xref: /openbmc/linux/drivers/net/ethernet/sfc/efx.c (revision 5ef12cb4a3a78ffb331c03a795a15eea4ae35155)
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 <net/gre.h>
27 #include <net/udp_tunnel.h>
28 #include "efx.h"
29 #include "nic.h"
30 #include "io.h"
31 #include "selftest.h"
32 #include "sriov.h"
33 
34 #include "mcdi.h"
35 #include "mcdi_pcol.h"
36 #include "workarounds.h"
37 
38 /**************************************************************************
39  *
40  * Type name strings
41  *
42  **************************************************************************
43  */
44 
45 /* Loopback mode names (see LOOPBACK_MODE()) */
46 const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
47 const char *const efx_loopback_mode_names[] = {
48 	[LOOPBACK_NONE]		= "NONE",
49 	[LOOPBACK_DATA]		= "DATAPATH",
50 	[LOOPBACK_GMAC]		= "GMAC",
51 	[LOOPBACK_XGMII]	= "XGMII",
52 	[LOOPBACK_XGXS]		= "XGXS",
53 	[LOOPBACK_XAUI]		= "XAUI",
54 	[LOOPBACK_GMII]		= "GMII",
55 	[LOOPBACK_SGMII]	= "SGMII",
56 	[LOOPBACK_XGBR]		= "XGBR",
57 	[LOOPBACK_XFI]		= "XFI",
58 	[LOOPBACK_XAUI_FAR]	= "XAUI_FAR",
59 	[LOOPBACK_GMII_FAR]	= "GMII_FAR",
60 	[LOOPBACK_SGMII_FAR]	= "SGMII_FAR",
61 	[LOOPBACK_XFI_FAR]	= "XFI_FAR",
62 	[LOOPBACK_GPHY]		= "GPHY",
63 	[LOOPBACK_PHYXS]	= "PHYXS",
64 	[LOOPBACK_PCS]		= "PCS",
65 	[LOOPBACK_PMAPMD]	= "PMA/PMD",
66 	[LOOPBACK_XPORT]	= "XPORT",
67 	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
68 	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
69 	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
70 	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
71 	[LOOPBACK_GMII_WS]	= "GMII_WS",
72 	[LOOPBACK_XFI_WS]	= "XFI_WS",
73 	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
74 	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
75 };
76 
77 const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
78 const char *const efx_reset_type_names[] = {
79 	[RESET_TYPE_INVISIBLE]          = "INVISIBLE",
80 	[RESET_TYPE_ALL]                = "ALL",
81 	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
82 	[RESET_TYPE_WORLD]              = "WORLD",
83 	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
84 	[RESET_TYPE_DATAPATH]           = "DATAPATH",
85 	[RESET_TYPE_MC_BIST]		= "MC_BIST",
86 	[RESET_TYPE_DISABLE]            = "DISABLE",
87 	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
88 	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
89 	[RESET_TYPE_DMA_ERROR]          = "DMA_ERROR",
90 	[RESET_TYPE_TX_SKIP]            = "TX_SKIP",
91 	[RESET_TYPE_MC_FAILURE]         = "MC_FAILURE",
92 	[RESET_TYPE_MCDI_TIMEOUT]	= "MCDI_TIMEOUT (FLR)",
93 };
94 
95 /* UDP tunnel type names */
96 static const char *const efx_udp_tunnel_type_names[] = {
97 	[TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN] = "vxlan",
98 	[TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE] = "geneve",
99 };
100 
101 void efx_get_udp_tunnel_type_name(u16 type, char *buf, size_t buflen)
102 {
103 	if (type < ARRAY_SIZE(efx_udp_tunnel_type_names) &&
104 	    efx_udp_tunnel_type_names[type] != NULL)
105 		snprintf(buf, buflen, "%s", efx_udp_tunnel_type_names[type]);
106 	else
107 		snprintf(buf, buflen, "type %d", type);
108 }
109 
110 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
111  * queued onto this work queue. This is not a per-nic work queue, because
112  * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
113  */
114 static struct workqueue_struct *reset_workqueue;
115 
116 /* How often and how many times to poll for a reset while waiting for a
117  * BIST that another function started to complete.
118  */
119 #define BIST_WAIT_DELAY_MS	100
120 #define BIST_WAIT_DELAY_COUNT	100
121 
122 /**************************************************************************
123  *
124  * Configurable values
125  *
126  *************************************************************************/
127 
128 /*
129  * Use separate channels for TX and RX events
130  *
131  * Set this to 1 to use separate channels for TX and RX. It allows us
132  * to control interrupt affinity separately for TX and RX.
133  *
134  * This is only used in MSI-X interrupt mode
135  */
136 bool efx_separate_tx_channels;
137 module_param(efx_separate_tx_channels, bool, 0444);
138 MODULE_PARM_DESC(efx_separate_tx_channels,
139 		 "Use separate channels for TX and RX");
140 
141 /* This is the weight assigned to each of the (per-channel) virtual
142  * NAPI devices.
143  */
144 static int napi_weight = 64;
145 
146 /* This is the time (in jiffies) between invocations of the hardware
147  * monitor.
148  * On Falcon-based NICs, this will:
149  * - Check the on-board hardware monitor;
150  * - Poll the link state and reconfigure the hardware as necessary.
151  * On Siena-based NICs for power systems with EEH support, this will give EEH a
152  * chance to start.
153  */
154 static unsigned int efx_monitor_interval = 1 * HZ;
155 
156 /* Initial interrupt moderation settings.  They can be modified after
157  * module load with ethtool.
158  *
159  * The default for RX should strike a balance between increasing the
160  * round-trip latency and reducing overhead.
161  */
162 static unsigned int rx_irq_mod_usec = 60;
163 
164 /* Initial interrupt moderation settings.  They can be modified after
165  * module load with ethtool.
166  *
167  * This default is chosen to ensure that a 10G link does not go idle
168  * while a TX queue is stopped after it has become full.  A queue is
169  * restarted when it drops below half full.  The time this takes (assuming
170  * worst case 3 descriptors per packet and 1024 descriptors) is
171  *   512 / 3 * 1.2 = 205 usec.
172  */
173 static unsigned int tx_irq_mod_usec = 150;
174 
175 /* This is the first interrupt mode to try out of:
176  * 0 => MSI-X
177  * 1 => MSI
178  * 2 => legacy
179  */
180 static unsigned int interrupt_mode;
181 
182 /* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
183  * i.e. the number of CPUs among which we may distribute simultaneous
184  * interrupt handling.
185  *
186  * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
187  * The default (0) means to assign an interrupt to each core.
188  */
189 static unsigned int rss_cpus;
190 module_param(rss_cpus, uint, 0444);
191 MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");
192 
193 static bool phy_flash_cfg;
194 module_param(phy_flash_cfg, bool, 0644);
195 MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");
196 
197 static unsigned irq_adapt_low_thresh = 8000;
198 module_param(irq_adapt_low_thresh, uint, 0644);
199 MODULE_PARM_DESC(irq_adapt_low_thresh,
200 		 "Threshold score for reducing IRQ moderation");
201 
202 static unsigned irq_adapt_high_thresh = 16000;
203 module_param(irq_adapt_high_thresh, uint, 0644);
204 MODULE_PARM_DESC(irq_adapt_high_thresh,
205 		 "Threshold score for increasing IRQ moderation");
206 
207 static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
208 			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
209 			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
210 			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
211 module_param(debug, uint, 0);
212 MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
213 
214 /**************************************************************************
215  *
216  * Utility functions and prototypes
217  *
218  *************************************************************************/
219 
220 static int efx_soft_enable_interrupts(struct efx_nic *efx);
221 static void efx_soft_disable_interrupts(struct efx_nic *efx);
222 static void efx_remove_channel(struct efx_channel *channel);
223 static void efx_remove_channels(struct efx_nic *efx);
224 static const struct efx_channel_type efx_default_channel_type;
225 static void efx_remove_port(struct efx_nic *efx);
226 static void efx_init_napi_channel(struct efx_channel *channel);
227 static void efx_fini_napi(struct efx_nic *efx);
228 static void efx_fini_napi_channel(struct efx_channel *channel);
229 static void efx_fini_struct(struct efx_nic *efx);
230 static void efx_start_all(struct efx_nic *efx);
231 static void efx_stop_all(struct efx_nic *efx);
232 
233 #define EFX_ASSERT_RESET_SERIALISED(efx)		\
234 	do {						\
235 		if ((efx->state == STATE_READY) ||	\
236 		    (efx->state == STATE_RECOVERY) ||	\
237 		    (efx->state == STATE_DISABLED))	\
238 			ASSERT_RTNL();			\
239 	} while (0)
240 
241 static int efx_check_disabled(struct efx_nic *efx)
242 {
243 	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
244 		netif_err(efx, drv, efx->net_dev,
245 			  "device is disabled due to earlier errors\n");
246 		return -EIO;
247 	}
248 	return 0;
249 }
250 
251 /**************************************************************************
252  *
253  * Event queue processing
254  *
255  *************************************************************************/
256 
257 /* Process channel's event queue
258  *
259  * This function is responsible for processing the event queue of a
260  * single channel.  The caller must guarantee that this function will
261  * never be concurrently called more than once on the same channel,
262  * though different channels may be being processed concurrently.
263  */
264 static int efx_process_channel(struct efx_channel *channel, int budget)
265 {
266 	struct efx_tx_queue *tx_queue;
267 	int spent;
268 
269 	if (unlikely(!channel->enabled))
270 		return 0;
271 
272 	efx_for_each_channel_tx_queue(tx_queue, channel) {
273 		tx_queue->pkts_compl = 0;
274 		tx_queue->bytes_compl = 0;
275 	}
276 
277 	spent = efx_nic_process_eventq(channel, budget);
278 	if (spent && efx_channel_has_rx_queue(channel)) {
279 		struct efx_rx_queue *rx_queue =
280 			efx_channel_get_rx_queue(channel);
281 
282 		efx_rx_flush_packet(channel);
283 		efx_fast_push_rx_descriptors(rx_queue, true);
284 	}
285 
286 	/* Update BQL */
287 	efx_for_each_channel_tx_queue(tx_queue, channel) {
288 		if (tx_queue->bytes_compl) {
289 			netdev_tx_completed_queue(tx_queue->core_txq,
290 				tx_queue->pkts_compl, tx_queue->bytes_compl);
291 		}
292 	}
293 
294 	return spent;
295 }
296 
297 /* NAPI poll handler
298  *
299  * NAPI guarantees serialisation of polls of the same device, which
300  * provides the guarantee required by efx_process_channel().
301  */
302 static void efx_update_irq_mod(struct efx_nic *efx, struct efx_channel *channel)
303 {
304 	int step = efx->irq_mod_step_us;
305 
306 	if (channel->irq_mod_score < irq_adapt_low_thresh) {
307 		if (channel->irq_moderation_us > step) {
308 			channel->irq_moderation_us -= step;
309 			efx->type->push_irq_moderation(channel);
310 		}
311 	} else if (channel->irq_mod_score > irq_adapt_high_thresh) {
312 		if (channel->irq_moderation_us <
313 		    efx->irq_rx_moderation_us) {
314 			channel->irq_moderation_us += step;
315 			efx->type->push_irq_moderation(channel);
316 		}
317 	}
318 
319 	channel->irq_count = 0;
320 	channel->irq_mod_score = 0;
321 }
322 
323 static int efx_poll(struct napi_struct *napi, int budget)
324 {
325 	struct efx_channel *channel =
326 		container_of(napi, struct efx_channel, napi_str);
327 	struct efx_nic *efx = channel->efx;
328 	int spent;
329 
330 	netif_vdbg(efx, intr, efx->net_dev,
331 		   "channel %d NAPI poll executing on CPU %d\n",
332 		   channel->channel, raw_smp_processor_id());
333 
334 	spent = efx_process_channel(channel, budget);
335 
336 	if (spent < budget) {
337 		if (efx_channel_has_rx_queue(channel) &&
338 		    efx->irq_rx_adaptive &&
339 		    unlikely(++channel->irq_count == 1000)) {
340 			efx_update_irq_mod(efx, channel);
341 		}
342 
343 #ifdef CONFIG_RFS_ACCEL
344 		/* Perhaps expire some ARFS filters */
345 		schedule_work(&channel->filter_work);
346 #endif
347 
348 		/* There is no race here; although napi_disable() will
349 		 * only wait for napi_complete(), this isn't a problem
350 		 * since efx_nic_eventq_read_ack() will have no effect if
351 		 * interrupts have already been disabled.
352 		 */
353 		if (napi_complete_done(napi, spent))
354 			efx_nic_eventq_read_ack(channel);
355 	}
356 
357 	return spent;
358 }
359 
360 /* Create event queue
361  * Event queue memory allocations are done only once.  If the channel
362  * is reset, the memory buffer will be reused; this guards against
363  * errors during channel reset and also simplifies interrupt handling.
364  */
365 static int efx_probe_eventq(struct efx_channel *channel)
366 {
367 	struct efx_nic *efx = channel->efx;
368 	unsigned long entries;
369 
370 	netif_dbg(efx, probe, efx->net_dev,
371 		  "chan %d create event queue\n", channel->channel);
372 
373 	/* Build an event queue with room for one event per tx and rx buffer,
374 	 * plus some extra for link state events and MCDI completions. */
375 	entries = roundup_pow_of_two(efx->rxq_entries + efx->txq_entries + 128);
376 	EFX_WARN_ON_PARANOID(entries > EFX_MAX_EVQ_SIZE);
377 	channel->eventq_mask = max(entries, EFX_MIN_EVQ_SIZE) - 1;
378 
379 	return efx_nic_probe_eventq(channel);
380 }
381 
382 /* Prepare channel's event queue */
383 static int efx_init_eventq(struct efx_channel *channel)
384 {
385 	struct efx_nic *efx = channel->efx;
386 	int rc;
387 
388 	EFX_WARN_ON_PARANOID(channel->eventq_init);
389 
390 	netif_dbg(efx, drv, efx->net_dev,
391 		  "chan %d init event queue\n", channel->channel);
392 
393 	rc = efx_nic_init_eventq(channel);
394 	if (rc == 0) {
395 		efx->type->push_irq_moderation(channel);
396 		channel->eventq_read_ptr = 0;
397 		channel->eventq_init = true;
398 	}
399 	return rc;
400 }
401 
402 /* Enable event queue processing and NAPI */
403 void efx_start_eventq(struct efx_channel *channel)
404 {
405 	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
406 		  "chan %d start event queue\n", channel->channel);
407 
408 	/* Make sure the NAPI handler sees the enabled flag set */
409 	channel->enabled = true;
410 	smp_wmb();
411 
412 	napi_enable(&channel->napi_str);
413 	efx_nic_eventq_read_ack(channel);
414 }
415 
416 /* Disable event queue processing and NAPI */
417 void efx_stop_eventq(struct efx_channel *channel)
418 {
419 	if (!channel->enabled)
420 		return;
421 
422 	napi_disable(&channel->napi_str);
423 	channel->enabled = false;
424 }
425 
426 static void efx_fini_eventq(struct efx_channel *channel)
427 {
428 	if (!channel->eventq_init)
429 		return;
430 
431 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
432 		  "chan %d fini event queue\n", channel->channel);
433 
434 	efx_nic_fini_eventq(channel);
435 	channel->eventq_init = false;
436 }
437 
438 static void efx_remove_eventq(struct efx_channel *channel)
439 {
440 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
441 		  "chan %d remove event queue\n", channel->channel);
442 
443 	efx_nic_remove_eventq(channel);
444 }
445 
446 /**************************************************************************
447  *
448  * Channel handling
449  *
450  *************************************************************************/
451 
452 /* Allocate and initialise a channel structure. */
453 static struct efx_channel *
454 efx_alloc_channel(struct efx_nic *efx, int i, struct efx_channel *old_channel)
455 {
456 	struct efx_channel *channel;
457 	struct efx_rx_queue *rx_queue;
458 	struct efx_tx_queue *tx_queue;
459 	int j;
460 
461 	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
462 	if (!channel)
463 		return NULL;
464 
465 	channel->efx = efx;
466 	channel->channel = i;
467 	channel->type = &efx_default_channel_type;
468 
469 	for (j = 0; j < EFX_TXQ_TYPES; j++) {
470 		tx_queue = &channel->tx_queue[j];
471 		tx_queue->efx = efx;
472 		tx_queue->queue = i * EFX_TXQ_TYPES + j;
473 		tx_queue->channel = channel;
474 	}
475 
476 #ifdef CONFIG_RFS_ACCEL
477 	INIT_WORK(&channel->filter_work, efx_filter_rfs_expire);
478 #endif
479 
480 	rx_queue = &channel->rx_queue;
481 	rx_queue->efx = efx;
482 	timer_setup(&rx_queue->slow_fill, efx_rx_slow_fill, 0);
483 
484 	return channel;
485 }
486 
487 /* Allocate and initialise a channel structure, copying parameters
488  * (but not resources) from an old channel structure.
489  */
490 static struct efx_channel *
491 efx_copy_channel(const struct efx_channel *old_channel)
492 {
493 	struct efx_channel *channel;
494 	struct efx_rx_queue *rx_queue;
495 	struct efx_tx_queue *tx_queue;
496 	int j;
497 
498 	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
499 	if (!channel)
500 		return NULL;
501 
502 	*channel = *old_channel;
503 
504 	channel->napi_dev = NULL;
505 	INIT_HLIST_NODE(&channel->napi_str.napi_hash_node);
506 	channel->napi_str.napi_id = 0;
507 	channel->napi_str.state = 0;
508 	memset(&channel->eventq, 0, sizeof(channel->eventq));
509 
510 	for (j = 0; j < EFX_TXQ_TYPES; j++) {
511 		tx_queue = &channel->tx_queue[j];
512 		if (tx_queue->channel)
513 			tx_queue->channel = channel;
514 		tx_queue->buffer = NULL;
515 		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
516 	}
517 
518 	rx_queue = &channel->rx_queue;
519 	rx_queue->buffer = NULL;
520 	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
521 	timer_setup(&rx_queue->slow_fill, efx_rx_slow_fill, 0);
522 #ifdef CONFIG_RFS_ACCEL
523 	INIT_WORK(&channel->filter_work, efx_filter_rfs_expire);
524 #endif
525 
526 	return channel;
527 }
528 
529 static int efx_probe_channel(struct efx_channel *channel)
530 {
531 	struct efx_tx_queue *tx_queue;
532 	struct efx_rx_queue *rx_queue;
533 	int rc;
534 
535 	netif_dbg(channel->efx, probe, channel->efx->net_dev,
536 		  "creating channel %d\n", channel->channel);
537 
538 	rc = channel->type->pre_probe(channel);
539 	if (rc)
540 		goto fail;
541 
542 	rc = efx_probe_eventq(channel);
543 	if (rc)
544 		goto fail;
545 
546 	efx_for_each_channel_tx_queue(tx_queue, channel) {
547 		rc = efx_probe_tx_queue(tx_queue);
548 		if (rc)
549 			goto fail;
550 	}
551 
552 	efx_for_each_channel_rx_queue(rx_queue, channel) {
553 		rc = efx_probe_rx_queue(rx_queue);
554 		if (rc)
555 			goto fail;
556 	}
557 
558 	return 0;
559 
560 fail:
561 	efx_remove_channel(channel);
562 	return rc;
563 }
564 
565 static void
566 efx_get_channel_name(struct efx_channel *channel, char *buf, size_t len)
567 {
568 	struct efx_nic *efx = channel->efx;
569 	const char *type;
570 	int number;
571 
572 	number = channel->channel;
573 	if (efx->tx_channel_offset == 0) {
574 		type = "";
575 	} else if (channel->channel < efx->tx_channel_offset) {
576 		type = "-rx";
577 	} else {
578 		type = "-tx";
579 		number -= efx->tx_channel_offset;
580 	}
581 	snprintf(buf, len, "%s%s-%d", efx->name, type, number);
582 }
583 
584 static void efx_set_channel_names(struct efx_nic *efx)
585 {
586 	struct efx_channel *channel;
587 
588 	efx_for_each_channel(channel, efx)
589 		channel->type->get_name(channel,
590 					efx->msi_context[channel->channel].name,
591 					sizeof(efx->msi_context[0].name));
592 }
593 
594 static int efx_probe_channels(struct efx_nic *efx)
595 {
596 	struct efx_channel *channel;
597 	int rc;
598 
599 	/* Restart special buffer allocation */
600 	efx->next_buffer_table = 0;
601 
602 	/* Probe channels in reverse, so that any 'extra' channels
603 	 * use the start of the buffer table. This allows the traffic
604 	 * channels to be resized without moving them or wasting the
605 	 * entries before them.
606 	 */
607 	efx_for_each_channel_rev(channel, efx) {
608 		rc = efx_probe_channel(channel);
609 		if (rc) {
610 			netif_err(efx, probe, efx->net_dev,
611 				  "failed to create channel %d\n",
612 				  channel->channel);
613 			goto fail;
614 		}
615 	}
616 	efx_set_channel_names(efx);
617 
618 	return 0;
619 
620 fail:
621 	efx_remove_channels(efx);
622 	return rc;
623 }
624 
625 /* Channels are shutdown and reinitialised whilst the NIC is running
626  * to propagate configuration changes (mtu, checksum offload), or
627  * to clear hardware error conditions
628  */
629 static void efx_start_datapath(struct efx_nic *efx)
630 {
631 	netdev_features_t old_features = efx->net_dev->features;
632 	bool old_rx_scatter = efx->rx_scatter;
633 	struct efx_tx_queue *tx_queue;
634 	struct efx_rx_queue *rx_queue;
635 	struct efx_channel *channel;
636 	size_t rx_buf_len;
637 
638 	/* Calculate the rx buffer allocation parameters required to
639 	 * support the current MTU, including padding for header
640 	 * alignment and overruns.
641 	 */
642 	efx->rx_dma_len = (efx->rx_prefix_size +
643 			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
644 			   efx->type->rx_buffer_padding);
645 	rx_buf_len = (sizeof(struct efx_rx_page_state) +
646 		      efx->rx_ip_align + efx->rx_dma_len);
647 	if (rx_buf_len <= PAGE_SIZE) {
648 		efx->rx_scatter = efx->type->always_rx_scatter;
649 		efx->rx_buffer_order = 0;
650 	} else if (efx->type->can_rx_scatter) {
651 		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
652 		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
653 			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
654 				       EFX_RX_BUF_ALIGNMENT) >
655 			     PAGE_SIZE);
656 		efx->rx_scatter = true;
657 		efx->rx_dma_len = EFX_RX_USR_BUF_SIZE;
658 		efx->rx_buffer_order = 0;
659 	} else {
660 		efx->rx_scatter = false;
661 		efx->rx_buffer_order = get_order(rx_buf_len);
662 	}
663 
664 	efx_rx_config_page_split(efx);
665 	if (efx->rx_buffer_order)
666 		netif_dbg(efx, drv, efx->net_dev,
667 			  "RX buf len=%u; page order=%u batch=%u\n",
668 			  efx->rx_dma_len, efx->rx_buffer_order,
669 			  efx->rx_pages_per_batch);
670 	else
671 		netif_dbg(efx, drv, efx->net_dev,
672 			  "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
673 			  efx->rx_dma_len, efx->rx_page_buf_step,
674 			  efx->rx_bufs_per_page, efx->rx_pages_per_batch);
675 
676 	/* Restore previously fixed features in hw_features and remove
677 	 * features which are fixed now
678 	 */
679 	efx->net_dev->hw_features |= efx->net_dev->features;
680 	efx->net_dev->hw_features &= ~efx->fixed_features;
681 	efx->net_dev->features |= efx->fixed_features;
682 	if (efx->net_dev->features != old_features)
683 		netdev_features_change(efx->net_dev);
684 
685 	/* RX filters may also have scatter-enabled flags */
686 	if (efx->rx_scatter != old_rx_scatter)
687 		efx->type->filter_update_rx_scatter(efx);
688 
689 	/* We must keep at least one descriptor in a TX ring empty.
690 	 * We could avoid this when the queue size does not exactly
691 	 * match the hardware ring size, but it's not that important.
692 	 * Therefore we stop the queue when one more skb might fill
693 	 * the ring completely.  We wake it when half way back to
694 	 * empty.
695 	 */
696 	efx->txq_stop_thresh = efx->txq_entries - efx_tx_max_skb_descs(efx);
697 	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;
698 
699 	/* Initialise the channels */
700 	efx_for_each_channel(channel, efx) {
701 		efx_for_each_channel_tx_queue(tx_queue, channel) {
702 			efx_init_tx_queue(tx_queue);
703 			atomic_inc(&efx->active_queues);
704 		}
705 
706 		efx_for_each_channel_rx_queue(rx_queue, channel) {
707 			efx_init_rx_queue(rx_queue);
708 			atomic_inc(&efx->active_queues);
709 			efx_stop_eventq(channel);
710 			efx_fast_push_rx_descriptors(rx_queue, false);
711 			efx_start_eventq(channel);
712 		}
713 
714 		WARN_ON(channel->rx_pkt_n_frags);
715 	}
716 
717 	efx_ptp_start_datapath(efx);
718 
719 	if (netif_device_present(efx->net_dev))
720 		netif_tx_wake_all_queues(efx->net_dev);
721 }
722 
723 static void efx_stop_datapath(struct efx_nic *efx)
724 {
725 	struct efx_channel *channel;
726 	struct efx_tx_queue *tx_queue;
727 	struct efx_rx_queue *rx_queue;
728 	int rc;
729 
730 	EFX_ASSERT_RESET_SERIALISED(efx);
731 	BUG_ON(efx->port_enabled);
732 
733 	efx_ptp_stop_datapath(efx);
734 
735 	/* Stop RX refill */
736 	efx_for_each_channel(channel, efx) {
737 		efx_for_each_channel_rx_queue(rx_queue, channel)
738 			rx_queue->refill_enabled = false;
739 	}
740 
741 	efx_for_each_channel(channel, efx) {
742 		/* RX packet processing is pipelined, so wait for the
743 		 * NAPI handler to complete.  At least event queue 0
744 		 * might be kept active by non-data events, so don't
745 		 * use napi_synchronize() but actually disable NAPI
746 		 * temporarily.
747 		 */
748 		if (efx_channel_has_rx_queue(channel)) {
749 			efx_stop_eventq(channel);
750 			efx_start_eventq(channel);
751 		}
752 	}
753 
754 	rc = efx->type->fini_dmaq(efx);
755 	if (rc) {
756 		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
757 	} else {
758 		netif_dbg(efx, drv, efx->net_dev,
759 			  "successfully flushed all queues\n");
760 	}
761 
762 	efx_for_each_channel(channel, efx) {
763 		efx_for_each_channel_rx_queue(rx_queue, channel)
764 			efx_fini_rx_queue(rx_queue);
765 		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
766 			efx_fini_tx_queue(tx_queue);
767 	}
768 }
769 
770 static void efx_remove_channel(struct efx_channel *channel)
771 {
772 	struct efx_tx_queue *tx_queue;
773 	struct efx_rx_queue *rx_queue;
774 
775 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
776 		  "destroy chan %d\n", channel->channel);
777 
778 	efx_for_each_channel_rx_queue(rx_queue, channel)
779 		efx_remove_rx_queue(rx_queue);
780 	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
781 		efx_remove_tx_queue(tx_queue);
782 	efx_remove_eventq(channel);
783 	channel->type->post_remove(channel);
784 }
785 
786 static void efx_remove_channels(struct efx_nic *efx)
787 {
788 	struct efx_channel *channel;
789 
790 	efx_for_each_channel(channel, efx)
791 		efx_remove_channel(channel);
792 }
793 
794 int
795 efx_realloc_channels(struct efx_nic *efx, u32 rxq_entries, u32 txq_entries)
796 {
797 	struct efx_channel *other_channel[EFX_MAX_CHANNELS], *channel;
798 	u32 old_rxq_entries, old_txq_entries;
799 	unsigned i, next_buffer_table = 0;
800 	int rc, rc2;
801 
802 	rc = efx_check_disabled(efx);
803 	if (rc)
804 		return rc;
805 
806 	/* Not all channels should be reallocated. We must avoid
807 	 * reallocating their buffer table entries.
808 	 */
809 	efx_for_each_channel(channel, efx) {
810 		struct efx_rx_queue *rx_queue;
811 		struct efx_tx_queue *tx_queue;
812 
813 		if (channel->type->copy)
814 			continue;
815 		next_buffer_table = max(next_buffer_table,
816 					channel->eventq.index +
817 					channel->eventq.entries);
818 		efx_for_each_channel_rx_queue(rx_queue, channel)
819 			next_buffer_table = max(next_buffer_table,
820 						rx_queue->rxd.index +
821 						rx_queue->rxd.entries);
822 		efx_for_each_channel_tx_queue(tx_queue, channel)
823 			next_buffer_table = max(next_buffer_table,
824 						tx_queue->txd.index +
825 						tx_queue->txd.entries);
826 	}
827 
828 	efx_device_detach_sync(efx);
829 	efx_stop_all(efx);
830 	efx_soft_disable_interrupts(efx);
831 
832 	/* Clone channels (where possible) */
833 	memset(other_channel, 0, sizeof(other_channel));
834 	for (i = 0; i < efx->n_channels; i++) {
835 		channel = efx->channel[i];
836 		if (channel->type->copy)
837 			channel = channel->type->copy(channel);
838 		if (!channel) {
839 			rc = -ENOMEM;
840 			goto out;
841 		}
842 		other_channel[i] = channel;
843 	}
844 
845 	/* Swap entry counts and channel pointers */
846 	old_rxq_entries = efx->rxq_entries;
847 	old_txq_entries = efx->txq_entries;
848 	efx->rxq_entries = rxq_entries;
849 	efx->txq_entries = txq_entries;
850 	for (i = 0; i < efx->n_channels; i++) {
851 		channel = efx->channel[i];
852 		efx->channel[i] = other_channel[i];
853 		other_channel[i] = channel;
854 	}
855 
856 	/* Restart buffer table allocation */
857 	efx->next_buffer_table = next_buffer_table;
858 
859 	for (i = 0; i < efx->n_channels; i++) {
860 		channel = efx->channel[i];
861 		if (!channel->type->copy)
862 			continue;
863 		rc = efx_probe_channel(channel);
864 		if (rc)
865 			goto rollback;
866 		efx_init_napi_channel(efx->channel[i]);
867 	}
868 
869 out:
870 	/* Destroy unused channel structures */
871 	for (i = 0; i < efx->n_channels; i++) {
872 		channel = other_channel[i];
873 		if (channel && channel->type->copy) {
874 			efx_fini_napi_channel(channel);
875 			efx_remove_channel(channel);
876 			kfree(channel);
877 		}
878 	}
879 
880 	rc2 = efx_soft_enable_interrupts(efx);
881 	if (rc2) {
882 		rc = rc ? rc : rc2;
883 		netif_err(efx, drv, efx->net_dev,
884 			  "unable to restart interrupts on channel reallocation\n");
885 		efx_schedule_reset(efx, RESET_TYPE_DISABLE);
886 	} else {
887 		efx_start_all(efx);
888 		efx_device_attach_if_not_resetting(efx);
889 	}
890 	return rc;
891 
892 rollback:
893 	/* Swap back */
894 	efx->rxq_entries = old_rxq_entries;
895 	efx->txq_entries = old_txq_entries;
896 	for (i = 0; i < efx->n_channels; i++) {
897 		channel = efx->channel[i];
898 		efx->channel[i] = other_channel[i];
899 		other_channel[i] = channel;
900 	}
901 	goto out;
902 }
903 
904 void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
905 {
906 	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
907 }
908 
909 static bool efx_default_channel_want_txqs(struct efx_channel *channel)
910 {
911 	return channel->channel - channel->efx->tx_channel_offset <
912 		channel->efx->n_tx_channels;
913 }
914 
915 static const struct efx_channel_type efx_default_channel_type = {
916 	.pre_probe		= efx_channel_dummy_op_int,
917 	.post_remove		= efx_channel_dummy_op_void,
918 	.get_name		= efx_get_channel_name,
919 	.copy			= efx_copy_channel,
920 	.want_txqs		= efx_default_channel_want_txqs,
921 	.keep_eventq		= false,
922 	.want_pio		= true,
923 };
924 
925 int efx_channel_dummy_op_int(struct efx_channel *channel)
926 {
927 	return 0;
928 }
929 
930 void efx_channel_dummy_op_void(struct efx_channel *channel)
931 {
932 }
933 
934 /**************************************************************************
935  *
936  * Port handling
937  *
938  **************************************************************************/
939 
940 /* This ensures that the kernel is kept informed (via
941  * netif_carrier_on/off) of the link status, and also maintains the
942  * link status's stop on the port's TX queue.
943  */
944 void efx_link_status_changed(struct efx_nic *efx)
945 {
946 	struct efx_link_state *link_state = &efx->link_state;
947 
948 	/* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
949 	 * that no events are triggered between unregister_netdev() and the
950 	 * driver unloading. A more general condition is that NETDEV_CHANGE
951 	 * can only be generated between NETDEV_UP and NETDEV_DOWN */
952 	if (!netif_running(efx->net_dev))
953 		return;
954 
955 	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
956 		efx->n_link_state_changes++;
957 
958 		if (link_state->up)
959 			netif_carrier_on(efx->net_dev);
960 		else
961 			netif_carrier_off(efx->net_dev);
962 	}
963 
964 	/* Status message for kernel log */
965 	if (link_state->up)
966 		netif_info(efx, link, efx->net_dev,
967 			   "link up at %uMbps %s-duplex (MTU %d)\n",
968 			   link_state->speed, link_state->fd ? "full" : "half",
969 			   efx->net_dev->mtu);
970 	else
971 		netif_info(efx, link, efx->net_dev, "link down\n");
972 }
973 
974 void efx_link_set_advertising(struct efx_nic *efx,
975 			      const unsigned long *advertising)
976 {
977 	memcpy(efx->link_advertising, advertising,
978 	       sizeof(__ETHTOOL_DECLARE_LINK_MODE_MASK()));
979 
980 	efx->link_advertising[0] |= ADVERTISED_Autoneg;
981 	if (advertising[0] & ADVERTISED_Pause)
982 		efx->wanted_fc |= (EFX_FC_TX | EFX_FC_RX);
983 	else
984 		efx->wanted_fc &= ~(EFX_FC_TX | EFX_FC_RX);
985 	if (advertising[0] & ADVERTISED_Asym_Pause)
986 		efx->wanted_fc ^= EFX_FC_TX;
987 }
988 
989 /* Equivalent to efx_link_set_advertising with all-zeroes, except does not
990  * force the Autoneg bit on.
991  */
992 void efx_link_clear_advertising(struct efx_nic *efx)
993 {
994 	bitmap_zero(efx->link_advertising, __ETHTOOL_LINK_MODE_MASK_NBITS);
995 	efx->wanted_fc &= ~(EFX_FC_TX | EFX_FC_RX);
996 }
997 
998 void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
999 {
1000 	efx->wanted_fc = wanted_fc;
1001 	if (efx->link_advertising[0]) {
1002 		if (wanted_fc & EFX_FC_RX)
1003 			efx->link_advertising[0] |= (ADVERTISED_Pause |
1004 						     ADVERTISED_Asym_Pause);
1005 		else
1006 			efx->link_advertising[0] &= ~(ADVERTISED_Pause |
1007 						      ADVERTISED_Asym_Pause);
1008 		if (wanted_fc & EFX_FC_TX)
1009 			efx->link_advertising[0] ^= ADVERTISED_Asym_Pause;
1010 	}
1011 }
1012 
1013 static void efx_fini_port(struct efx_nic *efx);
1014 
1015 /* We assume that efx->type->reconfigure_mac will always try to sync RX
1016  * filters and therefore needs to read-lock the filter table against freeing
1017  */
1018 void efx_mac_reconfigure(struct efx_nic *efx)
1019 {
1020 	down_read(&efx->filter_sem);
1021 	efx->type->reconfigure_mac(efx);
1022 	up_read(&efx->filter_sem);
1023 }
1024 
1025 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
1026  * the MAC appropriately. All other PHY configuration changes are pushed
1027  * through phy_op->set_settings(), and pushed asynchronously to the MAC
1028  * through efx_monitor().
1029  *
1030  * Callers must hold the mac_lock
1031  */
1032 int __efx_reconfigure_port(struct efx_nic *efx)
1033 {
1034 	enum efx_phy_mode phy_mode;
1035 	int rc;
1036 
1037 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
1038 
1039 	/* Disable PHY transmit in mac level loopbacks */
1040 	phy_mode = efx->phy_mode;
1041 	if (LOOPBACK_INTERNAL(efx))
1042 		efx->phy_mode |= PHY_MODE_TX_DISABLED;
1043 	else
1044 		efx->phy_mode &= ~PHY_MODE_TX_DISABLED;
1045 
1046 	rc = efx->type->reconfigure_port(efx);
1047 
1048 	if (rc)
1049 		efx->phy_mode = phy_mode;
1050 
1051 	return rc;
1052 }
1053 
1054 /* Reinitialise the MAC to pick up new PHY settings, even if the port is
1055  * disabled. */
1056 int efx_reconfigure_port(struct efx_nic *efx)
1057 {
1058 	int rc;
1059 
1060 	EFX_ASSERT_RESET_SERIALISED(efx);
1061 
1062 	mutex_lock(&efx->mac_lock);
1063 	rc = __efx_reconfigure_port(efx);
1064 	mutex_unlock(&efx->mac_lock);
1065 
1066 	return rc;
1067 }
1068 
1069 /* Asynchronous work item for changing MAC promiscuity and multicast
1070  * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
1071  * MAC directly. */
1072 static void efx_mac_work(struct work_struct *data)
1073 {
1074 	struct efx_nic *efx = container_of(data, struct efx_nic, mac_work);
1075 
1076 	mutex_lock(&efx->mac_lock);
1077 	if (efx->port_enabled)
1078 		efx_mac_reconfigure(efx);
1079 	mutex_unlock(&efx->mac_lock);
1080 }
1081 
1082 static int efx_probe_port(struct efx_nic *efx)
1083 {
1084 	int rc;
1085 
1086 	netif_dbg(efx, probe, efx->net_dev, "create port\n");
1087 
1088 	if (phy_flash_cfg)
1089 		efx->phy_mode = PHY_MODE_SPECIAL;
1090 
1091 	/* Connect up MAC/PHY operations table */
1092 	rc = efx->type->probe_port(efx);
1093 	if (rc)
1094 		return rc;
1095 
1096 	/* Initialise MAC address to permanent address */
1097 	ether_addr_copy(efx->net_dev->dev_addr, efx->net_dev->perm_addr);
1098 
1099 	return 0;
1100 }
1101 
1102 static int efx_init_port(struct efx_nic *efx)
1103 {
1104 	int rc;
1105 
1106 	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1107 
1108 	mutex_lock(&efx->mac_lock);
1109 
1110 	rc = efx->phy_op->init(efx);
1111 	if (rc)
1112 		goto fail1;
1113 
1114 	efx->port_initialized = true;
1115 
1116 	/* Reconfigure the MAC before creating dma queues (required for
1117 	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1118 	efx_mac_reconfigure(efx);
1119 
1120 	/* Ensure the PHY advertises the correct flow control settings */
1121 	rc = efx->phy_op->reconfigure(efx);
1122 	if (rc && rc != -EPERM)
1123 		goto fail2;
1124 
1125 	mutex_unlock(&efx->mac_lock);
1126 	return 0;
1127 
1128 fail2:
1129 	efx->phy_op->fini(efx);
1130 fail1:
1131 	mutex_unlock(&efx->mac_lock);
1132 	return rc;
1133 }
1134 
1135 static void efx_start_port(struct efx_nic *efx)
1136 {
1137 	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1138 	BUG_ON(efx->port_enabled);
1139 
1140 	mutex_lock(&efx->mac_lock);
1141 	efx->port_enabled = true;
1142 
1143 	/* Ensure MAC ingress/egress is enabled */
1144 	efx_mac_reconfigure(efx);
1145 
1146 	mutex_unlock(&efx->mac_lock);
1147 }
1148 
1149 /* Cancel work for MAC reconfiguration, periodic hardware monitoring
1150  * and the async self-test, wait for them to finish and prevent them
1151  * being scheduled again.  This doesn't cover online resets, which
1152  * should only be cancelled when removing the device.
1153  */
1154 static void efx_stop_port(struct efx_nic *efx)
1155 {
1156 	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1157 
1158 	EFX_ASSERT_RESET_SERIALISED(efx);
1159 
1160 	mutex_lock(&efx->mac_lock);
1161 	efx->port_enabled = false;
1162 	mutex_unlock(&efx->mac_lock);
1163 
1164 	/* Serialise against efx_set_multicast_list() */
1165 	netif_addr_lock_bh(efx->net_dev);
1166 	netif_addr_unlock_bh(efx->net_dev);
1167 
1168 	cancel_delayed_work_sync(&efx->monitor_work);
1169 	efx_selftest_async_cancel(efx);
1170 	cancel_work_sync(&efx->mac_work);
1171 }
1172 
1173 static void efx_fini_port(struct efx_nic *efx)
1174 {
1175 	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1176 
1177 	if (!efx->port_initialized)
1178 		return;
1179 
1180 	efx->phy_op->fini(efx);
1181 	efx->port_initialized = false;
1182 
1183 	efx->link_state.up = false;
1184 	efx_link_status_changed(efx);
1185 }
1186 
1187 static void efx_remove_port(struct efx_nic *efx)
1188 {
1189 	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1190 
1191 	efx->type->remove_port(efx);
1192 }
1193 
1194 /**************************************************************************
1195  *
1196  * NIC handling
1197  *
1198  **************************************************************************/
1199 
1200 static LIST_HEAD(efx_primary_list);
1201 static LIST_HEAD(efx_unassociated_list);
1202 
1203 static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right)
1204 {
1205 	return left->type == right->type &&
1206 		left->vpd_sn && right->vpd_sn &&
1207 		!strcmp(left->vpd_sn, right->vpd_sn);
1208 }
1209 
1210 static void efx_associate(struct efx_nic *efx)
1211 {
1212 	struct efx_nic *other, *next;
1213 
1214 	if (efx->primary == efx) {
1215 		/* Adding primary function; look for secondaries */
1216 
1217 		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
1218 		list_add_tail(&efx->node, &efx_primary_list);
1219 
1220 		list_for_each_entry_safe(other, next, &efx_unassociated_list,
1221 					 node) {
1222 			if (efx_same_controller(efx, other)) {
1223 				list_del(&other->node);
1224 				netif_dbg(other, probe, other->net_dev,
1225 					  "moving to secondary list of %s %s\n",
1226 					  pci_name(efx->pci_dev),
1227 					  efx->net_dev->name);
1228 				list_add_tail(&other->node,
1229 					      &efx->secondary_list);
1230 				other->primary = efx;
1231 			}
1232 		}
1233 	} else {
1234 		/* Adding secondary function; look for primary */
1235 
1236 		list_for_each_entry(other, &efx_primary_list, node) {
1237 			if (efx_same_controller(efx, other)) {
1238 				netif_dbg(efx, probe, efx->net_dev,
1239 					  "adding to secondary list of %s %s\n",
1240 					  pci_name(other->pci_dev),
1241 					  other->net_dev->name);
1242 				list_add_tail(&efx->node,
1243 					      &other->secondary_list);
1244 				efx->primary = other;
1245 				return;
1246 			}
1247 		}
1248 
1249 		netif_dbg(efx, probe, efx->net_dev,
1250 			  "adding to unassociated list\n");
1251 		list_add_tail(&efx->node, &efx_unassociated_list);
1252 	}
1253 }
1254 
1255 static void efx_dissociate(struct efx_nic *efx)
1256 {
1257 	struct efx_nic *other, *next;
1258 
1259 	list_del(&efx->node);
1260 	efx->primary = NULL;
1261 
1262 	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
1263 		list_del(&other->node);
1264 		netif_dbg(other, probe, other->net_dev,
1265 			  "moving to unassociated list\n");
1266 		list_add_tail(&other->node, &efx_unassociated_list);
1267 		other->primary = NULL;
1268 	}
1269 }
1270 
1271 /* This configures the PCI device to enable I/O and DMA. */
1272 static int efx_init_io(struct efx_nic *efx)
1273 {
1274 	struct pci_dev *pci_dev = efx->pci_dev;
1275 	dma_addr_t dma_mask = efx->type->max_dma_mask;
1276 	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1277 	int rc, bar;
1278 
1279 	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1280 
1281 	bar = efx->type->mem_bar(efx);
1282 
1283 	rc = pci_enable_device(pci_dev);
1284 	if (rc) {
1285 		netif_err(efx, probe, efx->net_dev,
1286 			  "failed to enable PCI device\n");
1287 		goto fail1;
1288 	}
1289 
1290 	pci_set_master(pci_dev);
1291 
1292 	/* Set the PCI DMA mask.  Try all possibilities from our
1293 	 * genuine mask down to 32 bits, because some architectures
1294 	 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
1295 	 * masks event though they reject 46 bit masks.
1296 	 */
1297 	while (dma_mask > 0x7fffffffUL) {
1298 		rc = dma_set_mask_and_coherent(&pci_dev->dev, dma_mask);
1299 		if (rc == 0)
1300 			break;
1301 		dma_mask >>= 1;
1302 	}
1303 	if (rc) {
1304 		netif_err(efx, probe, efx->net_dev,
1305 			  "could not find a suitable DMA mask\n");
1306 		goto fail2;
1307 	}
1308 	netif_dbg(efx, probe, efx->net_dev,
1309 		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1310 
1311 	efx->membase_phys = pci_resource_start(efx->pci_dev, bar);
1312 	rc = pci_request_region(pci_dev, bar, "sfc");
1313 	if (rc) {
1314 		netif_err(efx, probe, efx->net_dev,
1315 			  "request for memory BAR failed\n");
1316 		rc = -EIO;
1317 		goto fail3;
1318 	}
1319 	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1320 	if (!efx->membase) {
1321 		netif_err(efx, probe, efx->net_dev,
1322 			  "could not map memory BAR at %llx+%x\n",
1323 			  (unsigned long long)efx->membase_phys, mem_map_size);
1324 		rc = -ENOMEM;
1325 		goto fail4;
1326 	}
1327 	netif_dbg(efx, probe, efx->net_dev,
1328 		  "memory BAR at %llx+%x (virtual %p)\n",
1329 		  (unsigned long long)efx->membase_phys, mem_map_size,
1330 		  efx->membase);
1331 
1332 	return 0;
1333 
1334  fail4:
1335 	pci_release_region(efx->pci_dev, bar);
1336  fail3:
1337 	efx->membase_phys = 0;
1338  fail2:
1339 	pci_disable_device(efx->pci_dev);
1340  fail1:
1341 	return rc;
1342 }
1343 
1344 static void efx_fini_io(struct efx_nic *efx)
1345 {
1346 	int bar;
1347 
1348 	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1349 
1350 	if (efx->membase) {
1351 		iounmap(efx->membase);
1352 		efx->membase = NULL;
1353 	}
1354 
1355 	if (efx->membase_phys) {
1356 		bar = efx->type->mem_bar(efx);
1357 		pci_release_region(efx->pci_dev, bar);
1358 		efx->membase_phys = 0;
1359 	}
1360 
1361 	/* Don't disable bus-mastering if VFs are assigned */
1362 	if (!pci_vfs_assigned(efx->pci_dev))
1363 		pci_disable_device(efx->pci_dev);
1364 }
1365 
1366 void efx_set_default_rx_indir_table(struct efx_nic *efx,
1367 				    struct efx_rss_context *ctx)
1368 {
1369 	size_t i;
1370 
1371 	for (i = 0; i < ARRAY_SIZE(ctx->rx_indir_table); i++)
1372 		ctx->rx_indir_table[i] =
1373 			ethtool_rxfh_indir_default(i, efx->rss_spread);
1374 }
1375 
1376 static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1377 {
1378 	cpumask_var_t thread_mask;
1379 	unsigned int count;
1380 	int cpu;
1381 
1382 	if (rss_cpus) {
1383 		count = rss_cpus;
1384 	} else {
1385 		if (unlikely(!zalloc_cpumask_var(&thread_mask, GFP_KERNEL))) {
1386 			netif_warn(efx, probe, efx->net_dev,
1387 				   "RSS disabled due to allocation failure\n");
1388 			return 1;
1389 		}
1390 
1391 		count = 0;
1392 		for_each_online_cpu(cpu) {
1393 			if (!cpumask_test_cpu(cpu, thread_mask)) {
1394 				++count;
1395 				cpumask_or(thread_mask, thread_mask,
1396 					   topology_sibling_cpumask(cpu));
1397 			}
1398 		}
1399 
1400 		free_cpumask_var(thread_mask);
1401 	}
1402 
1403 	if (count > EFX_MAX_RX_QUEUES) {
1404 		netif_cond_dbg(efx, probe, efx->net_dev, !rss_cpus, warn,
1405 			       "Reducing number of rx queues from %u to %u.\n",
1406 			       count, EFX_MAX_RX_QUEUES);
1407 		count = EFX_MAX_RX_QUEUES;
1408 	}
1409 
1410 	/* If RSS is requested for the PF *and* VFs then we can't write RSS
1411 	 * table entries that are inaccessible to VFs
1412 	 */
1413 #ifdef CONFIG_SFC_SRIOV
1414 	if (efx->type->sriov_wanted) {
1415 		if (efx->type->sriov_wanted(efx) && efx_vf_size(efx) > 1 &&
1416 		    count > efx_vf_size(efx)) {
1417 			netif_warn(efx, probe, efx->net_dev,
1418 				   "Reducing number of RSS channels from %u to %u for "
1419 				   "VF support. Increase vf-msix-limit to use more "
1420 				   "channels on the PF.\n",
1421 				   count, efx_vf_size(efx));
1422 			count = efx_vf_size(efx);
1423 		}
1424 	}
1425 #endif
1426 
1427 	return count;
1428 }
1429 
1430 /* Probe the number and type of interrupts we are able to obtain, and
1431  * the resulting numbers of channels and RX queues.
1432  */
1433 static int efx_probe_interrupts(struct efx_nic *efx)
1434 {
1435 	unsigned int extra_channels = 0;
1436 	unsigned int i, j;
1437 	int rc;
1438 
1439 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
1440 		if (efx->extra_channel_type[i])
1441 			++extra_channels;
1442 
1443 	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1444 		struct msix_entry xentries[EFX_MAX_CHANNELS];
1445 		unsigned int n_channels;
1446 
1447 		n_channels = efx_wanted_parallelism(efx);
1448 		if (efx_separate_tx_channels)
1449 			n_channels *= 2;
1450 		n_channels += extra_channels;
1451 		n_channels = min(n_channels, efx->max_channels);
1452 
1453 		for (i = 0; i < n_channels; i++)
1454 			xentries[i].entry = i;
1455 		rc = pci_enable_msix_range(efx->pci_dev,
1456 					   xentries, 1, n_channels);
1457 		if (rc < 0) {
1458 			/* Fall back to single channel MSI */
1459 			netif_err(efx, drv, efx->net_dev,
1460 				  "could not enable MSI-X\n");
1461 			if (efx->type->min_interrupt_mode >= EFX_INT_MODE_MSI)
1462 				efx->interrupt_mode = EFX_INT_MODE_MSI;
1463 			else
1464 				return rc;
1465 		} else if (rc < n_channels) {
1466 			netif_err(efx, drv, efx->net_dev,
1467 				  "WARNING: Insufficient MSI-X vectors"
1468 				  " available (%d < %u).\n", rc, n_channels);
1469 			netif_err(efx, drv, efx->net_dev,
1470 				  "WARNING: Performance may be reduced.\n");
1471 			n_channels = rc;
1472 		}
1473 
1474 		if (rc > 0) {
1475 			efx->n_channels = n_channels;
1476 			if (n_channels > extra_channels)
1477 				n_channels -= extra_channels;
1478 			if (efx_separate_tx_channels) {
1479 				efx->n_tx_channels = min(max(n_channels / 2,
1480 							     1U),
1481 							 efx->max_tx_channels);
1482 				efx->n_rx_channels = max(n_channels -
1483 							 efx->n_tx_channels,
1484 							 1U);
1485 			} else {
1486 				efx->n_tx_channels = min(n_channels,
1487 							 efx->max_tx_channels);
1488 				efx->n_rx_channels = n_channels;
1489 			}
1490 			for (i = 0; i < efx->n_channels; i++)
1491 				efx_get_channel(efx, i)->irq =
1492 					xentries[i].vector;
1493 		}
1494 	}
1495 
1496 	/* Try single interrupt MSI */
1497 	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1498 		efx->n_channels = 1;
1499 		efx->n_rx_channels = 1;
1500 		efx->n_tx_channels = 1;
1501 		rc = pci_enable_msi(efx->pci_dev);
1502 		if (rc == 0) {
1503 			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1504 		} else {
1505 			netif_err(efx, drv, efx->net_dev,
1506 				  "could not enable MSI\n");
1507 			if (efx->type->min_interrupt_mode >= EFX_INT_MODE_LEGACY)
1508 				efx->interrupt_mode = EFX_INT_MODE_LEGACY;
1509 			else
1510 				return rc;
1511 		}
1512 	}
1513 
1514 	/* Assume legacy interrupts */
1515 	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1516 		efx->n_channels = 1 + (efx_separate_tx_channels ? 1 : 0);
1517 		efx->n_rx_channels = 1;
1518 		efx->n_tx_channels = 1;
1519 		efx->legacy_irq = efx->pci_dev->irq;
1520 	}
1521 
1522 	/* Assign extra channels if possible */
1523 	efx->n_extra_tx_channels = 0;
1524 	j = efx->n_channels;
1525 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++) {
1526 		if (!efx->extra_channel_type[i])
1527 			continue;
1528 		if (efx->interrupt_mode != EFX_INT_MODE_MSIX ||
1529 		    efx->n_channels <= extra_channels) {
1530 			efx->extra_channel_type[i]->handle_no_channel(efx);
1531 		} else {
1532 			--j;
1533 			efx_get_channel(efx, j)->type =
1534 				efx->extra_channel_type[i];
1535 			if (efx_channel_has_tx_queues(efx_get_channel(efx, j)))
1536 				efx->n_extra_tx_channels++;
1537 		}
1538 	}
1539 
1540 	/* RSS might be usable on VFs even if it is disabled on the PF */
1541 #ifdef CONFIG_SFC_SRIOV
1542 	if (efx->type->sriov_wanted) {
1543 		efx->rss_spread = ((efx->n_rx_channels > 1 ||
1544 				    !efx->type->sriov_wanted(efx)) ?
1545 				   efx->n_rx_channels : efx_vf_size(efx));
1546 		return 0;
1547 	}
1548 #endif
1549 	efx->rss_spread = efx->n_rx_channels;
1550 
1551 	return 0;
1552 }
1553 
1554 static int efx_soft_enable_interrupts(struct efx_nic *efx)
1555 {
1556 	struct efx_channel *channel, *end_channel;
1557 	int rc;
1558 
1559 	BUG_ON(efx->state == STATE_DISABLED);
1560 
1561 	efx->irq_soft_enabled = true;
1562 	smp_wmb();
1563 
1564 	efx_for_each_channel(channel, efx) {
1565 		if (!channel->type->keep_eventq) {
1566 			rc = efx_init_eventq(channel);
1567 			if (rc)
1568 				goto fail;
1569 		}
1570 		efx_start_eventq(channel);
1571 	}
1572 
1573 	efx_mcdi_mode_event(efx);
1574 
1575 	return 0;
1576 fail:
1577 	end_channel = channel;
1578 	efx_for_each_channel(channel, efx) {
1579 		if (channel == end_channel)
1580 			break;
1581 		efx_stop_eventq(channel);
1582 		if (!channel->type->keep_eventq)
1583 			efx_fini_eventq(channel);
1584 	}
1585 
1586 	return rc;
1587 }
1588 
1589 static void efx_soft_disable_interrupts(struct efx_nic *efx)
1590 {
1591 	struct efx_channel *channel;
1592 
1593 	if (efx->state == STATE_DISABLED)
1594 		return;
1595 
1596 	efx_mcdi_mode_poll(efx);
1597 
1598 	efx->irq_soft_enabled = false;
1599 	smp_wmb();
1600 
1601 	if (efx->legacy_irq)
1602 		synchronize_irq(efx->legacy_irq);
1603 
1604 	efx_for_each_channel(channel, efx) {
1605 		if (channel->irq)
1606 			synchronize_irq(channel->irq);
1607 
1608 		efx_stop_eventq(channel);
1609 		if (!channel->type->keep_eventq)
1610 			efx_fini_eventq(channel);
1611 	}
1612 
1613 	/* Flush the asynchronous MCDI request queue */
1614 	efx_mcdi_flush_async(efx);
1615 }
1616 
1617 static int efx_enable_interrupts(struct efx_nic *efx)
1618 {
1619 	struct efx_channel *channel, *end_channel;
1620 	int rc;
1621 
1622 	BUG_ON(efx->state == STATE_DISABLED);
1623 
1624 	if (efx->eeh_disabled_legacy_irq) {
1625 		enable_irq(efx->legacy_irq);
1626 		efx->eeh_disabled_legacy_irq = false;
1627 	}
1628 
1629 	efx->type->irq_enable_master(efx);
1630 
1631 	efx_for_each_channel(channel, efx) {
1632 		if (channel->type->keep_eventq) {
1633 			rc = efx_init_eventq(channel);
1634 			if (rc)
1635 				goto fail;
1636 		}
1637 	}
1638 
1639 	rc = efx_soft_enable_interrupts(efx);
1640 	if (rc)
1641 		goto fail;
1642 
1643 	return 0;
1644 
1645 fail:
1646 	end_channel = channel;
1647 	efx_for_each_channel(channel, efx) {
1648 		if (channel == end_channel)
1649 			break;
1650 		if (channel->type->keep_eventq)
1651 			efx_fini_eventq(channel);
1652 	}
1653 
1654 	efx->type->irq_disable_non_ev(efx);
1655 
1656 	return rc;
1657 }
1658 
1659 static void efx_disable_interrupts(struct efx_nic *efx)
1660 {
1661 	struct efx_channel *channel;
1662 
1663 	efx_soft_disable_interrupts(efx);
1664 
1665 	efx_for_each_channel(channel, efx) {
1666 		if (channel->type->keep_eventq)
1667 			efx_fini_eventq(channel);
1668 	}
1669 
1670 	efx->type->irq_disable_non_ev(efx);
1671 }
1672 
1673 static void efx_remove_interrupts(struct efx_nic *efx)
1674 {
1675 	struct efx_channel *channel;
1676 
1677 	/* Remove MSI/MSI-X interrupts */
1678 	efx_for_each_channel(channel, efx)
1679 		channel->irq = 0;
1680 	pci_disable_msi(efx->pci_dev);
1681 	pci_disable_msix(efx->pci_dev);
1682 
1683 	/* Remove legacy interrupt */
1684 	efx->legacy_irq = 0;
1685 }
1686 
1687 static void efx_set_channels(struct efx_nic *efx)
1688 {
1689 	struct efx_channel *channel;
1690 	struct efx_tx_queue *tx_queue;
1691 
1692 	efx->tx_channel_offset =
1693 		efx_separate_tx_channels ?
1694 		efx->n_channels - efx->n_tx_channels : 0;
1695 
1696 	/* We need to mark which channels really have RX and TX
1697 	 * queues, and adjust the TX queue numbers if we have separate
1698 	 * RX-only and TX-only channels.
1699 	 */
1700 	efx_for_each_channel(channel, efx) {
1701 		if (channel->channel < efx->n_rx_channels)
1702 			channel->rx_queue.core_index = channel->channel;
1703 		else
1704 			channel->rx_queue.core_index = -1;
1705 
1706 		efx_for_each_channel_tx_queue(tx_queue, channel)
1707 			tx_queue->queue -= (efx->tx_channel_offset *
1708 					    EFX_TXQ_TYPES);
1709 	}
1710 }
1711 
1712 static int efx_probe_nic(struct efx_nic *efx)
1713 {
1714 	int rc;
1715 
1716 	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1717 
1718 	/* Carry out hardware-type specific initialisation */
1719 	rc = efx->type->probe(efx);
1720 	if (rc)
1721 		return rc;
1722 
1723 	do {
1724 		if (!efx->max_channels || !efx->max_tx_channels) {
1725 			netif_err(efx, drv, efx->net_dev,
1726 				  "Insufficient resources to allocate"
1727 				  " any channels\n");
1728 			rc = -ENOSPC;
1729 			goto fail1;
1730 		}
1731 
1732 		/* Determine the number of channels and queues by trying
1733 		 * to hook in MSI-X interrupts.
1734 		 */
1735 		rc = efx_probe_interrupts(efx);
1736 		if (rc)
1737 			goto fail1;
1738 
1739 		efx_set_channels(efx);
1740 
1741 		/* dimension_resources can fail with EAGAIN */
1742 		rc = efx->type->dimension_resources(efx);
1743 		if (rc != 0 && rc != -EAGAIN)
1744 			goto fail2;
1745 
1746 		if (rc == -EAGAIN)
1747 			/* try again with new max_channels */
1748 			efx_remove_interrupts(efx);
1749 
1750 	} while (rc == -EAGAIN);
1751 
1752 	if (efx->n_channels > 1)
1753 		netdev_rss_key_fill(efx->rss_context.rx_hash_key,
1754 				    sizeof(efx->rss_context.rx_hash_key));
1755 	efx_set_default_rx_indir_table(efx, &efx->rss_context);
1756 
1757 	netif_set_real_num_tx_queues(efx->net_dev, efx->n_tx_channels);
1758 	netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels);
1759 
1760 	/* Initialise the interrupt moderation settings */
1761 	efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000);
1762 	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
1763 				true);
1764 
1765 	return 0;
1766 
1767 fail2:
1768 	efx_remove_interrupts(efx);
1769 fail1:
1770 	efx->type->remove(efx);
1771 	return rc;
1772 }
1773 
1774 static void efx_remove_nic(struct efx_nic *efx)
1775 {
1776 	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1777 
1778 	efx_remove_interrupts(efx);
1779 	efx->type->remove(efx);
1780 }
1781 
1782 static int efx_probe_filters(struct efx_nic *efx)
1783 {
1784 	int rc;
1785 
1786 	init_rwsem(&efx->filter_sem);
1787 	mutex_lock(&efx->mac_lock);
1788 	down_write(&efx->filter_sem);
1789 	rc = efx->type->filter_table_probe(efx);
1790 	if (rc)
1791 		goto out_unlock;
1792 
1793 #ifdef CONFIG_RFS_ACCEL
1794 	if (efx->type->offload_features & NETIF_F_NTUPLE) {
1795 		struct efx_channel *channel;
1796 		int i, success = 1;
1797 
1798 		efx_for_each_channel(channel, efx) {
1799 			channel->rps_flow_id =
1800 				kcalloc(efx->type->max_rx_ip_filters,
1801 					sizeof(*channel->rps_flow_id),
1802 					GFP_KERNEL);
1803 			if (!channel->rps_flow_id)
1804 				success = 0;
1805 			else
1806 				for (i = 0;
1807 				     i < efx->type->max_rx_ip_filters;
1808 				     ++i)
1809 					channel->rps_flow_id[i] =
1810 						RPS_FLOW_ID_INVALID;
1811 		}
1812 
1813 		if (!success) {
1814 			efx_for_each_channel(channel, efx)
1815 				kfree(channel->rps_flow_id);
1816 			efx->type->filter_table_remove(efx);
1817 			rc = -ENOMEM;
1818 			goto out_unlock;
1819 		}
1820 
1821 		efx->rps_expire_index = efx->rps_expire_channel = 0;
1822 	}
1823 #endif
1824 out_unlock:
1825 	up_write(&efx->filter_sem);
1826 	mutex_unlock(&efx->mac_lock);
1827 	return rc;
1828 }
1829 
1830 static void efx_remove_filters(struct efx_nic *efx)
1831 {
1832 #ifdef CONFIG_RFS_ACCEL
1833 	struct efx_channel *channel;
1834 
1835 	efx_for_each_channel(channel, efx)
1836 		kfree(channel->rps_flow_id);
1837 #endif
1838 	down_write(&efx->filter_sem);
1839 	efx->type->filter_table_remove(efx);
1840 	up_write(&efx->filter_sem);
1841 }
1842 
1843 static void efx_restore_filters(struct efx_nic *efx)
1844 {
1845 	down_read(&efx->filter_sem);
1846 	efx->type->filter_table_restore(efx);
1847 	up_read(&efx->filter_sem);
1848 }
1849 
1850 /**************************************************************************
1851  *
1852  * NIC startup/shutdown
1853  *
1854  *************************************************************************/
1855 
1856 static int efx_probe_all(struct efx_nic *efx)
1857 {
1858 	int rc;
1859 
1860 	rc = efx_probe_nic(efx);
1861 	if (rc) {
1862 		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1863 		goto fail1;
1864 	}
1865 
1866 	rc = efx_probe_port(efx);
1867 	if (rc) {
1868 		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1869 		goto fail2;
1870 	}
1871 
1872 	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
1873 	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
1874 		rc = -EINVAL;
1875 		goto fail3;
1876 	}
1877 	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1878 
1879 #ifdef CONFIG_SFC_SRIOV
1880 	rc = efx->type->vswitching_probe(efx);
1881 	if (rc) /* not fatal; the PF will still work fine */
1882 		netif_warn(efx, probe, efx->net_dev,
1883 			   "failed to setup vswitching rc=%d;"
1884 			   " VFs may not function\n", rc);
1885 #endif
1886 
1887 	rc = efx_probe_filters(efx);
1888 	if (rc) {
1889 		netif_err(efx, probe, efx->net_dev,
1890 			  "failed to create filter tables\n");
1891 		goto fail4;
1892 	}
1893 
1894 	rc = efx_probe_channels(efx);
1895 	if (rc)
1896 		goto fail5;
1897 
1898 	return 0;
1899 
1900  fail5:
1901 	efx_remove_filters(efx);
1902  fail4:
1903 #ifdef CONFIG_SFC_SRIOV
1904 	efx->type->vswitching_remove(efx);
1905 #endif
1906  fail3:
1907 	efx_remove_port(efx);
1908  fail2:
1909 	efx_remove_nic(efx);
1910  fail1:
1911 	return rc;
1912 }
1913 
1914 /* If the interface is supposed to be running but is not, start
1915  * the hardware and software data path, regular activity for the port
1916  * (MAC statistics, link polling, etc.) and schedule the port to be
1917  * reconfigured.  Interrupts must already be enabled.  This function
1918  * is safe to call multiple times, so long as the NIC is not disabled.
1919  * Requires the RTNL lock.
1920  */
1921 static void efx_start_all(struct efx_nic *efx)
1922 {
1923 	EFX_ASSERT_RESET_SERIALISED(efx);
1924 	BUG_ON(efx->state == STATE_DISABLED);
1925 
1926 	/* Check that it is appropriate to restart the interface. All
1927 	 * of these flags are safe to read under just the rtnl lock */
1928 	if (efx->port_enabled || !netif_running(efx->net_dev) ||
1929 	    efx->reset_pending)
1930 		return;
1931 
1932 	efx_start_port(efx);
1933 	efx_start_datapath(efx);
1934 
1935 	/* Start the hardware monitor if there is one */
1936 	if (efx->type->monitor != NULL)
1937 		queue_delayed_work(efx->workqueue, &efx->monitor_work,
1938 				   efx_monitor_interval);
1939 
1940 	/* Link state detection is normally event-driven; we have
1941 	 * to poll now because we could have missed a change
1942 	 */
1943 	mutex_lock(&efx->mac_lock);
1944 	if (efx->phy_op->poll(efx))
1945 		efx_link_status_changed(efx);
1946 	mutex_unlock(&efx->mac_lock);
1947 
1948 	efx->type->start_stats(efx);
1949 	efx->type->pull_stats(efx);
1950 	spin_lock_bh(&efx->stats_lock);
1951 	efx->type->update_stats(efx, NULL, NULL);
1952 	spin_unlock_bh(&efx->stats_lock);
1953 }
1954 
1955 /* Quiesce the hardware and software data path, and regular activity
1956  * for the port without bringing the link down.  Safe to call multiple
1957  * times with the NIC in almost any state, but interrupts should be
1958  * enabled.  Requires the RTNL lock.
1959  */
1960 static void efx_stop_all(struct efx_nic *efx)
1961 {
1962 	EFX_ASSERT_RESET_SERIALISED(efx);
1963 
1964 	/* port_enabled can be read safely under the rtnl lock */
1965 	if (!efx->port_enabled)
1966 		return;
1967 
1968 	/* update stats before we go down so we can accurately count
1969 	 * rx_nodesc_drops
1970 	 */
1971 	efx->type->pull_stats(efx);
1972 	spin_lock_bh(&efx->stats_lock);
1973 	efx->type->update_stats(efx, NULL, NULL);
1974 	spin_unlock_bh(&efx->stats_lock);
1975 	efx->type->stop_stats(efx);
1976 	efx_stop_port(efx);
1977 
1978 	/* Stop the kernel transmit interface.  This is only valid if
1979 	 * the device is stopped or detached; otherwise the watchdog
1980 	 * may fire immediately.
1981 	 */
1982 	WARN_ON(netif_running(efx->net_dev) &&
1983 		netif_device_present(efx->net_dev));
1984 	netif_tx_disable(efx->net_dev);
1985 
1986 	efx_stop_datapath(efx);
1987 }
1988 
1989 static void efx_remove_all(struct efx_nic *efx)
1990 {
1991 	efx_remove_channels(efx);
1992 	efx_remove_filters(efx);
1993 #ifdef CONFIG_SFC_SRIOV
1994 	efx->type->vswitching_remove(efx);
1995 #endif
1996 	efx_remove_port(efx);
1997 	efx_remove_nic(efx);
1998 }
1999 
2000 /**************************************************************************
2001  *
2002  * Interrupt moderation
2003  *
2004  **************************************************************************/
2005 unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs)
2006 {
2007 	if (usecs == 0)
2008 		return 0;
2009 	if (usecs * 1000 < efx->timer_quantum_ns)
2010 		return 1; /* never round down to 0 */
2011 	return usecs * 1000 / efx->timer_quantum_ns;
2012 }
2013 
2014 unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks)
2015 {
2016 	/* We must round up when converting ticks to microseconds
2017 	 * because we round down when converting the other way.
2018 	 */
2019 	return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000);
2020 }
2021 
2022 /* Set interrupt moderation parameters */
2023 int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
2024 			    unsigned int rx_usecs, bool rx_adaptive,
2025 			    bool rx_may_override_tx)
2026 {
2027 	struct efx_channel *channel;
2028 	unsigned int timer_max_us;
2029 
2030 	EFX_ASSERT_RESET_SERIALISED(efx);
2031 
2032 	timer_max_us = efx->timer_max_ns / 1000;
2033 
2034 	if (tx_usecs > timer_max_us || rx_usecs > timer_max_us)
2035 		return -EINVAL;
2036 
2037 	if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 &&
2038 	    !rx_may_override_tx) {
2039 		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
2040 			  "RX and TX IRQ moderation must be equal\n");
2041 		return -EINVAL;
2042 	}
2043 
2044 	efx->irq_rx_adaptive = rx_adaptive;
2045 	efx->irq_rx_moderation_us = rx_usecs;
2046 	efx_for_each_channel(channel, efx) {
2047 		if (efx_channel_has_rx_queue(channel))
2048 			channel->irq_moderation_us = rx_usecs;
2049 		else if (efx_channel_has_tx_queues(channel))
2050 			channel->irq_moderation_us = tx_usecs;
2051 	}
2052 
2053 	return 0;
2054 }
2055 
2056 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
2057 			    unsigned int *rx_usecs, bool *rx_adaptive)
2058 {
2059 	*rx_adaptive = efx->irq_rx_adaptive;
2060 	*rx_usecs = efx->irq_rx_moderation_us;
2061 
2062 	/* If channels are shared between RX and TX, so is IRQ
2063 	 * moderation.  Otherwise, IRQ moderation is the same for all
2064 	 * TX channels and is not adaptive.
2065 	 */
2066 	if (efx->tx_channel_offset == 0) {
2067 		*tx_usecs = *rx_usecs;
2068 	} else {
2069 		struct efx_channel *tx_channel;
2070 
2071 		tx_channel = efx->channel[efx->tx_channel_offset];
2072 		*tx_usecs = tx_channel->irq_moderation_us;
2073 	}
2074 }
2075 
2076 /**************************************************************************
2077  *
2078  * Hardware monitor
2079  *
2080  **************************************************************************/
2081 
2082 /* Run periodically off the general workqueue */
2083 static void efx_monitor(struct work_struct *data)
2084 {
2085 	struct efx_nic *efx = container_of(data, struct efx_nic,
2086 					   monitor_work.work);
2087 
2088 	netif_vdbg(efx, timer, efx->net_dev,
2089 		   "hardware monitor executing on CPU %d\n",
2090 		   raw_smp_processor_id());
2091 	BUG_ON(efx->type->monitor == NULL);
2092 
2093 	/* If the mac_lock is already held then it is likely a port
2094 	 * reconfiguration is already in place, which will likely do
2095 	 * most of the work of monitor() anyway. */
2096 	if (mutex_trylock(&efx->mac_lock)) {
2097 		if (efx->port_enabled)
2098 			efx->type->monitor(efx);
2099 		mutex_unlock(&efx->mac_lock);
2100 	}
2101 
2102 	queue_delayed_work(efx->workqueue, &efx->monitor_work,
2103 			   efx_monitor_interval);
2104 }
2105 
2106 /**************************************************************************
2107  *
2108  * ioctls
2109  *
2110  *************************************************************************/
2111 
2112 /* Net device ioctl
2113  * Context: process, rtnl_lock() held.
2114  */
2115 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
2116 {
2117 	struct efx_nic *efx = netdev_priv(net_dev);
2118 	struct mii_ioctl_data *data = if_mii(ifr);
2119 
2120 	if (cmd == SIOCSHWTSTAMP)
2121 		return efx_ptp_set_ts_config(efx, ifr);
2122 	if (cmd == SIOCGHWTSTAMP)
2123 		return efx_ptp_get_ts_config(efx, ifr);
2124 
2125 	/* Convert phy_id from older PRTAD/DEVAD format */
2126 	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
2127 	    (data->phy_id & 0xfc00) == 0x0400)
2128 		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;
2129 
2130 	return mdio_mii_ioctl(&efx->mdio, data, cmd);
2131 }
2132 
2133 /**************************************************************************
2134  *
2135  * NAPI interface
2136  *
2137  **************************************************************************/
2138 
2139 static void efx_init_napi_channel(struct efx_channel *channel)
2140 {
2141 	struct efx_nic *efx = channel->efx;
2142 
2143 	channel->napi_dev = efx->net_dev;
2144 	netif_napi_add(channel->napi_dev, &channel->napi_str,
2145 		       efx_poll, napi_weight);
2146 }
2147 
2148 static void efx_init_napi(struct efx_nic *efx)
2149 {
2150 	struct efx_channel *channel;
2151 
2152 	efx_for_each_channel(channel, efx)
2153 		efx_init_napi_channel(channel);
2154 }
2155 
2156 static void efx_fini_napi_channel(struct efx_channel *channel)
2157 {
2158 	if (channel->napi_dev)
2159 		netif_napi_del(&channel->napi_str);
2160 
2161 	channel->napi_dev = NULL;
2162 }
2163 
2164 static void efx_fini_napi(struct efx_nic *efx)
2165 {
2166 	struct efx_channel *channel;
2167 
2168 	efx_for_each_channel(channel, efx)
2169 		efx_fini_napi_channel(channel);
2170 }
2171 
2172 /**************************************************************************
2173  *
2174  * Kernel netpoll interface
2175  *
2176  *************************************************************************/
2177 
2178 #ifdef CONFIG_NET_POLL_CONTROLLER
2179 
2180 /* Although in the common case interrupts will be disabled, this is not
2181  * guaranteed. However, all our work happens inside the NAPI callback,
2182  * so no locking is required.
2183  */
2184 static void efx_netpoll(struct net_device *net_dev)
2185 {
2186 	struct efx_nic *efx = netdev_priv(net_dev);
2187 	struct efx_channel *channel;
2188 
2189 	efx_for_each_channel(channel, efx)
2190 		efx_schedule_channel(channel);
2191 }
2192 
2193 #endif
2194 
2195 /**************************************************************************
2196  *
2197  * Kernel net device interface
2198  *
2199  *************************************************************************/
2200 
2201 /* Context: process, rtnl_lock() held. */
2202 int efx_net_open(struct net_device *net_dev)
2203 {
2204 	struct efx_nic *efx = netdev_priv(net_dev);
2205 	int rc;
2206 
2207 	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
2208 		  raw_smp_processor_id());
2209 
2210 	rc = efx_check_disabled(efx);
2211 	if (rc)
2212 		return rc;
2213 	if (efx->phy_mode & PHY_MODE_SPECIAL)
2214 		return -EBUSY;
2215 	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
2216 		return -EIO;
2217 
2218 	/* Notify the kernel of the link state polled during driver load,
2219 	 * before the monitor starts running */
2220 	efx_link_status_changed(efx);
2221 
2222 	efx_start_all(efx);
2223 	if (efx->state == STATE_DISABLED || efx->reset_pending)
2224 		netif_device_detach(efx->net_dev);
2225 	efx_selftest_async_start(efx);
2226 	return 0;
2227 }
2228 
2229 /* Context: process, rtnl_lock() held.
2230  * Note that the kernel will ignore our return code; this method
2231  * should really be a void.
2232  */
2233 int efx_net_stop(struct net_device *net_dev)
2234 {
2235 	struct efx_nic *efx = netdev_priv(net_dev);
2236 
2237 	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
2238 		  raw_smp_processor_id());
2239 
2240 	/* Stop the device and flush all the channels */
2241 	efx_stop_all(efx);
2242 
2243 	return 0;
2244 }
2245 
2246 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
2247 static void efx_net_stats(struct net_device *net_dev,
2248 			  struct rtnl_link_stats64 *stats)
2249 {
2250 	struct efx_nic *efx = netdev_priv(net_dev);
2251 
2252 	spin_lock_bh(&efx->stats_lock);
2253 	efx->type->update_stats(efx, NULL, stats);
2254 	spin_unlock_bh(&efx->stats_lock);
2255 }
2256 
2257 /* Context: netif_tx_lock held, BHs disabled. */
2258 static void efx_watchdog(struct net_device *net_dev)
2259 {
2260 	struct efx_nic *efx = netdev_priv(net_dev);
2261 
2262 	netif_err(efx, tx_err, efx->net_dev,
2263 		  "TX stuck with port_enabled=%d: resetting channels\n",
2264 		  efx->port_enabled);
2265 
2266 	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2267 }
2268 
2269 
2270 /* Context: process, rtnl_lock() held. */
2271 static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
2272 {
2273 	struct efx_nic *efx = netdev_priv(net_dev);
2274 	int rc;
2275 
2276 	rc = efx_check_disabled(efx);
2277 	if (rc)
2278 		return rc;
2279 
2280 	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2281 
2282 	efx_device_detach_sync(efx);
2283 	efx_stop_all(efx);
2284 
2285 	mutex_lock(&efx->mac_lock);
2286 	net_dev->mtu = new_mtu;
2287 	efx_mac_reconfigure(efx);
2288 	mutex_unlock(&efx->mac_lock);
2289 
2290 	efx_start_all(efx);
2291 	efx_device_attach_if_not_resetting(efx);
2292 	return 0;
2293 }
2294 
2295 static int efx_set_mac_address(struct net_device *net_dev, void *data)
2296 {
2297 	struct efx_nic *efx = netdev_priv(net_dev);
2298 	struct sockaddr *addr = data;
2299 	u8 *new_addr = addr->sa_data;
2300 	u8 old_addr[6];
2301 	int rc;
2302 
2303 	if (!is_valid_ether_addr(new_addr)) {
2304 		netif_err(efx, drv, efx->net_dev,
2305 			  "invalid ethernet MAC address requested: %pM\n",
2306 			  new_addr);
2307 		return -EADDRNOTAVAIL;
2308 	}
2309 
2310 	/* save old address */
2311 	ether_addr_copy(old_addr, net_dev->dev_addr);
2312 	ether_addr_copy(net_dev->dev_addr, new_addr);
2313 	if (efx->type->set_mac_address) {
2314 		rc = efx->type->set_mac_address(efx);
2315 		if (rc) {
2316 			ether_addr_copy(net_dev->dev_addr, old_addr);
2317 			return rc;
2318 		}
2319 	}
2320 
2321 	/* Reconfigure the MAC */
2322 	mutex_lock(&efx->mac_lock);
2323 	efx_mac_reconfigure(efx);
2324 	mutex_unlock(&efx->mac_lock);
2325 
2326 	return 0;
2327 }
2328 
2329 /* Context: netif_addr_lock held, BHs disabled. */
2330 static void efx_set_rx_mode(struct net_device *net_dev)
2331 {
2332 	struct efx_nic *efx = netdev_priv(net_dev);
2333 
2334 	if (efx->port_enabled)
2335 		queue_work(efx->workqueue, &efx->mac_work);
2336 	/* Otherwise efx_start_port() will do this */
2337 }
2338 
2339 static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2340 {
2341 	struct efx_nic *efx = netdev_priv(net_dev);
2342 	int rc;
2343 
2344 	/* If disabling RX n-tuple filtering, clear existing filters */
2345 	if (net_dev->features & ~data & NETIF_F_NTUPLE) {
2346 		rc = efx->type->filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);
2347 		if (rc)
2348 			return rc;
2349 	}
2350 
2351 	/* If Rx VLAN filter is changed, update filters via mac_reconfigure.
2352 	 * If rx-fcs is changed, mac_reconfigure updates that too.
2353 	 */
2354 	if ((net_dev->features ^ data) & (NETIF_F_HW_VLAN_CTAG_FILTER |
2355 					  NETIF_F_RXFCS)) {
2356 		/* efx_set_rx_mode() will schedule MAC work to update filters
2357 		 * when a new features are finally set in net_dev.
2358 		 */
2359 		efx_set_rx_mode(net_dev);
2360 	}
2361 
2362 	return 0;
2363 }
2364 
2365 static int efx_get_phys_port_id(struct net_device *net_dev,
2366 				struct netdev_phys_item_id *ppid)
2367 {
2368 	struct efx_nic *efx = netdev_priv(net_dev);
2369 
2370 	if (efx->type->get_phys_port_id)
2371 		return efx->type->get_phys_port_id(efx, ppid);
2372 	else
2373 		return -EOPNOTSUPP;
2374 }
2375 
2376 static int efx_get_phys_port_name(struct net_device *net_dev,
2377 				  char *name, size_t len)
2378 {
2379 	struct efx_nic *efx = netdev_priv(net_dev);
2380 
2381 	if (snprintf(name, len, "p%u", efx->port_num) >= len)
2382 		return -EINVAL;
2383 	return 0;
2384 }
2385 
2386 static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid)
2387 {
2388 	struct efx_nic *efx = netdev_priv(net_dev);
2389 
2390 	if (efx->type->vlan_rx_add_vid)
2391 		return efx->type->vlan_rx_add_vid(efx, proto, vid);
2392 	else
2393 		return -EOPNOTSUPP;
2394 }
2395 
2396 static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid)
2397 {
2398 	struct efx_nic *efx = netdev_priv(net_dev);
2399 
2400 	if (efx->type->vlan_rx_kill_vid)
2401 		return efx->type->vlan_rx_kill_vid(efx, proto, vid);
2402 	else
2403 		return -EOPNOTSUPP;
2404 }
2405 
2406 static int efx_udp_tunnel_type_map(enum udp_parsable_tunnel_type in)
2407 {
2408 	switch (in) {
2409 	case UDP_TUNNEL_TYPE_VXLAN:
2410 		return TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN;
2411 	case UDP_TUNNEL_TYPE_GENEVE:
2412 		return TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE;
2413 	default:
2414 		return -1;
2415 	}
2416 }
2417 
2418 static void efx_udp_tunnel_add(struct net_device *dev, struct udp_tunnel_info *ti)
2419 {
2420 	struct efx_nic *efx = netdev_priv(dev);
2421 	struct efx_udp_tunnel tnl;
2422 	int efx_tunnel_type;
2423 
2424 	efx_tunnel_type = efx_udp_tunnel_type_map(ti->type);
2425 	if (efx_tunnel_type < 0)
2426 		return;
2427 
2428 	tnl.type = (u16)efx_tunnel_type;
2429 	tnl.port = ti->port;
2430 
2431 	if (efx->type->udp_tnl_add_port)
2432 		(void)efx->type->udp_tnl_add_port(efx, tnl);
2433 }
2434 
2435 static void efx_udp_tunnel_del(struct net_device *dev, struct udp_tunnel_info *ti)
2436 {
2437 	struct efx_nic *efx = netdev_priv(dev);
2438 	struct efx_udp_tunnel tnl;
2439 	int efx_tunnel_type;
2440 
2441 	efx_tunnel_type = efx_udp_tunnel_type_map(ti->type);
2442 	if (efx_tunnel_type < 0)
2443 		return;
2444 
2445 	tnl.type = (u16)efx_tunnel_type;
2446 	tnl.port = ti->port;
2447 
2448 	if (efx->type->udp_tnl_del_port)
2449 		(void)efx->type->udp_tnl_del_port(efx, tnl);
2450 }
2451 
2452 static const struct net_device_ops efx_netdev_ops = {
2453 	.ndo_open		= efx_net_open,
2454 	.ndo_stop		= efx_net_stop,
2455 	.ndo_get_stats64	= efx_net_stats,
2456 	.ndo_tx_timeout		= efx_watchdog,
2457 	.ndo_start_xmit		= efx_hard_start_xmit,
2458 	.ndo_validate_addr	= eth_validate_addr,
2459 	.ndo_do_ioctl		= efx_ioctl,
2460 	.ndo_change_mtu		= efx_change_mtu,
2461 	.ndo_set_mac_address	= efx_set_mac_address,
2462 	.ndo_set_rx_mode	= efx_set_rx_mode,
2463 	.ndo_set_features	= efx_set_features,
2464 	.ndo_vlan_rx_add_vid	= efx_vlan_rx_add_vid,
2465 	.ndo_vlan_rx_kill_vid	= efx_vlan_rx_kill_vid,
2466 #ifdef CONFIG_SFC_SRIOV
2467 	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
2468 	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
2469 	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
2470 	.ndo_get_vf_config	= efx_sriov_get_vf_config,
2471 	.ndo_set_vf_link_state  = efx_sriov_set_vf_link_state,
2472 #endif
2473 	.ndo_get_phys_port_id   = efx_get_phys_port_id,
2474 	.ndo_get_phys_port_name	= efx_get_phys_port_name,
2475 #ifdef CONFIG_NET_POLL_CONTROLLER
2476 	.ndo_poll_controller = efx_netpoll,
2477 #endif
2478 	.ndo_setup_tc		= efx_setup_tc,
2479 #ifdef CONFIG_RFS_ACCEL
2480 	.ndo_rx_flow_steer	= efx_filter_rfs,
2481 #endif
2482 	.ndo_udp_tunnel_add	= efx_udp_tunnel_add,
2483 	.ndo_udp_tunnel_del	= efx_udp_tunnel_del,
2484 };
2485 
2486 static void efx_update_name(struct efx_nic *efx)
2487 {
2488 	strcpy(efx->name, efx->net_dev->name);
2489 	efx_mtd_rename(efx);
2490 	efx_set_channel_names(efx);
2491 }
2492 
2493 static int efx_netdev_event(struct notifier_block *this,
2494 			    unsigned long event, void *ptr)
2495 {
2496 	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2497 
2498 	if ((net_dev->netdev_ops == &efx_netdev_ops) &&
2499 	    event == NETDEV_CHANGENAME)
2500 		efx_update_name(netdev_priv(net_dev));
2501 
2502 	return NOTIFY_DONE;
2503 }
2504 
2505 static struct notifier_block efx_netdev_notifier = {
2506 	.notifier_call = efx_netdev_event,
2507 };
2508 
2509 static ssize_t
2510 show_phy_type(struct device *dev, struct device_attribute *attr, char *buf)
2511 {
2512 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2513 	return sprintf(buf, "%d\n", efx->phy_type);
2514 }
2515 static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
2516 
2517 #ifdef CONFIG_SFC_MCDI_LOGGING
2518 static ssize_t show_mcdi_log(struct device *dev, struct device_attribute *attr,
2519 			     char *buf)
2520 {
2521 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2522 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2523 
2524 	return scnprintf(buf, PAGE_SIZE, "%d\n", mcdi->logging_enabled);
2525 }
2526 static ssize_t set_mcdi_log(struct device *dev, struct device_attribute *attr,
2527 			    const char *buf, size_t count)
2528 {
2529 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2530 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2531 	bool enable = count > 0 && *buf != '0';
2532 
2533 	mcdi->logging_enabled = enable;
2534 	return count;
2535 }
2536 static DEVICE_ATTR(mcdi_logging, 0644, show_mcdi_log, set_mcdi_log);
2537 #endif
2538 
2539 static int efx_register_netdev(struct efx_nic *efx)
2540 {
2541 	struct net_device *net_dev = efx->net_dev;
2542 	struct efx_channel *channel;
2543 	int rc;
2544 
2545 	net_dev->watchdog_timeo = 5 * HZ;
2546 	net_dev->irq = efx->pci_dev->irq;
2547 	net_dev->netdev_ops = &efx_netdev_ops;
2548 	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0)
2549 		net_dev->priv_flags |= IFF_UNICAST_FLT;
2550 	net_dev->ethtool_ops = &efx_ethtool_ops;
2551 	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2552 	net_dev->min_mtu = EFX_MIN_MTU;
2553 	net_dev->max_mtu = EFX_MAX_MTU;
2554 
2555 	rtnl_lock();
2556 
2557 	/* Enable resets to be scheduled and check whether any were
2558 	 * already requested.  If so, the NIC is probably hosed so we
2559 	 * abort.
2560 	 */
2561 	efx->state = STATE_READY;
2562 	smp_mb(); /* ensure we change state before checking reset_pending */
2563 	if (efx->reset_pending) {
2564 		netif_err(efx, probe, efx->net_dev,
2565 			  "aborting probe due to scheduled reset\n");
2566 		rc = -EIO;
2567 		goto fail_locked;
2568 	}
2569 
2570 	rc = dev_alloc_name(net_dev, net_dev->name);
2571 	if (rc < 0)
2572 		goto fail_locked;
2573 	efx_update_name(efx);
2574 
2575 	/* Always start with carrier off; PHY events will detect the link */
2576 	netif_carrier_off(net_dev);
2577 
2578 	rc = register_netdevice(net_dev);
2579 	if (rc)
2580 		goto fail_locked;
2581 
2582 	efx_for_each_channel(channel, efx) {
2583 		struct efx_tx_queue *tx_queue;
2584 		efx_for_each_channel_tx_queue(tx_queue, channel)
2585 			efx_init_tx_queue_core_txq(tx_queue);
2586 	}
2587 
2588 	efx_associate(efx);
2589 
2590 	rtnl_unlock();
2591 
2592 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2593 	if (rc) {
2594 		netif_err(efx, drv, efx->net_dev,
2595 			  "failed to init net dev attributes\n");
2596 		goto fail_registered;
2597 	}
2598 #ifdef CONFIG_SFC_MCDI_LOGGING
2599 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2600 	if (rc) {
2601 		netif_err(efx, drv, efx->net_dev,
2602 			  "failed to init net dev attributes\n");
2603 		goto fail_attr_mcdi_logging;
2604 	}
2605 #endif
2606 
2607 	return 0;
2608 
2609 #ifdef CONFIG_SFC_MCDI_LOGGING
2610 fail_attr_mcdi_logging:
2611 	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2612 #endif
2613 fail_registered:
2614 	rtnl_lock();
2615 	efx_dissociate(efx);
2616 	unregister_netdevice(net_dev);
2617 fail_locked:
2618 	efx->state = STATE_UNINIT;
2619 	rtnl_unlock();
2620 	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2621 	return rc;
2622 }
2623 
2624 static void efx_unregister_netdev(struct efx_nic *efx)
2625 {
2626 	if (!efx->net_dev)
2627 		return;
2628 
2629 	BUG_ON(netdev_priv(efx->net_dev) != efx);
2630 
2631 	if (efx_dev_registered(efx)) {
2632 		strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
2633 #ifdef CONFIG_SFC_MCDI_LOGGING
2634 		device_remove_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2635 #endif
2636 		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2637 		unregister_netdev(efx->net_dev);
2638 	}
2639 }
2640 
2641 /**************************************************************************
2642  *
2643  * Device reset and suspend
2644  *
2645  **************************************************************************/
2646 
2647 /* Tears down the entire software state and most of the hardware state
2648  * before reset.  */
2649 void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2650 {
2651 	EFX_ASSERT_RESET_SERIALISED(efx);
2652 
2653 	if (method == RESET_TYPE_MCDI_TIMEOUT)
2654 		efx->type->prepare_flr(efx);
2655 
2656 	efx_stop_all(efx);
2657 	efx_disable_interrupts(efx);
2658 
2659 	mutex_lock(&efx->mac_lock);
2660 	mutex_lock(&efx->rss_lock);
2661 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2662 	    method != RESET_TYPE_DATAPATH)
2663 		efx->phy_op->fini(efx);
2664 	efx->type->fini(efx);
2665 }
2666 
2667 /* This function will always ensure that the locks acquired in
2668  * efx_reset_down() are released. A failure return code indicates
2669  * that we were unable to reinitialise the hardware, and the
2670  * driver should be disabled. If ok is false, then the rx and tx
2671  * engines are not restarted, pending a RESET_DISABLE. */
2672 int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2673 {
2674 	int rc;
2675 
2676 	EFX_ASSERT_RESET_SERIALISED(efx);
2677 
2678 	if (method == RESET_TYPE_MCDI_TIMEOUT)
2679 		efx->type->finish_flr(efx);
2680 
2681 	/* Ensure that SRAM is initialised even if we're disabling the device */
2682 	rc = efx->type->init(efx);
2683 	if (rc) {
2684 		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2685 		goto fail;
2686 	}
2687 
2688 	if (!ok)
2689 		goto fail;
2690 
2691 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2692 	    method != RESET_TYPE_DATAPATH) {
2693 		rc = efx->phy_op->init(efx);
2694 		if (rc)
2695 			goto fail;
2696 		rc = efx->phy_op->reconfigure(efx);
2697 		if (rc && rc != -EPERM)
2698 			netif_err(efx, drv, efx->net_dev,
2699 				  "could not restore PHY settings\n");
2700 	}
2701 
2702 	rc = efx_enable_interrupts(efx);
2703 	if (rc)
2704 		goto fail;
2705 
2706 #ifdef CONFIG_SFC_SRIOV
2707 	rc = efx->type->vswitching_restore(efx);
2708 	if (rc) /* not fatal; the PF will still work fine */
2709 		netif_warn(efx, probe, efx->net_dev,
2710 			   "failed to restore vswitching rc=%d;"
2711 			   " VFs may not function\n", rc);
2712 #endif
2713 
2714 	if (efx->type->rx_restore_rss_contexts)
2715 		efx->type->rx_restore_rss_contexts(efx);
2716 	mutex_unlock(&efx->rss_lock);
2717 	down_read(&efx->filter_sem);
2718 	efx_restore_filters(efx);
2719 	up_read(&efx->filter_sem);
2720 	if (efx->type->sriov_reset)
2721 		efx->type->sriov_reset(efx);
2722 
2723 	mutex_unlock(&efx->mac_lock);
2724 
2725 	efx_start_all(efx);
2726 
2727 	if (efx->type->udp_tnl_push_ports)
2728 		efx->type->udp_tnl_push_ports(efx);
2729 
2730 	return 0;
2731 
2732 fail:
2733 	efx->port_initialized = false;
2734 
2735 	mutex_unlock(&efx->rss_lock);
2736 	mutex_unlock(&efx->mac_lock);
2737 
2738 	return rc;
2739 }
2740 
2741 /* Reset the NIC using the specified method.  Note that the reset may
2742  * fail, in which case the card will be left in an unusable state.
2743  *
2744  * Caller must hold the rtnl_lock.
2745  */
2746 int efx_reset(struct efx_nic *efx, enum reset_type method)
2747 {
2748 	int rc, rc2;
2749 	bool disabled;
2750 
2751 	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
2752 		   RESET_TYPE(method));
2753 
2754 	efx_device_detach_sync(efx);
2755 	efx_reset_down(efx, method);
2756 
2757 	rc = efx->type->reset(efx, method);
2758 	if (rc) {
2759 		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2760 		goto out;
2761 	}
2762 
2763 	/* Clear flags for the scopes we covered.  We assume the NIC and
2764 	 * driver are now quiescent so that there is no race here.
2765 	 */
2766 	if (method < RESET_TYPE_MAX_METHOD)
2767 		efx->reset_pending &= -(1 << (method + 1));
2768 	else /* it doesn't fit into the well-ordered scope hierarchy */
2769 		__clear_bit(method, &efx->reset_pending);
2770 
2771 	/* Reinitialise bus-mastering, which may have been turned off before
2772 	 * the reset was scheduled. This is still appropriate, even in the
2773 	 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
2774 	 * can respond to requests. */
2775 	pci_set_master(efx->pci_dev);
2776 
2777 out:
2778 	/* Leave device stopped if necessary */
2779 	disabled = rc ||
2780 		method == RESET_TYPE_DISABLE ||
2781 		method == RESET_TYPE_RECOVER_OR_DISABLE;
2782 	rc2 = efx_reset_up(efx, method, !disabled);
2783 	if (rc2) {
2784 		disabled = true;
2785 		if (!rc)
2786 			rc = rc2;
2787 	}
2788 
2789 	if (disabled) {
2790 		dev_close(efx->net_dev);
2791 		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2792 		efx->state = STATE_DISABLED;
2793 	} else {
2794 		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2795 		efx_device_attach_if_not_resetting(efx);
2796 	}
2797 	return rc;
2798 }
2799 
2800 /* Try recovery mechanisms.
2801  * For now only EEH is supported.
2802  * Returns 0 if the recovery mechanisms are unsuccessful.
2803  * Returns a non-zero value otherwise.
2804  */
2805 int efx_try_recovery(struct efx_nic *efx)
2806 {
2807 #ifdef CONFIG_EEH
2808 	/* A PCI error can occur and not be seen by EEH because nothing
2809 	 * happens on the PCI bus. In this case the driver may fail and
2810 	 * schedule a 'recover or reset', leading to this recovery handler.
2811 	 * Manually call the eeh failure check function.
2812 	 */
2813 	struct eeh_dev *eehdev = pci_dev_to_eeh_dev(efx->pci_dev);
2814 	if (eeh_dev_check_failure(eehdev)) {
2815 		/* The EEH mechanisms will handle the error and reset the
2816 		 * device if necessary.
2817 		 */
2818 		return 1;
2819 	}
2820 #endif
2821 	return 0;
2822 }
2823 
2824 static void efx_wait_for_bist_end(struct efx_nic *efx)
2825 {
2826 	int i;
2827 
2828 	for (i = 0; i < BIST_WAIT_DELAY_COUNT; ++i) {
2829 		if (efx_mcdi_poll_reboot(efx))
2830 			goto out;
2831 		msleep(BIST_WAIT_DELAY_MS);
2832 	}
2833 
2834 	netif_err(efx, drv, efx->net_dev, "Warning: No MC reboot after BIST mode\n");
2835 out:
2836 	/* Either way unset the BIST flag. If we found no reboot we probably
2837 	 * won't recover, but we should try.
2838 	 */
2839 	efx->mc_bist_for_other_fn = false;
2840 }
2841 
2842 /* The worker thread exists so that code that cannot sleep can
2843  * schedule a reset for later.
2844  */
2845 static void efx_reset_work(struct work_struct *data)
2846 {
2847 	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2848 	unsigned long pending;
2849 	enum reset_type method;
2850 
2851 	pending = READ_ONCE(efx->reset_pending);
2852 	method = fls(pending) - 1;
2853 
2854 	if (method == RESET_TYPE_MC_BIST)
2855 		efx_wait_for_bist_end(efx);
2856 
2857 	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
2858 	     method == RESET_TYPE_RECOVER_OR_ALL) &&
2859 	    efx_try_recovery(efx))
2860 		return;
2861 
2862 	if (!pending)
2863 		return;
2864 
2865 	rtnl_lock();
2866 
2867 	/* We checked the state in efx_schedule_reset() but it may
2868 	 * have changed by now.  Now that we have the RTNL lock,
2869 	 * it cannot change again.
2870 	 */
2871 	if (efx->state == STATE_READY)
2872 		(void)efx_reset(efx, method);
2873 
2874 	rtnl_unlock();
2875 }
2876 
2877 void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
2878 {
2879 	enum reset_type method;
2880 
2881 	if (efx->state == STATE_RECOVERY) {
2882 		netif_dbg(efx, drv, efx->net_dev,
2883 			  "recovering: skip scheduling %s reset\n",
2884 			  RESET_TYPE(type));
2885 		return;
2886 	}
2887 
2888 	switch (type) {
2889 	case RESET_TYPE_INVISIBLE:
2890 	case RESET_TYPE_ALL:
2891 	case RESET_TYPE_RECOVER_OR_ALL:
2892 	case RESET_TYPE_WORLD:
2893 	case RESET_TYPE_DISABLE:
2894 	case RESET_TYPE_RECOVER_OR_DISABLE:
2895 	case RESET_TYPE_DATAPATH:
2896 	case RESET_TYPE_MC_BIST:
2897 	case RESET_TYPE_MCDI_TIMEOUT:
2898 		method = type;
2899 		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
2900 			  RESET_TYPE(method));
2901 		break;
2902 	default:
2903 		method = efx->type->map_reset_reason(type);
2904 		netif_dbg(efx, drv, efx->net_dev,
2905 			  "scheduling %s reset for %s\n",
2906 			  RESET_TYPE(method), RESET_TYPE(type));
2907 		break;
2908 	}
2909 
2910 	set_bit(method, &efx->reset_pending);
2911 	smp_mb(); /* ensure we change reset_pending before checking state */
2912 
2913 	/* If we're not READY then just leave the flags set as the cue
2914 	 * to abort probing or reschedule the reset later.
2915 	 */
2916 	if (READ_ONCE(efx->state) != STATE_READY)
2917 		return;
2918 
2919 	/* efx_process_channel() will no longer read events once a
2920 	 * reset is scheduled. So switch back to poll'd MCDI completions. */
2921 	efx_mcdi_mode_poll(efx);
2922 
2923 	queue_work(reset_workqueue, &efx->reset_work);
2924 }
2925 
2926 /**************************************************************************
2927  *
2928  * List of NICs we support
2929  *
2930  **************************************************************************/
2931 
2932 /* PCI device ID table */
2933 static const struct pci_device_id efx_pci_table[] = {
2934 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2935 	 .driver_data = (unsigned long) &siena_a0_nic_type},
2936 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2937 	 .driver_data = (unsigned long) &siena_a0_nic_type},
2938 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
2939 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2940 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903),  /* SFC9120 VF */
2941 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2942 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923),  /* SFC9140 PF */
2943 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2944 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923),  /* SFC9140 VF */
2945 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2946 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03),  /* SFC9220 PF */
2947 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2948 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03),  /* SFC9220 VF */
2949 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2950 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0b03),  /* SFC9250 PF */
2951 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2952 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1b03),  /* SFC9250 VF */
2953 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2954 	{0}			/* end of list */
2955 };
2956 
2957 /**************************************************************************
2958  *
2959  * Dummy PHY/MAC operations
2960  *
2961  * Can be used for some unimplemented operations
2962  * Needed so all function pointers are valid and do not have to be tested
2963  * before use
2964  *
2965  **************************************************************************/
2966 int efx_port_dummy_op_int(struct efx_nic *efx)
2967 {
2968 	return 0;
2969 }
2970 void efx_port_dummy_op_void(struct efx_nic *efx) {}
2971 
2972 static bool efx_port_dummy_op_poll(struct efx_nic *efx)
2973 {
2974 	return false;
2975 }
2976 
2977 static const struct efx_phy_operations efx_dummy_phy_operations = {
2978 	.init		 = efx_port_dummy_op_int,
2979 	.reconfigure	 = efx_port_dummy_op_int,
2980 	.poll		 = efx_port_dummy_op_poll,
2981 	.fini		 = efx_port_dummy_op_void,
2982 };
2983 
2984 /**************************************************************************
2985  *
2986  * Data housekeeping
2987  *
2988  **************************************************************************/
2989 
2990 /* This zeroes out and then fills in the invariants in a struct
2991  * efx_nic (including all sub-structures).
2992  */
2993 static int efx_init_struct(struct efx_nic *efx,
2994 			   struct pci_dev *pci_dev, struct net_device *net_dev)
2995 {
2996 	int rc = -ENOMEM, i;
2997 
2998 	/* Initialise common structures */
2999 	INIT_LIST_HEAD(&efx->node);
3000 	INIT_LIST_HEAD(&efx->secondary_list);
3001 	spin_lock_init(&efx->biu_lock);
3002 #ifdef CONFIG_SFC_MTD
3003 	INIT_LIST_HEAD(&efx->mtd_list);
3004 #endif
3005 	INIT_WORK(&efx->reset_work, efx_reset_work);
3006 	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
3007 	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
3008 	efx->pci_dev = pci_dev;
3009 	efx->msg_enable = debug;
3010 	efx->state = STATE_UNINIT;
3011 	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
3012 
3013 	efx->net_dev = net_dev;
3014 	efx->rx_prefix_size = efx->type->rx_prefix_size;
3015 	efx->rx_ip_align =
3016 		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
3017 	efx->rx_packet_hash_offset =
3018 		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
3019 	efx->rx_packet_ts_offset =
3020 		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
3021 	INIT_LIST_HEAD(&efx->rss_context.list);
3022 	mutex_init(&efx->rss_lock);
3023 	spin_lock_init(&efx->stats_lock);
3024 	efx->vi_stride = EFX_DEFAULT_VI_STRIDE;
3025 	efx->num_mac_stats = MC_CMD_MAC_NSTATS;
3026 	BUILD_BUG_ON(MC_CMD_MAC_NSTATS - 1 != MC_CMD_MAC_GENERATION_END);
3027 	mutex_init(&efx->mac_lock);
3028 #ifdef CONFIG_RFS_ACCEL
3029 	mutex_init(&efx->rps_mutex);
3030 	spin_lock_init(&efx->rps_hash_lock);
3031 	/* Failure to allocate is not fatal, but may degrade ARFS performance */
3032 	efx->rps_hash_table = kcalloc(EFX_ARFS_HASH_TABLE_SIZE,
3033 				      sizeof(*efx->rps_hash_table), GFP_KERNEL);
3034 #endif
3035 	efx->phy_op = &efx_dummy_phy_operations;
3036 	efx->mdio.dev = net_dev;
3037 	INIT_WORK(&efx->mac_work, efx_mac_work);
3038 	init_waitqueue_head(&efx->flush_wq);
3039 
3040 	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
3041 		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
3042 		if (!efx->channel[i])
3043 			goto fail;
3044 		efx->msi_context[i].efx = efx;
3045 		efx->msi_context[i].index = i;
3046 	}
3047 
3048 	/* Higher numbered interrupt modes are less capable! */
3049 	if (WARN_ON_ONCE(efx->type->max_interrupt_mode >
3050 			 efx->type->min_interrupt_mode)) {
3051 		rc = -EIO;
3052 		goto fail;
3053 	}
3054 	efx->interrupt_mode = max(efx->type->max_interrupt_mode,
3055 				  interrupt_mode);
3056 	efx->interrupt_mode = min(efx->type->min_interrupt_mode,
3057 				  interrupt_mode);
3058 
3059 	/* Would be good to use the net_dev name, but we're too early */
3060 	snprintf(efx->workqueue_name, sizeof(efx->workqueue_name), "sfc%s",
3061 		 pci_name(pci_dev));
3062 	efx->workqueue = create_singlethread_workqueue(efx->workqueue_name);
3063 	if (!efx->workqueue)
3064 		goto fail;
3065 
3066 	return 0;
3067 
3068 fail:
3069 	efx_fini_struct(efx);
3070 	return rc;
3071 }
3072 
3073 static void efx_fini_struct(struct efx_nic *efx)
3074 {
3075 	int i;
3076 
3077 #ifdef CONFIG_RFS_ACCEL
3078 	kfree(efx->rps_hash_table);
3079 #endif
3080 
3081 	for (i = 0; i < EFX_MAX_CHANNELS; i++)
3082 		kfree(efx->channel[i]);
3083 
3084 	kfree(efx->vpd_sn);
3085 
3086 	if (efx->workqueue) {
3087 		destroy_workqueue(efx->workqueue);
3088 		efx->workqueue = NULL;
3089 	}
3090 }
3091 
3092 void efx_update_sw_stats(struct efx_nic *efx, u64 *stats)
3093 {
3094 	u64 n_rx_nodesc_trunc = 0;
3095 	struct efx_channel *channel;
3096 
3097 	efx_for_each_channel(channel, efx)
3098 		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
3099 	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
3100 	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
3101 }
3102 
3103 bool efx_filter_spec_equal(const struct efx_filter_spec *left,
3104 			   const struct efx_filter_spec *right)
3105 {
3106 	if ((left->match_flags ^ right->match_flags) |
3107 	    ((left->flags ^ right->flags) &
3108 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
3109 		return false;
3110 
3111 	return memcmp(&left->outer_vid, &right->outer_vid,
3112 		      sizeof(struct efx_filter_spec) -
3113 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
3114 }
3115 
3116 u32 efx_filter_spec_hash(const struct efx_filter_spec *spec)
3117 {
3118 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
3119 	return jhash2((const u32 *)&spec->outer_vid,
3120 		      (sizeof(struct efx_filter_spec) -
3121 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
3122 		      0);
3123 }
3124 
3125 #ifdef CONFIG_RFS_ACCEL
3126 bool efx_rps_check_rule(struct efx_arfs_rule *rule, unsigned int filter_idx,
3127 			bool *force)
3128 {
3129 	if (rule->filter_id == EFX_ARFS_FILTER_ID_PENDING) {
3130 		/* ARFS is currently updating this entry, leave it */
3131 		return false;
3132 	}
3133 	if (rule->filter_id == EFX_ARFS_FILTER_ID_ERROR) {
3134 		/* ARFS tried and failed to update this, so it's probably out
3135 		 * of date.  Remove the filter and the ARFS rule entry.
3136 		 */
3137 		rule->filter_id = EFX_ARFS_FILTER_ID_REMOVING;
3138 		*force = true;
3139 		return true;
3140 	} else if (WARN_ON(rule->filter_id != filter_idx)) { /* can't happen */
3141 		/* ARFS has moved on, so old filter is not needed.  Since we did
3142 		 * not mark the rule with EFX_ARFS_FILTER_ID_REMOVING, it will
3143 		 * not be removed by efx_rps_hash_del() subsequently.
3144 		 */
3145 		*force = true;
3146 		return true;
3147 	}
3148 	/* Remove it iff ARFS wants to. */
3149 	return true;
3150 }
3151 
3152 struct hlist_head *efx_rps_hash_bucket(struct efx_nic *efx,
3153 				       const struct efx_filter_spec *spec)
3154 {
3155 	u32 hash = efx_filter_spec_hash(spec);
3156 
3157 	WARN_ON(!spin_is_locked(&efx->rps_hash_lock));
3158 	if (!efx->rps_hash_table)
3159 		return NULL;
3160 	return &efx->rps_hash_table[hash % EFX_ARFS_HASH_TABLE_SIZE];
3161 }
3162 
3163 struct efx_arfs_rule *efx_rps_hash_find(struct efx_nic *efx,
3164 					const struct efx_filter_spec *spec)
3165 {
3166 	struct efx_arfs_rule *rule;
3167 	struct hlist_head *head;
3168 	struct hlist_node *node;
3169 
3170 	head = efx_rps_hash_bucket(efx, spec);
3171 	if (!head)
3172 		return NULL;
3173 	hlist_for_each(node, head) {
3174 		rule = container_of(node, struct efx_arfs_rule, node);
3175 		if (efx_filter_spec_equal(spec, &rule->spec))
3176 			return rule;
3177 	}
3178 	return NULL;
3179 }
3180 
3181 struct efx_arfs_rule *efx_rps_hash_add(struct efx_nic *efx,
3182 				       const struct efx_filter_spec *spec,
3183 				       bool *new)
3184 {
3185 	struct efx_arfs_rule *rule;
3186 	struct hlist_head *head;
3187 	struct hlist_node *node;
3188 
3189 	head = efx_rps_hash_bucket(efx, spec);
3190 	if (!head)
3191 		return NULL;
3192 	hlist_for_each(node, head) {
3193 		rule = container_of(node, struct efx_arfs_rule, node);
3194 		if (efx_filter_spec_equal(spec, &rule->spec)) {
3195 			*new = false;
3196 			return rule;
3197 		}
3198 	}
3199 	rule = kmalloc(sizeof(*rule), GFP_ATOMIC);
3200 	*new = true;
3201 	if (rule) {
3202 		memcpy(&rule->spec, spec, sizeof(rule->spec));
3203 		hlist_add_head(&rule->node, head);
3204 	}
3205 	return rule;
3206 }
3207 
3208 void efx_rps_hash_del(struct efx_nic *efx, const struct efx_filter_spec *spec)
3209 {
3210 	struct efx_arfs_rule *rule;
3211 	struct hlist_head *head;
3212 	struct hlist_node *node;
3213 
3214 	head = efx_rps_hash_bucket(efx, spec);
3215 	if (WARN_ON(!head))
3216 		return;
3217 	hlist_for_each(node, head) {
3218 		rule = container_of(node, struct efx_arfs_rule, node);
3219 		if (efx_filter_spec_equal(spec, &rule->spec)) {
3220 			/* Someone already reused the entry.  We know that if
3221 			 * this check doesn't fire (i.e. filter_id == REMOVING)
3222 			 * then the REMOVING mark was put there by our caller,
3223 			 * because caller is holding a lock on filter table and
3224 			 * only holders of that lock set REMOVING.
3225 			 */
3226 			if (rule->filter_id != EFX_ARFS_FILTER_ID_REMOVING)
3227 				return;
3228 			hlist_del(node);
3229 			kfree(rule);
3230 			return;
3231 		}
3232 	}
3233 	/* We didn't find it. */
3234 	WARN_ON(1);
3235 }
3236 #endif
3237 
3238 /* RSS contexts.  We're using linked lists and crappy O(n) algorithms, because
3239  * (a) this is an infrequent control-plane operation and (b) n is small (max 64)
3240  */
3241 struct efx_rss_context *efx_alloc_rss_context_entry(struct efx_nic *efx)
3242 {
3243 	struct list_head *head = &efx->rss_context.list;
3244 	struct efx_rss_context *ctx, *new;
3245 	u32 id = 1; /* Don't use zero, that refers to the master RSS context */
3246 
3247 	WARN_ON(!mutex_is_locked(&efx->rss_lock));
3248 
3249 	/* Search for first gap in the numbering */
3250 	list_for_each_entry(ctx, head, list) {
3251 		if (ctx->user_id != id)
3252 			break;
3253 		id++;
3254 		/* Check for wrap.  If this happens, we have nearly 2^32
3255 		 * allocated RSS contexts, which seems unlikely.
3256 		 */
3257 		if (WARN_ON_ONCE(!id))
3258 			return NULL;
3259 	}
3260 
3261 	/* Create the new entry */
3262 	new = kmalloc(sizeof(struct efx_rss_context), GFP_KERNEL);
3263 	if (!new)
3264 		return NULL;
3265 	new->context_id = EFX_EF10_RSS_CONTEXT_INVALID;
3266 	new->rx_hash_udp_4tuple = false;
3267 
3268 	/* Insert the new entry into the gap */
3269 	new->user_id = id;
3270 	list_add_tail(&new->list, &ctx->list);
3271 	return new;
3272 }
3273 
3274 struct efx_rss_context *efx_find_rss_context_entry(struct efx_nic *efx, u32 id)
3275 {
3276 	struct list_head *head = &efx->rss_context.list;
3277 	struct efx_rss_context *ctx;
3278 
3279 	WARN_ON(!mutex_is_locked(&efx->rss_lock));
3280 
3281 	list_for_each_entry(ctx, head, list)
3282 		if (ctx->user_id == id)
3283 			return ctx;
3284 	return NULL;
3285 }
3286 
3287 void efx_free_rss_context_entry(struct efx_rss_context *ctx)
3288 {
3289 	list_del(&ctx->list);
3290 	kfree(ctx);
3291 }
3292 
3293 /**************************************************************************
3294  *
3295  * PCI interface
3296  *
3297  **************************************************************************/
3298 
3299 /* Main body of final NIC shutdown code
3300  * This is called only at module unload (or hotplug removal).
3301  */
3302 static void efx_pci_remove_main(struct efx_nic *efx)
3303 {
3304 	/* Flush reset_work. It can no longer be scheduled since we
3305 	 * are not READY.
3306 	 */
3307 	BUG_ON(efx->state == STATE_READY);
3308 	cancel_work_sync(&efx->reset_work);
3309 
3310 	efx_disable_interrupts(efx);
3311 	efx_nic_fini_interrupt(efx);
3312 	efx_fini_port(efx);
3313 	efx->type->fini(efx);
3314 	efx_fini_napi(efx);
3315 	efx_remove_all(efx);
3316 }
3317 
3318 /* Final NIC shutdown
3319  * This is called only at module unload (or hotplug removal).  A PF can call
3320  * this on its VFs to ensure they are unbound first.
3321  */
3322 static void efx_pci_remove(struct pci_dev *pci_dev)
3323 {
3324 	struct efx_nic *efx;
3325 
3326 	efx = pci_get_drvdata(pci_dev);
3327 	if (!efx)
3328 		return;
3329 
3330 	/* Mark the NIC as fini, then stop the interface */
3331 	rtnl_lock();
3332 	efx_dissociate(efx);
3333 	dev_close(efx->net_dev);
3334 	efx_disable_interrupts(efx);
3335 	efx->state = STATE_UNINIT;
3336 	rtnl_unlock();
3337 
3338 	if (efx->type->sriov_fini)
3339 		efx->type->sriov_fini(efx);
3340 
3341 	efx_unregister_netdev(efx);
3342 
3343 	efx_mtd_remove(efx);
3344 
3345 	efx_pci_remove_main(efx);
3346 
3347 	efx_fini_io(efx);
3348 	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
3349 
3350 	efx_fini_struct(efx);
3351 	free_netdev(efx->net_dev);
3352 
3353 	pci_disable_pcie_error_reporting(pci_dev);
3354 };
3355 
3356 /* NIC VPD information
3357  * Called during probe to display the part number of the
3358  * installed NIC.  VPD is potentially very large but this should
3359  * always appear within the first 512 bytes.
3360  */
3361 #define SFC_VPD_LEN 512
3362 static void efx_probe_vpd_strings(struct efx_nic *efx)
3363 {
3364 	struct pci_dev *dev = efx->pci_dev;
3365 	char vpd_data[SFC_VPD_LEN];
3366 	ssize_t vpd_size;
3367 	int ro_start, ro_size, i, j;
3368 
3369 	/* Get the vpd data from the device */
3370 	vpd_size = pci_read_vpd(dev, 0, sizeof(vpd_data), vpd_data);
3371 	if (vpd_size <= 0) {
3372 		netif_err(efx, drv, efx->net_dev, "Unable to read VPD\n");
3373 		return;
3374 	}
3375 
3376 	/* Get the Read only section */
3377 	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size, PCI_VPD_LRDT_RO_DATA);
3378 	if (ro_start < 0) {
3379 		netif_err(efx, drv, efx->net_dev, "VPD Read-only not found\n");
3380 		return;
3381 	}
3382 
3383 	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
3384 	j = ro_size;
3385 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3386 	if (i + j > vpd_size)
3387 		j = vpd_size - i;
3388 
3389 	/* Get the Part number */
3390 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "PN");
3391 	if (i < 0) {
3392 		netif_err(efx, drv, efx->net_dev, "Part number not found\n");
3393 		return;
3394 	}
3395 
3396 	j = pci_vpd_info_field_size(&vpd_data[i]);
3397 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3398 	if (i + j > vpd_size) {
3399 		netif_err(efx, drv, efx->net_dev, "Incomplete part number\n");
3400 		return;
3401 	}
3402 
3403 	netif_info(efx, drv, efx->net_dev,
3404 		   "Part Number : %.*s\n", j, &vpd_data[i]);
3405 
3406 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3407 	j = ro_size;
3408 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "SN");
3409 	if (i < 0) {
3410 		netif_err(efx, drv, efx->net_dev, "Serial number not found\n");
3411 		return;
3412 	}
3413 
3414 	j = pci_vpd_info_field_size(&vpd_data[i]);
3415 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3416 	if (i + j > vpd_size) {
3417 		netif_err(efx, drv, efx->net_dev, "Incomplete serial number\n");
3418 		return;
3419 	}
3420 
3421 	efx->vpd_sn = kmalloc(j + 1, GFP_KERNEL);
3422 	if (!efx->vpd_sn)
3423 		return;
3424 
3425 	snprintf(efx->vpd_sn, j + 1, "%s", &vpd_data[i]);
3426 }
3427 
3428 
3429 /* Main body of NIC initialisation
3430  * This is called at module load (or hotplug insertion, theoretically).
3431  */
3432 static int efx_pci_probe_main(struct efx_nic *efx)
3433 {
3434 	int rc;
3435 
3436 	/* Do start-of-day initialisation */
3437 	rc = efx_probe_all(efx);
3438 	if (rc)
3439 		goto fail1;
3440 
3441 	efx_init_napi(efx);
3442 
3443 	rc = efx->type->init(efx);
3444 	if (rc) {
3445 		netif_err(efx, probe, efx->net_dev,
3446 			  "failed to initialise NIC\n");
3447 		goto fail3;
3448 	}
3449 
3450 	rc = efx_init_port(efx);
3451 	if (rc) {
3452 		netif_err(efx, probe, efx->net_dev,
3453 			  "failed to initialise port\n");
3454 		goto fail4;
3455 	}
3456 
3457 	rc = efx_nic_init_interrupt(efx);
3458 	if (rc)
3459 		goto fail5;
3460 	rc = efx_enable_interrupts(efx);
3461 	if (rc)
3462 		goto fail6;
3463 
3464 	return 0;
3465 
3466  fail6:
3467 	efx_nic_fini_interrupt(efx);
3468  fail5:
3469 	efx_fini_port(efx);
3470  fail4:
3471 	efx->type->fini(efx);
3472  fail3:
3473 	efx_fini_napi(efx);
3474 	efx_remove_all(efx);
3475  fail1:
3476 	return rc;
3477 }
3478 
3479 static int efx_pci_probe_post_io(struct efx_nic *efx)
3480 {
3481 	struct net_device *net_dev = efx->net_dev;
3482 	int rc = efx_pci_probe_main(efx);
3483 
3484 	if (rc)
3485 		return rc;
3486 
3487 	if (efx->type->sriov_init) {
3488 		rc = efx->type->sriov_init(efx);
3489 		if (rc)
3490 			netif_err(efx, probe, efx->net_dev,
3491 				  "SR-IOV can't be enabled rc %d\n", rc);
3492 	}
3493 
3494 	/* Determine netdevice features */
3495 	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
3496 			      NETIF_F_TSO | NETIF_F_RXCSUM | NETIF_F_RXALL);
3497 	if (efx->type->offload_features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
3498 		net_dev->features |= NETIF_F_TSO6;
3499 	/* Check whether device supports TSO */
3500 	if (!efx->type->tso_versions || !efx->type->tso_versions(efx))
3501 		net_dev->features &= ~NETIF_F_ALL_TSO;
3502 	/* Mask for features that also apply to VLAN devices */
3503 	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
3504 				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
3505 				   NETIF_F_RXCSUM);
3506 
3507 	net_dev->hw_features |= net_dev->features & ~efx->fixed_features;
3508 
3509 	/* Disable receiving frames with bad FCS, by default. */
3510 	net_dev->features &= ~NETIF_F_RXALL;
3511 
3512 	/* Disable VLAN filtering by default.  It may be enforced if
3513 	 * the feature is fixed (i.e. VLAN filters are required to
3514 	 * receive VLAN tagged packets due to vPort restrictions).
3515 	 */
3516 	net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3517 	net_dev->features |= efx->fixed_features;
3518 
3519 	rc = efx_register_netdev(efx);
3520 	if (!rc)
3521 		return 0;
3522 
3523 	efx_pci_remove_main(efx);
3524 	return rc;
3525 }
3526 
3527 /* NIC initialisation
3528  *
3529  * This is called at module load (or hotplug insertion,
3530  * theoretically).  It sets up PCI mappings, resets the NIC,
3531  * sets up and registers the network devices with the kernel and hooks
3532  * the interrupt service routine.  It does not prepare the device for
3533  * transmission; this is left to the first time one of the network
3534  * interfaces is brought up (i.e. efx_net_open).
3535  */
3536 static int efx_pci_probe(struct pci_dev *pci_dev,
3537 			 const struct pci_device_id *entry)
3538 {
3539 	struct net_device *net_dev;
3540 	struct efx_nic *efx;
3541 	int rc;
3542 
3543 	/* Allocate and initialise a struct net_device and struct efx_nic */
3544 	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
3545 				     EFX_MAX_RX_QUEUES);
3546 	if (!net_dev)
3547 		return -ENOMEM;
3548 	efx = netdev_priv(net_dev);
3549 	efx->type = (const struct efx_nic_type *) entry->driver_data;
3550 	efx->fixed_features |= NETIF_F_HIGHDMA;
3551 
3552 	pci_set_drvdata(pci_dev, efx);
3553 	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
3554 	rc = efx_init_struct(efx, pci_dev, net_dev);
3555 	if (rc)
3556 		goto fail1;
3557 
3558 	netif_info(efx, probe, efx->net_dev,
3559 		   "Solarflare NIC detected\n");
3560 
3561 	if (!efx->type->is_vf)
3562 		efx_probe_vpd_strings(efx);
3563 
3564 	/* Set up basic I/O (BAR mappings etc) */
3565 	rc = efx_init_io(efx);
3566 	if (rc)
3567 		goto fail2;
3568 
3569 	rc = efx_pci_probe_post_io(efx);
3570 	if (rc) {
3571 		/* On failure, retry once immediately.
3572 		 * If we aborted probe due to a scheduled reset, dismiss it.
3573 		 */
3574 		efx->reset_pending = 0;
3575 		rc = efx_pci_probe_post_io(efx);
3576 		if (rc) {
3577 			/* On another failure, retry once more
3578 			 * after a 50-305ms delay.
3579 			 */
3580 			unsigned char r;
3581 
3582 			get_random_bytes(&r, 1);
3583 			msleep((unsigned int)r + 50);
3584 			efx->reset_pending = 0;
3585 			rc = efx_pci_probe_post_io(efx);
3586 		}
3587 	}
3588 	if (rc)
3589 		goto fail3;
3590 
3591 	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
3592 
3593 	/* Try to create MTDs, but allow this to fail */
3594 	rtnl_lock();
3595 	rc = efx_mtd_probe(efx);
3596 	rtnl_unlock();
3597 	if (rc && rc != -EPERM)
3598 		netif_warn(efx, probe, efx->net_dev,
3599 			   "failed to create MTDs (%d)\n", rc);
3600 
3601 	rc = pci_enable_pcie_error_reporting(pci_dev);
3602 	if (rc && rc != -EINVAL)
3603 		netif_notice(efx, probe, efx->net_dev,
3604 			     "PCIE error reporting unavailable (%d).\n",
3605 			     rc);
3606 
3607 	if (efx->type->udp_tnl_push_ports)
3608 		efx->type->udp_tnl_push_ports(efx);
3609 
3610 	return 0;
3611 
3612  fail3:
3613 	efx_fini_io(efx);
3614  fail2:
3615 	efx_fini_struct(efx);
3616  fail1:
3617 	WARN_ON(rc > 0);
3618 	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
3619 	free_netdev(net_dev);
3620 	return rc;
3621 }
3622 
3623 /* efx_pci_sriov_configure returns the actual number of Virtual Functions
3624  * enabled on success
3625  */
3626 #ifdef CONFIG_SFC_SRIOV
3627 static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
3628 {
3629 	int rc;
3630 	struct efx_nic *efx = pci_get_drvdata(dev);
3631 
3632 	if (efx->type->sriov_configure) {
3633 		rc = efx->type->sriov_configure(efx, num_vfs);
3634 		if (rc)
3635 			return rc;
3636 		else
3637 			return num_vfs;
3638 	} else
3639 		return -EOPNOTSUPP;
3640 }
3641 #endif
3642 
3643 static int efx_pm_freeze(struct device *dev)
3644 {
3645 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3646 
3647 	rtnl_lock();
3648 
3649 	if (efx->state != STATE_DISABLED) {
3650 		efx->state = STATE_UNINIT;
3651 
3652 		efx_device_detach_sync(efx);
3653 
3654 		efx_stop_all(efx);
3655 		efx_disable_interrupts(efx);
3656 	}
3657 
3658 	rtnl_unlock();
3659 
3660 	return 0;
3661 }
3662 
3663 static int efx_pm_thaw(struct device *dev)
3664 {
3665 	int rc;
3666 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3667 
3668 	rtnl_lock();
3669 
3670 	if (efx->state != STATE_DISABLED) {
3671 		rc = efx_enable_interrupts(efx);
3672 		if (rc)
3673 			goto fail;
3674 
3675 		mutex_lock(&efx->mac_lock);
3676 		efx->phy_op->reconfigure(efx);
3677 		mutex_unlock(&efx->mac_lock);
3678 
3679 		efx_start_all(efx);
3680 
3681 		efx_device_attach_if_not_resetting(efx);
3682 
3683 		efx->state = STATE_READY;
3684 
3685 		efx->type->resume_wol(efx);
3686 	}
3687 
3688 	rtnl_unlock();
3689 
3690 	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
3691 	queue_work(reset_workqueue, &efx->reset_work);
3692 
3693 	return 0;
3694 
3695 fail:
3696 	rtnl_unlock();
3697 
3698 	return rc;
3699 }
3700 
3701 static int efx_pm_poweroff(struct device *dev)
3702 {
3703 	struct pci_dev *pci_dev = to_pci_dev(dev);
3704 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3705 
3706 	efx->type->fini(efx);
3707 
3708 	efx->reset_pending = 0;
3709 
3710 	pci_save_state(pci_dev);
3711 	return pci_set_power_state(pci_dev, PCI_D3hot);
3712 }
3713 
3714 /* Used for both resume and restore */
3715 static int efx_pm_resume(struct device *dev)
3716 {
3717 	struct pci_dev *pci_dev = to_pci_dev(dev);
3718 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3719 	int rc;
3720 
3721 	rc = pci_set_power_state(pci_dev, PCI_D0);
3722 	if (rc)
3723 		return rc;
3724 	pci_restore_state(pci_dev);
3725 	rc = pci_enable_device(pci_dev);
3726 	if (rc)
3727 		return rc;
3728 	pci_set_master(efx->pci_dev);
3729 	rc = efx->type->reset(efx, RESET_TYPE_ALL);
3730 	if (rc)
3731 		return rc;
3732 	rc = efx->type->init(efx);
3733 	if (rc)
3734 		return rc;
3735 	rc = efx_pm_thaw(dev);
3736 	return rc;
3737 }
3738 
3739 static int efx_pm_suspend(struct device *dev)
3740 {
3741 	int rc;
3742 
3743 	efx_pm_freeze(dev);
3744 	rc = efx_pm_poweroff(dev);
3745 	if (rc)
3746 		efx_pm_resume(dev);
3747 	return rc;
3748 }
3749 
3750 static const struct dev_pm_ops efx_pm_ops = {
3751 	.suspend	= efx_pm_suspend,
3752 	.resume		= efx_pm_resume,
3753 	.freeze		= efx_pm_freeze,
3754 	.thaw		= efx_pm_thaw,
3755 	.poweroff	= efx_pm_poweroff,
3756 	.restore	= efx_pm_resume,
3757 };
3758 
3759 /* A PCI error affecting this device was detected.
3760  * At this point MMIO and DMA may be disabled.
3761  * Stop the software path and request a slot reset.
3762  */
3763 static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
3764 					      enum pci_channel_state state)
3765 {
3766 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3767 	struct efx_nic *efx = pci_get_drvdata(pdev);
3768 
3769 	if (state == pci_channel_io_perm_failure)
3770 		return PCI_ERS_RESULT_DISCONNECT;
3771 
3772 	rtnl_lock();
3773 
3774 	if (efx->state != STATE_DISABLED) {
3775 		efx->state = STATE_RECOVERY;
3776 		efx->reset_pending = 0;
3777 
3778 		efx_device_detach_sync(efx);
3779 
3780 		efx_stop_all(efx);
3781 		efx_disable_interrupts(efx);
3782 
3783 		status = PCI_ERS_RESULT_NEED_RESET;
3784 	} else {
3785 		/* If the interface is disabled we don't want to do anything
3786 		 * with it.
3787 		 */
3788 		status = PCI_ERS_RESULT_RECOVERED;
3789 	}
3790 
3791 	rtnl_unlock();
3792 
3793 	pci_disable_device(pdev);
3794 
3795 	return status;
3796 }
3797 
3798 /* Fake a successful reset, which will be performed later in efx_io_resume. */
3799 static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
3800 {
3801 	struct efx_nic *efx = pci_get_drvdata(pdev);
3802 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3803 	int rc;
3804 
3805 	if (pci_enable_device(pdev)) {
3806 		netif_err(efx, hw, efx->net_dev,
3807 			  "Cannot re-enable PCI device after reset.\n");
3808 		status =  PCI_ERS_RESULT_DISCONNECT;
3809 	}
3810 
3811 	rc = pci_cleanup_aer_uncorrect_error_status(pdev);
3812 	if (rc) {
3813 		netif_err(efx, hw, efx->net_dev,
3814 		"pci_cleanup_aer_uncorrect_error_status failed (%d)\n", rc);
3815 		/* Non-fatal error. Continue. */
3816 	}
3817 
3818 	return status;
3819 }
3820 
3821 /* Perform the actual reset and resume I/O operations. */
3822 static void efx_io_resume(struct pci_dev *pdev)
3823 {
3824 	struct efx_nic *efx = pci_get_drvdata(pdev);
3825 	int rc;
3826 
3827 	rtnl_lock();
3828 
3829 	if (efx->state == STATE_DISABLED)
3830 		goto out;
3831 
3832 	rc = efx_reset(efx, RESET_TYPE_ALL);
3833 	if (rc) {
3834 		netif_err(efx, hw, efx->net_dev,
3835 			  "efx_reset failed after PCI error (%d)\n", rc);
3836 	} else {
3837 		efx->state = STATE_READY;
3838 		netif_dbg(efx, hw, efx->net_dev,
3839 			  "Done resetting and resuming IO after PCI error.\n");
3840 	}
3841 
3842 out:
3843 	rtnl_unlock();
3844 }
3845 
3846 /* For simplicity and reliability, we always require a slot reset and try to
3847  * reset the hardware when a pci error affecting the device is detected.
3848  * We leave both the link_reset and mmio_enabled callback unimplemented:
3849  * with our request for slot reset the mmio_enabled callback will never be
3850  * called, and the link_reset callback is not used by AER or EEH mechanisms.
3851  */
3852 static const struct pci_error_handlers efx_err_handlers = {
3853 	.error_detected = efx_io_error_detected,
3854 	.slot_reset	= efx_io_slot_reset,
3855 	.resume		= efx_io_resume,
3856 };
3857 
3858 static struct pci_driver efx_pci_driver = {
3859 	.name		= KBUILD_MODNAME,
3860 	.id_table	= efx_pci_table,
3861 	.probe		= efx_pci_probe,
3862 	.remove		= efx_pci_remove,
3863 	.driver.pm	= &efx_pm_ops,
3864 	.err_handler	= &efx_err_handlers,
3865 #ifdef CONFIG_SFC_SRIOV
3866 	.sriov_configure = efx_pci_sriov_configure,
3867 #endif
3868 };
3869 
3870 /**************************************************************************
3871  *
3872  * Kernel module interface
3873  *
3874  *************************************************************************/
3875 
3876 module_param(interrupt_mode, uint, 0444);
3877 MODULE_PARM_DESC(interrupt_mode,
3878 		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
3879 
3880 static int __init efx_init_module(void)
3881 {
3882 	int rc;
3883 
3884 	printk(KERN_INFO "Solarflare NET driver v" EFX_DRIVER_VERSION "\n");
3885 
3886 	rc = register_netdevice_notifier(&efx_netdev_notifier);
3887 	if (rc)
3888 		goto err_notifier;
3889 
3890 #ifdef CONFIG_SFC_SRIOV
3891 	rc = efx_init_sriov();
3892 	if (rc)
3893 		goto err_sriov;
3894 #endif
3895 
3896 	reset_workqueue = create_singlethread_workqueue("sfc_reset");
3897 	if (!reset_workqueue) {
3898 		rc = -ENOMEM;
3899 		goto err_reset;
3900 	}
3901 
3902 	rc = pci_register_driver(&efx_pci_driver);
3903 	if (rc < 0)
3904 		goto err_pci;
3905 
3906 	return 0;
3907 
3908  err_pci:
3909 	destroy_workqueue(reset_workqueue);
3910  err_reset:
3911 #ifdef CONFIG_SFC_SRIOV
3912 	efx_fini_sriov();
3913  err_sriov:
3914 #endif
3915 	unregister_netdevice_notifier(&efx_netdev_notifier);
3916  err_notifier:
3917 	return rc;
3918 }
3919 
3920 static void __exit efx_exit_module(void)
3921 {
3922 	printk(KERN_INFO "Solarflare NET driver unloading\n");
3923 
3924 	pci_unregister_driver(&efx_pci_driver);
3925 	destroy_workqueue(reset_workqueue);
3926 #ifdef CONFIG_SFC_SRIOV
3927 	efx_fini_sriov();
3928 #endif
3929 	unregister_netdevice_notifier(&efx_netdev_notifier);
3930 
3931 }
3932 
3933 module_init(efx_init_module);
3934 module_exit(efx_exit_module);
3935 
3936 MODULE_AUTHOR("Solarflare Communications and "
3937 	      "Michael Brown <mbrown@fensystems.co.uk>");
3938 MODULE_DESCRIPTION("Solarflare network driver");
3939 MODULE_LICENSE("GPL");
3940 MODULE_DEVICE_TABLE(pci, efx_pci_table);
3941 MODULE_VERSION(EFX_DRIVER_VERSION);
3942