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

   1 // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
   2 /*
   3  * Copyright 2015-2020 Amazon.com, Inc. or its affiliates. All rights reserved.
   4  */
   5
   6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   7
   8 #ifdef CONFIG_RFS_ACCEL
   9 #include <linux/cpu_rmap.h>
  10 #endif /* CONFIG_RFS_ACCEL */
  11 #include <linux/ethtool.h>
  12 #include <linux/kernel.h>
  13 #include <linux/module.h>
  14 #include <linux/numa.h>
  15 #include <linux/pci.h>
  16 #include <linux/utsname.h>
  17 #include <linux/version.h>
  18 #include <linux/vmalloc.h>
  19 #include <net/ip.h>
  20
  21 #include "ena_netdev.h"
  22 #include <linux/bpf_trace.h>
  23 #include "ena_pci_id_tbl.h"
  24
  25 MODULE_AUTHOR("Amazon.com, Inc. or its affiliates");
  26 MODULE_DESCRIPTION(DEVICE_NAME);
  27 MODULE_LICENSE("GPL");
  28
  29 /* Time in jiffies before concluding the transmitter is hung. */
  30 #define TX_TIMEOUT  (5 * HZ)
  31
  32 #define ENA_MAX_RINGS min_t(unsigned int, ENA_MAX_NUM_IO_QUEUES, num_possible_cpus())
  33
  34 #define ENA_NAPI_BUDGET 64
  35
  36 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_IFUP | \
  37                 NETIF_MSG_TX_DONE | NETIF_MSG_TX_ERR | NETIF_MSG_RX_ERR)
  38 static int debug = -1;
  39 module_param(debug, int, 0);
  40 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  41
  42 static struct ena_aenq_handlers aenq_handlers;
  43
  44 static struct workqueue_struct *ena_wq;
  45
  46 MODULE_DEVICE_TABLE(pci, ena_pci_tbl);
  47
  48 static int ena_rss_init_default(struct ena_adapter *adapter);
  49 static void check_for_admin_com_state(struct ena_adapter *adapter);
  50 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful);
  51 static int ena_restore_device(struct ena_adapter *adapter);
  52
  53 static void ena_init_io_rings(struct ena_adapter *adapter,
  54                               int first_index, int count);
  55 static void ena_init_napi_in_range(struct ena_adapter *adapter, int first_index,
  56                                    int count);
  57 static void ena_del_napi_in_range(struct ena_adapter *adapter, int first_index,
  58                                   int count);
  59 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid);
  60 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
  61                                            int first_index,
  62                                            int count);
  63 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid);
  64 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid);
  65 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget);
  66 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter);
  67 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter);
  68 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
  69                                       int first_index, int count);
  70 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
  71                                      int first_index, int count);
  72 static int ena_up(struct ena_adapter *adapter);
  73 static void ena_down(struct ena_adapter *adapter);
  74 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
  75                                  struct ena_ring *rx_ring);
  76 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
  77                                       struct ena_ring *rx_ring);
  78 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
  79                               struct ena_tx_buffer *tx_info);
  80 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
  81                                             int first_index, int count);
  82
  83 /* Increase a stat by cnt while holding syncp seqlock on 32bit machines */
  84 static void ena_increase_stat(u64 *statp, u64 cnt,
  85                               struct u64_stats_sync *syncp)
  86 {
  87         u64_stats_update_begin(syncp);
  88         (*statp) += cnt;
  89         u64_stats_update_end(syncp);
  90 }
  91
  92 static void ena_tx_timeout(struct net_device *dev, unsigned int txqueue)
  93 {
  94         struct ena_adapter *adapter = netdev_priv(dev);
  95
  96         /* Change the state of the device to trigger reset
  97          * Check that we are not in the middle or a trigger already
  98          */
  99
 100         if (test_and_set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
 101                 return;
 102
 103         adapter->reset_reason = ENA_REGS_RESET_OS_NETDEV_WD;
 104         ena_increase_stat(&adapter->dev_stats.tx_timeout, 1, &adapter->syncp);
 105
 106         netif_err(adapter, tx_err, dev, "Transmit time out\n");
 107 }
 108
 109 static void update_rx_ring_mtu(struct ena_adapter *adapter, int mtu)
 110 {
 111         int i;
 112
 113         for (i = 0; i < adapter->num_io_queues; i++)
 114                 adapter->rx_ring[i].mtu = mtu;
 115 }
 116
 117 static int ena_change_mtu(struct net_device *dev, int new_mtu)
 118 {
 119         struct ena_adapter *adapter = netdev_priv(dev);
 120         int ret;
 121
 122         ret = ena_com_set_dev_mtu(adapter->ena_dev, new_mtu);
 123         if (!ret) {
 124                 netif_dbg(adapter, drv, dev, "Set MTU to %d\n", new_mtu);
 125                 update_rx_ring_mtu(adapter, new_mtu);
 126                 dev->mtu = new_mtu;
 127         } else {
 128                 netif_err(adapter, drv, dev, "Failed to set MTU to %d\n",
 129                           new_mtu);
 130         }
 131
 132         return ret;
 133 }
 134
 135 static int ena_xmit_common(struct net_device *dev,
 136                            struct ena_ring *ring,
 137                            struct ena_tx_buffer *tx_info,
 138                            struct ena_com_tx_ctx *ena_tx_ctx,
 139                            u16 next_to_use,
 140                            u32 bytes)
 141 {
 142         struct ena_adapter *adapter = netdev_priv(dev);
 143         int rc, nb_hw_desc;
 144
 145         if (unlikely(ena_com_is_doorbell_needed(ring->ena_com_io_sq,
 146                                                 ena_tx_ctx))) {
 147                 netif_dbg(adapter, tx_queued, dev,
 148                           "llq tx max burst size of queue %d achieved, writing doorbell to send burst\n",
 149                           ring->qid);
 150                 ena_com_write_sq_doorbell(ring->ena_com_io_sq);
 151         }
 152
 153         /* prepare the packet's descriptors to dma engine */
 154         rc = ena_com_prepare_tx(ring->ena_com_io_sq, ena_tx_ctx,
 155                                 &nb_hw_desc);
 156
 157         /* In case there isn't enough space in the queue for the packet,
 158          * we simply drop it. All other failure reasons of
 159          * ena_com_prepare_tx() are fatal and therefore require a device reset.
 160          */
 161         if (unlikely(rc)) {
 162                 netif_err(adapter, tx_queued, dev,
 163                           "Failed to prepare tx bufs\n");
 164                 ena_increase_stat(&ring->tx_stats.prepare_ctx_err, 1,
 165                                   &ring->syncp);
 166                 if (rc != -ENOMEM) {
 167                         adapter->reset_reason =
 168                                 ENA_REGS_RESET_DRIVER_INVALID_STATE;
 169                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
 170                 }
 171                 return rc;
 172         }
 173
 174         u64_stats_update_begin(&ring->syncp);
 175         ring->tx_stats.cnt++;
 176         ring->tx_stats.bytes += bytes;
 177         u64_stats_update_end(&ring->syncp);
 178
 179         tx_info->tx_descs = nb_hw_desc;
 180         tx_info->last_jiffies = jiffies;
 181         tx_info->print_once = 0;
 182
 183         ring->next_to_use = ENA_TX_RING_IDX_NEXT(next_to_use,
 184                                                  ring->ring_size);
 185         return 0;
 186 }
 187
 188 /* This is the XDP napi callback. XDP queues use a separate napi callback
 189  * than Rx/Tx queues.
 190  */
 191 static int ena_xdp_io_poll(struct napi_struct *napi, int budget)
 192 {
 193         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
 194         u32 xdp_work_done, xdp_budget;
 195         struct ena_ring *xdp_ring;
 196         int napi_comp_call = 0;
 197         int ret;
 198
 199         xdp_ring = ena_napi->xdp_ring;
 200         xdp_ring->first_interrupt = ena_napi->first_interrupt;
 201
 202         xdp_budget = budget;
 203
 204         if (!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags) ||
 205             test_bit(ENA_FLAG_TRIGGER_RESET, &xdp_ring->adapter->flags)) {
 206                 napi_complete_done(napi, 0);
 207                 return 0;
 208         }
 209
 210         xdp_work_done = ena_clean_xdp_irq(xdp_ring, xdp_budget);
 211
 212         /* If the device is about to reset or down, avoid unmask
 213          * the interrupt and return 0 so NAPI won't reschedule
 214          */
 215         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &xdp_ring->adapter->flags))) {
 216                 napi_complete_done(napi, 0);
 217                 ret = 0;
 218         } else if (xdp_budget > xdp_work_done) {
 219                 napi_comp_call = 1;
 220                 if (napi_complete_done(napi, xdp_work_done))
 221                         ena_unmask_interrupt(xdp_ring, NULL);
 222                 ena_update_ring_numa_node(xdp_ring, NULL);
 223                 ret = xdp_work_done;
 224         } else {
 225                 ret = xdp_budget;
 226         }
 227
 228         u64_stats_update_begin(&xdp_ring->syncp);
 229         xdp_ring->tx_stats.napi_comp += napi_comp_call;
 230         xdp_ring->tx_stats.tx_poll++;
 231         u64_stats_update_end(&xdp_ring->syncp);
 232
 233         return ret;
 234 }
 235
 236 static int ena_xdp_tx_map_frame(struct ena_ring *xdp_ring,
 237                                 struct ena_tx_buffer *tx_info,
 238                                 struct xdp_frame *xdpf,
 239                                 void **push_hdr,
 240                                 u32 *push_len)
 241 {
 242         struct ena_adapter *adapter = xdp_ring->adapter;
 243         struct ena_com_buf *ena_buf;
 244         dma_addr_t dma = 0;
 245         u32 size;
 246
 247         tx_info->xdpf = xdpf;
 248         size = tx_info->xdpf->len;
 249         ena_buf = tx_info->bufs;
 250
 251         /* llq push buffer */
 252         *push_len = min_t(u32, size, xdp_ring->tx_max_header_size);
 253         *push_hdr = tx_info->xdpf->data;
 254
 255         if (size - *push_len > 0) {
 256                 dma = dma_map_single(xdp_ring->dev,
 257                                      *push_hdr + *push_len,
 258                                      size - *push_len,
 259                                      DMA_TO_DEVICE);
 260                 if (unlikely(dma_mapping_error(xdp_ring->dev, dma)))
 261                         goto error_report_dma_error;
 262
 263                 tx_info->map_linear_data = 1;
 264                 tx_info->num_of_bufs = 1;
 265         }
 266
 267         ena_buf->paddr = dma;
 268         ena_buf->len = size;
 269
 270         return 0;
 271
 272 error_report_dma_error:
 273         ena_increase_stat(&xdp_ring->tx_stats.dma_mapping_err, 1,
 274                           &xdp_ring->syncp);
 275         netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map xdp buff\n");
 276
 277         xdp_return_frame_rx_napi(tx_info->xdpf);
 278         tx_info->xdpf = NULL;
 279         tx_info->num_of_bufs = 0;
 280
 281         return -EINVAL;
 282 }
 283
 284 static int ena_xdp_xmit_frame(struct ena_ring *xdp_ring,
 285                               struct net_device *dev,
 286                               struct xdp_frame *xdpf,
 287                               int flags)
 288 {
 289         struct ena_com_tx_ctx ena_tx_ctx = {};
 290         struct ena_tx_buffer *tx_info;
 291         u16 next_to_use, req_id;
 292         void *push_hdr;
 293         u32 push_len;
 294         int rc;
 295
 296         next_to_use = xdp_ring->next_to_use;
 297         req_id = xdp_ring->free_ids[next_to_use];
 298         tx_info = &xdp_ring->tx_buffer_info[req_id];
 299         tx_info->num_of_bufs = 0;
 300
 301         rc = ena_xdp_tx_map_frame(xdp_ring, tx_info, xdpf, &push_hdr, &push_len);
 302         if (unlikely(rc))
 303                 return rc;
 304
 305         ena_tx_ctx.ena_bufs = tx_info->bufs;
 306         ena_tx_ctx.push_header = push_hdr;
 307         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
 308         ena_tx_ctx.req_id = req_id;
 309         ena_tx_ctx.header_len = push_len;
 310
 311         rc = ena_xmit_common(dev,
 312                              xdp_ring,
 313                              tx_info,
 314                              &ena_tx_ctx,
 315                              next_to_use,
 316                              xdpf->len);
 317         if (rc)
 318                 goto error_unmap_dma;
 319         /* trigger the dma engine. ena_com_write_sq_doorbell()
 320          * has a mb
 321          */
 322         if (flags & XDP_XMIT_FLUSH) {
 323                 ena_com_write_sq_doorbell(xdp_ring->ena_com_io_sq);
 324                 ena_increase_stat(&xdp_ring->tx_stats.doorbells, 1,
 325                                   &xdp_ring->syncp);
 326         }
 327
 328         return rc;
 329
 330 error_unmap_dma:
 331         ena_unmap_tx_buff(xdp_ring, tx_info);
 332         tx_info->xdpf = NULL;
 333         return rc;
 334 }
 335
 336 static int ena_xdp_xmit(struct net_device *dev, int n,
 337                         struct xdp_frame **frames, u32 flags)
 338 {
 339         struct ena_adapter *adapter = netdev_priv(dev);
 340         struct ena_ring *xdp_ring;
 341         int qid, i, nxmit = 0;
 342
 343         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 344                 return -EINVAL;
 345
 346         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
 347                 return -ENETDOWN;
 348
 349         /* We assume that all rings have the same XDP program */
 350         if (!READ_ONCE(adapter->rx_ring->xdp_bpf_prog))
 351                 return -ENXIO;
 352
 353         qid = smp_processor_id() % adapter->xdp_num_queues;
 354         qid += adapter->xdp_first_ring;
 355         xdp_ring = &adapter->tx_ring[qid];
 356
 357         /* Other CPU ids might try to send thorugh this queue */
 358         spin_lock(&xdp_ring->xdp_tx_lock);
 359
 360         for (i = 0; i < n; i++) {
 361                 if (ena_xdp_xmit_frame(xdp_ring, dev, frames[i], 0))
 362                         break;
 363                 nxmit++;
 364         }
 365
 366         /* Ring doorbell to make device aware of the packets */
 367         if (flags & XDP_XMIT_FLUSH) {
 368                 ena_com_write_sq_doorbell(xdp_ring->ena_com_io_sq);
 369                 ena_increase_stat(&xdp_ring->tx_stats.doorbells, 1,
 370                                   &xdp_ring->syncp);
 371         }
 372
 373         spin_unlock(&xdp_ring->xdp_tx_lock);
 374
 375         /* Return number of packets sent */
 376         return nxmit;
 377 }
 378
 379 static int ena_xdp_execute(struct ena_ring *rx_ring, struct xdp_buff *xdp)
 380 {
 381         struct bpf_prog *xdp_prog;
 382         struct ena_ring *xdp_ring;
 383         u32 verdict = XDP_PASS;
 384         struct xdp_frame *xdpf;
 385         u64 *xdp_stat;
 386         int qid;
 387
 388         rcu_read_lock();
 389         xdp_prog = READ_ONCE(rx_ring->xdp_bpf_prog);
 390
 391         if (!xdp_prog)
 392                 goto out;
 393
 394         verdict = bpf_prog_run_xdp(xdp_prog, xdp);
 395
 396         switch (verdict) {
 397         case XDP_TX:
 398                 xdpf = xdp_convert_buff_to_frame(xdp);
 399                 if (unlikely(!xdpf)) {
 400                         trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 401                         xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 402                         verdict = XDP_ABORTED;
 403                         break;
 404                 }
 405
 406                 /* Find xmit queue */
 407                 qid = rx_ring->qid + rx_ring->adapter->num_io_queues;
 408                 xdp_ring = &rx_ring->adapter->tx_ring[qid];
 409
 410                 /* The XDP queues are shared between XDP_TX and XDP_REDIRECT */
 411                 spin_lock(&xdp_ring->xdp_tx_lock);
 412
 413                 if (ena_xdp_xmit_frame(xdp_ring, rx_ring->netdev, xdpf,
 414                                        XDP_XMIT_FLUSH))
 415                         xdp_return_frame(xdpf);
 416
 417                 spin_unlock(&xdp_ring->xdp_tx_lock);
 418                 xdp_stat = &rx_ring->rx_stats.xdp_tx;
 419                 break;
 420         case XDP_REDIRECT:
 421                 if (likely(!xdp_do_redirect(rx_ring->netdev, xdp, xdp_prog))) {
 422                         xdp_stat = &rx_ring->rx_stats.xdp_redirect;
 423                         break;
 424                 }
 425                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 426                 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 427                 verdict = XDP_ABORTED;
 428                 break;
 429         case XDP_ABORTED:
 430                 trace_xdp_exception(rx_ring->netdev, xdp_prog, verdict);
 431                 xdp_stat = &rx_ring->rx_stats.xdp_aborted;
 432                 break;
 433         case XDP_DROP:
 434                 xdp_stat = &rx_ring->rx_stats.xdp_drop;
 435                 break;
 436         case XDP_PASS:
 437                 xdp_stat = &rx_ring->rx_stats.xdp_pass;
 438                 break;
 439         default:
 440                 bpf_warn_invalid_xdp_action(verdict);
 441                 xdp_stat = &rx_ring->rx_stats.xdp_invalid;
 442         }
 443
 444         ena_increase_stat(xdp_stat, 1, &rx_ring->syncp);
 445 out:
 446         rcu_read_unlock();
 447
 448         return verdict;
 449 }
 450
 451 static void ena_init_all_xdp_queues(struct ena_adapter *adapter)
 452 {
 453         adapter->xdp_first_ring = adapter->num_io_queues;
 454         adapter->xdp_num_queues = adapter->num_io_queues;
 455
 456         ena_init_io_rings(adapter,
 457                           adapter->xdp_first_ring,
 458                           adapter->xdp_num_queues);
 459 }
 460
 461 static int ena_setup_and_create_all_xdp_queues(struct ena_adapter *adapter)
 462 {
 463         int rc = 0;
 464
 465         rc = ena_setup_tx_resources_in_range(adapter, adapter->xdp_first_ring,
 466                                              adapter->xdp_num_queues);
 467         if (rc)
 468                 goto setup_err;
 469
 470         rc = ena_create_io_tx_queues_in_range(adapter,
 471                                               adapter->xdp_first_ring,
 472                                               adapter->xdp_num_queues);
 473         if (rc)
 474                 goto create_err;
 475
 476         return 0;
 477
 478 create_err:
 479         ena_free_all_io_tx_resources(adapter);
 480 setup_err:
 481         return rc;
 482 }
 483
 484 /* Provides a way for both kernel and bpf-prog to know
 485  * more about the RX-queue a given XDP frame arrived on.
 486  */
 487 static int ena_xdp_register_rxq_info(struct ena_ring *rx_ring)
 488 {
 489         int rc;
 490
 491         rc = xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev, rx_ring->qid, 0);
 492
 493         if (rc) {
 494                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 495                           "Failed to register xdp rx queue info. RX queue num %d rc: %d\n",
 496                           rx_ring->qid, rc);
 497                 goto err;
 498         }
 499
 500         rc = xdp_rxq_info_reg_mem_model(&rx_ring->xdp_rxq, MEM_TYPE_PAGE_SHARED,
 501                                         NULL);
 502
 503         if (rc) {
 504                 netif_err(rx_ring->adapter, ifup, rx_ring->netdev,
 505                           "Failed to register xdp rx queue info memory model. RX queue num %d rc: %d\n",
 506                           rx_ring->qid, rc);
 507                 xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 508         }
 509
 510 err:
 511         return rc;
 512 }
 513
 514 static void ena_xdp_unregister_rxq_info(struct ena_ring *rx_ring)
 515 {
 516         xdp_rxq_info_unreg_mem_model(&rx_ring->xdp_rxq);
 517         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 518 }
 519
 520 static void ena_xdp_exchange_program_rx_in_range(struct ena_adapter *adapter,
 521                                                  struct bpf_prog *prog,
 522                                                  int first, int count)
 523 {
 524         struct ena_ring *rx_ring;
 525         int i = 0;
 526
 527         for (i = first; i < count; i++) {
 528                 rx_ring = &adapter->rx_ring[i];
 529                 xchg(&rx_ring->xdp_bpf_prog, prog);
 530                 if (prog) {
 531                         ena_xdp_register_rxq_info(rx_ring);
 532                         rx_ring->rx_headroom = XDP_PACKET_HEADROOM;
 533                 } else {
 534                         ena_xdp_unregister_rxq_info(rx_ring);
 535                         rx_ring->rx_headroom = 0;
 536                 }
 537         }
 538 }
 539
 540 static void ena_xdp_exchange_program(struct ena_adapter *adapter,
 541                                      struct bpf_prog *prog)
 542 {
 543         struct bpf_prog *old_bpf_prog = xchg(&adapter->xdp_bpf_prog, prog);
 544
 545         ena_xdp_exchange_program_rx_in_range(adapter,
 546                                              prog,
 547                                              0,
 548                                              adapter->num_io_queues);
 549
 550         if (old_bpf_prog)
 551                 bpf_prog_put(old_bpf_prog);
 552 }
 553
 554 static int ena_destroy_and_free_all_xdp_queues(struct ena_adapter *adapter)
 555 {
 556         bool was_up;
 557         int rc;
 558
 559         was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 560
 561         if (was_up)
 562                 ena_down(adapter);
 563
 564         adapter->xdp_first_ring = 0;
 565         adapter->xdp_num_queues = 0;
 566         ena_xdp_exchange_program(adapter, NULL);
 567         if (was_up) {
 568                 rc = ena_up(adapter);
 569                 if (rc)
 570                         return rc;
 571         }
 572         return 0;
 573 }
 574
 575 static int ena_xdp_set(struct net_device *netdev, struct netdev_bpf *bpf)
 576 {
 577         struct ena_adapter *adapter = netdev_priv(netdev);
 578         struct bpf_prog *prog = bpf->prog;
 579         struct bpf_prog *old_bpf_prog;
 580         int rc, prev_mtu;
 581         bool is_up;
 582
 583         is_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
 584         rc = ena_xdp_allowed(adapter);
 585         if (rc == ENA_XDP_ALLOWED) {
 586                 old_bpf_prog = adapter->xdp_bpf_prog;
 587                 if (prog) {
 588                         if (!is_up) {
 589                                 ena_init_all_xdp_queues(adapter);
 590                         } else if (!old_bpf_prog) {
 591                                 ena_down(adapter);
 592                                 ena_init_all_xdp_queues(adapter);
 593                         }
 594                         ena_xdp_exchange_program(adapter, prog);
 595
 596                         if (is_up && !old_bpf_prog) {
 597                                 rc = ena_up(adapter);
 598                                 if (rc)
 599                                         return rc;
 600                         }
 601                 } else if (old_bpf_prog) {
 602                         rc = ena_destroy_and_free_all_xdp_queues(adapter);
 603                         if (rc)
 604                                 return rc;
 605                 }
 606
 607                 prev_mtu = netdev->max_mtu;
 608                 netdev->max_mtu = prog ? ENA_XDP_MAX_MTU : adapter->max_mtu;
 609
 610                 if (!old_bpf_prog)
 611                         netif_info(adapter, drv, adapter->netdev,
 612                                    "XDP program is set, changing the max_mtu from %d to %d",
 613                                    prev_mtu, netdev->max_mtu);
 614
 615         } else if (rc == ENA_XDP_CURRENT_MTU_TOO_LARGE) {
 616                 netif_err(adapter, drv, adapter->netdev,
 617                           "Failed to set xdp program, the current MTU (%d) is larger than the maximum allowed MTU (%lu) while xdp is on",
 618                           netdev->mtu, ENA_XDP_MAX_MTU);
 619                 NL_SET_ERR_MSG_MOD(bpf->extack,
 620                                    "Failed to set xdp program, the current MTU is larger than the maximum allowed MTU. Check the dmesg for more info");
 621                 return -EINVAL;
 622         } else if (rc == ENA_XDP_NO_ENOUGH_QUEUES) {
 623                 netif_err(adapter, drv, adapter->netdev,
 624                           "Failed to set xdp program, the Rx/Tx channel count should be at most half of the maximum allowed channel count. The current queue count (%d), the maximal queue count (%d)\n",
 625                           adapter->num_io_queues, adapter->max_num_io_queues);
 626                 NL_SET_ERR_MSG_MOD(bpf->extack,
 627                                    "Failed to set xdp program, there is no enough space for allocating XDP queues, Check the dmesg for more info");
 628                 return -EINVAL;
 629         }
 630
 631         return 0;
 632 }
 633
 634 /* This is the main xdp callback, it's used by the kernel to set/unset the xdp
 635  * program as well as to query the current xdp program id.
 636  */
 637 static int ena_xdp(struct net_device *netdev, struct netdev_bpf *bpf)
 638 {
 639         switch (bpf->command) {
 640         case XDP_SETUP_PROG:
 641                 return ena_xdp_set(netdev, bpf);
 642         default:
 643                 return -EINVAL;
 644         }
 645         return 0;
 646 }
 647
 648 static int ena_init_rx_cpu_rmap(struct ena_adapter *adapter)
 649 {
 650 #ifdef CONFIG_RFS_ACCEL
 651         u32 i;
 652         int rc;
 653
 654         adapter->netdev->rx_cpu_rmap = alloc_irq_cpu_rmap(adapter->num_io_queues);
 655         if (!adapter->netdev->rx_cpu_rmap)
 656                 return -ENOMEM;
 657         for (i = 0; i < adapter->num_io_queues; i++) {
 658                 int irq_idx = ENA_IO_IRQ_IDX(i);
 659
 660                 rc = irq_cpu_rmap_add(adapter->netdev->rx_cpu_rmap,
 661                                       pci_irq_vector(adapter->pdev, irq_idx));
 662                 if (rc) {
 663                         free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
 664                         adapter->netdev->rx_cpu_rmap = NULL;
 665                         return rc;
 666                 }
 667         }
 668 #endif /* CONFIG_RFS_ACCEL */
 669         return 0;
 670 }
 671
 672 static void ena_init_io_rings_common(struct ena_adapter *adapter,
 673                                      struct ena_ring *ring, u16 qid)
 674 {
 675         ring->qid = qid;
 676         ring->pdev = adapter->pdev;
 677         ring->dev = &adapter->pdev->dev;
 678         ring->netdev = adapter->netdev;
 679         ring->napi = &adapter->ena_napi[qid].napi;
 680         ring->adapter = adapter;
 681         ring->ena_dev = adapter->ena_dev;
 682         ring->per_napi_packets = 0;
 683         ring->cpu = 0;
 684         ring->first_interrupt = false;
 685         ring->no_interrupt_event_cnt = 0;
 686         u64_stats_init(&ring->syncp);
 687 }
 688
 689 static void ena_init_io_rings(struct ena_adapter *adapter,
 690                               int first_index, int count)
 691 {
 692         struct ena_com_dev *ena_dev;
 693         struct ena_ring *txr, *rxr;
 694         int i;
 695
 696         ena_dev = adapter->ena_dev;
 697
 698         for (i = first_index; i < first_index + count; i++) {
 699                 txr = &adapter->tx_ring[i];
 700                 rxr = &adapter->rx_ring[i];
 701
 702                 /* TX common ring state */
 703                 ena_init_io_rings_common(adapter, txr, i);
 704
 705                 /* TX specific ring state */
 706                 txr->ring_size = adapter->requested_tx_ring_size;
 707                 txr->tx_max_header_size = ena_dev->tx_max_header_size;
 708                 txr->tx_mem_queue_type = ena_dev->tx_mem_queue_type;
 709                 txr->sgl_size = adapter->max_tx_sgl_size;
 710                 txr->smoothed_interval =
 711                         ena_com_get_nonadaptive_moderation_interval_tx(ena_dev);
 712                 txr->disable_meta_caching = adapter->disable_meta_caching;
 713                 spin_lock_init(&txr->xdp_tx_lock);
 714
 715                 /* Don't init RX queues for xdp queues */
 716                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
 717                         /* RX common ring state */
 718                         ena_init_io_rings_common(adapter, rxr, i);
 719
 720                         /* RX specific ring state */
 721                         rxr->ring_size = adapter->requested_rx_ring_size;
 722                         rxr->rx_copybreak = adapter->rx_copybreak;
 723                         rxr->sgl_size = adapter->max_rx_sgl_size;
 724                         rxr->smoothed_interval =
 725                                 ena_com_get_nonadaptive_moderation_interval_rx(ena_dev);
 726                         rxr->empty_rx_queue = 0;
 727                         adapter->ena_napi[i].dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
 728                 }
 729         }
 730 }
 731
 732 /* ena_setup_tx_resources - allocate I/O Tx resources (Descriptors)
 733  * @adapter: network interface device structure
 734  * @qid: queue index
 735  *
 736  * Return 0 on success, negative on failure
 737  */
 738 static int ena_setup_tx_resources(struct ena_adapter *adapter, int qid)
 739 {
 740         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 741         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 742         int size, i, node;
 743
 744         if (tx_ring->tx_buffer_info) {
 745                 netif_err(adapter, ifup,
 746                           adapter->netdev, "tx_buffer_info info is not NULL");
 747                 return -EEXIST;
 748         }
 749
 750         size = sizeof(struct ena_tx_buffer) * tx_ring->ring_size;
 751         node = cpu_to_node(ena_irq->cpu);
 752
 753         tx_ring->tx_buffer_info = vzalloc_node(size, node);
 754         if (!tx_ring->tx_buffer_info) {
 755                 tx_ring->tx_buffer_info = vzalloc(size);
 756                 if (!tx_ring->tx_buffer_info)
 757                         goto err_tx_buffer_info;
 758         }
 759
 760         size = sizeof(u16) * tx_ring->ring_size;
 761         tx_ring->free_ids = vzalloc_node(size, node);
 762         if (!tx_ring->free_ids) {
 763                 tx_ring->free_ids = vzalloc(size);
 764                 if (!tx_ring->free_ids)
 765                         goto err_tx_free_ids;
 766         }
 767
 768         size = tx_ring->tx_max_header_size;
 769         tx_ring->push_buf_intermediate_buf = vzalloc_node(size, node);
 770         if (!tx_ring->push_buf_intermediate_buf) {
 771                 tx_ring->push_buf_intermediate_buf = vzalloc(size);
 772                 if (!tx_ring->push_buf_intermediate_buf)
 773                         goto err_push_buf_intermediate_buf;
 774         }
 775
 776         /* Req id ring for TX out of order completions */
 777         for (i = 0; i < tx_ring->ring_size; i++)
 778                 tx_ring->free_ids[i] = i;
 779
 780         /* Reset tx statistics */
 781         memset(&tx_ring->tx_stats, 0x0, sizeof(tx_ring->tx_stats));
 782
 783         tx_ring->next_to_use = 0;
 784         tx_ring->next_to_clean = 0;
 785         tx_ring->cpu = ena_irq->cpu;
 786         return 0;
 787
 788 err_push_buf_intermediate_buf:
 789         vfree(tx_ring->free_ids);
 790         tx_ring->free_ids = NULL;
 791 err_tx_free_ids:
 792         vfree(tx_ring->tx_buffer_info);
 793         tx_ring->tx_buffer_info = NULL;
 794 err_tx_buffer_info:
 795         return -ENOMEM;
 796 }
 797
 798 /* ena_free_tx_resources - Free I/O Tx Resources per Queue
 799  * @adapter: network interface device structure
 800  * @qid: queue index
 801  *
 802  * Free all transmit software resources
 803  */
 804 static void ena_free_tx_resources(struct ena_adapter *adapter, int qid)
 805 {
 806         struct ena_ring *tx_ring = &adapter->tx_ring[qid];
 807
 808         vfree(tx_ring->tx_buffer_info);
 809         tx_ring->tx_buffer_info = NULL;
 810
 811         vfree(tx_ring->free_ids);
 812         tx_ring->free_ids = NULL;
 813
 814         vfree(tx_ring->push_buf_intermediate_buf);
 815         tx_ring->push_buf_intermediate_buf = NULL;
 816 }
 817
 818 static int ena_setup_tx_resources_in_range(struct ena_adapter *adapter,
 819                                            int first_index,
 820                                            int count)
 821 {
 822         int i, rc = 0;
 823
 824         for (i = first_index; i < first_index + count; i++) {
 825                 rc = ena_setup_tx_resources(adapter, i);
 826                 if (rc)
 827                         goto err_setup_tx;
 828         }
 829
 830         return 0;
 831
 832 err_setup_tx:
 833
 834         netif_err(adapter, ifup, adapter->netdev,
 835                   "Tx queue %d: allocation failed\n", i);
 836
 837         /* rewind the index freeing the rings as we go */
 838         while (first_index < i--)
 839                 ena_free_tx_resources(adapter, i);
 840         return rc;
 841 }
 842
 843 static void ena_free_all_io_tx_resources_in_range(struct ena_adapter *adapter,
 844                                                   int first_index, int count)
 845 {
 846         int i;
 847
 848         for (i = first_index; i < first_index + count; i++)
 849                 ena_free_tx_resources(adapter, i);
 850 }
 851
 852 /* ena_free_all_io_tx_resources - Free I/O Tx Resources for All Queues
 853  * @adapter: board private structure
 854  *
 855  * Free all transmit software resources
 856  */
 857 static void ena_free_all_io_tx_resources(struct ena_adapter *adapter)
 858 {
 859         ena_free_all_io_tx_resources_in_range(adapter,
 860                                               0,
 861                                               adapter->xdp_num_queues +
 862                                               adapter->num_io_queues);
 863 }
 864
 865 /* ena_setup_rx_resources - allocate I/O Rx resources (Descriptors)
 866  * @adapter: network interface device structure
 867  * @qid: queue index
 868  *
 869  * Returns 0 on success, negative on failure
 870  */
 871 static int ena_setup_rx_resources(struct ena_adapter *adapter,
 872                                   u32 qid)
 873 {
 874         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 875         struct ena_irq *ena_irq = &adapter->irq_tbl[ENA_IO_IRQ_IDX(qid)];
 876         int size, node, i;
 877
 878         if (rx_ring->rx_buffer_info) {
 879                 netif_err(adapter, ifup, adapter->netdev,
 880                           "rx_buffer_info is not NULL");
 881                 return -EEXIST;
 882         }
 883
 884         /* alloc extra element so in rx path
 885          * we can always prefetch rx_info + 1
 886          */
 887         size = sizeof(struct ena_rx_buffer) * (rx_ring->ring_size + 1);
 888         node = cpu_to_node(ena_irq->cpu);
 889
 890         rx_ring->rx_buffer_info = vzalloc_node(size, node);
 891         if (!rx_ring->rx_buffer_info) {
 892                 rx_ring->rx_buffer_info = vzalloc(size);
 893                 if (!rx_ring->rx_buffer_info)
 894                         return -ENOMEM;
 895         }
 896
 897         size = sizeof(u16) * rx_ring->ring_size;
 898         rx_ring->free_ids = vzalloc_node(size, node);
 899         if (!rx_ring->free_ids) {
 900                 rx_ring->free_ids = vzalloc(size);
 901                 if (!rx_ring->free_ids) {
 902                         vfree(rx_ring->rx_buffer_info);
 903                         rx_ring->rx_buffer_info = NULL;
 904                         return -ENOMEM;
 905                 }
 906         }
 907
 908         /* Req id ring for receiving RX pkts out of order */
 909         for (i = 0; i < rx_ring->ring_size; i++)
 910                 rx_ring->free_ids[i] = i;
 911
 912         /* Reset rx statistics */
 913         memset(&rx_ring->rx_stats, 0x0, sizeof(rx_ring->rx_stats));
 914
 915         rx_ring->next_to_clean = 0;
 916         rx_ring->next_to_use = 0;
 917         rx_ring->cpu = ena_irq->cpu;
 918
 919         return 0;
 920 }
 921
 922 /* ena_free_rx_resources - Free I/O Rx Resources
 923  * @adapter: network interface device structure
 924  * @qid: queue index
 925  *
 926  * Free all receive software resources
 927  */
 928 static void ena_free_rx_resources(struct ena_adapter *adapter,
 929                                   u32 qid)
 930 {
 931         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
 932
 933         vfree(rx_ring->rx_buffer_info);
 934         rx_ring->rx_buffer_info = NULL;
 935
 936         vfree(rx_ring->free_ids);
 937         rx_ring->free_ids = NULL;
 938 }
 939
 940 /* ena_setup_all_rx_resources - allocate I/O Rx queues resources for all queues
 941  * @adapter: board private structure
 942  *
 943  * Return 0 on success, negative on failure
 944  */
 945 static int ena_setup_all_rx_resources(struct ena_adapter *adapter)
 946 {
 947         int i, rc = 0;
 948
 949         for (i = 0; i < adapter->num_io_queues; i++) {
 950                 rc = ena_setup_rx_resources(adapter, i);
 951                 if (rc)
 952                         goto err_setup_rx;
 953         }
 954
 955         return 0;
 956
 957 err_setup_rx:
 958
 959         netif_err(adapter, ifup, adapter->netdev,
 960                   "Rx queue %d: allocation failed\n", i);
 961
 962         /* rewind the index freeing the rings as we go */
 963         while (i--)
 964                 ena_free_rx_resources(adapter, i);
 965         return rc;
 966 }
 967
 968 /* ena_free_all_io_rx_resources - Free I/O Rx Resources for All Queues
 969  * @adapter: board private structure
 970  *
 971  * Free all receive software resources
 972  */
 973 static void ena_free_all_io_rx_resources(struct ena_adapter *adapter)
 974 {
 975         int i;
 976
 977         for (i = 0; i < adapter->num_io_queues; i++)
 978                 ena_free_rx_resources(adapter, i);
 979 }
 980
 981 static int ena_alloc_rx_page(struct ena_ring *rx_ring,
 982                                     struct ena_rx_buffer *rx_info, gfp_t gfp)
 983 {
 984         int headroom = rx_ring->rx_headroom;
 985         struct ena_com_buf *ena_buf;
 986         struct page *page;
 987         dma_addr_t dma;
 988
 989         /* restore page offset value in case it has been changed by device */
 990         rx_info->page_offset = headroom;
 991
 992         /* if previous allocated page is not used */
 993         if (unlikely(rx_info->page))
 994                 return 0;
 995
 996         page = alloc_page(gfp);
 997         if (unlikely(!page)) {
 998                 ena_increase_stat(&rx_ring->rx_stats.page_alloc_fail, 1,
 999                                   &rx_ring->syncp);
1000                 return -ENOMEM;
1001         }
1002
1003         /* To enable NIC-side port-mirroring, AKA SPAN port,
1004          * we make the buffer readable from the nic as well
1005          */
1006         dma = dma_map_page(rx_ring->dev, page, 0, ENA_PAGE_SIZE,
1007                            DMA_BIDIRECTIONAL);
1008         if (unlikely(dma_mapping_error(rx_ring->dev, dma))) {
1009                 ena_increase_stat(&rx_ring->rx_stats.dma_mapping_err, 1,
1010                                   &rx_ring->syncp);
1011
1012                 __free_page(page);
1013                 return -EIO;
1014         }
1015         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1016                   "Allocate page %p, rx_info %p\n", page, rx_info);
1017
1018         rx_info->page = page;
1019         ena_buf = &rx_info->ena_buf;
1020         ena_buf->paddr = dma + headroom;
1021         ena_buf->len = ENA_PAGE_SIZE - headroom;
1022
1023         return 0;
1024 }
1025
1026 static void ena_unmap_rx_buff(struct ena_ring *rx_ring,
1027                               struct ena_rx_buffer *rx_info)
1028 {
1029         struct ena_com_buf *ena_buf = &rx_info->ena_buf;
1030
1031         dma_unmap_page(rx_ring->dev, ena_buf->paddr - rx_ring->rx_headroom,
1032                        ENA_PAGE_SIZE,
1033                        DMA_BIDIRECTIONAL);
1034 }
1035
1036 static void ena_free_rx_page(struct ena_ring *rx_ring,
1037                              struct ena_rx_buffer *rx_info)
1038 {
1039         struct page *page = rx_info->page;
1040
1041         if (unlikely(!page)) {
1042                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1043                            "Trying to free unallocated buffer\n");
1044                 return;
1045         }
1046
1047         ena_unmap_rx_buff(rx_ring, rx_info);
1048
1049         __free_page(page);
1050         rx_info->page = NULL;
1051 }
1052
1053 static int ena_refill_rx_bufs(struct ena_ring *rx_ring, u32 num)
1054 {
1055         u16 next_to_use, req_id;
1056         u32 i;
1057         int rc;
1058
1059         next_to_use = rx_ring->next_to_use;
1060
1061         for (i = 0; i < num; i++) {
1062                 struct ena_rx_buffer *rx_info;
1063
1064                 req_id = rx_ring->free_ids[next_to_use];
1065
1066                 rx_info = &rx_ring->rx_buffer_info[req_id];
1067
1068                 rc = ena_alloc_rx_page(rx_ring, rx_info,
1069                                        GFP_ATOMIC | __GFP_COMP);
1070                 if (unlikely(rc < 0)) {
1071                         netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1072                                    "Failed to allocate buffer for rx queue %d\n",
1073                                    rx_ring->qid);
1074                         break;
1075                 }
1076                 rc = ena_com_add_single_rx_desc(rx_ring->ena_com_io_sq,
1077                                                 &rx_info->ena_buf,
1078                                                 req_id);
1079                 if (unlikely(rc)) {
1080                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1081                                    "Failed to add buffer for rx queue %d\n",
1082                                    rx_ring->qid);
1083                         break;
1084                 }
1085                 next_to_use = ENA_RX_RING_IDX_NEXT(next_to_use,
1086                                                    rx_ring->ring_size);
1087         }
1088
1089         if (unlikely(i < num)) {
1090                 ena_increase_stat(&rx_ring->rx_stats.refil_partial, 1,
1091                                   &rx_ring->syncp);
1092                 netif_warn(rx_ring->adapter, rx_err, rx_ring->netdev,
1093                            "Refilled rx qid %d with only %d buffers (from %d)\n",
1094                            rx_ring->qid, i, num);
1095         }
1096
1097         /* ena_com_write_sq_doorbell issues a wmb() */
1098         if (likely(i))
1099                 ena_com_write_sq_doorbell(rx_ring->ena_com_io_sq);
1100
1101         rx_ring->next_to_use = next_to_use;
1102
1103         return i;
1104 }
1105
1106 static void ena_free_rx_bufs(struct ena_adapter *adapter,
1107                              u32 qid)
1108 {
1109         struct ena_ring *rx_ring = &adapter->rx_ring[qid];
1110         u32 i;
1111
1112         for (i = 0; i < rx_ring->ring_size; i++) {
1113                 struct ena_rx_buffer *rx_info = &rx_ring->rx_buffer_info[i];
1114
1115                 if (rx_info->page)
1116                         ena_free_rx_page(rx_ring, rx_info);
1117         }
1118 }
1119
1120 /* ena_refill_all_rx_bufs - allocate all queues Rx buffers
1121  * @adapter: board private structure
1122  */
1123 static void ena_refill_all_rx_bufs(struct ena_adapter *adapter)
1124 {
1125         struct ena_ring *rx_ring;
1126         int i, rc, bufs_num;
1127
1128         for (i = 0; i < adapter->num_io_queues; i++) {
1129                 rx_ring = &adapter->rx_ring[i];
1130                 bufs_num = rx_ring->ring_size - 1;
1131                 rc = ena_refill_rx_bufs(rx_ring, bufs_num);
1132
1133                 if (unlikely(rc != bufs_num))
1134                         netif_warn(rx_ring->adapter, rx_status, rx_ring->netdev,
1135                                    "Refilling Queue %d failed. allocated %d buffers from: %d\n",
1136                                    i, rc, bufs_num);
1137         }
1138 }
1139
1140 static void ena_free_all_rx_bufs(struct ena_adapter *adapter)
1141 {
1142         int i;
1143
1144         for (i = 0; i < adapter->num_io_queues; i++)
1145                 ena_free_rx_bufs(adapter, i);
1146 }
1147
1148 static void ena_unmap_tx_buff(struct ena_ring *tx_ring,
1149                               struct ena_tx_buffer *tx_info)
1150 {
1151         struct ena_com_buf *ena_buf;
1152         u32 cnt;
1153         int i;
1154
1155         ena_buf = tx_info->bufs;
1156         cnt = tx_info->num_of_bufs;
1157
1158         if (unlikely(!cnt))
1159                 return;
1160
1161         if (tx_info->map_linear_data) {
1162                 dma_unmap_single(tx_ring->dev,
1163                                  dma_unmap_addr(ena_buf, paddr),
1164                                  dma_unmap_len(ena_buf, len),
1165                                  DMA_TO_DEVICE);
1166                 ena_buf++;
1167                 cnt--;
1168         }
1169
1170         /* unmap remaining mapped pages */
1171         for (i = 0; i < cnt; i++) {
1172                 dma_unmap_page(tx_ring->dev, dma_unmap_addr(ena_buf, paddr),
1173                                dma_unmap_len(ena_buf, len), DMA_TO_DEVICE);
1174                 ena_buf++;
1175         }
1176 }
1177
1178 /* ena_free_tx_bufs - Free Tx Buffers per Queue
1179  * @tx_ring: TX ring for which buffers be freed
1180  */
1181 static void ena_free_tx_bufs(struct ena_ring *tx_ring)
1182 {
1183         bool print_once = true;
1184         u32 i;
1185
1186         for (i = 0; i < tx_ring->ring_size; i++) {
1187                 struct ena_tx_buffer *tx_info = &tx_ring->tx_buffer_info[i];
1188
1189                 if (!tx_info->skb)
1190                         continue;
1191
1192                 if (print_once) {
1193                         netif_notice(tx_ring->adapter, ifdown, tx_ring->netdev,
1194                                      "Free uncompleted tx skb qid %d idx 0x%x\n",
1195                                      tx_ring->qid, i);
1196                         print_once = false;
1197                 } else {
1198                         netif_dbg(tx_ring->adapter, ifdown, tx_ring->netdev,
1199                                   "Free uncompleted tx skb qid %d idx 0x%x\n",
1200                                   tx_ring->qid, i);
1201                 }
1202
1203                 ena_unmap_tx_buff(tx_ring, tx_info);
1204
1205                 dev_kfree_skb_any(tx_info->skb);
1206         }
1207         netdev_tx_reset_queue(netdev_get_tx_queue(tx_ring->netdev,
1208                                                   tx_ring->qid));
1209 }
1210
1211 static void ena_free_all_tx_bufs(struct ena_adapter *adapter)
1212 {
1213         struct ena_ring *tx_ring;
1214         int i;
1215
1216         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1217                 tx_ring = &adapter->tx_ring[i];
1218                 ena_free_tx_bufs(tx_ring);
1219         }
1220 }
1221
1222 static void ena_destroy_all_tx_queues(struct ena_adapter *adapter)
1223 {
1224         u16 ena_qid;
1225         int i;
1226
1227         for (i = 0; i < adapter->num_io_queues + adapter->xdp_num_queues; i++) {
1228                 ena_qid = ENA_IO_TXQ_IDX(i);
1229                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1230         }
1231 }
1232
1233 static void ena_destroy_all_rx_queues(struct ena_adapter *adapter)
1234 {
1235         u16 ena_qid;
1236         int i;
1237
1238         for (i = 0; i < adapter->num_io_queues; i++) {
1239                 ena_qid = ENA_IO_RXQ_IDX(i);
1240                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
1241                 ena_com_destroy_io_queue(adapter->ena_dev, ena_qid);
1242         }
1243 }
1244
1245 static void ena_destroy_all_io_queues(struct ena_adapter *adapter)
1246 {
1247         ena_destroy_all_tx_queues(adapter);
1248         ena_destroy_all_rx_queues(adapter);
1249 }
1250
1251 static int handle_invalid_req_id(struct ena_ring *ring, u16 req_id,
1252                                  struct ena_tx_buffer *tx_info, bool is_xdp)
1253 {
1254         if (tx_info)
1255                 netif_err(ring->adapter,
1256                           tx_done,
1257                           ring->netdev,
1258                           "tx_info doesn't have valid %s",
1259                            is_xdp ? "xdp frame" : "skb");
1260         else
1261                 netif_err(ring->adapter,
1262                           tx_done,
1263                           ring->netdev,
1264                           "Invalid req_id: %hu\n",
1265                           req_id);
1266
1267         ena_increase_stat(&ring->tx_stats.bad_req_id, 1, &ring->syncp);
1268
1269         /* Trigger device reset */
1270         ring->adapter->reset_reason = ENA_REGS_RESET_INV_TX_REQ_ID;
1271         set_bit(ENA_FLAG_TRIGGER_RESET, &ring->adapter->flags);
1272         return -EFAULT;
1273 }
1274
1275 static int validate_tx_req_id(struct ena_ring *tx_ring, u16 req_id)
1276 {
1277         struct ena_tx_buffer *tx_info = NULL;
1278
1279         if (likely(req_id < tx_ring->ring_size)) {
1280                 tx_info = &tx_ring->tx_buffer_info[req_id];
1281                 if (likely(tx_info->skb))
1282                         return 0;
1283         }
1284
1285         return handle_invalid_req_id(tx_ring, req_id, tx_info, false);
1286 }
1287
1288 static int validate_xdp_req_id(struct ena_ring *xdp_ring, u16 req_id)
1289 {
1290         struct ena_tx_buffer *tx_info = NULL;
1291
1292         if (likely(req_id < xdp_ring->ring_size)) {
1293                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1294                 if (likely(tx_info->xdpf))
1295                         return 0;
1296         }
1297
1298         return handle_invalid_req_id(xdp_ring, req_id, tx_info, true);
1299 }
1300
1301 static int ena_clean_tx_irq(struct ena_ring *tx_ring, u32 budget)
1302 {
1303         struct netdev_queue *txq;
1304         bool above_thresh;
1305         u32 tx_bytes = 0;
1306         u32 total_done = 0;
1307         u16 next_to_clean;
1308         u16 req_id;
1309         int tx_pkts = 0;
1310         int rc;
1311
1312         next_to_clean = tx_ring->next_to_clean;
1313         txq = netdev_get_tx_queue(tx_ring->netdev, tx_ring->qid);
1314
1315         while (tx_pkts < budget) {
1316                 struct ena_tx_buffer *tx_info;
1317                 struct sk_buff *skb;
1318
1319                 rc = ena_com_tx_comp_req_id_get(tx_ring->ena_com_io_cq,
1320                                                 &req_id);
1321                 if (rc)
1322                         break;
1323
1324                 rc = validate_tx_req_id(tx_ring, req_id);
1325                 if (rc)
1326                         break;
1327
1328                 tx_info = &tx_ring->tx_buffer_info[req_id];
1329                 skb = tx_info->skb;
1330
1331                 /* prefetch skb_end_pointer() to speedup skb_shinfo(skb) */
1332                 prefetch(&skb->end);
1333
1334                 tx_info->skb = NULL;
1335                 tx_info->last_jiffies = 0;
1336
1337                 ena_unmap_tx_buff(tx_ring, tx_info);
1338
1339                 netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1340                           "tx_poll: q %d skb %p completed\n", tx_ring->qid,
1341                           skb);
1342
1343                 tx_bytes += skb->len;
1344                 dev_kfree_skb(skb);
1345                 tx_pkts++;
1346                 total_done += tx_info->tx_descs;
1347
1348                 tx_ring->free_ids[next_to_clean] = req_id;
1349                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1350                                                      tx_ring->ring_size);
1351         }
1352
1353         tx_ring->next_to_clean = next_to_clean;
1354         ena_com_comp_ack(tx_ring->ena_com_io_sq, total_done);
1355         ena_com_update_dev_comp_head(tx_ring->ena_com_io_cq);
1356
1357         netdev_tx_completed_queue(txq, tx_pkts, tx_bytes);
1358
1359         netif_dbg(tx_ring->adapter, tx_done, tx_ring->netdev,
1360                   "tx_poll: q %d done. total pkts: %d\n",
1361                   tx_ring->qid, tx_pkts);
1362
1363         /* need to make the rings circular update visible to
1364          * ena_start_xmit() before checking for netif_queue_stopped().
1365          */
1366         smp_mb();
1367
1368         above_thresh = ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1369                                                     ENA_TX_WAKEUP_THRESH);
1370         if (unlikely(netif_tx_queue_stopped(txq) && above_thresh)) {
1371                 __netif_tx_lock(txq, smp_processor_id());
1372                 above_thresh =
1373                         ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
1374                                                      ENA_TX_WAKEUP_THRESH);
1375                 if (netif_tx_queue_stopped(txq) && above_thresh &&
1376                     test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags)) {
1377                         netif_tx_wake_queue(txq);
1378                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
1379                                           &tx_ring->syncp);
1380                 }
1381                 __netif_tx_unlock(txq);
1382         }
1383
1384         return tx_pkts;
1385 }
1386
1387 static struct sk_buff *ena_alloc_skb(struct ena_ring *rx_ring, bool frags)
1388 {
1389         struct sk_buff *skb;
1390
1391         if (frags)
1392                 skb = napi_get_frags(rx_ring->napi);
1393         else
1394                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
1395                                                 rx_ring->rx_copybreak);
1396
1397         if (unlikely(!skb)) {
1398                 ena_increase_stat(&rx_ring->rx_stats.skb_alloc_fail, 1,
1399                                   &rx_ring->syncp);
1400                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1401                           "Failed to allocate skb. frags: %d\n", frags);
1402                 return NULL;
1403         }
1404
1405         return skb;
1406 }
1407
1408 static struct sk_buff *ena_rx_skb(struct ena_ring *rx_ring,
1409                                   struct ena_com_rx_buf_info *ena_bufs,
1410                                   u32 descs,
1411                                   u16 *next_to_clean)
1412 {
1413         struct sk_buff *skb;
1414         struct ena_rx_buffer *rx_info;
1415         u16 len, req_id, buf = 0;
1416         void *va;
1417
1418         len = ena_bufs[buf].len;
1419         req_id = ena_bufs[buf].req_id;
1420
1421         rx_info = &rx_ring->rx_buffer_info[req_id];
1422
1423         if (unlikely(!rx_info->page)) {
1424                 netif_err(rx_ring->adapter, rx_err, rx_ring->netdev,
1425                           "Page is NULL\n");
1426                 return NULL;
1427         }
1428
1429         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1430                   "rx_info %p page %p\n",
1431                   rx_info, rx_info->page);
1432
1433         /* save virt address of first buffer */
1434         va = page_address(rx_info->page) + rx_info->page_offset;
1435
1436         prefetch(va);
1437
1438         if (len <= rx_ring->rx_copybreak) {
1439                 skb = ena_alloc_skb(rx_ring, false);
1440                 if (unlikely(!skb))
1441                         return NULL;
1442
1443                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1444                           "RX allocated small packet. len %d. data_len %d\n",
1445                           skb->len, skb->data_len);
1446
1447                 /* sync this buffer for CPU use */
1448                 dma_sync_single_for_cpu(rx_ring->dev,
1449                                         dma_unmap_addr(&rx_info->ena_buf, paddr),
1450                                         len,
1451                                         DMA_FROM_DEVICE);
1452                 skb_copy_to_linear_data(skb, va, len);
1453                 dma_sync_single_for_device(rx_ring->dev,
1454                                            dma_unmap_addr(&rx_info->ena_buf, paddr),
1455                                            len,
1456                                            DMA_FROM_DEVICE);
1457
1458                 skb_put(skb, len);
1459                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1460                 rx_ring->free_ids[*next_to_clean] = req_id;
1461                 *next_to_clean = ENA_RX_RING_IDX_ADD(*next_to_clean, descs,
1462                                                      rx_ring->ring_size);
1463                 return skb;
1464         }
1465
1466         skb = ena_alloc_skb(rx_ring, true);
1467         if (unlikely(!skb))
1468                 return NULL;
1469
1470         do {
1471                 ena_unmap_rx_buff(rx_ring, rx_info);
1472
1473                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_info->page,
1474                                 rx_info->page_offset, len, ENA_PAGE_SIZE);
1475
1476                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1477                           "RX skb updated. len %d. data_len %d\n",
1478                           skb->len, skb->data_len);
1479
1480                 rx_info->page = NULL;
1481
1482                 rx_ring->free_ids[*next_to_clean] = req_id;
1483                 *next_to_clean =
1484                         ENA_RX_RING_IDX_NEXT(*next_to_clean,
1485                                              rx_ring->ring_size);
1486                 if (likely(--descs == 0))
1487                         break;
1488
1489                 buf++;
1490                 len = ena_bufs[buf].len;
1491                 req_id = ena_bufs[buf].req_id;
1492
1493                 rx_info = &rx_ring->rx_buffer_info[req_id];
1494         } while (1);
1495
1496         return skb;
1497 }
1498
1499 /* ena_rx_checksum - indicate in skb if hw indicated a good cksum
1500  * @adapter: structure containing adapter specific data
1501  * @ena_rx_ctx: received packet context/metadata
1502  * @skb: skb currently being received and modified
1503  */
1504 static void ena_rx_checksum(struct ena_ring *rx_ring,
1505                                    struct ena_com_rx_ctx *ena_rx_ctx,
1506                                    struct sk_buff *skb)
1507 {
1508         /* Rx csum disabled */
1509         if (unlikely(!(rx_ring->netdev->features & NETIF_F_RXCSUM))) {
1510                 skb->ip_summed = CHECKSUM_NONE;
1511                 return;
1512         }
1513
1514         /* For fragmented packets the checksum isn't valid */
1515         if (ena_rx_ctx->frag) {
1516                 skb->ip_summed = CHECKSUM_NONE;
1517                 return;
1518         }
1519
1520         /* if IP and error */
1521         if (unlikely((ena_rx_ctx->l3_proto == ENA_ETH_IO_L3_PROTO_IPV4) &&
1522                      (ena_rx_ctx->l3_csum_err))) {
1523                 /* ipv4 checksum error */
1524                 skb->ip_summed = CHECKSUM_NONE;
1525                 ena_increase_stat(&rx_ring->rx_stats.bad_csum, 1,
1526                                   &rx_ring->syncp);
1527                 netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1528                           "RX IPv4 header checksum error\n");
1529                 return;
1530         }
1531
1532         /* if TCP/UDP */
1533         if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1534                    (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP))) {
1535                 if (unlikely(ena_rx_ctx->l4_csum_err)) {
1536                         /* TCP/UDP checksum error */
1537                         ena_increase_stat(&rx_ring->rx_stats.bad_csum, 1,
1538                                           &rx_ring->syncp);
1539                         netif_dbg(rx_ring->adapter, rx_err, rx_ring->netdev,
1540                                   "RX L4 checksum error\n");
1541                         skb->ip_summed = CHECKSUM_NONE;
1542                         return;
1543                 }
1544
1545                 if (likely(ena_rx_ctx->l4_csum_checked)) {
1546                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1547                         ena_increase_stat(&rx_ring->rx_stats.csum_good, 1,
1548                                           &rx_ring->syncp);
1549                 } else {
1550                         ena_increase_stat(&rx_ring->rx_stats.csum_unchecked, 1,
1551                                           &rx_ring->syncp);
1552                         skb->ip_summed = CHECKSUM_NONE;
1553                 }
1554         } else {
1555                 skb->ip_summed = CHECKSUM_NONE;
1556                 return;
1557         }
1558
1559 }
1560
1561 static void ena_set_rx_hash(struct ena_ring *rx_ring,
1562                             struct ena_com_rx_ctx *ena_rx_ctx,
1563                             struct sk_buff *skb)
1564 {
1565         enum pkt_hash_types hash_type;
1566
1567         if (likely(rx_ring->netdev->features & NETIF_F_RXHASH)) {
1568                 if (likely((ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_TCP) ||
1569                            (ena_rx_ctx->l4_proto == ENA_ETH_IO_L4_PROTO_UDP)))
1570
1571                         hash_type = PKT_HASH_TYPE_L4;
1572                 else
1573                         hash_type = PKT_HASH_TYPE_NONE;
1574
1575                 /* Override hash type if the packet is fragmented */
1576                 if (ena_rx_ctx->frag)
1577                         hash_type = PKT_HASH_TYPE_NONE;
1578
1579                 skb_set_hash(skb, ena_rx_ctx->hash, hash_type);
1580         }
1581 }
1582
1583 static int ena_xdp_handle_buff(struct ena_ring *rx_ring, struct xdp_buff *xdp)
1584 {
1585         struct ena_rx_buffer *rx_info;
1586         int ret;
1587
1588         rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1589         xdp_prepare_buff(xdp, page_address(rx_info->page),
1590                          rx_info->page_offset,
1591                          rx_ring->ena_bufs[0].len, false);
1592         /* If for some reason we received a bigger packet than
1593          * we expect, then we simply drop it
1594          */
1595         if (unlikely(rx_ring->ena_bufs[0].len > ENA_XDP_MAX_MTU))
1596                 return XDP_DROP;
1597
1598         ret = ena_xdp_execute(rx_ring, xdp);
1599
1600         /* The xdp program might expand the headers */
1601         if (ret == XDP_PASS) {
1602                 rx_info->page_offset = xdp->data - xdp->data_hard_start;
1603                 rx_ring->ena_bufs[0].len = xdp->data_end - xdp->data;
1604         }
1605
1606         return ret;
1607 }
1608 /* ena_clean_rx_irq - Cleanup RX irq
1609  * @rx_ring: RX ring to clean
1610  * @napi: napi handler
1611  * @budget: how many packets driver is allowed to clean
1612  *
1613  * Returns the number of cleaned buffers.
1614  */
1615 static int ena_clean_rx_irq(struct ena_ring *rx_ring, struct napi_struct *napi,
1616                             u32 budget)
1617 {
1618         u16 next_to_clean = rx_ring->next_to_clean;
1619         struct ena_com_rx_ctx ena_rx_ctx;
1620         struct ena_rx_buffer *rx_info;
1621         struct ena_adapter *adapter;
1622         u32 res_budget, work_done;
1623         int rx_copybreak_pkt = 0;
1624         int refill_threshold;
1625         struct sk_buff *skb;
1626         int refill_required;
1627         struct xdp_buff xdp;
1628         int xdp_flags = 0;
1629         int total_len = 0;
1630         int xdp_verdict;
1631         int rc = 0;
1632         int i;
1633
1634         netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1635                   "%s qid %d\n", __func__, rx_ring->qid);
1636         res_budget = budget;
1637         xdp_init_buff(&xdp, ENA_PAGE_SIZE, &rx_ring->xdp_rxq);
1638
1639         do {
1640                 xdp_verdict = XDP_PASS;
1641                 skb = NULL;
1642                 ena_rx_ctx.ena_bufs = rx_ring->ena_bufs;
1643                 ena_rx_ctx.max_bufs = rx_ring->sgl_size;
1644                 ena_rx_ctx.descs = 0;
1645                 ena_rx_ctx.pkt_offset = 0;
1646                 rc = ena_com_rx_pkt(rx_ring->ena_com_io_cq,
1647                                     rx_ring->ena_com_io_sq,
1648                                     &ena_rx_ctx);
1649                 if (unlikely(rc))
1650                         goto error;
1651
1652                 if (unlikely(ena_rx_ctx.descs == 0))
1653                         break;
1654
1655                 /* First descriptor might have an offset set by the device */
1656                 rx_info = &rx_ring->rx_buffer_info[rx_ring->ena_bufs[0].req_id];
1657                 rx_info->page_offset += ena_rx_ctx.pkt_offset;
1658
1659                 netif_dbg(rx_ring->adapter, rx_status, rx_ring->netdev,
1660                           "rx_poll: q %d got packet from ena. descs #: %d l3 proto %d l4 proto %d hash: %x\n",
1661                           rx_ring->qid, ena_rx_ctx.descs, ena_rx_ctx.l3_proto,
1662                           ena_rx_ctx.l4_proto, ena_rx_ctx.hash);
1663
1664                 if (ena_xdp_present_ring(rx_ring))
1665                         xdp_verdict = ena_xdp_handle_buff(rx_ring, &xdp);
1666
1667                 /* allocate skb and fill it */
1668                 if (xdp_verdict == XDP_PASS)
1669                         skb = ena_rx_skb(rx_ring,
1670                                          rx_ring->ena_bufs,
1671                                          ena_rx_ctx.descs,
1672                                          &next_to_clean);
1673
1674                 if (unlikely(!skb)) {
1675                         for (i = 0; i < ena_rx_ctx.descs; i++) {
1676                                 int req_id = rx_ring->ena_bufs[i].req_id;
1677
1678                                 rx_ring->free_ids[next_to_clean] = req_id;
1679                                 next_to_clean =
1680                                         ENA_RX_RING_IDX_NEXT(next_to_clean,
1681                                                              rx_ring->ring_size);
1682
1683                                 /* Packets was passed for transmission, unmap it
1684                                  * from RX side.
1685                                  */
1686                                 if (xdp_verdict == XDP_TX || xdp_verdict == XDP_REDIRECT) {
1687                                         ena_unmap_rx_buff(rx_ring,
1688                                                           &rx_ring->rx_buffer_info[req_id]);
1689                                         rx_ring->rx_buffer_info[req_id].page = NULL;
1690                                 }
1691                         }
1692                         if (xdp_verdict != XDP_PASS) {
1693                                 xdp_flags |= xdp_verdict;
1694                                 res_budget--;
1695                                 continue;
1696                         }
1697                         break;
1698                 }
1699
1700                 ena_rx_checksum(rx_ring, &ena_rx_ctx, skb);
1701
1702                 ena_set_rx_hash(rx_ring, &ena_rx_ctx, skb);
1703
1704                 skb_record_rx_queue(skb, rx_ring->qid);
1705
1706                 if (rx_ring->ena_bufs[0].len <= rx_ring->rx_copybreak) {
1707                         total_len += rx_ring->ena_bufs[0].len;
1708                         rx_copybreak_pkt++;
1709                         napi_gro_receive(napi, skb);
1710                 } else {
1711                         total_len += skb->len;
1712                         napi_gro_frags(napi);
1713                 }
1714
1715                 res_budget--;
1716         } while (likely(res_budget));
1717
1718         work_done = budget - res_budget;
1719         rx_ring->per_napi_packets += work_done;
1720         u64_stats_update_begin(&rx_ring->syncp);
1721         rx_ring->rx_stats.bytes += total_len;
1722         rx_ring->rx_stats.cnt += work_done;
1723         rx_ring->rx_stats.rx_copybreak_pkt += rx_copybreak_pkt;
1724         u64_stats_update_end(&rx_ring->syncp);
1725
1726         rx_ring->next_to_clean = next_to_clean;
1727
1728         refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
1729         refill_threshold =
1730                 min_t(int, rx_ring->ring_size / ENA_RX_REFILL_THRESH_DIVIDER,
1731                       ENA_RX_REFILL_THRESH_PACKET);
1732
1733         /* Optimization, try to batch new rx buffers */
1734         if (refill_required > refill_threshold) {
1735                 ena_com_update_dev_comp_head(rx_ring->ena_com_io_cq);
1736                 ena_refill_rx_bufs(rx_ring, refill_required);
1737         }
1738
1739         if (xdp_flags & XDP_REDIRECT)
1740                 xdp_do_flush_map();
1741
1742         return work_done;
1743
1744 error:
1745         adapter = netdev_priv(rx_ring->netdev);
1746
1747         if (rc == -ENOSPC) {
1748                 ena_increase_stat(&rx_ring->rx_stats.bad_desc_num, 1,
1749                                   &rx_ring->syncp);
1750                 adapter->reset_reason = ENA_REGS_RESET_TOO_MANY_RX_DESCS;
1751         } else {
1752                 ena_increase_stat(&rx_ring->rx_stats.bad_req_id, 1,
1753                                   &rx_ring->syncp);
1754                 adapter->reset_reason = ENA_REGS_RESET_INV_RX_REQ_ID;
1755         }
1756
1757         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
1758
1759         return 0;
1760 }
1761
1762 static void ena_dim_work(struct work_struct *w)
1763 {
1764         struct dim *dim = container_of(w, struct dim, work);
1765         struct dim_cq_moder cur_moder =
1766                 net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1767         struct ena_napi *ena_napi = container_of(dim, struct ena_napi, dim);
1768
1769         ena_napi->rx_ring->smoothed_interval = cur_moder.usec;
1770         dim->state = DIM_START_MEASURE;
1771 }
1772
1773 static void ena_adjust_adaptive_rx_intr_moderation(struct ena_napi *ena_napi)
1774 {
1775         struct dim_sample dim_sample;
1776         struct ena_ring *rx_ring = ena_napi->rx_ring;
1777
1778         if (!rx_ring->per_napi_packets)
1779                 return;
1780
1781         rx_ring->non_empty_napi_events++;
1782
1783         dim_update_sample(rx_ring->non_empty_napi_events,
1784                           rx_ring->rx_stats.cnt,
1785                           rx_ring->rx_stats.bytes,
1786                           &dim_sample);
1787
1788         net_dim(&ena_napi->dim, dim_sample);
1789
1790         rx_ring->per_napi_packets = 0;
1791 }
1792
1793 static void ena_unmask_interrupt(struct ena_ring *tx_ring,
1794                                         struct ena_ring *rx_ring)
1795 {
1796         struct ena_eth_io_intr_reg intr_reg;
1797         u32 rx_interval = 0;
1798         /* Rx ring can be NULL when for XDP tx queues which don't have an
1799          * accompanying rx_ring pair.
1800          */
1801         if (rx_ring)
1802                 rx_interval = ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev) ?
1803                         rx_ring->smoothed_interval :
1804                         ena_com_get_nonadaptive_moderation_interval_rx(rx_ring->ena_dev);
1805
1806         /* Update intr register: rx intr delay,
1807          * tx intr delay and interrupt unmask
1808          */
1809         ena_com_update_intr_reg(&intr_reg,
1810                                 rx_interval,
1811                                 tx_ring->smoothed_interval,
1812                                 true);
1813
1814         ena_increase_stat(&tx_ring->tx_stats.unmask_interrupt, 1,
1815                           &tx_ring->syncp);
1816
1817         /* It is a shared MSI-X.
1818          * Tx and Rx CQ have pointer to it.
1819          * So we use one of them to reach the intr reg
1820          * The Tx ring is used because the rx_ring is NULL for XDP queues
1821          */
1822         ena_com_unmask_intr(tx_ring->ena_com_io_cq, &intr_reg);
1823 }
1824
1825 static void ena_update_ring_numa_node(struct ena_ring *tx_ring,
1826                                              struct ena_ring *rx_ring)
1827 {
1828         int cpu = get_cpu();
1829         int numa_node;
1830
1831         /* Check only one ring since the 2 rings are running on the same cpu */
1832         if (likely(tx_ring->cpu == cpu))
1833                 goto out;
1834
1835         numa_node = cpu_to_node(cpu);
1836         put_cpu();
1837
1838         if (numa_node != NUMA_NO_NODE) {
1839                 ena_com_update_numa_node(tx_ring->ena_com_io_cq, numa_node);
1840                 if (rx_ring)
1841                         ena_com_update_numa_node(rx_ring->ena_com_io_cq,
1842                                                  numa_node);
1843         }
1844
1845         tx_ring->cpu = cpu;
1846         if (rx_ring)
1847                 rx_ring->cpu = cpu;
1848
1849         return;
1850 out:
1851         put_cpu();
1852 }
1853
1854 static int ena_clean_xdp_irq(struct ena_ring *xdp_ring, u32 budget)
1855 {
1856         u32 total_done = 0;
1857         u16 next_to_clean;
1858         u32 tx_bytes = 0;
1859         int tx_pkts = 0;
1860         u16 req_id;
1861         int rc;
1862
1863         if (unlikely(!xdp_ring))
1864                 return 0;
1865         next_to_clean = xdp_ring->next_to_clean;
1866
1867         while (tx_pkts < budget) {
1868                 struct ena_tx_buffer *tx_info;
1869                 struct xdp_frame *xdpf;
1870
1871                 rc = ena_com_tx_comp_req_id_get(xdp_ring->ena_com_io_cq,
1872                                                 &req_id);
1873                 if (rc)
1874                         break;
1875
1876                 rc = validate_xdp_req_id(xdp_ring, req_id);
1877                 if (rc)
1878                         break;
1879
1880                 tx_info = &xdp_ring->tx_buffer_info[req_id];
1881                 xdpf = tx_info->xdpf;
1882
1883                 tx_info->xdpf = NULL;
1884                 tx_info->last_jiffies = 0;
1885                 ena_unmap_tx_buff(xdp_ring, tx_info);
1886
1887                 netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1888                           "tx_poll: q %d skb %p completed\n", xdp_ring->qid,
1889                           xdpf);
1890
1891                 tx_bytes += xdpf->len;
1892                 tx_pkts++;
1893                 total_done += tx_info->tx_descs;
1894
1895                 xdp_return_frame(xdpf);
1896                 xdp_ring->free_ids[next_to_clean] = req_id;
1897                 next_to_clean = ENA_TX_RING_IDX_NEXT(next_to_clean,
1898                                                      xdp_ring->ring_size);
1899         }
1900
1901         xdp_ring->next_to_clean = next_to_clean;
1902         ena_com_comp_ack(xdp_ring->ena_com_io_sq, total_done);
1903         ena_com_update_dev_comp_head(xdp_ring->ena_com_io_cq);
1904
1905         netif_dbg(xdp_ring->adapter, tx_done, xdp_ring->netdev,
1906                   "tx_poll: q %d done. total pkts: %d\n",
1907                   xdp_ring->qid, tx_pkts);
1908
1909         return tx_pkts;
1910 }
1911
1912 static int ena_io_poll(struct napi_struct *napi, int budget)
1913 {
1914         struct ena_napi *ena_napi = container_of(napi, struct ena_napi, napi);
1915         struct ena_ring *tx_ring, *rx_ring;
1916         int tx_work_done;
1917         int rx_work_done = 0;
1918         int tx_budget;
1919         int napi_comp_call = 0;
1920         int ret;
1921
1922         tx_ring = ena_napi->tx_ring;
1923         rx_ring = ena_napi->rx_ring;
1924
1925         tx_ring->first_interrupt = ena_napi->first_interrupt;
1926         rx_ring->first_interrupt = ena_napi->first_interrupt;
1927
1928         tx_budget = tx_ring->ring_size / ENA_TX_POLL_BUDGET_DIVIDER;
1929
1930         if (!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1931             test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags)) {
1932                 napi_complete_done(napi, 0);
1933                 return 0;
1934         }
1935
1936         tx_work_done = ena_clean_tx_irq(tx_ring, tx_budget);
1937         /* On netpoll the budget is zero and the handler should only clean the
1938          * tx completions.
1939          */
1940         if (likely(budget))
1941                 rx_work_done = ena_clean_rx_irq(rx_ring, napi, budget);
1942
1943         /* If the device is about to reset or down, avoid unmask
1944          * the interrupt and return 0 so NAPI won't reschedule
1945          */
1946         if (unlikely(!test_bit(ENA_FLAG_DEV_UP, &tx_ring->adapter->flags) ||
1947                      test_bit(ENA_FLAG_TRIGGER_RESET, &tx_ring->adapter->flags))) {
1948                 napi_complete_done(napi, 0);
1949                 ret = 0;
1950
1951         } else if ((budget > rx_work_done) && (tx_budget > tx_work_done)) {
1952                 napi_comp_call = 1;
1953
1954                 /* Update numa and unmask the interrupt only when schedule
1955                  * from the interrupt context (vs from sk_busy_loop)
1956                  */
1957                 if (napi_complete_done(napi, rx_work_done) &&
1958                     READ_ONCE(ena_napi->interrupts_masked)) {
1959                         smp_rmb(); /* make sure interrupts_masked is read */
1960                         WRITE_ONCE(ena_napi->interrupts_masked, false);
1961                         /* We apply adaptive moderation on Rx path only.
1962                          * Tx uses static interrupt moderation.
1963                          */
1964                         if (ena_com_get_adaptive_moderation_enabled(rx_ring->ena_dev))
1965                                 ena_adjust_adaptive_rx_intr_moderation(ena_napi);
1966
1967                         ena_unmask_interrupt(tx_ring, rx_ring);
1968                 }
1969
1970                 ena_update_ring_numa_node(tx_ring, rx_ring);
1971
1972                 ret = rx_work_done;
1973         } else {
1974                 ret = budget;
1975         }
1976
1977         u64_stats_update_begin(&tx_ring->syncp);
1978         tx_ring->tx_stats.napi_comp += napi_comp_call;
1979         tx_ring->tx_stats.tx_poll++;
1980         u64_stats_update_end(&tx_ring->syncp);
1981
1982         return ret;
1983 }
1984
1985 static irqreturn_t ena_intr_msix_mgmnt(int irq, void *data)
1986 {
1987         struct ena_adapter *adapter = (struct ena_adapter *)data;
1988
1989         ena_com_admin_q_comp_intr_handler(adapter->ena_dev);
1990
1991         /* Don't call the aenq handler before probe is done */
1992         if (likely(test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags)))
1993                 ena_com_aenq_intr_handler(adapter->ena_dev, data);
1994
1995         return IRQ_HANDLED;
1996 }
1997
1998 /* ena_intr_msix_io - MSI-X Interrupt Handler for Tx/Rx
1999  * @irq: interrupt number
2000  * @data: pointer to a network interface private napi device structure
2001  */
2002 static irqreturn_t ena_intr_msix_io(int irq, void *data)
2003 {
2004         struct ena_napi *ena_napi = data;
2005
2006         ena_napi->first_interrupt = true;
2007
2008         WRITE_ONCE(ena_napi->interrupts_masked, true);
2009         smp_wmb(); /* write interrupts_masked before calling napi */
2010
2011         napi_schedule_irqoff(&ena_napi->napi);
2012
2013         return IRQ_HANDLED;
2014 }
2015
2016 /* Reserve a single MSI-X vector for management (admin + aenq).
2017  * plus reserve one vector for each potential io queue.
2018  * the number of potential io queues is the minimum of what the device
2019  * supports and the number of vCPUs.
2020  */
2021 static int ena_enable_msix(struct ena_adapter *adapter)
2022 {
2023         int msix_vecs, irq_cnt;
2024
2025         if (test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2026                 netif_err(adapter, probe, adapter->netdev,
2027                           "Error, MSI-X is already enabled\n");
2028                 return -EPERM;
2029         }
2030
2031         /* Reserved the max msix vectors we might need */
2032         msix_vecs = ENA_MAX_MSIX_VEC(adapter->max_num_io_queues);
2033         netif_dbg(adapter, probe, adapter->netdev,
2034                   "Trying to enable MSI-X, vectors %d\n", msix_vecs);
2035
2036         irq_cnt = pci_alloc_irq_vectors(adapter->pdev, ENA_MIN_MSIX_VEC,
2037                                         msix_vecs, PCI_IRQ_MSIX);
2038
2039         if (irq_cnt < 0) {
2040                 netif_err(adapter, probe, adapter->netdev,
2041                           "Failed to enable MSI-X. irq_cnt %d\n", irq_cnt);
2042                 return -ENOSPC;
2043         }
2044
2045         if (irq_cnt != msix_vecs) {
2046                 netif_notice(adapter, probe, adapter->netdev,
2047                              "Enable only %d MSI-X (out of %d), reduce the number of queues\n",
2048                              irq_cnt, msix_vecs);
2049                 adapter->num_io_queues = irq_cnt - ENA_ADMIN_MSIX_VEC;
2050         }
2051
2052         if (ena_init_rx_cpu_rmap(adapter))
2053                 netif_warn(adapter, probe, adapter->netdev,
2054                            "Failed to map IRQs to CPUs\n");
2055
2056         adapter->msix_vecs = irq_cnt;
2057         set_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags);
2058
2059         return 0;
2060 }
2061
2062 static void ena_setup_mgmnt_intr(struct ena_adapter *adapter)
2063 {
2064         u32 cpu;
2065
2066         snprintf(adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].name,
2067                  ENA_IRQNAME_SIZE, "ena-mgmnt@pci:%s",
2068                  pci_name(adapter->pdev));
2069         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].handler =
2070                 ena_intr_msix_mgmnt;
2071         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].data = adapter;
2072         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].vector =
2073                 pci_irq_vector(adapter->pdev, ENA_MGMNT_IRQ_IDX);
2074         cpu = cpumask_first(cpu_online_mask);
2075         adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].cpu = cpu;
2076         cpumask_set_cpu(cpu,
2077                         &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX].affinity_hint_mask);
2078 }
2079
2080 static void ena_setup_io_intr(struct ena_adapter *adapter)
2081 {
2082         struct net_device *netdev;
2083         int irq_idx, i, cpu;
2084         int io_queue_count;
2085
2086         netdev = adapter->netdev;
2087         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2088
2089         for (i = 0; i < io_queue_count; i++) {
2090                 irq_idx = ENA_IO_IRQ_IDX(i);
2091                 cpu = i % num_online_cpus();
2092
2093                 snprintf(adapter->irq_tbl[irq_idx].name, ENA_IRQNAME_SIZE,
2094                          "%s-Tx-Rx-%d", netdev->name, i);
2095                 adapter->irq_tbl[irq_idx].handler = ena_intr_msix_io;
2096                 adapter->irq_tbl[irq_idx].data = &adapter->ena_napi[i];
2097                 adapter->irq_tbl[irq_idx].vector =
2098                         pci_irq_vector(adapter->pdev, irq_idx);
2099                 adapter->irq_tbl[irq_idx].cpu = cpu;
2100
2101                 cpumask_set_cpu(cpu,
2102                                 &adapter->irq_tbl[irq_idx].affinity_hint_mask);
2103         }
2104 }
2105
2106 static int ena_request_mgmnt_irq(struct ena_adapter *adapter)
2107 {
2108         unsigned long flags = 0;
2109         struct ena_irq *irq;
2110         int rc;
2111
2112         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2113         rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2114                          irq->data);
2115         if (rc) {
2116                 netif_err(adapter, probe, adapter->netdev,
2117                           "Failed to request admin irq\n");
2118                 return rc;
2119         }
2120
2121         netif_dbg(adapter, probe, adapter->netdev,
2122                   "Set affinity hint of mgmnt irq.to 0x%lx (irq vector: %d)\n",
2123                   irq->affinity_hint_mask.bits[0], irq->vector);
2124
2125         irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2126
2127         return rc;
2128 }
2129
2130 static int ena_request_io_irq(struct ena_adapter *adapter)
2131 {
2132         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2133         unsigned long flags = 0;
2134         struct ena_irq *irq;
2135         int rc = 0, i, k;
2136
2137         if (!test_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags)) {
2138                 netif_err(adapter, ifup, adapter->netdev,
2139                           "Failed to request I/O IRQ: MSI-X is not enabled\n");
2140                 return -EINVAL;
2141         }
2142
2143         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2144                 irq = &adapter->irq_tbl[i];
2145                 rc = request_irq(irq->vector, irq->handler, flags, irq->name,
2146                                  irq->data);
2147                 if (rc) {
2148                         netif_err(adapter, ifup, adapter->netdev,
2149                                   "Failed to request I/O IRQ. index %d rc %d\n",
2150                                    i, rc);
2151                         goto err;
2152                 }
2153
2154                 netif_dbg(adapter, ifup, adapter->netdev,
2155                           "Set affinity hint of irq. index %d to 0x%lx (irq vector: %d)\n",
2156                           i, irq->affinity_hint_mask.bits[0], irq->vector);
2157
2158                 irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask);
2159         }
2160
2161         return rc;
2162
2163 err:
2164         for (k = ENA_IO_IRQ_FIRST_IDX; k < i; k++) {
2165                 irq = &adapter->irq_tbl[k];
2166                 free_irq(irq->vector, irq->data);
2167         }
2168
2169         return rc;
2170 }
2171
2172 static void ena_free_mgmnt_irq(struct ena_adapter *adapter)
2173 {
2174         struct ena_irq *irq;
2175
2176         irq = &adapter->irq_tbl[ENA_MGMNT_IRQ_IDX];
2177         synchronize_irq(irq->vector);
2178         irq_set_affinity_hint(irq->vector, NULL);
2179         free_irq(irq->vector, irq->data);
2180 }
2181
2182 static void ena_free_io_irq(struct ena_adapter *adapter)
2183 {
2184         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2185         struct ena_irq *irq;
2186         int i;
2187
2188 #ifdef CONFIG_RFS_ACCEL
2189         if (adapter->msix_vecs >= 1) {
2190                 free_irq_cpu_rmap(adapter->netdev->rx_cpu_rmap);
2191                 adapter->netdev->rx_cpu_rmap = NULL;
2192         }
2193 #endif /* CONFIG_RFS_ACCEL */
2194
2195         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++) {
2196                 irq = &adapter->irq_tbl[i];
2197                 irq_set_affinity_hint(irq->vector, NULL);
2198                 free_irq(irq->vector, irq->data);
2199         }
2200 }
2201
2202 static void ena_disable_msix(struct ena_adapter *adapter)
2203 {
2204         if (test_and_clear_bit(ENA_FLAG_MSIX_ENABLED, &adapter->flags))
2205                 pci_free_irq_vectors(adapter->pdev);
2206 }
2207
2208 static void ena_disable_io_intr_sync(struct ena_adapter *adapter)
2209 {
2210         u32 io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2211         int i;
2212
2213         if (!netif_running(adapter->netdev))
2214                 return;
2215
2216         for (i = ENA_IO_IRQ_FIRST_IDX; i < ENA_MAX_MSIX_VEC(io_queue_count); i++)
2217                 synchronize_irq(adapter->irq_tbl[i].vector);
2218 }
2219
2220 static void ena_del_napi_in_range(struct ena_adapter *adapter,
2221                                   int first_index,
2222                                   int count)
2223 {
2224         int i;
2225
2226         for (i = first_index; i < first_index + count; i++) {
2227                 netif_napi_del(&adapter->ena_napi[i].napi);
2228
2229                 WARN_ON(!ENA_IS_XDP_INDEX(adapter, i) &&
2230                         adapter->ena_napi[i].xdp_ring);
2231         }
2232 }
2233
2234 static void ena_init_napi_in_range(struct ena_adapter *adapter,
2235                                    int first_index, int count)
2236 {
2237         int i;
2238
2239         for (i = first_index; i < first_index + count; i++) {
2240                 struct ena_napi *napi = &adapter->ena_napi[i];
2241
2242                 netif_napi_add(adapter->netdev,
2243                                &napi->napi,
2244                                ENA_IS_XDP_INDEX(adapter, i) ? ena_xdp_io_poll : ena_io_poll,
2245                                ENA_NAPI_BUDGET);
2246
2247                 if (!ENA_IS_XDP_INDEX(adapter, i)) {
2248                         napi->rx_ring = &adapter->rx_ring[i];
2249                         napi->tx_ring = &adapter->tx_ring[i];
2250                 } else {
2251                         napi->xdp_ring = &adapter->tx_ring[i];
2252                 }
2253                 napi->qid = i;
2254         }
2255 }
2256
2257 static void ena_napi_disable_in_range(struct ena_adapter *adapter,
2258                                       int first_index,
2259                                       int count)
2260 {
2261         int i;
2262
2263         for (i = first_index; i < first_index + count; i++)
2264                 napi_disable(&adapter->ena_napi[i].napi);
2265 }
2266
2267 static void ena_napi_enable_in_range(struct ena_adapter *adapter,
2268                                      int first_index,
2269                                      int count)
2270 {
2271         int i;
2272
2273         for (i = first_index; i < first_index + count; i++)
2274                 napi_enable(&adapter->ena_napi[i].napi);
2275 }
2276
2277 /* Configure the Rx forwarding */
2278 static int ena_rss_configure(struct ena_adapter *adapter)
2279 {
2280         struct ena_com_dev *ena_dev = adapter->ena_dev;
2281         int rc;
2282
2283         /* In case the RSS table wasn't initialized by probe */
2284         if (!ena_dev->rss.tbl_log_size) {
2285                 rc = ena_rss_init_default(adapter);
2286                 if (rc && (rc != -EOPNOTSUPP)) {
2287                         netif_err(adapter, ifup, adapter->netdev,
2288                                   "Failed to init RSS rc: %d\n", rc);
2289                         return rc;
2290                 }
2291         }
2292
2293         /* Set indirect table */
2294         rc = ena_com_indirect_table_set(ena_dev);
2295         if (unlikely(rc && rc != -EOPNOTSUPP))
2296                 return rc;
2297
2298         /* Configure hash function (if supported) */
2299         rc = ena_com_set_hash_function(ena_dev);
2300         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2301                 return rc;
2302
2303         /* Configure hash inputs (if supported) */
2304         rc = ena_com_set_hash_ctrl(ena_dev);
2305         if (unlikely(rc && (rc != -EOPNOTSUPP)))
2306                 return rc;
2307
2308         return 0;
2309 }
2310
2311 static int ena_up_complete(struct ena_adapter *adapter)
2312 {
2313         int rc;
2314
2315         rc = ena_rss_configure(adapter);
2316         if (rc)
2317                 return rc;
2318
2319         ena_change_mtu(adapter->netdev, adapter->netdev->mtu);
2320
2321         ena_refill_all_rx_bufs(adapter);
2322
2323         /* enable transmits */
2324         netif_tx_start_all_queues(adapter->netdev);
2325
2326         ena_napi_enable_in_range(adapter,
2327                                  0,
2328                                  adapter->xdp_num_queues + adapter->num_io_queues);
2329
2330         return 0;
2331 }
2332
2333 static int ena_create_io_tx_queue(struct ena_adapter *adapter, int qid)
2334 {
2335         struct ena_com_create_io_ctx ctx;
2336         struct ena_com_dev *ena_dev;
2337         struct ena_ring *tx_ring;
2338         u32 msix_vector;
2339         u16 ena_qid;
2340         int rc;
2341
2342         ena_dev = adapter->ena_dev;
2343
2344         tx_ring = &adapter->tx_ring[qid];
2345         msix_vector = ENA_IO_IRQ_IDX(qid);
2346         ena_qid = ENA_IO_TXQ_IDX(qid);
2347
2348         memset(&ctx, 0x0, sizeof(ctx));
2349
2350         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_TX;
2351         ctx.qid = ena_qid;
2352         ctx.mem_queue_type = ena_dev->tx_mem_queue_type;
2353         ctx.msix_vector = msix_vector;
2354         ctx.queue_size = tx_ring->ring_size;
2355         ctx.numa_node = cpu_to_node(tx_ring->cpu);
2356
2357         rc = ena_com_create_io_queue(ena_dev, &ctx);
2358         if (rc) {
2359                 netif_err(adapter, ifup, adapter->netdev,
2360                           "Failed to create I/O TX queue num %d rc: %d\n",
2361                           qid, rc);
2362                 return rc;
2363         }
2364
2365         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2366                                      &tx_ring->ena_com_io_sq,
2367                                      &tx_ring->ena_com_io_cq);
2368         if (rc) {
2369                 netif_err(adapter, ifup, adapter->netdev,
2370                           "Failed to get TX queue handlers. TX queue num %d rc: %d\n",
2371                           qid, rc);
2372                 ena_com_destroy_io_queue(ena_dev, ena_qid);
2373                 return rc;
2374         }
2375
2376         ena_com_update_numa_node(tx_ring->ena_com_io_cq, ctx.numa_node);
2377         return rc;
2378 }
2379
2380 static int ena_create_io_tx_queues_in_range(struct ena_adapter *adapter,
2381                                             int first_index, int count)
2382 {
2383         struct ena_com_dev *ena_dev = adapter->ena_dev;
2384         int rc, i;
2385
2386         for (i = first_index; i < first_index + count; i++) {
2387                 rc = ena_create_io_tx_queue(adapter, i);
2388                 if (rc)
2389                         goto create_err;
2390         }
2391
2392         return 0;
2393
2394 create_err:
2395         while (i-- > first_index)
2396                 ena_com_destroy_io_queue(ena_dev, ENA_IO_TXQ_IDX(i));
2397
2398         return rc;
2399 }
2400
2401 static int ena_create_io_rx_queue(struct ena_adapter *adapter, int qid)
2402 {
2403         struct ena_com_dev *ena_dev;
2404         struct ena_com_create_io_ctx ctx;
2405         struct ena_ring *rx_ring;
2406         u32 msix_vector;
2407         u16 ena_qid;
2408         int rc;
2409
2410         ena_dev = adapter->ena_dev;
2411
2412         rx_ring = &adapter->rx_ring[qid];
2413         msix_vector = ENA_IO_IRQ_IDX(qid);
2414         ena_qid = ENA_IO_RXQ_IDX(qid);
2415
2416         memset(&ctx, 0x0, sizeof(ctx));
2417
2418         ctx.qid = ena_qid;
2419         ctx.direction = ENA_COM_IO_QUEUE_DIRECTION_RX;
2420         ctx.mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
2421         ctx.msix_vector = msix_vector;
2422         ctx.queue_size = rx_ring->ring_size;
2423         ctx.numa_node = cpu_to_node(rx_ring->cpu);
2424
2425         rc = ena_com_create_io_queue(ena_dev, &ctx);
2426         if (rc) {
2427                 netif_err(adapter, ifup, adapter->netdev,
2428                           "Failed to create I/O RX queue num %d rc: %d\n",
2429                           qid, rc);
2430                 return rc;
2431         }
2432
2433         rc = ena_com_get_io_handlers(ena_dev, ena_qid,
2434                                      &rx_ring->ena_com_io_sq,
2435                                      &rx_ring->ena_com_io_cq);
2436         if (rc) {
2437                 netif_err(adapter, ifup, adapter->netdev,
2438                           "Failed to get RX queue handlers. RX queue num %d rc: %d\n",
2439                           qid, rc);
2440                 goto err;
2441         }
2442
2443         ena_com_update_numa_node(rx_ring->ena_com_io_cq, ctx.numa_node);
2444
2445         return rc;
2446 err:
2447         ena_com_destroy_io_queue(ena_dev, ena_qid);
2448         return rc;
2449 }
2450
2451 static int ena_create_all_io_rx_queues(struct ena_adapter *adapter)
2452 {
2453         struct ena_com_dev *ena_dev = adapter->ena_dev;
2454         int rc, i;
2455
2456         for (i = 0; i < adapter->num_io_queues; i++) {
2457                 rc = ena_create_io_rx_queue(adapter, i);
2458                 if (rc)
2459                         goto create_err;
2460                 INIT_WORK(&adapter->ena_napi[i].dim.work, ena_dim_work);
2461         }
2462
2463         return 0;
2464
2465 create_err:
2466         while (i--) {
2467                 cancel_work_sync(&adapter->ena_napi[i].dim.work);
2468                 ena_com_destroy_io_queue(ena_dev, ENA_IO_RXQ_IDX(i));
2469         }
2470
2471         return rc;
2472 }
2473
2474 static void set_io_rings_size(struct ena_adapter *adapter,
2475                               int new_tx_size,
2476                               int new_rx_size)
2477 {
2478         int i;
2479
2480         for (i = 0; i < adapter->num_io_queues; i++) {
2481                 adapter->tx_ring[i].ring_size = new_tx_size;
2482                 adapter->rx_ring[i].ring_size = new_rx_size;
2483         }
2484 }
2485
2486 /* This function allows queue allocation to backoff when the system is
2487  * low on memory. If there is not enough memory to allocate io queues
2488  * the driver will try to allocate smaller queues.
2489  *
2490  * The backoff algorithm is as follows:
2491  *  1. Try to allocate TX and RX and if successful.
2492  *  1.1. return success
2493  *
2494  *  2. Divide by 2 the size of the larger of RX and TX queues (or both if their size is the same).
2495  *
2496  *  3. If TX or RX is smaller than 256
2497  *  3.1. return failure.
2498  *  4. else
2499  *  4.1. go back to 1.
2500  */
2501 static int create_queues_with_size_backoff(struct ena_adapter *adapter)
2502 {
2503         int rc, cur_rx_ring_size, cur_tx_ring_size;
2504         int new_rx_ring_size, new_tx_ring_size;
2505
2506         /* current queue sizes might be set to smaller than the requested
2507          * ones due to past queue allocation failures.
2508          */
2509         set_io_rings_size(adapter, adapter->requested_tx_ring_size,
2510                           adapter->requested_rx_ring_size);
2511
2512         while (1) {
2513                 if (ena_xdp_present(adapter)) {
2514                         rc = ena_setup_and_create_all_xdp_queues(adapter);
2515
2516                         if (rc)
2517                                 goto err_setup_tx;
2518                 }
2519                 rc = ena_setup_tx_resources_in_range(adapter,
2520                                                      0,
2521                                                      adapter->num_io_queues);
2522                 if (rc)
2523                         goto err_setup_tx;
2524
2525                 rc = ena_create_io_tx_queues_in_range(adapter,
2526                                                       0,
2527                                                       adapter->num_io_queues);
2528                 if (rc)
2529                         goto err_create_tx_queues;
2530
2531                 rc = ena_setup_all_rx_resources(adapter);
2532                 if (rc)
2533                         goto err_setup_rx;
2534
2535                 rc = ena_create_all_io_rx_queues(adapter);
2536                 if (rc)
2537                         goto err_create_rx_queues;
2538
2539                 return 0;
2540
2541 err_create_rx_queues:
2542                 ena_free_all_io_rx_resources(adapter);
2543 err_setup_rx:
2544                 ena_destroy_all_tx_queues(adapter);
2545 err_create_tx_queues:
2546                 ena_free_all_io_tx_resources(adapter);
2547 err_setup_tx:
2548                 if (rc != -ENOMEM) {
2549                         netif_err(adapter, ifup, adapter->netdev,
2550                                   "Queue creation failed with error code %d\n",
2551                                   rc);
2552                         return rc;
2553                 }
2554
2555                 cur_tx_ring_size = adapter->tx_ring[0].ring_size;
2556                 cur_rx_ring_size = adapter->rx_ring[0].ring_size;
2557
2558                 netif_err(adapter, ifup, adapter->netdev,
2559                           "Not enough memory to create queues with sizes TX=%d, RX=%d\n",
2560                           cur_tx_ring_size, cur_rx_ring_size);
2561
2562                 new_tx_ring_size = cur_tx_ring_size;
2563                 new_rx_ring_size = cur_rx_ring_size;
2564
2565                 /* Decrease the size of the larger queue, or
2566                  * decrease both if they are the same size.
2567                  */
2568                 if (cur_rx_ring_size <= cur_tx_ring_size)
2569                         new_tx_ring_size = cur_tx_ring_size / 2;
2570                 if (cur_rx_ring_size >= cur_tx_ring_size)
2571                         new_rx_ring_size = cur_rx_ring_size / 2;
2572
2573                 if (new_tx_ring_size < ENA_MIN_RING_SIZE ||
2574                     new_rx_ring_size < ENA_MIN_RING_SIZE) {
2575                         netif_err(adapter, ifup, adapter->netdev,
2576                                   "Queue creation failed with the smallest possible queue size of %d for both queues. Not retrying with smaller queues\n",
2577                                   ENA_MIN_RING_SIZE);
2578                         return rc;
2579                 }
2580
2581                 netif_err(adapter, ifup, adapter->netdev,
2582                           "Retrying queue creation with sizes TX=%d, RX=%d\n",
2583                           new_tx_ring_size,
2584                           new_rx_ring_size);
2585
2586                 set_io_rings_size(adapter, new_tx_ring_size,
2587                                   new_rx_ring_size);
2588         }
2589 }
2590
2591 static int ena_up(struct ena_adapter *adapter)
2592 {
2593         int io_queue_count, rc, i;
2594
2595         netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
2596
2597         io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2598         ena_setup_io_intr(adapter);
2599
2600         /* napi poll functions should be initialized before running
2601          * request_irq(), to handle a rare condition where there is a pending
2602          * interrupt, causing the ISR to fire immediately while the poll
2603          * function wasn't set yet, causing a null dereference
2604          */
2605         ena_init_napi_in_range(adapter, 0, io_queue_count);
2606
2607         rc = ena_request_io_irq(adapter);
2608         if (rc)
2609                 goto err_req_irq;
2610
2611         rc = create_queues_with_size_backoff(adapter);
2612         if (rc)
2613                 goto err_create_queues_with_backoff;
2614
2615         rc = ena_up_complete(adapter);
2616         if (rc)
2617                 goto err_up;
2618
2619         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
2620                 netif_carrier_on(adapter->netdev);
2621
2622         ena_increase_stat(&adapter->dev_stats.interface_up, 1,
2623                           &adapter->syncp);
2624
2625         set_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2626
2627         /* Enable completion queues interrupt */
2628         for (i = 0; i < adapter->num_io_queues; i++)
2629                 ena_unmask_interrupt(&adapter->tx_ring[i],
2630                                      &adapter->rx_ring[i]);
2631
2632         /* schedule napi in case we had pending packets
2633          * from the last time we disable napi
2634          */
2635         for (i = 0; i < io_queue_count; i++)
2636                 napi_schedule(&adapter->ena_napi[i].napi);
2637
2638         return rc;
2639
2640 err_up:
2641         ena_destroy_all_tx_queues(adapter);
2642         ena_free_all_io_tx_resources(adapter);
2643         ena_destroy_all_rx_queues(adapter);
2644         ena_free_all_io_rx_resources(adapter);
2645 err_create_queues_with_backoff:
2646         ena_free_io_irq(adapter);
2647 err_req_irq:
2648         ena_del_napi_in_range(adapter, 0, io_queue_count);
2649
2650         return rc;
2651 }
2652
2653 static void ena_down(struct ena_adapter *adapter)
2654 {
2655         int io_queue_count = adapter->num_io_queues + adapter->xdp_num_queues;
2656
2657         netif_info(adapter, ifdown, adapter->netdev, "%s\n", __func__);
2658
2659         clear_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2660
2661         ena_increase_stat(&adapter->dev_stats.interface_down, 1,
2662                           &adapter->syncp);
2663
2664         netif_carrier_off(adapter->netdev);
2665         netif_tx_disable(adapter->netdev);
2666
2667         /* After this point the napi handler won't enable the tx queue */
2668         ena_napi_disable_in_range(adapter, 0, io_queue_count);
2669
2670         /* After destroy the queue there won't be any new interrupts */
2671
2672         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags)) {
2673                 int rc;
2674
2675                 rc = ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
2676                 if (rc)
2677                         netif_err(adapter, ifdown, adapter->netdev,
2678                                   "Device reset failed\n");
2679                 /* stop submitting admin commands on a device that was reset */
2680                 ena_com_set_admin_running_state(adapter->ena_dev, false);
2681         }
2682
2683         ena_destroy_all_io_queues(adapter);
2684
2685         ena_disable_io_intr_sync(adapter);
2686         ena_free_io_irq(adapter);
2687         ena_del_napi_in_range(adapter, 0, io_queue_count);
2688
2689         ena_free_all_tx_bufs(adapter);
2690         ena_free_all_rx_bufs(adapter);
2691         ena_free_all_io_tx_resources(adapter);
2692         ena_free_all_io_rx_resources(adapter);
2693 }
2694
2695 /* ena_open - Called when a network interface is made active
2696  * @netdev: network interface device structure
2697  *
2698  * Returns 0 on success, negative value on failure
2699  *
2700  * The open entry point is called when a network interface is made
2701  * active by the system (IFF_UP).  At this point all resources needed
2702  * for transmit and receive operations are allocated, the interrupt
2703  * handler is registered with the OS, the watchdog timer is started,
2704  * and the stack is notified that the interface is ready.
2705  */
2706 static int ena_open(struct net_device *netdev)
2707 {
2708         struct ena_adapter *adapter = netdev_priv(netdev);
2709         int rc;
2710
2711         /* Notify the stack of the actual queue counts. */
2712         rc = netif_set_real_num_tx_queues(netdev, adapter->num_io_queues);
2713         if (rc) {
2714                 netif_err(adapter, ifup, netdev, "Can't set num tx queues\n");
2715                 return rc;
2716         }
2717
2718         rc = netif_set_real_num_rx_queues(netdev, adapter->num_io_queues);
2719         if (rc) {
2720                 netif_err(adapter, ifup, netdev, "Can't set num rx queues\n");
2721                 return rc;
2722         }
2723
2724         rc = ena_up(adapter);
2725         if (rc)
2726                 return rc;
2727
2728         return rc;
2729 }
2730
2731 /* ena_close - Disables a network interface
2732  * @netdev: network interface device structure
2733  *
2734  * Returns 0, this is not allowed to fail
2735  *
2736  * The close entry point is called when an interface is de-activated
2737  * by the OS.  The hardware is still under the drivers control, but
2738  * needs to be disabled.  A global MAC reset is issued to stop the
2739  * hardware, and all transmit and receive resources are freed.
2740  */
2741 static int ena_close(struct net_device *netdev)
2742 {
2743         struct ena_adapter *adapter = netdev_priv(netdev);
2744
2745         netif_dbg(adapter, ifdown, netdev, "%s\n", __func__);
2746
2747         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
2748                 return 0;
2749
2750         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
2751                 ena_down(adapter);
2752
2753         /* Check for device status and issue reset if needed*/
2754         check_for_admin_com_state(adapter);
2755         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
2756                 netif_err(adapter, ifdown, adapter->netdev,
2757                           "Destroy failure, restarting device\n");
2758                 ena_dump_stats_to_dmesg(adapter);
2759                 /* rtnl lock already obtained in dev_ioctl() layer */
2760                 ena_destroy_device(adapter, false);
2761                 ena_restore_device(adapter);
2762         }
2763
2764         return 0;
2765 }
2766
2767 int ena_update_queue_sizes(struct ena_adapter *adapter,
2768                            u32 new_tx_size,
2769                            u32 new_rx_size)
2770 {
2771         bool dev_was_up;
2772
2773         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2774         ena_close(adapter->netdev);
2775         adapter->requested_tx_ring_size = new_tx_size;
2776         adapter->requested_rx_ring_size = new_rx_size;
2777         ena_init_io_rings(adapter,
2778                           0,
2779                           adapter->xdp_num_queues +
2780                           adapter->num_io_queues);
2781         return dev_was_up ? ena_up(adapter) : 0;
2782 }
2783
2784 int ena_update_queue_count(struct ena_adapter *adapter, u32 new_channel_count)
2785 {
2786         struct ena_com_dev *ena_dev = adapter->ena_dev;
2787         int prev_channel_count;
2788         bool dev_was_up;
2789
2790         dev_was_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
2791         ena_close(adapter->netdev);
2792         prev_channel_count = adapter->num_io_queues;
2793         adapter->num_io_queues = new_channel_count;
2794         if (ena_xdp_present(adapter) &&
2795             ena_xdp_allowed(adapter) == ENA_XDP_ALLOWED) {
2796                 adapter->xdp_first_ring = new_channel_count;
2797                 adapter->xdp_num_queues = new_channel_count;
2798                 if (prev_channel_count > new_channel_count)
2799                         ena_xdp_exchange_program_rx_in_range(adapter,
2800                                                              NULL,
2801                                                              new_channel_count,
2802                                                              prev_channel_count);
2803                 else
2804                         ena_xdp_exchange_program_rx_in_range(adapter,
2805                                                              adapter->xdp_bpf_prog,
2806                                                              prev_channel_count,
2807                                                              new_channel_count);
2808         }
2809
2810         /* We need to destroy the rss table so that the indirection
2811          * table will be reinitialized by ena_up()
2812          */
2813         ena_com_rss_destroy(ena_dev);
2814         ena_init_io_rings(adapter,
2815                           0,
2816                           adapter->xdp_num_queues +
2817                           adapter->num_io_queues);
2818         return dev_was_up ? ena_open(adapter->netdev) : 0;
2819 }
2820
2821 static void ena_tx_csum(struct ena_com_tx_ctx *ena_tx_ctx,
2822                         struct sk_buff *skb,
2823                         bool disable_meta_caching)
2824 {
2825         u32 mss = skb_shinfo(skb)->gso_size;
2826         struct ena_com_tx_meta *ena_meta = &ena_tx_ctx->ena_meta;
2827         u8 l4_protocol = 0;
2828
2829         if ((skb->ip_summed == CHECKSUM_PARTIAL) || mss) {
2830                 ena_tx_ctx->l4_csum_enable = 1;
2831                 if (mss) {
2832                         ena_tx_ctx->tso_enable = 1;
2833                         ena_meta->l4_hdr_len = tcp_hdr(skb)->doff;
2834                         ena_tx_ctx->l4_csum_partial = 0;
2835                 } else {
2836                         ena_tx_ctx->tso_enable = 0;
2837                         ena_meta->l4_hdr_len = 0;
2838                         ena_tx_ctx->l4_csum_partial = 1;
2839                 }
2840
2841                 switch (ip_hdr(skb)->version) {
2842                 case IPVERSION:
2843                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV4;
2844                         if (ip_hdr(skb)->frag_off & htons(IP_DF))
2845                                 ena_tx_ctx->df = 1;
2846                         if (mss)
2847                                 ena_tx_ctx->l3_csum_enable = 1;
2848                         l4_protocol = ip_hdr(skb)->protocol;
2849                         break;
2850                 case 6:
2851                         ena_tx_ctx->l3_proto = ENA_ETH_IO_L3_PROTO_IPV6;
2852                         l4_protocol = ipv6_hdr(skb)->nexthdr;
2853                         break;
2854                 default:
2855                         break;
2856                 }
2857
2858                 if (l4_protocol == IPPROTO_TCP)
2859                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_TCP;
2860                 else
2861                         ena_tx_ctx->l4_proto = ENA_ETH_IO_L4_PROTO_UDP;
2862
2863                 ena_meta->mss = mss;
2864                 ena_meta->l3_hdr_len = skb_network_header_len(skb);
2865                 ena_meta->l3_hdr_offset = skb_network_offset(skb);
2866                 ena_tx_ctx->meta_valid = 1;
2867         } else if (disable_meta_caching) {
2868                 memset(ena_meta, 0, sizeof(*ena_meta));
2869                 ena_tx_ctx->meta_valid = 1;
2870         } else {
2871                 ena_tx_ctx->meta_valid = 0;
2872         }
2873 }
2874
2875 static int ena_check_and_linearize_skb(struct ena_ring *tx_ring,
2876                                        struct sk_buff *skb)
2877 {
2878         int num_frags, header_len, rc;
2879
2880         num_frags = skb_shinfo(skb)->nr_frags;
2881         header_len = skb_headlen(skb);
2882
2883         if (num_frags < tx_ring->sgl_size)
2884                 return 0;
2885
2886         if ((num_frags == tx_ring->sgl_size) &&
2887             (header_len < tx_ring->tx_max_header_size))
2888                 return 0;
2889
2890         ena_increase_stat(&tx_ring->tx_stats.linearize, 1, &tx_ring->syncp);
2891
2892         rc = skb_linearize(skb);
2893         if (unlikely(rc)) {
2894                 ena_increase_stat(&tx_ring->tx_stats.linearize_failed, 1,
2895                                   &tx_ring->syncp);
2896         }
2897
2898         return rc;
2899 }
2900
2901 static int ena_tx_map_skb(struct ena_ring *tx_ring,
2902                           struct ena_tx_buffer *tx_info,
2903                           struct sk_buff *skb,
2904                           void **push_hdr,
2905                           u16 *header_len)
2906 {
2907         struct ena_adapter *adapter = tx_ring->adapter;
2908         struct ena_com_buf *ena_buf;
2909         dma_addr_t dma;
2910         u32 skb_head_len, frag_len, last_frag;
2911         u16 push_len = 0;
2912         u16 delta = 0;
2913         int i = 0;
2914
2915         skb_head_len = skb_headlen(skb);
2916         tx_info->skb = skb;
2917         ena_buf = tx_info->bufs;
2918
2919         if (tx_ring->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
2920                 /* When the device is LLQ mode, the driver will copy
2921                  * the header into the device memory space.
2922                  * the ena_com layer assume the header is in a linear
2923                  * memory space.
2924                  * This assumption might be wrong since part of the header
2925                  * can be in the fragmented buffers.
2926                  * Use skb_header_pointer to make sure the header is in a
2927                  * linear memory space.
2928                  */
2929
2930                 push_len = min_t(u32, skb->len, tx_ring->tx_max_header_size);
2931                 *push_hdr = skb_header_pointer(skb, 0, push_len,
2932                                                tx_ring->push_buf_intermediate_buf);
2933                 *header_len = push_len;
2934                 if (unlikely(skb->data != *push_hdr)) {
2935                         ena_increase_stat(&tx_ring->tx_stats.llq_buffer_copy, 1,
2936                                           &tx_ring->syncp);
2937
2938                         delta = push_len - skb_head_len;
2939                 }
2940         } else {
2941                 *push_hdr = NULL;
2942                 *header_len = min_t(u32, skb_head_len,
2943                                     tx_ring->tx_max_header_size);
2944         }
2945
2946         netif_dbg(adapter, tx_queued, adapter->netdev,
2947                   "skb: %p header_buf->vaddr: %p push_len: %d\n", skb,
2948                   *push_hdr, push_len);
2949
2950         if (skb_head_len > push_len) {
2951                 dma = dma_map_single(tx_ring->dev, skb->data + push_len,
2952                                      skb_head_len - push_len, DMA_TO_DEVICE);
2953                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2954                         goto error_report_dma_error;
2955
2956                 ena_buf->paddr = dma;
2957                 ena_buf->len = skb_head_len - push_len;
2958
2959                 ena_buf++;
2960                 tx_info->num_of_bufs++;
2961                 tx_info->map_linear_data = 1;
2962         } else {
2963                 tx_info->map_linear_data = 0;
2964         }
2965
2966         last_frag = skb_shinfo(skb)->nr_frags;
2967
2968         for (i = 0; i < last_frag; i++) {
2969                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2970
2971                 frag_len = skb_frag_size(frag);
2972
2973                 if (unlikely(delta >= frag_len)) {
2974                         delta -= frag_len;
2975                         continue;
2976                 }
2977
2978                 dma = skb_frag_dma_map(tx_ring->dev, frag, delta,
2979                                        frag_len - delta, DMA_TO_DEVICE);
2980                 if (unlikely(dma_mapping_error(tx_ring->dev, dma)))
2981                         goto error_report_dma_error;
2982
2983                 ena_buf->paddr = dma;
2984                 ena_buf->len = frag_len - delta;
2985                 ena_buf++;
2986                 tx_info->num_of_bufs++;
2987                 delta = 0;
2988         }
2989
2990         return 0;
2991
2992 error_report_dma_error:
2993         ena_increase_stat(&tx_ring->tx_stats.dma_mapping_err, 1,
2994                           &tx_ring->syncp);
2995         netif_warn(adapter, tx_queued, adapter->netdev, "Failed to map skb\n");
2996
2997         tx_info->skb = NULL;
2998
2999         tx_info->num_of_bufs += i;
3000         ena_unmap_tx_buff(tx_ring, tx_info);
3001
3002         return -EINVAL;
3003 }
3004
3005 /* Called with netif_tx_lock. */
3006 static netdev_tx_t ena_start_xmit(struct sk_buff *skb, struct net_device *dev)
3007 {
3008         struct ena_adapter *adapter = netdev_priv(dev);
3009         struct ena_tx_buffer *tx_info;
3010         struct ena_com_tx_ctx ena_tx_ctx;
3011         struct ena_ring *tx_ring;
3012         struct netdev_queue *txq;
3013         void *push_hdr;
3014         u16 next_to_use, req_id, header_len;
3015         int qid, rc;
3016
3017         netif_dbg(adapter, tx_queued, dev, "%s skb %p\n", __func__, skb);
3018         /*  Determine which tx ring we will be placed on */
3019         qid = skb_get_queue_mapping(skb);
3020         tx_ring = &adapter->tx_ring[qid];
3021         txq = netdev_get_tx_queue(dev, qid);
3022
3023         rc = ena_check_and_linearize_skb(tx_ring, skb);
3024         if (unlikely(rc))
3025                 goto error_drop_packet;
3026
3027         skb_tx_timestamp(skb);
3028
3029         next_to_use = tx_ring->next_to_use;
3030         req_id = tx_ring->free_ids[next_to_use];
3031         tx_info = &tx_ring->tx_buffer_info[req_id];
3032         tx_info->num_of_bufs = 0;
3033
3034         WARN(tx_info->skb, "SKB isn't NULL req_id %d\n", req_id);
3035
3036         rc = ena_tx_map_skb(tx_ring, tx_info, skb, &push_hdr, &header_len);
3037         if (unlikely(rc))
3038                 goto error_drop_packet;
3039
3040         memset(&ena_tx_ctx, 0x0, sizeof(struct ena_com_tx_ctx));
3041         ena_tx_ctx.ena_bufs = tx_info->bufs;
3042         ena_tx_ctx.push_header = push_hdr;
3043         ena_tx_ctx.num_bufs = tx_info->num_of_bufs;
3044         ena_tx_ctx.req_id = req_id;
3045         ena_tx_ctx.header_len = header_len;
3046
3047         /* set flags and meta data */
3048         ena_tx_csum(&ena_tx_ctx, skb, tx_ring->disable_meta_caching);
3049
3050         rc = ena_xmit_common(dev,
3051                              tx_ring,
3052                              tx_info,
3053                              &ena_tx_ctx,
3054                              next_to_use,
3055                              skb->len);
3056         if (rc)
3057                 goto error_unmap_dma;
3058
3059         netdev_tx_sent_queue(txq, skb->len);
3060
3061         /* stop the queue when no more space available, the packet can have up
3062          * to sgl_size + 2. one for the meta descriptor and one for header
3063          * (if the header is larger than tx_max_header_size).
3064          */
3065         if (unlikely(!ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3066                                                    tx_ring->sgl_size + 2))) {
3067                 netif_dbg(adapter, tx_queued, dev, "%s stop queue %d\n",
3068                           __func__, qid);
3069
3070                 netif_tx_stop_queue(txq);
3071                 ena_increase_stat(&tx_ring->tx_stats.queue_stop, 1,
3072                                   &tx_ring->syncp);
3073
3074                 /* There is a rare condition where this function decide to
3075                  * stop the queue but meanwhile clean_tx_irq updates
3076                  * next_to_completion and terminates.
3077                  * The queue will remain stopped forever.
3078                  * To solve this issue add a mb() to make sure that
3079                  * netif_tx_stop_queue() write is vissible before checking if
3080                  * there is additional space in the queue.
3081                  */
3082                 smp_mb();
3083
3084                 if (ena_com_sq_have_enough_space(tx_ring->ena_com_io_sq,
3085                                                  ENA_TX_WAKEUP_THRESH)) {
3086                         netif_tx_wake_queue(txq);
3087                         ena_increase_stat(&tx_ring->tx_stats.queue_wakeup, 1,
3088                                           &tx_ring->syncp);
3089                 }
3090         }
3091
3092         if (netif_xmit_stopped(txq) || !netdev_xmit_more()) {
3093                 /* trigger the dma engine. ena_com_write_sq_doorbell()
3094                  * has a mb
3095                  */
3096                 ena_com_write_sq_doorbell(tx_ring->ena_com_io_sq);
3097                 ena_increase_stat(&tx_ring->tx_stats.doorbells, 1,
3098                                   &tx_ring->syncp);
3099         }
3100
3101         return NETDEV_TX_OK;
3102
3103 error_unmap_dma:
3104         ena_unmap_tx_buff(tx_ring, tx_info);
3105         tx_info->skb = NULL;
3106
3107 error_drop_packet:
3108         dev_kfree_skb(skb);
3109         return NETDEV_TX_OK;
3110 }
3111
3112 static u16 ena_select_queue(struct net_device *dev, struct sk_buff *skb,
3113                             struct net_device *sb_dev)
3114 {
3115         u16 qid;
3116         /* we suspect that this is good for in--kernel network services that
3117          * want to loop incoming skb rx to tx in normal user generated traffic,
3118          * most probably we will not get to this
3119          */
3120         if (skb_rx_queue_recorded(skb))
3121                 qid = skb_get_rx_queue(skb);
3122         else
3123                 qid = netdev_pick_tx(dev, skb, NULL);
3124
3125         return qid;
3126 }
3127
3128 static void ena_config_host_info(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
3129 {
3130         struct device *dev = &pdev->dev;
3131         struct ena_admin_host_info *host_info;
3132         int rc;
3133
3134         /* Allocate only the host info */
3135         rc = ena_com_allocate_host_info(ena_dev);
3136         if (rc) {
3137                 dev_err(dev, "Cannot allocate host info\n");
3138                 return;
3139         }
3140
3141         host_info = ena_dev->host_attr.host_info;
3142
3143         host_info->bdf = (pdev->bus->number << 8) | pdev->devfn;
3144         host_info->os_type = ENA_ADMIN_OS_LINUX;
3145         host_info->kernel_ver = LINUX_VERSION_CODE;
3146         strlcpy(host_info->kernel_ver_str, utsname()->version,
3147                 sizeof(host_info->kernel_ver_str) - 1);
3148         host_info->os_dist = 0;
3149         strncpy(host_info->os_dist_str, utsname()->release,
3150                 sizeof(host_info->os_dist_str) - 1);
3151         host_info->driver_version =
3152                 (DRV_MODULE_GEN_MAJOR) |
3153                 (DRV_MODULE_GEN_MINOR << ENA_ADMIN_HOST_INFO_MINOR_SHIFT) |
3154                 (DRV_MODULE_GEN_SUBMINOR << ENA_ADMIN_HOST_INFO_SUB_MINOR_SHIFT) |
3155                 ("K"[0] << ENA_ADMIN_HOST_INFO_MODULE_TYPE_SHIFT);
3156         host_info->num_cpus = num_online_cpus();
3157
3158         host_info->driver_supported_features =
3159                 ENA_ADMIN_HOST_INFO_RX_OFFSET_MASK |
3160                 ENA_ADMIN_HOST_INFO_INTERRUPT_MODERATION_MASK |
3161                 ENA_ADMIN_HOST_INFO_RX_BUF_MIRRORING_MASK |
3162                 ENA_ADMIN_HOST_INFO_RSS_CONFIGURABLE_FUNCTION_KEY_MASK;
3163
3164         rc = ena_com_set_host_attributes(ena_dev);
3165         if (rc) {
3166                 if (rc == -EOPNOTSUPP)
3167                         dev_warn(dev, "Cannot set host attributes\n");
3168                 else
3169                         dev_err(dev, "Cannot set host attributes\n");
3170
3171                 goto err;
3172         }
3173
3174         return;
3175
3176 err:
3177         ena_com_delete_host_info(ena_dev);
3178 }
3179
3180 static void ena_config_debug_area(struct ena_adapter *adapter)
3181 {
3182         u32 debug_area_size;
3183         int rc, ss_count;
3184
3185         ss_count = ena_get_sset_count(adapter->netdev, ETH_SS_STATS);
3186         if (ss_count <= 0) {
3187                 netif_err(adapter, drv, adapter->netdev,
3188                           "SS count is negative\n");
3189                 return;
3190         }
3191
3192         /* allocate 32 bytes for each string and 64bit for the value */
3193         debug_area_size = ss_count * ETH_GSTRING_LEN + sizeof(u64) * ss_count;
3194
3195         rc = ena_com_allocate_debug_area(adapter->ena_dev, debug_area_size);
3196         if (rc) {
3197                 netif_err(adapter, drv, adapter->netdev,
3198                           "Cannot allocate debug area\n");
3199                 return;
3200         }
3201
3202         rc = ena_com_set_host_attributes(adapter->ena_dev);
3203         if (rc) {
3204                 if (rc == -EOPNOTSUPP)
3205                         netif_warn(adapter, drv, adapter->netdev,
3206                                    "Cannot set host attributes\n");
3207                 else
3208                         netif_err(adapter, drv, adapter->netdev,
3209                                   "Cannot set host attributes\n");
3210                 goto err;
3211         }
3212
3213         return;
3214 err:
3215         ena_com_delete_debug_area(adapter->ena_dev);
3216 }
3217
3218 int ena_update_hw_stats(struct ena_adapter *adapter)
3219 {
3220         int rc = 0;
3221
3222         rc = ena_com_get_eni_stats(adapter->ena_dev, &adapter->eni_stats);
3223         if (rc) {
3224                 dev_info_once(&adapter->pdev->dev, "Failed to get ENI stats\n");
3225                 return rc;
3226         }
3227
3228         return 0;
3229 }
3230
3231 static void ena_get_stats64(struct net_device *netdev,
3232                             struct rtnl_link_stats64 *stats)
3233 {
3234         struct ena_adapter *adapter = netdev_priv(netdev);
3235         struct ena_ring *rx_ring, *tx_ring;
3236         unsigned int start;
3237         u64 rx_drops;
3238         u64 tx_drops;
3239         int i;
3240
3241         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3242                 return;
3243
3244         for (i = 0; i < adapter->num_io_queues; i++) {
3245                 u64 bytes, packets;
3246
3247                 tx_ring = &adapter->tx_ring[i];
3248
3249                 do {
3250                         start = u64_stats_fetch_begin_irq(&tx_ring->syncp);
3251                         packets = tx_ring->tx_stats.cnt;
3252                         bytes = tx_ring->tx_stats.bytes;
3253                 } while (u64_stats_fetch_retry_irq(&tx_ring->syncp, start));
3254
3255                 stats->tx_packets += packets;
3256                 stats->tx_bytes += bytes;
3257
3258                 rx_ring = &adapter->rx_ring[i];
3259
3260                 do {
3261                         start = u64_stats_fetch_begin_irq(&rx_ring->syncp);
3262                         packets = rx_ring->rx_stats.cnt;
3263                         bytes = rx_ring->rx_stats.bytes;
3264                 } while (u64_stats_fetch_retry_irq(&rx_ring->syncp, start));
3265
3266                 stats->rx_packets += packets;
3267                 stats->rx_bytes += bytes;
3268         }
3269
3270         do {
3271                 start = u64_stats_fetch_begin_irq(&adapter->syncp);
3272                 rx_drops = adapter->dev_stats.rx_drops;
3273                 tx_drops = adapter->dev_stats.tx_drops;
3274         } while (u64_stats_fetch_retry_irq(&adapter->syncp, start));
3275
3276         stats->rx_dropped = rx_drops;
3277         stats->tx_dropped = tx_drops;
3278
3279         stats->multicast = 0;
3280         stats->collisions = 0;
3281
3282         stats->rx_length_errors = 0;
3283         stats->rx_crc_errors = 0;
3284         stats->rx_frame_errors = 0;
3285         stats->rx_fifo_errors = 0;
3286         stats->rx_missed_errors = 0;
3287         stats->tx_window_errors = 0;
3288
3289         stats->rx_errors = 0;
3290         stats->tx_errors = 0;
3291 }
3292
3293 static const struct net_device_ops ena_netdev_ops = {
3294         .ndo_open               = ena_open,
3295         .ndo_stop               = ena_close,
3296         .ndo_start_xmit         = ena_start_xmit,
3297         .ndo_select_queue       = ena_select_queue,
3298         .ndo_get_stats64        = ena_get_stats64,
3299         .ndo_tx_timeout         = ena_tx_timeout,
3300         .ndo_change_mtu         = ena_change_mtu,
3301         .ndo_set_mac_address    = NULL,
3302         .ndo_validate_addr      = eth_validate_addr,
3303         .ndo_bpf                = ena_xdp,
3304         .ndo_xdp_xmit           = ena_xdp_xmit,
3305 };
3306
3307 static int ena_device_validate_params(struct ena_adapter *adapter,
3308                                       struct ena_com_dev_get_features_ctx *get_feat_ctx)
3309 {
3310         struct net_device *netdev = adapter->netdev;
3311         int rc;
3312
3313         rc = ether_addr_equal(get_feat_ctx->dev_attr.mac_addr,
3314                               adapter->mac_addr);
3315         if (!rc) {
3316                 netif_err(adapter, drv, netdev,
3317                           "Error, mac address are different\n");
3318                 return -EINVAL;
3319         }
3320
3321         if (get_feat_ctx->dev_attr.max_mtu < netdev->mtu) {
3322                 netif_err(adapter, drv, netdev,
3323                           "Error, device max mtu is smaller than netdev MTU\n");
3324                 return -EINVAL;
3325         }
3326
3327         return 0;
3328 }
3329
3330 static void set_default_llq_configurations(struct ena_llq_configurations *llq_config)
3331 {
3332         llq_config->llq_header_location = ENA_ADMIN_INLINE_HEADER;
3333         llq_config->llq_stride_ctrl = ENA_ADMIN_MULTIPLE_DESCS_PER_ENTRY;
3334         llq_config->llq_num_decs_before_header = ENA_ADMIN_LLQ_NUM_DESCS_BEFORE_HEADER_2;
3335         llq_config->llq_ring_entry_size = ENA_ADMIN_LIST_ENTRY_SIZE_128B;
3336         llq_config->llq_ring_entry_size_value = 128;
3337 }
3338
3339 static int ena_set_queues_placement_policy(struct pci_dev *pdev,
3340                                            struct ena_com_dev *ena_dev,
3341                                            struct ena_admin_feature_llq_desc *llq,
3342                                            struct ena_llq_configurations *llq_default_configurations)
3343 {
3344         int rc;
3345         u32 llq_feature_mask;
3346
3347         llq_feature_mask = 1 << ENA_ADMIN_LLQ;
3348         if (!(ena_dev->supported_features & llq_feature_mask)) {
3349                 dev_err(&pdev->dev,
3350                         "LLQ is not supported Fallback to host mode policy.\n");
3351                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3352                 return 0;
3353         }
3354
3355         rc = ena_com_config_dev_mode(ena_dev, llq, llq_default_configurations);
3356         if (unlikely(rc)) {
3357                 dev_err(&pdev->dev,
3358                         "Failed to configure the device mode.  Fallback to host mode policy.\n");
3359                 ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3360         }
3361
3362         return 0;
3363 }
3364
3365 static int ena_map_llq_mem_bar(struct pci_dev *pdev, struct ena_com_dev *ena_dev,
3366                                int bars)
3367 {
3368         bool has_mem_bar = !!(bars & BIT(ENA_MEM_BAR));
3369
3370         if (!has_mem_bar) {
3371                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV) {
3372                         dev_err(&pdev->dev,
3373                                 "ENA device does not expose LLQ bar. Fallback to host mode policy.\n");
3374                         ena_dev->tx_mem_queue_type = ENA_ADMIN_PLACEMENT_POLICY_HOST;
3375                 }
3376
3377                 return 0;
3378         }
3379
3380         ena_dev->mem_bar = devm_ioremap_wc(&pdev->dev,
3381                                            pci_resource_start(pdev, ENA_MEM_BAR),
3382                                            pci_resource_len(pdev, ENA_MEM_BAR));
3383
3384         if (!ena_dev->mem_bar)
3385                 return -EFAULT;
3386
3387         return 0;
3388 }
3389
3390 static int ena_device_init(struct ena_com_dev *ena_dev, struct pci_dev *pdev,
3391                            struct ena_com_dev_get_features_ctx *get_feat_ctx,
3392                            bool *wd_state)
3393 {
3394         struct ena_llq_configurations llq_config;
3395         struct device *dev = &pdev->dev;
3396         bool readless_supported;
3397         u32 aenq_groups;
3398         int dma_width;
3399         int rc;
3400
3401         rc = ena_com_mmio_reg_read_request_init(ena_dev);
3402         if (rc) {
3403                 dev_err(dev, "Failed to init mmio read less\n");
3404                 return rc;
3405         }
3406
3407         /* The PCIe configuration space revision id indicate if mmio reg
3408          * read is disabled
3409          */
3410         readless_supported = !(pdev->revision & ENA_MMIO_DISABLE_REG_READ);
3411         ena_com_set_mmio_read_mode(ena_dev, readless_supported);
3412
3413         rc = ena_com_dev_reset(ena_dev, ENA_REGS_RESET_NORMAL);
3414         if (rc) {
3415                 dev_err(dev, "Can not reset device\n");
3416                 goto err_mmio_read_less;
3417         }
3418
3419         rc = ena_com_validate_version(ena_dev);
3420         if (rc) {
3421                 dev_err(dev, "Device version is too low\n");
3422                 goto err_mmio_read_less;
3423         }
3424
3425         dma_width = ena_com_get_dma_width(ena_dev);
3426         if (dma_width < 0) {
3427                 dev_err(dev, "Invalid dma width value %d", dma_width);
3428                 rc = dma_width;
3429                 goto err_mmio_read_less;
3430         }
3431
3432         rc = dma_set_mask_and_coherent(dev, DMA_BIT_MASK(dma_width));
3433         if (rc) {
3434                 dev_err(dev, "dma_set_mask_and_coherent failed %d\n", rc);
3435                 goto err_mmio_read_less;
3436         }
3437
3438         /* ENA admin level init */
3439         rc = ena_com_admin_init(ena_dev, &aenq_handlers);
3440         if (rc) {
3441                 dev_err(dev,
3442                         "Can not initialize ena admin queue with device\n");
3443                 goto err_mmio_read_less;
3444         }
3445
3446         /* To enable the msix interrupts the driver needs to know the number
3447          * of queues. So the driver uses polling mode to retrieve this
3448          * information
3449          */
3450         ena_com_set_admin_polling_mode(ena_dev, true);
3451
3452         ena_config_host_info(ena_dev, pdev);
3453
3454         /* Get Device Attributes*/
3455         rc = ena_com_get_dev_attr_feat(ena_dev, get_feat_ctx);
3456         if (rc) {
3457                 dev_err(dev, "Cannot get attribute for ena device rc=%d\n", rc);
3458                 goto err_admin_init;
3459         }
3460
3461         /* Try to turn all the available aenq groups */
3462         aenq_groups = BIT(ENA_ADMIN_LINK_CHANGE) |
3463                 BIT(ENA_ADMIN_FATAL_ERROR) |
3464                 BIT(ENA_ADMIN_WARNING) |
3465                 BIT(ENA_ADMIN_NOTIFICATION) |
3466                 BIT(ENA_ADMIN_KEEP_ALIVE);
3467
3468         aenq_groups &= get_feat_ctx->aenq.supported_groups;
3469
3470         rc = ena_com_set_aenq_config(ena_dev, aenq_groups);
3471         if (rc) {
3472                 dev_err(dev, "Cannot configure aenq groups rc= %d\n", rc);
3473                 goto err_admin_init;
3474         }
3475
3476         *wd_state = !!(aenq_groups & BIT(ENA_ADMIN_KEEP_ALIVE));
3477
3478         set_default_llq_configurations(&llq_config);
3479
3480         rc = ena_set_queues_placement_policy(pdev, ena_dev, &get_feat_ctx->llq,
3481                                              &llq_config);
3482         if (rc) {
3483                 dev_err(dev, "ENA device init failed\n");
3484                 goto err_admin_init;
3485         }
3486
3487         return 0;
3488
3489 err_admin_init:
3490         ena_com_delete_host_info(ena_dev);
3491         ena_com_admin_destroy(ena_dev);
3492 err_mmio_read_less:
3493         ena_com_mmio_reg_read_request_destroy(ena_dev);
3494
3495         return rc;
3496 }
3497
3498 static int ena_enable_msix_and_set_admin_interrupts(struct ena_adapter *adapter)
3499 {
3500         struct ena_com_dev *ena_dev = adapter->ena_dev;
3501         struct device *dev = &adapter->pdev->dev;
3502         int rc;
3503
3504         rc = ena_enable_msix(adapter);
3505         if (rc) {
3506                 dev_err(dev, "Can not reserve msix vectors\n");
3507                 return rc;
3508         }
3509
3510         ena_setup_mgmnt_intr(adapter);
3511
3512         rc = ena_request_mgmnt_irq(adapter);
3513         if (rc) {
3514                 dev_err(dev, "Can not setup management interrupts\n");
3515                 goto err_disable_msix;
3516         }
3517
3518         ena_com_set_admin_polling_mode(ena_dev, false);
3519
3520         ena_com_admin_aenq_enable(ena_dev);
3521
3522         return 0;
3523
3524 err_disable_msix:
3525         ena_disable_msix(adapter);
3526
3527         return rc;
3528 }
3529
3530 static void ena_destroy_device(struct ena_adapter *adapter, bool graceful)
3531 {
3532         struct net_device *netdev = adapter->netdev;
3533         struct ena_com_dev *ena_dev = adapter->ena_dev;
3534         bool dev_up;
3535
3536         if (!test_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags))
3537                 return;
3538
3539         netif_carrier_off(netdev);
3540
3541         del_timer_sync(&adapter->timer_service);
3542
3543         dev_up = test_bit(ENA_FLAG_DEV_UP, &adapter->flags);
3544         adapter->dev_up_before_reset = dev_up;
3545         if (!graceful)
3546                 ena_com_set_admin_running_state(ena_dev, false);
3547
3548         if (test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3549                 ena_down(adapter);
3550
3551         /* Stop the device from sending AENQ events (in case reset flag is set
3552          *  and device is up, ena_down() already reset the device.
3553          */
3554         if (!(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags) && dev_up))
3555                 ena_com_dev_reset(adapter->ena_dev, adapter->reset_reason);
3556
3557         ena_free_mgmnt_irq(adapter);
3558
3559         ena_disable_msix(adapter);
3560
3561         ena_com_abort_admin_commands(ena_dev);
3562
3563         ena_com_wait_for_abort_completion(ena_dev);
3564
3565         ena_com_admin_destroy(ena_dev);
3566
3567         ena_com_mmio_reg_read_request_destroy(ena_dev);
3568
3569         /* return reset reason to default value */
3570         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
3571
3572         clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3573         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3574 }
3575
3576 static int ena_restore_device(struct ena_adapter *adapter)
3577 {
3578         struct ena_com_dev_get_features_ctx get_feat_ctx;
3579         struct ena_com_dev *ena_dev = adapter->ena_dev;
3580         struct pci_dev *pdev = adapter->pdev;
3581         bool wd_state;
3582         int rc;
3583
3584         set_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3585         rc = ena_device_init(ena_dev, adapter->pdev, &get_feat_ctx, &wd_state);
3586         if (rc) {
3587                 dev_err(&pdev->dev, "Can not initialize device\n");
3588                 goto err;
3589         }
3590         adapter->wd_state = wd_state;
3591
3592         rc = ena_device_validate_params(adapter, &get_feat_ctx);
3593         if (rc) {
3594                 dev_err(&pdev->dev, "Validation of device parameters failed\n");
3595                 goto err_device_destroy;
3596         }
3597
3598         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
3599         if (rc) {
3600                 dev_err(&pdev->dev, "Enable MSI-X failed\n");
3601                 goto err_device_destroy;
3602         }
3603         /* If the interface was up before the reset bring it up */
3604         if (adapter->dev_up_before_reset) {
3605                 rc = ena_up(adapter);
3606                 if (rc) {
3607                         dev_err(&pdev->dev, "Failed to create I/O queues\n");
3608                         goto err_disable_msix;
3609                 }
3610         }
3611
3612         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3613
3614         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3615         if (test_bit(ENA_FLAG_LINK_UP, &adapter->flags))
3616                 netif_carrier_on(adapter->netdev);
3617
3618         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3619         adapter->last_keep_alive_jiffies = jiffies;
3620
3621         dev_err(&pdev->dev, "Device reset completed successfully\n");
3622
3623         return rc;
3624 err_disable_msix:
3625         ena_free_mgmnt_irq(adapter);
3626         ena_disable_msix(adapter);
3627 err_device_destroy:
3628         ena_com_abort_admin_commands(ena_dev);
3629         ena_com_wait_for_abort_completion(ena_dev);
3630         ena_com_admin_destroy(ena_dev);
3631         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_DRIVER_INVALID_STATE);
3632         ena_com_mmio_reg_read_request_destroy(ena_dev);
3633 err:
3634         clear_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
3635         clear_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags);
3636         dev_err(&pdev->dev,
3637                 "Reset attempt failed. Can not reset the device\n");
3638
3639         return rc;
3640 }
3641
3642 static void ena_fw_reset_device(struct work_struct *work)
3643 {
3644         struct ena_adapter *adapter =
3645                 container_of(work, struct ena_adapter, reset_task);
3646
3647         rtnl_lock();
3648
3649         if (likely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3650                 ena_destroy_device(adapter, false);
3651                 ena_restore_device(adapter);
3652         }
3653
3654         rtnl_unlock();
3655 }
3656
3657 static int check_for_rx_interrupt_queue(struct ena_adapter *adapter,
3658                                         struct ena_ring *rx_ring)
3659 {
3660         if (likely(rx_ring->first_interrupt))
3661                 return 0;
3662
3663         if (ena_com_cq_empty(rx_ring->ena_com_io_cq))
3664                 return 0;
3665
3666         rx_ring->no_interrupt_event_cnt++;
3667
3668         if (rx_ring->no_interrupt_event_cnt == ENA_MAX_NO_INTERRUPT_ITERATIONS) {
3669                 netif_err(adapter, rx_err, adapter->netdev,
3670                           "Potential MSIX issue on Rx side Queue = %d. Reset the device\n",
3671                           rx_ring->qid);
3672                 adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3673                 smp_mb__before_atomic();
3674                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3675                 return -EIO;
3676         }
3677
3678         return 0;
3679 }
3680
3681 static int check_missing_comp_in_tx_queue(struct ena_adapter *adapter,
3682                                           struct ena_ring *tx_ring)
3683 {
3684         struct ena_tx_buffer *tx_buf;
3685         unsigned long last_jiffies;
3686         u32 missed_tx = 0;
3687         int i, rc = 0;
3688
3689         for (i = 0; i < tx_ring->ring_size; i++) {
3690                 tx_buf = &tx_ring->tx_buffer_info[i];
3691                 last_jiffies = tx_buf->last_jiffies;
3692
3693                 if (last_jiffies == 0)
3694                         /* no pending Tx at this location */
3695                         continue;
3696
3697                 if (unlikely(!tx_ring->first_interrupt && time_is_before_jiffies(last_jiffies +
3698                              2 * adapter->missing_tx_completion_to))) {
3699                         /* If after graceful period interrupt is still not
3700                          * received, we schedule a reset
3701                          */
3702                         netif_err(adapter, tx_err, adapter->netdev,
3703                                   "Potential MSIX issue on Tx side Queue = %d. Reset the device\n",
3704                                   tx_ring->qid);
3705                         adapter->reset_reason = ENA_REGS_RESET_MISS_INTERRUPT;
3706                         smp_mb__before_atomic();
3707                         set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3708                         return -EIO;
3709                 }
3710
3711                 if (unlikely(time_is_before_jiffies(last_jiffies +
3712                                 adapter->missing_tx_completion_to))) {
3713                         if (!tx_buf->print_once)
3714                                 netif_notice(adapter, tx_err, adapter->netdev,
3715                                              "Found a Tx that wasn't completed on time, qid %d, index %d.\n",
3716                                              tx_ring->qid, i);
3717
3718                         tx_buf->print_once = 1;
3719                         missed_tx++;
3720                 }
3721         }
3722
3723         if (unlikely(missed_tx > adapter->missing_tx_completion_threshold)) {
3724                 netif_err(adapter, tx_err, adapter->netdev,
3725                           "The number of lost tx completions is above the threshold (%d > %d). Reset the device\n",
3726                           missed_tx,
3727                           adapter->missing_tx_completion_threshold);
3728                 adapter->reset_reason =
3729                         ENA_REGS_RESET_MISS_TX_CMPL;
3730                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3731                 rc = -EIO;
3732         }
3733
3734         ena_increase_stat(&tx_ring->tx_stats.missed_tx, missed_tx,
3735                           &tx_ring->syncp);
3736
3737         return rc;
3738 }
3739
3740 static void check_for_missing_completions(struct ena_adapter *adapter)
3741 {
3742         struct ena_ring *tx_ring;
3743         struct ena_ring *rx_ring;
3744         int i, budget, rc;
3745         int io_queue_count;
3746
3747         io_queue_count = adapter->xdp_num_queues + adapter->num_io_queues;
3748         /* Make sure the driver doesn't turn the device in other process */
3749         smp_rmb();
3750
3751         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3752                 return;
3753
3754         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3755                 return;
3756
3757         if (adapter->missing_tx_completion_to == ENA_HW_HINTS_NO_TIMEOUT)
3758                 return;
3759
3760         budget = ENA_MONITORED_TX_QUEUES;
3761
3762         for (i = adapter->last_monitored_tx_qid; i < io_queue_count; i++) {
3763                 tx_ring = &adapter->tx_ring[i];
3764                 rx_ring = &adapter->rx_ring[i];
3765
3766                 rc = check_missing_comp_in_tx_queue(adapter, tx_ring);
3767                 if (unlikely(rc))
3768                         return;
3769
3770                 rc =  !ENA_IS_XDP_INDEX(adapter, i) ?
3771                         check_for_rx_interrupt_queue(adapter, rx_ring) : 0;
3772                 if (unlikely(rc))
3773                         return;
3774
3775                 budget--;
3776                 if (!budget)
3777                         break;
3778         }
3779
3780         adapter->last_monitored_tx_qid = i % io_queue_count;
3781 }
3782
3783 /* trigger napi schedule after 2 consecutive detections */
3784 #define EMPTY_RX_REFILL 2
3785 /* For the rare case where the device runs out of Rx descriptors and the
3786  * napi handler failed to refill new Rx descriptors (due to a lack of memory
3787  * for example).
3788  * This case will lead to a deadlock:
3789  * The device won't send interrupts since all the new Rx packets will be dropped
3790  * The napi handler won't allocate new Rx descriptors so the device will be
3791  * able to send new packets.
3792  *
3793  * This scenario can happen when the kernel's vm.min_free_kbytes is too small.
3794  * It is recommended to have at least 512MB, with a minimum of 128MB for
3795  * constrained environment).
3796  *
3797  * When such a situation is detected - Reschedule napi
3798  */
3799 static void check_for_empty_rx_ring(struct ena_adapter *adapter)
3800 {
3801         struct ena_ring *rx_ring;
3802         int i, refill_required;
3803
3804         if (!test_bit(ENA_FLAG_DEV_UP, &adapter->flags))
3805                 return;
3806
3807         if (test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))
3808                 return;
3809
3810         for (i = 0; i < adapter->num_io_queues; i++) {
3811                 rx_ring = &adapter->rx_ring[i];
3812
3813                 refill_required = ena_com_free_q_entries(rx_ring->ena_com_io_sq);
3814                 if (unlikely(refill_required == (rx_ring->ring_size - 1))) {
3815                         rx_ring->empty_rx_queue++;
3816
3817                         if (rx_ring->empty_rx_queue >= EMPTY_RX_REFILL) {
3818                                 ena_increase_stat(&rx_ring->rx_stats.empty_rx_ring, 1,
3819                                                   &rx_ring->syncp);
3820
3821                                 netif_err(adapter, drv, adapter->netdev,
3822                                           "Trigger refill for ring %d\n", i);
3823
3824                                 napi_schedule(rx_ring->napi);
3825                                 rx_ring->empty_rx_queue = 0;
3826                         }
3827                 } else {
3828                         rx_ring->empty_rx_queue = 0;
3829                 }
3830         }
3831 }
3832
3833 /* Check for keep alive expiration */
3834 static void check_for_missing_keep_alive(struct ena_adapter *adapter)
3835 {
3836         unsigned long keep_alive_expired;
3837
3838         if (!adapter->wd_state)
3839                 return;
3840
3841         if (adapter->keep_alive_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3842                 return;
3843
3844         keep_alive_expired = adapter->last_keep_alive_jiffies +
3845                              adapter->keep_alive_timeout;
3846         if (unlikely(time_is_before_jiffies(keep_alive_expired))) {
3847                 netif_err(adapter, drv, adapter->netdev,
3848                           "Keep alive watchdog timeout.\n");
3849                 ena_increase_stat(&adapter->dev_stats.wd_expired, 1,
3850                                   &adapter->syncp);
3851                 adapter->reset_reason = ENA_REGS_RESET_KEEP_ALIVE_TO;
3852                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3853         }
3854 }
3855
3856 static void check_for_admin_com_state(struct ena_adapter *adapter)
3857 {
3858         if (unlikely(!ena_com_get_admin_running_state(adapter->ena_dev))) {
3859                 netif_err(adapter, drv, adapter->netdev,
3860                           "ENA admin queue is not in running state!\n");
3861                 ena_increase_stat(&adapter->dev_stats.admin_q_pause, 1,
3862                                   &adapter->syncp);
3863                 adapter->reset_reason = ENA_REGS_RESET_ADMIN_TO;
3864                 set_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
3865         }
3866 }
3867
3868 static void ena_update_hints(struct ena_adapter *adapter,
3869                              struct ena_admin_ena_hw_hints *hints)
3870 {
3871         struct net_device *netdev = adapter->netdev;
3872
3873         if (hints->admin_completion_tx_timeout)
3874                 adapter->ena_dev->admin_queue.completion_timeout =
3875                         hints->admin_completion_tx_timeout * 1000;
3876
3877         if (hints->mmio_read_timeout)
3878                 /* convert to usec */
3879                 adapter->ena_dev->mmio_read.reg_read_to =
3880                         hints->mmio_read_timeout * 1000;
3881
3882         if (hints->missed_tx_completion_count_threshold_to_reset)
3883                 adapter->missing_tx_completion_threshold =
3884                         hints->missed_tx_completion_count_threshold_to_reset;
3885
3886         if (hints->missing_tx_completion_timeout) {
3887                 if (hints->missing_tx_completion_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3888                         adapter->missing_tx_completion_to = ENA_HW_HINTS_NO_TIMEOUT;
3889                 else
3890                         adapter->missing_tx_completion_to =
3891                                 msecs_to_jiffies(hints->missing_tx_completion_timeout);
3892         }
3893
3894         if (hints->netdev_wd_timeout)
3895                 netdev->watchdog_timeo = msecs_to_jiffies(hints->netdev_wd_timeout);
3896
3897         if (hints->driver_watchdog_timeout) {
3898                 if (hints->driver_watchdog_timeout == ENA_HW_HINTS_NO_TIMEOUT)
3899                         adapter->keep_alive_timeout = ENA_HW_HINTS_NO_TIMEOUT;
3900                 else
3901                         adapter->keep_alive_timeout =
3902                                 msecs_to_jiffies(hints->driver_watchdog_timeout);
3903         }
3904 }
3905
3906 static void ena_update_host_info(struct ena_admin_host_info *host_info,
3907                                  struct net_device *netdev)
3908 {
3909         host_info->supported_network_features[0] =
3910                 netdev->features & GENMASK_ULL(31, 0);
3911         host_info->supported_network_features[1] =
3912                 (netdev->features & GENMASK_ULL(63, 32)) >> 32;
3913 }
3914
3915 static void ena_timer_service(struct timer_list *t)
3916 {
3917         struct ena_adapter *adapter = from_timer(adapter, t, timer_service);
3918         u8 *debug_area = adapter->ena_dev->host_attr.debug_area_virt_addr;
3919         struct ena_admin_host_info *host_info =
3920                 adapter->ena_dev->host_attr.host_info;
3921
3922         check_for_missing_keep_alive(adapter);
3923
3924         check_for_admin_com_state(adapter);
3925
3926         check_for_missing_completions(adapter);
3927
3928         check_for_empty_rx_ring(adapter);
3929
3930         if (debug_area)
3931                 ena_dump_stats_to_buf(adapter, debug_area);
3932
3933         if (host_info)
3934                 ena_update_host_info(host_info, adapter->netdev);
3935
3936         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
3937                 netif_err(adapter, drv, adapter->netdev,
3938                           "Trigger reset is on\n");
3939                 ena_dump_stats_to_dmesg(adapter);
3940                 queue_work(ena_wq, &adapter->reset_task);
3941                 return;
3942         }
3943
3944         /* Reset the timer */
3945         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
3946 }
3947
3948 static u32 ena_calc_max_io_queue_num(struct pci_dev *pdev,
3949                                      struct ena_com_dev *ena_dev,
3950                                      struct ena_com_dev_get_features_ctx *get_feat_ctx)
3951 {
3952         u32 io_tx_sq_num, io_tx_cq_num, io_rx_num, max_num_io_queues;
3953
3954         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
3955                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
3956                         &get_feat_ctx->max_queue_ext.max_queue_ext;
3957                 io_rx_num = min_t(u32, max_queue_ext->max_rx_sq_num,
3958                                   max_queue_ext->max_rx_cq_num);
3959
3960                 io_tx_sq_num = max_queue_ext->max_tx_sq_num;
3961                 io_tx_cq_num = max_queue_ext->max_tx_cq_num;
3962         } else {
3963                 struct ena_admin_queue_feature_desc *max_queues =
3964                         &get_feat_ctx->max_queues;
3965                 io_tx_sq_num = max_queues->max_sq_num;
3966                 io_tx_cq_num = max_queues->max_cq_num;
3967                 io_rx_num = min_t(u32, io_tx_sq_num, io_tx_cq_num);
3968         }
3969
3970         /* In case of LLQ use the llq fields for the tx SQ/CQ */
3971         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
3972                 io_tx_sq_num = get_feat_ctx->llq.max_llq_num;
3973
3974         max_num_io_queues = min_t(u32, num_online_cpus(), ENA_MAX_NUM_IO_QUEUES);
3975         max_num_io_queues = min_t(u32, max_num_io_queues, io_rx_num);
3976         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_sq_num);
3977         max_num_io_queues = min_t(u32, max_num_io_queues, io_tx_cq_num);
3978         /* 1 IRQ for mgmnt and 1 IRQs for each IO direction */
3979         max_num_io_queues = min_t(u32, max_num_io_queues, pci_msix_vec_count(pdev) - 1);
3980         if (unlikely(!max_num_io_queues)) {
3981                 dev_err(&pdev->dev, "The device doesn't have io queues\n");
3982                 return -EFAULT;
3983         }
3984
3985         return max_num_io_queues;
3986 }
3987
3988 static void ena_set_dev_offloads(struct ena_com_dev_get_features_ctx *feat,
3989                                  struct net_device *netdev)
3990 {
3991         netdev_features_t dev_features = 0;
3992
3993         /* Set offload features */
3994         if (feat->offload.tx &
3995                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV4_CSUM_PART_MASK)
3996                 dev_features |= NETIF_F_IP_CSUM;
3997
3998         if (feat->offload.tx &
3999                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_TX_L4_IPV6_CSUM_PART_MASK)
4000                 dev_features |= NETIF_F_IPV6_CSUM;
4001
4002         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV4_MASK)
4003                 dev_features |= NETIF_F_TSO;
4004
4005         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_IPV6_MASK)
4006                 dev_features |= NETIF_F_TSO6;
4007
4008         if (feat->offload.tx & ENA_ADMIN_FEATURE_OFFLOAD_DESC_TSO_ECN_MASK)
4009                 dev_features |= NETIF_F_TSO_ECN;
4010
4011         if (feat->offload.rx_supported &
4012                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV4_CSUM_MASK)
4013                 dev_features |= NETIF_F_RXCSUM;
4014
4015         if (feat->offload.rx_supported &
4016                 ENA_ADMIN_FEATURE_OFFLOAD_DESC_RX_L4_IPV6_CSUM_MASK)
4017                 dev_features |= NETIF_F_RXCSUM;
4018
4019         netdev->features =
4020                 dev_features |
4021                 NETIF_F_SG |
4022                 NETIF_F_RXHASH |
4023                 NETIF_F_HIGHDMA;
4024
4025         netdev->hw_features |= netdev->features;
4026         netdev->vlan_features |= netdev->features;
4027 }
4028
4029 static void ena_set_conf_feat_params(struct ena_adapter *adapter,
4030                                      struct ena_com_dev_get_features_ctx *feat)
4031 {
4032         struct net_device *netdev = adapter->netdev;
4033
4034         /* Copy mac address */
4035         if (!is_valid_ether_addr(feat->dev_attr.mac_addr)) {
4036                 eth_hw_addr_random(netdev);
4037                 ether_addr_copy(adapter->mac_addr, netdev->dev_addr);
4038         } else {
4039                 ether_addr_copy(adapter->mac_addr, feat->dev_attr.mac_addr);
4040                 ether_addr_copy(netdev->dev_addr, adapter->mac_addr);
4041         }
4042
4043         /* Set offload features */
4044         ena_set_dev_offloads(feat, netdev);
4045
4046         adapter->max_mtu = feat->dev_attr.max_mtu;
4047         netdev->max_mtu = adapter->max_mtu;
4048         netdev->min_mtu = ENA_MIN_MTU;
4049 }
4050
4051 static int ena_rss_init_default(struct ena_adapter *adapter)
4052 {
4053         struct ena_com_dev *ena_dev = adapter->ena_dev;
4054         struct device *dev = &adapter->pdev->dev;
4055         int rc, i;
4056         u32 val;
4057
4058         rc = ena_com_rss_init(ena_dev, ENA_RX_RSS_TABLE_LOG_SIZE);
4059         if (unlikely(rc)) {
4060                 dev_err(dev, "Cannot init indirect table\n");
4061                 goto err_rss_init;
4062         }
4063
4064         for (i = 0; i < ENA_RX_RSS_TABLE_SIZE; i++) {
4065                 val = ethtool_rxfh_indir_default(i, adapter->num_io_queues);
4066                 rc = ena_com_indirect_table_fill_entry(ena_dev, i,
4067                                                        ENA_IO_RXQ_IDX(val));
4068                 if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4069                         dev_err(dev, "Cannot fill indirect table\n");
4070                         goto err_fill_indir;
4071                 }
4072         }
4073
4074         rc = ena_com_fill_hash_function(ena_dev, ENA_ADMIN_TOEPLITZ, NULL,
4075                                         ENA_HASH_KEY_SIZE, 0xFFFFFFFF);
4076         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4077                 dev_err(dev, "Cannot fill hash function\n");
4078                 goto err_fill_indir;
4079         }
4080
4081         rc = ena_com_set_default_hash_ctrl(ena_dev);
4082         if (unlikely(rc && (rc != -EOPNOTSUPP))) {
4083                 dev_err(dev, "Cannot fill hash control\n");
4084                 goto err_fill_indir;
4085         }
4086
4087         return 0;
4088
4089 err_fill_indir:
4090         ena_com_rss_destroy(ena_dev);
4091 err_rss_init:
4092
4093         return rc;
4094 }
4095
4096 static void ena_release_bars(struct ena_com_dev *ena_dev, struct pci_dev *pdev)
4097 {
4098         int release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4099
4100         pci_release_selected_regions(pdev, release_bars);
4101 }
4102
4103
4104 static int ena_calc_io_queue_size(struct ena_calc_queue_size_ctx *ctx)
4105 {
4106         struct ena_admin_feature_llq_desc *llq = &ctx->get_feat_ctx->llq;
4107         struct ena_com_dev *ena_dev = ctx->ena_dev;
4108         u32 tx_queue_size = ENA_DEFAULT_RING_SIZE;
4109         u32 rx_queue_size = ENA_DEFAULT_RING_SIZE;
4110         u32 max_tx_queue_size;
4111         u32 max_rx_queue_size;
4112
4113         if (ena_dev->supported_features & BIT(ENA_ADMIN_MAX_QUEUES_EXT)) {
4114                 struct ena_admin_queue_ext_feature_fields *max_queue_ext =
4115                         &ctx->get_feat_ctx->max_queue_ext.max_queue_ext;
4116                 max_rx_queue_size = min_t(u32, max_queue_ext->max_rx_cq_depth,
4117                                           max_queue_ext->max_rx_sq_depth);
4118                 max_tx_queue_size = max_queue_ext->max_tx_cq_depth;
4119
4120                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4121                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4122                                                   llq->max_llq_depth);
4123                 else
4124                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4125                                                   max_queue_ext->max_tx_sq_depth);
4126
4127                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4128                                              max_queue_ext->max_per_packet_tx_descs);
4129                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4130                                              max_queue_ext->max_per_packet_rx_descs);
4131         } else {
4132                 struct ena_admin_queue_feature_desc *max_queues =
4133                         &ctx->get_feat_ctx->max_queues;
4134                 max_rx_queue_size = min_t(u32, max_queues->max_cq_depth,
4135                                           max_queues->max_sq_depth);
4136                 max_tx_queue_size = max_queues->max_cq_depth;
4137
4138                 if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4139                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4140                                                   llq->max_llq_depth);
4141                 else
4142                         max_tx_queue_size = min_t(u32, max_tx_queue_size,
4143                                                   max_queues->max_sq_depth);
4144
4145                 ctx->max_tx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4146                                              max_queues->max_packet_tx_descs);
4147                 ctx->max_rx_sgl_size = min_t(u16, ENA_PKT_MAX_BUFS,
4148                                              max_queues->max_packet_rx_descs);
4149         }
4150
4151         max_tx_queue_size = rounddown_pow_of_two(max_tx_queue_size);
4152         max_rx_queue_size = rounddown_pow_of_two(max_rx_queue_size);
4153
4154         tx_queue_size = clamp_val(tx_queue_size, ENA_MIN_RING_SIZE,
4155                                   max_tx_queue_size);
4156         rx_queue_size = clamp_val(rx_queue_size, ENA_MIN_RING_SIZE,
4157                                   max_rx_queue_size);
4158
4159         tx_queue_size = rounddown_pow_of_two(tx_queue_size);
4160         rx_queue_size = rounddown_pow_of_two(rx_queue_size);
4161
4162         ctx->max_tx_queue_size = max_tx_queue_size;
4163         ctx->max_rx_queue_size = max_rx_queue_size;
4164         ctx->tx_queue_size = tx_queue_size;
4165         ctx->rx_queue_size = rx_queue_size;
4166
4167         return 0;
4168 }
4169
4170 /* ena_probe - Device Initialization Routine
4171  * @pdev: PCI device information struct
4172  * @ent: entry in ena_pci_tbl
4173  *
4174  * Returns 0 on success, negative on failure
4175  *
4176  * ena_probe initializes an adapter identified by a pci_dev structure.
4177  * The OS initialization, configuring of the adapter private structure,
4178  * and a hardware reset occur.
4179  */
4180 static int ena_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
4181 {
4182         struct ena_calc_queue_size_ctx calc_queue_ctx = {};
4183         struct ena_com_dev_get_features_ctx get_feat_ctx;
4184         struct ena_com_dev *ena_dev = NULL;
4185         struct ena_adapter *adapter;
4186         struct net_device *netdev;
4187         static int adapters_found;
4188         u32 max_num_io_queues;
4189         bool wd_state;
4190         int bars, rc;
4191
4192         dev_dbg(&pdev->dev, "%s\n", __func__);
4193
4194         rc = pci_enable_device_mem(pdev);
4195         if (rc) {
4196                 dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n");
4197                 return rc;
4198         }
4199
4200         rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ENA_MAX_PHYS_ADDR_SIZE_BITS));
4201         if (rc) {
4202                 dev_err(&pdev->dev, "dma_set_mask_and_coherent failed %d\n", rc);
4203                 goto err_disable_device;
4204         }
4205
4206         pci_set_master(pdev);
4207
4208         ena_dev = vzalloc(sizeof(*ena_dev));
4209         if (!ena_dev) {
4210                 rc = -ENOMEM;
4211                 goto err_disable_device;
4212         }
4213
4214         bars = pci_select_bars(pdev, IORESOURCE_MEM) & ENA_BAR_MASK;
4215         rc = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME);
4216         if (rc) {
4217                 dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n",
4218                         rc);
4219                 goto err_free_ena_dev;
4220         }
4221
4222         ena_dev->reg_bar = devm_ioremap(&pdev->dev,
4223                                         pci_resource_start(pdev, ENA_REG_BAR),
4224                                         pci_resource_len(pdev, ENA_REG_BAR));
4225         if (!ena_dev->reg_bar) {
4226                 dev_err(&pdev->dev, "Failed to remap regs bar\n");
4227                 rc = -EFAULT;
4228                 goto err_free_region;
4229         }
4230
4231         ena_dev->ena_min_poll_delay_us = ENA_ADMIN_POLL_DELAY_US;
4232
4233         ena_dev->dmadev = &pdev->dev;
4234
4235         netdev = alloc_etherdev_mq(sizeof(struct ena_adapter), ENA_MAX_RINGS);
4236         if (!netdev) {
4237                 dev_err(&pdev->dev, "alloc_etherdev_mq failed\n");
4238                 rc = -ENOMEM;
4239                 goto err_free_region;
4240         }
4241
4242         SET_NETDEV_DEV(netdev, &pdev->dev);
4243         adapter = netdev_priv(netdev);
4244         adapter->ena_dev = ena_dev;
4245         adapter->netdev = netdev;
4246         adapter->pdev = pdev;
4247         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
4248
4249         ena_dev->net_device = netdev;
4250
4251         pci_set_drvdata(pdev, adapter);
4252
4253         rc = ena_device_init(ena_dev, pdev, &get_feat_ctx, &wd_state);
4254         if (rc) {
4255                 dev_err(&pdev->dev, "ENA device init failed\n");
4256                 if (rc == -ETIME)
4257                         rc = -EPROBE_DEFER;
4258                 goto err_netdev_destroy;
4259         }
4260
4261         rc = ena_map_llq_mem_bar(pdev, ena_dev, bars);
4262         if (rc) {
4263                 dev_err(&pdev->dev, "ENA llq bar mapping failed\n");
4264                 goto err_device_destroy;
4265         }
4266
4267         calc_queue_ctx.ena_dev = ena_dev;
4268         calc_queue_ctx.get_feat_ctx = &get_feat_ctx;
4269         calc_queue_ctx.pdev = pdev;
4270
4271         /* Initial TX and RX interrupt delay. Assumes 1 usec granularity.
4272          * Updated during device initialization with the real granularity
4273          */
4274         ena_dev->intr_moder_tx_interval = ENA_INTR_INITIAL_TX_INTERVAL_USECS;
4275         ena_dev->intr_moder_rx_interval = ENA_INTR_INITIAL_RX_INTERVAL_USECS;
4276         ena_dev->intr_delay_resolution = ENA_DEFAULT_INTR_DELAY_RESOLUTION;
4277         max_num_io_queues = ena_calc_max_io_queue_num(pdev, ena_dev, &get_feat_ctx);
4278         rc = ena_calc_io_queue_size(&calc_queue_ctx);
4279         if (rc || !max_num_io_queues) {
4280                 rc = -EFAULT;
4281                 goto err_device_destroy;
4282         }
4283
4284         ena_set_conf_feat_params(adapter, &get_feat_ctx);
4285
4286         adapter->reset_reason = ENA_REGS_RESET_NORMAL;
4287
4288         adapter->requested_tx_ring_size = calc_queue_ctx.tx_queue_size;
4289         adapter->requested_rx_ring_size = calc_queue_ctx.rx_queue_size;
4290         adapter->max_tx_ring_size = calc_queue_ctx.max_tx_queue_size;
4291         adapter->max_rx_ring_size = calc_queue_ctx.max_rx_queue_size;
4292         adapter->max_tx_sgl_size = calc_queue_ctx.max_tx_sgl_size;
4293         adapter->max_rx_sgl_size = calc_queue_ctx.max_rx_sgl_size;
4294
4295         adapter->num_io_queues = max_num_io_queues;
4296         adapter->max_num_io_queues = max_num_io_queues;
4297         adapter->last_monitored_tx_qid = 0;
4298
4299         adapter->xdp_first_ring = 0;
4300         adapter->xdp_num_queues = 0;
4301
4302         adapter->rx_copybreak = ENA_DEFAULT_RX_COPYBREAK;
4303         if (ena_dev->tx_mem_queue_type == ENA_ADMIN_PLACEMENT_POLICY_DEV)
4304                 adapter->disable_meta_caching =
4305                         !!(get_feat_ctx.llq.accel_mode.u.get.supported_flags &
4306                            BIT(ENA_ADMIN_DISABLE_META_CACHING));
4307
4308         adapter->wd_state = wd_state;
4309
4310         snprintf(adapter->name, ENA_NAME_MAX_LEN, "ena_%d", adapters_found);
4311
4312         rc = ena_com_init_interrupt_moderation(adapter->ena_dev);
4313         if (rc) {
4314                 dev_err(&pdev->dev,
4315                         "Failed to query interrupt moderation feature\n");
4316                 goto err_device_destroy;
4317         }
4318         ena_init_io_rings(adapter,
4319                           0,
4320                           adapter->xdp_num_queues +
4321                           adapter->num_io_queues);
4322
4323         netdev->netdev_ops = &ena_netdev_ops;
4324         netdev->watchdog_timeo = TX_TIMEOUT;
4325         ena_set_ethtool_ops(netdev);
4326
4327         netdev->priv_flags |= IFF_UNICAST_FLT;
4328
4329         u64_stats_init(&adapter->syncp);
4330
4331         rc = ena_enable_msix_and_set_admin_interrupts(adapter);
4332         if (rc) {
4333                 dev_err(&pdev->dev,
4334                         "Failed to enable and set the admin interrupts\n");
4335                 goto err_worker_destroy;
4336         }
4337         rc = ena_rss_init_default(adapter);
4338         if (rc && (rc != -EOPNOTSUPP)) {
4339                 dev_err(&pdev->dev, "Cannot init RSS rc: %d\n", rc);
4340                 goto err_free_msix;
4341         }
4342
4343         ena_config_debug_area(adapter);
4344
4345         if (!ena_update_hw_stats(adapter))
4346                 adapter->eni_stats_supported = true;
4347         else
4348                 adapter->eni_stats_supported = false;
4349
4350         memcpy(adapter->netdev->perm_addr, adapter->mac_addr, netdev->addr_len);
4351
4352         netif_carrier_off(netdev);
4353
4354         rc = register_netdev(netdev);
4355         if (rc) {
4356                 dev_err(&pdev->dev, "Cannot register net device\n");
4357                 goto err_rss;
4358         }
4359
4360         INIT_WORK(&adapter->reset_task, ena_fw_reset_device);
4361
4362         adapter->last_keep_alive_jiffies = jiffies;
4363         adapter->keep_alive_timeout = ENA_DEVICE_KALIVE_TIMEOUT;
4364         adapter->missing_tx_completion_to = TX_TIMEOUT;
4365         adapter->missing_tx_completion_threshold = MAX_NUM_OF_TIMEOUTED_PACKETS;
4366
4367         ena_update_hints(adapter, &get_feat_ctx.hw_hints);
4368
4369         timer_setup(&adapter->timer_service, ena_timer_service, 0);
4370         mod_timer(&adapter->timer_service, round_jiffies(jiffies + HZ));
4371
4372         dev_info(&pdev->dev,
4373                  "%s found at mem %lx, mac addr %pM\n",
4374                  DEVICE_NAME, (long)pci_resource_start(pdev, 0),
4375                  netdev->dev_addr);
4376
4377         set_bit(ENA_FLAG_DEVICE_RUNNING, &adapter->flags);
4378
4379         adapters_found++;
4380
4381         return 0;
4382
4383 err_rss:
4384         ena_com_delete_debug_area(ena_dev);
4385         ena_com_rss_destroy(ena_dev);
4386 err_free_msix:
4387         ena_com_dev_reset(ena_dev, ENA_REGS_RESET_INIT_ERR);
4388         /* stop submitting admin commands on a device that was reset */
4389         ena_com_set_admin_running_state(ena_dev, false);
4390         ena_free_mgmnt_irq(adapter);
4391         ena_disable_msix(adapter);
4392 err_worker_destroy:
4393         del_timer(&adapter->timer_service);
4394 err_device_destroy:
4395         ena_com_delete_host_info(ena_dev);
4396         ena_com_admin_destroy(ena_dev);
4397 err_netdev_destroy:
4398         free_netdev(netdev);
4399 err_free_region:
4400         ena_release_bars(ena_dev, pdev);
4401 err_free_ena_dev:
4402         vfree(ena_dev);
4403 err_disable_device:
4404         pci_disable_device(pdev);
4405         return rc;
4406 }
4407
4408 /*****************************************************************************/
4409
4410 /* __ena_shutoff - Helper used in both PCI remove/shutdown routines
4411  * @pdev: PCI device information struct
4412  * @shutdown: Is it a shutdown operation? If false, means it is a removal
4413  *
4414  * __ena_shutoff is a helper routine that does the real work on shutdown and
4415  * removal paths; the difference between those paths is with regards to whether
4416  * dettach or unregister the netdevice.
4417  */
4418 static void __ena_shutoff(struct pci_dev *pdev, bool shutdown)
4419 {
4420         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4421         struct ena_com_dev *ena_dev;
4422         struct net_device *netdev;
4423
4424         ena_dev = adapter->ena_dev;
4425         netdev = adapter->netdev;
4426
4427 #ifdef CONFIG_RFS_ACCEL
4428         if ((adapter->msix_vecs >= 1) && (netdev->rx_cpu_rmap)) {
4429                 free_irq_cpu_rmap(netdev->rx_cpu_rmap);
4430                 netdev->rx_cpu_rmap = NULL;
4431         }
4432 #endif /* CONFIG_RFS_ACCEL */
4433
4434         /* Make sure timer and reset routine won't be called after
4435          * freeing device resources.
4436          */
4437         del_timer_sync(&adapter->timer_service);
4438         cancel_work_sync(&adapter->reset_task);
4439
4440         rtnl_lock(); /* lock released inside the below if-else block */
4441         adapter->reset_reason = ENA_REGS_RESET_SHUTDOWN;
4442         ena_destroy_device(adapter, true);
4443         if (shutdown) {
4444                 netif_device_detach(netdev);
4445                 dev_close(netdev);
4446                 rtnl_unlock();
4447         } else {
4448                 rtnl_unlock();
4449                 unregister_netdev(netdev);
4450                 free_netdev(netdev);
4451         }
4452
4453         ena_com_rss_destroy(ena_dev);
4454
4455         ena_com_delete_debug_area(ena_dev);
4456
4457         ena_com_delete_host_info(ena_dev);
4458
4459         ena_release_bars(ena_dev, pdev);
4460
4461         pci_disable_device(pdev);
4462
4463         vfree(ena_dev);
4464 }
4465
4466 /* ena_remove - Device Removal Routine
4467  * @pdev: PCI device information struct
4468  *
4469  * ena_remove is called by the PCI subsystem to alert the driver
4470  * that it should release a PCI device.
4471  */
4472
4473 static void ena_remove(struct pci_dev *pdev)
4474 {
4475         __ena_shutoff(pdev, false);
4476 }
4477
4478 /* ena_shutdown - Device Shutdown Routine
4479  * @pdev: PCI device information struct
4480  *
4481  * ena_shutdown is called by the PCI subsystem to alert the driver that
4482  * a shutdown/reboot (or kexec) is happening and device must be disabled.
4483  */
4484
4485 static void ena_shutdown(struct pci_dev *pdev)
4486 {
4487         __ena_shutoff(pdev, true);
4488 }
4489
4490 /* ena_suspend - PM suspend callback
4491  * @dev_d: Device information struct
4492  */
4493 static int __maybe_unused ena_suspend(struct device *dev_d)
4494 {
4495         struct pci_dev *pdev = to_pci_dev(dev_d);
4496         struct ena_adapter *adapter = pci_get_drvdata(pdev);
4497
4498         ena_increase_stat(&adapter->dev_stats.suspend, 1, &adapter->syncp);
4499
4500         rtnl_lock();
4501         if (unlikely(test_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags))) {
4502                 dev_err(&pdev->dev,
4503                         "Ignoring device reset request as the device is being suspended\n");
4504                 clear_bit(ENA_FLAG_TRIGGER_RESET, &adapter->flags);
4505         }
4506         ena_destroy_device(adapter, true);
4507         rtnl_unlock();
4508         return 0;
4509 }
4510
4511 /* ena_resume - PM resume callback
4512  * @dev_d: Device information struct
4513  */
4514 static int __maybe_unused ena_resume(struct device *dev_d)
4515 {
4516         struct ena_adapter *adapter = dev_get_drvdata(dev_d);
4517         int rc;
4518
4519         ena_increase_stat(&adapter->dev_stats.resume, 1, &adapter->syncp);
4520
4521         rtnl_lock();
4522         rc = ena_restore_device(adapter);
4523         rtnl_unlock();
4524         return rc;
4525 }
4526
4527 static SIMPLE_DEV_PM_OPS(ena_pm_ops, ena_suspend, ena_resume);
4528
4529 static struct pci_driver ena_pci_driver = {
4530         .name           = DRV_MODULE_NAME,
4531         .id_table       = ena_pci_tbl,
4532         .probe          = ena_probe,
4533         .remove         = ena_remove,
4534         .shutdown       = ena_shutdown,
4535         .driver.pm      = &ena_pm_ops,
4536         .sriov_configure = pci_sriov_configure_simple,
4537 };
4538
4539 static int __init ena_init(void)
4540 {
4541         ena_wq = create_singlethread_workqueue(DRV_MODULE_NAME);
4542         if (!ena_wq) {
4543                 pr_err("Failed to create workqueue\n");
4544                 return -ENOMEM;
4545         }
4546
4547         return pci_register_driver(&ena_pci_driver);
4548 }
4549
4550 static void __exit ena_cleanup(void)
4551 {
4552         pci_unregister_driver(&ena_pci_driver);
4553
4554         if (ena_wq) {
4555                 destroy_workqueue(ena_wq);
4556                 ena_wq = NULL;
4557         }
4558 }
4559
4560 /******************************************************************************
4561  ******************************** AENQ Handlers *******************************
4562  *****************************************************************************/
4563 /* ena_update_on_link_change:
4564  * Notify the network interface about the change in link status
4565  */
4566 static void ena_update_on_link_change(void *adapter_data,
4567                                       struct ena_admin_aenq_entry *aenq_e)
4568 {
4569         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4570         struct ena_admin_aenq_link_change_desc *aenq_desc =
4571                 (struct ena_admin_aenq_link_change_desc *)aenq_e;
4572         int status = aenq_desc->flags &
4573                 ENA_ADMIN_AENQ_LINK_CHANGE_DESC_LINK_STATUS_MASK;
4574
4575         if (status) {
4576                 netif_dbg(adapter, ifup, adapter->netdev, "%s\n", __func__);
4577                 set_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4578                 if (!test_bit(ENA_FLAG_ONGOING_RESET, &adapter->flags))
4579                         netif_carrier_on(adapter->netdev);
4580         } else {
4581                 clear_bit(ENA_FLAG_LINK_UP, &adapter->flags);
4582                 netif_carrier_off(adapter->netdev);
4583         }
4584 }
4585
4586 static void ena_keep_alive_wd(void *adapter_data,
4587                               struct ena_admin_aenq_entry *aenq_e)
4588 {
4589         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4590         struct ena_admin_aenq_keep_alive_desc *desc;
4591         u64 rx_drops;
4592         u64 tx_drops;
4593
4594         desc = (struct ena_admin_aenq_keep_alive_desc *)aenq_e;
4595         adapter->last_keep_alive_jiffies = jiffies;
4596
4597         rx_drops = ((u64)desc->rx_drops_high << 32) | desc->rx_drops_low;
4598         tx_drops = ((u64)desc->tx_drops_high << 32) | desc->tx_drops_low;
4599
4600         u64_stats_update_begin(&adapter->syncp);
4601         /* These stats are accumulated by the device, so the counters indicate
4602          * all drops since last reset.
4603          */
4604         adapter->dev_stats.rx_drops = rx_drops;
4605         adapter->dev_stats.tx_drops = tx_drops;
4606         u64_stats_update_end(&adapter->syncp);
4607 }
4608
4609 static void ena_notification(void *adapter_data,
4610                              struct ena_admin_aenq_entry *aenq_e)
4611 {
4612         struct ena_adapter *adapter = (struct ena_adapter *)adapter_data;
4613         struct ena_admin_ena_hw_hints *hints;
4614
4615         WARN(aenq_e->aenq_common_desc.group != ENA_ADMIN_NOTIFICATION,
4616              "Invalid group(%x) expected %x\n",
4617              aenq_e->aenq_common_desc.group,
4618              ENA_ADMIN_NOTIFICATION);
4619
4620         switch (aenq_e->aenq_common_desc.syndrome) {
4621         case ENA_ADMIN_UPDATE_HINTS:
4622                 hints = (struct ena_admin_ena_hw_hints *)
4623                         (&aenq_e->inline_data_w4);
4624                 ena_update_hints(adapter, hints);
4625                 break;
4626         default:
4627                 netif_err(adapter, drv, adapter->netdev,
4628                           "Invalid aenq notification link state %d\n",
4629                           aenq_e->aenq_common_desc.syndrome);
4630         }
4631 }
4632
4633 /* This handler will called for unknown event group or unimplemented handlers*/
4634 static void unimplemented_aenq_handler(void *data,
4635                                        struct ena_admin_aenq_entry *aenq_e)
4636 {
4637         struct ena_adapter *adapter = (struct ena_adapter *)data;
4638
4639         netif_err(adapter, drv, adapter->netdev,
4640                   "Unknown event was received or event with unimplemented handler\n");
4641 }
4642
4643 static struct ena_aenq_handlers aenq_handlers = {
4644         .handlers = {
4645                 [ENA_ADMIN_LINK_CHANGE] = ena_update_on_link_change,
4646                 [ENA_ADMIN_NOTIFICATION] = ena_notification,
4647                 [ENA_ADMIN_KEEP_ALIVE] = ena_keep_alive_wd,
4648         },
4649         .unimplemented_handler = unimplemented_aenq_handler
4650 };
4651
4652 module_init(ena_init);
4653 module_exit(ena_cleanup);