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);
1383