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

   1 /*
   2  * Copyright 2015 Amazon.com, Inc. or its affiliates.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  34
  35 #ifdef CONFIG_RFS_ACCEL
  36 #include <linux/cpu_rmap.h>
  37 #endif /* CONFIG_RFS_ACCEL */
  38 #include <linux/ethtool.h>
  39 #include <linux/if_vlan.h>
  40 #include <linux/kernel.h>
  41 #include <linux/module.h>
  42 #include <linux/moduleparam.h>
  43 #include <linux/numa.h>
  44 #include <linux/pci.h>
  45 #include <linux/utsname.h>
  46 #include <linux/version.h>
  47 #include <linux/vmalloc.h>
  48 #include <net/ip.h>
  49
  50 #include "ena_netdev.h"
  51 #include "ena_pci_id_tbl.h"
  52
  53 static char version[] = DEVICE_NAME " v" DRV_MODULE_VERSION "\n";
  54
  55 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
  56 MODULE_DESCRIPTION(DEVICE_NAME);
  57 MODULE_LICENSE("GPL");
  58 MODULE_VERSION(DRV_MODULE_VERSION);
  59
  60 /* Time in jiffies before concluding the transmitter is hung. */
  61 #define TX_TIMEOUT  (5 * HZ)
  62
  63 #define ENA_NAPI_BUDGET 64
  64
  65 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
  66                 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
  67 static int debug = -1;
  68 module_param(debug, int, 0);
  69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  70
  71 static struct ena_aenq_handlers aenq_handlers;
  72
  73 static struct workqueue_struct *ena_wq;
  74
  75 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
  76
  77 static int ena_rss_init_default(struct ena_adapter *adapter);
  78
  79 static void ena_tx_timeout(struct net_device *dev)
  80 {
  81         struct ena_adapter *adapter = netdev_priv(dev);
  82
  83         /* Change the state of the device to trigger reset
  84          * Check that we are not in the middle or a trigger already
  85          */
  86
  87         if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
  88                 return;
  89
  90         u64_stats_update_begin(&adapter->syncp);
  91         adapter->dev_stats.tx_timeout++;
  92         u64_stats_update_end(&adapter->syncp);
  93
  94         netif_err(adapter, tx_err, dev, "Transmit time out\n");
  95 }
  96
  97 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
  98 {
  99         int i;
 100
 101         for (i = 0; i < adapter->num_queues; i++)
 102                 adapter->rx_ring[i].mtu = mtu;
 103 }
 104
 105 static int ena_change_mtu(struct net_device *dev, int new_mtu)
 106 {
 107         struct ena_adapter *adapter = netdev_priv(dev);
 108         int ret;
 109
 110         ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
 111         if (!ret) {
 112                 netif_dbg(adapter, drv, dev, "set MTU to %d\n", new_mtu);
 113                 update_rx_ring_mtu(adapter, new_mtu);
 114                 dev->mtu = new_mtu;
 115         } else {
 116                 netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
 117                           new_mtu);
 118         }
 119
 120         return ret;
 121 }
 122
 123 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
 124 {
 125 #ifdef CONFIG_RFS_ACCEL
 126         u32 i;
 127         int rc;
 128
 129         adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_queues);
 130         if (!adapter->netdev->rx_cpu_rmap)
 131                 return -ENOMEM;
 132         for (i = 0; i < adapter->num_queues; i++) {
 133                 int irq_idx = ENA_IO_IRQ_IDX(i);
 134
 135                 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
 136                                       pci_irq_vector(adapter->pdev, irq_idx));
 137                 if (rc) {
 138                         free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
 139                         adapter->netdev->rx_cpu_rmap = NULL;
 140                         return rc;
 141                 }
 142         }
 143 #endif /* CONFIG_RFS_ACCEL */
 144         return 0;
 145 }
 146
 147 static void ena_init_io_rings_common(struct ena_adapter *adapter,
 148                                      struct ena_ring *ring, u16 qid)
 149 {
 150         ring->qid = qid;
 151         ring->pdev = adapter->pdev;
 152         ring->dev = &adapter->pdev->dev;
 153         ring->netdev = adapter->netdev;
 154         ring->napi = &adapter->ena_napi[qid].napi;
 155         ring->adapter = adapter;
 156         ring->ena_dev = adapter->ena_dev;
 157         ring->per_napi_packets = 0;
 158         ring->per_napi_bytes = 0;
 159         ring->cpu = 0;
 160         u64_stats_init(&ring->syncp);
 161 }
 162
 163 static void ena_init_io_rings(struct ena_adapter *adapter)
 164 {
 165         struct ena_com_dev *ena_dev;
 166         struct ena_ring *txr, *rxr;
 167         int i;
 168
 169         ena_dev = adapter->ena_dev;
 170
 171         for (i = 0; i < adapter->num_queues; i++) {
 172                 txr = &adapter->tx_ring[i];
 173                 rxr = &adapter->rx_ring[i];
 174
 175                 /* TX/RX common ring state */
 176                 ena_init_io_rings_common(adapter, txr, i);
 177                 ena_init_io_rings_common(adapter, rxr, i);
 178
 179                 /* TX specific ring state */
 180                 txr->ring_size = adapter->tx_ring_size;
 181                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
 182                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
 183                 txr->sgl_size = adapter->max_tx_sgl_size;
 184                 txr->smoothed_interval =
 185                         ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
 186
 187                 /* RX specific ring state */
 188                 rxr->ring_size = adapter->rx_ring_size;
 189                 rxr->rx_copybreak = adapter->rx_copybreak;
 190                 rxr->sgl_size = adapter->max_rx_sgl_size;
 191                 rxr->smoothed_interval =
 192                         ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
 193                 rxr->empty_rx_queue = 0;
 194         }
 195 }
 196
 197 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
 198  * @adapter: network interface device structure
 199  * @qid: queue index
 200  *
 201  * Return 0 on success, negative on failure
 202  */
 203 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
 204 {
 205         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 206         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 207         int size, i, node;
 208
 209         if (tx_ring->tx_buffer_info) {
 210                 netif_err(adapter, ifup,
 211                           adapter->netdev, "tx_buffer_info info is not NULL");
 212                 return -EEXIST;
 213         }
 214
 215         size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
 216         node = cpu_to_node(ena_irq->cpu);
 217
 218         tx_ring->tx_buffer_info = vzalloc_node(size, node);
 219         if (!tx_ring->tx_buffer_info) {
 220                 tx_ring->tx_buffer_info = vzalloc(size);
 221                 if (!tx_ring->tx_buffer_info)
 222                         return -ENOMEM;
 223         }
 224
 225         size = sizeof(u16) * tx_ring->ring_size;
 226         tx_ring->free_tx_ids = vzalloc_node(size, node);
 227         if (!tx_ring->free_tx_ids) {
 228                 tx_ring->free_tx_ids = vzalloc(size);
 229                 if (!tx_ring->free_tx_ids) {
 230                         vfree(tx_ring->tx_buffer_info);
 231                         return -ENOMEM;
 232                 }
 233         }
 234
 235         /* Req id ring for TX out of order completions */
 236         for (i = 0; i < tx_ring->ring_size; i++)
 237                 tx_ring->free_tx_ids[i] = i;
 238
 239         /* Reset tx statistics */
 240         memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
 241
 242         tx_ring->next_to_use = 0;
 243         tx_ring->next_to_clean = 0;
 244         tx_ring->cpu = ena_irq->cpu;
 245         return 0;
 246 }
 247
 248 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
 249  * @adapter: network interface device structure
 250  * @qid: queue index
 251  *
 252  * Free all transmit software resources
 253  */
 254 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
 255 {
 256         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 257
 258         vfree(tx_ring->tx_buffer_info);
 259         tx_ring->tx_buffer_info = NULL;
 260
 261         vfree(tx_ring->free_tx_ids);
 262         tx_ring->free_tx_ids = NULL;
 263 }
 264
 265 /* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
 266  * @adapter: private structure
 267  *
 268  * Return 0 on success, negative on failure
 269  */
 270 static int ena_setup_all_tx_resources(struct ena_adapter *adapter)
 271 {
 272         int i, rc = 0;
 273
 274         for (i = 0; i < adapter->num_queues; i++) {
 275                 rc = ena_setup_tx_resources(adapter, i);
 276                 if (rc)
 277                         goto err_setup_tx;
 278         }
 279
 280         return 0;
 281
 282 err_setup_tx:
 283
 284         netif_err(adapter, ifup, adapter->netdev,
 285                   "Tx queue %d: allocation failed\n", i);
 286
 287         /* rewind the index freeing the rings as we go */
 288         while (i--)
 289                 ena_free_tx_resources(adapter, i);
 290         return rc;
 291 }
 292
 293 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
 294  * @adapter: board private structure
 295  *
 296  * Free all transmit software resources
 297  */
 298 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
 299 {
 300         int i;
 301
 302         for (i = 0; i < adapter->num_queues; i++)
 303                 ena_free_tx_resources(adapter, i);
 304 }
 305
 306 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
 307  * @adapter: network interface device structure
 308  * @qid: queue index
 309  *
 310  * Returns 0 on success, negative on failure
 311  */
 312 static int ena_setup_rx_resources(struct ena_adapter *adapter,
 313                                   u32 qid)
 314 {
 315         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 316         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 317         int size, node;
 318
 319         if (rx_ring->rx_buffer_info) {
 320                 netif_err(adapter, ifup, adapter->netdev,
 321                           "rx_buffer_info is not NULL");
 322                 return -EEXIST;
 323         }
 324
 325         /* alloc extra element so in rx path
 326          * we can always prefetch rx_info + 1
 327          */
 328         size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
 329         node = cpu_to_node(ena_irq->cpu);
 330
 331         rx_ring->rx_buffer_info = vzalloc_node(size, node);
 332         if (!rx_ring->rx_buffer_info) {
 333                 rx_ring->rx_buffer_info = vzalloc(size);
 334                 if (!rx_ring->rx_buffer_info)
 335                         return -ENOMEM;
 336         }
 337
 338         /* Reset rx statistics */
 339         memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
 340
 341         rx_ring->next_to_clean = 0;
 342         rx_ring->next_to_use = 0;
 343         rx_ring->cpu = ena_irq->cpu;
 344
 345         return 0;
 346 }
 347
 348 /* ena_free_rx_resources - Free I/O Rx Resources
 349  * @adapter: network interface device structure
 350  * @qid: queue index
 351  *
 352  * Free all receive software resources
 353  */
 354 static void ena_free_rx_resources(struct ena_adapter *adapter,
 355                                   u32 qid)
 356 {
 357         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 358
 359         vfree(rx_ring->rx_buffer_info);
 360         rx_ring->rx_buffer_info = NULL;
 361 }
 362
 363 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
 364  * @adapter: board private structure
 365  *
 366  * Return 0 on success, negative on failure
 367  */
 368 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
 369 {
 370         int i, rc = 0;
 371
 372         for (i = 0; i < adapter->num_queues; i++) {
 373                 rc = ena_setup_rx_resources(adapter, i);
 374                 if (rc)
 375                         goto err_setup_rx;
 376         }
 377
 378         return 0;
 379
 380 err_setup_rx:
 381
 382         netif_err(adapter, ifup, adapter->netdev,
 383                   "Rx queue %d: allocation failed\n", i);
 384
 385         /* rewind the index freeing the rings as we go */
 386         while (i--)
 387                 ena_free_rx_resources(adapter, i);
 388         return rc;
 389 }
 390
 391 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
 392  * @adapter: board private structure
 393  *
 394  * Free all receive software resources
 395  */
 396 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
 397 {
 398         int i;
 399
 400         for (i = 0; i < adapter->num_queues; i++)
 401                 ena_free_rx_resources(adapter, i);
 402 }
 403
 404 static inline int ena_alloc_rx_page(struct ena_ring *rx_ring,
 405                                     struct ena_rx_buffer *rx_info, gfp_t gfp)
 406 {
 407         struct ena_com_buf *ena_buf;
 408         struct page *page;
 409         dma_addr_t dma;
 410
 411         /* if previous allocated page is not used */
 412         if (unlikely(rx_info->page))
 413                 return 0;
 414
 415         page = alloc_page(gfp);
 416         if (unlikely(!page)) {
 417                 u64_stats_update_begin(&rx_ring->syncp);
 418                 rx_ring->rx_stats.page_alloc_fail++;
 419                 u64_stats_update_end(&rx_ring->syncp);
 420                 return -ENOMEM;
 421         }
 422
 423         dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE,
 424                            DMA_FROM_DEVICE);
 425         if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
 426                 u64_stats_update_begin(&rx_ring->syncp);
 427                 rx_ring->rx_stats.dma_mapping_err++;
 428                 u64_stats_update_end(&rx_ring->syncp);
 429
 430                 __free_page(page);
 431                 return -EIO;
 432         }
 433         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 434                   "alloc page %p, rx_info %p\n", page, rx_info);
 435
 436         rx_info->page = page;
 437         rx_info->page_offset = 0;
 438         ena_buf = &rx_info->ena_buf;
 439         ena_buf->paddr = dma;
 440         ena_buf->len = PAGE_SIZE;
 441
 442         return 0;
 443 }
 444
 445 static void ena_free_rx_page(struct ena_ring *rx_ring,
 446                              struct ena_rx_buffer *rx_info)
 447 {
 448         struct page *page = rx_info->page;
 449         struct ena_com_buf *ena_buf = &rx_info->ena_buf;
 450
 451         if (unlikely(!page)) {
 452                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
 453                            "Trying to free unallocated buffer\n");
 454                 return;
 455         }
 456
 457         dma_unmap_page(rx_ring->dev, ena_buf->paddr, PAGE_SIZE,
 458                        DMA_FROM_DEVICE);
 459
 460         __free_page(page);
 461         rx_info->page = NULL;
 462 }
 463
 464 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
 465 {
 466         u16 next_to_use;
 467         u32 i;
 468         int rc;
 469
 470         next_to_use = rx_ring->next_to_use;
 471
 472         for (i = 0; i < num; i++) {
 473                 struct ena_rx_buffer *rx_info =
 474                         &rx_ring->rx_buffer_info[next_to_use];
 475
 476                 rc = ena_alloc_rx_page(rx_ring, rx_info,
 477                                        __GFP_COLD | GFP_ATOMIC | __GFP_COMP);
 478                 if (unlikely(rc < 0)) {
 479                         netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
 480                                    "failed to alloc buffer for rx queue %d\n",
 481                                    rx_ring->qid);
 482                         break;
 483                 }
 484                 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
 485                                                 &rx_info->ena_buf,
 486                                                 next_to_use);
 487                 if (unlikely(rc)) {
 488                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
 489                                    "failed to add buffer for rx queue %d\n",
 490                                    rx_ring->qid);
 491                         break;
 492                 }
 493                 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
 494                                                    rx_ring->ring_size);
 495         }
 496
 497         if (unlikely(i < num)) {
 498                 u64_stats_update_begin(&rx_ring->syncp);
 499                 rx_ring->rx_stats.refil_partial++;
 500                 u64_stats_update_end(&rx_ring->syncp);
 501                 netdev_warn(rx_ring->netdev,
 502                             "refilled rx qid %d with only %d buffers (from %d)\n",
 503                             rx_ring->qid, i, num);
 504         }
 505
 506         if (likely(i)) {
 507                 /* Add memory barrier to make sure the desc were written before
 508                  * issue a doorbell
 509                  */
 510                 wmb();
 511                 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
 512         }
 513
 514         rx_ring->next_to_use = next_to_use;
 515
 516         return i;
 517 }
 518
 519 static void ena_free_rx_bufs(struct ena_adapter *adapter,
 520                              u32 qid)
 521 {
 522         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 523         u32 i;
 524
 525         for (i = 0; i < rx_ring->ring_size; i++) {
 526                 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
 527
 528                 if (rx_info->page)
 529                         ena_free_rx_page(rx_ring, rx_info);
 530         }
 531 }
 532
 533 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
 534  * @adapter: board private structure
 535  *
 536  */
 537 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
 538 {
 539         struct ena_ring *rx_ring;
 540         int i, rc, bufs_num;
 541
 542         for (i = 0; i < adapter->num_queues; i++) {
 543                 rx_ring = &adapter->rx_ring[i];
 544                 bufs_num = rx_ring->ring_size - 1;
 545                 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
 546
 547                 if (unlikely(rc != bufs_num))
 548                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
 549                                    "refilling Queue %d failed. allocated %d buffers from: %d\n",
 550                                    i, rc, bufs_num);
 551         }
 552 }
 553
 554 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
 555 {
 556         int i;
 557
 558         for (i = 0; i < adapter->num_queues; i++)
 559                 ena_free_rx_bufs(adapter, i);
 560 }
 561
 562 /* ena_free_tx_bufs - Free Tx Buffers per Queue
 563  * @tx_ring: TX ring for which buffers be freed
 564  */
 565 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
 566 {
 567         bool print_once = true;
 568         u32 i;
 569
 570         for (i = 0; i < tx_ring->ring_size; i++) {
 571                 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
 572                 struct ena_com_buf *ena_buf;
 573                 int nr_frags;
 574                 int j;
 575
 576                 if (!tx_info->skb)
 577                         continue;
 578
 579                 if (print_once) {
 580                         netdev_notice(tx_ring->netdev,
 581                                       "free uncompleted tx skb qid %d idx 0x%x\n",
 582                                       tx_ring->qid, i);
 583                         print_once = false;
 584                 } else {
 585                         netdev_dbg(tx_ring->netdev,
 586                                    "free uncompleted tx skb qid %d idx 0x%x\n",
 587                                    tx_ring->qid, i);
 588                 }
 589
 590                 ena_buf = tx_info->bufs;
 591                 dma_unmap_single(tx_ring->dev,
 592                                  ena_buf->paddr,
 593                                  ena_buf->len,
 594                                  DMA_TO_DEVICE);
 595
 596                 /* unmap remaining mapped pages */
 597                 nr_frags = tx_info->num_of_bufs - 1;
 598                 for (j = 0; j < nr_frags; j++) {
 599                         ena_buf++;
 600                         dma_unmap_page(tx_ring->dev,
 601                                        ena_buf->paddr,
 602                                        ena_buf->len,
 603                                        DMA_TO_DEVICE);
 604                 }
 605
 606                 dev_kfree_skb_any(tx_info->skb);
 607         }
 608         netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
 609                                                   tx_ring->qid));
 610 }
 611
 612 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
 613 {
 614         struct ena_ring *tx_ring;
 615         int i;
 616
 617         for (i = 0; i < adapter->num_queues; i++) {
 618                 tx_ring = &adapter->tx_ring[i];
 619                 ena_free_tx_bufs(tx_ring);
 620         }
 621 }
 622
 623 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
 624 {
 625         u16 ena_qid;
 626         int i;
 627
 628         for (i = 0; i < adapter->num_queues; i++) {
 629                 ena_qid = ENA_IO_TXQ_IDX(i);
 630                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
 631         }
 632 }
 633
 634 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
 635 {
 636         u16 ena_qid;
 637         int i;
 638
 639         for (i = 0; i < adapter->num_queues; i++) {
 640                 ena_qid = ENA_IO_RXQ_IDX(i);
 641                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
 642         }
 643 }
 644
 645 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
 646 {
 647         ena_destroy_all_tx_queues(adapter);
 648         ena_destroy_all_rx_queues(adapter);
 649 }
 650
 651 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
 652 {
 653         struct ena_tx_buffer *tx_info = NULL;
 654
 655         if (likely(req_id < tx_ring->ring_size)) {
 656                 tx_info = &tx_ring->tx_buffer_info[req_id];
 657                 if (likely(tx_info->skb))
 658                         return 0;
 659         }
 660
 661         if (tx_info)
 662                 netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
 663                           "tx_info doesn't have valid skb\n");
 664         else
 665                 netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
 666                           "Invalid req_id: %hu\n", req_id);
 667
 668         u64_stats_update_begin(&tx_ring->syncp);
 669         tx_ring->tx_stats.bad_req_id++;
 670         u64_stats_update_end(&tx_ring->syncp);
 671
 672         /* Trigger device reset */
 673         set_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags);
 674         return -EFAULT;
 675 }
 676
 677 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
 678 {
 679         struct netdev_queue *txq;
 680         bool above_thresh;
 681         u32 tx_bytes = 0;
 682         u32 total_done = 0;
 683         u16 next_to_clean;
 684         u16 req_id;
 685         int tx_pkts = 0;
 686         int rc;
 687
 688         next_to_clean = tx_ring->next_to_clean;
 689         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
 690
 691         while (tx_pkts < budget) {
 692                 struct ena_tx_buffer *tx_info;
 693                 struct sk_buff *skb;
 694                 struct ena_com_buf *ena_buf;
 695                 int i, nr_frags;
 696
 697                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
 698                                                 &req_id);
 699                 if (rc)
 700                         break;
 701
 702                 rc = validate_tx_req_id(tx_ring, req_id);
 703                 if (rc)
 704                         break;
 705
 706                 tx_info = &tx_ring->tx_buffer_info[req_id];
 707                 skb = tx_info->skb;
 708
 709                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
 710                 prefetch(&skb->end);
 711
 712                 tx_info->skb = NULL;
 713                 tx_info->last_jiffies = 0;
 714
 715                 if (likely(tx_info->num_of_bufs != 0)) {
 716                         ena_buf = tx_info->bufs;
 717
 718                         dma_unmap_single(tx_ring->dev,
 719                                          dma_unmap_addr(ena_buf, paddr),
 720                                          dma_unmap_len(ena_buf, len),
 721                                          DMA_TO_DEVICE);
 722
 723                         /* unmap remaining mapped pages */
 724                         nr_frags = tx_info->num_of_bufs - 1;
 725                         for (i = 0; i < nr_frags; i++) {
 726                                 ena_buf++;
 727                                 dma_unmap_page(tx_ring->dev,
 728                                                dma_unmap_addr(ena_buf, paddr),
 729                                                dma_unmap_len(ena_buf, len),
 730                                                DMA_TO_DEVICE);
 731                         }
 732                 }
 733
 734                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
 735                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
 736                           skb);
 737
 738                 tx_bytes += skb->len;
 739                 dev_kfree_skb(skb);
 740                 tx_pkts++;
 741                 total_done += tx_info->tx_descs;
 742
 743                 tx_ring->free_tx_ids[next_to_clean] = req_id;
 744                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
 745                                                      tx_ring->ring_size);
 746         }
 747
 748         tx_ring->next_to_clean = next_to_clean;
 749         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
 750         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
 751
 752         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
 753
 754         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
 755                   "tx_poll: q %d done. total pkts: %d\n",
 756                   tx_ring->qid, tx_pkts);
 757
 758         /* need to make the rings circular update visible to
 759          * ena_start_xmit() before checking for netif_queue_stopped().
 760          */
 761         smp_mb();
 762
 763         above_thresh = ena_com_sq_empty_space(tx_ring->ena_com_io_sq) >
 764                 ENA_TX_WAKEUP_THRESH;
 765         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
 766                 __netif_tx_lock(txq, smp_processor_id());
 767                 above_thresh = ena_com_sq_empty_space(tx_ring->ena_com_io_sq) >
 768                         ENA_TX_WAKEUP_THRESH;
 769                 if (netif_tx_queue_stopped(txq) && above_thresh) {
 770                         netif_tx_wake_queue(txq);
 771                         u64_stats_update_begin(&tx_ring->syncp);
 772                         tx_ring->tx_stats.queue_wakeup++;
 773                         u64_stats_update_end(&tx_ring->syncp);
 774                 }
 775                 __netif_tx_unlock(txq);
 776         }
 777
 778         tx_ring->per_napi_bytes += tx_bytes;
 779         tx_ring->per_napi_packets += tx_pkts;
 780
 781         return tx_pkts;
 782 }
 783
 784 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
 785                                   struct ena_com_rx_buf_info *ena_bufs,
 786                                   u32 descs,
 787                                   u16 *next_to_clean)
 788 {
 789         struct sk_buff *skb;
 790         struct ena_rx_buffer *rx_info =
 791                 &rx_ring->rx_buffer_info[*next_to_clean];
 792         u32 len;
 793         u32 buf = 0;
 794         void *va;
 795
 796         len = ena_bufs[0].len;
 797         if (unlikely(!rx_info->page)) {
 798                 netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 799                           "Page is NULL\n");
 800                 return NULL;
 801         }
 802
 803         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 804                   "rx_info %p page %p\n",
 805                   rx_info, rx_info->page);
 806
 807         /* save virt address of first buffer */
 808         va = page_address(rx_info->page) + rx_info->page_offset;
 809         prefetch(va + NET_IP_ALIGN);
 810
 811         if (len <= rx_ring->rx_copybreak) {
 812                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 813                                                 rx_ring->rx_copybreak);
 814                 if (unlikely(!skb)) {
 815                         u64_stats_update_begin(&rx_ring->syncp);
 816                         rx_ring->rx_stats.skb_alloc_fail++;
 817                         u64_stats_update_end(&rx_ring->syncp);
 818                         netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 819                                   "Failed to allocate skb\n");
 820                         return NULL;
 821                 }
 822
 823                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 824                           "rx allocated small packet. len %d. data_len %d\n",
 825                           skb->len, skb->data_len);
 826
 827                 /* sync this buffer for CPU use */
 828                 dma_sync_single_for_cpu(rx_ring->dev,
 829                                         dma_unmap_addr(&rx_info->ena_buf, paddr),
 830                                         len,
 831                                         DMA_FROM_DEVICE);
 832                 skb_copy_to_linear_data(skb, va, len);
 833                 dma_sync_single_for_device(rx_ring->dev,
 834                                            dma_unmap_addr(&rx_info->ena_buf, paddr),
 835                                            len,
 836                                            DMA_FROM_DEVICE);
 837
 838                 skb_put(skb, len);
 839                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 840                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
 841                                                      rx_ring->ring_size);
 842                 return skb;
 843         }
 844
 845         skb = napi_get_frags(rx_ring->napi);
 846         if (unlikely(!skb)) {
 847                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 848                           "Failed allocating skb\n");
 849                 u64_stats_update_begin(&rx_ring->syncp);
 850                 rx_ring->rx_stats.skb_alloc_fail++;
 851                 u64_stats_update_end(&rx_ring->syncp);
 852                 return NULL;
 853         }
 854
 855         do {
 856                 dma_unmap_page(rx_ring->dev,
 857                                dma_unmap_addr(&rx_info->ena_buf, paddr),
 858                                PAGE_SIZE, DMA_FROM_DEVICE);
 859
 860                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
 861                                 rx_info->page_offset, len, PAGE_SIZE);
 862
 863                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 864                           "rx skb updated. len %d. data_len %d\n",
 865                           skb->len, skb->data_len);
 866
 867                 rx_info->page = NULL;
 868                 *next_to_clean =
 869                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
 870                                              rx_ring->ring_size);
 871                 if (likely(--descs == 0))
 872                         break;
 873                 rx_info = &rx_ring->rx_buffer_info[*next_to_clean];
 874                 len = ena_bufs[++buf].len;
 875         } while (1);
 876
 877         return skb;
 878 }
 879
 880 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
 881  * @adapter: structure containing adapter specific data
 882  * @ena_rx_ctx: received packet context/metadata
 883  * @skb: skb currently being received and modified
 884  */
 885 static inline void ena_rx_checksum(struct ena_ring *rx_ring,
 886                                    struct ena_com_rx_ctx *ena_rx_ctx,
 887                                    struct sk_buff *skb)
 888 {
 889         /* Rx csum disabled */
 890         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
 891                 skb->ip_summed = CHECKSUM_NONE;
 892                 return;
 893         }
 894
 895         /* For fragmented packets the checksum isn't valid */
 896         if (ena_rx_ctx->frag) {
 897                 skb->ip_summed = CHECKSUM_NONE;
 898                 return;
 899         }
 900
 901         /* if IP and error */
 902         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
 903                      (ena_rx_ctx->l3_csum_err))) {
 904                 /* ipv4 checksum error */
 905                 skb->ip_summed = CHECKSUM_NONE;
 906                 u64_stats_update_begin(&rx_ring->syncp);
 907                 rx_ring->rx_stats.bad_csum++;
 908                 u64_stats_update_end(&rx_ring->syncp);
 909                 netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 910                           "RX IPv4 header checksum error\n");
 911                 return;
 912         }
 913
 914         /* if TCP/UDP */
 915         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
 916                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
 917                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
 918                         /* TCP/UDP checksum error */
 919                         u64_stats_update_begin(&rx_ring->syncp);
 920                         rx_ring->rx_stats.bad_csum++;
 921                         u64_stats_update_end(&rx_ring->syncp);
 922                         netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 923                                   "RX L4 checksum error\n");
 924                         skb->ip_summed = CHECKSUM_NONE;
 925                         return;
 926                 }
 927
 928                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 929         }
 930 }
 931
 932 static void ena_set_rx_hash(struct ena_ring *rx_ring,
 933                             struct ena_com_rx_ctx *ena_rx_ctx,
 934                             struct sk_buff *skb)
 935 {
 936         enum pkt_hash_types hash_type;
 937
 938         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
 939                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
 940                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
 941
 942                         hash_type = PKT_HASH_TYPE_L4;
 943                 else
 944                         hash_type = PKT_HASH_TYPE_NONE;
 945
 946                 /* Override hash type if the packet is fragmented */
 947                 if (ena_rx_ctx->frag)
 948                         hash_type = PKT_HASH_TYPE_NONE;
 949
 950                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
 951         }
 952 }
 953
 954 /* ena_clean_rx_irq - Cleanup RX irq
 955  * @rx_ring: RX ring to clean
 956  * @napi: napi handler
 957  * @budget: how many packets driver is allowed to clean
 958  *
 959  * Returns the number of cleaned buffers.
 960  */
 961 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
 962                             u32 budget)
 963 {
 964         u16 next_to_clean = rx_ring->next_to_clean;
 965         u32 res_budget, work_done;
 966
 967         struct ena_com_rx_ctx ena_rx_ctx;
 968         struct ena_adapter *adapter;
 969         struct sk_buff *skb;
 970         int refill_required;
 971         int refill_threshold;
 972         int rc = 0;
 973         int total_len = 0;
 974         int rx_copybreak_pkt = 0;
 975
 976         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 977                   "%s qid %d\n", __func__, rx_ring->qid);
 978         res_budget = budget;
 979
 980         do {
 981                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
 982                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
 983                 ena_rx_ctx.descs = 0;
 984                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
 985                                     rx_ring->ena_com_io_sq,
 986                                     &ena_rx_ctx);
 987                 if (unlikely(rc))
 988                         goto error;
 989
 990                 if (unlikely(ena_rx_ctx.descs == 0))
 991                         break;
 992
 993                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 994                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
 995                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
 996                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
 997
 998                 /* allocate skb and fill it */
 999                 skb = ena_rx_skb(rx_ring, rx_ring->ena_bufs, ena_rx_ctx.descs,
1000                                  &next_to_clean);
1001
1002                 /* exit if we failed to retrieve a buffer */
1003                 if (unlikely(!skb)) {
1004                         next_to_clean = ENA_RX_RING_IDX_ADD(next_to_clean,
1005                                                             ena_rx_ctx.descs,
1006                                                             rx_ring->ring_size);
1007                         break;
1008                 }
1009
1010                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1011
1012                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1013
1014                 skb_record_rx_queue(skb, rx_ring->qid);
1015
1016                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1017                         total_len += rx_ring->ena_bufs[0].len;
1018                         rx_copybreak_pkt++;
1019                         napi_gro_receive(napi, skb);
1020                 } else {
1021                         total_len += skb->len;
1022                         napi_gro_frags(napi);
1023                 }
1024
1025                 res_budget--;
1026         } while (likely(res_budget));
1027
1028         work_done = budget - res_budget;
1029         rx_ring->per_napi_bytes += total_len;
1030         rx_ring->per_napi_packets += work_done;
1031         u64_stats_update_begin(&rx_ring->syncp);
1032         rx_ring->rx_stats.bytes += total_len;
1033         rx_ring->rx_stats.cnt += work_done;
1034         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1035         u64_stats_update_end(&rx_ring->syncp);
1036
1037         rx_ring->next_to_clean = next_to_clean;
1038
1039         refill_required = ena_com_sq_empty_space(rx_ring->ena_com_io_sq);
1040         refill_threshold = rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER;
1041
1042         /* Optimization, try to batch new rx buffers */
1043         if (refill_required > refill_threshold) {
1044                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1045                 ena_refill_rx_bufs(rx_ring, refill_required);
1046         }
1047
1048         return work_done;
1049
1050 error:
1051         adapter = netdev_priv(rx_ring->netdev);
1052
1053         u64_stats_update_begin(&rx_ring->syncp);
1054         rx_ring->rx_stats.bad_desc_num++;
1055         u64_stats_update_end(&rx_ring->syncp);
1056
1057         /* Too many desc from the device. Trigger reset */
1058         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1059
1060         return 0;
1061 }
1062
1063 inline void ena_adjust_intr_moderation(struct ena_ring *rx_ring,
1064                                        struct ena_ring *tx_ring)
1065 {
1066         /* We apply adaptive moderation on Rx path only.
1067          * Tx uses static interrupt moderation.
1068          */
1069         ena_com_calculate_interrupt_delay(rx_ring->ena_dev,
1070                                           rx_ring->per_napi_packets,
1071                                           rx_ring->per_napi_bytes,
1072                                           &rx_ring->smoothed_interval,
1073                                           &rx_ring->moder_tbl_idx);
1074
1075         /* Reset per napi packets/bytes */
1076         tx_ring->per_napi_packets = 0;
1077         tx_ring->per_napi_bytes = 0;
1078         rx_ring->per_napi_packets = 0;
1079         rx_ring->per_napi_bytes = 0;
1080 }
1081
1082 static inline void ena_unmask_interrupt(struct ena_ring *tx_ring,
1083                                         struct ena_ring *rx_ring)
1084 {
1085         struct ena_eth_io_intr_reg intr_reg;
1086
1087         /* Update intr register: rx intr delay,
1088          * tx intr delay and interrupt unmask
1089          */
1090         ena_com_update_intr_reg(&intr_reg,
1091                                 rx_ring->smoothed_interval,
1092                                 tx_ring->smoothed_interval,
1093                                 true);
1094
1095         /* It is a shared MSI-X.
1096          * Tx and Rx CQ have pointer to it.
1097          * So we use one of them to reach the intr reg
1098          */
1099         ena_com_unmask_intr(rx_ring->ena_com_io_cq, &intr_reg);
1100 }
1101
1102 static inline void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1103                                              struct ena_ring *rx_ring)
1104 {
1105         int cpu = get_cpu();
1106         int numa_node;
1107
1108         /* Check only one ring since the 2 rings are running on the same cpu */
1109         if (likely(tx_ring->cpu == cpu))
1110                 goto out;
1111
1112         numa_node = cpu_to_node(cpu);
1113         put_cpu();
1114
1115         if (numa_node != NUMA_NO_NODE) {
1116                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1117                 ena_com_update_numa_node(rx_ring->ena_com_io_cq, numa_node);
1118         }
1119
1120         tx_ring->cpu = cpu;
1121         rx_ring->cpu = cpu;
1122
1123         return;
1124 out:
1125         put_cpu();
1126 }
1127
1128 static int ena_io_poll(struct napi_struct *napi, int budget)
1129 {
1130         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1131         struct ena_ring *tx_ring, *rx_ring;
1132
1133         u32 tx_work_done;
1134         u32 rx_work_done;
1135         int tx_budget;
1136         int napi_comp_call = 0;
1137         int ret;
1138
1139         tx_ring = ena_napi->tx_ring;
1140         rx_ring = ena_napi->rx_ring;
1141
1142         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1143
1144         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1145             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1146                 napi_complete_done(napi, 0);
1147                 return 0;
1148         }
1149
1150         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1151         rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1152
1153         /* If the device is about to reset or down, avoid unmask
1154          * the interrupt and return 0 so NAPI won't reschedule
1155          */
1156         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1157                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1158                 napi_complete_done(napi, 0);
1159                 ret = 0;
1160
1161         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1162                 napi_comp_call = 1;
1163
1164                 /* Update numa and unmask the interrupt only when schedule
1165                  * from the interrupt context (vs from sk_busy_loop)
1166                  */
1167                 if (napi_complete_done(napi, rx_work_done)) {
1168                         /* Tx and Rx share the same interrupt vector */
1169                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1170                                 ena_adjust_intr_moderation(rx_ring, tx_ring);
1171
1172                         ena_unmask_interrupt(tx_ring, rx_ring);
1173                 }
1174
1175                 ena_update_ring_numa_node(tx_ring, rx_ring);
1176
1177                 ret = rx_work_done;
1178         } else {
1179                 ret = budget;
1180         }
1181
1182         u64_stats_update_begin(&tx_ring->syncp);
1183         tx_ring->tx_stats.napi_comp += napi_comp_call;
1184         tx_ring->tx_stats.tx_poll++;
1185         u64_stats_update_end(&tx_ring->syncp);
1186
1187         return ret;
1188 }
1189
1190 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1191 {
1192         struct ena_adapter *adapter = (struct ena_adapter *)data;
1193
1194         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1195
1196         /* Don't call the aenq handler before probe is done */
1197         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1198                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
1199
1200         return IRQ_HANDLED;
1201 }
1202
1203 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1204  * @irq: interrupt number
1205  * @data: pointer to a network interface private napi device structure
1206  */
1207 static irqreturn_t ena_intr_msix_io(int irq, void *data)
1208 {
1209         struct ena_napi *ena_napi = data;
1210
1211         napi_schedule(&ena_napi->napi);
1212
1213         return IRQ_HANDLED;
1214 }
1215
1216 static int ena_enable_msix(struct ena_adapter *adapter, int num_queues)
1217 {
1218         int msix_vecs, rc;
1219
1220         /* Reserved the max msix vectors we might need */
1221         msix_vecs = ENA_MAX_MSIX_VEC(num_queues);
1222
1223         netif_dbg(adapter, probe, adapter->netdev,
1224                   "trying to enable MSI-X, vectors %d\n", msix_vecs);
1225
1226         rc = pci_alloc_irq_vectors(adapter->pdev, msix_vecs, msix_vecs,
1227                         PCI_IRQ_MSIX);
1228         if (rc < 0) {
1229                 netif_err(adapter, probe, adapter->netdev,
1230                           "Failed to enable MSI-X, vectors %d rc %d\n",
1231                           msix_vecs, rc);
1232                 return -ENOSPC;
1233         }
1234
1235         netif_dbg(adapter, probe, adapter->netdev, "enable MSI-X, vectors %d\n",
1236                   msix_vecs);
1237
1238         if (msix_vecs >= 1) {
1239                 if (ena_init_rx_cpu_rmap(adapter))
1240                         netif_warn(adapter, probe, adapter->netdev,
1241                                    "Failed to map IRQs to CPUs\n");
1242         }
1243
1244         adapter->msix_vecs = msix_vecs;
1245
1246         return 0;
1247 }
1248
1249 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1250 {
1251         u32 cpu;
1252
1253         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1254                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1255                  pci_name(adapter->pdev));
1256         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
1257                 ena_intr_msix_mgmnt;
1258         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1259         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1260                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
1261         cpu = cpumask_first(cpu_online_mask);
1262         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
1263         cpumask_set_cpu(cpu,
1264                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
1265 }
1266
1267 static void ena_setup_io_intr(struct ena_adapter *adapter)
1268 {
1269         struct net_device *netdev;
1270         int irq_idx, i, cpu;
1271
1272         netdev = adapter->netdev;
1273
1274         for (i = 0; i < adapter->num_queues; i++) {
1275                 irq_idx = ENA_IO_IRQ_IDX(i);
1276                 cpu = i % num_online_cpus();
1277
1278                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1279                          "%s-Tx-Rx-%d", netdev->name, i);
1280                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
1281                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
1282                 adapter->irq_tbl[irq_idx].vector =
1283                         pci_irq_vector(adapter->pdev, irq_idx);
1284                 adapter->irq_tbl[irq_idx].cpu = cpu;
1285
1286                 cpumask_set_cpu(cpu,
1287                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
1288         }
1289 }
1290
1291 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
1292 {
1293         unsigned long flags = 0;
1294         struct ena_irq *irq;
1295         int rc;
1296
1297         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1298         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1299                          irq->data);
1300         if (rc) {
1301                 netif_err(adapter, probe, adapter->netdev,
1302                           "failed to request admin irq\n");
1303                 return rc;
1304         }
1305
1306         netif_dbg(adapter, probe, adapter->netdev,
1307                   "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
1308                   irq->affinity_hint_mask.bits[0], irq->vector);
1309
1310         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1311
1312         return rc;
1313 }
1314
1315 static int ena_request_io_irq(struct ena_adapter *adapter)
1316 {
1317         unsigned long flags = 0;
1318         struct ena_irq *irq;
1319         int rc = 0, i, k;
1320
1321         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1322                 irq = &adapter->irq_tbl[i];
1323                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1324                                  irq->data);
1325                 if (rc) {
1326                         netif_err(adapter, ifup, adapter->netdev,
1327                                   "Failed to request I/O IRQ. index %d rc %d\n",
1328                                    i, rc);
1329                         goto err;
1330                 }
1331
1332                 netif_dbg(adapter, ifup, adapter->netdev,
1333                           "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
1334                           i, irq->affinity_hint_mask.bits[0], irq->vector);
1335
1336                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1337         }
1338
1339         return rc;
1340
1341 err:
1342         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
1343                 irq = &adapter->irq_tbl[k];
1344                 free_irq(irq->vector, irq->data);
1345         }
1346
1347         return rc;
1348 }
1349
1350 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
1351 {
1352         struct ena_irq *irq;
1353
1354         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1355         synchronize_irq(irq->vector);
1356         irq_set_affinity_hint(irq->vector, NULL);
1357         free_irq(irq->vector, irq->data);
1358 }
1359
1360 static void ena_free_io_irq(struct ena_adapter *adapter)
1361 {
1362         struct ena_irq *irq;
1363         int i;
1364
1365 #ifdef CONFIG_RFS_ACCEL
1366         if (adapter->msix_vecs >= 1) {
1367                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
1368                 adapter->netdev->rx_cpu_rmap = NULL;
1369         }
1370 #endif /* CONFIG_RFS_ACCEL */
1371
1372         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1373                 irq = &adapter->irq_tbl[i];
1374                 irq_set_affinity_hint(irq->vector, NULL);
1375                 free_irq(irq->vector, irq->data);
1376         }
1377 }
1378
1379 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
1380 {
1381         int i;
1382
1383         if (!netif_running(adapter->netdev))
1384                 return;
1385
1386         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++)
1387                 synchronize_irq(adapter->irq_tbl[i].vector);
1388 }
1389
1390 static void ena_del_napi(struct ena_adapter *adapter)
1391 {
1392         int i;
1393
1394         for (i = 0; i < adapter->num_queues; i++)
1395                 netif_napi_del(&adapter->ena_napi[i].napi);
1396 }
1397
1398 static void ena_init_napi(struct ena_adapter *adapter)
1399 {
1400         struct ena_napi *napi;
1401         int i;
1402
1403         for (i = 0; i < adapter->num_queues; i++) {
1404                 napi = &adapter->ena_napi[i];
1405
1406                 netif_napi_add(adapter->netdev,
1407                                &adapter->ena_napi[i].napi,
1408                                ena_io_poll,
1409                                ENA_NAPI_BUDGET);
1410                 napi->rx_ring = &adapter->rx_ring[i];
1411                 napi->tx_ring = &adapter->tx_ring[i];
1412                 napi->qid = i;
1413         }
1414 }
1415
1416 static void ena_napi_disable_all(struct ena_adapter *adapter)
1417 {
1418         int i;
1419
1420         for (i = 0; i < adapter->num_queues; i++)
1421                 napi_disable(&adapter->ena_napi[i].napi);
1422 }
1423
1424 static void ena_napi_enable_all(struct ena_adapter *adapter)
1425 {
1426         int i;
1427
1428         for (i = 0; i < adapter->num_queues; i++)
1429                 napi_enable(&adapter->ena_napi[i].napi);
1430 }
1431
1432 static void ena_restore_ethtool_params(struct ena_adapter *adapter)
1433 {
1434         adapter->tx_usecs = 0;
1435         adapter->rx_usecs = 0;
1436         adapter->tx_frames = 1;
1437         adapter->rx_frames = 1;
1438 }
1439
1440 /* Configure the Rx forwarding */
1441 static int ena_rss_configure(struct ena_adapter *adapter)
1442 {
1443         struct ena_com_dev *ena_dev = adapter->ena_dev;
1444         int rc;
1445
1446         /* In case the RSS table wasn't initialized by probe */
1447         if (!ena_dev->rss.tbl_log_size) {
1448                 rc = ena_rss_init_default(adapter);
1449                 if (rc && (rc != -EPERM)) {
1450                         netif_err(adapter, ifup, adapter->netdev,
1451                                   "Failed to init RSS rc: %d\n", rc);
1452                         return rc;
1453                 }
1454         }
1455
1456         /* Set indirect table */
1457         rc = ena_com_indirect_table_set(ena_dev);
1458         if (unlikely(rc && rc != -EPERM))
1459                 return rc;
1460
1461         /* Configure hash function (if supported) */
1462         rc = ena_com_set_hash_function(ena_dev);
1463         if (unlikely(rc && (rc != -EPERM)))
1464                 return rc;
1465
1466         /* Configure hash inputs (if supported) */
1467         rc = ena_com_set_hash_ctrl(ena_dev);
1468         if (unlikely(rc && (rc != -EPERM)))
1469                 return rc;
1470
1471         return 0;
1472 }
1473
1474 static int ena_up_complete(struct ena_adapter *adapter)
1475 {
1476         int rc, i;
1477
1478         rc = ena_rss_configure(adapter);
1479         if (rc)
1480                 return rc;
1481
1482         ena_init_napi(adapter);
1483
1484         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
1485
1486         ena_refill_all_rx_bufs(adapter);
1487
1488         /* enable transmits */
1489         netif_tx_start_all_queues(adapter->netdev);
1490
1491         ena_restore_ethtool_params(adapter);
1492
1493         ena_napi_enable_all(adapter);
1494
1495         /* Enable completion queues interrupt */
1496         for (i = 0; i < adapter->num_queues; i++)
1497                 ena_unmask_interrupt(&adapter->tx_ring[i],
1498                                      &adapter->rx_ring[i]);
1499
1500         /* schedule napi in case we had pending packets
1501          * from the last time we disable napi
1502          */
1503         for (i = 0; i < adapter->num_queues; i++)
1504                 napi_schedule(&adapter->ena_napi[i].napi);
1505
1506         return 0;
1507 }
1508
1509 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
1510 {
1511         struct ena_com_create_io_ctx ctx = { 0 };
1512         struct ena_com_dev *ena_dev;
1513         struct ena_ring *tx_ring;
1514         u32 msix_vector;
1515         u16 ena_qid;
1516         int rc;
1517
1518         ena_dev = adapter->ena_dev;
1519
1520         tx_ring = &adapter->tx_ring[qid];
1521         msix_vector = ENA_IO_IRQ_IDX(qid);
1522         ena_qid = ENA_IO_TXQ_IDX(qid);
1523
1524         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1525         ctx.qid = ena_qid;
1526         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1527         ctx.msix_vector = msix_vector;
1528         ctx.queue_size = adapter->tx_ring_size;
1529         ctx.numa_node = cpu_to_node(tx_ring->cpu);
1530
1531         rc = ena_com_create_io_queue(ena_dev, &ctx);
1532         if (rc) {
1533                 netif_err(adapter, ifup, adapter->netdev,
1534                           "Failed to create I/O TX queue num %d rc: %d\n",
1535                           qid, rc);
1536                 return rc;
1537         }
1538
1539         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1540                                      &tx_ring->ena_com_io_sq,
1541                                      &tx_ring->ena_com_io_cq);
1542         if (rc) {
1543                 netif_err(adapter, ifup, adapter->netdev,
1544                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1545                           qid, rc);
1546                 ena_com_destroy_io_queue(ena_dev, ena_qid);
1547                 return rc;
1548         }
1549
1550         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
1551         return rc;
1552 }
1553
1554 static int ena_create_all_io_tx_queues(struct ena_adapter *adapter)
1555 {
1556         struct ena_com_dev *ena_dev = adapter->ena_dev;
1557         int rc, i;
1558
1559         for (i = 0; i < adapter->num_queues; i++) {
1560                 rc = ena_create_io_tx_queue(adapter, i);
1561                 if (rc)
1562                         goto create_err;
1563         }
1564
1565         return 0;
1566
1567 create_err:
1568         while (i--)
1569                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1570
1571         return rc;
1572 }
1573
1574 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
1575 {
1576         struct ena_com_dev *ena_dev;
1577         struct ena_com_create_io_ctx ctx = { 0 };
1578         struct ena_ring *rx_ring;
1579         u32 msix_vector;
1580         u16 ena_qid;
1581         int rc;
1582
1583         ena_dev = adapter->ena_dev;
1584
1585         rx_ring = &adapter->rx_ring[qid];
1586         msix_vector = ENA_IO_IRQ_IDX(qid);
1587         ena_qid = ENA_IO_RXQ_IDX(qid);
1588
1589         ctx.qid = ena_qid;
1590         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1591         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1592         ctx.msix_vector = msix_vector;
1593         ctx.queue_size = adapter->rx_ring_size;
1594         ctx.numa_node = cpu_to_node(rx_ring->cpu);
1595
1596         rc = ena_com_create_io_queue(ena_dev, &ctx);
1597         if (rc) {
1598                 netif_err(adapter, ifup, adapter->netdev,
1599                           "Failed to create I/O RX queue num %d rc: %d\n",
1600                           qid, rc);
1601                 return rc;
1602         }
1603
1604         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1605                                      &rx_ring->ena_com_io_sq,
1606                                      &rx_ring->ena_com_io_cq);
1607         if (rc) {
1608                 netif_err(adapter, ifup, adapter->netdev,
1609                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1610                           qid, rc);
1611                 ena_com_destroy_io_queue(ena_dev, ena_qid);
1612                 return rc;
1613         }
1614
1615         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
1616
1617         return rc;
1618 }
1619
1620 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
1621 {
1622         struct ena_com_dev *ena_dev = adapter->ena_dev;
1623         int rc, i;
1624
1625         for (i = 0; i < adapter->num_queues; i++) {
1626                 rc = ena_create_io_rx_queue(adapter, i);
1627                 if (rc)
1628                         goto create_err;
1629         }
1630
1631         return 0;
1632
1633 create_err:
1634         while (i--)
1635                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1636
1637         return rc;
1638 }
1639
1640 static int ena_up(struct ena_adapter *adapter)
1641 {
1642         int rc;
1643
1644         netdev_dbg(adapter->netdev, "%s\n", __func__);
1645
1646         ena_setup_io_intr(adapter);
1647
1648         rc = ena_request_io_irq(adapter);
1649         if (rc)
1650                 goto err_req_irq;
1651
1652         /* allocate transmit descriptors */
1653         rc = ena_setup_all_tx_resources(adapter);
1654         if (rc)
1655                 goto err_setup_tx;
1656
1657         /* allocate receive descriptors */
1658         rc = ena_setup_all_rx_resources(adapter);
1659         if (rc)
1660                 goto err_setup_rx;
1661
1662         /* Create TX queues */
1663         rc = ena_create_all_io_tx_queues(adapter);
1664         if (rc)
1665                 goto err_create_tx_queues;
1666
1667         /* Create RX queues */
1668         rc = ena_create_all_io_rx_queues(adapter);
1669         if (rc)
1670                 goto err_create_rx_queues;
1671
1672         rc = ena_up_complete(adapter);
1673         if (rc)
1674                 goto err_up;
1675
1676         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
1677                 netif_carrier_on(adapter->netdev);
1678
1679         u64_stats_update_begin(&adapter->syncp);
1680         adapter->dev_stats.interface_up++;
1681         u64_stats_update_end(&adapter->syncp);
1682
1683         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1684
1685         return rc;
1686
1687 err_up:
1688         ena_destroy_all_rx_queues(adapter);
1689 err_create_rx_queues:
1690         ena_destroy_all_tx_queues(adapter);
1691 err_create_tx_queues:
1692         ena_free_all_io_rx_resources(adapter);
1693 err_setup_rx:
1694         ena_free_all_io_tx_resources(adapter);
1695 err_setup_tx:
1696         ena_free_io_irq(adapter);
1697 err_req_irq:
1698
1699         return rc;
1700 }
1701
1702 static void ena_down(struct ena_adapter *adapter)
1703 {
1704         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
1705
1706         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1707
1708         u64_stats_update_begin(&adapter->syncp);
1709         adapter->dev_stats.interface_down++;
1710         u64_stats_update_end(&adapter->syncp);
1711
1712         netif_carrier_off(adapter->netdev);
1713         netif_tx_disable(adapter->netdev);
1714
1715         /* After this point the napi handler won't enable the tx queue */
1716         ena_napi_disable_all(adapter);
1717
1718         /* After destroy the queue there won't be any new interrupts */
1719
1720         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
1721                 int rc;
1722
1723                 rc = ena_com_dev_reset(adapter->ena_dev);
1724                 if (rc)
1725                         dev_err(&adapter->pdev->dev, "Device reset failed\n");
1726         }
1727
1728         ena_destroy_all_io_queues(adapter);
1729
1730         ena_disable_io_intr_sync(adapter);
1731         ena_free_io_irq(adapter);
1732         ena_del_napi(adapter);
1733
1734         ena_free_all_tx_bufs(adapter);
1735         ena_free_all_rx_bufs(adapter);
1736         ena_free_all_io_tx_resources(adapter);
1737         ena_free_all_io_rx_resources(adapter);
1738 }
1739
1740 /* ena_open - Called when a network interface is made active
1741  * @netdev: network interface device structure
1742  *
1743  * Returns 0 on success, negative value on failure
1744  *
1745  * The open entry point is called when a network interface is made
1746  * active by the system (IFF_UP).  At this point all resources needed
1747  * for transmit and receive operations are allocated, the interrupt
1748  * handler is registered with the OS, the watchdog timer is started,
1749  * and the stack is notified that the interface is ready.
1750  */
1751 static int ena_open(struct net_device *netdev)
1752 {
1753         struct ena_adapter *adapter = netdev_priv(netdev);
1754         int rc;
1755
1756         /* Notify the stack of the actual queue counts. */
1757         rc = netif_set_real_num_tx_queues(netdev, adapter->num_queues);
1758         if (rc) {
1759                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
1760                 return rc;
1761         }
1762
1763         rc = netif_set_real_num_rx_queues(netdev, adapter->num_queues);
1764         if (rc) {
1765                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
1766                 return rc;
1767         }
1768
1769         rc = ena_up(adapter);
1770         if (rc)
1771                 return rc;
1772
1773         return rc;
1774 }
1775
1776 /* ena_close - Disables a network interface
1777  * @netdev: network interface device structure
1778  *
1779  * Returns 0, this is not allowed to fail
1780  *
1781  * The close entry point is called when an interface is de-activated
1782  * by the OS.  The hardware is still under the drivers control, but
1783  * needs to be disabled.  A global MAC reset is issued to stop the
1784  * hardware, and all transmit and receive resources are freed.
1785  */
1786 static int ena_close(struct net_device *netdev)
1787 {
1788         struct ena_adapter *adapter = netdev_priv(netdev);
1789
1790         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
1791
1792         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
1793                 ena_down(adapter);
1794
1795         return 0;
1796 }
1797
1798 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct sk_buff *skb)
1799 {
1800         u32 mss = skb_shinfo(skb)->gso_size;
1801         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
1802         u8 l4_protocol = 0;
1803
1804         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
1805                 ena_tx_ctx->l4_csum_enable = 1;
1806                 if (mss) {
1807                         ena_tx_ctx->tso_enable = 1;
1808                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
1809                         ena_tx_ctx->l4_csum_partial = 0;
1810                 } else {
1811                         ena_tx_ctx->tso_enable = 0;
1812                         ena_meta->l4_hdr_len = 0;
1813                         ena_tx_ctx->l4_csum_partial = 1;
1814                 }
1815
1816                 switch (ip_hdr(skb)->version) {
1817                 case IPVERSION:
1818                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
1819                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
1820                                 ena_tx_ctx->df = 1;
1821                         if (mss)
1822                                 ena_tx_ctx->l3_csum_enable = 1;
1823                         l4_protocol = ip_hdr(skb)->protocol;
1824                         break;
1825                 case 6:
1826                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
1827                         l4_protocol = ipv6_hdr(skb)->nexthdr;
1828                         break;
1829                 default:
1830                         break;
1831                 }
1832
1833                 if (l4_protocol == IPPROTO_TCP)
1834                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
1835                 else
1836                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
1837
1838                 ena_meta->mss = mss;
1839                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
1840                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
1841                 ena_tx_ctx->meta_valid = 1;
1842
1843         } else {
1844                 ena_tx_ctx->meta_valid = 0;
1845         }
1846 }
1847
1848 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
1849                                        struct sk_buff *skb)
1850 {
1851         int num_frags, header_len, rc;
1852
1853         num_frags = skb_shinfo(skb)->nr_frags;
1854         header_len = skb_headlen(skb);
1855
1856         if (num_frags < tx_ring->sgl_size)
1857                 return 0;
1858
1859         if ((num_frags == tx_ring->sgl_size) &&
1860             (header_len < tx_ring->tx_max_header_size))
1861                 return 0;
1862
1863         u64_stats_update_begin(&tx_ring->syncp);
1864         tx_ring->tx_stats.linearize++;
1865         u64_stats_update_end(&tx_ring->syncp);
1866
1867         rc = skb_linearize(skb);
1868         if (unlikely(rc)) {
1869                 u64_stats_update_begin(&tx_ring->syncp);
1870                 tx_ring->tx_stats.linearize_failed++;
1871                 u64_stats_update_end(&tx_ring->syncp);
1872         }
1873
1874         return rc;
1875 }
1876
1877 /* Called with netif_tx_lock. */
1878 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
1879 {
1880         struct ena_adapter *adapter = netdev_priv(dev);
1881         struct ena_tx_buffer *tx_info;
1882         struct ena_com_tx_ctx ena_tx_ctx;
1883         struct ena_ring *tx_ring;
1884         struct netdev_queue *txq;
1885         struct ena_com_buf *ena_buf;
1886         void *push_hdr;
1887         u32 len, last_frag;
1888         u16 next_to_use;
1889         u16 req_id;
1890         u16 push_len;
1891         u16 header_len;
1892         dma_addr_t dma;
1893         int qid, rc, nb_hw_desc;
1894         int i = -1;
1895
1896         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
1897         /*  Determine which tx ring we will be placed on */
1898         qid = skb_get_queue_mapping(skb);
1899         tx_ring = &adapter->tx_ring[qid];
1900         txq = netdev_get_tx_queue(dev, qid);
1901
1902         rc = ena_check_and_linearize_skb(tx_ring, skb);
1903         if (unlikely(rc))
1904                 goto error_drop_packet;
1905
1906         skb_tx_timestamp(skb);
1907         len = skb_headlen(skb);
1908
1909         next_to_use = tx_ring->next_to_use;
1910         req_id = tx_ring->free_tx_ids[next_to_use];
1911         tx_info = &tx_ring->tx_buffer_info[req_id];
1912         tx_info->num_of_bufs = 0;
1913
1914         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
1915         ena_buf = tx_info->bufs;
1916         tx_info->skb = skb;
1917
1918         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
1919                 /* prepared the push buffer */
1920                 push_len = min_t(u32, len, tx_ring->tx_max_header_size);
1921                 header_len = push_len;
1922                 push_hdr = skb->data;
1923         } else {
1924                 push_len = 0;
1925                 header_len = min_t(u32, len, tx_ring->tx_max_header_size);
1926                 push_hdr = NULL;
1927         }
1928
1929         netif_dbg(adapter, tx_queued, dev,
1930                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
1931                   push_hdr, push_len);
1932
1933         if (len > push_len) {
1934                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
1935                                      len - push_len, DMA_TO_DEVICE);
1936                 if (dma_mapping_error(tx_ring->dev, dma))
1937                         goto error_report_dma_error;
1938
1939                 ena_buf->paddr = dma;
1940                 ena_buf->len = len - push_len;
1941
1942                 ena_buf++;
1943                 tx_info->num_of_bufs++;
1944         }
1945
1946         last_frag = skb_shinfo(skb)->nr_frags;
1947
1948         for (i = 0; i < last_frag; i++) {
1949                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1950
1951                 len = skb_frag_size(frag);
1952                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, len,
1953                                        DMA_TO_DEVICE);
1954                 if (dma_mapping_error(tx_ring->dev, dma))
1955                         goto error_report_dma_error;
1956
1957                 ena_buf->paddr = dma;
1958                 ena_buf->len = len;
1959                 ena_buf++;
1960         }
1961
1962         tx_info->num_of_bufs += last_frag;
1963
1964         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
1965         ena_tx_ctx.ena_bufs = tx_info->bufs;
1966         ena_tx_ctx.push_header = push_hdr;
1967         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
1968         ena_tx_ctx.req_id = req_id;
1969         ena_tx_ctx.header_len = header_len;
1970
1971         /* set flags and meta data */
1972         ena_tx_csum(&ena_tx_ctx, skb);
1973
1974         /* prepare the packet's descriptors to dma engine */
1975         rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq, &ena_tx_ctx,
1976                                 &nb_hw_desc);
1977
1978         if (unlikely(rc)) {
1979                 netif_err(adapter, tx_queued, dev,
1980                           "failed to prepare tx bufs\n");
1981                 u64_stats_update_begin(&tx_ring->syncp);
1982                 tx_ring->tx_stats.queue_stop++;
1983                 tx_ring->tx_stats.prepare_ctx_err++;
1984                 u64_stats_update_end(&tx_ring->syncp);
1985                 netif_tx_stop_queue(txq);
1986                 goto error_unmap_dma;
1987         }
1988
1989         netdev_tx_sent_queue(txq, skb->len);
1990
1991         u64_stats_update_begin(&tx_ring->syncp);
1992         tx_ring->tx_stats.cnt++;
1993         tx_ring->tx_stats.bytes += skb->len;
1994         u64_stats_update_end(&tx_ring->syncp);
1995
1996         tx_info->tx_descs = nb_hw_desc;
1997         tx_info->last_jiffies = jiffies;
1998         tx_info->print_once = 0;
1999
2000         tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
2001                 tx_ring->ring_size);
2002
2003         /* This WMB is aimed to:
2004          * 1 - perform smp barrier before reading next_to_completion
2005          * 2 - make sure the desc were written before trigger DB
2006          */
2007         wmb();
2008
2009         /* stop the queue when no more space available, the packet can have up
2010          * to sgl_size + 2. one for the meta descriptor and one for header
2011          * (if the header is larger than tx_max_header_size).
2012          */
2013         if (unlikely(ena_com_sq_empty_space(tx_ring->ena_com_io_sq) <
2014                      (tx_ring->sgl_size + 2))) {
2015                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
2016                           __func__, qid);
2017
2018                 netif_tx_stop_queue(txq);
2019                 u64_stats_update_begin(&tx_ring->syncp);
2020                 tx_ring->tx_stats.queue_stop++;
2021                 u64_stats_update_end(&tx_ring->syncp);
2022
2023                 /* There is a rare condition where this function decide to
2024                  * stop the queue but meanwhile clean_tx_irq updates
2025                  * next_to_completion and terminates.
2026                  * The queue will remain stopped forever.
2027                  * To solve this issue this function perform rmb, check
2028                  * the wakeup condition and wake up the queue if needed.
2029                  */
2030                 smp_rmb();
2031
2032                 if (ena_com_sq_empty_space(tx_ring->ena_com_io_sq)
2033                                 > ENA_TX_WAKEUP_THRESH) {
2034                         netif_tx_wake_queue(txq);
2035                         u64_stats_update_begin(&tx_ring->syncp);
2036                         tx_ring->tx_stats.queue_wakeup++;
2037                         u64_stats_update_end(&tx_ring->syncp);
2038                 }
2039         }
2040
2041         if (netif_xmit_stopped(txq) || !skb->xmit_more) {
2042                 /* trigger the dma engine */
2043                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2044                 u64_stats_update_begin(&tx_ring->syncp);
2045                 tx_ring->tx_stats.doorbells++;
2046                 u64_stats_update_end(&tx_ring->syncp);
2047         }
2048
2049         return NETDEV_TX_OK;
2050
2051 error_report_dma_error:
2052         u64_stats_update_begin(&tx_ring->syncp);
2053         tx_ring->tx_stats.dma_mapping_err++;
2054         u64_stats_update_end(&tx_ring->syncp);
2055         netdev_warn(adapter->netdev, "failed to map skb\n");
2056
2057         tx_info->skb = NULL;
2058
2059 error_unmap_dma:
2060         if (i >= 0) {
2061                 /* save value of frag that failed */
2062                 last_frag = i;
2063
2064                 /* start back at beginning and unmap skb */
2065                 tx_info->skb = NULL;
2066                 ena_buf = tx_info->bufs;
2067                 dma_unmap_single(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
2068                                  dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
2069
2070                 /* unmap remaining mapped pages */
2071                 for (i = 0; i < last_frag; i++) {
2072                         ena_buf++;
2073                         dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
2074                                        dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
2075                 }
2076         }
2077
2078 error_drop_packet:
2079
2080         dev_kfree_skb(skb);
2081         return NETDEV_TX_OK;
2082 }
2083
2084 #ifdef CONFIG_NET_POLL_CONTROLLER
2085 static void ena_netpoll(struct net_device *netdev)
2086 {
2087         struct ena_adapter *adapter = netdev_priv(netdev);
2088         int i;
2089
2090         /* Dont schedule NAPI if the driver is in the middle of reset
2091          * or netdev is down.
2092          */
2093
2094         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags) ||
2095             test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2096                 return;
2097
2098         for (i = 0; i < adapter->num_queues; i++)
2099                 napi_schedule(&adapter->ena_napi[i].napi);
2100 }
2101 #endif /* CONFIG_NET_POLL_CONTROLLER */
2102
2103 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
2104                             void *accel_priv, select_queue_fallback_t fallback)
2105 {
2106         u16 qid;
2107         /* we suspect that this is good for in--kernel network services that
2108          * want to loop incoming skb rx to tx in normal user generated traffic,
2109          * most probably we will not get to this
2110          */
2111         if (skb_rx_queue_recorded(skb))
2112                 qid = skb_get_rx_queue(skb);
2113         else
2114                 qid = fallback(dev, skb);
2115
2116         return qid;
2117 }
2118
2119 static void ena_config_host_info(struct ena_com_dev *ena_dev)
2120 {
2121         struct ena_admin_host_info *host_info;
2122         int rc;
2123
2124         /* Allocate only the host info */
2125         rc = ena_com_allocate_host_info(ena_dev);
2126         if (rc) {
2127                 pr_err("Cannot allocate host info\n");
2128                 return;
2129         }
2130
2131         host_info = ena_dev->host_attr.host_info;
2132
2133         host_info->os_type = ENA_ADMIN_OS_LINUX;
2134         host_info->kernel_ver = LINUX_VERSION_CODE;
2135         strncpy(host_info->kernel_ver_str, utsname()->version,
2136                 sizeof(host_info->kernel_ver_str) - 1);
2137         host_info->os_dist = 0;
2138         strncpy(host_info->os_dist_str, utsname()->release,
2139                 sizeof(host_info->os_dist_str) - 1);
2140         host_info->driver_version =
2141                 (DRV_MODULE_VER_MAJOR) |
2142                 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
2143                 (DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
2144
2145         rc = ena_com_set_host_attributes(ena_dev);
2146         if (rc) {
2147                 if (rc == -EPERM)
2148                         pr_warn("Cannot set host attributes\n");
2149                 else
2150                         pr_err("Cannot set host attributes\n");
2151
2152                 goto err;
2153         }
2154
2155         return;
2156
2157 err:
2158         ena_com_delete_host_info(ena_dev);
2159 }
2160
2161 static void ena_config_debug_area(struct ena_adapter *adapter)
2162 {
2163         u32 debug_area_size;
2164         int rc, ss_count;
2165
2166         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
2167         if (ss_count <= 0) {
2168                 netif_err(adapter, drv, adapter->netdev,
2169                           "SS count is negative\n");
2170                 return;
2171         }
2172
2173         /* allocate 32 bytes for each string and 64bit for the value */
2174         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
2175
2176         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
2177         if (rc) {
2178                 pr_err("Cannot allocate debug area\n");
2179                 return;
2180         }
2181
2182         rc = ena_com_set_host_attributes(adapter->ena_dev);
2183         if (rc) {
2184                 if (rc == -EPERM)
2185                         netif_warn(adapter, drv, adapter->netdev,
2186                                    "Cannot set host attributes\n");
2187                 else
2188                         netif_err(adapter, drv, adapter->netdev,
2189                                   "Cannot set host attributes\n");
2190                 goto err;
2191         }
2192
2193         return;
2194 err:
2195         ena_com_delete_debug_area(adapter->ena_dev);
2196 }
2197
2198 static void ena_get_stats64(struct net_device *netdev,
2199                             struct rtnl_link_stats64 *stats)
2200 {
2201         struct ena_adapter *adapter = netdev_priv(netdev);
2202         struct ena_ring *rx_ring, *tx_ring;
2203         unsigned int start;
2204         u64 rx_drops;
2205         int i;
2206
2207         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2208                 return;
2209
2210         for (i = 0; i < adapter->num_queues; i++) {
2211                 u64 bytes, packets;
2212
2213                 tx_ring = &adapter->tx_ring[i];
2214
2215                 do {
2216                         start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
2217                         packets = tx_ring->tx_stats.cnt;
2218                         bytes = tx_ring->tx_stats.bytes;
2219                 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
2220
2221                 stats->tx_packets += packets;
2222                 stats->tx_bytes += bytes;
2223
2224                 rx_ring = &adapter->rx_ring[i];
2225
2226                 do {
2227                         start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
2228                         packets = rx_ring->rx_stats.cnt;
2229                         bytes = rx_ring->rx_stats.bytes;
2230                 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
2231
2232                 stats->rx_packets += packets;
2233                 stats->rx_bytes += bytes;
2234         }
2235
2236         do {
2237                 start = u64_stats_fetch_begin_irq(&adapter->syncp);
2238                 rx_drops = adapter->dev_stats.rx_drops;
2239         } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
2240
2241         stats->rx_dropped = rx_drops;
2242
2243         stats->multicast = 0;
2244         stats->collisions = 0;
2245
2246         stats->rx_length_errors = 0;
2247         stats->rx_crc_errors = 0;
2248         stats->rx_frame_errors = 0;
2249         stats->rx_fifo_errors = 0;
2250         stats->rx_missed_errors = 0;
2251         stats->tx_window_errors = 0;
2252
2253         stats->rx_errors = 0;
2254         stats->tx_errors = 0;
2255 }
2256
2257 static const struct net_device_ops ena_netdev_ops = {
2258         .ndo_open               = ena_open,
2259         .ndo_stop               = ena_close,
2260         .ndo_start_xmit         = ena_start_xmit,
2261         .ndo_select_queue       = ena_select_queue,
2262         .ndo_get_stats64        = ena_get_stats64,
2263         .ndo_tx_timeout         = ena_tx_timeout,
2264         .ndo_change_mtu         = ena_change_mtu,
2265         .ndo_set_mac_address    = NULL,
2266         .ndo_validate_addr      = eth_validate_addr,
2267 #ifdef CONFIG_NET_POLL_CONTROLLER
2268         .ndo_poll_controller    = ena_netpoll,
2269 #endif /* CONFIG_NET_POLL_CONTROLLER */
2270 };
2271
2272 static void ena_device_io_suspend(struct work_struct *work)
2273 {
2274         struct ena_adapter *adapter =
2275                 container_of(work, struct ena_adapter, suspend_io_task);
2276         struct net_device *netdev = adapter->netdev;
2277
2278         /* ena_napi_disable_all disables only the IO handling.
2279          * We are still subject to AENQ keep alive watchdog.
2280          */
2281         u64_stats_update_begin(&adapter->syncp);
2282         adapter->dev_stats.io_suspend++;
2283         u64_stats_update_begin(&adapter->syncp);
2284         ena_napi_disable_all(adapter);
2285         netif_tx_lock(netdev);
2286         netif_device_detach(netdev);
2287         netif_tx_unlock(netdev);
2288 }
2289
2290 static void ena_device_io_resume(struct work_struct *work)
2291 {
2292         struct ena_adapter *adapter =
2293                 container_of(work, struct ena_adapter, resume_io_task);
2294         struct net_device *netdev = adapter->netdev;
2295
2296         u64_stats_update_begin(&adapter->syncp);
2297         adapter->dev_stats.io_resume++;
2298         u64_stats_update_end(&adapter->syncp);
2299
2300         netif_device_attach(netdev);
2301         ena_napi_enable_all(adapter);
2302 }
2303
2304 static int ena_device_validate_params(struct ena_adapter *adapter,
2305                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
2306 {
2307         struct net_device *netdev = adapter->netdev;
2308         int rc;
2309
2310         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
2311                               adapter->mac_addr);
2312         if (!rc) {
2313                 netif_err(adapter, drv, netdev,
2314                           "Error, mac address are different\n");
2315                 return -EINVAL;
2316         }
2317
2318         if ((get_feat_ctx->max_queues.max_cq_num < adapter->num_queues) ||
2319             (get_feat_ctx->max_queues.max_sq_num < adapter->num_queues)) {
2320                 netif_err(adapter, drv, netdev,
2321                           "Error, device doesn't support enough queues\n");
2322                 return -EINVAL;
2323         }
2324
2325         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
2326                 netif_err(adapter, drv, netdev,
2327                           "Error, device max mtu is smaller than netdev MTU\n");
2328                 return -EINVAL;
2329         }
2330
2331         return 0;
2332 }
2333
2334 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
2335                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
2336                            bool *wd_state)
2337 {
2338         struct device *dev = &pdev->dev;
2339         bool readless_supported;
2340         u32 aenq_groups;
2341         int dma_width;
2342         int rc;
2343
2344         rc = ena_com_mmio_reg_read_request_init(ena_dev);
2345         if (rc) {
2346                 dev_err(dev, "failed to init mmio read less\n");
2347                 return rc;
2348         }
2349
2350         /* The PCIe configuration space revision id indicate if mmio reg
2351          * read is disabled
2352          */
2353         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
2354         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
2355
2356         rc = ena_com_dev_reset(ena_dev);
2357         if (rc) {
2358                 dev_err(dev, "Can not reset device\n");
2359                 goto err_mmio_read_less;
2360         }
2361
2362         rc = ena_com_validate_version(ena_dev);
2363         if (rc) {
2364                 dev_err(dev, "device version is too low\n");
2365                 goto err_mmio_read_less;
2366         }
2367
2368         dma_width = ena_com_get_dma_width(ena_dev);
2369         if (dma_width < 0) {
2370                 dev_err(dev, "Invalid dma width value %d", dma_width);
2371                 rc = dma_width;
2372                 goto err_mmio_read_less;
2373         }
2374
2375         rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2376         if (rc) {
2377                 dev_err(dev, "pci_set_dma_mask failed 0x%x\n", rc);
2378                 goto err_mmio_read_less;
2379         }
2380
2381         rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2382         if (rc) {
2383                 dev_err(dev, "err_pci_set_consistent_dma_mask failed 0x%x\n",
2384                         rc);
2385                 goto err_mmio_read_less;
2386         }
2387
2388         /* ENA admin level init */
2389         rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
2390         if (rc) {
2391                 dev_err(dev,
2392                         "Can not initialize ena admin queue with device\n");
2393                 goto err_mmio_read_less;
2394         }
2395
2396         /* To enable the msix interrupts the driver needs to know the number
2397          * of queues. So the driver uses polling mode to retrieve this
2398          * information
2399          */
2400         ena_com_set_admin_polling_mode(ena_dev, true);
2401
2402         ena_config_host_info(ena_dev);
2403
2404         /* Get Device Attributes*/
2405         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
2406         if (rc) {
2407                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
2408                 goto err_admin_init;
2409         }
2410
2411         /* Try to turn all the available aenq groups */
2412         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
2413                 BIT(ENA_ADMIN_FATAL_ERROR) |
2414                 BIT(ENA_ADMIN_WARNING) |
2415                 BIT(ENA_ADMIN_NOTIFICATION) |
2416                 BIT(ENA_ADMIN_KEEP_ALIVE);
2417
2418         aenq_groups &= get_feat_ctx->aenq.supported_groups;
2419
2420         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
2421         if (rc) {
2422                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
2423                 goto err_admin_init;
2424         }
2425
2426         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
2427
2428         return 0;
2429
2430 err_admin_init:
2431         ena_com_delete_host_info(ena_dev);
2432         ena_com_admin_destroy(ena_dev);
2433 err_mmio_read_less:
2434         ena_com_mmio_reg_read_request_destroy(ena_dev);
2435
2436         return rc;
2437 }
2438
2439 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
2440                                                     int io_vectors)
2441 {
2442         struct ena_com_dev *ena_dev = adapter->ena_dev;
2443         struct device *dev = &adapter->pdev->dev;
2444         int rc;
2445
2446         rc = ena_enable_msix(adapter, io_vectors);
2447         if (rc) {
2448                 dev_err(dev, "Can not reserve msix vectors\n");
2449                 return rc;
2450         }
2451
2452         ena_setup_mgmnt_intr(adapter);
2453
2454         rc = ena_request_mgmnt_irq(adapter);
2455         if (rc) {
2456                 dev_err(dev, "Can not setup management interrupts\n");
2457                 goto err_disable_msix;
2458         }
2459
2460         ena_com_set_admin_polling_mode(ena_dev, false);
2461
2462         ena_com_admin_aenq_enable(ena_dev);
2463
2464         return 0;
2465
2466 err_disable_msix:
2467         pci_free_irq_vectors(adapter->pdev);
2468         return rc;
2469 }
2470
2471 static void ena_fw_reset_device(struct work_struct *work)
2472 {
2473         struct ena_com_dev_get_features_ctx get_feat_ctx;
2474         struct ena_adapter *adapter =
2475                 container_of(work, struct ena_adapter, reset_task);
2476         struct net_device *netdev = adapter->netdev;
2477         struct ena_com_dev *ena_dev = adapter->ena_dev;
2478         struct pci_dev *pdev = adapter->pdev;
2479         bool dev_up, wd_state;
2480         int rc;
2481
2482         if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2483                 dev_err(&pdev->dev,
2484                         "device reset schedule while reset bit is off\n");
2485                 return;
2486         }
2487
2488         netif_carrier_off(netdev);
2489
2490         del_timer_sync(&adapter->timer_service);
2491
2492         rtnl_lock();
2493
2494         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2495         ena_com_set_admin_running_state(ena_dev, false);
2496
2497         /* After calling ena_close the tx queues and the napi
2498          * are disabled so no one can interfere or touch the
2499          * data structures
2500          */
2501         ena_close(netdev);
2502
2503         ena_free_mgmnt_irq(adapter);
2504
2505         pci_free_irq_vectors(adapter->pdev);
2506
2507         ena_com_abort_admin_commands(ena_dev);
2508
2509         ena_com_wait_for_abort_completion(ena_dev);
2510
2511         ena_com_admin_destroy(ena_dev);
2512
2513         ena_com_mmio_reg_read_request_destroy(ena_dev);
2514
2515         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2516
2517         /* Finish with the destroy part. Start the init part */
2518
2519         rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
2520         if (rc) {
2521                 dev_err(&pdev->dev, "Can not initialize device\n");
2522                 goto err;
2523         }
2524         adapter->wd_state = wd_state;
2525
2526         rc = ena_device_validate_params(adapter, &get_feat_ctx);
2527         if (rc) {
2528                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
2529                 goto err_device_destroy;
2530         }
2531
2532         rc = ena_enable_msix_and_set_admin_interrupts(adapter,
2533                                                       adapter->num_queues);
2534         if (rc) {
2535                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
2536                 goto err_device_destroy;
2537         }
2538         /* If the interface was up before the reset bring it up */
2539         if (dev_up) {
2540                 rc = ena_up(adapter);
2541                 if (rc) {
2542                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
2543                         goto err_disable_msix;
2544                 }
2545         }
2546
2547         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
2548
2549         rtnl_unlock();
2550
2551         dev_err(&pdev->dev, "Device reset completed successfully\n");
2552
2553         return;
2554 err_disable_msix:
2555         ena_free_mgmnt_irq(adapter);
2556         pci_free_irq_vectors(adapter->pdev);
2557 err_device_destroy:
2558         ena_com_admin_destroy(ena_dev);
2559 err:
2560         rtnl_unlock();
2561
2562         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2563
2564         dev_err(&pdev->dev,
2565                 "Reset attempt failed. Can not reset the device\n");
2566 }
2567
2568 static int check_missing_comp_in_queue(struct ena_adapter *adapter,
2569                                        struct ena_ring *tx_ring)
2570 {
2571         struct ena_tx_buffer *tx_buf;
2572         unsigned long last_jiffies;
2573         u32 missed_tx = 0;
2574         int i;
2575
2576         for (i = 0; i < tx_ring->ring_size; i++) {
2577                 tx_buf = &tx_ring->tx_buffer_info[i];
2578                 last_jiffies = tx_buf->last_jiffies;
2579                 if (unlikely(last_jiffies &&
2580                              time_is_before_jiffies(last_jiffies + TX_TIMEOUT))) {
2581                         if (!tx_buf->print_once)
2582                                 netif_notice(adapter, tx_err, adapter->netdev,
2583                                              "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
2584                                              tx_ring->qid, i);
2585
2586                         tx_buf->print_once = 1;
2587                         missed_tx++;
2588
2589                         if (unlikely(missed_tx > MAX_NUM_OF_TIMEOUTED_PACKETS)) {
2590                                 netif_err(adapter, tx_err, adapter->netdev,
2591                                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
2592                                           missed_tx, MAX_NUM_OF_TIMEOUTED_PACKETS);
2593                                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2594                                 return -EIO;
2595                         }
2596                 }
2597         }
2598
2599         return 0;
2600 }
2601
2602 static void check_for_missing_tx_completions(struct ena_adapter *adapter)
2603 {
2604         struct ena_ring *tx_ring;
2605         int i, budget, rc;
2606
2607         /* Make sure the driver doesn't turn the device in other process */
2608         smp_rmb();
2609
2610         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2611                 return;
2612
2613         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2614                 return;
2615
2616         budget = ENA_MONITORED_TX_QUEUES;
2617
2618         for (i = adapter->last_monitored_tx_qid; i < adapter->num_queues; i++) {
2619                 tx_ring = &adapter->tx_ring[i];
2620
2621                 rc = check_missing_comp_in_queue(adapter, tx_ring);
2622                 if (unlikely(rc))
2623                         return;
2624
2625                 budget--;
2626                 if (!budget)
2627                         break;
2628         }
2629
2630         adapter->last_monitored_tx_qid = i % adapter->num_queues;
2631 }
2632
2633 /* trigger napi schedule after 2 consecutive detections */
2634 #define EMPTY_RX_REFILL 2
2635 /* For the rare case where the device runs out of Rx descriptors and the
2636  * napi handler failed to refill new Rx descriptors (due to a lack of memory
2637  * for example).
2638  * This case will lead to a deadlock:
2639  * The device won't send interrupts since all the new Rx packets will be dropped
2640  * The napi handler won't allocate new Rx descriptors so the device will be
2641  * able to send new packets.
2642  *
2643  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
2644  * It is recommended to have at least 512MB, with a minimum of 128MB for
2645  * constrained environment).
2646  *
2647  * When such a situation is detected - Reschedule napi
2648  */
2649 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
2650 {
2651         struct ena_ring *rx_ring;
2652         int i, refill_required;
2653
2654         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2655                 return;
2656
2657         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2658                 return;
2659
2660         for (i = 0; i < adapter->num_queues; i++) {
2661                 rx_ring = &adapter->rx_ring[i];
2662
2663                 refill_required =
2664                         ena_com_sq_empty_space(rx_ring->ena_com_io_sq);
2665                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
2666                         rx_ring->empty_rx_queue++;
2667
2668                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
2669                                 u64_stats_update_begin(&rx_ring->syncp);
2670                                 rx_ring->rx_stats.empty_rx_ring++;
2671                                 u64_stats_update_end(&rx_ring->syncp);
2672
2673                                 netif_err(adapter, drv, adapter->netdev,
2674                                           "trigger refill for ring %d\n", i);
2675
2676                                 napi_schedule(rx_ring->napi);
2677                                 rx_ring->empty_rx_queue = 0;
2678                         }
2679                 } else {
2680                         rx_ring->empty_rx_queue = 0;
2681                 }
2682         }
2683 }
2684
2685 /* Check for keep alive expiration */
2686 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
2687 {
2688         unsigned long keep_alive_expired;
2689
2690         if (!adapter->wd_state)
2691                 return;
2692
2693         keep_alive_expired = round_jiffies(adapter->last_keep_alive_jiffies
2694                                            + ENA_DEVICE_KALIVE_TIMEOUT);
2695         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
2696                 netif_err(adapter, drv, adapter->netdev,
2697                           "Keep alive watchdog timeout.\n");
2698                 u64_stats_update_begin(&adapter->syncp);
2699                 adapter->dev_stats.wd_expired++;
2700                 u64_stats_update_end(&adapter->syncp);
2701                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2702         }
2703 }
2704
2705 static void check_for_admin_com_state(struct ena_adapter *adapter)
2706 {
2707         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
2708                 netif_err(adapter, drv, adapter->netdev,
2709                           "ENA admin queue is not in running state!\n");
2710                 u64_stats_update_begin(&adapter->syncp);
2711                 adapter->dev_stats.admin_q_pause++;
2712                 u64_stats_update_end(&adapter->syncp);
2713                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2714         }
2715 }
2716
2717 static void ena_update_host_info(struct ena_admin_host_info *host_info,
2718                                  struct net_device *netdev)
2719 {
2720         host_info->supported_network_features[0] =
2721                 netdev->features & GENMASK_ULL(31, 0);
2722         host_info->supported_network_features[1] =
2723                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
2724 }
2725
2726 static void ena_timer_service(unsigned long data)
2727 {
2728         struct ena_adapter *adapter = (struct ena_adapter *)data;
2729         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
2730         struct ena_admin_host_info *host_info =
2731                 adapter->ena_dev->host_attr.host_info;
2732
2733         check_for_missing_keep_alive(adapter);
2734
2735         check_for_admin_com_state(adapter);
2736
2737         check_for_missing_tx_completions(adapter);
2738
2739         check_for_empty_rx_ring(adapter);
2740
2741         if (debug_area)
2742                 ena_dump_stats_to_buf(adapter, debug_area);
2743
2744         if (host_info)
2745                 ena_update_host_info(host_info, adapter->netdev);
2746
2747         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2748                 netif_err(adapter, drv, adapter->netdev,
2749                           "Trigger reset is on\n");
2750                 ena_dump_stats_to_dmesg(adapter);
2751                 queue_work(ena_wq, &adapter->reset_task);
2752                 return;
2753         }
2754
2755         /* Reset the timer */
2756         mod_timer(&adapter->timer_service, jiffies + HZ);
2757 }
2758
2759 static int ena_calc_io_queue_num(struct pci_dev *pdev,
2760                                  struct ena_com_dev *ena_dev,
2761                                  struct ena_com_dev_get_features_ctx *get_feat_ctx)
2762 {
2763         int io_sq_num, io_queue_num;
2764
2765         /* In case of LLQ use the llq number in the get feature cmd */
2766         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2767                 io_sq_num = get_feat_ctx->max_queues.max_llq_num;
2768
2769                 if (io_sq_num == 0) {
2770                         dev_err(&pdev->dev,
2771                                 "Trying to use LLQ but llq_num is 0. Fall back into regular queues\n");
2772
2773                         ena_dev->tx_mem_queue_type =
2774                                 ENA_ADMIN_PLACEMENT_POLICY_HOST;
2775                         io_sq_num = get_feat_ctx->max_queues.max_sq_num;
2776                 }
2777         } else {
2778                 io_sq_num = get_feat_ctx->max_queues.max_sq_num;
2779         }
2780
2781         io_queue_num = min_t(int, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
2782         io_queue_num = min_t(int, io_queue_num, io_sq_num);
2783         io_queue_num = min_t(int, io_queue_num,
2784                              get_feat_ctx->max_queues.max_cq_num);
2785         /* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
2786         io_queue_num = min_t(int, io_queue_num, pci_msix_vec_count(pdev) - 1);
2787         if (unlikely(!io_queue_num)) {
2788                 dev_err(&pdev->dev, "The device doesn't have io queues\n");
2789                 return -EFAULT;
2790         }
2791
2792         return io_queue_num;
2793 }
2794
2795 static void ena_set_push_mode(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
2796                               struct ena_com_dev_get_features_ctx *get_feat_ctx)
2797 {
2798         bool has_mem_bar;
2799
2800         has_mem_bar = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(ENA_MEM_BAR);
2801
2802         /* Enable push mode if device supports LLQ */
2803         if (has_mem_bar && (get_feat_ctx->max_queues.max_llq_num > 0))
2804                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
2805         else
2806                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2807 }
2808
2809 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
2810                                  struct net_device *netdev)
2811 {
2812         netdev_features_t dev_features = 0;
2813
2814         /* Set offload features */
2815         if (feat->offload.tx &
2816                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
2817                 dev_features |= NETIF_F_IP_CSUM;
2818
2819         if (feat->offload.tx &
2820                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
2821                 dev_features |= NETIF_F_IPV6_CSUM;
2822
2823         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
2824                 dev_features |= NETIF_F_TSO;
2825
2826         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
2827                 dev_features |= NETIF_F_TSO6;
2828
2829         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
2830                 dev_features |= NETIF_F_TSO_ECN;
2831
2832         if (feat->offload.rx_supported &
2833                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
2834                 dev_features |= NETIF_F_RXCSUM;
2835
2836         if (feat->offload.rx_supported &
2837                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
2838                 dev_features |= NETIF_F_RXCSUM;
2839
2840         netdev->features =
2841                 dev_features |
2842                 NETIF_F_SG |
2843                 NETIF_F_RXHASH |
2844                 NETIF_F_HIGHDMA;
2845
2846         netdev->hw_features |= netdev->features;
2847         netdev->vlan_features |= netdev->features;
2848 }
2849
2850 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
2851                                      struct ena_com_dev_get_features_ctx *feat)
2852 {
2853         struct net_device *netdev = adapter->netdev;
2854
2855         /* Copy mac address */
2856         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
2857                 eth_hw_addr_random(netdev);
2858                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
2859         } else {
2860                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
2861                 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
2862         }
2863
2864         /* Set offload features */
2865         ena_set_dev_offloads(feat, netdev);
2866
2867         adapter->max_mtu = feat->dev_attr.max_mtu;
2868         netdev->max_mtu = adapter->max_mtu;
2869         netdev->min_mtu = ENA_MIN_MTU;
2870 }
2871
2872 static int ena_rss_init_default(struct ena_adapter *adapter)
2873 {
2874         struct ena_com_dev *ena_dev = adapter->ena_dev;
2875         struct device *dev = &adapter->pdev->dev;
2876         int rc, i;
2877         u32 val;
2878
2879         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
2880         if (unlikely(rc)) {
2881                 dev_err(dev, "Cannot init indirect table\n");
2882                 goto err_rss_init;
2883         }
2884
2885         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
2886                 val = ethtool_rxfh_indir_default(i, adapter->num_queues);
2887                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
2888                                                        ENA_IO_RXQ_IDX(val));
2889                 if (unlikely(rc && (rc != -EPERM))) {
2890                         dev_err(dev, "Cannot fill indirect table\n");
2891                         goto err_fill_indir;
2892                 }
2893         }
2894
2895         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
2896                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
2897         if (unlikely(rc && (rc != -EPERM))) {
2898                 dev_err(dev, "Cannot fill hash function\n");
2899                 goto err_fill_indir;
2900         }
2901
2902         rc = ena_com_set_default_hash_ctrl(ena_dev);
2903         if (unlikely(rc && (rc != -EPERM))) {
2904                 dev_err(dev, "Cannot fill hash control\n");
2905                 goto err_fill_indir;
2906         }
2907
2908         return 0;
2909
2910 err_fill_indir:
2911         ena_com_rss_destroy(ena_dev);
2912 err_rss_init:
2913
2914         return rc;
2915 }
2916
2917 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
2918 {
2919         int release_bars;
2920
2921         if (ena_dev->mem_bar)
2922                 devm_iounmap(&pdev->dev, ena_dev->mem_bar);
2923
2924         devm_iounmap(&pdev->dev, ena_dev->reg_bar);
2925
2926         release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
2927         pci_release_selected_regions(pdev, release_bars);
2928 }
2929
2930 static int ena_calc_queue_size(struct pci_dev *pdev,
2931                                struct ena_com_dev *ena_dev,
2932                                u16 *max_tx_sgl_size,
2933                                u16 *max_rx_sgl_size,
2934                                struct ena_com_dev_get_features_ctx *get_feat_ctx)
2935 {
2936         u32 queue_size = ENA_DEFAULT_RING_SIZE;
2937
2938         queue_size = min_t(u32, queue_size,
2939                            get_feat_ctx->max_queues.max_cq_depth);
2940         queue_size = min_t(u32, queue_size,
2941                            get_feat_ctx->max_queues.max_sq_depth);
2942
2943         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
2944                 queue_size = min_t(u32, queue_size,
2945                                    get_feat_ctx->max_queues.max_llq_depth);
2946
2947         queue_size = rounddown_pow_of_two(queue_size);
2948
2949         if (unlikely(!queue_size)) {
2950                 dev_err(&pdev->dev, "Invalid queue size\n");
2951                 return -EFAULT;
2952         }
2953
2954         *max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
2955                                  get_feat_ctx->max_queues.max_packet_tx_descs);
2956         *max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
2957                                  get_feat_ctx->max_queues.max_packet_rx_descs);
2958
2959         return queue_size;
2960 }
2961
2962 /* ena_probe - Device Initialization Routine
2963  * @pdev: PCI device information struct
2964  * @ent: entry in ena_pci_tbl
2965  *
2966  * Returns 0 on success, negative on failure
2967  *
2968  * ena_probe initializes an adapter identified by a pci_dev structure.
2969  * The OS initialization, configuring of the adapter private structure,
2970  * and a hardware reset occur.
2971  */
2972 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2973 {
2974         struct ena_com_dev_get_features_ctx get_feat_ctx;
2975         static int version_printed;
2976         struct net_device *netdev;
2977         struct ena_adapter *adapter;
2978         struct ena_com_dev *ena_dev = NULL;
2979         static int adapters_found;
2980         int io_queue_num, bars, rc;
2981         int queue_size;
2982         u16 tx_sgl_size = 0;
2983         u16 rx_sgl_size = 0;
2984         bool wd_state;
2985
2986         dev_dbg(&pdev->dev, "%s\n", __func__);
2987
2988         if (version_printed++ == 0)
2989                 dev_info(&pdev->dev, "%s", version);
2990
2991         rc = pci_enable_device_mem(pdev);
2992         if (rc) {
2993                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
2994                 return rc;
2995         }
2996
2997         pci_set_master(pdev);
2998
2999         ena_dev = vzalloc(sizeof(*ena_dev));
3000         if (!ena_dev) {
3001                 rc = -ENOMEM;
3002                 goto err_disable_device;
3003         }
3004
3005         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
3006         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
3007         if (rc) {
3008                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
3009                         rc);
3010                 goto err_free_ena_dev;
3011         }
3012
3013         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
3014                                         pci_resource_start(pdev, ENA_REG_BAR),
3015                                         pci_resource_len(pdev, ENA_REG_BAR));
3016         if (!ena_dev->reg_bar) {
3017                 dev_err(&pdev->dev, "failed to remap regs bar\n");
3018                 rc = -EFAULT;
3019                 goto err_free_region;
3020         }
3021
3022         ena_dev->dmadev = &pdev->dev;
3023
3024         rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
3025         if (rc) {
3026                 dev_err(&pdev->dev, "ena device init failed\n");
3027                 if (rc == -ETIME)
3028                         rc = -EPROBE_DEFER;
3029                 goto err_free_region;
3030         }
3031
3032         ena_set_push_mode(pdev, ena_dev, &get_feat_ctx);
3033
3034         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3035                 ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3036                                                    pci_resource_start(pdev, ENA_MEM_BAR),
3037                                                    pci_resource_len(pdev, ENA_MEM_BAR));
3038                 if (!ena_dev->mem_bar) {
3039                         rc = -EFAULT;
3040                         goto err_device_destroy;
3041                 }
3042         }
3043
3044         /* initial Tx interrupt delay, Assumes 1 usec granularity.
3045         * Updated during device initialization with the real granularity
3046         */
3047         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
3048         io_queue_num = ena_calc_io_queue_num(pdev, ena_dev, &get_feat_ctx);
3049         queue_size = ena_calc_queue_size(pdev, ena_dev, &tx_sgl_size,
3050                                          &rx_sgl_size, &get_feat_ctx);
3051         if ((queue_size <= 0) || (io_queue_num <= 0)) {
3052                 rc = -EFAULT;
3053                 goto err_device_destroy;
3054         }
3055
3056         dev_info(&pdev->dev, "creating %d io queues. queue size: %d\n",
3057                  io_queue_num, queue_size);
3058
3059         /* dev zeroed in init_etherdev */
3060         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), io_queue_num);
3061         if (!netdev) {
3062                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
3063                 rc = -ENOMEM;
3064                 goto err_device_destroy;
3065         }
3066
3067         SET_NETDEV_DEV(netdev, &pdev->dev);
3068
3069         adapter = netdev_priv(netdev);
3070         pci_set_drvdata(pdev, adapter);
3071
3072         adapter->ena_dev = ena_dev;
3073         adapter->netdev = netdev;
3074         adapter->pdev = pdev;
3075
3076         ena_set_conf_feat_params(adapter, &get_feat_ctx);
3077
3078         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3079
3080         adapter->tx_ring_size = queue_size;
3081         adapter->rx_ring_size = queue_size;
3082
3083         adapter->max_tx_sgl_size = tx_sgl_size;
3084         adapter->max_rx_sgl_size = rx_sgl_size;
3085
3086         adapter->num_queues = io_queue_num;
3087         adapter->last_monitored_tx_qid = 0;
3088
3089         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
3090         adapter->wd_state = wd_state;
3091
3092         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
3093
3094         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
3095         if (rc) {
3096                 dev_err(&pdev->dev,
3097                         "Failed to query interrupt moderation feature\n");
3098                 goto err_netdev_destroy;
3099         }
3100         ena_init_io_rings(adapter);
3101
3102         netdev->netdev_ops = &ena_netdev_ops;
3103         netdev->watchdog_timeo = TX_TIMEOUT;
3104         ena_set_ethtool_ops(netdev);
3105
3106         netdev->priv_flags |= IFF_UNICAST_FLT;
3107
3108         u64_stats_init(&adapter->syncp);
3109
3110         rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3111         if (rc) {
3112                 dev_err(&pdev->dev,
3113                         "Failed to enable and set the admin interrupts\n");
3114                 goto err_worker_destroy;
3115         }
3116         rc = ena_rss_init_default(adapter);
3117         if (rc && (rc != -EPERM)) {
3118                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
3119                 goto err_free_msix;
3120         }
3121
3122         ena_config_debug_area(adapter);
3123
3124         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
3125
3126         netif_carrier_off(netdev);
3127
3128         rc = register_netdev(netdev);
3129         if (rc) {
3130                 dev_err(&pdev->dev, "Cannot register net device\n");
3131                 goto err_rss;
3132         }
3133
3134         INIT_WORK(&adapter->suspend_io_task, ena_device_io_suspend);
3135         INIT_WORK(&adapter->resume_io_task, ena_device_io_resume);
3136         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
3137
3138         adapter->last_keep_alive_jiffies = jiffies;
3139
3140         setup_timer(&adapter->timer_service, ena_timer_service,
3141                     (unsigned long)adapter);
3142         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3143
3144         dev_info(&pdev->dev, "%s found at mem %lx, mac addr %pM Queues %d\n",
3145                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
3146                  netdev->dev_addr, io_queue_num);
3147
3148         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3149
3150         adapters_found++;
3151
3152         return 0;
3153
3154 err_rss:
3155         ena_com_delete_debug_area(ena_dev);
3156         ena_com_rss_destroy(ena_dev);
3157 err_free_msix:
3158         ena_com_dev_reset(ena_dev);
3159         ena_free_mgmnt_irq(adapter);
3160         pci_free_irq_vectors(adapter->pdev);
3161 err_worker_destroy:
3162         ena_com_destroy_interrupt_moderation(ena_dev);
3163         del_timer(&adapter->timer_service);
3164         cancel_work_sync(&adapter->suspend_io_task);
3165         cancel_work_sync(&adapter->resume_io_task);
3166 err_netdev_destroy:
3167         free_netdev(netdev);
3168 err_device_destroy:
3169         ena_com_delete_host_info(ena_dev);
3170         ena_com_admin_destroy(ena_dev);
3171 err_free_region:
3172         ena_release_bars(ena_dev, pdev);
3173 err_free_ena_dev:
3174         vfree(ena_dev);
3175 err_disable_device:
3176         pci_disable_device(pdev);
3177         return rc;
3178 }
3179
3180 /*****************************************************************************/
3181 static int ena_sriov_configure(struct pci_dev *dev, int numvfs)
3182 {
3183         int rc;
3184
3185         if (numvfs > 0) {
3186                 rc = pci_enable_sriov(dev, numvfs);
3187                 if (rc != 0) {
3188                         dev_err(&dev->dev,
3189                                 "pci_enable_sriov failed to enable: %d vfs with the error: %d\n",
3190                                 numvfs, rc);
3191                         return rc;
3192                 }
3193
3194                 return numvfs;
3195         }
3196
3197         if (numvfs == 0) {
3198                 pci_disable_sriov(dev);
3199                 return 0;
3200         }
3201
3202         return -EINVAL;
3203 }
3204
3205 /*****************************************************************************/
3206 /*****************************************************************************/
3207
3208 /* ena_remove - Device Removal Routine
3209  * @pdev: PCI device information struct
3210  *
3211  * ena_remove is called by the PCI subsystem to alert the driver
3212  * that it should release a PCI device.
3213  */
3214 static void ena_remove(struct pci_dev *pdev)
3215 {
3216         struct ena_adapter *adapter = pci_get_drvdata(pdev);
3217         struct ena_com_dev *ena_dev;
3218         struct net_device *netdev;
3219
3220         ena_dev = adapter->ena_dev;
3221         netdev = adapter->netdev;
3222
3223 #ifdef CONFIG_RFS_ACCEL
3224         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
3225                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
3226                 netdev->rx_cpu_rmap = NULL;
3227         }
3228 #endif /* CONFIG_RFS_ACCEL */
3229
3230         unregister_netdev(netdev);
3231         del_timer_sync(&adapter->timer_service);
3232
3233         cancel_work_sync(&adapter->reset_task);
3234
3235         cancel_work_sync(&adapter->suspend_io_task);
3236
3237         cancel_work_sync(&adapter->resume_io_task);
3238
3239         /* Reset the device only if the device is running. */
3240         if (test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3241                 ena_com_dev_reset(ena_dev);
3242
3243         ena_free_mgmnt_irq(adapter);
3244
3245         pci_free_irq_vectors(adapter->pdev);
3246
3247         free_netdev(netdev);
3248
3249         ena_com_mmio_reg_read_request_destroy(ena_dev);
3250
3251         ena_com_abort_admin_commands(ena_dev);
3252
3253         ena_com_wait_for_abort_completion(ena_dev);
3254
3255         ena_com_admin_destroy(ena_dev);
3256
3257         ena_com_rss_destroy(ena_dev);
3258
3259         ena_com_delete_debug_area(ena_dev);
3260
3261         ena_com_delete_host_info(ena_dev);
3262
3263         ena_release_bars(ena_dev, pdev);
3264
3265         pci_disable_device(pdev);
3266
3267         ena_com_destroy_interrupt_moderation(ena_dev);
3268
3269         vfree(ena_dev);
3270 }
3271
3272 static struct pci_driver ena_pci_driver = {
3273         .name           = DRV_MODULE_NAME,
3274         .id_table       = ena_pci_tbl,
3275         .probe          = ena_probe,
3276         .remove         = ena_remove,
3277         .sriov_configure = ena_sriov_configure,
3278 };
3279
3280 static int __init ena_init(void)
3281 {
3282         pr_info("%s", version);
3283
3284         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
3285         if (!ena_wq) {
3286                 pr_err("Failed to create workqueue\n");
3287                 return -ENOMEM;
3288         }
3289
3290         return pci_register_driver(&ena_pci_driver);
3291 }
3292
3293 static void __exit ena_cleanup(void)
3294 {
3295         pci_unregister_driver(&ena_pci_driver);
3296
3297         if (ena_wq) {
3298                 destroy_workqueue(ena_wq);
3299                 ena_wq = NULL;
3300         }
3301 }
3302
3303 /******************************************************************************
3304  ******************************** AENQ Handlers *******************************
3305  *****************************************************************************/
3306 /* ena_update_on_link_change:
3307  * Notify the network interface about the change in link status
3308  */
3309 static void ena_update_on_link_change(void *adapter_data,
3310                                       struct ena_admin_aenq_entry *aenq_e)
3311 {
3312         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3313         struct ena_admin_aenq_link_change_desc *aenq_desc =
3314                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
3315         int status = aenq_desc->flags &
3316                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3317
3318         if (status) {
3319                 netdev_dbg(adapter->netdev, "%s\n", __func__);
3320                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3321                 netif_carrier_on(adapter->netdev);
3322         } else {
3323                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3324                 netif_carrier_off(adapter->netdev);
3325         }
3326 }
3327
3328 static void ena_keep_alive_wd(void *adapter_data,
3329                               struct ena_admin_aenq_entry *aenq_e)
3330 {
3331         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3332
3333         adapter->last_keep_alive_jiffies = jiffies;
3334 }
3335
3336 static void ena_notification(void *adapter_data,
3337                              struct ena_admin_aenq_entry *aenq_e)
3338 {
3339         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3340
3341         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
3342              "Invalid group(%x) expected %x\n",
3343              aenq_e->aenq_common_desc.group,
3344              ENA_ADMIN_NOTIFICATION);
3345
3346         switch (aenq_e->aenq_common_desc.syndrom) {
3347         case ENA_ADMIN_SUSPEND:
3348                 /* Suspend just the IO queues.
3349                  * We deliberately don't suspend admin so the timer and
3350                  * the keep_alive events should remain.
3351                  */
3352                 queue_work(ena_wq, &adapter->suspend_io_task);
3353                 break;
3354         case ENA_ADMIN_RESUME:
3355                 queue_work(ena_wq, &adapter->resume_io_task);
3356                 break;
3357         default:
3358                 netif_err(adapter, drv, adapter->netdev,
3359                           "Invalid aenq notification link state %d\n",
3360                           aenq_e->aenq_common_desc.syndrom);
3361         }
3362 }
3363
3364 /* This handler will called for unknown event group or unimplemented handlers*/
3365 static void unimplemented_aenq_handler(void *data,
3366                                        struct ena_admin_aenq_entry *aenq_e)
3367 {
3368         struct ena_adapter *adapter = (struct ena_adapter *)data;
3369
3370         netif_err(adapter, drv, adapter->netdev,
3371                   "Unknown event was received or event with unimplemented handler\n");
3372 }
3373
3374 static struct ena_aenq_handlers aenq_handlers = {
3375         .handlers = {
3376                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3377                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
3378                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3379         },
3380         .unimplemented_handler = unimplemented_aenq_handler
3381 };
3382
3383 module_init(ena_init);
3384 module_exit(ena_cleanup);