for (port = 0; port < info->num_ports; port++) {
 371                                         sent = rte_eth_tx_buffer_flush(
 372                                                         info->id[port],
 373                                                         node_id,
 374                                                         tx_buffer[port]);
 375                                         if (unlikely(sent))
 376                                                 tx_stats->tx[port] += sent;
 377                                 }
 378                         need_flush = 0;
 379                         continue;
 380                 }
 381
 382                 handle_packets(h, (struct rte_mbuf **)pkts, rx_pkts);
 383
 384                 need_flush = 1;
 385         }
 386 }