drivers/net/avp/avp_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2013-2017 Wind River Systems, Inc.
   3  */
   4
   5 #include <stdint.h>
   6 #include <string.h>
   7 #include <stdio.h>
   8 #include <errno.h>
   9 #include <unistd.h>
  10
  11 #include <ethdev_driver.h>
  12 #include <ethdev_pci.h>
  13 #include <rte_memcpy.h>
  14 #include <rte_string_fns.h>
  15 #include <rte_malloc.h>
  16 #include <rte_atomic.h>
  17 #include <rte_branch_prediction.h>
  18 #include <rte_pci.h>
  19 #include <rte_bus_pci.h>
  20 #include <rte_ether.h>
  21 #include <rte_common.h>
  22 #include <rte_cycles.h>
  23 #include <rte_spinlock.h>
  24 #include <rte_byteorder.h>
  25 #include <rte_dev.h>
  26 #include <rte_memory.h>
  27 #include <rte_eal.h>
  28 #include <rte_io.h>
  29
  30 #include "rte_avp_common.h"
  31 #include "rte_avp_fifo.h"
  32
  33 #include "avp_logs.h"
  34
  35 static int avp_dev_create(struct rte_pci_device *pci_dev,
  36                           struct rte_eth_dev *eth_dev);
  37
  38 static int avp_dev_configure(struct rte_eth_dev *dev);
  39 static int avp_dev_start(struct rte_eth_dev *dev);
  40 static int avp_dev_stop(struct rte_eth_dev *dev);
  41 static int avp_dev_close(struct rte_eth_dev *dev);
  42 static int avp_dev_info_get(struct rte_eth_dev *dev,
  43                             struct rte_eth_dev_info *dev_info);
  44 static int avp_vlan_offload_set(struct rte_eth_dev *dev, int mask);
  45 static int avp_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete);
  46 static int avp_dev_promiscuous_enable(struct rte_eth_dev *dev);
  47 static int avp_dev_promiscuous_disable(struct rte_eth_dev *dev);
  48
  49 static int avp_dev_rx_queue_setup(struct rte_eth_dev *dev,
  50                                   uint16_t rx_queue_id,
  51                                   uint16_t nb_rx_desc,
  52                                   unsigned int socket_id,
  53                                   const struct rte_eth_rxconf *rx_conf,
  54                                   struct rte_mempool *pool);
  55
  56 static int avp_dev_tx_queue_setup(struct rte_eth_dev *dev,
  57                                   uint16_t tx_queue_id,
  58                                   uint16_t nb_tx_desc,
  59                                   unsigned int socket_id,
  60                                   const struct rte_eth_txconf *tx_conf);
  61
  62 static uint16_t avp_recv_scattered_pkts(void *rx_queue,
  63                                         struct rte_mbuf **rx_pkts,
  64                                         uint16_t nb_pkts);
  65
  66 static uint16_t avp_recv_pkts(void *rx_queue,
  67                               struct rte_mbuf **rx_pkts,
  68                               uint16_t nb_pkts);
  69
  70 static uint16_t avp_xmit_scattered_pkts(void *tx_queue,
  71                                         struct rte_mbuf **tx_pkts,
  72                                         uint16_t nb_pkts);
  73
  74 static uint16_t avp_xmit_pkts(void *tx_queue,
  75                               struct rte_mbuf **tx_pkts,
  76                               uint16_t nb_pkts);
  77
  78 static void avp_dev_rx_queue_release(struct rte_eth_dev *dev, uint16_t qid);
  79 static void avp_dev_tx_queue_release(struct rte_eth_dev *dev, uint16_t qid);
  80
  81 static int avp_dev_stats_get(struct rte_eth_dev *dev,
  82                               struct rte_eth_stats *stats);
  83 static int avp_dev_stats_reset(struct rte_eth_dev *dev);
  84
  85
  86 #define AVP_MAX_RX_BURST 64
  87 #define AVP_MAX_TX_BURST 64
  88 #define AVP_MAX_MAC_ADDRS 1
  89 #define AVP_MIN_RX_BUFSIZE RTE_ETHER_MIN_LEN
  90
  91
  92 /*
  93  * Defines the number of microseconds to wait before checking the response
  94  * queue for completion.
  95  */
  96 #define AVP_REQUEST_DELAY_USECS (5000)
  97
  98 /*
  99  * Defines the number times to check the response queue for completion before
 100  * declaring a timeout.
 101  */
 102 #define AVP_MAX_REQUEST_RETRY (100)
 103
 104 /* Defines the current PCI driver version number */
 105 #define AVP_DPDK_DRIVER_VERSION RTE_AVP_CURRENT_GUEST_VERSION
 106
 107 /*
 108  * The set of PCI devices this driver supports
 109  */
 110 static const struct rte_pci_id pci_id_avp_map[] = {
 111         { .vendor_id = RTE_AVP_PCI_VENDOR_ID,
 112           .device_id = RTE_AVP_PCI_DEVICE_ID,
 113           .subsystem_vendor_id = RTE_AVP_PCI_SUB_VENDOR_ID,
 114           .subsystem_device_id = RTE_AVP_PCI_SUB_DEVICE_ID,
 115           .class_id = RTE_CLASS_ANY_ID,
 116         },
 117
 118         { .vendor_id = 0, /* sentinel */
 119         },
 120 };
 121
 122 /*
 123  * dev_ops for avp, bare necessities for basic operation
 124  */
 125 static const struct eth_dev_ops avp_eth_dev_ops = {
 126         .dev_configure       = avp_dev_configure,
 127         .dev_start           = avp_dev_start,
 128         .dev_stop            = avp_dev_stop,
 129         .dev_close           = avp_dev_close,
 130         .dev_infos_get       = avp_dev_info_get,
 131         .vlan_offload_set    = avp_vlan_offload_set,
 132         .stats_get           = avp_dev_stats_get,
 133         .stats_reset         = avp_dev_stats_reset,
 134         .link_update         = avp_dev_link_update,
 135         .promiscuous_enable  = avp_dev_promiscuous_enable,
 136         .promiscuous_disable = avp_dev_promiscuous_disable,
 137         .rx_queue_setup      = avp_dev_rx_queue_setup,
 138         .rx_queue_release    = avp_dev_rx_queue_release,
 139         .tx_queue_setup      = avp_dev_tx_queue_setup,
 140         .tx_queue_release    = avp_dev_tx_queue_release,
 141 };
 142
 143 /**@{ AVP device flags */
 144 #define AVP_F_PROMISC (1 << 1)
 145 #define AVP_F_CONFIGURED (1 << 2)
 146 #define AVP_F_LINKUP (1 << 3)
 147 #define AVP_F_DETACHED (1 << 4)
 148 /**@} */
 149
 150 /* Ethernet device validation marker */
 151 #define AVP_ETHDEV_MAGIC 0x92972862
 152
 153 /*
 154  * Defines the AVP device attributes which are attached to an RTE ethernet
 155  * device
 156  */
 157 struct avp_dev {
 158         uint32_t magic; /**< Memory validation marker */
 159         uint64_t device_id; /**< Unique system identifier */
 160         struct rte_ether_addr ethaddr; /**< Host specified MAC address */
 161         struct rte_eth_dev_data *dev_data;
 162         /**< Back pointer to ethernet device data */
 163         volatile uint32_t flags; /**< Device operational flags */
 164         uint16_t port_id; /**< Ethernet port identifier */
 165         struct rte_mempool *pool; /**< pkt mbuf mempool */
 166         unsigned int guest_mbuf_size; /**< local pool mbuf size */
 167         unsigned int host_mbuf_size; /**< host mbuf size */
 168         unsigned int max_rx_pkt_len; /**< maximum receive unit */
 169         uint32_t host_features; /**< Supported feature bitmap */
 170         uint32_t features; /**< Enabled feature bitmap */
 171         unsigned int num_tx_queues; /**< Negotiated number of transmit queues */
 172         unsigned int max_tx_queues; /**< Maximum number of transmit queues */
 173         unsigned int num_rx_queues; /**< Negotiated number of receive queues */
 174         unsigned int max_rx_queues; /**< Maximum number of receive queues */
 175
 176         struct rte_avp_fifo *tx_q[RTE_AVP_MAX_QUEUES]; /**< TX queue */
 177         struct rte_avp_fifo *rx_q[RTE_AVP_MAX_QUEUES]; /**< RX queue */
 178         struct rte_avp_fifo *alloc_q[RTE_AVP_MAX_QUEUES];
 179         /**< Allocated mbufs queue */
 180         struct rte_avp_fifo *free_q[RTE_AVP_MAX_QUEUES];
 181         /**< To be freed mbufs queue */
 182
 183         /* mutual exclusion over the 'flag' and 'resp_q/req_q' fields */
 184         rte_spinlock_t lock;
 185
 186         /* For request & response */
 187         struct rte_avp_fifo *req_q; /**< Request queue */
 188         struct rte_avp_fifo *resp_q; /**< Response queue */
 189         void *host_sync_addr; /**< (host) Req/Resp Mem address */
 190         void *sync_addr; /**< Req/Resp Mem address */
 191         void *host_mbuf_addr; /**< (host) MBUF pool start address */
 192         void *mbuf_addr; /**< MBUF pool start address */
 193 } __rte_cache_aligned;
 194
 195 /* RTE ethernet private data */
 196 struct avp_adapter {
 197         struct avp_dev avp;
 198 } __rte_cache_aligned;
 199
 200
 201 /* 32-bit MMIO register write */
 202 #define AVP_WRITE32(_value, _addr) rte_write32_relaxed((_value), (_addr))
 203
 204 /* 32-bit MMIO register read */
 205 #define AVP_READ32(_addr) rte_read32_relaxed((_addr))
 206
 207 /* Macro to cast the ethernet device private data to a AVP object */
 208 #define AVP_DEV_PRIVATE_TO_HW(adapter) \
 209         (&((struct avp_adapter *)adapter)->avp)
 210
 211 /*
 212  * Defines the structure of a AVP device queue for the purpose of handling the
 213  * receive and transmit burst callback functions
 214  */
 215 struct avp_queue {
 216         struct rte_eth_dev_data *dev_data;
 217         /**< Backpointer to ethernet device data */
 218         struct avp_dev *avp; /**< Backpointer to AVP device */
 219         uint16_t queue_id;
 220         /**< Queue identifier used for indexing current queue */
 221         uint16_t queue_base;
 222         /**< Base queue identifier for queue servicing */
 223         uint16_t queue_limit;
 224         /**< Maximum queue identifier for queue servicing */
 225
 226         uint64_t packets;
 227         uint64_t bytes;
 228         uint64_t errors;
 229 };
 230
 231 /* send a request and wait for a response
 232  *
 233  * @warning must be called while holding the avp->lock spinlock.
 234  */
 235 static int
 236 avp_dev_process_request(struct avp_dev *avp, struct rte_avp_request *request)
 237 {
 238         unsigned int retry = AVP_MAX_REQUEST_RETRY;
 239         void *resp_addr = NULL;
 240         unsigned int count;
 241         int ret;
 242
 243         PMD_DRV_LOG(DEBUG, "Sending request %u to host\n", request->req_id);
 244
 245         request->result = -ENOTSUP;
 246
 247         /* Discard any stale responses before starting a new request */
 248         while (avp_fifo_get(avp->resp_q, (void **)&resp_addr, 1))
 249                 PMD_DRV_LOG(DEBUG, "Discarding stale response\n");
 250
 251         rte_memcpy(avp->sync_addr, request, sizeof(*request));
 252         count = avp_fifo_put(avp->req_q, &avp->host_sync_addr, 1);
 253         if (count < 1) {
 254                 PMD_DRV_LOG(ERR, "Cannot send request %u to host\n",
 255                             request->req_id);
 256                 ret = -EBUSY;
 257                 goto done;
 258         }
 259
 260         while (retry--) {
 261                 /* wait for a response */
 262                 usleep(AVP_REQUEST_DELAY_USECS);
 263
 264                 count = avp_fifo_count(avp->resp_q);
 265                 if (count >= 1) {
 266                         /* response received */
 267                         break;
 268                 }
 269
 270                 if (retry == 0) {
 271                         PMD_DRV_LOG(ERR, "Timeout while waiting for a response for %u\n",
 272                                     request->req_id);
 273                         ret = -ETIME;
 274                         goto done;
 275                 }
 276         }
 277
 278         /* retrieve the response */
 279         count = avp_fifo_get(avp->resp_q, (void **)&resp_addr, 1);
 280         if ((count != 1) || (resp_addr != avp->host_sync_addr)) {
 281                 PMD_DRV_LOG(ERR, "Invalid response from host, count=%u resp=%p host_sync_addr=%p\n",
 282                             count, resp_addr, avp->host_sync_addr);
 283                 ret = -ENODATA;
 284                 goto done;
 285         }
 286
 287         /* copy to user buffer */
 288         rte_memcpy(request, avp->sync_addr, sizeof(*request));
 289         ret = 0;
 290
 291         PMD_DRV_LOG(DEBUG, "Result %d received for request %u\n",
 292                     request->result, request->req_id);
 293
 294 done:
 295         return ret;
 296 }
 297
 298 static int
 299 avp_dev_ctrl_set_link_state(struct rte_eth_dev *eth_dev, unsigned int state)
 300 {
 301         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 302         struct rte_avp_request request;
 303         int ret;
 304
 305         /* setup a link state change request */
 306         memset(&request, 0, sizeof(request));
 307         request.req_id = RTE_AVP_REQ_CFG_NETWORK_IF;
 308         request.if_up = state;
 309
 310         ret = avp_dev_process_request(avp, &request);
 311
 312         return ret == 0 ? request.result : ret;
 313 }
 314
 315 static int
 316 avp_dev_ctrl_set_config(struct rte_eth_dev *eth_dev,
 317                         struct rte_avp_device_config *config)
 318 {
 319         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 320         struct rte_avp_request request;
 321         int ret;
 322
 323         /* setup a configure request */
 324         memset(&request, 0, sizeof(request));
 325         request.req_id = RTE_AVP_REQ_CFG_DEVICE;
 326         memcpy(&request.config, config, sizeof(request.config));
 327
 328         ret = avp_dev_process_request(avp, &request);
 329
 330         return ret == 0 ? request.result : ret;
 331 }
 332
 333 static int
 334 avp_dev_ctrl_shutdown(struct rte_eth_dev *eth_dev)
 335 {
 336         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 337         struct rte_avp_request request;
 338         int ret;
 339
 340         /* setup a shutdown request */
 341         memset(&request, 0, sizeof(request));
 342         request.req_id = RTE_AVP_REQ_SHUTDOWN_DEVICE;
 343
 344         ret = avp_dev_process_request(avp, &request);
 345
 346         return ret == 0 ? request.result : ret;
 347 }
 348
 349 /* translate from host mbuf virtual address to guest virtual address */
 350 static inline void *
 351 avp_dev_translate_buffer(struct avp_dev *avp, void *host_mbuf_address)
 352 {
 353         return RTE_PTR_ADD(RTE_PTR_SUB(host_mbuf_address,
 354                                        (uintptr_t)avp->host_mbuf_addr),
 355                            (uintptr_t)avp->mbuf_addr);
 356 }
 357
 358 /* translate from host physical address to guest virtual address */
 359 static void *
 360 avp_dev_translate_address(struct rte_eth_dev *eth_dev,
 361                           rte_iova_t host_phys_addr)
 362 {
 363         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 364         struct rte_mem_resource *resource;
 365         struct rte_avp_memmap_info *info;
 366         struct rte_avp_memmap *map;
 367         off_t offset;
 368         void *addr;
 369         unsigned int i;
 370
 371         addr = pci_dev->mem_resource[RTE_AVP_PCI_MEMORY_BAR].addr;
 372         resource = &pci_dev->mem_resource[RTE_AVP_PCI_MEMMAP_BAR];
 373         info = (struct rte_avp_memmap_info *)resource->addr;
 374
 375         offset = 0;
 376         for (i = 0; i < info->nb_maps; i++) {
 377                 /* search all segments looking for a matching address */
 378                 map = &info->maps[i];
 379
 380                 if ((host_phys_addr >= map->phys_addr) &&
 381                         (host_phys_addr < (map->phys_addr + map->length))) {
 382                         /* address is within this segment */
 383                         offset += (host_phys_addr - map->phys_addr);
 384                         addr = RTE_PTR_ADD(addr, (uintptr_t)offset);
 385
 386                         PMD_DRV_LOG(DEBUG, "Translating host physical 0x%" PRIx64 " to guest virtual 0x%p\n",
 387                                     host_phys_addr, addr);
 388
 389                         return addr;
 390                 }
 391                 offset += map->length;
 392         }
 393
 394         return NULL;
 395 }
 396
 397 /* verify that the incoming device version is compatible with our version */
 398 static int
 399 avp_dev_version_check(uint32_t version)
 400 {
 401         uint32_t driver = RTE_AVP_STRIP_MINOR_VERSION(AVP_DPDK_DRIVER_VERSION);
 402         uint32_t device = RTE_AVP_STRIP_MINOR_VERSION(version);
 403
 404         if (device <= driver) {
 405                 /* the host driver version is less than or equal to ours */
 406                 return 0;
 407         }
 408
 409         return 1;
 410 }
 411
 412 /* verify that memory regions have expected version and validation markers */
 413 static int
 414 avp_dev_check_regions(struct rte_eth_dev *eth_dev)
 415 {
 416         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 417         struct rte_avp_memmap_info *memmap;
 418         struct rte_avp_device_info *info;
 419         struct rte_mem_resource *resource;
 420         unsigned int i;
 421
 422         /* Dump resource info for debug */
 423         for (i = 0; i < PCI_MAX_RESOURCE; i++) {
 424                 resource = &pci_dev->mem_resource[i];
 425                 if ((resource->phys_addr == 0) || (resource->len == 0))
 426                         continue;
 427
 428                 PMD_DRV_LOG(DEBUG, "resource[%u]: phys=0x%" PRIx64 " len=%" PRIu64 " addr=%p\n",
 429                             i, resource->phys_addr,
 430                             resource->len, resource->addr);
 431
 432                 switch (i) {
 433                 case RTE_AVP_PCI_MEMMAP_BAR:
 434                         memmap = (struct rte_avp_memmap_info *)resource->addr;
 435                         if ((memmap->magic != RTE_AVP_MEMMAP_MAGIC) ||
 436                             (memmap->version != RTE_AVP_MEMMAP_VERSION)) {
 437                                 PMD_DRV_LOG(ERR, "Invalid memmap magic 0x%08x and version %u\n",
 438                                             memmap->magic, memmap->version);
 439                                 return -EINVAL;
 440                         }
 441                         break;
 442
 443                 case RTE_AVP_PCI_DEVICE_BAR:
 444                         info = (struct rte_avp_device_info *)resource->addr;
 445                         if ((info->magic != RTE_AVP_DEVICE_MAGIC) ||
 446                             avp_dev_version_check(info->version)) {
 447                                 PMD_DRV_LOG(ERR, "Invalid device info magic 0x%08x or version 0x%08x > 0x%08x\n",
 448                                             info->magic, info->version,
 449                                             AVP_DPDK_DRIVER_VERSION);
 450                                 return -EINVAL;
 451                         }
 452                         break;
 453
 454                 case RTE_AVP_PCI_MEMORY_BAR:
 455                 case RTE_AVP_PCI_MMIO_BAR:
 456                         if (resource->addr == NULL) {
 457                                 PMD_DRV_LOG(ERR, "Missing address space for BAR%u\n",
 458                                             i);
 459                                 return -EINVAL;
 460                         }
 461                         break;
 462
 463                 case RTE_AVP_PCI_MSIX_BAR:
 464                 default:
 465                         /* no validation required */
 466                         break;
 467                 }
 468         }
 469
 470         return 0;
 471 }
 472
 473 static int
 474 avp_dev_detach(struct rte_eth_dev *eth_dev)
 475 {
 476         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 477         int ret;
 478
 479         PMD_DRV_LOG(NOTICE, "Detaching port %u from AVP device 0x%" PRIx64 "\n",
 480                     eth_dev->data->port_id, avp->device_id);
 481
 482         rte_spinlock_lock(&avp->lock);
 483
 484         if (avp->flags & AVP_F_DETACHED) {
 485                 PMD_DRV_LOG(NOTICE, "port %u already detached\n",
 486                             eth_dev->data->port_id);
 487                 ret = 0;
 488                 goto unlock;
 489         }
 490
 491         /* shutdown the device first so the host stops sending us packets. */
 492         ret = avp_dev_ctrl_shutdown(eth_dev);
 493         if (ret < 0) {
 494                 PMD_DRV_LOG(ERR, "Failed to send/recv shutdown to host, ret=%d\n",
 495                             ret);
 496                 avp->flags &= ~AVP_F_DETACHED;
 497                 goto unlock;
 498         }
 499
 500         avp->flags |= AVP_F_DETACHED;
 501         rte_wmb();
 502
 503         /* wait for queues to acknowledge the presence of the detach flag */
 504         rte_delay_ms(1);
 505
 506         ret = 0;
 507
 508 unlock:
 509         rte_spinlock_unlock(&avp->lock);
 510         return ret;
 511 }
 512
 513 static void
 514 _avp_set_rx_queue_mappings(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
 515 {
 516         struct avp_dev *avp =
 517                 AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 518         struct avp_queue *rxq;
 519         uint16_t queue_count;
 520         uint16_t remainder;
 521
 522         rxq = (struct avp_queue *)eth_dev->data->rx_queues[rx_queue_id];
 523
 524         /*
 525          * Must map all AVP fifos as evenly as possible between the configured
 526          * device queues.  Each device queue will service a subset of the AVP
 527          * fifos. If there is an odd number of device queues the first set of
 528          * device queues will get the extra AVP fifos.
 529          */
 530         queue_count = avp->num_rx_queues / eth_dev->data->nb_rx_queues;
 531         remainder = avp->num_rx_queues % eth_dev->data->nb_rx_queues;
 532         if (rx_queue_id < remainder) {
 533                 /* these queues must service one extra FIFO */
 534                 rxq->queue_base = rx_queue_id * (queue_count + 1);
 535                 rxq->queue_limit = rxq->queue_base + (queue_count + 1) - 1;
 536         } else {
 537                 /* these queues service the regular number of FIFO */
 538                 rxq->queue_base = ((remainder * (queue_count + 1)) +
 539                                    ((rx_queue_id - remainder) * queue_count));
 540                 rxq->queue_limit = rxq->queue_base + queue_count - 1;
 541         }
 542
 543         PMD_DRV_LOG(DEBUG, "rxq %u at %p base %u limit %u\n",
 544                     rx_queue_id, rxq, rxq->queue_base, rxq->queue_limit);
 545
 546         rxq->queue_id = rxq->queue_base;
 547 }
 548
 549 static void
 550 _avp_set_queue_counts(struct rte_eth_dev *eth_dev)
 551 {
 552         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 553         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 554         struct rte_avp_device_info *host_info;
 555         void *addr;
 556
 557         addr = pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR].addr;
 558         host_info = (struct rte_avp_device_info *)addr;
 559
 560         /*
 561          * the transmit direction is not negotiated beyond respecting the max
 562          * number of queues because the host can handle arbitrary guest tx
 563          * queues (host rx queues).
 564          */
 565         avp->num_tx_queues = eth_dev->data->nb_tx_queues;
 566
 567         /*
 568          * the receive direction is more restrictive.  The host requires a
 569          * minimum number of guest rx queues (host tx queues) therefore
 570          * negotiate a value that is at least as large as the host minimum
 571          * requirement.  If the host and guest values are not identical then a
 572          * mapping will be established in the receive_queue_setup function.
 573          */
 574         avp->num_rx_queues = RTE_MAX(host_info->min_rx_queues,
 575                                      eth_dev->data->nb_rx_queues);
 576
 577         PMD_DRV_LOG(DEBUG, "Requesting %u Tx and %u Rx queues from host\n",
 578                     avp->num_tx_queues, avp->num_rx_queues);
 579 }
 580
 581 static int
 582 avp_dev_attach(struct rte_eth_dev *eth_dev)
 583 {
 584         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 585         struct rte_avp_device_config config;
 586         unsigned int i;
 587         int ret;
 588
 589         PMD_DRV_LOG(NOTICE, "Attaching port %u to AVP device 0x%" PRIx64 "\n",
 590                     eth_dev->data->port_id, avp->device_id);
 591
 592         rte_spinlock_lock(&avp->lock);
 593
 594         if (!(avp->flags & AVP_F_DETACHED)) {
 595                 PMD_DRV_LOG(NOTICE, "port %u already attached\n",
 596                             eth_dev->data->port_id);
 597                 ret = 0;
 598                 goto unlock;
 599         }
 600
 601         /*
 602          * make sure that the detached flag is set prior to reconfiguring the
 603          * queues.
 604          */
 605         avp->flags |= AVP_F_DETACHED;
 606         rte_wmb();
 607
 608         /*
 609          * re-run the device create utility which will parse the new host info
 610          * and setup the AVP device queue pointers.
 611          */
 612         ret = avp_dev_create(RTE_ETH_DEV_TO_PCI(eth_dev), eth_dev);
 613         if (ret < 0) {
 614                 PMD_DRV_LOG(ERR, "Failed to re-create AVP device, ret=%d\n",
 615                             ret);
 616                 goto unlock;
 617         }
 618
 619         if (avp->flags & AVP_F_CONFIGURED) {
 620                 /*
 621                  * Update the receive queue mapping to handle cases where the
 622                  * source and destination hosts have different queue
 623                  * requirements.  As long as the DETACHED flag is asserted the
 624                  * queue table should not be referenced so it should be safe to
 625                  * update it.
 626                  */
 627                 _avp_set_queue_counts(eth_dev);
 628                 for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
 629                         _avp_set_rx_queue_mappings(eth_dev, i);
 630
 631                 /*
 632                  * Update the host with our config details so that it knows the
 633                  * device is active.
 634                  */
 635                 memset(&config, 0, sizeof(config));
 636                 config.device_id = avp->device_id;
 637                 config.driver_type = RTE_AVP_DRIVER_TYPE_DPDK;
 638                 config.driver_version = AVP_DPDK_DRIVER_VERSION;
 639                 config.features = avp->features;
 640                 config.num_tx_queues = avp->num_tx_queues;
 641                 config.num_rx_queues = avp->num_rx_queues;
 642                 config.if_up = !!(avp->flags & AVP_F_LINKUP);
 643
 644                 ret = avp_dev_ctrl_set_config(eth_dev, &config);
 645                 if (ret < 0) {
 646                         PMD_DRV_LOG(ERR, "Config request failed by host, ret=%d\n",
 647                                     ret);
 648                         goto unlock;
 649                 }
 650         }
 651
 652         rte_wmb();
 653         avp->flags &= ~AVP_F_DETACHED;
 654
 655         ret = 0;
 656
 657 unlock:
 658         rte_spinlock_unlock(&avp->lock);
 659         return ret;
 660 }
 661
 662 static void
 663 avp_dev_interrupt_handler(void *data)
 664 {
 665         struct rte_eth_dev *eth_dev = data;
 666         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 667         void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
 668         uint32_t status, value;
 669         int ret;
 670
 671         if (registers == NULL)
 672                 rte_panic("no mapped MMIO register space\n");
 673
 674         /* read the interrupt status register
 675          * note: this register clears on read so all raised interrupts must be
 676          *    handled or remembered for later processing
 677          */
 678         status = AVP_READ32(
 679                 RTE_PTR_ADD(registers,
 680                             RTE_AVP_INTERRUPT_STATUS_OFFSET));
 681
 682         if (status & RTE_AVP_MIGRATION_INTERRUPT_MASK) {
 683                 /* handle interrupt based on current status */
 684                 value = AVP_READ32(
 685                         RTE_PTR_ADD(registers,
 686                                     RTE_AVP_MIGRATION_STATUS_OFFSET));
 687                 switch (value) {
 688                 case RTE_AVP_MIGRATION_DETACHED:
 689                         ret = avp_dev_detach(eth_dev);
 690                         break;
 691                 case RTE_AVP_MIGRATION_ATTACHED:
 692                         ret = avp_dev_attach(eth_dev);
 693                         break;
 694                 default:
 695                         PMD_DRV_LOG(ERR, "unexpected migration status, status=%u\n",
 696                                     value);
 697                         ret = -EINVAL;
 698                 }
 699
 700                 /* acknowledge the request by writing out our current status */
 701                 value = (ret == 0 ? value : RTE_AVP_MIGRATION_ERROR);
 702                 AVP_WRITE32(value,
 703                             RTE_PTR_ADD(registers,
 704                                         RTE_AVP_MIGRATION_ACK_OFFSET));
 705
 706                 PMD_DRV_LOG(NOTICE, "AVP migration interrupt handled\n");
 707         }
 708
 709         if (status & ~RTE_AVP_MIGRATION_INTERRUPT_MASK)
 710                 PMD_DRV_LOG(WARNING, "AVP unexpected interrupt, status=0x%08x\n",
 711                             status);
 712
 713         /* re-enable UIO interrupt handling */
 714         ret = rte_intr_ack(pci_dev->intr_handle);
 715         if (ret < 0) {
 716                 PMD_DRV_LOG(ERR, "Failed to re-enable UIO interrupts, ret=%d\n",
 717                             ret);
 718                 /* continue */
 719         }
 720 }
 721
 722 static int
 723 avp_dev_enable_interrupts(struct rte_eth_dev *eth_dev)
 724 {
 725         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 726         void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
 727         int ret;
 728
 729         if (registers == NULL)
 730                 return -EINVAL;
 731
 732         /* enable UIO interrupt handling */
 733         ret = rte_intr_enable(pci_dev->intr_handle);
 734         if (ret < 0) {
 735                 PMD_DRV_LOG(ERR, "Failed to enable UIO interrupts, ret=%d\n",
 736                             ret);
 737                 return ret;
 738         }
 739
 740         /* inform the device that all interrupts are enabled */
 741         AVP_WRITE32(RTE_AVP_APP_INTERRUPTS_MASK,
 742                     RTE_PTR_ADD(registers, RTE_AVP_INTERRUPT_MASK_OFFSET));
 743
 744         return 0;
 745 }
 746
 747 static int
 748 avp_dev_disable_interrupts(struct rte_eth_dev *eth_dev)
 749 {
 750         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 751         void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
 752         int ret;
 753
 754         if (registers == NULL)
 755                 return 0;
 756
 757         /* inform the device that all interrupts are disabled */
 758         AVP_WRITE32(RTE_AVP_NO_INTERRUPTS_MASK,
 759                     RTE_PTR_ADD(registers, RTE_AVP_INTERRUPT_MASK_OFFSET));
 760
 761         /* enable UIO interrupt handling */
 762         ret = rte_intr_disable(pci_dev->intr_handle);
 763         if (ret < 0) {
 764                 PMD_DRV_LOG(ERR, "Failed to disable UIO interrupts, ret=%d\n",
 765                             ret);
 766                 return ret;
 767         }
 768
 769         return 0;
 770 }
 771
 772 static int
 773 avp_dev_setup_interrupts(struct rte_eth_dev *eth_dev)
 774 {
 775         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 776         int ret;
 777
 778         /* register a callback handler with UIO for interrupt notifications */
 779         ret = rte_intr_callback_register(pci_dev->intr_handle,
 780                                          avp_dev_interrupt_handler,
 781                                          (void *)eth_dev);
 782         if (ret < 0) {
 783                 PMD_DRV_LOG(ERR, "Failed to register UIO interrupt callback, ret=%d\n",
 784                             ret);
 785                 return ret;
 786         }
 787
 788         /* enable interrupt processing */
 789         return avp_dev_enable_interrupts(eth_dev);
 790 }
 791
 792 static int
 793 avp_dev_migration_pending(struct rte_eth_dev *eth_dev)
 794 {
 795         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 796         void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
 797         uint32_t value;
 798
 799         if (registers == NULL)
 800                 return 0;
 801
 802         value = AVP_READ32(RTE_PTR_ADD(registers,
 803                                        RTE_AVP_MIGRATION_STATUS_OFFSET));
 804         if (value == RTE_AVP_MIGRATION_DETACHED) {
 805                 /* migration is in progress; ack it if we have not already */
 806                 AVP_WRITE32(value,
 807                             RTE_PTR_ADD(registers,
 808                                         RTE_AVP_MIGRATION_ACK_OFFSET));
 809                 return 1;
 810         }
 811         return 0;
 812 }
 813
 814 /*
 815  * create a AVP device using the supplied device info by first translating it
 816  * to guest address space(s).
 817  */
 818 static int
 819 avp_dev_create(struct rte_pci_device *pci_dev,
 820                struct rte_eth_dev *eth_dev)
 821 {
 822         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 823         struct rte_avp_device_info *host_info;
 824         struct rte_mem_resource *resource;
 825         unsigned int i;
 826
 827         resource = &pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR];
 828         if (resource->addr == NULL) {
 829                 PMD_DRV_LOG(ERR, "BAR%u is not mapped\n",
 830                             RTE_AVP_PCI_DEVICE_BAR);
 831                 return -EFAULT;
 832         }
 833         host_info = (struct rte_avp_device_info *)resource->addr;
 834
 835         if ((host_info->magic != RTE_AVP_DEVICE_MAGIC) ||
 836                 avp_dev_version_check(host_info->version)) {
 837                 PMD_DRV_LOG(ERR, "Invalid AVP PCI device, magic 0x%08x version 0x%08x > 0x%08x\n",
 838                             host_info->magic, host_info->version,
 839                             AVP_DPDK_DRIVER_VERSION);
 840                 return -EINVAL;
 841         }
 842
 843         PMD_DRV_LOG(DEBUG, "AVP host device is v%u.%u.%u\n",
 844                     RTE_AVP_GET_RELEASE_VERSION(host_info->version),
 845                     RTE_AVP_GET_MAJOR_VERSION(host_info->version),
 846                     RTE_AVP_GET_MINOR_VERSION(host_info->version));
 847
 848         PMD_DRV_LOG(DEBUG, "AVP host supports %u to %u TX queue(s)\n",
 849                     host_info->min_tx_queues, host_info->max_tx_queues);
 850         PMD_DRV_LOG(DEBUG, "AVP host supports %u to %u RX queue(s)\n",
 851                     host_info->min_rx_queues, host_info->max_rx_queues);
 852         PMD_DRV_LOG(DEBUG, "AVP host supports features 0x%08x\n",
 853                     host_info->features);
 854
 855         if (avp->magic != AVP_ETHDEV_MAGIC) {
 856                 /*
 857                  * First time initialization (i.e., not during a VM
 858                  * migration)
 859                  */
 860                 memset(avp, 0, sizeof(*avp));
 861                 avp->magic = AVP_ETHDEV_MAGIC;
 862                 avp->dev_data = eth_dev->data;
 863                 avp->port_id = eth_dev->data->port_id;
 864                 avp->host_mbuf_size = host_info->mbuf_size;
 865                 avp->host_features = host_info->features;
 866                 rte_spinlock_init(&avp->lock);
 867                 memcpy(&avp->ethaddr.addr_bytes[0],
 868                        host_info->ethaddr, RTE_ETHER_ADDR_LEN);
 869                 /* adjust max values to not exceed our max */
 870                 avp->max_tx_queues =
 871                         RTE_MIN(host_info->max_tx_queues, RTE_AVP_MAX_QUEUES);
 872                 avp->max_rx_queues =
 873                         RTE_MIN(host_info->max_rx_queues, RTE_AVP_MAX_QUEUES);
 874         } else {
 875                 /* Re-attaching during migration */
 876
 877                 /* TODO... requires validation of host values */
 878                 if ((host_info->features & avp->features) != avp->features) {
 879                         PMD_DRV_LOG(ERR, "AVP host features mismatched; 0x%08x, host=0x%08x\n",
 880                                     avp->features, host_info->features);
 881                         /* this should not be possible; continue for now */
 882                 }
 883         }
 884
 885         /* the device id is allowed to change over migrations */
 886         avp->device_id = host_info->device_id;
 887
 888         /* translate incoming host addresses to guest address space */
 889         PMD_DRV_LOG(DEBUG, "AVP first host tx queue at 0x%" PRIx64 "\n",
 890                     host_info->tx_phys);
 891         PMD_DRV_LOG(DEBUG, "AVP first host alloc queue at 0x%" PRIx64 "\n",
 892                     host_info->alloc_phys);
 893         for (i = 0; i < avp->max_tx_queues; i++) {
 894                 avp->tx_q[i] = avp_dev_translate_address(eth_dev,
 895                         host_info->tx_phys + (i * host_info->tx_size));
 896
 897                 avp->alloc_q[i] = avp_dev_translate_address(eth_dev,
 898                         host_info->alloc_phys + (i * host_info->alloc_size));
 899         }
 900
 901         PMD_DRV_LOG(DEBUG, "AVP first host rx queue at 0x%" PRIx64 "\n",
 902                     host_info->rx_phys);
 903         PMD_DRV_LOG(DEBUG, "AVP first host free queue at 0x%" PRIx64 "\n",
 904                     host_info->free_phys);
 905         for (i = 0; i < avp->max_rx_queues; i++) {
 906                 avp->rx_q[i] = avp_dev_translate_address(eth_dev,
 907                         host_info->rx_phys + (i * host_info->rx_size));
 908                 avp->free_q[i] = avp_dev_translate_address(eth_dev,
 909                         host_info->free_phys + (i * host_info->free_size));
 910         }
 911
 912         PMD_DRV_LOG(DEBUG, "AVP host request queue at 0x%" PRIx64 "\n",
 913                     host_info->req_phys);
 914         PMD_DRV_LOG(DEBUG, "AVP host response queue at 0x%" PRIx64 "\n",
 915                     host_info->resp_phys);
 916         PMD_DRV_LOG(DEBUG, "AVP host sync address at 0x%" PRIx64 "\n",
 917                     host_info->sync_phys);
 918         PMD_DRV_LOG(DEBUG, "AVP host mbuf address at 0x%" PRIx64 "\n",
 919                     host_info->mbuf_phys);
 920         avp->req_q = avp_dev_translate_address(eth_dev, host_info->req_phys);
 921         avp->resp_q = avp_dev_translate_address(eth_dev, host_info->resp_phys);
 922         avp->sync_addr =
 923                 avp_dev_translate_address(eth_dev, host_info->sync_phys);
 924         avp->mbuf_addr =
 925                 avp_dev_translate_address(eth_dev, host_info->mbuf_phys);
 926
 927         /*
 928          * store the host mbuf virtual address so that we can calculate
 929          * relative offsets for each mbuf as they are processed
 930          */
 931         avp->host_mbuf_addr = host_info->mbuf_va;
 932         avp->host_sync_addr = host_info->sync_va;
 933
 934         /*
 935          * store the maximum packet length that is supported by the host.
 936          */
 937         avp->max_rx_pkt_len = host_info->max_rx_pkt_len;
 938         PMD_DRV_LOG(DEBUG, "AVP host max receive packet length is %u\n",
 939                                 host_info->max_rx_pkt_len);
 940
 941         return 0;
 942 }
 943
 944 /*
 945  * This function is based on probe() function in avp_pci.c
 946  * It returns 0 on success.
 947  */
 948 static int
 949 eth_avp_dev_init(struct rte_eth_dev *eth_dev)
 950 {
 951         struct avp_dev *avp =
 952                 AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 953         struct rte_pci_device *pci_dev;
 954         int ret;
 955
 956         pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 957         eth_dev->dev_ops = &avp_eth_dev_ops;
 958         eth_dev->rx_pkt_burst = &avp_recv_pkts;
 959         eth_dev->tx_pkt_burst = &avp_xmit_pkts;
 960
 961         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 962                 /*
 963                  * no setup required on secondary processes.  All data is saved
 964                  * in dev_private by the primary process. All resource should
 965                  * be mapped to the same virtual address so all pointers should
 966                  * be valid.
 967                  */
 968                 if (eth_dev->data->scattered_rx) {
 969                         PMD_DRV_LOG(NOTICE, "AVP device configured for chained mbufs\n");
 970                         eth_dev->rx_pkt_burst = avp_recv_scattered_pkts;
 971                         eth_dev->tx_pkt_burst = avp_xmit_scattered_pkts;
 972                 }
 973                 return 0;
 974         }
 975
 976         rte_eth_copy_pci_info(eth_dev, pci_dev);
 977         eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
 978
 979         /* Check current migration status */
 980         if (avp_dev_migration_pending(eth_dev)) {
 981                 PMD_DRV_LOG(ERR, "VM live migration operation in progress\n");
 982                 return -EBUSY;
 983         }
 984
 985         /* Check BAR resources */
 986         ret = avp_dev_check_regions(eth_dev);
 987         if (ret < 0) {
 988                 PMD_DRV_LOG(ERR, "Failed to validate BAR resources, ret=%d\n",
 989                             ret);
 990                 return ret;
 991         }
 992
 993         /* Enable interrupts */
 994         ret = avp_dev_setup_interrupts(eth_dev);
 995         if (ret < 0) {
 996                 PMD_DRV_LOG(ERR, "Failed to enable interrupts, ret=%d\n", ret);
 997                 return ret;
 998         }
 999
1000         /* Handle each subtype */
1001         ret = avp_dev_create(pci_dev, eth_dev);
1002         if (ret < 0) {
1003                 PMD_DRV_LOG(ERR, "Failed to create device, ret=%d\n", ret);
1004                 return ret;
1005         }
1006
1007         /* Allocate memory for storing MAC addresses */
1008         eth_dev->data->mac_addrs = rte_zmalloc("avp_ethdev",
1009                                         RTE_ETHER_ADDR_LEN, 0);
1010         if (eth_dev->data->mac_addrs == NULL) {
1011                 PMD_DRV_LOG(ERR, "Failed to allocate %d bytes needed to store MAC addresses\n",
1012                             RTE_ETHER_ADDR_LEN);
1013                 return -ENOMEM;
1014         }
1015
1016         /* Get a mac from device config */
1017         rte_ether_addr_copy(&avp->ethaddr, &eth_dev->data->mac_addrs[0]);
1018
1019         return 0;
1020 }
1021
1022 static int
1023 eth_avp_dev_uninit(struct rte_eth_dev *eth_dev)
1024 {
1025         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1026                 return -EPERM;
1027
1028         if (eth_dev->data == NULL)
1029                 return 0;
1030
1031         avp_dev_close(eth_dev);
1032
1033         return 0;
1034 }
1035
1036 static int
1037 eth_avp_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1038                   struct rte_pci_device *pci_dev)
1039 {
1040         return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct avp_adapter),
1041                         eth_avp_dev_init);
1042 }
1043
1044 static int
1045 eth_avp_pci_remove(struct rte_pci_device *pci_dev)
1046 {
1047         return rte_eth_dev_pci_generic_remove(pci_dev,
1048                                               eth_avp_dev_uninit);
1049 }
1050
1051 static struct rte_pci_driver rte_avp_pmd = {
1052         .id_table = pci_id_avp_map,
1053         .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1054         .probe = eth_avp_pci_probe,
1055         .remove = eth_avp_pci_remove,
1056 };
1057
1058 static int
1059 avp_dev_enable_scattered(struct rte_eth_dev *eth_dev,
1060                          struct avp_dev *avp)
1061 {
1062         unsigned int max_rx_pktlen;
1063
1064         max_rx_pktlen = eth_dev->data->mtu + RTE_ETHER_HDR_LEN +
1065                 RTE_ETHER_CRC_LEN;
1066
1067         if (max_rx_pktlen > avp->guest_mbuf_size ||
1068             max_rx_pktlen > avp->host_mbuf_size) {
1069                 /*
1070                  * If the guest MTU is greater than either the host or guest
1071                  * buffers then chained mbufs have to be enabled in the TX
1072                  * direction.  It is assumed that the application will not need
1073                  * to send packets larger than their MTU.
1074                  */
1075                 return 1;
1076         }
1077
1078         if ((avp->max_rx_pkt_len > avp->guest_mbuf_size) ||
1079             (avp->max_rx_pkt_len > avp->host_mbuf_size)) {
1080                 /*
1081                  * If the host MRU is greater than its own mbuf size or the
1082                  * guest mbuf size then chained mbufs have to be enabled in the
1083                  * RX direction.
1084                  */
1085                 return 1;
1086         }
1087
1088         return 0;
1089 }
1090
1091 static int
1092 avp_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
1093                        uint16_t rx_queue_id,
1094                        uint16_t nb_rx_desc,
1095                        unsigned int socket_id,
1096                        const struct rte_eth_rxconf *rx_conf,
1097                        struct rte_mempool *pool)
1098 {
1099         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1100         struct rte_pktmbuf_pool_private *mbp_priv;
1101         struct avp_queue *rxq;
1102
1103         if (rx_queue_id >= eth_dev->data->nb_rx_queues) {
1104                 PMD_DRV_LOG(ERR, "RX queue id is out of range: rx_queue_id=%u, nb_rx_queues=%u\n",
1105                             rx_queue_id, eth_dev->data->nb_rx_queues);
1106                 return -EINVAL;
1107         }
1108
1109         /* Save mbuf pool pointer */
1110         avp->pool = pool;
1111
1112         /* Save the local mbuf size */
1113         mbp_priv = rte_mempool_get_priv(pool);
1114         avp->guest_mbuf_size = (uint16_t)(mbp_priv->mbuf_data_room_size);
1115         avp->guest_mbuf_size -= RTE_PKTMBUF_HEADROOM;
1116
1117         if (avp_dev_enable_scattered(eth_dev, avp)) {
1118                 if (!eth_dev->data->scattered_rx) {
1119                         PMD_DRV_LOG(NOTICE, "AVP device configured for chained mbufs\n");
1120                         eth_dev->data->scattered_rx = 1;
1121                         eth_dev->rx_pkt_burst = avp_recv_scattered_pkts;
1122                         eth_dev->tx_pkt_burst = avp_xmit_scattered_pkts;
1123                 }
1124         }
1125
1126         PMD_DRV_LOG(DEBUG, "AVP max_rx_pkt_len=(%u,%u) mbuf_size=(%u,%u)\n",
1127                     avp->max_rx_pkt_len,
1128                     eth_dev->data->mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN,
1129                     avp->host_mbuf_size,
1130                     avp->guest_mbuf_size);
1131
1132         /* allocate a queue object */
1133         rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct avp_queue),
1134                                  RTE_CACHE_LINE_SIZE, socket_id);
1135         if (rxq == NULL) {
1136                 PMD_DRV_LOG(ERR, "Failed to allocate new Rx queue object\n");
1137                 return -ENOMEM;
1138         }
1139
1140         /* save back pointers to AVP and Ethernet devices */
1141         rxq->avp = avp;
1142         rxq->dev_data = eth_dev->data;
1143         eth_dev->data->rx_queues[rx_queue_id] = (void *)rxq;
1144
1145         /* setup the queue receive mapping for the current queue. */
1146         _avp_set_rx_queue_mappings(eth_dev, rx_queue_id);
1147
1148         PMD_DRV_LOG(DEBUG, "Rx queue %u setup at %p\n", rx_queue_id, rxq);
1149
1150         (void)nb_rx_desc;
1151         (void)rx_conf;
1152         return 0;
1153 }
1154
1155 static int
1156 avp_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
1157                        uint16_t tx_queue_id,
1158                        uint16_t nb_tx_desc,
1159                        unsigned int socket_id,
1160                        const struct rte_eth_txconf *tx_conf)
1161 {
1162         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1163         struct avp_queue *txq;
1164
1165         if (tx_queue_id >= eth_dev->data->nb_tx_queues) {
1166                 PMD_DRV_LOG(ERR, "TX queue id is out of range: tx_queue_id=%u, nb_tx_queues=%u\n",
1167                             tx_queue_id, eth_dev->data->nb_tx_queues);
1168                 return -EINVAL;
1169         }
1170
1171         /* allocate a queue object */
1172         txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct avp_queue),
1173                                  RTE_CACHE_LINE_SIZE, socket_id);
1174         if (txq == NULL) {
1175                 PMD_DRV_LOG(ERR, "Failed to allocate new Tx queue object\n");
1176                 return -ENOMEM;
1177         }
1178
1179         /* only the configured set of transmit queues are used */
1180         txq->queue_id = tx_queue_id;
1181         txq->queue_base = tx_queue_id;
1182         txq->queue_limit = tx_queue_id;
1183
1184         /* save back pointers to AVP and Ethernet devices */
1185         txq->avp = avp;
1186         txq->dev_data = eth_dev->data;
1187         eth_dev->data->tx_queues[tx_queue_id] = (void *)txq;
1188
1189         PMD_DRV_LOG(DEBUG, "Tx queue %u setup at %p\n", tx_queue_id, txq);
1190
1191         (void)nb_tx_desc;
1192         (void)tx_conf;
1193         return 0;
1194 }
1195
1196 static inline int
1197 _avp_cmp_ether_addr(struct rte_ether_addr *a, struct rte_ether_addr *b)
1198 {
1199         uint16_t *_a = (uint16_t *)&a->addr_bytes[0];
1200         uint16_t *_b = (uint16_t *)&b->addr_bytes[0];
1201         return (_a[0] ^ _b[0]) | (_a[1] ^ _b[1]) | (_a[2] ^ _b[2]);
1202 }
1203
1204 static inline int
1205 _avp_mac_filter(struct avp_dev *avp, struct rte_mbuf *m)
1206 {
1207         struct rte_ether_hdr *eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1208
1209         if (likely(_avp_cmp_ether_addr(&avp->ethaddr, &eth->dst_addr) == 0)) {
1210                 /* allow all packets destined to our address */
1211                 return 0;
1212         }
1213
1214         if (likely(rte_is_broadcast_ether_addr(&eth->dst_addr))) {
1215                 /* allow all broadcast packets */
1216                 return 0;
1217         }
1218
1219         if (likely(rte_is_multicast_ether_addr(&eth->dst_addr))) {
1220                 /* allow all multicast packets */
1221                 return 0;
1222         }
1223
1224         if (avp->flags & AVP_F_PROMISC) {
1225                 /* allow all packets when in promiscuous mode */
1226                 return 0;
1227         }
1228
1229         return -1;
1230 }
1231
1232 #ifdef RTE_LIBRTE_AVP_DEBUG_BUFFERS
1233 static inline void
1234 __avp_dev_buffer_sanity_check(struct avp_dev *avp, struct rte_avp_desc *buf)
1235 {
1236         struct rte_avp_desc *first_buf;
1237         struct rte_avp_desc *pkt_buf;
1238         unsigned int pkt_len;
1239         unsigned int nb_segs;
1240         void *pkt_data;
1241         unsigned int i;
1242
1243         first_buf = avp_dev_translate_buffer(avp, buf);
1244
1245         i = 0;
1246         pkt_len = 0;
1247         nb_segs = first_buf->nb_segs;
1248         do {
1249                 /* Adjust pointers for guest addressing */
1250                 pkt_buf = avp_dev_translate_buffer(avp, buf);
1251                 if (pkt_buf == NULL)
1252                         rte_panic("bad buffer: segment %u has an invalid address %p\n",
1253                                   i, buf);
1254                 pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1255                 if (pkt_data == NULL)
1256                         rte_panic("bad buffer: segment %u has a NULL data pointer\n",
1257                                   i);
1258                 if (pkt_buf->data_len == 0)
1259                         rte_panic("bad buffer: segment %u has 0 data length\n",
1260                                   i);
1261                 pkt_len += pkt_buf->data_len;
1262                 nb_segs--;
1263                 i++;
1264
1265         } while (nb_segs && (buf = pkt_buf->next) != NULL);
1266
1267         if (nb_segs != 0)
1268                 rte_panic("bad buffer: expected %u segments found %u\n",
1269                           first_buf->nb_segs, (first_buf->nb_segs - nb_segs));
1270         if (pkt_len != first_buf->pkt_len)
1271                 rte_panic("bad buffer: expected length %u found %u\n",
1272                           first_buf->pkt_len, pkt_len);
1273 }
1274
1275 #define avp_dev_buffer_sanity_check(a, b) \
1276         __avp_dev_buffer_sanity_check((a), (b))
1277
1278 #else /* RTE_LIBRTE_AVP_DEBUG_BUFFERS */
1279
1280 #define avp_dev_buffer_sanity_check(a, b) do {} while (0)
1281
1282 #endif
1283
1284 /*
1285  * Copy a host buffer chain to a set of mbufs.  This function assumes that
1286  * there exactly the required number of mbufs to copy all source bytes.
1287  */
1288 static inline struct rte_mbuf *
1289 avp_dev_copy_from_buffers(struct avp_dev *avp,
1290                           struct rte_avp_desc *buf,
1291                           struct rte_mbuf **mbufs,
1292                           unsigned int count)
1293 {
1294         struct rte_mbuf *m_previous = NULL;
1295         struct rte_avp_desc *pkt_buf;
1296         unsigned int total_length = 0;
1297         unsigned int copy_length;
1298         unsigned int src_offset;
1299         struct rte_mbuf *m;
1300         uint16_t ol_flags;
1301         uint16_t vlan_tci;
1302         void *pkt_data;
1303         unsigned int i;
1304
1305         avp_dev_buffer_sanity_check(avp, buf);
1306
1307         /* setup the first source buffer */
1308         pkt_buf = avp_dev_translate_buffer(avp, buf);
1309         pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1310         total_length = pkt_buf->pkt_len;
1311         src_offset = 0;
1312
1313         if (pkt_buf->ol_flags & RTE_AVP_RX_VLAN_PKT) {
1314                 ol_flags = RTE_MBUF_F_RX_VLAN;
1315                 vlan_tci = pkt_buf->vlan_tci;
1316         } else {
1317                 ol_flags = 0;
1318                 vlan_tci = 0;
1319         }
1320
1321         for (i = 0; (i < count) && (buf != NULL); i++) {
1322                 /* fill each destination buffer */
1323                 m = mbufs[i];
1324
1325                 if (m_previous != NULL)
1326                         m_previous->next = m;
1327
1328                 m_previous = m;
1329
1330                 do {
1331                         /*
1332                          * Copy as many source buffers as will fit in the
1333                          * destination buffer.
1334                          */
1335                         copy_length = RTE_MIN((avp->guest_mbuf_size -
1336                                                rte_pktmbuf_data_len(m)),
1337                                               (pkt_buf->data_len -
1338                                                src_offset));
1339                         rte_memcpy(RTE_PTR_ADD(rte_pktmbuf_mtod(m, void *),
1340                                                rte_pktmbuf_data_len(m)),
1341                                    RTE_PTR_ADD(pkt_data, src_offset),
1342                                    copy_length);
1343                         rte_pktmbuf_data_len(m) += copy_length;
1344                         src_offset += copy_length;
1345
1346                         if (likely(src_offset == pkt_buf->data_len)) {
1347                                 /* need a new source buffer */
1348                                 buf = pkt_buf->next;
1349                                 if (buf != NULL) {
1350                                         pkt_buf = avp_dev_translate_buffer(
1351                                                 avp, buf);
1352                                         pkt_data = avp_dev_translate_buffer(
1353                                                 avp, pkt_buf->data);
1354                                         src_offset = 0;
1355                                 }
1356                         }
1357
1358                         if (unlikely(rte_pktmbuf_data_len(m) ==
1359                                      avp->guest_mbuf_size)) {
1360                                 /* need a new destination mbuf */
1361                                 break;
1362                         }
1363
1364                 } while (buf != NULL);
1365         }
1366
1367         m = mbufs[0];
1368         m->ol_flags = ol_flags;
1369         m->nb_segs = count;
1370         rte_pktmbuf_pkt_len(m) = total_length;
1371         m->vlan_tci = vlan_tci;
1372
1373         __rte_mbuf_sanity_check(m, 1);
1374
1375         return m;
1376 }
1377
1378 static uint16_t
1379 avp_recv_scattered_pkts(void *rx_queue,
1380                         struct rte_mbuf **rx_pkts,
1381                         uint16_t nb_pkts)
1382 {
1383         struct avp_queue *rxq = (struct avp_queue *)rx_queue;
1384         struct rte_avp_desc *avp_bufs[AVP_MAX_RX_BURST];
1385         struct rte_mbuf *mbufs[RTE_AVP_MAX_MBUF_SEGMENTS];
1386         struct avp_dev *avp = rxq->avp;
1387         struct rte_avp_desc *pkt_buf;
1388         struct rte_avp_fifo *free_q;
1389         struct rte_avp_fifo *rx_q;
1390         struct rte_avp_desc *buf;
1391         unsigned int count, avail, n;
1392         unsigned int guest_mbuf_size;
1393         struct rte_mbuf *m;
1394         unsigned int required;
1395         unsigned int buf_len;
1396         unsigned int port_id;
1397         unsigned int i;
1398
1399         if (unlikely(avp->flags & AVP_F_DETACHED)) {
1400                 /* VM live migration in progress */
1401                 return 0;
1402         }
1403
1404         guest_mbuf_size = avp->guest_mbuf_size;
1405         port_id = avp->port_id;
1406         rx_q = avp->rx_q[rxq->queue_id];
1407         free_q = avp->free_q[rxq->queue_id];
1408
1409         /* setup next queue to service */
1410         rxq->queue_id = (rxq->queue_id < rxq->queue_limit) ?
1411                 (rxq->queue_id + 1) : rxq->queue_base;
1412
1413         /* determine how many slots are available in the free queue */
1414         count = avp_fifo_free_count(free_q);
1415
1416         /* determine how many packets are available in the rx queue */
1417         avail = avp_fifo_count(rx_q);
1418
1419         /* determine how many packets can be received */
1420         count = RTE_MIN(count, avail);
1421         count = RTE_MIN(count, nb_pkts);
1422         count = RTE_MIN(count, (unsigned int)AVP_MAX_RX_BURST);
1423
1424         if (unlikely(count == 0)) {
1425                 /* no free buffers, or no buffers on the rx queue */
1426                 return 0;
1427         }
1428
1429         /* retrieve pending packets */
1430         n = avp_fifo_get(rx_q, (void **)&avp_bufs, count);
1431         PMD_RX_LOG(DEBUG, "Receiving %u packets from Rx queue at %p\n",
1432                    count, rx_q);
1433
1434         count = 0;
1435         for (i = 0; i < n; i++) {
1436                 /* prefetch next entry while processing current one */
1437                 if (i + 1 < n) {
1438                         pkt_buf = avp_dev_translate_buffer(avp,
1439                                                            avp_bufs[i + 1]);
1440                         rte_prefetch0(pkt_buf);
1441                 }
1442                 buf = avp_bufs[i];
1443
1444                 /* Peek into the first buffer to determine the total length */
1445                 pkt_buf = avp_dev_translate_buffer(avp, buf);
1446                 buf_len = pkt_buf->pkt_len;
1447
1448                 /* Allocate enough mbufs to receive the entire packet */
1449                 required = (buf_len + guest_mbuf_size - 1) / guest_mbuf_size;
1450                 if (rte_pktmbuf_alloc_bulk(avp->pool, mbufs, required)) {
1451                         rxq->dev_data->rx_mbuf_alloc_failed++;
1452                         continue;
1453                 }
1454
1455                 /* Copy the data from the buffers to our mbufs */
1456                 m = avp_dev_copy_from_buffers(avp, buf, mbufs, required);
1457
1458                 /* finalize mbuf */
1459                 m->port = port_id;
1460
1461                 if (_avp_mac_filter(avp, m) != 0) {
1462                         /* silently discard packets not destined to our MAC */
1463                         rte_pktmbuf_free(m);
1464                         continue;
1465                 }
1466
1467                 /* return new mbuf to caller */
1468                 rx_pkts[count++] = m;
1469                 rxq->bytes += buf_len;
1470         }
1471
1472         rxq->packets += count;
1473
1474         /* return the buffers to the free queue */
1475         avp_fifo_put(free_q, (void **)&avp_bufs[0], n);
1476
1477         return count;
1478 }
1479
1480
1481 static uint16_t
1482 avp_recv_pkts(void *rx_queue,
1483               struct rte_mbuf **rx_pkts,
1484               uint16_t nb_pkts)
1485 {
1486         struct avp_queue *rxq = (struct avp_queue *)rx_queue;
1487         struct rte_avp_desc *avp_bufs[AVP_MAX_RX_BURST];
1488         struct avp_dev *avp = rxq->avp;
1489         struct rte_avp_desc *pkt_buf;
1490         struct rte_avp_fifo *free_q;
1491         struct rte_avp_fifo *rx_q;
1492         unsigned int count, avail, n;
1493         unsigned int pkt_len;
1494         struct rte_mbuf *m;
1495         char *pkt_data;
1496         unsigned int i;
1497
1498         if (unlikely(avp->flags & AVP_F_DETACHED)) {
1499                 /* VM live migration in progress */
1500                 return 0;
1501         }
1502
1503         rx_q = avp->rx_q[rxq->queue_id];
1504         free_q = avp->free_q[rxq->queue_id];
1505
1506         /* setup next queue to service */
1507         rxq->queue_id = (rxq->queue_id < rxq->queue_limit) ?
1508                 (rxq->queue_id + 1) : rxq->queue_base;
1509
1510         /* determine how many slots are available in the free queue */
1511         count = avp_fifo_free_count(free_q);
1512
1513         /* determine how many packets are available in the rx queue */
1514         avail = avp_fifo_count(rx_q);
1515
1516         /* determine how many packets can be received */
1517         count = RTE_MIN(count, avail);
1518         count = RTE_MIN(count, nb_pkts);
1519         count = RTE_MIN(count, (unsigned int)AVP_MAX_RX_BURST);
1520
1521         if (unlikely(count == 0)) {
1522                 /* no free buffers, or no buffers on the rx queue */
1523                 return 0;
1524         }
1525
1526         /* retrieve pending packets */
1527         n = avp_fifo_get(rx_q, (void **)&avp_bufs, count);
1528         PMD_RX_LOG(DEBUG, "Receiving %u packets from Rx queue at %p\n",
1529                    count, rx_q);
1530
1531         count = 0;
1532         for (i = 0; i < n; i++) {
1533                 /* prefetch next entry while processing current one */
1534                 if (i < n - 1) {
1535                         pkt_buf = avp_dev_translate_buffer(avp,
1536                                                            avp_bufs[i + 1]);
1537                         rte_prefetch0(pkt_buf);
1538                 }
1539
1540                 /* Adjust host pointers for guest addressing */
1541                 pkt_buf = avp_dev_translate_buffer(avp, avp_bufs[i]);
1542                 pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1543                 pkt_len = pkt_buf->pkt_len;
1544
1545                 if (unlikely((pkt_len > avp->guest_mbuf_size) ||
1546                              (pkt_buf->nb_segs > 1))) {
1547                         /*
1548                          * application should be using the scattered receive
1549                          * function
1550                          */
1551                         rxq->errors++;
1552                         continue;
1553                 }
1554
1555                 /* process each packet to be transmitted */
1556                 m = rte_pktmbuf_alloc(avp->pool);
1557                 if (unlikely(m == NULL)) {
1558                         rxq->dev_data->rx_mbuf_alloc_failed++;
1559                         continue;
1560                 }
1561
1562                 /* copy data out of the host buffer to our buffer */
1563                 m->data_off = RTE_PKTMBUF_HEADROOM;
1564                 rte_memcpy(rte_pktmbuf_mtod(m, void *), pkt_data, pkt_len);
1565
1566                 /* initialize the local mbuf */
1567                 rte_pktmbuf_data_len(m) = pkt_len;
1568                 rte_pktmbuf_pkt_len(m) = pkt_len;
1569                 m->port = avp->port_id;
1570
1571                 if (pkt_buf->ol_flags & RTE_AVP_RX_VLAN_PKT) {
1572                         m->ol_flags = RTE_MBUF_F_RX_VLAN;
1573                         m->vlan_tci = pkt_buf->vlan_tci;
1574                 }
1575
1576                 if (_avp_mac_filter(avp, m) != 0) {
1577                         /* silently discard packets not destined to our MAC */
1578                         rte_pktmbuf_free(m);
1579                         continue;
1580                 }
1581
1582                 /* return new mbuf to caller */
1583                 rx_pkts[count++] = m;
1584                 rxq->bytes += pkt_len;
1585         }
1586
1587         rxq->packets += count;
1588
1589         /* return the buffers to the free queue */
1590         avp_fifo_put(free_q, (void **)&avp_bufs[0], n);
1591
1592         return count;
1593 }
1594
1595 /*
1596  * Copy a chained mbuf to a set of host buffers.  This function assumes that
1597  * there are sufficient destination buffers to contain the entire source
1598  * packet.
1599  */
1600 static inline uint16_t
1601 avp_dev_copy_to_buffers(struct avp_dev *avp,
1602                         struct rte_mbuf *mbuf,
1603                         struct rte_avp_desc **buffers,
1604                         unsigned int count)
1605 {
1606         struct rte_avp_desc *previous_buf = NULL;
1607         struct rte_avp_desc *first_buf = NULL;
1608         struct rte_avp_desc *pkt_buf;
1609         struct rte_avp_desc *buf;
1610         size_t total_length;
1611         struct rte_mbuf *m;
1612         size_t copy_length;
1613         size_t src_offset;
1614         char *pkt_data;
1615         unsigned int i;
1616
1617         __rte_mbuf_sanity_check(mbuf, 1);
1618
1619         m = mbuf;
1620         src_offset = 0;
1621         total_length = rte_pktmbuf_pkt_len(m);
1622         for (i = 0; (i < count) && (m != NULL); i++) {
1623                 /* fill each destination buffer */
1624                 buf = buffers[i];
1625
1626                 if (i < count - 1) {
1627                         /* prefetch next entry while processing this one */
1628                         pkt_buf = avp_dev_translate_buffer(avp, buffers[i + 1]);
1629                         rte_prefetch0(pkt_buf);
1630                 }
1631
1632                 /* Adjust pointers for guest addressing */
1633                 pkt_buf = avp_dev_translate_buffer(avp, buf);
1634                 pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1635
1636                 /* setup the buffer chain */
1637                 if (previous_buf != NULL)
1638                         previous_buf->next = buf;
1639                 else
1640                         first_buf = pkt_buf;
1641
1642                 previous_buf = pkt_buf;
1643
1644                 do {
1645                         /*
1646                          * copy as many source mbuf segments as will fit in the
1647                          * destination buffer.
1648                          */
1649                         copy_length = RTE_MIN((avp->host_mbuf_size -
1650                                                pkt_buf->data_len),
1651                                               (rte_pktmbuf_data_len(m) -
1652                                                src_offset));
1653                         rte_memcpy(RTE_PTR_ADD(pkt_data, pkt_buf->data_len),
1654                                    RTE_PTR_ADD(rte_pktmbuf_mtod(m, void *),
1655                                                src_offset),
1656                                    copy_length);
1657                         pkt_buf->data_len += copy_length;
1658                         src_offset += copy_length;
1659
1660                         if (likely(src_offset == rte_pktmbuf_data_len(m))) {
1661                                 /* need a new source buffer */
1662                                 m = m->next;
1663                                 src_offset = 0;
1664                         }
1665
1666                         if (unlikely(pkt_buf->data_len ==
1667                                      avp->host_mbuf_size)) {
1668                                 /* need a new destination buffer */
1669                                 break;
1670                         }
1671
1672                 } while (m != NULL);
1673         }
1674
1675         first_buf->nb_segs = count;
1676         first_buf->pkt_len = total_length;
1677
1678         if (mbuf->ol_flags & RTE_MBUF_F_TX_VLAN) {
1679                 first_buf->ol_flags |= RTE_AVP_TX_VLAN_PKT;
1680                 first_buf->vlan_tci = mbuf->vlan_tci;
1681         }
1682
1683         avp_dev_buffer_sanity_check(avp, buffers[0]);
1684
1685         return total_length;
1686 }
1687
1688
1689 static uint16_t
1690 avp_xmit_scattered_pkts(void *tx_queue,
1691                         struct rte_mbuf **tx_pkts,
1692                         uint16_t nb_pkts)
1693 {
1694         struct rte_avp_desc *avp_bufs[(AVP_MAX_TX_BURST *
1695                                        RTE_AVP_MAX_MBUF_SEGMENTS)] = {};
1696         struct avp_queue *txq = (struct avp_queue *)tx_queue;
1697         struct rte_avp_desc *tx_bufs[AVP_MAX_TX_BURST];
1698         struct avp_dev *avp = txq->avp;
1699         struct rte_avp_fifo *alloc_q;
1700         struct rte_avp_fifo *tx_q;
1701         unsigned int count, avail, n;
1702         unsigned int orig_nb_pkts;
1703         struct rte_mbuf *m;
1704         unsigned int required;
1705         unsigned int segments;
1706         unsigned int tx_bytes;
1707         unsigned int i;
1708
1709         orig_nb_pkts = nb_pkts;
1710         if (unlikely(avp->flags & AVP_F_DETACHED)) {
1711                 /* VM live migration in progress */
1712                 /* TODO ... buffer for X packets then drop? */
1713                 txq->errors += nb_pkts;
1714                 return 0;
1715         }
1716
1717         tx_q = avp->tx_q[txq->queue_id];
1718         alloc_q = avp->alloc_q[txq->queue_id];
1719
1720         /* limit the number of transmitted packets to the max burst size */
1721         if (unlikely(nb_pkts > AVP_MAX_TX_BURST))
1722                 nb_pkts = AVP_MAX_TX_BURST;
1723
1724         /* determine how many buffers are available to copy into */
1725         avail = avp_fifo_count(alloc_q);
1726         if (unlikely(avail > (AVP_MAX_TX_BURST *
1727                               RTE_AVP_MAX_MBUF_SEGMENTS)))
1728                 avail = AVP_MAX_TX_BURST * RTE_AVP_MAX_MBUF_SEGMENTS;
1729
1730         /* determine how many slots are available in the transmit queue */
1731         count = avp_fifo_free_count(tx_q);
1732
1733         /* determine how many packets can be sent */
1734         nb_pkts = RTE_MIN(count, nb_pkts);
1735
1736         /* determine how many packets will fit in the available buffers */
1737         count = 0;
1738         segments = 0;
1739         for (i = 0; i < nb_pkts; i++) {
1740                 m = tx_pkts[i];
1741                 if (likely(i < (unsigned int)nb_pkts - 1)) {
1742                         /* prefetch next entry while processing this one */
1743                         rte_prefetch0(tx_pkts[i + 1]);
1744                 }
1745                 required = (rte_pktmbuf_pkt_len(m) + avp->host_mbuf_size - 1) /
1746                         avp->host_mbuf_size;
1747
1748                 if (unlikely((required == 0) ||
1749                              (required > RTE_AVP_MAX_MBUF_SEGMENTS)))
1750                         break;
1751                 else if (unlikely(required + segments > avail))
1752                         break;
1753                 segments += required;
1754                 count++;
1755         }
1756         nb_pkts = count;
1757
1758         if (unlikely(nb_pkts == 0)) {
1759                 /* no available buffers, or no space on the tx queue */
1760                 txq->errors += orig_nb_pkts;
1761                 return 0;
1762         }
1763
1764         PMD_TX_LOG(DEBUG, "Sending %u packets on Tx queue at %p\n",
1765                    nb_pkts, tx_q);
1766
1767         /* retrieve sufficient send buffers */
1768         n = avp_fifo_get(alloc_q, (void **)&avp_bufs, segments);
1769         if (unlikely(n != segments)) {
1770                 PMD_TX_LOG(DEBUG, "Failed to allocate buffers "
1771                            "n=%u, segments=%u, orig=%u\n",
1772                            n, segments, orig_nb_pkts);
1773                 txq->errors += orig_nb_pkts;
1774                 return 0;
1775         }
1776
1777         tx_bytes = 0;
1778         count = 0;
1779         for (i = 0; i < nb_pkts; i++) {
1780                 /* process each packet to be transmitted */
1781                 m = tx_pkts[i];
1782
1783                 /* determine how many buffers are required for this packet */
1784                 required = (rte_pktmbuf_pkt_len(m) + avp->host_mbuf_size - 1) /
1785                         avp->host_mbuf_size;
1786
1787                 tx_bytes += avp_dev_copy_to_buffers(avp, m,
1788                                                     &avp_bufs[count], required);
1789                 tx_bufs[i] = avp_bufs[count];
1790                 count += required;
1791
1792                 /* free the original mbuf */
1793                 rte_pktmbuf_free(m);
1794         }
1795
1796         txq->packets += nb_pkts;
1797         txq->bytes += tx_bytes;
1798
1799 #ifdef RTE_LIBRTE_AVP_DEBUG_BUFFERS
1800         for (i = 0; i < nb_pkts; i++)
1801                 avp_dev_buffer_sanity_check(avp, tx_bufs[i]);
1802 #endif
1803
1804         /* send the packets */
1805         n = avp_fifo_put(tx_q, (void **)&tx_bufs[0], nb_pkts);
1806         if (unlikely(n != orig_nb_pkts))
1807                 txq->errors += (orig_nb_pkts - n);
1808
1809         return n;
1810 }
1811
1812
1813 static uint16_t
1814 avp_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1815 {
1816         struct avp_queue *txq = (struct avp_queue *)tx_queue;
1817         struct rte_avp_desc *avp_bufs[AVP_MAX_TX_BURST];
1818         struct avp_dev *avp = txq->avp;
1819         struct rte_avp_desc *pkt_buf;
1820         struct rte_avp_fifo *alloc_q;
1821         struct rte_avp_fifo *tx_q;
1822         unsigned int count, avail, n;
1823         struct rte_mbuf *m;
1824         unsigned int pkt_len;
1825         unsigned int tx_bytes;
1826         char *pkt_data;
1827         unsigned int i;
1828
1829         if (unlikely(avp->flags & AVP_F_DETACHED)) {
1830                 /* VM live migration in progress */
1831                 /* TODO ... buffer for X packets then drop?! */
1832                 txq->errors++;
1833                 return 0;
1834         }
1835
1836         tx_q = avp->tx_q[txq->queue_id];
1837         alloc_q = avp->alloc_q[txq->queue_id];
1838
1839         /* limit the number of transmitted packets to the max burst size */
1840         if (unlikely(nb_pkts > AVP_MAX_TX_BURST))
1841                 nb_pkts = AVP_MAX_TX_BURST;
1842
1843         /* determine how many buffers are available to copy into */
1844         avail = avp_fifo_count(alloc_q);
1845
1846         /* determine how many slots are available in the transmit queue */
1847         count = avp_fifo_free_count(tx_q);
1848
1849         /* determine how many packets can be sent */
1850         count = RTE_MIN(count, avail);
1851         count = RTE_MIN(count, nb_pkts);
1852
1853         if (unlikely(count == 0)) {
1854                 /* no available buffers, or no space on the tx queue */
1855                 txq->errors += nb_pkts;
1856                 return 0;
1857         }
1858
1859         PMD_TX_LOG(DEBUG, "Sending %u packets on Tx queue at %p\n",
1860                    count, tx_q);
1861
1862         /* retrieve sufficient send buffers */
1863         n = avp_fifo_get(alloc_q, (void **)&avp_bufs, count);
1864         if (unlikely(n != count)) {
1865                 txq->errors++;
1866                 return 0;
1867         }
1868
1869         tx_bytes = 0;
1870         for (i = 0; i < count; i++) {
1871                 /* prefetch next entry while processing the current one */
1872                 if (i < count - 1) {
1873                         pkt_buf = avp_dev_translate_buffer(avp,
1874                                                            avp_bufs[i + 1]);
1875                         rte_prefetch0(pkt_buf);
1876                 }
1877
1878                 /* process each packet to be transmitted */
1879                 m = tx_pkts[i];
1880
1881                 /* Adjust pointers for guest addressing */
1882                 pkt_buf = avp_dev_translate_buffer(avp, avp_bufs[i]);
1883                 pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1884                 pkt_len = rte_pktmbuf_pkt_len(m);
1885
1886                 if (unlikely((pkt_len > avp->guest_mbuf_size) ||
1887                                          (pkt_len > avp->host_mbuf_size))) {
1888                         /*
1889                          * application should be using the scattered transmit
1890                          * function; send it truncated to avoid the performance
1891                          * hit of having to manage returning the already
1892                          * allocated buffer to the free list.  This should not
1893                          * happen since the application should have not send
1894                          * packages larger than its MTU and it should be
1895                          * policing its own packet sizes.
1896                          */
1897                         txq->errors++;
1898                         pkt_len = RTE_MIN(avp->guest_mbuf_size,
1899                                           avp->host_mbuf_size);
1900                 }
1901
1902                 /* copy data out of our mbuf and into the AVP buffer */
1903                 rte_memcpy(pkt_data, rte_pktmbuf_mtod(m, void *), pkt_len);
1904                 pkt_buf->pkt_len = pkt_len;
1905                 pkt_buf->data_len = pkt_len;
1906                 pkt_buf->nb_segs = 1;
1907                 pkt_buf->next = NULL;
1908
1909                 if (m->ol_flags & RTE_MBUF_F_TX_VLAN) {
1910                         pkt_buf->ol_flags |= RTE_AVP_TX_VLAN_PKT;
1911                         pkt_buf->vlan_tci = m->vlan_tci;
1912                 }
1913
1914                 tx_bytes += pkt_len;
1915
1916                 /* free the original mbuf */
1917                 rte_pktmbuf_free(m);
1918         }
1919
1920         txq->packets += count;
1921         txq->bytes += tx_bytes;
1922
1923         /* send the packets */
1924         n = avp_fifo_put(tx_q, (void **)&avp_bufs[0], count);
1925
1926         return n;
1927 }
1928
1929 static void
1930 avp_dev_rx_queue_release(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
1931 {
1932         if (eth_dev->data->rx_queues[rx_queue_id] != NULL) {
1933                 rte_free(eth_dev->data->rx_queues[rx_queue_id]);
1934                 eth_dev->data->rx_queues[rx_queue_id] = NULL;
1935         }
1936 }
1937
1938 static void
1939 avp_dev_rx_queue_release_all(struct rte_eth_dev *eth_dev)
1940 {
1941         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1942         struct rte_eth_dev_data *data = avp->dev_data;
1943         unsigned int i;
1944
1945         for (i = 0; i < avp->num_rx_queues; i++) {
1946                 if (data->rx_queues[i]) {
1947                         rte_free(data->rx_queues[i]);
1948                         data->rx_queues[i] = NULL;
1949                 }
1950         }
1951 }
1952
1953 static void
1954 avp_dev_tx_queue_release(struct rte_eth_dev *eth_dev, uint16_t tx_queue_id)
1955 {
1956         if (eth_dev->data->tx_queues[tx_queue_id] != NULL) {
1957                 rte_free(eth_dev->data->tx_queues[tx_queue_id]);
1958                 eth_dev->data->tx_queues[tx_queue_id] = NULL;
1959         }
1960 }
1961
1962 static void
1963 avp_dev_tx_queue_release_all(struct rte_eth_dev *eth_dev)
1964 {
1965         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1966         struct rte_eth_dev_data *data = avp->dev_data;
1967         unsigned int i;
1968
1969         for (i = 0; i < avp->num_tx_queues; i++) {
1970                 if (data->tx_queues[i]) {
1971                         rte_free(data->tx_queues[i]);
1972                         data->tx_queues[i] = NULL;
1973                 }
1974         }
1975 }
1976
1977 static int
1978 avp_dev_configure(struct rte_eth_dev *eth_dev)
1979 {
1980         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1981         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1982         struct rte_avp_device_info *host_info;
1983         struct rte_avp_device_config config;
1984         int mask = 0;
1985         void *addr;
1986         int ret;
1987
1988         rte_spinlock_lock(&avp->lock);
1989         if (avp->flags & AVP_F_DETACHED) {
1990                 PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
1991                 ret = -ENOTSUP;
1992                 goto unlock;
1993         }
1994
1995         addr = pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR].addr;
1996         host_info = (struct rte_avp_device_info *)addr;
1997
1998         /* Setup required number of queues */
1999         _avp_set_queue_counts(eth_dev);
2000
2001         mask = (RTE_ETH_VLAN_STRIP_MASK |
2002                 RTE_ETH_VLAN_FILTER_MASK |
2003                 RTE_ETH_VLAN_EXTEND_MASK);
2004         ret = avp_vlan_offload_set(eth_dev, mask);
2005         if (ret < 0) {
2006                 PMD_DRV_LOG(ERR, "VLAN offload set failed by host, ret=%d\n",
2007                             ret);
2008                 goto unlock;
2009         }
2010
2011         /* update device config */
2012         memset(&config, 0, sizeof(config));
2013         config.device_id = host_info->device_id;
2014         config.driver_type = RTE_AVP_DRIVER_TYPE_DPDK;
2015         config.driver_version = AVP_DPDK_DRIVER_VERSION;
2016         config.features = avp->features;
2017         config.num_tx_queues = avp->num_tx_queues;
2018         config.num_rx_queues = avp->num_rx_queues;
2019
2020         ret = avp_dev_ctrl_set_config(eth_dev, &config);
2021         if (ret < 0) {
2022                 PMD_DRV_LOG(ERR, "Config request failed by host, ret=%d\n",
2023                             ret);
2024                 goto unlock;
2025         }
2026
2027         avp->flags |= AVP_F_CONFIGURED;
2028         ret = 0;
2029
2030 unlock:
2031         rte_spinlock_unlock(&avp->lock);
2032         return ret;
2033 }
2034
2035 static int
2036 avp_dev_start(struct rte_eth_dev *eth_dev)
2037 {
2038         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2039         int ret;
2040
2041         rte_spinlock_lock(&avp->lock);
2042         if (avp->flags & AVP_F_DETACHED) {
2043                 PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
2044                 ret = -ENOTSUP;
2045                 goto unlock;
2046         }
2047
2048         /* update link state */
2049         ret = avp_dev_ctrl_set_link_state(eth_dev, 1);
2050         if (ret < 0) {
2051                 PMD_DRV_LOG(ERR, "Link state change failed by host, ret=%d\n",
2052                             ret);
2053                 goto unlock;
2054         }
2055
2056         /* remember current link state */
2057         avp->flags |= AVP_F_LINKUP;
2058
2059         ret = 0;
2060
2061 unlock:
2062         rte_spinlock_unlock(&avp->lock);
2063         return ret;
2064 }
2065
2066 static int
2067 avp_dev_stop(struct rte_eth_dev *eth_dev)
2068 {
2069         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2070         int ret;
2071
2072         rte_spinlock_lock(&avp->lock);
2073         if (avp->flags & AVP_F_DETACHED) {
2074                 PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
2075                 ret = -ENOTSUP;
2076                 goto unlock;
2077         }
2078
2079         /* remember current link state */
2080         avp->flags &= ~AVP_F_LINKUP;
2081
2082         /* update link state */
2083         ret = avp_dev_ctrl_set_link_state(eth_dev, 0);
2084         if (ret < 0) {
2085                 PMD_DRV_LOG(ERR, "Link state change failed by host, ret=%d\n",
2086                             ret);
2087         }
2088
2089 unlock:
2090         rte_spinlock_unlock(&avp->lock);
2091         return ret;
2092 }
2093
2094 static int
2095 avp_dev_close(struct rte_eth_dev *eth_dev)
2096 {
2097         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2098         int ret;
2099
2100         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
2101                 return 0;
2102
2103         rte_spinlock_lock(&avp->lock);
2104         if (avp->flags & AVP_F_DETACHED) {
2105                 PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
2106                 goto unlock;
2107         }
2108
2109         /* remember current link state */
2110         avp->flags &= ~AVP_F_LINKUP;
2111         avp->flags &= ~AVP_F_CONFIGURED;
2112
2113         ret = avp_dev_disable_interrupts(eth_dev);
2114         if (ret < 0) {
2115                 PMD_DRV_LOG(ERR, "Failed to disable interrupts\n");
2116                 /* continue */
2117         }
2118
2119         /* update device state */
2120         ret = avp_dev_ctrl_shutdown(eth_dev);
2121         if (ret < 0) {
2122                 PMD_DRV_LOG(ERR, "Device shutdown failed by host, ret=%d\n",
2123                             ret);
2124                 /* continue */
2125         }
2126
2127         /* release dynamic storage for rx/tx queues */
2128         avp_dev_rx_queue_release_all(eth_dev);
2129         avp_dev_tx_queue_release_all(eth_dev);
2130
2131 unlock:
2132         rte_spinlock_unlock(&avp->lock);
2133         return 0;
2134 }
2135
2136 static int
2137 avp_dev_link_update(struct rte_eth_dev *eth_dev,
2138                                         __rte_unused int wait_to_complete)
2139 {
2140         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2141         struct rte_eth_link *link = &eth_dev->data->dev_link;
2142
2143         link->link_speed = RTE_ETH_SPEED_NUM_10G;
2144         link->link_duplex = RTE_ETH_LINK_FULL_DUPLEX;
2145         link->link_status = !!(avp->flags & AVP_F_LINKUP);
2146
2147         return -1;
2148 }
2149
2150 static int
2151 avp_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
2152 {
2153         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2154
2155         rte_spinlock_lock(&avp->lock);
2156         if ((avp->flags & AVP_F_PROMISC) == 0) {
2157                 avp->flags |= AVP_F_PROMISC;
2158                 PMD_DRV_LOG(DEBUG, "Promiscuous mode enabled on %u\n",
2159                             eth_dev->data->port_id);
2160         }
2161         rte_spinlock_unlock(&avp->lock);
2162
2163         return 0;
2164 }
2165
2166 static int
2167 avp_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
2168 {
2169         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2170
2171         rte_spinlock_lock(&avp->lock);
2172         if ((avp->flags & AVP_F_PROMISC) != 0) {
2173                 avp->flags &= ~AVP_F_PROMISC;
2174                 PMD_DRV_LOG(DEBUG, "Promiscuous mode disabled on %u\n",
2175                             eth_dev->data->port_id);
2176         }
2177         rte_spinlock_unlock(&avp->lock);
2178
2179         return 0;
2180 }
2181
2182 static int
2183 avp_dev_info_get(struct rte_eth_dev *eth_dev,
2184                  struct rte_eth_dev_info *dev_info)
2185 {
2186         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2187
2188         dev_info->max_rx_queues = avp->max_rx_queues;
2189         dev_info->max_tx_queues = avp->max_tx_queues;
2190         dev_info->min_rx_bufsize = AVP_MIN_RX_BUFSIZE;
2191         dev_info->max_rx_pktlen = avp->max_rx_pkt_len;
2192         dev_info->max_mac_addrs = AVP_MAX_MAC_ADDRS;
2193         if (avp->host_features & RTE_AVP_FEATURE_VLAN_OFFLOAD) {
2194                 dev_info->rx_offload_capa = RTE_ETH_RX_OFFLOAD_VLAN_STRIP;
2195                 dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_VLAN_INSERT;
2196         }
2197
2198         return 0;
2199 }
2200
2201 static int
2202 avp_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
2203 {
2204         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2205         struct rte_eth_conf *dev_conf = &eth_dev->data->dev_conf;
2206         uint64_t offloads = dev_conf->rxmode.offloads;
2207
2208         if (mask & RTE_ETH_VLAN_STRIP_MASK) {
2209                 if (avp->host_features & RTE_AVP_FEATURE_VLAN_OFFLOAD) {
2210                         if (offloads & RTE_ETH_RX_OFFLOAD_VLAN_STRIP)
2211                                 avp->features |= RTE_AVP_FEATURE_VLAN_OFFLOAD;
2212                         else
2213                                 avp->features &= ~RTE_AVP_FEATURE_VLAN_OFFLOAD;
2214                 } else {
2215                         PMD_DRV_LOG(ERR, "VLAN strip offload not supported\n");
2216                 }
2217         }
2218
2219         if (mask & RTE_ETH_VLAN_FILTER_MASK) {
2220                 if (offloads & RTE_ETH_RX_OFFLOAD_VLAN_FILTER)
2221                         PMD_DRV_LOG(ERR, "VLAN filter offload not supported\n");
2222         }
2223
2224         if (mask & RTE_ETH_VLAN_EXTEND_MASK) {
2225                 if (offloads & RTE_ETH_RX_OFFLOAD_VLAN_EXTEND)
2226                         PMD_DRV_LOG(ERR, "VLAN extend offload not supported\n");
2227         }
2228
2229         return 0;
2230 }
2231
2232 static int
2233 avp_dev_stats_get(struct rte_eth_dev *eth_dev, struct rte_eth_stats *stats)
2234 {
2235         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2236         unsigned int i;
2237
2238         for (i = 0; i < avp->num_rx_queues; i++) {
2239                 struct avp_queue *rxq = avp->dev_data->rx_queues[i];
2240
2241                 if (rxq) {
2242                         stats->ipackets += rxq->packets;
2243                         stats->ibytes += rxq->bytes;
2244                         stats->ierrors += rxq->errors;
2245
2246                         stats->q_ipackets[i] += rxq->packets;
2247                         stats->q_ibytes[i] += rxq->bytes;
2248                         stats->q_errors[i] += rxq->errors;
2249                 }
2250         }
2251
2252         for (i = 0; i < avp->num_tx_queues; i++) {
2253                 struct avp_queue *txq = avp->dev_data->tx_queues[i];
2254
2255                 if (txq) {
2256                         stats->opackets += txq->packets;
2257                         stats->obytes += txq->bytes;
2258                         stats->oerrors += txq->errors;
2259
2260                         stats->q_opackets[i] += txq->packets;
2261                         stats->q_obytes[i] += txq->bytes;
2262                 }
2263         }
2264
2265         return 0;
2266 }
2267
2268 static int
2269 avp_dev_stats_reset(struct rte_eth_dev *eth_dev)
2270 {
2271         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2272         unsigned int i;
2273
2274         for (i = 0; i < avp->num_rx_queues; i++) {
2275                 struct avp_queue *rxq = avp->dev_data->rx_queues[i];
2276
2277                 if (rxq) {
2278                         rxq->bytes = 0;
2279                         rxq->packets = 0;
2280                         rxq->errors = 0;
2281                 }
2282         }
2283
2284         for (i = 0; i < avp->num_tx_queues; i++) {
2285                 struct avp_queue *txq = avp->dev_data->tx_queues[i];
2286
2287                 if (txq) {
2288                         txq->bytes = 0;
2289                         txq->packets = 0;
2290                         txq->errors = 0;
2291                 }
2292         }
2293
2294         return 0;
2295 }
2296
2297 RTE_PMD_REGISTER_PCI(net_avp, rte_avp_pmd);
2298 RTE_PMD_REGISTER_PCI_TABLE(net_avp, pci_id_avp_map);
2299 RTE_LOG_REGISTER_SUFFIX(avp_logtype_driver, driver, NOTICE);