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

   1 /*
   2  * Copyright 2015 Amazon.com, Inc. or its affiliates.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  34
  35 #ifdef CONFIG_RFS_ACCEL
  36 #include <linux/cpu_rmap.h>
  37 #endif /* CONFIG_RFS_ACCEL */
  38 #include <linux/ethtool.h>
  39 #include <linux/if_vlan.h>
  40 #include <linux/kernel.h>
  41 #include <linux/module.h>
  42 #include <linux/moduleparam.h>
  43 #include <linux/numa.h>
  44 #include <linux/pci.h>
  45 #include <linux/utsname.h>
  46 #include <linux/version.h>
  47 #include <linux/vmalloc.h>
  48 #include <net/ip.h>
  49
  50 #include "ena_netdev.h"
  51 #include "ena_pci_id_tbl.h"
  52
  53 static char version[] = DEVICE_NAME " v" DRV_MODULE_VERSION "\n";
  54
  55 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
  56 MODULE_DESCRIPTION(DEVICE_NAME);
  57 MODULE_LICENSE("GPL");
  58 MODULE_VERSION(DRV_MODULE_VERSION);
  59
  60 /* Time in jiffies before concluding the transmitter is hung. */
  61 #define TX_TIMEOUT  (5 * HZ)
  62
  63 #define ENA_NAPI_BUDGET 64
  64
  65 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
  66                 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
  67 static int debug = -1;
  68 module_param(debug, int, 0);
  69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  70
  71 static struct ena_aenq_handlers aenq_handlers;
  72
  73 static struct workqueue_struct *ena_wq;
  74
  75 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
  76
  77 static int ena_rss_init_default(struct ena_adapter *adapter);
  78 static void check_for_admin_com_state(struct ena_adapter *adapter);
  79 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
  80 static int ena_restore_device(struct ena_adapter *adapter);
  81
  82 static void ena_tx_timeout(struct net_device *dev)
  83 {
  84         struct ena_adapter *adapter = netdev_priv(dev);
  85
  86         /* Change the state of the device to trigger reset
  87          * Check that we are not in the middle or a trigger already
  88          */
  89
  90         if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
  91                 return;
  92
  93         adapter->reset_reason = ENA_REGS_RESET_OS_NETDEV_WD;
  94         u64_stats_update_begin(&adapter->syncp);
  95         adapter->dev_stats.tx_timeout++;
  96         u64_stats_update_end(&adapter->syncp);
  97
  98         netif_err(adapter, tx_err, dev, "Transmit time out\n");
  99 }
 100
 101 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
 102 {
 103         int i;
 104
 105         for (i = 0; i < adapter->num_queues; i++)
 106                 adapter->rx_ring[i].mtu = mtu;
 107 }
 108
 109 static int ena_change_mtu(struct net_device *dev, int new_mtu)
 110 {
 111         struct ena_adapter *adapter = netdev_priv(dev);
 112         int ret;
 113
 114         ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
 115         if (!ret) {
 116                 netif_dbg(adapter, drv, dev, "set MTU to %d\n", new_mtu);
 117                 update_rx_ring_mtu(adapter, new_mtu);
 118                 dev->mtu = new_mtu;
 119         } else {
 120                 netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
 121                           new_mtu);
 122         }
 123
 124         return ret;
 125 }
 126
 127 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
 128 {
 129 #ifdef CONFIG_RFS_ACCEL
 130         u32 i;
 131         int rc;
 132
 133         adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_queues);
 134         if (!adapter->netdev->rx_cpu_rmap)
 135                 return -ENOMEM;
 136         for (i = 0; i < adapter->num_queues; i++) {
 137                 int irq_idx = ENA_IO_IRQ_IDX(i);
 138
 139                 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
 140                                       pci_irq_vector(adapter->pdev, irq_idx));
 141                 if (rc) {
 142                         free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
 143                         adapter->netdev->rx_cpu_rmap = NULL;
 144                         return rc;
 145                 }
 146         }
 147 #endif /* CONFIG_RFS_ACCEL */
 148         return 0;
 149 }
 150
 151 static void ena_init_io_rings_common(struct ena_adapter *adapter,
 152                                      struct ena_ring *ring, u16 qid)
 153 {
 154         ring->qid = qid;
 155         ring->pdev = adapter->pdev;
 156         ring->dev = &adapter->pdev->dev;
 157         ring->netdev = adapter->netdev;
 158         ring->napi = &adapter->ena_napi[qid].napi;
 159         ring->adapter = adapter;
 160         ring->ena_dev = adapter->ena_dev;
 161         ring->per_napi_packets = 0;
 162         ring->per_napi_bytes = 0;
 163         ring->cpu = 0;
 164         ring->first_interrupt = false;
 165         ring->no_interrupt_event_cnt = 0;
 166         u64_stats_init(&ring->syncp);
 167 }
 168
 169 static void ena_init_io_rings(struct ena_adapter *adapter)
 170 {
 171         struct ena_com_dev *ena_dev;
 172         struct ena_ring *txr, *rxr;
 173         int i;
 174
 175         ena_dev = adapter->ena_dev;
 176
 177         for (i = 0; i < adapter->num_queues; i++) {
 178                 txr = &adapter->tx_ring[i];
 179                 rxr = &adapter->rx_ring[i];
 180
 181                 /* TX/RX common ring state */
 182                 ena_init_io_rings_common(adapter, txr, i);
 183                 ena_init_io_rings_common(adapter, rxr, i);
 184
 185                 /* TX specific ring state */
 186                 txr->ring_size = adapter->tx_ring_size;
 187                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
 188                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
 189                 txr->sgl_size = adapter->max_tx_sgl_size;
 190                 txr->smoothed_interval =
 191                         ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
 192
 193                 /* RX specific ring state */
 194                 rxr->ring_size = adapter->rx_ring_size;
 195                 rxr->rx_copybreak = adapter->rx_copybreak;
 196                 rxr->sgl_size = adapter->max_rx_sgl_size;
 197                 rxr->smoothed_interval =
 198                         ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
 199                 rxr->empty_rx_queue = 0;
 200         }
 201 }
 202
 203 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
 204  * @adapter: network interface device structure
 205  * @qid: queue index
 206  *
 207  * Return 0 on success, negative on failure
 208  */
 209 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
 210 {
 211         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 212         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 213         int size, i, node;
 214
 215         if (tx_ring->tx_buffer_info) {
 216                 netif_err(adapter, ifup,
 217                           adapter->netdev, "tx_buffer_info info is not NULL");
 218                 return -EEXIST;
 219         }
 220
 221         size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
 222         node = cpu_to_node(ena_irq->cpu);
 223
 224         tx_ring->tx_buffer_info = vzalloc_node(size, node);
 225         if (!tx_ring->tx_buffer_info) {
 226                 tx_ring->tx_buffer_info = vzalloc(size);
 227                 if (!tx_ring->tx_buffer_info)
 228                         return -ENOMEM;
 229         }
 230
 231         size = sizeof(u16) * tx_ring->ring_size;
 232         tx_ring->free_tx_ids = vzalloc_node(size, node);
 233         if (!tx_ring->free_tx_ids) {
 234                 tx_ring->free_tx_ids = vzalloc(size);
 235                 if (!tx_ring->free_tx_ids) {
 236                         vfree(tx_ring->tx_buffer_info);
 237                         return -ENOMEM;
 238                 }
 239         }
 240
 241         /* Req id ring for TX out of order completions */
 242         for (i = 0; i < tx_ring->ring_size; i++)
 243                 tx_ring->free_tx_ids[i] = i;
 244
 245         /* Reset tx statistics */
 246         memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
 247
 248         tx_ring->next_to_use = 0;
 249         tx_ring->next_to_clean = 0;
 250         tx_ring->cpu = ena_irq->cpu;
 251         return 0;
 252 }
 253
 254 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
 255  * @adapter: network interface device structure
 256  * @qid: queue index
 257  *
 258  * Free all transmit software resources
 259  */
 260 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
 261 {
 262         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 263
 264         vfree(tx_ring->tx_buffer_info);
 265         tx_ring->tx_buffer_info = NULL;
 266
 267         vfree(tx_ring->free_tx_ids);
 268         tx_ring->free_tx_ids = NULL;
 269 }
 270
 271 /* ena_setup_all_tx_resources - allocate I/O Tx queues resources for All queues
 272  * @adapter: private structure
 273  *
 274  * Return 0 on success, negative on failure
 275  */
 276 static int ena_setup_all_tx_resources(struct ena_adapter *adapter)
 277 {
 278         int i, rc = 0;
 279
 280         for (i = 0; i < adapter->num_queues; i++) {
 281                 rc = ena_setup_tx_resources(adapter, i);
 282                 if (rc)
 283                         goto err_setup_tx;
 284         }
 285
 286         return 0;
 287
 288 err_setup_tx:
 289
 290         netif_err(adapter, ifup, adapter->netdev,
 291                   "Tx queue %d: allocation failed\n", i);
 292
 293         /* rewind the index freeing the rings as we go */
 294         while (i--)
 295                 ena_free_tx_resources(adapter, i);
 296         return rc;
 297 }
 298
 299 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
 300  * @adapter: board private structure
 301  *
 302  * Free all transmit software resources
 303  */
 304 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
 305 {
 306         int i;
 307
 308         for (i = 0; i < adapter->num_queues; i++)
 309                 ena_free_tx_resources(adapter, i);
 310 }
 311
 312 static inline int validate_rx_req_id(struct ena_ring *rx_ring, u16 req_id)
 313 {
 314         if (likely(req_id < rx_ring->ring_size))
 315                 return 0;
 316
 317         netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 318                   "Invalid rx req_id: %hu\n", req_id);
 319
 320         u64_stats_update_begin(&rx_ring->syncp);
 321         rx_ring->rx_stats.bad_req_id++;
 322         u64_stats_update_end(&rx_ring->syncp);
 323
 324         /* Trigger device reset */
 325         rx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
 326         set_bit(ENA_FLAG_TRIGGER_RESET, &rx_ring->adapter->flags);
 327         return -EFAULT;
 328 }
 329
 330 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
 331  * @adapter: network interface device structure
 332  * @qid: queue index
 333  *
 334  * Returns 0 on success, negative on failure
 335  */
 336 static int ena_setup_rx_resources(struct ena_adapter *adapter,
 337                                   u32 qid)
 338 {
 339         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 340         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 341         int size, node, i;
 342
 343         if (rx_ring->rx_buffer_info) {
 344                 netif_err(adapter, ifup, adapter->netdev,
 345                           "rx_buffer_info is not NULL");
 346                 return -EEXIST;
 347         }
 348
 349         /* alloc extra element so in rx path
 350          * we can always prefetch rx_info + 1
 351          */
 352         size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
 353         node = cpu_to_node(ena_irq->cpu);
 354
 355         rx_ring->rx_buffer_info = vzalloc_node(size, node);
 356         if (!rx_ring->rx_buffer_info) {
 357                 rx_ring->rx_buffer_info = vzalloc(size);
 358                 if (!rx_ring->rx_buffer_info)
 359                         return -ENOMEM;
 360         }
 361
 362         size = sizeof(u16) * rx_ring->ring_size;
 363         rx_ring->free_rx_ids = vzalloc_node(size, node);
 364         if (!rx_ring->free_rx_ids) {
 365                 rx_ring->free_rx_ids = vzalloc(size);
 366                 if (!rx_ring->free_rx_ids) {
 367                         vfree(rx_ring->rx_buffer_info);
 368                         return -ENOMEM;
 369                 }
 370         }
 371
 372         /* Req id ring for receiving RX pkts out of order */
 373         for (i = 0; i < rx_ring->ring_size; i++)
 374                 rx_ring->free_rx_ids[i] = i;
 375
 376         /* Reset rx statistics */
 377         memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
 378
 379         rx_ring->next_to_clean = 0;
 380         rx_ring->next_to_use = 0;
 381         rx_ring->cpu = ena_irq->cpu;
 382
 383         return 0;
 384 }
 385
 386 /* ena_free_rx_resources - Free I/O Rx Resources
 387  * @adapter: network interface device structure
 388  * @qid: queue index
 389  *
 390  * Free all receive software resources
 391  */
 392 static void ena_free_rx_resources(struct ena_adapter *adapter,
 393                                   u32 qid)
 394 {
 395         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 396
 397         vfree(rx_ring->rx_buffer_info);
 398         rx_ring->rx_buffer_info = NULL;
 399
 400         vfree(rx_ring->free_rx_ids);
 401         rx_ring->free_rx_ids = NULL;
 402 }
 403
 404 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
 405  * @adapter: board private structure
 406  *
 407  * Return 0 on success, negative on failure
 408  */
 409 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
 410 {
 411         int i, rc = 0;
 412
 413         for (i = 0; i < adapter->num_queues; i++) {
 414                 rc = ena_setup_rx_resources(adapter, i);
 415                 if (rc)
 416                         goto err_setup_rx;
 417         }
 418
 419         return 0;
 420
 421 err_setup_rx:
 422
 423         netif_err(adapter, ifup, adapter->netdev,
 424                   "Rx queue %d: allocation failed\n", i);
 425
 426         /* rewind the index freeing the rings as we go */
 427         while (i--)
 428                 ena_free_rx_resources(adapter, i);
 429         return rc;
 430 }
 431
 432 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
 433  * @adapter: board private structure
 434  *
 435  * Free all receive software resources
 436  */
 437 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
 438 {
 439         int i;
 440
 441         for (i = 0; i < adapter->num_queues; i++)
 442                 ena_free_rx_resources(adapter, i);
 443 }
 444
 445 static inline int ena_alloc_rx_page(struct ena_ring *rx_ring,
 446                                     struct ena_rx_buffer *rx_info, gfp_t gfp)
 447 {
 448         struct ena_com_buf *ena_buf;
 449         struct page *page;
 450         dma_addr_t dma;
 451
 452         /* if previous allocated page is not used */
 453         if (unlikely(rx_info->page))
 454                 return 0;
 455
 456         page = alloc_page(gfp);
 457         if (unlikely(!page)) {
 458                 u64_stats_update_begin(&rx_ring->syncp);
 459                 rx_ring->rx_stats.page_alloc_fail++;
 460                 u64_stats_update_end(&rx_ring->syncp);
 461                 return -ENOMEM;
 462         }
 463
 464         dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
 465                            DMA_FROM_DEVICE);
 466         if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
 467                 u64_stats_update_begin(&rx_ring->syncp);
 468                 rx_ring->rx_stats.dma_mapping_err++;
 469                 u64_stats_update_end(&rx_ring->syncp);
 470
 471                 __free_page(page);
 472                 return -EIO;
 473         }
 474         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 475                   "alloc page %p, rx_info %p\n", page, rx_info);
 476
 477         rx_info->page = page;
 478         rx_info->page_offset = 0;
 479         ena_buf = &rx_info->ena_buf;
 480         ena_buf->paddr = dma;
 481         ena_buf->len = ENA_PAGE_SIZE;
 482
 483         return 0;
 484 }
 485
 486 static void ena_free_rx_page(struct ena_ring *rx_ring,
 487                              struct ena_rx_buffer *rx_info)
 488 {
 489         struct page *page = rx_info->page;
 490         struct ena_com_buf *ena_buf = &rx_info->ena_buf;
 491
 492         if (unlikely(!page)) {
 493                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
 494                            "Trying to free unallocated buffer\n");
 495                 return;
 496         }
 497
 498         dma_unmap_page(rx_ring->dev, ena_buf->paddr, ENA_PAGE_SIZE,
 499                        DMA_FROM_DEVICE);
 500
 501         __free_page(page);
 502         rx_info->page = NULL;
 503 }
 504
 505 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
 506 {
 507         u16 next_to_use, req_id;
 508         u32 i;
 509         int rc;
 510
 511         next_to_use = rx_ring->next_to_use;
 512
 513         for (i = 0; i < num; i++) {
 514                 struct ena_rx_buffer *rx_info;
 515
 516                 req_id = rx_ring->free_rx_ids[next_to_use];
 517                 rc = validate_rx_req_id(rx_ring, req_id);
 518                 if (unlikely(rc < 0))
 519                         break;
 520
 521                 rx_info = &rx_ring->rx_buffer_info[req_id];
 522
 523
 524                 rc = ena_alloc_rx_page(rx_ring, rx_info,
 525                                        GFP_ATOMIC | __GFP_COMP);
 526                 if (unlikely(rc < 0)) {
 527                         netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
 528                                    "failed to alloc buffer for rx queue %d\n",
 529                                    rx_ring->qid);
 530                         break;
 531                 }
 532                 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
 533                                                 &rx_info->ena_buf,
 534                                                 req_id);
 535                 if (unlikely(rc)) {
 536                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
 537                                    "failed to add buffer for rx queue %d\n",
 538                                    rx_ring->qid);
 539                         break;
 540                 }
 541                 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
 542                                                    rx_ring->ring_size);
 543         }
 544
 545         if (unlikely(i < num)) {
 546                 u64_stats_update_begin(&rx_ring->syncp);
 547                 rx_ring->rx_stats.refil_partial++;
 548                 u64_stats_update_end(&rx_ring->syncp);
 549                 netdev_warn(rx_ring->netdev,
 550                             "refilled rx qid %d with only %d buffers (from %d)\n",
 551                             rx_ring->qid, i, num);
 552         }
 553
 554         /* ena_com_write_sq_doorbell issues a wmb() */
 555         if (likely(i))
 556                 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
 557
 558         rx_ring->next_to_use = next_to_use;
 559
 560         return i;
 561 }
 562
 563 static void ena_free_rx_bufs(struct ena_adapter *adapter,
 564                              u32 qid)
 565 {
 566         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 567         u32 i;
 568
 569         for (i = 0; i < rx_ring->ring_size; i++) {
 570                 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
 571
 572                 if (rx_info->page)
 573                         ena_free_rx_page(rx_ring, rx_info);
 574         }
 575 }
 576
 577 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
 578  * @adapter: board private structure
 579  *
 580  */
 581 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
 582 {
 583         struct ena_ring *rx_ring;
 584         int i, rc, bufs_num;
 585
 586         for (i = 0; i < adapter->num_queues; i++) {
 587                 rx_ring = &adapter->rx_ring[i];
 588                 bufs_num = rx_ring->ring_size - 1;
 589                 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
 590
 591                 if (unlikely(rc != bufs_num))
 592                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
 593                                    "refilling Queue %d failed. allocated %d buffers from: %d\n",
 594                                    i, rc, bufs_num);
 595         }
 596 }
 597
 598 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
 599 {
 600         int i;
 601
 602         for (i = 0; i < adapter->num_queues; i++)
 603                 ena_free_rx_bufs(adapter, i);
 604 }
 605
 606 /* ena_free_tx_bufs - Free Tx Buffers per Queue
 607  * @tx_ring: TX ring for which buffers be freed
 608  */
 609 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
 610 {
 611         bool print_once = true;
 612         u32 i;
 613
 614         for (i = 0; i < tx_ring->ring_size; i++) {
 615                 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
 616                 struct ena_com_buf *ena_buf;
 617                 int nr_frags;
 618                 int j;
 619
 620                 if (!tx_info->skb)
 621                         continue;
 622
 623                 if (print_once) {
 624                         netdev_notice(tx_ring->netdev,
 625                                       "free uncompleted tx skb qid %d idx 0x%x\n",
 626                                       tx_ring->qid, i);
 627                         print_once = false;
 628                 } else {
 629                         netdev_dbg(tx_ring->netdev,
 630                                    "free uncompleted tx skb qid %d idx 0x%x\n",
 631                                    tx_ring->qid, i);
 632                 }
 633
 634                 ena_buf = tx_info->bufs;
 635                 dma_unmap_single(tx_ring->dev,
 636                                  ena_buf->paddr,
 637                                  ena_buf->len,
 638                                  DMA_TO_DEVICE);
 639
 640                 /* unmap remaining mapped pages */
 641                 nr_frags = tx_info->num_of_bufs - 1;
 642                 for (j = 0; j < nr_frags; j++) {
 643                         ena_buf++;
 644                         dma_unmap_page(tx_ring->dev,
 645                                        ena_buf->paddr,
 646                                        ena_buf->len,
 647                                        DMA_TO_DEVICE);
 648                 }
 649
 650                 dev_kfree_skb_any(tx_info->skb);
 651         }
 652         netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
 653                                                   tx_ring->qid));
 654 }
 655
 656 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
 657 {
 658         struct ena_ring *tx_ring;
 659         int i;
 660
 661         for (i = 0; i < adapter->num_queues; i++) {
 662                 tx_ring = &adapter->tx_ring[i];
 663                 ena_free_tx_bufs(tx_ring);
 664         }
 665 }
 666
 667 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
 668 {
 669         u16 ena_qid;
 670         int i;
 671
 672         for (i = 0; i < adapter->num_queues; i++) {
 673                 ena_qid = ENA_IO_TXQ_IDX(i);
 674                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
 675         }
 676 }
 677
 678 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
 679 {
 680         u16 ena_qid;
 681         int i;
 682
 683         for (i = 0; i < adapter->num_queues; i++) {
 684                 ena_qid = ENA_IO_RXQ_IDX(i);
 685                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
 686         }
 687 }
 688
 689 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
 690 {
 691         ena_destroy_all_tx_queues(adapter);
 692         ena_destroy_all_rx_queues(adapter);
 693 }
 694
 695 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
 696 {
 697         struct ena_tx_buffer *tx_info = NULL;
 698
 699         if (likely(req_id < tx_ring->ring_size)) {
 700                 tx_info = &tx_ring->tx_buffer_info[req_id];
 701                 if (likely(tx_info->skb))
 702                         return 0;
 703         }
 704
 705         if (tx_info)
 706                 netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
 707                           "tx_info doesn't have valid skb\n");
 708         else
 709                 netif_err(tx_ring->adapter, tx_done, tx_ring->netdev,
 710                           "Invalid req_id: %hu\n", req_id);
 711
 712         u64_stats_update_begin(&tx_ring->syncp);
 713         tx_ring->tx_stats.bad_req_id++;
 714         u64_stats_update_end(&tx_ring->syncp);
 715
 716         /* Trigger device reset */
 717         tx_ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
 718         set_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags);
 719         return -EFAULT;
 720 }
 721
 722 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
 723 {
 724         struct netdev_queue *txq;
 725         bool above_thresh;
 726         u32 tx_bytes = 0;
 727         u32 total_done = 0;
 728         u16 next_to_clean;
 729         u16 req_id;
 730         int tx_pkts = 0;
 731         int rc;
 732
 733         next_to_clean = tx_ring->next_to_clean;
 734         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
 735
 736         while (tx_pkts < budget) {
 737                 struct ena_tx_buffer *tx_info;
 738                 struct sk_buff *skb;
 739                 struct ena_com_buf *ena_buf;
 740                 int i, nr_frags;
 741
 742                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
 743                                                 &req_id);
 744                 if (rc)
 745                         break;
 746
 747                 rc = validate_tx_req_id(tx_ring, req_id);
 748                 if (rc)
 749                         break;
 750
 751                 tx_info = &tx_ring->tx_buffer_info[req_id];
 752                 skb = tx_info->skb;
 753
 754                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
 755                 prefetch(&skb->end);
 756
 757                 tx_info->skb = NULL;
 758                 tx_info->last_jiffies = 0;
 759
 760                 if (likely(tx_info->num_of_bufs != 0)) {
 761                         ena_buf = tx_info->bufs;
 762
 763                         dma_unmap_single(tx_ring->dev,
 764                                          dma_unmap_addr(ena_buf, paddr),
 765                                          dma_unmap_len(ena_buf, len),
 766                                          DMA_TO_DEVICE);
 767
 768                         /* unmap remaining mapped pages */
 769                         nr_frags = tx_info->num_of_bufs - 1;
 770                         for (i = 0; i < nr_frags; i++) {
 771                                 ena_buf++;
 772                                 dma_unmap_page(tx_ring->dev,
 773                                                dma_unmap_addr(ena_buf, paddr),
 774                                                dma_unmap_len(ena_buf, len),
 775                                                DMA_TO_DEVICE);
 776                         }
 777                 }
 778
 779                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
 780                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
 781                           skb);
 782
 783                 tx_bytes += skb->len;
 784                 dev_kfree_skb(skb);
 785                 tx_pkts++;
 786                 total_done += tx_info->tx_descs;
 787
 788                 tx_ring->free_tx_ids[next_to_clean] = req_id;
 789                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
 790                                                      tx_ring->ring_size);
 791         }
 792
 793         tx_ring->next_to_clean = next_to_clean;
 794         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
 795         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
 796
 797         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
 798
 799         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
 800                   "tx_poll: q %d done. total pkts: %d\n",
 801                   tx_ring->qid, tx_pkts);
 802
 803         /* need to make the rings circular update visible to
 804          * ena_start_xmit() before checking for netif_queue_stopped().
 805          */
 806         smp_mb();
 807
 808         above_thresh = ena_com_sq_empty_space(tx_ring->ena_com_io_sq) >
 809                 ENA_TX_WAKEUP_THRESH;
 810         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
 811                 __netif_tx_lock(txq, smp_processor_id());
 812                 above_thresh = ena_com_sq_empty_space(tx_ring->ena_com_io_sq) >
 813                         ENA_TX_WAKEUP_THRESH;
 814                 if (netif_tx_queue_stopped(txq) && above_thresh) {
 815                         netif_tx_wake_queue(txq);
 816                         u64_stats_update_begin(&tx_ring->syncp);
 817                         tx_ring->tx_stats.queue_wakeup++;
 818                         u64_stats_update_end(&tx_ring->syncp);
 819                 }
 820                 __netif_tx_unlock(txq);
 821         }
 822
 823         tx_ring->per_napi_bytes += tx_bytes;
 824         tx_ring->per_napi_packets += tx_pkts;
 825
 826         return tx_pkts;
 827 }
 828
 829 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, bool frags)
 830 {
 831         struct sk_buff *skb;
 832
 833         if (frags)
 834                 skb = napi_get_frags(rx_ring->napi);
 835         else
 836                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 837                                                 rx_ring->rx_copybreak);
 838
 839         if (unlikely(!skb)) {
 840                 u64_stats_update_begin(&rx_ring->syncp);
 841                 rx_ring->rx_stats.skb_alloc_fail++;
 842                 u64_stats_update_end(&rx_ring->syncp);
 843                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
 844                           "Failed to allocate skb. frags: %d\n", frags);
 845                 return NULL;
 846         }
 847
 848         return skb;
 849 }
 850
 851 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
 852                                   struct ena_com_rx_buf_info *ena_bufs,
 853                                   u32 descs,
 854                                   u16 *next_to_clean)
 855 {
 856         struct sk_buff *skb;
 857         struct ena_rx_buffer *rx_info;
 858         u16 len, req_id, buf = 0;
 859         void *va;
 860
 861         len = ena_bufs[buf].len;
 862         req_id = ena_bufs[buf].req_id;
 863         rx_info = &rx_ring->rx_buffer_info[req_id];
 864
 865         if (unlikely(!rx_info->page)) {
 866                 netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
 867                           "Page is NULL\n");
 868                 return NULL;
 869         }
 870
 871         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 872                   "rx_info %p page %p\n",
 873                   rx_info, rx_info->page);
 874
 875         /* save virt address of first buffer */
 876         va = page_address(rx_info->page) + rx_info->page_offset;
 877         prefetch(va + NET_IP_ALIGN);
 878
 879         if (len <= rx_ring->rx_copybreak) {
 880                 skb = ena_alloc_skb(rx_ring, false);
 881                 if (unlikely(!skb))
 882                         return NULL;
 883
 884                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 885                           "rx allocated small packet. len %d. data_len %d\n",
 886                           skb->len, skb->data_len);
 887
 888                 /* sync this buffer for CPU use */
 889                 dma_sync_single_for_cpu(rx_ring->dev,
 890                                         dma_unmap_addr(&rx_info->ena_buf, paddr),
 891                                         len,
 892                                         DMA_FROM_DEVICE);
 893                 skb_copy_to_linear_data(skb, va, len);
 894                 dma_sync_single_for_device(rx_ring->dev,
 895                                            dma_unmap_addr(&rx_info->ena_buf, paddr),
 896                                            len,
 897                                            DMA_FROM_DEVICE);
 898
 899                 skb_put(skb, len);
 900                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 901                 rx_ring->free_rx_ids[*next_to_clean] = req_id;
 902                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
 903                                                      rx_ring->ring_size);
 904                 return skb;
 905         }
 906
 907         skb = ena_alloc_skb(rx_ring, true);
 908         if (unlikely(!skb))
 909                 return NULL;
 910
 911         do {
 912                 dma_unmap_page(rx_ring->dev,
 913                                dma_unmap_addr(&rx_info->ena_buf, paddr),
 914                                ENA_PAGE_SIZE, DMA_FROM_DEVICE);
 915
 916                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
 917                                 rx_info->page_offset, len, ENA_PAGE_SIZE);
 918
 919                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
 920                           "rx skb updated. len %d. data_len %d\n",
 921                           skb->len, skb->data_len);
 922
 923                 rx_info->page = NULL;
 924
 925                 rx_ring->free_rx_ids[*next_to_clean] = req_id;
 926                 *next_to_clean =
 927                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
 928                                              rx_ring->ring_size);
 929                 if (likely(--descs == 0))
 930                         break;
 931
 932                 buf++;
 933                 len = ena_bufs[buf].len;
 934                 req_id = ena_bufs[buf].req_id;
 935                 rx_info = &rx_ring->rx_buffer_info[req_id];
 936         } while (1);
 937
 938         return skb;
 939 }
 940
 941 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
 942  * @adapter: structure containing adapter specific data
 943  * @ena_rx_ctx: received packet context/metadata
 944  * @skb: skb currently being received and modified
 945  */
 946 static inline void ena_rx_checksum(struct ena_ring *rx_ring,
 947                                    struct ena_com_rx_ctx *ena_rx_ctx,
 948                                    struct sk_buff *skb)
 949 {
 950         /* Rx csum disabled */
 951         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
 952                 skb->ip_summed = CHECKSUM_NONE;
 953                 return;
 954         }
 955
 956         /* For fragmented packets the checksum isn't valid */
 957         if (ena_rx_ctx->frag) {
 958                 skb->ip_summed = CHECKSUM_NONE;
 959                 return;
 960         }
 961
 962         /* if IP and error */
 963         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
 964                      (ena_rx_ctx->l3_csum_err))) {
 965                 /* ipv4 checksum error */
 966                 skb->ip_summed = CHECKSUM_NONE;
 967                 u64_stats_update_begin(&rx_ring->syncp);
 968                 rx_ring->rx_stats.bad_csum++;
 969                 u64_stats_update_end(&rx_ring->syncp);
 970                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
 971                           "RX IPv4 header checksum error\n");
 972                 return;
 973         }
 974
 975         /* if TCP/UDP */
 976         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
 977                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
 978                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
 979                         /* TCP/UDP checksum error */
 980                         u64_stats_update_begin(&rx_ring->syncp);
 981                         rx_ring->rx_stats.bad_csum++;
 982                         u64_stats_update_end(&rx_ring->syncp);
 983                         netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
 984                                   "RX L4 checksum error\n");
 985                         skb->ip_summed = CHECKSUM_NONE;
 986                         return;
 987                 }
 988
 989                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 990         }
 991 }
 992
 993 static void ena_set_rx_hash(struct ena_ring *rx_ring,
 994                             struct ena_com_rx_ctx *ena_rx_ctx,
 995                             struct sk_buff *skb)
 996 {
 997         enum pkt_hash_types hash_type;
 998
 999         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1000                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1001                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1002
1003                         hash_type = PKT_HASH_TYPE_L4;
1004                 else
1005                         hash_type = PKT_HASH_TYPE_NONE;
1006
1007                 /* Override hash type if the packet is fragmented */
1008                 if (ena_rx_ctx->frag)
1009                         hash_type = PKT_HASH_TYPE_NONE;
1010
1011                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1012         }
1013 }
1014
1015 /* ena_clean_rx_irq - Cleanup RX irq
1016  * @rx_ring: RX ring to clean
1017  * @napi: napi handler
1018  * @budget: how many packets driver is allowed to clean
1019  *
1020  * Returns the number of cleaned buffers.
1021  */
1022 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1023                             u32 budget)
1024 {
1025         u16 next_to_clean = rx_ring->next_to_clean;
1026         u32 res_budget, work_done;
1027
1028         struct ena_com_rx_ctx ena_rx_ctx;
1029         struct ena_adapter *adapter;
1030         struct sk_buff *skb;
1031         int refill_required;
1032         int refill_threshold;
1033         int rc = 0;
1034         int total_len = 0;
1035         int rx_copybreak_pkt = 0;
1036         int i;
1037
1038         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1039                   "%s qid %d\n", __func__, rx_ring->qid);
1040         res_budget = budget;
1041
1042         do {
1043                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1044                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1045                 ena_rx_ctx.descs = 0;
1046                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1047                                     rx_ring->ena_com_io_sq,
1048                                     &ena_rx_ctx);
1049                 if (unlikely(rc))
1050                         goto error;
1051
1052                 if (unlikely(ena_rx_ctx.descs == 0))
1053                         break;
1054
1055                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1056                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1057                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1058                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1059
1060                 /* allocate skb and fill it */
1061                 skb = ena_rx_skb(rx_ring, rx_ring->ena_bufs, ena_rx_ctx.descs,
1062                                  &next_to_clean);
1063
1064                 /* exit if we failed to retrieve a buffer */
1065                 if (unlikely(!skb)) {
1066                         for (i = 0; i < ena_rx_ctx.descs; i++) {
1067                                 rx_ring->free_tx_ids[next_to_clean] =
1068                                         rx_ring->ena_bufs[i].req_id;
1069                                 next_to_clean =
1070                                         ENA_RX_RING_IDX_NEXT(next_to_clean,
1071                                                              rx_ring->ring_size);
1072                         }
1073                         break;
1074                 }
1075
1076                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1077
1078                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1079
1080                 skb_record_rx_queue(skb, rx_ring->qid);
1081
1082                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1083                         total_len += rx_ring->ena_bufs[0].len;
1084                         rx_copybreak_pkt++;
1085                         napi_gro_receive(napi, skb);
1086                 } else {
1087                         total_len += skb->len;
1088                         napi_gro_frags(napi);
1089                 }
1090
1091                 res_budget--;
1092         } while (likely(res_budget));
1093
1094         work_done = budget - res_budget;
1095         rx_ring->per_napi_bytes += total_len;
1096         rx_ring->per_napi_packets += work_done;
1097         u64_stats_update_begin(&rx_ring->syncp);
1098         rx_ring->rx_stats.bytes += total_len;
1099         rx_ring->rx_stats.cnt += work_done;
1100         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1101         u64_stats_update_end(&rx_ring->syncp);
1102
1103         rx_ring->next_to_clean = next_to_clean;
1104
1105         refill_required = ena_com_sq_empty_space(rx_ring->ena_com_io_sq);
1106         refill_threshold = rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER;
1107
1108         /* Optimization, try to batch new rx buffers */
1109         if (refill_required > refill_threshold) {
1110                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1111                 ena_refill_rx_bufs(rx_ring, refill_required);
1112         }
1113
1114         return work_done;
1115
1116 error:
1117         adapter = netdev_priv(rx_ring->netdev);
1118
1119         u64_stats_update_begin(&rx_ring->syncp);
1120         rx_ring->rx_stats.bad_desc_num++;
1121         u64_stats_update_end(&rx_ring->syncp);
1122
1123         /* Too many desc from the device. Trigger reset */
1124         adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1125         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1126
1127         return 0;
1128 }
1129
1130 inline void ena_adjust_intr_moderation(struct ena_ring *rx_ring,
1131                                        struct ena_ring *tx_ring)
1132 {
1133         /* We apply adaptive moderation on Rx path only.
1134          * Tx uses static interrupt moderation.
1135          */
1136         ena_com_calculate_interrupt_delay(rx_ring->ena_dev,
1137                                           rx_ring->per_napi_packets,
1138                                           rx_ring->per_napi_bytes,
1139                                           &rx_ring->smoothed_interval,
1140                                           &rx_ring->moder_tbl_idx);
1141
1142         /* Reset per napi packets/bytes */
1143         tx_ring->per_napi_packets = 0;
1144         tx_ring->per_napi_bytes = 0;
1145         rx_ring->per_napi_packets = 0;
1146         rx_ring->per_napi_bytes = 0;
1147 }
1148
1149 static inline void ena_unmask_interrupt(struct ena_ring *tx_ring,
1150                                         struct ena_ring *rx_ring)
1151 {
1152         struct ena_eth_io_intr_reg intr_reg;
1153
1154         /* Update intr register: rx intr delay,
1155          * tx intr delay and interrupt unmask
1156          */
1157         ena_com_update_intr_reg(&intr_reg,
1158                                 rx_ring->smoothed_interval,
1159                                 tx_ring->smoothed_interval,
1160                                 true);
1161
1162         /* It is a shared MSI-X.
1163          * Tx and Rx CQ have pointer to it.
1164          * So we use one of them to reach the intr reg
1165          */
1166         ena_com_unmask_intr(rx_ring->ena_com_io_cq, &intr_reg);
1167 }
1168
1169 static inline void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1170                                              struct ena_ring *rx_ring)
1171 {
1172         int cpu = get_cpu();
1173         int numa_node;
1174
1175         /* Check only one ring since the 2 rings are running on the same cpu */
1176         if (likely(tx_ring->cpu == cpu))
1177                 goto out;
1178
1179         numa_node = cpu_to_node(cpu);
1180         put_cpu();
1181
1182         if (numa_node != NUMA_NO_NODE) {
1183                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1184                 ena_com_update_numa_node(rx_ring->ena_com_io_cq, numa_node);
1185         }
1186
1187         tx_ring->cpu = cpu;
1188         rx_ring->cpu = cpu;
1189
1190         return;
1191 out:
1192         put_cpu();
1193 }
1194
1195 static int ena_io_poll(struct napi_struct *napi, int budget)
1196 {
1197         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1198         struct ena_ring *tx_ring, *rx_ring;
1199
1200         u32 tx_work_done;
1201         u32 rx_work_done;
1202         int tx_budget;
1203         int napi_comp_call = 0;
1204         int ret;
1205
1206         tx_ring = ena_napi->tx_ring;
1207         rx_ring = ena_napi->rx_ring;
1208
1209         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1210
1211         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1212             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1213                 napi_complete_done(napi, 0);
1214                 return 0;
1215         }
1216
1217         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1218         rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1219
1220         /* If the device is about to reset or down, avoid unmask
1221          * the interrupt and return 0 so NAPI won't reschedule
1222          */
1223         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1224                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1225                 napi_complete_done(napi, 0);
1226                 ret = 0;
1227
1228         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1229                 napi_comp_call = 1;
1230
1231                 /* Update numa and unmask the interrupt only when schedule
1232                  * from the interrupt context (vs from sk_busy_loop)
1233                  */
1234                 if (napi_complete_done(napi, rx_work_done)) {
1235                         /* Tx and Rx share the same interrupt vector */
1236                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1237                                 ena_adjust_intr_moderation(rx_ring, tx_ring);
1238
1239                         ena_unmask_interrupt(tx_ring, rx_ring);
1240                 }
1241
1242                 ena_update_ring_numa_node(tx_ring, rx_ring);
1243
1244                 ret = rx_work_done;
1245         } else {
1246                 ret = budget;
1247         }
1248
1249         u64_stats_update_begin(&tx_ring->syncp);
1250         tx_ring->tx_stats.napi_comp += napi_comp_call;
1251         tx_ring->tx_stats.tx_poll++;
1252         u64_stats_update_end(&tx_ring->syncp);
1253
1254         return ret;
1255 }
1256
1257 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1258 {
1259         struct ena_adapter *adapter = (struct ena_adapter *)data;
1260
1261         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1262
1263         /* Don't call the aenq handler before probe is done */
1264         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1265                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
1266
1267         return IRQ_HANDLED;
1268 }
1269
1270 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1271  * @irq: interrupt number
1272  * @data: pointer to a network interface private napi device structure
1273  */
1274 static irqreturn_t ena_intr_msix_io(int irq, void *data)
1275 {
1276         struct ena_napi *ena_napi = data;
1277
1278         ena_napi->tx_ring->first_interrupt = true;
1279         ena_napi->rx_ring->first_interrupt = true;
1280
1281         napi_schedule_irqoff(&ena_napi->napi);
1282
1283         return IRQ_HANDLED;
1284 }
1285
1286 /* Reserve a single MSI-X vector for management (admin + aenq).
1287  * plus reserve one vector for each potential io queue.
1288  * the number of potential io queues is the minimum of what the device
1289  * supports and the number of vCPUs.
1290  */
1291 static int ena_enable_msix(struct ena_adapter *adapter, int num_queues)
1292 {
1293         int msix_vecs, irq_cnt;
1294
1295         if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1296                 netif_err(adapter, probe, adapter->netdev,
1297                           "Error, MSI-X is already enabled\n");
1298                 return -EPERM;
1299         }
1300
1301         /* Reserved the max msix vectors we might need */
1302         msix_vecs = ENA_MAX_MSIX_VEC(num_queues);
1303
1304         netif_dbg(adapter, probe, adapter->netdev,
1305                   "trying to enable MSI-X, vectors %d\n", msix_vecs);
1306
1307         irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
1308                                         msix_vecs, PCI_IRQ_MSIX);
1309
1310         if (irq_cnt < 0) {
1311                 netif_err(adapter, probe, adapter->netdev,
1312                           "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
1313                 return -ENOSPC;
1314         }
1315
1316         if (irq_cnt != msix_vecs) {
1317                 netif_notice(adapter, probe, adapter->netdev,
1318                              "enable only %d MSI-X (out of %d), reduce the number of queues\n",
1319                              irq_cnt, msix_vecs);
1320                 adapter->num_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
1321         }
1322
1323         if (ena_init_rx_cpu_rmap(adapter))
1324                 netif_warn(adapter, probe, adapter->netdev,
1325                            "Failed to map IRQs to CPUs\n");
1326
1327         adapter->msix_vecs = irq_cnt;
1328         set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
1329
1330         return 0;
1331 }
1332
1333 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
1334 {
1335         u32 cpu;
1336
1337         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
1338                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
1339                  pci_name(adapter->pdev));
1340         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
1341                 ena_intr_msix_mgmnt;
1342         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
1343         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
1344                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
1345         cpu = cpumask_first(cpu_online_mask);
1346         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
1347         cpumask_set_cpu(cpu,
1348                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
1349 }
1350
1351 static void ena_setup_io_intr(struct ena_adapter *adapter)
1352 {
1353         struct net_device *netdev;
1354         int irq_idx, i, cpu;
1355
1356         netdev = adapter->netdev;
1357
1358         for (i = 0; i < adapter->num_queues; i++) {
1359                 irq_idx = ENA_IO_IRQ_IDX(i);
1360                 cpu = i % num_online_cpus();
1361
1362                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
1363                          "%s-Tx-Rx-%d", netdev->name, i);
1364                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
1365                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
1366                 adapter->irq_tbl[irq_idx].vector =
1367                         pci_irq_vector(adapter->pdev, irq_idx);
1368                 adapter->irq_tbl[irq_idx].cpu = cpu;
1369
1370                 cpumask_set_cpu(cpu,
1371                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
1372         }
1373 }
1374
1375 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
1376 {
1377         unsigned long flags = 0;
1378         struct ena_irq *irq;
1379         int rc;
1380
1381         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1382         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1383                          irq->data);
1384         if (rc) {
1385                 netif_err(adapter, probe, adapter->netdev,
1386                           "failed to request admin irq\n");
1387                 return rc;
1388         }
1389
1390         netif_dbg(adapter, probe, adapter->netdev,
1391                   "set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
1392                   irq->affinity_hint_mask.bits[0], irq->vector);
1393
1394         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1395
1396         return rc;
1397 }
1398
1399 static int ena_request_io_irq(struct ena_adapter *adapter)
1400 {
1401         unsigned long flags = 0;
1402         struct ena_irq *irq;
1403         int rc = 0, i, k;
1404
1405         if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
1406                 netif_err(adapter, ifup, adapter->netdev,
1407                           "Failed to request I/O IRQ: MSI-X is not enabled\n");
1408                 return -EINVAL;
1409         }
1410
1411         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1412                 irq = &adapter->irq_tbl[i];
1413                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
1414                                  irq->data);
1415                 if (rc) {
1416                         netif_err(adapter, ifup, adapter->netdev,
1417                                   "Failed to request I/O IRQ. index %d rc %d\n",
1418                                    i, rc);
1419                         goto err;
1420                 }
1421
1422                 netif_dbg(adapter, ifup, adapter->netdev,
1423                           "set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
1424                           i, irq->affinity_hint_mask.bits[0], irq->vector);
1425
1426                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
1427         }
1428
1429         return rc;
1430
1431 err:
1432         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
1433                 irq = &adapter->irq_tbl[k];
1434                 free_irq(irq->vector, irq->data);
1435         }
1436
1437         return rc;
1438 }
1439
1440 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
1441 {
1442         struct ena_irq *irq;
1443
1444         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
1445         synchronize_irq(irq->vector);
1446         irq_set_affinity_hint(irq->vector, NULL);
1447         free_irq(irq->vector, irq->data);
1448 }
1449
1450 static void ena_free_io_irq(struct ena_adapter *adapter)
1451 {
1452         struct ena_irq *irq;
1453         int i;
1454
1455 #ifdef CONFIG_RFS_ACCEL
1456         if (adapter->msix_vecs >= 1) {
1457                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
1458                 adapter->netdev->rx_cpu_rmap = NULL;
1459         }
1460 #endif /* CONFIG_RFS_ACCEL */
1461
1462         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++) {
1463                 irq = &adapter->irq_tbl[i];
1464                 irq_set_affinity_hint(irq->vector, NULL);
1465                 free_irq(irq->vector, irq->data);
1466         }
1467 }
1468
1469 static void ena_disable_msix(struct ena_adapter *adapter)
1470 {
1471         if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
1472                 pci_free_irq_vectors(adapter->pdev);
1473 }
1474
1475 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
1476 {
1477         int i;
1478
1479         if (!netif_running(adapter->netdev))
1480                 return;
1481
1482         for (i = ENA_IO_IRQ_FIRST_IDX; i < adapter->msix_vecs; i++)
1483                 synchronize_irq(adapter->irq_tbl[i].vector);
1484 }
1485
1486 static void ena_del_napi(struct ena_adapter *adapter)
1487 {
1488         int i;
1489
1490         for (i = 0; i < adapter->num_queues; i++)
1491                 netif_napi_del(&adapter->ena_napi[i].napi);
1492 }
1493
1494 static void ena_init_napi(struct ena_adapter *adapter)
1495 {
1496         struct ena_napi *napi;
1497         int i;
1498
1499         for (i = 0; i < adapter->num_queues; i++) {
1500                 napi = &adapter->ena_napi[i];
1501
1502                 netif_napi_add(adapter->netdev,
1503                                &adapter->ena_napi[i].napi,
1504                                ena_io_poll,
1505                                ENA_NAPI_BUDGET);
1506                 napi->rx_ring = &adapter->rx_ring[i];
1507                 napi->tx_ring = &adapter->tx_ring[i];
1508                 napi->qid = i;
1509         }
1510 }
1511
1512 static void ena_napi_disable_all(struct ena_adapter *adapter)
1513 {
1514         int i;
1515
1516         for (i = 0; i < adapter->num_queues; i++)
1517                 napi_disable(&adapter->ena_napi[i].napi);
1518 }
1519
1520 static void ena_napi_enable_all(struct ena_adapter *adapter)
1521 {
1522         int i;
1523
1524         for (i = 0; i < adapter->num_queues; i++)
1525                 napi_enable(&adapter->ena_napi[i].napi);
1526 }
1527
1528 static void ena_restore_ethtool_params(struct ena_adapter *adapter)
1529 {
1530         adapter->tx_usecs = 0;
1531         adapter->rx_usecs = 0;
1532         adapter->tx_frames = 1;
1533         adapter->rx_frames = 1;
1534 }
1535
1536 /* Configure the Rx forwarding */
1537 static int ena_rss_configure(struct ena_adapter *adapter)
1538 {
1539         struct ena_com_dev *ena_dev = adapter->ena_dev;
1540         int rc;
1541
1542         /* In case the RSS table wasn't initialized by probe */
1543         if (!ena_dev->rss.tbl_log_size) {
1544                 rc = ena_rss_init_default(adapter);
1545                 if (rc && (rc != -EOPNOTSUPP)) {
1546                         netif_err(adapter, ifup, adapter->netdev,
1547                                   "Failed to init RSS rc: %d\n", rc);
1548                         return rc;
1549                 }
1550         }
1551
1552         /* Set indirect table */
1553         rc = ena_com_indirect_table_set(ena_dev);
1554         if (unlikely(rc && rc != -EOPNOTSUPP))
1555                 return rc;
1556
1557         /* Configure hash function (if supported) */
1558         rc = ena_com_set_hash_function(ena_dev);
1559         if (unlikely(rc && (rc != -EOPNOTSUPP)))
1560                 return rc;
1561
1562         /* Configure hash inputs (if supported) */
1563         rc = ena_com_set_hash_ctrl(ena_dev);
1564         if (unlikely(rc && (rc != -EOPNOTSUPP)))
1565                 return rc;
1566
1567         return 0;
1568 }
1569
1570 static int ena_up_complete(struct ena_adapter *adapter)
1571 {
1572         int rc;
1573
1574         rc = ena_rss_configure(adapter);
1575         if (rc)
1576                 return rc;
1577
1578         ena_init_napi(adapter);
1579
1580         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
1581
1582         ena_refill_all_rx_bufs(adapter);
1583
1584         /* enable transmits */
1585         netif_tx_start_all_queues(adapter->netdev);
1586
1587         ena_restore_ethtool_params(adapter);
1588
1589         ena_napi_enable_all(adapter);
1590
1591         return 0;
1592 }
1593
1594 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
1595 {
1596         struct ena_com_create_io_ctx ctx = { 0 };
1597         struct ena_com_dev *ena_dev;
1598         struct ena_ring *tx_ring;
1599         u32 msix_vector;
1600         u16 ena_qid;
1601         int rc;
1602
1603         ena_dev = adapter->ena_dev;
1604
1605         tx_ring = &adapter->tx_ring[qid];
1606         msix_vector = ENA_IO_IRQ_IDX(qid);
1607         ena_qid = ENA_IO_TXQ_IDX(qid);
1608
1609         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
1610         ctx.qid = ena_qid;
1611         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
1612         ctx.msix_vector = msix_vector;
1613         ctx.queue_size = adapter->tx_ring_size;
1614         ctx.numa_node = cpu_to_node(tx_ring->cpu);
1615
1616         rc = ena_com_create_io_queue(ena_dev, &ctx);
1617         if (rc) {
1618                 netif_err(adapter, ifup, adapter->netdev,
1619                           "Failed to create I/O TX queue num %d rc: %d\n",
1620                           qid, rc);
1621                 return rc;
1622         }
1623
1624         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1625                                      &tx_ring->ena_com_io_sq,
1626                                      &tx_ring->ena_com_io_cq);
1627         if (rc) {
1628                 netif_err(adapter, ifup, adapter->netdev,
1629                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
1630                           qid, rc);
1631                 ena_com_destroy_io_queue(ena_dev, ena_qid);
1632                 return rc;
1633         }
1634
1635         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
1636         return rc;
1637 }
1638
1639 static int ena_create_all_io_tx_queues(struct ena_adapter *adapter)
1640 {
1641         struct ena_com_dev *ena_dev = adapter->ena_dev;
1642         int rc, i;
1643
1644         for (i = 0; i < adapter->num_queues; i++) {
1645                 rc = ena_create_io_tx_queue(adapter, i);
1646                 if (rc)
1647                         goto create_err;
1648         }
1649
1650         return 0;
1651
1652 create_err:
1653         while (i--)
1654                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
1655
1656         return rc;
1657 }
1658
1659 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
1660 {
1661         struct ena_com_dev *ena_dev;
1662         struct ena_com_create_io_ctx ctx = { 0 };
1663         struct ena_ring *rx_ring;
1664         u32 msix_vector;
1665         u16 ena_qid;
1666         int rc;
1667
1668         ena_dev = adapter->ena_dev;
1669
1670         rx_ring = &adapter->rx_ring[qid];
1671         msix_vector = ENA_IO_IRQ_IDX(qid);
1672         ena_qid = ENA_IO_RXQ_IDX(qid);
1673
1674         ctx.qid = ena_qid;
1675         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
1676         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
1677         ctx.msix_vector = msix_vector;
1678         ctx.queue_size = adapter->rx_ring_size;
1679         ctx.numa_node = cpu_to_node(rx_ring->cpu);
1680
1681         rc = ena_com_create_io_queue(ena_dev, &ctx);
1682         if (rc) {
1683                 netif_err(adapter, ifup, adapter->netdev,
1684                           "Failed to create I/O RX queue num %d rc: %d\n",
1685                           qid, rc);
1686                 return rc;
1687         }
1688
1689         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
1690                                      &rx_ring->ena_com_io_sq,
1691                                      &rx_ring->ena_com_io_cq);
1692         if (rc) {
1693                 netif_err(adapter, ifup, adapter->netdev,
1694                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
1695                           qid, rc);
1696                 ena_com_destroy_io_queue(ena_dev, ena_qid);
1697                 return rc;
1698         }
1699
1700         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
1701
1702         return rc;
1703 }
1704
1705 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
1706 {
1707         struct ena_com_dev *ena_dev = adapter->ena_dev;
1708         int rc, i;
1709
1710         for (i = 0; i < adapter->num_queues; i++) {
1711                 rc = ena_create_io_rx_queue(adapter, i);
1712                 if (rc)
1713                         goto create_err;
1714         }
1715
1716         return 0;
1717
1718 create_err:
1719         while (i--)
1720                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
1721
1722         return rc;
1723 }
1724
1725 static int ena_up(struct ena_adapter *adapter)
1726 {
1727         int rc, i;
1728
1729         netdev_dbg(adapter->netdev, "%s\n", __func__);
1730
1731         ena_setup_io_intr(adapter);
1732
1733         rc = ena_request_io_irq(adapter);
1734         if (rc)
1735                 goto err_req_irq;
1736
1737         /* allocate transmit descriptors */
1738         rc = ena_setup_all_tx_resources(adapter);
1739         if (rc)
1740                 goto err_setup_tx;
1741
1742         /* allocate receive descriptors */
1743         rc = ena_setup_all_rx_resources(adapter);
1744         if (rc)
1745                 goto err_setup_rx;
1746
1747         /* Create TX queues */
1748         rc = ena_create_all_io_tx_queues(adapter);
1749         if (rc)
1750                 goto err_create_tx_queues;
1751
1752         /* Create RX queues */
1753         rc = ena_create_all_io_rx_queues(adapter);
1754         if (rc)
1755                 goto err_create_rx_queues;
1756
1757         rc = ena_up_complete(adapter);
1758         if (rc)
1759                 goto err_up;
1760
1761         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
1762                 netif_carrier_on(adapter->netdev);
1763
1764         u64_stats_update_begin(&adapter->syncp);
1765         adapter->dev_stats.interface_up++;
1766         u64_stats_update_end(&adapter->syncp);
1767
1768         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1769
1770         /* Enable completion queues interrupt */
1771         for (i = 0; i < adapter->num_queues; i++)
1772                 ena_unmask_interrupt(&adapter->tx_ring[i],
1773                                      &adapter->rx_ring[i]);
1774
1775         /* schedule napi in case we had pending packets
1776          * from the last time we disable napi
1777          */
1778         for (i = 0; i < adapter->num_queues; i++)
1779                 napi_schedule(&adapter->ena_napi[i].napi);
1780
1781         return rc;
1782
1783 err_up:
1784         ena_destroy_all_rx_queues(adapter);
1785 err_create_rx_queues:
1786         ena_destroy_all_tx_queues(adapter);
1787 err_create_tx_queues:
1788         ena_free_all_io_rx_resources(adapter);
1789 err_setup_rx:
1790         ena_free_all_io_tx_resources(adapter);
1791 err_setup_tx:
1792         ena_free_io_irq(adapter);
1793 err_req_irq:
1794
1795         return rc;
1796 }
1797
1798 static void ena_down(struct ena_adapter *adapter)
1799 {
1800         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
1801
1802         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
1803
1804         u64_stats_update_begin(&adapter->syncp);
1805         adapter->dev_stats.interface_down++;
1806         u64_stats_update_end(&adapter->syncp);
1807
1808         netif_carrier_off(adapter->netdev);
1809         netif_tx_disable(adapter->netdev);
1810
1811         /* After this point the napi handler won't enable the tx queue */
1812         ena_napi_disable_all(adapter);
1813
1814         /* After destroy the queue there won't be any new interrupts */
1815
1816         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
1817                 int rc;
1818
1819                 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
1820                 if (rc)
1821                         dev_err(&adapter->pdev->dev, "Device reset failed\n");
1822         }
1823
1824         ena_destroy_all_io_queues(adapter);
1825
1826         ena_disable_io_intr_sync(adapter);
1827         ena_free_io_irq(adapter);
1828         ena_del_napi(adapter);
1829
1830         ena_free_all_tx_bufs(adapter);
1831         ena_free_all_rx_bufs(adapter);
1832         ena_free_all_io_tx_resources(adapter);
1833         ena_free_all_io_rx_resources(adapter);
1834 }
1835
1836 /* ena_open - Called when a network interface is made active
1837  * @netdev: network interface device structure
1838  *
1839  * Returns 0 on success, negative value on failure
1840  *
1841  * The open entry point is called when a network interface is made
1842  * active by the system (IFF_UP).  At this point all resources needed
1843  * for transmit and receive operations are allocated, the interrupt
1844  * handler is registered with the OS, the watchdog timer is started,
1845  * and the stack is notified that the interface is ready.
1846  */
1847 static int ena_open(struct net_device *netdev)
1848 {
1849         struct ena_adapter *adapter = netdev_priv(netdev);
1850         int rc;
1851
1852         /* Notify the stack of the actual queue counts. */
1853         rc = netif_set_real_num_tx_queues(netdev, adapter->num_queues);
1854         if (rc) {
1855                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
1856                 return rc;
1857         }
1858
1859         rc = netif_set_real_num_rx_queues(netdev, adapter->num_queues);
1860         if (rc) {
1861                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
1862                 return rc;
1863         }
1864
1865         rc = ena_up(adapter);
1866         if (rc)
1867                 return rc;
1868
1869         return rc;
1870 }
1871
1872 /* ena_close - Disables a network interface
1873  * @netdev: network interface device structure
1874  *
1875  * Returns 0, this is not allowed to fail
1876  *
1877  * The close entry point is called when an interface is de-activated
1878  * by the OS.  The hardware is still under the drivers control, but
1879  * needs to be disabled.  A global MAC reset is issued to stop the
1880  * hardware, and all transmit and receive resources are freed.
1881  */
1882 static int ena_close(struct net_device *netdev)
1883 {
1884         struct ena_adapter *adapter = netdev_priv(netdev);
1885
1886         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
1887
1888         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
1889                 ena_down(adapter);
1890
1891         /* Check for device status and issue reset if needed*/
1892         check_for_admin_com_state(adapter);
1893         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
1894                 netif_err(adapter, ifdown, adapter->netdev,
1895                           "Destroy failure, restarting device\n");
1896                 ena_dump_stats_to_dmesg(adapter);
1897                 /* rtnl lock already obtained in dev_ioctl() layer */
1898                 ena_destroy_device(adapter, false);
1899                 ena_restore_device(adapter);
1900         }
1901
1902         return 0;
1903 }
1904
1905 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx, struct sk_buff *skb)
1906 {
1907         u32 mss = skb_shinfo(skb)->gso_size;
1908         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
1909         u8 l4_protocol = 0;
1910
1911         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
1912                 ena_tx_ctx->l4_csum_enable = 1;
1913                 if (mss) {
1914                         ena_tx_ctx->tso_enable = 1;
1915                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
1916                         ena_tx_ctx->l4_csum_partial = 0;
1917                 } else {
1918                         ena_tx_ctx->tso_enable = 0;
1919                         ena_meta->l4_hdr_len = 0;
1920                         ena_tx_ctx->l4_csum_partial = 1;
1921                 }
1922
1923                 switch (ip_hdr(skb)->version) {
1924                 case IPVERSION:
1925                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
1926                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
1927                                 ena_tx_ctx->df = 1;
1928                         if (mss)
1929                                 ena_tx_ctx->l3_csum_enable = 1;
1930                         l4_protocol = ip_hdr(skb)->protocol;
1931                         break;
1932                 case 6:
1933                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
1934                         l4_protocol = ipv6_hdr(skb)->nexthdr;
1935                         break;
1936                 default:
1937                         break;
1938                 }
1939
1940                 if (l4_protocol == IPPROTO_TCP)
1941                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
1942                 else
1943                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
1944
1945                 ena_meta->mss = mss;
1946                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
1947                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
1948                 ena_tx_ctx->meta_valid = 1;
1949
1950         } else {
1951                 ena_tx_ctx->meta_valid = 0;
1952         }
1953 }
1954
1955 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
1956                                        struct sk_buff *skb)
1957 {
1958         int num_frags, header_len, rc;
1959
1960         num_frags = skb_shinfo(skb)->nr_frags;
1961         header_len = skb_headlen(skb);
1962
1963         if (num_frags < tx_ring->sgl_size)
1964                 return 0;
1965
1966         if ((num_frags == tx_ring->sgl_size) &&
1967             (header_len < tx_ring->tx_max_header_size))
1968                 return 0;
1969
1970         u64_stats_update_begin(&tx_ring->syncp);
1971         tx_ring->tx_stats.linearize++;
1972         u64_stats_update_end(&tx_ring->syncp);
1973
1974         rc = skb_linearize(skb);
1975         if (unlikely(rc)) {
1976                 u64_stats_update_begin(&tx_ring->syncp);
1977                 tx_ring->tx_stats.linearize_failed++;
1978                 u64_stats_update_end(&tx_ring->syncp);
1979         }
1980
1981         return rc;
1982 }
1983
1984 /* Called with netif_tx_lock. */
1985 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
1986 {
1987         struct ena_adapter *adapter = netdev_priv(dev);
1988         struct ena_tx_buffer *tx_info;
1989         struct ena_com_tx_ctx ena_tx_ctx;
1990         struct ena_ring *tx_ring;
1991         struct netdev_queue *txq;
1992         struct ena_com_buf *ena_buf;
1993         void *push_hdr;
1994         u32 len, last_frag;
1995         u16 next_to_use;
1996         u16 req_id;
1997         u16 push_len;
1998         u16 header_len;
1999         dma_addr_t dma;
2000         int qid, rc, nb_hw_desc;
2001         int i = -1;
2002
2003         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
2004         /*  Determine which tx ring we will be placed on */
2005         qid = skb_get_queue_mapping(skb);
2006         tx_ring = &adapter->tx_ring[qid];
2007         txq = netdev_get_tx_queue(dev, qid);
2008
2009         rc = ena_check_and_linearize_skb(tx_ring, skb);
2010         if (unlikely(rc))
2011                 goto error_drop_packet;
2012
2013         skb_tx_timestamp(skb);
2014         len = skb_headlen(skb);
2015
2016         next_to_use = tx_ring->next_to_use;
2017         req_id = tx_ring->free_tx_ids[next_to_use];
2018         tx_info = &tx_ring->tx_buffer_info[req_id];
2019         tx_info->num_of_bufs = 0;
2020
2021         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
2022         ena_buf = tx_info->bufs;
2023         tx_info->skb = skb;
2024
2025         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2026                 /* prepared the push buffer */
2027                 push_len = min_t(u32, len, tx_ring->tx_max_header_size);
2028                 header_len = push_len;
2029                 push_hdr = skb->data;
2030         } else {
2031                 push_len = 0;
2032                 header_len = min_t(u32, len, tx_ring->tx_max_header_size);
2033                 push_hdr = NULL;
2034         }
2035
2036         netif_dbg(adapter, tx_queued, dev,
2037                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2038                   push_hdr, push_len);
2039
2040         if (len > push_len) {
2041                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2042                                      len - push_len, DMA_TO_DEVICE);
2043                 if (dma_mapping_error(tx_ring->dev, dma))
2044                         goto error_report_dma_error;
2045
2046                 ena_buf->paddr = dma;
2047                 ena_buf->len = len - push_len;
2048
2049                 ena_buf++;
2050                 tx_info->num_of_bufs++;
2051         }
2052
2053         last_frag = skb_shinfo(skb)->nr_frags;
2054
2055         for (i = 0; i < last_frag; i++) {
2056                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2057
2058                 len = skb_frag_size(frag);
2059                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, len,
2060                                        DMA_TO_DEVICE);
2061                 if (dma_mapping_error(tx_ring->dev, dma))
2062                         goto error_report_dma_error;
2063
2064                 ena_buf->paddr = dma;
2065                 ena_buf->len = len;
2066                 ena_buf++;
2067         }
2068
2069         tx_info->num_of_bufs += last_frag;
2070
2071         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
2072         ena_tx_ctx.ena_bufs = tx_info->bufs;
2073         ena_tx_ctx.push_header = push_hdr;
2074         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
2075         ena_tx_ctx.req_id = req_id;
2076         ena_tx_ctx.header_len = header_len;
2077
2078         /* set flags and meta data */
2079         ena_tx_csum(&ena_tx_ctx, skb);
2080
2081         /* prepare the packet's descriptors to dma engine */
2082         rc = ena_com_prepare_tx(tx_ring->ena_com_io_sq, &ena_tx_ctx,
2083                                 &nb_hw_desc);
2084
2085         if (unlikely(rc)) {
2086                 netif_err(adapter, tx_queued, dev,
2087                           "failed to prepare tx bufs\n");
2088                 u64_stats_update_begin(&tx_ring->syncp);
2089                 tx_ring->tx_stats.queue_stop++;
2090                 tx_ring->tx_stats.prepare_ctx_err++;
2091                 u64_stats_update_end(&tx_ring->syncp);
2092                 netif_tx_stop_queue(txq);
2093                 goto error_unmap_dma;
2094         }
2095
2096         netdev_tx_sent_queue(txq, skb->len);
2097
2098         u64_stats_update_begin(&tx_ring->syncp);
2099         tx_ring->tx_stats.cnt++;
2100         tx_ring->tx_stats.bytes += skb->len;
2101         u64_stats_update_end(&tx_ring->syncp);
2102
2103         tx_info->tx_descs = nb_hw_desc;
2104         tx_info->last_jiffies = jiffies;
2105         tx_info->print_once = 0;
2106
2107         tx_ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
2108                 tx_ring->ring_size);
2109
2110         /* stop the queue when no more space available, the packet can have up
2111          * to sgl_size + 2. one for the meta descriptor and one for header
2112          * (if the header is larger than tx_max_header_size).
2113          */
2114         if (unlikely(ena_com_sq_empty_space(tx_ring->ena_com_io_sq) <
2115                      (tx_ring->sgl_size + 2))) {
2116                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
2117                           __func__, qid);
2118
2119                 netif_tx_stop_queue(txq);
2120                 u64_stats_update_begin(&tx_ring->syncp);
2121                 tx_ring->tx_stats.queue_stop++;
2122                 u64_stats_update_end(&tx_ring->syncp);
2123
2124                 /* There is a rare condition where this function decide to
2125                  * stop the queue but meanwhile clean_tx_irq updates
2126                  * next_to_completion and terminates.
2127                  * The queue will remain stopped forever.
2128                  * To solve this issue add a mb() to make sure that
2129                  * netif_tx_stop_queue() write is vissible before checking if
2130                  * there is additional space in the queue.
2131                  */
2132                 smp_mb();
2133
2134                 if (ena_com_sq_empty_space(tx_ring->ena_com_io_sq)
2135                                 > ENA_TX_WAKEUP_THRESH) {
2136                         netif_tx_wake_queue(txq);
2137                         u64_stats_update_begin(&tx_ring->syncp);
2138                         tx_ring->tx_stats.queue_wakeup++;
2139                         u64_stats_update_end(&tx_ring->syncp);
2140                 }
2141         }
2142
2143         if (netif_xmit_stopped(txq) || !skb->xmit_more) {
2144                 /* trigger the dma engine. ena_com_write_sq_doorbell()
2145                  * has a mb
2146                  */
2147                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
2148                 u64_stats_update_begin(&tx_ring->syncp);
2149                 tx_ring->tx_stats.doorbells++;
2150                 u64_stats_update_end(&tx_ring->syncp);
2151         }
2152
2153         return NETDEV_TX_OK;
2154
2155 error_report_dma_error:
2156         u64_stats_update_begin(&tx_ring->syncp);
2157         tx_ring->tx_stats.dma_mapping_err++;
2158         u64_stats_update_end(&tx_ring->syncp);
2159         netdev_warn(adapter->netdev, "failed to map skb\n");
2160
2161         tx_info->skb = NULL;
2162
2163 error_unmap_dma:
2164         if (i >= 0) {
2165                 /* save value of frag that failed */
2166                 last_frag = i;
2167
2168                 /* start back at beginning and unmap skb */
2169                 tx_info->skb = NULL;
2170                 ena_buf = tx_info->bufs;
2171                 dma_unmap_single(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
2172                                  dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
2173
2174                 /* unmap remaining mapped pages */
2175                 for (i = 0; i < last_frag; i++) {
2176                         ena_buf++;
2177                         dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
2178                                        dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
2179                 }
2180         }
2181
2182 error_drop_packet:
2183
2184         dev_kfree_skb(skb);
2185         return NETDEV_TX_OK;
2186 }
2187
2188 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
2189                             struct net_device *sb_dev,
2190                             select_queue_fallback_t fallback)
2191 {
2192         u16 qid;
2193         /* we suspect that this is good for in--kernel network services that
2194          * want to loop incoming skb rx to tx in normal user generated traffic,
2195          * most probably we will not get to this
2196          */
2197         if (skb_rx_queue_recorded(skb))
2198                 qid = skb_get_rx_queue(skb);
2199         else
2200                 qid = fallback(dev, skb, NULL);
2201
2202         return qid;
2203 }
2204
2205 static void ena_config_host_info(struct ena_com_dev *ena_dev)
2206 {
2207         struct ena_admin_host_info *host_info;
2208         int rc;
2209
2210         /* Allocate only the host info */
2211         rc = ena_com_allocate_host_info(ena_dev);
2212         if (rc) {
2213                 pr_err("Cannot allocate host info\n");
2214                 return;
2215         }
2216
2217         host_info = ena_dev->host_attr.host_info;
2218
2219         host_info->os_type = ENA_ADMIN_OS_LINUX;
2220         host_info->kernel_ver = LINUX_VERSION_CODE;
2221         strncpy(host_info->kernel_ver_str, utsname()->version,
2222                 sizeof(host_info->kernel_ver_str) - 1);
2223         host_info->os_dist = 0;
2224         strncpy(host_info->os_dist_str, utsname()->release,
2225                 sizeof(host_info->os_dist_str) - 1);
2226         host_info->driver_version =
2227                 (DRV_MODULE_VER_MAJOR) |
2228                 (DRV_MODULE_VER_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
2229                 (DRV_MODULE_VER_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT);
2230
2231         rc = ena_com_set_host_attributes(ena_dev);
2232         if (rc) {
2233                 if (rc == -EOPNOTSUPP)
2234                         pr_warn("Cannot set host attributes\n");
2235                 else
2236                         pr_err("Cannot set host attributes\n");
2237
2238                 goto err;
2239         }
2240
2241         return;
2242
2243 err:
2244         ena_com_delete_host_info(ena_dev);
2245 }
2246
2247 static void ena_config_debug_area(struct ena_adapter *adapter)
2248 {
2249         u32 debug_area_size;
2250         int rc, ss_count;
2251
2252         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
2253         if (ss_count <= 0) {
2254                 netif_err(adapter, drv, adapter->netdev,
2255                           "SS count is negative\n");
2256                 return;
2257         }
2258
2259         /* allocate 32 bytes for each string and 64bit for the value */
2260         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
2261
2262         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
2263         if (rc) {
2264                 pr_err("Cannot allocate debug area\n");
2265                 return;
2266         }
2267
2268         rc = ena_com_set_host_attributes(adapter->ena_dev);
2269         if (rc) {
2270                 if (rc == -EOPNOTSUPP)
2271                         netif_warn(adapter, drv, adapter->netdev,
2272                                    "Cannot set host attributes\n");
2273                 else
2274                         netif_err(adapter, drv, adapter->netdev,
2275                                   "Cannot set host attributes\n");
2276                 goto err;
2277         }
2278
2279         return;
2280 err:
2281         ena_com_delete_debug_area(adapter->ena_dev);
2282 }
2283
2284 static void ena_get_stats64(struct net_device *netdev,
2285                             struct rtnl_link_stats64 *stats)
2286 {
2287         struct ena_adapter *adapter = netdev_priv(netdev);
2288         struct ena_ring *rx_ring, *tx_ring;
2289         unsigned int start;
2290         u64 rx_drops;
2291         int i;
2292
2293         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2294                 return;
2295
2296         for (i = 0; i < adapter->num_queues; i++) {
2297                 u64 bytes, packets;
2298
2299                 tx_ring = &adapter->tx_ring[i];
2300
2301                 do {
2302                         start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
2303                         packets = tx_ring->tx_stats.cnt;
2304                         bytes = tx_ring->tx_stats.bytes;
2305                 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
2306
2307                 stats->tx_packets += packets;
2308                 stats->tx_bytes += bytes;
2309
2310                 rx_ring = &adapter->rx_ring[i];
2311
2312                 do {
2313                         start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
2314                         packets = rx_ring->rx_stats.cnt;
2315                         bytes = rx_ring->rx_stats.bytes;
2316                 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
2317
2318                 stats->rx_packets += packets;
2319                 stats->rx_bytes += bytes;
2320         }
2321
2322         do {
2323                 start = u64_stats_fetch_begin_irq(&adapter->syncp);
2324                 rx_drops = adapter->dev_stats.rx_drops;
2325         } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
2326
2327         stats->rx_dropped = rx_drops;
2328
2329         stats->multicast = 0;
2330         stats->collisions = 0;
2331
2332         stats->rx_length_errors = 0;
2333         stats->rx_crc_errors = 0;
2334         stats->rx_frame_errors = 0;
2335         stats->rx_fifo_errors = 0;
2336         stats->rx_missed_errors = 0;
2337         stats->tx_window_errors = 0;
2338
2339         stats->rx_errors = 0;
2340         stats->tx_errors = 0;
2341 }
2342
2343 static const struct net_device_ops ena_netdev_ops = {
2344         .ndo_open               = ena_open,
2345         .ndo_stop               = ena_close,
2346         .ndo_start_xmit         = ena_start_xmit,
2347         .ndo_select_queue       = ena_select_queue,
2348         .ndo_get_stats64        = ena_get_stats64,
2349         .ndo_tx_timeout         = ena_tx_timeout,
2350         .ndo_change_mtu         = ena_change_mtu,
2351         .ndo_set_mac_address    = NULL,
2352         .ndo_validate_addr      = eth_validate_addr,
2353 };
2354
2355 static int ena_device_validate_params(struct ena_adapter *adapter,
2356                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
2357 {
2358         struct net_device *netdev = adapter->netdev;
2359         int rc;
2360
2361         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
2362                               adapter->mac_addr);
2363         if (!rc) {
2364                 netif_err(adapter, drv, netdev,
2365                           "Error, mac address are different\n");
2366                 return -EINVAL;
2367         }
2368
2369         if ((get_feat_ctx->max_queues.max_cq_num < adapter->num_queues) ||
2370             (get_feat_ctx->max_queues.max_sq_num < adapter->num_queues)) {
2371                 netif_err(adapter, drv, netdev,
2372                           "Error, device doesn't support enough queues\n");
2373                 return -EINVAL;
2374         }
2375
2376         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
2377                 netif_err(adapter, drv, netdev,
2378                           "Error, device max mtu is smaller than netdev MTU\n");
2379                 return -EINVAL;
2380         }
2381
2382         return 0;
2383 }
2384
2385 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
2386                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
2387                            bool *wd_state)
2388 {
2389         struct device *dev = &pdev->dev;
2390         bool readless_supported;
2391         u32 aenq_groups;
2392         int dma_width;
2393         int rc;
2394
2395         rc = ena_com_mmio_reg_read_request_init(ena_dev);
2396         if (rc) {
2397                 dev_err(dev, "failed to init mmio read less\n");
2398                 return rc;
2399         }
2400
2401         /* The PCIe configuration space revision id indicate if mmio reg
2402          * read is disabled
2403          */
2404         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
2405         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
2406
2407         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
2408         if (rc) {
2409                 dev_err(dev, "Can not reset device\n");
2410                 goto err_mmio_read_less;
2411         }
2412
2413         rc = ena_com_validate_version(ena_dev);
2414         if (rc) {
2415                 dev_err(dev, "device version is too low\n");
2416                 goto err_mmio_read_less;
2417         }
2418
2419         dma_width = ena_com_get_dma_width(ena_dev);
2420         if (dma_width < 0) {
2421                 dev_err(dev, "Invalid dma width value %d", dma_width);
2422                 rc = dma_width;
2423                 goto err_mmio_read_less;
2424         }
2425
2426         rc = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2427         if (rc) {
2428                 dev_err(dev, "pci_set_dma_mask failed 0x%x\n", rc);
2429                 goto err_mmio_read_less;
2430         }
2431
2432         rc = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width));
2433         if (rc) {
2434                 dev_err(dev, "err_pci_set_consistent_dma_mask failed 0x%x\n",
2435                         rc);
2436                 goto err_mmio_read_less;
2437         }
2438
2439         /* ENA admin level init */
2440         rc = ena_com_admin_init(ena_dev, &aenq_handlers, true);
2441         if (rc) {
2442                 dev_err(dev,
2443                         "Can not initialize ena admin queue with device\n");
2444                 goto err_mmio_read_less;
2445         }
2446
2447         /* To enable the msix interrupts the driver needs to know the number
2448          * of queues. So the driver uses polling mode to retrieve this
2449          * information
2450          */
2451         ena_com_set_admin_polling_mode(ena_dev, true);
2452
2453         ena_config_host_info(ena_dev);
2454
2455         /* Get Device Attributes*/
2456         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
2457         if (rc) {
2458                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
2459                 goto err_admin_init;
2460         }
2461
2462         /* Try to turn all the available aenq groups */
2463         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
2464                 BIT(ENA_ADMIN_FATAL_ERROR) |
2465                 BIT(ENA_ADMIN_WARNING) |
2466                 BIT(ENA_ADMIN_NOTIFICATION) |
2467                 BIT(ENA_ADMIN_KEEP_ALIVE);
2468
2469         aenq_groups &= get_feat_ctx->aenq.supported_groups;
2470
2471         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
2472         if (rc) {
2473                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
2474                 goto err_admin_init;
2475         }
2476
2477         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
2478
2479         return 0;
2480
2481 err_admin_init:
2482         ena_com_delete_host_info(ena_dev);
2483         ena_com_admin_destroy(ena_dev);
2484 err_mmio_read_less:
2485         ena_com_mmio_reg_read_request_destroy(ena_dev);
2486
2487         return rc;
2488 }
2489
2490 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter,
2491                                                     int io_vectors)
2492 {
2493         struct ena_com_dev *ena_dev = adapter->ena_dev;
2494         struct device *dev = &adapter->pdev->dev;
2495         int rc;
2496
2497         rc = ena_enable_msix(adapter, io_vectors);
2498         if (rc) {
2499                 dev_err(dev, "Can not reserve msix vectors\n");
2500                 return rc;
2501         }
2502
2503         ena_setup_mgmnt_intr(adapter);
2504
2505         rc = ena_request_mgmnt_irq(adapter);
2506         if (rc) {
2507                 dev_err(dev, "Can not setup management interrupts\n");
2508                 goto err_disable_msix;
2509         }
2510
2511         ena_com_set_admin_polling_mode(ena_dev, false);
2512
2513         ena_com_admin_aenq_enable(ena_dev);
2514
2515         return 0;
2516
2517 err_disable_msix:
2518         ena_disable_msix(adapter);
2519
2520         return rc;
2521 }
2522
2523 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
2524 {
2525         struct net_device *netdev = adapter->netdev;
2526         struct ena_com_dev *ena_dev = adapter->ena_dev;
2527         bool dev_up;
2528
2529         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2530                 return;
2531
2532         netif_carrier_off(netdev);
2533
2534         del_timer_sync(&adapter->timer_service);
2535
2536         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2537         adapter->dev_up_before_reset = dev_up;
2538
2539         if (!graceful)
2540                 ena_com_set_admin_running_state(ena_dev, false);
2541
2542         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2543                 ena_down(adapter);
2544
2545         /* Before releasing the ENA resources, a device reset is required.
2546          * (to prevent the device from accessing them).
2547          * In case the reset flag is set and the device is up, ena_down()
2548          * already perform the reset, so it can be skipped.
2549          */
2550         if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
2551                 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2552
2553         ena_free_mgmnt_irq(adapter);
2554
2555         ena_disable_msix(adapter);
2556
2557         ena_com_abort_admin_commands(ena_dev);
2558
2559         ena_com_wait_for_abort_completion(ena_dev);
2560
2561         ena_com_admin_destroy(ena_dev);
2562
2563         ena_com_mmio_reg_read_request_destroy(ena_dev);
2564
2565         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
2566
2567         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2568         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2569 }
2570
2571 static int ena_restore_device(struct ena_adapter *adapter)
2572 {
2573         struct ena_com_dev_get_features_ctx get_feat_ctx;
2574         struct ena_com_dev *ena_dev = adapter->ena_dev;
2575         struct pci_dev *pdev = adapter->pdev;
2576         bool wd_state;
2577         int rc;
2578
2579         set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2580         rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
2581         if (rc) {
2582                 dev_err(&pdev->dev, "Can not initialize device\n");
2583                 goto err;
2584         }
2585         adapter->wd_state = wd_state;
2586
2587         rc = ena_device_validate_params(adapter, &get_feat_ctx);
2588         if (rc) {
2589                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
2590                 goto err_device_destroy;
2591         }
2592
2593         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2594         /* Make sure we don't have a race with AENQ Links state handler */
2595         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2596                 netif_carrier_on(adapter->netdev);
2597
2598         rc = ena_enable_msix_and_set_admin_interrupts(adapter,
2599                                                       adapter->num_queues);
2600         if (rc) {
2601                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
2602                 goto err_device_destroy;
2603         }
2604         /* If the interface was up before the reset bring it up */
2605         if (adapter->dev_up_before_reset) {
2606                 rc = ena_up(adapter);
2607                 if (rc) {
2608                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
2609                         goto err_disable_msix;
2610                 }
2611         }
2612
2613         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2614         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
2615         dev_err(&pdev->dev, "Device reset completed successfully\n");
2616
2617         return rc;
2618 err_disable_msix:
2619         ena_free_mgmnt_irq(adapter);
2620         ena_disable_msix(adapter);
2621 err_device_destroy:
2622         ena_com_admin_destroy(ena_dev);
2623 err:
2624         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
2625         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
2626         dev_err(&pdev->dev,
2627                 "Reset attempt failed. Can not reset the device\n");
2628
2629         return rc;
2630 }
2631
2632 static void ena_fw_reset_device(struct work_struct *work)
2633 {
2634         struct ena_adapter *adapter =
2635                 container_of(work, struct ena_adapter, reset_task);
2636         struct pci_dev *pdev = adapter->pdev;
2637
2638         if (unlikely(!test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2639                 dev_err(&pdev->dev,
2640                         "device reset schedule while reset bit is off\n");
2641                 return;
2642         }
2643         rtnl_lock();
2644         ena_destroy_device(adapter, false);
2645         ena_restore_device(adapter);
2646         rtnl_unlock();
2647 }
2648
2649 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
2650                                         struct ena_ring *rx_ring)
2651 {
2652         if (likely(rx_ring->first_interrupt))
2653                 return 0;
2654
2655         if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
2656                 return 0;
2657
2658         rx_ring->no_interrupt_event_cnt++;
2659
2660         if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
2661                 netif_err(adapter, rx_err, adapter->netdev,
2662                           "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
2663                           rx_ring->qid);
2664                 adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
2665                 smp_mb__before_atomic();
2666                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2667                 return -EIO;
2668         }
2669
2670         return 0;
2671 }
2672
2673 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
2674                                           struct ena_ring *tx_ring)
2675 {
2676         struct ena_tx_buffer *tx_buf;
2677         unsigned long last_jiffies;
2678         u32 missed_tx = 0;
2679         int i, rc = 0;
2680
2681         for (i = 0; i < tx_ring->ring_size; i++) {
2682                 tx_buf = &tx_ring->tx_buffer_info[i];
2683                 last_jiffies = tx_buf->last_jiffies;
2684
2685                 if (last_jiffies == 0)
2686                         /* no pending Tx at this location */
2687                         continue;
2688
2689                 if (unlikely(!tx_ring->first_interrupt && time_is_before_jiffies(last_jiffies +
2690                              2 * adapter->missing_tx_completion_to))) {
2691                         /* If after graceful period interrupt is still not
2692                          * received, we schedule a reset
2693                          */
2694                         netif_err(adapter, tx_err, adapter->netdev,
2695                                   "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
2696                                   tx_ring->qid);
2697                         adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
2698                         smp_mb__before_atomic();
2699                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2700                         return -EIO;
2701                 }
2702
2703                 if (unlikely(time_is_before_jiffies(last_jiffies +
2704                                 adapter->missing_tx_completion_to))) {
2705                         if (!tx_buf->print_once)
2706                                 netif_notice(adapter, tx_err, adapter->netdev,
2707                                              "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
2708                                              tx_ring->qid, i);
2709
2710                         tx_buf->print_once = 1;
2711                         missed_tx++;
2712                 }
2713         }
2714
2715         if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
2716                 netif_err(adapter, tx_err, adapter->netdev,
2717                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
2718                           missed_tx,
2719                           adapter->missing_tx_completion_threshold);
2720                 adapter->reset_reason =
2721                         ENA_REGS_RESET_MISS_TX_CMPL;
2722                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2723                 rc = -EIO;
2724         }
2725
2726         u64_stats_update_begin(&tx_ring->syncp);
2727         tx_ring->tx_stats.missed_tx = missed_tx;
2728         u64_stats_update_end(&tx_ring->syncp);
2729
2730         return rc;
2731 }
2732
2733 static void check_for_missing_completions(struct ena_adapter *adapter)
2734 {
2735         struct ena_ring *tx_ring;
2736         struct ena_ring *rx_ring;
2737         int i, budget, rc;
2738
2739         /* Make sure the driver doesn't turn the device in other process */
2740         smp_rmb();
2741
2742         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2743                 return;
2744
2745         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2746                 return;
2747
2748         if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
2749                 return;
2750
2751         budget = ENA_MONITORED_TX_QUEUES;
2752
2753         for (i = adapter->last_monitored_tx_qid; i < adapter->num_queues; i++) {
2754                 tx_ring = &adapter->tx_ring[i];
2755                 rx_ring = &adapter->rx_ring[i];
2756
2757                 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
2758                 if (unlikely(rc))
2759                         return;
2760
2761                 rc = check_for_rx_interrupt_queue(adapter, rx_ring);
2762                 if (unlikely(rc))
2763                         return;
2764
2765                 budget--;
2766                 if (!budget)
2767                         break;
2768         }
2769
2770         adapter->last_monitored_tx_qid = i % adapter->num_queues;
2771 }
2772
2773 /* trigger napi schedule after 2 consecutive detections */
2774 #define EMPTY_RX_REFILL 2
2775 /* For the rare case where the device runs out of Rx descriptors and the
2776  * napi handler failed to refill new Rx descriptors (due to a lack of memory
2777  * for example).
2778  * This case will lead to a deadlock:
2779  * The device won't send interrupts since all the new Rx packets will be dropped
2780  * The napi handler won't allocate new Rx descriptors so the device will be
2781  * able to send new packets.
2782  *
2783  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
2784  * It is recommended to have at least 512MB, with a minimum of 128MB for
2785  * constrained environment).
2786  *
2787  * When such a situation is detected - Reschedule napi
2788  */
2789 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
2790 {
2791         struct ena_ring *rx_ring;
2792         int i, refill_required;
2793
2794         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2795                 return;
2796
2797         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
2798                 return;
2799
2800         for (i = 0; i < adapter->num_queues; i++) {
2801                 rx_ring = &adapter->rx_ring[i];
2802
2803                 refill_required =
2804                         ena_com_sq_empty_space(rx_ring->ena_com_io_sq);
2805                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
2806                         rx_ring->empty_rx_queue++;
2807
2808                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
2809                                 u64_stats_update_begin(&rx_ring->syncp);
2810                                 rx_ring->rx_stats.empty_rx_ring++;
2811                                 u64_stats_update_end(&rx_ring->syncp);
2812
2813                                 netif_err(adapter, drv, adapter->netdev,
2814                                           "trigger refill for ring %d\n", i);
2815
2816                                 napi_schedule(rx_ring->napi);
2817                                 rx_ring->empty_rx_queue = 0;
2818                         }
2819                 } else {
2820                         rx_ring->empty_rx_queue = 0;
2821                 }
2822         }
2823 }
2824
2825 /* Check for keep alive expiration */
2826 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
2827 {
2828         unsigned long keep_alive_expired;
2829
2830         if (!adapter->wd_state)
2831                 return;
2832
2833         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2834                 return;
2835
2836         keep_alive_expired = round_jiffies(adapter->last_keep_alive_jiffies +
2837                                            adapter->keep_alive_timeout);
2838         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
2839                 netif_err(adapter, drv, adapter->netdev,
2840                           "Keep alive watchdog timeout.\n");
2841                 u64_stats_update_begin(&adapter->syncp);
2842                 adapter->dev_stats.wd_expired++;
2843                 u64_stats_update_end(&adapter->syncp);
2844                 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
2845                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2846         }
2847 }
2848
2849 static void check_for_admin_com_state(struct ena_adapter *adapter)
2850 {
2851         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
2852                 netif_err(adapter, drv, adapter->netdev,
2853                           "ENA admin queue is not in running state!\n");
2854                 u64_stats_update_begin(&adapter->syncp);
2855                 adapter->dev_stats.admin_q_pause++;
2856                 u64_stats_update_end(&adapter->syncp);
2857                 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
2858                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
2859         }
2860 }
2861
2862 static void ena_update_hints(struct ena_adapter *adapter,
2863                              struct ena_admin_ena_hw_hints *hints)
2864 {
2865         struct net_device *netdev = adapter->netdev;
2866
2867         if (hints->admin_completion_tx_timeout)
2868                 adapter->ena_dev->admin_queue.completion_timeout =
2869                         hints->admin_completion_tx_timeout * 1000;
2870
2871         if (hints->mmio_read_timeout)
2872                 /* convert to usec */
2873                 adapter->ena_dev->mmio_read.reg_read_to =
2874                         hints->mmio_read_timeout * 1000;
2875
2876         if (hints->missed_tx_completion_count_threshold_to_reset)
2877                 adapter->missing_tx_completion_threshold =
2878                         hints->missed_tx_completion_count_threshold_to_reset;
2879
2880         if (hints->missing_tx_completion_timeout) {
2881                 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2882                         adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
2883                 else
2884                         adapter->missing_tx_completion_to =
2885                                 msecs_to_jiffies(hints->missing_tx_completion_timeout);
2886         }
2887
2888         if (hints->netdev_wd_timeout)
2889                 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
2890
2891         if (hints->driver_watchdog_timeout) {
2892                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
2893                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
2894                 else
2895                         adapter->keep_alive_timeout =
2896                                 msecs_to_jiffies(hints->driver_watchdog_timeout);
2897         }
2898 }
2899
2900 static void ena_update_host_info(struct ena_admin_host_info *host_info,
2901                                  struct net_device *netdev)
2902 {
2903         host_info->supported_network_features[0] =
2904                 netdev->features & GENMASK_ULL(31, 0);
2905         host_info->supported_network_features[1] =
2906                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
2907 }
2908
2909 static void ena_timer_service(struct timer_list *t)
2910 {
2911         struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
2912         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
2913         struct ena_admin_host_info *host_info =
2914                 adapter->ena_dev->host_attr.host_info;
2915
2916         check_for_missing_keep_alive(adapter);
2917
2918         check_for_admin_com_state(adapter);
2919
2920         check_for_missing_completions(adapter);
2921
2922         check_for_empty_rx_ring(adapter);
2923
2924         if (debug_area)
2925                 ena_dump_stats_to_buf(adapter, debug_area);
2926
2927         if (host_info)
2928                 ena_update_host_info(host_info, adapter->netdev);
2929
2930         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2931                 netif_err(adapter, drv, adapter->netdev,
2932                           "Trigger reset is on\n");
2933                 ena_dump_stats_to_dmesg(adapter);
2934                 queue_work(ena_wq, &adapter->reset_task);
2935                 return;
2936         }
2937
2938         /* Reset the timer */
2939         mod_timer(&adapter->timer_service, jiffies + HZ);
2940 }
2941
2942 static int ena_calc_io_queue_num(struct pci_dev *pdev,
2943                                  struct ena_com_dev *ena_dev,
2944                                  struct ena_com_dev_get_features_ctx *get_feat_ctx)
2945 {
2946         int io_sq_num, io_queue_num;
2947
2948         /* In case of LLQ use the llq number in the get feature cmd */
2949         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2950                 io_sq_num = get_feat_ctx->max_queues.max_llq_num;
2951
2952                 if (io_sq_num == 0) {
2953                         dev_err(&pdev->dev,
2954                                 "Trying to use LLQ but llq_num is 0. Fall back into regular queues\n");
2955
2956                         ena_dev->tx_mem_queue_type =
2957                                 ENA_ADMIN_PLACEMENT_POLICY_HOST;
2958                         io_sq_num = get_feat_ctx->max_queues.max_sq_num;
2959                 }
2960         } else {
2961                 io_sq_num = get_feat_ctx->max_queues.max_sq_num;
2962         }
2963
2964         io_queue_num = min_t(int, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
2965         io_queue_num = min_t(int, io_queue_num, io_sq_num);
2966         io_queue_num = min_t(int, io_queue_num,
2967                              get_feat_ctx->max_queues.max_cq_num);
2968         /* 1 IRQ for for mgmnt and 1 IRQs for each IO direction */
2969         io_queue_num = min_t(int, io_queue_num, pci_msix_vec_count(pdev) - 1);
2970         if (unlikely(!io_queue_num)) {
2971                 dev_err(&pdev->dev, "The device doesn't have io queues\n");
2972                 return -EFAULT;
2973         }
2974
2975         return io_queue_num;
2976 }
2977
2978 static void ena_set_push_mode(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
2979                               struct ena_com_dev_get_features_ctx *get_feat_ctx)
2980 {
2981         bool has_mem_bar;
2982
2983         has_mem_bar = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(ENA_MEM_BAR);
2984
2985         /* Enable push mode if device supports LLQ */
2986         if (has_mem_bar && (get_feat_ctx->max_queues.max_llq_num > 0))
2987                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_DEV;
2988         else
2989                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2990 }
2991
2992 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
2993                                  struct net_device *netdev)
2994 {
2995         netdev_features_t dev_features = 0;
2996
2997         /* Set offload features */
2998         if (feat->offload.tx &
2999                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
3000                 dev_features |= NETIF_F_IP_CSUM;
3001
3002         if (feat->offload.tx &
3003                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
3004                 dev_features |= NETIF_F_IPV6_CSUM;
3005
3006         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
3007                 dev_features |= NETIF_F_TSO;
3008
3009         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
3010                 dev_features |= NETIF_F_TSO6;
3011
3012         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
3013                 dev_features |= NETIF_F_TSO_ECN;
3014
3015         if (feat->offload.rx_supported &
3016                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
3017                 dev_features |= NETIF_F_RXCSUM;
3018
3019         if (feat->offload.rx_supported &
3020                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
3021                 dev_features |= NETIF_F_RXCSUM;
3022
3023         netdev->features =
3024                 dev_features |
3025                 NETIF_F_SG |
3026                 NETIF_F_RXHASH |
3027                 NETIF_F_HIGHDMA;
3028
3029         netdev->hw_features |= netdev->features;
3030         netdev->vlan_features |= netdev->features;
3031 }
3032
3033 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
3034                                      struct ena_com_dev_get_features_ctx *feat)
3035 {
3036         struct net_device *netdev = adapter->netdev;
3037
3038         /* Copy mac address */
3039         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
3040                 eth_hw_addr_random(netdev);
3041                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
3042         } else {
3043                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
3044                 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
3045         }
3046
3047         /* Set offload features */
3048         ena_set_dev_offloads(feat, netdev);
3049
3050         adapter->max_mtu = feat->dev_attr.max_mtu;
3051         netdev->max_mtu = adapter->max_mtu;
3052         netdev->min_mtu = ENA_MIN_MTU;
3053 }
3054
3055 static int ena_rss_init_default(struct ena_adapter *adapter)
3056 {
3057         struct ena_com_dev *ena_dev = adapter->ena_dev;
3058         struct device *dev = &adapter->pdev->dev;
3059         int rc, i;
3060         u32 val;
3061
3062         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
3063         if (unlikely(rc)) {
3064                 dev_err(dev, "Cannot init indirect table\n");
3065                 goto err_rss_init;
3066         }
3067
3068         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
3069                 val = ethtool_rxfh_indir_default(i, adapter->num_queues);
3070                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
3071                                                        ENA_IO_RXQ_IDX(val));
3072                 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3073                         dev_err(dev, "Cannot fill indirect table\n");
3074                         goto err_fill_indir;
3075                 }
3076         }
3077
3078         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_CRC32, NULL,
3079                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
3080         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3081                 dev_err(dev, "Cannot fill hash function\n");
3082                 goto err_fill_indir;
3083         }
3084
3085         rc = ena_com_set_default_hash_ctrl(ena_dev);
3086         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
3087                 dev_err(dev, "Cannot fill hash control\n");
3088                 goto err_fill_indir;
3089         }
3090
3091         return 0;
3092
3093 err_fill_indir:
3094         ena_com_rss_destroy(ena_dev);
3095 err_rss_init:
3096
3097         return rc;
3098 }
3099
3100 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3101 {
3102         int release_bars;
3103
3104         if (ena_dev->mem_bar)
3105                 devm_iounmap(&pdev->dev, ena_dev->mem_bar);
3106
3107         if (ena_dev->reg_bar)
3108                 devm_iounmap(&pdev->dev, ena_dev->reg_bar);
3109
3110         release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
3111         pci_release_selected_regions(pdev, release_bars);
3112 }
3113
3114 static int ena_calc_queue_size(struct pci_dev *pdev,
3115                                struct ena_com_dev *ena_dev,
3116                                u16 *max_tx_sgl_size,
3117                                u16 *max_rx_sgl_size,
3118                                struct ena_com_dev_get_features_ctx *get_feat_ctx)
3119 {
3120         u32 queue_size = ENA_DEFAULT_RING_SIZE;
3121
3122         queue_size = min_t(u32, queue_size,
3123                            get_feat_ctx->max_queues.max_cq_depth);
3124         queue_size = min_t(u32, queue_size,
3125                            get_feat_ctx->max_queues.max_sq_depth);
3126
3127         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3128                 queue_size = min_t(u32, queue_size,
3129                                    get_feat_ctx->max_queues.max_llq_depth);
3130
3131         queue_size = rounddown_pow_of_two(queue_size);
3132
3133         if (unlikely(!queue_size)) {
3134                 dev_err(&pdev->dev, "Invalid queue size\n");
3135                 return -EFAULT;
3136         }
3137
3138         *max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3139                                  get_feat_ctx->max_queues.max_packet_tx_descs);
3140         *max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
3141                                  get_feat_ctx->max_queues.max_packet_rx_descs);
3142
3143         return queue_size;
3144 }
3145
3146 /* ena_probe - Device Initialization Routine
3147  * @pdev: PCI device information struct
3148  * @ent: entry in ena_pci_tbl
3149  *
3150  * Returns 0 on success, negative on failure
3151  *
3152  * ena_probe initializes an adapter identified by a pci_dev structure.
3153  * The OS initialization, configuring of the adapter private structure,
3154  * and a hardware reset occur.
3155  */
3156 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3157 {
3158         struct ena_com_dev_get_features_ctx get_feat_ctx;
3159         static int version_printed;
3160         struct net_device *netdev;
3161         struct ena_adapter *adapter;
3162         struct ena_com_dev *ena_dev = NULL;
3163         static int adapters_found;
3164         int io_queue_num, bars, rc;
3165         int queue_size;
3166         u16 tx_sgl_size = 0;
3167         u16 rx_sgl_size = 0;
3168         bool wd_state;
3169
3170         dev_dbg(&pdev->dev, "%s\n", __func__);
3171
3172         if (version_printed++ == 0)
3173                 dev_info(&pdev->dev, "%s", version);
3174
3175         rc = pci_enable_device_mem(pdev);
3176         if (rc) {
3177                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
3178                 return rc;
3179         }
3180
3181         pci_set_master(pdev);
3182
3183         ena_dev = vzalloc(sizeof(*ena_dev));
3184         if (!ena_dev) {
3185                 rc = -ENOMEM;
3186                 goto err_disable_device;
3187         }
3188
3189         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
3190         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
3191         if (rc) {
3192                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
3193                         rc);
3194                 goto err_free_ena_dev;
3195         }
3196
3197         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
3198                                         pci_resource_start(pdev, ENA_REG_BAR),
3199                                         pci_resource_len(pdev, ENA_REG_BAR));
3200         if (!ena_dev->reg_bar) {
3201                 dev_err(&pdev->dev, "failed to remap regs bar\n");
3202                 rc = -EFAULT;
3203                 goto err_free_region;
3204         }
3205
3206         ena_dev->dmadev = &pdev->dev;
3207
3208         rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
3209         if (rc) {
3210                 dev_err(&pdev->dev, "ena device init failed\n");
3211                 if (rc == -ETIME)
3212                         rc = -EPROBE_DEFER;
3213                 goto err_free_region;
3214         }
3215
3216         ena_set_push_mode(pdev, ena_dev, &get_feat_ctx);
3217
3218         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3219                 ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3220                                                    pci_resource_start(pdev, ENA_MEM_BAR),
3221                                                    pci_resource_len(pdev, ENA_MEM_BAR));
3222                 if (!ena_dev->mem_bar) {
3223                         rc = -EFAULT;
3224                         goto err_device_destroy;
3225                 }
3226         }
3227
3228         /* initial Tx interrupt delay, Assumes 1 usec granularity.
3229         * Updated during device initialization with the real granularity
3230         */
3231         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
3232         io_queue_num = ena_calc_io_queue_num(pdev, ena_dev, &get_feat_ctx);
3233         queue_size = ena_calc_queue_size(pdev, ena_dev, &tx_sgl_size,
3234                                          &rx_sgl_size, &get_feat_ctx);
3235         if ((queue_size <= 0) || (io_queue_num <= 0)) {
3236                 rc = -EFAULT;
3237                 goto err_device_destroy;
3238         }
3239
3240         dev_info(&pdev->dev, "creating %d io queues. queue size: %d\n",
3241                  io_queue_num, queue_size);
3242
3243         /* dev zeroed in init_etherdev */
3244         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), io_queue_num);
3245         if (!netdev) {
3246                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
3247                 rc = -ENOMEM;
3248                 goto err_device_destroy;
3249         }
3250
3251         SET_NETDEV_DEV(netdev, &pdev->dev);
3252
3253         adapter = netdev_priv(netdev);
3254         pci_set_drvdata(pdev, adapter);
3255
3256         adapter->ena_dev = ena_dev;
3257         adapter->netdev = netdev;
3258         adapter->pdev = pdev;
3259
3260         ena_set_conf_feat_params(adapter, &get_feat_ctx);
3261
3262         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3263         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3264
3265         adapter->tx_ring_size = queue_size;
3266         adapter->rx_ring_size = queue_size;
3267
3268         adapter->max_tx_sgl_size = tx_sgl_size;
3269         adapter->max_rx_sgl_size = rx_sgl_size;
3270
3271         adapter->num_queues = io_queue_num;
3272         adapter->last_monitored_tx_qid = 0;
3273
3274         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
3275         adapter->wd_state = wd_state;
3276
3277         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
3278
3279         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
3280         if (rc) {
3281                 dev_err(&pdev->dev,
3282                         "Failed to query interrupt moderation feature\n");
3283                 goto err_netdev_destroy;
3284         }
3285         ena_init_io_rings(adapter);
3286
3287         netdev->netdev_ops = &ena_netdev_ops;
3288         netdev->watchdog_timeo = TX_TIMEOUT;
3289         ena_set_ethtool_ops(netdev);
3290
3291         netdev->priv_flags |= IFF_UNICAST_FLT;
3292
3293         u64_stats_init(&adapter->syncp);
3294
3295         rc = ena_enable_msix_and_set_admin_interrupts(adapter, io_queue_num);
3296         if (rc) {
3297                 dev_err(&pdev->dev,
3298                         "Failed to enable and set the admin interrupts\n");
3299                 goto err_worker_destroy;
3300         }
3301         rc = ena_rss_init_default(adapter);
3302         if (rc && (rc != -EOPNOTSUPP)) {
3303                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
3304                 goto err_free_msix;
3305         }
3306
3307         ena_config_debug_area(adapter);
3308
3309         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
3310
3311         netif_carrier_off(netdev);
3312
3313         rc = register_netdev(netdev);
3314         if (rc) {
3315                 dev_err(&pdev->dev, "Cannot register net device\n");
3316                 goto err_rss;
3317         }
3318
3319         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
3320
3321         adapter->last_keep_alive_jiffies = jiffies;
3322         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
3323         adapter->missing_tx_completion_to = TX_TIMEOUT;
3324         adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
3325
3326         ena_update_hints(adapter, &get_feat_ctx.hw_hints);
3327
3328         timer_setup(&adapter->timer_service, ena_timer_service, 0);
3329         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3330
3331         dev_info(&pdev->dev, "%s found at mem %lx, mac addr %pM Queues %d\n",
3332                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
3333                  netdev->dev_addr, io_queue_num);
3334
3335         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3336
3337         adapters_found++;
3338
3339         return 0;
3340
3341 err_rss:
3342         ena_com_delete_debug_area(ena_dev);
3343         ena_com_rss_destroy(ena_dev);
3344 err_free_msix:
3345         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
3346         ena_free_mgmnt_irq(adapter);
3347         ena_disable_msix(adapter);
3348 err_worker_destroy:
3349         ena_com_destroy_interrupt_moderation(ena_dev);
3350         del_timer(&adapter->timer_service);
3351 err_netdev_destroy:
3352         free_netdev(netdev);
3353 err_device_destroy:
3354         ena_com_delete_host_info(ena_dev);
3355         ena_com_admin_destroy(ena_dev);
3356 err_free_region:
3357         ena_release_bars(ena_dev, pdev);
3358 err_free_ena_dev:
3359         vfree(ena_dev);
3360 err_disable_device:
3361         pci_disable_device(pdev);
3362         return rc;
3363 }
3364
3365 /*****************************************************************************/
3366
3367 /* ena_remove - Device Removal Routine
3368  * @pdev: PCI device information struct
3369  *
3370  * ena_remove is called by the PCI subsystem to alert the driver
3371  * that it should release a PCI device.
3372  */
3373 static void ena_remove(struct pci_dev *pdev)
3374 {
3375         struct ena_adapter *adapter = pci_get_drvdata(pdev);
3376         struct ena_com_dev *ena_dev;
3377         struct net_device *netdev;
3378
3379         ena_dev = adapter->ena_dev;
3380         netdev = adapter->netdev;
3381
3382 #ifdef CONFIG_RFS_ACCEL
3383         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
3384                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
3385                 netdev->rx_cpu_rmap = NULL;
3386         }
3387 #endif /* CONFIG_RFS_ACCEL */
3388         del_timer_sync(&adapter->timer_service);
3389
3390         cancel_work_sync(&adapter->reset_task);
3391
3392         unregister_netdev(netdev);
3393
3394         /* If the device is running then we want to make sure the device will be
3395          * reset to make sure no more events will be issued by the device.
3396          */
3397         if (test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3398                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3399
3400         rtnl_lock();
3401         ena_destroy_device(adapter, true);
3402         rtnl_unlock();
3403
3404         free_netdev(netdev);
3405
3406         ena_com_rss_destroy(ena_dev);
3407
3408         ena_com_delete_debug_area(ena_dev);
3409
3410         ena_com_delete_host_info(ena_dev);
3411
3412         ena_release_bars(ena_dev, pdev);
3413
3414         pci_disable_device(pdev);
3415
3416         ena_com_destroy_interrupt_moderation(ena_dev);
3417
3418         vfree(ena_dev);
3419 }
3420
3421 #ifdef CONFIG_PM
3422 /* ena_suspend - PM suspend callback
3423  * @pdev: PCI device information struct
3424  * @state:power state
3425  */
3426 static int ena_suspend(struct pci_dev *pdev,  pm_message_t state)
3427 {
3428         struct ena_adapter *adapter = pci_get_drvdata(pdev);
3429
3430         u64_stats_update_begin(&adapter->syncp);
3431         adapter->dev_stats.suspend++;
3432         u64_stats_update_end(&adapter->syncp);
3433
3434         rtnl_lock();
3435         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3436                 dev_err(&pdev->dev,
3437                         "ignoring device reset request as the device is being suspended\n");
3438                 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3439         }
3440         ena_destroy_device(adapter, true);
3441         rtnl_unlock();
3442         return 0;
3443 }
3444
3445 /* ena_resume - PM resume callback
3446  * @pdev: PCI device information struct
3447  *
3448  */
3449 static int ena_resume(struct pci_dev *pdev)
3450 {
3451         struct ena_adapter *adapter = pci_get_drvdata(pdev);
3452         int rc;
3453
3454         u64_stats_update_begin(&adapter->syncp);
3455         adapter->dev_stats.resume++;
3456         u64_stats_update_end(&adapter->syncp);
3457
3458         rtnl_lock();
3459         rc = ena_restore_device(adapter);
3460         rtnl_unlock();
3461         return rc;
3462 }
3463 #endif
3464
3465 static struct pci_driver ena_pci_driver = {
3466         .name           = DRV_MODULE_NAME,
3467         .id_table       = ena_pci_tbl,
3468         .probe          = ena_probe,
3469         .remove         = ena_remove,
3470 #ifdef CONFIG_PM
3471         .suspend    = ena_suspend,
3472         .resume     = ena_resume,
3473 #endif
3474         .sriov_configure = pci_sriov_configure_simple,
3475 };
3476
3477 static int __init ena_init(void)
3478 {
3479         pr_info("%s", version);
3480
3481         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
3482         if (!ena_wq) {
3483                 pr_err("Failed to create workqueue\n");
3484                 return -ENOMEM;
3485         }
3486
3487         return pci_register_driver(&ena_pci_driver);
3488 }
3489
3490 static void __exit ena_cleanup(void)
3491 {
3492         pci_unregister_driver(&ena_pci_driver);
3493
3494         if (ena_wq) {
3495                 destroy_workqueue(ena_wq);
3496                 ena_wq = NULL;
3497         }
3498 }
3499
3500 /******************************************************************************
3501  ******************************** AENQ Handlers *******************************
3502  *****************************************************************************/
3503 /* ena_update_on_link_change:
3504  * Notify the network interface about the change in link status
3505  */
3506 static void ena_update_on_link_change(void *adapter_data,
3507                                       struct ena_admin_aenq_entry *aenq_e)
3508 {
3509         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3510         struct ena_admin_aenq_link_change_desc *aenq_desc =
3511                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
3512         int status = aenq_desc->flags &
3513                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
3514
3515         if (status) {
3516                 netdev_dbg(adapter->netdev, "%s\n", __func__);
3517                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3518                 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
3519                         netif_carrier_on(adapter->netdev);
3520         } else {
3521                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
3522                 netif_carrier_off(adapter->netdev);
3523         }
3524 }
3525
3526 static void ena_keep_alive_wd(void *adapter_data,
3527                               struct ena_admin_aenq_entry *aenq_e)
3528 {
3529         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3530         struct ena_admin_aenq_keep_alive_desc *desc;
3531         u64 rx_drops;
3532
3533         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
3534         adapter->last_keep_alive_jiffies = jiffies;
3535
3536         rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
3537
3538         u64_stats_update_begin(&adapter->syncp);
3539         adapter->dev_stats.rx_drops = rx_drops;
3540         u64_stats_update_end(&adapter->syncp);
3541 }
3542
3543 static void ena_notification(void *adapter_data,
3544                              struct ena_admin_aenq_entry *aenq_e)
3545 {
3546         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
3547         struct ena_admin_ena_hw_hints *hints;
3548
3549         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
3550              "Invalid group(%x) expected %x\n",
3551              aenq_e->aenq_common_desc.group,
3552              ENA_ADMIN_NOTIFICATION);
3553
3554         switch (aenq_e->aenq_common_desc.syndrom) {
3555         case ENA_ADMIN_UPDATE_HINTS:
3556                 hints = (struct ena_admin_ena_hw_hints *)
3557                         (&aenq_e->inline_data_w4);
3558                 ena_update_hints(adapter, hints);
3559                 break;
3560         default:
3561                 netif_err(adapter, drv, adapter->netdev,
3562                           "Invalid aenq notification link state %d\n",
3563                           aenq_e->aenq_common_desc.syndrom);
3564         }
3565 }
3566
3567 /* This handler will called for unknown event group or unimplemented handlers*/
3568 static void unimplemented_aenq_handler(void *data,
3569                                        struct ena_admin_aenq_entry *aenq_e)
3570 {
3571         struct ena_adapter *adapter = (struct ena_adapter *)data;
3572
3573         netif_err(adapter, drv, adapter->netdev,
3574                   "Unknown event was received or event with unimplemented handler\n");
3575 }
3576
3577 static struct ena_aenq_handlers aenq_handlers = {
3578         .handlers = {
3579                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
3580                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
3581                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
3582         },
3583         .unimplemented_handler = unimplemented_aenq_handler
3584 };
3585
3586 module_init(ena_init);
3587 module_exit(ena_cleanup);