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