xref: /openbmc/linux/drivers/net/ethernet/sfc/efx.c (revision 5f32c314)
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 "mcdi.h"
31 #include "workarounds.h"
32 
33 /**************************************************************************
34  *
35  * Type name strings
36  *
37  **************************************************************************
38  */
39 
40 /* Loopback mode names (see LOOPBACK_MODE()) */
41 const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
42 const char *const efx_loopback_mode_names[] = {
43 	[LOOPBACK_NONE]		= "NONE",
44 	[LOOPBACK_DATA]		= "DATAPATH",
45 	[LOOPBACK_GMAC]		= "GMAC",
46 	[LOOPBACK_XGMII]	= "XGMII",
47 	[LOOPBACK_XGXS]		= "XGXS",
48 	[LOOPBACK_XAUI]		= "XAUI",
49 	[LOOPBACK_GMII]		= "GMII",
50 	[LOOPBACK_SGMII]	= "SGMII",
51 	[LOOPBACK_XGBR]		= "XGBR",
52 	[LOOPBACK_XFI]		= "XFI",
53 	[LOOPBACK_XAUI_FAR]	= "XAUI_FAR",
54 	[LOOPBACK_GMII_FAR]	= "GMII_FAR",
55 	[LOOPBACK_SGMII_FAR]	= "SGMII_FAR",
56 	[LOOPBACK_XFI_FAR]	= "XFI_FAR",
57 	[LOOPBACK_GPHY]		= "GPHY",
58 	[LOOPBACK_PHYXS]	= "PHYXS",
59 	[LOOPBACK_PCS]		= "PCS",
60 	[LOOPBACK_PMAPMD]	= "PMA/PMD",
61 	[LOOPBACK_XPORT]	= "XPORT",
62 	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
63 	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
64 	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
65 	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
66 	[LOOPBACK_GMII_WS]	= "GMII_WS",
67 	[LOOPBACK_XFI_WS]	= "XFI_WS",
68 	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
69 	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
70 };
71 
72 const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
73 const char *const efx_reset_type_names[] = {
74 	[RESET_TYPE_INVISIBLE]          = "INVISIBLE",
75 	[RESET_TYPE_ALL]                = "ALL",
76 	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
77 	[RESET_TYPE_WORLD]              = "WORLD",
78 	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
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 	[RESET_TYPE_MC_FAILURE]         = "MC_FAILURE",
86 	[RESET_TYPE_MC_BIST]		= "MC_BIST",
87 };
88 
89 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
90  * queued onto this work queue. This is not a per-nic work queue, because
91  * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
92  */
93 static struct workqueue_struct *reset_workqueue;
94 
95 /* How often and how many times to poll for a reset while waiting for a
96  * BIST that another function started to complete.
97  */
98 #define BIST_WAIT_DELAY_MS	100
99 #define BIST_WAIT_DELAY_COUNT	100
100 
101 /**************************************************************************
102  *
103  * Configurable values
104  *
105  *************************************************************************/
106 
107 /*
108  * Use separate channels for TX and RX events
109  *
110  * Set this to 1 to use separate channels for TX and RX. It allows us
111  * to control interrupt affinity separately for TX and RX.
112  *
113  * This is only used in MSI-X interrupt mode
114  */
115 static bool separate_tx_channels;
116 module_param(separate_tx_channels, bool, 0444);
117 MODULE_PARM_DESC(separate_tx_channels,
118 		 "Use separate channels for TX and RX");
119 
120 /* This is the weight assigned to each of the (per-channel) virtual
121  * NAPI devices.
122  */
123 static int napi_weight = 64;
124 
125 /* This is the time (in jiffies) between invocations of the hardware
126  * monitor.
127  * On Falcon-based NICs, this will:
128  * - Check the on-board hardware monitor;
129  * - Poll the link state and reconfigure the hardware as necessary.
130  * On Siena-based NICs for power systems with EEH support, this will give EEH a
131  * chance to start.
132  */
133 static unsigned int efx_monitor_interval = 1 * HZ;
134 
135 /* Initial interrupt moderation settings.  They can be modified after
136  * module load with ethtool.
137  *
138  * The default for RX should strike a balance between increasing the
139  * round-trip latency and reducing overhead.
140  */
141 static unsigned int rx_irq_mod_usec = 60;
142 
143 /* Initial interrupt moderation settings.  They can be modified after
144  * module load with ethtool.
145  *
146  * This default is chosen to ensure that a 10G link does not go idle
147  * while a TX queue is stopped after it has become full.  A queue is
148  * restarted when it drops below half full.  The time this takes (assuming
149  * worst case 3 descriptors per packet and 1024 descriptors) is
150  *   512 / 3 * 1.2 = 205 usec.
151  */
152 static unsigned int tx_irq_mod_usec = 150;
153 
154 /* This is the first interrupt mode to try out of:
155  * 0 => MSI-X
156  * 1 => MSI
157  * 2 => legacy
158  */
159 static unsigned int interrupt_mode;
160 
161 /* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
162  * i.e. the number of CPUs among which we may distribute simultaneous
163  * interrupt handling.
164  *
165  * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
166  * The default (0) means to assign an interrupt to each core.
167  */
168 static unsigned int rss_cpus;
169 module_param(rss_cpus, uint, 0444);
170 MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");
171 
172 static bool phy_flash_cfg;
173 module_param(phy_flash_cfg, bool, 0644);
174 MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");
175 
176 static unsigned irq_adapt_low_thresh = 8000;
177 module_param(irq_adapt_low_thresh, uint, 0644);
178 MODULE_PARM_DESC(irq_adapt_low_thresh,
179 		 "Threshold score for reducing IRQ moderation");
180 
181 static unsigned irq_adapt_high_thresh = 16000;
182 module_param(irq_adapt_high_thresh, uint, 0644);
183 MODULE_PARM_DESC(irq_adapt_high_thresh,
184 		 "Threshold score for increasing IRQ moderation");
185 
186 static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
187 			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
188 			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
189 			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
190 module_param(debug, uint, 0);
191 MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
192 
193 /**************************************************************************
194  *
195  * Utility functions and prototypes
196  *
197  *************************************************************************/
198 
199 static int efx_soft_enable_interrupts(struct efx_nic *efx);
200 static void efx_soft_disable_interrupts(struct efx_nic *efx);
201 static void efx_remove_channel(struct efx_channel *channel);
202 static void efx_remove_channels(struct efx_nic *efx);
203 static const struct efx_channel_type efx_default_channel_type;
204 static void efx_remove_port(struct efx_nic *efx);
205 static void efx_init_napi_channel(struct efx_channel *channel);
206 static void efx_fini_napi(struct efx_nic *efx);
207 static void efx_fini_napi_channel(struct efx_channel *channel);
208 static void efx_fini_struct(struct efx_nic *efx);
209 static void efx_start_all(struct efx_nic *efx);
210 static void efx_stop_all(struct efx_nic *efx);
211 
212 #define EFX_ASSERT_RESET_SERIALISED(efx)		\
213 	do {						\
214 		if ((efx->state == STATE_READY) ||	\
215 		    (efx->state == STATE_RECOVERY) ||	\
216 		    (efx->state == STATE_DISABLED))	\
217 			ASSERT_RTNL();			\
218 	} while (0)
219 
220 static int efx_check_disabled(struct efx_nic *efx)
221 {
222 	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
223 		netif_err(efx, drv, efx->net_dev,
224 			  "device is disabled due to earlier errors\n");
225 		return -EIO;
226 	}
227 	return 0;
228 }
229 
230 /**************************************************************************
231  *
232  * Event queue processing
233  *
234  *************************************************************************/
235 
236 /* Process channel's event queue
237  *
238  * This function is responsible for processing the event queue of a
239  * single channel.  The caller must guarantee that this function will
240  * never be concurrently called more than once on the same channel,
241  * though different channels may be being processed concurrently.
242  */
243 static int efx_process_channel(struct efx_channel *channel, int budget)
244 {
245 	int spent;
246 
247 	if (unlikely(!channel->enabled))
248 		return 0;
249 
250 	spent = efx_nic_process_eventq(channel, budget);
251 	if (spent && efx_channel_has_rx_queue(channel)) {
252 		struct efx_rx_queue *rx_queue =
253 			efx_channel_get_rx_queue(channel);
254 
255 		efx_rx_flush_packet(channel);
256 		efx_fast_push_rx_descriptors(rx_queue, true);
257 	}
258 
259 	return spent;
260 }
261 
262 /* NAPI poll handler
263  *
264  * NAPI guarantees serialisation of polls of the same device, which
265  * provides the guarantee required by efx_process_channel().
266  */
267 static int efx_poll(struct napi_struct *napi, int budget)
268 {
269 	struct efx_channel *channel =
270 		container_of(napi, struct efx_channel, napi_str);
271 	struct efx_nic *efx = channel->efx;
272 	int spent;
273 
274 	netif_vdbg(efx, intr, efx->net_dev,
275 		   "channel %d NAPI poll executing on CPU %d\n",
276 		   channel->channel, raw_smp_processor_id());
277 
278 	spent = efx_process_channel(channel, budget);
279 
280 	if (spent < budget) {
281 		if (efx_channel_has_rx_queue(channel) &&
282 		    efx->irq_rx_adaptive &&
283 		    unlikely(++channel->irq_count == 1000)) {
284 			if (unlikely(channel->irq_mod_score <
285 				     irq_adapt_low_thresh)) {
286 				if (channel->irq_moderation > 1) {
287 					channel->irq_moderation -= 1;
288 					efx->type->push_irq_moderation(channel);
289 				}
290 			} else if (unlikely(channel->irq_mod_score >
291 					    irq_adapt_high_thresh)) {
292 				if (channel->irq_moderation <
293 				    efx->irq_rx_moderation) {
294 					channel->irq_moderation += 1;
295 					efx->type->push_irq_moderation(channel);
296 				}
297 			}
298 			channel->irq_count = 0;
299 			channel->irq_mod_score = 0;
300 		}
301 
302 		efx_filter_rfs_expire(channel);
303 
304 		/* There is no race here; although napi_disable() will
305 		 * only wait for napi_complete(), this isn't a problem
306 		 * since efx_nic_eventq_read_ack() will have no effect if
307 		 * interrupts have already been disabled.
308 		 */
309 		napi_complete(napi);
310 		efx_nic_eventq_read_ack(channel);
311 	}
312 
313 	return spent;
314 }
315 
316 /* Create event queue
317  * Event queue memory allocations are done only once.  If the channel
318  * is reset, the memory buffer will be reused; this guards against
319  * errors during channel reset and also simplifies interrupt handling.
320  */
321 static int efx_probe_eventq(struct efx_channel *channel)
322 {
323 	struct efx_nic *efx = channel->efx;
324 	unsigned long entries;
325 
326 	netif_dbg(efx, probe, efx->net_dev,
327 		  "chan %d create event queue\n", channel->channel);
328 
329 	/* Build an event queue with room for one event per tx and rx buffer,
330 	 * plus some extra for link state events and MCDI completions. */
331 	entries = roundup_pow_of_two(efx->rxq_entries + efx->txq_entries + 128);
332 	EFX_BUG_ON_PARANOID(entries > EFX_MAX_EVQ_SIZE);
333 	channel->eventq_mask = max(entries, EFX_MIN_EVQ_SIZE) - 1;
334 
335 	return efx_nic_probe_eventq(channel);
336 }
337 
338 /* Prepare channel's event queue */
339 static int efx_init_eventq(struct efx_channel *channel)
340 {
341 	struct efx_nic *efx = channel->efx;
342 	int rc;
343 
344 	EFX_WARN_ON_PARANOID(channel->eventq_init);
345 
346 	netif_dbg(efx, drv, efx->net_dev,
347 		  "chan %d init event queue\n", channel->channel);
348 
349 	rc = efx_nic_init_eventq(channel);
350 	if (rc == 0) {
351 		efx->type->push_irq_moderation(channel);
352 		channel->eventq_read_ptr = 0;
353 		channel->eventq_init = true;
354 	}
355 	return rc;
356 }
357 
358 /* Enable event queue processing and NAPI */
359 static void efx_start_eventq(struct efx_channel *channel)
360 {
361 	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
362 		  "chan %d start event queue\n", channel->channel);
363 
364 	/* Make sure the NAPI handler sees the enabled flag set */
365 	channel->enabled = true;
366 	smp_wmb();
367 
368 	napi_enable(&channel->napi_str);
369 	efx_nic_eventq_read_ack(channel);
370 }
371 
372 /* Disable event queue processing and NAPI */
373 static void efx_stop_eventq(struct efx_channel *channel)
374 {
375 	if (!channel->enabled)
376 		return;
377 
378 	napi_disable(&channel->napi_str);
379 	channel->enabled = false;
380 }
381 
382 static void efx_fini_eventq(struct efx_channel *channel)
383 {
384 	if (!channel->eventq_init)
385 		return;
386 
387 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
388 		  "chan %d fini event queue\n", channel->channel);
389 
390 	efx_nic_fini_eventq(channel);
391 	channel->eventq_init = false;
392 }
393 
394 static void efx_remove_eventq(struct efx_channel *channel)
395 {
396 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
397 		  "chan %d remove event queue\n", channel->channel);
398 
399 	efx_nic_remove_eventq(channel);
400 }
401 
402 /**************************************************************************
403  *
404  * Channel handling
405  *
406  *************************************************************************/
407 
408 /* Allocate and initialise a channel structure. */
409 static struct efx_channel *
410 efx_alloc_channel(struct efx_nic *efx, int i, struct efx_channel *old_channel)
411 {
412 	struct efx_channel *channel;
413 	struct efx_rx_queue *rx_queue;
414 	struct efx_tx_queue *tx_queue;
415 	int j;
416 
417 	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
418 	if (!channel)
419 		return NULL;
420 
421 	channel->efx = efx;
422 	channel->channel = i;
423 	channel->type = &efx_default_channel_type;
424 
425 	for (j = 0; j < EFX_TXQ_TYPES; j++) {
426 		tx_queue = &channel->tx_queue[j];
427 		tx_queue->efx = efx;
428 		tx_queue->queue = i * EFX_TXQ_TYPES + j;
429 		tx_queue->channel = channel;
430 	}
431 
432 	rx_queue = &channel->rx_queue;
433 	rx_queue->efx = efx;
434 	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
435 		    (unsigned long)rx_queue);
436 
437 	return channel;
438 }
439 
440 /* Allocate and initialise a channel structure, copying parameters
441  * (but not resources) from an old channel structure.
442  */
443 static struct efx_channel *
444 efx_copy_channel(const struct efx_channel *old_channel)
445 {
446 	struct efx_channel *channel;
447 	struct efx_rx_queue *rx_queue;
448 	struct efx_tx_queue *tx_queue;
449 	int j;
450 
451 	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
452 	if (!channel)
453 		return NULL;
454 
455 	*channel = *old_channel;
456 
457 	channel->napi_dev = NULL;
458 	memset(&channel->eventq, 0, sizeof(channel->eventq));
459 
460 	for (j = 0; j < EFX_TXQ_TYPES; j++) {
461 		tx_queue = &channel->tx_queue[j];
462 		if (tx_queue->channel)
463 			tx_queue->channel = channel;
464 		tx_queue->buffer = NULL;
465 		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
466 	}
467 
468 	rx_queue = &channel->rx_queue;
469 	rx_queue->buffer = NULL;
470 	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
471 	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
472 		    (unsigned long)rx_queue);
473 
474 	return channel;
475 }
476 
477 static int efx_probe_channel(struct efx_channel *channel)
478 {
479 	struct efx_tx_queue *tx_queue;
480 	struct efx_rx_queue *rx_queue;
481 	int rc;
482 
483 	netif_dbg(channel->efx, probe, channel->efx->net_dev,
484 		  "creating channel %d\n", channel->channel);
485 
486 	rc = channel->type->pre_probe(channel);
487 	if (rc)
488 		goto fail;
489 
490 	rc = efx_probe_eventq(channel);
491 	if (rc)
492 		goto fail;
493 
494 	efx_for_each_channel_tx_queue(tx_queue, channel) {
495 		rc = efx_probe_tx_queue(tx_queue);
496 		if (rc)
497 			goto fail;
498 	}
499 
500 	efx_for_each_channel_rx_queue(rx_queue, channel) {
501 		rc = efx_probe_rx_queue(rx_queue);
502 		if (rc)
503 			goto fail;
504 	}
505 
506 	channel->n_rx_frm_trunc = 0;
507 
508 	return 0;
509 
510 fail:
511 	efx_remove_channel(channel);
512 	return rc;
513 }
514 
515 static void
516 efx_get_channel_name(struct efx_channel *channel, char *buf, size_t len)
517 {
518 	struct efx_nic *efx = channel->efx;
519 	const char *type;
520 	int number;
521 
522 	number = channel->channel;
523 	if (efx->tx_channel_offset == 0) {
524 		type = "";
525 	} else if (channel->channel < efx->tx_channel_offset) {
526 		type = "-rx";
527 	} else {
528 		type = "-tx";
529 		number -= efx->tx_channel_offset;
530 	}
531 	snprintf(buf, len, "%s%s-%d", efx->name, type, number);
532 }
533 
534 static void efx_set_channel_names(struct efx_nic *efx)
535 {
536 	struct efx_channel *channel;
537 
538 	efx_for_each_channel(channel, efx)
539 		channel->type->get_name(channel,
540 					efx->msi_context[channel->channel].name,
541 					sizeof(efx->msi_context[0].name));
542 }
543 
544 static int efx_probe_channels(struct efx_nic *efx)
545 {
546 	struct efx_channel *channel;
547 	int rc;
548 
549 	/* Restart special buffer allocation */
550 	efx->next_buffer_table = 0;
551 
552 	/* Probe channels in reverse, so that any 'extra' channels
553 	 * use the start of the buffer table. This allows the traffic
554 	 * channels to be resized without moving them or wasting the
555 	 * entries before them.
556 	 */
557 	efx_for_each_channel_rev(channel, efx) {
558 		rc = efx_probe_channel(channel);
559 		if (rc) {
560 			netif_err(efx, probe, efx->net_dev,
561 				  "failed to create channel %d\n",
562 				  channel->channel);
563 			goto fail;
564 		}
565 	}
566 	efx_set_channel_names(efx);
567 
568 	return 0;
569 
570 fail:
571 	efx_remove_channels(efx);
572 	return rc;
573 }
574 
575 /* Channels are shutdown and reinitialised whilst the NIC is running
576  * to propagate configuration changes (mtu, checksum offload), or
577  * to clear hardware error conditions
578  */
579 static void efx_start_datapath(struct efx_nic *efx)
580 {
581 	bool old_rx_scatter = efx->rx_scatter;
582 	struct efx_tx_queue *tx_queue;
583 	struct efx_rx_queue *rx_queue;
584 	struct efx_channel *channel;
585 	size_t rx_buf_len;
586 
587 	/* Calculate the rx buffer allocation parameters required to
588 	 * support the current MTU, including padding for header
589 	 * alignment and overruns.
590 	 */
591 	efx->rx_dma_len = (efx->rx_prefix_size +
592 			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
593 			   efx->type->rx_buffer_padding);
594 	rx_buf_len = (sizeof(struct efx_rx_page_state) +
595 		      efx->rx_ip_align + efx->rx_dma_len);
596 	if (rx_buf_len <= PAGE_SIZE) {
597 		efx->rx_scatter = efx->type->always_rx_scatter;
598 		efx->rx_buffer_order = 0;
599 	} else if (efx->type->can_rx_scatter) {
600 		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
601 		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
602 			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
603 				       EFX_RX_BUF_ALIGNMENT) >
604 			     PAGE_SIZE);
605 		efx->rx_scatter = true;
606 		efx->rx_dma_len = EFX_RX_USR_BUF_SIZE;
607 		efx->rx_buffer_order = 0;
608 	} else {
609 		efx->rx_scatter = false;
610 		efx->rx_buffer_order = get_order(rx_buf_len);
611 	}
612 
613 	efx_rx_config_page_split(efx);
614 	if (efx->rx_buffer_order)
615 		netif_dbg(efx, drv, efx->net_dev,
616 			  "RX buf len=%u; page order=%u batch=%u\n",
617 			  efx->rx_dma_len, efx->rx_buffer_order,
618 			  efx->rx_pages_per_batch);
619 	else
620 		netif_dbg(efx, drv, efx->net_dev,
621 			  "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
622 			  efx->rx_dma_len, efx->rx_page_buf_step,
623 			  efx->rx_bufs_per_page, efx->rx_pages_per_batch);
624 
625 	/* RX filters may also have scatter-enabled flags */
626 	if (efx->rx_scatter != old_rx_scatter)
627 		efx->type->filter_update_rx_scatter(efx);
628 
629 	/* We must keep at least one descriptor in a TX ring empty.
630 	 * We could avoid this when the queue size does not exactly
631 	 * match the hardware ring size, but it's not that important.
632 	 * Therefore we stop the queue when one more skb might fill
633 	 * the ring completely.  We wake it when half way back to
634 	 * empty.
635 	 */
636 	efx->txq_stop_thresh = efx->txq_entries - efx_tx_max_skb_descs(efx);
637 	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;
638 
639 	/* Initialise the channels */
640 	efx_for_each_channel(channel, efx) {
641 		efx_for_each_channel_tx_queue(tx_queue, channel) {
642 			efx_init_tx_queue(tx_queue);
643 			atomic_inc(&efx->active_queues);
644 		}
645 
646 		efx_for_each_channel_rx_queue(rx_queue, channel) {
647 			efx_init_rx_queue(rx_queue);
648 			atomic_inc(&efx->active_queues);
649 			efx_stop_eventq(channel);
650 			efx_fast_push_rx_descriptors(rx_queue, false);
651 			efx_start_eventq(channel);
652 		}
653 
654 		WARN_ON(channel->rx_pkt_n_frags);
655 	}
656 
657 	efx_ptp_start_datapath(efx);
658 
659 	if (netif_device_present(efx->net_dev))
660 		netif_tx_wake_all_queues(efx->net_dev);
661 }
662 
663 static void efx_stop_datapath(struct efx_nic *efx)
664 {
665 	struct efx_channel *channel;
666 	struct efx_tx_queue *tx_queue;
667 	struct efx_rx_queue *rx_queue;
668 	int rc;
669 
670 	EFX_ASSERT_RESET_SERIALISED(efx);
671 	BUG_ON(efx->port_enabled);
672 
673 	efx_ptp_stop_datapath(efx);
674 
675 	/* Stop RX refill */
676 	efx_for_each_channel(channel, efx) {
677 		efx_for_each_channel_rx_queue(rx_queue, channel)
678 			rx_queue->refill_enabled = false;
679 	}
680 
681 	efx_for_each_channel(channel, efx) {
682 		/* RX packet processing is pipelined, so wait for the
683 		 * NAPI handler to complete.  At least event queue 0
684 		 * might be kept active by non-data events, so don't
685 		 * use napi_synchronize() but actually disable NAPI
686 		 * temporarily.
687 		 */
688 		if (efx_channel_has_rx_queue(channel)) {
689 			efx_stop_eventq(channel);
690 			efx_start_eventq(channel);
691 		}
692 	}
693 
694 	rc = efx->type->fini_dmaq(efx);
695 	if (rc && EFX_WORKAROUND_7803(efx)) {
696 		/* Schedule a reset to recover from the flush failure. The
697 		 * descriptor caches reference memory we're about to free,
698 		 * but falcon_reconfigure_mac_wrapper() won't reconnect
699 		 * the MACs because of the pending reset.
700 		 */
701 		netif_err(efx, drv, efx->net_dev,
702 			  "Resetting to recover from flush failure\n");
703 		efx_schedule_reset(efx, RESET_TYPE_ALL);
704 	} else if (rc) {
705 		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
706 	} else {
707 		netif_dbg(efx, drv, efx->net_dev,
708 			  "successfully flushed all queues\n");
709 	}
710 
711 	efx_for_each_channel(channel, efx) {
712 		efx_for_each_channel_rx_queue(rx_queue, channel)
713 			efx_fini_rx_queue(rx_queue);
714 		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
715 			efx_fini_tx_queue(tx_queue);
716 	}
717 }
718 
719 static void efx_remove_channel(struct efx_channel *channel)
720 {
721 	struct efx_tx_queue *tx_queue;
722 	struct efx_rx_queue *rx_queue;
723 
724 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
725 		  "destroy chan %d\n", channel->channel);
726 
727 	efx_for_each_channel_rx_queue(rx_queue, channel)
728 		efx_remove_rx_queue(rx_queue);
729 	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
730 		efx_remove_tx_queue(tx_queue);
731 	efx_remove_eventq(channel);
732 	channel->type->post_remove(channel);
733 }
734 
735 static void efx_remove_channels(struct efx_nic *efx)
736 {
737 	struct efx_channel *channel;
738 
739 	efx_for_each_channel(channel, efx)
740 		efx_remove_channel(channel);
741 }
742 
743 int
744 efx_realloc_channels(struct efx_nic *efx, u32 rxq_entries, u32 txq_entries)
745 {
746 	struct efx_channel *other_channel[EFX_MAX_CHANNELS], *channel;
747 	u32 old_rxq_entries, old_txq_entries;
748 	unsigned i, next_buffer_table = 0;
749 	int rc, rc2;
750 
751 	rc = efx_check_disabled(efx);
752 	if (rc)
753 		return rc;
754 
755 	/* Not all channels should be reallocated. We must avoid
756 	 * reallocating their buffer table entries.
757 	 */
758 	efx_for_each_channel(channel, efx) {
759 		struct efx_rx_queue *rx_queue;
760 		struct efx_tx_queue *tx_queue;
761 
762 		if (channel->type->copy)
763 			continue;
764 		next_buffer_table = max(next_buffer_table,
765 					channel->eventq.index +
766 					channel->eventq.entries);
767 		efx_for_each_channel_rx_queue(rx_queue, channel)
768 			next_buffer_table = max(next_buffer_table,
769 						rx_queue->rxd.index +
770 						rx_queue->rxd.entries);
771 		efx_for_each_channel_tx_queue(tx_queue, channel)
772 			next_buffer_table = max(next_buffer_table,
773 						tx_queue->txd.index +
774 						tx_queue->txd.entries);
775 	}
776 
777 	efx_device_detach_sync(efx);
778 	efx_stop_all(efx);
779 	efx_soft_disable_interrupts(efx);
780 
781 	/* Clone channels (where possible) */
782 	memset(other_channel, 0, sizeof(other_channel));
783 	for (i = 0; i < efx->n_channels; i++) {
784 		channel = efx->channel[i];
785 		if (channel->type->copy)
786 			channel = channel->type->copy(channel);
787 		if (!channel) {
788 			rc = -ENOMEM;
789 			goto out;
790 		}
791 		other_channel[i] = channel;
792 	}
793 
794 	/* Swap entry counts and channel pointers */
795 	old_rxq_entries = efx->rxq_entries;
796 	old_txq_entries = efx->txq_entries;
797 	efx->rxq_entries = rxq_entries;
798 	efx->txq_entries = txq_entries;
799 	for (i = 0; i < efx->n_channels; i++) {
800 		channel = efx->channel[i];
801 		efx->channel[i] = other_channel[i];
802 		other_channel[i] = channel;
803 	}
804 
805 	/* Restart buffer table allocation */
806 	efx->next_buffer_table = next_buffer_table;
807 
808 	for (i = 0; i < efx->n_channels; i++) {
809 		channel = efx->channel[i];
810 		if (!channel->type->copy)
811 			continue;
812 		rc = efx_probe_channel(channel);
813 		if (rc)
814 			goto rollback;
815 		efx_init_napi_channel(efx->channel[i]);
816 	}
817 
818 out:
819 	/* Destroy unused channel structures */
820 	for (i = 0; i < efx->n_channels; i++) {
821 		channel = other_channel[i];
822 		if (channel && channel->type->copy) {
823 			efx_fini_napi_channel(channel);
824 			efx_remove_channel(channel);
825 			kfree(channel);
826 		}
827 	}
828 
829 	rc2 = efx_soft_enable_interrupts(efx);
830 	if (rc2) {
831 		rc = rc ? rc : rc2;
832 		netif_err(efx, drv, efx->net_dev,
833 			  "unable to restart interrupts on channel reallocation\n");
834 		efx_schedule_reset(efx, RESET_TYPE_DISABLE);
835 	} else {
836 		efx_start_all(efx);
837 		netif_device_attach(efx->net_dev);
838 	}
839 	return rc;
840 
841 rollback:
842 	/* Swap back */
843 	efx->rxq_entries = old_rxq_entries;
844 	efx->txq_entries = old_txq_entries;
845 	for (i = 0; i < efx->n_channels; i++) {
846 		channel = efx->channel[i];
847 		efx->channel[i] = other_channel[i];
848 		other_channel[i] = channel;
849 	}
850 	goto out;
851 }
852 
853 void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
854 {
855 	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
856 }
857 
858 static const struct efx_channel_type efx_default_channel_type = {
859 	.pre_probe		= efx_channel_dummy_op_int,
860 	.post_remove		= efx_channel_dummy_op_void,
861 	.get_name		= efx_get_channel_name,
862 	.copy			= efx_copy_channel,
863 	.keep_eventq		= false,
864 };
865 
866 int efx_channel_dummy_op_int(struct efx_channel *channel)
867 {
868 	return 0;
869 }
870 
871 void efx_channel_dummy_op_void(struct efx_channel *channel)
872 {
873 }
874 
875 /**************************************************************************
876  *
877  * Port handling
878  *
879  **************************************************************************/
880 
881 /* This ensures that the kernel is kept informed (via
882  * netif_carrier_on/off) of the link status, and also maintains the
883  * link status's stop on the port's TX queue.
884  */
885 void efx_link_status_changed(struct efx_nic *efx)
886 {
887 	struct efx_link_state *link_state = &efx->link_state;
888 
889 	/* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
890 	 * that no events are triggered between unregister_netdev() and the
891 	 * driver unloading. A more general condition is that NETDEV_CHANGE
892 	 * can only be generated between NETDEV_UP and NETDEV_DOWN */
893 	if (!netif_running(efx->net_dev))
894 		return;
895 
896 	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
897 		efx->n_link_state_changes++;
898 
899 		if (link_state->up)
900 			netif_carrier_on(efx->net_dev);
901 		else
902 			netif_carrier_off(efx->net_dev);
903 	}
904 
905 	/* Status message for kernel log */
906 	if (link_state->up)
907 		netif_info(efx, link, efx->net_dev,
908 			   "link up at %uMbps %s-duplex (MTU %d)\n",
909 			   link_state->speed, link_state->fd ? "full" : "half",
910 			   efx->net_dev->mtu);
911 	else
912 		netif_info(efx, link, efx->net_dev, "link down\n");
913 }
914 
915 void efx_link_set_advertising(struct efx_nic *efx, u32 advertising)
916 {
917 	efx->link_advertising = advertising;
918 	if (advertising) {
919 		if (advertising & ADVERTISED_Pause)
920 			efx->wanted_fc |= (EFX_FC_TX | EFX_FC_RX);
921 		else
922 			efx->wanted_fc &= ~(EFX_FC_TX | EFX_FC_RX);
923 		if (advertising & ADVERTISED_Asym_Pause)
924 			efx->wanted_fc ^= EFX_FC_TX;
925 	}
926 }
927 
928 void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
929 {
930 	efx->wanted_fc = wanted_fc;
931 	if (efx->link_advertising) {
932 		if (wanted_fc & EFX_FC_RX)
933 			efx->link_advertising |= (ADVERTISED_Pause |
934 						  ADVERTISED_Asym_Pause);
935 		else
936 			efx->link_advertising &= ~(ADVERTISED_Pause |
937 						   ADVERTISED_Asym_Pause);
938 		if (wanted_fc & EFX_FC_TX)
939 			efx->link_advertising ^= ADVERTISED_Asym_Pause;
940 	}
941 }
942 
943 static void efx_fini_port(struct efx_nic *efx);
944 
945 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
946  * the MAC appropriately. All other PHY configuration changes are pushed
947  * through phy_op->set_settings(), and pushed asynchronously to the MAC
948  * through efx_monitor().
949  *
950  * Callers must hold the mac_lock
951  */
952 int __efx_reconfigure_port(struct efx_nic *efx)
953 {
954 	enum efx_phy_mode phy_mode;
955 	int rc;
956 
957 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
958 
959 	/* Disable PHY transmit in mac level loopbacks */
960 	phy_mode = efx->phy_mode;
961 	if (LOOPBACK_INTERNAL(efx))
962 		efx->phy_mode |= PHY_MODE_TX_DISABLED;
963 	else
964 		efx->phy_mode &= ~PHY_MODE_TX_DISABLED;
965 
966 	rc = efx->type->reconfigure_port(efx);
967 
968 	if (rc)
969 		efx->phy_mode = phy_mode;
970 
971 	return rc;
972 }
973 
974 /* Reinitialise the MAC to pick up new PHY settings, even if the port is
975  * disabled. */
976 int efx_reconfigure_port(struct efx_nic *efx)
977 {
978 	int rc;
979 
980 	EFX_ASSERT_RESET_SERIALISED(efx);
981 
982 	mutex_lock(&efx->mac_lock);
983 	rc = __efx_reconfigure_port(efx);
984 	mutex_unlock(&efx->mac_lock);
985 
986 	return rc;
987 }
988 
989 /* Asynchronous work item for changing MAC promiscuity and multicast
990  * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
991  * MAC directly. */
992 static void efx_mac_work(struct work_struct *data)
993 {
994 	struct efx_nic *efx = container_of(data, struct efx_nic, mac_work);
995 
996 	mutex_lock(&efx->mac_lock);
997 	if (efx->port_enabled)
998 		efx->type->reconfigure_mac(efx);
999 	mutex_unlock(&efx->mac_lock);
1000 }
1001 
1002 static int efx_probe_port(struct efx_nic *efx)
1003 {
1004 	int rc;
1005 
1006 	netif_dbg(efx, probe, efx->net_dev, "create port\n");
1007 
1008 	if (phy_flash_cfg)
1009 		efx->phy_mode = PHY_MODE_SPECIAL;
1010 
1011 	/* Connect up MAC/PHY operations table */
1012 	rc = efx->type->probe_port(efx);
1013 	if (rc)
1014 		return rc;
1015 
1016 	/* Initialise MAC address to permanent address */
1017 	memcpy(efx->net_dev->dev_addr, efx->net_dev->perm_addr, ETH_ALEN);
1018 
1019 	return 0;
1020 }
1021 
1022 static int efx_init_port(struct efx_nic *efx)
1023 {
1024 	int rc;
1025 
1026 	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1027 
1028 	mutex_lock(&efx->mac_lock);
1029 
1030 	rc = efx->phy_op->init(efx);
1031 	if (rc)
1032 		goto fail1;
1033 
1034 	efx->port_initialized = true;
1035 
1036 	/* Reconfigure the MAC before creating dma queues (required for
1037 	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1038 	efx->type->reconfigure_mac(efx);
1039 
1040 	/* Ensure the PHY advertises the correct flow control settings */
1041 	rc = efx->phy_op->reconfigure(efx);
1042 	if (rc)
1043 		goto fail2;
1044 
1045 	mutex_unlock(&efx->mac_lock);
1046 	return 0;
1047 
1048 fail2:
1049 	efx->phy_op->fini(efx);
1050 fail1:
1051 	mutex_unlock(&efx->mac_lock);
1052 	return rc;
1053 }
1054 
1055 static void efx_start_port(struct efx_nic *efx)
1056 {
1057 	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1058 	BUG_ON(efx->port_enabled);
1059 
1060 	mutex_lock(&efx->mac_lock);
1061 	efx->port_enabled = true;
1062 
1063 	/* Ensure MAC ingress/egress is enabled */
1064 	efx->type->reconfigure_mac(efx);
1065 
1066 	mutex_unlock(&efx->mac_lock);
1067 }
1068 
1069 /* Cancel work for MAC reconfiguration, periodic hardware monitoring
1070  * and the async self-test, wait for them to finish and prevent them
1071  * being scheduled again.  This doesn't cover online resets, which
1072  * should only be cancelled when removing the device.
1073  */
1074 static void efx_stop_port(struct efx_nic *efx)
1075 {
1076 	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1077 
1078 	EFX_ASSERT_RESET_SERIALISED(efx);
1079 
1080 	mutex_lock(&efx->mac_lock);
1081 	efx->port_enabled = false;
1082 	mutex_unlock(&efx->mac_lock);
1083 
1084 	/* Serialise against efx_set_multicast_list() */
1085 	netif_addr_lock_bh(efx->net_dev);
1086 	netif_addr_unlock_bh(efx->net_dev);
1087 
1088 	cancel_delayed_work_sync(&efx->monitor_work);
1089 	efx_selftest_async_cancel(efx);
1090 	cancel_work_sync(&efx->mac_work);
1091 }
1092 
1093 static void efx_fini_port(struct efx_nic *efx)
1094 {
1095 	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1096 
1097 	if (!efx->port_initialized)
1098 		return;
1099 
1100 	efx->phy_op->fini(efx);
1101 	efx->port_initialized = false;
1102 
1103 	efx->link_state.up = false;
1104 	efx_link_status_changed(efx);
1105 }
1106 
1107 static void efx_remove_port(struct efx_nic *efx)
1108 {
1109 	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1110 
1111 	efx->type->remove_port(efx);
1112 }
1113 
1114 /**************************************************************************
1115  *
1116  * NIC handling
1117  *
1118  **************************************************************************/
1119 
1120 static LIST_HEAD(efx_primary_list);
1121 static LIST_HEAD(efx_unassociated_list);
1122 
1123 static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right)
1124 {
1125 	return left->type == right->type &&
1126 		left->vpd_sn && right->vpd_sn &&
1127 		!strcmp(left->vpd_sn, right->vpd_sn);
1128 }
1129 
1130 static void efx_associate(struct efx_nic *efx)
1131 {
1132 	struct efx_nic *other, *next;
1133 
1134 	if (efx->primary == efx) {
1135 		/* Adding primary function; look for secondaries */
1136 
1137 		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
1138 		list_add_tail(&efx->node, &efx_primary_list);
1139 
1140 		list_for_each_entry_safe(other, next, &efx_unassociated_list,
1141 					 node) {
1142 			if (efx_same_controller(efx, other)) {
1143 				list_del(&other->node);
1144 				netif_dbg(other, probe, other->net_dev,
1145 					  "moving to secondary list of %s %s\n",
1146 					  pci_name(efx->pci_dev),
1147 					  efx->net_dev->name);
1148 				list_add_tail(&other->node,
1149 					      &efx->secondary_list);
1150 				other->primary = efx;
1151 			}
1152 		}
1153 	} else {
1154 		/* Adding secondary function; look for primary */
1155 
1156 		list_for_each_entry(other, &efx_primary_list, node) {
1157 			if (efx_same_controller(efx, other)) {
1158 				netif_dbg(efx, probe, efx->net_dev,
1159 					  "adding to secondary list of %s %s\n",
1160 					  pci_name(other->pci_dev),
1161 					  other->net_dev->name);
1162 				list_add_tail(&efx->node,
1163 					      &other->secondary_list);
1164 				efx->primary = other;
1165 				return;
1166 			}
1167 		}
1168 
1169 		netif_dbg(efx, probe, efx->net_dev,
1170 			  "adding to unassociated list\n");
1171 		list_add_tail(&efx->node, &efx_unassociated_list);
1172 	}
1173 }
1174 
1175 static void efx_dissociate(struct efx_nic *efx)
1176 {
1177 	struct efx_nic *other, *next;
1178 
1179 	list_del(&efx->node);
1180 	efx->primary = NULL;
1181 
1182 	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
1183 		list_del(&other->node);
1184 		netif_dbg(other, probe, other->net_dev,
1185 			  "moving to unassociated list\n");
1186 		list_add_tail(&other->node, &efx_unassociated_list);
1187 		other->primary = NULL;
1188 	}
1189 }
1190 
1191 /* This configures the PCI device to enable I/O and DMA. */
1192 static int efx_init_io(struct efx_nic *efx)
1193 {
1194 	struct pci_dev *pci_dev = efx->pci_dev;
1195 	dma_addr_t dma_mask = efx->type->max_dma_mask;
1196 	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1197 	int rc;
1198 
1199 	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1200 
1201 	rc = pci_enable_device(pci_dev);
1202 	if (rc) {
1203 		netif_err(efx, probe, efx->net_dev,
1204 			  "failed to enable PCI device\n");
1205 		goto fail1;
1206 	}
1207 
1208 	pci_set_master(pci_dev);
1209 
1210 	/* Set the PCI DMA mask.  Try all possibilities from our
1211 	 * genuine mask down to 32 bits, because some architectures
1212 	 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
1213 	 * masks event though they reject 46 bit masks.
1214 	 */
1215 	while (dma_mask > 0x7fffffffUL) {
1216 		if (dma_supported(&pci_dev->dev, dma_mask)) {
1217 			rc = dma_set_mask_and_coherent(&pci_dev->dev, dma_mask);
1218 			if (rc == 0)
1219 				break;
1220 		}
1221 		dma_mask >>= 1;
1222 	}
1223 	if (rc) {
1224 		netif_err(efx, probe, efx->net_dev,
1225 			  "could not find a suitable DMA mask\n");
1226 		goto fail2;
1227 	}
1228 	netif_dbg(efx, probe, efx->net_dev,
1229 		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1230 
1231 	efx->membase_phys = pci_resource_start(efx->pci_dev, EFX_MEM_BAR);
1232 	rc = pci_request_region(pci_dev, EFX_MEM_BAR, "sfc");
1233 	if (rc) {
1234 		netif_err(efx, probe, efx->net_dev,
1235 			  "request for memory BAR failed\n");
1236 		rc = -EIO;
1237 		goto fail3;
1238 	}
1239 	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1240 	if (!efx->membase) {
1241 		netif_err(efx, probe, efx->net_dev,
1242 			  "could not map memory BAR at %llx+%x\n",
1243 			  (unsigned long long)efx->membase_phys, mem_map_size);
1244 		rc = -ENOMEM;
1245 		goto fail4;
1246 	}
1247 	netif_dbg(efx, probe, efx->net_dev,
1248 		  "memory BAR at %llx+%x (virtual %p)\n",
1249 		  (unsigned long long)efx->membase_phys, mem_map_size,
1250 		  efx->membase);
1251 
1252 	return 0;
1253 
1254  fail4:
1255 	pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1256  fail3:
1257 	efx->membase_phys = 0;
1258  fail2:
1259 	pci_disable_device(efx->pci_dev);
1260  fail1:
1261 	return rc;
1262 }
1263 
1264 static void efx_fini_io(struct efx_nic *efx)
1265 {
1266 	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1267 
1268 	if (efx->membase) {
1269 		iounmap(efx->membase);
1270 		efx->membase = NULL;
1271 	}
1272 
1273 	if (efx->membase_phys) {
1274 		pci_release_region(efx->pci_dev, EFX_MEM_BAR);
1275 		efx->membase_phys = 0;
1276 	}
1277 
1278 	pci_disable_device(efx->pci_dev);
1279 }
1280 
1281 static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1282 {
1283 	cpumask_var_t thread_mask;
1284 	unsigned int count;
1285 	int cpu;
1286 
1287 	if (rss_cpus) {
1288 		count = rss_cpus;
1289 	} else {
1290 		if (unlikely(!zalloc_cpumask_var(&thread_mask, GFP_KERNEL))) {
1291 			netif_warn(efx, probe, efx->net_dev,
1292 				   "RSS disabled due to allocation failure\n");
1293 			return 1;
1294 		}
1295 
1296 		count = 0;
1297 		for_each_online_cpu(cpu) {
1298 			if (!cpumask_test_cpu(cpu, thread_mask)) {
1299 				++count;
1300 				cpumask_or(thread_mask, thread_mask,
1301 					   topology_thread_cpumask(cpu));
1302 			}
1303 		}
1304 
1305 		free_cpumask_var(thread_mask);
1306 	}
1307 
1308 	/* If RSS is requested for the PF *and* VFs then we can't write RSS
1309 	 * table entries that are inaccessible to VFs
1310 	 */
1311 	if (efx_sriov_wanted(efx) && efx_vf_size(efx) > 1 &&
1312 	    count > efx_vf_size(efx)) {
1313 		netif_warn(efx, probe, efx->net_dev,
1314 			   "Reducing number of RSS channels from %u to %u for "
1315 			   "VF support. Increase vf-msix-limit to use more "
1316 			   "channels on the PF.\n",
1317 			   count, efx_vf_size(efx));
1318 		count = efx_vf_size(efx);
1319 	}
1320 
1321 	return count;
1322 }
1323 
1324 /* Probe the number and type of interrupts we are able to obtain, and
1325  * the resulting numbers of channels and RX queues.
1326  */
1327 static int efx_probe_interrupts(struct efx_nic *efx)
1328 {
1329 	unsigned int extra_channels = 0;
1330 	unsigned int i, j;
1331 	int rc;
1332 
1333 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
1334 		if (efx->extra_channel_type[i])
1335 			++extra_channels;
1336 
1337 	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1338 		struct msix_entry xentries[EFX_MAX_CHANNELS];
1339 		unsigned int n_channels;
1340 
1341 		n_channels = efx_wanted_parallelism(efx);
1342 		if (separate_tx_channels)
1343 			n_channels *= 2;
1344 		n_channels += extra_channels;
1345 		n_channels = min(n_channels, efx->max_channels);
1346 
1347 		for (i = 0; i < n_channels; i++)
1348 			xentries[i].entry = i;
1349 		rc = pci_enable_msix(efx->pci_dev, xentries, n_channels);
1350 		if (rc > 0) {
1351 			netif_err(efx, drv, efx->net_dev,
1352 				  "WARNING: Insufficient MSI-X vectors"
1353 				  " available (%d < %u).\n", rc, n_channels);
1354 			netif_err(efx, drv, efx->net_dev,
1355 				  "WARNING: Performance may be reduced.\n");
1356 			EFX_BUG_ON_PARANOID(rc >= n_channels);
1357 			n_channels = rc;
1358 			rc = pci_enable_msix(efx->pci_dev, xentries,
1359 					     n_channels);
1360 		}
1361 
1362 		if (rc == 0) {
1363 			efx->n_channels = n_channels;
1364 			if (n_channels > extra_channels)
1365 				n_channels -= extra_channels;
1366 			if (separate_tx_channels) {
1367 				efx->n_tx_channels = max(n_channels / 2, 1U);
1368 				efx->n_rx_channels = max(n_channels -
1369 							 efx->n_tx_channels,
1370 							 1U);
1371 			} else {
1372 				efx->n_tx_channels = n_channels;
1373 				efx->n_rx_channels = n_channels;
1374 			}
1375 			for (i = 0; i < efx->n_channels; i++)
1376 				efx_get_channel(efx, i)->irq =
1377 					xentries[i].vector;
1378 		} else {
1379 			/* Fall back to single channel MSI */
1380 			efx->interrupt_mode = EFX_INT_MODE_MSI;
1381 			netif_err(efx, drv, efx->net_dev,
1382 				  "could not enable MSI-X\n");
1383 		}
1384 	}
1385 
1386 	/* Try single interrupt MSI */
1387 	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1388 		efx->n_channels = 1;
1389 		efx->n_rx_channels = 1;
1390 		efx->n_tx_channels = 1;
1391 		rc = pci_enable_msi(efx->pci_dev);
1392 		if (rc == 0) {
1393 			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1394 		} else {
1395 			netif_err(efx, drv, efx->net_dev,
1396 				  "could not enable MSI\n");
1397 			efx->interrupt_mode = EFX_INT_MODE_LEGACY;
1398 		}
1399 	}
1400 
1401 	/* Assume legacy interrupts */
1402 	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1403 		efx->n_channels = 1 + (separate_tx_channels ? 1 : 0);
1404 		efx->n_rx_channels = 1;
1405 		efx->n_tx_channels = 1;
1406 		efx->legacy_irq = efx->pci_dev->irq;
1407 	}
1408 
1409 	/* Assign extra channels if possible */
1410 	j = efx->n_channels;
1411 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++) {
1412 		if (!efx->extra_channel_type[i])
1413 			continue;
1414 		if (efx->interrupt_mode != EFX_INT_MODE_MSIX ||
1415 		    efx->n_channels <= extra_channels) {
1416 			efx->extra_channel_type[i]->handle_no_channel(efx);
1417 		} else {
1418 			--j;
1419 			efx_get_channel(efx, j)->type =
1420 				efx->extra_channel_type[i];
1421 		}
1422 	}
1423 
1424 	/* RSS might be usable on VFs even if it is disabled on the PF */
1425 	efx->rss_spread = ((efx->n_rx_channels > 1 || !efx_sriov_wanted(efx)) ?
1426 			   efx->n_rx_channels : efx_vf_size(efx));
1427 
1428 	return 0;
1429 }
1430 
1431 static int efx_soft_enable_interrupts(struct efx_nic *efx)
1432 {
1433 	struct efx_channel *channel, *end_channel;
1434 	int rc;
1435 
1436 	BUG_ON(efx->state == STATE_DISABLED);
1437 
1438 	efx->irq_soft_enabled = true;
1439 	smp_wmb();
1440 
1441 	efx_for_each_channel(channel, efx) {
1442 		if (!channel->type->keep_eventq) {
1443 			rc = efx_init_eventq(channel);
1444 			if (rc)
1445 				goto fail;
1446 		}
1447 		efx_start_eventq(channel);
1448 	}
1449 
1450 	efx_mcdi_mode_event(efx);
1451 
1452 	return 0;
1453 fail:
1454 	end_channel = channel;
1455 	efx_for_each_channel(channel, efx) {
1456 		if (channel == end_channel)
1457 			break;
1458 		efx_stop_eventq(channel);
1459 		if (!channel->type->keep_eventq)
1460 			efx_fini_eventq(channel);
1461 	}
1462 
1463 	return rc;
1464 }
1465 
1466 static void efx_soft_disable_interrupts(struct efx_nic *efx)
1467 {
1468 	struct efx_channel *channel;
1469 
1470 	if (efx->state == STATE_DISABLED)
1471 		return;
1472 
1473 	efx_mcdi_mode_poll(efx);
1474 
1475 	efx->irq_soft_enabled = false;
1476 	smp_wmb();
1477 
1478 	if (efx->legacy_irq)
1479 		synchronize_irq(efx->legacy_irq);
1480 
1481 	efx_for_each_channel(channel, efx) {
1482 		if (channel->irq)
1483 			synchronize_irq(channel->irq);
1484 
1485 		efx_stop_eventq(channel);
1486 		if (!channel->type->keep_eventq)
1487 			efx_fini_eventq(channel);
1488 	}
1489 
1490 	/* Flush the asynchronous MCDI request queue */
1491 	efx_mcdi_flush_async(efx);
1492 }
1493 
1494 static int efx_enable_interrupts(struct efx_nic *efx)
1495 {
1496 	struct efx_channel *channel, *end_channel;
1497 	int rc;
1498 
1499 	BUG_ON(efx->state == STATE_DISABLED);
1500 
1501 	if (efx->eeh_disabled_legacy_irq) {
1502 		enable_irq(efx->legacy_irq);
1503 		efx->eeh_disabled_legacy_irq = false;
1504 	}
1505 
1506 	efx->type->irq_enable_master(efx);
1507 
1508 	efx_for_each_channel(channel, efx) {
1509 		if (channel->type->keep_eventq) {
1510 			rc = efx_init_eventq(channel);
1511 			if (rc)
1512 				goto fail;
1513 		}
1514 	}
1515 
1516 	rc = efx_soft_enable_interrupts(efx);
1517 	if (rc)
1518 		goto fail;
1519 
1520 	return 0;
1521 
1522 fail:
1523 	end_channel = channel;
1524 	efx_for_each_channel(channel, efx) {
1525 		if (channel == end_channel)
1526 			break;
1527 		if (channel->type->keep_eventq)
1528 			efx_fini_eventq(channel);
1529 	}
1530 
1531 	efx->type->irq_disable_non_ev(efx);
1532 
1533 	return rc;
1534 }
1535 
1536 static void efx_disable_interrupts(struct efx_nic *efx)
1537 {
1538 	struct efx_channel *channel;
1539 
1540 	efx_soft_disable_interrupts(efx);
1541 
1542 	efx_for_each_channel(channel, efx) {
1543 		if (channel->type->keep_eventq)
1544 			efx_fini_eventq(channel);
1545 	}
1546 
1547 	efx->type->irq_disable_non_ev(efx);
1548 }
1549 
1550 static void efx_remove_interrupts(struct efx_nic *efx)
1551 {
1552 	struct efx_channel *channel;
1553 
1554 	/* Remove MSI/MSI-X interrupts */
1555 	efx_for_each_channel(channel, efx)
1556 		channel->irq = 0;
1557 	pci_disable_msi(efx->pci_dev);
1558 	pci_disable_msix(efx->pci_dev);
1559 
1560 	/* Remove legacy interrupt */
1561 	efx->legacy_irq = 0;
1562 }
1563 
1564 static void efx_set_channels(struct efx_nic *efx)
1565 {
1566 	struct efx_channel *channel;
1567 	struct efx_tx_queue *tx_queue;
1568 
1569 	efx->tx_channel_offset =
1570 		separate_tx_channels ? efx->n_channels - efx->n_tx_channels : 0;
1571 
1572 	/* We need to mark which channels really have RX and TX
1573 	 * queues, and adjust the TX queue numbers if we have separate
1574 	 * RX-only and TX-only channels.
1575 	 */
1576 	efx_for_each_channel(channel, efx) {
1577 		if (channel->channel < efx->n_rx_channels)
1578 			channel->rx_queue.core_index = channel->channel;
1579 		else
1580 			channel->rx_queue.core_index = -1;
1581 
1582 		efx_for_each_channel_tx_queue(tx_queue, channel)
1583 			tx_queue->queue -= (efx->tx_channel_offset *
1584 					    EFX_TXQ_TYPES);
1585 	}
1586 }
1587 
1588 static int efx_probe_nic(struct efx_nic *efx)
1589 {
1590 	size_t i;
1591 	int rc;
1592 
1593 	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1594 
1595 	/* Carry out hardware-type specific initialisation */
1596 	rc = efx->type->probe(efx);
1597 	if (rc)
1598 		return rc;
1599 
1600 	/* Determine the number of channels and queues by trying to hook
1601 	 * in MSI-X interrupts. */
1602 	rc = efx_probe_interrupts(efx);
1603 	if (rc)
1604 		goto fail1;
1605 
1606 	rc = efx->type->dimension_resources(efx);
1607 	if (rc)
1608 		goto fail2;
1609 
1610 	if (efx->n_channels > 1)
1611 		get_random_bytes(&efx->rx_hash_key, sizeof(efx->rx_hash_key));
1612 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
1613 		efx->rx_indir_table[i] =
1614 			ethtool_rxfh_indir_default(i, efx->rss_spread);
1615 
1616 	efx_set_channels(efx);
1617 	netif_set_real_num_tx_queues(efx->net_dev, efx->n_tx_channels);
1618 	netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels);
1619 
1620 	/* Initialise the interrupt moderation settings */
1621 	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
1622 				true);
1623 
1624 	return 0;
1625 
1626 fail2:
1627 	efx_remove_interrupts(efx);
1628 fail1:
1629 	efx->type->remove(efx);
1630 	return rc;
1631 }
1632 
1633 static void efx_remove_nic(struct efx_nic *efx)
1634 {
1635 	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1636 
1637 	efx_remove_interrupts(efx);
1638 	efx->type->remove(efx);
1639 }
1640 
1641 static int efx_probe_filters(struct efx_nic *efx)
1642 {
1643 	int rc;
1644 
1645 	spin_lock_init(&efx->filter_lock);
1646 
1647 	rc = efx->type->filter_table_probe(efx);
1648 	if (rc)
1649 		return rc;
1650 
1651 #ifdef CONFIG_RFS_ACCEL
1652 	if (efx->type->offload_features & NETIF_F_NTUPLE) {
1653 		efx->rps_flow_id = kcalloc(efx->type->max_rx_ip_filters,
1654 					   sizeof(*efx->rps_flow_id),
1655 					   GFP_KERNEL);
1656 		if (!efx->rps_flow_id) {
1657 			efx->type->filter_table_remove(efx);
1658 			return -ENOMEM;
1659 		}
1660 	}
1661 #endif
1662 
1663 	return 0;
1664 }
1665 
1666 static void efx_remove_filters(struct efx_nic *efx)
1667 {
1668 #ifdef CONFIG_RFS_ACCEL
1669 	kfree(efx->rps_flow_id);
1670 #endif
1671 	efx->type->filter_table_remove(efx);
1672 }
1673 
1674 static void efx_restore_filters(struct efx_nic *efx)
1675 {
1676 	efx->type->filter_table_restore(efx);
1677 }
1678 
1679 /**************************************************************************
1680  *
1681  * NIC startup/shutdown
1682  *
1683  *************************************************************************/
1684 
1685 static int efx_probe_all(struct efx_nic *efx)
1686 {
1687 	int rc;
1688 
1689 	rc = efx_probe_nic(efx);
1690 	if (rc) {
1691 		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1692 		goto fail1;
1693 	}
1694 
1695 	rc = efx_probe_port(efx);
1696 	if (rc) {
1697 		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1698 		goto fail2;
1699 	}
1700 
1701 	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
1702 	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
1703 		rc = -EINVAL;
1704 		goto fail3;
1705 	}
1706 	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1707 
1708 	rc = efx_probe_filters(efx);
1709 	if (rc) {
1710 		netif_err(efx, probe, efx->net_dev,
1711 			  "failed to create filter tables\n");
1712 		goto fail3;
1713 	}
1714 
1715 	rc = efx_probe_channels(efx);
1716 	if (rc)
1717 		goto fail4;
1718 
1719 	return 0;
1720 
1721  fail4:
1722 	efx_remove_filters(efx);
1723  fail3:
1724 	efx_remove_port(efx);
1725  fail2:
1726 	efx_remove_nic(efx);
1727  fail1:
1728 	return rc;
1729 }
1730 
1731 /* If the interface is supposed to be running but is not, start
1732  * the hardware and software data path, regular activity for the port
1733  * (MAC statistics, link polling, etc.) and schedule the port to be
1734  * reconfigured.  Interrupts must already be enabled.  This function
1735  * is safe to call multiple times, so long as the NIC is not disabled.
1736  * Requires the RTNL lock.
1737  */
1738 static void efx_start_all(struct efx_nic *efx)
1739 {
1740 	EFX_ASSERT_RESET_SERIALISED(efx);
1741 	BUG_ON(efx->state == STATE_DISABLED);
1742 
1743 	/* Check that it is appropriate to restart the interface. All
1744 	 * of these flags are safe to read under just the rtnl lock */
1745 	if (efx->port_enabled || !netif_running(efx->net_dev))
1746 		return;
1747 
1748 	efx_start_port(efx);
1749 	efx_start_datapath(efx);
1750 
1751 	/* Start the hardware monitor if there is one */
1752 	if (efx->type->monitor != NULL)
1753 		queue_delayed_work(efx->workqueue, &efx->monitor_work,
1754 				   efx_monitor_interval);
1755 
1756 	/* If link state detection is normally event-driven, we have
1757 	 * to poll now because we could have missed a change
1758 	 */
1759 	if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0) {
1760 		mutex_lock(&efx->mac_lock);
1761 		if (efx->phy_op->poll(efx))
1762 			efx_link_status_changed(efx);
1763 		mutex_unlock(&efx->mac_lock);
1764 	}
1765 
1766 	efx->type->start_stats(efx);
1767 	efx->type->pull_stats(efx);
1768 	spin_lock_bh(&efx->stats_lock);
1769 	efx->type->update_stats(efx, NULL, NULL);
1770 	spin_unlock_bh(&efx->stats_lock);
1771 }
1772 
1773 /* Quiesce the hardware and software data path, and regular activity
1774  * for the port without bringing the link down.  Safe to call multiple
1775  * times with the NIC in almost any state, but interrupts should be
1776  * enabled.  Requires the RTNL lock.
1777  */
1778 static void efx_stop_all(struct efx_nic *efx)
1779 {
1780 	EFX_ASSERT_RESET_SERIALISED(efx);
1781 
1782 	/* port_enabled can be read safely under the rtnl lock */
1783 	if (!efx->port_enabled)
1784 		return;
1785 
1786 	/* update stats before we go down so we can accurately count
1787 	 * rx_nodesc_drops
1788 	 */
1789 	efx->type->pull_stats(efx);
1790 	spin_lock_bh(&efx->stats_lock);
1791 	efx->type->update_stats(efx, NULL, NULL);
1792 	spin_unlock_bh(&efx->stats_lock);
1793 	efx->type->stop_stats(efx);
1794 	efx_stop_port(efx);
1795 
1796 	/* Stop the kernel transmit interface.  This is only valid if
1797 	 * the device is stopped or detached; otherwise the watchdog
1798 	 * may fire immediately.
1799 	 */
1800 	WARN_ON(netif_running(efx->net_dev) &&
1801 		netif_device_present(efx->net_dev));
1802 	netif_tx_disable(efx->net_dev);
1803 
1804 	efx_stop_datapath(efx);
1805 }
1806 
1807 static void efx_remove_all(struct efx_nic *efx)
1808 {
1809 	efx_remove_channels(efx);
1810 	efx_remove_filters(efx);
1811 	efx_remove_port(efx);
1812 	efx_remove_nic(efx);
1813 }
1814 
1815 /**************************************************************************
1816  *
1817  * Interrupt moderation
1818  *
1819  **************************************************************************/
1820 
1821 static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int quantum_ns)
1822 {
1823 	if (usecs == 0)
1824 		return 0;
1825 	if (usecs * 1000 < quantum_ns)
1826 		return 1; /* never round down to 0 */
1827 	return usecs * 1000 / quantum_ns;
1828 }
1829 
1830 /* Set interrupt moderation parameters */
1831 int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
1832 			    unsigned int rx_usecs, bool rx_adaptive,
1833 			    bool rx_may_override_tx)
1834 {
1835 	struct efx_channel *channel;
1836 	unsigned int irq_mod_max = DIV_ROUND_UP(efx->type->timer_period_max *
1837 						efx->timer_quantum_ns,
1838 						1000);
1839 	unsigned int tx_ticks;
1840 	unsigned int rx_ticks;
1841 
1842 	EFX_ASSERT_RESET_SERIALISED(efx);
1843 
1844 	if (tx_usecs > irq_mod_max || rx_usecs > irq_mod_max)
1845 		return -EINVAL;
1846 
1847 	tx_ticks = irq_mod_ticks(tx_usecs, efx->timer_quantum_ns);
1848 	rx_ticks = irq_mod_ticks(rx_usecs, efx->timer_quantum_ns);
1849 
1850 	if (tx_ticks != rx_ticks && efx->tx_channel_offset == 0 &&
1851 	    !rx_may_override_tx) {
1852 		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
1853 			  "RX and TX IRQ moderation must be equal\n");
1854 		return -EINVAL;
1855 	}
1856 
1857 	efx->irq_rx_adaptive = rx_adaptive;
1858 	efx->irq_rx_moderation = rx_ticks;
1859 	efx_for_each_channel(channel, efx) {
1860 		if (efx_channel_has_rx_queue(channel))
1861 			channel->irq_moderation = rx_ticks;
1862 		else if (efx_channel_has_tx_queues(channel))
1863 			channel->irq_moderation = tx_ticks;
1864 	}
1865 
1866 	return 0;
1867 }
1868 
1869 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
1870 			    unsigned int *rx_usecs, bool *rx_adaptive)
1871 {
1872 	/* We must round up when converting ticks to microseconds
1873 	 * because we round down when converting the other way.
1874 	 */
1875 
1876 	*rx_adaptive = efx->irq_rx_adaptive;
1877 	*rx_usecs = DIV_ROUND_UP(efx->irq_rx_moderation *
1878 				 efx->timer_quantum_ns,
1879 				 1000);
1880 
1881 	/* If channels are shared between RX and TX, so is IRQ
1882 	 * moderation.  Otherwise, IRQ moderation is the same for all
1883 	 * TX channels and is not adaptive.
1884 	 */
1885 	if (efx->tx_channel_offset == 0)
1886 		*tx_usecs = *rx_usecs;
1887 	else
1888 		*tx_usecs = DIV_ROUND_UP(
1889 			efx->channel[efx->tx_channel_offset]->irq_moderation *
1890 			efx->timer_quantum_ns,
1891 			1000);
1892 }
1893 
1894 /**************************************************************************
1895  *
1896  * Hardware monitor
1897  *
1898  **************************************************************************/
1899 
1900 /* Run periodically off the general workqueue */
1901 static void efx_monitor(struct work_struct *data)
1902 {
1903 	struct efx_nic *efx = container_of(data, struct efx_nic,
1904 					   monitor_work.work);
1905 
1906 	netif_vdbg(efx, timer, efx->net_dev,
1907 		   "hardware monitor executing on CPU %d\n",
1908 		   raw_smp_processor_id());
1909 	BUG_ON(efx->type->monitor == NULL);
1910 
1911 	/* If the mac_lock is already held then it is likely a port
1912 	 * reconfiguration is already in place, which will likely do
1913 	 * most of the work of monitor() anyway. */
1914 	if (mutex_trylock(&efx->mac_lock)) {
1915 		if (efx->port_enabled)
1916 			efx->type->monitor(efx);
1917 		mutex_unlock(&efx->mac_lock);
1918 	}
1919 
1920 	queue_delayed_work(efx->workqueue, &efx->monitor_work,
1921 			   efx_monitor_interval);
1922 }
1923 
1924 /**************************************************************************
1925  *
1926  * ioctls
1927  *
1928  *************************************************************************/
1929 
1930 /* Net device ioctl
1931  * Context: process, rtnl_lock() held.
1932  */
1933 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
1934 {
1935 	struct efx_nic *efx = netdev_priv(net_dev);
1936 	struct mii_ioctl_data *data = if_mii(ifr);
1937 
1938 	if (cmd == SIOCSHWTSTAMP)
1939 		return efx_ptp_set_ts_config(efx, ifr);
1940 	if (cmd == SIOCGHWTSTAMP)
1941 		return efx_ptp_get_ts_config(efx, ifr);
1942 
1943 	/* Convert phy_id from older PRTAD/DEVAD format */
1944 	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
1945 	    (data->phy_id & 0xfc00) == 0x0400)
1946 		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;
1947 
1948 	return mdio_mii_ioctl(&efx->mdio, data, cmd);
1949 }
1950 
1951 /**************************************************************************
1952  *
1953  * NAPI interface
1954  *
1955  **************************************************************************/
1956 
1957 static void efx_init_napi_channel(struct efx_channel *channel)
1958 {
1959 	struct efx_nic *efx = channel->efx;
1960 
1961 	channel->napi_dev = efx->net_dev;
1962 	netif_napi_add(channel->napi_dev, &channel->napi_str,
1963 		       efx_poll, napi_weight);
1964 }
1965 
1966 static void efx_init_napi(struct efx_nic *efx)
1967 {
1968 	struct efx_channel *channel;
1969 
1970 	efx_for_each_channel(channel, efx)
1971 		efx_init_napi_channel(channel);
1972 }
1973 
1974 static void efx_fini_napi_channel(struct efx_channel *channel)
1975 {
1976 	if (channel->napi_dev)
1977 		netif_napi_del(&channel->napi_str);
1978 	channel->napi_dev = NULL;
1979 }
1980 
1981 static void efx_fini_napi(struct efx_nic *efx)
1982 {
1983 	struct efx_channel *channel;
1984 
1985 	efx_for_each_channel(channel, efx)
1986 		efx_fini_napi_channel(channel);
1987 }
1988 
1989 /**************************************************************************
1990  *
1991  * Kernel netpoll interface
1992  *
1993  *************************************************************************/
1994 
1995 #ifdef CONFIG_NET_POLL_CONTROLLER
1996 
1997 /* Although in the common case interrupts will be disabled, this is not
1998  * guaranteed. However, all our work happens inside the NAPI callback,
1999  * so no locking is required.
2000  */
2001 static void efx_netpoll(struct net_device *net_dev)
2002 {
2003 	struct efx_nic *efx = netdev_priv(net_dev);
2004 	struct efx_channel *channel;
2005 
2006 	efx_for_each_channel(channel, efx)
2007 		efx_schedule_channel(channel);
2008 }
2009 
2010 #endif
2011 
2012 /**************************************************************************
2013  *
2014  * Kernel net device interface
2015  *
2016  *************************************************************************/
2017 
2018 /* Context: process, rtnl_lock() held. */
2019 static int efx_net_open(struct net_device *net_dev)
2020 {
2021 	struct efx_nic *efx = netdev_priv(net_dev);
2022 	int rc;
2023 
2024 	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
2025 		  raw_smp_processor_id());
2026 
2027 	rc = efx_check_disabled(efx);
2028 	if (rc)
2029 		return rc;
2030 	if (efx->phy_mode & PHY_MODE_SPECIAL)
2031 		return -EBUSY;
2032 	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
2033 		return -EIO;
2034 
2035 	/* Notify the kernel of the link state polled during driver load,
2036 	 * before the monitor starts running */
2037 	efx_link_status_changed(efx);
2038 
2039 	efx_start_all(efx);
2040 	efx_selftest_async_start(efx);
2041 	return 0;
2042 }
2043 
2044 /* Context: process, rtnl_lock() held.
2045  * Note that the kernel will ignore our return code; this method
2046  * should really be a void.
2047  */
2048 static int efx_net_stop(struct net_device *net_dev)
2049 {
2050 	struct efx_nic *efx = netdev_priv(net_dev);
2051 
2052 	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
2053 		  raw_smp_processor_id());
2054 
2055 	/* Stop the device and flush all the channels */
2056 	efx_stop_all(efx);
2057 
2058 	return 0;
2059 }
2060 
2061 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
2062 static struct rtnl_link_stats64 *efx_net_stats(struct net_device *net_dev,
2063 					       struct rtnl_link_stats64 *stats)
2064 {
2065 	struct efx_nic *efx = netdev_priv(net_dev);
2066 
2067 	spin_lock_bh(&efx->stats_lock);
2068 	efx->type->update_stats(efx, NULL, stats);
2069 	spin_unlock_bh(&efx->stats_lock);
2070 
2071 	return stats;
2072 }
2073 
2074 /* Context: netif_tx_lock held, BHs disabled. */
2075 static void efx_watchdog(struct net_device *net_dev)
2076 {
2077 	struct efx_nic *efx = netdev_priv(net_dev);
2078 
2079 	netif_err(efx, tx_err, efx->net_dev,
2080 		  "TX stuck with port_enabled=%d: resetting channels\n",
2081 		  efx->port_enabled);
2082 
2083 	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2084 }
2085 
2086 
2087 /* Context: process, rtnl_lock() held. */
2088 static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
2089 {
2090 	struct efx_nic *efx = netdev_priv(net_dev);
2091 	int rc;
2092 
2093 	rc = efx_check_disabled(efx);
2094 	if (rc)
2095 		return rc;
2096 	if (new_mtu > EFX_MAX_MTU)
2097 		return -EINVAL;
2098 
2099 	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2100 
2101 	efx_device_detach_sync(efx);
2102 	efx_stop_all(efx);
2103 
2104 	mutex_lock(&efx->mac_lock);
2105 	net_dev->mtu = new_mtu;
2106 	efx->type->reconfigure_mac(efx);
2107 	mutex_unlock(&efx->mac_lock);
2108 
2109 	efx_start_all(efx);
2110 	netif_device_attach(efx->net_dev);
2111 	return 0;
2112 }
2113 
2114 static int efx_set_mac_address(struct net_device *net_dev, void *data)
2115 {
2116 	struct efx_nic *efx = netdev_priv(net_dev);
2117 	struct sockaddr *addr = data;
2118 	char *new_addr = addr->sa_data;
2119 
2120 	if (!is_valid_ether_addr(new_addr)) {
2121 		netif_err(efx, drv, efx->net_dev,
2122 			  "invalid ethernet MAC address requested: %pM\n",
2123 			  new_addr);
2124 		return -EADDRNOTAVAIL;
2125 	}
2126 
2127 	memcpy(net_dev->dev_addr, new_addr, net_dev->addr_len);
2128 	efx_sriov_mac_address_changed(efx);
2129 
2130 	/* Reconfigure the MAC */
2131 	mutex_lock(&efx->mac_lock);
2132 	efx->type->reconfigure_mac(efx);
2133 	mutex_unlock(&efx->mac_lock);
2134 
2135 	return 0;
2136 }
2137 
2138 /* Context: netif_addr_lock held, BHs disabled. */
2139 static void efx_set_rx_mode(struct net_device *net_dev)
2140 {
2141 	struct efx_nic *efx = netdev_priv(net_dev);
2142 
2143 	if (efx->port_enabled)
2144 		queue_work(efx->workqueue, &efx->mac_work);
2145 	/* Otherwise efx_start_port() will do this */
2146 }
2147 
2148 static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2149 {
2150 	struct efx_nic *efx = netdev_priv(net_dev);
2151 
2152 	/* If disabling RX n-tuple filtering, clear existing filters */
2153 	if (net_dev->features & ~data & NETIF_F_NTUPLE)
2154 		return efx->type->filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);
2155 
2156 	return 0;
2157 }
2158 
2159 static const struct net_device_ops efx_farch_netdev_ops = {
2160 	.ndo_open		= efx_net_open,
2161 	.ndo_stop		= efx_net_stop,
2162 	.ndo_get_stats64	= efx_net_stats,
2163 	.ndo_tx_timeout		= efx_watchdog,
2164 	.ndo_start_xmit		= efx_hard_start_xmit,
2165 	.ndo_validate_addr	= eth_validate_addr,
2166 	.ndo_do_ioctl		= efx_ioctl,
2167 	.ndo_change_mtu		= efx_change_mtu,
2168 	.ndo_set_mac_address	= efx_set_mac_address,
2169 	.ndo_set_rx_mode	= efx_set_rx_mode,
2170 	.ndo_set_features	= efx_set_features,
2171 #ifdef CONFIG_SFC_SRIOV
2172 	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
2173 	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
2174 	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
2175 	.ndo_get_vf_config	= efx_sriov_get_vf_config,
2176 #endif
2177 #ifdef CONFIG_NET_POLL_CONTROLLER
2178 	.ndo_poll_controller = efx_netpoll,
2179 #endif
2180 	.ndo_setup_tc		= efx_setup_tc,
2181 #ifdef CONFIG_RFS_ACCEL
2182 	.ndo_rx_flow_steer	= efx_filter_rfs,
2183 #endif
2184 };
2185 
2186 static const struct net_device_ops efx_ef10_netdev_ops = {
2187 	.ndo_open		= efx_net_open,
2188 	.ndo_stop		= efx_net_stop,
2189 	.ndo_get_stats64	= efx_net_stats,
2190 	.ndo_tx_timeout		= efx_watchdog,
2191 	.ndo_start_xmit		= efx_hard_start_xmit,
2192 	.ndo_validate_addr	= eth_validate_addr,
2193 	.ndo_do_ioctl		= efx_ioctl,
2194 	.ndo_change_mtu		= efx_change_mtu,
2195 	.ndo_set_mac_address	= efx_set_mac_address,
2196 	.ndo_set_rx_mode	= efx_set_rx_mode,
2197 	.ndo_set_features	= efx_set_features,
2198 #ifdef CONFIG_NET_POLL_CONTROLLER
2199 	.ndo_poll_controller	= efx_netpoll,
2200 #endif
2201 #ifdef CONFIG_RFS_ACCEL
2202 	.ndo_rx_flow_steer	= efx_filter_rfs,
2203 #endif
2204 };
2205 
2206 static void efx_update_name(struct efx_nic *efx)
2207 {
2208 	strcpy(efx->name, efx->net_dev->name);
2209 	efx_mtd_rename(efx);
2210 	efx_set_channel_names(efx);
2211 }
2212 
2213 static int efx_netdev_event(struct notifier_block *this,
2214 			    unsigned long event, void *ptr)
2215 {
2216 	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2217 
2218 	if ((net_dev->netdev_ops == &efx_farch_netdev_ops ||
2219 	     net_dev->netdev_ops == &efx_ef10_netdev_ops) &&
2220 	    event == NETDEV_CHANGENAME)
2221 		efx_update_name(netdev_priv(net_dev));
2222 
2223 	return NOTIFY_DONE;
2224 }
2225 
2226 static struct notifier_block efx_netdev_notifier = {
2227 	.notifier_call = efx_netdev_event,
2228 };
2229 
2230 static ssize_t
2231 show_phy_type(struct device *dev, struct device_attribute *attr, char *buf)
2232 {
2233 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2234 	return sprintf(buf, "%d\n", efx->phy_type);
2235 }
2236 static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
2237 
2238 static int efx_register_netdev(struct efx_nic *efx)
2239 {
2240 	struct net_device *net_dev = efx->net_dev;
2241 	struct efx_channel *channel;
2242 	int rc;
2243 
2244 	net_dev->watchdog_timeo = 5 * HZ;
2245 	net_dev->irq = efx->pci_dev->irq;
2246 	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0) {
2247 		net_dev->netdev_ops = &efx_ef10_netdev_ops;
2248 		net_dev->priv_flags |= IFF_UNICAST_FLT;
2249 	} else {
2250 		net_dev->netdev_ops = &efx_farch_netdev_ops;
2251 	}
2252 	SET_ETHTOOL_OPS(net_dev, &efx_ethtool_ops);
2253 	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2254 
2255 	rtnl_lock();
2256 
2257 	/* Enable resets to be scheduled and check whether any were
2258 	 * already requested.  If so, the NIC is probably hosed so we
2259 	 * abort.
2260 	 */
2261 	efx->state = STATE_READY;
2262 	smp_mb(); /* ensure we change state before checking reset_pending */
2263 	if (efx->reset_pending) {
2264 		netif_err(efx, probe, efx->net_dev,
2265 			  "aborting probe due to scheduled reset\n");
2266 		rc = -EIO;
2267 		goto fail_locked;
2268 	}
2269 
2270 	rc = dev_alloc_name(net_dev, net_dev->name);
2271 	if (rc < 0)
2272 		goto fail_locked;
2273 	efx_update_name(efx);
2274 
2275 	/* Always start with carrier off; PHY events will detect the link */
2276 	netif_carrier_off(net_dev);
2277 
2278 	rc = register_netdevice(net_dev);
2279 	if (rc)
2280 		goto fail_locked;
2281 
2282 	efx_for_each_channel(channel, efx) {
2283 		struct efx_tx_queue *tx_queue;
2284 		efx_for_each_channel_tx_queue(tx_queue, channel)
2285 			efx_init_tx_queue_core_txq(tx_queue);
2286 	}
2287 
2288 	efx_associate(efx);
2289 
2290 	rtnl_unlock();
2291 
2292 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2293 	if (rc) {
2294 		netif_err(efx, drv, efx->net_dev,
2295 			  "failed to init net dev attributes\n");
2296 		goto fail_registered;
2297 	}
2298 
2299 	return 0;
2300 
2301 fail_registered:
2302 	rtnl_lock();
2303 	efx_dissociate(efx);
2304 	unregister_netdevice(net_dev);
2305 fail_locked:
2306 	efx->state = STATE_UNINIT;
2307 	rtnl_unlock();
2308 	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2309 	return rc;
2310 }
2311 
2312 static void efx_unregister_netdev(struct efx_nic *efx)
2313 {
2314 	if (!efx->net_dev)
2315 		return;
2316 
2317 	BUG_ON(netdev_priv(efx->net_dev) != efx);
2318 
2319 	strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
2320 	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2321 
2322 	rtnl_lock();
2323 	unregister_netdevice(efx->net_dev);
2324 	efx->state = STATE_UNINIT;
2325 	rtnl_unlock();
2326 }
2327 
2328 /**************************************************************************
2329  *
2330  * Device reset and suspend
2331  *
2332  **************************************************************************/
2333 
2334 /* Tears down the entire software state and most of the hardware state
2335  * before reset.  */
2336 void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2337 {
2338 	EFX_ASSERT_RESET_SERIALISED(efx);
2339 
2340 	efx_stop_all(efx);
2341 	efx_disable_interrupts(efx);
2342 
2343 	mutex_lock(&efx->mac_lock);
2344 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE)
2345 		efx->phy_op->fini(efx);
2346 	efx->type->fini(efx);
2347 }
2348 
2349 /* This function will always ensure that the locks acquired in
2350  * efx_reset_down() are released. A failure return code indicates
2351  * that we were unable to reinitialise the hardware, and the
2352  * driver should be disabled. If ok is false, then the rx and tx
2353  * engines are not restarted, pending a RESET_DISABLE. */
2354 int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2355 {
2356 	int rc;
2357 
2358 	EFX_ASSERT_RESET_SERIALISED(efx);
2359 
2360 	rc = efx->type->init(efx);
2361 	if (rc) {
2362 		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2363 		goto fail;
2364 	}
2365 
2366 	if (!ok)
2367 		goto fail;
2368 
2369 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE) {
2370 		rc = efx->phy_op->init(efx);
2371 		if (rc)
2372 			goto fail;
2373 		if (efx->phy_op->reconfigure(efx))
2374 			netif_err(efx, drv, efx->net_dev,
2375 				  "could not restore PHY settings\n");
2376 	}
2377 
2378 	rc = efx_enable_interrupts(efx);
2379 	if (rc)
2380 		goto fail;
2381 	efx_restore_filters(efx);
2382 	efx_sriov_reset(efx);
2383 
2384 	mutex_unlock(&efx->mac_lock);
2385 
2386 	efx_start_all(efx);
2387 
2388 	return 0;
2389 
2390 fail:
2391 	efx->port_initialized = false;
2392 
2393 	mutex_unlock(&efx->mac_lock);
2394 
2395 	return rc;
2396 }
2397 
2398 /* Reset the NIC using the specified method.  Note that the reset may
2399  * fail, in which case the card will be left in an unusable state.
2400  *
2401  * Caller must hold the rtnl_lock.
2402  */
2403 int efx_reset(struct efx_nic *efx, enum reset_type method)
2404 {
2405 	int rc, rc2;
2406 	bool disabled;
2407 
2408 	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
2409 		   RESET_TYPE(method));
2410 
2411 	efx_device_detach_sync(efx);
2412 	efx_reset_down(efx, method);
2413 
2414 	rc = efx->type->reset(efx, method);
2415 	if (rc) {
2416 		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2417 		goto out;
2418 	}
2419 
2420 	/* Clear flags for the scopes we covered.  We assume the NIC and
2421 	 * driver are now quiescent so that there is no race here.
2422 	 */
2423 	efx->reset_pending &= -(1 << (method + 1));
2424 
2425 	/* Reinitialise bus-mastering, which may have been turned off before
2426 	 * the reset was scheduled. This is still appropriate, even in the
2427 	 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
2428 	 * can respond to requests. */
2429 	pci_set_master(efx->pci_dev);
2430 
2431 out:
2432 	/* Leave device stopped if necessary */
2433 	disabled = rc ||
2434 		method == RESET_TYPE_DISABLE ||
2435 		method == RESET_TYPE_RECOVER_OR_DISABLE;
2436 	rc2 = efx_reset_up(efx, method, !disabled);
2437 	if (rc2) {
2438 		disabled = true;
2439 		if (!rc)
2440 			rc = rc2;
2441 	}
2442 
2443 	if (disabled) {
2444 		dev_close(efx->net_dev);
2445 		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2446 		efx->state = STATE_DISABLED;
2447 	} else {
2448 		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2449 		netif_device_attach(efx->net_dev);
2450 	}
2451 	return rc;
2452 }
2453 
2454 /* Try recovery mechanisms.
2455  * For now only EEH is supported.
2456  * Returns 0 if the recovery mechanisms are unsuccessful.
2457  * Returns a non-zero value otherwise.
2458  */
2459 int efx_try_recovery(struct efx_nic *efx)
2460 {
2461 #ifdef CONFIG_EEH
2462 	/* A PCI error can occur and not be seen by EEH because nothing
2463 	 * happens on the PCI bus. In this case the driver may fail and
2464 	 * schedule a 'recover or reset', leading to this recovery handler.
2465 	 * Manually call the eeh failure check function.
2466 	 */
2467 	struct eeh_dev *eehdev =
2468 		of_node_to_eeh_dev(pci_device_to_OF_node(efx->pci_dev));
2469 
2470 	if (eeh_dev_check_failure(eehdev)) {
2471 		/* The EEH mechanisms will handle the error and reset the
2472 		 * device if necessary.
2473 		 */
2474 		return 1;
2475 	}
2476 #endif
2477 	return 0;
2478 }
2479 
2480 static void efx_wait_for_bist_end(struct efx_nic *efx)
2481 {
2482 	int i;
2483 
2484 	for (i = 0; i < BIST_WAIT_DELAY_COUNT; ++i) {
2485 		if (efx_mcdi_poll_reboot(efx))
2486 			goto out;
2487 		msleep(BIST_WAIT_DELAY_MS);
2488 	}
2489 
2490 	netif_err(efx, drv, efx->net_dev, "Warning: No MC reboot after BIST mode\n");
2491 out:
2492 	/* Either way unset the BIST flag. If we found no reboot we probably
2493 	 * won't recover, but we should try.
2494 	 */
2495 	efx->mc_bist_for_other_fn = false;
2496 }
2497 
2498 /* The worker thread exists so that code that cannot sleep can
2499  * schedule a reset for later.
2500  */
2501 static void efx_reset_work(struct work_struct *data)
2502 {
2503 	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2504 	unsigned long pending;
2505 	enum reset_type method;
2506 
2507 	pending = ACCESS_ONCE(efx->reset_pending);
2508 	method = fls(pending) - 1;
2509 
2510 	if (method == RESET_TYPE_MC_BIST)
2511 		efx_wait_for_bist_end(efx);
2512 
2513 	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
2514 	     method == RESET_TYPE_RECOVER_OR_ALL) &&
2515 	    efx_try_recovery(efx))
2516 		return;
2517 
2518 	if (!pending)
2519 		return;
2520 
2521 	rtnl_lock();
2522 
2523 	/* We checked the state in efx_schedule_reset() but it may
2524 	 * have changed by now.  Now that we have the RTNL lock,
2525 	 * it cannot change again.
2526 	 */
2527 	if (efx->state == STATE_READY)
2528 		(void)efx_reset(efx, method);
2529 
2530 	rtnl_unlock();
2531 }
2532 
2533 void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
2534 {
2535 	enum reset_type method;
2536 
2537 	if (efx->state == STATE_RECOVERY) {
2538 		netif_dbg(efx, drv, efx->net_dev,
2539 			  "recovering: skip scheduling %s reset\n",
2540 			  RESET_TYPE(type));
2541 		return;
2542 	}
2543 
2544 	switch (type) {
2545 	case RESET_TYPE_INVISIBLE:
2546 	case RESET_TYPE_ALL:
2547 	case RESET_TYPE_RECOVER_OR_ALL:
2548 	case RESET_TYPE_WORLD:
2549 	case RESET_TYPE_DISABLE:
2550 	case RESET_TYPE_RECOVER_OR_DISABLE:
2551 	case RESET_TYPE_MC_BIST:
2552 		method = type;
2553 		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
2554 			  RESET_TYPE(method));
2555 		break;
2556 	default:
2557 		method = efx->type->map_reset_reason(type);
2558 		netif_dbg(efx, drv, efx->net_dev,
2559 			  "scheduling %s reset for %s\n",
2560 			  RESET_TYPE(method), RESET_TYPE(type));
2561 		break;
2562 	}
2563 
2564 	set_bit(method, &efx->reset_pending);
2565 	smp_mb(); /* ensure we change reset_pending before checking state */
2566 
2567 	/* If we're not READY then just leave the flags set as the cue
2568 	 * to abort probing or reschedule the reset later.
2569 	 */
2570 	if (ACCESS_ONCE(efx->state) != STATE_READY)
2571 		return;
2572 
2573 	/* efx_process_channel() will no longer read events once a
2574 	 * reset is scheduled. So switch back to poll'd MCDI completions. */
2575 	efx_mcdi_mode_poll(efx);
2576 
2577 	queue_work(reset_workqueue, &efx->reset_work);
2578 }
2579 
2580 /**************************************************************************
2581  *
2582  * List of NICs we support
2583  *
2584  **************************************************************************/
2585 
2586 /* PCI device ID table */
2587 static DEFINE_PCI_DEVICE_TABLE(efx_pci_table) = {
2588 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
2589 		    PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0),
2590 	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2591 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
2592 		    PCI_DEVICE_ID_SOLARFLARE_SFC4000B),
2593 	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2594 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2595 	 .driver_data = (unsigned long) &siena_a0_nic_type},
2596 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2597 	 .driver_data = (unsigned long) &siena_a0_nic_type},
2598 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
2599 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2600 	{0}			/* end of list */
2601 };
2602 
2603 /**************************************************************************
2604  *
2605  * Dummy PHY/MAC operations
2606  *
2607  * Can be used for some unimplemented operations
2608  * Needed so all function pointers are valid and do not have to be tested
2609  * before use
2610  *
2611  **************************************************************************/
2612 int efx_port_dummy_op_int(struct efx_nic *efx)
2613 {
2614 	return 0;
2615 }
2616 void efx_port_dummy_op_void(struct efx_nic *efx) {}
2617 
2618 static bool efx_port_dummy_op_poll(struct efx_nic *efx)
2619 {
2620 	return false;
2621 }
2622 
2623 static const struct efx_phy_operations efx_dummy_phy_operations = {
2624 	.init		 = efx_port_dummy_op_int,
2625 	.reconfigure	 = efx_port_dummy_op_int,
2626 	.poll		 = efx_port_dummy_op_poll,
2627 	.fini		 = efx_port_dummy_op_void,
2628 };
2629 
2630 /**************************************************************************
2631  *
2632  * Data housekeeping
2633  *
2634  **************************************************************************/
2635 
2636 /* This zeroes out and then fills in the invariants in a struct
2637  * efx_nic (including all sub-structures).
2638  */
2639 static int efx_init_struct(struct efx_nic *efx,
2640 			   struct pci_dev *pci_dev, struct net_device *net_dev)
2641 {
2642 	int i;
2643 
2644 	/* Initialise common structures */
2645 	INIT_LIST_HEAD(&efx->node);
2646 	INIT_LIST_HEAD(&efx->secondary_list);
2647 	spin_lock_init(&efx->biu_lock);
2648 #ifdef CONFIG_SFC_MTD
2649 	INIT_LIST_HEAD(&efx->mtd_list);
2650 #endif
2651 	INIT_WORK(&efx->reset_work, efx_reset_work);
2652 	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
2653 	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
2654 	efx->pci_dev = pci_dev;
2655 	efx->msg_enable = debug;
2656 	efx->state = STATE_UNINIT;
2657 	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
2658 
2659 	efx->net_dev = net_dev;
2660 	efx->rx_prefix_size = efx->type->rx_prefix_size;
2661 	efx->rx_ip_align =
2662 		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
2663 	efx->rx_packet_hash_offset =
2664 		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
2665 	efx->rx_packet_ts_offset =
2666 		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
2667 	spin_lock_init(&efx->stats_lock);
2668 	mutex_init(&efx->mac_lock);
2669 	efx->phy_op = &efx_dummy_phy_operations;
2670 	efx->mdio.dev = net_dev;
2671 	INIT_WORK(&efx->mac_work, efx_mac_work);
2672 	init_waitqueue_head(&efx->flush_wq);
2673 
2674 	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2675 		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
2676 		if (!efx->channel[i])
2677 			goto fail;
2678 		efx->msi_context[i].efx = efx;
2679 		efx->msi_context[i].index = i;
2680 	}
2681 
2682 	/* Higher numbered interrupt modes are less capable! */
2683 	efx->interrupt_mode = max(efx->type->max_interrupt_mode,
2684 				  interrupt_mode);
2685 
2686 	/* Would be good to use the net_dev name, but we're too early */
2687 	snprintf(efx->workqueue_name, sizeof(efx->workqueue_name), "sfc%s",
2688 		 pci_name(pci_dev));
2689 	efx->workqueue = create_singlethread_workqueue(efx->workqueue_name);
2690 	if (!efx->workqueue)
2691 		goto fail;
2692 
2693 	return 0;
2694 
2695 fail:
2696 	efx_fini_struct(efx);
2697 	return -ENOMEM;
2698 }
2699 
2700 static void efx_fini_struct(struct efx_nic *efx)
2701 {
2702 	int i;
2703 
2704 	for (i = 0; i < EFX_MAX_CHANNELS; i++)
2705 		kfree(efx->channel[i]);
2706 
2707 	kfree(efx->vpd_sn);
2708 
2709 	if (efx->workqueue) {
2710 		destroy_workqueue(efx->workqueue);
2711 		efx->workqueue = NULL;
2712 	}
2713 }
2714 
2715 /**************************************************************************
2716  *
2717  * PCI interface
2718  *
2719  **************************************************************************/
2720 
2721 /* Main body of final NIC shutdown code
2722  * This is called only at module unload (or hotplug removal).
2723  */
2724 static void efx_pci_remove_main(struct efx_nic *efx)
2725 {
2726 	/* Flush reset_work. It can no longer be scheduled since we
2727 	 * are not READY.
2728 	 */
2729 	BUG_ON(efx->state == STATE_READY);
2730 	cancel_work_sync(&efx->reset_work);
2731 
2732 	efx_disable_interrupts(efx);
2733 	efx_nic_fini_interrupt(efx);
2734 	efx_fini_port(efx);
2735 	efx->type->fini(efx);
2736 	efx_fini_napi(efx);
2737 	efx_remove_all(efx);
2738 }
2739 
2740 /* Final NIC shutdown
2741  * This is called only at module unload (or hotplug removal).
2742  */
2743 static void efx_pci_remove(struct pci_dev *pci_dev)
2744 {
2745 	struct efx_nic *efx;
2746 
2747 	efx = pci_get_drvdata(pci_dev);
2748 	if (!efx)
2749 		return;
2750 
2751 	/* Mark the NIC as fini, then stop the interface */
2752 	rtnl_lock();
2753 	efx_dissociate(efx);
2754 	dev_close(efx->net_dev);
2755 	efx_disable_interrupts(efx);
2756 	rtnl_unlock();
2757 
2758 	efx_sriov_fini(efx);
2759 	efx_unregister_netdev(efx);
2760 
2761 	efx_mtd_remove(efx);
2762 
2763 	efx_pci_remove_main(efx);
2764 
2765 	efx_fini_io(efx);
2766 	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
2767 
2768 	efx_fini_struct(efx);
2769 	free_netdev(efx->net_dev);
2770 
2771 	pci_disable_pcie_error_reporting(pci_dev);
2772 };
2773 
2774 /* NIC VPD information
2775  * Called during probe to display the part number of the
2776  * installed NIC.  VPD is potentially very large but this should
2777  * always appear within the first 512 bytes.
2778  */
2779 #define SFC_VPD_LEN 512
2780 static void efx_probe_vpd_strings(struct efx_nic *efx)
2781 {
2782 	struct pci_dev *dev = efx->pci_dev;
2783 	char vpd_data[SFC_VPD_LEN];
2784 	ssize_t vpd_size;
2785 	int ro_start, ro_size, i, j;
2786 
2787 	/* Get the vpd data from the device */
2788 	vpd_size = pci_read_vpd(dev, 0, sizeof(vpd_data), vpd_data);
2789 	if (vpd_size <= 0) {
2790 		netif_err(efx, drv, efx->net_dev, "Unable to read VPD\n");
2791 		return;
2792 	}
2793 
2794 	/* Get the Read only section */
2795 	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size, PCI_VPD_LRDT_RO_DATA);
2796 	if (ro_start < 0) {
2797 		netif_err(efx, drv, efx->net_dev, "VPD Read-only not found\n");
2798 		return;
2799 	}
2800 
2801 	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
2802 	j = ro_size;
2803 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
2804 	if (i + j > vpd_size)
2805 		j = vpd_size - i;
2806 
2807 	/* Get the Part number */
2808 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "PN");
2809 	if (i < 0) {
2810 		netif_err(efx, drv, efx->net_dev, "Part number not found\n");
2811 		return;
2812 	}
2813 
2814 	j = pci_vpd_info_field_size(&vpd_data[i]);
2815 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
2816 	if (i + j > vpd_size) {
2817 		netif_err(efx, drv, efx->net_dev, "Incomplete part number\n");
2818 		return;
2819 	}
2820 
2821 	netif_info(efx, drv, efx->net_dev,
2822 		   "Part Number : %.*s\n", j, &vpd_data[i]);
2823 
2824 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
2825 	j = ro_size;
2826 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "SN");
2827 	if (i < 0) {
2828 		netif_err(efx, drv, efx->net_dev, "Serial number not found\n");
2829 		return;
2830 	}
2831 
2832 	j = pci_vpd_info_field_size(&vpd_data[i]);
2833 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
2834 	if (i + j > vpd_size) {
2835 		netif_err(efx, drv, efx->net_dev, "Incomplete serial number\n");
2836 		return;
2837 	}
2838 
2839 	efx->vpd_sn = kmalloc(j + 1, GFP_KERNEL);
2840 	if (!efx->vpd_sn)
2841 		return;
2842 
2843 	snprintf(efx->vpd_sn, j + 1, "%s", &vpd_data[i]);
2844 }
2845 
2846 
2847 /* Main body of NIC initialisation
2848  * This is called at module load (or hotplug insertion, theoretically).
2849  */
2850 static int efx_pci_probe_main(struct efx_nic *efx)
2851 {
2852 	int rc;
2853 
2854 	/* Do start-of-day initialisation */
2855 	rc = efx_probe_all(efx);
2856 	if (rc)
2857 		goto fail1;
2858 
2859 	efx_init_napi(efx);
2860 
2861 	rc = efx->type->init(efx);
2862 	if (rc) {
2863 		netif_err(efx, probe, efx->net_dev,
2864 			  "failed to initialise NIC\n");
2865 		goto fail3;
2866 	}
2867 
2868 	rc = efx_init_port(efx);
2869 	if (rc) {
2870 		netif_err(efx, probe, efx->net_dev,
2871 			  "failed to initialise port\n");
2872 		goto fail4;
2873 	}
2874 
2875 	rc = efx_nic_init_interrupt(efx);
2876 	if (rc)
2877 		goto fail5;
2878 	rc = efx_enable_interrupts(efx);
2879 	if (rc)
2880 		goto fail6;
2881 
2882 	return 0;
2883 
2884  fail6:
2885 	efx_nic_fini_interrupt(efx);
2886  fail5:
2887 	efx_fini_port(efx);
2888  fail4:
2889 	efx->type->fini(efx);
2890  fail3:
2891 	efx_fini_napi(efx);
2892 	efx_remove_all(efx);
2893  fail1:
2894 	return rc;
2895 }
2896 
2897 /* NIC initialisation
2898  *
2899  * This is called at module load (or hotplug insertion,
2900  * theoretically).  It sets up PCI mappings, resets the NIC,
2901  * sets up and registers the network devices with the kernel and hooks
2902  * the interrupt service routine.  It does not prepare the device for
2903  * transmission; this is left to the first time one of the network
2904  * interfaces is brought up (i.e. efx_net_open).
2905  */
2906 static int efx_pci_probe(struct pci_dev *pci_dev,
2907 			 const struct pci_device_id *entry)
2908 {
2909 	struct net_device *net_dev;
2910 	struct efx_nic *efx;
2911 	int rc;
2912 
2913 	/* Allocate and initialise a struct net_device and struct efx_nic */
2914 	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
2915 				     EFX_MAX_RX_QUEUES);
2916 	if (!net_dev)
2917 		return -ENOMEM;
2918 	efx = netdev_priv(net_dev);
2919 	efx->type = (const struct efx_nic_type *) entry->driver_data;
2920 	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
2921 			      NETIF_F_HIGHDMA | NETIF_F_TSO |
2922 			      NETIF_F_RXCSUM);
2923 	if (efx->type->offload_features & NETIF_F_V6_CSUM)
2924 		net_dev->features |= NETIF_F_TSO6;
2925 	/* Mask for features that also apply to VLAN devices */
2926 	net_dev->vlan_features |= (NETIF_F_ALL_CSUM | NETIF_F_SG |
2927 				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
2928 				   NETIF_F_RXCSUM);
2929 	/* All offloads can be toggled */
2930 	net_dev->hw_features = net_dev->features & ~NETIF_F_HIGHDMA;
2931 	pci_set_drvdata(pci_dev, efx);
2932 	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
2933 	rc = efx_init_struct(efx, pci_dev, net_dev);
2934 	if (rc)
2935 		goto fail1;
2936 
2937 	netif_info(efx, probe, efx->net_dev,
2938 		   "Solarflare NIC detected\n");
2939 
2940 	efx_probe_vpd_strings(efx);
2941 
2942 	/* Set up basic I/O (BAR mappings etc) */
2943 	rc = efx_init_io(efx);
2944 	if (rc)
2945 		goto fail2;
2946 
2947 	rc = efx_pci_probe_main(efx);
2948 	if (rc)
2949 		goto fail3;
2950 
2951 	rc = efx_register_netdev(efx);
2952 	if (rc)
2953 		goto fail4;
2954 
2955 	rc = efx_sriov_init(efx);
2956 	if (rc)
2957 		netif_err(efx, probe, efx->net_dev,
2958 			  "SR-IOV can't be enabled rc %d\n", rc);
2959 
2960 	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
2961 
2962 	/* Try to create MTDs, but allow this to fail */
2963 	rtnl_lock();
2964 	rc = efx_mtd_probe(efx);
2965 	rtnl_unlock();
2966 	if (rc)
2967 		netif_warn(efx, probe, efx->net_dev,
2968 			   "failed to create MTDs (%d)\n", rc);
2969 
2970 	rc = pci_enable_pcie_error_reporting(pci_dev);
2971 	if (rc && rc != -EINVAL)
2972 		netif_warn(efx, probe, efx->net_dev,
2973 			   "pci_enable_pcie_error_reporting failed (%d)\n", rc);
2974 
2975 	return 0;
2976 
2977  fail4:
2978 	efx_pci_remove_main(efx);
2979  fail3:
2980 	efx_fini_io(efx);
2981  fail2:
2982 	efx_fini_struct(efx);
2983  fail1:
2984 	WARN_ON(rc > 0);
2985 	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
2986 	free_netdev(net_dev);
2987 	return rc;
2988 }
2989 
2990 static int efx_pm_freeze(struct device *dev)
2991 {
2992 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2993 
2994 	rtnl_lock();
2995 
2996 	if (efx->state != STATE_DISABLED) {
2997 		efx->state = STATE_UNINIT;
2998 
2999 		efx_device_detach_sync(efx);
3000 
3001 		efx_stop_all(efx);
3002 		efx_disable_interrupts(efx);
3003 	}
3004 
3005 	rtnl_unlock();
3006 
3007 	return 0;
3008 }
3009 
3010 static int efx_pm_thaw(struct device *dev)
3011 {
3012 	int rc;
3013 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3014 
3015 	rtnl_lock();
3016 
3017 	if (efx->state != STATE_DISABLED) {
3018 		rc = efx_enable_interrupts(efx);
3019 		if (rc)
3020 			goto fail;
3021 
3022 		mutex_lock(&efx->mac_lock);
3023 		efx->phy_op->reconfigure(efx);
3024 		mutex_unlock(&efx->mac_lock);
3025 
3026 		efx_start_all(efx);
3027 
3028 		netif_device_attach(efx->net_dev);
3029 
3030 		efx->state = STATE_READY;
3031 
3032 		efx->type->resume_wol(efx);
3033 	}
3034 
3035 	rtnl_unlock();
3036 
3037 	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
3038 	queue_work(reset_workqueue, &efx->reset_work);
3039 
3040 	return 0;
3041 
3042 fail:
3043 	rtnl_unlock();
3044 
3045 	return rc;
3046 }
3047 
3048 static int efx_pm_poweroff(struct device *dev)
3049 {
3050 	struct pci_dev *pci_dev = to_pci_dev(dev);
3051 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3052 
3053 	efx->type->fini(efx);
3054 
3055 	efx->reset_pending = 0;
3056 
3057 	pci_save_state(pci_dev);
3058 	return pci_set_power_state(pci_dev, PCI_D3hot);
3059 }
3060 
3061 /* Used for both resume and restore */
3062 static int efx_pm_resume(struct device *dev)
3063 {
3064 	struct pci_dev *pci_dev = to_pci_dev(dev);
3065 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3066 	int rc;
3067 
3068 	rc = pci_set_power_state(pci_dev, PCI_D0);
3069 	if (rc)
3070 		return rc;
3071 	pci_restore_state(pci_dev);
3072 	rc = pci_enable_device(pci_dev);
3073 	if (rc)
3074 		return rc;
3075 	pci_set_master(efx->pci_dev);
3076 	rc = efx->type->reset(efx, RESET_TYPE_ALL);
3077 	if (rc)
3078 		return rc;
3079 	rc = efx->type->init(efx);
3080 	if (rc)
3081 		return rc;
3082 	rc = efx_pm_thaw(dev);
3083 	return rc;
3084 }
3085 
3086 static int efx_pm_suspend(struct device *dev)
3087 {
3088 	int rc;
3089 
3090 	efx_pm_freeze(dev);
3091 	rc = efx_pm_poweroff(dev);
3092 	if (rc)
3093 		efx_pm_resume(dev);
3094 	return rc;
3095 }
3096 
3097 static const struct dev_pm_ops efx_pm_ops = {
3098 	.suspend	= efx_pm_suspend,
3099 	.resume		= efx_pm_resume,
3100 	.freeze		= efx_pm_freeze,
3101 	.thaw		= efx_pm_thaw,
3102 	.poweroff	= efx_pm_poweroff,
3103 	.restore	= efx_pm_resume,
3104 };
3105 
3106 /* A PCI error affecting this device was detected.
3107  * At this point MMIO and DMA may be disabled.
3108  * Stop the software path and request a slot reset.
3109  */
3110 static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
3111 					      enum pci_channel_state state)
3112 {
3113 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3114 	struct efx_nic *efx = pci_get_drvdata(pdev);
3115 
3116 	if (state == pci_channel_io_perm_failure)
3117 		return PCI_ERS_RESULT_DISCONNECT;
3118 
3119 	rtnl_lock();
3120 
3121 	if (efx->state != STATE_DISABLED) {
3122 		efx->state = STATE_RECOVERY;
3123 		efx->reset_pending = 0;
3124 
3125 		efx_device_detach_sync(efx);
3126 
3127 		efx_stop_all(efx);
3128 		efx_disable_interrupts(efx);
3129 
3130 		status = PCI_ERS_RESULT_NEED_RESET;
3131 	} else {
3132 		/* If the interface is disabled we don't want to do anything
3133 		 * with it.
3134 		 */
3135 		status = PCI_ERS_RESULT_RECOVERED;
3136 	}
3137 
3138 	rtnl_unlock();
3139 
3140 	pci_disable_device(pdev);
3141 
3142 	return status;
3143 }
3144 
3145 /* Fake a successfull reset, which will be performed later in efx_io_resume. */
3146 static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
3147 {
3148 	struct efx_nic *efx = pci_get_drvdata(pdev);
3149 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3150 	int rc;
3151 
3152 	if (pci_enable_device(pdev)) {
3153 		netif_err(efx, hw, efx->net_dev,
3154 			  "Cannot re-enable PCI device after reset.\n");
3155 		status =  PCI_ERS_RESULT_DISCONNECT;
3156 	}
3157 
3158 	rc = pci_cleanup_aer_uncorrect_error_status(pdev);
3159 	if (rc) {
3160 		netif_err(efx, hw, efx->net_dev,
3161 		"pci_cleanup_aer_uncorrect_error_status failed (%d)\n", rc);
3162 		/* Non-fatal error. Continue. */
3163 	}
3164 
3165 	return status;
3166 }
3167 
3168 /* Perform the actual reset and resume I/O operations. */
3169 static void efx_io_resume(struct pci_dev *pdev)
3170 {
3171 	struct efx_nic *efx = pci_get_drvdata(pdev);
3172 	int rc;
3173 
3174 	rtnl_lock();
3175 
3176 	if (efx->state == STATE_DISABLED)
3177 		goto out;
3178 
3179 	rc = efx_reset(efx, RESET_TYPE_ALL);
3180 	if (rc) {
3181 		netif_err(efx, hw, efx->net_dev,
3182 			  "efx_reset failed after PCI error (%d)\n", rc);
3183 	} else {
3184 		efx->state = STATE_READY;
3185 		netif_dbg(efx, hw, efx->net_dev,
3186 			  "Done resetting and resuming IO after PCI error.\n");
3187 	}
3188 
3189 out:
3190 	rtnl_unlock();
3191 }
3192 
3193 /* For simplicity and reliability, we always require a slot reset and try to
3194  * reset the hardware when a pci error affecting the device is detected.
3195  * We leave both the link_reset and mmio_enabled callback unimplemented:
3196  * with our request for slot reset the mmio_enabled callback will never be
3197  * called, and the link_reset callback is not used by AER or EEH mechanisms.
3198  */
3199 static struct pci_error_handlers efx_err_handlers = {
3200 	.error_detected = efx_io_error_detected,
3201 	.slot_reset	= efx_io_slot_reset,
3202 	.resume		= efx_io_resume,
3203 };
3204 
3205 static struct pci_driver efx_pci_driver = {
3206 	.name		= KBUILD_MODNAME,
3207 	.id_table	= efx_pci_table,
3208 	.probe		= efx_pci_probe,
3209 	.remove		= efx_pci_remove,
3210 	.driver.pm	= &efx_pm_ops,
3211 	.err_handler	= &efx_err_handlers,
3212 };
3213 
3214 /**************************************************************************
3215  *
3216  * Kernel module interface
3217  *
3218  *************************************************************************/
3219 
3220 module_param(interrupt_mode, uint, 0444);
3221 MODULE_PARM_DESC(interrupt_mode,
3222 		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
3223 
3224 static int __init efx_init_module(void)
3225 {
3226 	int rc;
3227 
3228 	printk(KERN_INFO "Solarflare NET driver v" EFX_DRIVER_VERSION "\n");
3229 
3230 	rc = register_netdevice_notifier(&efx_netdev_notifier);
3231 	if (rc)
3232 		goto err_notifier;
3233 
3234 	rc = efx_init_sriov();
3235 	if (rc)
3236 		goto err_sriov;
3237 
3238 	reset_workqueue = create_singlethread_workqueue("sfc_reset");
3239 	if (!reset_workqueue) {
3240 		rc = -ENOMEM;
3241 		goto err_reset;
3242 	}
3243 
3244 	rc = pci_register_driver(&efx_pci_driver);
3245 	if (rc < 0)
3246 		goto err_pci;
3247 
3248 	return 0;
3249 
3250  err_pci:
3251 	destroy_workqueue(reset_workqueue);
3252  err_reset:
3253 	efx_fini_sriov();
3254  err_sriov:
3255 	unregister_netdevice_notifier(&efx_netdev_notifier);
3256  err_notifier:
3257 	return rc;
3258 }
3259 
3260 static void __exit efx_exit_module(void)
3261 {
3262 	printk(KERN_INFO "Solarflare NET driver unloading\n");
3263 
3264 	pci_unregister_driver(&efx_pci_driver);
3265 	destroy_workqueue(reset_workqueue);
3266 	efx_fini_sriov();
3267 	unregister_netdevice_notifier(&efx_netdev_notifier);
3268 
3269 }
3270 
3271 module_init(efx_init_module);
3272 module_exit(efx_exit_module);
3273 
3274 MODULE_AUTHOR("Solarflare Communications and "
3275 	      "Michael Brown <mbrown@fensystems.co.uk>");
3276 MODULE_DESCRIPTION("Solarflare Communications network driver");
3277 MODULE_LICENSE("GPL");
3278 MODULE_DEVICE_TABLE(pci, efx_pci_table);
3279