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 static int debug = -1;
  39 module_param(debug, int, 0);
  40 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  41
  42 static struct ena_aenq_handlers aenq_handlers;
  43
  44 static struct workqueue_struct *ena_wq;
  45
  46 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
  47
  48 static int ena_rss_init_default(struct ena_adapter *adapter);
  49 static void check_for_admin_com_state(struct ena_adapter *adapter);
  50 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
  51 static int ena_restore_device(struct ena_adapter *adapter);
  52
  53 static void ena_init_io_rings(struct ena_adapter *adapter,
  54                               int first_index, int count);
  55 static void ena_init_napi_in_range(struct ena_adapter *adapter, int first_index,
  56                                    int count);
  57 static void ena_del_napi_in_range(struct ena_adapter *adapter, int first_index,
  58                                   int count);
  59 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid);
  60 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
  61                                            int first_index,
  62                                            int count);
  63 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid);
  64 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid);
  65 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget);
  66 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter);
  67 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter);
  68 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
  69                                       int first_index, int count);
  70 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
  71                                      int first_index, int count);
  72 static int ena_up(struct ena_adapter *adapter);
  73 static void ena_down(struct ena_adapter *adapter);
  74 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
  75                                  struct ena_ring *rx_ring);
  76 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
  77                                       struct ena_ring *rx_ring);
  78 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
  79                               struct ena_tx_buffer *tx_info);
  80 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
  81                                             int first_index, int count);
  82
  83 /* Increase a stat by cnt while holding syncp seqlock on 32bit machines */
  84 static void ena_increase_stat(u64 *statp, u64 cnt,
  85                               struct u64_stats_sync *syncp)
  86 {
  87         u64_stats_update_begin(syncp);
  88         (*statp) += cnt;
  89         u64_stats_update_end(syncp);
  90 }
  91
  92 static void ena_tx_timeout(struct net_device *dev, unsigned int txqueue)
  93 {
  94         struct ena_adapter *adapter = netdev_priv(dev);
  95
  96         /* Change the state of the device to trigger reset
  97          * Check that we are not in the middle or a trigger already
  98          */
  99
 100         if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
 101                 return;
 102
 103         adapter->reset_reason = ENA_REGS_RESET_OS_NETDEV_WD;
 104         ena_increase_stat(&adapter->dev_stats.tx_timeout, 1, &adapter->syncp);
 105
 106         netif_err(adapter, tx_err, dev, "Transmit time out\n");
 107 }
 108
 109 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
 110 {
 111         int i;
 112
 113         for (i = 0; i < adapter->num_io_queues; i++)
 114                 adapter->rx_ring[i].mtu = mtu;
 115 }
 116
 117 static int ena_change_mtu(struct net_device *dev, int new_mtu)
 118 {
 119         struct ena_adapter *adapter = netdev_priv(dev);
 120         int ret;
 121
 122         ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
 123         if (!ret) {
 124                 netif_dbg(adapter, drv, dev, "Set MTU to %d\n", new_mtu);
 125                 update_rx_ring_mtu(adapter, new_mtu);
 126                 dev->mtu = new_mtu;
 127         } else {
 128                 netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
 129                           new_mtu);
 130         }
 131
 132         return ret;
 133 }
 134
 135 static int ena_xmit_common(struct net_device *dev,
 136                            struct ena_ring *ring,
 137                            struct ena_tx_buffer *tx_info,
 138                            struct ena_com_tx_ctx *ena_tx_ctx,
 139                            u16 next_to_use,
 140                            u32 bytes)
 141 {
 142         struct ena_adapter *adapter = netdev_priv(dev);
 143         int rc, nb_hw_desc;
 144
 145         if (unlikely(ena_com_is_doorbell_needed(ring->ena_com_io_sq,
 146                                                 ena_tx_ctx))) {
 147                 netif_dbg(adapter, tx_queued, dev,
 148                           "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
 149                           ring->qid);
 150                 ena_com_write_sq_doorbell(ring->ena_com_io_sq);
 151         }
 152
 153         /* prepare the packet's descriptors to dma engine */
 154         rc = ena_com_prepare_tx(ring->ena_com_io_sq, ena_tx_ctx,
 155                                 &nb_hw_desc);
 156
 157         /* In case there isn't enough space in the queue for the packet,
 158          * we simply drop it. All other failure reasons of
 159          * ena_com_prepare_tx() are fatal and therefore require a device reset.
 160          */
 161         if (unlikely(rc)) {
 162                 netif_err(adapter, tx_queued, dev,
 163                           "Failed to prepare tx bufs\n");
 164                 ena_increase_stat(&ring->tx_stats.prepare_ctx_err, 1,
 165                                   &ring->syncp);
 166                 if (rc != -ENOMEM) {
 167                         adapter->reset_reason =
 168                                 ENA_REGS_RESET_DRIVER_INVALID_STATE;
 169                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
 170                 }
 171                 return rc;
 172         }
 173
 174         u64_stats_update_begin(&ring->syncp);
 175         ring->tx_stats.cnt++;
 176         ring->tx_stats.bytes += bytes;
 177         u64_stats_update_end(&ring->syncp);
 178
 179         tx_info->tx_descs = nb_hw_desc;
 180         tx_info->last_jiffies = jiffies;
 181         tx_info->print_once = 0;
 182
 183         ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
 184                                                  ring->ring_size);
 185         return 0;
 186 }
 187
 188 /* This is the XDP napi callback. XDP queues use a separate napi callback
 189  * than Rx/Tx queues.
 190  */
 191 static int ena_xdp_io_poll(struct napi_struct *napi, int budget)
 192 {
 193         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
 194         u32 xdp_work_done, xdp_budget;
 195         struct ena_ring *xdp_ring;
 196         int napi_comp_call = 0;
 197         int ret;
 198
 199         xdp_ring = ena_napi->xdp_ring;
 200         xdp_ring->first_interrupt = ena_napi->first_interrupt;
 201
 202         xdp_budget = budget;
 203
 204         if (!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags) ||
 205             test_bit(ENA_FLAG_TRIGGER_RESET, &xdp_ring->adapter->flags)) {
 206                 napi_complete_done(napi, 0);
 207                 return 0;
 208         }
 209
 210         xdp_work_done = ena_clean_xdp_irq(xdp_ring, xdp_budget);
 211
 212         /* If the device is about to reset or down, avoid unmask
 213          * the interrupt and return 0 so NAPI won't reschedule
 214          */
 215         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags))) {
 216                 napi_complete_done(napi, 0);
 217                 ret = 0;
 218         } else if (xdp_budget > xdp_work_done) {
 219                 napi_comp_call = 1;
 220                 if (napi_complete_done(napi, xdp_work_done))
 221                         ena_unmask_interrupt(xdp_ring, NULL);
 222                 ena_update_ring_numa_node(xdp_ring, NULL);
 223                 ret = xdp_work_done;
 224         } else {
 225                 ret = xdp_budget;
 226         }
 227
 228         u64_stats_update_begin(&xdp_ring->syncp);
 229         xdp_ring->tx_stats.napi_comp += napi_comp_call;
 230         xdp_ring->tx_stats.tx_poll++;
 231         u64_stats_update_end(&xdp_ring->syncp);
 232
 233         return ret;
 234 }
 235
 236 static int ena_xdp_tx_map_frame(struct ena_ring *xdp_ring,
 237                                 struct ena_tx_buffer *tx_info,
 238                                 struct xdp_frame *xdpf,
 239                                 struct ena_com_tx_ctx *ena_tx_ctx)
 240 {
 241         struct ena_adapter *adapter = xdp_ring->adapter;
 242         struct ena_com_buf *ena_buf;
 243         int push_len = 0;
 244         dma_addr_t dma;
 245         void *data;
 246         u32 size;
 247
 248         tx_info->xdpf = xdpf;
 249         data = tx_info->xdpf->data;
 250         size = tx_info->xdpf->len;
 251
 252         if (xdp_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
 253                 /* Designate part of the packet for LLQ */
 254                 push_len = min_t(u32, size, xdp_ring->tx_max_header_size);
 255
 256                 ena_tx_ctx->push_header = data;
 257
 258                 size -= push_len;
 259                 data += push_len;
 260         }
 261
 262         ena_tx_ctx->header_len = push_len;
 263
 264         if (size > 0) {
 265                 dma = dma_map_single(xdp_ring->dev,
 266                                      data,
 267                                      size,
 268                                      DMA_TO_DEVICE);
 269                 if (unlikely(dma_mapping_error(xdp_ring->dev, dma)))
 270                         goto error_report_dma_error;
 271
 272                 tx_info->map_linear_data = 0;
 273
 274                 ena_buf = tx_info->bufs;
 275                 ena_buf->paddr = dma;
 276                 ena_buf->len = size;
 277
 278                 ena_tx_ctx->ena_bufs = ena_buf;
 279                 ena_tx_ctx->num_bufs = tx_info->num_of_bufs = 1;
 280         }
 281
 282         return 0;
 283
 284 error_report_dma_error:
 285         ena_increase_stat(&xdp_ring->tx_stats.dma_mapping_err, 1,
 286                           &xdp_ring->syncp);
 287         netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map xdp buff\n");
 288
 289         return -EINVAL;
 290 }
 291
 292 static int ena_xdp_xmit_frame(struct ena_ring *xdp_ring,
 293                               struct net_device *dev,
 294                               struct xdp_frame *xdpf,
 295                               int flags)
 296 {
 297         struct ena_com_tx_ctx ena_tx_ctx = {};
 298         struct ena_tx_buffer *tx_info;
 299         u16 next_to_use, req_id;
 300         int rc;
 301
 302         next_to_use = xdp_ring->next_to_use;
 303         req_id = xdp_ring->free_ids[next_to_use];
 304         tx_info = &xdp_ring->tx_buffer_info[req_id];
 305         tx_info->num_of_bufs = 0;
 306
 307         rc = ena_xdp_tx_map_frame(xdp_ring, tx_info, xdpf, &ena_tx_ctx);
 308         if (unlikely(rc))
 309                 return rc;
 310
 311         ena_tx_ctx.req_id = req_id;
 312
 313         rc = ena_xmit_common(dev,
 314                              xdp_ring,
 315                              tx_info,
 316                              &ena_tx_ctx,
 317                              next_to_use,
 318                              xdpf->len);
 319         if (rc)
 320                 goto error_unmap_dma;
 321         /* trigger the dma engine. ena_com_write_sq_doorbell()
 322          * has a mb
 323          */
 324         if (flags & XDP_XMIT_FLUSH) {
 325                 ena_com_write_sq_doorbell(xdp_ring->ena_com_io_sq);
 326                 ena_increase_stat(&xdp_ring->tx_stats.doorbells, 1,
 327                                   &xdp_ring->syncp);
 328         }
 329
 330         return rc;
 331
 332 error_unmap_dma:
 333         ena_unmap_tx_buff(xdp_ring, tx_info);
 334         tx_info->xdpf = NULL;
 335         return rc;
 336 }
 337
 338 static int ena_xdp_xmit(struct net_device *dev, int n,
 339                         struct xdp_frame **frames, u32 flags)
 340 {
 341         struct ena_adapter *adapter = netdev_priv(dev);
 342         struct ena_ring *xdp_ring;
 343         int qid, i, nxmit = 0;
 344
 345         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 346                 return -EINVAL;
 347
 348         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
 349                 return -ENETDOWN;
 350
 351         /* We assume that all rings have the same XDP program */
 352         if (!READ_ONCE(adapter->rx_ring->xdp_bpf_prog))
 353                 return -ENXIO;
 354
 355         qid = smp_processor_id() % adapter->xdp_num_queues;
 356         qid += adapter->xdp_first_ring;
 357         xdp_ring = &adapter->tx_ring[qid];
 358
 359         /* Other CPU ids might try to send thorugh this queue */
 360         spin_lock(&xdp_ring->xdp_tx_lock);
 361
 362         for (i = 0; i < n; i++) {
 363                 if (ena_xdp_xmit_frame(xdp_ring, dev, frames[i], 0))
 364                         break;
 365                 nxmit++;
 366         }
 367
 368         /* Ring doorbell to make device aware of the packets */
 369         if (flags & XDP_XMIT_FLUSH) {
 370                 ena_com_write_sq_doorbell(xdp_ring->ena_com_io_sq);
 371                 ena_increase_stat(&xdp_ring->tx_stats.doorbells, 1,
 372                                   &xdp_ring->syncp);
 373         }
 374
 375         spin_unlock(&xdp_ring->xdp_tx_lock);
 376
 377         /* Return number of packets sent */
 378         return nxmit;
 379 }
 380
 381 static int ena_xdp_execute(struct ena_ring *rx_ring, struct xdp_buff *xdp)
 382 {
 383         struct bpf_prog *xdp_prog;
 384         struct ena_ring *xdp_ring;
 385         u32 verdict = XDP_PASS;
 386         struct xdp_frame *xdpf;
 387         u64 *xdp_stat;
 388         int qid;
 389
 390         rcu_read_lock();
 391         xdp_prog = READ_ONCE(rx_ring->xdp_bpf_prog);
 392
 393         if (!xdp_prog)
 394                 goto out;
 395
 396         verdict = bpf_prog_run_xdp(xdp_prog, xdp);
 397
 398         switch (verdict) {
 399         case XDP_TX:
 400                 xdpf = xdp_convert_buff_to_frame(xdp);
 401                 if (unlikely(!xdpf)) {
 402                         trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 403                         xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 404                         verdict = XDP_ABORTED;
 405                         break;
 406                 }
 407
 408                 /* Find xmit queue */
 409                 qid = rx_ring->qid + rx_ring->adapter->num_io_queues;
 410                 xdp_ring = &rx_ring->adapter->tx_ring[qid];
 411
 412                 /* The XDP queues are shared between XDP_TX and XDP_REDIRECT */
 413                 spin_lock(&xdp_ring->xdp_tx_lock);
 414
 415                 if (ena_xdp_xmit_frame(xdp_ring, rx_ring->netdev, xdpf,
 416                                        XDP_XMIT_FLUSH))
 417                         xdp_return_frame(xdpf);
 418
 419                 spin_unlock(&xdp_ring->xdp_tx_lock);
 420                 xdp_stat = &rx_ring->rx_stats.xdp_tx;
 421                 break;
 422         case XDP_REDIRECT:
 423                 if (likely(!xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog))) {
 424                         xdp_stat = &rx_ring->rx_stats.xdp_redirect;
 425                         break;
 426                 }
 427                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 428                 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 429                 verdict = XDP_ABORTED;
 430                 break;
 431         case XDP_ABORTED:
 432                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 433                 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 434                 break;
 435         case XDP_DROP:
 436                 xdp_stat = &rx_ring->rx_stats.xdp_drop;
 437                 break;
 438         case XDP_PASS:
 439                 xdp_stat = &rx_ring->rx_stats.xdp_pass;
 440                 break;
 441         default:
 442                 bpf_warn_invalid_xdp_action(verdict);
 443                 xdp_stat = &rx_ring->rx_stats.xdp_invalid;
 444         }
 445
 446         ena_increase_stat(xdp_stat, 1, &rx_ring->syncp);
 447 out:
 448         rcu_read_unlock();
 449
 450         return verdict;
 451 }
 452
 453 static void ena_init_all_xdp_queues(struct ena_adapter *adapter)
 454 {
 455         adapter->xdp_first_ring = adapter->num_io_queues;
 456         adapter->xdp_num_queues = adapter->num_io_queues;
 457
 458         ena_init_io_rings(adapter,
 459                           adapter->xdp_first_ring,
 460                           adapter->xdp_num_queues);
 461 }
 462
 463 static int ena_setup_and_create_all_xdp_queues(struct ena_adapter *adapter)
 464 {
 465         int rc = 0;
 466
 467         rc = ena_setup_tx_resources_in_range(adapter, adapter->xdp_first_ring,
 468                                              adapter->xdp_num_queues);
 469         if (rc)
 470                 goto setup_err;
 471
 472         rc = ena_create_io_tx_queues_in_range(adapter,
 473                                               adapter->xdp_first_ring,
 474                                               adapter->xdp_num_queues);
 475         if (rc)
 476                 goto create_err;
 477
 478         return 0;
 479
 480 create_err:
 481         ena_free_all_io_tx_resources(adapter);
 482 setup_err:
 483         return rc;
 484 }
 485
 486 /* Provides a way for both kernel and bpf-prog to know
 487  * more about the RX-queue a given XDP frame arrived on.
 488  */
 489 static int ena_xdp_register_rxq_info(struct ena_ring *rx_ring)
 490 {
 491         int rc;
 492
 493         rc = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev, rx_ring->qid, 0);
 494
 495         if (rc) {
 496                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 497                           "Failed to register xdp rx queue info. RX queue num %d rc: %d\n",
 498                           rx_ring->qid, rc);
 499                 goto err;
 500         }
 501
 502         rc = xdp_rxq_info_reg_mem_model(&rx_ring->xdp_rxq, MEM_TYPE_PAGE_SHARED,
 503                                         NULL);
 504
 505         if (rc) {
 506                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 507                           "Failed to register xdp rx queue info memory model. RX queue num %d rc: %d\n",
 508                           rx_ring->qid, rc);
 509                 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 510         }
 511
 512 err:
 513         return rc;
 514 }
 515
 516 static void ena_xdp_unregister_rxq_info(struct ena_ring *rx_ring)
 517 {
 518         xdp_rxq_info_unreg_mem_model(&rx_ring->xdp_rxq);
 519         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 520 }
 521
 522 static void ena_xdp_exchange_program_rx_in_range(struct ena_adapter *adapter,
 523                                                  struct bpf_prog *prog,
 524                                                  int first, int count)
 525 {
 526         struct ena_ring *rx_ring;
 527         int i = 0;
 528
 529         for (i = first; i < count; i++) {
 530                 rx_ring = &adapter->rx_ring[i];
 531                 xchg(&rx_ring->xdp_bpf_prog, prog);
 532                 if (prog) {
 533                         ena_xdp_register_rxq_info(rx_ring);
 534                         rx_ring->rx_headroom = XDP_PACKET_HEADROOM;
 535                 } else {
 536                         ena_xdp_unregister_rxq_info(rx_ring);
 537                         rx_ring->rx_headroom = 0;
 538                 }
 539         }
 540 }
 541
 542 static void ena_xdp_exchange_program(struct ena_adapter *adapter,
 543                                      struct bpf_prog *prog)
 544 {
 545         struct bpf_prog *old_bpf_prog = xchg(&adapter->xdp_bpf_prog, prog);
 546
 547         ena_xdp_exchange_program_rx_in_range(adapter,
 548                                              prog,
 549                                              0,
 550                                              adapter->num_io_queues);
 551
 552         if (old_bpf_prog)
 553                 bpf_prog_put(old_bpf_prog);
 554 }
 555
 556 static int ena_destroy_and_free_all_xdp_queues(struct ena_adapter *adapter)
 557 {
 558         bool was_up;
 559         int rc;
 560
 561         was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 562
 563         if (was_up)
 564                 ena_down(adapter);
 565
 566         adapter->xdp_first_ring = 0;
 567         adapter->xdp_num_queues = 0;
 568         ena_xdp_exchange_program(adapter, NULL);
 569         if (was_up) {
 570                 rc = ena_up(adapter);
 571                 if (rc)
 572                         return rc;
 573         }
 574         return 0;
 575 }
 576
 577 static int ena_xdp_set(struct net_device *netdev, struct netdev_bpf *bpf)
 578 {
 579         struct ena_adapter *adapter = netdev_priv(netdev);
 580         struct bpf_prog *prog = bpf->prog;
 581         struct bpf_prog *old_bpf_prog;
 582         int rc, prev_mtu;
 583         bool is_up;
 584
 585         is_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 586         rc = ena_xdp_allowed(adapter);
 587         if (rc == ENA_XDP_ALLOWED) {
 588                 old_bpf_prog = adapter->xdp_bpf_prog;
 589                 if (prog) {
 590                         if (!is_up) {
 591                                 ena_init_all_xdp_queues(adapter);
 592                         } else if (!old_bpf_prog) {
 593                                 ena_down(adapter);
 594                                 ena_init_all_xdp_queues(adapter);
 595                         }
 596                         ena_xdp_exchange_program(adapter, prog);
 597
 598                         if (is_up && !old_bpf_prog) {
 599                                 rc = ena_up(adapter);
 600                                 if (rc)
 601                                         return rc;
 602                         }
 603                 } else if (old_bpf_prog) {
 604                         rc = ena_destroy_and_free_all_xdp_queues(adapter);
 605                         if (rc)
 606                                 return rc;
 607                 }
 608
 609                 prev_mtu = netdev->max_mtu;
 610                 netdev->max_mtu = prog ? ENA_XDP_MAX_MTU : adapter->max_mtu;
 611
 612                 if (!old_bpf_prog)
 613                         netif_info(adapter, drv, adapter->netdev,
 614                                    "XDP program is set, changing the max_mtu from %d to %d",
 615                                    prev_mtu, netdev->max_mtu);
 616
 617         } else if (rc == ENA_XDP_CURRENT_MTU_TOO_LARGE) {
 618                 netif_err(adapter, drv, adapter->netdev,
 619                           "Failed to set xdp program, the current MTU (%d) is larger than the maximum allowed MTU (%lu) while xdp is on",
 620                           netdev->mtu, ENA_XDP_MAX_MTU);
 621                 NL_SET_ERR_MSG_MOD(bpf->extack,
 622                                    "Failed to set xdp program, the current MTU is larger than the maximum allowed MTU. Check the dmesg for more info");
 623                 return -EINVAL;
 624         } else if (rc == ENA_XDP_NO_ENOUGH_QUEUES) {
 625                 netif_err(adapter, drv, adapter->netdev,
 626                           "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",
 627                           adapter->num_io_queues, adapter->max_num_io_queues);
 628                 NL_SET_ERR_MSG_MOD(bpf->extack,
 629                                    "Failed to set xdp program, there is no enough space for allocating XDP queues, Check the dmesg for more info");
 630                 return -EINVAL;
 631         }
 632
 633         return 0;
 634 }
 635
 636 /* This is the main xdp callback, it's used by the kernel to set/unset the xdp
 637  * program as well as to query the current xdp program id.
 638  */
 639 static int ena_xdp(struct net_device *netdev, struct netdev_bpf *bpf)
 640 {
 641         switch (bpf->command) {
 642         case XDP_SETUP_PROG:
 643                 return ena_xdp_set(netdev, bpf);
 644         default:
 645                 return -EINVAL;
 646         }
 647         return 0;
 648 }
 649
 650 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
 651 {
 652 #ifdef CONFIG_RFS_ACCEL
 653         u32 i;
 654         int rc;
 655
 656         adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_io_queues);
 657         if (!adapter->netdev->rx_cpu_rmap)
 658                 return -ENOMEM;
 659         for (i = 0; i < adapter->num_io_queues; i++) {
 660                 int irq_idx = ENA_IO_IRQ_IDX(i);
 661
 662                 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
 663                                       pci_irq_vector(adapter->pdev, irq_idx));
 664                 if (rc) {
 665                         free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
 666                         adapter->netdev->rx_cpu_rmap = NULL;
 667                         return rc;
 668                 }
 669         }
 670 #endif /* CONFIG_RFS_ACCEL */
 671         return 0;
 672 }
 673
 674 static void ena_init_io_rings_common(struct ena_adapter *adapter,
 675                                      struct ena_ring *ring, u16 qid)
 676 {
 677         ring->qid = qid;
 678         ring->pdev = adapter->pdev;
 679         ring->dev = &adapter->pdev->dev;
 680         ring->netdev = adapter->netdev;
 681         ring->napi = &adapter->ena_napi[qid].napi;
 682         ring->adapter = adapter;
 683         ring->ena_dev = adapter->ena_dev;
 684         ring->per_napi_packets = 0;
 685         ring->cpu = 0;
 686         ring->first_interrupt = false;
 687         ring->no_interrupt_event_cnt = 0;
 688         u64_stats_init(&ring->syncp);
 689 }
 690
 691 static void ena_init_io_rings(struct ena_adapter *adapter,
 692                               int first_index, int count)
 693 {
 694         struct ena_com_dev *ena_dev;
 695         struct ena_ring *txr, *rxr;
 696         int i;
 697
 698         ena_dev = adapter->ena_dev;
 699
 700         for (i = first_index; i < first_index + count; i++) {
 701                 txr = &adapter->tx_ring[i];
 702                 rxr = &adapter->rx_ring[i];
 703
 704                 /* TX common ring state */
 705                 ena_init_io_rings_common(adapter, txr, i);
 706
 707                 /* TX specific ring state */
 708                 txr->ring_size = adapter->requested_tx_ring_size;
 709                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
 710                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
 711                 txr->sgl_size = adapter->max_tx_sgl_size;
 712                 txr->smoothed_interval =
 713                         ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
 714                 txr->disable_meta_caching = adapter->disable_meta_caching;
 715                 spin_lock_init(&txr->xdp_tx_lock);
 716
 717                 /* Don't init RX queues for xdp queues */
 718                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
 719                         /* RX common ring state */
 720                         ena_init_io_rings_common(adapter, rxr, i);
 721
 722                         /* RX specific ring state */
 723                         rxr->ring_size = adapter->requested_rx_ring_size;
 724                         rxr->rx_copybreak = adapter->rx_copybreak;
 725                         rxr->sgl_size = adapter->max_rx_sgl_size;
 726                         rxr->smoothed_interval =
 727                                 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
 728                         rxr->empty_rx_queue = 0;
 729                         adapter->ena_napi[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
 730                 }
 731         }
 732 }
 733
 734 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
 735  * @adapter: network interface device structure
 736  * @qid: queue index
 737  *
 738  * Return 0 on success, negative on failure
 739  */
 740 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
 741 {
 742         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 743         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 744         int size, i, node;
 745
 746         if (tx_ring->tx_buffer_info) {
 747                 netif_err(adapter, ifup,
 748                           adapter->netdev, "tx_buffer_info info is not NULL");
 749                 return -EEXIST;
 750         }
 751
 752         size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
 753         node = cpu_to_node(ena_irq->cpu);
 754
 755         tx_ring->tx_buffer_info = vzalloc_node(size, node);
 756         if (!tx_ring->tx_buffer_info) {
 757                 tx_ring->tx_buffer_info = vzalloc(size);
 758                 if (!tx_ring->tx_buffer_info)
 759                         goto err_tx_buffer_info;
 760         }
 761
 762         size = sizeof(u16) * tx_ring->ring_size;
 763         tx_ring->free_ids = vzalloc_node(size, node);
 764         if (!tx_ring->free_ids) {
 765                 tx_ring->free_ids = vzalloc(size);
 766                 if (!tx_ring->free_ids)
 767                         goto err_tx_free_ids;
 768         }
 769
 770         size = tx_ring->tx_max_header_size;
 771         tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
 772         if (!tx_ring->push_buf_intermediate_buf) {
 773                 tx_ring->push_buf_intermediate_buf = vzalloc(size);
 774                 if (!tx_ring->push_buf_intermediate_buf)
 775                         goto err_push_buf_intermediate_buf;
 776         }
 777
 778         /* Req id ring for TX out of order completions */
 779         for (i = 0; i < tx_ring->ring_size; i++)
 780                 tx_ring->free_ids[i] = i;
 781
 782         /* Reset tx statistics */
 783         memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
 784
 785         tx_ring->next_to_use = 0;
 786         tx_ring->next_to_clean = 0;
 787         tx_ring->cpu = ena_irq->cpu;
 788         return 0;
 789
 790 err_push_buf_intermediate_buf:
 791         vfree(tx_ring->free_ids);
 792         tx_ring->free_ids = NULL;
 793 err_tx_free_ids:
 794         vfree(tx_ring->tx_buffer_info);
 795         tx_ring->tx_buffer_info = NULL;
 796 err_tx_buffer_info:
 797         return -ENOMEM;
 798 }
 799
 800 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
 801  * @adapter: network interface device structure
 802  * @qid: queue index
 803  *
 804  * Free all transmit software resources
 805  */
 806 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
 807 {
 808         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 809
 810         vfree(tx_ring->tx_buffer_info);
 811         tx_ring->tx_buffer_info = NULL;
 812
 813         vfree(tx_ring->free_ids);
 814         tx_ring->free_ids = NULL;
 815
 816         vfree(tx_ring->push_buf_intermediate_buf);
 817         tx_ring->push_buf_intermediate_buf = NULL;
 818 }
 819
 820 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
 821                                            int first_index,
 822                                            int count)
 823 {
 824         int i, rc = 0;
 825
 826         for (i = first_index; i < first_index + count; i++) {
 827                 rc = ena_setup_tx_resources(adapter, i);
 828                 if (rc)
 829                         goto err_setup_tx;
 830         }
 831
 832         return 0;
 833
 834 err_setup_tx:
 835
 836         netif_err(adapter, ifup, adapter->netdev,
 837                   "Tx queue %d: allocation failed\n", i);
 838
 839         /* rewind the index freeing the rings as we go */
 840         while (first_index < i--)
 841                 ena_free_tx_resources(adapter, i);
 842         return rc;
 843 }
 844
 845 static void ena_free_all_io_tx_resources_in_range(struct ena_adapter *adapter,
 846                                                   int first_index, int count)
 847 {
 848         int i;
 849
 850         for (i = first_index; i < first_index + count; i++)
 851                 ena_free_tx_resources(adapter, i);
 852 }
 853
 854 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
 855  * @adapter: board private structure
 856  *
 857  * Free all transmit software resources
 858  */
 859 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
 860 {
 861         ena_free_all_io_tx_resources_in_range(adapter,
 862                                               0,
 863                                               adapter->xdp_num_queues +
 864                                               adapter->num_io_queues);
 865 }
 866
 867 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
 868  * @adapter: network interface device structure
 869  * @qid: queue index
 870  *
 871  * Returns 0 on success, negative on failure
 872  */
 873 static int ena_setup_rx_resources(struct ena_adapter *adapter,
 874                                   u32 qid)
 875 {
 876         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 877         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 878         int size, node, i;
 879
 880         if (rx_ring->rx_buffer_info) {
 881                 netif_err(adapter, ifup, adapter->netdev,
 882                           "rx_buffer_info is not NULL");
 883                 return -EEXIST;
 884         }
 885
 886         /* alloc extra element so in rx path
 887          * we can always prefetch rx_info + 1
 888          */
 889         size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
 890         node = cpu_to_node(ena_irq->cpu);
 891
 892         rx_ring->rx_buffer_info = vzalloc_node(size, node);
 893         if (!rx_ring->rx_buffer_info) {
 894                 rx_ring->rx_buffer_info = vzalloc(size);
 895                 if (!rx_ring->rx_buffer_info)
 896                         return -ENOMEM;
 897         }
 898
 899         size = sizeof(u16) * rx_ring->ring_size;
 900         rx_ring->free_ids = vzalloc_node(size, node);
 901         if (!rx_ring->free_ids) {
 902                 rx_ring->free_ids = vzalloc(size);
 903                 if (!rx_ring->free_ids) {
 904                         vfree(rx_ring->rx_buffer_info);
 905                         rx_ring->rx_buffer_info = NULL;
 906                         return -ENOMEM;
 907                 }
 908         }
 909
 910         /* Req id ring for receiving RX pkts out of order */
 911         for (i = 0; i < rx_ring->ring_size; i++)
 912                 rx_ring->free_ids[i] = i;
 913
 914         /* Reset rx statistics */
 915         memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
 916
 917         rx_ring->next_to_clean = 0;
 918         rx_ring->next_to_use = 0;
 919         rx_ring->cpu = ena_irq->cpu;
 920
 921         return 0;
 922 }
 923
 924 /* ena_free_rx_resources - Free I/O Rx Resources
 925  * @adapter: network interface device structure
 926  * @qid: queue index
 927  *
 928  * Free all receive software resources
 929  */
 930 static void ena_free_rx_resources(struct ena_adapter *adapter,
 931                                   u32 qid)
 932 {
 933         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 934
 935         vfree(rx_ring->rx_buffer_info);
 936         rx_ring->rx_buffer_info = NULL;
 937
 938         vfree(rx_ring->free_ids);
 939         rx_ring->free_ids = NULL;
 940 }
 941
 942 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
 943  * @adapter: board private structure
 944  *
 945  * Return 0 on success, negative on failure
 946  */
 947 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
 948 {
 949         int i, rc = 0;
 950
 951         for (i = 0; i < adapter->num_io_queues; i++) {
 952                 rc = ena_setup_rx_resources(adapter, i);
 953                 if (rc)
 954                         goto err_setup_rx;
 955         }
 956
 957         return 0;
 958
 959 err_setup_rx:
 960
 961         netif_err(adapter, ifup, adapter->netdev,
 962                   "Rx queue %d: allocation failed\n", i);
 963
 964         /* rewind the index freeing the rings as we go */
 965         while (i--)
 966                 ena_free_rx_resources(adapter, i);
 967         return rc;
 968 }
 969
 970 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
 971  * @adapter: board private structure
 972  *
 973  * Free all receive software resources
 974  */
 975 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
 976 {
 977         int i;
 978
 979         for (i = 0; i < adapter->num_io_queues; i++)
 980                 ena_free_rx_resources(adapter, i);
 981 }
 982
 983 static int ena_alloc_rx_page(struct ena_ring *rx_ring,
 984                                     struct ena_rx_buffer *rx_info, gfp_t gfp)
 985 {
 986         int headroom = rx_ring->rx_headroom;
 987         struct ena_com_buf *ena_buf;
 988         struct page *page;
 989         dma_addr_t dma;
 990
 991         /* restore page offset value in case it has been changed by device */
 992         rx_info->page_offset = headroom;
 993
 994         /* if previous allocated page is not used */
 995         if (unlikely(rx_info->page))
 996                 return 0;
 997
 998         page = alloc_page(gfp);
 999         if (unlikely(!page)) {
1000                 ena_increase_stat(&rx_ring->rx_stats.page_alloc_fail, 1,
1001                                   &rx_ring->syncp);
1002                 return -ENOMEM;
1003         }
1004
1005         /* To enable NIC-side port-mirroring, AKA SPAN port,
1006          * we make the buffer readable from the nic as well
1007          */
1008         dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
1009                            DMA_BIDIRECTIONAL);
1010         if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
1011                 ena_increase_stat(&rx_ring->rx_stats.dma_mapping_err, 1,
1012                                   &rx_ring->syncp);
1013
1014                 __free_page(page);
1015                 return -EIO;
1016         }
1017         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1018                   "Allocate page %p, rx_info %p\n", page, rx_info);
1019
1020         rx_info->page = page;
1021         ena_buf = &rx_info->ena_buf;
1022         ena_buf->paddr = dma + headroom;
1023         ena_buf->len = ENA_PAGE_SIZE - headroom;
1024
1025         return 0;
1026 }
1027
1028 static void ena_unmap_rx_buff(struct ena_ring *rx_ring,
1029                               struct ena_rx_buffer *rx_info)
1030 {
1031         struct ena_com_buf *ena_buf = &rx_info->ena_buf;
1032
1033         dma_unmap_page(rx_ring->dev, ena_buf->paddr - rx_ring->rx_headroom,
1034                        ENA_PAGE_SIZE,
1035                        DMA_BIDIRECTIONAL);
1036 }
1037
1038 static void ena_free_rx_page(struct ena_ring *rx_ring,
1039                              struct ena_rx_buffer *rx_info)
1040 {
1041         struct page *page = rx_info->page;
1042
1043         if (unlikely(!page)) {
1044                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1045                            "Trying to free unallocated buffer\n");
1046                 return;
1047         }
1048
1049         ena_unmap_rx_buff(rx_ring, rx_info);
1050
1051         __free_page(page);
1052         rx_info->page = NULL;
1053 }
1054
1055 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
1056 {
1057         u16 next_to_use, req_id;
1058         u32 i;
1059         int rc;
1060
1061         next_to_use = rx_ring->next_to_use;
1062
1063         for (i = 0; i < num; i++) {
1064                 struct ena_rx_buffer *rx_info;
1065
1066                 req_id = rx_ring->free_ids[next_to_use];
1067
1068                 rx_info = &rx_ring->rx_buffer_info[req_id];
1069
1070                 rc = ena_alloc_rx_page(rx_ring, rx_info,
1071                                        GFP_ATOMIC | __GFP_COMP);
1072                 if (unlikely(rc < 0)) {
1073                         netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1074                                    "Failed to allocate buffer for rx queue %d\n",
1075                                    rx_ring->qid);
1076                         break;
1077                 }
1078                 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1079                                                 &rx_info->ena_buf,
1080                                                 req_id);
1081                 if (unlikely(rc)) {
1082                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1083                                    "Failed to add buffer for rx queue %d\n",
1084                                    rx_ring->qid);
1085                         break;
1086                 }
1087                 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1088                                                    rx_ring->ring_size);
1089         }
1090
1091         if (unlikely(i < num)) {
1092                 ena_increase_stat(&rx_ring->rx_stats.refil_partial, 1,
1093                                   &rx_ring->syncp);
1094                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1095                            "Refilled rx qid %d with only %d buffers (from %d)\n",
1096                            rx_ring->qid, i, num);
1097         }
1098
1099         /* ena_com_write_sq_doorbell issues a wmb() */
1100         if (likely(i))
1101                 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1102
1103         rx_ring->next_to_use = next_to_use;
1104
1105         return i;
1106 }
1107
1108 static void ena_free_rx_bufs(struct ena_adapter *adapter,
1109                              u32 qid)
1110 {
1111         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1112         u32 i;
1113
1114         for (i = 0; i < rx_ring->ring_size; i++) {
1115                 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1116
1117                 if (rx_info->page)
1118                         ena_free_rx_page(rx_ring, rx_info);
1119         }
1120 }
1121
1122 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
1123  * @adapter: board private structure
1124  */
1125 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1126 {
1127         struct ena_ring *rx_ring;
1128         int i, rc, bufs_num;
1129
1130         for (i = 0; i < adapter->num_io_queues; i++) {
1131                 rx_ring = &adapter->rx_ring[i];
1132                 bufs_num = rx_ring->ring_size - 1;
1133                 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1134
1135                 if (unlikely(rc != bufs_num))
1136                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1137                                    "Refilling Queue %d failed. allocated %d buffers from: %d\n",
1138                                    i, rc, bufs_num);
1139         }
1140 }
1141
1142 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
1143 {
1144         int i;
1145
1146         for (i = 0; i < adapter->num_io_queues; i++)
1147                 ena_free_rx_bufs(adapter, i);
1148 }
1149
1150 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
1151                               struct ena_tx_buffer *tx_info)
1152 {
1153         struct ena_com_buf *ena_buf;
1154         u32 cnt;
1155         int i;
1156
1157         ena_buf = tx_info->bufs;
1158         cnt = tx_info->num_of_bufs;
1159
1160         if (unlikely(!cnt))
1161                 return;
1162
1163         if (tx_info->map_linear_data) {
1164                 dma_unmap_single(tx_ring->dev,
1165                                  dma_unmap_addr(ena_buf, paddr),
1166                                  dma_unmap_len(ena_buf, len),
1167                                  DMA_TO_DEVICE);
1168                 ena_buf++;
1169                 cnt--;
1170         }
1171
1172         /* unmap remaining mapped pages */
1173         for (i = 0; i < cnt; i++) {
1174                 dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
1175                                dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
1176                 ena_buf++;
1177         }
1178 }
1179
1180 /* ena_free_tx_bufs - Free Tx Buffers per Queue
1181  * @tx_ring: TX ring for which buffers be freed
1182  */
1183 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
1184 {
1185         bool print_once = true;
1186         u32 i;
1187
1188         for (i = 0; i < tx_ring->ring_size; i++) {
1189                 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1190
1191                 if (!tx_info->skb)
1192                         continue;
1193
1194                 if (print_once) {
1195                         netif_notice(tx_ring->adapter, ifdown, tx_ring->netdev,
1196                                      "Free uncompleted tx skb qid %d idx 0x%x\n",
1197                                      tx_ring->qid, i);
1198                         print_once = false;
1199                 } else {
1200                         netif_dbg(tx_ring->adapter, ifdown, tx_ring->netdev,
1201                                   "Free uncompleted tx skb qid %d idx 0x%x\n",
1202                                   tx_ring->qid, i);
1203                 }
1204
1205                 ena_unmap_tx_buff(tx_ring, tx_info);
1206
1207                 dev_kfree_skb_any(tx_info->skb);
1208         }
1209         netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
1210                                                   tx_ring->qid));
1211 }
1212
1213 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
1214 {
1215         struct ena_ring *tx_ring;
1216         int i;
1217
1218         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1219                 tx_ring = &adapter->tx_ring[i];
1220                 ena_free_tx_bufs(tx_ring);
1221         }
1222 }
1223
1224 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1225 {
1226         u16 ena_qid;
1227         int i;
1228
1229         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1230                 ena_qid = ENA_IO_TXQ_IDX(i);
1231                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1232         }
1233 }
1234
1235 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1236 {
1237         u16 ena_qid;
1238         int i;
1239
1240         for (i = 0; i < adapter->num_io_queues; i++) {
1241                 ena_qid = ENA_IO_RXQ_IDX(i);
1242                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
1243                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1244         }
1245 }
1246
1247 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
1248 {
1249         ena_destroy_all_tx_queues(adapter);
1250         ena_destroy_all_rx_queues(adapter);
1251 }
1252
1253 static int handle_invalid_req_id(struct ena_ring *ring, u16 req_id,
1254                                  struct ena_tx_buffer *tx_info, bool is_xdp)
1255 {
1256         if (tx_info)
1257                 netif_err(ring->adapter,
1258                           tx_done,
1259                           ring->netdev,
1260                           "tx_info doesn't have valid %s",
1261                            is_xdp ? "xdp frame" : "skb");
1262         else
1263                 netif_err(ring->adapter,
1264                           tx_done,
1265                           ring->netdev,
1266                           "Invalid req_id: %hu\n",
1267                           req_id);
1268
1269         ena_increase_stat(&ring->tx_stats.bad_req_id, 1, &ring->syncp);
1270
1271         /* Trigger device reset */
1272         ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
1273         set_bit(ENA_FLAG_TRIGGER_RESET, &ring->adapter->flags);
1274         return -EFAULT;
1275 }
1276
1277 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
1278 {
1279         struct ena_tx_buffer *tx_info = NULL;
1280
1281         if (likely(req_id < tx_ring->ring_size)) {
1282                 tx_info = &tx_ring->tx_buffer_info[req_id];
1283                 if (likely(tx_info->skb))
1284                         return 0;
1285         }
1286
1287         return handle_invalid_req_id(tx_ring, req_id, tx_info, false);
1288 }
1289
1290 static int validate_xdp_req_id(struct ena_ring *xdp_ring, u16 req_id)
1291 {
1292         struct ena_tx_buffer *tx_info = NULL;
1293
1294         if (likely(req_id < xdp_ring->ring_size)) {
1295                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1296                 if (likely(tx_info->xdpf))
1297                         return 0;
1298         }
1299
1300         return handle_invalid_req_id(xdp_ring, req_id, tx_info, true);
1301 }
1302
1303 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
1304 {
1305         struct netdev_queue *txq;
1306         bool above_thresh;
1307         u32 tx_bytes = 0;
1308         u32 total_done = 0;
1309         u16 next_to_clean;
1310         u16 req_id;
1311         int tx_pkts = 0;
1312         int rc;
1313
1314         next_to_clean = tx_ring->next_to_clean;
1315         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
1316
1317         while (tx_pkts < budget) {
1318                 struct ena_tx_buffer *tx_info;
1319                 struct sk_buff *skb;
1320
1321                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
1322                                                 &req_id);
1323                 if (rc)
1324                         break;
1325
1326                 rc = validate_tx_req_id(tx_ring, req_id);
1327                 if (rc)
1328                         break;
1329
1330                 tx_info = &tx_ring->tx_buffer_info[req_id];
1331                 skb = tx_info->skb;
1332
1333                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
1334                 prefetch(&skb->end);
1335
1336                 tx_info->skb = NULL;
1337                 tx_info->last_jiffies = 0;
1338
1339                 ena_unmap_tx_buff(tx_ring, tx_info);
1340
1341                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1342                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
1343                           skb);
1344
1345                 tx_bytes += skb->len;
1346                 dev_kfree_skb(skb);
1347                 tx_pkts++;
1348                 total_done += tx_info->tx_descs;
1349
1350                 tx_ring->free_ids[next_to_clean] = req_id;
1351                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1352                                                      tx_ring->ring_size);
1353         }
1354
1355         tx_ring->next_to_clean = next_to_clean;
1356         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
1357         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
1358
1359         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
1360
1361         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1362                   "tx_poll: q %d done. total pkts: %d\n",
1363                   tx_ring->qid, tx_pkts);
1364
1365         /* need to make the rings circular update visible to
1366          * ena_start_xmit() before checking for netif_queue_stopped().
1367          */
1368         smp_mb();
1369
1370         above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1371                                                     ENA_TX_WAKEUP_THRESH);
1372         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
1373                 __netif_tx_lock(txq, smp_processor_id());
1374                 above_thresh =
1375                         ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1376                                                      ENA_TX_WAKEUP_THRESH);
1377                 if (netif_tx_queue_stopped(txq) && above_thresh &&
1378                     test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
1379                         netif_tx_wake_queue(txq);
1380                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
1381                                           &tx_ring->syncp);
1382                 }
1383                 __netif_tx_unlock(txq);
1384         }
1385
1386         return tx_pkts;
1387 }
1388
1389 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, bool frags)
1390 {
1391         struct sk_buff *skb;
1392
1393         if (frags)
1394                 skb = napi_get_frags(rx_ring->napi);
1395         else
1396                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1397                                                 rx_ring->rx_copybreak);
1398
1399         if (unlikely(!skb)) {
1400                 ena_increase_stat(&rx_ring->rx_stats.skb_alloc_fail, 1,
1401                                   &rx_ring->syncp);
1402                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1403                           "Failed to allocate skb. frags: %d\n", frags);
1404                 return NULL;
1405         }
1406
1407         return skb;
1408 }
1409
1410 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
1411                                   struct ena_com_rx_buf_info *ena_bufs,
1412                                   u32 descs,
1413                                   u16 *next_to_clean)
1414 {
1415         struct sk_buff *skb;
1416         struct ena_rx_buffer *rx_info;
1417         u16 len, req_id, buf = 0;
1418         void *va;
1419
1420         len = ena_bufs[buf].len;
1421         req_id = ena_bufs[buf].req_id;
1422
1423         rx_info = &rx_ring->rx_buffer_info[req_id];
1424
1425         if (unlikely(!rx_info->page)) {
1426                 netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
1427                           "Page is NULL\n");
1428                 return NULL;
1429         }
1430
1431         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1432                   "rx_info %p page %p\n",
1433                   rx_info, rx_info->page);
1434
1435         /* save virt address of first buffer */
1436         va = page_address(rx_info->page) + rx_info->page_offset;
1437
1438         prefetch(va);
1439
1440         if (len <= rx_ring->rx_copybreak) {
1441                 skb = ena_alloc_skb(rx_ring, false);
1442                 if (unlikely(!skb))
1443                         return NULL;
1444
1445                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1446                           "RX allocated small packet. len %d. data_len %d\n",
1447                           skb->len, skb->data_len);
1448
1449                 /* sync this buffer for CPU use */
1450                 dma_sync_single_for_cpu(rx_ring->dev,
1451                                         dma_unmap_addr(&rx_info->ena_buf, paddr),
1452                                         len,
1453                                         DMA_FROM_DEVICE);
1454                 skb_copy_to_linear_data(skb, va, len);
1455                 dma_sync_single_for_device(rx_ring->dev,
1456                                            dma_unmap_addr(&rx_info->ena_buf, paddr),
1457                                            len,
1458                                            DMA_FROM_DEVICE);
1459
1460                 skb_put(skb, len);
1461                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1462                 rx_ring->free_ids[*next_to_clean] = req_id;
1463                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
1464                                                      rx_ring->ring_size);
1465                 return skb;
1466         }
1467
1468         skb = ena_alloc_skb(rx_ring, true);
1469         if (unlikely(!skb))
1470                 return NULL;
1471
1472         do {
1473                 ena_unmap_rx_buff(rx_ring, rx_info);
1474
1475                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
1476                                 rx_info->page_offset, len, ENA_PAGE_SIZE);
1477
1478                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1479                           "RX skb updated. len %d. data_len %d\n",
1480                           skb->len, skb->data_len);
1481
1482                 rx_info->page = NULL;
1483
1484                 rx_ring->free_ids[*next_to_clean] = req_id;
1485                 *next_to_clean =
1486                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
1487                                              rx_ring->ring_size);
1488                 if (likely(--descs == 0))
1489                         break;
1490
1491                 buf++;
1492                 len = ena_bufs[buf].len;
1493                 req_id = ena_bufs[buf].req_id;
1494
1495                 rx_info = &rx_ring->rx_buffer_info[req_id];
1496         } while (1);
1497
1498         return skb;
1499 }
1500
1501 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
1502  * @adapter: structure containing adapter specific data
1503  * @ena_rx_ctx: received packet context/metadata
1504  * @skb: skb currently being received and modified
1505  */
1506 static void ena_rx_checksum(struct ena_ring *rx_ring,
1507                                    struct ena_com_rx_ctx *ena_rx_ctx,
1508                                    struct sk_buff *skb)
1509 {
1510         /* Rx csum disabled */
1511         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
1512                 skb->ip_summed = CHECKSUM_NONE;
1513                 return;
1514         }
1515
1516         /* For fragmented packets the checksum isn't valid */
1517         if (ena_rx_ctx->frag) {
1518                 skb->ip_summed = CHECKSUM_NONE;
1519                 return;
1520         }
1521
1522         /* if IP and error */
1523         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1524                      (ena_rx_ctx->l3_csum_err))) {
1525                 /* ipv4 checksum error */
1526                 skb->ip_summed = CHECKSUM_NONE;
1527                 ena_increase_stat(&rx_ring->rx_stats.bad_csum, 1,
1528                                   &rx_ring->syncp);
1529                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1530                           "RX IPv4 header checksum error\n");
1531                 return;
1532         }
1533
1534         /* if TCP/UDP */
1535         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1536                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
1537                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
1538                         /* TCP/UDP checksum error */
1539                         ena_increase_stat(&rx_ring->rx_stats.bad_csum, 1,
1540                                           &rx_ring->syncp);
1541                         netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1542                                   "RX L4 checksum error\n");
1543                         skb->ip_summed = CHECKSUM_NONE;
1544                         return;
1545                 }
1546
1547                 if (likely(ena_rx_ctx->l4_csum_checked)) {
1548                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1549                         ena_increase_stat(&rx_ring->rx_stats.csum_good, 1,
1550                                           &rx_ring->syncp);
1551                 } else {
1552                         ena_increase_stat(&rx_ring->rx_stats.csum_unchecked, 1,
1553                                           &rx_ring->syncp);
1554                         skb->ip_summed = CHECKSUM_NONE;
1555                 }
1556         } else {
1557                 skb->ip_summed = CHECKSUM_NONE;
1558                 return;
1559         }
1560
1561 }
1562
1563 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1564                             struct ena_com_rx_ctx *ena_rx_ctx,
1565                             struct sk_buff *skb)
1566 {
1567         enum pkt_hash_types hash_type;
1568
1569         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1570                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1571                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1572
1573                         hash_type = PKT_HASH_TYPE_L4;
1574                 else
1575                         hash_type = PKT_HASH_TYPE_NONE;
1576
1577                 /* Override hash type if the packet is fragmented */
1578                 if (ena_rx_ctx->frag)
1579                         hash_type = PKT_HASH_TYPE_NONE;
1580
1581                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1582         }
1583 }
1584
1585 static int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)
1586 {
1587         struct ena_rx_buffer *rx_info;
1588         int ret;
1589
1590         rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1591         xdp_prepare_buff(xdp, page_address(rx_info->page),
1592                          rx_info->page_offset,
1593                          rx_ring->ena_bufs[0].len, false);
1594         /* If for some reason we received a bigger packet than
1595          * we expect, then we simply drop it
1596          */
1597         if (unlikely(rx_ring->ena_bufs[0].len > ENA_XDP_MAX_MTU))
1598                 return XDP_DROP;
1599
1600         ret = ena_xdp_execute(rx_ring, xdp);
1601
1602         /* The xdp program might expand the headers */
1603         if (ret == XDP_PASS) {
1604                 rx_info->page_offset = xdp->data - xdp->data_hard_start;
1605                 rx_ring->ena_bufs[0].len = xdp->data_end - xdp->data;
1606         }
1607
1608         return ret;
1609 }
1610 /* ena_clean_rx_irq - Cleanup RX irq
1611  * @rx_ring: RX ring to clean
1612  * @napi: napi handler
1613  * @budget: how many packets driver is allowed to clean
1614  *
1615  * Returns the number of cleaned buffers.
1616  */
1617 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1618                             u32 budget)
1619 {
1620         u16 next_to_clean = rx_ring->next_to_clean;
1621         struct ena_com_rx_ctx ena_rx_ctx;
1622         struct ena_rx_buffer *rx_info;
1623         struct ena_adapter *adapter;
1624         u32 res_budget, work_done;
1625         int rx_copybreak_pkt = 0;
1626         int refill_threshold;
1627         struct sk_buff *skb;
1628         int refill_required;
1629         struct xdp_buff xdp;
1630         int xdp_flags = 0;
1631         int total_len = 0;
1632         int xdp_verdict;
1633         int rc = 0;
1634         int i;
1635
1636         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1637                   "%s qid %d\n", __func__, rx_ring->qid);
1638         res_budget = budget;
1639         xdp_init_buff(&xdp, ENA_PAGE_SIZE, &rx_ring->xdp_rxq);
1640
1641         do {
1642                 xdp_verdict = XDP_PASS;
1643                 skb = NULL;
1644                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1645                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1646                 ena_rx_ctx.descs = 0;
1647                 ena_rx_ctx.pkt_offset = 0;
1648                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1649                                     rx_ring->ena_com_io_sq,
1650                                     &ena_rx_ctx);
1651                 if (unlikely(rc))
1652                         goto error;
1653
1654                 if (unlikely(ena_rx_ctx.descs == 0))
1655                         break;
1656
1657                 /* First descriptor might have an offset set by the device */
1658                 rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1659                 rx_info->page_offset += ena_rx_ctx.pkt_offset;
1660
1661                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1662                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1663                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1664                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1665
1666                 if (ena_xdp_present_ring(rx_ring))
1667                         xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1668
1669                 /* allocate skb and fill it */
1670                 if (xdp_verdict == XDP_PASS)
1671                         skb = ena_rx_skb(rx_ring,
1672                                          rx_ring->ena_bufs,
1673                                          ena_rx_ctx.descs,
1674                                          &next_to_clean);
1675
1676                 if (unlikely(!skb)) {
1677                         for (i = 0; i < ena_rx_ctx.descs; i++) {
1678                                 int req_id = rx_ring->ena_bufs[i].req_id;
1679
1680                                 rx_ring->free_ids[next_to_clean] = req_id;
1681                                 next_to_clean =
1682                                         ENA_RX_RING_IDX_NEXT(next_to_clean,
1683                                                              rx_ring->ring_size);
1684
1685                                 /* Packets was passed for transmission, unmap it
1686                                  * from RX side.
1687                                  */
1688                                 if (xdp_verdict == XDP_TX || xdp_verdict == XDP_REDIRECT) {
1689                                         ena_unmap_rx_buff(rx_ring,
1690                                                           &rx_ring->rx_buffer_info[req_id]);
1691                                         rx_ring->rx_buffer_info[req_id].page = NULL;
1692                                 }
1693                         }
1694                         if (xdp_verdict != XDP_PASS) {
1695                                 xdp_flags |= xdp_verdict;
1696                                 res_budget--;
1697                                 continue;
1698                         }
1699                         break;
1700                 }
1701
1702                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1703
1704                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1705
1706                 skb_record_rx_queue(skb, rx_ring->qid);
1707
1708                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1709                         total_len += rx_ring->ena_bufs[0].len;
1710                         rx_copybreak_pkt++;
1711                         napi_gro_receive(napi, skb);
1712                 } else {
1713                         total_len += skb->len;
1714                         napi_gro_frags(napi);
1715                 }
1716
1717                 res_budget--;
1718         } while (likely(res_budget));
1719
1720         work_done = budget - res_budget;
1721         rx_ring->per_napi_packets += work_done;
1722         u64_stats_update_begin(&rx_ring->syncp);
1723         rx_ring->rx_stats.bytes += total_len;
1724         rx_ring->rx_stats.cnt += work_done;
1725         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1726         u64_stats_update_end(&rx_ring->syncp);
1727
1728         rx_ring->next_to_clean = next_to_clean;
1729
1730         refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
1731         refill_threshold =
1732                 min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1733                       ENA_RX_REFILL_THRESH_PACKET);
1734
1735         /* Optimization, try to batch new rx buffers */
1736         if (refill_required > refill_threshold) {
1737                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1738                 ena_refill_rx_bufs(rx_ring, refill_required);
1739         }
1740
1741         if (xdp_flags & XDP_REDIRECT)
1742                 xdp_do_flush_map();
1743
1744         return work_done;
1745
1746 error:
1747         adapter = netdev_priv(rx_ring->netdev);
1748
1749         if (rc == -ENOSPC) {
1750                 ena_increase_stat(&rx_ring->rx_stats.bad_desc_num, 1,
1751                                   &rx_ring->syncp);
1752                 adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1753         } else {
1754                 ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1,
1755                                   &rx_ring->syncp);
1756                 adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
1757         }
1758
1759         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1760
1761         return 0;
1762 }
1763
1764 static void ena_dim_work(struct work_struct *w)
1765 {
1766         struct dim *dim = container_of(w, struct dim, work);
1767         struct dim_cq_moder cur_moder =
1768                 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1769         struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1770
1771         ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1772         dim->state = DIM_START_MEASURE;
1773 }
1774
1775 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1776 {
1777         struct dim_sample dim_sample;
1778         struct ena_ring *rx_ring = ena_napi->rx_ring;
1779
1780         if (!rx_ring->per_napi_packets)
1781                 return;
1782
1783         rx_ring->non_empty_napi_events++;
1784
1785         dim_update_sample(rx_ring->non_empty_napi_events,
1786                           rx_ring->rx_stats.cnt,
1787                           rx_ring->rx_stats.bytes,
1788                           &dim_sample);
1789
1790         net_dim(&ena_napi->dim, dim_sample);
1791
1792         rx_ring->per_napi_packets = 0;
1793 }
1794
1795 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1796                                         struct ena_ring *rx_ring)
1797 {
1798         struct ena_eth_io_intr_reg intr_reg;
1799         u32 rx_interval = 0;
1800         /* Rx ring can be NULL when for XDP tx queues which don't have an
1801          * accompanying rx_ring pair.
1802          */
1803         if (rx_ring)
1804                 rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1805                         rx_ring->smoothed_interval :
1806                         ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1807
1808         /* Update intr register: rx intr delay,
1809          * tx intr delay and interrupt unmask
1810          */
1811         ena_com_update_intr_reg(&intr_reg,
1812                                 rx_interval,
1813                                 tx_ring->smoothed_interval,
1814                                 true);
1815
1816         ena_increase_stat(&tx_ring->tx_stats.unmask_interrupt, 1,
1817                           &tx_ring->syncp);
1818
1819         /* It is a shared MSI-X.
1820          * Tx and Rx CQ have pointer to it.
1821          * So we use one of them to reach the intr reg
1822          * The Tx ring is used because the rx_ring is NULL for XDP queues
1823          */
1824         ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1825 }
1826
1827 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1828                                              struct ena_ring *rx_ring)
1829 {
1830         int cpu = get_cpu();
1831         int numa_node;
1832
1833         /* Check only one ring since the 2 rings are running on the same cpu */
1834         if (likely(tx_ring->cpu == cpu))
1835                 goto out;
1836
1837         numa_node = cpu_to_node(cpu);
1838         put_cpu();
1839
1840         if (numa_node != NUMA_NO_NODE) {
1841                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1842                 if (rx_ring)
1843                         ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1844                                                  numa_node);
1845         }
1846
1847         tx_ring->cpu = cpu;
1848         if (rx_ring)
1849                 rx_ring->cpu = cpu;
1850
1851         return;
1852 out:
1853         put_cpu();
1854 }
1855
1856 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1857 {
1858         u32 total_done = 0;
1859         u16 next_to_clean;
1860         u32 tx_bytes = 0;
1861         int tx_pkts = 0;
1862         u16 req_id;
1863         int rc;
1864
1865         if (unlikely(!xdp_ring))
1866                 return 0;
1867         next_to_clean = xdp_ring->next_to_clean;
1868
1869         while (tx_pkts < budget) {
1870                 struct ena_tx_buffer *tx_info;
1871                 struct xdp_frame *xdpf;
1872
1873                 rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1874                                                 &req_id);
1875                 if (rc)
1876                         break;
1877
1878                 rc = validate_xdp_req_id(xdp_ring, req_id);
1879                 if (rc)
1880                         break;
1881
1882                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1883                 xdpf = tx_info->xdpf;
1884
1885                 tx_info->xdpf = NULL;
1886                 tx_info->last_jiffies = 0;
1887                 ena_unmap_tx_buff(xdp_ring, tx_info);
1888
1889                 netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1890                           "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1891                           xdpf);
1892
1893                 tx_bytes += xdpf->len;
1894                 tx_pkts++;
1895                 total_done += tx_info->tx_descs;
1896
1897                 xdp_return_frame(xdpf);
1898                 xdp_ring->free_ids[next_to_clean] = req_id;
1899                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1900                                                      xdp_ring->ring_size);
1901         }
1902
1903         xdp_ring->next_to_clean = next_to_clean;
1904         ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
1905         ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
1906
1907         netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1908                   "tx_poll: q %d done. total pkts: %d\n",
1909                   xdp_ring->qid, tx_pkts);
1910
1911         return tx_pkts;
1912 }
1913
1914 static int ena_io_poll(struct napi_struct *napi, int budget)
1915 {
1916         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1917         struct ena_ring *tx_ring, *rx_ring;
1918         int tx_work_done;
1919         int rx_work_done = 0;
1920         int tx_budget;
1921         int napi_comp_call = 0;
1922         int ret;
1923
1924         tx_ring = ena_napi->tx_ring;
1925         rx_ring = ena_napi->rx_ring;
1926
1927         tx_ring->first_interrupt = ena_napi->first_interrupt;
1928         rx_ring->first_interrupt = ena_napi->first_interrupt;
1929
1930         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1931
1932         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1933             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1934                 napi_complete_done(napi, 0);
1935                 return 0;
1936         }
1937
1938         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1939         /* On netpoll the budget is zero and the handler should only clean the
1940          * tx completions.
1941          */
1942         if (likely(budget))
1943                 rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1944
1945         /* If the device is about to reset or down, avoid unmask
1946          * the interrupt and return 0 so NAPI won't reschedule
1947          */
1948         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1949                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1950                 napi_complete_done(napi, 0);
1951                 ret = 0;
1952
1953         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1954                 napi_comp_call = 1;
1955
1956                 /* Update numa and unmask the interrupt only when schedule
1957                  * from the interrupt context (vs from sk_busy_loop)
1958                  */
1959                 if (napi_complete_done(napi, rx_work_done) &&
1960                     READ_ONCE(ena_napi->interrupts_masked)) {
1961                         smp_rmb(); /* make sure interrupts_masked is read */
1962                         WRITE_ONCE(ena_napi->interrupts_masked, false);
1963                         /* We apply adaptive moderation on Rx path only.
1964                          * Tx uses static interrupt moderation.
1965                          */
1966                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1967                                 ena_adjust_adaptive_rx_intr_moderation(ena_napi);
1968
1969                         ena_unmask_interrupt(tx_ring, rx_ring);
1970                 }
1971
1972                 ena_update_ring_numa_node(tx_ring, rx_ring);
1973
1974                 ret = rx_work_done;
1975         } else {
1976                 ret = budget;
1977         }
1978
1979         u64_stats_update_begin(&tx_ring->syncp);
1980         tx_ring->tx_stats.napi_comp += napi_comp_call;
1981         tx_ring->tx_stats.tx_poll++;
1982         u64_stats_update_end(&tx_ring->syncp);
1983
1984         return ret;
1985 }
1986
1987 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1988 {
1989         struct ena_adapter *adapter = (struct ena_adapter *)data;
1990
1991         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1992
1993         /* Don't call the aenq handler before probe is done */
1994         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1995                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
1996
1997         return IRQ_HANDLED;
1998 }
1999
2000 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
2001  * @irq: interrupt number
2002  * @data: pointer to a network interface private napi device structure
2003  */
2004 static irqreturn_t ena_intr_msix_io(int irq, void *data)
2005 {
2006         struct ena_napi *ena_napi = data;
2007
2008         ena_napi->first_interrupt = true;
2009
2010         WRITE_ONCE(ena_napi->interrupts_masked, true);
2011         smp_wmb(); /* write interrupts_masked before calling napi */
2012
2013         napi_schedule_irqoff(&ena_napi->napi);
2014
2015         return IRQ_HANDLED;
2016 }
2017
2018 /* Reserve a single MSI-X vector for management (admin + aenq).
2019  * plus reserve one vector for each potential io queue.
2020  * the number of potential io queues is the minimum of what the device
2021  * supports and the number of vCPUs.
2022  */
2023 static int ena_enable_msix(struct ena_adapter *adapter)
2024 {
2025         int msix_vecs, irq_cnt;
2026
2027         if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2028                 netif_err(adapter, probe, adapter->netdev,
2029                           "Error, MSI-X is already enabled\n");
2030                 return -EPERM;
2031         }
2032
2033         /* Reserved the max msix vectors we might need */
2034         msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
2035         netif_dbg(adapter, probe, adapter->netdev,
2036                   "Trying to enable MSI-X, vectors %d\n", msix_vecs);
2037
2038         irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
2039                                         msix_vecs, PCI_IRQ_MSIX);
2040
2041         if (irq_cnt < 0) {
2042                 netif_err(adapter, probe, adapter->netdev,
2043                           "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
2044                 return -ENOSPC;
2045         }
2046
2047         if (irq_cnt != msix_vecs) {
2048                 netif_notice(adapter, probe, adapter->netdev,
2049                              "Enable only %d MSI-X (out of %d), reduce the number of queues\n",
2050                              irq_cnt, msix_vecs);
2051                 adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
2052         }
2053
2054         if (ena_init_rx_cpu_rmap(adapter))
2055                 netif_warn(adapter, probe, adapter->netdev,
2056                            "Failed to map IRQs to CPUs\n");
2057
2058         adapter->msix_vecs = irq_cnt;
2059         set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2060
2061         return 0;
2062 }
2063
2064 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2065 {
2066         u32 cpu;
2067
2068         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2069                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2070                  pci_name(adapter->pdev));
2071         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2072                 ena_intr_msix_mgmnt;
2073         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2074         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2075                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2076         cpu = cpumask_first(cpu_online_mask);
2077         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2078         cpumask_set_cpu(cpu,
2079                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2080 }
2081
2082 static void ena_setup_io_intr(struct ena_adapter *adapter)
2083 {
2084         struct net_device *netdev;
2085         int irq_idx, i, cpu;
2086         int io_queue_count;
2087
2088         netdev = adapter->netdev;
2089         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2090
2091         for (i = 0; i < io_queue_count; i++) {
2092                 irq_idx = ENA_IO_IRQ_IDX(i);
2093                 cpu = i % num_online_cpus();
2094
2095                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2096                          "%s-Tx-Rx-%d", netdev->name, i);
2097                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2098                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2099                 adapter->irq_tbl[irq_idx].vector =
2100                         pci_irq_vector(adapter->pdev, irq_idx);
2101                 adapter->irq_tbl[irq_idx].cpu = cpu;
2102
2103                 cpumask_set_cpu(cpu,
2104                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
2105         }
2106 }
2107
2108 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2109 {
2110         unsigned long flags = 0;
2111         struct ena_irq *irq;
2112         int rc;
2113
2114         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2115         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2116                          irq->data);
2117         if (rc) {
2118                 netif_err(adapter, probe, adapter->netdev,
2119                           "Failed to request admin irq\n");
2120                 return rc;
2121         }
2122
2123         netif_dbg(adapter, probe, adapter->netdev,
2124                   "Set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2125                   irq->affinity_hint_mask.bits[0], irq->vector);
2126
2127         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2128
2129         return rc;
2130 }
2131
2132 static int ena_request_io_irq(struct ena_adapter *adapter)
2133 {
2134         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2135         unsigned long flags = 0;
2136         struct ena_irq *irq;
2137         int rc = 0, i, k;
2138
2139         if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2140                 netif_err(adapter, ifup, adapter->netdev,
2141                           "Failed to request I/O IRQ: MSI-X is not enabled\n");
2142                 return -EINVAL;
2143         }
2144
2145         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2146                 irq = &adapter->irq_tbl[i];
2147                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2148                                  irq->data);
2149                 if (rc) {
2150                         netif_err(adapter, ifup, adapter->netdev,
2151                                   "Failed to request I/O IRQ. index %d rc %d\n",
2152                                    i, rc);
2153                         goto err;
2154                 }
2155
2156                 netif_dbg(adapter, ifup, adapter->netdev,
2157                           "Set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2158                           i, irq->affinity_hint_mask.bits[0], irq->vector);
2159
2160                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2161         }
2162
2163         return rc;
2164
2165 err:
2166         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2167                 irq = &adapter->irq_tbl[k];
2168                 free_irq(irq->vector, irq->data);
2169         }
2170
2171         return rc;
2172 }
2173
2174 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2175 {
2176         struct ena_irq *irq;
2177
2178         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2179         synchronize_irq(irq->vector);
2180         irq_set_affinity_hint(irq->vector, NULL);
2181         free_irq(irq->vector, irq->data);
2182 }
2183
2184 static void ena_free_io_irq(struct ena_adapter *adapter)
2185 {
2186         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2187         struct ena_irq *irq;
2188         int i;
2189
2190 #ifdef CONFIG_RFS_ACCEL
2191         if (adapter->msix_vecs >= 1) {
2192                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2193                 adapter->netdev->rx_cpu_rmap = NULL;
2194         }
2195 #endif /* CONFIG_RFS_ACCEL */
2196
2197         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2198                 irq = &adapter->irq_tbl[i];
2199                 irq_set_affinity_hint(irq->vector, NULL);
2200                 free_irq(irq->vector, irq->data);
2201         }
2202 }
2203
2204 static void ena_disable_msix(struct ena_adapter *adapter)
2205 {
2206         if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2207                 pci_free_irq_vectors(adapter->pdev);
2208 }
2209
2210 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2211 {
2212         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2213         int i;
2214
2215         if (!netif_running(adapter->netdev))
2216                 return;
2217
2218         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2219                 synchronize_irq(adapter->irq_tbl[i].vector);
2220 }
2221
2222 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2223                                   int first_index,
2224                                   int count)
2225 {
2226         int i;
2227
2228         for (i = first_index; i < first_index + count; i++) {
2229                 netif_napi_del(&adapter->ena_napi[i].napi);
2230
2231                 WARN_ON(!ENA_IS_XDP_INDEX(adapter, i) &&
2232                         adapter->ena_napi[i].xdp_ring);
2233         }
2234 }
2235
2236 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2237                                    int first_index, int count)
2238 {
2239         int i;
2240
2241         for (i = first_index; i < first_index + count; i++) {
2242                 struct ena_napi *napi = &adapter->ena_napi[i];
2243
2244                 netif_napi_add(adapter->netdev,
2245                                &napi->napi,
2246                                ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll,
2247                                ENA_NAPI_BUDGET);
2248
2249                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
2250                         napi->rx_ring = &adapter->rx_ring[i];
2251                         napi->tx_ring = &adapter->tx_ring[i];
2252                 } else {
2253                         napi->xdp_ring = &adapter->tx_ring[i];
2254                 }
2255                 napi->qid = i;
2256         }
2257 }
2258
2259 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2260                                       int first_index,
2261                                       int count)
2262 {
2263         int i;
2264
2265         for (i = first_index; i < first_index + count; i++)
2266                 napi_disable(&adapter->ena_napi[i].napi);
2267 }
2268
2269 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2270                                      int first_index,
2271                                      int count)
2272 {
2273         int i;
2274
2275         for (i = first_index; i < first_index + count; i++)
2276                 napi_enable(&adapter->ena_napi[i].napi);
2277 }
2278
2279 /* Configure the Rx forwarding */
2280 static int ena_rss_configure(struct ena_adapter *adapter)
2281 {
2282         struct ena_com_dev *ena_dev = adapter->ena_dev;
2283         int rc;
2284
2285         /* In case the RSS table wasn't initialized by probe */
2286         if (!ena_dev->rss.tbl_log_size) {
2287                 rc = ena_rss_init_default(adapter);
2288                 if (rc && (rc != -EOPNOTSUPP)) {
2289                         netif_err(adapter, ifup, adapter->netdev,
2290                                   "Failed to init RSS rc: %d\n", rc);
2291                         return rc;
2292                 }
2293         }
2294
2295         /* Set indirect table */
2296         rc = ena_com_indirect_table_set(ena_dev);
2297         if (unlikely(rc && rc != -EOPNOTSUPP))
2298                 return rc;
2299
2300         /* Configure hash function (if supported) */
2301         rc = ena_com_set_hash_function(ena_dev);
2302         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2303                 return rc;
2304
2305         /* Configure hash inputs (if supported) */
2306         rc = ena_com_set_hash_ctrl(ena_dev);
2307         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2308                 return rc;
2309
2310         return 0;
2311 }
2312
2313 static int ena_up_complete(struct ena_adapter *adapter)
2314 {
2315         int rc;
2316
2317         rc = ena_rss_configure(adapter);
2318         if (rc)
2319                 return rc;
2320
2321         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2322
2323         ena_refill_all_rx_bufs(adapter);
2324
2325         /* enable transmits */
2326         netif_tx_start_all_queues(adapter->netdev);
2327
2328         ena_napi_enable_in_range(adapter,
2329                                  0,
2330                                  adapter->xdp_num_queues + adapter->num_io_queues);
2331
2332         return 0;
2333 }
2334
2335 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2336 {
2337         struct ena_com_create_io_ctx ctx;
2338         struct ena_com_dev *ena_dev;
2339         struct ena_ring *tx_ring;
2340         u32 msix_vector;
2341         u16 ena_qid;
2342         int rc;
2343
2344         ena_dev = adapter->ena_dev;
2345
2346         tx_ring = &adapter->tx_ring[qid];
2347         msix_vector = ENA_IO_IRQ_IDX(qid);
2348         ena_qid = ENA_IO_TXQ_IDX(qid);
2349
2350         memset(&ctx, 0x0, sizeof(ctx));
2351
2352         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2353         ctx.qid = ena_qid;
2354         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2355         ctx.msix_vector = msix_vector;
2356         ctx.queue_size = tx_ring->ring_size;
2357         ctx.numa_node = cpu_to_node(tx_ring->cpu);
2358
2359         rc = ena_com_create_io_queue(ena_dev, &ctx);
2360         if (rc) {
2361                 netif_err(adapter, ifup, adapter->netdev,
2362                           "Failed to create I/O TX queue num %d rc: %d\n",
2363                           qid, rc);
2364                 return rc;
2365         }
2366
2367         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2368                                      &tx_ring->ena_com_io_sq,
2369                                      &tx_ring->ena_com_io_cq);
2370         if (rc) {
2371                 netif_err(adapter, ifup, adapter->netdev,
2372                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2373                           qid, rc);
2374                 ena_com_destroy_io_queue(ena_dev, ena_qid);
2375                 return rc;
2376         }
2377
2378         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2379         return rc;
2380 }
2381
2382 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2383                                             int first_index, int count)
2384 {
2385         struct ena_com_dev *ena_dev = adapter->ena_dev;
2386         int rc, i;
2387
2388         for (i = first_index; i < first_index + count; i++) {
2389                 rc = ena_create_io_tx_queue(adapter, i);
2390                 if (rc)
2391                         goto create_err;
2392         }
2393
2394         return 0;
2395
2396 create_err:
2397         while (i-- > first_index)
2398                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2399
2400         return rc;
2401 }
2402
2403 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2404 {
2405         struct ena_com_dev *ena_dev;
2406         struct ena_com_create_io_ctx ctx;
2407         struct ena_ring *rx_ring;
2408         u32 msix_vector;
2409         u16 ena_qid;
2410         int rc;
2411
2412         ena_dev = adapter->ena_dev;
2413
2414         rx_ring = &adapter->rx_ring[qid];
2415         msix_vector = ENA_IO_IRQ_IDX(qid);
2416         ena_qid = ENA_IO_RXQ_IDX(qid);
2417
2418         memset(&ctx, 0x0, sizeof(ctx));
2419
2420         ctx.qid = ena_qid;
2421         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2422         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2423         ctx.msix_vector = msix_vector;
2424         ctx.queue_size = rx_ring->ring_size;
2425         ctx.numa_node = cpu_to_node(rx_ring->cpu);
2426
2427         rc = ena_com_create_io_queue(ena_dev, &ctx);
2428         if (rc) {
2429                 netif_err(adapter, ifup, adapter->netdev,
2430                           "Failed to create I/O RX queue num %d rc: %d\n",
2431                           qid, rc);
2432                 return rc;
2433         }
2434
2435         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2436                                      &rx_ring->ena_com_io_sq,
2437                                      &rx_ring->ena_com_io_cq);
2438         if (rc) {
2439                 netif_err(adapter, ifup, adapter->netdev,
2440                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2441                           qid, rc);
2442                 goto err;
2443         }
2444
2445         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2446
2447         return rc;
2448 err:
2449         ena_com_destroy_io_queue(ena_dev, ena_qid);
2450         return rc;
2451 }
2452
2453 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2454 {
2455         struct ena_com_dev *ena_dev = adapter->ena_dev;
2456         int rc, i;
2457
2458         for (i = 0; i < adapter->num_io_queues; i++) {
2459                 rc = ena_create_io_rx_queue(adapter, i);
2460                 if (rc)
2461                         goto create_err;
2462                 INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2463         }
2464
2465         return 0;
2466
2467 create_err:
2468         while (i--) {
2469                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
2470                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2471         }
2472
2473         return rc;
2474 }
2475
2476 static void set_io_rings_size(struct ena_adapter *adapter,
2477                               int new_tx_size,
2478                               int new_rx_size)
2479 {
2480         int i;
2481
2482         for (i = 0; i < adapter->num_io_queues; i++) {
2483                 adapter->tx_ring[i].ring_size = new_tx_size;
2484                 adapter->rx_ring[i].ring_size = new_rx_size;
2485         }
2486 }
2487
2488 /* This function allows queue allocation to backoff when the system is
2489  * low on memory. If there is not enough memory to allocate io queues
2490  * the driver will try to allocate smaller queues.
2491  *
2492  * The backoff algorithm is as follows:
2493  *  1. Try to allocate TX and RX and if successful.
2494  *  1.1. return success
2495  *
2496  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2497  *
2498  *  3. If TX or RX is smaller than 256
2499  *  3.1. return failure.
2500  *  4. else
2501  *  4.1. go back to 1.
2502  */
2503 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2504 {
2505         int rc, cur_rx_ring_size, cur_tx_ring_size;
2506         int new_rx_ring_size, new_tx_ring_size;
2507
2508         /* current queue sizes might be set to smaller than the requested
2509          * ones due to past queue allocation failures.
2510          */
2511         set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2512                           adapter->requested_rx_ring_size);
2513
2514         while (1) {
2515                 if (ena_xdp_present(adapter)) {
2516                         rc = ena_setup_and_create_all_xdp_queues(adapter);
2517
2518                         if (rc)
2519                                 goto err_setup_tx;
2520                 }
2521                 rc = ena_setup_tx_resources_in_range(adapter,
2522                                                      0,
2523                                                      adapter->num_io_queues);
2524                 if (rc)
2525                         goto err_setup_tx;
2526
2527                 rc = ena_create_io_tx_queues_in_range(adapter,
2528                                                       0,
2529                                                       adapter->num_io_queues);
2530                 if (rc)
2531                         goto err_create_tx_queues;
2532
2533                 rc = ena_setup_all_rx_resources(adapter);
2534                 if (rc)
2535                         goto err_setup_rx;
2536
2537                 rc = ena_create_all_io_rx_queues(adapter);
2538                 if (rc)
2539                         goto err_create_rx_queues;
2540
2541                 return 0;
2542
2543 err_create_rx_queues:
2544                 ena_free_all_io_rx_resources(adapter);
2545 err_setup_rx:
2546                 ena_destroy_all_tx_queues(adapter);
2547 err_create_tx_queues:
2548                 ena_free_all_io_tx_resources(adapter);
2549 err_setup_tx:
2550                 if (rc != -ENOMEM) {
2551                         netif_err(adapter, ifup, adapter->netdev,
2552                                   "Queue creation failed with error code %d\n",
2553                                   rc);
2554                         return rc;
2555                 }
2556
2557                 cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2558                 cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2559
2560                 netif_err(adapter, ifup, adapter->netdev,
2561                           "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2562                           cur_tx_ring_size, cur_rx_ring_size);
2563
2564                 new_tx_ring_size = cur_tx_ring_size;
2565                 new_rx_ring_size = cur_rx_ring_size;
2566
2567                 /* Decrease the size of the larger queue, or
2568                  * decrease both if they are the same size.
2569                  */
2570                 if (cur_rx_ring_size <= cur_tx_ring_size)
2571                         new_tx_ring_size = cur_tx_ring_size / 2;
2572                 if (cur_rx_ring_size >= cur_tx_ring_size)
2573                         new_rx_ring_size = cur_rx_ring_size / 2;
2574
2575                 if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2576                     new_rx_ring_size < ENA_MIN_RING_SIZE) {
2577                         netif_err(adapter, ifup, adapter->netdev,
2578                                   "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2579                                   ENA_MIN_RING_SIZE);
2580                         return rc;
2581                 }
2582
2583                 netif_err(adapter, ifup, adapter->netdev,
2584                           "Retrying queue creation with sizes TX=%d, RX=%d\n",
2585                           new_tx_ring_size,
2586                           new_rx_ring_size);
2587
2588                 set_io_rings_size(adapter, new_tx_ring_size,
2589                                   new_rx_ring_size);
2590         }
2591 }
2592
2593 static int ena_up(struct ena_adapter *adapter)
2594 {
2595         int io_queue_count, rc, i;
2596
2597         netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
2598
2599         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2600         ena_setup_io_intr(adapter);
2601
2602         /* napi poll functions should be initialized before running
2603          * request_irq(), to handle a rare condition where there is a pending
2604          * interrupt, causing the ISR to fire immediately while the poll
2605          * function wasn't set yet, causing a null dereference
2606          */
2607         ena_init_napi_in_range(adapter, 0, io_queue_count);
2608
2609         rc = ena_request_io_irq(adapter);
2610         if (rc)
2611                 goto err_req_irq;
2612
2613         rc = create_queues_with_size_backoff(adapter);
2614         if (rc)
2615                 goto err_create_queues_with_backoff;
2616
2617         rc = ena_up_complete(adapter);
2618         if (rc)
2619                 goto err_up;
2620
2621         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2622                 netif_carrier_on(adapter->netdev);
2623
2624         ena_increase_stat(&adapter->dev_stats.interface_up, 1,
2625                           &adapter->syncp);
2626
2627         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2628
2629         /* Enable completion queues interrupt */
2630         for (i = 0; i < adapter->num_io_queues; i++)
2631                 ena_unmask_interrupt(&adapter->tx_ring[i],
2632                                      &adapter->rx_ring[i]);
2633
2634         /* schedule napi in case we had pending packets
2635          * from the last time we disable napi
2636          */
2637         for (i = 0; i < io_queue_count; i++)
2638                 napi_schedule(&adapter->ena_napi[i].napi);
2639
2640         return rc;
2641
2642 err_up:
2643         ena_destroy_all_tx_queues(adapter);
2644         ena_free_all_io_tx_resources(adapter);
2645         ena_destroy_all_rx_queues(adapter);
2646         ena_free_all_io_rx_resources(adapter);
2647 err_create_queues_with_backoff:
2648         ena_free_io_irq(adapter);
2649 err_req_irq:
2650         ena_del_napi_in_range(adapter, 0, io_queue_count);
2651
2652         return rc;
2653 }
2654
2655 static void ena_down(struct ena_adapter *adapter)
2656 {
2657         int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2658
2659         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2660
2661         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2662
2663         ena_increase_stat(&adapter->dev_stats.interface_down, 1,
2664                           &adapter->syncp);
2665
2666         netif_carrier_off(adapter->netdev);
2667         netif_tx_disable(adapter->netdev);
2668
2669         /* After this point the napi handler won't enable the tx queue */
2670         ena_napi_disable_in_range(adapter, 0, io_queue_count);
2671
2672         /* After destroy the queue there won't be any new interrupts */
2673
2674         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2675                 int rc;
2676
2677                 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2678                 if (rc)
2679                         netif_err(adapter, ifdown, adapter->netdev,
2680                                   "Device reset failed\n");
2681                 /* stop submitting admin commands on a device that was reset */
2682                 ena_com_set_admin_running_state(adapter->ena_dev, false);
2683         }
2684
2685         ena_destroy_all_io_queues(adapter);
2686
2687         ena_disable_io_intr_sync(adapter);
2688         ena_free_io_irq(adapter);
2689         ena_del_napi_in_range(adapter, 0, io_queue_count);
2690
2691         ena_free_all_tx_bufs(adapter);
2692         ena_free_all_rx_bufs(adapter);
2693         ena_free_all_io_tx_resources(adapter);
2694         ena_free_all_io_rx_resources(adapter);
2695 }
2696
2697 /* ena_open - Called when a network interface is made active
2698  * @netdev: network interface device structure
2699  *
2700  * Returns 0 on success, negative value on failure
2701  *
2702  * The open entry point is called when a network interface is made
2703  * active by the system (IFF_UP).  At this point all resources needed
2704  * for transmit and receive operations are allocated, the interrupt
2705  * handler is registered with the OS, the watchdog timer is started,
2706  * and the stack is notified that the interface is ready.
2707  */
2708 static int ena_open(struct net_device *netdev)
2709 {
2710         struct ena_adapter *adapter = netdev_priv(netdev);
2711         int rc;
2712
2713         /* Notify the stack of the actual queue counts. */
2714         rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2715         if (rc) {
2716                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2717                 return rc;
2718         }
2719
2720         rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2721         if (rc) {
2722                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2723                 return rc;
2724         }
2725
2726         rc = ena_up(adapter);
2727         if (rc)
2728                 return rc;
2729
2730         return rc;
2731 }
2732
2733 /* ena_close - Disables a network interface
2734  * @netdev: network interface device structure
2735  *
2736  * Returns 0, this is not allowed to fail
2737  *
2738  * The close entry point is called when an interface is de-activated
2739  * by the OS.  The hardware is still under the drivers control, but
2740  * needs to be disabled.  A global MAC reset is issued to stop the
2741  * hardware, and all transmit and receive resources are freed.
2742  */
2743 static int ena_close(struct net_device *netdev)
2744 {
2745         struct ena_adapter *adapter = netdev_priv(netdev);
2746
2747         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2748
2749         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2750                 return 0;
2751
2752         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2753                 ena_down(adapter);
2754
2755         /* Check for device status and issue reset if needed*/
2756         check_for_admin_com_state(adapter);
2757         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2758                 netif_err(adapter, ifdown, adapter->netdev,
2759                           "Destroy failure, restarting device\n");
2760                 ena_dump_stats_to_dmesg(adapter);
2761                 /* rtnl lock already obtained in dev_ioctl() layer */
2762                 ena_destroy_device(adapter, false);
2763                 ena_restore_device(adapter);
2764         }
2765
2766         return 0;
2767 }
2768
2769 int ena_update_queue_sizes(struct ena_adapter *adapter,
2770                            u32 new_tx_size,
2771                            u32 new_rx_size)
2772 {
2773         bool dev_was_up;
2774
2775         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2776         ena_close(adapter->netdev);
2777         adapter->requested_tx_ring_size = new_tx_size;
2778         adapter->requested_rx_ring_size = new_rx_size;
2779         ena_init_io_rings(adapter,
2780                           0,
2781                           adapter->xdp_num_queues +
2782                           adapter->num_io_queues);
2783         return dev_was_up ? ena_up(adapter) : 0;
2784 }
2785
2786 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2787 {
2788         struct ena_com_dev *ena_dev = adapter->ena_dev;
2789         int prev_channel_count;
2790         bool dev_was_up;
2791
2792         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2793         ena_close(adapter->netdev);
2794         prev_channel_count = adapter->num_io_queues;
2795         adapter->num_io_queues = new_channel_count;
2796         if (ena_xdp_present(adapter) &&
2797             ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2798                 adapter->xdp_first_ring = new_channel_count;
2799                 adapter->xdp_num_queues = new_channel_count;
2800                 if (prev_channel_count > new_channel_count)
2801                         ena_xdp_exchange_program_rx_in_range(adapter,
2802                                                              NULL,
2803                                                              new_channel_count,
2804                                                              prev_channel_count);
2805                 else
2806                         ena_xdp_exchange_program_rx_in_range(adapter,
2807                                                              adapter->xdp_bpf_prog,
2808                                                              prev_channel_count,
2809                                                              new_channel_count);
2810         }
2811
2812         /* We need to destroy the rss table so that the indirection
2813          * table will be reinitialized by ena_up()
2814          */
2815         ena_com_rss_destroy(ena_dev);
2816         ena_init_io_rings(adapter,
2817                           0,
2818                           adapter->xdp_num_queues +
2819                           adapter->num_io_queues);
2820         return dev_was_up ? ena_open(adapter->netdev) : 0;
2821 }
2822
2823 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx,
2824                         struct sk_buff *skb,
2825                         bool disable_meta_caching)
2826 {
2827         u32 mss = skb_shinfo(skb)->gso_size;
2828         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2829         u8 l4_protocol = 0;
2830
2831         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2832                 ena_tx_ctx->l4_csum_enable = 1;
2833                 if (mss) {
2834                         ena_tx_ctx->tso_enable = 1;
2835                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2836                         ena_tx_ctx->l4_csum_partial = 0;
2837                 } else {
2838                         ena_tx_ctx->tso_enable = 0;
2839                         ena_meta->l4_hdr_len = 0;
2840                         ena_tx_ctx->l4_csum_partial = 1;
2841                 }
2842
2843                 switch (ip_hdr(skb)->version) {
2844                 case IPVERSION:
2845                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2846                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
2847                                 ena_tx_ctx->df = 1;
2848                         if (mss)
2849                                 ena_tx_ctx->l3_csum_enable = 1;
2850                         l4_protocol = ip_hdr(skb)->protocol;
2851                         break;
2852                 case 6:
2853                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2854                         l4_protocol = ipv6_hdr(skb)->nexthdr;
2855                         break;
2856                 default:
2857                         break;
2858                 }
2859
2860                 if (l4_protocol == IPPROTO_TCP)
2861                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2862                 else
2863                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2864
2865                 ena_meta->mss = mss;
2866                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
2867                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
2868                 ena_tx_ctx->meta_valid = 1;
2869         } else if (disable_meta_caching) {
2870                 memset(ena_meta, 0, sizeof(*ena_meta));
2871                 ena_tx_ctx->meta_valid = 1;
2872         } else {
2873                 ena_tx_ctx->meta_valid = 0;
2874         }
2875 }
2876
2877 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2878                                        struct sk_buff *skb)
2879 {
2880         int num_frags, header_len, rc;
2881
2882         num_frags = skb_shinfo(skb)->nr_frags;
2883         header_len = skb_headlen(skb);
2884
2885         if (num_frags < tx_ring->sgl_size)
2886                 return 0;
2887
2888         if ((num_frags == tx_ring->sgl_size) &&
2889             (header_len < tx_ring->tx_max_header_size))
2890                 return 0;
2891
2892         ena_increase_stat(&tx_ring->tx_stats.linearize, 1, &tx_ring->syncp);
2893
2894         rc = skb_linearize(skb);
2895         if (unlikely(rc)) {
2896                 ena_increase_stat(&tx_ring->tx_stats.linearize_failed, 1,
2897                                   &tx_ring->syncp);
2898         }
2899
2900         return rc;
2901 }
2902
2903 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2904                           struct ena_tx_buffer *tx_info,
2905                           struct sk_buff *skb,
2906                           void **push_hdr,
2907                           u16 *header_len)
2908 {
2909         struct ena_adapter *adapter = tx_ring->adapter;
2910         struct ena_com_buf *ena_buf;
2911         dma_addr_t dma;
2912         u32 skb_head_len, frag_len, last_frag;
2913         u16 push_len = 0;
2914         u16 delta = 0;
2915         int i = 0;
2916
2917         skb_head_len = skb_headlen(skb);
2918         tx_info->skb = skb;
2919         ena_buf = tx_info->bufs;
2920
2921         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2922                 /* When the device is LLQ mode, the driver will copy
2923                  * the header into the device memory space.
2924                  * the ena_com layer assume the header is in a linear
2925                  * memory space.
2926                  * This assumption might be wrong since part of the header
2927                  * can be in the fragmented buffers.
2928                  * Use skb_header_pointer to make sure the header is in a
2929                  * linear memory space.
2930                  */
2931
2932                 push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2933                 *push_hdr = skb_header_pointer(skb, 0, push_len,
2934                                                tx_ring->push_buf_intermediate_buf);
2935                 *header_len = push_len;
2936                 if (unlikely(skb->data != *push_hdr)) {
2937                         ena_increase_stat(&tx_ring->tx_stats.llq_buffer_copy, 1,
2938                                           &tx_ring->syncp);
2939
2940                         delta = push_len - skb_head_len;
2941                 }
2942         } else {
2943                 *push_hdr = NULL;
2944                 *header_len = min_t(u32, skb_head_len,
2945                                     tx_ring->tx_max_header_size);
2946         }
2947
2948         netif_dbg(adapter, tx_queued, adapter->netdev,
2949                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2950                   *push_hdr, push_len);
2951
2952         if (skb_head_len > push_len) {
2953                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2954                                      skb_head_len - push_len, DMA_TO_DEVICE);
2955                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2956                         goto error_report_dma_error;
2957
2958                 ena_buf->paddr = dma;
2959                 ena_buf->len = skb_head_len - push_len;
2960
2961                 ena_buf++;
2962                 tx_info->num_of_bufs++;
2963                 tx_info->map_linear_data = 1;
2964         } else {
2965                 tx_info->map_linear_data = 0;
2966         }
2967
2968         last_frag = skb_shinfo(skb)->nr_frags;
2969
2970         for (i = 0; i < last_frag; i++) {
2971                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2972
2973                 frag_len = skb_frag_size(frag);
2974
2975                 if (unlikely(delta >= frag_len)) {
2976                         delta -= frag_len;
2977                         continue;
2978                 }
2979
2980                 dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
2981                                        frag_len - delta, DMA_TO_DEVICE);
2982                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2983                         goto error_report_dma_error;
2984
2985                 ena_buf->paddr = dma;
2986                 ena_buf->len = frag_len - delta;
2987                 ena_buf++;
2988                 tx_info->num_of_bufs++;
2989                 delta = 0;
2990         }
2991
2992         return 0;
2993
2994 error_report_dma_error:
2995         ena_increase_stat(&tx_ring->tx_stats.dma_mapping_err, 1,
2996                           &tx_ring->syncp);
2997         netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map skb\n");
2998
2999         tx_info->skb = NULL;
3000
3001         tx_info->num_of_bufs += i;
3002         ena_unmap_tx_buff(tx_ring, tx_info);
3003
3004         return -EINVAL;
3005 }
3006
3007 /* Called with netif_tx_lock. */
3008 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
3009 {
3010         struct ena_adapter *adapter = netdev_priv(dev);
3011         struct ena_tx_buffer *tx_info;
3012         struct ena_com_tx_ctx ena_tx_ctx;
3013         struct ena_ring *tx_ring;
3014         struct netdev_queue *txq;
3015         void *push_hdr;
3016         u16 next_to_use, req_id, header_len;
3017         int qid, rc;
3018
3019         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
3020         /*  Determine which tx ring we will be placed on */
3021         qid = skb_get_queue_mapping(skb);
3022         tx_ring = &adapter->tx_ring[qid];
3023         txq = netdev_get_tx_queue(dev, qid);
3024
3025         rc = ena_check_and_linearize_skb(tx_ring, skb);
3026         if (unlikely(rc))
3027                 goto error_drop_packet;
3028
3029         skb_tx_timestamp(skb);
3030
3031         next_to_use = tx_ring->next_to_use;
3032         req_id = tx_ring->free_ids[next_to_use];
3033         tx_info = &tx_ring->tx_buffer_info[req_id];
3034         tx_info->num_of_bufs = 0;
3035
3036         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
3037
3038         rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
3039         if (unlikely(rc))
3040                 goto error_drop_packet;
3041
3042         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
3043         ena_tx_ctx.ena_bufs = tx_info->bufs;
3044         ena_tx_ctx.push_header = push_hdr;
3045         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
3046         ena_tx_ctx.req_id = req_id;
3047         ena_tx_ctx.header_len = header_len;
3048
3049         /* set flags and meta data */
3050         ena_tx_csum(&ena_tx_ctx, skb, tx_ring->disable_meta_caching);
3051
3052         rc = ena_xmit_common(dev,
3053                              tx_ring,
3054                              tx_info,
3055                              &ena_tx_ctx,
3056                              next_to_use,
3057                              skb->len);
3058         if (rc)
3059                 goto error_unmap_dma;
3060
3061         netdev_tx_sent_queue(txq, skb->len);
3062
3063         /* stop the queue when no more space available, the packet can have up
3064          * to sgl_size + 2. one for the meta descriptor and one for header
3065          * (if the header is larger than tx_max_header_size).
3066          */
3067         if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3068                                                    tx_ring->sgl_size + 2))) {
3069                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3070                           __func__, qid);
3071
3072                 netif_tx_stop_queue(txq);
3073                 ena_increase_stat(&tx_ring->tx_stats.queue_stop, 1,
3074                                   &tx_ring->syncp);
3075
3076                 /* There is a rare condition where this function decide to
3077                  * stop the queue but meanwhile clean_tx_irq updates
3078                  * next_to_completion and terminates.
3079                  * The queue will remain stopped forever.
3080                  * To solve this issue add a mb() to make sure that
3081                  * netif_tx_stop_queue() write is vissible before checking if
3082                  * there is additional space in the queue.
3083                  */
3084                 smp_mb();
3085
3086                 if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3087                                                  ENA_TX_WAKEUP_THRESH)) {
3088                         netif_tx_wake_queue(txq);
3089                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
3090                                           &tx_ring->syncp);
3091                 }
3092         }
3093
3094         if (netif_xmit_stopped(txq) || !netdev_xmit_more()) {
3095                 /* trigger the dma engine. ena_com_write_sq_doorbell()
3096                  * has a mb
3097                  */
3098                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
3099                 ena_increase_stat(&tx_ring->tx_stats.doorbells, 1,
3100                                   &tx_ring->syncp);
3101         }
3102
3103         return NETDEV_TX_OK;
3104
3105 error_unmap_dma:
3106         ena_unmap_tx_buff(tx_ring, tx_info);
3107         tx_info->skb = NULL;
3108
3109 error_drop_packet:
3110         dev_kfree_skb(skb);
3111         return NETDEV_TX_OK;
3112 }
3113
3114 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3115                             struct net_device *sb_dev)
3116 {
3117         u16 qid;
3118         /* we suspect that this is good for in--kernel network services that
3119          * want to loop incoming skb rx to tx in normal user generated traffic,
3120          * most probably we will not get to this
3121          */
3122         if (skb_rx_queue_recorded(skb))
3123                 qid = skb_get_rx_queue(skb);
3124         else
3125                 qid = netdev_pick_tx(dev, skb, NULL);
3126
3127         return qid;
3128 }
3129
3130 static void ena_config_host_info(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3131 {
3132         struct device *dev = &pdev->dev;
3133         struct ena_admin_host_info *host_info;
3134         int rc;
3135
3136         /* Allocate only the host info */
3137         rc = ena_com_allocate_host_info(ena_dev);
3138         if (rc) {
3139                 dev_err(dev, "Cannot allocate host info\n");
3140                 return;
3141         }
3142
3143         host_info = ena_dev->host_attr.host_info;
3144
3145         host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
3146         host_info->os_type = ENA_ADMIN_OS_LINUX;
3147         host_info->kernel_ver = LINUX_VERSION_CODE;
3148         strlcpy(host_info->kernel_ver_str, utsname()->version,
3149                 sizeof(host_info->kernel_ver_str) - 1);
3150         host_info->os_dist = 0;
3151         strncpy(host_info->os_dist_str, utsname()->release,
3152                 sizeof(host_info->os_dist_str) - 1);
3153         host_info->driver_version =
3154                 (DRV_MODULE_GEN_MAJOR) |
3155                 (DRV_MODULE_GEN_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3156                 (DRV_MODULE_GEN_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3157                 ("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3158         host_info->num_cpus = num_online_cpus();
3159
3160         host_info->driver_supported_features =
3161                 ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK |
3162                 ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK |
3163                 ENA_ADMIN_HOST_INFO_RX_BUF_MIRRORING_MASK |
3164                 ENA_ADMIN_HOST_INFO_RSS_CONFIGURABLE_FUNCTION_KEY_MASK;
3165
3166         rc = ena_com_set_host_attributes(ena_dev);
3167         if (rc) {
3168                 if (rc == -EOPNOTSUPP)
3169                         dev_warn(dev, "Cannot set host attributes\n");
3170                 else
3171                         dev_err(dev, "Cannot set host attributes\n");
3172
3173                 goto err;
3174         }
3175
3176         return;
3177
3178 err:
3179         ena_com_delete_host_info(ena_dev);
3180 }
3181
3182 static void ena_config_debug_area(struct ena_adapter *adapter)
3183 {
3184         u32 debug_area_size;
3185         int rc, ss_count;
3186
3187         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3188         if (ss_count <= 0) {
3189                 netif_err(adapter, drv, adapter->netdev,
3190                           "SS count is negative\n");
3191                 return;
3192         }
3193
3194         /* allocate 32 bytes for each string and 64bit for the value */
3195         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3196
3197         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3198         if (rc) {
3199                 netif_err(adapter, drv, adapter->netdev,
3200                           "Cannot allocate debug area\n");
3201                 return;
3202         }
3203
3204         rc = ena_com_set_host_attributes(adapter->ena_dev);
3205         if (rc) {
3206                 if (rc == -EOPNOTSUPP)
3207                         netif_warn(adapter, drv, adapter->netdev,
3208                                    "Cannot set host attributes\n");
3209                 else
3210                         netif_err(adapter, drv, adapter->netdev,
3211                                   "Cannot set host attributes\n");
3212                 goto err;
3213         }
3214
3215         return;
3216 err:
3217         ena_com_delete_debug_area(adapter->ena_dev);
3218 }
3219
3220 int ena_update_hw_stats(struct ena_adapter *adapter)
3221 {
3222         int rc = 0;
3223
3224         rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_stats);
3225         if (rc) {
3226                 dev_info_once(&adapter->pdev->dev, "Failed to get ENI stats\n");
3227                 return rc;
3228         }
3229
3230         return 0;
3231 }
3232
3233 static void ena_get_stats64(struct net_device *netdev,
3234                             struct rtnl_link_stats64 *stats)
3235 {
3236         struct ena_adapter *adapter = netdev_priv(netdev);
3237         struct ena_ring *rx_ring, *tx_ring;
3238         unsigned int start;
3239         u64 rx_drops;
3240         u64 tx_drops;
3241         int i;
3242
3243         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3244                 return;
3245
3246         for (i = 0; i < adapter->num_io_queues; i++) {
3247                 u64 bytes, packets;
3248
3249                 tx_ring = &adapter->tx_ring[i];
3250
3251                 do {
3252                         start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
3253                         packets = tx_ring->tx_stats.cnt;
3254                         bytes = tx_ring->tx_stats.bytes;
3255                 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
3256
3257                 stats->tx_packets += packets;
3258                 stats->tx_bytes += bytes;
3259
3260                 rx_ring = &adapter->rx_ring[i];
3261
3262                 do {
3263                         start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
3264                         packets = rx_ring->rx_stats.cnt;
3265                         bytes = rx_ring->rx_stats.bytes;
3266                 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
3267
3268                 stats->rx_packets += packets;
3269                 stats->rx_bytes += bytes;
3270         }
3271
3272         do {
3273                 start = u64_stats_fetch_begin_irq(&adapter->syncp);
3274                 rx_drops = adapter->dev_stats.rx_drops;
3275                 tx_drops = adapter->dev_stats.tx_drops;
3276         } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
3277
3278         stats->rx_dropped = rx_drops;
3279         stats->tx_dropped = tx_drops;
3280
3281         stats->multicast = 0;
3282         stats->collisions = 0;
3283
3284         stats->rx_length_errors = 0;
3285         stats->rx_crc_errors = 0;
3286         stats->rx_frame_errors = 0;
3287         stats->rx_fifo_errors = 0;
3288         stats->rx_missed_errors = 0;
3289         stats->tx_window_errors = 0;
3290
3291         stats->rx_errors = 0;
3292         stats->tx_errors = 0;
3293 }
3294
3295 static const struct net_device_ops ena_netdev_ops = {
3296         .ndo_open               = ena_open,
3297         .ndo_stop               = ena_close,
3298         .ndo_start_xmit         = ena_start_xmit,
3299         .ndo_select_queue       = ena_select_queue,
3300         .ndo_get_stats64        = ena_get_stats64,
3301         .ndo_tx_timeout         = ena_tx_timeout,
3302         .ndo_change_mtu         = ena_change_mtu,
3303         .ndo_set_mac_address    = NULL,
3304         .ndo_validate_addr      = eth_validate_addr,
3305         .ndo_bpf                = ena_xdp,
3306         .ndo_xdp_xmit           = ena_xdp_xmit,
3307 };
3308
3309 static int ena_device_validate_params(struct ena_adapter *adapter,
3310                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
3311 {
3312         struct net_device *netdev = adapter->netdev;
3313         int rc;
3314
3315         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3316                               adapter->mac_addr);
3317         if (!rc) {
3318                 netif_err(adapter, drv, netdev,
3319                           "Error, mac address are different\n");
3320                 return -EINVAL;
3321         }
3322
3323         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3324                 netif_err(adapter, drv, netdev,
3325                           "Error, device max mtu is smaller than netdev MTU\n");
3326                 return -EINVAL;
3327         }
3328
3329         return 0;
3330 }
3331
3332 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
3333 {
3334         llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
3335         llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
3336         llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
3337         llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
3338         llq_config->llq_ring_entry_size_value = 128;
3339 }
3340
3341 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3342                                            struct ena_com_dev *ena_dev,
3343                                            struct ena_admin_feature_llq_desc *llq,
3344                                            struct ena_llq_configurations *llq_default_configurations)
3345 {
3346         int rc;
3347         u32 llq_feature_mask;
3348
3349         llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3350         if (!(ena_dev->supported_features & llq_feature_mask)) {
3351                 dev_err(&pdev->dev,
3352                         "LLQ is not supported Fallback to host mode policy.\n");
3353                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3354                 return 0;
3355         }
3356
3357         rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3358         if (unlikely(rc)) {
3359                 dev_err(&pdev->dev,
3360                         "Failed to configure the device mode.  Fallback to host mode policy.\n");
3361                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3362         }
3363
3364         return 0;
3365 }
3366
3367 static int ena_map_llq_mem_bar(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
3368                                int bars)
3369 {
3370         bool has_mem_bar = !!(bars & BIT(ENA_MEM_BAR));
3371
3372         if (!has_mem_bar) {
3373                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3374                         dev_err(&pdev->dev,
3375                                 "ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3376                         ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3377                 }
3378
3379                 return 0;
3380         }
3381
3382         ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3383                                            pci_resource_start(pdev, ENA_MEM_BAR),
3384                                            pci_resource_len(pdev, ENA_MEM_BAR));
3385
3386         if (!ena_dev->mem_bar)
3387                 return -EFAULT;
3388
3389         return 0;
3390 }
3391
3392 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
3393                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
3394                            bool *wd_state)
3395 {
3396         struct ena_llq_configurations llq_config;
3397         struct device *dev = &pdev->dev;
3398         bool readless_supported;
3399         u32 aenq_groups;
3400         int dma_width;
3401         int rc;
3402
3403         rc = ena_com_mmio_reg_read_request_init(ena_dev);
3404         if (rc) {
3405                 dev_err(dev, "Failed to init mmio read less\n");
3406                 return rc;
3407         }
3408
3409         /* The PCIe configuration space revision id indicate if mmio reg
3410          * read is disabled
3411          */
3412         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3413         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3414
3415         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3416         if (rc) {
3417                 dev_err(dev, "Can not reset device\n");
3418                 goto err_mmio_read_less;
3419         }
3420
3421         rc = ena_com_validate_version(ena_dev);
3422         if (rc) {
3423                 dev_err(dev, "Device version is too low\n");
3424                 goto err_mmio_read_less;
3425         }
3426
3427         dma_width = ena_com_get_dma_width(ena_dev);
3428         if (dma_width < 0) {
3429                 dev_err(dev, "Invalid dma width value %d", dma_width);
3430                 rc = dma_width;
3431                 goto err_mmio_read_less;
3432         }
3433
3434         rc = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(dma_width));
3435         if (rc) {
3436                 dev_err(dev, "dma_set_mask_and_coherent failed %d\n", rc);
3437                 goto err_mmio_read_less;
3438         }
3439
3440         /* ENA admin level init */
3441         rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3442         if (rc) {
3443                 dev_err(dev,
3444                         "Can not initialize ena admin queue with device\n");
3445                 goto err_mmio_read_less;
3446         }
3447
3448         /* To enable the msix interrupts the driver needs to know the number
3449          * of queues. So the driver uses polling mode to retrieve this
3450          * information
3451          */
3452         ena_com_set_admin_polling_mode(ena_dev, true);
3453
3454         ena_config_host_info(ena_dev, pdev);
3455
3456         /* Get Device Attributes*/
3457         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3458         if (rc) {
3459                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3460                 goto err_admin_init;
3461         }
3462
3463         /* Try to turn all the available aenq groups */
3464         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3465                 BIT(ENA_ADMIN_FATAL_ERROR) |
3466                 BIT(ENA_ADMIN_WARNING) |
3467                 BIT(ENA_ADMIN_NOTIFICATION) |
3468                 BIT(ENA_ADMIN_KEEP_ALIVE);
3469
3470         aenq_groups &= get_feat_ctx->aenq.supported_groups;
3471
3472         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3473         if (rc) {
3474                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3475                 goto err_admin_init;
3476         }
3477
3478         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3479
3480         set_default_llq_configurations(&llq_config);
3481
3482         rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx->llq,
3483                                              &llq_config);
3484         if (rc) {
3485                 dev_err(dev, "ENA device init failed\n");
3486                 goto err_admin_init;
3487         }
3488
3489         return 0;
3490
3491 err_admin_init:
3492         ena_com_delete_host_info(ena_dev);
3493         ena_com_admin_destroy(ena_dev);
3494 err_mmio_read_less:
3495         ena_com_mmio_reg_read_request_destroy(ena_dev);
3496
3497         return rc;
3498 }
3499
3500 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3501 {
3502         struct ena_com_dev *ena_dev = adapter->ena_dev;
3503         struct device *dev = &adapter->pdev->dev;
3504         int rc;
3505
3506         rc = ena_enable_msix(adapter);
3507         if (rc) {
3508                 dev_err(dev, "Can not reserve msix vectors\n");
3509                 return rc;
3510         }
3511
3512         ena_setup_mgmnt_intr(adapter);
3513
3514         rc = ena_request_mgmnt_irq(adapter);
3515         if (rc) {
3516                 dev_err(dev, "Can not setup management interrupts\n");
3517                 goto err_disable_msix;
3518         }
3519
3520         ena_com_set_admin_polling_mode(ena_dev, false);
3521
3522         ena_com_admin_aenq_enable(ena_dev);
3523
3524         return 0;
3525
3526 err_disable_msix:
3527         ena_disable_msix(adapter);
3528
3529         return rc;
3530 }
3531
3532 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3533 {
3534         struct net_device *netdev = adapter->netdev;
3535         struct ena_com_dev *ena_dev = adapter->ena_dev;
3536         bool dev_up;
3537
3538         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3539                 return;
3540
3541         netif_carrier_off(netdev);
3542
3543         del_timer_sync(&adapter->timer_service);
3544
3545         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3546         adapter->dev_up_before_reset = dev_up;
3547         if (!graceful)
3548                 ena_com_set_admin_running_state(ena_dev, false);
3549
3550         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3551                 ena_down(adapter);
3552
3553         /* Stop the device from sending AENQ events (in case reset flag is set
3554          *  and device is up, ena_down() already reset the device.
3555          */
3556         if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3557                 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3558
3559         ena_free_mgmnt_irq(adapter);
3560
3561         ena_disable_msix(adapter);
3562
3563         ena_com_abort_admin_commands(ena_dev);
3564
3565         ena_com_wait_for_abort_completion(ena_dev);
3566
3567         ena_com_admin_destroy(ena_dev);
3568
3569         ena_com_mmio_reg_read_request_destroy(ena_dev);
3570
3571         /* return reset reason to default value */
3572         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3573
3574         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3575         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3576 }
3577
3578 static int ena_restore_device(struct ena_adapter *adapter)
3579 {
3580         struct ena_com_dev_get_features_ctx get_feat_ctx;
3581         struct ena_com_dev *ena_dev = adapter->ena_dev;
3582         struct pci_dev *pdev = adapter->pdev;
3583         bool wd_state;
3584         int rc;
3585
3586         set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3587         rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
3588         if (rc) {
3589                 dev_err(&pdev->dev, "Can not initialize device\n");
3590                 goto err;
3591         }
3592         adapter->wd_state = wd_state;
3593
3594         rc = ena_device_validate_params(adapter, &get_feat_ctx);
3595         if (rc) {
3596                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
3597                 goto err_device_destroy;
3598         }
3599
3600         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3601         if (rc) {
3602                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
3603                 goto err_device_destroy;
3604         }
3605         /* If the interface was up before the reset bring it up */
3606         if (adapter->dev_up_before_reset) {
3607                 rc = ena_up(adapter);
3608                 if (rc) {
3609                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
3610                         goto err_disable_msix;
3611                 }
3612         }
3613
3614         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3615
3616         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3617         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3618                 netif_carrier_on(adapter->netdev);
3619
3620         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3621         adapter->last_keep_alive_jiffies = jiffies;
3622
3623         dev_err(&pdev->dev, "Device reset completed successfully\n");
3624
3625         return rc;
3626 err_disable_msix:
3627         ena_free_mgmnt_irq(adapter);
3628         ena_disable_msix(adapter);
3629 err_device_destroy:
3630         ena_com_abort_admin_commands(ena_dev);
3631         ena_com_wait_for_abort_completion(ena_dev);
3632         ena_com_admin_destroy(ena_dev);
3633         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3634         ena_com_mmio_reg_read_request_destroy(ena_dev);
3635 err:
3636         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3637         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3638         dev_err(&pdev->dev,
3639                 "Reset attempt failed. Can not reset the device\n");
3640
3641         return rc;
3642 }
3643
3644 static void ena_fw_reset_device(struct work_struct *work)
3645 {
3646         struct ena_adapter *adapter =
3647                 container_of(work, struct ena_adapter, reset_task);
3648
3649         rtnl_lock();
3650
3651         if (likely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3652                 ena_destroy_device(adapter, false);
3653                 ena_restore_device(adapter);
3654         }
3655
3656         rtnl_unlock();
3657 }
3658
3659 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3660                                         struct ena_ring *rx_ring)
3661 {
3662         if (likely(rx_ring->first_interrupt))
3663                 return 0;
3664
3665         if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3666                 return 0;
3667
3668         rx_ring->no_interrupt_event_cnt++;
3669
3670         if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3671                 netif_err(adapter, rx_err, adapter->netdev,
3672                           "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3673                           rx_ring->qid);
3674                 adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3675                 smp_mb__before_atomic();
3676                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3677                 return -EIO;
3678         }
3679
3680         return 0;
3681 }
3682
3683 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3684                                           struct ena_ring *tx_ring)
3685 {
3686         struct ena_tx_buffer *tx_buf;
3687         unsigned long last_jiffies;
3688         u32 missed_tx = 0;
3689         int i, rc = 0;
3690
3691         for (i = 0; i < tx_ring->ring_size; i++) {
3692                 tx_buf = &tx_ring->tx_buffer_info[i];
3693                 last_jiffies = tx_buf->last_jiffies;
3694
3695                 if (last_jiffies == 0)
3696                         /* no pending Tx at this location */
3697                         continue;
3698
3699                 if (unlikely(!tx_ring->first_interrupt && time_is_before_jiffies(last_jiffies +
3700                              2 * adapter->missing_tx_completion_to))) {
3701                         /* If after graceful period interrupt is still not
3702                          * received, we schedule a reset
3703                          */
3704                         netif_err(adapter, tx_err, adapter->netdev,
3705                                   "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3706                                   tx_ring->qid);
3707                         adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3708                         smp_mb__before_atomic();
3709                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3710                         return -EIO;
3711                 }
3712
3713                 if (unlikely(time_is_before_jiffies(last_jiffies +
3714                                 adapter->missing_tx_completion_to))) {
3715                         if (!tx_buf->print_once)
3716                                 netif_notice(adapter, tx_err, adapter->netdev,
3717                                              "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
3718                                              tx_ring->qid, i);
3719
3720                         tx_buf->print_once = 1;
3721                         missed_tx++;
3722                 }
3723         }
3724
3725         if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3726                 netif_err(adapter, tx_err, adapter->netdev,
3727                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3728                           missed_tx,
3729                           adapter->missing_tx_completion_threshold);
3730                 adapter->reset_reason =
3731                         ENA_REGS_RESET_MISS_TX_CMPL;
3732                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3733                 rc = -EIO;
3734         }
3735
3736         ena_increase_stat(&tx_ring->tx_stats.missed_tx, missed_tx,
3737                           &tx_ring->syncp);
3738
3739         return rc;
3740 }
3741
3742 static void check_for_missing_completions(struct ena_adapter *adapter)
3743 {
3744         struct ena_ring *tx_ring;
3745         struct ena_ring *rx_ring;
3746         int i, budget, rc;
3747         int io_queue_count;
3748
3749         io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
3750         /* Make sure the driver doesn't turn the device in other process */
3751         smp_rmb();
3752
3753         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3754                 return;
3755
3756         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3757                 return;
3758
3759         if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
3760                 return;
3761
3762         budget = ENA_MONITORED_TX_QUEUES;
3763
3764         for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
3765                 tx_ring = &adapter->tx_ring[i];
3766                 rx_ring = &adapter->rx_ring[i];
3767
3768                 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3769                 if (unlikely(rc))
3770                         return;
3771
3772                 rc =  !ENA_IS_XDP_INDEX(adapter, i) ?
3773                         check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
3774                 if (unlikely(rc))
3775                         return;
3776
3777                 budget--;
3778                 if (!budget)
3779                         break;
3780         }
3781
3782         adapter->last_monitored_tx_qid = i % io_queue_count;
3783 }
3784
3785 /* trigger napi schedule after 2 consecutive detections */
3786 #define EMPTY_RX_REFILL 2
3787 /* For the rare case where the device runs out of Rx descriptors and the
3788  * napi handler failed to refill new Rx descriptors (due to a lack of memory
3789  * for example).
3790  * This case will lead to a deadlock:
3791  * The device won't send interrupts since all the new Rx packets will be dropped
3792  * The napi handler won't allocate new Rx descriptors so the device will be
3793  * able to send new packets.
3794  *
3795  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
3796  * It is recommended to have at least 512MB, with a minimum of 128MB for
3797  * constrained environment).
3798  *
3799  * When such a situation is detected - Reschedule napi
3800  */
3801 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
3802 {
3803         struct ena_ring *rx_ring;
3804         int i, refill_required;
3805
3806         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3807                 return;
3808
3809         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3810                 return;
3811
3812         for (i = 0; i < adapter->num_io_queues; i++) {
3813                 rx_ring = &adapter->rx_ring[i];
3814
3815                 refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
3816                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3817                         rx_ring->empty_rx_queue++;
3818
3819                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3820                                 ena_increase_stat(&rx_ring->rx_stats.empty_rx_ring, 1,
3821                                                   &rx_ring->syncp);
3822
3823                                 netif_err(adapter, drv, adapter->netdev,
3824                                           "Trigger refill for ring %d\n", i);
3825
3826                                 napi_schedule(rx_ring->napi);
3827                                 rx_ring->empty_rx_queue = 0;
3828                         }
3829                 } else {
3830                         rx_ring->empty_rx_queue = 0;
3831                 }
3832         }
3833 }
3834
3835 /* Check for keep alive expiration */
3836 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3837 {
3838         unsigned long keep_alive_expired;
3839
3840         if (!adapter->wd_state)
3841                 return;
3842
3843         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3844                 return;
3845
3846         keep_alive_expired = adapter->last_keep_alive_jiffies +
3847                              adapter->keep_alive_timeout;
3848         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3849                 netif_err(adapter, drv, adapter->netdev,
3850                           "Keep alive watchdog timeout.\n");
3851                 ena_increase_stat(&adapter->dev_stats.wd_expired, 1,
3852                                   &adapter->syncp);
3853                 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3854                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3855         }
3856 }
3857
3858 static void check_for_admin_com_state(struct ena_adapter *adapter)
3859 {
3860         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3861                 netif_err(adapter, drv, adapter->netdev,
3862                           "ENA admin queue is not in running state!\n");
3863                 ena_increase_stat(&adapter->dev_stats.admin_q_pause, 1,
3864                                   &adapter->syncp);
3865                 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3866                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3867         }
3868 }
3869
3870 static void ena_update_hints(struct ena_adapter *adapter,
3871                              struct ena_admin_ena_hw_hints *hints)
3872 {
3873         struct net_device *netdev = adapter->netdev;
3874
3875         if (hints->admin_completion_tx_timeout)
3876                 adapter->ena_dev->admin_queue.completion_timeout =
3877                         hints->admin_completion_tx_timeout * 1000;
3878
3879         if (hints->mmio_read_timeout)
3880                 /* convert to usec */
3881                 adapter->ena_dev->mmio_read.reg_read_to =
3882                         hints->mmio_read_timeout * 1000;
3883
3884         if (hints->missed_tx_completion_count_threshold_to_reset)
3885                 adapter->missing_tx_completion_threshold =
3886                         hints->missed_tx_completion_count_threshold_to_reset;
3887
3888         if (hints->missing_tx_completion_timeout) {
3889                 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3890                         adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3891                 else
3892                         adapter->missing_tx_completion_to =
3893                                 msecs_to_jiffies(hints->missing_tx_completion_timeout);
3894         }
3895
3896         if (hints->netdev_wd_timeout)
3897                 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3898
3899         if (hints->driver_watchdog_timeout) {
3900                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3901                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3902                 else
3903                         adapter->keep_alive_timeout =
3904                                 msecs_to_jiffies(hints->driver_watchdog_timeout);
3905         }
3906 }
3907
3908 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3909                                  struct net_device *netdev)
3910 {
3911         host_info->supported_network_features[0] =
3912                 netdev->features & GENMASK_ULL(31, 0);
3913         host_info->supported_network_features[1] =
3914                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
3915 }
3916
3917 static void ena_timer_service(struct timer_list *t)
3918 {
3919         struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3920         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3921         struct ena_admin_host_info *host_info =
3922                 adapter->ena_dev->host_attr.host_info;
3923
3924         check_for_missing_keep_alive(adapter);
3925
3926         check_for_admin_com_state(adapter);
3927
3928         check_for_missing_completions(adapter);
3929
3930         check_for_empty_rx_ring(adapter);
3931
3932         if (debug_area)
3933                 ena_dump_stats_to_buf(adapter, debug_area);
3934
3935         if (host_info)
3936                 ena_update_host_info(host_info, adapter->netdev);
3937
3938         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3939                 netif_err(adapter, drv, adapter->netdev,
3940                           "Trigger reset is on\n");
3941                 ena_dump_stats_to_dmesg(adapter);
3942                 queue_work(ena_wq, &adapter->reset_task);
3943                 return;
3944         }
3945
3946         /* Reset the timer */
3947         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3948 }
3949
3950 static u32 ena_calc_max_io_queue_num(struct pci_dev *pdev,
3951                                      struct ena_com_dev *ena_dev,
3952                                      struct ena_com_dev_get_features_ctx *get_feat_ctx)
3953 {
3954         u32 io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
3955
3956         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3957                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3958                         &get_feat_ctx->max_queue_ext.max_queue_ext;
3959                 io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
3960                                   max_queue_ext->max_rx_cq_num);
3961
3962                 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
3963                 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
3964         } else {
3965                 struct ena_admin_queue_feature_desc *max_queues =
3966                         &get_feat_ctx->max_queues;
3967                 io_tx_sq_num = max_queues->max_sq_num;
3968                 io_tx_cq_num = max_queues->max_cq_num;
3969                 io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
3970         }
3971
3972         /* In case of LLQ use the llq fields for the tx SQ/CQ */
3973         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3974                 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
3975
3976         max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
3977         max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
3978         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
3979         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
3980         /* 1 IRQ for mgmnt and 1 IRQs for each IO direction */
3981         max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
3982         if (unlikely(!max_num_io_queues)) {
3983                 dev_err(&pdev->dev, "The device doesn't have io queues\n");
3984                 return -EFAULT;
3985         }
3986
3987         return max_num_io_queues;
3988 }
3989
3990 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
3991                                  struct net_device *netdev)
3992 {
3993         netdev_features_t dev_features = 0;
3994
3995         /* Set offload features */
3996         if (feat->offload.tx &
3997                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
3998                 dev_features |= NETIF_F_IP_CSUM;
3999
4000         if (feat->offload.tx &
4001                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
4002                 dev_features |= NETIF_F_IPV6_CSUM;
4003
4004         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
4005                 dev_features |= NETIF_F_TSO;
4006
4007         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
4008                 dev_features |= NETIF_F_TSO6;
4009
4010         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
4011                 dev_features |= NETIF_F_TSO_ECN;
4012
4013         if (feat->offload.rx_supported &
4014                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
4015                 dev_features |= NETIF_F_RXCSUM;
4016
4017         if (feat->offload.rx_supported &
4018                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
4019                 dev_features |= NETIF_F_RXCSUM;
4020
4021         netdev->features =
4022                 dev_features |
4023                 NETIF_F_SG |
4024                 NETIF_F_RXHASH |
4025                 NETIF_F_HIGHDMA;
4026
4027         netdev->hw_features |= netdev->features;
4028         netdev->vlan_features |= netdev->features;
4029 }
4030
4031 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
4032                                      struct ena_com_dev_get_features_ctx *feat)
4033 {
4034         struct net_device *netdev = adapter->netdev;
4035
4036         /* Copy mac address */
4037         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
4038                 eth_hw_addr_random(netdev);
4039                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
4040         } else {
4041                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
4042                 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
4043         }
4044
4045         /* Set offload features */
4046         ena_set_dev_offloads(feat, netdev);
4047
4048         adapter->max_mtu = feat->dev_attr.max_mtu;
4049         netdev->max_mtu = adapter->max_mtu;
4050         netdev->min_mtu = ENA_MIN_MTU;
4051 }
4052
4053 static int ena_rss_init_default(struct ena_adapter *adapter)
4054 {
4055         struct ena_com_dev *ena_dev = adapter->ena_dev;
4056         struct device *dev = &adapter->pdev->dev;
4057         int rc, i;
4058         u32 val;
4059
4060         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
4061         if (unlikely(rc)) {
4062                 dev_err(dev, "Cannot init indirect table\n");
4063                 goto err_rss_init;
4064         }
4065
4066         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
4067                 val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
4068                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
4069                                                        ENA_IO_RXQ_IDX(val));
4070                 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4071                         dev_err(dev, "Cannot fill indirect table\n");
4072                         goto err_fill_indir;
4073                 }
4074         }
4075
4076         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ, NULL,
4077                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4078         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4079                 dev_err(dev, "Cannot fill hash function\n");
4080                 goto err_fill_indir;
4081         }
4082
4083         rc = ena_com_set_default_hash_ctrl(ena_dev);
4084         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4085                 dev_err(dev, "Cannot fill hash control\n");
4086                 goto err_fill_indir;
4087         }
4088
4089         return 0;
4090
4091 err_fill_indir:
4092         ena_com_rss_destroy(ena_dev);
4093 err_rss_init:
4094
4095         return rc;
4096 }
4097
4098 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4099 {
4100         int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4101
4102         pci_release_selected_regions(pdev, release_bars);
4103 }
4104
4105
4106 static int ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
4107 {
4108         struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
4109         struct ena_com_dev *ena_dev = ctx->ena_dev;
4110         u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
4111         u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
4112         u32 max_tx_queue_size;
4113         u32 max_rx_queue_size;
4114
4115         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4116                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4117                         &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
4118                 max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
4119                                           max_queue_ext->max_rx_sq_depth);
4120                 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
4121
4122                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4123                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4124                                                   llq->max_llq_depth);
4125                 else
4126                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4127                                                   max_queue_ext->max_tx_sq_depth);
4128
4129                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4130                                              max_queue_ext->max_per_packet_tx_descs);
4131                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4132                                              max_queue_ext->max_per_packet_rx_descs);
4133         } else {
4134                 struct ena_admin_queue_feature_desc *max_queues =
4135                         &ctx->get_feat_ctx->max_queues;
4136                 max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
4137                                           max_queues->max_sq_depth);
4138                 max_tx_queue_size = max_queues->max_cq_depth;
4139
4140                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4141                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4142                                                   llq->max_llq_depth);
4143                 else
4144                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4145                                                   max_queues->max_sq_depth);
4146
4147                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4148                                              max_queues->max_packet_tx_descs);
4149                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4150                                              max_queues->max_packet_rx_descs);
4151         }
4152
4153         max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
4154         max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
4155
4156         tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
4157                                   max_tx_queue_size);
4158         rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
4159                                   max_rx_queue_size);
4160
4161         tx_queue_size = rounddown_pow_of_two(tx_queue_size);
4162         rx_queue_size = rounddown_pow_of_two(rx_queue_size);
4163
4164         ctx->max_tx_queue_size = max_tx_queue_size;
4165         ctx->max_rx_queue_size = max_rx_queue_size;
4166         ctx->tx_queue_size = tx_queue_size;
4167         ctx->rx_queue_size = rx_queue_size;
4168
4169         return 0;
4170 }
4171
4172 /* ena_probe - Device Initialization Routine
4173  * @pdev: PCI device information struct
4174  * @ent: entry in ena_pci_tbl
4175  *
4176  * Returns 0 on success, negative on failure
4177  *
4178  * ena_probe initializes an adapter identified by a pci_dev structure.
4179  * The OS initialization, configuring of the adapter private structure,
4180  * and a hardware reset occur.
4181  */
4182 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4183 {
4184         struct ena_calc_queue_size_ctx calc_queue_ctx = {};
4185         struct ena_com_dev_get_features_ctx get_feat_ctx;
4186         struct ena_com_dev *ena_dev = NULL;
4187         struct ena_adapter *adapter;
4188         struct net_device *netdev;
4189         static int adapters_found;
4190         u32 max_num_io_queues;
4191         bool wd_state;
4192         int bars, rc;
4193
4194         dev_dbg(&pdev->dev, "%s\n", __func__);
4195
4196         rc = pci_enable_device_mem(pdev);
4197         if (rc) {
4198                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4199                 return rc;
4200         }
4201
4202         rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
4203         if (rc) {
4204                 dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n", rc);
4205                 goto err_disable_device;
4206         }
4207
4208         pci_set_master(pdev);
4209
4210         ena_dev = vzalloc(sizeof(*ena_dev));
4211         if (!ena_dev) {
4212                 rc = -ENOMEM;
4213                 goto err_disable_device;
4214         }
4215
4216         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4217         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4218         if (rc) {
4219                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4220                         rc);
4221                 goto err_free_ena_dev;
4222         }
4223
4224         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4225                                         pci_resource_start(pdev, ENA_REG_BAR),
4226                                         pci_resource_len(pdev, ENA_REG_BAR));
4227         if (!ena_dev->reg_bar) {
4228                 dev_err(&pdev->dev, "Failed to remap regs bar\n");
4229                 rc = -EFAULT;
4230                 goto err_free_region;
4231         }
4232
4233         ena_dev->ena_min_poll_delay_us = ENA_ADMIN_POLL_DELAY_US;
4234
4235         ena_dev->dmadev = &pdev->dev;
4236
4237         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), ENA_MAX_RINGS);
4238         if (!netdev) {
4239                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4240                 rc = -ENOMEM;
4241                 goto err_free_region;
4242         }
4243
4244         SET_NETDEV_DEV(netdev, &pdev->dev);
4245         adapter = netdev_priv(netdev);
4246         adapter->ena_dev = ena_dev;
4247         adapter->netdev = netdev;
4248         adapter->pdev = pdev;
4249         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4250
4251         ena_dev->net_device = netdev;
4252
4253         pci_set_drvdata(pdev, adapter);
4254
4255         rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
4256         if (rc) {
4257                 dev_err(&pdev->dev, "ENA device init failed\n");
4258                 if (rc == -ETIME)
4259                         rc = -EPROBE_DEFER;
4260                 goto err_netdev_destroy;
4261         }
4262
4263         rc = ena_map_llq_mem_bar(pdev, ena_dev, bars);
4264         if (rc) {
4265                 dev_err(&pdev->dev, "ENA llq bar mapping failed\n");
4266                 goto err_device_destroy;
4267         }
4268
4269         calc_queue_ctx.ena_dev = ena_dev;
4270         calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
4271         calc_queue_ctx.pdev = pdev;
4272
4273         /* Initial TX and RX interrupt delay. Assumes 1 usec granularity.
4274          * Updated during device initialization with the real granularity
4275          */
4276         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4277         ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4278         ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4279         max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4280         rc = ena_calc_io_queue_size(&calc_queue_ctx);
4281         if (rc || !max_num_io_queues) {
4282                 rc = -EFAULT;
4283                 goto err_device_destroy;
4284         }
4285
4286         ena_set_conf_feat_params(adapter, &get_feat_ctx);
4287
4288         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4289
4290         adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
4291         adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
4292         adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
4293         adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
4294         adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
4295         adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
4296
4297         adapter->num_io_queues = max_num_io_queues;
4298         adapter->max_num_io_queues = max_num_io_queues;
4299         adapter->last_monitored_tx_qid = 0;
4300
4301         adapter->xdp_first_ring = 0;
4302         adapter->xdp_num_queues = 0;
4303
4304         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4305         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4306                 adapter->disable_meta_caching =
4307                         !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
4308                            BIT(ENA_ADMIN_DISABLE_META_CACHING));
4309
4310         adapter->wd_state = wd_state;
4311
4312         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4313
4314         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4315         if (rc) {
4316                 dev_err(&pdev->dev,
4317                         "Failed to query interrupt moderation feature\n");
4318                 goto err_device_destroy;
4319         }
4320         ena_init_io_rings(adapter,
4321                           0,
4322                           adapter->xdp_num_queues +
4323                           adapter->num_io_queues);
4324
4325         netdev->netdev_ops = &ena_netdev_ops;
4326         netdev->watchdog_timeo = TX_TIMEOUT;
4327         ena_set_ethtool_ops(netdev);
4328
4329         netdev->priv_flags |= IFF_UNICAST_FLT;
4330
4331         u64_stats_init(&adapter->syncp);
4332
4333         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4334         if (rc) {
4335                 dev_err(&pdev->dev,
4336                         "Failed to enable and set the admin interrupts\n");
4337                 goto err_worker_destroy;
4338         }
4339         rc = ena_rss_init_default(adapter);
4340         if (rc && (rc != -EOPNOTSUPP)) {
4341                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4342                 goto err_free_msix;
4343         }
4344
4345         ena_config_debug_area(adapter);
4346
4347         if (!ena_update_hw_stats(adapter))
4348                 adapter->eni_stats_supported = true;
4349         else
4350                 adapter->eni_stats_supported = false;
4351
4352         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4353
4354         netif_carrier_off(netdev);
4355
4356         rc = register_netdev(netdev);
4357         if (rc) {
4358                 dev_err(&pdev->dev, "Cannot register net device\n");
4359                 goto err_rss;
4360         }
4361
4362         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4363
4364         adapter->last_keep_alive_jiffies = jiffies;
4365         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4366         adapter->missing_tx_completion_to = TX_TIMEOUT;
4367         adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4368
4369         ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4370
4371         timer_setup(&adapter->timer_service, ena_timer_service, 0);
4372         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4373
4374         dev_info(&pdev->dev,
4375                  "%s found at mem %lx, mac addr %pM\n",
4376                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4377                  netdev->dev_addr);
4378
4379         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4380
4381         adapters_found++;
4382
4383         return 0;
4384
4385 err_rss:
4386         ena_com_delete_debug_area(ena_dev);
4387         ena_com_rss_destroy(ena_dev);
4388 err_free_msix:
4389         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4390         /* stop submitting admin commands on a device that was reset */
4391         ena_com_set_admin_running_state(ena_dev, false);
4392         ena_free_mgmnt_irq(adapter);
4393         ena_disable_msix(adapter);
4394 err_worker_destroy:
4395         del_timer(&adapter->timer_service);
4396 err_device_destroy:
4397         ena_com_delete_host_info(ena_dev);
4398         ena_com_admin_destroy(ena_dev);
4399 err_netdev_destroy:
4400         free_netdev(netdev);
4401 err_free_region:
4402         ena_release_bars(ena_dev, pdev);
4403 err_free_ena_dev:
4404         vfree(ena_dev);
4405 err_disable_device:
4406         pci_disable_device(pdev);
4407         return rc;
4408 }
4409
4410 /*****************************************************************************/
4411
4412 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4413  * @pdev: PCI device information struct
4414  * @shutdown: Is it a shutdown operation? If false, means it is a removal
4415  *
4416  * __ena_shutoff is a helper routine that does the real work on shutdown and
4417  * removal paths; the difference between those paths is with regards to whether
4418  * dettach or unregister the netdevice.
4419  */
4420 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4421 {
4422         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4423         struct ena_com_dev *ena_dev;
4424         struct net_device *netdev;
4425
4426         ena_dev = adapter->ena_dev;
4427         netdev = adapter->netdev;
4428
4429 #ifdef CONFIG_RFS_ACCEL
4430         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4431                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4432                 netdev->rx_cpu_rmap = NULL;
4433         }
4434 #endif /* CONFIG_RFS_ACCEL */
4435
4436         /* Make sure timer and reset routine won't be called after
4437          * freeing device resources.
4438          */
4439         del_timer_sync(&adapter->timer_service);
4440         cancel_work_sync(&adapter->reset_task);
4441
4442         rtnl_lock(); /* lock released inside the below if-else block */
4443         adapter->reset_reason = ENA_REGS_RESET_SHUTDOWN;
4444         ena_destroy_device(adapter, true);
4445         if (shutdown) {
4446                 netif_device_detach(netdev);
4447                 dev_close(netdev);
4448                 rtnl_unlock();
4449         } else {
4450                 rtnl_unlock();
4451                 unregister_netdev(netdev);
4452                 free_netdev(netdev);
4453         }
4454
4455         ena_com_rss_destroy(ena_dev);
4456
4457         ena_com_delete_debug_area(ena_dev);
4458
4459         ena_com_delete_host_info(ena_dev);
4460
4461         ena_release_bars(ena_dev, pdev);
4462
4463         pci_disable_device(pdev);
4464
4465         vfree(ena_dev);
4466 }
4467
4468 /* ena_remove - Device Removal Routine
4469  * @pdev: PCI device information struct
4470  *
4471  * ena_remove is called by the PCI subsystem to alert the driver
4472  * that it should release a PCI device.
4473  */
4474
4475 static void ena_remove(struct pci_dev *pdev)
4476 {
4477         __ena_shutoff(pdev, false);
4478 }
4479
4480 /* ena_shutdown - Device Shutdown Routine
4481  * @pdev: PCI device information struct
4482  *
4483  * ena_shutdown is called by the PCI subsystem to alert the driver that
4484  * a shutdown/reboot (or kexec) is happening and device must be disabled.
4485  */
4486
4487 static void ena_shutdown(struct pci_dev *pdev)
4488 {
4489         __ena_shutoff(pdev, true);
4490 }
4491
4492 /* ena_suspend - PM suspend callback
4493  * @dev_d: Device information struct
4494  */
4495 static int __maybe_unused ena_suspend(struct device *dev_d)
4496 {
4497         struct pci_dev *pdev = to_pci_dev(dev_d);
4498         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4499
4500         ena_increase_stat(&adapter->dev_stats.suspend, 1, &adapter->syncp);
4501
4502         rtnl_lock();
4503         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4504                 dev_err(&pdev->dev,
4505                         "Ignoring device reset request as the device is being suspended\n");
4506                 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4507         }
4508         ena_destroy_device(adapter, true);
4509         rtnl_unlock();
4510         return 0;
4511 }
4512
4513 /* ena_resume - PM resume callback
4514  * @dev_d: Device information struct
4515  */
4516 static int __maybe_unused ena_resume(struct device *dev_d)
4517 {
4518         struct ena_adapter *adapter = dev_get_drvdata(dev_d);
4519         int rc;
4520
4521         ena_increase_stat(&adapter->dev_stats.resume, 1, &adapter->syncp);
4522
4523         rtnl_lock();
4524         rc = ena_restore_device(adapter);
4525         rtnl_unlock();
4526         return rc;
4527 }
4528
4529 static SIMPLE_DEV_PM_OPS(ena_pm_ops, ena_suspend, ena_resume);
4530
4531 static struct pci_driver ena_pci_driver = {
4532         .name           = DRV_MODULE_NAME,
4533         .id_table       = ena_pci_tbl,
4534         .probe          = ena_probe,
4535         .remove         = ena_remove,
4536         .shutdown       = ena_shutdown,
4537         .driver.pm      = &ena_pm_ops,
4538         .sriov_configure = pci_sriov_configure_simple,
4539 };
4540
4541 static int __init ena_init(void)
4542 {
4543         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4544         if (!ena_wq) {
4545                 pr_err("Failed to create workqueue\n");
4546                 return -ENOMEM;
4547         }
4548
4549         return pci_register_driver(&ena_pci_driver);
4550 }
4551
4552 static void __exit ena_cleanup(void)
4553 {
4554         pci_unregister_driver(&ena_pci_driver);
4555
4556         if (ena_wq) {
4557                 destroy_workqueue(ena_wq);
4558                 ena_wq = NULL;
4559         }
4560 }
4561
4562 /******************************************************************************
4563  ******************************** AENQ Handlers *******************************
4564  *****************************************************************************/
4565 /* ena_update_on_link_change:
4566  * Notify the network interface about the change in link status
4567  */
4568 static void ena_update_on_link_change(void *adapter_data,
4569                                       struct ena_admin_aenq_entry *aenq_e)
4570 {
4571         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4572         struct ena_admin_aenq_link_change_desc *aenq_desc =
4573                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
4574         int status = aenq_desc->flags &
4575                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4576
4577         if (status) {
4578                 netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
4579                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4580                 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4581                         netif_carrier_on(adapter->netdev);
4582         } else {
4583                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4584                 netif_carrier_off(adapter->netdev);
4585         }
4586 }
4587
4588 static void ena_keep_alive_wd(void *adapter_data,
4589                               struct ena_admin_aenq_entry *aenq_e)
4590 {
4591         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4592         struct ena_admin_aenq_keep_alive_desc *desc;
4593         u64 rx_drops;
4594         u64 tx_drops;
4595
4596         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4597         adapter->last_keep_alive_jiffies = jiffies;
4598
4599         rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4600         tx_drops = ((u64)desc->tx_drops_high << 32) | desc->tx_drops_low;
4601
4602         u64_stats_update_begin(&adapter->syncp);
4603         /* These stats are accumulated by the device, so the counters indicate
4604          * all drops since last reset.
4605          */
4606         adapter->dev_stats.rx_drops = rx_drops;
4607         adapter->dev_stats.tx_drops = tx_drops;
4608         u64_stats_update_end(&adapter->syncp);
4609 }
4610
4611 static void ena_notification(void *adapter_data,
4612                              struct ena_admin_aenq_entry *aenq_e)
4613 {
4614         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4615         struct ena_admin_ena_hw_hints *hints;
4616
4617         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4618              "Invalid group(%x) expected %x\n",
4619              aenq_e->aenq_common_desc.group,
4620              ENA_ADMIN_NOTIFICATION);
4621
4622         switch (aenq_e->aenq_common_desc.syndrome) {
4623         case ENA_ADMIN_UPDATE_HINTS:
4624                 hints = (struct ena_admin_ena_hw_hints *)
4625                         (&aenq_e->inline_data_w4);
4626                 ena_update_hints(adapter, hints);
4627                 break;
4628         default:
4629                 netif_err(adapter, drv, adapter->netdev,
4630                           "Invalid aenq notification link state %d\n",
4631                           aenq_e->aenq_common_desc.syndrome);
4632         }
4633 }
4634
4635 /* This handler will called for unknown event group or unimplemented handlers*/
4636 static void unimplemented_aenq_handler(void *data,
4637                                        struct ena_admin_aenq_entry *aenq_e)
4638 {
4639         struct ena_adapter *adapter = (struct ena_adapter *)data;
4640
4641         netif_err(adapter, drv, adapter->netdev,
4642                   "Unknown event was received or event with unimplemented handler\n");
4643 }
4644
4645 static struct ena_aenq_handlers aenq_handlers = {
4646         .handlers = {
4647                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4648                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
4649                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4650         },
4651         .unimplemented_handler = unimplemented_aenq_handler
4652 };
4653
4654 module_init(ena_init);
4655 module_exit(ena_cleanup);