Linux Audio

Check our new training course

Loading...
v6.8
   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 "efx.h"
  26#include "efx_common.h"
  27#include "efx_channels.h"
  28#include "ef100.h"
  29#include "rx_common.h"
  30#include "tx_common.h"
  31#include "nic.h"
  32#include "io.h"
  33#include "selftest.h"
  34#include "sriov.h"
  35#include "efx_devlink.h"
  36
  37#include "mcdi_port_common.h"
  38#include "mcdi_pcol.h"
  39#include "workarounds.h"
  40
  41/**************************************************************************
  42 *
  43 * Configurable values
  44 *
  45 *************************************************************************/
 
  46
  47module_param_named(interrupt_mode, efx_interrupt_mode, uint, 0444);
  48MODULE_PARM_DESC(interrupt_mode,
  49		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  50
  51module_param(rss_cpus, uint, 0444);
  52MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  53
  54/*
  55 * Use separate channels for TX and RX events
  56 *
  57 * Set this to 1 to use separate channels for TX and RX. It allows us
  58 * to control interrupt affinity separately for TX and RX.
  59 *
  60 * This is only used in MSI-X interrupt mode
  61 */
  62bool efx_separate_tx_channels;
  63module_param(efx_separate_tx_channels, bool, 0444);
  64MODULE_PARM_DESC(efx_separate_tx_channels,
  65		 "Use separate channels for TX and RX");
  66
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  67/* Initial interrupt moderation settings.  They can be modified after
  68 * module load with ethtool.
  69 *
  70 * The default for RX should strike a balance between increasing the
  71 * round-trip latency and reducing overhead.
  72 */
  73static unsigned int rx_irq_mod_usec = 60;
  74
  75/* Initial interrupt moderation settings.  They can be modified after
  76 * module load with ethtool.
  77 *
  78 * This default is chosen to ensure that a 10G link does not go idle
  79 * while a TX queue is stopped after it has become full.  A queue is
  80 * restarted when it drops below half full.  The time this takes (assuming
  81 * worst case 3 descriptors per packet and 1024 descriptors) is
  82 *   512 / 3 * 1.2 = 205 usec.
  83 */
  84static unsigned int tx_irq_mod_usec = 150;
  85
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  86static bool phy_flash_cfg;
  87module_param(phy_flash_cfg, bool, 0644);
  88MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");
  89
 
 
 
 
 
 
 
 
 
 
  90static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
  91			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
  92			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
  93			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
  94module_param(debug, uint, 0);
  95MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
  96
  97/**************************************************************************
  98 *
  99 * Utility functions and prototypes
 100 *
 101 *************************************************************************/
 102
 
 
 
 
 
 103static void efx_remove_port(struct efx_nic *efx);
 104static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog);
 105static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp);
 106static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs,
 107			u32 flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 108
 109/**************************************************************************
 110 *
 111 * Port handling
 112 *
 113 **************************************************************************/
 114
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 115static void efx_fini_port(struct efx_nic *efx);
 116
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 117static int efx_probe_port(struct efx_nic *efx)
 118{
 119	int rc;
 120
 121	netif_dbg(efx, probe, efx->net_dev, "create port\n");
 122
 123	if (phy_flash_cfg)
 124		efx->phy_mode = PHY_MODE_SPECIAL;
 125
 126	/* Connect up MAC/PHY operations table */
 127	rc = efx->type->probe_port(efx);
 128	if (rc)
 129		return rc;
 130
 131	/* Initialise MAC address to permanent address */
 132	eth_hw_addr_set(efx->net_dev, efx->net_dev->perm_addr);
 133
 134	return 0;
 135}
 136
 137static int efx_init_port(struct efx_nic *efx)
 138{
 139	int rc;
 140
 141	netif_dbg(efx, drv, efx->net_dev, "init port\n");
 142
 143	mutex_lock(&efx->mac_lock);
 144
 
 
 
 
 145	efx->port_initialized = true;
 146
 
 
 
 
 147	/* Ensure the PHY advertises the correct flow control settings */
 148	rc = efx_mcdi_port_reconfigure(efx);
 149	if (rc && rc != -EPERM)
 150		goto fail;
 151
 152	mutex_unlock(&efx->mac_lock);
 153	return 0;
 154
 155fail:
 
 
 156	mutex_unlock(&efx->mac_lock);
 157	return rc;
 158}
 159
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 160static void efx_fini_port(struct efx_nic *efx)
 161{
 162	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
 163
 164	if (!efx->port_initialized)
 165		return;
 166
 
 167	efx->port_initialized = false;
 168
 169	efx->link_state.up = false;
 170	efx_link_status_changed(efx);
 171}
 172
 173static void efx_remove_port(struct efx_nic *efx)
 174{
 175	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
 176
 177	efx->type->remove_port(efx);
 178}
 179
 180/**************************************************************************
 181 *
 182 * NIC handling
 183 *
 184 **************************************************************************/
 185
 186static LIST_HEAD(efx_primary_list);
 187static LIST_HEAD(efx_unassociated_list);
 188
 189static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right)
 190{
 191	return left->type == right->type &&
 192		left->vpd_sn && right->vpd_sn &&
 193		!strcmp(left->vpd_sn, right->vpd_sn);
 194}
 195
 196static void efx_associate(struct efx_nic *efx)
 197{
 198	struct efx_nic *other, *next;
 199
 200	if (efx->primary == efx) {
 201		/* Adding primary function; look for secondaries */
 202
 203		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
 204		list_add_tail(&efx->node, &efx_primary_list);
 205
 206		list_for_each_entry_safe(other, next, &efx_unassociated_list,
 207					 node) {
 208			if (efx_same_controller(efx, other)) {
 209				list_del(&other->node);
 210				netif_dbg(other, probe, other->net_dev,
 211					  "moving to secondary list of %s %s\n",
 212					  pci_name(efx->pci_dev),
 213					  efx->net_dev->name);
 214				list_add_tail(&other->node,
 215					      &efx->secondary_list);
 216				other->primary = efx;
 217			}
 218		}
 219	} else {
 220		/* Adding secondary function; look for primary */
 221
 222		list_for_each_entry(other, &efx_primary_list, node) {
 223			if (efx_same_controller(efx, other)) {
 224				netif_dbg(efx, probe, efx->net_dev,
 225					  "adding to secondary list of %s %s\n",
 226					  pci_name(other->pci_dev),
 227					  other->net_dev->name);
 228				list_add_tail(&efx->node,
 229					      &other->secondary_list);
 230				efx->primary = other;
 231				return;
 232			}
 233		}
 234
 235		netif_dbg(efx, probe, efx->net_dev,
 236			  "adding to unassociated list\n");
 237		list_add_tail(&efx->node, &efx_unassociated_list);
 238	}
 239}
 240
 241static void efx_dissociate(struct efx_nic *efx)
 242{
 243	struct efx_nic *other, *next;
 244
 245	list_del(&efx->node);
 246	efx->primary = NULL;
 247
 248	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
 249		list_del(&other->node);
 250		netif_dbg(other, probe, other->net_dev,
 251			  "moving to unassociated list\n");
 252		list_add_tail(&other->node, &efx_unassociated_list);
 253		other->primary = NULL;
 254	}
 255}
 256
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 257static int efx_probe_nic(struct efx_nic *efx)
 258{
 259	int rc;
 260
 261	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
 262
 263	/* Carry out hardware-type specific initialisation */
 264	rc = efx->type->probe(efx);
 265	if (rc)
 266		return rc;
 267
 268	do {
 269		if (!efx->max_channels || !efx->max_tx_channels) {
 270			netif_err(efx, drv, efx->net_dev,
 271				  "Insufficient resources to allocate"
 272				  " any channels\n");
 273			rc = -ENOSPC;
 274			goto fail1;
 275		}
 276
 277		/* Determine the number of channels and queues by trying
 278		 * to hook in MSI-X interrupts.
 279		 */
 280		rc = efx_probe_interrupts(efx);
 281		if (rc)
 282			goto fail1;
 283
 284		rc = efx_set_channels(efx);
 285		if (rc)
 286			goto fail1;
 287
 288		/* dimension_resources can fail with EAGAIN */
 289		rc = efx->type->dimension_resources(efx);
 290		if (rc != 0 && rc != -EAGAIN)
 291			goto fail2;
 292
 293		if (rc == -EAGAIN)
 294			/* try again with new max_channels */
 295			efx_remove_interrupts(efx);
 296
 297	} while (rc == -EAGAIN);
 298
 299	if (efx->n_channels > 1)
 300		netdev_rss_key_fill(efx->rss_context.rx_hash_key,
 301				    sizeof(efx->rss_context.rx_hash_key));
 302	efx_set_default_rx_indir_table(efx, &efx->rss_context);
 
 
 
 303
 304	/* Initialise the interrupt moderation settings */
 305	efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000);
 306	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
 307				true);
 308
 309	return 0;
 310
 311fail2:
 312	efx_remove_interrupts(efx);
 313fail1:
 314	efx->type->remove(efx);
 315	return rc;
 316}
 317
 318static void efx_remove_nic(struct efx_nic *efx)
 319{
 320	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
 321
 322	efx_remove_interrupts(efx);
 323	efx->type->remove(efx);
 324}
 325
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 326/**************************************************************************
 327 *
 328 * NIC startup/shutdown
 329 *
 330 *************************************************************************/
 331
 332static int efx_probe_all(struct efx_nic *efx)
 333{
 334	int rc;
 335
 336	rc = efx_probe_nic(efx);
 337	if (rc) {
 338		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
 339		goto fail1;
 340	}
 341
 342	rc = efx_probe_port(efx);
 343	if (rc) {
 344		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
 345		goto fail2;
 346	}
 347
 348	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
 349	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
 350		rc = -EINVAL;
 351		goto fail3;
 352	}
 
 353
 354#ifdef CONFIG_SFC_SRIOV
 355	rc = efx->type->vswitching_probe(efx);
 356	if (rc) /* not fatal; the PF will still work fine */
 357		netif_warn(efx, probe, efx->net_dev,
 358			   "failed to setup vswitching rc=%d;"
 359			   " VFs may not function\n", rc);
 360#endif
 361
 362	rc = efx_probe_filters(efx);
 363	if (rc) {
 364		netif_err(efx, probe, efx->net_dev,
 365			  "failed to create filter tables\n");
 366		goto fail4;
 367	}
 368
 369	rc = efx_probe_channels(efx);
 370	if (rc)
 371		goto fail5;
 372
 373	efx->state = STATE_NET_DOWN;
 374
 375	return 0;
 376
 377 fail5:
 378	efx_remove_filters(efx);
 379 fail4:
 380#ifdef CONFIG_SFC_SRIOV
 381	efx->type->vswitching_remove(efx);
 382#endif
 383 fail3:
 384	efx_remove_port(efx);
 385 fail2:
 386	efx_remove_nic(efx);
 387 fail1:
 388	return rc;
 389}
 390
 391static void efx_remove_all(struct efx_nic *efx)
 
 
 
 
 
 
 
 392{
 393	rtnl_lock();
 394	efx_xdp_setup_prog(efx, NULL);
 395	rtnl_unlock();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 396
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 397	efx_remove_channels(efx);
 398	efx_remove_filters(efx);
 399#ifdef CONFIG_SFC_SRIOV
 400	efx->type->vswitching_remove(efx);
 401#endif
 402	efx_remove_port(efx);
 403	efx_remove_nic(efx);
 404}
 405
 406/**************************************************************************
 407 *
 408 * Interrupt moderation
 409 *
 410 **************************************************************************/
 411unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs)
 
 412{
 413	if (usecs == 0)
 414		return 0;
 415	if (usecs * 1000 < efx->timer_quantum_ns)
 416		return 1; /* never round down to 0 */
 417	return usecs * 1000 / efx->timer_quantum_ns;
 418}
 419
 420unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks)
 421{
 422	/* We must round up when converting ticks to microseconds
 423	 * because we round down when converting the other way.
 424	 */
 425	return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000);
 426}
 427
 428/* Set interrupt moderation parameters */
 429int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
 430			    unsigned int rx_usecs, bool rx_adaptive,
 431			    bool rx_may_override_tx)
 432{
 433	struct efx_channel *channel;
 434	unsigned int timer_max_us;
 
 
 
 
 435
 436	EFX_ASSERT_RESET_SERIALISED(efx);
 437
 438	timer_max_us = efx->timer_max_ns / 1000;
 439
 440	if (tx_usecs > timer_max_us || rx_usecs > timer_max_us)
 441		return -EINVAL;
 442
 443	if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 &&
 
 
 
 444	    !rx_may_override_tx) {
 445		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
 446			  "RX and TX IRQ moderation must be equal\n");
 447		return -EINVAL;
 448	}
 449
 450	efx->irq_rx_adaptive = rx_adaptive;
 451	efx->irq_rx_moderation_us = rx_usecs;
 452	efx_for_each_channel(channel, efx) {
 453		if (efx_channel_has_rx_queue(channel))
 454			channel->irq_moderation_us = rx_usecs;
 455		else if (efx_channel_has_tx_queues(channel))
 456			channel->irq_moderation_us = tx_usecs;
 457		else if (efx_channel_is_xdp_tx(channel))
 458			channel->irq_moderation_us = tx_usecs;
 459	}
 460
 461	return 0;
 462}
 463
 464void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
 465			    unsigned int *rx_usecs, bool *rx_adaptive)
 466{
 
 
 
 
 467	*rx_adaptive = efx->irq_rx_adaptive;
 468	*rx_usecs = efx->irq_rx_moderation_us;
 
 
 469
 470	/* If channels are shared between RX and TX, so is IRQ
 471	 * moderation.  Otherwise, IRQ moderation is the same for all
 472	 * TX channels and is not adaptive.
 473	 */
 474	if (efx->tx_channel_offset == 0) {
 475		*tx_usecs = *rx_usecs;
 476	} else {
 477		struct efx_channel *tx_channel;
 
 
 
 
 
 
 
 
 
 
 478
 479		tx_channel = efx->channel[efx->tx_channel_offset];
 480		*tx_usecs = tx_channel->irq_moderation_us;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 481	}
 
 
 
 482}
 483
 484/**************************************************************************
 485 *
 486 * ioctls
 487 *
 488 *************************************************************************/
 489
 490/* Net device ioctl
 491 * Context: process, rtnl_lock() held.
 492 */
 493static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
 494{
 495	struct efx_nic *efx = efx_netdev_priv(net_dev);
 496	struct mii_ioctl_data *data = if_mii(ifr);
 497
 
 
 
 
 
 498	/* Convert phy_id from older PRTAD/DEVAD format */
 499	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
 500	    (data->phy_id & 0xfc00) == 0x0400)
 501		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;
 502
 503	return mdio_mii_ioctl(&efx->mdio, data, cmd);
 504}
 505
 506/**************************************************************************
 507 *
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 508 * Kernel net device interface
 509 *
 510 *************************************************************************/
 511
 512/* Context: process, rtnl_lock() held. */
 513int efx_net_open(struct net_device *net_dev)
 514{
 515	struct efx_nic *efx = efx_netdev_priv(net_dev);
 516	int rc;
 517
 518	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
 519		  raw_smp_processor_id());
 520
 521	rc = efx_check_disabled(efx);
 522	if (rc)
 523		return rc;
 524	if (efx->phy_mode & PHY_MODE_SPECIAL)
 525		return -EBUSY;
 526	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
 527		return -EIO;
 528
 529	/* Notify the kernel of the link state polled during driver load,
 530	 * before the monitor starts running */
 531	efx_link_status_changed(efx);
 532
 533	efx_start_all(efx);
 534	if (efx->state == STATE_DISABLED || efx->reset_pending)
 535		netif_device_detach(efx->net_dev);
 536	else
 537		efx->state = STATE_NET_UP;
 538
 539	return 0;
 540}
 541
 542/* Context: process, rtnl_lock() held.
 543 * Note that the kernel will ignore our return code; this method
 544 * should really be a void.
 545 */
 546int efx_net_stop(struct net_device *net_dev)
 547{
 548	struct efx_nic *efx = efx_netdev_priv(net_dev);
 549
 550	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
 551		  raw_smp_processor_id());
 552
 553	/* Stop the device and flush all the channels */
 554	efx_stop_all(efx);
 555
 556	return 0;
 557}
 558
 559static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid)
 
 
 560{
 561	struct efx_nic *efx = efx_netdev_priv(net_dev);
 562
 563	if (efx->type->vlan_rx_add_vid)
 564		return efx->type->vlan_rx_add_vid(efx, proto, vid);
 565	else
 566		return -EOPNOTSUPP;
 
 567}
 568
 569static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid)
 
 570{
 571	struct efx_nic *efx = efx_netdev_priv(net_dev);
 572
 573	if (efx->type->vlan_rx_kill_vid)
 574		return efx->type->vlan_rx_kill_vid(efx, proto, vid);
 575	else
 576		return -EOPNOTSUPP;
 
 577}
 578
 579static int efx_hwtstamp_set(struct net_device *net_dev,
 580			    struct kernel_hwtstamp_config *config,
 581			    struct netlink_ext_ack *extack)
 582{
 583	struct efx_nic *efx = efx_netdev_priv(net_dev);
 
 
 
 
 
 
 
 
 
 
 
 
 584
 585	return efx_ptp_set_ts_config(efx, config, extack);
 
 
 
 
 
 
 
 586}
 587
 588static int efx_hwtstamp_get(struct net_device *net_dev,
 589			    struct kernel_hwtstamp_config *config)
 590{
 591	struct efx_nic *efx = efx_netdev_priv(net_dev);
 
 
 
 
 
 
 
 
 
 
 
 592
 593	return efx_ptp_get_ts_config(efx, config);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 594}
 595
 596static const struct net_device_ops efx_netdev_ops = {
 597	.ndo_open		= efx_net_open,
 598	.ndo_stop		= efx_net_stop,
 599	.ndo_get_stats64	= efx_net_stats,
 600	.ndo_tx_timeout		= efx_watchdog,
 601	.ndo_start_xmit		= efx_hard_start_xmit,
 602	.ndo_validate_addr	= eth_validate_addr,
 603	.ndo_eth_ioctl		= efx_ioctl,
 604	.ndo_change_mtu		= efx_change_mtu,
 605	.ndo_set_mac_address	= efx_set_mac_address,
 606	.ndo_set_rx_mode	= efx_set_rx_mode,
 607	.ndo_set_features	= efx_set_features,
 608	.ndo_features_check	= efx_features_check,
 609	.ndo_vlan_rx_add_vid	= efx_vlan_rx_add_vid,
 610	.ndo_vlan_rx_kill_vid	= efx_vlan_rx_kill_vid,
 611	.ndo_hwtstamp_set	= efx_hwtstamp_set,
 612	.ndo_hwtstamp_get	= efx_hwtstamp_get,
 613#ifdef CONFIG_SFC_SRIOV
 614	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
 615	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
 616	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
 617	.ndo_get_vf_config	= efx_sriov_get_vf_config,
 618	.ndo_set_vf_link_state  = efx_sriov_set_vf_link_state,
 
 
 
 
 
 
 
 
 619#endif
 620	.ndo_get_phys_port_id   = efx_get_phys_port_id,
 621	.ndo_get_phys_port_name	= efx_get_phys_port_name,
 622#ifdef CONFIG_RFS_ACCEL
 623	.ndo_rx_flow_steer	= efx_filter_rfs,
 624#endif
 625	.ndo_xdp_xmit		= efx_xdp_xmit,
 626	.ndo_bpf		= efx_xdp
 627};
 628
 629static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog)
 630{
 631	struct bpf_prog *old_prog;
 632
 633	if (efx->xdp_rxq_info_failed) {
 634		netif_err(efx, drv, efx->net_dev,
 635			  "Unable to bind XDP program due to previous failure of rxq_info\n");
 636		return -EINVAL;
 637	}
 638
 639	if (prog && efx->net_dev->mtu > efx_xdp_max_mtu(efx)) {
 640		netif_err(efx, drv, efx->net_dev,
 641			  "Unable to configure XDP with MTU of %d (max: %d)\n",
 642			  efx->net_dev->mtu, efx_xdp_max_mtu(efx));
 643		return -EINVAL;
 644	}
 645
 646	old_prog = rtnl_dereference(efx->xdp_prog);
 647	rcu_assign_pointer(efx->xdp_prog, prog);
 648	/* Release the reference that was originally passed by the caller. */
 649	if (old_prog)
 650		bpf_prog_put(old_prog);
 651
 652	return 0;
 653}
 654
 655/* Context: process, rtnl_lock() held. */
 656static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp)
 657{
 658	struct efx_nic *efx = efx_netdev_priv(dev);
 659
 660	switch (xdp->command) {
 661	case XDP_SETUP_PROG:
 662		return efx_xdp_setup_prog(efx, xdp->prog);
 663	default:
 664		return -EINVAL;
 665	}
 666}
 667
 668static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs,
 669			u32 flags)
 670{
 671	struct efx_nic *efx = efx_netdev_priv(dev);
 672
 673	if (!netif_running(dev))
 674		return -EINVAL;
 675
 676	return efx_xdp_tx_buffers(efx, n, xdpfs, flags & XDP_XMIT_FLUSH);
 677}
 678
 679static void efx_update_name(struct efx_nic *efx)
 680{
 681	strcpy(efx->name, efx->net_dev->name);
 682	efx_mtd_rename(efx);
 683	efx_set_channel_names(efx);
 684}
 685
 686static int efx_netdev_event(struct notifier_block *this,
 687			    unsigned long event, void *ptr)
 688{
 689	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
 690
 691	if ((net_dev->netdev_ops == &efx_netdev_ops) &&
 692	    event == NETDEV_CHANGENAME)
 693		efx_update_name(efx_netdev_priv(net_dev));
 694
 695	return NOTIFY_DONE;
 696}
 697
 698static struct notifier_block efx_netdev_notifier = {
 699	.notifier_call = efx_netdev_event,
 700};
 701
 702static ssize_t phy_type_show(struct device *dev,
 703			     struct device_attribute *attr, char *buf)
 704{
 705	struct efx_nic *efx = dev_get_drvdata(dev);
 706	return sprintf(buf, "%d\n", efx->phy_type);
 707}
 708static DEVICE_ATTR_RO(phy_type);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 709
 710static int efx_register_netdev(struct efx_nic *efx)
 711{
 712	struct net_device *net_dev = efx->net_dev;
 713	struct efx_channel *channel;
 714	int rc;
 715
 716	net_dev->watchdog_timeo = 5 * HZ;
 717	net_dev->irq = efx->pci_dev->irq;
 718	net_dev->netdev_ops = &efx_netdev_ops;
 719	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0)
 720		net_dev->priv_flags |= IFF_UNICAST_FLT;
 721	net_dev->ethtool_ops = &efx_ethtool_ops;
 722	netif_set_tso_max_segs(net_dev, EFX_TSO_MAX_SEGS);
 723	net_dev->min_mtu = EFX_MIN_MTU;
 724	net_dev->max_mtu = EFX_MAX_MTU;
 725
 726	rtnl_lock();
 727
 728	/* Enable resets to be scheduled and check whether any were
 729	 * already requested.  If so, the NIC is probably hosed so we
 730	 * abort.
 731	 */
 
 
 732	if (efx->reset_pending) {
 733		pci_err(efx->pci_dev, "aborting probe due to scheduled reset\n");
 
 734		rc = -EIO;
 735		goto fail_locked;
 736	}
 737
 738	rc = dev_alloc_name(net_dev, net_dev->name);
 739	if (rc < 0)
 740		goto fail_locked;
 741	efx_update_name(efx);
 742
 743	/* Always start with carrier off; PHY events will detect the link */
 744	netif_carrier_off(net_dev);
 745
 746	rc = register_netdevice(net_dev);
 747	if (rc)
 748		goto fail_locked;
 749
 750	efx_for_each_channel(channel, efx) {
 751		struct efx_tx_queue *tx_queue;
 752		efx_for_each_channel_tx_queue(tx_queue, channel)
 753			efx_init_tx_queue_core_txq(tx_queue);
 754	}
 755
 756	efx_associate(efx);
 757
 758	efx->state = STATE_NET_DOWN;
 759
 760	rtnl_unlock();
 761
 762	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
 763	if (rc) {
 764		netif_err(efx, drv, efx->net_dev,
 765			  "failed to init net dev attributes\n");
 766		goto fail_registered;
 767	}
 768
 769	efx_init_mcdi_logging(efx);
 
 
 
 
 
 
 770
 771	return 0;
 772
 
 
 
 
 773fail_registered:
 774	rtnl_lock();
 775	efx_dissociate(efx);
 776	unregister_netdevice(net_dev);
 777fail_locked:
 778	efx->state = STATE_UNINIT;
 779	rtnl_unlock();
 780	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
 781	return rc;
 782}
 783
 784static void efx_unregister_netdev(struct efx_nic *efx)
 785{
 786	if (!efx->net_dev)
 787		return;
 788
 789	if (WARN_ON(efx_netdev_priv(efx->net_dev) != efx))
 790		return;
 791
 792	if (efx_dev_registered(efx)) {
 793		strscpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
 794		efx_fini_mcdi_logging(efx);
 
 
 795		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
 796		unregister_netdev(efx->net_dev);
 797	}
 798}
 799
 800/**************************************************************************
 801 *
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 802 * List of NICs we support
 803 *
 804 **************************************************************************/
 805
 806/* PCI device ID table */
 807static const struct pci_device_id efx_pci_table[] = {
 
 
 
 
 
 
 
 
 
 
 808	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
 809	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
 810	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903),  /* SFC9120 VF */
 811	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
 812	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923),  /* SFC9140 PF */
 813	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
 814	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923),  /* SFC9140 VF */
 815	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
 816	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03),  /* SFC9220 PF */
 817	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
 818	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03),  /* SFC9220 VF */
 819	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
 820	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0b03),  /* SFC9250 PF */
 821	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
 822	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1b03),  /* SFC9250 VF */
 823	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
 824	{0}			/* end of list */
 825};
 826
 827/**************************************************************************
 828 *
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 829 * Data housekeeping
 830 *
 831 **************************************************************************/
 832
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 833void efx_update_sw_stats(struct efx_nic *efx, u64 *stats)
 834{
 835	u64 n_rx_nodesc_trunc = 0;
 836	struct efx_channel *channel;
 837
 838	efx_for_each_channel(channel, efx)
 839		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
 840	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
 841	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
 842}
 843
 844/**************************************************************************
 845 *
 846 * PCI interface
 847 *
 848 **************************************************************************/
 849
 850/* Main body of final NIC shutdown code
 851 * This is called only at module unload (or hotplug removal).
 852 */
 853static void efx_pci_remove_main(struct efx_nic *efx)
 854{
 855	/* Flush reset_work. It can no longer be scheduled since we
 856	 * are not READY.
 857	 */
 858	WARN_ON(efx_net_active(efx->state));
 859	efx_flush_reset_workqueue(efx);
 860
 861	efx_disable_interrupts(efx);
 862	efx_clear_interrupt_affinity(efx);
 863	efx_nic_fini_interrupt(efx);
 864	efx_fini_port(efx);
 865	efx->type->fini(efx);
 866	efx_fini_napi(efx);
 867	efx_remove_all(efx);
 868}
 869
 870/* Final NIC shutdown
 871 * This is called only at module unload (or hotplug removal).  A PF can call
 872 * this on its VFs to ensure they are unbound first.
 873 */
 874static void efx_pci_remove(struct pci_dev *pci_dev)
 875{
 876	struct efx_probe_data *probe_data;
 877	struct efx_nic *efx;
 878
 879	efx = pci_get_drvdata(pci_dev);
 880	if (!efx)
 881		return;
 882
 883	/* Mark the NIC as fini, then stop the interface */
 884	rtnl_lock();
 885	efx_dissociate(efx);
 886	dev_close(efx->net_dev);
 887	efx_disable_interrupts(efx);
 888	efx->state = STATE_UNINIT;
 889	rtnl_unlock();
 890
 891	if (efx->type->sriov_fini)
 892		efx->type->sriov_fini(efx);
 893
 894	efx_fini_devlink_lock(efx);
 895	efx_unregister_netdev(efx);
 896
 897	efx_mtd_remove(efx);
 898
 899	efx_pci_remove_main(efx);
 900
 901	efx_fini_io(efx);
 902	pci_dbg(efx->pci_dev, "shutdown successful\n");
 903
 904	efx_fini_devlink_and_unlock(efx);
 905	efx_fini_struct(efx);
 906	free_netdev(efx->net_dev);
 907	probe_data = container_of(efx, struct efx_probe_data, efx);
 908	kfree(probe_data);
 909};
 910
 911/* NIC VPD information
 912 * Called during probe to display the part number of the
 913 * installed NIC.
 
 914 */
 
 915static void efx_probe_vpd_strings(struct efx_nic *efx)
 916{
 917	struct pci_dev *dev = efx->pci_dev;
 918	unsigned int vpd_size, kw_len;
 919	u8 *vpd_data;
 920	int start;
 921
 922	vpd_data = pci_vpd_alloc(dev, &vpd_size);
 923	if (IS_ERR(vpd_data)) {
 924		pci_warn(dev, "Unable to read VPD\n");
 
 925		return;
 926	}
 927
 928	start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size,
 929					     PCI_VPD_RO_KEYWORD_PARTNO, &kw_len);
 930	if (start < 0)
 931		pci_err(dev, "Part number not found or incomplete\n");
 932	else
 933		pci_info(dev, "Part Number : %.*s\n", kw_len, vpd_data + start);
 934
 935	start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size,
 936					     PCI_VPD_RO_KEYWORD_SERIALNO, &kw_len);
 937	if (start < 0)
 938		pci_err(dev, "Serial number not found or incomplete\n");
 939	else
 940		efx->vpd_sn = kmemdup_nul(vpd_data + start, kw_len, GFP_KERNEL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 941
 942	kfree(vpd_data);
 943}
 944
 945
 946/* Main body of NIC initialisation
 947 * This is called at module load (or hotplug insertion, theoretically).
 948 */
 949static int efx_pci_probe_main(struct efx_nic *efx)
 950{
 951	int rc;
 952
 953	/* Do start-of-day initialisation */
 954	rc = efx_probe_all(efx);
 955	if (rc)
 956		goto fail1;
 957
 958	efx_init_napi(efx);
 959
 960	down_write(&efx->filter_sem);
 961	rc = efx->type->init(efx);
 962	up_write(&efx->filter_sem);
 963	if (rc) {
 964		pci_err(efx->pci_dev, "failed to initialise NIC\n");
 
 965		goto fail3;
 966	}
 967
 968	rc = efx_init_port(efx);
 969	if (rc) {
 970		netif_err(efx, probe, efx->net_dev,
 971			  "failed to initialise port\n");
 972		goto fail4;
 973	}
 974
 975	rc = efx_nic_init_interrupt(efx);
 976	if (rc)
 977		goto fail5;
 978
 979	efx_set_interrupt_affinity(efx);
 980	rc = efx_enable_interrupts(efx);
 981	if (rc)
 982		goto fail6;
 983
 984	return 0;
 985
 986 fail6:
 987	efx_clear_interrupt_affinity(efx);
 988	efx_nic_fini_interrupt(efx);
 989 fail5:
 990	efx_fini_port(efx);
 991 fail4:
 992	efx->type->fini(efx);
 993 fail3:
 994	efx_fini_napi(efx);
 995	efx_remove_all(efx);
 996 fail1:
 997	return rc;
 998}
 999
1000static int efx_pci_probe_post_io(struct efx_nic *efx)
1001{
1002	struct net_device *net_dev = efx->net_dev;
1003	int rc = efx_pci_probe_main(efx);
1004
1005	if (rc)
1006		return rc;
1007
1008	if (efx->type->sriov_init) {
1009		rc = efx->type->sriov_init(efx);
1010		if (rc)
1011			pci_err(efx->pci_dev, "SR-IOV can't be enabled rc %d\n",
1012				rc);
1013	}
1014
1015	/* Determine netdevice features */
1016	net_dev->features |= efx->type->offload_features;
1017
1018	/* Add TSO features */
1019	if (efx->type->tso_versions && efx->type->tso_versions(efx))
1020		net_dev->features |= NETIF_F_TSO | NETIF_F_TSO6;
1021
1022	/* Mask for features that also apply to VLAN devices */
1023	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
1024				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
1025				   NETIF_F_RXCSUM);
1026
1027	/* Determine user configurable features */
1028	net_dev->hw_features |= net_dev->features & ~efx->fixed_features;
1029
1030	/* Disable receiving frames with bad FCS, by default. */
1031	net_dev->features &= ~NETIF_F_RXALL;
1032
1033	/* Disable VLAN filtering by default.  It may be enforced if
1034	 * the feature is fixed (i.e. VLAN filters are required to
1035	 * receive VLAN tagged packets due to vPort restrictions).
1036	 */
1037	net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
1038	net_dev->features |= efx->fixed_features;
1039
1040	net_dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1041				NETDEV_XDP_ACT_REDIRECT |
1042				NETDEV_XDP_ACT_NDO_XMIT;
1043
1044	/* devlink creation, registration and lock */
1045	rc = efx_probe_devlink_and_lock(efx);
1046	if (rc)
1047		pci_err(efx->pci_dev, "devlink registration failed");
1048
1049	rc = efx_register_netdev(efx);
1050	efx_probe_devlink_unlock(efx);
1051	if (!rc)
1052		return 0;
1053
1054	efx_pci_remove_main(efx);
1055	return rc;
1056}
1057
1058/* NIC initialisation
1059 *
1060 * This is called at module load (or hotplug insertion,
1061 * theoretically).  It sets up PCI mappings, resets the NIC,
1062 * sets up and registers the network devices with the kernel and hooks
1063 * the interrupt service routine.  It does not prepare the device for
1064 * transmission; this is left to the first time one of the network
1065 * interfaces is brought up (i.e. efx_net_open).
1066 */
1067static int efx_pci_probe(struct pci_dev *pci_dev,
1068			 const struct pci_device_id *entry)
1069{
1070	struct efx_probe_data *probe_data, **probe_ptr;
1071	struct net_device *net_dev;
1072	struct efx_nic *efx;
1073	int rc;
1074
1075	/* Allocate probe data and struct efx_nic */
1076	probe_data = kzalloc(sizeof(*probe_data), GFP_KERNEL);
1077	if (!probe_data)
 
1078		return -ENOMEM;
1079	probe_data->pci_dev = pci_dev;
1080	efx = &probe_data->efx;
1081
1082	/* Allocate and initialise a struct net_device */
1083	net_dev = alloc_etherdev_mq(sizeof(probe_data), EFX_MAX_CORE_TX_QUEUES);
1084	if (!net_dev) {
1085		rc = -ENOMEM;
1086		goto fail0;
1087	}
1088	probe_ptr = netdev_priv(net_dev);
1089	*probe_ptr = probe_data;
1090	efx->net_dev = net_dev;
1091	efx->type = (const struct efx_nic_type *) entry->driver_data;
1092	efx->fixed_features |= NETIF_F_HIGHDMA;
1093
 
 
 
 
 
 
 
 
 
1094	pci_set_drvdata(pci_dev, efx);
1095	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
1096	rc = efx_init_struct(efx, pci_dev);
1097	if (rc)
1098		goto fail1;
1099	efx->mdio.dev = net_dev;
1100
1101	pci_info(pci_dev, "Solarflare NIC detected\n");
 
1102
1103	if (!efx->type->is_vf)
1104		efx_probe_vpd_strings(efx);
1105
1106	/* Set up basic I/O (BAR mappings etc) */
1107	rc = efx_init_io(efx, efx->type->mem_bar(efx), efx->type->max_dma_mask,
1108			 efx->type->mem_map_size(efx));
1109	if (rc)
1110		goto fail2;
1111
1112	rc = efx_pci_probe_post_io(efx);
1113	if (rc) {
1114		/* On failure, retry once immediately.
1115		 * If we aborted probe due to a scheduled reset, dismiss it.
1116		 */
1117		efx->reset_pending = 0;
1118		rc = efx_pci_probe_post_io(efx);
1119		if (rc) {
1120			/* On another failure, retry once more
1121			 * after a 50-305ms delay.
1122			 */
1123			unsigned char r;
1124
1125			get_random_bytes(&r, 1);
1126			msleep((unsigned int)r + 50);
1127			efx->reset_pending = 0;
1128			rc = efx_pci_probe_post_io(efx);
1129		}
1130	}
1131	if (rc)
1132		goto fail3;
1133
 
 
 
 
 
 
 
 
 
 
 
1134	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
1135
1136	/* Try to create MTDs, but allow this to fail */
1137	rtnl_lock();
1138	rc = efx_mtd_probe(efx);
1139	rtnl_unlock();
1140	if (rc && rc != -EPERM)
1141		netif_warn(efx, probe, efx->net_dev,
1142			   "failed to create MTDs (%d)\n", rc);
1143
1144	if (efx->type->udp_tnl_push_ports)
1145		efx->type->udp_tnl_push_ports(efx);
 
 
 
1146
1147	return 0;
1148
 
 
1149 fail3:
1150	efx_fini_io(efx);
1151 fail2:
1152	efx_fini_struct(efx);
1153 fail1:
1154	WARN_ON(rc > 0);
1155	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
1156	free_netdev(net_dev);
1157 fail0:
1158	kfree(probe_data);
1159	return rc;
1160}
1161
1162/* efx_pci_sriov_configure returns the actual number of Virtual Functions
1163 * enabled on success
1164 */
1165#ifdef CONFIG_SFC_SRIOV
1166static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
1167{
1168	int rc;
1169	struct efx_nic *efx = pci_get_drvdata(dev);
1170
1171	if (efx->type->sriov_configure) {
1172		rc = efx->type->sriov_configure(efx, num_vfs);
1173		if (rc)
1174			return rc;
1175		else
1176			return num_vfs;
1177	} else
1178		return -EOPNOTSUPP;
1179}
1180#endif
1181
1182static int efx_pm_freeze(struct device *dev)
1183{
1184	struct efx_nic *efx = dev_get_drvdata(dev);
1185
1186	rtnl_lock();
1187
1188	if (efx_net_active(efx->state)) {
 
 
1189		efx_device_detach_sync(efx);
1190
1191		efx_stop_all(efx);
1192		efx_disable_interrupts(efx);
1193
1194		efx->state = efx_freeze(efx->state);
1195	}
1196
1197	rtnl_unlock();
1198
1199	return 0;
1200}
1201
1202static void efx_pci_shutdown(struct pci_dev *pci_dev)
1203{
1204	struct efx_nic *efx = pci_get_drvdata(pci_dev);
1205
1206	if (!efx)
1207		return;
1208
1209	efx_pm_freeze(&pci_dev->dev);
1210	pci_disable_device(pci_dev);
1211}
1212
1213static int efx_pm_thaw(struct device *dev)
1214{
1215	int rc;
1216	struct efx_nic *efx = dev_get_drvdata(dev);
1217
1218	rtnl_lock();
1219
1220	if (efx_frozen(efx->state)) {
1221		rc = efx_enable_interrupts(efx);
1222		if (rc)
1223			goto fail;
1224
1225		mutex_lock(&efx->mac_lock);
1226		efx_mcdi_port_reconfigure(efx);
1227		mutex_unlock(&efx->mac_lock);
1228
1229		efx_start_all(efx);
1230
1231		efx_device_attach_if_not_resetting(efx);
1232
1233		efx->state = efx_thaw(efx->state);
1234
1235		efx->type->resume_wol(efx);
1236	}
1237
1238	rtnl_unlock();
1239
1240	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
1241	efx_queue_reset_work(efx);
1242
1243	return 0;
1244
1245fail:
1246	rtnl_unlock();
1247
1248	return rc;
1249}
1250
1251static int efx_pm_poweroff(struct device *dev)
1252{
1253	struct pci_dev *pci_dev = to_pci_dev(dev);
1254	struct efx_nic *efx = pci_get_drvdata(pci_dev);
1255
1256	efx->type->fini(efx);
1257
1258	efx->reset_pending = 0;
1259
1260	pci_save_state(pci_dev);
1261	return pci_set_power_state(pci_dev, PCI_D3hot);
1262}
1263
1264/* Used for both resume and restore */
1265static int efx_pm_resume(struct device *dev)
1266{
1267	struct pci_dev *pci_dev = to_pci_dev(dev);
1268	struct efx_nic *efx = pci_get_drvdata(pci_dev);
1269	int rc;
1270
1271	rc = pci_set_power_state(pci_dev, PCI_D0);
1272	if (rc)
1273		return rc;
1274	pci_restore_state(pci_dev);
1275	rc = pci_enable_device(pci_dev);
1276	if (rc)
1277		return rc;
1278	pci_set_master(efx->pci_dev);
1279	rc = efx->type->reset(efx, RESET_TYPE_ALL);
1280	if (rc)
1281		return rc;
1282	down_write(&efx->filter_sem);
1283	rc = efx->type->init(efx);
1284	up_write(&efx->filter_sem);
1285	if (rc)
1286		return rc;
1287	rc = efx_pm_thaw(dev);
1288	return rc;
1289}
1290
1291static int efx_pm_suspend(struct device *dev)
1292{
1293	int rc;
1294
1295	efx_pm_freeze(dev);
1296	rc = efx_pm_poweroff(dev);
1297	if (rc)
1298		efx_pm_resume(dev);
1299	return rc;
1300}
1301
1302static const struct dev_pm_ops efx_pm_ops = {
1303	.suspend	= efx_pm_suspend,
1304	.resume		= efx_pm_resume,
1305	.freeze		= efx_pm_freeze,
1306	.thaw		= efx_pm_thaw,
1307	.poweroff	= efx_pm_poweroff,
1308	.restore	= efx_pm_resume,
1309};
1310
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1311static struct pci_driver efx_pci_driver = {
1312	.name		= KBUILD_MODNAME,
1313	.id_table	= efx_pci_table,
1314	.probe		= efx_pci_probe,
1315	.remove		= efx_pci_remove,
1316	.driver.pm	= &efx_pm_ops,
1317	.shutdown	= efx_pci_shutdown,
1318	.err_handler	= &efx_err_handlers,
1319#ifdef CONFIG_SFC_SRIOV
1320	.sriov_configure = efx_pci_sriov_configure,
1321#endif
1322};
1323
1324/**************************************************************************
1325 *
1326 * Kernel module interface
1327 *
1328 *************************************************************************/
1329
 
 
 
 
1330static int __init efx_init_module(void)
1331{
1332	int rc;
1333
1334	printk(KERN_INFO "Solarflare NET driver\n");
1335
1336	rc = register_netdevice_notifier(&efx_netdev_notifier);
1337	if (rc)
1338		goto err_notifier;
1339
1340	rc = efx_create_reset_workqueue();
 
1341	if (rc)
 
 
 
 
 
 
1342		goto err_reset;
 
1343
1344	rc = pci_register_driver(&efx_pci_driver);
1345	if (rc < 0)
1346		goto err_pci;
1347
1348	rc = pci_register_driver(&ef100_pci_driver);
1349	if (rc < 0)
1350		goto err_pci_ef100;
1351
1352	return 0;
1353
1354 err_pci_ef100:
1355	pci_unregister_driver(&efx_pci_driver);
1356 err_pci:
1357	efx_destroy_reset_workqueue();
1358 err_reset:
 
 
 
 
1359	unregister_netdevice_notifier(&efx_netdev_notifier);
1360 err_notifier:
1361	return rc;
1362}
1363
1364static void __exit efx_exit_module(void)
1365{
1366	printk(KERN_INFO "Solarflare NET driver unloading\n");
1367
1368	pci_unregister_driver(&ef100_pci_driver);
1369	pci_unregister_driver(&efx_pci_driver);
1370	efx_destroy_reset_workqueue();
 
 
 
1371	unregister_netdevice_notifier(&efx_netdev_notifier);
1372
1373}
1374
1375module_init(efx_init_module);
1376module_exit(efx_exit_module);
1377
1378MODULE_AUTHOR("Solarflare Communications and "
1379	      "Michael Brown <mbrown@fensystems.co.uk>");
1380MODULE_DESCRIPTION("Solarflare network driver");
1381MODULE_LICENSE("GPL");
1382MODULE_DEVICE_TABLE(pci, efx_pci_table);
v4.6
 
   1/****************************************************************************
   2 * Driver for Solarflare network controllers and boards
   3 * Copyright 2005-2006 Fen Systems Ltd.
   4 * Copyright 2005-2013 Solarflare Communications Inc.
   5 *
   6 * This program is free software; you can redistribute it and/or modify it
   7 * under the terms of the GNU General Public License version 2 as published
   8 * by the Free Software Foundation, incorporated herein by reference.
   9 */
  10
 
  11#include <linux/module.h>
  12#include <linux/pci.h>
  13#include <linux/netdevice.h>
  14#include <linux/etherdevice.h>
  15#include <linux/delay.h>
  16#include <linux/notifier.h>
  17#include <linux/ip.h>
  18#include <linux/tcp.h>
  19#include <linux/in.h>
  20#include <linux/ethtool.h>
  21#include <linux/topology.h>
  22#include <linux/gfp.h>
  23#include <linux/aer.h>
  24#include <linux/interrupt.h>
  25#include "net_driver.h"
 
 
  26#include "efx.h"
 
 
 
 
 
  27#include "nic.h"
 
  28#include "selftest.h"
  29#include "sriov.h"
 
  30
  31#include "mcdi.h"
 
  32#include "workarounds.h"
  33
  34/**************************************************************************
  35 *
  36 * Type name strings
  37 *
  38 **************************************************************************
  39 */
  40
  41/* Loopback mode names (see LOOPBACK_MODE()) */
  42const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
  43const char *const efx_loopback_mode_names[] = {
  44	[LOOPBACK_NONE]		= "NONE",
  45	[LOOPBACK_DATA]		= "DATAPATH",
  46	[LOOPBACK_GMAC]		= "GMAC",
  47	[LOOPBACK_XGMII]	= "XGMII",
  48	[LOOPBACK_XGXS]		= "XGXS",
  49	[LOOPBACK_XAUI]		= "XAUI",
  50	[LOOPBACK_GMII]		= "GMII",
  51	[LOOPBACK_SGMII]	= "SGMII",
  52	[LOOPBACK_XGBR]		= "XGBR",
  53	[LOOPBACK_XFI]		= "XFI",
  54	[LOOPBACK_XAUI_FAR]	= "XAUI_FAR",
  55	[LOOPBACK_GMII_FAR]	= "GMII_FAR",
  56	[LOOPBACK_SGMII_FAR]	= "SGMII_FAR",
  57	[LOOPBACK_XFI_FAR]	= "XFI_FAR",
  58	[LOOPBACK_GPHY]		= "GPHY",
  59	[LOOPBACK_PHYXS]	= "PHYXS",
  60	[LOOPBACK_PCS]		= "PCS",
  61	[LOOPBACK_PMAPMD]	= "PMA/PMD",
  62	[LOOPBACK_XPORT]	= "XPORT",
  63	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
  64	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
  65	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
  66	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
  67	[LOOPBACK_GMII_WS]	= "GMII_WS",
  68	[LOOPBACK_XFI_WS]	= "XFI_WS",
  69	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
  70	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
  71};
  72
  73const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
  74const char *const efx_reset_type_names[] = {
  75	[RESET_TYPE_INVISIBLE]          = "INVISIBLE",
  76	[RESET_TYPE_ALL]                = "ALL",
  77	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
  78	[RESET_TYPE_WORLD]              = "WORLD",
  79	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
  80	[RESET_TYPE_DATAPATH]           = "DATAPATH",
  81	[RESET_TYPE_MC_BIST]		= "MC_BIST",
  82	[RESET_TYPE_DISABLE]            = "DISABLE",
  83	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
  84	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
  85	[RESET_TYPE_RX_RECOVERY]        = "RX_RECOVERY",
  86	[RESET_TYPE_DMA_ERROR]          = "DMA_ERROR",
  87	[RESET_TYPE_TX_SKIP]            = "TX_SKIP",
  88	[RESET_TYPE_MC_FAILURE]         = "MC_FAILURE",
  89	[RESET_TYPE_MCDI_TIMEOUT]	= "MCDI_TIMEOUT (FLR)",
  90};
  91
  92/* Reset workqueue. If any NIC has a hardware failure then a reset will be
  93 * queued onto this work queue. This is not a per-nic work queue, because
  94 * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
  95 */
  96static struct workqueue_struct *reset_workqueue;
  97
  98/* How often and how many times to poll for a reset while waiting for a
  99 * BIST that another function started to complete.
 100 */
 101#define BIST_WAIT_DELAY_MS	100
 102#define BIST_WAIT_DELAY_COUNT	100
 103
 104/**************************************************************************
 105 *
 106 * Configurable values
 107 *
 108 *************************************************************************/
 109
 110/*
 111 * Use separate channels for TX and RX events
 112 *
 113 * Set this to 1 to use separate channels for TX and RX. It allows us
 114 * to control interrupt affinity separately for TX and RX.
 115 *
 116 * This is only used in MSI-X interrupt mode
 117 */
 118bool efx_separate_tx_channels;
 119module_param(efx_separate_tx_channels, bool, 0444);
 120MODULE_PARM_DESC(efx_separate_tx_channels,
 121		 "Use separate channels for TX and RX");
 122
 123/* This is the weight assigned to each of the (per-channel) virtual
 124 * NAPI devices.
 125 */
 126static int napi_weight = 64;
 127
 128/* This is the time (in jiffies) between invocations of the hardware
 129 * monitor.
 130 * On Falcon-based NICs, this will:
 131 * - Check the on-board hardware monitor;
 132 * - Poll the link state and reconfigure the hardware as necessary.
 133 * On Siena-based NICs for power systems with EEH support, this will give EEH a
 134 * chance to start.
 135 */
 136static unsigned int efx_monitor_interval = 1 * HZ;
 137
 138/* Initial interrupt moderation settings.  They can be modified after
 139 * module load with ethtool.
 140 *
 141 * The default for RX should strike a balance between increasing the
 142 * round-trip latency and reducing overhead.
 143 */
 144static unsigned int rx_irq_mod_usec = 60;
 145
 146/* Initial interrupt moderation settings.  They can be modified after
 147 * module load with ethtool.
 148 *
 149 * This default is chosen to ensure that a 10G link does not go idle
 150 * while a TX queue is stopped after it has become full.  A queue is
 151 * restarted when it drops below half full.  The time this takes (assuming
 152 * worst case 3 descriptors per packet and 1024 descriptors) is
 153 *   512 / 3 * 1.2 = 205 usec.
 154 */
 155static unsigned int tx_irq_mod_usec = 150;
 156
 157/* This is the first interrupt mode to try out of:
 158 * 0 => MSI-X
 159 * 1 => MSI
 160 * 2 => legacy
 161 */
 162static unsigned int interrupt_mode;
 163
 164/* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
 165 * i.e. the number of CPUs among which we may distribute simultaneous
 166 * interrupt handling.
 167 *
 168 * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
 169 * The default (0) means to assign an interrupt to each core.
 170 */
 171static unsigned int rss_cpus;
 172module_param(rss_cpus, uint, 0444);
 173MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");
 174
 175static bool phy_flash_cfg;
 176module_param(phy_flash_cfg, bool, 0644);
 177MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");
 178
 179static unsigned irq_adapt_low_thresh = 8000;
 180module_param(irq_adapt_low_thresh, uint, 0644);
 181MODULE_PARM_DESC(irq_adapt_low_thresh,
 182		 "Threshold score for reducing IRQ moderation");
 183
 184static unsigned irq_adapt_high_thresh = 16000;
 185module_param(irq_adapt_high_thresh, uint, 0644);
 186MODULE_PARM_DESC(irq_adapt_high_thresh,
 187		 "Threshold score for increasing IRQ moderation");
 188
 189static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
 190			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
 191			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
 192			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
 193module_param(debug, uint, 0);
 194MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
 195
 196/**************************************************************************
 197 *
 198 * Utility functions and prototypes
 199 *
 200 *************************************************************************/
 201
 202static int efx_soft_enable_interrupts(struct efx_nic *efx);
 203static void efx_soft_disable_interrupts(struct efx_nic *efx);
 204static void efx_remove_channel(struct efx_channel *channel);
 205static void efx_remove_channels(struct efx_nic *efx);
 206static const struct efx_channel_type efx_default_channel_type;
 207static void efx_remove_port(struct efx_nic *efx);
 208static void efx_init_napi_channel(struct efx_channel *channel);
 209static void efx_fini_napi(struct efx_nic *efx);
 210static void efx_fini_napi_channel(struct efx_channel *channel);
 211static void efx_fini_struct(struct efx_nic *efx);
 212static void efx_start_all(struct efx_nic *efx);
 213static void efx_stop_all(struct efx_nic *efx);
 214
 215#define EFX_ASSERT_RESET_SERIALISED(efx)		\
 216	do {						\
 217		if ((efx->state == STATE_READY) ||	\
 218		    (efx->state == STATE_RECOVERY) ||	\
 219		    (efx->state == STATE_DISABLED))	\
 220			ASSERT_RTNL();			\
 221	} while (0)
 222
 223static int efx_check_disabled(struct efx_nic *efx)
 224{
 225	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
 226		netif_err(efx, drv, efx->net_dev,
 227			  "device is disabled due to earlier errors\n");
 228		return -EIO;
 229	}
 230	return 0;
 231}
 232
 233/**************************************************************************
 234 *
 235 * Event queue processing
 236 *
 237 *************************************************************************/
 238
 239/* Process channel's event queue
 240 *
 241 * This function is responsible for processing the event queue of a
 242 * single channel.  The caller must guarantee that this function will
 243 * never be concurrently called more than once on the same channel,
 244 * though different channels may be being processed concurrently.
 245 */
 246static int efx_process_channel(struct efx_channel *channel, int budget)
 247{
 248	struct efx_tx_queue *tx_queue;
 249	int spent;
 250
 251	if (unlikely(!channel->enabled))
 252		return 0;
 253
 254	efx_for_each_channel_tx_queue(tx_queue, channel) {
 255		tx_queue->pkts_compl = 0;
 256		tx_queue->bytes_compl = 0;
 257	}
 258
 259	spent = efx_nic_process_eventq(channel, budget);
 260	if (spent && efx_channel_has_rx_queue(channel)) {
 261		struct efx_rx_queue *rx_queue =
 262			efx_channel_get_rx_queue(channel);
 263
 264		efx_rx_flush_packet(channel);
 265		efx_fast_push_rx_descriptors(rx_queue, true);
 266	}
 267
 268	/* Update BQL */
 269	efx_for_each_channel_tx_queue(tx_queue, channel) {
 270		if (tx_queue->bytes_compl) {
 271			netdev_tx_completed_queue(tx_queue->core_txq,
 272				tx_queue->pkts_compl, tx_queue->bytes_compl);
 273		}
 274	}
 275
 276	return spent;
 277}
 278
 279/* NAPI poll handler
 280 *
 281 * NAPI guarantees serialisation of polls of the same device, which
 282 * provides the guarantee required by efx_process_channel().
 283 */
 284static int efx_poll(struct napi_struct *napi, int budget)
 285{
 286	struct efx_channel *channel =
 287		container_of(napi, struct efx_channel, napi_str);
 288	struct efx_nic *efx = channel->efx;
 289	int spent;
 290
 291	if (!efx_channel_lock_napi(channel))
 292		return budget;
 293
 294	netif_vdbg(efx, intr, efx->net_dev,
 295		   "channel %d NAPI poll executing on CPU %d\n",
 296		   channel->channel, raw_smp_processor_id());
 297
 298	spent = efx_process_channel(channel, budget);
 299
 300	if (spent < budget) {
 301		if (efx_channel_has_rx_queue(channel) &&
 302		    efx->irq_rx_adaptive &&
 303		    unlikely(++channel->irq_count == 1000)) {
 304			if (unlikely(channel->irq_mod_score <
 305				     irq_adapt_low_thresh)) {
 306				if (channel->irq_moderation > 1) {
 307					channel->irq_moderation -= 1;
 308					efx->type->push_irq_moderation(channel);
 309				}
 310			} else if (unlikely(channel->irq_mod_score >
 311					    irq_adapt_high_thresh)) {
 312				if (channel->irq_moderation <
 313				    efx->irq_rx_moderation) {
 314					channel->irq_moderation += 1;
 315					efx->type->push_irq_moderation(channel);
 316				}
 317			}
 318			channel->irq_count = 0;
 319			channel->irq_mod_score = 0;
 320		}
 321
 322		efx_filter_rfs_expire(channel);
 323
 324		/* There is no race here; although napi_disable() will
 325		 * only wait for napi_complete(), this isn't a problem
 326		 * since efx_nic_eventq_read_ack() will have no effect if
 327		 * interrupts have already been disabled.
 328		 */
 329		napi_complete(napi);
 330		efx_nic_eventq_read_ack(channel);
 331	}
 332
 333	efx_channel_unlock_napi(channel);
 334	return spent;
 335}
 336
 337/* Create event queue
 338 * Event queue memory allocations are done only once.  If the channel
 339 * is reset, the memory buffer will be reused; this guards against
 340 * errors during channel reset and also simplifies interrupt handling.
 341 */
 342static int efx_probe_eventq(struct efx_channel *channel)
 343{
 344	struct efx_nic *efx = channel->efx;
 345	unsigned long entries;
 346
 347	netif_dbg(efx, probe, efx->net_dev,
 348		  "chan %d create event queue\n", channel->channel);
 349
 350	/* Build an event queue with room for one event per tx and rx buffer,
 351	 * plus some extra for link state events and MCDI completions. */
 352	entries = roundup_pow_of_two(efx->rxq_entries + efx->txq_entries + 128);
 353	EFX_BUG_ON_PARANOID(entries > EFX_MAX_EVQ_SIZE);
 354	channel->eventq_mask = max(entries, EFX_MIN_EVQ_SIZE) - 1;
 355
 356	return efx_nic_probe_eventq(channel);
 357}
 358
 359/* Prepare channel's event queue */
 360static int efx_init_eventq(struct efx_channel *channel)
 361{
 362	struct efx_nic *efx = channel->efx;
 363	int rc;
 364
 365	EFX_WARN_ON_PARANOID(channel->eventq_init);
 366
 367	netif_dbg(efx, drv, efx->net_dev,
 368		  "chan %d init event queue\n", channel->channel);
 369
 370	rc = efx_nic_init_eventq(channel);
 371	if (rc == 0) {
 372		efx->type->push_irq_moderation(channel);
 373		channel->eventq_read_ptr = 0;
 374		channel->eventq_init = true;
 375	}
 376	return rc;
 377}
 378
 379/* Enable event queue processing and NAPI */
 380void efx_start_eventq(struct efx_channel *channel)
 381{
 382	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
 383		  "chan %d start event queue\n", channel->channel);
 384
 385	/* Make sure the NAPI handler sees the enabled flag set */
 386	channel->enabled = true;
 387	smp_wmb();
 388
 389	efx_channel_enable(channel);
 390	napi_enable(&channel->napi_str);
 391	efx_nic_eventq_read_ack(channel);
 392}
 393
 394/* Disable event queue processing and NAPI */
 395void efx_stop_eventq(struct efx_channel *channel)
 396{
 397	if (!channel->enabled)
 398		return;
 399
 400	napi_disable(&channel->napi_str);
 401	while (!efx_channel_disable(channel))
 402		usleep_range(1000, 20000);
 403	channel->enabled = false;
 404}
 405
 406static void efx_fini_eventq(struct efx_channel *channel)
 407{
 408	if (!channel->eventq_init)
 409		return;
 410
 411	netif_dbg(channel->efx, drv, channel->efx->net_dev,
 412		  "chan %d fini event queue\n", channel->channel);
 413
 414	efx_nic_fini_eventq(channel);
 415	channel->eventq_init = false;
 416}
 417
 418static void efx_remove_eventq(struct efx_channel *channel)
 419{
 420	netif_dbg(channel->efx, drv, channel->efx->net_dev,
 421		  "chan %d remove event queue\n", channel->channel);
 422
 423	efx_nic_remove_eventq(channel);
 424}
 425
 426/**************************************************************************
 427 *
 428 * Channel handling
 429 *
 430 *************************************************************************/
 431
 432/* Allocate and initialise a channel structure. */
 433static struct efx_channel *
 434efx_alloc_channel(struct efx_nic *efx, int i, struct efx_channel *old_channel)
 435{
 436	struct efx_channel *channel;
 437	struct efx_rx_queue *rx_queue;
 438	struct efx_tx_queue *tx_queue;
 439	int j;
 440
 441	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
 442	if (!channel)
 443		return NULL;
 444
 445	channel->efx = efx;
 446	channel->channel = i;
 447	channel->type = &efx_default_channel_type;
 448
 449	for (j = 0; j < EFX_TXQ_TYPES; j++) {
 450		tx_queue = &channel->tx_queue[j];
 451		tx_queue->efx = efx;
 452		tx_queue->queue = i * EFX_TXQ_TYPES + j;
 453		tx_queue->channel = channel;
 454	}
 455
 456	rx_queue = &channel->rx_queue;
 457	rx_queue->efx = efx;
 458	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
 459		    (unsigned long)rx_queue);
 460
 461	return channel;
 462}
 463
 464/* Allocate and initialise a channel structure, copying parameters
 465 * (but not resources) from an old channel structure.
 466 */
 467static struct efx_channel *
 468efx_copy_channel(const struct efx_channel *old_channel)
 469{
 470	struct efx_channel *channel;
 471	struct efx_rx_queue *rx_queue;
 472	struct efx_tx_queue *tx_queue;
 473	int j;
 474
 475	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
 476	if (!channel)
 477		return NULL;
 478
 479	*channel = *old_channel;
 480
 481	channel->napi_dev = NULL;
 482	memset(&channel->eventq, 0, sizeof(channel->eventq));
 483
 484	for (j = 0; j < EFX_TXQ_TYPES; j++) {
 485		tx_queue = &channel->tx_queue[j];
 486		if (tx_queue->channel)
 487			tx_queue->channel = channel;
 488		tx_queue->buffer = NULL;
 489		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
 490	}
 491
 492	rx_queue = &channel->rx_queue;
 493	rx_queue->buffer = NULL;
 494	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
 495	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
 496		    (unsigned long)rx_queue);
 497
 498	return channel;
 499}
 500
 501static int efx_probe_channel(struct efx_channel *channel)
 502{
 503	struct efx_tx_queue *tx_queue;
 504	struct efx_rx_queue *rx_queue;
 505	int rc;
 506
 507	netif_dbg(channel->efx, probe, channel->efx->net_dev,
 508		  "creating channel %d\n", channel->channel);
 509
 510	rc = channel->type->pre_probe(channel);
 511	if (rc)
 512		goto fail;
 513
 514	rc = efx_probe_eventq(channel);
 515	if (rc)
 516		goto fail;
 517
 518	efx_for_each_channel_tx_queue(tx_queue, channel) {
 519		rc = efx_probe_tx_queue(tx_queue);
 520		if (rc)
 521			goto fail;
 522	}
 523
 524	efx_for_each_channel_rx_queue(rx_queue, channel) {
 525		rc = efx_probe_rx_queue(rx_queue);
 526		if (rc)
 527			goto fail;
 528	}
 529
 530	return 0;
 531
 532fail:
 533	efx_remove_channel(channel);
 534	return rc;
 535}
 536
 537static void
 538efx_get_channel_name(struct efx_channel *channel, char *buf, size_t len)
 539{
 540	struct efx_nic *efx = channel->efx;
 541	const char *type;
 542	int number;
 543
 544	number = channel->channel;
 545	if (efx->tx_channel_offset == 0) {
 546		type = "";
 547	} else if (channel->channel < efx->tx_channel_offset) {
 548		type = "-rx";
 549	} else {
 550		type = "-tx";
 551		number -= efx->tx_channel_offset;
 552	}
 553	snprintf(buf, len, "%s%s-%d", efx->name, type, number);
 554}
 555
 556static void efx_set_channel_names(struct efx_nic *efx)
 557{
 558	struct efx_channel *channel;
 559
 560	efx_for_each_channel(channel, efx)
 561		channel->type->get_name(channel,
 562					efx->msi_context[channel->channel].name,
 563					sizeof(efx->msi_context[0].name));
 564}
 565
 566static int efx_probe_channels(struct efx_nic *efx)
 567{
 568	struct efx_channel *channel;
 569	int rc;
 570
 571	/* Restart special buffer allocation */
 572	efx->next_buffer_table = 0;
 573
 574	/* Probe channels in reverse, so that any 'extra' channels
 575	 * use the start of the buffer table. This allows the traffic
 576	 * channels to be resized without moving them or wasting the
 577	 * entries before them.
 578	 */
 579	efx_for_each_channel_rev(channel, efx) {
 580		rc = efx_probe_channel(channel);
 581		if (rc) {
 582			netif_err(efx, probe, efx->net_dev,
 583				  "failed to create channel %d\n",
 584				  channel->channel);
 585			goto fail;
 586		}
 587	}
 588	efx_set_channel_names(efx);
 589
 590	return 0;
 591
 592fail:
 593	efx_remove_channels(efx);
 594	return rc;
 595}
 596
 597/* Channels are shutdown and reinitialised whilst the NIC is running
 598 * to propagate configuration changes (mtu, checksum offload), or
 599 * to clear hardware error conditions
 600 */
 601static void efx_start_datapath(struct efx_nic *efx)
 602{
 603	bool old_rx_scatter = efx->rx_scatter;
 604	struct efx_tx_queue *tx_queue;
 605	struct efx_rx_queue *rx_queue;
 606	struct efx_channel *channel;
 607	size_t rx_buf_len;
 608
 609	/* Calculate the rx buffer allocation parameters required to
 610	 * support the current MTU, including padding for header
 611	 * alignment and overruns.
 612	 */
 613	efx->rx_dma_len = (efx->rx_prefix_size +
 614			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
 615			   efx->type->rx_buffer_padding);
 616	rx_buf_len = (sizeof(struct efx_rx_page_state) +
 617		      efx->rx_ip_align + efx->rx_dma_len);
 618	if (rx_buf_len <= PAGE_SIZE) {
 619		efx->rx_scatter = efx->type->always_rx_scatter;
 620		efx->rx_buffer_order = 0;
 621	} else if (efx->type->can_rx_scatter) {
 622		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
 623		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
 624			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
 625				       EFX_RX_BUF_ALIGNMENT) >
 626			     PAGE_SIZE);
 627		efx->rx_scatter = true;
 628		efx->rx_dma_len = EFX_RX_USR_BUF_SIZE;
 629		efx->rx_buffer_order = 0;
 630	} else {
 631		efx->rx_scatter = false;
 632		efx->rx_buffer_order = get_order(rx_buf_len);
 633	}
 634
 635	efx_rx_config_page_split(efx);
 636	if (efx->rx_buffer_order)
 637		netif_dbg(efx, drv, efx->net_dev,
 638			  "RX buf len=%u; page order=%u batch=%u\n",
 639			  efx->rx_dma_len, efx->rx_buffer_order,
 640			  efx->rx_pages_per_batch);
 641	else
 642		netif_dbg(efx, drv, efx->net_dev,
 643			  "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
 644			  efx->rx_dma_len, efx->rx_page_buf_step,
 645			  efx->rx_bufs_per_page, efx->rx_pages_per_batch);
 646
 647	/* RX filters may also have scatter-enabled flags */
 648	if (efx->rx_scatter != old_rx_scatter)
 649		efx->type->filter_update_rx_scatter(efx);
 650
 651	/* We must keep at least one descriptor in a TX ring empty.
 652	 * We could avoid this when the queue size does not exactly
 653	 * match the hardware ring size, but it's not that important.
 654	 * Therefore we stop the queue when one more skb might fill
 655	 * the ring completely.  We wake it when half way back to
 656	 * empty.
 657	 */
 658	efx->txq_stop_thresh = efx->txq_entries - efx_tx_max_skb_descs(efx);
 659	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;
 660
 661	/* Initialise the channels */
 662	efx_for_each_channel(channel, efx) {
 663		efx_for_each_channel_tx_queue(tx_queue, channel) {
 664			efx_init_tx_queue(tx_queue);
 665			atomic_inc(&efx->active_queues);
 666		}
 667
 668		efx_for_each_channel_rx_queue(rx_queue, channel) {
 669			efx_init_rx_queue(rx_queue);
 670			atomic_inc(&efx->active_queues);
 671			efx_stop_eventq(channel);
 672			efx_fast_push_rx_descriptors(rx_queue, false);
 673			efx_start_eventq(channel);
 674		}
 675
 676		WARN_ON(channel->rx_pkt_n_frags);
 677	}
 678
 679	efx_ptp_start_datapath(efx);
 680
 681	if (netif_device_present(efx->net_dev))
 682		netif_tx_wake_all_queues(efx->net_dev);
 683}
 684
 685static void efx_stop_datapath(struct efx_nic *efx)
 686{
 687	struct efx_channel *channel;
 688	struct efx_tx_queue *tx_queue;
 689	struct efx_rx_queue *rx_queue;
 690	int rc;
 691
 692	EFX_ASSERT_RESET_SERIALISED(efx);
 693	BUG_ON(efx->port_enabled);
 694
 695	efx_ptp_stop_datapath(efx);
 696
 697	/* Stop RX refill */
 698	efx_for_each_channel(channel, efx) {
 699		efx_for_each_channel_rx_queue(rx_queue, channel)
 700			rx_queue->refill_enabled = false;
 701	}
 702
 703	efx_for_each_channel(channel, efx) {
 704		/* RX packet processing is pipelined, so wait for the
 705		 * NAPI handler to complete.  At least event queue 0
 706		 * might be kept active by non-data events, so don't
 707		 * use napi_synchronize() but actually disable NAPI
 708		 * temporarily.
 709		 */
 710		if (efx_channel_has_rx_queue(channel)) {
 711			efx_stop_eventq(channel);
 712			efx_start_eventq(channel);
 713		}
 714	}
 715
 716	rc = efx->type->fini_dmaq(efx);
 717	if (rc && EFX_WORKAROUND_7803(efx)) {
 718		/* Schedule a reset to recover from the flush failure. The
 719		 * descriptor caches reference memory we're about to free,
 720		 * but falcon_reconfigure_mac_wrapper() won't reconnect
 721		 * the MACs because of the pending reset.
 722		 */
 723		netif_err(efx, drv, efx->net_dev,
 724			  "Resetting to recover from flush failure\n");
 725		efx_schedule_reset(efx, RESET_TYPE_ALL);
 726	} else if (rc) {
 727		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
 728	} else {
 729		netif_dbg(efx, drv, efx->net_dev,
 730			  "successfully flushed all queues\n");
 731	}
 732
 733	efx_for_each_channel(channel, efx) {
 734		efx_for_each_channel_rx_queue(rx_queue, channel)
 735			efx_fini_rx_queue(rx_queue);
 736		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
 737			efx_fini_tx_queue(tx_queue);
 738	}
 739}
 740
 741static void efx_remove_channel(struct efx_channel *channel)
 742{
 743	struct efx_tx_queue *tx_queue;
 744	struct efx_rx_queue *rx_queue;
 745
 746	netif_dbg(channel->efx, drv, channel->efx->net_dev,
 747		  "destroy chan %d\n", channel->channel);
 748
 749	efx_for_each_channel_rx_queue(rx_queue, channel)
 750		efx_remove_rx_queue(rx_queue);
 751	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
 752		efx_remove_tx_queue(tx_queue);
 753	efx_remove_eventq(channel);
 754	channel->type->post_remove(channel);
 755}
 756
 757static void efx_remove_channels(struct efx_nic *efx)
 758{
 759	struct efx_channel *channel;
 760
 761	efx_for_each_channel(channel, efx)
 762		efx_remove_channel(channel);
 763}
 764
 765int
 766efx_realloc_channels(struct efx_nic *efx, u32 rxq_entries, u32 txq_entries)
 767{
 768	struct efx_channel *other_channel[EFX_MAX_CHANNELS], *channel;
 769	u32 old_rxq_entries, old_txq_entries;
 770	unsigned i, next_buffer_table = 0;
 771	int rc, rc2;
 772
 773	rc = efx_check_disabled(efx);
 774	if (rc)
 775		return rc;
 776
 777	/* Not all channels should be reallocated. We must avoid
 778	 * reallocating their buffer table entries.
 779	 */
 780	efx_for_each_channel(channel, efx) {
 781		struct efx_rx_queue *rx_queue;
 782		struct efx_tx_queue *tx_queue;
 783
 784		if (channel->type->copy)
 785			continue;
 786		next_buffer_table = max(next_buffer_table,
 787					channel->eventq.index +
 788					channel->eventq.entries);
 789		efx_for_each_channel_rx_queue(rx_queue, channel)
 790			next_buffer_table = max(next_buffer_table,
 791						rx_queue->rxd.index +
 792						rx_queue->rxd.entries);
 793		efx_for_each_channel_tx_queue(tx_queue, channel)
 794			next_buffer_table = max(next_buffer_table,
 795						tx_queue->txd.index +
 796						tx_queue->txd.entries);
 797	}
 798
 799	efx_device_detach_sync(efx);
 800	efx_stop_all(efx);
 801	efx_soft_disable_interrupts(efx);
 802
 803	/* Clone channels (where possible) */
 804	memset(other_channel, 0, sizeof(other_channel));
 805	for (i = 0; i < efx->n_channels; i++) {
 806		channel = efx->channel[i];
 807		if (channel->type->copy)
 808			channel = channel->type->copy(channel);
 809		if (!channel) {
 810			rc = -ENOMEM;
 811			goto out;
 812		}
 813		other_channel[i] = channel;
 814	}
 815
 816	/* Swap entry counts and channel pointers */
 817	old_rxq_entries = efx->rxq_entries;
 818	old_txq_entries = efx->txq_entries;
 819	efx->rxq_entries = rxq_entries;
 820	efx->txq_entries = txq_entries;
 821	for (i = 0; i < efx->n_channels; i++) {
 822		channel = efx->channel[i];
 823		efx->channel[i] = other_channel[i];
 824		other_channel[i] = channel;
 825	}
 826
 827	/* Restart buffer table allocation */
 828	efx->next_buffer_table = next_buffer_table;
 829
 830	for (i = 0; i < efx->n_channels; i++) {
 831		channel = efx->channel[i];
 832		if (!channel->type->copy)
 833			continue;
 834		rc = efx_probe_channel(channel);
 835		if (rc)
 836			goto rollback;
 837		efx_init_napi_channel(efx->channel[i]);
 838	}
 839
 840out:
 841	/* Destroy unused channel structures */
 842	for (i = 0; i < efx->n_channels; i++) {
 843		channel = other_channel[i];
 844		if (channel && channel->type->copy) {
 845			efx_fini_napi_channel(channel);
 846			efx_remove_channel(channel);
 847			kfree(channel);
 848		}
 849	}
 850
 851	rc2 = efx_soft_enable_interrupts(efx);
 852	if (rc2) {
 853		rc = rc ? rc : rc2;
 854		netif_err(efx, drv, efx->net_dev,
 855			  "unable to restart interrupts on channel reallocation\n");
 856		efx_schedule_reset(efx, RESET_TYPE_DISABLE);
 857	} else {
 858		efx_start_all(efx);
 859		netif_device_attach(efx->net_dev);
 860	}
 861	return rc;
 862
 863rollback:
 864	/* Swap back */
 865	efx->rxq_entries = old_rxq_entries;
 866	efx->txq_entries = old_txq_entries;
 867	for (i = 0; i < efx->n_channels; i++) {
 868		channel = efx->channel[i];
 869		efx->channel[i] = other_channel[i];
 870		other_channel[i] = channel;
 871	}
 872	goto out;
 873}
 874
 875void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
 876{
 877	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
 878}
 879
 880static const struct efx_channel_type efx_default_channel_type = {
 881	.pre_probe		= efx_channel_dummy_op_int,
 882	.post_remove		= efx_channel_dummy_op_void,
 883	.get_name		= efx_get_channel_name,
 884	.copy			= efx_copy_channel,
 885	.keep_eventq		= false,
 886};
 887
 888int efx_channel_dummy_op_int(struct efx_channel *channel)
 889{
 890	return 0;
 891}
 892
 893void efx_channel_dummy_op_void(struct efx_channel *channel)
 894{
 895}
 896
 897/**************************************************************************
 898 *
 899 * Port handling
 900 *
 901 **************************************************************************/
 902
 903/* This ensures that the kernel is kept informed (via
 904 * netif_carrier_on/off) of the link status, and also maintains the
 905 * link status's stop on the port's TX queue.
 906 */
 907void efx_link_status_changed(struct efx_nic *efx)
 908{
 909	struct efx_link_state *link_state = &efx->link_state;
 910
 911	/* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
 912	 * that no events are triggered between unregister_netdev() and the
 913	 * driver unloading. A more general condition is that NETDEV_CHANGE
 914	 * can only be generated between NETDEV_UP and NETDEV_DOWN */
 915	if (!netif_running(efx->net_dev))
 916		return;
 917
 918	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
 919		efx->n_link_state_changes++;
 920
 921		if (link_state->up)
 922			netif_carrier_on(efx->net_dev);
 923		else
 924			netif_carrier_off(efx->net_dev);
 925	}
 926
 927	/* Status message for kernel log */
 928	if (link_state->up)
 929		netif_info(efx, link, efx->net_dev,
 930			   "link up at %uMbps %s-duplex (MTU %d)\n",
 931			   link_state->speed, link_state->fd ? "full" : "half",
 932			   efx->net_dev->mtu);
 933	else
 934		netif_info(efx, link, efx->net_dev, "link down\n");
 935}
 936
 937void efx_link_set_advertising(struct efx_nic *efx, u32 advertising)
 938{
 939	efx->link_advertising = advertising;
 940	if (advertising) {
 941		if (advertising & ADVERTISED_Pause)
 942			efx->wanted_fc |= (EFX_FC_TX | EFX_FC_RX);
 943		else
 944			efx->wanted_fc &= ~(EFX_FC_TX | EFX_FC_RX);
 945		if (advertising & ADVERTISED_Asym_Pause)
 946			efx->wanted_fc ^= EFX_FC_TX;
 947	}
 948}
 949
 950void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
 951{
 952	efx->wanted_fc = wanted_fc;
 953	if (efx->link_advertising) {
 954		if (wanted_fc & EFX_FC_RX)
 955			efx->link_advertising |= (ADVERTISED_Pause |
 956						  ADVERTISED_Asym_Pause);
 957		else
 958			efx->link_advertising &= ~(ADVERTISED_Pause |
 959						   ADVERTISED_Asym_Pause);
 960		if (wanted_fc & EFX_FC_TX)
 961			efx->link_advertising ^= ADVERTISED_Asym_Pause;
 962	}
 963}
 964
 965static void efx_fini_port(struct efx_nic *efx);
 966
 967/* We assume that efx->type->reconfigure_mac will always try to sync RX
 968 * filters and therefore needs to read-lock the filter table against freeing
 969 */
 970void efx_mac_reconfigure(struct efx_nic *efx)
 971{
 972	down_read(&efx->filter_sem);
 973	efx->type->reconfigure_mac(efx);
 974	up_read(&efx->filter_sem);
 975}
 976
 977/* Push loopback/power/transmit disable settings to the PHY, and reconfigure
 978 * the MAC appropriately. All other PHY configuration changes are pushed
 979 * through phy_op->set_settings(), and pushed asynchronously to the MAC
 980 * through efx_monitor().
 981 *
 982 * Callers must hold the mac_lock
 983 */
 984int __efx_reconfigure_port(struct efx_nic *efx)
 985{
 986	enum efx_phy_mode phy_mode;
 987	int rc;
 988
 989	WARN_ON(!mutex_is_locked(&efx->mac_lock));
 990
 991	/* Disable PHY transmit in mac level loopbacks */
 992	phy_mode = efx->phy_mode;
 993	if (LOOPBACK_INTERNAL(efx))
 994		efx->phy_mode |= PHY_MODE_TX_DISABLED;
 995	else
 996		efx->phy_mode &= ~PHY_MODE_TX_DISABLED;
 997
 998	rc = efx->type->reconfigure_port(efx);
 999
1000	if (rc)
1001		efx->phy_mode = phy_mode;
1002
1003	return rc;
1004}
1005
1006/* Reinitialise the MAC to pick up new PHY settings, even if the port is
1007 * disabled. */
1008int efx_reconfigure_port(struct efx_nic *efx)
1009{
1010	int rc;
1011
1012	EFX_ASSERT_RESET_SERIALISED(efx);
1013
1014	mutex_lock(&efx->mac_lock);
1015	rc = __efx_reconfigure_port(efx);
1016	mutex_unlock(&efx->mac_lock);
1017
1018	return rc;
1019}
1020
1021/* Asynchronous work item for changing MAC promiscuity and multicast
1022 * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
1023 * MAC directly. */
1024static void efx_mac_work(struct work_struct *data)
1025{
1026	struct efx_nic *efx = container_of(data, struct efx_nic, mac_work);
1027
1028	mutex_lock(&efx->mac_lock);
1029	if (efx->port_enabled)
1030		efx_mac_reconfigure(efx);
1031	mutex_unlock(&efx->mac_lock);
1032}
1033
1034static int efx_probe_port(struct efx_nic *efx)
1035{
1036	int rc;
1037
1038	netif_dbg(efx, probe, efx->net_dev, "create port\n");
1039
1040	if (phy_flash_cfg)
1041		efx->phy_mode = PHY_MODE_SPECIAL;
1042
1043	/* Connect up MAC/PHY operations table */
1044	rc = efx->type->probe_port(efx);
1045	if (rc)
1046		return rc;
1047
1048	/* Initialise MAC address to permanent address */
1049	ether_addr_copy(efx->net_dev->dev_addr, efx->net_dev->perm_addr);
1050
1051	return 0;
1052}
1053
1054static int efx_init_port(struct efx_nic *efx)
1055{
1056	int rc;
1057
1058	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1059
1060	mutex_lock(&efx->mac_lock);
1061
1062	rc = efx->phy_op->init(efx);
1063	if (rc)
1064		goto fail1;
1065
1066	efx->port_initialized = true;
1067
1068	/* Reconfigure the MAC before creating dma queues (required for
1069	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1070	efx_mac_reconfigure(efx);
1071
1072	/* Ensure the PHY advertises the correct flow control settings */
1073	rc = efx->phy_op->reconfigure(efx);
1074	if (rc && rc != -EPERM)
1075		goto fail2;
1076
1077	mutex_unlock(&efx->mac_lock);
1078	return 0;
1079
1080fail2:
1081	efx->phy_op->fini(efx);
1082fail1:
1083	mutex_unlock(&efx->mac_lock);
1084	return rc;
1085}
1086
1087static void efx_start_port(struct efx_nic *efx)
1088{
1089	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1090	BUG_ON(efx->port_enabled);
1091
1092	mutex_lock(&efx->mac_lock);
1093	efx->port_enabled = true;
1094
1095	/* Ensure MAC ingress/egress is enabled */
1096	efx_mac_reconfigure(efx);
1097
1098	mutex_unlock(&efx->mac_lock);
1099}
1100
1101/* Cancel work for MAC reconfiguration, periodic hardware monitoring
1102 * and the async self-test, wait for them to finish and prevent them
1103 * being scheduled again.  This doesn't cover online resets, which
1104 * should only be cancelled when removing the device.
1105 */
1106static void efx_stop_port(struct efx_nic *efx)
1107{
1108	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1109
1110	EFX_ASSERT_RESET_SERIALISED(efx);
1111
1112	mutex_lock(&efx->mac_lock);
1113	efx->port_enabled = false;
1114	mutex_unlock(&efx->mac_lock);
1115
1116	/* Serialise against efx_set_multicast_list() */
1117	netif_addr_lock_bh(efx->net_dev);
1118	netif_addr_unlock_bh(efx->net_dev);
1119
1120	cancel_delayed_work_sync(&efx->monitor_work);
1121	efx_selftest_async_cancel(efx);
1122	cancel_work_sync(&efx->mac_work);
1123}
1124
1125static void efx_fini_port(struct efx_nic *efx)
1126{
1127	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1128
1129	if (!efx->port_initialized)
1130		return;
1131
1132	efx->phy_op->fini(efx);
1133	efx->port_initialized = false;
1134
1135	efx->link_state.up = false;
1136	efx_link_status_changed(efx);
1137}
1138
1139static void efx_remove_port(struct efx_nic *efx)
1140{
1141	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1142
1143	efx->type->remove_port(efx);
1144}
1145
1146/**************************************************************************
1147 *
1148 * NIC handling
1149 *
1150 **************************************************************************/
1151
1152static LIST_HEAD(efx_primary_list);
1153static LIST_HEAD(efx_unassociated_list);
1154
1155static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right)
1156{
1157	return left->type == right->type &&
1158		left->vpd_sn && right->vpd_sn &&
1159		!strcmp(left->vpd_sn, right->vpd_sn);
1160}
1161
1162static void efx_associate(struct efx_nic *efx)
1163{
1164	struct efx_nic *other, *next;
1165
1166	if (efx->primary == efx) {
1167		/* Adding primary function; look for secondaries */
1168
1169		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
1170		list_add_tail(&efx->node, &efx_primary_list);
1171
1172		list_for_each_entry_safe(other, next, &efx_unassociated_list,
1173					 node) {
1174			if (efx_same_controller(efx, other)) {
1175				list_del(&other->node);
1176				netif_dbg(other, probe, other->net_dev,
1177					  "moving to secondary list of %s %s\n",
1178					  pci_name(efx->pci_dev),
1179					  efx->net_dev->name);
1180				list_add_tail(&other->node,
1181					      &efx->secondary_list);
1182				other->primary = efx;
1183			}
1184		}
1185	} else {
1186		/* Adding secondary function; look for primary */
1187
1188		list_for_each_entry(other, &efx_primary_list, node) {
1189			if (efx_same_controller(efx, other)) {
1190				netif_dbg(efx, probe, efx->net_dev,
1191					  "adding to secondary list of %s %s\n",
1192					  pci_name(other->pci_dev),
1193					  other->net_dev->name);
1194				list_add_tail(&efx->node,
1195					      &other->secondary_list);
1196				efx->primary = other;
1197				return;
1198			}
1199		}
1200
1201		netif_dbg(efx, probe, efx->net_dev,
1202			  "adding to unassociated list\n");
1203		list_add_tail(&efx->node, &efx_unassociated_list);
1204	}
1205}
1206
1207static void efx_dissociate(struct efx_nic *efx)
1208{
1209	struct efx_nic *other, *next;
1210
1211	list_del(&efx->node);
1212	efx->primary = NULL;
1213
1214	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
1215		list_del(&other->node);
1216		netif_dbg(other, probe, other->net_dev,
1217			  "moving to unassociated list\n");
1218		list_add_tail(&other->node, &efx_unassociated_list);
1219		other->primary = NULL;
1220	}
1221}
1222
1223/* This configures the PCI device to enable I/O and DMA. */
1224static int efx_init_io(struct efx_nic *efx)
1225{
1226	struct pci_dev *pci_dev = efx->pci_dev;
1227	dma_addr_t dma_mask = efx->type->max_dma_mask;
1228	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1229	int rc, bar;
1230
1231	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1232
1233	bar = efx->type->mem_bar;
1234
1235	rc = pci_enable_device(pci_dev);
1236	if (rc) {
1237		netif_err(efx, probe, efx->net_dev,
1238			  "failed to enable PCI device\n");
1239		goto fail1;
1240	}
1241
1242	pci_set_master(pci_dev);
1243
1244	/* Set the PCI DMA mask.  Try all possibilities from our
1245	 * genuine mask down to 32 bits, because some architectures
1246	 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
1247	 * masks event though they reject 46 bit masks.
1248	 */
1249	while (dma_mask > 0x7fffffffUL) {
1250		rc = dma_set_mask_and_coherent(&pci_dev->dev, dma_mask);
1251		if (rc == 0)
1252			break;
1253		dma_mask >>= 1;
1254	}
1255	if (rc) {
1256		netif_err(efx, probe, efx->net_dev,
1257			  "could not find a suitable DMA mask\n");
1258		goto fail2;
1259	}
1260	netif_dbg(efx, probe, efx->net_dev,
1261		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1262
1263	efx->membase_phys = pci_resource_start(efx->pci_dev, bar);
1264	rc = pci_request_region(pci_dev, bar, "sfc");
1265	if (rc) {
1266		netif_err(efx, probe, efx->net_dev,
1267			  "request for memory BAR failed\n");
1268		rc = -EIO;
1269		goto fail3;
1270	}
1271	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1272	if (!efx->membase) {
1273		netif_err(efx, probe, efx->net_dev,
1274			  "could not map memory BAR at %llx+%x\n",
1275			  (unsigned long long)efx->membase_phys, mem_map_size);
1276		rc = -ENOMEM;
1277		goto fail4;
1278	}
1279	netif_dbg(efx, probe, efx->net_dev,
1280		  "memory BAR at %llx+%x (virtual %p)\n",
1281		  (unsigned long long)efx->membase_phys, mem_map_size,
1282		  efx->membase);
1283
1284	return 0;
1285
1286 fail4:
1287	pci_release_region(efx->pci_dev, bar);
1288 fail3:
1289	efx->membase_phys = 0;
1290 fail2:
1291	pci_disable_device(efx->pci_dev);
1292 fail1:
1293	return rc;
1294}
1295
1296static void efx_fini_io(struct efx_nic *efx)
1297{
1298	int bar;
1299
1300	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1301
1302	if (efx->membase) {
1303		iounmap(efx->membase);
1304		efx->membase = NULL;
1305	}
1306
1307	if (efx->membase_phys) {
1308		bar = efx->type->mem_bar;
1309		pci_release_region(efx->pci_dev, bar);
1310		efx->membase_phys = 0;
1311	}
1312
1313	/* Don't disable bus-mastering if VFs are assigned */
1314	if (!pci_vfs_assigned(efx->pci_dev))
1315		pci_disable_device(efx->pci_dev);
1316}
1317
1318void efx_set_default_rx_indir_table(struct efx_nic *efx)
1319{
1320	size_t i;
1321
1322	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
1323		efx->rx_indir_table[i] =
1324			ethtool_rxfh_indir_default(i, efx->rss_spread);
1325}
1326
1327static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1328{
1329	cpumask_var_t thread_mask;
1330	unsigned int count;
1331	int cpu;
1332
1333	if (rss_cpus) {
1334		count = rss_cpus;
1335	} else {
1336		if (unlikely(!zalloc_cpumask_var(&thread_mask, GFP_KERNEL))) {
1337			netif_warn(efx, probe, efx->net_dev,
1338				   "RSS disabled due to allocation failure\n");
1339			return 1;
1340		}
1341
1342		count = 0;
1343		for_each_online_cpu(cpu) {
1344			if (!cpumask_test_cpu(cpu, thread_mask)) {
1345				++count;
1346				cpumask_or(thread_mask, thread_mask,
1347					   topology_sibling_cpumask(cpu));
1348			}
1349		}
1350
1351		free_cpumask_var(thread_mask);
1352	}
1353
1354	/* If RSS is requested for the PF *and* VFs then we can't write RSS
1355	 * table entries that are inaccessible to VFs
1356	 */
1357#ifdef CONFIG_SFC_SRIOV
1358	if (efx->type->sriov_wanted) {
1359		if (efx->type->sriov_wanted(efx) && efx_vf_size(efx) > 1 &&
1360		    count > efx_vf_size(efx)) {
1361			netif_warn(efx, probe, efx->net_dev,
1362				   "Reducing number of RSS channels from %u to %u for "
1363				   "VF support. Increase vf-msix-limit to use more "
1364				   "channels on the PF.\n",
1365				   count, efx_vf_size(efx));
1366			count = efx_vf_size(efx);
1367		}
1368	}
1369#endif
1370
1371	return count;
1372}
1373
1374/* Probe the number and type of interrupts we are able to obtain, and
1375 * the resulting numbers of channels and RX queues.
1376 */
1377static int efx_probe_interrupts(struct efx_nic *efx)
1378{
1379	unsigned int extra_channels = 0;
1380	unsigned int i, j;
1381	int rc;
1382
1383	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
1384		if (efx->extra_channel_type[i])
1385			++extra_channels;
1386
1387	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1388		struct msix_entry xentries[EFX_MAX_CHANNELS];
1389		unsigned int n_channels;
1390
1391		n_channels = efx_wanted_parallelism(efx);
1392		if (efx_separate_tx_channels)
1393			n_channels *= 2;
1394		n_channels += extra_channels;
1395		n_channels = min(n_channels, efx->max_channels);
1396
1397		for (i = 0; i < n_channels; i++)
1398			xentries[i].entry = i;
1399		rc = pci_enable_msix_range(efx->pci_dev,
1400					   xentries, 1, n_channels);
1401		if (rc < 0) {
1402			/* Fall back to single channel MSI */
1403			efx->interrupt_mode = EFX_INT_MODE_MSI;
1404			netif_err(efx, drv, efx->net_dev,
1405				  "could not enable MSI-X\n");
1406		} else if (rc < n_channels) {
1407			netif_err(efx, drv, efx->net_dev,
1408				  "WARNING: Insufficient MSI-X vectors"
1409				  " available (%d < %u).\n", rc, n_channels);
1410			netif_err(efx, drv, efx->net_dev,
1411				  "WARNING: Performance may be reduced.\n");
1412			n_channels = rc;
1413		}
1414
1415		if (rc > 0) {
1416			efx->n_channels = n_channels;
1417			if (n_channels > extra_channels)
1418				n_channels -= extra_channels;
1419			if (efx_separate_tx_channels) {
1420				efx->n_tx_channels = min(max(n_channels / 2,
1421							     1U),
1422							 efx->max_tx_channels);
1423				efx->n_rx_channels = max(n_channels -
1424							 efx->n_tx_channels,
1425							 1U);
1426			} else {
1427				efx->n_tx_channels = min(n_channels,
1428							 efx->max_tx_channels);
1429				efx->n_rx_channels = n_channels;
1430			}
1431			for (i = 0; i < efx->n_channels; i++)
1432				efx_get_channel(efx, i)->irq =
1433					xentries[i].vector;
1434		}
1435	}
1436
1437	/* Try single interrupt MSI */
1438	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1439		efx->n_channels = 1;
1440		efx->n_rx_channels = 1;
1441		efx->n_tx_channels = 1;
1442		rc = pci_enable_msi(efx->pci_dev);
1443		if (rc == 0) {
1444			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1445		} else {
1446			netif_err(efx, drv, efx->net_dev,
1447				  "could not enable MSI\n");
1448			efx->interrupt_mode = EFX_INT_MODE_LEGACY;
1449		}
1450	}
1451
1452	/* Assume legacy interrupts */
1453	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1454		efx->n_channels = 1 + (efx_separate_tx_channels ? 1 : 0);
1455		efx->n_rx_channels = 1;
1456		efx->n_tx_channels = 1;
1457		efx->legacy_irq = efx->pci_dev->irq;
1458	}
1459
1460	/* Assign extra channels if possible */
1461	j = efx->n_channels;
1462	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++) {
1463		if (!efx->extra_channel_type[i])
1464			continue;
1465		if (efx->interrupt_mode != EFX_INT_MODE_MSIX ||
1466		    efx->n_channels <= extra_channels) {
1467			efx->extra_channel_type[i]->handle_no_channel(efx);
1468		} else {
1469			--j;
1470			efx_get_channel(efx, j)->type =
1471				efx->extra_channel_type[i];
1472		}
1473	}
1474
1475	/* RSS might be usable on VFs even if it is disabled on the PF */
1476#ifdef CONFIG_SFC_SRIOV
1477	if (efx->type->sriov_wanted) {
1478		efx->rss_spread = ((efx->n_rx_channels > 1 ||
1479				    !efx->type->sriov_wanted(efx)) ?
1480				   efx->n_rx_channels : efx_vf_size(efx));
1481		return 0;
1482	}
1483#endif
1484	efx->rss_spread = efx->n_rx_channels;
1485
1486	return 0;
1487}
1488
1489static int efx_soft_enable_interrupts(struct efx_nic *efx)
1490{
1491	struct efx_channel *channel, *end_channel;
1492	int rc;
1493
1494	BUG_ON(efx->state == STATE_DISABLED);
1495
1496	efx->irq_soft_enabled = true;
1497	smp_wmb();
1498
1499	efx_for_each_channel(channel, efx) {
1500		if (!channel->type->keep_eventq) {
1501			rc = efx_init_eventq(channel);
1502			if (rc)
1503				goto fail;
1504		}
1505		efx_start_eventq(channel);
1506	}
1507
1508	efx_mcdi_mode_event(efx);
1509
1510	return 0;
1511fail:
1512	end_channel = channel;
1513	efx_for_each_channel(channel, efx) {
1514		if (channel == end_channel)
1515			break;
1516		efx_stop_eventq(channel);
1517		if (!channel->type->keep_eventq)
1518			efx_fini_eventq(channel);
1519	}
1520
1521	return rc;
1522}
1523
1524static void efx_soft_disable_interrupts(struct efx_nic *efx)
1525{
1526	struct efx_channel *channel;
1527
1528	if (efx->state == STATE_DISABLED)
1529		return;
1530
1531	efx_mcdi_mode_poll(efx);
1532
1533	efx->irq_soft_enabled = false;
1534	smp_wmb();
1535
1536	if (efx->legacy_irq)
1537		synchronize_irq(efx->legacy_irq);
1538
1539	efx_for_each_channel(channel, efx) {
1540		if (channel->irq)
1541			synchronize_irq(channel->irq);
1542
1543		efx_stop_eventq(channel);
1544		if (!channel->type->keep_eventq)
1545			efx_fini_eventq(channel);
1546	}
1547
1548	/* Flush the asynchronous MCDI request queue */
1549	efx_mcdi_flush_async(efx);
1550}
1551
1552static int efx_enable_interrupts(struct efx_nic *efx)
1553{
1554	struct efx_channel *channel, *end_channel;
1555	int rc;
1556
1557	BUG_ON(efx->state == STATE_DISABLED);
1558
1559	if (efx->eeh_disabled_legacy_irq) {
1560		enable_irq(efx->legacy_irq);
1561		efx->eeh_disabled_legacy_irq = false;
1562	}
1563
1564	efx->type->irq_enable_master(efx);
1565
1566	efx_for_each_channel(channel, efx) {
1567		if (channel->type->keep_eventq) {
1568			rc = efx_init_eventq(channel);
1569			if (rc)
1570				goto fail;
1571		}
1572	}
1573
1574	rc = efx_soft_enable_interrupts(efx);
1575	if (rc)
1576		goto fail;
1577
1578	return 0;
1579
1580fail:
1581	end_channel = channel;
1582	efx_for_each_channel(channel, efx) {
1583		if (channel == end_channel)
1584			break;
1585		if (channel->type->keep_eventq)
1586			efx_fini_eventq(channel);
1587	}
1588
1589	efx->type->irq_disable_non_ev(efx);
1590
1591	return rc;
1592}
1593
1594static void efx_disable_interrupts(struct efx_nic *efx)
1595{
1596	struct efx_channel *channel;
1597
1598	efx_soft_disable_interrupts(efx);
1599
1600	efx_for_each_channel(channel, efx) {
1601		if (channel->type->keep_eventq)
1602			efx_fini_eventq(channel);
1603	}
1604
1605	efx->type->irq_disable_non_ev(efx);
1606}
1607
1608static void efx_remove_interrupts(struct efx_nic *efx)
1609{
1610	struct efx_channel *channel;
1611
1612	/* Remove MSI/MSI-X interrupts */
1613	efx_for_each_channel(channel, efx)
1614		channel->irq = 0;
1615	pci_disable_msi(efx->pci_dev);
1616	pci_disable_msix(efx->pci_dev);
1617
1618	/* Remove legacy interrupt */
1619	efx->legacy_irq = 0;
1620}
1621
1622static void efx_set_channels(struct efx_nic *efx)
1623{
1624	struct efx_channel *channel;
1625	struct efx_tx_queue *tx_queue;
1626
1627	efx->tx_channel_offset =
1628		efx_separate_tx_channels ?
1629		efx->n_channels - efx->n_tx_channels : 0;
1630
1631	/* We need to mark which channels really have RX and TX
1632	 * queues, and adjust the TX queue numbers if we have separate
1633	 * RX-only and TX-only channels.
1634	 */
1635	efx_for_each_channel(channel, efx) {
1636		if (channel->channel < efx->n_rx_channels)
1637			channel->rx_queue.core_index = channel->channel;
1638		else
1639			channel->rx_queue.core_index = -1;
1640
1641		efx_for_each_channel_tx_queue(tx_queue, channel)
1642			tx_queue->queue -= (efx->tx_channel_offset *
1643					    EFX_TXQ_TYPES);
1644	}
1645}
1646
1647static int efx_probe_nic(struct efx_nic *efx)
1648{
1649	int rc;
1650
1651	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1652
1653	/* Carry out hardware-type specific initialisation */
1654	rc = efx->type->probe(efx);
1655	if (rc)
1656		return rc;
1657
1658	do {
1659		if (!efx->max_channels || !efx->max_tx_channels) {
1660			netif_err(efx, drv, efx->net_dev,
1661				  "Insufficient resources to allocate"
1662				  " any channels\n");
1663			rc = -ENOSPC;
1664			goto fail1;
1665		}
1666
1667		/* Determine the number of channels and queues by trying
1668		 * to hook in MSI-X interrupts.
1669		 */
1670		rc = efx_probe_interrupts(efx);
1671		if (rc)
1672			goto fail1;
1673
1674		efx_set_channels(efx);
 
 
1675
1676		/* dimension_resources can fail with EAGAIN */
1677		rc = efx->type->dimension_resources(efx);
1678		if (rc != 0 && rc != -EAGAIN)
1679			goto fail2;
1680
1681		if (rc == -EAGAIN)
1682			/* try again with new max_channels */
1683			efx_remove_interrupts(efx);
1684
1685	} while (rc == -EAGAIN);
1686
1687	if (efx->n_channels > 1)
1688		netdev_rss_key_fill(&efx->rx_hash_key,
1689				    sizeof(efx->rx_hash_key));
1690	efx_set_default_rx_indir_table(efx);
1691
1692	netif_set_real_num_tx_queues(efx->net_dev, efx->n_tx_channels);
1693	netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels);
1694
1695	/* Initialise the interrupt moderation settings */
 
1696	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
1697				true);
1698
1699	return 0;
1700
1701fail2:
1702	efx_remove_interrupts(efx);
1703fail1:
1704	efx->type->remove(efx);
1705	return rc;
1706}
1707
1708static void efx_remove_nic(struct efx_nic *efx)
1709{
1710	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1711
1712	efx_remove_interrupts(efx);
1713	efx->type->remove(efx);
1714}
1715
1716static int efx_probe_filters(struct efx_nic *efx)
1717{
1718	int rc;
1719
1720	spin_lock_init(&efx->filter_lock);
1721	init_rwsem(&efx->filter_sem);
1722	down_write(&efx->filter_sem);
1723	rc = efx->type->filter_table_probe(efx);
1724	if (rc)
1725		goto out_unlock;
1726
1727#ifdef CONFIG_RFS_ACCEL
1728	if (efx->type->offload_features & NETIF_F_NTUPLE) {
1729		efx->rps_flow_id = kcalloc(efx->type->max_rx_ip_filters,
1730					   sizeof(*efx->rps_flow_id),
1731					   GFP_KERNEL);
1732		if (!efx->rps_flow_id) {
1733			efx->type->filter_table_remove(efx);
1734			rc = -ENOMEM;
1735			goto out_unlock;
1736		}
1737	}
1738#endif
1739out_unlock:
1740	up_write(&efx->filter_sem);
1741	return rc;
1742}
1743
1744static void efx_remove_filters(struct efx_nic *efx)
1745{
1746#ifdef CONFIG_RFS_ACCEL
1747	kfree(efx->rps_flow_id);
1748#endif
1749	down_write(&efx->filter_sem);
1750	efx->type->filter_table_remove(efx);
1751	up_write(&efx->filter_sem);
1752}
1753
1754static void efx_restore_filters(struct efx_nic *efx)
1755{
1756	down_read(&efx->filter_sem);
1757	efx->type->filter_table_restore(efx);
1758	up_read(&efx->filter_sem);
1759}
1760
1761/**************************************************************************
1762 *
1763 * NIC startup/shutdown
1764 *
1765 *************************************************************************/
1766
1767static int efx_probe_all(struct efx_nic *efx)
1768{
1769	int rc;
1770
1771	rc = efx_probe_nic(efx);
1772	if (rc) {
1773		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1774		goto fail1;
1775	}
1776
1777	rc = efx_probe_port(efx);
1778	if (rc) {
1779		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1780		goto fail2;
1781	}
1782
1783	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
1784	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
1785		rc = -EINVAL;
1786		goto fail3;
1787	}
1788	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1789
1790#ifdef CONFIG_SFC_SRIOV
1791	rc = efx->type->vswitching_probe(efx);
1792	if (rc) /* not fatal; the PF will still work fine */
1793		netif_warn(efx, probe, efx->net_dev,
1794			   "failed to setup vswitching rc=%d;"
1795			   " VFs may not function\n", rc);
1796#endif
1797
1798	rc = efx_probe_filters(efx);
1799	if (rc) {
1800		netif_err(efx, probe, efx->net_dev,
1801			  "failed to create filter tables\n");
1802		goto fail4;
1803	}
1804
1805	rc = efx_probe_channels(efx);
1806	if (rc)
1807		goto fail5;
1808
 
 
1809	return 0;
1810
1811 fail5:
1812	efx_remove_filters(efx);
1813 fail4:
1814#ifdef CONFIG_SFC_SRIOV
1815	efx->type->vswitching_remove(efx);
1816#endif
1817 fail3:
1818	efx_remove_port(efx);
1819 fail2:
1820	efx_remove_nic(efx);
1821 fail1:
1822	return rc;
1823}
1824
1825/* If the interface is supposed to be running but is not, start
1826 * the hardware and software data path, regular activity for the port
1827 * (MAC statistics, link polling, etc.) and schedule the port to be
1828 * reconfigured.  Interrupts must already be enabled.  This function
1829 * is safe to call multiple times, so long as the NIC is not disabled.
1830 * Requires the RTNL lock.
1831 */
1832static void efx_start_all(struct efx_nic *efx)
1833{
1834	EFX_ASSERT_RESET_SERIALISED(efx);
1835	BUG_ON(efx->state == STATE_DISABLED);
1836
1837	/* Check that it is appropriate to restart the interface. All
1838	 * of these flags are safe to read under just the rtnl lock */
1839	if (efx->port_enabled || !netif_running(efx->net_dev) ||
1840	    efx->reset_pending)
1841		return;
1842
1843	efx_start_port(efx);
1844	efx_start_datapath(efx);
1845
1846	/* Start the hardware monitor if there is one */
1847	if (efx->type->monitor != NULL)
1848		queue_delayed_work(efx->workqueue, &efx->monitor_work,
1849				   efx_monitor_interval);
1850
1851	/* If link state detection is normally event-driven, we have
1852	 * to poll now because we could have missed a change
1853	 */
1854	if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0) {
1855		mutex_lock(&efx->mac_lock);
1856		if (efx->phy_op->poll(efx))
1857			efx_link_status_changed(efx);
1858		mutex_unlock(&efx->mac_lock);
1859	}
1860
1861	efx->type->start_stats(efx);
1862	efx->type->pull_stats(efx);
1863	spin_lock_bh(&efx->stats_lock);
1864	efx->type->update_stats(efx, NULL, NULL);
1865	spin_unlock_bh(&efx->stats_lock);
1866}
1867
1868/* Quiesce the hardware and software data path, and regular activity
1869 * for the port without bringing the link down.  Safe to call multiple
1870 * times with the NIC in almost any state, but interrupts should be
1871 * enabled.  Requires the RTNL lock.
1872 */
1873static void efx_stop_all(struct efx_nic *efx)
1874{
1875	EFX_ASSERT_RESET_SERIALISED(efx);
1876
1877	/* port_enabled can be read safely under the rtnl lock */
1878	if (!efx->port_enabled)
1879		return;
1880
1881	/* update stats before we go down so we can accurately count
1882	 * rx_nodesc_drops
1883	 */
1884	efx->type->pull_stats(efx);
1885	spin_lock_bh(&efx->stats_lock);
1886	efx->type->update_stats(efx, NULL, NULL);
1887	spin_unlock_bh(&efx->stats_lock);
1888	efx->type->stop_stats(efx);
1889	efx_stop_port(efx);
1890
1891	/* Stop the kernel transmit interface.  This is only valid if
1892	 * the device is stopped or detached; otherwise the watchdog
1893	 * may fire immediately.
1894	 */
1895	WARN_ON(netif_running(efx->net_dev) &&
1896		netif_device_present(efx->net_dev));
1897	netif_tx_disable(efx->net_dev);
1898
1899	efx_stop_datapath(efx);
1900}
1901
1902static void efx_remove_all(struct efx_nic *efx)
1903{
1904	efx_remove_channels(efx);
1905	efx_remove_filters(efx);
1906#ifdef CONFIG_SFC_SRIOV
1907	efx->type->vswitching_remove(efx);
1908#endif
1909	efx_remove_port(efx);
1910	efx_remove_nic(efx);
1911}
1912
1913/**************************************************************************
1914 *
1915 * Interrupt moderation
1916 *
1917 **************************************************************************/
1918
1919static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int quantum_ns)
1920{
1921	if (usecs == 0)
1922		return 0;
1923	if (usecs * 1000 < quantum_ns)
1924		return 1; /* never round down to 0 */
1925	return usecs * 1000 / quantum_ns;
 
 
 
 
 
 
 
 
1926}
1927
1928/* Set interrupt moderation parameters */
1929int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
1930			    unsigned int rx_usecs, bool rx_adaptive,
1931			    bool rx_may_override_tx)
1932{
1933	struct efx_channel *channel;
1934	unsigned int irq_mod_max = DIV_ROUND_UP(efx->type->timer_period_max *
1935						efx->timer_quantum_ns,
1936						1000);
1937	unsigned int tx_ticks;
1938	unsigned int rx_ticks;
1939
1940	EFX_ASSERT_RESET_SERIALISED(efx);
1941
1942	if (tx_usecs > irq_mod_max || rx_usecs > irq_mod_max)
 
 
1943		return -EINVAL;
1944
1945	tx_ticks = irq_mod_ticks(tx_usecs, efx->timer_quantum_ns);
1946	rx_ticks = irq_mod_ticks(rx_usecs, efx->timer_quantum_ns);
1947
1948	if (tx_ticks != rx_ticks && efx->tx_channel_offset == 0 &&
1949	    !rx_may_override_tx) {
1950		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
1951			  "RX and TX IRQ moderation must be equal\n");
1952		return -EINVAL;
1953	}
1954
1955	efx->irq_rx_adaptive = rx_adaptive;
1956	efx->irq_rx_moderation = rx_ticks;
1957	efx_for_each_channel(channel, efx) {
1958		if (efx_channel_has_rx_queue(channel))
1959			channel->irq_moderation = rx_ticks;
1960		else if (efx_channel_has_tx_queues(channel))
1961			channel->irq_moderation = tx_ticks;
 
 
1962	}
1963
1964	return 0;
1965}
1966
1967void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
1968			    unsigned int *rx_usecs, bool *rx_adaptive)
1969{
1970	/* We must round up when converting ticks to microseconds
1971	 * because we round down when converting the other way.
1972	 */
1973
1974	*rx_adaptive = efx->irq_rx_adaptive;
1975	*rx_usecs = DIV_ROUND_UP(efx->irq_rx_moderation *
1976				 efx->timer_quantum_ns,
1977				 1000);
1978
1979	/* If channels are shared between RX and TX, so is IRQ
1980	 * moderation.  Otherwise, IRQ moderation is the same for all
1981	 * TX channels and is not adaptive.
1982	 */
1983	if (efx->tx_channel_offset == 0)
1984		*tx_usecs = *rx_usecs;
1985	else
1986		*tx_usecs = DIV_ROUND_UP(
1987			efx->channel[efx->tx_channel_offset]->irq_moderation *
1988			efx->timer_quantum_ns,
1989			1000);
1990}
1991
1992/**************************************************************************
1993 *
1994 * Hardware monitor
1995 *
1996 **************************************************************************/
1997
1998/* Run periodically off the general workqueue */
1999static void efx_monitor(struct work_struct *data)
2000{
2001	struct efx_nic *efx = container_of(data, struct efx_nic,
2002					   monitor_work.work);
2003
2004	netif_vdbg(efx, timer, efx->net_dev,
2005		   "hardware monitor executing on CPU %d\n",
2006		   raw_smp_processor_id());
2007	BUG_ON(efx->type->monitor == NULL);
2008
2009	/* If the mac_lock is already held then it is likely a port
2010	 * reconfiguration is already in place, which will likely do
2011	 * most of the work of monitor() anyway. */
2012	if (mutex_trylock(&efx->mac_lock)) {
2013		if (efx->port_enabled)
2014			efx->type->monitor(efx);
2015		mutex_unlock(&efx->mac_lock);
2016	}
2017
2018	queue_delayed_work(efx->workqueue, &efx->monitor_work,
2019			   efx_monitor_interval);
2020}
2021
2022/**************************************************************************
2023 *
2024 * ioctls
2025 *
2026 *************************************************************************/
2027
2028/* Net device ioctl
2029 * Context: process, rtnl_lock() held.
2030 */
2031static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
2032{
2033	struct efx_nic *efx = netdev_priv(net_dev);
2034	struct mii_ioctl_data *data = if_mii(ifr);
2035
2036	if (cmd == SIOCSHWTSTAMP)
2037		return efx_ptp_set_ts_config(efx, ifr);
2038	if (cmd == SIOCGHWTSTAMP)
2039		return efx_ptp_get_ts_config(efx, ifr);
2040
2041	/* Convert phy_id from older PRTAD/DEVAD format */
2042	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
2043	    (data->phy_id & 0xfc00) == 0x0400)
2044		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;
2045
2046	return mdio_mii_ioctl(&efx->mdio, data, cmd);
2047}
2048
2049/**************************************************************************
2050 *
2051 * NAPI interface
2052 *
2053 **************************************************************************/
2054
2055static void efx_init_napi_channel(struct efx_channel *channel)
2056{
2057	struct efx_nic *efx = channel->efx;
2058
2059	channel->napi_dev = efx->net_dev;
2060	netif_napi_add(channel->napi_dev, &channel->napi_str,
2061		       efx_poll, napi_weight);
2062	efx_channel_busy_poll_init(channel);
2063}
2064
2065static void efx_init_napi(struct efx_nic *efx)
2066{
2067	struct efx_channel *channel;
2068
2069	efx_for_each_channel(channel, efx)
2070		efx_init_napi_channel(channel);
2071}
2072
2073static void efx_fini_napi_channel(struct efx_channel *channel)
2074{
2075	if (channel->napi_dev) {
2076		netif_napi_del(&channel->napi_str);
2077		napi_hash_del(&channel->napi_str);
2078	}
2079	channel->napi_dev = NULL;
2080}
2081
2082static void efx_fini_napi(struct efx_nic *efx)
2083{
2084	struct efx_channel *channel;
2085
2086	efx_for_each_channel(channel, efx)
2087		efx_fini_napi_channel(channel);
2088}
2089
2090/**************************************************************************
2091 *
2092 * Kernel netpoll interface
2093 *
2094 *************************************************************************/
2095
2096#ifdef CONFIG_NET_POLL_CONTROLLER
2097
2098/* Although in the common case interrupts will be disabled, this is not
2099 * guaranteed. However, all our work happens inside the NAPI callback,
2100 * so no locking is required.
2101 */
2102static void efx_netpoll(struct net_device *net_dev)
2103{
2104	struct efx_nic *efx = netdev_priv(net_dev);
2105	struct efx_channel *channel;
2106
2107	efx_for_each_channel(channel, efx)
2108		efx_schedule_channel(channel);
2109}
2110
2111#endif
2112
2113#ifdef CONFIG_NET_RX_BUSY_POLL
2114static int efx_busy_poll(struct napi_struct *napi)
2115{
2116	struct efx_channel *channel =
2117		container_of(napi, struct efx_channel, napi_str);
2118	struct efx_nic *efx = channel->efx;
2119	int budget = 4;
2120	int old_rx_packets, rx_packets;
2121
2122	if (!netif_running(efx->net_dev))
2123		return LL_FLUSH_FAILED;
2124
2125	if (!efx_channel_try_lock_poll(channel))
2126		return LL_FLUSH_BUSY;
2127
2128	old_rx_packets = channel->rx_queue.rx_packets;
2129	efx_process_channel(channel, budget);
2130
2131	rx_packets = channel->rx_queue.rx_packets - old_rx_packets;
2132
2133	/* There is no race condition with NAPI here.
2134	 * NAPI will automatically be rescheduled if it yielded during busy
2135	 * polling, because it was not able to take the lock and thus returned
2136	 * the full budget.
2137	 */
2138	efx_channel_unlock_poll(channel);
2139
2140	return rx_packets;
2141}
2142#endif
2143
2144/**************************************************************************
2145 *
2146 * Kernel net device interface
2147 *
2148 *************************************************************************/
2149
2150/* Context: process, rtnl_lock() held. */
2151int efx_net_open(struct net_device *net_dev)
2152{
2153	struct efx_nic *efx = netdev_priv(net_dev);
2154	int rc;
2155
2156	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
2157		  raw_smp_processor_id());
2158
2159	rc = efx_check_disabled(efx);
2160	if (rc)
2161		return rc;
2162	if (efx->phy_mode & PHY_MODE_SPECIAL)
2163		return -EBUSY;
2164	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
2165		return -EIO;
2166
2167	/* Notify the kernel of the link state polled during driver load,
2168	 * before the monitor starts running */
2169	efx_link_status_changed(efx);
2170
2171	efx_start_all(efx);
2172	efx_selftest_async_start(efx);
 
 
 
 
2173	return 0;
2174}
2175
2176/* Context: process, rtnl_lock() held.
2177 * Note that the kernel will ignore our return code; this method
2178 * should really be a void.
2179 */
2180int efx_net_stop(struct net_device *net_dev)
2181{
2182	struct efx_nic *efx = netdev_priv(net_dev);
2183
2184	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
2185		  raw_smp_processor_id());
2186
2187	/* Stop the device and flush all the channels */
2188	efx_stop_all(efx);
2189
2190	return 0;
2191}
2192
2193/* Context: process, dev_base_lock or RTNL held, non-blocking. */
2194static struct rtnl_link_stats64 *efx_net_stats(struct net_device *net_dev,
2195					       struct rtnl_link_stats64 *stats)
2196{
2197	struct efx_nic *efx = netdev_priv(net_dev);
2198
2199	spin_lock_bh(&efx->stats_lock);
2200	efx->type->update_stats(efx, NULL, stats);
2201	spin_unlock_bh(&efx->stats_lock);
2202
2203	return stats;
2204}
2205
2206/* Context: netif_tx_lock held, BHs disabled. */
2207static void efx_watchdog(struct net_device *net_dev)
2208{
2209	struct efx_nic *efx = netdev_priv(net_dev);
2210
2211	netif_err(efx, tx_err, efx->net_dev,
2212		  "TX stuck with port_enabled=%d: resetting channels\n",
2213		  efx->port_enabled);
2214
2215	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2216}
2217
2218
2219/* Context: process, rtnl_lock() held. */
2220static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
2221{
2222	struct efx_nic *efx = netdev_priv(net_dev);
2223	int rc;
2224
2225	rc = efx_check_disabled(efx);
2226	if (rc)
2227		return rc;
2228	if (new_mtu > EFX_MAX_MTU)
2229		return -EINVAL;
2230
2231	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2232
2233	efx_device_detach_sync(efx);
2234	efx_stop_all(efx);
2235
2236	mutex_lock(&efx->mac_lock);
2237	net_dev->mtu = new_mtu;
2238	efx_mac_reconfigure(efx);
2239	mutex_unlock(&efx->mac_lock);
2240
2241	efx_start_all(efx);
2242	netif_device_attach(efx->net_dev);
2243	return 0;
2244}
2245
2246static int efx_set_mac_address(struct net_device *net_dev, void *data)
 
2247{
2248	struct efx_nic *efx = netdev_priv(net_dev);
2249	struct sockaddr *addr = data;
2250	u8 *new_addr = addr->sa_data;
2251	u8 old_addr[6];
2252	int rc;
2253
2254	if (!is_valid_ether_addr(new_addr)) {
2255		netif_err(efx, drv, efx->net_dev,
2256			  "invalid ethernet MAC address requested: %pM\n",
2257			  new_addr);
2258		return -EADDRNOTAVAIL;
2259	}
2260
2261	/* save old address */
2262	ether_addr_copy(old_addr, net_dev->dev_addr);
2263	ether_addr_copy(net_dev->dev_addr, new_addr);
2264	if (efx->type->set_mac_address) {
2265		rc = efx->type->set_mac_address(efx);
2266		if (rc) {
2267			ether_addr_copy(net_dev->dev_addr, old_addr);
2268			return rc;
2269		}
2270	}
2271
2272	/* Reconfigure the MAC */
2273	mutex_lock(&efx->mac_lock);
2274	efx_mac_reconfigure(efx);
2275	mutex_unlock(&efx->mac_lock);
2276
2277	return 0;
2278}
2279
2280/* Context: netif_addr_lock held, BHs disabled. */
2281static void efx_set_rx_mode(struct net_device *net_dev)
2282{
2283	struct efx_nic *efx = netdev_priv(net_dev);
2284
2285	if (efx->port_enabled)
2286		queue_work(efx->workqueue, &efx->mac_work);
2287	/* Otherwise efx_start_port() will do this */
2288}
2289
2290static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2291{
2292	struct efx_nic *efx = netdev_priv(net_dev);
2293
2294	/* If disabling RX n-tuple filtering, clear existing filters */
2295	if (net_dev->features & ~data & NETIF_F_NTUPLE)
2296		return efx->type->filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);
2297
2298	return 0;
2299}
2300
2301static const struct net_device_ops efx_netdev_ops = {
2302	.ndo_open		= efx_net_open,
2303	.ndo_stop		= efx_net_stop,
2304	.ndo_get_stats64	= efx_net_stats,
2305	.ndo_tx_timeout		= efx_watchdog,
2306	.ndo_start_xmit		= efx_hard_start_xmit,
2307	.ndo_validate_addr	= eth_validate_addr,
2308	.ndo_do_ioctl		= efx_ioctl,
2309	.ndo_change_mtu		= efx_change_mtu,
2310	.ndo_set_mac_address	= efx_set_mac_address,
2311	.ndo_set_rx_mode	= efx_set_rx_mode,
2312	.ndo_set_features	= efx_set_features,
 
 
 
 
 
2313#ifdef CONFIG_SFC_SRIOV
2314	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
2315	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
2316	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
2317	.ndo_get_vf_config	= efx_sriov_get_vf_config,
2318	.ndo_set_vf_link_state  = efx_sriov_set_vf_link_state,
2319	.ndo_get_phys_port_id   = efx_sriov_get_phys_port_id,
2320#endif
2321#ifdef CONFIG_NET_POLL_CONTROLLER
2322	.ndo_poll_controller = efx_netpoll,
2323#endif
2324	.ndo_setup_tc		= efx_setup_tc,
2325#ifdef CONFIG_NET_RX_BUSY_POLL
2326	.ndo_busy_poll		= efx_busy_poll,
2327#endif
 
 
2328#ifdef CONFIG_RFS_ACCEL
2329	.ndo_rx_flow_steer	= efx_filter_rfs,
2330#endif
 
 
2331};
2332
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2333static void efx_update_name(struct efx_nic *efx)
2334{
2335	strcpy(efx->name, efx->net_dev->name);
2336	efx_mtd_rename(efx);
2337	efx_set_channel_names(efx);
2338}
2339
2340static int efx_netdev_event(struct notifier_block *this,
2341			    unsigned long event, void *ptr)
2342{
2343	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2344
2345	if ((net_dev->netdev_ops == &efx_netdev_ops) &&
2346	    event == NETDEV_CHANGENAME)
2347		efx_update_name(netdev_priv(net_dev));
2348
2349	return NOTIFY_DONE;
2350}
2351
2352static struct notifier_block efx_netdev_notifier = {
2353	.notifier_call = efx_netdev_event,
2354};
2355
2356static ssize_t
2357show_phy_type(struct device *dev, struct device_attribute *attr, char *buf)
2358{
2359	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2360	return sprintf(buf, "%d\n", efx->phy_type);
2361}
2362static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
2363
2364#ifdef CONFIG_SFC_MCDI_LOGGING
2365static ssize_t show_mcdi_log(struct device *dev, struct device_attribute *attr,
2366			     char *buf)
2367{
2368	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2369	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2370
2371	return scnprintf(buf, PAGE_SIZE, "%d\n", mcdi->logging_enabled);
2372}
2373static ssize_t set_mcdi_log(struct device *dev, struct device_attribute *attr,
2374			    const char *buf, size_t count)
2375{
2376	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2377	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2378	bool enable = count > 0 && *buf != '0';
2379
2380	mcdi->logging_enabled = enable;
2381	return count;
2382}
2383static DEVICE_ATTR(mcdi_logging, 0644, show_mcdi_log, set_mcdi_log);
2384#endif
2385
2386static int efx_register_netdev(struct efx_nic *efx)
2387{
2388	struct net_device *net_dev = efx->net_dev;
2389	struct efx_channel *channel;
2390	int rc;
2391
2392	net_dev->watchdog_timeo = 5 * HZ;
2393	net_dev->irq = efx->pci_dev->irq;
2394	net_dev->netdev_ops = &efx_netdev_ops;
2395	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0)
2396		net_dev->priv_flags |= IFF_UNICAST_FLT;
2397	net_dev->ethtool_ops = &efx_ethtool_ops;
2398	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
 
 
2399
2400	rtnl_lock();
2401
2402	/* Enable resets to be scheduled and check whether any were
2403	 * already requested.  If so, the NIC is probably hosed so we
2404	 * abort.
2405	 */
2406	efx->state = STATE_READY;
2407	smp_mb(); /* ensure we change state before checking reset_pending */
2408	if (efx->reset_pending) {
2409		netif_err(efx, probe, efx->net_dev,
2410			  "aborting probe due to scheduled reset\n");
2411		rc = -EIO;
2412		goto fail_locked;
2413	}
2414
2415	rc = dev_alloc_name(net_dev, net_dev->name);
2416	if (rc < 0)
2417		goto fail_locked;
2418	efx_update_name(efx);
2419
2420	/* Always start with carrier off; PHY events will detect the link */
2421	netif_carrier_off(net_dev);
2422
2423	rc = register_netdevice(net_dev);
2424	if (rc)
2425		goto fail_locked;
2426
2427	efx_for_each_channel(channel, efx) {
2428		struct efx_tx_queue *tx_queue;
2429		efx_for_each_channel_tx_queue(tx_queue, channel)
2430			efx_init_tx_queue_core_txq(tx_queue);
2431	}
2432
2433	efx_associate(efx);
2434
 
 
2435	rtnl_unlock();
2436
2437	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2438	if (rc) {
2439		netif_err(efx, drv, efx->net_dev,
2440			  "failed to init net dev attributes\n");
2441		goto fail_registered;
2442	}
2443#ifdef CONFIG_SFC_MCDI_LOGGING
2444	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2445	if (rc) {
2446		netif_err(efx, drv, efx->net_dev,
2447			  "failed to init net dev attributes\n");
2448		goto fail_attr_mcdi_logging;
2449	}
2450#endif
2451
2452	return 0;
2453
2454#ifdef CONFIG_SFC_MCDI_LOGGING
2455fail_attr_mcdi_logging:
2456	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2457#endif
2458fail_registered:
2459	rtnl_lock();
2460	efx_dissociate(efx);
2461	unregister_netdevice(net_dev);
2462fail_locked:
2463	efx->state = STATE_UNINIT;
2464	rtnl_unlock();
2465	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2466	return rc;
2467}
2468
2469static void efx_unregister_netdev(struct efx_nic *efx)
2470{
2471	if (!efx->net_dev)
2472		return;
2473
2474	BUG_ON(netdev_priv(efx->net_dev) != efx);
 
2475
2476	if (efx_dev_registered(efx)) {
2477		strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
2478#ifdef CONFIG_SFC_MCDI_LOGGING
2479		device_remove_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2480#endif
2481		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2482		unregister_netdev(efx->net_dev);
2483	}
2484}
2485
2486/**************************************************************************
2487 *
2488 * Device reset and suspend
2489 *
2490 **************************************************************************/
2491
2492/* Tears down the entire software state and most of the hardware state
2493 * before reset.  */
2494void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2495{
2496	EFX_ASSERT_RESET_SERIALISED(efx);
2497
2498	if (method == RESET_TYPE_MCDI_TIMEOUT)
2499		efx->type->prepare_flr(efx);
2500
2501	efx_stop_all(efx);
2502	efx_disable_interrupts(efx);
2503
2504	mutex_lock(&efx->mac_lock);
2505	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2506	    method != RESET_TYPE_DATAPATH)
2507		efx->phy_op->fini(efx);
2508	efx->type->fini(efx);
2509}
2510
2511/* This function will always ensure that the locks acquired in
2512 * efx_reset_down() are released. A failure return code indicates
2513 * that we were unable to reinitialise the hardware, and the
2514 * driver should be disabled. If ok is false, then the rx and tx
2515 * engines are not restarted, pending a RESET_DISABLE. */
2516int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2517{
2518	int rc;
2519
2520	EFX_ASSERT_RESET_SERIALISED(efx);
2521
2522	if (method == RESET_TYPE_MCDI_TIMEOUT)
2523		efx->type->finish_flr(efx);
2524
2525	/* Ensure that SRAM is initialised even if we're disabling the device */
2526	rc = efx->type->init(efx);
2527	if (rc) {
2528		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2529		goto fail;
2530	}
2531
2532	if (!ok)
2533		goto fail;
2534
2535	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2536	    method != RESET_TYPE_DATAPATH) {
2537		rc = efx->phy_op->init(efx);
2538		if (rc)
2539			goto fail;
2540		rc = efx->phy_op->reconfigure(efx);
2541		if (rc && rc != -EPERM)
2542			netif_err(efx, drv, efx->net_dev,
2543				  "could not restore PHY settings\n");
2544	}
2545
2546	rc = efx_enable_interrupts(efx);
2547	if (rc)
2548		goto fail;
2549
2550#ifdef CONFIG_SFC_SRIOV
2551	rc = efx->type->vswitching_restore(efx);
2552	if (rc) /* not fatal; the PF will still work fine */
2553		netif_warn(efx, probe, efx->net_dev,
2554			   "failed to restore vswitching rc=%d;"
2555			   " VFs may not function\n", rc);
2556#endif
2557
2558	down_read(&efx->filter_sem);
2559	efx_restore_filters(efx);
2560	up_read(&efx->filter_sem);
2561	if (efx->type->sriov_reset)
2562		efx->type->sriov_reset(efx);
2563
2564	mutex_unlock(&efx->mac_lock);
2565
2566	efx_start_all(efx);
2567
2568	return 0;
2569
2570fail:
2571	efx->port_initialized = false;
2572
2573	mutex_unlock(&efx->mac_lock);
2574
2575	return rc;
2576}
2577
2578/* Reset the NIC using the specified method.  Note that the reset may
2579 * fail, in which case the card will be left in an unusable state.
2580 *
2581 * Caller must hold the rtnl_lock.
2582 */
2583int efx_reset(struct efx_nic *efx, enum reset_type method)
2584{
2585	int rc, rc2;
2586	bool disabled;
2587
2588	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
2589		   RESET_TYPE(method));
2590
2591	efx_device_detach_sync(efx);
2592	efx_reset_down(efx, method);
2593
2594	rc = efx->type->reset(efx, method);
2595	if (rc) {
2596		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2597		goto out;
2598	}
2599
2600	/* Clear flags for the scopes we covered.  We assume the NIC and
2601	 * driver are now quiescent so that there is no race here.
2602	 */
2603	if (method < RESET_TYPE_MAX_METHOD)
2604		efx->reset_pending &= -(1 << (method + 1));
2605	else /* it doesn't fit into the well-ordered scope hierarchy */
2606		__clear_bit(method, &efx->reset_pending);
2607
2608	/* Reinitialise bus-mastering, which may have been turned off before
2609	 * the reset was scheduled. This is still appropriate, even in the
2610	 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
2611	 * can respond to requests. */
2612	pci_set_master(efx->pci_dev);
2613
2614out:
2615	/* Leave device stopped if necessary */
2616	disabled = rc ||
2617		method == RESET_TYPE_DISABLE ||
2618		method == RESET_TYPE_RECOVER_OR_DISABLE;
2619	rc2 = efx_reset_up(efx, method, !disabled);
2620	if (rc2) {
2621		disabled = true;
2622		if (!rc)
2623			rc = rc2;
2624	}
2625
2626	if (disabled) {
2627		dev_close(efx->net_dev);
2628		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2629		efx->state = STATE_DISABLED;
2630	} else {
2631		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2632		netif_device_attach(efx->net_dev);
2633	}
2634	return rc;
2635}
2636
2637/* Try recovery mechanisms.
2638 * For now only EEH is supported.
2639 * Returns 0 if the recovery mechanisms are unsuccessful.
2640 * Returns a non-zero value otherwise.
2641 */
2642int efx_try_recovery(struct efx_nic *efx)
2643{
2644#ifdef CONFIG_EEH
2645	/* A PCI error can occur and not be seen by EEH because nothing
2646	 * happens on the PCI bus. In this case the driver may fail and
2647	 * schedule a 'recover or reset', leading to this recovery handler.
2648	 * Manually call the eeh failure check function.
2649	 */
2650	struct eeh_dev *eehdev = pci_dev_to_eeh_dev(efx->pci_dev);
2651	if (eeh_dev_check_failure(eehdev)) {
2652		/* The EEH mechanisms will handle the error and reset the
2653		 * device if necessary.
2654		 */
2655		return 1;
2656	}
2657#endif
2658	return 0;
2659}
2660
2661static void efx_wait_for_bist_end(struct efx_nic *efx)
2662{
2663	int i;
2664
2665	for (i = 0; i < BIST_WAIT_DELAY_COUNT; ++i) {
2666		if (efx_mcdi_poll_reboot(efx))
2667			goto out;
2668		msleep(BIST_WAIT_DELAY_MS);
2669	}
2670
2671	netif_err(efx, drv, efx->net_dev, "Warning: No MC reboot after BIST mode\n");
2672out:
2673	/* Either way unset the BIST flag. If we found no reboot we probably
2674	 * won't recover, but we should try.
2675	 */
2676	efx->mc_bist_for_other_fn = false;
2677}
2678
2679/* The worker thread exists so that code that cannot sleep can
2680 * schedule a reset for later.
2681 */
2682static void efx_reset_work(struct work_struct *data)
2683{
2684	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2685	unsigned long pending;
2686	enum reset_type method;
2687
2688	pending = ACCESS_ONCE(efx->reset_pending);
2689	method = fls(pending) - 1;
2690
2691	if (method == RESET_TYPE_MC_BIST)
2692		efx_wait_for_bist_end(efx);
2693
2694	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
2695	     method == RESET_TYPE_RECOVER_OR_ALL) &&
2696	    efx_try_recovery(efx))
2697		return;
2698
2699	if (!pending)
2700		return;
2701
2702	rtnl_lock();
2703
2704	/* We checked the state in efx_schedule_reset() but it may
2705	 * have changed by now.  Now that we have the RTNL lock,
2706	 * it cannot change again.
2707	 */
2708	if (efx->state == STATE_READY)
2709		(void)efx_reset(efx, method);
2710
2711	rtnl_unlock();
2712}
2713
2714void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
2715{
2716	enum reset_type method;
2717
2718	if (efx->state == STATE_RECOVERY) {
2719		netif_dbg(efx, drv, efx->net_dev,
2720			  "recovering: skip scheduling %s reset\n",
2721			  RESET_TYPE(type));
2722		return;
2723	}
2724
2725	switch (type) {
2726	case RESET_TYPE_INVISIBLE:
2727	case RESET_TYPE_ALL:
2728	case RESET_TYPE_RECOVER_OR_ALL:
2729	case RESET_TYPE_WORLD:
2730	case RESET_TYPE_DISABLE:
2731	case RESET_TYPE_RECOVER_OR_DISABLE:
2732	case RESET_TYPE_DATAPATH:
2733	case RESET_TYPE_MC_BIST:
2734	case RESET_TYPE_MCDI_TIMEOUT:
2735		method = type;
2736		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
2737			  RESET_TYPE(method));
2738		break;
2739	default:
2740		method = efx->type->map_reset_reason(type);
2741		netif_dbg(efx, drv, efx->net_dev,
2742			  "scheduling %s reset for %s\n",
2743			  RESET_TYPE(method), RESET_TYPE(type));
2744		break;
2745	}
2746
2747	set_bit(method, &efx->reset_pending);
2748	smp_mb(); /* ensure we change reset_pending before checking state */
2749
2750	/* If we're not READY then just leave the flags set as the cue
2751	 * to abort probing or reschedule the reset later.
2752	 */
2753	if (ACCESS_ONCE(efx->state) != STATE_READY)
2754		return;
2755
2756	/* efx_process_channel() will no longer read events once a
2757	 * reset is scheduled. So switch back to poll'd MCDI completions. */
2758	efx_mcdi_mode_poll(efx);
2759
2760	queue_work(reset_workqueue, &efx->reset_work);
2761}
2762
2763/**************************************************************************
2764 *
2765 * List of NICs we support
2766 *
2767 **************************************************************************/
2768
2769/* PCI device ID table */
2770static const struct pci_device_id efx_pci_table[] = {
2771	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
2772		    PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0),
2773	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2774	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
2775		    PCI_DEVICE_ID_SOLARFLARE_SFC4000B),
2776	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2777	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2778	 .driver_data = (unsigned long) &siena_a0_nic_type},
2779	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2780	 .driver_data = (unsigned long) &siena_a0_nic_type},
2781	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
2782	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2783	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903),  /* SFC9120 VF */
2784	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2785	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923),  /* SFC9140 PF */
2786	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2787	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923),  /* SFC9140 VF */
2788	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2789	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03),  /* SFC9220 PF */
2790	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2791	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03),  /* SFC9220 VF */
2792	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
 
 
 
 
2793	{0}			/* end of list */
2794};
2795
2796/**************************************************************************
2797 *
2798 * Dummy PHY/MAC operations
2799 *
2800 * Can be used for some unimplemented operations
2801 * Needed so all function pointers are valid and do not have to be tested
2802 * before use
2803 *
2804 **************************************************************************/
2805int efx_port_dummy_op_int(struct efx_nic *efx)
2806{
2807	return 0;
2808}
2809void efx_port_dummy_op_void(struct efx_nic *efx) {}
2810
2811static bool efx_port_dummy_op_poll(struct efx_nic *efx)
2812{
2813	return false;
2814}
2815
2816static const struct efx_phy_operations efx_dummy_phy_operations = {
2817	.init		 = efx_port_dummy_op_int,
2818	.reconfigure	 = efx_port_dummy_op_int,
2819	.poll		 = efx_port_dummy_op_poll,
2820	.fini		 = efx_port_dummy_op_void,
2821};
2822
2823/**************************************************************************
2824 *
2825 * Data housekeeping
2826 *
2827 **************************************************************************/
2828
2829/* This zeroes out and then fills in the invariants in a struct
2830 * efx_nic (including all sub-structures).
2831 */
2832static int efx_init_struct(struct efx_nic *efx,
2833			   struct pci_dev *pci_dev, struct net_device *net_dev)
2834{
2835	int i;
2836
2837	/* Initialise common structures */
2838	INIT_LIST_HEAD(&efx->node);
2839	INIT_LIST_HEAD(&efx->secondary_list);
2840	spin_lock_init(&efx->biu_lock);
2841#ifdef CONFIG_SFC_MTD
2842	INIT_LIST_HEAD(&efx->mtd_list);
2843#endif
2844	INIT_WORK(&efx->reset_work, efx_reset_work);
2845	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
2846	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
2847	efx->pci_dev = pci_dev;
2848	efx->msg_enable = debug;
2849	efx->state = STATE_UNINIT;
2850	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
2851
2852	efx->net_dev = net_dev;
2853	efx->rx_prefix_size = efx->type->rx_prefix_size;
2854	efx->rx_ip_align =
2855		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
2856	efx->rx_packet_hash_offset =
2857		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
2858	efx->rx_packet_ts_offset =
2859		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
2860	spin_lock_init(&efx->stats_lock);
2861	mutex_init(&efx->mac_lock);
2862	efx->phy_op = &efx_dummy_phy_operations;
2863	efx->mdio.dev = net_dev;
2864	INIT_WORK(&efx->mac_work, efx_mac_work);
2865	init_waitqueue_head(&efx->flush_wq);
2866
2867	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2868		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
2869		if (!efx->channel[i])
2870			goto fail;
2871		efx->msi_context[i].efx = efx;
2872		efx->msi_context[i].index = i;
2873	}
2874
2875	/* Higher numbered interrupt modes are less capable! */
2876	efx->interrupt_mode = max(efx->type->max_interrupt_mode,
2877				  interrupt_mode);
2878
2879	/* Would be good to use the net_dev name, but we're too early */
2880	snprintf(efx->workqueue_name, sizeof(efx->workqueue_name), "sfc%s",
2881		 pci_name(pci_dev));
2882	efx->workqueue = create_singlethread_workqueue(efx->workqueue_name);
2883	if (!efx->workqueue)
2884		goto fail;
2885
2886	return 0;
2887
2888fail:
2889	efx_fini_struct(efx);
2890	return -ENOMEM;
2891}
2892
2893static void efx_fini_struct(struct efx_nic *efx)
2894{
2895	int i;
2896
2897	for (i = 0; i < EFX_MAX_CHANNELS; i++)
2898		kfree(efx->channel[i]);
2899
2900	kfree(efx->vpd_sn);
2901
2902	if (efx->workqueue) {
2903		destroy_workqueue(efx->workqueue);
2904		efx->workqueue = NULL;
2905	}
2906}
2907
2908void efx_update_sw_stats(struct efx_nic *efx, u64 *stats)
2909{
2910	u64 n_rx_nodesc_trunc = 0;
2911	struct efx_channel *channel;
2912
2913	efx_for_each_channel(channel, efx)
2914		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
2915	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
2916	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
2917}
2918
2919/**************************************************************************
2920 *
2921 * PCI interface
2922 *
2923 **************************************************************************/
2924
2925/* Main body of final NIC shutdown code
2926 * This is called only at module unload (or hotplug removal).
2927 */
2928static void efx_pci_remove_main(struct efx_nic *efx)
2929{
2930	/* Flush reset_work. It can no longer be scheduled since we
2931	 * are not READY.
2932	 */
2933	BUG_ON(efx->state == STATE_READY);
2934	cancel_work_sync(&efx->reset_work);
2935
2936	efx_disable_interrupts(efx);
 
2937	efx_nic_fini_interrupt(efx);
2938	efx_fini_port(efx);
2939	efx->type->fini(efx);
2940	efx_fini_napi(efx);
2941	efx_remove_all(efx);
2942}
2943
2944/* Final NIC shutdown
2945 * This is called only at module unload (or hotplug removal).  A PF can call
2946 * this on its VFs to ensure they are unbound first.
2947 */
2948static void efx_pci_remove(struct pci_dev *pci_dev)
2949{
 
2950	struct efx_nic *efx;
2951
2952	efx = pci_get_drvdata(pci_dev);
2953	if (!efx)
2954		return;
2955
2956	/* Mark the NIC as fini, then stop the interface */
2957	rtnl_lock();
2958	efx_dissociate(efx);
2959	dev_close(efx->net_dev);
2960	efx_disable_interrupts(efx);
2961	efx->state = STATE_UNINIT;
2962	rtnl_unlock();
2963
2964	if (efx->type->sriov_fini)
2965		efx->type->sriov_fini(efx);
2966
 
2967	efx_unregister_netdev(efx);
2968
2969	efx_mtd_remove(efx);
2970
2971	efx_pci_remove_main(efx);
2972
2973	efx_fini_io(efx);
2974	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
2975
 
2976	efx_fini_struct(efx);
2977	free_netdev(efx->net_dev);
2978
2979	pci_disable_pcie_error_reporting(pci_dev);
2980};
2981
2982/* NIC VPD information
2983 * Called during probe to display the part number of the
2984 * installed NIC.  VPD is potentially very large but this should
2985 * always appear within the first 512 bytes.
2986 */
2987#define SFC_VPD_LEN 512
2988static void efx_probe_vpd_strings(struct efx_nic *efx)
2989{
2990	struct pci_dev *dev = efx->pci_dev;
2991	char vpd_data[SFC_VPD_LEN];
2992	ssize_t vpd_size;
2993	int ro_start, ro_size, i, j;
2994
2995	/* Get the vpd data from the device */
2996	vpd_size = pci_read_vpd(dev, 0, sizeof(vpd_data), vpd_data);
2997	if (vpd_size <= 0) {
2998		netif_err(efx, drv, efx->net_dev, "Unable to read VPD\n");
2999		return;
3000	}
3001
3002	/* Get the Read only section */
3003	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size, PCI_VPD_LRDT_RO_DATA);
3004	if (ro_start < 0) {
3005		netif_err(efx, drv, efx->net_dev, "VPD Read-only not found\n");
3006		return;
3007	}
3008
3009	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
3010	j = ro_size;
3011	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3012	if (i + j > vpd_size)
3013		j = vpd_size - i;
3014
3015	/* Get the Part number */
3016	i = pci_vpd_find_info_keyword(vpd_data, i, j, "PN");
3017	if (i < 0) {
3018		netif_err(efx, drv, efx->net_dev, "Part number not found\n");
3019		return;
3020	}
3021
3022	j = pci_vpd_info_field_size(&vpd_data[i]);
3023	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3024	if (i + j > vpd_size) {
3025		netif_err(efx, drv, efx->net_dev, "Incomplete part number\n");
3026		return;
3027	}
3028
3029	netif_info(efx, drv, efx->net_dev,
3030		   "Part Number : %.*s\n", j, &vpd_data[i]);
3031
3032	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3033	j = ro_size;
3034	i = pci_vpd_find_info_keyword(vpd_data, i, j, "SN");
3035	if (i < 0) {
3036		netif_err(efx, drv, efx->net_dev, "Serial number not found\n");
3037		return;
3038	}
3039
3040	j = pci_vpd_info_field_size(&vpd_data[i]);
3041	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3042	if (i + j > vpd_size) {
3043		netif_err(efx, drv, efx->net_dev, "Incomplete serial number\n");
3044		return;
3045	}
3046
3047	efx->vpd_sn = kmalloc(j + 1, GFP_KERNEL);
3048	if (!efx->vpd_sn)
3049		return;
3050
3051	snprintf(efx->vpd_sn, j + 1, "%s", &vpd_data[i]);
3052}
3053
3054
3055/* Main body of NIC initialisation
3056 * This is called at module load (or hotplug insertion, theoretically).
3057 */
3058static int efx_pci_probe_main(struct efx_nic *efx)
3059{
3060	int rc;
3061
3062	/* Do start-of-day initialisation */
3063	rc = efx_probe_all(efx);
3064	if (rc)
3065		goto fail1;
3066
3067	efx_init_napi(efx);
3068
 
3069	rc = efx->type->init(efx);
 
3070	if (rc) {
3071		netif_err(efx, probe, efx->net_dev,
3072			  "failed to initialise NIC\n");
3073		goto fail3;
3074	}
3075
3076	rc = efx_init_port(efx);
3077	if (rc) {
3078		netif_err(efx, probe, efx->net_dev,
3079			  "failed to initialise port\n");
3080		goto fail4;
3081	}
3082
3083	rc = efx_nic_init_interrupt(efx);
3084	if (rc)
3085		goto fail5;
 
 
3086	rc = efx_enable_interrupts(efx);
3087	if (rc)
3088		goto fail6;
3089
3090	return 0;
3091
3092 fail6:
 
3093	efx_nic_fini_interrupt(efx);
3094 fail5:
3095	efx_fini_port(efx);
3096 fail4:
3097	efx->type->fini(efx);
3098 fail3:
3099	efx_fini_napi(efx);
3100	efx_remove_all(efx);
3101 fail1:
3102	return rc;
3103}
3104
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3105/* NIC initialisation
3106 *
3107 * This is called at module load (or hotplug insertion,
3108 * theoretically).  It sets up PCI mappings, resets the NIC,
3109 * sets up and registers the network devices with the kernel and hooks
3110 * the interrupt service routine.  It does not prepare the device for
3111 * transmission; this is left to the first time one of the network
3112 * interfaces is brought up (i.e. efx_net_open).
3113 */
3114static int efx_pci_probe(struct pci_dev *pci_dev,
3115			 const struct pci_device_id *entry)
3116{
 
3117	struct net_device *net_dev;
3118	struct efx_nic *efx;
3119	int rc;
3120
3121	/* Allocate and initialise a struct net_device and struct efx_nic */
3122	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
3123				     EFX_MAX_RX_QUEUES);
3124	if (!net_dev)
3125		return -ENOMEM;
3126	efx = netdev_priv(net_dev);
 
 
 
 
 
 
 
 
 
 
 
3127	efx->type = (const struct efx_nic_type *) entry->driver_data;
3128	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
3129			      NETIF_F_HIGHDMA | NETIF_F_TSO |
3130			      NETIF_F_RXCSUM);
3131	if (efx->type->offload_features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
3132		net_dev->features |= NETIF_F_TSO6;
3133	/* Mask for features that also apply to VLAN devices */
3134	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
3135				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
3136				   NETIF_F_RXCSUM);
3137	/* All offloads can be toggled */
3138	net_dev->hw_features = net_dev->features & ~NETIF_F_HIGHDMA;
3139	pci_set_drvdata(pci_dev, efx);
3140	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
3141	rc = efx_init_struct(efx, pci_dev, net_dev);
3142	if (rc)
3143		goto fail1;
 
3144
3145	netif_info(efx, probe, efx->net_dev,
3146		   "Solarflare NIC detected\n");
3147
3148	if (!efx->type->is_vf)
3149		efx_probe_vpd_strings(efx);
3150
3151	/* Set up basic I/O (BAR mappings etc) */
3152	rc = efx_init_io(efx);
 
3153	if (rc)
3154		goto fail2;
3155
3156	rc = efx_pci_probe_main(efx);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3157	if (rc)
3158		goto fail3;
3159
3160	rc = efx_register_netdev(efx);
3161	if (rc)
3162		goto fail4;
3163
3164	if (efx->type->sriov_init) {
3165		rc = efx->type->sriov_init(efx);
3166		if (rc)
3167			netif_err(efx, probe, efx->net_dev,
3168				  "SR-IOV can't be enabled rc %d\n", rc);
3169	}
3170
3171	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
3172
3173	/* Try to create MTDs, but allow this to fail */
3174	rtnl_lock();
3175	rc = efx_mtd_probe(efx);
3176	rtnl_unlock();
3177	if (rc && rc != -EPERM)
3178		netif_warn(efx, probe, efx->net_dev,
3179			   "failed to create MTDs (%d)\n", rc);
3180
3181	rc = pci_enable_pcie_error_reporting(pci_dev);
3182	if (rc && rc != -EINVAL)
3183		netif_notice(efx, probe, efx->net_dev,
3184			     "PCIE error reporting unavailable (%d).\n",
3185			     rc);
3186
3187	return 0;
3188
3189 fail4:
3190	efx_pci_remove_main(efx);
3191 fail3:
3192	efx_fini_io(efx);
3193 fail2:
3194	efx_fini_struct(efx);
3195 fail1:
3196	WARN_ON(rc > 0);
3197	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
3198	free_netdev(net_dev);
 
 
3199	return rc;
3200}
3201
3202/* efx_pci_sriov_configure returns the actual number of Virtual Functions
3203 * enabled on success
3204 */
3205#ifdef CONFIG_SFC_SRIOV
3206static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
3207{
3208	int rc;
3209	struct efx_nic *efx = pci_get_drvdata(dev);
3210
3211	if (efx->type->sriov_configure) {
3212		rc = efx->type->sriov_configure(efx, num_vfs);
3213		if (rc)
3214			return rc;
3215		else
3216			return num_vfs;
3217	} else
3218		return -EOPNOTSUPP;
3219}
3220#endif
3221
3222static int efx_pm_freeze(struct device *dev)
3223{
3224	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3225
3226	rtnl_lock();
3227
3228	if (efx->state != STATE_DISABLED) {
3229		efx->state = STATE_UNINIT;
3230
3231		efx_device_detach_sync(efx);
3232
3233		efx_stop_all(efx);
3234		efx_disable_interrupts(efx);
 
 
3235	}
3236
3237	rtnl_unlock();
3238
3239	return 0;
3240}
3241
 
 
 
 
 
 
 
 
 
 
 
3242static int efx_pm_thaw(struct device *dev)
3243{
3244	int rc;
3245	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3246
3247	rtnl_lock();
3248
3249	if (efx->state != STATE_DISABLED) {
3250		rc = efx_enable_interrupts(efx);
3251		if (rc)
3252			goto fail;
3253
3254		mutex_lock(&efx->mac_lock);
3255		efx->phy_op->reconfigure(efx);
3256		mutex_unlock(&efx->mac_lock);
3257
3258		efx_start_all(efx);
3259
3260		netif_device_attach(efx->net_dev);
3261
3262		efx->state = STATE_READY;
3263
3264		efx->type->resume_wol(efx);
3265	}
3266
3267	rtnl_unlock();
3268
3269	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
3270	queue_work(reset_workqueue, &efx->reset_work);
3271
3272	return 0;
3273
3274fail:
3275	rtnl_unlock();
3276
3277	return rc;
3278}
3279
3280static int efx_pm_poweroff(struct device *dev)
3281{
3282	struct pci_dev *pci_dev = to_pci_dev(dev);
3283	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3284
3285	efx->type->fini(efx);
3286
3287	efx->reset_pending = 0;
3288
3289	pci_save_state(pci_dev);
3290	return pci_set_power_state(pci_dev, PCI_D3hot);
3291}
3292
3293/* Used for both resume and restore */
3294static int efx_pm_resume(struct device *dev)
3295{
3296	struct pci_dev *pci_dev = to_pci_dev(dev);
3297	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3298	int rc;
3299
3300	rc = pci_set_power_state(pci_dev, PCI_D0);
3301	if (rc)
3302		return rc;
3303	pci_restore_state(pci_dev);
3304	rc = pci_enable_device(pci_dev);
3305	if (rc)
3306		return rc;
3307	pci_set_master(efx->pci_dev);
3308	rc = efx->type->reset(efx, RESET_TYPE_ALL);
3309	if (rc)
3310		return rc;
 
3311	rc = efx->type->init(efx);
 
3312	if (rc)
3313		return rc;
3314	rc = efx_pm_thaw(dev);
3315	return rc;
3316}
3317
3318static int efx_pm_suspend(struct device *dev)
3319{
3320	int rc;
3321
3322	efx_pm_freeze(dev);
3323	rc = efx_pm_poweroff(dev);
3324	if (rc)
3325		efx_pm_resume(dev);
3326	return rc;
3327}
3328
3329static const struct dev_pm_ops efx_pm_ops = {
3330	.suspend	= efx_pm_suspend,
3331	.resume		= efx_pm_resume,
3332	.freeze		= efx_pm_freeze,
3333	.thaw		= efx_pm_thaw,
3334	.poweroff	= efx_pm_poweroff,
3335	.restore	= efx_pm_resume,
3336};
3337
3338/* A PCI error affecting this device was detected.
3339 * At this point MMIO and DMA may be disabled.
3340 * Stop the software path and request a slot reset.
3341 */
3342static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
3343					      enum pci_channel_state state)
3344{
3345	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3346	struct efx_nic *efx = pci_get_drvdata(pdev);
3347
3348	if (state == pci_channel_io_perm_failure)
3349		return PCI_ERS_RESULT_DISCONNECT;
3350
3351	rtnl_lock();
3352
3353	if (efx->state != STATE_DISABLED) {
3354		efx->state = STATE_RECOVERY;
3355		efx->reset_pending = 0;
3356
3357		efx_device_detach_sync(efx);
3358
3359		efx_stop_all(efx);
3360		efx_disable_interrupts(efx);
3361
3362		status = PCI_ERS_RESULT_NEED_RESET;
3363	} else {
3364		/* If the interface is disabled we don't want to do anything
3365		 * with it.
3366		 */
3367		status = PCI_ERS_RESULT_RECOVERED;
3368	}
3369
3370	rtnl_unlock();
3371
3372	pci_disable_device(pdev);
3373
3374	return status;
3375}
3376
3377/* Fake a successful reset, which will be performed later in efx_io_resume. */
3378static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
3379{
3380	struct efx_nic *efx = pci_get_drvdata(pdev);
3381	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3382	int rc;
3383
3384	if (pci_enable_device(pdev)) {
3385		netif_err(efx, hw, efx->net_dev,
3386			  "Cannot re-enable PCI device after reset.\n");
3387		status =  PCI_ERS_RESULT_DISCONNECT;
3388	}
3389
3390	rc = pci_cleanup_aer_uncorrect_error_status(pdev);
3391	if (rc) {
3392		netif_err(efx, hw, efx->net_dev,
3393		"pci_cleanup_aer_uncorrect_error_status failed (%d)\n", rc);
3394		/* Non-fatal error. Continue. */
3395	}
3396
3397	return status;
3398}
3399
3400/* Perform the actual reset and resume I/O operations. */
3401static void efx_io_resume(struct pci_dev *pdev)
3402{
3403	struct efx_nic *efx = pci_get_drvdata(pdev);
3404	int rc;
3405
3406	rtnl_lock();
3407
3408	if (efx->state == STATE_DISABLED)
3409		goto out;
3410
3411	rc = efx_reset(efx, RESET_TYPE_ALL);
3412	if (rc) {
3413		netif_err(efx, hw, efx->net_dev,
3414			  "efx_reset failed after PCI error (%d)\n", rc);
3415	} else {
3416		efx->state = STATE_READY;
3417		netif_dbg(efx, hw, efx->net_dev,
3418			  "Done resetting and resuming IO after PCI error.\n");
3419	}
3420
3421out:
3422	rtnl_unlock();
3423}
3424
3425/* For simplicity and reliability, we always require a slot reset and try to
3426 * reset the hardware when a pci error affecting the device is detected.
3427 * We leave both the link_reset and mmio_enabled callback unimplemented:
3428 * with our request for slot reset the mmio_enabled callback will never be
3429 * called, and the link_reset callback is not used by AER or EEH mechanisms.
3430 */
3431static const struct pci_error_handlers efx_err_handlers = {
3432	.error_detected = efx_io_error_detected,
3433	.slot_reset	= efx_io_slot_reset,
3434	.resume		= efx_io_resume,
3435};
3436
3437static struct pci_driver efx_pci_driver = {
3438	.name		= KBUILD_MODNAME,
3439	.id_table	= efx_pci_table,
3440	.probe		= efx_pci_probe,
3441	.remove		= efx_pci_remove,
3442	.driver.pm	= &efx_pm_ops,
 
3443	.err_handler	= &efx_err_handlers,
3444#ifdef CONFIG_SFC_SRIOV
3445	.sriov_configure = efx_pci_sriov_configure,
3446#endif
3447};
3448
3449/**************************************************************************
3450 *
3451 * Kernel module interface
3452 *
3453 *************************************************************************/
3454
3455module_param(interrupt_mode, uint, 0444);
3456MODULE_PARM_DESC(interrupt_mode,
3457		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
3458
3459static int __init efx_init_module(void)
3460{
3461	int rc;
3462
3463	printk(KERN_INFO "Solarflare NET driver v" EFX_DRIVER_VERSION "\n");
3464
3465	rc = register_netdevice_notifier(&efx_netdev_notifier);
3466	if (rc)
3467		goto err_notifier;
3468
3469#ifdef CONFIG_SFC_SRIOV
3470	rc = efx_init_sriov();
3471	if (rc)
3472		goto err_sriov;
3473#endif
3474
3475	reset_workqueue = create_singlethread_workqueue("sfc_reset");
3476	if (!reset_workqueue) {
3477		rc = -ENOMEM;
3478		goto err_reset;
3479	}
3480
3481	rc = pci_register_driver(&efx_pci_driver);
3482	if (rc < 0)
3483		goto err_pci;
3484
 
 
 
 
3485	return 0;
3486
 
 
3487 err_pci:
3488	destroy_workqueue(reset_workqueue);
3489 err_reset:
3490#ifdef CONFIG_SFC_SRIOV
3491	efx_fini_sriov();
3492 err_sriov:
3493#endif
3494	unregister_netdevice_notifier(&efx_netdev_notifier);
3495 err_notifier:
3496	return rc;
3497}
3498
3499static void __exit efx_exit_module(void)
3500{
3501	printk(KERN_INFO "Solarflare NET driver unloading\n");
3502
 
3503	pci_unregister_driver(&efx_pci_driver);
3504	destroy_workqueue(reset_workqueue);
3505#ifdef CONFIG_SFC_SRIOV
3506	efx_fini_sriov();
3507#endif
3508	unregister_netdevice_notifier(&efx_netdev_notifier);
3509
3510}
3511
3512module_init(efx_init_module);
3513module_exit(efx_exit_module);
3514
3515MODULE_AUTHOR("Solarflare Communications and "
3516	      "Michael Brown <mbrown@fensystems.co.uk>");
3517MODULE_DESCRIPTION("Solarflare network driver");
3518MODULE_LICENSE("GPL");
3519MODULE_DEVICE_TABLE(pci, efx_pci_table);