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 = NULL;
1292
1293         if (likely(req_id < tx_ring->ring_size)) {
1294                 tx_info = &tx_ring->tx_buffer_info[req_id];
1295                 if (likely(tx_info->skb))
1296                         return 0;
1297         }
1298
1299         return handle_invalid_req_id(tx_ring, req_id, tx_info, false);
1300 }
1301
1302 static int validate_xdp_req_id(struct ena_ring *xdp_ring, u16 req_id)
1303 {
1304         struct ena_tx_buffer *tx_info = NULL;
1305
1306         if (likely(req_id < xdp_ring->ring_size)) {
1307                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1308                 if (likely(tx_info->xdpf))
1309                         return 0;
1310         }
1311
1312         return handle_invalid_req_id(xdp_ring, req_id, tx_info, true);
1313 }
1314
1315 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
1316 {
1317         struct netdev_queue *txq;
1318         bool above_thresh;
1319         u32 tx_bytes = 0;
1320         u32 total_done = 0;
1321         u16 next_to_clean;
1322         u16 req_id;
1323         int tx_pkts = 0;
1324         int rc;
1325
1326         next_to_clean = tx_ring->next_to_clean;
1327         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
1328
1329         while (tx_pkts < budget) {
1330                 struct ena_tx_buffer *tx_info;
1331                 struct sk_buff *skb;
1332
1333                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
1334                                                 &req_id);
1335                 if (rc)
1336                         break;
1337
1338                 rc = validate_tx_req_id(tx_ring, req_id);
1339                 if (rc)
1340                         break;
1341
1342                 tx_info = &tx_ring->tx_buffer_info[req_id];
1343                 skb = tx_info->skb;
1344
1345                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
1346                 prefetch(&skb->end);
1347
1348                 tx_info->skb = NULL;
1349                 tx_info->last_jiffies = 0;
1350
1351                 ena_unmap_tx_buff(tx_ring, tx_info);
1352
1353                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1354                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
1355                           skb);
1356
1357                 tx_bytes += skb->len;
1358                 dev_kfree_skb(skb);
1359                 tx_pkts++;
1360                 total_done += tx_info->tx_descs;
1361
1362                 tx_ring->free_ids[next_to_clean] = req_id;
1363                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1364                                                      tx_ring->ring_size);
1365         }
1366
1367         tx_ring->next_to_clean = next_to_clean;
1368         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
1369         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
1370
1371         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
1372
1373         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1374                   "tx_poll: q %d done. total pkts: %d\n",
1375                   tx_ring->qid, tx_pkts);
1376
1377         /* need to make the rings circular update visible to
1378          * ena_start_xmit() before checking for netif_queue_stopped().
1379          */
1380         smp_mb();
1381
1382         above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1383                                                     ENA_TX_WAKEUP_THRESH);
1384         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
1385                 __netif_tx_lock(txq, smp_processor_id());
1386                 above_thresh =
1387                         ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1388                                                      ENA_TX_WAKEUP_THRESH);
1389                 if (netif_tx_queue_stopped(txq) && above_thresh &&
1390                     test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
1391                         netif_tx_wake_queue(txq);
1392                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
1393                                           &tx_ring->syncp);
1394                 }
1395                 __netif_tx_unlock(txq);
1396         }
1397
1398         return tx_pkts;
1399 }
1400
1401 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, void *first_frag)
1402 {
1403         struct sk_buff *skb;
1404
1405         if (!first_frag)
1406                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1407                                                 rx_ring->rx_copybreak);
1408         else
1409                 skb = build_skb(first_frag, ENA_PAGE_SIZE);
1410
1411         if (unlikely(!skb)) {
1412                 ena_increase_stat(&rx_ring->rx_stats.skb_alloc_fail, 1,
1413                                   &rx_ring->syncp);
1414
1415                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1416                           "Failed to allocate skb. first_frag %s\n",
1417                           first_frag ? "provided" : "not provided");
1418                 return NULL;
1419         }
1420
1421         return skb;
1422 }
1423
1424 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
1425                                   struct ena_com_rx_buf_info *ena_bufs,
1426                                   u32 descs,
1427                                   u16 *next_to_clean)
1428 {
1429         struct ena_rx_buffer *rx_info;
1430         u16 len, req_id, buf = 0;
1431         struct sk_buff *skb;
1432         void *page_addr;
1433         u32 page_offset;
1434         void *data_addr;
1435
1436         len = ena_bufs[buf].len;
1437         req_id = ena_bufs[buf].req_id;
1438
1439         rx_info = &rx_ring->rx_buffer_info[req_id];
1440
1441         if (unlikely(!rx_info->page)) {
1442                 netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
1443                           "Page is NULL\n");
1444                 return NULL;
1445         }
1446
1447         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1448                   "rx_info %p page %p\n",
1449                   rx_info, rx_info->page);
1450
1451         /* save virt address of first buffer */
1452         page_addr = page_address(rx_info->page);
1453         page_offset = rx_info->page_offset;
1454         data_addr = page_addr + page_offset;
1455
1456         prefetch(data_addr);
1457
1458         if (len <= rx_ring->rx_copybreak) {
1459                 skb = ena_alloc_skb(rx_ring, NULL);
1460                 if (unlikely(!skb))
1461                         return NULL;
1462
1463                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1464                           "RX allocated small packet. len %d. data_len %d\n",
1465                           skb->len, skb->data_len);
1466
1467                 /* sync this buffer for CPU use */
1468                 dma_sync_single_for_cpu(rx_ring->dev,
1469                                         dma_unmap_addr(&rx_info->ena_buf, paddr),
1470                                         len,
1471                                         DMA_FROM_DEVICE);
1472                 skb_copy_to_linear_data(skb, data_addr, len);
1473                 dma_sync_single_for_device(rx_ring->dev,
1474                                            dma_unmap_addr(&rx_info->ena_buf, paddr),
1475                                            len,
1476                                            DMA_FROM_DEVICE);
1477
1478                 skb_put(skb, len);
1479                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1480                 rx_ring->free_ids[*next_to_clean] = req_id;
1481                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
1482                                                      rx_ring->ring_size);
1483                 return skb;
1484         }
1485
1486         ena_unmap_rx_buff(rx_ring, rx_info);
1487
1488         skb = ena_alloc_skb(rx_ring, page_addr);
1489         if (unlikely(!skb))
1490                 return NULL;
1491
1492         /* Populate skb's linear part */
1493         skb_reserve(skb, page_offset);
1494         skb_put(skb, len);
1495         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1496
1497         do {
1498                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1499                           "RX skb updated. len %d. data_len %d\n",
1500                           skb->len, skb->data_len);
1501
1502                 rx_info->page = NULL;
1503
1504                 rx_ring->free_ids[*next_to_clean] = req_id;
1505                 *next_to_clean =
1506                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
1507                                              rx_ring->ring_size);
1508                 if (likely(--descs == 0))
1509                         break;
1510
1511                 buf++;
1512                 len = ena_bufs[buf].len;
1513                 req_id = ena_bufs[buf].req_id;
1514
1515                 rx_info = &rx_ring->rx_buffer_info[req_id];
1516
1517                 ena_unmap_rx_buff(rx_ring, rx_info);
1518
1519                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
1520                                 rx_info->page_offset, len, ENA_PAGE_SIZE);
1521
1522         } while (1);
1523
1524         return skb;
1525 }
1526
1527 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
1528  * @adapter: structure containing adapter specific data
1529  * @ena_rx_ctx: received packet context/metadata
1530  * @skb: skb currently being received and modified
1531  */
1532 static void ena_rx_checksum(struct ena_ring *rx_ring,
1533                                    struct ena_com_rx_ctx *ena_rx_ctx,
1534                                    struct sk_buff *skb)
1535 {
1536         /* Rx csum disabled */
1537         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
1538                 skb->ip_summed = CHECKSUM_NONE;
1539                 return;
1540         }
1541
1542         /* For fragmented packets the checksum isn't valid */
1543         if (ena_rx_ctx->frag) {
1544                 skb->ip_summed = CHECKSUM_NONE;
1545                 return;
1546         }
1547
1548         /* if IP and error */
1549         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1550                      (ena_rx_ctx->l3_csum_err))) {
1551                 /* ipv4 checksum error */
1552                 skb->ip_summed = CHECKSUM_NONE;
1553                 ena_increase_stat(&rx_ring->rx_stats.bad_csum, 1,
1554                                   &rx_ring->syncp);
1555                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1556                           "RX IPv4 header checksum error\n");
1557                 return;
1558         }
1559
1560         /* if TCP/UDP */
1561         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1562                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
1563                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
1564                         /* TCP/UDP checksum error */
1565                         ena_increase_stat(&rx_ring->rx_stats.bad_csum, 1,
1566                                           &rx_ring->syncp);
1567                         netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1568                                   "RX L4 checksum error\n");
1569                         skb->ip_summed = CHECKSUM_NONE;
1570                         return;
1571                 }
1572
1573                 if (likely(ena_rx_ctx->l4_csum_checked)) {
1574                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1575                         ena_increase_stat(&rx_ring->rx_stats.csum_good, 1,
1576                                           &rx_ring->syncp);
1577                 } else {
1578                         ena_increase_stat(&rx_ring->rx_stats.csum_unchecked, 1,
1579                                           &rx_ring->syncp);
1580                         skb->ip_summed = CHECKSUM_NONE;
1581                 }
1582         } else {
1583                 skb->ip_summed = CHECKSUM_NONE;
1584                 return;
1585         }
1586
1587 }
1588
1589 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1590                             struct ena_com_rx_ctx *ena_rx_ctx,
1591                             struct sk_buff *skb)
1592 {
1593         enum pkt_hash_types hash_type;
1594
1595         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1596                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1597                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1598
1599                         hash_type = PKT_HASH_TYPE_L4;
1600                 else
1601                         hash_type = PKT_HASH_TYPE_NONE;
1602
1603                 /* Override hash type if the packet is fragmented */
1604                 if (ena_rx_ctx->frag)
1605                         hash_type = PKT_HASH_TYPE_NONE;
1606
1607                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1608         }
1609 }
1610
1611 static int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)
1612 {
1613         struct ena_rx_buffer *rx_info;
1614         int ret;
1615
1616         rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1617         xdp_prepare_buff(xdp, page_address(rx_info->page),
1618                          rx_info->page_offset,
1619                          rx_ring->ena_bufs[0].len, false);
1620         /* If for some reason we received a bigger packet than
1621          * we expect, then we simply drop it
1622          */
1623         if (unlikely(rx_ring->ena_bufs[0].len > ENA_XDP_MAX_MTU))
1624                 return XDP_DROP;
1625
1626         ret = ena_xdp_execute(rx_ring, xdp);
1627
1628         /* The xdp program might expand the headers */
1629         if (ret == XDP_PASS) {
1630                 rx_info->page_offset = xdp->data - xdp->data_hard_start;
1631                 rx_ring->ena_bufs[0].len = xdp->data_end - xdp->data;
1632         }
1633
1634         return ret;
1635 }
1636 /* ena_clean_rx_irq - Cleanup RX irq
1637  * @rx_ring: RX ring to clean
1638  * @napi: napi handler
1639  * @budget: how many packets driver is allowed to clean
1640  *
1641  * Returns the number of cleaned buffers.
1642  */
1643 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1644                             u32 budget)
1645 {
1646         u16 next_to_clean = rx_ring->next_to_clean;
1647         struct ena_com_rx_ctx ena_rx_ctx;
1648         struct ena_rx_buffer *rx_info;
1649         struct ena_adapter *adapter;
1650         u32 res_budget, work_done;
1651         int rx_copybreak_pkt = 0;
1652         int refill_threshold;
1653         struct sk_buff *skb;
1654         int refill_required;
1655         struct xdp_buff xdp;
1656         int xdp_flags = 0;
1657         int total_len = 0;
1658         int xdp_verdict;
1659         int rc = 0;
1660         int i;
1661
1662         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1663                   "%s qid %d\n", __func__, rx_ring->qid);
1664         res_budget = budget;
1665         xdp_init_buff(&xdp, ENA_PAGE_SIZE, &rx_ring->xdp_rxq);
1666
1667         do {
1668                 xdp_verdict = XDP_PASS;
1669                 skb = NULL;
1670                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1671                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1672                 ena_rx_ctx.descs = 0;
1673                 ena_rx_ctx.pkt_offset = 0;
1674                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1675                                     rx_ring->ena_com_io_sq,
1676                                     &ena_rx_ctx);
1677                 if (unlikely(rc))
1678                         goto error;
1679
1680                 if (unlikely(ena_rx_ctx.descs == 0))
1681                         break;
1682
1683                 /* First descriptor might have an offset set by the device */
1684                 rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1685                 rx_info->page_offset += ena_rx_ctx.pkt_offset;
1686
1687                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1688                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1689                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1690                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1691
1692                 if (ena_xdp_present_ring(rx_ring))
1693                         xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1694
1695                 /* allocate skb and fill it */
1696                 if (xdp_verdict == XDP_PASS)
1697                         skb = ena_rx_skb(rx_ring,
1698                                          rx_ring->ena_bufs,
1699                                          ena_rx_ctx.descs,
1700                                          &next_to_clean);
1701
1702                 if (unlikely(!skb)) {
1703                         for (i = 0; i < ena_rx_ctx.descs; i++) {
1704                                 int req_id = rx_ring->ena_bufs[i].req_id;
1705
1706                                 rx_ring->free_ids[next_to_clean] = req_id;
1707                                 next_to_clean =
1708                                         ENA_RX_RING_IDX_NEXT(next_to_clean,
1709                                                              rx_ring->ring_size);
1710
1711                                 /* Packets was passed for transmission, unmap it
1712                                  * from RX side.
1713                                  */
1714                                 if (xdp_verdict == XDP_TX || xdp_verdict == XDP_REDIRECT) {
1715                                         ena_unmap_rx_buff(rx_ring,
1716                                                           &rx_ring->rx_buffer_info[req_id]);
1717                                         rx_ring->rx_buffer_info[req_id].page = NULL;
1718                                 }
1719                         }
1720                         if (xdp_verdict != XDP_PASS) {
1721                                 xdp_flags |= xdp_verdict;
1722                                 res_budget--;
1723                                 continue;
1724                         }
1725                         break;
1726                 }
1727
1728                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1729
1730                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1731
1732                 skb_record_rx_queue(skb, rx_ring->qid);
1733
1734                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak)
1735                         rx_copybreak_pkt++;
1736
1737                 total_len += skb->len;
1738
1739                 napi_gro_receive(napi, skb);
1740
1741                 res_budget--;
1742         } while (likely(res_budget));
1743
1744         work_done = budget - res_budget;
1745         rx_ring->per_napi_packets += work_done;
1746         u64_stats_update_begin(&rx_ring->syncp);
1747         rx_ring->rx_stats.bytes += total_len;
1748         rx_ring->rx_stats.cnt += work_done;
1749         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1750         u64_stats_update_end(&rx_ring->syncp);
1751
1752         rx_ring->next_to_clean = next_to_clean;
1753
1754         refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
1755         refill_threshold =
1756                 min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1757                       ENA_RX_REFILL_THRESH_PACKET);
1758
1759         /* Optimization, try to batch new rx buffers */
1760         if (refill_required > refill_threshold) {
1761                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1762                 ena_refill_rx_bufs(rx_ring, refill_required);
1763         }
1764
1765         if (xdp_flags & XDP_REDIRECT)
1766                 xdp_do_flush_map();
1767
1768         return work_done;
1769
1770 error:
1771         adapter = netdev_priv(rx_ring->netdev);
1772
1773         if (rc == -ENOSPC) {
1774                 ena_increase_stat(&rx_ring->rx_stats.bad_desc_num, 1,
1775                                   &rx_ring->syncp);
1776                 adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1777         } else {
1778                 ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1,
1779                                   &rx_ring->syncp);
1780                 adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
1781         }
1782
1783         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1784
1785         return 0;
1786 }
1787
1788 static void ena_dim_work(struct work_struct *w)
1789 {
1790         struct dim *dim = container_of(w, struct dim, work);
1791         struct dim_cq_moder cur_moder =
1792                 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1793         struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1794
1795         ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1796         dim->state = DIM_START_MEASURE;
1797 }
1798
1799 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1800 {
1801         struct dim_sample dim_sample;
1802         struct ena_ring *rx_ring = ena_napi->rx_ring;
1803
1804         if (!rx_ring->per_napi_packets)
1805                 return;
1806
1807         rx_ring->non_empty_napi_events++;
1808
1809         dim_update_sample(rx_ring->non_empty_napi_events,
1810                           rx_ring->rx_stats.cnt,
1811                           rx_ring->rx_stats.bytes,
1812                           &dim_sample);
1813
1814         net_dim(&ena_napi->dim, dim_sample);
1815
1816         rx_ring->per_napi_packets = 0;
1817 }
1818
1819 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1820                                         struct ena_ring *rx_ring)
1821 {
1822         struct ena_eth_io_intr_reg intr_reg;
1823         u32 rx_interval = 0;
1824         /* Rx ring can be NULL when for XDP tx queues which don't have an
1825          * accompanying rx_ring pair.
1826          */
1827         if (rx_ring)
1828                 rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1829                         rx_ring->smoothed_interval :
1830                         ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1831
1832         /* Update intr register: rx intr delay,
1833          * tx intr delay and interrupt unmask
1834          */
1835         ena_com_update_intr_reg(&intr_reg,
1836                                 rx_interval,
1837                                 tx_ring->smoothed_interval,
1838                                 true);
1839
1840         ena_increase_stat(&tx_ring->tx_stats.unmask_interrupt, 1,
1841                           &tx_ring->syncp);
1842
1843         /* It is a shared MSI-X.
1844          * Tx and Rx CQ have pointer to it.
1845          * So we use one of them to reach the intr reg
1846          * The Tx ring is used because the rx_ring is NULL for XDP queues
1847          */
1848         ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1849 }
1850
1851 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1852                                              struct ena_ring *rx_ring)
1853 {
1854         int cpu = get_cpu();
1855         int numa_node;
1856
1857         /* Check only one ring since the 2 rings are running on the same cpu */
1858         if (likely(tx_ring->cpu == cpu))
1859                 goto out;
1860
1861         numa_node = cpu_to_node(cpu);
1862         put_cpu();
1863
1864         if (numa_node != NUMA_NO_NODE) {
1865                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1866                 if (rx_ring)
1867                         ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1868                                                  numa_node);
1869         }
1870
1871         tx_ring->cpu = cpu;
1872         if (rx_ring)
1873                 rx_ring->cpu = cpu;
1874
1875         return;
1876 out:
1877         put_cpu();
1878 }
1879
1880 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1881 {
1882         u32 total_done = 0;
1883         u16 next_to_clean;
1884         u32 tx_bytes = 0;
1885         int tx_pkts = 0;
1886         u16 req_id;
1887         int rc;
1888
1889         if (unlikely(!xdp_ring))
1890                 return 0;
1891         next_to_clean = xdp_ring->next_to_clean;
1892
1893         while (tx_pkts < budget) {
1894                 struct ena_tx_buffer *tx_info;
1895                 struct xdp_frame *xdpf;
1896
1897                 rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1898                                                 &req_id);
1899                 if (rc)
1900                         break;
1901
1902                 rc = validate_xdp_req_id(xdp_ring, req_id);
1903                 if (rc)
1904                         break;
1905
1906                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1907                 xdpf = tx_info->xdpf;
1908
1909                 tx_info->xdpf = NULL;
1910                 tx_info->last_jiffies = 0;
1911                 ena_unmap_tx_buff(xdp_ring, tx_info);
1912
1913                 netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1914                           "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1915                           xdpf);
1916
1917                 tx_bytes += xdpf->len;
1918                 tx_pkts++;
1919                 total_done += tx_info->tx_descs;
1920
1921                 xdp_return_frame(xdpf);
1922                 xdp_ring->free_ids[next_to_clean] = req_id;
1923                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1924                                                      xdp_ring->ring_size);
1925         }
1926
1927         xdp_ring->next_to_clean = next_to_clean;
1928         ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
1929         ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
1930
1931         netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1932                   "tx_poll: q %d done. total pkts: %d\n",
1933                   xdp_ring->qid, tx_pkts);
1934
1935         return tx_pkts;
1936 }
1937
1938 static int ena_io_poll(struct napi_struct *napi, int budget)
1939 {
1940         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1941         struct ena_ring *tx_ring, *rx_ring;
1942         int tx_work_done;
1943         int rx_work_done = 0;
1944         int tx_budget;
1945         int napi_comp_call = 0;
1946         int ret;
1947
1948         tx_ring = ena_napi->tx_ring;
1949         rx_ring = ena_napi->rx_ring;
1950
1951         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1952
1953         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1954             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1955                 napi_complete_done(napi, 0);
1956                 return 0;
1957         }
1958
1959         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1960         /* On netpoll the budget is zero and the handler should only clean the
1961          * tx completions.
1962          */
1963         if (likely(budget))
1964                 rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1965
1966         /* If the device is about to reset or down, avoid unmask
1967          * the interrupt and return 0 so NAPI won't reschedule
1968          */
1969         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1970                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1971                 napi_complete_done(napi, 0);
1972                 ret = 0;
1973
1974         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1975                 napi_comp_call = 1;
1976
1977                 /* Update numa and unmask the interrupt only when schedule
1978                  * from the interrupt context (vs from sk_busy_loop)
1979                  */
1980                 if (napi_complete_done(napi, rx_work_done) &&
1981                     READ_ONCE(ena_napi->interrupts_masked)) {
1982                         smp_rmb(); /* make sure interrupts_masked is read */
1983                         WRITE_ONCE(ena_napi->interrupts_masked, false);
1984                         /* We apply adaptive moderation on Rx path only.
1985                          * Tx uses static interrupt moderation.
1986                          */
1987                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1988                                 ena_adjust_adaptive_rx_intr_moderation(ena_napi);
1989
1990                         ena_unmask_interrupt(tx_ring, rx_ring);
1991                 }
1992
1993                 ena_update_ring_numa_node(tx_ring, rx_ring);
1994
1995                 ret = rx_work_done;
1996         } else {
1997                 ret = budget;
1998         }
1999
2000         u64_stats_update_begin(&tx_ring->syncp);
2001         tx_ring->tx_stats.napi_comp += napi_comp_call;
2002         tx_ring->tx_stats.tx_poll++;
2003         u64_stats_update_end(&tx_ring->syncp);
2004
2005         tx_ring->tx_stats.last_napi_jiffies = jiffies;
2006
2007         return ret;
2008 }
2009
2010 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
2011 {
2012         struct ena_adapter *adapter = (struct ena_adapter *)data;
2013
2014         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
2015
2016         /* Don't call the aenq handler before probe is done */
2017         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
2018                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
2019
2020         return IRQ_HANDLED;
2021 }
2022
2023 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
2024  * @irq: interrupt number
2025  * @data: pointer to a network interface private napi device structure
2026  */
2027 static irqreturn_t ena_intr_msix_io(int irq, void *data)
2028 {
2029         struct ena_napi *ena_napi = data;
2030
2031         /* Used to check HW health */
2032         WRITE_ONCE(ena_napi->first_interrupt, true);
2033
2034         WRITE_ONCE(ena_napi->interrupts_masked, true);
2035         smp_wmb(); /* write interrupts_masked before calling napi */
2036
2037         napi_schedule_irqoff(&ena_napi->napi);
2038
2039         return IRQ_HANDLED;
2040 }
2041
2042 /* Reserve a single MSI-X vector for management (admin + aenq).
2043  * plus reserve one vector for each potential io queue.
2044  * the number of potential io queues is the minimum of what the device
2045  * supports and the number of vCPUs.
2046  */
2047 static int ena_enable_msix(struct ena_adapter *adapter)
2048 {
2049         int msix_vecs, irq_cnt;
2050
2051         if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2052                 netif_err(adapter, probe, adapter->netdev,
2053                           "Error, MSI-X is already enabled\n");
2054                 return -EPERM;
2055         }
2056
2057         /* Reserved the max msix vectors we might need */
2058         msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
2059         netif_dbg(adapter, probe, adapter->netdev,
2060                   "Trying to enable MSI-X, vectors %d\n", msix_vecs);
2061
2062         irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
2063                                         msix_vecs, PCI_IRQ_MSIX);
2064
2065         if (irq_cnt < 0) {
2066                 netif_err(adapter, probe, adapter->netdev,
2067                           "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
2068                 return -ENOSPC;
2069         }
2070
2071         if (irq_cnt != msix_vecs) {
2072                 netif_notice(adapter, probe, adapter->netdev,
2073                              "Enable only %d MSI-X (out of %d), reduce the number of queues\n",
2074                              irq_cnt, msix_vecs);
2075                 adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
2076         }
2077
2078         if (ena_init_rx_cpu_rmap(adapter))
2079                 netif_warn(adapter, probe, adapter->netdev,
2080                            "Failed to map IRQs to CPUs\n");
2081
2082         adapter->msix_vecs = irq_cnt;
2083         set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2084
2085         return 0;
2086 }
2087
2088 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2089 {
2090         u32 cpu;
2091
2092         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2093                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2094                  pci_name(adapter->pdev));
2095         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2096                 ena_intr_msix_mgmnt;
2097         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2098         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2099                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2100         cpu = cpumask_first(cpu_online_mask);
2101         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2102         cpumask_set_cpu(cpu,
2103                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2104 }
2105
2106 static void ena_setup_io_intr(struct ena_adapter *adapter)
2107 {
2108         struct net_device *netdev;
2109         int irq_idx, i, cpu;
2110         int io_queue_count;
2111
2112         netdev = adapter->netdev;
2113         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2114
2115         for (i = 0; i < io_queue_count; i++) {
2116                 irq_idx = ENA_IO_IRQ_IDX(i);
2117                 cpu = i % num_online_cpus();
2118
2119                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2120                          "%s-Tx-Rx-%d", netdev->name, i);
2121                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2122                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2123                 adapter->irq_tbl[irq_idx].vector =
2124                         pci_irq_vector(adapter->pdev, irq_idx);
2125                 adapter->irq_tbl[irq_idx].cpu = cpu;
2126
2127                 cpumask_set_cpu(cpu,
2128                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
2129         }
2130 }
2131
2132 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2133 {
2134         unsigned long flags = 0;
2135         struct ena_irq *irq;
2136         int rc;
2137
2138         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2139         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2140                          irq->data);
2141         if (rc) {
2142                 netif_err(adapter, probe, adapter->netdev,
2143                           "Failed to request admin irq\n");
2144                 return rc;
2145         }
2146
2147         netif_dbg(adapter, probe, adapter->netdev,
2148                   "Set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2149                   irq->affinity_hint_mask.bits[0], irq->vector);
2150
2151         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2152
2153         return rc;
2154 }
2155
2156 static int ena_request_io_irq(struct ena_adapter *adapter)
2157 {
2158         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2159         unsigned long flags = 0;
2160         struct ena_irq *irq;
2161         int rc = 0, i, k;
2162
2163         if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2164                 netif_err(adapter, ifup, adapter->netdev,
2165                           "Failed to request I/O IRQ: MSI-X is not enabled\n");
2166                 return -EINVAL;
2167         }
2168
2169         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2170                 irq = &adapter->irq_tbl[i];
2171                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2172                                  irq->data);
2173                 if (rc) {
2174                         netif_err(adapter, ifup, adapter->netdev,
2175                                   "Failed to request I/O IRQ. index %d rc %d\n",
2176                                    i, rc);
2177                         goto err;
2178                 }
2179
2180                 netif_dbg(adapter, ifup, adapter->netdev,
2181                           "Set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2182                           i, irq->affinity_hint_mask.bits[0], irq->vector);
2183
2184                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2185         }
2186
2187         return rc;
2188
2189 err:
2190         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2191                 irq = &adapter->irq_tbl[k];
2192                 free_irq(irq->vector, irq->data);
2193         }
2194
2195         return rc;
2196 }
2197
2198 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2199 {
2200         struct ena_irq *irq;
2201
2202         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2203         synchronize_irq(irq->vector);
2204         irq_set_affinity_hint(irq->vector, NULL);
2205         free_irq(irq->vector, irq->data);
2206 }
2207
2208 static void ena_free_io_irq(struct ena_adapter *adapter)
2209 {
2210         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2211         struct ena_irq *irq;
2212         int i;
2213
2214 #ifdef CONFIG_RFS_ACCEL
2215         if (adapter->msix_vecs >= 1) {
2216                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2217                 adapter->netdev->rx_cpu_rmap = NULL;
2218         }
2219 #endif /* CONFIG_RFS_ACCEL */
2220
2221         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2222                 irq = &adapter->irq_tbl[i];
2223                 irq_set_affinity_hint(irq->vector, NULL);
2224                 free_irq(irq->vector, irq->data);
2225         }
2226 }
2227
2228 static void ena_disable_msix(struct ena_adapter *adapter)
2229 {
2230         if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2231                 pci_free_irq_vectors(adapter->pdev);
2232 }
2233
2234 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2235 {
2236         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2237         int i;
2238
2239         if (!netif_running(adapter->netdev))
2240                 return;
2241
2242         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2243                 synchronize_irq(adapter->irq_tbl[i].vector);
2244 }
2245
2246 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2247                                   int first_index,
2248                                   int count)
2249 {
2250         int i;
2251
2252         for (i = first_index; i < first_index + count; i++) {
2253                 netif_napi_del(&adapter->ena_napi[i].napi);
2254
2255                 WARN_ON(!ENA_IS_XDP_INDEX(adapter, i) &&
2256                         adapter->ena_napi[i].xdp_ring);
2257         }
2258 }
2259
2260 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2261                                    int first_index, int count)
2262 {
2263         int i;
2264
2265         for (i = first_index; i < first_index + count; i++) {
2266                 struct ena_napi *napi = &adapter->ena_napi[i];
2267
2268                 netif_napi_add(adapter->netdev,
2269                                &napi->napi,
2270                                ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll,
2271                                ENA_NAPI_BUDGET);
2272
2273                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
2274                         napi->rx_ring = &adapter->rx_ring[i];
2275                         napi->tx_ring = &adapter->tx_ring[i];
2276                 } else {
2277                         napi->xdp_ring = &adapter->tx_ring[i];
2278                 }
2279                 napi->qid = i;
2280         }
2281 }
2282
2283 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2284                                       int first_index,
2285                                       int count)
2286 {
2287         int i;
2288
2289         for (i = first_index; i < first_index + count; i++)
2290                 napi_disable(&adapter->ena_napi[i].napi);
2291 }
2292
2293 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2294                                      int first_index,
2295                                      int count)
2296 {
2297         int i;
2298
2299         for (i = first_index; i < first_index + count; i++)
2300                 napi_enable(&adapter->ena_napi[i].napi);
2301 }
2302
2303 /* Configure the Rx forwarding */
2304 static int ena_rss_configure(struct ena_adapter *adapter)
2305 {
2306         struct ena_com_dev *ena_dev = adapter->ena_dev;
2307         int rc;
2308
2309         /* In case the RSS table wasn't initialized by probe */
2310         if (!ena_dev->rss.tbl_log_size) {
2311                 rc = ena_rss_init_default(adapter);
2312                 if (rc && (rc != -EOPNOTSUPP)) {
2313                         netif_err(adapter, ifup, adapter->netdev,
2314                                   "Failed to init RSS rc: %d\n", rc);
2315                         return rc;
2316                 }
2317         }
2318
2319         /* Set indirect table */
2320         rc = ena_com_indirect_table_set(ena_dev);
2321         if (unlikely(rc && rc != -EOPNOTSUPP))
2322                 return rc;
2323
2324         /* Configure hash function (if supported) */
2325         rc = ena_com_set_hash_function(ena_dev);
2326         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2327                 return rc;
2328
2329         /* Configure hash inputs (if supported) */
2330         rc = ena_com_set_hash_ctrl(ena_dev);
2331         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2332                 return rc;
2333
2334         return 0;
2335 }
2336
2337 static int ena_up_complete(struct ena_adapter *adapter)
2338 {
2339         int rc;
2340
2341         rc = ena_rss_configure(adapter);
2342         if (rc)
2343                 return rc;
2344
2345         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2346
2347         ena_refill_all_rx_bufs(adapter);
2348
2349         /* enable transmits */
2350         netif_tx_start_all_queues(adapter->netdev);
2351
2352         ena_napi_enable_in_range(adapter,
2353                                  0,
2354                                  adapter->xdp_num_queues + adapter->num_io_queues);
2355
2356         return 0;
2357 }
2358
2359 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2360 {
2361         struct ena_com_create_io_ctx ctx;
2362         struct ena_com_dev *ena_dev;
2363         struct ena_ring *tx_ring;
2364         u32 msix_vector;
2365         u16 ena_qid;
2366         int rc;
2367
2368         ena_dev = adapter->ena_dev;
2369
2370         tx_ring = &adapter->tx_ring[qid];
2371         msix_vector = ENA_IO_IRQ_IDX(qid);
2372         ena_qid = ENA_IO_TXQ_IDX(qid);
2373
2374         memset(&ctx, 0x0, sizeof(ctx));
2375
2376         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2377         ctx.qid = ena_qid;
2378         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2379         ctx.msix_vector = msix_vector;
2380         ctx.queue_size = tx_ring->ring_size;
2381         ctx.numa_node = cpu_to_node(tx_ring->cpu);
2382
2383         rc = ena_com_create_io_queue(ena_dev, &ctx);
2384         if (rc) {
2385                 netif_err(adapter, ifup, adapter->netdev,
2386                           "Failed to create I/O TX queue num %d rc: %d\n",
2387                           qid, rc);
2388                 return rc;
2389         }
2390
2391         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2392                                      &tx_ring->ena_com_io_sq,
2393                                      &tx_ring->ena_com_io_cq);
2394         if (rc) {
2395                 netif_err(adapter, ifup, adapter->netdev,
2396                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2397                           qid, rc);
2398                 ena_com_destroy_io_queue(ena_dev, ena_qid);
2399                 return rc;
2400         }
2401
2402         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2403         return rc;
2404 }
2405
2406 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2407                                             int first_index, int count)
2408 {
2409         struct ena_com_dev *ena_dev = adapter->ena_dev;
2410         int rc, i;
2411
2412         for (i = first_index; i < first_index + count; i++) {
2413                 rc = ena_create_io_tx_queue(adapter, i);
2414                 if (rc)
2415                         goto create_err;
2416         }
2417
2418         return 0;
2419
2420 create_err:
2421         while (i-- > first_index)
2422                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2423
2424         return rc;
2425 }
2426
2427 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2428 {
2429         struct ena_com_dev *ena_dev;
2430         struct ena_com_create_io_ctx ctx;
2431         struct ena_ring *rx_ring;
2432         u32 msix_vector;
2433         u16 ena_qid;
2434         int rc;
2435
2436         ena_dev = adapter->ena_dev;
2437
2438         rx_ring = &adapter->rx_ring[qid];
2439         msix_vector = ENA_IO_IRQ_IDX(qid);
2440         ena_qid = ENA_IO_RXQ_IDX(qid);
2441
2442         memset(&ctx, 0x0, sizeof(ctx));
2443
2444         ctx.qid = ena_qid;
2445         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2446         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2447         ctx.msix_vector = msix_vector;
2448         ctx.queue_size = rx_ring->ring_size;
2449         ctx.numa_node = cpu_to_node(rx_ring->cpu);
2450
2451         rc = ena_com_create_io_queue(ena_dev, &ctx);
2452         if (rc) {
2453                 netif_err(adapter, ifup, adapter->netdev,
2454                           "Failed to create I/O RX queue num %d rc: %d\n",
2455                           qid, rc);
2456                 return rc;
2457         }
2458
2459         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2460                                      &rx_ring->ena_com_io_sq,
2461                                      &rx_ring->ena_com_io_cq);
2462         if (rc) {
2463                 netif_err(adapter, ifup, adapter->netdev,
2464                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2465                           qid, rc);
2466                 goto err;
2467         }
2468
2469         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2470
2471         return rc;
2472 err:
2473         ena_com_destroy_io_queue(ena_dev, ena_qid);
2474         return rc;
2475 }
2476
2477 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2478 {
2479         struct ena_com_dev *ena_dev = adapter->ena_dev;
2480         int rc, i;
2481
2482         for (i = 0; i < adapter->num_io_queues; i++) {
2483                 rc = ena_create_io_rx_queue(adapter, i);
2484                 if (rc)
2485                         goto create_err;
2486                 INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2487         }
2488
2489         return 0;
2490
2491 create_err:
2492         while (i--) {
2493                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
2494                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2495         }
2496
2497         return rc;
2498 }
2499
2500 static void set_io_rings_size(struct ena_adapter *adapter,
2501                               int new_tx_size,
2502                               int new_rx_size)
2503 {
2504         int i;
2505
2506         for (i = 0; i < adapter->num_io_queues; i++) {
2507                 adapter->tx_ring[i].ring_size = new_tx_size;
2508                 adapter->rx_ring[i].ring_size = new_rx_size;
2509         }
2510 }
2511
2512 /* This function allows queue allocation to backoff when the system is
2513  * low on memory. If there is not enough memory to allocate io queues
2514  * the driver will try to allocate smaller queues.
2515  *
2516  * The backoff algorithm is as follows:
2517  *  1. Try to allocate TX and RX and if successful.
2518  *  1.1. return success
2519  *
2520  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2521  *
2522  *  3. If TX or RX is smaller than 256
2523  *  3.1. return failure.
2524  *  4. else
2525  *  4.1. go back to 1.
2526  */
2527 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2528 {
2529         int rc, cur_rx_ring_size, cur_tx_ring_size;
2530         int new_rx_ring_size, new_tx_ring_size;
2531
2532         /* current queue sizes might be set to smaller than the requested
2533          * ones due to past queue allocation failures.
2534          */
2535         set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2536                           adapter->requested_rx_ring_size);
2537
2538         while (1) {
2539                 if (ena_xdp_present(adapter)) {
2540                         rc = ena_setup_and_create_all_xdp_queues(adapter);
2541
2542                         if (rc)
2543                                 goto err_setup_tx;
2544                 }
2545                 rc = ena_setup_tx_resources_in_range(adapter,
2546                                                      0,
2547                                                      adapter->num_io_queues);
2548                 if (rc)
2549                         goto err_setup_tx;
2550
2551                 rc = ena_create_io_tx_queues_in_range(adapter,
2552                                                       0,
2553                                                       adapter->num_io_queues);
2554                 if (rc)
2555                         goto err_create_tx_queues;
2556
2557                 rc = ena_setup_all_rx_resources(adapter);
2558                 if (rc)
2559                         goto err_setup_rx;
2560
2561                 rc = ena_create_all_io_rx_queues(adapter);
2562                 if (rc)
2563                         goto err_create_rx_queues;
2564
2565                 return 0;
2566
2567 err_create_rx_queues:
2568                 ena_free_all_io_rx_resources(adapter);
2569 err_setup_rx:
2570                 ena_destroy_all_tx_queues(adapter);
2571 err_create_tx_queues:
2572                 ena_free_all_io_tx_resources(adapter);
2573 err_setup_tx:
2574                 if (rc != -ENOMEM) {
2575                         netif_err(adapter, ifup, adapter->netdev,
2576                                   "Queue creation failed with error code %d\n",
2577                                   rc);
2578                         return rc;
2579                 }
2580
2581                 cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2582                 cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2583
2584                 netif_err(adapter, ifup, adapter->netdev,
2585                           "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2586                           cur_tx_ring_size, cur_rx_ring_size);
2587
2588                 new_tx_ring_size = cur_tx_ring_size;
2589                 new_rx_ring_size = cur_rx_ring_size;
2590
2591                 /* Decrease the size of the larger queue, or
2592                  * decrease both if they are the same size.
2593                  */
2594                 if (cur_rx_ring_size <= cur_tx_ring_size)
2595                         new_tx_ring_size = cur_tx_ring_size / 2;
2596                 if (cur_rx_ring_size >= cur_tx_ring_size)
2597                         new_rx_ring_size = cur_rx_ring_size / 2;
2598
2599                 if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2600                     new_rx_ring_size < ENA_MIN_RING_SIZE) {
2601                         netif_err(adapter, ifup, adapter->netdev,
2602                                   "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2603                                   ENA_MIN_RING_SIZE);
2604                         return rc;
2605                 }
2606
2607                 netif_err(adapter, ifup, adapter->netdev,
2608                           "Retrying queue creation with sizes TX=%d, RX=%d\n",
2609                           new_tx_ring_size,
2610                           new_rx_ring_size);
2611
2612                 set_io_rings_size(adapter, new_tx_ring_size,
2613                                   new_rx_ring_size);
2614         }
2615 }
2616
2617 static int ena_up(struct ena_adapter *adapter)
2618 {
2619         int io_queue_count, rc, i;
2620
2621         netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
2622
2623         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2624         ena_setup_io_intr(adapter);
2625
2626         /* napi poll functions should be initialized before running
2627          * request_irq(), to handle a rare condition where there is a pending
2628          * interrupt, causing the ISR to fire immediately while the poll
2629          * function wasn't set yet, causing a null dereference
2630          */
2631         ena_init_napi_in_range(adapter, 0, io_queue_count);
2632
2633         rc = ena_request_io_irq(adapter);
2634         if (rc)
2635                 goto err_req_irq;
2636
2637         rc = create_queues_with_size_backoff(adapter);
2638         if (rc)
2639                 goto err_create_queues_with_backoff;
2640
2641         rc = ena_up_complete(adapter);
2642         if (rc)
2643                 goto err_up;
2644
2645         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2646                 netif_carrier_on(adapter->netdev);
2647
2648         ena_increase_stat(&adapter->dev_stats.interface_up, 1,
2649                           &adapter->syncp);
2650
2651         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2652
2653         /* Enable completion queues interrupt */
2654         for (i = 0; i < adapter->num_io_queues; i++)
2655                 ena_unmask_interrupt(&adapter->tx_ring[i],
2656                                      &adapter->rx_ring[i]);
2657
2658         /* schedule napi in case we had pending packets
2659          * from the last time we disable napi
2660          */
2661         for (i = 0; i < io_queue_count; i++)
2662                 napi_schedule(&adapter->ena_napi[i].napi);
2663
2664         return rc;
2665
2666 err_up:
2667         ena_destroy_all_tx_queues(adapter);
2668         ena_free_all_io_tx_resources(adapter);
2669         ena_destroy_all_rx_queues(adapter);
2670         ena_free_all_io_rx_resources(adapter);
2671 err_create_queues_with_backoff:
2672         ena_free_io_irq(adapter);
2673 err_req_irq:
2674         ena_del_napi_in_range(adapter, 0, io_queue_count);
2675
2676         return rc;
2677 }
2678
2679 static void ena_down(struct ena_adapter *adapter)
2680 {
2681         int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2682
2683         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2684
2685         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2686
2687         ena_increase_stat(&adapter->dev_stats.interface_down, 1,
2688                           &adapter->syncp);
2689
2690         netif_carrier_off(adapter->netdev);
2691         netif_tx_disable(adapter->netdev);
2692
2693         /* After this point the napi handler won't enable the tx queue */
2694         ena_napi_disable_in_range(adapter, 0, io_queue_count);
2695
2696         /* After destroy the queue there won't be any new interrupts */
2697
2698         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2699                 int rc;
2700
2701                 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2702                 if (rc)
2703                         netif_err(adapter, ifdown, adapter->netdev,
2704                                   "Device reset failed\n");
2705                 /* stop submitting admin commands on a device that was reset */
2706                 ena_com_set_admin_running_state(adapter->ena_dev, false);
2707         }
2708
2709         ena_destroy_all_io_queues(adapter);
2710
2711         ena_disable_io_intr_sync(adapter);
2712         ena_free_io_irq(adapter);
2713         ena_del_napi_in_range(adapter, 0, io_queue_count);
2714
2715         ena_free_all_tx_bufs(adapter);
2716         ena_free_all_rx_bufs(adapter);
2717         ena_free_all_io_tx_resources(adapter);
2718         ena_free_all_io_rx_resources(adapter);
2719 }
2720
2721 /* ena_open - Called when a network interface is made active
2722  * @netdev: network interface device structure
2723  *
2724  * Returns 0 on success, negative value on failure
2725  *
2726  * The open entry point is called when a network interface is made
2727  * active by the system (IFF_UP).  At this point all resources needed
2728  * for transmit and receive operations are allocated, the interrupt
2729  * handler is registered with the OS, the watchdog timer is started,
2730  * and the stack is notified that the interface is ready.
2731  */
2732 static int ena_open(struct net_device *netdev)
2733 {
2734         struct ena_adapter *adapter = netdev_priv(netdev);
2735         int rc;
2736
2737         /* Notify the stack of the actual queue counts. */
2738         rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2739         if (rc) {
2740                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2741                 return rc;
2742         }
2743
2744         rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2745         if (rc) {
2746                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2747                 return rc;
2748         }
2749
2750         rc = ena_up(adapter);
2751         if (rc)
2752                 return rc;
2753
2754         return rc;
2755 }
2756
2757 /* ena_close - Disables a network interface
2758  * @netdev: network interface device structure
2759  *
2760  * Returns 0, this is not allowed to fail
2761  *
2762  * The close entry point is called when an interface is de-activated
2763  * by the OS.  The hardware is still under the drivers control, but
2764  * needs to be disabled.  A global MAC reset is issued to stop the
2765  * hardware, and all transmit and receive resources are freed.
2766  */
2767 static int ena_close(struct net_device *netdev)
2768 {
2769         struct ena_adapter *adapter = netdev_priv(netdev);
2770
2771         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2772
2773         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2774                 return 0;
2775
2776         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2777                 ena_down(adapter);
2778
2779         /* Check for device status and issue reset if needed*/
2780         check_for_admin_com_state(adapter);
2781         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2782                 netif_err(adapter, ifdown, adapter->netdev,
2783                           "Destroy failure, restarting device\n");
2784                 ena_dump_stats_to_dmesg(adapter);
2785                 /* rtnl lock already obtained in dev_ioctl() layer */
2786                 ena_destroy_device(adapter, false);
2787                 ena_restore_device(adapter);
2788         }
2789
2790         return 0;
2791 }
2792
2793 int ena_update_queue_sizes(struct ena_adapter *adapter,
2794                            u32 new_tx_size,
2795                            u32 new_rx_size)
2796 {
2797         bool dev_was_up;
2798
2799         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2800         ena_close(adapter->netdev);
2801         adapter->requested_tx_ring_size = new_tx_size;
2802         adapter->requested_rx_ring_size = new_rx_size;
2803         ena_init_io_rings(adapter,
2804                           0,
2805                           adapter->xdp_num_queues +
2806                           adapter->num_io_queues);
2807         return dev_was_up ? ena_up(adapter) : 0;
2808 }
2809
2810 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2811 {
2812         struct ena_com_dev *ena_dev = adapter->ena_dev;
2813         int prev_channel_count;
2814         bool dev_was_up;
2815
2816         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2817         ena_close(adapter->netdev);
2818         prev_channel_count = adapter->num_io_queues;
2819         adapter->num_io_queues = new_channel_count;
2820         if (ena_xdp_present(adapter) &&
2821             ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2822                 adapter->xdp_first_ring = new_channel_count;
2823                 adapter->xdp_num_queues = new_channel_count;
2824                 if (prev_channel_count > new_channel_count)
2825                         ena_xdp_exchange_program_rx_in_range(adapter,
2826                                                              NULL,
2827                                                              new_channel_count,
2828                                                              prev_channel_count);
2829                 else
2830                         ena_xdp_exchange_program_rx_in_range(adapter,
2831                                                              adapter->xdp_bpf_prog,
2832                                                              prev_channel_count,
2833                                                              new_channel_count);
2834         }
2835
2836         /* We need to destroy the rss table so that the indirection
2837          * table will be reinitialized by ena_up()
2838          */
2839         ena_com_rss_destroy(ena_dev);
2840         ena_init_io_rings(adapter,
2841                           0,
2842                           adapter->xdp_num_queues +
2843                           adapter->num_io_queues);
2844         return dev_was_up ? ena_open(adapter->netdev) : 0;
2845 }
2846
2847 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx,
2848                         struct sk_buff *skb,
2849                         bool disable_meta_caching)
2850 {
2851         u32 mss = skb_shinfo(skb)->gso_size;
2852         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2853         u8 l4_protocol = 0;
2854
2855         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2856                 ena_tx_ctx->l4_csum_enable = 1;
2857                 if (mss) {
2858                         ena_tx_ctx->tso_enable = 1;
2859                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2860                         ena_tx_ctx->l4_csum_partial = 0;
2861                 } else {
2862                         ena_tx_ctx->tso_enable = 0;
2863                         ena_meta->l4_hdr_len = 0;
2864                         ena_tx_ctx->l4_csum_partial = 1;
2865                 }
2866
2867                 switch (ip_hdr(skb)->version) {
2868                 case IPVERSION:
2869                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2870                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
2871                                 ena_tx_ctx->df = 1;
2872                         if (mss)
2873                                 ena_tx_ctx->l3_csum_enable = 1;
2874                         l4_protocol = ip_hdr(skb)->protocol;
2875                         break;
2876                 case 6:
2877                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2878                         l4_protocol = ipv6_hdr(skb)->nexthdr;
2879                         break;
2880                 default:
2881                         break;
2882                 }
2883
2884                 if (l4_protocol == IPPROTO_TCP)
2885                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2886                 else
2887                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2888
2889                 ena_meta->mss = mss;
2890                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
2891                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
2892                 ena_tx_ctx->meta_valid = 1;
2893         } else if (disable_meta_caching) {
2894                 memset(ena_meta, 0, sizeof(*ena_meta));
2895                 ena_tx_ctx->meta_valid = 1;
2896         } else {
2897                 ena_tx_ctx->meta_valid = 0;
2898         }
2899 }
2900
2901 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2902                                        struct sk_buff *skb)
2903 {
2904         int num_frags, header_len, rc;
2905
2906         num_frags = skb_shinfo(skb)->nr_frags;
2907         header_len = skb_headlen(skb);
2908
2909         if (num_frags < tx_ring->sgl_size)
2910                 return 0;
2911
2912         if ((num_frags == tx_ring->sgl_size) &&
2913             (header_len < tx_ring->tx_max_header_size))
2914                 return 0;
2915
2916         ena_increase_stat(&tx_ring->tx_stats.linearize, 1, &tx_ring->syncp);
2917
2918         rc = skb_linearize(skb);
2919         if (unlikely(rc)) {
2920                 ena_increase_stat(&tx_ring->tx_stats.linearize_failed, 1,
2921                                   &tx_ring->syncp);
2922         }
2923
2924         return rc;
2925 }
2926
2927 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2928                           struct ena_tx_buffer *tx_info,
2929                           struct sk_buff *skb,
2930                           void **push_hdr,
2931                           u16 *header_len)
2932 {
2933         struct ena_adapter *adapter = tx_ring->adapter;
2934         struct ena_com_buf *ena_buf;
2935         dma_addr_t dma;
2936         u32 skb_head_len, frag_len, last_frag;
2937         u16 push_len = 0;
2938         u16 delta = 0;
2939         int i = 0;
2940
2941         skb_head_len = skb_headlen(skb);
2942         tx_info->skb = skb;
2943         ena_buf = tx_info->bufs;
2944
2945         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2946                 /* When the device is LLQ mode, the driver will copy
2947                  * the header into the device memory space.
2948                  * the ena_com layer assume the header is in a linear
2949                  * memory space.
2950                  * This assumption might be wrong since part of the header
2951                  * can be in the fragmented buffers.
2952                  * Use skb_header_pointer to make sure the header is in a
2953                  * linear memory space.
2954                  */
2955
2956                 push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2957                 *push_hdr = skb_header_pointer(skb, 0, push_len,
2958                                                tx_ring->push_buf_intermediate_buf);
2959                 *header_len = push_len;
2960                 if (unlikely(skb->data != *push_hdr)) {
2961                         ena_increase_stat(&tx_ring->tx_stats.llq_buffer_copy, 1,
2962                                           &tx_ring->syncp);
2963
2964                         delta = push_len - skb_head_len;
2965                 }
2966         } else {
2967                 *push_hdr = NULL;
2968                 *header_len = min_t(u32, skb_head_len,
2969                                     tx_ring->tx_max_header_size);
2970         }
2971
2972         netif_dbg(adapter, tx_queued, adapter->netdev,
2973                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2974                   *push_hdr, push_len);
2975
2976         if (skb_head_len > push_len) {
2977                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2978                                      skb_head_len - push_len, DMA_TO_DEVICE);
2979                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2980                         goto error_report_dma_error;
2981
2982                 ena_buf->paddr = dma;
2983                 ena_buf->len = skb_head_len - push_len;
2984
2985                 ena_buf++;
2986                 tx_info->num_of_bufs++;
2987                 tx_info->map_linear_data = 1;
2988         } else {
2989                 tx_info->map_linear_data = 0;
2990         }
2991
2992         last_frag = skb_shinfo(skb)->nr_frags;
2993
2994         for (i = 0; i < last_frag; i++) {
2995                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2996
2997                 frag_len = skb_frag_size(frag);
2998
2999                 if (unlikely(delta >= frag_len)) {
3000                         delta -= frag_len;
3001                         continue;
3002                 }
3003
3004                 dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
3005                                        frag_len - delta, DMA_TO_DEVICE);
3006                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
3007                         goto error_report_dma_error;
3008
3009                 ena_buf->paddr = dma;
3010                 ena_buf->len = frag_len - delta;
3011                 ena_buf++;
3012                 tx_info->num_of_bufs++;
3013                 delta = 0;
3014         }
3015
3016         return 0;
3017
3018 error_report_dma_error:
3019         ena_increase_stat(&tx_ring->tx_stats.dma_mapping_err, 1,
3020                           &tx_ring->syncp);
3021         netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map skb\n");
3022
3023         tx_info->skb = NULL;
3024
3025         tx_info->num_of_bufs += i;
3026         ena_unmap_tx_buff(tx_ring, tx_info);
3027
3028         return -EINVAL;
3029 }
3030
3031 /* Called with netif_tx_lock. */
3032 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
3033 {
3034         struct ena_adapter *adapter = netdev_priv(dev);
3035         struct ena_tx_buffer *tx_info;
3036         struct ena_com_tx_ctx ena_tx_ctx;
3037         struct ena_ring *tx_ring;
3038         struct netdev_queue *txq;
3039         void *push_hdr;
3040         u16 next_to_use, req_id, header_len;
3041         int qid, rc;
3042
3043         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
3044         /*  Determine which tx ring we will be placed on */
3045         qid = skb_get_queue_mapping(skb);
3046         tx_ring = &adapter->tx_ring[qid];
3047         txq = netdev_get_tx_queue(dev, qid);
3048
3049         rc = ena_check_and_linearize_skb(tx_ring, skb);
3050         if (unlikely(rc))
3051                 goto error_drop_packet;
3052
3053         skb_tx_timestamp(skb);
3054
3055         next_to_use = tx_ring->next_to_use;
3056         req_id = tx_ring->free_ids[next_to_use];
3057         tx_info = &tx_ring->tx_buffer_info[req_id];
3058         tx_info->num_of_bufs = 0;
3059
3060         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
3061
3062         rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
3063         if (unlikely(rc))
3064                 goto error_drop_packet;
3065
3066         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
3067         ena_tx_ctx.ena_bufs = tx_info->bufs;
3068         ena_tx_ctx.push_header = push_hdr;
3069         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
3070         ena_tx_ctx.req_id = req_id;
3071         ena_tx_ctx.header_len = header_len;
3072
3073         /* set flags and meta data */
3074         ena_tx_csum(&ena_tx_ctx, skb, tx_ring->disable_meta_caching);
3075
3076         rc = ena_xmit_common(dev,
3077                              tx_ring,
3078                              tx_info,
3079                              &ena_tx_ctx,
3080                              next_to_use,
3081                              skb->len);
3082         if (rc)
3083                 goto error_unmap_dma;
3084
3085         netdev_tx_sent_queue(txq, skb->len);
3086
3087         /* stop the queue when no more space available, the packet can have up
3088          * to sgl_size + 2. one for the meta descriptor and one for header
3089          * (if the header is larger than tx_max_header_size).
3090          */
3091         if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3092                                                    tx_ring->sgl_size + 2))) {
3093                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3094                           __func__, qid);
3095
3096                 netif_tx_stop_queue(txq);
3097                 ena_increase_stat(&tx_ring->tx_stats.queue_stop, 1,
3098                                   &tx_ring->syncp);
3099
3100                 /* There is a rare condition where this function decide to
3101                  * stop the queue but meanwhile clean_tx_irq updates
3102                  * next_to_completion and terminates.
3103                  * The queue will remain stopped forever.
3104                  * To solve this issue add a mb() to make sure that
3105                  * netif_tx_stop_queue() write is vissible before checking if
3106                  * there is additional space in the queue.
3107                  */
3108                 smp_mb();
3109
3110                 if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3111                                                  ENA_TX_WAKEUP_THRESH)) {
3112                         netif_tx_wake_queue(txq);
3113                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
3114                                           &tx_ring->syncp);
3115                 }
3116         }
3117
3118         if (netif_xmit_stopped(txq) || !netdev_xmit_more())
3119                 /* trigger the dma engine. ena_ring_tx_doorbell()
3120                  * calls a memory barrier inside it.
3121                  */
3122                 ena_ring_tx_doorbell(tx_ring);
3123
3124         return NETDEV_TX_OK;
3125
3126 error_unmap_dma:
3127         ena_unmap_tx_buff(tx_ring, tx_info);
3128         tx_info->skb = NULL;
3129
3130 error_drop_packet:
3131         dev_kfree_skb(skb);
3132         return NETDEV_TX_OK;
3133 }
3134
3135 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3136                             struct net_device *sb_dev)
3137 {
3138         u16 qid;
3139         /* we suspect that this is good for in--kernel network services that
3140          * want to loop incoming skb rx to tx in normal user generated traffic,
3141          * most probably we will not get to this
3142          */
3143         if (skb_rx_queue_recorded(skb))
3144                 qid = skb_get_rx_queue(skb);
3145         else
3146                 qid = netdev_pick_tx(dev, skb, NULL);
3147
3148         return qid;
3149 }
3150
3151 static void ena_config_host_info(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3152 {
3153         struct device *dev = &pdev->dev;
3154         struct ena_admin_host_info *host_info;
3155         int rc;
3156
3157         /* Allocate only the host info */
3158         rc = ena_com_allocate_host_info(ena_dev);
3159         if (rc) {
3160                 dev_err(dev, "Cannot allocate host info\n");
3161                 return;
3162         }
3163
3164         host_info = ena_dev->host_attr.host_info;
3165
3166         host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
3167         host_info->os_type = ENA_ADMIN_OS_LINUX;
3168         host_info->kernel_ver = LINUX_VERSION_CODE;
3169         strlcpy(host_info->kernel_ver_str, utsname()->version,
3170                 sizeof(host_info->kernel_ver_str) - 1);
3171         host_info->os_dist = 0;
3172         strncpy(host_info->os_dist_str, utsname()->release,
3173                 sizeof(host_info->os_dist_str) - 1);
3174         host_info->driver_version =
3175                 (DRV_MODULE_GEN_MAJOR) |
3176                 (DRV_MODULE_GEN_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3177                 (DRV_MODULE_GEN_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3178                 ("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3179         host_info->num_cpus = num_online_cpus();
3180
3181         host_info->driver_supported_features =
3182                 ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK |
3183                 ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK |
3184                 ENA_ADMIN_HOST_INFO_RX_BUF_MIRRORING_MASK |
3185                 ENA_ADMIN_HOST_INFO_RSS_CONFIGURABLE_FUNCTION_KEY_MASK;
3186
3187         rc = ena_com_set_host_attributes(ena_dev);
3188         if (rc) {
3189                 if (rc == -EOPNOTSUPP)
3190                         dev_warn(dev, "Cannot set host attributes\n");
3191                 else
3192                         dev_err(dev, "Cannot set host attributes\n");
3193
3194                 goto err;
3195         }
3196
3197         return;
3198
3199 err:
3200         ena_com_delete_host_info(ena_dev);
3201 }
3202
3203 static void ena_config_debug_area(struct ena_adapter *adapter)
3204 {
3205         u32 debug_area_size;
3206         int rc, ss_count;
3207
3208         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3209         if (ss_count <= 0) {
3210                 netif_err(adapter, drv, adapter->netdev,
3211                           "SS count is negative\n");
3212                 return;
3213         }
3214
3215         /* allocate 32 bytes for each string and 64bit for the value */
3216         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3217
3218         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3219         if (rc) {
3220                 netif_err(adapter, drv, adapter->netdev,
3221                           "Cannot allocate debug area\n");
3222                 return;
3223         }
3224
3225         rc = ena_com_set_host_attributes(adapter->ena_dev);
3226         if (rc) {
3227                 if (rc == -EOPNOTSUPP)
3228                         netif_warn(adapter, drv, adapter->netdev,
3229                                    "Cannot set host attributes\n");
3230                 else
3231                         netif_err(adapter, drv, adapter->netdev,
3232                                   "Cannot set host attributes\n");
3233                 goto err;
3234         }
3235
3236         return;
3237 err:
3238         ena_com_delete_debug_area(adapter->ena_dev);
3239 }
3240
3241 int ena_update_hw_stats(struct ena_adapter *adapter)
3242 {
3243         int rc = 0;
3244
3245         rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_stats);
3246         if (rc) {
3247                 dev_info_once(&adapter->pdev->dev, "Failed to get ENI stats\n");
3248                 return rc;
3249         }
3250
3251         return 0;
3252 }
3253
3254 static void ena_get_stats64(struct net_device *netdev,
3255                             struct rtnl_link_stats64 *stats)
3256 {
3257         struct ena_adapter *adapter = netdev_priv(netdev);
3258         struct ena_ring *rx_ring, *tx_ring;
3259         unsigned int start;
3260         u64 rx_drops;
3261         u64 tx_drops;
3262         int i;
3263
3264         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3265                 return;
3266
3267         for (i = 0; i < adapter->num_io_queues; i++) {
3268                 u64 bytes, packets;
3269
3270                 tx_ring = &adapter->tx_ring[i];
3271
3272                 do {
3273                         start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
3274                         packets = tx_ring->tx_stats.cnt;
3275                         bytes = tx_ring->tx_stats.bytes;
3276                 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
3277
3278                 stats->tx_packets += packets;
3279                 stats->tx_bytes += bytes;
3280
3281                 rx_ring = &adapter->rx_ring[i];
3282
3283                 do {
3284                         start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
3285                         packets = rx_ring->rx_stats.cnt;
3286                         bytes = rx_ring->rx_stats.bytes;
3287                 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
3288
3289                 stats->rx_packets += packets;
3290                 stats->rx_bytes += bytes;
3291         }
3292
3293         do {
3294                 start = u64_stats_fetch_begin_irq(&adapter->syncp);
3295                 rx_drops = adapter->dev_stats.rx_drops;
3296                 tx_drops = adapter->dev_stats.tx_drops;
3297         } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
3298
3299         stats->rx_dropped = rx_drops;
3300         stats->tx_dropped = tx_drops;
3301
3302         stats->multicast = 0;
3303         stats->collisions = 0;
3304
3305         stats->rx_length_errors = 0;
3306         stats->rx_crc_errors = 0;
3307         stats->rx_frame_errors = 0;
3308         stats->rx_fifo_errors = 0;
3309         stats->rx_missed_errors = 0;
3310         stats->tx_window_errors = 0;
3311
3312         stats->rx_errors = 0;
3313         stats->tx_errors = 0;
3314 }
3315
3316 static const struct net_device_ops ena_netdev_ops = {
3317         .ndo_open               = ena_open,
3318         .ndo_stop               = ena_close,
3319         .ndo_start_xmit         = ena_start_xmit,
3320         .ndo_select_queue       = ena_select_queue,
3321         .ndo_get_stats64        = ena_get_stats64,
3322         .ndo_tx_timeout         = ena_tx_timeout,
3323         .ndo_change_mtu         = ena_change_mtu,
3324         .ndo_set_mac_address    = NULL,
3325         .ndo_validate_addr      = eth_validate_addr,
3326         .ndo_bpf                = ena_xdp,
3327         .ndo_xdp_xmit           = ena_xdp_xmit,
3328 };
3329
3330 static int ena_device_validate_params(struct ena_adapter *adapter,
3331                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
3332 {
3333         struct net_device *netdev = adapter->netdev;
3334         int rc;
3335
3336         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3337                               adapter->mac_addr);
3338         if (!rc) {
3339                 netif_err(adapter, drv, netdev,
3340                           "Error, mac address are different\n");
3341                 return -EINVAL;
3342         }
3343
3344         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3345                 netif_err(adapter, drv, netdev,
3346                           "Error, device max mtu is smaller than netdev MTU\n");
3347                 return -EINVAL;
3348         }
3349
3350         return 0;
3351 }
3352
3353 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
3354 {
3355         llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
3356         llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
3357         llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
3358         llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
3359         llq_config->llq_ring_entry_size_value = 128;
3360 }
3361
3362 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3363                                            struct ena_com_dev *ena_dev,
3364                                            struct ena_admin_feature_llq_desc *llq,
3365                                            struct ena_llq_configurations *llq_default_configurations)
3366 {
3367         int rc;
3368         u32 llq_feature_mask;
3369
3370         llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3371         if (!(ena_dev->supported_features & llq_feature_mask)) {
3372                 dev_warn(&pdev->dev,
3373                         "LLQ is not supported Fallback to host mode policy.\n");
3374                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3375                 return 0;
3376         }
3377
3378         rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3379         if (unlikely(rc)) {
3380                 dev_err(&pdev->dev,
3381                         "Failed to configure the device mode.  Fallback to host mode policy.\n");
3382                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3383         }
3384
3385         return 0;
3386 }
3387
3388 static int ena_map_llq_mem_bar(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
3389                                int bars)
3390 {
3391         bool has_mem_bar = !!(bars & BIT(ENA_MEM_BAR));
3392
3393         if (!has_mem_bar) {
3394                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3395                         dev_err(&pdev->dev,
3396                                 "ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3397                         ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3398                 }
3399
3400                 return 0;
3401         }
3402
3403         ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3404                                            pci_resource_start(pdev, ENA_MEM_BAR),
3405                                            pci_resource_len(pdev, ENA_MEM_BAR));
3406
3407         if (!ena_dev->mem_bar)
3408                 return -EFAULT;
3409
3410         return 0;
3411 }
3412
3413 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
3414                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
3415                            bool *wd_state)
3416 {
3417         struct ena_llq_configurations llq_config;
3418         struct device *dev = &pdev->dev;
3419         bool readless_supported;
3420         u32 aenq_groups;
3421         int dma_width;
3422         int rc;
3423
3424         rc = ena_com_mmio_reg_read_request_init(ena_dev);
3425         if (rc) {
3426                 dev_err(dev, "Failed to init mmio read less\n");
3427                 return rc;
3428         }
3429
3430         /* The PCIe configuration space revision id indicate if mmio reg
3431          * read is disabled
3432          */
3433         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3434         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3435
3436         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3437         if (rc) {
3438                 dev_err(dev, "Can not reset device\n");
3439                 goto err_mmio_read_less;
3440         }
3441
3442         rc = ena_com_validate_version(ena_dev);
3443         if (rc) {
3444                 dev_err(dev, "Device version is too low\n");
3445                 goto err_mmio_read_less;
3446         }
3447
3448         dma_width = ena_com_get_dma_width(ena_dev);
3449         if (dma_width < 0) {
3450                 dev_err(dev, "Invalid dma width value %d", dma_width);
3451                 rc = dma_width;
3452                 goto err_mmio_read_less;
3453         }
3454
3455         rc = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(dma_width));
3456         if (rc) {
3457                 dev_err(dev, "dma_set_mask_and_coherent failed %d\n", rc);
3458                 goto err_mmio_read_less;
3459         }
3460
3461         /* ENA admin level init */
3462         rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3463         if (rc) {
3464                 dev_err(dev,
3465                         "Can not initialize ena admin queue with device\n");
3466                 goto err_mmio_read_less;
3467         }
3468
3469         /* To enable the msix interrupts the driver needs to know the number
3470          * of queues. So the driver uses polling mode to retrieve this
3471          * information
3472          */
3473         ena_com_set_admin_polling_mode(ena_dev, true);
3474
3475         ena_config_host_info(ena_dev, pdev);
3476
3477         /* Get Device Attributes*/
3478         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3479         if (rc) {
3480                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3481                 goto err_admin_init;
3482         }
3483
3484         /* Try to turn all the available aenq groups */
3485         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3486                 BIT(ENA_ADMIN_FATAL_ERROR) |
3487                 BIT(ENA_ADMIN_WARNING) |
3488                 BIT(ENA_ADMIN_NOTIFICATION) |
3489                 BIT(ENA_ADMIN_KEEP_ALIVE);
3490
3491         aenq_groups &= get_feat_ctx->aenq.supported_groups;
3492
3493         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3494         if (rc) {
3495                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3496                 goto err_admin_init;
3497         }
3498
3499         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3500
3501         set_default_llq_configurations(&llq_config);
3502
3503         rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx->llq,
3504                                              &llq_config);
3505         if (rc) {
3506                 dev_err(dev, "ENA device init failed\n");
3507                 goto err_admin_init;
3508         }
3509
3510         return 0;
3511
3512 err_admin_init:
3513         ena_com_delete_host_info(ena_dev);
3514         ena_com_admin_destroy(ena_dev);
3515 err_mmio_read_less:
3516         ena_com_mmio_reg_read_request_destroy(ena_dev);
3517
3518         return rc;
3519 }
3520
3521 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3522 {
3523         struct ena_com_dev *ena_dev = adapter->ena_dev;
3524         struct device *dev = &adapter->pdev->dev;
3525         int rc;
3526
3527         rc = ena_enable_msix(adapter);
3528         if (rc) {
3529                 dev_err(dev, "Can not reserve msix vectors\n");
3530                 return rc;
3531         }
3532
3533         ena_setup_mgmnt_intr(adapter);
3534
3535         rc = ena_request_mgmnt_irq(adapter);
3536         if (rc) {
3537                 dev_err(dev, "Can not setup management interrupts\n");
3538                 goto err_disable_msix;
3539         }
3540
3541         ena_com_set_admin_polling_mode(ena_dev, false);
3542
3543         ena_com_admin_aenq_enable(ena_dev);
3544
3545         return 0;
3546
3547 err_disable_msix:
3548         ena_disable_msix(adapter);
3549
3550         return rc;
3551 }
3552
3553 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3554 {
3555         struct net_device *netdev = adapter->netdev;
3556         struct ena_com_dev *ena_dev = adapter->ena_dev;
3557         bool dev_up;
3558
3559         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3560                 return;
3561
3562         netif_carrier_off(netdev);
3563
3564         del_timer_sync(&adapter->timer_service);
3565
3566         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3567         adapter->dev_up_before_reset = dev_up;
3568         if (!graceful)
3569                 ena_com_set_admin_running_state(ena_dev, false);
3570
3571         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3572                 ena_down(adapter);
3573
3574         /* Stop the device from sending AENQ events (in case reset flag is set
3575          *  and device is up, ena_down() already reset the device.
3576          */
3577         if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3578                 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3579
3580         ena_free_mgmnt_irq(adapter);
3581
3582         ena_disable_msix(adapter);
3583
3584         ena_com_abort_admin_commands(ena_dev);
3585
3586         ena_com_wait_for_abort_completion(ena_dev);
3587
3588         ena_com_admin_destroy(ena_dev);
3589
3590         ena_com_mmio_reg_read_request_destroy(ena_dev);
3591
3592         /* return reset reason to default value */
3593         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3594
3595         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3596         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3597 }
3598
3599 static int ena_restore_device(struct ena_adapter *adapter)
3600 {
3601         struct ena_com_dev_get_features_ctx get_feat_ctx;
3602         struct ena_com_dev *ena_dev = adapter->ena_dev;
3603         struct pci_dev *pdev = adapter->pdev;
3604         bool wd_state;
3605         int rc;
3606
3607         set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3608         rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
3609         if (rc) {
3610                 dev_err(&pdev->dev, "Can not initialize device\n");
3611                 goto err;
3612         }
3613         adapter->wd_state = wd_state;
3614
3615         rc = ena_device_validate_params(adapter, &get_feat_ctx);
3616         if (rc) {
3617                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
3618                 goto err_device_destroy;
3619         }
3620
3621         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3622         if (rc) {
3623                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
3624                 goto err_device_destroy;
3625         }
3626         /* If the interface was up before the reset bring it up */
3627         if (adapter->dev_up_before_reset) {
3628                 rc = ena_up(adapter);
3629                 if (rc) {
3630                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
3631                         goto err_disable_msix;
3632                 }
3633         }
3634
3635         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3636
3637         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3638         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3639                 netif_carrier_on(adapter->netdev);
3640
3641         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3642         adapter->last_keep_alive_jiffies = jiffies;
3643
3644         dev_err(&pdev->dev, "Device reset completed successfully\n");
3645
3646         return rc;
3647 err_disable_msix:
3648         ena_free_mgmnt_irq(adapter);
3649         ena_disable_msix(adapter);
3650 err_device_destroy:
3651         ena_com_abort_admin_commands(ena_dev);
3652         ena_com_wait_for_abort_completion(ena_dev);
3653         ena_com_admin_destroy(ena_dev);
3654         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3655         ena_com_mmio_reg_read_request_destroy(ena_dev);
3656 err:
3657         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3658         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3659         dev_err(&pdev->dev,
3660                 "Reset attempt failed. Can not reset the device\n");
3661
3662         return rc;
3663 }
3664
3665 static void ena_fw_reset_device(struct work_struct *work)
3666 {
3667         struct ena_adapter *adapter =
3668                 container_of(work, struct ena_adapter, reset_task);
3669
3670         rtnl_lock();
3671
3672         if (likely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3673                 ena_destroy_device(adapter, false);
3674                 ena_restore_device(adapter);
3675         }
3676
3677         rtnl_unlock();
3678 }
3679
3680 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3681                                         struct ena_ring *rx_ring)
3682 {
3683         struct ena_napi *ena_napi = container_of(rx_ring->napi, struct ena_napi, napi);
3684
3685         if (likely(READ_ONCE(ena_napi->first_interrupt)))
3686                 return 0;
3687
3688         if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3689                 return 0;
3690
3691         rx_ring->no_interrupt_event_cnt++;
3692
3693         if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3694                 netif_err(adapter, rx_err, adapter->netdev,
3695                           "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3696                           rx_ring->qid);
3697                 adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3698                 smp_mb__before_atomic();
3699                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3700                 return -EIO;
3701         }
3702
3703         return 0;
3704 }
3705
3706 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3707                                           struct ena_ring *tx_ring)
3708 {
3709         struct ena_napi *ena_napi = container_of(tx_ring->napi, struct ena_napi, napi);
3710         unsigned int time_since_last_napi;
3711         unsigned int missing_tx_comp_to;
3712         bool is_tx_comp_time_expired;
3713         struct ena_tx_buffer *tx_buf;
3714         unsigned long last_jiffies;
3715         u32 missed_tx = 0;
3716         int i, rc = 0;
3717
3718         for (i = 0; i < tx_ring->ring_size; i++) {
3719                 tx_buf = &tx_ring->tx_buffer_info[i];
3720                 last_jiffies = tx_buf->last_jiffies;
3721
3722                 if (last_jiffies == 0)
3723                         /* no pending Tx at this location */
3724                         continue;
3725
3726                 is_tx_comp_time_expired = time_is_before_jiffies(last_jiffies +
3727                          2 * adapter->missing_tx_completion_to);
3728
3729                 if (unlikely(!READ_ONCE(ena_napi->first_interrupt) && is_tx_comp_time_expired)) {
3730                         /* If after graceful period interrupt is still not
3731                          * received, we schedule a reset
3732                          */
3733                         netif_err(adapter, tx_err, adapter->netdev,
3734                                   "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3735                                   tx_ring->qid);
3736                         adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3737                         smp_mb__before_atomic();
3738                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3739                         return -EIO;
3740                 }
3741
3742                 is_tx_comp_time_expired = time_is_before_jiffies(last_jiffies +
3743                         adapter->missing_tx_completion_to);
3744
3745                 if (unlikely(is_tx_comp_time_expired)) {
3746                         if (!tx_buf->print_once) {
3747                                 time_since_last_napi = jiffies_to_usecs(jiffies - tx_ring->tx_stats.last_napi_jiffies);
3748                                 missing_tx_comp_to = jiffies_to_msecs(adapter->missing_tx_completion_to);
3749                                 netif_notice(adapter, tx_err, adapter->netdev,
3750                                              "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",
3751                                              tx_ring->qid, i, time_since_last_napi, missing_tx_comp_to);
3752                         }
3753
3754                         tx_buf->print_once = 1;
3755                         missed_tx++;
3756                 }
3757         }
3758
3759         if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3760                 netif_err(adapter, tx_err, adapter->netdev,
3761                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3762                           missed_tx,
3763                           adapter->missing_tx_completion_threshold);
3764                 adapter->reset_reason =
3765                         ENA_REGS_RESET_MISS_TX_CMPL;
3766                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3767                 rc = -EIO;
3768         }
3769
3770         ena_increase_stat(&tx_ring->tx_stats.missed_tx, missed_tx,
3771                           &tx_ring->syncp);
3772
3773         return rc;
3774 }
3775
3776 static void check_for_missing_completions(struct ena_adapter *adapter)
3777 {
3778         struct ena_ring *tx_ring;
3779         struct ena_ring *rx_ring;
3780         int i, budget, rc;
3781         int io_queue_count;
3782
3783         io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
3784         /* Make sure the driver doesn't turn the device in other process */
3785         smp_rmb();
3786
3787         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3788                 return;
3789
3790         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3791                 return;
3792
3793         if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
3794                 return;
3795
3796         budget = ENA_MONITORED_TX_QUEUES;
3797
3798         for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
3799                 tx_ring = &adapter->tx_ring[i];
3800                 rx_ring = &adapter->rx_ring[i];
3801
3802                 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3803                 if (unlikely(rc))
3804                         return;
3805
3806                 rc =  !ENA_IS_XDP_INDEX(adapter, i) ?
3807                         check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
3808                 if (unlikely(rc))
3809                         return;
3810
3811                 budget--;
3812                 if (!budget)
3813                         break;
3814         }
3815
3816         adapter->last_monitored_tx_qid = i % io_queue_count;
3817 }
3818
3819 /* trigger napi schedule after 2 consecutive detections */
3820 #define EMPTY_RX_REFILL 2
3821 /* For the rare case where the device runs out of Rx descriptors and the
3822  * napi handler failed to refill new Rx descriptors (due to a lack of memory
3823  * for example).
3824  * This case will lead to a deadlock:
3825  * The device won't send interrupts since all the new Rx packets will be dropped
3826  * The napi handler won't allocate new Rx descriptors so the device will be
3827  * able to send new packets.
3828  *
3829  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
3830  * It is recommended to have at least 512MB, with a minimum of 128MB for
3831  * constrained environment).
3832  *
3833  * When such a situation is detected - Reschedule napi
3834  */
3835 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
3836 {
3837         struct ena_ring *rx_ring;
3838         int i, refill_required;
3839
3840         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3841                 return;
3842
3843         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3844                 return;
3845
3846         for (i = 0; i < adapter->num_io_queues; i++) {
3847                 rx_ring = &adapter->rx_ring[i];
3848
3849                 refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
3850                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3851                         rx_ring->empty_rx_queue++;
3852
3853                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3854                                 ena_increase_stat(&rx_ring->rx_stats.empty_rx_ring, 1,
3855                                                   &rx_ring->syncp);
3856
3857                                 netif_err(adapter, drv, adapter->netdev,
3858                                           "Trigger refill for ring %d\n", i);
3859
3860                                 napi_schedule(rx_ring->napi);
3861                                 rx_ring->empty_rx_queue = 0;
3862                         }
3863                 } else {
3864                         rx_ring->empty_rx_queue = 0;
3865                 }
3866         }
3867 }
3868
3869 /* Check for keep alive expiration */
3870 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3871 {
3872         unsigned long keep_alive_expired;
3873
3874         if (!adapter->wd_state)
3875                 return;
3876
3877         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3878                 return;
3879
3880         keep_alive_expired = adapter->last_keep_alive_jiffies +
3881                              adapter->keep_alive_timeout;
3882         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3883                 netif_err(adapter, drv, adapter->netdev,
3884                           "Keep alive watchdog timeout.\n");
3885                 ena_increase_stat(&adapter->dev_stats.wd_expired, 1,
3886                                   &adapter->syncp);
3887                 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3888                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3889         }
3890 }
3891
3892 static void check_for_admin_com_state(struct ena_adapter *adapter)
3893 {
3894         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3895                 netif_err(adapter, drv, adapter->netdev,
3896                           "ENA admin queue is not in running state!\n");
3897                 ena_increase_stat(&adapter->dev_stats.admin_q_pause, 1,
3898                                   &adapter->syncp);
3899                 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3900                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3901         }
3902 }
3903
3904 static void ena_update_hints(struct ena_adapter *adapter,
3905                              struct ena_admin_ena_hw_hints *hints)
3906 {
3907         struct net_device *netdev = adapter->netdev;
3908
3909         if (hints->admin_completion_tx_timeout)
3910                 adapter->ena_dev->admin_queue.completion_timeout =
3911                         hints->admin_completion_tx_timeout * 1000;
3912
3913         if (hints->mmio_read_timeout)
3914                 /* convert to usec */
3915                 adapter->ena_dev->mmio_read.reg_read_to =
3916                         hints->mmio_read_timeout * 1000;
3917
3918         if (hints->missed_tx_completion_count_threshold_to_reset)
3919                 adapter->missing_tx_completion_threshold =
3920                         hints->missed_tx_completion_count_threshold_to_reset;
3921
3922         if (hints->missing_tx_completion_timeout) {
3923                 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3924                         adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3925                 else
3926                         adapter->missing_tx_completion_to =
3927                                 msecs_to_jiffies(hints->missing_tx_completion_timeout);
3928         }
3929
3930         if (hints->netdev_wd_timeout)
3931                 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3932
3933         if (hints->driver_watchdog_timeout) {
3934                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3935                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3936                 else
3937                         adapter->keep_alive_timeout =
3938                                 msecs_to_jiffies(hints->driver_watchdog_timeout);
3939         }
3940 }
3941
3942 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3943                                  struct net_device *netdev)
3944 {
3945         host_info->supported_network_features[0] =
3946                 netdev->features & GENMASK_ULL(31, 0);
3947         host_info->supported_network_features[1] =
3948                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
3949 }
3950
3951 static void ena_timer_service(struct timer_list *t)
3952 {
3953         struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3954         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3955         struct ena_admin_host_info *host_info =
3956                 adapter->ena_dev->host_attr.host_info;
3957
3958         check_for_missing_keep_alive(adapter);
3959
3960         check_for_admin_com_state(adapter);
3961
3962         check_for_missing_completions(adapter);
3963
3964         check_for_empty_rx_ring(adapter);
3965
3966         if (debug_area)
3967                 ena_dump_stats_to_buf(adapter, debug_area);
3968
3969         if (host_info)
3970                 ena_update_host_info(host_info, adapter->netdev);
3971
3972         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3973                 netif_err(adapter, drv, adapter->netdev,
3974                           "Trigger reset is on\n");
3975                 ena_dump_stats_to_dmesg(adapter);
3976                 queue_work(ena_wq, &adapter->reset_task);
3977                 return;
3978         }
3979
3980         /* Reset the timer */
3981         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3982 }
3983
3984 static u32 ena_calc_max_io_queue_num(struct pci_dev *pdev,
3985                                      struct ena_com_dev *ena_dev,
3986                                      struct ena_com_dev_get_features_ctx *get_feat_ctx)
3987 {
3988         u32 io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
3989
3990         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3991                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3992                         &get_feat_ctx->max_queue_ext.max_queue_ext;
3993                 io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
3994                                   max_queue_ext->max_rx_cq_num);
3995
3996                 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
3997                 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
3998         } else {
3999                 struct ena_admin_queue_feature_desc *max_queues =
4000                         &get_feat_ctx->max_queues;
4001                 io_tx_sq_num = max_queues->max_sq_num;
4002                 io_tx_cq_num = max_queues->max_cq_num;
4003                 io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
4004         }
4005
4006         /* In case of LLQ use the llq fields for the tx SQ/CQ */
4007         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4008                 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
4009
4010         max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
4011         max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
4012         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
4013         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
4014         /* 1 IRQ for mgmnt and 1 IRQs for each IO direction */
4015         max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
4016         if (unlikely(!max_num_io_queues)) {
4017                 dev_err(&pdev->dev, "The device doesn't have io queues\n");
4018                 return -EFAULT;
4019         }
4020
4021         return max_num_io_queues;
4022 }
4023
4024 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
4025                                  struct net_device *netdev)
4026 {
4027         netdev_features_t dev_features = 0;
4028
4029         /* Set offload features */
4030         if (feat->offload.tx &
4031                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
4032                 dev_features |= NETIF_F_IP_CSUM;
4033
4034         if (feat->offload.tx &
4035                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
4036                 dev_features |= NETIF_F_IPV6_CSUM;
4037
4038         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
4039                 dev_features |= NETIF_F_TSO;
4040
4041         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
4042                 dev_features |= NETIF_F_TSO6;
4043
4044         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
4045                 dev_features |= NETIF_F_TSO_ECN;
4046
4047         if (feat->offload.rx_supported &
4048                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
4049                 dev_features |= NETIF_F_RXCSUM;
4050
4051         if (feat->offload.rx_supported &
4052                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
4053                 dev_features |= NETIF_F_RXCSUM;
4054
4055         netdev->features =
4056                 dev_features |
4057                 NETIF_F_SG |
4058                 NETIF_F_RXHASH |
4059                 NETIF_F_HIGHDMA;
4060
4061         netdev->hw_features |= netdev->features;
4062         netdev->vlan_features |= netdev->features;
4063 }
4064
4065 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
4066                                      struct ena_com_dev_get_features_ctx *feat)
4067 {
4068         struct net_device *netdev = adapter->netdev;
4069
4070         /* Copy mac address */
4071         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
4072                 eth_hw_addr_random(netdev);
4073                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
4074         } else {
4075                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
4076                 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
4077         }
4078
4079         /* Set offload features */
4080         ena_set_dev_offloads(feat, netdev);
4081
4082         adapter->max_mtu = feat->dev_attr.max_mtu;
4083         netdev->max_mtu = adapter->max_mtu;
4084         netdev->min_mtu = ENA_MIN_MTU;
4085 }
4086
4087 static int ena_rss_init_default(struct ena_adapter *adapter)
4088 {
4089         struct ena_com_dev *ena_dev = adapter->ena_dev;
4090         struct device *dev = &adapter->pdev->dev;
4091         int rc, i;
4092         u32 val;
4093
4094         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
4095         if (unlikely(rc)) {
4096                 dev_err(dev, "Cannot init indirect table\n");
4097                 goto err_rss_init;
4098         }
4099
4100         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
4101                 val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
4102                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
4103                                                        ENA_IO_RXQ_IDX(val));
4104                 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4105                         dev_err(dev, "Cannot fill indirect table\n");
4106                         goto err_fill_indir;
4107                 }
4108         }
4109
4110         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ, NULL,
4111                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4112         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4113                 dev_err(dev, "Cannot fill hash function\n");
4114                 goto err_fill_indir;
4115         }
4116
4117         rc = ena_com_set_default_hash_ctrl(ena_dev);
4118         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4119                 dev_err(dev, "Cannot fill hash control\n");
4120                 goto err_fill_indir;
4121         }
4122
4123         return 0;
4124
4125 err_fill_indir:
4126         ena_com_rss_destroy(ena_dev);
4127 err_rss_init:
4128
4129         return rc;
4130 }
4131
4132 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4133 {
4134         int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4135
4136         pci_release_selected_regions(pdev, release_bars);
4137 }
4138
4139
4140 static int ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
4141 {
4142         struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
4143         struct ena_com_dev *ena_dev = ctx->ena_dev;
4144         u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
4145         u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
4146         u32 max_tx_queue_size;
4147         u32 max_rx_queue_size;
4148
4149         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4150                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4151                         &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
4152                 max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
4153                                           max_queue_ext->max_rx_sq_depth);
4154                 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
4155
4156                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4157                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4158                                                   llq->max_llq_depth);
4159                 else
4160                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4161                                                   max_queue_ext->max_tx_sq_depth);
4162
4163                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4164                                              max_queue_ext->max_per_packet_tx_descs);
4165                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4166                                              max_queue_ext->max_per_packet_rx_descs);
4167         } else {
4168                 struct ena_admin_queue_feature_desc *max_queues =
4169                         &ctx->get_feat_ctx->max_queues;
4170                 max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
4171                                           max_queues->max_sq_depth);
4172                 max_tx_queue_size = max_queues->max_cq_depth;
4173
4174                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4175                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4176                                                   llq->max_llq_depth);
4177                 else
4178                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4179                                                   max_queues->max_sq_depth);
4180
4181                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4182                                              max_queues->max_packet_tx_descs);
4183                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4184                                              max_queues->max_packet_rx_descs);
4185         }
4186
4187         max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
4188         max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
4189
4190         tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
4191                                   max_tx_queue_size);
4192         rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
4193                                   max_rx_queue_size);
4194
4195         tx_queue_size = rounddown_pow_of_two(tx_queue_size);
4196         rx_queue_size = rounddown_pow_of_two(rx_queue_size);
4197
4198         ctx->max_tx_queue_size = max_tx_queue_size;
4199         ctx->max_rx_queue_size = max_rx_queue_size;
4200         ctx->tx_queue_size = tx_queue_size;
4201         ctx->rx_queue_size = rx_queue_size;
4202
4203         return 0;
4204 }
4205
4206 /* ena_probe - Device Initialization Routine
4207  * @pdev: PCI device information struct
4208  * @ent: entry in ena_pci_tbl
4209  *
4210  * Returns 0 on success, negative on failure
4211  *
4212  * ena_probe initializes an adapter identified by a pci_dev structure.
4213  * The OS initialization, configuring of the adapter private structure,
4214  * and a hardware reset occur.
4215  */
4216 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4217 {
4218         struct ena_calc_queue_size_ctx calc_queue_ctx = {};
4219         struct ena_com_dev_get_features_ctx get_feat_ctx;
4220         struct ena_com_dev *ena_dev = NULL;
4221         struct ena_adapter *adapter;
4222         struct net_device *netdev;
4223         static int adapters_found;
4224         u32 max_num_io_queues;
4225         bool wd_state;
4226         int bars, rc;
4227
4228         dev_dbg(&pdev->dev, "%s\n", __func__);
4229
4230         rc = pci_enable_device_mem(pdev);
4231         if (rc) {
4232                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4233                 return rc;
4234         }
4235
4236         rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
4237         if (rc) {
4238                 dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n", rc);
4239                 goto err_disable_device;
4240         }
4241
4242         pci_set_master(pdev);
4243
4244         ena_dev = vzalloc(sizeof(*ena_dev));
4245         if (!ena_dev) {
4246                 rc = -ENOMEM;
4247                 goto err_disable_device;
4248         }
4249
4250         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4251         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4252         if (rc) {
4253                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4254                         rc);
4255                 goto err_free_ena_dev;
4256         }
4257
4258         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4259                                         pci_resource_start(pdev, ENA_REG_BAR),
4260                                         pci_resource_len(pdev, ENA_REG_BAR));
4261         if (!ena_dev->reg_bar) {
4262                 dev_err(&pdev->dev, "Failed to remap regs bar\n");
4263                 rc = -EFAULT;
4264                 goto err_free_region;
4265         }
4266
4267         ena_dev->ena_min_poll_delay_us = ENA_ADMIN_POLL_DELAY_US;
4268
4269         ena_dev->dmadev = &pdev->dev;
4270
4271         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), ENA_MAX_RINGS);
4272         if (!netdev) {
4273                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4274                 rc = -ENOMEM;
4275                 goto err_free_region;
4276         }
4277
4278         SET_NETDEV_DEV(netdev, &pdev->dev);
4279         adapter = netdev_priv(netdev);
4280         adapter->ena_dev = ena_dev;
4281         adapter->netdev = netdev;
4282         adapter->pdev = pdev;
4283         adapter->msg_enable = DEFAULT_MSG_ENABLE;
4284
4285         ena_dev->net_device = netdev;
4286
4287         pci_set_drvdata(pdev, adapter);
4288
4289         rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
4290         if (rc) {
4291                 dev_err(&pdev->dev, "ENA device init failed\n");
4292                 if (rc == -ETIME)
4293                         rc = -EPROBE_DEFER;
4294                 goto err_netdev_destroy;
4295         }
4296
4297         rc = ena_map_llq_mem_bar(pdev, ena_dev, bars);
4298         if (rc) {
4299                 dev_err(&pdev->dev, "ENA llq bar mapping failed\n");
4300                 goto err_device_destroy;
4301         }
4302
4303         calc_queue_ctx.ena_dev = ena_dev;
4304         calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
4305         calc_queue_ctx.pdev = pdev;
4306
4307         /* Initial TX and RX interrupt delay. Assumes 1 usec granularity.
4308          * Updated during device initialization with the real granularity
4309          */
4310         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4311         ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4312         ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4313         max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4314         rc = ena_calc_io_queue_size(&calc_queue_ctx);
4315         if (rc || !max_num_io_queues) {
4316                 rc = -EFAULT;
4317                 goto err_device_destroy;
4318         }
4319
4320         ena_set_conf_feat_params(adapter, &get_feat_ctx);
4321
4322         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4323
4324         adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
4325         adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
4326         adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
4327         adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
4328         adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
4329         adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
4330
4331         adapter->num_io_queues = max_num_io_queues;
4332         adapter->max_num_io_queues = max_num_io_queues;
4333         adapter->last_monitored_tx_qid = 0;
4334
4335         adapter->xdp_first_ring = 0;
4336         adapter->xdp_num_queues = 0;
4337
4338         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4339         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4340                 adapter->disable_meta_caching =
4341                         !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
4342                            BIT(ENA_ADMIN_DISABLE_META_CACHING));
4343
4344         adapter->wd_state = wd_state;
4345
4346         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4347
4348         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4349         if (rc) {
4350                 dev_err(&pdev->dev,
4351                         "Failed to query interrupt moderation feature\n");
4352                 goto err_device_destroy;
4353         }
4354         ena_init_io_rings(adapter,
4355                           0,
4356                           adapter->xdp_num_queues +
4357                           adapter->num_io_queues);
4358
4359         netdev->netdev_ops = &ena_netdev_ops;
4360         netdev->watchdog_timeo = TX_TIMEOUT;
4361         ena_set_ethtool_ops(netdev);
4362
4363         netdev->priv_flags |= IFF_UNICAST_FLT;
4364
4365         u64_stats_init(&adapter->syncp);
4366
4367         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4368         if (rc) {
4369                 dev_err(&pdev->dev,
4370                         "Failed to enable and set the admin interrupts\n");
4371                 goto err_worker_destroy;
4372         }
4373         rc = ena_rss_init_default(adapter);
4374         if (rc && (rc != -EOPNOTSUPP)) {
4375                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4376                 goto err_free_msix;
4377         }
4378
4379         ena_config_debug_area(adapter);
4380
4381         if (!ena_update_hw_stats(adapter))
4382                 adapter->eni_stats_supported = true;
4383         else
4384                 adapter->eni_stats_supported = false;
4385
4386         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4387
4388         netif_carrier_off(netdev);
4389
4390         rc = register_netdev(netdev);
4391         if (rc) {
4392                 dev_err(&pdev->dev, "Cannot register net device\n");
4393                 goto err_rss;
4394         }
4395
4396         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4397
4398         adapter->last_keep_alive_jiffies = jiffies;
4399         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4400         adapter->missing_tx_completion_to = TX_TIMEOUT;
4401         adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4402
4403         ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4404
4405         timer_setup(&adapter->timer_service, ena_timer_service, 0);
4406         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4407
4408         dev_info(&pdev->dev,
4409                  "%s found at mem %lx, mac addr %pM\n",
4410                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4411                  netdev->dev_addr);
4412
4413         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4414
4415         adapters_found++;
4416
4417         return 0;
4418
4419 err_rss:
4420         ena_com_delete_debug_area(ena_dev);
4421         ena_com_rss_destroy(ena_dev);
4422 err_free_msix:
4423         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4424         /* stop submitting admin commands on a device that was reset */
4425         ena_com_set_admin_running_state(ena_dev, false);
4426         ena_free_mgmnt_irq(adapter);
4427         ena_disable_msix(adapter);
4428 err_worker_destroy:
4429         del_timer(&adapter->timer_service);
4430 err_device_destroy:
4431         ena_com_delete_host_info(ena_dev);
4432         ena_com_admin_destroy(ena_dev);
4433 err_netdev_destroy:
4434         free_netdev(netdev);
4435 err_free_region:
4436         ena_release_bars(ena_dev, pdev);
4437 err_free_ena_dev:
4438         vfree(ena_dev);
4439 err_disable_device:
4440         pci_disable_device(pdev);
4441         return rc;
4442 }
4443
4444 /*****************************************************************************/
4445
4446 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4447  * @pdev: PCI device information struct
4448  * @shutdown: Is it a shutdown operation? If false, means it is a removal
4449  *
4450  * __ena_shutoff is a helper routine that does the real work on shutdown and
4451  * removal paths; the difference between those paths is with regards to whether
4452  * dettach or unregister the netdevice.
4453  */
4454 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4455 {
4456         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4457         struct ena_com_dev *ena_dev;
4458         struct net_device *netdev;
4459
4460         ena_dev = adapter->ena_dev;
4461         netdev = adapter->netdev;
4462
4463 #ifdef CONFIG_RFS_ACCEL
4464         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4465                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4466                 netdev->rx_cpu_rmap = NULL;
4467         }
4468 #endif /* CONFIG_RFS_ACCEL */
4469
4470         /* Make sure timer and reset routine won't be called after
4471          * freeing device resources.
4472          */
4473         del_timer_sync(&adapter->timer_service);
4474         cancel_work_sync(&adapter->reset_task);
4475
4476         rtnl_lock(); /* lock released inside the below if-else block */
4477         adapter->reset_reason = ENA_REGS_RESET_SHUTDOWN;
4478         ena_destroy_device(adapter, true);
4479         if (shutdown) {
4480                 netif_device_detach(netdev);
4481                 dev_close(netdev);
4482                 rtnl_unlock();
4483         } else {
4484                 rtnl_unlock();
4485                 unregister_netdev(netdev);
4486                 free_netdev(netdev);
4487         }
4488
4489         ena_com_rss_destroy(ena_dev);
4490
4491         ena_com_delete_debug_area(ena_dev);
4492
4493         ena_com_delete_host_info(ena_dev);
4494
4495         ena_release_bars(ena_dev, pdev);
4496
4497         pci_disable_device(pdev);
4498
4499         vfree(ena_dev);
4500 }
4501
4502 /* ena_remove - Device Removal Routine
4503  * @pdev: PCI device information struct
4504  *
4505  * ena_remove is called by the PCI subsystem to alert the driver
4506  * that it should release a PCI device.
4507  */
4508
4509 static void ena_remove(struct pci_dev *pdev)
4510 {
4511         __ena_shutoff(pdev, false);
4512 }
4513
4514 /* ena_shutdown - Device Shutdown Routine
4515  * @pdev: PCI device information struct
4516  *
4517  * ena_shutdown is called by the PCI subsystem to alert the driver that
4518  * a shutdown/reboot (or kexec) is happening and device must be disabled.
4519  */
4520
4521 static void ena_shutdown(struct pci_dev *pdev)
4522 {
4523         __ena_shutoff(pdev, true);
4524 }
4525
4526 /* ena_suspend - PM suspend callback
4527  * @dev_d: Device information struct
4528  */
4529 static int __maybe_unused ena_suspend(struct device *dev_d)
4530 {
4531         struct pci_dev *pdev = to_pci_dev(dev_d);
4532         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4533
4534         ena_increase_stat(&adapter->dev_stats.suspend, 1, &adapter->syncp);
4535
4536         rtnl_lock();
4537         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4538                 dev_err(&pdev->dev,
4539                         "Ignoring device reset request as the device is being suspended\n");
4540                 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4541         }
4542         ena_destroy_device(adapter, true);
4543         rtnl_unlock();
4544         return 0;
4545 }
4546
4547 /* ena_resume - PM resume callback
4548  * @dev_d: Device information struct
4549  */
4550 static int __maybe_unused ena_resume(struct device *dev_d)
4551 {
4552         struct ena_adapter *adapter = dev_get_drvdata(dev_d);
4553         int rc;
4554
4555         ena_increase_stat(&adapter->dev_stats.resume, 1, &adapter->syncp);
4556
4557         rtnl_lock();
4558         rc = ena_restore_device(adapter);
4559         rtnl_unlock();
4560         return rc;
4561 }
4562
4563 static SIMPLE_DEV_PM_OPS(ena_pm_ops, ena_suspend, ena_resume);
4564
4565 static struct pci_driver ena_pci_driver = {
4566         .name           = DRV_MODULE_NAME,
4567         .id_table       = ena_pci_tbl,
4568         .probe          = ena_probe,
4569         .remove         = ena_remove,
4570         .shutdown       = ena_shutdown,
4571         .driver.pm      = &ena_pm_ops,
4572         .sriov_configure = pci_sriov_configure_simple,
4573 };
4574
4575 static int __init ena_init(void)
4576 {
4577         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4578         if (!ena_wq) {
4579                 pr_err("Failed to create workqueue\n");
4580                 return -ENOMEM;
4581         }
4582
4583         return pci_register_driver(&ena_pci_driver);
4584 }
4585
4586 static void __exit ena_cleanup(void)
4587 {
4588         pci_unregister_driver(&ena_pci_driver);
4589
4590         if (ena_wq) {
4591                 destroy_workqueue(ena_wq);
4592                 ena_wq = NULL;
4593         }
4594 }
4595
4596 /******************************************************************************
4597  ******************************** AENQ Handlers *******************************
4598  *****************************************************************************/
4599 /* ena_update_on_link_change:
4600  * Notify the network interface about the change in link status
4601  */
4602 static void ena_update_on_link_change(void *adapter_data,
4603                                       struct ena_admin_aenq_entry *aenq_e)
4604 {
4605         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4606         struct ena_admin_aenq_link_change_desc *aenq_desc =
4607                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
4608         int status = aenq_desc->flags &
4609                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4610
4611         if (status) {
4612                 netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
4613                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4614                 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4615                         netif_carrier_on(adapter->netdev);
4616         } else {
4617                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4618                 netif_carrier_off(adapter->netdev);
4619         }
4620 }
4621
4622 static void ena_keep_alive_wd(void *adapter_data,
4623                               struct ena_admin_aenq_entry *aenq_e)
4624 {
4625         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4626         struct ena_admin_aenq_keep_alive_desc *desc;
4627         u64 rx_drops;
4628         u64 tx_drops;
4629
4630         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4631         adapter->last_keep_alive_jiffies = jiffies;
4632
4633         rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4634         tx_drops = ((u64)desc->tx_drops_high << 32) | desc->tx_drops_low;
4635
4636         u64_stats_update_begin(&adapter->syncp);
4637         /* These stats are accumulated by the device, so the counters indicate
4638          * all drops since last reset.
4639          */
4640         adapter->dev_stats.rx_drops = rx_drops;
4641         adapter->dev_stats.tx_drops = tx_drops;
4642         u64_stats_update_end(&adapter->syncp);
4643 }
4644
4645 static void ena_notification(void *adapter_data,
4646                              struct ena_admin_aenq_entry *aenq_e)
4647 {
4648         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4649         struct ena_admin_ena_hw_hints *hints;
4650
4651         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4652              "Invalid group(%x) expected %x\n",
4653              aenq_e->aenq_common_desc.group,
4654              ENA_ADMIN_NOTIFICATION);
4655
4656         switch (aenq_e->aenq_common_desc.syndrome) {
4657         case ENA_ADMIN_UPDATE_HINTS:
4658                 hints = (struct ena_admin_ena_hw_hints *)
4659                         (&aenq_e->inline_data_w4);
4660                 ena_update_hints(adapter, hints);
4661                 break;
4662         default:
4663                 netif_err(adapter, drv, adapter->netdev,
4664                           "Invalid aenq notification link state %d\n",
4665                           aenq_e->aenq_common_desc.syndrome);
4666         }
4667 }
4668
4669 /* This handler will called for unknown event group or unimplemented handlers*/
4670 static void unimplemented_aenq_handler(void *data,
4671                                        struct ena_admin_aenq_entry *aenq_e)
4672 {
4673         struct ena_adapter *adapter = (struct ena_adapter *)data;
4674
4675         netif_err(adapter, drv, adapter->netdev,
4676                   "Unknown event was received or event with unimplemented handler\n");
4677 }
4678
4679 static struct ena_aenq_handlers aenq_handlers = {
4680         .handlers = {
4681                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4682                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
4683                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4684         },
4685         .unimplemented_handler = unimplemented_aenq_handler
4686 };
4687
4688 module_init(ena_init);
4689 module_exit(ena_cleanup);