drivers/net/ethernet/amazon/ena/ena_netdev.c

   1 /*
   2  * Copyright 2015 Amazon.com, Inc. or its affiliates.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  34
  35 #ifdef CONFIG_RFS_ACCEL
  36 #include <linux/cpu_rmap.h>
  37 #endif /* CONFIG_RFS_ACCEL */
  38 #include <linux/ethtool.h>
  39 #include <linux/kernel.h>
  40 #include <linux/module.h>
  41 #include <linux/numa.h>
  42 #include <linux/pci.h>
  43 #include <linux/utsname.h>
  44 #include <linux/version.h>
  45 #include <linux/vmalloc.h>
  46 #include <net/ip.h>
  47
  48 #include "ena_netdev.h"
  49 #include <linux/bpf_trace.h>
  50 #include "ena_pci_id_tbl.h"
  51
  52 static char version[] = DEVICE_NAME " v" DRV_MODULE_VERSION "\n";
  53
  54 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
  55 MODULE_DESCRIPTION(DEVICE_NAME);
  56 MODULE_LICENSE("GPL");
  57 MODULE_VERSION(DRV_MODULE_VERSION);
  58
  59 /* Time in jiffies before concluding the transmitter is hung. */
  60 #define TX_TIMEOUT  (5 * HZ)
  61
  62 #define ENA_NAPI_BUDGET 64
  63
  64 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
  65                 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
  66 static int debug = -1;
  67 module_param(debug, int, 0);
  68 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  69
  70 static struct ena_aenq_handlers aenq_handlers;
  71
  72 static struct workqueue_struct *ena_wq;
  73
  74 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
  75
  76 static int ena_rss_init_default(struct ena_adapter *adapter);
  77 static void check_for_admin_com_state(struct ena_adapter *adapter);
  78 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
  79 static int ena_restore_device(struct ena_adapter *adapter);
  80
  81 static void ena_init_io_rings(struct ena_adapter *adapter,
  82                               int first_index, int count);
  83 static void ena_init_napi_in_range(struct ena_adapter *adapter, int first_index,
  84                                    int count);
  85 static void ena_del_napi_in_range(struct ena_adapter *adapter, int first_index,
  86                                   int count);
  87 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid);
  88 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
  89                                            int first_index,
  90                                            int count);
  91 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid);
  92 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid);
  93 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget);
  94 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter);
  95 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter);
  96 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
  97                                       int first_index, int count);
  98 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
  99                                      int first_index, int count);
 100 static int ena_up(struct ena_adapter *adapter);
 101 static void ena_down(struct ena_adapter *adapter);
 102 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
 103                                  struct ena_ring *rx_ring);
 104 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
 105                                       struct ena_ring *rx_ring);
 106 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
 107                               struct ena_tx_buffer *tx_info);
 108 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
 109                                             int first_index, int count);
 110
 111 static void ena_tx_timeout(struct net_device *dev, unsigned int txqueue)
 112 {
 113         struct ena_adapter *adapter = netdev_priv(dev);
 114
 115         /* Change the state of the device to trigger reset
 116          * Check that we are not in the middle or a trigger already
 117          */
 118
 119         if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
 120                 return;
 121
 122         adapter->reset_reason = ENA_REGS_RESET_OS_NETDEV_WD;
 123         u64_stats_update_begin(&adapter->syncp);
 124         adapter->dev_stats.tx_timeout++;
 125         u64_stats_update_end(&adapter->syncp);
 126
 127         netif_err(adapter, tx_err, dev, "Transmit time out\n");
 128 }
 129
 130 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
 131 {
 132         int i;
 133
 134         for (i = 0; i < adapter->num_io_queues; i++)
 135                 adapter->rx_ring[i].mtu = mtu;
 136 }
 137
 138 static int ena_change_mtu(struct net_device *dev, int new_mtu)
 139 {
 140         struct ena_adapter *adapter = netdev_priv(dev);
 141         int ret;
 142
 143         ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
 144         if (!ret) {
 145                 netif_dbg(adapter, drv, dev, "set MTU to %d\n", new_mtu);
 146                 update_rx_ring_mtu(adapter, new_mtu);
 147                 dev->mtu = new_mtu;
 148         } else {
 149                 netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
 150                           new_mtu);
 151         }
 152
 153         return ret;
 154 }
 155
 156 static int ena_xmit_common(struct net_device *dev,
 157                            struct ena_ring *ring,
 158                            struct ena_tx_buffer *tx_info,
 159                            struct ena_com_tx_ctx *ena_tx_ctx,
 160                            u16 next_to_use,
 161                            u32 bytes)
 162 {
 163         struct ena_adapter *adapter = netdev_priv(dev);
 164         int rc, nb_hw_desc;
 165
 166         if (unlikely(ena_com_is_doorbell_needed(ring->ena_com_io_sq,
 167                                                 ena_tx_ctx))) {
 168                 netif_dbg(adapter, tx_queued, dev,
 169                           "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
 170                           ring->qid);
 171                 ena_com_write_sq_doorbell(ring->ena_com_io_sq);
 172         }
 173
 174         /* prepare the packet's descriptors to dma engine */
 175         rc = ena_com_prepare_tx(ring->ena_com_io_sq, ena_tx_ctx,
 176                                 &nb_hw_desc);
 177
 178         /* In case there isn't enough space in the queue for the packet,
 179          * we simply drop it. All other failure reasons of
 180          * ena_com_prepare_tx() are fatal and therefore require a device reset.
 181          */
 182         if (unlikely(rc)) {
 183                 netif_err(adapter, tx_queued, dev,
 184                           "failed to prepare tx bufs\n");
 185                 u64_stats_update_begin(&ring->syncp);
 186                 ring->tx_stats.prepare_ctx_err++;
 187                 u64_stats_update_end(&ring->syncp);
 188                 if (rc != -ENOMEM) {
 189                         adapter->reset_reason =
 190                                 ENA_REGS_RESET_DRIVER_INVALID_STATE;
 191                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
 192                 }
 193                 return rc;
 194         }
 195
 196         u64_stats_update_begin(&ring->syncp);
 197         ring->tx_stats.cnt++;
 198         ring->tx_stats.bytes += bytes;
 199         u64_stats_update_end(&ring->syncp);
 200
 201         tx_info->tx_descs = nb_hw_desc;
 202         tx_info->last_jiffies = jiffies;
 203         tx_info->print_once = 0;
 204
 205         ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
 206                                                  ring->ring_size);
 207         return 0;
 208 }
 209
 210 /* This is the XDP napi callback. XDP queues use a separate napi callback
 211  * than Rx/Tx queues.
 212  */
 213 static int ena_xdp_io_poll(struct napi_struct *napi, int budget)
 214 {
 215         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
 216         u32 xdp_work_done, xdp_budget;
 217         struct ena_ring *xdp_ring;
 218         int napi_comp_call = 0;
 219         int ret;
 220
 221         xdp_ring = ena_napi->xdp_ring;
 222         xdp_ring->first_interrupt = ena_napi->first_interrupt;
 223
 224         xdp_budget = budget;
 225
 226         if (!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags) ||
 227             test_bit(ENA_FLAG_TRIGGER_RESET, &xdp_ring->adapter->flags)) {
 228                 napi_complete_done(napi, 0);
 229                 return 0;
 230         }
 231
 232         xdp_work_done = ena_clean_xdp_irq(xdp_ring, xdp_budget);
 233
 234         /* If the device is about to reset or down, avoid unmask
 235          * the interrupt and return 0 so NAPI won't reschedule
 236          */
 237         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags))) {
 238                 napi_complete_done(napi, 0);
 239                 ret = 0;
 240         } else if (xdp_budget > xdp_work_done) {
 241                 napi_comp_call = 1;
 242                 if (napi_complete_done(napi, xdp_work_done))
 243                         ena_unmask_interrupt(xdp_ring, NULL);
 244                 ena_update_ring_numa_node(xdp_ring, NULL);
 245                 ret = xdp_work_done;
 246         } else {
 247                 ret = xdp_budget;
 248         }
 249
 250         u64_stats_update_begin(&xdp_ring->syncp);
 251         xdp_ring->tx_stats.napi_comp += napi_comp_call;
 252         xdp_ring->tx_stats.tx_poll++;
 253         u64_stats_update_end(&xdp_ring->syncp);
 254
 255         return ret;
 256 }
 257
 258 static int ena_xdp_tx_map_buff(struct ena_ring *xdp_ring,
 259                                struct ena_tx_buffer *tx_info,
 260                                struct xdp_buff *xdp,
 261                                void **push_hdr,
 262                                u32 *push_len)
 263 {
 264         struct ena_adapter *adapter = xdp_ring->adapter;
 265         struct ena_com_buf *ena_buf;
 266         dma_addr_t dma = 0;
 267         u32 size;
 268
 269         tx_info->xdpf = convert_to_xdp_frame(xdp);
 270         size = tx_info->xdpf->len;
 271         ena_buf = tx_info->bufs;
 272
 273         /* llq push buffer */
 274         *push_len = min_t(u32, size, xdp_ring->tx_max_header_size);
 275         *push_hdr = tx_info->xdpf->data;
 276
 277         if (size - *push_len > 0) {
 278                 dma = dma_map_single(xdp_ring->dev,
 279                                      *push_hdr + *push_len,
 280                                      size - *push_len,
 281                                      DMA_TO_DEVICE);
 282                 if (unlikely(dma_mapping_error(xdp_ring->dev, dma)))
 283                         goto error_report_dma_error;
 284
 285                 tx_info->map_linear_data = 1;
 286                 tx_info->num_of_bufs = 1;
 287         }
 288
 289         ena_buf->paddr = dma;
 290         ena_buf->len = size;
 291
 292         return 0;
 293
 294 error_report_dma_error:
 295         u64_stats_update_begin(&xdp_ring->syncp);
 296         xdp_ring->tx_stats.dma_mapping_err++;
 297         u64_stats_update_end(&xdp_ring->syncp);
 298         netdev_warn(adapter->netdev, "failed to map xdp buff\n");
 299
 300         xdp_return_frame_rx_napi(tx_info->xdpf);
 301         tx_info->xdpf = NULL;
 302         tx_info->num_of_bufs = 0;
 303
 304         return -EINVAL;
 305 }
 306
 307 static int ena_xdp_xmit_buff(struct net_device *dev,
 308                              struct xdp_buff *xdp,
 309                              int qid,
 310                              struct ena_rx_buffer *rx_info)
 311 {
 312         struct ena_adapter *adapter = netdev_priv(dev);
 313         struct ena_com_tx_ctx ena_tx_ctx = {0};
 314         struct ena_tx_buffer *tx_info;
 315         struct ena_ring *xdp_ring;
 316         u16 next_to_use, req_id;
 317         int rc;
 318         void *push_hdr;
 319         u32 push_len;
 320
 321         xdp_ring = &adapter->tx_ring[qid];
 322         next_to_use = xdp_ring->next_to_use;
 323         req_id = xdp_ring->free_ids[next_to_use];
 324         tx_info = &xdp_ring->tx_buffer_info[req_id];
 325         tx_info->num_of_bufs = 0;
 326         page_ref_inc(rx_info->page);
 327         tx_info->xdp_rx_page = rx_info->page;
 328
 329         rc = ena_xdp_tx_map_buff(xdp_ring, tx_info, xdp, &push_hdr, &push_len);
 330         if (unlikely(rc))
 331                 goto error_drop_packet;
 332
 333         ena_tx_ctx.ena_bufs = tx_info->bufs;
 334         ena_tx_ctx.push_header = push_hdr;
 335         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
 336         ena_tx_ctx.req_id = req_id;
 337         ena_tx_ctx.header_len = push_len;
 338
 339         rc = ena_xmit_common(dev,
 340                              xdp_ring,
 341                              tx_info,
 342                              &ena_tx_ctx,
 343                              next_to_use,
 344                              xdp->data_end - xdp->data);
 345         if (rc)
 346                 goto error_unmap_dma;
 347         /* trigger the dma engine. ena_com_write_sq_doorbell()
 348          * has a mb
 349          */
 350         ena_com_write_sq_doorbell(xdp_ring->ena_com_io_sq);
 351         u64_stats_update_begin(&xdp_ring->syncp);
 352         xdp_ring->tx_stats.doorbells++;
 353         u64_stats_update_end(&xdp_ring->syncp);
 354
 355         return NETDEV_TX_OK;
 356
 357 error_unmap_dma:
 358         ena_unmap_tx_buff(xdp_ring, tx_info);
 359         tx_info->xdpf = NULL;
 360 error_drop_packet:
 361
 362         return NETDEV_TX_OK;
 363 }
 364
 365 static int ena_xdp_execute(struct ena_ring *rx_ring,
 366                            struct xdp_buff *xdp,
 367                            struct ena_rx_buffer *rx_info)
 368 {
 369         struct bpf_prog *xdp_prog;
 370         u32 verdict = XDP_PASS;
 371
 372         rcu_read_lock();
 373         xdp_prog = READ_ONCE(rx_ring->xdp_bpf_prog);
 374
 375         if (!xdp_prog)
 376                 goto out;
 377
 378         verdict = bpf_prog_run_xdp(xdp_prog, xdp);
 379
 380         if (verdict == XDP_TX)
 381                 ena_xdp_xmit_buff(rx_ring->netdev,
 382                                   xdp,
 383                                   rx_ring->qid + rx_ring->adapter->num_io_queues,
 384                                   rx_info);
 385         else if (unlikely(verdict == XDP_ABORTED))
 386                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 387         else if (unlikely(verdict > XDP_TX))
 388                 bpf_warn_invalid_xdp_action(verdict);
 389 out:
 390         rcu_read_unlock();
 391         return verdict;
 392 }
 393
 394 static void ena_init_all_xdp_queues(struct ena_adapter *adapter)
 395 {
 396         adapter->xdp_first_ring = adapter->num_io_queues;
 397         adapter->xdp_num_queues = adapter->num_io_queues;
 398
 399         ena_init_io_rings(adapter,
 400                           adapter->xdp_first_ring,
 401                           adapter->xdp_num_queues);
 402 }
 403
 404 static int ena_setup_and_create_all_xdp_queues(struct ena_adapter *adapter)
 405 {
 406         int rc = 0;
 407
 408         rc = ena_setup_tx_resources_in_range(adapter, adapter->xdp_first_ring,
 409                                              adapter->xdp_num_queues);
 410         if (rc)
 411                 goto setup_err;
 412
 413         rc = ena_create_io_tx_queues_in_range(adapter,
 414                                               adapter->xdp_first_ring,
 415                                               adapter->xdp_num_queues);
 416         if (rc)
 417                 goto create_err;
 418
 419         return 0;
 420
 421 create_err:
 422         ena_free_all_io_tx_resources(adapter);
 423 setup_err:
 424         return rc;
 425 }
 426
 427 /* Provides a way for both kernel and bpf-prog to know
 428  * more about the RX-queue a given XDP frame arrived on.
 429  */
 430 static int ena_xdp_register_rxq_info(struct ena_ring *rx_ring)
 431 {
 432         int rc;
 433
 434         rc = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev, rx_ring->qid);
 435
 436         if (rc) {
 437                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 438                           "Failed to register xdp rx queue info. RX queue num %d rc: %d\n",
 439                           rx_ring->qid, rc);
 440                 goto err;
 441         }
 442
 443         rc = xdp_rxq_info_reg_mem_model(&rx_ring->xdp_rxq, MEM_TYPE_PAGE_SHARED,
 444                                         NULL);
 445
 446         if (rc) {
 447                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 448                           "Failed to register xdp rx queue info memory model. RX queue num %d rc: %d\n",
 449                           rx_ring->qid, rc);
 450                 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 451         }
 452
 453 err:
 454         return rc;
 455 }
 456
 457 static void ena_xdp_unregister_rxq_info(struct ena_ring *rx_ring)
 458 {
 459         xdp_rxq_info_unreg_mem_model(&rx_ring->xdp_rxq);
 460         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 461 }
 462
 463 void ena_xdp_exchange_program_rx_in_range(struct ena_adapter *adapter,
 464                                           struct bpf_prog *prog,
 465                                           int first,
 466                                           int count)
 467 {
 468         struct ena_ring *rx_ring;
 469         int i = 0;
 470
 471         for (i = first; i < count; i++) {
 472                 rx_ring = &adapter->rx_ring[i];
 473                 xchg(&rx_ring->xdp_bpf_prog, prog);
 474                 if (prog) {
 475                         ena_xdp_register_rxq_info(rx_ring);
 476                         rx_ring->rx_headroom = XDP_PACKET_HEADROOM;
 477                 } else {
 478                         ena_xdp_unregister_rxq_info(rx_ring);
 479                         rx_ring->rx_headroom = 0;
 480                 }
 481         }
 482 }
 483
 484 void ena_xdp_exchange_program(struct ena_adapter *adapter,
 485                               struct bpf_prog *prog)
 486 {
 487         struct bpf_prog *old_bpf_prog = xchg(&adapter->xdp_bpf_prog, prog);
 488
 489         ena_xdp_exchange_program_rx_in_range(adapter,
 490                                              prog,
 491                                              0,
 492                                              adapter->num_io_queues);
 493
 494         if (old_bpf_prog)
 495                 bpf_prog_put(old_bpf_prog);
 496 }
 497
 498 static int ena_destroy_and_free_all_xdp_queues(struct ena_adapter *adapter)
 499 {
 500         bool was_up;
 501         int rc;
 502
 503         was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 504
 505         if (was_up)
 506                 ena_down(adapter);
 507
 508         adapter->xdp_first_ring = 0;
 509         adapter->xdp_num_queues = 0;
 510         ena_xdp_exchange_program(adapter, NULL);
 511         if (was_up) {
 512                 rc = ena_up(adapter);
 513                 if (rc)
 514                         return rc;
 515         }
 516         return 0;
 517 }
 518
 519 static int ena_xdp_set(struct net_device *netdev, struct netdev_bpf *bpf)
 520 {
 521         struct ena_adapter *adapter = netdev_priv(netdev);
 522         struct bpf_prog *prog = bpf->prog;
 523         struct bpf_prog *old_bpf_prog;
 524         int rc, prev_mtu;
 525         bool is_up;
 526
 527         is_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 528         rc = ena_xdp_allowed(adapter);
 529         if (rc == ENA_XDP_ALLOWED) {
 530                 old_bpf_prog = adapter->xdp_bpf_prog;
 531                 if (prog) {
 532                         if (!is_up) {
 533                                 ena_init_all_xdp_queues(adapter);
 534                         } else if (!old_bpf_prog) {
 535                                 ena_down(adapter);
 536                                 ena_init_all_xdp_queues(adapter);
 537                         }
 538                         ena_xdp_exchange_program(adapter, prog);
 539
 540                         if (is_up && !old_bpf_prog) {
 541                                 rc = ena_up(adapter);
 542                                 if (rc)
 543                                         return rc;
 544                         }
 545                 } else if (old_bpf_prog) {
 546                         rc = ena_destroy_and_free_all_xdp_queues(adapter);
 547                         if (rc)
 548                                 return rc;
 549                 }
 550
 551                 prev_mtu = netdev->max_mtu;
 552                 netdev->max_mtu = prog ? ENA_XDP_MAX_MTU : adapter->max_mtu;
 553
 554                 if (!old_bpf_prog)
 555                         netif_info(adapter, drv, adapter->netdev,
 556                                    "xdp program set, changing the max_mtu from %d to %d",
 557                                    prev_mtu, netdev->max_mtu);
 558
 559         } else if (rc == ENA_XDP_CURRENT_MTU_TOO_LARGE) {
 560                 netif_err(adapter, drv, adapter->netdev,
 561                           "Failed to set xdp program, the current MTU (%d) is larger than the maximum allowed MTU (%lu) while xdp is on",
 562                           netdev->mtu, ENA_XDP_MAX_MTU);
 563                 NL_SET_ERR_MSG_MOD(bpf->extack,
 564                                    "Failed to set xdp program, the current MTU is larger than the maximum allowed MTU. Check the dmesg for more info");
 565                 return -EINVAL;
 566         } else if (rc == ENA_XDP_NO_ENOUGH_QUEUES) {
 567                 netif_err(adapter, drv, adapter->netdev,
 568                           "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",
 569                           adapter->num_io_queues, adapter->max_num_io_queues);
 570                 NL_SET_ERR_MSG_MOD(bpf->extack,
 571                                    "Failed to set xdp program, there is no enough space for allocating XDP queues, Check the dmesg for more info");
 572                 return -EINVAL;
 573         }
 574
 575         return 0;
 576 }
 577
 578 /* This is the main xdp callback, it's used by the kernel to set/unset the xdp
 579  * program as well as to query the current xdp program id.
 580  */
 581 static int ena_xdp(struct net_device *netdev, struct netdev_bpf *bpf)
 582 {
 583         struct ena_adapter *adapter = netdev_priv(netdev);
 584
 585         switch (bpf->command) {
 586         case XDP_SETUP_PROG:
 587                 return ena_xdp_set(netdev, bpf);
 588         case XDP_QUERY_PROG:
 589                 bpf->prog_id = adapter->xdp_bpf_prog ?
 590                         adapter->xdp_bpf_prog->aux->id : 0;
 591                 break;
 592         default:
 593                 return -EINVAL;
 594         }
 595         return 0;
 596 }
 597
 598 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
 599 {
 600 #ifdef CONFIG_RFS_ACCEL
 601         u32 i;
 602         int rc;
 603
 604         adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_io_queues);
 605         if (!adapter->netdev->rx_cpu_rmap)
 606                 return -ENOMEM;
 607         for (i = 0; i < adapter->num_io_queues; i++) {
 608                 int irq_idx = ENA_IO_IRQ_IDX(i);
 609
 610                 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
 611                                       pci_irq_vector(adapter->pdev, irq_idx));
 612                 if (rc) {
 613                         free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
 614                         adapter->netdev->rx_cpu_rmap = NULL;
 615                         return rc;
 616                 }
 617         }
 618 #endif /* CONFIG_RFS_ACCEL */
 619         return 0;
 620 }
 621
 622 static void ena_init_io_rings_common(struct ena_adapter *adapter,
 623                                      struct ena_ring *ring, u16 qid)
 624 {
 625         ring->qid = qid;
 626         ring->pdev = adapter->pdev;
 627         ring->dev = &adapter->pdev->dev;
 628         ring->netdev = adapter->netdev;
 629         ring->napi = &adapter->ena_napi[qid].napi;
 630         ring->adapter = adapter;
 631         ring->ena_dev = adapter->ena_dev;
 632         ring->per_napi_packets = 0;
 633         ring->cpu = 0;
 634         ring->first_interrupt = false;
 635         ring->no_interrupt_event_cnt = 0;
 636         u64_stats_init(&ring->syncp);
 637 }
 638
 639 static void ena_init_io_rings(struct ena_adapter *adapter,
 640                               int first_index, int count)
 641 {
 642         struct ena_com_dev *ena_dev;
 643         struct ena_ring *txr, *rxr;
 644         int i;
 645
 646         ena_dev = adapter->ena_dev;
 647
 648         for (i = first_index; i < first_index + count; i++) {
 649                 txr = &adapter->tx_ring[i];
 650                 rxr = &adapter->rx_ring[i];
 651
 652                 /* TX common ring state */
 653                 ena_init_io_rings_common(adapter, txr, i);
 654
 655                 /* TX specific ring state */
 656                 txr->ring_size = adapter->requested_tx_ring_size;
 657                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
 658                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
 659                 txr->sgl_size = adapter->max_tx_sgl_size;
 660                 txr->smoothed_interval =
 661                         ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
 662
 663                 /* Don't init RX queues for xdp queues */
 664                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
 665                         /* RX common ring state */
 666                         ena_init_io_rings_common(adapter, rxr, i);
 667
 668                         /* RX specific ring state */
 669                         rxr->ring_size = adapter->requested_rx_ring_size;
 670                         rxr->rx_copybreak = adapter->rx_copybreak;
 671                         rxr->sgl_size = adapter->max_rx_sgl_size;
 672                         rxr->smoothed_interval =
 673                                 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
 674                         rxr->empty_rx_queue = 0;
 675                         adapter->ena_napi[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
 676                 }
 677         }
 678 }
 679
 680 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
 681  * @adapter: network interface device structure
 682  * @qid: queue index
 683  *
 684  * Return 0 on success, negative on failure
 685  */
 686 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
 687 {
 688         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 689         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 690         int size, i, node;
 691
 692         if (tx_ring->tx_buffer_info) {
 693                 netif_err(adapter, ifup,
 694                           adapter->netdev, "tx_buffer_info info is not NULL");
 695                 return -EEXIST;
 696         }
 697
 698         size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
 699         node = cpu_to_node(ena_irq->cpu);
 700
 701         tx_ring->tx_buffer_info = vzalloc_node(size, node);
 702         if (!tx_ring->tx_buffer_info) {
 703                 tx_ring->tx_buffer_info = vzalloc(size);
 704                 if (!tx_ring->tx_buffer_info)
 705                         goto err_tx_buffer_info;
 706         }
 707
 708         size = sizeof(u16) * tx_ring->ring_size;
 709         tx_ring->free_ids = vzalloc_node(size, node);
 710         if (!tx_ring->free_ids) {
 711                 tx_ring->free_ids = vzalloc(size);
 712                 if (!tx_ring->free_ids)
 713                         goto err_tx_free_ids;
 714         }
 715
 716         size = tx_ring->tx_max_header_size;
 717         tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
 718         if (!tx_ring->push_buf_intermediate_buf) {
 719                 tx_ring->push_buf_intermediate_buf = vzalloc(size);
 720                 if (!tx_ring->push_buf_intermediate_buf)
 721                         goto err_push_buf_intermediate_buf;
 722         }
 723
 724         /* Req id ring for TX out of order completions */
 725         for (i = 0; i < tx_ring->ring_size; i++)
 726                 tx_ring->free_ids[i] = i;
 727
 728         /* Reset tx statistics */
 729         memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
 730
 731         tx_ring->next_to_use = 0;
 732         tx_ring->next_to_clean = 0;
 733         tx_ring->cpu = ena_irq->cpu;
 734         return 0;
 735
 736 err_push_buf_intermediate_buf:
 737         vfree(tx_ring->free_ids);
 738         tx_ring->free_ids = NULL;
 739 err_tx_free_ids:
 740         vfree(tx_ring->tx_buffer_info);
 741         tx_ring->tx_buffer_info = NULL;
 742 err_tx_buffer_info:
 743         return -ENOMEM;
 744 }
 745
 746 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
 747  * @adapter: network interface device structure
 748  * @qid: queue index
 749  *
 750  * Free all transmit software resources
 751  */
 752 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
 753 {
 754         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 755
 756         vfree(tx_ring->tx_buffer_info);
 757         tx_ring->tx_buffer_info = NULL;
 758
 759         vfree(tx_ring->free_ids);
 760         tx_ring->free_ids = NULL;
 761
 762         vfree(tx_ring->push_buf_intermediate_buf);
 763         tx_ring->push_buf_intermediate_buf = NULL;
 764 }
 765
 766 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
 767                                            int first_index,
 768                                            int count)
 769 {
 770         int i, rc = 0;
 771
 772         for (i = first_index; i < first_index + count; i++) {
 773                 rc = ena_setup_tx_resources(adapter, i);
 774                 if (rc)
 775                         goto err_setup_tx;
 776         }
 777
 778         return 0;
 779
 780 err_setup_tx:
 781
 782         netif_err(adapter, ifup, adapter->netdev,
 783                   "Tx queue %d: allocation failed\n", i);
 784
 785         /* rewind the index freeing the rings as we go */
 786         while (first_index < i--)
 787                 ena_free_tx_resources(adapter, i);
 788         return rc;
 789 }
 790
 791 static void ena_free_all_io_tx_resources_in_range(struct ena_adapter *adapter,
 792                                                   int first_index, int count)
 793 {
 794         int i;
 795
 796         for (i = first_index; i < first_index + count; i++)
 797                 ena_free_tx_resources(adapter, i);
 798 }
 799
 800 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
 801  * @adapter: board private structure
 802  *
 803  * Free all transmit software resources
 804  */
 805 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
 806 {
 807         ena_free_all_io_tx_resources_in_range(adapter,
 808                                               0,
 809                                               adapter->xdp_num_queues +
 810                                               adapter->num_io_queues);
 811 }
 812
 813 static int validate_rx_req_id(struct ena_ring *rx_ring, u16 req_id)
 814 {
 815         if (likely(req_id < rx_ring->ring_size))
 816                 return 0;
 817
 818         netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 819                   "Invalid rx req_id: %hu\n", req_id);
 820
 821         u64_stats_update_begin(&rx_ring->syncp);
 822         rx_ring->rx_stats.bad_req_id++;
 823         u64_stats_update_end(&rx_ring->syncp);
 824
 825         /* Trigger device reset */
 826         rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
 827         set_bit(ENA_FLAG_TRIGGER_RESET, &rx_ring->adapter->flags);
 828         return -EFAULT;
 829 }
 830
 831 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
 832  * @adapter: network interface device structure
 833  * @qid: queue index
 834  *
 835  * Returns 0 on success, negative on failure
 836  */
 837 static int ena_setup_rx_resources(struct ena_adapter *adapter,
 838                                   u32 qid)
 839 {
 840         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 841         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 842         int size, node, i;
 843
 844         if (rx_ring->rx_buffer_info) {
 845                 netif_err(adapter, ifup, adapter->netdev,
 846                           "rx_buffer_info is not NULL");
 847                 return -EEXIST;
 848         }
 849
 850         /* alloc extra element so in rx path
 851          * we can always prefetch rx_info + 1
 852          */
 853         size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
 854         node = cpu_to_node(ena_irq->cpu);
 855
 856         rx_ring->rx_buffer_info = vzalloc_node(size, node);
 857         if (!rx_ring->rx_buffer_info) {
 858                 rx_ring->rx_buffer_info = vzalloc(size);
 859                 if (!rx_ring->rx_buffer_info)
 860                         return -ENOMEM;
 861         }
 862
 863         size = sizeof(u16) * rx_ring->ring_size;
 864         rx_ring->free_ids = vzalloc_node(size, node);
 865         if (!rx_ring->free_ids) {
 866                 rx_ring->free_ids = vzalloc(size);
 867                 if (!rx_ring->free_ids) {
 868                         vfree(rx_ring->rx_buffer_info);
 869                         rx_ring->rx_buffer_info = NULL;
 870                         return -ENOMEM;
 871                 }
 872         }
 873
 874         /* Req id ring for receiving RX pkts out of order */
 875         for (i = 0; i < rx_ring->ring_size; i++)
 876                 rx_ring->free_ids[i] = i;
 877
 878         /* Reset rx statistics */
 879         memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
 880
 881         rx_ring->next_to_clean = 0;
 882         rx_ring->next_to_use = 0;
 883         rx_ring->cpu = ena_irq->cpu;
 884
 885         return 0;
 886 }
 887
 888 /* ena_free_rx_resources - Free I/O Rx Resources
 889  * @adapter: network interface device structure
 890  * @qid: queue index
 891  *
 892  * Free all receive software resources
 893  */
 894 static void ena_free_rx_resources(struct ena_adapter *adapter,
 895                                   u32 qid)
 896 {
 897         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 898
 899         vfree(rx_ring->rx_buffer_info);
 900         rx_ring->rx_buffer_info = NULL;
 901
 902         vfree(rx_ring->free_ids);
 903         rx_ring->free_ids = NULL;
 904 }
 905
 906 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
 907  * @adapter: board private structure
 908  *
 909  * Return 0 on success, negative on failure
 910  */
 911 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
 912 {
 913         int i, rc = 0;
 914
 915         for (i = 0; i < adapter->num_io_queues; i++) {
 916                 rc = ena_setup_rx_resources(adapter, i);
 917                 if (rc)
 918                         goto err_setup_rx;
 919         }
 920
 921         return 0;
 922
 923 err_setup_rx:
 924
 925         netif_err(adapter, ifup, adapter->netdev,
 926                   "Rx queue %d: allocation failed\n", i);
 927
 928         /* rewind the index freeing the rings as we go */
 929         while (i--)
 930                 ena_free_rx_resources(adapter, i);
 931         return rc;
 932 }
 933
 934 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
 935  * @adapter: board private structure
 936  *
 937  * Free all receive software resources
 938  */
 939 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
 940 {
 941         int i;
 942
 943         for (i = 0; i < adapter->num_io_queues; i++)
 944                 ena_free_rx_resources(adapter, i);
 945 }
 946
 947 static int ena_alloc_rx_page(struct ena_ring *rx_ring,
 948                                     struct ena_rx_buffer *rx_info, gfp_t gfp)
 949 {
 950         struct ena_com_buf *ena_buf;
 951         struct page *page;
 952         dma_addr_t dma;
 953
 954         /* if previous allocated page is not used */
 955         if (unlikely(rx_info->page))
 956                 return 0;
 957
 958         page = alloc_page(gfp);
 959         if (unlikely(!page)) {
 960                 u64_stats_update_begin(&rx_ring->syncp);
 961                 rx_ring->rx_stats.page_alloc_fail++;
 962                 u64_stats_update_end(&rx_ring->syncp);
 963                 return -ENOMEM;
 964         }
 965
 966         dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
 967                            DMA_FROM_DEVICE);
 968         if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
 969                 u64_stats_update_begin(&rx_ring->syncp);
 970                 rx_ring->rx_stats.dma_mapping_err++;
 971                 u64_stats_update_end(&rx_ring->syncp);
 972
 973                 __free_page(page);
 974                 return -EIO;
 975         }
 976         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 977                   "alloc page %p, rx_info %p\n", page, rx_info);
 978
 979         rx_info->page = page;
 980         rx_info->page_offset = 0;
 981         ena_buf = &rx_info->ena_buf;
 982         ena_buf->paddr = dma + rx_ring->rx_headroom;
 983         ena_buf->len = ENA_PAGE_SIZE - rx_ring->rx_headroom;
 984
 985         return 0;
 986 }
 987
 988 static void ena_free_rx_page(struct ena_ring *rx_ring,
 989                              struct ena_rx_buffer *rx_info)
 990 {
 991         struct page *page = rx_info->page;
 992         struct ena_com_buf *ena_buf = &rx_info->ena_buf;
 993
 994         if (unlikely(!page)) {
 995                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
 996                            "Trying to free unallocated buffer\n");
 997                 return;
 998         }
 999
1000         dma_unmap_page(rx_ring->dev,
1001                        ena_buf->paddr - rx_ring->rx_headroom,
1002                        ENA_PAGE_SIZE,
1003                        DMA_FROM_DEVICE);
1004
1005         __free_page(page);
1006         rx_info->page = NULL;
1007 }
1008
1009 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
1010 {
1011         u16 next_to_use, req_id;
1012         u32 i;
1013         int rc;
1014
1015         next_to_use = rx_ring->next_to_use;
1016
1017         for (i = 0; i < num; i++) {
1018                 struct ena_rx_buffer *rx_info;
1019
1020                 req_id = rx_ring->free_ids[next_to_use];
1021
1022                 rx_info = &rx_ring->rx_buffer_info[req_id];
1023
1024                 rc = ena_alloc_rx_page(rx_ring, rx_info,
1025                                        GFP_ATOMIC | __GFP_COMP);
1026                 if (unlikely(rc < 0)) {
1027                         netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1028                                    "failed to alloc buffer for rx queue %d\n",
1029                                    rx_ring->qid);
1030                         break;
1031                 }
1032                 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1033                                                 &rx_info->ena_buf,
1034                                                 req_id);
1035                 if (unlikely(rc)) {
1036                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1037                                    "failed to add buffer for rx queue %d\n",
1038                                    rx_ring->qid);
1039                         break;
1040                 }
1041                 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1042                                                    rx_ring->ring_size);
1043         }
1044
1045         if (unlikely(i < num)) {
1046                 u64_stats_update_begin(&rx_ring->syncp);
1047                 rx_ring->rx_stats.refil_partial++;
1048                 u64_stats_update_end(&rx_ring->syncp);
1049                 netdev_warn(rx_ring->netdev,
1050                             "refilled rx qid %d with only %d buffers (from %d)\n",
1051                             rx_ring->qid, i, num);
1052         }
1053
1054         /* ena_com_write_sq_doorbell issues a wmb() */
1055         if (likely(i))
1056                 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1057
1058         rx_ring->next_to_use = next_to_use;
1059
1060         return i;
1061 }
1062
1063 static void ena_free_rx_bufs(struct ena_adapter *adapter,
1064                              u32 qid)
1065 {
1066         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1067         u32 i;
1068
1069         for (i = 0; i < rx_ring->ring_size; i++) {
1070                 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1071
1072                 if (rx_info->page)
1073                         ena_free_rx_page(rx_ring, rx_info);
1074         }
1075 }
1076
1077 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
1078  * @adapter: board private structure
1079  */
1080 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1081 {
1082         struct ena_ring *rx_ring;
1083         int i, rc, bufs_num;
1084
1085         for (i = 0; i < adapter->num_io_queues; i++) {
1086                 rx_ring = &adapter->rx_ring[i];
1087                 bufs_num = rx_ring->ring_size - 1;
1088                 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1089
1090                 if (unlikely(rc != bufs_num))
1091                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1092                                    "refilling Queue %d failed. allocated %d buffers from: %d\n",
1093                                    i, rc, bufs_num);
1094         }
1095 }
1096
1097 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
1098 {
1099         int i;
1100
1101         for (i = 0; i < adapter->num_io_queues; i++)
1102                 ena_free_rx_bufs(adapter, i);
1103 }
1104
1105 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
1106                               struct ena_tx_buffer *tx_info)
1107 {
1108         struct ena_com_buf *ena_buf;
1109         u32 cnt;
1110         int i;
1111
1112         ena_buf = tx_info->bufs;
1113         cnt = tx_info->num_of_bufs;
1114
1115         if (unlikely(!cnt))
1116                 return;
1117
1118         if (tx_info->map_linear_data) {
1119                 dma_unmap_single(tx_ring->dev,
1120                                  dma_unmap_addr(ena_buf, paddr),
1121                                  dma_unmap_len(ena_buf, len),
1122                                  DMA_TO_DEVICE);
1123                 ena_buf++;
1124                 cnt--;
1125         }
1126
1127         /* unmap remaining mapped pages */
1128         for (i = 0; i < cnt; i++) {
1129                 dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
1130                                dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
1131                 ena_buf++;
1132         }
1133 }
1134
1135 /* ena_free_tx_bufs - Free Tx Buffers per Queue
1136  * @tx_ring: TX ring for which buffers be freed
1137  */
1138 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
1139 {
1140         bool print_once = true;
1141         u32 i;
1142
1143         for (i = 0; i < tx_ring->ring_size; i++) {
1144                 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1145
1146                 if (!tx_info->skb)
1147                         continue;
1148
1149                 if (print_once) {
1150                         netdev_notice(tx_ring->netdev,
1151                                       "free uncompleted tx skb qid %d idx 0x%x\n",
1152                                       tx_ring->qid, i);
1153                         print_once = false;
1154                 } else {
1155                         netdev_dbg(tx_ring->netdev,
1156                                    "free uncompleted tx skb qid %d idx 0x%x\n",
1157                                    tx_ring->qid, i);
1158                 }
1159
1160                 ena_unmap_tx_buff(tx_ring, tx_info);
1161
1162                 dev_kfree_skb_any(tx_info->skb);
1163         }
1164         netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
1165                                                   tx_ring->qid));
1166 }
1167
1168 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
1169 {
1170         struct ena_ring *tx_ring;
1171         int i;
1172
1173         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1174                 tx_ring = &adapter->tx_ring[i];
1175                 ena_free_tx_bufs(tx_ring);
1176         }
1177 }
1178
1179 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1180 {
1181         u16 ena_qid;
1182         int i;
1183
1184         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1185                 ena_qid = ENA_IO_TXQ_IDX(i);
1186                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1187         }
1188 }
1189
1190 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1191 {
1192         u16 ena_qid;
1193         int i;
1194
1195         for (i = 0; i < adapter->num_io_queues; i++) {
1196                 ena_qid = ENA_IO_RXQ_IDX(i);
1197                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
1198                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1199         }
1200 }
1201
1202 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
1203 {
1204         ena_destroy_all_tx_queues(adapter);
1205         ena_destroy_all_rx_queues(adapter);
1206 }
1207
1208 static int handle_invalid_req_id(struct ena_ring *ring, u16 req_id,
1209                                  struct ena_tx_buffer *tx_info, bool is_xdp)
1210 {
1211         if (tx_info)
1212                 netif_err(ring->adapter,
1213                           tx_done,
1214                           ring->netdev,
1215                           "tx_info doesn't have valid %s",
1216                            is_xdp ? "xdp frame" : "skb");
1217         else
1218                 netif_err(ring->adapter,
1219                           tx_done,
1220                           ring->netdev,
1221                           "Invalid req_id: %hu\n",
1222                           req_id);
1223
1224         u64_stats_update_begin(&ring->syncp);
1225         ring->tx_stats.bad_req_id++;
1226         u64_stats_update_end(&ring->syncp);
1227
1228         /* Trigger device reset */
1229         ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
1230         set_bit(ENA_FLAG_TRIGGER_RESET, &ring->adapter->flags);
1231         return -EFAULT;
1232 }
1233
1234 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
1235 {
1236         struct ena_tx_buffer *tx_info = NULL;
1237
1238         if (likely(req_id < tx_ring->ring_size)) {
1239                 tx_info = &tx_ring->tx_buffer_info[req_id];
1240                 if (likely(tx_info->skb))
1241                         return 0;
1242         }
1243
1244         return handle_invalid_req_id(tx_ring, req_id, tx_info, false);
1245 }
1246
1247 static int validate_xdp_req_id(struct ena_ring *xdp_ring, u16 req_id)
1248 {
1249         struct ena_tx_buffer *tx_info = NULL;
1250
1251         if (likely(req_id < xdp_ring->ring_size)) {
1252                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1253                 if (likely(tx_info->xdpf))
1254                         return 0;
1255         }
1256
1257         return handle_invalid_req_id(xdp_ring, req_id, tx_info, true);
1258 }
1259
1260 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
1261 {
1262         struct netdev_queue *txq;
1263         bool above_thresh;
1264         u32 tx_bytes = 0;
1265         u32 total_done = 0;
1266         u16 next_to_clean;
1267         u16 req_id;
1268         int tx_pkts = 0;
1269         int rc;
1270
1271         next_to_clean = tx_ring->next_to_clean;
1272         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
1273
1274         while (tx_pkts < budget) {
1275                 struct ena_tx_buffer *tx_info;
1276                 struct sk_buff *skb;
1277
1278                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
1279                                                 &req_id);
1280                 if (rc)
1281                         break;
1282
1283                 rc = validate_tx_req_id(tx_ring, req_id);
1284                 if (rc)
1285                         break;
1286
1287                 tx_info = &tx_ring->tx_buffer_info[req_id];
1288                 skb = tx_info->skb;
1289
1290                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
1291                 prefetch(&skb->end);
1292
1293                 tx_info->skb = NULL;
1294                 tx_info->last_jiffies = 0;
1295
1296                 ena_unmap_tx_buff(tx_ring, tx_info);
1297
1298                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1299                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
1300                           skb);
1301
1302                 tx_bytes += skb->len;
1303                 dev_kfree_skb(skb);
1304                 tx_pkts++;
1305                 total_done += tx_info->tx_descs;
1306
1307                 tx_ring->free_ids[next_to_clean] = req_id;
1308                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1309                                                      tx_ring->ring_size);
1310         }
1311
1312         tx_ring->next_to_clean = next_to_clean;
1313         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
1314         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
1315
1316         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
1317
1318         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1319                   "tx_poll: q %d done. total pkts: %d\n",
1320                   tx_ring->qid, tx_pkts);
1321
1322         /* need to make the rings circular update visible to
1323          * ena_start_xmit() before checking for netif_queue_stopped().
1324          */
1325         smp_mb();
1326
1327         above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1328                                                     ENA_TX_WAKEUP_THRESH);
1329         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
1330                 __netif_tx_lock(txq, smp_processor_id());
1331                 above_thresh =
1332                         ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1333                                                      ENA_TX_WAKEUP_THRESH);
1334                 if (netif_tx_queue_stopped(txq) && above_thresh &&
1335                     test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
1336                         netif_tx_wake_queue(txq);
1337                         u64_stats_update_begin(&tx_ring->syncp);
1338                         tx_ring->tx_stats.queue_wakeup++;
1339                         u64_stats_update_end(&tx_ring->syncp);
1340                 }
1341                 __netif_tx_unlock(txq);
1342         }
1343
1344         return tx_pkts;
1345 }
1346
1347 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, bool frags)
1348 {
1349         struct sk_buff *skb;
1350
1351         if (frags)
1352                 skb = napi_get_frags(rx_ring->napi);
1353         else
1354                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1355                                                 rx_ring->rx_copybreak);
1356
1357         if (unlikely(!skb)) {
1358                 u64_stats_update_begin(&rx_ring->syncp);
1359                 rx_ring->rx_stats.skb_alloc_fail++;
1360                 u64_stats_update_end(&rx_ring->syncp);
1361                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1362                           "Failed to allocate skb. frags: %d\n", frags);
1363                 return NULL;
1364         }
1365
1366         return skb;
1367 }
1368
1369 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
1370                                   struct ena_com_rx_buf_info *ena_bufs,
1371                                   u32 descs,
1372                                   u16 *next_to_clean)
1373 {
1374         struct sk_buff *skb;
1375         struct ena_rx_buffer *rx_info;
1376         u16 len, req_id, buf = 0;
1377         void *va;
1378         int rc;
1379
1380         len = ena_bufs[buf].len;
1381         req_id = ena_bufs[buf].req_id;
1382
1383         rc = validate_rx_req_id(rx_ring, req_id);
1384         if (unlikely(rc < 0))
1385                 return NULL;
1386
1387         rx_info = &rx_ring->rx_buffer_info[req_id];
1388
1389         if (unlikely(!rx_info->page)) {
1390                 netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
1391                           "Page is NULL\n");
1392                 return NULL;
1393         }
1394
1395         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1396                   "rx_info %p page %p\n",
1397                   rx_info, rx_info->page);
1398
1399         /* save virt address of first buffer */
1400         va = page_address(rx_info->page) + rx_info->page_offset;
1401         prefetch(va + NET_IP_ALIGN);
1402
1403         if (len <= rx_ring->rx_copybreak) {
1404                 skb = ena_alloc_skb(rx_ring, false);
1405                 if (unlikely(!skb))
1406                         return NULL;
1407
1408                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1409                           "rx allocated small packet. len %d. data_len %d\n",
1410                           skb->len, skb->data_len);
1411
1412                 /* sync this buffer for CPU use */
1413                 dma_sync_single_for_cpu(rx_ring->dev,
1414                                         dma_unmap_addr(&rx_info->ena_buf, paddr),
1415                                         len,
1416                                         DMA_FROM_DEVICE);
1417                 skb_copy_to_linear_data(skb, va, len);
1418                 dma_sync_single_for_device(rx_ring->dev,
1419                                            dma_unmap_addr(&rx_info->ena_buf, paddr),
1420                                            len,
1421                                            DMA_FROM_DEVICE);
1422
1423                 skb_put(skb, len);
1424                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1425                 rx_ring->free_ids[*next_to_clean] = req_id;
1426                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
1427                                                      rx_ring->ring_size);
1428                 return skb;
1429         }
1430
1431         skb = ena_alloc_skb(rx_ring, true);
1432         if (unlikely(!skb))
1433                 return NULL;
1434
1435         do {
1436                 dma_unmap_page(rx_ring->dev,
1437                                dma_unmap_addr(&rx_info->ena_buf, paddr),
1438                                ENA_PAGE_SIZE, DMA_FROM_DEVICE);
1439
1440                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
1441                                 rx_info->page_offset, len, ENA_PAGE_SIZE);
1442
1443                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1444                           "rx skb updated. len %d. data_len %d\n",
1445                           skb->len, skb->data_len);
1446
1447                 rx_info->page = NULL;
1448
1449                 rx_ring->free_ids[*next_to_clean] = req_id;
1450                 *next_to_clean =
1451                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
1452                                              rx_ring->ring_size);
1453                 if (likely(--descs == 0))
1454                         break;
1455
1456                 buf++;
1457                 len = ena_bufs[buf].len;
1458                 req_id = ena_bufs[buf].req_id;
1459
1460                 rc = validate_rx_req_id(rx_ring, req_id);
1461                 if (unlikely(rc < 0))
1462                         return NULL;
1463
1464                 rx_info = &rx_ring->rx_buffer_info[req_id];
1465         } while (1);
1466
1467         return skb;
1468 }
1469
1470 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
1471  * @adapter: structure containing adapter specific data
1472  * @ena_rx_ctx: received packet context/metadata
1473  * @skb: skb currently being received and modified
1474  */
1475 static void ena_rx_checksum(struct ena_ring *rx_ring,
1476                                    struct ena_com_rx_ctx *ena_rx_ctx,
1477                                    struct sk_buff *skb)
1478 {
1479         /* Rx csum disabled */
1480         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
1481                 skb->ip_summed = CHECKSUM_NONE;
1482                 return;
1483         }
1484
1485         /* For fragmented packets the checksum isn't valid */
1486         if (ena_rx_ctx->frag) {
1487                 skb->ip_summed = CHECKSUM_NONE;
1488                 return;
1489         }
1490
1491         /* if IP and error */
1492         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1493                      (ena_rx_ctx->l3_csum_err))) {
1494                 /* ipv4 checksum error */
1495                 skb->ip_summed = CHECKSUM_NONE;
1496                 u64_stats_update_begin(&rx_ring->syncp);
1497                 rx_ring->rx_stats.bad_csum++;
1498                 u64_stats_update_end(&rx_ring->syncp);
1499                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1500                           "RX IPv4 header checksum error\n");
1501                 return;
1502         }
1503
1504         /* if TCP/UDP */
1505         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1506                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
1507                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
1508                         /* TCP/UDP checksum error */
1509                         u64_stats_update_begin(&rx_ring->syncp);
1510                         rx_ring->rx_stats.bad_csum++;
1511                         u64_stats_update_end(&rx_ring->syncp);
1512                         netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1513                                   "RX L4 checksum error\n");
1514                         skb->ip_summed = CHECKSUM_NONE;
1515                         return;
1516                 }
1517
1518                 if (likely(ena_rx_ctx->l4_csum_checked)) {
1519                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1520                         u64_stats_update_begin(&rx_ring->syncp);
1521                         rx_ring->rx_stats.csum_good++;
1522                         u64_stats_update_end(&rx_ring->syncp);
1523                 } else {
1524                         u64_stats_update_begin(&rx_ring->syncp);
1525                         rx_ring->rx_stats.csum_unchecked++;
1526                         u64_stats_update_end(&rx_ring->syncp);
1527                         skb->ip_summed = CHECKSUM_NONE;
1528                 }
1529         } else {
1530                 skb->ip_summed = CHECKSUM_NONE;
1531                 return;
1532         }
1533
1534 }
1535
1536 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1537                             struct ena_com_rx_ctx *ena_rx_ctx,
1538                             struct sk_buff *skb)
1539 {
1540         enum pkt_hash_types hash_type;
1541
1542         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1543                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1544                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1545
1546                         hash_type = PKT_HASH_TYPE_L4;
1547                 else
1548                         hash_type = PKT_HASH_TYPE_NONE;
1549
1550                 /* Override hash type if the packet is fragmented */
1551                 if (ena_rx_ctx->frag)
1552                         hash_type = PKT_HASH_TYPE_NONE;
1553
1554                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1555         }
1556 }
1557
1558 int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)
1559 {
1560         struct ena_rx_buffer *rx_info;
1561         int ret;
1562
1563         rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1564         xdp->data = page_address(rx_info->page) +
1565                 rx_info->page_offset + rx_ring->rx_headroom;
1566         xdp_set_data_meta_invalid(xdp);
1567         xdp->data_hard_start = page_address(rx_info->page);
1568         xdp->data_end = xdp->data + rx_ring->ena_bufs[0].len;
1569         /* If for some reason we received a bigger packet than
1570          * we expect, then we simply drop it
1571          */
1572         if (unlikely(rx_ring->ena_bufs[0].len > ENA_XDP_MAX_MTU))
1573                 return XDP_DROP;
1574
1575         ret = ena_xdp_execute(rx_ring, xdp, rx_info);
1576
1577         /* The xdp program might expand the headers */
1578         if (ret == XDP_PASS) {
1579                 rx_info->page_offset = xdp->data - xdp->data_hard_start;
1580                 rx_ring->ena_bufs[0].len = xdp->data_end - xdp->data;
1581         }
1582
1583         return ret;
1584 }
1585 /* ena_clean_rx_irq - Cleanup RX irq
1586  * @rx_ring: RX ring to clean
1587  * @napi: napi handler
1588  * @budget: how many packets driver is allowed to clean
1589  *
1590  * Returns the number of cleaned buffers.
1591  */
1592 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1593                             u32 budget)
1594 {
1595         u16 next_to_clean = rx_ring->next_to_clean;
1596         struct ena_com_rx_ctx ena_rx_ctx;
1597         struct ena_adapter *adapter;
1598         u32 res_budget, work_done;
1599         int rx_copybreak_pkt = 0;
1600         int refill_threshold;
1601         struct sk_buff *skb;
1602         int refill_required;
1603         struct xdp_buff xdp;
1604         int total_len = 0;
1605         int xdp_verdict;
1606         int rc = 0;
1607         int i;
1608
1609         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1610                   "%s qid %d\n", __func__, rx_ring->qid);
1611         res_budget = budget;
1612         xdp.rxq = &rx_ring->xdp_rxq;
1613
1614         do {
1615                 xdp_verdict = XDP_PASS;
1616                 skb = NULL;
1617                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1618                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1619                 ena_rx_ctx.descs = 0;
1620                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1621                                     rx_ring->ena_com_io_sq,
1622                                     &ena_rx_ctx);
1623                 if (unlikely(rc))
1624                         goto error;
1625
1626                 if (unlikely(ena_rx_ctx.descs == 0))
1627                         break;
1628
1629                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1630                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1631                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1632                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1633
1634                 if (ena_xdp_present_ring(rx_ring))
1635                         xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1636
1637                 /* allocate skb and fill it */
1638                 if (xdp_verdict == XDP_PASS)
1639                         skb = ena_rx_skb(rx_ring,
1640                                          rx_ring->ena_bufs,
1641                                          ena_rx_ctx.descs,
1642                                          &next_to_clean);
1643
1644                 if (unlikely(!skb)) {
1645                         if (xdp_verdict == XDP_TX) {
1646                                 ena_free_rx_page(rx_ring,
1647                                                  &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id]);
1648                                 res_budget--;
1649                         }
1650                         for (i = 0; i < ena_rx_ctx.descs; i++) {
1651                                 rx_ring->free_ids[next_to_clean] =
1652                                         rx_ring->ena_bufs[i].req_id;
1653                                 next_to_clean =
1654                                         ENA_RX_RING_IDX_NEXT(next_to_clean,
1655                                                              rx_ring->ring_size);
1656                         }
1657                         if (xdp_verdict == XDP_TX || xdp_verdict == XDP_DROP)
1658                                 continue;
1659                         break;
1660                 }
1661
1662                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1663
1664                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1665
1666                 skb_record_rx_queue(skb, rx_ring->qid);
1667
1668                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1669                         total_len += rx_ring->ena_bufs[0].len;
1670                         rx_copybreak_pkt++;
1671                         napi_gro_receive(napi, skb);
1672                 } else {
1673                         total_len += skb->len;
1674                         napi_gro_frags(napi);
1675                 }
1676
1677                 res_budget--;
1678         } while (likely(res_budget));
1679
1680         work_done = budget - res_budget;
1681         rx_ring->per_napi_packets += work_done;
1682         u64_stats_update_begin(&rx_ring->syncp);
1683         rx_ring->rx_stats.bytes += total_len;
1684         rx_ring->rx_stats.cnt += work_done;
1685         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1686         u64_stats_update_end(&rx_ring->syncp);
1687
1688         rx_ring->next_to_clean = next_to_clean;
1689
1690         refill_required = ena_com_free_desc(rx_ring->ena_com_io_sq);
1691         refill_threshold =
1692                 min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1693                       ENA_RX_REFILL_THRESH_PACKET);
1694
1695         /* Optimization, try to batch new rx buffers */
1696         if (refill_required > refill_threshold) {
1697                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1698                 ena_refill_rx_bufs(rx_ring, refill_required);
1699         }
1700
1701         return work_done;
1702
1703 error:
1704         adapter = netdev_priv(rx_ring->netdev);
1705
1706         u64_stats_update_begin(&rx_ring->syncp);
1707         rx_ring->rx_stats.bad_desc_num++;
1708         u64_stats_update_end(&rx_ring->syncp);
1709
1710         /* Too many desc from the device. Trigger reset */
1711         adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1712         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1713
1714         return 0;
1715 }
1716
1717 static void ena_dim_work(struct work_struct *w)
1718 {
1719         struct dim *dim = container_of(w, struct dim, work);
1720         struct dim_cq_moder cur_moder =
1721                 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1722         struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1723
1724         ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1725         dim->state = DIM_START_MEASURE;
1726 }
1727
1728 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1729 {
1730         struct dim_sample dim_sample;
1731         struct ena_ring *rx_ring = ena_napi->rx_ring;
1732
1733         if (!rx_ring->per_napi_packets)
1734                 return;
1735
1736         rx_ring->non_empty_napi_events++;
1737
1738         dim_update_sample(rx_ring->non_empty_napi_events,
1739                           rx_ring->rx_stats.cnt,
1740                           rx_ring->rx_stats.bytes,
1741                           &dim_sample);
1742
1743         net_dim(&ena_napi->dim, dim_sample);
1744
1745         rx_ring->per_napi_packets = 0;
1746 }
1747
1748 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1749                                         struct ena_ring *rx_ring)
1750 {
1751         struct ena_eth_io_intr_reg intr_reg;
1752         u32 rx_interval = 0;
1753         /* Rx ring can be NULL when for XDP tx queues which don't have an
1754          * accompanying rx_ring pair.
1755          */
1756         if (rx_ring)
1757                 rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1758                         rx_ring->smoothed_interval :
1759                         ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1760
1761         /* Update intr register: rx intr delay,
1762          * tx intr delay and interrupt unmask
1763          */
1764         ena_com_update_intr_reg(&intr_reg,
1765                                 rx_interval,
1766                                 tx_ring->smoothed_interval,
1767                                 true);
1768
1769         /* It is a shared MSI-X.
1770          * Tx and Rx CQ have pointer to it.
1771          * So we use one of them to reach the intr reg
1772          * The Tx ring is used because the rx_ring is NULL for XDP queues
1773          */
1774         ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1775 }
1776
1777 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1778                                              struct ena_ring *rx_ring)
1779 {
1780         int cpu = get_cpu();
1781         int numa_node;
1782
1783         /* Check only one ring since the 2 rings are running on the same cpu */
1784         if (likely(tx_ring->cpu == cpu))
1785                 goto out;
1786
1787         numa_node = cpu_to_node(cpu);
1788         put_cpu();
1789
1790         if (numa_node != NUMA_NO_NODE) {
1791                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1792                 if (rx_ring)
1793                         ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1794                                                  numa_node);
1795         }
1796
1797         tx_ring->cpu = cpu;
1798         if (rx_ring)
1799                 rx_ring->cpu = cpu;
1800
1801         return;
1802 out:
1803         put_cpu();
1804 }
1805
1806 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1807 {
1808         u32 total_done = 0;
1809         u16 next_to_clean;
1810         u32 tx_bytes = 0;
1811         int tx_pkts = 0;
1812         u16 req_id;
1813         int rc;
1814
1815         if (unlikely(!xdp_ring))
1816                 return 0;
1817         next_to_clean = xdp_ring->next_to_clean;
1818
1819         while (tx_pkts < budget) {
1820                 struct ena_tx_buffer *tx_info;
1821                 struct xdp_frame *xdpf;
1822
1823                 rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1824                                                 &req_id);
1825                 if (rc)
1826                         break;
1827
1828                 rc = validate_xdp_req_id(xdp_ring, req_id);
1829                 if (rc)
1830                         break;
1831
1832                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1833                 xdpf = tx_info->xdpf;
1834
1835                 tx_info->xdpf = NULL;
1836                 tx_info->last_jiffies = 0;
1837                 ena_unmap_tx_buff(xdp_ring, tx_info);
1838
1839                 netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1840                           "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1841                           xdpf);
1842
1843                 tx_bytes += xdpf->len;
1844                 tx_pkts++;
1845                 total_done += tx_info->tx_descs;
1846
1847                 __free_page(tx_info->xdp_rx_page);
1848                 xdp_ring->free_ids[next_to_clean] = req_id;
1849                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1850                                                      xdp_ring->ring_size);
1851         }
1852
1853         xdp_ring->next_to_clean = next_to_clean;
1854         ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
1855         ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
1856
1857         netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1858                   "tx_poll: q %d done. total pkts: %d\n",
1859                   xdp_ring->qid, tx_pkts);
1860
1861         return tx_pkts;
1862 }
1863
1864 static int ena_io_poll(struct napi_struct *napi, int budget)
1865 {
1866         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1867         struct ena_ring *tx_ring, *rx_ring;
1868         int tx_work_done;
1869         int rx_work_done = 0;
1870         int tx_budget;
1871         int napi_comp_call = 0;
1872         int ret;
1873
1874         tx_ring = ena_napi->tx_ring;
1875         rx_ring = ena_napi->rx_ring;
1876
1877         tx_ring->first_interrupt = ena_napi->first_interrupt;
1878         rx_ring->first_interrupt = ena_napi->first_interrupt;
1879
1880         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1881
1882         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1883             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1884                 napi_complete_done(napi, 0);
1885                 return 0;
1886         }
1887
1888         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1889         /* On netpoll the budget is zero and the handler should only clean the
1890          * tx completions.
1891          */
1892         if (likely(budget))
1893                 rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1894
1895         /* If the device is about to reset or down, avoid unmask
1896          * the interrupt and return 0 so NAPI won't reschedule
1897          */
1898         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1899                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1900                 napi_complete_done(napi, 0);
1901                 ret = 0;
1902
1903         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1904                 napi_comp_call = 1;
1905
1906                 /* Update numa and unmask the interrupt only when schedule
1907                  * from the interrupt context (vs from sk_busy_loop)
1908                  */
1909                 if (napi_complete_done(napi, rx_work_done)) {
1910                         /* We apply adaptive moderation on Rx path only.
1911                          * Tx uses static interrupt moderation.
1912                          */
1913                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1914                                 ena_adjust_adaptive_rx_intr_moderation(ena_napi);
1915
1916                         ena_unmask_interrupt(tx_ring, rx_ring);
1917                 }
1918
1919                 ena_update_ring_numa_node(tx_ring, rx_ring);
1920
1921                 ret = rx_work_done;
1922         } else {
1923                 ret = budget;
1924         }
1925
1926         u64_stats_update_begin(&tx_ring->syncp);
1927         tx_ring->tx_stats.napi_comp += napi_comp_call;
1928         tx_ring->tx_stats.tx_poll++;
1929         u64_stats_update_end(&tx_ring->syncp);
1930
1931         return ret;
1932 }
1933
1934 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1935 {
1936         struct ena_adapter *adapter = (struct ena_adapter *)data;
1937
1938         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1939
1940         /* Don't call the aenq handler before probe is done */
1941         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1942                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
1943
1944         return IRQ_HANDLED;
1945 }
1946
1947 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1948  * @irq: interrupt number
1949  * @data: pointer to a network interface private napi device structure
1950  */
1951 static irqreturn_t ena_intr_msix_io(int irq, void *data)
1952 {
1953         struct ena_napi *ena_napi = data;
1954
1955         ena_napi->first_interrupt = true;
1956
1957         napi_schedule_irqoff(&ena_napi->napi);
1958
1959         return IRQ_HANDLED;
1960 }
1961
1962 /* Reserve a single MSI-X vector for management (admin + aenq).
1963  * plus reserve one vector for each potential io queue.
1964  * the number of potential io queues is the minimum of what the device
1965  * supports and the number of vCPUs.
1966  */
1967 static int ena_enable_msix(struct ena_adapter *adapter)
1968 {
1969         int msix_vecs, irq_cnt;
1970
1971         if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1972                 netif_err(adapter, probe, adapter->netdev,
1973                           "Error, MSI-X is already enabled\n");
1974                 return -EPERM;
1975         }
1976
1977         /* Reserved the max msix vectors we might need */
1978         msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
1979         netif_dbg(adapter, probe, adapter->netdev,
1980                   "trying to enable MSI-X, vectors %d\n", msix_vecs);
1981
1982         irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
1983                                         msix_vecs, PCI_IRQ_MSIX);
1984
1985         if (irq_cnt < 0) {
1986                 netif_err(adapter, probe, adapter->netdev,
1987                           "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
1988                 return -ENOSPC;
1989         }
1990
1991         if (irq_cnt != msix_vecs) {
1992                 netif_notice(adapter, probe, adapter->netdev,
1993                              "enable only %d MSI-X (out of %d), reduce the number of queues\n",
1994                              irq_cnt, msix_vecs);
1995                 adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
1996         }
1997
1998         if (ena_init_rx_cpu_rmap(adapter))
1999                 netif_warn(adapter, probe, adapter->netdev,
2000                            "Failed to map IRQs to CPUs\n");
2001
2002         adapter->msix_vecs = irq_cnt;
2003         set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2004
2005         return 0;
2006 }
2007
2008 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2009 {
2010         u32 cpu;
2011
2012         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2013                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2014                  pci_name(adapter->pdev));
2015         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2016                 ena_intr_msix_mgmnt;
2017         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2018         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2019                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2020         cpu = cpumask_first(cpu_online_mask);
2021         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2022         cpumask_set_cpu(cpu,
2023                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2024 }
2025
2026 static void ena_setup_io_intr(struct ena_adapter *adapter)
2027 {
2028         struct net_device *netdev;
2029         int irq_idx, i, cpu;
2030         int io_queue_count;
2031
2032         netdev = adapter->netdev;
2033         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2034
2035         for (i = 0; i < io_queue_count; i++) {
2036                 irq_idx = ENA_IO_IRQ_IDX(i);
2037                 cpu = i % num_online_cpus();
2038
2039                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2040                          "%s-Tx-Rx-%d", netdev->name, i);
2041                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2042                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2043                 adapter->irq_tbl[irq_idx].vector =
2044                         pci_irq_vector(adapter->pdev, irq_idx);
2045                 adapter->irq_tbl[irq_idx].cpu = cpu;
2046
2047                 cpumask_set_cpu(cpu,
2048                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
2049         }
2050 }
2051
2052 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2053 {
2054         unsigned long flags = 0;
2055         struct ena_irq *irq;
2056         int rc;
2057
2058         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2059         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2060                          irq->data);
2061         if (rc) {
2062                 netif_err(adapter, probe, adapter->netdev,
2063                           "failed to request admin irq\n");
2064                 return rc;
2065         }
2066
2067         netif_dbg(adapter, probe, adapter->netdev,
2068                   "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2069                   irq->affinity_hint_mask.bits[0], irq->vector);
2070
2071         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2072
2073         return rc;
2074 }
2075
2076 static int ena_request_io_irq(struct ena_adapter *adapter)
2077 {
2078         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2079         unsigned long flags = 0;
2080         struct ena_irq *irq;
2081         int rc = 0, i, k;
2082
2083         if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2084                 netif_err(adapter, ifup, adapter->netdev,
2085                           "Failed to request I/O IRQ: MSI-X is not enabled\n");
2086                 return -EINVAL;
2087         }
2088
2089         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2090                 irq = &adapter->irq_tbl[i];
2091                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2092                                  irq->data);
2093                 if (rc) {
2094                         netif_err(adapter, ifup, adapter->netdev,
2095                                   "Failed to request I/O IRQ. index %d rc %d\n",
2096                                    i, rc);
2097                         goto err;
2098                 }
2099
2100                 netif_dbg(adapter, ifup, adapter->netdev,
2101                           "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2102                           i, irq->affinity_hint_mask.bits[0], irq->vector);
2103
2104                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2105         }
2106
2107         return rc;
2108
2109 err:
2110         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2111                 irq = &adapter->irq_tbl[k];
2112                 free_irq(irq->vector, irq->data);
2113         }
2114
2115         return rc;
2116 }
2117
2118 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2119 {
2120         struct ena_irq *irq;
2121
2122         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2123         synchronize_irq(irq->vector);
2124         irq_set_affinity_hint(irq->vector, NULL);
2125         free_irq(irq->vector, irq->data);
2126 }
2127
2128 static void ena_free_io_irq(struct ena_adapter *adapter)
2129 {
2130         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2131         struct ena_irq *irq;
2132         int i;
2133
2134 #ifdef CONFIG_RFS_ACCEL
2135         if (adapter->msix_vecs >= 1) {
2136                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2137                 adapter->netdev->rx_cpu_rmap = NULL;
2138         }
2139 #endif /* CONFIG_RFS_ACCEL */
2140
2141         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2142                 irq = &adapter->irq_tbl[i];
2143                 irq_set_affinity_hint(irq->vector, NULL);
2144                 free_irq(irq->vector, irq->data);
2145         }
2146 }
2147
2148 static void ena_disable_msix(struct ena_adapter *adapter)
2149 {
2150         if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2151                 pci_free_irq_vectors(adapter->pdev);
2152 }
2153
2154 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2155 {
2156         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2157         int i;
2158
2159         if (!netif_running(adapter->netdev))
2160                 return;
2161
2162         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2163                 synchronize_irq(adapter->irq_tbl[i].vector);
2164 }
2165
2166 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2167                                   int first_index,
2168                                   int count)
2169 {
2170         int i;
2171
2172         for (i = first_index; i < first_index + count; i++) {
2173                 /* Check if napi was initialized before */
2174                 if (!ENA_IS_XDP_INDEX(adapter, i) ||
2175                     adapter->ena_napi[i].xdp_ring)
2176                         netif_napi_del(&adapter->ena_napi[i].napi);
2177                 else
2178                         WARN_ON(ENA_IS_XDP_INDEX(adapter, i) &&
2179                                 adapter->ena_napi[i].xdp_ring);
2180         }
2181 }
2182
2183 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2184                                    int first_index, int count)
2185 {
2186         struct ena_napi *napi = {0};
2187         int i;
2188
2189         for (i = first_index; i < first_index + count; i++) {
2190                 napi = &adapter->ena_napi[i];
2191
2192                 netif_napi_add(adapter->netdev,
2193                                &adapter->ena_napi[i].napi,
2194                                ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll,
2195                                ENA_NAPI_BUDGET);
2196
2197                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
2198                         napi->rx_ring = &adapter->rx_ring[i];
2199                         napi->tx_ring = &adapter->tx_ring[i];
2200                 } else {
2201                         napi->xdp_ring = &adapter->tx_ring[i];
2202                 }
2203                 napi->qid = i;
2204         }
2205 }
2206
2207 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2208                                       int first_index,
2209                                       int count)
2210 {
2211         int i;
2212
2213         for (i = first_index; i < first_index + count; i++)
2214                 napi_disable(&adapter->ena_napi[i].napi);
2215 }
2216
2217 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2218                                      int first_index,
2219                                      int count)
2220 {
2221         int i;
2222
2223         for (i = first_index; i < first_index + count; i++)
2224                 napi_enable(&adapter->ena_napi[i].napi);
2225 }
2226
2227 /* Configure the Rx forwarding */
2228 static int ena_rss_configure(struct ena_adapter *adapter)
2229 {
2230         struct ena_com_dev *ena_dev = adapter->ena_dev;
2231         int rc;
2232
2233         /* In case the RSS table wasn't initialized by probe */
2234         if (!ena_dev->rss.tbl_log_size) {
2235                 rc = ena_rss_init_default(adapter);
2236                 if (rc && (rc != -EOPNOTSUPP)) {
2237                         netif_err(adapter, ifup, adapter->netdev,
2238                                         "Failed to init RSS rc: %d\n", rc);
2239                         return rc;
2240                 }
2241         }
2242
2243         /* Set indirect table */
2244         rc = ena_com_indirect_table_set(ena_dev);
2245         if (unlikely(rc && rc != -EOPNOTSUPP))
2246                 return rc;
2247
2248         /* Configure hash function (if supported) */
2249         rc = ena_com_set_hash_function(ena_dev);
2250         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2251                 return rc;
2252
2253         /* Configure hash inputs (if supported) */
2254         rc = ena_com_set_hash_ctrl(ena_dev);
2255         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2256                 return rc;
2257
2258         return 0;
2259 }
2260
2261 static int ena_up_complete(struct ena_adapter *adapter)
2262 {
2263         int rc;
2264
2265         rc = ena_rss_configure(adapter);
2266         if (rc)
2267                 return rc;
2268
2269         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2270
2271         ena_refill_all_rx_bufs(adapter);
2272
2273         /* enable transmits */
2274         netif_tx_start_all_queues(adapter->netdev);
2275
2276         ena_napi_enable_in_range(adapter,
2277                                  0,
2278                                  adapter->xdp_num_queues + adapter->num_io_queues);
2279
2280         return 0;
2281 }
2282
2283 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2284 {
2285         struct ena_com_create_io_ctx ctx;
2286         struct ena_com_dev *ena_dev;
2287         struct ena_ring *tx_ring;
2288         u32 msix_vector;
2289         u16 ena_qid;
2290         int rc;
2291
2292         ena_dev = adapter->ena_dev;
2293
2294         tx_ring = &adapter->tx_ring[qid];
2295         msix_vector = ENA_IO_IRQ_IDX(qid);
2296         ena_qid = ENA_IO_TXQ_IDX(qid);
2297
2298         memset(&ctx, 0x0, sizeof(ctx));
2299
2300         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2301         ctx.qid = ena_qid;
2302         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2303         ctx.msix_vector = msix_vector;
2304         ctx.queue_size = tx_ring->ring_size;
2305         ctx.numa_node = cpu_to_node(tx_ring->cpu);
2306
2307         rc = ena_com_create_io_queue(ena_dev, &ctx);
2308         if (rc) {
2309                 netif_err(adapter, ifup, adapter->netdev,
2310                           "Failed to create I/O TX queue num %d rc: %d\n",
2311                            qid, rc);
2312                 return rc;
2313         }
2314
2315         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2316                                      &tx_ring->ena_com_io_sq,
2317                                      &tx_ring->ena_com_io_cq);
2318         if (rc) {
2319                 netif_err(adapter, ifup, adapter->netdev,
2320                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2321                           qid, rc);
2322                 ena_com_destroy_io_queue(ena_dev, ena_qid);
2323                 return rc;
2324         }
2325
2326         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2327         return rc;
2328 }
2329
2330 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2331                                             int first_index, int count)
2332 {
2333         struct ena_com_dev *ena_dev = adapter->ena_dev;
2334         int rc, i;
2335
2336         for (i = first_index; i < first_index + count; i++) {
2337                 rc = ena_create_io_tx_queue(adapter, i);
2338                 if (rc)
2339                         goto create_err;
2340         }
2341
2342         return 0;
2343
2344 create_err:
2345         while (i-- > first_index)
2346                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2347
2348         return rc;
2349 }
2350
2351 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2352 {
2353         struct ena_com_dev *ena_dev;
2354         struct ena_com_create_io_ctx ctx;
2355         struct ena_ring *rx_ring;
2356         u32 msix_vector;
2357         u16 ena_qid;
2358         int rc;
2359
2360         ena_dev = adapter->ena_dev;
2361
2362         rx_ring = &adapter->rx_ring[qid];
2363         msix_vector = ENA_IO_IRQ_IDX(qid);
2364         ena_qid = ENA_IO_RXQ_IDX(qid);
2365
2366         memset(&ctx, 0x0, sizeof(ctx));
2367
2368         ctx.qid = ena_qid;
2369         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2370         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2371         ctx.msix_vector = msix_vector;
2372         ctx.queue_size = rx_ring->ring_size;
2373         ctx.numa_node = cpu_to_node(rx_ring->cpu);
2374
2375         rc = ena_com_create_io_queue(ena_dev, &ctx);
2376         if (rc) {
2377                 netif_err(adapter, ifup, adapter->netdev,
2378                           "Failed to create I/O RX queue num %d rc: %d\n",
2379                           qid, rc);
2380                 return rc;
2381         }
2382
2383         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2384                                      &rx_ring->ena_com_io_sq,
2385                                      &rx_ring->ena_com_io_cq);
2386         if (rc) {
2387                 netif_err(adapter, ifup, adapter->netdev,
2388                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2389                           qid, rc);
2390                 goto err;
2391         }
2392
2393         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2394
2395         return rc;
2396 err:
2397         ena_com_destroy_io_queue(ena_dev, ena_qid);
2398         return rc;
2399 }
2400
2401 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2402 {
2403         struct ena_com_dev *ena_dev = adapter->ena_dev;
2404         int rc, i;
2405
2406         for (i = 0; i < adapter->num_io_queues; i++) {
2407                 rc = ena_create_io_rx_queue(adapter, i);
2408                 if (rc)
2409                         goto create_err;
2410                 INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2411         }
2412
2413         return 0;
2414
2415 create_err:
2416         while (i--) {
2417                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
2418                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2419         }
2420
2421         return rc;
2422 }
2423
2424 static void set_io_rings_size(struct ena_adapter *adapter,
2425                               int new_tx_size,
2426                               int new_rx_size)
2427 {
2428         int i;
2429
2430         for (i = 0; i < adapter->num_io_queues; i++) {
2431                 adapter->tx_ring[i].ring_size = new_tx_size;
2432                 adapter->rx_ring[i].ring_size = new_rx_size;
2433         }
2434 }
2435
2436 /* This function allows queue allocation to backoff when the system is
2437  * low on memory. If there is not enough memory to allocate io queues
2438  * the driver will try to allocate smaller queues.
2439  *
2440  * The backoff algorithm is as follows:
2441  *  1. Try to allocate TX and RX and if successful.
2442  *  1.1. return success
2443  *
2444  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2445  *
2446  *  3. If TX or RX is smaller than 256
2447  *  3.1. return failure.
2448  *  4. else
2449  *  4.1. go back to 1.
2450  */
2451 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2452 {
2453         int rc, cur_rx_ring_size, cur_tx_ring_size;
2454         int new_rx_ring_size, new_tx_ring_size;
2455
2456         /* current queue sizes might be set to smaller than the requested
2457          * ones due to past queue allocation failures.
2458          */
2459         set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2460                         adapter->requested_rx_ring_size);
2461
2462         while (1) {
2463                 if (ena_xdp_present(adapter)) {
2464                         rc = ena_setup_and_create_all_xdp_queues(adapter);
2465
2466                         if (rc)
2467                                 goto err_setup_tx;
2468                 }
2469                 rc = ena_setup_tx_resources_in_range(adapter,
2470                                                      0,
2471                                                      adapter->num_io_queues);
2472                 if (rc)
2473                         goto err_setup_tx;
2474
2475                 rc = ena_create_io_tx_queues_in_range(adapter,
2476                                                       0,
2477                                                       adapter->num_io_queues);
2478                 if (rc)
2479                         goto err_create_tx_queues;
2480
2481                 rc = ena_setup_all_rx_resources(adapter);
2482                 if (rc)
2483                         goto err_setup_rx;
2484
2485                 rc = ena_create_all_io_rx_queues(adapter);
2486                 if (rc)
2487                         goto err_create_rx_queues;
2488
2489                 return 0;
2490
2491 err_create_rx_queues:
2492                 ena_free_all_io_rx_resources(adapter);
2493 err_setup_rx:
2494                 ena_destroy_all_tx_queues(adapter);
2495 err_create_tx_queues:
2496                 ena_free_all_io_tx_resources(adapter);
2497 err_setup_tx:
2498                 if (rc != -ENOMEM) {
2499                         netif_err(adapter, ifup, adapter->netdev,
2500                                   "Queue creation failed with error code %d\n",
2501                                    rc);
2502                         return rc;
2503                 }
2504
2505                 cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2506                 cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2507
2508                 netif_err(adapter, ifup, adapter->netdev,
2509                           "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2510                           cur_tx_ring_size, cur_rx_ring_size);
2511
2512                 new_tx_ring_size = cur_tx_ring_size;
2513                 new_rx_ring_size = cur_rx_ring_size;
2514
2515                 /* Decrease the size of the larger queue, or
2516                  * decrease both if they are the same size.
2517                  */
2518                 if (cur_rx_ring_size <= cur_tx_ring_size)
2519                         new_tx_ring_size = cur_tx_ring_size / 2;
2520                 if (cur_rx_ring_size >= cur_tx_ring_size)
2521                         new_rx_ring_size = cur_rx_ring_size / 2;
2522
2523                 if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2524                                 new_rx_ring_size < ENA_MIN_RING_SIZE) {
2525                         netif_err(adapter, ifup, adapter->netdev,
2526                                   "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2527                                   ENA_MIN_RING_SIZE);
2528                         return rc;
2529                 }
2530
2531                 netif_err(adapter, ifup, adapter->netdev,
2532                           "Retrying queue creation with sizes TX=%d, RX=%d\n",
2533                           new_tx_ring_size,
2534                           new_rx_ring_size);
2535
2536                 set_io_rings_size(adapter, new_tx_ring_size,
2537                                   new_rx_ring_size);
2538         }
2539 }
2540
2541 static int ena_up(struct ena_adapter *adapter)
2542 {
2543         int io_queue_count, rc, i;
2544
2545         netdev_dbg(adapter->netdev, "%s\n", __func__);
2546
2547         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2548         ena_setup_io_intr(adapter);
2549
2550         /* napi poll functions should be initialized before running
2551          * request_irq(), to handle a rare condition where there is a pending
2552          * interrupt, causing the ISR to fire immediately while the poll
2553          * function wasn't set yet, causing a null dereference
2554          */
2555         ena_init_napi_in_range(adapter, 0, io_queue_count);
2556
2557         rc = ena_request_io_irq(adapter);
2558         if (rc)
2559                 goto err_req_irq;
2560
2561         rc = create_queues_with_size_backoff(adapter);
2562         if (rc)
2563                 goto err_create_queues_with_backoff;
2564
2565         rc = ena_up_complete(adapter);
2566         if (rc)
2567                 goto err_up;
2568
2569         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2570                 netif_carrier_on(adapter->netdev);
2571
2572         u64_stats_update_begin(&adapter->syncp);
2573         adapter->dev_stats.interface_up++;
2574         u64_stats_update_end(&adapter->syncp);
2575
2576         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2577
2578         /* Enable completion queues interrupt */
2579         for (i = 0; i < adapter->num_io_queues; i++)
2580                 ena_unmask_interrupt(&adapter->tx_ring[i],
2581                                      &adapter->rx_ring[i]);
2582
2583         /* schedule napi in case we had pending packets
2584          * from the last time we disable napi
2585          */
2586         for (i = 0; i < io_queue_count; i++)
2587                 napi_schedule(&adapter->ena_napi[i].napi);
2588
2589         return rc;
2590
2591 err_up:
2592         ena_destroy_all_tx_queues(adapter);
2593         ena_free_all_io_tx_resources(adapter);
2594         ena_destroy_all_rx_queues(adapter);
2595         ena_free_all_io_rx_resources(adapter);
2596 err_create_queues_with_backoff:
2597         ena_free_io_irq(adapter);
2598 err_req_irq:
2599         ena_del_napi_in_range(adapter, 0, io_queue_count);
2600
2601         return rc;
2602 }
2603
2604 static void ena_down(struct ena_adapter *adapter)
2605 {
2606         int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2607
2608         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2609
2610         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2611
2612         u64_stats_update_begin(&adapter->syncp);
2613         adapter->dev_stats.interface_down++;
2614         u64_stats_update_end(&adapter->syncp);
2615
2616         netif_carrier_off(adapter->netdev);
2617         netif_tx_disable(adapter->netdev);
2618
2619         /* After this point the napi handler won't enable the tx queue */
2620         ena_napi_disable_in_range(adapter, 0, io_queue_count);
2621
2622         /* After destroy the queue there won't be any new interrupts */
2623
2624         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2625                 int rc;
2626
2627                 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2628                 if (rc)
2629                         dev_err(&adapter->pdev->dev, "Device reset failed\n");
2630                 /* stop submitting admin commands on a device that was reset */
2631                 ena_com_set_admin_running_state(adapter->ena_dev, false);
2632         }
2633
2634         ena_destroy_all_io_queues(adapter);
2635
2636         ena_disable_io_intr_sync(adapter);
2637         ena_free_io_irq(adapter);
2638         ena_del_napi_in_range(adapter, 0, io_queue_count);
2639
2640         ena_free_all_tx_bufs(adapter);
2641         ena_free_all_rx_bufs(adapter);
2642         ena_free_all_io_tx_resources(adapter);
2643         ena_free_all_io_rx_resources(adapter);
2644 }
2645
2646 /* ena_open - Called when a network interface is made active
2647  * @netdev: network interface device structure
2648  *
2649  * Returns 0 on success, negative value on failure
2650  *
2651  * The open entry point is called when a network interface is made
2652  * active by the system (IFF_UP).  At this point all resources needed
2653  * for transmit and receive operations are allocated, the interrupt
2654  * handler is registered with the OS, the watchdog timer is started,
2655  * and the stack is notified that the interface is ready.
2656  */
2657 static int ena_open(struct net_device *netdev)
2658 {
2659         struct ena_adapter *adapter = netdev_priv(netdev);
2660         int rc;
2661
2662         /* Notify the stack of the actual queue counts. */
2663         rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2664         if (rc) {
2665                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2666                 return rc;
2667         }
2668
2669         rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2670         if (rc) {
2671                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2672                 return rc;
2673         }
2674
2675         rc = ena_up(adapter);
2676         if (rc)
2677                 return rc;
2678
2679         return rc;
2680 }
2681
2682 /* ena_close - Disables a network interface
2683  * @netdev: network interface device structure
2684  *
2685  * Returns 0, this is not allowed to fail
2686  *
2687  * The close entry point is called when an interface is de-activated
2688  * by the OS.  The hardware is still under the drivers control, but
2689  * needs to be disabled.  A global MAC reset is issued to stop the
2690  * hardware, and all transmit and receive resources are freed.
2691  */
2692 static int ena_close(struct net_device *netdev)
2693 {
2694         struct ena_adapter *adapter = netdev_priv(netdev);
2695
2696         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2697
2698         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2699                 return 0;
2700
2701         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2702                 ena_down(adapter);
2703
2704         /* Check for device status and issue reset if needed*/
2705         check_for_admin_com_state(adapter);
2706         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2707                 netif_err(adapter, ifdown, adapter->netdev,
2708                           "Destroy failure, restarting device\n");
2709                 ena_dump_stats_to_dmesg(adapter);
2710                 /* rtnl lock already obtained in dev_ioctl() layer */
2711                 ena_destroy_device(adapter, false);
2712                 ena_restore_device(adapter);
2713         }
2714
2715         return 0;
2716 }
2717
2718 int ena_update_queue_sizes(struct ena_adapter *adapter,
2719                            u32 new_tx_size,
2720                            u32 new_rx_size)
2721 {
2722         bool dev_was_up;
2723
2724         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2725         ena_close(adapter->netdev);
2726         adapter->requested_tx_ring_size = new_tx_size;
2727         adapter->requested_rx_ring_size = new_rx_size;
2728         ena_init_io_rings(adapter,
2729                           0,
2730                           adapter->xdp_num_queues +
2731                           adapter->num_io_queues);
2732         return dev_was_up ? ena_up(adapter) : 0;
2733 }
2734
2735 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2736 {
2737         struct ena_com_dev *ena_dev = adapter->ena_dev;
2738         int prev_channel_count;
2739         bool dev_was_up;
2740
2741         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2742         ena_close(adapter->netdev);
2743         prev_channel_count = adapter->num_io_queues;
2744         adapter->num_io_queues = new_channel_count;
2745         if (ena_xdp_present(adapter) &&
2746             ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2747                 adapter->xdp_first_ring = new_channel_count;
2748                 adapter->xdp_num_queues = new_channel_count;
2749                 if (prev_channel_count > new_channel_count)
2750                         ena_xdp_exchange_program_rx_in_range(adapter,
2751                                                              NULL,
2752                                                              new_channel_count,
2753                                                              prev_channel_count);
2754                 else
2755                         ena_xdp_exchange_program_rx_in_range(adapter,
2756                                                              adapter->xdp_bpf_prog,
2757                                                              prev_channel_count,
2758                                                              new_channel_count);
2759         }
2760
2761         /* We need to destroy the rss table so that the indirection
2762          * table will be reinitialized by ena_up()
2763          */
2764         ena_com_rss_destroy(ena_dev);
2765         ena_init_io_rings(adapter,
2766                           0,
2767                           adapter->xdp_num_queues +
2768                           adapter->num_io_queues);
2769         return dev_was_up ? ena_open(adapter->netdev) : 0;
2770 }
2771
2772 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct sk_buff *skb)
2773 {
2774         u32 mss = skb_shinfo(skb)->gso_size;
2775         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2776         u8 l4_protocol = 0;
2777
2778         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2779                 ena_tx_ctx->l4_csum_enable = 1;
2780                 if (mss) {
2781                         ena_tx_ctx->tso_enable = 1;
2782                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2783                         ena_tx_ctx->l4_csum_partial = 0;
2784                 } else {
2785                         ena_tx_ctx->tso_enable = 0;
2786                         ena_meta->l4_hdr_len = 0;
2787                         ena_tx_ctx->l4_csum_partial = 1;
2788                 }
2789
2790                 switch (ip_hdr(skb)->version) {
2791                 case IPVERSION:
2792                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2793                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
2794                                 ena_tx_ctx->df = 1;
2795                         if (mss)
2796                                 ena_tx_ctx->l3_csum_enable = 1;
2797                         l4_protocol = ip_hdr(skb)->protocol;
2798                         break;
2799                 case 6:
2800                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2801                         l4_protocol = ipv6_hdr(skb)->nexthdr;
2802                         break;
2803                 default:
2804                         break;
2805                 }
2806
2807                 if (l4_protocol == IPPROTO_TCP)
2808                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2809                 else
2810                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2811
2812                 ena_meta->mss = mss;
2813                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
2814                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
2815                 ena_tx_ctx->meta_valid = 1;
2816
2817         } else {
2818                 ena_tx_ctx->meta_valid = 0;
2819         }
2820 }
2821
2822 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2823                                        struct sk_buff *skb)
2824 {
2825         int num_frags, header_len, rc;
2826
2827         num_frags = skb_shinfo(skb)->nr_frags;
2828         header_len = skb_headlen(skb);
2829
2830         if (num_frags < tx_ring->sgl_size)
2831                 return 0;
2832
2833         if ((num_frags == tx_ring->sgl_size) &&
2834             (header_len < tx_ring->tx_max_header_size))
2835                 return 0;
2836
2837         u64_stats_update_begin(&tx_ring->syncp);
2838         tx_ring->tx_stats.linearize++;
2839         u64_stats_update_end(&tx_ring->syncp);
2840
2841         rc = skb_linearize(skb);
2842         if (unlikely(rc)) {
2843                 u64_stats_update_begin(&tx_ring->syncp);
2844                 tx_ring->tx_stats.linearize_failed++;
2845                 u64_stats_update_end(&tx_ring->syncp);
2846         }
2847
2848         return rc;
2849 }
2850
2851 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2852                           struct ena_tx_buffer *tx_info,
2853                           struct sk_buff *skb,
2854                           void **push_hdr,
2855                           u16 *header_len)
2856 {
2857         struct ena_adapter *adapter = tx_ring->adapter;
2858         struct ena_com_buf *ena_buf;
2859         dma_addr_t dma;
2860         u32 skb_head_len, frag_len, last_frag;
2861         u16 push_len = 0;
2862         u16 delta = 0;
2863         int i = 0;
2864
2865         skb_head_len = skb_headlen(skb);
2866         tx_info->skb = skb;
2867         ena_buf = tx_info->bufs;
2868
2869         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2870                 /* When the device is LLQ mode, the driver will copy
2871                  * the header into the device memory space.
2872                  * the ena_com layer assume the header is in a linear
2873                  * memory space.
2874                  * This assumption might be wrong since part of the header
2875                  * can be in the fragmented buffers.
2876                  * Use skb_header_pointer to make sure the header is in a
2877                  * linear memory space.
2878                  */
2879
2880                 push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2881                 *push_hdr = skb_header_pointer(skb, 0, push_len,
2882                                                tx_ring->push_buf_intermediate_buf);
2883                 *header_len = push_len;
2884                 if (unlikely(skb->data != *push_hdr)) {
2885                         u64_stats_update_begin(&tx_ring->syncp);
2886                         tx_ring->tx_stats.llq_buffer_copy++;
2887                         u64_stats_update_end(&tx_ring->syncp);
2888
2889                         delta = push_len - skb_head_len;
2890                 }
2891         } else {
2892                 *push_hdr = NULL;
2893                 *header_len = min_t(u32, skb_head_len,
2894                                     tx_ring->tx_max_header_size);
2895         }
2896
2897         netif_dbg(adapter, tx_queued, adapter->netdev,
2898                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2899                   *push_hdr, push_len);
2900
2901         if (skb_head_len > push_len) {
2902                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2903                                      skb_head_len - push_len, DMA_TO_DEVICE);
2904                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2905                         goto error_report_dma_error;
2906
2907                 ena_buf->paddr = dma;
2908                 ena_buf->len = skb_head_len - push_len;
2909
2910                 ena_buf++;
2911                 tx_info->num_of_bufs++;
2912                 tx_info->map_linear_data = 1;
2913         } else {
2914                 tx_info->map_linear_data = 0;
2915         }
2916
2917         last_frag = skb_shinfo(skb)->nr_frags;
2918
2919         for (i = 0; i < last_frag; i++) {
2920                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2921
2922                 frag_len = skb_frag_size(frag);
2923
2924                 if (unlikely(delta >= frag_len)) {
2925                         delta -= frag_len;
2926                         continue;
2927                 }
2928
2929                 dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
2930                                        frag_len - delta, DMA_TO_DEVICE);
2931                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2932                         goto error_report_dma_error;
2933
2934                 ena_buf->paddr = dma;
2935                 ena_buf->len = frag_len - delta;
2936                 ena_buf++;
2937                 tx_info->num_of_bufs++;
2938                 delta = 0;
2939         }
2940
2941         return 0;
2942
2943 error_report_dma_error:
2944         u64_stats_update_begin(&tx_ring->syncp);
2945         tx_ring->tx_stats.dma_mapping_err++;
2946         u64_stats_update_end(&tx_ring->syncp);
2947         netdev_warn(adapter->netdev, "failed to map skb\n");
2948
2949         tx_info->skb = NULL;
2950
2951         tx_info->num_of_bufs += i;
2952         ena_unmap_tx_buff(tx_ring, tx_info);
2953
2954         return -EINVAL;
2955 }
2956
2957 /* Called with netif_tx_lock. */
2958 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
2959 {
2960         struct ena_adapter *adapter = netdev_priv(dev);
2961         struct ena_tx_buffer *tx_info;
2962         struct ena_com_tx_ctx ena_tx_ctx;
2963         struct ena_ring *tx_ring;
2964         struct netdev_queue *txq;
2965         void *push_hdr;
2966         u16 next_to_use, req_id, header_len;
2967         int qid, rc;
2968
2969         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
2970         /*  Determine which tx ring we will be placed on */
2971         qid = skb_get_queue_mapping(skb);
2972         tx_ring = &adapter->tx_ring[qid];
2973         txq = netdev_get_tx_queue(dev, qid);
2974
2975         rc = ena_check_and_linearize_skb(tx_ring, skb);
2976         if (unlikely(rc))
2977                 goto error_drop_packet;
2978
2979         skb_tx_timestamp(skb);
2980
2981         next_to_use = tx_ring->next_to_use;
2982         req_id = tx_ring->free_ids[next_to_use];
2983         tx_info = &tx_ring->tx_buffer_info[req_id];
2984         tx_info->num_of_bufs = 0;
2985
2986         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
2987
2988         rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
2989         if (unlikely(rc))
2990                 goto error_drop_packet;
2991
2992         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2993         ena_tx_ctx.ena_bufs = tx_info->bufs;
2994         ena_tx_ctx.push_header = push_hdr;
2995         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2996         ena_tx_ctx.req_id = req_id;
2997         ena_tx_ctx.header_len = header_len;
2998
2999         /* set flags and meta data */
3000         ena_tx_csum(&ena_tx_ctx, skb);
3001
3002         rc = ena_xmit_common(dev,
3003                              tx_ring,
3004                              tx_info,
3005                              &ena_tx_ctx,
3006                              next_to_use,
3007                              skb->len);
3008         if (rc)
3009                 goto error_unmap_dma;
3010
3011         netdev_tx_sent_queue(txq, skb->len);
3012
3013         /* stop the queue when no more space available, the packet can have up
3014          * to sgl_size + 2. one for the meta descriptor and one for header
3015          * (if the header is larger than tx_max_header_size).
3016          */
3017         if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3018                                                    tx_ring->sgl_size + 2))) {
3019                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3020                           __func__, qid);
3021
3022                 netif_tx_stop_queue(txq);
3023                 u64_stats_update_begin(&tx_ring->syncp);
3024                 tx_ring->tx_stats.queue_stop++;
3025                 u64_stats_update_end(&tx_ring->syncp);
3026
3027                 /* There is a rare condition where this function decide to
3028                  * stop the queue but meanwhile clean_tx_irq updates
3029                  * next_to_completion and terminates.
3030                  * The queue will remain stopped forever.
3031                  * To solve this issue add a mb() to make sure that
3032                  * netif_tx_stop_queue() write is vissible before checking if
3033                  * there is additional space in the queue.
3034                  */
3035                 smp_mb();
3036
3037                 if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3038                                                  ENA_TX_WAKEUP_THRESH)) {
3039                         netif_tx_wake_queue(txq);
3040                         u64_stats_update_begin(&tx_ring->syncp);
3041                         tx_ring->tx_stats.queue_wakeup++;
3042                         u64_stats_update_end(&tx_ring->syncp);
3043                 }
3044         }
3045
3046         if (netif_xmit_stopped(txq) || !netdev_xmit_more()) {
3047                 /* trigger the dma engine. ena_com_write_sq_doorbell()
3048                  * has a mb
3049                  */
3050                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
3051                 u64_stats_update_begin(&tx_ring->syncp);
3052                 tx_ring->tx_stats.doorbells++;
3053                 u64_stats_update_end(&tx_ring->syncp);
3054         }
3055
3056         return NETDEV_TX_OK;
3057
3058 error_unmap_dma:
3059         ena_unmap_tx_buff(tx_ring, tx_info);
3060         tx_info->skb = NULL;
3061
3062 error_drop_packet:
3063         dev_kfree_skb(skb);
3064         return NETDEV_TX_OK;
3065 }
3066
3067 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3068                             struct net_device *sb_dev)
3069 {
3070         u16 qid;
3071         /* we suspect that this is good for in--kernel network services that
3072          * want to loop incoming skb rx to tx in normal user generated traffic,
3073          * most probably we will not get to this
3074          */
3075         if (skb_rx_queue_recorded(skb))
3076                 qid = skb_get_rx_queue(skb);
3077         else
3078                 qid = netdev_pick_tx(dev, skb, NULL);
3079
3080         return qid;
3081 }
3082
3083 static void ena_config_host_info(struct ena_com_dev *ena_dev,
3084                                  struct pci_dev *pdev)
3085 {
3086         struct ena_admin_host_info *host_info;
3087         int rc;
3088
3089         /* Allocate only the host info */
3090         rc = ena_com_allocate_host_info(ena_dev);
3091         if (rc) {
3092                 pr_err("Cannot allocate host info\n");
3093                 return;
3094         }
3095
3096         host_info = ena_dev->host_attr.host_info;
3097
3098         host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
3099         host_info->os_type = ENA_ADMIN_OS_LINUX;
3100         host_info->kernel_ver = LINUX_VERSION_CODE;
3101         strlcpy(host_info->kernel_ver_str, utsname()->version,
3102                 sizeof(host_info->kernel_ver_str) - 1);
3103         host_info->os_dist = 0;
3104         strncpy(host_info->os_dist_str, utsname()->release,
3105                 sizeof(host_info->os_dist_str) - 1);
3106         host_info->driver_version =
3107                 (DRV_MODULE_VER_MAJOR) |
3108                 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3109                 (DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3110                 ("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3111         host_info->num_cpus = num_online_cpus();
3112
3113         host_info->driver_supported_features =
3114                 ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK;
3115
3116         rc = ena_com_set_host_attributes(ena_dev);
3117         if (rc) {
3118                 if (rc == -EOPNOTSUPP)
3119                         pr_warn("Cannot set host attributes\n");
3120                 else
3121                         pr_err("Cannot set host attributes\n");
3122
3123                 goto err;
3124         }
3125
3126         return;
3127
3128 err:
3129         ena_com_delete_host_info(ena_dev);
3130 }
3131
3132 static void ena_config_debug_area(struct ena_adapter *adapter)
3133 {
3134         u32 debug_area_size;
3135         int rc, ss_count;
3136
3137         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3138         if (ss_count <= 0) {
3139                 netif_err(adapter, drv, adapter->netdev,
3140                           "SS count is negative\n");
3141                 return;
3142         }
3143
3144         /* allocate 32 bytes for each string and 64bit for the value */
3145         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3146
3147         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3148         if (rc) {
3149                 pr_err("Cannot allocate debug area\n");
3150                 return;
3151         }
3152
3153         rc = ena_com_set_host_attributes(adapter->ena_dev);
3154         if (rc) {
3155                 if (rc == -EOPNOTSUPP)
3156                         netif_warn(adapter, drv, adapter->netdev,
3157                                    "Cannot set host attributes\n");
3158                 else
3159                         netif_err(adapter, drv, adapter->netdev,
3160                                   "Cannot set host attributes\n");
3161                 goto err;
3162         }
3163
3164         return;
3165 err:
3166         ena_com_delete_debug_area(adapter->ena_dev);
3167 }
3168
3169 static void ena_get_stats64(struct net_device *netdev,
3170                             struct rtnl_link_stats64 *stats)
3171 {
3172         struct ena_adapter *adapter = netdev_priv(netdev);
3173         struct ena_ring *rx_ring, *tx_ring;
3174         unsigned int start;
3175         u64 rx_drops;
3176         int i;
3177
3178         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3179                 return;
3180
3181         for (i = 0; i < adapter->num_io_queues; i++) {
3182                 u64 bytes, packets;
3183
3184                 tx_ring = &adapter->tx_ring[i];
3185
3186                 do {
3187                         start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
3188                         packets = tx_ring->tx_stats.cnt;
3189                         bytes = tx_ring->tx_stats.bytes;
3190                 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
3191
3192                 stats->tx_packets += packets;
3193                 stats->tx_bytes += bytes;
3194
3195                 rx_ring = &adapter->rx_ring[i];
3196
3197                 do {
3198                         start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
3199                         packets = rx_ring->rx_stats.cnt;
3200                         bytes = rx_ring->rx_stats.bytes;
3201                 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
3202
3203                 stats->rx_packets += packets;
3204                 stats->rx_bytes += bytes;
3205         }
3206
3207         do {
3208                 start = u64_stats_fetch_begin_irq(&adapter->syncp);
3209                 rx_drops = adapter->dev_stats.rx_drops;
3210         } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
3211
3212         stats->rx_dropped = rx_drops;
3213
3214         stats->multicast = 0;
3215         stats->collisions = 0;
3216
3217         stats->rx_length_errors = 0;
3218         stats->rx_crc_errors = 0;
3219         stats->rx_frame_errors = 0;
3220         stats->rx_fifo_errors = 0;
3221         stats->rx_missed_errors = 0;
3222         stats->tx_window_errors = 0;
3223
3224         stats->rx_errors = 0;
3225         stats->tx_errors = 0;
3226 }
3227
3228 static const struct net_device_ops ena_netdev_ops = {
3229         .ndo_open               = ena_open,
3230         .ndo_stop               = ena_close,
3231         .ndo_start_xmit         = ena_start_xmit,
3232         .ndo_select_queue       = ena_select_queue,
3233         .ndo_get_stats64        = ena_get_stats64,
3234         .ndo_tx_timeout         = ena_tx_timeout,
3235         .ndo_change_mtu         = ena_change_mtu,
3236         .ndo_set_mac_address    = NULL,
3237         .ndo_validate_addr      = eth_validate_addr,
3238         .ndo_bpf                = ena_xdp,
3239 };
3240
3241 static int ena_device_validate_params(struct ena_adapter *adapter,
3242                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
3243 {
3244         struct net_device *netdev = adapter->netdev;
3245         int rc;
3246
3247         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3248                               adapter->mac_addr);
3249         if (!rc) {
3250                 netif_err(adapter, drv, netdev,
3251                           "Error, mac address are different\n");
3252                 return -EINVAL;
3253         }
3254
3255         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3256                 netif_err(adapter, drv, netdev,
3257                           "Error, device max mtu is smaller than netdev MTU\n");
3258                 return -EINVAL;
3259         }
3260
3261         return 0;
3262 }
3263
3264 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
3265                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
3266                            bool *wd_state)
3267 {
3268         struct device *dev = &pdev->dev;
3269         bool readless_supported;
3270         u32 aenq_groups;
3271         int dma_width;
3272         int rc;
3273
3274         rc = ena_com_mmio_reg_read_request_init(ena_dev);
3275         if (rc) {
3276                 dev_err(dev, "failed to init mmio read less\n");
3277                 return rc;
3278         }
3279
3280         /* The PCIe configuration space revision id indicate if mmio reg
3281          * read is disabled
3282          */
3283         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3284         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3285
3286         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3287         if (rc) {
3288                 dev_err(dev, "Can not reset device\n");
3289                 goto err_mmio_read_less;
3290         }
3291
3292         rc = ena_com_validate_version(ena_dev);
3293         if (rc) {
3294                 dev_err(dev, "device version is too low\n");
3295                 goto err_mmio_read_less;
3296         }
3297
3298         dma_width = ena_com_get_dma_width(ena_dev);
3299         if (dma_width < 0) {
3300                 dev_err(dev, "Invalid dma width value %d", dma_width);
3301                 rc = dma_width;
3302                 goto err_mmio_read_less;
3303         }
3304
3305         rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
3306         if (rc) {
3307                 dev_err(dev, "pci_set_dma_mask failed 0x%x\n", rc);
3308                 goto err_mmio_read_less;
3309         }
3310
3311         rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
3312         if (rc) {
3313                 dev_err(dev, "err_pci_set_consistent_dma_mask failed 0x%x\n",
3314                         rc);
3315                 goto err_mmio_read_less;
3316         }
3317
3318         /* ENA admin level init */
3319         rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3320         if (rc) {
3321                 dev_err(dev,
3322                         "Can not initialize ena admin queue with device\n");
3323                 goto err_mmio_read_less;
3324         }
3325
3326         /* To enable the msix interrupts the driver needs to know the number
3327          * of queues. So the driver uses polling mode to retrieve this
3328          * information
3329          */
3330         ena_com_set_admin_polling_mode(ena_dev, true);
3331
3332         ena_config_host_info(ena_dev, pdev);
3333
3334         /* Get Device Attributes*/
3335         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3336         if (rc) {
3337                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3338                 goto err_admin_init;
3339         }
3340
3341         /* Try to turn all the available aenq groups */
3342         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3343                 BIT(ENA_ADMIN_FATAL_ERROR) |
3344                 BIT(ENA_ADMIN_WARNING) |
3345                 BIT(ENA_ADMIN_NOTIFICATION) |
3346                 BIT(ENA_ADMIN_KEEP_ALIVE);
3347
3348         aenq_groups &= get_feat_ctx->aenq.supported_groups;
3349
3350         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3351         if (rc) {
3352                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3353                 goto err_admin_init;
3354         }
3355
3356         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3357
3358         return 0;
3359
3360 err_admin_init:
3361         ena_com_delete_host_info(ena_dev);
3362         ena_com_admin_destroy(ena_dev);
3363 err_mmio_read_less:
3364         ena_com_mmio_reg_read_request_destroy(ena_dev);
3365
3366         return rc;
3367 }
3368
3369 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3370 {
3371         struct ena_com_dev *ena_dev = adapter->ena_dev;
3372         struct device *dev = &adapter->pdev->dev;
3373         int rc;
3374
3375         rc = ena_enable_msix(adapter);
3376         if (rc) {
3377                 dev_err(dev, "Can not reserve msix vectors\n");
3378                 return rc;
3379         }
3380
3381         ena_setup_mgmnt_intr(adapter);
3382
3383         rc = ena_request_mgmnt_irq(adapter);
3384         if (rc) {
3385                 dev_err(dev, "Can not setup management interrupts\n");
3386                 goto err_disable_msix;
3387         }
3388
3389         ena_com_set_admin_polling_mode(ena_dev, false);
3390
3391         ena_com_admin_aenq_enable(ena_dev);
3392
3393         return 0;
3394
3395 err_disable_msix:
3396         ena_disable_msix(adapter);
3397
3398         return rc;
3399 }
3400
3401 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3402 {
3403         struct net_device *netdev = adapter->netdev;
3404         struct ena_com_dev *ena_dev = adapter->ena_dev;
3405         bool dev_up;
3406
3407         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3408                 return;
3409
3410         netif_carrier_off(netdev);
3411
3412         del_timer_sync(&adapter->timer_service);
3413
3414         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3415         adapter->dev_up_before_reset = dev_up;
3416         if (!graceful)
3417                 ena_com_set_admin_running_state(ena_dev, false);
3418
3419         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3420                 ena_down(adapter);
3421
3422         /* Stop the device from sending AENQ events (in case reset flag is set
3423          *  and device is up, ena_down() already reset the device.
3424          */
3425         if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3426                 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3427
3428         ena_free_mgmnt_irq(adapter);
3429
3430         ena_disable_msix(adapter);
3431
3432         ena_com_abort_admin_commands(ena_dev);
3433
3434         ena_com_wait_for_abort_completion(ena_dev);
3435
3436         ena_com_admin_destroy(ena_dev);
3437
3438         ena_com_mmio_reg_read_request_destroy(ena_dev);
3439
3440         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3441
3442         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3443         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3444 }
3445
3446 static int ena_restore_device(struct ena_adapter *adapter)
3447 {
3448         struct ena_com_dev_get_features_ctx get_feat_ctx;
3449         struct ena_com_dev *ena_dev = adapter->ena_dev;
3450         struct pci_dev *pdev = adapter->pdev;
3451         bool wd_state;
3452         int rc;
3453
3454         set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3455         rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
3456         if (rc) {
3457                 dev_err(&pdev->dev, "Can not initialize device\n");
3458                 goto err;
3459         }
3460         adapter->wd_state = wd_state;
3461
3462         rc = ena_device_validate_params(adapter, &get_feat_ctx);
3463         if (rc) {
3464                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
3465                 goto err_device_destroy;
3466         }
3467
3468         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3469         if (rc) {
3470                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
3471                 goto err_device_destroy;
3472         }
3473         /* If the interface was up before the reset bring it up */
3474         if (adapter->dev_up_before_reset) {
3475                 rc = ena_up(adapter);
3476                 if (rc) {
3477                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
3478                         goto err_disable_msix;
3479                 }
3480         }
3481
3482         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3483
3484         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3485         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3486                 netif_carrier_on(adapter->netdev);
3487
3488         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3489         adapter->last_keep_alive_jiffies = jiffies;
3490         dev_err(&pdev->dev,
3491                 "Device reset completed successfully, Driver info: %s\n",
3492                 version);
3493
3494         return rc;
3495 err_disable_msix:
3496         ena_free_mgmnt_irq(adapter);
3497         ena_disable_msix(adapter);
3498 err_device_destroy:
3499         ena_com_abort_admin_commands(ena_dev);
3500         ena_com_wait_for_abort_completion(ena_dev);
3501         ena_com_admin_destroy(ena_dev);
3502         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3503         ena_com_mmio_reg_read_request_destroy(ena_dev);
3504 err:
3505         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3506         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3507         dev_err(&pdev->dev,
3508                 "Reset attempt failed. Can not reset the device\n");
3509
3510         return rc;
3511 }
3512
3513 static void ena_fw_reset_device(struct work_struct *work)
3514 {
3515         struct ena_adapter *adapter =
3516                 container_of(work, struct ena_adapter, reset_task);
3517         struct pci_dev *pdev = adapter->pdev;
3518
3519         if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3520                 dev_err(&pdev->dev,
3521                         "device reset schedule while reset bit is off\n");
3522                 return;
3523         }
3524         rtnl_lock();
3525         ena_destroy_device(adapter, false);
3526         ena_restore_device(adapter);
3527         rtnl_unlock();
3528 }
3529
3530 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3531                                         struct ena_ring *rx_ring)
3532 {
3533         if (likely(rx_ring->first_interrupt))
3534                 return 0;
3535
3536         if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3537                 return 0;
3538
3539         rx_ring->no_interrupt_event_cnt++;
3540
3541         if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3542                 netif_err(adapter, rx_err, adapter->netdev,
3543                           "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3544                           rx_ring->qid);
3545                 adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3546                 smp_mb__before_atomic();
3547                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3548                 return -EIO;
3549         }
3550
3551         return 0;
3552 }
3553
3554 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3555                                           struct ena_ring *tx_ring)
3556 {
3557         struct ena_tx_buffer *tx_buf;
3558         unsigned long last_jiffies;
3559         u32 missed_tx = 0;
3560         int i, rc = 0;
3561
3562         for (i = 0; i < tx_ring->ring_size; i++) {
3563                 tx_buf = &tx_ring->tx_buffer_info[i];
3564                 last_jiffies = tx_buf->last_jiffies;
3565
3566                 if (last_jiffies == 0)
3567                         /* no pending Tx at this location */
3568                         continue;
3569
3570                 if (unlikely(!tx_ring->first_interrupt && time_is_before_jiffies(last_jiffies +
3571                              2 * adapter->missing_tx_completion_to))) {
3572                         /* If after graceful period interrupt is still not
3573                          * received, we schedule a reset
3574                          */
3575                         netif_err(adapter, tx_err, adapter->netdev,
3576                                   "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3577                                   tx_ring->qid);
3578                         adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3579                         smp_mb__before_atomic();
3580                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3581                         return -EIO;
3582                 }
3583
3584                 if (unlikely(time_is_before_jiffies(last_jiffies +
3585                                 adapter->missing_tx_completion_to))) {
3586                         if (!tx_buf->print_once)
3587                                 netif_notice(adapter, tx_err, adapter->netdev,
3588                                              "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
3589                                              tx_ring->qid, i);
3590
3591                         tx_buf->print_once = 1;
3592                         missed_tx++;
3593                 }
3594         }
3595
3596         if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3597                 netif_err(adapter, tx_err, adapter->netdev,
3598                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3599                           missed_tx,
3600                           adapter->missing_tx_completion_threshold);
3601                 adapter->reset_reason =
3602                         ENA_REGS_RESET_MISS_TX_CMPL;
3603                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3604                 rc = -EIO;
3605         }
3606
3607         u64_stats_update_begin(&tx_ring->syncp);
3608         tx_ring->tx_stats.missed_tx = missed_tx;
3609         u64_stats_update_end(&tx_ring->syncp);
3610
3611         return rc;
3612 }
3613
3614 static void check_for_missing_completions(struct ena_adapter *adapter)
3615 {
3616         struct ena_ring *tx_ring;
3617         struct ena_ring *rx_ring;
3618         int i, budget, rc;
3619         int io_queue_count;
3620
3621         io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
3622         /* Make sure the driver doesn't turn the device in other process */
3623         smp_rmb();
3624
3625         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3626                 return;
3627
3628         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3629                 return;
3630
3631         if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
3632                 return;
3633
3634         budget = ENA_MONITORED_TX_QUEUES;
3635
3636         for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
3637                 tx_ring = &adapter->tx_ring[i];
3638                 rx_ring = &adapter->rx_ring[i];
3639
3640                 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3641                 if (unlikely(rc))
3642                         return;
3643
3644                 rc =  !ENA_IS_XDP_INDEX(adapter, i) ?
3645                         check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
3646                 if (unlikely(rc))
3647                         return;
3648
3649                 budget--;
3650                 if (!budget)
3651                         break;
3652         }
3653
3654         adapter->last_monitored_tx_qid = i % io_queue_count;
3655 }
3656
3657 /* trigger napi schedule after 2 consecutive detections */
3658 #define EMPTY_RX_REFILL 2
3659 /* For the rare case where the device runs out of Rx descriptors and the
3660  * napi handler failed to refill new Rx descriptors (due to a lack of memory
3661  * for example).
3662  * This case will lead to a deadlock:
3663  * The device won't send interrupts since all the new Rx packets will be dropped
3664  * The napi handler won't allocate new Rx descriptors so the device will be
3665  * able to send new packets.
3666  *
3667  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
3668  * It is recommended to have at least 512MB, with a minimum of 128MB for
3669  * constrained environment).
3670  *
3671  * When such a situation is detected - Reschedule napi
3672  */
3673 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
3674 {
3675         struct ena_ring *rx_ring;
3676         int i, refill_required;
3677
3678         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3679                 return;
3680
3681         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3682                 return;
3683
3684         for (i = 0; i < adapter->num_io_queues; i++) {
3685                 rx_ring = &adapter->rx_ring[i];
3686
3687                 refill_required =
3688                         ena_com_free_desc(rx_ring->ena_com_io_sq);
3689                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3690                         rx_ring->empty_rx_queue++;
3691
3692                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3693                                 u64_stats_update_begin(&rx_ring->syncp);
3694                                 rx_ring->rx_stats.empty_rx_ring++;
3695                                 u64_stats_update_end(&rx_ring->syncp);
3696
3697                                 netif_err(adapter, drv, adapter->netdev,
3698                                           "trigger refill for ring %d\n", i);
3699
3700                                 napi_schedule(rx_ring->napi);
3701                                 rx_ring->empty_rx_queue = 0;
3702                         }
3703                 } else {
3704                         rx_ring->empty_rx_queue = 0;
3705                 }
3706         }
3707 }
3708
3709 /* Check for keep alive expiration */
3710 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3711 {
3712         unsigned long keep_alive_expired;
3713
3714         if (!adapter->wd_state)
3715                 return;
3716
3717         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3718                 return;
3719
3720         keep_alive_expired = adapter->last_keep_alive_jiffies +
3721                              adapter->keep_alive_timeout;
3722         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3723                 netif_err(adapter, drv, adapter->netdev,
3724                           "Keep alive watchdog timeout.\n");
3725                 u64_stats_update_begin(&adapter->syncp);
3726                 adapter->dev_stats.wd_expired++;
3727                 u64_stats_update_end(&adapter->syncp);
3728                 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3729                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3730         }
3731 }
3732
3733 static void check_for_admin_com_state(struct ena_adapter *adapter)
3734 {
3735         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3736                 netif_err(adapter, drv, adapter->netdev,
3737                           "ENA admin queue is not in running state!\n");
3738                 u64_stats_update_begin(&adapter->syncp);
3739                 adapter->dev_stats.admin_q_pause++;
3740                 u64_stats_update_end(&adapter->syncp);
3741                 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3742                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3743         }
3744 }
3745
3746 static void ena_update_hints(struct ena_adapter *adapter,
3747                              struct ena_admin_ena_hw_hints *hints)
3748 {
3749         struct net_device *netdev = adapter->netdev;
3750
3751         if (hints->admin_completion_tx_timeout)
3752                 adapter->ena_dev->admin_queue.completion_timeout =
3753                         hints->admin_completion_tx_timeout * 1000;
3754
3755         if (hints->mmio_read_timeout)
3756                 /* convert to usec */
3757                 adapter->ena_dev->mmio_read.reg_read_to =
3758                         hints->mmio_read_timeout * 1000;
3759
3760         if (hints->missed_tx_completion_count_threshold_to_reset)
3761                 adapter->missing_tx_completion_threshold =
3762                         hints->missed_tx_completion_count_threshold_to_reset;
3763
3764         if (hints->missing_tx_completion_timeout) {
3765                 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3766                         adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3767                 else
3768                         adapter->missing_tx_completion_to =
3769                                 msecs_to_jiffies(hints->missing_tx_completion_timeout);
3770         }
3771
3772         if (hints->netdev_wd_timeout)
3773                 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3774
3775         if (hints->driver_watchdog_timeout) {
3776                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3777                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3778                 else
3779                         adapter->keep_alive_timeout =
3780                                 msecs_to_jiffies(hints->driver_watchdog_timeout);
3781         }
3782 }
3783
3784 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3785                                  struct net_device *netdev)
3786 {
3787         host_info->supported_network_features[0] =
3788                 netdev->features & GENMASK_ULL(31, 0);
3789         host_info->supported_network_features[1] =
3790                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
3791 }
3792
3793 static void ena_timer_service(struct timer_list *t)
3794 {
3795         struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3796         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3797         struct ena_admin_host_info *host_info =
3798                 adapter->ena_dev->host_attr.host_info;
3799
3800         check_for_missing_keep_alive(adapter);
3801
3802         check_for_admin_com_state(adapter);
3803
3804         check_for_missing_completions(adapter);
3805
3806         check_for_empty_rx_ring(adapter);
3807
3808         if (debug_area)
3809                 ena_dump_stats_to_buf(adapter, debug_area);
3810
3811         if (host_info)
3812                 ena_update_host_info(host_info, adapter->netdev);
3813
3814         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3815                 netif_err(adapter, drv, adapter->netdev,
3816                           "Trigger reset is on\n");
3817                 ena_dump_stats_to_dmesg(adapter);
3818                 queue_work(ena_wq, &adapter->reset_task);
3819                 return;
3820         }
3821
3822         /* Reset the timer */
3823         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3824 }
3825
3826 static int ena_calc_max_io_queue_num(struct pci_dev *pdev,
3827                                      struct ena_com_dev *ena_dev,
3828                                      struct ena_com_dev_get_features_ctx *get_feat_ctx)
3829 {
3830         int io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
3831
3832         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3833                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3834                         &get_feat_ctx->max_queue_ext.max_queue_ext;
3835                 io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
3836                                   max_queue_ext->max_rx_cq_num);
3837
3838                 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
3839                 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
3840         } else {
3841                 struct ena_admin_queue_feature_desc *max_queues =
3842                         &get_feat_ctx->max_queues;
3843                 io_tx_sq_num = max_queues->max_sq_num;
3844                 io_tx_cq_num = max_queues->max_cq_num;
3845                 io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
3846         }
3847
3848         /* In case of LLQ use the llq fields for the tx SQ/CQ */
3849         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3850                 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
3851
3852         max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
3853         max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
3854         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
3855         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
3856         /* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
3857         max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
3858         if (unlikely(!max_num_io_queues)) {
3859                 dev_err(&pdev->dev, "The device doesn't have io queues\n");
3860                 return -EFAULT;
3861         }
3862
3863         return max_num_io_queues;
3864 }
3865
3866 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3867                                            struct ena_com_dev *ena_dev,
3868                                            struct ena_admin_feature_llq_desc *llq,
3869                                            struct ena_llq_configurations *llq_default_configurations)
3870 {
3871         bool has_mem_bar;
3872         int rc;
3873         u32 llq_feature_mask;
3874
3875         llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3876         if (!(ena_dev->supported_features & llq_feature_mask)) {
3877                 dev_err(&pdev->dev,
3878                         "LLQ is not supported Fallback to host mode policy.\n");
3879                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3880                 return 0;
3881         }
3882
3883         has_mem_bar = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(ENA_MEM_BAR);
3884
3885         rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3886         if (unlikely(rc)) {
3887                 dev_err(&pdev->dev,
3888                         "Failed to configure the device mode.  Fallback to host mode policy.\n");
3889                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3890                 return 0;
3891         }
3892
3893         /* Nothing to config, exit */
3894         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
3895                 return 0;
3896
3897         if (!has_mem_bar) {
3898                 dev_err(&pdev->dev,
3899                         "ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3900                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3901                 return 0;
3902         }
3903
3904         ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3905                                            pci_resource_start(pdev, ENA_MEM_BAR),
3906                                            pci_resource_len(pdev, ENA_MEM_BAR));
3907
3908         if (!ena_dev->mem_bar)
3909                 return -EFAULT;
3910
3911         return 0;
3912 }
3913
3914 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
3915                                  struct net_device *netdev)
3916 {
3917         netdev_features_t dev_features = 0;
3918
3919         /* Set offload features */
3920         if (feat->offload.tx &
3921                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
3922                 dev_features |= NETIF_F_IP_CSUM;
3923
3924         if (feat->offload.tx &
3925                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
3926                 dev_features |= NETIF_F_IPV6_CSUM;
3927
3928         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
3929                 dev_features |= NETIF_F_TSO;
3930
3931         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
3932                 dev_features |= NETIF_F_TSO6;
3933
3934         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
3935                 dev_features |= NETIF_F_TSO_ECN;
3936
3937         if (feat->offload.rx_supported &
3938                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
3939                 dev_features |= NETIF_F_RXCSUM;
3940
3941         if (feat->offload.rx_supported &
3942                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
3943                 dev_features |= NETIF_F_RXCSUM;
3944
3945         netdev->features =
3946                 dev_features |
3947                 NETIF_F_SG |
3948                 NETIF_F_RXHASH |
3949                 NETIF_F_HIGHDMA;
3950
3951         netdev->hw_features |= netdev->features;
3952         netdev->vlan_features |= netdev->features;
3953 }
3954
3955 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
3956                                      struct ena_com_dev_get_features_ctx *feat)
3957 {
3958         struct net_device *netdev = adapter->netdev;
3959
3960         /* Copy mac address */
3961         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
3962                 eth_hw_addr_random(netdev);
3963                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
3964         } else {
3965                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
3966                 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
3967         }
3968
3969         /* Set offload features */
3970         ena_set_dev_offloads(feat, netdev);
3971
3972         adapter->max_mtu = feat->dev_attr.max_mtu;
3973         netdev->max_mtu = adapter->max_mtu;
3974         netdev->min_mtu = ENA_MIN_MTU;
3975 }
3976
3977 static int ena_rss_init_default(struct ena_adapter *adapter)
3978 {
3979         struct ena_com_dev *ena_dev = adapter->ena_dev;
3980         struct device *dev = &adapter->pdev->dev;
3981         int rc, i;
3982         u32 val;
3983
3984         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3985         if (unlikely(rc)) {
3986                 dev_err(dev, "Cannot init indirect table\n");
3987                 goto err_rss_init;
3988         }
3989
3990         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3991                 val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
3992                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3993                                                        ENA_IO_RXQ_IDX(val));
3994                 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3995                         dev_err(dev, "Cannot fill indirect table\n");
3996                         goto err_fill_indir;
3997                 }
3998         }
3999
4000         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
4001                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4002         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4003                 dev_err(dev, "Cannot fill hash function\n");
4004                 goto err_fill_indir;
4005         }
4006
4007         rc = ena_com_set_default_hash_ctrl(ena_dev);
4008         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4009                 dev_err(dev, "Cannot fill hash control\n");
4010                 goto err_fill_indir;
4011         }
4012
4013         return 0;
4014
4015 err_fill_indir:
4016         ena_com_rss_destroy(ena_dev);
4017 err_rss_init:
4018
4019         return rc;
4020 }
4021
4022 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4023 {
4024         int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4025
4026         pci_release_selected_regions(pdev, release_bars);
4027 }
4028
4029 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
4030 {
4031         llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
4032         llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
4033         llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
4034         llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
4035         llq_config->llq_ring_entry_size_value = 128;
4036 }
4037
4038 static int ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
4039 {
4040         struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
4041         struct ena_com_dev *ena_dev = ctx->ena_dev;
4042         u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
4043         u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
4044         u32 max_tx_queue_size;
4045         u32 max_rx_queue_size;
4046
4047         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4048                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4049                         &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
4050                 max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
4051                                           max_queue_ext->max_rx_sq_depth);
4052                 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
4053
4054                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4055                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4056                                                   llq->max_llq_depth);
4057                 else
4058                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4059                                                   max_queue_ext->max_tx_sq_depth);
4060
4061                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4062                                              max_queue_ext->max_per_packet_tx_descs);
4063                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4064                                              max_queue_ext->max_per_packet_rx_descs);
4065         } else {
4066                 struct ena_admin_queue_feature_desc *max_queues =
4067                         &ctx->get_feat_ctx->max_queues;
4068                 max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
4069                                           max_queues->max_sq_depth);
4070                 max_tx_queue_size = max_queues->max_cq_depth;
4071
4072                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4073                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4074                                                   llq->max_llq_depth);
4075                 else
4076                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4077                                                   max_queues->max_sq_depth);
4078
4079                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4080                                              max_queues->max_packet_tx_descs);
4081                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4082                                              max_queues->max_packet_rx_descs);
4083         }
4084
4085         max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
4086         max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
4087
4088         tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
4089                                   max_tx_queue_size);
4090         rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
4091                                   max_rx_queue_size);
4092
4093         tx_queue_size = rounddown_pow_of_two(tx_queue_size);
4094         rx_queue_size = rounddown_pow_of_two(rx_queue_size);
4095
4096         ctx->max_tx_queue_size = max_tx_queue_size;
4097         ctx->max_rx_queue_size = max_rx_queue_size;
4098         ctx->tx_queue_size = tx_queue_size;
4099         ctx->rx_queue_size = rx_queue_size;
4100
4101         return 0;
4102 }
4103
4104 /* ena_probe - Device Initialization Routine
4105  * @pdev: PCI device information struct
4106  * @ent: entry in ena_pci_tbl
4107  *
4108  * Returns 0 on success, negative on failure
4109  *
4110  * ena_probe initializes an adapter identified by a pci_dev structure.
4111  * The OS initialization, configuring of the adapter private structure,
4112  * and a hardware reset occur.
4113  */
4114 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4115 {
4116         struct ena_com_dev_get_features_ctx get_feat_ctx;
4117         struct ena_calc_queue_size_ctx calc_queue_ctx = { 0 };
4118         struct ena_llq_configurations llq_config;
4119         struct ena_com_dev *ena_dev = NULL;
4120         struct ena_adapter *adapter;
4121         struct net_device *netdev;
4122         static int adapters_found;
4123         u32 max_num_io_queues;
4124         char *queue_type_str;
4125         bool wd_state;
4126         int bars, rc;
4127
4128         dev_dbg(&pdev->dev, "%s\n", __func__);
4129
4130         dev_info_once(&pdev->dev, "%s", version);
4131
4132         rc = pci_enable_device_mem(pdev);
4133         if (rc) {
4134                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4135                 return rc;
4136         }
4137
4138         pci_set_master(pdev);
4139
4140         ena_dev = vzalloc(sizeof(*ena_dev));
4141         if (!ena_dev) {
4142                 rc = -ENOMEM;
4143                 goto err_disable_device;
4144         }
4145
4146         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4147         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4148         if (rc) {
4149                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4150                         rc);
4151                 goto err_free_ena_dev;
4152         }
4153
4154         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4155                                         pci_resource_start(pdev, ENA_REG_BAR),
4156                                         pci_resource_len(pdev, ENA_REG_BAR));
4157         if (!ena_dev->reg_bar) {
4158                 dev_err(&pdev->dev, "failed to remap regs bar\n");
4159                 rc = -EFAULT;
4160                 goto err_free_region;
4161         }
4162
4163         ena_dev->dmadev = &pdev->dev;
4164
4165         rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
4166         if (rc) {
4167                 dev_err(&pdev->dev, "ena device init failed\n");
4168                 if (rc == -ETIME)
4169                         rc = -EPROBE_DEFER;
4170                 goto err_free_region;
4171         }
4172
4173         set_default_llq_configurations(&llq_config);
4174
4175         rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx.llq,
4176                                              &llq_config);
4177         if (rc) {
4178                 dev_err(&pdev->dev, "ena device init failed\n");
4179                 goto err_device_destroy;
4180         }
4181
4182         calc_queue_ctx.ena_dev = ena_dev;
4183         calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
4184         calc_queue_ctx.pdev = pdev;
4185
4186         /* Initial Tx and RX interrupt delay. Assumes 1 usec granularity.
4187          * Updated during device initialization with the real granularity
4188          */
4189         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4190         ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4191         ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4192         max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4193         rc = ena_calc_io_queue_size(&calc_queue_ctx);
4194         if (rc || !max_num_io_queues) {
4195                 rc = -EFAULT;
4196                 goto err_device_destroy;
4197         }
4198
4199         /* dev zeroed in init_etherdev */
4200         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), max_num_io_queues);
4201         if (!netdev) {
4202                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4203                 rc = -ENOMEM;
4204                 goto err_device_destroy;
4205         }
4206
4207         SET_NETDEV_DEV(netdev, &pdev->dev);
4208
4209         adapter = netdev_priv(netdev);
4210         pci_set_drvdata(pdev, adapter);
4211
4212         adapter->ena_dev = ena_dev;
4213         adapter->netdev = netdev;
4214         adapter->pdev = pdev;
4215
4216         ena_set_conf_feat_params(adapter, &get_feat_ctx);
4217
4218         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4219         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4220
4221         adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
4222         adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
4223         adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
4224         adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
4225         adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
4226         adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
4227
4228         adapter->num_io_queues = max_num_io_queues;
4229         adapter->max_num_io_queues = max_num_io_queues;
4230
4231         adapter->xdp_first_ring = 0;
4232         adapter->xdp_num_queues = 0;
4233
4234         adapter->last_monitored_tx_qid = 0;
4235
4236         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4237         adapter->wd_state = wd_state;
4238
4239         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4240
4241         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4242         if (rc) {
4243                 dev_err(&pdev->dev,
4244                         "Failed to query interrupt moderation feature\n");
4245                 goto err_netdev_destroy;
4246         }
4247         ena_init_io_rings(adapter,
4248                           0,
4249                           adapter->xdp_num_queues +
4250                           adapter->num_io_queues);
4251
4252         netdev->netdev_ops = &ena_netdev_ops;
4253         netdev->watchdog_timeo = TX_TIMEOUT;
4254         ena_set_ethtool_ops(netdev);
4255
4256         netdev->priv_flags |= IFF_UNICAST_FLT;
4257
4258         u64_stats_init(&adapter->syncp);
4259
4260         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4261         if (rc) {
4262                 dev_err(&pdev->dev,
4263                         "Failed to enable and set the admin interrupts\n");
4264                 goto err_worker_destroy;
4265         }
4266         rc = ena_rss_init_default(adapter);
4267         if (rc && (rc != -EOPNOTSUPP)) {
4268                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4269                 goto err_free_msix;
4270         }
4271
4272         ena_config_debug_area(adapter);
4273
4274         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4275
4276         netif_carrier_off(netdev);
4277
4278         rc = register_netdev(netdev);
4279         if (rc) {
4280                 dev_err(&pdev->dev, "Cannot register net device\n");
4281                 goto err_rss;
4282         }
4283
4284         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4285
4286         adapter->last_keep_alive_jiffies = jiffies;
4287         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4288         adapter->missing_tx_completion_to = TX_TIMEOUT;
4289         adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4290
4291         ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4292
4293         timer_setup(&adapter->timer_service, ena_timer_service, 0);
4294         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4295
4296         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_HOST)
4297                 queue_type_str = "Regular";
4298         else
4299                 queue_type_str = "Low Latency";
4300
4301         dev_info(&pdev->dev,
4302                  "%s found at mem %lx, mac addr %pM, Placement policy: %s\n",
4303                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4304                  netdev->dev_addr, queue_type_str);
4305
4306         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4307
4308         adapters_found++;
4309
4310         return 0;
4311
4312 err_rss:
4313         ena_com_delete_debug_area(ena_dev);
4314         ena_com_rss_destroy(ena_dev);
4315 err_free_msix:
4316         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4317         /* stop submitting admin commands on a device that was reset */
4318         ena_com_set_admin_running_state(ena_dev, false);
4319         ena_free_mgmnt_irq(adapter);
4320         ena_disable_msix(adapter);
4321 err_worker_destroy:
4322         del_timer(&adapter->timer_service);
4323 err_netdev_destroy:
4324         free_netdev(netdev);
4325 err_device_destroy:
4326         ena_com_delete_host_info(ena_dev);
4327         ena_com_admin_destroy(ena_dev);
4328 err_free_region:
4329         ena_release_bars(ena_dev, pdev);
4330 err_free_ena_dev:
4331         vfree(ena_dev);
4332 err_disable_device:
4333         pci_disable_device(pdev);
4334         return rc;
4335 }
4336
4337 /*****************************************************************************/
4338
4339 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4340  * @pdev: PCI device information struct
4341  * @shutdown: Is it a shutdown operation? If false, means it is a removal
4342  *
4343  * __ena_shutoff is a helper routine that does the real work on shutdown and
4344  * removal paths; the difference between those paths is with regards to whether
4345  * dettach or unregister the netdevice.
4346  */
4347 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4348 {
4349         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4350         struct ena_com_dev *ena_dev;
4351         struct net_device *netdev;
4352
4353         ena_dev = adapter->ena_dev;
4354         netdev = adapter->netdev;
4355
4356 #ifdef CONFIG_RFS_ACCEL
4357         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4358                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4359                 netdev->rx_cpu_rmap = NULL;
4360         }
4361 #endif /* CONFIG_RFS_ACCEL */
4362         del_timer_sync(&adapter->timer_service);
4363
4364         cancel_work_sync(&adapter->reset_task);
4365
4366         rtnl_lock(); /* lock released inside the below if-else block */
4367         ena_destroy_device(adapter, true);
4368         if (shutdown) {
4369                 netif_device_detach(netdev);
4370                 dev_close(netdev);
4371                 rtnl_unlock();
4372         } else {
4373                 rtnl_unlock();
4374                 unregister_netdev(netdev);
4375                 free_netdev(netdev);
4376         }
4377
4378         ena_com_rss_destroy(ena_dev);
4379
4380         ena_com_delete_debug_area(ena_dev);
4381
4382         ena_com_delete_host_info(ena_dev);
4383
4384         ena_release_bars(ena_dev, pdev);
4385
4386         pci_disable_device(pdev);
4387
4388         vfree(ena_dev);
4389 }
4390
4391 /* ena_remove - Device Removal Routine
4392  * @pdev: PCI device information struct
4393  *
4394  * ena_remove is called by the PCI subsystem to alert the driver
4395  * that it should release a PCI device.
4396  */
4397
4398 static void ena_remove(struct pci_dev *pdev)
4399 {
4400         __ena_shutoff(pdev, false);
4401 }
4402
4403 /* ena_shutdown - Device Shutdown Routine
4404  * @pdev: PCI device information struct
4405  *
4406  * ena_shutdown is called by the PCI subsystem to alert the driver that
4407  * a shutdown/reboot (or kexec) is happening and device must be disabled.
4408  */
4409
4410 static void ena_shutdown(struct pci_dev *pdev)
4411 {
4412         __ena_shutoff(pdev, true);
4413 }
4414
4415 #ifdef CONFIG_PM
4416 /* ena_suspend - PM suspend callback
4417  * @pdev: PCI device information struct
4418  * @state:power state
4419  */
4420 static int ena_suspend(struct pci_dev *pdev,  pm_message_t state)
4421 {
4422         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4423
4424         u64_stats_update_begin(&adapter->syncp);
4425         adapter->dev_stats.suspend++;
4426         u64_stats_update_end(&adapter->syncp);
4427
4428         rtnl_lock();
4429         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4430                 dev_err(&pdev->dev,
4431                         "ignoring device reset request as the device is being suspended\n");
4432                 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4433         }
4434         ena_destroy_device(adapter, true);
4435         rtnl_unlock();
4436         return 0;
4437 }
4438
4439 /* ena_resume - PM resume callback
4440  * @pdev: PCI device information struct
4441  *
4442  */
4443 static int ena_resume(struct pci_dev *pdev)
4444 {
4445         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4446         int rc;
4447
4448         u64_stats_update_begin(&adapter->syncp);
4449         adapter->dev_stats.resume++;
4450         u64_stats_update_end(&adapter->syncp);
4451
4452         rtnl_lock();
4453         rc = ena_restore_device(adapter);
4454         rtnl_unlock();
4455         return rc;
4456 }
4457 #endif
4458
4459 static struct pci_driver ena_pci_driver = {
4460         .name           = DRV_MODULE_NAME,
4461         .id_table       = ena_pci_tbl,
4462         .probe          = ena_probe,
4463         .remove         = ena_remove,
4464         .shutdown       = ena_shutdown,
4465 #ifdef CONFIG_PM
4466         .suspend    = ena_suspend,
4467         .resume     = ena_resume,
4468 #endif
4469         .sriov_configure = pci_sriov_configure_simple,
4470 };
4471
4472 static int __init ena_init(void)
4473 {
4474         pr_info("%s", version);
4475
4476         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4477         if (!ena_wq) {
4478                 pr_err("Failed to create workqueue\n");
4479                 return -ENOMEM;
4480         }
4481
4482         return pci_register_driver(&ena_pci_driver);
4483 }
4484
4485 static void __exit ena_cleanup(void)
4486 {
4487         pci_unregister_driver(&ena_pci_driver);
4488
4489         if (ena_wq) {
4490                 destroy_workqueue(ena_wq);
4491                 ena_wq = NULL;
4492         }
4493 }
4494
4495 /******************************************************************************
4496  ******************************** AENQ Handlers *******************************
4497  *****************************************************************************/
4498 /* ena_update_on_link_change:
4499  * Notify the network interface about the change in link status
4500  */
4501 static void ena_update_on_link_change(void *adapter_data,
4502                                       struct ena_admin_aenq_entry *aenq_e)
4503 {
4504         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4505         struct ena_admin_aenq_link_change_desc *aenq_desc =
4506                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
4507         int status = aenq_desc->flags &
4508                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4509
4510         if (status) {
4511                 netdev_dbg(adapter->netdev, "%s\n", __func__);
4512                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4513                 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4514                         netif_carrier_on(adapter->netdev);
4515         } else {
4516                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4517                 netif_carrier_off(adapter->netdev);
4518         }
4519 }
4520
4521 static void ena_keep_alive_wd(void *adapter_data,
4522                               struct ena_admin_aenq_entry *aenq_e)
4523 {
4524         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4525         struct ena_admin_aenq_keep_alive_desc *desc;
4526         u64 rx_drops;
4527
4528         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4529         adapter->last_keep_alive_jiffies = jiffies;
4530
4531         rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4532
4533         u64_stats_update_begin(&adapter->syncp);
4534         adapter->dev_stats.rx_drops = rx_drops;
4535         u64_stats_update_end(&adapter->syncp);
4536 }
4537
4538 static void ena_notification(void *adapter_data,
4539                              struct ena_admin_aenq_entry *aenq_e)
4540 {
4541         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4542         struct ena_admin_ena_hw_hints *hints;
4543
4544         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4545              "Invalid group(%x) expected %x\n",
4546              aenq_e->aenq_common_desc.group,
4547              ENA_ADMIN_NOTIFICATION);
4548
4549         switch (aenq_e->aenq_common_desc.syndrom) {
4550         case ENA_ADMIN_UPDATE_HINTS:
4551                 hints = (struct ena_admin_ena_hw_hints *)
4552                         (&aenq_e->inline_data_w4);
4553                 ena_update_hints(adapter, hints);
4554                 break;
4555         default:
4556                 netif_err(adapter, drv, adapter->netdev,
4557                           "Invalid aenq notification link state %d\n",
4558                           aenq_e->aenq_common_desc.syndrom);
4559         }
4560 }
4561
4562 /* This handler will called for unknown event group or unimplemented handlers*/
4563 static void unimplemented_aenq_handler(void *data,
4564                                        struct ena_admin_aenq_entry *aenq_e)
4565 {
4566         struct ena_adapter *adapter = (struct ena_adapter *)data;
4567
4568         netif_err(adapter, drv, adapter->netdev,
4569                   "Unknown event was received or event with unimplemented handler\n");
4570 }
4571
4572 static struct ena_aenq_handlers aenq_handlers = {
4573         .handlers = {
4574                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4575                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
4576                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4577         },
4578         .unimplemented_handler = unimplemented_aenq_handler
4579 };
4580
4581 module_init(ena_init);
4582 module_exit(ena_cleanup);