xref: /linux/drivers/net/ethernet/sfc/falcon/efx.c (revision 32786fdc9506aeba98278c1844d4bfb766863832)
1 /****************************************************************************
2  * Driver for Solarflare network controllers and boards
3  * Copyright 2005-2006 Fen Systems Ltd.
4  * Copyright 2005-2013 Solarflare Communications Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License version 2 as published
8  * by the Free Software Foundation, incorporated herein by reference.
9  */
10 
11 #include <linux/module.h>
12 #include <linux/pci.h>
13 #include <linux/netdevice.h>
14 #include <linux/etherdevice.h>
15 #include <linux/delay.h>
16 #include <linux/notifier.h>
17 #include <linux/ip.h>
18 #include <linux/tcp.h>
19 #include <linux/in.h>
20 #include <linux/ethtool.h>
21 #include <linux/topology.h>
22 #include <linux/gfp.h>
23 #include <linux/aer.h>
24 #include <linux/interrupt.h>
25 #include "net_driver.h"
26 #include "efx.h"
27 #include "nic.h"
28 #include "selftest.h"
29 
30 #include "workarounds.h"
31 
32 /**************************************************************************
33  *
34  * Type name strings
35  *
36  **************************************************************************
37  */
38 
39 /* Loopback mode names (see LOOPBACK_MODE()) */
40 const unsigned int ef4_loopback_mode_max = LOOPBACK_MAX;
41 const char *const ef4_loopback_mode_names[] = {
42 	[LOOPBACK_NONE]		= "NONE",
43 	[LOOPBACK_DATA]		= "DATAPATH",
44 	[LOOPBACK_GMAC]		= "GMAC",
45 	[LOOPBACK_XGMII]	= "XGMII",
46 	[LOOPBACK_XGXS]		= "XGXS",
47 	[LOOPBACK_XAUI]		= "XAUI",
48 	[LOOPBACK_GMII]		= "GMII",
49 	[LOOPBACK_SGMII]	= "SGMII",
50 	[LOOPBACK_XGBR]		= "XGBR",
51 	[LOOPBACK_XFI]		= "XFI",
52 	[LOOPBACK_XAUI_FAR]	= "XAUI_FAR",
53 	[LOOPBACK_GMII_FAR]	= "GMII_FAR",
54 	[LOOPBACK_SGMII_FAR]	= "SGMII_FAR",
55 	[LOOPBACK_XFI_FAR]	= "XFI_FAR",
56 	[LOOPBACK_GPHY]		= "GPHY",
57 	[LOOPBACK_PHYXS]	= "PHYXS",
58 	[LOOPBACK_PCS]		= "PCS",
59 	[LOOPBACK_PMAPMD]	= "PMA/PMD",
60 	[LOOPBACK_XPORT]	= "XPORT",
61 	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
62 	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
63 	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
64 	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
65 	[LOOPBACK_GMII_WS]	= "GMII_WS",
66 	[LOOPBACK_XFI_WS]	= "XFI_WS",
67 	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
68 	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
69 };
70 
71 const unsigned int ef4_reset_type_max = RESET_TYPE_MAX;
72 const char *const ef4_reset_type_names[] = {
73 	[RESET_TYPE_INVISIBLE]          = "INVISIBLE",
74 	[RESET_TYPE_ALL]                = "ALL",
75 	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
76 	[RESET_TYPE_WORLD]              = "WORLD",
77 	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
78 	[RESET_TYPE_DATAPATH]           = "DATAPATH",
79 	[RESET_TYPE_DISABLE]            = "DISABLE",
80 	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
81 	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
82 	[RESET_TYPE_RX_RECOVERY]        = "RX_RECOVERY",
83 	[RESET_TYPE_DMA_ERROR]          = "DMA_ERROR",
84 	[RESET_TYPE_TX_SKIP]            = "TX_SKIP",
85 };
86 
87 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
88  * queued onto this work queue. This is not a per-nic work queue, because
89  * ef4_reset_work() acquires the rtnl lock, so resets are naturally serialised.
90  */
91 static struct workqueue_struct *reset_workqueue;
92 
93 /* How often and how many times to poll for a reset while waiting for a
94  * BIST that another function started to complete.
95  */
96 #define BIST_WAIT_DELAY_MS	100
97 #define BIST_WAIT_DELAY_COUNT	100
98 
99 /**************************************************************************
100  *
101  * Configurable values
102  *
103  *************************************************************************/
104 
105 /*
106  * Use separate channels for TX and RX events
107  *
108  * Set this to 1 to use separate channels for TX and RX. It allows us
109  * to control interrupt affinity separately for TX and RX.
110  *
111  * This is only used in MSI-X interrupt mode
112  */
113 bool ef4_separate_tx_channels;
114 module_param(ef4_separate_tx_channels, bool, 0444);
115 MODULE_PARM_DESC(ef4_separate_tx_channels,
116 		 "Use separate channels for TX and RX");
117 
118 /* This is the weight assigned to each of the (per-channel) virtual
119  * NAPI devices.
120  */
121 static int napi_weight = 64;
122 
123 /* This is the time (in jiffies) between invocations of the hardware
124  * monitor.
125  * On Falcon-based NICs, this will:
126  * - Check the on-board hardware monitor;
127  * - Poll the link state and reconfigure the hardware as necessary.
128  * On Siena-based NICs for power systems with EEH support, this will give EEH a
129  * chance to start.
130  */
131 static unsigned int ef4_monitor_interval = 1 * HZ;
132 
133 /* Initial interrupt moderation settings.  They can be modified after
134  * module load with ethtool.
135  *
136  * The default for RX should strike a balance between increasing the
137  * round-trip latency and reducing overhead.
138  */
139 static unsigned int rx_irq_mod_usec = 60;
140 
141 /* Initial interrupt moderation settings.  They can be modified after
142  * module load with ethtool.
143  *
144  * This default is chosen to ensure that a 10G link does not go idle
145  * while a TX queue is stopped after it has become full.  A queue is
146  * restarted when it drops below half full.  The time this takes (assuming
147  * worst case 3 descriptors per packet and 1024 descriptors) is
148  *   512 / 3 * 1.2 = 205 usec.
149  */
150 static unsigned int tx_irq_mod_usec = 150;
151 
152 /* This is the first interrupt mode to try out of:
153  * 0 => MSI-X
154  * 1 => MSI
155  * 2 => legacy
156  */
157 static unsigned int interrupt_mode;
158 
159 /* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
160  * i.e. the number of CPUs among which we may distribute simultaneous
161  * interrupt handling.
162  *
163  * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
164  * The default (0) means to assign an interrupt to each core.
165  */
166 static unsigned int rss_cpus;
167 module_param(rss_cpus, uint, 0444);
168 MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");
169 
170 static bool phy_flash_cfg;
171 module_param(phy_flash_cfg, bool, 0644);
172 MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");
173 
174 static unsigned irq_adapt_low_thresh = 8000;
175 module_param(irq_adapt_low_thresh, uint, 0644);
176 MODULE_PARM_DESC(irq_adapt_low_thresh,
177 		 "Threshold score for reducing IRQ moderation");
178 
179 static unsigned irq_adapt_high_thresh = 16000;
180 module_param(irq_adapt_high_thresh, uint, 0644);
181 MODULE_PARM_DESC(irq_adapt_high_thresh,
182 		 "Threshold score for increasing IRQ moderation");
183 
184 static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
185 			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
186 			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
187 			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
188 module_param(debug, uint, 0);
189 MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
190 
191 /**************************************************************************
192  *
193  * Utility functions and prototypes
194  *
195  *************************************************************************/
196 
197 static int ef4_soft_enable_interrupts(struct ef4_nic *efx);
198 static void ef4_soft_disable_interrupts(struct ef4_nic *efx);
199 static void ef4_remove_channel(struct ef4_channel *channel);
200 static void ef4_remove_channels(struct ef4_nic *efx);
201 static const struct ef4_channel_type ef4_default_channel_type;
202 static void ef4_remove_port(struct ef4_nic *efx);
203 static void ef4_init_napi_channel(struct ef4_channel *channel);
204 static void ef4_fini_napi(struct ef4_nic *efx);
205 static void ef4_fini_napi_channel(struct ef4_channel *channel);
206 static void ef4_fini_struct(struct ef4_nic *efx);
207 static void ef4_start_all(struct ef4_nic *efx);
208 static void ef4_stop_all(struct ef4_nic *efx);
209 
210 #define EF4_ASSERT_RESET_SERIALISED(efx)		\
211 	do {						\
212 		if ((efx->state == STATE_READY) ||	\
213 		    (efx->state == STATE_RECOVERY) ||	\
214 		    (efx->state == STATE_DISABLED))	\
215 			ASSERT_RTNL();			\
216 	} while (0)
217 
218 static int ef4_check_disabled(struct ef4_nic *efx)
219 {
220 	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
221 		netif_err(efx, drv, efx->net_dev,
222 			  "device is disabled due to earlier errors\n");
223 		return -EIO;
224 	}
225 	return 0;
226 }
227 
228 /**************************************************************************
229  *
230  * Event queue processing
231  *
232  *************************************************************************/
233 
234 /* Process channel's event queue
235  *
236  * This function is responsible for processing the event queue of a
237  * single channel.  The caller must guarantee that this function will
238  * never be concurrently called more than once on the same channel,
239  * though different channels may be being processed concurrently.
240  */
241 static int ef4_process_channel(struct ef4_channel *channel, int budget)
242 {
243 	struct ef4_tx_queue *tx_queue;
244 	int spent;
245 
246 	if (unlikely(!channel->enabled))
247 		return 0;
248 
249 	ef4_for_each_channel_tx_queue(tx_queue, channel) {
250 		tx_queue->pkts_compl = 0;
251 		tx_queue->bytes_compl = 0;
252 	}
253 
254 	spent = ef4_nic_process_eventq(channel, budget);
255 	if (spent && ef4_channel_has_rx_queue(channel)) {
256 		struct ef4_rx_queue *rx_queue =
257 			ef4_channel_get_rx_queue(channel);
258 
259 		ef4_rx_flush_packet(channel);
260 		ef4_fast_push_rx_descriptors(rx_queue, true);
261 	}
262 
263 	/* Update BQL */
264 	ef4_for_each_channel_tx_queue(tx_queue, channel) {
265 		if (tx_queue->bytes_compl) {
266 			netdev_tx_completed_queue(tx_queue->core_txq,
267 				tx_queue->pkts_compl, tx_queue->bytes_compl);
268 		}
269 	}
270 
271 	return spent;
272 }
273 
274 /* NAPI poll handler
275  *
276  * NAPI guarantees serialisation of polls of the same device, which
277  * provides the guarantee required by ef4_process_channel().
278  */
279 static void ef4_update_irq_mod(struct ef4_nic *efx, struct ef4_channel *channel)
280 {
281 	int step = efx->irq_mod_step_us;
282 
283 	if (channel->irq_mod_score < irq_adapt_low_thresh) {
284 		if (channel->irq_moderation_us > step) {
285 			channel->irq_moderation_us -= step;
286 			efx->type->push_irq_moderation(channel);
287 		}
288 	} else if (channel->irq_mod_score > irq_adapt_high_thresh) {
289 		if (channel->irq_moderation_us <
290 		    efx->irq_rx_moderation_us) {
291 			channel->irq_moderation_us += step;
292 			efx->type->push_irq_moderation(channel);
293 		}
294 	}
295 
296 	channel->irq_count = 0;
297 	channel->irq_mod_score = 0;
298 }
299 
300 static int ef4_poll(struct napi_struct *napi, int budget)
301 {
302 	struct ef4_channel *channel =
303 		container_of(napi, struct ef4_channel, napi_str);
304 	struct ef4_nic *efx = channel->efx;
305 	int spent;
306 
307 	if (!ef4_channel_lock_napi(channel))
308 		return budget;
309 
310 	netif_vdbg(efx, intr, efx->net_dev,
311 		   "channel %d NAPI poll executing on CPU %d\n",
312 		   channel->channel, raw_smp_processor_id());
313 
314 	spent = ef4_process_channel(channel, budget);
315 
316 	if (spent < budget) {
317 		if (ef4_channel_has_rx_queue(channel) &&
318 		    efx->irq_rx_adaptive &&
319 		    unlikely(++channel->irq_count == 1000)) {
320 			ef4_update_irq_mod(efx, channel);
321 		}
322 
323 		ef4_filter_rfs_expire(channel);
324 
325 		/* There is no race here; although napi_disable() will
326 		 * only wait for napi_complete(), this isn't a problem
327 		 * since ef4_nic_eventq_read_ack() will have no effect if
328 		 * interrupts have already been disabled.
329 		 */
330 		napi_complete(napi);
331 		ef4_nic_eventq_read_ack(channel);
332 	}
333 
334 	ef4_channel_unlock_napi(channel);
335 	return spent;
336 }
337 
338 /* Create event queue
339  * Event queue memory allocations are done only once.  If the channel
340  * is reset, the memory buffer will be reused; this guards against
341  * errors during channel reset and also simplifies interrupt handling.
342  */
343 static int ef4_probe_eventq(struct ef4_channel *channel)
344 {
345 	struct ef4_nic *efx = channel->efx;
346 	unsigned long entries;
347 
348 	netif_dbg(efx, probe, efx->net_dev,
349 		  "chan %d create event queue\n", channel->channel);
350 
351 	/* Build an event queue with room for one event per tx and rx buffer,
352 	 * plus some extra for link state events and MCDI completions. */
353 	entries = roundup_pow_of_two(efx->rxq_entries + efx->txq_entries + 128);
354 	EF4_BUG_ON_PARANOID(entries > EF4_MAX_EVQ_SIZE);
355 	channel->eventq_mask = max(entries, EF4_MIN_EVQ_SIZE) - 1;
356 
357 	return ef4_nic_probe_eventq(channel);
358 }
359 
360 /* Prepare channel's event queue */
361 static int ef4_init_eventq(struct ef4_channel *channel)
362 {
363 	struct ef4_nic *efx = channel->efx;
364 	int rc;
365 
366 	EF4_WARN_ON_PARANOID(channel->eventq_init);
367 
368 	netif_dbg(efx, drv, efx->net_dev,
369 		  "chan %d init event queue\n", channel->channel);
370 
371 	rc = ef4_nic_init_eventq(channel);
372 	if (rc == 0) {
373 		efx->type->push_irq_moderation(channel);
374 		channel->eventq_read_ptr = 0;
375 		channel->eventq_init = true;
376 	}
377 	return rc;
378 }
379 
380 /* Enable event queue processing and NAPI */
381 void ef4_start_eventq(struct ef4_channel *channel)
382 {
383 	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
384 		  "chan %d start event queue\n", channel->channel);
385 
386 	/* Make sure the NAPI handler sees the enabled flag set */
387 	channel->enabled = true;
388 	smp_wmb();
389 
390 	ef4_channel_enable(channel);
391 	napi_enable(&channel->napi_str);
392 	ef4_nic_eventq_read_ack(channel);
393 }
394 
395 /* Disable event queue processing and NAPI */
396 void ef4_stop_eventq(struct ef4_channel *channel)
397 {
398 	if (!channel->enabled)
399 		return;
400 
401 	napi_disable(&channel->napi_str);
402 	while (!ef4_channel_disable(channel))
403 		usleep_range(1000, 20000);
404 	channel->enabled = false;
405 }
406 
407 static void ef4_fini_eventq(struct ef4_channel *channel)
408 {
409 	if (!channel->eventq_init)
410 		return;
411 
412 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
413 		  "chan %d fini event queue\n", channel->channel);
414 
415 	ef4_nic_fini_eventq(channel);
416 	channel->eventq_init = false;
417 }
418 
419 static void ef4_remove_eventq(struct ef4_channel *channel)
420 {
421 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
422 		  "chan %d remove event queue\n", channel->channel);
423 
424 	ef4_nic_remove_eventq(channel);
425 }
426 
427 /**************************************************************************
428  *
429  * Channel handling
430  *
431  *************************************************************************/
432 
433 /* Allocate and initialise a channel structure. */
434 static struct ef4_channel *
435 ef4_alloc_channel(struct ef4_nic *efx, int i, struct ef4_channel *old_channel)
436 {
437 	struct ef4_channel *channel;
438 	struct ef4_rx_queue *rx_queue;
439 	struct ef4_tx_queue *tx_queue;
440 	int j;
441 
442 	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
443 	if (!channel)
444 		return NULL;
445 
446 	channel->efx = efx;
447 	channel->channel = i;
448 	channel->type = &ef4_default_channel_type;
449 
450 	for (j = 0; j < EF4_TXQ_TYPES; j++) {
451 		tx_queue = &channel->tx_queue[j];
452 		tx_queue->efx = efx;
453 		tx_queue->queue = i * EF4_TXQ_TYPES + j;
454 		tx_queue->channel = channel;
455 	}
456 
457 	rx_queue = &channel->rx_queue;
458 	rx_queue->efx = efx;
459 	setup_timer(&rx_queue->slow_fill, ef4_rx_slow_fill,
460 		    (unsigned long)rx_queue);
461 
462 	return channel;
463 }
464 
465 /* Allocate and initialise a channel structure, copying parameters
466  * (but not resources) from an old channel structure.
467  */
468 static struct ef4_channel *
469 ef4_copy_channel(const struct ef4_channel *old_channel)
470 {
471 	struct ef4_channel *channel;
472 	struct ef4_rx_queue *rx_queue;
473 	struct ef4_tx_queue *tx_queue;
474 	int j;
475 
476 	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
477 	if (!channel)
478 		return NULL;
479 
480 	*channel = *old_channel;
481 
482 	channel->napi_dev = NULL;
483 	INIT_HLIST_NODE(&channel->napi_str.napi_hash_node);
484 	channel->napi_str.napi_id = 0;
485 	channel->napi_str.state = 0;
486 	memset(&channel->eventq, 0, sizeof(channel->eventq));
487 
488 	for (j = 0; j < EF4_TXQ_TYPES; j++) {
489 		tx_queue = &channel->tx_queue[j];
490 		if (tx_queue->channel)
491 			tx_queue->channel = channel;
492 		tx_queue->buffer = NULL;
493 		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
494 	}
495 
496 	rx_queue = &channel->rx_queue;
497 	rx_queue->buffer = NULL;
498 	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
499 	setup_timer(&rx_queue->slow_fill, ef4_rx_slow_fill,
500 		    (unsigned long)rx_queue);
501 
502 	return channel;
503 }
504 
505 static int ef4_probe_channel(struct ef4_channel *channel)
506 {
507 	struct ef4_tx_queue *tx_queue;
508 	struct ef4_rx_queue *rx_queue;
509 	int rc;
510 
511 	netif_dbg(channel->efx, probe, channel->efx->net_dev,
512 		  "creating channel %d\n", channel->channel);
513 
514 	rc = channel->type->pre_probe(channel);
515 	if (rc)
516 		goto fail;
517 
518 	rc = ef4_probe_eventq(channel);
519 	if (rc)
520 		goto fail;
521 
522 	ef4_for_each_channel_tx_queue(tx_queue, channel) {
523 		rc = ef4_probe_tx_queue(tx_queue);
524 		if (rc)
525 			goto fail;
526 	}
527 
528 	ef4_for_each_channel_rx_queue(rx_queue, channel) {
529 		rc = ef4_probe_rx_queue(rx_queue);
530 		if (rc)
531 			goto fail;
532 	}
533 
534 	return 0;
535 
536 fail:
537 	ef4_remove_channel(channel);
538 	return rc;
539 }
540 
541 static void
542 ef4_get_channel_name(struct ef4_channel *channel, char *buf, size_t len)
543 {
544 	struct ef4_nic *efx = channel->efx;
545 	const char *type;
546 	int number;
547 
548 	number = channel->channel;
549 	if (efx->tx_channel_offset == 0) {
550 		type = "";
551 	} else if (channel->channel < efx->tx_channel_offset) {
552 		type = "-rx";
553 	} else {
554 		type = "-tx";
555 		number -= efx->tx_channel_offset;
556 	}
557 	snprintf(buf, len, "%s%s-%d", efx->name, type, number);
558 }
559 
560 static void ef4_set_channel_names(struct ef4_nic *efx)
561 {
562 	struct ef4_channel *channel;
563 
564 	ef4_for_each_channel(channel, efx)
565 		channel->type->get_name(channel,
566 					efx->msi_context[channel->channel].name,
567 					sizeof(efx->msi_context[0].name));
568 }
569 
570 static int ef4_probe_channels(struct ef4_nic *efx)
571 {
572 	struct ef4_channel *channel;
573 	int rc;
574 
575 	/* Restart special buffer allocation */
576 	efx->next_buffer_table = 0;
577 
578 	/* Probe channels in reverse, so that any 'extra' channels
579 	 * use the start of the buffer table. This allows the traffic
580 	 * channels to be resized without moving them or wasting the
581 	 * entries before them.
582 	 */
583 	ef4_for_each_channel_rev(channel, efx) {
584 		rc = ef4_probe_channel(channel);
585 		if (rc) {
586 			netif_err(efx, probe, efx->net_dev,
587 				  "failed to create channel %d\n",
588 				  channel->channel);
589 			goto fail;
590 		}
591 	}
592 	ef4_set_channel_names(efx);
593 
594 	return 0;
595 
596 fail:
597 	ef4_remove_channels(efx);
598 	return rc;
599 }
600 
601 /* Channels are shutdown and reinitialised whilst the NIC is running
602  * to propagate configuration changes (mtu, checksum offload), or
603  * to clear hardware error conditions
604  */
605 static void ef4_start_datapath(struct ef4_nic *efx)
606 {
607 	netdev_features_t old_features = efx->net_dev->features;
608 	bool old_rx_scatter = efx->rx_scatter;
609 	struct ef4_tx_queue *tx_queue;
610 	struct ef4_rx_queue *rx_queue;
611 	struct ef4_channel *channel;
612 	size_t rx_buf_len;
613 
614 	/* Calculate the rx buffer allocation parameters required to
615 	 * support the current MTU, including padding for header
616 	 * alignment and overruns.
617 	 */
618 	efx->rx_dma_len = (efx->rx_prefix_size +
619 			   EF4_MAX_FRAME_LEN(efx->net_dev->mtu) +
620 			   efx->type->rx_buffer_padding);
621 	rx_buf_len = (sizeof(struct ef4_rx_page_state) +
622 		      efx->rx_ip_align + efx->rx_dma_len);
623 	if (rx_buf_len <= PAGE_SIZE) {
624 		efx->rx_scatter = efx->type->always_rx_scatter;
625 		efx->rx_buffer_order = 0;
626 	} else if (efx->type->can_rx_scatter) {
627 		BUILD_BUG_ON(EF4_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
628 		BUILD_BUG_ON(sizeof(struct ef4_rx_page_state) +
629 			     2 * ALIGN(NET_IP_ALIGN + EF4_RX_USR_BUF_SIZE,
630 				       EF4_RX_BUF_ALIGNMENT) >
631 			     PAGE_SIZE);
632 		efx->rx_scatter = true;
633 		efx->rx_dma_len = EF4_RX_USR_BUF_SIZE;
634 		efx->rx_buffer_order = 0;
635 	} else {
636 		efx->rx_scatter = false;
637 		efx->rx_buffer_order = get_order(rx_buf_len);
638 	}
639 
640 	ef4_rx_config_page_split(efx);
641 	if (efx->rx_buffer_order)
642 		netif_dbg(efx, drv, efx->net_dev,
643 			  "RX buf len=%u; page order=%u batch=%u\n",
644 			  efx->rx_dma_len, efx->rx_buffer_order,
645 			  efx->rx_pages_per_batch);
646 	else
647 		netif_dbg(efx, drv, efx->net_dev,
648 			  "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
649 			  efx->rx_dma_len, efx->rx_page_buf_step,
650 			  efx->rx_bufs_per_page, efx->rx_pages_per_batch);
651 
652 	/* Restore previously fixed features in hw_features and remove
653 	 * features which are fixed now
654 	 */
655 	efx->net_dev->hw_features |= efx->net_dev->features;
656 	efx->net_dev->hw_features &= ~efx->fixed_features;
657 	efx->net_dev->features |= efx->fixed_features;
658 	if (efx->net_dev->features != old_features)
659 		netdev_features_change(efx->net_dev);
660 
661 	/* RX filters may also have scatter-enabled flags */
662 	if (efx->rx_scatter != old_rx_scatter)
663 		efx->type->filter_update_rx_scatter(efx);
664 
665 	/* We must keep at least one descriptor in a TX ring empty.
666 	 * We could avoid this when the queue size does not exactly
667 	 * match the hardware ring size, but it's not that important.
668 	 * Therefore we stop the queue when one more skb might fill
669 	 * the ring completely.  We wake it when half way back to
670 	 * empty.
671 	 */
672 	efx->txq_stop_thresh = efx->txq_entries - ef4_tx_max_skb_descs(efx);
673 	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;
674 
675 	/* Initialise the channels */
676 	ef4_for_each_channel(channel, efx) {
677 		ef4_for_each_channel_tx_queue(tx_queue, channel) {
678 			ef4_init_tx_queue(tx_queue);
679 			atomic_inc(&efx->active_queues);
680 		}
681 
682 		ef4_for_each_channel_rx_queue(rx_queue, channel) {
683 			ef4_init_rx_queue(rx_queue);
684 			atomic_inc(&efx->active_queues);
685 			ef4_stop_eventq(channel);
686 			ef4_fast_push_rx_descriptors(rx_queue, false);
687 			ef4_start_eventq(channel);
688 		}
689 
690 		WARN_ON(channel->rx_pkt_n_frags);
691 	}
692 
693 	if (netif_device_present(efx->net_dev))
694 		netif_tx_wake_all_queues(efx->net_dev);
695 }
696 
697 static void ef4_stop_datapath(struct ef4_nic *efx)
698 {
699 	struct ef4_channel *channel;
700 	struct ef4_tx_queue *tx_queue;
701 	struct ef4_rx_queue *rx_queue;
702 	int rc;
703 
704 	EF4_ASSERT_RESET_SERIALISED(efx);
705 	BUG_ON(efx->port_enabled);
706 
707 	/* Stop RX refill */
708 	ef4_for_each_channel(channel, efx) {
709 		ef4_for_each_channel_rx_queue(rx_queue, channel)
710 			rx_queue->refill_enabled = false;
711 	}
712 
713 	ef4_for_each_channel(channel, efx) {
714 		/* RX packet processing is pipelined, so wait for the
715 		 * NAPI handler to complete.  At least event queue 0
716 		 * might be kept active by non-data events, so don't
717 		 * use napi_synchronize() but actually disable NAPI
718 		 * temporarily.
719 		 */
720 		if (ef4_channel_has_rx_queue(channel)) {
721 			ef4_stop_eventq(channel);
722 			ef4_start_eventq(channel);
723 		}
724 	}
725 
726 	rc = efx->type->fini_dmaq(efx);
727 	if (rc && EF4_WORKAROUND_7803(efx)) {
728 		/* Schedule a reset to recover from the flush failure. The
729 		 * descriptor caches reference memory we're about to free,
730 		 * but falcon_reconfigure_mac_wrapper() won't reconnect
731 		 * the MACs because of the pending reset.
732 		 */
733 		netif_err(efx, drv, efx->net_dev,
734 			  "Resetting to recover from flush failure\n");
735 		ef4_schedule_reset(efx, RESET_TYPE_ALL);
736 	} else if (rc) {
737 		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
738 	} else {
739 		netif_dbg(efx, drv, efx->net_dev,
740 			  "successfully flushed all queues\n");
741 	}
742 
743 	ef4_for_each_channel(channel, efx) {
744 		ef4_for_each_channel_rx_queue(rx_queue, channel)
745 			ef4_fini_rx_queue(rx_queue);
746 		ef4_for_each_possible_channel_tx_queue(tx_queue, channel)
747 			ef4_fini_tx_queue(tx_queue);
748 	}
749 }
750 
751 static void ef4_remove_channel(struct ef4_channel *channel)
752 {
753 	struct ef4_tx_queue *tx_queue;
754 	struct ef4_rx_queue *rx_queue;
755 
756 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
757 		  "destroy chan %d\n", channel->channel);
758 
759 	ef4_for_each_channel_rx_queue(rx_queue, channel)
760 		ef4_remove_rx_queue(rx_queue);
761 	ef4_for_each_possible_channel_tx_queue(tx_queue, channel)
762 		ef4_remove_tx_queue(tx_queue);
763 	ef4_remove_eventq(channel);
764 	channel->type->post_remove(channel);
765 }
766 
767 static void ef4_remove_channels(struct ef4_nic *efx)
768 {
769 	struct ef4_channel *channel;
770 
771 	ef4_for_each_channel(channel, efx)
772 		ef4_remove_channel(channel);
773 }
774 
775 int
776 ef4_realloc_channels(struct ef4_nic *efx, u32 rxq_entries, u32 txq_entries)
777 {
778 	struct ef4_channel *other_channel[EF4_MAX_CHANNELS], *channel;
779 	u32 old_rxq_entries, old_txq_entries;
780 	unsigned i, next_buffer_table = 0;
781 	int rc, rc2;
782 
783 	rc = ef4_check_disabled(efx);
784 	if (rc)
785 		return rc;
786 
787 	/* Not all channels should be reallocated. We must avoid
788 	 * reallocating their buffer table entries.
789 	 */
790 	ef4_for_each_channel(channel, efx) {
791 		struct ef4_rx_queue *rx_queue;
792 		struct ef4_tx_queue *tx_queue;
793 
794 		if (channel->type->copy)
795 			continue;
796 		next_buffer_table = max(next_buffer_table,
797 					channel->eventq.index +
798 					channel->eventq.entries);
799 		ef4_for_each_channel_rx_queue(rx_queue, channel)
800 			next_buffer_table = max(next_buffer_table,
801 						rx_queue->rxd.index +
802 						rx_queue->rxd.entries);
803 		ef4_for_each_channel_tx_queue(tx_queue, channel)
804 			next_buffer_table = max(next_buffer_table,
805 						tx_queue->txd.index +
806 						tx_queue->txd.entries);
807 	}
808 
809 	ef4_device_detach_sync(efx);
810 	ef4_stop_all(efx);
811 	ef4_soft_disable_interrupts(efx);
812 
813 	/* Clone channels (where possible) */
814 	memset(other_channel, 0, sizeof(other_channel));
815 	for (i = 0; i < efx->n_channels; i++) {
816 		channel = efx->channel[i];
817 		if (channel->type->copy)
818 			channel = channel->type->copy(channel);
819 		if (!channel) {
820 			rc = -ENOMEM;
821 			goto out;
822 		}
823 		other_channel[i] = channel;
824 	}
825 
826 	/* Swap entry counts and channel pointers */
827 	old_rxq_entries = efx->rxq_entries;
828 	old_txq_entries = efx->txq_entries;
829 	efx->rxq_entries = rxq_entries;
830 	efx->txq_entries = txq_entries;
831 	for (i = 0; i < efx->n_channels; i++) {
832 		channel = efx->channel[i];
833 		efx->channel[i] = other_channel[i];
834 		other_channel[i] = channel;
835 	}
836 
837 	/* Restart buffer table allocation */
838 	efx->next_buffer_table = next_buffer_table;
839 
840 	for (i = 0; i < efx->n_channels; i++) {
841 		channel = efx->channel[i];
842 		if (!channel->type->copy)
843 			continue;
844 		rc = ef4_probe_channel(channel);
845 		if (rc)
846 			goto rollback;
847 		ef4_init_napi_channel(efx->channel[i]);
848 	}
849 
850 out:
851 	/* Destroy unused channel structures */
852 	for (i = 0; i < efx->n_channels; i++) {
853 		channel = other_channel[i];
854 		if (channel && channel->type->copy) {
855 			ef4_fini_napi_channel(channel);
856 			ef4_remove_channel(channel);
857 			kfree(channel);
858 		}
859 	}
860 
861 	rc2 = ef4_soft_enable_interrupts(efx);
862 	if (rc2) {
863 		rc = rc ? rc : rc2;
864 		netif_err(efx, drv, efx->net_dev,
865 			  "unable to restart interrupts on channel reallocation\n");
866 		ef4_schedule_reset(efx, RESET_TYPE_DISABLE);
867 	} else {
868 		ef4_start_all(efx);
869 		netif_device_attach(efx->net_dev);
870 	}
871 	return rc;
872 
873 rollback:
874 	/* Swap back */
875 	efx->rxq_entries = old_rxq_entries;
876 	efx->txq_entries = old_txq_entries;
877 	for (i = 0; i < efx->n_channels; i++) {
878 		channel = efx->channel[i];
879 		efx->channel[i] = other_channel[i];
880 		other_channel[i] = channel;
881 	}
882 	goto out;
883 }
884 
885 void ef4_schedule_slow_fill(struct ef4_rx_queue *rx_queue)
886 {
887 	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
888 }
889 
890 static const struct ef4_channel_type ef4_default_channel_type = {
891 	.pre_probe		= ef4_channel_dummy_op_int,
892 	.post_remove		= ef4_channel_dummy_op_void,
893 	.get_name		= ef4_get_channel_name,
894 	.copy			= ef4_copy_channel,
895 	.keep_eventq		= false,
896 };
897 
898 int ef4_channel_dummy_op_int(struct ef4_channel *channel)
899 {
900 	return 0;
901 }
902 
903 void ef4_channel_dummy_op_void(struct ef4_channel *channel)
904 {
905 }
906 
907 /**************************************************************************
908  *
909  * Port handling
910  *
911  **************************************************************************/
912 
913 /* This ensures that the kernel is kept informed (via
914  * netif_carrier_on/off) of the link status, and also maintains the
915  * link status's stop on the port's TX queue.
916  */
917 void ef4_link_status_changed(struct ef4_nic *efx)
918 {
919 	struct ef4_link_state *link_state = &efx->link_state;
920 
921 	/* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
922 	 * that no events are triggered between unregister_netdev() and the
923 	 * driver unloading. A more general condition is that NETDEV_CHANGE
924 	 * can only be generated between NETDEV_UP and NETDEV_DOWN */
925 	if (!netif_running(efx->net_dev))
926 		return;
927 
928 	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
929 		efx->n_link_state_changes++;
930 
931 		if (link_state->up)
932 			netif_carrier_on(efx->net_dev);
933 		else
934 			netif_carrier_off(efx->net_dev);
935 	}
936 
937 	/* Status message for kernel log */
938 	if (link_state->up)
939 		netif_info(efx, link, efx->net_dev,
940 			   "link up at %uMbps %s-duplex (MTU %d)\n",
941 			   link_state->speed, link_state->fd ? "full" : "half",
942 			   efx->net_dev->mtu);
943 	else
944 		netif_info(efx, link, efx->net_dev, "link down\n");
945 }
946 
947 void ef4_link_set_advertising(struct ef4_nic *efx, u32 advertising)
948 {
949 	efx->link_advertising = advertising;
950 	if (advertising) {
951 		if (advertising & ADVERTISED_Pause)
952 			efx->wanted_fc |= (EF4_FC_TX | EF4_FC_RX);
953 		else
954 			efx->wanted_fc &= ~(EF4_FC_TX | EF4_FC_RX);
955 		if (advertising & ADVERTISED_Asym_Pause)
956 			efx->wanted_fc ^= EF4_FC_TX;
957 	}
958 }
959 
960 void ef4_link_set_wanted_fc(struct ef4_nic *efx, u8 wanted_fc)
961 {
962 	efx->wanted_fc = wanted_fc;
963 	if (efx->link_advertising) {
964 		if (wanted_fc & EF4_FC_RX)
965 			efx->link_advertising |= (ADVERTISED_Pause |
966 						  ADVERTISED_Asym_Pause);
967 		else
968 			efx->link_advertising &= ~(ADVERTISED_Pause |
969 						   ADVERTISED_Asym_Pause);
970 		if (wanted_fc & EF4_FC_TX)
971 			efx->link_advertising ^= ADVERTISED_Asym_Pause;
972 	}
973 }
974 
975 static void ef4_fini_port(struct ef4_nic *efx);
976 
977 /* We assume that efx->type->reconfigure_mac will always try to sync RX
978  * filters and therefore needs to read-lock the filter table against freeing
979  */
980 void ef4_mac_reconfigure(struct ef4_nic *efx)
981 {
982 	down_read(&efx->filter_sem);
983 	efx->type->reconfigure_mac(efx);
984 	up_read(&efx->filter_sem);
985 }
986 
987 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
988  * the MAC appropriately. All other PHY configuration changes are pushed
989  * through phy_op->set_settings(), and pushed asynchronously to the MAC
990  * through ef4_monitor().
991  *
992  * Callers must hold the mac_lock
993  */
994 int __ef4_reconfigure_port(struct ef4_nic *efx)
995 {
996 	enum ef4_phy_mode phy_mode;
997 	int rc;
998 
999 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
1000 
1001 	/* Disable PHY transmit in mac level loopbacks */
1002 	phy_mode = efx->phy_mode;
1003 	if (LOOPBACK_INTERNAL(efx))
1004 		efx->phy_mode |= PHY_MODE_TX_DISABLED;
1005 	else
1006 		efx->phy_mode &= ~PHY_MODE_TX_DISABLED;
1007 
1008 	rc = efx->type->reconfigure_port(efx);
1009 
1010 	if (rc)
1011 		efx->phy_mode = phy_mode;
1012 
1013 	return rc;
1014 }
1015 
1016 /* Reinitialise the MAC to pick up new PHY settings, even if the port is
1017  * disabled. */
1018 int ef4_reconfigure_port(struct ef4_nic *efx)
1019 {
1020 	int rc;
1021 
1022 	EF4_ASSERT_RESET_SERIALISED(efx);
1023 
1024 	mutex_lock(&efx->mac_lock);
1025 	rc = __ef4_reconfigure_port(efx);
1026 	mutex_unlock(&efx->mac_lock);
1027 
1028 	return rc;
1029 }
1030 
1031 /* Asynchronous work item for changing MAC promiscuity and multicast
1032  * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
1033  * MAC directly. */
1034 static void ef4_mac_work(struct work_struct *data)
1035 {
1036 	struct ef4_nic *efx = container_of(data, struct ef4_nic, mac_work);
1037 
1038 	mutex_lock(&efx->mac_lock);
1039 	if (efx->port_enabled)
1040 		ef4_mac_reconfigure(efx);
1041 	mutex_unlock(&efx->mac_lock);
1042 }
1043 
1044 static int ef4_probe_port(struct ef4_nic *efx)
1045 {
1046 	int rc;
1047 
1048 	netif_dbg(efx, probe, efx->net_dev, "create port\n");
1049 
1050 	if (phy_flash_cfg)
1051 		efx->phy_mode = PHY_MODE_SPECIAL;
1052 
1053 	/* Connect up MAC/PHY operations table */
1054 	rc = efx->type->probe_port(efx);
1055 	if (rc)
1056 		return rc;
1057 
1058 	/* Initialise MAC address to permanent address */
1059 	ether_addr_copy(efx->net_dev->dev_addr, efx->net_dev->perm_addr);
1060 
1061 	return 0;
1062 }
1063 
1064 static int ef4_init_port(struct ef4_nic *efx)
1065 {
1066 	int rc;
1067 
1068 	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1069 
1070 	mutex_lock(&efx->mac_lock);
1071 
1072 	rc = efx->phy_op->init(efx);
1073 	if (rc)
1074 		goto fail1;
1075 
1076 	efx->port_initialized = true;
1077 
1078 	/* Reconfigure the MAC before creating dma queues (required for
1079 	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1080 	ef4_mac_reconfigure(efx);
1081 
1082 	/* Ensure the PHY advertises the correct flow control settings */
1083 	rc = efx->phy_op->reconfigure(efx);
1084 	if (rc && rc != -EPERM)
1085 		goto fail2;
1086 
1087 	mutex_unlock(&efx->mac_lock);
1088 	return 0;
1089 
1090 fail2:
1091 	efx->phy_op->fini(efx);
1092 fail1:
1093 	mutex_unlock(&efx->mac_lock);
1094 	return rc;
1095 }
1096 
1097 static void ef4_start_port(struct ef4_nic *efx)
1098 {
1099 	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1100 	BUG_ON(efx->port_enabled);
1101 
1102 	mutex_lock(&efx->mac_lock);
1103 	efx->port_enabled = true;
1104 
1105 	/* Ensure MAC ingress/egress is enabled */
1106 	ef4_mac_reconfigure(efx);
1107 
1108 	mutex_unlock(&efx->mac_lock);
1109 }
1110 
1111 /* Cancel work for MAC reconfiguration, periodic hardware monitoring
1112  * and the async self-test, wait for them to finish and prevent them
1113  * being scheduled again.  This doesn't cover online resets, which
1114  * should only be cancelled when removing the device.
1115  */
1116 static void ef4_stop_port(struct ef4_nic *efx)
1117 {
1118 	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1119 
1120 	EF4_ASSERT_RESET_SERIALISED(efx);
1121 
1122 	mutex_lock(&efx->mac_lock);
1123 	efx->port_enabled = false;
1124 	mutex_unlock(&efx->mac_lock);
1125 
1126 	/* Serialise against ef4_set_multicast_list() */
1127 	netif_addr_lock_bh(efx->net_dev);
1128 	netif_addr_unlock_bh(efx->net_dev);
1129 
1130 	cancel_delayed_work_sync(&efx->monitor_work);
1131 	ef4_selftest_async_cancel(efx);
1132 	cancel_work_sync(&efx->mac_work);
1133 }
1134 
1135 static void ef4_fini_port(struct ef4_nic *efx)
1136 {
1137 	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1138 
1139 	if (!efx->port_initialized)
1140 		return;
1141 
1142 	efx->phy_op->fini(efx);
1143 	efx->port_initialized = false;
1144 
1145 	efx->link_state.up = false;
1146 	ef4_link_status_changed(efx);
1147 }
1148 
1149 static void ef4_remove_port(struct ef4_nic *efx)
1150 {
1151 	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1152 
1153 	efx->type->remove_port(efx);
1154 }
1155 
1156 /**************************************************************************
1157  *
1158  * NIC handling
1159  *
1160  **************************************************************************/
1161 
1162 static LIST_HEAD(ef4_primary_list);
1163 static LIST_HEAD(ef4_unassociated_list);
1164 
1165 static bool ef4_same_controller(struct ef4_nic *left, struct ef4_nic *right)
1166 {
1167 	return left->type == right->type &&
1168 		left->vpd_sn && right->vpd_sn &&
1169 		!strcmp(left->vpd_sn, right->vpd_sn);
1170 }
1171 
1172 static void ef4_associate(struct ef4_nic *efx)
1173 {
1174 	struct ef4_nic *other, *next;
1175 
1176 	if (efx->primary == efx) {
1177 		/* Adding primary function; look for secondaries */
1178 
1179 		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
1180 		list_add_tail(&efx->node, &ef4_primary_list);
1181 
1182 		list_for_each_entry_safe(other, next, &ef4_unassociated_list,
1183 					 node) {
1184 			if (ef4_same_controller(efx, other)) {
1185 				list_del(&other->node);
1186 				netif_dbg(other, probe, other->net_dev,
1187 					  "moving to secondary list of %s %s\n",
1188 					  pci_name(efx->pci_dev),
1189 					  efx->net_dev->name);
1190 				list_add_tail(&other->node,
1191 					      &efx->secondary_list);
1192 				other->primary = efx;
1193 			}
1194 		}
1195 	} else {
1196 		/* Adding secondary function; look for primary */
1197 
1198 		list_for_each_entry(other, &ef4_primary_list, node) {
1199 			if (ef4_same_controller(efx, other)) {
1200 				netif_dbg(efx, probe, efx->net_dev,
1201 					  "adding to secondary list of %s %s\n",
1202 					  pci_name(other->pci_dev),
1203 					  other->net_dev->name);
1204 				list_add_tail(&efx->node,
1205 					      &other->secondary_list);
1206 				efx->primary = other;
1207 				return;
1208 			}
1209 		}
1210 
1211 		netif_dbg(efx, probe, efx->net_dev,
1212 			  "adding to unassociated list\n");
1213 		list_add_tail(&efx->node, &ef4_unassociated_list);
1214 	}
1215 }
1216 
1217 static void ef4_dissociate(struct ef4_nic *efx)
1218 {
1219 	struct ef4_nic *other, *next;
1220 
1221 	list_del(&efx->node);
1222 	efx->primary = NULL;
1223 
1224 	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
1225 		list_del(&other->node);
1226 		netif_dbg(other, probe, other->net_dev,
1227 			  "moving to unassociated list\n");
1228 		list_add_tail(&other->node, &ef4_unassociated_list);
1229 		other->primary = NULL;
1230 	}
1231 }
1232 
1233 /* This configures the PCI device to enable I/O and DMA. */
1234 static int ef4_init_io(struct ef4_nic *efx)
1235 {
1236 	struct pci_dev *pci_dev = efx->pci_dev;
1237 	dma_addr_t dma_mask = efx->type->max_dma_mask;
1238 	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1239 	int rc, bar;
1240 
1241 	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1242 
1243 	bar = efx->type->mem_bar;
1244 
1245 	rc = pci_enable_device(pci_dev);
1246 	if (rc) {
1247 		netif_err(efx, probe, efx->net_dev,
1248 			  "failed to enable PCI device\n");
1249 		goto fail1;
1250 	}
1251 
1252 	pci_set_master(pci_dev);
1253 
1254 	/* Set the PCI DMA mask.  Try all possibilities from our
1255 	 * genuine mask down to 32 bits, because some architectures
1256 	 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
1257 	 * masks event though they reject 46 bit masks.
1258 	 */
1259 	while (dma_mask > 0x7fffffffUL) {
1260 		rc = dma_set_mask_and_coherent(&pci_dev->dev, dma_mask);
1261 		if (rc == 0)
1262 			break;
1263 		dma_mask >>= 1;
1264 	}
1265 	if (rc) {
1266 		netif_err(efx, probe, efx->net_dev,
1267 			  "could not find a suitable DMA mask\n");
1268 		goto fail2;
1269 	}
1270 	netif_dbg(efx, probe, efx->net_dev,
1271 		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1272 
1273 	efx->membase_phys = pci_resource_start(efx->pci_dev, bar);
1274 	rc = pci_request_region(pci_dev, bar, "sfc");
1275 	if (rc) {
1276 		netif_err(efx, probe, efx->net_dev,
1277 			  "request for memory BAR failed\n");
1278 		rc = -EIO;
1279 		goto fail3;
1280 	}
1281 	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1282 	if (!efx->membase) {
1283 		netif_err(efx, probe, efx->net_dev,
1284 			  "could not map memory BAR at %llx+%x\n",
1285 			  (unsigned long long)efx->membase_phys, mem_map_size);
1286 		rc = -ENOMEM;
1287 		goto fail4;
1288 	}
1289 	netif_dbg(efx, probe, efx->net_dev,
1290 		  "memory BAR at %llx+%x (virtual %p)\n",
1291 		  (unsigned long long)efx->membase_phys, mem_map_size,
1292 		  efx->membase);
1293 
1294 	return 0;
1295 
1296  fail4:
1297 	pci_release_region(efx->pci_dev, bar);
1298  fail3:
1299 	efx->membase_phys = 0;
1300  fail2:
1301 	pci_disable_device(efx->pci_dev);
1302  fail1:
1303 	return rc;
1304 }
1305 
1306 static void ef4_fini_io(struct ef4_nic *efx)
1307 {
1308 	int bar;
1309 
1310 	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1311 
1312 	if (efx->membase) {
1313 		iounmap(efx->membase);
1314 		efx->membase = NULL;
1315 	}
1316 
1317 	if (efx->membase_phys) {
1318 		bar = efx->type->mem_bar;
1319 		pci_release_region(efx->pci_dev, bar);
1320 		efx->membase_phys = 0;
1321 	}
1322 
1323 	/* Don't disable bus-mastering if VFs are assigned */
1324 	if (!pci_vfs_assigned(efx->pci_dev))
1325 		pci_disable_device(efx->pci_dev);
1326 }
1327 
1328 void ef4_set_default_rx_indir_table(struct ef4_nic *efx)
1329 {
1330 	size_t i;
1331 
1332 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
1333 		efx->rx_indir_table[i] =
1334 			ethtool_rxfh_indir_default(i, efx->rss_spread);
1335 }
1336 
1337 static unsigned int ef4_wanted_parallelism(struct ef4_nic *efx)
1338 {
1339 	cpumask_var_t thread_mask;
1340 	unsigned int count;
1341 	int cpu;
1342 
1343 	if (rss_cpus) {
1344 		count = rss_cpus;
1345 	} else {
1346 		if (unlikely(!zalloc_cpumask_var(&thread_mask, GFP_KERNEL))) {
1347 			netif_warn(efx, probe, efx->net_dev,
1348 				   "RSS disabled due to allocation failure\n");
1349 			return 1;
1350 		}
1351 
1352 		count = 0;
1353 		for_each_online_cpu(cpu) {
1354 			if (!cpumask_test_cpu(cpu, thread_mask)) {
1355 				++count;
1356 				cpumask_or(thread_mask, thread_mask,
1357 					   topology_sibling_cpumask(cpu));
1358 			}
1359 		}
1360 
1361 		free_cpumask_var(thread_mask);
1362 	}
1363 
1364 	return count;
1365 }
1366 
1367 /* Probe the number and type of interrupts we are able to obtain, and
1368  * the resulting numbers of channels and RX queues.
1369  */
1370 static int ef4_probe_interrupts(struct ef4_nic *efx)
1371 {
1372 	unsigned int extra_channels = 0;
1373 	unsigned int i, j;
1374 	int rc;
1375 
1376 	for (i = 0; i < EF4_MAX_EXTRA_CHANNELS; i++)
1377 		if (efx->extra_channel_type[i])
1378 			++extra_channels;
1379 
1380 	if (efx->interrupt_mode == EF4_INT_MODE_MSIX) {
1381 		struct msix_entry xentries[EF4_MAX_CHANNELS];
1382 		unsigned int n_channels;
1383 
1384 		n_channels = ef4_wanted_parallelism(efx);
1385 		if (ef4_separate_tx_channels)
1386 			n_channels *= 2;
1387 		n_channels += extra_channels;
1388 		n_channels = min(n_channels, efx->max_channels);
1389 
1390 		for (i = 0; i < n_channels; i++)
1391 			xentries[i].entry = i;
1392 		rc = pci_enable_msix_range(efx->pci_dev,
1393 					   xentries, 1, n_channels);
1394 		if (rc < 0) {
1395 			/* Fall back to single channel MSI */
1396 			efx->interrupt_mode = EF4_INT_MODE_MSI;
1397 			netif_err(efx, drv, efx->net_dev,
1398 				  "could not enable MSI-X\n");
1399 		} else if (rc < n_channels) {
1400 			netif_err(efx, drv, efx->net_dev,
1401 				  "WARNING: Insufficient MSI-X vectors"
1402 				  " available (%d < %u).\n", rc, n_channels);
1403 			netif_err(efx, drv, efx->net_dev,
1404 				  "WARNING: Performance may be reduced.\n");
1405 			n_channels = rc;
1406 		}
1407 
1408 		if (rc > 0) {
1409 			efx->n_channels = n_channels;
1410 			if (n_channels > extra_channels)
1411 				n_channels -= extra_channels;
1412 			if (ef4_separate_tx_channels) {
1413 				efx->n_tx_channels = min(max(n_channels / 2,
1414 							     1U),
1415 							 efx->max_tx_channels);
1416 				efx->n_rx_channels = max(n_channels -
1417 							 efx->n_tx_channels,
1418 							 1U);
1419 			} else {
1420 				efx->n_tx_channels = min(n_channels,
1421 							 efx->max_tx_channels);
1422 				efx->n_rx_channels = n_channels;
1423 			}
1424 			for (i = 0; i < efx->n_channels; i++)
1425 				ef4_get_channel(efx, i)->irq =
1426 					xentries[i].vector;
1427 		}
1428 	}
1429 
1430 	/* Try single interrupt MSI */
1431 	if (efx->interrupt_mode == EF4_INT_MODE_MSI) {
1432 		efx->n_channels = 1;
1433 		efx->n_rx_channels = 1;
1434 		efx->n_tx_channels = 1;
1435 		rc = pci_enable_msi(efx->pci_dev);
1436 		if (rc == 0) {
1437 			ef4_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1438 		} else {
1439 			netif_err(efx, drv, efx->net_dev,
1440 				  "could not enable MSI\n");
1441 			efx->interrupt_mode = EF4_INT_MODE_LEGACY;
1442 		}
1443 	}
1444 
1445 	/* Assume legacy interrupts */
1446 	if (efx->interrupt_mode == EF4_INT_MODE_LEGACY) {
1447 		efx->n_channels = 1 + (ef4_separate_tx_channels ? 1 : 0);
1448 		efx->n_rx_channels = 1;
1449 		efx->n_tx_channels = 1;
1450 		efx->legacy_irq = efx->pci_dev->irq;
1451 	}
1452 
1453 	/* Assign extra channels if possible */
1454 	j = efx->n_channels;
1455 	for (i = 0; i < EF4_MAX_EXTRA_CHANNELS; i++) {
1456 		if (!efx->extra_channel_type[i])
1457 			continue;
1458 		if (efx->interrupt_mode != EF4_INT_MODE_MSIX ||
1459 		    efx->n_channels <= extra_channels) {
1460 			efx->extra_channel_type[i]->handle_no_channel(efx);
1461 		} else {
1462 			--j;
1463 			ef4_get_channel(efx, j)->type =
1464 				efx->extra_channel_type[i];
1465 		}
1466 	}
1467 
1468 	efx->rss_spread = efx->n_rx_channels;
1469 
1470 	return 0;
1471 }
1472 
1473 static int ef4_soft_enable_interrupts(struct ef4_nic *efx)
1474 {
1475 	struct ef4_channel *channel, *end_channel;
1476 	int rc;
1477 
1478 	BUG_ON(efx->state == STATE_DISABLED);
1479 
1480 	efx->irq_soft_enabled = true;
1481 	smp_wmb();
1482 
1483 	ef4_for_each_channel(channel, efx) {
1484 		if (!channel->type->keep_eventq) {
1485 			rc = ef4_init_eventq(channel);
1486 			if (rc)
1487 				goto fail;
1488 		}
1489 		ef4_start_eventq(channel);
1490 	}
1491 
1492 	return 0;
1493 fail:
1494 	end_channel = channel;
1495 	ef4_for_each_channel(channel, efx) {
1496 		if (channel == end_channel)
1497 			break;
1498 		ef4_stop_eventq(channel);
1499 		if (!channel->type->keep_eventq)
1500 			ef4_fini_eventq(channel);
1501 	}
1502 
1503 	return rc;
1504 }
1505 
1506 static void ef4_soft_disable_interrupts(struct ef4_nic *efx)
1507 {
1508 	struct ef4_channel *channel;
1509 
1510 	if (efx->state == STATE_DISABLED)
1511 		return;
1512 
1513 	efx->irq_soft_enabled = false;
1514 	smp_wmb();
1515 
1516 	if (efx->legacy_irq)
1517 		synchronize_irq(efx->legacy_irq);
1518 
1519 	ef4_for_each_channel(channel, efx) {
1520 		if (channel->irq)
1521 			synchronize_irq(channel->irq);
1522 
1523 		ef4_stop_eventq(channel);
1524 		if (!channel->type->keep_eventq)
1525 			ef4_fini_eventq(channel);
1526 	}
1527 }
1528 
1529 static int ef4_enable_interrupts(struct ef4_nic *efx)
1530 {
1531 	struct ef4_channel *channel, *end_channel;
1532 	int rc;
1533 
1534 	BUG_ON(efx->state == STATE_DISABLED);
1535 
1536 	if (efx->eeh_disabled_legacy_irq) {
1537 		enable_irq(efx->legacy_irq);
1538 		efx->eeh_disabled_legacy_irq = false;
1539 	}
1540 
1541 	efx->type->irq_enable_master(efx);
1542 
1543 	ef4_for_each_channel(channel, efx) {
1544 		if (channel->type->keep_eventq) {
1545 			rc = ef4_init_eventq(channel);
1546 			if (rc)
1547 				goto fail;
1548 		}
1549 	}
1550 
1551 	rc = ef4_soft_enable_interrupts(efx);
1552 	if (rc)
1553 		goto fail;
1554 
1555 	return 0;
1556 
1557 fail:
1558 	end_channel = channel;
1559 	ef4_for_each_channel(channel, efx) {
1560 		if (channel == end_channel)
1561 			break;
1562 		if (channel->type->keep_eventq)
1563 			ef4_fini_eventq(channel);
1564 	}
1565 
1566 	efx->type->irq_disable_non_ev(efx);
1567 
1568 	return rc;
1569 }
1570 
1571 static void ef4_disable_interrupts(struct ef4_nic *efx)
1572 {
1573 	struct ef4_channel *channel;
1574 
1575 	ef4_soft_disable_interrupts(efx);
1576 
1577 	ef4_for_each_channel(channel, efx) {
1578 		if (channel->type->keep_eventq)
1579 			ef4_fini_eventq(channel);
1580 	}
1581 
1582 	efx->type->irq_disable_non_ev(efx);
1583 }
1584 
1585 static void ef4_remove_interrupts(struct ef4_nic *efx)
1586 {
1587 	struct ef4_channel *channel;
1588 
1589 	/* Remove MSI/MSI-X interrupts */
1590 	ef4_for_each_channel(channel, efx)
1591 		channel->irq = 0;
1592 	pci_disable_msi(efx->pci_dev);
1593 	pci_disable_msix(efx->pci_dev);
1594 
1595 	/* Remove legacy interrupt */
1596 	efx->legacy_irq = 0;
1597 }
1598 
1599 static void ef4_set_channels(struct ef4_nic *efx)
1600 {
1601 	struct ef4_channel *channel;
1602 	struct ef4_tx_queue *tx_queue;
1603 
1604 	efx->tx_channel_offset =
1605 		ef4_separate_tx_channels ?
1606 		efx->n_channels - efx->n_tx_channels : 0;
1607 
1608 	/* We need to mark which channels really have RX and TX
1609 	 * queues, and adjust the TX queue numbers if we have separate
1610 	 * RX-only and TX-only channels.
1611 	 */
1612 	ef4_for_each_channel(channel, efx) {
1613 		if (channel->channel < efx->n_rx_channels)
1614 			channel->rx_queue.core_index = channel->channel;
1615 		else
1616 			channel->rx_queue.core_index = -1;
1617 
1618 		ef4_for_each_channel_tx_queue(tx_queue, channel)
1619 			tx_queue->queue -= (efx->tx_channel_offset *
1620 					    EF4_TXQ_TYPES);
1621 	}
1622 }
1623 
1624 static int ef4_probe_nic(struct ef4_nic *efx)
1625 {
1626 	int rc;
1627 
1628 	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1629 
1630 	/* Carry out hardware-type specific initialisation */
1631 	rc = efx->type->probe(efx);
1632 	if (rc)
1633 		return rc;
1634 
1635 	do {
1636 		if (!efx->max_channels || !efx->max_tx_channels) {
1637 			netif_err(efx, drv, efx->net_dev,
1638 				  "Insufficient resources to allocate"
1639 				  " any channels\n");
1640 			rc = -ENOSPC;
1641 			goto fail1;
1642 		}
1643 
1644 		/* Determine the number of channels and queues by trying
1645 		 * to hook in MSI-X interrupts.
1646 		 */
1647 		rc = ef4_probe_interrupts(efx);
1648 		if (rc)
1649 			goto fail1;
1650 
1651 		ef4_set_channels(efx);
1652 
1653 		/* dimension_resources can fail with EAGAIN */
1654 		rc = efx->type->dimension_resources(efx);
1655 		if (rc != 0 && rc != -EAGAIN)
1656 			goto fail2;
1657 
1658 		if (rc == -EAGAIN)
1659 			/* try again with new max_channels */
1660 			ef4_remove_interrupts(efx);
1661 
1662 	} while (rc == -EAGAIN);
1663 
1664 	if (efx->n_channels > 1)
1665 		netdev_rss_key_fill(&efx->rx_hash_key,
1666 				    sizeof(efx->rx_hash_key));
1667 	ef4_set_default_rx_indir_table(efx);
1668 
1669 	netif_set_real_num_tx_queues(efx->net_dev, efx->n_tx_channels);
1670 	netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels);
1671 
1672 	/* Initialise the interrupt moderation settings */
1673 	efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000);
1674 	ef4_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
1675 				true);
1676 
1677 	return 0;
1678 
1679 fail2:
1680 	ef4_remove_interrupts(efx);
1681 fail1:
1682 	efx->type->remove(efx);
1683 	return rc;
1684 }
1685 
1686 static void ef4_remove_nic(struct ef4_nic *efx)
1687 {
1688 	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1689 
1690 	ef4_remove_interrupts(efx);
1691 	efx->type->remove(efx);
1692 }
1693 
1694 static int ef4_probe_filters(struct ef4_nic *efx)
1695 {
1696 	int rc;
1697 
1698 	spin_lock_init(&efx->filter_lock);
1699 	init_rwsem(&efx->filter_sem);
1700 	mutex_lock(&efx->mac_lock);
1701 	down_write(&efx->filter_sem);
1702 	rc = efx->type->filter_table_probe(efx);
1703 	if (rc)
1704 		goto out_unlock;
1705 
1706 #ifdef CONFIG_RFS_ACCEL
1707 	if (efx->type->offload_features & NETIF_F_NTUPLE) {
1708 		struct ef4_channel *channel;
1709 		int i, success = 1;
1710 
1711 		ef4_for_each_channel(channel, efx) {
1712 			channel->rps_flow_id =
1713 				kcalloc(efx->type->max_rx_ip_filters,
1714 					sizeof(*channel->rps_flow_id),
1715 					GFP_KERNEL);
1716 			if (!channel->rps_flow_id)
1717 				success = 0;
1718 			else
1719 				for (i = 0;
1720 				     i < efx->type->max_rx_ip_filters;
1721 				     ++i)
1722 					channel->rps_flow_id[i] =
1723 						RPS_FLOW_ID_INVALID;
1724 		}
1725 
1726 		if (!success) {
1727 			ef4_for_each_channel(channel, efx)
1728 				kfree(channel->rps_flow_id);
1729 			efx->type->filter_table_remove(efx);
1730 			rc = -ENOMEM;
1731 			goto out_unlock;
1732 		}
1733 
1734 		efx->rps_expire_index = efx->rps_expire_channel = 0;
1735 	}
1736 #endif
1737 out_unlock:
1738 	up_write(&efx->filter_sem);
1739 	mutex_unlock(&efx->mac_lock);
1740 	return rc;
1741 }
1742 
1743 static void ef4_remove_filters(struct ef4_nic *efx)
1744 {
1745 #ifdef CONFIG_RFS_ACCEL
1746 	struct ef4_channel *channel;
1747 
1748 	ef4_for_each_channel(channel, efx)
1749 		kfree(channel->rps_flow_id);
1750 #endif
1751 	down_write(&efx->filter_sem);
1752 	efx->type->filter_table_remove(efx);
1753 	up_write(&efx->filter_sem);
1754 }
1755 
1756 static void ef4_restore_filters(struct ef4_nic *efx)
1757 {
1758 	down_read(&efx->filter_sem);
1759 	efx->type->filter_table_restore(efx);
1760 	up_read(&efx->filter_sem);
1761 }
1762 
1763 /**************************************************************************
1764  *
1765  * NIC startup/shutdown
1766  *
1767  *************************************************************************/
1768 
1769 static int ef4_probe_all(struct ef4_nic *efx)
1770 {
1771 	int rc;
1772 
1773 	rc = ef4_probe_nic(efx);
1774 	if (rc) {
1775 		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1776 		goto fail1;
1777 	}
1778 
1779 	rc = ef4_probe_port(efx);
1780 	if (rc) {
1781 		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1782 		goto fail2;
1783 	}
1784 
1785 	BUILD_BUG_ON(EF4_DEFAULT_DMAQ_SIZE < EF4_RXQ_MIN_ENT);
1786 	if (WARN_ON(EF4_DEFAULT_DMAQ_SIZE < EF4_TXQ_MIN_ENT(efx))) {
1787 		rc = -EINVAL;
1788 		goto fail3;
1789 	}
1790 	efx->rxq_entries = efx->txq_entries = EF4_DEFAULT_DMAQ_SIZE;
1791 
1792 	rc = ef4_probe_filters(efx);
1793 	if (rc) {
1794 		netif_err(efx, probe, efx->net_dev,
1795 			  "failed to create filter tables\n");
1796 		goto fail4;
1797 	}
1798 
1799 	rc = ef4_probe_channels(efx);
1800 	if (rc)
1801 		goto fail5;
1802 
1803 	return 0;
1804 
1805  fail5:
1806 	ef4_remove_filters(efx);
1807  fail4:
1808  fail3:
1809 	ef4_remove_port(efx);
1810  fail2:
1811 	ef4_remove_nic(efx);
1812  fail1:
1813 	return rc;
1814 }
1815 
1816 /* If the interface is supposed to be running but is not, start
1817  * the hardware and software data path, regular activity for the port
1818  * (MAC statistics, link polling, etc.) and schedule the port to be
1819  * reconfigured.  Interrupts must already be enabled.  This function
1820  * is safe to call multiple times, so long as the NIC is not disabled.
1821  * Requires the RTNL lock.
1822  */
1823 static void ef4_start_all(struct ef4_nic *efx)
1824 {
1825 	EF4_ASSERT_RESET_SERIALISED(efx);
1826 	BUG_ON(efx->state == STATE_DISABLED);
1827 
1828 	/* Check that it is appropriate to restart the interface. All
1829 	 * of these flags are safe to read under just the rtnl lock */
1830 	if (efx->port_enabled || !netif_running(efx->net_dev) ||
1831 	    efx->reset_pending)
1832 		return;
1833 
1834 	ef4_start_port(efx);
1835 	ef4_start_datapath(efx);
1836 
1837 	/* Start the hardware monitor if there is one */
1838 	if (efx->type->monitor != NULL)
1839 		queue_delayed_work(efx->workqueue, &efx->monitor_work,
1840 				   ef4_monitor_interval);
1841 
1842 	efx->type->start_stats(efx);
1843 	efx->type->pull_stats(efx);
1844 	spin_lock_bh(&efx->stats_lock);
1845 	efx->type->update_stats(efx, NULL, NULL);
1846 	spin_unlock_bh(&efx->stats_lock);
1847 }
1848 
1849 /* Quiesce the hardware and software data path, and regular activity
1850  * for the port without bringing the link down.  Safe to call multiple
1851  * times with the NIC in almost any state, but interrupts should be
1852  * enabled.  Requires the RTNL lock.
1853  */
1854 static void ef4_stop_all(struct ef4_nic *efx)
1855 {
1856 	EF4_ASSERT_RESET_SERIALISED(efx);
1857 
1858 	/* port_enabled can be read safely under the rtnl lock */
1859 	if (!efx->port_enabled)
1860 		return;
1861 
1862 	/* update stats before we go down so we can accurately count
1863 	 * rx_nodesc_drops
1864 	 */
1865 	efx->type->pull_stats(efx);
1866 	spin_lock_bh(&efx->stats_lock);
1867 	efx->type->update_stats(efx, NULL, NULL);
1868 	spin_unlock_bh(&efx->stats_lock);
1869 	efx->type->stop_stats(efx);
1870 	ef4_stop_port(efx);
1871 
1872 	/* Stop the kernel transmit interface.  This is only valid if
1873 	 * the device is stopped or detached; otherwise the watchdog
1874 	 * may fire immediately.
1875 	 */
1876 	WARN_ON(netif_running(efx->net_dev) &&
1877 		netif_device_present(efx->net_dev));
1878 	netif_tx_disable(efx->net_dev);
1879 
1880 	ef4_stop_datapath(efx);
1881 }
1882 
1883 static void ef4_remove_all(struct ef4_nic *efx)
1884 {
1885 	ef4_remove_channels(efx);
1886 	ef4_remove_filters(efx);
1887 	ef4_remove_port(efx);
1888 	ef4_remove_nic(efx);
1889 }
1890 
1891 /**************************************************************************
1892  *
1893  * Interrupt moderation
1894  *
1895  **************************************************************************/
1896 unsigned int ef4_usecs_to_ticks(struct ef4_nic *efx, unsigned int usecs)
1897 {
1898 	if (usecs == 0)
1899 		return 0;
1900 	if (usecs * 1000 < efx->timer_quantum_ns)
1901 		return 1; /* never round down to 0 */
1902 	return usecs * 1000 / efx->timer_quantum_ns;
1903 }
1904 
1905 unsigned int ef4_ticks_to_usecs(struct ef4_nic *efx, unsigned int ticks)
1906 {
1907 	/* We must round up when converting ticks to microseconds
1908 	 * because we round down when converting the other way.
1909 	 */
1910 	return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000);
1911 }
1912 
1913 /* Set interrupt moderation parameters */
1914 int ef4_init_irq_moderation(struct ef4_nic *efx, unsigned int tx_usecs,
1915 			    unsigned int rx_usecs, bool rx_adaptive,
1916 			    bool rx_may_override_tx)
1917 {
1918 	struct ef4_channel *channel;
1919 	unsigned int timer_max_us;
1920 
1921 	EF4_ASSERT_RESET_SERIALISED(efx);
1922 
1923 	timer_max_us = efx->timer_max_ns / 1000;
1924 
1925 	if (tx_usecs > timer_max_us || rx_usecs > timer_max_us)
1926 		return -EINVAL;
1927 
1928 	if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 &&
1929 	    !rx_may_override_tx) {
1930 		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
1931 			  "RX and TX IRQ moderation must be equal\n");
1932 		return -EINVAL;
1933 	}
1934 
1935 	efx->irq_rx_adaptive = rx_adaptive;
1936 	efx->irq_rx_moderation_us = rx_usecs;
1937 	ef4_for_each_channel(channel, efx) {
1938 		if (ef4_channel_has_rx_queue(channel))
1939 			channel->irq_moderation_us = rx_usecs;
1940 		else if (ef4_channel_has_tx_queues(channel))
1941 			channel->irq_moderation_us = tx_usecs;
1942 	}
1943 
1944 	return 0;
1945 }
1946 
1947 void ef4_get_irq_moderation(struct ef4_nic *efx, unsigned int *tx_usecs,
1948 			    unsigned int *rx_usecs, bool *rx_adaptive)
1949 {
1950 	*rx_adaptive = efx->irq_rx_adaptive;
1951 	*rx_usecs = efx->irq_rx_moderation_us;
1952 
1953 	/* If channels are shared between RX and TX, so is IRQ
1954 	 * moderation.  Otherwise, IRQ moderation is the same for all
1955 	 * TX channels and is not adaptive.
1956 	 */
1957 	if (efx->tx_channel_offset == 0) {
1958 		*tx_usecs = *rx_usecs;
1959 	} else {
1960 		struct ef4_channel *tx_channel;
1961 
1962 		tx_channel = efx->channel[efx->tx_channel_offset];
1963 		*tx_usecs = tx_channel->irq_moderation_us;
1964 	}
1965 }
1966 
1967 /**************************************************************************
1968  *
1969  * Hardware monitor
1970  *
1971  **************************************************************************/
1972 
1973 /* Run periodically off the general workqueue */
1974 static void ef4_monitor(struct work_struct *data)
1975 {
1976 	struct ef4_nic *efx = container_of(data, struct ef4_nic,
1977 					   monitor_work.work);
1978 
1979 	netif_vdbg(efx, timer, efx->net_dev,
1980 		   "hardware monitor executing on CPU %d\n",
1981 		   raw_smp_processor_id());
1982 	BUG_ON(efx->type->monitor == NULL);
1983 
1984 	/* If the mac_lock is already held then it is likely a port
1985 	 * reconfiguration is already in place, which will likely do
1986 	 * most of the work of monitor() anyway. */
1987 	if (mutex_trylock(&efx->mac_lock)) {
1988 		if (efx->port_enabled)
1989 			efx->type->monitor(efx);
1990 		mutex_unlock(&efx->mac_lock);
1991 	}
1992 
1993 	queue_delayed_work(efx->workqueue, &efx->monitor_work,
1994 			   ef4_monitor_interval);
1995 }
1996 
1997 /**************************************************************************
1998  *
1999  * ioctls
2000  *
2001  *************************************************************************/
2002 
2003 /* Net device ioctl
2004  * Context: process, rtnl_lock() held.
2005  */
2006 static int ef4_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
2007 {
2008 	struct ef4_nic *efx = netdev_priv(net_dev);
2009 	struct mii_ioctl_data *data = if_mii(ifr);
2010 
2011 	/* Convert phy_id from older PRTAD/DEVAD format */
2012 	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
2013 	    (data->phy_id & 0xfc00) == 0x0400)
2014 		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;
2015 
2016 	return mdio_mii_ioctl(&efx->mdio, data, cmd);
2017 }
2018 
2019 /**************************************************************************
2020  *
2021  * NAPI interface
2022  *
2023  **************************************************************************/
2024 
2025 static void ef4_init_napi_channel(struct ef4_channel *channel)
2026 {
2027 	struct ef4_nic *efx = channel->efx;
2028 
2029 	channel->napi_dev = efx->net_dev;
2030 	netif_napi_add(channel->napi_dev, &channel->napi_str,
2031 		       ef4_poll, napi_weight);
2032 	ef4_channel_busy_poll_init(channel);
2033 }
2034 
2035 static void ef4_init_napi(struct ef4_nic *efx)
2036 {
2037 	struct ef4_channel *channel;
2038 
2039 	ef4_for_each_channel(channel, efx)
2040 		ef4_init_napi_channel(channel);
2041 }
2042 
2043 static void ef4_fini_napi_channel(struct ef4_channel *channel)
2044 {
2045 	if (channel->napi_dev)
2046 		netif_napi_del(&channel->napi_str);
2047 
2048 	channel->napi_dev = NULL;
2049 }
2050 
2051 static void ef4_fini_napi(struct ef4_nic *efx)
2052 {
2053 	struct ef4_channel *channel;
2054 
2055 	ef4_for_each_channel(channel, efx)
2056 		ef4_fini_napi_channel(channel);
2057 }
2058 
2059 /**************************************************************************
2060  *
2061  * Kernel netpoll interface
2062  *
2063  *************************************************************************/
2064 
2065 #ifdef CONFIG_NET_POLL_CONTROLLER
2066 
2067 /* Although in the common case interrupts will be disabled, this is not
2068  * guaranteed. However, all our work happens inside the NAPI callback,
2069  * so no locking is required.
2070  */
2071 static void ef4_netpoll(struct net_device *net_dev)
2072 {
2073 	struct ef4_nic *efx = netdev_priv(net_dev);
2074 	struct ef4_channel *channel;
2075 
2076 	ef4_for_each_channel(channel, efx)
2077 		ef4_schedule_channel(channel);
2078 }
2079 
2080 #endif
2081 
2082 #ifdef CONFIG_NET_RX_BUSY_POLL
2083 static int ef4_busy_poll(struct napi_struct *napi)
2084 {
2085 	struct ef4_channel *channel =
2086 		container_of(napi, struct ef4_channel, napi_str);
2087 	struct ef4_nic *efx = channel->efx;
2088 	int budget = 4;
2089 	int old_rx_packets, rx_packets;
2090 
2091 	if (!netif_running(efx->net_dev))
2092 		return LL_FLUSH_FAILED;
2093 
2094 	if (!ef4_channel_try_lock_poll(channel))
2095 		return LL_FLUSH_BUSY;
2096 
2097 	old_rx_packets = channel->rx_queue.rx_packets;
2098 	ef4_process_channel(channel, budget);
2099 
2100 	rx_packets = channel->rx_queue.rx_packets - old_rx_packets;
2101 
2102 	/* There is no race condition with NAPI here.
2103 	 * NAPI will automatically be rescheduled if it yielded during busy
2104 	 * polling, because it was not able to take the lock and thus returned
2105 	 * the full budget.
2106 	 */
2107 	ef4_channel_unlock_poll(channel);
2108 
2109 	return rx_packets;
2110 }
2111 #endif
2112 
2113 /**************************************************************************
2114  *
2115  * Kernel net device interface
2116  *
2117  *************************************************************************/
2118 
2119 /* Context: process, rtnl_lock() held. */
2120 int ef4_net_open(struct net_device *net_dev)
2121 {
2122 	struct ef4_nic *efx = netdev_priv(net_dev);
2123 	int rc;
2124 
2125 	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
2126 		  raw_smp_processor_id());
2127 
2128 	rc = ef4_check_disabled(efx);
2129 	if (rc)
2130 		return rc;
2131 	if (efx->phy_mode & PHY_MODE_SPECIAL)
2132 		return -EBUSY;
2133 
2134 	/* Notify the kernel of the link state polled during driver load,
2135 	 * before the monitor starts running */
2136 	ef4_link_status_changed(efx);
2137 
2138 	ef4_start_all(efx);
2139 	ef4_selftest_async_start(efx);
2140 	return 0;
2141 }
2142 
2143 /* Context: process, rtnl_lock() held.
2144  * Note that the kernel will ignore our return code; this method
2145  * should really be a void.
2146  */
2147 int ef4_net_stop(struct net_device *net_dev)
2148 {
2149 	struct ef4_nic *efx = netdev_priv(net_dev);
2150 
2151 	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
2152 		  raw_smp_processor_id());
2153 
2154 	/* Stop the device and flush all the channels */
2155 	ef4_stop_all(efx);
2156 
2157 	return 0;
2158 }
2159 
2160 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
2161 static struct rtnl_link_stats64 *ef4_net_stats(struct net_device *net_dev,
2162 					       struct rtnl_link_stats64 *stats)
2163 {
2164 	struct ef4_nic *efx = netdev_priv(net_dev);
2165 
2166 	spin_lock_bh(&efx->stats_lock);
2167 	efx->type->update_stats(efx, NULL, stats);
2168 	spin_unlock_bh(&efx->stats_lock);
2169 
2170 	return stats;
2171 }
2172 
2173 /* Context: netif_tx_lock held, BHs disabled. */
2174 static void ef4_watchdog(struct net_device *net_dev)
2175 {
2176 	struct ef4_nic *efx = netdev_priv(net_dev);
2177 
2178 	netif_err(efx, tx_err, efx->net_dev,
2179 		  "TX stuck with port_enabled=%d: resetting channels\n",
2180 		  efx->port_enabled);
2181 
2182 	ef4_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2183 }
2184 
2185 
2186 /* Context: process, rtnl_lock() held. */
2187 static int ef4_change_mtu(struct net_device *net_dev, int new_mtu)
2188 {
2189 	struct ef4_nic *efx = netdev_priv(net_dev);
2190 	int rc;
2191 
2192 	rc = ef4_check_disabled(efx);
2193 	if (rc)
2194 		return rc;
2195 
2196 	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2197 
2198 	ef4_device_detach_sync(efx);
2199 	ef4_stop_all(efx);
2200 
2201 	mutex_lock(&efx->mac_lock);
2202 	net_dev->mtu = new_mtu;
2203 	ef4_mac_reconfigure(efx);
2204 	mutex_unlock(&efx->mac_lock);
2205 
2206 	ef4_start_all(efx);
2207 	netif_device_attach(efx->net_dev);
2208 	return 0;
2209 }
2210 
2211 static int ef4_set_mac_address(struct net_device *net_dev, void *data)
2212 {
2213 	struct ef4_nic *efx = netdev_priv(net_dev);
2214 	struct sockaddr *addr = data;
2215 	u8 *new_addr = addr->sa_data;
2216 	u8 old_addr[6];
2217 	int rc;
2218 
2219 	if (!is_valid_ether_addr(new_addr)) {
2220 		netif_err(efx, drv, efx->net_dev,
2221 			  "invalid ethernet MAC address requested: %pM\n",
2222 			  new_addr);
2223 		return -EADDRNOTAVAIL;
2224 	}
2225 
2226 	/* save old address */
2227 	ether_addr_copy(old_addr, net_dev->dev_addr);
2228 	ether_addr_copy(net_dev->dev_addr, new_addr);
2229 	if (efx->type->set_mac_address) {
2230 		rc = efx->type->set_mac_address(efx);
2231 		if (rc) {
2232 			ether_addr_copy(net_dev->dev_addr, old_addr);
2233 			return rc;
2234 		}
2235 	}
2236 
2237 	/* Reconfigure the MAC */
2238 	mutex_lock(&efx->mac_lock);
2239 	ef4_mac_reconfigure(efx);
2240 	mutex_unlock(&efx->mac_lock);
2241 
2242 	return 0;
2243 }
2244 
2245 /* Context: netif_addr_lock held, BHs disabled. */
2246 static void ef4_set_rx_mode(struct net_device *net_dev)
2247 {
2248 	struct ef4_nic *efx = netdev_priv(net_dev);
2249 
2250 	if (efx->port_enabled)
2251 		queue_work(efx->workqueue, &efx->mac_work);
2252 	/* Otherwise ef4_start_port() will do this */
2253 }
2254 
2255 static int ef4_set_features(struct net_device *net_dev, netdev_features_t data)
2256 {
2257 	struct ef4_nic *efx = netdev_priv(net_dev);
2258 	int rc;
2259 
2260 	/* If disabling RX n-tuple filtering, clear existing filters */
2261 	if (net_dev->features & ~data & NETIF_F_NTUPLE) {
2262 		rc = efx->type->filter_clear_rx(efx, EF4_FILTER_PRI_MANUAL);
2263 		if (rc)
2264 			return rc;
2265 	}
2266 
2267 	/* If Rx VLAN filter is changed, update filters via mac_reconfigure */
2268 	if ((net_dev->features ^ data) & NETIF_F_HW_VLAN_CTAG_FILTER) {
2269 		/* ef4_set_rx_mode() will schedule MAC work to update filters
2270 		 * when a new features are finally set in net_dev.
2271 		 */
2272 		ef4_set_rx_mode(net_dev);
2273 	}
2274 
2275 	return 0;
2276 }
2277 
2278 static const struct net_device_ops ef4_netdev_ops = {
2279 	.ndo_open		= ef4_net_open,
2280 	.ndo_stop		= ef4_net_stop,
2281 	.ndo_get_stats64	= ef4_net_stats,
2282 	.ndo_tx_timeout		= ef4_watchdog,
2283 	.ndo_start_xmit		= ef4_hard_start_xmit,
2284 	.ndo_validate_addr	= eth_validate_addr,
2285 	.ndo_do_ioctl		= ef4_ioctl,
2286 	.ndo_change_mtu		= ef4_change_mtu,
2287 	.ndo_set_mac_address	= ef4_set_mac_address,
2288 	.ndo_set_rx_mode	= ef4_set_rx_mode,
2289 	.ndo_set_features	= ef4_set_features,
2290 #ifdef CONFIG_NET_POLL_CONTROLLER
2291 	.ndo_poll_controller = ef4_netpoll,
2292 #endif
2293 	.ndo_setup_tc		= ef4_setup_tc,
2294 #ifdef CONFIG_NET_RX_BUSY_POLL
2295 	.ndo_busy_poll		= ef4_busy_poll,
2296 #endif
2297 #ifdef CONFIG_RFS_ACCEL
2298 	.ndo_rx_flow_steer	= ef4_filter_rfs,
2299 #endif
2300 };
2301 
2302 static void ef4_update_name(struct ef4_nic *efx)
2303 {
2304 	strcpy(efx->name, efx->net_dev->name);
2305 	ef4_mtd_rename(efx);
2306 	ef4_set_channel_names(efx);
2307 }
2308 
2309 static int ef4_netdev_event(struct notifier_block *this,
2310 			    unsigned long event, void *ptr)
2311 {
2312 	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2313 
2314 	if ((net_dev->netdev_ops == &ef4_netdev_ops) &&
2315 	    event == NETDEV_CHANGENAME)
2316 		ef4_update_name(netdev_priv(net_dev));
2317 
2318 	return NOTIFY_DONE;
2319 }
2320 
2321 static struct notifier_block ef4_netdev_notifier = {
2322 	.notifier_call = ef4_netdev_event,
2323 };
2324 
2325 static ssize_t
2326 show_phy_type(struct device *dev, struct device_attribute *attr, char *buf)
2327 {
2328 	struct ef4_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2329 	return sprintf(buf, "%d\n", efx->phy_type);
2330 }
2331 static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
2332 
2333 static int ef4_register_netdev(struct ef4_nic *efx)
2334 {
2335 	struct net_device *net_dev = efx->net_dev;
2336 	struct ef4_channel *channel;
2337 	int rc;
2338 
2339 	net_dev->watchdog_timeo = 5 * HZ;
2340 	net_dev->irq = efx->pci_dev->irq;
2341 	net_dev->netdev_ops = &ef4_netdev_ops;
2342 	net_dev->ethtool_ops = &ef4_ethtool_ops;
2343 	net_dev->gso_max_segs = EF4_TSO_MAX_SEGS;
2344 	net_dev->min_mtu = EF4_MIN_MTU;
2345 	net_dev->max_mtu = EF4_MAX_MTU;
2346 
2347 	rtnl_lock();
2348 
2349 	/* Enable resets to be scheduled and check whether any were
2350 	 * already requested.  If so, the NIC is probably hosed so we
2351 	 * abort.
2352 	 */
2353 	efx->state = STATE_READY;
2354 	smp_mb(); /* ensure we change state before checking reset_pending */
2355 	if (efx->reset_pending) {
2356 		netif_err(efx, probe, efx->net_dev,
2357 			  "aborting probe due to scheduled reset\n");
2358 		rc = -EIO;
2359 		goto fail_locked;
2360 	}
2361 
2362 	rc = dev_alloc_name(net_dev, net_dev->name);
2363 	if (rc < 0)
2364 		goto fail_locked;
2365 	ef4_update_name(efx);
2366 
2367 	/* Always start with carrier off; PHY events will detect the link */
2368 	netif_carrier_off(net_dev);
2369 
2370 	rc = register_netdevice(net_dev);
2371 	if (rc)
2372 		goto fail_locked;
2373 
2374 	ef4_for_each_channel(channel, efx) {
2375 		struct ef4_tx_queue *tx_queue;
2376 		ef4_for_each_channel_tx_queue(tx_queue, channel)
2377 			ef4_init_tx_queue_core_txq(tx_queue);
2378 	}
2379 
2380 	ef4_associate(efx);
2381 
2382 	rtnl_unlock();
2383 
2384 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2385 	if (rc) {
2386 		netif_err(efx, drv, efx->net_dev,
2387 			  "failed to init net dev attributes\n");
2388 		goto fail_registered;
2389 	}
2390 	return 0;
2391 
2392 fail_registered:
2393 	rtnl_lock();
2394 	ef4_dissociate(efx);
2395 	unregister_netdevice(net_dev);
2396 fail_locked:
2397 	efx->state = STATE_UNINIT;
2398 	rtnl_unlock();
2399 	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2400 	return rc;
2401 }
2402 
2403 static void ef4_unregister_netdev(struct ef4_nic *efx)
2404 {
2405 	if (!efx->net_dev)
2406 		return;
2407 
2408 	BUG_ON(netdev_priv(efx->net_dev) != efx);
2409 
2410 	if (ef4_dev_registered(efx)) {
2411 		strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
2412 		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2413 		unregister_netdev(efx->net_dev);
2414 	}
2415 }
2416 
2417 /**************************************************************************
2418  *
2419  * Device reset and suspend
2420  *
2421  **************************************************************************/
2422 
2423 /* Tears down the entire software state and most of the hardware state
2424  * before reset.  */
2425 void ef4_reset_down(struct ef4_nic *efx, enum reset_type method)
2426 {
2427 	EF4_ASSERT_RESET_SERIALISED(efx);
2428 
2429 	ef4_stop_all(efx);
2430 	ef4_disable_interrupts(efx);
2431 
2432 	mutex_lock(&efx->mac_lock);
2433 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2434 	    method != RESET_TYPE_DATAPATH)
2435 		efx->phy_op->fini(efx);
2436 	efx->type->fini(efx);
2437 }
2438 
2439 /* This function will always ensure that the locks acquired in
2440  * ef4_reset_down() are released. A failure return code indicates
2441  * that we were unable to reinitialise the hardware, and the
2442  * driver should be disabled. If ok is false, then the rx and tx
2443  * engines are not restarted, pending a RESET_DISABLE. */
2444 int ef4_reset_up(struct ef4_nic *efx, enum reset_type method, bool ok)
2445 {
2446 	int rc;
2447 
2448 	EF4_ASSERT_RESET_SERIALISED(efx);
2449 
2450 	/* Ensure that SRAM is initialised even if we're disabling the device */
2451 	rc = efx->type->init(efx);
2452 	if (rc) {
2453 		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2454 		goto fail;
2455 	}
2456 
2457 	if (!ok)
2458 		goto fail;
2459 
2460 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2461 	    method != RESET_TYPE_DATAPATH) {
2462 		rc = efx->phy_op->init(efx);
2463 		if (rc)
2464 			goto fail;
2465 		rc = efx->phy_op->reconfigure(efx);
2466 		if (rc && rc != -EPERM)
2467 			netif_err(efx, drv, efx->net_dev,
2468 				  "could not restore PHY settings\n");
2469 	}
2470 
2471 	rc = ef4_enable_interrupts(efx);
2472 	if (rc)
2473 		goto fail;
2474 
2475 	down_read(&efx->filter_sem);
2476 	ef4_restore_filters(efx);
2477 	up_read(&efx->filter_sem);
2478 
2479 	mutex_unlock(&efx->mac_lock);
2480 
2481 	ef4_start_all(efx);
2482 
2483 	return 0;
2484 
2485 fail:
2486 	efx->port_initialized = false;
2487 
2488 	mutex_unlock(&efx->mac_lock);
2489 
2490 	return rc;
2491 }
2492 
2493 /* Reset the NIC using the specified method.  Note that the reset may
2494  * fail, in which case the card will be left in an unusable state.
2495  *
2496  * Caller must hold the rtnl_lock.
2497  */
2498 int ef4_reset(struct ef4_nic *efx, enum reset_type method)
2499 {
2500 	int rc, rc2;
2501 	bool disabled;
2502 
2503 	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
2504 		   RESET_TYPE(method));
2505 
2506 	ef4_device_detach_sync(efx);
2507 	ef4_reset_down(efx, method);
2508 
2509 	rc = efx->type->reset(efx, method);
2510 	if (rc) {
2511 		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2512 		goto out;
2513 	}
2514 
2515 	/* Clear flags for the scopes we covered.  We assume the NIC and
2516 	 * driver are now quiescent so that there is no race here.
2517 	 */
2518 	if (method < RESET_TYPE_MAX_METHOD)
2519 		efx->reset_pending &= -(1 << (method + 1));
2520 	else /* it doesn't fit into the well-ordered scope hierarchy */
2521 		__clear_bit(method, &efx->reset_pending);
2522 
2523 	/* Reinitialise bus-mastering, which may have been turned off before
2524 	 * the reset was scheduled. This is still appropriate, even in the
2525 	 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
2526 	 * can respond to requests. */
2527 	pci_set_master(efx->pci_dev);
2528 
2529 out:
2530 	/* Leave device stopped if necessary */
2531 	disabled = rc ||
2532 		method == RESET_TYPE_DISABLE ||
2533 		method == RESET_TYPE_RECOVER_OR_DISABLE;
2534 	rc2 = ef4_reset_up(efx, method, !disabled);
2535 	if (rc2) {
2536 		disabled = true;
2537 		if (!rc)
2538 			rc = rc2;
2539 	}
2540 
2541 	if (disabled) {
2542 		dev_close(efx->net_dev);
2543 		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2544 		efx->state = STATE_DISABLED;
2545 	} else {
2546 		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2547 		netif_device_attach(efx->net_dev);
2548 	}
2549 	return rc;
2550 }
2551 
2552 /* Try recovery mechanisms.
2553  * For now only EEH is supported.
2554  * Returns 0 if the recovery mechanisms are unsuccessful.
2555  * Returns a non-zero value otherwise.
2556  */
2557 int ef4_try_recovery(struct ef4_nic *efx)
2558 {
2559 #ifdef CONFIG_EEH
2560 	/* A PCI error can occur and not be seen by EEH because nothing
2561 	 * happens on the PCI bus. In this case the driver may fail and
2562 	 * schedule a 'recover or reset', leading to this recovery handler.
2563 	 * Manually call the eeh failure check function.
2564 	 */
2565 	struct eeh_dev *eehdev = pci_dev_to_eeh_dev(efx->pci_dev);
2566 	if (eeh_dev_check_failure(eehdev)) {
2567 		/* The EEH mechanisms will handle the error and reset the
2568 		 * device if necessary.
2569 		 */
2570 		return 1;
2571 	}
2572 #endif
2573 	return 0;
2574 }
2575 
2576 /* The worker thread exists so that code that cannot sleep can
2577  * schedule a reset for later.
2578  */
2579 static void ef4_reset_work(struct work_struct *data)
2580 {
2581 	struct ef4_nic *efx = container_of(data, struct ef4_nic, reset_work);
2582 	unsigned long pending;
2583 	enum reset_type method;
2584 
2585 	pending = ACCESS_ONCE(efx->reset_pending);
2586 	method = fls(pending) - 1;
2587 
2588 	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
2589 	     method == RESET_TYPE_RECOVER_OR_ALL) &&
2590 	    ef4_try_recovery(efx))
2591 		return;
2592 
2593 	if (!pending)
2594 		return;
2595 
2596 	rtnl_lock();
2597 
2598 	/* We checked the state in ef4_schedule_reset() but it may
2599 	 * have changed by now.  Now that we have the RTNL lock,
2600 	 * it cannot change again.
2601 	 */
2602 	if (efx->state == STATE_READY)
2603 		(void)ef4_reset(efx, method);
2604 
2605 	rtnl_unlock();
2606 }
2607 
2608 void ef4_schedule_reset(struct ef4_nic *efx, enum reset_type type)
2609 {
2610 	enum reset_type method;
2611 
2612 	if (efx->state == STATE_RECOVERY) {
2613 		netif_dbg(efx, drv, efx->net_dev,
2614 			  "recovering: skip scheduling %s reset\n",
2615 			  RESET_TYPE(type));
2616 		return;
2617 	}
2618 
2619 	switch (type) {
2620 	case RESET_TYPE_INVISIBLE:
2621 	case RESET_TYPE_ALL:
2622 	case RESET_TYPE_RECOVER_OR_ALL:
2623 	case RESET_TYPE_WORLD:
2624 	case RESET_TYPE_DISABLE:
2625 	case RESET_TYPE_RECOVER_OR_DISABLE:
2626 	case RESET_TYPE_DATAPATH:
2627 		method = type;
2628 		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
2629 			  RESET_TYPE(method));
2630 		break;
2631 	default:
2632 		method = efx->type->map_reset_reason(type);
2633 		netif_dbg(efx, drv, efx->net_dev,
2634 			  "scheduling %s reset for %s\n",
2635 			  RESET_TYPE(method), RESET_TYPE(type));
2636 		break;
2637 	}
2638 
2639 	set_bit(method, &efx->reset_pending);
2640 	smp_mb(); /* ensure we change reset_pending before checking state */
2641 
2642 	/* If we're not READY then just leave the flags set as the cue
2643 	 * to abort probing or reschedule the reset later.
2644 	 */
2645 	if (ACCESS_ONCE(efx->state) != STATE_READY)
2646 		return;
2647 
2648 	queue_work(reset_workqueue, &efx->reset_work);
2649 }
2650 
2651 /**************************************************************************
2652  *
2653  * List of NICs we support
2654  *
2655  **************************************************************************/
2656 
2657 /* PCI device ID table */
2658 static const struct pci_device_id ef4_pci_table[] = {
2659 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
2660 		    PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0),
2661 	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2662 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
2663 		    PCI_DEVICE_ID_SOLARFLARE_SFC4000B),
2664 	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2665 	{0}			/* end of list */
2666 };
2667 
2668 /**************************************************************************
2669  *
2670  * Dummy PHY/MAC operations
2671  *
2672  * Can be used for some unimplemented operations
2673  * Needed so all function pointers are valid and do not have to be tested
2674  * before use
2675  *
2676  **************************************************************************/
2677 int ef4_port_dummy_op_int(struct ef4_nic *efx)
2678 {
2679 	return 0;
2680 }
2681 void ef4_port_dummy_op_void(struct ef4_nic *efx) {}
2682 
2683 static bool ef4_port_dummy_op_poll(struct ef4_nic *efx)
2684 {
2685 	return false;
2686 }
2687 
2688 static const struct ef4_phy_operations ef4_dummy_phy_operations = {
2689 	.init		 = ef4_port_dummy_op_int,
2690 	.reconfigure	 = ef4_port_dummy_op_int,
2691 	.poll		 = ef4_port_dummy_op_poll,
2692 	.fini		 = ef4_port_dummy_op_void,
2693 };
2694 
2695 /**************************************************************************
2696  *
2697  * Data housekeeping
2698  *
2699  **************************************************************************/
2700 
2701 /* This zeroes out and then fills in the invariants in a struct
2702  * ef4_nic (including all sub-structures).
2703  */
2704 static int ef4_init_struct(struct ef4_nic *efx,
2705 			   struct pci_dev *pci_dev, struct net_device *net_dev)
2706 {
2707 	int i;
2708 
2709 	/* Initialise common structures */
2710 	INIT_LIST_HEAD(&efx->node);
2711 	INIT_LIST_HEAD(&efx->secondary_list);
2712 	spin_lock_init(&efx->biu_lock);
2713 #ifdef CONFIG_SFC_FALCON_MTD
2714 	INIT_LIST_HEAD(&efx->mtd_list);
2715 #endif
2716 	INIT_WORK(&efx->reset_work, ef4_reset_work);
2717 	INIT_DELAYED_WORK(&efx->monitor_work, ef4_monitor);
2718 	INIT_DELAYED_WORK(&efx->selftest_work, ef4_selftest_async_work);
2719 	efx->pci_dev = pci_dev;
2720 	efx->msg_enable = debug;
2721 	efx->state = STATE_UNINIT;
2722 	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
2723 
2724 	efx->net_dev = net_dev;
2725 	efx->rx_prefix_size = efx->type->rx_prefix_size;
2726 	efx->rx_ip_align =
2727 		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
2728 	efx->rx_packet_hash_offset =
2729 		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
2730 	efx->rx_packet_ts_offset =
2731 		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
2732 	spin_lock_init(&efx->stats_lock);
2733 	mutex_init(&efx->mac_lock);
2734 	efx->phy_op = &ef4_dummy_phy_operations;
2735 	efx->mdio.dev = net_dev;
2736 	INIT_WORK(&efx->mac_work, ef4_mac_work);
2737 	init_waitqueue_head(&efx->flush_wq);
2738 
2739 	for (i = 0; i < EF4_MAX_CHANNELS; i++) {
2740 		efx->channel[i] = ef4_alloc_channel(efx, i, NULL);
2741 		if (!efx->channel[i])
2742 			goto fail;
2743 		efx->msi_context[i].efx = efx;
2744 		efx->msi_context[i].index = i;
2745 	}
2746 
2747 	/* Higher numbered interrupt modes are less capable! */
2748 	efx->interrupt_mode = max(efx->type->max_interrupt_mode,
2749 				  interrupt_mode);
2750 
2751 	/* Would be good to use the net_dev name, but we're too early */
2752 	snprintf(efx->workqueue_name, sizeof(efx->workqueue_name), "sfc%s",
2753 		 pci_name(pci_dev));
2754 	efx->workqueue = create_singlethread_workqueue(efx->workqueue_name);
2755 	if (!efx->workqueue)
2756 		goto fail;
2757 
2758 	return 0;
2759 
2760 fail:
2761 	ef4_fini_struct(efx);
2762 	return -ENOMEM;
2763 }
2764 
2765 static void ef4_fini_struct(struct ef4_nic *efx)
2766 {
2767 	int i;
2768 
2769 	for (i = 0; i < EF4_MAX_CHANNELS; i++)
2770 		kfree(efx->channel[i]);
2771 
2772 	kfree(efx->vpd_sn);
2773 
2774 	if (efx->workqueue) {
2775 		destroy_workqueue(efx->workqueue);
2776 		efx->workqueue = NULL;
2777 	}
2778 }
2779 
2780 void ef4_update_sw_stats(struct ef4_nic *efx, u64 *stats)
2781 {
2782 	u64 n_rx_nodesc_trunc = 0;
2783 	struct ef4_channel *channel;
2784 
2785 	ef4_for_each_channel(channel, efx)
2786 		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
2787 	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
2788 	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
2789 }
2790 
2791 /**************************************************************************
2792  *
2793  * PCI interface
2794  *
2795  **************************************************************************/
2796 
2797 /* Main body of final NIC shutdown code
2798  * This is called only at module unload (or hotplug removal).
2799  */
2800 static void ef4_pci_remove_main(struct ef4_nic *efx)
2801 {
2802 	/* Flush reset_work. It can no longer be scheduled since we
2803 	 * are not READY.
2804 	 */
2805 	BUG_ON(efx->state == STATE_READY);
2806 	cancel_work_sync(&efx->reset_work);
2807 
2808 	ef4_disable_interrupts(efx);
2809 	ef4_nic_fini_interrupt(efx);
2810 	ef4_fini_port(efx);
2811 	efx->type->fini(efx);
2812 	ef4_fini_napi(efx);
2813 	ef4_remove_all(efx);
2814 }
2815 
2816 /* Final NIC shutdown
2817  * This is called only at module unload (or hotplug removal).  A PF can call
2818  * this on its VFs to ensure they are unbound first.
2819  */
2820 static void ef4_pci_remove(struct pci_dev *pci_dev)
2821 {
2822 	struct ef4_nic *efx;
2823 
2824 	efx = pci_get_drvdata(pci_dev);
2825 	if (!efx)
2826 		return;
2827 
2828 	/* Mark the NIC as fini, then stop the interface */
2829 	rtnl_lock();
2830 	ef4_dissociate(efx);
2831 	dev_close(efx->net_dev);
2832 	ef4_disable_interrupts(efx);
2833 	efx->state = STATE_UNINIT;
2834 	rtnl_unlock();
2835 
2836 	ef4_unregister_netdev(efx);
2837 
2838 	ef4_mtd_remove(efx);
2839 
2840 	ef4_pci_remove_main(efx);
2841 
2842 	ef4_fini_io(efx);
2843 	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
2844 
2845 	ef4_fini_struct(efx);
2846 	free_netdev(efx->net_dev);
2847 
2848 	pci_disable_pcie_error_reporting(pci_dev);
2849 };
2850 
2851 /* NIC VPD information
2852  * Called during probe to display the part number of the
2853  * installed NIC.  VPD is potentially very large but this should
2854  * always appear within the first 512 bytes.
2855  */
2856 #define SFC_VPD_LEN 512
2857 static void ef4_probe_vpd_strings(struct ef4_nic *efx)
2858 {
2859 	struct pci_dev *dev = efx->pci_dev;
2860 	char vpd_data[SFC_VPD_LEN];
2861 	ssize_t vpd_size;
2862 	int ro_start, ro_size, i, j;
2863 
2864 	/* Get the vpd data from the device */
2865 	vpd_size = pci_read_vpd(dev, 0, sizeof(vpd_data), vpd_data);
2866 	if (vpd_size <= 0) {
2867 		netif_err(efx, drv, efx->net_dev, "Unable to read VPD\n");
2868 		return;
2869 	}
2870 
2871 	/* Get the Read only section */
2872 	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size, PCI_VPD_LRDT_RO_DATA);
2873 	if (ro_start < 0) {
2874 		netif_err(efx, drv, efx->net_dev, "VPD Read-only not found\n");
2875 		return;
2876 	}
2877 
2878 	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
2879 	j = ro_size;
2880 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
2881 	if (i + j > vpd_size)
2882 		j = vpd_size - i;
2883 
2884 	/* Get the Part number */
2885 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "PN");
2886 	if (i < 0) {
2887 		netif_err(efx, drv, efx->net_dev, "Part number not found\n");
2888 		return;
2889 	}
2890 
2891 	j = pci_vpd_info_field_size(&vpd_data[i]);
2892 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
2893 	if (i + j > vpd_size) {
2894 		netif_err(efx, drv, efx->net_dev, "Incomplete part number\n");
2895 		return;
2896 	}
2897 
2898 	netif_info(efx, drv, efx->net_dev,
2899 		   "Part Number : %.*s\n", j, &vpd_data[i]);
2900 
2901 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
2902 	j = ro_size;
2903 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "SN");
2904 	if (i < 0) {
2905 		netif_err(efx, drv, efx->net_dev, "Serial number not found\n");
2906 		return;
2907 	}
2908 
2909 	j = pci_vpd_info_field_size(&vpd_data[i]);
2910 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
2911 	if (i + j > vpd_size) {
2912 		netif_err(efx, drv, efx->net_dev, "Incomplete serial number\n");
2913 		return;
2914 	}
2915 
2916 	efx->vpd_sn = kmalloc(j + 1, GFP_KERNEL);
2917 	if (!efx->vpd_sn)
2918 		return;
2919 
2920 	snprintf(efx->vpd_sn, j + 1, "%s", &vpd_data[i]);
2921 }
2922 
2923 
2924 /* Main body of NIC initialisation
2925  * This is called at module load (or hotplug insertion, theoretically).
2926  */
2927 static int ef4_pci_probe_main(struct ef4_nic *efx)
2928 {
2929 	int rc;
2930 
2931 	/* Do start-of-day initialisation */
2932 	rc = ef4_probe_all(efx);
2933 	if (rc)
2934 		goto fail1;
2935 
2936 	ef4_init_napi(efx);
2937 
2938 	rc = efx->type->init(efx);
2939 	if (rc) {
2940 		netif_err(efx, probe, efx->net_dev,
2941 			  "failed to initialise NIC\n");
2942 		goto fail3;
2943 	}
2944 
2945 	rc = ef4_init_port(efx);
2946 	if (rc) {
2947 		netif_err(efx, probe, efx->net_dev,
2948 			  "failed to initialise port\n");
2949 		goto fail4;
2950 	}
2951 
2952 	rc = ef4_nic_init_interrupt(efx);
2953 	if (rc)
2954 		goto fail5;
2955 	rc = ef4_enable_interrupts(efx);
2956 	if (rc)
2957 		goto fail6;
2958 
2959 	return 0;
2960 
2961  fail6:
2962 	ef4_nic_fini_interrupt(efx);
2963  fail5:
2964 	ef4_fini_port(efx);
2965  fail4:
2966 	efx->type->fini(efx);
2967  fail3:
2968 	ef4_fini_napi(efx);
2969 	ef4_remove_all(efx);
2970  fail1:
2971 	return rc;
2972 }
2973 
2974 /* NIC initialisation
2975  *
2976  * This is called at module load (or hotplug insertion,
2977  * theoretically).  It sets up PCI mappings, resets the NIC,
2978  * sets up and registers the network devices with the kernel and hooks
2979  * the interrupt service routine.  It does not prepare the device for
2980  * transmission; this is left to the first time one of the network
2981  * interfaces is brought up (i.e. ef4_net_open).
2982  */
2983 static int ef4_pci_probe(struct pci_dev *pci_dev,
2984 			 const struct pci_device_id *entry)
2985 {
2986 	struct net_device *net_dev;
2987 	struct ef4_nic *efx;
2988 	int rc;
2989 
2990 	/* Allocate and initialise a struct net_device and struct ef4_nic */
2991 	net_dev = alloc_etherdev_mqs(sizeof(*efx), EF4_MAX_CORE_TX_QUEUES,
2992 				     EF4_MAX_RX_QUEUES);
2993 	if (!net_dev)
2994 		return -ENOMEM;
2995 	efx = netdev_priv(net_dev);
2996 	efx->type = (const struct ef4_nic_type *) entry->driver_data;
2997 	efx->fixed_features |= NETIF_F_HIGHDMA;
2998 
2999 	pci_set_drvdata(pci_dev, efx);
3000 	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
3001 	rc = ef4_init_struct(efx, pci_dev, net_dev);
3002 	if (rc)
3003 		goto fail1;
3004 
3005 	netif_info(efx, probe, efx->net_dev,
3006 		   "Solarflare NIC detected\n");
3007 
3008 	ef4_probe_vpd_strings(efx);
3009 
3010 	/* Set up basic I/O (BAR mappings etc) */
3011 	rc = ef4_init_io(efx);
3012 	if (rc)
3013 		goto fail2;
3014 
3015 	rc = ef4_pci_probe_main(efx);
3016 	if (rc)
3017 		goto fail3;
3018 
3019 	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
3020 			      NETIF_F_RXCSUM);
3021 	/* Mask for features that also apply to VLAN devices */
3022 	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
3023 				   NETIF_F_HIGHDMA | NETIF_F_RXCSUM);
3024 
3025 	net_dev->hw_features = net_dev->features & ~efx->fixed_features;
3026 
3027 	/* Disable VLAN filtering by default.  It may be enforced if
3028 	 * the feature is fixed (i.e. VLAN filters are required to
3029 	 * receive VLAN tagged packets due to vPort restrictions).
3030 	 */
3031 	net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3032 	net_dev->features |= efx->fixed_features;
3033 
3034 	rc = ef4_register_netdev(efx);
3035 	if (rc)
3036 		goto fail4;
3037 
3038 	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
3039 
3040 	/* Try to create MTDs, but allow this to fail */
3041 	rtnl_lock();
3042 	rc = ef4_mtd_probe(efx);
3043 	rtnl_unlock();
3044 	if (rc && rc != -EPERM)
3045 		netif_warn(efx, probe, efx->net_dev,
3046 			   "failed to create MTDs (%d)\n", rc);
3047 
3048 	rc = pci_enable_pcie_error_reporting(pci_dev);
3049 	if (rc && rc != -EINVAL)
3050 		netif_notice(efx, probe, efx->net_dev,
3051 			     "PCIE error reporting unavailable (%d).\n",
3052 			     rc);
3053 
3054 	return 0;
3055 
3056  fail4:
3057 	ef4_pci_remove_main(efx);
3058  fail3:
3059 	ef4_fini_io(efx);
3060  fail2:
3061 	ef4_fini_struct(efx);
3062  fail1:
3063 	WARN_ON(rc > 0);
3064 	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
3065 	free_netdev(net_dev);
3066 	return rc;
3067 }
3068 
3069 static int ef4_pm_freeze(struct device *dev)
3070 {
3071 	struct ef4_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3072 
3073 	rtnl_lock();
3074 
3075 	if (efx->state != STATE_DISABLED) {
3076 		efx->state = STATE_UNINIT;
3077 
3078 		ef4_device_detach_sync(efx);
3079 
3080 		ef4_stop_all(efx);
3081 		ef4_disable_interrupts(efx);
3082 	}
3083 
3084 	rtnl_unlock();
3085 
3086 	return 0;
3087 }
3088 
3089 static int ef4_pm_thaw(struct device *dev)
3090 {
3091 	int rc;
3092 	struct ef4_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3093 
3094 	rtnl_lock();
3095 
3096 	if (efx->state != STATE_DISABLED) {
3097 		rc = ef4_enable_interrupts(efx);
3098 		if (rc)
3099 			goto fail;
3100 
3101 		mutex_lock(&efx->mac_lock);
3102 		efx->phy_op->reconfigure(efx);
3103 		mutex_unlock(&efx->mac_lock);
3104 
3105 		ef4_start_all(efx);
3106 
3107 		netif_device_attach(efx->net_dev);
3108 
3109 		efx->state = STATE_READY;
3110 
3111 		efx->type->resume_wol(efx);
3112 	}
3113 
3114 	rtnl_unlock();
3115 
3116 	/* Reschedule any quenched resets scheduled during ef4_pm_freeze() */
3117 	queue_work(reset_workqueue, &efx->reset_work);
3118 
3119 	return 0;
3120 
3121 fail:
3122 	rtnl_unlock();
3123 
3124 	return rc;
3125 }
3126 
3127 static int ef4_pm_poweroff(struct device *dev)
3128 {
3129 	struct pci_dev *pci_dev = to_pci_dev(dev);
3130 	struct ef4_nic *efx = pci_get_drvdata(pci_dev);
3131 
3132 	efx->type->fini(efx);
3133 
3134 	efx->reset_pending = 0;
3135 
3136 	pci_save_state(pci_dev);
3137 	return pci_set_power_state(pci_dev, PCI_D3hot);
3138 }
3139 
3140 /* Used for both resume and restore */
3141 static int ef4_pm_resume(struct device *dev)
3142 {
3143 	struct pci_dev *pci_dev = to_pci_dev(dev);
3144 	struct ef4_nic *efx = pci_get_drvdata(pci_dev);
3145 	int rc;
3146 
3147 	rc = pci_set_power_state(pci_dev, PCI_D0);
3148 	if (rc)
3149 		return rc;
3150 	pci_restore_state(pci_dev);
3151 	rc = pci_enable_device(pci_dev);
3152 	if (rc)
3153 		return rc;
3154 	pci_set_master(efx->pci_dev);
3155 	rc = efx->type->reset(efx, RESET_TYPE_ALL);
3156 	if (rc)
3157 		return rc;
3158 	rc = efx->type->init(efx);
3159 	if (rc)
3160 		return rc;
3161 	rc = ef4_pm_thaw(dev);
3162 	return rc;
3163 }
3164 
3165 static int ef4_pm_suspend(struct device *dev)
3166 {
3167 	int rc;
3168 
3169 	ef4_pm_freeze(dev);
3170 	rc = ef4_pm_poweroff(dev);
3171 	if (rc)
3172 		ef4_pm_resume(dev);
3173 	return rc;
3174 }
3175 
3176 static const struct dev_pm_ops ef4_pm_ops = {
3177 	.suspend	= ef4_pm_suspend,
3178 	.resume		= ef4_pm_resume,
3179 	.freeze		= ef4_pm_freeze,
3180 	.thaw		= ef4_pm_thaw,
3181 	.poweroff	= ef4_pm_poweroff,
3182 	.restore	= ef4_pm_resume,
3183 };
3184 
3185 /* A PCI error affecting this device was detected.
3186  * At this point MMIO and DMA may be disabled.
3187  * Stop the software path and request a slot reset.
3188  */
3189 static pci_ers_result_t ef4_io_error_detected(struct pci_dev *pdev,
3190 					      enum pci_channel_state state)
3191 {
3192 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3193 	struct ef4_nic *efx = pci_get_drvdata(pdev);
3194 
3195 	if (state == pci_channel_io_perm_failure)
3196 		return PCI_ERS_RESULT_DISCONNECT;
3197 
3198 	rtnl_lock();
3199 
3200 	if (efx->state != STATE_DISABLED) {
3201 		efx->state = STATE_RECOVERY;
3202 		efx->reset_pending = 0;
3203 
3204 		ef4_device_detach_sync(efx);
3205 
3206 		ef4_stop_all(efx);
3207 		ef4_disable_interrupts(efx);
3208 
3209 		status = PCI_ERS_RESULT_NEED_RESET;
3210 	} else {
3211 		/* If the interface is disabled we don't want to do anything
3212 		 * with it.
3213 		 */
3214 		status = PCI_ERS_RESULT_RECOVERED;
3215 	}
3216 
3217 	rtnl_unlock();
3218 
3219 	pci_disable_device(pdev);
3220 
3221 	return status;
3222 }
3223 
3224 /* Fake a successful reset, which will be performed later in ef4_io_resume. */
3225 static pci_ers_result_t ef4_io_slot_reset(struct pci_dev *pdev)
3226 {
3227 	struct ef4_nic *efx = pci_get_drvdata(pdev);
3228 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3229 	int rc;
3230 
3231 	if (pci_enable_device(pdev)) {
3232 		netif_err(efx, hw, efx->net_dev,
3233 			  "Cannot re-enable PCI device after reset.\n");
3234 		status =  PCI_ERS_RESULT_DISCONNECT;
3235 	}
3236 
3237 	rc = pci_cleanup_aer_uncorrect_error_status(pdev);
3238 	if (rc) {
3239 		netif_err(efx, hw, efx->net_dev,
3240 		"pci_cleanup_aer_uncorrect_error_status failed (%d)\n", rc);
3241 		/* Non-fatal error. Continue. */
3242 	}
3243 
3244 	return status;
3245 }
3246 
3247 /* Perform the actual reset and resume I/O operations. */
3248 static void ef4_io_resume(struct pci_dev *pdev)
3249 {
3250 	struct ef4_nic *efx = pci_get_drvdata(pdev);
3251 	int rc;
3252 
3253 	rtnl_lock();
3254 
3255 	if (efx->state == STATE_DISABLED)
3256 		goto out;
3257 
3258 	rc = ef4_reset(efx, RESET_TYPE_ALL);
3259 	if (rc) {
3260 		netif_err(efx, hw, efx->net_dev,
3261 			  "ef4_reset failed after PCI error (%d)\n", rc);
3262 	} else {
3263 		efx->state = STATE_READY;
3264 		netif_dbg(efx, hw, efx->net_dev,
3265 			  "Done resetting and resuming IO after PCI error.\n");
3266 	}
3267 
3268 out:
3269 	rtnl_unlock();
3270 }
3271 
3272 /* For simplicity and reliability, we always require a slot reset and try to
3273  * reset the hardware when a pci error affecting the device is detected.
3274  * We leave both the link_reset and mmio_enabled callback unimplemented:
3275  * with our request for slot reset the mmio_enabled callback will never be
3276  * called, and the link_reset callback is not used by AER or EEH mechanisms.
3277  */
3278 static const struct pci_error_handlers ef4_err_handlers = {
3279 	.error_detected = ef4_io_error_detected,
3280 	.slot_reset	= ef4_io_slot_reset,
3281 	.resume		= ef4_io_resume,
3282 };
3283 
3284 static struct pci_driver ef4_pci_driver = {
3285 	.name		= KBUILD_MODNAME,
3286 	.id_table	= ef4_pci_table,
3287 	.probe		= ef4_pci_probe,
3288 	.remove		= ef4_pci_remove,
3289 	.driver.pm	= &ef4_pm_ops,
3290 	.err_handler	= &ef4_err_handlers,
3291 };
3292 
3293 /**************************************************************************
3294  *
3295  * Kernel module interface
3296  *
3297  *************************************************************************/
3298 
3299 module_param(interrupt_mode, uint, 0444);
3300 MODULE_PARM_DESC(interrupt_mode,
3301 		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
3302 
3303 static int __init ef4_init_module(void)
3304 {
3305 	int rc;
3306 
3307 	printk(KERN_INFO "Solarflare Falcon driver v" EF4_DRIVER_VERSION "\n");
3308 
3309 	rc = register_netdevice_notifier(&ef4_netdev_notifier);
3310 	if (rc)
3311 		goto err_notifier;
3312 
3313 	reset_workqueue = create_singlethread_workqueue("sfc_reset");
3314 	if (!reset_workqueue) {
3315 		rc = -ENOMEM;
3316 		goto err_reset;
3317 	}
3318 
3319 	rc = pci_register_driver(&ef4_pci_driver);
3320 	if (rc < 0)
3321 		goto err_pci;
3322 
3323 	return 0;
3324 
3325  err_pci:
3326 	destroy_workqueue(reset_workqueue);
3327  err_reset:
3328 	unregister_netdevice_notifier(&ef4_netdev_notifier);
3329  err_notifier:
3330 	return rc;
3331 }
3332 
3333 static void __exit ef4_exit_module(void)
3334 {
3335 	printk(KERN_INFO "Solarflare Falcon driver unloading\n");
3336 
3337 	pci_unregister_driver(&ef4_pci_driver);
3338 	destroy_workqueue(reset_workqueue);
3339 	unregister_netdevice_notifier(&ef4_netdev_notifier);
3340 
3341 }
3342 
3343 module_init(ef4_init_module);
3344 module_exit(ef4_exit_module);
3345 
3346 MODULE_AUTHOR("Solarflare Communications and "
3347 	      "Michael Brown <mbrown@fensystems.co.uk>");
3348 MODULE_DESCRIPTION("Solarflare Falcon network driver");
3349 MODULE_LICENSE("GPL");
3350 MODULE_DEVICE_TABLE(pci, ef4_pci_table);
3351