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

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