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    3
  31 #define DRV_MODULE_VER_SUBMINOR 0
  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) RTE_DIM(x)
  51
  52 #define ENA_MIN_RING_DESC       128
  53
  54 #define ENA_PTYPE_HAS_HASH      (RTE_PTYPE_L4_TCP | RTE_PTYPE_L4_UDP)
  55
  56 enum ethtool_stringset {
  57         ETH_SS_TEST             = 0,
  58         ETH_SS_STATS,
  59 };
  60
  61 struct ena_stats {
  62         char name[ETH_GSTRING_LEN];
  63         int stat_offset;
  64 };
  65
  66 #define ENA_STAT_ENTRY(stat, stat_type) { \
  67         .name = #stat, \
  68         .stat_offset = offsetof(struct ena_stats_##stat_type, stat) \
  69 }
  70
  71 #define ENA_STAT_RX_ENTRY(stat) \
  72         ENA_STAT_ENTRY(stat, rx)
  73
  74 #define ENA_STAT_TX_ENTRY(stat) \
  75         ENA_STAT_ENTRY(stat, tx)
  76
  77 #define ENA_STAT_ENI_ENTRY(stat) \
  78         ENA_STAT_ENTRY(stat, eni)
  79
  80 #define ENA_STAT_GLOBAL_ENTRY(stat) \
  81         ENA_STAT_ENTRY(stat, dev)
  82
  83 /* Device arguments */
  84 #define ENA_DEVARG_LARGE_LLQ_HDR "large_llq_hdr"
  85
  86 /*
  87  * Each rte_memzone should have unique name.
  88  * To satisfy it, count number of allocation and add it to name.
  89  */
  90 rte_atomic64_t ena_alloc_cnt;
  91
  92 static const struct ena_stats ena_stats_global_strings[] = {
  93         ENA_STAT_GLOBAL_ENTRY(wd_expired),
  94         ENA_STAT_GLOBAL_ENTRY(dev_start),
  95         ENA_STAT_GLOBAL_ENTRY(dev_stop),
  96         ENA_STAT_GLOBAL_ENTRY(tx_drops),
  97 };
  98
  99 static const struct ena_stats ena_stats_eni_strings[] = {
 100         ENA_STAT_ENI_ENTRY(bw_in_allowance_exceeded),
 101         ENA_STAT_ENI_ENTRY(bw_out_allowance_exceeded),
 102         ENA_STAT_ENI_ENTRY(pps_allowance_exceeded),
 103         ENA_STAT_ENI_ENTRY(conntrack_allowance_exceeded),
 104         ENA_STAT_ENI_ENTRY(linklocal_allowance_exceeded),
 105 };
 106
 107 static const struct ena_stats ena_stats_tx_strings[] = {
 108         ENA_STAT_TX_ENTRY(cnt),
 109         ENA_STAT_TX_ENTRY(bytes),
 110         ENA_STAT_TX_ENTRY(prepare_ctx_err),
 111         ENA_STAT_TX_ENTRY(linearize),
 112         ENA_STAT_TX_ENTRY(linearize_failed),
 113         ENA_STAT_TX_ENTRY(tx_poll),
 114         ENA_STAT_TX_ENTRY(doorbells),
 115         ENA_STAT_TX_ENTRY(bad_req_id),
 116         ENA_STAT_TX_ENTRY(available_desc),
 117 };
 118
 119 static const struct ena_stats ena_stats_rx_strings[] = {
 120         ENA_STAT_RX_ENTRY(cnt),
 121         ENA_STAT_RX_ENTRY(bytes),
 122         ENA_STAT_RX_ENTRY(refill_partial),
 123         ENA_STAT_RX_ENTRY(bad_csum),
 124         ENA_STAT_RX_ENTRY(mbuf_alloc_fail),
 125         ENA_STAT_RX_ENTRY(bad_desc_num),
 126         ENA_STAT_RX_ENTRY(bad_req_id),
 127 };
 128
 129 #define ENA_STATS_ARRAY_GLOBAL  ARRAY_SIZE(ena_stats_global_strings)
 130 #define ENA_STATS_ARRAY_ENI     ARRAY_SIZE(ena_stats_eni_strings)
 131 #define ENA_STATS_ARRAY_TX      ARRAY_SIZE(ena_stats_tx_strings)
 132 #define ENA_STATS_ARRAY_RX      ARRAY_SIZE(ena_stats_rx_strings)
 133
 134 #define QUEUE_OFFLOADS (DEV_TX_OFFLOAD_TCP_CKSUM |\
 135                         DEV_TX_OFFLOAD_UDP_CKSUM |\
 136                         DEV_TX_OFFLOAD_IPV4_CKSUM |\
 137                         DEV_TX_OFFLOAD_TCP_TSO)
 138 #define MBUF_OFFLOADS (PKT_TX_L4_MASK |\
 139                        PKT_TX_IP_CKSUM |\
 140                        PKT_TX_TCP_SEG)
 141
 142 /** Vendor ID used by Amazon devices */
 143 #define PCI_VENDOR_ID_AMAZON 0x1D0F
 144 /** Amazon devices */
 145 #define PCI_DEVICE_ID_ENA_VF            0xEC20
 146 #define PCI_DEVICE_ID_ENA_VF_RSERV0     0xEC21
 147
 148 #define ENA_TX_OFFLOAD_MASK     (\
 149         PKT_TX_L4_MASK |         \
 150         PKT_TX_IPV6 |            \
 151         PKT_TX_IPV4 |            \
 152         PKT_TX_IP_CKSUM |        \
 153         PKT_TX_TCP_SEG)
 154
 155 #define ENA_TX_OFFLOAD_NOTSUP_MASK      \
 156         (PKT_TX_OFFLOAD_MASK ^ ENA_TX_OFFLOAD_MASK)
 157
 158 static const struct rte_pci_id pci_id_ena_map[] = {
 159         { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF) },
 160         { RTE_PCI_DEVICE(PCI_VENDOR_ID_AMAZON, PCI_DEVICE_ID_ENA_VF_RSERV0) },
 161         { .device_id = 0 },
 162 };
 163
 164 static struct ena_aenq_handlers aenq_handlers;
 165
 166 static int ena_device_init(struct ena_com_dev *ena_dev,
 167                            struct rte_pci_device *pdev,
 168                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
 169                            bool *wd_state);
 170 static int ena_dev_configure(struct rte_eth_dev *dev);
 171 static void ena_tx_map_mbuf(struct ena_ring *tx_ring,
 172         struct ena_tx_buffer *tx_info,
 173         struct rte_mbuf *mbuf,
 174         void **push_header,
 175         uint16_t *header_len);
 176 static int ena_xmit_mbuf(struct ena_ring *tx_ring, struct rte_mbuf *mbuf);
 177 static void ena_tx_cleanup(struct ena_ring *tx_ring);
 178 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 179                                   uint16_t nb_pkts);
 180 static uint16_t eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
 181                 uint16_t nb_pkts);
 182 static int ena_tx_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_txconf *tx_conf);
 185 static int ena_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
 186                               uint16_t nb_desc, unsigned int socket_id,
 187                               const struct rte_eth_rxconf *rx_conf,
 188                               struct rte_mempool *mp);
 189 static inline void ena_init_rx_mbuf(struct rte_mbuf *mbuf, uint16_t len);
 190 static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
 191                                     struct ena_com_rx_buf_info *ena_bufs,
 192                                     uint32_t descs,
 193                                     uint16_t *next_to_clean,
 194                                     uint8_t offset);
 195 static uint16_t eth_ena_recv_pkts(void *rx_queue,
 196                                   struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
 197 static int ena_add_single_rx_desc(struct ena_com_io_sq *io_sq,
 198                                   struct rte_mbuf *mbuf, uint16_t id);
 199 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count);
 200 static void ena_init_rings(struct ena_adapter *adapter,
 201                            bool disable_meta_caching);
 202 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu);
 203 static int ena_start(struct rte_eth_dev *dev);
 204 static int ena_stop(struct rte_eth_dev *dev);
 205 static int ena_close(struct rte_eth_dev *dev);
 206 static int ena_dev_reset(struct rte_eth_dev *dev);
 207 static int ena_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats);
 208 static void ena_rx_queue_release_all(struct rte_eth_dev *dev);
 209 static void ena_tx_queue_release_all(struct rte_eth_dev *dev);
 210 static void ena_rx_queue_release(void *queue);
 211 static void ena_tx_queue_release(void *queue);
 212 static void ena_rx_queue_release_bufs(struct ena_ring *ring);
 213 static void ena_tx_queue_release_bufs(struct ena_ring *ring);
 214 static int ena_link_update(struct rte_eth_dev *dev,
 215                            int wait_to_complete);
 216 static int ena_create_io_queue(struct ena_ring *ring);
 217 static void ena_queue_stop(struct ena_ring *ring);
 218 static void ena_queue_stop_all(struct rte_eth_dev *dev,
 219                               enum ena_ring_type ring_type);
 220 static int ena_queue_start(struct ena_ring *ring);
 221 static int ena_queue_start_all(struct rte_eth_dev *dev,
 222                                enum ena_ring_type ring_type);
 223 static void ena_stats_restart(struct rte_eth_dev *dev);
 224 static int ena_infos_get(struct rte_eth_dev *dev,
 225                          struct rte_eth_dev_info *dev_info);
 226 static int ena_rss_reta_update(struct rte_eth_dev *dev,
 227                                struct rte_eth_rss_reta_entry64 *reta_conf,
 228                                uint16_t reta_size);
 229 static int ena_rss_reta_query(struct rte_eth_dev *dev,
 230                               struct rte_eth_rss_reta_entry64 *reta_conf,
 231                               uint16_t reta_size);
 232 static void ena_interrupt_handler_rte(void *cb_arg);
 233 static void ena_timer_wd_callback(struct rte_timer *timer, void *arg);
 234 static void ena_destroy_device(struct rte_eth_dev *eth_dev);
 235 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev);
 236 static int ena_xstats_get_names(struct rte_eth_dev *dev,
 237                                 struct rte_eth_xstat_name *xstats_names,
 238                                 unsigned int n);
 239 static int ena_xstats_get(struct rte_eth_dev *dev,
 240                           struct rte_eth_xstat *stats,
 241                           unsigned int n);
 242 static int ena_xstats_get_by_id(struct rte_eth_dev *dev,
 243                                 const uint64_t *ids,
 244                                 uint64_t *values,
 245                                 unsigned int n);
 246 static int ena_process_bool_devarg(const char *key,
 247                                    const char *value,
 248                                    void *opaque);
 249 static int ena_parse_devargs(struct ena_adapter *adapter,
 250                              struct rte_devargs *devargs);
 251 static int ena_copy_eni_stats(struct ena_adapter *adapter);
 252
 253 static const struct eth_dev_ops ena_dev_ops = {
 254         .dev_configure        = ena_dev_configure,
 255         .dev_infos_get        = ena_infos_get,
 256         .rx_queue_setup       = ena_rx_queue_setup,
 257         .tx_queue_setup       = ena_tx_queue_setup,
 258         .dev_start            = ena_start,
 259         .dev_stop             = ena_stop,
 260         .link_update          = ena_link_update,
 261         .stats_get            = ena_stats_get,
 262         .xstats_get_names     = ena_xstats_get_names,
 263         .xstats_get           = ena_xstats_get,
 264         .xstats_get_by_id     = ena_xstats_get_by_id,
 265         .mtu_set              = ena_mtu_set,
 266         .rx_queue_release     = ena_rx_queue_release,
 267         .tx_queue_release     = ena_tx_queue_release,
 268         .dev_close            = ena_close,
 269         .dev_reset            = ena_dev_reset,
 270         .reta_update          = ena_rss_reta_update,
 271         .reta_query           = ena_rss_reta_query,
 272 };
 273
 274 void ena_rss_key_fill(void *key, size_t size)
 275 {
 276         static bool key_generated;
 277         static uint8_t default_key[ENA_HASH_KEY_SIZE];
 278         size_t i;
 279
 280         RTE_ASSERT(size <= ENA_HASH_KEY_SIZE);
 281
 282         if (!key_generated) {
 283                 for (i = 0; i < ENA_HASH_KEY_SIZE; ++i)
 284                         default_key[i] = rte_rand() & 0xff;
 285                 key_generated = true;
 286         }
 287
 288         rte_memcpy(key, default_key, size);
 289 }
 290
 291 static inline void ena_rx_mbuf_prepare(struct rte_mbuf *mbuf,
 292                                        struct ena_com_rx_ctx *ena_rx_ctx)
 293 {
 294         uint64_t ol_flags = 0;
 295         uint32_t packet_type = 0;
 296
 297         if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP)
 298                 packet_type |= RTE_PTYPE_L4_TCP;
 299         else if (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)
 300                 packet_type |= RTE_PTYPE_L4_UDP;
 301
 302         if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) {
 303                 packet_type |= RTE_PTYPE_L3_IPV4;
 304                 if (unlikely(ena_rx_ctx->l3_csum_err))
 305                         ol_flags |= PKT_RX_IP_CKSUM_BAD;
 306                 else
 307                         ol_flags |= PKT_RX_IP_CKSUM_GOOD;
 308         } else if (ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV6) {
 309                 packet_type |= RTE_PTYPE_L3_IPV6;
 310         }
 311
 312         if (!ena_rx_ctx->l4_csum_checked || ena_rx_ctx->frag)
 313                 ol_flags |= PKT_RX_L4_CKSUM_UNKNOWN;
 314         else
 315                 if (unlikely(ena_rx_ctx->l4_csum_err))
 316                         ol_flags |= PKT_RX_L4_CKSUM_BAD;
 317                 else
 318                         ol_flags |= PKT_RX_L4_CKSUM_GOOD;
 319
 320         if (likely((packet_type & ENA_PTYPE_HAS_HASH) && !ena_rx_ctx->frag)) {
 321                 ol_flags |= PKT_RX_RSS_HASH;
 322                 mbuf->hash.rss = ena_rx_ctx->hash;
 323         }
 324
 325         mbuf->ol_flags = ol_flags;
 326         mbuf->packet_type = packet_type;
 327 }
 328
 329 static inline void ena_tx_mbuf_prepare(struct rte_mbuf *mbuf,
 330                                        struct ena_com_tx_ctx *ena_tx_ctx,
 331                                        uint64_t queue_offloads,
 332                                        bool disable_meta_caching)
 333 {
 334         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
 335
 336         if ((mbuf->ol_flags & MBUF_OFFLOADS) &&
 337             (queue_offloads & QUEUE_OFFLOADS)) {
 338                 /* check if TSO is required */
 339                 if ((mbuf->ol_flags & PKT_TX_TCP_SEG) &&
 340                     (queue_offloads & DEV_TX_OFFLOAD_TCP_TSO)) {
 341                         ena_tx_ctx->tso_enable = true;
 342
 343                         ena_meta->l4_hdr_len = GET_L4_HDR_LEN(mbuf);
 344                 }
 345
 346                 /* check if L3 checksum is needed */
 347                 if ((mbuf->ol_flags & PKT_TX_IP_CKSUM) &&
 348                     (queue_offloads & DEV_TX_OFFLOAD_IPV4_CKSUM))
 349                         ena_tx_ctx->l3_csum_enable = true;
 350
 351                 if (mbuf->ol_flags & PKT_TX_IPV6) {
 352                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
 353                 } else {
 354                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
 355
 356                         /* set don't fragment (DF) flag */
 357                         if (mbuf->packet_type &
 358                                 (RTE_PTYPE_L4_NONFRAG
 359                                  | RTE_PTYPE_INNER_L4_NONFRAG))
 360                                 ena_tx_ctx->df = true;
 361                 }
 362
 363                 /* check if L4 checksum is needed */
 364                 if (((mbuf->ol_flags & PKT_TX_L4_MASK) == PKT_TX_TCP_CKSUM) &&
 365                     (queue_offloads & DEV_TX_OFFLOAD_TCP_CKSUM)) {
 366                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
 367                         ena_tx_ctx->l4_csum_enable = true;
 368                 } else if (((mbuf->ol_flags & PKT_TX_L4_MASK) ==
 369                                 PKT_TX_UDP_CKSUM) &&
 370                                 (queue_offloads & DEV_TX_OFFLOAD_UDP_CKSUM)) {
 371                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
 372                         ena_tx_ctx->l4_csum_enable = true;
 373                 } else {
 374                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UNKNOWN;
 375                         ena_tx_ctx->l4_csum_enable = false;
 376                 }
 377
 378                 ena_meta->mss = mbuf->tso_segsz;
 379                 ena_meta->l3_hdr_len = mbuf->l3_len;
 380                 ena_meta->l3_hdr_offset = mbuf->l2_len;
 381
 382                 ena_tx_ctx->meta_valid = true;
 383         } else if (disable_meta_caching) {
 384                 memset(ena_meta, 0, sizeof(*ena_meta));
 385                 ena_tx_ctx->meta_valid = true;
 386         } else {
 387                 ena_tx_ctx->meta_valid = false;
 388         }
 389 }
 390
 391 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
 392 {
 393         struct ena_tx_buffer *tx_info = NULL;
 394
 395         if (likely(req_id < tx_ring->ring_size)) {
 396                 tx_info = &tx_ring->tx_buffer_info[req_id];
 397                 if (likely(tx_info->mbuf))
 398                         return 0;
 399         }
 400
 401         if (tx_info)
 402                 PMD_DRV_LOG(ERR, "tx_info doesn't have valid mbuf\n");
 403         else
 404                 PMD_DRV_LOG(ERR, "Invalid req_id: %hu\n", req_id);
 405
 406         /* Trigger device reset */
 407         ++tx_ring->tx_stats.bad_req_id;
 408         tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
 409         tx_ring->adapter->trigger_reset = true;
 410         return -EFAULT;
 411 }
 412
 413 static void ena_config_host_info(struct ena_com_dev *ena_dev)
 414 {
 415         struct ena_admin_host_info *host_info;
 416         int rc;
 417
 418         /* Allocate only the host info */
 419         rc = ena_com_allocate_host_info(ena_dev);
 420         if (rc) {
 421                 PMD_DRV_LOG(ERR, "Cannot allocate host info\n");
 422                 return;
 423         }
 424
 425         host_info = ena_dev->host_attr.host_info;
 426
 427         host_info->os_type = ENA_ADMIN_OS_DPDK;
 428         host_info->kernel_ver = RTE_VERSION;
 429         strlcpy((char *)host_info->kernel_ver_str, rte_version(),
 430                 sizeof(host_info->kernel_ver_str));
 431         host_info->os_dist = RTE_VERSION;
 432         strlcpy((char *)host_info->os_dist_str, rte_version(),
 433                 sizeof(host_info->os_dist_str));
 434         host_info->driver_version =
 435                 (DRV_MODULE_VER_MAJOR) |
 436                 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
 437                 (DRV_MODULE_VER_SUBMINOR <<
 438                         ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
 439         host_info->num_cpus = rte_lcore_count();
 440
 441         host_info->driver_supported_features =
 442                 ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK;
 443
 444         rc = ena_com_set_host_attributes(ena_dev);
 445         if (rc) {
 446                 if (rc == -ENA_COM_UNSUPPORTED)
 447                         PMD_DRV_LOG(WARNING, "Cannot set host attributes\n");
 448                 else
 449                         PMD_DRV_LOG(ERR, "Cannot set host attributes\n");
 450
 451                 goto err;
 452         }
 453
 454         return;
 455
 456 err:
 457         ena_com_delete_host_info(ena_dev);
 458 }
 459
 460 /* This function calculates the number of xstats based on the current config */
 461 static unsigned int ena_xstats_calc_num(struct rte_eth_dev_data *data)
 462 {
 463         return ENA_STATS_ARRAY_GLOBAL + ENA_STATS_ARRAY_ENI +
 464                 (data->nb_tx_queues * ENA_STATS_ARRAY_TX) +
 465                 (data->nb_rx_queues * ENA_STATS_ARRAY_RX);
 466 }
 467
 468 static void ena_config_debug_area(struct ena_adapter *adapter)
 469 {
 470         u32 debug_area_size;
 471         int rc, ss_count;
 472
 473         ss_count = ena_xstats_calc_num(adapter->edev_data);
 474
 475         /* allocate 32 bytes for each string and 64bit for the value */
 476         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
 477
 478         rc = ena_com_allocate_debug_area(&adapter->ena_dev, debug_area_size);
 479         if (rc) {
 480                 PMD_DRV_LOG(ERR, "Cannot allocate debug area\n");
 481                 return;
 482         }
 483
 484         rc = ena_com_set_host_attributes(&adapter->ena_dev);
 485         if (rc) {
 486                 if (rc == -ENA_COM_UNSUPPORTED)
 487                         PMD_DRV_LOG(WARNING, "Cannot set host attributes\n");
 488                 else
 489                         PMD_DRV_LOG(ERR, "Cannot set host attributes\n");
 490
 491                 goto err;
 492         }
 493
 494         return;
 495 err:
 496         ena_com_delete_debug_area(&adapter->ena_dev);
 497 }
 498
 499 static int ena_close(struct rte_eth_dev *dev)
 500 {
 501         struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(dev);
 502         struct rte_intr_handle *intr_handle = &pci_dev->intr_handle;
 503         struct ena_adapter *adapter = dev->data->dev_private;
 504         int ret = 0;
 505
 506         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 507                 return 0;
 508
 509         if (adapter->state == ENA_ADAPTER_STATE_RUNNING)
 510                 ret = ena_stop(dev);
 511         adapter->state = ENA_ADAPTER_STATE_CLOSED;
 512
 513         ena_rx_queue_release_all(dev);
 514         ena_tx_queue_release_all(dev);
 515
 516         rte_free(adapter->drv_stats);
 517         adapter->drv_stats = NULL;
 518
 519         rte_intr_disable(intr_handle);
 520         rte_intr_callback_unregister(intr_handle,
 521                                      ena_interrupt_handler_rte,
 522                                      dev);
 523
 524         /*
 525          * MAC is not allocated dynamically. Setting NULL should prevent from
 526          * release of the resource in the rte_eth_dev_release_port().
 527          */
 528         dev->data->mac_addrs = NULL;
 529
 530         return ret;
 531 }
 532
 533 static int
 534 ena_dev_reset(struct rte_eth_dev *dev)
 535 {
 536         int rc = 0;
 537
 538         /* Cannot release memory in secondary process */
 539         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
 540                 PMD_DRV_LOG(WARNING, "dev_reset not supported in secondary.\n");
 541                 return -EPERM;
 542         }
 543
 544         ena_destroy_device(dev);
 545         rc = eth_ena_dev_init(dev);
 546         if (rc)
 547                 PMD_INIT_LOG(CRIT, "Cannot initialize device\n");
 548
 549         return rc;
 550 }
 551
 552 static int ena_rss_reta_update(struct rte_eth_dev *dev,
 553                                struct rte_eth_rss_reta_entry64 *reta_conf,
 554                                uint16_t reta_size)
 555 {
 556         struct ena_adapter *adapter = dev->data->dev_private;
 557         struct ena_com_dev *ena_dev = &adapter->ena_dev;
 558         int rc, i;
 559         u16 entry_value;
 560         int conf_idx;
 561         int idx;
 562
 563         if ((reta_size == 0) || (reta_conf == NULL))
 564                 return -EINVAL;
 565
 566         if (reta_size > ENA_RX_RSS_TABLE_SIZE) {
 567                 PMD_DRV_LOG(WARNING,
 568                         "Requested indirection table size (%d) is bigger than supported: %d\n",
 569                         reta_size, ENA_RX_RSS_TABLE_SIZE);
 570                 return -EINVAL;
 571         }
 572
 573         for (i = 0 ; i < reta_size ; i++) {
 574                 /* each reta_conf is for 64 entries.
 575                  * to support 128 we use 2 conf of 64
 576                  */
 577                 conf_idx = i / RTE_RETA_GROUP_SIZE;
 578                 idx = i % RTE_RETA_GROUP_SIZE;
 579                 if (TEST_BIT(reta_conf[conf_idx].mask, idx)) {
 580                         entry_value =
 581                                 ENA_IO_RXQ_IDX(reta_conf[conf_idx].reta[idx]);
 582
 583                         rc = ena_com_indirect_table_fill_entry(ena_dev,
 584                                                                i,
 585                                                                entry_value);
 586                         if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
 587                                 PMD_DRV_LOG(ERR,
 588                                         "Cannot fill indirect table\n");
 589                                 return rc;
 590                         }
 591                 }
 592         }
 593
 594         rte_spinlock_lock(&adapter->admin_lock);
 595         rc = ena_com_indirect_table_set(ena_dev);
 596         rte_spinlock_unlock(&adapter->admin_lock);
 597         if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
 598                 PMD_DRV_LOG(ERR, "Cannot flush the indirect table\n");
 599                 return rc;
 600         }
 601
 602         PMD_DRV_LOG(DEBUG, "RSS configured %d entries for port %d\n",
 603                 reta_size, dev->data->port_id);
 604
 605         return 0;
 606 }
 607
 608 /* Query redirection table. */
 609 static int ena_rss_reta_query(struct rte_eth_dev *dev,
 610                               struct rte_eth_rss_reta_entry64 *reta_conf,
 611                               uint16_t reta_size)
 612 {
 613         struct ena_adapter *adapter = dev->data->dev_private;
 614         struct ena_com_dev *ena_dev = &adapter->ena_dev;
 615         int rc;
 616         int i;
 617         u32 indirect_table[ENA_RX_RSS_TABLE_SIZE] = {0};
 618         int reta_conf_idx;
 619         int reta_idx;
 620
 621         if (reta_size == 0 || reta_conf == NULL ||
 622             (reta_size > RTE_RETA_GROUP_SIZE && ((reta_conf + 1) == NULL)))
 623                 return -EINVAL;
 624
 625         rte_spinlock_lock(&adapter->admin_lock);
 626         rc = ena_com_indirect_table_get(ena_dev, indirect_table);
 627         rte_spinlock_unlock(&adapter->admin_lock);
 628         if (unlikely(rc && rc != ENA_COM_UNSUPPORTED)) {
 629                 PMD_DRV_LOG(ERR, "Cannot get indirection table\n");
 630                 return -ENOTSUP;
 631         }
 632
 633         for (i = 0 ; i < reta_size ; i++) {
 634                 reta_conf_idx = i / RTE_RETA_GROUP_SIZE;
 635                 reta_idx = i % RTE_RETA_GROUP_SIZE;
 636                 if (TEST_BIT(reta_conf[reta_conf_idx].mask, reta_idx))
 637                         reta_conf[reta_conf_idx].reta[reta_idx] =
 638                                 ENA_IO_RXQ_IDX_REV(indirect_table[i]);
 639         }
 640
 641         return 0;
 642 }
 643
 644 static int ena_rss_init_default(struct ena_adapter *adapter)
 645 {
 646         struct ena_com_dev *ena_dev = &adapter->ena_dev;
 647         uint16_t nb_rx_queues = adapter->edev_data->nb_rx_queues;
 648         int rc, i;
 649         u32 val;
 650
 651         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
 652         if (unlikely(rc)) {
 653                 PMD_DRV_LOG(ERR, "Cannot init indirection table\n");
 654                 goto err_rss_init;
 655         }
 656
 657         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
 658                 val = i % nb_rx_queues;
 659                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
 660                                                        ENA_IO_RXQ_IDX(val));
 661                 if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
 662                         PMD_DRV_LOG(ERR, "Cannot fill indirection table\n");
 663                         goto err_fill_indir;
 664                 }
 665         }
 666
 667         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
 668                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
 669         if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
 670                 PMD_DRV_LOG(INFO, "Cannot fill hash function\n");
 671                 goto err_fill_indir;
 672         }
 673
 674         rc = ena_com_set_default_hash_ctrl(ena_dev);
 675         if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
 676                 PMD_DRV_LOG(INFO, "Cannot fill hash control\n");
 677                 goto err_fill_indir;
 678         }
 679
 680         rc = ena_com_indirect_table_set(ena_dev);
 681         if (unlikely(rc && (rc != ENA_COM_UNSUPPORTED))) {
 682                 PMD_DRV_LOG(ERR, "Cannot flush indirection table\n");
 683                 goto err_fill_indir;
 684         }
 685         PMD_DRV_LOG(DEBUG, "RSS configured for port %d\n",
 686                 adapter->edev_data->port_id);
 687
 688         return 0;
 689
 690 err_fill_indir:
 691         ena_com_rss_destroy(ena_dev);
 692 err_rss_init:
 693
 694         return rc;
 695 }
 696
 697 static void ena_rx_queue_release_all(struct rte_eth_dev *dev)
 698 {
 699         struct ena_ring **queues = (struct ena_ring **)dev->data->rx_queues;
 700         int nb_queues = dev->data->nb_rx_queues;
 701         int i;
 702
 703         for (i = 0; i < nb_queues; i++)
 704                 ena_rx_queue_release(queues[i]);
 705 }
 706
 707 static void ena_tx_queue_release_all(struct rte_eth_dev *dev)
 708 {
 709         struct ena_ring **queues = (struct ena_ring **)dev->data->tx_queues;
 710         int nb_queues = dev->data->nb_tx_queues;
 711         int i;
 712
 713         for (i = 0; i < nb_queues; i++)
 714                 ena_tx_queue_release(queues[i]);
 715 }
 716
 717 static void ena_rx_queue_release(void *queue)
 718 {
 719         struct ena_ring *ring = (struct ena_ring *)queue;
 720
 721         /* Free ring resources */
 722         if (ring->rx_buffer_info)
 723                 rte_free(ring->rx_buffer_info);
 724         ring->rx_buffer_info = NULL;
 725
 726         if (ring->rx_refill_buffer)
 727                 rte_free(ring->rx_refill_buffer);
 728         ring->rx_refill_buffer = NULL;
 729
 730         if (ring->empty_rx_reqs)
 731                 rte_free(ring->empty_rx_reqs);
 732         ring->empty_rx_reqs = NULL;
 733
 734         ring->configured = 0;
 735
 736         PMD_DRV_LOG(NOTICE, "Rx queue %d:%d released\n",
 737                 ring->port_id, ring->id);
 738 }
 739
 740 static void ena_tx_queue_release(void *queue)
 741 {
 742         struct ena_ring *ring = (struct ena_ring *)queue;
 743
 744         /* Free ring resources */
 745         if (ring->push_buf_intermediate_buf)
 746                 rte_free(ring->push_buf_intermediate_buf);
 747
 748         if (ring->tx_buffer_info)
 749                 rte_free(ring->tx_buffer_info);
 750
 751         if (ring->empty_tx_reqs)
 752                 rte_free(ring->empty_tx_reqs);
 753
 754         ring->empty_tx_reqs = NULL;
 755         ring->tx_buffer_info = NULL;
 756         ring->push_buf_intermediate_buf = NULL;
 757
 758         ring->configured = 0;
 759
 760         PMD_DRV_LOG(NOTICE, "Tx queue %d:%d released\n",
 761                 ring->port_id, ring->id);
 762 }
 763
 764 static void ena_rx_queue_release_bufs(struct ena_ring *ring)
 765 {
 766         unsigned int i;
 767
 768         for (i = 0; i < ring->ring_size; ++i) {
 769                 struct ena_rx_buffer *rx_info = &ring->rx_buffer_info[i];
 770                 if (rx_info->mbuf) {
 771                         rte_mbuf_raw_free(rx_info->mbuf);
 772                         rx_info->mbuf = NULL;
 773                 }
 774         }
 775 }
 776
 777 static void ena_tx_queue_release_bufs(struct ena_ring *ring)
 778 {
 779         unsigned int i;
 780
 781         for (i = 0; i < ring->ring_size; ++i) {
 782                 struct ena_tx_buffer *tx_buf = &ring->tx_buffer_info[i];
 783
 784                 if (tx_buf->mbuf) {
 785                         rte_pktmbuf_free(tx_buf->mbuf);
 786                         tx_buf->mbuf = NULL;
 787                 }
 788         }
 789 }
 790
 791 static int ena_link_update(struct rte_eth_dev *dev,
 792                            __rte_unused int wait_to_complete)
 793 {
 794         struct rte_eth_link *link = &dev->data->dev_link;
 795         struct ena_adapter *adapter = dev->data->dev_private;
 796
 797         link->link_status = adapter->link_status ? ETH_LINK_UP : ETH_LINK_DOWN;
 798         link->link_speed = ETH_SPEED_NUM_NONE;
 799         link->link_duplex = ETH_LINK_FULL_DUPLEX;
 800
 801         return 0;
 802 }
 803
 804 static int ena_queue_start_all(struct rte_eth_dev *dev,
 805                                enum ena_ring_type ring_type)
 806 {
 807         struct ena_adapter *adapter = dev->data->dev_private;
 808         struct ena_ring *queues = NULL;
 809         int nb_queues;
 810         int i = 0;
 811         int rc = 0;
 812
 813         if (ring_type == ENA_RING_TYPE_RX) {
 814                 queues = adapter->rx_ring;
 815                 nb_queues = dev->data->nb_rx_queues;
 816         } else {
 817                 queues = adapter->tx_ring;
 818                 nb_queues = dev->data->nb_tx_queues;
 819         }
 820         for (i = 0; i < nb_queues; i++) {
 821                 if (queues[i].configured) {
 822                         if (ring_type == ENA_RING_TYPE_RX) {
 823                                 ena_assert_msg(
 824                                         dev->data->rx_queues[i] == &queues[i],
 825                                         "Inconsistent state of Rx queues\n");
 826                         } else {
 827                                 ena_assert_msg(
 828                                         dev->data->tx_queues[i] == &queues[i],
 829                                         "Inconsistent state of Tx queues\n");
 830                         }
 831
 832                         rc = ena_queue_start(&queues[i]);
 833
 834                         if (rc) {
 835                                 PMD_INIT_LOG(ERR,
 836                                         "Failed to start queue[%d] of type(%d)\n",
 837                                         i, ring_type);
 838                                 goto err;
 839                         }
 840                 }
 841         }
 842
 843         return 0;
 844
 845 err:
 846         while (i--)
 847                 if (queues[i].configured)
 848                         ena_queue_stop(&queues[i]);
 849
 850         return rc;
 851 }
 852
 853 static uint32_t ena_get_mtu_conf(struct ena_adapter *adapter)
 854 {
 855         uint32_t max_frame_len = adapter->max_mtu;
 856
 857         if (adapter->edev_data->dev_conf.rxmode.offloads &
 858             DEV_RX_OFFLOAD_JUMBO_FRAME)
 859                 max_frame_len =
 860                         adapter->edev_data->dev_conf.rxmode.max_rx_pkt_len;
 861
 862         return max_frame_len;
 863 }
 864
 865 static int ena_check_valid_conf(struct ena_adapter *adapter)
 866 {
 867         uint32_t max_frame_len = ena_get_mtu_conf(adapter);
 868
 869         if (max_frame_len > adapter->max_mtu || max_frame_len < ENA_MIN_MTU) {
 870                 PMD_INIT_LOG(ERR,
 871                         "Unsupported MTU of %d. Max MTU: %d, min MTU: %d\n",
 872                         max_frame_len, adapter->max_mtu, ENA_MIN_MTU);
 873                 return ENA_COM_UNSUPPORTED;
 874         }
 875
 876         return 0;
 877 }
 878
 879 static int
 880 ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx,
 881                        bool use_large_llq_hdr)
 882 {
 883         struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
 884         struct ena_com_dev *ena_dev = ctx->ena_dev;
 885         uint32_t max_tx_queue_size;
 886         uint32_t max_rx_queue_size;
 887
 888         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
 889                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
 890                         &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
 891                 max_rx_queue_size = RTE_MIN(max_queue_ext->max_rx_cq_depth,
 892                         max_queue_ext->max_rx_sq_depth);
 893                 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
 894
 895                 if (ena_dev->tx_mem_queue_type ==
 896                     ENA_ADMIN_PLACEMENT_POLICY_DEV) {
 897                         max_tx_queue_size = RTE_MIN(max_tx_queue_size,
 898                                 llq->max_llq_depth);
 899                 } else {
 900                         max_tx_queue_size = RTE_MIN(max_tx_queue_size,
 901                                 max_queue_ext->max_tx_sq_depth);
 902                 }
 903
 904                 ctx->max_rx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
 905                         max_queue_ext->max_per_packet_rx_descs);
 906                 ctx->max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
 907                         max_queue_ext->max_per_packet_tx_descs);
 908         } else {
 909                 struct ena_admin_queue_feature_desc *max_queues =
 910                         &ctx->get_feat_ctx->max_queues;
 911                 max_rx_queue_size = RTE_MIN(max_queues->max_cq_depth,
 912                         max_queues->max_sq_depth);
 913                 max_tx_queue_size = max_queues->max_cq_depth;
 914
 915                 if (ena_dev->tx_mem_queue_type ==
 916                     ENA_ADMIN_PLACEMENT_POLICY_DEV) {
 917                         max_tx_queue_size = RTE_MIN(max_tx_queue_size,
 918                                 llq->max_llq_depth);
 919                 } else {
 920                         max_tx_queue_size = RTE_MIN(max_tx_queue_size,
 921                                 max_queues->max_sq_depth);
 922                 }
 923
 924                 ctx->max_rx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
 925                         max_queues->max_packet_rx_descs);
 926                 ctx->max_tx_sgl_size = RTE_MIN(ENA_PKT_MAX_BUFS,
 927                         max_queues->max_packet_tx_descs);
 928         }
 929
 930         /* Round down to the nearest power of 2 */
 931         max_rx_queue_size = rte_align32prevpow2(max_rx_queue_size);
 932         max_tx_queue_size = rte_align32prevpow2(max_tx_queue_size);
 933
 934         if (use_large_llq_hdr) {
 935                 if ((llq->entry_size_ctrl_supported &
 936                      ENA_ADMIN_LIST_ENTRY_SIZE_256B) &&
 937                     (ena_dev->tx_mem_queue_type ==
 938                      ENA_ADMIN_PLACEMENT_POLICY_DEV)) {
 939                         max_tx_queue_size /= 2;
 940                         PMD_INIT_LOG(INFO,
 941                                 "Forcing large headers and decreasing maximum Tx queue size to %d\n",
 942                                 max_tx_queue_size);
 943                 } else {
 944                         PMD_INIT_LOG(ERR,
 945                                 "Forcing large headers failed: LLQ is disabled or device does not support large headers\n");
 946                 }
 947         }
 948
 949         if (unlikely(max_rx_queue_size == 0 || max_tx_queue_size == 0)) {
 950                 PMD_INIT_LOG(ERR, "Invalid queue size\n");
 951                 return -EFAULT;
 952         }
 953
 954         ctx->max_tx_queue_size = max_tx_queue_size;
 955         ctx->max_rx_queue_size = max_rx_queue_size;
 956
 957         return 0;
 958 }
 959
 960 static void ena_stats_restart(struct rte_eth_dev *dev)
 961 {
 962         struct ena_adapter *adapter = dev->data->dev_private;
 963
 964         rte_atomic64_init(&adapter->drv_stats->ierrors);
 965         rte_atomic64_init(&adapter->drv_stats->oerrors);
 966         rte_atomic64_init(&adapter->drv_stats->rx_nombuf);
 967         adapter->drv_stats->rx_drops = 0;
 968 }
 969
 970 static int ena_stats_get(struct rte_eth_dev *dev,
 971                           struct rte_eth_stats *stats)
 972 {
 973         struct ena_admin_basic_stats ena_stats;
 974         struct ena_adapter *adapter = dev->data->dev_private;
 975         struct ena_com_dev *ena_dev = &adapter->ena_dev;
 976         int rc;
 977         int i;
 978         int max_rings_stats;
 979
 980         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
 981                 return -ENOTSUP;
 982
 983         memset(&ena_stats, 0, sizeof(ena_stats));
 984
 985         rte_spinlock_lock(&adapter->admin_lock);
 986         rc = ena_com_get_dev_basic_stats(ena_dev, &ena_stats);
 987         rte_spinlock_unlock(&adapter->admin_lock);
 988         if (unlikely(rc)) {
 989                 PMD_DRV_LOG(ERR, "Could not retrieve statistics from ENA\n");
 990                 return rc;
 991         }
 992
 993         /* Set of basic statistics from ENA */
 994         stats->ipackets = __MERGE_64B_H_L(ena_stats.rx_pkts_high,
 995                                           ena_stats.rx_pkts_low);
 996         stats->opackets = __MERGE_64B_H_L(ena_stats.tx_pkts_high,
 997                                           ena_stats.tx_pkts_low);
 998         stats->ibytes = __MERGE_64B_H_L(ena_stats.rx_bytes_high,
 999                                         ena_stats.rx_bytes_low);
1000         stats->obytes = __MERGE_64B_H_L(ena_stats.tx_bytes_high,
1001                                         ena_stats.tx_bytes_low);
1002
1003         /* Driver related stats */
1004         stats->imissed = adapter->drv_stats->rx_drops;
1005         stats->ierrors = rte_atomic64_read(&adapter->drv_stats->ierrors);
1006         stats->oerrors = rte_atomic64_read(&adapter->drv_stats->oerrors);
1007         stats->rx_nombuf = rte_atomic64_read(&adapter->drv_stats->rx_nombuf);
1008
1009         max_rings_stats = RTE_MIN(dev->data->nb_rx_queues,
1010                 RTE_ETHDEV_QUEUE_STAT_CNTRS);
1011         for (i = 0; i < max_rings_stats; ++i) {
1012                 struct ena_stats_rx *rx_stats = &adapter->rx_ring[i].rx_stats;
1013
1014                 stats->q_ibytes[i] = rx_stats->bytes;
1015                 stats->q_ipackets[i] = rx_stats->cnt;
1016                 stats->q_errors[i] = rx_stats->bad_desc_num +
1017                         rx_stats->bad_req_id;
1018         }
1019
1020         max_rings_stats = RTE_MIN(dev->data->nb_tx_queues,
1021                 RTE_ETHDEV_QUEUE_STAT_CNTRS);
1022         for (i = 0; i < max_rings_stats; ++i) {
1023                 struct ena_stats_tx *tx_stats = &adapter->tx_ring[i].tx_stats;
1024
1025                 stats->q_obytes[i] = tx_stats->bytes;
1026                 stats->q_opackets[i] = tx_stats->cnt;
1027         }
1028
1029         return 0;
1030 }
1031
1032 static int ena_mtu_set(struct rte_eth_dev *dev, uint16_t mtu)
1033 {
1034         struct ena_adapter *adapter;
1035         struct ena_com_dev *ena_dev;
1036         int rc = 0;
1037
1038         ena_assert_msg(dev->data != NULL, "Uninitialized device\n");
1039         ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device\n");
1040         adapter = dev->data->dev_private;
1041
1042         ena_dev = &adapter->ena_dev;
1043         ena_assert_msg(ena_dev != NULL, "Uninitialized device\n");
1044
1045         if (mtu > ena_get_mtu_conf(adapter) || mtu < ENA_MIN_MTU) {
1046                 PMD_DRV_LOG(ERR,
1047                         "Invalid MTU setting. New MTU: %d, max MTU: %d, min MTU: %d\n",
1048                         mtu, ena_get_mtu_conf(adapter), ENA_MIN_MTU);
1049                 return -EINVAL;
1050         }
1051
1052         rc = ena_com_set_dev_mtu(ena_dev, mtu);
1053         if (rc)
1054                 PMD_DRV_LOG(ERR, "Could not set MTU: %d\n", mtu);
1055         else
1056                 PMD_DRV_LOG(NOTICE, "MTU set to: %d\n", mtu);
1057
1058         return rc;
1059 }
1060
1061 static int ena_start(struct rte_eth_dev *dev)
1062 {
1063         struct ena_adapter *adapter = dev->data->dev_private;
1064         uint64_t ticks;
1065         int rc = 0;
1066
1067         /* Cannot allocate memory in secondary process */
1068         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1069                 PMD_DRV_LOG(WARNING, "dev_start not supported in secondary.\n");
1070                 return -EPERM;
1071         }
1072
1073         rc = ena_check_valid_conf(adapter);
1074         if (rc)
1075                 return rc;
1076
1077         rc = ena_queue_start_all(dev, ENA_RING_TYPE_RX);
1078         if (rc)
1079                 return rc;
1080
1081         rc = ena_queue_start_all(dev, ENA_RING_TYPE_TX);
1082         if (rc)
1083                 goto err_start_tx;
1084
1085         if (adapter->edev_data->dev_conf.rxmode.mq_mode &
1086             ETH_MQ_RX_RSS_FLAG && adapter->edev_data->nb_rx_queues > 0) {
1087                 rc = ena_rss_init_default(adapter);
1088                 if (rc)
1089                         goto err_rss_init;
1090         }
1091
1092         ena_stats_restart(dev);
1093
1094         adapter->timestamp_wd = rte_get_timer_cycles();
1095         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
1096
1097         ticks = rte_get_timer_hz();
1098         rte_timer_reset(&adapter->timer_wd, ticks, PERIODICAL, rte_lcore_id(),
1099                         ena_timer_wd_callback, dev);
1100
1101         ++adapter->dev_stats.dev_start;
1102         adapter->state = ENA_ADAPTER_STATE_RUNNING;
1103
1104         return 0;
1105
1106 err_rss_init:
1107         ena_queue_stop_all(dev, ENA_RING_TYPE_TX);
1108 err_start_tx:
1109         ena_queue_stop_all(dev, ENA_RING_TYPE_RX);
1110         return rc;
1111 }
1112
1113 static int ena_stop(struct rte_eth_dev *dev)
1114 {
1115         struct ena_adapter *adapter = dev->data->dev_private;
1116         struct ena_com_dev *ena_dev = &adapter->ena_dev;
1117         int rc;
1118
1119         /* Cannot free memory in secondary process */
1120         if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
1121                 PMD_DRV_LOG(WARNING, "dev_stop not supported in secondary.\n");
1122                 return -EPERM;
1123         }
1124
1125         rte_timer_stop_sync(&adapter->timer_wd);
1126         ena_queue_stop_all(dev, ENA_RING_TYPE_TX);
1127         ena_queue_stop_all(dev, ENA_RING_TYPE_RX);
1128
1129         if (adapter->trigger_reset) {
1130                 rc = ena_com_dev_reset(ena_dev, adapter->reset_reason);
1131                 if (rc)
1132                         PMD_DRV_LOG(ERR, "Device reset failed, rc: %d\n", rc);
1133         }
1134
1135         ++adapter->dev_stats.dev_stop;
1136         adapter->state = ENA_ADAPTER_STATE_STOPPED;
1137         dev->data->dev_started = 0;
1138
1139         return 0;
1140 }
1141
1142 static int ena_create_io_queue(struct ena_ring *ring)
1143 {
1144         struct ena_adapter *adapter;
1145         struct ena_com_dev *ena_dev;
1146         struct ena_com_create_io_ctx ctx =
1147                 /* policy set to _HOST just to satisfy icc compiler */
1148                 { ENA_ADMIN_PLACEMENT_POLICY_HOST,
1149                   0, 0, 0, 0, 0 };
1150         uint16_t ena_qid;
1151         unsigned int i;
1152         int rc;
1153
1154         adapter = ring->adapter;
1155         ena_dev = &adapter->ena_dev;
1156
1157         if (ring->type == ENA_RING_TYPE_TX) {
1158                 ena_qid = ENA_IO_TXQ_IDX(ring->id);
1159                 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1160                 ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1161                 for (i = 0; i < ring->ring_size; i++)
1162                         ring->empty_tx_reqs[i] = i;
1163         } else {
1164                 ena_qid = ENA_IO_RXQ_IDX(ring->id);
1165                 ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1166                 for (i = 0; i < ring->ring_size; i++)
1167                         ring->empty_rx_reqs[i] = i;
1168         }
1169         ctx.queue_size = ring->ring_size;
1170         ctx.qid = ena_qid;
1171         ctx.msix_vector = -1; /* interrupts not used */
1172         ctx.numa_node = ring->numa_socket_id;
1173
1174         rc = ena_com_create_io_queue(ena_dev, &ctx);
1175         if (rc) {
1176                 PMD_DRV_LOG(ERR,
1177                         "Failed to create IO queue[%d] (qid:%d), rc: %d\n",
1178                         ring->id, ena_qid, rc);
1179                 return rc;
1180         }
1181
1182         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1183                                      &ring->ena_com_io_sq,
1184                                      &ring->ena_com_io_cq);
1185         if (rc) {
1186                 PMD_DRV_LOG(ERR,
1187                         "Failed to get IO queue[%d] handlers, rc: %d\n",
1188                         ring->id, rc);
1189                 ena_com_destroy_io_queue(ena_dev, ena_qid);
1190                 return rc;
1191         }
1192
1193         if (ring->type == ENA_RING_TYPE_TX)
1194                 ena_com_update_numa_node(ring->ena_com_io_cq, ctx.numa_node);
1195
1196         return 0;
1197 }
1198
1199 static void ena_queue_stop(struct ena_ring *ring)
1200 {
1201         struct ena_com_dev *ena_dev = &ring->adapter->ena_dev;
1202
1203         if (ring->type == ENA_RING_TYPE_RX) {
1204                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(ring->id));
1205                 ena_rx_queue_release_bufs(ring);
1206         } else {
1207                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(ring->id));
1208                 ena_tx_queue_release_bufs(ring);
1209         }
1210 }
1211
1212 static void ena_queue_stop_all(struct rte_eth_dev *dev,
1213                               enum ena_ring_type ring_type)
1214 {
1215         struct ena_adapter *adapter = dev->data->dev_private;
1216         struct ena_ring *queues = NULL;
1217         uint16_t nb_queues, i;
1218
1219         if (ring_type == ENA_RING_TYPE_RX) {
1220                 queues = adapter->rx_ring;
1221                 nb_queues = dev->data->nb_rx_queues;
1222         } else {
1223                 queues = adapter->tx_ring;
1224                 nb_queues = dev->data->nb_tx_queues;
1225         }
1226
1227         for (i = 0; i < nb_queues; ++i)
1228                 if (queues[i].configured)
1229                         ena_queue_stop(&queues[i]);
1230 }
1231
1232 static int ena_queue_start(struct ena_ring *ring)
1233 {
1234         int rc, bufs_num;
1235
1236         ena_assert_msg(ring->configured == 1,
1237                        "Trying to start unconfigured queue\n");
1238
1239         rc = ena_create_io_queue(ring);
1240         if (rc) {
1241                 PMD_INIT_LOG(ERR, "Failed to create IO queue\n");
1242                 return rc;
1243         }
1244
1245         ring->next_to_clean = 0;
1246         ring->next_to_use = 0;
1247
1248         if (ring->type == ENA_RING_TYPE_TX) {
1249                 ring->tx_stats.available_desc =
1250                         ena_com_free_q_entries(ring->ena_com_io_sq);
1251                 return 0;
1252         }
1253
1254         bufs_num = ring->ring_size - 1;
1255         rc = ena_populate_rx_queue(ring, bufs_num);
1256         if (rc != bufs_num) {
1257                 ena_com_destroy_io_queue(&ring->adapter->ena_dev,
1258                                          ENA_IO_RXQ_IDX(ring->id));
1259                 PMD_INIT_LOG(ERR, "Failed to populate Rx ring\n");
1260                 return ENA_COM_FAULT;
1261         }
1262         /* Flush per-core RX buffers pools cache as they can be used on other
1263          * cores as well.
1264          */
1265         rte_mempool_cache_flush(NULL, ring->mb_pool);
1266
1267         return 0;
1268 }
1269
1270 static int ena_tx_queue_setup(struct rte_eth_dev *dev,
1271                               uint16_t queue_idx,
1272                               uint16_t nb_desc,
1273                               unsigned int socket_id,
1274                               const struct rte_eth_txconf *tx_conf)
1275 {
1276         struct ena_ring *txq = NULL;
1277         struct ena_adapter *adapter = dev->data->dev_private;
1278         unsigned int i;
1279
1280         txq = &adapter->tx_ring[queue_idx];
1281
1282         if (txq->configured) {
1283                 PMD_DRV_LOG(CRIT,
1284                         "API violation. Queue[%d] is already configured\n",
1285                         queue_idx);
1286                 return ENA_COM_FAULT;
1287         }
1288
1289         if (!rte_is_power_of_2(nb_desc)) {
1290                 PMD_DRV_LOG(ERR,
1291                         "Unsupported size of Tx queue: %d is not a power of 2.\n",
1292                         nb_desc);
1293                 return -EINVAL;
1294         }
1295
1296         if (nb_desc > adapter->max_tx_ring_size) {
1297                 PMD_DRV_LOG(ERR,
1298                         "Unsupported size of Tx queue (max size: %d)\n",
1299                         adapter->max_tx_ring_size);
1300                 return -EINVAL;
1301         }
1302
1303         txq->port_id = dev->data->port_id;
1304         txq->next_to_clean = 0;
1305         txq->next_to_use = 0;
1306         txq->ring_size = nb_desc;
1307         txq->size_mask = nb_desc - 1;
1308         txq->numa_socket_id = socket_id;
1309         txq->pkts_without_db = false;
1310
1311         txq->tx_buffer_info = rte_zmalloc("txq->tx_buffer_info",
1312                                           sizeof(struct ena_tx_buffer) *
1313                                           txq->ring_size,
1314                                           RTE_CACHE_LINE_SIZE);
1315         if (!txq->tx_buffer_info) {
1316                 PMD_DRV_LOG(ERR,
1317                         "Failed to allocate memory for Tx buffer info\n");
1318                 return -ENOMEM;
1319         }
1320
1321         txq->empty_tx_reqs = rte_zmalloc("txq->empty_tx_reqs",
1322                                          sizeof(u16) * txq->ring_size,
1323                                          RTE_CACHE_LINE_SIZE);
1324         if (!txq->empty_tx_reqs) {
1325                 PMD_DRV_LOG(ERR,
1326                         "Failed to allocate memory for empty Tx requests\n");
1327                 rte_free(txq->tx_buffer_info);
1328                 return -ENOMEM;
1329         }
1330
1331         txq->push_buf_intermediate_buf =
1332                 rte_zmalloc("txq->push_buf_intermediate_buf",
1333                             txq->tx_max_header_size,
1334                             RTE_CACHE_LINE_SIZE);
1335         if (!txq->push_buf_intermediate_buf) {
1336                 PMD_DRV_LOG(ERR, "Failed to alloc push buffer for LLQ\n");
1337                 rte_free(txq->tx_buffer_info);
1338                 rte_free(txq->empty_tx_reqs);
1339                 return -ENOMEM;
1340         }
1341
1342         for (i = 0; i < txq->ring_size; i++)
1343                 txq->empty_tx_reqs[i] = i;
1344
1345         if (tx_conf != NULL) {
1346                 txq->offloads =
1347                         tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
1348         }
1349         /* Store pointer to this queue in upper layer */
1350         txq->configured = 1;
1351         dev->data->tx_queues[queue_idx] = txq;
1352
1353         return 0;
1354 }
1355
1356 static int ena_rx_queue_setup(struct rte_eth_dev *dev,
1357                               uint16_t queue_idx,
1358                               uint16_t nb_desc,
1359                               unsigned int socket_id,
1360                               __rte_unused const struct rte_eth_rxconf *rx_conf,
1361                               struct rte_mempool *mp)
1362 {
1363         struct ena_adapter *adapter = dev->data->dev_private;
1364         struct ena_ring *rxq = NULL;
1365         size_t buffer_size;
1366         int i;
1367
1368         rxq = &adapter->rx_ring[queue_idx];
1369         if (rxq->configured) {
1370                 PMD_DRV_LOG(CRIT,
1371                         "API violation. Queue[%d] is already configured\n",
1372                         queue_idx);
1373                 return ENA_COM_FAULT;
1374         }
1375
1376         if (!rte_is_power_of_2(nb_desc)) {
1377                 PMD_DRV_LOG(ERR,
1378                         "Unsupported size of Rx queue: %d is not a power of 2.\n",
1379                         nb_desc);
1380                 return -EINVAL;
1381         }
1382
1383         if (nb_desc > adapter->max_rx_ring_size) {
1384                 PMD_DRV_LOG(ERR,
1385                         "Unsupported size of Rx queue (max size: %d)\n",
1386                         adapter->max_rx_ring_size);
1387                 return -EINVAL;
1388         }
1389
1390         /* ENA isn't supporting buffers smaller than 1400 bytes */
1391         buffer_size = rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
1392         if (buffer_size < ENA_RX_BUF_MIN_SIZE) {
1393                 PMD_DRV_LOG(ERR,
1394                         "Unsupported size of Rx buffer: %zu (min size: %d)\n",
1395                         buffer_size, ENA_RX_BUF_MIN_SIZE);
1396                 return -EINVAL;
1397         }
1398
1399         rxq->port_id = dev->data->port_id;
1400         rxq->next_to_clean = 0;
1401         rxq->next_to_use = 0;
1402         rxq->ring_size = nb_desc;
1403         rxq->size_mask = nb_desc - 1;
1404         rxq->numa_socket_id = socket_id;
1405         rxq->mb_pool = mp;
1406
1407         rxq->rx_buffer_info = rte_zmalloc("rxq->buffer_info",
1408                 sizeof(struct ena_rx_buffer) * nb_desc,
1409                 RTE_CACHE_LINE_SIZE);
1410         if (!rxq->rx_buffer_info) {
1411                 PMD_DRV_LOG(ERR,
1412                         "Failed to allocate memory for Rx buffer info\n");
1413                 return -ENOMEM;
1414         }
1415
1416         rxq->rx_refill_buffer = rte_zmalloc("rxq->rx_refill_buffer",
1417                                             sizeof(struct rte_mbuf *) * nb_desc,
1418                                             RTE_CACHE_LINE_SIZE);
1419
1420         if (!rxq->rx_refill_buffer) {
1421                 PMD_DRV_LOG(ERR,
1422                         "Failed to allocate memory for Rx refill buffer\n");
1423                 rte_free(rxq->rx_buffer_info);
1424                 rxq->rx_buffer_info = NULL;
1425                 return -ENOMEM;
1426         }
1427
1428         rxq->empty_rx_reqs = rte_zmalloc("rxq->empty_rx_reqs",
1429                                          sizeof(uint16_t) * nb_desc,
1430                                          RTE_CACHE_LINE_SIZE);
1431         if (!rxq->empty_rx_reqs) {
1432                 PMD_DRV_LOG(ERR,
1433                         "Failed to allocate memory for empty Rx requests\n");
1434                 rte_free(rxq->rx_buffer_info);
1435                 rxq->rx_buffer_info = NULL;
1436                 rte_free(rxq->rx_refill_buffer);
1437                 rxq->rx_refill_buffer = NULL;
1438                 return -ENOMEM;
1439         }
1440
1441         for (i = 0; i < nb_desc; i++)
1442                 rxq->empty_rx_reqs[i] = i;
1443
1444         /* Store pointer to this queue in upper layer */
1445         rxq->configured = 1;
1446         dev->data->rx_queues[queue_idx] = rxq;
1447
1448         return 0;
1449 }
1450
1451 static int ena_add_single_rx_desc(struct ena_com_io_sq *io_sq,
1452                                   struct rte_mbuf *mbuf, uint16_t id)
1453 {
1454         struct ena_com_buf ebuf;
1455         int rc;
1456
1457         /* prepare physical address for DMA transaction */
1458         ebuf.paddr = mbuf->buf_iova + RTE_PKTMBUF_HEADROOM;
1459         ebuf.len = mbuf->buf_len - RTE_PKTMBUF_HEADROOM;
1460
1461         /* pass resource to device */
1462         rc = ena_com_add_single_rx_desc(io_sq, &ebuf, id);
1463         if (unlikely(rc != 0))
1464                 PMD_DRV_LOG(WARNING, "Failed adding Rx desc\n");
1465
1466         return rc;
1467 }
1468
1469 static int ena_populate_rx_queue(struct ena_ring *rxq, unsigned int count)
1470 {
1471         unsigned int i;
1472         int rc;
1473         uint16_t next_to_use = rxq->next_to_use;
1474         uint16_t in_use, req_id;
1475         struct rte_mbuf **mbufs = rxq->rx_refill_buffer;
1476
1477         if (unlikely(!count))
1478                 return 0;
1479
1480         in_use = rxq->ring_size - 1 -
1481                 ena_com_free_q_entries(rxq->ena_com_io_sq);
1482         ena_assert_msg(((in_use + count) < rxq->ring_size),
1483                 "bad ring state\n");
1484
1485         /* get resources for incoming packets */
1486         rc = rte_pktmbuf_alloc_bulk(rxq->mb_pool, mbufs, count);
1487         if (unlikely(rc < 0)) {
1488                 rte_atomic64_inc(&rxq->adapter->drv_stats->rx_nombuf);
1489                 ++rxq->rx_stats.mbuf_alloc_fail;
1490                 PMD_RX_LOG(DEBUG, "There are not enough free buffers\n");
1491                 return 0;
1492         }
1493
1494         for (i = 0; i < count; i++) {
1495                 struct rte_mbuf *mbuf = mbufs[i];
1496                 struct ena_rx_buffer *rx_info;
1497
1498                 if (likely((i + 4) < count))
1499                         rte_prefetch0(mbufs[i + 4]);
1500
1501                 req_id = rxq->empty_rx_reqs[next_to_use];
1502                 rx_info = &rxq->rx_buffer_info[req_id];
1503
1504                 rc = ena_add_single_rx_desc(rxq->ena_com_io_sq, mbuf, req_id);
1505                 if (unlikely(rc != 0))
1506                         break;
1507
1508                 rx_info->mbuf = mbuf;
1509                 next_to_use = ENA_IDX_NEXT_MASKED(next_to_use, rxq->size_mask);
1510         }
1511
1512         if (unlikely(i < count)) {
1513                 PMD_DRV_LOG(WARNING,
1514                         "Refilled Rx queue[%d] with only %d/%d buffers\n",
1515                         rxq->id, i, count);
1516                 rte_pktmbuf_free_bulk(&mbufs[i], count - i);
1517                 ++rxq->rx_stats.refill_partial;
1518         }
1519
1520         /* When we submitted free recources to device... */
1521         if (likely(i > 0)) {
1522                 /* ...let HW know that it can fill buffers with data. */
1523                 ena_com_write_sq_doorbell(rxq->ena_com_io_sq);
1524
1525                 rxq->next_to_use = next_to_use;
1526         }
1527
1528         return i;
1529 }
1530
1531 static int ena_device_init(struct ena_com_dev *ena_dev,
1532                            struct rte_pci_device *pdev,
1533                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
1534                            bool *wd_state)
1535 {
1536         uint32_t aenq_groups;
1537         int rc;
1538         bool readless_supported;
1539
1540         /* Initialize mmio registers */
1541         rc = ena_com_mmio_reg_read_request_init(ena_dev);
1542         if (rc) {
1543                 PMD_DRV_LOG(ERR, "Failed to init MMIO read less\n");
1544                 return rc;
1545         }
1546
1547         /* The PCIe configuration space revision id indicate if mmio reg
1548          * read is disabled.
1549          */
1550         readless_supported = !(pdev->id.class_id & ENA_MMIO_DISABLE_REG_READ);
1551         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
1552
1553         /* reset device */
1554         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
1555         if (rc) {
1556                 PMD_DRV_LOG(ERR, "Cannot reset device\n");
1557                 goto err_mmio_read_less;
1558         }
1559
1560         /* check FW version */
1561         rc = ena_com_validate_version(ena_dev);
1562         if (rc) {
1563                 PMD_DRV_LOG(ERR, "Device version is too low\n");
1564                 goto err_mmio_read_less;
1565         }
1566
1567         ena_dev->dma_addr_bits = ena_com_get_dma_width(ena_dev);
1568
1569         /* ENA device administration layer init */
1570         rc = ena_com_admin_init(ena_dev, &aenq_handlers);
1571         if (rc) {
1572                 PMD_DRV_LOG(ERR,
1573                         "Cannot initialize ENA admin queue\n");
1574                 goto err_mmio_read_less;
1575         }
1576
1577         /* To enable the msix interrupts the driver needs to know the number
1578          * of queues. So the driver uses polling mode to retrieve this
1579          * information.
1580          */
1581         ena_com_set_admin_polling_mode(ena_dev, true);
1582
1583         ena_config_host_info(ena_dev);
1584
1585         /* Get Device Attributes and features */
1586         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
1587         if (rc) {
1588                 PMD_DRV_LOG(ERR,
1589                         "Cannot get attribute for ENA device, rc: %d\n", rc);
1590                 goto err_admin_init;
1591         }
1592
1593         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
1594                       BIT(ENA_ADMIN_NOTIFICATION) |
1595                       BIT(ENA_ADMIN_KEEP_ALIVE) |
1596                       BIT(ENA_ADMIN_FATAL_ERROR) |
1597                       BIT(ENA_ADMIN_WARNING);
1598
1599         aenq_groups &= get_feat_ctx->aenq.supported_groups;
1600         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
1601         if (rc) {
1602                 PMD_DRV_LOG(ERR, "Cannot configure AENQ groups, rc: %d\n", rc);
1603                 goto err_admin_init;
1604         }
1605
1606         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
1607
1608         return 0;
1609
1610 err_admin_init:
1611         ena_com_admin_destroy(ena_dev);
1612
1613 err_mmio_read_less:
1614         ena_com_mmio_reg_read_request_destroy(ena_dev);
1615
1616         return rc;
1617 }
1618
1619 static void ena_interrupt_handler_rte(void *cb_arg)
1620 {
1621         struct rte_eth_dev *dev = cb_arg;
1622         struct ena_adapter *adapter = dev->data->dev_private;
1623         struct ena_com_dev *ena_dev = &adapter->ena_dev;
1624
1625         ena_com_admin_q_comp_intr_handler(ena_dev);
1626         if (likely(adapter->state != ENA_ADAPTER_STATE_CLOSED))
1627                 ena_com_aenq_intr_handler(ena_dev, dev);
1628 }
1629
1630 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
1631 {
1632         if (!adapter->wd_state)
1633                 return;
1634
1635         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
1636                 return;
1637
1638         if (unlikely((rte_get_timer_cycles() - adapter->timestamp_wd) >=
1639             adapter->keep_alive_timeout)) {
1640                 PMD_DRV_LOG(ERR, "Keep alive timeout\n");
1641                 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
1642                 adapter->trigger_reset = true;
1643                 ++adapter->dev_stats.wd_expired;
1644         }
1645 }
1646
1647 /* Check if admin queue is enabled */
1648 static void check_for_admin_com_state(struct ena_adapter *adapter)
1649 {
1650         if (unlikely(!ena_com_get_admin_running_state(&adapter->ena_dev))) {
1651                 PMD_DRV_LOG(ERR, "ENA admin queue is not in running state\n");
1652                 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
1653                 adapter->trigger_reset = true;
1654         }
1655 }
1656
1657 static void ena_timer_wd_callback(__rte_unused struct rte_timer *timer,
1658                                   void *arg)
1659 {
1660         struct rte_eth_dev *dev = arg;
1661         struct ena_adapter *adapter = dev->data->dev_private;
1662
1663         check_for_missing_keep_alive(adapter);
1664         check_for_admin_com_state(adapter);
1665
1666         if (unlikely(adapter->trigger_reset)) {
1667                 PMD_DRV_LOG(ERR, "Trigger reset is on\n");
1668                 rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_INTR_RESET,
1669                         NULL);
1670         }
1671 }
1672
1673 static inline void
1674 set_default_llq_configurations(struct ena_llq_configurations *llq_config,
1675                                struct ena_admin_feature_llq_desc *llq,
1676                                bool use_large_llq_hdr)
1677 {
1678         llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
1679         llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
1680         llq_config->llq_num_decs_before_header =
1681                 ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
1682
1683         if (use_large_llq_hdr &&
1684             (llq->entry_size_ctrl_supported & ENA_ADMIN_LIST_ENTRY_SIZE_256B)) {
1685                 llq_config->llq_ring_entry_size =
1686                         ENA_ADMIN_LIST_ENTRY_SIZE_256B;
1687                 llq_config->llq_ring_entry_size_value = 256;
1688         } else {
1689                 llq_config->llq_ring_entry_size =
1690                         ENA_ADMIN_LIST_ENTRY_SIZE_128B;
1691                 llq_config->llq_ring_entry_size_value = 128;
1692         }
1693 }
1694
1695 static int
1696 ena_set_queues_placement_policy(struct ena_adapter *adapter,
1697                                 struct ena_com_dev *ena_dev,
1698                                 struct ena_admin_feature_llq_desc *llq,
1699                                 struct ena_llq_configurations *llq_default_configurations)
1700 {
1701         int rc;
1702         u32 llq_feature_mask;
1703
1704         llq_feature_mask = 1 << ENA_ADMIN_LLQ;
1705         if (!(ena_dev->supported_features & llq_feature_mask)) {
1706                 PMD_DRV_LOG(INFO,
1707                         "LLQ is not supported. Fallback to host mode policy.\n");
1708                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1709                 return 0;
1710         }
1711
1712         rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
1713         if (unlikely(rc)) {
1714                 PMD_INIT_LOG(WARNING,
1715                         "Failed to config dev mode. Fallback to host mode policy.\n");
1716                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1717                 return 0;
1718         }
1719
1720         /* Nothing to config, exit */
1721         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
1722                 return 0;
1723
1724         if (!adapter->dev_mem_base) {
1725                 PMD_DRV_LOG(ERR,
1726                         "Unable to access LLQ BAR resource. Fallback to host mode policy.\n");
1727                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1728                 return 0;
1729         }
1730
1731         ena_dev->mem_bar = adapter->dev_mem_base;
1732
1733         return 0;
1734 }
1735
1736 static uint32_t ena_calc_max_io_queue_num(struct ena_com_dev *ena_dev,
1737         struct ena_com_dev_get_features_ctx *get_feat_ctx)
1738 {
1739         uint32_t io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
1740
1741         /* Regular queues capabilities */
1742         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
1743                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
1744                         &get_feat_ctx->max_queue_ext.max_queue_ext;
1745                 io_rx_num = RTE_MIN(max_queue_ext->max_rx_sq_num,
1746                                     max_queue_ext->max_rx_cq_num);
1747                 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
1748                 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
1749         } else {
1750                 struct ena_admin_queue_feature_desc *max_queues =
1751                         &get_feat_ctx->max_queues;
1752                 io_tx_sq_num = max_queues->max_sq_num;
1753                 io_tx_cq_num = max_queues->max_cq_num;
1754                 io_rx_num = RTE_MIN(io_tx_sq_num, io_tx_cq_num);
1755         }
1756
1757         /* In case of LLQ use the llq number in the get feature cmd */
1758         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
1759                 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
1760
1761         max_num_io_queues = RTE_MIN(ENA_MAX_NUM_IO_QUEUES, io_rx_num);
1762         max_num_io_queues = RTE_MIN(max_num_io_queues, io_tx_sq_num);
1763         max_num_io_queues = RTE_MIN(max_num_io_queues, io_tx_cq_num);
1764
1765         if (unlikely(max_num_io_queues == 0)) {
1766                 PMD_DRV_LOG(ERR, "Number of IO queues cannot not be 0\n");
1767                 return -EFAULT;
1768         }
1769
1770         return max_num_io_queues;
1771 }
1772
1773 static int eth_ena_dev_init(struct rte_eth_dev *eth_dev)
1774 {
1775         struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
1776         struct rte_pci_device *pci_dev;
1777         struct rte_intr_handle *intr_handle;
1778         struct ena_adapter *adapter = eth_dev->data->dev_private;
1779         struct ena_com_dev *ena_dev = &adapter->ena_dev;
1780         struct ena_com_dev_get_features_ctx get_feat_ctx;
1781         struct ena_llq_configurations llq_config;
1782         const char *queue_type_str;
1783         uint32_t max_num_io_queues;
1784         int rc;
1785         static int adapters_found;
1786         bool disable_meta_caching;
1787         bool wd_state = false;
1788
1789         eth_dev->dev_ops = &ena_dev_ops;
1790         eth_dev->rx_pkt_burst = &eth_ena_recv_pkts;
1791         eth_dev->tx_pkt_burst = &eth_ena_xmit_pkts;
1792         eth_dev->tx_pkt_prepare = &eth_ena_prep_pkts;
1793
1794         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1795                 return 0;
1796
1797         eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1798
1799         memset(adapter, 0, sizeof(struct ena_adapter));
1800         ena_dev = &adapter->ena_dev;
1801
1802         adapter->edev_data = eth_dev->data;
1803
1804         pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1805
1806         PMD_INIT_LOG(INFO, "Initializing %x:%x:%x.%d\n",
1807                      pci_dev->addr.domain,
1808                      pci_dev->addr.bus,
1809                      pci_dev->addr.devid,
1810                      pci_dev->addr.function);
1811
1812         intr_handle = &pci_dev->intr_handle;
1813
1814         adapter->regs = pci_dev->mem_resource[ENA_REGS_BAR].addr;
1815         adapter->dev_mem_base = pci_dev->mem_resource[ENA_MEM_BAR].addr;
1816
1817         if (!adapter->regs) {
1818                 PMD_INIT_LOG(CRIT, "Failed to access registers BAR(%d)\n",
1819                              ENA_REGS_BAR);
1820                 return -ENXIO;
1821         }
1822
1823         ena_dev->reg_bar = adapter->regs;
1824         /* This is a dummy pointer for ena_com functions. */
1825         ena_dev->dmadev = adapter;
1826
1827         adapter->id_number = adapters_found;
1828
1829         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d",
1830                  adapter->id_number);
1831
1832         rc = ena_parse_devargs(adapter, pci_dev->device.devargs);
1833         if (rc != 0) {
1834                 PMD_INIT_LOG(CRIT, "Failed to parse devargs\n");
1835                 goto err;
1836         }
1837
1838         /* device specific initialization routine */
1839         rc = ena_device_init(ena_dev, pci_dev, &get_feat_ctx, &wd_state);
1840         if (rc) {
1841                 PMD_INIT_LOG(CRIT, "Failed to init ENA device\n");
1842                 goto err;
1843         }
1844         adapter->wd_state = wd_state;
1845
1846         set_default_llq_configurations(&llq_config, &get_feat_ctx.llq,
1847                 adapter->use_large_llq_hdr);
1848         rc = ena_set_queues_placement_policy(adapter, ena_dev,
1849                                              &get_feat_ctx.llq, &llq_config);
1850         if (unlikely(rc)) {
1851                 PMD_INIT_LOG(CRIT, "Failed to set placement policy\n");
1852                 return rc;
1853         }
1854
1855         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
1856                 queue_type_str = "Regular";
1857         else
1858                 queue_type_str = "Low latency";
1859         PMD_DRV_LOG(INFO, "Placement policy: %s\n", queue_type_str);
1860
1861         calc_queue_ctx.ena_dev = ena_dev;
1862         calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
1863
1864         max_num_io_queues = ena_calc_max_io_queue_num(ena_dev, &get_feat_ctx);
1865         rc = ena_calc_io_queue_size(&calc_queue_ctx,
1866                 adapter->use_large_llq_hdr);
1867         if (unlikely((rc != 0) || (max_num_io_queues == 0))) {
1868                 rc = -EFAULT;
1869                 goto err_device_destroy;
1870         }
1871
1872         adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
1873         adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
1874         adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
1875         adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
1876         adapter->max_num_io_queues = max_num_io_queues;
1877
1878         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1879                 disable_meta_caching =
1880                         !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
1881                         BIT(ENA_ADMIN_DISABLE_META_CACHING));
1882         } else {
1883                 disable_meta_caching = false;
1884         }
1885
1886         /* prepare ring structures */
1887         ena_init_rings(adapter, disable_meta_caching);
1888
1889         ena_config_debug_area(adapter);
1890
1891         /* Set max MTU for this device */
1892         adapter->max_mtu = get_feat_ctx.dev_attr.max_mtu;
1893
1894         /* set device support for offloads */
1895         adapter->offloads.tso4_supported = (get_feat_ctx.offload.tx &
1896                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK) != 0;
1897         adapter->offloads.tx_csum_supported = (get_feat_ctx.offload.tx &
1898                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK) != 0;
1899         adapter->offloads.rx_csum_supported =
1900                 (get_feat_ctx.offload.rx_supported &
1901                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK) != 0;
1902
1903         /* Copy MAC address and point DPDK to it */
1904         eth_dev->data->mac_addrs = (struct rte_ether_addr *)adapter->mac_addr;
1905         rte_ether_addr_copy((struct rte_ether_addr *)
1906                         get_feat_ctx.dev_attr.mac_addr,
1907                         (struct rte_ether_addr *)adapter->mac_addr);
1908
1909         adapter->drv_stats = rte_zmalloc("adapter stats",
1910                                          sizeof(*adapter->drv_stats),
1911                                          RTE_CACHE_LINE_SIZE);
1912         if (!adapter->drv_stats) {
1913                 PMD_DRV_LOG(ERR,
1914                         "Failed to allocate memory for adapter statistics\n");
1915                 rc = -ENOMEM;
1916                 goto err_delete_debug_area;
1917         }
1918
1919         rte_spinlock_init(&adapter->admin_lock);
1920
1921         rte_intr_callback_register(intr_handle,
1922                                    ena_interrupt_handler_rte,
1923                                    eth_dev);
1924         rte_intr_enable(intr_handle);
1925         ena_com_set_admin_polling_mode(ena_dev, false);
1926         ena_com_admin_aenq_enable(ena_dev);
1927
1928         if (adapters_found == 0)
1929                 rte_timer_subsystem_init();
1930         rte_timer_init(&adapter->timer_wd);
1931
1932         adapters_found++;
1933         adapter->state = ENA_ADAPTER_STATE_INIT;
1934
1935         return 0;
1936
1937 err_delete_debug_area:
1938         ena_com_delete_debug_area(ena_dev);
1939
1940 err_device_destroy:
1941         ena_com_delete_host_info(ena_dev);
1942         ena_com_admin_destroy(ena_dev);
1943
1944 err:
1945         return rc;
1946 }
1947
1948 static void ena_destroy_device(struct rte_eth_dev *eth_dev)
1949 {
1950         struct ena_adapter *adapter = eth_dev->data->dev_private;
1951         struct ena_com_dev *ena_dev = &adapter->ena_dev;
1952
1953         if (adapter->state == ENA_ADAPTER_STATE_FREE)
1954                 return;
1955
1956         ena_com_set_admin_running_state(ena_dev, false);
1957
1958         if (adapter->state != ENA_ADAPTER_STATE_CLOSED)
1959                 ena_close(eth_dev);
1960
1961         ena_com_delete_debug_area(ena_dev);
1962         ena_com_delete_host_info(ena_dev);
1963
1964         ena_com_abort_admin_commands(ena_dev);
1965         ena_com_wait_for_abort_completion(ena_dev);
1966         ena_com_admin_destroy(ena_dev);
1967         ena_com_mmio_reg_read_request_destroy(ena_dev);
1968
1969         adapter->state = ENA_ADAPTER_STATE_FREE;
1970 }
1971
1972 static int eth_ena_dev_uninit(struct rte_eth_dev *eth_dev)
1973 {
1974         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1975                 return 0;
1976
1977         ena_destroy_device(eth_dev);
1978
1979         return 0;
1980 }
1981
1982 static int ena_dev_configure(struct rte_eth_dev *dev)
1983 {
1984         struct ena_adapter *adapter = dev->data->dev_private;
1985
1986         adapter->state = ENA_ADAPTER_STATE_CONFIG;
1987
1988         if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
1989                 dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
1990
1991         adapter->tx_selected_offloads = dev->data->dev_conf.txmode.offloads;
1992         adapter->rx_selected_offloads = dev->data->dev_conf.rxmode.offloads;
1993         return 0;
1994 }
1995
1996 static void ena_init_rings(struct ena_adapter *adapter,
1997                            bool disable_meta_caching)
1998 {
1999         size_t i;
2000
2001         for (i = 0; i < adapter->max_num_io_queues; i++) {
2002                 struct ena_ring *ring = &adapter->tx_ring[i];
2003
2004                 ring->configured = 0;
2005                 ring->type = ENA_RING_TYPE_TX;
2006                 ring->adapter = adapter;
2007                 ring->id = i;
2008                 ring->tx_mem_queue_type = adapter->ena_dev.tx_mem_queue_type;
2009                 ring->tx_max_header_size = adapter->ena_dev.tx_max_header_size;
2010                 ring->sgl_size = adapter->max_tx_sgl_size;
2011                 ring->disable_meta_caching = disable_meta_caching;
2012         }
2013
2014         for (i = 0; i < adapter->max_num_io_queues; i++) {
2015                 struct ena_ring *ring = &adapter->rx_ring[i];
2016
2017                 ring->configured = 0;
2018                 ring->type = ENA_RING_TYPE_RX;
2019                 ring->adapter = adapter;
2020                 ring->id = i;
2021                 ring->sgl_size = adapter->max_rx_sgl_size;
2022         }
2023 }
2024
2025 static int ena_infos_get(struct rte_eth_dev *dev,
2026                           struct rte_eth_dev_info *dev_info)
2027 {
2028         struct ena_adapter *adapter;
2029         struct ena_com_dev *ena_dev;
2030         uint64_t rx_feat = 0, tx_feat = 0;
2031
2032         ena_assert_msg(dev->data != NULL, "Uninitialized device\n");
2033         ena_assert_msg(dev->data->dev_private != NULL, "Uninitialized device\n");
2034         adapter = dev->data->dev_private;
2035
2036         ena_dev = &adapter->ena_dev;
2037         ena_assert_msg(ena_dev != NULL, "Uninitialized device\n");
2038
2039         dev_info->speed_capa =
2040                         ETH_LINK_SPEED_1G   |
2041                         ETH_LINK_SPEED_2_5G |
2042                         ETH_LINK_SPEED_5G   |
2043                         ETH_LINK_SPEED_10G  |
2044                         ETH_LINK_SPEED_25G  |
2045                         ETH_LINK_SPEED_40G  |
2046                         ETH_LINK_SPEED_50G  |
2047                         ETH_LINK_SPEED_100G;
2048
2049         /* Set Tx & Rx features available for device */
2050         if (adapter->offloads.tso4_supported)
2051                 tx_feat |= DEV_TX_OFFLOAD_TCP_TSO;
2052
2053         if (adapter->offloads.tx_csum_supported)
2054                 tx_feat |= DEV_TX_OFFLOAD_IPV4_CKSUM |
2055                         DEV_TX_OFFLOAD_UDP_CKSUM |
2056                         DEV_TX_OFFLOAD_TCP_CKSUM;
2057
2058         if (adapter->offloads.rx_csum_supported)
2059                 rx_feat |= DEV_RX_OFFLOAD_IPV4_CKSUM |
2060                         DEV_RX_OFFLOAD_UDP_CKSUM  |
2061                         DEV_RX_OFFLOAD_TCP_CKSUM;
2062
2063         rx_feat |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2064
2065         /* Inform framework about available features */
2066         dev_info->rx_offload_capa = rx_feat;
2067         dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_RSS_HASH;
2068         dev_info->rx_queue_offload_capa = rx_feat;
2069         dev_info->tx_offload_capa = tx_feat;
2070         dev_info->tx_queue_offload_capa = tx_feat;
2071
2072         dev_info->flow_type_rss_offloads = ETH_RSS_IP | ETH_RSS_TCP |
2073                                            ETH_RSS_UDP;
2074
2075         dev_info->min_rx_bufsize = ENA_MIN_FRAME_LEN;
2076         dev_info->max_rx_pktlen  = adapter->max_mtu;
2077         dev_info->max_mac_addrs = 1;
2078
2079         dev_info->max_rx_queues = adapter->max_num_io_queues;
2080         dev_info->max_tx_queues = adapter->max_num_io_queues;
2081         dev_info->reta_size = ENA_RX_RSS_TABLE_SIZE;
2082
2083         adapter->tx_supported_offloads = tx_feat;
2084         adapter->rx_supported_offloads = rx_feat;
2085
2086         dev_info->rx_desc_lim.nb_max = adapter->max_rx_ring_size;
2087         dev_info->rx_desc_lim.nb_min = ENA_MIN_RING_DESC;
2088         dev_info->rx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2089                                         adapter->max_rx_sgl_size);
2090         dev_info->rx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2091                                         adapter->max_rx_sgl_size);
2092
2093         dev_info->tx_desc_lim.nb_max = adapter->max_tx_ring_size;
2094         dev_info->tx_desc_lim.nb_min = ENA_MIN_RING_DESC;
2095         dev_info->tx_desc_lim.nb_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2096                                         adapter->max_tx_sgl_size);
2097         dev_info->tx_desc_lim.nb_mtu_seg_max = RTE_MIN(ENA_PKT_MAX_BUFS,
2098                                         adapter->max_tx_sgl_size);
2099
2100         dev_info->default_rxportconf.ring_size = ENA_DEFAULT_RING_SIZE;
2101         dev_info->default_txportconf.ring_size = ENA_DEFAULT_RING_SIZE;
2102
2103         return 0;
2104 }
2105
2106 static inline void ena_init_rx_mbuf(struct rte_mbuf *mbuf, uint16_t len)
2107 {
2108         mbuf->data_len = len;
2109         mbuf->data_off = RTE_PKTMBUF_HEADROOM;
2110         mbuf->refcnt = 1;
2111         mbuf->next = NULL;
2112 }
2113
2114 static struct rte_mbuf *ena_rx_mbuf(struct ena_ring *rx_ring,
2115                                     struct ena_com_rx_buf_info *ena_bufs,
2116                                     uint32_t descs,
2117                                     uint16_t *next_to_clean,
2118                                     uint8_t offset)
2119 {
2120         struct rte_mbuf *mbuf;
2121         struct rte_mbuf *mbuf_head;
2122         struct ena_rx_buffer *rx_info;
2123         int rc;
2124         uint16_t ntc, len, req_id, buf = 0;
2125
2126         if (unlikely(descs == 0))
2127                 return NULL;
2128
2129         ntc = *next_to_clean;
2130
2131         len = ena_bufs[buf].len;
2132         req_id = ena_bufs[buf].req_id;
2133
2134         rx_info = &rx_ring->rx_buffer_info[req_id];
2135
2136         mbuf = rx_info->mbuf;
2137         RTE_ASSERT(mbuf != NULL);
2138
2139         ena_init_rx_mbuf(mbuf, len);
2140
2141         /* Fill the mbuf head with the data specific for 1st segment. */
2142         mbuf_head = mbuf;
2143         mbuf_head->nb_segs = descs;
2144         mbuf_head->port = rx_ring->port_id;
2145         mbuf_head->pkt_len = len;
2146         mbuf_head->data_off += offset;
2147
2148         rx_info->mbuf = NULL;
2149         rx_ring->empty_rx_reqs[ntc] = req_id;
2150         ntc = ENA_IDX_NEXT_MASKED(ntc, rx_ring->size_mask);
2151
2152         while (--descs) {
2153                 ++buf;
2154                 len = ena_bufs[buf].len;
2155                 req_id = ena_bufs[buf].req_id;
2156
2157                 rx_info = &rx_ring->rx_buffer_info[req_id];
2158                 RTE_ASSERT(rx_info->mbuf != NULL);
2159
2160                 if (unlikely(len == 0)) {
2161                         /*
2162                          * Some devices can pass descriptor with the length 0.
2163                          * To avoid confusion, the PMD is simply putting the
2164                          * descriptor back, as it was never used. We'll avoid
2165                          * mbuf allocation that way.
2166                          */
2167                         rc = ena_add_single_rx_desc(rx_ring->ena_com_io_sq,
2168                                 rx_info->mbuf, req_id);
2169                         if (unlikely(rc != 0)) {
2170                                 /* Free the mbuf in case of an error. */
2171                                 rte_mbuf_raw_free(rx_info->mbuf);
2172                         } else {
2173                                 /*
2174                                  * If there was no error, just exit the loop as
2175                                  * 0 length descriptor is always the last one.
2176                                  */
2177                                 break;
2178                         }
2179                 } else {
2180                         /* Create an mbuf chain. */
2181                         mbuf->next = rx_info->mbuf;
2182                         mbuf = mbuf->next;
2183
2184                         ena_init_rx_mbuf(mbuf, len);
2185                         mbuf_head->pkt_len += len;
2186                 }
2187
2188                 /*
2189                  * Mark the descriptor as depleted and perform necessary
2190                  * cleanup.
2191                  * This code will execute in two cases:
2192                  *  1. Descriptor len was greater than 0 - normal situation.
2193                  *  2. Descriptor len was 0 and we failed to add the descriptor
2194                  *     to the device. In that situation, we should try to add
2195                  *     the mbuf again in the populate routine and mark the
2196                  *     descriptor as used up by the device.
2197                  */
2198                 rx_info->mbuf = NULL;
2199                 rx_ring->empty_rx_reqs[ntc] = req_id;
2200                 ntc = ENA_IDX_NEXT_MASKED(ntc, rx_ring->size_mask);
2201         }
2202
2203         *next_to_clean = ntc;
2204
2205         return mbuf_head;
2206 }
2207
2208 static uint16_t eth_ena_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
2209                                   uint16_t nb_pkts)
2210 {
2211         struct ena_ring *rx_ring = (struct ena_ring *)(rx_queue);
2212         unsigned int free_queue_entries;
2213         unsigned int refill_threshold;
2214         uint16_t next_to_clean = rx_ring->next_to_clean;
2215         uint16_t descs_in_use;
2216         struct rte_mbuf *mbuf;
2217         uint16_t completed;
2218         struct ena_com_rx_ctx ena_rx_ctx;
2219         int i, rc = 0;
2220
2221         /* Check adapter state */
2222         if (unlikely(rx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2223                 PMD_DRV_LOG(ALERT,
2224                         "Trying to receive pkts while device is NOT running\n");
2225                 return 0;
2226         }
2227
2228         descs_in_use = rx_ring->ring_size -
2229                 ena_com_free_q_entries(rx_ring->ena_com_io_sq) - 1;
2230         nb_pkts = RTE_MIN(descs_in_use, nb_pkts);
2231
2232         for (completed = 0; completed < nb_pkts; completed++) {
2233                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
2234                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
2235                 ena_rx_ctx.descs = 0;
2236                 ena_rx_ctx.pkt_offset = 0;
2237                 /* receive packet context */
2238                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
2239                                     rx_ring->ena_com_io_sq,
2240                                     &ena_rx_ctx);
2241                 if (unlikely(rc)) {
2242                         PMD_DRV_LOG(ERR,
2243                                 "Failed to get the packet from the device, rc: %d\n",
2244                                 rc);
2245                         if (rc == ENA_COM_NO_SPACE) {
2246                                 ++rx_ring->rx_stats.bad_desc_num;
2247                                 rx_ring->adapter->reset_reason =
2248                                         ENA_REGS_RESET_TOO_MANY_RX_DESCS;
2249                         } else {
2250                                 ++rx_ring->rx_stats.bad_req_id;
2251                                 rx_ring->adapter->reset_reason =
2252                                         ENA_REGS_RESET_INV_RX_REQ_ID;
2253                         }
2254                         rx_ring->adapter->trigger_reset = true;
2255                         return 0;
2256                 }
2257
2258                 mbuf = ena_rx_mbuf(rx_ring,
2259                         ena_rx_ctx.ena_bufs,
2260                         ena_rx_ctx.descs,
2261                         &next_to_clean,
2262                         ena_rx_ctx.pkt_offset);
2263                 if (unlikely(mbuf == NULL)) {
2264                         for (i = 0; i < ena_rx_ctx.descs; ++i) {
2265                                 rx_ring->empty_rx_reqs[next_to_clean] =
2266                                         rx_ring->ena_bufs[i].req_id;
2267                                 next_to_clean = ENA_IDX_NEXT_MASKED(
2268                                         next_to_clean, rx_ring->size_mask);
2269                         }
2270                         break;
2271                 }
2272
2273                 /* fill mbuf attributes if any */
2274                 ena_rx_mbuf_prepare(mbuf, &ena_rx_ctx);
2275
2276                 if (unlikely(mbuf->ol_flags &
2277                                 (PKT_RX_IP_CKSUM_BAD | PKT_RX_L4_CKSUM_BAD))) {
2278                         rte_atomic64_inc(&rx_ring->adapter->drv_stats->ierrors);
2279                         ++rx_ring->rx_stats.bad_csum;
2280                 }
2281
2282                 rx_pkts[completed] = mbuf;
2283                 rx_ring->rx_stats.bytes += mbuf->pkt_len;
2284         }
2285
2286         rx_ring->rx_stats.cnt += completed;
2287         rx_ring->next_to_clean = next_to_clean;
2288
2289         free_queue_entries = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
2290         refill_threshold =
2291                 RTE_MIN(rx_ring->ring_size / ENA_REFILL_THRESH_DIVIDER,
2292                 (unsigned int)ENA_REFILL_THRESH_PACKET);
2293
2294         /* Burst refill to save doorbells, memory barriers, const interval */
2295         if (free_queue_entries > refill_threshold) {
2296                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
2297                 ena_populate_rx_queue(rx_ring, free_queue_entries);
2298         }
2299
2300         return completed;
2301 }
2302
2303 static uint16_t
2304 eth_ena_prep_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2305                 uint16_t nb_pkts)
2306 {
2307         int32_t ret;
2308         uint32_t i;
2309         struct rte_mbuf *m;
2310         struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2311         struct rte_ipv4_hdr *ip_hdr;
2312         uint64_t ol_flags;
2313         uint16_t frag_field;
2314
2315         for (i = 0; i != nb_pkts; i++) {
2316                 m = tx_pkts[i];
2317                 ol_flags = m->ol_flags;
2318
2319                 if (!(ol_flags & PKT_TX_IPV4))
2320                         continue;
2321
2322                 /* If there was not L2 header length specified, assume it is
2323                  * length of the ethernet header.
2324                  */
2325                 if (unlikely(m->l2_len == 0))
2326                         m->l2_len = sizeof(struct rte_ether_hdr);
2327
2328                 ip_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
2329                                                  m->l2_len);
2330                 frag_field = rte_be_to_cpu_16(ip_hdr->fragment_offset);
2331
2332                 if ((frag_field & RTE_IPV4_HDR_DF_FLAG) != 0) {
2333                         m->packet_type |= RTE_PTYPE_L4_NONFRAG;
2334
2335                         /* If IPv4 header has DF flag enabled and TSO support is
2336                          * disabled, partial chcecksum should not be calculated.
2337                          */
2338                         if (!tx_ring->adapter->offloads.tso4_supported)
2339                                 continue;
2340                 }
2341
2342                 if ((ol_flags & ENA_TX_OFFLOAD_NOTSUP_MASK) != 0 ||
2343                                 (ol_flags & PKT_TX_L4_MASK) ==
2344                                 PKT_TX_SCTP_CKSUM) {
2345                         rte_errno = ENOTSUP;
2346                         return i;
2347                 }
2348
2349 #ifdef RTE_LIBRTE_ETHDEV_DEBUG
2350                 ret = rte_validate_tx_offload(m);
2351                 if (ret != 0) {
2352                         rte_errno = -ret;
2353                         return i;
2354                 }
2355 #endif
2356
2357                 /* In case we are supposed to TSO and have DF not set (DF=0)
2358                  * hardware must be provided with partial checksum, otherwise
2359                  * it will take care of necessary calculations.
2360                  */
2361
2362                 ret = rte_net_intel_cksum_flags_prepare(m,
2363                         ol_flags & ~PKT_TX_TCP_SEG);
2364                 if (ret != 0) {
2365                         rte_errno = -ret;
2366                         return i;
2367                 }
2368         }
2369
2370         return i;
2371 }
2372
2373 static void ena_update_hints(struct ena_adapter *adapter,
2374                              struct ena_admin_ena_hw_hints *hints)
2375 {
2376         if (hints->admin_completion_tx_timeout)
2377                 adapter->ena_dev.admin_queue.completion_timeout =
2378                         hints->admin_completion_tx_timeout * 1000;
2379
2380         if (hints->mmio_read_timeout)
2381                 /* convert to usec */
2382                 adapter->ena_dev.mmio_read.reg_read_to =
2383                         hints->mmio_read_timeout * 1000;
2384
2385         if (hints->driver_watchdog_timeout) {
2386                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2387                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2388                 else
2389                         // Convert msecs to ticks
2390                         adapter->keep_alive_timeout =
2391                                 (hints->driver_watchdog_timeout *
2392                                 rte_get_timer_hz()) / 1000;
2393         }
2394 }
2395
2396 static int ena_check_space_and_linearize_mbuf(struct ena_ring *tx_ring,
2397                                               struct rte_mbuf *mbuf)
2398 {
2399         struct ena_com_dev *ena_dev;
2400         int num_segments, header_len, rc;
2401
2402         ena_dev = &tx_ring->adapter->ena_dev;
2403         num_segments = mbuf->nb_segs;
2404         header_len = mbuf->data_len;
2405
2406         if (likely(num_segments < tx_ring->sgl_size))
2407                 goto checkspace;
2408
2409         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV &&
2410             (num_segments == tx_ring->sgl_size) &&
2411             (header_len < tx_ring->tx_max_header_size))
2412                 goto checkspace;
2413
2414         /* Checking for space for 2 additional metadata descriptors due to
2415          * possible header split and metadata descriptor. Linearization will
2416          * be needed so we reduce the segments number from num_segments to 1
2417          */
2418         if (!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq, 3)) {
2419                 PMD_DRV_LOG(DEBUG, "Not enough space in the Tx queue\n");
2420                 return ENA_COM_NO_MEM;
2421         }
2422         ++tx_ring->tx_stats.linearize;
2423         rc = rte_pktmbuf_linearize(mbuf);
2424         if (unlikely(rc)) {
2425                 PMD_DRV_LOG(WARNING, "Mbuf linearize failed\n");
2426                 rte_atomic64_inc(&tx_ring->adapter->drv_stats->ierrors);
2427                 ++tx_ring->tx_stats.linearize_failed;
2428                 return rc;
2429         }
2430
2431         return 0;
2432
2433 checkspace:
2434         /* Checking for space for 2 additional metadata descriptors due to
2435          * possible header split and metadata descriptor
2436          */
2437         if (!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
2438                                           num_segments + 2)) {
2439                 PMD_DRV_LOG(DEBUG, "Not enough space in the Tx queue\n");
2440                 return ENA_COM_NO_MEM;
2441         }
2442
2443         return 0;
2444 }
2445
2446 static void ena_tx_map_mbuf(struct ena_ring *tx_ring,
2447         struct ena_tx_buffer *tx_info,
2448         struct rte_mbuf *mbuf,
2449         void **push_header,
2450         uint16_t *header_len)
2451 {
2452         struct ena_com_buf *ena_buf;
2453         uint16_t delta, seg_len, push_len;
2454
2455         delta = 0;
2456         seg_len = mbuf->data_len;
2457
2458         tx_info->mbuf = mbuf;
2459         ena_buf = tx_info->bufs;
2460
2461         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2462                 /*
2463                  * Tx header might be (and will be in most cases) smaller than
2464                  * tx_max_header_size. But it's not an issue to send more data
2465                  * to the device, than actually needed if the mbuf size is
2466                  * greater than tx_max_header_size.
2467                  */
2468                 push_len = RTE_MIN(mbuf->pkt_len, tx_ring->tx_max_header_size);
2469                 *header_len = push_len;
2470
2471                 if (likely(push_len <= seg_len)) {
2472                         /* If the push header is in the single segment, then
2473                          * just point it to the 1st mbuf data.
2474                          */
2475                         *push_header = rte_pktmbuf_mtod(mbuf, uint8_t *);
2476                 } else {
2477                         /* If the push header lays in the several segments, copy
2478                          * it to the intermediate buffer.
2479                          */
2480                         rte_pktmbuf_read(mbuf, 0, push_len,
2481                                 tx_ring->push_buf_intermediate_buf);
2482                         *push_header = tx_ring->push_buf_intermediate_buf;
2483                         delta = push_len - seg_len;
2484                 }
2485         } else {
2486                 *push_header = NULL;
2487                 *header_len = 0;
2488                 push_len = 0;
2489         }
2490
2491         /* Process first segment taking into consideration pushed header */
2492         if (seg_len > push_len) {
2493                 ena_buf->paddr = mbuf->buf_iova +
2494                                 mbuf->data_off +
2495                                 push_len;
2496                 ena_buf->len = seg_len - push_len;
2497                 ena_buf++;
2498                 tx_info->num_of_bufs++;
2499         }
2500
2501         while ((mbuf = mbuf->next) != NULL) {
2502                 seg_len = mbuf->data_len;
2503
2504                 /* Skip mbufs if whole data is pushed as a header */
2505                 if (unlikely(delta > seg_len)) {
2506                         delta -= seg_len;
2507                         continue;
2508                 }
2509
2510                 ena_buf->paddr = mbuf->buf_iova + mbuf->data_off + delta;
2511                 ena_buf->len = seg_len - delta;
2512                 ena_buf++;
2513                 tx_info->num_of_bufs++;
2514
2515                 delta = 0;
2516         }
2517 }
2518
2519 static int ena_xmit_mbuf(struct ena_ring *tx_ring, struct rte_mbuf *mbuf)
2520 {
2521         struct ena_tx_buffer *tx_info;
2522         struct ena_com_tx_ctx ena_tx_ctx = { { 0 } };
2523         uint16_t next_to_use;
2524         uint16_t header_len;
2525         uint16_t req_id;
2526         void *push_header;
2527         int nb_hw_desc;
2528         int rc;
2529
2530         rc = ena_check_space_and_linearize_mbuf(tx_ring, mbuf);
2531         if (unlikely(rc))
2532                 return rc;
2533
2534         next_to_use = tx_ring->next_to_use;
2535
2536         req_id = tx_ring->empty_tx_reqs[next_to_use];
2537         tx_info = &tx_ring->tx_buffer_info[req_id];
2538         tx_info->num_of_bufs = 0;
2539
2540         ena_tx_map_mbuf(tx_ring, tx_info, mbuf, &push_header, &header_len);
2541
2542         ena_tx_ctx.ena_bufs = tx_info->bufs;
2543         ena_tx_ctx.push_header = push_header;
2544         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2545         ena_tx_ctx.req_id = req_id;
2546         ena_tx_ctx.header_len = header_len;
2547
2548         /* Set Tx offloads flags, if applicable */
2549         ena_tx_mbuf_prepare(mbuf, &ena_tx_ctx, tx_ring->offloads,
2550                 tx_ring->disable_meta_caching);
2551
2552         if (unlikely(ena_com_is_doorbell_needed(tx_ring->ena_com_io_sq,
2553                         &ena_tx_ctx))) {
2554                 PMD_DRV_LOG(DEBUG,
2555                         "LLQ Tx max burst size of queue %d achieved, writing doorbell to send burst\n",
2556                         tx_ring->id);
2557                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2558                 tx_ring->tx_stats.doorbells++;
2559                 tx_ring->pkts_without_db = false;
2560         }
2561
2562         /* prepare the packet's descriptors to dma engine */
2563         rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq, &ena_tx_ctx,
2564                 &nb_hw_desc);
2565         if (unlikely(rc)) {
2566                 ++tx_ring->tx_stats.prepare_ctx_err;
2567                 return rc;
2568         }
2569
2570         tx_info->tx_descs = nb_hw_desc;
2571
2572         tx_ring->tx_stats.cnt++;
2573         tx_ring->tx_stats.bytes += mbuf->pkt_len;
2574
2575         tx_ring->next_to_use = ENA_IDX_NEXT_MASKED(next_to_use,
2576                 tx_ring->size_mask);
2577
2578         return 0;
2579 }
2580
2581 static void ena_tx_cleanup(struct ena_ring *tx_ring)
2582 {
2583         unsigned int cleanup_budget;
2584         unsigned int total_tx_descs = 0;
2585         uint16_t next_to_clean = tx_ring->next_to_clean;
2586
2587         cleanup_budget = RTE_MIN(tx_ring->ring_size / ENA_REFILL_THRESH_DIVIDER,
2588                 (unsigned int)ENA_REFILL_THRESH_PACKET);
2589
2590         while (likely(total_tx_descs < cleanup_budget)) {
2591                 struct rte_mbuf *mbuf;
2592                 struct ena_tx_buffer *tx_info;
2593                 uint16_t req_id;
2594
2595                 if (ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq, &req_id) != 0)
2596                         break;
2597
2598                 if (unlikely(validate_tx_req_id(tx_ring, req_id) != 0))
2599                         break;
2600
2601                 /* Get Tx info & store how many descs were processed  */
2602                 tx_info = &tx_ring->tx_buffer_info[req_id];
2603
2604                 mbuf = tx_info->mbuf;
2605                 rte_pktmbuf_free(mbuf);
2606
2607                 tx_info->mbuf = NULL;
2608                 tx_ring->empty_tx_reqs[next_to_clean] = req_id;
2609
2610                 total_tx_descs += tx_info->tx_descs;
2611
2612                 /* Put back descriptor to the ring for reuse */
2613                 next_to_clean = ENA_IDX_NEXT_MASKED(next_to_clean,
2614                         tx_ring->size_mask);
2615         }
2616
2617         if (likely(total_tx_descs > 0)) {
2618                 /* acknowledge completion of sent packets */
2619                 tx_ring->next_to_clean = next_to_clean;
2620                 ena_com_comp_ack(tx_ring->ena_com_io_sq, total_tx_descs);
2621                 ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
2622         }
2623 }
2624
2625 static uint16_t eth_ena_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts,
2626                                   uint16_t nb_pkts)
2627 {
2628         struct ena_ring *tx_ring = (struct ena_ring *)(tx_queue);
2629         uint16_t sent_idx = 0;
2630
2631         /* Check adapter state */
2632         if (unlikely(tx_ring->adapter->state != ENA_ADAPTER_STATE_RUNNING)) {
2633                 PMD_DRV_LOG(ALERT,
2634                         "Trying to xmit pkts while device is NOT running\n");
2635                 return 0;
2636         }
2637
2638         for (sent_idx = 0; sent_idx < nb_pkts; sent_idx++) {
2639                 if (ena_xmit_mbuf(tx_ring, tx_pkts[sent_idx]))
2640                         break;
2641                 tx_ring->pkts_without_db = true;
2642                 rte_prefetch0(tx_pkts[ENA_IDX_ADD_MASKED(sent_idx, 4,
2643                         tx_ring->size_mask)]);
2644         }
2645
2646         tx_ring->tx_stats.available_desc =
2647                 ena_com_free_q_entries(tx_ring->ena_com_io_sq);
2648
2649         /* If there are ready packets to be xmitted... */
2650         if (likely(tx_ring->pkts_without_db)) {
2651                 /* ...let HW do its best :-) */
2652                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2653                 tx_ring->tx_stats.doorbells++;
2654                 tx_ring->pkts_without_db = false;
2655         }
2656
2657         ena_tx_cleanup(tx_ring);
2658
2659         tx_ring->tx_stats.available_desc =
2660                 ena_com_free_q_entries(tx_ring->ena_com_io_sq);
2661         tx_ring->tx_stats.tx_poll++;
2662
2663         return sent_idx;
2664 }
2665
2666 int ena_copy_eni_stats(struct ena_adapter *adapter)
2667 {
2668         struct ena_admin_eni_stats admin_eni_stats;
2669         int rc;
2670
2671         rte_spinlock_lock(&adapter->admin_lock);
2672         rc = ena_com_get_eni_stats(&adapter->ena_dev, &admin_eni_stats);
2673         rte_spinlock_unlock(&adapter->admin_lock);
2674         if (rc != 0) {
2675                 if (rc == ENA_COM_UNSUPPORTED) {
2676                         PMD_DRV_LOG(DEBUG,
2677                                 "Retrieving ENI metrics is not supported\n");
2678                 } else {
2679                         PMD_DRV_LOG(WARNING,
2680                                 "Failed to get ENI metrics, rc: %d\n", rc);
2681                 }
2682                 return rc;
2683         }
2684
2685         rte_memcpy(&adapter->eni_stats, &admin_eni_stats,
2686                 sizeof(struct ena_stats_eni));
2687
2688         return 0;
2689 }
2690
2691 /**
2692  * DPDK callback to retrieve names of extended device statistics
2693  *
2694  * @param dev
2695  *   Pointer to Ethernet device structure.
2696  * @param[out] xstats_names
2697  *   Buffer to insert names into.
2698  * @param n
2699  *   Number of names.
2700  *
2701  * @return
2702  *   Number of xstats names.
2703  */
2704 static int ena_xstats_get_names(struct rte_eth_dev *dev,
2705                                 struct rte_eth_xstat_name *xstats_names,
2706                                 unsigned int n)
2707 {
2708         unsigned int xstats_count = ena_xstats_calc_num(dev->data);
2709         unsigned int stat, i, count = 0;
2710
2711         if (n < xstats_count || !xstats_names)
2712                 return xstats_count;
2713
2714         for (stat = 0; stat < ENA_STATS_ARRAY_GLOBAL; stat++, count++)
2715                 strcpy(xstats_names[count].name,
2716                         ena_stats_global_strings[stat].name);
2717
2718         for (stat = 0; stat < ENA_STATS_ARRAY_ENI; stat++, count++)
2719                 strcpy(xstats_names[count].name,
2720                         ena_stats_eni_strings[stat].name);
2721
2722         for (stat = 0; stat < ENA_STATS_ARRAY_RX; stat++)
2723                 for (i = 0; i < dev->data->nb_rx_queues; i++, count++)
2724                         snprintf(xstats_names[count].name,
2725                                 sizeof(xstats_names[count].name),
2726                                 "rx_q%d_%s", i,
2727                                 ena_stats_rx_strings[stat].name);
2728
2729         for (stat = 0; stat < ENA_STATS_ARRAY_TX; stat++)
2730                 for (i = 0; i < dev->data->nb_tx_queues; i++, count++)
2731                         snprintf(xstats_names[count].name,
2732                                 sizeof(xstats_names[count].name),
2733                                 "tx_q%d_%s", i,
2734                                 ena_stats_tx_strings[stat].name);
2735
2736         return xstats_count;
2737 }
2738
2739 /**
2740  * DPDK callback to get extended device statistics.
2741  *
2742  * @param dev
2743  *   Pointer to Ethernet device structure.
2744  * @param[out] stats
2745  *   Stats table output buffer.
2746  * @param n
2747  *   The size of the stats table.
2748  *
2749  * @return
2750  *   Number of xstats on success, negative on failure.
2751  */
2752 static int ena_xstats_get(struct rte_eth_dev *dev,
2753                           struct rte_eth_xstat *xstats,
2754                           unsigned int n)
2755 {
2756         struct ena_adapter *adapter = dev->data->dev_private;
2757         unsigned int xstats_count = ena_xstats_calc_num(dev->data);
2758         unsigned int stat, i, count = 0;
2759         int stat_offset;
2760         void *stats_begin;
2761
2762         if (n < xstats_count)
2763                 return xstats_count;
2764
2765         if (!xstats)
2766                 return 0;
2767
2768         for (stat = 0; stat < ENA_STATS_ARRAY_GLOBAL; stat++, count++) {
2769                 stat_offset = ena_stats_global_strings[stat].stat_offset;
2770                 stats_begin = &adapter->dev_stats;
2771
2772                 xstats[count].id = count;
2773                 xstats[count].value = *((uint64_t *)
2774                         ((char *)stats_begin + stat_offset));
2775         }
2776
2777         /* Even if the function below fails, we should copy previous (or initial
2778          * values) to keep structure of rte_eth_xstat consistent.
2779          */
2780         ena_copy_eni_stats(adapter);
2781         for (stat = 0; stat < ENA_STATS_ARRAY_ENI; stat++, count++) {
2782                 stat_offset = ena_stats_eni_strings[stat].stat_offset;
2783                 stats_begin = &adapter->eni_stats;
2784
2785                 xstats[count].id = count;
2786                 xstats[count].value = *((uint64_t *)
2787                     ((char *)stats_begin + stat_offset));
2788         }
2789
2790         for (stat = 0; stat < ENA_STATS_ARRAY_RX; stat++) {
2791                 for (i = 0; i < dev->data->nb_rx_queues; i++, count++) {
2792                         stat_offset = ena_stats_rx_strings[stat].stat_offset;
2793                         stats_begin = &adapter->rx_ring[i].rx_stats;
2794
2795                         xstats[count].id = count;
2796                         xstats[count].value = *((uint64_t *)
2797                                 ((char *)stats_begin + stat_offset));
2798                 }
2799         }
2800
2801         for (stat = 0; stat < ENA_STATS_ARRAY_TX; stat++) {
2802                 for (i = 0; i < dev->data->nb_tx_queues; i++, count++) {
2803                         stat_offset = ena_stats_tx_strings[stat].stat_offset;
2804                         stats_begin = &adapter->tx_ring[i].rx_stats;
2805
2806                         xstats[count].id = count;
2807                         xstats[count].value = *((uint64_t *)
2808                                 ((char *)stats_begin + stat_offset));
2809                 }
2810         }
2811
2812         return count;
2813 }
2814
2815 static int ena_xstats_get_by_id(struct rte_eth_dev *dev,
2816                                 const uint64_t *ids,
2817                                 uint64_t *values,
2818                                 unsigned int n)
2819 {
2820         struct ena_adapter *adapter = dev->data->dev_private;
2821         uint64_t id;
2822         uint64_t rx_entries, tx_entries;
2823         unsigned int i;
2824         int qid;
2825         int valid = 0;
2826         bool was_eni_copied = false;
2827
2828         for (i = 0; i < n; ++i) {
2829                 id = ids[i];
2830                 /* Check if id belongs to global statistics */
2831                 if (id < ENA_STATS_ARRAY_GLOBAL) {
2832                         values[i] = *((uint64_t *)&adapter->dev_stats + id);
2833                         ++valid;
2834                         continue;
2835                 }
2836
2837                 /* Check if id belongs to ENI statistics */
2838                 id -= ENA_STATS_ARRAY_GLOBAL;
2839                 if (id < ENA_STATS_ARRAY_ENI) {
2840                         /* Avoid reading ENI stats multiple times in a single
2841                          * function call, as it requires communication with the
2842                          * admin queue.
2843                          */
2844                         if (!was_eni_copied) {
2845                                 was_eni_copied = true;
2846                                 ena_copy_eni_stats(adapter);
2847                         }
2848                         values[i] = *((uint64_t *)&adapter->eni_stats + id);
2849                         ++valid;
2850                         continue;
2851                 }
2852
2853                 /* Check if id belongs to rx queue statistics */
2854                 id -= ENA_STATS_ARRAY_ENI;
2855                 rx_entries = ENA_STATS_ARRAY_RX * dev->data->nb_rx_queues;
2856                 if (id < rx_entries) {
2857                         qid = id % dev->data->nb_rx_queues;
2858                         id /= dev->data->nb_rx_queues;
2859                         values[i] = *((uint64_t *)
2860                                 &adapter->rx_ring[qid].rx_stats + id);
2861                         ++valid;
2862                         continue;
2863                 }
2864                                 /* Check if id belongs to rx queue statistics */
2865                 id -= rx_entries;
2866                 tx_entries = ENA_STATS_ARRAY_TX * dev->data->nb_tx_queues;
2867                 if (id < tx_entries) {
2868                         qid = id % dev->data->nb_tx_queues;
2869                         id /= dev->data->nb_tx_queues;
2870                         values[i] = *((uint64_t *)
2871                                 &adapter->tx_ring[qid].tx_stats + id);
2872                         ++valid;
2873                         continue;
2874                 }
2875         }
2876
2877         return valid;
2878 }
2879
2880 static int ena_process_bool_devarg(const char *key,
2881                                    const char *value,
2882                                    void *opaque)
2883 {
2884         struct ena_adapter *adapter = opaque;
2885         bool bool_value;
2886
2887         /* Parse the value. */
2888         if (strcmp(value, "1") == 0) {
2889                 bool_value = true;
2890         } else if (strcmp(value, "0") == 0) {
2891                 bool_value = false;
2892         } else {
2893                 PMD_INIT_LOG(ERR,
2894                         "Invalid value: '%s' for key '%s'. Accepted: '0' or '1'\n",
2895                         value, key);
2896                 return -EINVAL;
2897         }
2898
2899         /* Now, assign it to the proper adapter field. */
2900         if (strcmp(key, ENA_DEVARG_LARGE_LLQ_HDR) == 0)
2901                 adapter->use_large_llq_hdr = bool_value;
2902
2903         return 0;
2904 }
2905
2906 static int ena_parse_devargs(struct ena_adapter *adapter,
2907                              struct rte_devargs *devargs)
2908 {
2909         static const char * const allowed_args[] = {
2910                 ENA_DEVARG_LARGE_LLQ_HDR,
2911                 NULL,
2912         };
2913         struct rte_kvargs *kvlist;
2914         int rc;
2915
2916         if (devargs == NULL)
2917                 return 0;
2918
2919         kvlist = rte_kvargs_parse(devargs->args, allowed_args);
2920         if (kvlist == NULL) {
2921                 PMD_INIT_LOG(ERR, "Invalid device arguments: %s\n",
2922                         devargs->args);
2923                 return -EINVAL;
2924         }
2925
2926         rc = rte_kvargs_process(kvlist, ENA_DEVARG_LARGE_LLQ_HDR,
2927                 ena_process_bool_devarg, adapter);
2928
2929         rte_kvargs_free(kvlist);
2930
2931         return rc;
2932 }
2933
2934 /*********************************************************************
2935  *  PMD configuration
2936  *********************************************************************/
2937 static int eth_ena_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
2938         struct rte_pci_device *pci_dev)
2939 {
2940         return rte_eth_dev_pci_generic_probe(pci_dev,
2941                 sizeof(struct ena_adapter), eth_ena_dev_init);
2942 }
2943
2944 static int eth_ena_pci_remove(struct rte_pci_device *pci_dev)
2945 {
2946         return rte_eth_dev_pci_generic_remove(pci_dev, eth_ena_dev_uninit);
2947 }
2948
2949 static struct rte_pci_driver rte_ena_pmd = {
2950         .id_table = pci_id_ena_map,
2951         .drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC |
2952                      RTE_PCI_DRV_WC_ACTIVATE,
2953         .probe = eth_ena_pci_probe,
2954         .remove = eth_ena_pci_remove,
2955 };
2956
2957 RTE_PMD_REGISTER_PCI(net_ena, rte_ena_pmd);
2958 RTE_PMD_REGISTER_PCI_TABLE(net_ena, pci_id_ena_map);
2959 RTE_PMD_REGISTER_KMOD_DEP(net_ena, "* igb_uio | uio_pci_generic | vfio-pci");
2960 RTE_PMD_REGISTER_PARAM_STRING(net_ena, ENA_DEVARG_LARGE_LLQ_HDR "=<0|1>");
2961 RTE_LOG_REGISTER_SUFFIX(ena_logtype_init, init, NOTICE);
2962 RTE_LOG_REGISTER_SUFFIX(ena_logtype_driver, driver, NOTICE);
2963 #ifdef RTE_LIBRTE_ENA_DEBUG_RX
2964 RTE_LOG_REGISTER_SUFFIX(ena_logtype_rx, rx, NOTICE);
2965 #endif
2966 #ifdef RTE_LIBRTE_ENA_DEBUG_TX
2967 RTE_LOG_REGISTER_SUFFIX(ena_logtype_tx, tx, NOTICE);
2968 #endif
2969 #ifdef RTE_LIBRTE_ENA_DEBUG_TX_FREE
2970 RTE_LOG_REGISTER_SUFFIX(ena_logtype_tx_free, tx_free, NOTICE);
2971 #endif
2972 #ifdef RTE_LIBRTE_ENA_COM_DEBUG
2973 RTE_LOG_REGISTER_SUFFIX(ena_logtype_com, com, NOTICE);
2974 #endif
2975
2976 /******************************************************************************
2977  ******************************** AENQ Handlers *******************************
2978  *****************************************************************************/
2979 static void ena_update_on_link_change(void *adapter_data,
2980                                       struct ena_admin_aenq_entry *aenq_e)
2981 {
2982         struct rte_eth_dev *eth_dev = adapter_data;
2983         struct ena_adapter *adapter = eth_dev->data->dev_private;
2984         struct ena_admin_aenq_link_change_desc *aenq_link_desc;
2985         uint32_t status;
2986
2987         aenq_link_desc = (struct ena_admin_aenq_link_change_desc *)aenq_e;
2988
2989         status = get_ena_admin_aenq_link_change_desc_link_status(aenq_link_desc);
2990         adapter->link_status = status;
2991
2992         ena_link_update(eth_dev, 0);
2993         rte_eth_dev_callback_process(eth_dev, RTE_ETH_EVENT_INTR_LSC, NULL);
2994 }
2995
2996 static void ena_notification(void *adapter_data,
2997                              struct ena_admin_aenq_entry *aenq_e)
2998 {
2999         struct rte_eth_dev *eth_dev = adapter_data;
3000         struct ena_adapter *adapter = eth_dev->data->dev_private;
3001         struct ena_admin_ena_hw_hints *hints;
3002
3003         if (aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION)
3004                 PMD_DRV_LOG(WARNING, "Invalid AENQ group: %x. Expected: %x\n",
3005                         aenq_e->aenq_common_desc.group,
3006                         ENA_ADMIN_NOTIFICATION);
3007
3008         switch (aenq_e->aenq_common_desc.syndrome) {
3009         case ENA_ADMIN_UPDATE_HINTS:
3010                 hints = (struct ena_admin_ena_hw_hints *)
3011                         (&aenq_e->inline_data_w4);
3012                 ena_update_hints(adapter, hints);
3013                 break;
3014         default:
3015                 PMD_DRV_LOG(ERR, "Invalid AENQ notification link state: %d\n",
3016                         aenq_e->aenq_common_desc.syndrome);
3017         }
3018 }
3019
3020 static void ena_keep_alive(void *adapter_data,
3021                            __rte_unused struct ena_admin_aenq_entry *aenq_e)
3022 {
3023         struct rte_eth_dev *eth_dev = adapter_data;
3024         struct ena_adapter *adapter = eth_dev->data->dev_private;
3025         struct ena_admin_aenq_keep_alive_desc *desc;
3026         uint64_t rx_drops;
3027         uint64_t tx_drops;
3028
3029         adapter->timestamp_wd = rte_get_timer_cycles();
3030
3031         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
3032         rx_drops = ((uint64_t)desc->rx_drops_high << 32) | desc->rx_drops_low;
3033         tx_drops = ((uint64_t)desc->tx_drops_high << 32) | desc->tx_drops_low;
3034
3035         adapter->drv_stats->rx_drops = rx_drops;
3036         adapter->dev_stats.tx_drops = tx_drops;
3037 }
3038
3039 /**
3040  * This handler will called for unknown event group or unimplemented handlers
3041  **/
3042 static void unimplemented_aenq_handler(__rte_unused void *data,
3043                                        __rte_unused struct ena_admin_aenq_entry *aenq_e)
3044 {
3045         PMD_DRV_LOG(ERR,
3046                 "Unknown event was received or event with unimplemented handler\n");
3047 }
3048
3049 static struct ena_aenq_handlers aenq_handlers = {
3050         .handlers = {
3051                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3052                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
3053                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive
3054         },
3055         .unimplemented_handler = unimplemented_aenq_handler
3056 };