drivers/net/ena/ena_ethdev.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright (c) 2015-2020 Amazon.com, Inc. or its affiliates.
   3  * All rights reserved.
   4  */
   5
   6 #include <rte_string_fns.h>
   7 #include <rte_ether.h>
   8 #include <ethdev_driver.h>
   9 #include <ethdev_pci.h>
  10 #include <rte_tcp.h>
  11 #include <rte_atomic.h>
  12 #include <rte_dev.h>
  13 #include <rte_errno.h>
  14 #include <rte_version.h>
  15 #include <rte_net.h>
  16 #include <rte_kvargs.h>
  17
  18 #include "ena_ethdev.h"
  19 #include "ena_logs.h"
  20 #include "ena_platform.h"
  21 #include "ena_com.h"
  22 #include "ena_eth_com.h"
  23
  24 #include <ena_common_defs.h>
  25 #include <ena_regs_defs.h>
  26 #include <ena_admin_defs.h>
  27 #include <ena_eth_io_defs.h>
  28
  29 #define DRV_MODULE_VER_MAJOR    2
  30 #define DRV_MODULE_VER_MINOR    2
  31 #define DRV_MODULE_VER_SUBMINOR 1
  32
  33 #define ENA_IO_TXQ_IDX(q)       (2 * (q))
  34 #define ENA_IO_RXQ_IDX(q)       (2 * (q) + 1)
  35 /*reverse version of ENA_IO_RXQ_IDX*/
  36 #define ENA_IO_RXQ_IDX_REV(q)   ((q - 1) / 2)
  37
  38 #define __MERGE_64B_H_L(h, l) (((uint64_t)h << 32) | l)
  39 #define TEST_BIT(val, bit_shift) (val & (1UL << bit_shift))
  40
  41 #define GET_L4_HDR_LEN(mbuf)                                    \
  42         ((rte_pktmbuf_mtod_offset(mbuf, struct rte_tcp_hdr *,   \
  43                 mbuf->l3_len + mbuf->l2_len)->data_off) >> 4)
  44
  45 #define ENA_RX_RSS_TABLE_LOG_SIZE  7
  46 #define ENA_RX_RSS_TABLE_SIZE   (1 << ENA_RX_RSS_TABLE_LOG_SIZE)
  47 #define ENA_HASH_KEY_SIZE       40
  48 #define ETH_GSTRING_LEN 32
  49
  50 #define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
  51
  52 #define ENA_MIN_RING_DESC       128
  53
  54 enum ethtool_stringset {
  55         ETH_SS_TEST             = 0,
  56         ETH_SS_STATS,
  57 };
  58
  59 struct ena_stats {
  60         char name[ETH_GSTRING_LEN];
  61         int stat_offset;
  62 };
  63
  64 #define ENA_STAT_ENTRY(stat, stat_type) { \
  65         .name = #stat, \
  66         .stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
  67 }
  68
  69 #define ENA_STAT_RX_ENTRY(stat) \
  70         ENA_STAT_ENTRY(stat, rx)
  71
  72 #define ENA_STAT_TX_ENTRY(stat) \
  73         ENA_STAT_ENTRY(stat, tx)
  74
  75 #define ENA_STAT_ENI_ENTRY(stat) \
  76         ENA_STAT_ENTRY(stat, eni)
  77
  78 #define ENA_STAT_GLOBAL_ENTRY(stat) \
  79         ENA_STAT_ENTRY(stat, dev)
  80
  81 /* Device arguments */
  82 #define ENA_DEVARG_LARGE_LLQ_HDR "large_llq_hdr"
  83
  84 /*
  85  * Each rte_memzone should have unique name.
  86  * To satisfy it, count number of allocation and add it to name.
  87  */
  88 rte_atomic32_t ena_alloc_cnt;
  89
  90 static const struct ena_stats ena_stats_global_strings[] = {
  91         ENA_STAT_GLOBAL_ENTRY(wd_expired),
  92         ENA_STAT_GLOBAL_ENTRY(dev_start),
  93         ENA_STAT_GLOBAL_ENTRY(dev_stop),
  94         ENA_STAT_GLOBAL_ENTRY(tx_drops),
  95 };
  96
  97 static const struct ena_stats ena_stats_eni_strings[] = {
  98         ENA_STAT_ENI_ENTRY(bw_in_allowance_exceeded),
  99         ENA_STAT_ENI_ENTRY(bw_out_allowance_exceeded),
 100         ENA_STAT_ENI_ENTRY(pps_allowance_exceeded),
 101         ENA_STAT_ENI_ENTRY(conntrack_allowance_exceeded),
 102         ENA_STAT_ENI_ENTRY(linklocal_allowance_exceeded),
 103 };
 104
 105 static const struct ena_stats ena_stats_tx_strings[] = {
 106         ENA_STAT_TX_ENTRY(cnt),
 107         ENA_STAT_TX_ENTRY(bytes),
 108         ENA_STAT_TX_ENTRY(prepare_ctx_err),
 109         ENA_STAT_TX_ENTRY(linearize),
 110         ENA_STAT_TX_ENTRY(linearize_failed),
 111         ENA_STAT_TX_ENTRY(tx_poll),
 112         ENA_STAT_TX_ENTRY(doorbells),
 113         ENA_STAT_TX_ENTRY(bad_req_id),
 114         ENA_STAT_TX_ENTRY(available_desc),
 115 };
 116
 117 static const struct ena_stats ena_stats_rx_strings[] = {
 118         ENA_STAT_RX_ENTRY(cnt),
 119         ENA_STAT_RX_ENTRY(bytes),
 120         ENA_STAT_RX_ENTRY(refill_partial),
 121         ENA_STAT_RX_ENTRY(bad_csum),
 122         ENA_STAT_RX_ENTRY(mbuf_alloc_fail),
 123         ENA_STAT_RX_ENTRY(bad_desc_num),
 124         ENA_STAT_RX_ENTRY(bad_req_id),
 125 };
 126
 127 #define ENA_STATS_ARRAY_GLOBAL  ARRAY_SIZE(ena_stats_global_strings)
 128 #define ENA_STATS_ARRAY_ENI     ARRAY_SIZE(ena_stats_eni_strings)
 129 #define ENA_STATS_ARRAY_TX      ARRAY_SIZE(ena_stats_tx_strings)
 130 #define ENA_STATS_ARRAY_RX      ARRAY_SIZE(ena_stats_rx_strings)
 131
 132 #define QUEUE_OFFLOADS (DEV_TX_OFFLOAD_TCP_CKSUM |\
 133                         DEV_TX_OFFLOAD_UDP_CKSUM |\
 134                         DEV_TX_OFFLOAD_IPV4_CKSUM |\
 135                         DEV_TX_OFFLOAD_TCP_TSO)
 136 #define MBUF_OFFLOADS (PKT_TX_L4_MASK |\
 137                        PKT_TX_IP_CKSUM |\
 138                        PKT_TX_TCP_SEG)
 139
 140 /** Vendor ID used by Amazon devices */
 141 #define PCI_VENDOR_ID_AMAZON 0x1D0F
 142 /** Amazon devices */
 143 #define PCI_DEVICE_ID_ENA_VF            0xEC20
 144 #define PCI_DEVICE_ID_ENA_VF_RSERV0     0xEC21
 145
 146 #define ENA_TX_OFFLOAD_MASK     (\
 147         PKT_TX_L4_MASK |         \
 148         PKT_TX_IPV6 |            \
 149         PKT_TX_IPV4 |            \
 150         PKT_TX_IP_CKSUM |        \
 151         PKT_TX_TCP_SEG)
 152
 153 #define ENA_TX_OFFLOAD_NOTSUP_MASK      \
 154         (PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
 155
 156 static const struct rte_pci_id pci_id_ena_map[] = {
 157         { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
 158         { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF_RSERV0) },
 159         { .device_id = 0 },
 160 };
 161
 162 static struct ena_aenq_handlers aenq_handlers;
 163
 164 static int ena_device_init(struct ena_com_dev *ena_dev,
 165                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
 166                            bool *wd_state);
 167 static int ena_dev_configure(struct rte_eth_dev *dev);
 168 static void ena_tx_map_mbuf(struct ena_ring *tx_ring,
 169         struct ena_tx_buffer *tx_info,
 170         struct rte_mbuf *mbuf,
 171         void **push_header,
 172         uint16_t *header_len);
 173 static int ena_xmit_mbuf(struct ena_ring *tx_ring, struct rte_mbuf *mbuf);
 174 static void ena_tx_cleanup(struct ena_ring *tx_ring);
 175 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 176                                   uint16_t nb_pkts);
 177 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 178                 uint16_t nb_pkts);
 179 static int ena_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 180                               uint16_t nb_desc, unsigned int socket_id,
 181                               const struct rte_eth_txconf *tx_conf);
 182 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 183                               uint16_t nb_desc, unsigned int socket_id,
 184                               const struct rte_eth_rxconf *rx_conf,
 185                               struct rte_mempool *mp);
 186 static inline void ena_init_rx_mbuf(struct rte_mbuf *mbuf, uint16_t len);
 187 static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
 188                                     struct ena_com_rx_buf_info *ena_bufs,
 189                                     uint32_t descs,
 190                                     uint16_t *next_to_clean,
 191                                     uint8_t offset);
 192 static uint16_t eth_ena_recv_pkts(void *rx_queue,
 193                                   struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
 194 static int ena_add_single_rx_desc(struct ena_com_io_sq *io_sq,
 195                                   struct rte_mbuf *mbuf, uint16_t id);
 196 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
 197 static void ena_init_rings(struct ena_adapter *adapter,
 198                            bool disable_meta_caching);
 199 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
 200 static int ena_start(struct rte_eth_dev *dev);
 201 static int ena_stop(struct rte_eth_dev *dev);
 202 static int ena_close(struct rte_eth_dev *dev);
 203 static int ena_dev_reset(struct rte_eth_dev *dev);
 204 static int ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
 205 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
 206 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
 207 static void ena_rx_queue_release(void *queue);
 208 static void ena_tx_queue_release(void *queue);
 209 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
 210 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
 211 static int ena_link_update(struct rte_eth_dev *dev,
 212                            int wait_to_complete);
 213 static int ena_create_io_queue(struct ena_ring *ring);
 214 static void ena_queue_stop(struct ena_ring *ring);
 215 static void ena_queue_stop_all(struct rte_eth_dev *dev,
 216                               enum ena_ring_type ring_type);
 217 static int ena_queue_start(struct ena_ring *ring);
 218 static int ena_queue_start_all(struct rte_eth_dev *dev,
 219                                enum ena_ring_type ring_type);
 220 static void ena_stats_restart(struct rte_eth_dev *dev);
 221 static int ena_infos_get(struct rte_eth_dev *dev,
 222                          struct rte_eth_dev_info *dev_info);
 223 static int ena_rss_reta_update(struct rte_eth_dev *dev,
 224                                struct rte_eth_rss_reta_entry64 *reta_conf,
 225                                uint16_t reta_size);
 226 static int ena_rss_reta_query(struct rte_eth_dev *dev,
 227                               struct rte_eth_rss_reta_entry64 *reta_conf,
 228                               uint16_t reta_size);
 229 static void ena_interrupt_handler_rte(void *cb_arg);
 230 static void ena_timer_wd_callback(struct rte_timer *timer, void *arg);
 231 static void ena_destroy_device(struct rte_eth_dev *eth_dev);
 232 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev);
 233 static int ena_xstats_get_names(struct rte_eth_dev *dev,
 234                                 struct rte_eth_xstat_name *xstats_names,
 235                                 unsigned int n);
 236 static int ena_xstats_get(struct rte_eth_dev *dev,
 237                           struct rte_eth_xstat *stats,
 238                           unsigned int n);
 239 static int ena_xstats_get_by_id(struct rte_eth_dev *dev,
 240                                 const uint64_t *ids,
 241                                 uint64_t *values,
 242                                 unsigned int n);
 243 static int ena_process_bool_devarg(const char *key,
 244                                    const char *value,
 245                                    void *opaque);
 246 static int ena_parse_devargs(struct ena_adapter *adapter,
 247                              struct rte_devargs *devargs);
 248 static int ena_copy_eni_stats(struct ena_adapter *adapter);
 249
 250 static const struct eth_dev_ops ena_dev_ops = {
 251         .dev_configure        = ena_dev_configure,
 252         .dev_infos_get        = ena_infos_get,
 253         .rx_queue_setup       = ena_rx_queue_setup,
 254         .tx_queue_setup       = ena_tx_queue_setup,
 255         .dev_start            = ena_start,
 256         .dev_stop             = ena_stop,
 257         .link_update          = ena_link_update,
 258         .stats_get            = ena_stats_get,
 259         .xstats_get_names     = ena_xstats_get_names,
 260         .xstats_get           = ena_xstats_get,
 261         .xstats_get_by_id     = ena_xstats_get_by_id,
 262         .mtu_set              = ena_mtu_set,
 263         .rx_queue_release     = ena_rx_queue_release,
 264         .tx_queue_release     = ena_tx_queue_release,
 265         .dev_close            = ena_close,
 266         .dev_reset            = ena_dev_reset,
 267         .reta_update          = ena_rss_reta_update,
 268         .reta_query           = ena_rss_reta_query,
 269 };
 270
 271 void ena_rss_key_fill(void *key, size_t size)
 272 {
 273         static bool key_generated;
 274         static uint8_t default_key[ENA_HASH_KEY_SIZE];
 275         size_t i;
 276
 277         RTE_ASSERT(size <= ENA_HASH_KEY_SIZE);
 278
 279         if (!key_generated) {
 280                 for (i = 0; i < ENA_HASH_KEY_SIZE; ++i)
 281                         default_key[i] = rte_rand() & 0xff;
 282                 key_generated = true;
 283         }
 284
 285         rte_memcpy(key, default_key, size);
 286 }
 287
 288 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
 289                                        struct ena_com_rx_ctx *ena_rx_ctx)
 290 {
 291         uint64_t ol_flags = 0;
 292         uint32_t packet_type = 0;
 293
 294         if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
 295                 packet_type |= RTE_PTYPE_L4_TCP;
 296         else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
 297                 packet_type |= RTE_PTYPE_L4_UDP;
 298
 299         if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) {
 300                 packet_type |= RTE_PTYPE_L3_IPV4;
 301                 if (unlikely(ena_rx_ctx->l3_csum_err))
 302                         ol_flags |= PKT_RX_IP_CKSUM_BAD;
 303                 else
 304                         ol_flags |= PKT_RX_IP_CKSUM_GOOD;
 305         } else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6) {
 306                 packet_type |= RTE_PTYPE_L3_IPV6;
 307         }
 308
 309         if (!ena_rx_ctx->l4_csum_checked || ena_rx_ctx->frag)
 310                 ol_flags |= PKT_RX_L4_CKSUM_UNKNOWN;
 311         else
 312                 if (unlikely(ena_rx_ctx->l4_csum_err))
 313                         ol_flags |= PKT_RX_L4_CKSUM_BAD;
 314                 else
 315                         ol_flags |= PKT_RX_L4_CKSUM_GOOD;
 316
 317         mbuf->ol_flags = ol_flags;
 318         mbuf->packet_type = packet_type;
 319 }
 320
 321 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
 322                                        struct ena_com_tx_ctx *ena_tx_ctx,
 323                                        uint64_t queue_offloads,
 324                                        bool disable_meta_caching)
 325 {
 326         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
 327
 328         if ((mbuf->ol_flags & MBUF_OFFLOADS) &&
 329             (queue_offloads & QUEUE_OFFLOADS)) {
 330                 /* check if TSO is required */
 331                 if ((mbuf->ol_flags & PKT_TX_TCP_SEG) &&
 332                     (queue_offloads & DEV_TX_OFFLOAD_TCP_TSO)) {
 333                         ena_tx_ctx->tso_enable = true;
 334
 335                         ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
 336                 }
 337
 338                 /* check if L3 checksum is needed */
 339                 if ((mbuf->ol_flags & PKT_TX_IP_CKSUM) &&
 340                     (queue_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM))
 341                         ena_tx_ctx->l3_csum_enable = true;
 342
 343                 if (mbuf->ol_flags & PKT_TX_IPV6) {
 344                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
 345                 } else {
 346                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
 347
 348                         /* set don't fragment (DF) flag */
 349                         if (mbuf->packet_type &
 350                                 (RTE_PTYPE_L4_NONFRAG
 351                                  | RTE_PTYPE_INNER_L4_NONFRAG))
 352                                 ena_tx_ctx->df = true;
 353                 }
 354
 355                 /* check if L4 checksum is needed */
 356                 if (((mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) &&
 357                     (queue_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)) {
 358                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
 359                         ena_tx_ctx->l4_csum_enable = true;
 360                 } else if (((mbuf->ol_flags & PKT_TX_L4_MASK) ==
 361                                 PKT_TX_UDP_CKSUM) &&
 362                                 (queue_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)) {
 363                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
 364                         ena_tx_ctx->l4_csum_enable = true;
 365                 } else {
 366                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
 367                         ena_tx_ctx->l4_csum_enable = false;
 368                 }
 369
 370                 ena_meta->mss = mbuf->tso_segsz;
 371                 ena_meta->l3_hdr_len = mbuf->l3_len;
 372                 ena_meta->l3_hdr_offset = mbuf->l2_len;
 373
 374                 ena_tx_ctx->meta_valid = true;
 375         } else if (disable_meta_caching) {
 376                 memset(ena_meta, 0, sizeof(*ena_meta));
 377                 ena_tx_ctx->meta_valid = true;
 378         } else {
 379                 ena_tx_ctx->meta_valid = false;
 380         }
 381 }
 382
 383 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
 384 {
 385         struct ena_tx_buffer *tx_info = NULL;
 386
 387         if (likely(req_id < tx_ring->ring_size)) {
 388                 tx_info = &tx_ring->tx_buffer_info[req_id];
 389                 if (likely(tx_info->mbuf))
 390                         return 0;
 391         }
 392
 393         if (tx_info)
 394                 PMD_DRV_LOG(ERR, "tx_info doesn't have valid mbuf\n");
 395         else
 396                 PMD_DRV_LOG(ERR, "Invalid req_id: %hu\n", req_id);
 397
 398         /* Trigger device reset */
 399         ++tx_ring->tx_stats.bad_req_id;
 400         tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
 401         tx_ring->adapter->trigger_reset = true;
 402         return -EFAULT;
 403 }
 404
 405 static void ena_config_host_info(struct ena_com_dev *ena_dev)
 406 {
 407         struct ena_admin_host_info *host_info;
 408         int rc;
 409
 410         /* Allocate only the host info */
 411         rc = ena_com_allocate_host_info(ena_dev);
 412         if (rc) {
 413                 PMD_DRV_LOG(ERR, "Cannot allocate host info\n");
 414                 return;
 415         }
 416
 417         host_info = ena_dev->host_attr.host_info;
 418
 419         host_info->os_type = ENA_ADMIN_OS_DPDK;
 420         host_info->kernel_ver = RTE_VERSION;
 421         strlcpy((char *)host_info->kernel_ver_str, rte_version(),
 422                 sizeof(host_info->kernel_ver_str));
 423         host_info->os_dist = RTE_VERSION;
 424         strlcpy((char *)host_info->os_dist_str, rte_version(),
 425                 sizeof(host_info->os_dist_str));
 426         host_info->driver_version =
 427                 (DRV_MODULE_VER_MAJOR) |
 428                 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
 429                 (DRV_MODULE_VER_SUBMINOR <<
 430                         ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
 431         host_info->num_cpus = rte_lcore_count();
 432
 433         host_info->driver_supported_features =
 434                 ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK;
 435
 436         rc = ena_com_set_host_attributes(ena_dev);
 437         if (rc) {
 438                 if (rc == -ENA_COM_UNSUPPORTED)
 439                         PMD_DRV_LOG(WARNING, "Cannot set host attributes\n");
 440                 else
 441                         PMD_DRV_LOG(ERR, "Cannot set host attributes\n");
 442
 443                 goto err;
 444         }
 445
 446         return;
 447
 448 err:
 449         ena_com_delete_host_info(ena_dev);
 450 }
 451
 452 /* This function calculates the number of xstats based on the current config */
 453 static unsigned int ena_xstats_calc_num(struct rte_eth_dev *dev)
 454 {
 455         return ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENI +
 456                 (dev->data->nb_tx_queues * ENA_STATS_ARRAY_TX) +
 457                 (dev->data->nb_rx_queues * ENA_STATS_ARRAY_RX);
 458 }
 459
 460 static void ena_config_debug_area(struct ena_adapter *adapter)
 461 {
 462         u32 debug_area_size;
 463         int rc, ss_count;
 464
 465         ss_count = ena_xstats_calc_num(adapter->rte_dev);
 466
 467         /* allocate 32 bytes for each string and 64bit for the value */
 468         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
 469
 470         rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
 471         if (rc) {
 472                 PMD_DRV_LOG(ERR, "Cannot allocate debug area\n");
 473                 return;
 474         }
 475
 476         rc = ena_com_set_host_attributes(&adapter->ena_dev);
 477         if (rc) {
 478                 if (rc == -ENA_COM_UNSUPPORTED)
 479                         PMD_DRV_LOG(WARNING, "Cannot set host attributes\n");
 480                 else
 481                         PMD_DRV_LOG(ERR, "Cannot set host attributes\n");
 482
 483                 goto err;
 484         }
 485
 486         return;
 487 err:
 488         ena_com_delete_debug_area(&adapter->ena_dev);
 489 }
 490
 491 static int ena_close(struct rte_eth_dev *dev)
 492 {
 493         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 494         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
 495         struct ena_adapter *adapter = dev->data->dev_private;
 496         int ret = 0;
 497
 498         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 499                 return 0;
 500
 501         if (adapter->state == ENA_ADAPTER_STATE_RUNNING)
 502                 ret = ena_stop(dev);
 503         adapter->state = ENA_ADAPTER_STATE_CLOSED;
 504
 505         ena_rx_queue_release_all(dev);
 506         ena_tx_queue_release_all(dev);
 507
 508         rte_free(adapter->drv_stats);
 509         adapter->drv_stats = NULL;
 510
 511         rte_intr_disable(intr_handle);
 512         rte_intr_callback_unregister(intr_handle,
 513                                      ena_interrupt_handler_rte,
 514                                      adapter);
 515
 516         /*
 517          * MAC is not allocated dynamically. Setting NULL should prevent from
 518          * release of the resource in the rte_eth_dev_release_port().
 519          */
 520         dev->data->mac_addrs = NULL;
 521
 522         return ret;
 523 }
 524
 525 static int
 526 ena_dev_reset(struct rte_eth_dev *dev)
 527 {
 528         int rc = 0;
 529
 530         ena_destroy_device(dev);
 531         rc = eth_ena_dev_init(dev);
 532         if (rc)
 533                 PMD_INIT_LOG(CRIT, "Cannot initialize device");
 534
 535         return rc;
 536 }
 537
 538 static int ena_rss_reta_update(struct rte_eth_dev *dev,
 539                                struct rte_eth_rss_reta_entry64 *reta_conf,
 540                                uint16_t reta_size)
 541 {
 542         struct ena_adapter *adapter = dev->data->dev_private;
 543         struct ena_com_dev *ena_dev = &adapter->ena_dev;
 544         int rc, i;
 545         u16 entry_value;
 546         int conf_idx;
 547         int idx;
 548
 549         if ((reta_size == 0) || (reta_conf == NULL))
 550                 return -EINVAL;
 551
 552         if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
 553                 PMD_DRV_LOG(WARNING,
 554                         "indirection table %d is bigger than supported (%d)\n",
 555                         reta_size, ENA_RX_RSS_TABLE_SIZE);
 556                 return -EINVAL;
 557         }
 558
 559         for (i = 0 ; i < reta_size ; i++) {
 560                 /* each reta_conf is for 64 entries.
 561                  * to support 128 we use 2 conf of 64
 562                  */
 563                 conf_idx = i / RTE_RETA_GROUP_SIZE;
 564                 idx = i % RTE_RETA_GROUP_SIZE;
 565                 if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
 566                         entry_value =
 567                                 ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
 568
 569                         rc = ena_com_indirect_table_fill_entry(ena_dev,
 570                                                                i,
 571                                                                entry_value);
 572                         if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
 573                                 PMD_DRV_LOG(ERR,
 574                                         "Cannot fill indirect table\n");
 575                                 return rc;
 576                         }
 577                 }
 578         }
 579
 580         rte_spinlock_lock(&adapter->admin_lock);
 581         rc = ena_com_indirect_table_set(ena_dev);
 582         rte_spinlock_unlock(&adapter->admin_lock);
 583         if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
 584                 PMD_DRV_LOG(ERR, "Cannot flush the indirect table\n");
 585                 return rc;
 586         }
 587
 588         PMD_DRV_LOG(DEBUG, "%s(): RSS configured %d entries  for port %d\n",
 589                 __func__, reta_size, adapter->rte_dev->data->port_id);
 590
 591         return 0;
 592 }
 593
 594 /* Query redirection table. */
 595 static int ena_rss_reta_query(struct rte_eth_dev *dev,
 596                               struct rte_eth_rss_reta_entry64 *reta_conf,
 597                               uint16_t reta_size)
 598 {
 599         struct ena_adapter *adapter = dev->data->dev_private;
 600         struct ena_com_dev *ena_dev = &adapter->ena_dev;
 601         int rc;
 602         int i;
 603         u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
 604         int reta_conf_idx;
 605         int reta_idx;
 606
 607         if (reta_size == 0 || reta_conf == NULL ||
 608             (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
 609                 return -EINVAL;
 610
 611         rte_spinlock_lock(&adapter->admin_lock);
 612         rc = ena_com_indirect_table_get(ena_dev, indirect_table);
 613         rte_spinlock_unlock(&adapter->admin_lock);
 614         if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
 615                 PMD_DRV_LOG(ERR, "cannot get indirect table\n");
 616                 return -ENOTSUP;
 617         }
 618
 619         for (i = 0 ; i < reta_size ; i++) {
 620                 reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
 621                 reta_idx = i % RTE_RETA_GROUP_SIZE;
 622                 if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
 623                         reta_conf[reta_conf_idx].reta[reta_idx] =
 624                                 ENA_IO_RXQ_IDX_REV(indirect_table[i]);
 625         }
 626
 627         return 0;
 628 }
 629
 630 static int ena_rss_init_default(struct ena_adapter *adapter)
 631 {
 632         struct ena_com_dev *ena_dev = &adapter->ena_dev;
 633         uint16_t nb_rx_queues = adapter->rte_dev->data->nb_rx_queues;
 634         int rc, i;
 635         u32 val;
 636
 637         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
 638         if (unlikely(rc)) {
 639                 PMD_DRV_LOG(ERR, "Cannot init indirect table\n");
 640                 goto err_rss_init;
 641         }
 642
 643         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
 644                 val = i % nb_rx_queues;
 645                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
 646                                                        ENA_IO_RXQ_IDX(val));
 647                 if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
 648                         PMD_DRV_LOG(ERR, "Cannot fill indirect table\n");
 649                         goto err_fill_indir;
 650                 }
 651         }
 652
 653         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
 654                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
 655         if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
 656                 PMD_DRV_LOG(INFO, "Cannot fill hash function\n");
 657                 goto err_fill_indir;
 658         }
 659
 660         rc = ena_com_set_default_hash_ctrl(ena_dev);
 661         if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
 662                 PMD_DRV_LOG(INFO, "Cannot fill hash control\n");
 663                 goto err_fill_indir;
 664         }
 665
 666         rc = ena_com_indirect_table_set(ena_dev);
 667         if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
 668                 PMD_DRV_LOG(ERR, "Cannot flush the indirect table\n");
 669                 goto err_fill_indir;
 670         }
 671         PMD_DRV_LOG(DEBUG, "RSS configured for port %d\n",
 672                 adapter->rte_dev->data->port_id);
 673
 674         return 0;
 675
 676 err_fill_indir:
 677         ena_com_rss_destroy(ena_dev);
 678 err_rss_init:
 679
 680         return rc;
 681 }
 682
 683 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
 684 {
 685         struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
 686         int nb_queues = dev->data->nb_rx_queues;
 687         int i;
 688
 689         for (i = 0; i < nb_queues; i++)
 690                 ena_rx_queue_release(queues[i]);
 691 }
 692
 693 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
 694 {
 695         struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
 696         int nb_queues = dev->data->nb_tx_queues;
 697         int i;
 698
 699         for (i = 0; i < nb_queues; i++)
 700                 ena_tx_queue_release(queues[i]);
 701 }
 702
 703 static void ena_rx_queue_release(void *queue)
 704 {
 705         struct ena_ring *ring = (struct ena_ring *)queue;
 706
 707         /* Free ring resources */
 708         if (ring->rx_buffer_info)
 709                 rte_free(ring->rx_buffer_info);
 710         ring->rx_buffer_info = NULL;
 711
 712         if (ring->rx_refill_buffer)
 713                 rte_free(ring->rx_refill_buffer);
 714         ring->rx_refill_buffer = NULL;
 715
 716         if (ring->empty_rx_reqs)
 717                 rte_free(ring->empty_rx_reqs);
 718         ring->empty_rx_reqs = NULL;
 719
 720         ring->configured = 0;
 721
 722         PMD_DRV_LOG(NOTICE, "RX Queue %d:%d released\n",
 723                 ring->port_id, ring->id);
 724 }
 725
 726 static void ena_tx_queue_release(void *queue)
 727 {
 728         struct ena_ring *ring = (struct ena_ring *)queue;
 729
 730         /* Free ring resources */
 731         if (ring->push_buf_intermediate_buf)
 732                 rte_free(ring->push_buf_intermediate_buf);
 733
 734         if (ring->tx_buffer_info)
 735                 rte_free(ring->tx_buffer_info);
 736
 737         if (ring->empty_tx_reqs)
 738                 rte_free(ring->empty_tx_reqs);
 739
 740         ring->empty_tx_reqs = NULL;
 741         ring->tx_buffer_info = NULL;
 742         ring->push_buf_intermediate_buf = NULL;
 743
 744         ring->configured = 0;
 745
 746         PMD_DRV_LOG(NOTICE, "TX Queue %d:%d released\n",
 747                 ring->port_id, ring->id);
 748 }
 749
 750 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
 751 {
 752         unsigned int i;
 753
 754         for (i = 0; i < ring->ring_size; ++i) {
 755                 struct ena_rx_buffer *rx_info = &ring->rx_buffer_info[i];
 756                 if (rx_info->mbuf) {
 757                         rte_mbuf_raw_free(rx_info->mbuf);
 758                         rx_info->mbuf = NULL;
 759                 }
 760         }
 761 }
 762
 763 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
 764 {
 765         unsigned int i;
 766
 767         for (i = 0; i < ring->ring_size; ++i) {
 768                 struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
 769
 770                 if (tx_buf->mbuf)
 771                         rte_pktmbuf_free(tx_buf->mbuf);
 772         }
 773 }
 774
 775 static int ena_link_update(struct rte_eth_dev *dev,
 776                            __rte_unused int wait_to_complete)
 777 {
 778         struct rte_eth_link *link = &dev->data->dev_link;
 779         struct ena_adapter *adapter = dev->data->dev_private;
 780
 781         link->link_status = adapter->link_status ? ETH_LINK_UP : ETH_LINK_DOWN;
 782         link->link_speed = ETH_SPEED_NUM_NONE;
 783         link->link_duplex = ETH_LINK_FULL_DUPLEX;
 784
 785         return 0;
 786 }
 787
 788 static int ena_queue_start_all(struct rte_eth_dev *dev,
 789                                enum ena_ring_type ring_type)
 790 {
 791         struct ena_adapter *adapter = dev->data->dev_private;
 792         struct ena_ring *queues = NULL;
 793         int nb_queues;
 794         int i = 0;
 795         int rc = 0;
 796
 797         if (ring_type == ENA_RING_TYPE_RX) {
 798                 queues = adapter->rx_ring;
 799                 nb_queues = dev->data->nb_rx_queues;
 800         } else {
 801                 queues = adapter->tx_ring;
 802                 nb_queues = dev->data->nb_tx_queues;
 803         }
 804         for (i = 0; i < nb_queues; i++) {
 805                 if (queues[i].configured) {
 806                         if (ring_type == ENA_RING_TYPE_RX) {
 807                                 ena_assert_msg(
 808                                         dev->data->rx_queues[i] == &queues[i],
 809                                         "Inconsistent state of rx queues\n");
 810                         } else {
 811                                 ena_assert_msg(
 812                                         dev->data->tx_queues[i] == &queues[i],
 813                                         "Inconsistent state of tx queues\n");
 814                         }
 815
 816                         rc = ena_queue_start(&queues[i]);
 817
 818                         if (rc) {
 819                                 PMD_INIT_LOG(ERR,
 820                                              "failed to start queue %d type(%d)",
 821                                              i, ring_type);
 822                                 goto err;
 823                         }
 824                 }
 825         }
 826
 827         return 0;
 828
 829 err:
 830         while (i--)
 831                 if (queues[i].configured)
 832                         ena_queue_stop(&queues[i]);
 833
 834         return rc;
 835 }
 836
 837 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
 838 {
 839         uint32_t max_frame_len = adapter->max_mtu;
 840
 841         if (adapter->rte_eth_dev_data->dev_conf.rxmode.offloads &
 842             DEV_RX_OFFLOAD_JUMBO_FRAME)
 843                 max_frame_len =
 844                         adapter->rte_eth_dev_data->dev_conf.rxmode.max_rx_pkt_len;
 845
 846         return max_frame_len;
 847 }
 848
 849 static int ena_check_valid_conf(struct ena_adapter *adapter)
 850 {
 851         uint32_t max_frame_len = ena_get_mtu_conf(adapter);
 852
 853         if (max_frame_len > adapter->max_mtu || max_frame_len < ENA_MIN_MTU) {
 854                 PMD_INIT_LOG(ERR, "Unsupported MTU of %d. "
 855                                   "max mtu: %d, min mtu: %d",
 856                              max_frame_len, adapter->max_mtu, ENA_MIN_MTU);
 857                 return ENA_COM_UNSUPPORTED;
 858         }
 859
 860         return 0;
 861 }
 862
 863 static int
 864 ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx,
 865                        bool use_large_llq_hdr)
 866 {
 867         struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
 868         struct ena_com_dev *ena_dev = ctx->ena_dev;
 869         uint32_t max_tx_queue_size;
 870         uint32_t max_rx_queue_size;
 871
 872         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
 873                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
 874                         &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
 875                 max_rx_queue_size = RTE_MIN(max_queue_ext->max_rx_cq_depth,
 876                         max_queue_ext->max_rx_sq_depth);
 877                 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
 878
 879                 if (ena_dev->tx_mem_queue_type ==
 880                     ENA_ADMIN_PLACEMENT_POLICY_DEV) {
 881                         max_tx_queue_size = RTE_MIN(max_tx_queue_size,
 882                                 llq->max_llq_depth);
 883                 } else {
 884                         max_tx_queue_size = RTE_MIN(max_tx_queue_size,
 885                                 max_queue_ext->max_tx_sq_depth);
 886                 }
 887
 888                 ctx->max_rx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
 889                         max_queue_ext->max_per_packet_rx_descs);
 890                 ctx->max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
 891                         max_queue_ext->max_per_packet_tx_descs);
 892         } else {
 893                 struct ena_admin_queue_feature_desc *max_queues =
 894                         &ctx->get_feat_ctx->max_queues;
 895                 max_rx_queue_size = RTE_MIN(max_queues->max_cq_depth,
 896                         max_queues->max_sq_depth);
 897                 max_tx_queue_size = max_queues->max_cq_depth;
 898
 899                 if (ena_dev->tx_mem_queue_type ==
 900                     ENA_ADMIN_PLACEMENT_POLICY_DEV) {
 901                         max_tx_queue_size = RTE_MIN(max_tx_queue_size,
 902                                 llq->max_llq_depth);
 903                 } else {
 904                         max_tx_queue_size = RTE_MIN(max_tx_queue_size,
 905                                 max_queues->max_sq_depth);
 906                 }
 907
 908                 ctx->max_rx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
 909                         max_queues->max_packet_rx_descs);
 910                 ctx->max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
 911                         max_queues->max_packet_tx_descs);
 912         }
 913
 914         /* Round down to the nearest power of 2 */
 915         max_rx_queue_size = rte_align32prevpow2(max_rx_queue_size);
 916         max_tx_queue_size = rte_align32prevpow2(max_tx_queue_size);
 917
 918         if (use_large_llq_hdr) {
 919                 if ((llq->entry_size_ctrl_supported &
 920                      ENA_ADMIN_LIST_ENTRY_SIZE_256B) &&
 921                     (ena_dev->tx_mem_queue_type ==
 922                      ENA_ADMIN_PLACEMENT_POLICY_DEV)) {
 923                         max_tx_queue_size /= 2;
 924                         PMD_INIT_LOG(INFO,
 925                                 "Forcing large headers and decreasing maximum TX queue size to %d\n",
 926                                 max_tx_queue_size);
 927                 } else {
 928                         PMD_INIT_LOG(ERR,
 929                                 "Forcing large headers failed: LLQ is disabled or device does not support large headers\n");
 930                 }
 931         }
 932
 933         if (unlikely(max_rx_queue_size == 0 || max_tx_queue_size == 0)) {
 934                 PMD_INIT_LOG(ERR, "Invalid queue size");
 935                 return -EFAULT;
 936         }
 937
 938         ctx->max_tx_queue_size = max_tx_queue_size;
 939         ctx->max_rx_queue_size = max_rx_queue_size;
 940
 941         return 0;
 942 }
 943
 944 static void ena_stats_restart(struct rte_eth_dev *dev)
 945 {
 946         struct ena_adapter *adapter = dev->data->dev_private;
 947
 948         rte_atomic64_init(&adapter->drv_stats->ierrors);
 949         rte_atomic64_init(&adapter->drv_stats->oerrors);
 950         rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
 951         adapter->drv_stats->rx_drops = 0;
 952 }
 953
 954 static int ena_stats_get(struct rte_eth_dev *dev,
 955                           struct rte_eth_stats *stats)
 956 {
 957         struct ena_admin_basic_stats ena_stats;
 958         struct ena_adapter *adapter = dev->data->dev_private;
 959         struct ena_com_dev *ena_dev = &adapter->ena_dev;
 960         int rc;
 961         int i;
 962         int max_rings_stats;
 963
 964         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 965                 return -ENOTSUP;
 966
 967         memset(&ena_stats, 0, sizeof(ena_stats));
 968
 969         rte_spinlock_lock(&adapter->admin_lock);
 970         rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
 971         rte_spinlock_unlock(&adapter->admin_lock);
 972         if (unlikely(rc)) {
 973                 PMD_DRV_LOG(ERR, "Could not retrieve statistics from ENA\n");
 974                 return rc;
 975         }
 976
 977         /* Set of basic statistics from ENA */
 978         stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
 979                                           ena_stats.rx_pkts_low);
 980         stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
 981                                           ena_stats.tx_pkts_low);
 982         stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
 983                                         ena_stats.rx_bytes_low);
 984         stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
 985                                         ena_stats.tx_bytes_low);
 986
 987         /* Driver related stats */
 988         stats->imissed = adapter->drv_stats->rx_drops;
 989         stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
 990         stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
 991         stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
 992
 993         max_rings_stats = RTE_MIN(dev->data->nb_rx_queues,
 994                 RTE_ETHDEV_QUEUE_STAT_CNTRS);
 995         for (i = 0; i < max_rings_stats; ++i) {
 996                 struct ena_stats_rx *rx_stats = &adapter->rx_ring[i].rx_stats;
 997
 998                 stats->q_ibytes[i] = rx_stats->bytes;
 999                 stats->q_ipackets[i] = rx_stats->cnt;
1000                 stats->q_errors[i] = rx_stats->bad_desc_num +
1001                         rx_stats->bad_req_id;
1002         }
1003
1004         max_rings_stats = RTE_MIN(dev->data->nb_tx_queues,
1005                 RTE_ETHDEV_QUEUE_STAT_CNTRS);
1006         for (i = 0; i < max_rings_stats; ++i) {
1007                 struct ena_stats_tx *tx_stats = &adapter->tx_ring[i].tx_stats;
1008
1009                 stats->q_obytes[i] = tx_stats->bytes;
1010                 stats->q_opackets[i] = tx_stats->cnt;
1011         }
1012
1013         return 0;
1014 }
1015
1016 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1017 {
1018         struct ena_adapter *adapter;
1019         struct ena_com_dev *ena_dev;
1020         int rc = 0;
1021
1022         ena_assert_msg(dev->data != NULL, "Uninitialized device\n");
1023         ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device\n");
1024         adapter = dev->data->dev_private;
1025
1026         ena_dev = &adapter->ena_dev;
1027         ena_assert_msg(ena_dev != NULL, "Uninitialized device\n");
1028
1029         if (mtu > ena_get_mtu_conf(adapter) || mtu < ENA_MIN_MTU) {
1030                 PMD_DRV_LOG(ERR,
1031                         "Invalid MTU setting. new_mtu: %d "
1032                         "max mtu: %d min mtu: %d\n",
1033                         mtu, ena_get_mtu_conf(adapter), ENA_MIN_MTU);
1034                 return -EINVAL;
1035         }
1036
1037         rc = ena_com_set_dev_mtu(ena_dev, mtu);
1038         if (rc)
1039                 PMD_DRV_LOG(ERR, "Could not set MTU: %d\n", mtu);
1040         else
1041                 PMD_DRV_LOG(NOTICE, "Set MTU: %d\n", mtu);
1042
1043         return rc;
1044 }
1045
1046 static int ena_start(struct rte_eth_dev *dev)
1047 {
1048         struct ena_adapter *adapter = dev->data->dev_private;
1049         uint64_t ticks;
1050         int rc = 0;
1051
1052         rc = ena_check_valid_conf(adapter);
1053         if (rc)
1054                 return rc;
1055
1056         rc = ena_queue_start_all(dev, ENA_RING_TYPE_RX);
1057         if (rc)
1058                 return rc;
1059
1060         rc = ena_queue_start_all(dev, ENA_RING_TYPE_TX);
1061         if (rc)
1062                 goto err_start_tx;
1063
1064         if (adapter->rte_dev->data->dev_conf.rxmode.mq_mode &
1065             ETH_MQ_RX_RSS_FLAG && adapter->rte_dev->data->nb_rx_queues > 0) {
1066                 rc = ena_rss_init_default(adapter);
1067                 if (rc)
1068                         goto err_rss_init;
1069         }
1070
1071         ena_stats_restart(dev);
1072
1073         adapter->timestamp_wd = rte_get_timer_cycles();
1074         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
1075
1076         ticks = rte_get_timer_hz();
1077         rte_timer_reset(&adapter->timer_wd, ticks, PERIODICAL, rte_lcore_id(),
1078                         ena_timer_wd_callback, adapter);
1079
1080         ++adapter->dev_stats.dev_start;
1081         adapter->state = ENA_ADAPTER_STATE_RUNNING;
1082
1083         return 0;
1084
1085 err_rss_init:
1086         ena_queue_stop_all(dev, ENA_RING_TYPE_TX);
1087 err_start_tx:
1088         ena_queue_stop_all(dev, ENA_RING_TYPE_RX);
1089         return rc;
1090 }
1091
1092 static int ena_stop(struct rte_eth_dev *dev)
1093 {
1094         struct ena_adapter *adapter = dev->data->dev_private;
1095         struct ena_com_dev *ena_dev = &adapter->ena_dev;
1096         int rc;
1097
1098         rte_timer_stop_sync(&adapter->timer_wd);
1099         ena_queue_stop_all(dev, ENA_RING_TYPE_TX);
1100         ena_queue_stop_all(dev, ENA_RING_TYPE_RX);
1101
1102         if (adapter->trigger_reset) {
1103                 rc = ena_com_dev_reset(ena_dev, adapter->reset_reason);
1104                 if (rc)
1105                         PMD_DRV_LOG(ERR, "Device reset failed rc=%d\n", rc);
1106         }
1107
1108         ++adapter->dev_stats.dev_stop;
1109         adapter->state = ENA_ADAPTER_STATE_STOPPED;
1110         dev->data->dev_started = 0;
1111
1112         return 0;
1113 }
1114
1115 static int ena_create_io_queue(struct ena_ring *ring)
1116 {
1117         struct ena_adapter *adapter;
1118         struct ena_com_dev *ena_dev;
1119         struct ena_com_create_io_ctx ctx =
1120                 /* policy set to _HOST just to satisfy icc compiler */
1121                 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
1122                   0, 0, 0, 0, 0 };
1123         uint16_t ena_qid;
1124         unsigned int i;
1125         int rc;
1126
1127         adapter = ring->adapter;
1128         ena_dev = &adapter->ena_dev;
1129
1130         if (ring->type == ENA_RING_TYPE_TX) {
1131                 ena_qid = ENA_IO_TXQ_IDX(ring->id);
1132                 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1133                 ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1134                 for (i = 0; i < ring->ring_size; i++)
1135                         ring->empty_tx_reqs[i] = i;
1136         } else {
1137                 ena_qid = ENA_IO_RXQ_IDX(ring->id);
1138                 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1139                 for (i = 0; i < ring->ring_size; i++)
1140                         ring->empty_rx_reqs[i] = i;
1141         }
1142         ctx.queue_size = ring->ring_size;
1143         ctx.qid = ena_qid;
1144         ctx.msix_vector = -1; /* interrupts not used */
1145         ctx.numa_node = ring->numa_socket_id;
1146
1147         rc = ena_com_create_io_queue(ena_dev, &ctx);
1148         if (rc) {
1149                 PMD_DRV_LOG(ERR,
1150                         "failed to create io queue #%d (qid:%d) rc: %d\n",
1151                         ring->id, ena_qid, rc);
1152                 return rc;
1153         }
1154
1155         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1156                                      &ring->ena_com_io_sq,
1157                                      &ring->ena_com_io_cq);
1158         if (rc) {
1159                 PMD_DRV_LOG(ERR,
1160                         "Failed to get io queue handlers. queue num %d rc: %d\n",
1161                         ring->id, rc);
1162                 ena_com_destroy_io_queue(ena_dev, ena_qid);
1163                 return rc;
1164         }
1165
1166         if (ring->type == ENA_RING_TYPE_TX)
1167                 ena_com_update_numa_node(ring->ena_com_io_cq, ctx.numa_node);
1168
1169         return 0;
1170 }
1171
1172 static void ena_queue_stop(struct ena_ring *ring)
1173 {
1174         struct ena_com_dev *ena_dev = &ring->adapter->ena_dev;
1175
1176         if (ring->type == ENA_RING_TYPE_RX) {
1177                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(ring->id));
1178                 ena_rx_queue_release_bufs(ring);
1179         } else {
1180                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(ring->id));
1181                 ena_tx_queue_release_bufs(ring);
1182         }
1183 }
1184
1185 static void ena_queue_stop_all(struct rte_eth_dev *dev,
1186                               enum ena_ring_type ring_type)
1187 {
1188         struct ena_adapter *adapter = dev->data->dev_private;
1189         struct ena_ring *queues = NULL;
1190         uint16_t nb_queues, i;
1191
1192         if (ring_type == ENA_RING_TYPE_RX) {
1193                 queues = adapter->rx_ring;
1194                 nb_queues = dev->data->nb_rx_queues;
1195         } else {
1196                 queues = adapter->tx_ring;
1197                 nb_queues = dev->data->nb_tx_queues;
1198         }
1199
1200         for (i = 0; i < nb_queues; ++i)
1201                 if (queues[i].configured)
1202                         ena_queue_stop(&queues[i]);
1203 }
1204
1205 static int ena_queue_start(struct ena_ring *ring)
1206 {
1207         int rc, bufs_num;
1208
1209         ena_assert_msg(ring->configured == 1,
1210                        "Trying to start unconfigured queue\n");
1211
1212         rc = ena_create_io_queue(ring);
1213         if (rc) {
1214                 PMD_INIT_LOG(ERR, "Failed to create IO queue!");
1215                 return rc;
1216         }
1217
1218         ring->next_to_clean = 0;
1219         ring->next_to_use = 0;
1220
1221         if (ring->type == ENA_RING_TYPE_TX) {
1222                 ring->tx_stats.available_desc =
1223                         ena_com_free_q_entries(ring->ena_com_io_sq);
1224                 return 0;
1225         }
1226
1227         bufs_num = ring->ring_size - 1;
1228         rc = ena_populate_rx_queue(ring, bufs_num);
1229         if (rc != bufs_num) {
1230                 ena_com_destroy_io_queue(&ring->adapter->ena_dev,
1231                                          ENA_IO_RXQ_IDX(ring->id));
1232                 PMD_INIT_LOG(ERR, "Failed to populate rx ring !");
1233                 return ENA_COM_FAULT;
1234         }
1235         /* Flush per-core RX buffers pools cache as they can be used on other
1236          * cores as well.
1237          */
1238         rte_mempool_cache_flush(NULL, ring->mb_pool);
1239
1240         return 0;
1241 }
1242
1243 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
1244                               uint16_t queue_idx,
1245                               uint16_t nb_desc,
1246                               unsigned int socket_id,
1247                               const struct rte_eth_txconf *tx_conf)
1248 {
1249         struct ena_ring *txq = NULL;
1250         struct ena_adapter *adapter = dev->data->dev_private;
1251         unsigned int i;
1252
1253         txq = &adapter->tx_ring[queue_idx];
1254
1255         if (txq->configured) {
1256                 PMD_DRV_LOG(CRIT,
1257                         "API violation. Queue %d is already configured\n",
1258                         queue_idx);
1259                 return ENA_COM_FAULT;
1260         }
1261
1262         if (!rte_is_power_of_2(nb_desc)) {
1263                 PMD_DRV_LOG(ERR,
1264                         "Unsupported size of TX queue: %d is not a power of 2.\n",
1265                         nb_desc);
1266                 return -EINVAL;
1267         }
1268
1269         if (nb_desc > adapter->max_tx_ring_size) {
1270                 PMD_DRV_LOG(ERR,
1271                         "Unsupported size of TX queue (max size: %d)\n",
1272                         adapter->max_tx_ring_size);
1273                 return -EINVAL;
1274         }
1275
1276         if (nb_desc == RTE_ETH_DEV_FALLBACK_TX_RINGSIZE)
1277                 nb_desc = adapter->max_tx_ring_size;
1278
1279         txq->port_id = dev->data->port_id;
1280         txq->next_to_clean = 0;
1281         txq->next_to_use = 0;
1282         txq->ring_size = nb_desc;
1283         txq->size_mask = nb_desc - 1;
1284         txq->numa_socket_id = socket_id;
1285
1286         txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1287                                           sizeof(struct ena_tx_buffer) *
1288                                           txq->ring_size,
1289                                           RTE_CACHE_LINE_SIZE);
1290         if (!txq->tx_buffer_info) {
1291                 PMD_DRV_LOG(ERR, "failed to alloc mem for tx buffer info\n");
1292                 return -ENOMEM;
1293         }
1294
1295         txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1296                                          sizeof(u16) * txq->ring_size,
1297                                          RTE_CACHE_LINE_SIZE);
1298         if (!txq->empty_tx_reqs) {
1299                 PMD_DRV_LOG(ERR, "failed to alloc mem for tx reqs\n");
1300                 rte_free(txq->tx_buffer_info);
1301                 return -ENOMEM;
1302         }
1303
1304         txq->push_buf_intermediate_buf =
1305                 rte_zmalloc("txq->push_buf_intermediate_buf",
1306                             txq->tx_max_header_size,
1307                             RTE_CACHE_LINE_SIZE);
1308         if (!txq->push_buf_intermediate_buf) {
1309                 PMD_DRV_LOG(ERR, "failed to alloc push buff for LLQ\n");
1310                 rte_free(txq->tx_buffer_info);
1311                 rte_free(txq->empty_tx_reqs);
1312                 return -ENOMEM;
1313         }
1314
1315         for (i = 0; i < txq->ring_size; i++)
1316                 txq->empty_tx_reqs[i] = i;
1317
1318         if (tx_conf != NULL) {
1319                 txq->offloads =
1320                         tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1321         }
1322         /* Store pointer to this queue in upper layer */
1323         txq->configured = 1;
1324         dev->data->tx_queues[queue_idx] = txq;
1325
1326         return 0;
1327 }
1328
1329 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1330                               uint16_t queue_idx,
1331                               uint16_t nb_desc,
1332                               unsigned int socket_id,
1333                               __rte_unused const struct rte_eth_rxconf *rx_conf,
1334                               struct rte_mempool *mp)
1335 {
1336         struct ena_adapter *adapter = dev->data->dev_private;
1337         struct ena_ring *rxq = NULL;
1338         size_t buffer_size;
1339         int i;
1340
1341         rxq = &adapter->rx_ring[queue_idx];
1342         if (rxq->configured) {
1343                 PMD_DRV_LOG(CRIT,
1344                         "API violation. Queue %d is already configured\n",
1345                         queue_idx);
1346                 return ENA_COM_FAULT;
1347         }
1348
1349         if (nb_desc == RTE_ETH_DEV_FALLBACK_RX_RINGSIZE)
1350                 nb_desc = adapter->max_rx_ring_size;
1351
1352         if (!rte_is_power_of_2(nb_desc)) {
1353                 PMD_DRV_LOG(ERR,
1354                         "Unsupported size of RX queue: %d is not a power of 2.\n",
1355                         nb_desc);
1356                 return -EINVAL;
1357         }
1358
1359         if (nb_desc > adapter->max_rx_ring_size) {
1360                 PMD_DRV_LOG(ERR,
1361                         "Unsupported size of RX queue (max size: %d)\n",
1362                         adapter->max_rx_ring_size);
1363                 return -EINVAL;
1364         }
1365
1366         /* ENA isn't supporting buffers smaller than 1400 bytes */
1367         buffer_size = rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
1368         if (buffer_size < ENA_RX_BUF_MIN_SIZE) {
1369                 PMD_DRV_LOG(ERR,
1370                         "Unsupported size of RX buffer: %zu (min size: %d)\n",
1371                         buffer_size, ENA_RX_BUF_MIN_SIZE);
1372                 return -EINVAL;
1373         }
1374
1375         rxq->port_id = dev->data->port_id;
1376         rxq->next_to_clean = 0;
1377         rxq->next_to_use = 0;
1378         rxq->ring_size = nb_desc;
1379         rxq->size_mask = nb_desc - 1;
1380         rxq->numa_socket_id = socket_id;
1381         rxq->mb_pool = mp;
1382
1383         rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1384                 sizeof(struct ena_rx_buffer) * nb_desc,
1385                 RTE_CACHE_LINE_SIZE);
1386         if (!rxq->rx_buffer_info) {
1387                 PMD_DRV_LOG(ERR, "failed to alloc mem for rx buffer info\n");
1388                 return -ENOMEM;
1389         }
1390
1391         rxq->rx_refill_buffer = rte_zmalloc("rxq->rx_refill_buffer",
1392                                             sizeof(struct rte_mbuf *) * nb_desc,
1393                                             RTE_CACHE_LINE_SIZE);
1394
1395         if (!rxq->rx_refill_buffer) {
1396                 PMD_DRV_LOG(ERR, "failed to alloc mem for rx refill buffer\n");
1397                 rte_free(rxq->rx_buffer_info);
1398                 rxq->rx_buffer_info = NULL;
1399                 return -ENOMEM;
1400         }
1401
1402         rxq->empty_rx_reqs = rte_zmalloc("rxq->empty_rx_reqs",
1403                                          sizeof(uint16_t) * nb_desc,
1404                                          RTE_CACHE_LINE_SIZE);
1405         if (!rxq->empty_rx_reqs) {
1406                 PMD_DRV_LOG(ERR, "failed to alloc mem for empty rx reqs\n");
1407                 rte_free(rxq->rx_buffer_info);
1408                 rxq->rx_buffer_info = NULL;
1409                 rte_free(rxq->rx_refill_buffer);
1410                 rxq->rx_refill_buffer = NULL;
1411                 return -ENOMEM;
1412         }
1413
1414         for (i = 0; i < nb_desc; i++)
1415                 rxq->empty_rx_reqs[i] = i;
1416
1417         /* Store pointer to this queue in upper layer */
1418         rxq->configured = 1;
1419         dev->data->rx_queues[queue_idx] = rxq;
1420
1421         return 0;
1422 }
1423
1424 static int ena_add_single_rx_desc(struct ena_com_io_sq *io_sq,
1425                                   struct rte_mbuf *mbuf, uint16_t id)
1426 {
1427         struct ena_com_buf ebuf;
1428         int rc;
1429
1430         /* prepare physical address for DMA transaction */
1431         ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
1432         ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1433
1434         /* pass resource to device */
1435         rc = ena_com_add_single_rx_desc(io_sq, &ebuf, id);
1436         if (unlikely(rc != 0))
1437                 PMD_DRV_LOG(WARNING, "failed adding rx desc\n");
1438
1439         return rc;
1440 }
1441
1442 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1443 {
1444         unsigned int i;
1445         int rc;
1446         uint16_t next_to_use = rxq->next_to_use;
1447         uint16_t in_use, req_id;
1448         struct rte_mbuf **mbufs = rxq->rx_refill_buffer;
1449
1450         if (unlikely(!count))
1451                 return 0;
1452
1453         in_use = rxq->ring_size - 1 -
1454                 ena_com_free_q_entries(rxq->ena_com_io_sq);
1455         ena_assert_msg(((in_use + count) < rxq->ring_size),
1456                 "bad ring state\n");
1457
1458         /* get resources for incoming packets */
1459         rc = rte_mempool_get_bulk(rxq->mb_pool, (void **)mbufs, count);
1460         if (unlikely(rc < 0)) {
1461                 rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1462                 ++rxq->rx_stats.mbuf_alloc_fail;
1463                 PMD_RX_LOG(DEBUG, "there are no enough free buffers");
1464                 return 0;
1465         }
1466
1467         for (i = 0; i < count; i++) {
1468                 struct rte_mbuf *mbuf = mbufs[i];
1469                 struct ena_rx_buffer *rx_info;
1470
1471                 if (likely((i + 4) < count))
1472                         rte_prefetch0(mbufs[i + 4]);
1473
1474                 req_id = rxq->empty_rx_reqs[next_to_use];
1475                 rx_info = &rxq->rx_buffer_info[req_id];
1476
1477                 rc = ena_add_single_rx_desc(rxq->ena_com_io_sq, mbuf, req_id);
1478                 if (unlikely(rc != 0))
1479                         break;
1480
1481                 rx_info->mbuf = mbuf;
1482                 next_to_use = ENA_IDX_NEXT_MASKED(next_to_use, rxq->size_mask);
1483         }
1484
1485         if (unlikely(i < count)) {
1486                 PMD_DRV_LOG(WARNING, "refilled rx qid %d with only %d "
1487                         "buffers (from %d)\n", rxq->id, i, count);
1488                 rte_mempool_put_bulk(rxq->mb_pool, (void **)(&mbufs[i]),
1489                                      count - i);
1490                 ++rxq->rx_stats.refill_partial;
1491         }
1492
1493         /* When we submitted free recources to device... */
1494         if (likely(i > 0)) {
1495                 /* ...let HW know that it can fill buffers with data. */
1496                 ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1497
1498                 rxq->next_to_use = next_to_use;
1499         }
1500
1501         return i;
1502 }
1503
1504 static int ena_device_init(struct ena_com_dev *ena_dev,
1505                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
1506                            bool *wd_state)
1507 {
1508         uint32_t aenq_groups;
1509         int rc;
1510         bool readless_supported;
1511
1512         /* Initialize mmio registers */
1513         rc = ena_com_mmio_reg_read_request_init(ena_dev);
1514         if (rc) {
1515                 PMD_DRV_LOG(ERR, "failed to init mmio read less\n");
1516                 return rc;
1517         }
1518
1519         /* The PCIe configuration space revision id indicate if mmio reg
1520          * read is disabled.
1521          */
1522         readless_supported =
1523                 !(((struct rte_pci_device *)ena_dev->dmadev)->id.class_id
1524                                & ENA_MMIO_DISABLE_REG_READ);
1525         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1526
1527         /* reset device */
1528         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
1529         if (rc) {
1530                 PMD_DRV_LOG(ERR, "cannot reset device\n");
1531                 goto err_mmio_read_less;
1532         }
1533
1534         /* check FW version */
1535         rc = ena_com_validate_version(ena_dev);
1536         if (rc) {
1537                 PMD_DRV_LOG(ERR, "device version is too low\n");
1538                 goto err_mmio_read_less;
1539         }
1540
1541         ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1542
1543         /* ENA device administration layer init */
1544         rc = ena_com_admin_init(ena_dev, &aenq_handlers);
1545         if (rc) {
1546                 PMD_DRV_LOG(ERR,
1547                         "cannot initialize ena admin queue with device\n");
1548                 goto err_mmio_read_less;
1549         }
1550
1551         /* To enable the msix interrupts the driver needs to know the number
1552          * of queues. So the driver uses polling mode to retrieve this
1553          * information.
1554          */
1555         ena_com_set_admin_polling_mode(ena_dev, true);
1556
1557         ena_config_host_info(ena_dev);
1558
1559         /* Get Device Attributes and features */
1560         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1561         if (rc) {
1562                 PMD_DRV_LOG(ERR,
1563                         "cannot get attribute for ena device rc= %d\n", rc);
1564                 goto err_admin_init;
1565         }
1566
1567         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
1568                       BIT(ENA_ADMIN_NOTIFICATION) |
1569                       BIT(ENA_ADMIN_KEEP_ALIVE) |
1570                       BIT(ENA_ADMIN_FATAL_ERROR) |
1571                       BIT(ENA_ADMIN_WARNING);
1572
1573         aenq_groups &= get_feat_ctx->aenq.supported_groups;
1574         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
1575         if (rc) {
1576                 PMD_DRV_LOG(ERR, "Cannot configure aenq groups rc: %d\n", rc);
1577                 goto err_admin_init;
1578         }
1579
1580         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
1581
1582         return 0;
1583
1584 err_admin_init:
1585         ena_com_admin_destroy(ena_dev);
1586
1587 err_mmio_read_less:
1588         ena_com_mmio_reg_read_request_destroy(ena_dev);
1589
1590         return rc;
1591 }
1592
1593 static void ena_interrupt_handler_rte(void *cb_arg)
1594 {
1595         struct ena_adapter *adapter = cb_arg;
1596         struct ena_com_dev *ena_dev = &adapter->ena_dev;
1597
1598         ena_com_admin_q_comp_intr_handler(ena_dev);
1599         if (likely(adapter->state != ENA_ADAPTER_STATE_CLOSED))
1600                 ena_com_aenq_intr_handler(ena_dev, adapter);
1601 }
1602
1603 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
1604 {
1605         if (!adapter->wd_state)
1606                 return;
1607
1608         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
1609                 return;
1610
1611         if (unlikely((rte_get_timer_cycles() - adapter->timestamp_wd) >=
1612             adapter->keep_alive_timeout)) {
1613                 PMD_DRV_LOG(ERR, "Keep alive timeout\n");
1614                 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
1615                 adapter->trigger_reset = true;
1616                 ++adapter->dev_stats.wd_expired;
1617         }
1618 }
1619
1620 /* Check if admin queue is enabled */
1621 static void check_for_admin_com_state(struct ena_adapter *adapter)
1622 {
1623         if (unlikely(!ena_com_get_admin_running_state(&adapter->ena_dev))) {
1624                 PMD_DRV_LOG(ERR, "ENA admin queue is not in running state!\n");
1625                 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
1626                 adapter->trigger_reset = true;
1627         }
1628 }
1629
1630 static void ena_timer_wd_callback(__rte_unused struct rte_timer *timer,
1631                                   void *arg)
1632 {
1633         struct ena_adapter *adapter = arg;
1634         struct rte_eth_dev *dev = adapter->rte_dev;
1635
1636         check_for_missing_keep_alive(adapter);
1637         check_for_admin_com_state(adapter);
1638
1639         if (unlikely(adapter->trigger_reset)) {
1640                 PMD_DRV_LOG(ERR, "Trigger reset is on\n");
1641                 rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
1642                         NULL);
1643         }
1644 }
1645
1646 static inline void
1647 set_default_llq_configurations(struct ena_llq_configurations *llq_config,
1648                                struct ena_admin_feature_llq_desc *llq,
1649                                bool use_large_llq_hdr)
1650 {
1651         llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
1652         llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
1653         llq_config->llq_num_decs_before_header =
1654                 ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
1655
1656         if (use_large_llq_hdr &&
1657             (llq->entry_size_ctrl_supported & ENA_ADMIN_LIST_ENTRY_SIZE_256B)) {
1658                 llq_config->llq_ring_entry_size =
1659                         ENA_ADMIN_LIST_ENTRY_SIZE_256B;
1660                 llq_config->llq_ring_entry_size_value = 256;
1661         } else {
1662                 llq_config->llq_ring_entry_size =
1663                         ENA_ADMIN_LIST_ENTRY_SIZE_128B;
1664                 llq_config->llq_ring_entry_size_value = 128;
1665         }
1666 }
1667
1668 static int
1669 ena_set_queues_placement_policy(struct ena_adapter *adapter,
1670                                 struct ena_com_dev *ena_dev,
1671                                 struct ena_admin_feature_llq_desc *llq,
1672                                 struct ena_llq_configurations *llq_default_configurations)
1673 {
1674         int rc;
1675         u32 llq_feature_mask;
1676
1677         llq_feature_mask = 1 << ENA_ADMIN_LLQ;
1678         if (!(ena_dev->supported_features & llq_feature_mask)) {
1679                 PMD_DRV_LOG(INFO,
1680                         "LLQ is not supported. Fallback to host mode policy.\n");
1681                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1682                 return 0;
1683         }
1684
1685         rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
1686         if (unlikely(rc)) {
1687                 PMD_INIT_LOG(WARNING, "Failed to config dev mode. "
1688                         "Fallback to host mode policy.");
1689                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1690                 return 0;
1691         }
1692
1693         /* Nothing to config, exit */
1694         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
1695                 return 0;
1696
1697         if (!adapter->dev_mem_base) {
1698                 PMD_DRV_LOG(ERR, "Unable to access LLQ bar resource. "
1699                         "Fallback to host mode policy.\n.");
1700                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1701                 return 0;
1702         }
1703
1704         ena_dev->mem_bar = adapter->dev_mem_base;
1705
1706         return 0;
1707 }
1708
1709 static uint32_t ena_calc_max_io_queue_num(struct ena_com_dev *ena_dev,
1710         struct ena_com_dev_get_features_ctx *get_feat_ctx)
1711 {
1712         uint32_t io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
1713
1714         /* Regular queues capabilities */
1715         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
1716                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
1717                         &get_feat_ctx->max_queue_ext.max_queue_ext;
1718                 io_rx_num = RTE_MIN(max_queue_ext->max_rx_sq_num,
1719                                     max_queue_ext->max_rx_cq_num);
1720                 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
1721                 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
1722         } else {
1723                 struct ena_admin_queue_feature_desc *max_queues =
1724                         &get_feat_ctx->max_queues;
1725                 io_tx_sq_num = max_queues->max_sq_num;
1726                 io_tx_cq_num = max_queues->max_cq_num;
1727                 io_rx_num = RTE_MIN(io_tx_sq_num, io_tx_cq_num);
1728         }
1729
1730         /* In case of LLQ use the llq number in the get feature cmd */
1731         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
1732                 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
1733
1734         max_num_io_queues = RTE_MIN(ENA_MAX_NUM_IO_QUEUES, io_rx_num);
1735         max_num_io_queues = RTE_MIN(max_num_io_queues, io_tx_sq_num);
1736         max_num_io_queues = RTE_MIN(max_num_io_queues, io_tx_cq_num);
1737
1738         if (unlikely(max_num_io_queues == 0)) {
1739                 PMD_DRV_LOG(ERR, "Number of IO queues should not be 0\n");
1740                 return -EFAULT;
1741         }
1742
1743         return max_num_io_queues;
1744 }
1745
1746 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1747 {
1748         struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
1749         struct rte_pci_device *pci_dev;
1750         struct rte_intr_handle *intr_handle;
1751         struct ena_adapter *adapter = eth_dev->data->dev_private;
1752         struct ena_com_dev *ena_dev = &adapter->ena_dev;
1753         struct ena_com_dev_get_features_ctx get_feat_ctx;
1754         struct ena_llq_configurations llq_config;
1755         const char *queue_type_str;
1756         uint32_t max_num_io_queues;
1757         int rc;
1758         static int adapters_found;
1759         bool disable_meta_caching;
1760         bool wd_state = false;
1761
1762         eth_dev->dev_ops = &ena_dev_ops;
1763         eth_dev->rx_pkt_burst = &eth_ena_recv_pkts;
1764         eth_dev->tx_pkt_burst = &eth_ena_xmit_pkts;
1765         eth_dev->tx_pkt_prepare = &eth_ena_prep_pkts;
1766
1767         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1768                 return 0;
1769
1770         eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1771
1772         memset(adapter, 0, sizeof(struct ena_adapter));
1773         ena_dev = &adapter->ena_dev;
1774
1775         adapter->rte_eth_dev_data = eth_dev->data;
1776         adapter->rte_dev = eth_dev;
1777
1778         pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1779         adapter->pdev = pci_dev;
1780
1781         PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d",
1782                      pci_dev->addr.domain,
1783                      pci_dev->addr.bus,
1784                      pci_dev->addr.devid,
1785                      pci_dev->addr.function);
1786
1787         intr_handle = &pci_dev->intr_handle;
1788
1789         adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1790         adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1791
1792         if (!adapter->regs) {
1793                 PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)",
1794                              ENA_REGS_BAR);
1795                 return -ENXIO;
1796         }
1797
1798         ena_dev->reg_bar = adapter->regs;
1799         ena_dev->dmadev = adapter->pdev;
1800
1801         adapter->id_number = adapters_found;
1802
1803         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1804                  adapter->id_number);
1805
1806         rc = ena_parse_devargs(adapter, pci_dev->device.devargs);
1807         if (rc != 0) {
1808                 PMD_INIT_LOG(CRIT, "Failed to parse devargs\n");
1809                 goto err;
1810         }
1811
1812         /* device specific initialization routine */
1813         rc = ena_device_init(ena_dev, &get_feat_ctx, &wd_state);
1814         if (rc) {
1815                 PMD_INIT_LOG(CRIT, "Failed to init ENA device");
1816                 goto err;
1817         }
1818         adapter->wd_state = wd_state;
1819
1820         set_default_llq_configurations(&llq_config, &get_feat_ctx.llq,
1821                 adapter->use_large_llq_hdr);
1822         rc = ena_set_queues_placement_policy(adapter, ena_dev,
1823                                              &get_feat_ctx.llq, &llq_config);
1824         if (unlikely(rc)) {
1825                 PMD_INIT_LOG(CRIT, "Failed to set placement policy");
1826                 return rc;
1827         }
1828
1829         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
1830                 queue_type_str = "Regular";
1831         else
1832                 queue_type_str = "Low latency";
1833         PMD_DRV_LOG(INFO, "Placement policy: %s\n", queue_type_str);
1834
1835         calc_queue_ctx.ena_dev = ena_dev;
1836         calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
1837
1838         max_num_io_queues = ena_calc_max_io_queue_num(ena_dev, &get_feat_ctx);
1839         rc = ena_calc_io_queue_size(&calc_queue_ctx,
1840                 adapter->use_large_llq_hdr);
1841         if (unlikely((rc != 0) || (max_num_io_queues == 0))) {
1842                 rc = -EFAULT;
1843                 goto err_device_destroy;
1844         }
1845
1846         adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
1847         adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
1848         adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
1849         adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
1850         adapter->max_num_io_queues = max_num_io_queues;
1851
1852         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1853                 disable_meta_caching =
1854                         !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
1855                         BIT(ENA_ADMIN_DISABLE_META_CACHING));
1856         } else {
1857                 disable_meta_caching = false;
1858         }
1859
1860         /* prepare ring structures */
1861         ena_init_rings(adapter, disable_meta_caching);
1862
1863         ena_config_debug_area(adapter);
1864
1865         /* Set max MTU for this device */
1866         adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1867
1868         /* set device support for offloads */
1869         adapter->offloads.tso4_supported = (get_feat_ctx.offload.tx &
1870                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0;
1871         adapter->offloads.tx_csum_supported = (get_feat_ctx.offload.tx &
1872                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK) != 0;
1873         adapter->offloads.rx_csum_supported =
1874                 (get_feat_ctx.offload.rx_supported &
1875                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK) != 0;
1876
1877         /* Copy MAC address and point DPDK to it */
1878         eth_dev->data->mac_addrs = (struct rte_ether_addr *)adapter->mac_addr;
1879         rte_ether_addr_copy((struct rte_ether_addr *)
1880                         get_feat_ctx.dev_attr.mac_addr,
1881                         (struct rte_ether_addr *)adapter->mac_addr);
1882
1883         adapter->drv_stats = rte_zmalloc("adapter stats",
1884                                          sizeof(*adapter->drv_stats),
1885                                          RTE_CACHE_LINE_SIZE);
1886         if (!adapter->drv_stats) {
1887                 PMD_DRV_LOG(ERR, "failed to alloc mem for adapter stats\n");
1888                 rc = -ENOMEM;
1889                 goto err_delete_debug_area;
1890         }
1891
1892         rte_spinlock_init(&adapter->admin_lock);
1893
1894         rte_intr_callback_register(intr_handle,
1895                                    ena_interrupt_handler_rte,
1896                                    adapter);
1897         rte_intr_enable(intr_handle);
1898         ena_com_set_admin_polling_mode(ena_dev, false);
1899         ena_com_admin_aenq_enable(ena_dev);
1900
1901         if (adapters_found == 0)
1902                 rte_timer_subsystem_init();
1903         rte_timer_init(&adapter->timer_wd);
1904
1905         adapters_found++;
1906         adapter->state = ENA_ADAPTER_STATE_INIT;
1907
1908         return 0;
1909
1910 err_delete_debug_area:
1911         ena_com_delete_debug_area(ena_dev);
1912
1913 err_device_destroy:
1914         ena_com_delete_host_info(ena_dev);
1915         ena_com_admin_destroy(ena_dev);
1916
1917 err:
1918         return rc;
1919 }
1920
1921 static void ena_destroy_device(struct rte_eth_dev *eth_dev)
1922 {
1923         struct ena_adapter *adapter = eth_dev->data->dev_private;
1924         struct ena_com_dev *ena_dev = &adapter->ena_dev;
1925
1926         if (adapter->state == ENA_ADAPTER_STATE_FREE)
1927                 return;
1928
1929         ena_com_set_admin_running_state(ena_dev, false);
1930
1931         if (adapter->state != ENA_ADAPTER_STATE_CLOSED)
1932                 ena_close(eth_dev);
1933
1934         ena_com_delete_debug_area(ena_dev);
1935         ena_com_delete_host_info(ena_dev);
1936
1937         ena_com_abort_admin_commands(ena_dev);
1938         ena_com_wait_for_abort_completion(ena_dev);
1939         ena_com_admin_destroy(ena_dev);
1940         ena_com_mmio_reg_read_request_destroy(ena_dev);
1941
1942         adapter->state = ENA_ADAPTER_STATE_FREE;
1943 }
1944
1945 static int eth_ena_dev_uninit(struct rte_eth_dev *eth_dev)
1946 {
1947         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1948                 return 0;
1949
1950         ena_destroy_device(eth_dev);
1951
1952         return 0;
1953 }
1954
1955 static int ena_dev_configure(struct rte_eth_dev *dev)
1956 {
1957         struct ena_adapter *adapter = dev->data->dev_private;
1958
1959         adapter->state = ENA_ADAPTER_STATE_CONFIG;
1960
1961         adapter->tx_selected_offloads = dev->data->dev_conf.txmode.offloads;
1962         adapter->rx_selected_offloads = dev->data->dev_conf.rxmode.offloads;
1963         return 0;
1964 }
1965
1966 static void ena_init_rings(struct ena_adapter *adapter,
1967                            bool disable_meta_caching)
1968 {
1969         size_t i;
1970
1971         for (i = 0; i < adapter->max_num_io_queues; i++) {
1972                 struct ena_ring *ring = &adapter->tx_ring[i];
1973
1974                 ring->configured = 0;
1975                 ring->type = ENA_RING_TYPE_TX;
1976                 ring->adapter = adapter;
1977                 ring->id = i;
1978                 ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
1979                 ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
1980                 ring->sgl_size = adapter->max_tx_sgl_size;
1981                 ring->disable_meta_caching = disable_meta_caching;
1982         }
1983
1984         for (i = 0; i < adapter->max_num_io_queues; i++) {
1985                 struct ena_ring *ring = &adapter->rx_ring[i];
1986
1987                 ring->configured = 0;
1988                 ring->type = ENA_RING_TYPE_RX;
1989                 ring->adapter = adapter;
1990                 ring->id = i;
1991                 ring->sgl_size = adapter->max_rx_sgl_size;
1992         }
1993 }
1994
1995 static int ena_infos_get(struct rte_eth_dev *dev,
1996                           struct rte_eth_dev_info *dev_info)
1997 {
1998         struct ena_adapter *adapter;
1999         struct ena_com_dev *ena_dev;
2000         uint64_t rx_feat = 0, tx_feat = 0;
2001
2002         ena_assert_msg(dev->data != NULL, "Uninitialized device\n");
2003         ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device\n");
2004         adapter = dev->data->dev_private;
2005
2006         ena_dev = &adapter->ena_dev;
2007         ena_assert_msg(ena_dev != NULL, "Uninitialized device\n");
2008
2009         dev_info->speed_capa =
2010                         ETH_LINK_SPEED_1G   |
2011                         ETH_LINK_SPEED_2_5G |
2012                         ETH_LINK_SPEED_5G   |
2013                         ETH_LINK_SPEED_10G  |
2014                         ETH_LINK_SPEED_25G  |
2015                         ETH_LINK_SPEED_40G  |
2016                         ETH_LINK_SPEED_50G  |
2017                         ETH_LINK_SPEED_100G;
2018
2019         /* Set Tx & Rx features available for device */
2020         if (adapter->offloads.tso4_supported)
2021                 tx_feat |= DEV_TX_OFFLOAD_TCP_TSO;
2022
2023         if (adapter->offloads.tx_csum_supported)
2024                 tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
2025                         DEV_TX_OFFLOAD_UDP_CKSUM |
2026                         DEV_TX_OFFLOAD_TCP_CKSUM;
2027
2028         if (adapter->offloads.rx_csum_supported)
2029                 rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
2030                         DEV_RX_OFFLOAD_UDP_CKSUM  |
2031                         DEV_RX_OFFLOAD_TCP_CKSUM;
2032
2033         rx_feat |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2034
2035         /* Inform framework about available features */
2036         dev_info->rx_offload_capa = rx_feat;
2037         dev_info->rx_queue_offload_capa = rx_feat;
2038         dev_info->tx_offload_capa = tx_feat;
2039         dev_info->tx_queue_offload_capa = tx_feat;
2040
2041         dev_info->flow_type_rss_offloads = ETH_RSS_IP | ETH_RSS_TCP |
2042                                            ETH_RSS_UDP;
2043
2044         dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
2045         dev_info->max_rx_pktlen  = adapter->max_mtu;
2046         dev_info->max_mac_addrs = 1;
2047
2048         dev_info->max_rx_queues = adapter->max_num_io_queues;
2049         dev_info->max_tx_queues = adapter->max_num_io_queues;
2050         dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
2051
2052         adapter->tx_supported_offloads = tx_feat;
2053         adapter->rx_supported_offloads = rx_feat;
2054
2055         dev_info->rx_desc_lim.nb_max = adapter->max_rx_ring_size;
2056         dev_info->rx_desc_lim.nb_min = ENA_MIN_RING_DESC;
2057         dev_info->rx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2058                                         adapter->max_rx_sgl_size);
2059         dev_info->rx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2060                                         adapter->max_rx_sgl_size);
2061
2062         dev_info->tx_desc_lim.nb_max = adapter->max_tx_ring_size;
2063         dev_info->tx_desc_lim.nb_min = ENA_MIN_RING_DESC;
2064         dev_info->tx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2065                                         adapter->max_tx_sgl_size);
2066         dev_info->tx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2067                                         adapter->max_tx_sgl_size);
2068
2069         return 0;
2070 }
2071
2072 static inline void ena_init_rx_mbuf(struct rte_mbuf *mbuf, uint16_t len)
2073 {
2074         mbuf->data_len = len;
2075         mbuf->data_off = RTE_PKTMBUF_HEADROOM;
2076         mbuf->refcnt = 1;
2077         mbuf->next = NULL;
2078 }
2079
2080 static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
2081                                     struct ena_com_rx_buf_info *ena_bufs,
2082                                     uint32_t descs,
2083                                     uint16_t *next_to_clean,
2084                                     uint8_t offset)
2085 {
2086         struct rte_mbuf *mbuf;
2087         struct rte_mbuf *mbuf_head;
2088         struct ena_rx_buffer *rx_info;
2089         int rc;
2090         uint16_t ntc, len, req_id, buf = 0;
2091
2092         if (unlikely(descs == 0))
2093                 return NULL;
2094
2095         ntc = *next_to_clean;
2096
2097         len = ena_bufs[buf].len;
2098         req_id = ena_bufs[buf].req_id;
2099
2100         rx_info = &rx_ring->rx_buffer_info[req_id];
2101
2102         mbuf = rx_info->mbuf;
2103         RTE_ASSERT(mbuf != NULL);
2104
2105         ena_init_rx_mbuf(mbuf, len);
2106
2107         /* Fill the mbuf head with the data specific for 1st segment. */
2108         mbuf_head = mbuf;
2109         mbuf_head->nb_segs = descs;
2110         mbuf_head->port = rx_ring->port_id;
2111         mbuf_head->pkt_len = len;
2112         mbuf_head->data_off += offset;
2113
2114         rx_info->mbuf = NULL;
2115         rx_ring->empty_rx_reqs[ntc] = req_id;
2116         ntc = ENA_IDX_NEXT_MASKED(ntc, rx_ring->size_mask);
2117
2118         while (--descs) {
2119                 ++buf;
2120                 len = ena_bufs[buf].len;
2121                 req_id = ena_bufs[buf].req_id;
2122
2123                 rx_info = &rx_ring->rx_buffer_info[req_id];
2124                 RTE_ASSERT(rx_info->mbuf != NULL);
2125
2126                 if (unlikely(len == 0)) {
2127                         /*
2128                          * Some devices can pass descriptor with the length 0.
2129                          * To avoid confusion, the PMD is simply putting the
2130                          * descriptor back, as it was never used. We'll avoid
2131                          * mbuf allocation that way.
2132                          */
2133                         rc = ena_add_single_rx_desc(rx_ring->ena_com_io_sq,
2134                                 rx_info->mbuf, req_id);
2135                         if (unlikely(rc != 0)) {
2136                                 /* Free the mbuf in case of an error. */
2137                                 rte_mbuf_raw_free(rx_info->mbuf);
2138                         } else {
2139                                 /*
2140                                  * If there was no error, just exit the loop as
2141                                  * 0 length descriptor is always the last one.
2142                                  */
2143                                 break;
2144                         }
2145                 } else {
2146                         /* Create an mbuf chain. */
2147                         mbuf->next = rx_info->mbuf;
2148                         mbuf = mbuf->next;
2149
2150                         ena_init_rx_mbuf(mbuf, len);
2151                         mbuf_head->pkt_len += len;
2152                 }
2153
2154                 /*
2155                  * Mark the descriptor as depleted and perform necessary
2156                  * cleanup.
2157                  * This code will execute in two cases:
2158                  *  1. Descriptor len was greater than 0 - normal situation.
2159                  *  2. Descriptor len was 0 and we failed to add the descriptor
2160                  *     to the device. In that situation, we should try to add
2161                  *     the mbuf again in the populate routine and mark the
2162                  *     descriptor as used up by the device.
2163                  */
2164                 rx_info->mbuf = NULL;
2165                 rx_ring->empty_rx_reqs[ntc] = req_id;
2166                 ntc = ENA_IDX_NEXT_MASKED(ntc, rx_ring->size_mask);
2167         }
2168
2169         *next_to_clean = ntc;
2170
2171         return mbuf_head;
2172 }
2173
2174 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
2175                                   uint16_t nb_pkts)
2176 {
2177         struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
2178         unsigned int free_queue_entries;
2179         unsigned int refill_threshold;
2180         uint16_t next_to_clean = rx_ring->next_to_clean;
2181         uint16_t descs_in_use;
2182         struct rte_mbuf *mbuf;
2183         uint16_t completed;
2184         struct ena_com_rx_ctx ena_rx_ctx;
2185         int i, rc = 0;
2186
2187         /* Check adapter state */
2188         if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2189                 PMD_DRV_LOG(ALERT,
2190                         "Trying to receive pkts while device is NOT running\n");
2191                 return 0;
2192         }
2193
2194         descs_in_use = rx_ring->ring_size -
2195                 ena_com_free_q_entries(rx_ring->ena_com_io_sq) - 1;
2196         nb_pkts = RTE_MIN(descs_in_use, nb_pkts);
2197
2198         for (completed = 0; completed < nb_pkts; completed++) {
2199                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
2200                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
2201                 ena_rx_ctx.descs = 0;
2202                 ena_rx_ctx.pkt_offset = 0;
2203                 /* receive packet context */
2204                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
2205                                     rx_ring->ena_com_io_sq,
2206                                     &ena_rx_ctx);
2207                 if (unlikely(rc)) {
2208                         PMD_DRV_LOG(ERR, "ena_com_rx_pkt error %d\n", rc);
2209                         if (rc == ENA_COM_NO_SPACE) {
2210                                 ++rx_ring->rx_stats.bad_desc_num;
2211                                 rx_ring->adapter->reset_reason =
2212                                         ENA_REGS_RESET_TOO_MANY_RX_DESCS;
2213                         } else {
2214                                 ++rx_ring->rx_stats.bad_req_id;
2215                                 rx_ring->adapter->reset_reason =
2216                                         ENA_REGS_RESET_INV_RX_REQ_ID;
2217                         }
2218                         rx_ring->adapter->trigger_reset = true;
2219                         return 0;
2220                 }
2221
2222                 mbuf = ena_rx_mbuf(rx_ring,
2223                         ena_rx_ctx.ena_bufs,
2224                         ena_rx_ctx.descs,
2225                         &next_to_clean,
2226                         ena_rx_ctx.pkt_offset);
2227                 if (unlikely(mbuf == NULL)) {
2228                         for (i = 0; i < ena_rx_ctx.descs; ++i) {
2229                                 rx_ring->empty_rx_reqs[next_to_clean] =
2230                                         rx_ring->ena_bufs[i].req_id;
2231                                 next_to_clean = ENA_IDX_NEXT_MASKED(
2232                                         next_to_clean, rx_ring->size_mask);
2233                         }
2234                         break;
2235                 }
2236
2237                 /* fill mbuf attributes if any */
2238                 ena_rx_mbuf_prepare(mbuf, &ena_rx_ctx);
2239
2240                 if (unlikely(mbuf->ol_flags &
2241                                 (PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD))) {
2242                         rte_atomic64_inc(&rx_ring->adapter->drv_stats->ierrors);
2243                         ++rx_ring->rx_stats.bad_csum;
2244                 }
2245
2246                 mbuf->hash.rss = ena_rx_ctx.hash;
2247
2248                 rx_pkts[completed] = mbuf;
2249                 rx_ring->rx_stats.bytes += mbuf->pkt_len;
2250         }
2251
2252         rx_ring->rx_stats.cnt += completed;
2253         rx_ring->next_to_clean = next_to_clean;
2254
2255         free_queue_entries = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
2256         refill_threshold =
2257                 RTE_MIN(rx_ring->ring_size / ENA_REFILL_THRESH_DIVIDER,
2258                 (unsigned int)ENA_REFILL_THRESH_PACKET);
2259
2260         /* Burst refill to save doorbells, memory barriers, const interval */
2261         if (free_queue_entries > refill_threshold) {
2262                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
2263                 ena_populate_rx_queue(rx_ring, free_queue_entries);
2264         }
2265
2266         return completed;
2267 }
2268
2269 static uint16_t
2270 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2271                 uint16_t nb_pkts)
2272 {
2273         int32_t ret;
2274         uint32_t i;
2275         struct rte_mbuf *m;
2276         struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2277         struct rte_ipv4_hdr *ip_hdr;
2278         uint64_t ol_flags;
2279         uint16_t frag_field;
2280
2281         for (i = 0; i != nb_pkts; i++) {
2282                 m = tx_pkts[i];
2283                 ol_flags = m->ol_flags;
2284
2285                 if (!(ol_flags & PKT_TX_IPV4))
2286                         continue;
2287
2288                 /* If there was not L2 header length specified, assume it is
2289                  * length of the ethernet header.
2290                  */
2291                 if (unlikely(m->l2_len == 0))
2292                         m->l2_len = sizeof(struct rte_ether_hdr);
2293
2294                 ip_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
2295                                                  m->l2_len);
2296                 frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
2297
2298                 if ((frag_field & RTE_IPV4_HDR_DF_FLAG) != 0) {
2299                         m->packet_type |= RTE_PTYPE_L4_NONFRAG;
2300
2301                         /* If IPv4 header has DF flag enabled and TSO support is
2302                          * disabled, partial chcecksum should not be calculated.
2303                          */
2304                         if (!tx_ring->adapter->offloads.tso4_supported)
2305                                 continue;
2306                 }
2307
2308                 if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
2309                                 (ol_flags & PKT_TX_L4_MASK) ==
2310                                 PKT_TX_SCTP_CKSUM) {
2311                         rte_errno = ENOTSUP;
2312                         return i;
2313                 }
2314
2315 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2316                 ret = rte_validate_tx_offload(m);
2317                 if (ret != 0) {
2318                         rte_errno = -ret;
2319                         return i;
2320                 }
2321 #endif
2322
2323                 /* In case we are supposed to TSO and have DF not set (DF=0)
2324                  * hardware must be provided with partial checksum, otherwise
2325                  * it will take care of necessary calculations.
2326                  */
2327
2328                 ret = rte_net_intel_cksum_flags_prepare(m,
2329                         ol_flags & ~PKT_TX_TCP_SEG);
2330                 if (ret != 0) {
2331                         rte_errno = -ret;
2332                         return i;
2333                 }
2334         }
2335
2336         return i;
2337 }
2338
2339 static void ena_update_hints(struct ena_adapter *adapter,
2340                              struct ena_admin_ena_hw_hints *hints)
2341 {
2342         if (hints->admin_completion_tx_timeout)
2343                 adapter->ena_dev.admin_queue.completion_timeout =
2344                         hints->admin_completion_tx_timeout * 1000;
2345
2346         if (hints->mmio_read_timeout)
2347                 /* convert to usec */
2348                 adapter->ena_dev.mmio_read.reg_read_to =
2349                         hints->mmio_read_timeout * 1000;
2350
2351         if (hints->driver_watchdog_timeout) {
2352                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2353                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2354                 else
2355                         // Convert msecs to ticks
2356                         adapter->keep_alive_timeout =
2357                                 (hints->driver_watchdog_timeout *
2358                                 rte_get_timer_hz()) / 1000;
2359         }
2360 }
2361
2362 static int ena_check_space_and_linearize_mbuf(struct ena_ring *tx_ring,
2363                                               struct rte_mbuf *mbuf)
2364 {
2365         struct ena_com_dev *ena_dev;
2366         int num_segments, header_len, rc;
2367
2368         ena_dev = &tx_ring->adapter->ena_dev;
2369         num_segments = mbuf->nb_segs;
2370         header_len = mbuf->data_len;
2371
2372         if (likely(num_segments < tx_ring->sgl_size))
2373                 goto checkspace;
2374
2375         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV &&
2376             (num_segments == tx_ring->sgl_size) &&
2377             (header_len < tx_ring->tx_max_header_size))
2378                 goto checkspace;
2379
2380         /* Checking for space for 2 additional metadata descriptors due to
2381          * possible header split and metadata descriptor. Linearization will
2382          * be needed so we reduce the segments number from num_segments to 1
2383          */
2384         if (!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq, 3)) {
2385                 PMD_DRV_LOG(DEBUG, "Not enough space in the tx queue\n");
2386                 return ENA_COM_NO_MEM;
2387         }
2388         ++tx_ring->tx_stats.linearize;
2389         rc = rte_pktmbuf_linearize(mbuf);
2390         if (unlikely(rc)) {
2391                 PMD_DRV_LOG(WARNING, "Mbuf linearize failed\n");
2392                 rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
2393                 ++tx_ring->tx_stats.linearize_failed;
2394                 return rc;
2395         }
2396
2397         return 0;
2398
2399 checkspace:
2400         /* Checking for space for 2 additional metadata descriptors due to
2401          * possible header split and metadata descriptor
2402          */
2403         if (!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
2404                                           num_segments + 2)) {
2405                 PMD_DRV_LOG(DEBUG, "Not enough space in the tx queue\n");
2406                 return ENA_COM_NO_MEM;
2407         }
2408
2409         return 0;
2410 }
2411
2412 static void ena_tx_map_mbuf(struct ena_ring *tx_ring,
2413         struct ena_tx_buffer *tx_info,
2414         struct rte_mbuf *mbuf,
2415         void **push_header,
2416         uint16_t *header_len)
2417 {
2418         struct ena_com_buf *ena_buf;
2419         uint16_t delta, seg_len, push_len;
2420
2421         delta = 0;
2422         seg_len = mbuf->data_len;
2423
2424         tx_info->mbuf = mbuf;
2425         ena_buf = tx_info->bufs;
2426
2427         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2428                 /*
2429                  * Tx header might be (and will be in most cases) smaller than
2430                  * tx_max_header_size. But it's not an issue to send more data
2431                  * to the device, than actually needed if the mbuf size is
2432                  * greater than tx_max_header_size.
2433                  */
2434                 push_len = RTE_MIN(mbuf->pkt_len, tx_ring->tx_max_header_size);
2435                 *header_len = push_len;
2436
2437                 if (likely(push_len <= seg_len)) {
2438                         /* If the push header is in the single segment, then
2439                          * just point it to the 1st mbuf data.
2440                          */
2441                         *push_header = rte_pktmbuf_mtod(mbuf, uint8_t *);
2442                 } else {
2443                         /* If the push header lays in the several segments, copy
2444                          * it to the intermediate buffer.
2445                          */
2446                         rte_pktmbuf_read(mbuf, 0, push_len,
2447                                 tx_ring->push_buf_intermediate_buf);
2448                         *push_header = tx_ring->push_buf_intermediate_buf;
2449                         delta = push_len - seg_len;
2450                 }
2451         } else {
2452                 *push_header = NULL;
2453                 *header_len = 0;
2454                 push_len = 0;
2455         }
2456
2457         /* Process first segment taking into consideration pushed header */
2458         if (seg_len > push_len) {
2459                 ena_buf->paddr = mbuf->buf_iova +
2460                                 mbuf->data_off +
2461                                 push_len;
2462                 ena_buf->len = seg_len - push_len;
2463                 ena_buf++;
2464                 tx_info->num_of_bufs++;
2465         }
2466
2467         while ((mbuf = mbuf->next) != NULL) {
2468                 seg_len = mbuf->data_len;
2469
2470                 /* Skip mbufs if whole data is pushed as a header */
2471                 if (unlikely(delta > seg_len)) {
2472                         delta -= seg_len;
2473                         continue;
2474                 }
2475
2476                 ena_buf->paddr = mbuf->buf_iova + mbuf->data_off + delta;
2477                 ena_buf->len = seg_len - delta;
2478                 ena_buf++;
2479                 tx_info->num_of_bufs++;
2480
2481                 delta = 0;
2482         }
2483 }
2484
2485 static int ena_xmit_mbuf(struct ena_ring *tx_ring, struct rte_mbuf *mbuf)
2486 {
2487         struct ena_tx_buffer *tx_info;
2488         struct ena_com_tx_ctx ena_tx_ctx = { { 0 } };
2489         uint16_t next_to_use;
2490         uint16_t header_len;
2491         uint16_t req_id;
2492         void *push_header;
2493         int nb_hw_desc;
2494         int rc;
2495
2496         rc = ena_check_space_and_linearize_mbuf(tx_ring, mbuf);
2497         if (unlikely(rc))
2498                 return rc;
2499
2500         next_to_use = tx_ring->next_to_use;
2501
2502         req_id = tx_ring->empty_tx_reqs[next_to_use];
2503         tx_info = &tx_ring->tx_buffer_info[req_id];
2504         tx_info->num_of_bufs = 0;
2505
2506         ena_tx_map_mbuf(tx_ring, tx_info, mbuf, &push_header, &header_len);
2507
2508         ena_tx_ctx.ena_bufs = tx_info->bufs;
2509         ena_tx_ctx.push_header = push_header;
2510         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2511         ena_tx_ctx.req_id = req_id;
2512         ena_tx_ctx.header_len = header_len;
2513
2514         /* Set Tx offloads flags, if applicable */
2515         ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx, tx_ring->offloads,
2516                 tx_ring->disable_meta_caching);
2517
2518         if (unlikely(ena_com_is_doorbell_needed(tx_ring->ena_com_io_sq,
2519                         &ena_tx_ctx))) {
2520                 PMD_DRV_LOG(DEBUG,
2521                         "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
2522                         tx_ring->id);
2523                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2524                 tx_ring->tx_stats.doorbells++;
2525         }
2526
2527         /* prepare the packet's descriptors to dma engine */
2528         rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq, &ena_tx_ctx,
2529                 &nb_hw_desc);
2530         if (unlikely(rc)) {
2531                 ++tx_ring->tx_stats.prepare_ctx_err;
2532                 return rc;
2533         }
2534
2535         tx_info->tx_descs = nb_hw_desc;
2536
2537         tx_ring->tx_stats.cnt++;
2538         tx_ring->tx_stats.bytes += mbuf->pkt_len;
2539
2540         tx_ring->next_to_use = ENA_IDX_NEXT_MASKED(next_to_use,
2541                 tx_ring->size_mask);
2542
2543         return 0;
2544 }
2545
2546 static void ena_tx_cleanup(struct ena_ring *tx_ring)
2547 {
2548         unsigned int cleanup_budget;
2549         unsigned int total_tx_descs = 0;
2550         uint16_t next_to_clean = tx_ring->next_to_clean;
2551
2552         cleanup_budget = RTE_MIN(tx_ring->ring_size / ENA_REFILL_THRESH_DIVIDER,
2553                 (unsigned int)ENA_REFILL_THRESH_PACKET);
2554
2555         while (likely(total_tx_descs < cleanup_budget)) {
2556                 struct rte_mbuf *mbuf;
2557                 struct ena_tx_buffer *tx_info;
2558                 uint16_t req_id;
2559
2560                 if (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) != 0)
2561                         break;
2562
2563                 if (unlikely(validate_tx_req_id(tx_ring, req_id) != 0))
2564                         break;
2565
2566                 /* Get Tx info & store how many descs were processed  */
2567                 tx_info = &tx_ring->tx_buffer_info[req_id];
2568
2569                 mbuf = tx_info->mbuf;
2570                 rte_pktmbuf_free(mbuf);
2571
2572                 tx_info->mbuf = NULL;
2573                 tx_ring->empty_tx_reqs[next_to_clean] = req_id;
2574
2575                 total_tx_descs += tx_info->tx_descs;
2576
2577                 /* Put back descriptor to the ring for reuse */
2578                 next_to_clean = ENA_IDX_NEXT_MASKED(next_to_clean,
2579                         tx_ring->size_mask);
2580         }
2581
2582         if (likely(total_tx_descs > 0)) {
2583                 /* acknowledge completion of sent packets */
2584                 tx_ring->next_to_clean = next_to_clean;
2585                 ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
2586                 ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
2587         }
2588 }
2589
2590 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2591                                   uint16_t nb_pkts)
2592 {
2593         struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2594         uint16_t sent_idx = 0;
2595
2596         /* Check adapter state */
2597         if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2598                 PMD_DRV_LOG(ALERT,
2599                         "Trying to xmit pkts while device is NOT running\n");
2600                 return 0;
2601         }
2602
2603         for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
2604                 if (ena_xmit_mbuf(tx_ring, tx_pkts[sent_idx]))
2605                         break;
2606
2607                 rte_prefetch0(tx_pkts[ENA_IDX_ADD_MASKED(sent_idx, 4,
2608                         tx_ring->size_mask)]);
2609         }
2610
2611         tx_ring->tx_stats.available_desc =
2612                 ena_com_free_q_entries(tx_ring->ena_com_io_sq);
2613
2614         /* If there are ready packets to be xmitted... */
2615         if (sent_idx > 0) {
2616                 /* ...let HW do its best :-) */
2617                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2618                 tx_ring->tx_stats.doorbells++;
2619         }
2620
2621         ena_tx_cleanup(tx_ring);
2622
2623         tx_ring->tx_stats.available_desc =
2624                 ena_com_free_q_entries(tx_ring->ena_com_io_sq);
2625         tx_ring->tx_stats.tx_poll++;
2626
2627         return sent_idx;
2628 }
2629
2630 int ena_copy_eni_stats(struct ena_adapter *adapter)
2631 {
2632         struct ena_admin_eni_stats admin_eni_stats;
2633         int rc;
2634
2635         rte_spinlock_lock(&adapter->admin_lock);
2636         rc = ena_com_get_eni_stats(&adapter->ena_dev, &admin_eni_stats);
2637         rte_spinlock_unlock(&adapter->admin_lock);
2638         if (rc != 0) {
2639                 if (rc == ENA_COM_UNSUPPORTED) {
2640                         PMD_DRV_LOG(DEBUG,
2641                                 "Retrieving ENI metrics is not supported.\n");
2642                 } else {
2643                         PMD_DRV_LOG(WARNING,
2644                                 "Failed to get ENI metrics: %d\n", rc);
2645                 }
2646                 return rc;
2647         }
2648
2649         rte_memcpy(&adapter->eni_stats, &admin_eni_stats,
2650                 sizeof(struct ena_stats_eni));
2651
2652         return 0;
2653 }
2654
2655 /**
2656  * DPDK callback to retrieve names of extended device statistics
2657  *
2658  * @param dev
2659  *   Pointer to Ethernet device structure.
2660  * @param[out] xstats_names
2661  *   Buffer to insert names into.
2662  * @param n
2663  *   Number of names.
2664  *
2665  * @return
2666  *   Number of xstats names.
2667  */
2668 static int ena_xstats_get_names(struct rte_eth_dev *dev,
2669                                 struct rte_eth_xstat_name *xstats_names,
2670                                 unsigned int n)
2671 {
2672         unsigned int xstats_count = ena_xstats_calc_num(dev);
2673         unsigned int stat, i, count = 0;
2674
2675         if (n < xstats_count || !xstats_names)
2676                 return xstats_count;
2677
2678         for (stat = 0; stat < ENA_STATS_ARRAY_GLOBAL; stat++, count++)
2679                 strcpy(xstats_names[count].name,
2680                         ena_stats_global_strings[stat].name);
2681
2682         for (stat = 0; stat < ENA_STATS_ARRAY_ENI; stat++, count++)
2683                 strcpy(xstats_names[count].name,
2684                         ena_stats_eni_strings[stat].name);
2685
2686         for (stat = 0; stat < ENA_STATS_ARRAY_RX; stat++)
2687                 for (i = 0; i < dev->data->nb_rx_queues; i++, count++)
2688                         snprintf(xstats_names[count].name,
2689                                 sizeof(xstats_names[count].name),
2690                                 "rx_q%d_%s", i,
2691                                 ena_stats_rx_strings[stat].name);
2692
2693         for (stat = 0; stat < ENA_STATS_ARRAY_TX; stat++)
2694                 for (i = 0; i < dev->data->nb_tx_queues; i++, count++)
2695                         snprintf(xstats_names[count].name,
2696                                 sizeof(xstats_names[count].name),
2697                                 "tx_q%d_%s", i,
2698                                 ena_stats_tx_strings[stat].name);
2699
2700         return xstats_count;
2701 }
2702
2703 /**
2704  * DPDK callback to get extended device statistics.
2705  *
2706  * @param dev
2707  *   Pointer to Ethernet device structure.
2708  * @param[out] stats
2709  *   Stats table output buffer.
2710  * @param n
2711  *   The size of the stats table.
2712  *
2713  * @return
2714  *   Number of xstats on success, negative on failure.
2715  */
2716 static int ena_xstats_get(struct rte_eth_dev *dev,
2717                           struct rte_eth_xstat *xstats,
2718                           unsigned int n)
2719 {
2720         struct ena_adapter *adapter = dev->data->dev_private;
2721         unsigned int xstats_count = ena_xstats_calc_num(dev);
2722         unsigned int stat, i, count = 0;
2723         int stat_offset;
2724         void *stats_begin;
2725
2726         if (n < xstats_count)
2727                 return xstats_count;
2728
2729         if (!xstats)
2730                 return 0;
2731
2732         for (stat = 0; stat < ENA_STATS_ARRAY_GLOBAL; stat++, count++) {
2733                 stat_offset = ena_stats_global_strings[stat].stat_offset;
2734                 stats_begin = &adapter->dev_stats;
2735
2736                 xstats[count].id = count;
2737                 xstats[count].value = *((uint64_t *)
2738                         ((char *)stats_begin + stat_offset));
2739         }
2740
2741         /* Even if the function below fails, we should copy previous (or initial
2742          * values) to keep structure of rte_eth_xstat consistent.
2743          */
2744         ena_copy_eni_stats(adapter);
2745         for (stat = 0; stat < ENA_STATS_ARRAY_ENI; stat++, count++) {
2746                 stat_offset = ena_stats_eni_strings[stat].stat_offset;
2747                 stats_begin = &adapter->eni_stats;
2748
2749                 xstats[count].id = count;
2750                 xstats[count].value = *((uint64_t *)
2751                     ((char *)stats_begin + stat_offset));
2752         }
2753
2754         for (stat = 0; stat < ENA_STATS_ARRAY_RX; stat++) {
2755                 for (i = 0; i < dev->data->nb_rx_queues; i++, count++) {
2756                         stat_offset = ena_stats_rx_strings[stat].stat_offset;
2757                         stats_begin = &adapter->rx_ring[i].rx_stats;
2758
2759                         xstats[count].id = count;
2760                         xstats[count].value = *((uint64_t *)
2761                                 ((char *)stats_begin + stat_offset));
2762                 }
2763         }
2764
2765         for (stat = 0; stat < ENA_STATS_ARRAY_TX; stat++) {
2766                 for (i = 0; i < dev->data->nb_tx_queues; i++, count++) {
2767                         stat_offset = ena_stats_tx_strings[stat].stat_offset;
2768                         stats_begin = &adapter->tx_ring[i].rx_stats;
2769
2770                         xstats[count].id = count;
2771                         xstats[count].value = *((uint64_t *)
2772                                 ((char *)stats_begin + stat_offset));
2773                 }
2774         }
2775
2776         return count;
2777 }
2778
2779 static int ena_xstats_get_by_id(struct rte_eth_dev *dev,
2780                                 const uint64_t *ids,
2781                                 uint64_t *values,
2782                                 unsigned int n)
2783 {
2784         struct ena_adapter *adapter = dev->data->dev_private;
2785         uint64_t id;
2786         uint64_t rx_entries, tx_entries;
2787         unsigned int i;
2788         int qid;
2789         int valid = 0;
2790         bool was_eni_copied = false;
2791
2792         for (i = 0; i < n; ++i) {
2793                 id = ids[i];
2794                 /* Check if id belongs to global statistics */
2795                 if (id < ENA_STATS_ARRAY_GLOBAL) {
2796                         values[i] = *((uint64_t *)&adapter->dev_stats + id);
2797                         ++valid;
2798                         continue;
2799                 }
2800
2801                 /* Check if id belongs to ENI statistics */
2802                 id -= ENA_STATS_ARRAY_GLOBAL;
2803                 if (id < ENA_STATS_ARRAY_ENI) {
2804                         /* Avoid reading ENI stats multiple times in a single
2805                          * function call, as it requires communication with the
2806                          * admin queue.
2807                          */
2808                         if (!was_eni_copied) {
2809                                 was_eni_copied = true;
2810                                 ena_copy_eni_stats(adapter);
2811                         }
2812                         values[i] = *((uint64_t *)&adapter->eni_stats + id);
2813                         ++valid;
2814                         continue;
2815                 }
2816
2817                 /* Check if id belongs to rx queue statistics */
2818                 id -= ENA_STATS_ARRAY_ENI;
2819                 rx_entries = ENA_STATS_ARRAY_RX * dev->data->nb_rx_queues;
2820                 if (id < rx_entries) {
2821                         qid = id % dev->data->nb_rx_queues;
2822                         id /= dev->data->nb_rx_queues;
2823                         values[i] = *((uint64_t *)
2824                                 &adapter->rx_ring[qid].rx_stats + id);
2825                         ++valid;
2826                         continue;
2827                 }
2828                                 /* Check if id belongs to rx queue statistics */
2829                 id -= rx_entries;
2830                 tx_entries = ENA_STATS_ARRAY_TX * dev->data->nb_tx_queues;
2831                 if (id < tx_entries) {
2832                         qid = id % dev->data->nb_tx_queues;
2833                         id /= dev->data->nb_tx_queues;
2834                         values[i] = *((uint64_t *)
2835                                 &adapter->tx_ring[qid].tx_stats + id);
2836                         ++valid;
2837                         continue;
2838                 }
2839         }
2840
2841         return valid;
2842 }
2843
2844 static int ena_process_bool_devarg(const char *key,
2845                                    const char *value,
2846                                    void *opaque)
2847 {
2848         struct ena_adapter *adapter = opaque;
2849         bool bool_value;
2850
2851         /* Parse the value. */
2852         if (strcmp(value, "1") == 0) {
2853                 bool_value = true;
2854         } else if (strcmp(value, "0") == 0) {
2855                 bool_value = false;
2856         } else {
2857                 PMD_INIT_LOG(ERR,
2858                         "Invalid value: '%s' for key '%s'. Accepted: '0' or '1'\n",
2859                         value, key);
2860                 return -EINVAL;
2861         }
2862
2863         /* Now, assign it to the proper adapter field. */
2864         if (strcmp(key, ENA_DEVARG_LARGE_LLQ_HDR))
2865                 adapter->use_large_llq_hdr = bool_value;
2866
2867         return 0;
2868 }
2869
2870 static int ena_parse_devargs(struct ena_adapter *adapter,
2871                              struct rte_devargs *devargs)
2872 {
2873         static const char * const allowed_args[] = {
2874                 ENA_DEVARG_LARGE_LLQ_HDR,
2875         };
2876         struct rte_kvargs *kvlist;
2877         int rc;
2878
2879         if (devargs == NULL)
2880                 return 0;
2881
2882         kvlist = rte_kvargs_parse(devargs->args, allowed_args);
2883         if (kvlist == NULL) {
2884                 PMD_INIT_LOG(ERR, "Invalid device arguments: %s\n",
2885                         devargs->args);
2886                 return -EINVAL;
2887         }
2888
2889         rc = rte_kvargs_process(kvlist, ENA_DEVARG_LARGE_LLQ_HDR,
2890                 ena_process_bool_devarg, adapter);
2891
2892         rte_kvargs_free(kvlist);
2893
2894         return rc;
2895 }
2896
2897 /*********************************************************************
2898  *  PMD configuration
2899  *********************************************************************/
2900 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2901         struct rte_pci_device *pci_dev)
2902 {
2903         return rte_eth_dev_pci_generic_probe(pci_dev,
2904                 sizeof(struct ena_adapter), eth_ena_dev_init);
2905 }
2906
2907 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
2908 {
2909         return rte_eth_dev_pci_generic_remove(pci_dev, eth_ena_dev_uninit);
2910 }
2911
2912 static struct rte_pci_driver rte_ena_pmd = {
2913         .id_table = pci_id_ena_map,
2914         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
2915                      RTE_PCI_DRV_WC_ACTIVATE,
2916         .probe = eth_ena_pci_probe,
2917         .remove = eth_ena_pci_remove,
2918 };
2919
2920 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
2921 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
2922 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
2923 RTE_PMD_REGISTER_PARAM_STRING(net_ena, ENA_DEVARG_LARGE_LLQ_HDR "=<0|1>");
2924 RTE_LOG_REGISTER(ena_logtype_init, pmd.net.ena.init, NOTICE);
2925 RTE_LOG_REGISTER(ena_logtype_driver, pmd.net.ena.driver, NOTICE);
2926 #ifdef RTE_LIBRTE_ENA_DEBUG_RX
2927 RTE_LOG_REGISTER(ena_logtype_rx, pmd.net.ena.rx, NOTICE);
2928 #endif
2929 #ifdef RTE_LIBRTE_ENA_DEBUG_TX
2930 RTE_LOG_REGISTER(ena_logtype_tx, pmd.net.ena.tx, NOTICE);
2931 #endif
2932 #ifdef RTE_LIBRTE_ENA_DEBUG_TX_FREE
2933 RTE_LOG_REGISTER(ena_logtype_tx_free, pmd.net.ena.tx_free, NOTICE);
2934 #endif
2935 #ifdef RTE_LIBRTE_ENA_COM_DEBUG
2936 RTE_LOG_REGISTER(ena_logtype_com, pmd.net.ena.com, NOTICE);
2937 #endif
2938
2939 /******************************************************************************
2940  ******************************** AENQ Handlers *******************************
2941  *****************************************************************************/
2942 static void ena_update_on_link_change(void *adapter_data,
2943                                       struct ena_admin_aenq_entry *aenq_e)
2944 {
2945         struct rte_eth_dev *eth_dev;
2946         struct ena_adapter *adapter;
2947         struct ena_admin_aenq_link_change_desc *aenq_link_desc;
2948         uint32_t status;
2949
2950         adapter = adapter_data;
2951         aenq_link_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
2952         eth_dev = adapter->rte_dev;
2953
2954         status = get_ena_admin_aenq_link_change_desc_link_status(aenq_link_desc);
2955         adapter->link_status = status;
2956
2957         ena_link_update(eth_dev, 0);
2958         rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
2959 }
2960
2961 static void ena_notification(void *data,
2962                              struct ena_admin_aenq_entry *aenq_e)
2963 {
2964         struct ena_adapter *adapter = data;
2965         struct ena_admin_ena_hw_hints *hints;
2966
2967         if (aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION)
2968                 PMD_DRV_LOG(WARNING, "Invalid group(%x) expected %x\n",
2969                         aenq_e->aenq_common_desc.group,
2970                         ENA_ADMIN_NOTIFICATION);
2971
2972         switch (aenq_e->aenq_common_desc.syndrom) {
2973         case ENA_ADMIN_UPDATE_HINTS:
2974                 hints = (struct ena_admin_ena_hw_hints *)
2975                         (&aenq_e->inline_data_w4);
2976                 ena_update_hints(adapter, hints);
2977                 break;
2978         default:
2979                 PMD_DRV_LOG(ERR, "Invalid aenq notification link state %d\n",
2980                         aenq_e->aenq_common_desc.syndrom);
2981         }
2982 }
2983
2984 static void ena_keep_alive(void *adapter_data,
2985                            __rte_unused struct ena_admin_aenq_entry *aenq_e)
2986 {
2987         struct ena_adapter *adapter = adapter_data;
2988         struct ena_admin_aenq_keep_alive_desc *desc;
2989         uint64_t rx_drops;
2990         uint64_t tx_drops;
2991
2992         adapter->timestamp_wd = rte_get_timer_cycles();
2993
2994         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
2995         rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
2996         tx_drops = ((uint64_t)desc->tx_drops_high << 32) | desc->tx_drops_low;
2997
2998         adapter->drv_stats->rx_drops = rx_drops;
2999         adapter->dev_stats.tx_drops = tx_drops;
3000 }
3001
3002 /**
3003  * This handler will called for unknown event group or unimplemented handlers
3004  **/
3005 static void unimplemented_aenq_handler(__rte_unused void *data,
3006                                        __rte_unused struct ena_admin_aenq_entry *aenq_e)
3007 {
3008         PMD_DRV_LOG(ERR, "Unknown event was received or event with "
3009                           "unimplemented handler\n");
3010 }
3011
3012 static struct ena_aenq_handlers aenq_handlers = {
3013         .handlers = {
3014                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3015                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
3016                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive
3017         },
3018         .unimplemented_handler = unimplemented_aenq_handler
3019 };