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/tcp.h>
   8#include <linux/udp.h>
   9#include <linux/ip.h>
  10#include <linux/pm_runtime.h>
  11#include <net/pkt_sched.h>
  12#include <linux/bpf_trace.h>
  13#include <net/xdp_sock_drv.h>
  14#include <linux/pci.h>
  15
  16#include <net/ipv6.h>
  17
  18#include "igc.h"
  19#include "igc_hw.h"
  20#include "igc_tsn.h"
  21#include "igc_xdp.h"
  22
  23#define DRV_SUMMARY	"Intel(R) 2.5G Ethernet Linux Driver"
  24
  25#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK)
  26
  27#define IGC_XDP_PASS		0
  28#define IGC_XDP_CONSUMED	BIT(0)
  29#define IGC_XDP_TX		BIT(1)
  30#define IGC_XDP_REDIRECT	BIT(2)
  31
  32static int debug = -1;
  33
  34MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  35MODULE_DESCRIPTION(DRV_SUMMARY);
  36MODULE_LICENSE("GPL v2");
  37module_param(debug, int, 0);
  38MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  39
  40char igc_driver_name[] = "igc";
  41static const char igc_driver_string[] = DRV_SUMMARY;
  42static const char igc_copyright[] =
  43	"Copyright(c) 2018 Intel Corporation.";
  44
  45static const struct igc_info *igc_info_tbl[] = {
  46	[board_base] = &igc_base_info,
  47};
  48
  49static const struct pci_device_id igc_pci_tbl[] = {
  50	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_LM), board_base },
  51	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_V), board_base },
  52	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_I), board_base },
  53	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I220_V), board_base },
  54	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_K), board_base },
  55	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_K2), board_base },
  56	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_K), board_base },
  57	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_LMVP), board_base },
  58	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_LMVP), board_base },
  59	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_IT), board_base },
  60	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_LM), board_base },
  61	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_V), board_base },
  62	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_IT), board_base },
  63	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I221_V), board_base },
  64	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I226_BLANK_NVM), board_base },
  65	{ PCI_VDEVICE(INTEL, IGC_DEV_ID_I225_BLANK_NVM), board_base },
  66	/* required last entry */
  67	{0, }
  68};
  69
  70MODULE_DEVICE_TABLE(pci, igc_pci_tbl);
  71
  72enum latency_range {
  73	lowest_latency = 0,
  74	low_latency = 1,
  75	bulk_latency = 2,
  76	latency_invalid = 255
  77};
  78
  79void igc_reset(struct igc_adapter *adapter)
  80{
  81	struct net_device *dev = adapter->netdev;
  82	struct igc_hw *hw = &adapter->hw;
  83	struct igc_fc_info *fc = &hw->fc;
  84	u32 pba, hwm;
  85
  86	/* Repartition PBA for greater than 9k MTU if required */
  87	pba = IGC_PBA_34K;
  88
  89	/* flow control settings
  90	 * The high water mark must be low enough to fit one full frame
  91	 * after transmitting the pause frame.  As such we must have enough
  92	 * space to allow for us to complete our current transmit and then
  93	 * receive the frame that is in progress from the link partner.
  94	 * Set it to:
  95	 * - the full Rx FIFO size minus one full Tx plus one full Rx frame
  96	 */
  97	hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
  98
  99	fc->high_water = hwm & 0xFFFFFFF0;	/* 16-byte granularity */
 100	fc->low_water = fc->high_water - 16;
 101	fc->pause_time = 0xFFFF;
 102	fc->send_xon = 1;
 103	fc->current_mode = fc->requested_mode;
 104
 105	hw->mac.ops.reset_hw(hw);
 106
 107	if (hw->mac.ops.init_hw(hw))
 108		netdev_err(dev, "Error on hardware initialization\n");
 109
 110	/* Re-establish EEE setting */
 111	igc_set_eee_i225(hw, true, true, true);
 112
 113	if (!netif_running(adapter->netdev))
 114		igc_power_down_phy_copper_base(&adapter->hw);
 115
 116	/* Enable HW to recognize an 802.1Q VLAN Ethernet packet */
 117	wr32(IGC_VET, ETH_P_8021Q);
 118
 119	/* Re-enable PTP, where applicable. */
 120	igc_ptp_reset(adapter);
 121
 122	/* Re-enable TSN offloading, where applicable. */
 123	igc_tsn_reset(adapter);
 124
 125	igc_get_phy_info(hw);
 126}
 127
 128/**
 129 * igc_power_up_link - Power up the phy link
 130 * @adapter: address of board private structure
 131 */
 132static void igc_power_up_link(struct igc_adapter *adapter)
 133{
 134	igc_reset_phy(&adapter->hw);
 135
 136	igc_power_up_phy_copper(&adapter->hw);
 137
 138	igc_setup_link(&adapter->hw);
 139}
 140
 141/**
 142 * igc_release_hw_control - release control of the h/w to f/w
 143 * @adapter: address of board private structure
 144 *
 145 * igc_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
 146 * For ASF and Pass Through versions of f/w this means that the
 147 * driver is no longer loaded.
 148 */
 149static void igc_release_hw_control(struct igc_adapter *adapter)
 150{
 151	struct igc_hw *hw = &adapter->hw;
 152	u32 ctrl_ext;
 153
 154	if (!pci_device_is_present(adapter->pdev))
 155		return;
 156
 157	/* Let firmware take over control of h/w */
 158	ctrl_ext = rd32(IGC_CTRL_EXT);
 159	wr32(IGC_CTRL_EXT,
 160	     ctrl_ext & ~IGC_CTRL_EXT_DRV_LOAD);
 161}
 162
 163/**
 164 * igc_get_hw_control - get control of the h/w from f/w
 165 * @adapter: address of board private structure
 166 *
 167 * igc_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
 168 * For ASF and Pass Through versions of f/w this means that
 169 * the driver is loaded.
 170 */
 171static void igc_get_hw_control(struct igc_adapter *adapter)
 172{
 173	struct igc_hw *hw = &adapter->hw;
 174	u32 ctrl_ext;
 175
 176	/* Let firmware know the driver has taken over */
 177	ctrl_ext = rd32(IGC_CTRL_EXT);
 178	wr32(IGC_CTRL_EXT,
 179	     ctrl_ext | IGC_CTRL_EXT_DRV_LOAD);
 180}
 181
 182static void igc_unmap_tx_buffer(struct device *dev, struct igc_tx_buffer *buf)
 183{
 184	dma_unmap_single(dev, dma_unmap_addr(buf, dma),
 185			 dma_unmap_len(buf, len), DMA_TO_DEVICE);
 186
 187	dma_unmap_len_set(buf, len, 0);
 188}
 189
 190/**
 191 * igc_clean_tx_ring - Free Tx Buffers
 192 * @tx_ring: ring to be cleaned
 193 */
 194static void igc_clean_tx_ring(struct igc_ring *tx_ring)
 195{
 196	u16 i = tx_ring->next_to_clean;
 197	struct igc_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
 198	u32 xsk_frames = 0;
 199
 200	while (i != tx_ring->next_to_use) {
 201		union igc_adv_tx_desc *eop_desc, *tx_desc;
 202
 203		switch (tx_buffer->type) {
 204		case IGC_TX_BUFFER_TYPE_XSK:
 205			xsk_frames++;
 206			break;
 207		case IGC_TX_BUFFER_TYPE_XDP:
 208			xdp_return_frame(tx_buffer->xdpf);
 209			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 210			break;
 211		case IGC_TX_BUFFER_TYPE_SKB:
 212			dev_kfree_skb_any(tx_buffer->skb);
 213			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 214			break;
 215		default:
 216			netdev_warn_once(tx_ring->netdev, "Unknown Tx buffer type\n");
 217			break;
 218		}
 219
 220		/* check for eop_desc to determine the end of the packet */
 221		eop_desc = tx_buffer->next_to_watch;
 222		tx_desc = IGC_TX_DESC(tx_ring, i);
 223
 224		/* unmap remaining buffers */
 225		while (tx_desc != eop_desc) {
 226			tx_buffer++;
 227			tx_desc++;
 228			i++;
 229			if (unlikely(i == tx_ring->count)) {
 230				i = 0;
 231				tx_buffer = tx_ring->tx_buffer_info;
 232				tx_desc = IGC_TX_DESC(tx_ring, 0);
 233			}
 234
 235			/* unmap any remaining paged data */
 236			if (dma_unmap_len(tx_buffer, len))
 237				igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 
 
 
 238		}
 239
 240		tx_buffer->next_to_watch = NULL;
 241
 242		/* move us one more past the eop_desc for start of next pkt */
 243		tx_buffer++;
 244		i++;
 245		if (unlikely(i == tx_ring->count)) {
 246			i = 0;
 247			tx_buffer = tx_ring->tx_buffer_info;
 248		}
 249	}
 250
 251	if (tx_ring->xsk_pool && xsk_frames)
 252		xsk_tx_completed(tx_ring->xsk_pool, xsk_frames);
 253
 254	/* reset BQL for queue */
 255	netdev_tx_reset_queue(txring_txq(tx_ring));
 256
 257	/* Zero out the buffer ring */
 258	memset(tx_ring->tx_buffer_info, 0,
 259	       sizeof(*tx_ring->tx_buffer_info) * tx_ring->count);
 260
 261	/* Zero out the descriptor ring */
 262	memset(tx_ring->desc, 0, tx_ring->size);
 263
 264	/* reset next_to_use and next_to_clean */
 265	tx_ring->next_to_use = 0;
 266	tx_ring->next_to_clean = 0;
 267}
 268
 269/**
 270 * igc_free_tx_resources - Free Tx Resources per Queue
 271 * @tx_ring: Tx descriptor ring for a specific queue
 272 *
 273 * Free all transmit software resources
 274 */
 275void igc_free_tx_resources(struct igc_ring *tx_ring)
 276{
 277	igc_disable_tx_ring(tx_ring);
 278
 279	vfree(tx_ring->tx_buffer_info);
 280	tx_ring->tx_buffer_info = NULL;
 281
 282	/* if not set, then don't free */
 283	if (!tx_ring->desc)
 284		return;
 285
 286	dma_free_coherent(tx_ring->dev, tx_ring->size,
 287			  tx_ring->desc, tx_ring->dma);
 288
 289	tx_ring->desc = NULL;
 290}
 291
 292/**
 293 * igc_free_all_tx_resources - Free Tx Resources for All Queues
 294 * @adapter: board private structure
 295 *
 296 * Free all transmit software resources
 297 */
 298static void igc_free_all_tx_resources(struct igc_adapter *adapter)
 299{
 300	int i;
 301
 302	for (i = 0; i < adapter->num_tx_queues; i++)
 303		igc_free_tx_resources(adapter->tx_ring[i]);
 304}
 305
 306/**
 307 * igc_clean_all_tx_rings - Free Tx Buffers for all queues
 308 * @adapter: board private structure
 309 */
 310static void igc_clean_all_tx_rings(struct igc_adapter *adapter)
 311{
 312	int i;
 313
 314	for (i = 0; i < adapter->num_tx_queues; i++)
 315		if (adapter->tx_ring[i])
 316			igc_clean_tx_ring(adapter->tx_ring[i]);
 317}
 318
 319static void igc_disable_tx_ring_hw(struct igc_ring *ring)
 320{
 321	struct igc_hw *hw = &ring->q_vector->adapter->hw;
 322	u8 idx = ring->reg_idx;
 323	u32 txdctl;
 324
 325	txdctl = rd32(IGC_TXDCTL(idx));
 326	txdctl &= ~IGC_TXDCTL_QUEUE_ENABLE;
 327	txdctl |= IGC_TXDCTL_SWFLUSH;
 328	wr32(IGC_TXDCTL(idx), txdctl);
 329}
 330
 331/**
 332 * igc_disable_all_tx_rings_hw - Disable all transmit queue operation
 333 * @adapter: board private structure
 334 */
 335static void igc_disable_all_tx_rings_hw(struct igc_adapter *adapter)
 336{
 337	int i;
 338
 339	for (i = 0; i < adapter->num_tx_queues; i++) {
 340		struct igc_ring *tx_ring = adapter->tx_ring[i];
 341
 342		igc_disable_tx_ring_hw(tx_ring);
 343	}
 344}
 345
 346/**
 347 * igc_setup_tx_resources - allocate Tx resources (Descriptors)
 348 * @tx_ring: tx descriptor ring (for a specific queue) to setup
 349 *
 350 * Return 0 on success, negative on failure
 351 */
 352int igc_setup_tx_resources(struct igc_ring *tx_ring)
 353{
 354	struct net_device *ndev = tx_ring->netdev;
 355	struct device *dev = tx_ring->dev;
 356	int size = 0;
 357
 358	size = sizeof(struct igc_tx_buffer) * tx_ring->count;
 359	tx_ring->tx_buffer_info = vzalloc(size);
 360	if (!tx_ring->tx_buffer_info)
 361		goto err;
 362
 363	/* round up to nearest 4K */
 364	tx_ring->size = tx_ring->count * sizeof(union igc_adv_tx_desc);
 365	tx_ring->size = ALIGN(tx_ring->size, 4096);
 366
 367	tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
 368					   &tx_ring->dma, GFP_KERNEL);
 369
 370	if (!tx_ring->desc)
 371		goto err;
 372
 373	tx_ring->next_to_use = 0;
 374	tx_ring->next_to_clean = 0;
 375
 376	return 0;
 377
 378err:
 379	vfree(tx_ring->tx_buffer_info);
 380	netdev_err(ndev, "Unable to allocate memory for Tx descriptor ring\n");
 381	return -ENOMEM;
 382}
 383
 384/**
 385 * igc_setup_all_tx_resources - wrapper to allocate Tx resources for all queues
 386 * @adapter: board private structure
 387 *
 388 * Return 0 on success, negative on failure
 389 */
 390static int igc_setup_all_tx_resources(struct igc_adapter *adapter)
 391{
 392	struct net_device *dev = adapter->netdev;
 393	int i, err = 0;
 394
 395	for (i = 0; i < adapter->num_tx_queues; i++) {
 396		err = igc_setup_tx_resources(adapter->tx_ring[i]);
 397		if (err) {
 398			netdev_err(dev, "Error on Tx queue %u setup\n", i);
 399			for (i--; i >= 0; i--)
 400				igc_free_tx_resources(adapter->tx_ring[i]);
 401			break;
 402		}
 403	}
 404
 405	return err;
 406}
 407
 408static void igc_clean_rx_ring_page_shared(struct igc_ring *rx_ring)
 
 
 
 
 409{
 410	u16 i = rx_ring->next_to_clean;
 411
 412	dev_kfree_skb(rx_ring->skb);
 413	rx_ring->skb = NULL;
 414
 415	/* Free all the Rx ring sk_buffs */
 416	while (i != rx_ring->next_to_alloc) {
 417		struct igc_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
 418
 419		/* Invalidate cache lines that may have been written to by
 420		 * device so that we avoid corrupting memory.
 421		 */
 422		dma_sync_single_range_for_cpu(rx_ring->dev,
 423					      buffer_info->dma,
 424					      buffer_info->page_offset,
 425					      igc_rx_bufsz(rx_ring),
 426					      DMA_FROM_DEVICE);
 427
 428		/* free resources associated with mapping */
 429		dma_unmap_page_attrs(rx_ring->dev,
 430				     buffer_info->dma,
 431				     igc_rx_pg_size(rx_ring),
 432				     DMA_FROM_DEVICE,
 433				     IGC_RX_DMA_ATTR);
 434		__page_frag_cache_drain(buffer_info->page,
 435					buffer_info->pagecnt_bias);
 436
 437		i++;
 438		if (i == rx_ring->count)
 439			i = 0;
 440	}
 441}
 442
 443static void igc_clean_rx_ring_xsk_pool(struct igc_ring *ring)
 444{
 445	struct igc_rx_buffer *bi;
 446	u16 i;
 447
 448	for (i = 0; i < ring->count; i++) {
 449		bi = &ring->rx_buffer_info[i];
 450		if (!bi->xdp)
 451			continue;
 452
 453		xsk_buff_free(bi->xdp);
 454		bi->xdp = NULL;
 455	}
 456}
 457
 458/**
 459 * igc_clean_rx_ring - Free Rx Buffers per Queue
 460 * @ring: ring to free buffers from
 461 */
 462static void igc_clean_rx_ring(struct igc_ring *ring)
 463{
 464	if (ring->xsk_pool)
 465		igc_clean_rx_ring_xsk_pool(ring);
 466	else
 467		igc_clean_rx_ring_page_shared(ring);
 468
 469	clear_ring_uses_large_buffer(ring);
 470
 471	ring->next_to_alloc = 0;
 472	ring->next_to_clean = 0;
 473	ring->next_to_use = 0;
 474}
 475
 476/**
 477 * igc_clean_all_rx_rings - Free Rx Buffers for all queues
 478 * @adapter: board private structure
 479 */
 480static void igc_clean_all_rx_rings(struct igc_adapter *adapter)
 481{
 482	int i;
 483
 484	for (i = 0; i < adapter->num_rx_queues; i++)
 485		if (adapter->rx_ring[i])
 486			igc_clean_rx_ring(adapter->rx_ring[i]);
 487}
 488
 489/**
 490 * igc_free_rx_resources - Free Rx Resources
 491 * @rx_ring: ring to clean the resources from
 492 *
 493 * Free all receive software resources
 494 */
 495void igc_free_rx_resources(struct igc_ring *rx_ring)
 496{
 497	igc_clean_rx_ring(rx_ring);
 498
 499	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 500
 501	vfree(rx_ring->rx_buffer_info);
 502	rx_ring->rx_buffer_info = NULL;
 503
 504	/* if not set, then don't free */
 505	if (!rx_ring->desc)
 506		return;
 507
 508	dma_free_coherent(rx_ring->dev, rx_ring->size,
 509			  rx_ring->desc, rx_ring->dma);
 510
 511	rx_ring->desc = NULL;
 512}
 513
 514/**
 515 * igc_free_all_rx_resources - Free Rx Resources for All Queues
 516 * @adapter: board private structure
 517 *
 518 * Free all receive software resources
 519 */
 520static void igc_free_all_rx_resources(struct igc_adapter *adapter)
 521{
 522	int i;
 523
 524	for (i = 0; i < adapter->num_rx_queues; i++)
 525		igc_free_rx_resources(adapter->rx_ring[i]);
 526}
 527
 528/**
 529 * igc_setup_rx_resources - allocate Rx resources (Descriptors)
 530 * @rx_ring:    rx descriptor ring (for a specific queue) to setup
 531 *
 532 * Returns 0 on success, negative on failure
 533 */
 534int igc_setup_rx_resources(struct igc_ring *rx_ring)
 535{
 536	struct net_device *ndev = rx_ring->netdev;
 537	struct device *dev = rx_ring->dev;
 538	u8 index = rx_ring->queue_index;
 539	int size, desc_len, res;
 540
 541	/* XDP RX-queue info */
 542	if (xdp_rxq_info_is_reg(&rx_ring->xdp_rxq))
 543		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 544	res = xdp_rxq_info_reg(&rx_ring->xdp_rxq, ndev, index,
 545			       rx_ring->q_vector->napi.napi_id);
 546	if (res < 0) {
 547		netdev_err(ndev, "Failed to register xdp_rxq index %u\n",
 548			   index);
 549		return res;
 550	}
 551
 552	size = sizeof(struct igc_rx_buffer) * rx_ring->count;
 553	rx_ring->rx_buffer_info = vzalloc(size);
 554	if (!rx_ring->rx_buffer_info)
 555		goto err;
 556
 557	desc_len = sizeof(union igc_adv_rx_desc);
 558
 559	/* Round up to nearest 4K */
 560	rx_ring->size = rx_ring->count * desc_len;
 561	rx_ring->size = ALIGN(rx_ring->size, 4096);
 562
 563	rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
 564					   &rx_ring->dma, GFP_KERNEL);
 565
 566	if (!rx_ring->desc)
 567		goto err;
 568
 569	rx_ring->next_to_alloc = 0;
 570	rx_ring->next_to_clean = 0;
 571	rx_ring->next_to_use = 0;
 572
 573	return 0;
 574
 575err:
 576	xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
 577	vfree(rx_ring->rx_buffer_info);
 578	rx_ring->rx_buffer_info = NULL;
 579	netdev_err(ndev, "Unable to allocate memory for Rx descriptor ring\n");
 580	return -ENOMEM;
 581}
 582
 583/**
 584 * igc_setup_all_rx_resources - wrapper to allocate Rx resources
 585 *                                (Descriptors) for all queues
 586 * @adapter: board private structure
 587 *
 588 * Return 0 on success, negative on failure
 589 */
 590static int igc_setup_all_rx_resources(struct igc_adapter *adapter)
 591{
 592	struct net_device *dev = adapter->netdev;
 593	int i, err = 0;
 594
 595	for (i = 0; i < adapter->num_rx_queues; i++) {
 596		err = igc_setup_rx_resources(adapter->rx_ring[i]);
 597		if (err) {
 598			netdev_err(dev, "Error on Rx queue %u setup\n", i);
 599			for (i--; i >= 0; i--)
 600				igc_free_rx_resources(adapter->rx_ring[i]);
 601			break;
 602		}
 603	}
 604
 605	return err;
 606}
 607
 608static struct xsk_buff_pool *igc_get_xsk_pool(struct igc_adapter *adapter,
 609					      struct igc_ring *ring)
 610{
 611	if (!igc_xdp_is_enabled(adapter) ||
 612	    !test_bit(IGC_RING_FLAG_AF_XDP_ZC, &ring->flags))
 613		return NULL;
 614
 615	return xsk_get_pool_from_qid(ring->netdev, ring->queue_index);
 616}
 617
 618/**
 619 * igc_configure_rx_ring - Configure a receive ring after Reset
 620 * @adapter: board private structure
 621 * @ring: receive ring to be configured
 622 *
 623 * Configure the Rx unit of the MAC after a reset.
 624 */
 625static void igc_configure_rx_ring(struct igc_adapter *adapter,
 626				  struct igc_ring *ring)
 627{
 628	struct igc_hw *hw = &adapter->hw;
 629	union igc_adv_rx_desc *rx_desc;
 630	int reg_idx = ring->reg_idx;
 631	u32 srrctl = 0, rxdctl = 0;
 632	u64 rdba = ring->dma;
 633	u32 buf_size;
 634
 635	xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
 636	ring->xsk_pool = igc_get_xsk_pool(adapter, ring);
 637	if (ring->xsk_pool) {
 638		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
 639						   MEM_TYPE_XSK_BUFF_POOL,
 640						   NULL));
 641		xsk_pool_set_rxq_info(ring->xsk_pool, &ring->xdp_rxq);
 642	} else {
 643		WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
 644						   MEM_TYPE_PAGE_SHARED,
 645						   NULL));
 646	}
 647
 648	if (igc_xdp_is_enabled(adapter))
 649		set_ring_uses_large_buffer(ring);
 650
 651	/* disable the queue */
 652	wr32(IGC_RXDCTL(reg_idx), 0);
 653
 654	/* Set DMA base address registers */
 655	wr32(IGC_RDBAL(reg_idx),
 656	     rdba & 0x00000000ffffffffULL);
 657	wr32(IGC_RDBAH(reg_idx), rdba >> 32);
 658	wr32(IGC_RDLEN(reg_idx),
 659	     ring->count * sizeof(union igc_adv_rx_desc));
 660
 661	/* initialize head and tail */
 662	ring->tail = adapter->io_addr + IGC_RDT(reg_idx);
 663	wr32(IGC_RDH(reg_idx), 0);
 664	writel(0, ring->tail);
 665
 666	/* reset next-to- use/clean to place SW in sync with hardware */
 667	ring->next_to_clean = 0;
 668	ring->next_to_use = 0;
 669
 670	if (ring->xsk_pool)
 671		buf_size = xsk_pool_get_rx_frame_size(ring->xsk_pool);
 672	else if (ring_uses_large_buffer(ring))
 673		buf_size = IGC_RXBUFFER_3072;
 674	else
 675		buf_size = IGC_RXBUFFER_2048;
 676
 677	srrctl = rd32(IGC_SRRCTL(reg_idx));
 678	srrctl &= ~(IGC_SRRCTL_BSIZEPKT_MASK | IGC_SRRCTL_BSIZEHDR_MASK |
 679		    IGC_SRRCTL_DESCTYPE_MASK);
 680	srrctl |= IGC_SRRCTL_BSIZEHDR(IGC_RX_HDR_LEN);
 681	srrctl |= IGC_SRRCTL_BSIZEPKT(buf_size);
 682	srrctl |= IGC_SRRCTL_DESCTYPE_ADV_ONEBUF;
 683
 684	wr32(IGC_SRRCTL(reg_idx), srrctl);
 685
 686	rxdctl |= IGC_RX_PTHRESH;
 687	rxdctl |= IGC_RX_HTHRESH << 8;
 688	rxdctl |= IGC_RX_WTHRESH << 16;
 689
 690	/* initialize rx_buffer_info */
 691	memset(ring->rx_buffer_info, 0,
 692	       sizeof(struct igc_rx_buffer) * ring->count);
 693
 694	/* initialize Rx descriptor 0 */
 695	rx_desc = IGC_RX_DESC(ring, 0);
 696	rx_desc->wb.upper.length = 0;
 697
 698	/* enable receive descriptor fetching */
 699	rxdctl |= IGC_RXDCTL_QUEUE_ENABLE;
 700
 701	wr32(IGC_RXDCTL(reg_idx), rxdctl);
 702}
 703
 704/**
 705 * igc_configure_rx - Configure receive Unit after Reset
 706 * @adapter: board private structure
 707 *
 708 * Configure the Rx unit of the MAC after a reset.
 709 */
 710static void igc_configure_rx(struct igc_adapter *adapter)
 711{
 712	int i;
 713
 714	/* Setup the HW Rx Head and Tail Descriptor Pointers and
 715	 * the Base and Length of the Rx Descriptor Ring
 716	 */
 717	for (i = 0; i < adapter->num_rx_queues; i++)
 718		igc_configure_rx_ring(adapter, adapter->rx_ring[i]);
 719}
 720
 721/**
 722 * igc_configure_tx_ring - Configure transmit ring after Reset
 723 * @adapter: board private structure
 724 * @ring: tx ring to configure
 725 *
 726 * Configure a transmit ring after a reset.
 727 */
 728static void igc_configure_tx_ring(struct igc_adapter *adapter,
 729				  struct igc_ring *ring)
 730{
 731	struct igc_hw *hw = &adapter->hw;
 732	int reg_idx = ring->reg_idx;
 733	u64 tdba = ring->dma;
 734	u32 txdctl = 0;
 735
 736	ring->xsk_pool = igc_get_xsk_pool(adapter, ring);
 737
 738	/* disable the queue */
 739	wr32(IGC_TXDCTL(reg_idx), 0);
 740	wrfl();
 
 741
 742	wr32(IGC_TDLEN(reg_idx),
 743	     ring->count * sizeof(union igc_adv_tx_desc));
 744	wr32(IGC_TDBAL(reg_idx),
 745	     tdba & 0x00000000ffffffffULL);
 746	wr32(IGC_TDBAH(reg_idx), tdba >> 32);
 747
 748	ring->tail = adapter->io_addr + IGC_TDT(reg_idx);
 749	wr32(IGC_TDH(reg_idx), 0);
 750	writel(0, ring->tail);
 751
 752	txdctl |= IGC_TX_PTHRESH;
 753	txdctl |= IGC_TX_HTHRESH << 8;
 754	txdctl |= IGC_TX_WTHRESH << 16;
 755
 756	txdctl |= IGC_TXDCTL_QUEUE_ENABLE;
 757	wr32(IGC_TXDCTL(reg_idx), txdctl);
 758}
 759
 760/**
 761 * igc_configure_tx - Configure transmit Unit after Reset
 762 * @adapter: board private structure
 763 *
 764 * Configure the Tx unit of the MAC after a reset.
 765 */
 766static void igc_configure_tx(struct igc_adapter *adapter)
 767{
 768	int i;
 769
 770	for (i = 0; i < adapter->num_tx_queues; i++)
 771		igc_configure_tx_ring(adapter, adapter->tx_ring[i]);
 772}
 773
 774/**
 775 * igc_setup_mrqc - configure the multiple receive queue control registers
 776 * @adapter: Board private structure
 777 */
 778static void igc_setup_mrqc(struct igc_adapter *adapter)
 779{
 780	struct igc_hw *hw = &adapter->hw;
 781	u32 j, num_rx_queues;
 782	u32 mrqc, rxcsum;
 783	u32 rss_key[10];
 784
 785	netdev_rss_key_fill(rss_key, sizeof(rss_key));
 786	for (j = 0; j < 10; j++)
 787		wr32(IGC_RSSRK(j), rss_key[j]);
 788
 789	num_rx_queues = adapter->rss_queues;
 790
 791	if (adapter->rss_indir_tbl_init != num_rx_queues) {
 792		for (j = 0; j < IGC_RETA_SIZE; j++)
 793			adapter->rss_indir_tbl[j] =
 794			(j * num_rx_queues) / IGC_RETA_SIZE;
 795		adapter->rss_indir_tbl_init = num_rx_queues;
 796	}
 797	igc_write_rss_indir_tbl(adapter);
 798
 799	/* Disable raw packet checksumming so that RSS hash is placed in
 800	 * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
 801	 * offloads as they are enabled by default
 802	 */
 803	rxcsum = rd32(IGC_RXCSUM);
 804	rxcsum |= IGC_RXCSUM_PCSD;
 805
 806	/* Enable Receive Checksum Offload for SCTP */
 807	rxcsum |= IGC_RXCSUM_CRCOFL;
 808
 809	/* Don't need to set TUOFL or IPOFL, they default to 1 */
 810	wr32(IGC_RXCSUM, rxcsum);
 811
 812	/* Generate RSS hash based on packet types, TCP/UDP
 813	 * port numbers and/or IPv4/v6 src and dst addresses
 814	 */
 815	mrqc = IGC_MRQC_RSS_FIELD_IPV4 |
 816	       IGC_MRQC_RSS_FIELD_IPV4_TCP |
 817	       IGC_MRQC_RSS_FIELD_IPV6 |
 818	       IGC_MRQC_RSS_FIELD_IPV6_TCP |
 819	       IGC_MRQC_RSS_FIELD_IPV6_TCP_EX;
 820
 821	if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV4_UDP)
 822		mrqc |= IGC_MRQC_RSS_FIELD_IPV4_UDP;
 823	if (adapter->flags & IGC_FLAG_RSS_FIELD_IPV6_UDP)
 824		mrqc |= IGC_MRQC_RSS_FIELD_IPV6_UDP;
 825
 826	mrqc |= IGC_MRQC_ENABLE_RSS_MQ;
 827
 828	wr32(IGC_MRQC, mrqc);
 829}
 830
 831/**
 832 * igc_setup_rctl - configure the receive control registers
 833 * @adapter: Board private structure
 834 */
 835static void igc_setup_rctl(struct igc_adapter *adapter)
 836{
 837	struct igc_hw *hw = &adapter->hw;
 838	u32 rctl;
 839
 840	rctl = rd32(IGC_RCTL);
 841
 842	rctl &= ~(3 << IGC_RCTL_MO_SHIFT);
 843	rctl &= ~(IGC_RCTL_LBM_TCVR | IGC_RCTL_LBM_MAC);
 844
 845	rctl |= IGC_RCTL_EN | IGC_RCTL_BAM | IGC_RCTL_RDMTS_HALF |
 846		(hw->mac.mc_filter_type << IGC_RCTL_MO_SHIFT);
 847
 848	/* enable stripping of CRC. Newer features require
 849	 * that the HW strips the CRC.
 850	 */
 851	rctl |= IGC_RCTL_SECRC;
 852
 853	/* disable store bad packets and clear size bits. */
 854	rctl &= ~(IGC_RCTL_SBP | IGC_RCTL_SZ_256);
 855
 856	/* enable LPE to allow for reception of jumbo frames */
 857	rctl |= IGC_RCTL_LPE;
 858
 859	/* disable queue 0 to prevent tail write w/o re-config */
 860	wr32(IGC_RXDCTL(0), 0);
 861
 862	/* This is useful for sniffing bad packets. */
 863	if (adapter->netdev->features & NETIF_F_RXALL) {
 864		/* UPE and MPE will be handled by normal PROMISC logic
 865		 * in set_rx_mode
 866		 */
 867		rctl |= (IGC_RCTL_SBP | /* Receive bad packets */
 868			 IGC_RCTL_BAM | /* RX All Bcast Pkts */
 869			 IGC_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
 870
 871		rctl &= ~(IGC_RCTL_DPF | /* Allow filtered pause */
 872			  IGC_RCTL_CFIEN); /* Disable VLAN CFIEN Filter */
 873	}
 874
 875	wr32(IGC_RCTL, rctl);
 876}
 877
 878/**
 879 * igc_setup_tctl - configure the transmit control registers
 880 * @adapter: Board private structure
 881 */
 882static void igc_setup_tctl(struct igc_adapter *adapter)
 883{
 884	struct igc_hw *hw = &adapter->hw;
 885	u32 tctl;
 886
 887	/* disable queue 0 which icould be enabled by default */
 888	wr32(IGC_TXDCTL(0), 0);
 889
 890	/* Program the Transmit Control Register */
 891	tctl = rd32(IGC_TCTL);
 892	tctl &= ~IGC_TCTL_CT;
 893	tctl |= IGC_TCTL_PSP | IGC_TCTL_RTLC |
 894		(IGC_COLLISION_THRESHOLD << IGC_CT_SHIFT);
 895
 896	/* Enable transmits */
 897	tctl |= IGC_TCTL_EN;
 898
 899	wr32(IGC_TCTL, tctl);
 900}
 901
 902/**
 903 * igc_set_mac_filter_hw() - Set MAC address filter in hardware
 904 * @adapter: Pointer to adapter where the filter should be set
 905 * @index: Filter index
 906 * @type: MAC address filter type (source or destination)
 907 * @addr: MAC address
 908 * @queue: If non-negative, queue assignment feature is enabled and frames
 909 *         matching the filter are enqueued onto 'queue'. Otherwise, queue
 910 *         assignment is disabled.
 911 */
 912static void igc_set_mac_filter_hw(struct igc_adapter *adapter, int index,
 913				  enum igc_mac_filter_type type,
 914				  const u8 *addr, int queue)
 915{
 916	struct net_device *dev = adapter->netdev;
 917	struct igc_hw *hw = &adapter->hw;
 918	u32 ral, rah;
 919
 920	if (WARN_ON(index >= hw->mac.rar_entry_count))
 921		return;
 922
 923	ral = le32_to_cpup((__le32 *)(addr));
 924	rah = le16_to_cpup((__le16 *)(addr + 4));
 925
 926	if (type == IGC_MAC_FILTER_TYPE_SRC) {
 927		rah &= ~IGC_RAH_ASEL_MASK;
 928		rah |= IGC_RAH_ASEL_SRC_ADDR;
 929	}
 930
 931	if (queue >= 0) {
 932		rah &= ~IGC_RAH_QSEL_MASK;
 933		rah |= (queue << IGC_RAH_QSEL_SHIFT);
 934		rah |= IGC_RAH_QSEL_ENABLE;
 935	}
 936
 937	rah |= IGC_RAH_AV;
 938
 939	wr32(IGC_RAL(index), ral);
 940	wr32(IGC_RAH(index), rah);
 941
 942	netdev_dbg(dev, "MAC address filter set in HW: index %d", index);
 943}
 944
 945/**
 946 * igc_clear_mac_filter_hw() - Clear MAC address filter in hardware
 947 * @adapter: Pointer to adapter where the filter should be cleared
 948 * @index: Filter index
 949 */
 950static void igc_clear_mac_filter_hw(struct igc_adapter *adapter, int index)
 951{
 952	struct net_device *dev = adapter->netdev;
 953	struct igc_hw *hw = &adapter->hw;
 954
 955	if (WARN_ON(index >= hw->mac.rar_entry_count))
 956		return;
 957
 958	wr32(IGC_RAL(index), 0);
 959	wr32(IGC_RAH(index), 0);
 960
 961	netdev_dbg(dev, "MAC address filter cleared in HW: index %d", index);
 962}
 963
 964/* Set default MAC address for the PF in the first RAR entry */
 965static void igc_set_default_mac_filter(struct igc_adapter *adapter)
 966{
 967	struct net_device *dev = adapter->netdev;
 968	u8 *addr = adapter->hw.mac.addr;
 969
 970	netdev_dbg(dev, "Set default MAC address filter: address %pM", addr);
 971
 972	igc_set_mac_filter_hw(adapter, 0, IGC_MAC_FILTER_TYPE_DST, addr, -1);
 973}
 974
 975/**
 976 * igc_set_mac - Change the Ethernet Address of the NIC
 977 * @netdev: network interface device structure
 978 * @p: pointer to an address structure
 979 *
 980 * Returns 0 on success, negative on failure
 981 */
 982static int igc_set_mac(struct net_device *netdev, void *p)
 983{
 984	struct igc_adapter *adapter = netdev_priv(netdev);
 985	struct igc_hw *hw = &adapter->hw;
 986	struct sockaddr *addr = p;
 987
 988	if (!is_valid_ether_addr(addr->sa_data))
 989		return -EADDRNOTAVAIL;
 990
 991	eth_hw_addr_set(netdev, addr->sa_data);
 992	memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
 993
 994	/* set the correct pool for the new PF MAC address in entry 0 */
 995	igc_set_default_mac_filter(adapter);
 996
 997	return 0;
 998}
 999
1000/**
1001 *  igc_write_mc_addr_list - write multicast addresses to MTA
1002 *  @netdev: network interface device structure
1003 *
1004 *  Writes multicast address list to the MTA hash table.
1005 *  Returns: -ENOMEM on failure
1006 *           0 on no addresses written
1007 *           X on writing X addresses to MTA
1008 **/
1009static int igc_write_mc_addr_list(struct net_device *netdev)
1010{
1011	struct igc_adapter *adapter = netdev_priv(netdev);
1012	struct igc_hw *hw = &adapter->hw;
1013	struct netdev_hw_addr *ha;
1014	u8  *mta_list;
1015	int i;
1016
1017	if (netdev_mc_empty(netdev)) {
1018		/* nothing to program, so clear mc list */
1019		igc_update_mc_addr_list(hw, NULL, 0);
1020		return 0;
1021	}
1022
1023	mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
1024	if (!mta_list)
1025		return -ENOMEM;
1026
1027	/* The shared function expects a packed array of only addresses. */
1028	i = 0;
1029	netdev_for_each_mc_addr(ha, netdev)
1030		memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
1031
1032	igc_update_mc_addr_list(hw, mta_list, i);
1033	kfree(mta_list);
1034
1035	return netdev_mc_count(netdev);
1036}
1037
1038static __le32 igc_tx_launchtime(struct igc_ring *ring, ktime_t txtime,
1039				bool *first_flag, bool *insert_empty)
1040{
1041	struct igc_adapter *adapter = netdev_priv(ring->netdev);
1042	ktime_t cycle_time = adapter->cycle_time;
1043	ktime_t base_time = adapter->base_time;
1044	ktime_t now = ktime_get_clocktai();
1045	ktime_t baset_est, end_of_cycle;
1046	s32 launchtime;
1047	s64 n;
1048
1049	n = div64_s64(ktime_sub_ns(now, base_time), cycle_time);
1050
1051	baset_est = ktime_add_ns(base_time, cycle_time * (n));
1052	end_of_cycle = ktime_add_ns(baset_est, cycle_time);
1053
1054	if (ktime_compare(txtime, end_of_cycle) >= 0) {
1055		if (baset_est != ring->last_ff_cycle) {
1056			*first_flag = true;
1057			ring->last_ff_cycle = baset_est;
1058
1059			if (ktime_compare(end_of_cycle, ring->last_tx_cycle) > 0)
1060				*insert_empty = true;
1061		}
1062	}
1063
1064	/* Introducing a window at end of cycle on which packets
1065	 * potentially not honor launchtime. Window of 5us chosen
1066	 * considering software update the tail pointer and packets
1067	 * are dma'ed to packet buffer.
1068	 */
1069	if ((ktime_sub_ns(end_of_cycle, now) < 5 * NSEC_PER_USEC))
1070		netdev_warn(ring->netdev, "Packet with txtime=%llu may not be honoured\n",
1071			    txtime);
1072
1073	ring->last_tx_cycle = end_of_cycle;
1074
1075	launchtime = ktime_sub_ns(txtime, baset_est);
1076	if (launchtime > 0)
1077		div_s64_rem(launchtime, cycle_time, &launchtime);
1078	else
1079		launchtime = 0;
1080
1081	return cpu_to_le32(launchtime);
1082}
1083
1084static int igc_init_empty_frame(struct igc_ring *ring,
1085				struct igc_tx_buffer *buffer,
1086				struct sk_buff *skb)
1087{
1088	unsigned int size;
1089	dma_addr_t dma;
1090
1091	size = skb_headlen(skb);
1092
1093	dma = dma_map_single(ring->dev, skb->data, size, DMA_TO_DEVICE);
1094	if (dma_mapping_error(ring->dev, dma)) {
1095		netdev_err_once(ring->netdev, "Failed to map DMA for TX\n");
1096		return -ENOMEM;
1097	}
1098
1099	buffer->skb = skb;
1100	buffer->protocol = 0;
1101	buffer->bytecount = skb->len;
1102	buffer->gso_segs = 1;
1103	buffer->time_stamp = jiffies;
1104	dma_unmap_len_set(buffer, len, skb->len);
1105	dma_unmap_addr_set(buffer, dma, dma);
1106
1107	return 0;
1108}
1109
1110static int igc_init_tx_empty_descriptor(struct igc_ring *ring,
1111					struct sk_buff *skb,
1112					struct igc_tx_buffer *first)
1113{
1114	union igc_adv_tx_desc *desc;
1115	u32 cmd_type, olinfo_status;
1116	int err;
1117
1118	if (!igc_desc_unused(ring))
1119		return -EBUSY;
1120
1121	err = igc_init_empty_frame(ring, first, skb);
1122	if (err)
1123		return err;
1124
1125	cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
1126		   IGC_ADVTXD_DCMD_IFCS | IGC_TXD_DCMD |
1127		   first->bytecount;
1128	olinfo_status = first->bytecount << IGC_ADVTXD_PAYLEN_SHIFT;
1129
1130	desc = IGC_TX_DESC(ring, ring->next_to_use);
1131	desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1132	desc->read.olinfo_status = cpu_to_le32(olinfo_status);
1133	desc->read.buffer_addr = cpu_to_le64(dma_unmap_addr(first, dma));
1134
1135	netdev_tx_sent_queue(txring_txq(ring), skb->len);
1136
1137	first->next_to_watch = desc;
1138
1139	ring->next_to_use++;
1140	if (ring->next_to_use == ring->count)
1141		ring->next_to_use = 0;
1142
1143	return 0;
1144}
1145
1146#define IGC_EMPTY_FRAME_SIZE 60
1147
1148static void igc_tx_ctxtdesc(struct igc_ring *tx_ring,
1149			    __le32 launch_time, bool first_flag,
1150			    u32 vlan_macip_lens, u32 type_tucmd,
1151			    u32 mss_l4len_idx)
1152{
1153	struct igc_adv_tx_context_desc *context_desc;
1154	u16 i = tx_ring->next_to_use;
1155
1156	context_desc = IGC_TX_CTXTDESC(tx_ring, i);
1157
1158	i++;
1159	tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
1160
1161	/* set bits to identify this as an advanced context descriptor */
1162	type_tucmd |= IGC_TXD_CMD_DEXT | IGC_ADVTXD_DTYP_CTXT;
1163
1164	/* For i225, context index must be unique per ring. */
1165	if (test_bit(IGC_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
1166		mss_l4len_idx |= tx_ring->reg_idx << 4;
1167
1168	if (first_flag)
1169		mss_l4len_idx |= IGC_ADVTXD_TSN_CNTX_FIRST;
1170
1171	context_desc->vlan_macip_lens	= cpu_to_le32(vlan_macip_lens);
1172	context_desc->type_tucmd_mlhl	= cpu_to_le32(type_tucmd);
1173	context_desc->mss_l4len_idx	= cpu_to_le32(mss_l4len_idx);
1174	context_desc->launch_time	= launch_time;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1175}
1176
1177static void igc_tx_csum(struct igc_ring *tx_ring, struct igc_tx_buffer *first,
1178			__le32 launch_time, bool first_flag)
1179{
1180	struct sk_buff *skb = first->skb;
1181	u32 vlan_macip_lens = 0;
1182	u32 type_tucmd = 0;
1183
1184	if (skb->ip_summed != CHECKSUM_PARTIAL) {
1185csum_failed:
1186		if (!(first->tx_flags & IGC_TX_FLAGS_VLAN) &&
1187		    !tx_ring->launchtime_enable)
1188			return;
1189		goto no_csum;
1190	}
1191
1192	switch (skb->csum_offset) {
1193	case offsetof(struct tcphdr, check):
1194		type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP;
1195		fallthrough;
1196	case offsetof(struct udphdr, check):
1197		break;
1198	case offsetof(struct sctphdr, checksum):
1199		/* validate that this is actually an SCTP request */
1200		if (skb_csum_is_sctp(skb)) {
 
 
 
1201			type_tucmd = IGC_ADVTXD_TUCMD_L4T_SCTP;
1202			break;
1203		}
1204		fallthrough;
1205	default:
1206		skb_checksum_help(skb);
1207		goto csum_failed;
1208	}
1209
1210	/* update TX checksum flag */
1211	first->tx_flags |= IGC_TX_FLAGS_CSUM;
1212	vlan_macip_lens = skb_checksum_start_offset(skb) -
1213			  skb_network_offset(skb);
1214no_csum:
1215	vlan_macip_lens |= skb_network_offset(skb) << IGC_ADVTXD_MACLEN_SHIFT;
1216	vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK;
1217
1218	igc_tx_ctxtdesc(tx_ring, launch_time, first_flag,
1219			vlan_macip_lens, type_tucmd, 0);
1220}
1221
1222static int __igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size)
1223{
1224	struct net_device *netdev = tx_ring->netdev;
1225
1226	netif_stop_subqueue(netdev, tx_ring->queue_index);
1227
1228	/* memory barriier comment */
1229	smp_mb();
1230
1231	/* We need to check again in a case another CPU has just
1232	 * made room available.
1233	 */
1234	if (igc_desc_unused(tx_ring) < size)
1235		return -EBUSY;
1236
1237	/* A reprieve! */
1238	netif_wake_subqueue(netdev, tx_ring->queue_index);
1239
1240	u64_stats_update_begin(&tx_ring->tx_syncp2);
1241	tx_ring->tx_stats.restart_queue2++;
1242	u64_stats_update_end(&tx_ring->tx_syncp2);
1243
1244	return 0;
1245}
1246
1247static inline int igc_maybe_stop_tx(struct igc_ring *tx_ring, const u16 size)
1248{
1249	if (igc_desc_unused(tx_ring) >= size)
1250		return 0;
1251	return __igc_maybe_stop_tx(tx_ring, size);
1252}
1253
1254#define IGC_SET_FLAG(_input, _flag, _result) \
1255	(((_flag) <= (_result)) ?				\
1256	 ((u32)((_input) & (_flag)) * ((_result) / (_flag))) :	\
1257	 ((u32)((_input) & (_flag)) / ((_flag) / (_result))))
1258
1259static u32 igc_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
1260{
1261	/* set type for advanced descriptor with frame checksum insertion */
1262	u32 cmd_type = IGC_ADVTXD_DTYP_DATA |
1263		       IGC_ADVTXD_DCMD_DEXT |
1264		       IGC_ADVTXD_DCMD_IFCS;
1265
1266	/* set HW vlan bit if vlan is present */
1267	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_VLAN,
1268				 IGC_ADVTXD_DCMD_VLE);
1269
1270	/* set segmentation bits for TSO */
1271	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSO,
1272				 (IGC_ADVTXD_DCMD_TSE));
1273
1274	/* set timestamp bit if present, will select the register set
1275	 * based on the _TSTAMP(_X) bit.
1276	 */
1277	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP,
1278				 (IGC_ADVTXD_MAC_TSTAMP));
1279
1280	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP_1,
1281				 (IGC_ADVTXD_TSTAMP_REG_1));
1282
1283	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP_2,
1284				 (IGC_ADVTXD_TSTAMP_REG_2));
1285
1286	cmd_type |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP_3,
1287				 (IGC_ADVTXD_TSTAMP_REG_3));
1288
1289	/* insert frame checksum */
1290	cmd_type ^= IGC_SET_FLAG(skb->no_fcs, 1, IGC_ADVTXD_DCMD_IFCS);
1291
1292	return cmd_type;
1293}
1294
1295static void igc_tx_olinfo_status(struct igc_ring *tx_ring,
1296				 union igc_adv_tx_desc *tx_desc,
1297				 u32 tx_flags, unsigned int paylen)
1298{
1299	u32 olinfo_status = paylen << IGC_ADVTXD_PAYLEN_SHIFT;
1300
1301	/* insert L4 checksum */
1302	olinfo_status |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_CSUM,
1303				      (IGC_TXD_POPTS_TXSM << 8));
 
1304
1305	/* insert IPv4 checksum */
1306	olinfo_status |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_IPV4,
1307				      (IGC_TXD_POPTS_IXSM << 8));
1308
1309	/* Use the second timer (free running, in general) for the timestamp */
1310	olinfo_status |= IGC_SET_FLAG(tx_flags, IGC_TX_FLAGS_TSTAMP_TIMER_1,
1311				      IGC_TXD_PTP2_TIMER_1);
1312
1313	tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
1314}
1315
1316static int igc_tx_map(struct igc_ring *tx_ring,
1317		      struct igc_tx_buffer *first,
1318		      const u8 hdr_len)
1319{
1320	struct sk_buff *skb = first->skb;
1321	struct igc_tx_buffer *tx_buffer;
1322	union igc_adv_tx_desc *tx_desc;
1323	u32 tx_flags = first->tx_flags;
1324	skb_frag_t *frag;
1325	u16 i = tx_ring->next_to_use;
1326	unsigned int data_len, size;
1327	dma_addr_t dma;
1328	u32 cmd_type;
1329
1330	cmd_type = igc_tx_cmd_type(skb, tx_flags);
1331	tx_desc = IGC_TX_DESC(tx_ring, i);
1332
1333	igc_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
1334
1335	size = skb_headlen(skb);
1336	data_len = skb->data_len;
1337
1338	dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
1339
1340	tx_buffer = first;
1341
1342	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
1343		if (dma_mapping_error(tx_ring->dev, dma))
1344			goto dma_error;
1345
1346		/* record length, and DMA address */
1347		dma_unmap_len_set(tx_buffer, len, size);
1348		dma_unmap_addr_set(tx_buffer, dma, dma);
1349
1350		tx_desc->read.buffer_addr = cpu_to_le64(dma);
1351
1352		while (unlikely(size > IGC_MAX_DATA_PER_TXD)) {
1353			tx_desc->read.cmd_type_len =
1354				cpu_to_le32(cmd_type ^ IGC_MAX_DATA_PER_TXD);
1355
1356			i++;
1357			tx_desc++;
1358			if (i == tx_ring->count) {
1359				tx_desc = IGC_TX_DESC(tx_ring, 0);
1360				i = 0;
1361			}
1362			tx_desc->read.olinfo_status = 0;
1363
1364			dma += IGC_MAX_DATA_PER_TXD;
1365			size -= IGC_MAX_DATA_PER_TXD;
1366
1367			tx_desc->read.buffer_addr = cpu_to_le64(dma);
1368		}
1369
1370		if (likely(!data_len))
1371			break;
1372
1373		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
1374
1375		i++;
1376		tx_desc++;
1377		if (i == tx_ring->count) {
1378			tx_desc = IGC_TX_DESC(tx_ring, 0);
1379			i = 0;
1380		}
1381		tx_desc->read.olinfo_status = 0;
1382
1383		size = skb_frag_size(frag);
1384		data_len -= size;
1385
1386		dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
1387				       size, DMA_TO_DEVICE);
1388
1389		tx_buffer = &tx_ring->tx_buffer_info[i];
1390	}
1391
1392	/* write last descriptor with RS and EOP bits */
1393	cmd_type |= size | IGC_TXD_DCMD;
1394	tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
1395
1396	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
1397
1398	/* set the timestamp */
1399	first->time_stamp = jiffies;
1400
1401	skb_tx_timestamp(skb);
1402
1403	/* Force memory writes to complete before letting h/w know there
1404	 * are new descriptors to fetch.  (Only applicable for weak-ordered
1405	 * memory model archs, such as IA-64).
1406	 *
1407	 * We also need this memory barrier to make certain all of the
1408	 * status bits have been updated before next_to_watch is written.
1409	 */
1410	wmb();
1411
1412	/* set next_to_watch value indicating a packet is present */
1413	first->next_to_watch = tx_desc;
1414
1415	i++;
1416	if (i == tx_ring->count)
1417		i = 0;
1418
1419	tx_ring->next_to_use = i;
1420
1421	/* Make sure there is space in the ring for the next send. */
1422	igc_maybe_stop_tx(tx_ring, DESC_NEEDED);
1423
1424	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
1425		writel(i, tx_ring->tail);
1426	}
1427
1428	return 0;
1429dma_error:
1430	netdev_err(tx_ring->netdev, "TX DMA map failed\n");
1431	tx_buffer = &tx_ring->tx_buffer_info[i];
1432
1433	/* clear dma mappings for failed tx_buffer_info map */
1434	while (tx_buffer != first) {
1435		if (dma_unmap_len(tx_buffer, len))
1436			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 
 
 
 
1437
1438		if (i-- == 0)
1439			i += tx_ring->count;
1440		tx_buffer = &tx_ring->tx_buffer_info[i];
1441	}
1442
1443	if (dma_unmap_len(tx_buffer, len))
1444		igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 
 
 
 
1445
1446	dev_kfree_skb_any(tx_buffer->skb);
1447	tx_buffer->skb = NULL;
1448
1449	tx_ring->next_to_use = i;
1450
1451	return -1;
1452}
1453
1454static int igc_tso(struct igc_ring *tx_ring,
1455		   struct igc_tx_buffer *first,
1456		   __le32 launch_time, bool first_flag,
1457		   u8 *hdr_len)
1458{
1459	u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
1460	struct sk_buff *skb = first->skb;
1461	union {
1462		struct iphdr *v4;
1463		struct ipv6hdr *v6;
1464		unsigned char *hdr;
1465	} ip;
1466	union {
1467		struct tcphdr *tcp;
1468		struct udphdr *udp;
1469		unsigned char *hdr;
1470	} l4;
1471	u32 paylen, l4_offset;
1472	int err;
1473
1474	if (skb->ip_summed != CHECKSUM_PARTIAL)
1475		return 0;
1476
1477	if (!skb_is_gso(skb))
1478		return 0;
1479
1480	err = skb_cow_head(skb, 0);
1481	if (err < 0)
1482		return err;
1483
1484	ip.hdr = skb_network_header(skb);
1485	l4.hdr = skb_checksum_start(skb);
1486
1487	/* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
1488	type_tucmd = IGC_ADVTXD_TUCMD_L4T_TCP;
1489
1490	/* initialize outer IP header fields */
1491	if (ip.v4->version == 4) {
1492		unsigned char *csum_start = skb_checksum_start(skb);
1493		unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
1494
1495		/* IP header will have to cancel out any data that
1496		 * is not a part of the outer IP header
1497		 */
1498		ip.v4->check = csum_fold(csum_partial(trans_start,
1499						      csum_start - trans_start,
1500						      0));
1501		type_tucmd |= IGC_ADVTXD_TUCMD_IPV4;
1502
1503		ip.v4->tot_len = 0;
1504		first->tx_flags |= IGC_TX_FLAGS_TSO |
1505				   IGC_TX_FLAGS_CSUM |
1506				   IGC_TX_FLAGS_IPV4;
1507	} else {
1508		ip.v6->payload_len = 0;
1509		first->tx_flags |= IGC_TX_FLAGS_TSO |
1510				   IGC_TX_FLAGS_CSUM;
1511	}
1512
1513	/* determine offset of inner transport header */
1514	l4_offset = l4.hdr - skb->data;
1515
1516	/* remove payload length from inner checksum */
1517	paylen = skb->len - l4_offset;
1518	if (type_tucmd & IGC_ADVTXD_TUCMD_L4T_TCP) {
1519		/* compute length of segmentation header */
1520		*hdr_len = (l4.tcp->doff * 4) + l4_offset;
1521		csum_replace_by_diff(&l4.tcp->check,
1522				     (__force __wsum)htonl(paylen));
1523	} else {
1524		/* compute length of segmentation header */
1525		*hdr_len = sizeof(*l4.udp) + l4_offset;
1526		csum_replace_by_diff(&l4.udp->check,
1527				     (__force __wsum)htonl(paylen));
1528	}
1529
1530	/* update gso size and bytecount with header size */
1531	first->gso_segs = skb_shinfo(skb)->gso_segs;
1532	first->bytecount += (first->gso_segs - 1) * *hdr_len;
1533
1534	/* MSS L4LEN IDX */
1535	mss_l4len_idx = (*hdr_len - l4_offset) << IGC_ADVTXD_L4LEN_SHIFT;
1536	mss_l4len_idx |= skb_shinfo(skb)->gso_size << IGC_ADVTXD_MSS_SHIFT;
1537
1538	/* VLAN MACLEN IPLEN */
1539	vlan_macip_lens = l4.hdr - ip.hdr;
1540	vlan_macip_lens |= (ip.hdr - skb->data) << IGC_ADVTXD_MACLEN_SHIFT;
1541	vlan_macip_lens |= first->tx_flags & IGC_TX_FLAGS_VLAN_MASK;
1542
1543	igc_tx_ctxtdesc(tx_ring, launch_time, first_flag,
1544			vlan_macip_lens, type_tucmd, mss_l4len_idx);
1545
1546	return 1;
1547}
1548
1549static bool igc_request_tx_tstamp(struct igc_adapter *adapter, struct sk_buff *skb, u32 *flags)
1550{
1551	int i;
1552
1553	for (i = 0; i < IGC_MAX_TX_TSTAMP_REGS; i++) {
1554		struct igc_tx_timestamp_request *tstamp = &adapter->tx_tstamp[i];
1555
1556		if (tstamp->skb)
1557			continue;
1558
1559		tstamp->skb = skb_get(skb);
1560		tstamp->start = jiffies;
1561		*flags = tstamp->flags;
1562
1563		return true;
1564	}
1565
1566	return false;
1567}
1568
1569static netdev_tx_t igc_xmit_frame_ring(struct sk_buff *skb,
1570				       struct igc_ring *tx_ring)
1571{
1572	struct igc_adapter *adapter = netdev_priv(tx_ring->netdev);
1573	bool first_flag = false, insert_empty = false;
1574	u16 count = TXD_USE_COUNT(skb_headlen(skb));
1575	__be16 protocol = vlan_get_protocol(skb);
1576	struct igc_tx_buffer *first;
1577	__le32 launch_time = 0;
1578	u32 tx_flags = 0;
1579	unsigned short f;
1580	ktime_t txtime;
1581	u8 hdr_len = 0;
1582	int tso = 0;
1583
1584	/* need: 1 descriptor per page * PAGE_SIZE/IGC_MAX_DATA_PER_TXD,
1585	 *	+ 1 desc for skb_headlen/IGC_MAX_DATA_PER_TXD,
1586	 *	+ 2 desc gap to keep tail from touching head,
1587	 *	+ 1 desc for context descriptor,
1588	 * otherwise try next time
1589	 */
1590	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1591		count += TXD_USE_COUNT(skb_frag_size(
1592						&skb_shinfo(skb)->frags[f]));
1593
1594	if (igc_maybe_stop_tx(tx_ring, count + 5)) {
1595		/* this is a hard error */
1596		return NETDEV_TX_BUSY;
1597	}
1598
1599	if (!tx_ring->launchtime_enable)
1600		goto done;
1601
1602	txtime = skb->tstamp;
1603	skb->tstamp = ktime_set(0, 0);
1604	launch_time = igc_tx_launchtime(tx_ring, txtime, &first_flag, &insert_empty);
1605
1606	if (insert_empty) {
1607		struct igc_tx_buffer *empty_info;
1608		struct sk_buff *empty;
1609		void *data;
1610
1611		empty_info = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
1612		empty = alloc_skb(IGC_EMPTY_FRAME_SIZE, GFP_ATOMIC);
1613		if (!empty)
1614			goto done;
1615
1616		data = skb_put(empty, IGC_EMPTY_FRAME_SIZE);
1617		memset(data, 0, IGC_EMPTY_FRAME_SIZE);
1618
1619		igc_tx_ctxtdesc(tx_ring, 0, false, 0, 0, 0);
1620
1621		if (igc_init_tx_empty_descriptor(tx_ring,
1622						 empty,
1623						 empty_info) < 0)
1624			dev_kfree_skb_any(empty);
1625	}
1626
1627done:
1628	/* record the location of the first descriptor for this packet */
1629	first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
1630	first->type = IGC_TX_BUFFER_TYPE_SKB;
1631	first->skb = skb;
1632	first->bytecount = skb->len;
1633	first->gso_segs = 1;
1634
1635	if (adapter->qbv_transition || tx_ring->oper_gate_closed)
1636		goto out_drop;
1637
1638	if (tx_ring->max_sdu > 0 && first->bytecount > tx_ring->max_sdu) {
1639		adapter->stats.txdrop++;
1640		goto out_drop;
1641	}
1642
1643	if (unlikely(test_bit(IGC_RING_FLAG_TX_HWTSTAMP, &tx_ring->flags) &&
1644		     skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
1645		/* FIXME: add support for retrieving timestamps from
1646		 * the other timer registers before skipping the
1647		 * timestamping request.
1648		 */
1649		unsigned long flags;
1650		u32 tstamp_flags;
1651
1652		spin_lock_irqsave(&adapter->ptp_tx_lock, flags);
1653		if (igc_request_tx_tstamp(adapter, skb, &tstamp_flags)) {
1654			skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
1655			tx_flags |= IGC_TX_FLAGS_TSTAMP | tstamp_flags;
1656			if (skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP_USE_CYCLES)
1657				tx_flags |= IGC_TX_FLAGS_TSTAMP_TIMER_1;
 
1658		} else {
1659			adapter->tx_hwtstamp_skipped++;
1660		}
1661
1662		spin_unlock_irqrestore(&adapter->ptp_tx_lock, flags);
1663	}
1664
1665	if (skb_vlan_tag_present(skb)) {
1666		tx_flags |= IGC_TX_FLAGS_VLAN;
1667		tx_flags |= (skb_vlan_tag_get(skb) << IGC_TX_FLAGS_VLAN_SHIFT);
1668	}
1669
1670	/* record initial flags and protocol */
1671	first->tx_flags = tx_flags;
1672	first->protocol = protocol;
1673
1674	tso = igc_tso(tx_ring, first, launch_time, first_flag, &hdr_len);
1675	if (tso < 0)
1676		goto out_drop;
1677	else if (!tso)
1678		igc_tx_csum(tx_ring, first, launch_time, first_flag);
1679
1680	igc_tx_map(tx_ring, first, hdr_len);
1681
1682	return NETDEV_TX_OK;
1683
1684out_drop:
1685	dev_kfree_skb_any(first->skb);
1686	first->skb = NULL;
1687
1688	return NETDEV_TX_OK;
1689}
1690
1691static inline struct igc_ring *igc_tx_queue_mapping(struct igc_adapter *adapter,
1692						    struct sk_buff *skb)
1693{
1694	unsigned int r_idx = skb->queue_mapping;
1695
1696	if (r_idx >= adapter->num_tx_queues)
1697		r_idx = r_idx % adapter->num_tx_queues;
1698
1699	return adapter->tx_ring[r_idx];
1700}
1701
1702static netdev_tx_t igc_xmit_frame(struct sk_buff *skb,
1703				  struct net_device *netdev)
1704{
1705	struct igc_adapter *adapter = netdev_priv(netdev);
1706
1707	/* The minimum packet size with TCTL.PSP set is 17 so pad the skb
1708	 * in order to meet this minimum size requirement.
1709	 */
1710	if (skb->len < 17) {
1711		if (skb_padto(skb, 17))
1712			return NETDEV_TX_OK;
1713		skb->len = 17;
1714	}
1715
1716	return igc_xmit_frame_ring(skb, igc_tx_queue_mapping(adapter, skb));
1717}
1718
1719static void igc_rx_checksum(struct igc_ring *ring,
1720			    union igc_adv_rx_desc *rx_desc,
1721			    struct sk_buff *skb)
1722{
1723	skb_checksum_none_assert(skb);
1724
1725	/* Ignore Checksum bit is set */
1726	if (igc_test_staterr(rx_desc, IGC_RXD_STAT_IXSM))
1727		return;
1728
1729	/* Rx checksum disabled via ethtool */
1730	if (!(ring->netdev->features & NETIF_F_RXCSUM))
1731		return;
1732
1733	/* TCP/UDP checksum error bit is set */
1734	if (igc_test_staterr(rx_desc,
1735			     IGC_RXDEXT_STATERR_L4E |
1736			     IGC_RXDEXT_STATERR_IPE)) {
1737		/* work around errata with sctp packets where the TCPE aka
1738		 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
1739		 * packets (aka let the stack check the crc32c)
1740		 */
1741		if (!(skb->len == 60 &&
1742		      test_bit(IGC_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
1743			u64_stats_update_begin(&ring->rx_syncp);
1744			ring->rx_stats.csum_err++;
1745			u64_stats_update_end(&ring->rx_syncp);
1746		}
1747		/* let the stack verify checksum errors */
1748		return;
1749	}
1750	/* It must be a TCP or UDP packet with a valid checksum */
1751	if (igc_test_staterr(rx_desc, IGC_RXD_STAT_TCPCS |
1752				      IGC_RXD_STAT_UDPCS))
1753		skb->ip_summed = CHECKSUM_UNNECESSARY;
1754
1755	netdev_dbg(ring->netdev, "cksum success: bits %08X\n",
1756		   le32_to_cpu(rx_desc->wb.upper.status_error));
1757}
1758
1759/* Mapping HW RSS Type to enum pkt_hash_types */
1760static const enum pkt_hash_types igc_rss_type_table[IGC_RSS_TYPE_MAX_TABLE] = {
1761	[IGC_RSS_TYPE_NO_HASH]		= PKT_HASH_TYPE_L2,
1762	[IGC_RSS_TYPE_HASH_TCP_IPV4]	= PKT_HASH_TYPE_L4,
1763	[IGC_RSS_TYPE_HASH_IPV4]	= PKT_HASH_TYPE_L3,
1764	[IGC_RSS_TYPE_HASH_TCP_IPV6]	= PKT_HASH_TYPE_L4,
1765	[IGC_RSS_TYPE_HASH_IPV6_EX]	= PKT_HASH_TYPE_L3,
1766	[IGC_RSS_TYPE_HASH_IPV6]	= PKT_HASH_TYPE_L3,
1767	[IGC_RSS_TYPE_HASH_TCP_IPV6_EX] = PKT_HASH_TYPE_L4,
1768	[IGC_RSS_TYPE_HASH_UDP_IPV4]	= PKT_HASH_TYPE_L4,
1769	[IGC_RSS_TYPE_HASH_UDP_IPV6]	= PKT_HASH_TYPE_L4,
1770	[IGC_RSS_TYPE_HASH_UDP_IPV6_EX] = PKT_HASH_TYPE_L4,
1771	[10] = PKT_HASH_TYPE_NONE, /* RSS Type above 9 "Reserved" by HW  */
1772	[11] = PKT_HASH_TYPE_NONE, /* keep array sized for SW bit-mask   */
1773	[12] = PKT_HASH_TYPE_NONE, /* to handle future HW revisons       */
1774	[13] = PKT_HASH_TYPE_NONE,
1775	[14] = PKT_HASH_TYPE_NONE,
1776	[15] = PKT_HASH_TYPE_NONE,
1777};
1778
1779static inline void igc_rx_hash(struct igc_ring *ring,
1780			       union igc_adv_rx_desc *rx_desc,
1781			       struct sk_buff *skb)
1782{
1783	if (ring->netdev->features & NETIF_F_RXHASH) {
1784		u32 rss_hash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
1785		u32 rss_type = igc_rss_type(rx_desc);
1786
1787		skb_set_hash(skb, rss_hash, igc_rss_type_table[rss_type]);
1788	}
1789}
1790
1791static void igc_rx_vlan(struct igc_ring *rx_ring,
1792			union igc_adv_rx_desc *rx_desc,
1793			struct sk_buff *skb)
1794{
1795	struct net_device *dev = rx_ring->netdev;
1796	u16 vid;
1797
1798	if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1799	    igc_test_staterr(rx_desc, IGC_RXD_STAT_VP)) {
1800		if (igc_test_staterr(rx_desc, IGC_RXDEXT_STATERR_LB) &&
1801		    test_bit(IGC_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
1802			vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan);
1803		else
1804			vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1805
1806		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1807	}
1808}
1809
1810/**
1811 * igc_process_skb_fields - Populate skb header fields from Rx descriptor
1812 * @rx_ring: rx descriptor ring packet is being transacted on
1813 * @rx_desc: pointer to the EOP Rx descriptor
1814 * @skb: pointer to current skb being populated
1815 *
1816 * This function checks the ring, descriptor, and packet information in order
1817 * to populate the hash, checksum, VLAN, protocol, and other fields within the
1818 * skb.
1819 */
1820static void igc_process_skb_fields(struct igc_ring *rx_ring,
1821				   union igc_adv_rx_desc *rx_desc,
1822				   struct sk_buff *skb)
1823{
1824	igc_rx_hash(rx_ring, rx_desc, skb);
1825
1826	igc_rx_checksum(rx_ring, rx_desc, skb);
1827
1828	igc_rx_vlan(rx_ring, rx_desc, skb);
1829
1830	skb_record_rx_queue(skb, rx_ring->queue_index);
1831
1832	skb->protocol = eth_type_trans(skb, rx_ring->netdev);
1833}
1834
1835static void igc_vlan_mode(struct net_device *netdev, netdev_features_t features)
1836{
1837	bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
1838	struct igc_adapter *adapter = netdev_priv(netdev);
1839	struct igc_hw *hw = &adapter->hw;
1840	u32 ctrl;
1841
1842	ctrl = rd32(IGC_CTRL);
1843
1844	if (enable) {
1845		/* enable VLAN tag insert/strip */
1846		ctrl |= IGC_CTRL_VME;
1847	} else {
1848		/* disable VLAN tag insert/strip */
1849		ctrl &= ~IGC_CTRL_VME;
1850	}
1851	wr32(IGC_CTRL, ctrl);
1852}
1853
1854static void igc_restore_vlan(struct igc_adapter *adapter)
1855{
1856	igc_vlan_mode(adapter->netdev, adapter->netdev->features);
1857}
1858
1859static struct igc_rx_buffer *igc_get_rx_buffer(struct igc_ring *rx_ring,
1860					       const unsigned int size,
1861					       int *rx_buffer_pgcnt)
1862{
1863	struct igc_rx_buffer *rx_buffer;
1864
1865	rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1866	*rx_buffer_pgcnt =
1867#if (PAGE_SIZE < 8192)
1868		page_count(rx_buffer->page);
1869#else
1870		0;
1871#endif
1872	prefetchw(rx_buffer->page);
1873
1874	/* we are reusing so sync this buffer for CPU use */
1875	dma_sync_single_range_for_cpu(rx_ring->dev,
1876				      rx_buffer->dma,
1877				      rx_buffer->page_offset,
1878				      size,
1879				      DMA_FROM_DEVICE);
1880
1881	rx_buffer->pagecnt_bias--;
1882
1883	return rx_buffer;
1884}
1885
1886static void igc_rx_buffer_flip(struct igc_rx_buffer *buffer,
1887			       unsigned int truesize)
1888{
1889#if (PAGE_SIZE < 8192)
1890	buffer->page_offset ^= truesize;
1891#else
1892	buffer->page_offset += truesize;
1893#endif
1894}
1895
1896static unsigned int igc_get_rx_frame_truesize(struct igc_ring *ring,
1897					      unsigned int size)
1898{
1899	unsigned int truesize;
1900
1901#if (PAGE_SIZE < 8192)
1902	truesize = igc_rx_pg_size(ring) / 2;
1903#else
1904	truesize = ring_uses_build_skb(ring) ?
1905		   SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
1906		   SKB_DATA_ALIGN(IGC_SKB_PAD + size) :
1907		   SKB_DATA_ALIGN(size);
1908#endif
1909	return truesize;
1910}
1911
1912/**
1913 * igc_add_rx_frag - Add contents of Rx buffer to sk_buff
1914 * @rx_ring: rx descriptor ring to transact packets on
1915 * @rx_buffer: buffer containing page to add
1916 * @skb: sk_buff to place the data into
1917 * @size: size of buffer to be added
1918 *
1919 * This function will add the data contained in rx_buffer->page to the skb.
1920 */
1921static void igc_add_rx_frag(struct igc_ring *rx_ring,
1922			    struct igc_rx_buffer *rx_buffer,
1923			    struct sk_buff *skb,
1924			    unsigned int size)
1925{
1926	unsigned int truesize;
1927
1928#if (PAGE_SIZE < 8192)
1929	truesize = igc_rx_pg_size(rx_ring) / 2;
 
 
 
 
1930#else
1931	truesize = ring_uses_build_skb(rx_ring) ?
1932		   SKB_DATA_ALIGN(IGC_SKB_PAD + size) :
1933		   SKB_DATA_ALIGN(size);
1934#endif
1935	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
1936			rx_buffer->page_offset, size, truesize);
1937
1938	igc_rx_buffer_flip(rx_buffer, truesize);
1939}
1940
1941static struct sk_buff *igc_build_skb(struct igc_ring *rx_ring,
1942				     struct igc_rx_buffer *rx_buffer,
1943				     struct xdp_buff *xdp)
 
1944{
1945	unsigned int size = xdp->data_end - xdp->data;
1946	unsigned int truesize = igc_get_rx_frame_truesize(rx_ring, size);
1947	unsigned int metasize = xdp->data - xdp->data_meta;
 
 
 
 
1948	struct sk_buff *skb;
1949
1950	/* prefetch first cache line of first page */
1951	net_prefetch(xdp->data_meta);
 
 
 
1952
1953	/* build an skb around the page buffer */
1954	skb = napi_build_skb(xdp->data_hard_start, truesize);
1955	if (unlikely(!skb))
1956		return NULL;
1957
1958	/* update pointers within the skb to store the data */
1959	skb_reserve(skb, xdp->data - xdp->data_hard_start);
1960	__skb_put(skb, size);
1961	if (metasize)
1962		skb_metadata_set(skb, metasize);
1963
1964	igc_rx_buffer_flip(rx_buffer, truesize);
 
 
 
 
 
 
1965	return skb;
1966}
1967
1968static struct sk_buff *igc_construct_skb(struct igc_ring *rx_ring,
1969					 struct igc_rx_buffer *rx_buffer,
1970					 struct igc_xdp_buff *ctx)
 
1971{
1972	struct xdp_buff *xdp = &ctx->xdp;
1973	unsigned int metasize = xdp->data - xdp->data_meta;
1974	unsigned int size = xdp->data_end - xdp->data;
1975	unsigned int truesize = igc_get_rx_frame_truesize(rx_ring, size);
1976	void *va = xdp->data;
 
1977	unsigned int headlen;
1978	struct sk_buff *skb;
1979
1980	/* prefetch first cache line of first page */
1981	net_prefetch(xdp->data_meta);
 
 
 
1982
1983	/* allocate a skb to store the frags */
1984	skb = napi_alloc_skb(&rx_ring->q_vector->napi,
1985			     IGC_RX_HDR_LEN + metasize);
1986	if (unlikely(!skb))
1987		return NULL;
1988
1989	if (ctx->rx_ts) {
1990		skb_shinfo(skb)->tx_flags |= SKBTX_HW_TSTAMP_NETDEV;
1991		skb_hwtstamps(skb)->netdev_data = ctx->rx_ts;
 
1992	}
1993
1994	/* Determine available headroom for copy */
1995	headlen = size;
1996	if (headlen > IGC_RX_HDR_LEN)
1997		headlen = eth_get_headlen(skb->dev, va, IGC_RX_HDR_LEN);
1998
1999	/* align pull length to size of long to optimize memcpy performance */
2000	memcpy(__skb_put(skb, headlen + metasize), xdp->data_meta,
2001	       ALIGN(headlen + metasize, sizeof(long)));
2002
2003	if (metasize) {
2004		skb_metadata_set(skb, metasize);
2005		__skb_pull(skb, metasize);
2006	}
2007
2008	/* update all of the pointers */
2009	size -= headlen;
2010	if (size) {
2011		skb_add_rx_frag(skb, 0, rx_buffer->page,
2012				(va + headlen) - page_address(rx_buffer->page),
2013				size, truesize);
2014		igc_rx_buffer_flip(rx_buffer, truesize);
 
 
 
 
2015	} else {
2016		rx_buffer->pagecnt_bias++;
2017	}
2018
2019	return skb;
2020}
2021
2022/**
2023 * igc_reuse_rx_page - page flip buffer and store it back on the ring
2024 * @rx_ring: rx descriptor ring to store buffers on
2025 * @old_buff: donor buffer to have page reused
2026 *
2027 * Synchronizes page for reuse by the adapter
2028 */
2029static void igc_reuse_rx_page(struct igc_ring *rx_ring,
2030			      struct igc_rx_buffer *old_buff)
2031{
2032	u16 nta = rx_ring->next_to_alloc;
2033	struct igc_rx_buffer *new_buff;
2034
2035	new_buff = &rx_ring->rx_buffer_info[nta];
2036
2037	/* update, and store next to alloc */
2038	nta++;
2039	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
2040
2041	/* Transfer page from old buffer to new buffer.
2042	 * Move each member individually to avoid possible store
2043	 * forwarding stalls.
2044	 */
2045	new_buff->dma		= old_buff->dma;
2046	new_buff->page		= old_buff->page;
2047	new_buff->page_offset	= old_buff->page_offset;
2048	new_buff->pagecnt_bias	= old_buff->pagecnt_bias;
2049}
2050
2051static bool igc_can_reuse_rx_page(struct igc_rx_buffer *rx_buffer,
2052				  int rx_buffer_pgcnt)
 
 
 
 
2053{
2054	unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
2055	struct page *page = rx_buffer->page;
2056
2057	/* avoid re-using remote and pfmemalloc pages */
2058	if (!dev_page_is_reusable(page))
2059		return false;
2060
2061#if (PAGE_SIZE < 8192)
2062	/* if we are only owner of page we can reuse it */
2063	if (unlikely((rx_buffer_pgcnt - pagecnt_bias) > 1))
2064		return false;
2065#else
2066#define IGC_LAST_OFFSET \
2067	(SKB_WITH_OVERHEAD(PAGE_SIZE) - IGC_RXBUFFER_2048)
2068
2069	if (rx_buffer->page_offset > IGC_LAST_OFFSET)
2070		return false;
2071#endif
2072
2073	/* If we have drained the page fragment pool we need to update
2074	 * the pagecnt_bias and page count so that we fully restock the
2075	 * number of references the driver holds.
2076	 */
2077	if (unlikely(pagecnt_bias == 1)) {
2078		page_ref_add(page, USHRT_MAX - 1);
2079		rx_buffer->pagecnt_bias = USHRT_MAX;
2080	}
2081
2082	return true;
2083}
2084
2085/**
2086 * igc_is_non_eop - process handling of non-EOP buffers
2087 * @rx_ring: Rx ring being processed
2088 * @rx_desc: Rx descriptor for current buffer
2089 *
2090 * This function updates next to clean.  If the buffer is an EOP buffer
2091 * this function exits returning false, otherwise it will place the
2092 * sk_buff in the next buffer to be chained and return true indicating
2093 * that this is in fact a non-EOP buffer.
2094 */
2095static bool igc_is_non_eop(struct igc_ring *rx_ring,
2096			   union igc_adv_rx_desc *rx_desc)
2097{
2098	u32 ntc = rx_ring->next_to_clean + 1;
2099
2100	/* fetch, update, and store next to clean */
2101	ntc = (ntc < rx_ring->count) ? ntc : 0;
2102	rx_ring->next_to_clean = ntc;
2103
2104	prefetch(IGC_RX_DESC(rx_ring, ntc));
2105
2106	if (likely(igc_test_staterr(rx_desc, IGC_RXD_STAT_EOP)))
2107		return false;
2108
2109	return true;
2110}
2111
2112/**
2113 * igc_cleanup_headers - Correct corrupted or empty headers
2114 * @rx_ring: rx descriptor ring packet is being transacted on
2115 * @rx_desc: pointer to the EOP Rx descriptor
2116 * @skb: pointer to current skb being fixed
2117 *
2118 * Address the case where we are pulling data in on pages only
2119 * and as such no data is present in the skb header.
2120 *
2121 * In addition if skb is not at least 60 bytes we need to pad it so that
2122 * it is large enough to qualify as a valid Ethernet frame.
2123 *
2124 * Returns true if an error was encountered and skb was freed.
2125 */
2126static bool igc_cleanup_headers(struct igc_ring *rx_ring,
2127				union igc_adv_rx_desc *rx_desc,
2128				struct sk_buff *skb)
2129{
2130	/* XDP packets use error pointer so abort at this point */
2131	if (IS_ERR(skb))
2132		return true;
2133
2134	if (unlikely(igc_test_staterr(rx_desc, IGC_RXDEXT_STATERR_RXE))) {
2135		struct net_device *netdev = rx_ring->netdev;
2136
2137		if (!(netdev->features & NETIF_F_RXALL)) {
2138			dev_kfree_skb_any(skb);
2139			return true;
2140		}
2141	}
2142
2143	/* if eth_skb_pad returns an error the skb was freed */
2144	if (eth_skb_pad(skb))
2145		return true;
2146
2147	return false;
2148}
2149
2150static void igc_put_rx_buffer(struct igc_ring *rx_ring,
2151			      struct igc_rx_buffer *rx_buffer,
2152			      int rx_buffer_pgcnt)
2153{
2154	if (igc_can_reuse_rx_page(rx_buffer, rx_buffer_pgcnt)) {
2155		/* hand second half of page back to the ring */
2156		igc_reuse_rx_page(rx_ring, rx_buffer);
2157	} else {
2158		/* We are not reusing the buffer so unmap it and free
2159		 * any references we are holding to it
2160		 */
2161		dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2162				     igc_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
2163				     IGC_RX_DMA_ATTR);
2164		__page_frag_cache_drain(rx_buffer->page,
2165					rx_buffer->pagecnt_bias);
2166	}
2167
2168	/* clear contents of rx_buffer */
2169	rx_buffer->page = NULL;
2170}
2171
2172static inline unsigned int igc_rx_offset(struct igc_ring *rx_ring)
2173{
2174	struct igc_adapter *adapter = rx_ring->q_vector->adapter;
2175
2176	if (ring_uses_build_skb(rx_ring))
2177		return IGC_SKB_PAD;
2178	if (igc_xdp_is_enabled(adapter))
2179		return XDP_PACKET_HEADROOM;
2180
2181	return 0;
2182}
2183
2184static bool igc_alloc_mapped_page(struct igc_ring *rx_ring,
2185				  struct igc_rx_buffer *bi)
2186{
2187	struct page *page = bi->page;
2188	dma_addr_t dma;
2189
2190	/* since we are recycling buffers we should seldom need to alloc */
2191	if (likely(page))
2192		return true;
2193
2194	/* alloc new page for storage */
2195	page = dev_alloc_pages(igc_rx_pg_order(rx_ring));
2196	if (unlikely(!page)) {
2197		rx_ring->rx_stats.alloc_failed++;
2198		return false;
2199	}
2200
2201	/* map page for use */
2202	dma = dma_map_page_attrs(rx_ring->dev, page, 0,
2203				 igc_rx_pg_size(rx_ring),
2204				 DMA_FROM_DEVICE,
2205				 IGC_RX_DMA_ATTR);
2206
2207	/* if mapping failed free memory back to system since
2208	 * there isn't much point in holding memory we can't use
2209	 */
2210	if (dma_mapping_error(rx_ring->dev, dma)) {
2211		__free_page(page);
2212
2213		rx_ring->rx_stats.alloc_failed++;
2214		return false;
2215	}
2216
2217	bi->dma = dma;
2218	bi->page = page;
2219	bi->page_offset = igc_rx_offset(rx_ring);
2220	page_ref_add(page, USHRT_MAX - 1);
2221	bi->pagecnt_bias = USHRT_MAX;
2222
2223	return true;
2224}
2225
2226/**
2227 * igc_alloc_rx_buffers - Replace used receive buffers; packet split
2228 * @rx_ring: rx descriptor ring
2229 * @cleaned_count: number of buffers to clean
2230 */
2231static void igc_alloc_rx_buffers(struct igc_ring *rx_ring, u16 cleaned_count)
2232{
2233	union igc_adv_rx_desc *rx_desc;
2234	u16 i = rx_ring->next_to_use;
2235	struct igc_rx_buffer *bi;
2236	u16 bufsz;
2237
2238	/* nothing to do */
2239	if (!cleaned_count)
2240		return;
2241
2242	rx_desc = IGC_RX_DESC(rx_ring, i);
2243	bi = &rx_ring->rx_buffer_info[i];
2244	i -= rx_ring->count;
2245
2246	bufsz = igc_rx_bufsz(rx_ring);
2247
2248	do {
2249		if (!igc_alloc_mapped_page(rx_ring, bi))
2250			break;
2251
2252		/* sync the buffer for use by the device */
2253		dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
2254						 bi->page_offset, bufsz,
2255						 DMA_FROM_DEVICE);
2256
2257		/* Refresh the desc even if buffer_addrs didn't change
2258		 * because each write-back erases this info.
2259		 */
2260		rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
2261
2262		rx_desc++;
2263		bi++;
2264		i++;
2265		if (unlikely(!i)) {
2266			rx_desc = IGC_RX_DESC(rx_ring, 0);
2267			bi = rx_ring->rx_buffer_info;
2268			i -= rx_ring->count;
2269		}
2270
2271		/* clear the length for the next_to_use descriptor */
2272		rx_desc->wb.upper.length = 0;
2273
2274		cleaned_count--;
2275	} while (cleaned_count);
2276
2277	i += rx_ring->count;
2278
2279	if (rx_ring->next_to_use != i) {
2280		/* record the next descriptor to use */
2281		rx_ring->next_to_use = i;
2282
2283		/* update next to alloc since we have filled the ring */
2284		rx_ring->next_to_alloc = i;
2285
2286		/* Force memory writes to complete before letting h/w
2287		 * know there are new descriptors to fetch.  (Only
2288		 * applicable for weak-ordered memory model archs,
2289		 * such as IA-64).
2290		 */
2291		wmb();
2292		writel(i, rx_ring->tail);
2293	}
2294}
2295
2296static bool igc_alloc_rx_buffers_zc(struct igc_ring *ring, u16 count)
2297{
2298	union igc_adv_rx_desc *desc;
2299	u16 i = ring->next_to_use;
2300	struct igc_rx_buffer *bi;
2301	dma_addr_t dma;
2302	bool ok = true;
2303
2304	if (!count)
2305		return ok;
2306
2307	XSK_CHECK_PRIV_TYPE(struct igc_xdp_buff);
2308
2309	desc = IGC_RX_DESC(ring, i);
2310	bi = &ring->rx_buffer_info[i];
2311	i -= ring->count;
2312
2313	do {
2314		bi->xdp = xsk_buff_alloc(ring->xsk_pool);
2315		if (!bi->xdp) {
2316			ok = false;
2317			break;
2318		}
2319
2320		dma = xsk_buff_xdp_get_dma(bi->xdp);
2321		desc->read.pkt_addr = cpu_to_le64(dma);
2322
2323		desc++;
2324		bi++;
2325		i++;
2326		if (unlikely(!i)) {
2327			desc = IGC_RX_DESC(ring, 0);
2328			bi = ring->rx_buffer_info;
2329			i -= ring->count;
2330		}
2331
2332		/* Clear the length for the next_to_use descriptor. */
2333		desc->wb.upper.length = 0;
2334
2335		count--;
2336	} while (count);
2337
2338	i += ring->count;
2339
2340	if (ring->next_to_use != i) {
2341		ring->next_to_use = i;
2342
2343		/* Force memory writes to complete before letting h/w
2344		 * know there are new descriptors to fetch.  (Only
2345		 * applicable for weak-ordered memory model archs,
2346		 * such as IA-64).
2347		 */
2348		wmb();
2349		writel(i, ring->tail);
2350	}
2351
2352	return ok;
2353}
2354
2355/* This function requires __netif_tx_lock is held by the caller. */
2356static int igc_xdp_init_tx_descriptor(struct igc_ring *ring,
2357				      struct xdp_frame *xdpf)
2358{
2359	struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
2360	u8 nr_frags = unlikely(xdp_frame_has_frags(xdpf)) ? sinfo->nr_frags : 0;
2361	u16 count, index = ring->next_to_use;
2362	struct igc_tx_buffer *head = &ring->tx_buffer_info[index];
2363	struct igc_tx_buffer *buffer = head;
2364	union igc_adv_tx_desc *desc = IGC_TX_DESC(ring, index);
2365	u32 olinfo_status, len = xdpf->len, cmd_type;
2366	void *data = xdpf->data;
2367	u16 i;
2368
2369	count = TXD_USE_COUNT(len);
2370	for (i = 0; i < nr_frags; i++)
2371		count += TXD_USE_COUNT(skb_frag_size(&sinfo->frags[i]));
2372
2373	if (igc_maybe_stop_tx(ring, count + 3)) {
2374		/* this is a hard error */
2375		return -EBUSY;
2376	}
2377
2378	i = 0;
2379	head->bytecount = xdp_get_frame_len(xdpf);
2380	head->type = IGC_TX_BUFFER_TYPE_XDP;
2381	head->gso_segs = 1;
2382	head->xdpf = xdpf;
2383
2384	olinfo_status = head->bytecount << IGC_ADVTXD_PAYLEN_SHIFT;
2385	desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2386
2387	for (;;) {
2388		dma_addr_t dma;
2389
2390		dma = dma_map_single(ring->dev, data, len, DMA_TO_DEVICE);
2391		if (dma_mapping_error(ring->dev, dma)) {
2392			netdev_err_once(ring->netdev,
2393					"Failed to map DMA for TX\n");
2394			goto unmap;
2395		}
2396
2397		dma_unmap_len_set(buffer, len, len);
2398		dma_unmap_addr_set(buffer, dma, dma);
2399
2400		cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
2401			   IGC_ADVTXD_DCMD_IFCS | len;
2402
2403		desc->read.cmd_type_len = cpu_to_le32(cmd_type);
2404		desc->read.buffer_addr = cpu_to_le64(dma);
2405
2406		buffer->protocol = 0;
2407
2408		if (++index == ring->count)
2409			index = 0;
2410
2411		if (i == nr_frags)
2412			break;
2413
2414		buffer = &ring->tx_buffer_info[index];
2415		desc = IGC_TX_DESC(ring, index);
2416		desc->read.olinfo_status = 0;
2417
2418		data = skb_frag_address(&sinfo->frags[i]);
2419		len = skb_frag_size(&sinfo->frags[i]);
2420		i++;
2421	}
2422	desc->read.cmd_type_len |= cpu_to_le32(IGC_TXD_DCMD);
2423
2424	netdev_tx_sent_queue(txring_txq(ring), head->bytecount);
2425	/* set the timestamp */
2426	head->time_stamp = jiffies;
2427	/* set next_to_watch value indicating a packet is present */
2428	head->next_to_watch = desc;
2429	ring->next_to_use = index;
2430
2431	return 0;
2432
2433unmap:
2434	for (;;) {
2435		buffer = &ring->tx_buffer_info[index];
2436		if (dma_unmap_len(buffer, len))
2437			dma_unmap_page(ring->dev,
2438				       dma_unmap_addr(buffer, dma),
2439				       dma_unmap_len(buffer, len),
2440				       DMA_TO_DEVICE);
2441		dma_unmap_len_set(buffer, len, 0);
2442		if (buffer == head)
2443			break;
2444
2445		if (!index)
2446			index += ring->count;
2447		index--;
2448	}
2449
2450	return -ENOMEM;
2451}
2452
2453static struct igc_ring *igc_xdp_get_tx_ring(struct igc_adapter *adapter,
2454					    int cpu)
2455{
2456	int index = cpu;
2457
2458	if (unlikely(index < 0))
2459		index = 0;
2460
2461	while (index >= adapter->num_tx_queues)
2462		index -= adapter->num_tx_queues;
2463
2464	return adapter->tx_ring[index];
2465}
2466
2467static int igc_xdp_xmit_back(struct igc_adapter *adapter, struct xdp_buff *xdp)
2468{
2469	struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2470	int cpu = smp_processor_id();
2471	struct netdev_queue *nq;
2472	struct igc_ring *ring;
2473	int res;
2474
2475	if (unlikely(!xdpf))
2476		return -EFAULT;
2477
2478	ring = igc_xdp_get_tx_ring(adapter, cpu);
2479	nq = txring_txq(ring);
2480
2481	__netif_tx_lock(nq, cpu);
2482	/* Avoid transmit queue timeout since we share it with the slow path */
2483	txq_trans_cond_update(nq);
2484	res = igc_xdp_init_tx_descriptor(ring, xdpf);
2485	__netif_tx_unlock(nq);
2486	return res;
2487}
2488
2489/* This function assumes rcu_read_lock() is held by the caller. */
2490static int __igc_xdp_run_prog(struct igc_adapter *adapter,
2491			      struct bpf_prog *prog,
2492			      struct xdp_buff *xdp)
2493{
2494	u32 act = bpf_prog_run_xdp(prog, xdp);
2495
2496	switch (act) {
2497	case XDP_PASS:
2498		return IGC_XDP_PASS;
2499	case XDP_TX:
2500		if (igc_xdp_xmit_back(adapter, xdp) < 0)
2501			goto out_failure;
2502		return IGC_XDP_TX;
2503	case XDP_REDIRECT:
2504		if (xdp_do_redirect(adapter->netdev, xdp, prog) < 0)
2505			goto out_failure;
2506		return IGC_XDP_REDIRECT;
2507		break;
2508	default:
2509		bpf_warn_invalid_xdp_action(adapter->netdev, prog, act);
2510		fallthrough;
2511	case XDP_ABORTED:
2512out_failure:
2513		trace_xdp_exception(adapter->netdev, prog, act);
2514		fallthrough;
2515	case XDP_DROP:
2516		return IGC_XDP_CONSUMED;
2517	}
2518}
2519
2520static struct sk_buff *igc_xdp_run_prog(struct igc_adapter *adapter,
2521					struct xdp_buff *xdp)
2522{
2523	struct bpf_prog *prog;
2524	int res;
2525
2526	prog = READ_ONCE(adapter->xdp_prog);
2527	if (!prog) {
2528		res = IGC_XDP_PASS;
2529		goto out;
2530	}
2531
2532	res = __igc_xdp_run_prog(adapter, prog, xdp);
2533
2534out:
2535	return ERR_PTR(-res);
2536}
2537
2538/* This function assumes __netif_tx_lock is held by the caller. */
2539static void igc_flush_tx_descriptors(struct igc_ring *ring)
2540{
2541	/* Once tail pointer is updated, hardware can fetch the descriptors
2542	 * any time so we issue a write membar here to ensure all memory
2543	 * writes are complete before the tail pointer is updated.
2544	 */
2545	wmb();
2546	writel(ring->next_to_use, ring->tail);
2547}
2548
2549static void igc_finalize_xdp(struct igc_adapter *adapter, int status)
2550{
2551	int cpu = smp_processor_id();
2552	struct netdev_queue *nq;
2553	struct igc_ring *ring;
2554
2555	if (status & IGC_XDP_TX) {
2556		ring = igc_xdp_get_tx_ring(adapter, cpu);
2557		nq = txring_txq(ring);
2558
2559		__netif_tx_lock(nq, cpu);
2560		igc_flush_tx_descriptors(ring);
2561		__netif_tx_unlock(nq);
2562	}
2563
2564	if (status & IGC_XDP_REDIRECT)
2565		xdp_do_flush();
2566}
2567
2568static void igc_update_rx_stats(struct igc_q_vector *q_vector,
2569				unsigned int packets, unsigned int bytes)
2570{
2571	struct igc_ring *ring = q_vector->rx.ring;
2572
2573	u64_stats_update_begin(&ring->rx_syncp);
2574	ring->rx_stats.packets += packets;
2575	ring->rx_stats.bytes += bytes;
2576	u64_stats_update_end(&ring->rx_syncp);
2577
2578	q_vector->rx.total_packets += packets;
2579	q_vector->rx.total_bytes += bytes;
2580}
2581
2582static int igc_clean_rx_irq(struct igc_q_vector *q_vector, const int budget)
2583{
2584	unsigned int total_bytes = 0, total_packets = 0;
2585	struct igc_adapter *adapter = q_vector->adapter;
2586	struct igc_ring *rx_ring = q_vector->rx.ring;
2587	struct sk_buff *skb = rx_ring->skb;
2588	u16 cleaned_count = igc_desc_unused(rx_ring);
2589	int xdp_status = 0, rx_buffer_pgcnt;
2590
2591	while (likely(total_packets < budget)) {
2592		struct igc_xdp_buff ctx = { .rx_ts = NULL };
2593		struct igc_rx_buffer *rx_buffer;
2594		union igc_adv_rx_desc *rx_desc;
2595		unsigned int size, truesize;
2596		int pkt_offset = 0;
2597		void *pktbuf;
2598
2599		/* return some buffers to hardware, one at a time is too slow */
2600		if (cleaned_count >= IGC_RX_BUFFER_WRITE) {
2601			igc_alloc_rx_buffers(rx_ring, cleaned_count);
2602			cleaned_count = 0;
2603		}
2604
2605		rx_desc = IGC_RX_DESC(rx_ring, rx_ring->next_to_clean);
2606		size = le16_to_cpu(rx_desc->wb.upper.length);
2607		if (!size)
2608			break;
2609
2610		/* This memory barrier is needed to keep us from reading
2611		 * any other fields out of the rx_desc until we know the
2612		 * descriptor has been written back
2613		 */
2614		dma_rmb();
2615
2616		rx_buffer = igc_get_rx_buffer(rx_ring, size, &rx_buffer_pgcnt);
2617		truesize = igc_get_rx_frame_truesize(rx_ring, size);
2618
2619		pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset;
2620
2621		if (igc_test_staterr(rx_desc, IGC_RXDADV_STAT_TSIP)) {
2622			ctx.rx_ts = pktbuf;
2623			pkt_offset = IGC_TS_HDR_LEN;
2624			size -= IGC_TS_HDR_LEN;
2625		}
2626
2627		if (!skb) {
2628			xdp_init_buff(&ctx.xdp, truesize, &rx_ring->xdp_rxq);
2629			xdp_prepare_buff(&ctx.xdp, pktbuf - igc_rx_offset(rx_ring),
2630					 igc_rx_offset(rx_ring) + pkt_offset,
2631					 size, true);
2632			xdp_buff_clear_frags_flag(&ctx.xdp);
2633			ctx.rx_desc = rx_desc;
2634
2635			skb = igc_xdp_run_prog(adapter, &ctx.xdp);
2636		}
2637
2638		if (IS_ERR(skb)) {
2639			unsigned int xdp_res = -PTR_ERR(skb);
2640
2641			switch (xdp_res) {
2642			case IGC_XDP_CONSUMED:
2643				rx_buffer->pagecnt_bias++;
2644				break;
2645			case IGC_XDP_TX:
2646			case IGC_XDP_REDIRECT:
2647				igc_rx_buffer_flip(rx_buffer, truesize);
2648				xdp_status |= xdp_res;
2649				break;
2650			}
2651
2652			total_packets++;
2653			total_bytes += size;
2654		} else if (skb)
2655			igc_add_rx_frag(rx_ring, rx_buffer, skb, size);
2656		else if (ring_uses_build_skb(rx_ring))
2657			skb = igc_build_skb(rx_ring, rx_buffer, &ctx.xdp);
2658		else
2659			skb = igc_construct_skb(rx_ring, rx_buffer, &ctx);
 
2660
2661		/* exit if we failed to retrieve a buffer */
2662		if (!skb) {
2663			rx_ring->rx_stats.alloc_failed++;
2664			rx_buffer->pagecnt_bias++;
2665			break;
2666		}
2667
2668		igc_put_rx_buffer(rx_ring, rx_buffer, rx_buffer_pgcnt);
2669		cleaned_count++;
2670
2671		/* fetch next buffer in frame if non-eop */
2672		if (igc_is_non_eop(rx_ring, rx_desc))
2673			continue;
2674
2675		/* verify the packet layout is correct */
2676		if (igc_cleanup_headers(rx_ring, rx_desc, skb)) {
2677			skb = NULL;
2678			continue;
2679		}
2680
2681		/* probably a little skewed due to removing CRC */
2682		total_bytes += skb->len;
2683
2684		/* populate checksum, VLAN, and protocol */
2685		igc_process_skb_fields(rx_ring, rx_desc, skb);
2686
2687		napi_gro_receive(&q_vector->napi, skb);
2688
2689		/* reset skb pointer */
2690		skb = NULL;
2691
2692		/* update budget accounting */
2693		total_packets++;
2694	}
2695
2696	if (xdp_status)
2697		igc_finalize_xdp(adapter, xdp_status);
2698
2699	/* place incomplete frames back on ring for completion */
2700	rx_ring->skb = skb;
2701
2702	igc_update_rx_stats(q_vector, total_packets, total_bytes);
 
 
 
 
 
2703
2704	if (cleaned_count)
2705		igc_alloc_rx_buffers(rx_ring, cleaned_count);
2706
2707	return total_packets;
2708}
2709
2710static struct sk_buff *igc_construct_skb_zc(struct igc_ring *ring,
2711					    struct xdp_buff *xdp)
2712{
2713	unsigned int totalsize = xdp->data_end - xdp->data_meta;
2714	unsigned int metasize = xdp->data - xdp->data_meta;
2715	struct sk_buff *skb;
2716
2717	net_prefetch(xdp->data_meta);
2718
2719	skb = __napi_alloc_skb(&ring->q_vector->napi, totalsize,
2720			       GFP_ATOMIC | __GFP_NOWARN);
2721	if (unlikely(!skb))
2722		return NULL;
2723
2724	memcpy(__skb_put(skb, totalsize), xdp->data_meta,
2725	       ALIGN(totalsize, sizeof(long)));
2726
2727	if (metasize) {
2728		skb_metadata_set(skb, metasize);
2729		__skb_pull(skb, metasize);
2730	}
2731
2732	return skb;
2733}
2734
2735static void igc_dispatch_skb_zc(struct igc_q_vector *q_vector,
2736				union igc_adv_rx_desc *desc,
2737				struct xdp_buff *xdp,
2738				ktime_t timestamp)
2739{
2740	struct igc_ring *ring = q_vector->rx.ring;
2741	struct sk_buff *skb;
2742
2743	skb = igc_construct_skb_zc(ring, xdp);
2744	if (!skb) {
2745		ring->rx_stats.alloc_failed++;
2746		return;
2747	}
2748
2749	if (timestamp)
2750		skb_hwtstamps(skb)->hwtstamp = timestamp;
2751
2752	if (igc_cleanup_headers(ring, desc, skb))
2753		return;
2754
2755	igc_process_skb_fields(ring, desc, skb);
2756	napi_gro_receive(&q_vector->napi, skb);
2757}
2758
2759static struct igc_xdp_buff *xsk_buff_to_igc_ctx(struct xdp_buff *xdp)
2760{
2761	/* xdp_buff pointer used by ZC code path is alloc as xdp_buff_xsk. The
2762	 * igc_xdp_buff shares its layout with xdp_buff_xsk and private
2763	 * igc_xdp_buff fields fall into xdp_buff_xsk->cb
2764	 */
2765       return (struct igc_xdp_buff *)xdp;
2766}
2767
2768static int igc_clean_rx_irq_zc(struct igc_q_vector *q_vector, const int budget)
2769{
2770	struct igc_adapter *adapter = q_vector->adapter;
2771	struct igc_ring *ring = q_vector->rx.ring;
2772	u16 cleaned_count = igc_desc_unused(ring);
2773	int total_bytes = 0, total_packets = 0;
2774	u16 ntc = ring->next_to_clean;
2775	struct bpf_prog *prog;
2776	bool failure = false;
2777	int xdp_status = 0;
2778
2779	rcu_read_lock();
2780
2781	prog = READ_ONCE(adapter->xdp_prog);
2782
2783	while (likely(total_packets < budget)) {
2784		union igc_adv_rx_desc *desc;
2785		struct igc_rx_buffer *bi;
2786		struct igc_xdp_buff *ctx;
2787		ktime_t timestamp = 0;
2788		unsigned int size;
2789		int res;
2790
2791		desc = IGC_RX_DESC(ring, ntc);
2792		size = le16_to_cpu(desc->wb.upper.length);
2793		if (!size)
2794			break;
2795
2796		/* This memory barrier is needed to keep us from reading
2797		 * any other fields out of the rx_desc until we know the
2798		 * descriptor has been written back
2799		 */
2800		dma_rmb();
2801
2802		bi = &ring->rx_buffer_info[ntc];
2803
2804		ctx = xsk_buff_to_igc_ctx(bi->xdp);
2805		ctx->rx_desc = desc;
2806
2807		if (igc_test_staterr(desc, IGC_RXDADV_STAT_TSIP)) {
2808			ctx->rx_ts = bi->xdp->data;
2809
2810			bi->xdp->data += IGC_TS_HDR_LEN;
2811
2812			/* HW timestamp has been copied into local variable. Metadata
2813			 * length when XDP program is called should be 0.
2814			 */
2815			bi->xdp->data_meta += IGC_TS_HDR_LEN;
2816			size -= IGC_TS_HDR_LEN;
2817		}
2818
2819		bi->xdp->data_end = bi->xdp->data + size;
2820		xsk_buff_dma_sync_for_cpu(bi->xdp, ring->xsk_pool);
2821
2822		res = __igc_xdp_run_prog(adapter, prog, bi->xdp);
2823		switch (res) {
2824		case IGC_XDP_PASS:
2825			igc_dispatch_skb_zc(q_vector, desc, bi->xdp, timestamp);
2826			fallthrough;
2827		case IGC_XDP_CONSUMED:
2828			xsk_buff_free(bi->xdp);
2829			break;
2830		case IGC_XDP_TX:
2831		case IGC_XDP_REDIRECT:
2832			xdp_status |= res;
2833			break;
2834		}
2835
2836		bi->xdp = NULL;
2837		total_bytes += size;
2838		total_packets++;
2839		cleaned_count++;
2840		ntc++;
2841		if (ntc == ring->count)
2842			ntc = 0;
2843	}
2844
2845	ring->next_to_clean = ntc;
2846	rcu_read_unlock();
2847
2848	if (cleaned_count >= IGC_RX_BUFFER_WRITE)
2849		failure = !igc_alloc_rx_buffers_zc(ring, cleaned_count);
2850
2851	if (xdp_status)
2852		igc_finalize_xdp(adapter, xdp_status);
2853
2854	igc_update_rx_stats(q_vector, total_packets, total_bytes);
2855
2856	if (xsk_uses_need_wakeup(ring->xsk_pool)) {
2857		if (failure || ring->next_to_clean == ring->next_to_use)
2858			xsk_set_rx_need_wakeup(ring->xsk_pool);
2859		else
2860			xsk_clear_rx_need_wakeup(ring->xsk_pool);
2861		return total_packets;
2862	}
2863
2864	return failure ? budget : total_packets;
2865}
2866
2867static void igc_update_tx_stats(struct igc_q_vector *q_vector,
2868				unsigned int packets, unsigned int bytes)
2869{
2870	struct igc_ring *ring = q_vector->tx.ring;
2871
2872	u64_stats_update_begin(&ring->tx_syncp);
2873	ring->tx_stats.bytes += bytes;
2874	ring->tx_stats.packets += packets;
2875	u64_stats_update_end(&ring->tx_syncp);
2876
2877	q_vector->tx.total_bytes += bytes;
2878	q_vector->tx.total_packets += packets;
2879}
2880
2881static void igc_xdp_xmit_zc(struct igc_ring *ring)
2882{
2883	struct xsk_buff_pool *pool = ring->xsk_pool;
2884	struct netdev_queue *nq = txring_txq(ring);
2885	union igc_adv_tx_desc *tx_desc = NULL;
2886	int cpu = smp_processor_id();
2887	struct xdp_desc xdp_desc;
2888	u16 budget, ntu;
2889
2890	if (!netif_carrier_ok(ring->netdev))
2891		return;
2892
2893	__netif_tx_lock(nq, cpu);
2894
2895	/* Avoid transmit queue timeout since we share it with the slow path */
2896	txq_trans_cond_update(nq);
2897
2898	ntu = ring->next_to_use;
2899	budget = igc_desc_unused(ring);
2900
2901	while (xsk_tx_peek_desc(pool, &xdp_desc) && budget--) {
2902		u32 cmd_type, olinfo_status;
2903		struct igc_tx_buffer *bi;
2904		dma_addr_t dma;
2905
2906		cmd_type = IGC_ADVTXD_DTYP_DATA | IGC_ADVTXD_DCMD_DEXT |
2907			   IGC_ADVTXD_DCMD_IFCS | IGC_TXD_DCMD |
2908			   xdp_desc.len;
2909		olinfo_status = xdp_desc.len << IGC_ADVTXD_PAYLEN_SHIFT;
2910
2911		dma = xsk_buff_raw_get_dma(pool, xdp_desc.addr);
2912		xsk_buff_raw_dma_sync_for_device(pool, dma, xdp_desc.len);
2913
2914		tx_desc = IGC_TX_DESC(ring, ntu);
2915		tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
2916		tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2917		tx_desc->read.buffer_addr = cpu_to_le64(dma);
2918
2919		bi = &ring->tx_buffer_info[ntu];
2920		bi->type = IGC_TX_BUFFER_TYPE_XSK;
2921		bi->protocol = 0;
2922		bi->bytecount = xdp_desc.len;
2923		bi->gso_segs = 1;
2924		bi->time_stamp = jiffies;
2925		bi->next_to_watch = tx_desc;
2926
2927		netdev_tx_sent_queue(txring_txq(ring), xdp_desc.len);
2928
2929		ntu++;
2930		if (ntu == ring->count)
2931			ntu = 0;
2932	}
2933
2934	ring->next_to_use = ntu;
2935	if (tx_desc) {
2936		igc_flush_tx_descriptors(ring);
2937		xsk_tx_release(pool);
2938	}
2939
2940	__netif_tx_unlock(nq);
2941}
2942
2943/**
2944 * igc_clean_tx_irq - Reclaim resources after transmit completes
2945 * @q_vector: pointer to q_vector containing needed info
2946 * @napi_budget: Used to determine if we are in netpoll
2947 *
2948 * returns true if ring is completely cleaned
2949 */
2950static bool igc_clean_tx_irq(struct igc_q_vector *q_vector, int napi_budget)
2951{
2952	struct igc_adapter *adapter = q_vector->adapter;
2953	unsigned int total_bytes = 0, total_packets = 0;
2954	unsigned int budget = q_vector->tx.work_limit;
2955	struct igc_ring *tx_ring = q_vector->tx.ring;
2956	unsigned int i = tx_ring->next_to_clean;
2957	struct igc_tx_buffer *tx_buffer;
2958	union igc_adv_tx_desc *tx_desc;
2959	u32 xsk_frames = 0;
2960
2961	if (test_bit(__IGC_DOWN, &adapter->state))
2962		return true;
2963
2964	tx_buffer = &tx_ring->tx_buffer_info[i];
2965	tx_desc = IGC_TX_DESC(tx_ring, i);
2966	i -= tx_ring->count;
2967
2968	do {
2969		union igc_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
2970
2971		/* if next_to_watch is not set then there is no work pending */
2972		if (!eop_desc)
2973			break;
2974
2975		/* prevent any other reads prior to eop_desc */
2976		smp_rmb();
2977
2978		/* if DD is not set pending work has not been completed */
2979		if (!(eop_desc->wb.status & cpu_to_le32(IGC_TXD_STAT_DD)))
2980			break;
2981
2982		/* clear next_to_watch to prevent false hangs */
2983		tx_buffer->next_to_watch = NULL;
2984
2985		/* update the statistics for this packet */
2986		total_bytes += tx_buffer->bytecount;
2987		total_packets += tx_buffer->gso_segs;
2988
2989		switch (tx_buffer->type) {
2990		case IGC_TX_BUFFER_TYPE_XSK:
2991			xsk_frames++;
2992			break;
2993		case IGC_TX_BUFFER_TYPE_XDP:
2994			xdp_return_frame(tx_buffer->xdpf);
2995			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
2996			break;
2997		case IGC_TX_BUFFER_TYPE_SKB:
2998			napi_consume_skb(tx_buffer->skb, napi_budget);
2999			igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
3000			break;
3001		default:
3002			netdev_warn_once(tx_ring->netdev, "Unknown Tx buffer type\n");
3003			break;
3004		}
3005
3006		/* clear last DMA location and unmap remaining buffers */
3007		while (tx_desc != eop_desc) {
3008			tx_buffer++;
3009			tx_desc++;
3010			i++;
3011			if (unlikely(!i)) {
3012				i -= tx_ring->count;
3013				tx_buffer = tx_ring->tx_buffer_info;
3014				tx_desc = IGC_TX_DESC(tx_ring, 0);
3015			}
3016
3017			/* unmap any remaining paged data */
3018			if (dma_unmap_len(tx_buffer, len))
3019				igc_unmap_tx_buffer(tx_ring->dev, tx_buffer);
 
 
 
 
 
3020		}
3021
3022		/* move us one more past the eop_desc for start of next pkt */
3023		tx_buffer++;
3024		tx_desc++;
3025		i++;
3026		if (unlikely(!i)) {
3027			i -= tx_ring->count;
3028			tx_buffer = tx_ring->tx_buffer_info;
3029			tx_desc = IGC_TX_DESC(tx_ring, 0);
3030		}
3031
3032		/* issue prefetch for next Tx descriptor */
3033		prefetch(tx_desc);
3034
3035		/* update budget accounting */
3036		budget--;
3037	} while (likely(budget));
3038
3039	netdev_tx_completed_queue(txring_txq(tx_ring),
3040				  total_packets, total_bytes);
3041
3042	i += tx_ring->count;
3043	tx_ring->next_to_clean = i;
3044
3045	igc_update_tx_stats(q_vector, total_packets, total_bytes);
3046
3047	if (tx_ring->xsk_pool) {
3048		if (xsk_frames)
3049			xsk_tx_completed(tx_ring->xsk_pool, xsk_frames);
3050		if (xsk_uses_need_wakeup(tx_ring->xsk_pool))
3051			xsk_set_tx_need_wakeup(tx_ring->xsk_pool);
3052		igc_xdp_xmit_zc(tx_ring);
3053	}
3054
3055	if (test_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
3056		struct igc_hw *hw = &adapter->hw;
3057
3058		/* Detect a transmit hang in hardware, this serializes the
3059		 * check with the clearing of time_stamp and movement of i
3060		 */
3061		clear_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
3062		if (tx_buffer->next_to_watch &&
3063		    time_after(jiffies, tx_buffer->time_stamp +
3064		    (adapter->tx_timeout_factor * HZ)) &&
3065		    !(rd32(IGC_STATUS) & IGC_STATUS_TXOFF) &&
3066		    (rd32(IGC_TDH(tx_ring->reg_idx)) != readl(tx_ring->tail)) &&
3067		    !tx_ring->oper_gate_closed) {
3068			/* detected Tx unit hang */
3069			netdev_err(tx_ring->netdev,
3070				   "Detected Tx Unit Hang\n"
3071				   "  Tx Queue             <%d>\n"
3072				   "  TDH                  <%x>\n"
3073				   "  TDT                  <%x>\n"
3074				   "  next_to_use          <%x>\n"
3075				   "  next_to_clean        <%x>\n"
3076				   "buffer_info[next_to_clean]\n"
3077				   "  time_stamp           <%lx>\n"
3078				   "  next_to_watch        <%p>\n"
3079				   "  jiffies              <%lx>\n"
3080				   "  desc.status          <%x>\n",
3081				   tx_ring->queue_index,
3082				   rd32(IGC_TDH(tx_ring->reg_idx)),
3083				   readl(tx_ring->tail),
3084				   tx_ring->next_to_use,
3085				   tx_ring->next_to_clean,
3086				   tx_buffer->time_stamp,
3087				   tx_buffer->next_to_watch,
3088				   jiffies,
3089				   tx_buffer->next_to_watch->wb.status);
3090			netif_stop_subqueue(tx_ring->netdev,
3091					    tx_ring->queue_index);
3092
3093			/* we are about to reset, no point in enabling stuff */
3094			return true;
3095		}
3096	}
3097
3098#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
3099	if (unlikely(total_packets &&
3100		     netif_carrier_ok(tx_ring->netdev) &&
3101		     igc_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
3102		/* Make sure that anybody stopping the queue after this
3103		 * sees the new next_to_clean.
3104		 */
3105		smp_mb();
3106		if (__netif_subqueue_stopped(tx_ring->netdev,
3107					     tx_ring->queue_index) &&
3108		    !(test_bit(__IGC_DOWN, &adapter->state))) {
3109			netif_wake_subqueue(tx_ring->netdev,
3110					    tx_ring->queue_index);
3111
3112			u64_stats_update_begin(&tx_ring->tx_syncp);
3113			tx_ring->tx_stats.restart_queue++;
3114			u64_stats_update_end(&tx_ring->tx_syncp);
3115		}
3116	}
3117
3118	return !!budget;
3119}
3120
3121static int igc_find_mac_filter(struct igc_adapter *adapter,
3122			       enum igc_mac_filter_type type, const u8 *addr)
3123{
3124	struct igc_hw *hw = &adapter->hw;
3125	int max_entries = hw->mac.rar_entry_count;
3126	u32 ral, rah;
3127	int i;
3128
3129	for (i = 0; i < max_entries; i++) {
3130		ral = rd32(IGC_RAL(i));
3131		rah = rd32(IGC_RAH(i));
3132
3133		if (!(rah & IGC_RAH_AV))
3134			continue;
3135		if (!!(rah & IGC_RAH_ASEL_SRC_ADDR) != type)
3136			continue;
3137		if ((rah & IGC_RAH_RAH_MASK) !=
3138		    le16_to_cpup((__le16 *)(addr + 4)))
3139			continue;
3140		if (ral != le32_to_cpup((__le32 *)(addr)))
3141			continue;
3142
3143		return i;
3144	}
3145
3146	return -1;
3147}
3148
3149static int igc_get_avail_mac_filter_slot(struct igc_adapter *adapter)
3150{
3151	struct igc_hw *hw = &adapter->hw;
3152	int max_entries = hw->mac.rar_entry_count;
3153	u32 rah;
3154	int i;
3155
3156	for (i = 0; i < max_entries; i++) {
3157		rah = rd32(IGC_RAH(i));
3158
3159		if (!(rah & IGC_RAH_AV))
3160			return i;
3161	}
3162
3163	return -1;
3164}
3165
3166/**
3167 * igc_add_mac_filter() - Add MAC address filter
3168 * @adapter: Pointer to adapter where the filter should be added
3169 * @type: MAC address filter type (source or destination)
3170 * @addr: MAC address
3171 * @queue: If non-negative, queue assignment feature is enabled and frames
3172 *         matching the filter are enqueued onto 'queue'. Otherwise, queue
3173 *         assignment is disabled.
3174 *
3175 * Return: 0 in case of success, negative errno code otherwise.
3176 */
3177static int igc_add_mac_filter(struct igc_adapter *adapter,
3178			      enum igc_mac_filter_type type, const u8 *addr,
3179			      int queue)
3180{
3181	struct net_device *dev = adapter->netdev;
3182	int index;
3183
3184	index = igc_find_mac_filter(adapter, type, addr);
3185	if (index >= 0)
3186		goto update_filter;
3187
3188	index = igc_get_avail_mac_filter_slot(adapter);
3189	if (index < 0)
3190		return -ENOSPC;
3191
3192	netdev_dbg(dev, "Add MAC address filter: index %d type %s address %pM queue %d\n",
3193		   index, type == IGC_MAC_FILTER_TYPE_DST ? "dst" : "src",
3194		   addr, queue);
3195
3196update_filter:
3197	igc_set_mac_filter_hw(adapter, index, type, addr, queue);
3198	return 0;
3199}
3200
3201/**
3202 * igc_del_mac_filter() - Delete MAC address filter
3203 * @adapter: Pointer to adapter where the filter should be deleted from
3204 * @type: MAC address filter type (source or destination)
3205 * @addr: MAC address
3206 */
3207static void igc_del_mac_filter(struct igc_adapter *adapter,
3208			       enum igc_mac_filter_type type, const u8 *addr)
3209{
3210	struct net_device *dev = adapter->netdev;
3211	int index;
3212
3213	index = igc_find_mac_filter(adapter, type, addr);
3214	if (index < 0)
3215		return;
3216
3217	if (index == 0) {
3218		/* If this is the default filter, we don't actually delete it.
3219		 * We just reset to its default value i.e. disable queue
3220		 * assignment.
3221		 */
3222		netdev_dbg(dev, "Disable default MAC filter queue assignment");
3223
3224		igc_set_mac_filter_hw(adapter, 0, type, addr, -1);
3225	} else {
3226		netdev_dbg(dev, "Delete MAC address filter: index %d type %s address %pM\n",
3227			   index,
3228			   type == IGC_MAC_FILTER_TYPE_DST ? "dst" : "src",
3229			   addr);
3230
3231		igc_clear_mac_filter_hw(adapter, index);
3232	}
3233}
3234
3235/**
3236 * igc_add_vlan_prio_filter() - Add VLAN priority filter
3237 * @adapter: Pointer to adapter where the filter should be added
3238 * @prio: VLAN priority value
3239 * @queue: Queue number which matching frames are assigned to
3240 *
3241 * Return: 0 in case of success, negative errno code otherwise.
3242 */
3243static int igc_add_vlan_prio_filter(struct igc_adapter *adapter, int prio,
3244				    int queue)
3245{
3246	struct net_device *dev = adapter->netdev;
3247	struct igc_hw *hw = &adapter->hw;
3248	u32 vlanpqf;
3249
3250	vlanpqf = rd32(IGC_VLANPQF);
3251
3252	if (vlanpqf & IGC_VLANPQF_VALID(prio)) {
3253		netdev_dbg(dev, "VLAN priority filter already in use\n");
3254		return -EEXIST;
3255	}
3256
3257	vlanpqf |= IGC_VLANPQF_QSEL(prio, queue);
3258	vlanpqf |= IGC_VLANPQF_VALID(prio);
3259
3260	wr32(IGC_VLANPQF, vlanpqf);
3261
3262	netdev_dbg(dev, "Add VLAN priority filter: prio %d queue %d\n",
3263		   prio, queue);
3264	return 0;
3265}
3266
3267/**
3268 * igc_del_vlan_prio_filter() - Delete VLAN priority filter
3269 * @adapter: Pointer to adapter where the filter should be deleted from
3270 * @prio: VLAN priority value
3271 */
3272static void igc_del_vlan_prio_filter(struct igc_adapter *adapter, int prio)
3273{
3274	struct igc_hw *hw = &adapter->hw;
3275	u32 vlanpqf;
3276
3277	vlanpqf = rd32(IGC_VLANPQF);
3278
3279	vlanpqf &= ~IGC_VLANPQF_VALID(prio);
3280	vlanpqf &= ~IGC_VLANPQF_QSEL(prio, IGC_VLANPQF_QUEUE_MASK);
3281
3282	wr32(IGC_VLANPQF, vlanpqf);
3283
3284	netdev_dbg(adapter->netdev, "Delete VLAN priority filter: prio %d\n",
3285		   prio);
3286}
3287
3288static int igc_get_avail_etype_filter_slot(struct igc_adapter *adapter)
3289{
3290	struct igc_hw *hw = &adapter->hw;
3291	int i;
3292
3293	for (i = 0; i < MAX_ETYPE_FILTER; i++) {
3294		u32 etqf = rd32(IGC_ETQF(i));
3295
3296		if (!(etqf & IGC_ETQF_FILTER_ENABLE))
3297			return i;
3298	}
3299
3300	return -1;
3301}
3302
3303/**
3304 * igc_add_etype_filter() - Add ethertype filter
3305 * @adapter: Pointer to adapter where the filter should be added
3306 * @etype: Ethertype value
3307 * @queue: If non-negative, queue assignment feature is enabled and frames
3308 *         matching the filter are enqueued onto 'queue'. Otherwise, queue
3309 *         assignment is disabled.
3310 *
3311 * Return: 0 in case of success, negative errno code otherwise.
3312 */
3313static int igc_add_etype_filter(struct igc_adapter *adapter, u16 etype,
3314				int queue)
3315{
3316	struct igc_hw *hw = &adapter->hw;
3317	int index;
3318	u32 etqf;
3319
3320	index = igc_get_avail_etype_filter_slot(adapter);
3321	if (index < 0)
3322		return -ENOSPC;
3323
3324	etqf = rd32(IGC_ETQF(index));
3325
3326	etqf &= ~IGC_ETQF_ETYPE_MASK;
3327	etqf |= etype;
3328
3329	if (queue >= 0) {
3330		etqf &= ~IGC_ETQF_QUEUE_MASK;
3331		etqf |= (queue << IGC_ETQF_QUEUE_SHIFT);
3332		etqf |= IGC_ETQF_QUEUE_ENABLE;
3333	}
3334
3335	etqf |= IGC_ETQF_FILTER_ENABLE;
3336
3337	wr32(IGC_ETQF(index), etqf);
3338
3339	netdev_dbg(adapter->netdev, "Add ethertype filter: etype %04x queue %d\n",
3340		   etype, queue);
3341	return 0;
3342}
3343
3344static int igc_find_etype_filter(struct igc_adapter *adapter, u16 etype)
3345{
3346	struct igc_hw *hw = &adapter->hw;
3347	int i;
3348
3349	for (i = 0; i < MAX_ETYPE_FILTER; i++) {
3350		u32 etqf = rd32(IGC_ETQF(i));
3351
3352		if ((etqf & IGC_ETQF_ETYPE_MASK) == etype)
3353			return i;
3354	}
3355
3356	return -1;
3357}
3358
3359/**
3360 * igc_del_etype_filter() - Delete ethertype filter
3361 * @adapter: Pointer to adapter where the filter should be deleted from
3362 * @etype: Ethertype value
3363 */
3364static void igc_del_etype_filter(struct igc_adapter *adapter, u16 etype)
3365{
3366	struct igc_hw *hw = &adapter->hw;
3367	int index;
3368
3369	index = igc_find_etype_filter(adapter, etype);
3370	if (index < 0)
3371		return;
3372
3373	wr32(IGC_ETQF(index), 0);
3374
3375	netdev_dbg(adapter->netdev, "Delete ethertype filter: etype %04x\n",
3376		   etype);
3377}
3378
3379static int igc_flex_filter_select(struct igc_adapter *adapter,
3380				  struct igc_flex_filter *input,
3381				  u32 *fhft)
3382{
3383	struct igc_hw *hw = &adapter->hw;
3384	u8 fhft_index;
3385	u32 fhftsl;
3386
3387	if (input->index >= MAX_FLEX_FILTER) {
3388		dev_err(&adapter->pdev->dev, "Wrong Flex Filter index selected!\n");
3389		return -EINVAL;
3390	}
3391
3392	/* Indirect table select register */
3393	fhftsl = rd32(IGC_FHFTSL);
3394	fhftsl &= ~IGC_FHFTSL_FTSL_MASK;
3395	switch (input->index) {
3396	case 0 ... 7:
3397		fhftsl |= 0x00;
3398		break;
3399	case 8 ... 15:
3400		fhftsl |= 0x01;
3401		break;
3402	case 16 ... 23:
3403		fhftsl |= 0x02;
3404		break;
3405	case 24 ... 31:
3406		fhftsl |= 0x03;
3407		break;
3408	}
3409	wr32(IGC_FHFTSL, fhftsl);
3410
3411	/* Normalize index down to host table register */
3412	fhft_index = input->index % 8;
3413
3414	*fhft = (fhft_index < 4) ? IGC_FHFT(fhft_index) :
3415		IGC_FHFT_EXT(fhft_index - 4);
3416
3417	return 0;
3418}
3419
3420static int igc_write_flex_filter_ll(struct igc_adapter *adapter,
3421				    struct igc_flex_filter *input)
3422{
3423	struct device *dev = &adapter->pdev->dev;
3424	struct igc_hw *hw = &adapter->hw;
3425	u8 *data = input->data;
3426	u8 *mask = input->mask;
3427	u32 queuing;
3428	u32 fhft;
3429	u32 wufc;
3430	int ret;
3431	int i;
3432
3433	/* Length has to be aligned to 8. Otherwise the filter will fail. Bail
3434	 * out early to avoid surprises later.
3435	 */
3436	if (input->length % 8 != 0) {
3437		dev_err(dev, "The length of a flex filter has to be 8 byte aligned!\n");
3438		return -EINVAL;
3439	}
3440
3441	/* Select corresponding flex filter register and get base for host table. */
3442	ret = igc_flex_filter_select(adapter, input, &fhft);
3443	if (ret)
3444		return ret;
3445
3446	/* When adding a filter globally disable flex filter feature. That is
3447	 * recommended within the datasheet.
3448	 */
3449	wufc = rd32(IGC_WUFC);
3450	wufc &= ~IGC_WUFC_FLEX_HQ;
3451	wr32(IGC_WUFC, wufc);
3452
3453	/* Configure filter */
3454	queuing = input->length & IGC_FHFT_LENGTH_MASK;
3455	queuing |= FIELD_PREP(IGC_FHFT_QUEUE_MASK, input->rx_queue);
3456	queuing |= FIELD_PREP(IGC_FHFT_PRIO_MASK, input->prio);
3457
3458	if (input->immediate_irq)
3459		queuing |= IGC_FHFT_IMM_INT;
3460
3461	if (input->drop)
3462		queuing |= IGC_FHFT_DROP;
3463
3464	wr32(fhft + 0xFC, queuing);
3465
3466	/* Write data (128 byte) and mask (128 bit) */
3467	for (i = 0; i < 16; ++i) {
3468		const size_t data_idx = i * 8;
3469		const size_t row_idx = i * 16;
3470		u32 dw0 =
3471			(data[data_idx + 0] << 0) |
3472			(data[data_idx + 1] << 8) |
3473			(data[data_idx + 2] << 16) |
3474			(data[data_idx + 3] << 24);
3475		u32 dw1 =
3476			(data[data_idx + 4] << 0) |
3477			(data[data_idx + 5] << 8) |
3478			(data[data_idx + 6] << 16) |
3479			(data[data_idx + 7] << 24);
3480		u32 tmp;
3481
3482		/* Write row: dw0, dw1 and mask */
3483		wr32(fhft + row_idx, dw0);
3484		wr32(fhft + row_idx + 4, dw1);
3485
3486		/* mask is only valid for MASK(7, 0) */
3487		tmp = rd32(fhft + row_idx + 8);
3488		tmp &= ~GENMASK(7, 0);
3489		tmp |= mask[i];
3490		wr32(fhft + row_idx + 8, tmp);
3491	}
3492
3493	/* Enable filter. */
3494	wufc |= IGC_WUFC_FLEX_HQ;
3495	if (input->index > 8) {
3496		/* Filter 0-7 are enabled via WUFC. The other 24 filters are not. */
3497		u32 wufc_ext = rd32(IGC_WUFC_EXT);
3498
3499		wufc_ext |= (IGC_WUFC_EXT_FLX8 << (input->index - 8));
3500
3501		wr32(IGC_WUFC_EXT, wufc_ext);
3502	} else {
3503		wufc |= (IGC_WUFC_FLX0 << input->index);
3504	}
3505	wr32(IGC_WUFC, wufc);
3506
3507	dev_dbg(&adapter->pdev->dev, "Added flex filter %u to HW.\n",
3508		input->index);
3509
3510	return 0;
3511}
3512
3513static void igc_flex_filter_add_field(struct igc_flex_filter *flex,
3514				      const void *src, unsigned int offset,
3515				      size_t len, const void *mask)
3516{
3517	int i;
3518
3519	/* data */
3520	memcpy(&flex->data[offset], src, len);
3521
3522	/* mask */
3523	for (i = 0; i < len; ++i) {
3524		const unsigned int idx = i + offset;
3525		const u8 *ptr = mask;
3526
3527		if (mask) {
3528			if (ptr[i] & 0xff)
3529				flex->mask[idx / 8] |= BIT(idx % 8);
3530
3531			continue;
3532		}
3533
3534		flex->mask[idx / 8] |= BIT(idx % 8);
3535	}
3536}
3537
3538static int igc_find_avail_flex_filter_slot(struct igc_adapter *adapter)
3539{
3540	struct igc_hw *hw = &adapter->hw;
3541	u32 wufc, wufc_ext;
3542	int i;
3543
3544	wufc = rd32(IGC_WUFC);
3545	wufc_ext = rd32(IGC_WUFC_EXT);
3546
3547	for (i = 0; i < MAX_FLEX_FILTER; i++) {
3548		if (i < 8) {
3549			if (!(wufc & (IGC_WUFC_FLX0 << i)))
3550				return i;
3551		} else {
3552			if (!(wufc_ext & (IGC_WUFC_EXT_FLX8 << (i - 8))))
3553				return i;
3554		}
3555	}
3556
3557	return -ENOSPC;
3558}
3559
3560static bool igc_flex_filter_in_use(struct igc_adapter *adapter)
3561{
3562	struct igc_hw *hw = &adapter->hw;
3563	u32 wufc, wufc_ext;
3564
3565	wufc = rd32(IGC_WUFC);
3566	wufc_ext = rd32(IGC_WUFC_EXT);
3567
3568	if (wufc & IGC_WUFC_FILTER_MASK)
3569		return true;
3570
3571	if (wufc_ext & IGC_WUFC_EXT_FILTER_MASK)
3572		return true;
3573
3574	return false;
3575}
3576
3577static int igc_add_flex_filter(struct igc_adapter *adapter,
3578			       struct igc_nfc_rule *rule)
3579{
3580	struct igc_flex_filter flex = { };
3581	struct igc_nfc_filter *filter = &rule->filter;
3582	unsigned int eth_offset, user_offset;
3583	int ret, index;
3584	bool vlan;
3585
3586	index = igc_find_avail_flex_filter_slot(adapter);
3587	if (index < 0)
3588		return -ENOSPC;
3589
3590	/* Construct the flex filter:
3591	 *  -> dest_mac [6]
3592	 *  -> src_mac [6]
3593	 *  -> tpid [2]
3594	 *  -> vlan tci [2]
3595	 *  -> ether type [2]
3596	 *  -> user data [8]
3597	 *  -> = 26 bytes => 32 length
3598	 */
3599	flex.index    = index;
3600	flex.length   = 32;
3601	flex.rx_queue = rule->action;
3602
3603	vlan = rule->filter.vlan_tci || rule->filter.vlan_etype;
3604	eth_offset = vlan ? 16 : 12;
3605	user_offset = vlan ? 18 : 14;
3606
3607	/* Add destination MAC  */
3608	if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR)
3609		igc_flex_filter_add_field(&flex, &filter->dst_addr, 0,
3610					  ETH_ALEN, NULL);
3611
3612	/* Add source MAC */
3613	if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR)
3614		igc_flex_filter_add_field(&flex, &filter->src_addr, 6,
3615					  ETH_ALEN, NULL);
3616
3617	/* Add VLAN etype */
3618	if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_ETYPE)
3619		igc_flex_filter_add_field(&flex, &filter->vlan_etype, 12,
3620					  sizeof(filter->vlan_etype),
3621					  NULL);
3622
3623	/* Add VLAN TCI */
3624	if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI)
3625		igc_flex_filter_add_field(&flex, &filter->vlan_tci, 14,
3626					  sizeof(filter->vlan_tci), NULL);
3627
3628	/* Add Ether type */
3629	if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE) {
3630		__be16 etype = cpu_to_be16(filter->etype);
3631
3632		igc_flex_filter_add_field(&flex, &etype, eth_offset,
3633					  sizeof(etype), NULL);
3634	}
3635
3636	/* Add user data */
3637	if (rule->filter.match_flags & IGC_FILTER_FLAG_USER_DATA)
3638		igc_flex_filter_add_field(&flex, &filter->user_data,
3639					  user_offset,
3640					  sizeof(filter->user_data),
3641					  filter->user_mask);
3642
3643	/* Add it down to the hardware and enable it. */
3644	ret = igc_write_flex_filter_ll(adapter, &flex);
3645	if (ret)
3646		return ret;
3647
3648	filter->flex_index = index;
3649
3650	return 0;
3651}
3652
3653static void igc_del_flex_filter(struct igc_adapter *adapter,
3654				u16 reg_index)
3655{
3656	struct igc_hw *hw = &adapter->hw;
3657	u32 wufc;
3658
3659	/* Just disable the filter. The filter table itself is kept
3660	 * intact. Another flex_filter_add() should override the "old" data
3661	 * then.
3662	 */
3663	if (reg_index > 8) {
3664		u32 wufc_ext = rd32(IGC_WUFC_EXT);
3665
3666		wufc_ext &= ~(IGC_WUFC_EXT_FLX8 << (reg_index - 8));
3667		wr32(IGC_WUFC_EXT, wufc_ext);
3668	} else {
3669		wufc = rd32(IGC_WUFC);
3670
3671		wufc &= ~(IGC_WUFC_FLX0 << reg_index);
3672		wr32(IGC_WUFC, wufc);
3673	}
3674
3675	if (igc_flex_filter_in_use(adapter))
3676		return;
3677
3678	/* No filters are in use, we may disable flex filters */
3679	wufc = rd32(IGC_WUFC);
3680	wufc &= ~IGC_WUFC_FLEX_HQ;
3681	wr32(IGC_WUFC, wufc);
3682}
3683
3684static int igc_enable_nfc_rule(struct igc_adapter *adapter,
3685			       struct igc_nfc_rule *rule)
3686{
3687	int err;
3688
3689	if (rule->flex) {
3690		return igc_add_flex_filter(adapter, rule);
3691	}
3692
3693	if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE) {
3694		err = igc_add_etype_filter(adapter, rule->filter.etype,
3695					   rule->action);
3696		if (err)
3697			return err;
3698	}
3699
3700	if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR) {
3701		err = igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_SRC,
3702					 rule->filter.src_addr, rule->action);
3703		if (err)
3704			return err;
3705	}
3706
3707	if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR) {
3708		err = igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST,
3709					 rule->filter.dst_addr, rule->action);
3710		if (err)
3711			return err;
3712	}
3713
3714	if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI) {
3715		int prio = FIELD_GET(VLAN_PRIO_MASK, rule->filter.vlan_tci);
 
3716
3717		err = igc_add_vlan_prio_filter(adapter, prio, rule->action);
3718		if (err)
3719			return err;
3720	}
3721
3722	return 0;
3723}
3724
3725static void igc_disable_nfc_rule(struct igc_adapter *adapter,
3726				 const struct igc_nfc_rule *rule)
3727{
3728	if (rule->flex) {
3729		igc_del_flex_filter(adapter, rule->filter.flex_index);
3730		return;
3731	}
3732
3733	if (rule->filter.match_flags & IGC_FILTER_FLAG_ETHER_TYPE)
3734		igc_del_etype_filter(adapter, rule->filter.etype);
3735
3736	if (rule->filter.match_flags & IGC_FILTER_FLAG_VLAN_TCI) {
3737		int prio = FIELD_GET(VLAN_PRIO_MASK, rule->filter.vlan_tci);
 
3738
3739		igc_del_vlan_prio_filter(adapter, prio);
3740	}
3741
3742	if (rule->filter.match_flags & IGC_FILTER_FLAG_SRC_MAC_ADDR)
3743		igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_SRC,
3744				   rule->filter.src_addr);
3745
3746	if (rule->filter.match_flags & IGC_FILTER_FLAG_DST_MAC_ADDR)
3747		igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST,
3748				   rule->filter.dst_addr);
3749}
3750
3751/**
3752 * igc_get_nfc_rule() - Get NFC rule
3753 * @adapter: Pointer to adapter
3754 * @location: Rule location
3755 *
3756 * Context: Expects adapter->nfc_rule_lock to be held by caller.
3757 *
3758 * Return: Pointer to NFC rule at @location. If not found, NULL.
3759 */
3760struct igc_nfc_rule *igc_get_nfc_rule(struct igc_adapter *adapter,
3761				      u32 location)
3762{
3763	struct igc_nfc_rule *rule;
3764
3765	list_for_each_entry(rule, &adapter->nfc_rule_list, list) {
3766		if (rule->location == location)
3767			return rule;
3768		if (rule->location > location)
3769			break;
3770	}
3771
3772	return NULL;
3773}
3774
3775/**
3776 * igc_del_nfc_rule() - Delete NFC rule
3777 * @adapter: Pointer to adapter
3778 * @rule: Pointer to rule to be deleted
3779 *
3780 * Disable NFC rule in hardware and delete it from adapter.
3781 *
3782 * Context: Expects adapter->nfc_rule_lock to be held by caller.
3783 */
3784void igc_del_nfc_rule(struct igc_adapter *adapter, struct igc_nfc_rule *rule)
3785{
3786	igc_disable_nfc_rule(adapter, rule);
3787
3788	list_del(&rule->list);
3789	adapter->nfc_rule_count--;
3790
3791	kfree(rule);
3792}
3793
3794static void igc_flush_nfc_rules(struct igc_adapter *adapter)
3795{
3796	struct igc_nfc_rule *rule, *tmp;
3797
3798	mutex_lock(&adapter->nfc_rule_lock);
3799
3800	list_for_each_entry_safe(rule, tmp, &adapter->nfc_rule_list, list)
3801		igc_del_nfc_rule(adapter, rule);
3802
3803	mutex_unlock(&adapter->nfc_rule_lock);
3804}
3805
3806/**
3807 * igc_add_nfc_rule() - Add NFC rule
3808 * @adapter: Pointer to adapter
3809 * @rule: Pointer to rule to be added
3810 *
3811 * Enable NFC rule in hardware and add it to adapter.
3812 *
3813 * Context: Expects adapter->nfc_rule_lock to be held by caller.
3814 *
3815 * Return: 0 on success, negative errno on failure.
3816 */
3817int igc_add_nfc_rule(struct igc_adapter *adapter, struct igc_nfc_rule *rule)
3818{
3819	struct igc_nfc_rule *pred, *cur;
3820	int err;
3821
3822	err = igc_enable_nfc_rule(adapter, rule);
3823	if (err)
3824		return err;
3825
3826	pred = NULL;
3827	list_for_each_entry(cur, &adapter->nfc_rule_list, list) {
3828		if (cur->location >= rule->location)
3829			break;
3830		pred = cur;
3831	}
3832
3833	list_add(&rule->list, pred ? &pred->list : &adapter->nfc_rule_list);
3834	adapter->nfc_rule_count++;
3835	return 0;
3836}
3837
3838static void igc_restore_nfc_rules(struct igc_adapter *adapter)
3839{
3840	struct igc_nfc_rule *rule;
3841
3842	mutex_lock(&adapter->nfc_rule_lock);
3843
3844	list_for_each_entry_reverse(rule, &adapter->nfc_rule_list, list)
3845		igc_enable_nfc_rule(adapter, rule);
3846
3847	mutex_unlock(&adapter->nfc_rule_lock);
3848}
3849
3850static int igc_uc_sync(struct net_device *netdev, const unsigned char *addr)
3851{
3852	struct igc_adapter *adapter = netdev_priv(netdev);
3853
3854	return igc_add_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST, addr, -1);
3855}
3856
3857static int igc_uc_unsync(struct net_device *netdev, const unsigned char *addr)
3858{
3859	struct igc_adapter *adapter = netdev_priv(netdev);
3860
3861	igc_del_mac_filter(adapter, IGC_MAC_FILTER_TYPE_DST, addr);
3862	return 0;
3863}
3864
3865/**
3866 * igc_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3867 * @netdev: network interface device structure
3868 *
3869 * The set_rx_mode entry point is called whenever the unicast or multicast
3870 * address lists or the network interface flags are updated.  This routine is
3871 * responsible for configuring the hardware for proper unicast, multicast,
3872 * promiscuous mode, and all-multi behavior.
3873 */
3874static void igc_set_rx_mode(struct net_device *netdev)
3875{
3876	struct igc_adapter *adapter = netdev_priv(netdev);
3877	struct igc_hw *hw = &adapter->hw;
3878	u32 rctl = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
3879	int count;
3880
3881	/* Check for Promiscuous and All Multicast modes */
3882	if (netdev->flags & IFF_PROMISC) {
3883		rctl |= IGC_RCTL_UPE | IGC_RCTL_MPE;
3884	} else {
3885		if (netdev->flags & IFF_ALLMULTI) {
3886			rctl |= IGC_RCTL_MPE;
3887		} else {
3888			/* Write addresses to the MTA, if the attempt fails
3889			 * then we should just turn on promiscuous mode so
3890			 * that we can at least receive multicast traffic
3891			 */
3892			count = igc_write_mc_addr_list(netdev);
3893			if (count < 0)
3894				rctl |= IGC_RCTL_MPE;
3895		}
3896	}
3897
3898	/* Write addresses to available RAR registers, if there is not
3899	 * sufficient space to store all the addresses then enable
3900	 * unicast promiscuous mode
3901	 */
3902	if (__dev_uc_sync(netdev, igc_uc_sync, igc_uc_unsync))
3903		rctl |= IGC_RCTL_UPE;
3904
3905	/* update state of unicast and multicast */
3906	rctl |= rd32(IGC_RCTL) & ~(IGC_RCTL_UPE | IGC_RCTL_MPE);
3907	wr32(IGC_RCTL, rctl);
3908
3909#if (PAGE_SIZE < 8192)
3910	if (adapter->max_frame_size <= IGC_MAX_FRAME_BUILD_SKB)
3911		rlpml = IGC_MAX_FRAME_BUILD_SKB;
3912#endif
3913	wr32(IGC_RLPML, rlpml);
3914}
3915
3916/**
3917 * igc_configure - configure the hardware for RX and TX
3918 * @adapter: private board structure
3919 */
3920static void igc_configure(struct igc_adapter *adapter)
3921{
3922	struct net_device *netdev = adapter->netdev;
3923	int i = 0;
3924
3925	igc_get_hw_control(adapter);
3926	igc_set_rx_mode(netdev);
3927
3928	igc_restore_vlan(adapter);
3929
3930	igc_setup_tctl(adapter);
3931	igc_setup_mrqc(adapter);
3932	igc_setup_rctl(adapter);
3933
3934	igc_set_default_mac_filter(adapter);
3935	igc_restore_nfc_rules(adapter);
3936
3937	igc_configure_tx(adapter);
3938	igc_configure_rx(adapter);
3939
3940	igc_rx_fifo_flush_base(&adapter->hw);
3941
3942	/* call igc_desc_unused which always leaves
3943	 * at least 1 descriptor unused to make sure
3944	 * next_to_use != next_to_clean
3945	 */
3946	for (i = 0; i < adapter->num_rx_queues; i++) {
3947		struct igc_ring *ring = adapter->rx_ring[i];
3948
3949		if (ring->xsk_pool)
3950			igc_alloc_rx_buffers_zc(ring, igc_desc_unused(ring));
3951		else
3952			igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
3953	}
3954}
3955
3956/**
3957 * igc_write_ivar - configure ivar for given MSI-X vector
3958 * @hw: pointer to the HW structure
3959 * @msix_vector: vector number we are allocating to a given ring
3960 * @index: row index of IVAR register to write within IVAR table
3961 * @offset: column offset of in IVAR, should be multiple of 8
3962 *
3963 * The IVAR table consists of 2 columns,
3964 * each containing an cause allocation for an Rx and Tx ring, and a
3965 * variable number of rows depending on the number of queues supported.
3966 */
3967static void igc_write_ivar(struct igc_hw *hw, int msix_vector,
3968			   int index, int offset)
3969{
3970	u32 ivar = array_rd32(IGC_IVAR0, index);
3971
3972	/* clear any bits that are currently set */
3973	ivar &= ~((u32)0xFF << offset);
3974
3975	/* write vector and valid bit */
3976	ivar |= (msix_vector | IGC_IVAR_VALID) << offset;
3977
3978	array_wr32(IGC_IVAR0, index, ivar);
3979}
3980
3981static void igc_assign_vector(struct igc_q_vector *q_vector, int msix_vector)
3982{
3983	struct igc_adapter *adapter = q_vector->adapter;
3984	struct igc_hw *hw = &adapter->hw;
3985	int rx_queue = IGC_N0_QUEUE;
3986	int tx_queue = IGC_N0_QUEUE;
3987
3988	if (q_vector->rx.ring)
3989		rx_queue = q_vector->rx.ring->reg_idx;
3990	if (q_vector->tx.ring)
3991		tx_queue = q_vector->tx.ring->reg_idx;
3992
3993	switch (hw->mac.type) {
3994	case igc_i225:
3995		if (rx_queue > IGC_N0_QUEUE)
3996			igc_write_ivar(hw, msix_vector,
3997				       rx_queue >> 1,
3998				       (rx_queue & 0x1) << 4);
3999		if (tx_queue > IGC_N0_QUEUE)
4000			igc_write_ivar(hw, msix_vector,
4001				       tx_queue >> 1,
4002				       ((tx_queue & 0x1) << 4) + 8);
4003		q_vector->eims_value = BIT(msix_vector);
4004		break;
4005	default:
4006		WARN_ONCE(hw->mac.type != igc_i225, "Wrong MAC type\n");
4007		break;
4008	}
4009
4010	/* add q_vector eims value to global eims_enable_mask */
4011	adapter->eims_enable_mask |= q_vector->eims_value;
4012
4013	/* configure q_vector to set itr on first interrupt */
4014	q_vector->set_itr = 1;
4015}
4016
4017/**
4018 * igc_configure_msix - Configure MSI-X hardware
4019 * @adapter: Pointer to adapter structure
4020 *
4021 * igc_configure_msix sets up the hardware to properly
4022 * generate MSI-X interrupts.
4023 */
4024static void igc_configure_msix(struct igc_adapter *adapter)
4025{
4026	struct igc_hw *hw = &adapter->hw;
4027	int i, vector = 0;
4028	u32 tmp;
4029
4030	adapter->eims_enable_mask = 0;
4031
4032	/* set vector for other causes, i.e. link changes */
4033	switch (hw->mac.type) {
4034	case igc_i225:
4035		/* Turn on MSI-X capability first, or our settings
4036		 * won't stick.  And it will take days to debug.
4037		 */
4038		wr32(IGC_GPIE, IGC_GPIE_MSIX_MODE |
4039		     IGC_GPIE_PBA | IGC_GPIE_EIAME |
4040		     IGC_GPIE_NSICR);
4041
4042		/* enable msix_other interrupt */
4043		adapter->eims_other = BIT(vector);
4044		tmp = (vector++ | IGC_IVAR_VALID) << 8;
4045
4046		wr32(IGC_IVAR_MISC, tmp);
4047		break;
4048	default:
4049		/* do nothing, since nothing else supports MSI-X */
4050		break;
4051	} /* switch (hw->mac.type) */
4052
4053	adapter->eims_enable_mask |= adapter->eims_other;
4054
4055	for (i = 0; i < adapter->num_q_vectors; i++)
4056		igc_assign_vector(adapter->q_vector[i], vector++);
4057
4058	wrfl();
4059}
4060
4061/**
4062 * igc_irq_enable - Enable default interrupt generation settings
4063 * @adapter: board private structure
4064 */
4065static void igc_irq_enable(struct igc_adapter *adapter)
4066{
4067	struct igc_hw *hw = &adapter->hw;
4068
4069	if (adapter->msix_entries) {
4070		u32 ims = IGC_IMS_LSC | IGC_IMS_DOUTSYNC | IGC_IMS_DRSTA;
4071		u32 regval = rd32(IGC_EIAC);
4072
4073		wr32(IGC_EIAC, regval | adapter->eims_enable_mask);
4074		regval = rd32(IGC_EIAM);
4075		wr32(IGC_EIAM, regval | adapter->eims_enable_mask);
4076		wr32(IGC_EIMS, adapter->eims_enable_mask);
4077		wr32(IGC_IMS, ims);
4078	} else {
4079		wr32(IGC_IMS, IMS_ENABLE_MASK | IGC_IMS_DRSTA);
4080		wr32(IGC_IAM, IMS_ENABLE_MASK | IGC_IMS_DRSTA);
4081	}
4082}
4083
4084/**
4085 * igc_irq_disable - Mask off interrupt generation on the NIC
4086 * @adapter: board private structure
4087 */
4088static void igc_irq_disable(struct igc_adapter *adapter)
4089{
4090	struct igc_hw *hw = &adapter->hw;
4091
4092	if (adapter->msix_entries) {
4093		u32 regval = rd32(IGC_EIAM);
4094
4095		wr32(IGC_EIAM, regval & ~adapter->eims_enable_mask);
4096		wr32(IGC_EIMC, adapter->eims_enable_mask);
4097		regval = rd32(IGC_EIAC);
4098		wr32(IGC_EIAC, regval & ~adapter->eims_enable_mask);
4099	}
4100
4101	wr32(IGC_IAM, 0);
4102	wr32(IGC_IMC, ~0);
4103	wrfl();
4104
4105	if (adapter->msix_entries) {
4106		int vector = 0, i;
4107
4108		synchronize_irq(adapter->msix_entries[vector++].vector);
4109
4110		for (i = 0; i < adapter->num_q_vectors; i++)
4111			synchronize_irq(adapter->msix_entries[vector++].vector);
4112	} else {
4113		synchronize_irq(adapter->pdev->irq);
4114	}
4115}
4116
4117void igc_set_flag_queue_pairs(struct igc_adapter *adapter,
4118			      const u32 max_rss_queues)
4119{
4120	/* Determine if we need to pair queues. */
4121	/* If rss_queues > half of max_rss_queues, pair the queues in
4122	 * order to conserve interrupts due to limited supply.
4123	 */
4124	if (adapter->rss_queues > (max_rss_queues / 2))
4125		adapter->flags |= IGC_FLAG_QUEUE_PAIRS;
4126	else
4127		adapter->flags &= ~IGC_FLAG_QUEUE_PAIRS;
4128}
4129
4130unsigned int igc_get_max_rss_queues(struct igc_adapter *adapter)
4131{
4132	return IGC_MAX_RX_QUEUES;
4133}
4134
4135static void igc_init_queue_configuration(struct igc_adapter *adapter)
4136{
4137	u32 max_rss_queues;
4138
4139	max_rss_queues = igc_get_max_rss_queues(adapter);
4140	adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
4141
4142	igc_set_flag_queue_pairs(adapter, max_rss_queues);
4143}
4144
4145/**
4146 * igc_reset_q_vector - Reset config for interrupt vector
4147 * @adapter: board private structure to initialize
4148 * @v_idx: Index of vector to be reset
4149 *
4150 * If NAPI is enabled it will delete any references to the
4151 * NAPI struct. This is preparation for igc_free_q_vector.
4152 */
4153static void igc_reset_q_vector(struct igc_adapter *adapter, int v_idx)
4154{
4155	struct igc_q_vector *q_vector = adapter->q_vector[v_idx];
4156
4157	/* if we're coming from igc_set_interrupt_capability, the vectors are
4158	 * not yet allocated
4159	 */
4160	if (!q_vector)
4161		return;
4162
4163	if (q_vector->tx.ring)
4164		adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
4165
4166	if (q_vector->rx.ring)
4167		adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
4168
4169	netif_napi_del(&q_vector->napi);
4170}
4171
4172/**
4173 * igc_free_q_vector - Free memory allocated for specific interrupt vector
4174 * @adapter: board private structure to initialize
4175 * @v_idx: Index of vector to be freed
4176 *
4177 * This function frees the memory allocated to the q_vector.
4178 */
4179static void igc_free_q_vector(struct igc_adapter *adapter, int v_idx)
4180{
4181	struct igc_q_vector *q_vector = adapter->q_vector[v_idx];
4182
4183	adapter->q_vector[v_idx] = NULL;
4184
4185	/* igc_get_stats64() might access the rings on this vector,
4186	 * we must wait a grace period before freeing it.
4187	 */
4188	if (q_vector)
4189		kfree_rcu(q_vector, rcu);
4190}
4191
4192/**
4193 * igc_free_q_vectors - Free memory allocated for interrupt vectors
4194 * @adapter: board private structure to initialize
4195 *
4196 * This function frees the memory allocated to the q_vectors.  In addition if
4197 * NAPI is enabled it will delete any references to the NAPI struct prior
4198 * to freeing the q_vector.
4199 */
4200static void igc_free_q_vectors(struct igc_adapter *adapter)
4201{
4202	int v_idx = adapter->num_q_vectors;
4203
4204	adapter->num_tx_queues = 0;
4205	adapter->num_rx_queues = 0;
4206	adapter->num_q_vectors = 0;
4207
4208	while (v_idx--) {
4209		igc_reset_q_vector(adapter, v_idx);
4210		igc_free_q_vector(adapter, v_idx);
4211	}
4212}
4213
4214/**
4215 * igc_update_itr - update the dynamic ITR value based on statistics
4216 * @q_vector: pointer to q_vector
4217 * @ring_container: ring info to update the itr for
4218 *
4219 * Stores a new ITR value based on packets and byte
4220 * counts during the last interrupt.  The advantage of per interrupt
4221 * computation is faster updates and more accurate ITR for the current
4222 * traffic pattern.  Constants in this function were computed
4223 * based on theoretical maximum wire speed and thresholds were set based
4224 * on testing data as well as attempting to minimize response time
4225 * while increasing bulk throughput.
4226 * NOTE: These calculations are only valid when operating in a single-
4227 * queue environment.
4228 */
4229static void igc_update_itr(struct igc_q_vector *q_vector,
4230			   struct igc_ring_container *ring_container)
4231{
4232	unsigned int packets = ring_container->total_packets;
4233	unsigned int bytes = ring_container->total_bytes;
4234	u8 itrval = ring_container->itr;
4235
4236	/* no packets, exit with status unchanged */
4237	if (packets == 0)
4238		return;
4239
4240	switch (itrval) {
4241	case lowest_latency:
4242		/* handle TSO and jumbo frames */
4243		if (bytes / packets > 8000)
4244			itrval = bulk_latency;
4245		else if ((packets < 5) && (bytes > 512))
4246			itrval = low_latency;
4247		break;
4248	case low_latency:  /* 50 usec aka 20000 ints/s */
4249		if (bytes > 10000) {
4250			/* this if handles the TSO accounting */
4251			if (bytes / packets > 8000)
4252				itrval = bulk_latency;
4253			else if ((packets < 10) || ((bytes / packets) > 1200))
4254				itrval = bulk_latency;
4255			else if ((packets > 35))
4256				itrval = lowest_latency;
4257		} else if (bytes / packets > 2000) {
4258			itrval = bulk_latency;
4259		} else if (packets <= 2 && bytes < 512) {
4260			itrval = lowest_latency;
4261		}
4262		break;
4263	case bulk_latency: /* 250 usec aka 4000 ints/s */
4264		if (bytes > 25000) {
4265			if (packets > 35)
4266				itrval = low_latency;
4267		} else if (bytes < 1500) {
4268			itrval = low_latency;
4269		}
4270		break;
4271	}
4272
4273	/* clear work counters since we have the values we need */
4274	ring_container->total_bytes = 0;
4275	ring_container->total_packets = 0;
4276
4277	/* write updated itr to ring container */
4278	ring_container->itr = itrval;
4279}
4280
4281static void igc_set_itr(struct igc_q_vector *q_vector)
4282{
4283	struct igc_adapter *adapter = q_vector->adapter;
4284	u32 new_itr = q_vector->itr_val;
4285	u8 current_itr = 0;
4286
4287	/* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4288	switch (adapter->link_speed) {
4289	case SPEED_10:
4290	case SPEED_100:
4291		current_itr = 0;
4292		new_itr = IGC_4K_ITR;
4293		goto set_itr_now;
4294	default:
4295		break;
4296	}
4297
4298	igc_update_itr(q_vector, &q_vector->tx);
4299	igc_update_itr(q_vector, &q_vector->rx);
4300
4301	current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4302
4303	/* conservative mode (itr 3) eliminates the lowest_latency setting */
4304	if (current_itr == lowest_latency &&
4305	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4306	    (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4307		current_itr = low_latency;
4308
4309	switch (current_itr) {
4310	/* counts and packets in update_itr are dependent on these numbers */
4311	case lowest_latency:
4312		new_itr = IGC_70K_ITR; /* 70,000 ints/sec */
4313		break;
4314	case low_latency:
4315		new_itr = IGC_20K_ITR; /* 20,000 ints/sec */
4316		break;
4317	case bulk_latency:
4318		new_itr = IGC_4K_ITR;  /* 4,000 ints/sec */
4319		break;
4320	default:
4321		break;
4322	}
4323
4324set_itr_now:
4325	if (new_itr != q_vector->itr_val) {
4326		/* this attempts to bias the interrupt rate towards Bulk
4327		 * by adding intermediate steps when interrupt rate is
4328		 * increasing
4329		 */
4330		new_itr = new_itr > q_vector->itr_val ?
4331			  max((new_itr * q_vector->itr_val) /
4332			  (new_itr + (q_vector->itr_val >> 2)),
4333			  new_itr) : new_itr;
4334		/* Don't write the value here; it resets the adapter's
4335		 * internal timer, and causes us to delay far longer than
4336		 * we should between interrupts.  Instead, we write the ITR
4337		 * value at the beginning of the next interrupt so the timing
4338		 * ends up being correct.
4339		 */
4340		q_vector->itr_val = new_itr;
4341		q_vector->set_itr = 1;
4342	}
4343}
4344
4345static void igc_reset_interrupt_capability(struct igc_adapter *adapter)
4346{
4347	int v_idx = adapter->num_q_vectors;
4348
4349	if (adapter->msix_entries) {
4350		pci_disable_msix(adapter->pdev);
4351		kfree(adapter->msix_entries);
4352		adapter->msix_entries = NULL;
4353	} else if (adapter->flags & IGC_FLAG_HAS_MSI) {
4354		pci_disable_msi(adapter->pdev);
4355	}
4356
4357	while (v_idx--)
4358		igc_reset_q_vector(adapter, v_idx);
4359}
4360
4361/**
4362 * igc_set_interrupt_capability - set MSI or MSI-X if supported
4363 * @adapter: Pointer to adapter structure
4364 * @msix: boolean value for MSI-X capability
4365 *
4366 * Attempt to configure interrupts using the best available
4367 * capabilities of the hardware and kernel.
4368 */
4369static void igc_set_interrupt_capability(struct igc_adapter *adapter,
4370					 bool msix)
4371{
4372	int numvecs, i;
4373	int err;
4374
4375	if (!msix)
4376		goto msi_only;
4377	adapter->flags |= IGC_FLAG_HAS_MSIX;
4378
4379	/* Number of supported queues. */
4380	adapter->num_rx_queues = adapter->rss_queues;
4381
4382	adapter->num_tx_queues = adapter->rss_queues;
4383
4384	/* start with one vector for every Rx queue */
4385	numvecs = adapter->num_rx_queues;
4386
4387	/* if Tx handler is separate add 1 for every Tx queue */
4388	if (!(adapter->flags & IGC_FLAG_QUEUE_PAIRS))
4389		numvecs += adapter->num_tx_queues;
4390
4391	/* store the number of vectors reserved for queues */
4392	adapter->num_q_vectors = numvecs;
4393
4394	/* add 1 vector for link status interrupts */
4395	numvecs++;
4396
4397	adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
4398					GFP_KERNEL);
4399
4400	if (!adapter->msix_entries)
4401		return;
4402
4403	/* populate entry values */
4404	for (i = 0; i < numvecs; i++)
4405		adapter->msix_entries[i].entry = i;
4406
4407	err = pci_enable_msix_range(adapter->pdev,
4408				    adapter->msix_entries,
4409				    numvecs,
4410				    numvecs);
4411	if (err > 0)
4412		return;
4413
4414	kfree(adapter->msix_entries);
4415	adapter->msix_entries = NULL;
4416
4417	igc_reset_interrupt_capability(adapter);
4418
4419msi_only:
4420	adapter->flags &= ~IGC_FLAG_HAS_MSIX;
4421
4422	adapter->rss_queues = 1;
4423	adapter->flags |= IGC_FLAG_QUEUE_PAIRS;
4424	adapter->num_rx_queues = 1;
4425	adapter->num_tx_queues = 1;
4426	adapter->num_q_vectors = 1;
4427	if (!pci_enable_msi(adapter->pdev))
4428		adapter->flags |= IGC_FLAG_HAS_MSI;
4429}
4430
4431/**
4432 * igc_update_ring_itr - update the dynamic ITR value based on packet size
4433 * @q_vector: pointer to q_vector
4434 *
4435 * Stores a new ITR value based on strictly on packet size.  This
4436 * algorithm is less sophisticated than that used in igc_update_itr,
4437 * due to the difficulty of synchronizing statistics across multiple
4438 * receive rings.  The divisors and thresholds used by this function
4439 * were determined based on theoretical maximum wire speed and testing
4440 * data, in order to minimize response time while increasing bulk
4441 * throughput.
4442 * NOTE: This function is called only when operating in a multiqueue
4443 * receive environment.
4444 */
4445static void igc_update_ring_itr(struct igc_q_vector *q_vector)
4446{
4447	struct igc_adapter *adapter = q_vector->adapter;
4448	int new_val = q_vector->itr_val;
4449	int avg_wire_size = 0;
4450	unsigned int packets;
4451
4452	/* For non-gigabit speeds, just fix the interrupt rate at 4000
4453	 * ints/sec - ITR timer value of 120 ticks.
4454	 */
4455	switch (adapter->link_speed) {
4456	case SPEED_10:
4457	case SPEED_100:
4458		new_val = IGC_4K_ITR;
4459		goto set_itr_val;
4460	default:
4461		break;
4462	}
4463
4464	packets = q_vector->rx.total_packets;
4465	if (packets)
4466		avg_wire_size = q_vector->rx.total_bytes / packets;
4467
4468	packets = q_vector->tx.total_packets;
4469	if (packets)
4470		avg_wire_size = max_t(u32, avg_wire_size,
4471				      q_vector->tx.total_bytes / packets);
4472
4473	/* if avg_wire_size isn't set no work was done */
4474	if (!avg_wire_size)
4475		goto clear_counts;
4476
4477	/* Add 24 bytes to size to account for CRC, preamble, and gap */
4478	avg_wire_size += 24;
4479
4480	/* Don't starve jumbo frames */
4481	avg_wire_size = min(avg_wire_size, 3000);
4482
4483	/* Give a little boost to mid-size frames */
4484	if (avg_wire_size > 300 && avg_wire_size < 1200)
4485		new_val = avg_wire_size / 3;
4486	else
4487		new_val = avg_wire_size / 2;
4488
4489	/* conservative mode (itr 3) eliminates the lowest_latency setting */
4490	if (new_val < IGC_20K_ITR &&
4491	    ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4492	    (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4493		new_val = IGC_20K_ITR;
4494
4495set_itr_val:
4496	if (new_val != q_vector->itr_val) {
4497		q_vector->itr_val = new_val;
4498		q_vector->set_itr = 1;
4499	}
4500clear_counts:
4501	q_vector->rx.total_bytes = 0;
4502	q_vector->rx.total_packets = 0;
4503	q_vector->tx.total_bytes = 0;
4504	q_vector->tx.total_packets = 0;
4505}
4506
4507static void igc_ring_irq_enable(struct igc_q_vector *q_vector)
4508{
4509	struct igc_adapter *adapter = q_vector->adapter;
4510	struct igc_hw *hw = &adapter->hw;
4511
4512	if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
4513	    (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
4514		if (adapter->num_q_vectors == 1)
4515			igc_set_itr(q_vector);
4516		else
4517			igc_update_ring_itr(q_vector);
4518	}
4519
4520	if (!test_bit(__IGC_DOWN, &adapter->state)) {
4521		if (adapter->msix_entries)
4522			wr32(IGC_EIMS, q_vector->eims_value);
4523		else
4524			igc_irq_enable(adapter);
4525	}
4526}
4527
4528static void igc_add_ring(struct igc_ring *ring,
4529			 struct igc_ring_container *head)
4530{
4531	head->ring = ring;
4532	head->count++;
4533}
4534
4535/**
4536 * igc_cache_ring_register - Descriptor ring to register mapping
4537 * @adapter: board private structure to initialize
4538 *
4539 * Once we know the feature-set enabled for the device, we'll cache
4540 * the register offset the descriptor ring is assigned to.
4541 */
4542static void igc_cache_ring_register(struct igc_adapter *adapter)
4543{
4544	int i = 0, j = 0;
4545
4546	switch (adapter->hw.mac.type) {
4547	case igc_i225:
4548	default:
4549		for (; i < adapter->num_rx_queues; i++)
4550			adapter->rx_ring[i]->reg_idx = i;
4551		for (; j < adapter->num_tx_queues; j++)
4552			adapter->tx_ring[j]->reg_idx = j;
4553		break;
4554	}
4555}
4556
4557/**
4558 * igc_poll - NAPI Rx polling callback
4559 * @napi: napi polling structure
4560 * @budget: count of how many packets we should handle
4561 */
4562static int igc_poll(struct napi_struct *napi, int budget)
4563{
4564	struct igc_q_vector *q_vector = container_of(napi,
4565						     struct igc_q_vector,
4566						     napi);
4567	struct igc_ring *rx_ring = q_vector->rx.ring;
4568	bool clean_complete = true;
4569	int work_done = 0;
4570
4571	if (q_vector->tx.ring)
4572		clean_complete = igc_clean_tx_irq(q_vector, budget);
4573
4574	if (rx_ring) {
4575		int cleaned = rx_ring->xsk_pool ?
4576			      igc_clean_rx_irq_zc(q_vector, budget) :
4577			      igc_clean_rx_irq(q_vector, budget);
4578
4579		work_done += cleaned;
4580		if (cleaned >= budget)
4581			clean_complete = false;
4582	}
4583
4584	/* If all work not completed, return budget and keep polling */
4585	if (!clean_complete)
4586		return budget;
4587
4588	/* Exit the polling mode, but don't re-enable interrupts if stack might
4589	 * poll us due to busy-polling
4590	 */
4591	if (likely(napi_complete_done(napi, work_done)))
4592		igc_ring_irq_enable(q_vector);
4593
4594	return min(work_done, budget - 1);
4595}
4596
4597/**
4598 * igc_alloc_q_vector - Allocate memory for a single interrupt vector
4599 * @adapter: board private structure to initialize
4600 * @v_count: q_vectors allocated on adapter, used for ring interleaving
4601 * @v_idx: index of vector in adapter struct
4602 * @txr_count: total number of Tx rings to allocate
4603 * @txr_idx: index of first Tx ring to allocate
4604 * @rxr_count: total number of Rx rings to allocate
4605 * @rxr_idx: index of first Rx ring to allocate
4606 *
4607 * We allocate one q_vector.  If allocation fails we return -ENOMEM.
4608 */
4609static int igc_alloc_q_vector(struct igc_adapter *adapter,
4610			      unsigned int v_count, unsigned int v_idx,
4611			      unsigned int txr_count, unsigned int txr_idx,
4612			      unsigned int rxr_count, unsigned int rxr_idx)
4613{
4614	struct igc_q_vector *q_vector;
4615	struct igc_ring *ring;
4616	int ring_count;
4617
4618	/* igc only supports 1 Tx and/or 1 Rx queue per vector */
4619	if (txr_count > 1 || rxr_count > 1)
4620		return -ENOMEM;
4621
4622	ring_count = txr_count + rxr_count;
4623
4624	/* allocate q_vector and rings */
4625	q_vector = adapter->q_vector[v_idx];
4626	if (!q_vector)
4627		q_vector = kzalloc(struct_size(q_vector, ring, ring_count),
4628				   GFP_KERNEL);
4629	else
4630		memset(q_vector, 0, struct_size(q_vector, ring, ring_count));
4631	if (!q_vector)
4632		return -ENOMEM;
4633
4634	/* initialize NAPI */
4635	netif_napi_add(adapter->netdev, &q_vector->napi, igc_poll);
 
4636
4637	/* tie q_vector and adapter together */
4638	adapter->q_vector[v_idx] = q_vector;
4639	q_vector->adapter = adapter;
4640
4641	/* initialize work limits */
4642	q_vector->tx.work_limit = adapter->tx_work_limit;
4643
4644	/* initialize ITR configuration */
4645	q_vector->itr_register = adapter->io_addr + IGC_EITR(0);
4646	q_vector->itr_val = IGC_START_ITR;
4647
4648	/* initialize pointer to rings */
4649	ring = q_vector->ring;
4650
4651	/* initialize ITR */
4652	if (rxr_count) {
4653		/* rx or rx/tx vector */
4654		if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
4655			q_vector->itr_val = adapter->rx_itr_setting;
4656	} else {
4657		/* tx only vector */
4658		if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
4659			q_vector->itr_val = adapter->tx_itr_setting;
4660	}
4661
4662	if (txr_count) {
4663		/* assign generic ring traits */
4664		ring->dev = &adapter->pdev->dev;
4665		ring->netdev = adapter->netdev;
4666
4667		/* configure backlink on ring */
4668		ring->q_vector = q_vector;
4669
4670		/* update q_vector Tx values */
4671		igc_add_ring(ring, &q_vector->tx);
4672
4673		/* apply Tx specific ring traits */
4674		ring->count = adapter->tx_ring_count;
4675		ring->queue_index = txr_idx;
4676
4677		/* assign ring to adapter */
4678		adapter->tx_ring[txr_idx] = ring;
4679
4680		/* push pointer to next ring */
4681		ring++;
4682	}
4683
4684	if (rxr_count) {
4685		/* assign generic ring traits */
4686		ring->dev = &adapter->pdev->dev;
4687		ring->netdev = adapter->netdev;
4688
4689		/* configure backlink on ring */
4690		ring->q_vector = q_vector;
4691
4692		/* update q_vector Rx values */
4693		igc_add_ring(ring, &q_vector->rx);
4694
4695		/* apply Rx specific ring traits */
4696		ring->count = adapter->rx_ring_count;
4697		ring->queue_index = rxr_idx;
4698
4699		/* assign ring to adapter */
4700		adapter->rx_ring[rxr_idx] = ring;
4701	}
4702
4703	return 0;
4704}
4705
4706/**
4707 * igc_alloc_q_vectors - Allocate memory for interrupt vectors
4708 * @adapter: board private structure to initialize
4709 *
4710 * We allocate one q_vector per queue interrupt.  If allocation fails we
4711 * return -ENOMEM.
4712 */
4713static int igc_alloc_q_vectors(struct igc_adapter *adapter)
4714{
4715	int rxr_remaining = adapter->num_rx_queues;
4716	int txr_remaining = adapter->num_tx_queues;
4717	int rxr_idx = 0, txr_idx = 0, v_idx = 0;
4718	int q_vectors = adapter->num_q_vectors;
4719	int err;
4720
4721	if (q_vectors >= (rxr_remaining + txr_remaining)) {
4722		for (; rxr_remaining; v_idx++) {
4723			err = igc_alloc_q_vector(adapter, q_vectors, v_idx,
4724						 0, 0, 1, rxr_idx);
4725
4726			if (err)
4727				goto err_out;
4728
4729			/* update counts and index */
4730			rxr_remaining--;
4731			rxr_idx++;
4732		}
4733	}
4734
4735	for (; v_idx < q_vectors; v_idx++) {
4736		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
4737		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
4738
4739		err = igc_alloc_q_vector(adapter, q_vectors, v_idx,
4740					 tqpv, txr_idx, rqpv, rxr_idx);
4741
4742		if (err)
4743			goto err_out;
4744
4745		/* update counts and index */
4746		rxr_remaining -= rqpv;
4747		txr_remaining -= tqpv;
4748		rxr_idx++;
4749		txr_idx++;
4750	}
4751
4752	return 0;
4753
4754err_out:
4755	adapter->num_tx_queues = 0;
4756	adapter->num_rx_queues = 0;
4757	adapter->num_q_vectors = 0;
4758
4759	while (v_idx--)
4760		igc_free_q_vector(adapter, v_idx);
4761
4762	return -ENOMEM;
4763}
4764
4765/**
4766 * igc_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
4767 * @adapter: Pointer to adapter structure
4768 * @msix: boolean for MSI-X capability
4769 *
4770 * This function initializes the interrupts and allocates all of the queues.
4771 */
4772static int igc_init_interrupt_scheme(struct igc_adapter *adapter, bool msix)
4773{
4774	struct net_device *dev = adapter->netdev;
4775	int err = 0;
4776
4777	igc_set_interrupt_capability(adapter, msix);
4778
4779	err = igc_alloc_q_vectors(adapter);
4780	if (err) {
4781		netdev_err(dev, "Unable to allocate memory for vectors\n");
4782		goto err_alloc_q_vectors;
4783	}
4784
4785	igc_cache_ring_register(adapter);
4786
4787	return 0;
4788
4789err_alloc_q_vectors:
4790	igc_reset_interrupt_capability(adapter);
4791	return err;
4792}
4793
4794/**
4795 * igc_sw_init - Initialize general software structures (struct igc_adapter)
4796 * @adapter: board private structure to initialize
4797 *
4798 * igc_sw_init initializes the Adapter private data structure.
4799 * Fields are initialized based on PCI device information and
4800 * OS network device settings (MTU size).
4801 */
4802static int igc_sw_init(struct igc_adapter *adapter)
4803{
4804	struct net_device *netdev = adapter->netdev;
4805	struct pci_dev *pdev = adapter->pdev;
4806	struct igc_hw *hw = &adapter->hw;
4807
4808	pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
4809
4810	/* set default ring sizes */
4811	adapter->tx_ring_count = IGC_DEFAULT_TXD;
4812	adapter->rx_ring_count = IGC_DEFAULT_RXD;
4813
4814	/* set default ITR values */
4815	adapter->rx_itr_setting = IGC_DEFAULT_ITR;
4816	adapter->tx_itr_setting = IGC_DEFAULT_ITR;
4817
4818	/* set default work limits */
4819	adapter->tx_work_limit = IGC_DEFAULT_TX_WORK;
4820
4821	/* adjust max frame to be at least the size of a standard frame */
4822	adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
4823				VLAN_HLEN;
4824	adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
4825
4826	mutex_init(&adapter->nfc_rule_lock);
4827	INIT_LIST_HEAD(&adapter->nfc_rule_list);
4828	adapter->nfc_rule_count = 0;
4829
4830	spin_lock_init(&adapter->stats64_lock);
4831	spin_lock_init(&adapter->qbv_tx_lock);
4832	/* Assume MSI-X interrupts, will be checked during IRQ allocation */
4833	adapter->flags |= IGC_FLAG_HAS_MSIX;
4834
4835	igc_init_queue_configuration(adapter);
4836
4837	/* This call may decrease the number of queues */
4838	if (igc_init_interrupt_scheme(adapter, true)) {
4839		netdev_err(netdev, "Unable to allocate memory for queues\n");
4840		return -ENOMEM;
4841	}
4842
4843	/* Explicitly disable IRQ since the NIC can be in any state. */
4844	igc_irq_disable(adapter);
4845
4846	set_bit(__IGC_DOWN, &adapter->state);
4847
4848	return 0;
4849}
4850
4851/**
4852 * igc_up - Open the interface and prepare it to handle traffic
4853 * @adapter: board private structure
4854 */
4855void igc_up(struct igc_adapter *adapter)
4856{
4857	struct igc_hw *hw = &adapter->hw;
4858	int i = 0;
4859
4860	/* hardware has been reset, we need to reload some things */
4861	igc_configure(adapter);
4862
4863	clear_bit(__IGC_DOWN, &adapter->state);
4864
4865	for (i = 0; i < adapter->num_q_vectors; i++)
4866		napi_enable(&adapter->q_vector[i]->napi);
4867
4868	if (adapter->msix_entries)
4869		igc_configure_msix(adapter);
4870	else
4871		igc_assign_vector(adapter->q_vector[0], 0);
4872
4873	/* Clear any pending interrupts. */
4874	rd32(IGC_ICR);
4875	igc_irq_enable(adapter);
4876
4877	netif_tx_start_all_queues(adapter->netdev);
4878
4879	/* start the watchdog. */
4880	hw->mac.get_link_status = true;
4881	schedule_work(&adapter->watchdog_task);
4882}
4883
4884/**
4885 * igc_update_stats - Update the board statistics counters
4886 * @adapter: board private structure
4887 */
4888void igc_update_stats(struct igc_adapter *adapter)
4889{
4890	struct rtnl_link_stats64 *net_stats = &adapter->stats64;
4891	struct pci_dev *pdev = adapter->pdev;
4892	struct igc_hw *hw = &adapter->hw;
4893	u64 _bytes, _packets;
4894	u64 bytes, packets;
4895	unsigned int start;
4896	u32 mpc;
4897	int i;
4898
4899	/* Prevent stats update while adapter is being reset, or if the pci
4900	 * connection is down.
4901	 */
4902	if (adapter->link_speed == 0)
4903		return;
4904	if (pci_channel_offline(pdev))
4905		return;
4906
4907	packets = 0;
4908	bytes = 0;
4909
4910	rcu_read_lock();
4911	for (i = 0; i < adapter->num_rx_queues; i++) {
4912		struct igc_ring *ring = adapter->rx_ring[i];
4913		u32 rqdpc = rd32(IGC_RQDPC(i));
4914
4915		if (hw->mac.type >= igc_i225)
4916			wr32(IGC_RQDPC(i), 0);
4917
4918		if (rqdpc) {
4919			ring->rx_stats.drops += rqdpc;
4920			net_stats->rx_fifo_errors += rqdpc;
4921		}
4922
4923		do {
4924			start = u64_stats_fetch_begin(&ring->rx_syncp);
4925			_bytes = ring->rx_stats.bytes;
4926			_packets = ring->rx_stats.packets;
4927		} while (u64_stats_fetch_retry(&ring->rx_syncp, start));
4928		bytes += _bytes;
4929		packets += _packets;
4930	}
4931
4932	net_stats->rx_bytes = bytes;
4933	net_stats->rx_packets = packets;
4934
4935	packets = 0;
4936	bytes = 0;
4937	for (i = 0; i < adapter->num_tx_queues; i++) {
4938		struct igc_ring *ring = adapter->tx_ring[i];
4939
4940		do {
4941			start = u64_stats_fetch_begin(&ring->tx_syncp);
4942			_bytes = ring->tx_stats.bytes;
4943			_packets = ring->tx_stats.packets;
4944		} while (u64_stats_fetch_retry(&ring->tx_syncp, start));
4945		bytes += _bytes;
4946		packets += _packets;
4947	}
4948	net_stats->tx_bytes = bytes;
4949	net_stats->tx_packets = packets;
4950	rcu_read_unlock();
4951
4952	/* read stats registers */
4953	adapter->stats.crcerrs += rd32(IGC_CRCERRS);
4954	adapter->stats.gprc += rd32(IGC_GPRC);
4955	adapter->stats.gorc += rd32(IGC_GORCL);
4956	rd32(IGC_GORCH); /* clear GORCL */
4957	adapter->stats.bprc += rd32(IGC_BPRC);
4958	adapter->stats.mprc += rd32(IGC_MPRC);
4959	adapter->stats.roc += rd32(IGC_ROC);
4960
4961	adapter->stats.prc64 += rd32(IGC_PRC64);
4962	adapter->stats.prc127 += rd32(IGC_PRC127);
4963	adapter->stats.prc255 += rd32(IGC_PRC255);
4964	adapter->stats.prc511 += rd32(IGC_PRC511);
4965	adapter->stats.prc1023 += rd32(IGC_PRC1023);
4966	adapter->stats.prc1522 += rd32(IGC_PRC1522);
4967	adapter->stats.tlpic += rd32(IGC_TLPIC);
4968	adapter->stats.rlpic += rd32(IGC_RLPIC);
4969	adapter->stats.hgptc += rd32(IGC_HGPTC);
4970
4971	mpc = rd32(IGC_MPC);
4972	adapter->stats.mpc += mpc;
4973	net_stats->rx_fifo_errors += mpc;
4974	adapter->stats.scc += rd32(IGC_SCC);
4975	adapter->stats.ecol += rd32(IGC_ECOL);
4976	adapter->stats.mcc += rd32(IGC_MCC);
4977	adapter->stats.latecol += rd32(IGC_LATECOL);
4978	adapter->stats.dc += rd32(IGC_DC);
4979	adapter->stats.rlec += rd32(IGC_RLEC);
4980	adapter->stats.xonrxc += rd32(IGC_XONRXC);
4981	adapter->stats.xontxc += rd32(IGC_XONTXC);
4982	adapter->stats.xoffrxc += rd32(IGC_XOFFRXC);
4983	adapter->stats.xofftxc += rd32(IGC_XOFFTXC);
4984	adapter->stats.fcruc += rd32(IGC_FCRUC);
4985	adapter->stats.gptc += rd32(IGC_GPTC);
4986	adapter->stats.gotc += rd32(IGC_GOTCL);
4987	rd32(IGC_GOTCH); /* clear GOTCL */
4988	adapter->stats.rnbc += rd32(IGC_RNBC);
4989	adapter->stats.ruc += rd32(IGC_RUC);
4990	adapter->stats.rfc += rd32(IGC_RFC);
4991	adapter->stats.rjc += rd32(IGC_RJC);
4992	adapter->stats.tor += rd32(IGC_TORH);
4993	adapter->stats.tot += rd32(IGC_TOTH);
4994	adapter->stats.tpr += rd32(IGC_TPR);
4995
4996	adapter->stats.ptc64 += rd32(IGC_PTC64);
4997	adapter->stats.ptc127 += rd32(IGC_PTC127);
4998	adapter->stats.ptc255 += rd32(IGC_PTC255);
4999	adapter->stats.ptc511 += rd32(IGC_PTC511);
5000	adapter->stats.ptc1023 += rd32(IGC_PTC1023);
5001	adapter->stats.ptc1522 += rd32(IGC_PTC1522);
5002
5003	adapter->stats.mptc += rd32(IGC_MPTC);
5004	adapter->stats.bptc += rd32(IGC_BPTC);
5005
5006	adapter->stats.tpt += rd32(IGC_TPT);
5007	adapter->stats.colc += rd32(IGC_COLC);
5008	adapter->stats.colc += rd32(IGC_RERC);
5009
5010	adapter->stats.algnerrc += rd32(IGC_ALGNERRC);
5011
5012	adapter->stats.tsctc += rd32(IGC_TSCTC);
5013
5014	adapter->stats.iac += rd32(IGC_IAC);
5015
5016	/* Fill out the OS statistics structure */
5017	net_stats->multicast = adapter->stats.mprc;
5018	net_stats->collisions = adapter->stats.colc;
5019
5020	/* Rx Errors */
5021
5022	/* RLEC on some newer hardware can be incorrect so build
5023	 * our own version based on RUC and ROC
5024	 */
5025	net_stats->rx_errors = adapter->stats.rxerrc +
5026		adapter->stats.crcerrs + adapter->stats.algnerrc +
5027		adapter->stats.ruc + adapter->stats.roc +
5028		adapter->stats.cexterr;
5029	net_stats->rx_length_errors = adapter->stats.ruc +
5030				      adapter->stats.roc;
5031	net_stats->rx_crc_errors = adapter->stats.crcerrs;
5032	net_stats->rx_frame_errors = adapter->stats.algnerrc;
5033	net_stats->rx_missed_errors = adapter->stats.mpc;
5034
5035	/* Tx Errors */
5036	net_stats->tx_errors = adapter->stats.ecol +
5037			       adapter->stats.latecol;
5038	net_stats->tx_aborted_errors = adapter->stats.ecol;
5039	net_stats->tx_window_errors = adapter->stats.latecol;
5040	net_stats->tx_carrier_errors = adapter->stats.tncrs;
5041
5042	/* Tx Dropped */
5043	net_stats->tx_dropped = adapter->stats.txdrop;
5044
5045	/* Management Stats */
5046	adapter->stats.mgptc += rd32(IGC_MGTPTC);
5047	adapter->stats.mgprc += rd32(IGC_MGTPRC);
5048	adapter->stats.mgpdc += rd32(IGC_MGTPDC);
5049}
5050
5051/**
5052 * igc_down - Close the interface
5053 * @adapter: board private structure
5054 */
5055void igc_down(struct igc_adapter *adapter)
5056{
5057	struct net_device *netdev = adapter->netdev;
5058	struct igc_hw *hw = &adapter->hw;
5059	u32 tctl, rctl;
5060	int i = 0;
5061
5062	set_bit(__IGC_DOWN, &adapter->state);
5063
5064	igc_ptp_suspend(adapter);
 
 
 
5065
5066	if (pci_device_is_present(adapter->pdev)) {
5067		/* disable receives in the hardware */
5068		rctl = rd32(IGC_RCTL);
5069		wr32(IGC_RCTL, rctl & ~IGC_RCTL_EN);
5070		/* flush and sleep below */
5071	}
5072	/* set trans_start so we don't get spurious watchdogs during reset */
5073	netif_trans_update(netdev);
5074
5075	netif_carrier_off(netdev);
5076	netif_tx_stop_all_queues(netdev);
5077
5078	if (pci_device_is_present(adapter->pdev)) {
5079		/* disable transmits in the hardware */
5080		tctl = rd32(IGC_TCTL);
5081		tctl &= ~IGC_TCTL_EN;
5082		wr32(IGC_TCTL, tctl);
5083		/* flush both disables and wait for them to finish */
5084		wrfl();
5085		usleep_range(10000, 20000);
5086
5087		igc_irq_disable(adapter);
5088	}
5089
5090	adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
5091
5092	for (i = 0; i < adapter->num_q_vectors; i++) {
5093		if (adapter->q_vector[i]) {
5094			napi_synchronize(&adapter->q_vector[i]->napi);
5095			napi_disable(&adapter->q_vector[i]->napi);
5096		}
5097	}
5098
5099	del_timer_sync(&adapter->watchdog_timer);
5100	del_timer_sync(&adapter->phy_info_timer);
5101
5102	/* record the stats before reset*/
5103	spin_lock(&adapter->stats64_lock);
5104	igc_update_stats(adapter);
5105	spin_unlock(&adapter->stats64_lock);
5106
5107	adapter->link_speed = 0;
5108	adapter->link_duplex = 0;
5109
5110	if (!pci_channel_offline(adapter->pdev))
5111		igc_reset(adapter);
5112
5113	/* clear VLAN promisc flag so VFTA will be updated if necessary */
5114	adapter->flags &= ~IGC_FLAG_VLAN_PROMISC;
5115
5116	igc_disable_all_tx_rings_hw(adapter);
5117	igc_clean_all_tx_rings(adapter);
5118	igc_clean_all_rx_rings(adapter);
5119}
5120
5121void igc_reinit_locked(struct igc_adapter *adapter)
5122{
 
5123	while (test_and_set_bit(__IGC_RESETTING, &adapter->state))
5124		usleep_range(1000, 2000);
5125	igc_down(adapter);
5126	igc_up(adapter);
5127	clear_bit(__IGC_RESETTING, &adapter->state);
5128}
5129
5130static void igc_reset_task(struct work_struct *work)
5131{
5132	struct igc_adapter *adapter;
5133
5134	adapter = container_of(work, struct igc_adapter, reset_task);
5135
5136	rtnl_lock();
5137	/* If we're already down or resetting, just bail */
5138	if (test_bit(__IGC_DOWN, &adapter->state) ||
5139	    test_bit(__IGC_RESETTING, &adapter->state)) {
5140		rtnl_unlock();
5141		return;
5142	}
5143
5144	igc_rings_dump(adapter);
5145	igc_regs_dump(adapter);
5146	netdev_err(adapter->netdev, "Reset adapter\n");
5147	igc_reinit_locked(adapter);
5148	rtnl_unlock();
5149}
5150
5151/**
5152 * igc_change_mtu - Change the Maximum Transfer Unit
5153 * @netdev: network interface device structure
5154 * @new_mtu: new value for maximum frame size
5155 *
5156 * Returns 0 on success, negative on failure
5157 */
5158static int igc_change_mtu(struct net_device *netdev, int new_mtu)
5159{
5160	int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
5161	struct igc_adapter *adapter = netdev_priv(netdev);
5162
5163	if (igc_xdp_is_enabled(adapter) && new_mtu > ETH_DATA_LEN) {
5164		netdev_dbg(netdev, "Jumbo frames not supported with XDP");
5165		return -EINVAL;
5166	}
5167
5168	/* adjust max frame to be at least the size of a standard frame */
5169	if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
5170		max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
5171
5172	while (test_and_set_bit(__IGC_RESETTING, &adapter->state))
5173		usleep_range(1000, 2000);
5174
5175	/* igc_down has a dependency on max_frame_size */
5176	adapter->max_frame_size = max_frame;
5177
5178	if (netif_running(netdev))
5179		igc_down(adapter);
5180
5181	netdev_dbg(netdev, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5182	netdev->mtu = new_mtu;
5183
5184	if (netif_running(netdev))
5185		igc_up(adapter);
5186	else
5187		igc_reset(adapter);
5188
5189	clear_bit(__IGC_RESETTING, &adapter->state);
5190
5191	return 0;
5192}
5193
5194/**
5195 * igc_tx_timeout - Respond to a Tx Hang
5196 * @netdev: network interface device structure
5197 * @txqueue: queue number that timed out
5198 **/
5199static void igc_tx_timeout(struct net_device *netdev,
5200			   unsigned int __always_unused txqueue)
5201{
5202	struct igc_adapter *adapter = netdev_priv(netdev);
5203	struct igc_hw *hw = &adapter->hw;
5204
5205	/* Do the reset outside of interrupt context */
5206	adapter->tx_timeout_count++;
5207	schedule_work(&adapter->reset_task);
5208	wr32(IGC_EICS,
5209	     (adapter->eims_enable_mask & ~adapter->eims_other));
5210}
5211
5212/**
5213 * igc_get_stats64 - Get System Network Statistics
5214 * @netdev: network interface device structure
5215 * @stats: rtnl_link_stats64 pointer
5216 *
5217 * Returns the address of the device statistics structure.
5218 * The statistics are updated here and also from the timer callback.
5219 */
5220static void igc_get_stats64(struct net_device *netdev,
5221			    struct rtnl_link_stats64 *stats)
5222{
5223	struct igc_adapter *adapter = netdev_priv(netdev);
5224
5225	spin_lock(&adapter->stats64_lock);
5226	if (!test_bit(__IGC_RESETTING, &adapter->state))
5227		igc_update_stats(adapter);
5228	memcpy(stats, &adapter->stats64, sizeof(*stats));
5229	spin_unlock(&adapter->stats64_lock);
 
5230}
5231
5232static netdev_features_t igc_fix_features(struct net_device *netdev,
5233					  netdev_features_t features)
5234{
5235	/* Since there is no support for separate Rx/Tx vlan accel
5236	 * enable/disable make sure Tx flag is always in same state as Rx.
5237	 */
5238	if (features & NETIF_F_HW_VLAN_CTAG_RX)
5239		features |= NETIF_F_HW_VLAN_CTAG_TX;
5240	else
5241		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
5242
5243	return features;
5244}
5245
5246static int igc_set_features(struct net_device *netdev,
5247			    netdev_features_t features)
5248{
5249	netdev_features_t changed = netdev->features ^ features;
5250	struct igc_adapter *adapter = netdev_priv(netdev);
5251
5252	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
5253		igc_vlan_mode(netdev, features);
5254
5255	/* Add VLAN support */
5256	if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
5257		return 0;
5258
5259	if (!(features & NETIF_F_NTUPLE))
5260		igc_flush_nfc_rules(adapter);
5261
5262	netdev->features = features;
5263
5264	if (netif_running(netdev))
5265		igc_reinit_locked(adapter);
5266	else
5267		igc_reset(adapter);
5268
5269	return 1;
5270}
5271
5272static netdev_features_t
5273igc_features_check(struct sk_buff *skb, struct net_device *dev,
5274		   netdev_features_t features)
5275{
5276	unsigned int network_hdr_len, mac_hdr_len;
5277
5278	/* Make certain the headers can be described by a context descriptor */
5279	mac_hdr_len = skb_network_header(skb) - skb->data;
5280	if (unlikely(mac_hdr_len > IGC_MAX_MAC_HDR_LEN))
5281		return features & ~(NETIF_F_HW_CSUM |
5282				    NETIF_F_SCTP_CRC |
5283				    NETIF_F_HW_VLAN_CTAG_TX |
5284				    NETIF_F_TSO |
5285				    NETIF_F_TSO6);
5286
5287	network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
5288	if (unlikely(network_hdr_len >  IGC_MAX_NETWORK_HDR_LEN))
5289		return features & ~(NETIF_F_HW_CSUM |
5290				    NETIF_F_SCTP_CRC |
5291				    NETIF_F_TSO |
5292				    NETIF_F_TSO6);
5293
5294	/* We can only support IPv4 TSO in tunnels if we can mangle the
5295	 * inner IP ID field, so strip TSO if MANGLEID is not supported.
5296	 */
5297	if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
5298		features &= ~NETIF_F_TSO;
5299
5300	return features;
5301}
5302
5303static void igc_tsync_interrupt(struct igc_adapter *adapter)
5304{
5305	u32 ack, tsauxc, sec, nsec, tsicr;
5306	struct igc_hw *hw = &adapter->hw;
5307	struct ptp_clock_event event;
5308	struct timespec64 ts;
5309
5310	tsicr = rd32(IGC_TSICR);
5311	ack = 0;
5312
5313	if (tsicr & IGC_TSICR_SYS_WRAP) {
5314		event.type = PTP_CLOCK_PPS;
5315		if (adapter->ptp_caps.pps)
5316			ptp_clock_event(adapter->ptp_clock, &event);
5317		ack |= IGC_TSICR_SYS_WRAP;
5318	}
5319
5320	if (tsicr & IGC_TSICR_TXTS) {
5321		/* retrieve hardware timestamp */
5322		igc_ptp_tx_tstamp_event(adapter);
5323		ack |= IGC_TSICR_TXTS;
5324	}
5325
5326	if (tsicr & IGC_TSICR_TT0) {
5327		spin_lock(&adapter->tmreg_lock);
5328		ts = timespec64_add(adapter->perout[0].start,
5329				    adapter->perout[0].period);
5330		wr32(IGC_TRGTTIML0, ts.tv_nsec | IGC_TT_IO_TIMER_SEL_SYSTIM0);
5331		wr32(IGC_TRGTTIMH0, (u32)ts.tv_sec);
5332		tsauxc = rd32(IGC_TSAUXC);
5333		tsauxc |= IGC_TSAUXC_EN_TT0;
5334		wr32(IGC_TSAUXC, tsauxc);
5335		adapter->perout[0].start = ts;
5336		spin_unlock(&adapter->tmreg_lock);
5337		ack |= IGC_TSICR_TT0;
5338	}
5339
5340	if (tsicr & IGC_TSICR_TT1) {
5341		spin_lock(&adapter->tmreg_lock);
5342		ts = timespec64_add(adapter->perout[1].start,
5343				    adapter->perout[1].period);
5344		wr32(IGC_TRGTTIML1, ts.tv_nsec | IGC_TT_IO_TIMER_SEL_SYSTIM0);
5345		wr32(IGC_TRGTTIMH1, (u32)ts.tv_sec);
5346		tsauxc = rd32(IGC_TSAUXC);
5347		tsauxc |= IGC_TSAUXC_EN_TT1;
5348		wr32(IGC_TSAUXC, tsauxc);
5349		adapter->perout[1].start = ts;
5350		spin_unlock(&adapter->tmreg_lock);
5351		ack |= IGC_TSICR_TT1;
5352	}
5353
5354	if (tsicr & IGC_TSICR_AUTT0) {
5355		nsec = rd32(IGC_AUXSTMPL0);
5356		sec  = rd32(IGC_AUXSTMPH0);
5357		event.type = PTP_CLOCK_EXTTS;
5358		event.index = 0;
5359		event.timestamp = sec * NSEC_PER_SEC + nsec;
5360		ptp_clock_event(adapter->ptp_clock, &event);
5361		ack |= IGC_TSICR_AUTT0;
5362	}
5363
5364	if (tsicr & IGC_TSICR_AUTT1) {
5365		nsec = rd32(IGC_AUXSTMPL1);
5366		sec  = rd32(IGC_AUXSTMPH1);
5367		event.type = PTP_CLOCK_EXTTS;
5368		event.index = 1;
5369		event.timestamp = sec * NSEC_PER_SEC + nsec;
5370		ptp_clock_event(adapter->ptp_clock, &event);
5371		ack |= IGC_TSICR_AUTT1;
5372	}
5373
5374	/* acknowledge the interrupts */
5375	wr32(IGC_TSICR, ack);
5376}
5377
5378/**
5379 * igc_msix_other - msix other interrupt handler
5380 * @irq: interrupt number
5381 * @data: pointer to a q_vector
5382 */
5383static irqreturn_t igc_msix_other(int irq, void *data)
5384{
5385	struct igc_adapter *adapter = data;
5386	struct igc_hw *hw = &adapter->hw;
5387	u32 icr = rd32(IGC_ICR);
5388
5389	/* reading ICR causes bit 31 of EICR to be cleared */
5390	if (icr & IGC_ICR_DRSTA)
5391		schedule_work(&adapter->reset_task);
5392
5393	if (icr & IGC_ICR_DOUTSYNC) {
5394		/* HW is reporting DMA is out of sync */
5395		adapter->stats.doosync++;
5396	}
5397
5398	if (icr & IGC_ICR_LSC) {
5399		hw->mac.get_link_status = true;
5400		/* guard against interrupt when we're going down */
5401		if (!test_bit(__IGC_DOWN, &adapter->state))
5402			mod_timer(&adapter->watchdog_timer, jiffies + 1);
5403	}
5404
5405	if (icr & IGC_ICR_TS)
5406		igc_tsync_interrupt(adapter);
5407
5408	wr32(IGC_EIMS, adapter->eims_other);
5409
5410	return IRQ_HANDLED;
5411}
5412
5413static void igc_write_itr(struct igc_q_vector *q_vector)
5414{
5415	u32 itr_val = q_vector->itr_val & IGC_QVECTOR_MASK;
5416
5417	if (!q_vector->set_itr)
5418		return;
5419
5420	if (!itr_val)
5421		itr_val = IGC_ITR_VAL_MASK;
5422
5423	itr_val |= IGC_EITR_CNT_IGNR;
5424
5425	writel(itr_val, q_vector->itr_register);
5426	q_vector->set_itr = 0;
5427}
5428
5429static irqreturn_t igc_msix_ring(int irq, void *data)
5430{
5431	struct igc_q_vector *q_vector = data;
5432
5433	/* Write the ITR value calculated from the previous interrupt. */
5434	igc_write_itr(q_vector);
5435
5436	napi_schedule(&q_vector->napi);
5437
5438	return IRQ_HANDLED;
5439}
5440
5441/**
5442 * igc_request_msix - Initialize MSI-X interrupts
5443 * @adapter: Pointer to adapter structure
5444 *
5445 * igc_request_msix allocates MSI-X vectors and requests interrupts from the
5446 * kernel.
5447 */
5448static int igc_request_msix(struct igc_adapter *adapter)
5449{
5450	unsigned int num_q_vectors = adapter->num_q_vectors;
5451	int i = 0, err = 0, vector = 0, free_vector = 0;
5452	struct net_device *netdev = adapter->netdev;
5453
5454	err = request_irq(adapter->msix_entries[vector].vector,
5455			  &igc_msix_other, 0, netdev->name, adapter);
5456	if (err)
5457		goto err_out;
5458
5459	if (num_q_vectors > MAX_Q_VECTORS) {
5460		num_q_vectors = MAX_Q_VECTORS;
5461		dev_warn(&adapter->pdev->dev,
5462			 "The number of queue vectors (%d) is higher than max allowed (%d)\n",
5463			 adapter->num_q_vectors, MAX_Q_VECTORS);
5464	}
5465	for (i = 0; i < num_q_vectors; i++) {
5466		struct igc_q_vector *q_vector = adapter->q_vector[i];
5467
5468		vector++;
5469
5470		q_vector->itr_register = adapter->io_addr + IGC_EITR(vector);
5471
5472		if (q_vector->rx.ring && q_vector->tx.ring)
5473			sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
5474				q_vector->rx.ring->queue_index);
5475		else if (q_vector->tx.ring)
5476			sprintf(q_vector->name, "%s-tx-%u", netdev->name,
5477				q_vector->tx.ring->queue_index);
5478		else if (q_vector->rx.ring)
5479			sprintf(q_vector->name, "%s-rx-%u", netdev->name,
5480				q_vector->rx.ring->queue_index);
5481		else
5482			sprintf(q_vector->name, "%s-unused", netdev->name);
5483
5484		err = request_irq(adapter->msix_entries[vector].vector,
5485				  igc_msix_ring, 0, q_vector->name,
5486				  q_vector);
5487		if (err)
5488			goto err_free;
5489	}
5490
5491	igc_configure_msix(adapter);
5492	return 0;
5493
5494err_free:
5495	/* free already assigned IRQs */
5496	free_irq(adapter->msix_entries[free_vector++].vector, adapter);
5497
5498	vector--;
5499	for (i = 0; i < vector; i++) {
5500		free_irq(adapter->msix_entries[free_vector++].vector,
5501			 adapter->q_vector[i]);
5502	}
5503err_out:
5504	return err;
5505}
5506
5507/**
5508 * igc_clear_interrupt_scheme - reset the device to a state of no interrupts
5509 * @adapter: Pointer to adapter structure
5510 *
5511 * This function resets the device so that it has 0 rx queues, tx queues, and
5512 * MSI-X interrupts allocated.
5513 */
5514static void igc_clear_interrupt_scheme(struct igc_adapter *adapter)
5515{
5516	igc_free_q_vectors(adapter);
5517	igc_reset_interrupt_capability(adapter);
5518}
5519
5520/* Need to wait a few seconds after link up to get diagnostic information from
5521 * the phy
5522 */
5523static void igc_update_phy_info(struct timer_list *t)
5524{
5525	struct igc_adapter *adapter = from_timer(adapter, t, phy_info_timer);
5526
5527	igc_get_phy_info(&adapter->hw);
5528}
5529
5530/**
5531 * igc_has_link - check shared code for link and determine up/down
5532 * @adapter: pointer to driver private info
5533 */
5534bool igc_has_link(struct igc_adapter *adapter)
5535{
5536	struct igc_hw *hw = &adapter->hw;
5537	bool link_active = false;
5538
5539	/* get_link_status is set on LSC (link status) interrupt or
5540	 * rx sequence error interrupt.  get_link_status will stay
5541	 * false until the igc_check_for_link establishes link
5542	 * for copper adapters ONLY
5543	 */
5544	if (!hw->mac.get_link_status)
5545		return true;
5546	hw->mac.ops.check_for_link(hw);
5547	link_active = !hw->mac.get_link_status;
 
 
 
 
 
 
 
5548
5549	if (hw->mac.type == igc_i225) {
 
5550		if (!netif_carrier_ok(adapter->netdev)) {
5551			adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
5552		} else if (!(adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)) {
5553			adapter->flags |= IGC_FLAG_NEED_LINK_UPDATE;
5554			adapter->link_check_timeout = jiffies;
5555		}
5556	}
5557
5558	return link_active;
5559}
5560
5561/**
5562 * igc_watchdog - Timer Call-back
5563 * @t: timer for the watchdog
5564 */
5565static void igc_watchdog(struct timer_list *t)
5566{
5567	struct igc_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5568	/* Do the rest outside of interrupt context */
5569	schedule_work(&adapter->watchdog_task);
5570}
5571
5572static void igc_watchdog_task(struct work_struct *work)
5573{
5574	struct igc_adapter *adapter = container_of(work,
5575						   struct igc_adapter,
5576						   watchdog_task);
5577	struct net_device *netdev = adapter->netdev;
5578	struct igc_hw *hw = &adapter->hw;
5579	struct igc_phy_info *phy = &hw->phy;
5580	u16 phy_data, retry_count = 20;
5581	u32 link;
5582	int i;
5583
5584	link = igc_has_link(adapter);
5585
5586	if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE) {
5587		if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
5588			adapter->flags &= ~IGC_FLAG_NEED_LINK_UPDATE;
5589		else
5590			link = false;
5591	}
5592
5593	if (link) {
5594		/* Cancel scheduled suspend requests. */
5595		pm_runtime_resume(netdev->dev.parent);
5596
5597		if (!netif_carrier_ok(netdev)) {
5598			u32 ctrl;
5599
5600			hw->mac.ops.get_speed_and_duplex(hw,
5601							 &adapter->link_speed,
5602							 &adapter->link_duplex);
5603
5604			ctrl = rd32(IGC_CTRL);
5605			/* Link status message must follow this format */
5606			netdev_info(netdev,
5607				    "NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5608				    adapter->link_speed,
5609				    adapter->link_duplex == FULL_DUPLEX ?
5610				    "Full" : "Half",
5611				    (ctrl & IGC_CTRL_TFCE) &&
5612				    (ctrl & IGC_CTRL_RFCE) ? "RX/TX" :
5613				    (ctrl & IGC_CTRL_RFCE) ?  "RX" :
5614				    (ctrl & IGC_CTRL_TFCE) ?  "TX" : "None");
5615
5616			/* disable EEE if enabled */
5617			if ((adapter->flags & IGC_FLAG_EEE) &&
5618			    adapter->link_duplex == HALF_DUPLEX) {
5619				netdev_info(netdev,
5620					    "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex\n");
5621				adapter->hw.dev_spec._base.eee_enable = false;
5622				adapter->flags &= ~IGC_FLAG_EEE;
5623			}
5624
5625			/* check if SmartSpeed worked */
5626			igc_check_downshift(hw);
5627			if (phy->speed_downgraded)
5628				netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
5629
5630			/* adjust timeout factor according to speed/duplex */
5631			adapter->tx_timeout_factor = 1;
5632			switch (adapter->link_speed) {
5633			case SPEED_10:
5634				adapter->tx_timeout_factor = 14;
5635				break;
5636			case SPEED_100:
5637			case SPEED_1000:
5638			case SPEED_2500:
5639				adapter->tx_timeout_factor = 1;
5640				break;
5641			}
5642
5643			/* Once the launch time has been set on the wire, there
5644			 * is a delay before the link speed can be determined
5645			 * based on link-up activity. Write into the register
5646			 * as soon as we know the correct link speed.
5647			 */
5648			igc_tsn_adjust_txtime_offset(adapter);
5649
5650			if (adapter->link_speed != SPEED_1000)
5651				goto no_wait;
5652
5653			/* wait for Remote receiver status OK */
5654retry_read_status:
5655			if (!igc_read_phy_reg(hw, PHY_1000T_STATUS,
5656					      &phy_data)) {
5657				if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5658				    retry_count) {
5659					msleep(100);
5660					retry_count--;
5661					goto retry_read_status;
5662				} else if (!retry_count) {
5663					netdev_err(netdev, "exceed max 2 second\n");
5664				}
5665			} else {
5666				netdev_err(netdev, "read 1000Base-T Status Reg\n");
5667			}
5668no_wait:
5669			netif_carrier_on(netdev);
5670
5671			/* link state has changed, schedule phy info update */
5672			if (!test_bit(__IGC_DOWN, &adapter->state))
5673				mod_timer(&adapter->phy_info_timer,
5674					  round_jiffies(jiffies + 2 * HZ));
5675		}
5676	} else {
5677		if (netif_carrier_ok(netdev)) {
5678			adapter->link_speed = 0;
5679			adapter->link_duplex = 0;
5680
5681			/* Links status message must follow this format */
5682			netdev_info(netdev, "NIC Link is Down\n");
5683			netif_carrier_off(netdev);
5684
5685			/* link state has changed, schedule phy info update */
5686			if (!test_bit(__IGC_DOWN, &adapter->state))
5687				mod_timer(&adapter->phy_info_timer,
5688					  round_jiffies(jiffies + 2 * HZ));
5689
 
 
 
 
 
 
 
 
5690			pm_schedule_suspend(netdev->dev.parent,
5691					    MSEC_PER_SEC * 5);
 
 
 
 
 
 
 
 
 
5692		}
5693	}
5694
5695	spin_lock(&adapter->stats64_lock);
5696	igc_update_stats(adapter);
5697	spin_unlock(&adapter->stats64_lock);
5698
5699	for (i = 0; i < adapter->num_tx_queues; i++) {
5700		struct igc_ring *tx_ring = adapter->tx_ring[i];
5701
5702		if (!netif_carrier_ok(netdev)) {
5703			/* We've lost link, so the controller stops DMA,
5704			 * but we've got queued Tx work that's never going
5705			 * to get done, so reset controller to flush Tx.
5706			 * (Do the reset outside of interrupt context).
5707			 */
5708			if (igc_desc_unused(tx_ring) + 1 < tx_ring->count) {
5709				adapter->tx_timeout_count++;
5710				schedule_work(&adapter->reset_task);
5711				/* return immediately since reset is imminent */
5712				return;
5713			}
5714		}
5715
5716		/* Force detection of hung controller every watchdog period */
5717		set_bit(IGC_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5718	}
5719
5720	/* Cause software interrupt to ensure Rx ring is cleaned */
5721	if (adapter->flags & IGC_FLAG_HAS_MSIX) {
5722		u32 eics = 0;
5723
5724		for (i = 0; i < adapter->num_q_vectors; i++)
5725			eics |= adapter->q_vector[i]->eims_value;
5726		wr32(IGC_EICS, eics);
5727	} else {
5728		wr32(IGC_ICS, IGC_ICS_RXDMT0);
5729	}
5730
5731	igc_ptp_tx_hang(adapter);
5732
5733	/* Reset the timer */
5734	if (!test_bit(__IGC_DOWN, &adapter->state)) {
5735		if (adapter->flags & IGC_FLAG_NEED_LINK_UPDATE)
5736			mod_timer(&adapter->watchdog_timer,
5737				  round_jiffies(jiffies +  HZ));
5738		else
5739			mod_timer(&adapter->watchdog_timer,
5740				  round_jiffies(jiffies + 2 * HZ));
5741	}
5742}
5743
5744/**
5745 * igc_intr_msi - Interrupt Handler
5746 * @irq: interrupt number
5747 * @data: pointer to a network interface device structure
5748 */
5749static irqreturn_t igc_intr_msi(int irq, void *data)
5750{
5751	struct igc_adapter *adapter = data;
5752	struct igc_q_vector *q_vector = adapter->q_vector[0];
5753	struct igc_hw *hw = &adapter->hw;
5754	/* read ICR disables interrupts using IAM */
5755	u32 icr = rd32(IGC_ICR);
5756
5757	igc_write_itr(q_vector);
5758
5759	if (icr & IGC_ICR_DRSTA)
5760		schedule_work(&adapter->reset_task);
5761
5762	if (icr & IGC_ICR_DOUTSYNC) {
5763		/* HW is reporting DMA is out of sync */
5764		adapter->stats.doosync++;
5765	}
5766
5767	if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) {
5768		hw->mac.get_link_status = true;
5769		if (!test_bit(__IGC_DOWN, &adapter->state))
5770			mod_timer(&adapter->watchdog_timer, jiffies + 1);
5771	}
5772
5773	if (icr & IGC_ICR_TS)
5774		igc_tsync_interrupt(adapter);
5775
5776	napi_schedule(&q_vector->napi);
5777
5778	return IRQ_HANDLED;
5779}
5780
5781/**
5782 * igc_intr - Legacy Interrupt Handler
5783 * @irq: interrupt number
5784 * @data: pointer to a network interface device structure
5785 */
5786static irqreturn_t igc_intr(int irq, void *data)
5787{
5788	struct igc_adapter *adapter = data;
5789	struct igc_q_vector *q_vector = adapter->q_vector[0];
5790	struct igc_hw *hw = &adapter->hw;
5791	/* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
5792	 * need for the IMC write
5793	 */
5794	u32 icr = rd32(IGC_ICR);
5795
5796	/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5797	 * not set, then the adapter didn't send an interrupt
5798	 */
5799	if (!(icr & IGC_ICR_INT_ASSERTED))
5800		return IRQ_NONE;
5801
5802	igc_write_itr(q_vector);
5803
5804	if (icr & IGC_ICR_DRSTA)
5805		schedule_work(&adapter->reset_task);
5806
5807	if (icr & IGC_ICR_DOUTSYNC) {
5808		/* HW is reporting DMA is out of sync */
5809		adapter->stats.doosync++;
5810	}
5811
5812	if (icr & (IGC_ICR_RXSEQ | IGC_ICR_LSC)) {
5813		hw->mac.get_link_status = true;
5814		/* guard against interrupt when we're going down */
5815		if (!test_bit(__IGC_DOWN, &adapter->state))
5816			mod_timer(&adapter->watchdog_timer, jiffies + 1);
5817	}
5818
5819	if (icr & IGC_ICR_TS)
5820		igc_tsync_interrupt(adapter);
5821
5822	napi_schedule(&q_vector->napi);
5823
5824	return IRQ_HANDLED;
5825}
5826
5827static void igc_free_irq(struct igc_adapter *adapter)
5828{
5829	if (adapter->msix_entries) {
5830		int vector = 0, i;
5831
5832		free_irq(adapter->msix_entries[vector++].vector, adapter);
5833
5834		for (i = 0; i < adapter->num_q_vectors; i++)
5835			free_irq(adapter->msix_entries[vector++].vector,
5836				 adapter->q_vector[i]);
5837	} else {
5838		free_irq(adapter->pdev->irq, adapter);
5839	}
5840}
5841
5842/**
5843 * igc_request_irq - initialize interrupts
5844 * @adapter: Pointer to adapter structure
5845 *
5846 * Attempts to configure interrupts using the best available
5847 * capabilities of the hardware and kernel.
5848 */
5849static int igc_request_irq(struct igc_adapter *adapter)
5850{
5851	struct net_device *netdev = adapter->netdev;
5852	struct pci_dev *pdev = adapter->pdev;
5853	int err = 0;
5854
5855	if (adapter->flags & IGC_FLAG_HAS_MSIX) {
5856		err = igc_request_msix(adapter);
5857		if (!err)
5858			goto request_done;
5859		/* fall back to MSI */
5860		igc_free_all_tx_resources(adapter);
5861		igc_free_all_rx_resources(adapter);
5862
5863		igc_clear_interrupt_scheme(adapter);
5864		err = igc_init_interrupt_scheme(adapter, false);
5865		if (err)
5866			goto request_done;
5867		igc_setup_all_tx_resources(adapter);
5868		igc_setup_all_rx_resources(adapter);
5869		igc_configure(adapter);
5870	}
5871
5872	igc_assign_vector(adapter->q_vector[0], 0);
5873
5874	if (adapter->flags & IGC_FLAG_HAS_MSI) {
5875		err = request_irq(pdev->irq, &igc_intr_msi, 0,
5876				  netdev->name, adapter);
5877		if (!err)
5878			goto request_done;
5879
5880		/* fall back to legacy interrupts */
5881		igc_reset_interrupt_capability(adapter);
5882		adapter->flags &= ~IGC_FLAG_HAS_MSI;
5883	}
5884
5885	err = request_irq(pdev->irq, &igc_intr, IRQF_SHARED,
5886			  netdev->name, adapter);
5887
5888	if (err)
5889		netdev_err(netdev, "Error %d getting interrupt\n", err);
5890
5891request_done:
5892	return err;
5893}
5894
5895/**
5896 * __igc_open - Called when a network interface is made active
5897 * @netdev: network interface device structure
5898 * @resuming: boolean indicating if the device is resuming
5899 *
5900 * Returns 0 on success, negative value on failure
5901 *
5902 * The open entry point is called when a network interface is made
5903 * active by the system (IFF_UP).  At this point all resources needed
5904 * for transmit and receive operations are allocated, the interrupt
5905 * handler is registered with the OS, the watchdog timer is started,
5906 * and the stack is notified that the interface is ready.
5907 */
5908static int __igc_open(struct net_device *netdev, bool resuming)
5909{
5910	struct igc_adapter *adapter = netdev_priv(netdev);
5911	struct pci_dev *pdev = adapter->pdev;
5912	struct igc_hw *hw = &adapter->hw;
5913	int err = 0;
5914	int i = 0;
5915
5916	/* disallow open during test */
5917
5918	if (test_bit(__IGC_TESTING, &adapter->state)) {
5919		WARN_ON(resuming);
5920		return -EBUSY;
5921	}
5922
5923	if (!resuming)
5924		pm_runtime_get_sync(&pdev->dev);
5925
5926	netif_carrier_off(netdev);
5927
5928	/* allocate transmit descriptors */
5929	err = igc_setup_all_tx_resources(adapter);
5930	if (err)
5931		goto err_setup_tx;
5932
5933	/* allocate receive descriptors */
5934	err = igc_setup_all_rx_resources(adapter);
5935	if (err)
5936		goto err_setup_rx;
5937
5938	igc_power_up_link(adapter);
5939
5940	igc_configure(adapter);
5941
5942	err = igc_request_irq(adapter);
5943	if (err)
5944		goto err_req_irq;
5945
5946	/* Notify the stack of the actual queue counts. */
5947	err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
5948	if (err)
5949		goto err_set_queues;
5950
5951	err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
5952	if (err)
5953		goto err_set_queues;
5954
5955	clear_bit(__IGC_DOWN, &adapter->state);
5956
5957	for (i = 0; i < adapter->num_q_vectors; i++)
5958		napi_enable(&adapter->q_vector[i]->napi);
5959
5960	/* Clear any pending interrupts. */
5961	rd32(IGC_ICR);
5962	igc_irq_enable(adapter);
5963
5964	if (!resuming)
5965		pm_runtime_put(&pdev->dev);
5966
5967	netif_tx_start_all_queues(netdev);
5968
5969	/* start the watchdog. */
5970	hw->mac.get_link_status = true;
5971	schedule_work(&adapter->watchdog_task);
5972
5973	return IGC_SUCCESS;
5974
5975err_set_queues:
5976	igc_free_irq(adapter);
5977err_req_irq:
5978	igc_release_hw_control(adapter);
5979	igc_power_down_phy_copper_base(&adapter->hw);
5980	igc_free_all_rx_resources(adapter);
5981err_setup_rx:
5982	igc_free_all_tx_resources(adapter);
5983err_setup_tx:
5984	igc_reset(adapter);
5985	if (!resuming)
5986		pm_runtime_put(&pdev->dev);
5987
5988	return err;
5989}
5990
5991int igc_open(struct net_device *netdev)
5992{
5993	return __igc_open(netdev, false);
5994}
5995
5996/**
5997 * __igc_close - Disables a network interface
5998 * @netdev: network interface device structure
5999 * @suspending: boolean indicating the device is suspending
6000 *
6001 * Returns 0, this is not allowed to fail
6002 *
6003 * The close entry point is called when an interface is de-activated
6004 * by the OS.  The hardware is still under the driver's control, but
6005 * needs to be disabled.  A global MAC reset is issued to stop the
6006 * hardware, and all transmit and receive resources are freed.
6007 */
6008static int __igc_close(struct net_device *netdev, bool suspending)
6009{
6010	struct igc_adapter *adapter = netdev_priv(netdev);
6011	struct pci_dev *pdev = adapter->pdev;
6012
6013	WARN_ON(test_bit(__IGC_RESETTING, &adapter->state));
6014
6015	if (!suspending)
6016		pm_runtime_get_sync(&pdev->dev);
6017
6018	igc_down(adapter);
6019
6020	igc_release_hw_control(adapter);
6021
6022	igc_free_irq(adapter);
6023
6024	igc_free_all_tx_resources(adapter);
6025	igc_free_all_rx_resources(adapter);
6026
6027	if (!suspending)
6028		pm_runtime_put_sync(&pdev->dev);
6029
6030	return 0;
6031}
6032
6033int igc_close(struct net_device *netdev)
6034{
6035	if (netif_device_present(netdev) || netdev->dismantle)
6036		return __igc_close(netdev, false);
6037	return 0;
6038}
6039
6040/**
6041 * igc_ioctl - Access the hwtstamp interface
6042 * @netdev: network interface device structure
6043 * @ifr: interface request data
6044 * @cmd: ioctl command
6045 **/
6046static int igc_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6047{
6048	switch (cmd) {
6049	case SIOCGHWTSTAMP:
6050		return igc_ptp_get_ts_config(netdev, ifr);
6051	case SIOCSHWTSTAMP:
6052		return igc_ptp_set_ts_config(netdev, ifr);
6053	default:
6054		return -EOPNOTSUPP;
6055	}
6056}
6057
6058static int igc_save_launchtime_params(struct igc_adapter *adapter, int queue,
6059				      bool enable)
6060{
6061	struct igc_ring *ring;
 
6062
6063	if (queue < 0 || queue >= adapter->num_tx_queues)
6064		return -EINVAL;
6065
6066	ring = adapter->tx_ring[queue];
6067	ring->launchtime_enable = enable;
6068
6069	return 0;
6070}
6071
6072static bool is_base_time_past(ktime_t base_time, const struct timespec64 *now)
6073{
6074	struct timespec64 b;
6075
6076	b = ktime_to_timespec64(base_time);
 
 
 
 
6077
6078	return timespec64_compare(now, &b) > 0;
6079}
6080
6081static bool validate_schedule(struct igc_adapter *adapter,
6082			      const struct tc_taprio_qopt_offload *qopt)
6083{
6084	int queue_uses[IGC_MAX_TX_QUEUES] = { };
6085	struct igc_hw *hw = &adapter->hw;
6086	struct timespec64 now;
6087	size_t n;
6088
6089	if (qopt->cycle_time_extension)
6090		return false;
6091
6092	igc_ptp_read(adapter, &now);
6093
6094	/* If we program the controller's BASET registers with a time
6095	 * in the future, it will hold all the packets until that
6096	 * time, causing a lot of TX Hangs, so to avoid that, we
6097	 * reject schedules that would start in the future.
6098	 * Note: Limitation above is no longer in i226.
6099	 */
6100	if (!is_base_time_past(qopt->base_time, &now) &&
6101	    igc_is_device_id_i225(hw))
6102		return false;
6103
6104	for (n = 0; n < qopt->num_entries; n++) {
6105		const struct tc_taprio_sched_entry *e, *prev;
6106		int i;
6107
6108		prev = n ? &qopt->entries[n - 1] : NULL;
6109		e = &qopt->entries[n];
6110
6111		/* i225 only supports "global" frame preemption
6112		 * settings.
6113		 */
6114		if (e->command != TC_TAPRIO_CMD_SET_GATES)
6115			return false;
6116
6117		for (i = 0; i < adapter->num_tx_queues; i++)
6118			if (e->gate_mask & BIT(i)) {
6119				queue_uses[i]++;
6120
6121				/* There are limitations: A single queue cannot
6122				 * be opened and closed multiple times per cycle
6123				 * unless the gate stays open. Check for it.
6124				 */
6125				if (queue_uses[i] > 1 &&
6126				    !(prev->gate_mask & BIT(i)))
6127					return false;
6128			}
6129	}
6130
6131	return true;
6132}
6133
6134static int igc_tsn_enable_launchtime(struct igc_adapter *adapter,
6135				     struct tc_etf_qopt_offload *qopt)
6136{
6137	struct igc_hw *hw = &adapter->hw;
6138	int err;
6139
6140	if (hw->mac.type != igc_i225)
6141		return -EOPNOTSUPP;
6142
6143	err = igc_save_launchtime_params(adapter, qopt->queue, qopt->enable);
6144	if (err)
6145		return err;
6146
6147	return igc_tsn_offload_apply(adapter);
6148}
6149
6150static int igc_qbv_clear_schedule(struct igc_adapter *adapter)
6151{
6152	unsigned long flags;
6153	int i;
6154
6155	adapter->base_time = 0;
6156	adapter->cycle_time = NSEC_PER_SEC;
6157	adapter->taprio_offload_enable = false;
6158	adapter->qbv_config_change_errors = 0;
6159	adapter->qbv_count = 0;
6160
6161	for (i = 0; i < adapter->num_tx_queues; i++) {
6162		struct igc_ring *ring = adapter->tx_ring[i];
6163
6164		ring->start_time = 0;
6165		ring->end_time = NSEC_PER_SEC;
6166		ring->max_sdu = 0;
6167	}
6168
6169	spin_lock_irqsave(&adapter->qbv_tx_lock, flags);
6170
6171	adapter->qbv_transition = false;
6172
6173	for (i = 0; i < adapter->num_tx_queues; i++) {
6174		struct igc_ring *ring = adapter->tx_ring[i];
6175
6176		ring->oper_gate_closed = false;
6177		ring->admin_gate_closed = false;
6178	}
6179
6180	spin_unlock_irqrestore(&adapter->qbv_tx_lock, flags);
6181
6182	return 0;
6183}
6184
6185static int igc_tsn_clear_schedule(struct igc_adapter *adapter)
6186{
6187	igc_qbv_clear_schedule(adapter);
6188
6189	return 0;
6190}
6191
6192static void igc_taprio_stats(struct net_device *dev,
6193			     struct tc_taprio_qopt_stats *stats)
6194{
6195	/* When Strict_End is enabled, the tx_overruns counter
6196	 * will always be zero.
6197	 */
6198	stats->tx_overruns = 0;
6199}
6200
6201static void igc_taprio_queue_stats(struct net_device *dev,
6202				   struct tc_taprio_qopt_queue_stats *queue_stats)
6203{
6204	struct tc_taprio_qopt_stats *stats = &queue_stats->stats;
6205
6206	/* When Strict_End is enabled, the tx_overruns counter
6207	 * will always be zero.
6208	 */
6209	stats->tx_overruns = 0;
6210}
6211
6212static int igc_save_qbv_schedule(struct igc_adapter *adapter,
6213				 struct tc_taprio_qopt_offload *qopt)
6214{
6215	bool queue_configured[IGC_MAX_TX_QUEUES] = { };
6216	struct igc_hw *hw = &adapter->hw;
6217	u32 start_time = 0, end_time = 0;
6218	struct timespec64 now;
6219	unsigned long flags;
6220	size_t n;
6221	int i;
6222
6223	switch (qopt->cmd) {
6224	case TAPRIO_CMD_REPLACE:
6225		break;
6226	case TAPRIO_CMD_DESTROY:
6227		return igc_tsn_clear_schedule(adapter);
6228	case TAPRIO_CMD_STATS:
6229		igc_taprio_stats(adapter->netdev, &qopt->stats);
6230		return 0;
6231	case TAPRIO_CMD_QUEUE_STATS:
6232		igc_taprio_queue_stats(adapter->netdev, &qopt->queue_stats);
6233		return 0;
6234	default:
6235		return -EOPNOTSUPP;
6236	}
6237
6238	if (qopt->base_time < 0)
6239		return -ERANGE;
6240
6241	if (igc_is_device_id_i225(hw) && adapter->taprio_offload_enable)
6242		return -EALREADY;
6243
6244	if (!validate_schedule(adapter, qopt))
6245		return -EINVAL;
6246
6247	adapter->cycle_time = qopt->cycle_time;
6248	adapter->base_time = qopt->base_time;
6249	adapter->taprio_offload_enable = true;
6250
6251	igc_ptp_read(adapter, &now);
6252
 
 
 
6253	for (n = 0; n < qopt->num_entries; n++) {
6254		struct tc_taprio_sched_entry *e = &qopt->entries[n];
 
6255
6256		end_time += e->interval;
6257
6258		/* If any of the conditions below are true, we need to manually
6259		 * control the end time of the cycle.
6260		 * 1. Qbv users can specify a cycle time that is not equal
6261		 * to the total GCL intervals. Hence, recalculation is
6262		 * necessary here to exclude the time interval that
6263		 * exceeds the cycle time.
6264		 * 2. According to IEEE Std. 802.1Q-2018 section 8.6.9.2,
6265		 * once the end of the list is reached, it will switch
6266		 * to the END_OF_CYCLE state and leave the gates in the
6267		 * same state until the next cycle is started.
6268		 */
6269		if (end_time > adapter->cycle_time ||
6270		    n + 1 == qopt->num_entries)
6271			end_time = adapter->cycle_time;
6272
6273		for (i = 0; i < adapter->num_tx_queues; i++) {
6274			struct igc_ring *ring = adapter->tx_ring[i];
6275
6276			if (!(e->gate_mask & BIT(i)))
6277				continue;
6278
6279			/* Check whether a queue stays open for more than one
6280			 * entry. If so, keep the start and advance the end
6281			 * time.
6282			 */
6283			if (!queue_configured[i])
6284				ring->start_time = start_time;
6285			ring->end_time = end_time;
6286
6287			if (ring->start_time >= adapter->cycle_time)
6288				queue_configured[i] = false;
6289			else
6290				queue_configured[i] = true;
6291		}
6292
6293		start_time += e->interval;
6294	}
6295
6296	spin_lock_irqsave(&adapter->qbv_tx_lock, flags);
6297
6298	/* Check whether a queue gets configured.
6299	 * If not, set the start and end time to be end time.
6300	 */
6301	for (i = 0; i < adapter->num_tx_queues; i++) {
6302		struct igc_ring *ring = adapter->tx_ring[i];
6303
6304		if (!is_base_time_past(qopt->base_time, &now)) {
6305			ring->admin_gate_closed = false;
6306		} else {
6307			ring->oper_gate_closed = false;
6308			ring->admin_gate_closed = false;
6309		}
6310
6311		if (!queue_configured[i]) {
6312			if (!is_base_time_past(qopt->base_time, &now))
6313				ring->admin_gate_closed = true;
6314			else
6315				ring->oper_gate_closed = true;
6316
6317			ring->start_time = end_time;
6318			ring->end_time = end_time;
6319		}
6320	}
6321
6322	spin_unlock_irqrestore(&adapter->qbv_tx_lock, flags);
6323
6324	for (i = 0; i < adapter->num_tx_queues; i++) {
6325		struct igc_ring *ring = adapter->tx_ring[i];
6326		struct net_device *dev = adapter->netdev;
6327
6328		if (qopt->max_sdu[i])
6329			ring->max_sdu = qopt->max_sdu[i] + dev->hard_header_len - ETH_TLEN;
6330		else
6331			ring->max_sdu = 0;
6332	}
6333
6334	return 0;
6335}
6336
6337static int igc_tsn_enable_qbv_scheduling(struct igc_adapter *adapter,
6338					 struct tc_taprio_qopt_offload *qopt)
6339{
6340	struct igc_hw *hw = &adapter->hw;
6341	int err;
6342
6343	if (hw->mac.type != igc_i225)
6344		return -EOPNOTSUPP;
6345
6346	err = igc_save_qbv_schedule(adapter, qopt);
6347	if (err)
6348		return err;
6349
6350	return igc_tsn_offload_apply(adapter);
6351}
6352
6353static int igc_save_cbs_params(struct igc_adapter *adapter, int queue,
6354			       bool enable, int idleslope, int sendslope,
6355			       int hicredit, int locredit)
6356{
6357	bool cbs_status[IGC_MAX_SR_QUEUES] = { false };
6358	struct net_device *netdev = adapter->netdev;
6359	struct igc_ring *ring;
6360	int i;
6361
6362	/* i225 has two sets of credit-based shaper logic.
6363	 * Supporting it only on the top two priority queues
6364	 */
6365	if (queue < 0 || queue > 1)
6366		return -EINVAL;
6367
6368	ring = adapter->tx_ring[queue];
6369
6370	for (i = 0; i < IGC_MAX_SR_QUEUES; i++)
6371		if (adapter->tx_ring[i])
6372			cbs_status[i] = adapter->tx_ring[i]->cbs_enable;
6373
6374	/* CBS should be enabled on the highest priority queue first in order
6375	 * for the CBS algorithm to operate as intended.
6376	 */
6377	if (enable) {
6378		if (queue == 1 && !cbs_status[0]) {
6379			netdev_err(netdev,
6380				   "Enabling CBS on queue1 before queue0\n");
6381			return -EINVAL;
6382		}
6383	} else {
6384		if (queue == 0 && cbs_status[1]) {
6385			netdev_err(netdev,
6386				   "Disabling CBS on queue0 before queue1\n");
6387			return -EINVAL;
6388		}
6389	}
6390
6391	ring->cbs_enable = enable;
6392	ring->idleslope = idleslope;
6393	ring->sendslope = sendslope;
6394	ring->hicredit = hicredit;
6395	ring->locredit = locredit;
6396
6397	return 0;
6398}
6399
6400static int igc_tsn_enable_cbs(struct igc_adapter *adapter,
6401			      struct tc_cbs_qopt_offload *qopt)
6402{
6403	struct igc_hw *hw = &adapter->hw;
6404	int err;
6405
6406	if (hw->mac.type != igc_i225)
6407		return -EOPNOTSUPP;
6408
6409	if (qopt->queue < 0 || qopt->queue > 1)
6410		return -EINVAL;
6411
6412	err = igc_save_cbs_params(adapter, qopt->queue, qopt->enable,
6413				  qopt->idleslope, qopt->sendslope,
6414				  qopt->hicredit, qopt->locredit);
6415	if (err)
6416		return err;
6417
6418	return igc_tsn_offload_apply(adapter);
6419}
6420
6421static int igc_tc_query_caps(struct igc_adapter *adapter,
6422			     struct tc_query_caps_base *base)
6423{
6424	struct igc_hw *hw = &adapter->hw;
6425
6426	switch (base->type) {
6427	case TC_SETUP_QDISC_TAPRIO: {
6428		struct tc_taprio_caps *caps = base->caps;
6429
6430		caps->broken_mqprio = true;
6431
6432		if (hw->mac.type == igc_i225) {
6433			caps->supports_queue_max_sdu = true;
6434			caps->gate_mask_per_txq = true;
6435		}
6436
6437		return 0;
6438	}
6439	default:
6440		return -EOPNOTSUPP;
6441	}
6442}
6443
6444static int igc_setup_tc(struct net_device *dev, enum tc_setup_type type,
6445			void *type_data)
6446{
6447	struct igc_adapter *adapter = netdev_priv(dev);
6448
6449	adapter->tc_setup_type = type;
6450
6451	switch (type) {
6452	case TC_QUERY_CAPS:
6453		return igc_tc_query_caps(adapter, type_data);
6454	case TC_SETUP_QDISC_TAPRIO:
6455		return igc_tsn_enable_qbv_scheduling(adapter, type_data);
6456
6457	case TC_SETUP_QDISC_ETF:
6458		return igc_tsn_enable_launchtime(adapter, type_data);
6459
6460	case TC_SETUP_QDISC_CBS:
6461		return igc_tsn_enable_cbs(adapter, type_data);
6462
6463	default:
6464		return -EOPNOTSUPP;
6465	}
6466}
6467
6468static int igc_bpf(struct net_device *dev, struct netdev_bpf *bpf)
6469{
6470	struct igc_adapter *adapter = netdev_priv(dev);
6471
6472	switch (bpf->command) {
6473	case XDP_SETUP_PROG:
6474		return igc_xdp_set_prog(adapter, bpf->prog, bpf->extack);
6475	case XDP_SETUP_XSK_POOL:
6476		return igc_xdp_setup_pool(adapter, bpf->xsk.pool,
6477					  bpf->xsk.queue_id);
6478	default:
6479		return -EOPNOTSUPP;
6480	}
6481}
6482
6483static int igc_xdp_xmit(struct net_device *dev, int num_frames,
6484			struct xdp_frame **frames, u32 flags)
6485{
6486	struct igc_adapter *adapter = netdev_priv(dev);
6487	int cpu = smp_processor_id();
6488	struct netdev_queue *nq;
6489	struct igc_ring *ring;
6490	int i, nxmit;
6491
6492	if (unlikely(!netif_carrier_ok(dev)))
6493		return -ENETDOWN;
6494
6495	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
6496		return -EINVAL;
6497
6498	ring = igc_xdp_get_tx_ring(adapter, cpu);
6499	nq = txring_txq(ring);
6500
6501	__netif_tx_lock(nq, cpu);
6502
6503	/* Avoid transmit queue timeout since we share it with the slow path */
6504	txq_trans_cond_update(nq);
6505
6506	nxmit = 0;
6507	for (i = 0; i < num_frames; i++) {
6508		int err;
6509		struct xdp_frame *xdpf = frames[i];
6510
6511		err = igc_xdp_init_tx_descriptor(ring, xdpf);
6512		if (err)
6513			break;
6514		nxmit++;
6515	}
6516
6517	if (flags & XDP_XMIT_FLUSH)
6518		igc_flush_tx_descriptors(ring);
6519
6520	__netif_tx_unlock(nq);
6521
6522	return nxmit;
6523}
6524
6525static void igc_trigger_rxtxq_interrupt(struct igc_adapter *adapter,
6526					struct igc_q_vector *q_vector)
6527{
6528	struct igc_hw *hw = &adapter->hw;
6529	u32 eics = 0;
6530
6531	eics |= q_vector->eims_value;
6532	wr32(IGC_EICS, eics);
6533}
6534
6535int igc_xsk_wakeup(struct net_device *dev, u32 queue_id, u32 flags)
6536{
6537	struct igc_adapter *adapter = netdev_priv(dev);
6538	struct igc_q_vector *q_vector;
6539	struct igc_ring *ring;
6540
6541	if (test_bit(__IGC_DOWN, &adapter->state))
6542		return -ENETDOWN;
6543
6544	if (!igc_xdp_is_enabled(adapter))
6545		return -ENXIO;
6546
6547	if (queue_id >= adapter->num_rx_queues)
6548		return -EINVAL;
6549
6550	ring = adapter->rx_ring[queue_id];
6551
6552	if (!ring->xsk_pool)
6553		return -ENXIO;
6554
6555	q_vector = adapter->q_vector[queue_id];
6556	if (!napi_if_scheduled_mark_missed(&q_vector->napi))
6557		igc_trigger_rxtxq_interrupt(adapter, q_vector);
6558
6559	return 0;
6560}
6561
6562static ktime_t igc_get_tstamp(struct net_device *dev,
6563			      const struct skb_shared_hwtstamps *hwtstamps,
6564			      bool cycles)
6565{
6566	struct igc_adapter *adapter = netdev_priv(dev);
6567	struct igc_inline_rx_tstamps *tstamp;
6568	ktime_t timestamp;
6569
6570	tstamp = hwtstamps->netdev_data;
6571
6572	if (cycles)
6573		timestamp = igc_ptp_rx_pktstamp(adapter, tstamp->timer1);
6574	else
6575		timestamp = igc_ptp_rx_pktstamp(adapter, tstamp->timer0);
6576
6577	return timestamp;
6578}
6579
6580static const struct net_device_ops igc_netdev_ops = {
6581	.ndo_open		= igc_open,
6582	.ndo_stop		= igc_close,
6583	.ndo_start_xmit		= igc_xmit_frame,
6584	.ndo_set_rx_mode	= igc_set_rx_mode,
6585	.ndo_set_mac_address	= igc_set_mac,
6586	.ndo_change_mtu		= igc_change_mtu,
6587	.ndo_tx_timeout		= igc_tx_timeout,
6588	.ndo_get_stats64	= igc_get_stats64,
6589	.ndo_fix_features	= igc_fix_features,
6590	.ndo_set_features	= igc_set_features,
6591	.ndo_features_check	= igc_features_check,
6592	.ndo_eth_ioctl		= igc_ioctl,
6593	.ndo_setup_tc		= igc_setup_tc,
6594	.ndo_bpf		= igc_bpf,
6595	.ndo_xdp_xmit		= igc_xdp_xmit,
6596	.ndo_xsk_wakeup		= igc_xsk_wakeup,
6597	.ndo_get_tstamp		= igc_get_tstamp,
6598};
6599
6600/* PCIe configuration access */
6601void igc_read_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value)
6602{
6603	struct igc_adapter *adapter = hw->back;
6604
6605	pci_read_config_word(adapter->pdev, reg, value);
6606}
6607
6608void igc_write_pci_cfg(struct igc_hw *hw, u32 reg, u16 *value)
6609{
6610	struct igc_adapter *adapter = hw->back;
6611
6612	pci_write_config_word(adapter->pdev, reg, *value);
6613}
6614
6615s32 igc_read_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value)
6616{
6617	struct igc_adapter *adapter = hw->back;
6618
6619	if (!pci_is_pcie(adapter->pdev))
6620		return -IGC_ERR_CONFIG;
6621
6622	pcie_capability_read_word(adapter->pdev, reg, value);
6623
6624	return IGC_SUCCESS;
6625}
6626
6627s32 igc_write_pcie_cap_reg(struct igc_hw *hw, u32 reg, u16 *value)
6628{
6629	struct igc_adapter *adapter = hw->back;
6630
6631	if (!pci_is_pcie(adapter->pdev))
6632		return -IGC_ERR_CONFIG;
6633
6634	pcie_capability_write_word(adapter->pdev, reg, *value);
6635
6636	return IGC_SUCCESS;
6637}
6638
6639u32 igc_rd32(struct igc_hw *hw, u32 reg)
6640{
6641	struct igc_adapter *igc = container_of(hw, struct igc_adapter, hw);
6642	u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
6643	u32 value = 0;
6644
6645	if (IGC_REMOVED(hw_addr))
6646		return ~value;
6647
6648	value = readl(&hw_addr[reg]);
6649
6650	/* reads should not return all F's */
6651	if (!(~value) && (!reg || !(~readl(hw_addr)))) {
6652		struct net_device *netdev = igc->netdev;
6653
6654		hw->hw_addr = NULL;
6655		netif_device_detach(netdev);
6656		netdev_err(netdev, "PCIe link lost, device now detached\n");
6657		WARN(pci_device_is_present(igc->pdev),
6658		     "igc: Failed to read reg 0x%x!\n", reg);
6659	}
6660
6661	return value;
6662}
6663
6664/* Mapping HW RSS Type to enum xdp_rss_hash_type */
6665static enum xdp_rss_hash_type igc_xdp_rss_type[IGC_RSS_TYPE_MAX_TABLE] = {
6666	[IGC_RSS_TYPE_NO_HASH]		= XDP_RSS_TYPE_L2,
6667	[IGC_RSS_TYPE_HASH_TCP_IPV4]	= XDP_RSS_TYPE_L4_IPV4_TCP,
6668	[IGC_RSS_TYPE_HASH_IPV4]	= XDP_RSS_TYPE_L3_IPV4,
6669	[IGC_RSS_TYPE_HASH_TCP_IPV6]	= XDP_RSS_TYPE_L4_IPV6_TCP,
6670	[IGC_RSS_TYPE_HASH_IPV6_EX]	= XDP_RSS_TYPE_L3_IPV6_EX,
6671	[IGC_RSS_TYPE_HASH_IPV6]	= XDP_RSS_TYPE_L3_IPV6,
6672	[IGC_RSS_TYPE_HASH_TCP_IPV6_EX] = XDP_RSS_TYPE_L4_IPV6_TCP_EX,
6673	[IGC_RSS_TYPE_HASH_UDP_IPV4]	= XDP_RSS_TYPE_L4_IPV4_UDP,
6674	[IGC_RSS_TYPE_HASH_UDP_IPV6]	= XDP_RSS_TYPE_L4_IPV6_UDP,
6675	[IGC_RSS_TYPE_HASH_UDP_IPV6_EX] = XDP_RSS_TYPE_L4_IPV6_UDP_EX,
6676	[10] = XDP_RSS_TYPE_NONE, /* RSS Type above 9 "Reserved" by HW  */
6677	[11] = XDP_RSS_TYPE_NONE, /* keep array sized for SW bit-mask   */
6678	[12] = XDP_RSS_TYPE_NONE, /* to handle future HW revisons       */
6679	[13] = XDP_RSS_TYPE_NONE,
6680	[14] = XDP_RSS_TYPE_NONE,
6681	[15] = XDP_RSS_TYPE_NONE,
6682};
6683
6684static int igc_xdp_rx_hash(const struct xdp_md *_ctx, u32 *hash,
6685			   enum xdp_rss_hash_type *rss_type)
6686{
6687	const struct igc_xdp_buff *ctx = (void *)_ctx;
6688
6689	if (!(ctx->xdp.rxq->dev->features & NETIF_F_RXHASH))
6690		return -ENODATA;
6691
6692	*hash = le32_to_cpu(ctx->rx_desc->wb.lower.hi_dword.rss);
6693	*rss_type = igc_xdp_rss_type[igc_rss_type(ctx->rx_desc)];
6694
6695	return 0;
6696}
6697
6698static int igc_xdp_rx_timestamp(const struct xdp_md *_ctx, u64 *timestamp)
6699{
6700	const struct igc_xdp_buff *ctx = (void *)_ctx;
6701	struct igc_adapter *adapter = netdev_priv(ctx->xdp.rxq->dev);
6702	struct igc_inline_rx_tstamps *tstamp = ctx->rx_ts;
6703
6704	if (igc_test_staterr(ctx->rx_desc, IGC_RXDADV_STAT_TSIP)) {
6705		*timestamp = igc_ptp_rx_pktstamp(adapter, tstamp->timer0);
6706
6707		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6708	}
6709
6710	return -ENODATA;
6711}
6712
6713static const struct xdp_metadata_ops igc_xdp_metadata_ops = {
6714	.xmo_rx_hash			= igc_xdp_rx_hash,
6715	.xmo_rx_timestamp		= igc_xdp_rx_timestamp,
6716};
6717
6718static enum hrtimer_restart igc_qbv_scheduling_timer(struct hrtimer *timer)
6719{
6720	struct igc_adapter *adapter = container_of(timer, struct igc_adapter,
6721						   hrtimer);
6722	unsigned long flags;
6723	unsigned int i;
6724
6725	spin_lock_irqsave(&adapter->qbv_tx_lock, flags);
6726
6727	adapter->qbv_transition = true;
6728	for (i = 0; i < adapter->num_tx_queues; i++) {
6729		struct igc_ring *tx_ring = adapter->tx_ring[i];
6730
6731		if (tx_ring->admin_gate_closed) {
6732			tx_ring->admin_gate_closed = false;
6733			tx_ring->oper_gate_closed = true;
6734		} else {
6735			tx_ring->oper_gate_closed = false;
6736		}
6737	}
6738	adapter->qbv_transition = false;
6739
6740	spin_unlock_irqrestore(&adapter->qbv_tx_lock, flags);
6741
6742	return HRTIMER_NORESTART;
 
 
6743}
6744
6745/**
6746 * igc_probe - Device Initialization Routine
6747 * @pdev: PCI device information struct
6748 * @ent: entry in igc_pci_tbl
6749 *
6750 * Returns 0 on success, negative on failure
6751 *
6752 * igc_probe initializes an adapter identified by a pci_dev structure.
6753 * The OS initialization, configuring the adapter private structure,
6754 * and a hardware reset occur.
6755 */
6756static int igc_probe(struct pci_dev *pdev,
6757		     const struct pci_device_id *ent)
6758{
6759	struct igc_adapter *adapter;
6760	struct net_device *netdev;
6761	struct igc_hw *hw;
6762	const struct igc_info *ei = igc_info_tbl[ent->driver_data];
6763	int err;
6764
6765	err = pci_enable_device_mem(pdev);
6766	if (err)
6767		return err;
6768
 
6769	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
6770	if (err) {
6771		dev_err(&pdev->dev,
6772			"No usable DMA configuration, aborting\n");
6773		goto err_dma;
 
 
 
 
 
6774	}
6775
6776	err = pci_request_mem_regions(pdev, igc_driver_name);
6777	if (err)
6778		goto err_pci_reg;
6779
6780	err = pci_enable_ptm(pdev, NULL);
6781	if (err < 0)
6782		dev_info(&pdev->dev, "PCIe PTM not supported by PCIe bus/controller\n");
6783
6784	pci_set_master(pdev);
6785
6786	err = -ENOMEM;
6787	netdev = alloc_etherdev_mq(sizeof(struct igc_adapter),
6788				   IGC_MAX_TX_QUEUES);
6789
6790	if (!netdev)
6791		goto err_alloc_etherdev;
6792
6793	SET_NETDEV_DEV(netdev, &pdev->dev);
6794
6795	pci_set_drvdata(pdev, netdev);
6796	adapter = netdev_priv(netdev);
6797	adapter->netdev = netdev;
6798	adapter->pdev = pdev;
6799	hw = &adapter->hw;
6800	hw->back = adapter;
6801	adapter->port_num = hw->bus.func;
6802	adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
6803
6804	err = pci_save_state(pdev);
6805	if (err)
6806		goto err_ioremap;
6807
6808	err = -EIO;
6809	adapter->io_addr = ioremap(pci_resource_start(pdev, 0),
6810				   pci_resource_len(pdev, 0));
6811	if (!adapter->io_addr)
6812		goto err_ioremap;
6813
6814	/* hw->hw_addr can be zeroed, so use adapter->io_addr for unmap */
6815	hw->hw_addr = adapter->io_addr;
6816
6817	netdev->netdev_ops = &igc_netdev_ops;
6818	netdev->xdp_metadata_ops = &igc_xdp_metadata_ops;
6819	igc_ethtool_set_ops(netdev);
6820	netdev->watchdog_timeo = 5 * HZ;
6821
6822	netdev->mem_start = pci_resource_start(pdev, 0);
6823	netdev->mem_end = pci_resource_end(pdev, 0);
6824
6825	/* PCI config space info */
6826	hw->vendor_id = pdev->vendor;
6827	hw->device_id = pdev->device;
6828	hw->revision_id = pdev->revision;
6829	hw->subsystem_vendor_id = pdev->subsystem_vendor;
6830	hw->subsystem_device_id = pdev->subsystem_device;
6831
6832	/* Copy the default MAC and PHY function pointers */
6833	memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
6834	memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
6835
6836	/* Initialize skew-specific constants */
6837	err = ei->get_invariants(hw);
6838	if (err)
6839		goto err_sw_init;
6840
6841	/* Add supported features to the features list*/
6842	netdev->features |= NETIF_F_SG;
6843	netdev->features |= NETIF_F_TSO;
6844	netdev->features |= NETIF_F_TSO6;
6845	netdev->features |= NETIF_F_TSO_ECN;
6846	netdev->features |= NETIF_F_RXHASH;
6847	netdev->features |= NETIF_F_RXCSUM;
6848	netdev->features |= NETIF_F_HW_CSUM;
6849	netdev->features |= NETIF_F_SCTP_CRC;
6850	netdev->features |= NETIF_F_HW_TC;
6851
6852#define IGC_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
6853				  NETIF_F_GSO_GRE_CSUM | \
6854				  NETIF_F_GSO_IPXIP4 | \
6855				  NETIF_F_GSO_IPXIP6 | \
6856				  NETIF_F_GSO_UDP_TUNNEL | \
6857				  NETIF_F_GSO_UDP_TUNNEL_CSUM)
6858
6859	netdev->gso_partial_features = IGC_GSO_PARTIAL_FEATURES;
6860	netdev->features |= NETIF_F_GSO_PARTIAL | IGC_GSO_PARTIAL_FEATURES;
6861
6862	/* setup the private structure */
6863	err = igc_sw_init(adapter);
6864	if (err)
6865		goto err_sw_init;
6866
6867	/* copy netdev features into list of user selectable features */
6868	netdev->hw_features |= NETIF_F_NTUPLE;
6869	netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
6870	netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
6871	netdev->hw_features |= netdev->features;
6872
6873	netdev->features |= NETIF_F_HIGHDMA;
6874
6875	netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
6876	netdev->mpls_features |= NETIF_F_HW_CSUM;
6877	netdev->hw_enc_features |= netdev->vlan_features;
6878
6879	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
6880			       NETDEV_XDP_ACT_XSK_ZEROCOPY;
6881
6882	/* MTU range: 68 - 9216 */
6883	netdev->min_mtu = ETH_MIN_MTU;
6884	netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
6885
6886	/* before reading the NVM, reset the controller to put the device in a
6887	 * known good starting state
6888	 */
6889	hw->mac.ops.reset_hw(hw);
6890
6891	if (igc_get_flash_presence_i225(hw)) {
6892		if (hw->nvm.ops.validate(hw) < 0) {
6893			dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
6894			err = -EIO;
6895			goto err_eeprom;
6896		}
6897	}
6898
6899	if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
6900		/* copy the MAC address out of the NVM */
6901		if (hw->mac.ops.read_mac_addr(hw))
6902			dev_err(&pdev->dev, "NVM Read Error\n");
6903	}
6904
6905	eth_hw_addr_set(netdev, hw->mac.addr);
6906
6907	if (!is_valid_ether_addr(netdev->dev_addr)) {
6908		dev_err(&pdev->dev, "Invalid MAC Address\n");
6909		err = -EIO;
6910		goto err_eeprom;
6911	}
6912
6913	/* configure RXPBSIZE and TXPBSIZE */
6914	wr32(IGC_RXPBS, I225_RXPBSIZE_DEFAULT);
6915	wr32(IGC_TXPBS, I225_TXPBSIZE_DEFAULT);
6916
6917	timer_setup(&adapter->watchdog_timer, igc_watchdog, 0);
6918	timer_setup(&adapter->phy_info_timer, igc_update_phy_info, 0);
6919
6920	INIT_WORK(&adapter->reset_task, igc_reset_task);
6921	INIT_WORK(&adapter->watchdog_task, igc_watchdog_task);
6922
6923	hrtimer_init(&adapter->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
6924	adapter->hrtimer.function = &igc_qbv_scheduling_timer;
6925
6926	/* Initialize link properties that are user-changeable */
6927	adapter->fc_autoneg = true;
6928	hw->mac.autoneg = true;
6929	hw->phy.autoneg_advertised = 0xaf;
6930
6931	hw->fc.requested_mode = igc_fc_default;
6932	hw->fc.current_mode = igc_fc_default;
6933
6934	/* By default, support wake on port A */
6935	adapter->flags |= IGC_FLAG_WOL_SUPPORTED;
6936
6937	/* initialize the wol settings based on the eeprom settings */
6938	if (adapter->flags & IGC_FLAG_WOL_SUPPORTED)
6939		adapter->wol |= IGC_WUFC_MAG;
6940
6941	device_set_wakeup_enable(&adapter->pdev->dev,
6942				 adapter->flags & IGC_FLAG_WOL_SUPPORTED);
6943
6944	igc_ptp_init(adapter);
6945
6946	igc_tsn_clear_schedule(adapter);
6947
6948	/* reset the hardware with the new settings */
6949	igc_reset(adapter);
6950
6951	/* let the f/w know that the h/w is now under the control of the
6952	 * driver.
6953	 */
6954	igc_get_hw_control(adapter);
6955
6956	strscpy(netdev->name, "eth%d", sizeof(netdev->name));
6957	err = register_netdev(netdev);
6958	if (err)
6959		goto err_register;
6960
6961	 /* carrier off reporting is important to ethtool even BEFORE open */
6962	netif_carrier_off(netdev);
6963
6964	/* Check if Media Autosense is enabled */
6965	adapter->ei = *ei;
6966
6967	/* print pcie link status and MAC address */
6968	pcie_print_link_status(pdev);
6969	netdev_info(netdev, "MAC: %pM\n", netdev->dev_addr);
6970
6971	dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
6972	/* Disable EEE for internal PHY devices */
6973	hw->dev_spec._base.eee_enable = false;
6974	adapter->flags &= ~IGC_FLAG_EEE;
6975	igc_set_eee_i225(hw, false, false, false);
6976
6977	pm_runtime_put_noidle(&pdev->dev);
6978
6979	return 0;
6980
6981err_register:
6982	igc_release_hw_control(adapter);
6983err_eeprom:
6984	if (!igc_check_reset_block(hw))
6985		igc_reset_phy(hw);
6986err_sw_init:
6987	igc_clear_interrupt_scheme(adapter);
6988	iounmap(adapter->io_addr);
6989err_ioremap:
6990	free_netdev(netdev);
6991err_alloc_etherdev:
6992	pci_release_mem_regions(pdev);
6993err_pci_reg:
6994err_dma:
6995	pci_disable_device(pdev);
6996	return err;
6997}
6998
6999/**
7000 * igc_remove - Device Removal Routine
7001 * @pdev: PCI device information struct
7002 *
7003 * igc_remove is called by the PCI subsystem to alert the driver
7004 * that it should release a PCI device.  This could be caused by a
7005 * Hot-Plug event, or because the driver is going to be removed from
7006 * memory.
7007 */
7008static void igc_remove(struct pci_dev *pdev)
7009{
7010	struct net_device *netdev = pci_get_drvdata(pdev);
7011	struct igc_adapter *adapter = netdev_priv(netdev);
7012
7013	pm_runtime_get_noresume(&pdev->dev);
7014
7015	igc_flush_nfc_rules(adapter);
7016
7017	igc_ptp_stop(adapter);
7018
7019	pci_disable_ptm(pdev);
7020	pci_clear_master(pdev);
7021
7022	set_bit(__IGC_DOWN, &adapter->state);
7023
7024	del_timer_sync(&adapter->watchdog_timer);
7025	del_timer_sync(&adapter->phy_info_timer);
7026
7027	cancel_work_sync(&adapter->reset_task);
7028	cancel_work_sync(&adapter->watchdog_task);
7029	hrtimer_cancel(&adapter->hrtimer);
7030
7031	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
7032	 * would have already happened in close and is redundant.
7033	 */
7034	igc_release_hw_control(adapter);
7035	unregister_netdev(netdev);
7036
7037	igc_clear_interrupt_scheme(adapter);
7038	pci_iounmap(pdev, adapter->io_addr);
7039	pci_release_mem_regions(pdev);
7040
7041	free_netdev(netdev);
7042
 
 
7043	pci_disable_device(pdev);
7044}
7045
7046static int __igc_shutdown(struct pci_dev *pdev, bool *enable_wake,
7047			  bool runtime)
7048{
7049	struct net_device *netdev = pci_get_drvdata(pdev);
7050	struct igc_adapter *adapter = netdev_priv(netdev);
7051	u32 wufc = runtime ? IGC_WUFC_LNKC : adapter->wol;
7052	struct igc_hw *hw = &adapter->hw;
7053	u32 ctrl, rctl, status;
7054	bool wake;
7055
7056	rtnl_lock();
7057	netif_device_detach(netdev);
7058
7059	if (netif_running(netdev))
7060		__igc_close(netdev, true);
7061
7062	igc_ptp_suspend(adapter);
7063
7064	igc_clear_interrupt_scheme(adapter);
7065	rtnl_unlock();
7066
7067	status = rd32(IGC_STATUS);
7068	if (status & IGC_STATUS_LU)
7069		wufc &= ~IGC_WUFC_LNKC;
7070
7071	if (wufc) {
7072		igc_setup_rctl(adapter);
7073		igc_set_rx_mode(netdev);
7074
7075		/* turn on all-multi mode if wake on multicast is enabled */
7076		if (wufc & IGC_WUFC_MC) {
7077			rctl = rd32(IGC_RCTL);
7078			rctl |= IGC_RCTL_MPE;
7079			wr32(IGC_RCTL, rctl);
7080		}
7081
7082		ctrl = rd32(IGC_CTRL);
7083		ctrl |= IGC_CTRL_ADVD3WUC;
7084		wr32(IGC_CTRL, ctrl);
7085
7086		/* Allow time for pending master requests to run */
7087		igc_disable_pcie_master(hw);
7088
7089		wr32(IGC_WUC, IGC_WUC_PME_EN);
7090		wr32(IGC_WUFC, wufc);
7091	} else {
7092		wr32(IGC_WUC, 0);
7093		wr32(IGC_WUFC, 0);
7094	}
7095
7096	wake = wufc || adapter->en_mng_pt;
7097	if (!wake)
7098		igc_power_down_phy_copper_base(&adapter->hw);
7099	else
7100		igc_power_up_link(adapter);
7101
7102	if (enable_wake)
7103		*enable_wake = wake;
7104
7105	/* Release control of h/w to f/w.  If f/w is AMT enabled, this
7106	 * would have already happened in close and is redundant.
7107	 */
7108	igc_release_hw_control(adapter);
7109
7110	pci_disable_device(pdev);
7111
7112	return 0;
7113}
7114
7115#ifdef CONFIG_PM
7116static int __maybe_unused igc_runtime_suspend(struct device *dev)
7117{
7118	return __igc_shutdown(to_pci_dev(dev), NULL, 1);
7119}
7120
7121static void igc_deliver_wake_packet(struct net_device *netdev)
7122{
7123	struct igc_adapter *adapter = netdev_priv(netdev);
7124	struct igc_hw *hw = &adapter->hw;
7125	struct sk_buff *skb;
7126	u32 wupl;
7127
7128	wupl = rd32(IGC_WUPL) & IGC_WUPL_MASK;
7129
7130	/* WUPM stores only the first 128 bytes of the wake packet.
7131	 * Read the packet only if we have the whole thing.
7132	 */
7133	if (wupl == 0 || wupl > IGC_WUPM_BYTES)
7134		return;
7135
7136	skb = netdev_alloc_skb_ip_align(netdev, IGC_WUPM_BYTES);
7137	if (!skb)
7138		return;
7139
7140	skb_put(skb, wupl);
7141
7142	/* Ensure reads are 32-bit aligned */
7143	wupl = roundup(wupl, 4);
7144
7145	memcpy_fromio(skb->data, hw->hw_addr + IGC_WUPM_REG(0), wupl);
7146
7147	skb->protocol = eth_type_trans(skb, netdev);
7148	netif_rx(skb);
7149}
7150
7151static int __maybe_unused igc_resume(struct device *dev)
7152{
7153	struct pci_dev *pdev = to_pci_dev(dev);
7154	struct net_device *netdev = pci_get_drvdata(pdev);
7155	struct igc_adapter *adapter = netdev_priv(netdev);
7156	struct igc_hw *hw = &adapter->hw;
7157	u32 err, val;
7158
7159	pci_set_power_state(pdev, PCI_D0);
7160	pci_restore_state(pdev);
7161	pci_save_state(pdev);
7162
7163	if (!pci_device_is_present(pdev))
7164		return -ENODEV;
7165	err = pci_enable_device_mem(pdev);
7166	if (err) {
7167		netdev_err(netdev, "Cannot enable PCI device from suspend\n");
7168		return err;
7169	}
7170	pci_set_master(pdev);
7171
7172	pci_enable_wake(pdev, PCI_D3hot, 0);
7173	pci_enable_wake(pdev, PCI_D3cold, 0);
7174
7175	if (igc_init_interrupt_scheme(adapter, true)) {
7176		netdev_err(netdev, "Unable to allocate memory for queues\n");
7177		return -ENOMEM;
7178	}
7179
7180	igc_reset(adapter);
7181
7182	/* let the f/w know that the h/w is now under the control of the
7183	 * driver.
7184	 */
7185	igc_get_hw_control(adapter);
7186
7187	val = rd32(IGC_WUS);
7188	if (val & WAKE_PKT_WUS)
7189		igc_deliver_wake_packet(netdev);
7190
7191	wr32(IGC_WUS, ~0);
7192
7193	rtnl_lock();
7194	if (!err && netif_running(netdev))
7195		err = __igc_open(netdev, true);
7196
7197	if (!err)
7198		netif_device_attach(netdev);
7199	rtnl_unlock();
7200
7201	return err;
7202}
7203
7204static int __maybe_unused igc_runtime_resume(struct device *dev)
7205{
7206	return igc_resume(dev);
7207}
7208
7209static int __maybe_unused igc_suspend(struct device *dev)
7210{
7211	return __igc_shutdown(to_pci_dev(dev), NULL, 0);
7212}
7213
7214static int __maybe_unused igc_runtime_idle(struct device *dev)
7215{
7216	struct net_device *netdev = dev_get_drvdata(dev);
7217	struct igc_adapter *adapter = netdev_priv(netdev);
7218
7219	if (!igc_has_link(adapter))
7220		pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7221
7222	return -EBUSY;
7223}
7224#endif /* CONFIG_PM */
7225
7226static void igc_shutdown(struct pci_dev *pdev)
7227{
7228	bool wake;
7229
7230	__igc_shutdown(pdev, &wake, 0);
7231
7232	if (system_state == SYSTEM_POWER_OFF) {
7233		pci_wake_from_d3(pdev, wake);
7234		pci_set_power_state(pdev, PCI_D3hot);
7235	}
7236}
7237
7238/**
7239 *  igc_io_error_detected - called when PCI error is detected
7240 *  @pdev: Pointer to PCI device
7241 *  @state: The current PCI connection state
7242 *
7243 *  This function is called after a PCI bus error affecting
7244 *  this device has been detected.
7245 **/
7246static pci_ers_result_t igc_io_error_detected(struct pci_dev *pdev,
7247					      pci_channel_state_t state)
7248{
7249	struct net_device *netdev = pci_get_drvdata(pdev);
7250	struct igc_adapter *adapter = netdev_priv(netdev);
7251
7252	netif_device_detach(netdev);
7253
7254	if (state == pci_channel_io_perm_failure)
7255		return PCI_ERS_RESULT_DISCONNECT;
7256
7257	if (netif_running(netdev))
7258		igc_down(adapter);
7259	pci_disable_device(pdev);
7260
7261	/* Request a slot reset. */
7262	return PCI_ERS_RESULT_NEED_RESET;
7263}
7264
7265/**
7266 *  igc_io_slot_reset - called after the PCI bus has been reset.
7267 *  @pdev: Pointer to PCI device
7268 *
7269 *  Restart the card from scratch, as if from a cold-boot. Implementation
7270 *  resembles the first-half of the igc_resume routine.
7271 **/
7272static pci_ers_result_t igc_io_slot_reset(struct pci_dev *pdev)
7273{
7274	struct net_device *netdev = pci_get_drvdata(pdev);
7275	struct igc_adapter *adapter = netdev_priv(netdev);
7276	struct igc_hw *hw = &adapter->hw;
7277	pci_ers_result_t result;
7278
7279	if (pci_enable_device_mem(pdev)) {
7280		netdev_err(netdev, "Could not re-enable PCI device after reset\n");
7281		result = PCI_ERS_RESULT_DISCONNECT;
7282	} else {
7283		pci_set_master(pdev);
7284		pci_restore_state(pdev);
7285		pci_save_state(pdev);
7286
7287		pci_enable_wake(pdev, PCI_D3hot, 0);
7288		pci_enable_wake(pdev, PCI_D3cold, 0);
7289
7290		/* In case of PCI error, adapter loses its HW address
7291		 * so we should re-assign it here.
7292		 */
7293		hw->hw_addr = adapter->io_addr;
7294
7295		igc_reset(adapter);
7296		wr32(IGC_WUS, ~0);
7297		result = PCI_ERS_RESULT_RECOVERED;
7298	}
7299
7300	return result;
7301}
7302
7303/**
7304 *  igc_io_resume - called when traffic can start to flow again.
7305 *  @pdev: Pointer to PCI device
7306 *
7307 *  This callback is called when the error recovery driver tells us that
7308 *  its OK to resume normal operation. Implementation resembles the
7309 *  second-half of the igc_resume routine.
7310 */
7311static void igc_io_resume(struct pci_dev *pdev)
7312{
7313	struct net_device *netdev = pci_get_drvdata(pdev);
7314	struct igc_adapter *adapter = netdev_priv(netdev);
7315
7316	rtnl_lock();
7317	if (netif_running(netdev)) {
7318		if (igc_open(netdev)) {
7319			netdev_err(netdev, "igc_open failed after reset\n");
7320			return;
7321		}
7322	}
7323
7324	netif_device_attach(netdev);
7325
7326	/* let the f/w know that the h/w is now under the control of the
7327	 * driver.
7328	 */
7329	igc_get_hw_control(adapter);
7330	rtnl_unlock();
7331}
7332
7333static const struct pci_error_handlers igc_err_handler = {
7334	.error_detected = igc_io_error_detected,
7335	.slot_reset = igc_io_slot_reset,
7336	.resume = igc_io_resume,
7337};
7338
7339#ifdef CONFIG_PM
7340static const struct dev_pm_ops igc_pm_ops = {
7341	SET_SYSTEM_SLEEP_PM_OPS(igc_suspend, igc_resume)
7342	SET_RUNTIME_PM_OPS(igc_runtime_suspend, igc_runtime_resume,
7343			   igc_runtime_idle)
7344};
7345#endif
7346
7347static struct pci_driver igc_driver = {
7348	.name     = igc_driver_name,
7349	.id_table = igc_pci_tbl,
7350	.probe    = igc_probe,
7351	.remove   = igc_remove,
7352#ifdef CONFIG_PM
7353	.driver.pm = &igc_pm_ops,
7354#endif
7355	.shutdown = igc_shutdown,
7356	.err_handler = &igc_err_handler,
7357};
7358
7359/**
7360 * igc_reinit_queues - return error
7361 * @adapter: pointer to adapter structure
7362 */
7363int igc_reinit_queues(struct igc_adapter *adapter)
7364{
7365	struct net_device *netdev = adapter->netdev;
7366	int err = 0;
7367
7368	if (netif_running(netdev))
7369		igc_close(netdev);
7370
7371	igc_reset_interrupt_capability(adapter);
7372
7373	if (igc_init_interrupt_scheme(adapter, true)) {
7374		netdev_err(netdev, "Unable to allocate memory for queues\n");
7375		return -ENOMEM;
7376	}
7377
7378	if (netif_running(netdev))
7379		err = igc_open(netdev);
7380
7381	return err;
7382}
7383
7384/**
7385 * igc_get_hw_dev - return device
7386 * @hw: pointer to hardware structure
7387 *
7388 * used by hardware layer to print debugging information
7389 */
7390struct net_device *igc_get_hw_dev(struct igc_hw *hw)
7391{
7392	struct igc_adapter *adapter = hw->back;
7393
7394	return adapter->netdev;
7395}
7396
7397static void igc_disable_rx_ring_hw(struct igc_ring *ring)
7398{
7399	struct igc_hw *hw = &ring->q_vector->adapter->hw;
7400	u8 idx = ring->reg_idx;
7401	u32 rxdctl;
7402
7403	rxdctl = rd32(IGC_RXDCTL(idx));
7404	rxdctl &= ~IGC_RXDCTL_QUEUE_ENABLE;
7405	rxdctl |= IGC_RXDCTL_SWFLUSH;
7406	wr32(IGC_RXDCTL(idx), rxdctl);
7407}
7408
7409void igc_disable_rx_ring(struct igc_ring *ring)
7410{
7411	igc_disable_rx_ring_hw(ring);
7412	igc_clean_rx_ring(ring);
7413}
7414
7415void igc_enable_rx_ring(struct igc_ring *ring)
7416{
7417	struct igc_adapter *adapter = ring->q_vector->adapter;
7418
7419	igc_configure_rx_ring(adapter, ring);
7420
7421	if (ring->xsk_pool)
7422		igc_alloc_rx_buffers_zc(ring, igc_desc_unused(ring));
7423	else
7424		igc_alloc_rx_buffers(ring, igc_desc_unused(ring));
7425}
7426
7427void igc_disable_tx_ring(struct igc_ring *ring)
7428{
7429	igc_disable_tx_ring_hw(ring);
7430	igc_clean_tx_ring(ring);
7431}
7432
7433void igc_enable_tx_ring(struct igc_ring *ring)
7434{
7435	struct igc_adapter *adapter = ring->q_vector->adapter;
7436
7437	igc_configure_tx_ring(adapter, ring);
7438}
7439
7440/**
7441 * igc_init_module - Driver Registration Routine
7442 *
7443 * igc_init_module is the first routine called when the driver is
7444 * loaded. All it does is register with the PCI subsystem.
7445 */
7446static int __init igc_init_module(void)
7447{
7448	int ret;
7449
7450	pr_info("%s\n", igc_driver_string);
7451	pr_info("%s\n", igc_copyright);
7452
7453	ret = pci_register_driver(&igc_driver);
7454	return ret;
7455}
7456
7457module_init(igc_init_module);
7458
7459/**
7460 * igc_exit_module - Driver Exit Cleanup Routine
7461 *
7462 * igc_exit_module is called just before the driver is removed
7463 * from memory.
7464 */
7465static void __exit igc_exit_module(void)
7466{
7467	pci_unregister_driver(&igc_driver);
7468}
7469
7470module_exit(igc_exit_module);
7471/* igc_main.c */