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 ENA_NAPI_BUDGET 64
  35
  36 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
  37                 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
  38
  39 static struct ena_aenq_handlers aenq_handlers;
  40
  41 static struct workqueue_struct *ena_wq;
  42
  43 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
  44
  45 static int ena_rss_init_default(struct ena_adapter *adapter);
  46 static void check_for_admin_com_state(struct ena_adapter *adapter);
  47 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
  48 static int ena_restore_device(struct ena_adapter *adapter);
  49
  50 static void ena_init_io_rings(struct ena_adapter *adapter,
  51                               int first_index, int count);
  52 static void ena_init_napi_in_range(struct ena_adapter *adapter, int first_index,
  53                                    int count);
  54 static void ena_del_napi_in_range(struct ena_adapter *adapter, int first_index,
  55                                   int count);
  56 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid);
  57 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
  58                                            int first_index,
  59                                            int count);
  60 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid);
  61 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid);
  62 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget);
  63 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter);
  64 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter);
  65 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
  66                                       int first_index, int count);
  67 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
  68                                      int first_index, int count);
  69 static int ena_up(struct ena_adapter *adapter);
  70 static void ena_down(struct ena_adapter *adapter);
  71 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
  72                                  struct ena_ring *rx_ring);
  73 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
  74                                       struct ena_ring *rx_ring);
  75 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
  76                               struct ena_tx_buffer *tx_info);
  77 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
  78                                             int first_index, int count);
  79
  80 /* Increase a stat by cnt while holding syncp seqlock on 32bit machines */
  81 static void ena_increase_stat(u64 *statp, u64 cnt,
  82                               struct u64_stats_sync *syncp)
  83 {
  84         u64_stats_update_begin(syncp);
  85         (*statp) += cnt;
  86         u64_stats_update_end(syncp);
  87 }
  88
  89 static void ena_ring_tx_doorbell(struct ena_ring *tx_ring)
  90 {
  91         ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
  92         ena_increase_stat(&tx_ring->tx_stats.doorbells, 1, &tx_ring->syncp);
  93 }
  94
  95 static void ena_tx_timeout(struct net_device *dev, unsigned int txqueue)
  96 {
  97         struct ena_adapter *adapter = netdev_priv(dev);
  98
  99         /* Change the state of the device to trigger reset
 100          * Check that we are not in the middle or a trigger already
 101          */
 102
 103         if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
 104                 return;
 105
 106         adapter->reset_reason = ENA_REGS_RESET_OS_NETDEV_WD;
 107         ena_increase_stat(&adapter->dev_stats.tx_timeout, 1, &adapter->syncp);
 108
 109         netif_err(adapter, tx_err, dev, "Transmit time out\n");
 110 }
 111
 112 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
 113 {
 114         int i;
 115
 116         for (i = 0; i < adapter->num_io_queues; i++)
 117                 adapter->rx_ring[i].mtu = mtu;
 118 }
 119
 120 static int ena_change_mtu(struct net_device *dev, int new_mtu)
 121 {
 122         struct ena_adapter *adapter = netdev_priv(dev);
 123         int ret;
 124
 125         ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
 126         if (!ret) {
 127                 netif_dbg(adapter, drv, dev, "Set MTU to %d\n", new_mtu);
 128                 update_rx_ring_mtu(adapter, new_mtu);
 129                 dev->mtu = new_mtu;
 130         } else {
 131                 netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
 132                           new_mtu);
 133         }
 134
 135         return ret;
 136 }
 137
 138 static int ena_xmit_common(struct net_device *dev,
 139                            struct ena_ring *ring,
 140                            struct ena_tx_buffer *tx_info,
 141                            struct ena_com_tx_ctx *ena_tx_ctx,
 142                            u16 next_to_use,
 143                            u32 bytes)
 144 {
 145         struct ena_adapter *adapter = netdev_priv(dev);
 146         int rc, nb_hw_desc;
 147
 148         if (unlikely(ena_com_is_doorbell_needed(ring->ena_com_io_sq,
 149                                                 ena_tx_ctx))) {
 150                 netif_dbg(adapter, tx_queued, dev,
 151                           "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
 152                           ring->qid);
 153                 ena_ring_tx_doorbell(ring);
 154         }
 155
 156         /* prepare the packet's descriptors to dma engine */
 157         rc = ena_com_prepare_tx(ring->ena_com_io_sq, ena_tx_ctx,
 158                                 &nb_hw_desc);
 159
 160         /* In case there isn't enough space in the queue for the packet,
 161          * we simply drop it. All other failure reasons of
 162          * ena_com_prepare_tx() are fatal and therefore require a device reset.
 163          */
 164         if (unlikely(rc)) {
 165                 netif_err(adapter, tx_queued, dev,
 166                           "Failed to prepare tx bufs\n");
 167                 ena_increase_stat(&ring->tx_stats.prepare_ctx_err, 1,
 168                                   &ring->syncp);
 169                 if (rc != -ENOMEM) {
 170                         adapter->reset_reason =
 171                                 ENA_REGS_RESET_DRIVER_INVALID_STATE;
 172                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
 173                 }
 174                 return rc;
 175         }
 176
 177         u64_stats_update_begin(&ring->syncp);
 178         ring->tx_stats.cnt++;
 179         ring->tx_stats.bytes += bytes;
 180         u64_stats_update_end(&ring->syncp);
 181
 182         tx_info->tx_descs = nb_hw_desc;
 183         tx_info->last_jiffies = jiffies;
 184         tx_info->print_once = 0;
 185
 186         ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
 187                                                  ring->ring_size);
 188         return 0;
 189 }
 190
 191 /* This is the XDP napi callback. XDP queues use a separate napi callback
 192  * than Rx/Tx queues.
 193  */
 194 static int ena_xdp_io_poll(struct napi_struct *napi, int budget)
 195 {
 196         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
 197         u32 xdp_work_done, xdp_budget;
 198         struct ena_ring *xdp_ring;
 199         int napi_comp_call = 0;
 200         int ret;
 201
 202         xdp_ring = ena_napi->xdp_ring;
 203
 204         xdp_budget = budget;
 205
 206         if (!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags) ||
 207             test_bit(ENA_FLAG_TRIGGER_RESET, &xdp_ring->adapter->flags)) {
 208                 napi_complete_done(napi, 0);
 209                 return 0;
 210         }
 211
 212         xdp_work_done = ena_clean_xdp_irq(xdp_ring, xdp_budget);
 213
 214         /* If the device is about to reset or down, avoid unmask
 215          * the interrupt and return 0 so NAPI won't reschedule
 216          */
 217         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags))) {
 218                 napi_complete_done(napi, 0);
 219                 ret = 0;
 220         } else if (xdp_budget > xdp_work_done) {
 221                 napi_comp_call = 1;
 222                 if (napi_complete_done(napi, xdp_work_done))
 223                         ena_unmask_interrupt(xdp_ring, NULL);
 224                 ena_update_ring_numa_node(xdp_ring, NULL);
 225                 ret = xdp_work_done;
 226         } else {
 227                 ret = xdp_budget;
 228         }
 229
 230         u64_stats_update_begin(&xdp_ring->syncp);
 231         xdp_ring->tx_stats.napi_comp += napi_comp_call;
 232         xdp_ring->tx_stats.tx_poll++;
 233         u64_stats_update_end(&xdp_ring->syncp);
 234         xdp_ring->tx_stats.last_napi_jiffies = jiffies;
 235
 236         return ret;
 237 }
 238
 239 static int ena_xdp_tx_map_frame(struct ena_ring *xdp_ring,
 240                                 struct ena_tx_buffer *tx_info,
 241                                 struct xdp_frame *xdpf,
 242                                 struct ena_com_tx_ctx *ena_tx_ctx)
 243 {
 244         struct ena_adapter *adapter = xdp_ring->adapter;
 245         struct ena_com_buf *ena_buf;
 246         int push_len = 0;
 247         dma_addr_t dma;
 248         void *data;
 249         u32 size;
 250
 251         tx_info->xdpf = xdpf;
 252         data = tx_info->xdpf->data;
 253         size = tx_info->xdpf->len;
 254
 255         if (xdp_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
 256                 /* Designate part of the packet for LLQ */
 257                 push_len = min_t(u32, size, xdp_ring->tx_max_header_size);
 258
 259                 ena_tx_ctx->push_header = data;
 260
 261                 size -= push_len;
 262                 data += push_len;
 263         }
 264
 265         ena_tx_ctx->header_len = push_len;
 266
 267         if (size > 0) {
 268                 dma = dma_map_single(xdp_ring->dev,
 269                                      data,
 270                                      size,
 271                                      DMA_TO_DEVICE);
 272                 if (unlikely(dma_mapping_error(xdp_ring->dev, dma)))
 273                         goto error_report_dma_error;
 274
 275                 tx_info->map_linear_data = 0;
 276
 277                 ena_buf = tx_info->bufs;
 278                 ena_buf->paddr = dma;
 279                 ena_buf->len = size;
 280
 281                 ena_tx_ctx->ena_bufs = ena_buf;
 282                 ena_tx_ctx->num_bufs = tx_info->num_of_bufs = 1;
 283         }
 284
 285         return 0;
 286
 287 error_report_dma_error:
 288         ena_increase_stat(&xdp_ring->tx_stats.dma_mapping_err, 1,
 289                           &xdp_ring->syncp);
 290         netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map xdp buff\n");
 291
 292         return -EINVAL;
 293 }
 294
 295 static int ena_xdp_xmit_frame(struct ena_ring *xdp_ring,
 296                               struct net_device *dev,
 297                               struct xdp_frame *xdpf,
 298                               int flags)
 299 {
 300         struct ena_com_tx_ctx ena_tx_ctx = {};
 301         struct ena_tx_buffer *tx_info;
 302         u16 next_to_use, req_id;
 303         int rc;
 304
 305         next_to_use = xdp_ring->next_to_use;
 306         req_id = xdp_ring->free_ids[next_to_use];
 307         tx_info = &xdp_ring->tx_buffer_info[req_id];
 308         tx_info->num_of_bufs = 0;
 309
 310         rc = ena_xdp_tx_map_frame(xdp_ring, tx_info, xdpf, &ena_tx_ctx);
 311         if (unlikely(rc))
 312                 return rc;
 313
 314         ena_tx_ctx.req_id = req_id;
 315
 316         rc = ena_xmit_common(dev,
 317                              xdp_ring,
 318                              tx_info,
 319                              &ena_tx_ctx,
 320                              next_to_use,
 321                              xdpf->len);
 322         if (rc)
 323                 goto error_unmap_dma;
 324
 325         /* trigger the dma engine. ena_ring_tx_doorbell()
 326          * calls a memory barrier inside it.
 327          */
 328         if (flags & XDP_XMIT_FLUSH)
 329                 ena_ring_tx_doorbell(xdp_ring);
 330
 331         return rc;
 332
 333 error_unmap_dma:
 334         ena_unmap_tx_buff(xdp_ring, tx_info);
 335         tx_info->xdpf = NULL;
 336         return rc;
 337 }
 338
 339 static int ena_xdp_xmit(struct net_device *dev, int n,
 340                         struct xdp_frame **frames, u32 flags)
 341 {
 342         struct ena_adapter *adapter = netdev_priv(dev);
 343         struct ena_ring *xdp_ring;
 344         int qid, i, nxmit = 0;
 345
 346         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 347                 return -EINVAL;
 348
 349         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
 350                 return -ENETDOWN;
 351
 352         /* We assume that all rings have the same XDP program */
 353         if (!READ_ONCE(adapter->rx_ring->xdp_bpf_prog))
 354                 return -ENXIO;
 355
 356         qid = smp_processor_id() % adapter->xdp_num_queues;
 357         qid += adapter->xdp_first_ring;
 358         xdp_ring = &adapter->tx_ring[qid];
 359
 360         /* Other CPU ids might try to send thorugh this queue */
 361         spin_lock(&xdp_ring->xdp_tx_lock);
 362
 363         for (i = 0; i < n; i++) {
 364                 if (ena_xdp_xmit_frame(xdp_ring, dev, frames[i], 0))
 365                         break;
 366                 nxmit++;
 367         }
 368
 369         /* Ring doorbell to make device aware of the packets */
 370         if (flags & XDP_XMIT_FLUSH)
 371                 ena_ring_tx_doorbell(xdp_ring);
 372
 373         spin_unlock(&xdp_ring->xdp_tx_lock);
 374
 375         /* Return number of packets sent */
 376         return nxmit;
 377 }
 378
 379 static int ena_xdp_execute(struct ena_ring *rx_ring, struct xdp_buff *xdp)
 380 {
 381         struct bpf_prog *xdp_prog;
 382         struct ena_ring *xdp_ring;
 383         u32 verdict = XDP_PASS;
 384         struct xdp_frame *xdpf;
 385         u64 *xdp_stat;
 386
 387         xdp_prog = READ_ONCE(rx_ring->xdp_bpf_prog);
 388
 389         if (!xdp_prog)
 390                 goto out;
 391
 392         verdict = bpf_prog_run_xdp(xdp_prog, xdp);
 393
 394         switch (verdict) {
 395         case XDP_TX:
 396                 xdpf = xdp_convert_buff_to_frame(xdp);
 397                 if (unlikely(!xdpf)) {
 398                         trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 399                         xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 400                         verdict = XDP_ABORTED;
 401                         break;
 402                 }
 403
 404                 /* Find xmit queue */
 405                 xdp_ring = rx_ring->xdp_ring;
 406
 407                 /* The XDP queues are shared between XDP_TX and XDP_REDIRECT */
 408                 spin_lock(&xdp_ring->xdp_tx_lock);
 409
 410                 if (ena_xdp_xmit_frame(xdp_ring, rx_ring->netdev, xdpf,
 411                                        XDP_XMIT_FLUSH))
 412                         xdp_return_frame(xdpf);
 413
 414                 spin_unlock(&xdp_ring->xdp_tx_lock);
 415                 xdp_stat = &rx_ring->rx_stats.xdp_tx;
 416                 break;
 417         case XDP_REDIRECT:
 418                 if (likely(!xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog))) {
 419                         xdp_stat = &rx_ring->rx_stats.xdp_redirect;
 420                         break;
 421                 }
 422                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 423                 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 424                 verdict = XDP_ABORTED;
 425                 break;
 426         case XDP_ABORTED:
 427                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 428                 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 429                 break;
 430         case XDP_DROP:
 431                 xdp_stat = &rx_ring->rx_stats.xdp_drop;
 432                 break;
 433         case XDP_PASS:
 434                 xdp_stat = &rx_ring->rx_stats.xdp_pass;
 435                 break;
 436         default:
 437                 bpf_warn_invalid_xdp_action(verdict);
 438                 xdp_stat = &rx_ring->rx_stats.xdp_invalid;
 439         }
 440
 441         ena_increase_stat(xdp_stat, 1, &rx_ring->syncp);
 442 out:
 443         return verdict;
 444 }
 445
 446 static void ena_init_all_xdp_queues(struct ena_adapter *adapter)
 447 {
 448         adapter->xdp_first_ring = adapter->num_io_queues;
 449         adapter->xdp_num_queues = adapter->num_io_queues;
 450
 451         ena_init_io_rings(adapter,
 452                           adapter->xdp_first_ring,
 453                           adapter->xdp_num_queues);
 454 }
 455
 456 static int ena_setup_and_create_all_xdp_queues(struct ena_adapter *adapter)
 457 {
 458         int rc = 0;
 459
 460         rc = ena_setup_tx_resources_in_range(adapter, adapter->xdp_first_ring,
 461                                              adapter->xdp_num_queues);
 462         if (rc)
 463                 goto setup_err;
 464
 465         rc = ena_create_io_tx_queues_in_range(adapter,
 466                                               adapter->xdp_first_ring,
 467                                               adapter->xdp_num_queues);
 468         if (rc)
 469                 goto create_err;
 470
 471         return 0;
 472
 473 create_err:
 474         ena_free_all_io_tx_resources(adapter);
 475 setup_err:
 476         return rc;
 477 }
 478
 479 /* Provides a way for both kernel and bpf-prog to know
 480  * more about the RX-queue a given XDP frame arrived on.
 481  */
 482 static int ena_xdp_register_rxq_info(struct ena_ring *rx_ring)
 483 {
 484         int rc;
 485
 486         rc = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev, rx_ring->qid, 0);
 487
 488         if (rc) {
 489                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 490                           "Failed to register xdp rx queue info. RX queue num %d rc: %d\n",
 491                           rx_ring->qid, rc);
 492                 goto err;
 493         }
 494
 495         rc = xdp_rxq_info_reg_mem_model(&rx_ring->xdp_rxq, MEM_TYPE_PAGE_SHARED,
 496                                         NULL);
 497
 498         if (rc) {
 499                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 500                           "Failed to register xdp rx queue info memory model. RX queue num %d rc: %d\n",
 501                           rx_ring->qid, rc);
 502                 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 503         }
 504
 505 err:
 506         return rc;
 507 }
 508
 509 static void ena_xdp_unregister_rxq_info(struct ena_ring *rx_ring)
 510 {
 511         xdp_rxq_info_unreg_mem_model(&rx_ring->xdp_rxq);
 512         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 513 }
 514
 515 static void ena_xdp_exchange_program_rx_in_range(struct ena_adapter *adapter,
 516                                                  struct bpf_prog *prog,
 517                                                  int first, int count)
 518 {
 519         struct ena_ring *rx_ring;
 520         int i = 0;
 521
 522         for (i = first; i < count; i++) {
 523                 rx_ring = &adapter->rx_ring[i];
 524                 xchg(&rx_ring->xdp_bpf_prog, prog);
 525                 if (prog) {
 526                         ena_xdp_register_rxq_info(rx_ring);
 527                         rx_ring->rx_headroom = XDP_PACKET_HEADROOM;
 528                 } else {
 529                         ena_xdp_unregister_rxq_info(rx_ring);
 530                         rx_ring->rx_headroom = NET_SKB_PAD;
 531                 }
 532         }
 533 }
 534
 535 static void ena_xdp_exchange_program(struct ena_adapter *adapter,
 536                                      struct bpf_prog *prog)
 537 {
 538         struct bpf_prog *old_bpf_prog = xchg(&adapter->xdp_bpf_prog, prog);
 539
 540         ena_xdp_exchange_program_rx_in_range(adapter,
 541                                              prog,
 542                                              0,
 543                                              adapter->num_io_queues);
 544
 545         if (old_bpf_prog)
 546                 bpf_prog_put(old_bpf_prog);
 547 }
 548
 549 static int ena_destroy_and_free_all_xdp_queues(struct ena_adapter *adapter)
 550 {
 551         bool was_up;
 552         int rc;
 553
 554         was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 555
 556         if (was_up)
 557                 ena_down(adapter);
 558
 559         adapter->xdp_first_ring = 0;
 560         adapter->xdp_num_queues = 0;
 561         ena_xdp_exchange_program(adapter, NULL);
 562         if (was_up) {
 563                 rc = ena_up(adapter);
 564                 if (rc)
 565                         return rc;
 566         }
 567         return 0;
 568 }
 569
 570 static int ena_xdp_set(struct net_device *netdev, struct netdev_bpf *bpf)
 571 {
 572         struct ena_adapter *adapter = netdev_priv(netdev);
 573         struct bpf_prog *prog = bpf->prog;
 574         struct bpf_prog *old_bpf_prog;
 575         int rc, prev_mtu;
 576         bool is_up;
 577
 578         is_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 579         rc = ena_xdp_allowed(adapter);
 580         if (rc == ENA_XDP_ALLOWED) {
 581                 old_bpf_prog = adapter->xdp_bpf_prog;
 582                 if (prog) {
 583                         if (!is_up) {
 584                                 ena_init_all_xdp_queues(adapter);
 585                         } else if (!old_bpf_prog) {
 586                                 ena_down(adapter);
 587                                 ena_init_all_xdp_queues(adapter);
 588                         }
 589                         ena_xdp_exchange_program(adapter, prog);
 590
 591                         if (is_up && !old_bpf_prog) {
 592                                 rc = ena_up(adapter);
 593                                 if (rc)
 594                                         return rc;
 595                         }
 596                 } else if (old_bpf_prog) {
 597                         rc = ena_destroy_and_free_all_xdp_queues(adapter);
 598                         if (rc)
 599                                 return rc;
 600                 }
 601
 602                 prev_mtu = netdev->max_mtu;
 603                 netdev->max_mtu = prog ? ENA_XDP_MAX_MTU : adapter->max_mtu;
 604
 605                 if (!old_bpf_prog)
 606                         netif_info(adapter, drv, adapter->netdev,
 607                                    "XDP program is set, changing the max_mtu from %d to %d",
 608                                    prev_mtu, netdev->max_mtu);
 609
 610         } else if (rc == ENA_XDP_CURRENT_MTU_TOO_LARGE) {
 611                 netif_err(adapter, drv, adapter->netdev,
 612                           "Failed to set xdp program, the current MTU (%d) is larger than the maximum allowed MTU (%lu) while xdp is on",
 613                           netdev->mtu, ENA_XDP_MAX_MTU);
 614                 NL_SET_ERR_MSG_MOD(bpf->extack,
 615                                    "Failed to set xdp program, the current MTU is larger than the maximum allowed MTU. Check the dmesg for more info");
 616                 return -EINVAL;
 617         } else if (rc == ENA_XDP_NO_ENOUGH_QUEUES) {
 618                 netif_err(adapter, drv, adapter->netdev,
 619                           "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",
 620                           adapter->num_io_queues, adapter->max_num_io_queues);
 621                 NL_SET_ERR_MSG_MOD(bpf->extack,
 622                                    "Failed to set xdp program, there is no enough space for allocating XDP queues, Check the dmesg for more info");
 623                 return -EINVAL;
 624         }
 625
 626         return 0;
 627 }
 628
 629 /* This is the main xdp callback, it's used by the kernel to set/unset the xdp
 630  * program as well as to query the current xdp program id.
 631  */
 632 static int ena_xdp(struct net_device *netdev, struct netdev_bpf *bpf)
 633 {
 634         switch (bpf->command) {
 635         case XDP_SETUP_PROG:
 636                 return ena_xdp_set(netdev, bpf);
 637         default:
 638                 return -EINVAL;
 639         }
 640         return 0;
 641 }
 642
 643 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
 644 {
 645 #ifdef CONFIG_RFS_ACCEL
 646         u32 i;
 647         int rc;
 648
 649         adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_io_queues);
 650         if (!adapter->netdev->rx_cpu_rmap)
 651                 return -ENOMEM;
 652         for (i = 0; i < adapter->num_io_queues; i++) {
 653                 int irq_idx = ENA_IO_IRQ_IDX(i);
 654
 655                 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
 656                                       pci_irq_vector(adapter->pdev, irq_idx));
 657                 if (rc) {
 658                         free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
 659                         adapter->netdev->rx_cpu_rmap = NULL;
 660                         return rc;
 661                 }
 662         }
 663 #endif /* CONFIG_RFS_ACCEL */
 664         return 0;
 665 }
 666
 667 static void ena_init_io_rings_common(struct ena_adapter *adapter,
 668                                      struct ena_ring *ring, u16 qid)
 669 {
 670         ring->qid = qid;
 671         ring->pdev = adapter->pdev;
 672         ring->dev = &adapter->pdev->dev;
 673         ring->netdev = adapter->netdev;
 674         ring->napi = &adapter->ena_napi[qid].napi;
 675         ring->adapter = adapter;
 676         ring->ena_dev = adapter->ena_dev;
 677         ring->per_napi_packets = 0;
 678         ring->cpu = 0;
 679         ring->no_interrupt_event_cnt = 0;
 680         u64_stats_init(&ring->syncp);
 681 }
 682
 683 static void ena_init_io_rings(struct ena_adapter *adapter,
 684                               int first_index, int count)
 685 {
 686         struct ena_com_dev *ena_dev;
 687         struct ena_ring *txr, *rxr;
 688         int i;
 689
 690         ena_dev = adapter->ena_dev;
 691
 692         for (i = first_index; i < first_index + count; i++) {
 693                 txr = &adapter->tx_ring[i];
 694                 rxr = &adapter->rx_ring[i];
 695
 696                 /* TX common ring state */
 697                 ena_init_io_rings_common(adapter, txr, i);
 698
 699                 /* TX specific ring state */
 700                 txr->ring_size = adapter->requested_tx_ring_size;
 701                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
 702                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
 703                 txr->sgl_size = adapter->max_tx_sgl_size;
 704                 txr->smoothed_interval =
 705                         ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
 706                 txr->disable_meta_caching = adapter->disable_meta_caching;
 707                 spin_lock_init(&txr->xdp_tx_lock);
 708
 709                 /* Don't init RX queues for xdp queues */
 710                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
 711                         /* RX common ring state */
 712                         ena_init_io_rings_common(adapter, rxr, i);
 713
 714                         /* RX specific ring state */
 715                         rxr->ring_size = adapter->requested_rx_ring_size;
 716                         rxr->rx_copybreak = adapter->rx_copybreak;
 717                         rxr->sgl_size = adapter->max_rx_sgl_size;
 718                         rxr->smoothed_interval =
 719                                 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
 720                         rxr->empty_rx_queue = 0;
 721                         rxr->rx_headroom = NET_SKB_PAD;
 722                         adapter->ena_napi[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
 723                         rxr->xdp_ring = &adapter->tx_ring[i + adapter->num_io_queues];
 724                 }
 725         }
 726 }
 727
 728 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
 729  * @adapter: network interface device structure
 730  * @qid: queue index
 731  *
 732  * Return 0 on success, negative on failure
 733  */
 734 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
 735 {
 736         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 737         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 738         int size, i, node;
 739
 740         if (tx_ring->tx_buffer_info) {
 741                 netif_err(adapter, ifup,
 742                           adapter->netdev, "tx_buffer_info info is not NULL");
 743                 return -EEXIST;
 744         }
 745
 746         size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
 747         node = cpu_to_node(ena_irq->cpu);
 748
 749         tx_ring->tx_buffer_info = vzalloc_node(size, node);
 750         if (!tx_ring->tx_buffer_info) {
 751                 tx_ring->tx_buffer_info = vzalloc(size);
 752                 if (!tx_ring->tx_buffer_info)
 753                         goto err_tx_buffer_info;
 754         }
 755
 756         size = sizeof(u16) * tx_ring->ring_size;
 757         tx_ring->free_ids = vzalloc_node(size, node);
 758         if (!tx_ring->free_ids) {
 759                 tx_ring->free_ids = vzalloc(size);
 760                 if (!tx_ring->free_ids)
 761                         goto err_tx_free_ids;
 762         }
 763
 764         size = tx_ring->tx_max_header_size;
 765         tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
 766         if (!tx_ring->push_buf_intermediate_buf) {
 767                 tx_ring->push_buf_intermediate_buf = vzalloc(size);
 768                 if (!tx_ring->push_buf_intermediate_buf)
 769                         goto err_push_buf_intermediate_buf;
 770         }
 771
 772         /* Req id ring for TX out of order completions */
 773         for (i = 0; i < tx_ring->ring_size; i++)
 774                 tx_ring->free_ids[i] = i;
 775
 776         /* Reset tx statistics */
 777         memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
 778
 779         tx_ring->next_to_use = 0;
 780         tx_ring->next_to_clean = 0;
 781         tx_ring->cpu = ena_irq->cpu;
 782         return 0;
 783
 784 err_push_buf_intermediate_buf:
 785         vfree(tx_ring->free_ids);
 786         tx_ring->free_ids = NULL;
 787 err_tx_free_ids:
 788         vfree(tx_ring->tx_buffer_info);
 789         tx_ring->tx_buffer_info = NULL;
 790 err_tx_buffer_info:
 791         return -ENOMEM;
 792 }
 793
 794 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
 795  * @adapter: network interface device structure
 796  * @qid: queue index
 797  *
 798  * Free all transmit software resources
 799  */
 800 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
 801 {
 802         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 803
 804         vfree(tx_ring->tx_buffer_info);
 805         tx_ring->tx_buffer_info = NULL;
 806
 807         vfree(tx_ring->free_ids);
 808         tx_ring->free_ids = NULL;
 809
 810         vfree(tx_ring->push_buf_intermediate_buf);
 811         tx_ring->push_buf_intermediate_buf = NULL;
 812 }
 813
 814 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
 815                                            int first_index,
 816                                            int count)
 817 {
 818         int i, rc = 0;
 819
 820         for (i = first_index; i < first_index + count; i++) {
 821                 rc = ena_setup_tx_resources(adapter, i);
 822                 if (rc)
 823                         goto err_setup_tx;
 824         }
 825
 826         return 0;
 827
 828 err_setup_tx:
 829
 830         netif_err(adapter, ifup, adapter->netdev,
 831                   "Tx queue %d: allocation failed\n", i);
 832
 833         /* rewind the index freeing the rings as we go */
 834         while (first_index < i--)
 835                 ena_free_tx_resources(adapter, i);
 836         return rc;
 837 }
 838
 839 static void ena_free_all_io_tx_resources_in_range(struct ena_adapter *adapter,
 840                                                   int first_index, int count)
 841 {
 842         int i;
 843
 844         for (i = first_index; i < first_index + count; i++)
 845                 ena_free_tx_resources(adapter, i);
 846 }
 847
 848 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
 849  * @adapter: board private structure
 850  *
 851  * Free all transmit software resources
 852  */
 853 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
 854 {
 855         ena_free_all_io_tx_resources_in_range(adapter,
 856                                               0,
 857                                               adapter->xdp_num_queues +
 858                                               adapter->num_io_queues);
 859 }
 860
 861 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
 862  * @adapter: network interface device structure
 863  * @qid: queue index
 864  *
 865  * Returns 0 on success, negative on failure
 866  */
 867 static int ena_setup_rx_resources(struct ena_adapter *adapter,
 868                                   u32 qid)
 869 {
 870         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 871         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 872         int size, node, i;
 873
 874         if (rx_ring->rx_buffer_info) {
 875                 netif_err(adapter, ifup, adapter->netdev,
 876                           "rx_buffer_info is not NULL");
 877                 return -EEXIST;
 878         }
 879
 880         /* alloc extra element so in rx path
 881          * we can always prefetch rx_info + 1
 882          */
 883         size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
 884         node = cpu_to_node(ena_irq->cpu);
 885
 886         rx_ring->rx_buffer_info = vzalloc_node(size, node);
 887         if (!rx_ring->rx_buffer_info) {
 888                 rx_ring->rx_buffer_info = vzalloc(size);
 889                 if (!rx_ring->rx_buffer_info)
 890                         return -ENOMEM;
 891         }
 892
 893         size = sizeof(u16) * rx_ring->ring_size;
 894         rx_ring->free_ids = vzalloc_node(size, node);
 895         if (!rx_ring->free_ids) {
 896                 rx_ring->free_ids = vzalloc(size);
 897                 if (!rx_ring->free_ids) {
 898                         vfree(rx_ring->rx_buffer_info);
 899                         rx_ring->rx_buffer_info = NULL;
 900                         return -ENOMEM;
 901                 }
 902         }
 903
 904         /* Req id ring for receiving RX pkts out of order */
 905         for (i = 0; i < rx_ring->ring_size; i++)
 906                 rx_ring->free_ids[i] = i;
 907
 908         /* Reset rx statistics */
 909         memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
 910
 911         rx_ring->next_to_clean = 0;
 912         rx_ring->next_to_use = 0;
 913         rx_ring->cpu = ena_irq->cpu;
 914
 915         return 0;
 916 }
 917
 918 /* ena_free_rx_resources - Free I/O Rx Resources
 919  * @adapter: network interface device structure
 920  * @qid: queue index
 921  *
 922  * Free all receive software resources
 923  */
 924 static void ena_free_rx_resources(struct ena_adapter *adapter,
 925                                   u32 qid)
 926 {
 927         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 928
 929         vfree(rx_ring->rx_buffer_info);
 930         rx_ring->rx_buffer_info = NULL;
 931
 932         vfree(rx_ring->free_ids);
 933         rx_ring->free_ids = NULL;
 934 }
 935
 936 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
 937  * @adapter: board private structure
 938  *
 939  * Return 0 on success, negative on failure
 940  */
 941 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
 942 {
 943         int i, rc = 0;
 944
 945         for (i = 0; i < adapter->num_io_queues; i++) {
 946                 rc = ena_setup_rx_resources(adapter, i);
 947                 if (rc)
 948                         goto err_setup_rx;
 949         }
 950
 951         return 0;
 952
 953 err_setup_rx:
 954
 955         netif_err(adapter, ifup, adapter->netdev,
 956                   "Rx queue %d: allocation failed\n", i);
 957
 958         /* rewind the index freeing the rings as we go */
 959         while (i--)
 960                 ena_free_rx_resources(adapter, i);
 961         return rc;
 962 }
 963
 964 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
 965  * @adapter: board private structure
 966  *
 967  * Free all receive software resources
 968  */
 969 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
 970 {
 971         int i;
 972
 973         for (i = 0; i < adapter->num_io_queues; i++)
 974                 ena_free_rx_resources(adapter, i);
 975 }
 976
 977 static struct page *ena_alloc_map_page(struct ena_ring *rx_ring,
 978                                        dma_addr_t *dma)
 979 {
 980         struct page *page;
 981
 982         /* This would allocate the page on the same NUMA node the executing code
 983          * is running on.
 984          */
 985         page = dev_alloc_page();
 986         if (!page) {
 987                 ena_increase_stat(&rx_ring->rx_stats.page_alloc_fail, 1,
 988                                   &rx_ring->syncp);
 989                 return ERR_PTR(-ENOSPC);
 990         }
 991
 992         /* To enable NIC-side port-mirroring, AKA SPAN port,
 993          * we make the buffer readable from the nic as well
 994          */
 995         *dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
 996                             DMA_BIDIRECTIONAL);
 997         if (unlikely(dma_mapping_error(rx_ring->dev, *dma))) {
 998                 ena_increase_stat(&rx_ring->rx_stats.dma_mapping_err, 1,
 999                                   &rx_ring->syncp);
1000                 __free_page(page);
1001                 return ERR_PTR(-EIO);
1002         }
1003
1004         return page;
1005 }
1006
1007 static int ena_alloc_rx_buffer(struct ena_ring *rx_ring,
1008                                struct ena_rx_buffer *rx_info)
1009 {
1010         int headroom = rx_ring->rx_headroom;
1011         struct ena_com_buf *ena_buf;
1012         struct page *page;
1013         dma_addr_t dma;
1014         int tailroom;
1015
1016         /* restore page offset value in case it has been changed by device */
1017         rx_info->page_offset = headroom;
1018
1019         /* if previous allocated page is not used */
1020         if (unlikely(rx_info->page))
1021                 return 0;
1022
1023         /* We handle DMA here */
1024         page = ena_alloc_map_page(rx_ring, &dma);
1025         if (unlikely(IS_ERR(page)))
1026                 return PTR_ERR(page);
1027
1028         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1029                   "Allocate page %p, rx_info %p\n", page, rx_info);
1030
1031         tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1032
1033         rx_info->page = page;
1034         ena_buf = &rx_info->ena_buf;
1035         ena_buf->paddr = dma + headroom;
1036         ena_buf->len = ENA_PAGE_SIZE - headroom - tailroom;
1037
1038         return 0;
1039 }
1040
1041 static void ena_unmap_rx_buff(struct ena_ring *rx_ring,
1042                               struct ena_rx_buffer *rx_info)
1043 {
1044         struct ena_com_buf *ena_buf = &rx_info->ena_buf;
1045
1046         dma_unmap_page(rx_ring->dev, ena_buf->paddr - rx_ring->rx_headroom,
1047                        ENA_PAGE_SIZE,
1048                        DMA_BIDIRECTIONAL);
1049 }
1050
1051 static void ena_free_rx_page(struct ena_ring *rx_ring,
1052                              struct ena_rx_buffer *rx_info)
1053 {
1054         struct page *page = rx_info->page;
1055
1056         if (unlikely(!page)) {
1057                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1058                            "Trying to free unallocated buffer\n");
1059                 return;
1060         }
1061
1062         ena_unmap_rx_buff(rx_ring, rx_info);
1063
1064         __free_page(page);
1065         rx_info->page = NULL;
1066 }
1067
1068 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
1069 {
1070         u16 next_to_use, req_id;
1071         u32 i;
1072         int rc;
1073
1074         next_to_use = rx_ring->next_to_use;
1075
1076         for (i = 0; i < num; i++) {
1077                 struct ena_rx_buffer *rx_info;
1078
1079                 req_id = rx_ring->free_ids[next_to_use];
1080
1081                 rx_info = &rx_ring->rx_buffer_info[req_id];
1082
1083                 rc = ena_alloc_rx_buffer(rx_ring, rx_info);
1084                 if (unlikely(rc < 0)) {
1085                         netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1086                                    "Failed to allocate buffer for rx queue %d\n",
1087                                    rx_ring->qid);
1088                         break;
1089                 }
1090                 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1091                                                 &rx_info->ena_buf,
1092                                                 req_id);
1093                 if (unlikely(rc)) {
1094                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1095                                    "Failed to add buffer for rx queue %d\n",
1096                                    rx_ring->qid);
1097                         break;
1098                 }
1099                 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1100                                                    rx_ring->ring_size);
1101         }
1102
1103         if (unlikely(i < num)) {
1104                 ena_increase_stat(&rx_ring->rx_stats.refil_partial, 1,
1105                                   &rx_ring->syncp);
1106                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1107                            "Refilled rx qid %d with only %d buffers (from %d)\n",
1108                            rx_ring->qid, i, num);
1109         }
1110
1111         /* ena_com_write_sq_doorbell issues a wmb() */
1112         if (likely(i))
1113                 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1114
1115         rx_ring->next_to_use = next_to_use;
1116
1117         return i;
1118 }
1119
1120 static void ena_free_rx_bufs(struct ena_adapter *adapter,
1121                              u32 qid)
1122 {
1123         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1124         u32 i;
1125
1126         for (i = 0; i < rx_ring->ring_size; i++) {
1127                 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1128
1129                 if (rx_info->page)
1130                         ena_free_rx_page(rx_ring, rx_info);
1131         }
1132 }
1133
1134 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
1135  * @adapter: board private structure
1136  */
1137 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1138 {
1139         struct ena_ring *rx_ring;
1140         int i, rc, bufs_num;
1141
1142         for (i = 0; i < adapter->num_io_queues; i++) {
1143                 rx_ring = &adapter->rx_ring[i];
1144                 bufs_num = rx_ring->ring_size - 1;
1145                 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1146
1147                 if (unlikely(rc != bufs_num))
1148                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1149                                    "Refilling Queue %d failed. allocated %d buffers from: %d\n",
1150                                    i, rc, bufs_num);
1151         }
1152 }
1153
1154 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
1155 {
1156         int i;
1157
1158         for (i = 0; i < adapter->num_io_queues; i++)
1159                 ena_free_rx_bufs(adapter, i);
1160 }
1161
1162 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
1163                               struct ena_tx_buffer *tx_info)
1164 {
1165         struct ena_com_buf *ena_buf;
1166         u32 cnt;
1167         int i;
1168
1169         ena_buf = tx_info->bufs;
1170         cnt = tx_info->num_of_bufs;
1171
1172         if (unlikely(!cnt))
1173                 return;
1174
1175         if (tx_info->map_linear_data) {
1176                 dma_unmap_single(tx_ring->dev,
1177                                  dma_unmap_addr(ena_buf, paddr),
1178                                  dma_unmap_len(ena_buf, len),
1179                                  DMA_TO_DEVICE);
1180                 ena_buf++;
1181                 cnt--;
1182         }
1183
1184         /* unmap remaining mapped pages */
1185         for (i = 0; i < cnt; i++) {
1186                 dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
1187                                dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
1188                 ena_buf++;
1189         }
1190 }
1191
1192 /* ena_free_tx_bufs - Free Tx Buffers per Queue
1193  * @tx_ring: TX ring for which buffers be freed
1194  */
1195 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
1196 {
1197         bool print_once = true;
1198         u32 i;
1199
1200         for (i = 0; i < tx_ring->ring_size; i++) {
1201                 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1202
1203                 if (!tx_info->skb)
1204                         continue;
1205
1206                 if (print_once) {
1207                         netif_notice(tx_ring->adapter, ifdown, tx_ring->netdev,
1208                                      "Free uncompleted tx skb qid %d idx 0x%x\n",
1209                                      tx_ring->qid, i);
1210                         print_once = false;
1211                 } else {
1212                         netif_dbg(tx_ring->adapter, ifdown, tx_ring->netdev,
1213                                   "Free uncompleted tx skb qid %d idx 0x%x\n",
1214                                   tx_ring->qid, i);
1215                 }
1216
1217                 ena_unmap_tx_buff(tx_ring, tx_info);
1218
1219                 dev_kfree_skb_any(tx_info->skb);
1220         }
1221         netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
1222                                                   tx_ring->qid));
1223 }
1224
1225 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
1226 {
1227         struct ena_ring *tx_ring;
1228         int i;
1229
1230         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1231                 tx_ring = &adapter->tx_ring[i];
1232                 ena_free_tx_bufs(tx_ring);
1233         }
1234 }
1235
1236 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1237 {
1238         u16 ena_qid;
1239         int i;
1240
1241         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1242                 ena_qid = ENA_IO_TXQ_IDX(i);
1243                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1244         }
1245 }
1246
1247 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1248 {
1249         u16 ena_qid;
1250         int i;
1251
1252         for (i = 0; i < adapter->num_io_queues; i++) {
1253                 ena_qid = ENA_IO_RXQ_IDX(i);
1254                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
1255                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1256         }
1257 }
1258
1259 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
1260 {
1261         ena_destroy_all_tx_queues(adapter);
1262         ena_destroy_all_rx_queues(adapter);
1263 }
1264
1265 static int handle_invalid_req_id(struct ena_ring *ring, u16 req_id,
1266                                  struct ena_tx_buffer *tx_info, bool is_xdp)
1267 {
1268         if (tx_info)
1269                 netif_err(ring->adapter,
1270                           tx_done,
1271                           ring->netdev,
1272                           "tx_info doesn't have valid %s",
1273                            is_xdp ? "xdp frame" : "skb");
1274         else
1275                 netif_err(ring->adapter,
1276                           tx_done,
1277                           ring->netdev,
1278                           "Invalid req_id: %hu\n",
1279                           req_id);
1280
1281         ena_increase_stat(&ring->tx_stats.bad_req_id, 1, &ring->syncp);
1282
1283         /* Trigger device reset */
1284         ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
1285         set_bit(ENA_FLAG_TRIGGER_RESET, &ring->adapter->flags);
1286         return -EFAULT;
1287 }
1288
1289 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
1290 {
1291         struct ena_tx_buffer *tx_info;
1292
1293         tx_info = &tx_ring->tx_buffer_info[req_id];
1294         if (likely(tx_info->skb))
1295                 return 0;
1296
1297         return handle_invalid_req_id(tx_ring, req_id, tx_info, false);
1298 }
1299
1300 static int validate_xdp_req_id(struct ena_ring *xdp_ring, u16 req_id)
1301 {
1302         struct ena_tx_buffer *tx_info;
1303
1304         tx_info = &xdp_ring->tx_buffer_info[req_id];
1305         if (likely(tx_info->xdpf))
1306                 return 0;
1307
1308         return handle_invalid_req_id(xdp_ring, req_id, tx_info, true);
1309 }
1310
1311 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
1312 {
1313         struct netdev_queue *txq;
1314         bool above_thresh;
1315         u32 tx_bytes = 0;
1316         u32 total_done = 0;
1317         u16 next_to_clean;
1318         u16 req_id;
1319         int tx_pkts = 0;
1320         int rc;
1321
1322         next_to_clean = tx_ring->next_to_clean;
1323         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
1324
1325         while (tx_pkts < budget) {
1326                 struct ena_tx_buffer *tx_info;
1327                 struct sk_buff *skb;
1328
1329                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
1330                                                 &req_id);
1331                 if (rc) {
1332                         if (unlikely(rc == -EINVAL))
1333                                 handle_invalid_req_id(tx_ring, req_id, NULL,
1334                                                       false);
1335                         break;
1336                 }
1337
1338                 /* validate that the request id points to a valid skb */
1339                 rc = validate_tx_req_id(tx_ring, req_id);
1340                 if (rc)
1341                         break;
1342
1343                 tx_info = &tx_ring->tx_buffer_info[req_id];
1344                 skb = tx_info->skb;
1345
1346                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
1347                 prefetch(&skb->end);
1348
1349                 tx_info->skb = NULL;
1350                 tx_info->last_jiffies = 0;
1351
1352                 ena_unmap_tx_buff(tx_ring, tx_info);
1353
1354                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1355                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
1356                           skb);
1357
1358                 tx_bytes += skb->len;
1359                 dev_kfree_skb(skb);
1360                 tx_pkts++;
1361                 total_done += tx_info->tx_descs;
1362
1363                 tx_ring->free_ids[next_to_clean] = req_id;
1364                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1365                                                      tx_ring->ring_size);
1366         }
1367
1368         tx_ring->next_to_clean = next_to_clean;
1369         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
1370         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
1371
1372         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
1373
1374         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1375                   "tx_poll: q %d done. total pkts: %d\n",
1376                   tx_ring->qid, tx_pkts);
1377
1378         /* need to make the rings circular update visible to
1379          * ena_start_xmit() before checking for netif_queue_stopped().
1380          */
1381         smp_mb();
1382
1383         above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1384                                                     ENA_TX_WAKEUP_THRESH);
1385         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
1386                 __netif_tx_lock(txq, smp_processor_id());
1387                 above_thresh =
1388                         ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1389                                                      ENA_TX_WAKEUP_THRESH);
1390                 if (netif_tx_queue_stopped(txq) && above_thresh &&
1391                     test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
1392                         netif_tx_wake_queue(txq);
1393                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
1394                                           &tx_ring->syncp);
1395                 }
1396                 __netif_tx_unlock(txq);
1397         }
1398
1399         return tx_pkts;
1400 }
1401
1402 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, void *first_frag)
1403 {
1404         struct sk_buff *skb;
1405
1406         if (!first_frag)
1407                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1408                                                 rx_ring->rx_copybreak);
1409         else
1410                 skb = build_skb(first_frag, ENA_PAGE_SIZE);
1411
1412         if (unlikely(!skb)) {
1413                 ena_increase_stat(&rx_ring->rx_stats.skb_alloc_fail, 1,
1414                                   &rx_ring->syncp);
1415
1416                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1417                           "Failed to allocate skb. first_frag %s\n",
1418                           first_frag ? "provided" : "not provided");
1419                 return NULL;
1420         }
1421
1422         return skb;
1423 }
1424
1425 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
1426                                   struct ena_com_rx_buf_info *ena_bufs,
1427                                   u32 descs,
1428                                   u16 *next_to_clean)
1429 {
1430         struct ena_rx_buffer *rx_info;
1431         struct ena_adapter *adapter;
1432         u16 len, req_id, buf = 0;
1433         struct sk_buff *skb;
1434         void *page_addr;
1435         u32 page_offset;
1436         void *data_addr;
1437
1438         len = ena_bufs[buf].len;
1439         req_id = ena_bufs[buf].req_id;
1440
1441         rx_info = &rx_ring->rx_buffer_info[req_id];
1442
1443         if (unlikely(!rx_info->page)) {
1444                 adapter = rx_ring->adapter;
1445                 netif_err(adapter, rx_err, rx_ring->netdev,
1446                           "Page is NULL. qid %u req_id %u\n", rx_ring->qid, req_id);
1447                 ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1, &rx_ring->syncp);
1448                 adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
1449                 /* Make sure reset reason is set before triggering the reset */
1450                 smp_mb__before_atomic();
1451                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1452                 return NULL;
1453         }
1454
1455         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1456                   "rx_info %p page %p\n",
1457                   rx_info, rx_info->page);
1458
1459         /* save virt address of first buffer */
1460         page_addr = page_address(rx_info->page);
1461         page_offset = rx_info->page_offset;
1462         data_addr = page_addr + page_offset;
1463
1464         prefetch(data_addr);
1465
1466         if (len <= rx_ring->rx_copybreak) {
1467                 skb = ena_alloc_skb(rx_ring, NULL);
1468                 if (unlikely(!skb))
1469                         return NULL;
1470
1471                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1472                           "RX allocated small packet. len %d. data_len %d\n",
1473                           skb->len, skb->data_len);
1474
1475                 /* sync this buffer for CPU use */
1476                 dma_sync_single_for_cpu(rx_ring->dev,
1477                                         dma_unmap_addr(&rx_info->ena_buf, paddr),
1478                                         len,
1479                                         DMA_FROM_DEVICE);
1480                 skb_copy_to_linear_data(skb, data_addr, len);
1481                 dma_sync_single_for_device(rx_ring->dev,
1482                                            dma_unmap_addr(&rx_info->ena_buf, paddr),
1483                                            len,
1484                                            DMA_FROM_DEVICE);
1485
1486                 skb_put(skb, len);
1487                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1488                 rx_ring->free_ids[*next_to_clean] = req_id;
1489                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
1490                                                      rx_ring->ring_size);
1491                 return skb;
1492         }
1493
1494         ena_unmap_rx_buff(rx_ring, rx_info);
1495
1496         skb = ena_alloc_skb(rx_ring, page_addr);
1497         if (unlikely(!skb))
1498                 return NULL;
1499
1500         /* Populate skb's linear part */
1501         skb_reserve(skb, page_offset);
1502         skb_put(skb, len);
1503         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1504
1505         do {
1506                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1507                           "RX skb updated. len %d. data_len %d\n",
1508                           skb->len, skb->data_len);
1509
1510                 rx_info->page = NULL;
1511
1512                 rx_ring->free_ids[*next_to_clean] = req_id;
1513                 *next_to_clean =
1514                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
1515                                              rx_ring->ring_size);
1516                 if (likely(--descs == 0))
1517                         break;
1518
1519                 buf++;
1520                 len = ena_bufs[buf].len;
1521                 req_id = ena_bufs[buf].req_id;
1522
1523                 rx_info = &rx_ring->rx_buffer_info[req_id];
1524
1525                 ena_unmap_rx_buff(rx_ring, rx_info);
1526
1527                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
1528                                 rx_info->page_offset, len, ENA_PAGE_SIZE);
1529
1530         } while (1);
1531
1532         return skb;
1533 }
1534
1535 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
1536  * @adapter: structure containing adapter specific data
1537  * @ena_rx_ctx: received packet context/metadata
1538  * @skb: skb currently being received and modified
1539  */
1540 static void ena_rx_checksum(struct ena_ring *rx_ring,
1541                                    struct ena_com_rx_ctx *ena_rx_ctx,
1542                                    struct sk_buff *skb)
1543 {
1544         /* Rx csum disabled */
1545         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
1546                 skb->ip_summed = CHECKSUM_NONE;
1547                 return;
1548         }
1549
1550         /* For fragmented packets the checksum isn't valid */
1551         if (ena_rx_ctx->frag) {
1552                 skb->ip_summed = CHECKSUM_NONE;
1553                 return;
1554         }
1555
1556         /* if IP and error */
1557         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1558                      (ena_rx_ctx->l3_csum_err))) {
1559                 /* ipv4 checksum error */
1560                 skb->ip_summed = CHECKSUM_NONE;
1561                 ena_increase_stat(&rx_ring->rx_stats.bad_csum, 1,
1562                                   &rx_ring->syncp);
1563                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1564                           "RX IPv4 header checksum error\n");
1565                 return;
1566         }
1567
1568         /* if TCP/UDP */
1569         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1570                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
1571                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
1572                         /* TCP/UDP checksum error */
1573                         ena_increase_stat(&rx_ring->rx_stats.bad_csum, 1,
1574                                           &rx_ring->syncp);
1575                         netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1576                                   "RX L4 checksum error\n");
1577                         skb->ip_summed = CHECKSUM_NONE;
1578                         return;
1579                 }
1580
1581                 if (likely(ena_rx_ctx->l4_csum_checked)) {
1582                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1583                         ena_increase_stat(&rx_ring->rx_stats.csum_good, 1,
1584                                           &rx_ring->syncp);
1585                 } else {
1586                         ena_increase_stat(&rx_ring->rx_stats.csum_unchecked, 1,
1587                                           &rx_ring->syncp);
1588                         skb->ip_summed = CHECKSUM_NONE;
1589                 }
1590         } else {
1591                 skb->ip_summed = CHECKSUM_NONE;
1592                 return;
1593         }
1594
1595 }
1596
1597 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1598                             struct ena_com_rx_ctx *ena_rx_ctx,
1599                             struct sk_buff *skb)
1600 {
1601         enum pkt_hash_types hash_type;
1602
1603         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1604                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1605                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1606
1607                         hash_type = PKT_HASH_TYPE_L4;
1608                 else
1609                         hash_type = PKT_HASH_TYPE_NONE;
1610
1611                 /* Override hash type if the packet is fragmented */
1612                 if (ena_rx_ctx->frag)
1613                         hash_type = PKT_HASH_TYPE_NONE;
1614
1615                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1616         }
1617 }
1618
1619 static int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)
1620 {
1621         struct ena_rx_buffer *rx_info;
1622         int ret;
1623
1624         rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1625         xdp_prepare_buff(xdp, page_address(rx_info->page),
1626                          rx_info->page_offset,
1627                          rx_ring->ena_bufs[0].len, false);
1628         /* If for some reason we received a bigger packet than
1629          * we expect, then we simply drop it
1630          */
1631         if (unlikely(rx_ring->ena_bufs[0].len > ENA_XDP_MAX_MTU))
1632                 return XDP_DROP;
1633
1634         ret = ena_xdp_execute(rx_ring, xdp);
1635
1636         /* The xdp program might expand the headers */
1637         if (ret == XDP_PASS) {
1638                 rx_info->page_offset = xdp->data - xdp->data_hard_start;
1639                 rx_ring->ena_bufs[0].len = xdp->data_end - xdp->data;
1640         }
1641
1642         return ret;
1643 }
1644 /* ena_clean_rx_irq - Cleanup RX irq
1645  * @rx_ring: RX ring to clean
1646  * @napi: napi handler
1647  * @budget: how many packets driver is allowed to clean
1648  *
1649  * Returns the number of cleaned buffers.
1650  */
1651 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1652                             u32 budget)
1653 {
1654         u16 next_to_clean = rx_ring->next_to_clean;
1655         struct ena_com_rx_ctx ena_rx_ctx;
1656         struct ena_rx_buffer *rx_info;
1657         struct ena_adapter *adapter;
1658         u32 res_budget, work_done;
1659         int rx_copybreak_pkt = 0;
1660         int refill_threshold;
1661         struct sk_buff *skb;
1662         int refill_required;
1663         struct xdp_buff xdp;
1664         int xdp_flags = 0;
1665         int total_len = 0;
1666         int xdp_verdict;
1667         int rc = 0;
1668         int i;
1669
1670         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1671                   "%s qid %d\n", __func__, rx_ring->qid);
1672         res_budget = budget;
1673         xdp_init_buff(&xdp, ENA_PAGE_SIZE, &rx_ring->xdp_rxq);
1674
1675         do {
1676                 xdp_verdict = XDP_PASS;
1677                 skb = NULL;
1678                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1679                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1680                 ena_rx_ctx.descs = 0;
1681                 ena_rx_ctx.pkt_offset = 0;
1682                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1683                                     rx_ring->ena_com_io_sq,
1684                                     &ena_rx_ctx);
1685                 if (unlikely(rc))
1686                         goto error;
1687
1688                 if (unlikely(ena_rx_ctx.descs == 0))
1689                         break;
1690
1691                 /* First descriptor might have an offset set by the device */
1692                 rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1693                 rx_info->page_offset += ena_rx_ctx.pkt_offset;
1694
1695                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1696                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1697                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1698                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1699
1700                 if (ena_xdp_present_ring(rx_ring))
1701                         xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1702
1703                 /* allocate skb and fill it */
1704                 if (xdp_verdict == XDP_PASS)
1705                         skb = ena_rx_skb(rx_ring,
1706                                          rx_ring->ena_bufs,
1707                                          ena_rx_ctx.descs,
1708                                          &next_to_clean);
1709
1710                 if (unlikely(!skb)) {
1711                         for (i = 0; i < ena_rx_ctx.descs; i++) {
1712                                 int req_id = rx_ring->ena_bufs[i].req_id;
1713
1714                                 rx_ring->free_ids[next_to_clean] = req_id;
1715                                 next_to_clean =
1716                                         ENA_RX_RING_IDX_NEXT(next_to_clean,
1717                                                              rx_ring->ring_size);
1718
1719                                 /* Packets was passed for transmission, unmap it
1720                                  * from RX side.
1721                                  */
1722                                 if (xdp_verdict == XDP_TX || xdp_verdict == XDP_REDIRECT) {
1723                                         ena_unmap_rx_buff(rx_ring,
1724                                                           &rx_ring->rx_buffer_info[req_id]);
1725                                         rx_ring->rx_buffer_info[req_id].page = NULL;
1726                                 }
1727                         }
1728                         if (xdp_verdict != XDP_PASS) {
1729                                 xdp_flags |= xdp_verdict;
1730                                 res_budget--;
1731                                 continue;
1732                         }
1733                         break;
1734                 }
1735
1736                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1737
1738                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1739
1740                 skb_record_rx_queue(skb, rx_ring->qid);
1741
1742                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak)
1743                         rx_copybreak_pkt++;
1744
1745                 total_len += skb->len;
1746
1747                 napi_gro_receive(napi, skb);
1748
1749                 res_budget--;
1750         } while (likely(res_budget));
1751
1752         work_done = budget - res_budget;
1753         rx_ring->per_napi_packets += work_done;
1754         u64_stats_update_begin(&rx_ring->syncp);
1755         rx_ring->rx_stats.bytes += total_len;
1756         rx_ring->rx_stats.cnt += work_done;
1757         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1758         u64_stats_update_end(&rx_ring->syncp);
1759
1760         rx_ring->next_to_clean = next_to_clean;
1761
1762         refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
1763         refill_threshold =
1764                 min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1765                       ENA_RX_REFILL_THRESH_PACKET);
1766
1767         /* Optimization, try to batch new rx buffers */
1768         if (refill_required > refill_threshold) {
1769                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1770                 ena_refill_rx_bufs(rx_ring, refill_required);
1771         }
1772
1773         if (xdp_flags & XDP_REDIRECT)
1774                 xdp_do_flush_map();
1775
1776         return work_done;
1777
1778 error:
1779         adapter = netdev_priv(rx_ring->netdev);
1780
1781         if (rc == -ENOSPC) {
1782                 ena_increase_stat(&rx_ring->rx_stats.bad_desc_num, 1,
1783                                   &rx_ring->syncp);
1784                 adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1785         } else {
1786                 ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1,
1787                                   &rx_ring->syncp);
1788                 adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
1789         }
1790
1791         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1792
1793         return 0;
1794 }
1795
1796 static void ena_dim_work(struct work_struct *w)
1797 {
1798         struct dim *dim = container_of(w, struct dim, work);
1799         struct dim_cq_moder cur_moder =
1800                 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1801         struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1802
1803         ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1804         dim->state = DIM_START_MEASURE;
1805 }
1806
1807 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1808 {
1809         struct dim_sample dim_sample;
1810         struct ena_ring *rx_ring = ena_napi->rx_ring;
1811
1812         if (!rx_ring->per_napi_packets)
1813                 return;
1814
1815         rx_ring->non_empty_napi_events++;
1816
1817         dim_update_sample(rx_ring->non_empty_napi_events,
1818                           rx_ring->rx_stats.cnt,
1819                           rx_ring->rx_stats.bytes,
1820                           &dim_sample);
1821
1822         net_dim(&ena_napi->dim, dim_sample);
1823
1824         rx_ring->per_napi_packets = 0;
1825 }
1826
1827 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1828                                         struct ena_ring *rx_ring)
1829 {
1830         struct ena_eth_io_intr_reg intr_reg;
1831         u32 rx_interval = 0;
1832         /* Rx ring can be NULL when for XDP tx queues which don't have an
1833          * accompanying rx_ring pair.
1834          */
1835         if (rx_ring)
1836                 rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1837                         rx_ring->smoothed_interval :
1838                         ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1839
1840         /* Update intr register: rx intr delay,
1841          * tx intr delay and interrupt unmask
1842          */
1843         ena_com_update_intr_reg(&intr_reg,
1844                                 rx_interval,
1845                                 tx_ring->smoothed_interval,
1846                                 true);
1847
1848         ena_increase_stat(&tx_ring->tx_stats.unmask_interrupt, 1,
1849                           &tx_ring->syncp);
1850
1851         /* It is a shared MSI-X.
1852          * Tx and Rx CQ have pointer to it.
1853          * So we use one of them to reach the intr reg
1854          * The Tx ring is used because the rx_ring is NULL for XDP queues
1855          */
1856         ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1857 }
1858
1859 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1860                                              struct ena_ring *rx_ring)
1861 {
1862         int cpu = get_cpu();
1863         int numa_node;
1864
1865         /* Check only one ring since the 2 rings are running on the same cpu */
1866         if (likely(tx_ring->cpu == cpu))
1867                 goto out;
1868
1869         numa_node = cpu_to_node(cpu);
1870         put_cpu();
1871
1872         if (numa_node != NUMA_NO_NODE) {
1873                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1874                 if (rx_ring)
1875                         ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1876                                                  numa_node);
1877         }
1878
1879         tx_ring->cpu = cpu;
1880         if (rx_ring)
1881                 rx_ring->cpu = cpu;
1882
1883         return;
1884 out:
1885         put_cpu();
1886 }
1887
1888 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1889 {
1890         u32 total_done = 0;
1891         u16 next_to_clean;
1892         u32 tx_bytes = 0;
1893         int tx_pkts = 0;
1894         u16 req_id;
1895         int rc;
1896
1897         if (unlikely(!xdp_ring))
1898                 return 0;
1899         next_to_clean = xdp_ring->next_to_clean;
1900
1901         while (tx_pkts < budget) {
1902                 struct ena_tx_buffer *tx_info;
1903                 struct xdp_frame *xdpf;
1904
1905                 rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1906                                                 &req_id);
1907                 if (rc) {
1908                         if (unlikely(rc == -EINVAL))
1909                                 handle_invalid_req_id(xdp_ring, req_id, NULL,
1910                                                       true);
1911                         break;
1912                 }
1913
1914                 /* validate that the request id points to a valid xdp_frame */
1915                 rc = validate_xdp_req_id(xdp_ring, req_id);
1916                 if (rc)
1917                         break;
1918
1919                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1920                 xdpf = tx_info->xdpf;
1921
1922                 tx_info->xdpf = NULL;
1923                 tx_info->last_jiffies = 0;
1924                 ena_unmap_tx_buff(xdp_ring, tx_info);
1925
1926                 netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1927                           "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1928                           xdpf);
1929
1930                 tx_bytes += xdpf->len;
1931                 tx_pkts++;
1932                 total_done += tx_info->tx_descs;
1933
1934                 xdp_return_frame(xdpf);
1935                 xdp_ring->free_ids[next_to_clean] = req_id;
1936                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1937                                                      xdp_ring->ring_size);
1938         }
1939
1940         xdp_ring->next_to_clean = next_to_clean;
1941         ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
1942         ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
1943
1944         netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1945                   "tx_poll: q %d done. total pkts: %d\n",
1946                   xdp_ring->qid, tx_pkts);
1947
1948         return tx_pkts;
1949 }
1950
1951 static int ena_io_poll(struct napi_struct *napi, int budget)
1952 {
1953         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1954         struct ena_ring *tx_ring, *rx_ring;
1955         int tx_work_done;
1956         int rx_work_done = 0;
1957         int tx_budget;
1958         int napi_comp_call = 0;
1959         int ret;
1960
1961         tx_ring = ena_napi->tx_ring;
1962         rx_ring = ena_napi->rx_ring;
1963
1964         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1965
1966         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1967             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1968                 napi_complete_done(napi, 0);
1969                 return 0;
1970         }
1971
1972         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1973         /* On netpoll the budget is zero and the handler should only clean the
1974          * tx completions.
1975          */
1976         if (likely(budget))
1977                 rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1978
1979         /* If the device is about to reset or down, avoid unmask
1980          * the interrupt and return 0 so NAPI won't reschedule
1981          */
1982         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1983                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1984                 napi_complete_done(napi, 0);
1985                 ret = 0;
1986
1987         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1988                 napi_comp_call = 1;
1989
1990                 /* Update numa and unmask the interrupt only when schedule
1991                  * from the interrupt context (vs from sk_busy_loop)
1992                  */
1993                 if (napi_complete_done(napi, rx_work_done) &&
1994                     READ_ONCE(ena_napi->interrupts_masked)) {
1995                         smp_rmb(); /* make sure interrupts_masked is read */
1996                         WRITE_ONCE(ena_napi->interrupts_masked, false);
1997                         /* We apply adaptive moderation on Rx path only.
1998                          * Tx uses static interrupt moderation.
1999                          */
2000                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
2001                                 ena_adjust_adaptive_rx_intr_moderation(ena_napi);
2002
2003                         ena_unmask_interrupt(tx_ring, rx_ring);
2004                 }
2005
2006                 ena_update_ring_numa_node(tx_ring, rx_ring);
2007
2008                 ret = rx_work_done;
2009         } else {
2010                 ret = budget;
2011         }
2012
2013         u64_stats_update_begin(&tx_ring->syncp);
2014         tx_ring->tx_stats.napi_comp += napi_comp_call;
2015         tx_ring->tx_stats.tx_poll++;
2016         u64_stats_update_end(&tx_ring->syncp);
2017
2018         tx_ring->tx_stats.last_napi_jiffies = jiffies;
2019
2020         return ret;
2021 }
2022
2023 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
2024 {
2025         struct ena_adapter *adapter = (struct ena_adapter *)data;
2026
2027         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
2028
2029         /* Don't call the aenq handler before probe is done */
2030         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
2031                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
2032
2033         return IRQ_HANDLED;
2034 }
2035
2036 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
2037  * @irq: interrupt number
2038  * @data: pointer to a network interface private napi device structure
2039  */
2040 static irqreturn_t ena_intr_msix_io(int irq, void *data)
2041 {
2042         struct ena_napi *ena_napi = data;
2043
2044         /* Used to check HW health */
2045         WRITE_ONCE(ena_napi->first_interrupt, true);
2046
2047         WRITE_ONCE(ena_napi->interrupts_masked, true);
2048         smp_wmb(); /* write interrupts_masked before calling napi */
2049
2050         napi_schedule_irqoff(&ena_napi->napi);
2051
2052         return IRQ_HANDLED;
2053 }
2054
2055 /* Reserve a single MSI-X vector for management (admin + aenq).
2056  * plus reserve one vector for each potential io queue.
2057  * the number of potential io queues is the minimum of what the device
2058  * supports and the number of vCPUs.
2059  */
2060 static int ena_enable_msix(struct ena_adapter *adapter)
2061 {
2062         int msix_vecs, irq_cnt;
2063
2064         if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2065                 netif_err(adapter, probe, adapter->netdev,
2066                           "Error, MSI-X is already enabled\n");
2067                 return -EPERM;
2068         }
2069
2070         /* Reserved the max msix vectors we might need */
2071         msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
2072         netif_dbg(adapter, probe, adapter->netdev,
2073                   "Trying to enable MSI-X, vectors %d\n", msix_vecs);
2074
2075         irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
2076                                         msix_vecs, PCI_IRQ_MSIX);
2077
2078         if (irq_cnt < 0) {
2079                 netif_err(adapter, probe, adapter->netdev,
2080                           "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
2081                 return -ENOSPC;
2082         }
2083
2084         if (irq_cnt != msix_vecs) {
2085                 netif_notice(adapter, probe, adapter->netdev,
2086                              "Enable only %d MSI-X (out of %d), reduce the number of queues\n",
2087                              irq_cnt, msix_vecs);
2088                 adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
2089         }
2090
2091         if (ena_init_rx_cpu_rmap(adapter))
2092                 netif_warn(adapter, probe, adapter->netdev,
2093                            "Failed to map IRQs to CPUs\n");
2094
2095         adapter->msix_vecs = irq_cnt;
2096         set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2097
2098         return 0;
2099 }
2100
2101 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2102 {
2103         u32 cpu;
2104
2105         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2106                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2107                  pci_name(adapter->pdev));
2108         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2109                 ena_intr_msix_mgmnt;
2110         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2111         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2112                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2113         cpu = cpumask_first(cpu_online_mask);
2114         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2115         cpumask_set_cpu(cpu,
2116                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2117 }
2118
2119 static void ena_setup_io_intr(struct ena_adapter *adapter)
2120 {
2121         struct net_device *netdev;
2122         int irq_idx, i, cpu;
2123         int io_queue_count;
2124
2125         netdev = adapter->netdev;
2126         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2127
2128         for (i = 0; i < io_queue_count; i++) {
2129                 irq_idx = ENA_IO_IRQ_IDX(i);
2130                 cpu = i % num_online_cpus();
2131
2132                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2133                          "%s-Tx-Rx-%d", netdev->name, i);
2134                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2135                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2136                 adapter->irq_tbl[irq_idx].vector =
2137                         pci_irq_vector(adapter->pdev, irq_idx);
2138                 adapter->irq_tbl[irq_idx].cpu = cpu;
2139
2140                 cpumask_set_cpu(cpu,
2141                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
2142         }
2143 }
2144
2145 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2146 {
2147         unsigned long flags = 0;
2148         struct ena_irq *irq;
2149         int rc;
2150
2151         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2152         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2153                          irq->data);
2154         if (rc) {
2155                 netif_err(adapter, probe, adapter->netdev,
2156                           "Failed to request admin irq\n");
2157                 return rc;
2158         }
2159
2160         netif_dbg(adapter, probe, adapter->netdev,
2161                   "Set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2162                   irq->affinity_hint_mask.bits[0], irq->vector);
2163
2164         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2165
2166         return rc;
2167 }
2168
2169 static int ena_request_io_irq(struct ena_adapter *adapter)
2170 {
2171         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2172         unsigned long flags = 0;
2173         struct ena_irq *irq;
2174         int rc = 0, i, k;
2175
2176         if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2177                 netif_err(adapter, ifup, adapter->netdev,
2178                           "Failed to request I/O IRQ: MSI-X is not enabled\n");
2179                 return -EINVAL;
2180         }
2181
2182         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2183                 irq = &adapter->irq_tbl[i];
2184                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2185                                  irq->data);
2186                 if (rc) {
2187                         netif_err(adapter, ifup, adapter->netdev,
2188                                   "Failed to request I/O IRQ. index %d rc %d\n",
2189                                    i, rc);
2190                         goto err;
2191                 }
2192
2193                 netif_dbg(adapter, ifup, adapter->netdev,
2194                           "Set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2195                           i, irq->affinity_hint_mask.bits[0], irq->vector);
2196
2197                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2198         }
2199
2200         return rc;
2201
2202 err:
2203         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2204                 irq = &adapter->irq_tbl[k];
2205                 free_irq(irq->vector, irq->data);
2206         }
2207
2208         return rc;
2209 }
2210
2211 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2212 {
2213         struct ena_irq *irq;
2214
2215         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2216         synchronize_irq(irq->vector);
2217         irq_set_affinity_hint(irq->vector, NULL);
2218         free_irq(irq->vector, irq->data);
2219 }
2220
2221 static void ena_free_io_irq(struct ena_adapter *adapter)
2222 {
2223         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2224         struct ena_irq *irq;
2225         int i;
2226
2227 #ifdef CONFIG_RFS_ACCEL
2228         if (adapter->msix_vecs >= 1) {
2229                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2230                 adapter->netdev->rx_cpu_rmap = NULL;
2231         }
2232 #endif /* CONFIG_RFS_ACCEL */
2233
2234         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2235                 irq = &adapter->irq_tbl[i];
2236                 irq_set_affinity_hint(irq->vector, NULL);
2237                 free_irq(irq->vector, irq->data);
2238         }
2239 }
2240
2241 static void ena_disable_msix(struct ena_adapter *adapter)
2242 {
2243         if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2244                 pci_free_irq_vectors(adapter->pdev);
2245 }
2246
2247 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2248 {
2249         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2250         int i;
2251
2252         if (!netif_running(adapter->netdev))
2253                 return;
2254
2255         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2256                 synchronize_irq(adapter->irq_tbl[i].vector);
2257 }
2258
2259 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2260                                   int first_index,
2261                                   int count)
2262 {
2263         int i;
2264
2265         for (i = first_index; i < first_index + count; i++) {
2266                 netif_napi_del(&adapter->ena_napi[i].napi);
2267
2268                 WARN_ON(!ENA_IS_XDP_INDEX(adapter, i) &&
2269                         adapter->ena_napi[i].xdp_ring);
2270         }
2271 }
2272
2273 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2274                                    int first_index, int count)
2275 {
2276         int i;
2277
2278         for (i = first_index; i < first_index + count; i++) {
2279                 struct ena_napi *napi = &adapter->ena_napi[i];
2280
2281                 netif_napi_add(adapter->netdev,
2282                                &napi->napi,
2283                                ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll,
2284                                ENA_NAPI_BUDGET);
2285
2286                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
2287                         napi->rx_ring = &adapter->rx_ring[i];
2288                         napi->tx_ring = &adapter->tx_ring[i];
2289                 } else {
2290                         napi->xdp_ring = &adapter->tx_ring[i];
2291                 }
2292                 napi->qid = i;
2293         }
2294 }
2295
2296 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2297                                       int first_index,
2298                                       int count)
2299 {
2300         int i;
2301
2302         for (i = first_index; i < first_index + count; i++)
2303                 napi_disable(&adapter->ena_napi[i].napi);
2304 }
2305
2306 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2307                                      int first_index,
2308                                      int count)
2309 {
2310         int i;
2311
2312         for (i = first_index; i < first_index + count; i++)
2313                 napi_enable(&adapter->ena_napi[i].napi);
2314 }
2315
2316 /* Configure the Rx forwarding */
2317 static int ena_rss_configure(struct ena_adapter *adapter)
2318 {
2319         struct ena_com_dev *ena_dev = adapter->ena_dev;
2320         int rc;
2321
2322         /* In case the RSS table wasn't initialized by probe */
2323         if (!ena_dev->rss.tbl_log_size) {
2324                 rc = ena_rss_init_default(adapter);
2325                 if (rc && (rc != -EOPNOTSUPP)) {
2326                         netif_err(adapter, ifup, adapter->netdev,
2327                                   "Failed to init RSS rc: %d\n", rc);
2328                         return rc;
2329                 }
2330         }
2331
2332         /* Set indirect table */
2333         rc = ena_com_indirect_table_set(ena_dev);
2334         if (unlikely(rc && rc != -EOPNOTSUPP))
2335                 return rc;
2336
2337         /* Configure hash function (if supported) */
2338         rc = ena_com_set_hash_function(ena_dev);
2339         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2340                 return rc;
2341
2342         /* Configure hash inputs (if supported) */
2343         rc = ena_com_set_hash_ctrl(ena_dev);
2344         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2345                 return rc;
2346
2347         return 0;
2348 }
2349
2350 static int ena_up_complete(struct ena_adapter *adapter)
2351 {
2352         int rc;
2353
2354         rc = ena_rss_configure(adapter);
2355         if (rc)
2356                 return rc;
2357
2358         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2359
2360         ena_refill_all_rx_bufs(adapter);
2361
2362         /* enable transmits */
2363         netif_tx_start_all_queues(adapter->netdev);
2364
2365         ena_napi_enable_in_range(adapter,
2366                                  0,
2367                                  adapter->xdp_num_queues + adapter->num_io_queues);
2368
2369         return 0;
2370 }
2371
2372 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2373 {
2374         struct ena_com_create_io_ctx ctx;
2375         struct ena_com_dev *ena_dev;
2376         struct ena_ring *tx_ring;
2377         u32 msix_vector;
2378         u16 ena_qid;
2379         int rc;
2380
2381         ena_dev = adapter->ena_dev;
2382
2383         tx_ring = &adapter->tx_ring[qid];
2384         msix_vector = ENA_IO_IRQ_IDX(qid);
2385         ena_qid = ENA_IO_TXQ_IDX(qid);
2386
2387         memset(&ctx, 0x0, sizeof(ctx));
2388
2389         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2390         ctx.qid = ena_qid;
2391         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2392         ctx.msix_vector = msix_vector;
2393         ctx.queue_size = tx_ring->ring_size;
2394         ctx.numa_node = cpu_to_node(tx_ring->cpu);
2395
2396         rc = ena_com_create_io_queue(ena_dev, &ctx);
2397         if (rc) {
2398                 netif_err(adapter, ifup, adapter->netdev,
2399                           "Failed to create I/O TX queue num %d rc: %d\n",
2400                           qid, rc);
2401                 return rc;
2402         }
2403
2404         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2405                                      &tx_ring->ena_com_io_sq,
2406                                      &tx_ring->ena_com_io_cq);
2407         if (rc) {
2408                 netif_err(adapter, ifup, adapter->netdev,
2409                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2410                           qid, rc);
2411                 ena_com_destroy_io_queue(ena_dev, ena_qid);
2412                 return rc;
2413         }
2414
2415         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2416         return rc;
2417 }
2418
2419 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2420                                             int first_index, int count)
2421 {
2422         struct ena_com_dev *ena_dev = adapter->ena_dev;
2423         int rc, i;
2424
2425         for (i = first_index; i < first_index + count; i++) {
2426                 rc = ena_create_io_tx_queue(adapter, i);
2427                 if (rc)
2428                         goto create_err;
2429         }
2430
2431         return 0;
2432
2433 create_err:
2434         while (i-- > first_index)
2435                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2436
2437         return rc;
2438 }
2439
2440 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2441 {
2442         struct ena_com_dev *ena_dev;
2443         struct ena_com_create_io_ctx ctx;
2444         struct ena_ring *rx_ring;
2445         u32 msix_vector;
2446         u16 ena_qid;
2447         int rc;
2448
2449         ena_dev = adapter->ena_dev;
2450
2451         rx_ring = &adapter->rx_ring[qid];
2452         msix_vector = ENA_IO_IRQ_IDX(qid);
2453         ena_qid = ENA_IO_RXQ_IDX(qid);
2454
2455         memset(&ctx, 0x0, sizeof(ctx));
2456
2457         ctx.qid = ena_qid;
2458         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2459         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2460         ctx.msix_vector = msix_vector;
2461         ctx.queue_size = rx_ring->ring_size;
2462         ctx.numa_node = cpu_to_node(rx_ring->cpu);
2463
2464         rc = ena_com_create_io_queue(ena_dev, &ctx);
2465         if (rc) {
2466                 netif_err(adapter, ifup, adapter->netdev,
2467                           "Failed to create I/O RX queue num %d rc: %d\n",
2468                           qid, rc);
2469                 return rc;
2470         }
2471
2472         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2473                                      &rx_ring->ena_com_io_sq,
2474                                      &rx_ring->ena_com_io_cq);
2475         if (rc) {
2476                 netif_err(adapter, ifup, adapter->netdev,
2477                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2478                           qid, rc);
2479                 goto err;
2480         }
2481
2482         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2483
2484         return rc;
2485 err:
2486         ena_com_destroy_io_queue(ena_dev, ena_qid);
2487         return rc;
2488 }
2489
2490 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2491 {
2492         struct ena_com_dev *ena_dev = adapter->ena_dev;
2493         int rc, i;
2494
2495         for (i = 0; i < adapter->num_io_queues; i++) {
2496                 rc = ena_create_io_rx_queue(adapter, i);
2497                 if (rc)
2498                         goto create_err;
2499                 INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2500         }
2501
2502         return 0;
2503
2504 create_err:
2505         while (i--) {
2506                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
2507                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2508         }
2509
2510         return rc;
2511 }
2512
2513 static void set_io_rings_size(struct ena_adapter *adapter,
2514                               int new_tx_size,
2515                               int new_rx_size)
2516 {
2517         int i;
2518
2519         for (i = 0; i < adapter->num_io_queues; i++) {
2520                 adapter->tx_ring[i].ring_size = new_tx_size;
2521                 adapter->rx_ring[i].ring_size = new_rx_size;
2522         }
2523 }
2524
2525 /* This function allows queue allocation to backoff when the system is
2526  * low on memory. If there is not enough memory to allocate io queues
2527  * the driver will try to allocate smaller queues.
2528  *
2529  * The backoff algorithm is as follows:
2530  *  1. Try to allocate TX and RX and if successful.
2531  *  1.1. return success
2532  *
2533  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2534  *
2535  *  3. If TX or RX is smaller than 256
2536  *  3.1. return failure.
2537  *  4. else
2538  *  4.1. go back to 1.
2539  */
2540 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2541 {
2542         int rc, cur_rx_ring_size, cur_tx_ring_size;
2543         int new_rx_ring_size, new_tx_ring_size;
2544
2545         /* current queue sizes might be set to smaller than the requested
2546          * ones due to past queue allocation failures.
2547          */
2548         set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2549                           adapter->requested_rx_ring_size);
2550
2551         while (1) {
2552                 if (ena_xdp_present(adapter)) {
2553                         rc = ena_setup_and_create_all_xdp_queues(adapter);
2554
2555                         if (rc)
2556                                 goto err_setup_tx;
2557                 }
2558                 rc = ena_setup_tx_resources_in_range(adapter,
2559                                                      0,
2560                                                      adapter->num_io_queues);
2561                 if (rc)
2562                         goto err_setup_tx;
2563
2564                 rc = ena_create_io_tx_queues_in_range(adapter,
2565                                                       0,
2566                                                       adapter->num_io_queues);
2567                 if (rc)
2568                         goto err_create_tx_queues;
2569
2570                 rc = ena_setup_all_rx_resources(adapter);
2571                 if (rc)
2572                         goto err_setup_rx;
2573
2574                 rc = ena_create_all_io_rx_queues(adapter);
2575                 if (rc)
2576                         goto err_create_rx_queues;
2577
2578                 return 0;
2579
2580 err_create_rx_queues:
2581                 ena_free_all_io_rx_resources(adapter);
2582 err_setup_rx:
2583                 ena_destroy_all_tx_queues(adapter);
2584 err_create_tx_queues:
2585                 ena_free_all_io_tx_resources(adapter);
2586 err_setup_tx:
2587                 if (rc != -ENOMEM) {
2588                         netif_err(adapter, ifup, adapter->netdev,
2589                                   "Queue creation failed with error code %d\n",
2590                                   rc);
2591                         return rc;
2592                 }
2593
2594                 cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2595                 cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2596
2597                 netif_err(adapter, ifup, adapter->netdev,
2598                           "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2599                           cur_tx_ring_size, cur_rx_ring_size);
2600
2601                 new_tx_ring_size = cur_tx_ring_size;
2602                 new_rx_ring_size = cur_rx_ring_size;
2603
2604                 /* Decrease the size of the larger queue, or
2605                  * decrease both if they are the same size.
2606                  */
2607                 if (cur_rx_ring_size <= cur_tx_ring_size)
2608                         new_tx_ring_size = cur_tx_ring_size / 2;
2609                 if (cur_rx_ring_size >= cur_tx_ring_size)
2610                         new_rx_ring_size = cur_rx_ring_size / 2;
2611
2612                 if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2613                     new_rx_ring_size < ENA_MIN_RING_SIZE) {
2614                         netif_err(adapter, ifup, adapter->netdev,
2615                                   "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2616                                   ENA_MIN_RING_SIZE);
2617                         return rc;
2618                 }
2619
2620                 netif_err(adapter, ifup, adapter->netdev,
2621                           "Retrying queue creation with sizes TX=%d, RX=%d\n",
2622                           new_tx_ring_size,
2623                           new_rx_ring_size);
2624
2625                 set_io_rings_size(adapter, new_tx_ring_size,
2626                                   new_rx_ring_size);
2627         }
2628 }
2629
2630 static int ena_up(struct ena_adapter *adapter)
2631 {
2632         int io_queue_count, rc, i;
2633
2634         netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
2635
2636         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2637         ena_setup_io_intr(adapter);
2638
2639         /* napi poll functions should be initialized before running
2640          * request_irq(), to handle a rare condition where there is a pending
2641          * interrupt, causing the ISR to fire immediately while the poll
2642          * function wasn't set yet, causing a null dereference
2643          */
2644         ena_init_napi_in_range(adapter, 0, io_queue_count);
2645
2646         rc = ena_request_io_irq(adapter);
2647         if (rc)
2648                 goto err_req_irq;
2649
2650         rc = create_queues_with_size_backoff(adapter);
2651         if (rc)
2652                 goto err_create_queues_with_backoff;
2653
2654         rc = ena_up_complete(adapter);
2655         if (rc)
2656                 goto err_up;
2657
2658         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2659                 netif_carrier_on(adapter->netdev);
2660
2661         ena_increase_stat(&adapter->dev_stats.interface_up, 1,
2662                           &adapter->syncp);
2663
2664         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2665
2666         /* Enable completion queues interrupt */
2667         for (i = 0; i < adapter->num_io_queues; i++)
2668                 ena_unmask_interrupt(&adapter->tx_ring[i],
2669                                      &adapter->rx_ring[i]);
2670
2671         /* schedule napi in case we had pending packets
2672          * from the last time we disable napi
2673          */
2674         for (i = 0; i < io_queue_count; i++)
2675                 napi_schedule(&adapter->ena_napi[i].napi);
2676
2677         return rc;
2678
2679 err_up:
2680         ena_destroy_all_tx_queues(adapter);
2681         ena_free_all_io_tx_resources(adapter);
2682         ena_destroy_all_rx_queues(adapter);
2683         ena_free_all_io_rx_resources(adapter);
2684 err_create_queues_with_backoff:
2685         ena_free_io_irq(adapter);
2686 err_req_irq:
2687         ena_del_napi_in_range(adapter, 0, io_queue_count);
2688
2689         return rc;
2690 }
2691
2692 static void ena_down(struct ena_adapter *adapter)
2693 {
2694         int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2695
2696         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2697
2698         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2699
2700         ena_increase_stat(&adapter->dev_stats.interface_down, 1,
2701                           &adapter->syncp);
2702
2703         netif_carrier_off(adapter->netdev);
2704         netif_tx_disable(adapter->netdev);
2705
2706         /* After this point the napi handler won't enable the tx queue */
2707         ena_napi_disable_in_range(adapter, 0, io_queue_count);
2708
2709         /* After destroy the queue there won't be any new interrupts */
2710
2711         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2712                 int rc;
2713
2714                 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2715                 if (rc)
2716                         netif_err(adapter, ifdown, adapter->netdev,
2717                                   "Device reset failed\n");
2718                 /* stop submitting admin commands on a device that was reset */
2719                 ena_com_set_admin_running_state(adapter->ena_dev, false);
2720         }
2721
2722         ena_destroy_all_io_queues(adapter);
2723
2724         ena_disable_io_intr_sync(adapter);
2725         ena_free_io_irq(adapter);
2726         ena_del_napi_in_range(adapter, 0, io_queue_count);
2727
2728         ena_free_all_tx_bufs(adapter);
2729         ena_free_all_rx_bufs(adapter);
2730         ena_free_all_io_tx_resources(adapter);
2731         ena_free_all_io_rx_resources(adapter);
2732 }
2733
2734 /* ena_open - Called when a network interface is made active
2735  * @netdev: network interface device structure
2736  *
2737  * Returns 0 on success, negative value on failure
2738  *
2739  * The open entry point is called when a network interface is made
2740  * active by the system (IFF_UP).  At this point all resources needed
2741  * for transmit and receive operations are allocated, the interrupt
2742  * handler is registered with the OS, the watchdog timer is started,
2743  * and the stack is notified that the interface is ready.
2744  */
2745 static int ena_open(struct net_device *netdev)
2746 {
2747         struct ena_adapter *adapter = netdev_priv(netdev);
2748         int rc;
2749
2750         /* Notify the stack of the actual queue counts. */
2751         rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2752         if (rc) {
2753                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2754                 return rc;
2755         }
2756
2757         rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2758         if (rc) {
2759                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2760                 return rc;
2761         }
2762
2763         rc = ena_up(adapter);
2764         if (rc)
2765                 return rc;
2766
2767         return rc;
2768 }
2769
2770 /* ena_close - Disables a network interface
2771  * @netdev: network interface device structure
2772  *
2773  * Returns 0, this is not allowed to fail
2774  *
2775  * The close entry point is called when an interface is de-activated
2776  * by the OS.  The hardware is still under the drivers control, but
2777  * needs to be disabled.  A global MAC reset is issued to stop the
2778  * hardware, and all transmit and receive resources are freed.
2779  */
2780 static int ena_close(struct net_device *netdev)
2781 {
2782         struct ena_adapter *adapter = netdev_priv(netdev);
2783
2784         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2785
2786         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2787                 return 0;
2788
2789         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2790                 ena_down(adapter);
2791
2792         /* Check for device status and issue reset if needed*/
2793         check_for_admin_com_state(adapter);
2794         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2795                 netif_err(adapter, ifdown, adapter->netdev,
2796                           "Destroy failure, restarting device\n");
2797                 ena_dump_stats_to_dmesg(adapter);
2798                 /* rtnl lock already obtained in dev_ioctl() layer */
2799                 ena_destroy_device(adapter, false);
2800                 ena_restore_device(adapter);
2801         }
2802
2803         return 0;
2804 }
2805
2806 int ena_update_queue_sizes(struct ena_adapter *adapter,
2807                            u32 new_tx_size,
2808                            u32 new_rx_size)
2809 {
2810         bool dev_was_up;
2811
2812         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2813         ena_close(adapter->netdev);
2814         adapter->requested_tx_ring_size = new_tx_size;
2815         adapter->requested_rx_ring_size = new_rx_size;
2816         ena_init_io_rings(adapter,
2817                           0,
2818                           adapter->xdp_num_queues +
2819                           adapter->num_io_queues);
2820         return dev_was_up ? ena_up(adapter) : 0;
2821 }
2822
2823 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2824 {
2825         struct ena_com_dev *ena_dev = adapter->ena_dev;
2826         int prev_channel_count;
2827         bool dev_was_up;
2828
2829         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2830         ena_close(adapter->netdev);
2831         prev_channel_count = adapter->num_io_queues;
2832         adapter->num_io_queues = new_channel_count;
2833         if (ena_xdp_present(adapter) &&
2834             ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2835                 adapter->xdp_first_ring = new_channel_count;
2836                 adapter->xdp_num_queues = new_channel_count;
2837                 if (prev_channel_count > new_channel_count)
2838                         ena_xdp_exchange_program_rx_in_range(adapter,
2839                                                              NULL,
2840                                                              new_channel_count,
2841                                                              prev_channel_count);
2842                 else
2843                         ena_xdp_exchange_program_rx_in_range(adapter,
2844                                                              adapter->xdp_bpf_prog,
2845                                                              prev_channel_count,
2846                                                              new_channel_count);
2847         }
2848
2849         /* We need to destroy the rss table so that the indirection
2850          * table will be reinitialized by ena_up()
2851          */
2852         ena_com_rss_destroy(ena_dev);
2853         ena_init_io_rings(adapter,
2854                           0,
2855                           adapter->xdp_num_queues +
2856                           adapter->num_io_queues);
2857         return dev_was_up ? ena_open(adapter->netdev) : 0;
2858 }
2859
2860 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx,
2861                         struct sk_buff *skb,
2862                         bool disable_meta_caching)
2863 {
2864         u32 mss = skb_shinfo(skb)->gso_size;
2865         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2866         u8 l4_protocol = 0;
2867
2868         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2869                 ena_tx_ctx->l4_csum_enable = 1;
2870                 if (mss) {
2871                         ena_tx_ctx->tso_enable = 1;
2872                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2873                         ena_tx_ctx->l4_csum_partial = 0;
2874                 } else {
2875                         ena_tx_ctx->tso_enable = 0;
2876                         ena_meta->l4_hdr_len = 0;
2877                         ena_tx_ctx->l4_csum_partial = 1;
2878                 }
2879
2880                 switch (ip_hdr(skb)->version) {
2881                 case IPVERSION:
2882                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2883                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
2884                                 ena_tx_ctx->df = 1;
2885                         if (mss)
2886                                 ena_tx_ctx->l3_csum_enable = 1;
2887                         l4_protocol = ip_hdr(skb)->protocol;
2888                         break;
2889                 case 6:
2890                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2891                         l4_protocol = ipv6_hdr(skb)->nexthdr;
2892                         break;
2893                 default:
2894                         break;
2895                 }
2896
2897                 if (l4_protocol == IPPROTO_TCP)
2898                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2899                 else
2900                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2901
2902                 ena_meta->mss = mss;
2903                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
2904                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
2905                 ena_tx_ctx->meta_valid = 1;
2906         } else if (disable_meta_caching) {
2907                 memset(ena_meta, 0, sizeof(*ena_meta));
2908                 ena_tx_ctx->meta_valid = 1;
2909         } else {
2910                 ena_tx_ctx->meta_valid = 0;
2911         }
2912 }
2913
2914 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2915                                        struct sk_buff *skb)
2916 {
2917         int num_frags, header_len, rc;
2918
2919         num_frags = skb_shinfo(skb)->nr_frags;
2920         header_len = skb_headlen(skb);
2921
2922         if (num_frags < tx_ring->sgl_size)
2923                 return 0;
2924
2925         if ((num_frags == tx_ring->sgl_size) &&
2926             (header_len < tx_ring->tx_max_header_size))
2927                 return 0;
2928
2929         ena_increase_stat(&tx_ring->tx_stats.linearize, 1, &tx_ring->syncp);
2930
2931         rc = skb_linearize(skb);
2932         if (unlikely(rc)) {
2933                 ena_increase_stat(&tx_ring->tx_stats.linearize_failed, 1,
2934                                   &tx_ring->syncp);
2935         }
2936
2937         return rc;
2938 }
2939
2940 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2941                           struct ena_tx_buffer *tx_info,
2942                           struct sk_buff *skb,
2943                           void **push_hdr,
2944                           u16 *header_len)
2945 {
2946         struct ena_adapter *adapter = tx_ring->adapter;
2947         struct ena_com_buf *ena_buf;
2948         dma_addr_t dma;
2949         u32 skb_head_len, frag_len, last_frag;
2950         u16 push_len = 0;
2951         u16 delta = 0;
2952         int i = 0;
2953
2954         skb_head_len = skb_headlen(skb);
2955         tx_info->skb = skb;
2956         ena_buf = tx_info->bufs;
2957
2958         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2959                 /* When the device is LLQ mode, the driver will copy
2960                  * the header into the device memory space.
2961                  * the ena_com layer assume the header is in a linear
2962                  * memory space.
2963                  * This assumption might be wrong since part of the header
2964                  * can be in the fragmented buffers.
2965                  * Use skb_header_pointer to make sure the header is in a
2966                  * linear memory space.
2967                  */
2968
2969                 push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2970                 *push_hdr = skb_header_pointer(skb, 0, push_len,
2971                                                tx_ring->push_buf_intermediate_buf);
2972                 *header_len = push_len;
2973                 if (unlikely(skb->data != *push_hdr)) {
2974                         ena_increase_stat(&tx_ring->tx_stats.llq_buffer_copy, 1,
2975                                           &tx_ring->syncp);
2976
2977                         delta = push_len - skb_head_len;
2978                 }
2979         } else {
2980                 *push_hdr = NULL;
2981                 *header_len = min_t(u32, skb_head_len,
2982                                     tx_ring->tx_max_header_size);
2983         }
2984
2985         netif_dbg(adapter, tx_queued, adapter->netdev,
2986                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2987                   *push_hdr, push_len);
2988
2989         if (skb_head_len > push_len) {
2990                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2991                                      skb_head_len - push_len, DMA_TO_DEVICE);
2992                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2993                         goto error_report_dma_error;
2994
2995                 ena_buf->paddr = dma;
2996                 ena_buf->len = skb_head_len - push_len;
2997
2998                 ena_buf++;
2999                 tx_info->num_of_bufs++;
3000                 tx_info->map_linear_data = 1;
3001         } else {
3002                 tx_info->map_linear_data = 0;
3003         }
3004
3005         last_frag = skb_shinfo(skb)->nr_frags;
3006
3007         for (i = 0; i < last_frag; i++) {
3008                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3009
3010                 frag_len = skb_frag_size(frag);
3011
3012                 if (unlikely(delta >= frag_len)) {
3013                         delta -= frag_len;
3014                         continue;
3015                 }
3016
3017                 dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
3018                                        frag_len - delta, DMA_TO_DEVICE);
3019                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
3020                         goto error_report_dma_error;
3021
3022                 ena_buf->paddr = dma;
3023                 ena_buf->len = frag_len - delta;
3024                 ena_buf++;
3025                 tx_info->num_of_bufs++;
3026                 delta = 0;
3027         }
3028
3029         return 0;
3030
3031 error_report_dma_error:
3032         ena_increase_stat(&tx_ring->tx_stats.dma_mapping_err, 1,
3033                           &tx_ring->syncp);
3034         netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map skb\n");
3035
3036         tx_info->skb = NULL;
3037
3038         tx_info->num_of_bufs += i;
3039         ena_unmap_tx_buff(tx_ring, tx_info);
3040
3041         return -EINVAL;
3042 }
3043
3044 /* Called with netif_tx_lock. */
3045 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
3046 {
3047         struct ena_adapter *adapter = netdev_priv(dev);
3048         struct ena_tx_buffer *tx_info;
3049         struct ena_com_tx_ctx ena_tx_ctx;
3050         struct ena_ring *tx_ring;
3051         struct netdev_queue *txq;
3052         void *push_hdr;
3053         u16 next_to_use, req_id, header_len;
3054         int qid, rc;
3055
3056         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
3057         /*  Determine which tx ring we will be placed on */
3058         qid = skb_get_queue_mapping(skb);
3059         tx_ring = &adapter->tx_ring[qid];
3060         txq = netdev_get_tx_queue(dev, qid);
3061
3062         rc = ena_check_and_linearize_skb(tx_ring, skb);
3063         if (unlikely(rc))
3064                 goto error_drop_packet;
3065
3066         skb_tx_timestamp(skb);
3067
3068         next_to_use = tx_ring->next_to_use;
3069         req_id = tx_ring->free_ids[next_to_use];
3070         tx_info = &tx_ring->tx_buffer_info[req_id];
3071         tx_info->num_of_bufs = 0;
3072
3073         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
3074
3075         rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
3076         if (unlikely(rc))
3077                 goto error_drop_packet;
3078
3079         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
3080         ena_tx_ctx.ena_bufs = tx_info->bufs;
3081         ena_tx_ctx.push_header = push_hdr;
3082         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
3083         ena_tx_ctx.req_id = req_id;
3084         ena_tx_ctx.header_len = header_len;
3085
3086         /* set flags and meta data */
3087         ena_tx_csum(&ena_tx_ctx, skb, tx_ring->disable_meta_caching);
3088
3089         rc = ena_xmit_common(dev,
3090                              tx_ring,
3091                              tx_info,
3092                              &ena_tx_ctx,
3093                              next_to_use,
3094                              skb->len);
3095         if (rc)
3096                 goto error_unmap_dma;
3097
3098         netdev_tx_sent_queue(txq, skb->len);
3099
3100         /* stop the queue when no more space available, the packet can have up
3101          * to sgl_size + 2. one for the meta descriptor and one for header
3102          * (if the header is larger than tx_max_header_size).
3103          */
3104         if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3105                                                    tx_ring->sgl_size + 2))) {
3106                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3107                           __func__, qid);
3108
3109                 netif_tx_stop_queue(txq);
3110                 ena_increase_stat(&tx_ring->tx_stats.queue_stop, 1,
3111                                   &tx_ring->syncp);
3112
3113                 /* There is a rare condition where this function decide to
3114                  * stop the queue but meanwhile clean_tx_irq updates
3115                  * next_to_completion and terminates.
3116                  * The queue will remain stopped forever.
3117                  * To solve this issue add a mb() to make sure that
3118                  * netif_tx_stop_queue() write is vissible before checking if
3119                  * there is additional space in the queue.
3120                  */
3121                 smp_mb();
3122
3123                 if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3124                                                  ENA_TX_WAKEUP_THRESH)) {
3125                         netif_tx_wake_queue(txq);
3126                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
3127                                           &tx_ring->syncp);
3128                 }
3129         }
3130
3131         if (netif_xmit_stopped(txq) || !netdev_xmit_more())
3132                 /* trigger the dma engine. ena_ring_tx_doorbell()
3133                  * calls a memory barrier inside it.
3134                  */
3135                 ena_ring_tx_doorbell(tx_ring);
3136
3137         return NETDEV_TX_OK;
3138
3139 error_unmap_dma:
3140         ena_unmap_tx_buff(tx_ring, tx_info);
3141         tx_info->skb = NULL;
3142
3143 error_drop_packet:
3144         dev_kfree_skb(skb);
3145         return NETDEV_TX_OK;
3146 }
3147
3148 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3149                             struct net_device *sb_dev)
3150 {
3151         u16 qid;
3152         /* we suspect that this is good for in--kernel network services that
3153          * want to loop incoming skb rx to tx in normal user generated traffic,
3154          * most probably we will not get to this
3155          */
3156         if (skb_rx_queue_recorded(skb))
3157                 qid = skb_get_rx_queue(skb);
3158         else
3159                 qid = netdev_pick_tx(dev, skb, NULL);
3160
3161         return qid;
3162 }
3163
3164 static void ena_config_host_info(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3165 {
3166         struct device *dev = &pdev->dev;
3167         struct ena_admin_host_info *host_info;
3168         int rc;
3169
3170         /* Allocate only the host info */
3171         rc = ena_com_allocate_host_info(ena_dev);
3172         if (rc) {
3173                 dev_err(dev, "Cannot allocate host info\n");
3174                 return;
3175         }
3176
3177         host_info = ena_dev->host_attr.host_info;
3178
3179         host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
3180         host_info->os_type = ENA_ADMIN_OS_LINUX;
3181         host_info->kernel_ver = LINUX_VERSION_CODE;
3182         strlcpy(host_info->kernel_ver_str, utsname()->version,
3183                 sizeof(host_info->kernel_ver_str) - 1);
3184         host_info->os_dist = 0;
3185         strncpy(host_info->os_dist_str, utsname()->release,
3186                 sizeof(host_info->os_dist_str) - 1);
3187         host_info->driver_version =
3188                 (DRV_MODULE_GEN_MAJOR) |
3189                 (DRV_MODULE_GEN_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3190                 (DRV_MODULE_GEN_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3191                 ("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3192         host_info->num_cpus = num_online_cpus();
3193
3194         host_info->driver_supported_features =
3195                 ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK |
3196                 ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK |
3197                 ENA_ADMIN_HOST_INFO_RX_BUF_MIRRORING_MASK |
3198                 ENA_ADMIN_HOST_INFO_RSS_CONFIGURABLE_FUNCTION_KEY_MASK;
3199
3200         rc = ena_com_set_host_attributes(ena_dev);
3201         if (rc) {
3202                 if (rc == -EOPNOTSUPP)
3203                         dev_warn(dev, "Cannot set host attributes\n");
3204                 else
3205                         dev_err(dev, "Cannot set host attributes\n");
3206
3207                 goto err;
3208         }
3209
3210         return;
3211
3212 err:
3213         ena_com_delete_host_info(ena_dev);
3214 }
3215
3216 static void ena_config_debug_area(struct ena_adapter *adapter)
3217 {
3218         u32 debug_area_size;
3219         int rc, ss_count;
3220
3221         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3222         if (ss_count <= 0) {
3223                 netif_err(adapter, drv, adapter->netdev,
3224                           "SS count is negative\n");
3225                 return;
3226         }
3227
3228         /* allocate 32 bytes for each string and 64bit for the value */
3229         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3230
3231         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3232         if (rc) {
3233                 netif_err(adapter, drv, adapter->netdev,
3234                           "Cannot allocate debug area\n");
3235                 return;
3236         }
3237
3238         rc = ena_com_set_host_attributes(adapter->ena_dev);
3239         if (rc) {
3240                 if (rc == -EOPNOTSUPP)
3241                         netif_warn(adapter, drv, adapter->netdev,
3242                                    "Cannot set host attributes\n");
3243                 else
3244                         netif_err(adapter, drv, adapter->netdev,
3245                                   "Cannot set host attributes\n");
3246                 goto err;
3247         }
3248
3249         return;
3250 err:
3251         ena_com_delete_debug_area(adapter->ena_dev);
3252 }
3253
3254 int ena_update_hw_stats(struct ena_adapter *adapter)
3255 {
3256         int rc = 0;
3257
3258         rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_stats);
3259         if (rc) {
3260                 dev_info_once(&adapter->pdev->dev, "Failed to get ENI stats\n");
3261                 return rc;
3262         }
3263
3264         return 0;
3265 }
3266
3267 static void ena_get_stats64(struct net_device *netdev,
3268                             struct rtnl_link_stats64 *stats)
3269 {
3270         struct ena_adapter *adapter = netdev_priv(netdev);
3271         struct ena_ring *rx_ring, *tx_ring;
3272         unsigned int start;
3273         u64 rx_drops;
3274         u64 tx_drops;
3275         int i;
3276
3277         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3278                 return;
3279
3280         for (i = 0; i < adapter->num_io_queues; i++) {
3281                 u64 bytes, packets;
3282
3283                 tx_ring = &adapter->tx_ring[i];
3284
3285                 do {
3286                         start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
3287                         packets = tx_ring->tx_stats.cnt;
3288                         bytes = tx_ring->tx_stats.bytes;
3289                 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
3290
3291                 stats->tx_packets += packets;
3292                 stats->tx_bytes += bytes;
3293
3294                 rx_ring = &adapter->rx_ring[i];
3295
3296                 do {
3297                         start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
3298                         packets = rx_ring->rx_stats.cnt;
3299                         bytes = rx_ring->rx_stats.bytes;
3300                 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
3301
3302                 stats->rx_packets += packets;
3303                 stats->rx_bytes += bytes;
3304         }
3305
3306         do {
3307                 start = u64_stats_fetch_begin_irq(&adapter->syncp);
3308                 rx_drops = adapter->dev_stats.rx_drops;
3309                 tx_drops = adapter->dev_stats.tx_drops;
3310         } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
3311
3312         stats->rx_dropped = rx_drops;
3313         stats->tx_dropped = tx_drops;
3314
3315         stats->multicast = 0;
3316         stats->collisions = 0;
3317
3318         stats->rx_length_errors = 0;
3319         stats->rx_crc_errors = 0;
3320         stats->rx_frame_errors = 0;
3321         stats->rx_fifo_errors = 0;
3322         stats->rx_missed_errors = 0;
3323         stats->tx_window_errors = 0;
3324
3325         stats->rx_errors = 0;
3326         stats->tx_errors = 0;
3327 }
3328
3329 static const struct net_device_ops ena_netdev_ops = {
3330         .ndo_open               = ena_open,
3331         .ndo_stop               = ena_close,
3332         .ndo_start_xmit         = ena_start_xmit,
3333         .ndo_select_queue       = ena_select_queue,
3334         .ndo_get_stats64        = ena_get_stats64,
3335         .ndo_tx_timeout         = ena_tx_timeout,
3336         .ndo_change_mtu         = ena_change_mtu,
3337         .ndo_set_mac_address    = NULL,
3338         .ndo_validate_addr      = eth_validate_addr,
3339         .ndo_bpf                = ena_xdp,
3340         .ndo_xdp_xmit           = ena_xdp_xmit,
3341 };
3342
3343 static int ena_device_validate_params(struct ena_adapter *adapter,
3344                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
3345 {
3346         struct net_device *netdev = adapter->netdev;
3347         int rc;
3348
3349         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3350                               adapter->mac_addr);
3351         if (!rc) {
3352                 netif_err(adapter, drv, netdev,
3353                           "Error, mac address are different\n");
3354                 return -EINVAL;
3355         }
3356
3357         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3358                 netif_err(adapter, drv, netdev,
3359                           "Error, device max mtu is smaller than netdev MTU\n");
3360                 return -EINVAL;
3361         }
3362
3363         return 0;
3364 }
3365
3366 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
3367 {
3368         llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
3369         llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
3370         llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
3371         llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
3372         llq_config->llq_ring_entry_size_value = 128;
3373 }
3374
3375 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3376                                            struct ena_com_dev *ena_dev,
3377                                            struct ena_admin_feature_llq_desc *llq,
3378                                            struct ena_llq_configurations *llq_default_configurations)
3379 {
3380         int rc;
3381         u32 llq_feature_mask;
3382
3383         llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3384         if (!(ena_dev->supported_features & llq_feature_mask)) {
3385                 dev_warn(&pdev->dev,
3386                         "LLQ is not supported Fallback to host mode policy.\n");
3387                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3388                 return 0;
3389         }
3390
3391         rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3392         if (unlikely(rc)) {
3393                 dev_err(&pdev->dev,
3394                         "Failed to configure the device mode.  Fallback to host mode policy.\n");
3395                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3396         }
3397
3398         return 0;
3399 }
3400
3401 static int ena_map_llq_mem_bar(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
3402                                int bars)
3403 {
3404         bool has_mem_bar = !!(bars & BIT(ENA_MEM_BAR));
3405
3406         if (!has_mem_bar) {
3407                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3408                         dev_err(&pdev->dev,
3409                                 "ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3410                         ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3411                 }
3412
3413                 return 0;
3414         }
3415
3416         ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3417                                            pci_resource_start(pdev, ENA_MEM_BAR),
3418                                            pci_resource_len(pdev, ENA_MEM_BAR));
3419
3420         if (!ena_dev->mem_bar)
3421                 return -EFAULT;
3422
3423         return 0;
3424 }
3425
3426 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
3427                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
3428                            bool *wd_state)
3429 {
3430         struct ena_llq_configurations llq_config;
3431         struct device *dev = &pdev->dev;
3432         bool readless_supported;
3433         u32 aenq_groups;
3434         int dma_width;
3435         int rc;
3436
3437         rc = ena_com_mmio_reg_read_request_init(ena_dev);
3438         if (rc) {
3439                 dev_err(dev, "Failed to init mmio read less\n");
3440                 return rc;
3441         }
3442
3443         /* The PCIe configuration space revision id indicate if mmio reg
3444          * read is disabled
3445          */
3446         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3447         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3448
3449         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3450         if (rc) {
3451                 dev_err(dev, "Can not reset device\n");
3452                 goto err_mmio_read_less;
3453         }
3454
3455         rc = ena_com_validate_version(ena_dev);
3456         if (rc) {
3457                 dev_err(dev, "Device version is too low\n");
3458                 goto err_mmio_read_less;
3459         }
3460
3461         dma_width = ena_com_get_dma_width(ena_dev);
3462         if (dma_width < 0) {
3463                 dev_err(dev, "Invalid dma width value %d", dma_width);
3464                 rc = dma_width;
3465                 goto err_mmio_read_less;
3466         }
3467
3468         rc = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(dma_width));
3469         if (rc) {
3470                 dev_err(dev, "dma_set_mask_and_coherent failed %d\n", rc);
3471                 goto err_mmio_read_less;
3472         }
3473
3474         /* ENA admin level init */
3475         rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3476         if (rc) {
3477                 dev_err(dev,
3478                         "Can not initialize ena admin queue with device\n");
3479                 goto err_mmio_read_less;
3480         }
3481
3482         /* To enable the msix interrupts the driver needs to know the number
3483          * of queues. So the driver uses polling mode to retrieve this
3484          * information
3485          */
3486         ena_com_set_admin_polling_mode(ena_dev, true);
3487
3488         ena_config_host_info(ena_dev, pdev);
3489
3490         /* Get Device Attributes*/
3491         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3492         if (rc) {
3493                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3494                 goto err_admin_init;
3495         }
3496
3497         /* Try to turn all the available aenq groups */
3498         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3499                 BIT(ENA_ADMIN_FATAL_ERROR) |
3500                 BIT(ENA_ADMIN_WARNING) |
3501                 BIT(ENA_ADMIN_NOTIFICATION) |
3502                 BIT(ENA_ADMIN_KEEP_ALIVE);
3503
3504         aenq_groups &= get_feat_ctx->aenq.supported_groups;
3505
3506         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3507         if (rc) {
3508                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3509                 goto err_admin_init;
3510         }
3511
3512         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3513
3514         set_default_llq_configurations(&llq_config);
3515
3516         rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx->llq,
3517                                              &llq_config);
3518         if (rc) {
3519                 dev_err(dev, "ENA device init failed\n");
3520                 goto err_admin_init;
3521         }
3522
3523         return 0;
3524
3525 err_admin_init:
3526         ena_com_delete_host_info(ena_dev);
3527         ena_com_admin_destroy(ena_dev);
3528 err_mmio_read_less:
3529         ena_com_mmio_reg_read_request_destroy(ena_dev);
3530
3531         return rc;
3532 }
3533
3534 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3535 {
3536         struct ena_com_dev *ena_dev = adapter->ena_dev;
3537         struct device *dev = &adapter->pdev->dev;
3538         int rc;
3539
3540         rc = ena_enable_msix(adapter);
3541         if (rc) {
3542                 dev_err(dev, "Can not reserve msix vectors\n");
3543                 return rc;
3544         }
3545
3546         ena_setup_mgmnt_intr(adapter);
3547
3548         rc = ena_request_mgmnt_irq(adapter);
3549         if (rc) {
3550                 dev_err(dev, "Can not setup management interrupts\n");
3551                 goto err_disable_msix;
3552         }
3553
3554         ena_com_set_admin_polling_mode(ena_dev, false);
3555
3556         ena_com_admin_aenq_enable(ena_dev);
3557
3558         return 0;
3559
3560 err_disable_msix:
3561         ena_disable_msix(adapter);
3562
3563         return rc;
3564 }
3565
3566 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3567 {
3568         struct net_device *netdev = adapter->netdev;
3569         struct ena_com_dev *ena_dev = adapter->ena_dev;
3570         bool dev_up;
3571
3572         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3573                 return;
3574
3575         netif_carrier_off(netdev);
3576
3577         del_timer_sync(&adapter->timer_service);
3578
3579         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3580         adapter->dev_up_before_reset = dev_up;
3581         if (!graceful)
3582                 ena_com_set_admin_running_state(ena_dev, false);
3583
3584         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3585                 ena_down(adapter);
3586
3587         /* Stop the device from sending AENQ events (in case reset flag is set
3588          *  and device is up, ena_down() already reset the device.
3589          */
3590         if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3591                 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3592
3593         ena_free_mgmnt_irq(adapter);
3594
3595         ena_disable_msix(adapter);
3596
3597         ena_com_abort_admin_commands(ena_dev);
3598
3599         ena_com_wait_for_abort_completion(ena_dev);
3600
3601         ena_com_admin_destroy(ena_dev);
3602
3603         ena_com_mmio_reg_read_request_destroy(ena_dev);
3604
3605         /* return reset reason to default value */
3606         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3607
3608         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3609         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3610 }
3611
3612 static int ena_restore_device(struct ena_adapter *adapter)
3613 {
3614         struct ena_com_dev_get_features_ctx get_feat_ctx;
3615         struct ena_com_dev *ena_dev = adapter->ena_dev;
3616         struct pci_dev *pdev = adapter->pdev;
3617         bool wd_state;
3618         int rc;
3619
3620         set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3621         rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
3622         if (rc) {
3623                 dev_err(&pdev->dev, "Can not initialize device\n");
3624                 goto err;
3625         }
3626         adapter->wd_state = wd_state;
3627
3628         rc = ena_device_validate_params(adapter, &get_feat_ctx);
3629         if (rc) {
3630                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
3631                 goto err_device_destroy;
3632         }
3633
3634         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3635         if (rc) {
3636                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
3637                 goto err_device_destroy;
3638         }
3639         /* If the interface was up before the reset bring it up */
3640         if (adapter->dev_up_before_reset) {
3641                 rc = ena_up(adapter);
3642                 if (rc) {
3643                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
3644                         goto err_disable_msix;
3645                 }
3646         }
3647
3648         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3649
3650         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3651         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3652                 netif_carrier_on(adapter->netdev);
3653
3654         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3655         adapter->last_keep_alive_jiffies = jiffies;
3656
3657         dev_err(&pdev->dev, "Device reset completed successfully\n");
3658
3659         return rc;
3660 err_disable_msix:
3661         ena_free_mgmnt_irq(adapter);
3662         ena_disable_msix(adapter);
3663 err_device_destroy:
3664         ena_com_abort_admin_commands(ena_dev);
3665         ena_com_wait_for_abort_completion(ena_dev);
3666         ena_com_admin_destroy(ena_dev);
3667         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3668         ena_com_mmio_reg_read_request_destroy(ena_dev);
3669 err:
3670         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3671         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3672         dev_err(&pdev->dev,
3673                 "Reset attempt failed. Can not reset the device\n");
3674
3675         return rc;
3676 }
3677
3678 static void ena_fw_reset_device(struct work_struct *work)
3679 {
3680         struct ena_adapter *adapter =
3681                 container_of(work, struct ena_adapter, reset_task);
3682
3683         rtnl_lock();
3684
3685         if (likely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3686                 ena_destroy_device(adapter, false);
3687                 ena_restore_device(adapter);
3688         }
3689
3690         rtnl_unlock();
3691 }
3692
3693 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3694                                         struct ena_ring *rx_ring)
3695 {
3696         struct ena_napi *ena_napi = container_of(rx_ring->napi, struct ena_napi, napi);
3697
3698         if (likely(READ_ONCE(ena_napi->first_interrupt)))
3699                 return 0;
3700
3701         if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3702                 return 0;
3703
3704         rx_ring->no_interrupt_event_cnt++;
3705
3706         if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3707                 netif_err(adapter, rx_err, adapter->netdev,
3708                           "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3709                           rx_ring->qid);
3710                 adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3711                 smp_mb__before_atomic();
3712                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3713                 return -EIO;
3714         }
3715
3716         return 0;
3717 }
3718
3719 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3720                                           struct ena_ring *tx_ring)
3721 {
3722         struct ena_napi *ena_napi = container_of(tx_ring->napi, struct ena_napi, napi);
3723         unsigned int time_since_last_napi;
3724         unsigned int missing_tx_comp_to;
3725         bool is_tx_comp_time_expired;
3726         struct ena_tx_buffer *tx_buf;
3727         unsigned long last_jiffies;
3728         u32 missed_tx = 0;
3729         int i, rc = 0;
3730
3731         for (i = 0; i < tx_ring->ring_size; i++) {
3732                 tx_buf = &tx_ring->tx_buffer_info[i];
3733                 last_jiffies = tx_buf->last_jiffies;
3734
3735                 if (last_jiffies == 0)
3736                         /* no pending Tx at this location */
3737                         continue;
3738
3739                 is_tx_comp_time_expired = time_is_before_jiffies(last_jiffies +
3740                          2 * adapter->missing_tx_completion_to);
3741
3742                 if (unlikely(!READ_ONCE(ena_napi->first_interrupt) && is_tx_comp_time_expired)) {
3743                         /* If after graceful period interrupt is still not
3744                          * received, we schedule a reset
3745                          */
3746                         netif_err(adapter, tx_err, adapter->netdev,
3747                                   "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3748                                   tx_ring->qid);
3749                         adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3750                         smp_mb__before_atomic();
3751                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3752                         return -EIO;
3753                 }
3754
3755                 is_tx_comp_time_expired = time_is_before_jiffies(last_jiffies +
3756                         adapter->missing_tx_completion_to);
3757
3758                 if (unlikely(is_tx_comp_time_expired)) {
3759                         if (!tx_buf->print_once) {
3760                                 time_since_last_napi = jiffies_to_usecs(jiffies - tx_ring->tx_stats.last_napi_jiffies);
3761                                 missing_tx_comp_to = jiffies_to_msecs(adapter->missing_tx_completion_to);
3762                                 netif_notice(adapter, tx_err, adapter->netdev,
3763                                              "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",
3764                                              tx_ring->qid, i, time_since_last_napi, missing_tx_comp_to);
3765                         }
3766
3767                         tx_buf->print_once = 1;
3768                         missed_tx++;
3769                 }
3770         }
3771
3772         if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3773                 netif_err(adapter, tx_err, adapter->netdev,
3774                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3775                           missed_tx,
3776                           adapter->missing_tx_completion_threshold);
3777                 adapter->reset_reason =
3778                         ENA_REGS_RESET_MISS_TX_CMPL;
3779                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3780                 rc = -EIO;
3781         }
3782
3783         ena_increase_stat(&tx_ring->tx_stats.missed_tx, missed_tx,
3784                           &tx_ring->syncp);
3785
3786         return rc;
3787 }
3788
3789 static void check_for_missing_completions(struct ena_adapter *adapter)
3790 {
3791         struct ena_ring *tx_ring;
3792         struct ena_ring *rx_ring;
3793         int i, budget, rc;
3794         int io_queue_count;
3795
3796         io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
3797         /* Make sure the driver doesn't turn the device in other process */
3798         smp_rmb();
3799
3800         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3801                 return;
3802
3803         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3804                 return;
3805
3806         if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
3807                 return;
3808
3809         budget = ENA_MONITORED_TX_QUEUES;
3810
3811         for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
3812                 tx_ring = &adapter->tx_ring[i];
3813                 rx_ring = &adapter->rx_ring[i];
3814
3815                 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3816                 if (unlikely(rc))
3817                         return;
3818
3819                 rc =  !ENA_IS_XDP_INDEX(adapter, i) ?
3820                         check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
3821                 if (unlikely(rc))
3822                         return;
3823
3824                 budget--;
3825                 if (!budget)
3826                         break;
3827         }
3828
3829         adapter->last_monitored_tx_qid = i % io_queue_count;
3830 }
3831
3832 /* trigger napi schedule after 2 consecutive detections */
3833 #define EMPTY_RX_REFILL 2
3834 /* For the rare case where the device runs out of Rx descriptors and the
3835  * napi handler failed to refill new Rx descriptors (due to a lack of memory
3836  * for example).
3837  * This case will lead to a deadlock:
3838  * The device won't send interrupts since all the new Rx packets will be dropped
3839  * The napi handler won't allocate new Rx descriptors so the device will be
3840  * able to send new packets.
3841  *
3842  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
3843  * It is recommended to have at least 512MB, with a minimum of 128MB for
3844  * constrained environment).
3845  *
3846  * When such a situation is detected - Reschedule napi
3847  */
3848 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
3849 {
3850         struct ena_ring *rx_ring;
3851         int i, refill_required;
3852
3853         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3854                 return;
3855
3856         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3857                 return;
3858
3859         for (i = 0; i < adapter->num_io_queues; i++) {
3860                 rx_ring = &adapter->rx_ring[i];
3861
3862                 refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
3863                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3864                         rx_ring->empty_rx_queue++;
3865
3866                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3867                                 ena_increase_stat(&rx_ring->rx_stats.empty_rx_ring, 1,
3868                                                   &rx_ring->syncp);
3869
3870                                 netif_err(adapter, drv, adapter->netdev,
3871                                           "Trigger refill for ring %d\n", i);
3872
3873                                 napi_schedule(rx_ring->napi);
3874                                 rx_ring->empty_rx_queue = 0;
3875                         }
3876                 } else {
3877                         rx_ring->empty_rx_queue = 0;
3878                 }
3879         }
3880 }
3881
3882 /* Check for keep alive expiration */
3883 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3884 {
3885         unsigned long keep_alive_expired;
3886
3887         if (!adapter->wd_state)
3888                 return;
3889
3890         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3891                 return;
3892
3893         keep_alive_expired = adapter->last_keep_alive_jiffies +
3894                              adapter->keep_alive_timeout;
3895         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3896                 netif_err(adapter, drv, adapter->netdev,
3897                           "Keep alive watchdog timeout.\n");
3898                 ena_increase_stat(&adapter->dev_stats.wd_expired, 1,
3899                                   &adapter->syncp);
3900                 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3901                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3902         }
3903 }
3904
3905 static void check_for_admin_com_state(struct ena_adapter *adapter)
3906 {
3907         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3908                 netif_err(adapter, drv, adapter->netdev,
3909                           "ENA admin queue is not in running state!\n");
3910                 ena_increase_stat(&adapter->dev_stats.admin_q_pause, 1,
3911                                   &adapter->syncp);
3912                 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3913                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3914         }
3915 }
3916
3917 static void ena_update_hints(struct ena_adapter *adapter,
3918                              struct ena_admin_ena_hw_hints *hints)
3919 {
3920         struct net_device *netdev = adapter->netdev;
3921
3922         if (hints->admin_completion_tx_timeout)
3923                 adapter->ena_dev->admin_queue.completion_timeout =
3924                         hints->admin_completion_tx_timeout * 1000;
3925
3926         if (hints->mmio_read_timeout)
3927                 /* convert to usec */
3928                 adapter->ena_dev->mmio_read.reg_read_to =
3929                         hints->mmio_read_timeout * 1000;
3930
3931         if (hints->missed_tx_completion_count_threshold_to_reset)
3932                 adapter->missing_tx_completion_threshold =
3933                         hints->missed_tx_completion_count_threshold_to_reset;
3934
3935         if (hints->missing_tx_completion_timeout) {
3936                 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3937                         adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3938                 else
3939                         adapter->missing_tx_completion_to =
3940                                 msecs_to_jiffies(hints->missing_tx_completion_timeout);
3941         }
3942
3943         if (hints->netdev_wd_timeout)
3944                 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3945
3946         if (hints->driver_watchdog_timeout) {
3947                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3948                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3949                 else
3950                         adapter->keep_alive_timeout =
3951                                 msecs_to_jiffies(hints->driver_watchdog_timeout);
3952         }
3953 }
3954
3955 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3956                                  struct net_device *netdev)
3957 {
3958         host_info->supported_network_features[0] =
3959                 netdev->features & GENMASK_ULL(31, 0);
3960         host_info->supported_network_features[1] =
3961                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
3962 }
3963
3964 static void ena_timer_service(struct timer_list *t)
3965 {
3966         struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3967         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3968         struct ena_admin_host_info *host_info =
3969                 adapter->ena_dev->host_attr.host_info;
3970
3971         check_for_missing_keep_alive(adapter);
3972
3973         check_for_admin_com_state(adapter);
3974
3975         check_for_missing_completions(adapter);
3976
3977         check_for_empty_rx_ring(adapter);
3978
3979         if (debug_area)
3980                 ena_dump_stats_to_buf(adapter, debug_area);
3981
3982         if (host_info)
3983                 ena_update_host_info(host_info, adapter->netdev);
3984
3985         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3986                 netif_err(adapter, drv, adapter->netdev,
3987                           "Trigger reset is on\n");
3988                 ena_dump_stats_to_dmesg(adapter);
3989                 queue_work(ena_wq, &adapter->reset_task);
3990                 return;
3991         }
3992
3993         /* Reset the timer */
3994         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3995 }
3996
3997 static u32 ena_calc_max_io_queue_num(struct pci_dev *pdev,
3998                                      struct ena_com_dev *ena_dev,
3999                                      struct ena_com_dev_get_features_ctx *get_feat_ctx)
4000 {
4001         u32 io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
4002
4003         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4004                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4005                         &get_feat_ctx->max_queue_ext.max_queue_ext;
4006                 io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
4007                                   max_queue_ext->max_rx_cq_num);
4008
4009                 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
4010                 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
4011         } else {
4012                 struct ena_admin_queue_feature_desc *max_queues =
4013                         &get_feat_ctx->max_queues;
4014                 io_tx_sq_num = max_queues->max_sq_num;
4015                 io_tx_cq_num = max_queues->max_cq_num;
4016                 io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
4017         }
4018
4019         /* In case of LLQ use the llq fields for the tx SQ/CQ */
4020         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4021                 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
4022
4023         max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
4024         max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
4025         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
4026         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
4027         /* 1 IRQ for mgmnt and 1 IRQs for each IO direction */
4028         max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
4029
4030         return max_num_io_queues;
4031 }
4032
4033 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
4034                                  struct net_device *netdev)
4035 {
4036         netdev_features_t dev_features = 0;
4037
4038         /* Set offload features */
4039         if (feat->offload.tx &
4040                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
4041                 dev_features |= NETIF_F_IP_CSUM;
4042
4043         if (feat->offload.tx &
4044                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
4045                 dev_features |= NETIF_F_IPV6_CSUM;
4046
4047         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
4048                 dev_features |= NETIF_F_TSO;
4049
4050         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
4051                 dev_features |= NETIF_F_TSO6;
4052
4053         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
4054                 dev_features |= NETIF_F_TSO_ECN;
4055
4056         if (feat->offload.rx_supported &
4057                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
4058                 dev_features |= NETIF_F_RXCSUM;
4059
4060         if (feat->offload.rx_supported &
4061                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
4062                 dev_features |= NETIF_F_RXCSUM;
4063
4064         netdev->features =
4065                 dev_features |
4066                 NETIF_F_SG |
4067                 NETIF_F_RXHASH |
4068                 NETIF_F_HIGHDMA;
4069
4070         netdev->hw_features |= netdev->features;
4071         netdev->vlan_features |= netdev->features;
4072 }
4073
4074 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
4075                                      struct ena_com_dev_get_features_ctx *feat)
4076 {
4077         struct net_device *netdev = adapter->netdev;
4078
4079         /* Copy mac address */
4080         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
4081                 eth_hw_addr_random(netdev);
4082                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
4083         } else {
4084                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
4085                 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
4086         }
4087
4088         /* Set offload features */
4089         ena_set_dev_offloads(feat, netdev);
4090
4091         adapter->max_mtu = feat->dev_attr.max_mtu;
4092         netdev->max_mtu = adapter->max_mtu;
4093         netdev->min_mtu = ENA_MIN_MTU;
4094 }
4095
4096 static int ena_rss_init_default(struct ena_adapter *adapter)
4097 {
4098         struct ena_com_dev *ena_dev = adapter->ena_dev;
4099         struct device *dev = &adapter->pdev->dev;
4100         int rc, i;
4101         u32 val;
4102
4103         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
4104         if (unlikely(rc)) {
4105                 dev_err(dev, "Cannot init indirect table\n");
4106                 goto err_rss_init;
4107         }
4108
4109         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
4110                 val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
4111                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
4112                                                        ENA_IO_RXQ_IDX(val));
4113                 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4114                         dev_err(dev, "Cannot fill indirect table\n");
4115                         goto err_fill_indir;
4116                 }
4117         }
4118
4119         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ, NULL,
4120                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4121         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4122                 dev_err(dev, "Cannot fill hash function\n");
4123                 goto err_fill_indir;
4124         }
4125
4126         rc = ena_com_set_default_hash_ctrl(ena_dev);
4127         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4128                 dev_err(dev, "Cannot fill hash control\n");
4129                 goto err_fill_indir;
4130         }
4131
4132         return 0;
4133
4134 err_fill_indir:
4135         ena_com_rss_destroy(ena_dev);
4136 err_rss_init:
4137
4138         return rc;
4139 }
4140
4141 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4142 {
4143         int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4144
4145         pci_release_selected_regions(pdev, release_bars);
4146 }
4147
4148
4149 static int ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
4150 {
4151         struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
4152         struct ena_com_dev *ena_dev = ctx->ena_dev;
4153         u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
4154         u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
4155         u32 max_tx_queue_size;
4156         u32 max_rx_queue_size;
4157
4158         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4159                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4160                         &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
4161                 max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
4162                                           max_queue_ext->max_rx_sq_depth);
4163                 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
4164
4165                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4166                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4167                                                   llq->max_llq_depth);
4168                 else
4169                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4170                                                   max_queue_ext->max_tx_sq_depth);
4171
4172                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4173                                              max_queue_ext->max_per_packet_tx_descs);
4174                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4175                                              max_queue_ext->max_per_packet_rx_descs);
4176         } else {
4177                 struct ena_admin_queue_feature_desc *max_queues =
4178                         &ctx->get_feat_ctx->max_queues;
4179                 max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
4180                                           max_queues->max_sq_depth);
4181                 max_tx_queue_size = max_queues->max_cq_depth;
4182
4183                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4184                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4185                                                   llq->max_llq_depth);
4186                 else
4187                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4188                                                   max_queues->max_sq_depth);
4189
4190                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4191                                              max_queues->max_packet_tx_descs);
4192                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4193                                              max_queues->max_packet_rx_descs);
4194         }
4195
4196         max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
4197         max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
4198
4199         tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
4200                                   max_tx_queue_size);
4201         rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
4202                                   max_rx_queue_size);
4203
4204         tx_queue_size = rounddown_pow_of_two(tx_queue_size);
4205         rx_queue_size = rounddown_pow_of_two(rx_queue_size);
4206
4207         ctx->max_tx_queue_size = max_tx_queue_size;
4208         ctx->max_rx_queue_size = max_rx_queue_size;
4209         ctx->tx_queue_size = tx_queue_size;
4210         ctx->rx_queue_size = rx_queue_size;
4211
4212         return 0;
4213 }
4214
4215 /* ena_probe - Device Initialization Routine
4216  * @pdev: PCI device information struct
4217  * @ent: entry in ena_pci_tbl
4218  *
4219  * Returns 0 on success, negative on failure
4220  *
4221  * ena_probe initializes an adapter identified by a pci_dev structure.
4222  * The OS initialization, configuring of the adapter private structure,
4223  * and a hardware reset occur.
4224  */
4225 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4226 {
4227         struct ena_calc_queue_size_ctx calc_queue_ctx = {};
4228         struct ena_com_dev_get_features_ctx get_feat_ctx;
4229         struct ena_com_dev *ena_dev = NULL;
4230         struct ena_adapter *adapter;
4231         struct net_device *netdev;
4232         static int adapters_found;
4233         u32 max_num_io_queues;
4234         bool wd_state;
4235         int bars, rc;
4236
4237         dev_dbg(&pdev->dev, "%s\n", __func__);
4238
4239         rc = pci_enable_device_mem(pdev);
4240         if (rc) {
4241                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4242                 return rc;
4243         }
4244
4245         rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
4246         if (rc) {
4247                 dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n", rc);
4248                 goto err_disable_device;
4249         }
4250
4251         pci_set_master(pdev);
4252
4253         ena_dev = vzalloc(sizeof(*ena_dev));
4254         if (!ena_dev) {
4255                 rc = -ENOMEM;
4256                 goto err_disable_device;
4257         }
4258
4259         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4260         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4261         if (rc) {
4262                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4263                         rc);
4264                 goto err_free_ena_dev;
4265         }
4266
4267         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4268                                         pci_resource_start(pdev, ENA_REG_BAR),
4269                                         pci_resource_len(pdev, ENA_REG_BAR));
4270         if (!ena_dev->reg_bar) {
4271                 dev_err(&pdev->dev, "Failed to remap regs bar\n");
4272                 rc = -EFAULT;
4273                 goto err_free_region;
4274         }
4275
4276         ena_dev->ena_min_poll_delay_us = ENA_ADMIN_POLL_DELAY_US;
4277
4278         ena_dev->dmadev = &pdev->dev;
4279
4280         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), ENA_MAX_RINGS);
4281         if (!netdev) {
4282                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4283                 rc = -ENOMEM;
4284                 goto err_free_region;
4285         }
4286
4287         SET_NETDEV_DEV(netdev, &pdev->dev);
4288         adapter = netdev_priv(netdev);
4289         adapter->ena_dev = ena_dev;
4290         adapter->netdev = netdev;
4291         adapter->pdev = pdev;
4292         adapter->msg_enable = DEFAULT_MSG_ENABLE;
4293
4294         ena_dev->net_device = netdev;
4295
4296         pci_set_drvdata(pdev, adapter);
4297
4298         rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
4299         if (rc) {
4300                 dev_err(&pdev->dev, "ENA device init failed\n");
4301                 if (rc == -ETIME)
4302                         rc = -EPROBE_DEFER;
4303                 goto err_netdev_destroy;
4304         }
4305
4306         rc = ena_map_llq_mem_bar(pdev, ena_dev, bars);
4307         if (rc) {
4308                 dev_err(&pdev->dev, "ENA llq bar mapping failed\n");
4309                 goto err_device_destroy;
4310         }
4311
4312         calc_queue_ctx.ena_dev = ena_dev;
4313         calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
4314         calc_queue_ctx.pdev = pdev;
4315
4316         /* Initial TX and RX interrupt delay. Assumes 1 usec granularity.
4317          * Updated during device initialization with the real granularity
4318          */
4319         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4320         ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4321         ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4322         max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4323         rc = ena_calc_io_queue_size(&calc_queue_ctx);
4324         if (rc || !max_num_io_queues) {
4325                 rc = -EFAULT;
4326                 goto err_device_destroy;
4327         }
4328
4329         ena_set_conf_feat_params(adapter, &get_feat_ctx);
4330
4331         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4332
4333         adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
4334         adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
4335         adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
4336         adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
4337         adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
4338         adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
4339
4340         adapter->num_io_queues = max_num_io_queues;
4341         adapter->max_num_io_queues = max_num_io_queues;
4342         adapter->last_monitored_tx_qid = 0;
4343
4344         adapter->xdp_first_ring = 0;
4345         adapter->xdp_num_queues = 0;
4346
4347         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4348         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4349                 adapter->disable_meta_caching =
4350                         !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
4351                            BIT(ENA_ADMIN_DISABLE_META_CACHING));
4352
4353         adapter->wd_state = wd_state;
4354
4355         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4356
4357         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4358         if (rc) {
4359                 dev_err(&pdev->dev,
4360                         "Failed to query interrupt moderation feature\n");
4361                 goto err_device_destroy;
4362         }
4363         ena_init_io_rings(adapter,
4364                           0,
4365                           adapter->xdp_num_queues +
4366                           adapter->num_io_queues);
4367
4368         netdev->netdev_ops = &ena_netdev_ops;
4369         netdev->watchdog_timeo = TX_TIMEOUT;
4370         ena_set_ethtool_ops(netdev);
4371
4372         netdev->priv_flags |= IFF_UNICAST_FLT;
4373
4374         u64_stats_init(&adapter->syncp);
4375
4376         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4377         if (rc) {
4378                 dev_err(&pdev->dev,
4379                         "Failed to enable and set the admin interrupts\n");
4380                 goto err_worker_destroy;
4381         }
4382         rc = ena_rss_init_default(adapter);
4383         if (rc && (rc != -EOPNOTSUPP)) {
4384                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4385                 goto err_free_msix;
4386         }
4387
4388         ena_config_debug_area(adapter);
4389
4390         if (!ena_update_hw_stats(adapter))
4391                 adapter->eni_stats_supported = true;
4392         else
4393                 adapter->eni_stats_supported = false;
4394
4395         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4396
4397         netif_carrier_off(netdev);
4398
4399         rc = register_netdev(netdev);
4400         if (rc) {
4401                 dev_err(&pdev->dev, "Cannot register net device\n");
4402                 goto err_rss;
4403         }
4404
4405         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4406
4407         adapter->last_keep_alive_jiffies = jiffies;
4408         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4409         adapter->missing_tx_completion_to = TX_TIMEOUT;
4410         adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4411
4412         ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4413
4414         timer_setup(&adapter->timer_service, ena_timer_service, 0);
4415         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4416
4417         dev_info(&pdev->dev,
4418                  "%s found at mem %lx, mac addr %pM\n",
4419                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4420                  netdev->dev_addr);
4421
4422         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4423
4424         adapters_found++;
4425
4426         return 0;
4427
4428 err_rss:
4429         ena_com_delete_debug_area(ena_dev);
4430         ena_com_rss_destroy(ena_dev);
4431 err_free_msix:
4432         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4433         /* stop submitting admin commands on a device that was reset */
4434         ena_com_set_admin_running_state(ena_dev, false);
4435         ena_free_mgmnt_irq(adapter);
4436         ena_disable_msix(adapter);
4437 err_worker_destroy:
4438         del_timer(&adapter->timer_service);
4439 err_device_destroy:
4440         ena_com_delete_host_info(ena_dev);
4441         ena_com_admin_destroy(ena_dev);
4442 err_netdev_destroy:
4443         free_netdev(netdev);
4444 err_free_region:
4445         ena_release_bars(ena_dev, pdev);
4446 err_free_ena_dev:
4447         vfree(ena_dev);
4448 err_disable_device:
4449         pci_disable_device(pdev);
4450         return rc;
4451 }
4452
4453 /*****************************************************************************/
4454
4455 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4456  * @pdev: PCI device information struct
4457  * @shutdown: Is it a shutdown operation? If false, means it is a removal
4458  *
4459  * __ena_shutoff is a helper routine that does the real work on shutdown and
4460  * removal paths; the difference between those paths is with regards to whether
4461  * dettach or unregister the netdevice.
4462  */
4463 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4464 {
4465         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4466         struct ena_com_dev *ena_dev;
4467         struct net_device *netdev;
4468
4469         ena_dev = adapter->ena_dev;
4470         netdev = adapter->netdev;
4471
4472 #ifdef CONFIG_RFS_ACCEL
4473         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4474                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4475                 netdev->rx_cpu_rmap = NULL;
4476         }
4477 #endif /* CONFIG_RFS_ACCEL */
4478
4479         /* Make sure timer and reset routine won't be called after
4480          * freeing device resources.
4481          */
4482         del_timer_sync(&adapter->timer_service);
4483         cancel_work_sync(&adapter->reset_task);
4484
4485         rtnl_lock(); /* lock released inside the below if-else block */
4486         adapter->reset_reason = ENA_REGS_RESET_SHUTDOWN;
4487         ena_destroy_device(adapter, true);
4488         if (shutdown) {
4489                 netif_device_detach(netdev);
4490                 dev_close(netdev);
4491                 rtnl_unlock();
4492         } else {
4493                 rtnl_unlock();
4494                 unregister_netdev(netdev);
4495                 free_netdev(netdev);
4496         }
4497
4498         ena_com_rss_destroy(ena_dev);
4499
4500         ena_com_delete_debug_area(ena_dev);
4501
4502         ena_com_delete_host_info(ena_dev);
4503
4504         ena_release_bars(ena_dev, pdev);
4505
4506         pci_disable_device(pdev);
4507
4508         vfree(ena_dev);
4509 }
4510
4511 /* ena_remove - Device Removal Routine
4512  * @pdev: PCI device information struct
4513  *
4514  * ena_remove is called by the PCI subsystem to alert the driver
4515  * that it should release a PCI device.
4516  */
4517
4518 static void ena_remove(struct pci_dev *pdev)
4519 {
4520         __ena_shutoff(pdev, false);
4521 }
4522
4523 /* ena_shutdown - Device Shutdown Routine
4524  * @pdev: PCI device information struct
4525  *
4526  * ena_shutdown is called by the PCI subsystem to alert the driver that
4527  * a shutdown/reboot (or kexec) is happening and device must be disabled.
4528  */
4529
4530 static void ena_shutdown(struct pci_dev *pdev)
4531 {
4532         __ena_shutoff(pdev, true);
4533 }
4534
4535 /* ena_suspend - PM suspend callback
4536  * @dev_d: Device information struct
4537  */
4538 static int __maybe_unused ena_suspend(struct device *dev_d)
4539 {
4540         struct pci_dev *pdev = to_pci_dev(dev_d);
4541         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4542
4543         ena_increase_stat(&adapter->dev_stats.suspend, 1, &adapter->syncp);
4544
4545         rtnl_lock();
4546         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4547                 dev_err(&pdev->dev,
4548                         "Ignoring device reset request as the device is being suspended\n");
4549                 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4550         }
4551         ena_destroy_device(adapter, true);
4552         rtnl_unlock();
4553         return 0;
4554 }
4555
4556 /* ena_resume - PM resume callback
4557  * @dev_d: Device information struct
4558  */
4559 static int __maybe_unused ena_resume(struct device *dev_d)
4560 {
4561         struct ena_adapter *adapter = dev_get_drvdata(dev_d);
4562         int rc;
4563
4564         ena_increase_stat(&adapter->dev_stats.resume, 1, &adapter->syncp);
4565
4566         rtnl_lock();
4567         rc = ena_restore_device(adapter);
4568         rtnl_unlock();
4569         return rc;
4570 }
4571
4572 static SIMPLE_DEV_PM_OPS(ena_pm_ops, ena_suspend, ena_resume);
4573
4574 static struct pci_driver ena_pci_driver = {
4575         .name           = DRV_MODULE_NAME,
4576         .id_table       = ena_pci_tbl,
4577         .probe          = ena_probe,
4578         .remove         = ena_remove,
4579         .shutdown       = ena_shutdown,
4580         .driver.pm      = &ena_pm_ops,
4581         .sriov_configure = pci_sriov_configure_simple,
4582 };
4583
4584 static int __init ena_init(void)
4585 {
4586         int ret;
4587
4588         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4589         if (!ena_wq) {
4590                 pr_err("Failed to create workqueue\n");
4591                 return -ENOMEM;
4592         }
4593
4594         ret = pci_register_driver(&ena_pci_driver);
4595         if (ret)
4596                 destroy_workqueue(ena_wq);
4597
4598         return ret;
4599 }
4600
4601 static void __exit ena_cleanup(void)
4602 {
4603         pci_unregister_driver(&ena_pci_driver);
4604
4605         if (ena_wq) {
4606                 destroy_workqueue(ena_wq);
4607                 ena_wq = NULL;
4608         }
4609 }
4610
4611 /******************************************************************************
4612  ******************************** AENQ Handlers *******************************
4613  *****************************************************************************/
4614 /* ena_update_on_link_change:
4615  * Notify the network interface about the change in link status
4616  */
4617 static void ena_update_on_link_change(void *adapter_data,
4618                                       struct ena_admin_aenq_entry *aenq_e)
4619 {
4620         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4621         struct ena_admin_aenq_link_change_desc *aenq_desc =
4622                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
4623         int status = aenq_desc->flags &
4624                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4625
4626         if (status) {
4627                 netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
4628                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4629                 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4630                         netif_carrier_on(adapter->netdev);
4631         } else {
4632                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4633                 netif_carrier_off(adapter->netdev);
4634         }
4635 }
4636
4637 static void ena_keep_alive_wd(void *adapter_data,
4638                               struct ena_admin_aenq_entry *aenq_e)
4639 {
4640         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4641         struct ena_admin_aenq_keep_alive_desc *desc;
4642         u64 rx_drops;
4643         u64 tx_drops;
4644
4645         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4646         adapter->last_keep_alive_jiffies = jiffies;
4647
4648         rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4649         tx_drops = ((u64)desc->tx_drops_high << 32) | desc->tx_drops_low;
4650
4651         u64_stats_update_begin(&adapter->syncp);
4652         /* These stats are accumulated by the device, so the counters indicate
4653          * all drops since last reset.
4654          */
4655         adapter->dev_stats.rx_drops = rx_drops;
4656         adapter->dev_stats.tx_drops = tx_drops;
4657         u64_stats_update_end(&adapter->syncp);
4658 }
4659
4660 static void ena_notification(void *adapter_data,
4661                              struct ena_admin_aenq_entry *aenq_e)
4662 {
4663         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4664         struct ena_admin_ena_hw_hints *hints;
4665
4666         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4667              "Invalid group(%x) expected %x\n",
4668              aenq_e->aenq_common_desc.group,
4669              ENA_ADMIN_NOTIFICATION);
4670
4671         switch (aenq_e->aenq_common_desc.syndrome) {
4672         case ENA_ADMIN_UPDATE_HINTS:
4673                 hints = (struct ena_admin_ena_hw_hints *)
4674                         (&aenq_e->inline_data_w4);
4675                 ena_update_hints(adapter, hints);
4676                 break;
4677         default:
4678                 netif_err(adapter, drv, adapter->netdev,
4679                           "Invalid aenq notification link state %d\n",
4680                           aenq_e->aenq_common_desc.syndrome);
4681         }
4682 }
4683
4684 /* This handler will called for unknown event group or unimplemented handlers*/
4685 static void unimplemented_aenq_handler(void *data,
4686                                        struct ena_admin_aenq_entry *aenq_e)
4687 {
4688         struct ena_adapter *adapter = (struct ena_adapter *)data;
4689
4690         netif_err(adapter, drv, adapter->netdev,
4691                   "Unknown event was received or event with unimplemented handler\n");
4692 }
4693
4694 static struct ena_aenq_handlers aenq_handlers = {
4695         .handlers = {
4696                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4697                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
4698                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4699         },
4700         .unimplemented_handler = unimplemented_aenq_handler
4701 };
4702
4703 module_init(ena_init);
4704 module_exit(ena_cleanup);