drivers/net/ethernet/amazon/ena/ena_netdev.c

   1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
   2 /*
   3  * Copyright 2015-2020 Amazon.com, Inc. or its affiliates. All rights reserved.
   4  */
   5
   6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8 #ifdef CONFIG_RFS_ACCEL
   9 #include <linux/cpu_rmap.h>
  10 #endif /* CONFIG_RFS_ACCEL */
  11 #include <linux/ethtool.h>
  12 #include <linux/kernel.h>
  13 #include <linux/module.h>
  14 #include <linux/numa.h>
  15 #include <linux/pci.h>
  16 #include <linux/utsname.h>
  17 #include <linux/version.h>
  18 #include <linux/vmalloc.h>
  19 #include <net/ip.h>
  20
  21 #include "ena_netdev.h"
  22 #include <linux/bpf_trace.h>
  23 #include "ena_pci_id_tbl.h"
  24
  25 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
  26 MODULE_DESCRIPTION(DEVICE_NAME);
  27 MODULE_LICENSE("GPL");
  28
  29 /* Time in jiffies before concluding the transmitter is hung. */
  30 #define TX_TIMEOUT  (5 * HZ)
  31
  32 #define ENA_MAX_RINGS min_t(unsigned int, ENA_MAX_NUM_IO_QUEUES, num_possible_cpus())
  33
  34 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
  35                 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
  36
  37 static struct ena_aenq_handlers aenq_handlers;
  38
  39 static struct workqueue_struct *ena_wq;
  40
  41 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
  42
  43 static int ena_rss_init_default(struct ena_adapter *adapter);
  44 static void check_for_admin_com_state(struct ena_adapter *adapter);
  45 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
  46 static int ena_restore_device(struct ena_adapter *adapter);
  47
  48 static void ena_init_io_rings(struct ena_adapter *adapter,
  49                               int first_index, int count);
  50 static void ena_init_napi_in_range(struct ena_adapter *adapter, int first_index,
  51                                    int count);
  52 static void ena_del_napi_in_range(struct ena_adapter *adapter, int first_index,
  53                                   int count);
  54 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid);
  55 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
  56                                            int first_index,
  57                                            int count);
  58 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid);
  59 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid);
  60 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget);
  61 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter);
  62 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter);
  63 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
  64                                       int first_index, int count);
  65 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
  66                                      int first_index, int count);
  67 static int ena_up(struct ena_adapter *adapter);
  68 static void ena_down(struct ena_adapter *adapter);
  69 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
  70                                  struct ena_ring *rx_ring);
  71 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
  72                                       struct ena_ring *rx_ring);
  73 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
  74                               struct ena_tx_buffer *tx_info);
  75 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
  76                                             int first_index, int count);
  77
  78 /* Increase a stat by cnt while holding syncp seqlock on 32bit machines */
  79 static void ena_increase_stat(u64 *statp, u64 cnt,
  80                               struct u64_stats_sync *syncp)
  81 {
  82         u64_stats_update_begin(syncp);
  83         (*statp) += cnt;
  84         u64_stats_update_end(syncp);
  85 }
  86
  87 static void ena_ring_tx_doorbell(struct ena_ring *tx_ring)
  88 {
  89         ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
  90         ena_increase_stat(&tx_ring->tx_stats.doorbells, 1, &tx_ring->syncp);
  91 }
  92
  93 static void ena_tx_timeout(struct net_device *dev, unsigned int txqueue)
  94 {
  95         struct ena_adapter *adapter = netdev_priv(dev);
  96
  97         /* Change the state of the device to trigger reset
  98          * Check that we are not in the middle or a trigger already
  99          */
 100
 101         if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
 102                 return;
 103
 104         ena_reset_device(adapter, ENA_REGS_RESET_OS_NETDEV_WD);
 105         ena_increase_stat(&adapter->dev_stats.tx_timeout, 1, &adapter->syncp);
 106
 107         netif_err(adapter, tx_err, dev, "Transmit time out\n");
 108 }
 109
 110 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
 111 {
 112         int i;
 113
 114         for (i = 0; i < adapter->num_io_queues; i++)
 115                 adapter->rx_ring[i].mtu = mtu;
 116 }
 117
 118 static int ena_change_mtu(struct net_device *dev, int new_mtu)
 119 {
 120         struct ena_adapter *adapter = netdev_priv(dev);
 121         int ret;
 122
 123         ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
 124         if (!ret) {
 125                 netif_dbg(adapter, drv, dev, "Set MTU to %d\n", new_mtu);
 126                 update_rx_ring_mtu(adapter, new_mtu);
 127                 dev->mtu = new_mtu;
 128         } else {
 129                 netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
 130                           new_mtu);
 131         }
 132
 133         return ret;
 134 }
 135
 136 static int ena_xmit_common(struct net_device *dev,
 137                            struct ena_ring *ring,
 138                            struct ena_tx_buffer *tx_info,
 139                            struct ena_com_tx_ctx *ena_tx_ctx,
 140                            u16 next_to_use,
 141                            u32 bytes)
 142 {
 143         struct ena_adapter *adapter = netdev_priv(dev);
 144         int rc, nb_hw_desc;
 145
 146         if (unlikely(ena_com_is_doorbell_needed(ring->ena_com_io_sq,
 147                                                 ena_tx_ctx))) {
 148                 netif_dbg(adapter, tx_queued, dev,
 149                           "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
 150                           ring->qid);
 151                 ena_ring_tx_doorbell(ring);
 152         }
 153
 154         /* prepare the packet's descriptors to dma engine */
 155         rc = ena_com_prepare_tx(ring->ena_com_io_sq, ena_tx_ctx,
 156                                 &nb_hw_desc);
 157
 158         /* In case there isn't enough space in the queue for the packet,
 159          * we simply drop it. All other failure reasons of
 160          * ena_com_prepare_tx() are fatal and therefore require a device reset.
 161          */
 162         if (unlikely(rc)) {
 163                 netif_err(adapter, tx_queued, dev,
 164                           "Failed to prepare tx bufs\n");
 165                 ena_increase_stat(&ring->tx_stats.prepare_ctx_err, 1,
 166                                   &ring->syncp);
 167                 if (rc != -ENOMEM)
 168                         ena_reset_device(adapter,
 169                                          ENA_REGS_RESET_DRIVER_INVALID_STATE);
 170                 return rc;
 171         }
 172
 173         u64_stats_update_begin(&ring->syncp);
 174         ring->tx_stats.cnt++;
 175         ring->tx_stats.bytes += bytes;
 176         u64_stats_update_end(&ring->syncp);
 177
 178         tx_info->tx_descs = nb_hw_desc;
 179         tx_info->last_jiffies = jiffies;
 180         tx_info->print_once = 0;
 181
 182         ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
 183                                                  ring->ring_size);
 184         return 0;
 185 }
 186
 187 /* This is the XDP napi callback. XDP queues use a separate napi callback
 188  * than Rx/Tx queues.
 189  */
 190 static int ena_xdp_io_poll(struct napi_struct *napi, int budget)
 191 {
 192         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
 193         u32 xdp_work_done, xdp_budget;
 194         struct ena_ring *xdp_ring;
 195         int napi_comp_call = 0;
 196         int ret;
 197
 198         xdp_ring = ena_napi->xdp_ring;
 199
 200         xdp_budget = budget;
 201
 202         if (!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags) ||
 203             test_bit(ENA_FLAG_TRIGGER_RESET, &xdp_ring->adapter->flags)) {
 204                 napi_complete_done(napi, 0);
 205                 return 0;
 206         }
 207
 208         xdp_work_done = ena_clean_xdp_irq(xdp_ring, xdp_budget);
 209
 210         /* If the device is about to reset or down, avoid unmask
 211          * the interrupt and return 0 so NAPI won't reschedule
 212          */
 213         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags))) {
 214                 napi_complete_done(napi, 0);
 215                 ret = 0;
 216         } else if (xdp_budget > xdp_work_done) {
 217                 napi_comp_call = 1;
 218                 if (napi_complete_done(napi, xdp_work_done))
 219                         ena_unmask_interrupt(xdp_ring, NULL);
 220                 ena_update_ring_numa_node(xdp_ring, NULL);
 221                 ret = xdp_work_done;
 222         } else {
 223                 ret = xdp_budget;
 224         }
 225
 226         u64_stats_update_begin(&xdp_ring->syncp);
 227         xdp_ring->tx_stats.napi_comp += napi_comp_call;
 228         xdp_ring->tx_stats.tx_poll++;
 229         u64_stats_update_end(&xdp_ring->syncp);
 230         xdp_ring->tx_stats.last_napi_jiffies = jiffies;
 231
 232         return ret;
 233 }
 234
 235 static int ena_xdp_tx_map_frame(struct ena_ring *xdp_ring,
 236                                 struct ena_tx_buffer *tx_info,
 237                                 struct xdp_frame *xdpf,
 238                                 struct ena_com_tx_ctx *ena_tx_ctx)
 239 {
 240         struct ena_adapter *adapter = xdp_ring->adapter;
 241         struct ena_com_buf *ena_buf;
 242         int push_len = 0;
 243         dma_addr_t dma;
 244         void *data;
 245         u32 size;
 246
 247         tx_info->xdpf = xdpf;
 248         data = tx_info->xdpf->data;
 249         size = tx_info->xdpf->len;
 250
 251         if (xdp_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
 252                 /* Designate part of the packet for LLQ */
 253                 push_len = min_t(u32, size, xdp_ring->tx_max_header_size);
 254
 255                 ena_tx_ctx->push_header = data;
 256
 257                 size -= push_len;
 258                 data += push_len;
 259         }
 260
 261         ena_tx_ctx->header_len = push_len;
 262
 263         if (size > 0) {
 264                 dma = dma_map_single(xdp_ring->dev,
 265                                      data,
 266                                      size,
 267                                      DMA_TO_DEVICE);
 268                 if (unlikely(dma_mapping_error(xdp_ring->dev, dma)))
 269                         goto error_report_dma_error;
 270
 271                 tx_info->map_linear_data = 0;
 272
 273                 ena_buf = tx_info->bufs;
 274                 ena_buf->paddr = dma;
 275                 ena_buf->len = size;
 276
 277                 ena_tx_ctx->ena_bufs = ena_buf;
 278                 ena_tx_ctx->num_bufs = tx_info->num_of_bufs = 1;
 279         }
 280
 281         return 0;
 282
 283 error_report_dma_error:
 284         ena_increase_stat(&xdp_ring->tx_stats.dma_mapping_err, 1,
 285                           &xdp_ring->syncp);
 286         netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map xdp buff\n");
 287
 288         return -EINVAL;
 289 }
 290
 291 static int ena_xdp_xmit_frame(struct ena_ring *xdp_ring,
 292                               struct net_device *dev,
 293                               struct xdp_frame *xdpf,
 294                               int flags)
 295 {
 296         struct ena_com_tx_ctx ena_tx_ctx = {};
 297         struct ena_tx_buffer *tx_info;
 298         u16 next_to_use, req_id;
 299         int rc;
 300
 301         next_to_use = xdp_ring->next_to_use;
 302         req_id = xdp_ring->free_ids[next_to_use];
 303         tx_info = &xdp_ring->tx_buffer_info[req_id];
 304         tx_info->num_of_bufs = 0;
 305
 306         rc = ena_xdp_tx_map_frame(xdp_ring, tx_info, xdpf, &ena_tx_ctx);
 307         if (unlikely(rc))
 308                 return rc;
 309
 310         ena_tx_ctx.req_id = req_id;
 311
 312         rc = ena_xmit_common(dev,
 313                              xdp_ring,
 314                              tx_info,
 315                              &ena_tx_ctx,
 316                              next_to_use,
 317                              xdpf->len);
 318         if (rc)
 319                 goto error_unmap_dma;
 320
 321         /* trigger the dma engine. ena_ring_tx_doorbell()
 322          * calls a memory barrier inside it.
 323          */
 324         if (flags & XDP_XMIT_FLUSH)
 325                 ena_ring_tx_doorbell(xdp_ring);
 326
 327         return rc;
 328
 329 error_unmap_dma:
 330         ena_unmap_tx_buff(xdp_ring, tx_info);
 331         tx_info->xdpf = NULL;
 332         return rc;
 333 }
 334
 335 static int ena_xdp_xmit(struct net_device *dev, int n,
 336                         struct xdp_frame **frames, u32 flags)
 337 {
 338         struct ena_adapter *adapter = netdev_priv(dev);
 339         struct ena_ring *xdp_ring;
 340         int qid, i, nxmit = 0;
 341
 342         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 343                 return -EINVAL;
 344
 345         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
 346                 return -ENETDOWN;
 347
 348         /* We assume that all rings have the same XDP program */
 349         if (!READ_ONCE(adapter->rx_ring->xdp_bpf_prog))
 350                 return -ENXIO;
 351
 352         qid = smp_processor_id() % adapter->xdp_num_queues;
 353         qid += adapter->xdp_first_ring;
 354         xdp_ring = &adapter->tx_ring[qid];
 355
 356         /* Other CPU ids might try to send thorugh this queue */
 357         spin_lock(&xdp_ring->xdp_tx_lock);
 358
 359         for (i = 0; i < n; i++) {
 360                 if (ena_xdp_xmit_frame(xdp_ring, dev, frames[i], 0))
 361                         break;
 362                 nxmit++;
 363         }
 364
 365         /* Ring doorbell to make device aware of the packets */
 366         if (flags & XDP_XMIT_FLUSH)
 367                 ena_ring_tx_doorbell(xdp_ring);
 368
 369         spin_unlock(&xdp_ring->xdp_tx_lock);
 370
 371         /* Return number of packets sent */
 372         return nxmit;
 373 }
 374
 375 static int ena_xdp_execute(struct ena_ring *rx_ring, struct xdp_buff *xdp)
 376 {
 377         u32 verdict = ENA_XDP_PASS;
 378         struct bpf_prog *xdp_prog;
 379         struct ena_ring *xdp_ring;
 380         struct xdp_frame *xdpf;
 381         u64 *xdp_stat;
 382
 383         xdp_prog = READ_ONCE(rx_ring->xdp_bpf_prog);
 384
 385         if (!xdp_prog)
 386                 goto out;
 387
 388         verdict = bpf_prog_run_xdp(xdp_prog, xdp);
 389
 390         switch (verdict) {
 391         case XDP_TX:
 392                 xdpf = xdp_convert_buff_to_frame(xdp);
 393                 if (unlikely(!xdpf)) {
 394                         trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 395                         xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 396                         verdict = ENA_XDP_DROP;
 397                         break;
 398                 }
 399
 400                 /* Find xmit queue */
 401                 xdp_ring = rx_ring->xdp_ring;
 402
 403                 /* The XDP queues are shared between XDP_TX and XDP_REDIRECT */
 404                 spin_lock(&xdp_ring->xdp_tx_lock);
 405
 406                 if (ena_xdp_xmit_frame(xdp_ring, rx_ring->netdev, xdpf,
 407                                        XDP_XMIT_FLUSH))
 408                         xdp_return_frame(xdpf);
 409
 410                 spin_unlock(&xdp_ring->xdp_tx_lock);
 411                 xdp_stat = &rx_ring->rx_stats.xdp_tx;
 412                 verdict = ENA_XDP_TX;
 413                 break;
 414         case XDP_REDIRECT:
 415                 if (likely(!xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog))) {
 416                         xdp_stat = &rx_ring->rx_stats.xdp_redirect;
 417                         verdict = ENA_XDP_REDIRECT;
 418                         break;
 419                 }
 420                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 421                 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 422                 verdict = ENA_XDP_DROP;
 423                 break;
 424         case XDP_ABORTED:
 425                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 426                 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 427                 verdict = ENA_XDP_DROP;
 428                 break;
 429         case XDP_DROP:
 430                 xdp_stat = &rx_ring->rx_stats.xdp_drop;
 431                 verdict = ENA_XDP_DROP;
 432                 break;
 433         case XDP_PASS:
 434                 xdp_stat = &rx_ring->rx_stats.xdp_pass;
 435                 verdict = ENA_XDP_PASS;
 436                 break;
 437         default:
 438                 bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, verdict);
 439                 xdp_stat = &rx_ring->rx_stats.xdp_invalid;
 440                 verdict = ENA_XDP_DROP;
 441         }
 442
 443         ena_increase_stat(xdp_stat, 1, &rx_ring->syncp);
 444 out:
 445         return verdict;
 446 }
 447
 448 static void ena_init_all_xdp_queues(struct ena_adapter *adapter)
 449 {
 450         adapter->xdp_first_ring = adapter->num_io_queues;
 451         adapter->xdp_num_queues = adapter->num_io_queues;
 452
 453         ena_init_io_rings(adapter,
 454                           adapter->xdp_first_ring,
 455                           adapter->xdp_num_queues);
 456 }
 457
 458 static int ena_setup_and_create_all_xdp_queues(struct ena_adapter *adapter)
 459 {
 460         int rc = 0;
 461
 462         rc = ena_setup_tx_resources_in_range(adapter, adapter->xdp_first_ring,
 463                                              adapter->xdp_num_queues);
 464         if (rc)
 465                 goto setup_err;
 466
 467         rc = ena_create_io_tx_queues_in_range(adapter,
 468                                               adapter->xdp_first_ring,
 469                                               adapter->xdp_num_queues);
 470         if (rc)
 471                 goto create_err;
 472
 473         return 0;
 474
 475 create_err:
 476         ena_free_all_io_tx_resources(adapter);
 477 setup_err:
 478         return rc;
 479 }
 480
 481 /* Provides a way for both kernel and bpf-prog to know
 482  * more about the RX-queue a given XDP frame arrived on.
 483  */
 484 static int ena_xdp_register_rxq_info(struct ena_ring *rx_ring)
 485 {
 486         int rc;
 487
 488         rc = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev, rx_ring->qid, 0);
 489
 490         if (rc) {
 491                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 492                           "Failed to register xdp rx queue info. RX queue num %d rc: %d\n",
 493                           rx_ring->qid, rc);
 494                 goto err;
 495         }
 496
 497         rc = xdp_rxq_info_reg_mem_model(&rx_ring->xdp_rxq, MEM_TYPE_PAGE_SHARED,
 498                                         NULL);
 499
 500         if (rc) {
 501                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 502                           "Failed to register xdp rx queue info memory model. RX queue num %d rc: %d\n",
 503                           rx_ring->qid, rc);
 504                 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 505         }
 506
 507 err:
 508         return rc;
 509 }
 510
 511 static void ena_xdp_unregister_rxq_info(struct ena_ring *rx_ring)
 512 {
 513         xdp_rxq_info_unreg_mem_model(&rx_ring->xdp_rxq);
 514         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 515 }
 516
 517 static void ena_xdp_exchange_program_rx_in_range(struct ena_adapter *adapter,
 518                                                  struct bpf_prog *prog,
 519                                                  int first, int count)
 520 {
 521         struct bpf_prog *old_bpf_prog;
 522         struct ena_ring *rx_ring;
 523         int i = 0;
 524
 525         for (i = first; i < count; i++) {
 526                 rx_ring = &adapter->rx_ring[i];
 527                 old_bpf_prog = xchg(&rx_ring->xdp_bpf_prog, prog);
 528
 529                 if (!old_bpf_prog && prog) {
 530                         ena_xdp_register_rxq_info(rx_ring);
 531                         rx_ring->rx_headroom = XDP_PACKET_HEADROOM;
 532                 } else if (old_bpf_prog && !prog) {
 533                         ena_xdp_unregister_rxq_info(rx_ring);
 534                         rx_ring->rx_headroom = NET_SKB_PAD;
 535                 }
 536         }
 537 }
 538
 539 static void ena_xdp_exchange_program(struct ena_adapter *adapter,
 540                                      struct bpf_prog *prog)
 541 {
 542         struct bpf_prog *old_bpf_prog = xchg(&adapter->xdp_bpf_prog, prog);
 543
 544         ena_xdp_exchange_program_rx_in_range(adapter,
 545                                              prog,
 546                                              0,
 547                                              adapter->num_io_queues);
 548
 549         if (old_bpf_prog)
 550                 bpf_prog_put(old_bpf_prog);
 551 }
 552
 553 static int ena_destroy_and_free_all_xdp_queues(struct ena_adapter *adapter)
 554 {
 555         bool was_up;
 556         int rc;
 557
 558         was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 559
 560         if (was_up)
 561                 ena_down(adapter);
 562
 563         adapter->xdp_first_ring = 0;
 564         adapter->xdp_num_queues = 0;
 565         ena_xdp_exchange_program(adapter, NULL);
 566         if (was_up) {
 567                 rc = ena_up(adapter);
 568                 if (rc)
 569                         return rc;
 570         }
 571         return 0;
 572 }
 573
 574 static int ena_xdp_set(struct net_device *netdev, struct netdev_bpf *bpf)
 575 {
 576         struct ena_adapter *adapter = netdev_priv(netdev);
 577         struct bpf_prog *prog = bpf->prog;
 578         struct bpf_prog *old_bpf_prog;
 579         int rc, prev_mtu;
 580         bool is_up;
 581
 582         is_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 583         rc = ena_xdp_allowed(adapter);
 584         if (rc == ENA_XDP_ALLOWED) {
 585                 old_bpf_prog = adapter->xdp_bpf_prog;
 586                 if (prog) {
 587                         if (!is_up) {
 588                                 ena_init_all_xdp_queues(adapter);
 589                         } else if (!old_bpf_prog) {
 590                                 ena_down(adapter);
 591                                 ena_init_all_xdp_queues(adapter);
 592                         }
 593                         ena_xdp_exchange_program(adapter, prog);
 594
 595                         if (is_up && !old_bpf_prog) {
 596                                 rc = ena_up(adapter);
 597                                 if (rc)
 598                                         return rc;
 599                         }
 600                         xdp_features_set_redirect_target(netdev, false);
 601                 } else if (old_bpf_prog) {
 602                         xdp_features_clear_redirect_target(netdev);
 603                         rc = ena_destroy_and_free_all_xdp_queues(adapter);
 604                         if (rc)
 605                                 return rc;
 606                 }
 607
 608                 prev_mtu = netdev->max_mtu;
 609                 netdev->max_mtu = prog ? ENA_XDP_MAX_MTU : adapter->max_mtu;
 610
 611                 if (!old_bpf_prog)
 612                         netif_info(adapter, drv, adapter->netdev,
 613                                    "XDP program is set, changing the max_mtu from %d to %d",
 614                                    prev_mtu, netdev->max_mtu);
 615
 616         } else if (rc == ENA_XDP_CURRENT_MTU_TOO_LARGE) {
 617                 netif_err(adapter, drv, adapter->netdev,
 618                           "Failed to set xdp program, the current MTU (%d) is larger than the maximum allowed MTU (%lu) while xdp is on",
 619                           netdev->mtu, ENA_XDP_MAX_MTU);
 620                 NL_SET_ERR_MSG_MOD(bpf->extack,
 621                                    "Failed to set xdp program, the current MTU is larger than the maximum allowed MTU. Check the dmesg for more info");
 622                 return -EINVAL;
 623         } else if (rc == ENA_XDP_NO_ENOUGH_QUEUES) {
 624                 netif_err(adapter, drv, adapter->netdev,
 625                           "Failed to set xdp program, the Rx/Tx channel count should be at most half of the maximum allowed channel count. The current queue count (%d), the maximal queue count (%d)\n",
 626                           adapter->num_io_queues, adapter->max_num_io_queues);
 627                 NL_SET_ERR_MSG_MOD(bpf->extack,
 628                                    "Failed to set xdp program, there is no enough space for allocating XDP queues, Check the dmesg for more info");
 629                 return -EINVAL;
 630         }
 631
 632         return 0;
 633 }
 634
 635 /* This is the main xdp callback, it's used by the kernel to set/unset the xdp
 636  * program as well as to query the current xdp program id.
 637  */
 638 static int ena_xdp(struct net_device *netdev, struct netdev_bpf *bpf)
 639 {
 640         switch (bpf->command) {
 641         case XDP_SETUP_PROG:
 642                 return ena_xdp_set(netdev, bpf);
 643         default:
 644                 return -EINVAL;
 645         }
 646         return 0;
 647 }
 648
 649 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
 650 {
 651 #ifdef CONFIG_RFS_ACCEL
 652         u32 i;
 653         int rc;
 654
 655         adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_io_queues);
 656         if (!adapter->netdev->rx_cpu_rmap)
 657                 return -ENOMEM;
 658         for (i = 0; i < adapter->num_io_queues; i++) {
 659                 int irq_idx = ENA_IO_IRQ_IDX(i);
 660
 661                 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
 662                                       pci_irq_vector(adapter->pdev, irq_idx));
 663                 if (rc) {
 664                         free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
 665                         adapter->netdev->rx_cpu_rmap = NULL;
 666                         return rc;
 667                 }
 668         }
 669 #endif /* CONFIG_RFS_ACCEL */
 670         return 0;
 671 }
 672
 673 static void ena_init_io_rings_common(struct ena_adapter *adapter,
 674                                      struct ena_ring *ring, u16 qid)
 675 {
 676         ring->qid = qid;
 677         ring->pdev = adapter->pdev;
 678         ring->dev = &adapter->pdev->dev;
 679         ring->netdev = adapter->netdev;
 680         ring->napi = &adapter->ena_napi[qid].napi;
 681         ring->adapter = adapter;
 682         ring->ena_dev = adapter->ena_dev;
 683         ring->per_napi_packets = 0;
 684         ring->cpu = 0;
 685         ring->numa_node = 0;
 686         ring->no_interrupt_event_cnt = 0;
 687         u64_stats_init(&ring->syncp);
 688 }
 689
 690 static void ena_init_io_rings(struct ena_adapter *adapter,
 691                               int first_index, int count)
 692 {
 693         struct ena_com_dev *ena_dev;
 694         struct ena_ring *txr, *rxr;
 695         int i;
 696
 697         ena_dev = adapter->ena_dev;
 698
 699         for (i = first_index; i < first_index + count; i++) {
 700                 txr = &adapter->tx_ring[i];
 701                 rxr = &adapter->rx_ring[i];
 702
 703                 /* TX common ring state */
 704                 ena_init_io_rings_common(adapter, txr, i);
 705
 706                 /* TX specific ring state */
 707                 txr->ring_size = adapter->requested_tx_ring_size;
 708                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
 709                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
 710                 txr->sgl_size = adapter->max_tx_sgl_size;
 711                 txr->smoothed_interval =
 712                         ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
 713                 txr->disable_meta_caching = adapter->disable_meta_caching;
 714                 spin_lock_init(&txr->xdp_tx_lock);
 715
 716                 /* Don't init RX queues for xdp queues */
 717                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
 718                         /* RX common ring state */
 719                         ena_init_io_rings_common(adapter, rxr, i);
 720
 721                         /* RX specific ring state */
 722                         rxr->ring_size = adapter->requested_rx_ring_size;
 723                         rxr->rx_copybreak = adapter->rx_copybreak;
 724                         rxr->sgl_size = adapter->max_rx_sgl_size;
 725                         rxr->smoothed_interval =
 726                                 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
 727                         rxr->empty_rx_queue = 0;
 728                         rxr->rx_headroom = NET_SKB_PAD;
 729                         adapter->ena_napi[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
 730                         rxr->xdp_ring = &adapter->tx_ring[i + adapter->num_io_queues];
 731                 }
 732         }
 733 }
 734
 735 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
 736  * @adapter: network interface device structure
 737  * @qid: queue index
 738  *
 739  * Return 0 on success, negative on failure
 740  */
 741 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
 742 {
 743         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 744         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 745         int size, i, node;
 746
 747         if (tx_ring->tx_buffer_info) {
 748                 netif_err(adapter, ifup,
 749                           adapter->netdev, "tx_buffer_info info is not NULL");
 750                 return -EEXIST;
 751         }
 752
 753         size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
 754         node = cpu_to_node(ena_irq->cpu);
 755
 756         tx_ring->tx_buffer_info = vzalloc_node(size, node);
 757         if (!tx_ring->tx_buffer_info) {
 758                 tx_ring->tx_buffer_info = vzalloc(size);
 759                 if (!tx_ring->tx_buffer_info)
 760                         goto err_tx_buffer_info;
 761         }
 762
 763         size = sizeof(u16) * tx_ring->ring_size;
 764         tx_ring->free_ids = vzalloc_node(size, node);
 765         if (!tx_ring->free_ids) {
 766                 tx_ring->free_ids = vzalloc(size);
 767                 if (!tx_ring->free_ids)
 768                         goto err_tx_free_ids;
 769         }
 770
 771         size = tx_ring->tx_max_header_size;
 772         tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
 773         if (!tx_ring->push_buf_intermediate_buf) {
 774                 tx_ring->push_buf_intermediate_buf = vzalloc(size);
 775                 if (!tx_ring->push_buf_intermediate_buf)
 776                         goto err_push_buf_intermediate_buf;
 777         }
 778
 779         /* Req id ring for TX out of order completions */
 780         for (i = 0; i < tx_ring->ring_size; i++)
 781                 tx_ring->free_ids[i] = i;
 782
 783         /* Reset tx statistics */
 784         memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
 785
 786         tx_ring->next_to_use = 0;
 787         tx_ring->next_to_clean = 0;
 788         tx_ring->cpu = ena_irq->cpu;
 789         tx_ring->numa_node = node;
 790         return 0;
 791
 792 err_push_buf_intermediate_buf:
 793         vfree(tx_ring->free_ids);
 794         tx_ring->free_ids = NULL;
 795 err_tx_free_ids:
 796         vfree(tx_ring->tx_buffer_info);
 797         tx_ring->tx_buffer_info = NULL;
 798 err_tx_buffer_info:
 799         return -ENOMEM;
 800 }
 801
 802 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
 803  * @adapter: network interface device structure
 804  * @qid: queue index
 805  *
 806  * Free all transmit software resources
 807  */
 808 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
 809 {
 810         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 811
 812         vfree(tx_ring->tx_buffer_info);
 813         tx_ring->tx_buffer_info = NULL;
 814
 815         vfree(tx_ring->free_ids);
 816         tx_ring->free_ids = NULL;
 817
 818         vfree(tx_ring->push_buf_intermediate_buf);
 819         tx_ring->push_buf_intermediate_buf = NULL;
 820 }
 821
 822 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
 823                                            int first_index,
 824                                            int count)
 825 {
 826         int i, rc = 0;
 827
 828         for (i = first_index; i < first_index + count; i++) {
 829                 rc = ena_setup_tx_resources(adapter, i);
 830                 if (rc)
 831                         goto err_setup_tx;
 832         }
 833
 834         return 0;
 835
 836 err_setup_tx:
 837
 838         netif_err(adapter, ifup, adapter->netdev,
 839                   "Tx queue %d: allocation failed\n", i);
 840
 841         /* rewind the index freeing the rings as we go */
 842         while (first_index < i--)
 843                 ena_free_tx_resources(adapter, i);
 844         return rc;
 845 }
 846
 847 static void ena_free_all_io_tx_resources_in_range(struct ena_adapter *adapter,
 848                                                   int first_index, int count)
 849 {
 850         int i;
 851
 852         for (i = first_index; i < first_index + count; i++)
 853                 ena_free_tx_resources(adapter, i);
 854 }
 855
 856 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
 857  * @adapter: board private structure
 858  *
 859  * Free all transmit software resources
 860  */
 861 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
 862 {
 863         ena_free_all_io_tx_resources_in_range(adapter,
 864                                               0,
 865                                               adapter->xdp_num_queues +
 866                                               adapter->num_io_queues);
 867 }
 868
 869 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
 870  * @adapter: network interface device structure
 871  * @qid: queue index
 872  *
 873  * Returns 0 on success, negative on failure
 874  */
 875 static int ena_setup_rx_resources(struct ena_adapter *adapter,
 876                                   u32 qid)
 877 {
 878         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 879         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 880         int size, node, i;
 881
 882         if (rx_ring->rx_buffer_info) {
 883                 netif_err(adapter, ifup, adapter->netdev,
 884                           "rx_buffer_info is not NULL");
 885                 return -EEXIST;
 886         }
 887
 888         /* alloc extra element so in rx path
 889          * we can always prefetch rx_info + 1
 890          */
 891         size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
 892         node = cpu_to_node(ena_irq->cpu);
 893
 894         rx_ring->rx_buffer_info = vzalloc_node(size, node);
 895         if (!rx_ring->rx_buffer_info) {
 896                 rx_ring->rx_buffer_info = vzalloc(size);
 897                 if (!rx_ring->rx_buffer_info)
 898                         return -ENOMEM;
 899         }
 900
 901         size = sizeof(u16) * rx_ring->ring_size;
 902         rx_ring->free_ids = vzalloc_node(size, node);
 903         if (!rx_ring->free_ids) {
 904                 rx_ring->free_ids = vzalloc(size);
 905                 if (!rx_ring->free_ids) {
 906                         vfree(rx_ring->rx_buffer_info);
 907                         rx_ring->rx_buffer_info = NULL;
 908                         return -ENOMEM;
 909                 }
 910         }
 911
 912         /* Req id ring for receiving RX pkts out of order */
 913         for (i = 0; i < rx_ring->ring_size; i++)
 914                 rx_ring->free_ids[i] = i;
 915
 916         /* Reset rx statistics */
 917         memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
 918
 919         rx_ring->next_to_clean = 0;
 920         rx_ring->next_to_use = 0;
 921         rx_ring->cpu = ena_irq->cpu;
 922         rx_ring->numa_node = node;
 923
 924         return 0;
 925 }
 926
 927 /* ena_free_rx_resources - Free I/O Rx Resources
 928  * @adapter: network interface device structure
 929  * @qid: queue index
 930  *
 931  * Free all receive software resources
 932  */
 933 static void ena_free_rx_resources(struct ena_adapter *adapter,
 934                                   u32 qid)
 935 {
 936         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 937
 938         vfree(rx_ring->rx_buffer_info);
 939         rx_ring->rx_buffer_info = NULL;
 940
 941         vfree(rx_ring->free_ids);
 942         rx_ring->free_ids = NULL;
 943 }
 944
 945 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
 946  * @adapter: board private structure
 947  *
 948  * Return 0 on success, negative on failure
 949  */
 950 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
 951 {
 952         int i, rc = 0;
 953
 954         for (i = 0; i < adapter->num_io_queues; i++) {
 955                 rc = ena_setup_rx_resources(adapter, i);
 956                 if (rc)
 957                         goto err_setup_rx;
 958         }
 959
 960         return 0;
 961
 962 err_setup_rx:
 963
 964         netif_err(adapter, ifup, adapter->netdev,
 965                   "Rx queue %d: allocation failed\n", i);
 966
 967         /* rewind the index freeing the rings as we go */
 968         while (i--)
 969                 ena_free_rx_resources(adapter, i);
 970         return rc;
 971 }
 972
 973 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
 974  * @adapter: board private structure
 975  *
 976  * Free all receive software resources
 977  */
 978 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
 979 {
 980         int i;
 981
 982         for (i = 0; i < adapter->num_io_queues; i++)
 983                 ena_free_rx_resources(adapter, i);
 984 }
 985
 986 static struct page *ena_alloc_map_page(struct ena_ring *rx_ring,
 987                                        dma_addr_t *dma)
 988 {
 989         struct page *page;
 990
 991         /* This would allocate the page on the same NUMA node the executing code
 992          * is running on.
 993          */
 994         page = dev_alloc_page();
 995         if (!page) {
 996                 ena_increase_stat(&rx_ring->rx_stats.page_alloc_fail, 1,
 997                                   &rx_ring->syncp);
 998                 return ERR_PTR(-ENOSPC);
 999         }
1000
1001         /* To enable NIC-side port-mirroring, AKA SPAN port,
1002          * we make the buffer readable from the nic as well
1003          */
1004         *dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
1005                             DMA_BIDIRECTIONAL);
1006         if (unlikely(dma_mapping_error(rx_ring->dev, *dma))) {
1007                 ena_increase_stat(&rx_ring->rx_stats.dma_mapping_err, 1,
1008                                   &rx_ring->syncp);
1009                 __free_page(page);
1010                 return ERR_PTR(-EIO);
1011         }
1012
1013         return page;
1014 }
1015
1016 static int ena_alloc_rx_buffer(struct ena_ring *rx_ring,
1017                                struct ena_rx_buffer *rx_info)
1018 {
1019         int headroom = rx_ring->rx_headroom;
1020         struct ena_com_buf *ena_buf;
1021         struct page *page;
1022         dma_addr_t dma;
1023         int tailroom;
1024
1025         /* restore page offset value in case it has been changed by device */
1026         rx_info->buf_offset = headroom;
1027
1028         /* if previous allocated page is not used */
1029         if (unlikely(rx_info->page))
1030                 return 0;
1031
1032         /* We handle DMA here */
1033         page = ena_alloc_map_page(rx_ring, &dma);
1034         if (unlikely(IS_ERR(page)))
1035                 return PTR_ERR(page);
1036
1037         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1038                   "Allocate page %p, rx_info %p\n", page, rx_info);
1039
1040         tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1041
1042         rx_info->page = page;
1043         rx_info->dma_addr = dma;
1044         rx_info->page_offset = 0;
1045         ena_buf = &rx_info->ena_buf;
1046         ena_buf->paddr = dma + headroom;
1047         ena_buf->len = ENA_PAGE_SIZE - headroom - tailroom;
1048
1049         return 0;
1050 }
1051
1052 static void ena_unmap_rx_buff_attrs(struct ena_ring *rx_ring,
1053                                     struct ena_rx_buffer *rx_info,
1054                                     unsigned long attrs)
1055 {
1056         dma_unmap_page_attrs(rx_ring->dev, rx_info->dma_addr, ENA_PAGE_SIZE,
1057                              DMA_BIDIRECTIONAL, attrs);
1058 }
1059
1060 static void ena_free_rx_page(struct ena_ring *rx_ring,
1061                              struct ena_rx_buffer *rx_info)
1062 {
1063         struct page *page = rx_info->page;
1064
1065         if (unlikely(!page)) {
1066                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1067                            "Trying to free unallocated buffer\n");
1068                 return;
1069         }
1070
1071         ena_unmap_rx_buff_attrs(rx_ring, rx_info, 0);
1072
1073         __free_page(page);
1074         rx_info->page = NULL;
1075 }
1076
1077 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
1078 {
1079         u16 next_to_use, req_id;
1080         u32 i;
1081         int rc;
1082
1083         next_to_use = rx_ring->next_to_use;
1084
1085         for (i = 0; i < num; i++) {
1086                 struct ena_rx_buffer *rx_info;
1087
1088                 req_id = rx_ring->free_ids[next_to_use];
1089
1090                 rx_info = &rx_ring->rx_buffer_info[req_id];
1091
1092                 rc = ena_alloc_rx_buffer(rx_ring, rx_info);
1093                 if (unlikely(rc < 0)) {
1094                         netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1095                                    "Failed to allocate buffer for rx queue %d\n",
1096                                    rx_ring->qid);
1097                         break;
1098                 }
1099                 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1100                                                 &rx_info->ena_buf,
1101                                                 req_id);
1102                 if (unlikely(rc)) {
1103                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1104                                    "Failed to add buffer for rx queue %d\n",
1105                                    rx_ring->qid);
1106                         break;
1107                 }
1108                 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1109                                                    rx_ring->ring_size);
1110         }
1111
1112         if (unlikely(i < num)) {
1113                 ena_increase_stat(&rx_ring->rx_stats.refil_partial, 1,
1114                                   &rx_ring->syncp);
1115                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1116                            "Refilled rx qid %d with only %d buffers (from %d)\n",
1117                            rx_ring->qid, i, num);
1118         }
1119
1120         /* ena_com_write_sq_doorbell issues a wmb() */
1121         if (likely(i))
1122                 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1123
1124         rx_ring->next_to_use = next_to_use;
1125
1126         return i;
1127 }
1128
1129 static void ena_free_rx_bufs(struct ena_adapter *adapter,
1130                              u32 qid)
1131 {
1132         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1133         u32 i;
1134
1135         for (i = 0; i < rx_ring->ring_size; i++) {
1136                 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1137
1138                 if (rx_info->page)
1139                         ena_free_rx_page(rx_ring, rx_info);
1140         }
1141 }
1142
1143 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
1144  * @adapter: board private structure
1145  */
1146 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1147 {
1148         struct ena_ring *rx_ring;
1149         int i, rc, bufs_num;
1150
1151         for (i = 0; i < adapter->num_io_queues; i++) {
1152                 rx_ring = &adapter->rx_ring[i];
1153                 bufs_num = rx_ring->ring_size - 1;
1154                 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1155
1156                 if (unlikely(rc != bufs_num))
1157                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1158                                    "Refilling Queue %d failed. allocated %d buffers from: %d\n",
1159                                    i, rc, bufs_num);
1160         }
1161 }
1162
1163 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
1164 {
1165         int i;
1166
1167         for (i = 0; i < adapter->num_io_queues; i++)
1168                 ena_free_rx_bufs(adapter, i);
1169 }
1170
1171 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
1172                               struct ena_tx_buffer *tx_info)
1173 {
1174         struct ena_com_buf *ena_buf;
1175         u32 cnt;
1176         int i;
1177
1178         ena_buf = tx_info->bufs;
1179         cnt = tx_info->num_of_bufs;
1180
1181         if (unlikely(!cnt))
1182                 return;
1183
1184         if (tx_info->map_linear_data) {
1185                 dma_unmap_single(tx_ring->dev,
1186                                  dma_unmap_addr(ena_buf, paddr),
1187                                  dma_unmap_len(ena_buf, len),
1188                                  DMA_TO_DEVICE);
1189                 ena_buf++;
1190                 cnt--;
1191         }
1192
1193         /* unmap remaining mapped pages */
1194         for (i = 0; i < cnt; i++) {
1195                 dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
1196                                dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
1197                 ena_buf++;
1198         }
1199 }
1200
1201 /* ena_free_tx_bufs - Free Tx Buffers per Queue
1202  * @tx_ring: TX ring for which buffers be freed
1203  */
1204 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
1205 {
1206         bool print_once = true;
1207         u32 i;
1208
1209         for (i = 0; i < tx_ring->ring_size; i++) {
1210                 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1211
1212                 if (!tx_info->skb)
1213                         continue;
1214
1215                 if (print_once) {
1216                         netif_notice(tx_ring->adapter, ifdown, tx_ring->netdev,
1217                                      "Free uncompleted tx skb qid %d idx 0x%x\n",
1218                                      tx_ring->qid, i);
1219                         print_once = false;
1220                 } else {
1221                         netif_dbg(tx_ring->adapter, ifdown, tx_ring->netdev,
1222                                   "Free uncompleted tx skb qid %d idx 0x%x\n",
1223                                   tx_ring->qid, i);
1224                 }
1225
1226                 ena_unmap_tx_buff(tx_ring, tx_info);
1227
1228                 dev_kfree_skb_any(tx_info->skb);
1229         }
1230         netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
1231                                                   tx_ring->qid));
1232 }
1233
1234 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
1235 {
1236         struct ena_ring *tx_ring;
1237         int i;
1238
1239         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1240                 tx_ring = &adapter->tx_ring[i];
1241                 ena_free_tx_bufs(tx_ring);
1242         }
1243 }
1244
1245 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1246 {
1247         u16 ena_qid;
1248         int i;
1249
1250         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1251                 ena_qid = ENA_IO_TXQ_IDX(i);
1252                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1253         }
1254 }
1255
1256 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1257 {
1258         u16 ena_qid;
1259         int i;
1260
1261         for (i = 0; i < adapter->num_io_queues; i++) {
1262                 ena_qid = ENA_IO_RXQ_IDX(i);
1263                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
1264                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1265         }
1266 }
1267
1268 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
1269 {
1270         ena_destroy_all_tx_queues(adapter);
1271         ena_destroy_all_rx_queues(adapter);
1272 }
1273
1274 static int handle_invalid_req_id(struct ena_ring *ring, u16 req_id,
1275                                  struct ena_tx_buffer *tx_info, bool is_xdp)
1276 {
1277         if (tx_info)
1278                 netif_err(ring->adapter,
1279                           tx_done,
1280                           ring->netdev,
1281                           "tx_info doesn't have valid %s. qid %u req_id %u",
1282                            is_xdp ? "xdp frame" : "skb", ring->qid, req_id);
1283         else
1284                 netif_err(ring->adapter,
1285                           tx_done,
1286                           ring->netdev,
1287                           "Invalid req_id %u in qid %u\n",
1288                           req_id, ring->qid);
1289
1290         ena_increase_stat(&ring->tx_stats.bad_req_id, 1, &ring->syncp);
1291         ena_reset_device(ring->adapter, ENA_REGS_RESET_INV_TX_REQ_ID);
1292
1293         return -EFAULT;
1294 }
1295
1296 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
1297 {
1298         struct ena_tx_buffer *tx_info;
1299
1300         tx_info = &tx_ring->tx_buffer_info[req_id];
1301         if (likely(tx_info->skb))
1302                 return 0;
1303
1304         return handle_invalid_req_id(tx_ring, req_id, tx_info, false);
1305 }
1306
1307 static int validate_xdp_req_id(struct ena_ring *xdp_ring, u16 req_id)
1308 {
1309         struct ena_tx_buffer *tx_info;
1310
1311         tx_info = &xdp_ring->tx_buffer_info[req_id];
1312         if (likely(tx_info->xdpf))
1313                 return 0;
1314
1315         return handle_invalid_req_id(xdp_ring, req_id, tx_info, true);
1316 }
1317
1318 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
1319 {
1320         struct netdev_queue *txq;
1321         bool above_thresh;
1322         u32 tx_bytes = 0;
1323         u32 total_done = 0;
1324         u16 next_to_clean;
1325         u16 req_id;
1326         int tx_pkts = 0;
1327         int rc;
1328
1329         next_to_clean = tx_ring->next_to_clean;
1330         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
1331
1332         while (tx_pkts < budget) {
1333                 struct ena_tx_buffer *tx_info;
1334                 struct sk_buff *skb;
1335
1336                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
1337                                                 &req_id);
1338                 if (rc) {
1339                         if (unlikely(rc == -EINVAL))
1340                                 handle_invalid_req_id(tx_ring, req_id, NULL,
1341                                                       false);
1342                         break;
1343                 }
1344
1345                 /* validate that the request id points to a valid skb */
1346                 rc = validate_tx_req_id(tx_ring, req_id);
1347                 if (rc)
1348                         break;
1349
1350                 tx_info = &tx_ring->tx_buffer_info[req_id];
1351                 skb = tx_info->skb;
1352
1353                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
1354                 prefetch(&skb->end);
1355
1356                 tx_info->skb = NULL;
1357                 tx_info->last_jiffies = 0;
1358
1359                 ena_unmap_tx_buff(tx_ring, tx_info);
1360
1361                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1362                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
1363                           skb);
1364
1365                 tx_bytes += skb->len;
1366                 dev_kfree_skb(skb);
1367                 tx_pkts++;
1368                 total_done += tx_info->tx_descs;
1369
1370                 tx_ring->free_ids[next_to_clean] = req_id;
1371                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1372                                                      tx_ring->ring_size);
1373         }
1374
1375         tx_ring->next_to_clean = next_to_clean;
1376         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
1377         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
1378
1379         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
1380
1381         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1382                   "tx_poll: q %d done. total pkts: %d\n",
1383                   tx_ring->qid, tx_pkts);
1384
1385         /* need to make the rings circular update visible to
1386          * ena_start_xmit() before checking for netif_queue_stopped().
1387          */
1388         smp_mb();
1389
1390         above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1391                                                     ENA_TX_WAKEUP_THRESH);
1392         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
1393                 __netif_tx_lock(txq, smp_processor_id());
1394                 above_thresh =
1395                         ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1396                                                      ENA_TX_WAKEUP_THRESH);
1397                 if (netif_tx_queue_stopped(txq) && above_thresh &&
1398                     test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
1399                         netif_tx_wake_queue(txq);
1400                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
1401                                           &tx_ring->syncp);
1402                 }
1403                 __netif_tx_unlock(txq);
1404         }
1405
1406         return tx_pkts;
1407 }
1408
1409 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, void *first_frag, u16 len)
1410 {
1411         struct sk_buff *skb;
1412
1413         if (!first_frag)
1414                 skb = napi_alloc_skb(rx_ring->napi, len);
1415         else
1416                 skb = napi_build_skb(first_frag, len);
1417
1418         if (unlikely(!skb)) {
1419                 ena_increase_stat(&rx_ring->rx_stats.skb_alloc_fail, 1,
1420                                   &rx_ring->syncp);
1421
1422                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1423                           "Failed to allocate skb. first_frag %s\n",
1424                           first_frag ? "provided" : "not provided");
1425         }
1426
1427         return skb;
1428 }
1429
1430 static bool ena_try_rx_buf_page_reuse(struct ena_rx_buffer *rx_info, u16 buf_len,
1431                                       u16 len, int pkt_offset)
1432 {
1433         struct ena_com_buf *ena_buf = &rx_info->ena_buf;
1434
1435         /* More than ENA_MIN_RX_BUF_SIZE left in the reused buffer
1436          * for data + headroom + tailroom.
1437          */
1438         if (SKB_DATA_ALIGN(len + pkt_offset) + ENA_MIN_RX_BUF_SIZE <= ena_buf->len) {
1439                 page_ref_inc(rx_info->page);
1440                 rx_info->page_offset += buf_len;
1441                 ena_buf->paddr += buf_len;
1442                 ena_buf->len -= buf_len;
1443                 return true;
1444         }
1445
1446         return false;
1447 }
1448
1449 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
1450                                   struct ena_com_rx_buf_info *ena_bufs,
1451                                   u32 descs,
1452                                   u16 *next_to_clean)
1453 {
1454         int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1455         bool is_xdp_loaded = ena_xdp_present_ring(rx_ring);
1456         struct ena_rx_buffer *rx_info;
1457         struct ena_adapter *adapter;
1458         int page_offset, pkt_offset;
1459         dma_addr_t pre_reuse_paddr;
1460         u16 len, req_id, buf = 0;
1461         bool reuse_rx_buf_page;
1462         struct sk_buff *skb;
1463         void *buf_addr;
1464         int buf_offset;
1465         u16 buf_len;
1466
1467         len = ena_bufs[buf].len;
1468         req_id = ena_bufs[buf].req_id;
1469
1470         rx_info = &rx_ring->rx_buffer_info[req_id];
1471
1472         if (unlikely(!rx_info->page)) {
1473                 adapter = rx_ring->adapter;
1474                 netif_err(adapter, rx_err, rx_ring->netdev,
1475                           "Page is NULL. qid %u req_id %u\n", rx_ring->qid, req_id);
1476                 ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1, &rx_ring->syncp);
1477                 ena_reset_device(adapter, ENA_REGS_RESET_INV_RX_REQ_ID);
1478                 return NULL;
1479         }
1480
1481         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1482                   "rx_info %p page %p\n",
1483                   rx_info, rx_info->page);
1484
1485         buf_offset = rx_info->buf_offset;
1486         pkt_offset = buf_offset - rx_ring->rx_headroom;
1487         page_offset = rx_info->page_offset;
1488         buf_addr = page_address(rx_info->page) + page_offset;
1489
1490         if (len <= rx_ring->rx_copybreak) {
1491                 skb = ena_alloc_skb(rx_ring, NULL, len);
1492                 if (unlikely(!skb))
1493                         return NULL;
1494
1495                 /* sync this buffer for CPU use */
1496                 dma_sync_single_for_cpu(rx_ring->dev,
1497                                         dma_unmap_addr(&rx_info->ena_buf, paddr) + pkt_offset,
1498                                         len,
1499                                         DMA_FROM_DEVICE);
1500                 skb_copy_to_linear_data(skb, buf_addr + buf_offset, len);
1501                 dma_sync_single_for_device(rx_ring->dev,
1502                                            dma_unmap_addr(&rx_info->ena_buf, paddr) + pkt_offset,
1503                                            len,
1504                                            DMA_FROM_DEVICE);
1505
1506                 skb_put(skb, len);
1507                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1508                           "RX allocated small packet. len %d.\n", skb->len);
1509                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1510                 rx_ring->free_ids[*next_to_clean] = req_id;
1511                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
1512                                                      rx_ring->ring_size);
1513                 return skb;
1514         }
1515
1516         buf_len = SKB_DATA_ALIGN(len + buf_offset + tailroom);
1517
1518         pre_reuse_paddr = dma_unmap_addr(&rx_info->ena_buf, paddr);
1519
1520         /* If XDP isn't loaded try to reuse part of the RX buffer */
1521         reuse_rx_buf_page = !is_xdp_loaded &&
1522                             ena_try_rx_buf_page_reuse(rx_info, buf_len, len, pkt_offset);
1523
1524         dma_sync_single_for_cpu(rx_ring->dev,
1525                                 pre_reuse_paddr + pkt_offset,
1526                                 len,
1527                                 DMA_FROM_DEVICE);
1528
1529         if (!reuse_rx_buf_page)
1530                 ena_unmap_rx_buff_attrs(rx_ring, rx_info, DMA_ATTR_SKIP_CPU_SYNC);
1531
1532         skb = ena_alloc_skb(rx_ring, buf_addr, buf_len);
1533         if (unlikely(!skb))
1534                 return NULL;
1535
1536         /* Populate skb's linear part */
1537         skb_reserve(skb, buf_offset);
1538         skb_put(skb, len);
1539         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1540
1541         do {
1542                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1543                           "RX skb updated. len %d. data_len %d\n",
1544                           skb->len, skb->data_len);
1545
1546                 if (!reuse_rx_buf_page)
1547                         rx_info->page = NULL;
1548
1549                 rx_ring->free_ids[*next_to_clean] = req_id;
1550                 *next_to_clean =
1551                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
1552                                              rx_ring->ring_size);
1553                 if (likely(--descs == 0))
1554                         break;
1555
1556                 buf++;
1557                 len = ena_bufs[buf].len;
1558                 req_id = ena_bufs[buf].req_id;
1559
1560                 rx_info = &rx_ring->rx_buffer_info[req_id];
1561
1562                 /* rx_info->buf_offset includes rx_ring->rx_headroom */
1563                 buf_offset = rx_info->buf_offset;
1564                 pkt_offset = buf_offset - rx_ring->rx_headroom;
1565                 buf_len = SKB_DATA_ALIGN(len + buf_offset + tailroom);
1566                 page_offset = rx_info->page_offset;
1567
1568                 pre_reuse_paddr = dma_unmap_addr(&rx_info->ena_buf, paddr);
1569
1570                 reuse_rx_buf_page = !is_xdp_loaded &&
1571                                     ena_try_rx_buf_page_reuse(rx_info, buf_len, len, pkt_offset);
1572
1573                 dma_sync_single_for_cpu(rx_ring->dev,
1574                                         pre_reuse_paddr + pkt_offset,
1575                                         len,
1576                                         DMA_FROM_DEVICE);
1577
1578                 if (!reuse_rx_buf_page)
1579                         ena_unmap_rx_buff_attrs(rx_ring, rx_info,
1580                                                 DMA_ATTR_SKIP_CPU_SYNC);
1581
1582                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
1583                                 page_offset + buf_offset, len, buf_len);
1584
1585         } while (1);
1586
1587         return skb;
1588 }
1589
1590 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
1591  * @adapter: structure containing adapter specific data
1592  * @ena_rx_ctx: received packet context/metadata
1593  * @skb: skb currently being received and modified
1594  */
1595 static void ena_rx_checksum(struct ena_ring *rx_ring,
1596                                    struct ena_com_rx_ctx *ena_rx_ctx,
1597                                    struct sk_buff *skb)
1598 {
1599         /* Rx csum disabled */
1600         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
1601                 skb->ip_summed = CHECKSUM_NONE;
1602                 return;
1603         }
1604
1605         /* For fragmented packets the checksum isn't valid */
1606         if (ena_rx_ctx->frag) {
1607                 skb->ip_summed = CHECKSUM_NONE;
1608                 return;
1609         }
1610
1611         /* if IP and error */
1612         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1613                      (ena_rx_ctx->l3_csum_err))) {
1614                 /* ipv4 checksum error */
1615                 skb->ip_summed = CHECKSUM_NONE;
1616                 ena_increase_stat(&rx_ring->rx_stats.csum_bad, 1,
1617                                   &rx_ring->syncp);
1618                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1619                           "RX IPv4 header checksum error\n");
1620                 return;
1621         }
1622
1623         /* if TCP/UDP */
1624         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1625                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
1626                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
1627                         /* TCP/UDP checksum error */
1628                         ena_increase_stat(&rx_ring->rx_stats.csum_bad, 1,
1629                                           &rx_ring->syncp);
1630                         netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1631                                   "RX L4 checksum error\n");
1632                         skb->ip_summed = CHECKSUM_NONE;
1633                         return;
1634                 }
1635
1636                 if (likely(ena_rx_ctx->l4_csum_checked)) {
1637                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1638                         ena_increase_stat(&rx_ring->rx_stats.csum_good, 1,
1639                                           &rx_ring->syncp);
1640                 } else {
1641                         ena_increase_stat(&rx_ring->rx_stats.csum_unchecked, 1,
1642                                           &rx_ring->syncp);
1643                         skb->ip_summed = CHECKSUM_NONE;
1644                 }
1645         } else {
1646                 skb->ip_summed = CHECKSUM_NONE;
1647                 return;
1648         }
1649
1650 }
1651
1652 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1653                             struct ena_com_rx_ctx *ena_rx_ctx,
1654                             struct sk_buff *skb)
1655 {
1656         enum pkt_hash_types hash_type;
1657
1658         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1659                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1660                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1661
1662                         hash_type = PKT_HASH_TYPE_L4;
1663                 else
1664                         hash_type = PKT_HASH_TYPE_NONE;
1665
1666                 /* Override hash type if the packet is fragmented */
1667                 if (ena_rx_ctx->frag)
1668                         hash_type = PKT_HASH_TYPE_NONE;
1669
1670                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1671         }
1672 }
1673
1674 static int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)
1675 {
1676         struct ena_rx_buffer *rx_info;
1677         int ret;
1678
1679         rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1680         xdp_prepare_buff(xdp, page_address(rx_info->page),
1681                          rx_info->buf_offset,
1682                          rx_ring->ena_bufs[0].len, false);
1683         /* If for some reason we received a bigger packet than
1684          * we expect, then we simply drop it
1685          */
1686         if (unlikely(rx_ring->ena_bufs[0].len > ENA_XDP_MAX_MTU))
1687                 return ENA_XDP_DROP;
1688
1689         ret = ena_xdp_execute(rx_ring, xdp);
1690
1691         /* The xdp program might expand the headers */
1692         if (ret == ENA_XDP_PASS) {
1693                 rx_info->buf_offset = xdp->data - xdp->data_hard_start;
1694                 rx_ring->ena_bufs[0].len = xdp->data_end - xdp->data;
1695         }
1696
1697         return ret;
1698 }
1699 /* ena_clean_rx_irq - Cleanup RX irq
1700  * @rx_ring: RX ring to clean
1701  * @napi: napi handler
1702  * @budget: how many packets driver is allowed to clean
1703  *
1704  * Returns the number of cleaned buffers.
1705  */
1706 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1707                             u32 budget)
1708 {
1709         u16 next_to_clean = rx_ring->next_to_clean;
1710         struct ena_com_rx_ctx ena_rx_ctx;
1711         struct ena_rx_buffer *rx_info;
1712         struct ena_adapter *adapter;
1713         u32 res_budget, work_done;
1714         int rx_copybreak_pkt = 0;
1715         int refill_threshold;
1716         struct sk_buff *skb;
1717         int refill_required;
1718         struct xdp_buff xdp;
1719         int xdp_flags = 0;
1720         int total_len = 0;
1721         int xdp_verdict;
1722         int rc = 0;
1723         int i;
1724
1725         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1726                   "%s qid %d\n", __func__, rx_ring->qid);
1727         res_budget = budget;
1728         xdp_init_buff(&xdp, ENA_PAGE_SIZE, &rx_ring->xdp_rxq);
1729
1730         do {
1731                 xdp_verdict = ENA_XDP_PASS;
1732                 skb = NULL;
1733                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1734                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1735                 ena_rx_ctx.descs = 0;
1736                 ena_rx_ctx.pkt_offset = 0;
1737                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1738                                     rx_ring->ena_com_io_sq,
1739                                     &ena_rx_ctx);
1740                 if (unlikely(rc))
1741                         goto error;
1742
1743                 if (unlikely(ena_rx_ctx.descs == 0))
1744                         break;
1745
1746                 /* First descriptor might have an offset set by the device */
1747                 rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1748                 rx_info->buf_offset += ena_rx_ctx.pkt_offset;
1749
1750                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1751                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1752                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1753                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1754
1755                 if (ena_xdp_present_ring(rx_ring))
1756                         xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1757
1758                 /* allocate skb and fill it */
1759                 if (xdp_verdict == ENA_XDP_PASS)
1760                         skb = ena_rx_skb(rx_ring,
1761                                          rx_ring->ena_bufs,
1762                                          ena_rx_ctx.descs,
1763                                          &next_to_clean);
1764
1765                 if (unlikely(!skb)) {
1766                         for (i = 0; i < ena_rx_ctx.descs; i++) {
1767                                 int req_id = rx_ring->ena_bufs[i].req_id;
1768
1769                                 rx_ring->free_ids[next_to_clean] = req_id;
1770                                 next_to_clean =
1771                                         ENA_RX_RING_IDX_NEXT(next_to_clean,
1772                                                              rx_ring->ring_size);
1773
1774                                 /* Packets was passed for transmission, unmap it
1775                                  * from RX side.
1776                                  */
1777                                 if (xdp_verdict & ENA_XDP_FORWARDED) {
1778                                         ena_unmap_rx_buff_attrs(rx_ring,
1779                                                                 &rx_ring->rx_buffer_info[req_id],
1780                                                                 0);
1781                                         rx_ring->rx_buffer_info[req_id].page = NULL;
1782                                 }
1783                         }
1784                         if (xdp_verdict != ENA_XDP_PASS) {
1785                                 xdp_flags |= xdp_verdict;
1786                                 total_len += ena_rx_ctx.ena_bufs[0].len;
1787                                 res_budget--;
1788                                 continue;
1789                         }
1790                         break;
1791                 }
1792
1793                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1794
1795                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1796
1797                 skb_record_rx_queue(skb, rx_ring->qid);
1798
1799                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak)
1800                         rx_copybreak_pkt++;
1801
1802                 total_len += skb->len;
1803
1804                 napi_gro_receive(napi, skb);
1805
1806                 res_budget--;
1807         } while (likely(res_budget));
1808
1809         work_done = budget - res_budget;
1810         rx_ring->per_napi_packets += work_done;
1811         u64_stats_update_begin(&rx_ring->syncp);
1812         rx_ring->rx_stats.bytes += total_len;
1813         rx_ring->rx_stats.cnt += work_done;
1814         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1815         u64_stats_update_end(&rx_ring->syncp);
1816
1817         rx_ring->next_to_clean = next_to_clean;
1818
1819         refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
1820         refill_threshold =
1821                 min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1822                       ENA_RX_REFILL_THRESH_PACKET);
1823
1824         /* Optimization, try to batch new rx buffers */
1825         if (refill_required > refill_threshold) {
1826                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1827                 ena_refill_rx_bufs(rx_ring, refill_required);
1828         }
1829
1830         if (xdp_flags & ENA_XDP_REDIRECT)
1831                 xdp_do_flush_map();
1832
1833         return work_done;
1834
1835 error:
1836         adapter = netdev_priv(rx_ring->netdev);
1837
1838         if (rc == -ENOSPC) {
1839                 ena_increase_stat(&rx_ring->rx_stats.bad_desc_num, 1,
1840                                   &rx_ring->syncp);
1841                 ena_reset_device(adapter, ENA_REGS_RESET_TOO_MANY_RX_DESCS);
1842         } else {
1843                 ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1,
1844                                   &rx_ring->syncp);
1845                 ena_reset_device(adapter, ENA_REGS_RESET_INV_RX_REQ_ID);
1846         }
1847         return 0;
1848 }
1849
1850 static void ena_dim_work(struct work_struct *w)
1851 {
1852         struct dim *dim = container_of(w, struct dim, work);
1853         struct dim_cq_moder cur_moder =
1854                 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1855         struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1856
1857         ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1858         dim->state = DIM_START_MEASURE;
1859 }
1860
1861 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1862 {
1863         struct dim_sample dim_sample;
1864         struct ena_ring *rx_ring = ena_napi->rx_ring;
1865
1866         if (!rx_ring->per_napi_packets)
1867                 return;
1868
1869         rx_ring->non_empty_napi_events++;
1870
1871         dim_update_sample(rx_ring->non_empty_napi_events,
1872                           rx_ring->rx_stats.cnt,
1873                           rx_ring->rx_stats.bytes,
1874                           &dim_sample);
1875
1876         net_dim(&ena_napi->dim, dim_sample);
1877
1878         rx_ring->per_napi_packets = 0;
1879 }
1880
1881 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1882                                         struct ena_ring *rx_ring)
1883 {
1884         u32 rx_interval = tx_ring->smoothed_interval;
1885         struct ena_eth_io_intr_reg intr_reg;
1886
1887         /* Rx ring can be NULL when for XDP tx queues which don't have an
1888          * accompanying rx_ring pair.
1889          */
1890         if (rx_ring)
1891                 rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1892                         rx_ring->smoothed_interval :
1893                         ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1894
1895         /* Update intr register: rx intr delay,
1896          * tx intr delay and interrupt unmask
1897          */
1898         ena_com_update_intr_reg(&intr_reg,
1899                                 rx_interval,
1900                                 tx_ring->smoothed_interval,
1901                                 true);
1902
1903         ena_increase_stat(&tx_ring->tx_stats.unmask_interrupt, 1,
1904                           &tx_ring->syncp);
1905
1906         /* It is a shared MSI-X.
1907          * Tx and Rx CQ have pointer to it.
1908          * So we use one of them to reach the intr reg
1909          * The Tx ring is used because the rx_ring is NULL for XDP queues
1910          */
1911         ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1912 }
1913
1914 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1915                                              struct ena_ring *rx_ring)
1916 {
1917         int cpu = get_cpu();
1918         int numa_node;
1919
1920         /* Check only one ring since the 2 rings are running on the same cpu */
1921         if (likely(tx_ring->cpu == cpu))
1922                 goto out;
1923
1924         tx_ring->cpu = cpu;
1925         if (rx_ring)
1926                 rx_ring->cpu = cpu;
1927
1928         numa_node = cpu_to_node(cpu);
1929
1930         if (likely(tx_ring->numa_node == numa_node))
1931                 goto out;
1932
1933         put_cpu();
1934
1935         if (numa_node != NUMA_NO_NODE) {
1936                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1937                 tx_ring->numa_node = numa_node;
1938                 if (rx_ring) {
1939                         rx_ring->numa_node = numa_node;
1940                         ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1941                                                  numa_node);
1942                 }
1943         }
1944
1945         return;
1946 out:
1947         put_cpu();
1948 }
1949
1950 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1951 {
1952         u32 total_done = 0;
1953         u16 next_to_clean;
1954         int tx_pkts = 0;
1955         u16 req_id;
1956         int rc;
1957
1958         if (unlikely(!xdp_ring))
1959                 return 0;
1960         next_to_clean = xdp_ring->next_to_clean;
1961
1962         while (tx_pkts < budget) {
1963                 struct ena_tx_buffer *tx_info;
1964                 struct xdp_frame *xdpf;
1965
1966                 rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1967                                                 &req_id);
1968                 if (rc) {
1969                         if (unlikely(rc == -EINVAL))
1970                                 handle_invalid_req_id(xdp_ring, req_id, NULL,
1971                                                       true);
1972                         break;
1973                 }
1974
1975                 /* validate that the request id points to a valid xdp_frame */
1976                 rc = validate_xdp_req_id(xdp_ring, req_id);
1977                 if (rc)
1978                         break;
1979
1980                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1981                 xdpf = tx_info->xdpf;
1982
1983                 tx_info->xdpf = NULL;
1984                 tx_info->last_jiffies = 0;
1985                 ena_unmap_tx_buff(xdp_ring, tx_info);
1986
1987                 netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1988                           "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1989                           xdpf);
1990
1991                 tx_pkts++;
1992                 total_done += tx_info->tx_descs;
1993
1994                 xdp_return_frame(xdpf);
1995                 xdp_ring->free_ids[next_to_clean] = req_id;
1996                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1997                                                      xdp_ring->ring_size);
1998         }
1999
2000         xdp_ring->next_to_clean = next_to_clean;
2001         ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
2002         ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
2003
2004         netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
2005                   "tx_poll: q %d done. total pkts: %d\n",
2006                   xdp_ring->qid, tx_pkts);
2007
2008         return tx_pkts;
2009 }
2010
2011 static int ena_io_poll(struct napi_struct *napi, int budget)
2012 {
2013         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
2014         struct ena_ring *tx_ring, *rx_ring;
2015         int tx_work_done;
2016         int rx_work_done = 0;
2017         int tx_budget;
2018         int napi_comp_call = 0;
2019         int ret;
2020
2021         tx_ring = ena_napi->tx_ring;
2022         rx_ring = ena_napi->rx_ring;
2023
2024         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
2025
2026         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
2027             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
2028                 napi_complete_done(napi, 0);
2029                 return 0;
2030         }
2031
2032         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
2033         /* On netpoll the budget is zero and the handler should only clean the
2034          * tx completions.
2035          */
2036         if (likely(budget))
2037                 rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
2038
2039         /* If the device is about to reset or down, avoid unmask
2040          * the interrupt and return 0 so NAPI won't reschedule
2041          */
2042         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
2043                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
2044                 napi_complete_done(napi, 0);
2045                 ret = 0;
2046
2047         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
2048                 napi_comp_call = 1;
2049
2050                 /* Update numa and unmask the interrupt only when schedule
2051                  * from the interrupt context (vs from sk_busy_loop)
2052                  */
2053                 if (napi_complete_done(napi, rx_work_done) &&
2054                     READ_ONCE(ena_napi->interrupts_masked)) {
2055                         smp_rmb(); /* make sure interrupts_masked is read */
2056                         WRITE_ONCE(ena_napi->interrupts_masked, false);
2057                         /* We apply adaptive moderation on Rx path only.
2058                          * Tx uses static interrupt moderation.
2059                          */
2060                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
2061                                 ena_adjust_adaptive_rx_intr_moderation(ena_napi);
2062
2063                         ena_update_ring_numa_node(tx_ring, rx_ring);
2064                         ena_unmask_interrupt(tx_ring, rx_ring);
2065                 }
2066
2067                 ret = rx_work_done;
2068         } else {
2069                 ret = budget;
2070         }
2071
2072         u64_stats_update_begin(&tx_ring->syncp);
2073         tx_ring->tx_stats.napi_comp += napi_comp_call;
2074         tx_ring->tx_stats.tx_poll++;
2075         u64_stats_update_end(&tx_ring->syncp);
2076
2077         tx_ring->tx_stats.last_napi_jiffies = jiffies;
2078
2079         return ret;
2080 }
2081
2082 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
2083 {
2084         struct ena_adapter *adapter = (struct ena_adapter *)data;
2085
2086         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
2087
2088         /* Don't call the aenq handler before probe is done */
2089         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
2090                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
2091
2092         return IRQ_HANDLED;
2093 }
2094
2095 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
2096  * @irq: interrupt number
2097  * @data: pointer to a network interface private napi device structure
2098  */
2099 static irqreturn_t ena_intr_msix_io(int irq, void *data)
2100 {
2101         struct ena_napi *ena_napi = data;
2102
2103         /* Used to check HW health */
2104         WRITE_ONCE(ena_napi->first_interrupt, true);
2105
2106         WRITE_ONCE(ena_napi->interrupts_masked, true);
2107         smp_wmb(); /* write interrupts_masked before calling napi */
2108
2109         napi_schedule_irqoff(&ena_napi->napi);
2110
2111         return IRQ_HANDLED;
2112 }
2113
2114 /* Reserve a single MSI-X vector for management (admin + aenq).
2115  * plus reserve one vector for each potential io queue.
2116  * the number of potential io queues is the minimum of what the device
2117  * supports and the number of vCPUs.
2118  */
2119 static int ena_enable_msix(struct ena_adapter *adapter)
2120 {
2121         int msix_vecs, irq_cnt;
2122
2123         if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2124                 netif_err(adapter, probe, adapter->netdev,
2125                           "Error, MSI-X is already enabled\n");
2126                 return -EPERM;
2127         }
2128
2129         /* Reserved the max msix vectors we might need */
2130         msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
2131         netif_dbg(adapter, probe, adapter->netdev,
2132                   "Trying to enable MSI-X, vectors %d\n", msix_vecs);
2133
2134         irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
2135                                         msix_vecs, PCI_IRQ_MSIX);
2136
2137         if (irq_cnt < 0) {
2138                 netif_err(adapter, probe, adapter->netdev,
2139                           "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
2140                 return -ENOSPC;
2141         }
2142
2143         if (irq_cnt != msix_vecs) {
2144                 netif_notice(adapter, probe, adapter->netdev,
2145                              "Enable only %d MSI-X (out of %d), reduce the number of queues\n",
2146                              irq_cnt, msix_vecs);
2147                 adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
2148         }
2149
2150         if (ena_init_rx_cpu_rmap(adapter))
2151                 netif_warn(adapter, probe, adapter->netdev,
2152                            "Failed to map IRQs to CPUs\n");
2153
2154         adapter->msix_vecs = irq_cnt;
2155         set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2156
2157         return 0;
2158 }
2159
2160 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2161 {
2162         u32 cpu;
2163
2164         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2165                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2166                  pci_name(adapter->pdev));
2167         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2168                 ena_intr_msix_mgmnt;
2169         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2170         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2171                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2172         cpu = cpumask_first(cpu_online_mask);
2173         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2174         cpumask_set_cpu(cpu,
2175                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2176 }
2177
2178 static void ena_setup_io_intr(struct ena_adapter *adapter)
2179 {
2180         struct net_device *netdev;
2181         int irq_idx, i, cpu;
2182         int io_queue_count;
2183
2184         netdev = adapter->netdev;
2185         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2186
2187         for (i = 0; i < io_queue_count; i++) {
2188                 irq_idx = ENA_IO_IRQ_IDX(i);
2189                 cpu = i % num_online_cpus();
2190
2191                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2192                          "%s-Tx-Rx-%d", netdev->name, i);
2193                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2194                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2195                 adapter->irq_tbl[irq_idx].vector =
2196                         pci_irq_vector(adapter->pdev, irq_idx);
2197                 adapter->irq_tbl[irq_idx].cpu = cpu;
2198
2199                 cpumask_set_cpu(cpu,
2200                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
2201         }
2202 }
2203
2204 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2205 {
2206         unsigned long flags = 0;
2207         struct ena_irq *irq;
2208         int rc;
2209
2210         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2211         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2212                          irq->data);
2213         if (rc) {
2214                 netif_err(adapter, probe, adapter->netdev,
2215                           "Failed to request admin irq\n");
2216                 return rc;
2217         }
2218
2219         netif_dbg(adapter, probe, adapter->netdev,
2220                   "Set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2221                   irq->affinity_hint_mask.bits[0], irq->vector);
2222
2223         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2224
2225         return rc;
2226 }
2227
2228 static int ena_request_io_irq(struct ena_adapter *adapter)
2229 {
2230         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2231         unsigned long flags = 0;
2232         struct ena_irq *irq;
2233         int rc = 0, i, k;
2234
2235         if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2236                 netif_err(adapter, ifup, adapter->netdev,
2237                           "Failed to request I/O IRQ: MSI-X is not enabled\n");
2238                 return -EINVAL;
2239         }
2240
2241         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2242                 irq = &adapter->irq_tbl[i];
2243                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2244                                  irq->data);
2245                 if (rc) {
2246                         netif_err(adapter, ifup, adapter->netdev,
2247                                   "Failed to request I/O IRQ. index %d rc %d\n",
2248                                    i, rc);
2249                         goto err;
2250                 }
2251
2252                 netif_dbg(adapter, ifup, adapter->netdev,
2253                           "Set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2254                           i, irq->affinity_hint_mask.bits[0], irq->vector);
2255
2256                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2257         }
2258
2259         return rc;
2260
2261 err:
2262         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2263                 irq = &adapter->irq_tbl[k];
2264                 free_irq(irq->vector, irq->data);
2265         }
2266
2267         return rc;
2268 }
2269
2270 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2271 {
2272         struct ena_irq *irq;
2273
2274         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2275         synchronize_irq(irq->vector);
2276         irq_set_affinity_hint(irq->vector, NULL);
2277         free_irq(irq->vector, irq->data);
2278 }
2279
2280 static void ena_free_io_irq(struct ena_adapter *adapter)
2281 {
2282         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2283         struct ena_irq *irq;
2284         int i;
2285
2286 #ifdef CONFIG_RFS_ACCEL
2287         if (adapter->msix_vecs >= 1) {
2288                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2289                 adapter->netdev->rx_cpu_rmap = NULL;
2290         }
2291 #endif /* CONFIG_RFS_ACCEL */
2292
2293         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2294                 irq = &adapter->irq_tbl[i];
2295                 irq_set_affinity_hint(irq->vector, NULL);
2296                 free_irq(irq->vector, irq->data);
2297         }
2298 }
2299
2300 static void ena_disable_msix(struct ena_adapter *adapter)
2301 {
2302         if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2303                 pci_free_irq_vectors(adapter->pdev);
2304 }
2305
2306 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2307 {
2308         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2309         int i;
2310
2311         if (!netif_running(adapter->netdev))
2312                 return;
2313
2314         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2315                 synchronize_irq(adapter->irq_tbl[i].vector);
2316 }
2317
2318 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2319                                   int first_index,
2320                                   int count)
2321 {
2322         int i;
2323
2324         for (i = first_index; i < first_index + count; i++) {
2325                 netif_napi_del(&adapter->ena_napi[i].napi);
2326
2327                 WARN_ON(!ENA_IS_XDP_INDEX(adapter, i) &&
2328                         adapter->ena_napi[i].xdp_ring);
2329         }
2330 }
2331
2332 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2333                                    int first_index, int count)
2334 {
2335         int i;
2336
2337         for (i = first_index; i < first_index + count; i++) {
2338                 struct ena_napi *napi = &adapter->ena_napi[i];
2339
2340                 netif_napi_add(adapter->netdev, &napi->napi,
2341                                ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll);
2342
2343                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
2344                         napi->rx_ring = &adapter->rx_ring[i];
2345                         napi->tx_ring = &adapter->tx_ring[i];
2346                 } else {
2347                         napi->xdp_ring = &adapter->tx_ring[i];
2348                 }
2349                 napi->qid = i;
2350         }
2351 }
2352
2353 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2354                                       int first_index,
2355                                       int count)
2356 {
2357         int i;
2358
2359         for (i = first_index; i < first_index + count; i++)
2360                 napi_disable(&adapter->ena_napi[i].napi);
2361 }
2362
2363 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2364                                      int first_index,
2365                                      int count)
2366 {
2367         int i;
2368
2369         for (i = first_index; i < first_index + count; i++)
2370                 napi_enable(&adapter->ena_napi[i].napi);
2371 }
2372
2373 /* Configure the Rx forwarding */
2374 static int ena_rss_configure(struct ena_adapter *adapter)
2375 {
2376         struct ena_com_dev *ena_dev = adapter->ena_dev;
2377         int rc;
2378
2379         /* In case the RSS table wasn't initialized by probe */
2380         if (!ena_dev->rss.tbl_log_size) {
2381                 rc = ena_rss_init_default(adapter);
2382                 if (rc && (rc != -EOPNOTSUPP)) {
2383                         netif_err(adapter, ifup, adapter->netdev,
2384                                   "Failed to init RSS rc: %d\n", rc);
2385                         return rc;
2386                 }
2387         }
2388
2389         /* Set indirect table */
2390         rc = ena_com_indirect_table_set(ena_dev);
2391         if (unlikely(rc && rc != -EOPNOTSUPP))
2392                 return rc;
2393
2394         /* Configure hash function (if supported) */
2395         rc = ena_com_set_hash_function(ena_dev);
2396         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2397                 return rc;
2398
2399         /* Configure hash inputs (if supported) */
2400         rc = ena_com_set_hash_ctrl(ena_dev);
2401         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2402                 return rc;
2403
2404         return 0;
2405 }
2406
2407 static int ena_up_complete(struct ena_adapter *adapter)
2408 {
2409         int rc;
2410
2411         rc = ena_rss_configure(adapter);
2412         if (rc)
2413                 return rc;
2414
2415         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2416
2417         ena_refill_all_rx_bufs(adapter);
2418
2419         /* enable transmits */
2420         netif_tx_start_all_queues(adapter->netdev);
2421
2422         ena_napi_enable_in_range(adapter,
2423                                  0,
2424                                  adapter->xdp_num_queues + adapter->num_io_queues);
2425
2426         return 0;
2427 }
2428
2429 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2430 {
2431         struct ena_com_create_io_ctx ctx;
2432         struct ena_com_dev *ena_dev;
2433         struct ena_ring *tx_ring;
2434         u32 msix_vector;
2435         u16 ena_qid;
2436         int rc;
2437
2438         ena_dev = adapter->ena_dev;
2439
2440         tx_ring = &adapter->tx_ring[qid];
2441         msix_vector = ENA_IO_IRQ_IDX(qid);
2442         ena_qid = ENA_IO_TXQ_IDX(qid);
2443
2444         memset(&ctx, 0x0, sizeof(ctx));
2445
2446         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2447         ctx.qid = ena_qid;
2448         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2449         ctx.msix_vector = msix_vector;
2450         ctx.queue_size = tx_ring->ring_size;
2451         ctx.numa_node = tx_ring->numa_node;
2452
2453         rc = ena_com_create_io_queue(ena_dev, &ctx);
2454         if (rc) {
2455                 netif_err(adapter, ifup, adapter->netdev,
2456                           "Failed to create I/O TX queue num %d rc: %d\n",
2457                           qid, rc);
2458                 return rc;
2459         }
2460
2461         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2462                                      &tx_ring->ena_com_io_sq,
2463                                      &tx_ring->ena_com_io_cq);
2464         if (rc) {
2465                 netif_err(adapter, ifup, adapter->netdev,
2466                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2467                           qid, rc);
2468                 ena_com_destroy_io_queue(ena_dev, ena_qid);
2469                 return rc;
2470         }
2471
2472         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2473         return rc;
2474 }
2475
2476 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2477                                             int first_index, int count)
2478 {
2479         struct ena_com_dev *ena_dev = adapter->ena_dev;
2480         int rc, i;
2481
2482         for (i = first_index; i < first_index + count; i++) {
2483                 rc = ena_create_io_tx_queue(adapter, i);
2484                 if (rc)
2485                         goto create_err;
2486         }
2487
2488         return 0;
2489
2490 create_err:
2491         while (i-- > first_index)
2492                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2493
2494         return rc;
2495 }
2496
2497 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2498 {
2499         struct ena_com_dev *ena_dev;
2500         struct ena_com_create_io_ctx ctx;
2501         struct ena_ring *rx_ring;
2502         u32 msix_vector;
2503         u16 ena_qid;
2504         int rc;
2505
2506         ena_dev = adapter->ena_dev;
2507
2508         rx_ring = &adapter->rx_ring[qid];
2509         msix_vector = ENA_IO_IRQ_IDX(qid);
2510         ena_qid = ENA_IO_RXQ_IDX(qid);
2511
2512         memset(&ctx, 0x0, sizeof(ctx));
2513
2514         ctx.qid = ena_qid;
2515         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2516         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2517         ctx.msix_vector = msix_vector;
2518         ctx.queue_size = rx_ring->ring_size;
2519         ctx.numa_node = rx_ring->numa_node;
2520
2521         rc = ena_com_create_io_queue(ena_dev, &ctx);
2522         if (rc) {
2523                 netif_err(adapter, ifup, adapter->netdev,
2524                           "Failed to create I/O RX queue num %d rc: %d\n",
2525                           qid, rc);
2526                 return rc;
2527         }
2528
2529         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2530                                      &rx_ring->ena_com_io_sq,
2531                                      &rx_ring->ena_com_io_cq);
2532         if (rc) {
2533                 netif_err(adapter, ifup, adapter->netdev,
2534                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2535                           qid, rc);
2536                 goto err;
2537         }
2538
2539         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2540
2541         return rc;
2542 err:
2543         ena_com_destroy_io_queue(ena_dev, ena_qid);
2544         return rc;
2545 }
2546
2547 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2548 {
2549         struct ena_com_dev *ena_dev = adapter->ena_dev;
2550         int rc, i;
2551
2552         for (i = 0; i < adapter->num_io_queues; i++) {
2553                 rc = ena_create_io_rx_queue(adapter, i);
2554                 if (rc)
2555                         goto create_err;
2556                 INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2557         }
2558
2559         return 0;
2560
2561 create_err:
2562         while (i--) {
2563                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
2564                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2565         }
2566
2567         return rc;
2568 }
2569
2570 static void set_io_rings_size(struct ena_adapter *adapter,
2571                               int new_tx_size,
2572                               int new_rx_size)
2573 {
2574         int i;
2575
2576         for (i = 0; i < adapter->num_io_queues; i++) {
2577                 adapter->tx_ring[i].ring_size = new_tx_size;
2578                 adapter->rx_ring[i].ring_size = new_rx_size;
2579         }
2580 }
2581
2582 /* This function allows queue allocation to backoff when the system is
2583  * low on memory. If there is not enough memory to allocate io queues
2584  * the driver will try to allocate smaller queues.
2585  *
2586  * The backoff algorithm is as follows:
2587  *  1. Try to allocate TX and RX and if successful.
2588  *  1.1. return success
2589  *
2590  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2591  *
2592  *  3. If TX or RX is smaller than 256
2593  *  3.1. return failure.
2594  *  4. else
2595  *  4.1. go back to 1.
2596  */
2597 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2598 {
2599         int rc, cur_rx_ring_size, cur_tx_ring_size;
2600         int new_rx_ring_size, new_tx_ring_size;
2601
2602         /* current queue sizes might be set to smaller than the requested
2603          * ones due to past queue allocation failures.
2604          */
2605         set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2606                           adapter->requested_rx_ring_size);
2607
2608         while (1) {
2609                 if (ena_xdp_present(adapter)) {
2610                         rc = ena_setup_and_create_all_xdp_queues(adapter);
2611
2612                         if (rc)
2613                                 goto err_setup_tx;
2614                 }
2615                 rc = ena_setup_tx_resources_in_range(adapter,
2616                                                      0,
2617                                                      adapter->num_io_queues);
2618                 if (rc)
2619                         goto err_setup_tx;
2620
2621                 rc = ena_create_io_tx_queues_in_range(adapter,
2622                                                       0,
2623                                                       adapter->num_io_queues);
2624                 if (rc)
2625                         goto err_create_tx_queues;
2626
2627                 rc = ena_setup_all_rx_resources(adapter);
2628                 if (rc)
2629                         goto err_setup_rx;
2630
2631                 rc = ena_create_all_io_rx_queues(adapter);
2632                 if (rc)
2633                         goto err_create_rx_queues;
2634
2635                 return 0;
2636
2637 err_create_rx_queues:
2638                 ena_free_all_io_rx_resources(adapter);
2639 err_setup_rx:
2640                 ena_destroy_all_tx_queues(adapter);
2641 err_create_tx_queues:
2642                 ena_free_all_io_tx_resources(adapter);
2643 err_setup_tx:
2644                 if (rc != -ENOMEM) {
2645                         netif_err(adapter, ifup, adapter->netdev,
2646                                   "Queue creation failed with error code %d\n",
2647                                   rc);
2648                         return rc;
2649                 }
2650
2651                 cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2652                 cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2653
2654                 netif_err(adapter, ifup, adapter->netdev,
2655                           "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2656                           cur_tx_ring_size, cur_rx_ring_size);
2657
2658                 new_tx_ring_size = cur_tx_ring_size;
2659                 new_rx_ring_size = cur_rx_ring_size;
2660
2661                 /* Decrease the size of the larger queue, or
2662                  * decrease both if they are the same size.
2663                  */
2664                 if (cur_rx_ring_size <= cur_tx_ring_size)
2665                         new_tx_ring_size = cur_tx_ring_size / 2;
2666                 if (cur_rx_ring_size >= cur_tx_ring_size)
2667                         new_rx_ring_size = cur_rx_ring_size / 2;
2668
2669                 if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2670                     new_rx_ring_size < ENA_MIN_RING_SIZE) {
2671                         netif_err(adapter, ifup, adapter->netdev,
2672                                   "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2673                                   ENA_MIN_RING_SIZE);
2674                         return rc;
2675                 }
2676
2677                 netif_err(adapter, ifup, adapter->netdev,
2678                           "Retrying queue creation with sizes TX=%d, RX=%d\n",
2679                           new_tx_ring_size,
2680                           new_rx_ring_size);
2681
2682                 set_io_rings_size(adapter, new_tx_ring_size,
2683                                   new_rx_ring_size);
2684         }
2685 }
2686
2687 static int ena_up(struct ena_adapter *adapter)
2688 {
2689         int io_queue_count, rc, i;
2690
2691         netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
2692
2693         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2694         ena_setup_io_intr(adapter);
2695
2696         /* napi poll functions should be initialized before running
2697          * request_irq(), to handle a rare condition where there is a pending
2698          * interrupt, causing the ISR to fire immediately while the poll
2699          * function wasn't set yet, causing a null dereference
2700          */
2701         ena_init_napi_in_range(adapter, 0, io_queue_count);
2702
2703         rc = ena_request_io_irq(adapter);
2704         if (rc)
2705                 goto err_req_irq;
2706
2707         rc = create_queues_with_size_backoff(adapter);
2708         if (rc)
2709                 goto err_create_queues_with_backoff;
2710
2711         rc = ena_up_complete(adapter);
2712         if (rc)
2713                 goto err_up;
2714
2715         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2716                 netif_carrier_on(adapter->netdev);
2717
2718         ena_increase_stat(&adapter->dev_stats.interface_up, 1,
2719                           &adapter->syncp);
2720
2721         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2722
2723         /* Enable completion queues interrupt */
2724         for (i = 0; i < adapter->num_io_queues; i++)
2725                 ena_unmask_interrupt(&adapter->tx_ring[i],
2726                                      &adapter->rx_ring[i]);
2727
2728         /* schedule napi in case we had pending packets
2729          * from the last time we disable napi
2730          */
2731         for (i = 0; i < io_queue_count; i++)
2732                 napi_schedule(&adapter->ena_napi[i].napi);
2733
2734         return rc;
2735
2736 err_up:
2737         ena_destroy_all_tx_queues(adapter);
2738         ena_free_all_io_tx_resources(adapter);
2739         ena_destroy_all_rx_queues(adapter);
2740         ena_free_all_io_rx_resources(adapter);
2741 err_create_queues_with_backoff:
2742         ena_free_io_irq(adapter);
2743 err_req_irq:
2744         ena_del_napi_in_range(adapter, 0, io_queue_count);
2745
2746         return rc;
2747 }
2748
2749 static void ena_down(struct ena_adapter *adapter)
2750 {
2751         int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2752
2753         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2754
2755         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2756
2757         ena_increase_stat(&adapter->dev_stats.interface_down, 1,
2758                           &adapter->syncp);
2759
2760         netif_carrier_off(adapter->netdev);
2761         netif_tx_disable(adapter->netdev);
2762
2763         /* After this point the napi handler won't enable the tx queue */
2764         ena_napi_disable_in_range(adapter, 0, io_queue_count);
2765
2766         /* After destroy the queue there won't be any new interrupts */
2767
2768         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2769                 int rc;
2770
2771                 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2772                 if (rc)
2773                         netif_err(adapter, ifdown, adapter->netdev,
2774                                   "Device reset failed\n");
2775                 /* stop submitting admin commands on a device that was reset */
2776                 ena_com_set_admin_running_state(adapter->ena_dev, false);
2777         }
2778
2779         ena_destroy_all_io_queues(adapter);
2780
2781         ena_disable_io_intr_sync(adapter);
2782         ena_free_io_irq(adapter);
2783         ena_del_napi_in_range(adapter, 0, io_queue_count);
2784
2785         ena_free_all_tx_bufs(adapter);
2786         ena_free_all_rx_bufs(adapter);
2787         ena_free_all_io_tx_resources(adapter);
2788         ena_free_all_io_rx_resources(adapter);
2789 }
2790
2791 /* ena_open - Called when a network interface is made active
2792  * @netdev: network interface device structure
2793  *
2794  * Returns 0 on success, negative value on failure
2795  *
2796  * The open entry point is called when a network interface is made
2797  * active by the system (IFF_UP).  At this point all resources needed
2798  * for transmit and receive operations are allocated, the interrupt
2799  * handler is registered with the OS, the watchdog timer is started,
2800  * and the stack is notified that the interface is ready.
2801  */
2802 static int ena_open(struct net_device *netdev)
2803 {
2804         struct ena_adapter *adapter = netdev_priv(netdev);
2805         int rc;
2806
2807         /* Notify the stack of the actual queue counts. */
2808         rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2809         if (rc) {
2810                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2811                 return rc;
2812         }
2813
2814         rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2815         if (rc) {
2816                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2817                 return rc;
2818         }
2819
2820         rc = ena_up(adapter);
2821         if (rc)
2822                 return rc;
2823
2824         return rc;
2825 }
2826
2827 /* ena_close - Disables a network interface
2828  * @netdev: network interface device structure
2829  *
2830  * Returns 0, this is not allowed to fail
2831  *
2832  * The close entry point is called when an interface is de-activated
2833  * by the OS.  The hardware is still under the drivers control, but
2834  * needs to be disabled.  A global MAC reset is issued to stop the
2835  * hardware, and all transmit and receive resources are freed.
2836  */
2837 static int ena_close(struct net_device *netdev)
2838 {
2839         struct ena_adapter *adapter = netdev_priv(netdev);
2840
2841         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2842
2843         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2844                 return 0;
2845
2846         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2847                 ena_down(adapter);
2848
2849         /* Check for device status and issue reset if needed*/
2850         check_for_admin_com_state(adapter);
2851         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2852                 netif_err(adapter, ifdown, adapter->netdev,
2853                           "Destroy failure, restarting device\n");
2854                 ena_dump_stats_to_dmesg(adapter);
2855                 /* rtnl lock already obtained in dev_ioctl() layer */
2856                 ena_destroy_device(adapter, false);
2857                 ena_restore_device(adapter);
2858         }
2859
2860         return 0;
2861 }
2862
2863 int ena_update_queue_params(struct ena_adapter *adapter,
2864                             u32 new_tx_size,
2865                             u32 new_rx_size,
2866                             u32 new_llq_header_len)
2867 {
2868         bool dev_was_up, large_llq_changed = false;
2869         int rc = 0;
2870
2871         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2872         ena_close(adapter->netdev);
2873         adapter->requested_tx_ring_size = new_tx_size;
2874         adapter->requested_rx_ring_size = new_rx_size;
2875         ena_init_io_rings(adapter,
2876                           0,
2877                           adapter->xdp_num_queues +
2878                           adapter->num_io_queues);
2879
2880         large_llq_changed = adapter->ena_dev->tx_mem_queue_type ==
2881                             ENA_ADMIN_PLACEMENT_POLICY_DEV;
2882         large_llq_changed &=
2883                 new_llq_header_len != adapter->ena_dev->tx_max_header_size;
2884
2885         /* a check that the configuration is valid is done by caller */
2886         if (large_llq_changed) {
2887                 adapter->large_llq_header_enabled = !adapter->large_llq_header_enabled;
2888
2889                 ena_destroy_device(adapter, false);
2890                 rc = ena_restore_device(adapter);
2891         }
2892
2893         return dev_was_up && !rc ? ena_up(adapter) : rc;
2894 }
2895
2896 int ena_set_rx_copybreak(struct ena_adapter *adapter, u32 rx_copybreak)
2897 {
2898         struct ena_ring *rx_ring;
2899         int i;
2900
2901         if (rx_copybreak > min_t(u16, adapter->netdev->mtu, ENA_PAGE_SIZE))
2902                 return -EINVAL;
2903
2904         adapter->rx_copybreak = rx_copybreak;
2905
2906         for (i = 0; i < adapter->num_io_queues; i++) {
2907                 rx_ring = &adapter->rx_ring[i];
2908                 rx_ring->rx_copybreak = rx_copybreak;
2909         }
2910
2911         return 0;
2912 }
2913
2914 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2915 {
2916         struct ena_com_dev *ena_dev = adapter->ena_dev;
2917         int prev_channel_count;
2918         bool dev_was_up;
2919
2920         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2921         ena_close(adapter->netdev);
2922         prev_channel_count = adapter->num_io_queues;
2923         adapter->num_io_queues = new_channel_count;
2924         if (ena_xdp_present(adapter) &&
2925             ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2926                 adapter->xdp_first_ring = new_channel_count;
2927                 adapter->xdp_num_queues = new_channel_count;
2928                 if (prev_channel_count > new_channel_count)
2929                         ena_xdp_exchange_program_rx_in_range(adapter,
2930                                                              NULL,
2931                                                              new_channel_count,
2932                                                              prev_channel_count);
2933                 else
2934                         ena_xdp_exchange_program_rx_in_range(adapter,
2935                                                              adapter->xdp_bpf_prog,
2936                                                              prev_channel_count,
2937                                                              new_channel_count);
2938         }
2939
2940         /* We need to destroy the rss table so that the indirection
2941          * table will be reinitialized by ena_up()
2942          */
2943         ena_com_rss_destroy(ena_dev);
2944         ena_init_io_rings(adapter,
2945                           0,
2946                           adapter->xdp_num_queues +
2947                           adapter->num_io_queues);
2948         return dev_was_up ? ena_open(adapter->netdev) : 0;
2949 }
2950
2951 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx,
2952                         struct sk_buff *skb,
2953                         bool disable_meta_caching)
2954 {
2955         u32 mss = skb_shinfo(skb)->gso_size;
2956         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2957         u8 l4_protocol = 0;
2958
2959         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2960                 ena_tx_ctx->l4_csum_enable = 1;
2961                 if (mss) {
2962                         ena_tx_ctx->tso_enable = 1;
2963                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2964                         ena_tx_ctx->l4_csum_partial = 0;
2965                 } else {
2966                         ena_tx_ctx->tso_enable = 0;
2967                         ena_meta->l4_hdr_len = 0;
2968                         ena_tx_ctx->l4_csum_partial = 1;
2969                 }
2970
2971                 switch (ip_hdr(skb)->version) {
2972                 case IPVERSION:
2973                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2974                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
2975                                 ena_tx_ctx->df = 1;
2976                         if (mss)
2977                                 ena_tx_ctx->l3_csum_enable = 1;
2978                         l4_protocol = ip_hdr(skb)->protocol;
2979                         break;
2980                 case 6:
2981                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2982                         l4_protocol = ipv6_hdr(skb)->nexthdr;
2983                         break;
2984                 default:
2985                         break;
2986                 }
2987
2988                 if (l4_protocol == IPPROTO_TCP)
2989                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2990                 else
2991                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2992
2993                 ena_meta->mss = mss;
2994                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
2995                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
2996                 ena_tx_ctx->meta_valid = 1;
2997         } else if (disable_meta_caching) {
2998                 memset(ena_meta, 0, sizeof(*ena_meta));
2999                 ena_tx_ctx->meta_valid = 1;
3000         } else {
3001                 ena_tx_ctx->meta_valid = 0;
3002         }
3003 }
3004
3005 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
3006                                        struct sk_buff *skb)
3007 {
3008         int num_frags, header_len, rc;
3009
3010         num_frags = skb_shinfo(skb)->nr_frags;
3011         header_len = skb_headlen(skb);
3012
3013         if (num_frags < tx_ring->sgl_size)
3014                 return 0;
3015
3016         if ((num_frags == tx_ring->sgl_size) &&
3017             (header_len < tx_ring->tx_max_header_size))
3018                 return 0;
3019
3020         ena_increase_stat(&tx_ring->tx_stats.linearize, 1, &tx_ring->syncp);
3021
3022         rc = skb_linearize(skb);
3023         if (unlikely(rc)) {
3024                 ena_increase_stat(&tx_ring->tx_stats.linearize_failed, 1,
3025                                   &tx_ring->syncp);
3026         }
3027
3028         return rc;
3029 }
3030
3031 static int ena_tx_map_skb(struct ena_ring *tx_ring,
3032                           struct ena_tx_buffer *tx_info,
3033                           struct sk_buff *skb,
3034                           void **push_hdr,
3035                           u16 *header_len)
3036 {
3037         struct ena_adapter *adapter = tx_ring->adapter;
3038         struct ena_com_buf *ena_buf;
3039         dma_addr_t dma;
3040         u32 skb_head_len, frag_len, last_frag;
3041         u16 push_len = 0;
3042         u16 delta = 0;
3043         int i = 0;
3044
3045         skb_head_len = skb_headlen(skb);
3046         tx_info->skb = skb;
3047         ena_buf = tx_info->bufs;
3048
3049         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3050                 /* When the device is LLQ mode, the driver will copy
3051                  * the header into the device memory space.
3052                  * the ena_com layer assume the header is in a linear
3053                  * memory space.
3054                  * This assumption might be wrong since part of the header
3055                  * can be in the fragmented buffers.
3056                  * Use skb_header_pointer to make sure the header is in a
3057                  * linear memory space.
3058                  */
3059
3060                 push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
3061                 *push_hdr = skb_header_pointer(skb, 0, push_len,
3062                                                tx_ring->push_buf_intermediate_buf);
3063                 *header_len = push_len;
3064                 if (unlikely(skb->data != *push_hdr)) {
3065                         ena_increase_stat(&tx_ring->tx_stats.llq_buffer_copy, 1,
3066                                           &tx_ring->syncp);
3067
3068                         delta = push_len - skb_head_len;
3069                 }
3070         } else {
3071                 *push_hdr = NULL;
3072                 *header_len = min_t(u32, skb_head_len,
3073                                     tx_ring->tx_max_header_size);
3074         }
3075
3076         netif_dbg(adapter, tx_queued, adapter->netdev,
3077                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
3078                   *push_hdr, push_len);
3079
3080         if (skb_head_len > push_len) {
3081                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
3082                                      skb_head_len - push_len, DMA_TO_DEVICE);
3083                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
3084                         goto error_report_dma_error;
3085
3086                 ena_buf->paddr = dma;
3087                 ena_buf->len = skb_head_len - push_len;
3088
3089                 ena_buf++;
3090                 tx_info->num_of_bufs++;
3091                 tx_info->map_linear_data = 1;
3092         } else {
3093                 tx_info->map_linear_data = 0;
3094         }
3095
3096         last_frag = skb_shinfo(skb)->nr_frags;
3097
3098         for (i = 0; i < last_frag; i++) {
3099                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3100
3101                 frag_len = skb_frag_size(frag);
3102
3103                 if (unlikely(delta >= frag_len)) {
3104                         delta -= frag_len;
3105                         continue;
3106                 }
3107
3108                 dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
3109                                        frag_len - delta, DMA_TO_DEVICE);
3110                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
3111                         goto error_report_dma_error;
3112
3113                 ena_buf->paddr = dma;
3114                 ena_buf->len = frag_len - delta;
3115                 ena_buf++;
3116                 tx_info->num_of_bufs++;
3117                 delta = 0;
3118         }
3119
3120         return 0;
3121
3122 error_report_dma_error:
3123         ena_increase_stat(&tx_ring->tx_stats.dma_mapping_err, 1,
3124                           &tx_ring->syncp);
3125         netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map skb\n");
3126
3127         tx_info->skb = NULL;
3128
3129         tx_info->num_of_bufs += i;
3130         ena_unmap_tx_buff(tx_ring, tx_info);
3131
3132         return -EINVAL;
3133 }
3134
3135 /* Called with netif_tx_lock. */
3136 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
3137 {
3138         struct ena_adapter *adapter = netdev_priv(dev);
3139         struct ena_tx_buffer *tx_info;
3140         struct ena_com_tx_ctx ena_tx_ctx;
3141         struct ena_ring *tx_ring;
3142         struct netdev_queue *txq;
3143         void *push_hdr;
3144         u16 next_to_use, req_id, header_len;
3145         int qid, rc;
3146
3147         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
3148         /*  Determine which tx ring we will be placed on */
3149         qid = skb_get_queue_mapping(skb);
3150         tx_ring = &adapter->tx_ring[qid];
3151         txq = netdev_get_tx_queue(dev, qid);
3152
3153         rc = ena_check_and_linearize_skb(tx_ring, skb);
3154         if (unlikely(rc))
3155                 goto error_drop_packet;
3156
3157         skb_tx_timestamp(skb);
3158
3159         next_to_use = tx_ring->next_to_use;
3160         req_id = tx_ring->free_ids[next_to_use];
3161         tx_info = &tx_ring->tx_buffer_info[req_id];
3162         tx_info->num_of_bufs = 0;
3163
3164         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
3165
3166         rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
3167         if (unlikely(rc))
3168                 goto error_drop_packet;
3169
3170         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
3171         ena_tx_ctx.ena_bufs = tx_info->bufs;
3172         ena_tx_ctx.push_header = push_hdr;
3173         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
3174         ena_tx_ctx.req_id = req_id;
3175         ena_tx_ctx.header_len = header_len;
3176
3177         /* set flags and meta data */
3178         ena_tx_csum(&ena_tx_ctx, skb, tx_ring->disable_meta_caching);
3179
3180         rc = ena_xmit_common(dev,
3181                              tx_ring,
3182                              tx_info,
3183                              &ena_tx_ctx,
3184                              next_to_use,
3185                              skb->len);
3186         if (rc)
3187                 goto error_unmap_dma;
3188
3189         netdev_tx_sent_queue(txq, skb->len);
3190
3191         /* stop the queue when no more space available, the packet can have up
3192          * to sgl_size + 2. one for the meta descriptor and one for header
3193          * (if the header is larger than tx_max_header_size).
3194          */
3195         if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3196                                                    tx_ring->sgl_size + 2))) {
3197                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3198                           __func__, qid);
3199
3200                 netif_tx_stop_queue(txq);
3201                 ena_increase_stat(&tx_ring->tx_stats.queue_stop, 1,
3202                                   &tx_ring->syncp);
3203
3204                 /* There is a rare condition where this function decide to
3205                  * stop the queue but meanwhile clean_tx_irq updates
3206                  * next_to_completion and terminates.
3207                  * The queue will remain stopped forever.
3208                  * To solve this issue add a mb() to make sure that
3209                  * netif_tx_stop_queue() write is vissible before checking if
3210                  * there is additional space in the queue.
3211                  */
3212                 smp_mb();
3213
3214                 if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3215                                                  ENA_TX_WAKEUP_THRESH)) {
3216                         netif_tx_wake_queue(txq);
3217                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
3218                                           &tx_ring->syncp);
3219                 }
3220         }
3221
3222         if (netif_xmit_stopped(txq) || !netdev_xmit_more())
3223                 /* trigger the dma engine. ena_ring_tx_doorbell()
3224                  * calls a memory barrier inside it.
3225                  */
3226                 ena_ring_tx_doorbell(tx_ring);
3227
3228         return NETDEV_TX_OK;
3229
3230 error_unmap_dma:
3231         ena_unmap_tx_buff(tx_ring, tx_info);
3232         tx_info->skb = NULL;
3233
3234 error_drop_packet:
3235         dev_kfree_skb(skb);
3236         return NETDEV_TX_OK;
3237 }
3238
3239 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3240                             struct net_device *sb_dev)
3241 {
3242         u16 qid;
3243         /* we suspect that this is good for in--kernel network services that
3244          * want to loop incoming skb rx to tx in normal user generated traffic,
3245          * most probably we will not get to this
3246          */
3247         if (skb_rx_queue_recorded(skb))
3248                 qid = skb_get_rx_queue(skb);
3249         else
3250                 qid = netdev_pick_tx(dev, skb, NULL);
3251
3252         return qid;
3253 }
3254
3255 static void ena_config_host_info(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3256 {
3257         struct device *dev = &pdev->dev;
3258         struct ena_admin_host_info *host_info;
3259         int rc;
3260
3261         /* Allocate only the host info */
3262         rc = ena_com_allocate_host_info(ena_dev);
3263         if (rc) {
3264                 dev_err(dev, "Cannot allocate host info\n");
3265                 return;
3266         }
3267
3268         host_info = ena_dev->host_attr.host_info;
3269
3270         host_info->bdf = pci_dev_id(pdev);
3271         host_info->os_type = ENA_ADMIN_OS_LINUX;
3272         host_info->kernel_ver = LINUX_VERSION_CODE;
3273         strscpy(host_info->kernel_ver_str, utsname()->version,
3274                 sizeof(host_info->kernel_ver_str) - 1);
3275         host_info->os_dist = 0;
3276         strncpy(host_info->os_dist_str, utsname()->release,
3277                 sizeof(host_info->os_dist_str) - 1);
3278         host_info->driver_version =
3279                 (DRV_MODULE_GEN_MAJOR) |
3280                 (DRV_MODULE_GEN_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3281                 (DRV_MODULE_GEN_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3282                 ("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3283         host_info->num_cpus = num_online_cpus();
3284
3285         host_info->driver_supported_features =
3286                 ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK |
3287                 ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK |
3288                 ENA_ADMIN_HOST_INFO_RX_BUF_MIRRORING_MASK |
3289                 ENA_ADMIN_HOST_INFO_RSS_CONFIGURABLE_FUNCTION_KEY_MASK |
3290                 ENA_ADMIN_HOST_INFO_RX_PAGE_REUSE_MASK;
3291
3292         rc = ena_com_set_host_attributes(ena_dev);
3293         if (rc) {
3294                 if (rc == -EOPNOTSUPP)
3295                         dev_warn(dev, "Cannot set host attributes\n");
3296                 else
3297                         dev_err(dev, "Cannot set host attributes\n");
3298
3299                 goto err;
3300         }
3301
3302         return;
3303
3304 err:
3305         ena_com_delete_host_info(ena_dev);
3306 }
3307
3308 static void ena_config_debug_area(struct ena_adapter *adapter)
3309 {
3310         u32 debug_area_size;
3311         int rc, ss_count;
3312
3313         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3314         if (ss_count <= 0) {
3315                 netif_err(adapter, drv, adapter->netdev,
3316                           "SS count is negative\n");
3317                 return;
3318         }
3319
3320         /* allocate 32 bytes for each string and 64bit for the value */
3321         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3322
3323         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3324         if (rc) {
3325                 netif_err(adapter, drv, adapter->netdev,
3326                           "Cannot allocate debug area\n");
3327                 return;
3328         }
3329
3330         rc = ena_com_set_host_attributes(adapter->ena_dev);
3331         if (rc) {
3332                 if (rc == -EOPNOTSUPP)
3333                         netif_warn(adapter, drv, adapter->netdev,
3334                                    "Cannot set host attributes\n");
3335                 else
3336                         netif_err(adapter, drv, adapter->netdev,
3337                                   "Cannot set host attributes\n");
3338                 goto err;
3339         }
3340
3341         return;
3342 err:
3343         ena_com_delete_debug_area(adapter->ena_dev);
3344 }
3345
3346 int ena_update_hw_stats(struct ena_adapter *adapter)
3347 {
3348         int rc;
3349
3350         rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_stats);
3351         if (rc) {
3352                 netdev_err(adapter->netdev, "Failed to get ENI stats\n");
3353                 return rc;
3354         }
3355
3356         return 0;
3357 }
3358
3359 static void ena_get_stats64(struct net_device *netdev,
3360                             struct rtnl_link_stats64 *stats)
3361 {
3362         struct ena_adapter *adapter = netdev_priv(netdev);
3363         struct ena_ring *rx_ring, *tx_ring;
3364         unsigned int start;
3365         u64 rx_drops;
3366         u64 tx_drops;
3367         int i;
3368
3369         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3370                 return;
3371
3372         for (i = 0; i < adapter->num_io_queues; i++) {
3373                 u64 bytes, packets;
3374
3375                 tx_ring = &adapter->tx_ring[i];
3376
3377                 do {
3378                         start = u64_stats_fetch_begin(&tx_ring->syncp);
3379                         packets = tx_ring->tx_stats.cnt;
3380                         bytes = tx_ring->tx_stats.bytes;
3381                 } while (u64_stats_fetch_retry(&tx_ring->syncp, start));
3382
3383                 stats->tx_packets += packets;
3384                 stats->tx_bytes += bytes;
3385
3386                 rx_ring = &adapter->rx_ring[i];
3387
3388                 do {
3389                         start = u64_stats_fetch_begin(&rx_ring->syncp);
3390                         packets = rx_ring->rx_stats.cnt;
3391                         bytes = rx_ring->rx_stats.bytes;
3392                 } while (u64_stats_fetch_retry(&rx_ring->syncp, start));
3393
3394                 stats->rx_packets += packets;
3395                 stats->rx_bytes += bytes;
3396         }
3397
3398         do {
3399                 start = u64_stats_fetch_begin(&adapter->syncp);
3400                 rx_drops = adapter->dev_stats.rx_drops;
3401                 tx_drops = adapter->dev_stats.tx_drops;
3402         } while (u64_stats_fetch_retry(&adapter->syncp, start));
3403
3404         stats->rx_dropped = rx_drops;
3405         stats->tx_dropped = tx_drops;
3406
3407         stats->multicast = 0;
3408         stats->collisions = 0;
3409
3410         stats->rx_length_errors = 0;
3411         stats->rx_crc_errors = 0;
3412         stats->rx_frame_errors = 0;
3413         stats->rx_fifo_errors = 0;
3414         stats->rx_missed_errors = 0;
3415         stats->tx_window_errors = 0;
3416
3417         stats->rx_errors = 0;
3418         stats->tx_errors = 0;
3419 }
3420
3421 static const struct net_device_ops ena_netdev_ops = {
3422         .ndo_open               = ena_open,
3423         .ndo_stop               = ena_close,
3424         .ndo_start_xmit         = ena_start_xmit,
3425         .ndo_select_queue       = ena_select_queue,
3426         .ndo_get_stats64        = ena_get_stats64,
3427         .ndo_tx_timeout         = ena_tx_timeout,
3428         .ndo_change_mtu         = ena_change_mtu,
3429         .ndo_set_mac_address    = NULL,
3430         .ndo_validate_addr      = eth_validate_addr,
3431         .ndo_bpf                = ena_xdp,
3432         .ndo_xdp_xmit           = ena_xdp_xmit,
3433 };
3434
3435 static void ena_calc_io_queue_size(struct ena_adapter *adapter,
3436                                    struct ena_com_dev_get_features_ctx *get_feat_ctx)
3437 {
3438         struct ena_admin_feature_llq_desc *llq = &get_feat_ctx->llq;
3439         struct ena_com_dev *ena_dev = adapter->ena_dev;
3440         u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
3441         u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
3442         u32 max_tx_queue_size;
3443         u32 max_rx_queue_size;
3444
3445         /* If this function is called after driver load, the ring sizes have already
3446          * been configured. Take it into account when recalculating ring size.
3447          */
3448         if (adapter->tx_ring->ring_size)
3449                 tx_queue_size = adapter->tx_ring->ring_size;
3450
3451         if (adapter->rx_ring->ring_size)
3452                 rx_queue_size = adapter->rx_ring->ring_size;
3453
3454         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3455                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3456                         &get_feat_ctx->max_queue_ext.max_queue_ext;
3457                 max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
3458                                           max_queue_ext->max_rx_sq_depth);
3459                 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
3460
3461                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3462                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
3463                                                   llq->max_llq_depth);
3464                 else
3465                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
3466                                                   max_queue_ext->max_tx_sq_depth);
3467
3468                 adapter->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3469                                                  max_queue_ext->max_per_packet_tx_descs);
3470                 adapter->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3471                                                  max_queue_ext->max_per_packet_rx_descs);
3472         } else {
3473                 struct ena_admin_queue_feature_desc *max_queues =
3474                         &get_feat_ctx->max_queues;
3475                 max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
3476                                           max_queues->max_sq_depth);
3477                 max_tx_queue_size = max_queues->max_cq_depth;
3478
3479                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3480                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
3481                                                   llq->max_llq_depth);
3482                 else
3483                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
3484                                                   max_queues->max_sq_depth);
3485
3486                 adapter->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3487                                                  max_queues->max_packet_tx_descs);
3488                 adapter->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3489                                                  max_queues->max_packet_rx_descs);
3490         }
3491
3492         max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
3493         max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
3494
3495         /* When forcing large headers, we multiply the entry size by 2, and therefore divide
3496          * the queue size by 2, leaving the amount of memory used by the queues unchanged.
3497          */
3498         if (adapter->large_llq_header_enabled) {
3499                 if ((llq->entry_size_ctrl_supported & ENA_ADMIN_LIST_ENTRY_SIZE_256B) &&
3500                     ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3501                         max_tx_queue_size /= 2;
3502                         dev_info(&adapter->pdev->dev,
3503                                  "Forcing large headers and decreasing maximum TX queue size to %d\n",
3504                                  max_tx_queue_size);
3505                 } else {
3506                         dev_err(&adapter->pdev->dev,
3507                                 "Forcing large headers failed: LLQ is disabled or device does not support large headers\n");
3508
3509                         adapter->large_llq_header_enabled = false;
3510                 }
3511         }
3512
3513         tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
3514                                   max_tx_queue_size);
3515         rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
3516                                   max_rx_queue_size);
3517
3518         tx_queue_size = rounddown_pow_of_two(tx_queue_size);
3519         rx_queue_size = rounddown_pow_of_two(rx_queue_size);
3520
3521         adapter->max_tx_ring_size  = max_tx_queue_size;
3522         adapter->max_rx_ring_size = max_rx_queue_size;
3523         adapter->requested_tx_ring_size = tx_queue_size;
3524         adapter->requested_rx_ring_size = rx_queue_size;
3525 }
3526
3527 static int ena_device_validate_params(struct ena_adapter *adapter,
3528                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
3529 {
3530         struct net_device *netdev = adapter->netdev;
3531         int rc;
3532
3533         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3534                               adapter->mac_addr);
3535         if (!rc) {
3536                 netif_err(adapter, drv, netdev,
3537                           "Error, mac address are different\n");
3538                 return -EINVAL;
3539         }
3540
3541         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3542                 netif_err(adapter, drv, netdev,
3543                           "Error, device max mtu is smaller than netdev MTU\n");
3544                 return -EINVAL;
3545         }
3546
3547         return 0;
3548 }
3549
3550 static void set_default_llq_configurations(struct ena_adapter *adapter,
3551                                            struct ena_llq_configurations *llq_config,
3552                                            struct ena_admin_feature_llq_desc *llq)
3553 {
3554         struct ena_com_dev *ena_dev = adapter->ena_dev;
3555
3556         llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
3557         llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
3558         llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
3559
3560         adapter->large_llq_header_supported =
3561                 !!(ena_dev->supported_features & BIT(ENA_ADMIN_LLQ));
3562         adapter->large_llq_header_supported &=
3563                 !!(llq->entry_size_ctrl_supported &
3564                         ENA_ADMIN_LIST_ENTRY_SIZE_256B);
3565
3566         if ((llq->entry_size_ctrl_supported & ENA_ADMIN_LIST_ENTRY_SIZE_256B) &&
3567             adapter->large_llq_header_enabled) {
3568                 llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_256B;
3569                 llq_config->llq_ring_entry_size_value = 256;
3570         } else {
3571                 llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
3572                 llq_config->llq_ring_entry_size_value = 128;
3573         }
3574 }
3575
3576 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3577                                            struct ena_com_dev *ena_dev,
3578                                            struct ena_admin_feature_llq_desc *llq,
3579                                            struct ena_llq_configurations *llq_default_configurations)
3580 {
3581         int rc;
3582         u32 llq_feature_mask;
3583
3584         llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3585         if (!(ena_dev->supported_features & llq_feature_mask)) {
3586                 dev_warn(&pdev->dev,
3587                         "LLQ is not supported Fallback to host mode policy.\n");
3588                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3589                 return 0;
3590         }
3591
3592         if (!ena_dev->mem_bar) {
3593                 netdev_err(ena_dev->net_device,
3594                            "LLQ is advertised as supported but device doesn't expose mem bar\n");
3595                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3596                 return 0;
3597         }
3598
3599         rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3600         if (unlikely(rc)) {
3601                 dev_err(&pdev->dev,
3602                         "Failed to configure the device mode.  Fallback to host mode policy.\n");
3603                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3604         }
3605
3606         return 0;
3607 }
3608
3609 static int ena_map_llq_mem_bar(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
3610                                int bars)
3611 {
3612         bool has_mem_bar = !!(bars & BIT(ENA_MEM_BAR));
3613
3614         if (!has_mem_bar)
3615                 return 0;
3616
3617         ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3618                                            pci_resource_start(pdev, ENA_MEM_BAR),
3619                                            pci_resource_len(pdev, ENA_MEM_BAR));
3620
3621         if (!ena_dev->mem_bar)
3622                 return -EFAULT;
3623
3624         return 0;
3625 }
3626
3627 static int ena_device_init(struct ena_adapter *adapter, struct pci_dev *pdev,
3628                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
3629                            bool *wd_state)
3630 {
3631         struct ena_com_dev *ena_dev = adapter->ena_dev;
3632         struct ena_llq_configurations llq_config;
3633         struct device *dev = &pdev->dev;
3634         bool readless_supported;
3635         u32 aenq_groups;
3636         int dma_width;
3637         int rc;
3638
3639         rc = ena_com_mmio_reg_read_request_init(ena_dev);
3640         if (rc) {
3641                 dev_err(dev, "Failed to init mmio read less\n");
3642                 return rc;
3643         }
3644
3645         /* The PCIe configuration space revision id indicate if mmio reg
3646          * read is disabled
3647          */
3648         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3649         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3650
3651         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3652         if (rc) {
3653                 dev_err(dev, "Can not reset device\n");
3654                 goto err_mmio_read_less;
3655         }
3656
3657         rc = ena_com_validate_version(ena_dev);
3658         if (rc) {
3659                 dev_err(dev, "Device version is too low\n");
3660                 goto err_mmio_read_less;
3661         }
3662
3663         dma_width = ena_com_get_dma_width(ena_dev);
3664         if (dma_width < 0) {
3665                 dev_err(dev, "Invalid dma width value %d", dma_width);
3666                 rc = dma_width;
3667                 goto err_mmio_read_less;
3668         }
3669
3670         rc = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(dma_width));
3671         if (rc) {
3672                 dev_err(dev, "dma_set_mask_and_coherent failed %d\n", rc);
3673                 goto err_mmio_read_less;
3674         }
3675
3676         /* ENA admin level init */
3677         rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3678         if (rc) {
3679                 dev_err(dev,
3680                         "Can not initialize ena admin queue with device\n");
3681                 goto err_mmio_read_less;
3682         }
3683
3684         /* To enable the msix interrupts the driver needs to know the number
3685          * of queues. So the driver uses polling mode to retrieve this
3686          * information
3687          */
3688         ena_com_set_admin_polling_mode(ena_dev, true);
3689
3690         ena_config_host_info(ena_dev, pdev);
3691
3692         /* Get Device Attributes*/
3693         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3694         if (rc) {
3695                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3696                 goto err_admin_init;
3697         }
3698
3699         /* Try to turn all the available aenq groups */
3700         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3701                 BIT(ENA_ADMIN_FATAL_ERROR) |
3702                 BIT(ENA_ADMIN_WARNING) |
3703                 BIT(ENA_ADMIN_NOTIFICATION) |
3704                 BIT(ENA_ADMIN_KEEP_ALIVE);
3705
3706         aenq_groups &= get_feat_ctx->aenq.supported_groups;
3707
3708         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3709         if (rc) {
3710                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3711                 goto err_admin_init;
3712         }
3713
3714         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3715
3716         set_default_llq_configurations(adapter, &llq_config, &get_feat_ctx->llq);
3717
3718         rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx->llq,
3719                                              &llq_config);
3720         if (rc) {
3721                 dev_err(dev, "ENA device init failed\n");
3722                 goto err_admin_init;
3723         }
3724
3725         ena_calc_io_queue_size(adapter, get_feat_ctx);
3726
3727         return 0;
3728
3729 err_admin_init:
3730         ena_com_delete_host_info(ena_dev);
3731         ena_com_admin_destroy(ena_dev);
3732 err_mmio_read_less:
3733         ena_com_mmio_reg_read_request_destroy(ena_dev);
3734
3735         return rc;
3736 }
3737
3738 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3739 {
3740         struct ena_com_dev *ena_dev = adapter->ena_dev;
3741         struct device *dev = &adapter->pdev->dev;
3742         int rc;
3743
3744         rc = ena_enable_msix(adapter);
3745         if (rc) {
3746                 dev_err(dev, "Can not reserve msix vectors\n");
3747                 return rc;
3748         }
3749
3750         ena_setup_mgmnt_intr(adapter);
3751
3752         rc = ena_request_mgmnt_irq(adapter);
3753         if (rc) {
3754                 dev_err(dev, "Can not setup management interrupts\n");
3755                 goto err_disable_msix;
3756         }
3757
3758         ena_com_set_admin_polling_mode(ena_dev, false);
3759
3760         ena_com_admin_aenq_enable(ena_dev);
3761
3762         return 0;
3763
3764 err_disable_msix:
3765         ena_disable_msix(adapter);
3766
3767         return rc;
3768 }
3769
3770 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3771 {
3772         struct net_device *netdev = adapter->netdev;
3773         struct ena_com_dev *ena_dev = adapter->ena_dev;
3774         bool dev_up;
3775
3776         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3777                 return;
3778
3779         netif_carrier_off(netdev);
3780
3781         del_timer_sync(&adapter->timer_service);
3782
3783         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3784         adapter->dev_up_before_reset = dev_up;
3785         if (!graceful)
3786                 ena_com_set_admin_running_state(ena_dev, false);
3787
3788         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3789                 ena_down(adapter);
3790
3791         /* Stop the device from sending AENQ events (in case reset flag is set
3792          *  and device is up, ena_down() already reset the device.
3793          */
3794         if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3795                 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3796
3797         ena_free_mgmnt_irq(adapter);
3798
3799         ena_disable_msix(adapter);
3800
3801         ena_com_abort_admin_commands(ena_dev);
3802
3803         ena_com_wait_for_abort_completion(ena_dev);
3804
3805         ena_com_admin_destroy(ena_dev);
3806
3807         ena_com_mmio_reg_read_request_destroy(ena_dev);
3808
3809         /* return reset reason to default value */
3810         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3811
3812         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3813         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3814 }
3815
3816 static int ena_restore_device(struct ena_adapter *adapter)
3817 {
3818         struct ena_com_dev_get_features_ctx get_feat_ctx;
3819         struct ena_com_dev *ena_dev = adapter->ena_dev;
3820         struct pci_dev *pdev = adapter->pdev;
3821         struct ena_ring *txr;
3822         int rc, count, i;
3823         bool wd_state;
3824
3825         set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3826         rc = ena_device_init(adapter, adapter->pdev, &get_feat_ctx, &wd_state);
3827         if (rc) {
3828                 dev_err(&pdev->dev, "Can not initialize device\n");
3829                 goto err;
3830         }
3831         adapter->wd_state = wd_state;
3832
3833         count =  adapter->xdp_num_queues + adapter->num_io_queues;
3834         for (i = 0 ; i < count; i++) {
3835                 txr = &adapter->tx_ring[i];
3836                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
3837                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
3838         }
3839
3840         rc = ena_device_validate_params(adapter, &get_feat_ctx);
3841         if (rc) {
3842                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
3843                 goto err_device_destroy;
3844         }
3845
3846         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3847         if (rc) {
3848                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
3849                 goto err_device_destroy;
3850         }
3851         /* If the interface was up before the reset bring it up */
3852         if (adapter->dev_up_before_reset) {
3853                 rc = ena_up(adapter);
3854                 if (rc) {
3855                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
3856                         goto err_disable_msix;
3857                 }
3858         }
3859
3860         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3861
3862         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3863         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3864                 netif_carrier_on(adapter->netdev);
3865
3866         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3867         adapter->last_keep_alive_jiffies = jiffies;
3868
3869         return rc;
3870 err_disable_msix:
3871         ena_free_mgmnt_irq(adapter);
3872         ena_disable_msix(adapter);
3873 err_device_destroy:
3874         ena_com_abort_admin_commands(ena_dev);
3875         ena_com_wait_for_abort_completion(ena_dev);
3876         ena_com_admin_destroy(ena_dev);
3877         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3878         ena_com_mmio_reg_read_request_destroy(ena_dev);
3879 err:
3880         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3881         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3882         dev_err(&pdev->dev,
3883                 "Reset attempt failed. Can not reset the device\n");
3884
3885         return rc;
3886 }
3887
3888 static void ena_fw_reset_device(struct work_struct *work)
3889 {
3890         struct ena_adapter *adapter =
3891                 container_of(work, struct ena_adapter, reset_task);
3892
3893         rtnl_lock();
3894
3895         if (likely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3896                 ena_destroy_device(adapter, false);
3897                 ena_restore_device(adapter);
3898
3899                 dev_err(&adapter->pdev->dev, "Device reset completed successfully\n");
3900         }
3901
3902         rtnl_unlock();
3903 }
3904
3905 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3906                                         struct ena_ring *rx_ring)
3907 {
3908         struct ena_napi *ena_napi = container_of(rx_ring->napi, struct ena_napi, napi);
3909
3910         if (likely(READ_ONCE(ena_napi->first_interrupt)))
3911                 return 0;
3912
3913         if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3914                 return 0;
3915
3916         rx_ring->no_interrupt_event_cnt++;
3917
3918         if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3919                 netif_err(adapter, rx_err, adapter->netdev,
3920                           "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3921                           rx_ring->qid);
3922
3923                 ena_reset_device(adapter, ENA_REGS_RESET_MISS_INTERRUPT);
3924                 return -EIO;
3925         }
3926
3927         return 0;
3928 }
3929
3930 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3931                                           struct ena_ring *tx_ring)
3932 {
3933         struct ena_napi *ena_napi = container_of(tx_ring->napi, struct ena_napi, napi);
3934         unsigned int time_since_last_napi;
3935         unsigned int missing_tx_comp_to;
3936         bool is_tx_comp_time_expired;
3937         struct ena_tx_buffer *tx_buf;
3938         unsigned long last_jiffies;
3939         u32 missed_tx = 0;
3940         int i, rc = 0;
3941
3942         for (i = 0; i < tx_ring->ring_size; i++) {
3943                 tx_buf = &tx_ring->tx_buffer_info[i];
3944                 last_jiffies = tx_buf->last_jiffies;
3945
3946                 if (last_jiffies == 0)
3947                         /* no pending Tx at this location */
3948                         continue;
3949
3950                 is_tx_comp_time_expired = time_is_before_jiffies(last_jiffies +
3951                          2 * adapter->missing_tx_completion_to);
3952
3953                 if (unlikely(!READ_ONCE(ena_napi->first_interrupt) && is_tx_comp_time_expired)) {
3954                         /* If after graceful period interrupt is still not
3955                          * received, we schedule a reset
3956                          */
3957                         netif_err(adapter, tx_err, adapter->netdev,
3958                                   "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3959                                   tx_ring->qid);
3960                         ena_reset_device(adapter, ENA_REGS_RESET_MISS_INTERRUPT);
3961                         return -EIO;
3962                 }
3963
3964                 is_tx_comp_time_expired = time_is_before_jiffies(last_jiffies +
3965                         adapter->missing_tx_completion_to);
3966
3967                 if (unlikely(is_tx_comp_time_expired)) {
3968                         if (!tx_buf->print_once) {
3969                                 time_since_last_napi = jiffies_to_usecs(jiffies - tx_ring->tx_stats.last_napi_jiffies);
3970                                 missing_tx_comp_to = jiffies_to_msecs(adapter->missing_tx_completion_to);
3971                                 netif_notice(adapter, tx_err, adapter->netdev,
3972                                              "Found a Tx that wasn't completed on time, qid %d, index %d. %u usecs have passed since last napi execution. Missing Tx timeout value %u msecs\n",
3973                                              tx_ring->qid, i, time_since_last_napi, missing_tx_comp_to);
3974                         }
3975
3976                         tx_buf->print_once = 1;
3977                         missed_tx++;
3978                 }
3979         }
3980
3981         if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3982                 netif_err(adapter, tx_err, adapter->netdev,
3983                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3984                           missed_tx,
3985                           adapter->missing_tx_completion_threshold);
3986                 ena_reset_device(adapter, ENA_REGS_RESET_MISS_TX_CMPL);
3987                 rc = -EIO;
3988         }
3989
3990         ena_increase_stat(&tx_ring->tx_stats.missed_tx, missed_tx,
3991                           &tx_ring->syncp);
3992
3993         return rc;
3994 }
3995
3996 static void check_for_missing_completions(struct ena_adapter *adapter)
3997 {
3998         struct ena_ring *tx_ring;
3999         struct ena_ring *rx_ring;
4000         int i, budget, rc;
4001         int io_queue_count;
4002
4003         io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
4004         /* Make sure the driver doesn't turn the device in other process */
4005         smp_rmb();
4006
4007         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
4008                 return;
4009
4010         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
4011                 return;
4012
4013         if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
4014                 return;
4015
4016         budget = ENA_MONITORED_TX_QUEUES;
4017
4018         for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
4019                 tx_ring = &adapter->tx_ring[i];
4020                 rx_ring = &adapter->rx_ring[i];
4021
4022                 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
4023                 if (unlikely(rc))
4024                         return;
4025
4026                 rc =  !ENA_IS_XDP_INDEX(adapter, i) ?
4027                         check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
4028                 if (unlikely(rc))
4029                         return;
4030
4031                 budget--;
4032                 if (!budget)
4033                         break;
4034         }
4035
4036         adapter->last_monitored_tx_qid = i % io_queue_count;
4037 }
4038
4039 /* trigger napi schedule after 2 consecutive detections */
4040 #define EMPTY_RX_REFILL 2
4041 /* For the rare case where the device runs out of Rx descriptors and the
4042  * napi handler failed to refill new Rx descriptors (due to a lack of memory
4043  * for example).
4044  * This case will lead to a deadlock:
4045  * The device won't send interrupts since all the new Rx packets will be dropped
4046  * The napi handler won't allocate new Rx descriptors so the device will be
4047  * able to send new packets.
4048  *
4049  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
4050  * It is recommended to have at least 512MB, with a minimum of 128MB for
4051  * constrained environment).
4052  *
4053  * When such a situation is detected - Reschedule napi
4054  */
4055 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
4056 {
4057         struct ena_ring *rx_ring;
4058         int i, refill_required;
4059
4060         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
4061                 return;
4062
4063         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
4064                 return;
4065
4066         for (i = 0; i < adapter->num_io_queues; i++) {
4067                 rx_ring = &adapter->rx_ring[i];
4068
4069                 refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
4070                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
4071                         rx_ring->empty_rx_queue++;
4072
4073                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
4074                                 ena_increase_stat(&rx_ring->rx_stats.empty_rx_ring, 1,
4075                                                   &rx_ring->syncp);
4076
4077                                 netif_err(adapter, drv, adapter->netdev,
4078                                           "Trigger refill for ring %d\n", i);
4079
4080                                 napi_schedule(rx_ring->napi);
4081                                 rx_ring->empty_rx_queue = 0;
4082                         }
4083                 } else {
4084                         rx_ring->empty_rx_queue = 0;
4085                 }
4086         }
4087 }
4088
4089 /* Check for keep alive expiration */
4090 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
4091 {
4092         unsigned long keep_alive_expired;
4093
4094         if (!adapter->wd_state)
4095                 return;
4096
4097         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
4098                 return;
4099
4100         keep_alive_expired = adapter->last_keep_alive_jiffies +
4101                              adapter->keep_alive_timeout;
4102         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
4103                 netif_err(adapter, drv, adapter->netdev,
4104                           "Keep alive watchdog timeout.\n");
4105                 ena_increase_stat(&adapter->dev_stats.wd_expired, 1,
4106                                   &adapter->syncp);
4107                 ena_reset_device(adapter, ENA_REGS_RESET_KEEP_ALIVE_TO);
4108         }
4109 }
4110
4111 static void check_for_admin_com_state(struct ena_adapter *adapter)
4112 {
4113         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
4114                 netif_err(adapter, drv, adapter->netdev,
4115                           "ENA admin queue is not in running state!\n");
4116                 ena_increase_stat(&adapter->dev_stats.admin_q_pause, 1,
4117                                   &adapter->syncp);
4118                 ena_reset_device(adapter, ENA_REGS_RESET_ADMIN_TO);
4119         }
4120 }
4121
4122 static void ena_update_hints(struct ena_adapter *adapter,
4123                              struct ena_admin_ena_hw_hints *hints)
4124 {
4125         struct net_device *netdev = adapter->netdev;
4126
4127         if (hints->admin_completion_tx_timeout)
4128                 adapter->ena_dev->admin_queue.completion_timeout =
4129                         hints->admin_completion_tx_timeout * 1000;
4130
4131         if (hints->mmio_read_timeout)
4132                 /* convert to usec */
4133                 adapter->ena_dev->mmio_read.reg_read_to =
4134                         hints->mmio_read_timeout * 1000;
4135
4136         if (hints->missed_tx_completion_count_threshold_to_reset)
4137                 adapter->missing_tx_completion_threshold =
4138                         hints->missed_tx_completion_count_threshold_to_reset;
4139
4140         if (hints->missing_tx_completion_timeout) {
4141                 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
4142                         adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
4143                 else
4144                         adapter->missing_tx_completion_to =
4145                                 msecs_to_jiffies(hints->missing_tx_completion_timeout);
4146         }
4147
4148         if (hints->netdev_wd_timeout)
4149                 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
4150
4151         if (hints->driver_watchdog_timeout) {
4152                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
4153                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
4154                 else
4155                         adapter->keep_alive_timeout =
4156                                 msecs_to_jiffies(hints->driver_watchdog_timeout);
4157         }
4158 }
4159
4160 static void ena_update_host_info(struct ena_admin_host_info *host_info,
4161                                  struct net_device *netdev)
4162 {
4163         host_info->supported_network_features[0] =
4164                 netdev->features & GENMASK_ULL(31, 0);
4165         host_info->supported_network_features[1] =
4166                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
4167 }
4168
4169 static void ena_timer_service(struct timer_list *t)
4170 {
4171         struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
4172         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
4173         struct ena_admin_host_info *host_info =
4174                 adapter->ena_dev->host_attr.host_info;
4175
4176         check_for_missing_keep_alive(adapter);
4177
4178         check_for_admin_com_state(adapter);
4179
4180         check_for_missing_completions(adapter);
4181
4182         check_for_empty_rx_ring(adapter);
4183
4184         if (debug_area)
4185                 ena_dump_stats_to_buf(adapter, debug_area);
4186
4187         if (host_info)
4188                 ena_update_host_info(host_info, adapter->netdev);
4189
4190         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4191                 netif_err(adapter, drv, adapter->netdev,
4192                           "Trigger reset is on\n");
4193                 ena_dump_stats_to_dmesg(adapter);
4194                 queue_work(ena_wq, &adapter->reset_task);
4195                 return;
4196         }
4197
4198         /* Reset the timer */
4199         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4200 }
4201
4202 static u32 ena_calc_max_io_queue_num(struct pci_dev *pdev,
4203                                      struct ena_com_dev *ena_dev,
4204                                      struct ena_com_dev_get_features_ctx *get_feat_ctx)
4205 {
4206         u32 io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
4207
4208         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4209                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4210                         &get_feat_ctx->max_queue_ext.max_queue_ext;
4211                 io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
4212                                   max_queue_ext->max_rx_cq_num);
4213
4214                 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
4215                 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
4216         } else {
4217                 struct ena_admin_queue_feature_desc *max_queues =
4218                         &get_feat_ctx->max_queues;
4219                 io_tx_sq_num = max_queues->max_sq_num;
4220                 io_tx_cq_num = max_queues->max_cq_num;
4221                 io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
4222         }
4223
4224         /* In case of LLQ use the llq fields for the tx SQ/CQ */
4225         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4226                 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
4227
4228         max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
4229         max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
4230         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
4231         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
4232         /* 1 IRQ for mgmnt and 1 IRQs for each IO direction */
4233         max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
4234
4235         return max_num_io_queues;
4236 }
4237
4238 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
4239                                  struct net_device *netdev)
4240 {
4241         netdev_features_t dev_features = 0;
4242
4243         /* Set offload features */
4244         if (feat->offload.tx &
4245                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
4246                 dev_features |= NETIF_F_IP_CSUM;
4247
4248         if (feat->offload.tx &
4249                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
4250                 dev_features |= NETIF_F_IPV6_CSUM;
4251
4252         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
4253                 dev_features |= NETIF_F_TSO;
4254
4255         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
4256                 dev_features |= NETIF_F_TSO6;
4257
4258         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
4259                 dev_features |= NETIF_F_TSO_ECN;
4260
4261         if (feat->offload.rx_supported &
4262                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
4263                 dev_features |= NETIF_F_RXCSUM;
4264
4265         if (feat->offload.rx_supported &
4266                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
4267                 dev_features |= NETIF_F_RXCSUM;
4268
4269         netdev->features =
4270                 dev_features |
4271                 NETIF_F_SG |
4272                 NETIF_F_RXHASH |
4273                 NETIF_F_HIGHDMA;
4274
4275         netdev->hw_features |= netdev->features;
4276         netdev->vlan_features |= netdev->features;
4277 }
4278
4279 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
4280                                      struct ena_com_dev_get_features_ctx *feat)
4281 {
4282         struct net_device *netdev = adapter->netdev;
4283
4284         /* Copy mac address */
4285         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
4286                 eth_hw_addr_random(netdev);
4287                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
4288         } else {
4289                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
4290                 eth_hw_addr_set(netdev, adapter->mac_addr);
4291         }
4292
4293         /* Set offload features */
4294         ena_set_dev_offloads(feat, netdev);
4295
4296         adapter->max_mtu = feat->dev_attr.max_mtu;
4297         netdev->max_mtu = adapter->max_mtu;
4298         netdev->min_mtu = ENA_MIN_MTU;
4299 }
4300
4301 static int ena_rss_init_default(struct ena_adapter *adapter)
4302 {
4303         struct ena_com_dev *ena_dev = adapter->ena_dev;
4304         struct device *dev = &adapter->pdev->dev;
4305         int rc, i;
4306         u32 val;
4307
4308         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
4309         if (unlikely(rc)) {
4310                 dev_err(dev, "Cannot init indirect table\n");
4311                 goto err_rss_init;
4312         }
4313
4314         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
4315                 val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
4316                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
4317                                                        ENA_IO_RXQ_IDX(val));
4318                 if (unlikely(rc)) {
4319                         dev_err(dev, "Cannot fill indirect table\n");
4320                         goto err_fill_indir;
4321                 }
4322         }
4323
4324         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ, NULL,
4325                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4326         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4327                 dev_err(dev, "Cannot fill hash function\n");
4328                 goto err_fill_indir;
4329         }
4330
4331         rc = ena_com_set_default_hash_ctrl(ena_dev);
4332         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4333                 dev_err(dev, "Cannot fill hash control\n");
4334                 goto err_fill_indir;
4335         }
4336
4337         return 0;
4338
4339 err_fill_indir:
4340         ena_com_rss_destroy(ena_dev);
4341 err_rss_init:
4342
4343         return rc;
4344 }
4345
4346 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4347 {
4348         int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4349
4350         pci_release_selected_regions(pdev, release_bars);
4351 }
4352
4353 /* ena_probe - Device Initialization Routine
4354  * @pdev: PCI device information struct
4355  * @ent: entry in ena_pci_tbl
4356  *
4357  * Returns 0 on success, negative on failure
4358  *
4359  * ena_probe initializes an adapter identified by a pci_dev structure.
4360  * The OS initialization, configuring of the adapter private structure,
4361  * and a hardware reset occur.
4362  */
4363 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4364 {
4365         struct ena_com_dev_get_features_ctx get_feat_ctx;
4366         struct ena_com_dev *ena_dev = NULL;
4367         struct ena_adapter *adapter;
4368         struct net_device *netdev;
4369         static int adapters_found;
4370         u32 max_num_io_queues;
4371         bool wd_state;
4372         int bars, rc;
4373
4374         dev_dbg(&pdev->dev, "%s\n", __func__);
4375
4376         rc = pci_enable_device_mem(pdev);
4377         if (rc) {
4378                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4379                 return rc;
4380         }
4381
4382         rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
4383         if (rc) {
4384                 dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n", rc);
4385                 goto err_disable_device;
4386         }
4387
4388         pci_set_master(pdev);
4389
4390         ena_dev = vzalloc(sizeof(*ena_dev));
4391         if (!ena_dev) {
4392                 rc = -ENOMEM;
4393                 goto err_disable_device;
4394         }
4395
4396         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4397         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4398         if (rc) {
4399                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4400                         rc);
4401                 goto err_free_ena_dev;
4402         }
4403
4404         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4405                                         pci_resource_start(pdev, ENA_REG_BAR),
4406                                         pci_resource_len(pdev, ENA_REG_BAR));
4407         if (!ena_dev->reg_bar) {
4408                 dev_err(&pdev->dev, "Failed to remap regs bar\n");
4409                 rc = -EFAULT;
4410                 goto err_free_region;
4411         }
4412
4413         ena_dev->ena_min_poll_delay_us = ENA_ADMIN_POLL_DELAY_US;
4414
4415         ena_dev->dmadev = &pdev->dev;
4416
4417         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), ENA_MAX_RINGS);
4418         if (!netdev) {
4419                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4420                 rc = -ENOMEM;
4421                 goto err_free_region;
4422         }
4423
4424         SET_NETDEV_DEV(netdev, &pdev->dev);
4425         adapter = netdev_priv(netdev);
4426         adapter->ena_dev = ena_dev;
4427         adapter->netdev = netdev;
4428         adapter->pdev = pdev;
4429         adapter->msg_enable = DEFAULT_MSG_ENABLE;
4430
4431         ena_dev->net_device = netdev;
4432
4433         pci_set_drvdata(pdev, adapter);
4434
4435         rc = ena_map_llq_mem_bar(pdev, ena_dev, bars);
4436         if (rc) {
4437                 dev_err(&pdev->dev, "ENA LLQ bar mapping failed\n");
4438                 goto err_netdev_destroy;
4439         }
4440
4441         rc = ena_device_init(adapter, pdev, &get_feat_ctx, &wd_state);
4442         if (rc) {
4443                 dev_err(&pdev->dev, "ENA device init failed\n");
4444                 if (rc == -ETIME)
4445                         rc = -EPROBE_DEFER;
4446                 goto err_netdev_destroy;
4447         }
4448
4449         /* Initial TX and RX interrupt delay. Assumes 1 usec granularity.
4450          * Updated during device initialization with the real granularity
4451          */
4452         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4453         ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4454         ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4455         max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4456         if (unlikely(!max_num_io_queues)) {
4457                 rc = -EFAULT;
4458                 goto err_device_destroy;
4459         }
4460
4461         ena_set_conf_feat_params(adapter, &get_feat_ctx);
4462
4463         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4464
4465         adapter->num_io_queues = max_num_io_queues;
4466         adapter->max_num_io_queues = max_num_io_queues;
4467         adapter->last_monitored_tx_qid = 0;
4468
4469         adapter->xdp_first_ring = 0;
4470         adapter->xdp_num_queues = 0;
4471
4472         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4473         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4474                 adapter->disable_meta_caching =
4475                         !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
4476                            BIT(ENA_ADMIN_DISABLE_META_CACHING));
4477
4478         adapter->wd_state = wd_state;
4479
4480         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4481
4482         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4483         if (rc) {
4484                 dev_err(&pdev->dev,
4485                         "Failed to query interrupt moderation feature\n");
4486                 goto err_device_destroy;
4487         }
4488
4489         ena_init_io_rings(adapter,
4490                           0,
4491                           adapter->xdp_num_queues +
4492                           adapter->num_io_queues);
4493
4494         netdev->netdev_ops = &ena_netdev_ops;
4495         netdev->watchdog_timeo = TX_TIMEOUT;
4496         ena_set_ethtool_ops(netdev);
4497
4498         netdev->priv_flags |= IFF_UNICAST_FLT;
4499
4500         u64_stats_init(&adapter->syncp);
4501
4502         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4503         if (rc) {
4504                 dev_err(&pdev->dev,
4505                         "Failed to enable and set the admin interrupts\n");
4506                 goto err_worker_destroy;
4507         }
4508         rc = ena_rss_init_default(adapter);
4509         if (rc && (rc != -EOPNOTSUPP)) {
4510                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4511                 goto err_free_msix;
4512         }
4513
4514         ena_config_debug_area(adapter);
4515
4516         if (ena_xdp_legal_queue_count(adapter, adapter->num_io_queues))
4517                 netdev->xdp_features = NETDEV_XDP_ACT_BASIC |
4518                                        NETDEV_XDP_ACT_REDIRECT;
4519
4520         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4521
4522         netif_carrier_off(netdev);
4523
4524         rc = register_netdev(netdev);
4525         if (rc) {
4526                 dev_err(&pdev->dev, "Cannot register net device\n");
4527                 goto err_rss;
4528         }
4529
4530         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4531
4532         adapter->last_keep_alive_jiffies = jiffies;
4533         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4534         adapter->missing_tx_completion_to = TX_TIMEOUT;
4535         adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4536
4537         ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4538
4539         timer_setup(&adapter->timer_service, ena_timer_service, 0);
4540         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4541
4542         dev_info(&pdev->dev,
4543                  "%s found at mem %lx, mac addr %pM\n",
4544                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4545                  netdev->dev_addr);
4546
4547         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4548
4549         adapters_found++;
4550
4551         return 0;
4552
4553 err_rss:
4554         ena_com_delete_debug_area(ena_dev);
4555         ena_com_rss_destroy(ena_dev);
4556 err_free_msix:
4557         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4558         /* stop submitting admin commands on a device that was reset */
4559         ena_com_set_admin_running_state(ena_dev, false);
4560         ena_free_mgmnt_irq(adapter);
4561         ena_disable_msix(adapter);
4562 err_worker_destroy:
4563         del_timer(&adapter->timer_service);
4564 err_device_destroy:
4565         ena_com_delete_host_info(ena_dev);
4566         ena_com_admin_destroy(ena_dev);
4567 err_netdev_destroy:
4568         free_netdev(netdev);
4569 err_free_region:
4570         ena_release_bars(ena_dev, pdev);
4571 err_free_ena_dev:
4572         vfree(ena_dev);
4573 err_disable_device:
4574         pci_disable_device(pdev);
4575         return rc;
4576 }
4577
4578 /*****************************************************************************/
4579
4580 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4581  * @pdev: PCI device information struct
4582  * @shutdown: Is it a shutdown operation? If false, means it is a removal
4583  *
4584  * __ena_shutoff is a helper routine that does the real work on shutdown and
4585  * removal paths; the difference between those paths is with regards to whether
4586  * dettach or unregister the netdevice.
4587  */
4588 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4589 {
4590         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4591         struct ena_com_dev *ena_dev;
4592         struct net_device *netdev;
4593
4594         ena_dev = adapter->ena_dev;
4595         netdev = adapter->netdev;
4596
4597 #ifdef CONFIG_RFS_ACCEL
4598         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4599                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4600                 netdev->rx_cpu_rmap = NULL;
4601         }
4602 #endif /* CONFIG_RFS_ACCEL */
4603
4604         /* Make sure timer and reset routine won't be called after
4605          * freeing device resources.
4606          */
4607         del_timer_sync(&adapter->timer_service);
4608         cancel_work_sync(&adapter->reset_task);
4609
4610         rtnl_lock(); /* lock released inside the below if-else block */
4611         adapter->reset_reason = ENA_REGS_RESET_SHUTDOWN;
4612         ena_destroy_device(adapter, true);
4613
4614         if (shutdown) {
4615                 netif_device_detach(netdev);
4616                 dev_close(netdev);
4617                 rtnl_unlock();
4618         } else {
4619                 rtnl_unlock();
4620                 unregister_netdev(netdev);
4621                 free_netdev(netdev);
4622         }
4623
4624         ena_com_rss_destroy(ena_dev);
4625
4626         ena_com_delete_debug_area(ena_dev);
4627
4628         ena_com_delete_host_info(ena_dev);
4629
4630         ena_release_bars(ena_dev, pdev);
4631
4632         pci_disable_device(pdev);
4633
4634         vfree(ena_dev);
4635 }
4636
4637 /* ena_remove - Device Removal Routine
4638  * @pdev: PCI device information struct
4639  *
4640  * ena_remove is called by the PCI subsystem to alert the driver
4641  * that it should release a PCI device.
4642  */
4643
4644 static void ena_remove(struct pci_dev *pdev)
4645 {
4646         __ena_shutoff(pdev, false);
4647 }
4648
4649 /* ena_shutdown - Device Shutdown Routine
4650  * @pdev: PCI device information struct
4651  *
4652  * ena_shutdown is called by the PCI subsystem to alert the driver that
4653  * a shutdown/reboot (or kexec) is happening and device must be disabled.
4654  */
4655
4656 static void ena_shutdown(struct pci_dev *pdev)
4657 {
4658         __ena_shutoff(pdev, true);
4659 }
4660
4661 /* ena_suspend - PM suspend callback
4662  * @dev_d: Device information struct
4663  */
4664 static int __maybe_unused ena_suspend(struct device *dev_d)
4665 {
4666         struct pci_dev *pdev = to_pci_dev(dev_d);
4667         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4668
4669         ena_increase_stat(&adapter->dev_stats.suspend, 1, &adapter->syncp);
4670
4671         rtnl_lock();
4672         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4673                 dev_err(&pdev->dev,
4674                         "Ignoring device reset request as the device is being suspended\n");
4675                 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4676         }
4677         ena_destroy_device(adapter, true);
4678         rtnl_unlock();
4679         return 0;
4680 }
4681
4682 /* ena_resume - PM resume callback
4683  * @dev_d: Device information struct
4684  */
4685 static int __maybe_unused ena_resume(struct device *dev_d)
4686 {
4687         struct ena_adapter *adapter = dev_get_drvdata(dev_d);
4688         int rc;
4689
4690         ena_increase_stat(&adapter->dev_stats.resume, 1, &adapter->syncp);
4691
4692         rtnl_lock();
4693         rc = ena_restore_device(adapter);
4694         rtnl_unlock();
4695         return rc;
4696 }
4697
4698 static SIMPLE_DEV_PM_OPS(ena_pm_ops, ena_suspend, ena_resume);
4699
4700 static struct pci_driver ena_pci_driver = {
4701         .name           = DRV_MODULE_NAME,
4702         .id_table       = ena_pci_tbl,
4703         .probe          = ena_probe,
4704         .remove         = ena_remove,
4705         .shutdown       = ena_shutdown,
4706         .driver.pm      = &ena_pm_ops,
4707         .sriov_configure = pci_sriov_configure_simple,
4708 };
4709
4710 static int __init ena_init(void)
4711 {
4712         int ret;
4713
4714         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4715         if (!ena_wq) {
4716                 pr_err("Failed to create workqueue\n");
4717                 return -ENOMEM;
4718         }
4719
4720         ret = pci_register_driver(&ena_pci_driver);
4721         if (ret)
4722                 destroy_workqueue(ena_wq);
4723
4724         return ret;
4725 }
4726
4727 static void __exit ena_cleanup(void)
4728 {
4729         pci_unregister_driver(&ena_pci_driver);
4730
4731         if (ena_wq) {
4732                 destroy_workqueue(ena_wq);
4733                 ena_wq = NULL;
4734         }
4735 }
4736
4737 /******************************************************************************
4738  ******************************** AENQ Handlers *******************************
4739  *****************************************************************************/
4740 /* ena_update_on_link_change:
4741  * Notify the network interface about the change in link status
4742  */
4743 static void ena_update_on_link_change(void *adapter_data,
4744                                       struct ena_admin_aenq_entry *aenq_e)
4745 {
4746         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4747         struct ena_admin_aenq_link_change_desc *aenq_desc =
4748                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
4749         int status = aenq_desc->flags &
4750                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4751
4752         if (status) {
4753                 netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
4754                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4755                 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4756                         netif_carrier_on(adapter->netdev);
4757         } else {
4758                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4759                 netif_carrier_off(adapter->netdev);
4760         }
4761 }
4762
4763 static void ena_keep_alive_wd(void *adapter_data,
4764                               struct ena_admin_aenq_entry *aenq_e)
4765 {
4766         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4767         struct ena_admin_aenq_keep_alive_desc *desc;
4768         u64 rx_drops;
4769         u64 tx_drops;
4770
4771         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4772         adapter->last_keep_alive_jiffies = jiffies;
4773
4774         rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4775         tx_drops = ((u64)desc->tx_drops_high << 32) | desc->tx_drops_low;
4776
4777         u64_stats_update_begin(&adapter->syncp);
4778         /* These stats are accumulated by the device, so the counters indicate
4779          * all drops since last reset.
4780          */
4781         adapter->dev_stats.rx_drops = rx_drops;
4782         adapter->dev_stats.tx_drops = tx_drops;
4783         u64_stats_update_end(&adapter->syncp);
4784 }
4785
4786 static void ena_notification(void *adapter_data,
4787                              struct ena_admin_aenq_entry *aenq_e)
4788 {
4789         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4790         struct ena_admin_ena_hw_hints *hints;
4791
4792         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4793              "Invalid group(%x) expected %x\n",
4794              aenq_e->aenq_common_desc.group,
4795              ENA_ADMIN_NOTIFICATION);
4796
4797         switch (aenq_e->aenq_common_desc.syndrome) {
4798         case ENA_ADMIN_UPDATE_HINTS:
4799                 hints = (struct ena_admin_ena_hw_hints *)
4800                         (&aenq_e->inline_data_w4);
4801                 ena_update_hints(adapter, hints);
4802                 break;
4803         default:
4804                 netif_err(adapter, drv, adapter->netdev,
4805                           "Invalid aenq notification link state %d\n",
4806                           aenq_e->aenq_common_desc.syndrome);
4807         }
4808 }
4809
4810 /* This handler will called for unknown event group or unimplemented handlers*/
4811 static void unimplemented_aenq_handler(void *data,
4812                                        struct ena_admin_aenq_entry *aenq_e)
4813 {
4814         struct ena_adapter *adapter = (struct ena_adapter *)data;
4815
4816         netif_err(adapter, drv, adapter->netdev,
4817                   "Unknown event was received or event with unimplemented handler\n");
4818 }
4819
4820 static struct ena_aenq_handlers aenq_handlers = {
4821         .handlers = {
4822                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4823                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
4824                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4825         },
4826         .unimplemented_handler = unimplemented_aenq_handler
4827 };
4828
4829 module_init(ena_init);
4830 module_exit(ena_cleanup);