drivers/event/sw/sw_evdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016-2017 Intel Corporation
   3  */
   4
   5 #include <inttypes.h>
   6 #include <string.h>
   7
   8 #include <rte_bus_vdev.h>
   9 #include <rte_kvargs.h>
  10 #include <rte_ring.h>
  11 #include <rte_errno.h>
  12 #include <rte_event_ring.h>
  13 #include <rte_service_component.h>
  14
  15 #include "sw_evdev.h"
  16 #include "iq_chunk.h"
  17
  18 #define EVENTDEV_NAME_SW_PMD event_sw
  19 #define NUMA_NODE_ARG "numa_node"
  20 #define SCHED_QUANTA_ARG "sched_quanta"
  21 #define CREDIT_QUANTA_ARG "credit_quanta"
  22
  23 static void
  24 sw_info_get(struct rte_eventdev *dev, struct rte_event_dev_info *info);
  25
  26 static int
  27 sw_port_link(struct rte_eventdev *dev, void *port, const uint8_t queues[],
  28                 const uint8_t priorities[], uint16_t num)
  29 {
  30         struct sw_port *p = port;
  31         struct sw_evdev *sw = sw_pmd_priv(dev);
  32         int i;
  33
  34         RTE_SET_USED(priorities);
  35         for (i = 0; i < num; i++) {
  36                 struct sw_qid *q = &sw->qids[queues[i]];
  37                 unsigned int j;
  38
  39                 /* check for qid map overflow */
  40                 if (q->cq_num_mapped_cqs >= RTE_DIM(q->cq_map)) {
  41                         rte_errno = -EDQUOT;
  42                         break;
  43                 }
  44
  45                 if (p->is_directed && p->num_qids_mapped > 0) {
  46                         rte_errno = -EDQUOT;
  47                         break;
  48                 }
  49
  50                 for (j = 0; j < q->cq_num_mapped_cqs; j++) {
  51                         if (q->cq_map[j] == p->id)
  52                                 break;
  53                 }
  54
  55                 /* check if port is already linked */
  56                 if (j < q->cq_num_mapped_cqs)
  57                         continue;
  58
  59                 if (q->type == SW_SCHED_TYPE_DIRECT) {
  60                         /* check directed qids only map to one port */
  61                         if (p->num_qids_mapped > 0) {
  62                                 rte_errno = -EDQUOT;
  63                                 break;
  64                         }
  65                         /* check port only takes a directed flow */
  66                         if (num > 1) {
  67                                 rte_errno = -EDQUOT;
  68                                 break;
  69                         }
  70
  71                         p->is_directed = 1;
  72                         p->num_qids_mapped = 1;
  73                 } else if (q->type == RTE_SCHED_TYPE_ORDERED) {
  74                         p->num_ordered_qids++;
  75                         p->num_qids_mapped++;
  76                 } else if (q->type == RTE_SCHED_TYPE_ATOMIC ||
  77                                 q->type == RTE_SCHED_TYPE_PARALLEL) {
  78                         p->num_qids_mapped++;
  79                 }
  80
  81                 q->cq_map[q->cq_num_mapped_cqs] = p->id;
  82                 rte_smp_wmb();
  83                 q->cq_num_mapped_cqs++;
  84         }
  85         return i;
  86 }
  87
  88 static int
  89 sw_port_unlink(struct rte_eventdev *dev, void *port, uint8_t queues[],
  90                 uint16_t nb_unlinks)
  91 {
  92         struct sw_port *p = port;
  93         struct sw_evdev *sw = sw_pmd_priv(dev);
  94         unsigned int i, j;
  95
  96         int unlinked = 0;
  97         for (i = 0; i < nb_unlinks; i++) {
  98                 struct sw_qid *q = &sw->qids[queues[i]];
  99                 for (j = 0; j < q->cq_num_mapped_cqs; j++) {
 100                         if (q->cq_map[j] == p->id) {
 101                                 q->cq_map[j] =
 102                                         q->cq_map[q->cq_num_mapped_cqs - 1];
 103                                 rte_smp_wmb();
 104                                 q->cq_num_mapped_cqs--;
 105                                 unlinked++;
 106
 107                                 p->num_qids_mapped--;
 108
 109                                 if (q->type == RTE_SCHED_TYPE_ORDERED)
 110                                         p->num_ordered_qids--;
 111
 112                                 continue;
 113                         }
 114                 }
 115         }
 116
 117         p->unlinks_in_progress += unlinked;
 118         rte_smp_mb();
 119
 120         return unlinked;
 121 }
 122
 123 static int
 124 sw_port_unlinks_in_progress(struct rte_eventdev *dev, void *port)
 125 {
 126         RTE_SET_USED(dev);
 127         struct sw_port *p = port;
 128         return p->unlinks_in_progress;
 129 }
 130
 131 static int
 132 sw_port_setup(struct rte_eventdev *dev, uint8_t port_id,
 133                 const struct rte_event_port_conf *conf)
 134 {
 135         struct sw_evdev *sw = sw_pmd_priv(dev);
 136         struct sw_port *p = &sw->ports[port_id];
 137         char buf[RTE_RING_NAMESIZE];
 138         unsigned int i;
 139
 140         struct rte_event_dev_info info;
 141         sw_info_get(dev, &info);
 142
 143         /* detect re-configuring and return credits to instance if needed */
 144         if (p->initialized) {
 145                 /* taking credits from pool is done one quanta at a time, and
 146                  * credits may be spend (counted in p->inflights) or still
 147                  * available in the port (p->inflight_credits). We must return
 148                  * the sum to no leak credits
 149                  */
 150                 int possible_inflights = p->inflight_credits + p->inflights;
 151                 rte_atomic32_sub(&sw->inflights, possible_inflights);
 152         }
 153
 154         *p = (struct sw_port){0}; /* zero entire structure */
 155         p->id = port_id;
 156         p->sw = sw;
 157
 158         /* check to see if rings exists - port_setup() can be called multiple
 159          * times legally (assuming device is stopped). If ring exists, free it
 160          * to so it gets re-created with the correct size
 161          */
 162         snprintf(buf, sizeof(buf), "sw%d_p%u_%s", dev->data->dev_id,
 163                         port_id, "rx_worker_ring");
 164         struct rte_event_ring *existing_ring = rte_event_ring_lookup(buf);
 165         if (existing_ring)
 166                 rte_event_ring_free(existing_ring);
 167
 168         p->rx_worker_ring = rte_event_ring_create(buf, MAX_SW_PROD_Q_DEPTH,
 169                         dev->data->socket_id,
 170                         RING_F_SP_ENQ | RING_F_SC_DEQ | RING_F_EXACT_SZ);
 171         if (p->rx_worker_ring == NULL) {
 172                 SW_LOG_ERR("Error creating RX worker ring for port %d\n",
 173                                 port_id);
 174                 return -1;
 175         }
 176
 177         p->inflight_max = conf->new_event_threshold;
 178         p->implicit_release = !conf->disable_implicit_release;
 179
 180         /* check if ring exists, same as rx_worker above */
 181         snprintf(buf, sizeof(buf), "sw%d_p%u, %s", dev->data->dev_id,
 182                         port_id, "cq_worker_ring");
 183         existing_ring = rte_event_ring_lookup(buf);
 184         if (existing_ring)
 185                 rte_event_ring_free(existing_ring);
 186
 187         p->cq_worker_ring = rte_event_ring_create(buf, conf->dequeue_depth,
 188                         dev->data->socket_id,
 189                         RING_F_SP_ENQ | RING_F_SC_DEQ | RING_F_EXACT_SZ);
 190         if (p->cq_worker_ring == NULL) {
 191                 rte_event_ring_free(p->rx_worker_ring);
 192                 SW_LOG_ERR("Error creating CQ worker ring for port %d\n",
 193                                 port_id);
 194                 return -1;
 195         }
 196         sw->cq_ring_space[port_id] = conf->dequeue_depth;
 197
 198         /* set hist list contents to empty */
 199         for (i = 0; i < SW_PORT_HIST_LIST; i++) {
 200                 p->hist_list[i].fid = -1;
 201                 p->hist_list[i].qid = -1;
 202         }
 203         dev->data->ports[port_id] = p;
 204
 205         rte_smp_wmb();
 206         p->initialized = 1;
 207         return 0;
 208 }
 209
 210 static void
 211 sw_port_release(void *port)
 212 {
 213         struct sw_port *p = (void *)port;
 214         if (p == NULL)
 215                 return;
 216
 217         rte_event_ring_free(p->rx_worker_ring);
 218         rte_event_ring_free(p->cq_worker_ring);
 219         memset(p, 0, sizeof(*p));
 220 }
 221
 222 static int32_t
 223 qid_init(struct sw_evdev *sw, unsigned int idx, int type,
 224                 const struct rte_event_queue_conf *queue_conf)
 225 {
 226         unsigned int i;
 227         int dev_id = sw->data->dev_id;
 228         int socket_id = sw->data->socket_id;
 229         char buf[IQ_ROB_NAMESIZE];
 230         struct sw_qid *qid = &sw->qids[idx];
 231
 232         /* Initialize the FID structures to no pinning (-1), and zero packets */
 233         const struct sw_fid_t fid = {.cq = -1, .pcount = 0};
 234         for (i = 0; i < RTE_DIM(qid->fids); i++)
 235                 qid->fids[i] = fid;
 236
 237         qid->id = idx;
 238         qid->type = type;
 239         qid->priority = queue_conf->priority;
 240
 241         if (qid->type == RTE_SCHED_TYPE_ORDERED) {
 242                 char ring_name[RTE_RING_NAMESIZE];
 243                 uint32_t window_size;
 244
 245                 /* rte_ring and window_size_mask require require window_size to
 246                  * be a power-of-2.
 247                  */
 248                 window_size = rte_align32pow2(
 249                                 queue_conf->nb_atomic_order_sequences);
 250
 251                 qid->window_size = window_size - 1;
 252
 253                 if (!window_size) {
 254                         SW_LOG_DBG(
 255                                 "invalid reorder_window_size for ordered queue\n"
 256                                 );
 257                         goto cleanup;
 258                 }
 259
 260                 snprintf(buf, sizeof(buf), "sw%d_iq_%d_rob", dev_id, i);
 261                 qid->reorder_buffer = rte_zmalloc_socket(buf,
 262                                 window_size * sizeof(qid->reorder_buffer[0]),
 263                                 0, socket_id);
 264                 if (!qid->reorder_buffer) {
 265                         SW_LOG_DBG("reorder_buffer malloc failed\n");
 266                         goto cleanup;
 267                 }
 268
 269                 memset(&qid->reorder_buffer[0],
 270                        0,
 271                        window_size * sizeof(qid->reorder_buffer[0]));
 272
 273                 snprintf(ring_name, sizeof(ring_name), "sw%d_q%d_freelist",
 274                                 dev_id, idx);
 275
 276                 /* lookup the ring, and if it already exists, free it */
 277                 struct rte_ring *cleanup = rte_ring_lookup(ring_name);
 278                 if (cleanup)
 279                         rte_ring_free(cleanup);
 280
 281                 qid->reorder_buffer_freelist = rte_ring_create(ring_name,
 282                                 window_size,
 283                                 socket_id,
 284                                 RING_F_SP_ENQ | RING_F_SC_DEQ);
 285                 if (!qid->reorder_buffer_freelist) {
 286                         SW_LOG_DBG("freelist ring create failed");
 287                         goto cleanup;
 288                 }
 289
 290                 /* Populate the freelist with reorder buffer entries. Enqueue
 291                  * 'window_size - 1' entries because the rte_ring holds only
 292                  * that many.
 293                  */
 294                 for (i = 0; i < window_size - 1; i++) {
 295                         if (rte_ring_sp_enqueue(qid->reorder_buffer_freelist,
 296                                                 &qid->reorder_buffer[i]) < 0)
 297                                 goto cleanup;
 298                 }
 299
 300                 qid->reorder_buffer_index = 0;
 301                 qid->cq_next_tx = 0;
 302         }
 303
 304         qid->initialized = 1;
 305
 306         return 0;
 307
 308 cleanup:
 309         if (qid->reorder_buffer) {
 310                 rte_free(qid->reorder_buffer);
 311                 qid->reorder_buffer = NULL;
 312         }
 313
 314         if (qid->reorder_buffer_freelist) {
 315                 rte_ring_free(qid->reorder_buffer_freelist);
 316                 qid->reorder_buffer_freelist = NULL;
 317         }
 318
 319         return -EINVAL;
 320 }
 321
 322 static void
 323 sw_queue_release(struct rte_eventdev *dev, uint8_t id)
 324 {
 325         struct sw_evdev *sw = sw_pmd_priv(dev);
 326         struct sw_qid *qid = &sw->qids[id];
 327
 328         if (qid->type == RTE_SCHED_TYPE_ORDERED) {
 329                 rte_free(qid->reorder_buffer);
 330                 rte_ring_free(qid->reorder_buffer_freelist);
 331         }
 332         memset(qid, 0, sizeof(*qid));
 333 }
 334
 335 static int
 336 sw_queue_setup(struct rte_eventdev *dev, uint8_t queue_id,
 337                 const struct rte_event_queue_conf *conf)
 338 {
 339         int type;
 340
 341         type = conf->schedule_type;
 342
 343         if (RTE_EVENT_QUEUE_CFG_SINGLE_LINK & conf->event_queue_cfg) {
 344                 type = SW_SCHED_TYPE_DIRECT;
 345         } else if (RTE_EVENT_QUEUE_CFG_ALL_TYPES
 346                         & conf->event_queue_cfg) {
 347                 SW_LOG_ERR("QUEUE_CFG_ALL_TYPES not supported\n");
 348                 return -ENOTSUP;
 349         }
 350
 351         struct sw_evdev *sw = sw_pmd_priv(dev);
 352
 353         if (sw->qids[queue_id].initialized)
 354                 sw_queue_release(dev, queue_id);
 355
 356         return qid_init(sw, queue_id, type, conf);
 357 }
 358
 359 static void
 360 sw_init_qid_iqs(struct sw_evdev *sw)
 361 {
 362         int i, j;
 363
 364         /* Initialize the IQ memory of all configured qids */
 365         for (i = 0; i < RTE_EVENT_MAX_QUEUES_PER_DEV; i++) {
 366                 struct sw_qid *qid = &sw->qids[i];
 367
 368                 if (!qid->initialized)
 369                         continue;
 370
 371                 for (j = 0; j < SW_IQS_MAX; j++)
 372                         iq_init(sw, &qid->iq[j]);
 373         }
 374 }
 375
 376 static int
 377 sw_qids_empty(struct sw_evdev *sw)
 378 {
 379         unsigned int i, j;
 380
 381         for (i = 0; i < sw->qid_count; i++) {
 382                 for (j = 0; j < SW_IQS_MAX; j++) {
 383                         if (iq_count(&sw->qids[i].iq[j]))
 384                                 return 0;
 385                 }
 386         }
 387
 388         return 1;
 389 }
 390
 391 static int
 392 sw_ports_empty(struct sw_evdev *sw)
 393 {
 394         unsigned int i;
 395
 396         for (i = 0; i < sw->port_count; i++) {
 397                 if ((rte_event_ring_count(sw->ports[i].rx_worker_ring)) ||
 398                      rte_event_ring_count(sw->ports[i].cq_worker_ring))
 399                         return 0;
 400         }
 401
 402         return 1;
 403 }
 404
 405 static void
 406 sw_drain_ports(struct rte_eventdev *dev)
 407 {
 408         struct sw_evdev *sw = sw_pmd_priv(dev);
 409         eventdev_stop_flush_t flush;
 410         unsigned int i;
 411         uint8_t dev_id;
 412         void *arg;
 413
 414         flush = dev->dev_ops->dev_stop_flush;
 415         dev_id = dev->data->dev_id;
 416         arg = dev->data->dev_stop_flush_arg;
 417
 418         for (i = 0; i < sw->port_count; i++) {
 419                 struct rte_event ev;
 420
 421                 while (rte_event_dequeue_burst(dev_id, i, &ev, 1, 0)) {
 422                         if (flush)
 423                                 flush(dev_id, ev, arg);
 424
 425                         ev.op = RTE_EVENT_OP_RELEASE;
 426                         rte_event_enqueue_burst(dev_id, i, &ev, 1);
 427                 }
 428         }
 429 }
 430
 431 static void
 432 sw_drain_queue(struct rte_eventdev *dev, struct sw_iq *iq)
 433 {
 434         struct sw_evdev *sw = sw_pmd_priv(dev);
 435         eventdev_stop_flush_t flush;
 436         uint8_t dev_id;
 437         void *arg;
 438
 439         flush = dev->dev_ops->dev_stop_flush;
 440         dev_id = dev->data->dev_id;
 441         arg = dev->data->dev_stop_flush_arg;
 442
 443         while (iq_count(iq) > 0) {
 444                 struct rte_event ev;
 445
 446                 iq_dequeue_burst(sw, iq, &ev, 1);
 447
 448                 if (flush)
 449                         flush(dev_id, ev, arg);
 450         }
 451 }
 452
 453 static void
 454 sw_drain_queues(struct rte_eventdev *dev)
 455 {
 456         struct sw_evdev *sw = sw_pmd_priv(dev);
 457         unsigned int i, j;
 458
 459         for (i = 0; i < sw->qid_count; i++) {
 460                 for (j = 0; j < SW_IQS_MAX; j++)
 461                         sw_drain_queue(dev, &sw->qids[i].iq[j]);
 462         }
 463 }
 464
 465 static void
 466 sw_clean_qid_iqs(struct rte_eventdev *dev)
 467 {
 468         struct sw_evdev *sw = sw_pmd_priv(dev);
 469         int i, j;
 470
 471         /* Release the IQ memory of all configured qids */
 472         for (i = 0; i < RTE_EVENT_MAX_QUEUES_PER_DEV; i++) {
 473                 struct sw_qid *qid = &sw->qids[i];
 474
 475                 for (j = 0; j < SW_IQS_MAX; j++) {
 476                         if (!qid->iq[j].head)
 477                                 continue;
 478                         iq_free_chunk_list(sw, qid->iq[j].head);
 479                         qid->iq[j].head = NULL;
 480                 }
 481         }
 482 }
 483
 484 static void
 485 sw_queue_def_conf(struct rte_eventdev *dev, uint8_t queue_id,
 486                                  struct rte_event_queue_conf *conf)
 487 {
 488         RTE_SET_USED(dev);
 489         RTE_SET_USED(queue_id);
 490
 491         static const struct rte_event_queue_conf default_conf = {
 492                 .nb_atomic_flows = 4096,
 493                 .nb_atomic_order_sequences = 1,
 494                 .schedule_type = RTE_SCHED_TYPE_ATOMIC,
 495                 .priority = RTE_EVENT_DEV_PRIORITY_NORMAL,
 496         };
 497
 498         *conf = default_conf;
 499 }
 500
 501 static void
 502 sw_port_def_conf(struct rte_eventdev *dev, uint8_t port_id,
 503                  struct rte_event_port_conf *port_conf)
 504 {
 505         RTE_SET_USED(dev);
 506         RTE_SET_USED(port_id);
 507
 508         port_conf->new_event_threshold = 1024;
 509         port_conf->dequeue_depth = 16;
 510         port_conf->enqueue_depth = 16;
 511         port_conf->disable_implicit_release = 0;
 512 }
 513
 514 static int
 515 sw_dev_configure(const struct rte_eventdev *dev)
 516 {
 517         struct sw_evdev *sw = sw_pmd_priv(dev);
 518         const struct rte_eventdev_data *data = dev->data;
 519         const struct rte_event_dev_config *conf = &data->dev_conf;
 520         int num_chunks, i;
 521
 522         sw->qid_count = conf->nb_event_queues;
 523         sw->port_count = conf->nb_event_ports;
 524         sw->nb_events_limit = conf->nb_events_limit;
 525         rte_atomic32_set(&sw->inflights, 0);
 526
 527         /* Number of chunks sized for worst-case spread of events across IQs */
 528         num_chunks = ((SW_INFLIGHT_EVENTS_TOTAL/SW_EVS_PER_Q_CHUNK)+1) +
 529                         sw->qid_count*SW_IQS_MAX*2;
 530
 531         /* If this is a reconfiguration, free the previous IQ allocation. All
 532          * IQ chunk references were cleaned out of the QIDs in sw_stop(), and
 533          * will be reinitialized in sw_start().
 534          */
 535         if (sw->chunks)
 536                 rte_free(sw->chunks);
 537
 538         sw->chunks = rte_malloc_socket(NULL,
 539                                        sizeof(struct sw_queue_chunk) *
 540                                        num_chunks,
 541                                        0,
 542                                        sw->data->socket_id);
 543         if (!sw->chunks)
 544                 return -ENOMEM;
 545
 546         sw->chunk_list_head = NULL;
 547         for (i = 0; i < num_chunks; i++)
 548                 iq_free_chunk(sw, &sw->chunks[i]);
 549
 550         if (conf->event_dev_cfg & RTE_EVENT_DEV_CFG_PER_DEQUEUE_TIMEOUT)
 551                 return -ENOTSUP;
 552
 553         return 0;
 554 }
 555
 556 struct rte_eth_dev;
 557
 558 static int
 559 sw_eth_rx_adapter_caps_get(const struct rte_eventdev *dev,
 560                         const struct rte_eth_dev *eth_dev,
 561                         uint32_t *caps)
 562 {
 563         RTE_SET_USED(dev);
 564         RTE_SET_USED(eth_dev);
 565         *caps = RTE_EVENT_ETH_RX_ADAPTER_SW_CAP;
 566         return 0;
 567 }
 568
 569 static int
 570 sw_timer_adapter_caps_get(const struct rte_eventdev *dev,
 571                           uint64_t flags,
 572                           uint32_t *caps,
 573                           const struct rte_event_timer_adapter_ops **ops)
 574 {
 575         RTE_SET_USED(dev);
 576         RTE_SET_USED(flags);
 577         *caps = 0;
 578
 579         /* Use default SW ops */
 580         *ops = NULL;
 581
 582         return 0;
 583 }
 584
 585 static int
 586 sw_crypto_adapter_caps_get(const struct rte_eventdev *dev,
 587                            const struct rte_cryptodev *cdev,
 588                            uint32_t *caps)
 589 {
 590         RTE_SET_USED(dev);
 591         RTE_SET_USED(cdev);
 592         *caps = RTE_EVENT_CRYPTO_ADAPTER_SW_CAP;
 593         return 0;
 594 }
 595
 596 static void
 597 sw_info_get(struct rte_eventdev *dev, struct rte_event_dev_info *info)
 598 {
 599         RTE_SET_USED(dev);
 600
 601         static const struct rte_event_dev_info evdev_sw_info = {
 602                         .driver_name = SW_PMD_NAME,
 603                         .max_event_queues = RTE_EVENT_MAX_QUEUES_PER_DEV,
 604                         .max_event_queue_flows = SW_QID_NUM_FIDS,
 605                         .max_event_queue_priority_levels = SW_Q_PRIORITY_MAX,
 606                         .max_event_priority_levels = SW_IQS_MAX,
 607                         .max_event_ports = SW_PORTS_MAX,
 608                         .max_event_port_dequeue_depth = MAX_SW_CONS_Q_DEPTH,
 609                         .max_event_port_enqueue_depth = MAX_SW_PROD_Q_DEPTH,
 610                         .max_num_events = SW_INFLIGHT_EVENTS_TOTAL,
 611                         .event_dev_cap = (
 612                                 RTE_EVENT_DEV_CAP_QUEUE_QOS |
 613                                 RTE_EVENT_DEV_CAP_BURST_MODE |
 614                                 RTE_EVENT_DEV_CAP_EVENT_QOS |
 615                                 RTE_EVENT_DEV_CAP_IMPLICIT_RELEASE_DISABLE|
 616                                 RTE_EVENT_DEV_CAP_RUNTIME_PORT_LINK |
 617                                 RTE_EVENT_DEV_CAP_MULTIPLE_QUEUE_PORT |
 618                                 RTE_EVENT_DEV_CAP_NONSEQ_MODE),
 619         };
 620
 621         *info = evdev_sw_info;
 622 }
 623
 624 static void
 625 sw_dump(struct rte_eventdev *dev, FILE *f)
 626 {
 627         const struct sw_evdev *sw = sw_pmd_priv(dev);
 628
 629         static const char * const q_type_strings[] = {
 630                         "Ordered", "Atomic", "Parallel", "Directed"
 631         };
 632         uint32_t i;
 633         fprintf(f, "EventDev %s: ports %d, qids %d\n", "todo-fix-name",
 634                         sw->port_count, sw->qid_count);
 635
 636         fprintf(f, "\trx   %"PRIu64"\n\tdrop %"PRIu64"\n\ttx   %"PRIu64"\n",
 637                 sw->stats.rx_pkts, sw->stats.rx_dropped, sw->stats.tx_pkts);
 638         fprintf(f, "\tsched calls: %"PRIu64"\n", sw->sched_called);
 639         fprintf(f, "\tsched cq/qid call: %"PRIu64"\n", sw->sched_cq_qid_called);
 640         fprintf(f, "\tsched no IQ enq: %"PRIu64"\n", sw->sched_no_iq_enqueues);
 641         fprintf(f, "\tsched no CQ enq: %"PRIu64"\n", sw->sched_no_cq_enqueues);
 642         uint32_t inflights = rte_atomic32_read(&sw->inflights);
 643         uint32_t credits = sw->nb_events_limit - inflights;
 644         fprintf(f, "\tinflight %d, credits: %d\n", inflights, credits);
 645
 646 #define COL_RED "\x1b[31m"
 647 #define COL_RESET "\x1b[0m"
 648
 649         for (i = 0; i < sw->port_count; i++) {
 650                 int max, j;
 651                 const struct sw_port *p = &sw->ports[i];
 652                 if (!p->initialized) {
 653                         fprintf(f, "  %sPort %d not initialized.%s\n",
 654                                 COL_RED, i, COL_RESET);
 655                         continue;
 656                 }
 657                 fprintf(f, "  Port %d %s\n", i,
 658                         p->is_directed ? " (SingleCons)" : "");
 659                 fprintf(f, "\trx   %"PRIu64"\tdrop %"PRIu64"\ttx   %"PRIu64
 660                         "\t%sinflight %d%s\n", sw->ports[i].stats.rx_pkts,
 661                         sw->ports[i].stats.rx_dropped,
 662                         sw->ports[i].stats.tx_pkts,
 663                         (p->inflights == p->inflight_max) ?
 664                                 COL_RED : COL_RESET,
 665                         sw->ports[i].inflights, COL_RESET);
 666
 667                 fprintf(f, "\tMax New: %u"
 668                         "\tAvg cycles PP: %"PRIu64"\tCredits: %u\n",
 669                         sw->ports[i].inflight_max,
 670                         sw->ports[i].avg_pkt_ticks,
 671                         sw->ports[i].inflight_credits);
 672                 fprintf(f, "\tReceive burst distribution:\n");
 673                 float zp_percent = p->zero_polls * 100.0 / p->total_polls;
 674                 fprintf(f, zp_percent < 10 ? "\t\t0:%.02f%% " : "\t\t0:%.0f%% ",
 675                                 zp_percent);
 676                 for (max = (int)RTE_DIM(p->poll_buckets); max-- > 0;)
 677                         if (p->poll_buckets[max] != 0)
 678                                 break;
 679                 for (j = 0; j <= max; j++) {
 680                         if (p->poll_buckets[j] != 0) {
 681                                 float poll_pc = p->poll_buckets[j] * 100.0 /
 682                                         p->total_polls;
 683                                 fprintf(f, "%u-%u:%.02f%% ",
 684                                         ((j << SW_DEQ_STAT_BUCKET_SHIFT) + 1),
 685                                         ((j+1) << SW_DEQ_STAT_BUCKET_SHIFT),
 686                                         poll_pc);
 687                         }
 688                 }
 689                 fprintf(f, "\n");
 690
 691                 if (p->rx_worker_ring) {
 692                         uint64_t used = rte_event_ring_count(p->rx_worker_ring);
 693                         uint64_t space = rte_event_ring_free_count(
 694                                         p->rx_worker_ring);
 695                         const char *col = (space == 0) ? COL_RED : COL_RESET;
 696                         fprintf(f, "\t%srx ring used: %4"PRIu64"\tfree: %4"
 697                                         PRIu64 COL_RESET"\n", col, used, space);
 698                 } else
 699                         fprintf(f, "\trx ring not initialized.\n");
 700
 701                 if (p->cq_worker_ring) {
 702                         uint64_t used = rte_event_ring_count(p->cq_worker_ring);
 703                         uint64_t space = rte_event_ring_free_count(
 704                                         p->cq_worker_ring);
 705                         const char *col = (space == 0) ? COL_RED : COL_RESET;
 706                         fprintf(f, "\t%scq ring used: %4"PRIu64"\tfree: %4"
 707                                         PRIu64 COL_RESET"\n", col, used, space);
 708                 } else
 709                         fprintf(f, "\tcq ring not initialized.\n");
 710         }
 711
 712         for (i = 0; i < sw->qid_count; i++) {
 713                 const struct sw_qid *qid = &sw->qids[i];
 714                 if (!qid->initialized) {
 715                         fprintf(f, "  %sQueue %d not initialized.%s\n",
 716                                 COL_RED, i, COL_RESET);
 717                         continue;
 718                 }
 719                 int affinities_per_port[SW_PORTS_MAX] = {0};
 720                 uint32_t inflights = 0;
 721
 722                 fprintf(f, "  Queue %d (%s)\n", i, q_type_strings[qid->type]);
 723                 fprintf(f, "\trx   %"PRIu64"\tdrop %"PRIu64"\ttx   %"PRIu64"\n",
 724                         qid->stats.rx_pkts, qid->stats.rx_dropped,
 725                         qid->stats.tx_pkts);
 726                 if (qid->type == RTE_SCHED_TYPE_ORDERED) {
 727                         struct rte_ring *rob_buf_free =
 728                                 qid->reorder_buffer_freelist;
 729                         if (rob_buf_free)
 730                                 fprintf(f, "\tReorder entries in use: %u\n",
 731                                         rte_ring_free_count(rob_buf_free));
 732                         else
 733                                 fprintf(f,
 734                                         "\tReorder buffer not initialized\n");
 735                 }
 736
 737                 uint32_t flow;
 738                 for (flow = 0; flow < RTE_DIM(qid->fids); flow++)
 739                         if (qid->fids[flow].cq != -1) {
 740                                 affinities_per_port[qid->fids[flow].cq]++;
 741                                 inflights += qid->fids[flow].pcount;
 742                         }
 743
 744                 uint32_t port;
 745                 fprintf(f, "\tPer Port Stats:\n");
 746                 for (port = 0; port < sw->port_count; port++) {
 747                         fprintf(f, "\t  Port %d: Pkts: %"PRIu64, port,
 748                                         qid->to_port[port]);
 749                         fprintf(f, "\tFlows: %d\n", affinities_per_port[port]);
 750                 }
 751
 752                 uint32_t iq;
 753                 uint32_t iq_printed = 0;
 754                 for (iq = 0; iq < SW_IQS_MAX; iq++) {
 755                         if (!qid->iq[iq].head) {
 756                                 fprintf(f, "\tiq %d is not initialized.\n", iq);
 757                                 iq_printed = 1;
 758                                 continue;
 759                         }
 760                         uint32_t used = iq_count(&qid->iq[iq]);
 761                         const char *col = COL_RESET;
 762                         if (used > 0) {
 763                                 fprintf(f, "\t%siq %d: Used %d"
 764                                         COL_RESET"\n", col, iq, used);
 765                                 iq_printed = 1;
 766                         }
 767                 }
 768                 if (iq_printed == 0)
 769                         fprintf(f, "\t-- iqs empty --\n");
 770         }
 771 }
 772
 773 static int
 774 sw_start(struct rte_eventdev *dev)
 775 {
 776         unsigned int i, j;
 777         struct sw_evdev *sw = sw_pmd_priv(dev);
 778
 779         rte_service_component_runstate_set(sw->service_id, 1);
 780
 781         /* check a service core is mapped to this service */
 782         if (!rte_service_runstate_get(sw->service_id)) {
 783                 SW_LOG_ERR("Warning: No Service core enabled on service %s\n",
 784                                 sw->service_name);
 785                 return -ENOENT;
 786         }
 787
 788         /* check all ports are set up */
 789         for (i = 0; i < sw->port_count; i++)
 790                 if (sw->ports[i].rx_worker_ring == NULL) {
 791                         SW_LOG_ERR("Port %d not configured\n", i);
 792                         return -ESTALE;
 793                 }
 794
 795         /* check all queues are configured and mapped to ports*/
 796         for (i = 0; i < sw->qid_count; i++)
 797                 if (!sw->qids[i].initialized ||
 798                     sw->qids[i].cq_num_mapped_cqs == 0) {
 799                         SW_LOG_ERR("Queue %d not configured\n", i);
 800                         return -ENOLINK;
 801                 }
 802
 803         /* build up our prioritized array of qids */
 804         /* We don't use qsort here, as if all/multiple entries have the same
 805          * priority, the result is non-deterministic. From "man 3 qsort":
 806          * "If two members compare as equal, their order in the sorted
 807          * array is undefined."
 808          */
 809         uint32_t qidx = 0;
 810         for (j = 0; j <= RTE_EVENT_DEV_PRIORITY_LOWEST; j++) {
 811                 for (i = 0; i < sw->qid_count; i++) {
 812                         if (sw->qids[i].priority == j) {
 813                                 sw->qids_prioritized[qidx] = &sw->qids[i];
 814                                 qidx++;
 815                         }
 816                 }
 817         }
 818
 819         sw_init_qid_iqs(sw);
 820
 821         if (sw_xstats_init(sw) < 0)
 822                 return -EINVAL;
 823
 824         rte_smp_wmb();
 825         sw->started = 1;
 826
 827         return 0;
 828 }
 829
 830 static void
 831 sw_stop(struct rte_eventdev *dev)
 832 {
 833         struct sw_evdev *sw = sw_pmd_priv(dev);
 834         int32_t runstate;
 835
 836         /* Stop the scheduler if it's running */
 837         runstate = rte_service_runstate_get(sw->service_id);
 838         if (runstate == 1)
 839                 rte_service_runstate_set(sw->service_id, 0);
 840
 841         while (rte_service_may_be_active(sw->service_id))
 842                 rte_pause();
 843
 844         /* Flush all events out of the device */
 845         while (!(sw_qids_empty(sw) && sw_ports_empty(sw))) {
 846                 sw_event_schedule(dev);
 847                 sw_drain_ports(dev);
 848                 sw_drain_queues(dev);
 849         }
 850
 851         sw_clean_qid_iqs(dev);
 852         sw_xstats_uninit(sw);
 853         sw->started = 0;
 854         rte_smp_wmb();
 855
 856         if (runstate == 1)
 857                 rte_service_runstate_set(sw->service_id, 1);
 858 }
 859
 860 static int
 861 sw_close(struct rte_eventdev *dev)
 862 {
 863         struct sw_evdev *sw = sw_pmd_priv(dev);
 864         uint32_t i;
 865
 866         for (i = 0; i < sw->qid_count; i++)
 867                 sw_queue_release(dev, i);
 868         sw->qid_count = 0;
 869
 870         for (i = 0; i < sw->port_count; i++)
 871                 sw_port_release(&sw->ports[i]);
 872         sw->port_count = 0;
 873
 874         memset(&sw->stats, 0, sizeof(sw->stats));
 875         sw->sched_called = 0;
 876         sw->sched_no_iq_enqueues = 0;
 877         sw->sched_no_cq_enqueues = 0;
 878         sw->sched_cq_qid_called = 0;
 879
 880         return 0;
 881 }
 882
 883 static int
 884 assign_numa_node(const char *key __rte_unused, const char *value, void *opaque)
 885 {
 886         int *socket_id = opaque;
 887         *socket_id = atoi(value);
 888         if (*socket_id >= RTE_MAX_NUMA_NODES)
 889                 return -1;
 890         return 0;
 891 }
 892
 893 static int
 894 set_sched_quanta(const char *key __rte_unused, const char *value, void *opaque)
 895 {
 896         int *quanta = opaque;
 897         *quanta = atoi(value);
 898         if (*quanta < 0 || *quanta >= 4096)
 899                 return -1;
 900         return 0;
 901 }
 902
 903 static int
 904 set_credit_quanta(const char *key __rte_unused, const char *value, void *opaque)
 905 {
 906         int *credit = opaque;
 907         *credit = atoi(value);
 908         if (*credit < 0 || *credit >= 128)
 909                 return -1;
 910         return 0;
 911 }
 912
 913
 914 static int32_t sw_sched_service_func(void *args)
 915 {
 916         struct rte_eventdev *dev = args;
 917         sw_event_schedule(dev);
 918         return 0;
 919 }
 920
 921 static int
 922 sw_probe(struct rte_vdev_device *vdev)
 923 {
 924         static struct rte_eventdev_ops evdev_sw_ops = {
 925                         .dev_configure = sw_dev_configure,
 926                         .dev_infos_get = sw_info_get,
 927                         .dev_close = sw_close,
 928                         .dev_start = sw_start,
 929                         .dev_stop = sw_stop,
 930                         .dump = sw_dump,
 931
 932                         .queue_def_conf = sw_queue_def_conf,
 933                         .queue_setup = sw_queue_setup,
 934                         .queue_release = sw_queue_release,
 935                         .port_def_conf = sw_port_def_conf,
 936                         .port_setup = sw_port_setup,
 937                         .port_release = sw_port_release,
 938                         .port_link = sw_port_link,
 939                         .port_unlink = sw_port_unlink,
 940                         .port_unlinks_in_progress = sw_port_unlinks_in_progress,
 941
 942                         .eth_rx_adapter_caps_get = sw_eth_rx_adapter_caps_get,
 943
 944                         .timer_adapter_caps_get = sw_timer_adapter_caps_get,
 945
 946                         .crypto_adapter_caps_get = sw_crypto_adapter_caps_get,
 947
 948                         .xstats_get = sw_xstats_get,
 949                         .xstats_get_names = sw_xstats_get_names,
 950                         .xstats_get_by_name = sw_xstats_get_by_name,
 951                         .xstats_reset = sw_xstats_reset,
 952
 953                         .dev_selftest = test_sw_eventdev,
 954         };
 955
 956         static const char *const args[] = {
 957                 NUMA_NODE_ARG,
 958                 SCHED_QUANTA_ARG,
 959                 CREDIT_QUANTA_ARG,
 960                 NULL
 961         };
 962         const char *name;
 963         const char *params;
 964         struct rte_eventdev *dev;
 965         struct sw_evdev *sw;
 966         int socket_id = rte_socket_id();
 967         int sched_quanta  = SW_DEFAULT_SCHED_QUANTA;
 968         int credit_quanta = SW_DEFAULT_CREDIT_QUANTA;
 969
 970         name = rte_vdev_device_name(vdev);
 971         params = rte_vdev_device_args(vdev);
 972         if (params != NULL && params[0] != '\0') {
 973                 struct rte_kvargs *kvlist = rte_kvargs_parse(params, args);
 974
 975                 if (!kvlist) {
 976                         SW_LOG_INFO(
 977                                 "Ignoring unsupported parameters when creating device '%s'\n",
 978                                 name);
 979                 } else {
 980                         int ret = rte_kvargs_process(kvlist, NUMA_NODE_ARG,
 981                                         assign_numa_node, &socket_id);
 982                         if (ret != 0) {
 983                                 SW_LOG_ERR(
 984                                         "%s: Error parsing numa node parameter",
 985                                         name);
 986                                 rte_kvargs_free(kvlist);
 987                                 return ret;
 988                         }
 989
 990                         ret = rte_kvargs_process(kvlist, SCHED_QUANTA_ARG,
 991                                         set_sched_quanta, &sched_quanta);
 992                         if (ret != 0) {
 993                                 SW_LOG_ERR(
 994                                         "%s: Error parsing sched quanta parameter",
 995                                         name);
 996                                 rte_kvargs_free(kvlist);
 997                                 return ret;
 998                         }
 999
1000                         ret = rte_kvargs_process(kvlist, CREDIT_QUANTA_ARG,
1001                                         set_credit_quanta, &credit_quanta);
1002                         if (ret != 0) {
1003                                 SW_LOG_ERR(
1004                                         "%s: Error parsing credit quanta parameter",
1005                                         name);
1006                                 rte_kvargs_free(kvlist);
1007                                 return ret;
1008                         }
1009
1010                         rte_kvargs_free(kvlist);
1011                 }
1012         }
1013
1014         SW_LOG_INFO(
1015                         "Creating eventdev sw device %s, numa_node=%d, sched_quanta=%d, credit_quanta=%d\n",
1016                         name, socket_id, sched_quanta, credit_quanta);
1017
1018         dev = rte_event_pmd_vdev_init(name,
1019                         sizeof(struct sw_evdev), socket_id);
1020         if (dev == NULL) {
1021                 SW_LOG_ERR("eventdev vdev init() failed");
1022                 return -EFAULT;
1023         }
1024         dev->dev_ops = &evdev_sw_ops;
1025         dev->enqueue = sw_event_enqueue;
1026         dev->enqueue_burst = sw_event_enqueue_burst;
1027         dev->enqueue_new_burst = sw_event_enqueue_burst;
1028         dev->enqueue_forward_burst = sw_event_enqueue_burst;
1029         dev->dequeue = sw_event_dequeue;
1030         dev->dequeue_burst = sw_event_dequeue_burst;
1031
1032         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1033                 return 0;
1034
1035         sw = dev->data->dev_private;
1036         sw->data = dev->data;
1037
1038         /* copy values passed from vdev command line to instance */
1039         sw->credit_update_quanta = credit_quanta;
1040         sw->sched_quanta = sched_quanta;
1041
1042         /* register service with EAL */
1043         struct rte_service_spec service;
1044         memset(&service, 0, sizeof(struct rte_service_spec));
1045         snprintf(service.name, sizeof(service.name), "%s_service", name);
1046         snprintf(sw->service_name, sizeof(sw->service_name), "%s_service",
1047                         name);
1048         service.socket_id = socket_id;
1049         service.callback = sw_sched_service_func;
1050         service.callback_userdata = (void *)dev;
1051
1052         int32_t ret = rte_service_component_register(&service, &sw->service_id);
1053         if (ret) {
1054                 SW_LOG_ERR("service register() failed");
1055                 return -ENOEXEC;
1056         }
1057
1058         dev->data->service_inited = 1;
1059         dev->data->service_id = sw->service_id;
1060
1061         return 0;
1062 }
1063
1064 static int
1065 sw_remove(struct rte_vdev_device *vdev)
1066 {
1067         const char *name;
1068
1069         name = rte_vdev_device_name(vdev);
1070         if (name == NULL)
1071                 return -EINVAL;
1072
1073         SW_LOG_INFO("Closing eventdev sw device %s\n", name);
1074
1075         return rte_event_pmd_vdev_uninit(name);
1076 }
1077
1078 static struct rte_vdev_driver evdev_sw_pmd_drv = {
1079         .probe = sw_probe,
1080         .remove = sw_remove
1081 };
1082
1083 RTE_PMD_REGISTER_VDEV(EVENTDEV_NAME_SW_PMD, evdev_sw_pmd_drv);
1084 RTE_PMD_REGISTER_PARAM_STRING(event_sw, NUMA_NODE_ARG "=<int> "
1085                 SCHED_QUANTA_ARG "=<int>" CREDIT_QUANTA_ARG "=<int>");
1086
1087 /* declared extern in header, for access from other .c files */
1088 int eventdev_sw_log_level;
1089
1090 RTE_INIT(evdev_sw_init_log)
1091 {
1092         eventdev_sw_log_level = rte_log_register("pmd.event.sw");
1093         if (eventdev_sw_log_level >= 0)
1094                 rte_log_set_level(eventdev_sw_log_level, RTE_LOG_NOTICE);
1095 }