efx.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/****************************************************************************
   3 * Driver for Solarflare network controllers and boards
   4 * Copyright 2005-2006 Fen Systems Ltd.
   5 * Copyright 2005-2013 Solarflare Communications Inc.
   6 */
   7
   8#include <linux/filter.h>
   9#include <linux/module.h>
  10#include <linux/pci.h>
  11#include <linux/netdevice.h>
  12#include <linux/etherdevice.h>
  13#include <linux/delay.h>
  14#include <linux/notifier.h>
  15#include <linux/ip.h>
  16#include <linux/tcp.h>
  17#include <linux/in.h>
  18#include <linux/ethtool.h>
  19#include <linux/topology.h>
  20#include <linux/gfp.h>
  21#include <linux/interrupt.h>
  22#include "net_driver.h"
  23#include <net/gre.h>
  24#include <net/udp_tunnel.h>
  25#include <net/netdev_queues.h>
  26#include "efx.h"
  27#include "efx_common.h"
  28#include "efx_channels.h"
  29#include "ef100.h"
  30#include "rx_common.h"
  31#include "tx_common.h"
  32#include "nic.h"
  33#include "io.h"
  34#include "selftest.h"
  35#include "sriov.h"
  36#include "efx_devlink.h"
  37
  38#include "mcdi_port_common.h"
  39#include "mcdi_pcol.h"
  40#include "workarounds.h"
  41
  42/**************************************************************************
  43 *
  44 * Configurable values
  45 *
  46 *************************************************************************/
  47
  48module_param_named(interrupt_mode, efx_interrupt_mode, uint, 0444);
  49MODULE_PARM_DESC(interrupt_mode,
  50		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
  51
  52module_param(rss_cpus, uint, 0444);
  53MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");
  54
  55/*
  56 * Use separate channels for TX and RX events
  57 *
  58 * Set this to 1 to use separate channels for TX and RX. It allows us
  59 * to control interrupt affinity separately for TX and RX.
  60 *
  61 * This is only used in MSI-X interrupt mode
  62 */
  63bool efx_separate_tx_channels;
  64module_param(efx_separate_tx_channels, bool, 0444);
  65MODULE_PARM_DESC(efx_separate_tx_channels,
  66		 "Use separate channels for TX and RX");
  67
  68/* Initial interrupt moderation settings.  They can be modified after
  69 * module load with ethtool.
  70 *
  71 * The default for RX should strike a balance between increasing the
  72 * round-trip latency and reducing overhead.
  73 */
  74static unsigned int rx_irq_mod_usec = 60;
  75
  76/* Initial interrupt moderation settings.  They can be modified after
  77 * module load with ethtool.
  78 *
  79 * This default is chosen to ensure that a 10G link does not go idle
  80 * while a TX queue is stopped after it has become full.  A queue is
  81 * restarted when it drops below half full.  The time this takes (assuming
  82 * worst case 3 descriptors per packet and 1024 descriptors) is
  83 *   512 / 3 * 1.2 = 205 usec.
  84 */
  85static unsigned int tx_irq_mod_usec = 150;
  86
  87static bool phy_flash_cfg;
  88module_param(phy_flash_cfg, bool, 0644);
  89MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");
  90
  91static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
  92			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
  93			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
  94			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
  95module_param(debug, uint, 0);
  96MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
  97
  98/**************************************************************************
  99 *
 100 * Utility functions and prototypes
 101 *
 102 *************************************************************************/
 103
 104static void efx_remove_port(struct efx_nic *efx);
 105static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog);
 106static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp);
 107static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs,
 108			u32 flags);
 109
 110/**************************************************************************
 111 *
 112 * Port handling
 113 *
 114 **************************************************************************/
 115
 116static void efx_fini_port(struct efx_nic *efx);
 117
 118static int efx_probe_port(struct efx_nic *efx)
 119{
 120	int rc;
 121
 122	netif_dbg(efx, probe, efx->net_dev, "create port\n");
 123
 124	if (phy_flash_cfg)
 125		efx->phy_mode = PHY_MODE_SPECIAL;
 126
 127	/* Connect up MAC/PHY operations table */
 128	rc = efx->type->probe_port(efx);
 129	if (rc)
 130		return rc;
 131
 132	/* Initialise MAC address to permanent address */
 133	eth_hw_addr_set(efx->net_dev, efx->net_dev->perm_addr);
 134
 135	return 0;
 136}
 137
 138static int efx_init_port(struct efx_nic *efx)
 139{
 140	int rc;
 141
 142	netif_dbg(efx, drv, efx->net_dev, "init port\n");
 143
 144	mutex_lock(&efx->mac_lock);
 145
 146	efx->port_initialized = true;
 147
 148	/* Ensure the PHY advertises the correct flow control settings */
 149	rc = efx_mcdi_port_reconfigure(efx);
 150	if (rc && rc != -EPERM)
 151		goto fail;
 152
 153	mutex_unlock(&efx->mac_lock);
 154	return 0;
 155
 156fail:
 157	mutex_unlock(&efx->mac_lock);
 158	return rc;
 159}
 160
 161static void efx_fini_port(struct efx_nic *efx)
 162{
 163	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
 164
 165	if (!efx->port_initialized)
 166		return;
 167
 168	efx->port_initialized = false;
 169
 170	efx->link_state.up = false;
 171	efx_link_status_changed(efx);
 172}
 173
 174static void efx_remove_port(struct efx_nic *efx)
 175{
 176	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
 177
 178	efx->type->remove_port(efx);
 179}
 180
 181/**************************************************************************
 182 *
 183 * NIC handling
 184 *
 185 **************************************************************************/
 186
 187static LIST_HEAD(efx_primary_list);
 188static LIST_HEAD(efx_unassociated_list);
 189
 190static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right)
 191{
 192	return left->type == right->type &&
 193		left->vpd_sn && right->vpd_sn &&
 194		!strcmp(left->vpd_sn, right->vpd_sn);
 195}
 196
 197static void efx_associate(struct efx_nic *efx)
 198{
 199	struct efx_nic *other, *next;
 200
 201	if (efx->primary == efx) {
 202		/* Adding primary function; look for secondaries */
 203
 204		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
 205		list_add_tail(&efx->node, &efx_primary_list);
 206
 207		list_for_each_entry_safe(other, next, &efx_unassociated_list,
 208					 node) {
 209			if (efx_same_controller(efx, other)) {
 210				list_del(&other->node);
 211				netif_dbg(other, probe, other->net_dev,
 212					  "moving to secondary list of %s %s\n",
 213					  pci_name(efx->pci_dev),
 214					  efx->net_dev->name);
 215				list_add_tail(&other->node,
 216					      &efx->secondary_list);
 217				other->primary = efx;
 218			}
 219		}
 220	} else {
 221		/* Adding secondary function; look for primary */
 222
 223		list_for_each_entry(other, &efx_primary_list, node) {
 224			if (efx_same_controller(efx, other)) {
 225				netif_dbg(efx, probe, efx->net_dev,
 226					  "adding to secondary list of %s %s\n",
 227					  pci_name(other->pci_dev),
 228					  other->net_dev->name);
 229				list_add_tail(&efx->node,
 230					      &other->secondary_list);
 231				efx->primary = other;
 232				return;
 233			}
 234		}
 235
 236		netif_dbg(efx, probe, efx->net_dev,
 237			  "adding to unassociated list\n");
 238		list_add_tail(&efx->node, &efx_unassociated_list);
 239	}
 240}
 241
 242static void efx_dissociate(struct efx_nic *efx)
 243{
 244	struct efx_nic *other, *next;
 245
 246	list_del(&efx->node);
 247	efx->primary = NULL;
 248
 249	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
 250		list_del(&other->node);
 251		netif_dbg(other, probe, other->net_dev,
 252			  "moving to unassociated list\n");
 253		list_add_tail(&other->node, &efx_unassociated_list);
 254		other->primary = NULL;
 255	}
 256}
 257
 258static int efx_probe_nic(struct efx_nic *efx)
 259{
 260	int rc;
 261
 262	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
 263
 264	/* Carry out hardware-type specific initialisation */
 265	rc = efx->type->probe(efx);
 266	if (rc)
 267		return rc;
 268
 269	do {
 270		if (!efx->max_channels || !efx->max_tx_channels) {
 271			netif_err(efx, drv, efx->net_dev,
 272				  "Insufficient resources to allocate"
 273				  " any channels\n");
 274			rc = -ENOSPC;
 275			goto fail1;
 276		}
 277
 278		/* Determine the number of channels and queues by trying
 279		 * to hook in MSI-X interrupts.
 280		 */
 281		rc = efx_probe_interrupts(efx);
 282		if (rc)
 283			goto fail1;
 284
 285		rc = efx_set_channels(efx);
 286		if (rc)
 287			goto fail1;
 288
 289		/* dimension_resources can fail with EAGAIN */
 290		rc = efx->type->dimension_resources(efx);
 291		if (rc != 0 && rc != -EAGAIN)
 292			goto fail2;
 293
 294		if (rc == -EAGAIN)
 295			/* try again with new max_channels */
 296			efx_remove_interrupts(efx);
 297
 298	} while (rc == -EAGAIN);
 299
 300	if (efx->n_channels > 1)
 301		netdev_rss_key_fill(efx->rss_context.rx_hash_key,
 302				    sizeof(efx->rss_context.rx_hash_key));
 303	efx_set_default_rx_indir_table(efx, efx->rss_context.rx_indir_table);
 304
 305	/* Initialise the interrupt moderation settings */
 306	efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000);
 307	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
 308				true);
 309
 310	return 0;
 311
 312fail2:
 313	efx_remove_interrupts(efx);
 314fail1:
 315	efx->type->remove(efx);
 316	return rc;
 317}
 318
 319static void efx_remove_nic(struct efx_nic *efx)
 320{
 321	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
 322
 323	efx_remove_interrupts(efx);
 324	efx->type->remove(efx);
 325}
 326
 327/**************************************************************************
 328 *
 329 * NIC startup/shutdown
 330 *
 331 *************************************************************************/
 332
 333static int efx_probe_all(struct efx_nic *efx)
 334{
 335	int rc;
 336
 337	rc = efx_probe_nic(efx);
 338	if (rc) {
 339		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
 340		goto fail1;
 341	}
 342
 343	rc = efx_probe_port(efx);
 344	if (rc) {
 345		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
 346		goto fail2;
 347	}
 348
 349	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
 350	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
 351		rc = -EINVAL;
 352		goto fail3;
 353	}
 354
 355#ifdef CONFIG_SFC_SRIOV
 356	rc = efx->type->vswitching_probe(efx);
 357	if (rc) /* not fatal; the PF will still work fine */
 358		netif_warn(efx, probe, efx->net_dev,
 359			   "failed to setup vswitching rc=%d;"
 360			   " VFs may not function\n", rc);
 361#endif
 362
 363	rc = efx_probe_filters(efx);
 364	if (rc) {
 365		netif_err(efx, probe, efx->net_dev,
 366			  "failed to create filter tables\n");
 367		goto fail4;
 368	}
 369
 370	rc = efx_probe_channels(efx);
 371	if (rc)
 372		goto fail5;
 373
 374	efx->state = STATE_NET_DOWN;
 375
 376	return 0;
 377
 378 fail5:
 379	efx_remove_filters(efx);
 380 fail4:
 381#ifdef CONFIG_SFC_SRIOV
 382	efx->type->vswitching_remove(efx);
 383#endif
 384 fail3:
 385	efx_remove_port(efx);
 386 fail2:
 387	efx_remove_nic(efx);
 388 fail1:
 389	return rc;
 390}
 391
 392static void efx_remove_all(struct efx_nic *efx)
 393{
 394	rtnl_lock();
 395	efx_xdp_setup_prog(efx, NULL);
 396	rtnl_unlock();
 397
 398	efx_remove_channels(efx);
 399	efx_remove_filters(efx);
 400#ifdef CONFIG_SFC_SRIOV
 401	efx->type->vswitching_remove(efx);
 402#endif
 403	efx_remove_port(efx);
 404	efx_remove_nic(efx);
 405}
 406
 407/**************************************************************************
 408 *
 409 * Interrupt moderation
 410 *
 411 **************************************************************************/
 412unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs)
 413{
 414	if (usecs == 0)
 415		return 0;
 416	if (usecs * 1000 < efx->timer_quantum_ns)
 417		return 1; /* never round down to 0 */
 418	return usecs * 1000 / efx->timer_quantum_ns;
 419}
 420
 421/* Set interrupt moderation parameters */
 422int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
 423			    unsigned int rx_usecs, bool rx_adaptive,
 424			    bool rx_may_override_tx)
 425{
 426	struct efx_channel *channel;
 427	unsigned int timer_max_us;
 428
 429	EFX_ASSERT_RESET_SERIALISED(efx);
 430
 431	timer_max_us = efx->timer_max_ns / 1000;
 432
 433	if (tx_usecs > timer_max_us || rx_usecs > timer_max_us)
 434		return -EINVAL;
 435
 436	if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 &&
 437	    !rx_may_override_tx) {
 438		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
 439			  "RX and TX IRQ moderation must be equal\n");
 440		return -EINVAL;
 441	}
 442
 443	efx->irq_rx_adaptive = rx_adaptive;
 444	efx->irq_rx_moderation_us = rx_usecs;
 445	efx_for_each_channel(channel, efx) {
 446		if (efx_channel_has_rx_queue(channel))
 447			channel->irq_moderation_us = rx_usecs;
 448		else if (efx_channel_has_tx_queues(channel))
 449			channel->irq_moderation_us = tx_usecs;
 450		else if (efx_channel_is_xdp_tx(channel))
 451			channel->irq_moderation_us = tx_usecs;
 452	}
 453
 454	return 0;
 455}
 456
 457void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
 458			    unsigned int *rx_usecs, bool *rx_adaptive)
 459{
 460	*rx_adaptive = efx->irq_rx_adaptive;
 461	*rx_usecs = efx->irq_rx_moderation_us;
 462
 463	/* If channels are shared between RX and TX, so is IRQ
 464	 * moderation.  Otherwise, IRQ moderation is the same for all
 465	 * TX channels and is not adaptive.
 466	 */
 467	if (efx->tx_channel_offset == 0) {
 468		*tx_usecs = *rx_usecs;
 469	} else {
 470		struct efx_channel *tx_channel;
 471
 472		tx_channel = efx->channel[efx->tx_channel_offset];
 473		*tx_usecs = tx_channel->irq_moderation_us;
 474	}
 475}
 476
 477/**************************************************************************
 478 *
 479 * Kernel net device interface
 480 *
 481 *************************************************************************/
 482
 483/* Context: process, rtnl_lock() held. */
 484int efx_net_open(struct net_device *net_dev)
 485{
 486	struct efx_nic *efx = efx_netdev_priv(net_dev);
 487	int rc;
 488
 489	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
 490		  raw_smp_processor_id());
 491
 492	rc = efx_check_disabled(efx);
 493	if (rc)
 494		return rc;
 495	if (efx->phy_mode & PHY_MODE_SPECIAL)
 496		return -EBUSY;
 497	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
 498		return -EIO;
 499
 500	/* Notify the kernel of the link state polled during driver load,
 501	 * before the monitor starts running */
 502	efx_link_status_changed(efx);
 503
 504	efx_start_all(efx);
 505	if (efx->state == STATE_DISABLED || efx->reset_pending)
 506		netif_device_detach(efx->net_dev);
 507	else
 508		efx->state = STATE_NET_UP;
 509
 510	return 0;
 511}
 512
 513/* Context: process, rtnl_lock() held.
 514 * Note that the kernel will ignore our return code; this method
 515 * should really be a void.
 516 */
 517int efx_net_stop(struct net_device *net_dev)
 518{
 519	struct efx_nic *efx = efx_netdev_priv(net_dev);
 520
 521	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
 522		  raw_smp_processor_id());
 523
 524	/* Stop the device and flush all the channels */
 525	efx_stop_all(efx);
 526
 527	return 0;
 528}
 529
 530static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid)
 531{
 532	struct efx_nic *efx = efx_netdev_priv(net_dev);
 533
 534	if (efx->type->vlan_rx_add_vid)
 535		return efx->type->vlan_rx_add_vid(efx, proto, vid);
 536	else
 537		return -EOPNOTSUPP;
 538}
 539
 540static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid)
 541{
 542	struct efx_nic *efx = efx_netdev_priv(net_dev);
 543
 544	if (efx->type->vlan_rx_kill_vid)
 545		return efx->type->vlan_rx_kill_vid(efx, proto, vid);
 546	else
 547		return -EOPNOTSUPP;
 548}
 549
 550static int efx_hwtstamp_set(struct net_device *net_dev,
 551			    struct kernel_hwtstamp_config *config,
 552			    struct netlink_ext_ack *extack)
 553{
 554	struct efx_nic *efx = efx_netdev_priv(net_dev);
 555
 556	return efx_ptp_set_ts_config(efx, config, extack);
 557}
 558
 559static int efx_hwtstamp_get(struct net_device *net_dev,
 560			    struct kernel_hwtstamp_config *config)
 561{
 562	struct efx_nic *efx = efx_netdev_priv(net_dev);
 563
 564	return efx_ptp_get_ts_config(efx, config);
 565}
 566
 567static const struct net_device_ops efx_netdev_ops = {
 568	.ndo_open		= efx_net_open,
 569	.ndo_stop		= efx_net_stop,
 570	.ndo_get_stats64	= efx_net_stats,
 571	.ndo_tx_timeout		= efx_watchdog,
 572	.ndo_start_xmit		= efx_hard_start_xmit,
 573	.ndo_validate_addr	= eth_validate_addr,
 574	.ndo_change_mtu		= efx_change_mtu,
 575	.ndo_set_mac_address	= efx_set_mac_address,
 576	.ndo_set_rx_mode	= efx_set_rx_mode,
 577	.ndo_set_features	= efx_set_features,
 578	.ndo_features_check	= efx_features_check,
 579	.ndo_vlan_rx_add_vid	= efx_vlan_rx_add_vid,
 580	.ndo_vlan_rx_kill_vid	= efx_vlan_rx_kill_vid,
 581	.ndo_hwtstamp_set	= efx_hwtstamp_set,
 582	.ndo_hwtstamp_get	= efx_hwtstamp_get,
 583#ifdef CONFIG_SFC_SRIOV
 584	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
 585	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
 586	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
 587	.ndo_get_vf_config	= efx_sriov_get_vf_config,
 588	.ndo_set_vf_link_state  = efx_sriov_set_vf_link_state,
 589#endif
 590	.ndo_get_phys_port_id   = efx_get_phys_port_id,
 591	.ndo_get_phys_port_name	= efx_get_phys_port_name,
 592#ifdef CONFIG_RFS_ACCEL
 593	.ndo_rx_flow_steer	= efx_filter_rfs,
 594#endif
 595	.ndo_xdp_xmit		= efx_xdp_xmit,
 596	.ndo_bpf		= efx_xdp
 597};
 598
 599static void efx_get_queue_stats_rx(struct net_device *net_dev, int idx,
 600				   struct netdev_queue_stats_rx *stats)
 601{
 602	struct efx_nic *efx = efx_netdev_priv(net_dev);
 603	struct efx_rx_queue *rx_queue;
 604	struct efx_channel *channel;
 605
 606	channel = efx_get_channel(efx, idx);
 607	rx_queue = efx_channel_get_rx_queue(channel);
 608	/* Count only packets since last time datapath was started */
 609	stats->packets = rx_queue->rx_packets - rx_queue->old_rx_packets;
 610	stats->bytes = rx_queue->rx_bytes - rx_queue->old_rx_bytes;
 611	stats->hw_drops = efx_get_queue_stat_rx_hw_drops(channel) -
 612			  channel->old_n_rx_hw_drops;
 613	stats->hw_drop_overruns = channel->n_rx_nodesc_trunc -
 614				  channel->old_n_rx_hw_drop_overruns;
 615}
 616
 617static void efx_get_queue_stats_tx(struct net_device *net_dev, int idx,
 618				   struct netdev_queue_stats_tx *stats)
 619{
 620	struct efx_nic *efx = efx_netdev_priv(net_dev);
 621	struct efx_tx_queue *tx_queue;
 622	struct efx_channel *channel;
 623
 624	channel = efx_get_tx_channel(efx, idx);
 625	stats->packets = 0;
 626	stats->bytes = 0;
 627	stats->hw_gso_packets = 0;
 628	stats->hw_gso_wire_packets = 0;
 629	efx_for_each_channel_tx_queue(tx_queue, channel) {
 630		stats->packets += tx_queue->complete_packets -
 631				  tx_queue->old_complete_packets;
 632		stats->bytes += tx_queue->complete_bytes -
 633				tx_queue->old_complete_bytes;
 634		/* Note that, unlike stats->packets and stats->bytes,
 635		 * these count TXes enqueued, rather than completed,
 636		 * which may not be what users expect.
 637		 */
 638		stats->hw_gso_packets += tx_queue->tso_bursts -
 639					 tx_queue->old_tso_bursts;
 640		stats->hw_gso_wire_packets += tx_queue->tso_packets -
 641					      tx_queue->old_tso_packets;
 642	}
 643}
 644
 645static void efx_get_base_stats(struct net_device *net_dev,
 646			       struct netdev_queue_stats_rx *rx,
 647			       struct netdev_queue_stats_tx *tx)
 648{
 649	struct efx_nic *efx = efx_netdev_priv(net_dev);
 650	struct efx_tx_queue *tx_queue;
 651	struct efx_rx_queue *rx_queue;
 652	struct efx_channel *channel;
 653
 654	rx->packets = 0;
 655	rx->bytes = 0;
 656	rx->hw_drops = 0;
 657	rx->hw_drop_overruns = 0;
 658	tx->packets = 0;
 659	tx->bytes = 0;
 660	tx->hw_gso_packets = 0;
 661	tx->hw_gso_wire_packets = 0;
 662
 663	/* Count all packets on non-core queues, and packets before last
 664	 * datapath start on core queues.
 665	 */
 666	efx_for_each_channel(channel, efx) {
 667		rx_queue = efx_channel_get_rx_queue(channel);
 668		if (channel->channel >= net_dev->real_num_rx_queues) {
 669			rx->packets += rx_queue->rx_packets;
 670			rx->bytes += rx_queue->rx_bytes;
 671			rx->hw_drops += efx_get_queue_stat_rx_hw_drops(channel);
 672			rx->hw_drop_overruns += channel->n_rx_nodesc_trunc;
 673		} else {
 674			rx->packets += rx_queue->old_rx_packets;
 675			rx->bytes += rx_queue->old_rx_bytes;
 676			rx->hw_drops += channel->old_n_rx_hw_drops;
 677			rx->hw_drop_overruns += channel->old_n_rx_hw_drop_overruns;
 678		}
 679		efx_for_each_channel_tx_queue(tx_queue, channel) {
 680			if (channel->channel < efx->tx_channel_offset ||
 681			    channel->channel >= efx->tx_channel_offset +
 682						net_dev->real_num_tx_queues) {
 683				tx->packets += tx_queue->complete_packets;
 684				tx->bytes += tx_queue->complete_bytes;
 685				tx->hw_gso_packets += tx_queue->tso_bursts;
 686				tx->hw_gso_wire_packets += tx_queue->tso_packets;
 687			} else {
 688				tx->packets += tx_queue->old_complete_packets;
 689				tx->bytes += tx_queue->old_complete_bytes;
 690				tx->hw_gso_packets += tx_queue->old_tso_bursts;
 691				tx->hw_gso_wire_packets += tx_queue->old_tso_packets;
 692			}
 693			/* Include XDP TX in device-wide stats */
 694			tx->packets += tx_queue->complete_xdp_packets;
 695			tx->bytes += tx_queue->complete_xdp_bytes;
 696		}
 697	}
 698}
 699
 700static const struct netdev_stat_ops efx_stat_ops = {
 701	.get_queue_stats_rx	= efx_get_queue_stats_rx,
 702	.get_queue_stats_tx	= efx_get_queue_stats_tx,
 703	.get_base_stats		= efx_get_base_stats,
 704};
 705
 706static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog)
 707{
 708	struct bpf_prog *old_prog;
 709
 710	if (efx->xdp_rxq_info_failed) {
 711		netif_err(efx, drv, efx->net_dev,
 712			  "Unable to bind XDP program due to previous failure of rxq_info\n");
 713		return -EINVAL;
 714	}
 715
 716	if (prog && efx->net_dev->mtu > efx_xdp_max_mtu(efx)) {
 717		netif_err(efx, drv, efx->net_dev,
 718			  "Unable to configure XDP with MTU of %d (max: %d)\n",
 719			  efx->net_dev->mtu, efx_xdp_max_mtu(efx));
 720		return -EINVAL;
 721	}
 722
 723	old_prog = rtnl_dereference(efx->xdp_prog);
 724	rcu_assign_pointer(efx->xdp_prog, prog);
 725	/* Release the reference that was originally passed by the caller. */
 726	if (old_prog)
 727		bpf_prog_put(old_prog);
 728
 729	return 0;
 730}
 731
 732/* Context: process, rtnl_lock() held. */
 733static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp)
 734{
 735	struct efx_nic *efx = efx_netdev_priv(dev);
 736
 737	switch (xdp->command) {
 738	case XDP_SETUP_PROG:
 739		return efx_xdp_setup_prog(efx, xdp->prog);
 740	default:
 741		return -EINVAL;
 742	}
 743}
 744
 745static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs,
 746			u32 flags)
 747{
 748	struct efx_nic *efx = efx_netdev_priv(dev);
 749
 750	if (!netif_running(dev))
 751		return -EINVAL;
 752
 753	return efx_xdp_tx_buffers(efx, n, xdpfs, flags & XDP_XMIT_FLUSH);
 754}
 755
 756static void efx_update_name(struct efx_nic *efx)
 757{
 758	strcpy(efx->name, efx->net_dev->name);
 759	efx_mtd_rename(efx);
 760	efx_set_channel_names(efx);
 761}
 762
 763static int efx_netdev_event(struct notifier_block *this,
 764			    unsigned long event, void *ptr)
 765{
 766	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
 767
 768	if ((net_dev->netdev_ops == &efx_netdev_ops) &&
 769	    event == NETDEV_CHANGENAME)
 770		efx_update_name(efx_netdev_priv(net_dev));
 771
 772	return NOTIFY_DONE;
 773}
 774
 775static struct notifier_block efx_netdev_notifier = {
 776	.notifier_call = efx_netdev_event,
 777};
 778
 779static ssize_t phy_type_show(struct device *dev,
 780			     struct device_attribute *attr, char *buf)
 781{
 782	struct efx_nic *efx = dev_get_drvdata(dev);
 783	return sprintf(buf, "%d\n", efx->phy_type);
 784}
 785static DEVICE_ATTR_RO(phy_type);
 786
 787static int efx_register_netdev(struct efx_nic *efx)
 788{
 789	struct net_device *net_dev = efx->net_dev;
 790	struct efx_channel *channel;
 791	int rc;
 792
 793	net_dev->watchdog_timeo = 5 * HZ;
 794	net_dev->irq = efx->pci_dev->irq;
 795	net_dev->netdev_ops = &efx_netdev_ops;
 796	net_dev->stat_ops = &efx_stat_ops;
 797	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0)
 798		net_dev->priv_flags |= IFF_UNICAST_FLT;
 799	net_dev->ethtool_ops = &efx_ethtool_ops;
 800	netif_set_tso_max_segs(net_dev, EFX_TSO_MAX_SEGS);
 801	net_dev->min_mtu = EFX_MIN_MTU;
 802	net_dev->max_mtu = EFX_MAX_MTU;
 803
 804	rtnl_lock();
 805
 806	/* Enable resets to be scheduled and check whether any were
 807	 * already requested.  If so, the NIC is probably hosed so we
 808	 * abort.
 809	 */
 810	if (efx->reset_pending) {
 811		pci_err(efx->pci_dev, "aborting probe due to scheduled reset\n");
 812		rc = -EIO;
 813		goto fail_locked;
 814	}
 815
 816	rc = dev_alloc_name(net_dev, net_dev->name);
 817	if (rc < 0)
 818		goto fail_locked;
 819	efx_update_name(efx);
 820
 821	/* Always start with carrier off; PHY events will detect the link */
 822	netif_carrier_off(net_dev);
 823
 824	rc = register_netdevice(net_dev);
 825	if (rc)
 826		goto fail_locked;
 827
 828	efx_for_each_channel(channel, efx) {
 829		struct efx_tx_queue *tx_queue;
 830		efx_for_each_channel_tx_queue(tx_queue, channel)
 831			efx_init_tx_queue_core_txq(tx_queue);
 832	}
 833
 834	efx_associate(efx);
 835
 836	efx->state = STATE_NET_DOWN;
 837
 838	rtnl_unlock();
 839
 840	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
 841	if (rc) {
 842		netif_err(efx, drv, efx->net_dev,
 843			  "failed to init net dev attributes\n");
 844		goto fail_registered;
 845	}
 846
 847	efx_init_mcdi_logging(efx);
 848
 849	return 0;
 850
 851fail_registered:
 852	rtnl_lock();
 853	efx_dissociate(efx);
 854	unregister_netdevice(net_dev);
 855fail_locked:
 856	efx->state = STATE_UNINIT;
 857	rtnl_unlock();
 858	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
 859	return rc;
 860}
 861
 862static void efx_unregister_netdev(struct efx_nic *efx)
 863{
 864	if (!efx->net_dev)
 865		return;
 866
 867	if (WARN_ON(efx_netdev_priv(efx->net_dev) != efx))
 868		return;
 869
 870	if (efx_dev_registered(efx)) {
 871		strscpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
 872		efx_fini_mcdi_logging(efx);
 873		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
 874		unregister_netdev(efx->net_dev);
 875	}
 876}
 877
 878/**************************************************************************
 879 *
 880 * List of NICs we support
 881 *
 882 **************************************************************************/
 883
 884/* PCI device ID table */
 885static const struct pci_device_id efx_pci_table[] = {
 886	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
 887	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
 888	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903),  /* SFC9120 VF */
 889	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
 890	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923),  /* SFC9140 PF */
 891	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
 892	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923),  /* SFC9140 VF */
 893	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
 894	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03),  /* SFC9220 PF */
 895	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
 896	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03),  /* SFC9220 VF */
 897	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
 898	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0b03),  /* SFC9250 PF */
 899	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
 900	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1b03),  /* SFC9250 VF */
 901	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
 902	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0c03),  /* X4 PF (FF/LL) */
 903	 .driver_data = (unsigned long)&efx_x4_nic_type},
 904	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x2c03),  /* X4 PF (FF only) */
 905	 .driver_data = (unsigned long)&efx_x4_nic_type},
 906	{0}			/* end of list */
 907};
 908
 909/**************************************************************************
 910 *
 911 * Data housekeeping
 912 *
 913 **************************************************************************/
 914
 915void efx_update_sw_stats(struct efx_nic *efx, u64 *stats)
 916{
 917	u64 n_rx_nodesc_trunc = 0;
 918	struct efx_channel *channel;
 919
 920	efx_for_each_channel(channel, efx)
 921		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
 922	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
 923	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
 924}
 925
 926/**************************************************************************
 927 *
 928 * PCI interface
 929 *
 930 **************************************************************************/
 931
 932/* Main body of final NIC shutdown code
 933 * This is called only at module unload (or hotplug removal).
 934 */
 935static void efx_pci_remove_main(struct efx_nic *efx)
 936{
 937	/* Flush reset_work. It can no longer be scheduled since we
 938	 * are not READY.
 939	 */
 940	WARN_ON(efx_net_active(efx->state));
 941	efx_flush_reset_workqueue(efx);
 942
 943	efx_disable_interrupts(efx);
 944	efx_clear_interrupt_affinity(efx);
 945	efx_nic_fini_interrupt(efx);
 946	efx_fini_port(efx);
 947	efx->type->fini(efx);
 948	efx_fini_napi(efx);
 949	efx_remove_all(efx);
 950}
 951
 952/* Final NIC shutdown
 953 * This is called only at module unload (or hotplug removal).  A PF can call
 954 * this on its VFs to ensure they are unbound first.
 955 */
 956static void efx_pci_remove(struct pci_dev *pci_dev)
 957{
 958	struct efx_probe_data *probe_data;
 959	struct efx_nic *efx;
 960
 961	efx = pci_get_drvdata(pci_dev);
 962	if (!efx)
 963		return;
 964
 965	/* Mark the NIC as fini, then stop the interface */
 966	rtnl_lock();
 967	efx_dissociate(efx);
 968	dev_close(efx->net_dev);
 969	efx_disable_interrupts(efx);
 970	efx->state = STATE_UNINIT;
 971	rtnl_unlock();
 972
 973	if (efx->type->sriov_fini)
 974		efx->type->sriov_fini(efx);
 975
 976	efx_fini_devlink_lock(efx);
 977	efx_unregister_netdev(efx);
 978
 979	efx_mtd_remove(efx);
 980
 981	efx_pci_remove_main(efx);
 982
 983	efx_fini_io(efx);
 984	pci_dbg(efx->pci_dev, "shutdown successful\n");
 985
 986	efx_fini_devlink_and_unlock(efx);
 987	efx_fini_struct(efx);
 988	free_netdev(efx->net_dev);
 989	probe_data = container_of(efx, struct efx_probe_data, efx);
 990	kfree(probe_data);
 991};
 992
 993/* NIC VPD information
 994 * Called during probe to display the part number of the
 995 * installed NIC.
 996 */
 997static void efx_probe_vpd_strings(struct efx_nic *efx)
 998{
 999	struct pci_dev *dev = efx->pci_dev;
1000	unsigned int vpd_size, kw_len;
1001	u8 *vpd_data;
1002	int start;
1003
1004	vpd_data = pci_vpd_alloc(dev, &vpd_size);
1005	if (IS_ERR(vpd_data)) {
1006		pci_warn(dev, "Unable to read VPD\n");
1007		return;
1008	}
1009
1010	start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size,
1011					     PCI_VPD_RO_KEYWORD_PARTNO, &kw_len);
1012	if (start < 0)
1013		pci_err(dev, "Part number not found or incomplete\n");
1014	else
1015		pci_info(dev, "Part Number : %.*s\n", kw_len, vpd_data + start);
1016
1017	start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size,
1018					     PCI_VPD_RO_KEYWORD_SERIALNO, &kw_len);
1019	if (start < 0)
1020		pci_err(dev, "Serial number not found or incomplete\n");
1021	else
1022		efx->vpd_sn = kmemdup_nul(vpd_data + start, kw_len, GFP_KERNEL);
1023
1024	kfree(vpd_data);
1025}
1026
1027
1028/* Main body of NIC initialisation
1029 * This is called at module load (or hotplug insertion, theoretically).
1030 */
1031static int efx_pci_probe_main(struct efx_nic *efx)
1032{
1033	int rc;
1034
1035	/* Do start-of-day initialisation */
1036	rc = efx_probe_all(efx);
1037	if (rc)
1038		goto fail1;
1039
1040	efx_init_napi(efx);
1041
1042	down_write(&efx->filter_sem);
1043	rc = efx->type->init(efx);
1044	up_write(&efx->filter_sem);
1045	if (rc) {
1046		pci_err(efx->pci_dev, "failed to initialise NIC\n");
1047		goto fail3;
1048	}
1049
1050	rc = efx_init_port(efx);
1051	if (rc) {
1052		netif_err(efx, probe, efx->net_dev,
1053			  "failed to initialise port\n");
1054		goto fail4;
1055	}
1056
1057	rc = efx_nic_init_interrupt(efx);
1058	if (rc)
1059		goto fail5;
1060
1061	efx_set_interrupt_affinity(efx);
1062	rc = efx_enable_interrupts(efx);
1063	if (rc)
1064		goto fail6;
1065
1066	return 0;
1067
1068 fail6:
1069	efx_clear_interrupt_affinity(efx);
1070	efx_nic_fini_interrupt(efx);
1071 fail5:
1072	efx_fini_port(efx);
1073 fail4:
1074	efx->type->fini(efx);
1075 fail3:
1076	efx_fini_napi(efx);
1077	efx_remove_all(efx);
1078 fail1:
1079	return rc;
1080}
1081
1082static int efx_pci_probe_post_io(struct efx_nic *efx)
1083{
1084	struct net_device *net_dev = efx->net_dev;
1085	int rc = efx_pci_probe_main(efx);
1086
1087	if (rc)
1088		return rc;
1089
1090	if (efx->type->sriov_init) {
1091		rc = efx->type->sriov_init(efx);
1092		if (rc)
1093			pci_err(efx->pci_dev, "SR-IOV can't be enabled rc %d\n",
1094				rc);
1095	}
1096
1097	/* Determine netdevice features */
1098	net_dev->features |= efx->type->offload_features;
1099
1100	/* Add TSO features */
1101	if (efx->type->tso_versions && efx->type->tso_versions(efx))
1102		net_dev->features |= NETIF_F_TSO | NETIF_F_TSO6;
1103
1104	/* Mask for features that also apply to VLAN devices */
1105	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
1106				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
1107				   NETIF_F_RXCSUM);
1108
1109	/* Determine user configurable features */
1110	net_dev->hw_features |= net_dev->features & ~efx->fixed_features;
1111
1112	/* Disable receiving frames with bad FCS, by default. */
1113	net_dev->features &= ~NETIF_F_RXALL;
1114
1115	/* Disable VLAN filtering by default.  It may be enforced if
1116	 * the feature is fixed (i.e. VLAN filters are required to
1117	 * receive VLAN tagged packets due to vPort restrictions).
1118	 */
1119	net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
1120	net_dev->features |= efx->fixed_features;
1121
1122	net_dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1123				NETDEV_XDP_ACT_REDIRECT |
1124				NETDEV_XDP_ACT_NDO_XMIT;
1125
1126	/* devlink creation, registration and lock */
1127	rc = efx_probe_devlink_and_lock(efx);
1128	if (rc)
1129		pci_err(efx->pci_dev, "devlink registration failed");
1130
1131	rc = efx_register_netdev(efx);
1132	efx_probe_devlink_unlock(efx);
1133	if (!rc)
1134		return 0;
1135
1136	efx_pci_remove_main(efx);
1137	return rc;
1138}
1139
1140/* NIC initialisation
1141 *
1142 * This is called at module load (or hotplug insertion,
1143 * theoretically).  It sets up PCI mappings, resets the NIC,
1144 * sets up and registers the network devices with the kernel and hooks
1145 * the interrupt service routine.  It does not prepare the device for
1146 * transmission; this is left to the first time one of the network
1147 * interfaces is brought up (i.e. efx_net_open).
1148 */
1149static int efx_pci_probe(struct pci_dev *pci_dev,
1150			 const struct pci_device_id *entry)
1151{
1152	struct efx_probe_data *probe_data, **probe_ptr;
1153	struct net_device *net_dev;
1154	struct efx_nic *efx;
1155	int rc;
1156
1157	/* Allocate probe data and struct efx_nic */
1158	probe_data = kzalloc(sizeof(*probe_data), GFP_KERNEL);
1159	if (!probe_data)
1160		return -ENOMEM;
1161	probe_data->pci_dev = pci_dev;
1162	efx = &probe_data->efx;
1163
1164	/* Allocate and initialise a struct net_device */
1165	net_dev = alloc_etherdev_mq(sizeof(probe_data), EFX_MAX_CORE_TX_QUEUES);
1166	if (!net_dev) {
1167		rc = -ENOMEM;
1168		goto fail0;
1169	}
1170	probe_ptr = netdev_priv(net_dev);
1171	*probe_ptr = probe_data;
1172	efx->net_dev = net_dev;
1173	efx->type = (const struct efx_nic_type *) entry->driver_data;
1174	efx->fixed_features |= NETIF_F_HIGHDMA;
1175
1176	pci_set_drvdata(pci_dev, efx);
1177	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
1178	rc = efx_init_struct(efx, pci_dev);
1179	if (rc)
1180		goto fail1;
1181
1182	pci_info(pci_dev, "Solarflare NIC detected\n");
1183
1184	if (!efx->type->is_vf)
1185		efx_probe_vpd_strings(efx);
1186
1187	/* Set up basic I/O (BAR mappings etc) */
1188	rc = efx_init_io(efx, efx->type->mem_bar(efx), efx->type->max_dma_mask,
1189			 efx->type->mem_map_size(efx));
1190	if (rc)
1191		goto fail2;
1192
1193	rc = efx_pci_probe_post_io(efx);
1194	if (rc) {
1195		/* On failure, retry once immediately.
1196		 * If we aborted probe due to a scheduled reset, dismiss it.
1197		 */
1198		efx->reset_pending = 0;
1199		rc = efx_pci_probe_post_io(efx);
1200		if (rc) {
1201			/* On another failure, retry once more
1202			 * after a 50-305ms delay.
1203			 */
1204			unsigned char r;
1205
1206			get_random_bytes(&r, 1);
1207			msleep((unsigned int)r + 50);
1208			efx->reset_pending = 0;
1209			rc = efx_pci_probe_post_io(efx);
1210		}
1211	}
1212	if (rc)
1213		goto fail3;
1214
1215	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
1216
1217	/* Try to create MTDs, but allow this to fail */
1218	rtnl_lock();
1219	rc = efx_mtd_probe(efx);
1220	rtnl_unlock();
1221	if (rc && rc != -EPERM)
1222		netif_warn(efx, probe, efx->net_dev,
1223			   "failed to create MTDs (%d)\n", rc);
1224
1225	if (efx->type->udp_tnl_push_ports)
1226		efx->type->udp_tnl_push_ports(efx);
1227
1228	return 0;
1229
1230 fail3:
1231	efx_fini_io(efx);
1232 fail2:
1233	efx_fini_struct(efx);
1234 fail1:
1235	WARN_ON(rc > 0);
1236	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
1237	free_netdev(net_dev);
1238 fail0:
1239	kfree(probe_data);
1240	return rc;
1241}
1242
1243/* efx_pci_sriov_configure returns the actual number of Virtual Functions
1244 * enabled on success
1245 */
1246#ifdef CONFIG_SFC_SRIOV
1247static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
1248{
1249	int rc;
1250	struct efx_nic *efx = pci_get_drvdata(dev);
1251
1252	if (efx->type->sriov_configure) {
1253		rc = efx->type->sriov_configure(efx, num_vfs);
1254		if (rc)
1255			return rc;
1256		else
1257			return num_vfs;
1258	} else
1259		return -EOPNOTSUPP;
1260}
1261#endif
1262
1263static int efx_pm_freeze(struct device *dev)
1264{
1265	struct efx_nic *efx = dev_get_drvdata(dev);
1266
1267	rtnl_lock();
1268
1269	if (efx_net_active(efx->state)) {
1270		efx_device_detach_sync(efx);
1271
1272		efx_stop_all(efx);
1273		efx_disable_interrupts(efx);
1274
1275		efx->state = efx_freeze(efx->state);
1276	}
1277
1278	rtnl_unlock();
1279
1280	return 0;
1281}
1282
1283static void efx_pci_shutdown(struct pci_dev *pci_dev)
1284{
1285	struct efx_nic *efx = pci_get_drvdata(pci_dev);
1286
1287	if (!efx)
1288		return;
1289
1290	efx_pm_freeze(&pci_dev->dev);
1291	pci_disable_device(pci_dev);
1292}
1293
1294static int efx_pm_thaw(struct device *dev)
1295{
1296	int rc;
1297	struct efx_nic *efx = dev_get_drvdata(dev);
1298
1299	rtnl_lock();
1300
1301	if (efx_frozen(efx->state)) {
1302		rc = efx_enable_interrupts(efx);
1303		if (rc)
1304			goto fail;
1305
1306		mutex_lock(&efx->mac_lock);
1307		efx_mcdi_port_reconfigure(efx);
1308		mutex_unlock(&efx->mac_lock);
1309
1310		efx_start_all(efx);
1311
1312		efx_device_attach_if_not_resetting(efx);
1313
1314		efx->state = efx_thaw(efx->state);
1315
1316		efx->type->resume_wol(efx);
1317	}
1318
1319	rtnl_unlock();
1320
1321	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
1322	efx_queue_reset_work(efx);
1323
1324	return 0;
1325
1326fail:
1327	rtnl_unlock();
1328
1329	return rc;
1330}
1331
1332static int efx_pm_poweroff(struct device *dev)
1333{
1334	struct pci_dev *pci_dev = to_pci_dev(dev);
1335	struct efx_nic *efx = pci_get_drvdata(pci_dev);
1336
1337	efx->type->fini(efx);
1338
1339	efx->reset_pending = 0;
1340
1341	pci_save_state(pci_dev);
1342	return pci_set_power_state(pci_dev, PCI_D3hot);
1343}
1344
1345/* Used for both resume and restore */
1346static int efx_pm_resume(struct device *dev)
1347{
1348	struct pci_dev *pci_dev = to_pci_dev(dev);
1349	struct efx_nic *efx = pci_get_drvdata(pci_dev);
1350	int rc;
1351
1352	rc = pci_set_power_state(pci_dev, PCI_D0);
1353	if (rc)
1354		return rc;
1355	pci_restore_state(pci_dev);
1356	rc = pci_enable_device(pci_dev);
1357	if (rc)
1358		return rc;
1359	pci_set_master(efx->pci_dev);
1360	rc = efx->type->reset(efx, RESET_TYPE_ALL);
1361	if (rc)
1362		return rc;
1363	down_write(&efx->filter_sem);
1364	rc = efx->type->init(efx);
1365	up_write(&efx->filter_sem);
1366	if (rc)
1367		return rc;
1368	rc = efx_pm_thaw(dev);
1369	return rc;
1370}
1371
1372static int efx_pm_suspend(struct device *dev)
1373{
1374	int rc;
1375
1376	efx_pm_freeze(dev);
1377	rc = efx_pm_poweroff(dev);
1378	if (rc)
1379		efx_pm_resume(dev);
1380	return rc;
1381}
1382
1383static const struct dev_pm_ops efx_pm_ops = {
1384	.suspend	= efx_pm_suspend,
1385	.resume		= efx_pm_resume,
1386	.freeze		= efx_pm_freeze,
1387	.thaw		= efx_pm_thaw,
1388	.poweroff	= efx_pm_poweroff,
1389	.restore	= efx_pm_resume,
1390};
1391
1392static struct pci_driver efx_pci_driver = {
1393	.name		= KBUILD_MODNAME,
1394	.id_table	= efx_pci_table,
1395	.probe		= efx_pci_probe,
1396	.remove		= efx_pci_remove,
1397	.driver.pm	= &efx_pm_ops,
1398	.shutdown	= efx_pci_shutdown,
1399	.err_handler	= &efx_err_handlers,
1400#ifdef CONFIG_SFC_SRIOV
1401	.sriov_configure = efx_pci_sriov_configure,
1402#endif
1403};
1404
1405/**************************************************************************
1406 *
1407 * Kernel module interface
1408 *
1409 *************************************************************************/
1410
1411static int __init efx_init_module(void)
1412{
1413	int rc;
1414
1415	printk(KERN_INFO "Solarflare NET driver\n");
1416
1417	rc = register_netdevice_notifier(&efx_netdev_notifier);
1418	if (rc)
1419		goto err_notifier;
1420
1421	rc = efx_create_reset_workqueue();
1422	if (rc)
1423		goto err_reset;
1424
1425	rc = pci_register_driver(&efx_pci_driver);
1426	if (rc < 0)
1427		goto err_pci;
1428
1429	rc = pci_register_driver(&ef100_pci_driver);
1430	if (rc < 0)
1431		goto err_pci_ef100;
1432
1433	return 0;
1434
1435 err_pci_ef100:
1436	pci_unregister_driver(&efx_pci_driver);
1437 err_pci:
1438	efx_destroy_reset_workqueue();
1439 err_reset:
1440	unregister_netdevice_notifier(&efx_netdev_notifier);
1441 err_notifier:
1442	return rc;
1443}
1444
1445static void __exit efx_exit_module(void)
1446{
1447	printk(KERN_INFO "Solarflare NET driver unloading\n");
1448
1449	pci_unregister_driver(&ef100_pci_driver);
1450	pci_unregister_driver(&efx_pci_driver);
1451	efx_destroy_reset_workqueue();
1452	unregister_netdevice_notifier(&efx_netdev_notifier);
1453
1454}
1455
1456module_init(efx_init_module);
1457module_exit(efx_exit_module);
1458
1459MODULE_AUTHOR("Solarflare Communications and "
1460	      "Michael Brown <mbrown@fensystems.co.uk>");
1461MODULE_DESCRIPTION("Solarflare network driver");
1462MODULE_LICENSE("GPL");
1463MODULE_DEVICE_TABLE(pci, efx_pci_table);