xref: /linux/drivers/net/ethernet/sfc/efx.c (revision 91afa49a3eda8ebf902ce68f02cb16b48c879b6a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /****************************************************************************
3  * Driver for Solarflare network controllers and boards
4  * Copyright 2005-2006 Fen Systems Ltd.
5  * Copyright 2005-2013 Solarflare Communications Inc.
6  */
7 
8 #include <linux/filter.h>
9 #include <linux/module.h>
10 #include <linux/pci.h>
11 #include <linux/netdevice.h>
12 #include <linux/etherdevice.h>
13 #include <linux/delay.h>
14 #include <linux/notifier.h>
15 #include <linux/ip.h>
16 #include <linux/tcp.h>
17 #include <linux/in.h>
18 #include <linux/ethtool.h>
19 #include <linux/topology.h>
20 #include <linux/gfp.h>
21 #include <linux/interrupt.h>
22 #include "net_driver.h"
23 #include <net/gre.h>
24 #include <net/udp_tunnel.h>
25 #include <net/netdev_queues.h>
26 #include "efx.h"
27 #include "efx_common.h"
28 #include "efx_channels.h"
29 #include "ef100.h"
30 #include "rx_common.h"
31 #include "tx_common.h"
32 #include "nic.h"
33 #include "io.h"
34 #include "selftest.h"
35 #include "sriov.h"
36 #include "efx_devlink.h"
37 
38 #include "mcdi_port_common.h"
39 #include "mcdi_pcol.h"
40 #include "workarounds.h"
41 
42 /**************************************************************************
43  *
44  * Configurable values
45  *
46  *************************************************************************/
47 
48 module_param_named(interrupt_mode, efx_interrupt_mode, uint, 0444);
49 MODULE_PARM_DESC(interrupt_mode,
50 		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
51 
52 module_param(rss_cpus, uint, 0444);
53 MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");
54 
55 /*
56  * Use separate channels for TX and RX events
57  *
58  * Set this to 1 to use separate channels for TX and RX. It allows us
59  * to control interrupt affinity separately for TX and RX.
60  *
61  * This is only used in MSI-X interrupt mode
62  */
63 bool efx_separate_tx_channels;
64 module_param(efx_separate_tx_channels, bool, 0444);
65 MODULE_PARM_DESC(efx_separate_tx_channels,
66 		 "Use separate channels for TX and RX");
67 
68 /* Initial interrupt moderation settings.  They can be modified after
69  * module load with ethtool.
70  *
71  * The default for RX should strike a balance between increasing the
72  * round-trip latency and reducing overhead.
73  */
74 static unsigned int rx_irq_mod_usec = 60;
75 
76 /* Initial interrupt moderation settings.  They can be modified after
77  * module load with ethtool.
78  *
79  * This default is chosen to ensure that a 10G link does not go idle
80  * while a TX queue is stopped after it has become full.  A queue is
81  * restarted when it drops below half full.  The time this takes (assuming
82  * worst case 3 descriptors per packet and 1024 descriptors) is
83  *   512 / 3 * 1.2 = 205 usec.
84  */
85 static unsigned int tx_irq_mod_usec = 150;
86 
87 static bool phy_flash_cfg;
88 module_param(phy_flash_cfg, bool, 0644);
89 MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");
90 
91 static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
92 			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
93 			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
94 			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
95 module_param(debug, uint, 0);
96 MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
97 
98 /**************************************************************************
99  *
100  * Utility functions and prototypes
101  *
102  *************************************************************************/
103 
104 static void efx_remove_port(struct efx_nic *efx);
105 static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog);
106 static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp);
107 static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs,
108 			u32 flags);
109 
110 /**************************************************************************
111  *
112  * Port handling
113  *
114  **************************************************************************/
115 
116 static void efx_fini_port(struct efx_nic *efx);
117 
118 static int efx_probe_port(struct efx_nic *efx)
119 {
120 	int rc;
121 
122 	netif_dbg(efx, probe, efx->net_dev, "create port\n");
123 
124 	if (phy_flash_cfg)
125 		efx->phy_mode = PHY_MODE_SPECIAL;
126 
127 	/* Connect up MAC/PHY operations table */
128 	rc = efx->type->probe_port(efx);
129 	if (rc)
130 		return rc;
131 
132 	/* Initialise MAC address to permanent address */
133 	eth_hw_addr_set(efx->net_dev, efx->net_dev->perm_addr);
134 
135 	return 0;
136 }
137 
138 static int efx_init_port(struct efx_nic *efx)
139 {
140 	int rc;
141 
142 	netif_dbg(efx, drv, efx->net_dev, "init port\n");
143 
144 	mutex_lock(&efx->mac_lock);
145 
146 	efx->port_initialized = true;
147 
148 	/* Ensure the PHY advertises the correct flow control settings */
149 	rc = efx_mcdi_port_reconfigure(efx);
150 	if (rc && rc != -EPERM)
151 		goto fail;
152 
153 	mutex_unlock(&efx->mac_lock);
154 	return 0;
155 
156 fail:
157 	mutex_unlock(&efx->mac_lock);
158 	return rc;
159 }
160 
161 static void efx_fini_port(struct efx_nic *efx)
162 {
163 	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
164 
165 	if (!efx->port_initialized)
166 		return;
167 
168 	efx->port_initialized = false;
169 
170 	efx->link_state.up = false;
171 	efx_link_status_changed(efx);
172 }
173 
174 static void efx_remove_port(struct efx_nic *efx)
175 {
176 	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
177 
178 	efx->type->remove_port(efx);
179 }
180 
181 /**************************************************************************
182  *
183  * NIC handling
184  *
185  **************************************************************************/
186 
187 static LIST_HEAD(efx_primary_list);
188 static LIST_HEAD(efx_unassociated_list);
189 
190 static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right)
191 {
192 	return left->type == right->type &&
193 		left->vpd_sn && right->vpd_sn &&
194 		!strcmp(left->vpd_sn, right->vpd_sn);
195 }
196 
197 static void efx_associate(struct efx_nic *efx)
198 {
199 	struct efx_nic *other, *next;
200 
201 	if (efx->primary == efx) {
202 		/* Adding primary function; look for secondaries */
203 
204 		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
205 		list_add_tail(&efx->node, &efx_primary_list);
206 
207 		list_for_each_entry_safe(other, next, &efx_unassociated_list,
208 					 node) {
209 			if (efx_same_controller(efx, other)) {
210 				list_del(&other->node);
211 				netif_dbg(other, probe, other->net_dev,
212 					  "moving to secondary list of %s %s\n",
213 					  pci_name(efx->pci_dev),
214 					  efx->net_dev->name);
215 				list_add_tail(&other->node,
216 					      &efx->secondary_list);
217 				other->primary = efx;
218 			}
219 		}
220 	} else {
221 		/* Adding secondary function; look for primary */
222 
223 		list_for_each_entry(other, &efx_primary_list, node) {
224 			if (efx_same_controller(efx, other)) {
225 				netif_dbg(efx, probe, efx->net_dev,
226 					  "adding to secondary list of %s %s\n",
227 					  pci_name(other->pci_dev),
228 					  other->net_dev->name);
229 				list_add_tail(&efx->node,
230 					      &other->secondary_list);
231 				efx->primary = other;
232 				return;
233 			}
234 		}
235 
236 		netif_dbg(efx, probe, efx->net_dev,
237 			  "adding to unassociated list\n");
238 		list_add_tail(&efx->node, &efx_unassociated_list);
239 	}
240 }
241 
242 static void efx_dissociate(struct efx_nic *efx)
243 {
244 	struct efx_nic *other, *next;
245 
246 	list_del(&efx->node);
247 	efx->primary = NULL;
248 
249 	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
250 		list_del(&other->node);
251 		netif_dbg(other, probe, other->net_dev,
252 			  "moving to unassociated list\n");
253 		list_add_tail(&other->node, &efx_unassociated_list);
254 		other->primary = NULL;
255 	}
256 }
257 
258 static int efx_probe_nic(struct efx_nic *efx)
259 {
260 	int rc;
261 
262 	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
263 
264 	/* Carry out hardware-type specific initialisation */
265 	rc = efx->type->probe(efx);
266 	if (rc)
267 		return rc;
268 
269 	do {
270 		if (!efx->max_channels || !efx->max_tx_channels) {
271 			netif_err(efx, drv, efx->net_dev,
272 				  "Insufficient resources to allocate"
273 				  " any channels\n");
274 			rc = -ENOSPC;
275 			goto fail1;
276 		}
277 
278 		/* Determine the number of channels and queues by trying
279 		 * to hook in MSI-X interrupts.
280 		 */
281 		rc = efx_probe_interrupts(efx);
282 		if (rc)
283 			goto fail1;
284 
285 		rc = efx_set_channels(efx);
286 		if (rc)
287 			goto fail1;
288 
289 		/* dimension_resources can fail with EAGAIN */
290 		rc = efx->type->dimension_resources(efx);
291 		if (rc != 0 && rc != -EAGAIN)
292 			goto fail2;
293 
294 		if (rc == -EAGAIN)
295 			/* try again with new max_channels */
296 			efx_remove_interrupts(efx);
297 
298 	} while (rc == -EAGAIN);
299 
300 	if (efx->n_channels > 1)
301 		netdev_rss_key_fill(efx->rss_context.rx_hash_key,
302 				    sizeof(efx->rss_context.rx_hash_key));
303 	efx_set_default_rx_indir_table(efx, efx->rss_context.rx_indir_table);
304 
305 	/* Initialise the interrupt moderation settings */
306 	efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000);
307 	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
308 				true);
309 
310 	return 0;
311 
312 fail2:
313 	efx_remove_interrupts(efx);
314 fail1:
315 	efx->type->remove(efx);
316 	return rc;
317 }
318 
319 static void efx_remove_nic(struct efx_nic *efx)
320 {
321 	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
322 
323 	efx_remove_interrupts(efx);
324 	efx->type->remove(efx);
325 }
326 
327 /**************************************************************************
328  *
329  * NIC startup/shutdown
330  *
331  *************************************************************************/
332 
333 static int efx_probe_all(struct efx_nic *efx)
334 {
335 	int rc;
336 
337 	rc = efx_probe_nic(efx);
338 	if (rc) {
339 		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
340 		goto fail1;
341 	}
342 
343 	rc = efx_probe_port(efx);
344 	if (rc) {
345 		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
346 		goto fail2;
347 	}
348 
349 	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
350 	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
351 		rc = -EINVAL;
352 		goto fail3;
353 	}
354 
355 #ifdef CONFIG_SFC_SRIOV
356 	rc = efx->type->vswitching_probe(efx);
357 	if (rc) /* not fatal; the PF will still work fine */
358 		netif_warn(efx, probe, efx->net_dev,
359 			   "failed to setup vswitching rc=%d;"
360 			   " VFs may not function\n", rc);
361 #endif
362 
363 	rc = efx_probe_filters(efx);
364 	if (rc) {
365 		netif_err(efx, probe, efx->net_dev,
366 			  "failed to create filter tables\n");
367 		goto fail4;
368 	}
369 
370 	rc = efx_probe_channels(efx);
371 	if (rc)
372 		goto fail5;
373 
374 	efx->state = STATE_NET_DOWN;
375 
376 	return 0;
377 
378  fail5:
379 	efx_remove_filters(efx);
380  fail4:
381 #ifdef CONFIG_SFC_SRIOV
382 	efx->type->vswitching_remove(efx);
383 #endif
384  fail3:
385 	efx_remove_port(efx);
386  fail2:
387 	efx_remove_nic(efx);
388  fail1:
389 	return rc;
390 }
391 
392 static void efx_remove_all(struct efx_nic *efx)
393 {
394 	rtnl_lock();
395 	efx_xdp_setup_prog(efx, NULL);
396 	rtnl_unlock();
397 
398 	efx_remove_channels(efx);
399 	efx_remove_filters(efx);
400 #ifdef CONFIG_SFC_SRIOV
401 	efx->type->vswitching_remove(efx);
402 #endif
403 	efx_remove_port(efx);
404 	efx_remove_nic(efx);
405 }
406 
407 /**************************************************************************
408  *
409  * Interrupt moderation
410  *
411  **************************************************************************/
412 unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs)
413 {
414 	if (usecs == 0)
415 		return 0;
416 	if (usecs * 1000 < efx->timer_quantum_ns)
417 		return 1; /* never round down to 0 */
418 	return usecs * 1000 / efx->timer_quantum_ns;
419 }
420 
421 unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks)
422 {
423 	/* We must round up when converting ticks to microseconds
424 	 * because we round down when converting the other way.
425 	 */
426 	return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000);
427 }
428 
429 /* Set interrupt moderation parameters */
430 int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
431 			    unsigned int rx_usecs, bool rx_adaptive,
432 			    bool rx_may_override_tx)
433 {
434 	struct efx_channel *channel;
435 	unsigned int timer_max_us;
436 
437 	EFX_ASSERT_RESET_SERIALISED(efx);
438 
439 	timer_max_us = efx->timer_max_ns / 1000;
440 
441 	if (tx_usecs > timer_max_us || rx_usecs > timer_max_us)
442 		return -EINVAL;
443 
444 	if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 &&
445 	    !rx_may_override_tx) {
446 		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
447 			  "RX and TX IRQ moderation must be equal\n");
448 		return -EINVAL;
449 	}
450 
451 	efx->irq_rx_adaptive = rx_adaptive;
452 	efx->irq_rx_moderation_us = rx_usecs;
453 	efx_for_each_channel(channel, efx) {
454 		if (efx_channel_has_rx_queue(channel))
455 			channel->irq_moderation_us = rx_usecs;
456 		else if (efx_channel_has_tx_queues(channel))
457 			channel->irq_moderation_us = tx_usecs;
458 		else if (efx_channel_is_xdp_tx(channel))
459 			channel->irq_moderation_us = tx_usecs;
460 	}
461 
462 	return 0;
463 }
464 
465 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
466 			    unsigned int *rx_usecs, bool *rx_adaptive)
467 {
468 	*rx_adaptive = efx->irq_rx_adaptive;
469 	*rx_usecs = efx->irq_rx_moderation_us;
470 
471 	/* If channels are shared between RX and TX, so is IRQ
472 	 * moderation.  Otherwise, IRQ moderation is the same for all
473 	 * TX channels and is not adaptive.
474 	 */
475 	if (efx->tx_channel_offset == 0) {
476 		*tx_usecs = *rx_usecs;
477 	} else {
478 		struct efx_channel *tx_channel;
479 
480 		tx_channel = efx->channel[efx->tx_channel_offset];
481 		*tx_usecs = tx_channel->irq_moderation_us;
482 	}
483 }
484 
485 /**************************************************************************
486  *
487  * ioctls
488  *
489  *************************************************************************/
490 
491 /* Net device ioctl
492  * Context: process, rtnl_lock() held.
493  */
494 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
495 {
496 	struct efx_nic *efx = efx_netdev_priv(net_dev);
497 	struct mii_ioctl_data *data = if_mii(ifr);
498 
499 	/* Convert phy_id from older PRTAD/DEVAD format */
500 	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
501 	    (data->phy_id & 0xfc00) == 0x0400)
502 		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;
503 
504 	return mdio_mii_ioctl(&efx->mdio, data, cmd);
505 }
506 
507 /**************************************************************************
508  *
509  * Kernel net device interface
510  *
511  *************************************************************************/
512 
513 /* Context: process, rtnl_lock() held. */
514 int efx_net_open(struct net_device *net_dev)
515 {
516 	struct efx_nic *efx = efx_netdev_priv(net_dev);
517 	int rc;
518 
519 	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
520 		  raw_smp_processor_id());
521 
522 	rc = efx_check_disabled(efx);
523 	if (rc)
524 		return rc;
525 	if (efx->phy_mode & PHY_MODE_SPECIAL)
526 		return -EBUSY;
527 	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
528 		return -EIO;
529 
530 	/* Notify the kernel of the link state polled during driver load,
531 	 * before the monitor starts running */
532 	efx_link_status_changed(efx);
533 
534 	efx_start_all(efx);
535 	if (efx->state == STATE_DISABLED || efx->reset_pending)
536 		netif_device_detach(efx->net_dev);
537 	else
538 		efx->state = STATE_NET_UP;
539 
540 	return 0;
541 }
542 
543 /* Context: process, rtnl_lock() held.
544  * Note that the kernel will ignore our return code; this method
545  * should really be a void.
546  */
547 int efx_net_stop(struct net_device *net_dev)
548 {
549 	struct efx_nic *efx = efx_netdev_priv(net_dev);
550 
551 	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
552 		  raw_smp_processor_id());
553 
554 	/* Stop the device and flush all the channels */
555 	efx_stop_all(efx);
556 
557 	return 0;
558 }
559 
560 static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid)
561 {
562 	struct efx_nic *efx = efx_netdev_priv(net_dev);
563 
564 	if (efx->type->vlan_rx_add_vid)
565 		return efx->type->vlan_rx_add_vid(efx, proto, vid);
566 	else
567 		return -EOPNOTSUPP;
568 }
569 
570 static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid)
571 {
572 	struct efx_nic *efx = efx_netdev_priv(net_dev);
573 
574 	if (efx->type->vlan_rx_kill_vid)
575 		return efx->type->vlan_rx_kill_vid(efx, proto, vid);
576 	else
577 		return -EOPNOTSUPP;
578 }
579 
580 static int efx_hwtstamp_set(struct net_device *net_dev,
581 			    struct kernel_hwtstamp_config *config,
582 			    struct netlink_ext_ack *extack)
583 {
584 	struct efx_nic *efx = efx_netdev_priv(net_dev);
585 
586 	return efx_ptp_set_ts_config(efx, config, extack);
587 }
588 
589 static int efx_hwtstamp_get(struct net_device *net_dev,
590 			    struct kernel_hwtstamp_config *config)
591 {
592 	struct efx_nic *efx = efx_netdev_priv(net_dev);
593 
594 	return efx_ptp_get_ts_config(efx, config);
595 }
596 
597 static const struct net_device_ops efx_netdev_ops = {
598 	.ndo_open		= efx_net_open,
599 	.ndo_stop		= efx_net_stop,
600 	.ndo_get_stats64	= efx_net_stats,
601 	.ndo_tx_timeout		= efx_watchdog,
602 	.ndo_start_xmit		= efx_hard_start_xmit,
603 	.ndo_validate_addr	= eth_validate_addr,
604 	.ndo_eth_ioctl		= efx_ioctl,
605 	.ndo_change_mtu		= efx_change_mtu,
606 	.ndo_set_mac_address	= efx_set_mac_address,
607 	.ndo_set_rx_mode	= efx_set_rx_mode,
608 	.ndo_set_features	= efx_set_features,
609 	.ndo_features_check	= efx_features_check,
610 	.ndo_vlan_rx_add_vid	= efx_vlan_rx_add_vid,
611 	.ndo_vlan_rx_kill_vid	= efx_vlan_rx_kill_vid,
612 	.ndo_hwtstamp_set	= efx_hwtstamp_set,
613 	.ndo_hwtstamp_get	= efx_hwtstamp_get,
614 #ifdef CONFIG_SFC_SRIOV
615 	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
616 	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
617 	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
618 	.ndo_get_vf_config	= efx_sriov_get_vf_config,
619 	.ndo_set_vf_link_state  = efx_sriov_set_vf_link_state,
620 #endif
621 	.ndo_get_phys_port_id   = efx_get_phys_port_id,
622 	.ndo_get_phys_port_name	= efx_get_phys_port_name,
623 #ifdef CONFIG_RFS_ACCEL
624 	.ndo_rx_flow_steer	= efx_filter_rfs,
625 #endif
626 	.ndo_xdp_xmit		= efx_xdp_xmit,
627 	.ndo_bpf		= efx_xdp
628 };
629 
630 static void efx_get_queue_stats_rx(struct net_device *net_dev, int idx,
631 				   struct netdev_queue_stats_rx *stats)
632 {
633 	struct efx_nic *efx = efx_netdev_priv(net_dev);
634 	struct efx_rx_queue *rx_queue;
635 	struct efx_channel *channel;
636 
637 	channel = efx_get_channel(efx, idx);
638 	rx_queue = efx_channel_get_rx_queue(channel);
639 	/* Count only packets since last time datapath was started */
640 	stats->packets = rx_queue->rx_packets - rx_queue->old_rx_packets;
641 	stats->bytes = rx_queue->rx_bytes - rx_queue->old_rx_bytes;
642 	stats->hw_drops = efx_get_queue_stat_rx_hw_drops(channel) -
643 			  channel->old_n_rx_hw_drops;
644 	stats->hw_drop_overruns = channel->n_rx_nodesc_trunc -
645 				  channel->old_n_rx_hw_drop_overruns;
646 }
647 
648 static void efx_get_queue_stats_tx(struct net_device *net_dev, int idx,
649 				   struct netdev_queue_stats_tx *stats)
650 {
651 	struct efx_nic *efx = efx_netdev_priv(net_dev);
652 	struct efx_tx_queue *tx_queue;
653 	struct efx_channel *channel;
654 
655 	channel = efx_get_tx_channel(efx, idx);
656 	stats->packets = 0;
657 	stats->bytes = 0;
658 	stats->hw_gso_packets = 0;
659 	stats->hw_gso_wire_packets = 0;
660 	efx_for_each_channel_tx_queue(tx_queue, channel) {
661 		stats->packets += tx_queue->complete_packets -
662 				  tx_queue->old_complete_packets;
663 		stats->bytes += tx_queue->complete_bytes -
664 				tx_queue->old_complete_bytes;
665 		/* Note that, unlike stats->packets and stats->bytes,
666 		 * these count TXes enqueued, rather than completed,
667 		 * which may not be what users expect.
668 		 */
669 		stats->hw_gso_packets += tx_queue->tso_bursts -
670 					 tx_queue->old_tso_bursts;
671 		stats->hw_gso_wire_packets += tx_queue->tso_packets -
672 					      tx_queue->old_tso_packets;
673 	}
674 }
675 
676 static void efx_get_base_stats(struct net_device *net_dev,
677 			       struct netdev_queue_stats_rx *rx,
678 			       struct netdev_queue_stats_tx *tx)
679 {
680 	struct efx_nic *efx = efx_netdev_priv(net_dev);
681 	struct efx_tx_queue *tx_queue;
682 	struct efx_rx_queue *rx_queue;
683 	struct efx_channel *channel;
684 
685 	rx->packets = 0;
686 	rx->bytes = 0;
687 	rx->hw_drops = 0;
688 	rx->hw_drop_overruns = 0;
689 	tx->packets = 0;
690 	tx->bytes = 0;
691 	tx->hw_gso_packets = 0;
692 	tx->hw_gso_wire_packets = 0;
693 
694 	/* Count all packets on non-core queues, and packets before last
695 	 * datapath start on core queues.
696 	 */
697 	efx_for_each_channel(channel, efx) {
698 		rx_queue = efx_channel_get_rx_queue(channel);
699 		if (channel->channel >= net_dev->real_num_rx_queues) {
700 			rx->packets += rx_queue->rx_packets;
701 			rx->bytes += rx_queue->rx_bytes;
702 			rx->hw_drops += efx_get_queue_stat_rx_hw_drops(channel);
703 			rx->hw_drop_overruns += channel->n_rx_nodesc_trunc;
704 		} else {
705 			rx->packets += rx_queue->old_rx_packets;
706 			rx->bytes += rx_queue->old_rx_bytes;
707 			rx->hw_drops += channel->old_n_rx_hw_drops;
708 			rx->hw_drop_overruns += channel->old_n_rx_hw_drop_overruns;
709 		}
710 		efx_for_each_channel_tx_queue(tx_queue, channel) {
711 			if (channel->channel < efx->tx_channel_offset ||
712 			    channel->channel >= efx->tx_channel_offset +
713 						net_dev->real_num_tx_queues) {
714 				tx->packets += tx_queue->complete_packets;
715 				tx->bytes += tx_queue->complete_bytes;
716 				tx->hw_gso_packets += tx_queue->tso_bursts;
717 				tx->hw_gso_wire_packets += tx_queue->tso_packets;
718 			} else {
719 				tx->packets += tx_queue->old_complete_packets;
720 				tx->bytes += tx_queue->old_complete_bytes;
721 				tx->hw_gso_packets += tx_queue->old_tso_bursts;
722 				tx->hw_gso_wire_packets += tx_queue->old_tso_packets;
723 			}
724 			/* Include XDP TX in device-wide stats */
725 			tx->packets += tx_queue->complete_xdp_packets;
726 			tx->bytes += tx_queue->complete_xdp_bytes;
727 		}
728 	}
729 }
730 
731 static const struct netdev_stat_ops efx_stat_ops = {
732 	.get_queue_stats_rx	= efx_get_queue_stats_rx,
733 	.get_queue_stats_tx	= efx_get_queue_stats_tx,
734 	.get_base_stats		= efx_get_base_stats,
735 };
736 
737 static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog)
738 {
739 	struct bpf_prog *old_prog;
740 
741 	if (efx->xdp_rxq_info_failed) {
742 		netif_err(efx, drv, efx->net_dev,
743 			  "Unable to bind XDP program due to previous failure of rxq_info\n");
744 		return -EINVAL;
745 	}
746 
747 	if (prog && efx->net_dev->mtu > efx_xdp_max_mtu(efx)) {
748 		netif_err(efx, drv, efx->net_dev,
749 			  "Unable to configure XDP with MTU of %d (max: %d)\n",
750 			  efx->net_dev->mtu, efx_xdp_max_mtu(efx));
751 		return -EINVAL;
752 	}
753 
754 	old_prog = rtnl_dereference(efx->xdp_prog);
755 	rcu_assign_pointer(efx->xdp_prog, prog);
756 	/* Release the reference that was originally passed by the caller. */
757 	if (old_prog)
758 		bpf_prog_put(old_prog);
759 
760 	return 0;
761 }
762 
763 /* Context: process, rtnl_lock() held. */
764 static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp)
765 {
766 	struct efx_nic *efx = efx_netdev_priv(dev);
767 
768 	switch (xdp->command) {
769 	case XDP_SETUP_PROG:
770 		return efx_xdp_setup_prog(efx, xdp->prog);
771 	default:
772 		return -EINVAL;
773 	}
774 }
775 
776 static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs,
777 			u32 flags)
778 {
779 	struct efx_nic *efx = efx_netdev_priv(dev);
780 
781 	if (!netif_running(dev))
782 		return -EINVAL;
783 
784 	return efx_xdp_tx_buffers(efx, n, xdpfs, flags & XDP_XMIT_FLUSH);
785 }
786 
787 static void efx_update_name(struct efx_nic *efx)
788 {
789 	strcpy(efx->name, efx->net_dev->name);
790 	efx_mtd_rename(efx);
791 	efx_set_channel_names(efx);
792 }
793 
794 static int efx_netdev_event(struct notifier_block *this,
795 			    unsigned long event, void *ptr)
796 {
797 	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
798 
799 	if ((net_dev->netdev_ops == &efx_netdev_ops) &&
800 	    event == NETDEV_CHANGENAME)
801 		efx_update_name(efx_netdev_priv(net_dev));
802 
803 	return NOTIFY_DONE;
804 }
805 
806 static struct notifier_block efx_netdev_notifier = {
807 	.notifier_call = efx_netdev_event,
808 };
809 
810 static ssize_t phy_type_show(struct device *dev,
811 			     struct device_attribute *attr, char *buf)
812 {
813 	struct efx_nic *efx = dev_get_drvdata(dev);
814 	return sprintf(buf, "%d\n", efx->phy_type);
815 }
816 static DEVICE_ATTR_RO(phy_type);
817 
818 static int efx_register_netdev(struct efx_nic *efx)
819 {
820 	struct net_device *net_dev = efx->net_dev;
821 	struct efx_channel *channel;
822 	int rc;
823 
824 	net_dev->watchdog_timeo = 5 * HZ;
825 	net_dev->irq = efx->pci_dev->irq;
826 	net_dev->netdev_ops = &efx_netdev_ops;
827 	net_dev->stat_ops = &efx_stat_ops;
828 	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0)
829 		net_dev->priv_flags |= IFF_UNICAST_FLT;
830 	net_dev->ethtool_ops = &efx_ethtool_ops;
831 	netif_set_tso_max_segs(net_dev, EFX_TSO_MAX_SEGS);
832 	net_dev->min_mtu = EFX_MIN_MTU;
833 	net_dev->max_mtu = EFX_MAX_MTU;
834 
835 	rtnl_lock();
836 
837 	/* Enable resets to be scheduled and check whether any were
838 	 * already requested.  If so, the NIC is probably hosed so we
839 	 * abort.
840 	 */
841 	if (efx->reset_pending) {
842 		pci_err(efx->pci_dev, "aborting probe due to scheduled reset\n");
843 		rc = -EIO;
844 		goto fail_locked;
845 	}
846 
847 	rc = dev_alloc_name(net_dev, net_dev->name);
848 	if (rc < 0)
849 		goto fail_locked;
850 	efx_update_name(efx);
851 
852 	/* Always start with carrier off; PHY events will detect the link */
853 	netif_carrier_off(net_dev);
854 
855 	rc = register_netdevice(net_dev);
856 	if (rc)
857 		goto fail_locked;
858 
859 	efx_for_each_channel(channel, efx) {
860 		struct efx_tx_queue *tx_queue;
861 		efx_for_each_channel_tx_queue(tx_queue, channel)
862 			efx_init_tx_queue_core_txq(tx_queue);
863 	}
864 
865 	efx_associate(efx);
866 
867 	efx->state = STATE_NET_DOWN;
868 
869 	rtnl_unlock();
870 
871 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
872 	if (rc) {
873 		netif_err(efx, drv, efx->net_dev,
874 			  "failed to init net dev attributes\n");
875 		goto fail_registered;
876 	}
877 
878 	efx_init_mcdi_logging(efx);
879 
880 	return 0;
881 
882 fail_registered:
883 	rtnl_lock();
884 	efx_dissociate(efx);
885 	unregister_netdevice(net_dev);
886 fail_locked:
887 	efx->state = STATE_UNINIT;
888 	rtnl_unlock();
889 	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
890 	return rc;
891 }
892 
893 static void efx_unregister_netdev(struct efx_nic *efx)
894 {
895 	if (!efx->net_dev)
896 		return;
897 
898 	if (WARN_ON(efx_netdev_priv(efx->net_dev) != efx))
899 		return;
900 
901 	if (efx_dev_registered(efx)) {
902 		strscpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
903 		efx_fini_mcdi_logging(efx);
904 		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
905 		unregister_netdev(efx->net_dev);
906 	}
907 }
908 
909 /**************************************************************************
910  *
911  * List of NICs we support
912  *
913  **************************************************************************/
914 
915 /* PCI device ID table */
916 static const struct pci_device_id efx_pci_table[] = {
917 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
918 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
919 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903),  /* SFC9120 VF */
920 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
921 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923),  /* SFC9140 PF */
922 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
923 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923),  /* SFC9140 VF */
924 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
925 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03),  /* SFC9220 PF */
926 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
927 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03),  /* SFC9220 VF */
928 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
929 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0b03),  /* SFC9250 PF */
930 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
931 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1b03),  /* SFC9250 VF */
932 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
933 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0c03),  /* X4 PF (FF/LL) */
934 	 .driver_data = (unsigned long)&efx_x4_nic_type},
935 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x2c03),  /* X4 PF (FF only) */
936 	 .driver_data = (unsigned long)&efx_x4_nic_type},
937 	{0}			/* end of list */
938 };
939 
940 /**************************************************************************
941  *
942  * Data housekeeping
943  *
944  **************************************************************************/
945 
946 void efx_update_sw_stats(struct efx_nic *efx, u64 *stats)
947 {
948 	u64 n_rx_nodesc_trunc = 0;
949 	struct efx_channel *channel;
950 
951 	efx_for_each_channel(channel, efx)
952 		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
953 	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
954 	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
955 }
956 
957 /**************************************************************************
958  *
959  * PCI interface
960  *
961  **************************************************************************/
962 
963 /* Main body of final NIC shutdown code
964  * This is called only at module unload (or hotplug removal).
965  */
966 static void efx_pci_remove_main(struct efx_nic *efx)
967 {
968 	/* Flush reset_work. It can no longer be scheduled since we
969 	 * are not READY.
970 	 */
971 	WARN_ON(efx_net_active(efx->state));
972 	efx_flush_reset_workqueue(efx);
973 
974 	efx_disable_interrupts(efx);
975 	efx_clear_interrupt_affinity(efx);
976 	efx_nic_fini_interrupt(efx);
977 	efx_fini_port(efx);
978 	efx->type->fini(efx);
979 	efx_fini_napi(efx);
980 	efx_remove_all(efx);
981 }
982 
983 /* Final NIC shutdown
984  * This is called only at module unload (or hotplug removal).  A PF can call
985  * this on its VFs to ensure they are unbound first.
986  */
987 static void efx_pci_remove(struct pci_dev *pci_dev)
988 {
989 	struct efx_probe_data *probe_data;
990 	struct efx_nic *efx;
991 
992 	efx = pci_get_drvdata(pci_dev);
993 	if (!efx)
994 		return;
995 
996 	/* Mark the NIC as fini, then stop the interface */
997 	rtnl_lock();
998 	efx_dissociate(efx);
999 	dev_close(efx->net_dev);
1000 	efx_disable_interrupts(efx);
1001 	efx->state = STATE_UNINIT;
1002 	rtnl_unlock();
1003 
1004 	if (efx->type->sriov_fini)
1005 		efx->type->sriov_fini(efx);
1006 
1007 	efx_fini_devlink_lock(efx);
1008 	efx_unregister_netdev(efx);
1009 
1010 	efx_mtd_remove(efx);
1011 
1012 	efx_pci_remove_main(efx);
1013 
1014 	efx_fini_io(efx);
1015 	pci_dbg(efx->pci_dev, "shutdown successful\n");
1016 
1017 	efx_fini_devlink_and_unlock(efx);
1018 	efx_fini_struct(efx);
1019 	free_netdev(efx->net_dev);
1020 	probe_data = container_of(efx, struct efx_probe_data, efx);
1021 	kfree(probe_data);
1022 };
1023 
1024 /* NIC VPD information
1025  * Called during probe to display the part number of the
1026  * installed NIC.
1027  */
1028 static void efx_probe_vpd_strings(struct efx_nic *efx)
1029 {
1030 	struct pci_dev *dev = efx->pci_dev;
1031 	unsigned int vpd_size, kw_len;
1032 	u8 *vpd_data;
1033 	int start;
1034 
1035 	vpd_data = pci_vpd_alloc(dev, &vpd_size);
1036 	if (IS_ERR(vpd_data)) {
1037 		pci_warn(dev, "Unable to read VPD\n");
1038 		return;
1039 	}
1040 
1041 	start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size,
1042 					     PCI_VPD_RO_KEYWORD_PARTNO, &kw_len);
1043 	if (start < 0)
1044 		pci_err(dev, "Part number not found or incomplete\n");
1045 	else
1046 		pci_info(dev, "Part Number : %.*s\n", kw_len, vpd_data + start);
1047 
1048 	start = pci_vpd_find_ro_info_keyword(vpd_data, vpd_size,
1049 					     PCI_VPD_RO_KEYWORD_SERIALNO, &kw_len);
1050 	if (start < 0)
1051 		pci_err(dev, "Serial number not found or incomplete\n");
1052 	else
1053 		efx->vpd_sn = kmemdup_nul(vpd_data + start, kw_len, GFP_KERNEL);
1054 
1055 	kfree(vpd_data);
1056 }
1057 
1058 
1059 /* Main body of NIC initialisation
1060  * This is called at module load (or hotplug insertion, theoretically).
1061  */
1062 static int efx_pci_probe_main(struct efx_nic *efx)
1063 {
1064 	int rc;
1065 
1066 	/* Do start-of-day initialisation */
1067 	rc = efx_probe_all(efx);
1068 	if (rc)
1069 		goto fail1;
1070 
1071 	efx_init_napi(efx);
1072 
1073 	down_write(&efx->filter_sem);
1074 	rc = efx->type->init(efx);
1075 	up_write(&efx->filter_sem);
1076 	if (rc) {
1077 		pci_err(efx->pci_dev, "failed to initialise NIC\n");
1078 		goto fail3;
1079 	}
1080 
1081 	rc = efx_init_port(efx);
1082 	if (rc) {
1083 		netif_err(efx, probe, efx->net_dev,
1084 			  "failed to initialise port\n");
1085 		goto fail4;
1086 	}
1087 
1088 	rc = efx_nic_init_interrupt(efx);
1089 	if (rc)
1090 		goto fail5;
1091 
1092 	efx_set_interrupt_affinity(efx);
1093 	rc = efx_enable_interrupts(efx);
1094 	if (rc)
1095 		goto fail6;
1096 
1097 	return 0;
1098 
1099  fail6:
1100 	efx_clear_interrupt_affinity(efx);
1101 	efx_nic_fini_interrupt(efx);
1102  fail5:
1103 	efx_fini_port(efx);
1104  fail4:
1105 	efx->type->fini(efx);
1106  fail3:
1107 	efx_fini_napi(efx);
1108 	efx_remove_all(efx);
1109  fail1:
1110 	return rc;
1111 }
1112 
1113 static int efx_pci_probe_post_io(struct efx_nic *efx)
1114 {
1115 	struct net_device *net_dev = efx->net_dev;
1116 	int rc = efx_pci_probe_main(efx);
1117 
1118 	if (rc)
1119 		return rc;
1120 
1121 	if (efx->type->sriov_init) {
1122 		rc = efx->type->sriov_init(efx);
1123 		if (rc)
1124 			pci_err(efx->pci_dev, "SR-IOV can't be enabled rc %d\n",
1125 				rc);
1126 	}
1127 
1128 	/* Determine netdevice features */
1129 	net_dev->features |= efx->type->offload_features;
1130 
1131 	/* Add TSO features */
1132 	if (efx->type->tso_versions && efx->type->tso_versions(efx))
1133 		net_dev->features |= NETIF_F_TSO | NETIF_F_TSO6;
1134 
1135 	/* Mask for features that also apply to VLAN devices */
1136 	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
1137 				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
1138 				   NETIF_F_RXCSUM);
1139 
1140 	/* Determine user configurable features */
1141 	net_dev->hw_features |= net_dev->features & ~efx->fixed_features;
1142 
1143 	/* Disable receiving frames with bad FCS, by default. */
1144 	net_dev->features &= ~NETIF_F_RXALL;
1145 
1146 	/* Disable VLAN filtering by default.  It may be enforced if
1147 	 * the feature is fixed (i.e. VLAN filters are required to
1148 	 * receive VLAN tagged packets due to vPort restrictions).
1149 	 */
1150 	net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
1151 	net_dev->features |= efx->fixed_features;
1152 
1153 	net_dev->xdp_features = NETDEV_XDP_ACT_BASIC |
1154 				NETDEV_XDP_ACT_REDIRECT |
1155 				NETDEV_XDP_ACT_NDO_XMIT;
1156 
1157 	/* devlink creation, registration and lock */
1158 	rc = efx_probe_devlink_and_lock(efx);
1159 	if (rc)
1160 		pci_err(efx->pci_dev, "devlink registration failed");
1161 
1162 	rc = efx_register_netdev(efx);
1163 	efx_probe_devlink_unlock(efx);
1164 	if (!rc)
1165 		return 0;
1166 
1167 	efx_pci_remove_main(efx);
1168 	return rc;
1169 }
1170 
1171 /* NIC initialisation
1172  *
1173  * This is called at module load (or hotplug insertion,
1174  * theoretically).  It sets up PCI mappings, resets the NIC,
1175  * sets up and registers the network devices with the kernel and hooks
1176  * the interrupt service routine.  It does not prepare the device for
1177  * transmission; this is left to the first time one of the network
1178  * interfaces is brought up (i.e. efx_net_open).
1179  */
1180 static int efx_pci_probe(struct pci_dev *pci_dev,
1181 			 const struct pci_device_id *entry)
1182 {
1183 	struct efx_probe_data *probe_data, **probe_ptr;
1184 	struct net_device *net_dev;
1185 	struct efx_nic *efx;
1186 	int rc;
1187 
1188 	/* Allocate probe data and struct efx_nic */
1189 	probe_data = kzalloc(sizeof(*probe_data), GFP_KERNEL);
1190 	if (!probe_data)
1191 		return -ENOMEM;
1192 	probe_data->pci_dev = pci_dev;
1193 	efx = &probe_data->efx;
1194 
1195 	/* Allocate and initialise a struct net_device */
1196 	net_dev = alloc_etherdev_mq(sizeof(probe_data), EFX_MAX_CORE_TX_QUEUES);
1197 	if (!net_dev) {
1198 		rc = -ENOMEM;
1199 		goto fail0;
1200 	}
1201 	probe_ptr = netdev_priv(net_dev);
1202 	*probe_ptr = probe_data;
1203 	efx->net_dev = net_dev;
1204 	efx->type = (const struct efx_nic_type *) entry->driver_data;
1205 	efx->fixed_features |= NETIF_F_HIGHDMA;
1206 
1207 	pci_set_drvdata(pci_dev, efx);
1208 	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
1209 	rc = efx_init_struct(efx, pci_dev);
1210 	if (rc)
1211 		goto fail1;
1212 	efx->mdio.dev = net_dev;
1213 
1214 	pci_info(pci_dev, "Solarflare NIC detected\n");
1215 
1216 	if (!efx->type->is_vf)
1217 		efx_probe_vpd_strings(efx);
1218 
1219 	/* Set up basic I/O (BAR mappings etc) */
1220 	rc = efx_init_io(efx, efx->type->mem_bar(efx), efx->type->max_dma_mask,
1221 			 efx->type->mem_map_size(efx));
1222 	if (rc)
1223 		goto fail2;
1224 
1225 	rc = efx_pci_probe_post_io(efx);
1226 	if (rc) {
1227 		/* On failure, retry once immediately.
1228 		 * If we aborted probe due to a scheduled reset, dismiss it.
1229 		 */
1230 		efx->reset_pending = 0;
1231 		rc = efx_pci_probe_post_io(efx);
1232 		if (rc) {
1233 			/* On another failure, retry once more
1234 			 * after a 50-305ms delay.
1235 			 */
1236 			unsigned char r;
1237 
1238 			get_random_bytes(&r, 1);
1239 			msleep((unsigned int)r + 50);
1240 			efx->reset_pending = 0;
1241 			rc = efx_pci_probe_post_io(efx);
1242 		}
1243 	}
1244 	if (rc)
1245 		goto fail3;
1246 
1247 	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
1248 
1249 	/* Try to create MTDs, but allow this to fail */
1250 	rtnl_lock();
1251 	rc = efx_mtd_probe(efx);
1252 	rtnl_unlock();
1253 	if (rc && rc != -EPERM)
1254 		netif_warn(efx, probe, efx->net_dev,
1255 			   "failed to create MTDs (%d)\n", rc);
1256 
1257 	if (efx->type->udp_tnl_push_ports)
1258 		efx->type->udp_tnl_push_ports(efx);
1259 
1260 	return 0;
1261 
1262  fail3:
1263 	efx_fini_io(efx);
1264  fail2:
1265 	efx_fini_struct(efx);
1266  fail1:
1267 	WARN_ON(rc > 0);
1268 	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
1269 	free_netdev(net_dev);
1270  fail0:
1271 	kfree(probe_data);
1272 	return rc;
1273 }
1274 
1275 /* efx_pci_sriov_configure returns the actual number of Virtual Functions
1276  * enabled on success
1277  */
1278 #ifdef CONFIG_SFC_SRIOV
1279 static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
1280 {
1281 	int rc;
1282 	struct efx_nic *efx = pci_get_drvdata(dev);
1283 
1284 	if (efx->type->sriov_configure) {
1285 		rc = efx->type->sriov_configure(efx, num_vfs);
1286 		if (rc)
1287 			return rc;
1288 		else
1289 			return num_vfs;
1290 	} else
1291 		return -EOPNOTSUPP;
1292 }
1293 #endif
1294 
1295 static int efx_pm_freeze(struct device *dev)
1296 {
1297 	struct efx_nic *efx = dev_get_drvdata(dev);
1298 
1299 	rtnl_lock();
1300 
1301 	if (efx_net_active(efx->state)) {
1302 		efx_device_detach_sync(efx);
1303 
1304 		efx_stop_all(efx);
1305 		efx_disable_interrupts(efx);
1306 
1307 		efx->state = efx_freeze(efx->state);
1308 	}
1309 
1310 	rtnl_unlock();
1311 
1312 	return 0;
1313 }
1314 
1315 static void efx_pci_shutdown(struct pci_dev *pci_dev)
1316 {
1317 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
1318 
1319 	if (!efx)
1320 		return;
1321 
1322 	efx_pm_freeze(&pci_dev->dev);
1323 	pci_disable_device(pci_dev);
1324 }
1325 
1326 static int efx_pm_thaw(struct device *dev)
1327 {
1328 	int rc;
1329 	struct efx_nic *efx = dev_get_drvdata(dev);
1330 
1331 	rtnl_lock();
1332 
1333 	if (efx_frozen(efx->state)) {
1334 		rc = efx_enable_interrupts(efx);
1335 		if (rc)
1336 			goto fail;
1337 
1338 		mutex_lock(&efx->mac_lock);
1339 		efx_mcdi_port_reconfigure(efx);
1340 		mutex_unlock(&efx->mac_lock);
1341 
1342 		efx_start_all(efx);
1343 
1344 		efx_device_attach_if_not_resetting(efx);
1345 
1346 		efx->state = efx_thaw(efx->state);
1347 
1348 		efx->type->resume_wol(efx);
1349 	}
1350 
1351 	rtnl_unlock();
1352 
1353 	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
1354 	efx_queue_reset_work(efx);
1355 
1356 	return 0;
1357 
1358 fail:
1359 	rtnl_unlock();
1360 
1361 	return rc;
1362 }
1363 
1364 static int efx_pm_poweroff(struct device *dev)
1365 {
1366 	struct pci_dev *pci_dev = to_pci_dev(dev);
1367 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
1368 
1369 	efx->type->fini(efx);
1370 
1371 	efx->reset_pending = 0;
1372 
1373 	pci_save_state(pci_dev);
1374 	return pci_set_power_state(pci_dev, PCI_D3hot);
1375 }
1376 
1377 /* Used for both resume and restore */
1378 static int efx_pm_resume(struct device *dev)
1379 {
1380 	struct pci_dev *pci_dev = to_pci_dev(dev);
1381 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
1382 	int rc;
1383 
1384 	rc = pci_set_power_state(pci_dev, PCI_D0);
1385 	if (rc)
1386 		return rc;
1387 	pci_restore_state(pci_dev);
1388 	rc = pci_enable_device(pci_dev);
1389 	if (rc)
1390 		return rc;
1391 	pci_set_master(efx->pci_dev);
1392 	rc = efx->type->reset(efx, RESET_TYPE_ALL);
1393 	if (rc)
1394 		return rc;
1395 	down_write(&efx->filter_sem);
1396 	rc = efx->type->init(efx);
1397 	up_write(&efx->filter_sem);
1398 	if (rc)
1399 		return rc;
1400 	rc = efx_pm_thaw(dev);
1401 	return rc;
1402 }
1403 
1404 static int efx_pm_suspend(struct device *dev)
1405 {
1406 	int rc;
1407 
1408 	efx_pm_freeze(dev);
1409 	rc = efx_pm_poweroff(dev);
1410 	if (rc)
1411 		efx_pm_resume(dev);
1412 	return rc;
1413 }
1414 
1415 static const struct dev_pm_ops efx_pm_ops = {
1416 	.suspend	= efx_pm_suspend,
1417 	.resume		= efx_pm_resume,
1418 	.freeze		= efx_pm_freeze,
1419 	.thaw		= efx_pm_thaw,
1420 	.poweroff	= efx_pm_poweroff,
1421 	.restore	= efx_pm_resume,
1422 };
1423 
1424 static struct pci_driver efx_pci_driver = {
1425 	.name		= KBUILD_MODNAME,
1426 	.id_table	= efx_pci_table,
1427 	.probe		= efx_pci_probe,
1428 	.remove		= efx_pci_remove,
1429 	.driver.pm	= &efx_pm_ops,
1430 	.shutdown	= efx_pci_shutdown,
1431 	.err_handler	= &efx_err_handlers,
1432 #ifdef CONFIG_SFC_SRIOV
1433 	.sriov_configure = efx_pci_sriov_configure,
1434 #endif
1435 };
1436 
1437 /**************************************************************************
1438  *
1439  * Kernel module interface
1440  *
1441  *************************************************************************/
1442 
1443 static int __init efx_init_module(void)
1444 {
1445 	int rc;
1446 
1447 	printk(KERN_INFO "Solarflare NET driver\n");
1448 
1449 	rc = register_netdevice_notifier(&efx_netdev_notifier);
1450 	if (rc)
1451 		goto err_notifier;
1452 
1453 	rc = efx_create_reset_workqueue();
1454 	if (rc)
1455 		goto err_reset;
1456 
1457 	rc = pci_register_driver(&efx_pci_driver);
1458 	if (rc < 0)
1459 		goto err_pci;
1460 
1461 	rc = pci_register_driver(&ef100_pci_driver);
1462 	if (rc < 0)
1463 		goto err_pci_ef100;
1464 
1465 	return 0;
1466 
1467  err_pci_ef100:
1468 	pci_unregister_driver(&efx_pci_driver);
1469  err_pci:
1470 	efx_destroy_reset_workqueue();
1471  err_reset:
1472 	unregister_netdevice_notifier(&efx_netdev_notifier);
1473  err_notifier:
1474 	return rc;
1475 }
1476 
1477 static void __exit efx_exit_module(void)
1478 {
1479 	printk(KERN_INFO "Solarflare NET driver unloading\n");
1480 
1481 	pci_unregister_driver(&ef100_pci_driver);
1482 	pci_unregister_driver(&efx_pci_driver);
1483 	efx_destroy_reset_workqueue();
1484 	unregister_netdevice_notifier(&efx_netdev_notifier);
1485 
1486 }
1487 
1488 module_init(efx_init_module);
1489 module_exit(efx_exit_module);
1490 
1491 MODULE_AUTHOR("Solarflare Communications and "
1492 	      "Michael Brown <mbrown@fensystems.co.uk>");
1493 MODULE_DESCRIPTION("Solarflare network driver");
1494 MODULE_LICENSE("GPL");
1495 MODULE_DEVICE_TABLE(pci, efx_pci_table);
1496