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