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/if_vlan.h>
  40 #include <linux/kernel.h>
  41 #include <linux/module.h>
  42 #include <linux/moduleparam.h>
  43 #include <linux/numa.h>
  44 #include <linux/pci.h>
  45 #include <linux/utsname.h>
  46 #include <linux/version.h>
  47 #include <linux/vmalloc.h>
  48 #include <net/ip.h>
  49
  50 #include "ena_netdev.h"
  51 #include "ena_pci_id_tbl.h"
  52
  53 static char version[] = DEVICE_NAME " v" DRV_MODULE_VERSION "\n";
  54
  55 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
  56 MODULE_DESCRIPTION(DEVICE_NAME);
  57 MODULE_LICENSE("GPL");
  58 MODULE_VERSION(DRV_MODULE_VERSION);
  59
  60 /* Time in jiffies before concluding the transmitter is hung. */
  61 #define TX_TIMEOUT  (5 * HZ)
  62
  63 #define ENA_NAPI_BUDGET 64
  64
  65 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
  66                 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
  67 static int debug = -1;
  68 module_param(debug, int, 0);
  69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  70
  71 static struct ena_aenq_handlers aenq_handlers;
  72
  73 static struct workqueue_struct *ena_wq;
  74
  75 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
  76
  77 static int ena_rss_init_default(struct ena_adapter *adapter);
  78 static void check_for_admin_com_state(struct ena_adapter *adapter);
  79 static void ena_destroy_device(struct ena_adapter *adapter);
  80 static int ena_restore_device(struct ena_adapter *adapter);
  81
  82 static void ena_tx_timeout(struct net_device *dev)
  83 {
  84         struct ena_adapter *adapter = netdev_priv(dev);
  85
  86         /* Change the state of the device to trigger reset
  87          * Check that we are not in the middle or a trigger already
  88          */
  89
  90         if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
  91                 return;
  92
  93         adapter->reset_reason = ENA_REGS_RESET_OS_NETDEV_WD;
  94         u64_stats_update_begin(&adapter->syncp);
  95         adapter->dev_stats.tx_timeout++;
  96         u64_stats_update_end(&adapter->syncp);
  97
  98         netif_err(adapter, tx_err, dev, "Transmit time out\n");
  99 }
 100
 101 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
 102 {
 103         int i;
 104
 105         for (i = 0; i < adapter->num_queues; i++)
 106                 adapter->rx_ring[i].mtu = mtu;
 107 }
 108
 109 static int ena_change_mtu(struct net_device *dev, int new_mtu)
 110 {
 111         struct ena_adapter *adapter = netdev_priv(dev);
 112         int ret;
 113
 114         ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
 115         if (!ret) {
 116                 netif_dbg(adapter, drv, dev, "set MTU to %d\n", new_mtu);
 117                 update_rx_ring_mtu(adapter, new_mtu);
 118                 dev->mtu = new_mtu;
 119         } else {
 120                 netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
 121                           new_mtu);
 122         }
 123
 124         return ret;
 125 }
 126
 127 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
 128 {
 129 #ifdef CONFIG_RFS_ACCEL
 130         u32 i;
 131         int rc;
 132
 133         adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_queues);
 134         if (!adapter->netdev->rx_cpu_rmap)
 135                 return -ENOMEM;
 136         for (i = 0; i < adapter->num_queues; i++) {
 137                 int irq_idx = ENA_IO_IRQ_IDX(i);
 138
 139                 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
 140                                       pci_irq_vector(adapter->pdev, irq_idx));
 141                 if (rc) {
 142                         free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
 143                         adapter->netdev->rx_cpu_rmap = NULL;
 144                         return rc;
 145                 }
 146         }
 147 #endif /* CONFIG_RFS_ACCEL */
 148         return 0;
 149 }
 150
 151 static void ena_init_io_rings_common(struct ena_adapter *adapter,
 152                                      struct ena_ring *ring, u16 qid)
 153 {
 154         ring->qid = qid;
 155         ring->pdev = adapter->pdev;
 156         ring->dev = &adapter->pdev->dev;
 157         ring->netdev = adapter->netdev;
 158         ring->napi = &adapter->ena_napi[qid].napi;
 159         ring->adapter = adapter;
 160         ring->ena_dev = adapter->ena_dev;
 161         ring->per_napi_packets = 0;
 162         ring->per_napi_bytes = 0;
 163         ring->cpu = 0;
 164         ring->first_interrupt = false;
 165         ring->no_interrupt_event_cnt = 0;
 166         u64_stats_init(&ring->syncp);
 167 }
 168
 169 static void ena_init_io_rings(struct ena_adapter *adapter)
 170 {
 171         struct ena_com_dev *ena_dev;
 172         struct ena_ring *txr, *rxr;
 173         int i;
 174
 175         ena_dev = adapter->ena_dev;
 176
 177         for (i = 0; i < adapter->num_queues; i++) {
 178                 txr = &adapter->tx_ring[i];
 179                 rxr = &adapter->rx_ring[i];
 180
 181                 /* TX/RX common ring state */
 182                 ena_init_io_rings_common(adapter, txr, i);
 183                 ena_init_io_rings_common(adapter, rxr, i);
 184
 185                 /* TX specific ring state */
 186                 txr->ring_size = adapter->tx_ring_size;
 187                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
 188                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
 189                 txr->sgl_size = adapter->max_tx_sgl_size;
 190                 txr->smoothed_interval =
 191                         ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
 192
 193                 /* RX specific ring state */
 194                 rxr->ring_size = adapter->rx_ring_size;
 195                 rxr->rx_copybreak = adapter->rx_copybreak;
 196                 rxr->sgl_size = adapter->max_rx_sgl_size;
 197                 rxr->smoothed_interval =
 198                         ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
 199                 rxr->empty_rx_queue = 0;
 200         }
 201 }
 202
 203 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
 204  * @adapter: network interface device structure
 205  * @qid: queue index
 206  *
 207  * Return 0 on success, negative on failure
 208  */
 209 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
 210 {
 211         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 212         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 213         int size, i, node;
 214
 215         if (tx_ring->tx_buffer_info) {
 216                 netif_err(adapter, ifup,
 217                           adapter->netdev, "tx_buffer_info info is not NULL");
 218                 return -EEXIST;
 219         }
 220
 221         size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
 222         node = cpu_to_node(ena_irq->cpu);
 223
 224         tx_ring->tx_buffer_info = vzalloc_node(size, node);
 225         if (!tx_ring->tx_buffer_info) {
 226                 tx_ring->tx_buffer_info = vzalloc(size);
 227                 if (!tx_ring->tx_buffer_info)
 228                         return -ENOMEM;
 229         }
 230
 231         size = sizeof(u16) * tx_ring->ring_size;
 232         tx_ring->free_tx_ids = vzalloc_node(size, node);
 233         if (!tx_ring->free_tx_ids) {
 234                 tx_ring->free_tx_ids = vzalloc(size);
 235                 if (!tx_ring->free_tx_ids) {
 236                         vfree(tx_ring->tx_buffer_info);
 237                         return -ENOMEM;
 238                 }
 239         }
 240
 241         /* Req id ring for TX out of order completions */
 242         for (i = 0; i < tx_ring->ring_size; i++)
 243                 tx_ring->free_tx_ids[i] = i;
 244
 245         /* Reset tx statistics */
 246         memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
 247
 248         tx_ring->next_to_use = 0;
 249         tx_ring->next_to_clean = 0;
 250         tx_ring->cpu = ena_irq->cpu;
 251         return 0;
 252 }
 253
 254 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
 255  * @adapter: network interface device structure
 256  * @qid: queue index
 257  *
 258  * Free all transmit software resources
 259  */
 260 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
 261 {
 262         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 263
 264         vfree(tx_ring->tx_buffer_info);
 265         tx_ring->tx_buffer_info = NULL;
 266
 267         vfree(tx_ring->free_tx_ids);
 268         tx_ring->free_tx_ids = NULL;
 269 }
 270
 271 /* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
 272  * @adapter: private structure
 273  *
 274  * Return 0 on success, negative on failure
 275  */
 276 static int ena_setup_all_tx_resources(struct ena_adapter *adapter)
 277 {
 278         int i, rc = 0;
 279
 280         for (i = 0; i < adapter->num_queues; i++) {
 281                 rc = ena_setup_tx_resources(adapter, i);
 282                 if (rc)
 283                         goto err_setup_tx;
 284         }
 285
 286         return 0;
 287
 288 err_setup_tx:
 289
 290         netif_err(adapter, ifup, adapter->netdev,
 291                   "Tx queue %d: allocation failed\n", i);
 292
 293         /* rewind the index freeing the rings as we go */
 294         while (i--)
 295                 ena_free_tx_resources(adapter, i);
 296         return rc;
 297 }
 298
 299 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
 300  * @adapter: board private structure
 301  *
 302  * Free all transmit software resources
 303  */
 304 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
 305 {
 306         int i;
 307
 308         for (i = 0; i < adapter->num_queues; i++)
 309                 ena_free_tx_resources(adapter, i);
 310 }
 311
 312 static inline int validate_rx_req_id(struct ena_ring *rx_ring, u16 req_id)
 313 {
 314         if (likely(req_id < rx_ring->ring_size))
 315                 return 0;
 316
 317         netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 318                   "Invalid rx req_id: %hu\n", req_id);
 319
 320         u64_stats_update_begin(&rx_ring->syncp);
 321         rx_ring->rx_stats.bad_req_id++;
 322         u64_stats_update_end(&rx_ring->syncp);
 323
 324         /* Trigger device reset */
 325         rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
 326         set_bit(ENA_FLAG_TRIGGER_RESET, &rx_ring->adapter->flags);
 327         return -EFAULT;
 328 }
 329
 330 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
 331  * @adapter: network interface device structure
 332  * @qid: queue index
 333  *
 334  * Returns 0 on success, negative on failure
 335  */
 336 static int ena_setup_rx_resources(struct ena_adapter *adapter,
 337                                   u32 qid)
 338 {
 339         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 340         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 341         int size, node, i;
 342
 343         if (rx_ring->rx_buffer_info) {
 344                 netif_err(adapter, ifup, adapter->netdev,
 345                           "rx_buffer_info is not NULL");
 346                 return -EEXIST;
 347         }
 348
 349         /* alloc extra element so in rx path
 350          * we can always prefetch rx_info + 1
 351          */
 352         size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
 353         node = cpu_to_node(ena_irq->cpu);
 354
 355         rx_ring->rx_buffer_info = vzalloc_node(size, node);
 356         if (!rx_ring->rx_buffer_info) {
 357                 rx_ring->rx_buffer_info = vzalloc(size);
 358                 if (!rx_ring->rx_buffer_info)
 359                         return -ENOMEM;
 360         }
 361
 362         size = sizeof(u16) * rx_ring->ring_size;
 363         rx_ring->free_rx_ids = vzalloc_node(size, node);
 364         if (!rx_ring->free_rx_ids) {
 365                 rx_ring->free_rx_ids = vzalloc(size);
 366                 if (!rx_ring->free_rx_ids) {
 367                         vfree(rx_ring->rx_buffer_info);
 368                         return -ENOMEM;
 369                 }
 370         }
 371
 372         /* Req id ring for receiving RX pkts out of order */
 373         for (i = 0; i < rx_ring->ring_size; i++)
 374                 rx_ring->free_rx_ids[i] = i;
 375
 376         /* Reset rx statistics */
 377         memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
 378
 379         rx_ring->next_to_clean = 0;
 380         rx_ring->next_to_use = 0;
 381         rx_ring->cpu = ena_irq->cpu;
 382
 383         return 0;
 384 }
 385
 386 /* ena_free_rx_resources - Free I/O Rx Resources
 387  * @adapter: network interface device structure
 388  * @qid: queue index
 389  *
 390  * Free all receive software resources
 391  */
 392 static void ena_free_rx_resources(struct ena_adapter *adapter,
 393                                   u32 qid)
 394 {
 395         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 396
 397         vfree(rx_ring->rx_buffer_info);
 398         rx_ring->rx_buffer_info = NULL;
 399
 400         vfree(rx_ring->free_rx_ids);
 401         rx_ring->free_rx_ids = NULL;
 402 }
 403
 404 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
 405  * @adapter: board private structure
 406  *
 407  * Return 0 on success, negative on failure
 408  */
 409 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
 410 {
 411         int i, rc = 0;
 412
 413         for (i = 0; i < adapter->num_queues; i++) {
 414                 rc = ena_setup_rx_resources(adapter, i);
 415                 if (rc)
 416                         goto err_setup_rx;
 417         }
 418
 419         return 0;
 420
 421 err_setup_rx:
 422
 423         netif_err(adapter, ifup, adapter->netdev,
 424                   "Rx queue %d: allocation failed\n", i);
 425
 426         /* rewind the index freeing the rings as we go */
 427         while (i--)
 428                 ena_free_rx_resources(adapter, i);
 429         return rc;
 430 }
 431
 432 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
 433  * @adapter: board private structure
 434  *
 435  * Free all receive software resources
 436  */
 437 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
 438 {
 439         int i;
 440
 441         for (i = 0; i < adapter->num_queues; i++)
 442                 ena_free_rx_resources(adapter, i);
 443 }
 444
 445 static inline int ena_alloc_rx_page(struct ena_ring *rx_ring,
 446                                     struct ena_rx_buffer *rx_info, gfp_t gfp)
 447 {
 448         struct ena_com_buf *ena_buf;
 449         struct page *page;
 450         dma_addr_t dma;
 451
 452         /* if previous allocated page is not used */
 453         if (unlikely(rx_info->page))
 454                 return 0;
 455
 456         page = alloc_page(gfp);
 457         if (unlikely(!page)) {
 458                 u64_stats_update_begin(&rx_ring->syncp);
 459                 rx_ring->rx_stats.page_alloc_fail++;
 460                 u64_stats_update_end(&rx_ring->syncp);
 461                 return -ENOMEM;
 462         }
 463
 464         dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE,
 465                            DMA_FROM_DEVICE);
 466         if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
 467                 u64_stats_update_begin(&rx_ring->syncp);
 468                 rx_ring->rx_stats.dma_mapping_err++;
 469                 u64_stats_update_end(&rx_ring->syncp);
 470
 471                 __free_page(page);
 472                 return -EIO;
 473         }
 474         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 475                   "alloc page %p, rx_info %p\n", page, rx_info);
 476
 477         rx_info->page = page;
 478         rx_info->page_offset = 0;
 479         ena_buf = &rx_info->ena_buf;
 480         ena_buf->paddr = dma;
 481         ena_buf->len = PAGE_SIZE;
 482
 483         return 0;
 484 }
 485
 486 static void ena_free_rx_page(struct ena_ring *rx_ring,
 487                              struct ena_rx_buffer *rx_info)
 488 {
 489         struct page *page = rx_info->page;
 490         struct ena_com_buf *ena_buf = &rx_info->ena_buf;
 491
 492         if (unlikely(!page)) {
 493                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
 494                            "Trying to free unallocated buffer\n");
 495                 return;
 496         }
 497
 498         dma_unmap_page(rx_ring->dev, ena_buf->paddr, PAGE_SIZE,
 499                        DMA_FROM_DEVICE);
 500
 501         __free_page(page);
 502         rx_info->page = NULL;
 503 }
 504
 505 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
 506 {
 507         u16 next_to_use, req_id;
 508         u32 i;
 509         int rc;
 510
 511         next_to_use = rx_ring->next_to_use;
 512
 513         for (i = 0; i < num; i++) {
 514                 struct ena_rx_buffer *rx_info;
 515
 516                 req_id = rx_ring->free_rx_ids[next_to_use];
 517                 rc = validate_rx_req_id(rx_ring, req_id);
 518                 if (unlikely(rc < 0))
 519                         break;
 520
 521                 rx_info = &rx_ring->rx_buffer_info[req_id];
 522
 523
 524                 rc = ena_alloc_rx_page(rx_ring, rx_info,
 525                                        GFP_ATOMIC | __GFP_COMP);
 526                 if (unlikely(rc < 0)) {
 527                         netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
 528                                    "failed to alloc buffer for rx queue %d\n",
 529                                    rx_ring->qid);
 530                         break;
 531                 }
 532                 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
 533                                                 &rx_info->ena_buf,
 534                                                 req_id);
 535                 if (unlikely(rc)) {
 536                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
 537                                    "failed to add buffer for rx queue %d\n",
 538                                    rx_ring->qid);
 539                         break;
 540                 }
 541                 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
 542                                                    rx_ring->ring_size);
 543         }
 544
 545         if (unlikely(i < num)) {
 546                 u64_stats_update_begin(&rx_ring->syncp);
 547                 rx_ring->rx_stats.refil_partial++;
 548                 u64_stats_update_end(&rx_ring->syncp);
 549                 netdev_warn(rx_ring->netdev,
 550                             "refilled rx qid %d with only %d buffers (from %d)\n",
 551                             rx_ring->qid, i, num);
 552         }
 553
 554         if (likely(i)) {
 555                 /* Add memory barrier to make sure the desc were written before
 556                  * issue a doorbell
 557                  */
 558                 wmb();
 559                 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq, true);
 560                 mmiowb();
 561         }
 562
 563         rx_ring->next_to_use = next_to_use;
 564
 565         return i;
 566 }
 567
 568 static void ena_free_rx_bufs(struct ena_adapter *adapter,
 569                              u32 qid)
 570 {
 571         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 572         u32 i;
 573
 574         for (i = 0; i < rx_ring->ring_size; i++) {
 575                 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
 576
 577                 if (rx_info->page)
 578                         ena_free_rx_page(rx_ring, rx_info);
 579         }
 580 }
 581
 582 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
 583  * @adapter: board private structure
 584  *
 585  */
 586 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
 587 {
 588         struct ena_ring *rx_ring;
 589         int i, rc, bufs_num;
 590
 591         for (i = 0; i < adapter->num_queues; i++) {
 592                 rx_ring = &adapter->rx_ring[i];
 593                 bufs_num = rx_ring->ring_size - 1;
 594                 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
 595
 596                 if (unlikely(rc != bufs_num))
 597                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
 598                                    "refilling Queue %d failed. allocated %d buffers from: %d\n",
 599                                    i, rc, bufs_num);
 600         }
 601 }
 602
 603 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
 604 {
 605         int i;
 606
 607         for (i = 0; i < adapter->num_queues; i++)
 608                 ena_free_rx_bufs(adapter, i);
 609 }
 610
 611 /* ena_free_tx_bufs - Free Tx Buffers per Queue
 612  * @tx_ring: TX ring for which buffers be freed
 613  */
 614 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
 615 {
 616         bool print_once = true;
 617         u32 i;
 618
 619         for (i = 0; i < tx_ring->ring_size; i++) {
 620                 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
 621                 struct ena_com_buf *ena_buf;
 622                 int nr_frags;
 623                 int j;
 624
 625                 if (!tx_info->skb)
 626                         continue;
 627
 628                 if (print_once) {
 629                         netdev_notice(tx_ring->netdev,
 630                                       "free uncompleted tx skb qid %d idx 0x%x\n",
 631                                       tx_ring->qid, i);
 632                         print_once = false;
 633                 } else {
 634                         netdev_dbg(tx_ring->netdev,
 635                                    "free uncompleted tx skb qid %d idx 0x%x\n",
 636                                    tx_ring->qid, i);
 637                 }
 638
 639                 ena_buf = tx_info->bufs;
 640                 dma_unmap_single(tx_ring->dev,
 641                                  ena_buf->paddr,
 642                                  ena_buf->len,
 643                                  DMA_TO_DEVICE);
 644
 645                 /* unmap remaining mapped pages */
 646                 nr_frags = tx_info->num_of_bufs - 1;
 647                 for (j = 0; j < nr_frags; j++) {
 648                         ena_buf++;
 649                         dma_unmap_page(tx_ring->dev,
 650                                        ena_buf->paddr,
 651                                        ena_buf->len,
 652                                        DMA_TO_DEVICE);
 653                 }
 654
 655                 dev_kfree_skb_any(tx_info->skb);
 656         }
 657         netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
 658                                                   tx_ring->qid));
 659 }
 660
 661 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
 662 {
 663         struct ena_ring *tx_ring;
 664         int i;
 665
 666         for (i = 0; i < adapter->num_queues; i++) {
 667                 tx_ring = &adapter->tx_ring[i];
 668                 ena_free_tx_bufs(tx_ring);
 669         }
 670 }
 671
 672 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
 673 {
 674         u16 ena_qid;
 675         int i;
 676
 677         for (i = 0; i < adapter->num_queues; i++) {
 678                 ena_qid = ENA_IO_TXQ_IDX(i);
 679                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
 680         }
 681 }
 682
 683 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
 684 {
 685         u16 ena_qid;
 686         int i;
 687
 688         for (i = 0; i < adapter->num_queues; i++) {
 689                 ena_qid = ENA_IO_RXQ_IDX(i);
 690                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
 691         }
 692 }
 693
 694 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
 695 {
 696         ena_destroy_all_tx_queues(adapter);
 697         ena_destroy_all_rx_queues(adapter);
 698 }
 699
 700 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
 701 {
 702         struct ena_tx_buffer *tx_info = NULL;
 703
 704         if (likely(req_id < tx_ring->ring_size)) {
 705                 tx_info = &tx_ring->tx_buffer_info[req_id];
 706                 if (likely(tx_info->skb))
 707                         return 0;
 708         }
 709
 710         if (tx_info)
 711                 netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
 712                           "tx_info doesn't have valid skb\n");
 713         else
 714                 netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
 715                           "Invalid req_id: %hu\n", req_id);
 716
 717         u64_stats_update_begin(&tx_ring->syncp);
 718         tx_ring->tx_stats.bad_req_id++;
 719         u64_stats_update_end(&tx_ring->syncp);
 720
 721         /* Trigger device reset */
 722         tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
 723         set_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags);
 724         return -EFAULT;
 725 }
 726
 727 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
 728 {
 729         struct netdev_queue *txq;
 730         bool above_thresh;
 731         u32 tx_bytes = 0;
 732         u32 total_done = 0;
 733         u16 next_to_clean;
 734         u16 req_id;
 735         int tx_pkts = 0;
 736         int rc;
 737
 738         next_to_clean = tx_ring->next_to_clean;
 739         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
 740
 741         while (tx_pkts < budget) {
 742                 struct ena_tx_buffer *tx_info;
 743                 struct sk_buff *skb;
 744                 struct ena_com_buf *ena_buf;
 745                 int i, nr_frags;
 746
 747                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
 748                                                 &req_id);
 749                 if (rc)
 750                         break;
 751
 752                 rc = validate_tx_req_id(tx_ring, req_id);
 753                 if (rc)
 754                         break;
 755
 756                 tx_info = &tx_ring->tx_buffer_info[req_id];
 757                 skb = tx_info->skb;
 758
 759                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
 760                 prefetch(&skb->end);
 761
 762                 tx_info->skb = NULL;
 763                 tx_info->last_jiffies = 0;
 764
 765                 if (likely(tx_info->num_of_bufs != 0)) {
 766                         ena_buf = tx_info->bufs;
 767
 768                         dma_unmap_single(tx_ring->dev,
 769                                          dma_unmap_addr(ena_buf, paddr),
 770                                          dma_unmap_len(ena_buf, len),
 771                                          DMA_TO_DEVICE);
 772
 773                         /* unmap remaining mapped pages */
 774                         nr_frags = tx_info->num_of_bufs - 1;
 775                         for (i = 0; i < nr_frags; i++) {
 776                                 ena_buf++;
 777                                 dma_unmap_page(tx_ring->dev,
 778                                                dma_unmap_addr(ena_buf, paddr),
 779                                                dma_unmap_len(ena_buf, len),
 780                                                DMA_TO_DEVICE);
 781                         }
 782                 }
 783
 784                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
 785                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
 786                           skb);
 787
 788                 tx_bytes += skb->len;
 789                 dev_kfree_skb(skb);
 790                 tx_pkts++;
 791                 total_done += tx_info->tx_descs;
 792
 793                 tx_ring->free_tx_ids[next_to_clean] = req_id;
 794                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
 795                                                      tx_ring->ring_size);
 796         }
 797
 798         tx_ring->next_to_clean = next_to_clean;
 799         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
 800         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
 801
 802         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
 803
 804         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
 805                   "tx_poll: q %d done. total pkts: %d\n",
 806                   tx_ring->qid, tx_pkts);
 807
 808         /* need to make the rings circular update visible to
 809          * ena_start_xmit() before checking for netif_queue_stopped().
 810          */
 811         smp_mb();
 812
 813         above_thresh = ena_com_sq_empty_space(tx_ring->ena_com_io_sq) >
 814                 ENA_TX_WAKEUP_THRESH;
 815         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
 816                 __netif_tx_lock(txq, smp_processor_id());
 817                 above_thresh = ena_com_sq_empty_space(tx_ring->ena_com_io_sq) >
 818                         ENA_TX_WAKEUP_THRESH;
 819                 if (netif_tx_queue_stopped(txq) && above_thresh) {
 820                         netif_tx_wake_queue(txq);
 821                         u64_stats_update_begin(&tx_ring->syncp);
 822                         tx_ring->tx_stats.queue_wakeup++;
 823                         u64_stats_update_end(&tx_ring->syncp);
 824                 }
 825                 __netif_tx_unlock(txq);
 826         }
 827
 828         tx_ring->per_napi_bytes += tx_bytes;
 829         tx_ring->per_napi_packets += tx_pkts;
 830
 831         return tx_pkts;
 832 }
 833
 834 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, bool frags)
 835 {
 836         struct sk_buff *skb;
 837
 838         if (frags)
 839                 skb = napi_get_frags(rx_ring->napi);
 840         else
 841                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 842                                                 rx_ring->rx_copybreak);
 843
 844         if (unlikely(!skb)) {
 845                 u64_stats_update_begin(&rx_ring->syncp);
 846                 rx_ring->rx_stats.skb_alloc_fail++;
 847                 u64_stats_update_end(&rx_ring->syncp);
 848                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
 849                           "Failed to allocate skb. frags: %d\n", frags);
 850                 return NULL;
 851         }
 852
 853         return skb;
 854 }
 855
 856 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
 857                                   struct ena_com_rx_buf_info *ena_bufs,
 858                                   u32 descs,
 859                                   u16 *next_to_clean)
 860 {
 861         struct sk_buff *skb;
 862         struct ena_rx_buffer *rx_info;
 863         u16 len, req_id, buf = 0;
 864         void *va;
 865
 866         len = ena_bufs[buf].len;
 867         req_id = ena_bufs[buf].req_id;
 868         rx_info = &rx_ring->rx_buffer_info[req_id];
 869
 870         if (unlikely(!rx_info->page)) {
 871                 netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 872                           "Page is NULL\n");
 873                 return NULL;
 874         }
 875
 876         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 877                   "rx_info %p page %p\n",
 878                   rx_info, rx_info->page);
 879
 880         /* save virt address of first buffer */
 881         va = page_address(rx_info->page) + rx_info->page_offset;
 882         prefetch(va + NET_IP_ALIGN);
 883
 884         if (len <= rx_ring->rx_copybreak) {
 885                 skb = ena_alloc_skb(rx_ring, false);
 886                 if (unlikely(!skb))
 887                         return NULL;
 888
 889                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 890                           "rx allocated small packet. len %d. data_len %d\n",
 891                           skb->len, skb->data_len);
 892
 893                 /* sync this buffer for CPU use */
 894                 dma_sync_single_for_cpu(rx_ring->dev,
 895                                         dma_unmap_addr(&rx_info->ena_buf, paddr),
 896                                         len,
 897                                         DMA_FROM_DEVICE);
 898                 skb_copy_to_linear_data(skb, va, len);
 899                 dma_sync_single_for_device(rx_ring->dev,
 900                                            dma_unmap_addr(&rx_info->ena_buf, paddr),
 901                                            len,
 902                                            DMA_FROM_DEVICE);
 903
 904                 skb_put(skb, len);
 905                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 906                 rx_ring->free_rx_ids[*next_to_clean] = req_id;
 907                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
 908                                                      rx_ring->ring_size);
 909                 return skb;
 910         }
 911
 912         skb = ena_alloc_skb(rx_ring, true);
 913         if (unlikely(!skb))
 914                 return NULL;
 915
 916         do {
 917                 dma_unmap_page(rx_ring->dev,
 918                                dma_unmap_addr(&rx_info->ena_buf, paddr),
 919                                PAGE_SIZE, DMA_FROM_DEVICE);
 920
 921                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
 922                                 rx_info->page_offset, len, PAGE_SIZE);
 923
 924                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 925                           "rx skb updated. len %d. data_len %d\n",
 926                           skb->len, skb->data_len);
 927
 928                 rx_info->page = NULL;
 929
 930                 rx_ring->free_rx_ids[*next_to_clean] = req_id;
 931                 *next_to_clean =
 932                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
 933                                              rx_ring->ring_size);
 934                 if (likely(--descs == 0))
 935                         break;
 936
 937                 buf++;
 938                 len = ena_bufs[buf].len;
 939                 req_id = ena_bufs[buf].req_id;
 940                 rx_info = &rx_ring->rx_buffer_info[req_id];
 941         } while (1);
 942
 943         return skb;
 944 }
 945
 946 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
 947  * @adapter: structure containing adapter specific data
 948  * @ena_rx_ctx: received packet context/metadata
 949  * @skb: skb currently being received and modified
 950  */
 951 static inline void ena_rx_checksum(struct ena_ring *rx_ring,
 952                                    struct ena_com_rx_ctx *ena_rx_ctx,
 953                                    struct sk_buff *skb)
 954 {
 955         /* Rx csum disabled */
 956         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
 957                 skb->ip_summed = CHECKSUM_NONE;
 958                 return;
 959         }
 960
 961         /* For fragmented packets the checksum isn't valid */
 962         if (ena_rx_ctx->frag) {
 963                 skb->ip_summed = CHECKSUM_NONE;
 964                 return;
 965         }
 966
 967         /* if IP and error */
 968         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
 969                      (ena_rx_ctx->l3_csum_err))) {
 970                 /* ipv4 checksum error */
 971                 skb->ip_summed = CHECKSUM_NONE;
 972                 u64_stats_update_begin(&rx_ring->syncp);
 973                 rx_ring->rx_stats.bad_csum++;
 974                 u64_stats_update_end(&rx_ring->syncp);
 975                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
 976                           "RX IPv4 header checksum error\n");
 977                 return;
 978         }
 979
 980         /* if TCP/UDP */
 981         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
 982                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
 983                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
 984                         /* TCP/UDP checksum error */
 985                         u64_stats_update_begin(&rx_ring->syncp);
 986                         rx_ring->rx_stats.bad_csum++;
 987                         u64_stats_update_end(&rx_ring->syncp);
 988                         netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
 989                                   "RX L4 checksum error\n");
 990                         skb->ip_summed = CHECKSUM_NONE;
 991                         return;
 992                 }
 993
 994                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 995         }
 996 }
 997
 998 static void ena_set_rx_hash(struct ena_ring *rx_ring,
 999                             struct ena_com_rx_ctx *ena_rx_ctx,
1000                             struct sk_buff *skb)
1001 {
1002         enum pkt_hash_types hash_type;
1003
1004         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1005                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1006                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1007
1008                         hash_type = PKT_HASH_TYPE_L4;
1009                 else
1010                         hash_type = PKT_HASH_TYPE_NONE;
1011
1012                 /* Override hash type if the packet is fragmented */
1013                 if (ena_rx_ctx->frag)
1014                         hash_type = PKT_HASH_TYPE_NONE;
1015
1016                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1017         }
1018 }
1019
1020 /* ena_clean_rx_irq - Cleanup RX irq
1021  * @rx_ring: RX ring to clean
1022  * @napi: napi handler
1023  * @budget: how many packets driver is allowed to clean
1024  *
1025  * Returns the number of cleaned buffers.
1026  */
1027 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1028                             u32 budget)
1029 {
1030         u16 next_to_clean = rx_ring->next_to_clean;
1031         u32 res_budget, work_done;
1032
1033         struct ena_com_rx_ctx ena_rx_ctx;
1034         struct ena_adapter *adapter;
1035         struct sk_buff *skb;
1036         int refill_required;
1037         int refill_threshold;
1038         int rc = 0;
1039         int total_len = 0;
1040         int rx_copybreak_pkt = 0;
1041         int i;
1042
1043         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1044                   "%s qid %d\n", __func__, rx_ring->qid);
1045         res_budget = budget;
1046
1047         do {
1048                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1049                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1050                 ena_rx_ctx.descs = 0;
1051                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1052                                     rx_ring->ena_com_io_sq,
1053                                     &ena_rx_ctx);
1054                 if (unlikely(rc))
1055                         goto error;
1056
1057                 if (unlikely(ena_rx_ctx.descs == 0))
1058                         break;
1059
1060                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1061                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1062                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1063                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1064
1065                 /* allocate skb and fill it */
1066                 skb = ena_rx_skb(rx_ring, rx_ring->ena_bufs, ena_rx_ctx.descs,
1067                                  &next_to_clean);
1068
1069                 /* exit if we failed to retrieve a buffer */
1070                 if (unlikely(!skb)) {
1071                         for (i = 0; i < ena_rx_ctx.descs; i++) {
1072                                 rx_ring->free_tx_ids[next_to_clean] =
1073                                         rx_ring->ena_bufs[i].req_id;
1074                                 next_to_clean =
1075                                         ENA_RX_RING_IDX_NEXT(next_to_clean,
1076                                                              rx_ring->ring_size);
1077                         }
1078                         break;
1079                 }
1080
1081                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1082
1083                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1084
1085                 skb_record_rx_queue(skb, rx_ring->qid);
1086
1087                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1088                         total_len += rx_ring->ena_bufs[0].len;
1089                         rx_copybreak_pkt++;
1090                         napi_gro_receive(napi, skb);
1091                 } else {
1092                         total_len += skb->len;
1093                         napi_gro_frags(napi);
1094                 }
1095
1096                 res_budget--;
1097         } while (likely(res_budget));
1098
1099         work_done = budget - res_budget;
1100         rx_ring->per_napi_bytes += total_len;
1101         rx_ring->per_napi_packets += work_done;
1102         u64_stats_update_begin(&rx_ring->syncp);
1103         rx_ring->rx_stats.bytes += total_len;
1104         rx_ring->rx_stats.cnt += work_done;
1105         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1106         u64_stats_update_end(&rx_ring->syncp);
1107
1108         rx_ring->next_to_clean = next_to_clean;
1109
1110         refill_required = ena_com_sq_empty_space(rx_ring->ena_com_io_sq);
1111         refill_threshold = rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER;
1112
1113         /* Optimization, try to batch new rx buffers */
1114         if (refill_required > refill_threshold) {
1115                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1116                 ena_refill_rx_bufs(rx_ring, refill_required);
1117         }
1118
1119         return work_done;
1120
1121 error:
1122         adapter = netdev_priv(rx_ring->netdev);
1123
1124         u64_stats_update_begin(&rx_ring->syncp);
1125         rx_ring->rx_stats.bad_desc_num++;
1126         u64_stats_update_end(&rx_ring->syncp);
1127
1128         /* Too many desc from the device. Trigger reset */
1129         adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1130         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1131
1132         return 0;
1133 }
1134
1135 inline void ena_adjust_intr_moderation(struct ena_ring *rx_ring,
1136                                        struct ena_ring *tx_ring)
1137 {
1138         /* We apply adaptive moderation on Rx path only.
1139          * Tx uses static interrupt moderation.
1140          */
1141         ena_com_calculate_interrupt_delay(rx_ring->ena_dev,
1142                                           rx_ring->per_napi_packets,
1143                                           rx_ring->per_napi_bytes,
1144                                           &rx_ring->smoothed_interval,
1145                                           &rx_ring->moder_tbl_idx);
1146
1147         /* Reset per napi packets/bytes */
1148         tx_ring->per_napi_packets = 0;
1149         tx_ring->per_napi_bytes = 0;
1150         rx_ring->per_napi_packets = 0;
1151         rx_ring->per_napi_bytes = 0;
1152 }
1153
1154 static inline void ena_unmask_interrupt(struct ena_ring *tx_ring,
1155                                         struct ena_ring *rx_ring)
1156 {
1157         struct ena_eth_io_intr_reg intr_reg;
1158
1159         /* Update intr register: rx intr delay,
1160          * tx intr delay and interrupt unmask
1161          */
1162         ena_com_update_intr_reg(&intr_reg,
1163                                 rx_ring->smoothed_interval,
1164                                 tx_ring->smoothed_interval,
1165                                 true);
1166
1167         /* It is a shared MSI-X.
1168          * Tx and Rx CQ have pointer to it.
1169          * So we use one of them to reach the intr reg
1170          */
1171         ena_com_unmask_intr(rx_ring->ena_com_io_cq, &intr_reg);
1172 }
1173
1174 static inline void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1175                                              struct ena_ring *rx_ring)
1176 {
1177         int cpu = get_cpu();
1178         int numa_node;
1179
1180         /* Check only one ring since the 2 rings are running on the same cpu */
1181         if (likely(tx_ring->cpu == cpu))
1182                 goto out;
1183
1184         numa_node = cpu_to_node(cpu);
1185         put_cpu();
1186
1187         if (numa_node != NUMA_NO_NODE) {
1188                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1189                 ena_com_update_numa_node(rx_ring->ena_com_io_cq, numa_node);
1190         }
1191
1192         tx_ring->cpu = cpu;
1193         rx_ring->cpu = cpu;
1194
1195         return;
1196 out:
1197         put_cpu();
1198 }
1199
1200 static int ena_io_poll(struct napi_struct *napi, int budget)
1201 {
1202         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1203         struct ena_ring *tx_ring, *rx_ring;
1204
1205         u32 tx_work_done;
1206         u32 rx_work_done;
1207         int tx_budget;
1208         int napi_comp_call = 0;
1209         int ret;
1210
1211         tx_ring = ena_napi->tx_ring;
1212         rx_ring = ena_napi->rx_ring;
1213
1214         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1215
1216         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1217             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1218                 napi_complete_done(napi, 0);
1219                 return 0;
1220         }
1221
1222         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1223         rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1224
1225         /* If the device is about to reset or down, avoid unmask
1226          * the interrupt and return 0 so NAPI won't reschedule
1227          */
1228         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1229                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1230                 napi_complete_done(napi, 0);
1231                 ret = 0;
1232
1233         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1234                 napi_comp_call = 1;
1235
1236                 /* Update numa and unmask the interrupt only when schedule
1237                  * from the interrupt context (vs from sk_busy_loop)
1238                  */
1239                 if (napi_complete_done(napi, rx_work_done)) {
1240                         /* Tx and Rx share the same interrupt vector */
1241                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1242                                 ena_adjust_intr_moderation(rx_ring, tx_ring);
1243
1244                         ena_unmask_interrupt(tx_ring, rx_ring);
1245                 }
1246
1247                 ena_update_ring_numa_node(tx_ring, rx_ring);
1248
1249                 ret = rx_work_done;
1250         } else {
1251                 ret = budget;
1252         }
1253
1254         u64_stats_update_begin(&tx_ring->syncp);
1255         tx_ring->tx_stats.napi_comp += napi_comp_call;
1256         tx_ring->tx_stats.tx_poll++;
1257         u64_stats_update_end(&tx_ring->syncp);
1258
1259         return ret;
1260 }
1261
1262 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1263 {
1264         struct ena_adapter *adapter = (struct ena_adapter *)data;
1265
1266         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1267
1268         /* Don't call the aenq handler before probe is done */
1269         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1270                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
1271
1272         return IRQ_HANDLED;
1273 }
1274
1275 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1276  * @irq: interrupt number
1277  * @data: pointer to a network interface private napi device structure
1278  */
1279 static irqreturn_t ena_intr_msix_io(int irq, void *data)
1280 {
1281         struct ena_napi *ena_napi = data;
1282
1283         ena_napi->tx_ring->first_interrupt = true;
1284         ena_napi->rx_ring->first_interrupt = true;
1285
1286         napi_schedule_irqoff(&ena_napi->napi);
1287
1288         return IRQ_HANDLED;
1289 }
1290
1291 /* Reserve a single MSI-X vector for management (admin + aenq).
1292  * plus reserve one vector for each potential io queue.
1293  * the number of potential io queues is the minimum of what the device
1294  * supports and the number of vCPUs.
1295  */
1296 static int ena_enable_msix(struct ena_adapter *adapter, int num_queues)
1297 {
1298         int msix_vecs, irq_cnt;
1299
1300         if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1301                 netif_err(adapter, probe, adapter->netdev,
1302                           "Error, MSI-X is already enabled\n");
1303                 return -EPERM;
1304         }
1305
1306         /* Reserved the max msix vectors we might need */
1307         msix_vecs = ENA_MAX_MSIX_VEC(num_queues);
1308
1309         netif_dbg(adapter, probe, adapter->netdev,
1310                   "trying to enable MSI-X, vectors %d\n", msix_vecs);
1311
1312         irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
1313                                         msix_vecs, PCI_IRQ_MSIX);
1314
1315         if (irq_cnt < 0) {
1316                 netif_err(adapter, probe, adapter->netdev,
1317                           "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
1318                 return -ENOSPC;
1319         }
1320
1321         if (irq_cnt != msix_vecs) {
1322                 netif_notice(adapter, probe, adapter->netdev,
1323                              "enable only %d MSI-X (out of %d), reduce the number of queues\n",
1324                              irq_cnt, msix_vecs);
1325                 adapter->num_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
1326         }
1327
1328         if (ena_init_rx_cpu_rmap(adapter))
1329                 netif_warn(adapter, probe, adapter->netdev,
1330                            "Failed to map IRQs to CPUs\n");
1331
1332         adapter->msix_vecs = irq_cnt;
1333         set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
1334
1335         return 0;
1336 }
1337
1338 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1339 {
1340         u32 cpu;
1341
1342         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1343                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1344                  pci_name(adapter->pdev));
1345         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
1346                 ena_intr_msix_mgmnt;
1347         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1348         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1349                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
1350         cpu = cpumask_first(cpu_online_mask);
1351         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
1352         cpumask_set_cpu(cpu,
1353                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
1354 }
1355
1356 static void ena_setup_io_intr(struct ena_adapter *adapter)
1357 {
1358         struct net_device *netdev;
1359         int irq_idx, i, cpu;
1360
1361         netdev = adapter->netdev;
1362
1363         for (i = 0; i < adapter->num_queues; i++) {
1364                 irq_idx = ENA_IO_IRQ_IDX(i);
1365                 cpu = i % num_online_cpus();
1366
1367                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1368                          "%s-Tx-Rx-%d", netdev->name, i);
1369                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
1370                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
1371                 adapter->irq_tbl[irq_idx].vector =
1372                         pci_irq_vector(adapter->pdev, irq_idx);
1373                 adapter->irq_tbl[irq_idx].cpu = cpu;
1374
1375                 cpumask_set_cpu(cpu,
1376                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
1377         }
1378 }
1379
1380 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
1381 {
1382         unsigned long flags = 0;
1383         struct ena_irq *irq;
1384         int rc;
1385
1386         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1387         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1388                          irq->data);
1389         if (rc) {
1390                 netif_err(adapter, probe, adapter->netdev,
1391                           "failed to request admin irq\n");
1392                 return rc;
1393         }
1394
1395         netif_dbg(adapter, probe, adapter->netdev,
1396                   "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
1397                   irq->affinity_hint_mask.bits[0], irq->vector);
1398
1399         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1400
1401         return rc;
1402 }
1403
1404 static int ena_request_io_irq(struct ena_adapter *adapter)
1405 {
1406         unsigned long flags = 0;
1407         struct ena_irq *irq;
1408         int rc = 0, i, k;
1409
1410         if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1411                 netif_err(adapter, ifup, adapter->netdev,
1412                           "Failed to request I/O IRQ: MSI-X is not enabled\n");
1413                 return -EINVAL;
1414         }
1415
1416         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1417                 irq = &adapter->irq_tbl[i];
1418                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1419                                  irq->data);
1420                 if (rc) {
1421                         netif_err(adapter, ifup, adapter->netdev,
1422                                   "Failed to request I/O IRQ. index %d rc %d\n",
1423                                    i, rc);
1424                         goto err;
1425                 }
1426
1427                 netif_dbg(adapter, ifup, adapter->netdev,
1428                           "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
1429                           i, irq->affinity_hint_mask.bits[0], irq->vector);
1430
1431                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1432         }
1433
1434         return rc;
1435
1436 err:
1437         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
1438                 irq = &adapter->irq_tbl[k];
1439                 free_irq(irq->vector, irq->data);
1440         }
1441
1442         return rc;
1443 }
1444
1445 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
1446 {
1447         struct ena_irq *irq;
1448
1449         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1450         synchronize_irq(irq->vector);
1451         irq_set_affinity_hint(irq->vector, NULL);
1452         free_irq(irq->vector, irq->data);
1453 }
1454
1455 static void ena_free_io_irq(struct ena_adapter *adapter)
1456 {
1457         struct ena_irq *irq;
1458         int i;
1459
1460 #ifdef CONFIG_RFS_ACCEL
1461         if (adapter->msix_vecs >= 1) {
1462                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
1463                 adapter->netdev->rx_cpu_rmap = NULL;
1464         }
1465 #endif /* CONFIG_RFS_ACCEL */
1466
1467         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1468                 irq = &adapter->irq_tbl[i];
1469                 irq_set_affinity_hint(irq->vector, NULL);
1470                 free_irq(irq->vector, irq->data);
1471         }
1472 }
1473
1474 static void ena_disable_msix(struct ena_adapter *adapter)
1475 {
1476         if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
1477                 pci_free_irq_vectors(adapter->pdev);
1478 }
1479
1480 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
1481 {
1482         int i;
1483
1484         if (!netif_running(adapter->netdev))
1485                 return;
1486
1487         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++)
1488                 synchronize_irq(adapter->irq_tbl[i].vector);
1489 }
1490
1491 static void ena_del_napi(struct ena_adapter *adapter)
1492 {
1493         int i;
1494
1495         for (i = 0; i < adapter->num_queues; i++)
1496                 netif_napi_del(&adapter->ena_napi[i].napi);
1497 }
1498
1499 static void ena_init_napi(struct ena_adapter *adapter)
1500 {
1501         struct ena_napi *napi;
1502         int i;
1503
1504         for (i = 0; i < adapter->num_queues; i++) {
1505                 napi = &adapter->ena_napi[i];
1506
1507                 netif_napi_add(adapter->netdev,
1508                                &adapter->ena_napi[i].napi,
1509                                ena_io_poll,
1510                                ENA_NAPI_BUDGET);
1511                 napi->rx_ring = &adapter->rx_ring[i];
1512                 napi->tx_ring = &adapter->tx_ring[i];
1513                 napi->qid = i;
1514         }
1515 }
1516
1517 static void ena_napi_disable_all(struct ena_adapter *adapter)
1518 {
1519         int i;
1520
1521         for (i = 0; i < adapter->num_queues; i++)
1522                 napi_disable(&adapter->ena_napi[i].napi);
1523 }
1524
1525 static void ena_napi_enable_all(struct ena_adapter *adapter)
1526 {
1527         int i;
1528
1529         for (i = 0; i < adapter->num_queues; i++)
1530                 napi_enable(&adapter->ena_napi[i].napi);
1531 }
1532
1533 static void ena_restore_ethtool_params(struct ena_adapter *adapter)
1534 {
1535         adapter->tx_usecs = 0;
1536         adapter->rx_usecs = 0;
1537         adapter->tx_frames = 1;
1538         adapter->rx_frames = 1;
1539 }
1540
1541 /* Configure the Rx forwarding */
1542 static int ena_rss_configure(struct ena_adapter *adapter)
1543 {
1544         struct ena_com_dev *ena_dev = adapter->ena_dev;
1545         int rc;
1546
1547         /* In case the RSS table wasn't initialized by probe */
1548         if (!ena_dev->rss.tbl_log_size) {
1549                 rc = ena_rss_init_default(adapter);
1550                 if (rc && (rc != -EOPNOTSUPP)) {
1551                         netif_err(adapter, ifup, adapter->netdev,
1552                                   "Failed to init RSS rc: %d\n", rc);
1553                         return rc;
1554                 }
1555         }
1556
1557         /* Set indirect table */
1558         rc = ena_com_indirect_table_set(ena_dev);
1559         if (unlikely(rc && rc != -EOPNOTSUPP))
1560                 return rc;
1561
1562         /* Configure hash function (if supported) */
1563         rc = ena_com_set_hash_function(ena_dev);
1564         if (unlikely(rc && (rc != -EOPNOTSUPP)))
1565                 return rc;
1566
1567         /* Configure hash inputs (if supported) */
1568         rc = ena_com_set_hash_ctrl(ena_dev);
1569         if (unlikely(rc && (rc != -EOPNOTSUPP)))
1570                 return rc;
1571
1572         return 0;
1573 }
1574
1575 static int ena_up_complete(struct ena_adapter *adapter)
1576 {
1577         int rc;
1578
1579         rc = ena_rss_configure(adapter);
1580         if (rc)
1581                 return rc;
1582
1583         ena_init_napi(adapter);
1584
1585         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
1586
1587         ena_refill_all_rx_bufs(adapter);
1588
1589         /* enable transmits */
1590         netif_tx_start_all_queues(adapter->netdev);
1591
1592         ena_restore_ethtool_params(adapter);
1593
1594         ena_napi_enable_all(adapter);
1595
1596         return 0;
1597 }
1598
1599 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
1600 {
1601         struct ena_com_create_io_ctx ctx = { 0 };
1602         struct ena_com_dev *ena_dev;
1603         struct ena_ring *tx_ring;
1604         u32 msix_vector;
1605         u16 ena_qid;
1606         int rc;
1607
1608         ena_dev = adapter->ena_dev;
1609
1610         tx_ring = &adapter->tx_ring[qid];
1611         msix_vector = ENA_IO_IRQ_IDX(qid);
1612         ena_qid = ENA_IO_TXQ_IDX(qid);
1613
1614         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1615         ctx.qid = ena_qid;
1616         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1617         ctx.msix_vector = msix_vector;
1618         ctx.queue_size = adapter->tx_ring_size;
1619         ctx.numa_node = cpu_to_node(tx_ring->cpu);
1620
1621         rc = ena_com_create_io_queue(ena_dev, &ctx);
1622         if (rc) {
1623                 netif_err(adapter, ifup, adapter->netdev,
1624                           "Failed to create I/O TX queue num %d rc: %d\n",
1625                           qid, rc);
1626                 return rc;
1627         }
1628
1629         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1630                                      &tx_ring->ena_com_io_sq,
1631                                      &tx_ring->ena_com_io_cq);
1632         if (rc) {
1633                 netif_err(adapter, ifup, adapter->netdev,
1634                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1635                           qid, rc);
1636                 ena_com_destroy_io_queue(ena_dev, ena_qid);
1637                 return rc;
1638         }
1639
1640         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
1641         return rc;
1642 }
1643
1644 static int ena_create_all_io_tx_queues(struct ena_adapter *adapter)
1645 {
1646         struct ena_com_dev *ena_dev = adapter->ena_dev;
1647         int rc, i;
1648
1649         for (i = 0; i < adapter->num_queues; i++) {
1650                 rc = ena_create_io_tx_queue(adapter, i);
1651                 if (rc)
1652                         goto create_err;
1653         }
1654
1655         return 0;
1656
1657 create_err:
1658         while (i--)
1659                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1660
1661         return rc;
1662 }
1663
1664 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
1665 {
1666         struct ena_com_dev *ena_dev;
1667         struct ena_com_create_io_ctx ctx = { 0 };
1668         struct ena_ring *rx_ring;
1669         u32 msix_vector;
1670         u16 ena_qid;
1671         int rc;
1672
1673         ena_dev = adapter->ena_dev;
1674
1675         rx_ring = &adapter->rx_ring[qid];
1676         msix_vector = ENA_IO_IRQ_IDX(qid);
1677         ena_qid = ENA_IO_RXQ_IDX(qid);
1678
1679         ctx.qid = ena_qid;
1680         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1681         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1682         ctx.msix_vector = msix_vector;
1683         ctx.queue_size = adapter->rx_ring_size;
1684         ctx.numa_node = cpu_to_node(rx_ring->cpu);
1685
1686         rc = ena_com_create_io_queue(ena_dev, &ctx);
1687         if (rc) {
1688                 netif_err(adapter, ifup, adapter->netdev,
1689                           "Failed to create I/O RX queue num %d rc: %d\n",
1690                           qid, rc);
1691                 return rc;
1692         }
1693
1694         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1695                                      &rx_ring->ena_com_io_sq,
1696                                      &rx_ring->ena_com_io_cq);
1697         if (rc) {
1698                 netif_err(adapter, ifup, adapter->netdev,
1699                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1700                           qid, rc);
1701                 ena_com_destroy_io_queue(ena_dev, ena_qid);
1702                 return rc;
1703         }
1704
1705         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
1706
1707         return rc;
1708 }
1709
1710 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
1711 {
1712         struct ena_com_dev *ena_dev = adapter->ena_dev;
1713         int rc, i;
1714
1715         for (i = 0; i < adapter->num_queues; i++) {
1716                 rc = ena_create_io_rx_queue(adapter, i);
1717                 if (rc)
1718                         goto create_err;
1719         }
1720
1721         return 0;
1722
1723 create_err:
1724         while (i--)
1725                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1726
1727         return rc;
1728 }
1729
1730 static int ena_up(struct ena_adapter *adapter)
1731 {
1732         int rc, i;
1733
1734         netdev_dbg(adapter->netdev, "%s\n", __func__);
1735
1736         ena_setup_io_intr(adapter);
1737
1738         rc = ena_request_io_irq(adapter);
1739         if (rc)
1740                 goto err_req_irq;
1741
1742         /* allocate transmit descriptors */
1743         rc = ena_setup_all_tx_resources(adapter);
1744         if (rc)
1745                 goto err_setup_tx;
1746
1747         /* allocate receive descriptors */
1748         rc = ena_setup_all_rx_resources(adapter);
1749         if (rc)
1750                 goto err_setup_rx;
1751
1752         /* Create TX queues */
1753         rc = ena_create_all_io_tx_queues(adapter);
1754         if (rc)
1755                 goto err_create_tx_queues;
1756
1757         /* Create RX queues */
1758         rc = ena_create_all_io_rx_queues(adapter);
1759         if (rc)
1760                 goto err_create_rx_queues;
1761
1762         rc = ena_up_complete(adapter);
1763         if (rc)
1764                 goto err_up;
1765
1766         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
1767                 netif_carrier_on(adapter->netdev);
1768
1769         u64_stats_update_begin(&adapter->syncp);
1770         adapter->dev_stats.interface_up++;
1771         u64_stats_update_end(&adapter->syncp);
1772
1773         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1774
1775         /* Enable completion queues interrupt */
1776         for (i = 0; i < adapter->num_queues; i++)
1777                 ena_unmask_interrupt(&adapter->tx_ring[i],
1778                                      &adapter->rx_ring[i]);
1779
1780         /* schedule napi in case we had pending packets
1781          * from the last time we disable napi
1782          */
1783         for (i = 0; i < adapter->num_queues; i++)
1784                 napi_schedule(&adapter->ena_napi[i].napi);
1785
1786         return rc;
1787
1788 err_up:
1789         ena_destroy_all_rx_queues(adapter);
1790 err_create_rx_queues:
1791         ena_destroy_all_tx_queues(adapter);
1792 err_create_tx_queues:
1793         ena_free_all_io_rx_resources(adapter);
1794 err_setup_rx:
1795         ena_free_all_io_tx_resources(adapter);
1796 err_setup_tx:
1797         ena_free_io_irq(adapter);
1798 err_req_irq:
1799
1800         return rc;
1801 }
1802
1803 static void ena_down(struct ena_adapter *adapter)
1804 {
1805         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
1806
1807         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1808
1809         u64_stats_update_begin(&adapter->syncp);
1810         adapter->dev_stats.interface_down++;
1811         u64_stats_update_end(&adapter->syncp);
1812
1813         netif_carrier_off(adapter->netdev);
1814         netif_tx_disable(adapter->netdev);
1815
1816         /* After this point the napi handler won't enable the tx queue */
1817         ena_napi_disable_all(adapter);
1818
1819         /* After destroy the queue there won't be any new interrupts */
1820
1821         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
1822                 int rc;
1823
1824                 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
1825                 if (rc)
1826                         dev_err(&adapter->pdev->dev, "Device reset failed\n");
1827         }
1828
1829         ena_destroy_all_io_queues(adapter);
1830
1831         ena_disable_io_intr_sync(adapter);
1832         ena_free_io_irq(adapter);
1833         ena_del_napi(adapter);
1834
1835         ena_free_all_tx_bufs(adapter);
1836         ena_free_all_rx_bufs(adapter);
1837         ena_free_all_io_tx_resources(adapter);
1838         ena_free_all_io_rx_resources(adapter);
1839 }
1840
1841 /* ena_open - Called when a network interface is made active
1842  * @netdev: network interface device structure
1843  *
1844  * Returns 0 on success, negative value on failure
1845  *
1846  * The open entry point is called when a network interface is made
1847  * active by the system (IFF_UP).  At this point all resources needed
1848  * for transmit and receive operations are allocated, the interrupt
1849  * handler is registered with the OS, the watchdog timer is started,
1850  * and the stack is notified that the interface is ready.
1851  */
1852 static int ena_open(struct net_device *netdev)
1853 {
1854         struct ena_adapter *adapter = netdev_priv(netdev);
1855         int rc;
1856
1857         /* Notify the stack of the actual queue counts. */
1858         rc = netif_set_real_num_tx_queues(netdev, adapter->num_queues);
1859         if (rc) {
1860                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
1861                 return rc;
1862         }
1863
1864         rc = netif_set_real_num_rx_queues(netdev, adapter->num_queues);
1865         if (rc) {
1866                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
1867                 return rc;
1868         }
1869
1870         rc = ena_up(adapter);
1871         if (rc)
1872                 return rc;
1873
1874         return rc;
1875 }
1876
1877 /* ena_close - Disables a network interface
1878  * @netdev: network interface device structure
1879  *
1880  * Returns 0, this is not allowed to fail
1881  *
1882  * The close entry point is called when an interface is de-activated
1883  * by the OS.  The hardware is still under the drivers control, but
1884  * needs to be disabled.  A global MAC reset is issued to stop the
1885  * hardware, and all transmit and receive resources are freed.
1886  */
1887 static int ena_close(struct net_device *netdev)
1888 {
1889         struct ena_adapter *adapter = netdev_priv(netdev);
1890
1891         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
1892
1893         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
1894                 ena_down(adapter);
1895
1896         /* Check for device status and issue reset if needed*/
1897         check_for_admin_com_state(adapter);
1898         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
1899                 netif_err(adapter, ifdown, adapter->netdev,
1900                           "Destroy failure, restarting device\n");
1901                 ena_dump_stats_to_dmesg(adapter);
1902                 /* rtnl lock already obtained in dev_ioctl() layer */
1903                 ena_destroy_device(adapter);
1904                 ena_restore_device(adapter);
1905         }
1906
1907         return 0;
1908 }
1909
1910 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct sk_buff *skb)
1911 {
1912         u32 mss = skb_shinfo(skb)->gso_size;
1913         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
1914         u8 l4_protocol = 0;
1915
1916         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
1917                 ena_tx_ctx->l4_csum_enable = 1;
1918                 if (mss) {
1919                         ena_tx_ctx->tso_enable = 1;
1920                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
1921                         ena_tx_ctx->l4_csum_partial = 0;
1922                 } else {
1923                         ena_tx_ctx->tso_enable = 0;
1924                         ena_meta->l4_hdr_len = 0;
1925                         ena_tx_ctx->l4_csum_partial = 1;
1926                 }
1927
1928                 switch (ip_hdr(skb)->version) {
1929                 case IPVERSION:
1930                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
1931                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
1932                                 ena_tx_ctx->df = 1;
1933                         if (mss)
1934                                 ena_tx_ctx->l3_csum_enable = 1;
1935                         l4_protocol = ip_hdr(skb)->protocol;
1936                         break;
1937                 case 6:
1938                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
1939                         l4_protocol = ipv6_hdr(skb)->nexthdr;
1940                         break;
1941                 default:
1942                         break;
1943                 }
1944
1945                 if (l4_protocol == IPPROTO_TCP)
1946                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
1947                 else
1948                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
1949
1950                 ena_meta->mss = mss;
1951                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
1952                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
1953                 ena_tx_ctx->meta_valid = 1;
1954
1955         } else {
1956                 ena_tx_ctx->meta_valid = 0;
1957         }
1958 }
1959
1960 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
1961                                        struct sk_buff *skb)
1962 {
1963         int num_frags, header_len, rc;
1964
1965         num_frags = skb_shinfo(skb)->nr_frags;
1966         header_len = skb_headlen(skb);
1967
1968         if (num_frags < tx_ring->sgl_size)
1969                 return 0;
1970
1971         if ((num_frags == tx_ring->sgl_size) &&
1972             (header_len < tx_ring->tx_max_header_size))
1973                 return 0;
1974
1975         u64_stats_update_begin(&tx_ring->syncp);
1976         tx_ring->tx_stats.linearize++;
1977         u64_stats_update_end(&tx_ring->syncp);
1978
1979         rc = skb_linearize(skb);
1980         if (unlikely(rc)) {
1981                 u64_stats_update_begin(&tx_ring->syncp);
1982                 tx_ring->tx_stats.linearize_failed++;
1983                 u64_stats_update_end(&tx_ring->syncp);
1984         }
1985
1986         return rc;
1987 }
1988
1989 /* Called with netif_tx_lock. */
1990 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
1991 {
1992         struct ena_adapter *adapter = netdev_priv(dev);
1993         struct ena_tx_buffer *tx_info;
1994         struct ena_com_tx_ctx ena_tx_ctx;
1995         struct ena_ring *tx_ring;
1996         struct netdev_queue *txq;
1997         struct ena_com_buf *ena_buf;
1998         void *push_hdr;
1999         u32 len, last_frag;
2000         u16 next_to_use;
2001         u16 req_id;
2002         u16 push_len;
2003         u16 header_len;
2004         dma_addr_t dma;
2005         int qid, rc, nb_hw_desc;
2006         int i = -1;
2007
2008         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
2009         /*  Determine which tx ring we will be placed on */
2010         qid = skb_get_queue_mapping(skb);
2011         tx_ring = &adapter->tx_ring[qid];
2012         txq = netdev_get_tx_queue(dev, qid);
2013
2014         rc = ena_check_and_linearize_skb(tx_ring, skb);
2015         if (unlikely(rc))
2016                 goto error_drop_packet;
2017
2018         skb_tx_timestamp(skb);
2019         len = skb_headlen(skb);
2020
2021         next_to_use = tx_ring->next_to_use;
2022         req_id = tx_ring->free_tx_ids[next_to_use];
2023         tx_info = &tx_ring->tx_buffer_info[req_id];
2024         tx_info->num_of_bufs = 0;
2025
2026         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
2027         ena_buf = tx_info->bufs;
2028         tx_info->skb = skb;
2029
2030         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2031                 /* prepared the push buffer */
2032                 push_len = min_t(u32, len, tx_ring->tx_max_header_size);
2033                 header_len = push_len;
2034                 push_hdr = skb->data;
2035         } else {
2036                 push_len = 0;
2037                 header_len = min_t(u32, len, tx_ring->tx_max_header_size);
2038                 push_hdr = NULL;
2039         }
2040
2041         netif_dbg(adapter, tx_queued, dev,
2042                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2043                   push_hdr, push_len);
2044
2045         if (len > push_len) {
2046                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2047                                      len - push_len, DMA_TO_DEVICE);
2048                 if (dma_mapping_error(tx_ring->dev, dma))
2049                         goto error_report_dma_error;
2050
2051                 ena_buf->paddr = dma;
2052                 ena_buf->len = len - push_len;
2053
2054                 ena_buf++;
2055                 tx_info->num_of_bufs++;
2056         }
2057
2058         last_frag = skb_shinfo(skb)->nr_frags;
2059
2060         for (i = 0; i < last_frag; i++) {
2061                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2062
2063                 len = skb_frag_size(frag);
2064                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, len,
2065                                        DMA_TO_DEVICE);
2066                 if (dma_mapping_error(tx_ring->dev, dma))
2067                         goto error_report_dma_error;
2068
2069                 ena_buf->paddr = dma;
2070                 ena_buf->len = len;
2071                 ena_buf++;
2072         }
2073
2074         tx_info->num_of_bufs += last_frag;
2075
2076         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2077         ena_tx_ctx.ena_bufs = tx_info->bufs;
2078         ena_tx_ctx.push_header = push_hdr;
2079         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2080         ena_tx_ctx.req_id = req_id;
2081         ena_tx_ctx.header_len = header_len;
2082
2083         /* set flags and meta data */
2084         ena_tx_csum(&ena_tx_ctx, skb);
2085
2086         /* prepare the packet's descriptors to dma engine */
2087         rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq, &ena_tx_ctx,
2088                                 &nb_hw_desc);
2089
2090         if (unlikely(rc)) {
2091                 netif_err(adapter, tx_queued, dev,
2092                           "failed to prepare tx bufs\n");
2093                 u64_stats_update_begin(&tx_ring->syncp);
2094                 tx_ring->tx_stats.queue_stop++;
2095                 tx_ring->tx_stats.prepare_ctx_err++;
2096                 u64_stats_update_end(&tx_ring->syncp);
2097                 netif_tx_stop_queue(txq);
2098                 goto error_unmap_dma;
2099         }
2100
2101         netdev_tx_sent_queue(txq, skb->len);
2102
2103         u64_stats_update_begin(&tx_ring->syncp);
2104         tx_ring->tx_stats.cnt++;
2105         tx_ring->tx_stats.bytes += skb->len;
2106         u64_stats_update_end(&tx_ring->syncp);
2107
2108         tx_info->tx_descs = nb_hw_desc;
2109         tx_info->last_jiffies = jiffies;
2110         tx_info->print_once = 0;
2111
2112         tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
2113                 tx_ring->ring_size);
2114
2115         /* This WMB is aimed to:
2116          * 1 - perform smp barrier before reading next_to_completion
2117          * 2 - make sure the desc were written before trigger DB
2118          */
2119         wmb();
2120
2121         /* stop the queue when no more space available, the packet can have up
2122          * to sgl_size + 2. one for the meta descriptor and one for header
2123          * (if the header is larger than tx_max_header_size).
2124          */
2125         if (unlikely(ena_com_sq_empty_space(tx_ring->ena_com_io_sq) <
2126                      (tx_ring->sgl_size + 2))) {
2127                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
2128                           __func__, qid);
2129
2130                 netif_tx_stop_queue(txq);
2131                 u64_stats_update_begin(&tx_ring->syncp);
2132                 tx_ring->tx_stats.queue_stop++;
2133                 u64_stats_update_end(&tx_ring->syncp);
2134
2135                 /* There is a rare condition where this function decide to
2136                  * stop the queue but meanwhile clean_tx_irq updates
2137                  * next_to_completion and terminates.
2138                  * The queue will remain stopped forever.
2139                  * To solve this issue this function perform rmb, check
2140                  * the wakeup condition and wake up the queue if needed.
2141                  */
2142                 smp_rmb();
2143
2144                 if (ena_com_sq_empty_space(tx_ring->ena_com_io_sq)
2145                                 > ENA_TX_WAKEUP_THRESH) {
2146                         netif_tx_wake_queue(txq);
2147                         u64_stats_update_begin(&tx_ring->syncp);
2148                         tx_ring->tx_stats.queue_wakeup++;
2149                         u64_stats_update_end(&tx_ring->syncp);
2150                 }
2151         }
2152
2153         if (netif_xmit_stopped(txq) || !skb->xmit_more) {
2154                 /* trigger the dma engine */
2155                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq, false);
2156                 u64_stats_update_begin(&tx_ring->syncp);
2157                 tx_ring->tx_stats.doorbells++;
2158                 u64_stats_update_end(&tx_ring->syncp);
2159         }
2160
2161         return NETDEV_TX_OK;
2162
2163 error_report_dma_error:
2164         u64_stats_update_begin(&tx_ring->syncp);
2165         tx_ring->tx_stats.dma_mapping_err++;
2166         u64_stats_update_end(&tx_ring->syncp);
2167         netdev_warn(adapter->netdev, "failed to map skb\n");
2168
2169         tx_info->skb = NULL;
2170
2171 error_unmap_dma:
2172         if (i >= 0) {
2173                 /* save value of frag that failed */
2174                 last_frag = i;
2175
2176                 /* start back at beginning and unmap skb */
2177                 tx_info->skb = NULL;
2178                 ena_buf = tx_info->bufs;
2179                 dma_unmap_single(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
2180                                  dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
2181
2182                 /* unmap remaining mapped pages */
2183                 for (i = 0; i < last_frag; i++) {
2184                         ena_buf++;
2185                         dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
2186                                        dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
2187                 }
2188         }
2189
2190 error_drop_packet:
2191
2192         dev_kfree_skb(skb);
2193         return NETDEV_TX_OK;
2194 }
2195
2196 #ifdef CONFIG_NET_POLL_CONTROLLER
2197 static void ena_netpoll(struct net_device *netdev)
2198 {
2199         struct ena_adapter *adapter = netdev_priv(netdev);
2200         int i;
2201
2202         /* Dont schedule NAPI if the driver is in the middle of reset
2203          * or netdev is down.
2204          */
2205
2206         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags) ||
2207             test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2208                 return;
2209
2210         for (i = 0; i < adapter->num_queues; i++)
2211                 napi_schedule(&adapter->ena_napi[i].napi);
2212 }
2213 #endif /* CONFIG_NET_POLL_CONTROLLER */
2214
2215 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
2216                             void *accel_priv, select_queue_fallback_t fallback)
2217 {
2218         u16 qid;
2219         /* we suspect that this is good for in--kernel network services that
2220          * want to loop incoming skb rx to tx in normal user generated traffic,
2221          * most probably we will not get to this
2222          */
2223         if (skb_rx_queue_recorded(skb))
2224                 qid = skb_get_rx_queue(skb);
2225         else
2226                 qid = fallback(dev, skb);
2227
2228         return qid;
2229 }
2230
2231 static void ena_config_host_info(struct ena_com_dev *ena_dev)
2232 {
2233         struct ena_admin_host_info *host_info;
2234         int rc;
2235
2236         /* Allocate only the host info */
2237         rc = ena_com_allocate_host_info(ena_dev);
2238         if (rc) {
2239                 pr_err("Cannot allocate host info\n");
2240                 return;
2241         }
2242
2243         host_info = ena_dev->host_attr.host_info;
2244
2245         host_info->os_type = ENA_ADMIN_OS_LINUX;
2246         host_info->kernel_ver = LINUX_VERSION_CODE;
2247         strncpy(host_info->kernel_ver_str, utsname()->version,
2248                 sizeof(host_info->kernel_ver_str) - 1);
2249         host_info->os_dist = 0;
2250         strncpy(host_info->os_dist_str, utsname()->release,
2251                 sizeof(host_info->os_dist_str) - 1);
2252         host_info->driver_version =
2253                 (DRV_MODULE_VER_MAJOR) |
2254                 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
2255                 (DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
2256
2257         rc = ena_com_set_host_attributes(ena_dev);
2258         if (rc) {
2259                 if (rc == -EOPNOTSUPP)
2260                         pr_warn("Cannot set host attributes\n");
2261                 else
2262                         pr_err("Cannot set host attributes\n");
2263
2264                 goto err;
2265         }
2266
2267         return;
2268
2269 err:
2270         ena_com_delete_host_info(ena_dev);
2271 }
2272
2273 static void ena_config_debug_area(struct ena_adapter *adapter)
2274 {
2275         u32 debug_area_size;
2276         int rc, ss_count;
2277
2278         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
2279         if (ss_count <= 0) {
2280                 netif_err(adapter, drv, adapter->netdev,
2281                           "SS count is negative\n");
2282                 return;
2283         }
2284
2285         /* allocate 32 bytes for each string and 64bit for the value */
2286         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
2287
2288         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
2289         if (rc) {
2290                 pr_err("Cannot allocate debug area\n");
2291                 return;
2292         }
2293
2294         rc = ena_com_set_host_attributes(adapter->ena_dev);
2295         if (rc) {
2296                 if (rc == -EOPNOTSUPP)
2297                         netif_warn(adapter, drv, adapter->netdev,
2298                                    "Cannot set host attributes\n");
2299                 else
2300                         netif_err(adapter, drv, adapter->netdev,
2301                                   "Cannot set host attributes\n");
2302                 goto err;
2303         }
2304
2305         return;
2306 err:
2307         ena_com_delete_debug_area(adapter->ena_dev);
2308 }
2309
2310 static void ena_get_stats64(struct net_device *netdev,
2311                             struct rtnl_link_stats64 *stats)
2312 {
2313         struct ena_adapter *adapter = netdev_priv(netdev);
2314         struct ena_ring *rx_ring, *tx_ring;
2315         unsigned int start;
2316         u64 rx_drops;
2317         int i;
2318
2319         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2320                 return;
2321
2322         for (i = 0; i < adapter->num_queues; i++) {
2323                 u64 bytes, packets;
2324
2325                 tx_ring = &adapter->tx_ring[i];
2326
2327                 do {
2328                         start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
2329                         packets = tx_ring->tx_stats.cnt;
2330                         bytes = tx_ring->tx_stats.bytes;
2331                 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
2332
2333                 stats->tx_packets += packets;
2334                 stats->tx_bytes += bytes;
2335
2336                 rx_ring = &adapter->rx_ring[i];
2337
2338                 do {
2339                         start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
2340                         packets = rx_ring->rx_stats.cnt;
2341                         bytes = rx_ring->rx_stats.bytes;
2342                 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
2343
2344                 stats->rx_packets += packets;
2345                 stats->rx_bytes += bytes;
2346         }
2347
2348         do {
2349                 start = u64_stats_fetch_begin_irq(&adapter->syncp);
2350                 rx_drops = adapter->dev_stats.rx_drops;
2351         } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
2352
2353         stats->rx_dropped = rx_drops;
2354
2355         stats->multicast = 0;
2356         stats->collisions = 0;
2357
2358         stats->rx_length_errors = 0;
2359         stats->rx_crc_errors = 0;
2360         stats->rx_frame_errors = 0;
2361         stats->rx_fifo_errors = 0;
2362         stats->rx_missed_errors = 0;
2363         stats->tx_window_errors = 0;
2364
2365         stats->rx_errors = 0;
2366         stats->tx_errors = 0;
2367 }
2368
2369 static const struct net_device_ops ena_netdev_ops = {
2370         .ndo_open               = ena_open,
2371         .ndo_stop               = ena_close,
2372         .ndo_start_xmit         = ena_start_xmit,
2373         .ndo_select_queue       = ena_select_queue,
2374         .ndo_get_stats64        = ena_get_stats64,
2375         .ndo_tx_timeout         = ena_tx_timeout,
2376         .ndo_change_mtu         = ena_change_mtu,
2377         .ndo_set_mac_address    = NULL,
2378         .ndo_validate_addr      = eth_validate_addr,
2379 #ifdef CONFIG_NET_POLL_CONTROLLER
2380         .ndo_poll_controller    = ena_netpoll,
2381 #endif /* CONFIG_NET_POLL_CONTROLLER */
2382 };
2383
2384 static int ena_device_validate_params(struct ena_adapter *adapter,
2385                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
2386 {
2387         struct net_device *netdev = adapter->netdev;
2388         int rc;
2389
2390         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
2391                               adapter->mac_addr);
2392         if (!rc) {
2393                 netif_err(adapter, drv, netdev,
2394                           "Error, mac address are different\n");
2395                 return -EINVAL;
2396         }
2397
2398         if ((get_feat_ctx->max_queues.max_cq_num < adapter->num_queues) ||
2399             (get_feat_ctx->max_queues.max_sq_num < adapter->num_queues)) {
2400                 netif_err(adapter, drv, netdev,
2401                           "Error, device doesn't support enough queues\n");
2402                 return -EINVAL;
2403         }
2404
2405         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
2406                 netif_err(adapter, drv, netdev,
2407                           "Error, device max mtu is smaller than netdev MTU\n");
2408                 return -EINVAL;
2409         }
2410
2411         return 0;
2412 }
2413
2414 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
2415                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
2416                            bool *wd_state)
2417 {
2418         struct device *dev = &pdev->dev;
2419         bool readless_supported;
2420         u32 aenq_groups;
2421         int dma_width;
2422         int rc;
2423
2424         rc = ena_com_mmio_reg_read_request_init(ena_dev);
2425         if (rc) {
2426                 dev_err(dev, "failed to init mmio read less\n");
2427                 return rc;
2428         }
2429
2430         /* The PCIe configuration space revision id indicate if mmio reg
2431          * read is disabled
2432          */
2433         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
2434         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
2435
2436         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
2437         if (rc) {
2438                 dev_err(dev, "Can not reset device\n");
2439                 goto err_mmio_read_less;
2440         }
2441
2442         rc = ena_com_validate_version(ena_dev);
2443         if (rc) {
2444                 dev_err(dev, "device version is too low\n");
2445                 goto err_mmio_read_less;
2446         }
2447
2448         dma_width = ena_com_get_dma_width(ena_dev);
2449         if (dma_width < 0) {
2450                 dev_err(dev, "Invalid dma width value %d", dma_width);
2451                 rc = dma_width;
2452                 goto err_mmio_read_less;
2453         }
2454
2455         rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2456         if (rc) {
2457                 dev_err(dev, "pci_set_dma_mask failed 0x%x\n", rc);
2458                 goto err_mmio_read_less;
2459         }
2460
2461         rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2462         if (rc) {
2463                 dev_err(dev, "err_pci_set_consistent_dma_mask failed 0x%x\n",
2464                         rc);
2465                 goto err_mmio_read_less;
2466         }
2467
2468         /* ENA admin level init */
2469         rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
2470         if (rc) {
2471                 dev_err(dev,
2472                         "Can not initialize ena admin queue with device\n");
2473                 goto err_mmio_read_less;
2474         }
2475
2476         /* To enable the msix interrupts the driver needs to know the number
2477          * of queues. So the driver uses polling mode to retrieve this
2478          * information
2479          */
2480         ena_com_set_admin_polling_mode(ena_dev, true);
2481
2482         ena_config_host_info(ena_dev);
2483
2484         /* Get Device Attributes*/
2485         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
2486         if (rc) {
2487                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
2488                 goto err_admin_init;
2489         }
2490
2491         /* Try to turn all the available aenq groups */
2492         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
2493                 BIT(ENA_ADMIN_FATAL_ERROR) |
2494                 BIT(ENA_ADMIN_WARNING) |
2495                 BIT(ENA_ADMIN_NOTIFICATION) |
2496                 BIT(ENA_ADMIN_KEEP_ALIVE);
2497
2498         aenq_groups &= get_feat_ctx->aenq.supported_groups;
2499
2500         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
2501         if (rc) {
2502                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
2503                 goto err_admin_init;
2504         }
2505
2506         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
2507
2508         return 0;
2509
2510 err_admin_init:
2511         ena_com_delete_host_info(ena_dev);
2512         ena_com_admin_destroy(ena_dev);
2513 err_mmio_read_less:
2514         ena_com_mmio_reg_read_request_destroy(ena_dev);
2515
2516         return rc;
2517 }
2518
2519 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
2520                                                     int io_vectors)
2521 {
2522         struct ena_com_dev *ena_dev = adapter->ena_dev;
2523         struct device *dev = &adapter->pdev->dev;
2524         int rc;
2525
2526         rc = ena_enable_msix(adapter, io_vectors);
2527         if (rc) {
2528                 dev_err(dev, "Can not reserve msix vectors\n");
2529                 return rc;
2530         }
2531
2532         ena_setup_mgmnt_intr(adapter);
2533
2534         rc = ena_request_mgmnt_irq(adapter);
2535         if (rc) {
2536                 dev_err(dev, "Can not setup management interrupts\n");
2537                 goto err_disable_msix;
2538         }
2539
2540         ena_com_set_admin_polling_mode(ena_dev, false);
2541
2542         ena_com_admin_aenq_enable(ena_dev);
2543
2544         return 0;
2545
2546 err_disable_msix:
2547         ena_disable_msix(adapter);
2548
2549         return rc;
2550 }
2551
2552 static void ena_destroy_device(struct ena_adapter *adapter)
2553 {
2554         struct net_device *netdev = adapter->netdev;
2555         struct ena_com_dev *ena_dev = adapter->ena_dev;
2556         bool dev_up;
2557
2558         netif_carrier_off(netdev);
2559
2560         del_timer_sync(&adapter->timer_service);
2561
2562         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2563         adapter->dev_up_before_reset = dev_up;
2564
2565         ena_com_set_admin_running_state(ena_dev, false);
2566
2567         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2568                 ena_down(adapter);
2569
2570         /* Before releasing the ENA resources, a device reset is required.
2571          * (to prevent the device from accessing them).
2572          * In case the reset flag is set and the device is up, ena_down()
2573          * already perform the reset, so it can be skipped.
2574          */
2575         if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
2576                 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2577
2578         ena_free_mgmnt_irq(adapter);
2579
2580         ena_disable_msix(adapter);
2581
2582         ena_com_abort_admin_commands(ena_dev);
2583
2584         ena_com_wait_for_abort_completion(ena_dev);
2585
2586         ena_com_admin_destroy(ena_dev);
2587
2588         ena_com_mmio_reg_read_request_destroy(ena_dev);
2589
2590         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
2591
2592         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2593 }
2594
2595 static int ena_restore_device(struct ena_adapter *adapter)
2596 {
2597         struct ena_com_dev_get_features_ctx get_feat_ctx;
2598         struct ena_com_dev *ena_dev = adapter->ena_dev;
2599         struct pci_dev *pdev = adapter->pdev;
2600         bool wd_state;
2601         int rc;
2602
2603         set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2604         rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
2605         if (rc) {
2606                 dev_err(&pdev->dev, "Can not initialize device\n");
2607                 goto err;
2608         }
2609         adapter->wd_state = wd_state;
2610
2611         rc = ena_device_validate_params(adapter, &get_feat_ctx);
2612         if (rc) {
2613                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
2614                 goto err_device_destroy;
2615         }
2616
2617         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2618         /* Make sure we don't have a race with AENQ Links state handler */
2619         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2620                 netif_carrier_on(adapter->netdev);
2621
2622         rc = ena_enable_msix_and_set_admin_interrupts(adapter,
2623                                                       adapter->num_queues);
2624         if (rc) {
2625                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
2626                 goto err_device_destroy;
2627         }
2628         /* If the interface was up before the reset bring it up */
2629         if (adapter->dev_up_before_reset) {
2630                 rc = ena_up(adapter);
2631                 if (rc) {
2632                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
2633                         goto err_disable_msix;
2634                 }
2635         }
2636
2637         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
2638         dev_err(&pdev->dev, "Device reset completed successfully\n");
2639
2640         return rc;
2641 err_disable_msix:
2642         ena_free_mgmnt_irq(adapter);
2643         ena_disable_msix(adapter);
2644 err_device_destroy:
2645         ena_com_admin_destroy(ena_dev);
2646 err:
2647         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2648         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2649         dev_err(&pdev->dev,
2650                 "Reset attempt failed. Can not reset the device\n");
2651
2652         return rc;
2653 }
2654
2655 static void ena_fw_reset_device(struct work_struct *work)
2656 {
2657         struct ena_adapter *adapter =
2658                 container_of(work, struct ena_adapter, reset_task);
2659         struct pci_dev *pdev = adapter->pdev;
2660
2661         if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2662                 dev_err(&pdev->dev,
2663                         "device reset schedule while reset bit is off\n");
2664                 return;
2665         }
2666         rtnl_lock();
2667         ena_destroy_device(adapter);
2668         ena_restore_device(adapter);
2669         rtnl_unlock();
2670 }
2671
2672 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
2673                                         struct ena_ring *rx_ring)
2674 {
2675         if (likely(rx_ring->first_interrupt))
2676                 return 0;
2677
2678         if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
2679                 return 0;
2680
2681         rx_ring->no_interrupt_event_cnt++;
2682
2683         if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
2684                 netif_err(adapter, rx_err, adapter->netdev,
2685                           "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
2686                           rx_ring->qid);
2687                 adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
2688                 smp_mb__before_atomic();
2689                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2690                 return -EIO;
2691         }
2692
2693         return 0;
2694 }
2695
2696 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
2697                                           struct ena_ring *tx_ring)
2698 {
2699         struct ena_tx_buffer *tx_buf;
2700         unsigned long last_jiffies;
2701         u32 missed_tx = 0;
2702         int i, rc = 0;
2703
2704         for (i = 0; i < tx_ring->ring_size; i++) {
2705                 tx_buf = &tx_ring->tx_buffer_info[i];
2706                 last_jiffies = tx_buf->last_jiffies;
2707
2708                 if (last_jiffies == 0)
2709                         /* no pending Tx at this location */
2710                         continue;
2711
2712                 if (unlikely(!tx_ring->first_interrupt && time_is_before_jiffies(last_jiffies +
2713                              2 * adapter->missing_tx_completion_to))) {
2714                         /* If after graceful period interrupt is still not
2715                          * received, we schedule a reset
2716                          */
2717                         netif_err(adapter, tx_err, adapter->netdev,
2718                                   "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
2719                                   tx_ring->qid);
2720                         adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
2721                         smp_mb__before_atomic();
2722                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2723                         return -EIO;
2724                 }
2725
2726                 if (unlikely(time_is_before_jiffies(last_jiffies +
2727                                 adapter->missing_tx_completion_to))) {
2728                         if (!tx_buf->print_once)
2729                                 netif_notice(adapter, tx_err, adapter->netdev,
2730                                              "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
2731                                              tx_ring->qid, i);
2732
2733                         tx_buf->print_once = 1;
2734                         missed_tx++;
2735                 }
2736         }
2737
2738         if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
2739                 netif_err(adapter, tx_err, adapter->netdev,
2740                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
2741                           missed_tx,
2742                           adapter->missing_tx_completion_threshold);
2743                 adapter->reset_reason =
2744                         ENA_REGS_RESET_MISS_TX_CMPL;
2745                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2746                 rc = -EIO;
2747         }
2748
2749         u64_stats_update_begin(&tx_ring->syncp);
2750         tx_ring->tx_stats.missed_tx = missed_tx;
2751         u64_stats_update_end(&tx_ring->syncp);
2752
2753         return rc;
2754 }
2755
2756 static void check_for_missing_completions(struct ena_adapter *adapter)
2757 {
2758         struct ena_ring *tx_ring;
2759         struct ena_ring *rx_ring;
2760         int i, budget, rc;
2761
2762         /* Make sure the driver doesn't turn the device in other process */
2763         smp_rmb();
2764
2765         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2766                 return;
2767
2768         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2769                 return;
2770
2771         if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
2772                 return;
2773
2774         budget = ENA_MONITORED_TX_QUEUES;
2775
2776         for (i = adapter->last_monitored_tx_qid; i < adapter->num_queues; i++) {
2777                 tx_ring = &adapter->tx_ring[i];
2778                 rx_ring = &adapter->rx_ring[i];
2779
2780                 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
2781                 if (unlikely(rc))
2782                         return;
2783
2784                 rc = check_for_rx_interrupt_queue(adapter, rx_ring);
2785                 if (unlikely(rc))
2786                         return;
2787
2788                 budget--;
2789                 if (!budget)
2790                         break;
2791         }
2792
2793         adapter->last_monitored_tx_qid = i % adapter->num_queues;
2794 }
2795
2796 /* trigger napi schedule after 2 consecutive detections */
2797 #define EMPTY_RX_REFILL 2
2798 /* For the rare case where the device runs out of Rx descriptors and the
2799  * napi handler failed to refill new Rx descriptors (due to a lack of memory
2800  * for example).
2801  * This case will lead to a deadlock:
2802  * The device won't send interrupts since all the new Rx packets will be dropped
2803  * The napi handler won't allocate new Rx descriptors so the device will be
2804  * able to send new packets.
2805  *
2806  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
2807  * It is recommended to have at least 512MB, with a minimum of 128MB for
2808  * constrained environment).
2809  *
2810  * When such a situation is detected - Reschedule napi
2811  */
2812 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
2813 {
2814         struct ena_ring *rx_ring;
2815         int i, refill_required;
2816
2817         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2818                 return;
2819
2820         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2821                 return;
2822
2823         for (i = 0; i < adapter->num_queues; i++) {
2824                 rx_ring = &adapter->rx_ring[i];
2825
2826                 refill_required =
2827                         ena_com_sq_empty_space(rx_ring->ena_com_io_sq);
2828                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
2829                         rx_ring->empty_rx_queue++;
2830
2831                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
2832                                 u64_stats_update_begin(&rx_ring->syncp);
2833                                 rx_ring->rx_stats.empty_rx_ring++;
2834                                 u64_stats_update_end(&rx_ring->syncp);
2835
2836                                 netif_err(adapter, drv, adapter->netdev,
2837                                           "trigger refill for ring %d\n", i);
2838
2839                                 napi_schedule(rx_ring->napi);
2840                                 rx_ring->empty_rx_queue = 0;
2841                         }
2842                 } else {
2843                         rx_ring->empty_rx_queue = 0;
2844                 }
2845         }
2846 }
2847
2848 /* Check for keep alive expiration */
2849 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
2850 {
2851         unsigned long keep_alive_expired;
2852
2853         if (!adapter->wd_state)
2854                 return;
2855
2856         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2857                 return;
2858
2859         keep_alive_expired = round_jiffies(adapter->last_keep_alive_jiffies +
2860                                            adapter->keep_alive_timeout);
2861         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
2862                 netif_err(adapter, drv, adapter->netdev,
2863                           "Keep alive watchdog timeout.\n");
2864                 u64_stats_update_begin(&adapter->syncp);
2865                 adapter->dev_stats.wd_expired++;
2866                 u64_stats_update_end(&adapter->syncp);
2867                 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
2868                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2869         }
2870 }
2871
2872 static void check_for_admin_com_state(struct ena_adapter *adapter)
2873 {
2874         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
2875                 netif_err(adapter, drv, adapter->netdev,
2876                           "ENA admin queue is not in running state!\n");
2877                 u64_stats_update_begin(&adapter->syncp);
2878                 adapter->dev_stats.admin_q_pause++;
2879                 u64_stats_update_end(&adapter->syncp);
2880                 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
2881                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2882         }
2883 }
2884
2885 static void ena_update_hints(struct ena_adapter *adapter,
2886                              struct ena_admin_ena_hw_hints *hints)
2887 {
2888         struct net_device *netdev = adapter->netdev;
2889
2890         if (hints->admin_completion_tx_timeout)
2891                 adapter->ena_dev->admin_queue.completion_timeout =
2892                         hints->admin_completion_tx_timeout * 1000;
2893
2894         if (hints->mmio_read_timeout)
2895                 /* convert to usec */
2896                 adapter->ena_dev->mmio_read.reg_read_to =
2897                         hints->mmio_read_timeout * 1000;
2898
2899         if (hints->missed_tx_completion_count_threshold_to_reset)
2900                 adapter->missing_tx_completion_threshold =
2901                         hints->missed_tx_completion_count_threshold_to_reset;
2902
2903         if (hints->missing_tx_completion_timeout) {
2904                 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2905                         adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
2906                 else
2907                         adapter->missing_tx_completion_to =
2908                                 msecs_to_jiffies(hints->missing_tx_completion_timeout);
2909         }
2910
2911         if (hints->netdev_wd_timeout)
2912                 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
2913
2914         if (hints->driver_watchdog_timeout) {
2915                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2916                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2917                 else
2918                         adapter->keep_alive_timeout =
2919                                 msecs_to_jiffies(hints->driver_watchdog_timeout);
2920         }
2921 }
2922
2923 static void ena_update_host_info(struct ena_admin_host_info *host_info,
2924                                  struct net_device *netdev)
2925 {
2926         host_info->supported_network_features[0] =
2927                 netdev->features & GENMASK_ULL(31, 0);
2928         host_info->supported_network_features[1] =
2929                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
2930 }
2931
2932 static void ena_timer_service(struct timer_list *t)
2933 {
2934         struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
2935         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
2936         struct ena_admin_host_info *host_info =
2937                 adapter->ena_dev->host_attr.host_info;
2938
2939         check_for_missing_keep_alive(adapter);
2940
2941         check_for_admin_com_state(adapter);
2942
2943         check_for_missing_completions(adapter);
2944
2945         check_for_empty_rx_ring(adapter);
2946
2947         if (debug_area)
2948                 ena_dump_stats_to_buf(adapter, debug_area);
2949
2950         if (host_info)
2951                 ena_update_host_info(host_info, adapter->netdev);
2952
2953         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2954                 netif_err(adapter, drv, adapter->netdev,
2955                           "Trigger reset is on\n");
2956                 ena_dump_stats_to_dmesg(adapter);
2957                 queue_work(ena_wq, &adapter->reset_task);
2958                 return;
2959         }
2960
2961         /* Reset the timer */
2962         mod_timer(&adapter->timer_service, jiffies + HZ);
2963 }
2964
2965 static int ena_calc_io_queue_num(struct pci_dev *pdev,
2966                                  struct ena_com_dev *ena_dev,
2967                                  struct ena_com_dev_get_features_ctx *get_feat_ctx)
2968 {
2969         int io_sq_num, io_queue_num;
2970
2971         /* In case of LLQ use the llq number in the get feature cmd */
2972         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2973                 io_sq_num = get_feat_ctx->max_queues.max_llq_num;
2974
2975                 if (io_sq_num == 0) {
2976                         dev_err(&pdev->dev,
2977                                 "Trying to use LLQ but llq_num is 0. Fall back into regular queues\n");
2978
2979                         ena_dev->tx_mem_queue_type =
2980                                 ENA_ADMIN_PLACEMENT_POLICY_HOST;
2981                         io_sq_num = get_feat_ctx->max_queues.max_sq_num;
2982                 }
2983         } else {
2984                 io_sq_num = get_feat_ctx->max_queues.max_sq_num;
2985         }
2986
2987         io_queue_num = min_t(int, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
2988         io_queue_num = min_t(int, io_queue_num, io_sq_num);
2989         io_queue_num = min_t(int, io_queue_num,
2990                              get_feat_ctx->max_queues.max_cq_num);
2991         /* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
2992         io_queue_num = min_t(int, io_queue_num, pci_msix_vec_count(pdev) - 1);
2993         if (unlikely(!io_queue_num)) {
2994                 dev_err(&pdev->dev, "The device doesn't have io queues\n");
2995                 return -EFAULT;
2996         }
2997
2998         return io_queue_num;
2999 }
3000
3001 static void ena_set_push_mode(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
3002                               struct ena_com_dev_get_features_ctx *get_feat_ctx)
3003 {
3004         bool has_mem_bar;
3005
3006         has_mem_bar = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(ENA_MEM_BAR);
3007
3008         /* Enable push mode if device supports LLQ */
3009         if (has_mem_bar && (get_feat_ctx->max_queues.max_llq_num > 0))
3010                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
3011         else
3012                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3013 }
3014
3015 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
3016                                  struct net_device *netdev)
3017 {
3018         netdev_features_t dev_features = 0;
3019
3020         /* Set offload features */
3021         if (feat->offload.tx &
3022                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
3023                 dev_features |= NETIF_F_IP_CSUM;
3024
3025         if (feat->offload.tx &
3026                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
3027                 dev_features |= NETIF_F_IPV6_CSUM;
3028
3029         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
3030                 dev_features |= NETIF_F_TSO;
3031
3032         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
3033                 dev_features |= NETIF_F_TSO6;
3034
3035         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
3036                 dev_features |= NETIF_F_TSO_ECN;
3037
3038         if (feat->offload.rx_supported &
3039                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
3040                 dev_features |= NETIF_F_RXCSUM;
3041
3042         if (feat->offload.rx_supported &
3043                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
3044                 dev_features |= NETIF_F_RXCSUM;
3045
3046         netdev->features =
3047                 dev_features |
3048                 NETIF_F_SG |
3049                 NETIF_F_RXHASH |
3050                 NETIF_F_HIGHDMA;
3051
3052         netdev->hw_features |= netdev->features;
3053         netdev->vlan_features |= netdev->features;
3054 }
3055
3056 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
3057                                      struct ena_com_dev_get_features_ctx *feat)
3058 {
3059         struct net_device *netdev = adapter->netdev;
3060
3061         /* Copy mac address */
3062         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
3063                 eth_hw_addr_random(netdev);
3064                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
3065         } else {
3066                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
3067                 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
3068         }
3069
3070         /* Set offload features */
3071         ena_set_dev_offloads(feat, netdev);
3072
3073         adapter->max_mtu = feat->dev_attr.max_mtu;
3074         netdev->max_mtu = adapter->max_mtu;
3075         netdev->min_mtu = ENA_MIN_MTU;
3076 }
3077
3078 static int ena_rss_init_default(struct ena_adapter *adapter)
3079 {
3080         struct ena_com_dev *ena_dev = adapter->ena_dev;
3081         struct device *dev = &adapter->pdev->dev;
3082         int rc, i;
3083         u32 val;
3084
3085         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3086         if (unlikely(rc)) {
3087                 dev_err(dev, "Cannot init indirect table\n");
3088                 goto err_rss_init;
3089         }
3090
3091         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3092                 val = ethtool_rxfh_indir_default(i, adapter->num_queues);
3093                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3094                                                        ENA_IO_RXQ_IDX(val));
3095                 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3096                         dev_err(dev, "Cannot fill indirect table\n");
3097                         goto err_fill_indir;
3098                 }
3099         }
3100
3101         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
3102                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
3103         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3104                 dev_err(dev, "Cannot fill hash function\n");
3105                 goto err_fill_indir;
3106         }
3107
3108         rc = ena_com_set_default_hash_ctrl(ena_dev);
3109         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3110                 dev_err(dev, "Cannot fill hash control\n");
3111                 goto err_fill_indir;
3112         }
3113
3114         return 0;
3115
3116 err_fill_indir:
3117         ena_com_rss_destroy(ena_dev);
3118 err_rss_init:
3119
3120         return rc;
3121 }
3122
3123 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3124 {
3125         int release_bars;
3126
3127         if (ena_dev->mem_bar)
3128                 devm_iounmap(&pdev->dev, ena_dev->mem_bar);
3129
3130         if (ena_dev->reg_bar)
3131                 devm_iounmap(&pdev->dev, ena_dev->reg_bar);
3132
3133         release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
3134         pci_release_selected_regions(pdev, release_bars);
3135 }
3136
3137 static int ena_calc_queue_size(struct pci_dev *pdev,
3138                                struct ena_com_dev *ena_dev,
3139                                u16 *max_tx_sgl_size,
3140                                u16 *max_rx_sgl_size,
3141                                struct ena_com_dev_get_features_ctx *get_feat_ctx)
3142 {
3143         u32 queue_size = ENA_DEFAULT_RING_SIZE;
3144
3145         queue_size = min_t(u32, queue_size,
3146                            get_feat_ctx->max_queues.max_cq_depth);
3147         queue_size = min_t(u32, queue_size,
3148                            get_feat_ctx->max_queues.max_sq_depth);
3149
3150         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3151                 queue_size = min_t(u32, queue_size,
3152                                    get_feat_ctx->max_queues.max_llq_depth);
3153
3154         queue_size = rounddown_pow_of_two(queue_size);
3155
3156         if (unlikely(!queue_size)) {
3157                 dev_err(&pdev->dev, "Invalid queue size\n");
3158                 return -EFAULT;
3159         }
3160
3161         *max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3162                                  get_feat_ctx->max_queues.max_packet_tx_descs);
3163         *max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3164                                  get_feat_ctx->max_queues.max_packet_rx_descs);
3165
3166         return queue_size;
3167 }
3168
3169 /* ena_probe - Device Initialization Routine
3170  * @pdev: PCI device information struct
3171  * @ent: entry in ena_pci_tbl
3172  *
3173  * Returns 0 on success, negative on failure
3174  *
3175  * ena_probe initializes an adapter identified by a pci_dev structure.
3176  * The OS initialization, configuring of the adapter private structure,
3177  * and a hardware reset occur.
3178  */
3179 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3180 {
3181         struct ena_com_dev_get_features_ctx get_feat_ctx;
3182         static int version_printed;
3183         struct net_device *netdev;
3184         struct ena_adapter *adapter;
3185         struct ena_com_dev *ena_dev = NULL;
3186         static int adapters_found;
3187         int io_queue_num, bars, rc;
3188         int queue_size;
3189         u16 tx_sgl_size = 0;
3190         u16 rx_sgl_size = 0;
3191         bool wd_state;
3192
3193         dev_dbg(&pdev->dev, "%s\n", __func__);
3194
3195         if (version_printed++ == 0)
3196                 dev_info(&pdev->dev, "%s", version);
3197
3198         rc = pci_enable_device_mem(pdev);
3199         if (rc) {
3200                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
3201                 return rc;
3202         }
3203
3204         pci_set_master(pdev);
3205
3206         ena_dev = vzalloc(sizeof(*ena_dev));
3207         if (!ena_dev) {
3208                 rc = -ENOMEM;
3209                 goto err_disable_device;
3210         }
3211
3212         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
3213         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
3214         if (rc) {
3215                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
3216                         rc);
3217                 goto err_free_ena_dev;
3218         }
3219
3220         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
3221                                         pci_resource_start(pdev, ENA_REG_BAR),
3222                                         pci_resource_len(pdev, ENA_REG_BAR));
3223         if (!ena_dev->reg_bar) {
3224                 dev_err(&pdev->dev, "failed to remap regs bar\n");
3225                 rc = -EFAULT;
3226                 goto err_free_region;
3227         }
3228
3229         ena_dev->dmadev = &pdev->dev;
3230
3231         rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
3232         if (rc) {
3233                 dev_err(&pdev->dev, "ena device init failed\n");
3234                 if (rc == -ETIME)
3235                         rc = -EPROBE_DEFER;
3236                 goto err_free_region;
3237         }
3238
3239         ena_set_push_mode(pdev, ena_dev, &get_feat_ctx);
3240
3241         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3242                 ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3243                                                    pci_resource_start(pdev, ENA_MEM_BAR),
3244                                                    pci_resource_len(pdev, ENA_MEM_BAR));
3245                 if (!ena_dev->mem_bar) {
3246                         rc = -EFAULT;
3247                         goto err_device_destroy;
3248                 }
3249         }
3250
3251         /* initial Tx interrupt delay, Assumes 1 usec granularity.
3252         * Updated during device initialization with the real granularity
3253         */
3254         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
3255         io_queue_num = ena_calc_io_queue_num(pdev, ena_dev, &get_feat_ctx);
3256         queue_size = ena_calc_queue_size(pdev, ena_dev, &tx_sgl_size,
3257                                          &rx_sgl_size, &get_feat_ctx);
3258         if ((queue_size <= 0) || (io_queue_num <= 0)) {
3259                 rc = -EFAULT;
3260                 goto err_device_destroy;
3261         }
3262
3263         dev_info(&pdev->dev, "creating %d io queues. queue size: %d\n",
3264                  io_queue_num, queue_size);
3265
3266         /* dev zeroed in init_etherdev */
3267         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), io_queue_num);
3268         if (!netdev) {
3269                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
3270                 rc = -ENOMEM;
3271                 goto err_device_destroy;
3272         }
3273
3274         SET_NETDEV_DEV(netdev, &pdev->dev);
3275
3276         adapter = netdev_priv(netdev);
3277         pci_set_drvdata(pdev, adapter);
3278
3279         adapter->ena_dev = ena_dev;
3280         adapter->netdev = netdev;
3281         adapter->pdev = pdev;
3282
3283         ena_set_conf_feat_params(adapter, &get_feat_ctx);
3284
3285         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3286         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3287
3288         adapter->tx_ring_size = queue_size;
3289         adapter->rx_ring_size = queue_size;
3290
3291         adapter->max_tx_sgl_size = tx_sgl_size;
3292         adapter->max_rx_sgl_size = rx_sgl_size;
3293
3294         adapter->num_queues = io_queue_num;
3295         adapter->last_monitored_tx_qid = 0;
3296
3297         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
3298         adapter->wd_state = wd_state;
3299
3300         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
3301
3302         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
3303         if (rc) {
3304                 dev_err(&pdev->dev,
3305                         "Failed to query interrupt moderation feature\n");
3306                 goto err_netdev_destroy;
3307         }
3308         ena_init_io_rings(adapter);
3309
3310         netdev->netdev_ops = &ena_netdev_ops;
3311         netdev->watchdog_timeo = TX_TIMEOUT;
3312         ena_set_ethtool_ops(netdev);
3313
3314         netdev->priv_flags |= IFF_UNICAST_FLT;
3315
3316         u64_stats_init(&adapter->syncp);
3317
3318         rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3319         if (rc) {
3320                 dev_err(&pdev->dev,
3321                         "Failed to enable and set the admin interrupts\n");
3322                 goto err_worker_destroy;
3323         }
3324         rc = ena_rss_init_default(adapter);
3325         if (rc && (rc != -EOPNOTSUPP)) {
3326                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
3327                 goto err_free_msix;
3328         }
3329
3330         ena_config_debug_area(adapter);
3331
3332         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
3333
3334         netif_carrier_off(netdev);
3335
3336         rc = register_netdev(netdev);
3337         if (rc) {
3338                 dev_err(&pdev->dev, "Cannot register net device\n");
3339                 goto err_rss;
3340         }
3341
3342         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
3343
3344         adapter->last_keep_alive_jiffies = jiffies;
3345         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
3346         adapter->missing_tx_completion_to = TX_TIMEOUT;
3347         adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
3348
3349         ena_update_hints(adapter, &get_feat_ctx.hw_hints);
3350
3351         timer_setup(&adapter->timer_service, ena_timer_service, 0);
3352         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3353
3354         dev_info(&pdev->dev, "%s found at mem %lx, mac addr %pM Queues %d\n",
3355                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
3356                  netdev->dev_addr, io_queue_num);
3357
3358         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3359
3360         adapters_found++;
3361
3362         return 0;
3363
3364 err_rss:
3365         ena_com_delete_debug_area(ena_dev);
3366         ena_com_rss_destroy(ena_dev);
3367 err_free_msix:
3368         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
3369         ena_free_mgmnt_irq(adapter);
3370         ena_disable_msix(adapter);
3371 err_worker_destroy:
3372         ena_com_destroy_interrupt_moderation(ena_dev);
3373         del_timer(&adapter->timer_service);
3374 err_netdev_destroy:
3375         free_netdev(netdev);
3376 err_device_destroy:
3377         ena_com_delete_host_info(ena_dev);
3378         ena_com_admin_destroy(ena_dev);
3379 err_free_region:
3380         ena_release_bars(ena_dev, pdev);
3381 err_free_ena_dev:
3382         vfree(ena_dev);
3383 err_disable_device:
3384         pci_disable_device(pdev);
3385         return rc;
3386 }
3387
3388 /*****************************************************************************/
3389 static int ena_sriov_configure(struct pci_dev *dev, int numvfs)
3390 {
3391         int rc;
3392
3393         if (numvfs > 0) {
3394                 rc = pci_enable_sriov(dev, numvfs);
3395                 if (rc != 0) {
3396                         dev_err(&dev->dev,
3397                                 "pci_enable_sriov failed to enable: %d vfs with the error: %d\n",
3398                                 numvfs, rc);
3399                         return rc;
3400                 }
3401
3402                 return numvfs;
3403         }
3404
3405         if (numvfs == 0) {
3406                 pci_disable_sriov(dev);
3407                 return 0;
3408         }
3409
3410         return -EINVAL;
3411 }
3412
3413 /*****************************************************************************/
3414 /*****************************************************************************/
3415
3416 /* ena_remove - Device Removal Routine
3417  * @pdev: PCI device information struct
3418  *
3419  * ena_remove is called by the PCI subsystem to alert the driver
3420  * that it should release a PCI device.
3421  */
3422 static void ena_remove(struct pci_dev *pdev)
3423 {
3424         struct ena_adapter *adapter = pci_get_drvdata(pdev);
3425         struct ena_com_dev *ena_dev;
3426         struct net_device *netdev;
3427
3428         ena_dev = adapter->ena_dev;
3429         netdev = adapter->netdev;
3430
3431 #ifdef CONFIG_RFS_ACCEL
3432         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
3433                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
3434                 netdev->rx_cpu_rmap = NULL;
3435         }
3436 #endif /* CONFIG_RFS_ACCEL */
3437
3438         unregister_netdev(netdev);
3439         del_timer_sync(&adapter->timer_service);
3440
3441         cancel_work_sync(&adapter->reset_task);
3442
3443         /* Reset the device only if the device is running. */
3444         if (test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3445                 ena_com_dev_reset(ena_dev, adapter->reset_reason);
3446
3447         ena_free_mgmnt_irq(adapter);
3448
3449         ena_disable_msix(adapter);
3450
3451         free_netdev(netdev);
3452
3453         ena_com_mmio_reg_read_request_destroy(ena_dev);
3454
3455         ena_com_abort_admin_commands(ena_dev);
3456
3457         ena_com_wait_for_abort_completion(ena_dev);
3458
3459         ena_com_admin_destroy(ena_dev);
3460
3461         ena_com_rss_destroy(ena_dev);
3462
3463         ena_com_delete_debug_area(ena_dev);
3464
3465         ena_com_delete_host_info(ena_dev);
3466
3467         ena_release_bars(ena_dev, pdev);
3468
3469         pci_disable_device(pdev);
3470
3471         ena_com_destroy_interrupt_moderation(ena_dev);
3472
3473         vfree(ena_dev);
3474 }
3475
3476 #ifdef CONFIG_PM
3477 /* ena_suspend - PM suspend callback
3478  * @pdev: PCI device information struct
3479  * @state:power state
3480  */
3481 static int ena_suspend(struct pci_dev *pdev,  pm_message_t state)
3482 {
3483         struct ena_adapter *adapter = pci_get_drvdata(pdev);
3484
3485         u64_stats_update_begin(&adapter->syncp);
3486         adapter->dev_stats.suspend++;
3487         u64_stats_update_end(&adapter->syncp);
3488
3489         rtnl_lock();
3490         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3491                 dev_err(&pdev->dev,
3492                         "ignoring device reset request as the device is being suspended\n");
3493                 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3494         }
3495         ena_destroy_device(adapter);
3496         rtnl_unlock();
3497         return 0;
3498 }
3499
3500 /* ena_resume - PM resume callback
3501  * @pdev: PCI device information struct
3502  *
3503  */
3504 static int ena_resume(struct pci_dev *pdev)
3505 {
3506         struct ena_adapter *adapter = pci_get_drvdata(pdev);
3507         int rc;
3508
3509         u64_stats_update_begin(&adapter->syncp);
3510         adapter->dev_stats.resume++;
3511         u64_stats_update_end(&adapter->syncp);
3512
3513         rtnl_lock();
3514         rc = ena_restore_device(adapter);
3515         rtnl_unlock();
3516         return rc;
3517 }
3518 #endif
3519
3520 static struct pci_driver ena_pci_driver = {
3521         .name           = DRV_MODULE_NAME,
3522         .id_table       = ena_pci_tbl,
3523         .probe          = ena_probe,
3524         .remove         = ena_remove,
3525 #ifdef CONFIG_PM
3526         .suspend    = ena_suspend,
3527         .resume     = ena_resume,
3528 #endif
3529         .sriov_configure = ena_sriov_configure,
3530 };
3531
3532 static int __init ena_init(void)
3533 {
3534         pr_info("%s", version);
3535
3536         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
3537         if (!ena_wq) {
3538                 pr_err("Failed to create workqueue\n");
3539                 return -ENOMEM;
3540         }
3541
3542         return pci_register_driver(&ena_pci_driver);
3543 }
3544
3545 static void __exit ena_cleanup(void)
3546 {
3547         pci_unregister_driver(&ena_pci_driver);
3548
3549         if (ena_wq) {
3550                 destroy_workqueue(ena_wq);
3551                 ena_wq = NULL;
3552         }
3553 }
3554
3555 /******************************************************************************
3556  ******************************** AENQ Handlers *******************************
3557  *****************************************************************************/
3558 /* ena_update_on_link_change:
3559  * Notify the network interface about the change in link status
3560  */
3561 static void ena_update_on_link_change(void *adapter_data,
3562                                       struct ena_admin_aenq_entry *aenq_e)
3563 {
3564         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3565         struct ena_admin_aenq_link_change_desc *aenq_desc =
3566                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
3567         int status = aenq_desc->flags &
3568                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3569
3570         if (status) {
3571                 netdev_dbg(adapter->netdev, "%s\n", __func__);
3572                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3573                 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
3574                         netif_carrier_on(adapter->netdev);
3575         } else {
3576                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3577                 netif_carrier_off(adapter->netdev);
3578         }
3579 }
3580
3581 static void ena_keep_alive_wd(void *adapter_data,
3582                               struct ena_admin_aenq_entry *aenq_e)
3583 {
3584         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3585         struct ena_admin_aenq_keep_alive_desc *desc;
3586         u64 rx_drops;
3587
3588         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
3589         adapter->last_keep_alive_jiffies = jiffies;
3590
3591         rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
3592
3593         u64_stats_update_begin(&adapter->syncp);
3594         adapter->dev_stats.rx_drops = rx_drops;
3595         u64_stats_update_end(&adapter->syncp);
3596 }
3597
3598 static void ena_notification(void *adapter_data,
3599                              struct ena_admin_aenq_entry *aenq_e)
3600 {
3601         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3602         struct ena_admin_ena_hw_hints *hints;
3603
3604         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
3605              "Invalid group(%x) expected %x\n",
3606              aenq_e->aenq_common_desc.group,
3607              ENA_ADMIN_NOTIFICATION);
3608
3609         switch (aenq_e->aenq_common_desc.syndrom) {
3610         case ENA_ADMIN_UPDATE_HINTS:
3611                 hints = (struct ena_admin_ena_hw_hints *)
3612                         (&aenq_e->inline_data_w4);
3613                 ena_update_hints(adapter, hints);
3614                 break;
3615         default:
3616                 netif_err(adapter, drv, adapter->netdev,
3617                           "Invalid aenq notification link state %d\n",
3618                           aenq_e->aenq_common_desc.syndrom);
3619         }
3620 }
3621
3622 /* This handler will called for unknown event group or unimplemented handlers*/
3623 static void unimplemented_aenq_handler(void *data,
3624                                        struct ena_admin_aenq_entry *aenq_e)
3625 {
3626         struct ena_adapter *adapter = (struct ena_adapter *)data;
3627
3628         netif_err(adapter, drv, adapter->netdev,
3629                   "Unknown event was received or event with unimplemented handler\n");
3630 }
3631
3632 static struct ena_aenq_handlers aenq_handlers = {
3633         .handlers = {
3634                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3635                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
3636                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3637         },
3638         .unimplemented_handler = unimplemented_aenq_handler
3639 };
3640
3641 module_init(ena_init);
3642 module_exit(ena_cleanup);