drivers/net/ethernet/intel/igc/igc_main.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright (c)  2018 Intel Corporation */
   3
   4 #include <linux/module.h>
   5 #include <linux/types.h>
   6 #include <linux/if_vlan.h>
   7 #include <linux/aer.h>
   8 #include <linux/tcp.h>
   9 #include <linux/udp.h>
  10 #include <linux/ip.h>
  11 #include <linux/pm_runtime.h>
  12 #include <net/pkt_sched.h>
  13 #include <linux/bpf_trace.h>
  14 #include <net/xdp_sock_drv.h>
  15 #include <net/ipv6.h>
  16
  17 #include "igc.h"
  18 #include "igc_hw.h"
  19 #include "igc_tsn.h"
  20 #include "igc_xdp.h"
  21
  22 #define DRV_SUMMARY     "Intel(R) 2.5G Ethernet Linux Driver"
  23
  24 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK)
  25
  26 #define IGC_XDP_PASS            0
  27 #define IGC_XDP_CONSUMED        BIT(0)
  28 #define IGC_XDP_TX              BIT(1)
  29 #define IGC_XDP_REDIRECT        BIT(2)
  30
  31 static int debug = -1;
  32
  33 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  34 MODULE_DESCRIPTION(DRV_SUMMARY);
  35 MODULE_LICENSE("GPL v2");
  36 module_param(debug, int, 0);
  37 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  38
  39 char igc_driver_name[] = "igc";
  40 static const char igc_driver_string[] = DRV_SUMMARY;
  41 static const char igc_copyright[] =
  42         "Copyright(c) 2018 Intel Corporation.";
  43
  44 static const struct igc_info *igc_info_tbl[] = {
  45         [board_base] = &igc_base_info,
  46 };
  47
  48 static const struct pci_device_id igc_pci_tbl[] = {
  49         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_LM), board_base },
  50         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_V), board_base },
  51         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_I), board_base },
  52         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I220_V), board_base },
  53         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_K), board_base },
  54         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_K2), board_base },
  55         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_K), board_base },
  56         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_LMVP), board_base },
  57         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_IT), board_base },
  58         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_LM), board_base },
  59         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_V), board_base },
  60         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_IT), board_base },
  61         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I221_V), board_base },
  62         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_BLANK_NVM), board_base },
  63         { PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_BLANK_NVM), board_base },
  64         /* required last entry */
  65         {0, }
  66 };
  67
  68 MODULE_DEVICE_TABLE(pci, igc_pci_tbl);
  69
  70 enum latency_range {
  71         lowest_latency = 0,
  72         low_latency = 1,
  73         bulk_latency = 2,
  74         latency_invalid = 255
  75 };
  76
  77 void igc_reset(struct igc_adapter *adapter)
  78 {
  79         struct net_device *dev = adapter->netdev;
  80         struct igc_hw *hw = &adapter->hw;
  81         struct igc_fc_info *fc = &hw->fc;
  82         u32 pba, hwm;
  83
  84         /* Repartition PBA for greater than 9k MTU if required */
  85         pba = IGC_PBA_34K;
  86
  87         /* flow control settings
  88          * The high water mark must be low enough to fit one full frame
  89          * after transmitting the pause frame.  As such we must have enough
  90          * space to allow for us to complete our current transmit and then
  91          * receive the frame that is in progress from the link partner.
  92          * Set it to:
  93          * - the full Rx FIFO size minus one full Tx plus one full Rx frame
  94          */
  95         hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
  96
  97         fc->high_water = hwm & 0xFFFFFFF0;      /* 16-byte granularity */
  98         fc->low_water = fc->high_water - 16;
  99         fc->pause_time = 0xFFFF;
 100         fc->send_xon = 1;
 101         fc->current_mode = fc->requested_mode;
 102
 103         hw->mac.ops.reset_hw(hw);
 104
 105         if (hw->mac.ops.init_hw(hw))
 106                 netdev_err(dev, "Error on hardware initialization\n");
 107
 108         /* Re-establish EEE setting */
 109         igc_set_eee_i225(hw, true, true, true);
 110
 111         if (!netif_running(adapter->netdev))
 112                 igc_power_down_phy_copper_base(&adapter->hw);
 113
 114         /* Enable HW to recognize an 802.1Q VLAN Ethernet packet */
 115         wr32(IGC_VET, ETH_P_8021Q);
 116
 117         /* Re-enable PTP, where applicable. */
 118         igc_ptp_reset(adapter);
 119
 120         /* Re-enable TSN offloading, where applicable. */
 121         igc_tsn_offload_apply(adapter);
 122
 123         igc_get_phy_info(hw);
 124 }
 125
 126 /**
 127  * igc_power_up_link - Power up the phy link
 128  * @adapter: address of board private structure
 129  */
 130 static void igc_power_up_link(struct igc_adapter *adapter)
 131 {
 132         igc_reset_phy(&adapter->hw);
 133
 134         igc_power_up_phy_copper(&adapter->hw);
 135
 136         igc_setup_link(&adapter->hw);
 137 }
 138
 139 /**
 140  * igc_release_hw_control - release control of the h/w to f/w
 141  * @adapter: address of board private structure
 142  *
 143  * igc_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
 144  * For ASF and Pass Through versions of f/w this means that the
 145  * driver is no longer loaded.
 146  */
 147 static void igc_release_hw_control(struct igc_adapter *adapter)
 148 {
 149         struct igc_hw *hw = &adapter->hw;
 150         u32 ctrl_ext;
 151
 152         /* Let firmware take over control of h/w */
 153         ctrl_ext = rd32(IGC_CTRL_EXT);
 154         wr32(IGC_CTRL_EXT,
 155              ctrl_ext & ~IGC_CTRL_EXT_DRV_LOAD);
 156 }
 157
 158 /**
 159  * igc_get_hw_control - get control of the h/w from f/w
 160  * @adapter: address of board private structure
 161  *
 162  * igc_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
 163  * For ASF and Pass Through versions of f/w this means that
 164  * the driver is loaded.
 165  */
 166 static void igc_get_hw_control(struct igc_adapter *adapter)
 167 {
 168         struct igc_hw *hw = &adapter->hw;
 169         u32 ctrl_ext;
 170
 171         /* Let firmware know the driver has taken over */
 172         ctrl_ext = rd32(IGC_CTRL_EXT);
 173         wr32(IGC_CTRL_EXT,
 174              ctrl_ext | IGC_CTRL_EXT_DRV_LOAD);
 175 }
 176
 177 static void igc_unmap_tx_buffer(struct device *dev, struct igc_tx_buffer *buf)
 178 {
 179         dma_unmap_single(dev, dma_unmap_addr(buf, dma),
 180                          dma_unmap_len(buf, len), DMA_TO_DEVICE);
 181
 182         dma_unmap_len_set(buf, len, 0);
 183 }
 184
 185 /**
 186  * igc_clean_tx_ring - Free Tx Buffers
 187  * @tx_ring: ring to be cleaned
 188  */
 189 static void igc_clean_tx_ring(struct igc_ring *tx_ring)
 190 {
 191         u16 i = tx_ring->next_to_clean;
 192         struct igc_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
 193         u32 xsk_frames = 0;
 194
 195         while (i != tx_ring->next_to_use) {
 196                 union igc_adv_tx_desc *eop_desc, *tx_desc;
 197
 198                 switch (tx_buffer->type) {
 199                 case IGC_TX_BUFFER_TYPE_XSK:
 200                         xsk_frames++;
 201                         break;
 202                 case IGC_TX_BUFFER_TYPE_XDP:
 203                         xdp_return_frame(tx_buffer->xdpf);
 204                         igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 205                         break;
 206                 case IGC_TX_BUFFER_TYPE_SKB:
 207                         dev_kfree_skb_any(tx_buffer->skb);
 208                         igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 209                         break;
 210                 default:
 211                         netdev_warn_once(tx_ring->netdev, "Unknown Tx buffer type\n");
 212                         break;
 213                 }
 214
 215                 /* check for eop_desc to determine the end of the packet */
 216                 eop_desc = tx_buffer->next_to_watch;
 217                 tx_desc = IGC_TX_DESC(tx_ring, i);
 218
 219                 /* unmap remaining buffers */
 220                 while (tx_desc != eop_desc) {
 221                         tx_buffer++;
 222                         tx_desc++;
 223                         i++;
 224                         if (unlikely(i == tx_ring->count)) {
 225                                 i = 0;
 226                                 tx_buffer = tx_ring->tx_buffer_info;
 227                                 tx_desc = IGC_TX_DESC(tx_ring, 0);
 228                         }
 229
 230                         /* unmap any remaining paged data */
 231                         if (dma_unmap_len(tx_buffer, len))
 232                                 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 233                 }
 234
 235                 tx_buffer->next_to_watch = NULL;
 236
 237                 /* move us one more past the eop_desc for start of next pkt */
 238                 tx_buffer++;
 239                 i++;
 240                 if (unlikely(i == tx_ring->count)) {
 241                         i = 0;
 242                         tx_buffer = tx_ring->tx_buffer_info;
 243                 }
 244         }
 245
 246         if (tx_ring->xsk_pool && xsk_frames)
 247                 xsk_tx_completed(tx_ring->xsk_pool, xsk_frames);
 248
 249         /* reset BQL for queue */
 250         netdev_tx_reset_queue(txring_txq(tx_ring));
 251
 252         /* reset next_to_use and next_to_clean */
 253         tx_ring->next_to_use = 0;
 254         tx_ring->next_to_clean = 0;
 255 }
 256
 257 /**
 258  * igc_free_tx_resources - Free Tx Resources per Queue
 259  * @tx_ring: Tx descriptor ring for a specific queue
 260  *
 261  * Free all transmit software resources
 262  */
 263 void igc_free_tx_resources(struct igc_ring *tx_ring)
 264 {
 265         igc_clean_tx_ring(tx_ring);
 266
 267         vfree(tx_ring->tx_buffer_info);
 268         tx_ring->tx_buffer_info = NULL;
 269
 270         /* if not set, then don't free */
 271         if (!tx_ring->desc)
 272                 return;
 273
 274         dma_free_coherent(tx_ring->dev, tx_ring->size,
 275                           tx_ring->desc, tx_ring->dma);
 276
 277         tx_ring->desc = NULL;
 278 }
 279
 280 /**
 281  * igc_free_all_tx_resources - Free Tx Resources for All Queues
 282  * @adapter: board private structure
 283  *
 284  * Free all transmit software resources
 285  */
 286 static void igc_free_all_tx_resources(struct igc_adapter *adapter)
 287 {
 288         int i;
 289
 290         for (i = 0; i < adapter->num_tx_queues; i++)
 291                 igc_free_tx_resources(adapter->tx_ring[i]);
 292 }
 293
 294 /**
 295  * igc_clean_all_tx_rings - Free Tx Buffers for all queues
 296  * @adapter: board private structure
 297  */
 298 static void igc_clean_all_tx_rings(struct igc_adapter *adapter)
 299 {
 300         int i;
 301
 302         for (i = 0; i < adapter->num_tx_queues; i++)
 303                 if (adapter->tx_ring[i])
 304                         igc_clean_tx_ring(adapter->tx_ring[i]);
 305 }
 306
 307 /**
 308  * igc_setup_tx_resources - allocate Tx resources (Descriptors)
 309  * @tx_ring: tx descriptor ring (for a specific queue) to setup
 310  *
 311  * Return 0 on success, negative on failure
 312  */
 313 int igc_setup_tx_resources(struct igc_ring *tx_ring)
 314 {
 315         struct net_device *ndev = tx_ring->netdev;
 316         struct device *dev = tx_ring->dev;
 317         int size = 0;
 318
 319         size = sizeof(struct igc_tx_buffer) * tx_ring->count;
 320         tx_ring->tx_buffer_info = vzalloc(size);
 321         if (!tx_ring->tx_buffer_info)
 322                 goto err;
 323
 324         /* round up to nearest 4K */
 325         tx_ring->size = tx_ring->count * sizeof(union igc_adv_tx_desc);
 326         tx_ring->size = ALIGN(tx_ring->size, 4096);
 327
 328         tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
 329                                            &tx_ring->dma, GFP_KERNEL);
 330
 331         if (!tx_ring->desc)
 332                 goto err;
 333
 334         tx_ring->next_to_use = 0;
 335         tx_ring->next_to_clean = 0;
 336
 337         return 0;
 338
 339 err:
 340         vfree(tx_ring->tx_buffer_info);
 341         netdev_err(ndev, "Unable to allocate memory for Tx descriptor ring\n");
 342         return -ENOMEM;
 343 }
 344
 345 /**
 346  * igc_setup_all_tx_resources - wrapper to allocate Tx resources for all queues
 347  * @adapter: board private structure
 348  *
 349  * Return 0 on success, negative on failure
 350  */
 351 static int igc_setup_all_tx_resources(struct igc_adapter *adapter)
 352 {
 353         struct net_device *dev = adapter->netdev;
 354         int i, err = 0;
 355
 356         for (i = 0; i < adapter->num_tx_queues; i++) {
 357                 err = igc_setup_tx_resources(adapter->tx_ring[i]);
 358                 if (err) {
 359                         netdev_err(dev, "Error on Tx queue %u setup\n", i);
 360                         for (i--; i >= 0; i--)
 361                                 igc_free_tx_resources(adapter->tx_ring[i]);
 362                         break;
 363                 }
 364         }
 365
 366         return err;
 367 }
 368
 369 static void igc_clean_rx_ring_page_shared(struct igc_ring *rx_ring)
 370 {
 371         u16 i = rx_ring->next_to_clean;
 372
 373         dev_kfree_skb(rx_ring->skb);
 374         rx_ring->skb = NULL;
 375
 376         /* Free all the Rx ring sk_buffs */
 377         while (i != rx_ring->next_to_alloc) {
 378                 struct igc_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
 379
 380                 /* Invalidate cache lines that may have been written to by
 381                  * device so that we avoid corrupting memory.
 382                  */
 383                 dma_sync_single_range_for_cpu(rx_ring->dev,
 384                                               buffer_info->dma,
 385                                               buffer_info->page_offset,
 386                                               igc_rx_bufsz(rx_ring),
 387                                               DMA_FROM_DEVICE);
 388
 389                 /* free resources associated with mapping */
 390                 dma_unmap_page_attrs(rx_ring->dev,
 391                                      buffer_info->dma,
 392                                      igc_rx_pg_size(rx_ring),
 393                                      DMA_FROM_DEVICE,
 394                                      IGC_RX_DMA_ATTR);
 395                 __page_frag_cache_drain(buffer_info->page,
 396                                         buffer_info->pagecnt_bias);
 397
 398                 i++;
 399                 if (i == rx_ring->count)
 400                         i = 0;
 401         }
 402 }
 403
 404 static void igc_clean_rx_ring_xsk_pool(struct igc_ring *ring)
 405 {
 406         struct igc_rx_buffer *bi;
 407         u16 i;
 408
 409         for (i = 0; i < ring->count; i++) {
 410                 bi = &ring->rx_buffer_info[i];
 411                 if (!bi->xdp)
 412                         continue;
 413
 414                 xsk_buff_free(bi->xdp);
 415                 bi->xdp = NULL;
 416         }
 417 }
 418
 419 /**
 420  * igc_clean_rx_ring - Free Rx Buffers per Queue
 421  * @ring: ring to free buffers from
 422  */
 423 static void igc_clean_rx_ring(struct igc_ring *ring)
 424 {
 425         if (ring->xsk_pool)
 426                 igc_clean_rx_ring_xsk_pool(ring);
 427         else
 428                 igc_clean_rx_ring_page_shared(ring);
 429
 430         clear_ring_uses_large_buffer(ring);
 431
 432         ring->next_to_alloc = 0;
 433         ring->next_to_clean = 0;
 434         ring->next_to_use = 0;
 435 }
 436
 437 /**
 438  * igc_clean_all_rx_rings - Free Rx Buffers for all queues
 439  * @adapter: board private structure
 440  */
 441 static void igc_clean_all_rx_rings(struct igc_adapter *adapter)
 442 {
 443         int i;
 444
 445         for (i = 0; i < adapter->num_rx_queues; i++)
 446                 if (adapter->rx_ring[i])
 447                         igc_clean_rx_ring(adapter->rx_ring[i]);
 448 }
 449
 450 /**
 451  * igc_free_rx_resources - Free Rx Resources
 452  * @rx_ring: ring to clean the resources from
 453  *
 454  * Free all receive software resources
 455  */
 456 void igc_free_rx_resources(struct igc_ring *rx_ring)
 457 {
 458         igc_clean_rx_ring(rx_ring);
 459
 460         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 461
 462         vfree(rx_ring->rx_buffer_info);
 463         rx_ring->rx_buffer_info = NULL;
 464
 465         /* if not set, then don't free */
 466         if (!rx_ring->desc)
 467                 return;
 468
 469         dma_free_coherent(rx_ring->dev, rx_ring->size,
 470                           rx_ring->desc, rx_ring->dma);
 471
 472         rx_ring->desc = NULL;
 473 }
 474
 475 /**
 476  * igc_free_all_rx_resources - Free Rx Resources for All Queues
 477  * @adapter: board private structure
 478  *
 479  * Free all receive software resources
 480  */
 481 static void igc_free_all_rx_resources(struct igc_adapter *adapter)
 482 {
 483         int i;
 484
 485         for (i = 0; i < adapter->num_rx_queues; i++)
 486                 igc_free_rx_resources(adapter->rx_ring[i]);
 487 }
 488
 489 /**
 490  * igc_setup_rx_resources - allocate Rx resources (Descriptors)
 491  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
 492  *
 493  * Returns 0 on success, negative on failure
 494  */
 495 int igc_setup_rx_resources(struct igc_ring *rx_ring)
 496 {
 497         struct net_device *ndev = rx_ring->netdev;
 498         struct device *dev = rx_ring->dev;
 499         u8 index = rx_ring->queue_index;
 500         int size, desc_len, res;
 501
 502         res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, ndev, index,
 503                                rx_ring->q_vector->napi.napi_id);
 504         if (res < 0) {
 505                 netdev_err(ndev, "Failed to register xdp_rxq index %u\n",
 506                            index);
 507                 return res;
 508         }
 509
 510         size = sizeof(struct igc_rx_buffer) * rx_ring->count;
 511         rx_ring->rx_buffer_info = vzalloc(size);
 512         if (!rx_ring->rx_buffer_info)
 513                 goto err;
 514
 515         desc_len = sizeof(union igc_adv_rx_desc);
 516
 517         /* Round up to nearest 4K */
 518         rx_ring->size = rx_ring->count * desc_len;
 519         rx_ring->size = ALIGN(rx_ring->size, 4096);
 520
 521         rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
 522                                            &rx_ring->dma, GFP_KERNEL);
 523
 524         if (!rx_ring->desc)
 525                 goto err;
 526
 527         rx_ring->next_to_alloc = 0;
 528         rx_ring->next_to_clean = 0;
 529         rx_ring->next_to_use = 0;
 530
 531         return 0;
 532
 533 err:
 534         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 535         vfree(rx_ring->rx_buffer_info);
 536         rx_ring->rx_buffer_info = NULL;
 537         netdev_err(ndev, "Unable to allocate memory for Rx descriptor ring\n");
 538         return -ENOMEM;
 539 }
 540
 541 /**
 542  * igc_setup_all_rx_resources - wrapper to allocate Rx resources
 543  *                                (Descriptors) for all queues
 544  * @adapter: board private structure
 545  *
 546  * Return 0 on success, negative on failure
 547  */
 548 static int igc_setup_all_rx_resources(struct igc_adapter *adapter)
 549 {
 550         struct net_device *dev = adapter->netdev;
 551         int i, err = 0;
 552
 553         for (i = 0; i < adapter->num_rx_queues; i++) {
 554                 err = igc_setup_rx_resources(adapter->rx_ring[i]);
 555                 if (err) {
 556                         netdev_err(dev, "Error on Rx queue %u setup\n", i);
 557                         for (i--; i >= 0; i--)
 558                                 igc_free_rx_resources(adapter->rx_ring[i]);
 559                         break;
 560                 }
 561         }
 562
 563         return err;
 564 }
 565
 566 static struct xsk_buff_pool *igc_get_xsk_pool(struct igc_adapter *adapter,
 567                                               struct igc_ring *ring)
 568 {
 569         if (!igc_xdp_is_enabled(adapter) ||
 570             !test_bit(IGC_RING_FLAG_AF_XDP_ZC, &ring->flags))
 571                 return NULL;
 572
 573         return xsk_get_pool_from_qid(ring->netdev, ring->queue_index);
 574 }
 575
 576 /**
 577  * igc_configure_rx_ring - Configure a receive ring after Reset
 578  * @adapter: board private structure
 579  * @ring: receive ring to be configured
 580  *
 581  * Configure the Rx unit of the MAC after a reset.
 582  */
 583 static void igc_configure_rx_ring(struct igc_adapter *adapter,
 584                                   struct igc_ring *ring)
 585 {
 586         struct igc_hw *hw = &adapter->hw;
 587         union igc_adv_rx_desc *rx_desc;
 588         int reg_idx = ring->reg_idx;
 589         u32 srrctl = 0, rxdctl = 0;
 590         u64 rdba = ring->dma;
 591         u32 buf_size;
 592
 593         xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
 594         ring->xsk_pool = igc_get_xsk_pool(adapter, ring);
 595         if (ring->xsk_pool) {
 596                 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
 597                                                    MEM_TYPE_XSK_BUFF_POOL,
 598                                                    NULL));
 599                 xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
 600         } else {
 601                 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
 602                                                    MEM_TYPE_PAGE_SHARED,
 603                                                    NULL));
 604         }
 605
 606         if (igc_xdp_is_enabled(adapter))
 607                 set_ring_uses_large_buffer(ring);
 608
 609         /* disable the queue */
 610         wr32(IGC_RXDCTL(reg_idx), 0);
 611
 612         /* Set DMA base address registers */
 613         wr32(IGC_RDBAL(reg_idx),
 614              rdba & 0x00000000ffffffffULL);
 615         wr32(IGC_RDBAH(reg_idx), rdba >> 32);
 616         wr32(IGC_RDLEN(reg_idx),
 617              ring->count * sizeof(union igc_adv_rx_desc));
 618
 619         /* initialize head and tail */
 620         ring->tail = adapter->io_addr + IGC_RDT(reg_idx);
 621         wr32(IGC_RDH(reg_idx), 0);
 622         writel(0, ring->tail);
 623
 624         /* reset next-to- use/clean to place SW in sync with hardware */
 625         ring->next_to_clean = 0;
 626         ring->next_to_use = 0;
 627
 628         if (ring->xsk_pool)
 629                 buf_size = xsk_pool_get_rx_frame_size(ring->xsk_pool);
 630         else if (ring_uses_large_buffer(ring))
 631                 buf_size = IGC_RXBUFFER_3072;
 632         else
 633                 buf_size = IGC_RXBUFFER_2048;
 634
 635         srrctl = IGC_RX_HDR_LEN << IGC_SRRCTL_BSIZEHDRSIZE_SHIFT;
 636         srrctl |= buf_size >> IGC_SRRCTL_BSIZEPKT_SHIFT;
 637         srrctl |= IGC_SRRCTL_DESCTYPE_ADV_ONEBUF;
 638
 639         wr32(IGC_SRRCTL(reg_idx), srrctl);
 640
 641         rxdctl |= IGC_RX_PTHRESH;
 642         rxdctl |= IGC_RX_HTHRESH << 8;
 643         rxdctl |= IGC_RX_WTHRESH << 16;
 644
 645         /* initialize rx_buffer_info */
 646         memset(ring->rx_buffer_info, 0,
 647                sizeof(struct igc_rx_buffer) * ring->count);
 648
 649         /* initialize Rx descriptor 0 */
 650         rx_desc = IGC_RX_DESC(ring, 0);
 651         rx_desc->wb.upper.length = 0;
 652
 653         /* enable receive descriptor fetching */
 654         rxdctl |= IGC_RXDCTL_QUEUE_ENABLE;
 655
 656         wr32(IGC_RXDCTL(reg_idx), rxdctl);
 657 }
 658
 659 /**
 660  * igc_configure_rx - Configure receive Unit after Reset
 661  * @adapter: board private structure
 662  *
 663  * Configure the Rx unit of the MAC after a reset.
 664  */
 665 static void igc_configure_rx(struct igc_adapter *adapter)
 666 {
 667         int i;
 668
 669         /* Setup the HW Rx Head and Tail Descriptor Pointers and
 670          * the Base and Length of the Rx Descriptor Ring
 671          */
 672         for (i = 0; i < adapter->num_rx_queues; i++)
 673                 igc_configure_rx_ring(adapter, adapter->rx_ring[i]);
 674 }
 675
 676 /**
 677  * igc_configure_tx_ring - Configure transmit ring after Reset
 678  * @adapter: board private structure
 679  * @ring: tx ring to configure
 680  *
 681  * Configure a transmit ring after a reset.
 682  */
 683 static void igc_configure_tx_ring(struct igc_adapter *adapter,
 684                                   struct igc_ring *ring)
 685 {
 686         struct igc_hw *hw = &adapter->hw;
 687         int reg_idx = ring->reg_idx;
 688         u64 tdba = ring->dma;
 689         u32 txdctl = 0;
 690
 691         ring->xsk_pool = igc_get_xsk_pool(adapter, ring);
 692
 693         /* disable the queue */
 694         wr32(IGC_TXDCTL(reg_idx), 0);
 695         wrfl();
 696         mdelay(10);
 697
 698         wr32(IGC_TDLEN(reg_idx),
 699              ring->count * sizeof(union igc_adv_tx_desc));
 700         wr32(IGC_TDBAL(reg_idx),
 701              tdba & 0x00000000ffffffffULL);
 702         wr32(IGC_TDBAH(reg_idx), tdba >> 32);
 703
 704         ring->tail = adapter->io_addr + IGC_TDT(reg_idx);
 705         wr32(IGC_TDH(reg_idx), 0);
 706         writel(0, ring->tail);
 707
 708         txdctl |= IGC_TX_PTHRESH;
 709         txdctl |= IGC_TX_HTHRESH << 8;
 710         txdctl |= IGC_TX_WTHRESH << 16;
 711
 712         txdctl |= IGC_TXDCTL_QUEUE_ENABLE;
 713         wr32(IGC_TXDCTL(reg_idx), txdctl);
 714 }
 715
 716 /**
 717  * igc_configure_tx - Configure transmit Unit after Reset
 718  * @adapter: board private structure
 719  *
 720  * Configure the Tx unit of the MAC after a reset.
 721  */
 722 static void igc_configure_tx(struct igc_adapter *adapter)
 723 {
 724         int i;
 725
 726         for (i = 0; i < adapter->num_tx_queues; i++)
 727                 igc_configure_tx_ring(adapter, adapter->tx_ring[i]);
 728 }
 729
 730 /**
 731  * igc_setup_mrqc - configure the multiple receive queue control registers
 732  * @adapter: Board private structure
 733  */
 734 static void igc_setup_mrqc(struct igc_adapter *adapter)
 735 {
 736         struct igc_hw *hw = &adapter->hw;
 737         u32 j, num_rx_queues;
 738         u32 mrqc, rxcsum;
 739         u32 rss_key[10];
 740
 741         netdev_rss_key_fill(rss_key, sizeof(rss_key));
 742         for (j = 0; j < 10; j++)
 743                 wr32(IGC_RSSRK(j), rss_key[j]);
 744
 745         num_rx_queues = adapter->rss_queues;
 746
 747         if (adapter->rss_indir_tbl_init != num_rx_queues) {
 748                 for (j = 0; j < IGC_RETA_SIZE; j++)
 749                         adapter->rss_indir_tbl[j] =
 750                         (j * num_rx_queues) / IGC_RETA_SIZE;
 751                 adapter->rss_indir_tbl_init = num_rx_queues;
 752         }
 753         igc_write_rss_indir_tbl(adapter);
 754
 755         /* Disable raw packet checksumming so that RSS hash is placed in
 756          * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
 757          * offloads as they are enabled by default
 758          */
 759         rxcsum = rd32(IGC_RXCSUM);
 760         rxcsum |= IGC_RXCSUM_PCSD;
 761
 762         /* Enable Receive Checksum Offload for SCTP */
 763         rxcsum |= IGC_RXCSUM_CRCOFL;
 764
 765         /* Don't need to set TUOFL or IPOFL, they default to 1 */
 766         wr32(IGC_RXCSUM, rxcsum);
 767
 768         /* Generate RSS hash based on packet types, TCP/UDP
 769          * port numbers and/or IPv4/v6 src and dst addresses
 770          */
 771         mrqc = IGC_MRQC_RSS_FIELD_IPV4 |
 772                IGC_MRQC_RSS_FIELD_IPV4_TCP |
 773                IGC_MRQC_RSS_FIELD_IPV6 |
 774                IGC_MRQC_RSS_FIELD_IPV6_TCP |
 775                IGC_MRQC_RSS_FIELD_IPV6_TCP_EX;
 776
 777         if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV4_UDP)
 778                 mrqc |= IGC_MRQC_RSS_FIELD_IPV4_UDP;
 779         if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV6_UDP)
 780                 mrqc |= IGC_MRQC_RSS_FIELD_IPV6_UDP;
 781
 782         mrqc |= IGC_MRQC_ENABLE_RSS_MQ;
 783
 784         wr32(IGC_MRQC, mrqc);
 785 }
 786
 787 /**
 788  * igc_setup_rctl - configure the receive control registers
 789  * @adapter: Board private structure
 790  */
 791 static void igc_setup_rctl(struct igc_adapter *adapter)
 792 {
 793         struct igc_hw *hw = &adapter->hw;
 794         u32 rctl;
 795
 796         rctl = rd32(IGC_RCTL);
 797
 798         rctl &= ~(3 << IGC_RCTL_MO_SHIFT);
 799         rctl &= ~(IGC_RCTL_LBM_TCVR | IGC_RCTL_LBM_MAC);
 800
 801         rctl |= IGC_RCTL_EN | IGC_RCTL_BAM | IGC_RCTL_RDMTS_HALF |
 802                 (hw->mac.mc_filter_type << IGC_RCTL_MO_SHIFT);
 803
 804         /* enable stripping of CRC. Newer features require
 805          * that the HW strips the CRC.
 806          */
 807         rctl |= IGC_RCTL_SECRC;
 808
 809         /* disable store bad packets and clear size bits. */
 810         rctl &= ~(IGC_RCTL_SBP | IGC_RCTL_SZ_256);
 811
 812         /* enable LPE to allow for reception of jumbo frames */
 813         rctl |= IGC_RCTL_LPE;
 814
 815         /* disable queue 0 to prevent tail write w/o re-config */
 816         wr32(IGC_RXDCTL(0), 0);
 817
 818         /* This is useful for sniffing bad packets. */
 819         if (adapter->netdev->features & NETIF_F_RXALL) {
 820                 /* UPE and MPE will be handled by normal PROMISC logic
 821                  * in set_rx_mode
 822                  */
 823                 rctl |= (IGC_RCTL_SBP | /* Receive bad packets */
 824                          IGC_RCTL_BAM | /* RX All Bcast Pkts */
 825                          IGC_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
 826
 827                 rctl &= ~(IGC_RCTL_DPF | /* Allow filtered pause */
 828                           IGC_RCTL_CFIEN); /* Disable VLAN CFIEN Filter */
 829         }
 830
 831         wr32(IGC_RCTL, rctl);
 832 }
 833
 834 /**
 835  * igc_setup_tctl - configure the transmit control registers
 836  * @adapter: Board private structure
 837  */
 838 static void igc_setup_tctl(struct igc_adapter *adapter)
 839 {
 840         struct igc_hw *hw = &adapter->hw;
 841         u32 tctl;
 842
 843         /* disable queue 0 which icould be enabled by default */
 844         wr32(IGC_TXDCTL(0), 0);
 845
 846         /* Program the Transmit Control Register */
 847         tctl = rd32(IGC_TCTL);
 848         tctl &= ~IGC_TCTL_CT;
 849         tctl |= IGC_TCTL_PSP | IGC_TCTL_RTLC |
 850                 (IGC_COLLISION_THRESHOLD << IGC_CT_SHIFT);
 851
 852         /* Enable transmits */
 853         tctl |= IGC_TCTL_EN;
 854
 855         wr32(IGC_TCTL, tctl);
 856 }
 857
 858 /**
 859  * igc_set_mac_filter_hw() - Set MAC address filter in hardware
 860  * @adapter: Pointer to adapter where the filter should be set
 861  * @index: Filter index
 862  * @type: MAC address filter type (source or destination)
 863  * @addr: MAC address
 864  * @queue: If non-negative, queue assignment feature is enabled and frames
 865  *         matching the filter are enqueued onto 'queue'. Otherwise, queue
 866  *         assignment is disabled.
 867  */
 868 static void igc_set_mac_filter_hw(struct igc_adapter *adapter, int index,
 869                                   enum igc_mac_filter_type type,
 870                                   const u8 *addr, int queue)
 871 {
 872         struct net_device *dev = adapter->netdev;
 873         struct igc_hw *hw = &adapter->hw;
 874         u32 ral, rah;
 875
 876         if (WARN_ON(index >= hw->mac.rar_entry_count))
 877                 return;
 878
 879         ral = le32_to_cpup((__le32 *)(addr));
 880         rah = le16_to_cpup((__le16 *)(addr + 4));
 881
 882         if (type == IGC_MAC_FILTER_TYPE_SRC) {
 883                 rah &= ~IGC_RAH_ASEL_MASK;
 884                 rah |= IGC_RAH_ASEL_SRC_ADDR;
 885         }
 886
 887         if (queue >= 0) {
 888                 rah &= ~IGC_RAH_QSEL_MASK;
 889                 rah |= (queue << IGC_RAH_QSEL_SHIFT);
 890                 rah |= IGC_RAH_QSEL_ENABLE;
 891         }
 892
 893         rah |= IGC_RAH_AV;
 894
 895         wr32(IGC_RAL(index), ral);
 896         wr32(IGC_RAH(index), rah);
 897
 898         netdev_dbg(dev, "MAC address filter set in HW: index %d", index);
 899 }
 900
 901 /**
 902  * igc_clear_mac_filter_hw() - Clear MAC address filter in hardware
 903  * @adapter: Pointer to adapter where the filter should be cleared
 904  * @index: Filter index
 905  */
 906 static void igc_clear_mac_filter_hw(struct igc_adapter *adapter, int index)
 907 {
 908         struct net_device *dev = adapter->netdev;
 909         struct igc_hw *hw = &adapter->hw;
 910
 911         if (WARN_ON(index >= hw->mac.rar_entry_count))
 912                 return;
 913
 914         wr32(IGC_RAL(index), 0);
 915         wr32(IGC_RAH(index), 0);
 916
 917         netdev_dbg(dev, "MAC address filter cleared in HW: index %d", index);
 918 }
 919
 920 /* Set default MAC address for the PF in the first RAR entry */
 921 static void igc_set_default_mac_filter(struct igc_adapter *adapter)
 922 {
 923         struct net_device *dev = adapter->netdev;
 924         u8 *addr = adapter->hw.mac.addr;
 925
 926         netdev_dbg(dev, "Set default MAC address filter: address %pM", addr);
 927
 928         igc_set_mac_filter_hw(adapter, 0, IGC_MAC_FILTER_TYPE_DST, addr, -1);
 929 }
 930
 931 /**
 932  * igc_set_mac - Change the Ethernet Address of the NIC
 933  * @netdev: network interface device structure
 934  * @p: pointer to an address structure
 935  *
 936  * Returns 0 on success, negative on failure
 937  */
 938 static int igc_set_mac(struct net_device *netdev, void *p)
 939 {
 940         struct igc_adapter *adapter = netdev_priv(netdev);
 941         struct igc_hw *hw = &adapter->hw;
 942         struct sockaddr *addr = p;
 943
 944         if (!is_valid_ether_addr(addr->sa_data))
 945                 return -EADDRNOTAVAIL;
 946
 947         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
 948         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
 949
 950         /* set the correct pool for the new PF MAC address in entry 0 */
 951         igc_set_default_mac_filter(adapter);
 952
 953         return 0;
 954 }
 955
 956 /**
 957  *  igc_write_mc_addr_list - write multicast addresses to MTA
 958  *  @netdev: network interface device structure
 959  *
 960  *  Writes multicast address list to the MTA hash table.
 961  *  Returns: -ENOMEM on failure
 962  *           0 on no addresses written
 963  *           X on writing X addresses to MTA
 964  **/
 965 static int igc_write_mc_addr_list(struct net_device *netdev)
 966 {
 967         struct igc_adapter *adapter = netdev_priv(netdev);
 968         struct igc_hw *hw = &adapter->hw;
 969         struct netdev_hw_addr *ha;
 970         u8  *mta_list;
 971         int i;
 972
 973         if (netdev_mc_empty(netdev)) {
 974                 /* nothing to program, so clear mc list */
 975                 igc_update_mc_addr_list(hw, NULL, 0);
 976                 return 0;
 977         }
 978
 979         mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
 980         if (!mta_list)
 981                 return -ENOMEM;
 982
 983         /* The shared function expects a packed array of only addresses. */
 984         i = 0;
 985         netdev_for_each_mc_addr(ha, netdev)
 986                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
 987
 988         igc_update_mc_addr_list(hw, mta_list, i);
 989         kfree(mta_list);
 990
 991         return netdev_mc_count(netdev);
 992 }
 993
 994 static __le32 igc_tx_launchtime(struct igc_adapter *adapter, ktime_t txtime)
 995 {
 996         ktime_t cycle_time = adapter->cycle_time;
 997         ktime_t base_time = adapter->base_time;
 998         u32 launchtime;
 999
1000         /* FIXME: when using ETF together with taprio, we may have a
1001          * case where 'delta' is larger than the cycle_time, this may
1002          * cause problems if we don't read the current value of
1003          * IGC_BASET, as the value writen into the launchtime
1004          * descriptor field may be misinterpreted.
1005          */
1006         div_s64_rem(ktime_sub_ns(txtime, base_time), cycle_time, &launchtime);
1007
1008         return cpu_to_le32(launchtime);
1009 }
1010
1011 static void igc_tx_ctxtdesc(struct igc_ring *tx_ring,
1012                             struct igc_tx_buffer *first,
1013                             u32 vlan_macip_lens, u32 type_tucmd,
1014                             u32 mss_l4len_idx)
1015 {
1016         struct igc_adv_tx_context_desc *context_desc;
1017         u16 i = tx_ring->next_to_use;
1018
1019         context_desc = IGC_TX_CTXTDESC(tx_ring, i);
1020
1021         i++;
1022         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1023
1024         /* set bits to identify this as an advanced context descriptor */
1025         type_tucmd |= IGC_TXD_CMD_DEXT | IGC_ADVTXD_DTYP_CTXT;
1026
1027         /* For i225, context index must be unique per ring. */
1028         if (test_bit(IGC_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
1029                 mss_l4len_idx |= tx_ring->reg_idx << 4;
1030
1031         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
1032         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
1033         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
1034
1035         /* We assume there is always a valid Tx time available. Invalid times
1036          * should have been handled by the upper layers.
1037          */
1038         if (tx_ring->launchtime_enable) {
1039                 struct igc_adapter *adapter = netdev_priv(tx_ring->netdev);
1040                 ktime_t txtime = first->skb->tstamp;
1041
1042                 skb_txtime_consumed(first->skb);
1043                 context_desc->launch_time = igc_tx_launchtime(adapter,
1044                                                               txtime);
1045         } else {
1046                 context_desc->launch_time = 0;
1047         }
1048 }
1049
1050 static void igc_tx_csum(struct igc_ring *tx_ring, struct igc_tx_buffer *first)
1051 {
1052         struct sk_buff *skb = first->skb;
1053         u32 vlan_macip_lens = 0;
1054         u32 type_tucmd = 0;
1055
1056         if (skb->ip_summed != CHECKSUM_PARTIAL) {
1057 csum_failed:
1058                 if (!(first->tx_flags & IGC_TX_FLAGS_VLAN) &&
1059                     !tx_ring->launchtime_enable)
1060                         return;
1061                 goto no_csum;
1062         }
1063
1064         switch (skb->csum_offset) {
1065         case offsetof(struct tcphdr, check):
1066                 type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP;
1067                 fallthrough;
1068         case offsetof(struct udphdr, check):
1069                 break;
1070         case offsetof(struct sctphdr, checksum):
1071                 /* validate that this is actually an SCTP request */
1072                 if (skb_csum_is_sctp(skb)) {
1073                         type_tucmd = IGC_ADVTXD_TUCMD_L4T_SCTP;
1074                         break;
1075                 }
1076                 fallthrough;
1077         default:
1078                 skb_checksum_help(skb);
1079                 goto csum_failed;
1080         }
1081
1082         /* update TX checksum flag */
1083         first->tx_flags |= IGC_TX_FLAGS_CSUM;
1084         vlan_macip_lens = skb_checksum_start_offset(skb) -
1085                           skb_network_offset(skb);
1086 no_csum:
1087         vlan_macip_lens |= skb_network_offset(skb) << IGC_ADVTXD_MACLEN_SHIFT;
1088         vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK;
1089
1090         igc_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
1091 }
1092
1093 static int __igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size)
1094 {
1095         struct net_device *netdev = tx_ring->netdev;
1096
1097         netif_stop_subqueue(netdev, tx_ring->queue_index);
1098
1099         /* memory barriier comment */
1100         smp_mb();
1101
1102         /* We need to check again in a case another CPU has just
1103          * made room available.
1104          */
1105         if (igc_desc_unused(tx_ring) < size)
1106                 return -EBUSY;
1107
1108         /* A reprieve! */
1109         netif_wake_subqueue(netdev, tx_ring->queue_index);
1110
1111         u64_stats_update_begin(&tx_ring->tx_syncp2);
1112         tx_ring->tx_stats.restart_queue2++;
1113         u64_stats_update_end(&tx_ring->tx_syncp2);
1114
1115         return 0;
1116 }
1117
1118 static inline int igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size)
1119 {
1120         if (igc_desc_unused(tx_ring) >= size)
1121                 return 0;
1122         return __igc_maybe_stop_tx(tx_ring, size);
1123 }
1124
1125 #define IGC_SET_FLAG(_input, _flag, _result) \
1126         (((_flag) <= (_result)) ?                               \
1127          ((u32)((_input) & (_flag)) * ((_result) / (_flag))) :  \
1128          ((u32)((_input) & (_flag)) / ((_flag) / (_result))))
1129
1130 static u32 igc_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
1131 {
1132         /* set type for advanced descriptor with frame checksum insertion */
1133         u32 cmd_type = IGC_ADVTXD_DTYP_DATA |
1134                        IGC_ADVTXD_DCMD_DEXT |
1135                        IGC_ADVTXD_DCMD_IFCS;
1136
1137         /* set HW vlan bit if vlan is present */
1138         cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_VLAN,
1139                                  IGC_ADVTXD_DCMD_VLE);
1140
1141         /* set segmentation bits for TSO */
1142         cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSO,
1143                                  (IGC_ADVTXD_DCMD_TSE));
1144
1145         /* set timestamp bit if present */
1146         cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP,
1147                                  (IGC_ADVTXD_MAC_TSTAMP));
1148
1149         /* insert frame checksum */
1150         cmd_type ^= IGC_SET_FLAG(skb->no_fcs, 1, IGC_ADVTXD_DCMD_IFCS);
1151
1152         return cmd_type;
1153 }
1154
1155 static void igc_tx_olinfo_status(struct igc_ring *tx_ring,
1156                                  union igc_adv_tx_desc *tx_desc,
1157                                  u32 tx_flags, unsigned int paylen)
1158 {
1159         u32 olinfo_status = paylen << IGC_ADVTXD_PAYLEN_SHIFT;
1160
1161         /* insert L4 checksum */
1162         olinfo_status |= (tx_flags & IGC_TX_FLAGS_CSUM) *
1163                           ((IGC_TXD_POPTS_TXSM << 8) /
1164                           IGC_TX_FLAGS_CSUM);
1165
1166         /* insert IPv4 checksum */
1167         olinfo_status |= (tx_flags & IGC_TX_FLAGS_IPV4) *
1168                           (((IGC_TXD_POPTS_IXSM << 8)) /
1169                           IGC_TX_FLAGS_IPV4);
1170
1171         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
1172 }
1173
1174 static int igc_tx_map(struct igc_ring *tx_ring,
1175                       struct igc_tx_buffer *first,
1176                       const u8 hdr_len)
1177 {
1178         struct sk_buff *skb = first->skb;
1179         struct igc_tx_buffer *tx_buffer;
1180         union igc_adv_tx_desc *tx_desc;
1181         u32 tx_flags = first->tx_flags;
1182         skb_frag_t *frag;
1183         u16 i = tx_ring->next_to_use;
1184         unsigned int data_len, size;
1185         dma_addr_t dma;
1186         u32 cmd_type;
1187
1188         cmd_type = igc_tx_cmd_type(skb, tx_flags);
1189         tx_desc = IGC_TX_DESC(tx_ring, i);
1190
1191         igc_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
1192
1193         size = skb_headlen(skb);
1194         data_len = skb->data_len;
1195
1196         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1197
1198         tx_buffer = first;
1199
1200         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1201                 if (dma_mapping_error(tx_ring->dev, dma))
1202                         goto dma_error;
1203
1204                 /* record length, and DMA address */
1205                 dma_unmap_len_set(tx_buffer, len, size);
1206                 dma_unmap_addr_set(tx_buffer, dma, dma);
1207
1208                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
1209
1210                 while (unlikely(size > IGC_MAX_DATA_PER_TXD)) {
1211                         tx_desc->read.cmd_type_len =
1212                                 cpu_to_le32(cmd_type ^ IGC_MAX_DATA_PER_TXD);
1213
1214                         i++;
1215                         tx_desc++;
1216                         if (i == tx_ring->count) {
1217                                 tx_desc = IGC_TX_DESC(tx_ring, 0);
1218                                 i = 0;
1219                         }
1220                         tx_desc->read.olinfo_status = 0;
1221
1222                         dma += IGC_MAX_DATA_PER_TXD;
1223                         size -= IGC_MAX_DATA_PER_TXD;
1224
1225                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
1226                 }
1227
1228                 if (likely(!data_len))
1229                         break;
1230
1231                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
1232
1233                 i++;
1234                 tx_desc++;
1235                 if (i == tx_ring->count) {
1236                         tx_desc = IGC_TX_DESC(tx_ring, 0);
1237                         i = 0;
1238                 }
1239                 tx_desc->read.olinfo_status = 0;
1240
1241                 size = skb_frag_size(frag);
1242                 data_len -= size;
1243
1244                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
1245                                        size, DMA_TO_DEVICE);
1246
1247                 tx_buffer = &tx_ring->tx_buffer_info[i];
1248         }
1249
1250         /* write last descriptor with RS and EOP bits */
1251         cmd_type |= size | IGC_TXD_DCMD;
1252         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1253
1254         netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
1255
1256         /* set the timestamp */
1257         first->time_stamp = jiffies;
1258
1259         skb_tx_timestamp(skb);
1260
1261         /* Force memory writes to complete before letting h/w know there
1262          * are new descriptors to fetch.  (Only applicable for weak-ordered
1263          * memory model archs, such as IA-64).
1264          *
1265          * We also need this memory barrier to make certain all of the
1266          * status bits have been updated before next_to_watch is written.
1267          */
1268         wmb();
1269
1270         /* set next_to_watch value indicating a packet is present */
1271         first->next_to_watch = tx_desc;
1272
1273         i++;
1274         if (i == tx_ring->count)
1275                 i = 0;
1276
1277         tx_ring->next_to_use = i;
1278
1279         /* Make sure there is space in the ring for the next send. */
1280         igc_maybe_stop_tx(tx_ring, DESC_NEEDED);
1281
1282         if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
1283                 writel(i, tx_ring->tail);
1284         }
1285
1286         return 0;
1287 dma_error:
1288         netdev_err(tx_ring->netdev, "TX DMA map failed\n");
1289         tx_buffer = &tx_ring->tx_buffer_info[i];
1290
1291         /* clear dma mappings for failed tx_buffer_info map */
1292         while (tx_buffer != first) {
1293                 if (dma_unmap_len(tx_buffer, len))
1294                         igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
1295
1296                 if (i-- == 0)
1297                         i += tx_ring->count;
1298                 tx_buffer = &tx_ring->tx_buffer_info[i];
1299         }
1300
1301         if (dma_unmap_len(tx_buffer, len))
1302                 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
1303
1304         dev_kfree_skb_any(tx_buffer->skb);
1305         tx_buffer->skb = NULL;
1306
1307         tx_ring->next_to_use = i;
1308
1309         return -1;
1310 }
1311
1312 static int igc_tso(struct igc_ring *tx_ring,
1313                    struct igc_tx_buffer *first,
1314                    u8 *hdr_len)
1315 {
1316         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
1317         struct sk_buff *skb = first->skb;
1318         union {
1319                 struct iphdr *v4;
1320                 struct ipv6hdr *v6;
1321                 unsigned char *hdr;
1322         } ip;
1323         union {
1324                 struct tcphdr *tcp;
1325                 struct udphdr *udp;
1326                 unsigned char *hdr;
1327         } l4;
1328         u32 paylen, l4_offset;
1329         int err;
1330
1331         if (skb->ip_summed != CHECKSUM_PARTIAL)
1332                 return 0;
1333
1334         if (!skb_is_gso(skb))
1335                 return 0;
1336
1337         err = skb_cow_head(skb, 0);
1338         if (err < 0)
1339                 return err;
1340
1341         ip.hdr = skb_network_header(skb);
1342         l4.hdr = skb_checksum_start(skb);
1343
1344         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1345         type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP;
1346
1347         /* initialize outer IP header fields */
1348         if (ip.v4->version == 4) {
1349                 unsigned char *csum_start = skb_checksum_start(skb);
1350                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
1351
1352                 /* IP header will have to cancel out any data that
1353                  * is not a part of the outer IP header
1354                  */
1355                 ip.v4->check = csum_fold(csum_partial(trans_start,
1356                                                       csum_start - trans_start,
1357                                                       0));
1358                 type_tucmd |= IGC_ADVTXD_TUCMD_IPV4;
1359
1360                 ip.v4->tot_len = 0;
1361                 first->tx_flags |= IGC_TX_FLAGS_TSO |
1362                                    IGC_TX_FLAGS_CSUM |
1363                                    IGC_TX_FLAGS_IPV4;
1364         } else {
1365                 ip.v6->payload_len = 0;
1366                 first->tx_flags |= IGC_TX_FLAGS_TSO |
1367                                    IGC_TX_FLAGS_CSUM;
1368         }
1369
1370         /* determine offset of inner transport header */
1371         l4_offset = l4.hdr - skb->data;
1372
1373         /* remove payload length from inner checksum */
1374         paylen = skb->len - l4_offset;
1375         if (type_tucmd & IGC_ADVTXD_TUCMD_L4T_TCP) {
1376                 /* compute length of segmentation header */
1377                 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
1378                 csum_replace_by_diff(&l4.tcp->check,
1379                                      (__force __wsum)htonl(paylen));
1380         } else {
1381                 /* compute length of segmentation header */
1382                 *hdr_len = sizeof(*l4.udp) + l4_offset;
1383                 csum_replace_by_diff(&l4.udp->check,
1384                                      (__force __wsum)htonl(paylen));
1385         }
1386
1387         /* update gso size and bytecount with header size */
1388         first->gso_segs = skb_shinfo(skb)->gso_segs;
1389         first->bytecount += (first->gso_segs - 1) * *hdr_len;
1390
1391         /* MSS L4LEN IDX */
1392         mss_l4len_idx = (*hdr_len - l4_offset) << IGC_ADVTXD_L4LEN_SHIFT;
1393         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IGC_ADVTXD_MSS_SHIFT;
1394
1395         /* VLAN MACLEN IPLEN */
1396         vlan_macip_lens = l4.hdr - ip.hdr;
1397         vlan_macip_lens |= (ip.hdr - skb->data) << IGC_ADVTXD_MACLEN_SHIFT;
1398         vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK;
1399
1400         igc_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
1401                         type_tucmd, mss_l4len_idx);
1402
1403         return 1;
1404 }
1405
1406 static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,
1407                                        struct igc_ring *tx_ring)
1408 {
1409         u16 count = TXD_USE_COUNT(skb_headlen(skb));
1410         __be16 protocol = vlan_get_protocol(skb);
1411         struct igc_tx_buffer *first;
1412         u32 tx_flags = 0;
1413         unsigned short f;
1414         u8 hdr_len = 0;
1415         int tso = 0;
1416
1417         /* need: 1 descriptor per page * PAGE_SIZE/IGC_MAX_DATA_PER_TXD,
1418          *      + 1 desc for skb_headlen/IGC_MAX_DATA_PER_TXD,
1419          *      + 2 desc gap to keep tail from touching head,
1420          *      + 1 desc for context descriptor,
1421          * otherwise try next time
1422          */
1423         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1424                 count += TXD_USE_COUNT(skb_frag_size(
1425                                                 &skb_shinfo(skb)->frags[f]));
1426
1427         if (igc_maybe_stop_tx(tx_ring, count + 3)) {
1428                 /* this is a hard error */
1429                 return NETDEV_TX_BUSY;
1430         }
1431
1432         /* record the location of the first descriptor for this packet */
1433         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
1434         first->type = IGC_TX_BUFFER_TYPE_SKB;
1435         first->skb = skb;
1436         first->bytecount = skb->len;
1437         first->gso_segs = 1;
1438
1439         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
1440                 struct igc_adapter *adapter = netdev_priv(tx_ring->netdev);
1441
1442                 /* FIXME: add support for retrieving timestamps from
1443                  * the other timer registers before skipping the
1444                  * timestamping request.
1445                  */
1446                 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
1447                     !test_and_set_bit_lock(__IGC_PTP_TX_IN_PROGRESS,
1448                                            &adapter->state)) {
1449                         skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
1450                         tx_flags |= IGC_TX_FLAGS_TSTAMP;
1451
1452                         adapter->ptp_tx_skb = skb_get(skb);
1453                         adapter->ptp_tx_start = jiffies;
1454                 } else {
1455                         adapter->tx_hwtstamp_skipped++;
1456                 }
1457         }
1458
1459         if (skb_vlan_tag_present(skb)) {
1460                 tx_flags |= IGC_TX_FLAGS_VLAN;
1461                 tx_flags |= (skb_vlan_tag_get(skb) << IGC_TX_FLAGS_VLAN_SHIFT);
1462         }
1463
1464         /* record initial flags and protocol */
1465         first->tx_flags = tx_flags;
1466         first->protocol = protocol;
1467
1468         tso = igc_tso(tx_ring, first, &hdr_len);
1469         if (tso < 0)
1470                 goto out_drop;
1471         else if (!tso)
1472                 igc_tx_csum(tx_ring, first);
1473
1474         igc_tx_map(tx_ring, first, hdr_len);
1475
1476         return NETDEV_TX_OK;
1477
1478 out_drop:
1479         dev_kfree_skb_any(first->skb);
1480         first->skb = NULL;
1481
1482         return NETDEV_TX_OK;
1483 }
1484
1485 static inline struct igc_ring *igc_tx_queue_mapping(struct igc_adapter *adapter,
1486                                                     struct sk_buff *skb)
1487 {
1488         unsigned int r_idx = skb->queue_mapping;
1489
1490         if (r_idx >= adapter->num_tx_queues)
1491                 r_idx = r_idx % adapter->num_tx_queues;
1492
1493         return adapter->tx_ring[r_idx];
1494 }
1495
1496 static netdev_tx_t igc_xmit_frame(struct sk_buff *skb,
1497                                   struct net_device *netdev)
1498 {
1499         struct igc_adapter *adapter = netdev_priv(netdev);
1500
1501         /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
1502          * in order to meet this minimum size requirement.
1503          */
1504         if (skb->len < 17) {
1505                 if (skb_padto(skb, 17))
1506                         return NETDEV_TX_OK;
1507                 skb->len = 17;
1508         }
1509
1510         return igc_xmit_frame_ring(skb, igc_tx_queue_mapping(adapter, skb));
1511 }
1512
1513 static void igc_rx_checksum(struct igc_ring *ring,
1514                             union igc_adv_rx_desc *rx_desc,
1515                             struct sk_buff *skb)
1516 {
1517         skb_checksum_none_assert(skb);
1518
1519         /* Ignore Checksum bit is set */
1520         if (igc_test_staterr(rx_desc, IGC_RXD_STAT_IXSM))
1521                 return;
1522
1523         /* Rx checksum disabled via ethtool */
1524         if (!(ring->netdev->features & NETIF_F_RXCSUM))
1525                 return;
1526
1527         /* TCP/UDP checksum error bit is set */
1528         if (igc_test_staterr(rx_desc,
1529                              IGC_RXDEXT_STATERR_L4E |
1530                              IGC_RXDEXT_STATERR_IPE)) {
1531                 /* work around errata with sctp packets where the TCPE aka
1532                  * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
1533                  * packets (aka let the stack check the crc32c)
1534                  */
1535                 if (!(skb->len == 60 &&
1536                       test_bit(IGC_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
1537                         u64_stats_update_begin(&ring->rx_syncp);
1538                         ring->rx_stats.csum_err++;
1539                         u64_stats_update_end(&ring->rx_syncp);
1540                 }
1541                 /* let the stack verify checksum errors */
1542                 return;
1543         }
1544         /* It must be a TCP or UDP packet with a valid checksum */
1545         if (igc_test_staterr(rx_desc, IGC_RXD_STAT_TCPCS |
1546                                       IGC_RXD_STAT_UDPCS))
1547                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1548
1549         netdev_dbg(ring->netdev, "cksum success: bits %08X\n",
1550                    le32_to_cpu(rx_desc->wb.upper.status_error));
1551 }
1552
1553 static inline void igc_rx_hash(struct igc_ring *ring,
1554                                union igc_adv_rx_desc *rx_desc,
1555                                struct sk_buff *skb)
1556 {
1557         if (ring->netdev->features & NETIF_F_RXHASH)
1558                 skb_set_hash(skb,
1559                              le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1560                              PKT_HASH_TYPE_L3);
1561 }
1562
1563 static void igc_rx_vlan(struct igc_ring *rx_ring,
1564                         union igc_adv_rx_desc *rx_desc,
1565                         struct sk_buff *skb)
1566 {
1567         struct net_device *dev = rx_ring->netdev;
1568         u16 vid;
1569
1570         if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1571             igc_test_staterr(rx_desc, IGC_RXD_STAT_VP)) {
1572                 if (igc_test_staterr(rx_desc, IGC_RXDEXT_STATERR_LB) &&
1573                     test_bit(IGC_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
1574                         vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan);
1575                 else
1576                         vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1577
1578                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1579         }
1580 }
1581
1582 /**
1583  * igc_process_skb_fields - Populate skb header fields from Rx descriptor
1584  * @rx_ring: rx descriptor ring packet is being transacted on
1585  * @rx_desc: pointer to the EOP Rx descriptor
1586  * @skb: pointer to current skb being populated
1587  *
1588  * This function checks the ring, descriptor, and packet information in order
1589  * to populate the hash, checksum, VLAN, protocol, and other fields within the
1590  * skb.
1591  */
1592 static void igc_process_skb_fields(struct igc_ring *rx_ring,
1593                                    union igc_adv_rx_desc *rx_desc,
1594                                    struct sk_buff *skb)
1595 {
1596         igc_rx_hash(rx_ring, rx_desc, skb);
1597
1598         igc_rx_checksum(rx_ring, rx_desc, skb);
1599
1600         igc_rx_vlan(rx_ring, rx_desc, skb);
1601
1602         skb_record_rx_queue(skb, rx_ring->queue_index);
1603
1604         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1605 }
1606
1607 static void igc_vlan_mode(struct net_device *netdev, netdev_features_t features)
1608 {
1609         bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1610         struct igc_adapter *adapter = netdev_priv(netdev);
1611         struct igc_hw *hw = &adapter->hw;
1612         u32 ctrl;
1613
1614         ctrl = rd32(IGC_CTRL);
1615
1616         if (enable) {
1617                 /* enable VLAN tag insert/strip */
1618                 ctrl |= IGC_CTRL_VME;
1619         } else {
1620                 /* disable VLAN tag insert/strip */
1621                 ctrl &= ~IGC_CTRL_VME;
1622         }
1623         wr32(IGC_CTRL, ctrl);
1624 }
1625
1626 static void igc_restore_vlan(struct igc_adapter *adapter)
1627 {
1628         igc_vlan_mode(adapter->netdev, adapter->netdev->features);
1629 }
1630
1631 static struct igc_rx_buffer *igc_get_rx_buffer(struct igc_ring *rx_ring,
1632                                                const unsigned int size,
1633                                                int *rx_buffer_pgcnt)
1634 {
1635         struct igc_rx_buffer *rx_buffer;
1636
1637         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1638         *rx_buffer_pgcnt =
1639 #if (PAGE_SIZE < 8192)
1640                 page_count(rx_buffer->page);
1641 #else
1642                 0;
1643 #endif
1644         prefetchw(rx_buffer->page);
1645
1646         /* we are reusing so sync this buffer for CPU use */
1647         dma_sync_single_range_for_cpu(rx_ring->dev,
1648                                       rx_buffer->dma,
1649                                       rx_buffer->page_offset,
1650                                       size,
1651                                       DMA_FROM_DEVICE);
1652
1653         rx_buffer->pagecnt_bias--;
1654
1655         return rx_buffer;
1656 }
1657
1658 static void igc_rx_buffer_flip(struct igc_rx_buffer *buffer,
1659                                unsigned int truesize)
1660 {
1661 #if (PAGE_SIZE < 8192)
1662         buffer->page_offset ^= truesize;
1663 #else
1664         buffer->page_offset += truesize;
1665 #endif
1666 }
1667
1668 static unsigned int igc_get_rx_frame_truesize(struct igc_ring *ring,
1669                                               unsigned int size)
1670 {
1671         unsigned int truesize;
1672
1673 #if (PAGE_SIZE < 8192)
1674         truesize = igc_rx_pg_size(ring) / 2;
1675 #else
1676         truesize = ring_uses_build_skb(ring) ?
1677                    SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
1678                    SKB_DATA_ALIGN(IGC_SKB_PAD + size) :
1679                    SKB_DATA_ALIGN(size);
1680 #endif
1681         return truesize;
1682 }
1683
1684 /**
1685  * igc_add_rx_frag - Add contents of Rx buffer to sk_buff
1686  * @rx_ring: rx descriptor ring to transact packets on
1687  * @rx_buffer: buffer containing page to add
1688  * @skb: sk_buff to place the data into
1689  * @size: size of buffer to be added
1690  *
1691  * This function will add the data contained in rx_buffer->page to the skb.
1692  */
1693 static void igc_add_rx_frag(struct igc_ring *rx_ring,
1694                             struct igc_rx_buffer *rx_buffer,
1695                             struct sk_buff *skb,
1696                             unsigned int size)
1697 {
1698         unsigned int truesize;
1699
1700 #if (PAGE_SIZE < 8192)
1701         truesize = igc_rx_pg_size(rx_ring) / 2;
1702 #else
1703         truesize = ring_uses_build_skb(rx_ring) ?
1704                    SKB_DATA_ALIGN(IGC_SKB_PAD + size) :
1705                    SKB_DATA_ALIGN(size);
1706 #endif
1707         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1708                         rx_buffer->page_offset, size, truesize);
1709
1710         igc_rx_buffer_flip(rx_buffer, truesize);
1711 }
1712
1713 static struct sk_buff *igc_build_skb(struct igc_ring *rx_ring,
1714                                      struct igc_rx_buffer *rx_buffer,
1715                                      union igc_adv_rx_desc *rx_desc,
1716                                      unsigned int size)
1717 {
1718         void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
1719         unsigned int truesize = igc_get_rx_frame_truesize(rx_ring, size);
1720         struct sk_buff *skb;
1721
1722         /* prefetch first cache line of first page */
1723         net_prefetch(va);
1724
1725         /* build an skb around the page buffer */
1726         skb = build_skb(va - IGC_SKB_PAD, truesize);
1727         if (unlikely(!skb))
1728                 return NULL;
1729
1730         /* update pointers within the skb to store the data */
1731         skb_reserve(skb, IGC_SKB_PAD);
1732         __skb_put(skb, size);
1733
1734         igc_rx_buffer_flip(rx_buffer, truesize);
1735         return skb;
1736 }
1737
1738 static struct sk_buff *igc_construct_skb(struct igc_ring *rx_ring,
1739                                          struct igc_rx_buffer *rx_buffer,
1740                                          struct xdp_buff *xdp,
1741                                          ktime_t timestamp)
1742 {
1743         unsigned int size = xdp->data_end - xdp->data;
1744         unsigned int truesize = igc_get_rx_frame_truesize(rx_ring, size);
1745         void *va = xdp->data;
1746         unsigned int headlen;
1747         struct sk_buff *skb;
1748
1749         /* prefetch first cache line of first page */
1750         net_prefetch(va);
1751
1752         /* allocate a skb to store the frags */
1753         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGC_RX_HDR_LEN);
1754         if (unlikely(!skb))
1755                 return NULL;
1756
1757         if (timestamp)
1758                 skb_hwtstamps(skb)->hwtstamp = timestamp;
1759
1760         /* Determine available headroom for copy */
1761         headlen = size;
1762         if (headlen > IGC_RX_HDR_LEN)
1763                 headlen = eth_get_headlen(skb->dev, va, IGC_RX_HDR_LEN);
1764
1765         /* align pull length to size of long to optimize memcpy performance */
1766         memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
1767
1768         /* update all of the pointers */
1769         size -= headlen;
1770         if (size) {
1771                 skb_add_rx_frag(skb, 0, rx_buffer->page,
1772                                 (va + headlen) - page_address(rx_buffer->page),
1773                                 size, truesize);
1774                 igc_rx_buffer_flip(rx_buffer, truesize);
1775         } else {
1776                 rx_buffer->pagecnt_bias++;
1777         }
1778
1779         return skb;
1780 }
1781
1782 /**
1783  * igc_reuse_rx_page - page flip buffer and store it back on the ring
1784  * @rx_ring: rx descriptor ring to store buffers on
1785  * @old_buff: donor buffer to have page reused
1786  *
1787  * Synchronizes page for reuse by the adapter
1788  */
1789 static void igc_reuse_rx_page(struct igc_ring *rx_ring,
1790                               struct igc_rx_buffer *old_buff)
1791 {
1792         u16 nta = rx_ring->next_to_alloc;
1793         struct igc_rx_buffer *new_buff;
1794
1795         new_buff = &rx_ring->rx_buffer_info[nta];
1796
1797         /* update, and store next to alloc */
1798         nta++;
1799         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1800
1801         /* Transfer page from old buffer to new buffer.
1802          * Move each member individually to avoid possible store
1803          * forwarding stalls.
1804          */
1805         new_buff->dma           = old_buff->dma;
1806         new_buff->page          = old_buff->page;
1807         new_buff->page_offset   = old_buff->page_offset;
1808         new_buff->pagecnt_bias  = old_buff->pagecnt_bias;
1809 }
1810
1811 static bool igc_can_reuse_rx_page(struct igc_rx_buffer *rx_buffer,
1812                                   int rx_buffer_pgcnt)
1813 {
1814         unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1815         struct page *page = rx_buffer->page;
1816
1817         /* avoid re-using remote and pfmemalloc pages */
1818         if (!dev_page_is_reusable(page))
1819                 return false;
1820
1821 #if (PAGE_SIZE < 8192)
1822         /* if we are only owner of page we can reuse it */
1823         if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1))
1824                 return false;
1825 #else
1826 #define IGC_LAST_OFFSET \
1827         (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGC_RXBUFFER_2048)
1828
1829         if (rx_buffer->page_offset > IGC_LAST_OFFSET)
1830                 return false;
1831 #endif
1832
1833         /* If we have drained the page fragment pool we need to update
1834          * the pagecnt_bias and page count so that we fully restock the
1835          * number of references the driver holds.
1836          */
1837         if (unlikely(pagecnt_bias == 1)) {
1838                 page_ref_add(page, USHRT_MAX - 1);
1839                 rx_buffer->pagecnt_bias = USHRT_MAX;
1840         }
1841
1842         return true;
1843 }
1844
1845 /**
1846  * igc_is_non_eop - process handling of non-EOP buffers
1847  * @rx_ring: Rx ring being processed
1848  * @rx_desc: Rx descriptor for current buffer
1849  *
1850  * This function updates next to clean.  If the buffer is an EOP buffer
1851  * this function exits returning false, otherwise it will place the
1852  * sk_buff in the next buffer to be chained and return true indicating
1853  * that this is in fact a non-EOP buffer.
1854  */
1855 static bool igc_is_non_eop(struct igc_ring *rx_ring,
1856                            union igc_adv_rx_desc *rx_desc)
1857 {
1858         u32 ntc = rx_ring->next_to_clean + 1;
1859
1860         /* fetch, update, and store next to clean */
1861         ntc = (ntc < rx_ring->count) ? ntc : 0;
1862         rx_ring->next_to_clean = ntc;
1863
1864         prefetch(IGC_RX_DESC(rx_ring, ntc));
1865
1866         if (likely(igc_test_staterr(rx_desc, IGC_RXD_STAT_EOP)))
1867                 return false;
1868
1869         return true;
1870 }
1871
1872 /**
1873  * igc_cleanup_headers - Correct corrupted or empty headers
1874  * @rx_ring: rx descriptor ring packet is being transacted on
1875  * @rx_desc: pointer to the EOP Rx descriptor
1876  * @skb: pointer to current skb being fixed
1877  *
1878  * Address the case where we are pulling data in on pages only
1879  * and as such no data is present in the skb header.
1880  *
1881  * In addition if skb is not at least 60 bytes we need to pad it so that
1882  * it is large enough to qualify as a valid Ethernet frame.
1883  *
1884  * Returns true if an error was encountered and skb was freed.
1885  */
1886 static bool igc_cleanup_headers(struct igc_ring *rx_ring,
1887                                 union igc_adv_rx_desc *rx_desc,
1888                                 struct sk_buff *skb)
1889 {
1890         /* XDP packets use error pointer so abort at this point */
1891         if (IS_ERR(skb))
1892                 return true;
1893
1894         if (unlikely(igc_test_staterr(rx_desc, IGC_RXDEXT_STATERR_RXE))) {
1895                 struct net_device *netdev = rx_ring->netdev;
1896
1897                 if (!(netdev->features & NETIF_F_RXALL)) {
1898                         dev_kfree_skb_any(skb);
1899                         return true;
1900                 }
1901         }
1902
1903         /* if eth_skb_pad returns an error the skb was freed */
1904         if (eth_skb_pad(skb))
1905                 return true;
1906
1907         return false;
1908 }
1909
1910 static void igc_put_rx_buffer(struct igc_ring *rx_ring,
1911                               struct igc_rx_buffer *rx_buffer,
1912                               int rx_buffer_pgcnt)
1913 {
1914         if (igc_can_reuse_rx_page(rx_buffer, rx_buffer_pgcnt)) {
1915                 /* hand second half of page back to the ring */
1916                 igc_reuse_rx_page(rx_ring, rx_buffer);
1917         } else {
1918                 /* We are not reusing the buffer so unmap it and free
1919                  * any references we are holding to it
1920                  */
1921                 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
1922                                      igc_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
1923                                      IGC_RX_DMA_ATTR);
1924                 __page_frag_cache_drain(rx_buffer->page,
1925                                         rx_buffer->pagecnt_bias);
1926         }
1927
1928         /* clear contents of rx_buffer */
1929         rx_buffer->page = NULL;
1930 }
1931
1932 static inline unsigned int igc_rx_offset(struct igc_ring *rx_ring)
1933 {
1934         struct igc_adapter *adapter = rx_ring->q_vector->adapter;
1935
1936         if (ring_uses_build_skb(rx_ring))
1937                 return IGC_SKB_PAD;
1938         if (igc_xdp_is_enabled(adapter))
1939                 return XDP_PACKET_HEADROOM;
1940
1941         return 0;
1942 }
1943
1944 static bool igc_alloc_mapped_page(struct igc_ring *rx_ring,
1945                                   struct igc_rx_buffer *bi)
1946 {
1947         struct page *page = bi->page;
1948         dma_addr_t dma;
1949
1950         /* since we are recycling buffers we should seldom need to alloc */
1951         if (likely(page))
1952                 return true;
1953
1954         /* alloc new page for storage */
1955         page = dev_alloc_pages(igc_rx_pg_order(rx_ring));
1956         if (unlikely(!page)) {
1957                 rx_ring->rx_stats.alloc_failed++;
1958                 return false;
1959         }
1960
1961         /* map page for use */
1962         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1963                                  igc_rx_pg_size(rx_ring),
1964                                  DMA_FROM_DEVICE,
1965                                  IGC_RX_DMA_ATTR);
1966
1967         /* if mapping failed free memory back to system since
1968          * there isn't much point in holding memory we can't use
1969          */
1970         if (dma_mapping_error(rx_ring->dev, dma)) {
1971                 __free_page(page);
1972
1973                 rx_ring->rx_stats.alloc_failed++;
1974                 return false;
1975         }
1976
1977         bi->dma = dma;
1978         bi->page = page;
1979         bi->page_offset = igc_rx_offset(rx_ring);
1980         page_ref_add(page, USHRT_MAX - 1);
1981         bi->pagecnt_bias = USHRT_MAX;
1982
1983         return true;
1984 }
1985
1986 /**
1987  * igc_alloc_rx_buffers - Replace used receive buffers; packet split
1988  * @rx_ring: rx descriptor ring
1989  * @cleaned_count: number of buffers to clean
1990  */
1991 static void igc_alloc_rx_buffers(struct igc_ring *rx_ring, u16 cleaned_count)
1992 {
1993         union igc_adv_rx_desc *rx_desc;
1994         u16 i = rx_ring->next_to_use;
1995         struct igc_rx_buffer *bi;
1996         u16 bufsz;
1997
1998         /* nothing to do */
1999         if (!cleaned_count)
2000                 return;
2001
2002         rx_desc = IGC_RX_DESC(rx_ring, i);
2003         bi = &rx_ring->rx_buffer_info[i];
2004         i -= rx_ring->count;
2005
2006         bufsz = igc_rx_bufsz(rx_ring);
2007
2008         do {
2009                 if (!igc_alloc_mapped_page(rx_ring, bi))
2010                         break;
2011
2012                 /* sync the buffer for use by the device */
2013                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
2014                                                  bi->page_offset, bufsz,
2015                                                  DMA_FROM_DEVICE);
2016
2017                 /* Refresh the desc even if buffer_addrs didn't change
2018                  * because each write-back erases this info.
2019                  */
2020                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
2021
2022                 rx_desc++;
2023                 bi++;
2024                 i++;
2025                 if (unlikely(!i)) {
2026                         rx_desc = IGC_RX_DESC(rx_ring, 0);
2027                         bi = rx_ring->rx_buffer_info;
2028                         i -= rx_ring->count;
2029                 }
2030
2031                 /* clear the length for the next_to_use descriptor */
2032                 rx_desc->wb.upper.length = 0;
2033
2034                 cleaned_count--;
2035         } while (cleaned_count);
2036
2037         i += rx_ring->count;
2038
2039         if (rx_ring->next_to_use != i) {
2040                 /* record the next descriptor to use */
2041                 rx_ring->next_to_use = i;
2042
2043                 /* update next to alloc since we have filled the ring */
2044                 rx_ring->next_to_alloc = i;
2045
2046                 /* Force memory writes to complete before letting h/w
2047                  * know there are new descriptors to fetch.  (Only
2048                  * applicable for weak-ordered memory model archs,
2049                  * such as IA-64).
2050                  */
2051                 wmb();
2052                 writel(i, rx_ring->tail);
2053         }
2054 }
2055
2056 static bool igc_alloc_rx_buffers_zc(struct igc_ring *ring, u16 count)
2057 {
2058         union igc_adv_rx_desc *desc;
2059         u16 i = ring->next_to_use;
2060         struct igc_rx_buffer *bi;
2061         dma_addr_t dma;
2062         bool ok = true;
2063
2064         if (!count)
2065                 return ok;
2066
2067         desc = IGC_RX_DESC(ring, i);
2068         bi = &ring->rx_buffer_info[i];
2069         i -= ring->count;
2070
2071         do {
2072                 bi->xdp = xsk_buff_alloc(ring->xsk_pool);
2073                 if (!bi->xdp) {
2074                         ok = false;
2075                         break;
2076                 }
2077
2078                 dma = xsk_buff_xdp_get_dma(bi->xdp);
2079                 desc->read.pkt_addr = cpu_to_le64(dma);
2080
2081                 desc++;
2082                 bi++;
2083                 i++;
2084                 if (unlikely(!i)) {
2085                         desc = IGC_RX_DESC(ring, 0);
2086                         bi = ring->rx_buffer_info;
2087                         i -= ring->count;
2088                 }
2089
2090                 /* Clear the length for the next_to_use descriptor. */
2091                 desc->wb.upper.length = 0;
2092
2093                 count--;
2094         } while (count);
2095
2096         i += ring->count;
2097
2098         if (ring->next_to_use != i) {
2099                 ring->next_to_use = i;
2100
2101                 /* Force memory writes to complete before letting h/w
2102                  * know there are new descriptors to fetch.  (Only
2103                  * applicable for weak-ordered memory model archs,
2104                  * such as IA-64).
2105                  */
2106                 wmb();
2107                 writel(i, ring->tail);
2108         }
2109
2110         return ok;
2111 }
2112
2113 static int igc_xdp_init_tx_buffer(struct igc_tx_buffer *buffer,
2114                                   struct xdp_frame *xdpf,
2115                                   struct igc_ring *ring)
2116 {
2117         dma_addr_t dma;
2118
2119         dma = dma_map_single(ring->dev, xdpf->data, xdpf->len, DMA_TO_DEVICE);
2120         if (dma_mapping_error(ring->dev, dma)) {
2121                 netdev_err_once(ring->netdev, "Failed to map DMA for TX\n");
2122                 return -ENOMEM;
2123         }
2124
2125         buffer->type = IGC_TX_BUFFER_TYPE_XDP;
2126         buffer->xdpf = xdpf;
2127         buffer->protocol = 0;
2128         buffer->bytecount = xdpf->len;
2129         buffer->gso_segs = 1;
2130         buffer->time_stamp = jiffies;
2131         dma_unmap_len_set(buffer, len, xdpf->len);
2132         dma_unmap_addr_set(buffer, dma, dma);
2133         return 0;
2134 }
2135
2136 /* This function requires __netif_tx_lock is held by the caller. */
2137 static int igc_xdp_init_tx_descriptor(struct igc_ring *ring,
2138                                       struct xdp_frame *xdpf)
2139 {
2140         struct igc_tx_buffer *buffer;
2141         union igc_adv_tx_desc *desc;
2142         u32 cmd_type, olinfo_status;
2143         int err;
2144
2145         if (!igc_desc_unused(ring))
2146                 return -EBUSY;
2147
2148         buffer = &ring->tx_buffer_info[ring->next_to_use];
2149         err = igc_xdp_init_tx_buffer(buffer, xdpf, ring);
2150         if (err)
2151                 return err;
2152
2153         cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
2154                    IGC_ADVTXD_DCMD_IFCS | IGC_TXD_DCMD |
2155                    buffer->bytecount;
2156         olinfo_status = buffer->bytecount << IGC_ADVTXD_PAYLEN_SHIFT;
2157
2158         desc = IGC_TX_DESC(ring, ring->next_to_use);
2159         desc->read.cmd_type_len = cpu_to_le32(cmd_type);
2160         desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2161         desc->read.buffer_addr = cpu_to_le64(dma_unmap_addr(buffer, dma));
2162
2163         netdev_tx_sent_queue(txring_txq(ring), buffer->bytecount);
2164
2165         buffer->next_to_watch = desc;
2166
2167         ring->next_to_use++;
2168         if (ring->next_to_use == ring->count)
2169                 ring->next_to_use = 0;
2170
2171         return 0;
2172 }
2173
2174 static struct igc_ring *igc_xdp_get_tx_ring(struct igc_adapter *adapter,
2175                                             int cpu)
2176 {
2177         int index = cpu;
2178
2179         if (unlikely(index < 0))
2180                 index = 0;
2181
2182         while (index >= adapter->num_tx_queues)
2183                 index -= adapter->num_tx_queues;
2184
2185         return adapter->tx_ring[index];
2186 }
2187
2188 static int igc_xdp_xmit_back(struct igc_adapter *adapter, struct xdp_buff *xdp)
2189 {
2190         struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2191         int cpu = smp_processor_id();
2192         struct netdev_queue *nq;
2193         struct igc_ring *ring;
2194         int res;
2195
2196         if (unlikely(!xdpf))
2197                 return -EFAULT;
2198
2199         ring = igc_xdp_get_tx_ring(adapter, cpu);
2200         nq = txring_txq(ring);
2201
2202         __netif_tx_lock(nq, cpu);
2203         res = igc_xdp_init_tx_descriptor(ring, xdpf);
2204         __netif_tx_unlock(nq);
2205         return res;
2206 }
2207
2208 /* This function assumes rcu_read_lock() is held by the caller. */
2209 static int __igc_xdp_run_prog(struct igc_adapter *adapter,
2210                               struct bpf_prog *prog,
2211                               struct xdp_buff *xdp)
2212 {
2213         u32 act = bpf_prog_run_xdp(prog, xdp);
2214
2215         switch (act) {
2216         case XDP_PASS:
2217                 return IGC_XDP_PASS;
2218         case XDP_TX:
2219                 if (igc_xdp_xmit_back(adapter, xdp) < 0)
2220                         goto out_failure;
2221                 return IGC_XDP_TX;
2222         case XDP_REDIRECT:
2223                 if (xdp_do_redirect(adapter->netdev, xdp, prog) < 0)
2224                         goto out_failure;
2225                 return IGC_XDP_REDIRECT;
2226                 break;
2227         default:
2228                 bpf_warn_invalid_xdp_action(act);
2229                 fallthrough;
2230         case XDP_ABORTED:
2231 out_failure:
2232                 trace_xdp_exception(adapter->netdev, prog, act);
2233                 fallthrough;
2234         case XDP_DROP:
2235                 return IGC_XDP_CONSUMED;
2236         }
2237 }
2238
2239 static struct sk_buff *igc_xdp_run_prog(struct igc_adapter *adapter,
2240                                         struct xdp_buff *xdp)
2241 {
2242         struct bpf_prog *prog;
2243         int res;
2244
2245         prog = READ_ONCE(adapter->xdp_prog);
2246         if (!prog) {
2247                 res = IGC_XDP_PASS;
2248                 goto out;
2249         }
2250
2251         res = __igc_xdp_run_prog(adapter, prog, xdp);
2252
2253 out:
2254         return ERR_PTR(-res);
2255 }
2256
2257 /* This function assumes __netif_tx_lock is held by the caller. */
2258 static void igc_flush_tx_descriptors(struct igc_ring *ring)
2259 {
2260         /* Once tail pointer is updated, hardware can fetch the descriptors
2261          * any time so we issue a write membar here to ensure all memory
2262          * writes are complete before the tail pointer is updated.
2263          */
2264         wmb();
2265         writel(ring->next_to_use, ring->tail);
2266 }
2267
2268 static void igc_finalize_xdp(struct igc_adapter *adapter, int status)
2269 {
2270         int cpu = smp_processor_id();
2271         struct netdev_queue *nq;
2272         struct igc_ring *ring;
2273
2274         if (status & IGC_XDP_TX) {
2275                 ring = igc_xdp_get_tx_ring(adapter, cpu);
2276                 nq = txring_txq(ring);
2277
2278                 __netif_tx_lock(nq, cpu);
2279                 igc_flush_tx_descriptors(ring);
2280                 __netif_tx_unlock(nq);
2281         }
2282
2283         if (status & IGC_XDP_REDIRECT)
2284                 xdp_do_flush();
2285 }
2286
2287 static void igc_update_rx_stats(struct igc_q_vector *q_vector,
2288                                 unsigned int packets, unsigned int bytes)
2289 {
2290         struct igc_ring *ring = q_vector->rx.ring;
2291
2292         u64_stats_update_begin(&ring->rx_syncp);
2293         ring->rx_stats.packets += packets;
2294         ring->rx_stats.bytes += bytes;
2295         u64_stats_update_end(&ring->rx_syncp);
2296
2297         q_vector->rx.total_packets += packets;
2298         q_vector->rx.total_bytes += bytes;
2299 }
2300
2301 static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)
2302 {
2303         unsigned int total_bytes = 0, total_packets = 0;
2304         struct igc_adapter *adapter = q_vector->adapter;
2305         struct igc_ring *rx_ring = q_vector->rx.ring;
2306         struct sk_buff *skb = rx_ring->skb;
2307         u16 cleaned_count = igc_desc_unused(rx_ring);
2308         int xdp_status = 0, rx_buffer_pgcnt;
2309
2310         while (likely(total_packets < budget)) {
2311                 union igc_adv_rx_desc *rx_desc;
2312                 struct igc_rx_buffer *rx_buffer;
2313                 unsigned int size, truesize;
2314                 ktime_t timestamp = 0;
2315                 struct xdp_buff xdp;
2316                 int pkt_offset = 0;
2317                 void *pktbuf;
2318
2319                 /* return some buffers to hardware, one at a time is too slow */
2320                 if (cleaned_count >= IGC_RX_BUFFER_WRITE) {
2321                         igc_alloc_rx_buffers(rx_ring, cleaned_count);
2322                         cleaned_count = 0;
2323                 }
2324
2325                 rx_desc = IGC_RX_DESC(rx_ring, rx_ring->next_to_clean);
2326                 size = le16_to_cpu(rx_desc->wb.upper.length);
2327                 if (!size)
2328                         break;
2329
2330                 /* This memory barrier is needed to keep us from reading
2331                  * any other fields out of the rx_desc until we know the
2332                  * descriptor has been written back
2333                  */
2334                 dma_rmb();
2335
2336                 rx_buffer = igc_get_rx_buffer(rx_ring, size, &rx_buffer_pgcnt);
2337                 truesize = igc_get_rx_frame_truesize(rx_ring, size);
2338
2339                 pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset;
2340
2341                 if (igc_test_staterr(rx_desc, IGC_RXDADV_STAT_TSIP)) {
2342                         timestamp = igc_ptp_rx_pktstamp(q_vector->adapter,
2343                                                         pktbuf);
2344                         pkt_offset = IGC_TS_HDR_LEN;
2345                         size -= IGC_TS_HDR_LEN;
2346                 }
2347
2348                 if (!skb) {
2349                         xdp_init_buff(&xdp, truesize, &rx_ring->xdp_rxq);
2350                         xdp_prepare_buff(&xdp, pktbuf - igc_rx_offset(rx_ring),
2351                                          igc_rx_offset(rx_ring) + pkt_offset, size, false);
2352
2353                         skb = igc_xdp_run_prog(adapter, &xdp);
2354                 }
2355
2356                 if (IS_ERR(skb)) {
2357                         unsigned int xdp_res = -PTR_ERR(skb);
2358
2359                         switch (xdp_res) {
2360                         case IGC_XDP_CONSUMED:
2361                                 rx_buffer->pagecnt_bias++;
2362                                 break;
2363                         case IGC_XDP_TX:
2364                         case IGC_XDP_REDIRECT:
2365                                 igc_rx_buffer_flip(rx_buffer, truesize);
2366                                 xdp_status |= xdp_res;
2367                                 break;
2368                         }
2369
2370                         total_packets++;
2371                         total_bytes += size;
2372                 } else if (skb)
2373                         igc_add_rx_frag(rx_ring, rx_buffer, skb, size);
2374                 else if (ring_uses_build_skb(rx_ring))
2375                         skb = igc_build_skb(rx_ring, rx_buffer, rx_desc, size);
2376                 else
2377                         skb = igc_construct_skb(rx_ring, rx_buffer, &xdp,
2378                                                 timestamp);
2379
2380                 /* exit if we failed to retrieve a buffer */
2381                 if (!skb) {
2382                         rx_ring->rx_stats.alloc_failed++;
2383                         rx_buffer->pagecnt_bias++;
2384                         break;
2385                 }
2386
2387                 igc_put_rx_buffer(rx_ring, rx_buffer, rx_buffer_pgcnt);
2388                 cleaned_count++;
2389
2390                 /* fetch next buffer in frame if non-eop */
2391                 if (igc_is_non_eop(rx_ring, rx_desc))
2392                         continue;
2393
2394                 /* verify the packet layout is correct */
2395                 if (igc_cleanup_headers(rx_ring, rx_desc, skb)) {
2396                         skb = NULL;
2397                         continue;
2398                 }
2399
2400                 /* probably a little skewed due to removing CRC */
2401                 total_bytes += skb->len;
2402
2403                 /* populate checksum, VLAN, and protocol */
2404                 igc_process_skb_fields(rx_ring, rx_desc, skb);
2405
2406                 napi_gro_receive(&q_vector->napi, skb);
2407
2408                 /* reset skb pointer */
2409                 skb = NULL;
2410
2411                 /* update budget accounting */
2412                 total_packets++;
2413         }
2414
2415         if (xdp_status)
2416                 igc_finalize_xdp(adapter, xdp_status);
2417
2418         /* place incomplete frames back on ring for completion */
2419         rx_ring->skb = skb;
2420
2421         igc_update_rx_stats(q_vector, total_packets, total_bytes);
2422
2423         if (cleaned_count)
2424                 igc_alloc_rx_buffers(rx_ring, cleaned_count);
2425
2426         return total_packets;
2427 }
2428
2429 static struct sk_buff *igc_construct_skb_zc(struct igc_ring *ring,
2430                                             struct xdp_buff *xdp)
2431 {
2432         unsigned int metasize = xdp->data - xdp->data_meta;
2433         unsigned int datasize = xdp->data_end - xdp->data;
2434         unsigned int totalsize = metasize + datasize;
2435         struct sk_buff *skb;
2436
2437         skb = __napi_alloc_skb(&ring->q_vector->napi,
2438                                xdp->data_end - xdp->data_hard_start,
2439                                GFP_ATOMIC | __GFP_NOWARN);
2440         if (unlikely(!skb))
2441                 return NULL;
2442
2443         skb_reserve(skb, xdp->data_meta - xdp->data_hard_start);
2444         memcpy(__skb_put(skb, totalsize), xdp->data_meta, totalsize);
2445         if (metasize)
2446                 skb_metadata_set(skb, metasize);
2447
2448         return skb;
2449 }
2450
2451 static void igc_dispatch_skb_zc(struct igc_q_vector *q_vector,
2452                                 union igc_adv_rx_desc *desc,
2453                                 struct xdp_buff *xdp,
2454                                 ktime_t timestamp)
2455 {
2456         struct igc_ring *ring = q_vector->rx.ring;
2457         struct sk_buff *skb;
2458
2459         skb = igc_construct_skb_zc(ring, xdp);
2460         if (!skb) {
2461                 ring->rx_stats.alloc_failed++;
2462                 return;
2463         }
2464
2465         if (timestamp)
2466                 skb_hwtstamps(skb)->hwtstamp = timestamp;
2467
2468         if (igc_cleanup_headers(ring, desc, skb))
2469                 return;
2470
2471         igc_process_skb_fields(ring, desc, skb);
2472         napi_gro_receive(&q_vector->napi, skb);
2473 }
2474
2475 static int igc_clean_rx_irq_zc(struct igc_q_vector *q_vector, const int budget)
2476 {
2477         struct igc_adapter *adapter = q_vector->adapter;
2478         struct igc_ring *ring = q_vector->rx.ring;
2479         u16 cleaned_count = igc_desc_unused(ring);
2480         int total_bytes = 0, total_packets = 0;
2481         u16 ntc = ring->next_to_clean;
2482         struct bpf_prog *prog;
2483         bool failure = false;
2484         int xdp_status = 0;
2485
2486         rcu_read_lock();
2487
2488         prog = READ_ONCE(adapter->xdp_prog);
2489
2490         while (likely(total_packets < budget)) {
2491                 union igc_adv_rx_desc *desc;
2492                 struct igc_rx_buffer *bi;
2493                 ktime_t timestamp = 0;
2494                 unsigned int size;
2495                 int res;
2496
2497                 desc = IGC_RX_DESC(ring, ntc);
2498                 size = le16_to_cpu(desc->wb.upper.length);
2499                 if (!size)
2500                         break;
2501
2502                 /* This memory barrier is needed to keep us from reading
2503                  * any other fields out of the rx_desc until we know the
2504                  * descriptor has been written back
2505                  */
2506                 dma_rmb();
2507
2508                 bi = &ring->rx_buffer_info[ntc];
2509
2510                 if (igc_test_staterr(desc, IGC_RXDADV_STAT_TSIP)) {
2511                         timestamp = igc_ptp_rx_pktstamp(q_vector->adapter,
2512                                                         bi->xdp->data);
2513
2514                         bi->xdp->data += IGC_TS_HDR_LEN;
2515
2516                         /* HW timestamp has been copied into local variable. Metadata
2517                          * length when XDP program is called should be 0.
2518                          */
2519                         bi->xdp->data_meta += IGC_TS_HDR_LEN;
2520                         size -= IGC_TS_HDR_LEN;
2521                 }
2522
2523                 bi->xdp->data_end = bi->xdp->data + size;
2524                 xsk_buff_dma_sync_for_cpu(bi->xdp, ring->xsk_pool);
2525
2526                 res = __igc_xdp_run_prog(adapter, prog, bi->xdp);
2527                 switch (res) {
2528                 case IGC_XDP_PASS:
2529                         igc_dispatch_skb_zc(q_vector, desc, bi->xdp, timestamp);
2530                         fallthrough;
2531                 case IGC_XDP_CONSUMED:
2532                         xsk_buff_free(bi->xdp);
2533                         break;
2534                 case IGC_XDP_TX:
2535                 case IGC_XDP_REDIRECT:
2536                         xdp_status |= res;
2537                         break;
2538                 }
2539
2540                 bi->xdp = NULL;
2541                 total_bytes += size;
2542                 total_packets++;
2543                 cleaned_count++;
2544                 ntc++;
2545                 if (ntc == ring->count)
2546                         ntc = 0;
2547         }
2548
2549         ring->next_to_clean = ntc;
2550         rcu_read_unlock();
2551
2552         if (cleaned_count >= IGC_RX_BUFFER_WRITE)
2553                 failure = !igc_alloc_rx_buffers_zc(ring, cleaned_count);
2554
2555         if (xdp_status)
2556                 igc_finalize_xdp(adapter, xdp_status);
2557
2558         igc_update_rx_stats(q_vector, total_packets, total_bytes);
2559
2560         if (xsk_uses_need_wakeup(ring->xsk_pool)) {
2561                 if (failure || ring->next_to_clean == ring->next_to_use)
2562                         xsk_set_rx_need_wakeup(ring->xsk_pool);
2563                 else
2564                         xsk_clear_rx_need_wakeup(ring->xsk_pool);
2565                 return total_packets;
2566         }
2567
2568         return failure ? budget : total_packets;
2569 }
2570
2571 static void igc_update_tx_stats(struct igc_q_vector *q_vector,
2572                                 unsigned int packets, unsigned int bytes)
2573 {
2574         struct igc_ring *ring = q_vector->tx.ring;
2575
2576         u64_stats_update_begin(&ring->tx_syncp);
2577         ring->tx_stats.bytes += bytes;
2578         ring->tx_stats.packets += packets;
2579         u64_stats_update_end(&ring->tx_syncp);
2580
2581         q_vector->tx.total_bytes += bytes;
2582         q_vector->tx.total_packets += packets;
2583 }
2584
2585 static void igc_xdp_xmit_zc(struct igc_ring *ring)
2586 {
2587         struct xsk_buff_pool *pool = ring->xsk_pool;
2588         struct netdev_queue *nq = txring_txq(ring);
2589         union igc_adv_tx_desc *tx_desc = NULL;
2590         int cpu = smp_processor_id();
2591         u16 ntu = ring->next_to_use;
2592         struct xdp_desc xdp_desc;
2593         u16 budget;
2594
2595         if (!netif_carrier_ok(ring->netdev))
2596                 return;
2597
2598         __netif_tx_lock(nq, cpu);
2599
2600         budget = igc_desc_unused(ring);
2601
2602         while (xsk_tx_peek_desc(pool, &xdp_desc) && budget--) {
2603                 u32 cmd_type, olinfo_status;
2604                 struct igc_tx_buffer *bi;
2605                 dma_addr_t dma;
2606
2607                 cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
2608                            IGC_ADVTXD_DCMD_IFCS | IGC_TXD_DCMD |
2609                            xdp_desc.len;
2610                 olinfo_status = xdp_desc.len << IGC_ADVTXD_PAYLEN_SHIFT;
2611
2612                 dma = xsk_buff_raw_get_dma(pool, xdp_desc.addr);
2613                 xsk_buff_raw_dma_sync_for_device(pool, dma, xdp_desc.len);
2614
2615                 tx_desc = IGC_TX_DESC(ring, ntu);
2616                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
2617                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2618                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
2619
2620                 bi = &ring->tx_buffer_info[ntu];
2621                 bi->type = IGC_TX_BUFFER_TYPE_XSK;
2622                 bi->protocol = 0;
2623                 bi->bytecount = xdp_desc.len;
2624                 bi->gso_segs = 1;
2625                 bi->time_stamp = jiffies;
2626                 bi->next_to_watch = tx_desc;
2627
2628                 netdev_tx_sent_queue(txring_txq(ring), xdp_desc.len);
2629
2630                 ntu++;
2631                 if (ntu == ring->count)
2632                         ntu = 0;
2633         }
2634
2635         ring->next_to_use = ntu;
2636         if (tx_desc) {
2637                 igc_flush_tx_descriptors(ring);
2638                 xsk_tx_release(pool);
2639         }
2640
2641         __netif_tx_unlock(nq);
2642 }
2643
2644 /**
2645  * igc_clean_tx_irq - Reclaim resources after transmit completes
2646  * @q_vector: pointer to q_vector containing needed info
2647  * @napi_budget: Used to determine if we are in netpoll
2648  *
2649  * returns true if ring is completely cleaned
2650  */
2651 static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)
2652 {
2653         struct igc_adapter *adapter = q_vector->adapter;
2654         unsigned int total_bytes = 0, total_packets = 0;
2655         unsigned int budget = q_vector->tx.work_limit;
2656         struct igc_ring *tx_ring = q_vector->tx.ring;
2657         unsigned int i = tx_ring->next_to_clean;
2658         struct igc_tx_buffer *tx_buffer;
2659         union igc_adv_tx_desc *tx_desc;
2660         u32 xsk_frames = 0;
2661
2662         if (test_bit(__IGC_DOWN, &adapter->state))
2663                 return true;
2664
2665         tx_buffer = &tx_ring->tx_buffer_info[i];
2666         tx_desc = IGC_TX_DESC(tx_ring, i);
2667         i -= tx_ring->count;
2668
2669         do {
2670                 union igc_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
2671
2672                 /* if next_to_watch is not set then there is no work pending */
2673                 if (!eop_desc)
2674                         break;
2675
2676                 /* prevent any other reads prior to eop_desc */
2677                 smp_rmb();
2678
2679                 /* if DD is not set pending work has not been completed */
2680                 if (!(eop_desc->wb.status & cpu_to_le32(IGC_TXD_STAT_DD)))
2681                         break;
2682
2683                 /* clear next_to_watch to prevent false hangs */
2684                 tx_buffer->next_to_watch = NULL;
2685
2686                 /* update the statistics for this packet */
2687                 total_bytes += tx_buffer->bytecount;
2688                 total_packets += tx_buffer->gso_segs;
2689
2690                 switch (tx_buffer->type) {
2691                 case IGC_TX_BUFFER_TYPE_XSK:
2692                         xsk_frames++;
2693                         break;
2694                 case IGC_TX_BUFFER_TYPE_XDP:
2695                         xdp_return_frame(tx_buffer->xdpf);
2696                         igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
2697                         break;
2698                 case IGC_TX_BUFFER_TYPE_SKB:
2699                         napi_consume_skb(tx_buffer->skb, napi_budget);
2700                         igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
2701                         break;
2702                 default:
2703                         netdev_warn_once(tx_ring->netdev, "Unknown Tx buffer type\n");
2704                         break;
2705                 }
2706
2707                 /* clear last DMA location and unmap remaining buffers */
2708                 while (tx_desc != eop_desc) {
2709                         tx_buffer++;
2710                         tx_desc++;
2711                         i++;
2712                         if (unlikely(!i)) {
2713                                 i -= tx_ring->count;
2714                                 tx_buffer = tx_ring->tx_buffer_info;
2715                                 tx_desc = IGC_TX_DESC(tx_ring, 0);
2716                         }
2717
2718                         /* unmap any remaining paged data */
2719                         if (dma_unmap_len(tx_buffer, len))
2720                                 igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
2721                 }
2722
2723                 /* move us one more past the eop_desc for start of next pkt */
2724                 tx_buffer++;
2725                 tx_desc++;
2726                 i++;
2727                 if (unlikely(!i)) {
2728                         i -= tx_ring->count;
2729                         tx_buffer = tx_ring->tx_buffer_info;
2730                         tx_desc = IGC_TX_DESC(tx_ring, 0);
2731                 }
2732
2733                 /* issue prefetch for next Tx descriptor */
2734                 prefetch(tx_desc);
2735
2736                 /* update budget accounting */
2737                 budget--;
2738         } while (likely(budget));
2739
2740         netdev_tx_completed_queue(txring_txq(tx_ring),
2741                                   total_packets, total_bytes);
2742
2743         i += tx_ring->count;
2744         tx_ring->next_to_clean = i;
2745
2746         igc_update_tx_stats(q_vector, total_packets, total_bytes);
2747
2748         if (tx_ring->xsk_pool) {
2749                 if (xsk_frames)
2750                         xsk_tx_completed(tx_ring->xsk_pool, xsk_frames);
2751                 if (xsk_uses_need_wakeup(tx_ring->xsk_pool))
2752                         xsk_set_tx_need_wakeup(tx_ring->xsk_pool);
2753                 igc_xdp_xmit_zc(tx_ring);
2754         }
2755
2756         if (test_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
2757                 struct igc_hw *hw = &adapter->hw;
2758
2759                 /* Detect a transmit hang in hardware, this serializes the
2760                  * check with the clearing of time_stamp and movement of i
2761                  */
2762                 clear_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
2763                 if (tx_buffer->next_to_watch &&
2764                     time_after(jiffies, tx_buffer->time_stamp +
2765                     (adapter->tx_timeout_factor * HZ)) &&
2766                     !(rd32(IGC_STATUS) & IGC_STATUS_TXOFF)) {
2767                         /* detected Tx unit hang */
2768                         netdev_err(tx_ring->netdev,
2769                                    "Detected Tx Unit Hang\n"
2770                                    "  Tx Queue             <%d>\n"
2771                                    "  TDH                  <%x>\n"
2772                                    "  TDT                  <%x>\n"
2773                                    "  next_to_use          <%x>\n"
2774                                    "  next_to_clean        <%x>\n"
2775                                    "buffer_info[next_to_clean]\n"
2776                                    "  time_stamp           <%lx>\n"
2777                                    "  next_to_watch        <%p>\n"
2778                                    "  jiffies              <%lx>\n"
2779                                    "  desc.status          <%x>\n",
2780                                    tx_ring->queue_index,
2781                                    rd32(IGC_TDH(tx_ring->reg_idx)),
2782                                    readl(tx_ring->tail),
2783                                    tx_ring->next_to_use,
2784                                    tx_ring->next_to_clean,
2785                                    tx_buffer->time_stamp,
2786                                    tx_buffer->next_to_watch,
2787                                    jiffies,
2788                                    tx_buffer->next_to_watch->wb.status);
2789                         netif_stop_subqueue(tx_ring->netdev,
2790                                             tx_ring->queue_index);
2791
2792                         /* we are about to reset, no point in enabling stuff */
2793                         return true;
2794                 }
2795         }
2796
2797 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
2798         if (unlikely(total_packets &&
2799                      netif_carrier_ok(tx_ring->netdev) &&
2800                      igc_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
2801                 /* Make sure that anybody stopping the queue after this
2802                  * sees the new next_to_clean.
2803                  */
2804                 smp_mb();
2805                 if (__netif_subqueue_stopped(tx_ring->netdev,
2806                                              tx_ring->queue_index) &&
2807                     !(test_bit(__IGC_DOWN, &adapter->state))) {
2808                         netif_wake_subqueue(tx_ring->netdev,
2809                                             tx_ring->queue_index);
2810
2811                         u64_stats_update_begin(&tx_ring->tx_syncp);
2812                         tx_ring->tx_stats.restart_queue++;
2813                         u64_stats_update_end(&tx_ring->tx_syncp);
2814                 }
2815         }
2816
2817         return !!budget;
2818 }
2819
2820 static int igc_find_mac_filter(struct igc_adapter *adapter,
2821                                enum igc_mac_filter_type type, const u8 *addr)
2822 {
2823         struct igc_hw *hw = &adapter->hw;
2824         int max_entries = hw->mac.rar_entry_count;
2825         u32 ral, rah;
2826         int i;
2827
2828         for (i = 0; i < max_entries; i++) {
2829                 ral = rd32(IGC_RAL(i));
2830                 rah = rd32(IGC_RAH(i));
2831
2832                 if (!(rah & IGC_RAH_AV))
2833                         continue;
2834                 if (!!(rah & IGC_RAH_ASEL_SRC_ADDR) != type)
2835                         continue;
2836                 if ((rah & IGC_RAH_RAH_MASK) !=
2837                     le16_to_cpup((__le16 *)(addr + 4)))
2838                         continue;
2839                 if (ral != le32_to_cpup((__le32 *)(addr)))
2840                         continue;
2841
2842                 return i;
2843         }
2844
2845         return -1;
2846 }
2847
2848 static int igc_get_avail_mac_filter_slot(struct igc_adapter *adapter)
2849 {
2850         struct igc_hw *hw = &adapter->hw;
2851         int max_entries = hw->mac.rar_entry_count;
2852         u32 rah;
2853         int i;
2854
2855         for (i = 0; i < max_entries; i++) {
2856                 rah = rd32(IGC_RAH(i));
2857
2858                 if (!(rah & IGC_RAH_AV))
2859                         return i;
2860         }
2861
2862         return -1;
2863 }
2864
2865 /**
2866  * igc_add_mac_filter() - Add MAC address filter
2867  * @adapter: Pointer to adapter where the filter should be added
2868  * @type: MAC address filter type (source or destination)
2869  * @addr: MAC address
2870  * @queue: If non-negative, queue assignment feature is enabled and frames
2871  *         matching the filter are enqueued onto 'queue'. Otherwise, queue
2872  *         assignment is disabled.
2873  *
2874  * Return: 0 in case of success, negative errno code otherwise.
2875  */
2876 static int igc_add_mac_filter(struct igc_adapter *adapter,
2877                               enum igc_mac_filter_type type, const u8 *addr,
2878                               int queue)
2879 {
2880         struct net_device *dev = adapter->netdev;
2881         int index;
2882
2883         index = igc_find_mac_filter(adapter, type, addr);
2884         if (index >= 0)
2885                 goto update_filter;
2886
2887         index = igc_get_avail_mac_filter_slot(adapter);
2888         if (index < 0)
2889                 return -ENOSPC;
2890
2891         netdev_dbg(dev, "Add MAC address filter: index %d type %s address %pM queue %d\n",
2892                    index, type == IGC_MAC_FILTER_TYPE_DST ? "dst" : "src",
2893                    addr, queue);
2894
2895 update_filter:
2896         igc_set_mac_filter_hw(adapter, index, type, addr, queue);
2897         return 0;
2898 }
2899
2900 /**
2901  * igc_del_mac_filter() - Delete MAC address filter
2902  * @adapter: Pointer to adapter where the filter should be deleted from
2903  * @type: MAC address filter type (source or destination)
2904  * @addr: MAC address
2905  */
2906 static void igc_del_mac_filter(struct igc_adapter *adapter,
2907                                enum igc_mac_filter_type type, const u8 *addr)
2908 {
2909         struct net_device *dev = adapter->netdev;
2910         int index;
2911
2912         index = igc_find_mac_filter(adapter, type, addr);
2913         if (index < 0)
2914                 return;
2915
2916         if (index == 0) {
2917                 /* If this is the default filter, we don't actually delete it.
2918                  * We just reset to its default value i.e. disable queue
2919                  * assignment.
2920                  */
2921                 netdev_dbg(dev, "Disable default MAC filter queue assignment");
2922
2923                 igc_set_mac_filter_hw(adapter, 0, type, addr, -1);
2924         } else {
2925                 netdev_dbg(dev, "Delete MAC address filter: index %d type %s address %pM\n",
2926                            index,
2927                            type == IGC_MAC_FILTER_TYPE_DST ? "dst" : "src",
2928                            addr);
2929
2930                 igc_clear_mac_filter_hw(adapter, index);
2931         }
2932 }
2933
2934 /**
2935  * igc_add_vlan_prio_filter() - Add VLAN priority filter
2936  * @adapter: Pointer to adapter where the filter should be added
2937  * @prio: VLAN priority value
2938  * @queue: Queue number which matching frames are assigned to
2939  *
2940  * Return: 0 in case of success, negative errno code otherwise.
2941  */
2942 static int igc_add_vlan_prio_filter(struct igc_adapter *adapter, int prio,
2943                                     int queue)
2944 {
2945         struct net_device *dev = adapter->netdev;
2946         struct igc_hw *hw = &adapter->hw;
2947         u32 vlanpqf;
2948
2949         vlanpqf = rd32(IGC_VLANPQF);
2950
2951         if (vlanpqf & IGC_VLANPQF_VALID(prio)) {
2952                 netdev_dbg(dev, "VLAN priority filter already in use\n");
2953                 return -EEXIST;
2954         }
2955
2956         vlanpqf |= IGC_VLANPQF_QSEL(prio, queue);
2957         vlanpqf |= IGC_VLANPQF_VALID(prio);
2958
2959         wr32(IGC_VLANPQF, vlanpqf);
2960
2961         netdev_dbg(dev, "Add VLAN priority filter: prio %d queue %d\n",
2962                    prio, queue);
2963         return 0;
2964 }
2965
2966 /**
2967  * igc_del_vlan_prio_filter() - Delete VLAN priority filter
2968  * @adapter: Pointer to adapter where the filter should be deleted from
2969  * @prio: VLAN priority value
2970  */
2971 static void igc_del_vlan_prio_filter(struct igc_adapter *adapter, int prio)
2972 {
2973         struct igc_hw *hw = &adapter->hw;
2974         u32 vlanpqf;
2975
2976         vlanpqf = rd32(IGC_VLANPQF);
2977
2978         vlanpqf &= ~IGC_VLANPQF_VALID(prio);
2979         vlanpqf &= ~IGC_VLANPQF_QSEL(prio, IGC_VLANPQF_QUEUE_MASK);
2980
2981         wr32(IGC_VLANPQF, vlanpqf);
2982
2983         netdev_dbg(adapter->netdev, "Delete VLAN priority filter: prio %d\n",
2984                    prio);
2985 }
2986
2987 static int igc_get_avail_etype_filter_slot(struct igc_adapter *adapter)
2988 {
2989         struct igc_hw *hw = &adapter->hw;
2990         int i;
2991
2992         for (i = 0; i < MAX_ETYPE_FILTER; i++) {
2993                 u32 etqf = rd32(IGC_ETQF(i));
2994
2995                 if (!(etqf & IGC_ETQF_FILTER_ENABLE))
2996                         return i;
2997         }
2998
2999         return -1;
3000 }
3001
3002 /**
3003  * igc_add_etype_filter() - Add ethertype filter
3004  * @adapter: Pointer to adapter where the filter should be added
3005  * @etype: Ethertype value
3006  * @queue: If non-negative, queue assignment feature is enabled and frames
3007  *         matching the filter are enqueued onto 'queue'. Otherwise, queue
3008  *         assignment is disabled.
3009  *
3010  * Return: 0 in case of success, negative errno code otherwise.
3011  */
3012 static int igc_add_etype_filter(struct igc_adapter *adapter, u16 etype,
3013                                 int queue)
3014 {
3015         struct igc_hw *hw = &adapter->hw;
3016         int index;
3017         u32 etqf;
3018
3019         index = igc_get_avail_etype_filter_slot(adapter);
3020         if (index < 0)
3021                 return -ENOSPC;
3022
3023         etqf = rd32(IGC_ETQF(index));
3024
3025         etqf &= ~IGC_ETQF_ETYPE_MASK;
3026         etqf |= etype;
3027
3028         if (queue >= 0) {
3029                 etqf &= ~IGC_ETQF_QUEUE_MASK;
3030                 etqf |= (queue << IGC_ETQF_QUEUE_SHIFT);
3031                 etqf |= IGC_ETQF_QUEUE_ENABLE;
3032         }
3033
3034         etqf |= IGC_ETQF_FILTER_ENABLE;
3035
3036         wr32(IGC_ETQF(index), etqf);
3037
3038         netdev_dbg(adapter->netdev, "Add ethertype filter: etype %04x queue %d\n",
3039                    etype, queue);
3040         return 0;
3041 }
3042
3043 static int igc_find_etype_filter(struct igc_adapter *adapter, u16 etype)
3044 {
3045         struct igc_hw *hw = &adapter->hw;
3046         int i;
3047
3048         for (i = 0; i < MAX_ETYPE_FILTER; i++) {
3049                 u32 etqf = rd32(IGC_ETQF(i));
3050
3051                 if ((etqf & IGC_ETQF_ETYPE_MASK) == etype)
3052                         return i;
3053         }
3054
3055         return -1;
3056 }
3057
3058 /**
3059  * igc_del_etype_filter() - Delete ethertype filter
3060  * @adapter: Pointer to adapter where the filter should be deleted from
3061  * @etype: Ethertype value
3062  */
3063 static void igc_del_etype_filter(struct igc_adapter *adapter, u16 etype)
3064 {
3065         struct igc_hw *hw = &adapter->hw;
3066         int index;
3067
3068         index = igc_find_etype_filter(adapter, etype);
3069         if (index < 0)
3070                 return;
3071
3072         wr32(IGC_ETQF(index), 0);
3073
3074         netdev_dbg(adapter->netdev, "Delete ethertype filter: etype %04x\n",
3075                    etype);
3076 }
3077
3078 static int igc_flex_filter_select(struct igc_adapter *adapter,
3079                                   struct igc_flex_filter *input,
3080                                   u32 *fhft)
3081 {
3082         struct igc_hw *hw = &adapter->hw;
3083         u8 fhft_index;
3084         u32 fhftsl;
3085
3086         if (input->index >= MAX_FLEX_FILTER) {
3087                 dev_err(&adapter->pdev->dev, "Wrong Flex Filter index selected!\n");
3088                 return -EINVAL;
3089         }
3090
3091         /* Indirect table select register */
3092         fhftsl = rd32(IGC_FHFTSL);
3093         fhftsl &= ~IGC_FHFTSL_FTSL_MASK;
3094         switch (input->index) {
3095         case 0 ... 7:
3096                 fhftsl |= 0x00;
3097                 break;
3098         case 8 ... 15:
3099                 fhftsl |= 0x01;
3100                 break;
3101         case 16 ... 23:
3102                 fhftsl |= 0x02;
3103                 break;
3104         case 24 ... 31:
3105                 fhftsl |= 0x03;
3106                 break;
3107         }
3108         wr32(IGC_FHFTSL, fhftsl);
3109
3110         /* Normalize index down to host table register */
3111         fhft_index = input->index % 8;
3112
3113         *fhft = (fhft_index < 4) ? IGC_FHFT(fhft_index) :
3114                 IGC_FHFT_EXT(fhft_index - 4);
3115
3116         return 0;
3117 }
3118
3119 static int igc_write_flex_filter_ll(struct igc_adapter *adapter,
3120                                     struct igc_flex_filter *input)
3121 {
3122         struct device *dev = &adapter->pdev->dev;
3123         struct igc_hw *hw = &adapter->hw;
3124         u8 *data = input->data;
3125         u8 *mask = input->mask;
3126         u32 queuing;
3127         u32 fhft;
3128         u32 wufc;
3129         int ret;
3130         int i;
3131
3132         /* Length has to be aligned to 8. Otherwise the filter will fail. Bail
3133          * out early to avoid surprises later.
3134          */
3135         if (input->length % 8 != 0) {
3136                 dev_err(dev, "The length of a flex filter has to be 8 byte aligned!\n");
3137                 return -EINVAL;
3138         }
3139
3140         /* Select corresponding flex filter register and get base for host table. */
3141         ret = igc_flex_filter_select(adapter, input, &fhft);
3142         if (ret)
3143                 return ret;
3144
3145         /* When adding a filter globally disable flex filter feature. That is
3146          * recommended within the datasheet.
3147          */
3148         wufc = rd32(IGC_WUFC);
3149         wufc &= ~IGC_WUFC_FLEX_HQ;
3150         wr32(IGC_WUFC, wufc);
3151
3152         /* Configure filter */
3153         queuing = input->length & IGC_FHFT_LENGTH_MASK;
3154         queuing |= (input->rx_queue << IGC_FHFT_QUEUE_SHIFT) & IGC_FHFT_QUEUE_MASK;
3155         queuing |= (input->prio << IGC_FHFT_PRIO_SHIFT) & IGC_FHFT_PRIO_MASK;
3156
3157         if (input->immediate_irq)
3158                 queuing |= IGC_FHFT_IMM_INT;
3159
3160         if (input->drop)
3161                 queuing |= IGC_FHFT_DROP;
3162
3163         wr32(fhft + 0xFC, queuing);
3164
3165         /* Write data (128 byte) and mask (128 bit) */
3166         for (i = 0; i < 16; ++i) {
3167                 const size_t data_idx = i * 8;
3168                 const size_t row_idx = i * 16;
3169                 u32 dw0 =
3170                         (data[data_idx + 0] << 0) |
3171                         (data[data_idx + 1] << 8) |
3172                         (data[data_idx + 2] << 16) |
3173                         (data[data_idx + 3] << 24);
3174                 u32 dw1 =
3175                         (data[data_idx + 4] << 0) |
3176                         (data[data_idx + 5] << 8) |
3177                         (data[data_idx + 6] << 16) |
3178                         (data[data_idx + 7] << 24);
3179                 u32 tmp;
3180
3181                 /* Write row: dw0, dw1 and mask */
3182                 wr32(fhft + row_idx, dw0);
3183                 wr32(fhft + row_idx + 4, dw1);
3184
3185                 /* mask is only valid for MASK(7, 0) */
3186                 tmp = rd32(fhft + row_idx + 8);
3187                 tmp &= ~GENMASK(7, 0);
3188                 tmp |= mask[i];
3189                 wr32(fhft + row_idx + 8, tmp);
3190         }
3191
3192         /* Enable filter. */
3193         wufc |= IGC_WUFC_FLEX_HQ;
3194         if (input->index > 8) {
3195                 /* Filter 0-7 are enabled via WUFC. The other 24 filters are not. */
3196                 u32 wufc_ext = rd32(IGC_WUFC_EXT);
3197
3198                 wufc_ext |= (IGC_WUFC_EXT_FLX8 << (input->index - 8));
3199
3200                 wr32(IGC_WUFC_EXT, wufc_ext);
3201         } else {
3202                 wufc |= (IGC_WUFC_FLX0 << input->index);
3203         }
3204         wr32(IGC_WUFC, wufc);
3205
3206         dev_dbg(&adapter->pdev->dev, "Added flex filter %u to HW.\n",
3207                 input->index);
3208
3209         return 0;
3210 }
3211
3212 static void igc_flex_filter_add_field(struct igc_flex_filter *flex,
3213                                       const void *src, unsigned int offset,
3214                                       size_t len, const void *mask)
3215 {
3216         int i;
3217
3218         /* data */
3219         memcpy(&flex->data[offset], src, len);
3220
3221         /* mask */
3222         for (i = 0; i < len; ++i) {
3223                 const unsigned int idx = i + offset;
3224                 const u8 *ptr = mask;
3225
3226                 if (mask) {
3227                         if (ptr[i] & 0xff)
3228                                 flex->mask[idx / 8] |= BIT(idx % 8);
3229
3230                         continue;
3231                 }
3232
3233                 flex->mask[idx / 8] |= BIT(idx % 8);
3234         }
3235 }
3236
3237 static int igc_find_avail_flex_filter_slot(struct igc_adapter *adapter)
3238 {
3239         struct igc_hw *hw = &adapter->hw;
3240         u32 wufc, wufc_ext;
3241         int i;
3242
3243         wufc = rd32(IGC_WUFC);
3244         wufc_ext = rd32(IGC_WUFC_EXT);
3245
3246         for (i = 0; i < MAX_FLEX_FILTER; i++) {
3247                 if (i < 8) {
3248                         if (!(wufc & (IGC_WUFC_FLX0 << i)))
3249                                 return i;
3250                 } else {
3251                         if (!(wufc_ext & (IGC_WUFC_EXT_FLX8 << (i - 8))))
3252                                 return i;
3253                 }
3254         }
3255
3256         return -ENOSPC;
3257 }
3258
3259 static bool igc_flex_filter_in_use(struct igc_adapter *adapter)
3260 {
3261         struct igc_hw *hw = &adapter->hw;
3262         u32 wufc, wufc_ext;
3263
3264         wufc = rd32(IGC_WUFC);
3265         wufc_ext = rd32(IGC_WUFC_EXT);
3266
3267         if (wufc & IGC_WUFC_FILTER_MASK)
3268                 return true;
3269
3270         if (wufc_ext & IGC_WUFC_EXT_FILTER_MASK)
3271                 return true;
3272
3273         return false;
3274 }
3275
3276 static int igc_add_flex_filter(struct igc_adapter *adapter,
3277                                struct igc_nfc_rule *rule)
3278 {
3279         struct igc_flex_filter flex = { };
3280         struct igc_nfc_filter *filter = &rule->filter;
3281         unsigned int eth_offset, user_offset;
3282         int ret, index;
3283         bool vlan;
3284
3285         index = igc_find_avail_flex_filter_slot(adapter);
3286         if (index < 0)
3287                 return -ENOSPC;
3288
3289         /* Construct the flex filter:
3290          *  -> dest_mac [6]
3291          *  -> src_mac [6]
3292          *  -> tpid [2]
3293          *  -> vlan tci [2]
3294          *  -> ether type [2]
3295          *  -> user data [8]
3296          *  -> = 26 bytes => 32 length
3297          */
3298         flex.index    = index;
3299         flex.length   = 32;
3300         flex.rx_queue = rule->action;
3301
3302         vlan = rule->filter.vlan_tci || rule->filter.vlan_etype;
3303         eth_offset = vlan ? 16 : 12;
3304         user_offset = vlan ? 18 : 14;
3305
3306         /* Add destination MAC  */
3307         if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR)
3308                 igc_flex_filter_add_field(&flex, &filter->dst_addr, 0,
3309                                           ETH_ALEN, NULL);
3310
3311         /* Add source MAC */
3312         if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR)
3313                 igc_flex_filter_add_field(&flex, &filter->src_addr, 6,
3314                                           ETH_ALEN, NULL);
3315
3316         /* Add VLAN etype */
3317         if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_ETYPE)
3318                 igc_flex_filter_add_field(&flex, &filter->vlan_etype, 12,
3319                                           sizeof(filter->vlan_etype),
3320                                           NULL);
3321
3322         /* Add VLAN TCI */
3323         if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI)
3324                 igc_flex_filter_add_field(&flex, &filter->vlan_tci, 14,
3325                                           sizeof(filter->vlan_tci), NULL);
3326
3327         /* Add Ether type */
3328         if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE) {
3329                 __be16 etype = cpu_to_be16(filter->etype);
3330
3331                 igc_flex_filter_add_field(&flex, &etype, eth_offset,
3332                                           sizeof(etype), NULL);
3333         }
3334
3335         /* Add user data */
3336         if (rule->filter.match_flags & IGC_FILTER_FLAG_USER_DATA)
3337                 igc_flex_filter_add_field(&flex, &filter->user_data,
3338                                           user_offset,
3339                                           sizeof(filter->user_data),
3340                                           filter->user_mask);
3341
3342         /* Add it down to the hardware and enable it. */
3343         ret = igc_write_flex_filter_ll(adapter, &flex);
3344         if (ret)
3345                 return ret;
3346
3347         filter->flex_index = index;
3348
3349         return 0;
3350 }
3351
3352 static void igc_del_flex_filter(struct igc_adapter *adapter,
3353                                 u16 reg_index)
3354 {
3355         struct igc_hw *hw = &adapter->hw;
3356         u32 wufc;
3357
3358         /* Just disable the filter. The filter table itself is kept
3359          * intact. Another flex_filter_add() should override the "old" data
3360          * then.
3361          */
3362         if (reg_index > 8) {
3363                 u32 wufc_ext = rd32(IGC_WUFC_EXT);
3364
3365                 wufc_ext &= ~(IGC_WUFC_EXT_FLX8 << (reg_index - 8));
3366                 wr32(IGC_WUFC_EXT, wufc_ext);
3367         } else {
3368                 wufc = rd32(IGC_WUFC);
3369
3370                 wufc &= ~(IGC_WUFC_FLX0 << reg_index);
3371                 wr32(IGC_WUFC, wufc);
3372         }
3373
3374         if (igc_flex_filter_in_use(adapter))
3375                 return;
3376
3377         /* No filters are in use, we may disable flex filters */
3378         wufc = rd32(IGC_WUFC);
3379         wufc &= ~IGC_WUFC_FLEX_HQ;
3380         wr32(IGC_WUFC, wufc);
3381 }
3382
3383 static int igc_enable_nfc_rule(struct igc_adapter *adapter,
3384                                struct igc_nfc_rule *rule)
3385 {
3386         int err;
3387
3388         if (rule->flex) {
3389                 return igc_add_flex_filter(adapter, rule);
3390         }
3391
3392         if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE) {
3393                 err = igc_add_etype_filter(adapter, rule->filter.etype,
3394                                            rule->action);
3395                 if (err)
3396                         return err;
3397         }
3398
3399         if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR) {
3400                 err = igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_SRC,
3401                                          rule->filter.src_addr, rule->action);
3402                 if (err)
3403                         return err;
3404         }
3405
3406         if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR) {
3407                 err = igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST,
3408                                          rule->filter.dst_addr, rule->action);
3409                 if (err)
3410                         return err;
3411         }
3412
3413         if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI) {
3414                 int prio = (rule->filter.vlan_tci & VLAN_PRIO_MASK) >>
3415                            VLAN_PRIO_SHIFT;
3416
3417                 err = igc_add_vlan_prio_filter(adapter, prio, rule->action);
3418                 if (err)
3419                         return err;
3420         }
3421
3422         return 0;
3423 }
3424
3425 static void igc_disable_nfc_rule(struct igc_adapter *adapter,
3426                                  const struct igc_nfc_rule *rule)
3427 {
3428         if (rule->flex) {
3429                 igc_del_flex_filter(adapter, rule->filter.flex_index);
3430                 return;
3431         }
3432
3433         if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE)
3434                 igc_del_etype_filter(adapter, rule->filter.etype);
3435
3436         if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI) {
3437                 int prio = (rule->filter.vlan_tci & VLAN_PRIO_MASK) >>
3438                            VLAN_PRIO_SHIFT;
3439
3440                 igc_del_vlan_prio_filter(adapter, prio);
3441         }
3442
3443         if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR)
3444                 igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_SRC,
3445                                    rule->filter.src_addr);
3446
3447         if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR)
3448                 igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST,
3449                                    rule->filter.dst_addr);
3450 }
3451
3452 /**
3453  * igc_get_nfc_rule() - Get NFC rule
3454  * @adapter: Pointer to adapter
3455  * @location: Rule location
3456  *
3457  * Context: Expects adapter->nfc_rule_lock to be held by caller.
3458  *
3459  * Return: Pointer to NFC rule at @location. If not found, NULL.
3460  */
3461 struct igc_nfc_rule *igc_get_nfc_rule(struct igc_adapter *adapter,
3462                                       u32 location)
3463 {
3464         struct igc_nfc_rule *rule;
3465
3466         list_for_each_entry(rule, &adapter->nfc_rule_list, list) {
3467                 if (rule->location == location)
3468                         return rule;
3469                 if (rule->location > location)
3470                         break;
3471         }
3472
3473         return NULL;
3474 }
3475
3476 /**
3477  * igc_del_nfc_rule() - Delete NFC rule
3478  * @adapter: Pointer to adapter
3479  * @rule: Pointer to rule to be deleted
3480  *
3481  * Disable NFC rule in hardware and delete it from adapter.
3482  *
3483  * Context: Expects adapter->nfc_rule_lock to be held by caller.
3484  */
3485 void igc_del_nfc_rule(struct igc_adapter *adapter, struct igc_nfc_rule *rule)
3486 {
3487         igc_disable_nfc_rule(adapter, rule);
3488
3489         list_del(&rule->list);
3490         adapter->nfc_rule_count--;
3491
3492         kfree(rule);
3493 }
3494
3495 static void igc_flush_nfc_rules(struct igc_adapter *adapter)
3496 {
3497         struct igc_nfc_rule *rule, *tmp;
3498
3499         mutex_lock(&adapter->nfc_rule_lock);
3500
3501         list_for_each_entry_safe(rule, tmp, &adapter->nfc_rule_list, list)
3502                 igc_del_nfc_rule(adapter, rule);
3503
3504         mutex_unlock(&adapter->nfc_rule_lock);
3505 }
3506
3507 /**
3508  * igc_add_nfc_rule() - Add NFC rule
3509  * @adapter: Pointer to adapter
3510  * @rule: Pointer to rule to be added
3511  *
3512  * Enable NFC rule in hardware and add it to adapter.
3513  *
3514  * Context: Expects adapter->nfc_rule_lock to be held by caller.
3515  *
3516  * Return: 0 on success, negative errno on failure.
3517  */
3518 int igc_add_nfc_rule(struct igc_adapter *adapter, struct igc_nfc_rule *rule)
3519 {
3520         struct igc_nfc_rule *pred, *cur;
3521         int err;
3522
3523         err = igc_enable_nfc_rule(adapter, rule);
3524         if (err)
3525                 return err;
3526
3527         pred = NULL;
3528         list_for_each_entry(cur, &adapter->nfc_rule_list, list) {
3529                 if (cur->location >= rule->location)
3530                         break;
3531                 pred = cur;
3532         }
3533
3534         list_add(&rule->list, pred ? &pred->list : &adapter->nfc_rule_list);
3535         adapter->nfc_rule_count++;
3536         return 0;
3537 }
3538
3539 static void igc_restore_nfc_rules(struct igc_adapter *adapter)
3540 {
3541         struct igc_nfc_rule *rule;
3542
3543         mutex_lock(&adapter->nfc_rule_lock);
3544
3545         list_for_each_entry_reverse(rule, &adapter->nfc_rule_list, list)
3546                 igc_enable_nfc_rule(adapter, rule);
3547
3548         mutex_unlock(&adapter->nfc_rule_lock);
3549 }
3550
3551 static int igc_uc_sync(struct net_device *netdev, const unsigned char *addr)
3552 {
3553         struct igc_adapter *adapter = netdev_priv(netdev);
3554
3555         return igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST, addr, -1);
3556 }
3557
3558 static int igc_uc_unsync(struct net_device *netdev, const unsigned char *addr)
3559 {
3560         struct igc_adapter *adapter = netdev_priv(netdev);
3561
3562         igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST, addr);
3563         return 0;
3564 }
3565
3566 /**
3567  * igc_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3568  * @netdev: network interface device structure
3569  *
3570  * The set_rx_mode entry point is called whenever the unicast or multicast
3571  * address lists or the network interface flags are updated.  This routine is
3572  * responsible for configuring the hardware for proper unicast, multicast,
3573  * promiscuous mode, and all-multi behavior.
3574  */
3575 static void igc_set_rx_mode(struct net_device *netdev)
3576 {
3577         struct igc_adapter *adapter = netdev_priv(netdev);
3578         struct igc_hw *hw = &adapter->hw;
3579         u32 rctl = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
3580         int count;
3581
3582         /* Check for Promiscuous and All Multicast modes */
3583         if (netdev->flags & IFF_PROMISC) {
3584                 rctl |= IGC_RCTL_UPE | IGC_RCTL_MPE;
3585         } else {
3586                 if (netdev->flags & IFF_ALLMULTI) {
3587                         rctl |= IGC_RCTL_MPE;
3588                 } else {
3589                         /* Write addresses to the MTA, if the attempt fails
3590                          * then we should just turn on promiscuous mode so
3591                          * that we can at least receive multicast traffic
3592                          */
3593                         count = igc_write_mc_addr_list(netdev);
3594                         if (count < 0)
3595                                 rctl |= IGC_RCTL_MPE;
3596                 }
3597         }
3598
3599         /* Write addresses to available RAR registers, if there is not
3600          * sufficient space to store all the addresses then enable
3601          * unicast promiscuous mode
3602          */
3603         if (__dev_uc_sync(netdev, igc_uc_sync, igc_uc_unsync))
3604                 rctl |= IGC_RCTL_UPE;
3605
3606         /* update state of unicast and multicast */
3607         rctl |= rd32(IGC_RCTL) & ~(IGC_RCTL_UPE | IGC_RCTL_MPE);
3608         wr32(IGC_RCTL, rctl);
3609
3610 #if (PAGE_SIZE < 8192)
3611         if (adapter->max_frame_size <= IGC_MAX_FRAME_BUILD_SKB)
3612                 rlpml = IGC_MAX_FRAME_BUILD_SKB;
3613 #endif
3614         wr32(IGC_RLPML, rlpml);
3615 }
3616
3617 /**
3618  * igc_configure - configure the hardware for RX and TX
3619  * @adapter: private board structure
3620  */
3621 static void igc_configure(struct igc_adapter *adapter)
3622 {
3623         struct net_device *netdev = adapter->netdev;
3624         int i = 0;
3625
3626         igc_get_hw_control(adapter);
3627         igc_set_rx_mode(netdev);
3628
3629         igc_restore_vlan(adapter);
3630
3631         igc_setup_tctl(adapter);
3632         igc_setup_mrqc(adapter);
3633         igc_setup_rctl(adapter);
3634
3635         igc_set_default_mac_filter(adapter);
3636         igc_restore_nfc_rules(adapter);
3637
3638         igc_configure_tx(adapter);
3639         igc_configure_rx(adapter);
3640
3641         igc_rx_fifo_flush_base(&adapter->hw);
3642
3643         /* call igc_desc_unused which always leaves
3644          * at least 1 descriptor unused to make sure
3645          * next_to_use != next_to_clean
3646          */
3647         for (i = 0; i < adapter->num_rx_queues; i++) {
3648                 struct igc_ring *ring = adapter->rx_ring[i];
3649
3650                 if (ring->xsk_pool)
3651                         igc_alloc_rx_buffers_zc(ring, igc_desc_unused(ring));
3652                 else
3653                         igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
3654         }
3655 }
3656
3657 /**
3658  * igc_write_ivar - configure ivar for given MSI-X vector
3659  * @hw: pointer to the HW structure
3660  * @msix_vector: vector number we are allocating to a given ring
3661  * @index: row index of IVAR register to write within IVAR table
3662  * @offset: column offset of in IVAR, should be multiple of 8
3663  *
3664  * The IVAR table consists of 2 columns,
3665  * each containing an cause allocation for an Rx and Tx ring, and a
3666  * variable number of rows depending on the number of queues supported.
3667  */
3668 static void igc_write_ivar(struct igc_hw *hw, int msix_vector,
3669                            int index, int offset)
3670 {
3671         u32 ivar = array_rd32(IGC_IVAR0, index);
3672
3673         /* clear any bits that are currently set */
3674         ivar &= ~((u32)0xFF << offset);
3675
3676         /* write vector and valid bit */
3677         ivar |= (msix_vector | IGC_IVAR_VALID) << offset;
3678
3679         array_wr32(IGC_IVAR0, index, ivar);
3680 }
3681
3682 static void igc_assign_vector(struct igc_q_vector *q_vector, int msix_vector)
3683 {
3684         struct igc_adapter *adapter = q_vector->adapter;
3685         struct igc_hw *hw = &adapter->hw;
3686         int rx_queue = IGC_N0_QUEUE;
3687         int tx_queue = IGC_N0_QUEUE;
3688
3689         if (q_vector->rx.ring)
3690                 rx_queue = q_vector->rx.ring->reg_idx;
3691         if (q_vector->tx.ring)
3692                 tx_queue = q_vector->tx.ring->reg_idx;
3693
3694         switch (hw->mac.type) {
3695         case igc_i225:
3696                 if (rx_queue > IGC_N0_QUEUE)
3697                         igc_write_ivar(hw, msix_vector,
3698                                        rx_queue >> 1,
3699                                        (rx_queue & 0x1) << 4);
3700                 if (tx_queue > IGC_N0_QUEUE)
3701                         igc_write_ivar(hw, msix_vector,
3702                                        tx_queue >> 1,
3703                                        ((tx_queue & 0x1) << 4) + 8);
3704                 q_vector->eims_value = BIT(msix_vector);
3705                 break;
3706         default:
3707                 WARN_ONCE(hw->mac.type != igc_i225, "Wrong MAC type\n");
3708                 break;
3709         }
3710
3711         /* add q_vector eims value to global eims_enable_mask */
3712         adapter->eims_enable_mask |= q_vector->eims_value;
3713
3714         /* configure q_vector to set itr on first interrupt */
3715         q_vector->set_itr = 1;
3716 }
3717
3718 /**
3719  * igc_configure_msix - Configure MSI-X hardware
3720  * @adapter: Pointer to adapter structure
3721  *
3722  * igc_configure_msix sets up the hardware to properly
3723  * generate MSI-X interrupts.
3724  */
3725 static void igc_configure_msix(struct igc_adapter *adapter)
3726 {
3727         struct igc_hw *hw = &adapter->hw;
3728         int i, vector = 0;
3729         u32 tmp;
3730
3731         adapter->eims_enable_mask = 0;
3732
3733         /* set vector for other causes, i.e. link changes */
3734         switch (hw->mac.type) {
3735         case igc_i225:
3736                 /* Turn on MSI-X capability first, or our settings
3737                  * won't stick.  And it will take days to debug.
3738                  */
3739                 wr32(IGC_GPIE, IGC_GPIE_MSIX_MODE |
3740                      IGC_GPIE_PBA | IGC_GPIE_EIAME |
3741                      IGC_GPIE_NSICR);
3742
3743                 /* enable msix_other interrupt */
3744                 adapter->eims_other = BIT(vector);
3745                 tmp = (vector++ | IGC_IVAR_VALID) << 8;
3746
3747                 wr32(IGC_IVAR_MISC, tmp);
3748                 break;
3749         default:
3750                 /* do nothing, since nothing else supports MSI-X */
3751                 break;
3752         } /* switch (hw->mac.type) */
3753
3754         adapter->eims_enable_mask |= adapter->eims_other;
3755
3756         for (i = 0; i < adapter->num_q_vectors; i++)
3757                 igc_assign_vector(adapter->q_vector[i], vector++);
3758
3759         wrfl();
3760 }
3761
3762 /**
3763  * igc_irq_enable - Enable default interrupt generation settings
3764  * @adapter: board private structure
3765  */
3766 static void igc_irq_enable(struct igc_adapter *adapter)
3767 {
3768         struct igc_hw *hw = &adapter->hw;
3769
3770         if (adapter->msix_entries) {
3771                 u32 ims = IGC_IMS_LSC | IGC_IMS_DOUTSYNC | IGC_IMS_DRSTA;
3772                 u32 regval = rd32(IGC_EIAC);
3773
3774                 wr32(IGC_EIAC, regval | adapter->eims_enable_mask);
3775                 regval = rd32(IGC_EIAM);
3776                 wr32(IGC_EIAM, regval | adapter->eims_enable_mask);
3777                 wr32(IGC_EIMS, adapter->eims_enable_mask);
3778                 wr32(IGC_IMS, ims);
3779         } else {
3780                 wr32(IGC_IMS, IMS_ENABLE_MASK | IGC_IMS_DRSTA);
3781                 wr32(IGC_IAM, IMS_ENABLE_MASK | IGC_IMS_DRSTA);
3782         }
3783 }
3784
3785 /**
3786  * igc_irq_disable - Mask off interrupt generation on the NIC
3787  * @adapter: board private structure
3788  */
3789 static void igc_irq_disable(struct igc_adapter *adapter)
3790 {
3791         struct igc_hw *hw = &adapter->hw;
3792
3793         if (adapter->msix_entries) {
3794                 u32 regval = rd32(IGC_EIAM);
3795
3796                 wr32(IGC_EIAM, regval & ~adapter->eims_enable_mask);
3797                 wr32(IGC_EIMC, adapter->eims_enable_mask);
3798                 regval = rd32(IGC_EIAC);
3799                 wr32(IGC_EIAC, regval & ~adapter->eims_enable_mask);
3800         }
3801
3802         wr32(IGC_IAM, 0);
3803         wr32(IGC_IMC, ~0);
3804         wrfl();
3805
3806         if (adapter->msix_entries) {
3807                 int vector = 0, i;
3808
3809                 synchronize_irq(adapter->msix_entries[vector++].vector);
3810
3811                 for (i = 0; i < adapter->num_q_vectors; i++)
3812                         synchronize_irq(adapter->msix_entries[vector++].vector);
3813         } else {
3814                 synchronize_irq(adapter->pdev->irq);
3815         }
3816 }
3817
3818 void igc_set_flag_queue_pairs(struct igc_adapter *adapter,
3819                               const u32 max_rss_queues)
3820 {
3821         /* Determine if we need to pair queues. */
3822         /* If rss_queues > half of max_rss_queues, pair the queues in
3823          * order to conserve interrupts due to limited supply.
3824          */
3825         if (adapter->rss_queues > (max_rss_queues / 2))
3826                 adapter->flags |= IGC_FLAG_QUEUE_PAIRS;
3827         else
3828                 adapter->flags &= ~IGC_FLAG_QUEUE_PAIRS;
3829 }
3830
3831 unsigned int igc_get_max_rss_queues(struct igc_adapter *adapter)
3832 {
3833         return IGC_MAX_RX_QUEUES;
3834 }
3835
3836 static void igc_init_queue_configuration(struct igc_adapter *adapter)
3837 {
3838         u32 max_rss_queues;
3839
3840         max_rss_queues = igc_get_max_rss_queues(adapter);
3841         adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3842
3843         igc_set_flag_queue_pairs(adapter, max_rss_queues);
3844 }
3845
3846 /**
3847  * igc_reset_q_vector - Reset config for interrupt vector
3848  * @adapter: board private structure to initialize
3849  * @v_idx: Index of vector to be reset
3850  *
3851  * If NAPI is enabled it will delete any references to the
3852  * NAPI struct. This is preparation for igc_free_q_vector.
3853  */
3854 static void igc_reset_q_vector(struct igc_adapter *adapter, int v_idx)
3855 {
3856         struct igc_q_vector *q_vector = adapter->q_vector[v_idx];
3857
3858         /* if we're coming from igc_set_interrupt_capability, the vectors are
3859          * not yet allocated
3860          */
3861         if (!q_vector)
3862                 return;
3863
3864         if (q_vector->tx.ring)
3865                 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
3866
3867         if (q_vector->rx.ring)
3868                 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
3869
3870         netif_napi_del(&q_vector->napi);
3871 }
3872
3873 /**
3874  * igc_free_q_vector - Free memory allocated for specific interrupt vector
3875  * @adapter: board private structure to initialize
3876  * @v_idx: Index of vector to be freed
3877  *
3878  * This function frees the memory allocated to the q_vector.
3879  */
3880 static void igc_free_q_vector(struct igc_adapter *adapter, int v_idx)
3881 {
3882         struct igc_q_vector *q_vector = adapter->q_vector[v_idx];
3883
3884         adapter->q_vector[v_idx] = NULL;
3885
3886         /* igc_get_stats64() might access the rings on this vector,
3887          * we must wait a grace period before freeing it.
3888          */
3889         if (q_vector)
3890                 kfree_rcu(q_vector, rcu);
3891 }
3892
3893 /**
3894  * igc_free_q_vectors - Free memory allocated for interrupt vectors
3895  * @adapter: board private structure to initialize
3896  *
3897  * This function frees the memory allocated to the q_vectors.  In addition if
3898  * NAPI is enabled it will delete any references to the NAPI struct prior
3899  * to freeing the q_vector.
3900  */
3901 static void igc_free_q_vectors(struct igc_adapter *adapter)
3902 {
3903         int v_idx = adapter->num_q_vectors;
3904
3905         adapter->num_tx_queues = 0;
3906         adapter->num_rx_queues = 0;
3907         adapter->num_q_vectors = 0;
3908
3909         while (v_idx--) {
3910                 igc_reset_q_vector(adapter, v_idx);
3911                 igc_free_q_vector(adapter, v_idx);
3912         }
3913 }
3914
3915 /**
3916  * igc_update_itr - update the dynamic ITR value based on statistics
3917  * @q_vector: pointer to q_vector
3918  * @ring_container: ring info to update the itr for
3919  *
3920  * Stores a new ITR value based on packets and byte
3921  * counts during the last interrupt.  The advantage of per interrupt
3922  * computation is faster updates and more accurate ITR for the current
3923  * traffic pattern.  Constants in this function were computed
3924  * based on theoretical maximum wire speed and thresholds were set based
3925  * on testing data as well as attempting to minimize response time
3926  * while increasing bulk throughput.
3927  * NOTE: These calculations are only valid when operating in a single-
3928  * queue environment.
3929  */
3930 static void igc_update_itr(struct igc_q_vector *q_vector,
3931                            struct igc_ring_container *ring_container)
3932 {
3933         unsigned int packets = ring_container->total_packets;
3934         unsigned int bytes = ring_container->total_bytes;
3935         u8 itrval = ring_container->itr;
3936
3937         /* no packets, exit with status unchanged */
3938         if (packets == 0)
3939                 return;
3940
3941         switch (itrval) {
3942         case lowest_latency:
3943                 /* handle TSO and jumbo frames */
3944                 if (bytes / packets > 8000)
3945                         itrval = bulk_latency;
3946                 else if ((packets < 5) && (bytes > 512))
3947                         itrval = low_latency;
3948                 break;
3949         case low_latency:  /* 50 usec aka 20000 ints/s */
3950                 if (bytes > 10000) {
3951                         /* this if handles the TSO accounting */
3952                         if (bytes / packets > 8000)
3953                                 itrval = bulk_latency;
3954                         else if ((packets < 10) || ((bytes / packets) > 1200))
3955                                 itrval = bulk_latency;
3956                         else if ((packets > 35))
3957                                 itrval = lowest_latency;
3958                 } else if (bytes / packets > 2000) {
3959                         itrval = bulk_latency;
3960                 } else if (packets <= 2 && bytes < 512) {
3961                         itrval = lowest_latency;
3962                 }
3963                 break;
3964         case bulk_latency: /* 250 usec aka 4000 ints/s */
3965                 if (bytes > 25000) {
3966                         if (packets > 35)
3967                                 itrval = low_latency;
3968                 } else if (bytes < 1500) {
3969                         itrval = low_latency;
3970                 }
3971                 break;
3972         }
3973
3974         /* clear work counters since we have the values we need */
3975         ring_container->total_bytes = 0;
3976         ring_container->total_packets = 0;
3977
3978         /* write updated itr to ring container */
3979         ring_container->itr = itrval;
3980 }
3981
3982 static void igc_set_itr(struct igc_q_vector *q_vector)
3983 {
3984         struct igc_adapter *adapter = q_vector->adapter;
3985         u32 new_itr = q_vector->itr_val;
3986         u8 current_itr = 0;
3987
3988         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
3989         switch (adapter->link_speed) {
3990         case SPEED_10:
3991         case SPEED_100:
3992                 current_itr = 0;
3993                 new_itr = IGC_4K_ITR;
3994                 goto set_itr_now;
3995         default:
3996                 break;
3997         }
3998
3999         igc_update_itr(q_vector, &q_vector->tx);
4000         igc_update_itr(q_vector, &q_vector->rx);
4001
4002         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4003
4004         /* conservative mode (itr 3) eliminates the lowest_latency setting */
4005         if (current_itr == lowest_latency &&
4006             ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4007             (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4008                 current_itr = low_latency;
4009
4010         switch (current_itr) {
4011         /* counts and packets in update_itr are dependent on these numbers */
4012         case lowest_latency:
4013                 new_itr = IGC_70K_ITR; /* 70,000 ints/sec */
4014                 break;
4015         case low_latency:
4016                 new_itr = IGC_20K_ITR; /* 20,000 ints/sec */
4017                 break;
4018         case bulk_latency:
4019                 new_itr = IGC_4K_ITR;  /* 4,000 ints/sec */
4020                 break;
4021         default:
4022                 break;
4023         }
4024
4025 set_itr_now:
4026         if (new_itr != q_vector->itr_val) {
4027                 /* this attempts to bias the interrupt rate towards Bulk
4028                  * by adding intermediate steps when interrupt rate is
4029                  * increasing
4030                  */
4031                 new_itr = new_itr > q_vector->itr_val ?
4032                           max((new_itr * q_vector->itr_val) /
4033                           (new_itr + (q_vector->itr_val >> 2)),
4034                           new_itr) : new_itr;
4035                 /* Don't write the value here; it resets the adapter's
4036                  * internal timer, and causes us to delay far longer than
4037                  * we should between interrupts.  Instead, we write the ITR
4038                  * value at the beginning of the next interrupt so the timing
4039                  * ends up being correct.
4040                  */
4041                 q_vector->itr_val = new_itr;
4042                 q_vector->set_itr = 1;
4043         }
4044 }
4045
4046 static void igc_reset_interrupt_capability(struct igc_adapter *adapter)
4047 {
4048         int v_idx = adapter->num_q_vectors;
4049
4050         if (adapter->msix_entries) {
4051                 pci_disable_msix(adapter->pdev);
4052                 kfree(adapter->msix_entries);
4053                 adapter->msix_entries = NULL;
4054         } else if (adapter->flags & IGC_FLAG_HAS_MSI) {
4055                 pci_disable_msi(adapter->pdev);
4056         }
4057
4058         while (v_idx--)
4059                 igc_reset_q_vector(adapter, v_idx);
4060 }
4061
4062 /**
4063  * igc_set_interrupt_capability - set MSI or MSI-X if supported
4064  * @adapter: Pointer to adapter structure
4065  * @msix: boolean value for MSI-X capability
4066  *
4067  * Attempt to configure interrupts using the best available
4068  * capabilities of the hardware and kernel.
4069  */
4070 static void igc_set_interrupt_capability(struct igc_adapter *adapter,
4071                                          bool msix)
4072 {
4073         int numvecs, i;
4074         int err;
4075
4076         if (!msix)
4077                 goto msi_only;
4078         adapter->flags |= IGC_FLAG_HAS_MSIX;
4079
4080         /* Number of supported queues. */
4081         adapter->num_rx_queues = adapter->rss_queues;
4082
4083         adapter->num_tx_queues = adapter->rss_queues;
4084
4085         /* start with one vector for every Rx queue */
4086         numvecs = adapter->num_rx_queues;
4087
4088         /* if Tx handler is separate add 1 for every Tx queue */
4089         if (!(adapter->flags & IGC_FLAG_QUEUE_PAIRS))
4090                 numvecs += adapter->num_tx_queues;
4091
4092         /* store the number of vectors reserved for queues */
4093         adapter->num_q_vectors = numvecs;
4094
4095         /* add 1 vector for link status interrupts */
4096         numvecs++;
4097
4098         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
4099                                         GFP_KERNEL);
4100
4101         if (!adapter->msix_entries)
4102                 return;
4103
4104         /* populate entry values */
4105         for (i = 0; i < numvecs; i++)
4106                 adapter->msix_entries[i].entry = i;
4107
4108         err = pci_enable_msix_range(adapter->pdev,
4109                                     adapter->msix_entries,
4110                                     numvecs,
4111                                     numvecs);
4112         if (err > 0)
4113                 return;
4114
4115         kfree(adapter->msix_entries);
4116         adapter->msix_entries = NULL;
4117
4118         igc_reset_interrupt_capability(adapter);
4119
4120 msi_only:
4121         adapter->flags &= ~IGC_FLAG_HAS_MSIX;
4122
4123         adapter->rss_queues = 1;
4124         adapter->flags |= IGC_FLAG_QUEUE_PAIRS;
4125         adapter->num_rx_queues = 1;
4126         adapter->num_tx_queues = 1;
4127         adapter->num_q_vectors = 1;
4128         if (!pci_enable_msi(adapter->pdev))
4129                 adapter->flags |= IGC_FLAG_HAS_MSI;
4130 }
4131
4132 /**
4133  * igc_update_ring_itr - update the dynamic ITR value based on packet size
4134  * @q_vector: pointer to q_vector
4135  *
4136  * Stores a new ITR value based on strictly on packet size.  This
4137  * algorithm is less sophisticated than that used in igc_update_itr,
4138  * due to the difficulty of synchronizing statistics across multiple
4139  * receive rings.  The divisors and thresholds used by this function
4140  * were determined based on theoretical maximum wire speed and testing
4141  * data, in order to minimize response time while increasing bulk
4142  * throughput.
4143  * NOTE: This function is called only when operating in a multiqueue
4144  * receive environment.
4145  */
4146 static void igc_update_ring_itr(struct igc_q_vector *q_vector)
4147 {
4148         struct igc_adapter *adapter = q_vector->adapter;
4149         int new_val = q_vector->itr_val;
4150         int avg_wire_size = 0;
4151         unsigned int packets;
4152
4153         /* For non-gigabit speeds, just fix the interrupt rate at 4000
4154          * ints/sec - ITR timer value of 120 ticks.
4155          */
4156         switch (adapter->link_speed) {
4157         case SPEED_10:
4158         case SPEED_100:
4159                 new_val = IGC_4K_ITR;
4160                 goto set_itr_val;
4161         default:
4162                 break;
4163         }
4164
4165         packets = q_vector->rx.total_packets;
4166         if (packets)
4167                 avg_wire_size = q_vector->rx.total_bytes / packets;
4168
4169         packets = q_vector->tx.total_packets;
4170         if (packets)
4171                 avg_wire_size = max_t(u32, avg_wire_size,
4172                                       q_vector->tx.total_bytes / packets);
4173
4174         /* if avg_wire_size isn't set no work was done */
4175         if (!avg_wire_size)
4176                 goto clear_counts;
4177
4178         /* Add 24 bytes to size to account for CRC, preamble, and gap */
4179         avg_wire_size += 24;
4180
4181         /* Don't starve jumbo frames */
4182         avg_wire_size = min(avg_wire_size, 3000);
4183
4184         /* Give a little boost to mid-size frames */
4185         if (avg_wire_size > 300 && avg_wire_size < 1200)
4186                 new_val = avg_wire_size / 3;
4187         else
4188                 new_val = avg_wire_size / 2;
4189
4190         /* conservative mode (itr 3) eliminates the lowest_latency setting */
4191         if (new_val < IGC_20K_ITR &&
4192             ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4193             (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4194                 new_val = IGC_20K_ITR;
4195
4196 set_itr_val:
4197         if (new_val != q_vector->itr_val) {
4198                 q_vector->itr_val = new_val;
4199                 q_vector->set_itr = 1;
4200         }
4201 clear_counts:
4202         q_vector->rx.total_bytes = 0;
4203         q_vector->rx.total_packets = 0;
4204         q_vector->tx.total_bytes = 0;
4205         q_vector->tx.total_packets = 0;
4206 }
4207
4208 static void igc_ring_irq_enable(struct igc_q_vector *q_vector)
4209 {
4210         struct igc_adapter *adapter = q_vector->adapter;
4211         struct igc_hw *hw = &adapter->hw;
4212
4213         if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
4214             (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
4215                 if (adapter->num_q_vectors == 1)
4216                         igc_set_itr(q_vector);
4217                 else
4218                         igc_update_ring_itr(q_vector);
4219         }
4220
4221         if (!test_bit(__IGC_DOWN, &adapter->state)) {
4222                 if (adapter->msix_entries)
4223                         wr32(IGC_EIMS, q_vector->eims_value);
4224                 else
4225                         igc_irq_enable(adapter);
4226         }
4227 }
4228
4229 static void igc_add_ring(struct igc_ring *ring,
4230                          struct igc_ring_container *head)
4231 {
4232         head->ring = ring;
4233         head->count++;
4234 }
4235
4236 /**
4237  * igc_cache_ring_register - Descriptor ring to register mapping
4238  * @adapter: board private structure to initialize
4239  *
4240  * Once we know the feature-set enabled for the device, we'll cache
4241  * the register offset the descriptor ring is assigned to.
4242  */
4243 static void igc_cache_ring_register(struct igc_adapter *adapter)
4244 {
4245         int i = 0, j = 0;
4246
4247         switch (adapter->hw.mac.type) {
4248         case igc_i225:
4249         default:
4250                 for (; i < adapter->num_rx_queues; i++)
4251                         adapter->rx_ring[i]->reg_idx = i;
4252                 for (; j < adapter->num_tx_queues; j++)
4253                         adapter->tx_ring[j]->reg_idx = j;
4254                 break;
4255         }
4256 }
4257
4258 /**
4259  * igc_poll - NAPI Rx polling callback
4260  * @napi: napi polling structure
4261  * @budget: count of how many packets we should handle
4262  */
4263 static int igc_poll(struct napi_struct *napi, int budget)
4264 {
4265         struct igc_q_vector *q_vector = container_of(napi,
4266                                                      struct igc_q_vector,
4267                                                      napi);
4268         struct igc_ring *rx_ring = q_vector->rx.ring;
4269         bool clean_complete = true;
4270         int work_done = 0;
4271
4272         if (q_vector->tx.ring)
4273                 clean_complete = igc_clean_tx_irq(q_vector, budget);
4274
4275         if (rx_ring) {
4276                 int cleaned = rx_ring->xsk_pool ?
4277                               igc_clean_rx_irq_zc(q_vector, budget) :
4278                               igc_clean_rx_irq(q_vector, budget);
4279
4280                 work_done += cleaned;
4281                 if (cleaned >= budget)
4282                         clean_complete = false;
4283         }
4284
4285         /* If all work not completed, return budget and keep polling */
4286         if (!clean_complete)
4287                 return budget;
4288
4289         /* Exit the polling mode, but don't re-enable interrupts if stack might
4290          * poll us due to busy-polling
4291          */
4292         if (likely(napi_complete_done(napi, work_done)))
4293                 igc_ring_irq_enable(q_vector);
4294
4295         return min(work_done, budget - 1);
4296 }
4297
4298 /**
4299  * igc_alloc_q_vector - Allocate memory for a single interrupt vector
4300  * @adapter: board private structure to initialize
4301  * @v_count: q_vectors allocated on adapter, used for ring interleaving
4302  * @v_idx: index of vector in adapter struct
4303  * @txr_count: total number of Tx rings to allocate
4304  * @txr_idx: index of first Tx ring to allocate
4305  * @rxr_count: total number of Rx rings to allocate
4306  * @rxr_idx: index of first Rx ring to allocate
4307  *
4308  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
4309  */
4310 static int igc_alloc_q_vector(struct igc_adapter *adapter,
4311                               unsigned int v_count, unsigned int v_idx,
4312                               unsigned int txr_count, unsigned int txr_idx,
4313                               unsigned int rxr_count, unsigned int rxr_idx)
4314 {
4315         struct igc_q_vector *q_vector;
4316         struct igc_ring *ring;
4317         int ring_count;
4318
4319         /* igc only supports 1 Tx and/or 1 Rx queue per vector */
4320         if (txr_count > 1 || rxr_count > 1)
4321                 return -ENOMEM;
4322
4323         ring_count = txr_count + rxr_count;
4324
4325         /* allocate q_vector and rings */
4326         q_vector = adapter->q_vector[v_idx];
4327         if (!q_vector)
4328                 q_vector = kzalloc(struct_size(q_vector, ring, ring_count),
4329                                    GFP_KERNEL);
4330         else
4331                 memset(q_vector, 0, struct_size(q_vector, ring, ring_count));
4332         if (!q_vector)
4333                 return -ENOMEM;
4334
4335         /* initialize NAPI */
4336         netif_napi_add(adapter->netdev, &q_vector->napi,
4337                        igc_poll, 64);
4338
4339         /* tie q_vector and adapter together */
4340         adapter->q_vector[v_idx] = q_vector;
4341         q_vector->adapter = adapter;
4342
4343         /* initialize work limits */
4344         q_vector->tx.work_limit = adapter->tx_work_limit;
4345
4346         /* initialize ITR configuration */
4347         q_vector->itr_register = adapter->io_addr + IGC_EITR(0);
4348         q_vector->itr_val = IGC_START_ITR;
4349
4350         /* initialize pointer to rings */
4351         ring = q_vector->ring;
4352
4353         /* initialize ITR */
4354         if (rxr_count) {
4355                 /* rx or rx/tx vector */
4356                 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
4357                         q_vector->itr_val = adapter->rx_itr_setting;
4358         } else {
4359                 /* tx only vector */
4360                 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
4361                         q_vector->itr_val = adapter->tx_itr_setting;
4362         }
4363
4364         if (txr_count) {
4365                 /* assign generic ring traits */
4366                 ring->dev = &adapter->pdev->dev;
4367                 ring->netdev = adapter->netdev;
4368
4369                 /* configure backlink on ring */
4370                 ring->q_vector = q_vector;
4371
4372                 /* update q_vector Tx values */
4373                 igc_add_ring(ring, &q_vector->tx);
4374
4375                 /* apply Tx specific ring traits */
4376                 ring->count = adapter->tx_ring_count;
4377                 ring->queue_index = txr_idx;
4378
4379                 /* assign ring to adapter */
4380                 adapter->tx_ring[txr_idx] = ring;
4381
4382                 /* push pointer to next ring */
4383                 ring++;
4384         }
4385
4386         if (rxr_count) {
4387                 /* assign generic ring traits */
4388                 ring->dev = &adapter->pdev->dev;
4389                 ring->netdev = adapter->netdev;
4390
4391                 /* configure backlink on ring */
4392                 ring->q_vector = q_vector;
4393
4394                 /* update q_vector Rx values */
4395                 igc_add_ring(ring, &q_vector->rx);
4396
4397                 /* apply Rx specific ring traits */
4398                 ring->count = adapter->rx_ring_count;
4399                 ring->queue_index = rxr_idx;
4400
4401                 /* assign ring to adapter */
4402                 adapter->rx_ring[rxr_idx] = ring;
4403         }
4404
4405         return 0;
4406 }
4407
4408 /**
4409  * igc_alloc_q_vectors - Allocate memory for interrupt vectors
4410  * @adapter: board private structure to initialize
4411  *
4412  * We allocate one q_vector per queue interrupt.  If allocation fails we
4413  * return -ENOMEM.
4414  */
4415 static int igc_alloc_q_vectors(struct igc_adapter *adapter)
4416 {
4417         int rxr_remaining = adapter->num_rx_queues;
4418         int txr_remaining = adapter->num_tx_queues;
4419         int rxr_idx = 0, txr_idx = 0, v_idx = 0;
4420         int q_vectors = adapter->num_q_vectors;
4421         int err;
4422
4423         if (q_vectors >= (rxr_remaining + txr_remaining)) {
4424                 for (; rxr_remaining; v_idx++) {
4425                         err = igc_alloc_q_vector(adapter, q_vectors, v_idx,
4426                                                  0, 0, 1, rxr_idx);
4427
4428                         if (err)
4429                                 goto err_out;
4430
4431                         /* update counts and index */
4432                         rxr_remaining--;
4433                         rxr_idx++;
4434                 }
4435         }
4436
4437         for (; v_idx < q_vectors; v_idx++) {
4438                 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
4439                 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
4440
4441                 err = igc_alloc_q_vector(adapter, q_vectors, v_idx,
4442                                          tqpv, txr_idx, rqpv, rxr_idx);
4443
4444                 if (err)
4445                         goto err_out;
4446
4447                 /* update counts and index */
4448                 rxr_remaining -= rqpv;
4449                 txr_remaining -= tqpv;
4450                 rxr_idx++;
4451                 txr_idx++;
4452         }
4453
4454         return 0;
4455
4456 err_out:
4457         adapter->num_tx_queues = 0;
4458         adapter->num_rx_queues = 0;
4459         adapter->num_q_vectors = 0;
4460
4461         while (v_idx--)
4462                 igc_free_q_vector(adapter, v_idx);
4463
4464         return -ENOMEM;
4465 }
4466
4467 /**
4468  * igc_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
4469  * @adapter: Pointer to adapter structure
4470  * @msix: boolean for MSI-X capability
4471  *
4472  * This function initializes the interrupts and allocates all of the queues.
4473  */
4474 static int igc_init_interrupt_scheme(struct igc_adapter *adapter, bool msix)
4475 {
4476         struct net_device *dev = adapter->netdev;
4477         int err = 0;
4478
4479         igc_set_interrupt_capability(adapter, msix);
4480
4481         err = igc_alloc_q_vectors(adapter);
4482         if (err) {
4483                 netdev_err(dev, "Unable to allocate memory for vectors\n");
4484                 goto err_alloc_q_vectors;
4485         }
4486
4487         igc_cache_ring_register(adapter);
4488
4489         return 0;
4490
4491 err_alloc_q_vectors:
4492         igc_reset_interrupt_capability(adapter);
4493         return err;
4494 }
4495
4496 /**
4497  * igc_sw_init - Initialize general software structures (struct igc_adapter)
4498  * @adapter: board private structure to initialize
4499  *
4500  * igc_sw_init initializes the Adapter private data structure.
4501  * Fields are initialized based on PCI device information and
4502  * OS network device settings (MTU size).
4503  */
4504 static int igc_sw_init(struct igc_adapter *adapter)
4505 {
4506         struct net_device *netdev = adapter->netdev;
4507         struct pci_dev *pdev = adapter->pdev;
4508         struct igc_hw *hw = &adapter->hw;
4509
4510         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
4511
4512         /* set default ring sizes */
4513         adapter->tx_ring_count = IGC_DEFAULT_TXD;
4514         adapter->rx_ring_count = IGC_DEFAULT_RXD;
4515
4516         /* set default ITR values */
4517         adapter->rx_itr_setting = IGC_DEFAULT_ITR;
4518         adapter->tx_itr_setting = IGC_DEFAULT_ITR;
4519
4520         /* set default work limits */
4521         adapter->tx_work_limit = IGC_DEFAULT_TX_WORK;
4522
4523         /* adjust max frame to be at least the size of a standard frame */
4524         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
4525                                 VLAN_HLEN;
4526         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4527
4528         mutex_init(&adapter->nfc_rule_lock);
4529         INIT_LIST_HEAD(&adapter->nfc_rule_list);
4530         adapter->nfc_rule_count = 0;
4531
4532         spin_lock_init(&adapter->stats64_lock);
4533         /* Assume MSI-X interrupts, will be checked during IRQ allocation */
4534         adapter->flags |= IGC_FLAG_HAS_MSIX;
4535
4536         igc_init_queue_configuration(adapter);
4537
4538         /* This call may decrease the number of queues */
4539         if (igc_init_interrupt_scheme(adapter, true)) {
4540                 netdev_err(netdev, "Unable to allocate memory for queues\n");
4541                 return -ENOMEM;
4542         }
4543
4544         /* Explicitly disable IRQ since the NIC can be in any state. */
4545         igc_irq_disable(adapter);
4546
4547         set_bit(__IGC_DOWN, &adapter->state);
4548
4549         return 0;
4550 }
4551
4552 /**
4553  * igc_up - Open the interface and prepare it to handle traffic
4554  * @adapter: board private structure
4555  */
4556 void igc_up(struct igc_adapter *adapter)
4557 {
4558         struct igc_hw *hw = &adapter->hw;
4559         int i = 0;
4560
4561         /* hardware has been reset, we need to reload some things */
4562         igc_configure(adapter);
4563
4564         clear_bit(__IGC_DOWN, &adapter->state);
4565
4566         for (i = 0; i < adapter->num_q_vectors; i++)
4567                 napi_enable(&adapter->q_vector[i]->napi);
4568
4569         if (adapter->msix_entries)
4570                 igc_configure_msix(adapter);
4571         else
4572                 igc_assign_vector(adapter->q_vector[0], 0);
4573
4574         /* Clear any pending interrupts. */
4575         rd32(IGC_ICR);
4576         igc_irq_enable(adapter);
4577
4578         netif_tx_start_all_queues(adapter->netdev);
4579
4580         /* start the watchdog. */
4581         hw->mac.get_link_status = true;
4582         schedule_work(&adapter->watchdog_task);
4583 }
4584
4585 /**
4586  * igc_update_stats - Update the board statistics counters
4587  * @adapter: board private structure
4588  */
4589 void igc_update_stats(struct igc_adapter *adapter)
4590 {
4591         struct rtnl_link_stats64 *net_stats = &adapter->stats64;
4592         struct pci_dev *pdev = adapter->pdev;
4593         struct igc_hw *hw = &adapter->hw;
4594         u64 _bytes, _packets;
4595         u64 bytes, packets;
4596         unsigned int start;
4597         u32 mpc;
4598         int i;
4599
4600         /* Prevent stats update while adapter is being reset, or if the pci
4601          * connection is down.
4602          */
4603         if (adapter->link_speed == 0)
4604                 return;
4605         if (pci_channel_offline(pdev))
4606                 return;
4607
4608         packets = 0;
4609         bytes = 0;
4610
4611         rcu_read_lock();
4612         for (i = 0; i < adapter->num_rx_queues; i++) {
4613                 struct igc_ring *ring = adapter->rx_ring[i];
4614                 u32 rqdpc = rd32(IGC_RQDPC(i));
4615
4616                 if (hw->mac.type >= igc_i225)
4617                         wr32(IGC_RQDPC(i), 0);
4618
4619                 if (rqdpc) {
4620                         ring->rx_stats.drops += rqdpc;
4621                         net_stats->rx_fifo_errors += rqdpc;
4622                 }
4623
4624                 do {
4625                         start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
4626                         _bytes = ring->rx_stats.bytes;
4627                         _packets = ring->rx_stats.packets;
4628                 } while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
4629                 bytes += _bytes;
4630                 packets += _packets;
4631         }
4632
4633         net_stats->rx_bytes = bytes;
4634         net_stats->rx_packets = packets;
4635
4636         packets = 0;
4637         bytes = 0;
4638         for (i = 0; i < adapter->num_tx_queues; i++) {
4639                 struct igc_ring *ring = adapter->tx_ring[i];
4640
4641                 do {
4642                         start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
4643                         _bytes = ring->tx_stats.bytes;
4644                         _packets = ring->tx_stats.packets;
4645                 } while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
4646                 bytes += _bytes;
4647                 packets += _packets;
4648         }
4649         net_stats->tx_bytes = bytes;
4650         net_stats->tx_packets = packets;
4651         rcu_read_unlock();
4652
4653         /* read stats registers */
4654         adapter->stats.crcerrs += rd32(IGC_CRCERRS);
4655         adapter->stats.gprc += rd32(IGC_GPRC);
4656         adapter->stats.gorc += rd32(IGC_GORCL);
4657         rd32(IGC_GORCH); /* clear GORCL */
4658         adapter->stats.bprc += rd32(IGC_BPRC);
4659         adapter->stats.mprc += rd32(IGC_MPRC);
4660         adapter->stats.roc += rd32(IGC_ROC);
4661
4662         adapter->stats.prc64 += rd32(IGC_PRC64);
4663         adapter->stats.prc127 += rd32(IGC_PRC127);
4664         adapter->stats.prc255 += rd32(IGC_PRC255);
4665         adapter->stats.prc511 += rd32(IGC_PRC511);
4666         adapter->stats.prc1023 += rd32(IGC_PRC1023);
4667         adapter->stats.prc1522 += rd32(IGC_PRC1522);
4668         adapter->stats.tlpic += rd32(IGC_TLPIC);
4669         adapter->stats.rlpic += rd32(IGC_RLPIC);
4670         adapter->stats.hgptc += rd32(IGC_HGPTC);
4671
4672         mpc = rd32(IGC_MPC);
4673         adapter->stats.mpc += mpc;
4674         net_stats->rx_fifo_errors += mpc;
4675         adapter->stats.scc += rd32(IGC_SCC);
4676         adapter->stats.ecol += rd32(IGC_ECOL);
4677         adapter->stats.mcc += rd32(IGC_MCC);
4678         adapter->stats.latecol += rd32(IGC_LATECOL);
4679         adapter->stats.dc += rd32(IGC_DC);
4680         adapter->stats.rlec += rd32(IGC_RLEC);
4681         adapter->stats.xonrxc += rd32(IGC_XONRXC);
4682         adapter->stats.xontxc += rd32(IGC_XONTXC);
4683         adapter->stats.xoffrxc += rd32(IGC_XOFFRXC);
4684         adapter->stats.xofftxc += rd32(IGC_XOFFTXC);
4685         adapter->stats.fcruc += rd32(IGC_FCRUC);
4686         adapter->stats.gptc += rd32(IGC_GPTC);
4687         adapter->stats.gotc += rd32(IGC_GOTCL);
4688         rd32(IGC_GOTCH); /* clear GOTCL */
4689         adapter->stats.rnbc += rd32(IGC_RNBC);
4690         adapter->stats.ruc += rd32(IGC_RUC);
4691         adapter->stats.rfc += rd32(IGC_RFC);
4692         adapter->stats.rjc += rd32(IGC_RJC);
4693         adapter->stats.tor += rd32(IGC_TORH);
4694         adapter->stats.tot += rd32(IGC_TOTH);
4695         adapter->stats.tpr += rd32(IGC_TPR);
4696
4697         adapter->stats.ptc64 += rd32(IGC_PTC64);
4698         adapter->stats.ptc127 += rd32(IGC_PTC127);
4699         adapter->stats.ptc255 += rd32(IGC_PTC255);
4700         adapter->stats.ptc511 += rd32(IGC_PTC511);
4701         adapter->stats.ptc1023 += rd32(IGC_PTC1023);
4702         adapter->stats.ptc1522 += rd32(IGC_PTC1522);
4703
4704         adapter->stats.mptc += rd32(IGC_MPTC);
4705         adapter->stats.bptc += rd32(IGC_BPTC);
4706
4707         adapter->stats.tpt += rd32(IGC_TPT);
4708         adapter->stats.colc += rd32(IGC_COLC);
4709         adapter->stats.colc += rd32(IGC_RERC);
4710
4711         adapter->stats.algnerrc += rd32(IGC_ALGNERRC);
4712
4713         adapter->stats.tsctc += rd32(IGC_TSCTC);
4714
4715         adapter->stats.iac += rd32(IGC_IAC);
4716
4717         /* Fill out the OS statistics structure */
4718         net_stats->multicast = adapter->stats.mprc;
4719         net_stats->collisions = adapter->stats.colc;
4720
4721         /* Rx Errors */
4722
4723         /* RLEC on some newer hardware can be incorrect so build
4724          * our own version based on RUC and ROC
4725          */
4726         net_stats->rx_errors = adapter->stats.rxerrc +
4727                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4728                 adapter->stats.ruc + adapter->stats.roc +
4729                 adapter->stats.cexterr;
4730         net_stats->rx_length_errors = adapter->stats.ruc +
4731                                       adapter->stats.roc;
4732         net_stats->rx_crc_errors = adapter->stats.crcerrs;
4733         net_stats->rx_frame_errors = adapter->stats.algnerrc;
4734         net_stats->rx_missed_errors = adapter->stats.mpc;
4735
4736         /* Tx Errors */
4737         net_stats->tx_errors = adapter->stats.ecol +
4738                                adapter->stats.latecol;
4739         net_stats->tx_aborted_errors = adapter->stats.ecol;
4740         net_stats->tx_window_errors = adapter->stats.latecol;
4741         net_stats->tx_carrier_errors = adapter->stats.tncrs;
4742
4743         /* Tx Dropped needs to be maintained elsewhere */
4744
4745         /* Management Stats */
4746         adapter->stats.mgptc += rd32(IGC_MGTPTC);
4747         adapter->stats.mgprc += rd32(IGC_MGTPRC);
4748         adapter->stats.mgpdc += rd32(IGC_MGTPDC);
4749 }
4750
4751 /**
4752  * igc_down - Close the interface
4753  * @adapter: board private structure
4754  */
4755 void igc_down(struct igc_adapter *adapter)
4756 {
4757         struct net_device *netdev = adapter->netdev;
4758         struct igc_hw *hw = &adapter->hw;
4759         u32 tctl, rctl;
4760         int i = 0;
4761
4762         set_bit(__IGC_DOWN, &adapter->state);
4763
4764         igc_ptp_suspend(adapter);
4765
4766         /* disable receives in the hardware */
4767         rctl = rd32(IGC_RCTL);
4768         wr32(IGC_RCTL, rctl & ~IGC_RCTL_EN);
4769         /* flush and sleep below */
4770
4771         /* set trans_start so we don't get spurious watchdogs during reset */
4772         netif_trans_update(netdev);
4773
4774         netif_carrier_off(netdev);
4775         netif_tx_stop_all_queues(netdev);
4776
4777         /* disable transmits in the hardware */
4778         tctl = rd32(IGC_TCTL);
4779         tctl &= ~IGC_TCTL_EN;
4780         wr32(IGC_TCTL, tctl);
4781         /* flush both disables and wait for them to finish */
4782         wrfl();
4783         usleep_range(10000, 20000);
4784
4785         igc_irq_disable(adapter);
4786
4787         adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
4788
4789         for (i = 0; i < adapter->num_q_vectors; i++) {
4790                 if (adapter->q_vector[i]) {
4791                         napi_synchronize(&adapter->q_vector[i]->napi);
4792                         napi_disable(&adapter->q_vector[i]->napi);
4793                 }
4794         }
4795
4796         del_timer_sync(&adapter->watchdog_timer);
4797         del_timer_sync(&adapter->phy_info_timer);
4798
4799         /* record the stats before reset*/
4800         spin_lock(&adapter->stats64_lock);
4801         igc_update_stats(adapter);
4802         spin_unlock(&adapter->stats64_lock);
4803
4804         adapter->link_speed = 0;
4805         adapter->link_duplex = 0;
4806
4807         if (!pci_channel_offline(adapter->pdev))
4808                 igc_reset(adapter);
4809
4810         /* clear VLAN promisc flag so VFTA will be updated if necessary */
4811         adapter->flags &= ~IGC_FLAG_VLAN_PROMISC;
4812
4813         igc_clean_all_tx_rings(adapter);
4814         igc_clean_all_rx_rings(adapter);
4815 }
4816
4817 void igc_reinit_locked(struct igc_adapter *adapter)
4818 {
4819         while (test_and_set_bit(__IGC_RESETTING, &adapter->state))
4820                 usleep_range(1000, 2000);
4821         igc_down(adapter);
4822         igc_up(adapter);
4823         clear_bit(__IGC_RESETTING, &adapter->state);
4824 }
4825
4826 static void igc_reset_task(struct work_struct *work)
4827 {
4828         struct igc_adapter *adapter;
4829
4830         adapter = container_of(work, struct igc_adapter, reset_task);
4831
4832         rtnl_lock();
4833         /* If we're already down or resetting, just bail */
4834         if (test_bit(__IGC_DOWN, &adapter->state) ||
4835             test_bit(__IGC_RESETTING, &adapter->state)) {
4836                 rtnl_unlock();
4837                 return;
4838         }
4839
4840         igc_rings_dump(adapter);
4841         igc_regs_dump(adapter);
4842         netdev_err(adapter->netdev, "Reset adapter\n");
4843         igc_reinit_locked(adapter);
4844         rtnl_unlock();
4845 }
4846
4847 /**
4848  * igc_change_mtu - Change the Maximum Transfer Unit
4849  * @netdev: network interface device structure
4850  * @new_mtu: new value for maximum frame size
4851  *
4852  * Returns 0 on success, negative on failure
4853  */
4854 static int igc_change_mtu(struct net_device *netdev, int new_mtu)
4855 {
4856         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4857         struct igc_adapter *adapter = netdev_priv(netdev);
4858
4859         if (igc_xdp_is_enabled(adapter) && new_mtu > ETH_DATA_LEN) {
4860                 netdev_dbg(netdev, "Jumbo frames not supported with XDP");
4861                 return -EINVAL;
4862         }
4863
4864         /* adjust max frame to be at least the size of a standard frame */
4865         if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4866                 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
4867
4868         while (test_and_set_bit(__IGC_RESETTING, &adapter->state))
4869                 usleep_range(1000, 2000);
4870
4871         /* igc_down has a dependency on max_frame_size */
4872         adapter->max_frame_size = max_frame;
4873
4874         if (netif_running(netdev))
4875                 igc_down(adapter);
4876
4877         netdev_dbg(netdev, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
4878         netdev->mtu = new_mtu;
4879
4880         if (netif_running(netdev))
4881                 igc_up(adapter);
4882         else
4883                 igc_reset(adapter);
4884
4885         clear_bit(__IGC_RESETTING, &adapter->state);
4886
4887         return 0;
4888 }
4889
4890 /**
4891  * igc_get_stats64 - Get System Network Statistics
4892  * @netdev: network interface device structure
4893  * @stats: rtnl_link_stats64 pointer
4894  *
4895  * Returns the address of the device statistics structure.
4896  * The statistics are updated here and also from the timer callback.
4897  */
4898 static void igc_get_stats64(struct net_device *netdev,
4899                             struct rtnl_link_stats64 *stats)
4900 {
4901         struct igc_adapter *adapter = netdev_priv(netdev);
4902
4903         spin_lock(&adapter->stats64_lock);
4904         if (!test_bit(__IGC_RESETTING, &adapter->state))
4905                 igc_update_stats(adapter);
4906         memcpy(stats, &adapter->stats64, sizeof(*stats));
4907         spin_unlock(&adapter->stats64_lock);
4908 }
4909
4910 static netdev_features_t igc_fix_features(struct net_device *netdev,
4911                                           netdev_features_t features)
4912 {
4913         /* Since there is no support for separate Rx/Tx vlan accel
4914          * enable/disable make sure Tx flag is always in same state as Rx.
4915          */
4916         if (features & NETIF_F_HW_VLAN_CTAG_RX)
4917                 features |= NETIF_F_HW_VLAN_CTAG_TX;
4918         else
4919                 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
4920
4921         return features;
4922 }
4923
4924 static int igc_set_features(struct net_device *netdev,
4925                             netdev_features_t features)
4926 {
4927         netdev_features_t changed = netdev->features ^ features;
4928         struct igc_adapter *adapter = netdev_priv(netdev);
4929
4930         if (changed & NETIF_F_HW_VLAN_CTAG_RX)
4931                 igc_vlan_mode(netdev, features);
4932
4933         /* Add VLAN support */
4934         if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
4935                 return 0;
4936
4937         if (!(features & NETIF_F_NTUPLE))
4938                 igc_flush_nfc_rules(adapter);
4939
4940         netdev->features = features;
4941
4942         if (netif_running(netdev))
4943                 igc_reinit_locked(adapter);
4944         else
4945                 igc_reset(adapter);
4946
4947         return 1;
4948 }
4949
4950 static netdev_features_t
4951 igc_features_check(struct sk_buff *skb, struct net_device *dev,
4952                    netdev_features_t features)
4953 {
4954         unsigned int network_hdr_len, mac_hdr_len;
4955
4956         /* Make certain the headers can be described by a context descriptor */
4957         mac_hdr_len = skb_network_header(skb) - skb->data;
4958         if (unlikely(mac_hdr_len > IGC_MAX_MAC_HDR_LEN))
4959                 return features & ~(NETIF_F_HW_CSUM |
4960                                     NETIF_F_SCTP_CRC |
4961                                     NETIF_F_HW_VLAN_CTAG_TX |
4962                                     NETIF_F_TSO |
4963                                     NETIF_F_TSO6);
4964
4965         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
4966         if (unlikely(network_hdr_len >  IGC_MAX_NETWORK_HDR_LEN))
4967                 return features & ~(NETIF_F_HW_CSUM |
4968                                     NETIF_F_SCTP_CRC |
4969                                     NETIF_F_TSO |
4970                                     NETIF_F_TSO6);
4971
4972         /* We can only support IPv4 TSO in tunnels if we can mangle the
4973          * inner IP ID field, so strip TSO if MANGLEID is not supported.
4974          */
4975         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
4976                 features &= ~NETIF_F_TSO;
4977
4978         return features;
4979 }
4980
4981 static void igc_tsync_interrupt(struct igc_adapter *adapter)
4982 {
4983         u32 ack, tsauxc, sec, nsec, tsicr;
4984         struct igc_hw *hw = &adapter->hw;
4985         struct ptp_clock_event event;
4986         struct timespec64 ts;
4987
4988         tsicr = rd32(IGC_TSICR);
4989         ack = 0;
4990
4991         if (tsicr & IGC_TSICR_SYS_WRAP) {
4992                 event.type = PTP_CLOCK_PPS;
4993                 if (adapter->ptp_caps.pps)
4994                         ptp_clock_event(adapter->ptp_clock, &event);
4995                 ack |= IGC_TSICR_SYS_WRAP;
4996         }
4997
4998         if (tsicr & IGC_TSICR_TXTS) {
4999                 /* retrieve hardware timestamp */
5000                 schedule_work(&adapter->ptp_tx_work);
5001                 ack |= IGC_TSICR_TXTS;
5002         }
5003
5004         if (tsicr & IGC_TSICR_TT0) {
5005                 spin_lock(&adapter->tmreg_lock);
5006                 ts = timespec64_add(adapter->perout[0].start,
5007                                     adapter->perout[0].period);
5008                 wr32(IGC_TRGTTIML0, ts.tv_nsec | IGC_TT_IO_TIMER_SEL_SYSTIM0);
5009                 wr32(IGC_TRGTTIMH0, (u32)ts.tv_sec);
5010                 tsauxc = rd32(IGC_TSAUXC);
5011                 tsauxc |= IGC_TSAUXC_EN_TT0;
5012                 wr32(IGC_TSAUXC, tsauxc);
5013                 adapter->perout[0].start = ts;
5014                 spin_unlock(&adapter->tmreg_lock);
5015                 ack |= IGC_TSICR_TT0;
5016         }
5017
5018         if (tsicr & IGC_TSICR_TT1) {
5019                 spin_lock(&adapter->tmreg_lock);
5020                 ts = timespec64_add(adapter->perout[1].start,
5021                                     adapter->perout[1].period);
5022                 wr32(IGC_TRGTTIML1, ts.tv_nsec | IGC_TT_IO_TIMER_SEL_SYSTIM0);
5023                 wr32(IGC_TRGTTIMH1, (u32)ts.tv_sec);
5024                 tsauxc = rd32(IGC_TSAUXC);
5025                 tsauxc |= IGC_TSAUXC_EN_TT1;
5026                 wr32(IGC_TSAUXC, tsauxc);
5027                 adapter->perout[1].start = ts;
5028                 spin_unlock(&adapter->tmreg_lock);
5029                 ack |= IGC_TSICR_TT1;
5030         }
5031
5032         if (tsicr & IGC_TSICR_AUTT0) {
5033                 nsec = rd32(IGC_AUXSTMPL0);
5034                 sec  = rd32(IGC_AUXSTMPH0);
5035                 event.type = PTP_CLOCK_EXTTS;
5036                 event.index = 0;
5037                 event.timestamp = sec * NSEC_PER_SEC + nsec;
5038                 ptp_clock_event(adapter->ptp_clock, &event);
5039                 ack |= IGC_TSICR_AUTT0;
5040         }
5041
5042         if (tsicr & IGC_TSICR_AUTT1) {
5043                 nsec = rd32(IGC_AUXSTMPL1);
5044                 sec  = rd32(IGC_AUXSTMPH1);
5045                 event.type = PTP_CLOCK_EXTTS;
5046                 event.index = 1;
5047                 event.timestamp = sec * NSEC_PER_SEC + nsec;
5048                 ptp_clock_event(adapter->ptp_clock, &event);
5049                 ack |= IGC_TSICR_AUTT1;
5050         }
5051
5052         /* acknowledge the interrupts */
5053         wr32(IGC_TSICR, ack);
5054 }
5055
5056 /**
5057  * igc_msix_other - msix other interrupt handler
5058  * @irq: interrupt number
5059  * @data: pointer to a q_vector
5060  */
5061 static irqreturn_t igc_msix_other(int irq, void *data)
5062 {
5063         struct igc_adapter *adapter = data;
5064         struct igc_hw *hw = &adapter->hw;
5065         u32 icr = rd32(IGC_ICR);
5066
5067         /* reading ICR causes bit 31 of EICR to be cleared */
5068         if (icr & IGC_ICR_DRSTA)
5069                 schedule_work(&adapter->reset_task);
5070
5071         if (icr & IGC_ICR_DOUTSYNC) {
5072                 /* HW is reporting DMA is out of sync */
5073                 adapter->stats.doosync++;
5074         }
5075
5076         if (icr & IGC_ICR_LSC) {
5077                 hw->mac.get_link_status = true;
5078                 /* guard against interrupt when we're going down */
5079                 if (!test_bit(__IGC_DOWN, &adapter->state))
5080                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
5081         }
5082
5083         if (icr & IGC_ICR_TS)
5084                 igc_tsync_interrupt(adapter);
5085
5086         wr32(IGC_EIMS, adapter->eims_other);
5087
5088         return IRQ_HANDLED;
5089 }
5090
5091 static void igc_write_itr(struct igc_q_vector *q_vector)
5092 {
5093         u32 itr_val = q_vector->itr_val & IGC_QVECTOR_MASK;
5094
5095         if (!q_vector->set_itr)
5096                 return;
5097
5098         if (!itr_val)
5099                 itr_val = IGC_ITR_VAL_MASK;
5100
5101         itr_val |= IGC_EITR_CNT_IGNR;
5102
5103         writel(itr_val, q_vector->itr_register);
5104         q_vector->set_itr = 0;
5105 }
5106
5107 static irqreturn_t igc_msix_ring(int irq, void *data)
5108 {
5109         struct igc_q_vector *q_vector = data;
5110
5111         /* Write the ITR value calculated from the previous interrupt. */
5112         igc_write_itr(q_vector);
5113
5114         napi_schedule(&q_vector->napi);
5115
5116         return IRQ_HANDLED;
5117 }
5118
5119 /**
5120  * igc_request_msix - Initialize MSI-X interrupts
5121  * @adapter: Pointer to adapter structure
5122  *
5123  * igc_request_msix allocates MSI-X vectors and requests interrupts from the
5124  * kernel.
5125  */
5126 static int igc_request_msix(struct igc_adapter *adapter)
5127 {
5128         unsigned int num_q_vectors = adapter->num_q_vectors;
5129         int i = 0, err = 0, vector = 0, free_vector = 0;
5130         struct net_device *netdev = adapter->netdev;
5131
5132         err = request_irq(adapter->msix_entries[vector].vector,
5133                           &igc_msix_other, 0, netdev->name, adapter);
5134         if (err)
5135                 goto err_out;
5136
5137         if (num_q_vectors > MAX_Q_VECTORS) {
5138                 num_q_vectors = MAX_Q_VECTORS;
5139                 dev_warn(&adapter->pdev->dev,
5140                          "The number of queue vectors (%d) is higher than max allowed (%d)\n",
5141                          adapter->num_q_vectors, MAX_Q_VECTORS);
5142         }
5143         for (i = 0; i < num_q_vectors; i++) {
5144                 struct igc_q_vector *q_vector = adapter->q_vector[i];
5145
5146                 vector++;
5147
5148                 q_vector->itr_register = adapter->io_addr + IGC_EITR(vector);
5149
5150                 if (q_vector->rx.ring && q_vector->tx.ring)
5151                         sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
5152                                 q_vector->rx.ring->queue_index);
5153                 else if (q_vector->tx.ring)
5154                         sprintf(q_vector->name, "%s-tx-%u", netdev->name,
5155                                 q_vector->tx.ring->queue_index);
5156                 else if (q_vector->rx.ring)
5157                         sprintf(q_vector->name, "%s-rx-%u", netdev->name,
5158                                 q_vector->rx.ring->queue_index);
5159                 else
5160                         sprintf(q_vector->name, "%s-unused", netdev->name);
5161
5162                 err = request_irq(adapter->msix_entries[vector].vector,
5163                                   igc_msix_ring, 0, q_vector->name,
5164                                   q_vector);
5165                 if (err)
5166                         goto err_free;
5167         }
5168
5169         igc_configure_msix(adapter);
5170         return 0;
5171
5172 err_free:
5173         /* free already assigned IRQs */
5174         free_irq(adapter->msix_entries[free_vector++].vector, adapter);
5175
5176         vector--;
5177         for (i = 0; i < vector; i++) {
5178                 free_irq(adapter->msix_entries[free_vector++].vector,
5179                          adapter->q_vector[i]);
5180         }
5181 err_out:
5182         return err;
5183 }
5184
5185 /**
5186  * igc_clear_interrupt_scheme - reset the device to a state of no interrupts
5187  * @adapter: Pointer to adapter structure
5188  *
5189  * This function resets the device so that it has 0 rx queues, tx queues, and
5190  * MSI-X interrupts allocated.
5191  */
5192 static void igc_clear_interrupt_scheme(struct igc_adapter *adapter)
5193 {
5194         igc_free_q_vectors(adapter);
5195         igc_reset_interrupt_capability(adapter);
5196 }
5197
5198 /* Need to wait a few seconds after link up to get diagnostic information from
5199  * the phy
5200  */
5201 static void igc_update_phy_info(struct timer_list *t)
5202 {
5203         struct igc_adapter *adapter = from_timer(adapter, t, phy_info_timer);
5204
5205         igc_get_phy_info(&adapter->hw);
5206 }
5207
5208 /**
5209  * igc_has_link - check shared code for link and determine up/down
5210  * @adapter: pointer to driver private info
5211  */
5212 bool igc_has_link(struct igc_adapter *adapter)
5213 {
5214         struct igc_hw *hw = &adapter->hw;
5215         bool link_active = false;
5216
5217         /* get_link_status is set on LSC (link status) interrupt or
5218          * rx sequence error interrupt.  get_link_status will stay
5219          * false until the igc_check_for_link establishes link
5220          * for copper adapters ONLY
5221          */
5222         if (!hw->mac.get_link_status)
5223                 return true;
5224         hw->mac.ops.check_for_link(hw);
5225         link_active = !hw->mac.get_link_status;
5226
5227         if (hw->mac.type == igc_i225) {
5228                 if (!netif_carrier_ok(adapter->netdev)) {
5229                         adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
5230                 } else if (!(adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)) {
5231                         adapter->flags |= IGC_FLAG_NEED_LINK_UPDATE;
5232                         adapter->link_check_timeout = jiffies;
5233                 }
5234         }
5235
5236         return link_active;
5237 }
5238
5239 /**
5240  * igc_watchdog - Timer Call-back
5241  * @t: timer for the watchdog
5242  */
5243 static void igc_watchdog(struct timer_list *t)
5244 {
5245         struct igc_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5246         /* Do the rest outside of interrupt context */
5247         schedule_work(&adapter->watchdog_task);
5248 }
5249
5250 static void igc_watchdog_task(struct work_struct *work)
5251 {
5252         struct igc_adapter *adapter = container_of(work,
5253                                                    struct igc_adapter,
5254                                                    watchdog_task);
5255         struct net_device *netdev = adapter->netdev;
5256         struct igc_hw *hw = &adapter->hw;
5257         struct igc_phy_info *phy = &hw->phy;
5258         u16 phy_data, retry_count = 20;
5259         u32 link;
5260         int i;
5261
5262         link = igc_has_link(adapter);
5263
5264         if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE) {
5265                 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
5266                         adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
5267                 else
5268                         link = false;
5269         }
5270
5271         if (link) {
5272                 /* Cancel scheduled suspend requests. */
5273                 pm_runtime_resume(netdev->dev.parent);
5274
5275                 if (!netif_carrier_ok(netdev)) {
5276                         u32 ctrl;
5277
5278                         hw->mac.ops.get_speed_and_duplex(hw,
5279                                                          &adapter->link_speed,
5280                                                          &adapter->link_duplex);
5281
5282                         ctrl = rd32(IGC_CTRL);
5283                         /* Link status message must follow this format */
5284                         netdev_info(netdev,
5285                                     "NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5286                                     adapter->link_speed,
5287                                     adapter->link_duplex == FULL_DUPLEX ?
5288                                     "Full" : "Half",
5289                                     (ctrl & IGC_CTRL_TFCE) &&
5290                                     (ctrl & IGC_CTRL_RFCE) ? "RX/TX" :
5291                                     (ctrl & IGC_CTRL_RFCE) ?  "RX" :
5292                                     (ctrl & IGC_CTRL_TFCE) ?  "TX" : "None");
5293
5294                         /* disable EEE if enabled */
5295                         if ((adapter->flags & IGC_FLAG_EEE) &&
5296                             adapter->link_duplex == HALF_DUPLEX) {
5297                                 netdev_info(netdev,
5298                                             "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex\n");
5299                                 adapter->hw.dev_spec._base.eee_enable = false;
5300                                 adapter->flags &= ~IGC_FLAG_EEE;
5301                         }
5302
5303                         /* check if SmartSpeed worked */
5304                         igc_check_downshift(hw);
5305                         if (phy->speed_downgraded)
5306                                 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
5307
5308                         /* adjust timeout factor according to speed/duplex */
5309                         adapter->tx_timeout_factor = 1;
5310                         switch (adapter->link_speed) {
5311                         case SPEED_10:
5312                                 adapter->tx_timeout_factor = 14;
5313                                 break;
5314                         case SPEED_100:
5315                         case SPEED_1000:
5316                         case SPEED_2500:
5317                                 adapter->tx_timeout_factor = 7;
5318                                 break;
5319                         }
5320
5321                         if (adapter->link_speed != SPEED_1000)
5322                                 goto no_wait;
5323
5324                         /* wait for Remote receiver status OK */
5325 retry_read_status:
5326                         if (!igc_read_phy_reg(hw, PHY_1000T_STATUS,
5327                                               &phy_data)) {
5328                                 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5329                                     retry_count) {
5330                                         msleep(100);
5331                                         retry_count--;
5332                                         goto retry_read_status;
5333                                 } else if (!retry_count) {
5334                                         netdev_err(netdev, "exceed max 2 second\n");
5335                                 }
5336                         } else {
5337                                 netdev_err(netdev, "read 1000Base-T Status Reg\n");
5338                         }
5339 no_wait:
5340                         netif_carrier_on(netdev);
5341
5342                         /* link state has changed, schedule phy info update */
5343                         if (!test_bit(__IGC_DOWN, &adapter->state))
5344                                 mod_timer(&adapter->phy_info_timer,
5345                                           round_jiffies(jiffies + 2 * HZ));
5346                 }
5347         } else {
5348                 if (netif_carrier_ok(netdev)) {
5349                         adapter->link_speed = 0;
5350                         adapter->link_duplex = 0;
5351
5352                         /* Links status message must follow this format */
5353                         netdev_info(netdev, "NIC Link is Down\n");
5354                         netif_carrier_off(netdev);
5355
5356                         /* link state has changed, schedule phy info update */
5357                         if (!test_bit(__IGC_DOWN, &adapter->state))
5358                                 mod_timer(&adapter->phy_info_timer,
5359                                           round_jiffies(jiffies + 2 * HZ));
5360
5361                         /* link is down, time to check for alternate media */
5362                         if (adapter->flags & IGC_FLAG_MAS_ENABLE) {
5363                                 if (adapter->flags & IGC_FLAG_MEDIA_RESET) {
5364                                         schedule_work(&adapter->reset_task);
5365                                         /* return immediately */
5366                                         return;
5367                                 }
5368                         }
5369                         pm_schedule_suspend(netdev->dev.parent,
5370                                             MSEC_PER_SEC * 5);
5371
5372                 /* also check for alternate media here */
5373                 } else if (!netif_carrier_ok(netdev) &&
5374                            (adapter->flags & IGC_FLAG_MAS_ENABLE)) {
5375                         if (adapter->flags & IGC_FLAG_MEDIA_RESET) {
5376                                 schedule_work(&adapter->reset_task);
5377                                 /* return immediately */
5378                                 return;
5379                         }
5380                 }
5381         }
5382
5383         spin_lock(&adapter->stats64_lock);
5384         igc_update_stats(adapter);
5385         spin_unlock(&adapter->stats64_lock);
5386
5387         for (i = 0; i < adapter->num_tx_queues; i++) {
5388                 struct igc_ring *tx_ring = adapter->tx_ring[i];
5389
5390                 if (!netif_carrier_ok(netdev)) {
5391                         /* We've lost link, so the controller stops DMA,
5392                          * but we've got queued Tx work that's never going
5393                          * to get done, so reset controller to flush Tx.
5394                          * (Do the reset outside of interrupt context).
5395                          */
5396                         if (igc_desc_unused(tx_ring) + 1 < tx_ring->count) {
5397                                 adapter->tx_timeout_count++;
5398                                 schedule_work(&adapter->reset_task);
5399                                 /* return immediately since reset is imminent */
5400                                 return;
5401                         }
5402                 }
5403
5404                 /* Force detection of hung controller every watchdog period */
5405                 set_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5406         }
5407
5408         /* Cause software interrupt to ensure Rx ring is cleaned */
5409         if (adapter->flags & IGC_FLAG_HAS_MSIX) {
5410                 u32 eics = 0;
5411
5412                 for (i = 0; i < adapter->num_q_vectors; i++)
5413                         eics |= adapter->q_vector[i]->eims_value;
5414                 wr32(IGC_EICS, eics);
5415         } else {
5416                 wr32(IGC_ICS, IGC_ICS_RXDMT0);
5417         }
5418
5419         igc_ptp_tx_hang(adapter);
5420
5421         /* Reset the timer */
5422         if (!test_bit(__IGC_DOWN, &adapter->state)) {
5423                 if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)
5424                         mod_timer(&adapter->watchdog_timer,
5425                                   round_jiffies(jiffies +  HZ));
5426                 else
5427                         mod_timer(&adapter->watchdog_timer,
5428                                   round_jiffies(jiffies + 2 * HZ));
5429         }
5430 }
5431
5432 /**
5433  * igc_intr_msi - Interrupt Handler
5434  * @irq: interrupt number
5435  * @data: pointer to a network interface device structure
5436  */
5437 static irqreturn_t igc_intr_msi(int irq, void *data)
5438 {
5439         struct igc_adapter *adapter = data;
5440         struct igc_q_vector *q_vector = adapter->q_vector[0];
5441         struct igc_hw *hw = &adapter->hw;
5442         /* read ICR disables interrupts using IAM */
5443         u32 icr = rd32(IGC_ICR);
5444
5445         igc_write_itr(q_vector);
5446
5447         if (icr & IGC_ICR_DRSTA)
5448                 schedule_work(&adapter->reset_task);
5449
5450         if (icr & IGC_ICR_DOUTSYNC) {
5451                 /* HW is reporting DMA is out of sync */
5452                 adapter->stats.doosync++;
5453         }
5454
5455         if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) {
5456                 hw->mac.get_link_status = true;
5457                 if (!test_bit(__IGC_DOWN, &adapter->state))
5458                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
5459         }
5460
5461         napi_schedule(&q_vector->napi);
5462
5463         return IRQ_HANDLED;
5464 }
5465
5466 /**
5467  * igc_intr - Legacy Interrupt Handler
5468  * @irq: interrupt number
5469  * @data: pointer to a network interface device structure
5470  */
5471 static irqreturn_t igc_intr(int irq, void *data)
5472 {
5473         struct igc_adapter *adapter = data;
5474         struct igc_q_vector *q_vector = adapter->q_vector[0];
5475         struct igc_hw *hw = &adapter->hw;
5476         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
5477          * need for the IMC write
5478          */
5479         u32 icr = rd32(IGC_ICR);
5480
5481         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5482          * not set, then the adapter didn't send an interrupt
5483          */
5484         if (!(icr & IGC_ICR_INT_ASSERTED))
5485                 return IRQ_NONE;
5486
5487         igc_write_itr(q_vector);
5488
5489         if (icr & IGC_ICR_DRSTA)
5490                 schedule_work(&adapter->reset_task);
5491
5492         if (icr & IGC_ICR_DOUTSYNC) {
5493                 /* HW is reporting DMA is out of sync */
5494                 adapter->stats.doosync++;
5495         }
5496
5497         if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) {
5498                 hw->mac.get_link_status = true;
5499                 /* guard against interrupt when we're going down */
5500                 if (!test_bit(__IGC_DOWN, &adapter->state))
5501                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
5502         }
5503
5504         napi_schedule(&q_vector->napi);
5505
5506         return IRQ_HANDLED;
5507 }
5508
5509 static void igc_free_irq(struct igc_adapter *adapter)
5510 {
5511         if (adapter->msix_entries) {
5512                 int vector = 0, i;
5513
5514                 free_irq(adapter->msix_entries[vector++].vector, adapter);
5515
5516                 for (i = 0; i < adapter->num_q_vectors; i++)
5517                         free_irq(adapter->msix_entries[vector++].vector,
5518                                  adapter->q_vector[i]);
5519         } else {
5520                 free_irq(adapter->pdev->irq, adapter);
5521         }
5522 }
5523
5524 /**
5525  * igc_request_irq - initialize interrupts
5526  * @adapter: Pointer to adapter structure
5527  *
5528  * Attempts to configure interrupts using the best available
5529  * capabilities of the hardware and kernel.
5530  */
5531 static int igc_request_irq(struct igc_adapter *adapter)
5532 {
5533         struct net_device *netdev = adapter->netdev;
5534         struct pci_dev *pdev = adapter->pdev;
5535         int err = 0;
5536
5537         if (adapter->flags & IGC_FLAG_HAS_MSIX) {
5538                 err = igc_request_msix(adapter);
5539                 if (!err)
5540                         goto request_done;
5541                 /* fall back to MSI */
5542                 igc_free_all_tx_resources(adapter);
5543                 igc_free_all_rx_resources(adapter);
5544
5545                 igc_clear_interrupt_scheme(adapter);
5546                 err = igc_init_interrupt_scheme(adapter, false);
5547                 if (err)
5548                         goto request_done;
5549                 igc_setup_all_tx_resources(adapter);
5550                 igc_setup_all_rx_resources(adapter);
5551                 igc_configure(adapter);
5552         }
5553
5554         igc_assign_vector(adapter->q_vector[0], 0);
5555
5556         if (adapter->flags & IGC_FLAG_HAS_MSI) {
5557                 err = request_irq(pdev->irq, &igc_intr_msi, 0,
5558                                   netdev->name, adapter);
5559                 if (!err)
5560                         goto request_done;
5561
5562                 /* fall back to legacy interrupts */
5563                 igc_reset_interrupt_capability(adapter);
5564                 adapter->flags &= ~IGC_FLAG_HAS_MSI;
5565         }
5566
5567         err = request_irq(pdev->irq, &igc_intr, IRQF_SHARED,
5568                           netdev->name, adapter);
5569
5570         if (err)
5571                 netdev_err(netdev, "Error %d getting interrupt\n", err);
5572
5573 request_done:
5574         return err;
5575 }
5576
5577 /**
5578  * __igc_open - Called when a network interface is made active
5579  * @netdev: network interface device structure
5580  * @resuming: boolean indicating if the device is resuming
5581  *
5582  * Returns 0 on success, negative value on failure
5583  *
5584  * The open entry point is called when a network interface is made
5585  * active by the system (IFF_UP).  At this point all resources needed
5586  * for transmit and receive operations are allocated, the interrupt
5587  * handler is registered with the OS, the watchdog timer is started,
5588  * and the stack is notified that the interface is ready.
5589  */
5590 static int __igc_open(struct net_device *netdev, bool resuming)
5591 {
5592         struct igc_adapter *adapter = netdev_priv(netdev);
5593         struct pci_dev *pdev = adapter->pdev;
5594         struct igc_hw *hw = &adapter->hw;
5595         int err = 0;
5596         int i = 0;
5597
5598         /* disallow open during test */
5599
5600         if (test_bit(__IGC_TESTING, &adapter->state)) {
5601                 WARN_ON(resuming);
5602                 return -EBUSY;
5603         }
5604
5605         if (!resuming)
5606                 pm_runtime_get_sync(&pdev->dev);
5607
5608         netif_carrier_off(netdev);
5609
5610         /* allocate transmit descriptors */
5611         err = igc_setup_all_tx_resources(adapter);
5612         if (err)
5613                 goto err_setup_tx;
5614
5615         /* allocate receive descriptors */
5616         err = igc_setup_all_rx_resources(adapter);
5617         if (err)
5618                 goto err_setup_rx;
5619
5620         igc_power_up_link(adapter);
5621
5622         igc_configure(adapter);
5623
5624         err = igc_request_irq(adapter);
5625         if (err)
5626                 goto err_req_irq;
5627
5628         /* Notify the stack of the actual queue counts. */
5629         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
5630         if (err)
5631                 goto err_set_queues;
5632
5633         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
5634         if (err)
5635                 goto err_set_queues;
5636
5637         clear_bit(__IGC_DOWN, &adapter->state);
5638
5639         for (i = 0; i < adapter->num_q_vectors; i++)
5640                 napi_enable(&adapter->q_vector[i]->napi);
5641
5642         /* Clear any pending interrupts. */
5643         rd32(IGC_ICR);
5644         igc_irq_enable(adapter);
5645
5646         if (!resuming)
5647                 pm_runtime_put(&pdev->dev);
5648
5649         netif_tx_start_all_queues(netdev);
5650
5651         /* start the watchdog. */
5652         hw->mac.get_link_status = true;
5653         schedule_work(&adapter->watchdog_task);
5654
5655         return IGC_SUCCESS;
5656
5657 err_set_queues:
5658         igc_free_irq(adapter);
5659 err_req_irq:
5660         igc_release_hw_control(adapter);
5661         igc_power_down_phy_copper_base(&adapter->hw);
5662         igc_free_all_rx_resources(adapter);
5663 err_setup_rx:
5664         igc_free_all_tx_resources(adapter);
5665 err_setup_tx:
5666         igc_reset(adapter);
5667         if (!resuming)
5668                 pm_runtime_put(&pdev->dev);
5669
5670         return err;
5671 }
5672
5673 int igc_open(struct net_device *netdev)
5674 {
5675         return __igc_open(netdev, false);
5676 }
5677
5678 /**
5679  * __igc_close - Disables a network interface
5680  * @netdev: network interface device structure
5681  * @suspending: boolean indicating the device is suspending
5682  *
5683  * Returns 0, this is not allowed to fail
5684  *
5685  * The close entry point is called when an interface is de-activated
5686  * by the OS.  The hardware is still under the driver's control, but
5687  * needs to be disabled.  A global MAC reset is issued to stop the
5688  * hardware, and all transmit and receive resources are freed.
5689  */
5690 static int __igc_close(struct net_device *netdev, bool suspending)
5691 {
5692         struct igc_adapter *adapter = netdev_priv(netdev);
5693         struct pci_dev *pdev = adapter->pdev;
5694
5695         WARN_ON(test_bit(__IGC_RESETTING, &adapter->state));
5696
5697         if (!suspending)
5698                 pm_runtime_get_sync(&pdev->dev);
5699
5700         igc_down(adapter);
5701
5702         igc_release_hw_control(adapter);
5703
5704         igc_free_irq(adapter);
5705
5706         igc_free_all_tx_resources(adapter);
5707         igc_free_all_rx_resources(adapter);
5708
5709         if (!suspending)
5710                 pm_runtime_put_sync(&pdev->dev);
5711
5712         return 0;
5713 }
5714
5715 int igc_close(struct net_device *netdev)
5716 {
5717         if (netif_device_present(netdev) || netdev->dismantle)
5718                 return __igc_close(netdev, false);
5719         return 0;
5720 }
5721
5722 /**
5723  * igc_ioctl - Access the hwtstamp interface
5724  * @netdev: network interface device structure
5725  * @ifr: interface request data
5726  * @cmd: ioctl command
5727  **/
5728 static int igc_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
5729 {
5730         switch (cmd) {
5731         case SIOCGHWTSTAMP:
5732                 return igc_ptp_get_ts_config(netdev, ifr);
5733         case SIOCSHWTSTAMP:
5734                 return igc_ptp_set_ts_config(netdev, ifr);
5735         default:
5736                 return -EOPNOTSUPP;
5737         }
5738 }
5739
5740 static int igc_save_launchtime_params(struct igc_adapter *adapter, int queue,
5741                                       bool enable)
5742 {
5743         struct igc_ring *ring;
5744         int i;
5745
5746         if (queue < 0 || queue >= adapter->num_tx_queues)
5747                 return -EINVAL;
5748
5749         ring = adapter->tx_ring[queue];
5750         ring->launchtime_enable = enable;
5751
5752         if (adapter->base_time)
5753                 return 0;
5754
5755         adapter->cycle_time = NSEC_PER_SEC;
5756
5757         for (i = 0; i < adapter->num_tx_queues; i++) {
5758                 ring = adapter->tx_ring[i];
5759                 ring->start_time = 0;
5760                 ring->end_time = NSEC_PER_SEC;
5761         }
5762
5763         return 0;
5764 }
5765
5766 static bool is_base_time_past(ktime_t base_time, const struct timespec64 *now)
5767 {
5768         struct timespec64 b;
5769
5770         b = ktime_to_timespec64(base_time);
5771
5772         return timespec64_compare(now, &b) > 0;
5773 }
5774
5775 static bool validate_schedule(struct igc_adapter *adapter,
5776                               const struct tc_taprio_qopt_offload *qopt)
5777 {
5778         int queue_uses[IGC_MAX_TX_QUEUES] = { };
5779         struct timespec64 now;
5780         size_t n;
5781
5782         if (qopt->cycle_time_extension)
5783                 return false;
5784
5785         igc_ptp_read(adapter, &now);
5786
5787         /* If we program the controller's BASET registers with a time
5788          * in the future, it will hold all the packets until that
5789          * time, causing a lot of TX Hangs, so to avoid that, we
5790          * reject schedules that would start in the future.
5791          */
5792         if (!is_base_time_past(qopt->base_time, &now))
5793                 return false;
5794
5795         for (n = 0; n < qopt->num_entries; n++) {
5796                 const struct tc_taprio_sched_entry *e;
5797                 int i;
5798
5799                 e = &qopt->entries[n];
5800
5801                 /* i225 only supports "global" frame preemption
5802                  * settings.
5803                  */
5804                 if (e->command != TC_TAPRIO_CMD_SET_GATES)
5805                         return false;
5806
5807                 for (i = 0; i < IGC_MAX_TX_QUEUES; i++) {
5808                         if (e->gate_mask & BIT(i))
5809                                 queue_uses[i]++;
5810
5811                         if (queue_uses[i] > 1)
5812                                 return false;
5813                 }
5814         }
5815
5816         return true;
5817 }
5818
5819 static int igc_tsn_enable_launchtime(struct igc_adapter *adapter,
5820                                      struct tc_etf_qopt_offload *qopt)
5821 {
5822         struct igc_hw *hw = &adapter->hw;
5823         int err;
5824
5825         if (hw->mac.type != igc_i225)
5826                 return -EOPNOTSUPP;
5827
5828         err = igc_save_launchtime_params(adapter, qopt->queue, qopt->enable);
5829         if (err)
5830                 return err;
5831
5832         return igc_tsn_offload_apply(adapter);
5833 }
5834
5835 static int igc_save_qbv_schedule(struct igc_adapter *adapter,
5836                                  struct tc_taprio_qopt_offload *qopt)
5837 {
5838         u32 start_time = 0, end_time = 0;
5839         size_t n;
5840
5841         if (!qopt->enable) {
5842                 adapter->base_time = 0;
5843                 return 0;
5844         }
5845
5846         if (adapter->base_time)
5847                 return -EALREADY;
5848
5849         if (!validate_schedule(adapter, qopt))
5850                 return -EINVAL;
5851
5852         adapter->cycle_time = qopt->cycle_time;
5853         adapter->base_time = qopt->base_time;
5854
5855         /* FIXME: be a little smarter about cases when the gate for a
5856          * queue stays open for more than one entry.
5857          */
5858         for (n = 0; n < qopt->num_entries; n++) {
5859                 struct tc_taprio_sched_entry *e = &qopt->entries[n];
5860                 int i;
5861
5862                 end_time += e->interval;
5863
5864                 for (i = 0; i < IGC_MAX_TX_QUEUES; i++) {
5865                         struct igc_ring *ring = adapter->tx_ring[i];
5866
5867                         if (!(e->gate_mask & BIT(i)))
5868                                 continue;
5869
5870                         ring->start_time = start_time;
5871                         ring->end_time = end_time;
5872                 }
5873
5874                 start_time += e->interval;
5875         }
5876
5877         return 0;
5878 }
5879
5880 static int igc_tsn_enable_qbv_scheduling(struct igc_adapter *adapter,
5881                                          struct tc_taprio_qopt_offload *qopt)
5882 {
5883         struct igc_hw *hw = &adapter->hw;
5884         int err;
5885
5886         if (hw->mac.type != igc_i225)
5887                 return -EOPNOTSUPP;
5888
5889         err = igc_save_qbv_schedule(adapter, qopt);
5890         if (err)
5891                 return err;
5892
5893         return igc_tsn_offload_apply(adapter);
5894 }
5895
5896 static int igc_setup_tc(struct net_device *dev, enum tc_setup_type type,
5897                         void *type_data)
5898 {
5899         struct igc_adapter *adapter = netdev_priv(dev);
5900
5901         switch (type) {
5902         case TC_SETUP_QDISC_TAPRIO:
5903                 return igc_tsn_enable_qbv_scheduling(adapter, type_data);
5904
5905         case TC_SETUP_QDISC_ETF:
5906                 return igc_tsn_enable_launchtime(adapter, type_data);
5907
5908         default:
5909                 return -EOPNOTSUPP;
5910         }
5911 }
5912
5913 static int igc_bpf(struct net_device *dev, struct netdev_bpf *bpf)
5914 {
5915         struct igc_adapter *adapter = netdev_priv(dev);
5916
5917         switch (bpf->command) {
5918         case XDP_SETUP_PROG:
5919                 return igc_xdp_set_prog(adapter, bpf->prog, bpf->extack);
5920         case XDP_SETUP_XSK_POOL:
5921                 return igc_xdp_setup_pool(adapter, bpf->xsk.pool,
5922                                           bpf->xsk.queue_id);
5923         default:
5924                 return -EOPNOTSUPP;
5925         }
5926 }
5927
5928 static int igc_xdp_xmit(struct net_device *dev, int num_frames,
5929                         struct xdp_frame **frames, u32 flags)
5930 {
5931         struct igc_adapter *adapter = netdev_priv(dev);
5932         int cpu = smp_processor_id();
5933         struct netdev_queue *nq;
5934         struct igc_ring *ring;
5935         int i, drops;
5936
5937         if (unlikely(test_bit(__IGC_DOWN, &adapter->state)))
5938                 return -ENETDOWN;
5939
5940         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
5941                 return -EINVAL;
5942
5943         ring = igc_xdp_get_tx_ring(adapter, cpu);
5944         nq = txring_txq(ring);
5945
5946         __netif_tx_lock(nq, cpu);
5947
5948         drops = 0;
5949         for (i = 0; i < num_frames; i++) {
5950                 int err;
5951                 struct xdp_frame *xdpf = frames[i];
5952
5953                 err = igc_xdp_init_tx_descriptor(ring, xdpf);
5954                 if (err) {
5955                         xdp_return_frame_rx_napi(xdpf);
5956                         drops++;
5957                 }
5958         }
5959
5960         if (flags & XDP_XMIT_FLUSH)
5961                 igc_flush_tx_descriptors(ring);
5962
5963         __netif_tx_unlock(nq);
5964
5965         return num_frames - drops;
5966 }
5967
5968 static void igc_trigger_rxtxq_interrupt(struct igc_adapter *adapter,
5969                                         struct igc_q_vector *q_vector)
5970 {
5971         struct igc_hw *hw = &adapter->hw;
5972         u32 eics = 0;
5973
5974         eics |= q_vector->eims_value;
5975         wr32(IGC_EICS, eics);
5976 }
5977
5978 int igc_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags)
5979 {
5980         struct igc_adapter *adapter = netdev_priv(dev);
5981         struct igc_q_vector *q_vector;
5982         struct igc_ring *ring;
5983
5984         if (test_bit(__IGC_DOWN, &adapter->state))
5985                 return -ENETDOWN;
5986
5987         if (!igc_xdp_is_enabled(adapter))
5988                 return -ENXIO;
5989
5990         if (queue_id >= adapter->num_rx_queues)
5991                 return -EINVAL;
5992
5993         ring = adapter->rx_ring[queue_id];
5994
5995         if (!ring->xsk_pool)
5996                 return -ENXIO;
5997
5998         q_vector = adapter->q_vector[queue_id];
5999         if (!napi_if_scheduled_mark_missed(&q_vector->napi))
6000                 igc_trigger_rxtxq_interrupt(adapter, q_vector);
6001
6002         return 0;
6003 }
6004
6005 static const struct net_device_ops igc_netdev_ops = {
6006         .ndo_open               = igc_open,
6007         .ndo_stop               = igc_close,
6008         .ndo_start_xmit         = igc_xmit_frame,
6009         .ndo_set_rx_mode        = igc_set_rx_mode,
6010         .ndo_set_mac_address    = igc_set_mac,
6011         .ndo_change_mtu         = igc_change_mtu,
6012         .ndo_get_stats64        = igc_get_stats64,
6013         .ndo_fix_features       = igc_fix_features,
6014         .ndo_set_features       = igc_set_features,
6015         .ndo_features_check     = igc_features_check,
6016         .ndo_do_ioctl           = igc_ioctl,
6017         .ndo_setup_tc           = igc_setup_tc,
6018         .ndo_bpf                = igc_bpf,
6019         .ndo_xdp_xmit           = igc_xdp_xmit,
6020         .ndo_xsk_wakeup         = igc_xsk_wakeup,
6021 };
6022
6023 /* PCIe configuration access */
6024 void igc_read_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value)
6025 {
6026         struct igc_adapter *adapter = hw->back;
6027
6028         pci_read_config_word(adapter->pdev, reg, value);
6029 }
6030
6031 void igc_write_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value)
6032 {
6033         struct igc_adapter *adapter = hw->back;
6034
6035         pci_write_config_word(adapter->pdev, reg, *value);
6036 }
6037
6038 s32 igc_read_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value)
6039 {
6040         struct igc_adapter *adapter = hw->back;
6041
6042         if (!pci_is_pcie(adapter->pdev))
6043                 return -IGC_ERR_CONFIG;
6044
6045         pcie_capability_read_word(adapter->pdev, reg, value);
6046
6047         return IGC_SUCCESS;
6048 }
6049
6050 s32 igc_write_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value)
6051 {
6052         struct igc_adapter *adapter = hw->back;
6053
6054         if (!pci_is_pcie(adapter->pdev))
6055                 return -IGC_ERR_CONFIG;
6056
6057         pcie_capability_write_word(adapter->pdev, reg, *value);
6058
6059         return IGC_SUCCESS;
6060 }
6061
6062 u32 igc_rd32(struct igc_hw *hw, u32 reg)
6063 {
6064         struct igc_adapter *igc = container_of(hw, struct igc_adapter, hw);
6065         u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
6066         u32 value = 0;
6067
6068         value = readl(&hw_addr[reg]);
6069
6070         /* reads should not return all F's */
6071         if (!(~value) && (!reg || !(~readl(hw_addr)))) {
6072                 struct net_device *netdev = igc->netdev;
6073
6074                 hw->hw_addr = NULL;
6075                 netif_device_detach(netdev);
6076                 netdev_err(netdev, "PCIe link lost, device now detached\n");
6077                 WARN(pci_device_is_present(igc->pdev),
6078                      "igc: Failed to read reg 0x%x!\n", reg);
6079         }
6080
6081         return value;
6082 }
6083
6084 int igc_set_spd_dplx(struct igc_adapter *adapter, u32 spd, u8 dplx)
6085 {
6086         struct igc_mac_info *mac = &adapter->hw.mac;
6087
6088         mac->autoneg = false;
6089
6090         /* Make sure dplx is at most 1 bit and lsb of speed is not set
6091          * for the switch() below to work
6092          */
6093         if ((spd & 1) || (dplx & ~1))
6094                 goto err_inval;
6095
6096         switch (spd + dplx) {
6097         case SPEED_10 + DUPLEX_HALF:
6098                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
6099                 break;
6100         case SPEED_10 + DUPLEX_FULL:
6101                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
6102                 break;
6103         case SPEED_100 + DUPLEX_HALF:
6104                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
6105                 break;
6106         case SPEED_100 + DUPLEX_FULL:
6107                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
6108                 break;
6109         case SPEED_1000 + DUPLEX_FULL:
6110                 mac->autoneg = true;
6111                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
6112                 break;
6113         case SPEED_1000 + DUPLEX_HALF: /* not supported */
6114                 goto err_inval;
6115         case SPEED_2500 + DUPLEX_FULL:
6116                 mac->autoneg = true;
6117                 adapter->hw.phy.autoneg_advertised = ADVERTISE_2500_FULL;
6118                 break;
6119         case SPEED_2500 + DUPLEX_HALF: /* not supported */
6120         default:
6121                 goto err_inval;
6122         }
6123
6124         /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
6125         adapter->hw.phy.mdix = AUTO_ALL_MODES;
6126
6127         return 0;
6128
6129 err_inval:
6130         netdev_err(adapter->netdev, "Unsupported Speed/Duplex configuration\n");
6131         return -EINVAL;
6132 }
6133
6134 /**
6135  * igc_probe - Device Initialization Routine
6136  * @pdev: PCI device information struct
6137  * @ent: entry in igc_pci_tbl
6138  *
6139  * Returns 0 on success, negative on failure
6140  *
6141  * igc_probe initializes an adapter identified by a pci_dev structure.
6142  * The OS initialization, configuring the adapter private structure,
6143  * and a hardware reset occur.
6144  */
6145 static int igc_probe(struct pci_dev *pdev,
6146                      const struct pci_device_id *ent)
6147 {
6148         struct igc_adapter *adapter;
6149         struct net_device *netdev;
6150         struct igc_hw *hw;
6151         const struct igc_info *ei = igc_info_tbl[ent->driver_data];
6152         int err, pci_using_dac;
6153
6154         err = pci_enable_device_mem(pdev);
6155         if (err)
6156                 return err;
6157
6158         pci_using_dac = 0;
6159         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
6160         if (!err) {
6161                 pci_using_dac = 1;
6162         } else {
6163                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
6164                 if (err) {
6165                         dev_err(&pdev->dev,
6166                                 "No usable DMA configuration, aborting\n");
6167                         goto err_dma;
6168                 }
6169         }
6170
6171         err = pci_request_mem_regions(pdev, igc_driver_name);
6172         if (err)
6173                 goto err_pci_reg;
6174
6175         pci_enable_pcie_error_reporting(pdev);
6176
6177         pci_set_master(pdev);
6178
6179         err = -ENOMEM;
6180         netdev = alloc_etherdev_mq(sizeof(struct igc_adapter),
6181                                    IGC_MAX_TX_QUEUES);
6182
6183         if (!netdev)
6184                 goto err_alloc_etherdev;
6185
6186         SET_NETDEV_DEV(netdev, &pdev->dev);
6187
6188         pci_set_drvdata(pdev, netdev);
6189         adapter = netdev_priv(netdev);
6190         adapter->netdev = netdev;
6191         adapter->pdev = pdev;
6192         hw = &adapter->hw;
6193         hw->back = adapter;
6194         adapter->port_num = hw->bus.func;
6195         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6196
6197         err = pci_save_state(pdev);
6198         if (err)
6199                 goto err_ioremap;
6200
6201         err = -EIO;
6202         adapter->io_addr = ioremap(pci_resource_start(pdev, 0),
6203                                    pci_resource_len(pdev, 0));
6204         if (!adapter->io_addr)
6205                 goto err_ioremap;
6206
6207         /* hw->hw_addr can be zeroed, so use adapter->io_addr for unmap */
6208         hw->hw_addr = adapter->io_addr;
6209
6210         netdev->netdev_ops = &igc_netdev_ops;
6211         igc_ethtool_set_ops(netdev);
6212         netdev->watchdog_timeo = 5 * HZ;
6213
6214         netdev->mem_start = pci_resource_start(pdev, 0);
6215         netdev->mem_end = pci_resource_end(pdev, 0);
6216
6217         /* PCI config space info */
6218         hw->vendor_id = pdev->vendor;
6219         hw->device_id = pdev->device;
6220         hw->revision_id = pdev->revision;
6221         hw->subsystem_vendor_id = pdev->subsystem_vendor;
6222         hw->subsystem_device_id = pdev->subsystem_device;
6223
6224         /* Copy the default MAC and PHY function pointers */
6225         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6226         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6227
6228         /* Initialize skew-specific constants */
6229         err = ei->get_invariants(hw);
6230         if (err)
6231                 goto err_sw_init;
6232
6233         /* Add supported features to the features list*/
6234         netdev->features |= NETIF_F_SG;
6235         netdev->features |= NETIF_F_TSO;
6236         netdev->features |= NETIF_F_TSO6;
6237         netdev->features |= NETIF_F_TSO_ECN;
6238         netdev->features |= NETIF_F_RXCSUM;
6239         netdev->features |= NETIF_F_HW_CSUM;
6240         netdev->features |= NETIF_F_SCTP_CRC;
6241         netdev->features |= NETIF_F_HW_TC;
6242
6243 #define IGC_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
6244                                   NETIF_F_GSO_GRE_CSUM | \
6245                                   NETIF_F_GSO_IPXIP4 | \
6246                                   NETIF_F_GSO_IPXIP6 | \
6247                                   NETIF_F_GSO_UDP_TUNNEL | \
6248                                   NETIF_F_GSO_UDP_TUNNEL_CSUM)
6249
6250         netdev->gso_partial_features = IGC_GSO_PARTIAL_FEATURES;
6251         netdev->features |= NETIF_F_GSO_PARTIAL | IGC_GSO_PARTIAL_FEATURES;
6252
6253         /* setup the private structure */
6254         err = igc_sw_init(adapter);
6255         if (err)
6256                 goto err_sw_init;
6257
6258         /* copy netdev features into list of user selectable features */
6259         netdev->hw_features |= NETIF_F_NTUPLE;
6260         netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
6261         netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
6262         netdev->hw_features |= netdev->features;
6263
6264         if (pci_using_dac)
6265                 netdev->features |= NETIF_F_HIGHDMA;
6266
6267         netdev->vlan_features |= netdev->features;
6268
6269         /* MTU range: 68 - 9216 */
6270         netdev->min_mtu = ETH_MIN_MTU;
6271         netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
6272
6273         /* before reading the NVM, reset the controller to put the device in a
6274          * known good starting state
6275          */
6276         hw->mac.ops.reset_hw(hw);
6277
6278         if (igc_get_flash_presence_i225(hw)) {
6279                 if (hw->nvm.ops.validate(hw) < 0) {
6280                         dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
6281                         err = -EIO;
6282                         goto err_eeprom;
6283                 }
6284         }
6285
6286         if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
6287                 /* copy the MAC address out of the NVM */
6288                 if (hw->mac.ops.read_mac_addr(hw))
6289                         dev_err(&pdev->dev, "NVM Read Error\n");
6290         }
6291
6292         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
6293
6294         if (!is_valid_ether_addr(netdev->dev_addr)) {
6295                 dev_err(&pdev->dev, "Invalid MAC Address\n");
6296                 err = -EIO;
6297                 goto err_eeprom;
6298         }
6299
6300         /* configure RXPBSIZE and TXPBSIZE */
6301         wr32(IGC_RXPBS, I225_RXPBSIZE_DEFAULT);
6302         wr32(IGC_TXPBS, I225_TXPBSIZE_DEFAULT);
6303
6304         timer_setup(&adapter->watchdog_timer, igc_watchdog, 0);
6305         timer_setup(&adapter->phy_info_timer, igc_update_phy_info, 0);
6306
6307         INIT_WORK(&adapter->reset_task, igc_reset_task);
6308         INIT_WORK(&adapter->watchdog_task, igc_watchdog_task);
6309
6310         /* Initialize link properties that are user-changeable */
6311         adapter->fc_autoneg = true;
6312         hw->mac.autoneg = true;
6313         hw->phy.autoneg_advertised = 0xaf;
6314
6315         hw->fc.requested_mode = igc_fc_default;
6316         hw->fc.current_mode = igc_fc_default;
6317
6318         /* By default, support wake on port A */
6319         adapter->flags |= IGC_FLAG_WOL_SUPPORTED;
6320
6321         /* initialize the wol settings based on the eeprom settings */
6322         if (adapter->flags & IGC_FLAG_WOL_SUPPORTED)
6323                 adapter->wol |= IGC_WUFC_MAG;
6324
6325         device_set_wakeup_enable(&adapter->pdev->dev,
6326                                  adapter->flags & IGC_FLAG_WOL_SUPPORTED);
6327
6328         igc_ptp_init(adapter);
6329
6330         /* reset the hardware with the new settings */
6331         igc_reset(adapter);
6332
6333         /* let the f/w know that the h/w is now under the control of the
6334          * driver.
6335          */
6336         igc_get_hw_control(adapter);
6337
6338         strncpy(netdev->name, "eth%d", IFNAMSIZ);
6339         err = register_netdev(netdev);
6340         if (err)
6341                 goto err_register;
6342
6343          /* carrier off reporting is important to ethtool even BEFORE open */
6344         netif_carrier_off(netdev);
6345
6346         /* Check if Media Autosense is enabled */
6347         adapter->ei = *ei;
6348
6349         /* print pcie link status and MAC address */
6350         pcie_print_link_status(pdev);
6351         netdev_info(netdev, "MAC: %pM\n", netdev->dev_addr);
6352
6353         dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
6354         /* Disable EEE for internal PHY devices */
6355         hw->dev_spec._base.eee_enable = false;
6356         adapter->flags &= ~IGC_FLAG_EEE;
6357         igc_set_eee_i225(hw, false, false, false);
6358
6359         pm_runtime_put_noidle(&pdev->dev);
6360
6361         return 0;
6362
6363 err_register:
6364         igc_release_hw_control(adapter);
6365 err_eeprom:
6366         if (!igc_check_reset_block(hw))
6367                 igc_reset_phy(hw);
6368 err_sw_init:
6369         igc_clear_interrupt_scheme(adapter);
6370         iounmap(adapter->io_addr);
6371 err_ioremap:
6372         free_netdev(netdev);
6373 err_alloc_etherdev:
6374         pci_disable_pcie_error_reporting(pdev);
6375         pci_release_mem_regions(pdev);
6376 err_pci_reg:
6377 err_dma:
6378         pci_disable_device(pdev);
6379         return err;
6380 }
6381
6382 /**
6383  * igc_remove - Device Removal Routine
6384  * @pdev: PCI device information struct
6385  *
6386  * igc_remove is called by the PCI subsystem to alert the driver
6387  * that it should release a PCI device.  This could be caused by a
6388  * Hot-Plug event, or because the driver is going to be removed from
6389  * memory.
6390  */
6391 static void igc_remove(struct pci_dev *pdev)
6392 {
6393         struct net_device *netdev = pci_get_drvdata(pdev);
6394         struct igc_adapter *adapter = netdev_priv(netdev);
6395
6396         pm_runtime_get_noresume(&pdev->dev);
6397
6398         igc_flush_nfc_rules(adapter);
6399
6400         igc_ptp_stop(adapter);
6401
6402         set_bit(__IGC_DOWN, &adapter->state);
6403
6404         del_timer_sync(&adapter->watchdog_timer);
6405         del_timer_sync(&adapter->phy_info_timer);
6406
6407         cancel_work_sync(&adapter->reset_task);
6408         cancel_work_sync(&adapter->watchdog_task);
6409
6410         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
6411          * would have already happened in close and is redundant.
6412          */
6413         igc_release_hw_control(adapter);
6414         unregister_netdev(netdev);
6415
6416         igc_clear_interrupt_scheme(adapter);
6417         pci_iounmap(pdev, adapter->io_addr);
6418         pci_release_mem_regions(pdev);
6419
6420         free_netdev(netdev);
6421
6422         pci_disable_pcie_error_reporting(pdev);
6423
6424         pci_disable_device(pdev);
6425 }
6426
6427 static int __igc_shutdown(struct pci_dev *pdev, bool *enable_wake,
6428                           bool runtime)
6429 {
6430         struct net_device *netdev = pci_get_drvdata(pdev);
6431         struct igc_adapter *adapter = netdev_priv(netdev);
6432         u32 wufc = runtime ? IGC_WUFC_LNKC : adapter->wol;
6433         struct igc_hw *hw = &adapter->hw;
6434         u32 ctrl, rctl, status;
6435         bool wake;
6436
6437         rtnl_lock();
6438         netif_device_detach(netdev);
6439
6440         if (netif_running(netdev))
6441                 __igc_close(netdev, true);
6442
6443         igc_ptp_suspend(adapter);
6444
6445         igc_clear_interrupt_scheme(adapter);
6446         rtnl_unlock();
6447
6448         status = rd32(IGC_STATUS);
6449         if (status & IGC_STATUS_LU)
6450                 wufc &= ~IGC_WUFC_LNKC;
6451
6452         if (wufc) {
6453                 igc_setup_rctl(adapter);
6454                 igc_set_rx_mode(netdev);
6455
6456                 /* turn on all-multi mode if wake on multicast is enabled */
6457                 if (wufc & IGC_WUFC_MC) {
6458                         rctl = rd32(IGC_RCTL);
6459                         rctl |= IGC_RCTL_MPE;
6460                         wr32(IGC_RCTL, rctl);
6461                 }
6462
6463                 ctrl = rd32(IGC_CTRL);
6464                 ctrl |= IGC_CTRL_ADVD3WUC;
6465                 wr32(IGC_CTRL, ctrl);
6466
6467                 /* Allow time for pending master requests to run */
6468                 igc_disable_pcie_master(hw);
6469
6470                 wr32(IGC_WUC, IGC_WUC_PME_EN);
6471                 wr32(IGC_WUFC, wufc);
6472         } else {
6473                 wr32(IGC_WUC, 0);
6474                 wr32(IGC_WUFC, 0);
6475         }
6476
6477         wake = wufc || adapter->en_mng_pt;
6478         if (!wake)
6479                 igc_power_down_phy_copper_base(&adapter->hw);
6480         else
6481                 igc_power_up_link(adapter);
6482
6483         if (enable_wake)
6484                 *enable_wake = wake;
6485
6486         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
6487          * would have already happened in close and is redundant.
6488          */
6489         igc_release_hw_control(adapter);
6490
6491         pci_disable_device(pdev);
6492
6493         return 0;
6494 }
6495
6496 #ifdef CONFIG_PM
6497 static int __maybe_unused igc_runtime_suspend(struct device *dev)
6498 {
6499         return __igc_shutdown(to_pci_dev(dev), NULL, 1);
6500 }
6501
6502 static void igc_deliver_wake_packet(struct net_device *netdev)
6503 {
6504         struct igc_adapter *adapter = netdev_priv(netdev);
6505         struct igc_hw *hw = &adapter->hw;
6506         struct sk_buff *skb;
6507         u32 wupl;
6508
6509         wupl = rd32(IGC_WUPL) & IGC_WUPL_MASK;
6510
6511         /* WUPM stores only the first 128 bytes of the wake packet.
6512          * Read the packet only if we have the whole thing.
6513          */
6514         if (wupl == 0 || wupl > IGC_WUPM_BYTES)
6515                 return;
6516
6517         skb = netdev_alloc_skb_ip_align(netdev, IGC_WUPM_BYTES);
6518         if (!skb)
6519                 return;
6520
6521         skb_put(skb, wupl);
6522
6523         /* Ensure reads are 32-bit aligned */
6524         wupl = roundup(wupl, 4);
6525
6526         memcpy_fromio(skb->data, hw->hw_addr + IGC_WUPM_REG(0), wupl);
6527
6528         skb->protocol = eth_type_trans(skb, netdev);
6529         netif_rx(skb);
6530 }
6531
6532 static int __maybe_unused igc_resume(struct device *dev)
6533 {
6534         struct pci_dev *pdev = to_pci_dev(dev);
6535         struct net_device *netdev = pci_get_drvdata(pdev);
6536         struct igc_adapter *adapter = netdev_priv(netdev);
6537         struct igc_hw *hw = &adapter->hw;
6538         u32 err, val;
6539
6540         pci_set_power_state(pdev, PCI_D0);
6541         pci_restore_state(pdev);
6542         pci_save_state(pdev);
6543
6544         if (!pci_device_is_present(pdev))
6545                 return -ENODEV;
6546         err = pci_enable_device_mem(pdev);
6547         if (err) {
6548                 netdev_err(netdev, "Cannot enable PCI device from suspend\n");
6549                 return err;
6550         }
6551         pci_set_master(pdev);
6552
6553         pci_enable_wake(pdev, PCI_D3hot, 0);
6554         pci_enable_wake(pdev, PCI_D3cold, 0);
6555
6556         if (igc_init_interrupt_scheme(adapter, true)) {
6557                 netdev_err(netdev, "Unable to allocate memory for queues\n");
6558                 return -ENOMEM;
6559         }
6560
6561         igc_reset(adapter);
6562
6563         /* let the f/w know that the h/w is now under the control of the
6564          * driver.
6565          */
6566         igc_get_hw_control(adapter);
6567
6568         val = rd32(IGC_WUS);
6569         if (val & WAKE_PKT_WUS)
6570                 igc_deliver_wake_packet(netdev);
6571
6572         wr32(IGC_WUS, ~0);
6573
6574         rtnl_lock();
6575         if (!err && netif_running(netdev))
6576                 err = __igc_open(netdev, true);
6577
6578         if (!err)
6579                 netif_device_attach(netdev);
6580         rtnl_unlock();
6581
6582         return err;
6583 }
6584
6585 static int __maybe_unused igc_runtime_resume(struct device *dev)
6586 {
6587         return igc_resume(dev);
6588 }
6589
6590 static int __maybe_unused igc_suspend(struct device *dev)
6591 {
6592         return __igc_shutdown(to_pci_dev(dev), NULL, 0);
6593 }
6594
6595 static int __maybe_unused igc_runtime_idle(struct device *dev)
6596 {
6597         struct net_device *netdev = dev_get_drvdata(dev);
6598         struct igc_adapter *adapter = netdev_priv(netdev);
6599
6600         if (!igc_has_link(adapter))
6601                 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
6602
6603         return -EBUSY;
6604 }
6605 #endif /* CONFIG_PM */
6606
6607 static void igc_shutdown(struct pci_dev *pdev)
6608 {
6609         bool wake;
6610
6611         __igc_shutdown(pdev, &wake, 0);
6612
6613         if (system_state == SYSTEM_POWER_OFF) {
6614                 pci_wake_from_d3(pdev, wake);
6615                 pci_set_power_state(pdev, PCI_D3hot);
6616         }
6617 }
6618
6619 /**
6620  *  igc_io_error_detected - called when PCI error is detected
6621  *  @pdev: Pointer to PCI device
6622  *  @state: The current PCI connection state
6623  *
6624  *  This function is called after a PCI bus error affecting
6625  *  this device has been detected.
6626  **/
6627 static pci_ers_result_t igc_io_error_detected(struct pci_dev *pdev,
6628                                               pci_channel_state_t state)
6629 {
6630         struct net_device *netdev = pci_get_drvdata(pdev);
6631         struct igc_adapter *adapter = netdev_priv(netdev);
6632
6633         netif_device_detach(netdev);
6634
6635         if (state == pci_channel_io_perm_failure)
6636                 return PCI_ERS_RESULT_DISCONNECT;
6637
6638         if (netif_running(netdev))
6639                 igc_down(adapter);
6640         pci_disable_device(pdev);
6641
6642         /* Request a slot reset. */
6643         return PCI_ERS_RESULT_NEED_RESET;
6644 }
6645
6646 /**
6647  *  igc_io_slot_reset - called after the PCI bus has been reset.
6648  *  @pdev: Pointer to PCI device
6649  *
6650  *  Restart the card from scratch, as if from a cold-boot. Implementation
6651  *  resembles the first-half of the igc_resume routine.
6652  **/
6653 static pci_ers_result_t igc_io_slot_reset(struct pci_dev *pdev)
6654 {
6655         struct net_device *netdev = pci_get_drvdata(pdev);
6656         struct igc_adapter *adapter = netdev_priv(netdev);
6657         struct igc_hw *hw = &adapter->hw;
6658         pci_ers_result_t result;
6659
6660         if (pci_enable_device_mem(pdev)) {
6661                 netdev_err(netdev, "Could not re-enable PCI device after reset\n");
6662                 result = PCI_ERS_RESULT_DISCONNECT;
6663         } else {
6664                 pci_set_master(pdev);
6665                 pci_restore_state(pdev);
6666                 pci_save_state(pdev);
6667
6668                 pci_enable_wake(pdev, PCI_D3hot, 0);
6669                 pci_enable_wake(pdev, PCI_D3cold, 0);
6670
6671                 /* In case of PCI error, adapter loses its HW address
6672                  * so we should re-assign it here.
6673                  */
6674                 hw->hw_addr = adapter->io_addr;
6675
6676                 igc_reset(adapter);
6677                 wr32(IGC_WUS, ~0);
6678                 result = PCI_ERS_RESULT_RECOVERED;
6679         }
6680
6681         return result;
6682 }
6683
6684 /**
6685  *  igc_io_resume - called when traffic can start to flow again.
6686  *  @pdev: Pointer to PCI device
6687  *
6688  *  This callback is called when the error recovery driver tells us that
6689  *  its OK to resume normal operation. Implementation resembles the
6690  *  second-half of the igc_resume routine.
6691  */
6692 static void igc_io_resume(struct pci_dev *pdev)
6693 {
6694         struct net_device *netdev = pci_get_drvdata(pdev);
6695         struct igc_adapter *adapter = netdev_priv(netdev);
6696
6697         rtnl_lock();
6698         if (netif_running(netdev)) {
6699                 if (igc_open(netdev)) {
6700                         netdev_err(netdev, "igc_open failed after reset\n");
6701                         return;
6702                 }
6703         }
6704
6705         netif_device_attach(netdev);
6706
6707         /* let the f/w know that the h/w is now under the control of the
6708          * driver.
6709          */
6710         igc_get_hw_control(adapter);
6711         rtnl_unlock();
6712 }
6713
6714 static const struct pci_error_handlers igc_err_handler = {
6715         .error_detected = igc_io_error_detected,
6716         .slot_reset = igc_io_slot_reset,
6717         .resume = igc_io_resume,
6718 };
6719
6720 #ifdef CONFIG_PM
6721 static const struct dev_pm_ops igc_pm_ops = {
6722         SET_SYSTEM_SLEEP_PM_OPS(igc_suspend, igc_resume)
6723         SET_RUNTIME_PM_OPS(igc_runtime_suspend, igc_runtime_resume,
6724                            igc_runtime_idle)
6725 };
6726 #endif
6727
6728 static struct pci_driver igc_driver = {
6729         .name     = igc_driver_name,
6730         .id_table = igc_pci_tbl,
6731         .probe    = igc_probe,
6732         .remove   = igc_remove,
6733 #ifdef CONFIG_PM
6734         .driver.pm = &igc_pm_ops,
6735 #endif
6736         .shutdown = igc_shutdown,
6737         .err_handler = &igc_err_handler,
6738 };
6739
6740 /**
6741  * igc_reinit_queues - return error
6742  * @adapter: pointer to adapter structure
6743  */
6744 int igc_reinit_queues(struct igc_adapter *adapter)
6745 {
6746         struct net_device *netdev = adapter->netdev;
6747         int err = 0;
6748
6749         if (netif_running(netdev))
6750                 igc_close(netdev);
6751
6752         igc_reset_interrupt_capability(adapter);
6753
6754         if (igc_init_interrupt_scheme(adapter, true)) {
6755                 netdev_err(netdev, "Unable to allocate memory for queues\n");
6756                 return -ENOMEM;
6757         }
6758
6759         if (netif_running(netdev))
6760                 err = igc_open(netdev);
6761
6762         return err;
6763 }
6764
6765 /**
6766  * igc_get_hw_dev - return device
6767  * @hw: pointer to hardware structure
6768  *
6769  * used by hardware layer to print debugging information
6770  */
6771 struct net_device *igc_get_hw_dev(struct igc_hw *hw)
6772 {
6773         struct igc_adapter *adapter = hw->back;
6774
6775         return adapter->netdev;
6776 }
6777
6778 static void igc_disable_rx_ring_hw(struct igc_ring *ring)
6779 {
6780         struct igc_hw *hw = &ring->q_vector->adapter->hw;
6781         u8 idx = ring->reg_idx;
6782         u32 rxdctl;
6783
6784         rxdctl = rd32(IGC_RXDCTL(idx));
6785         rxdctl &= ~IGC_RXDCTL_QUEUE_ENABLE;
6786         rxdctl |= IGC_RXDCTL_SWFLUSH;
6787         wr32(IGC_RXDCTL(idx), rxdctl);
6788 }
6789
6790 void igc_disable_rx_ring(struct igc_ring *ring)
6791 {
6792         igc_disable_rx_ring_hw(ring);
6793         igc_clean_rx_ring(ring);
6794 }
6795
6796 void igc_enable_rx_ring(struct igc_ring *ring)
6797 {
6798         struct igc_adapter *adapter = ring->q_vector->adapter;
6799
6800         igc_configure_rx_ring(adapter, ring);
6801
6802         if (ring->xsk_pool)
6803                 igc_alloc_rx_buffers_zc(ring, igc_desc_unused(ring));
6804         else
6805                 igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
6806 }
6807
6808 static void igc_disable_tx_ring_hw(struct igc_ring *ring)
6809 {
6810         struct igc_hw *hw = &ring->q_vector->adapter->hw;
6811         u8 idx = ring->reg_idx;
6812         u32 txdctl;
6813
6814         txdctl = rd32(IGC_TXDCTL(idx));
6815         txdctl &= ~IGC_TXDCTL_QUEUE_ENABLE;
6816         txdctl |= IGC_TXDCTL_SWFLUSH;
6817         wr32(IGC_TXDCTL(idx), txdctl);
6818 }
6819
6820 void igc_disable_tx_ring(struct igc_ring *ring)
6821 {
6822         igc_disable_tx_ring_hw(ring);
6823         igc_clean_tx_ring(ring);
6824 }
6825
6826 void igc_enable_tx_ring(struct igc_ring *ring)
6827 {
6828         struct igc_adapter *adapter = ring->q_vector->adapter;
6829
6830         igc_configure_tx_ring(adapter, ring);
6831 }
6832
6833 /**
6834  * igc_init_module - Driver Registration Routine
6835  *
6836  * igc_init_module is the first routine called when the driver is
6837  * loaded. All it does is register with the PCI subsystem.
6838  */
6839 static int __init igc_init_module(void)
6840 {
6841         int ret;
6842
6843         pr_info("%s\n", igc_driver_string);
6844         pr_info("%s\n", igc_copyright);
6845
6846         ret = pci_register_driver(&igc_driver);
6847         return ret;
6848 }
6849
6850 module_init(igc_init_module);
6851
6852 /**
6853  * igc_exit_module - Driver Exit Cleanup Routine
6854  *
6855  * igc_exit_module is called just before the driver is removed
6856  * from memory.
6857  */
6858 static void __exit igc_exit_module(void)
6859 {
6860         pci_unregister_driver(&igc_driver);
6861 }
6862
6863 module_exit(igc_exit_module);
6864 /* igc_main.c */