drivers/net/mpipe/mpipe_tilegx.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2015 EZchip Semiconductor Ltd. All rights reserved.
   5  *
   6  *   Redistribution and use in source and binary forms, with or without
   7  *   modification, are permitted provided that the following conditions
   8  *   are met:
   9  *
  10  *     * Redistributions of source code must retain the above copyright
  11  *       notice, this list of conditions and the following disclaimer.
  12  *     * Redistributions in binary form must reproduce the above copyright
  13  *       notice, this list of conditions and the following disclaimer in
  14  *       the documentation and/or other materials provided with the
  15  *       distribution.
  16  *     * Neither the name of EZchip Semiconductor nor the names of its
  17  *       contributors may be used to endorse or promote products derived
  18  *       from this software without specific prior written permission.
  19  *
  20  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  21  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  22  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  23  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  24  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  25  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  26  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  27  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  28  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  29  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  30  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  31  */
  32
  33 #include <unistd.h>
  34
  35 #include <rte_eal.h>
  36 #include <rte_dev.h>
  37 #include <rte_eal_memconfig.h>
  38 #include <rte_ethdev.h>
  39 #include <rte_malloc.h>
  40 #include <rte_cycles.h>
  41
  42 #include <arch/mpipe_xaui_def.h>
  43 #include <arch/mpipe_gbe_def.h>
  44
  45 #include <gxio/mpipe.h>
  46
  47 #ifdef RTE_LIBRTE_MPIPE_PMD_DEBUG
  48 #define PMD_DEBUG_RX(...)       RTE_LOG(DEBUG, PMD, __VA_ARGS__)
  49 #define PMD_DEBUG_TX(...)       RTE_LOG(DEBUG, PMD, __VA_ARGS__)
  50 #else
  51 #define PMD_DEBUG_RX(...)
  52 #define PMD_DEBUG_TX(...)
  53 #endif
  54
  55 #define MPIPE_MAX_CHANNELS              128
  56 #define MPIPE_TX_MAX_QUEUES             128
  57 #define MPIPE_RX_MAX_QUEUES             16
  58 #define MPIPE_TX_DESCS                  512
  59 #define MPIPE_RX_BUCKETS                256
  60 #define MPIPE_RX_STACK_SIZE             65536
  61 #define MPIPE_RX_IP_ALIGN               2
  62 #define MPIPE_BSM_ALIGN                 128
  63
  64 #define MPIPE_LINK_UPDATE_TIMEOUT       10      /*  s */
  65 #define MPIPE_LINK_UPDATE_INTERVAL      100000  /* us */
  66
  67 struct mpipe_channel_config {
  68         int enable;
  69         int first_bucket;
  70         int num_buckets;
  71         int head_room;
  72         gxio_mpipe_rules_stacks_t stacks;
  73 };
  74
  75 struct mpipe_context {
  76         rte_spinlock_t        lock;
  77         gxio_mpipe_context_t  context;
  78         struct mpipe_channel_config channels[MPIPE_MAX_CHANNELS];
  79 };
  80
  81 static struct mpipe_context mpipe_contexts[GXIO_MPIPE_INSTANCE_MAX];
  82 static int mpipe_instances;
  83 static const char *drivername = "MPIPE PMD";
  84
  85 /* Per queue statistics. */
  86 struct mpipe_queue_stats {
  87         uint64_t packets, bytes, errors, nomem;
  88 };
  89
  90 /* Common tx/rx queue fields. */
  91 struct mpipe_queue {
  92         struct mpipe_dev_priv *priv;    /* "priv" data of its device. */
  93         uint16_t nb_desc;               /* Number of tx descriptors. */
  94         uint16_t port_id;               /* Device index. */
  95         uint16_t stat_idx;              /* Queue stats index. */
  96         uint8_t queue_idx;              /* Queue index. */
  97         uint8_t link_status;            /* 0 = link down. */
  98         struct mpipe_queue_stats stats; /* Stat data for the queue. */
  99 };
 100
 101 /* Transmit queue description. */
 102 struct mpipe_tx_queue {
 103         struct mpipe_queue q;           /* Common stuff. */
 104 };
 105
 106 /* Receive queue description. */
 107 struct mpipe_rx_queue {
 108         struct mpipe_queue q;           /* Common stuff. */
 109         gxio_mpipe_iqueue_t iqueue;     /* mPIPE iqueue. */
 110         gxio_mpipe_idesc_t *next_desc;  /* Next idesc to process. */
 111         int avail_descs;                /* Number of available descs. */
 112         void *rx_ring_mem;              /* DMA ring memory. */
 113 };
 114
 115 struct mpipe_dev_priv {
 116         gxio_mpipe_context_t *context;  /* mPIPE context. */
 117         gxio_mpipe_link_t link;         /* mPIPE link for the device. */
 118         gxio_mpipe_equeue_t equeue;     /* mPIPE equeue. */
 119         unsigned equeue_size;           /* mPIPE equeue desc count. */
 120         int instance;                   /* mPIPE instance. */
 121         int ering;                      /* mPIPE eDMA ring. */
 122         int stack;                      /* mPIPE buffer stack. */
 123         int channel;                    /* Device channel. */
 124         int port_id;                    /* DPDK port index. */
 125         struct rte_eth_dev *eth_dev;    /* DPDK device. */
 126         struct rte_pci_device pci_dev;  /* PCI device data. */
 127         struct rte_mbuf **tx_comps;     /* TX completion array. */
 128         struct rte_mempool *rx_mpool;   /* mpool used by the rx queues. */
 129         unsigned rx_offset;             /* Receive head room. */
 130         unsigned rx_size_code;          /* mPIPE rx buffer size code. */
 131         unsigned rx_buffers;            /* receive buffers on stack. */
 132         int is_xaui:1,                  /* Is this an xgbe or gbe? */
 133             initialized:1,              /* Initialized port? */
 134             running:1;                  /* Running port? */
 135         struct ether_addr mac_addr;     /* MAC address. */
 136         unsigned nb_rx_queues;          /* Configured tx queues. */
 137         unsigned nb_tx_queues;          /* Configured rx queues. */
 138         int first_bucket;               /* mPIPE bucket start index. */
 139         int first_ring;                 /* mPIPE notif ring start index. */
 140         int notif_group;                /* mPIPE notif group. */
 141         rte_atomic32_t dp_count;        /* Active datapath thread count. */
 142         int tx_stat_mapping[RTE_ETHDEV_QUEUE_STAT_CNTRS];
 143         int rx_stat_mapping[RTE_ETHDEV_QUEUE_STAT_CNTRS];
 144 };
 145
 146 #define mpipe_priv(dev)                 \
 147         ((struct mpipe_dev_priv*)(dev)->data->dev_private)
 148
 149 #define mpipe_name(priv)                \
 150         ((priv)->eth_dev->data->name)
 151
 152 #define mpipe_rx_queue(priv, n)         \
 153         ((struct mpipe_rx_queue *)(priv)->eth_dev->data->rx_queues[n])
 154
 155 #define mpipe_tx_queue(priv, n)         \
 156         ((struct mpipe_tx_queue *)(priv)->eth_dev->data->tx_queues[n])
 157
 158 static void
 159 mpipe_xmit_flush(struct mpipe_dev_priv *priv);
 160
 161 static void
 162 mpipe_recv_flush(struct mpipe_dev_priv *priv);
 163
 164 static int mpipe_equeue_sizes[] = {
 165         [GXIO_MPIPE_EQUEUE_ENTRY_512]   = 512,
 166         [GXIO_MPIPE_EQUEUE_ENTRY_2K]    = 2048,
 167         [GXIO_MPIPE_EQUEUE_ENTRY_8K]    = 8192,
 168         [GXIO_MPIPE_EQUEUE_ENTRY_64K]   = 65536,
 169 };
 170
 171 static int mpipe_iqueue_sizes[] = {
 172         [GXIO_MPIPE_IQUEUE_ENTRY_128]   = 128,
 173         [GXIO_MPIPE_IQUEUE_ENTRY_512]   = 512,
 174         [GXIO_MPIPE_IQUEUE_ENTRY_2K]    = 2048,
 175         [GXIO_MPIPE_IQUEUE_ENTRY_64K]   = 65536,
 176 };
 177
 178 static int mpipe_buffer_sizes[] = {
 179         [GXIO_MPIPE_BUFFER_SIZE_128]    = 128,
 180         [GXIO_MPIPE_BUFFER_SIZE_256]    = 256,
 181         [GXIO_MPIPE_BUFFER_SIZE_512]    = 512,
 182         [GXIO_MPIPE_BUFFER_SIZE_1024]   = 1024,
 183         [GXIO_MPIPE_BUFFER_SIZE_1664]   = 1664,
 184         [GXIO_MPIPE_BUFFER_SIZE_4096]   = 4096,
 185         [GXIO_MPIPE_BUFFER_SIZE_10368]  = 10368,
 186         [GXIO_MPIPE_BUFFER_SIZE_16384]  = 16384,
 187 };
 188
 189 static gxio_mpipe_context_t *
 190 mpipe_context(int instance)
 191 {
 192         if (instance < 0 || instance >= mpipe_instances)
 193                 return NULL;
 194         return &mpipe_contexts[instance].context;
 195 }
 196
 197 static int mpipe_channel_config(int instance, int channel,
 198                                 struct mpipe_channel_config *config)
 199 {
 200         struct mpipe_channel_config *data;
 201         struct mpipe_context *context;
 202         gxio_mpipe_rules_t rules;
 203         int idx, rc = 0;
 204
 205         if (instance < 0 || instance >= mpipe_instances ||
 206             channel < 0 || channel >= MPIPE_MAX_CHANNELS)
 207                 return -EINVAL;
 208
 209         context = &mpipe_contexts[instance];
 210
 211         rte_spinlock_lock(&context->lock);
 212
 213         gxio_mpipe_rules_init(&rules, &context->context);
 214
 215         for (idx = 0; idx < MPIPE_MAX_CHANNELS; idx++) {
 216                 data = (channel == idx) ? config : &context->channels[idx];
 217
 218                 if (!data->enable)
 219                         continue;
 220
 221                 rc = gxio_mpipe_rules_begin(&rules, data->first_bucket,
 222                                             data->num_buckets, &data->stacks);
 223                 if (rc < 0) {
 224                         goto done;
 225                 }
 226
 227                 rc = gxio_mpipe_rules_add_channel(&rules, idx);
 228                 if (rc < 0) {
 229                         goto done;
 230                 }
 231
 232                 rc = gxio_mpipe_rules_set_headroom(&rules, data->head_room);
 233                 if (rc < 0) {
 234                         goto done;
 235                 }
 236         }
 237
 238         rc = gxio_mpipe_rules_commit(&rules);
 239         if (rc == 0) {
 240                 memcpy(&context->channels[channel], config, sizeof(*config));
 241         }
 242
 243 done:
 244         rte_spinlock_unlock(&context->lock);
 245
 246         return rc;
 247 }
 248
 249 static int
 250 mpipe_get_size_index(int *array, int count, int size,
 251                      bool roundup)
 252 {
 253         int i, last = -1;
 254
 255         for (i = 0; i < count && array[i] < size; i++) {
 256                 if (array[i])
 257                         last = i;
 258         }
 259
 260         if (roundup)
 261                 return i < count ? (int)i : -ENOENT;
 262         else
 263                 return last >= 0 ? last : -ENOENT;
 264 }
 265
 266 static int
 267 mpipe_calc_size(int *array, int count, int size)
 268 {
 269         int index = mpipe_get_size_index(array, count, size, 1);
 270         return index < 0 ? index : array[index];
 271 }
 272
 273 static int mpipe_equeue_size(int size)
 274 {
 275         int result;
 276         result = mpipe_calc_size(mpipe_equeue_sizes,
 277                                  RTE_DIM(mpipe_equeue_sizes), size);
 278         return result;
 279 }
 280
 281 static int mpipe_iqueue_size(int size)
 282 {
 283         int result;
 284         result = mpipe_calc_size(mpipe_iqueue_sizes,
 285                                  RTE_DIM(mpipe_iqueue_sizes), size);
 286         return result;
 287 }
 288
 289 static int mpipe_buffer_size_index(int size)
 290 {
 291         int result;
 292         result = mpipe_get_size_index(mpipe_buffer_sizes,
 293                                       RTE_DIM(mpipe_buffer_sizes), size, 0);
 294         return result;
 295 }
 296
 297 static inline int
 298 mpipe_dev_atomic_read_link_status(struct rte_eth_dev *dev,
 299                                   struct rte_eth_link *link)
 300 {
 301         struct rte_eth_link *dst = link;
 302         struct rte_eth_link *src = &(dev->data->dev_link);
 303
 304         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 305                                 *(uint64_t *)src) == 0)
 306                 return -1;
 307
 308         return 0;
 309 }
 310
 311 static inline int
 312 mpipe_dev_atomic_write_link_status(struct rte_eth_dev *dev,
 313                                    struct rte_eth_link *link)
 314 {
 315         struct rte_eth_link *dst = &(dev->data->dev_link);
 316         struct rte_eth_link *src = link;
 317
 318         if (rte_atomic64_cmpset((uint64_t *)dst, *(uint64_t *)dst,
 319                                 *(uint64_t *)src) == 0)
 320                 return -1;
 321
 322         return 0;
 323 }
 324
 325 static void
 326 mpipe_infos_get(struct rte_eth_dev *dev __rte_unused,
 327                 struct rte_eth_dev_info *dev_info)
 328 {
 329         dev_info->min_rx_bufsize  = 128;
 330         dev_info->max_rx_pktlen   = 1518;
 331         dev_info->max_tx_queues   = MPIPE_TX_MAX_QUEUES;
 332         dev_info->max_rx_queues   = MPIPE_RX_MAX_QUEUES;
 333         dev_info->max_mac_addrs   = 1;
 334         dev_info->rx_offload_capa = 0;
 335         dev_info->tx_offload_capa = 0;
 336 }
 337
 338 static int
 339 mpipe_configure(struct rte_eth_dev *dev)
 340 {
 341         struct mpipe_dev_priv *priv = mpipe_priv(dev);
 342
 343         if (dev->data->nb_tx_queues > MPIPE_TX_MAX_QUEUES) {
 344                 RTE_LOG(ERR, PMD, "%s: Too many tx queues: %d > %d\n",
 345                         mpipe_name(priv), dev->data->nb_tx_queues,
 346                         MPIPE_TX_MAX_QUEUES);
 347                 return -EINVAL;
 348         }
 349         priv->nb_tx_queues = dev->data->nb_tx_queues;
 350
 351         if (dev->data->nb_rx_queues > MPIPE_RX_MAX_QUEUES) {
 352                 RTE_LOG(ERR, PMD, "%s: Too many rx queues: %d > %d\n",
 353                         mpipe_name(priv), dev->data->nb_rx_queues,
 354                         MPIPE_RX_MAX_QUEUES);
 355         }
 356         priv->nb_rx_queues = dev->data->nb_rx_queues;
 357
 358         return 0;
 359 }
 360
 361 static inline int
 362 mpipe_link_compare(struct rte_eth_link *link1,
 363                    struct rte_eth_link *link2)
 364 {
 365         return ((*(uint64_t *)link1 == *(uint64_t *)link2)
 366                 ? -1 : 0);
 367 }
 368
 369 static int
 370 mpipe_link_update(struct rte_eth_dev *dev, int wait_to_complete)
 371 {
 372         struct mpipe_dev_priv *priv = mpipe_priv(dev);
 373         struct rte_eth_link old, new;
 374         int64_t state, speed;
 375         int count, rc;
 376
 377         memset(&old, 0, sizeof(old));
 378         memset(&new, 0, sizeof(new));
 379         mpipe_dev_atomic_read_link_status(dev, &old);
 380
 381         for (count = 0, rc = 0; count < MPIPE_LINK_UPDATE_TIMEOUT; count++) {
 382                 if (!priv->initialized)
 383                         break;
 384
 385                 state = gxio_mpipe_link_get_attr(&priv->link,
 386                                                  GXIO_MPIPE_LINK_CURRENT_STATE);
 387                 if (state < 0)
 388                         break;
 389
 390                 speed = state & GXIO_MPIPE_LINK_SPEED_MASK;
 391
 392                 if (speed == GXIO_MPIPE_LINK_1G) {
 393                         new.link_speed = ETH_LINK_SPEED_1000;
 394                         new.link_duplex = ETH_LINK_FULL_DUPLEX;
 395                         new.link_status = 1;
 396                 } else if (speed == GXIO_MPIPE_LINK_10G) {
 397                         new.link_speed = ETH_LINK_SPEED_10000;
 398                         new.link_duplex = ETH_LINK_FULL_DUPLEX;
 399                         new.link_status = 1;
 400                 }
 401
 402                 rc = mpipe_link_compare(&old, &new);
 403                 if (rc == 0 || !wait_to_complete)
 404                         break;
 405
 406                 rte_delay_us(MPIPE_LINK_UPDATE_INTERVAL);
 407         }
 408
 409         mpipe_dev_atomic_write_link_status(dev, &new);
 410         return rc;
 411 }
 412
 413 static int
 414 mpipe_set_link(struct rte_eth_dev *dev, int up)
 415 {
 416         struct mpipe_dev_priv *priv = mpipe_priv(dev);
 417         int rc;
 418
 419         rc = gxio_mpipe_link_set_attr(&priv->link,
 420                                       GXIO_MPIPE_LINK_DESIRED_STATE,
 421                                       up ? GXIO_MPIPE_LINK_ANYSPEED : 0);
 422         if (rc < 0) {
 423                 RTE_LOG(ERR, PMD, "%s: Failed to set link %s.\n",
 424                         mpipe_name(priv), up ? "up" : "down");
 425         } else {
 426                 mpipe_link_update(dev, 0);
 427         }
 428
 429         return rc;
 430 }
 431
 432 static int
 433 mpipe_set_link_up(struct rte_eth_dev *dev)
 434 {
 435         return mpipe_set_link(dev, 1);
 436 }
 437
 438 static int
 439 mpipe_set_link_down(struct rte_eth_dev *dev)
 440 {
 441         return mpipe_set_link(dev, 0);
 442 }
 443
 444 static inline void
 445 mpipe_dp_enter(struct mpipe_dev_priv *priv)
 446 {
 447         __insn_mtspr(SPR_DSTREAM_PF, 0);
 448         rte_atomic32_inc(&priv->dp_count);
 449 }
 450
 451 static inline void
 452 mpipe_dp_exit(struct mpipe_dev_priv *priv)
 453 {
 454         rte_atomic32_dec(&priv->dp_count);
 455 }
 456
 457 static inline void
 458 mpipe_dp_wait(struct mpipe_dev_priv *priv)
 459 {
 460         while (rte_atomic32_read(&priv->dp_count) != 0) {
 461                 rte_pause();
 462         }
 463 }
 464
 465 static inline struct rte_mbuf *
 466 mpipe_recv_mbuf(struct mpipe_dev_priv *priv, gxio_mpipe_idesc_t *idesc,
 467                 int in_port)
 468 {
 469         void *va = gxio_mpipe_idesc_get_va(idesc);
 470         uint16_t size = gxio_mpipe_idesc_get_xfer_size(idesc);
 471         struct rte_mbuf *mbuf = RTE_PTR_SUB(va, priv->rx_offset);
 472
 473         rte_pktmbuf_reset(mbuf);
 474         mbuf->data_off = (uintptr_t)va - (uintptr_t)mbuf->buf_addr;
 475         mbuf->port     = in_port;
 476         mbuf->data_len = size;
 477         mbuf->pkt_len  = size;
 478         mbuf->hash.rss = gxio_mpipe_idesc_get_flow_hash(idesc);
 479
 480         PMD_DEBUG_RX("%s: RX mbuf %p, buffer %p, buf_addr %p, size %d\n",
 481                      mpipe_name(priv), mbuf, va, mbuf->buf_addr, size);
 482
 483         return mbuf;
 484 }
 485
 486 static inline void
 487 mpipe_recv_push(struct mpipe_dev_priv *priv, struct rte_mbuf *mbuf)
 488 {
 489         const int offset = RTE_PKTMBUF_HEADROOM + MPIPE_RX_IP_ALIGN;
 490         void *buf_addr = RTE_PTR_ADD(mbuf->buf_addr, offset);
 491
 492         gxio_mpipe_push_buffer(priv->context, priv->stack, buf_addr);
 493         PMD_DEBUG_RX("%s: Pushed mbuf %p, buffer %p into stack %d\n",
 494                      mpipe_name(priv), mbuf, buf_addr, priv->stack);
 495 }
 496
 497 static inline void
 498 mpipe_recv_fill_stack(struct mpipe_dev_priv *priv, int count)
 499 {
 500         struct rte_mbuf *mbuf;
 501         int i;
 502
 503         for (i = 0; i < count; i++) {
 504                 mbuf = __rte_mbuf_raw_alloc(priv->rx_mpool);
 505                 if (!mbuf)
 506                         break;
 507                 mpipe_recv_push(priv, mbuf);
 508         }
 509
 510         priv->rx_buffers += count;
 511         PMD_DEBUG_RX("%s: Filled %d/%d buffers\n", mpipe_name(priv), i, count);
 512 }
 513
 514 static inline void
 515 mpipe_recv_flush_stack(struct mpipe_dev_priv *priv)
 516 {
 517         const int offset = priv->rx_offset & ~RTE_MEMPOOL_ALIGN_MASK;
 518         uint8_t in_port = priv->port_id;
 519         struct rte_mbuf *mbuf;
 520         unsigned count;
 521         void *va;
 522
 523         for (count = 0; count < priv->rx_buffers; count++) {
 524                 va = gxio_mpipe_pop_buffer(priv->context, priv->stack);
 525                 if (!va)
 526                         break;
 527                 mbuf = RTE_PTR_SUB(va, offset);
 528
 529                 PMD_DEBUG_RX("%s: Flushing mbuf %p, va %p\n",
 530                              mpipe_name(priv), mbuf, va);
 531
 532                 mbuf->data_off    = (uintptr_t)va - (uintptr_t)mbuf->buf_addr;
 533                 mbuf->refcnt      = 1;
 534                 mbuf->nb_segs     = 1;
 535                 mbuf->port        = in_port;
 536                 mbuf->packet_type = 0;
 537                 mbuf->data_len    = 0;
 538                 mbuf->pkt_len     = 0;
 539
 540                 __rte_mbuf_raw_free(mbuf);
 541         }
 542
 543         PMD_DEBUG_RX("%s: Returned %d/%d buffers\n",
 544                      mpipe_name(priv), count, priv->rx_buffers);
 545         priv->rx_buffers -= count;
 546 }
 547
 548 static void
 549 mpipe_register_segment(struct mpipe_dev_priv *priv, const struct rte_memseg *ms)
 550 {
 551         size_t size = ms->hugepage_sz;
 552         uint8_t *addr, *end;
 553         int rc;
 554
 555         for (addr = ms->addr, end = addr + ms->len; addr < end; addr += size) {
 556                 rc = gxio_mpipe_register_page(priv->context, priv->stack, addr,
 557                                               size, 0);
 558                 if (rc < 0)
 559                         break;
 560         }
 561
 562         if (rc < 0) {
 563                 RTE_LOG(ERR, PMD, "%s: Could not register memseg @%p, %d.\n",
 564                         mpipe_name(priv), ms->addr, rc);
 565         } else {
 566                 RTE_LOG(DEBUG, PMD, "%s: Registered segment %p - %p\n",
 567                         mpipe_name(priv), ms->addr,
 568                         RTE_PTR_ADD(ms->addr, ms->len - 1));
 569         }
 570 }
 571
 572 static int
 573 mpipe_recv_init(struct mpipe_dev_priv *priv)
 574 {
 575         const struct rte_memseg *seg = rte_eal_get_physmem_layout();
 576         size_t stack_size;
 577         void *stack_mem;
 578         int rc;
 579
 580         if (!priv->rx_mpool) {
 581                 RTE_LOG(ERR, PMD, "%s: No buffer pool.\n",
 582                         mpipe_name(priv));
 583                 return -ENODEV;
 584         }
 585
 586         /* Allocate one NotifRing for each queue. */
 587         rc = gxio_mpipe_alloc_notif_rings(priv->context, MPIPE_RX_MAX_QUEUES,
 588                                           0, 0);
 589         if (rc < 0) {
 590                 RTE_LOG(ERR, PMD, "%s: Failed to allocate notif rings.\n",
 591                         mpipe_name(priv));
 592                 return rc;
 593         }
 594         priv->first_ring = rc;
 595
 596         /* Allocate a NotifGroup. */
 597         rc = gxio_mpipe_alloc_notif_groups(priv->context, 1, 0, 0);
 598         if (rc < 0) {
 599                 RTE_LOG(ERR, PMD, "%s: Failed to allocate rx group.\n",
 600                         mpipe_name(priv));
 601                 return rc;
 602         }
 603         priv->notif_group = rc;
 604
 605         /* Allocate required buckets. */
 606         rc = gxio_mpipe_alloc_buckets(priv->context, MPIPE_RX_BUCKETS, 0, 0);
 607         if (rc < 0) {
 608                 RTE_LOG(ERR, PMD, "%s: Failed to allocate buckets.\n",
 609                         mpipe_name(priv));
 610                 return rc;
 611         }
 612         priv->first_bucket = rc;
 613
 614         rc = gxio_mpipe_alloc_buffer_stacks(priv->context, 1, 0, 0);
 615         if (rc < 0) {
 616                 RTE_LOG(ERR, PMD, "%s: Failed to allocate buffer stack.\n",
 617                         mpipe_name(priv));
 618                 return rc;
 619         }
 620         priv->stack = rc;
 621
 622         while (seg && seg->addr)
 623                 mpipe_register_segment(priv, seg++);
 624
 625         stack_size = gxio_mpipe_calc_buffer_stack_bytes(MPIPE_RX_STACK_SIZE);
 626         stack_mem = rte_zmalloc(NULL, stack_size, 65536);
 627         if (!stack_mem) {
 628                 RTE_LOG(ERR, PMD, "%s: Failed to allocate buffer memory.\n",
 629                         mpipe_name(priv));
 630                 return -ENOMEM;
 631         } else {
 632                 RTE_LOG(DEBUG, PMD, "%s: Buffer stack memory %p - %p.\n",
 633                         mpipe_name(priv), stack_mem,
 634                         RTE_PTR_ADD(stack_mem, stack_size - 1));
 635         }
 636
 637         rc = gxio_mpipe_init_buffer_stack(priv->context, priv->stack,
 638                                           priv->rx_size_code, stack_mem,
 639                                           stack_size, 0);
 640         if (rc < 0) {
 641                 RTE_LOG(ERR, PMD, "%s: Failed to initialize buffer stack.\n",
 642                         mpipe_name(priv));
 643                 return rc;
 644         }
 645
 646         return 0;
 647 }
 648
 649 static int
 650 mpipe_xmit_init(struct mpipe_dev_priv *priv)
 651 {
 652         size_t ring_size;
 653         void *ring_mem;
 654         int rc;
 655
 656         /* Allocate eDMA ring. */
 657         rc = gxio_mpipe_alloc_edma_rings(priv->context, 1, 0, 0);
 658         if (rc < 0) {
 659                 RTE_LOG(ERR, PMD, "%s: Failed to alloc tx ring.\n",
 660                         mpipe_name(priv));
 661                 return rc;
 662         }
 663         priv->ering = rc;
 664
 665         rc = mpipe_equeue_size(MPIPE_TX_DESCS);
 666         if (rc < 0) {
 667                 RTE_LOG(ERR, PMD, "%s: Cannot allocate %d equeue descs.\n",
 668                         mpipe_name(priv), (int)MPIPE_TX_DESCS);
 669                 return -ENOMEM;
 670         }
 671         priv->equeue_size = rc;
 672
 673         /* Initialize completion array. */
 674         ring_size = sizeof(priv->tx_comps[0]) * priv->equeue_size;
 675         priv->tx_comps = rte_zmalloc(NULL, ring_size, RTE_CACHE_LINE_SIZE);
 676         if (!priv->tx_comps) {
 677                 RTE_LOG(ERR, PMD, "%s: Failed to allocate egress comps.\n",
 678                         mpipe_name(priv));
 679                 return -ENOMEM;
 680         }
 681
 682         /* Allocate eDMA ring memory. */
 683         ring_size = sizeof(gxio_mpipe_edesc_t) * priv->equeue_size;
 684         ring_mem = rte_zmalloc(NULL, ring_size, ring_size);
 685         if (!ring_mem) {
 686                 RTE_LOG(ERR, PMD, "%s: Failed to allocate egress descs.\n",
 687                         mpipe_name(priv));
 688                 return -ENOMEM;
 689         } else {
 690                 RTE_LOG(DEBUG, PMD, "%s: eDMA ring memory %p - %p.\n",
 691                         mpipe_name(priv), ring_mem,
 692                         RTE_PTR_ADD(ring_mem, ring_size - 1));
 693         }
 694
 695         /* Initialize eDMA ring. */
 696         rc = gxio_mpipe_equeue_init(&priv->equeue, priv->context, priv->ering,
 697                                     priv->channel, ring_mem, ring_size, 0);
 698         if (rc < 0) {
 699                 RTE_LOG(ERR, PMD, "%s: Failed to init equeue\n",
 700                         mpipe_name(priv));
 701                 return rc;
 702         }
 703
 704         return 0;
 705 }
 706
 707 static int
 708 mpipe_link_init(struct mpipe_dev_priv *priv)
 709 {
 710         int rc;
 711
 712         /* Open the link. */
 713         rc = gxio_mpipe_link_open(&priv->link, priv->context,
 714                                   mpipe_name(priv), GXIO_MPIPE_LINK_AUTO_NONE);
 715         if (rc < 0) {
 716                 RTE_LOG(ERR, PMD, "%s: Failed to open link.\n",
 717                         mpipe_name(priv));
 718                 return rc;
 719         }
 720
 721         /* Get the channel index. */
 722         rc = gxio_mpipe_link_channel(&priv->link);
 723         if (rc < 0) {
 724                 RTE_LOG(ERR, PMD, "%s: Bad channel\n",
 725                         mpipe_name(priv));
 726                 return rc;
 727         }
 728         priv->channel = rc;
 729
 730         return 0;
 731 }
 732
 733 static int
 734 mpipe_init(struct mpipe_dev_priv *priv)
 735 {
 736         int rc;
 737
 738         if (priv->initialized)
 739                 return 0;
 740
 741         rc = mpipe_link_init(priv);
 742         if (rc < 0) {
 743                 RTE_LOG(ERR, PMD, "%s: Failed to init link.\n",
 744                         mpipe_name(priv));
 745                 return rc;
 746         }
 747
 748         rc = mpipe_recv_init(priv);
 749         if (rc < 0) {
 750                 RTE_LOG(ERR, PMD, "%s: Failed to init rx.\n",
 751                         mpipe_name(priv));
 752                 return rc;
 753         }
 754
 755         rc = mpipe_xmit_init(priv);
 756         if (rc < 0) {
 757                 RTE_LOG(ERR, PMD, "%s: Failed to init tx.\n",
 758                         mpipe_name(priv));
 759                 rte_free(priv);
 760                 return rc;
 761         }
 762
 763         priv->initialized = 1;
 764
 765         return 0;
 766 }
 767
 768 static int
 769 mpipe_start(struct rte_eth_dev *dev)
 770 {
 771         struct mpipe_dev_priv *priv = mpipe_priv(dev);
 772         struct mpipe_channel_config config;
 773         struct mpipe_rx_queue *rx_queue;
 774         struct rte_eth_link eth_link;
 775         unsigned queue, buffers = 0;
 776         size_t ring_size;
 777         void *ring_mem;
 778         int rc;
 779
 780         memset(&eth_link, 0, sizeof(eth_link));
 781         mpipe_dev_atomic_write_link_status(dev, &eth_link);
 782
 783         rc = mpipe_init(priv);
 784         if (rc < 0)
 785                 return rc;
 786
 787         /* Initialize NotifRings. */
 788         for (queue = 0; queue < priv->nb_rx_queues; queue++) {
 789                 rx_queue = mpipe_rx_queue(priv, queue);
 790                 ring_size = rx_queue->q.nb_desc * sizeof(gxio_mpipe_idesc_t);
 791
 792                 ring_mem = rte_malloc(NULL, ring_size, ring_size);
 793                 if (!ring_mem) {
 794                         RTE_LOG(ERR, PMD, "%s: Failed to alloc rx descs.\n",
 795                                 mpipe_name(priv));
 796                         return -ENOMEM;
 797                 } else {
 798                         RTE_LOG(DEBUG, PMD, "%s: iDMA ring %d memory %p - %p.\n",
 799                                 mpipe_name(priv), queue, ring_mem,
 800                                 RTE_PTR_ADD(ring_mem, ring_size - 1));
 801                 }
 802
 803                 rc = gxio_mpipe_iqueue_init(&rx_queue->iqueue, priv->context,
 804                                             priv->first_ring + queue, ring_mem,
 805                                             ring_size, 0);
 806                 if (rc < 0) {
 807                         RTE_LOG(ERR, PMD, "%s: Failed to init rx queue.\n",
 808                                 mpipe_name(priv));
 809                         return rc;
 810                 }
 811
 812                 rx_queue->rx_ring_mem = ring_mem;
 813                 buffers += rx_queue->q.nb_desc;
 814         }
 815
 816         /* Initialize ingress NotifGroup and buckets. */
 817         rc = gxio_mpipe_init_notif_group_and_buckets(priv->context,
 818                         priv->notif_group, priv->first_ring, priv->nb_rx_queues,
 819                         priv->first_bucket, MPIPE_RX_BUCKETS,
 820                         GXIO_MPIPE_BUCKET_STATIC_FLOW_AFFINITY);
 821         if (rc < 0) {
 822                 RTE_LOG(ERR, PMD, "%s: Failed to init group and buckets.\n",
 823                         mpipe_name(priv));
 824                 return rc;
 825         }
 826
 827         /* Configure the classifier to deliver packets from this port. */
 828         config.enable = 1;
 829         config.first_bucket = priv->first_bucket;
 830         config.num_buckets = MPIPE_RX_BUCKETS;
 831         memset(&config.stacks, 0xff, sizeof(config.stacks));
 832         config.stacks.stacks[priv->rx_size_code] = priv->stack;
 833         config.head_room = priv->rx_offset & RTE_MEMPOOL_ALIGN_MASK;
 834
 835         rc = mpipe_channel_config(priv->instance, priv->channel,
 836                                   &config);
 837         if (rc < 0) {
 838                 RTE_LOG(ERR, PMD, "%s: Failed to setup classifier.\n",
 839                         mpipe_name(priv));
 840                 return rc;
 841         }
 842
 843         /* Fill empty buffers into the buffer stack. */
 844         mpipe_recv_fill_stack(priv, buffers);
 845
 846         /* Bring up the link. */
 847         mpipe_set_link_up(dev);
 848
 849         /* Start xmit/recv on queues. */
 850         for (queue = 0; queue < priv->nb_tx_queues; queue++)
 851                 mpipe_tx_queue(priv, queue)->q.link_status = 1;
 852         for (queue = 0; queue < priv->nb_rx_queues; queue++)
 853                 mpipe_rx_queue(priv, queue)->q.link_status = 1;
 854         priv->running = 1;
 855
 856         return 0;
 857 }
 858
 859 static void
 860 mpipe_stop(struct rte_eth_dev *dev)
 861 {
 862         struct mpipe_dev_priv *priv = mpipe_priv(dev);
 863         struct mpipe_channel_config config;
 864         unsigned queue;
 865         int rc;
 866
 867         for (queue = 0; queue < priv->nb_tx_queues; queue++)
 868                 mpipe_tx_queue(priv, queue)->q.link_status = 0;
 869         for (queue = 0; queue < priv->nb_rx_queues; queue++)
 870                 mpipe_rx_queue(priv, queue)->q.link_status = 0;
 871
 872         /* Make sure the link_status writes land. */
 873         rte_wmb();
 874
 875         /*
 876          * Wait for link_status change to register with straggling datapath
 877          * threads.
 878          */
 879         mpipe_dp_wait(priv);
 880
 881         /* Bring down the link. */
 882         mpipe_set_link_down(dev);
 883
 884         /* Remove classifier rules. */
 885         memset(&config, 0, sizeof(config));
 886         rc = mpipe_channel_config(priv->instance, priv->channel,
 887                                   &config);
 888         if (rc < 0) {
 889                 RTE_LOG(ERR, PMD, "%s: Failed to stop classifier.\n",
 890                         mpipe_name(priv));
 891         }
 892
 893         /* Flush completed xmit packets. */
 894         mpipe_xmit_flush(priv);
 895
 896         /* Flush buffer stacks. */
 897         mpipe_recv_flush(priv);
 898
 899         priv->running = 0;
 900 }
 901
 902 static void
 903 mpipe_close(struct rte_eth_dev *dev)
 904 {
 905         struct mpipe_dev_priv *priv = mpipe_priv(dev);
 906         if (priv->running)
 907                 mpipe_stop(dev);
 908 }
 909
 910 static void
 911 mpipe_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
 912 {
 913         struct mpipe_dev_priv *priv = mpipe_priv(dev);
 914         struct mpipe_tx_queue *tx_queue;
 915         struct mpipe_rx_queue *rx_queue;
 916         unsigned i;
 917         uint16_t idx;
 918
 919         memset(stats, 0, sizeof(*stats));
 920
 921         for (i = 0; i < priv->nb_tx_queues; i++) {
 922                 tx_queue = mpipe_tx_queue(priv, i);
 923
 924                 stats->opackets += tx_queue->q.stats.packets;
 925                 stats->obytes   += tx_queue->q.stats.bytes;
 926                 stats->oerrors  += tx_queue->q.stats.errors;
 927
 928                 idx = tx_queue->q.stat_idx;
 929                 if (idx != (uint16_t)-1) {
 930                         stats->q_opackets[idx] += tx_queue->q.stats.packets;
 931                         stats->q_obytes[idx]   += tx_queue->q.stats.bytes;
 932                         stats->q_errors[idx]   += tx_queue->q.stats.errors;
 933                 }
 934         }
 935
 936         for (i = 0; i < priv->nb_rx_queues; i++) {
 937                 rx_queue = mpipe_rx_queue(priv, i);
 938
 939                 stats->ipackets  += rx_queue->q.stats.packets;
 940                 stats->ibytes    += rx_queue->q.stats.bytes;
 941                 stats->ierrors   += rx_queue->q.stats.errors;
 942                 stats->rx_nombuf += rx_queue->q.stats.nomem;
 943
 944                 idx = rx_queue->q.stat_idx;
 945                 if (idx != (uint16_t)-1) {
 946                         stats->q_ipackets[idx] += rx_queue->q.stats.packets;
 947                         stats->q_ibytes[idx]   += rx_queue->q.stats.bytes;
 948                         stats->q_errors[idx]   += rx_queue->q.stats.errors;
 949                 }
 950         }
 951 }
 952
 953 static void
 954 mpipe_stats_reset(struct rte_eth_dev *dev)
 955 {
 956         struct mpipe_dev_priv *priv = mpipe_priv(dev);
 957         struct mpipe_tx_queue *tx_queue;
 958         struct mpipe_rx_queue *rx_queue;
 959         unsigned i;
 960
 961         for (i = 0; i < priv->nb_tx_queues; i++) {
 962                 tx_queue = mpipe_tx_queue(priv, i);
 963                 memset(&tx_queue->q.stats, 0, sizeof(tx_queue->q.stats));
 964         }
 965
 966         for (i = 0; i < priv->nb_rx_queues; i++) {
 967                 rx_queue = mpipe_rx_queue(priv, i);
 968                 memset(&rx_queue->q.stats, 0, sizeof(rx_queue->q.stats));
 969         }
 970 }
 971
 972 static int
 973 mpipe_queue_stats_mapping_set(struct rte_eth_dev *dev, uint16_t queue_id,
 974                               uint8_t stat_idx, uint8_t is_rx)
 975 {
 976         struct mpipe_dev_priv *priv = mpipe_priv(dev);
 977
 978         if (is_rx) {
 979                 priv->rx_stat_mapping[stat_idx] = queue_id;
 980         } else {
 981                 priv->tx_stat_mapping[stat_idx] = queue_id;
 982         }
 983
 984         return 0;
 985 }
 986
 987 static int
 988 mpipe_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 989                      uint16_t nb_desc, unsigned int socket_id __rte_unused,
 990                      const struct rte_eth_txconf *tx_conf __rte_unused)
 991 {
 992         struct mpipe_tx_queue *tx_queue = dev->data->tx_queues[queue_idx];
 993         struct mpipe_dev_priv *priv = mpipe_priv(dev);
 994         uint16_t idx;
 995
 996         tx_queue = rte_realloc(tx_queue, sizeof(*tx_queue),
 997                                RTE_CACHE_LINE_SIZE);
 998         if (!tx_queue) {
 999                 RTE_LOG(ERR, PMD, "%s: Failed to allocate TX queue.\n",
1000                         mpipe_name(priv));
1001                 return -ENOMEM;
1002         }
1003
1004         memset(&tx_queue->q, 0, sizeof(tx_queue->q));
1005         tx_queue->q.priv = priv;
1006         tx_queue->q.queue_idx = queue_idx;
1007         tx_queue->q.port_id = dev->data->port_id;
1008         tx_queue->q.nb_desc = nb_desc;
1009
1010         tx_queue->q.stat_idx = -1;
1011         for (idx = 0; idx < RTE_ETHDEV_QUEUE_STAT_CNTRS; idx++) {
1012                 if (priv->tx_stat_mapping[idx] == queue_idx)
1013                         tx_queue->q.stat_idx = idx;
1014         }
1015
1016         dev->data->tx_queues[queue_idx] = tx_queue;
1017
1018         return 0;
1019 }
1020
1021 static void
1022 mpipe_tx_queue_release(void *_txq)
1023 {
1024         rte_free(_txq);
1025 }
1026
1027 static int
1028 mpipe_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
1029                      uint16_t nb_desc, unsigned int socket_id __rte_unused,
1030                      const struct rte_eth_rxconf *rx_conf __rte_unused,
1031                      struct rte_mempool *mp)
1032 {
1033         struct mpipe_rx_queue *rx_queue = dev->data->rx_queues[queue_idx];
1034         struct mpipe_dev_priv *priv = mpipe_priv(dev);
1035         uint16_t idx;
1036         int size, rc;
1037
1038         rc = mpipe_iqueue_size(nb_desc);
1039         if (rc < 0) {
1040                 RTE_LOG(ERR, PMD, "%s: Cannot allocate %d iqueue descs.\n",
1041                         mpipe_name(priv), (int)nb_desc);
1042                 return -ENOMEM;
1043         }
1044
1045         if (rc != nb_desc) {
1046                 RTE_LOG(WARNING, PMD, "%s: Extending RX descs from %d to %d.\n",
1047                         mpipe_name(priv), (int)nb_desc, rc);
1048                 nb_desc = rc;
1049         }
1050
1051         size = sizeof(*rx_queue);
1052         rx_queue = rte_realloc(rx_queue, size, RTE_CACHE_LINE_SIZE);
1053         if (!rx_queue) {
1054                 RTE_LOG(ERR, PMD, "%s: Failed to allocate RX queue.\n",
1055                         mpipe_name(priv));
1056                 return -ENOMEM;
1057         }
1058
1059         memset(&rx_queue->q, 0, sizeof(rx_queue->q));
1060         rx_queue->q.priv = priv;
1061         rx_queue->q.nb_desc = nb_desc;
1062         rx_queue->q.port_id = dev->data->port_id;
1063         rx_queue->q.queue_idx = queue_idx;
1064
1065         if (!priv->rx_mpool) {
1066                 int size = (rte_pktmbuf_data_room_size(mp) -
1067                             RTE_PKTMBUF_HEADROOM -
1068                             MPIPE_RX_IP_ALIGN);
1069
1070                 priv->rx_offset = (sizeof(struct rte_mbuf) +
1071                                    rte_pktmbuf_priv_size(mp) +
1072                                    RTE_PKTMBUF_HEADROOM +
1073                                    MPIPE_RX_IP_ALIGN);
1074                 if (size < 0) {
1075                         RTE_LOG(ERR, PMD, "%s: Bad buffer size %d.\n",
1076                                 mpipe_name(priv),
1077                                 rte_pktmbuf_data_room_size(mp));
1078                         return -ENOMEM;
1079                 }
1080
1081                 priv->rx_size_code = mpipe_buffer_size_index(size);
1082                 priv->rx_mpool = mp;
1083         }
1084
1085         if (priv->rx_mpool != mp) {
1086                 RTE_LOG(WARNING, PMD, "%s: Ignoring multiple buffer pools.\n",
1087                         mpipe_name(priv));
1088         }
1089
1090         rx_queue->q.stat_idx = -1;
1091         for (idx = 0; idx < RTE_ETHDEV_QUEUE_STAT_CNTRS; idx++) {
1092                 if (priv->rx_stat_mapping[idx] == queue_idx)
1093                         rx_queue->q.stat_idx = idx;
1094         }
1095
1096         dev->data->rx_queues[queue_idx] = rx_queue;
1097
1098         return 0;
1099 }
1100
1101 static void
1102 mpipe_rx_queue_release(void *_rxq)
1103 {
1104         rte_free(_rxq);
1105 }
1106
1107 #define MPIPE_XGBE_ENA_HASH_MULTI       \
1108         (1UL << MPIPE_XAUI_RECEIVE_CONFIGURATION__ENA_HASH_MULTI_SHIFT)
1109 #define MPIPE_XGBE_ENA_HASH_UNI         \
1110         (1UL << MPIPE_XAUI_RECEIVE_CONFIGURATION__ENA_HASH_UNI_SHIFT)
1111 #define MPIPE_XGBE_COPY_ALL             \
1112         (1UL << MPIPE_XAUI_RECEIVE_CONFIGURATION__COPY_ALL_SHIFT)
1113 #define MPIPE_GBE_ENA_MULTI_HASH        \
1114         (1UL << MPIPE_GBE_NETWORK_CONFIGURATION__MULTI_HASH_ENA_SHIFT)
1115 #define MPIPE_GBE_ENA_UNI_HASH          \
1116         (1UL << MPIPE_GBE_NETWORK_CONFIGURATION__UNI_HASH_ENA_SHIFT)
1117 #define MPIPE_GBE_COPY_ALL              \
1118         (1UL << MPIPE_GBE_NETWORK_CONFIGURATION__COPY_ALL_SHIFT)
1119
1120 static void
1121 mpipe_promiscuous_enable(struct rte_eth_dev *dev)
1122 {
1123         struct mpipe_dev_priv *priv = mpipe_priv(dev);
1124         int64_t reg;
1125         int addr;
1126
1127         if (priv->is_xaui) {
1128                 addr = MPIPE_XAUI_RECEIVE_CONFIGURATION;
1129                 reg  = gxio_mpipe_link_mac_rd(&priv->link, addr);
1130                 reg &= ~MPIPE_XGBE_ENA_HASH_MULTI;
1131                 reg &= ~MPIPE_XGBE_ENA_HASH_UNI;
1132                 reg |=  MPIPE_XGBE_COPY_ALL;
1133                 gxio_mpipe_link_mac_wr(&priv->link, addr, reg);
1134         } else {
1135                 addr = MPIPE_GBE_NETWORK_CONFIGURATION;
1136                 reg  = gxio_mpipe_link_mac_rd(&priv->link, addr);
1137                 reg &= ~MPIPE_GBE_ENA_MULTI_HASH;
1138                 reg &= ~MPIPE_GBE_ENA_UNI_HASH;
1139                 reg |=  MPIPE_GBE_COPY_ALL;
1140                 gxio_mpipe_link_mac_wr(&priv->link, addr, reg);
1141         }
1142 }
1143
1144 static void
1145 mpipe_promiscuous_disable(struct rte_eth_dev *dev)
1146 {
1147         struct mpipe_dev_priv *priv = mpipe_priv(dev);
1148         int64_t reg;
1149         int addr;
1150
1151         if (priv->is_xaui) {
1152                 addr = MPIPE_XAUI_RECEIVE_CONFIGURATION;
1153                 reg  = gxio_mpipe_link_mac_rd(&priv->link, addr);
1154                 reg |=  MPIPE_XGBE_ENA_HASH_MULTI;
1155                 reg |=  MPIPE_XGBE_ENA_HASH_UNI;
1156                 reg &= ~MPIPE_XGBE_COPY_ALL;
1157                 gxio_mpipe_link_mac_wr(&priv->link, addr, reg);
1158         } else {
1159                 addr = MPIPE_GBE_NETWORK_CONFIGURATION;
1160                 reg  = gxio_mpipe_link_mac_rd(&priv->link, addr);
1161                 reg |=  MPIPE_GBE_ENA_MULTI_HASH;
1162                 reg |=  MPIPE_GBE_ENA_UNI_HASH;
1163                 reg &= ~MPIPE_GBE_COPY_ALL;
1164                 gxio_mpipe_link_mac_wr(&priv->link, addr, reg);
1165         }
1166 }
1167
1168 static struct eth_dev_ops mpipe_dev_ops = {
1169         .dev_infos_get           = mpipe_infos_get,
1170         .dev_configure           = mpipe_configure,
1171         .dev_start               = mpipe_start,
1172         .dev_stop                = mpipe_stop,
1173         .dev_close               = mpipe_close,
1174         .stats_get               = mpipe_stats_get,
1175         .stats_reset             = mpipe_stats_reset,
1176         .queue_stats_mapping_set = mpipe_queue_stats_mapping_set,
1177         .tx_queue_setup          = mpipe_tx_queue_setup,
1178         .rx_queue_setup          = mpipe_rx_queue_setup,
1179         .tx_queue_release        = mpipe_tx_queue_release,
1180         .rx_queue_release        = mpipe_rx_queue_release,
1181         .link_update             = mpipe_link_update,
1182         .dev_set_link_up         = mpipe_set_link_up,
1183         .dev_set_link_down       = mpipe_set_link_down,
1184         .promiscuous_enable      = mpipe_promiscuous_enable,
1185         .promiscuous_disable     = mpipe_promiscuous_disable,
1186 };
1187
1188 static inline void
1189 mpipe_xmit_null(struct mpipe_dev_priv *priv, int64_t start, int64_t end)
1190 {
1191         gxio_mpipe_edesc_t null_desc = { { .bound = 1, .ns = 1 } };
1192         gxio_mpipe_equeue_t *equeue = &priv->equeue;
1193         int64_t slot;
1194
1195         for (slot = start; slot < end; slot++) {
1196                 gxio_mpipe_equeue_put_at(equeue, null_desc, slot);
1197         }
1198 }
1199
1200 static void
1201 mpipe_xmit_flush(struct mpipe_dev_priv *priv)
1202 {
1203         gxio_mpipe_equeue_t *equeue = &priv->equeue;
1204         int64_t slot;
1205
1206         /* Post a dummy descriptor and wait for its return. */
1207         slot = gxio_mpipe_equeue_reserve(equeue, 1);
1208         if (slot < 0) {
1209                 RTE_LOG(ERR, PMD, "%s: Failed to reserve stop slot.\n",
1210                         mpipe_name(priv));
1211                 return;
1212         }
1213
1214         mpipe_xmit_null(priv, slot, slot + 1);
1215
1216         while (!gxio_mpipe_equeue_is_complete(equeue, slot, 1)) {
1217                 rte_pause();
1218         }
1219
1220         for (slot = 0; slot < priv->equeue_size; slot++) {
1221                 if (priv->tx_comps[slot])
1222                         rte_pktmbuf_free_seg(priv->tx_comps[slot]);
1223         }
1224 }
1225
1226 static void
1227 mpipe_recv_flush(struct mpipe_dev_priv *priv)
1228 {
1229         uint8_t in_port = priv->port_id;
1230         struct mpipe_rx_queue *rx_queue;
1231         gxio_mpipe_iqueue_t *iqueue;
1232         gxio_mpipe_idesc_t idesc;
1233         struct rte_mbuf *mbuf;
1234         int retries = 0;
1235         unsigned queue;
1236
1237         do {
1238                 mpipe_recv_flush_stack(priv);
1239
1240                 /* Flush packets sitting in recv queues. */
1241                 for (queue = 0; queue < priv->nb_rx_queues; queue++) {
1242                         rx_queue = mpipe_rx_queue(priv, queue);
1243                         iqueue = &rx_queue->iqueue;
1244                         while (gxio_mpipe_iqueue_try_get(iqueue, &idesc) >= 0) {
1245                                 mbuf = mpipe_recv_mbuf(priv, &idesc, in_port);
1246                                 rte_pktmbuf_free(mbuf);
1247                                 priv->rx_buffers--;
1248                         }
1249                         rte_free(rx_queue->rx_ring_mem);
1250                 }
1251         } while (retries++ < 10 && priv->rx_buffers);
1252
1253         if (priv->rx_buffers) {
1254                 RTE_LOG(ERR, PMD, "%s: Leaked %d receive buffers.\n",
1255                         mpipe_name(priv), priv->rx_buffers);
1256         } else {
1257                 PMD_DEBUG_RX("%s: Returned all receive buffers.\n",
1258                              mpipe_name(priv));
1259         }
1260 }
1261
1262 static inline uint16_t
1263 mpipe_do_xmit(struct mpipe_tx_queue *tx_queue, struct rte_mbuf **tx_pkts,
1264               uint16_t nb_pkts)
1265 {
1266         struct mpipe_dev_priv *priv = tx_queue->q.priv;
1267         gxio_mpipe_equeue_t *equeue = &priv->equeue;
1268         unsigned nb_bytes = 0;
1269         unsigned nb_sent = 0;
1270         int nb_slots, i;
1271
1272         PMD_DEBUG_TX("Trying to transmit %d packets on %s:%d.\n",
1273                      nb_pkts, mpipe_name(tx_queue->q.priv),
1274                      tx_queue->q.queue_idx);
1275
1276         /* Optimistic assumption that we need exactly one slot per packet. */
1277         nb_slots = RTE_MIN(nb_pkts, MPIPE_TX_DESCS / 2);
1278
1279         do {
1280                 struct rte_mbuf *mbuf = NULL, *pkt = NULL;
1281                 int64_t slot;
1282
1283                 /* Reserve eDMA ring slots. */
1284                 slot = gxio_mpipe_equeue_try_reserve_fast(equeue, nb_slots);
1285                 if (unlikely(slot < 0)) {
1286                         break;
1287                 }
1288
1289                 for (i = 0; i < nb_slots; i++) {
1290                         unsigned idx = (slot + i) & (priv->equeue_size - 1);
1291                         rte_prefetch0(priv->tx_comps[idx]);
1292                 }
1293
1294                 /* Fill up slots with descriptor and completion info. */
1295                 for (i = 0; i < nb_slots; i++) {
1296                         unsigned idx = (slot + i) & (priv->equeue_size - 1);
1297                         gxio_mpipe_edesc_t desc;
1298                         struct rte_mbuf *next;
1299
1300                         /* Starting on a new packet? */
1301                         if (likely(!mbuf)) {
1302                                 int room = nb_slots - i;
1303
1304                                 pkt = mbuf = tx_pkts[nb_sent];
1305
1306                                 /* Bail out if we run out of descs. */
1307                                 if (unlikely(pkt->nb_segs > room))
1308                                         break;
1309
1310                                 nb_sent++;
1311                         }
1312
1313                         /* We have a segment to send. */
1314                         next = mbuf->next;
1315
1316                         if (priv->tx_comps[idx])
1317                                 rte_pktmbuf_free_seg(priv->tx_comps[idx]);
1318
1319                         desc = (gxio_mpipe_edesc_t) { {
1320                                 .va        = rte_pktmbuf_mtod(mbuf, uintptr_t),
1321                                 .xfer_size = rte_pktmbuf_data_len(mbuf),
1322                                 .bound     = next ? 0 : 1,
1323                         } };
1324
1325                         nb_bytes += mbuf->data_len;
1326                         priv->tx_comps[idx] = mbuf;
1327                         gxio_mpipe_equeue_put_at(equeue, desc, slot + i);
1328
1329                         PMD_DEBUG_TX("%s:%d: Sending packet %p, len %d\n",
1330                                      mpipe_name(priv),
1331                                      tx_queue->q.queue_idx,
1332                                      rte_pktmbuf_mtod(mbuf, void *),
1333                                      rte_pktmbuf_data_len(mbuf));
1334
1335                         mbuf = next;
1336                 }
1337
1338                 if (unlikely(nb_sent < nb_pkts)) {
1339
1340                         /* Fill remaining slots with null descriptors. */
1341                         mpipe_xmit_null(priv, slot + i, slot + nb_slots);
1342
1343                         /*
1344                          * Calculate exact number of descriptors needed for
1345                          * the next go around.
1346                          */
1347                         nb_slots = 0;
1348                         for (i = nb_sent; i < nb_pkts; i++) {
1349                                 nb_slots += tx_pkts[i]->nb_segs;
1350                         }
1351
1352                         nb_slots = RTE_MIN(nb_slots, MPIPE_TX_DESCS / 2);
1353                 }
1354         } while (nb_sent < nb_pkts);
1355
1356         tx_queue->q.stats.packets += nb_sent;
1357         tx_queue->q.stats.bytes   += nb_bytes;
1358
1359         return nb_sent;
1360 }
1361
1362 static inline uint16_t
1363 mpipe_do_recv(struct mpipe_rx_queue *rx_queue, struct rte_mbuf **rx_pkts,
1364               uint16_t nb_pkts)
1365 {
1366         struct mpipe_dev_priv *priv = rx_queue->q.priv;
1367         gxio_mpipe_iqueue_t *iqueue = &rx_queue->iqueue;
1368         gxio_mpipe_idesc_t *first_idesc, *idesc, *last_idesc;
1369         uint8_t in_port = rx_queue->q.port_id;
1370         const unsigned look_ahead = 8;
1371         int room = nb_pkts, rc = 0;
1372         unsigned nb_packets = 0;
1373         unsigned nb_dropped = 0;
1374         unsigned nb_nomem = 0;
1375         unsigned nb_bytes = 0;
1376         unsigned nb_descs, i;
1377
1378         while (room && !rc) {
1379                 if (rx_queue->avail_descs < room) {
1380                         rc = gxio_mpipe_iqueue_try_peek(iqueue,
1381                                                         &rx_queue->next_desc);
1382                         rx_queue->avail_descs = rc < 0 ? 0 : rc;
1383                 }
1384
1385                 if (unlikely(!rx_queue->avail_descs)) {
1386                         break;
1387                 }
1388
1389                 nb_descs = RTE_MIN(room, rx_queue->avail_descs);
1390
1391                 first_idesc = rx_queue->next_desc;
1392                 last_idesc  = first_idesc + nb_descs;
1393
1394                 rx_queue->next_desc   += nb_descs;
1395                 rx_queue->avail_descs -= nb_descs;
1396
1397                 for (i = 1; i < look_ahead; i++) {
1398                         rte_prefetch0(first_idesc + i);
1399                 }
1400
1401                 PMD_DEBUG_RX("%s:%d: Trying to receive %d packets\n",
1402                              mpipe_name(rx_queue->q.priv),
1403                              rx_queue->q.queue_idx,
1404                              nb_descs);
1405
1406                 for (idesc = first_idesc; idesc < last_idesc; idesc++) {
1407                         struct rte_mbuf *mbuf;
1408
1409                         PMD_DEBUG_RX("%s:%d: processing idesc %d/%d\n",
1410                                      mpipe_name(priv),
1411                                      rx_queue->q.queue_idx,
1412                                      nb_packets, nb_descs);
1413
1414                         rte_prefetch0(idesc + look_ahead);
1415
1416                         PMD_DEBUG_RX("%s:%d: idesc %p, %s%s%s%s%s%s%s%s%s%s"
1417                                      "size: %d, bkt: %d, chan: %d, ring: %d, sqn: %lu, va: %lu\n",
1418                                      mpipe_name(priv),
1419                                      rx_queue->q.queue_idx,
1420                                      idesc,
1421                                      idesc->me ? "me, " : "",
1422                                      idesc->tr ? "tr, " : "",
1423                                      idesc->ce ? "ce, " : "",
1424                                      idesc->ct ? "ct, " : "",
1425                                      idesc->cs ? "cs, " : "",
1426                                      idesc->nr ? "nr, " : "",
1427                                      idesc->sq ? "sq, " : "",
1428                                      idesc->ts ? "ts, " : "",
1429                                      idesc->ps ? "ps, " : "",
1430                                      idesc->be ? "be, " : "",
1431                                      idesc->l2_size,
1432                                      idesc->bucket_id,
1433                                      idesc->channel,
1434                                      idesc->notif_ring,
1435                                      (unsigned long)idesc->packet_sqn,
1436                                      (unsigned long)idesc->va);
1437
1438                         if (unlikely(gxio_mpipe_idesc_has_error(idesc))) {
1439                                 nb_dropped++;
1440                                 gxio_mpipe_iqueue_drop(iqueue, idesc);
1441                                 PMD_DEBUG_RX("%s:%d: Descriptor error\n",
1442                                              mpipe_name(rx_queue->q.priv),
1443                                              rx_queue->q.queue_idx);
1444                                 continue;
1445                         }
1446
1447                         mbuf = __rte_mbuf_raw_alloc(priv->rx_mpool);
1448                         if (unlikely(!mbuf)) {
1449                                 nb_nomem++;
1450                                 gxio_mpipe_iqueue_drop(iqueue, idesc);
1451                                 PMD_DEBUG_RX("%s:%d: RX alloc failure\n",
1452                                              mpipe_name(rx_queue->q.priv),
1453                                              rx_queue->q.queue_idx);
1454                                 continue;
1455                         }
1456
1457                         mpipe_recv_push(priv, mbuf);
1458
1459                         /* Get and setup the mbuf for the received packet. */
1460                         mbuf = mpipe_recv_mbuf(priv, idesc, in_port);
1461
1462                         /* Update results and statistics counters. */
1463                         rx_pkts[nb_packets] = mbuf;
1464                         nb_bytes += mbuf->pkt_len;
1465                         nb_packets++;
1466                 }
1467
1468                 /*
1469                  * We release the ring in bursts, but do not track and release
1470                  * buckets.  This therefore breaks dynamic flow affinity, but
1471                  * we always operate in static affinity mode, and so we're OK
1472                  * with this optimization.
1473                  */
1474                 gxio_mpipe_iqueue_advance(iqueue, nb_descs);
1475                 gxio_mpipe_credit(iqueue->context, iqueue->ring, -1, nb_descs);
1476
1477                 /*
1478                  * Go around once more if we haven't yet peeked the queue, and
1479                  * if we have more room to receive.
1480                  */
1481                 room = nb_pkts - nb_packets;
1482         }
1483
1484         rx_queue->q.stats.packets += nb_packets;
1485         rx_queue->q.stats.bytes   += nb_bytes;
1486         rx_queue->q.stats.errors  += nb_dropped;
1487         rx_queue->q.stats.nomem   += nb_nomem;
1488
1489         PMD_DEBUG_RX("%s:%d: RX: %d/%d pkts/bytes, %d/%d drops/nomem\n",
1490                      mpipe_name(rx_queue->q.priv), rx_queue->q.queue_idx,
1491                      nb_packets, nb_bytes, nb_dropped, nb_nomem);
1492
1493         return nb_packets;
1494 }
1495
1496 static uint16_t
1497 mpipe_recv_pkts(void *_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
1498 {
1499         struct mpipe_rx_queue *rx_queue = _rxq;
1500         uint16_t result = 0;
1501
1502         if (rx_queue) {
1503                 mpipe_dp_enter(rx_queue->q.priv);
1504                 if (likely(rx_queue->q.link_status))
1505                         result = mpipe_do_recv(rx_queue, rx_pkts, nb_pkts);
1506                 mpipe_dp_exit(rx_queue->q.priv);
1507         }
1508
1509         return result;
1510 }
1511
1512 static uint16_t
1513 mpipe_xmit_pkts(void *_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1514 {
1515         struct mpipe_tx_queue *tx_queue = _txq;
1516         uint16_t result = 0;
1517
1518         if (tx_queue) {
1519                 mpipe_dp_enter(tx_queue->q.priv);
1520                 if (likely(tx_queue->q.link_status))
1521                         result = mpipe_do_xmit(tx_queue, tx_pkts, nb_pkts);
1522                 mpipe_dp_exit(tx_queue->q.priv);
1523         }
1524
1525         return result;
1526 }
1527
1528 static int
1529 mpipe_link_mac(const char *ifname, uint8_t *mac)
1530 {
1531         int rc, idx;
1532         char name[GXIO_MPIPE_LINK_NAME_LEN];
1533
1534         for (idx = 0, rc = 0; !rc; idx++) {
1535                 rc = gxio_mpipe_link_enumerate_mac(idx, name, mac);
1536                 if (!rc && !strncmp(name, ifname, GXIO_MPIPE_LINK_NAME_LEN))
1537                         return 0;
1538         }
1539         return -ENODEV;
1540 }
1541
1542 static int
1543 rte_pmd_mpipe_devinit(const char *ifname,
1544                       const char *params __rte_unused)
1545 {
1546         gxio_mpipe_context_t *context;
1547         struct rte_eth_dev *eth_dev;
1548         struct mpipe_dev_priv *priv;
1549         int instance, rc;
1550         uint8_t *mac;
1551
1552         /* Get the mPIPE instance that the device belongs to. */
1553         instance = gxio_mpipe_link_instance(ifname);
1554         context = mpipe_context(instance);
1555         if (!context) {
1556                 RTE_LOG(ERR, PMD, "%s: No device for link.\n", ifname);
1557                 return -ENODEV;
1558         }
1559
1560         priv = rte_zmalloc(NULL, sizeof(*priv), 0);
1561         if (!priv) {
1562                 RTE_LOG(ERR, PMD, "%s: Failed to allocate priv.\n", ifname);
1563                 return -ENOMEM;
1564         }
1565
1566         memset(&priv->tx_stat_mapping, 0xff, sizeof(priv->tx_stat_mapping));
1567         memset(&priv->rx_stat_mapping, 0xff, sizeof(priv->rx_stat_mapping));
1568         priv->context = context;
1569         priv->instance = instance;
1570         priv->is_xaui = (strncmp(ifname, "xgbe", 4) == 0);
1571         priv->pci_dev.numa_node = instance;
1572         priv->channel = -1;
1573
1574         mac = priv->mac_addr.addr_bytes;
1575         rc = mpipe_link_mac(ifname, mac);
1576         if (rc < 0) {
1577                 RTE_LOG(ERR, PMD, "%s: Failed to enumerate link.\n", ifname);
1578                 rte_free(priv);
1579                 return -ENODEV;
1580         }
1581
1582         eth_dev = rte_eth_dev_allocate(ifname, RTE_ETH_DEV_VIRTUAL);
1583         if (!eth_dev) {
1584                 RTE_LOG(ERR, PMD, "%s: Failed to allocate device.\n", ifname);
1585                 rte_free(priv);
1586                 return -ENOMEM;
1587         }
1588
1589         RTE_LOG(INFO, PMD, "%s: Initialized mpipe device"
1590                 "(mac %02x:%02x:%02x:%02x:%02x:%02x).\n",
1591                 ifname, mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]);
1592
1593         priv->eth_dev = eth_dev;
1594         priv->port_id = eth_dev->data->port_id;
1595         eth_dev->data->dev_private = priv;
1596         eth_dev->pci_dev = &priv->pci_dev;
1597         eth_dev->data->mac_addrs = &priv->mac_addr;
1598
1599         eth_dev->data->dev_flags = 0;
1600         eth_dev->data->kdrv = RTE_KDRV_NONE;
1601         eth_dev->driver = NULL;
1602         eth_dev->data->drv_name = drivername;
1603         eth_dev->data->numa_node = instance;
1604
1605         eth_dev->dev_ops      = &mpipe_dev_ops;
1606         eth_dev->rx_pkt_burst = &mpipe_recv_pkts;
1607         eth_dev->tx_pkt_burst = &mpipe_xmit_pkts;
1608
1609         return 0;
1610 }
1611
1612 static struct rte_driver pmd_mpipe_xgbe_drv = {
1613         .name = "xgbe",
1614         .type = PMD_VDEV,
1615         .init = rte_pmd_mpipe_devinit,
1616 };
1617
1618 static struct rte_driver pmd_mpipe_gbe_drv = {
1619         .name = "gbe",
1620         .type = PMD_VDEV,
1621         .init = rte_pmd_mpipe_devinit,
1622 };
1623
1624 PMD_REGISTER_DRIVER(pmd_mpipe_xgbe_drv);
1625 PMD_REGISTER_DRIVER(pmd_mpipe_gbe_drv);
1626
1627 static void __attribute__((constructor, used))
1628 mpipe_init_contexts(void)
1629 {
1630         struct mpipe_context *context;
1631         int rc, instance;
1632
1633         for (instance = 0; instance < GXIO_MPIPE_INSTANCE_MAX; instance++) {
1634                 context = &mpipe_contexts[instance];
1635
1636                 rte_spinlock_init(&context->lock);
1637                 rc = gxio_mpipe_init(&context->context, instance);
1638                 if (rc < 0)
1639                         break;
1640         }
1641
1642         mpipe_instances = instance;
1643 }