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