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 <rte_ethdev_driver.h>
  12 #include <rte_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 int avp_logtype_driver;
  36
  37 static int avp_dev_create(struct rte_pci_device *pci_dev,
  38                           struct rte_eth_dev *eth_dev);
  39
  40 static int avp_dev_configure(struct rte_eth_dev *dev);
  41 static int avp_dev_start(struct rte_eth_dev *dev);
  42 static void avp_dev_stop(struct rte_eth_dev *dev);
  43 static void avp_dev_close(struct rte_eth_dev *dev);
  44 static void avp_dev_info_get(struct rte_eth_dev *dev,
  45                              struct rte_eth_dev_info *dev_info);
  46 static int avp_vlan_offload_set(struct rte_eth_dev *dev, int mask);
  47 static int avp_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete);
  48 static void avp_dev_promiscuous_enable(struct rte_eth_dev *dev);
  49 static void avp_dev_promiscuous_disable(struct rte_eth_dev *dev);
  50
  51 static int avp_dev_rx_queue_setup(struct rte_eth_dev *dev,
  52                                   uint16_t rx_queue_id,
  53                                   uint16_t nb_rx_desc,
  54                                   unsigned int socket_id,
  55                                   const struct rte_eth_rxconf *rx_conf,
  56                                   struct rte_mempool *pool);
  57
  58 static int avp_dev_tx_queue_setup(struct rte_eth_dev *dev,
  59                                   uint16_t tx_queue_id,
  60                                   uint16_t nb_tx_desc,
  61                                   unsigned int socket_id,
  62                                   const struct rte_eth_txconf *tx_conf);
  63
  64 static uint16_t avp_recv_scattered_pkts(void *rx_queue,
  65                                         struct rte_mbuf **rx_pkts,
  66                                         uint16_t nb_pkts);
  67
  68 static uint16_t avp_recv_pkts(void *rx_queue,
  69                               struct rte_mbuf **rx_pkts,
  70                               uint16_t nb_pkts);
  71
  72 static uint16_t avp_xmit_scattered_pkts(void *tx_queue,
  73                                         struct rte_mbuf **tx_pkts,
  74                                         uint16_t nb_pkts);
  75
  76 static uint16_t avp_xmit_pkts(void *tx_queue,
  77                               struct rte_mbuf **tx_pkts,
  78                               uint16_t nb_pkts);
  79
  80 static void avp_dev_rx_queue_release(void *rxq);
  81 static void avp_dev_tx_queue_release(void *txq);
  82
  83 static int avp_dev_stats_get(struct rte_eth_dev *dev,
  84                               struct rte_eth_stats *stats);
  85 static void avp_dev_stats_reset(struct rte_eth_dev *dev);
  86
  87
  88 #define AVP_MAX_RX_BURST 64
  89 #define AVP_MAX_TX_BURST 64
  90 #define AVP_MAX_MAC_ADDRS 1
  91 #define AVP_MIN_RX_BUFSIZE ETHER_MIN_LEN
  92
  93
  94 /*
  95  * Defines the number of microseconds to wait before checking the response
  96  * queue for completion.
  97  */
  98 #define AVP_REQUEST_DELAY_USECS (5000)
  99
 100 /*
 101  * Defines the number times to check the response queue for completion before
 102  * declaring a timeout.
 103  */
 104 #define AVP_MAX_REQUEST_RETRY (100)
 105
 106 /* Defines the current PCI driver version number */
 107 #define AVP_DPDK_DRIVER_VERSION RTE_AVP_CURRENT_GUEST_VERSION
 108
 109 /*
 110  * The set of PCI devices this driver supports
 111  */
 112 static const struct rte_pci_id pci_id_avp_map[] = {
 113         { .vendor_id = RTE_AVP_PCI_VENDOR_ID,
 114           .device_id = RTE_AVP_PCI_DEVICE_ID,
 115           .subsystem_vendor_id = RTE_AVP_PCI_SUB_VENDOR_ID,
 116           .subsystem_device_id = RTE_AVP_PCI_SUB_DEVICE_ID,
 117           .class_id = RTE_CLASS_ANY_ID,
 118         },
 119
 120         { .vendor_id = 0, /* sentinel */
 121         },
 122 };
 123
 124 /*
 125  * dev_ops for avp, bare necessities for basic operation
 126  */
 127 static const struct eth_dev_ops avp_eth_dev_ops = {
 128         .dev_configure       = avp_dev_configure,
 129         .dev_start           = avp_dev_start,
 130         .dev_stop            = avp_dev_stop,
 131         .dev_close           = avp_dev_close,
 132         .dev_infos_get       = avp_dev_info_get,
 133         .vlan_offload_set    = avp_vlan_offload_set,
 134         .stats_get           = avp_dev_stats_get,
 135         .stats_reset         = avp_dev_stats_reset,
 136         .link_update         = avp_dev_link_update,
 137         .promiscuous_enable  = avp_dev_promiscuous_enable,
 138         .promiscuous_disable = avp_dev_promiscuous_disable,
 139         .rx_queue_setup      = avp_dev_rx_queue_setup,
 140         .rx_queue_release    = avp_dev_rx_queue_release,
 141         .tx_queue_setup      = avp_dev_tx_queue_setup,
 142         .tx_queue_release    = avp_dev_tx_queue_release,
 143 };
 144
 145 /**@{ AVP device flags */
 146 #define AVP_F_PROMISC (1 << 1)
 147 #define AVP_F_CONFIGURED (1 << 2)
 148 #define AVP_F_LINKUP (1 << 3)
 149 #define AVP_F_DETACHED (1 << 4)
 150 /**@} */
 151
 152 /* Ethernet device validation marker */
 153 #define AVP_ETHDEV_MAGIC 0x92972862
 154
 155 /*
 156  * Defines the AVP device attributes which are attached to an RTE ethernet
 157  * device
 158  */
 159 struct avp_dev {
 160         uint32_t magic; /**< Memory validation marker */
 161         uint64_t device_id; /**< Unique system identifier */
 162         struct ether_addr ethaddr; /**< Host specified MAC address */
 163         struct rte_eth_dev_data *dev_data;
 164         /**< Back pointer to ethernet device data */
 165         volatile uint32_t flags; /**< Device operational flags */
 166         uint16_t port_id; /**< Ethernet port identifier */
 167         struct rte_mempool *pool; /**< pkt mbuf mempool */
 168         unsigned int guest_mbuf_size; /**< local pool mbuf size */
 169         unsigned int host_mbuf_size; /**< host mbuf size */
 170         unsigned int max_rx_pkt_len; /**< maximum receive unit */
 171         uint32_t host_features; /**< Supported feature bitmap */
 172         uint32_t features; /**< Enabled feature bitmap */
 173         unsigned int num_tx_queues; /**< Negotiated number of transmit queues */
 174         unsigned int max_tx_queues; /**< Maximum number of transmit queues */
 175         unsigned int num_rx_queues; /**< Negotiated number of receive queues */
 176         unsigned int max_rx_queues; /**< Maximum number of receive queues */
 177
 178         struct rte_avp_fifo *tx_q[RTE_AVP_MAX_QUEUES]; /**< TX queue */
 179         struct rte_avp_fifo *rx_q[RTE_AVP_MAX_QUEUES]; /**< RX queue */
 180         struct rte_avp_fifo *alloc_q[RTE_AVP_MAX_QUEUES];
 181         /**< Allocated mbufs queue */
 182         struct rte_avp_fifo *free_q[RTE_AVP_MAX_QUEUES];
 183         /**< To be freed mbufs queue */
 184
 185         /* mutual exclusion over the 'flag' and 'resp_q/req_q' fields */
 186         rte_spinlock_t lock;
 187
 188         /* For request & response */
 189         struct rte_avp_fifo *req_q; /**< Request queue */
 190         struct rte_avp_fifo *resp_q; /**< Response queue */
 191         void *host_sync_addr; /**< (host) Req/Resp Mem address */
 192         void *sync_addr; /**< Req/Resp Mem address */
 193         void *host_mbuf_addr; /**< (host) MBUF pool start address */
 194         void *mbuf_addr; /**< MBUF pool start address */
 195 } __rte_cache_aligned;
 196
 197 /* RTE ethernet private data */
 198 struct avp_adapter {
 199         struct avp_dev avp;
 200 } __rte_cache_aligned;
 201
 202
 203 /* 32-bit MMIO register write */
 204 #define AVP_WRITE32(_value, _addr) rte_write32_relaxed((_value), (_addr))
 205
 206 /* 32-bit MMIO register read */
 207 #define AVP_READ32(_addr) rte_read32_relaxed((_addr))
 208
 209 /* Macro to cast the ethernet device private data to a AVP object */
 210 #define AVP_DEV_PRIVATE_TO_HW(adapter) \
 211         (&((struct avp_adapter *)adapter)->avp)
 212
 213 /*
 214  * Defines the structure of a AVP device queue for the purpose of handling the
 215  * receive and transmit burst callback functions
 216  */
 217 struct avp_queue {
 218         struct rte_eth_dev_data *dev_data;
 219         /**< Backpointer to ethernet device data */
 220         struct avp_dev *avp; /**< Backpointer to AVP device */
 221         uint16_t queue_id;
 222         /**< Queue identifier used for indexing current queue */
 223         uint16_t queue_base;
 224         /**< Base queue identifier for queue servicing */
 225         uint16_t queue_limit;
 226         /**< Maximum queue identifier for queue servicing */
 227
 228         uint64_t packets;
 229         uint64_t bytes;
 230         uint64_t errors;
 231 };
 232
 233 /* send a request and wait for a response
 234  *
 235  * @warning must be called while holding the avp->lock spinlock.
 236  */
 237 static int
 238 avp_dev_process_request(struct avp_dev *avp, struct rte_avp_request *request)
 239 {
 240         unsigned int retry = AVP_MAX_REQUEST_RETRY;
 241         void *resp_addr = NULL;
 242         unsigned int count;
 243         int ret;
 244
 245         PMD_DRV_LOG(DEBUG, "Sending request %u to host\n", request->req_id);
 246
 247         request->result = -ENOTSUP;
 248
 249         /* Discard any stale responses before starting a new request */
 250         while (avp_fifo_get(avp->resp_q, (void **)&resp_addr, 1))
 251                 PMD_DRV_LOG(DEBUG, "Discarding stale response\n");
 252
 253         rte_memcpy(avp->sync_addr, request, sizeof(*request));
 254         count = avp_fifo_put(avp->req_q, &avp->host_sync_addr, 1);
 255         if (count < 1) {
 256                 PMD_DRV_LOG(ERR, "Cannot send request %u to host\n",
 257                             request->req_id);
 258                 ret = -EBUSY;
 259                 goto done;
 260         }
 261
 262         while (retry--) {
 263                 /* wait for a response */
 264                 usleep(AVP_REQUEST_DELAY_USECS);
 265
 266                 count = avp_fifo_count(avp->resp_q);
 267                 if (count >= 1) {
 268                         /* response received */
 269                         break;
 270                 }
 271
 272                 if ((count < 1) && (retry == 0)) {
 273                         PMD_DRV_LOG(ERR, "Timeout while waiting for a response for %u\n",
 274                                     request->req_id);
 275                         ret = -ETIME;
 276                         goto done;
 277                 }
 278         }
 279
 280         /* retrieve the response */
 281         count = avp_fifo_get(avp->resp_q, (void **)&resp_addr, 1);
 282         if ((count != 1) || (resp_addr != avp->host_sync_addr)) {
 283                 PMD_DRV_LOG(ERR, "Invalid response from host, count=%u resp=%p host_sync_addr=%p\n",
 284                             count, resp_addr, avp->host_sync_addr);
 285                 ret = -ENODATA;
 286                 goto done;
 287         }
 288
 289         /* copy to user buffer */
 290         rte_memcpy(request, avp->sync_addr, sizeof(*request));
 291         ret = 0;
 292
 293         PMD_DRV_LOG(DEBUG, "Result %d received for request %u\n",
 294                     request->result, request->req_id);
 295
 296 done:
 297         return ret;
 298 }
 299
 300 static int
 301 avp_dev_ctrl_set_link_state(struct rte_eth_dev *eth_dev, unsigned int state)
 302 {
 303         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 304         struct rte_avp_request request;
 305         int ret;
 306
 307         /* setup a link state change request */
 308         memset(&request, 0, sizeof(request));
 309         request.req_id = RTE_AVP_REQ_CFG_NETWORK_IF;
 310         request.if_up = state;
 311
 312         ret = avp_dev_process_request(avp, &request);
 313
 314         return ret == 0 ? request.result : ret;
 315 }
 316
 317 static int
 318 avp_dev_ctrl_set_config(struct rte_eth_dev *eth_dev,
 319                         struct rte_avp_device_config *config)
 320 {
 321         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 322         struct rte_avp_request request;
 323         int ret;
 324
 325         /* setup a configure request */
 326         memset(&request, 0, sizeof(request));
 327         request.req_id = RTE_AVP_REQ_CFG_DEVICE;
 328         memcpy(&request.config, config, sizeof(request.config));
 329
 330         ret = avp_dev_process_request(avp, &request);
 331
 332         return ret == 0 ? request.result : ret;
 333 }
 334
 335 static int
 336 avp_dev_ctrl_shutdown(struct rte_eth_dev *eth_dev)
 337 {
 338         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 339         struct rte_avp_request request;
 340         int ret;
 341
 342         /* setup a shutdown request */
 343         memset(&request, 0, sizeof(request));
 344         request.req_id = RTE_AVP_REQ_SHUTDOWN_DEVICE;
 345
 346         ret = avp_dev_process_request(avp, &request);
 347
 348         return ret == 0 ? request.result : ret;
 349 }
 350
 351 /* translate from host mbuf virtual address to guest virtual address */
 352 static inline void *
 353 avp_dev_translate_buffer(struct avp_dev *avp, void *host_mbuf_address)
 354 {
 355         return RTE_PTR_ADD(RTE_PTR_SUB(host_mbuf_address,
 356                                        (uintptr_t)avp->host_mbuf_addr),
 357                            (uintptr_t)avp->mbuf_addr);
 358 }
 359
 360 /* translate from host physical address to guest virtual address */
 361 static void *
 362 avp_dev_translate_address(struct rte_eth_dev *eth_dev,
 363                           rte_iova_t host_phys_addr)
 364 {
 365         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 366         struct rte_mem_resource *resource;
 367         struct rte_avp_memmap_info *info;
 368         struct rte_avp_memmap *map;
 369         off_t offset;
 370         void *addr;
 371         unsigned int i;
 372
 373         addr = pci_dev->mem_resource[RTE_AVP_PCI_MEMORY_BAR].addr;
 374         resource = &pci_dev->mem_resource[RTE_AVP_PCI_MEMMAP_BAR];
 375         info = (struct rte_avp_memmap_info *)resource->addr;
 376
 377         offset = 0;
 378         for (i = 0; i < info->nb_maps; i++) {
 379                 /* search all segments looking for a matching address */
 380                 map = &info->maps[i];
 381
 382                 if ((host_phys_addr >= map->phys_addr) &&
 383                         (host_phys_addr < (map->phys_addr + map->length))) {
 384                         /* address is within this segment */
 385                         offset += (host_phys_addr - map->phys_addr);
 386                         addr = RTE_PTR_ADD(addr, offset);
 387
 388                         PMD_DRV_LOG(DEBUG, "Translating host physical 0x%" PRIx64 " to guest virtual 0x%p\n",
 389                                     host_phys_addr, addr);
 390
 391                         return addr;
 392                 }
 393                 offset += map->length;
 394         }
 395
 396         return NULL;
 397 }
 398
 399 /* verify that the incoming device version is compatible with our version */
 400 static int
 401 avp_dev_version_check(uint32_t version)
 402 {
 403         uint32_t driver = RTE_AVP_STRIP_MINOR_VERSION(AVP_DPDK_DRIVER_VERSION);
 404         uint32_t device = RTE_AVP_STRIP_MINOR_VERSION(version);
 405
 406         if (device <= driver) {
 407                 /* the host driver version is less than or equal to ours */
 408                 return 0;
 409         }
 410
 411         return 1;
 412 }
 413
 414 /* verify that memory regions have expected version and validation markers */
 415 static int
 416 avp_dev_check_regions(struct rte_eth_dev *eth_dev)
 417 {
 418         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 419         struct rte_avp_memmap_info *memmap;
 420         struct rte_avp_device_info *info;
 421         struct rte_mem_resource *resource;
 422         unsigned int i;
 423
 424         /* Dump resource info for debug */
 425         for (i = 0; i < PCI_MAX_RESOURCE; i++) {
 426                 resource = &pci_dev->mem_resource[i];
 427                 if ((resource->phys_addr == 0) || (resource->len == 0))
 428                         continue;
 429
 430                 PMD_DRV_LOG(DEBUG, "resource[%u]: phys=0x%" PRIx64 " len=%" PRIu64 " addr=%p\n",
 431                             i, resource->phys_addr,
 432                             resource->len, resource->addr);
 433
 434                 switch (i) {
 435                 case RTE_AVP_PCI_MEMMAP_BAR:
 436                         memmap = (struct rte_avp_memmap_info *)resource->addr;
 437                         if ((memmap->magic != RTE_AVP_MEMMAP_MAGIC) ||
 438                             (memmap->version != RTE_AVP_MEMMAP_VERSION)) {
 439                                 PMD_DRV_LOG(ERR, "Invalid memmap magic 0x%08x and version %u\n",
 440                                             memmap->magic, memmap->version);
 441                                 return -EINVAL;
 442                         }
 443                         break;
 444
 445                 case RTE_AVP_PCI_DEVICE_BAR:
 446                         info = (struct rte_avp_device_info *)resource->addr;
 447                         if ((info->magic != RTE_AVP_DEVICE_MAGIC) ||
 448                             avp_dev_version_check(info->version)) {
 449                                 PMD_DRV_LOG(ERR, "Invalid device info magic 0x%08x or version 0x%08x > 0x%08x\n",
 450                                             info->magic, info->version,
 451                                             AVP_DPDK_DRIVER_VERSION);
 452                                 return -EINVAL;
 453                         }
 454                         break;
 455
 456                 case RTE_AVP_PCI_MEMORY_BAR:
 457                 case RTE_AVP_PCI_MMIO_BAR:
 458                         if (resource->addr == NULL) {
 459                                 PMD_DRV_LOG(ERR, "Missing address space for BAR%u\n",
 460                                             i);
 461                                 return -EINVAL;
 462                         }
 463                         break;
 464
 465                 case RTE_AVP_PCI_MSIX_BAR:
 466                 default:
 467                         /* no validation required */
 468                         break;
 469                 }
 470         }
 471
 472         return 0;
 473 }
 474
 475 static int
 476 avp_dev_detach(struct rte_eth_dev *eth_dev)
 477 {
 478         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 479         int ret;
 480
 481         PMD_DRV_LOG(NOTICE, "Detaching port %u from AVP device 0x%" PRIx64 "\n",
 482                     eth_dev->data->port_id, avp->device_id);
 483
 484         rte_spinlock_lock(&avp->lock);
 485
 486         if (avp->flags & AVP_F_DETACHED) {
 487                 PMD_DRV_LOG(NOTICE, "port %u already detached\n",
 488                             eth_dev->data->port_id);
 489                 ret = 0;
 490                 goto unlock;
 491         }
 492
 493         /* shutdown the device first so the host stops sending us packets. */
 494         ret = avp_dev_ctrl_shutdown(eth_dev);
 495         if (ret < 0) {
 496                 PMD_DRV_LOG(ERR, "Failed to send/recv shutdown to host, ret=%d\n",
 497                             ret);
 498                 avp->flags &= ~AVP_F_DETACHED;
 499                 goto unlock;
 500         }
 501
 502         avp->flags |= AVP_F_DETACHED;
 503         rte_wmb();
 504
 505         /* wait for queues to acknowledge the presence of the detach flag */
 506         rte_delay_ms(1);
 507
 508         ret = 0;
 509
 510 unlock:
 511         rte_spinlock_unlock(&avp->lock);
 512         return ret;
 513 }
 514
 515 static void
 516 _avp_set_rx_queue_mappings(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
 517 {
 518         struct avp_dev *avp =
 519                 AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 520         struct avp_queue *rxq;
 521         uint16_t queue_count;
 522         uint16_t remainder;
 523
 524         rxq = (struct avp_queue *)eth_dev->data->rx_queues[rx_queue_id];
 525
 526         /*
 527          * Must map all AVP fifos as evenly as possible between the configured
 528          * device queues.  Each device queue will service a subset of the AVP
 529          * fifos. If there is an odd number of device queues the first set of
 530          * device queues will get the extra AVP fifos.
 531          */
 532         queue_count = avp->num_rx_queues / eth_dev->data->nb_rx_queues;
 533         remainder = avp->num_rx_queues % eth_dev->data->nb_rx_queues;
 534         if (rx_queue_id < remainder) {
 535                 /* these queues must service one extra FIFO */
 536                 rxq->queue_base = rx_queue_id * (queue_count + 1);
 537                 rxq->queue_limit = rxq->queue_base + (queue_count + 1) - 1;
 538         } else {
 539                 /* these queues service the regular number of FIFO */
 540                 rxq->queue_base = ((remainder * (queue_count + 1)) +
 541                                    ((rx_queue_id - remainder) * queue_count));
 542                 rxq->queue_limit = rxq->queue_base + queue_count - 1;
 543         }
 544
 545         PMD_DRV_LOG(DEBUG, "rxq %u at %p base %u limit %u\n",
 546                     rx_queue_id, rxq, rxq->queue_base, rxq->queue_limit);
 547
 548         rxq->queue_id = rxq->queue_base;
 549 }
 550
 551 static void
 552 _avp_set_queue_counts(struct rte_eth_dev *eth_dev)
 553 {
 554         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 555         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 556         struct rte_avp_device_info *host_info;
 557         void *addr;
 558
 559         addr = pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR].addr;
 560         host_info = (struct rte_avp_device_info *)addr;
 561
 562         /*
 563          * the transmit direction is not negotiated beyond respecting the max
 564          * number of queues because the host can handle arbitrary guest tx
 565          * queues (host rx queues).
 566          */
 567         avp->num_tx_queues = eth_dev->data->nb_tx_queues;
 568
 569         /*
 570          * the receive direction is more restrictive.  The host requires a
 571          * minimum number of guest rx queues (host tx queues) therefore
 572          * negotiate a value that is at least as large as the host minimum
 573          * requirement.  If the host and guest values are not identical then a
 574          * mapping will be established in the receive_queue_setup function.
 575          */
 576         avp->num_rx_queues = RTE_MAX(host_info->min_rx_queues,
 577                                      eth_dev->data->nb_rx_queues);
 578
 579         PMD_DRV_LOG(DEBUG, "Requesting %u Tx and %u Rx queues from host\n",
 580                     avp->num_tx_queues, avp->num_rx_queues);
 581 }
 582
 583 static int
 584 avp_dev_attach(struct rte_eth_dev *eth_dev)
 585 {
 586         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 587         struct rte_avp_device_config config;
 588         unsigned int i;
 589         int ret;
 590
 591         PMD_DRV_LOG(NOTICE, "Attaching port %u to AVP device 0x%" PRIx64 "\n",
 592                     eth_dev->data->port_id, avp->device_id);
 593
 594         rte_spinlock_lock(&avp->lock);
 595
 596         if (!(avp->flags & AVP_F_DETACHED)) {
 597                 PMD_DRV_LOG(NOTICE, "port %u already attached\n",
 598                             eth_dev->data->port_id);
 599                 ret = 0;
 600                 goto unlock;
 601         }
 602
 603         /*
 604          * make sure that the detached flag is set prior to reconfiguring the
 605          * queues.
 606          */
 607         avp->flags |= AVP_F_DETACHED;
 608         rte_wmb();
 609
 610         /*
 611          * re-run the device create utility which will parse the new host info
 612          * and setup the AVP device queue pointers.
 613          */
 614         ret = avp_dev_create(RTE_ETH_DEV_TO_PCI(eth_dev), eth_dev);
 615         if (ret < 0) {
 616                 PMD_DRV_LOG(ERR, "Failed to re-create AVP device, ret=%d\n",
 617                             ret);
 618                 goto unlock;
 619         }
 620
 621         if (avp->flags & AVP_F_CONFIGURED) {
 622                 /*
 623                  * Update the receive queue mapping to handle cases where the
 624                  * source and destination hosts have different queue
 625                  * requirements.  As long as the DETACHED flag is asserted the
 626                  * queue table should not be referenced so it should be safe to
 627                  * update it.
 628                  */
 629                 _avp_set_queue_counts(eth_dev);
 630                 for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
 631                         _avp_set_rx_queue_mappings(eth_dev, i);
 632
 633                 /*
 634                  * Update the host with our config details so that it knows the
 635                  * device is active.
 636                  */
 637                 memset(&config, 0, sizeof(config));
 638                 config.device_id = avp->device_id;
 639                 config.driver_type = RTE_AVP_DRIVER_TYPE_DPDK;
 640                 config.driver_version = AVP_DPDK_DRIVER_VERSION;
 641                 config.features = avp->features;
 642                 config.num_tx_queues = avp->num_tx_queues;
 643                 config.num_rx_queues = avp->num_rx_queues;
 644                 config.if_up = !!(avp->flags & AVP_F_LINKUP);
 645
 646                 ret = avp_dev_ctrl_set_config(eth_dev, &config);
 647                 if (ret < 0) {
 648                         PMD_DRV_LOG(ERR, "Config request failed by host, ret=%d\n",
 649                                     ret);
 650                         goto unlock;
 651                 }
 652         }
 653
 654         rte_wmb();
 655         avp->flags &= ~AVP_F_DETACHED;
 656
 657         ret = 0;
 658
 659 unlock:
 660         rte_spinlock_unlock(&avp->lock);
 661         return ret;
 662 }
 663
 664 static void
 665 avp_dev_interrupt_handler(void *data)
 666 {
 667         struct rte_eth_dev *eth_dev = data;
 668         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 669         void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
 670         uint32_t status, value;
 671         int ret;
 672
 673         if (registers == NULL)
 674                 rte_panic("no mapped MMIO register space\n");
 675
 676         /* read the interrupt status register
 677          * note: this register clears on read so all raised interrupts must be
 678          *    handled or remembered for later processing
 679          */
 680         status = AVP_READ32(
 681                 RTE_PTR_ADD(registers,
 682                             RTE_AVP_INTERRUPT_STATUS_OFFSET));
 683
 684         if (status & RTE_AVP_MIGRATION_INTERRUPT_MASK) {
 685                 /* handle interrupt based on current status */
 686                 value = AVP_READ32(
 687                         RTE_PTR_ADD(registers,
 688                                     RTE_AVP_MIGRATION_STATUS_OFFSET));
 689                 switch (value) {
 690                 case RTE_AVP_MIGRATION_DETACHED:
 691                         ret = avp_dev_detach(eth_dev);
 692                         break;
 693                 case RTE_AVP_MIGRATION_ATTACHED:
 694                         ret = avp_dev_attach(eth_dev);
 695                         break;
 696                 default:
 697                         PMD_DRV_LOG(ERR, "unexpected migration status, status=%u\n",
 698                                     value);
 699                         ret = -EINVAL;
 700                 }
 701
 702                 /* acknowledge the request by writing out our current status */
 703                 value = (ret == 0 ? value : RTE_AVP_MIGRATION_ERROR);
 704                 AVP_WRITE32(value,
 705                             RTE_PTR_ADD(registers,
 706                                         RTE_AVP_MIGRATION_ACK_OFFSET));
 707
 708                 PMD_DRV_LOG(NOTICE, "AVP migration interrupt handled\n");
 709         }
 710
 711         if (status & ~RTE_AVP_MIGRATION_INTERRUPT_MASK)
 712                 PMD_DRV_LOG(WARNING, "AVP unexpected interrupt, status=0x%08x\n",
 713                             status);
 714
 715         /* re-enable UIO interrupt handling */
 716         ret = rte_intr_enable(&pci_dev->intr_handle);
 717         if (ret < 0) {
 718                 PMD_DRV_LOG(ERR, "Failed to re-enable UIO interrupts, ret=%d\n",
 719                             ret);
 720                 /* continue */
 721         }
 722 }
 723
 724 static int
 725 avp_dev_enable_interrupts(struct rte_eth_dev *eth_dev)
 726 {
 727         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 728         void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
 729         int ret;
 730
 731         if (registers == NULL)
 732                 return -EINVAL;
 733
 734         /* enable UIO interrupt handling */
 735         ret = rte_intr_enable(&pci_dev->intr_handle);
 736         if (ret < 0) {
 737                 PMD_DRV_LOG(ERR, "Failed to enable UIO interrupts, ret=%d\n",
 738                             ret);
 739                 return ret;
 740         }
 741
 742         /* inform the device that all interrupts are enabled */
 743         AVP_WRITE32(RTE_AVP_APP_INTERRUPTS_MASK,
 744                     RTE_PTR_ADD(registers, RTE_AVP_INTERRUPT_MASK_OFFSET));
 745
 746         return 0;
 747 }
 748
 749 static int
 750 avp_dev_disable_interrupts(struct rte_eth_dev *eth_dev)
 751 {
 752         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 753         void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
 754         int ret;
 755
 756         if (registers == NULL)
 757                 return 0;
 758
 759         /* inform the device that all interrupts are disabled */
 760         AVP_WRITE32(RTE_AVP_NO_INTERRUPTS_MASK,
 761                     RTE_PTR_ADD(registers, RTE_AVP_INTERRUPT_MASK_OFFSET));
 762
 763         /* enable UIO interrupt handling */
 764         ret = rte_intr_disable(&pci_dev->intr_handle);
 765         if (ret < 0) {
 766                 PMD_DRV_LOG(ERR, "Failed to disable UIO interrupts, ret=%d\n",
 767                             ret);
 768                 return ret;
 769         }
 770
 771         return 0;
 772 }
 773
 774 static int
 775 avp_dev_setup_interrupts(struct rte_eth_dev *eth_dev)
 776 {
 777         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 778         int ret;
 779
 780         /* register a callback handler with UIO for interrupt notifications */
 781         ret = rte_intr_callback_register(&pci_dev->intr_handle,
 782                                          avp_dev_interrupt_handler,
 783                                          (void *)eth_dev);
 784         if (ret < 0) {
 785                 PMD_DRV_LOG(ERR, "Failed to register UIO interrupt callback, ret=%d\n",
 786                             ret);
 787                 return ret;
 788         }
 789
 790         /* enable interrupt processing */
 791         return avp_dev_enable_interrupts(eth_dev);
 792 }
 793
 794 static int
 795 avp_dev_migration_pending(struct rte_eth_dev *eth_dev)
 796 {
 797         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 798         void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
 799         uint32_t value;
 800
 801         if (registers == NULL)
 802                 return 0;
 803
 804         value = AVP_READ32(RTE_PTR_ADD(registers,
 805                                        RTE_AVP_MIGRATION_STATUS_OFFSET));
 806         if (value == RTE_AVP_MIGRATION_DETACHED) {
 807                 /* migration is in progress; ack it if we have not already */
 808                 AVP_WRITE32(value,
 809                             RTE_PTR_ADD(registers,
 810                                         RTE_AVP_MIGRATION_ACK_OFFSET));
 811                 return 1;
 812         }
 813         return 0;
 814 }
 815
 816 /*
 817  * create a AVP device using the supplied device info by first translating it
 818  * to guest address space(s).
 819  */
 820 static int
 821 avp_dev_create(struct rte_pci_device *pci_dev,
 822                struct rte_eth_dev *eth_dev)
 823 {
 824         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 825         struct rte_avp_device_info *host_info;
 826         struct rte_mem_resource *resource;
 827         unsigned int i;
 828
 829         resource = &pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR];
 830         if (resource->addr == NULL) {
 831                 PMD_DRV_LOG(ERR, "BAR%u is not mapped\n",
 832                             RTE_AVP_PCI_DEVICE_BAR);
 833                 return -EFAULT;
 834         }
 835         host_info = (struct rte_avp_device_info *)resource->addr;
 836
 837         if ((host_info->magic != RTE_AVP_DEVICE_MAGIC) ||
 838                 avp_dev_version_check(host_info->version)) {
 839                 PMD_DRV_LOG(ERR, "Invalid AVP PCI device, magic 0x%08x version 0x%08x > 0x%08x\n",
 840                             host_info->magic, host_info->version,
 841                             AVP_DPDK_DRIVER_VERSION);
 842                 return -EINVAL;
 843         }
 844
 845         PMD_DRV_LOG(DEBUG, "AVP host device is v%u.%u.%u\n",
 846                     RTE_AVP_GET_RELEASE_VERSION(host_info->version),
 847                     RTE_AVP_GET_MAJOR_VERSION(host_info->version),
 848                     RTE_AVP_GET_MINOR_VERSION(host_info->version));
 849
 850         PMD_DRV_LOG(DEBUG, "AVP host supports %u to %u TX queue(s)\n",
 851                     host_info->min_tx_queues, host_info->max_tx_queues);
 852         PMD_DRV_LOG(DEBUG, "AVP host supports %u to %u RX queue(s)\n",
 853                     host_info->min_rx_queues, host_info->max_rx_queues);
 854         PMD_DRV_LOG(DEBUG, "AVP host supports features 0x%08x\n",
 855                     host_info->features);
 856
 857         if (avp->magic != AVP_ETHDEV_MAGIC) {
 858                 /*
 859                  * First time initialization (i.e., not during a VM
 860                  * migration)
 861                  */
 862                 memset(avp, 0, sizeof(*avp));
 863                 avp->magic = AVP_ETHDEV_MAGIC;
 864                 avp->dev_data = eth_dev->data;
 865                 avp->port_id = eth_dev->data->port_id;
 866                 avp->host_mbuf_size = host_info->mbuf_size;
 867                 avp->host_features = host_info->features;
 868                 rte_spinlock_init(&avp->lock);
 869                 memcpy(&avp->ethaddr.addr_bytes[0],
 870                        host_info->ethaddr, ETHER_ADDR_LEN);
 871                 /* adjust max values to not exceed our max */
 872                 avp->max_tx_queues =
 873                         RTE_MIN(host_info->max_tx_queues, RTE_AVP_MAX_QUEUES);
 874                 avp->max_rx_queues =
 875                         RTE_MIN(host_info->max_rx_queues, RTE_AVP_MAX_QUEUES);
 876         } else {
 877                 /* Re-attaching during migration */
 878
 879                 /* TODO... requires validation of host values */
 880                 if ((host_info->features & avp->features) != avp->features) {
 881                         PMD_DRV_LOG(ERR, "AVP host features mismatched; 0x%08x, host=0x%08x\n",
 882                                     avp->features, host_info->features);
 883                         /* this should not be possible; continue for now */
 884                 }
 885         }
 886
 887         /* the device id is allowed to change over migrations */
 888         avp->device_id = host_info->device_id;
 889
 890         /* translate incoming host addresses to guest address space */
 891         PMD_DRV_LOG(DEBUG, "AVP first host tx queue at 0x%" PRIx64 "\n",
 892                     host_info->tx_phys);
 893         PMD_DRV_LOG(DEBUG, "AVP first host alloc queue at 0x%" PRIx64 "\n",
 894                     host_info->alloc_phys);
 895         for (i = 0; i < avp->max_tx_queues; i++) {
 896                 avp->tx_q[i] = avp_dev_translate_address(eth_dev,
 897                         host_info->tx_phys + (i * host_info->tx_size));
 898
 899                 avp->alloc_q[i] = avp_dev_translate_address(eth_dev,
 900                         host_info->alloc_phys + (i * host_info->alloc_size));
 901         }
 902
 903         PMD_DRV_LOG(DEBUG, "AVP first host rx queue at 0x%" PRIx64 "\n",
 904                     host_info->rx_phys);
 905         PMD_DRV_LOG(DEBUG, "AVP first host free queue at 0x%" PRIx64 "\n",
 906                     host_info->free_phys);
 907         for (i = 0; i < avp->max_rx_queues; i++) {
 908                 avp->rx_q[i] = avp_dev_translate_address(eth_dev,
 909                         host_info->rx_phys + (i * host_info->rx_size));
 910                 avp->free_q[i] = avp_dev_translate_address(eth_dev,
 911                         host_info->free_phys + (i * host_info->free_size));
 912         }
 913
 914         PMD_DRV_LOG(DEBUG, "AVP host request queue at 0x%" PRIx64 "\n",
 915                     host_info->req_phys);
 916         PMD_DRV_LOG(DEBUG, "AVP host response queue at 0x%" PRIx64 "\n",
 917                     host_info->resp_phys);
 918         PMD_DRV_LOG(DEBUG, "AVP host sync address at 0x%" PRIx64 "\n",
 919                     host_info->sync_phys);
 920         PMD_DRV_LOG(DEBUG, "AVP host mbuf address at 0x%" PRIx64 "\n",
 921                     host_info->mbuf_phys);
 922         avp->req_q = avp_dev_translate_address(eth_dev, host_info->req_phys);
 923         avp->resp_q = avp_dev_translate_address(eth_dev, host_info->resp_phys);
 924         avp->sync_addr =
 925                 avp_dev_translate_address(eth_dev, host_info->sync_phys);
 926         avp->mbuf_addr =
 927                 avp_dev_translate_address(eth_dev, host_info->mbuf_phys);
 928
 929         /*
 930          * store the host mbuf virtual address so that we can calculate
 931          * relative offsets for each mbuf as they are processed
 932          */
 933         avp->host_mbuf_addr = host_info->mbuf_va;
 934         avp->host_sync_addr = host_info->sync_va;
 935
 936         /*
 937          * store the maximum packet length that is supported by the host.
 938          */
 939         avp->max_rx_pkt_len = host_info->max_rx_pkt_len;
 940         PMD_DRV_LOG(DEBUG, "AVP host max receive packet length is %u\n",
 941                                 host_info->max_rx_pkt_len);
 942
 943         return 0;
 944 }
 945
 946 /*
 947  * This function is based on probe() function in avp_pci.c
 948  * It returns 0 on success.
 949  */
 950 static int
 951 eth_avp_dev_init(struct rte_eth_dev *eth_dev)
 952 {
 953         struct avp_dev *avp =
 954                 AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
 955         struct rte_pci_device *pci_dev;
 956         int ret;
 957
 958         pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 959         eth_dev->dev_ops = &avp_eth_dev_ops;
 960         eth_dev->rx_pkt_burst = &avp_recv_pkts;
 961         eth_dev->tx_pkt_burst = &avp_xmit_pkts;
 962
 963         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 964                 /*
 965                  * no setup required on secondary processes.  All data is saved
 966                  * in dev_private by the primary process. All resource should
 967                  * be mapped to the same virtual address so all pointers should
 968                  * be valid.
 969                  */
 970                 if (eth_dev->data->scattered_rx) {
 971                         PMD_DRV_LOG(NOTICE, "AVP device configured for chained mbufs\n");
 972                         eth_dev->rx_pkt_burst = avp_recv_scattered_pkts;
 973                         eth_dev->tx_pkt_burst = avp_xmit_scattered_pkts;
 974                 }
 975                 return 0;
 976         }
 977
 978         rte_eth_copy_pci_info(eth_dev, pci_dev);
 979
 980         /* Check current migration status */
 981         if (avp_dev_migration_pending(eth_dev)) {
 982                 PMD_DRV_LOG(ERR, "VM live migration operation in progress\n");
 983                 return -EBUSY;
 984         }
 985
 986         /* Check BAR resources */
 987         ret = avp_dev_check_regions(eth_dev);
 988         if (ret < 0) {
 989                 PMD_DRV_LOG(ERR, "Failed to validate BAR resources, ret=%d\n",
 990                             ret);
 991                 return ret;
 992         }
 993
 994         /* Enable interrupts */
 995         ret = avp_dev_setup_interrupts(eth_dev);
 996         if (ret < 0) {
 997                 PMD_DRV_LOG(ERR, "Failed to enable interrupts, ret=%d\n", ret);
 998                 return ret;
 999         }
1000
1001         /* Handle each subtype */
1002         ret = avp_dev_create(pci_dev, eth_dev);
1003         if (ret < 0) {
1004                 PMD_DRV_LOG(ERR, "Failed to create device, ret=%d\n", ret);
1005                 return ret;
1006         }
1007
1008         /* Allocate memory for storing MAC addresses */
1009         eth_dev->data->mac_addrs = rte_zmalloc("avp_ethdev", 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                             ETHER_ADDR_LEN);
1013                 return -ENOMEM;
1014         }
1015
1016         /* Get a mac from device config */
1017         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         int ret;
1026
1027         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1028                 return -EPERM;
1029
1030         if (eth_dev->data == NULL)
1031                 return 0;
1032
1033         ret = avp_dev_disable_interrupts(eth_dev);
1034         if (ret != 0) {
1035                 PMD_DRV_LOG(ERR, "Failed to disable interrupts, ret=%d\n", ret);
1036                 return ret;
1037         }
1038
1039         if (eth_dev->data->mac_addrs != NULL) {
1040                 rte_free(eth_dev->data->mac_addrs);
1041                 eth_dev->data->mac_addrs = NULL;
1042         }
1043
1044         return 0;
1045 }
1046
1047 static int
1048 eth_avp_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1049                   struct rte_pci_device *pci_dev)
1050 {
1051         struct rte_eth_dev *eth_dev;
1052         int ret;
1053
1054         eth_dev = rte_eth_dev_pci_allocate(pci_dev,
1055                                            sizeof(struct avp_adapter));
1056         if (eth_dev == NULL)
1057                 return -ENOMEM;
1058
1059         ret = eth_avp_dev_init(eth_dev);
1060         if (ret)
1061                 rte_eth_dev_pci_release(eth_dev);
1062
1063         return ret;
1064 }
1065
1066 static int
1067 eth_avp_pci_remove(struct rte_pci_device *pci_dev)
1068 {
1069         return rte_eth_dev_pci_generic_remove(pci_dev,
1070                                               eth_avp_dev_uninit);
1071 }
1072
1073 static struct rte_pci_driver rte_avp_pmd = {
1074         .id_table = pci_id_avp_map,
1075         .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1076         .probe = eth_avp_pci_probe,
1077         .remove = eth_avp_pci_remove,
1078 };
1079
1080 static int
1081 avp_dev_enable_scattered(struct rte_eth_dev *eth_dev,
1082                          struct avp_dev *avp)
1083 {
1084         unsigned int max_rx_pkt_len;
1085
1086         max_rx_pkt_len = eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
1087
1088         if ((max_rx_pkt_len > avp->guest_mbuf_size) ||
1089             (max_rx_pkt_len > avp->host_mbuf_size)) {
1090                 /*
1091                  * If the guest MTU is greater than either the host or guest
1092                  * buffers then chained mbufs have to be enabled in the TX
1093                  * direction.  It is assumed that the application will not need
1094                  * to send packets larger than their max_rx_pkt_len (MRU).
1095                  */
1096                 return 1;
1097         }
1098
1099         if ((avp->max_rx_pkt_len > avp->guest_mbuf_size) ||
1100             (avp->max_rx_pkt_len > avp->host_mbuf_size)) {
1101                 /*
1102                  * If the host MRU is greater than its own mbuf size or the
1103                  * guest mbuf size then chained mbufs have to be enabled in the
1104                  * RX direction.
1105                  */
1106                 return 1;
1107         }
1108
1109         return 0;
1110 }
1111
1112 static int
1113 avp_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
1114                        uint16_t rx_queue_id,
1115                        uint16_t nb_rx_desc,
1116                        unsigned int socket_id,
1117                        const struct rte_eth_rxconf *rx_conf,
1118                        struct rte_mempool *pool)
1119 {
1120         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1121         struct rte_pktmbuf_pool_private *mbp_priv;
1122         struct avp_queue *rxq;
1123
1124         if (rx_queue_id >= eth_dev->data->nb_rx_queues) {
1125                 PMD_DRV_LOG(ERR, "RX queue id is out of range: rx_queue_id=%u, nb_rx_queues=%u\n",
1126                             rx_queue_id, eth_dev->data->nb_rx_queues);
1127                 return -EINVAL;
1128         }
1129
1130         /* Save mbuf pool pointer */
1131         avp->pool = pool;
1132
1133         /* Save the local mbuf size */
1134         mbp_priv = rte_mempool_get_priv(pool);
1135         avp->guest_mbuf_size = (uint16_t)(mbp_priv->mbuf_data_room_size);
1136         avp->guest_mbuf_size -= RTE_PKTMBUF_HEADROOM;
1137
1138         if (avp_dev_enable_scattered(eth_dev, avp)) {
1139                 if (!eth_dev->data->scattered_rx) {
1140                         PMD_DRV_LOG(NOTICE, "AVP device configured for chained mbufs\n");
1141                         eth_dev->data->scattered_rx = 1;
1142                         eth_dev->rx_pkt_burst = avp_recv_scattered_pkts;
1143                         eth_dev->tx_pkt_burst = avp_xmit_scattered_pkts;
1144                 }
1145         }
1146
1147         PMD_DRV_LOG(DEBUG, "AVP max_rx_pkt_len=(%u,%u) mbuf_size=(%u,%u)\n",
1148                     avp->max_rx_pkt_len,
1149                     eth_dev->data->dev_conf.rxmode.max_rx_pkt_len,
1150                     avp->host_mbuf_size,
1151                     avp->guest_mbuf_size);
1152
1153         /* allocate a queue object */
1154         rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct avp_queue),
1155                                  RTE_CACHE_LINE_SIZE, socket_id);
1156         if (rxq == NULL) {
1157                 PMD_DRV_LOG(ERR, "Failed to allocate new Rx queue object\n");
1158                 return -ENOMEM;
1159         }
1160
1161         /* save back pointers to AVP and Ethernet devices */
1162         rxq->avp = avp;
1163         rxq->dev_data = eth_dev->data;
1164         eth_dev->data->rx_queues[rx_queue_id] = (void *)rxq;
1165
1166         /* setup the queue receive mapping for the current queue. */
1167         _avp_set_rx_queue_mappings(eth_dev, rx_queue_id);
1168
1169         PMD_DRV_LOG(DEBUG, "Rx queue %u setup at %p\n", rx_queue_id, rxq);
1170
1171         (void)nb_rx_desc;
1172         (void)rx_conf;
1173         return 0;
1174 }
1175
1176 static int
1177 avp_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
1178                        uint16_t tx_queue_id,
1179                        uint16_t nb_tx_desc,
1180                        unsigned int socket_id,
1181                        const struct rte_eth_txconf *tx_conf)
1182 {
1183         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1184         struct avp_queue *txq;
1185
1186         if (tx_queue_id >= eth_dev->data->nb_tx_queues) {
1187                 PMD_DRV_LOG(ERR, "TX queue id is out of range: tx_queue_id=%u, nb_tx_queues=%u\n",
1188                             tx_queue_id, eth_dev->data->nb_tx_queues);
1189                 return -EINVAL;
1190         }
1191
1192         /* allocate a queue object */
1193         txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct avp_queue),
1194                                  RTE_CACHE_LINE_SIZE, socket_id);
1195         if (txq == NULL) {
1196                 PMD_DRV_LOG(ERR, "Failed to allocate new Tx queue object\n");
1197                 return -ENOMEM;
1198         }
1199
1200         /* only the configured set of transmit queues are used */
1201         txq->queue_id = tx_queue_id;
1202         txq->queue_base = tx_queue_id;
1203         txq->queue_limit = tx_queue_id;
1204
1205         /* save back pointers to AVP and Ethernet devices */
1206         txq->avp = avp;
1207         txq->dev_data = eth_dev->data;
1208         eth_dev->data->tx_queues[tx_queue_id] = (void *)txq;
1209
1210         PMD_DRV_LOG(DEBUG, "Tx queue %u setup at %p\n", tx_queue_id, txq);
1211
1212         (void)nb_tx_desc;
1213         (void)tx_conf;
1214         return 0;
1215 }
1216
1217 static inline int
1218 _avp_cmp_ether_addr(struct ether_addr *a, struct ether_addr *b)
1219 {
1220         uint16_t *_a = (uint16_t *)&a->addr_bytes[0];
1221         uint16_t *_b = (uint16_t *)&b->addr_bytes[0];
1222         return (_a[0] ^ _b[0]) | (_a[1] ^ _b[1]) | (_a[2] ^ _b[2]);
1223 }
1224
1225 static inline int
1226 _avp_mac_filter(struct avp_dev *avp, struct rte_mbuf *m)
1227 {
1228         struct ether_hdr *eth = rte_pktmbuf_mtod(m, struct ether_hdr *);
1229
1230         if (likely(_avp_cmp_ether_addr(&avp->ethaddr, &eth->d_addr) == 0)) {
1231                 /* allow all packets destined to our address */
1232                 return 0;
1233         }
1234
1235         if (likely(is_broadcast_ether_addr(&eth->d_addr))) {
1236                 /* allow all broadcast packets */
1237                 return 0;
1238         }
1239
1240         if (likely(is_multicast_ether_addr(&eth->d_addr))) {
1241                 /* allow all multicast packets */
1242                 return 0;
1243         }
1244
1245         if (avp->flags & AVP_F_PROMISC) {
1246                 /* allow all packets when in promiscuous mode */
1247                 return 0;
1248         }
1249
1250         return -1;
1251 }
1252
1253 #ifdef RTE_LIBRTE_AVP_DEBUG_BUFFERS
1254 static inline void
1255 __avp_dev_buffer_sanity_check(struct avp_dev *avp, struct rte_avp_desc *buf)
1256 {
1257         struct rte_avp_desc *first_buf;
1258         struct rte_avp_desc *pkt_buf;
1259         unsigned int pkt_len;
1260         unsigned int nb_segs;
1261         void *pkt_data;
1262         unsigned int i;
1263
1264         first_buf = avp_dev_translate_buffer(avp, buf);
1265
1266         i = 0;
1267         pkt_len = 0;
1268         nb_segs = first_buf->nb_segs;
1269         do {
1270                 /* Adjust pointers for guest addressing */
1271                 pkt_buf = avp_dev_translate_buffer(avp, buf);
1272                 if (pkt_buf == NULL)
1273                         rte_panic("bad buffer: segment %u has an invalid address %p\n",
1274                                   i, buf);
1275                 pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1276                 if (pkt_data == NULL)
1277                         rte_panic("bad buffer: segment %u has a NULL data pointer\n",
1278                                   i);
1279                 if (pkt_buf->data_len == 0)
1280                         rte_panic("bad buffer: segment %u has 0 data length\n",
1281                                   i);
1282                 pkt_len += pkt_buf->data_len;
1283                 nb_segs--;
1284                 i++;
1285
1286         } while (nb_segs && (buf = pkt_buf->next) != NULL);
1287
1288         if (nb_segs != 0)
1289                 rte_panic("bad buffer: expected %u segments found %u\n",
1290                           first_buf->nb_segs, (first_buf->nb_segs - nb_segs));
1291         if (pkt_len != first_buf->pkt_len)
1292                 rte_panic("bad buffer: expected length %u found %u\n",
1293                           first_buf->pkt_len, pkt_len);
1294 }
1295
1296 #define avp_dev_buffer_sanity_check(a, b) \
1297         __avp_dev_buffer_sanity_check((a), (b))
1298
1299 #else /* RTE_LIBRTE_AVP_DEBUG_BUFFERS */
1300
1301 #define avp_dev_buffer_sanity_check(a, b) do {} while (0)
1302
1303 #endif
1304
1305 /*
1306  * Copy a host buffer chain to a set of mbufs.  This function assumes that
1307  * there exactly the required number of mbufs to copy all source bytes.
1308  */
1309 static inline struct rte_mbuf *
1310 avp_dev_copy_from_buffers(struct avp_dev *avp,
1311                           struct rte_avp_desc *buf,
1312                           struct rte_mbuf **mbufs,
1313                           unsigned int count)
1314 {
1315         struct rte_mbuf *m_previous = NULL;
1316         struct rte_avp_desc *pkt_buf;
1317         unsigned int total_length = 0;
1318         unsigned int copy_length;
1319         unsigned int src_offset;
1320         struct rte_mbuf *m;
1321         uint16_t ol_flags;
1322         uint16_t vlan_tci;
1323         void *pkt_data;
1324         unsigned int i;
1325
1326         avp_dev_buffer_sanity_check(avp, buf);
1327
1328         /* setup the first source buffer */
1329         pkt_buf = avp_dev_translate_buffer(avp, buf);
1330         pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1331         total_length = pkt_buf->pkt_len;
1332         src_offset = 0;
1333
1334         if (pkt_buf->ol_flags & RTE_AVP_RX_VLAN_PKT) {
1335                 ol_flags = PKT_RX_VLAN;
1336                 vlan_tci = pkt_buf->vlan_tci;
1337         } else {
1338                 ol_flags = 0;
1339                 vlan_tci = 0;
1340         }
1341
1342         for (i = 0; (i < count) && (buf != NULL); i++) {
1343                 /* fill each destination buffer */
1344                 m = mbufs[i];
1345
1346                 if (m_previous != NULL)
1347                         m_previous->next = m;
1348
1349                 m_previous = m;
1350
1351                 do {
1352                         /*
1353                          * Copy as many source buffers as will fit in the
1354                          * destination buffer.
1355                          */
1356                         copy_length = RTE_MIN((avp->guest_mbuf_size -
1357                                                rte_pktmbuf_data_len(m)),
1358                                               (pkt_buf->data_len -
1359                                                src_offset));
1360                         rte_memcpy(RTE_PTR_ADD(rte_pktmbuf_mtod(m, void *),
1361                                                rte_pktmbuf_data_len(m)),
1362                                    RTE_PTR_ADD(pkt_data, src_offset),
1363                                    copy_length);
1364                         rte_pktmbuf_data_len(m) += copy_length;
1365                         src_offset += copy_length;
1366
1367                         if (likely(src_offset == pkt_buf->data_len)) {
1368                                 /* need a new source buffer */
1369                                 buf = pkt_buf->next;
1370                                 if (buf != NULL) {
1371                                         pkt_buf = avp_dev_translate_buffer(
1372                                                 avp, buf);
1373                                         pkt_data = avp_dev_translate_buffer(
1374                                                 avp, pkt_buf->data);
1375                                         src_offset = 0;
1376                                 }
1377                         }
1378
1379                         if (unlikely(rte_pktmbuf_data_len(m) ==
1380                                      avp->guest_mbuf_size)) {
1381                                 /* need a new destination mbuf */
1382                                 break;
1383                         }
1384
1385                 } while (buf != NULL);
1386         }
1387
1388         m = mbufs[0];
1389         m->ol_flags = ol_flags;
1390         m->nb_segs = count;
1391         rte_pktmbuf_pkt_len(m) = total_length;
1392         m->vlan_tci = vlan_tci;
1393
1394         __rte_mbuf_sanity_check(m, 1);
1395
1396         return m;
1397 }
1398
1399 static uint16_t
1400 avp_recv_scattered_pkts(void *rx_queue,
1401                         struct rte_mbuf **rx_pkts,
1402                         uint16_t nb_pkts)
1403 {
1404         struct avp_queue *rxq = (struct avp_queue *)rx_queue;
1405         struct rte_avp_desc *avp_bufs[AVP_MAX_RX_BURST];
1406         struct rte_mbuf *mbufs[RTE_AVP_MAX_MBUF_SEGMENTS];
1407         struct avp_dev *avp = rxq->avp;
1408         struct rte_avp_desc *pkt_buf;
1409         struct rte_avp_fifo *free_q;
1410         struct rte_avp_fifo *rx_q;
1411         struct rte_avp_desc *buf;
1412         unsigned int count, avail, n;
1413         unsigned int guest_mbuf_size;
1414         struct rte_mbuf *m;
1415         unsigned int required;
1416         unsigned int buf_len;
1417         unsigned int port_id;
1418         unsigned int i;
1419
1420         if (unlikely(avp->flags & AVP_F_DETACHED)) {
1421                 /* VM live migration in progress */
1422                 return 0;
1423         }
1424
1425         guest_mbuf_size = avp->guest_mbuf_size;
1426         port_id = avp->port_id;
1427         rx_q = avp->rx_q[rxq->queue_id];
1428         free_q = avp->free_q[rxq->queue_id];
1429
1430         /* setup next queue to service */
1431         rxq->queue_id = (rxq->queue_id < rxq->queue_limit) ?
1432                 (rxq->queue_id + 1) : rxq->queue_base;
1433
1434         /* determine how many slots are available in the free queue */
1435         count = avp_fifo_free_count(free_q);
1436
1437         /* determine how many packets are available in the rx queue */
1438         avail = avp_fifo_count(rx_q);
1439
1440         /* determine how many packets can be received */
1441         count = RTE_MIN(count, avail);
1442         count = RTE_MIN(count, nb_pkts);
1443         count = RTE_MIN(count, (unsigned int)AVP_MAX_RX_BURST);
1444
1445         if (unlikely(count == 0)) {
1446                 /* no free buffers, or no buffers on the rx queue */
1447                 return 0;
1448         }
1449
1450         /* retrieve pending packets */
1451         n = avp_fifo_get(rx_q, (void **)&avp_bufs, count);
1452         PMD_RX_LOG(DEBUG, "Receiving %u packets from Rx queue at %p\n",
1453                    count, rx_q);
1454
1455         count = 0;
1456         for (i = 0; i < n; i++) {
1457                 /* prefetch next entry while processing current one */
1458                 if (i + 1 < n) {
1459                         pkt_buf = avp_dev_translate_buffer(avp,
1460                                                            avp_bufs[i + 1]);
1461                         rte_prefetch0(pkt_buf);
1462                 }
1463                 buf = avp_bufs[i];
1464
1465                 /* Peek into the first buffer to determine the total length */
1466                 pkt_buf = avp_dev_translate_buffer(avp, buf);
1467                 buf_len = pkt_buf->pkt_len;
1468
1469                 /* Allocate enough mbufs to receive the entire packet */
1470                 required = (buf_len + guest_mbuf_size - 1) / guest_mbuf_size;
1471                 if (rte_pktmbuf_alloc_bulk(avp->pool, mbufs, required)) {
1472                         rxq->dev_data->rx_mbuf_alloc_failed++;
1473                         continue;
1474                 }
1475
1476                 /* Copy the data from the buffers to our mbufs */
1477                 m = avp_dev_copy_from_buffers(avp, buf, mbufs, required);
1478
1479                 /* finalize mbuf */
1480                 m->port = port_id;
1481
1482                 if (_avp_mac_filter(avp, m) != 0) {
1483                         /* silently discard packets not destined to our MAC */
1484                         rte_pktmbuf_free(m);
1485                         continue;
1486                 }
1487
1488                 /* return new mbuf to caller */
1489                 rx_pkts[count++] = m;
1490                 rxq->bytes += buf_len;
1491         }
1492
1493         rxq->packets += count;
1494
1495         /* return the buffers to the free queue */
1496         avp_fifo_put(free_q, (void **)&avp_bufs[0], n);
1497
1498         return count;
1499 }
1500
1501
1502 static uint16_t
1503 avp_recv_pkts(void *rx_queue,
1504               struct rte_mbuf **rx_pkts,
1505               uint16_t nb_pkts)
1506 {
1507         struct avp_queue *rxq = (struct avp_queue *)rx_queue;
1508         struct rte_avp_desc *avp_bufs[AVP_MAX_RX_BURST];
1509         struct avp_dev *avp = rxq->avp;
1510         struct rte_avp_desc *pkt_buf;
1511         struct rte_avp_fifo *free_q;
1512         struct rte_avp_fifo *rx_q;
1513         unsigned int count, avail, n;
1514         unsigned int pkt_len;
1515         struct rte_mbuf *m;
1516         char *pkt_data;
1517         unsigned int i;
1518
1519         if (unlikely(avp->flags & AVP_F_DETACHED)) {
1520                 /* VM live migration in progress */
1521                 return 0;
1522         }
1523
1524         rx_q = avp->rx_q[rxq->queue_id];
1525         free_q = avp->free_q[rxq->queue_id];
1526
1527         /* setup next queue to service */
1528         rxq->queue_id = (rxq->queue_id < rxq->queue_limit) ?
1529                 (rxq->queue_id + 1) : rxq->queue_base;
1530
1531         /* determine how many slots are available in the free queue */
1532         count = avp_fifo_free_count(free_q);
1533
1534         /* determine how many packets are available in the rx queue */
1535         avail = avp_fifo_count(rx_q);
1536
1537         /* determine how many packets can be received */
1538         count = RTE_MIN(count, avail);
1539         count = RTE_MIN(count, nb_pkts);
1540         count = RTE_MIN(count, (unsigned int)AVP_MAX_RX_BURST);
1541
1542         if (unlikely(count == 0)) {
1543                 /* no free buffers, or no buffers on the rx queue */
1544                 return 0;
1545         }
1546
1547         /* retrieve pending packets */
1548         n = avp_fifo_get(rx_q, (void **)&avp_bufs, count);
1549         PMD_RX_LOG(DEBUG, "Receiving %u packets from Rx queue at %p\n",
1550                    count, rx_q);
1551
1552         count = 0;
1553         for (i = 0; i < n; i++) {
1554                 /* prefetch next entry while processing current one */
1555                 if (i < n - 1) {
1556                         pkt_buf = avp_dev_translate_buffer(avp,
1557                                                            avp_bufs[i + 1]);
1558                         rte_prefetch0(pkt_buf);
1559                 }
1560
1561                 /* Adjust host pointers for guest addressing */
1562                 pkt_buf = avp_dev_translate_buffer(avp, avp_bufs[i]);
1563                 pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1564                 pkt_len = pkt_buf->pkt_len;
1565
1566                 if (unlikely((pkt_len > avp->guest_mbuf_size) ||
1567                              (pkt_buf->nb_segs > 1))) {
1568                         /*
1569                          * application should be using the scattered receive
1570                          * function
1571                          */
1572                         rxq->errors++;
1573                         continue;
1574                 }
1575
1576                 /* process each packet to be transmitted */
1577                 m = rte_pktmbuf_alloc(avp->pool);
1578                 if (unlikely(m == NULL)) {
1579                         rxq->dev_data->rx_mbuf_alloc_failed++;
1580                         continue;
1581                 }
1582
1583                 /* copy data out of the host buffer to our buffer */
1584                 m->data_off = RTE_PKTMBUF_HEADROOM;
1585                 rte_memcpy(rte_pktmbuf_mtod(m, void *), pkt_data, pkt_len);
1586
1587                 /* initialize the local mbuf */
1588                 rte_pktmbuf_data_len(m) = pkt_len;
1589                 rte_pktmbuf_pkt_len(m) = pkt_len;
1590                 m->port = avp->port_id;
1591
1592                 if (pkt_buf->ol_flags & RTE_AVP_RX_VLAN_PKT) {
1593                         m->ol_flags = PKT_RX_VLAN;
1594                         m->vlan_tci = pkt_buf->vlan_tci;
1595                 }
1596
1597                 if (_avp_mac_filter(avp, m) != 0) {
1598                         /* silently discard packets not destined to our MAC */
1599                         rte_pktmbuf_free(m);
1600                         continue;
1601                 }
1602
1603                 /* return new mbuf to caller */
1604                 rx_pkts[count++] = m;
1605                 rxq->bytes += pkt_len;
1606         }
1607
1608         rxq->packets += count;
1609
1610         /* return the buffers to the free queue */
1611         avp_fifo_put(free_q, (void **)&avp_bufs[0], n);
1612
1613         return count;
1614 }
1615
1616 /*
1617  * Copy a chained mbuf to a set of host buffers.  This function assumes that
1618  * there are sufficient destination buffers to contain the entire source
1619  * packet.
1620  */
1621 static inline uint16_t
1622 avp_dev_copy_to_buffers(struct avp_dev *avp,
1623                         struct rte_mbuf *mbuf,
1624                         struct rte_avp_desc **buffers,
1625                         unsigned int count)
1626 {
1627         struct rte_avp_desc *previous_buf = NULL;
1628         struct rte_avp_desc *first_buf = NULL;
1629         struct rte_avp_desc *pkt_buf;
1630         struct rte_avp_desc *buf;
1631         size_t total_length;
1632         struct rte_mbuf *m;
1633         size_t copy_length;
1634         size_t src_offset;
1635         char *pkt_data;
1636         unsigned int i;
1637
1638         __rte_mbuf_sanity_check(mbuf, 1);
1639
1640         m = mbuf;
1641         src_offset = 0;
1642         total_length = rte_pktmbuf_pkt_len(m);
1643         for (i = 0; (i < count) && (m != NULL); i++) {
1644                 /* fill each destination buffer */
1645                 buf = buffers[i];
1646
1647                 if (i < count - 1) {
1648                         /* prefetch next entry while processing this one */
1649                         pkt_buf = avp_dev_translate_buffer(avp, buffers[i + 1]);
1650                         rte_prefetch0(pkt_buf);
1651                 }
1652
1653                 /* Adjust pointers for guest addressing */
1654                 pkt_buf = avp_dev_translate_buffer(avp, buf);
1655                 pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1656
1657                 /* setup the buffer chain */
1658                 if (previous_buf != NULL)
1659                         previous_buf->next = buf;
1660                 else
1661                         first_buf = pkt_buf;
1662
1663                 previous_buf = pkt_buf;
1664
1665                 do {
1666                         /*
1667                          * copy as many source mbuf segments as will fit in the
1668                          * destination buffer.
1669                          */
1670                         copy_length = RTE_MIN((avp->host_mbuf_size -
1671                                                pkt_buf->data_len),
1672                                               (rte_pktmbuf_data_len(m) -
1673                                                src_offset));
1674                         rte_memcpy(RTE_PTR_ADD(pkt_data, pkt_buf->data_len),
1675                                    RTE_PTR_ADD(rte_pktmbuf_mtod(m, void *),
1676                                                src_offset),
1677                                    copy_length);
1678                         pkt_buf->data_len += copy_length;
1679                         src_offset += copy_length;
1680
1681                         if (likely(src_offset == rte_pktmbuf_data_len(m))) {
1682                                 /* need a new source buffer */
1683                                 m = m->next;
1684                                 src_offset = 0;
1685                         }
1686
1687                         if (unlikely(pkt_buf->data_len ==
1688                                      avp->host_mbuf_size)) {
1689                                 /* need a new destination buffer */
1690                                 break;
1691                         }
1692
1693                 } while (m != NULL);
1694         }
1695
1696         first_buf->nb_segs = count;
1697         first_buf->pkt_len = total_length;
1698
1699         if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
1700                 first_buf->ol_flags |= RTE_AVP_TX_VLAN_PKT;
1701                 first_buf->vlan_tci = mbuf->vlan_tci;
1702         }
1703
1704         avp_dev_buffer_sanity_check(avp, buffers[0]);
1705
1706         return total_length;
1707 }
1708
1709
1710 static uint16_t
1711 avp_xmit_scattered_pkts(void *tx_queue,
1712                         struct rte_mbuf **tx_pkts,
1713                         uint16_t nb_pkts)
1714 {
1715         struct rte_avp_desc *avp_bufs[(AVP_MAX_TX_BURST *
1716                                        RTE_AVP_MAX_MBUF_SEGMENTS)];
1717         struct avp_queue *txq = (struct avp_queue *)tx_queue;
1718         struct rte_avp_desc *tx_bufs[AVP_MAX_TX_BURST];
1719         struct avp_dev *avp = txq->avp;
1720         struct rte_avp_fifo *alloc_q;
1721         struct rte_avp_fifo *tx_q;
1722         unsigned int count, avail, n;
1723         unsigned int orig_nb_pkts;
1724         struct rte_mbuf *m;
1725         unsigned int required;
1726         unsigned int segments;
1727         unsigned int tx_bytes;
1728         unsigned int i;
1729
1730         orig_nb_pkts = nb_pkts;
1731         if (unlikely(avp->flags & AVP_F_DETACHED)) {
1732                 /* VM live migration in progress */
1733                 /* TODO ... buffer for X packets then drop? */
1734                 txq->errors += nb_pkts;
1735                 return 0;
1736         }
1737
1738         tx_q = avp->tx_q[txq->queue_id];
1739         alloc_q = avp->alloc_q[txq->queue_id];
1740
1741         /* limit the number of transmitted packets to the max burst size */
1742         if (unlikely(nb_pkts > AVP_MAX_TX_BURST))
1743                 nb_pkts = AVP_MAX_TX_BURST;
1744
1745         /* determine how many buffers are available to copy into */
1746         avail = avp_fifo_count(alloc_q);
1747         if (unlikely(avail > (AVP_MAX_TX_BURST *
1748                               RTE_AVP_MAX_MBUF_SEGMENTS)))
1749                 avail = AVP_MAX_TX_BURST * RTE_AVP_MAX_MBUF_SEGMENTS;
1750
1751         /* determine how many slots are available in the transmit queue */
1752         count = avp_fifo_free_count(tx_q);
1753
1754         /* determine how many packets can be sent */
1755         nb_pkts = RTE_MIN(count, nb_pkts);
1756
1757         /* determine how many packets will fit in the available buffers */
1758         count = 0;
1759         segments = 0;
1760         for (i = 0; i < nb_pkts; i++) {
1761                 m = tx_pkts[i];
1762                 if (likely(i < (unsigned int)nb_pkts - 1)) {
1763                         /* prefetch next entry while processing this one */
1764                         rte_prefetch0(tx_pkts[i + 1]);
1765                 }
1766                 required = (rte_pktmbuf_pkt_len(m) + avp->host_mbuf_size - 1) /
1767                         avp->host_mbuf_size;
1768
1769                 if (unlikely((required == 0) ||
1770                              (required > RTE_AVP_MAX_MBUF_SEGMENTS)))
1771                         break;
1772                 else if (unlikely(required + segments > avail))
1773                         break;
1774                 segments += required;
1775                 count++;
1776         }
1777         nb_pkts = count;
1778
1779         if (unlikely(nb_pkts == 0)) {
1780                 /* no available buffers, or no space on the tx queue */
1781                 txq->errors += orig_nb_pkts;
1782                 return 0;
1783         }
1784
1785         PMD_TX_LOG(DEBUG, "Sending %u packets on Tx queue at %p\n",
1786                    nb_pkts, tx_q);
1787
1788         /* retrieve sufficient send buffers */
1789         n = avp_fifo_get(alloc_q, (void **)&avp_bufs, segments);
1790         if (unlikely(n != segments)) {
1791                 PMD_TX_LOG(DEBUG, "Failed to allocate buffers "
1792                            "n=%u, segments=%u, orig=%u\n",
1793                            n, segments, orig_nb_pkts);
1794                 txq->errors += orig_nb_pkts;
1795                 return 0;
1796         }
1797
1798         tx_bytes = 0;
1799         count = 0;
1800         for (i = 0; i < nb_pkts; i++) {
1801                 /* process each packet to be transmitted */
1802                 m = tx_pkts[i];
1803
1804                 /* determine how many buffers are required for this packet */
1805                 required = (rte_pktmbuf_pkt_len(m) + avp->host_mbuf_size - 1) /
1806                         avp->host_mbuf_size;
1807
1808                 tx_bytes += avp_dev_copy_to_buffers(avp, m,
1809                                                     &avp_bufs[count], required);
1810                 tx_bufs[i] = avp_bufs[count];
1811                 count += required;
1812
1813                 /* free the original mbuf */
1814                 rte_pktmbuf_free(m);
1815         }
1816
1817         txq->packets += nb_pkts;
1818         txq->bytes += tx_bytes;
1819
1820 #ifdef RTE_LIBRTE_AVP_DEBUG_BUFFERS
1821         for (i = 0; i < nb_pkts; i++)
1822                 avp_dev_buffer_sanity_check(avp, tx_bufs[i]);
1823 #endif
1824
1825         /* send the packets */
1826         n = avp_fifo_put(tx_q, (void **)&tx_bufs[0], nb_pkts);
1827         if (unlikely(n != orig_nb_pkts))
1828                 txq->errors += (orig_nb_pkts - n);
1829
1830         return n;
1831 }
1832
1833
1834 static uint16_t
1835 avp_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1836 {
1837         struct avp_queue *txq = (struct avp_queue *)tx_queue;
1838         struct rte_avp_desc *avp_bufs[AVP_MAX_TX_BURST];
1839         struct avp_dev *avp = txq->avp;
1840         struct rte_avp_desc *pkt_buf;
1841         struct rte_avp_fifo *alloc_q;
1842         struct rte_avp_fifo *tx_q;
1843         unsigned int count, avail, n;
1844         struct rte_mbuf *m;
1845         unsigned int pkt_len;
1846         unsigned int tx_bytes;
1847         char *pkt_data;
1848         unsigned int i;
1849
1850         if (unlikely(avp->flags & AVP_F_DETACHED)) {
1851                 /* VM live migration in progress */
1852                 /* TODO ... buffer for X packets then drop?! */
1853                 txq->errors++;
1854                 return 0;
1855         }
1856
1857         tx_q = avp->tx_q[txq->queue_id];
1858         alloc_q = avp->alloc_q[txq->queue_id];
1859
1860         /* limit the number of transmitted packets to the max burst size */
1861         if (unlikely(nb_pkts > AVP_MAX_TX_BURST))
1862                 nb_pkts = AVP_MAX_TX_BURST;
1863
1864         /* determine how many buffers are available to copy into */
1865         avail = avp_fifo_count(alloc_q);
1866
1867         /* determine how many slots are available in the transmit queue */
1868         count = avp_fifo_free_count(tx_q);
1869
1870         /* determine how many packets can be sent */
1871         count = RTE_MIN(count, avail);
1872         count = RTE_MIN(count, nb_pkts);
1873
1874         if (unlikely(count == 0)) {
1875                 /* no available buffers, or no space on the tx queue */
1876                 txq->errors += nb_pkts;
1877                 return 0;
1878         }
1879
1880         PMD_TX_LOG(DEBUG, "Sending %u packets on Tx queue at %p\n",
1881                    count, tx_q);
1882
1883         /* retrieve sufficient send buffers */
1884         n = avp_fifo_get(alloc_q, (void **)&avp_bufs, count);
1885         if (unlikely(n != count)) {
1886                 txq->errors++;
1887                 return 0;
1888         }
1889
1890         tx_bytes = 0;
1891         for (i = 0; i < count; i++) {
1892                 /* prefetch next entry while processing the current one */
1893                 if (i < count - 1) {
1894                         pkt_buf = avp_dev_translate_buffer(avp,
1895                                                            avp_bufs[i + 1]);
1896                         rte_prefetch0(pkt_buf);
1897                 }
1898
1899                 /* process each packet to be transmitted */
1900                 m = tx_pkts[i];
1901
1902                 /* Adjust pointers for guest addressing */
1903                 pkt_buf = avp_dev_translate_buffer(avp, avp_bufs[i]);
1904                 pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1905                 pkt_len = rte_pktmbuf_pkt_len(m);
1906
1907                 if (unlikely((pkt_len > avp->guest_mbuf_size) ||
1908                                          (pkt_len > avp->host_mbuf_size))) {
1909                         /*
1910                          * application should be using the scattered transmit
1911                          * function; send it truncated to avoid the performance
1912                          * hit of having to manage returning the already
1913                          * allocated buffer to the free list.  This should not
1914                          * happen since the application should have set the
1915                          * max_rx_pkt_len based on its MTU and it should be
1916                          * policing its own packet sizes.
1917                          */
1918                         txq->errors++;
1919                         pkt_len = RTE_MIN(avp->guest_mbuf_size,
1920                                           avp->host_mbuf_size);
1921                 }
1922
1923                 /* copy data out of our mbuf and into the AVP buffer */
1924                 rte_memcpy(pkt_data, rte_pktmbuf_mtod(m, void *), pkt_len);
1925                 pkt_buf->pkt_len = pkt_len;
1926                 pkt_buf->data_len = pkt_len;
1927                 pkt_buf->nb_segs = 1;
1928                 pkt_buf->next = NULL;
1929
1930                 if (m->ol_flags & PKT_TX_VLAN_PKT) {
1931                         pkt_buf->ol_flags |= RTE_AVP_TX_VLAN_PKT;
1932                         pkt_buf->vlan_tci = m->vlan_tci;
1933                 }
1934
1935                 tx_bytes += pkt_len;
1936
1937                 /* free the original mbuf */
1938                 rte_pktmbuf_free(m);
1939         }
1940
1941         txq->packets += count;
1942         txq->bytes += tx_bytes;
1943
1944         /* send the packets */
1945         n = avp_fifo_put(tx_q, (void **)&avp_bufs[0], count);
1946
1947         return n;
1948 }
1949
1950 static void
1951 avp_dev_rx_queue_release(void *rx_queue)
1952 {
1953         struct avp_queue *rxq = (struct avp_queue *)rx_queue;
1954         struct avp_dev *avp = rxq->avp;
1955         struct rte_eth_dev_data *data = avp->dev_data;
1956         unsigned int i;
1957
1958         for (i = 0; i < avp->num_rx_queues; i++) {
1959                 if (data->rx_queues[i] == rxq)
1960                         data->rx_queues[i] = NULL;
1961         }
1962 }
1963
1964 static void
1965 avp_dev_tx_queue_release(void *tx_queue)
1966 {
1967         struct avp_queue *txq = (struct avp_queue *)tx_queue;
1968         struct avp_dev *avp = txq->avp;
1969         struct rte_eth_dev_data *data = avp->dev_data;
1970         unsigned int i;
1971
1972         for (i = 0; i < avp->num_tx_queues; i++) {
1973                 if (data->tx_queues[i] == txq)
1974                         data->tx_queues[i] = NULL;
1975         }
1976 }
1977
1978 static int
1979 avp_dev_configure(struct rte_eth_dev *eth_dev)
1980 {
1981         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1982         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1983         struct rte_avp_device_info *host_info;
1984         struct rte_avp_device_config config;
1985         int mask = 0;
1986         void *addr;
1987         int ret;
1988
1989         rte_spinlock_lock(&avp->lock);
1990         if (avp->flags & AVP_F_DETACHED) {
1991                 PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
1992                 ret = -ENOTSUP;
1993                 goto unlock;
1994         }
1995
1996         addr = pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR].addr;
1997         host_info = (struct rte_avp_device_info *)addr;
1998
1999         /* Setup required number of queues */
2000         _avp_set_queue_counts(eth_dev);
2001
2002         mask = (ETH_VLAN_STRIP_MASK |
2003                 ETH_VLAN_FILTER_MASK |
2004                 ETH_VLAN_EXTEND_MASK);
2005         ret = avp_vlan_offload_set(eth_dev, mask);
2006         if (ret < 0) {
2007                 PMD_DRV_LOG(ERR, "VLAN offload set failed by host, ret=%d\n",
2008                             ret);
2009                 goto unlock;
2010         }
2011
2012         /* update device config */
2013         memset(&config, 0, sizeof(config));
2014         config.device_id = host_info->device_id;
2015         config.driver_type = RTE_AVP_DRIVER_TYPE_DPDK;
2016         config.driver_version = AVP_DPDK_DRIVER_VERSION;
2017         config.features = avp->features;
2018         config.num_tx_queues = avp->num_tx_queues;
2019         config.num_rx_queues = avp->num_rx_queues;
2020
2021         ret = avp_dev_ctrl_set_config(eth_dev, &config);
2022         if (ret < 0) {
2023                 PMD_DRV_LOG(ERR, "Config request failed by host, ret=%d\n",
2024                             ret);
2025                 goto unlock;
2026         }
2027
2028         avp->flags |= AVP_F_CONFIGURED;
2029         ret = 0;
2030
2031 unlock:
2032         rte_spinlock_unlock(&avp->lock);
2033         return ret;
2034 }
2035
2036 static int
2037 avp_dev_start(struct rte_eth_dev *eth_dev)
2038 {
2039         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2040         int ret;
2041
2042         rte_spinlock_lock(&avp->lock);
2043         if (avp->flags & AVP_F_DETACHED) {
2044                 PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
2045                 ret = -ENOTSUP;
2046                 goto unlock;
2047         }
2048
2049         /* update link state */
2050         ret = avp_dev_ctrl_set_link_state(eth_dev, 1);
2051         if (ret < 0) {
2052                 PMD_DRV_LOG(ERR, "Link state change failed by host, ret=%d\n",
2053                             ret);
2054                 goto unlock;
2055         }
2056
2057         /* remember current link state */
2058         avp->flags |= AVP_F_LINKUP;
2059
2060         ret = 0;
2061
2062 unlock:
2063         rte_spinlock_unlock(&avp->lock);
2064         return ret;
2065 }
2066
2067 static void
2068 avp_dev_stop(struct rte_eth_dev *eth_dev)
2069 {
2070         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2071         int ret;
2072
2073         rte_spinlock_lock(&avp->lock);
2074         if (avp->flags & AVP_F_DETACHED) {
2075                 PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
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 }
2092
2093 static void
2094 avp_dev_close(struct rte_eth_dev *eth_dev)
2095 {
2096         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2097         int ret;
2098
2099         rte_spinlock_lock(&avp->lock);
2100         if (avp->flags & AVP_F_DETACHED) {
2101                 PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
2102                 goto unlock;
2103         }
2104
2105         /* remember current link state */
2106         avp->flags &= ~AVP_F_LINKUP;
2107         avp->flags &= ~AVP_F_CONFIGURED;
2108
2109         ret = avp_dev_disable_interrupts(eth_dev);
2110         if (ret < 0) {
2111                 PMD_DRV_LOG(ERR, "Failed to disable interrupts\n");
2112                 /* continue */
2113         }
2114
2115         /* update device state */
2116         ret = avp_dev_ctrl_shutdown(eth_dev);
2117         if (ret < 0) {
2118                 PMD_DRV_LOG(ERR, "Device shutdown failed by host, ret=%d\n",
2119                             ret);
2120                 /* continue */
2121         }
2122
2123 unlock:
2124         rte_spinlock_unlock(&avp->lock);
2125 }
2126
2127 static int
2128 avp_dev_link_update(struct rte_eth_dev *eth_dev,
2129                                         __rte_unused int wait_to_complete)
2130 {
2131         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2132         struct rte_eth_link *link = &eth_dev->data->dev_link;
2133
2134         link->link_speed = ETH_SPEED_NUM_10G;
2135         link->link_duplex = ETH_LINK_FULL_DUPLEX;
2136         link->link_status = !!(avp->flags & AVP_F_LINKUP);
2137
2138         return -1;
2139 }
2140
2141 static void
2142 avp_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
2143 {
2144         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2145
2146         rte_spinlock_lock(&avp->lock);
2147         if ((avp->flags & AVP_F_PROMISC) == 0) {
2148                 avp->flags |= AVP_F_PROMISC;
2149                 PMD_DRV_LOG(DEBUG, "Promiscuous mode enabled on %u\n",
2150                             eth_dev->data->port_id);
2151         }
2152         rte_spinlock_unlock(&avp->lock);
2153 }
2154
2155 static void
2156 avp_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
2157 {
2158         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2159
2160         rte_spinlock_lock(&avp->lock);
2161         if ((avp->flags & AVP_F_PROMISC) != 0) {
2162                 avp->flags &= ~AVP_F_PROMISC;
2163                 PMD_DRV_LOG(DEBUG, "Promiscuous mode disabled on %u\n",
2164                             eth_dev->data->port_id);
2165         }
2166         rte_spinlock_unlock(&avp->lock);
2167 }
2168
2169 static void
2170 avp_dev_info_get(struct rte_eth_dev *eth_dev,
2171                  struct rte_eth_dev_info *dev_info)
2172 {
2173         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2174
2175         dev_info->max_rx_queues = avp->max_rx_queues;
2176         dev_info->max_tx_queues = avp->max_tx_queues;
2177         dev_info->min_rx_bufsize = AVP_MIN_RX_BUFSIZE;
2178         dev_info->max_rx_pktlen = avp->max_rx_pkt_len;
2179         dev_info->max_mac_addrs = AVP_MAX_MAC_ADDRS;
2180         if (avp->host_features & RTE_AVP_FEATURE_VLAN_OFFLOAD) {
2181                 dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2182                 dev_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
2183         }
2184 }
2185
2186 static int
2187 avp_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
2188 {
2189         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2190         struct rte_eth_conf *dev_conf = &eth_dev->data->dev_conf;
2191         uint64_t offloads = dev_conf->rxmode.offloads;
2192
2193         if (mask & ETH_VLAN_STRIP_MASK) {
2194                 if (avp->host_features & RTE_AVP_FEATURE_VLAN_OFFLOAD) {
2195                         if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2196                                 avp->features |= RTE_AVP_FEATURE_VLAN_OFFLOAD;
2197                         else
2198                                 avp->features &= ~RTE_AVP_FEATURE_VLAN_OFFLOAD;
2199                 } else {
2200                         PMD_DRV_LOG(ERR, "VLAN strip offload not supported\n");
2201                 }
2202         }
2203
2204         if (mask & ETH_VLAN_FILTER_MASK) {
2205                 if (offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2206                         PMD_DRV_LOG(ERR, "VLAN filter offload not supported\n");
2207         }
2208
2209         if (mask & ETH_VLAN_EXTEND_MASK) {
2210                 if (offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)
2211                         PMD_DRV_LOG(ERR, "VLAN extend offload not supported\n");
2212         }
2213
2214         return 0;
2215 }
2216
2217 static int
2218 avp_dev_stats_get(struct rte_eth_dev *eth_dev, struct rte_eth_stats *stats)
2219 {
2220         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2221         unsigned int i;
2222
2223         for (i = 0; i < avp->num_rx_queues; i++) {
2224                 struct avp_queue *rxq = avp->dev_data->rx_queues[i];
2225
2226                 if (rxq) {
2227                         stats->ipackets += rxq->packets;
2228                         stats->ibytes += rxq->bytes;
2229                         stats->ierrors += rxq->errors;
2230
2231                         stats->q_ipackets[i] += rxq->packets;
2232                         stats->q_ibytes[i] += rxq->bytes;
2233                         stats->q_errors[i] += rxq->errors;
2234                 }
2235         }
2236
2237         for (i = 0; i < avp->num_tx_queues; i++) {
2238                 struct avp_queue *txq = avp->dev_data->tx_queues[i];
2239
2240                 if (txq) {
2241                         stats->opackets += txq->packets;
2242                         stats->obytes += txq->bytes;
2243                         stats->oerrors += txq->errors;
2244
2245                         stats->q_opackets[i] += txq->packets;
2246                         stats->q_obytes[i] += txq->bytes;
2247                         stats->q_errors[i] += txq->errors;
2248                 }
2249         }
2250
2251         return 0;
2252 }
2253
2254 static void
2255 avp_dev_stats_reset(struct rte_eth_dev *eth_dev)
2256 {
2257         struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2258         unsigned int i;
2259
2260         for (i = 0; i < avp->num_rx_queues; i++) {
2261                 struct avp_queue *rxq = avp->dev_data->rx_queues[i];
2262
2263                 if (rxq) {
2264                         rxq->bytes = 0;
2265                         rxq->packets = 0;
2266                         rxq->errors = 0;
2267                 }
2268         }
2269
2270         for (i = 0; i < avp->num_tx_queues; i++) {
2271                 struct avp_queue *txq = avp->dev_data->tx_queues[i];
2272
2273                 if (txq) {
2274                         txq->bytes = 0;
2275                         txq->packets = 0;
2276                         txq->errors = 0;
2277                 }
2278         }
2279 }
2280
2281 RTE_PMD_REGISTER_PCI(net_avp, rte_avp_pmd);
2282 RTE_PMD_REGISTER_PCI_TABLE(net_avp, pci_id_avp_map);
2283
2284 RTE_INIT(avp_init_log);
2285 static void
2286 avp_init_log(void)
2287 {
2288         avp_logtype_driver = rte_log_register("pmd.net.avp.driver");
2289         if (avp_logtype_driver >= 0)
2290                 rte_log_set_level(avp_logtype_driver, RTE_LOG_NOTICE);
2291 }