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         struct bpf_prog *xdp_prog;
 378         struct ena_ring *xdp_ring;
 379         u32 verdict = XDP_PASS;
 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 = XDP_ABORTED;
 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                 break;
 413         case XDP_REDIRECT:
 414                 if (likely(!xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog))) {
 415                         xdp_stat = &rx_ring->rx_stats.xdp_redirect;
 416                         break;
 417                 }
 418                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 419                 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 420                 verdict = XDP_ABORTED;
 421                 break;
 422         case XDP_ABORTED:
 423                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 424                 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 425                 break;
 426         case XDP_DROP:
 427                 xdp_stat = &rx_ring->rx_stats.xdp_drop;
 428                 break;
 429         case XDP_PASS:
 430                 xdp_stat = &rx_ring->rx_stats.xdp_pass;
 431                 break;
 432         default:
 433                 bpf_warn_invalid_xdp_action(rx_ring->netdev, xdp_prog, verdict);
 434                 xdp_stat = &rx_ring->rx_stats.xdp_invalid;
 435         }
 436
 437         ena_increase_stat(xdp_stat, 1, &rx_ring->syncp);
 438 out:
 439         return verdict;
 440 }
 441
 442 static void ena_init_all_xdp_queues(struct ena_adapter *adapter)
 443 {
 444         adapter->xdp_first_ring = adapter->num_io_queues;
 445         adapter->xdp_num_queues = adapter->num_io_queues;
 446
 447         ena_init_io_rings(adapter,
 448                           adapter->xdp_first_ring,
 449                           adapter->xdp_num_queues);
 450 }
 451
 452 static int ena_setup_and_create_all_xdp_queues(struct ena_adapter *adapter)
 453 {
 454         int rc = 0;
 455
 456         rc = ena_setup_tx_resources_in_range(adapter, adapter->xdp_first_ring,
 457                                              adapter->xdp_num_queues);
 458         if (rc)
 459                 goto setup_err;
 460
 461         rc = ena_create_io_tx_queues_in_range(adapter,
 462                                               adapter->xdp_first_ring,
 463                                               adapter->xdp_num_queues);
 464         if (rc)
 465                 goto create_err;
 466
 467         return 0;
 468
 469 create_err:
 470         ena_free_all_io_tx_resources(adapter);
 471 setup_err:
 472         return rc;
 473 }
 474
 475 /* Provides a way for both kernel and bpf-prog to know
 476  * more about the RX-queue a given XDP frame arrived on.
 477  */
 478 static int ena_xdp_register_rxq_info(struct ena_ring *rx_ring)
 479 {
 480         int rc;
 481
 482         rc = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev, rx_ring->qid, 0);
 483
 484         if (rc) {
 485                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 486                           "Failed to register xdp rx queue info. RX queue num %d rc: %d\n",
 487                           rx_ring->qid, rc);
 488                 goto err;
 489         }
 490
 491         rc = xdp_rxq_info_reg_mem_model(&rx_ring->xdp_rxq, MEM_TYPE_PAGE_SHARED,
 492                                         NULL);
 493
 494         if (rc) {
 495                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 496                           "Failed to register xdp rx queue info memory model. RX queue num %d rc: %d\n",
 497                           rx_ring->qid, rc);
 498                 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 499         }
 500
 501 err:
 502         return rc;
 503 }
 504
 505 static void ena_xdp_unregister_rxq_info(struct ena_ring *rx_ring)
 506 {
 507         xdp_rxq_info_unreg_mem_model(&rx_ring->xdp_rxq);
 508         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 509 }
 510
 511 static void ena_xdp_exchange_program_rx_in_range(struct ena_adapter *adapter,
 512                                                  struct bpf_prog *prog,
 513                                                  int first, int count)
 514 {
 515         struct ena_ring *rx_ring;
 516         int i = 0;
 517
 518         for (i = first; i < count; i++) {
 519                 rx_ring = &adapter->rx_ring[i];
 520                 xchg(&rx_ring->xdp_bpf_prog, prog);
 521                 if (prog) {
 522                         ena_xdp_register_rxq_info(rx_ring);
 523                         rx_ring->rx_headroom = XDP_PACKET_HEADROOM;
 524                 } else {
 525                         ena_xdp_unregister_rxq_info(rx_ring);
 526                         rx_ring->rx_headroom = NET_SKB_PAD;
 527                 }
 528         }
 529 }
 530
 531 static void ena_xdp_exchange_program(struct ena_adapter *adapter,
 532                                      struct bpf_prog *prog)
 533 {
 534         struct bpf_prog *old_bpf_prog = xchg(&adapter->xdp_bpf_prog, prog);
 535
 536         ena_xdp_exchange_program_rx_in_range(adapter,
 537                                              prog,
 538                                              0,
 539                                              adapter->num_io_queues);
 540
 541         if (old_bpf_prog)
 542                 bpf_prog_put(old_bpf_prog);
 543 }
 544
 545 static int ena_destroy_and_free_all_xdp_queues(struct ena_adapter *adapter)
 546 {
 547         bool was_up;
 548         int rc;
 549
 550         was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 551
 552         if (was_up)
 553                 ena_down(adapter);
 554
 555         adapter->xdp_first_ring = 0;
 556         adapter->xdp_num_queues = 0;
 557         ena_xdp_exchange_program(adapter, NULL);
 558         if (was_up) {
 559                 rc = ena_up(adapter);
 560                 if (rc)
 561                         return rc;
 562         }
 563         return 0;
 564 }
 565
 566 static int ena_xdp_set(struct net_device *netdev, struct netdev_bpf *bpf)
 567 {
 568         struct ena_adapter *adapter = netdev_priv(netdev);
 569         struct bpf_prog *prog = bpf->prog;
 570         struct bpf_prog *old_bpf_prog;
 571         int rc, prev_mtu;
 572         bool is_up;
 573
 574         is_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 575         rc = ena_xdp_allowed(adapter);
 576         if (rc == ENA_XDP_ALLOWED) {
 577                 old_bpf_prog = adapter->xdp_bpf_prog;
 578                 if (prog) {
 579                         if (!is_up) {
 580                                 ena_init_all_xdp_queues(adapter);
 581                         } else if (!old_bpf_prog) {
 582                                 ena_down(adapter);
 583                                 ena_init_all_xdp_queues(adapter);
 584                         }
 585                         ena_xdp_exchange_program(adapter, prog);
 586
 587                         if (is_up && !old_bpf_prog) {
 588                                 rc = ena_up(adapter);
 589                                 if (rc)
 590                                         return rc;
 591                         }
 592                 } else if (old_bpf_prog) {
 593                         rc = ena_destroy_and_free_all_xdp_queues(adapter);
 594                         if (rc)
 595                                 return rc;
 596                 }
 597
 598                 prev_mtu = netdev->max_mtu;
 599                 netdev->max_mtu = prog ? ENA_XDP_MAX_MTU : adapter->max_mtu;
 600
 601                 if (!old_bpf_prog)
 602                         netif_info(adapter, drv, adapter->netdev,
 603                                    "XDP program is set, changing the max_mtu from %d to %d",
 604                                    prev_mtu, netdev->max_mtu);
 605
 606         } else if (rc == ENA_XDP_CURRENT_MTU_TOO_LARGE) {
 607                 netif_err(adapter, drv, adapter->netdev,
 608                           "Failed to set xdp program, the current MTU (%d) is larger than the maximum allowed MTU (%lu) while xdp is on",
 609                           netdev->mtu, ENA_XDP_MAX_MTU);
 610                 NL_SET_ERR_MSG_MOD(bpf->extack,
 611                                    "Failed to set xdp program, the current MTU is larger than the maximum allowed MTU. Check the dmesg for more info");
 612                 return -EINVAL;
 613         } else if (rc == ENA_XDP_NO_ENOUGH_QUEUES) {
 614                 netif_err(adapter, drv, adapter->netdev,
 615                           "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",
 616                           adapter->num_io_queues, adapter->max_num_io_queues);
 617                 NL_SET_ERR_MSG_MOD(bpf->extack,
 618                                    "Failed to set xdp program, there is no enough space for allocating XDP queues, Check the dmesg for more info");
 619                 return -EINVAL;
 620         }
 621
 622         return 0;
 623 }
 624
 625 /* This is the main xdp callback, it's used by the kernel to set/unset the xdp
 626  * program as well as to query the current xdp program id.
 627  */
 628 static int ena_xdp(struct net_device *netdev, struct netdev_bpf *bpf)
 629 {
 630         switch (bpf->command) {
 631         case XDP_SETUP_PROG:
 632                 return ena_xdp_set(netdev, bpf);
 633         default:
 634                 return -EINVAL;
 635         }
 636         return 0;
 637 }
 638
 639 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
 640 {
 641 #ifdef CONFIG_RFS_ACCEL
 642         u32 i;
 643         int rc;
 644
 645         adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_io_queues);
 646         if (!adapter->netdev->rx_cpu_rmap)
 647                 return -ENOMEM;
 648         for (i = 0; i < adapter->num_io_queues; i++) {
 649                 int irq_idx = ENA_IO_IRQ_IDX(i);
 650
 651                 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
 652                                       pci_irq_vector(adapter->pdev, irq_idx));
 653                 if (rc) {
 654                         free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
 655                         adapter->netdev->rx_cpu_rmap = NULL;
 656                         return rc;
 657                 }
 658         }
 659 #endif /* CONFIG_RFS_ACCEL */
 660         return 0;
 661 }
 662
 663 static void ena_init_io_rings_common(struct ena_adapter *adapter,
 664                                      struct ena_ring *ring, u16 qid)
 665 {
 666         ring->qid = qid;
 667         ring->pdev = adapter->pdev;
 668         ring->dev = &adapter->pdev->dev;
 669         ring->netdev = adapter->netdev;
 670         ring->napi = &adapter->ena_napi[qid].napi;
 671         ring->adapter = adapter;
 672         ring->ena_dev = adapter->ena_dev;
 673         ring->per_napi_packets = 0;
 674         ring->cpu = 0;
 675         ring->no_interrupt_event_cnt = 0;
 676         u64_stats_init(&ring->syncp);
 677 }
 678
 679 static void ena_init_io_rings(struct ena_adapter *adapter,
 680                               int first_index, int count)
 681 {
 682         struct ena_com_dev *ena_dev;
 683         struct ena_ring *txr, *rxr;
 684         int i;
 685
 686         ena_dev = adapter->ena_dev;
 687
 688         for (i = first_index; i < first_index + count; i++) {
 689                 txr = &adapter->tx_ring[i];
 690                 rxr = &adapter->rx_ring[i];
 691
 692                 /* TX common ring state */
 693                 ena_init_io_rings_common(adapter, txr, i);
 694
 695                 /* TX specific ring state */
 696                 txr->ring_size = adapter->requested_tx_ring_size;
 697                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
 698                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
 699                 txr->sgl_size = adapter->max_tx_sgl_size;
 700                 txr->smoothed_interval =
 701                         ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
 702                 txr->disable_meta_caching = adapter->disable_meta_caching;
 703                 spin_lock_init(&txr->xdp_tx_lock);
 704
 705                 /* Don't init RX queues for xdp queues */
 706                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
 707                         /* RX common ring state */
 708                         ena_init_io_rings_common(adapter, rxr, i);
 709
 710                         /* RX specific ring state */
 711                         rxr->ring_size = adapter->requested_rx_ring_size;
 712                         rxr->rx_copybreak = adapter->rx_copybreak;
 713                         rxr->sgl_size = adapter->max_rx_sgl_size;
 714                         rxr->smoothed_interval =
 715                                 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
 716                         rxr->empty_rx_queue = 0;
 717                         rxr->rx_headroom = NET_SKB_PAD;
 718                         adapter->ena_napi[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
 719                         rxr->xdp_ring = &adapter->tx_ring[i + adapter->num_io_queues];
 720                 }
 721         }
 722 }
 723
 724 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
 725  * @adapter: network interface device structure
 726  * @qid: queue index
 727  *
 728  * Return 0 on success, negative on failure
 729  */
 730 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
 731 {
 732         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 733         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 734         int size, i, node;
 735
 736         if (tx_ring->tx_buffer_info) {
 737                 netif_err(adapter, ifup,
 738                           adapter->netdev, "tx_buffer_info info is not NULL");
 739                 return -EEXIST;
 740         }
 741
 742         size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
 743         node = cpu_to_node(ena_irq->cpu);
 744
 745         tx_ring->tx_buffer_info = vzalloc_node(size, node);
 746         if (!tx_ring->tx_buffer_info) {
 747                 tx_ring->tx_buffer_info = vzalloc(size);
 748                 if (!tx_ring->tx_buffer_info)
 749                         goto err_tx_buffer_info;
 750         }
 751
 752         size = sizeof(u16) * tx_ring->ring_size;
 753         tx_ring->free_ids = vzalloc_node(size, node);
 754         if (!tx_ring->free_ids) {
 755                 tx_ring->free_ids = vzalloc(size);
 756                 if (!tx_ring->free_ids)
 757                         goto err_tx_free_ids;
 758         }
 759
 760         size = tx_ring->tx_max_header_size;
 761         tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
 762         if (!tx_ring->push_buf_intermediate_buf) {
 763                 tx_ring->push_buf_intermediate_buf = vzalloc(size);
 764                 if (!tx_ring->push_buf_intermediate_buf)
 765                         goto err_push_buf_intermediate_buf;
 766         }
 767
 768         /* Req id ring for TX out of order completions */
 769         for (i = 0; i < tx_ring->ring_size; i++)
 770                 tx_ring->free_ids[i] = i;
 771
 772         /* Reset tx statistics */
 773         memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
 774
 775         tx_ring->next_to_use = 0;
 776         tx_ring->next_to_clean = 0;
 777         tx_ring->cpu = ena_irq->cpu;
 778         return 0;
 779
 780 err_push_buf_intermediate_buf:
 781         vfree(tx_ring->free_ids);
 782         tx_ring->free_ids = NULL;
 783 err_tx_free_ids:
 784         vfree(tx_ring->tx_buffer_info);
 785         tx_ring->tx_buffer_info = NULL;
 786 err_tx_buffer_info:
 787         return -ENOMEM;
 788 }
 789
 790 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
 791  * @adapter: network interface device structure
 792  * @qid: queue index
 793  *
 794  * Free all transmit software resources
 795  */
 796 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
 797 {
 798         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 799
 800         vfree(tx_ring->tx_buffer_info);
 801         tx_ring->tx_buffer_info = NULL;
 802
 803         vfree(tx_ring->free_ids);
 804         tx_ring->free_ids = NULL;
 805
 806         vfree(tx_ring->push_buf_intermediate_buf);
 807         tx_ring->push_buf_intermediate_buf = NULL;
 808 }
 809
 810 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
 811                                            int first_index,
 812                                            int count)
 813 {
 814         int i, rc = 0;
 815
 816         for (i = first_index; i < first_index + count; i++) {
 817                 rc = ena_setup_tx_resources(adapter, i);
 818                 if (rc)
 819                         goto err_setup_tx;
 820         }
 821
 822         return 0;
 823
 824 err_setup_tx:
 825
 826         netif_err(adapter, ifup, adapter->netdev,
 827                   "Tx queue %d: allocation failed\n", i);
 828
 829         /* rewind the index freeing the rings as we go */
 830         while (first_index < i--)
 831                 ena_free_tx_resources(adapter, i);
 832         return rc;
 833 }
 834
 835 static void ena_free_all_io_tx_resources_in_range(struct ena_adapter *adapter,
 836                                                   int first_index, int count)
 837 {
 838         int i;
 839
 840         for (i = first_index; i < first_index + count; i++)
 841                 ena_free_tx_resources(adapter, i);
 842 }
 843
 844 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
 845  * @adapter: board private structure
 846  *
 847  * Free all transmit software resources
 848  */
 849 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
 850 {
 851         ena_free_all_io_tx_resources_in_range(adapter,
 852                                               0,
 853                                               adapter->xdp_num_queues +
 854                                               adapter->num_io_queues);
 855 }
 856
 857 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
 858  * @adapter: network interface device structure
 859  * @qid: queue index
 860  *
 861  * Returns 0 on success, negative on failure
 862  */
 863 static int ena_setup_rx_resources(struct ena_adapter *adapter,
 864                                   u32 qid)
 865 {
 866         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 867         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 868         int size, node, i;
 869
 870         if (rx_ring->rx_buffer_info) {
 871                 netif_err(adapter, ifup, adapter->netdev,
 872                           "rx_buffer_info is not NULL");
 873                 return -EEXIST;
 874         }
 875
 876         /* alloc extra element so in rx path
 877          * we can always prefetch rx_info + 1
 878          */
 879         size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
 880         node = cpu_to_node(ena_irq->cpu);
 881
 882         rx_ring->rx_buffer_info = vzalloc_node(size, node);
 883         if (!rx_ring->rx_buffer_info) {
 884                 rx_ring->rx_buffer_info = vzalloc(size);
 885                 if (!rx_ring->rx_buffer_info)
 886                         return -ENOMEM;
 887         }
 888
 889         size = sizeof(u16) * rx_ring->ring_size;
 890         rx_ring->free_ids = vzalloc_node(size, node);
 891         if (!rx_ring->free_ids) {
 892                 rx_ring->free_ids = vzalloc(size);
 893                 if (!rx_ring->free_ids) {
 894                         vfree(rx_ring->rx_buffer_info);
 895                         rx_ring->rx_buffer_info = NULL;
 896                         return -ENOMEM;
 897                 }
 898         }
 899
 900         /* Req id ring for receiving RX pkts out of order */
 901         for (i = 0; i < rx_ring->ring_size; i++)
 902                 rx_ring->free_ids[i] = i;
 903
 904         /* Reset rx statistics */
 905         memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
 906
 907         rx_ring->next_to_clean = 0;
 908         rx_ring->next_to_use = 0;
 909         rx_ring->cpu = ena_irq->cpu;
 910
 911         return 0;
 912 }
 913
 914 /* ena_free_rx_resources - Free I/O Rx Resources
 915  * @adapter: network interface device structure
 916  * @qid: queue index
 917  *
 918  * Free all receive software resources
 919  */
 920 static void ena_free_rx_resources(struct ena_adapter *adapter,
 921                                   u32 qid)
 922 {
 923         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 924
 925         vfree(rx_ring->rx_buffer_info);
 926         rx_ring->rx_buffer_info = NULL;
 927
 928         vfree(rx_ring->free_ids);
 929         rx_ring->free_ids = NULL;
 930 }
 931
 932 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
 933  * @adapter: board private structure
 934  *
 935  * Return 0 on success, negative on failure
 936  */
 937 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
 938 {
 939         int i, rc = 0;
 940
 941         for (i = 0; i < adapter->num_io_queues; i++) {
 942                 rc = ena_setup_rx_resources(adapter, i);
 943                 if (rc)
 944                         goto err_setup_rx;
 945         }
 946
 947         return 0;
 948
 949 err_setup_rx:
 950
 951         netif_err(adapter, ifup, adapter->netdev,
 952                   "Rx queue %d: allocation failed\n", i);
 953
 954         /* rewind the index freeing the rings as we go */
 955         while (i--)
 956                 ena_free_rx_resources(adapter, i);
 957         return rc;
 958 }
 959
 960 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
 961  * @adapter: board private structure
 962  *
 963  * Free all receive software resources
 964  */
 965 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
 966 {
 967         int i;
 968
 969         for (i = 0; i < adapter->num_io_queues; i++)
 970                 ena_free_rx_resources(adapter, i);
 971 }
 972
 973 static struct page *ena_alloc_map_page(struct ena_ring *rx_ring,
 974                                        dma_addr_t *dma)
 975 {
 976         struct page *page;
 977
 978         /* This would allocate the page on the same NUMA node the executing code
 979          * is running on.
 980          */
 981         page = dev_alloc_page();
 982         if (!page) {
 983                 ena_increase_stat(&rx_ring->rx_stats.page_alloc_fail, 1,
 984                                   &rx_ring->syncp);
 985                 return ERR_PTR(-ENOSPC);
 986         }
 987
 988         /* To enable NIC-side port-mirroring, AKA SPAN port,
 989          * we make the buffer readable from the nic as well
 990          */
 991         *dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
 992                             DMA_BIDIRECTIONAL);
 993         if (unlikely(dma_mapping_error(rx_ring->dev, *dma))) {
 994                 ena_increase_stat(&rx_ring->rx_stats.dma_mapping_err, 1,
 995                                   &rx_ring->syncp);
 996                 __free_page(page);
 997                 return ERR_PTR(-EIO);
 998         }
 999
1000         return page;
1001 }
1002
1003 static int ena_alloc_rx_buffer(struct ena_ring *rx_ring,
1004                                struct ena_rx_buffer *rx_info)
1005 {
1006         int headroom = rx_ring->rx_headroom;
1007         struct ena_com_buf *ena_buf;
1008         struct page *page;
1009         dma_addr_t dma;
1010         int tailroom;
1011
1012         /* restore page offset value in case it has been changed by device */
1013         rx_info->page_offset = headroom;
1014
1015         /* if previous allocated page is not used */
1016         if (unlikely(rx_info->page))
1017                 return 0;
1018
1019         /* We handle DMA here */
1020         page = ena_alloc_map_page(rx_ring, &dma);
1021         if (unlikely(IS_ERR(page)))
1022                 return PTR_ERR(page);
1023
1024         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1025                   "Allocate page %p, rx_info %p\n", page, rx_info);
1026
1027         tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1028
1029         rx_info->page = page;
1030         ena_buf = &rx_info->ena_buf;
1031         ena_buf->paddr = dma + headroom;
1032         ena_buf->len = ENA_PAGE_SIZE - headroom - tailroom;
1033
1034         return 0;
1035 }
1036
1037 static void ena_unmap_rx_buff(struct ena_ring *rx_ring,
1038                               struct ena_rx_buffer *rx_info)
1039 {
1040         struct ena_com_buf *ena_buf = &rx_info->ena_buf;
1041
1042         dma_unmap_page(rx_ring->dev, ena_buf->paddr - rx_ring->rx_headroom,
1043                        ENA_PAGE_SIZE,
1044                        DMA_BIDIRECTIONAL);
1045 }
1046
1047 static void ena_free_rx_page(struct ena_ring *rx_ring,
1048                              struct ena_rx_buffer *rx_info)
1049 {
1050         struct page *page = rx_info->page;
1051
1052         if (unlikely(!page)) {
1053                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1054                            "Trying to free unallocated buffer\n");
1055                 return;
1056         }
1057
1058         ena_unmap_rx_buff(rx_ring, rx_info);
1059
1060         __free_page(page);
1061         rx_info->page = NULL;
1062 }
1063
1064 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
1065 {
1066         u16 next_to_use, req_id;
1067         u32 i;
1068         int rc;
1069
1070         next_to_use = rx_ring->next_to_use;
1071
1072         for (i = 0; i < num; i++) {
1073                 struct ena_rx_buffer *rx_info;
1074
1075                 req_id = rx_ring->free_ids[next_to_use];
1076
1077                 rx_info = &rx_ring->rx_buffer_info[req_id];
1078
1079                 rc = ena_alloc_rx_buffer(rx_ring, rx_info);
1080                 if (unlikely(rc < 0)) {
1081                         netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1082                                    "Failed to allocate buffer for rx queue %d\n",
1083                                    rx_ring->qid);
1084                         break;
1085                 }
1086                 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1087                                                 &rx_info->ena_buf,
1088                                                 req_id);
1089                 if (unlikely(rc)) {
1090                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1091                                    "Failed to add buffer for rx queue %d\n",
1092                                    rx_ring->qid);
1093                         break;
1094                 }
1095                 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1096                                                    rx_ring->ring_size);
1097         }
1098
1099         if (unlikely(i < num)) {
1100                 ena_increase_stat(&rx_ring->rx_stats.refil_partial, 1,
1101                                   &rx_ring->syncp);
1102                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1103                            "Refilled rx qid %d with only %d buffers (from %d)\n",
1104                            rx_ring->qid, i, num);
1105         }
1106
1107         /* ena_com_write_sq_doorbell issues a wmb() */
1108         if (likely(i))
1109                 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1110
1111         rx_ring->next_to_use = next_to_use;
1112
1113         return i;
1114 }
1115
1116 static void ena_free_rx_bufs(struct ena_adapter *adapter,
1117                              u32 qid)
1118 {
1119         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1120         u32 i;
1121
1122         for (i = 0; i < rx_ring->ring_size; i++) {
1123                 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1124
1125                 if (rx_info->page)
1126                         ena_free_rx_page(rx_ring, rx_info);
1127         }
1128 }
1129
1130 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
1131  * @adapter: board private structure
1132  */
1133 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1134 {
1135         struct ena_ring *rx_ring;
1136         int i, rc, bufs_num;
1137
1138         for (i = 0; i < adapter->num_io_queues; i++) {
1139                 rx_ring = &adapter->rx_ring[i];
1140                 bufs_num = rx_ring->ring_size - 1;
1141                 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1142
1143                 if (unlikely(rc != bufs_num))
1144                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1145                                    "Refilling Queue %d failed. allocated %d buffers from: %d\n",
1146                                    i, rc, bufs_num);
1147         }
1148 }
1149
1150 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
1151 {
1152         int i;
1153
1154         for (i = 0; i < adapter->num_io_queues; i++)
1155                 ena_free_rx_bufs(adapter, i);
1156 }
1157
1158 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
1159                               struct ena_tx_buffer *tx_info)
1160 {
1161         struct ena_com_buf *ena_buf;
1162         u32 cnt;
1163         int i;
1164
1165         ena_buf = tx_info->bufs;
1166         cnt = tx_info->num_of_bufs;
1167
1168         if (unlikely(!cnt))
1169                 return;
1170
1171         if (tx_info->map_linear_data) {
1172                 dma_unmap_single(tx_ring->dev,
1173                                  dma_unmap_addr(ena_buf, paddr),
1174                                  dma_unmap_len(ena_buf, len),
1175                                  DMA_TO_DEVICE);
1176                 ena_buf++;
1177                 cnt--;
1178         }
1179
1180         /* unmap remaining mapped pages */
1181         for (i = 0; i < cnt; i++) {
1182                 dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
1183                                dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
1184                 ena_buf++;
1185         }
1186 }
1187
1188 /* ena_free_tx_bufs - Free Tx Buffers per Queue
1189  * @tx_ring: TX ring for which buffers be freed
1190  */
1191 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
1192 {
1193         bool print_once = true;
1194         u32 i;
1195
1196         for (i = 0; i < tx_ring->ring_size; i++) {
1197                 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1198
1199                 if (!tx_info->skb)
1200                         continue;
1201
1202                 if (print_once) {
1203                         netif_notice(tx_ring->adapter, ifdown, tx_ring->netdev,
1204                                      "Free uncompleted tx skb qid %d idx 0x%x\n",
1205                                      tx_ring->qid, i);
1206                         print_once = false;
1207                 } else {
1208                         netif_dbg(tx_ring->adapter, ifdown, tx_ring->netdev,
1209                                   "Free uncompleted tx skb qid %d idx 0x%x\n",
1210                                   tx_ring->qid, i);
1211                 }
1212
1213                 ena_unmap_tx_buff(tx_ring, tx_info);
1214
1215                 dev_kfree_skb_any(tx_info->skb);
1216         }
1217         netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
1218                                                   tx_ring->qid));
1219 }
1220
1221 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
1222 {
1223         struct ena_ring *tx_ring;
1224         int i;
1225
1226         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1227                 tx_ring = &adapter->tx_ring[i];
1228                 ena_free_tx_bufs(tx_ring);
1229         }
1230 }
1231
1232 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1233 {
1234         u16 ena_qid;
1235         int i;
1236
1237         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1238                 ena_qid = ENA_IO_TXQ_IDX(i);
1239                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1240         }
1241 }
1242
1243 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1244 {
1245         u16 ena_qid;
1246         int i;
1247
1248         for (i = 0; i < adapter->num_io_queues; i++) {
1249                 ena_qid = ENA_IO_RXQ_IDX(i);
1250                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
1251                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1252         }
1253 }
1254
1255 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
1256 {
1257         ena_destroy_all_tx_queues(adapter);
1258         ena_destroy_all_rx_queues(adapter);
1259 }
1260
1261 static int handle_invalid_req_id(struct ena_ring *ring, u16 req_id,
1262                                  struct ena_tx_buffer *tx_info, bool is_xdp)
1263 {
1264         if (tx_info)
1265                 netif_err(ring->adapter,
1266                           tx_done,
1267                           ring->netdev,
1268                           "tx_info doesn't have valid %s. qid %u req_id %u",
1269                            is_xdp ? "xdp frame" : "skb", ring->qid, req_id);
1270         else
1271                 netif_err(ring->adapter,
1272                           tx_done,
1273                           ring->netdev,
1274                           "Invalid req_id %u in qid %u\n",
1275                           req_id, ring->qid);
1276
1277         ena_increase_stat(&ring->tx_stats.bad_req_id, 1, &ring->syncp);
1278         ena_reset_device(ring->adapter, ENA_REGS_RESET_INV_TX_REQ_ID);
1279
1280         return -EFAULT;
1281 }
1282
1283 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
1284 {
1285         struct ena_tx_buffer *tx_info;
1286
1287         tx_info = &tx_ring->tx_buffer_info[req_id];
1288         if (likely(tx_info->skb))
1289                 return 0;
1290
1291         return handle_invalid_req_id(tx_ring, req_id, tx_info, false);
1292 }
1293
1294 static int validate_xdp_req_id(struct ena_ring *xdp_ring, u16 req_id)
1295 {
1296         struct ena_tx_buffer *tx_info;
1297
1298         tx_info = &xdp_ring->tx_buffer_info[req_id];
1299         if (likely(tx_info->xdpf))
1300                 return 0;
1301
1302         return handle_invalid_req_id(xdp_ring, req_id, tx_info, true);
1303 }
1304
1305 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
1306 {
1307         struct netdev_queue *txq;
1308         bool above_thresh;
1309         u32 tx_bytes = 0;
1310         u32 total_done = 0;
1311         u16 next_to_clean;
1312         u16 req_id;
1313         int tx_pkts = 0;
1314         int rc;
1315
1316         next_to_clean = tx_ring->next_to_clean;
1317         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
1318
1319         while (tx_pkts < budget) {
1320                 struct ena_tx_buffer *tx_info;
1321                 struct sk_buff *skb;
1322
1323                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
1324                                                 &req_id);
1325                 if (rc) {
1326                         if (unlikely(rc == -EINVAL))
1327                                 handle_invalid_req_id(tx_ring, req_id, NULL,
1328                                                       false);
1329                         break;
1330                 }
1331
1332                 /* validate that the request id points to a valid skb */
1333                 rc = validate_tx_req_id(tx_ring, req_id);
1334                 if (rc)
1335                         break;
1336
1337                 tx_info = &tx_ring->tx_buffer_info[req_id];
1338                 skb = tx_info->skb;
1339
1340                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
1341                 prefetch(&skb->end);
1342
1343                 tx_info->skb = NULL;
1344                 tx_info->last_jiffies = 0;
1345
1346                 ena_unmap_tx_buff(tx_ring, tx_info);
1347
1348                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1349                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
1350                           skb);
1351
1352                 tx_bytes += skb->len;
1353                 dev_kfree_skb(skb);
1354                 tx_pkts++;
1355                 total_done += tx_info->tx_descs;
1356
1357                 tx_ring->free_ids[next_to_clean] = req_id;
1358                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1359                                                      tx_ring->ring_size);
1360         }
1361
1362         tx_ring->next_to_clean = next_to_clean;
1363         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
1364         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
1365
1366         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
1367
1368         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1369                   "tx_poll: q %d done. total pkts: %d\n",
1370                   tx_ring->qid, tx_pkts);
1371
1372         /* need to make the rings circular update visible to
1373          * ena_start_xmit() before checking for netif_queue_stopped().
1374          */
1375         smp_mb();
1376
1377         above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1378                                                     ENA_TX_WAKEUP_THRESH);
1379         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
1380                 __netif_tx_lock(txq, smp_processor_id());
1381                 above_thresh =
1382                         ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1383                                                      ENA_TX_WAKEUP_THRESH);
1384                 if (netif_tx_queue_stopped(txq) && above_thresh &&
1385                     test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
1386                         netif_tx_wake_queue(txq);
1387                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
1388                                           &tx_ring->syncp);
1389                 }
1390                 __netif_tx_unlock(txq);
1391         }
1392
1393         return tx_pkts;
1394 }
1395
1396 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, void *first_frag)
1397 {
1398         struct sk_buff *skb;
1399
1400         if (!first_frag)
1401                 skb = napi_alloc_skb(rx_ring->napi, rx_ring->rx_copybreak);
1402         else
1403                 skb = napi_build_skb(first_frag, ENA_PAGE_SIZE);
1404
1405         if (unlikely(!skb)) {
1406                 ena_increase_stat(&rx_ring->rx_stats.skb_alloc_fail, 1,
1407                                   &rx_ring->syncp);
1408
1409                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1410                           "Failed to allocate skb. first_frag %s\n",
1411                           first_frag ? "provided" : "not provided");
1412                 return NULL;
1413         }
1414
1415         return skb;
1416 }
1417
1418 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
1419                                   struct ena_com_rx_buf_info *ena_bufs,
1420                                   u32 descs,
1421                                   u16 *next_to_clean)
1422 {
1423         struct ena_rx_buffer *rx_info;
1424         struct ena_adapter *adapter;
1425         u16 len, req_id, buf = 0;
1426         struct sk_buff *skb;
1427         void *page_addr;
1428         u32 page_offset;
1429         void *data_addr;
1430
1431         len = ena_bufs[buf].len;
1432         req_id = ena_bufs[buf].req_id;
1433
1434         rx_info = &rx_ring->rx_buffer_info[req_id];
1435
1436         if (unlikely(!rx_info->page)) {
1437                 adapter = rx_ring->adapter;
1438                 netif_err(adapter, rx_err, rx_ring->netdev,
1439                           "Page is NULL. qid %u req_id %u\n", rx_ring->qid, req_id);
1440                 ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1, &rx_ring->syncp);
1441                 ena_reset_device(adapter, ENA_REGS_RESET_INV_RX_REQ_ID);
1442                 return NULL;
1443         }
1444
1445         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1446                   "rx_info %p page %p\n",
1447                   rx_info, rx_info->page);
1448
1449         /* save virt address of first buffer */
1450         page_addr = page_address(rx_info->page);
1451         page_offset = rx_info->page_offset;
1452         data_addr = page_addr + page_offset;
1453
1454         prefetch(data_addr);
1455
1456         if (len <= rx_ring->rx_copybreak) {
1457                 skb = ena_alloc_skb(rx_ring, NULL);
1458                 if (unlikely(!skb))
1459                         return NULL;
1460
1461                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1462                           "RX allocated small packet. len %d. data_len %d\n",
1463                           skb->len, skb->data_len);
1464
1465                 /* sync this buffer for CPU use */
1466                 dma_sync_single_for_cpu(rx_ring->dev,
1467                                         dma_unmap_addr(&rx_info->ena_buf, paddr),
1468                                         len,
1469                                         DMA_FROM_DEVICE);
1470                 skb_copy_to_linear_data(skb, data_addr, len);
1471                 dma_sync_single_for_device(rx_ring->dev,
1472                                            dma_unmap_addr(&rx_info->ena_buf, paddr),
1473                                            len,
1474                                            DMA_FROM_DEVICE);
1475
1476                 skb_put(skb, len);
1477                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1478                 rx_ring->free_ids[*next_to_clean] = req_id;
1479                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
1480                                                      rx_ring->ring_size);
1481                 return skb;
1482         }
1483
1484         ena_unmap_rx_buff(rx_ring, rx_info);
1485
1486         skb = ena_alloc_skb(rx_ring, page_addr);
1487         if (unlikely(!skb))
1488                 return NULL;
1489
1490         /* Populate skb's linear part */
1491         skb_reserve(skb, page_offset);
1492         skb_put(skb, len);
1493         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1494
1495         do {
1496                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1497                           "RX skb updated. len %d. data_len %d\n",
1498                           skb->len, skb->data_len);
1499
1500                 rx_info->page = NULL;
1501
1502                 rx_ring->free_ids[*next_to_clean] = req_id;
1503                 *next_to_clean =
1504                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
1505                                              rx_ring->ring_size);
1506                 if (likely(--descs == 0))
1507                         break;
1508
1509                 buf++;
1510                 len = ena_bufs[buf].len;
1511                 req_id = ena_bufs[buf].req_id;
1512
1513                 rx_info = &rx_ring->rx_buffer_info[req_id];
1514
1515                 ena_unmap_rx_buff(rx_ring, rx_info);
1516
1517                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
1518                                 rx_info->page_offset, len, ENA_PAGE_SIZE);
1519
1520         } while (1);
1521
1522         return skb;
1523 }
1524
1525 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
1526  * @adapter: structure containing adapter specific data
1527  * @ena_rx_ctx: received packet context/metadata
1528  * @skb: skb currently being received and modified
1529  */
1530 static void ena_rx_checksum(struct ena_ring *rx_ring,
1531                                    struct ena_com_rx_ctx *ena_rx_ctx,
1532                                    struct sk_buff *skb)
1533 {
1534         /* Rx csum disabled */
1535         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
1536                 skb->ip_summed = CHECKSUM_NONE;
1537                 return;
1538         }
1539
1540         /* For fragmented packets the checksum isn't valid */
1541         if (ena_rx_ctx->frag) {
1542                 skb->ip_summed = CHECKSUM_NONE;
1543                 return;
1544         }
1545
1546         /* if IP and error */
1547         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1548                      (ena_rx_ctx->l3_csum_err))) {
1549                 /* ipv4 checksum error */
1550                 skb->ip_summed = CHECKSUM_NONE;
1551                 ena_increase_stat(&rx_ring->rx_stats.csum_bad, 1,
1552                                   &rx_ring->syncp);
1553                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1554                           "RX IPv4 header checksum error\n");
1555                 return;
1556         }
1557
1558         /* if TCP/UDP */
1559         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1560                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
1561                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
1562                         /* TCP/UDP checksum error */
1563                         ena_increase_stat(&rx_ring->rx_stats.csum_bad, 1,
1564                                           &rx_ring->syncp);
1565                         netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1566                                   "RX L4 checksum error\n");
1567                         skb->ip_summed = CHECKSUM_NONE;
1568                         return;
1569                 }
1570
1571                 if (likely(ena_rx_ctx->l4_csum_checked)) {
1572                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1573                         ena_increase_stat(&rx_ring->rx_stats.csum_good, 1,
1574                                           &rx_ring->syncp);
1575                 } else {
1576                         ena_increase_stat(&rx_ring->rx_stats.csum_unchecked, 1,
1577                                           &rx_ring->syncp);
1578                         skb->ip_summed = CHECKSUM_NONE;
1579                 }
1580         } else {
1581                 skb->ip_summed = CHECKSUM_NONE;
1582                 return;
1583         }
1584
1585 }
1586
1587 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1588                             struct ena_com_rx_ctx *ena_rx_ctx,
1589                             struct sk_buff *skb)
1590 {
1591         enum pkt_hash_types hash_type;
1592
1593         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1594                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1595                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1596
1597                         hash_type = PKT_HASH_TYPE_L4;
1598                 else
1599                         hash_type = PKT_HASH_TYPE_NONE;
1600
1601                 /* Override hash type if the packet is fragmented */
1602                 if (ena_rx_ctx->frag)
1603                         hash_type = PKT_HASH_TYPE_NONE;
1604
1605                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1606         }
1607 }
1608
1609 static int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)
1610 {
1611         struct ena_rx_buffer *rx_info;
1612         int ret;
1613
1614         rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1615         xdp_prepare_buff(xdp, page_address(rx_info->page),
1616                          rx_info->page_offset,
1617                          rx_ring->ena_bufs[0].len, false);
1618         /* If for some reason we received a bigger packet than
1619          * we expect, then we simply drop it
1620          */
1621         if (unlikely(rx_ring->ena_bufs[0].len > ENA_XDP_MAX_MTU))
1622                 return XDP_DROP;
1623
1624         ret = ena_xdp_execute(rx_ring, xdp);
1625
1626         /* The xdp program might expand the headers */
1627         if (ret == XDP_PASS) {
1628                 rx_info->page_offset = xdp->data - xdp->data_hard_start;
1629                 rx_ring->ena_bufs[0].len = xdp->data_end - xdp->data;
1630         }
1631
1632         return ret;
1633 }
1634 /* ena_clean_rx_irq - Cleanup RX irq
1635  * @rx_ring: RX ring to clean
1636  * @napi: napi handler
1637  * @budget: how many packets driver is allowed to clean
1638  *
1639  * Returns the number of cleaned buffers.
1640  */
1641 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1642                             u32 budget)
1643 {
1644         u16 next_to_clean = rx_ring->next_to_clean;
1645         struct ena_com_rx_ctx ena_rx_ctx;
1646         struct ena_rx_buffer *rx_info;
1647         struct ena_adapter *adapter;
1648         u32 res_budget, work_done;
1649         int rx_copybreak_pkt = 0;
1650         int refill_threshold;
1651         struct sk_buff *skb;
1652         int refill_required;
1653         struct xdp_buff xdp;
1654         int xdp_flags = 0;
1655         int total_len = 0;
1656         int xdp_verdict;
1657         int rc = 0;
1658         int i;
1659
1660         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1661                   "%s qid %d\n", __func__, rx_ring->qid);
1662         res_budget = budget;
1663         xdp_init_buff(&xdp, ENA_PAGE_SIZE, &rx_ring->xdp_rxq);
1664
1665         do {
1666                 xdp_verdict = XDP_PASS;
1667                 skb = NULL;
1668                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1669                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1670                 ena_rx_ctx.descs = 0;
1671                 ena_rx_ctx.pkt_offset = 0;
1672                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1673                                     rx_ring->ena_com_io_sq,
1674                                     &ena_rx_ctx);
1675                 if (unlikely(rc))
1676                         goto error;
1677
1678                 if (unlikely(ena_rx_ctx.descs == 0))
1679                         break;
1680
1681                 /* First descriptor might have an offset set by the device */
1682                 rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1683                 rx_info->page_offset += ena_rx_ctx.pkt_offset;
1684
1685                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1686                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1687                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1688                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1689
1690                 if (ena_xdp_present_ring(rx_ring))
1691                         xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1692
1693                 /* allocate skb and fill it */
1694                 if (xdp_verdict == XDP_PASS)
1695                         skb = ena_rx_skb(rx_ring,
1696                                          rx_ring->ena_bufs,
1697                                          ena_rx_ctx.descs,
1698                                          &next_to_clean);
1699
1700                 if (unlikely(!skb)) {
1701                         for (i = 0; i < ena_rx_ctx.descs; i++) {
1702                                 int req_id = rx_ring->ena_bufs[i].req_id;
1703
1704                                 rx_ring->free_ids[next_to_clean] = req_id;
1705                                 next_to_clean =
1706                                         ENA_RX_RING_IDX_NEXT(next_to_clean,
1707                                                              rx_ring->ring_size);
1708
1709                                 /* Packets was passed for transmission, unmap it
1710                                  * from RX side.
1711                                  */
1712                                 if (xdp_verdict == XDP_TX || xdp_verdict == XDP_REDIRECT) {
1713                                         ena_unmap_rx_buff(rx_ring,
1714                                                           &rx_ring->rx_buffer_info[req_id]);
1715                                         rx_ring->rx_buffer_info[req_id].page = NULL;
1716                                 }
1717                         }
1718                         if (xdp_verdict != XDP_PASS) {
1719                                 xdp_flags |= xdp_verdict;
1720                                 res_budget--;
1721                                 continue;
1722                         }
1723                         break;
1724                 }
1725
1726                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1727
1728                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1729
1730                 skb_record_rx_queue(skb, rx_ring->qid);
1731
1732                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak)
1733                         rx_copybreak_pkt++;
1734
1735                 total_len += skb->len;
1736
1737                 napi_gro_receive(napi, skb);
1738
1739                 res_budget--;
1740         } while (likely(res_budget));
1741
1742         work_done = budget - res_budget;
1743         rx_ring->per_napi_packets += work_done;
1744         u64_stats_update_begin(&rx_ring->syncp);
1745         rx_ring->rx_stats.bytes += total_len;
1746         rx_ring->rx_stats.cnt += work_done;
1747         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1748         u64_stats_update_end(&rx_ring->syncp);
1749
1750         rx_ring->next_to_clean = next_to_clean;
1751
1752         refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
1753         refill_threshold =
1754                 min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1755                       ENA_RX_REFILL_THRESH_PACKET);
1756
1757         /* Optimization, try to batch new rx buffers */
1758         if (refill_required > refill_threshold) {
1759                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1760                 ena_refill_rx_bufs(rx_ring, refill_required);
1761         }
1762
1763         if (xdp_flags & XDP_REDIRECT)
1764                 xdp_do_flush_map();
1765
1766         return work_done;
1767
1768 error:
1769         adapter = netdev_priv(rx_ring->netdev);
1770
1771         if (rc == -ENOSPC) {
1772                 ena_increase_stat(&rx_ring->rx_stats.bad_desc_num, 1,
1773                                   &rx_ring->syncp);
1774                 ena_reset_device(adapter, ENA_REGS_RESET_TOO_MANY_RX_DESCS);
1775         } else {
1776                 ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1,
1777                                   &rx_ring->syncp);
1778                 ena_reset_device(adapter, ENA_REGS_RESET_INV_RX_REQ_ID);
1779         }
1780         return 0;
1781 }
1782
1783 static void ena_dim_work(struct work_struct *w)
1784 {
1785         struct dim *dim = container_of(w, struct dim, work);
1786         struct dim_cq_moder cur_moder =
1787                 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1788         struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1789
1790         ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1791         dim->state = DIM_START_MEASURE;
1792 }
1793
1794 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1795 {
1796         struct dim_sample dim_sample;
1797         struct ena_ring *rx_ring = ena_napi->rx_ring;
1798
1799         if (!rx_ring->per_napi_packets)
1800                 return;
1801
1802         rx_ring->non_empty_napi_events++;
1803
1804         dim_update_sample(rx_ring->non_empty_napi_events,
1805                           rx_ring->rx_stats.cnt,
1806                           rx_ring->rx_stats.bytes,
1807                           &dim_sample);
1808
1809         net_dim(&ena_napi->dim, dim_sample);
1810
1811         rx_ring->per_napi_packets = 0;
1812 }
1813
1814 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1815                                         struct ena_ring *rx_ring)
1816 {
1817         struct ena_eth_io_intr_reg intr_reg;
1818         u32 rx_interval = 0;
1819         /* Rx ring can be NULL when for XDP tx queues which don't have an
1820          * accompanying rx_ring pair.
1821          */
1822         if (rx_ring)
1823                 rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1824                         rx_ring->smoothed_interval :
1825                         ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1826
1827         /* Update intr register: rx intr delay,
1828          * tx intr delay and interrupt unmask
1829          */
1830         ena_com_update_intr_reg(&intr_reg,
1831                                 rx_interval,
1832                                 tx_ring->smoothed_interval,
1833                                 true);
1834
1835         ena_increase_stat(&tx_ring->tx_stats.unmask_interrupt, 1,
1836                           &tx_ring->syncp);
1837
1838         /* It is a shared MSI-X.
1839          * Tx and Rx CQ have pointer to it.
1840          * So we use one of them to reach the intr reg
1841          * The Tx ring is used because the rx_ring is NULL for XDP queues
1842          */
1843         ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1844 }
1845
1846 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1847                                              struct ena_ring *rx_ring)
1848 {
1849         int cpu = get_cpu();
1850         int numa_node;
1851
1852         /* Check only one ring since the 2 rings are running on the same cpu */
1853         if (likely(tx_ring->cpu == cpu))
1854                 goto out;
1855
1856         numa_node = cpu_to_node(cpu);
1857         put_cpu();
1858
1859         if (numa_node != NUMA_NO_NODE) {
1860                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1861                 if (rx_ring)
1862                         ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1863                                                  numa_node);
1864         }
1865
1866         tx_ring->cpu = cpu;
1867         if (rx_ring)
1868                 rx_ring->cpu = cpu;
1869
1870         return;
1871 out:
1872         put_cpu();
1873 }
1874
1875 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1876 {
1877         u32 total_done = 0;
1878         u16 next_to_clean;
1879         u32 tx_bytes = 0;
1880         int tx_pkts = 0;
1881         u16 req_id;
1882         int rc;
1883
1884         if (unlikely(!xdp_ring))
1885                 return 0;
1886         next_to_clean = xdp_ring->next_to_clean;
1887
1888         while (tx_pkts < budget) {
1889                 struct ena_tx_buffer *tx_info;
1890                 struct xdp_frame *xdpf;
1891
1892                 rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1893                                                 &req_id);
1894                 if (rc) {
1895                         if (unlikely(rc == -EINVAL))
1896                                 handle_invalid_req_id(xdp_ring, req_id, NULL,
1897                                                       true);
1898                         break;
1899                 }
1900
1901                 /* validate that the request id points to a valid xdp_frame */
1902                 rc = validate_xdp_req_id(xdp_ring, req_id);
1903                 if (rc)
1904                         break;
1905
1906                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1907                 xdpf = tx_info->xdpf;
1908
1909                 tx_info->xdpf = NULL;
1910                 tx_info->last_jiffies = 0;
1911                 ena_unmap_tx_buff(xdp_ring, tx_info);
1912
1913                 netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1914                           "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1915                           xdpf);
1916
1917                 tx_bytes += xdpf->len;
1918                 tx_pkts++;
1919                 total_done += tx_info->tx_descs;
1920
1921                 xdp_return_frame(xdpf);
1922                 xdp_ring->free_ids[next_to_clean] = req_id;
1923                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1924                                                      xdp_ring->ring_size);
1925         }
1926
1927         xdp_ring->next_to_clean = next_to_clean;
1928         ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
1929         ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
1930
1931         netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1932                   "tx_poll: q %d done. total pkts: %d\n",
1933                   xdp_ring->qid, tx_pkts);
1934
1935         return tx_pkts;
1936 }
1937
1938 static int ena_io_poll(struct napi_struct *napi, int budget)
1939 {
1940         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1941         struct ena_ring *tx_ring, *rx_ring;
1942         int tx_work_done;
1943         int rx_work_done = 0;
1944         int tx_budget;
1945         int napi_comp_call = 0;
1946         int ret;
1947
1948         tx_ring = ena_napi->tx_ring;
1949         rx_ring = ena_napi->rx_ring;
1950
1951         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1952
1953         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1954             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1955                 napi_complete_done(napi, 0);
1956                 return 0;
1957         }
1958
1959         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1960         /* On netpoll the budget is zero and the handler should only clean the
1961          * tx completions.
1962          */
1963         if (likely(budget))
1964                 rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1965
1966         /* If the device is about to reset or down, avoid unmask
1967          * the interrupt and return 0 so NAPI won't reschedule
1968          */
1969         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1970                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1971                 napi_complete_done(napi, 0);
1972                 ret = 0;
1973
1974         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1975                 napi_comp_call = 1;
1976
1977                 /* Update numa and unmask the interrupt only when schedule
1978                  * from the interrupt context (vs from sk_busy_loop)
1979                  */
1980                 if (napi_complete_done(napi, rx_work_done) &&
1981                     READ_ONCE(ena_napi->interrupts_masked)) {
1982                         smp_rmb(); /* make sure interrupts_masked is read */
1983                         WRITE_ONCE(ena_napi->interrupts_masked, false);
1984                         /* We apply adaptive moderation on Rx path only.
1985                          * Tx uses static interrupt moderation.
1986                          */
1987                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1988                                 ena_adjust_adaptive_rx_intr_moderation(ena_napi);
1989
1990                         ena_unmask_interrupt(tx_ring, rx_ring);
1991                 }
1992
1993                 ena_update_ring_numa_node(tx_ring, rx_ring);
1994
1995                 ret = rx_work_done;
1996         } else {
1997                 ret = budget;
1998         }
1999
2000         u64_stats_update_begin(&tx_ring->syncp);
2001         tx_ring->tx_stats.napi_comp += napi_comp_call;
2002         tx_ring->tx_stats.tx_poll++;
2003         u64_stats_update_end(&tx_ring->syncp);
2004
2005         tx_ring->tx_stats.last_napi_jiffies = jiffies;
2006
2007         return ret;
2008 }
2009
2010 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
2011 {
2012         struct ena_adapter *adapter = (struct ena_adapter *)data;
2013
2014         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
2015
2016         /* Don't call the aenq handler before probe is done */
2017         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
2018                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
2019
2020         return IRQ_HANDLED;
2021 }
2022
2023 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
2024  * @irq: interrupt number
2025  * @data: pointer to a network interface private napi device structure
2026  */
2027 static irqreturn_t ena_intr_msix_io(int irq, void *data)
2028 {
2029         struct ena_napi *ena_napi = data;
2030
2031         /* Used to check HW health */
2032         WRITE_ONCE(ena_napi->first_interrupt, true);
2033
2034         WRITE_ONCE(ena_napi->interrupts_masked, true);
2035         smp_wmb(); /* write interrupts_masked before calling napi */
2036
2037         napi_schedule_irqoff(&ena_napi->napi);
2038
2039         return IRQ_HANDLED;
2040 }
2041
2042 /* Reserve a single MSI-X vector for management (admin + aenq).
2043  * plus reserve one vector for each potential io queue.
2044  * the number of potential io queues is the minimum of what the device
2045  * supports and the number of vCPUs.
2046  */
2047 static int ena_enable_msix(struct ena_adapter *adapter)
2048 {
2049         int msix_vecs, irq_cnt;
2050
2051         if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2052                 netif_err(adapter, probe, adapter->netdev,
2053                           "Error, MSI-X is already enabled\n");
2054                 return -EPERM;
2055         }
2056
2057         /* Reserved the max msix vectors we might need */
2058         msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
2059         netif_dbg(adapter, probe, adapter->netdev,
2060                   "Trying to enable MSI-X, vectors %d\n", msix_vecs);
2061
2062         irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
2063                                         msix_vecs, PCI_IRQ_MSIX);
2064
2065         if (irq_cnt < 0) {
2066                 netif_err(adapter, probe, adapter->netdev,
2067                           "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
2068                 return -ENOSPC;
2069         }
2070
2071         if (irq_cnt != msix_vecs) {
2072                 netif_notice(adapter, probe, adapter->netdev,
2073                              "Enable only %d MSI-X (out of %d), reduce the number of queues\n",
2074                              irq_cnt, msix_vecs);
2075                 adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
2076         }
2077
2078         if (ena_init_rx_cpu_rmap(adapter))
2079                 netif_warn(adapter, probe, adapter->netdev,
2080                            "Failed to map IRQs to CPUs\n");
2081
2082         adapter->msix_vecs = irq_cnt;
2083         set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2084
2085         return 0;
2086 }
2087
2088 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2089 {
2090         u32 cpu;
2091
2092         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2093                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2094                  pci_name(adapter->pdev));
2095         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2096                 ena_intr_msix_mgmnt;
2097         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2098         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2099                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2100         cpu = cpumask_first(cpu_online_mask);
2101         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2102         cpumask_set_cpu(cpu,
2103                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2104 }
2105
2106 static void ena_setup_io_intr(struct ena_adapter *adapter)
2107 {
2108         struct net_device *netdev;
2109         int irq_idx, i, cpu;
2110         int io_queue_count;
2111
2112         netdev = adapter->netdev;
2113         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2114
2115         for (i = 0; i < io_queue_count; i++) {
2116                 irq_idx = ENA_IO_IRQ_IDX(i);
2117                 cpu = i % num_online_cpus();
2118
2119                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2120                          "%s-Tx-Rx-%d", netdev->name, i);
2121                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2122                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2123                 adapter->irq_tbl[irq_idx].vector =
2124                         pci_irq_vector(adapter->pdev, irq_idx);
2125                 adapter->irq_tbl[irq_idx].cpu = cpu;
2126
2127                 cpumask_set_cpu(cpu,
2128                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
2129         }
2130 }
2131
2132 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2133 {
2134         unsigned long flags = 0;
2135         struct ena_irq *irq;
2136         int rc;
2137
2138         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2139         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2140                          irq->data);
2141         if (rc) {
2142                 netif_err(adapter, probe, adapter->netdev,
2143                           "Failed to request admin irq\n");
2144                 return rc;
2145         }
2146
2147         netif_dbg(adapter, probe, adapter->netdev,
2148                   "Set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2149                   irq->affinity_hint_mask.bits[0], irq->vector);
2150
2151         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2152
2153         return rc;
2154 }
2155
2156 static int ena_request_io_irq(struct ena_adapter *adapter)
2157 {
2158         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2159         unsigned long flags = 0;
2160         struct ena_irq *irq;
2161         int rc = 0, i, k;
2162
2163         if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2164                 netif_err(adapter, ifup, adapter->netdev,
2165                           "Failed to request I/O IRQ: MSI-X is not enabled\n");
2166                 return -EINVAL;
2167         }
2168
2169         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2170                 irq = &adapter->irq_tbl[i];
2171                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2172                                  irq->data);
2173                 if (rc) {
2174                         netif_err(adapter, ifup, adapter->netdev,
2175                                   "Failed to request I/O IRQ. index %d rc %d\n",
2176                                    i, rc);
2177                         goto err;
2178                 }
2179
2180                 netif_dbg(adapter, ifup, adapter->netdev,
2181                           "Set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2182                           i, irq->affinity_hint_mask.bits[0], irq->vector);
2183
2184                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2185         }
2186
2187         return rc;
2188
2189 err:
2190         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2191                 irq = &adapter->irq_tbl[k];
2192                 free_irq(irq->vector, irq->data);
2193         }
2194
2195         return rc;
2196 }
2197
2198 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2199 {
2200         struct ena_irq *irq;
2201
2202         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2203         synchronize_irq(irq->vector);
2204         irq_set_affinity_hint(irq->vector, NULL);
2205         free_irq(irq->vector, irq->data);
2206 }
2207
2208 static void ena_free_io_irq(struct ena_adapter *adapter)
2209 {
2210         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2211         struct ena_irq *irq;
2212         int i;
2213
2214 #ifdef CONFIG_RFS_ACCEL
2215         if (adapter->msix_vecs >= 1) {
2216                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2217                 adapter->netdev->rx_cpu_rmap = NULL;
2218         }
2219 #endif /* CONFIG_RFS_ACCEL */
2220
2221         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2222                 irq = &adapter->irq_tbl[i];
2223                 irq_set_affinity_hint(irq->vector, NULL);
2224                 free_irq(irq->vector, irq->data);
2225         }
2226 }
2227
2228 static void ena_disable_msix(struct ena_adapter *adapter)
2229 {
2230         if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2231                 pci_free_irq_vectors(adapter->pdev);
2232 }
2233
2234 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2235 {
2236         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2237         int i;
2238
2239         if (!netif_running(adapter->netdev))
2240                 return;
2241
2242         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2243                 synchronize_irq(adapter->irq_tbl[i].vector);
2244 }
2245
2246 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2247                                   int first_index,
2248                                   int count)
2249 {
2250         int i;
2251
2252         for (i = first_index; i < first_index + count; i++) {
2253                 netif_napi_del(&adapter->ena_napi[i].napi);
2254
2255                 WARN_ON(!ENA_IS_XDP_INDEX(adapter, i) &&
2256                         adapter->ena_napi[i].xdp_ring);
2257         }
2258 }
2259
2260 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2261                                    int first_index, int count)
2262 {
2263         int i;
2264
2265         for (i = first_index; i < first_index + count; i++) {
2266                 struct ena_napi *napi = &adapter->ena_napi[i];
2267
2268                 netif_napi_add(adapter->netdev,
2269                                &napi->napi,
2270                                ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll,
2271                                NAPI_POLL_WEIGHT);
2272
2273                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
2274                         napi->rx_ring = &adapter->rx_ring[i];
2275                         napi->tx_ring = &adapter->tx_ring[i];
2276                 } else {
2277                         napi->xdp_ring = &adapter->tx_ring[i];
2278                 }
2279                 napi->qid = i;
2280         }
2281 }
2282
2283 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2284                                       int first_index,
2285                                       int count)
2286 {
2287         int i;
2288
2289         for (i = first_index; i < first_index + count; i++)
2290                 napi_disable(&adapter->ena_napi[i].napi);
2291 }
2292
2293 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2294                                      int first_index,
2295                                      int count)
2296 {
2297         int i;
2298
2299         for (i = first_index; i < first_index + count; i++)
2300                 napi_enable(&adapter->ena_napi[i].napi);
2301 }
2302
2303 /* Configure the Rx forwarding */
2304 static int ena_rss_configure(struct ena_adapter *adapter)
2305 {
2306         struct ena_com_dev *ena_dev = adapter->ena_dev;
2307         int rc;
2308
2309         /* In case the RSS table wasn't initialized by probe */
2310         if (!ena_dev->rss.tbl_log_size) {
2311                 rc = ena_rss_init_default(adapter);
2312                 if (rc && (rc != -EOPNOTSUPP)) {
2313                         netif_err(adapter, ifup, adapter->netdev,
2314                                   "Failed to init RSS rc: %d\n", rc);
2315                         return rc;
2316                 }
2317         }
2318
2319         /* Set indirect table */
2320         rc = ena_com_indirect_table_set(ena_dev);
2321         if (unlikely(rc && rc != -EOPNOTSUPP))
2322                 return rc;
2323
2324         /* Configure hash function (if supported) */
2325         rc = ena_com_set_hash_function(ena_dev);
2326         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2327                 return rc;
2328
2329         /* Configure hash inputs (if supported) */
2330         rc = ena_com_set_hash_ctrl(ena_dev);
2331         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2332                 return rc;
2333
2334         return 0;
2335 }
2336
2337 static int ena_up_complete(struct ena_adapter *adapter)
2338 {
2339         int rc;
2340
2341         rc = ena_rss_configure(adapter);
2342         if (rc)
2343                 return rc;
2344
2345         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2346
2347         ena_refill_all_rx_bufs(adapter);
2348
2349         /* enable transmits */
2350         netif_tx_start_all_queues(adapter->netdev);
2351
2352         ena_napi_enable_in_range(adapter,
2353                                  0,
2354                                  adapter->xdp_num_queues + adapter->num_io_queues);
2355
2356         return 0;
2357 }
2358
2359 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2360 {
2361         struct ena_com_create_io_ctx ctx;
2362         struct ena_com_dev *ena_dev;
2363         struct ena_ring *tx_ring;
2364         u32 msix_vector;
2365         u16 ena_qid;
2366         int rc;
2367
2368         ena_dev = adapter->ena_dev;
2369
2370         tx_ring = &adapter->tx_ring[qid];
2371         msix_vector = ENA_IO_IRQ_IDX(qid);
2372         ena_qid = ENA_IO_TXQ_IDX(qid);
2373
2374         memset(&ctx, 0x0, sizeof(ctx));
2375
2376         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2377         ctx.qid = ena_qid;
2378         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2379         ctx.msix_vector = msix_vector;
2380         ctx.queue_size = tx_ring->ring_size;
2381         ctx.numa_node = cpu_to_node(tx_ring->cpu);
2382
2383         rc = ena_com_create_io_queue(ena_dev, &ctx);
2384         if (rc) {
2385                 netif_err(adapter, ifup, adapter->netdev,
2386                           "Failed to create I/O TX queue num %d rc: %d\n",
2387                           qid, rc);
2388                 return rc;
2389         }
2390
2391         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2392                                      &tx_ring->ena_com_io_sq,
2393                                      &tx_ring->ena_com_io_cq);
2394         if (rc) {
2395                 netif_err(adapter, ifup, adapter->netdev,
2396                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2397                           qid, rc);
2398                 ena_com_destroy_io_queue(ena_dev, ena_qid);
2399                 return rc;
2400         }
2401
2402         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2403         return rc;
2404 }
2405
2406 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2407                                             int first_index, int count)
2408 {
2409         struct ena_com_dev *ena_dev = adapter->ena_dev;
2410         int rc, i;
2411
2412         for (i = first_index; i < first_index + count; i++) {
2413                 rc = ena_create_io_tx_queue(adapter, i);
2414                 if (rc)
2415                         goto create_err;
2416         }
2417
2418         return 0;
2419
2420 create_err:
2421         while (i-- > first_index)
2422                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2423
2424         return rc;
2425 }
2426
2427 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2428 {
2429         struct ena_com_dev *ena_dev;
2430         struct ena_com_create_io_ctx ctx;
2431         struct ena_ring *rx_ring;
2432         u32 msix_vector;
2433         u16 ena_qid;
2434         int rc;
2435
2436         ena_dev = adapter->ena_dev;
2437
2438         rx_ring = &adapter->rx_ring[qid];
2439         msix_vector = ENA_IO_IRQ_IDX(qid);
2440         ena_qid = ENA_IO_RXQ_IDX(qid);
2441
2442         memset(&ctx, 0x0, sizeof(ctx));
2443
2444         ctx.qid = ena_qid;
2445         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2446         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2447         ctx.msix_vector = msix_vector;
2448         ctx.queue_size = rx_ring->ring_size;
2449         ctx.numa_node = cpu_to_node(rx_ring->cpu);
2450
2451         rc = ena_com_create_io_queue(ena_dev, &ctx);
2452         if (rc) {
2453                 netif_err(adapter, ifup, adapter->netdev,
2454                           "Failed to create I/O RX queue num %d rc: %d\n",
2455                           qid, rc);
2456                 return rc;
2457         }
2458
2459         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2460                                      &rx_ring->ena_com_io_sq,
2461                                      &rx_ring->ena_com_io_cq);
2462         if (rc) {
2463                 netif_err(adapter, ifup, adapter->netdev,
2464                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2465                           qid, rc);
2466                 goto err;
2467         }
2468
2469         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2470
2471         return rc;
2472 err:
2473         ena_com_destroy_io_queue(ena_dev, ena_qid);
2474         return rc;
2475 }
2476
2477 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2478 {
2479         struct ena_com_dev *ena_dev = adapter->ena_dev;
2480         int rc, i;
2481
2482         for (i = 0; i < adapter->num_io_queues; i++) {
2483                 rc = ena_create_io_rx_queue(adapter, i);
2484                 if (rc)
2485                         goto create_err;
2486                 INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2487         }
2488
2489         return 0;
2490
2491 create_err:
2492         while (i--) {
2493                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
2494                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2495         }
2496
2497         return rc;
2498 }
2499
2500 static void set_io_rings_size(struct ena_adapter *adapter,
2501                               int new_tx_size,
2502                               int new_rx_size)
2503 {
2504         int i;
2505
2506         for (i = 0; i < adapter->num_io_queues; i++) {
2507                 adapter->tx_ring[i].ring_size = new_tx_size;
2508                 adapter->rx_ring[i].ring_size = new_rx_size;
2509         }
2510 }
2511
2512 /* This function allows queue allocation to backoff when the system is
2513  * low on memory. If there is not enough memory to allocate io queues
2514  * the driver will try to allocate smaller queues.
2515  *
2516  * The backoff algorithm is as follows:
2517  *  1. Try to allocate TX and RX and if successful.
2518  *  1.1. return success
2519  *
2520  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2521  *
2522  *  3. If TX or RX is smaller than 256
2523  *  3.1. return failure.
2524  *  4. else
2525  *  4.1. go back to 1.
2526  */
2527 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2528 {
2529         int rc, cur_rx_ring_size, cur_tx_ring_size;
2530         int new_rx_ring_size, new_tx_ring_size;
2531
2532         /* current queue sizes might be set to smaller than the requested
2533          * ones due to past queue allocation failures.
2534          */
2535         set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2536                           adapter->requested_rx_ring_size);
2537
2538         while (1) {
2539                 if (ena_xdp_present(adapter)) {
2540                         rc = ena_setup_and_create_all_xdp_queues(adapter);
2541
2542                         if (rc)
2543                                 goto err_setup_tx;
2544                 }
2545                 rc = ena_setup_tx_resources_in_range(adapter,
2546                                                      0,
2547                                                      adapter->num_io_queues);
2548                 if (rc)
2549                         goto err_setup_tx;
2550
2551                 rc = ena_create_io_tx_queues_in_range(adapter,
2552                                                       0,
2553                                                       adapter->num_io_queues);
2554                 if (rc)
2555                         goto err_create_tx_queues;
2556
2557                 rc = ena_setup_all_rx_resources(adapter);
2558                 if (rc)
2559                         goto err_setup_rx;
2560
2561                 rc = ena_create_all_io_rx_queues(adapter);
2562                 if (rc)
2563                         goto err_create_rx_queues;
2564
2565                 return 0;
2566
2567 err_create_rx_queues:
2568                 ena_free_all_io_rx_resources(adapter);
2569 err_setup_rx:
2570                 ena_destroy_all_tx_queues(adapter);
2571 err_create_tx_queues:
2572                 ena_free_all_io_tx_resources(adapter);
2573 err_setup_tx:
2574                 if (rc != -ENOMEM) {
2575                         netif_err(adapter, ifup, adapter->netdev,
2576                                   "Queue creation failed with error code %d\n",
2577                                   rc);
2578                         return rc;
2579                 }
2580
2581                 cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2582                 cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2583
2584                 netif_err(adapter, ifup, adapter->netdev,
2585                           "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2586                           cur_tx_ring_size, cur_rx_ring_size);
2587
2588                 new_tx_ring_size = cur_tx_ring_size;
2589                 new_rx_ring_size = cur_rx_ring_size;
2590
2591                 /* Decrease the size of the larger queue, or
2592                  * decrease both if they are the same size.
2593                  */
2594                 if (cur_rx_ring_size <= cur_tx_ring_size)
2595                         new_tx_ring_size = cur_tx_ring_size / 2;
2596                 if (cur_rx_ring_size >= cur_tx_ring_size)
2597                         new_rx_ring_size = cur_rx_ring_size / 2;
2598
2599                 if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2600                     new_rx_ring_size < ENA_MIN_RING_SIZE) {
2601                         netif_err(adapter, ifup, adapter->netdev,
2602                                   "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2603                                   ENA_MIN_RING_SIZE);
2604                         return rc;
2605                 }
2606
2607                 netif_err(adapter, ifup, adapter->netdev,
2608                           "Retrying queue creation with sizes TX=%d, RX=%d\n",
2609                           new_tx_ring_size,
2610                           new_rx_ring_size);
2611
2612                 set_io_rings_size(adapter, new_tx_ring_size,
2613                                   new_rx_ring_size);
2614         }
2615 }
2616
2617 static int ena_up(struct ena_adapter *adapter)
2618 {
2619         int io_queue_count, rc, i;
2620
2621         netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
2622
2623         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2624         ena_setup_io_intr(adapter);
2625
2626         /* napi poll functions should be initialized before running
2627          * request_irq(), to handle a rare condition where there is a pending
2628          * interrupt, causing the ISR to fire immediately while the poll
2629          * function wasn't set yet, causing a null dereference
2630          */
2631         ena_init_napi_in_range(adapter, 0, io_queue_count);
2632
2633         rc = ena_request_io_irq(adapter);
2634         if (rc)
2635                 goto err_req_irq;
2636
2637         rc = create_queues_with_size_backoff(adapter);
2638         if (rc)
2639                 goto err_create_queues_with_backoff;
2640
2641         rc = ena_up_complete(adapter);
2642         if (rc)
2643                 goto err_up;
2644
2645         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2646                 netif_carrier_on(adapter->netdev);
2647
2648         ena_increase_stat(&adapter->dev_stats.interface_up, 1,
2649                           &adapter->syncp);
2650
2651         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2652
2653         /* Enable completion queues interrupt */
2654         for (i = 0; i < adapter->num_io_queues; i++)
2655                 ena_unmask_interrupt(&adapter->tx_ring[i],
2656                                      &adapter->rx_ring[i]);
2657
2658         /* schedule napi in case we had pending packets
2659          * from the last time we disable napi
2660          */
2661         for (i = 0; i < io_queue_count; i++)
2662                 napi_schedule(&adapter->ena_napi[i].napi);
2663
2664         return rc;
2665
2666 err_up:
2667         ena_destroy_all_tx_queues(adapter);
2668         ena_free_all_io_tx_resources(adapter);
2669         ena_destroy_all_rx_queues(adapter);
2670         ena_free_all_io_rx_resources(adapter);
2671 err_create_queues_with_backoff:
2672         ena_free_io_irq(adapter);
2673 err_req_irq:
2674         ena_del_napi_in_range(adapter, 0, io_queue_count);
2675
2676         return rc;
2677 }
2678
2679 static void ena_down(struct ena_adapter *adapter)
2680 {
2681         int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2682
2683         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2684
2685         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2686
2687         ena_increase_stat(&adapter->dev_stats.interface_down, 1,
2688                           &adapter->syncp);
2689
2690         netif_carrier_off(adapter->netdev);
2691         netif_tx_disable(adapter->netdev);
2692
2693         /* After this point the napi handler won't enable the tx queue */
2694         ena_napi_disable_in_range(adapter, 0, io_queue_count);
2695
2696         /* After destroy the queue there won't be any new interrupts */
2697
2698         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2699                 int rc;
2700
2701                 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2702                 if (rc)
2703                         netif_err(adapter, ifdown, adapter->netdev,
2704                                   "Device reset failed\n");
2705                 /* stop submitting admin commands on a device that was reset */
2706                 ena_com_set_admin_running_state(adapter->ena_dev, false);
2707         }
2708
2709         ena_destroy_all_io_queues(adapter);
2710
2711         ena_disable_io_intr_sync(adapter);
2712         ena_free_io_irq(adapter);
2713         ena_del_napi_in_range(adapter, 0, io_queue_count);
2714
2715         ena_free_all_tx_bufs(adapter);
2716         ena_free_all_rx_bufs(adapter);
2717         ena_free_all_io_tx_resources(adapter);
2718         ena_free_all_io_rx_resources(adapter);
2719 }
2720
2721 /* ena_open - Called when a network interface is made active
2722  * @netdev: network interface device structure
2723  *
2724  * Returns 0 on success, negative value on failure
2725  *
2726  * The open entry point is called when a network interface is made
2727  * active by the system (IFF_UP).  At this point all resources needed
2728  * for transmit and receive operations are allocated, the interrupt
2729  * handler is registered with the OS, the watchdog timer is started,
2730  * and the stack is notified that the interface is ready.
2731  */
2732 static int ena_open(struct net_device *netdev)
2733 {
2734         struct ena_adapter *adapter = netdev_priv(netdev);
2735         int rc;
2736
2737         /* Notify the stack of the actual queue counts. */
2738         rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2739         if (rc) {
2740                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2741                 return rc;
2742         }
2743
2744         rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2745         if (rc) {
2746                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2747                 return rc;
2748         }
2749
2750         rc = ena_up(adapter);
2751         if (rc)
2752                 return rc;
2753
2754         return rc;
2755 }
2756
2757 /* ena_close - Disables a network interface
2758  * @netdev: network interface device structure
2759  *
2760  * Returns 0, this is not allowed to fail
2761  *
2762  * The close entry point is called when an interface is de-activated
2763  * by the OS.  The hardware is still under the drivers control, but
2764  * needs to be disabled.  A global MAC reset is issued to stop the
2765  * hardware, and all transmit and receive resources are freed.
2766  */
2767 static int ena_close(struct net_device *netdev)
2768 {
2769         struct ena_adapter *adapter = netdev_priv(netdev);
2770
2771         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2772
2773         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2774                 return 0;
2775
2776         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2777                 ena_down(adapter);
2778
2779         /* Check for device status and issue reset if needed*/
2780         check_for_admin_com_state(adapter);
2781         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2782                 netif_err(adapter, ifdown, adapter->netdev,
2783                           "Destroy failure, restarting device\n");
2784                 ena_dump_stats_to_dmesg(adapter);
2785                 /* rtnl lock already obtained in dev_ioctl() layer */
2786                 ena_destroy_device(adapter, false);
2787                 ena_restore_device(adapter);
2788         }
2789
2790         return 0;
2791 }
2792
2793 int ena_update_queue_sizes(struct ena_adapter *adapter,
2794                            u32 new_tx_size,
2795                            u32 new_rx_size)
2796 {
2797         bool dev_was_up;
2798
2799         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2800         ena_close(adapter->netdev);
2801         adapter->requested_tx_ring_size = new_tx_size;
2802         adapter->requested_rx_ring_size = new_rx_size;
2803         ena_init_io_rings(adapter,
2804                           0,
2805                           adapter->xdp_num_queues +
2806                           adapter->num_io_queues);
2807         return dev_was_up ? ena_up(adapter) : 0;
2808 }
2809
2810 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2811 {
2812         struct ena_com_dev *ena_dev = adapter->ena_dev;
2813         int prev_channel_count;
2814         bool dev_was_up;
2815
2816         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2817         ena_close(adapter->netdev);
2818         prev_channel_count = adapter->num_io_queues;
2819         adapter->num_io_queues = new_channel_count;
2820         if (ena_xdp_present(adapter) &&
2821             ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2822                 adapter->xdp_first_ring = new_channel_count;
2823                 adapter->xdp_num_queues = new_channel_count;
2824                 if (prev_channel_count > new_channel_count)
2825                         ena_xdp_exchange_program_rx_in_range(adapter,
2826                                                              NULL,
2827                                                              new_channel_count,
2828                                                              prev_channel_count);
2829                 else
2830                         ena_xdp_exchange_program_rx_in_range(adapter,
2831                                                              adapter->xdp_bpf_prog,
2832                                                              prev_channel_count,
2833                                                              new_channel_count);
2834         }
2835
2836         /* We need to destroy the rss table so that the indirection
2837          * table will be reinitialized by ena_up()
2838          */
2839         ena_com_rss_destroy(ena_dev);
2840         ena_init_io_rings(adapter,
2841                           0,
2842                           adapter->xdp_num_queues +
2843                           adapter->num_io_queues);
2844         return dev_was_up ? ena_open(adapter->netdev) : 0;
2845 }
2846
2847 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx,
2848                         struct sk_buff *skb,
2849                         bool disable_meta_caching)
2850 {
2851         u32 mss = skb_shinfo(skb)->gso_size;
2852         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2853         u8 l4_protocol = 0;
2854
2855         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2856                 ena_tx_ctx->l4_csum_enable = 1;
2857                 if (mss) {
2858                         ena_tx_ctx->tso_enable = 1;
2859                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2860                         ena_tx_ctx->l4_csum_partial = 0;
2861                 } else {
2862                         ena_tx_ctx->tso_enable = 0;
2863                         ena_meta->l4_hdr_len = 0;
2864                         ena_tx_ctx->l4_csum_partial = 1;
2865                 }
2866
2867                 switch (ip_hdr(skb)->version) {
2868                 case IPVERSION:
2869                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2870                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
2871                                 ena_tx_ctx->df = 1;
2872                         if (mss)
2873                                 ena_tx_ctx->l3_csum_enable = 1;
2874                         l4_protocol = ip_hdr(skb)->protocol;
2875                         break;
2876                 case 6:
2877                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2878                         l4_protocol = ipv6_hdr(skb)->nexthdr;
2879                         break;
2880                 default:
2881                         break;
2882                 }
2883
2884                 if (l4_protocol == IPPROTO_TCP)
2885                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2886                 else
2887                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2888
2889                 ena_meta->mss = mss;
2890                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
2891                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
2892                 ena_tx_ctx->meta_valid = 1;
2893         } else if (disable_meta_caching) {
2894                 memset(ena_meta, 0, sizeof(*ena_meta));
2895                 ena_tx_ctx->meta_valid = 1;
2896         } else {
2897                 ena_tx_ctx->meta_valid = 0;
2898         }
2899 }
2900
2901 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2902                                        struct sk_buff *skb)
2903 {
2904         int num_frags, header_len, rc;
2905
2906         num_frags = skb_shinfo(skb)->nr_frags;
2907         header_len = skb_headlen(skb);
2908
2909         if (num_frags < tx_ring->sgl_size)
2910                 return 0;
2911
2912         if ((num_frags == tx_ring->sgl_size) &&
2913             (header_len < tx_ring->tx_max_header_size))
2914                 return 0;
2915
2916         ena_increase_stat(&tx_ring->tx_stats.linearize, 1, &tx_ring->syncp);
2917
2918         rc = skb_linearize(skb);
2919         if (unlikely(rc)) {
2920                 ena_increase_stat(&tx_ring->tx_stats.linearize_failed, 1,
2921                                   &tx_ring->syncp);
2922         }
2923
2924         return rc;
2925 }
2926
2927 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2928                           struct ena_tx_buffer *tx_info,
2929                           struct sk_buff *skb,
2930                           void **push_hdr,
2931                           u16 *header_len)
2932 {
2933         struct ena_adapter *adapter = tx_ring->adapter;
2934         struct ena_com_buf *ena_buf;
2935         dma_addr_t dma;
2936         u32 skb_head_len, frag_len, last_frag;
2937         u16 push_len = 0;
2938         u16 delta = 0;
2939         int i = 0;
2940
2941         skb_head_len = skb_headlen(skb);
2942         tx_info->skb = skb;
2943         ena_buf = tx_info->bufs;
2944
2945         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2946                 /* When the device is LLQ mode, the driver will copy
2947                  * the header into the device memory space.
2948                  * the ena_com layer assume the header is in a linear
2949                  * memory space.
2950                  * This assumption might be wrong since part of the header
2951                  * can be in the fragmented buffers.
2952                  * Use skb_header_pointer to make sure the header is in a
2953                  * linear memory space.
2954                  */
2955
2956                 push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2957                 *push_hdr = skb_header_pointer(skb, 0, push_len,
2958                                                tx_ring->push_buf_intermediate_buf);
2959                 *header_len = push_len;
2960                 if (unlikely(skb->data != *push_hdr)) {
2961                         ena_increase_stat(&tx_ring->tx_stats.llq_buffer_copy, 1,
2962                                           &tx_ring->syncp);
2963
2964                         delta = push_len - skb_head_len;
2965                 }
2966         } else {
2967                 *push_hdr = NULL;
2968                 *header_len = min_t(u32, skb_head_len,
2969                                     tx_ring->tx_max_header_size);
2970         }
2971
2972         netif_dbg(adapter, tx_queued, adapter->netdev,
2973                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2974                   *push_hdr, push_len);
2975
2976         if (skb_head_len > push_len) {
2977                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2978                                      skb_head_len - push_len, DMA_TO_DEVICE);
2979                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2980                         goto error_report_dma_error;
2981
2982                 ena_buf->paddr = dma;
2983                 ena_buf->len = skb_head_len - push_len;
2984
2985                 ena_buf++;
2986                 tx_info->num_of_bufs++;
2987                 tx_info->map_linear_data = 1;
2988         } else {
2989                 tx_info->map_linear_data = 0;
2990         }
2991
2992         last_frag = skb_shinfo(skb)->nr_frags;
2993
2994         for (i = 0; i < last_frag; i++) {
2995                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2996
2997                 frag_len = skb_frag_size(frag);
2998
2999                 if (unlikely(delta >= frag_len)) {
3000                         delta -= frag_len;
3001                         continue;
3002                 }
3003
3004                 dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
3005                                        frag_len - delta, DMA_TO_DEVICE);
3006                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
3007                         goto error_report_dma_error;
3008
3009                 ena_buf->paddr = dma;
3010                 ena_buf->len = frag_len - delta;
3011                 ena_buf++;
3012                 tx_info->num_of_bufs++;
3013                 delta = 0;
3014         }
3015
3016         return 0;
3017
3018 error_report_dma_error:
3019         ena_increase_stat(&tx_ring->tx_stats.dma_mapping_err, 1,
3020                           &tx_ring->syncp);
3021         netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map skb\n");
3022
3023         tx_info->skb = NULL;
3024
3025         tx_info->num_of_bufs += i;
3026         ena_unmap_tx_buff(tx_ring, tx_info);
3027
3028         return -EINVAL;
3029 }
3030
3031 /* Called with netif_tx_lock. */
3032 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
3033 {
3034         struct ena_adapter *adapter = netdev_priv(dev);
3035         struct ena_tx_buffer *tx_info;
3036         struct ena_com_tx_ctx ena_tx_ctx;
3037         struct ena_ring *tx_ring;
3038         struct netdev_queue *txq;
3039         void *push_hdr;
3040         u16 next_to_use, req_id, header_len;
3041         int qid, rc;
3042
3043         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
3044         /*  Determine which tx ring we will be placed on */
3045         qid = skb_get_queue_mapping(skb);
3046         tx_ring = &adapter->tx_ring[qid];
3047         txq = netdev_get_tx_queue(dev, qid);
3048
3049         rc = ena_check_and_linearize_skb(tx_ring, skb);
3050         if (unlikely(rc))
3051                 goto error_drop_packet;
3052
3053         skb_tx_timestamp(skb);
3054
3055         next_to_use = tx_ring->next_to_use;
3056         req_id = tx_ring->free_ids[next_to_use];
3057         tx_info = &tx_ring->tx_buffer_info[req_id];
3058         tx_info->num_of_bufs = 0;
3059
3060         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
3061
3062         rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
3063         if (unlikely(rc))
3064                 goto error_drop_packet;
3065
3066         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
3067         ena_tx_ctx.ena_bufs = tx_info->bufs;
3068         ena_tx_ctx.push_header = push_hdr;
3069         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
3070         ena_tx_ctx.req_id = req_id;
3071         ena_tx_ctx.header_len = header_len;
3072
3073         /* set flags and meta data */
3074         ena_tx_csum(&ena_tx_ctx, skb, tx_ring->disable_meta_caching);
3075
3076         rc = ena_xmit_common(dev,
3077                              tx_ring,
3078                              tx_info,
3079                              &ena_tx_ctx,
3080                              next_to_use,
3081                              skb->len);
3082         if (rc)
3083                 goto error_unmap_dma;
3084
3085         netdev_tx_sent_queue(txq, skb->len);
3086
3087         /* stop the queue when no more space available, the packet can have up
3088          * to sgl_size + 2. one for the meta descriptor and one for header
3089          * (if the header is larger than tx_max_header_size).
3090          */
3091         if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3092                                                    tx_ring->sgl_size + 2))) {
3093                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3094                           __func__, qid);
3095
3096                 netif_tx_stop_queue(txq);
3097                 ena_increase_stat(&tx_ring->tx_stats.queue_stop, 1,
3098                                   &tx_ring->syncp);
3099
3100                 /* There is a rare condition where this function decide to
3101                  * stop the queue but meanwhile clean_tx_irq updates
3102                  * next_to_completion and terminates.
3103                  * The queue will remain stopped forever.
3104                  * To solve this issue add a mb() to make sure that
3105                  * netif_tx_stop_queue() write is vissible before checking if
3106                  * there is additional space in the queue.
3107                  */
3108                 smp_mb();
3109
3110                 if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3111                                                  ENA_TX_WAKEUP_THRESH)) {
3112                         netif_tx_wake_queue(txq);
3113                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
3114                                           &tx_ring->syncp);
3115                 }
3116         }
3117
3118         if (netif_xmit_stopped(txq) || !netdev_xmit_more())
3119                 /* trigger the dma engine. ena_ring_tx_doorbell()
3120                  * calls a memory barrier inside it.
3121                  */
3122                 ena_ring_tx_doorbell(tx_ring);
3123
3124         return NETDEV_TX_OK;
3125
3126 error_unmap_dma:
3127         ena_unmap_tx_buff(tx_ring, tx_info);
3128         tx_info->skb = NULL;
3129
3130 error_drop_packet:
3131         dev_kfree_skb(skb);
3132         return NETDEV_TX_OK;
3133 }
3134
3135 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3136                             struct net_device *sb_dev)
3137 {
3138         u16 qid;
3139         /* we suspect that this is good for in--kernel network services that
3140          * want to loop incoming skb rx to tx in normal user generated traffic,
3141          * most probably we will not get to this
3142          */
3143         if (skb_rx_queue_recorded(skb))
3144                 qid = skb_get_rx_queue(skb);
3145         else
3146                 qid = netdev_pick_tx(dev, skb, NULL);
3147
3148         return qid;
3149 }
3150
3151 static void ena_config_host_info(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3152 {
3153         struct device *dev = &pdev->dev;
3154         struct ena_admin_host_info *host_info;
3155         int rc;
3156
3157         /* Allocate only the host info */
3158         rc = ena_com_allocate_host_info(ena_dev);
3159         if (rc) {
3160                 dev_err(dev, "Cannot allocate host info\n");
3161                 return;
3162         }
3163
3164         host_info = ena_dev->host_attr.host_info;
3165
3166         host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
3167         host_info->os_type = ENA_ADMIN_OS_LINUX;
3168         host_info->kernel_ver = LINUX_VERSION_CODE;
3169         strlcpy(host_info->kernel_ver_str, utsname()->version,
3170                 sizeof(host_info->kernel_ver_str) - 1);
3171         host_info->os_dist = 0;
3172         strncpy(host_info->os_dist_str, utsname()->release,
3173                 sizeof(host_info->os_dist_str) - 1);
3174         host_info->driver_version =
3175                 (DRV_MODULE_GEN_MAJOR) |
3176                 (DRV_MODULE_GEN_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3177                 (DRV_MODULE_GEN_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3178                 ("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3179         host_info->num_cpus = num_online_cpus();
3180
3181         host_info->driver_supported_features =
3182                 ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK |
3183                 ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK |
3184                 ENA_ADMIN_HOST_INFO_RX_BUF_MIRRORING_MASK |
3185                 ENA_ADMIN_HOST_INFO_RSS_CONFIGURABLE_FUNCTION_KEY_MASK;
3186
3187         rc = ena_com_set_host_attributes(ena_dev);
3188         if (rc) {
3189                 if (rc == -EOPNOTSUPP)
3190                         dev_warn(dev, "Cannot set host attributes\n");
3191                 else
3192                         dev_err(dev, "Cannot set host attributes\n");
3193
3194                 goto err;
3195         }
3196
3197         return;
3198
3199 err:
3200         ena_com_delete_host_info(ena_dev);
3201 }
3202
3203 static void ena_config_debug_area(struct ena_adapter *adapter)
3204 {
3205         u32 debug_area_size;
3206         int rc, ss_count;
3207
3208         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3209         if (ss_count <= 0) {
3210                 netif_err(adapter, drv, adapter->netdev,
3211                           "SS count is negative\n");
3212                 return;
3213         }
3214
3215         /* allocate 32 bytes for each string and 64bit for the value */
3216         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3217
3218         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3219         if (rc) {
3220                 netif_err(adapter, drv, adapter->netdev,
3221                           "Cannot allocate debug area\n");
3222                 return;
3223         }
3224
3225         rc = ena_com_set_host_attributes(adapter->ena_dev);
3226         if (rc) {
3227                 if (rc == -EOPNOTSUPP)
3228                         netif_warn(adapter, drv, adapter->netdev,
3229                                    "Cannot set host attributes\n");
3230                 else
3231                         netif_err(adapter, drv, adapter->netdev,
3232                                   "Cannot set host attributes\n");
3233                 goto err;
3234         }
3235
3236         return;
3237 err:
3238         ena_com_delete_debug_area(adapter->ena_dev);
3239 }
3240
3241 int ena_update_hw_stats(struct ena_adapter *adapter)
3242 {
3243         int rc;
3244
3245         rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_stats);
3246         if (rc) {
3247                 netdev_err(adapter->netdev, "Failed to get ENI stats\n");
3248                 return rc;
3249         }
3250
3251         return 0;
3252 }
3253
3254 static void ena_get_stats64(struct net_device *netdev,
3255                             struct rtnl_link_stats64 *stats)
3256 {
3257         struct ena_adapter *adapter = netdev_priv(netdev);
3258         struct ena_ring *rx_ring, *tx_ring;
3259         unsigned int start;
3260         u64 rx_drops;
3261         u64 tx_drops;
3262         int i;
3263
3264         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3265                 return;
3266
3267         for (i = 0; i < adapter->num_io_queues; i++) {
3268                 u64 bytes, packets;
3269
3270                 tx_ring = &adapter->tx_ring[i];
3271
3272                 do {
3273                         start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
3274                         packets = tx_ring->tx_stats.cnt;
3275                         bytes = tx_ring->tx_stats.bytes;
3276                 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
3277
3278                 stats->tx_packets += packets;
3279                 stats->tx_bytes += bytes;
3280
3281                 rx_ring = &adapter->rx_ring[i];
3282
3283                 do {
3284                         start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
3285                         packets = rx_ring->rx_stats.cnt;
3286                         bytes = rx_ring->rx_stats.bytes;
3287                 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
3288
3289                 stats->rx_packets += packets;
3290                 stats->rx_bytes += bytes;
3291         }
3292
3293         do {
3294                 start = u64_stats_fetch_begin_irq(&adapter->syncp);
3295                 rx_drops = adapter->dev_stats.rx_drops;
3296                 tx_drops = adapter->dev_stats.tx_drops;
3297         } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
3298
3299         stats->rx_dropped = rx_drops;
3300         stats->tx_dropped = tx_drops;
3301
3302         stats->multicast = 0;
3303         stats->collisions = 0;
3304
3305         stats->rx_length_errors = 0;
3306         stats->rx_crc_errors = 0;
3307         stats->rx_frame_errors = 0;
3308         stats->rx_fifo_errors = 0;
3309         stats->rx_missed_errors = 0;
3310         stats->tx_window_errors = 0;
3311
3312         stats->rx_errors = 0;
3313         stats->tx_errors = 0;
3314 }
3315
3316 static const struct net_device_ops ena_netdev_ops = {
3317         .ndo_open               = ena_open,
3318         .ndo_stop               = ena_close,
3319         .ndo_start_xmit         = ena_start_xmit,
3320         .ndo_select_queue       = ena_select_queue,
3321         .ndo_get_stats64        = ena_get_stats64,
3322         .ndo_tx_timeout         = ena_tx_timeout,
3323         .ndo_change_mtu         = ena_change_mtu,
3324         .ndo_set_mac_address    = NULL,
3325         .ndo_validate_addr      = eth_validate_addr,
3326         .ndo_bpf                = ena_xdp,
3327         .ndo_xdp_xmit           = ena_xdp_xmit,
3328 };
3329
3330 static int ena_device_validate_params(struct ena_adapter *adapter,
3331                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
3332 {
3333         struct net_device *netdev = adapter->netdev;
3334         int rc;
3335
3336         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3337                               adapter->mac_addr);
3338         if (!rc) {
3339                 netif_err(adapter, drv, netdev,
3340                           "Error, mac address are different\n");
3341                 return -EINVAL;
3342         }
3343
3344         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3345                 netif_err(adapter, drv, netdev,
3346                           "Error, device max mtu is smaller than netdev MTU\n");
3347                 return -EINVAL;
3348         }
3349
3350         return 0;
3351 }
3352
3353 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
3354 {
3355         llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
3356         llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
3357         llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
3358         llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
3359         llq_config->llq_ring_entry_size_value = 128;
3360 }
3361
3362 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3363                                            struct ena_com_dev *ena_dev,
3364                                            struct ena_admin_feature_llq_desc *llq,
3365                                            struct ena_llq_configurations *llq_default_configurations)
3366 {
3367         int rc;
3368         u32 llq_feature_mask;
3369
3370         llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3371         if (!(ena_dev->supported_features & llq_feature_mask)) {
3372                 dev_warn(&pdev->dev,
3373                         "LLQ is not supported Fallback to host mode policy.\n");
3374                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3375                 return 0;
3376         }
3377
3378         rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3379         if (unlikely(rc)) {
3380                 dev_err(&pdev->dev,
3381                         "Failed to configure the device mode.  Fallback to host mode policy.\n");
3382                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3383         }
3384
3385         return 0;
3386 }
3387
3388 static int ena_map_llq_mem_bar(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
3389                                int bars)
3390 {
3391         bool has_mem_bar = !!(bars & BIT(ENA_MEM_BAR));
3392
3393         if (!has_mem_bar) {
3394                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3395                         dev_err(&pdev->dev,
3396                                 "ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3397                         ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3398                 }
3399
3400                 return 0;
3401         }
3402
3403         ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3404                                            pci_resource_start(pdev, ENA_MEM_BAR),
3405                                            pci_resource_len(pdev, ENA_MEM_BAR));
3406
3407         if (!ena_dev->mem_bar)
3408                 return -EFAULT;
3409
3410         return 0;
3411 }
3412
3413 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
3414                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
3415                            bool *wd_state)
3416 {
3417         struct ena_llq_configurations llq_config;
3418         struct device *dev = &pdev->dev;
3419         bool readless_supported;
3420         u32 aenq_groups;
3421         int dma_width;
3422         int rc;
3423
3424         rc = ena_com_mmio_reg_read_request_init(ena_dev);
3425         if (rc) {
3426                 dev_err(dev, "Failed to init mmio read less\n");
3427                 return rc;
3428         }
3429
3430         /* The PCIe configuration space revision id indicate if mmio reg
3431          * read is disabled
3432          */
3433         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3434         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3435
3436         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3437         if (rc) {
3438                 dev_err(dev, "Can not reset device\n");
3439                 goto err_mmio_read_less;
3440         }
3441
3442         rc = ena_com_validate_version(ena_dev);
3443         if (rc) {
3444                 dev_err(dev, "Device version is too low\n");
3445                 goto err_mmio_read_less;
3446         }
3447
3448         dma_width = ena_com_get_dma_width(ena_dev);
3449         if (dma_width < 0) {
3450                 dev_err(dev, "Invalid dma width value %d", dma_width);
3451                 rc = dma_width;
3452                 goto err_mmio_read_less;
3453         }
3454
3455         rc = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(dma_width));
3456         if (rc) {
3457                 dev_err(dev, "dma_set_mask_and_coherent failed %d\n", rc);
3458                 goto err_mmio_read_less;
3459         }
3460
3461         /* ENA admin level init */
3462         rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3463         if (rc) {
3464                 dev_err(dev,
3465                         "Can not initialize ena admin queue with device\n");
3466                 goto err_mmio_read_less;
3467         }
3468
3469         /* To enable the msix interrupts the driver needs to know the number
3470          * of queues. So the driver uses polling mode to retrieve this
3471          * information
3472          */
3473         ena_com_set_admin_polling_mode(ena_dev, true);
3474
3475         ena_config_host_info(ena_dev, pdev);
3476
3477         /* Get Device Attributes*/
3478         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3479         if (rc) {
3480                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3481                 goto err_admin_init;
3482         }
3483
3484         /* Try to turn all the available aenq groups */
3485         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3486                 BIT(ENA_ADMIN_FATAL_ERROR) |
3487                 BIT(ENA_ADMIN_WARNING) |
3488                 BIT(ENA_ADMIN_NOTIFICATION) |
3489                 BIT(ENA_ADMIN_KEEP_ALIVE);
3490
3491         aenq_groups &= get_feat_ctx->aenq.supported_groups;
3492
3493         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3494         if (rc) {
3495                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3496                 goto err_admin_init;
3497         }
3498
3499         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3500
3501         set_default_llq_configurations(&llq_config);
3502
3503         rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx->llq,
3504                                              &llq_config);
3505         if (rc) {
3506                 dev_err(dev, "ENA device init failed\n");
3507                 goto err_admin_init;
3508         }
3509
3510         return 0;
3511
3512 err_admin_init:
3513         ena_com_delete_host_info(ena_dev);
3514         ena_com_admin_destroy(ena_dev);
3515 err_mmio_read_less:
3516         ena_com_mmio_reg_read_request_destroy(ena_dev);
3517
3518         return rc;
3519 }
3520
3521 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3522 {
3523         struct ena_com_dev *ena_dev = adapter->ena_dev;
3524         struct device *dev = &adapter->pdev->dev;
3525         int rc;
3526
3527         rc = ena_enable_msix(adapter);
3528         if (rc) {
3529                 dev_err(dev, "Can not reserve msix vectors\n");
3530                 return rc;
3531         }
3532
3533         ena_setup_mgmnt_intr(adapter);
3534
3535         rc = ena_request_mgmnt_irq(adapter);
3536         if (rc) {
3537                 dev_err(dev, "Can not setup management interrupts\n");
3538                 goto err_disable_msix;
3539         }
3540
3541         ena_com_set_admin_polling_mode(ena_dev, false);
3542
3543         ena_com_admin_aenq_enable(ena_dev);
3544
3545         return 0;
3546
3547 err_disable_msix:
3548         ena_disable_msix(adapter);
3549
3550         return rc;
3551 }
3552
3553 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3554 {
3555         struct net_device *netdev = adapter->netdev;
3556         struct ena_com_dev *ena_dev = adapter->ena_dev;
3557         bool dev_up;
3558
3559         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3560                 return;
3561
3562         netif_carrier_off(netdev);
3563
3564         del_timer_sync(&adapter->timer_service);
3565
3566         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3567         adapter->dev_up_before_reset = dev_up;
3568         if (!graceful)
3569                 ena_com_set_admin_running_state(ena_dev, false);
3570
3571         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3572                 ena_down(adapter);
3573
3574         /* Stop the device from sending AENQ events (in case reset flag is set
3575          *  and device is up, ena_down() already reset the device.
3576          */
3577         if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3578                 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3579
3580         ena_free_mgmnt_irq(adapter);
3581
3582         ena_disable_msix(adapter);
3583
3584         ena_com_abort_admin_commands(ena_dev);
3585
3586         ena_com_wait_for_abort_completion(ena_dev);
3587
3588         ena_com_admin_destroy(ena_dev);
3589
3590         ena_com_mmio_reg_read_request_destroy(ena_dev);
3591
3592         /* return reset reason to default value */
3593         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3594
3595         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3596         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3597 }
3598
3599 static int ena_restore_device(struct ena_adapter *adapter)
3600 {
3601         struct ena_com_dev_get_features_ctx get_feat_ctx;
3602         struct ena_com_dev *ena_dev = adapter->ena_dev;
3603         struct pci_dev *pdev = adapter->pdev;
3604         bool wd_state;
3605         int rc;
3606
3607         set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3608         rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
3609         if (rc) {
3610                 dev_err(&pdev->dev, "Can not initialize device\n");
3611                 goto err;
3612         }
3613         adapter->wd_state = wd_state;
3614
3615         rc = ena_device_validate_params(adapter, &get_feat_ctx);
3616         if (rc) {
3617                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
3618                 goto err_device_destroy;
3619         }
3620
3621         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3622         if (rc) {
3623                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
3624                 goto err_device_destroy;
3625         }
3626         /* If the interface was up before the reset bring it up */
3627         if (adapter->dev_up_before_reset) {
3628                 rc = ena_up(adapter);
3629                 if (rc) {
3630                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
3631                         goto err_disable_msix;
3632                 }
3633         }
3634
3635         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3636
3637         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3638         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3639                 netif_carrier_on(adapter->netdev);
3640
3641         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3642         adapter->last_keep_alive_jiffies = jiffies;
3643
3644         return rc;
3645 err_disable_msix:
3646         ena_free_mgmnt_irq(adapter);
3647         ena_disable_msix(adapter);
3648 err_device_destroy:
3649         ena_com_abort_admin_commands(ena_dev);
3650         ena_com_wait_for_abort_completion(ena_dev);
3651         ena_com_admin_destroy(ena_dev);
3652         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3653         ena_com_mmio_reg_read_request_destroy(ena_dev);
3654 err:
3655         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3656         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3657         dev_err(&pdev->dev,
3658                 "Reset attempt failed. Can not reset the device\n");
3659
3660         return rc;
3661 }
3662
3663 static void ena_fw_reset_device(struct work_struct *work)
3664 {
3665         struct ena_adapter *adapter =
3666                 container_of(work, struct ena_adapter, reset_task);
3667
3668         rtnl_lock();
3669
3670         if (likely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3671                 ena_destroy_device(adapter, false);
3672                 ena_restore_device(adapter);
3673
3674                 dev_err(&adapter->pdev->dev, "Device reset completed successfully\n");
3675         }
3676
3677         rtnl_unlock();
3678 }
3679
3680 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3681                                         struct ena_ring *rx_ring)
3682 {
3683         struct ena_napi *ena_napi = container_of(rx_ring->napi, struct ena_napi, napi);
3684
3685         if (likely(READ_ONCE(ena_napi->first_interrupt)))
3686                 return 0;
3687
3688         if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3689                 return 0;
3690
3691         rx_ring->no_interrupt_event_cnt++;
3692
3693         if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3694                 netif_err(adapter, rx_err, adapter->netdev,
3695                           "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3696                           rx_ring->qid);
3697
3698                 ena_reset_device(adapter, ENA_REGS_RESET_MISS_INTERRUPT);
3699                 return -EIO;
3700         }
3701
3702         return 0;
3703 }
3704
3705 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3706                                           struct ena_ring *tx_ring)
3707 {
3708         struct ena_napi *ena_napi = container_of(tx_ring->napi, struct ena_napi, napi);
3709         unsigned int time_since_last_napi;
3710         unsigned int missing_tx_comp_to;
3711         bool is_tx_comp_time_expired;
3712         struct ena_tx_buffer *tx_buf;
3713         unsigned long last_jiffies;
3714         u32 missed_tx = 0;
3715         int i, rc = 0;
3716
3717         for (i = 0; i < tx_ring->ring_size; i++) {
3718                 tx_buf = &tx_ring->tx_buffer_info[i];
3719                 last_jiffies = tx_buf->last_jiffies;
3720
3721                 if (last_jiffies == 0)
3722                         /* no pending Tx at this location */
3723                         continue;
3724
3725                 is_tx_comp_time_expired = time_is_before_jiffies(last_jiffies +
3726                          2 * adapter->missing_tx_completion_to);
3727
3728                 if (unlikely(!READ_ONCE(ena_napi->first_interrupt) && is_tx_comp_time_expired)) {
3729                         /* If after graceful period interrupt is still not
3730                          * received, we schedule a reset
3731                          */
3732                         netif_err(adapter, tx_err, adapter->netdev,
3733                                   "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3734                                   tx_ring->qid);
3735                         ena_reset_device(adapter, ENA_REGS_RESET_MISS_INTERRUPT);
3736                         return -EIO;
3737                 }
3738
3739                 is_tx_comp_time_expired = time_is_before_jiffies(last_jiffies +
3740                         adapter->missing_tx_completion_to);
3741
3742                 if (unlikely(is_tx_comp_time_expired)) {
3743                         if (!tx_buf->print_once) {
3744                                 time_since_last_napi = jiffies_to_usecs(jiffies - tx_ring->tx_stats.last_napi_jiffies);
3745                                 missing_tx_comp_to = jiffies_to_msecs(adapter->missing_tx_completion_to);
3746                                 netif_notice(adapter, tx_err, adapter->netdev,
3747                                              "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",
3748                                              tx_ring->qid, i, time_since_last_napi, missing_tx_comp_to);
3749                         }
3750
3751                         tx_buf->print_once = 1;
3752                         missed_tx++;
3753                 }
3754         }
3755
3756         if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3757                 netif_err(adapter, tx_err, adapter->netdev,
3758                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3759                           missed_tx,
3760                           adapter->missing_tx_completion_threshold);
3761                 ena_reset_device(adapter, ENA_REGS_RESET_MISS_TX_CMPL);
3762                 rc = -EIO;
3763         }
3764
3765         ena_increase_stat(&tx_ring->tx_stats.missed_tx, missed_tx,
3766                           &tx_ring->syncp);
3767
3768         return rc;
3769 }
3770
3771 static void check_for_missing_completions(struct ena_adapter *adapter)
3772 {
3773         struct ena_ring *tx_ring;
3774         struct ena_ring *rx_ring;
3775         int i, budget, rc;
3776         int io_queue_count;
3777
3778         io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
3779         /* Make sure the driver doesn't turn the device in other process */
3780         smp_rmb();
3781
3782         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3783                 return;
3784
3785         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3786                 return;
3787
3788         if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
3789                 return;
3790
3791         budget = ENA_MONITORED_TX_QUEUES;
3792
3793         for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
3794                 tx_ring = &adapter->tx_ring[i];
3795                 rx_ring = &adapter->rx_ring[i];
3796
3797                 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3798                 if (unlikely(rc))
3799                         return;
3800
3801                 rc =  !ENA_IS_XDP_INDEX(adapter, i) ?
3802                         check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
3803                 if (unlikely(rc))
3804                         return;
3805
3806                 budget--;
3807                 if (!budget)
3808                         break;
3809         }
3810
3811         adapter->last_monitored_tx_qid = i % io_queue_count;
3812 }
3813
3814 /* trigger napi schedule after 2 consecutive detections */
3815 #define EMPTY_RX_REFILL 2
3816 /* For the rare case where the device runs out of Rx descriptors and the
3817  * napi handler failed to refill new Rx descriptors (due to a lack of memory
3818  * for example).
3819  * This case will lead to a deadlock:
3820  * The device won't send interrupts since all the new Rx packets will be dropped
3821  * The napi handler won't allocate new Rx descriptors so the device will be
3822  * able to send new packets.
3823  *
3824  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
3825  * It is recommended to have at least 512MB, with a minimum of 128MB for
3826  * constrained environment).
3827  *
3828  * When such a situation is detected - Reschedule napi
3829  */
3830 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
3831 {
3832         struct ena_ring *rx_ring;
3833         int i, refill_required;
3834
3835         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3836                 return;
3837
3838         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3839                 return;
3840
3841         for (i = 0; i < adapter->num_io_queues; i++) {
3842                 rx_ring = &adapter->rx_ring[i];
3843
3844                 refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
3845                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3846                         rx_ring->empty_rx_queue++;
3847
3848                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3849                                 ena_increase_stat(&rx_ring->rx_stats.empty_rx_ring, 1,
3850                                                   &rx_ring->syncp);
3851
3852                                 netif_err(adapter, drv, adapter->netdev,
3853                                           "Trigger refill for ring %d\n", i);
3854
3855                                 napi_schedule(rx_ring->napi);
3856                                 rx_ring->empty_rx_queue = 0;
3857                         }
3858                 } else {
3859                         rx_ring->empty_rx_queue = 0;
3860                 }
3861         }
3862 }
3863
3864 /* Check for keep alive expiration */
3865 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3866 {
3867         unsigned long keep_alive_expired;
3868
3869         if (!adapter->wd_state)
3870                 return;
3871
3872         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3873                 return;
3874
3875         keep_alive_expired = adapter->last_keep_alive_jiffies +
3876                              adapter->keep_alive_timeout;
3877         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3878                 netif_err(adapter, drv, adapter->netdev,
3879                           "Keep alive watchdog timeout.\n");
3880                 ena_increase_stat(&adapter->dev_stats.wd_expired, 1,
3881                                   &adapter->syncp);
3882                 ena_reset_device(adapter, ENA_REGS_RESET_KEEP_ALIVE_TO);
3883         }
3884 }
3885
3886 static void check_for_admin_com_state(struct ena_adapter *adapter)
3887 {
3888         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3889                 netif_err(adapter, drv, adapter->netdev,
3890                           "ENA admin queue is not in running state!\n");
3891                 ena_increase_stat(&adapter->dev_stats.admin_q_pause, 1,
3892                                   &adapter->syncp);
3893                 ena_reset_device(adapter, ENA_REGS_RESET_ADMIN_TO);
3894         }
3895 }
3896
3897 static void ena_update_hints(struct ena_adapter *adapter,
3898                              struct ena_admin_ena_hw_hints *hints)
3899 {
3900         struct net_device *netdev = adapter->netdev;
3901
3902         if (hints->admin_completion_tx_timeout)
3903                 adapter->ena_dev->admin_queue.completion_timeout =
3904                         hints->admin_completion_tx_timeout * 1000;
3905
3906         if (hints->mmio_read_timeout)
3907                 /* convert to usec */
3908                 adapter->ena_dev->mmio_read.reg_read_to =
3909                         hints->mmio_read_timeout * 1000;
3910
3911         if (hints->missed_tx_completion_count_threshold_to_reset)
3912                 adapter->missing_tx_completion_threshold =
3913                         hints->missed_tx_completion_count_threshold_to_reset;
3914
3915         if (hints->missing_tx_completion_timeout) {
3916                 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3917                         adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3918                 else
3919                         adapter->missing_tx_completion_to =
3920                                 msecs_to_jiffies(hints->missing_tx_completion_timeout);
3921         }
3922
3923         if (hints->netdev_wd_timeout)
3924                 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3925
3926         if (hints->driver_watchdog_timeout) {
3927                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3928                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3929                 else
3930                         adapter->keep_alive_timeout =
3931                                 msecs_to_jiffies(hints->driver_watchdog_timeout);
3932         }
3933 }
3934
3935 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3936                                  struct net_device *netdev)
3937 {
3938         host_info->supported_network_features[0] =
3939                 netdev->features & GENMASK_ULL(31, 0);
3940         host_info->supported_network_features[1] =
3941                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
3942 }
3943
3944 static void ena_timer_service(struct timer_list *t)
3945 {
3946         struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3947         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3948         struct ena_admin_host_info *host_info =
3949                 adapter->ena_dev->host_attr.host_info;
3950
3951         check_for_missing_keep_alive(adapter);
3952
3953         check_for_admin_com_state(adapter);
3954
3955         check_for_missing_completions(adapter);
3956
3957         check_for_empty_rx_ring(adapter);
3958
3959         if (debug_area)
3960                 ena_dump_stats_to_buf(adapter, debug_area);
3961
3962         if (host_info)
3963                 ena_update_host_info(host_info, adapter->netdev);
3964
3965         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3966                 netif_err(adapter, drv, adapter->netdev,
3967                           "Trigger reset is on\n");
3968                 ena_dump_stats_to_dmesg(adapter);
3969                 queue_work(ena_wq, &adapter->reset_task);
3970                 return;
3971         }
3972
3973         /* Reset the timer */
3974         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3975 }
3976
3977 static u32 ena_calc_max_io_queue_num(struct pci_dev *pdev,
3978                                      struct ena_com_dev *ena_dev,
3979                                      struct ena_com_dev_get_features_ctx *get_feat_ctx)
3980 {
3981         u32 io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
3982
3983         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3984                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3985                         &get_feat_ctx->max_queue_ext.max_queue_ext;
3986                 io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
3987                                   max_queue_ext->max_rx_cq_num);
3988
3989                 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
3990                 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
3991         } else {
3992                 struct ena_admin_queue_feature_desc *max_queues =
3993                         &get_feat_ctx->max_queues;
3994                 io_tx_sq_num = max_queues->max_sq_num;
3995                 io_tx_cq_num = max_queues->max_cq_num;
3996                 io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
3997         }
3998
3999         /* In case of LLQ use the llq fields for the tx SQ/CQ */
4000         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4001                 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
4002
4003         max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
4004         max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
4005         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
4006         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
4007         /* 1 IRQ for mgmnt and 1 IRQs for each IO direction */
4008         max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
4009
4010         return max_num_io_queues;
4011 }
4012
4013 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
4014                                  struct net_device *netdev)
4015 {
4016         netdev_features_t dev_features = 0;
4017
4018         /* Set offload features */
4019         if (feat->offload.tx &
4020                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
4021                 dev_features |= NETIF_F_IP_CSUM;
4022
4023         if (feat->offload.tx &
4024                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
4025                 dev_features |= NETIF_F_IPV6_CSUM;
4026
4027         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
4028                 dev_features |= NETIF_F_TSO;
4029
4030         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
4031                 dev_features |= NETIF_F_TSO6;
4032
4033         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
4034                 dev_features |= NETIF_F_TSO_ECN;
4035
4036         if (feat->offload.rx_supported &
4037                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
4038                 dev_features |= NETIF_F_RXCSUM;
4039
4040         if (feat->offload.rx_supported &
4041                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
4042                 dev_features |= NETIF_F_RXCSUM;
4043
4044         netdev->features =
4045                 dev_features |
4046                 NETIF_F_SG |
4047                 NETIF_F_RXHASH |
4048                 NETIF_F_HIGHDMA;
4049
4050         netdev->hw_features |= netdev->features;
4051         netdev->vlan_features |= netdev->features;
4052 }
4053
4054 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
4055                                      struct ena_com_dev_get_features_ctx *feat)
4056 {
4057         struct net_device *netdev = adapter->netdev;
4058
4059         /* Copy mac address */
4060         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
4061                 eth_hw_addr_random(netdev);
4062                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
4063         } else {
4064                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
4065                 eth_hw_addr_set(netdev, adapter->mac_addr);
4066         }
4067
4068         /* Set offload features */
4069         ena_set_dev_offloads(feat, netdev);
4070
4071         adapter->max_mtu = feat->dev_attr.max_mtu;
4072         netdev->max_mtu = adapter->max_mtu;
4073         netdev->min_mtu = ENA_MIN_MTU;
4074 }
4075
4076 static int ena_rss_init_default(struct ena_adapter *adapter)
4077 {
4078         struct ena_com_dev *ena_dev = adapter->ena_dev;
4079         struct device *dev = &adapter->pdev->dev;
4080         int rc, i;
4081         u32 val;
4082
4083         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
4084         if (unlikely(rc)) {
4085                 dev_err(dev, "Cannot init indirect table\n");
4086                 goto err_rss_init;
4087         }
4088
4089         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
4090                 val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
4091                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
4092                                                        ENA_IO_RXQ_IDX(val));
4093                 if (unlikely(rc)) {
4094                         dev_err(dev, "Cannot fill indirect table\n");
4095                         goto err_fill_indir;
4096                 }
4097         }
4098
4099         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ, NULL,
4100                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4101         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4102                 dev_err(dev, "Cannot fill hash function\n");
4103                 goto err_fill_indir;
4104         }
4105
4106         rc = ena_com_set_default_hash_ctrl(ena_dev);
4107         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4108                 dev_err(dev, "Cannot fill hash control\n");
4109                 goto err_fill_indir;
4110         }
4111
4112         return 0;
4113
4114 err_fill_indir:
4115         ena_com_rss_destroy(ena_dev);
4116 err_rss_init:
4117
4118         return rc;
4119 }
4120
4121 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4122 {
4123         int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4124
4125         pci_release_selected_regions(pdev, release_bars);
4126 }
4127
4128
4129 static void ena_calc_io_queue_size(struct ena_adapter *adapter,
4130                                    struct ena_com_dev_get_features_ctx *get_feat_ctx)
4131 {
4132         struct ena_admin_feature_llq_desc *llq = &get_feat_ctx->llq;
4133         struct ena_com_dev *ena_dev = adapter->ena_dev;
4134         u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
4135         u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
4136         u32 max_tx_queue_size;
4137         u32 max_rx_queue_size;
4138
4139         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4140                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4141                         &get_feat_ctx->max_queue_ext.max_queue_ext;
4142                 max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
4143                                           max_queue_ext->max_rx_sq_depth);
4144                 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
4145
4146                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4147                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4148                                                   llq->max_llq_depth);
4149                 else
4150                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4151                                                   max_queue_ext->max_tx_sq_depth);
4152
4153                 adapter->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4154                                                  max_queue_ext->max_per_packet_tx_descs);
4155                 adapter->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4156                                                  max_queue_ext->max_per_packet_rx_descs);
4157         } else {
4158                 struct ena_admin_queue_feature_desc *max_queues =
4159                         &get_feat_ctx->max_queues;
4160                 max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
4161                                           max_queues->max_sq_depth);
4162                 max_tx_queue_size = max_queues->max_cq_depth;
4163
4164                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4165                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4166                                                   llq->max_llq_depth);
4167                 else
4168                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4169                                                   max_queues->max_sq_depth);
4170
4171                 adapter->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4172                                                  max_queues->max_packet_tx_descs);
4173                 adapter->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4174                                                  max_queues->max_packet_rx_descs);
4175         }
4176
4177         max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
4178         max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
4179
4180         tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
4181                                   max_tx_queue_size);
4182         rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
4183                                   max_rx_queue_size);
4184
4185         tx_queue_size = rounddown_pow_of_two(tx_queue_size);
4186         rx_queue_size = rounddown_pow_of_two(rx_queue_size);
4187
4188         adapter->max_tx_ring_size  = max_tx_queue_size;
4189         adapter->max_rx_ring_size = max_rx_queue_size;
4190         adapter->requested_tx_ring_size = tx_queue_size;
4191         adapter->requested_rx_ring_size = rx_queue_size;
4192 }
4193
4194 /* ena_probe - Device Initialization Routine
4195  * @pdev: PCI device information struct
4196  * @ent: entry in ena_pci_tbl
4197  *
4198  * Returns 0 on success, negative on failure
4199  *
4200  * ena_probe initializes an adapter identified by a pci_dev structure.
4201  * The OS initialization, configuring of the adapter private structure,
4202  * and a hardware reset occur.
4203  */
4204 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4205 {
4206         struct ena_com_dev_get_features_ctx get_feat_ctx;
4207         struct ena_com_dev *ena_dev = NULL;
4208         struct ena_adapter *adapter;
4209         struct net_device *netdev;
4210         static int adapters_found;
4211         u32 max_num_io_queues;
4212         bool wd_state;
4213         int bars, rc;
4214
4215         dev_dbg(&pdev->dev, "%s\n", __func__);
4216
4217         rc = pci_enable_device_mem(pdev);
4218         if (rc) {
4219                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4220                 return rc;
4221         }
4222
4223         rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
4224         if (rc) {
4225                 dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n", rc);
4226                 goto err_disable_device;
4227         }
4228
4229         pci_set_master(pdev);
4230
4231         ena_dev = vzalloc(sizeof(*ena_dev));
4232         if (!ena_dev) {
4233                 rc = -ENOMEM;
4234                 goto err_disable_device;
4235         }
4236
4237         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4238         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4239         if (rc) {
4240                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4241                         rc);
4242                 goto err_free_ena_dev;
4243         }
4244
4245         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4246                                         pci_resource_start(pdev, ENA_REG_BAR),
4247                                         pci_resource_len(pdev, ENA_REG_BAR));
4248         if (!ena_dev->reg_bar) {
4249                 dev_err(&pdev->dev, "Failed to remap regs bar\n");
4250                 rc = -EFAULT;
4251                 goto err_free_region;
4252         }
4253
4254         ena_dev->ena_min_poll_delay_us = ENA_ADMIN_POLL_DELAY_US;
4255
4256         ena_dev->dmadev = &pdev->dev;
4257
4258         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), ENA_MAX_RINGS);
4259         if (!netdev) {
4260                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4261                 rc = -ENOMEM;
4262                 goto err_free_region;
4263         }
4264
4265         SET_NETDEV_DEV(netdev, &pdev->dev);
4266         adapter = netdev_priv(netdev);
4267         adapter->ena_dev = ena_dev;
4268         adapter->netdev = netdev;
4269         adapter->pdev = pdev;
4270         adapter->msg_enable = DEFAULT_MSG_ENABLE;
4271
4272         ena_dev->net_device = netdev;
4273
4274         pci_set_drvdata(pdev, adapter);
4275
4276         rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
4277         if (rc) {
4278                 dev_err(&pdev->dev, "ENA device init failed\n");
4279                 if (rc == -ETIME)
4280                         rc = -EPROBE_DEFER;
4281                 goto err_netdev_destroy;
4282         }
4283
4284         rc = ena_map_llq_mem_bar(pdev, ena_dev, bars);
4285         if (rc) {
4286                 dev_err(&pdev->dev, "ENA llq bar mapping failed\n");
4287                 goto err_device_destroy;
4288         }
4289
4290         /* Initial TX and RX interrupt delay. Assumes 1 usec granularity.
4291          * Updated during device initialization with the real granularity
4292          */
4293         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4294         ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4295         ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4296         max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4297         ena_calc_io_queue_size(adapter, &get_feat_ctx);
4298         if (unlikely(!max_num_io_queues)) {
4299                 rc = -EFAULT;
4300                 goto err_device_destroy;
4301         }
4302
4303         ena_set_conf_feat_params(adapter, &get_feat_ctx);
4304
4305         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4306
4307         adapter->num_io_queues = max_num_io_queues;
4308         adapter->max_num_io_queues = max_num_io_queues;
4309         adapter->last_monitored_tx_qid = 0;
4310
4311         adapter->xdp_first_ring = 0;
4312         adapter->xdp_num_queues = 0;
4313
4314         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4315         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4316                 adapter->disable_meta_caching =
4317                         !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
4318                            BIT(ENA_ADMIN_DISABLE_META_CACHING));
4319
4320         adapter->wd_state = wd_state;
4321
4322         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4323
4324         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4325         if (rc) {
4326                 dev_err(&pdev->dev,
4327                         "Failed to query interrupt moderation feature\n");
4328                 goto err_device_destroy;
4329         }
4330         ena_init_io_rings(adapter,
4331                           0,
4332                           adapter->xdp_num_queues +
4333                           adapter->num_io_queues);
4334
4335         netdev->netdev_ops = &ena_netdev_ops;
4336         netdev->watchdog_timeo = TX_TIMEOUT;
4337         ena_set_ethtool_ops(netdev);
4338
4339         netdev->priv_flags |= IFF_UNICAST_FLT;
4340
4341         u64_stats_init(&adapter->syncp);
4342
4343         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4344         if (rc) {
4345                 dev_err(&pdev->dev,
4346                         "Failed to enable and set the admin interrupts\n");
4347                 goto err_worker_destroy;
4348         }
4349         rc = ena_rss_init_default(adapter);
4350         if (rc && (rc != -EOPNOTSUPP)) {
4351                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4352                 goto err_free_msix;
4353         }
4354
4355         ena_config_debug_area(adapter);
4356
4357         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4358
4359         netif_carrier_off(netdev);
4360
4361         rc = register_netdev(netdev);
4362         if (rc) {
4363                 dev_err(&pdev->dev, "Cannot register net device\n");
4364                 goto err_rss;
4365         }
4366
4367         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4368
4369         adapter->last_keep_alive_jiffies = jiffies;
4370         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4371         adapter->missing_tx_completion_to = TX_TIMEOUT;
4372         adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4373
4374         ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4375
4376         timer_setup(&adapter->timer_service, ena_timer_service, 0);
4377         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4378
4379         dev_info(&pdev->dev,
4380                  "%s found at mem %lx, mac addr %pM\n",
4381                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4382                  netdev->dev_addr);
4383
4384         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4385
4386         adapters_found++;
4387
4388         return 0;
4389
4390 err_rss:
4391         ena_com_delete_debug_area(ena_dev);
4392         ena_com_rss_destroy(ena_dev);
4393 err_free_msix:
4394         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4395         /* stop submitting admin commands on a device that was reset */
4396         ena_com_set_admin_running_state(ena_dev, false);
4397         ena_free_mgmnt_irq(adapter);
4398         ena_disable_msix(adapter);
4399 err_worker_destroy:
4400         del_timer(&adapter->timer_service);
4401 err_device_destroy:
4402         ena_com_delete_host_info(ena_dev);
4403         ena_com_admin_destroy(ena_dev);
4404 err_netdev_destroy:
4405         free_netdev(netdev);
4406 err_free_region:
4407         ena_release_bars(ena_dev, pdev);
4408 err_free_ena_dev:
4409         vfree(ena_dev);
4410 err_disable_device:
4411         pci_disable_device(pdev);
4412         return rc;
4413 }
4414
4415 /*****************************************************************************/
4416
4417 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4418  * @pdev: PCI device information struct
4419  * @shutdown: Is it a shutdown operation? If false, means it is a removal
4420  *
4421  * __ena_shutoff is a helper routine that does the real work on shutdown and
4422  * removal paths; the difference between those paths is with regards to whether
4423  * dettach or unregister the netdevice.
4424  */
4425 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4426 {
4427         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4428         struct ena_com_dev *ena_dev;
4429         struct net_device *netdev;
4430
4431         ena_dev = adapter->ena_dev;
4432         netdev = adapter->netdev;
4433
4434 #ifdef CONFIG_RFS_ACCEL
4435         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4436                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4437                 netdev->rx_cpu_rmap = NULL;
4438         }
4439 #endif /* CONFIG_RFS_ACCEL */
4440
4441         /* Make sure timer and reset routine won't be called after
4442          * freeing device resources.
4443          */
4444         del_timer_sync(&adapter->timer_service);
4445         cancel_work_sync(&adapter->reset_task);
4446
4447         rtnl_lock(); /* lock released inside the below if-else block */
4448         adapter->reset_reason = ENA_REGS_RESET_SHUTDOWN;
4449         ena_destroy_device(adapter, true);
4450         if (shutdown) {
4451                 netif_device_detach(netdev);
4452                 dev_close(netdev);
4453                 rtnl_unlock();
4454         } else {
4455                 rtnl_unlock();
4456                 unregister_netdev(netdev);
4457                 free_netdev(netdev);
4458         }
4459
4460         ena_com_rss_destroy(ena_dev);
4461
4462         ena_com_delete_debug_area(ena_dev);
4463
4464         ena_com_delete_host_info(ena_dev);
4465
4466         ena_release_bars(ena_dev, pdev);
4467
4468         pci_disable_device(pdev);
4469
4470         vfree(ena_dev);
4471 }
4472
4473 /* ena_remove - Device Removal Routine
4474  * @pdev: PCI device information struct
4475  *
4476  * ena_remove is called by the PCI subsystem to alert the driver
4477  * that it should release a PCI device.
4478  */
4479
4480 static void ena_remove(struct pci_dev *pdev)
4481 {
4482         __ena_shutoff(pdev, false);
4483 }
4484
4485 /* ena_shutdown - Device Shutdown Routine
4486  * @pdev: PCI device information struct
4487  *
4488  * ena_shutdown is called by the PCI subsystem to alert the driver that
4489  * a shutdown/reboot (or kexec) is happening and device must be disabled.
4490  */
4491
4492 static void ena_shutdown(struct pci_dev *pdev)
4493 {
4494         __ena_shutoff(pdev, true);
4495 }
4496
4497 /* ena_suspend - PM suspend callback
4498  * @dev_d: Device information struct
4499  */
4500 static int __maybe_unused ena_suspend(struct device *dev_d)
4501 {
4502         struct pci_dev *pdev = to_pci_dev(dev_d);
4503         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4504
4505         ena_increase_stat(&adapter->dev_stats.suspend, 1, &adapter->syncp);
4506
4507         rtnl_lock();
4508         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4509                 dev_err(&pdev->dev,
4510                         "Ignoring device reset request as the device is being suspended\n");
4511                 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4512         }
4513         ena_destroy_device(adapter, true);
4514         rtnl_unlock();
4515         return 0;
4516 }
4517
4518 /* ena_resume - PM resume callback
4519  * @dev_d: Device information struct
4520  */
4521 static int __maybe_unused ena_resume(struct device *dev_d)
4522 {
4523         struct ena_adapter *adapter = dev_get_drvdata(dev_d);
4524         int rc;
4525
4526         ena_increase_stat(&adapter->dev_stats.resume, 1, &adapter->syncp);
4527
4528         rtnl_lock();
4529         rc = ena_restore_device(adapter);
4530         rtnl_unlock();
4531         return rc;
4532 }
4533
4534 static SIMPLE_DEV_PM_OPS(ena_pm_ops, ena_suspend, ena_resume);
4535
4536 static struct pci_driver ena_pci_driver = {
4537         .name           = DRV_MODULE_NAME,
4538         .id_table       = ena_pci_tbl,
4539         .probe          = ena_probe,
4540         .remove         = ena_remove,
4541         .shutdown       = ena_shutdown,
4542         .driver.pm      = &ena_pm_ops,
4543         .sriov_configure = pci_sriov_configure_simple,
4544 };
4545
4546 static int __init ena_init(void)
4547 {
4548         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4549         if (!ena_wq) {
4550                 pr_err("Failed to create workqueue\n");
4551                 return -ENOMEM;
4552         }
4553
4554         return pci_register_driver(&ena_pci_driver);
4555 }
4556
4557 static void __exit ena_cleanup(void)
4558 {
4559         pci_unregister_driver(&ena_pci_driver);
4560
4561         if (ena_wq) {
4562                 destroy_workqueue(ena_wq);
4563                 ena_wq = NULL;
4564         }
4565 }
4566
4567 /******************************************************************************
4568  ******************************** AENQ Handlers *******************************
4569  *****************************************************************************/
4570 /* ena_update_on_link_change:
4571  * Notify the network interface about the change in link status
4572  */
4573 static void ena_update_on_link_change(void *adapter_data,
4574                                       struct ena_admin_aenq_entry *aenq_e)
4575 {
4576         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4577         struct ena_admin_aenq_link_change_desc *aenq_desc =
4578                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
4579         int status = aenq_desc->flags &
4580                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4581
4582         if (status) {
4583                 netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
4584                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4585                 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4586                         netif_carrier_on(adapter->netdev);
4587         } else {
4588                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4589                 netif_carrier_off(adapter->netdev);
4590         }
4591 }
4592
4593 static void ena_keep_alive_wd(void *adapter_data,
4594                               struct ena_admin_aenq_entry *aenq_e)
4595 {
4596         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4597         struct ena_admin_aenq_keep_alive_desc *desc;
4598         u64 rx_drops;
4599         u64 tx_drops;
4600
4601         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4602         adapter->last_keep_alive_jiffies = jiffies;
4603
4604         rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4605         tx_drops = ((u64)desc->tx_drops_high << 32) | desc->tx_drops_low;
4606
4607         u64_stats_update_begin(&adapter->syncp);
4608         /* These stats are accumulated by the device, so the counters indicate
4609          * all drops since last reset.
4610          */
4611         adapter->dev_stats.rx_drops = rx_drops;
4612         adapter->dev_stats.tx_drops = tx_drops;
4613         u64_stats_update_end(&adapter->syncp);
4614 }
4615
4616 static void ena_notification(void *adapter_data,
4617                              struct ena_admin_aenq_entry *aenq_e)
4618 {
4619         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4620         struct ena_admin_ena_hw_hints *hints;
4621
4622         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4623              "Invalid group(%x) expected %x\n",
4624              aenq_e->aenq_common_desc.group,
4625              ENA_ADMIN_NOTIFICATION);
4626
4627         switch (aenq_e->aenq_common_desc.syndrome) {
4628         case ENA_ADMIN_UPDATE_HINTS:
4629                 hints = (struct ena_admin_ena_hw_hints *)
4630                         (&aenq_e->inline_data_w4);
4631                 ena_update_hints(adapter, hints);
4632                 break;
4633         default:
4634                 netif_err(adapter, drv, adapter->netdev,
4635                           "Invalid aenq notification link state %d\n",
4636                           aenq_e->aenq_common_desc.syndrome);
4637         }
4638 }
4639
4640 /* This handler will called for unknown event group or unimplemented handlers*/
4641 static void unimplemented_aenq_handler(void *data,
4642                                        struct ena_admin_aenq_entry *aenq_e)
4643 {
4644         struct ena_adapter *adapter = (struct ena_adapter *)data;
4645
4646         netif_err(adapter, drv, adapter->netdev,
4647                   "Unknown event was received or event with unimplemented handler\n");
4648 }
4649
4650 static struct ena_aenq_handlers aenq_handlers = {
4651         .handlers = {
4652                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4653                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
4654                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4655         },
4656         .unimplemented_handler = unimplemented_aenq_handler
4657 };
4658
4659 module_init(ena_init);
4660 module_exit(ena_cleanup);