xref: /openbmc/linux/drivers/net/ethernet/sfc/efx.c (revision 232b0b08)
1 /****************************************************************************
2  * Driver for Solarflare network controllers and boards
3  * Copyright 2005-2006 Fen Systems Ltd.
4  * Copyright 2005-2013 Solarflare Communications Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License version 2 as published
8  * by the Free Software Foundation, incorporated herein by reference.
9  */
10 
11 #include <linux/module.h>
12 #include <linux/pci.h>
13 #include <linux/netdevice.h>
14 #include <linux/etherdevice.h>
15 #include <linux/delay.h>
16 #include <linux/notifier.h>
17 #include <linux/ip.h>
18 #include <linux/tcp.h>
19 #include <linux/in.h>
20 #include <linux/ethtool.h>
21 #include <linux/topology.h>
22 #include <linux/gfp.h>
23 #include <linux/aer.h>
24 #include <linux/interrupt.h>
25 #include "net_driver.h"
26 #include <net/gre.h>
27 #include <net/udp_tunnel.h>
28 #include "efx.h"
29 #include "nic.h"
30 #include "selftest.h"
31 #include "sriov.h"
32 
33 #include "mcdi.h"
34 #include "mcdi_pcol.h"
35 #include "workarounds.h"
36 
37 /**************************************************************************
38  *
39  * Type name strings
40  *
41  **************************************************************************
42  */
43 
44 /* Loopback mode names (see LOOPBACK_MODE()) */
45 const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
46 const char *const efx_loopback_mode_names[] = {
47 	[LOOPBACK_NONE]		= "NONE",
48 	[LOOPBACK_DATA]		= "DATAPATH",
49 	[LOOPBACK_GMAC]		= "GMAC",
50 	[LOOPBACK_XGMII]	= "XGMII",
51 	[LOOPBACK_XGXS]		= "XGXS",
52 	[LOOPBACK_XAUI]		= "XAUI",
53 	[LOOPBACK_GMII]		= "GMII",
54 	[LOOPBACK_SGMII]	= "SGMII",
55 	[LOOPBACK_XGBR]		= "XGBR",
56 	[LOOPBACK_XFI]		= "XFI",
57 	[LOOPBACK_XAUI_FAR]	= "XAUI_FAR",
58 	[LOOPBACK_GMII_FAR]	= "GMII_FAR",
59 	[LOOPBACK_SGMII_FAR]	= "SGMII_FAR",
60 	[LOOPBACK_XFI_FAR]	= "XFI_FAR",
61 	[LOOPBACK_GPHY]		= "GPHY",
62 	[LOOPBACK_PHYXS]	= "PHYXS",
63 	[LOOPBACK_PCS]		= "PCS",
64 	[LOOPBACK_PMAPMD]	= "PMA/PMD",
65 	[LOOPBACK_XPORT]	= "XPORT",
66 	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
67 	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
68 	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
69 	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
70 	[LOOPBACK_GMII_WS]	= "GMII_WS",
71 	[LOOPBACK_XFI_WS]	= "XFI_WS",
72 	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
73 	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
74 };
75 
76 const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
77 const char *const efx_reset_type_names[] = {
78 	[RESET_TYPE_INVISIBLE]          = "INVISIBLE",
79 	[RESET_TYPE_ALL]                = "ALL",
80 	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
81 	[RESET_TYPE_WORLD]              = "WORLD",
82 	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
83 	[RESET_TYPE_DATAPATH]           = "DATAPATH",
84 	[RESET_TYPE_MC_BIST]		= "MC_BIST",
85 	[RESET_TYPE_DISABLE]            = "DISABLE",
86 	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
87 	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
88 	[RESET_TYPE_DMA_ERROR]          = "DMA_ERROR",
89 	[RESET_TYPE_TX_SKIP]            = "TX_SKIP",
90 	[RESET_TYPE_MC_FAILURE]         = "MC_FAILURE",
91 	[RESET_TYPE_MCDI_TIMEOUT]	= "MCDI_TIMEOUT (FLR)",
92 };
93 
94 /* UDP tunnel type names */
95 static const char *const efx_udp_tunnel_type_names[] = {
96 	[TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN] = "vxlan",
97 	[TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE] = "geneve",
98 };
99 
100 void efx_get_udp_tunnel_type_name(u16 type, char *buf, size_t buflen)
101 {
102 	if (type < ARRAY_SIZE(efx_udp_tunnel_type_names) &&
103 	    efx_udp_tunnel_type_names[type] != NULL)
104 		snprintf(buf, buflen, "%s", efx_udp_tunnel_type_names[type]);
105 	else
106 		snprintf(buf, buflen, "type %d", type);
107 }
108 
109 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
110  * queued onto this work queue. This is not a per-nic work queue, because
111  * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
112  */
113 static struct workqueue_struct *reset_workqueue;
114 
115 /* How often and how many times to poll for a reset while waiting for a
116  * BIST that another function started to complete.
117  */
118 #define BIST_WAIT_DELAY_MS	100
119 #define BIST_WAIT_DELAY_COUNT	100
120 
121 /**************************************************************************
122  *
123  * Configurable values
124  *
125  *************************************************************************/
126 
127 /*
128  * Use separate channels for TX and RX events
129  *
130  * Set this to 1 to use separate channels for TX and RX. It allows us
131  * to control interrupt affinity separately for TX and RX.
132  *
133  * This is only used in MSI-X interrupt mode
134  */
135 bool efx_separate_tx_channels;
136 module_param(efx_separate_tx_channels, bool, 0444);
137 MODULE_PARM_DESC(efx_separate_tx_channels,
138 		 "Use separate channels for TX and RX");
139 
140 /* This is the weight assigned to each of the (per-channel) virtual
141  * NAPI devices.
142  */
143 static int napi_weight = 64;
144 
145 /* This is the time (in jiffies) between invocations of the hardware
146  * monitor.
147  * On Falcon-based NICs, this will:
148  * - Check the on-board hardware monitor;
149  * - Poll the link state and reconfigure the hardware as necessary.
150  * On Siena-based NICs for power systems with EEH support, this will give EEH a
151  * chance to start.
152  */
153 static unsigned int efx_monitor_interval = 1 * HZ;
154 
155 /* Initial interrupt moderation settings.  They can be modified after
156  * module load with ethtool.
157  *
158  * The default for RX should strike a balance between increasing the
159  * round-trip latency and reducing overhead.
160  */
161 static unsigned int rx_irq_mod_usec = 60;
162 
163 /* Initial interrupt moderation settings.  They can be modified after
164  * module load with ethtool.
165  *
166  * This default is chosen to ensure that a 10G link does not go idle
167  * while a TX queue is stopped after it has become full.  A queue is
168  * restarted when it drops below half full.  The time this takes (assuming
169  * worst case 3 descriptors per packet and 1024 descriptors) is
170  *   512 / 3 * 1.2 = 205 usec.
171  */
172 static unsigned int tx_irq_mod_usec = 150;
173 
174 /* This is the first interrupt mode to try out of:
175  * 0 => MSI-X
176  * 1 => MSI
177  * 2 => legacy
178  */
179 static unsigned int interrupt_mode;
180 
181 /* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
182  * i.e. the number of CPUs among which we may distribute simultaneous
183  * interrupt handling.
184  *
185  * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
186  * The default (0) means to assign an interrupt to each core.
187  */
188 static unsigned int rss_cpus;
189 module_param(rss_cpus, uint, 0444);
190 MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");
191 
192 static bool phy_flash_cfg;
193 module_param(phy_flash_cfg, bool, 0644);
194 MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");
195 
196 static unsigned irq_adapt_low_thresh = 8000;
197 module_param(irq_adapt_low_thresh, uint, 0644);
198 MODULE_PARM_DESC(irq_adapt_low_thresh,
199 		 "Threshold score for reducing IRQ moderation");
200 
201 static unsigned irq_adapt_high_thresh = 16000;
202 module_param(irq_adapt_high_thresh, uint, 0644);
203 MODULE_PARM_DESC(irq_adapt_high_thresh,
204 		 "Threshold score for increasing IRQ moderation");
205 
206 static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
207 			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
208 			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
209 			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
210 module_param(debug, uint, 0);
211 MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
212 
213 /**************************************************************************
214  *
215  * Utility functions and prototypes
216  *
217  *************************************************************************/
218 
219 static int efx_soft_enable_interrupts(struct efx_nic *efx);
220 static void efx_soft_disable_interrupts(struct efx_nic *efx);
221 static void efx_remove_channel(struct efx_channel *channel);
222 static void efx_remove_channels(struct efx_nic *efx);
223 static const struct efx_channel_type efx_default_channel_type;
224 static void efx_remove_port(struct efx_nic *efx);
225 static void efx_init_napi_channel(struct efx_channel *channel);
226 static void efx_fini_napi(struct efx_nic *efx);
227 static void efx_fini_napi_channel(struct efx_channel *channel);
228 static void efx_fini_struct(struct efx_nic *efx);
229 static void efx_start_all(struct efx_nic *efx);
230 static void efx_stop_all(struct efx_nic *efx);
231 
232 #define EFX_ASSERT_RESET_SERIALISED(efx)		\
233 	do {						\
234 		if ((efx->state == STATE_READY) ||	\
235 		    (efx->state == STATE_RECOVERY) ||	\
236 		    (efx->state == STATE_DISABLED))	\
237 			ASSERT_RTNL();			\
238 	} while (0)
239 
240 static int efx_check_disabled(struct efx_nic *efx)
241 {
242 	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
243 		netif_err(efx, drv, efx->net_dev,
244 			  "device is disabled due to earlier errors\n");
245 		return -EIO;
246 	}
247 	return 0;
248 }
249 
250 /**************************************************************************
251  *
252  * Event queue processing
253  *
254  *************************************************************************/
255 
256 /* Process channel's event queue
257  *
258  * This function is responsible for processing the event queue of a
259  * single channel.  The caller must guarantee that this function will
260  * never be concurrently called more than once on the same channel,
261  * though different channels may be being processed concurrently.
262  */
263 static int efx_process_channel(struct efx_channel *channel, int budget)
264 {
265 	struct efx_tx_queue *tx_queue;
266 	int spent;
267 
268 	if (unlikely(!channel->enabled))
269 		return 0;
270 
271 	efx_for_each_channel_tx_queue(tx_queue, channel) {
272 		tx_queue->pkts_compl = 0;
273 		tx_queue->bytes_compl = 0;
274 	}
275 
276 	spent = efx_nic_process_eventq(channel, budget);
277 	if (spent && efx_channel_has_rx_queue(channel)) {
278 		struct efx_rx_queue *rx_queue =
279 			efx_channel_get_rx_queue(channel);
280 
281 		efx_rx_flush_packet(channel);
282 		efx_fast_push_rx_descriptors(rx_queue, true);
283 	}
284 
285 	/* Update BQL */
286 	efx_for_each_channel_tx_queue(tx_queue, channel) {
287 		if (tx_queue->bytes_compl) {
288 			netdev_tx_completed_queue(tx_queue->core_txq,
289 				tx_queue->pkts_compl, tx_queue->bytes_compl);
290 		}
291 	}
292 
293 	return spent;
294 }
295 
296 /* NAPI poll handler
297  *
298  * NAPI guarantees serialisation of polls of the same device, which
299  * provides the guarantee required by efx_process_channel().
300  */
301 static void efx_update_irq_mod(struct efx_nic *efx, struct efx_channel *channel)
302 {
303 	int step = efx->irq_mod_step_us;
304 
305 	if (channel->irq_mod_score < irq_adapt_low_thresh) {
306 		if (channel->irq_moderation_us > step) {
307 			channel->irq_moderation_us -= step;
308 			efx->type->push_irq_moderation(channel);
309 		}
310 	} else if (channel->irq_mod_score > irq_adapt_high_thresh) {
311 		if (channel->irq_moderation_us <
312 		    efx->irq_rx_moderation_us) {
313 			channel->irq_moderation_us += step;
314 			efx->type->push_irq_moderation(channel);
315 		}
316 	}
317 
318 	channel->irq_count = 0;
319 	channel->irq_mod_score = 0;
320 }
321 
322 static int efx_poll(struct napi_struct *napi, int budget)
323 {
324 	struct efx_channel *channel =
325 		container_of(napi, struct efx_channel, napi_str);
326 	struct efx_nic *efx = channel->efx;
327 	int spent;
328 
329 	netif_vdbg(efx, intr, efx->net_dev,
330 		   "channel %d NAPI poll executing on CPU %d\n",
331 		   channel->channel, raw_smp_processor_id());
332 
333 	spent = efx_process_channel(channel, budget);
334 
335 	if (spent < budget) {
336 		if (efx_channel_has_rx_queue(channel) &&
337 		    efx->irq_rx_adaptive &&
338 		    unlikely(++channel->irq_count == 1000)) {
339 			efx_update_irq_mod(efx, channel);
340 		}
341 
342 		efx_filter_rfs_expire(channel);
343 
344 		/* There is no race here; although napi_disable() will
345 		 * only wait for napi_complete(), this isn't a problem
346 		 * since efx_nic_eventq_read_ack() will have no effect if
347 		 * interrupts have already been disabled.
348 		 */
349 		if (napi_complete_done(napi, spent))
350 			efx_nic_eventq_read_ack(channel);
351 	}
352 
353 	return spent;
354 }
355 
356 /* Create event queue
357  * Event queue memory allocations are done only once.  If the channel
358  * is reset, the memory buffer will be reused; this guards against
359  * errors during channel reset and also simplifies interrupt handling.
360  */
361 static int efx_probe_eventq(struct efx_channel *channel)
362 {
363 	struct efx_nic *efx = channel->efx;
364 	unsigned long entries;
365 
366 	netif_dbg(efx, probe, efx->net_dev,
367 		  "chan %d create event queue\n", channel->channel);
368 
369 	/* Build an event queue with room for one event per tx and rx buffer,
370 	 * plus some extra for link state events and MCDI completions. */
371 	entries = roundup_pow_of_two(efx->rxq_entries + efx->txq_entries + 128);
372 	EFX_WARN_ON_PARANOID(entries > EFX_MAX_EVQ_SIZE);
373 	channel->eventq_mask = max(entries, EFX_MIN_EVQ_SIZE) - 1;
374 
375 	return efx_nic_probe_eventq(channel);
376 }
377 
378 /* Prepare channel's event queue */
379 static int efx_init_eventq(struct efx_channel *channel)
380 {
381 	struct efx_nic *efx = channel->efx;
382 	int rc;
383 
384 	EFX_WARN_ON_PARANOID(channel->eventq_init);
385 
386 	netif_dbg(efx, drv, efx->net_dev,
387 		  "chan %d init event queue\n", channel->channel);
388 
389 	rc = efx_nic_init_eventq(channel);
390 	if (rc == 0) {
391 		efx->type->push_irq_moderation(channel);
392 		channel->eventq_read_ptr = 0;
393 		channel->eventq_init = true;
394 	}
395 	return rc;
396 }
397 
398 /* Enable event queue processing and NAPI */
399 void efx_start_eventq(struct efx_channel *channel)
400 {
401 	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
402 		  "chan %d start event queue\n", channel->channel);
403 
404 	/* Make sure the NAPI handler sees the enabled flag set */
405 	channel->enabled = true;
406 	smp_wmb();
407 
408 	napi_enable(&channel->napi_str);
409 	efx_nic_eventq_read_ack(channel);
410 }
411 
412 /* Disable event queue processing and NAPI */
413 void efx_stop_eventq(struct efx_channel *channel)
414 {
415 	if (!channel->enabled)
416 		return;
417 
418 	napi_disable(&channel->napi_str);
419 	channel->enabled = false;
420 }
421 
422 static void efx_fini_eventq(struct efx_channel *channel)
423 {
424 	if (!channel->eventq_init)
425 		return;
426 
427 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
428 		  "chan %d fini event queue\n", channel->channel);
429 
430 	efx_nic_fini_eventq(channel);
431 	channel->eventq_init = false;
432 }
433 
434 static void efx_remove_eventq(struct efx_channel *channel)
435 {
436 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
437 		  "chan %d remove event queue\n", channel->channel);
438 
439 	efx_nic_remove_eventq(channel);
440 }
441 
442 /**************************************************************************
443  *
444  * Channel handling
445  *
446  *************************************************************************/
447 
448 /* Allocate and initialise a channel structure. */
449 static struct efx_channel *
450 efx_alloc_channel(struct efx_nic *efx, int i, struct efx_channel *old_channel)
451 {
452 	struct efx_channel *channel;
453 	struct efx_rx_queue *rx_queue;
454 	struct efx_tx_queue *tx_queue;
455 	int j;
456 
457 	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
458 	if (!channel)
459 		return NULL;
460 
461 	channel->efx = efx;
462 	channel->channel = i;
463 	channel->type = &efx_default_channel_type;
464 
465 	for (j = 0; j < EFX_TXQ_TYPES; j++) {
466 		tx_queue = &channel->tx_queue[j];
467 		tx_queue->efx = efx;
468 		tx_queue->queue = i * EFX_TXQ_TYPES + j;
469 		tx_queue->channel = channel;
470 	}
471 
472 	rx_queue = &channel->rx_queue;
473 	rx_queue->efx = efx;
474 	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
475 		    (unsigned long)rx_queue);
476 
477 	return channel;
478 }
479 
480 /* Allocate and initialise a channel structure, copying parameters
481  * (but not resources) from an old channel structure.
482  */
483 static struct efx_channel *
484 efx_copy_channel(const struct efx_channel *old_channel)
485 {
486 	struct efx_channel *channel;
487 	struct efx_rx_queue *rx_queue;
488 	struct efx_tx_queue *tx_queue;
489 	int j;
490 
491 	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
492 	if (!channel)
493 		return NULL;
494 
495 	*channel = *old_channel;
496 
497 	channel->napi_dev = NULL;
498 	INIT_HLIST_NODE(&channel->napi_str.napi_hash_node);
499 	channel->napi_str.napi_id = 0;
500 	channel->napi_str.state = 0;
501 	memset(&channel->eventq, 0, sizeof(channel->eventq));
502 
503 	for (j = 0; j < EFX_TXQ_TYPES; j++) {
504 		tx_queue = &channel->tx_queue[j];
505 		if (tx_queue->channel)
506 			tx_queue->channel = channel;
507 		tx_queue->buffer = NULL;
508 		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
509 	}
510 
511 	rx_queue = &channel->rx_queue;
512 	rx_queue->buffer = NULL;
513 	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
514 	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
515 		    (unsigned long)rx_queue);
516 
517 	return channel;
518 }
519 
520 static int efx_probe_channel(struct efx_channel *channel)
521 {
522 	struct efx_tx_queue *tx_queue;
523 	struct efx_rx_queue *rx_queue;
524 	int rc;
525 
526 	netif_dbg(channel->efx, probe, channel->efx->net_dev,
527 		  "creating channel %d\n", channel->channel);
528 
529 	rc = channel->type->pre_probe(channel);
530 	if (rc)
531 		goto fail;
532 
533 	rc = efx_probe_eventq(channel);
534 	if (rc)
535 		goto fail;
536 
537 	efx_for_each_channel_tx_queue(tx_queue, channel) {
538 		rc = efx_probe_tx_queue(tx_queue);
539 		if (rc)
540 			goto fail;
541 	}
542 
543 	efx_for_each_channel_rx_queue(rx_queue, channel) {
544 		rc = efx_probe_rx_queue(rx_queue);
545 		if (rc)
546 			goto fail;
547 	}
548 
549 	return 0;
550 
551 fail:
552 	efx_remove_channel(channel);
553 	return rc;
554 }
555 
556 static void
557 efx_get_channel_name(struct efx_channel *channel, char *buf, size_t len)
558 {
559 	struct efx_nic *efx = channel->efx;
560 	const char *type;
561 	int number;
562 
563 	number = channel->channel;
564 	if (efx->tx_channel_offset == 0) {
565 		type = "";
566 	} else if (channel->channel < efx->tx_channel_offset) {
567 		type = "-rx";
568 	} else {
569 		type = "-tx";
570 		number -= efx->tx_channel_offset;
571 	}
572 	snprintf(buf, len, "%s%s-%d", efx->name, type, number);
573 }
574 
575 static void efx_set_channel_names(struct efx_nic *efx)
576 {
577 	struct efx_channel *channel;
578 
579 	efx_for_each_channel(channel, efx)
580 		channel->type->get_name(channel,
581 					efx->msi_context[channel->channel].name,
582 					sizeof(efx->msi_context[0].name));
583 }
584 
585 static int efx_probe_channels(struct efx_nic *efx)
586 {
587 	struct efx_channel *channel;
588 	int rc;
589 
590 	/* Restart special buffer allocation */
591 	efx->next_buffer_table = 0;
592 
593 	/* Probe channels in reverse, so that any 'extra' channels
594 	 * use the start of the buffer table. This allows the traffic
595 	 * channels to be resized without moving them or wasting the
596 	 * entries before them.
597 	 */
598 	efx_for_each_channel_rev(channel, efx) {
599 		rc = efx_probe_channel(channel);
600 		if (rc) {
601 			netif_err(efx, probe, efx->net_dev,
602 				  "failed to create channel %d\n",
603 				  channel->channel);
604 			goto fail;
605 		}
606 	}
607 	efx_set_channel_names(efx);
608 
609 	return 0;
610 
611 fail:
612 	efx_remove_channels(efx);
613 	return rc;
614 }
615 
616 /* Channels are shutdown and reinitialised whilst the NIC is running
617  * to propagate configuration changes (mtu, checksum offload), or
618  * to clear hardware error conditions
619  */
620 static void efx_start_datapath(struct efx_nic *efx)
621 {
622 	netdev_features_t old_features = efx->net_dev->features;
623 	bool old_rx_scatter = efx->rx_scatter;
624 	struct efx_tx_queue *tx_queue;
625 	struct efx_rx_queue *rx_queue;
626 	struct efx_channel *channel;
627 	size_t rx_buf_len;
628 
629 	/* Calculate the rx buffer allocation parameters required to
630 	 * support the current MTU, including padding for header
631 	 * alignment and overruns.
632 	 */
633 	efx->rx_dma_len = (efx->rx_prefix_size +
634 			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
635 			   efx->type->rx_buffer_padding);
636 	rx_buf_len = (sizeof(struct efx_rx_page_state) +
637 		      efx->rx_ip_align + efx->rx_dma_len);
638 	if (rx_buf_len <= PAGE_SIZE) {
639 		efx->rx_scatter = efx->type->always_rx_scatter;
640 		efx->rx_buffer_order = 0;
641 	} else if (efx->type->can_rx_scatter) {
642 		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
643 		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
644 			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
645 				       EFX_RX_BUF_ALIGNMENT) >
646 			     PAGE_SIZE);
647 		efx->rx_scatter = true;
648 		efx->rx_dma_len = EFX_RX_USR_BUF_SIZE;
649 		efx->rx_buffer_order = 0;
650 	} else {
651 		efx->rx_scatter = false;
652 		efx->rx_buffer_order = get_order(rx_buf_len);
653 	}
654 
655 	efx_rx_config_page_split(efx);
656 	if (efx->rx_buffer_order)
657 		netif_dbg(efx, drv, efx->net_dev,
658 			  "RX buf len=%u; page order=%u batch=%u\n",
659 			  efx->rx_dma_len, efx->rx_buffer_order,
660 			  efx->rx_pages_per_batch);
661 	else
662 		netif_dbg(efx, drv, efx->net_dev,
663 			  "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
664 			  efx->rx_dma_len, efx->rx_page_buf_step,
665 			  efx->rx_bufs_per_page, efx->rx_pages_per_batch);
666 
667 	/* Restore previously fixed features in hw_features and remove
668 	 * features which are fixed now
669 	 */
670 	efx->net_dev->hw_features |= efx->net_dev->features;
671 	efx->net_dev->hw_features &= ~efx->fixed_features;
672 	efx->net_dev->features |= efx->fixed_features;
673 	if (efx->net_dev->features != old_features)
674 		netdev_features_change(efx->net_dev);
675 
676 	/* RX filters may also have scatter-enabled flags */
677 	if (efx->rx_scatter != old_rx_scatter)
678 		efx->type->filter_update_rx_scatter(efx);
679 
680 	/* We must keep at least one descriptor in a TX ring empty.
681 	 * We could avoid this when the queue size does not exactly
682 	 * match the hardware ring size, but it's not that important.
683 	 * Therefore we stop the queue when one more skb might fill
684 	 * the ring completely.  We wake it when half way back to
685 	 * empty.
686 	 */
687 	efx->txq_stop_thresh = efx->txq_entries - efx_tx_max_skb_descs(efx);
688 	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;
689 
690 	/* Initialise the channels */
691 	efx_for_each_channel(channel, efx) {
692 		efx_for_each_channel_tx_queue(tx_queue, channel) {
693 			efx_init_tx_queue(tx_queue);
694 			atomic_inc(&efx->active_queues);
695 		}
696 
697 		efx_for_each_channel_rx_queue(rx_queue, channel) {
698 			efx_init_rx_queue(rx_queue);
699 			atomic_inc(&efx->active_queues);
700 			efx_stop_eventq(channel);
701 			efx_fast_push_rx_descriptors(rx_queue, false);
702 			efx_start_eventq(channel);
703 		}
704 
705 		WARN_ON(channel->rx_pkt_n_frags);
706 	}
707 
708 	efx_ptp_start_datapath(efx);
709 
710 	if (netif_device_present(efx->net_dev))
711 		netif_tx_wake_all_queues(efx->net_dev);
712 }
713 
714 static void efx_stop_datapath(struct efx_nic *efx)
715 {
716 	struct efx_channel *channel;
717 	struct efx_tx_queue *tx_queue;
718 	struct efx_rx_queue *rx_queue;
719 	int rc;
720 
721 	EFX_ASSERT_RESET_SERIALISED(efx);
722 	BUG_ON(efx->port_enabled);
723 
724 	efx_ptp_stop_datapath(efx);
725 
726 	/* Stop RX refill */
727 	efx_for_each_channel(channel, efx) {
728 		efx_for_each_channel_rx_queue(rx_queue, channel)
729 			rx_queue->refill_enabled = false;
730 	}
731 
732 	efx_for_each_channel(channel, efx) {
733 		/* RX packet processing is pipelined, so wait for the
734 		 * NAPI handler to complete.  At least event queue 0
735 		 * might be kept active by non-data events, so don't
736 		 * use napi_synchronize() but actually disable NAPI
737 		 * temporarily.
738 		 */
739 		if (efx_channel_has_rx_queue(channel)) {
740 			efx_stop_eventq(channel);
741 			efx_start_eventq(channel);
742 		}
743 	}
744 
745 	rc = efx->type->fini_dmaq(efx);
746 	if (rc) {
747 		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
748 	} else {
749 		netif_dbg(efx, drv, efx->net_dev,
750 			  "successfully flushed all queues\n");
751 	}
752 
753 	efx_for_each_channel(channel, efx) {
754 		efx_for_each_channel_rx_queue(rx_queue, channel)
755 			efx_fini_rx_queue(rx_queue);
756 		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
757 			efx_fini_tx_queue(tx_queue);
758 	}
759 }
760 
761 static void efx_remove_channel(struct efx_channel *channel)
762 {
763 	struct efx_tx_queue *tx_queue;
764 	struct efx_rx_queue *rx_queue;
765 
766 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
767 		  "destroy chan %d\n", channel->channel);
768 
769 	efx_for_each_channel_rx_queue(rx_queue, channel)
770 		efx_remove_rx_queue(rx_queue);
771 	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
772 		efx_remove_tx_queue(tx_queue);
773 	efx_remove_eventq(channel);
774 	channel->type->post_remove(channel);
775 }
776 
777 static void efx_remove_channels(struct efx_nic *efx)
778 {
779 	struct efx_channel *channel;
780 
781 	efx_for_each_channel(channel, efx)
782 		efx_remove_channel(channel);
783 }
784 
785 int
786 efx_realloc_channels(struct efx_nic *efx, u32 rxq_entries, u32 txq_entries)
787 {
788 	struct efx_channel *other_channel[EFX_MAX_CHANNELS], *channel;
789 	u32 old_rxq_entries, old_txq_entries;
790 	unsigned i, next_buffer_table = 0;
791 	int rc, rc2;
792 
793 	rc = efx_check_disabled(efx);
794 	if (rc)
795 		return rc;
796 
797 	/* Not all channels should be reallocated. We must avoid
798 	 * reallocating their buffer table entries.
799 	 */
800 	efx_for_each_channel(channel, efx) {
801 		struct efx_rx_queue *rx_queue;
802 		struct efx_tx_queue *tx_queue;
803 
804 		if (channel->type->copy)
805 			continue;
806 		next_buffer_table = max(next_buffer_table,
807 					channel->eventq.index +
808 					channel->eventq.entries);
809 		efx_for_each_channel_rx_queue(rx_queue, channel)
810 			next_buffer_table = max(next_buffer_table,
811 						rx_queue->rxd.index +
812 						rx_queue->rxd.entries);
813 		efx_for_each_channel_tx_queue(tx_queue, channel)
814 			next_buffer_table = max(next_buffer_table,
815 						tx_queue->txd.index +
816 						tx_queue->txd.entries);
817 	}
818 
819 	efx_device_detach_sync(efx);
820 	efx_stop_all(efx);
821 	efx_soft_disable_interrupts(efx);
822 
823 	/* Clone channels (where possible) */
824 	memset(other_channel, 0, sizeof(other_channel));
825 	for (i = 0; i < efx->n_channels; i++) {
826 		channel = efx->channel[i];
827 		if (channel->type->copy)
828 			channel = channel->type->copy(channel);
829 		if (!channel) {
830 			rc = -ENOMEM;
831 			goto out;
832 		}
833 		other_channel[i] = channel;
834 	}
835 
836 	/* Swap entry counts and channel pointers */
837 	old_rxq_entries = efx->rxq_entries;
838 	old_txq_entries = efx->txq_entries;
839 	efx->rxq_entries = rxq_entries;
840 	efx->txq_entries = txq_entries;
841 	for (i = 0; i < efx->n_channels; i++) {
842 		channel = efx->channel[i];
843 		efx->channel[i] = other_channel[i];
844 		other_channel[i] = channel;
845 	}
846 
847 	/* Restart buffer table allocation */
848 	efx->next_buffer_table = next_buffer_table;
849 
850 	for (i = 0; i < efx->n_channels; i++) {
851 		channel = efx->channel[i];
852 		if (!channel->type->copy)
853 			continue;
854 		rc = efx_probe_channel(channel);
855 		if (rc)
856 			goto rollback;
857 		efx_init_napi_channel(efx->channel[i]);
858 	}
859 
860 out:
861 	/* Destroy unused channel structures */
862 	for (i = 0; i < efx->n_channels; i++) {
863 		channel = other_channel[i];
864 		if (channel && channel->type->copy) {
865 			efx_fini_napi_channel(channel);
866 			efx_remove_channel(channel);
867 			kfree(channel);
868 		}
869 	}
870 
871 	rc2 = efx_soft_enable_interrupts(efx);
872 	if (rc2) {
873 		rc = rc ? rc : rc2;
874 		netif_err(efx, drv, efx->net_dev,
875 			  "unable to restart interrupts on channel reallocation\n");
876 		efx_schedule_reset(efx, RESET_TYPE_DISABLE);
877 	} else {
878 		efx_start_all(efx);
879 		efx_device_attach_if_not_resetting(efx);
880 	}
881 	return rc;
882 
883 rollback:
884 	/* Swap back */
885 	efx->rxq_entries = old_rxq_entries;
886 	efx->txq_entries = old_txq_entries;
887 	for (i = 0; i < efx->n_channels; i++) {
888 		channel = efx->channel[i];
889 		efx->channel[i] = other_channel[i];
890 		other_channel[i] = channel;
891 	}
892 	goto out;
893 }
894 
895 void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
896 {
897 	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
898 }
899 
900 static const struct efx_channel_type efx_default_channel_type = {
901 	.pre_probe		= efx_channel_dummy_op_int,
902 	.post_remove		= efx_channel_dummy_op_void,
903 	.get_name		= efx_get_channel_name,
904 	.copy			= efx_copy_channel,
905 	.keep_eventq		= false,
906 };
907 
908 int efx_channel_dummy_op_int(struct efx_channel *channel)
909 {
910 	return 0;
911 }
912 
913 void efx_channel_dummy_op_void(struct efx_channel *channel)
914 {
915 }
916 
917 /**************************************************************************
918  *
919  * Port handling
920  *
921  **************************************************************************/
922 
923 /* This ensures that the kernel is kept informed (via
924  * netif_carrier_on/off) of the link status, and also maintains the
925  * link status's stop on the port's TX queue.
926  */
927 void efx_link_status_changed(struct efx_nic *efx)
928 {
929 	struct efx_link_state *link_state = &efx->link_state;
930 
931 	/* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
932 	 * that no events are triggered between unregister_netdev() and the
933 	 * driver unloading. A more general condition is that NETDEV_CHANGE
934 	 * can only be generated between NETDEV_UP and NETDEV_DOWN */
935 	if (!netif_running(efx->net_dev))
936 		return;
937 
938 	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
939 		efx->n_link_state_changes++;
940 
941 		if (link_state->up)
942 			netif_carrier_on(efx->net_dev);
943 		else
944 			netif_carrier_off(efx->net_dev);
945 	}
946 
947 	/* Status message for kernel log */
948 	if (link_state->up)
949 		netif_info(efx, link, efx->net_dev,
950 			   "link up at %uMbps %s-duplex (MTU %d)\n",
951 			   link_state->speed, link_state->fd ? "full" : "half",
952 			   efx->net_dev->mtu);
953 	else
954 		netif_info(efx, link, efx->net_dev, "link down\n");
955 }
956 
957 void efx_link_set_advertising(struct efx_nic *efx, u32 advertising)
958 {
959 	efx->link_advertising = advertising;
960 	if (advertising) {
961 		if (advertising & ADVERTISED_Pause)
962 			efx->wanted_fc |= (EFX_FC_TX | EFX_FC_RX);
963 		else
964 			efx->wanted_fc &= ~(EFX_FC_TX | EFX_FC_RX);
965 		if (advertising & ADVERTISED_Asym_Pause)
966 			efx->wanted_fc ^= EFX_FC_TX;
967 	}
968 }
969 
970 void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
971 {
972 	efx->wanted_fc = wanted_fc;
973 	if (efx->link_advertising) {
974 		if (wanted_fc & EFX_FC_RX)
975 			efx->link_advertising |= (ADVERTISED_Pause |
976 						  ADVERTISED_Asym_Pause);
977 		else
978 			efx->link_advertising &= ~(ADVERTISED_Pause |
979 						   ADVERTISED_Asym_Pause);
980 		if (wanted_fc & EFX_FC_TX)
981 			efx->link_advertising ^= ADVERTISED_Asym_Pause;
982 	}
983 }
984 
985 static void efx_fini_port(struct efx_nic *efx);
986 
987 /* We assume that efx->type->reconfigure_mac will always try to sync RX
988  * filters and therefore needs to read-lock the filter table against freeing
989  */
990 void efx_mac_reconfigure(struct efx_nic *efx)
991 {
992 	down_read(&efx->filter_sem);
993 	efx->type->reconfigure_mac(efx);
994 	up_read(&efx->filter_sem);
995 }
996 
997 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
998  * the MAC appropriately. All other PHY configuration changes are pushed
999  * through phy_op->set_settings(), and pushed asynchronously to the MAC
1000  * through efx_monitor().
1001  *
1002  * Callers must hold the mac_lock
1003  */
1004 int __efx_reconfigure_port(struct efx_nic *efx)
1005 {
1006 	enum efx_phy_mode phy_mode;
1007 	int rc;
1008 
1009 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
1010 
1011 	/* Disable PHY transmit in mac level loopbacks */
1012 	phy_mode = efx->phy_mode;
1013 	if (LOOPBACK_INTERNAL(efx))
1014 		efx->phy_mode |= PHY_MODE_TX_DISABLED;
1015 	else
1016 		efx->phy_mode &= ~PHY_MODE_TX_DISABLED;
1017 
1018 	rc = efx->type->reconfigure_port(efx);
1019 
1020 	if (rc)
1021 		efx->phy_mode = phy_mode;
1022 
1023 	return rc;
1024 }
1025 
1026 /* Reinitialise the MAC to pick up new PHY settings, even if the port is
1027  * disabled. */
1028 int efx_reconfigure_port(struct efx_nic *efx)
1029 {
1030 	int rc;
1031 
1032 	EFX_ASSERT_RESET_SERIALISED(efx);
1033 
1034 	mutex_lock(&efx->mac_lock);
1035 	rc = __efx_reconfigure_port(efx);
1036 	mutex_unlock(&efx->mac_lock);
1037 
1038 	return rc;
1039 }
1040 
1041 /* Asynchronous work item for changing MAC promiscuity and multicast
1042  * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
1043  * MAC directly. */
1044 static void efx_mac_work(struct work_struct *data)
1045 {
1046 	struct efx_nic *efx = container_of(data, struct efx_nic, mac_work);
1047 
1048 	mutex_lock(&efx->mac_lock);
1049 	if (efx->port_enabled)
1050 		efx_mac_reconfigure(efx);
1051 	mutex_unlock(&efx->mac_lock);
1052 }
1053 
1054 static int efx_probe_port(struct efx_nic *efx)
1055 {
1056 	int rc;
1057 
1058 	netif_dbg(efx, probe, efx->net_dev, "create port\n");
1059 
1060 	if (phy_flash_cfg)
1061 		efx->phy_mode = PHY_MODE_SPECIAL;
1062 
1063 	/* Connect up MAC/PHY operations table */
1064 	rc = efx->type->probe_port(efx);
1065 	if (rc)
1066 		return rc;
1067 
1068 	/* Initialise MAC address to permanent address */
1069 	ether_addr_copy(efx->net_dev->dev_addr, efx->net_dev->perm_addr);
1070 
1071 	return 0;
1072 }
1073 
1074 static int efx_init_port(struct efx_nic *efx)
1075 {
1076 	int rc;
1077 
1078 	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1079 
1080 	mutex_lock(&efx->mac_lock);
1081 
1082 	rc = efx->phy_op->init(efx);
1083 	if (rc)
1084 		goto fail1;
1085 
1086 	efx->port_initialized = true;
1087 
1088 	/* Reconfigure the MAC before creating dma queues (required for
1089 	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1090 	efx_mac_reconfigure(efx);
1091 
1092 	/* Ensure the PHY advertises the correct flow control settings */
1093 	rc = efx->phy_op->reconfigure(efx);
1094 	if (rc && rc != -EPERM)
1095 		goto fail2;
1096 
1097 	mutex_unlock(&efx->mac_lock);
1098 	return 0;
1099 
1100 fail2:
1101 	efx->phy_op->fini(efx);
1102 fail1:
1103 	mutex_unlock(&efx->mac_lock);
1104 	return rc;
1105 }
1106 
1107 static void efx_start_port(struct efx_nic *efx)
1108 {
1109 	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1110 	BUG_ON(efx->port_enabled);
1111 
1112 	mutex_lock(&efx->mac_lock);
1113 	efx->port_enabled = true;
1114 
1115 	/* Ensure MAC ingress/egress is enabled */
1116 	efx_mac_reconfigure(efx);
1117 
1118 	mutex_unlock(&efx->mac_lock);
1119 }
1120 
1121 /* Cancel work for MAC reconfiguration, periodic hardware monitoring
1122  * and the async self-test, wait for them to finish and prevent them
1123  * being scheduled again.  This doesn't cover online resets, which
1124  * should only be cancelled when removing the device.
1125  */
1126 static void efx_stop_port(struct efx_nic *efx)
1127 {
1128 	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1129 
1130 	EFX_ASSERT_RESET_SERIALISED(efx);
1131 
1132 	mutex_lock(&efx->mac_lock);
1133 	efx->port_enabled = false;
1134 	mutex_unlock(&efx->mac_lock);
1135 
1136 	/* Serialise against efx_set_multicast_list() */
1137 	netif_addr_lock_bh(efx->net_dev);
1138 	netif_addr_unlock_bh(efx->net_dev);
1139 
1140 	cancel_delayed_work_sync(&efx->monitor_work);
1141 	efx_selftest_async_cancel(efx);
1142 	cancel_work_sync(&efx->mac_work);
1143 }
1144 
1145 static void efx_fini_port(struct efx_nic *efx)
1146 {
1147 	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1148 
1149 	if (!efx->port_initialized)
1150 		return;
1151 
1152 	efx->phy_op->fini(efx);
1153 	efx->port_initialized = false;
1154 
1155 	efx->link_state.up = false;
1156 	efx_link_status_changed(efx);
1157 }
1158 
1159 static void efx_remove_port(struct efx_nic *efx)
1160 {
1161 	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1162 
1163 	efx->type->remove_port(efx);
1164 }
1165 
1166 /**************************************************************************
1167  *
1168  * NIC handling
1169  *
1170  **************************************************************************/
1171 
1172 static LIST_HEAD(efx_primary_list);
1173 static LIST_HEAD(efx_unassociated_list);
1174 
1175 static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right)
1176 {
1177 	return left->type == right->type &&
1178 		left->vpd_sn && right->vpd_sn &&
1179 		!strcmp(left->vpd_sn, right->vpd_sn);
1180 }
1181 
1182 static void efx_associate(struct efx_nic *efx)
1183 {
1184 	struct efx_nic *other, *next;
1185 
1186 	if (efx->primary == efx) {
1187 		/* Adding primary function; look for secondaries */
1188 
1189 		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
1190 		list_add_tail(&efx->node, &efx_primary_list);
1191 
1192 		list_for_each_entry_safe(other, next, &efx_unassociated_list,
1193 					 node) {
1194 			if (efx_same_controller(efx, other)) {
1195 				list_del(&other->node);
1196 				netif_dbg(other, probe, other->net_dev,
1197 					  "moving to secondary list of %s %s\n",
1198 					  pci_name(efx->pci_dev),
1199 					  efx->net_dev->name);
1200 				list_add_tail(&other->node,
1201 					      &efx->secondary_list);
1202 				other->primary = efx;
1203 			}
1204 		}
1205 	} else {
1206 		/* Adding secondary function; look for primary */
1207 
1208 		list_for_each_entry(other, &efx_primary_list, node) {
1209 			if (efx_same_controller(efx, other)) {
1210 				netif_dbg(efx, probe, efx->net_dev,
1211 					  "adding to secondary list of %s %s\n",
1212 					  pci_name(other->pci_dev),
1213 					  other->net_dev->name);
1214 				list_add_tail(&efx->node,
1215 					      &other->secondary_list);
1216 				efx->primary = other;
1217 				return;
1218 			}
1219 		}
1220 
1221 		netif_dbg(efx, probe, efx->net_dev,
1222 			  "adding to unassociated list\n");
1223 		list_add_tail(&efx->node, &efx_unassociated_list);
1224 	}
1225 }
1226 
1227 static void efx_dissociate(struct efx_nic *efx)
1228 {
1229 	struct efx_nic *other, *next;
1230 
1231 	list_del(&efx->node);
1232 	efx->primary = NULL;
1233 
1234 	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
1235 		list_del(&other->node);
1236 		netif_dbg(other, probe, other->net_dev,
1237 			  "moving to unassociated list\n");
1238 		list_add_tail(&other->node, &efx_unassociated_list);
1239 		other->primary = NULL;
1240 	}
1241 }
1242 
1243 /* This configures the PCI device to enable I/O and DMA. */
1244 static int efx_init_io(struct efx_nic *efx)
1245 {
1246 	struct pci_dev *pci_dev = efx->pci_dev;
1247 	dma_addr_t dma_mask = efx->type->max_dma_mask;
1248 	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1249 	int rc, bar;
1250 
1251 	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1252 
1253 	bar = efx->type->mem_bar;
1254 
1255 	rc = pci_enable_device(pci_dev);
1256 	if (rc) {
1257 		netif_err(efx, probe, efx->net_dev,
1258 			  "failed to enable PCI device\n");
1259 		goto fail1;
1260 	}
1261 
1262 	pci_set_master(pci_dev);
1263 
1264 	/* Set the PCI DMA mask.  Try all possibilities from our
1265 	 * genuine mask down to 32 bits, because some architectures
1266 	 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
1267 	 * masks event though they reject 46 bit masks.
1268 	 */
1269 	while (dma_mask > 0x7fffffffUL) {
1270 		rc = dma_set_mask_and_coherent(&pci_dev->dev, dma_mask);
1271 		if (rc == 0)
1272 			break;
1273 		dma_mask >>= 1;
1274 	}
1275 	if (rc) {
1276 		netif_err(efx, probe, efx->net_dev,
1277 			  "could not find a suitable DMA mask\n");
1278 		goto fail2;
1279 	}
1280 	netif_dbg(efx, probe, efx->net_dev,
1281 		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1282 
1283 	efx->membase_phys = pci_resource_start(efx->pci_dev, bar);
1284 	rc = pci_request_region(pci_dev, bar, "sfc");
1285 	if (rc) {
1286 		netif_err(efx, probe, efx->net_dev,
1287 			  "request for memory BAR failed\n");
1288 		rc = -EIO;
1289 		goto fail3;
1290 	}
1291 	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1292 	if (!efx->membase) {
1293 		netif_err(efx, probe, efx->net_dev,
1294 			  "could not map memory BAR at %llx+%x\n",
1295 			  (unsigned long long)efx->membase_phys, mem_map_size);
1296 		rc = -ENOMEM;
1297 		goto fail4;
1298 	}
1299 	netif_dbg(efx, probe, efx->net_dev,
1300 		  "memory BAR at %llx+%x (virtual %p)\n",
1301 		  (unsigned long long)efx->membase_phys, mem_map_size,
1302 		  efx->membase);
1303 
1304 	return 0;
1305 
1306  fail4:
1307 	pci_release_region(efx->pci_dev, bar);
1308  fail3:
1309 	efx->membase_phys = 0;
1310  fail2:
1311 	pci_disable_device(efx->pci_dev);
1312  fail1:
1313 	return rc;
1314 }
1315 
1316 static void efx_fini_io(struct efx_nic *efx)
1317 {
1318 	int bar;
1319 
1320 	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1321 
1322 	if (efx->membase) {
1323 		iounmap(efx->membase);
1324 		efx->membase = NULL;
1325 	}
1326 
1327 	if (efx->membase_phys) {
1328 		bar = efx->type->mem_bar;
1329 		pci_release_region(efx->pci_dev, bar);
1330 		efx->membase_phys = 0;
1331 	}
1332 
1333 	/* Don't disable bus-mastering if VFs are assigned */
1334 	if (!pci_vfs_assigned(efx->pci_dev))
1335 		pci_disable_device(efx->pci_dev);
1336 }
1337 
1338 void efx_set_default_rx_indir_table(struct efx_nic *efx)
1339 {
1340 	size_t i;
1341 
1342 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
1343 		efx->rx_indir_table[i] =
1344 			ethtool_rxfh_indir_default(i, efx->rss_spread);
1345 }
1346 
1347 static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1348 {
1349 	cpumask_var_t thread_mask;
1350 	unsigned int count;
1351 	int cpu;
1352 
1353 	if (rss_cpus) {
1354 		count = rss_cpus;
1355 	} else {
1356 		if (unlikely(!zalloc_cpumask_var(&thread_mask, GFP_KERNEL))) {
1357 			netif_warn(efx, probe, efx->net_dev,
1358 				   "RSS disabled due to allocation failure\n");
1359 			return 1;
1360 		}
1361 
1362 		count = 0;
1363 		for_each_online_cpu(cpu) {
1364 			if (!cpumask_test_cpu(cpu, thread_mask)) {
1365 				++count;
1366 				cpumask_or(thread_mask, thread_mask,
1367 					   topology_sibling_cpumask(cpu));
1368 			}
1369 		}
1370 
1371 		free_cpumask_var(thread_mask);
1372 	}
1373 
1374 	/* If RSS is requested for the PF *and* VFs then we can't write RSS
1375 	 * table entries that are inaccessible to VFs
1376 	 */
1377 #ifdef CONFIG_SFC_SRIOV
1378 	if (efx->type->sriov_wanted) {
1379 		if (efx->type->sriov_wanted(efx) && efx_vf_size(efx) > 1 &&
1380 		    count > efx_vf_size(efx)) {
1381 			netif_warn(efx, probe, efx->net_dev,
1382 				   "Reducing number of RSS channels from %u to %u for "
1383 				   "VF support. Increase vf-msix-limit to use more "
1384 				   "channels on the PF.\n",
1385 				   count, efx_vf_size(efx));
1386 			count = efx_vf_size(efx);
1387 		}
1388 	}
1389 #endif
1390 
1391 	return count;
1392 }
1393 
1394 /* Probe the number and type of interrupts we are able to obtain, and
1395  * the resulting numbers of channels and RX queues.
1396  */
1397 static int efx_probe_interrupts(struct efx_nic *efx)
1398 {
1399 	unsigned int extra_channels = 0;
1400 	unsigned int i, j;
1401 	int rc;
1402 
1403 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
1404 		if (efx->extra_channel_type[i])
1405 			++extra_channels;
1406 
1407 	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1408 		struct msix_entry xentries[EFX_MAX_CHANNELS];
1409 		unsigned int n_channels;
1410 
1411 		n_channels = efx_wanted_parallelism(efx);
1412 		if (efx_separate_tx_channels)
1413 			n_channels *= 2;
1414 		n_channels += extra_channels;
1415 		n_channels = min(n_channels, efx->max_channels);
1416 
1417 		for (i = 0; i < n_channels; i++)
1418 			xentries[i].entry = i;
1419 		rc = pci_enable_msix_range(efx->pci_dev,
1420 					   xentries, 1, n_channels);
1421 		if (rc < 0) {
1422 			/* Fall back to single channel MSI */
1423 			netif_err(efx, drv, efx->net_dev,
1424 				  "could not enable MSI-X\n");
1425 			if (efx->type->min_interrupt_mode >= EFX_INT_MODE_MSI)
1426 				efx->interrupt_mode = EFX_INT_MODE_MSI;
1427 			else
1428 				return rc;
1429 		} else if (rc < n_channels) {
1430 			netif_err(efx, drv, efx->net_dev,
1431 				  "WARNING: Insufficient MSI-X vectors"
1432 				  " available (%d < %u).\n", rc, n_channels);
1433 			netif_err(efx, drv, efx->net_dev,
1434 				  "WARNING: Performance may be reduced.\n");
1435 			n_channels = rc;
1436 		}
1437 
1438 		if (rc > 0) {
1439 			efx->n_channels = n_channels;
1440 			if (n_channels > extra_channels)
1441 				n_channels -= extra_channels;
1442 			if (efx_separate_tx_channels) {
1443 				efx->n_tx_channels = min(max(n_channels / 2,
1444 							     1U),
1445 							 efx->max_tx_channels);
1446 				efx->n_rx_channels = max(n_channels -
1447 							 efx->n_tx_channels,
1448 							 1U);
1449 			} else {
1450 				efx->n_tx_channels = min(n_channels,
1451 							 efx->max_tx_channels);
1452 				efx->n_rx_channels = n_channels;
1453 			}
1454 			for (i = 0; i < efx->n_channels; i++)
1455 				efx_get_channel(efx, i)->irq =
1456 					xentries[i].vector;
1457 		}
1458 	}
1459 
1460 	/* Try single interrupt MSI */
1461 	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1462 		efx->n_channels = 1;
1463 		efx->n_rx_channels = 1;
1464 		efx->n_tx_channels = 1;
1465 		rc = pci_enable_msi(efx->pci_dev);
1466 		if (rc == 0) {
1467 			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1468 		} else {
1469 			netif_err(efx, drv, efx->net_dev,
1470 				  "could not enable MSI\n");
1471 			if (efx->type->min_interrupt_mode >= EFX_INT_MODE_LEGACY)
1472 				efx->interrupt_mode = EFX_INT_MODE_LEGACY;
1473 			else
1474 				return rc;
1475 		}
1476 	}
1477 
1478 	/* Assume legacy interrupts */
1479 	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1480 		efx->n_channels = 1 + (efx_separate_tx_channels ? 1 : 0);
1481 		efx->n_rx_channels = 1;
1482 		efx->n_tx_channels = 1;
1483 		efx->legacy_irq = efx->pci_dev->irq;
1484 	}
1485 
1486 	/* Assign extra channels if possible */
1487 	j = efx->n_channels;
1488 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++) {
1489 		if (!efx->extra_channel_type[i])
1490 			continue;
1491 		if (efx->interrupt_mode != EFX_INT_MODE_MSIX ||
1492 		    efx->n_channels <= extra_channels) {
1493 			efx->extra_channel_type[i]->handle_no_channel(efx);
1494 		} else {
1495 			--j;
1496 			efx_get_channel(efx, j)->type =
1497 				efx->extra_channel_type[i];
1498 		}
1499 	}
1500 
1501 	/* RSS might be usable on VFs even if it is disabled on the PF */
1502 #ifdef CONFIG_SFC_SRIOV
1503 	if (efx->type->sriov_wanted) {
1504 		efx->rss_spread = ((efx->n_rx_channels > 1 ||
1505 				    !efx->type->sriov_wanted(efx)) ?
1506 				   efx->n_rx_channels : efx_vf_size(efx));
1507 		return 0;
1508 	}
1509 #endif
1510 	efx->rss_spread = efx->n_rx_channels;
1511 
1512 	return 0;
1513 }
1514 
1515 static int efx_soft_enable_interrupts(struct efx_nic *efx)
1516 {
1517 	struct efx_channel *channel, *end_channel;
1518 	int rc;
1519 
1520 	BUG_ON(efx->state == STATE_DISABLED);
1521 
1522 	efx->irq_soft_enabled = true;
1523 	smp_wmb();
1524 
1525 	efx_for_each_channel(channel, efx) {
1526 		if (!channel->type->keep_eventq) {
1527 			rc = efx_init_eventq(channel);
1528 			if (rc)
1529 				goto fail;
1530 		}
1531 		efx_start_eventq(channel);
1532 	}
1533 
1534 	efx_mcdi_mode_event(efx);
1535 
1536 	return 0;
1537 fail:
1538 	end_channel = channel;
1539 	efx_for_each_channel(channel, efx) {
1540 		if (channel == end_channel)
1541 			break;
1542 		efx_stop_eventq(channel);
1543 		if (!channel->type->keep_eventq)
1544 			efx_fini_eventq(channel);
1545 	}
1546 
1547 	return rc;
1548 }
1549 
1550 static void efx_soft_disable_interrupts(struct efx_nic *efx)
1551 {
1552 	struct efx_channel *channel;
1553 
1554 	if (efx->state == STATE_DISABLED)
1555 		return;
1556 
1557 	efx_mcdi_mode_poll(efx);
1558 
1559 	efx->irq_soft_enabled = false;
1560 	smp_wmb();
1561 
1562 	if (efx->legacy_irq)
1563 		synchronize_irq(efx->legacy_irq);
1564 
1565 	efx_for_each_channel(channel, efx) {
1566 		if (channel->irq)
1567 			synchronize_irq(channel->irq);
1568 
1569 		efx_stop_eventq(channel);
1570 		if (!channel->type->keep_eventq)
1571 			efx_fini_eventq(channel);
1572 	}
1573 
1574 	/* Flush the asynchronous MCDI request queue */
1575 	efx_mcdi_flush_async(efx);
1576 }
1577 
1578 static int efx_enable_interrupts(struct efx_nic *efx)
1579 {
1580 	struct efx_channel *channel, *end_channel;
1581 	int rc;
1582 
1583 	BUG_ON(efx->state == STATE_DISABLED);
1584 
1585 	if (efx->eeh_disabled_legacy_irq) {
1586 		enable_irq(efx->legacy_irq);
1587 		efx->eeh_disabled_legacy_irq = false;
1588 	}
1589 
1590 	efx->type->irq_enable_master(efx);
1591 
1592 	efx_for_each_channel(channel, efx) {
1593 		if (channel->type->keep_eventq) {
1594 			rc = efx_init_eventq(channel);
1595 			if (rc)
1596 				goto fail;
1597 		}
1598 	}
1599 
1600 	rc = efx_soft_enable_interrupts(efx);
1601 	if (rc)
1602 		goto fail;
1603 
1604 	return 0;
1605 
1606 fail:
1607 	end_channel = channel;
1608 	efx_for_each_channel(channel, efx) {
1609 		if (channel == end_channel)
1610 			break;
1611 		if (channel->type->keep_eventq)
1612 			efx_fini_eventq(channel);
1613 	}
1614 
1615 	efx->type->irq_disable_non_ev(efx);
1616 
1617 	return rc;
1618 }
1619 
1620 static void efx_disable_interrupts(struct efx_nic *efx)
1621 {
1622 	struct efx_channel *channel;
1623 
1624 	efx_soft_disable_interrupts(efx);
1625 
1626 	efx_for_each_channel(channel, efx) {
1627 		if (channel->type->keep_eventq)
1628 			efx_fini_eventq(channel);
1629 	}
1630 
1631 	efx->type->irq_disable_non_ev(efx);
1632 }
1633 
1634 static void efx_remove_interrupts(struct efx_nic *efx)
1635 {
1636 	struct efx_channel *channel;
1637 
1638 	/* Remove MSI/MSI-X interrupts */
1639 	efx_for_each_channel(channel, efx)
1640 		channel->irq = 0;
1641 	pci_disable_msi(efx->pci_dev);
1642 	pci_disable_msix(efx->pci_dev);
1643 
1644 	/* Remove legacy interrupt */
1645 	efx->legacy_irq = 0;
1646 }
1647 
1648 static void efx_set_channels(struct efx_nic *efx)
1649 {
1650 	struct efx_channel *channel;
1651 	struct efx_tx_queue *tx_queue;
1652 
1653 	efx->tx_channel_offset =
1654 		efx_separate_tx_channels ?
1655 		efx->n_channels - efx->n_tx_channels : 0;
1656 
1657 	/* We need to mark which channels really have RX and TX
1658 	 * queues, and adjust the TX queue numbers if we have separate
1659 	 * RX-only and TX-only channels.
1660 	 */
1661 	efx_for_each_channel(channel, efx) {
1662 		if (channel->channel < efx->n_rx_channels)
1663 			channel->rx_queue.core_index = channel->channel;
1664 		else
1665 			channel->rx_queue.core_index = -1;
1666 
1667 		efx_for_each_channel_tx_queue(tx_queue, channel)
1668 			tx_queue->queue -= (efx->tx_channel_offset *
1669 					    EFX_TXQ_TYPES);
1670 	}
1671 }
1672 
1673 static int efx_probe_nic(struct efx_nic *efx)
1674 {
1675 	int rc;
1676 
1677 	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1678 
1679 	/* Carry out hardware-type specific initialisation */
1680 	rc = efx->type->probe(efx);
1681 	if (rc)
1682 		return rc;
1683 
1684 	do {
1685 		if (!efx->max_channels || !efx->max_tx_channels) {
1686 			netif_err(efx, drv, efx->net_dev,
1687 				  "Insufficient resources to allocate"
1688 				  " any channels\n");
1689 			rc = -ENOSPC;
1690 			goto fail1;
1691 		}
1692 
1693 		/* Determine the number of channels and queues by trying
1694 		 * to hook in MSI-X interrupts.
1695 		 */
1696 		rc = efx_probe_interrupts(efx);
1697 		if (rc)
1698 			goto fail1;
1699 
1700 		efx_set_channels(efx);
1701 
1702 		/* dimension_resources can fail with EAGAIN */
1703 		rc = efx->type->dimension_resources(efx);
1704 		if (rc != 0 && rc != -EAGAIN)
1705 			goto fail2;
1706 
1707 		if (rc == -EAGAIN)
1708 			/* try again with new max_channels */
1709 			efx_remove_interrupts(efx);
1710 
1711 	} while (rc == -EAGAIN);
1712 
1713 	if (efx->n_channels > 1)
1714 		netdev_rss_key_fill(&efx->rx_hash_key,
1715 				    sizeof(efx->rx_hash_key));
1716 	efx_set_default_rx_indir_table(efx);
1717 
1718 	netif_set_real_num_tx_queues(efx->net_dev, efx->n_tx_channels);
1719 	netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels);
1720 
1721 	/* Initialise the interrupt moderation settings */
1722 	efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000);
1723 	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
1724 				true);
1725 
1726 	return 0;
1727 
1728 fail2:
1729 	efx_remove_interrupts(efx);
1730 fail1:
1731 	efx->type->remove(efx);
1732 	return rc;
1733 }
1734 
1735 static void efx_remove_nic(struct efx_nic *efx)
1736 {
1737 	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1738 
1739 	efx_remove_interrupts(efx);
1740 	efx->type->remove(efx);
1741 }
1742 
1743 static int efx_probe_filters(struct efx_nic *efx)
1744 {
1745 	int rc;
1746 
1747 	spin_lock_init(&efx->filter_lock);
1748 	init_rwsem(&efx->filter_sem);
1749 	mutex_lock(&efx->mac_lock);
1750 	down_write(&efx->filter_sem);
1751 	rc = efx->type->filter_table_probe(efx);
1752 	if (rc)
1753 		goto out_unlock;
1754 
1755 #ifdef CONFIG_RFS_ACCEL
1756 	if (efx->type->offload_features & NETIF_F_NTUPLE) {
1757 		struct efx_channel *channel;
1758 		int i, success = 1;
1759 
1760 		efx_for_each_channel(channel, efx) {
1761 			channel->rps_flow_id =
1762 				kcalloc(efx->type->max_rx_ip_filters,
1763 					sizeof(*channel->rps_flow_id),
1764 					GFP_KERNEL);
1765 			if (!channel->rps_flow_id)
1766 				success = 0;
1767 			else
1768 				for (i = 0;
1769 				     i < efx->type->max_rx_ip_filters;
1770 				     ++i)
1771 					channel->rps_flow_id[i] =
1772 						RPS_FLOW_ID_INVALID;
1773 		}
1774 
1775 		if (!success) {
1776 			efx_for_each_channel(channel, efx)
1777 				kfree(channel->rps_flow_id);
1778 			efx->type->filter_table_remove(efx);
1779 			rc = -ENOMEM;
1780 			goto out_unlock;
1781 		}
1782 
1783 		efx->rps_expire_index = efx->rps_expire_channel = 0;
1784 	}
1785 #endif
1786 out_unlock:
1787 	up_write(&efx->filter_sem);
1788 	mutex_unlock(&efx->mac_lock);
1789 	return rc;
1790 }
1791 
1792 static void efx_remove_filters(struct efx_nic *efx)
1793 {
1794 #ifdef CONFIG_RFS_ACCEL
1795 	struct efx_channel *channel;
1796 
1797 	efx_for_each_channel(channel, efx)
1798 		kfree(channel->rps_flow_id);
1799 #endif
1800 	down_write(&efx->filter_sem);
1801 	efx->type->filter_table_remove(efx);
1802 	up_write(&efx->filter_sem);
1803 }
1804 
1805 static void efx_restore_filters(struct efx_nic *efx)
1806 {
1807 	down_read(&efx->filter_sem);
1808 	efx->type->filter_table_restore(efx);
1809 	up_read(&efx->filter_sem);
1810 }
1811 
1812 /**************************************************************************
1813  *
1814  * NIC startup/shutdown
1815  *
1816  *************************************************************************/
1817 
1818 static int efx_probe_all(struct efx_nic *efx)
1819 {
1820 	int rc;
1821 
1822 	rc = efx_probe_nic(efx);
1823 	if (rc) {
1824 		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1825 		goto fail1;
1826 	}
1827 
1828 	rc = efx_probe_port(efx);
1829 	if (rc) {
1830 		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1831 		goto fail2;
1832 	}
1833 
1834 	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
1835 	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
1836 		rc = -EINVAL;
1837 		goto fail3;
1838 	}
1839 	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1840 
1841 #ifdef CONFIG_SFC_SRIOV
1842 	rc = efx->type->vswitching_probe(efx);
1843 	if (rc) /* not fatal; the PF will still work fine */
1844 		netif_warn(efx, probe, efx->net_dev,
1845 			   "failed to setup vswitching rc=%d;"
1846 			   " VFs may not function\n", rc);
1847 #endif
1848 
1849 	rc = efx_probe_filters(efx);
1850 	if (rc) {
1851 		netif_err(efx, probe, efx->net_dev,
1852 			  "failed to create filter tables\n");
1853 		goto fail4;
1854 	}
1855 
1856 	rc = efx_probe_channels(efx);
1857 	if (rc)
1858 		goto fail5;
1859 
1860 	return 0;
1861 
1862  fail5:
1863 	efx_remove_filters(efx);
1864  fail4:
1865 #ifdef CONFIG_SFC_SRIOV
1866 	efx->type->vswitching_remove(efx);
1867 #endif
1868  fail3:
1869 	efx_remove_port(efx);
1870  fail2:
1871 	efx_remove_nic(efx);
1872  fail1:
1873 	return rc;
1874 }
1875 
1876 /* If the interface is supposed to be running but is not, start
1877  * the hardware and software data path, regular activity for the port
1878  * (MAC statistics, link polling, etc.) and schedule the port to be
1879  * reconfigured.  Interrupts must already be enabled.  This function
1880  * is safe to call multiple times, so long as the NIC is not disabled.
1881  * Requires the RTNL lock.
1882  */
1883 static void efx_start_all(struct efx_nic *efx)
1884 {
1885 	EFX_ASSERT_RESET_SERIALISED(efx);
1886 	BUG_ON(efx->state == STATE_DISABLED);
1887 
1888 	/* Check that it is appropriate to restart the interface. All
1889 	 * of these flags are safe to read under just the rtnl lock */
1890 	if (efx->port_enabled || !netif_running(efx->net_dev) ||
1891 	    efx->reset_pending)
1892 		return;
1893 
1894 	efx_start_port(efx);
1895 	efx_start_datapath(efx);
1896 
1897 	/* Start the hardware monitor if there is one */
1898 	if (efx->type->monitor != NULL)
1899 		queue_delayed_work(efx->workqueue, &efx->monitor_work,
1900 				   efx_monitor_interval);
1901 
1902 	/* Link state detection is normally event-driven; we have
1903 	 * to poll now because we could have missed a change
1904 	 */
1905 	mutex_lock(&efx->mac_lock);
1906 	if (efx->phy_op->poll(efx))
1907 		efx_link_status_changed(efx);
1908 	mutex_unlock(&efx->mac_lock);
1909 
1910 	efx->type->start_stats(efx);
1911 	efx->type->pull_stats(efx);
1912 	spin_lock_bh(&efx->stats_lock);
1913 	efx->type->update_stats(efx, NULL, NULL);
1914 	spin_unlock_bh(&efx->stats_lock);
1915 }
1916 
1917 /* Quiesce the hardware and software data path, and regular activity
1918  * for the port without bringing the link down.  Safe to call multiple
1919  * times with the NIC in almost any state, but interrupts should be
1920  * enabled.  Requires the RTNL lock.
1921  */
1922 static void efx_stop_all(struct efx_nic *efx)
1923 {
1924 	EFX_ASSERT_RESET_SERIALISED(efx);
1925 
1926 	/* port_enabled can be read safely under the rtnl lock */
1927 	if (!efx->port_enabled)
1928 		return;
1929 
1930 	/* update stats before we go down so we can accurately count
1931 	 * rx_nodesc_drops
1932 	 */
1933 	efx->type->pull_stats(efx);
1934 	spin_lock_bh(&efx->stats_lock);
1935 	efx->type->update_stats(efx, NULL, NULL);
1936 	spin_unlock_bh(&efx->stats_lock);
1937 	efx->type->stop_stats(efx);
1938 	efx_stop_port(efx);
1939 
1940 	/* Stop the kernel transmit interface.  This is only valid if
1941 	 * the device is stopped or detached; otherwise the watchdog
1942 	 * may fire immediately.
1943 	 */
1944 	WARN_ON(netif_running(efx->net_dev) &&
1945 		netif_device_present(efx->net_dev));
1946 	netif_tx_disable(efx->net_dev);
1947 
1948 	efx_stop_datapath(efx);
1949 }
1950 
1951 static void efx_remove_all(struct efx_nic *efx)
1952 {
1953 	efx_remove_channels(efx);
1954 	efx_remove_filters(efx);
1955 #ifdef CONFIG_SFC_SRIOV
1956 	efx->type->vswitching_remove(efx);
1957 #endif
1958 	efx_remove_port(efx);
1959 	efx_remove_nic(efx);
1960 }
1961 
1962 /**************************************************************************
1963  *
1964  * Interrupt moderation
1965  *
1966  **************************************************************************/
1967 unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs)
1968 {
1969 	if (usecs == 0)
1970 		return 0;
1971 	if (usecs * 1000 < efx->timer_quantum_ns)
1972 		return 1; /* never round down to 0 */
1973 	return usecs * 1000 / efx->timer_quantum_ns;
1974 }
1975 
1976 unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks)
1977 {
1978 	/* We must round up when converting ticks to microseconds
1979 	 * because we round down when converting the other way.
1980 	 */
1981 	return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000);
1982 }
1983 
1984 /* Set interrupt moderation parameters */
1985 int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
1986 			    unsigned int rx_usecs, bool rx_adaptive,
1987 			    bool rx_may_override_tx)
1988 {
1989 	struct efx_channel *channel;
1990 	unsigned int timer_max_us;
1991 
1992 	EFX_ASSERT_RESET_SERIALISED(efx);
1993 
1994 	timer_max_us = efx->timer_max_ns / 1000;
1995 
1996 	if (tx_usecs > timer_max_us || rx_usecs > timer_max_us)
1997 		return -EINVAL;
1998 
1999 	if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 &&
2000 	    !rx_may_override_tx) {
2001 		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
2002 			  "RX and TX IRQ moderation must be equal\n");
2003 		return -EINVAL;
2004 	}
2005 
2006 	efx->irq_rx_adaptive = rx_adaptive;
2007 	efx->irq_rx_moderation_us = rx_usecs;
2008 	efx_for_each_channel(channel, efx) {
2009 		if (efx_channel_has_rx_queue(channel))
2010 			channel->irq_moderation_us = rx_usecs;
2011 		else if (efx_channel_has_tx_queues(channel))
2012 			channel->irq_moderation_us = tx_usecs;
2013 	}
2014 
2015 	return 0;
2016 }
2017 
2018 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
2019 			    unsigned int *rx_usecs, bool *rx_adaptive)
2020 {
2021 	*rx_adaptive = efx->irq_rx_adaptive;
2022 	*rx_usecs = efx->irq_rx_moderation_us;
2023 
2024 	/* If channels are shared between RX and TX, so is IRQ
2025 	 * moderation.  Otherwise, IRQ moderation is the same for all
2026 	 * TX channels and is not adaptive.
2027 	 */
2028 	if (efx->tx_channel_offset == 0) {
2029 		*tx_usecs = *rx_usecs;
2030 	} else {
2031 		struct efx_channel *tx_channel;
2032 
2033 		tx_channel = efx->channel[efx->tx_channel_offset];
2034 		*tx_usecs = tx_channel->irq_moderation_us;
2035 	}
2036 }
2037 
2038 /**************************************************************************
2039  *
2040  * Hardware monitor
2041  *
2042  **************************************************************************/
2043 
2044 /* Run periodically off the general workqueue */
2045 static void efx_monitor(struct work_struct *data)
2046 {
2047 	struct efx_nic *efx = container_of(data, struct efx_nic,
2048 					   monitor_work.work);
2049 
2050 	netif_vdbg(efx, timer, efx->net_dev,
2051 		   "hardware monitor executing on CPU %d\n",
2052 		   raw_smp_processor_id());
2053 	BUG_ON(efx->type->monitor == NULL);
2054 
2055 	/* If the mac_lock is already held then it is likely a port
2056 	 * reconfiguration is already in place, which will likely do
2057 	 * most of the work of monitor() anyway. */
2058 	if (mutex_trylock(&efx->mac_lock)) {
2059 		if (efx->port_enabled)
2060 			efx->type->monitor(efx);
2061 		mutex_unlock(&efx->mac_lock);
2062 	}
2063 
2064 	queue_delayed_work(efx->workqueue, &efx->monitor_work,
2065 			   efx_monitor_interval);
2066 }
2067 
2068 /**************************************************************************
2069  *
2070  * ioctls
2071  *
2072  *************************************************************************/
2073 
2074 /* Net device ioctl
2075  * Context: process, rtnl_lock() held.
2076  */
2077 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
2078 {
2079 	struct efx_nic *efx = netdev_priv(net_dev);
2080 	struct mii_ioctl_data *data = if_mii(ifr);
2081 
2082 	if (cmd == SIOCSHWTSTAMP)
2083 		return efx_ptp_set_ts_config(efx, ifr);
2084 	if (cmd == SIOCGHWTSTAMP)
2085 		return efx_ptp_get_ts_config(efx, ifr);
2086 
2087 	/* Convert phy_id from older PRTAD/DEVAD format */
2088 	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
2089 	    (data->phy_id & 0xfc00) == 0x0400)
2090 		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;
2091 
2092 	return mdio_mii_ioctl(&efx->mdio, data, cmd);
2093 }
2094 
2095 /**************************************************************************
2096  *
2097  * NAPI interface
2098  *
2099  **************************************************************************/
2100 
2101 static void efx_init_napi_channel(struct efx_channel *channel)
2102 {
2103 	struct efx_nic *efx = channel->efx;
2104 
2105 	channel->napi_dev = efx->net_dev;
2106 	netif_napi_add(channel->napi_dev, &channel->napi_str,
2107 		       efx_poll, napi_weight);
2108 }
2109 
2110 static void efx_init_napi(struct efx_nic *efx)
2111 {
2112 	struct efx_channel *channel;
2113 
2114 	efx_for_each_channel(channel, efx)
2115 		efx_init_napi_channel(channel);
2116 }
2117 
2118 static void efx_fini_napi_channel(struct efx_channel *channel)
2119 {
2120 	if (channel->napi_dev)
2121 		netif_napi_del(&channel->napi_str);
2122 
2123 	channel->napi_dev = NULL;
2124 }
2125 
2126 static void efx_fini_napi(struct efx_nic *efx)
2127 {
2128 	struct efx_channel *channel;
2129 
2130 	efx_for_each_channel(channel, efx)
2131 		efx_fini_napi_channel(channel);
2132 }
2133 
2134 /**************************************************************************
2135  *
2136  * Kernel netpoll interface
2137  *
2138  *************************************************************************/
2139 
2140 #ifdef CONFIG_NET_POLL_CONTROLLER
2141 
2142 /* Although in the common case interrupts will be disabled, this is not
2143  * guaranteed. However, all our work happens inside the NAPI callback,
2144  * so no locking is required.
2145  */
2146 static void efx_netpoll(struct net_device *net_dev)
2147 {
2148 	struct efx_nic *efx = netdev_priv(net_dev);
2149 	struct efx_channel *channel;
2150 
2151 	efx_for_each_channel(channel, efx)
2152 		efx_schedule_channel(channel);
2153 }
2154 
2155 #endif
2156 
2157 /**************************************************************************
2158  *
2159  * Kernel net device interface
2160  *
2161  *************************************************************************/
2162 
2163 /* Context: process, rtnl_lock() held. */
2164 int efx_net_open(struct net_device *net_dev)
2165 {
2166 	struct efx_nic *efx = netdev_priv(net_dev);
2167 	int rc;
2168 
2169 	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
2170 		  raw_smp_processor_id());
2171 
2172 	rc = efx_check_disabled(efx);
2173 	if (rc)
2174 		return rc;
2175 	if (efx->phy_mode & PHY_MODE_SPECIAL)
2176 		return -EBUSY;
2177 	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
2178 		return -EIO;
2179 
2180 	/* Notify the kernel of the link state polled during driver load,
2181 	 * before the monitor starts running */
2182 	efx_link_status_changed(efx);
2183 
2184 	efx_start_all(efx);
2185 	if (efx->state == STATE_DISABLED || efx->reset_pending)
2186 		netif_device_detach(efx->net_dev);
2187 	efx_selftest_async_start(efx);
2188 	return 0;
2189 }
2190 
2191 /* Context: process, rtnl_lock() held.
2192  * Note that the kernel will ignore our return code; this method
2193  * should really be a void.
2194  */
2195 int efx_net_stop(struct net_device *net_dev)
2196 {
2197 	struct efx_nic *efx = netdev_priv(net_dev);
2198 
2199 	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
2200 		  raw_smp_processor_id());
2201 
2202 	/* Stop the device and flush all the channels */
2203 	efx_stop_all(efx);
2204 
2205 	return 0;
2206 }
2207 
2208 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
2209 static void efx_net_stats(struct net_device *net_dev,
2210 			  struct rtnl_link_stats64 *stats)
2211 {
2212 	struct efx_nic *efx = netdev_priv(net_dev);
2213 
2214 	spin_lock_bh(&efx->stats_lock);
2215 	efx->type->update_stats(efx, NULL, stats);
2216 	spin_unlock_bh(&efx->stats_lock);
2217 }
2218 
2219 /* Context: netif_tx_lock held, BHs disabled. */
2220 static void efx_watchdog(struct net_device *net_dev)
2221 {
2222 	struct efx_nic *efx = netdev_priv(net_dev);
2223 
2224 	netif_err(efx, tx_err, efx->net_dev,
2225 		  "TX stuck with port_enabled=%d: resetting channels\n",
2226 		  efx->port_enabled);
2227 
2228 	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2229 }
2230 
2231 
2232 /* Context: process, rtnl_lock() held. */
2233 static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
2234 {
2235 	struct efx_nic *efx = netdev_priv(net_dev);
2236 	int rc;
2237 
2238 	rc = efx_check_disabled(efx);
2239 	if (rc)
2240 		return rc;
2241 
2242 	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2243 
2244 	efx_device_detach_sync(efx);
2245 	efx_stop_all(efx);
2246 
2247 	mutex_lock(&efx->mac_lock);
2248 	net_dev->mtu = new_mtu;
2249 	efx_mac_reconfigure(efx);
2250 	mutex_unlock(&efx->mac_lock);
2251 
2252 	efx_start_all(efx);
2253 	efx_device_attach_if_not_resetting(efx);
2254 	return 0;
2255 }
2256 
2257 static int efx_set_mac_address(struct net_device *net_dev, void *data)
2258 {
2259 	struct efx_nic *efx = netdev_priv(net_dev);
2260 	struct sockaddr *addr = data;
2261 	u8 *new_addr = addr->sa_data;
2262 	u8 old_addr[6];
2263 	int rc;
2264 
2265 	if (!is_valid_ether_addr(new_addr)) {
2266 		netif_err(efx, drv, efx->net_dev,
2267 			  "invalid ethernet MAC address requested: %pM\n",
2268 			  new_addr);
2269 		return -EADDRNOTAVAIL;
2270 	}
2271 
2272 	/* save old address */
2273 	ether_addr_copy(old_addr, net_dev->dev_addr);
2274 	ether_addr_copy(net_dev->dev_addr, new_addr);
2275 	if (efx->type->set_mac_address) {
2276 		rc = efx->type->set_mac_address(efx);
2277 		if (rc) {
2278 			ether_addr_copy(net_dev->dev_addr, old_addr);
2279 			return rc;
2280 		}
2281 	}
2282 
2283 	/* Reconfigure the MAC */
2284 	mutex_lock(&efx->mac_lock);
2285 	efx_mac_reconfigure(efx);
2286 	mutex_unlock(&efx->mac_lock);
2287 
2288 	return 0;
2289 }
2290 
2291 /* Context: netif_addr_lock held, BHs disabled. */
2292 static void efx_set_rx_mode(struct net_device *net_dev)
2293 {
2294 	struct efx_nic *efx = netdev_priv(net_dev);
2295 
2296 	if (efx->port_enabled)
2297 		queue_work(efx->workqueue, &efx->mac_work);
2298 	/* Otherwise efx_start_port() will do this */
2299 }
2300 
2301 static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2302 {
2303 	struct efx_nic *efx = netdev_priv(net_dev);
2304 	int rc;
2305 
2306 	/* If disabling RX n-tuple filtering, clear existing filters */
2307 	if (net_dev->features & ~data & NETIF_F_NTUPLE) {
2308 		rc = efx->type->filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);
2309 		if (rc)
2310 			return rc;
2311 	}
2312 
2313 	/* If Rx VLAN filter is changed, update filters via mac_reconfigure */
2314 	if ((net_dev->features ^ data) & NETIF_F_HW_VLAN_CTAG_FILTER) {
2315 		/* efx_set_rx_mode() will schedule MAC work to update filters
2316 		 * when a new features are finally set in net_dev.
2317 		 */
2318 		efx_set_rx_mode(net_dev);
2319 	}
2320 
2321 	return 0;
2322 }
2323 
2324 static int efx_get_phys_port_id(struct net_device *net_dev,
2325 				struct netdev_phys_item_id *ppid)
2326 {
2327 	struct efx_nic *efx = netdev_priv(net_dev);
2328 
2329 	if (efx->type->get_phys_port_id)
2330 		return efx->type->get_phys_port_id(efx, ppid);
2331 	else
2332 		return -EOPNOTSUPP;
2333 }
2334 
2335 static int efx_get_phys_port_name(struct net_device *net_dev,
2336 				  char *name, size_t len)
2337 {
2338 	struct efx_nic *efx = netdev_priv(net_dev);
2339 
2340 	if (snprintf(name, len, "p%u", efx->port_num) >= len)
2341 		return -EINVAL;
2342 	return 0;
2343 }
2344 
2345 static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid)
2346 {
2347 	struct efx_nic *efx = netdev_priv(net_dev);
2348 
2349 	if (efx->type->vlan_rx_add_vid)
2350 		return efx->type->vlan_rx_add_vid(efx, proto, vid);
2351 	else
2352 		return -EOPNOTSUPP;
2353 }
2354 
2355 static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid)
2356 {
2357 	struct efx_nic *efx = netdev_priv(net_dev);
2358 
2359 	if (efx->type->vlan_rx_kill_vid)
2360 		return efx->type->vlan_rx_kill_vid(efx, proto, vid);
2361 	else
2362 		return -EOPNOTSUPP;
2363 }
2364 
2365 static int efx_udp_tunnel_type_map(enum udp_parsable_tunnel_type in)
2366 {
2367 	switch (in) {
2368 	case UDP_TUNNEL_TYPE_VXLAN:
2369 		return TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN;
2370 	case UDP_TUNNEL_TYPE_GENEVE:
2371 		return TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE;
2372 	default:
2373 		return -1;
2374 	}
2375 }
2376 
2377 static void efx_udp_tunnel_add(struct net_device *dev, struct udp_tunnel_info *ti)
2378 {
2379 	struct efx_nic *efx = netdev_priv(dev);
2380 	struct efx_udp_tunnel tnl;
2381 	int efx_tunnel_type;
2382 
2383 	efx_tunnel_type = efx_udp_tunnel_type_map(ti->type);
2384 	if (efx_tunnel_type < 0)
2385 		return;
2386 
2387 	tnl.type = (u16)efx_tunnel_type;
2388 	tnl.port = ti->port;
2389 
2390 	if (efx->type->udp_tnl_add_port)
2391 		(void)efx->type->udp_tnl_add_port(efx, tnl);
2392 }
2393 
2394 static void efx_udp_tunnel_del(struct net_device *dev, struct udp_tunnel_info *ti)
2395 {
2396 	struct efx_nic *efx = netdev_priv(dev);
2397 	struct efx_udp_tunnel tnl;
2398 	int efx_tunnel_type;
2399 
2400 	efx_tunnel_type = efx_udp_tunnel_type_map(ti->type);
2401 	if (efx_tunnel_type < 0)
2402 		return;
2403 
2404 	tnl.type = (u16)efx_tunnel_type;
2405 	tnl.port = ti->port;
2406 
2407 	if (efx->type->udp_tnl_del_port)
2408 		(void)efx->type->udp_tnl_del_port(efx, tnl);
2409 }
2410 
2411 static const struct net_device_ops efx_netdev_ops = {
2412 	.ndo_open		= efx_net_open,
2413 	.ndo_stop		= efx_net_stop,
2414 	.ndo_get_stats64	= efx_net_stats,
2415 	.ndo_tx_timeout		= efx_watchdog,
2416 	.ndo_start_xmit		= efx_hard_start_xmit,
2417 	.ndo_validate_addr	= eth_validate_addr,
2418 	.ndo_do_ioctl		= efx_ioctl,
2419 	.ndo_change_mtu		= efx_change_mtu,
2420 	.ndo_set_mac_address	= efx_set_mac_address,
2421 	.ndo_set_rx_mode	= efx_set_rx_mode,
2422 	.ndo_set_features	= efx_set_features,
2423 	.ndo_vlan_rx_add_vid	= efx_vlan_rx_add_vid,
2424 	.ndo_vlan_rx_kill_vid	= efx_vlan_rx_kill_vid,
2425 #ifdef CONFIG_SFC_SRIOV
2426 	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
2427 	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
2428 	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
2429 	.ndo_get_vf_config	= efx_sriov_get_vf_config,
2430 	.ndo_set_vf_link_state  = efx_sriov_set_vf_link_state,
2431 #endif
2432 	.ndo_get_phys_port_id   = efx_get_phys_port_id,
2433 	.ndo_get_phys_port_name	= efx_get_phys_port_name,
2434 #ifdef CONFIG_NET_POLL_CONTROLLER
2435 	.ndo_poll_controller = efx_netpoll,
2436 #endif
2437 	.ndo_setup_tc		= efx_setup_tc,
2438 #ifdef CONFIG_RFS_ACCEL
2439 	.ndo_rx_flow_steer	= efx_filter_rfs,
2440 #endif
2441 	.ndo_udp_tunnel_add	= efx_udp_tunnel_add,
2442 	.ndo_udp_tunnel_del	= efx_udp_tunnel_del,
2443 };
2444 
2445 static void efx_update_name(struct efx_nic *efx)
2446 {
2447 	strcpy(efx->name, efx->net_dev->name);
2448 	efx_mtd_rename(efx);
2449 	efx_set_channel_names(efx);
2450 }
2451 
2452 static int efx_netdev_event(struct notifier_block *this,
2453 			    unsigned long event, void *ptr)
2454 {
2455 	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2456 
2457 	if ((net_dev->netdev_ops == &efx_netdev_ops) &&
2458 	    event == NETDEV_CHANGENAME)
2459 		efx_update_name(netdev_priv(net_dev));
2460 
2461 	return NOTIFY_DONE;
2462 }
2463 
2464 static struct notifier_block efx_netdev_notifier = {
2465 	.notifier_call = efx_netdev_event,
2466 };
2467 
2468 static ssize_t
2469 show_phy_type(struct device *dev, struct device_attribute *attr, char *buf)
2470 {
2471 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2472 	return sprintf(buf, "%d\n", efx->phy_type);
2473 }
2474 static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
2475 
2476 #ifdef CONFIG_SFC_MCDI_LOGGING
2477 static ssize_t show_mcdi_log(struct device *dev, struct device_attribute *attr,
2478 			     char *buf)
2479 {
2480 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2481 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2482 
2483 	return scnprintf(buf, PAGE_SIZE, "%d\n", mcdi->logging_enabled);
2484 }
2485 static ssize_t set_mcdi_log(struct device *dev, struct device_attribute *attr,
2486 			    const char *buf, size_t count)
2487 {
2488 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2489 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2490 	bool enable = count > 0 && *buf != '0';
2491 
2492 	mcdi->logging_enabled = enable;
2493 	return count;
2494 }
2495 static DEVICE_ATTR(mcdi_logging, 0644, show_mcdi_log, set_mcdi_log);
2496 #endif
2497 
2498 static int efx_register_netdev(struct efx_nic *efx)
2499 {
2500 	struct net_device *net_dev = efx->net_dev;
2501 	struct efx_channel *channel;
2502 	int rc;
2503 
2504 	net_dev->watchdog_timeo = 5 * HZ;
2505 	net_dev->irq = efx->pci_dev->irq;
2506 	net_dev->netdev_ops = &efx_netdev_ops;
2507 	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0)
2508 		net_dev->priv_flags |= IFF_UNICAST_FLT;
2509 	net_dev->ethtool_ops = &efx_ethtool_ops;
2510 	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2511 	net_dev->min_mtu = EFX_MIN_MTU;
2512 	net_dev->max_mtu = EFX_MAX_MTU;
2513 
2514 	rtnl_lock();
2515 
2516 	/* Enable resets to be scheduled and check whether any were
2517 	 * already requested.  If so, the NIC is probably hosed so we
2518 	 * abort.
2519 	 */
2520 	efx->state = STATE_READY;
2521 	smp_mb(); /* ensure we change state before checking reset_pending */
2522 	if (efx->reset_pending) {
2523 		netif_err(efx, probe, efx->net_dev,
2524 			  "aborting probe due to scheduled reset\n");
2525 		rc = -EIO;
2526 		goto fail_locked;
2527 	}
2528 
2529 	rc = dev_alloc_name(net_dev, net_dev->name);
2530 	if (rc < 0)
2531 		goto fail_locked;
2532 	efx_update_name(efx);
2533 
2534 	/* Always start with carrier off; PHY events will detect the link */
2535 	netif_carrier_off(net_dev);
2536 
2537 	rc = register_netdevice(net_dev);
2538 	if (rc)
2539 		goto fail_locked;
2540 
2541 	efx_for_each_channel(channel, efx) {
2542 		struct efx_tx_queue *tx_queue;
2543 		efx_for_each_channel_tx_queue(tx_queue, channel)
2544 			efx_init_tx_queue_core_txq(tx_queue);
2545 	}
2546 
2547 	efx_associate(efx);
2548 
2549 	rtnl_unlock();
2550 
2551 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2552 	if (rc) {
2553 		netif_err(efx, drv, efx->net_dev,
2554 			  "failed to init net dev attributes\n");
2555 		goto fail_registered;
2556 	}
2557 #ifdef CONFIG_SFC_MCDI_LOGGING
2558 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2559 	if (rc) {
2560 		netif_err(efx, drv, efx->net_dev,
2561 			  "failed to init net dev attributes\n");
2562 		goto fail_attr_mcdi_logging;
2563 	}
2564 #endif
2565 
2566 	return 0;
2567 
2568 #ifdef CONFIG_SFC_MCDI_LOGGING
2569 fail_attr_mcdi_logging:
2570 	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2571 #endif
2572 fail_registered:
2573 	rtnl_lock();
2574 	efx_dissociate(efx);
2575 	unregister_netdevice(net_dev);
2576 fail_locked:
2577 	efx->state = STATE_UNINIT;
2578 	rtnl_unlock();
2579 	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2580 	return rc;
2581 }
2582 
2583 static void efx_unregister_netdev(struct efx_nic *efx)
2584 {
2585 	if (!efx->net_dev)
2586 		return;
2587 
2588 	BUG_ON(netdev_priv(efx->net_dev) != efx);
2589 
2590 	if (efx_dev_registered(efx)) {
2591 		strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
2592 #ifdef CONFIG_SFC_MCDI_LOGGING
2593 		device_remove_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2594 #endif
2595 		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2596 		unregister_netdev(efx->net_dev);
2597 	}
2598 }
2599 
2600 /**************************************************************************
2601  *
2602  * Device reset and suspend
2603  *
2604  **************************************************************************/
2605 
2606 /* Tears down the entire software state and most of the hardware state
2607  * before reset.  */
2608 void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2609 {
2610 	EFX_ASSERT_RESET_SERIALISED(efx);
2611 
2612 	if (method == RESET_TYPE_MCDI_TIMEOUT)
2613 		efx->type->prepare_flr(efx);
2614 
2615 	efx_stop_all(efx);
2616 	efx_disable_interrupts(efx);
2617 
2618 	mutex_lock(&efx->mac_lock);
2619 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2620 	    method != RESET_TYPE_DATAPATH)
2621 		efx->phy_op->fini(efx);
2622 	efx->type->fini(efx);
2623 }
2624 
2625 /* This function will always ensure that the locks acquired in
2626  * efx_reset_down() are released. A failure return code indicates
2627  * that we were unable to reinitialise the hardware, and the
2628  * driver should be disabled. If ok is false, then the rx and tx
2629  * engines are not restarted, pending a RESET_DISABLE. */
2630 int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2631 {
2632 	int rc;
2633 
2634 	EFX_ASSERT_RESET_SERIALISED(efx);
2635 
2636 	if (method == RESET_TYPE_MCDI_TIMEOUT)
2637 		efx->type->finish_flr(efx);
2638 
2639 	/* Ensure that SRAM is initialised even if we're disabling the device */
2640 	rc = efx->type->init(efx);
2641 	if (rc) {
2642 		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2643 		goto fail;
2644 	}
2645 
2646 	if (!ok)
2647 		goto fail;
2648 
2649 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2650 	    method != RESET_TYPE_DATAPATH) {
2651 		rc = efx->phy_op->init(efx);
2652 		if (rc)
2653 			goto fail;
2654 		rc = efx->phy_op->reconfigure(efx);
2655 		if (rc && rc != -EPERM)
2656 			netif_err(efx, drv, efx->net_dev,
2657 				  "could not restore PHY settings\n");
2658 	}
2659 
2660 	rc = efx_enable_interrupts(efx);
2661 	if (rc)
2662 		goto fail;
2663 
2664 #ifdef CONFIG_SFC_SRIOV
2665 	rc = efx->type->vswitching_restore(efx);
2666 	if (rc) /* not fatal; the PF will still work fine */
2667 		netif_warn(efx, probe, efx->net_dev,
2668 			   "failed to restore vswitching rc=%d;"
2669 			   " VFs may not function\n", rc);
2670 #endif
2671 
2672 	down_read(&efx->filter_sem);
2673 	efx_restore_filters(efx);
2674 	up_read(&efx->filter_sem);
2675 	if (efx->type->sriov_reset)
2676 		efx->type->sriov_reset(efx);
2677 
2678 	mutex_unlock(&efx->mac_lock);
2679 
2680 	efx_start_all(efx);
2681 
2682 	if (efx->type->udp_tnl_push_ports)
2683 		efx->type->udp_tnl_push_ports(efx);
2684 
2685 	return 0;
2686 
2687 fail:
2688 	efx->port_initialized = false;
2689 
2690 	mutex_unlock(&efx->mac_lock);
2691 
2692 	return rc;
2693 }
2694 
2695 /* Reset the NIC using the specified method.  Note that the reset may
2696  * fail, in which case the card will be left in an unusable state.
2697  *
2698  * Caller must hold the rtnl_lock.
2699  */
2700 int efx_reset(struct efx_nic *efx, enum reset_type method)
2701 {
2702 	int rc, rc2;
2703 	bool disabled;
2704 
2705 	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
2706 		   RESET_TYPE(method));
2707 
2708 	efx_device_detach_sync(efx);
2709 	efx_reset_down(efx, method);
2710 
2711 	rc = efx->type->reset(efx, method);
2712 	if (rc) {
2713 		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2714 		goto out;
2715 	}
2716 
2717 	/* Clear flags for the scopes we covered.  We assume the NIC and
2718 	 * driver are now quiescent so that there is no race here.
2719 	 */
2720 	if (method < RESET_TYPE_MAX_METHOD)
2721 		efx->reset_pending &= -(1 << (method + 1));
2722 	else /* it doesn't fit into the well-ordered scope hierarchy */
2723 		__clear_bit(method, &efx->reset_pending);
2724 
2725 	/* Reinitialise bus-mastering, which may have been turned off before
2726 	 * the reset was scheduled. This is still appropriate, even in the
2727 	 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
2728 	 * can respond to requests. */
2729 	pci_set_master(efx->pci_dev);
2730 
2731 out:
2732 	/* Leave device stopped if necessary */
2733 	disabled = rc ||
2734 		method == RESET_TYPE_DISABLE ||
2735 		method == RESET_TYPE_RECOVER_OR_DISABLE;
2736 	rc2 = efx_reset_up(efx, method, !disabled);
2737 	if (rc2) {
2738 		disabled = true;
2739 		if (!rc)
2740 			rc = rc2;
2741 	}
2742 
2743 	if (disabled) {
2744 		dev_close(efx->net_dev);
2745 		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2746 		efx->state = STATE_DISABLED;
2747 	} else {
2748 		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2749 		efx_device_attach_if_not_resetting(efx);
2750 	}
2751 	return rc;
2752 }
2753 
2754 /* Try recovery mechanisms.
2755  * For now only EEH is supported.
2756  * Returns 0 if the recovery mechanisms are unsuccessful.
2757  * Returns a non-zero value otherwise.
2758  */
2759 int efx_try_recovery(struct efx_nic *efx)
2760 {
2761 #ifdef CONFIG_EEH
2762 	/* A PCI error can occur and not be seen by EEH because nothing
2763 	 * happens on the PCI bus. In this case the driver may fail and
2764 	 * schedule a 'recover or reset', leading to this recovery handler.
2765 	 * Manually call the eeh failure check function.
2766 	 */
2767 	struct eeh_dev *eehdev = pci_dev_to_eeh_dev(efx->pci_dev);
2768 	if (eeh_dev_check_failure(eehdev)) {
2769 		/* The EEH mechanisms will handle the error and reset the
2770 		 * device if necessary.
2771 		 */
2772 		return 1;
2773 	}
2774 #endif
2775 	return 0;
2776 }
2777 
2778 static void efx_wait_for_bist_end(struct efx_nic *efx)
2779 {
2780 	int i;
2781 
2782 	for (i = 0; i < BIST_WAIT_DELAY_COUNT; ++i) {
2783 		if (efx_mcdi_poll_reboot(efx))
2784 			goto out;
2785 		msleep(BIST_WAIT_DELAY_MS);
2786 	}
2787 
2788 	netif_err(efx, drv, efx->net_dev, "Warning: No MC reboot after BIST mode\n");
2789 out:
2790 	/* Either way unset the BIST flag. If we found no reboot we probably
2791 	 * won't recover, but we should try.
2792 	 */
2793 	efx->mc_bist_for_other_fn = false;
2794 }
2795 
2796 /* The worker thread exists so that code that cannot sleep can
2797  * schedule a reset for later.
2798  */
2799 static void efx_reset_work(struct work_struct *data)
2800 {
2801 	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2802 	unsigned long pending;
2803 	enum reset_type method;
2804 
2805 	pending = ACCESS_ONCE(efx->reset_pending);
2806 	method = fls(pending) - 1;
2807 
2808 	if (method == RESET_TYPE_MC_BIST)
2809 		efx_wait_for_bist_end(efx);
2810 
2811 	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
2812 	     method == RESET_TYPE_RECOVER_OR_ALL) &&
2813 	    efx_try_recovery(efx))
2814 		return;
2815 
2816 	if (!pending)
2817 		return;
2818 
2819 	rtnl_lock();
2820 
2821 	/* We checked the state in efx_schedule_reset() but it may
2822 	 * have changed by now.  Now that we have the RTNL lock,
2823 	 * it cannot change again.
2824 	 */
2825 	if (efx->state == STATE_READY)
2826 		(void)efx_reset(efx, method);
2827 
2828 	rtnl_unlock();
2829 }
2830 
2831 void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
2832 {
2833 	enum reset_type method;
2834 
2835 	if (efx->state == STATE_RECOVERY) {
2836 		netif_dbg(efx, drv, efx->net_dev,
2837 			  "recovering: skip scheduling %s reset\n",
2838 			  RESET_TYPE(type));
2839 		return;
2840 	}
2841 
2842 	switch (type) {
2843 	case RESET_TYPE_INVISIBLE:
2844 	case RESET_TYPE_ALL:
2845 	case RESET_TYPE_RECOVER_OR_ALL:
2846 	case RESET_TYPE_WORLD:
2847 	case RESET_TYPE_DISABLE:
2848 	case RESET_TYPE_RECOVER_OR_DISABLE:
2849 	case RESET_TYPE_DATAPATH:
2850 	case RESET_TYPE_MC_BIST:
2851 	case RESET_TYPE_MCDI_TIMEOUT:
2852 		method = type;
2853 		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
2854 			  RESET_TYPE(method));
2855 		break;
2856 	default:
2857 		method = efx->type->map_reset_reason(type);
2858 		netif_dbg(efx, drv, efx->net_dev,
2859 			  "scheduling %s reset for %s\n",
2860 			  RESET_TYPE(method), RESET_TYPE(type));
2861 		break;
2862 	}
2863 
2864 	set_bit(method, &efx->reset_pending);
2865 	smp_mb(); /* ensure we change reset_pending before checking state */
2866 
2867 	/* If we're not READY then just leave the flags set as the cue
2868 	 * to abort probing or reschedule the reset later.
2869 	 */
2870 	if (ACCESS_ONCE(efx->state) != STATE_READY)
2871 		return;
2872 
2873 	/* efx_process_channel() will no longer read events once a
2874 	 * reset is scheduled. So switch back to poll'd MCDI completions. */
2875 	efx_mcdi_mode_poll(efx);
2876 
2877 	queue_work(reset_workqueue, &efx->reset_work);
2878 }
2879 
2880 /**************************************************************************
2881  *
2882  * List of NICs we support
2883  *
2884  **************************************************************************/
2885 
2886 /* PCI device ID table */
2887 static const struct pci_device_id efx_pci_table[] = {
2888 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2889 	 .driver_data = (unsigned long) &siena_a0_nic_type},
2890 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2891 	 .driver_data = (unsigned long) &siena_a0_nic_type},
2892 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
2893 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2894 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903),  /* SFC9120 VF */
2895 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2896 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923),  /* SFC9140 PF */
2897 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2898 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923),  /* SFC9140 VF */
2899 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2900 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03),  /* SFC9220 PF */
2901 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2902 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03),  /* SFC9220 VF */
2903 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2904 	{0}			/* end of list */
2905 };
2906 
2907 /**************************************************************************
2908  *
2909  * Dummy PHY/MAC operations
2910  *
2911  * Can be used for some unimplemented operations
2912  * Needed so all function pointers are valid and do not have to be tested
2913  * before use
2914  *
2915  **************************************************************************/
2916 int efx_port_dummy_op_int(struct efx_nic *efx)
2917 {
2918 	return 0;
2919 }
2920 void efx_port_dummy_op_void(struct efx_nic *efx) {}
2921 
2922 static bool efx_port_dummy_op_poll(struct efx_nic *efx)
2923 {
2924 	return false;
2925 }
2926 
2927 static const struct efx_phy_operations efx_dummy_phy_operations = {
2928 	.init		 = efx_port_dummy_op_int,
2929 	.reconfigure	 = efx_port_dummy_op_int,
2930 	.poll		 = efx_port_dummy_op_poll,
2931 	.fini		 = efx_port_dummy_op_void,
2932 };
2933 
2934 /**************************************************************************
2935  *
2936  * Data housekeeping
2937  *
2938  **************************************************************************/
2939 
2940 /* This zeroes out and then fills in the invariants in a struct
2941  * efx_nic (including all sub-structures).
2942  */
2943 static int efx_init_struct(struct efx_nic *efx,
2944 			   struct pci_dev *pci_dev, struct net_device *net_dev)
2945 {
2946 	int rc = -ENOMEM, i;
2947 
2948 	/* Initialise common structures */
2949 	INIT_LIST_HEAD(&efx->node);
2950 	INIT_LIST_HEAD(&efx->secondary_list);
2951 	spin_lock_init(&efx->biu_lock);
2952 #ifdef CONFIG_SFC_MTD
2953 	INIT_LIST_HEAD(&efx->mtd_list);
2954 #endif
2955 	INIT_WORK(&efx->reset_work, efx_reset_work);
2956 	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
2957 	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
2958 	efx->pci_dev = pci_dev;
2959 	efx->msg_enable = debug;
2960 	efx->state = STATE_UNINIT;
2961 	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
2962 
2963 	efx->net_dev = net_dev;
2964 	efx->rx_prefix_size = efx->type->rx_prefix_size;
2965 	efx->rx_ip_align =
2966 		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
2967 	efx->rx_packet_hash_offset =
2968 		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
2969 	efx->rx_packet_ts_offset =
2970 		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
2971 	spin_lock_init(&efx->stats_lock);
2972 	mutex_init(&efx->mac_lock);
2973 	efx->phy_op = &efx_dummy_phy_operations;
2974 	efx->mdio.dev = net_dev;
2975 	INIT_WORK(&efx->mac_work, efx_mac_work);
2976 	init_waitqueue_head(&efx->flush_wq);
2977 
2978 	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2979 		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
2980 		if (!efx->channel[i])
2981 			goto fail;
2982 		efx->msi_context[i].efx = efx;
2983 		efx->msi_context[i].index = i;
2984 	}
2985 
2986 	/* Higher numbered interrupt modes are less capable! */
2987 	if (WARN_ON_ONCE(efx->type->max_interrupt_mode >
2988 			 efx->type->min_interrupt_mode)) {
2989 		rc = -EIO;
2990 		goto fail;
2991 	}
2992 	efx->interrupt_mode = max(efx->type->max_interrupt_mode,
2993 				  interrupt_mode);
2994 	efx->interrupt_mode = min(efx->type->min_interrupt_mode,
2995 				  interrupt_mode);
2996 
2997 	/* Would be good to use the net_dev name, but we're too early */
2998 	snprintf(efx->workqueue_name, sizeof(efx->workqueue_name), "sfc%s",
2999 		 pci_name(pci_dev));
3000 	efx->workqueue = create_singlethread_workqueue(efx->workqueue_name);
3001 	if (!efx->workqueue)
3002 		goto fail;
3003 
3004 	return 0;
3005 
3006 fail:
3007 	efx_fini_struct(efx);
3008 	return rc;
3009 }
3010 
3011 static void efx_fini_struct(struct efx_nic *efx)
3012 {
3013 	int i;
3014 
3015 	for (i = 0; i < EFX_MAX_CHANNELS; i++)
3016 		kfree(efx->channel[i]);
3017 
3018 	kfree(efx->vpd_sn);
3019 
3020 	if (efx->workqueue) {
3021 		destroy_workqueue(efx->workqueue);
3022 		efx->workqueue = NULL;
3023 	}
3024 }
3025 
3026 void efx_update_sw_stats(struct efx_nic *efx, u64 *stats)
3027 {
3028 	u64 n_rx_nodesc_trunc = 0;
3029 	struct efx_channel *channel;
3030 
3031 	efx_for_each_channel(channel, efx)
3032 		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
3033 	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
3034 	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
3035 }
3036 
3037 /**************************************************************************
3038  *
3039  * PCI interface
3040  *
3041  **************************************************************************/
3042 
3043 /* Main body of final NIC shutdown code
3044  * This is called only at module unload (or hotplug removal).
3045  */
3046 static void efx_pci_remove_main(struct efx_nic *efx)
3047 {
3048 	/* Flush reset_work. It can no longer be scheduled since we
3049 	 * are not READY.
3050 	 */
3051 	BUG_ON(efx->state == STATE_READY);
3052 	cancel_work_sync(&efx->reset_work);
3053 
3054 	efx_disable_interrupts(efx);
3055 	efx_nic_fini_interrupt(efx);
3056 	efx_fini_port(efx);
3057 	efx->type->fini(efx);
3058 	efx_fini_napi(efx);
3059 	efx_remove_all(efx);
3060 }
3061 
3062 /* Final NIC shutdown
3063  * This is called only at module unload (or hotplug removal).  A PF can call
3064  * this on its VFs to ensure they are unbound first.
3065  */
3066 static void efx_pci_remove(struct pci_dev *pci_dev)
3067 {
3068 	struct efx_nic *efx;
3069 
3070 	efx = pci_get_drvdata(pci_dev);
3071 	if (!efx)
3072 		return;
3073 
3074 	/* Mark the NIC as fini, then stop the interface */
3075 	rtnl_lock();
3076 	efx_dissociate(efx);
3077 	dev_close(efx->net_dev);
3078 	efx_disable_interrupts(efx);
3079 	efx->state = STATE_UNINIT;
3080 	rtnl_unlock();
3081 
3082 	if (efx->type->sriov_fini)
3083 		efx->type->sriov_fini(efx);
3084 
3085 	efx_unregister_netdev(efx);
3086 
3087 	efx_mtd_remove(efx);
3088 
3089 	efx_pci_remove_main(efx);
3090 
3091 	efx_fini_io(efx);
3092 	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
3093 
3094 	efx_fini_struct(efx);
3095 	free_netdev(efx->net_dev);
3096 
3097 	pci_disable_pcie_error_reporting(pci_dev);
3098 };
3099 
3100 /* NIC VPD information
3101  * Called during probe to display the part number of the
3102  * installed NIC.  VPD is potentially very large but this should
3103  * always appear within the first 512 bytes.
3104  */
3105 #define SFC_VPD_LEN 512
3106 static void efx_probe_vpd_strings(struct efx_nic *efx)
3107 {
3108 	struct pci_dev *dev = efx->pci_dev;
3109 	char vpd_data[SFC_VPD_LEN];
3110 	ssize_t vpd_size;
3111 	int ro_start, ro_size, i, j;
3112 
3113 	/* Get the vpd data from the device */
3114 	vpd_size = pci_read_vpd(dev, 0, sizeof(vpd_data), vpd_data);
3115 	if (vpd_size <= 0) {
3116 		netif_err(efx, drv, efx->net_dev, "Unable to read VPD\n");
3117 		return;
3118 	}
3119 
3120 	/* Get the Read only section */
3121 	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size, PCI_VPD_LRDT_RO_DATA);
3122 	if (ro_start < 0) {
3123 		netif_err(efx, drv, efx->net_dev, "VPD Read-only not found\n");
3124 		return;
3125 	}
3126 
3127 	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
3128 	j = ro_size;
3129 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3130 	if (i + j > vpd_size)
3131 		j = vpd_size - i;
3132 
3133 	/* Get the Part number */
3134 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "PN");
3135 	if (i < 0) {
3136 		netif_err(efx, drv, efx->net_dev, "Part number not found\n");
3137 		return;
3138 	}
3139 
3140 	j = pci_vpd_info_field_size(&vpd_data[i]);
3141 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3142 	if (i + j > vpd_size) {
3143 		netif_err(efx, drv, efx->net_dev, "Incomplete part number\n");
3144 		return;
3145 	}
3146 
3147 	netif_info(efx, drv, efx->net_dev,
3148 		   "Part Number : %.*s\n", j, &vpd_data[i]);
3149 
3150 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3151 	j = ro_size;
3152 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "SN");
3153 	if (i < 0) {
3154 		netif_err(efx, drv, efx->net_dev, "Serial number not found\n");
3155 		return;
3156 	}
3157 
3158 	j = pci_vpd_info_field_size(&vpd_data[i]);
3159 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3160 	if (i + j > vpd_size) {
3161 		netif_err(efx, drv, efx->net_dev, "Incomplete serial number\n");
3162 		return;
3163 	}
3164 
3165 	efx->vpd_sn = kmalloc(j + 1, GFP_KERNEL);
3166 	if (!efx->vpd_sn)
3167 		return;
3168 
3169 	snprintf(efx->vpd_sn, j + 1, "%s", &vpd_data[i]);
3170 }
3171 
3172 
3173 /* Main body of NIC initialisation
3174  * This is called at module load (or hotplug insertion, theoretically).
3175  */
3176 static int efx_pci_probe_main(struct efx_nic *efx)
3177 {
3178 	int rc;
3179 
3180 	/* Do start-of-day initialisation */
3181 	rc = efx_probe_all(efx);
3182 	if (rc)
3183 		goto fail1;
3184 
3185 	efx_init_napi(efx);
3186 
3187 	rc = efx->type->init(efx);
3188 	if (rc) {
3189 		netif_err(efx, probe, efx->net_dev,
3190 			  "failed to initialise NIC\n");
3191 		goto fail3;
3192 	}
3193 
3194 	rc = efx_init_port(efx);
3195 	if (rc) {
3196 		netif_err(efx, probe, efx->net_dev,
3197 			  "failed to initialise port\n");
3198 		goto fail4;
3199 	}
3200 
3201 	rc = efx_nic_init_interrupt(efx);
3202 	if (rc)
3203 		goto fail5;
3204 	rc = efx_enable_interrupts(efx);
3205 	if (rc)
3206 		goto fail6;
3207 
3208 	return 0;
3209 
3210  fail6:
3211 	efx_nic_fini_interrupt(efx);
3212  fail5:
3213 	efx_fini_port(efx);
3214  fail4:
3215 	efx->type->fini(efx);
3216  fail3:
3217 	efx_fini_napi(efx);
3218 	efx_remove_all(efx);
3219  fail1:
3220 	return rc;
3221 }
3222 
3223 static int efx_pci_probe_post_io(struct efx_nic *efx)
3224 {
3225 	struct net_device *net_dev = efx->net_dev;
3226 	int rc = efx_pci_probe_main(efx);
3227 
3228 	if (rc)
3229 		return rc;
3230 
3231 	if (efx->type->sriov_init) {
3232 		rc = efx->type->sriov_init(efx);
3233 		if (rc)
3234 			netif_err(efx, probe, efx->net_dev,
3235 				  "SR-IOV can't be enabled rc %d\n", rc);
3236 	}
3237 
3238 	/* Determine netdevice features */
3239 	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
3240 			      NETIF_F_TSO | NETIF_F_RXCSUM);
3241 	if (efx->type->offload_features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
3242 		net_dev->features |= NETIF_F_TSO6;
3243 	/* Check whether device supports TSO */
3244 	if (!efx->type->tso_versions || !efx->type->tso_versions(efx))
3245 		net_dev->features &= ~NETIF_F_ALL_TSO;
3246 	/* Mask for features that also apply to VLAN devices */
3247 	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
3248 				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
3249 				   NETIF_F_RXCSUM);
3250 
3251 	net_dev->hw_features = net_dev->features & ~efx->fixed_features;
3252 
3253 	/* Disable VLAN filtering by default.  It may be enforced if
3254 	 * the feature is fixed (i.e. VLAN filters are required to
3255 	 * receive VLAN tagged packets due to vPort restrictions).
3256 	 */
3257 	net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3258 	net_dev->features |= efx->fixed_features;
3259 
3260 	rc = efx_register_netdev(efx);
3261 	if (!rc)
3262 		return 0;
3263 
3264 	efx_pci_remove_main(efx);
3265 	return rc;
3266 }
3267 
3268 /* NIC initialisation
3269  *
3270  * This is called at module load (or hotplug insertion,
3271  * theoretically).  It sets up PCI mappings, resets the NIC,
3272  * sets up and registers the network devices with the kernel and hooks
3273  * the interrupt service routine.  It does not prepare the device for
3274  * transmission; this is left to the first time one of the network
3275  * interfaces is brought up (i.e. efx_net_open).
3276  */
3277 static int efx_pci_probe(struct pci_dev *pci_dev,
3278 			 const struct pci_device_id *entry)
3279 {
3280 	struct net_device *net_dev;
3281 	struct efx_nic *efx;
3282 	int rc;
3283 
3284 	/* Allocate and initialise a struct net_device and struct efx_nic */
3285 	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
3286 				     EFX_MAX_RX_QUEUES);
3287 	if (!net_dev)
3288 		return -ENOMEM;
3289 	efx = netdev_priv(net_dev);
3290 	efx->type = (const struct efx_nic_type *) entry->driver_data;
3291 	efx->fixed_features |= NETIF_F_HIGHDMA;
3292 
3293 	pci_set_drvdata(pci_dev, efx);
3294 	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
3295 	rc = efx_init_struct(efx, pci_dev, net_dev);
3296 	if (rc)
3297 		goto fail1;
3298 
3299 	netif_info(efx, probe, efx->net_dev,
3300 		   "Solarflare NIC detected\n");
3301 
3302 	if (!efx->type->is_vf)
3303 		efx_probe_vpd_strings(efx);
3304 
3305 	/* Set up basic I/O (BAR mappings etc) */
3306 	rc = efx_init_io(efx);
3307 	if (rc)
3308 		goto fail2;
3309 
3310 	rc = efx_pci_probe_post_io(efx);
3311 	if (rc) {
3312 		/* On failure, retry once immediately.
3313 		 * If we aborted probe due to a scheduled reset, dismiss it.
3314 		 */
3315 		efx->reset_pending = 0;
3316 		rc = efx_pci_probe_post_io(efx);
3317 		if (rc) {
3318 			/* On another failure, retry once more
3319 			 * after a 50-305ms delay.
3320 			 */
3321 			unsigned char r;
3322 
3323 			get_random_bytes(&r, 1);
3324 			msleep((unsigned int)r + 50);
3325 			efx->reset_pending = 0;
3326 			rc = efx_pci_probe_post_io(efx);
3327 		}
3328 	}
3329 	if (rc)
3330 		goto fail3;
3331 
3332 	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
3333 
3334 	/* Try to create MTDs, but allow this to fail */
3335 	rtnl_lock();
3336 	rc = efx_mtd_probe(efx);
3337 	rtnl_unlock();
3338 	if (rc && rc != -EPERM)
3339 		netif_warn(efx, probe, efx->net_dev,
3340 			   "failed to create MTDs (%d)\n", rc);
3341 
3342 	rc = pci_enable_pcie_error_reporting(pci_dev);
3343 	if (rc && rc != -EINVAL)
3344 		netif_notice(efx, probe, efx->net_dev,
3345 			     "PCIE error reporting unavailable (%d).\n",
3346 			     rc);
3347 
3348 	if (efx->type->udp_tnl_push_ports)
3349 		efx->type->udp_tnl_push_ports(efx);
3350 
3351 	return 0;
3352 
3353  fail3:
3354 	efx_fini_io(efx);
3355  fail2:
3356 	efx_fini_struct(efx);
3357  fail1:
3358 	WARN_ON(rc > 0);
3359 	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
3360 	free_netdev(net_dev);
3361 	return rc;
3362 }
3363 
3364 /* efx_pci_sriov_configure returns the actual number of Virtual Functions
3365  * enabled on success
3366  */
3367 #ifdef CONFIG_SFC_SRIOV
3368 static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
3369 {
3370 	int rc;
3371 	struct efx_nic *efx = pci_get_drvdata(dev);
3372 
3373 	if (efx->type->sriov_configure) {
3374 		rc = efx->type->sriov_configure(efx, num_vfs);
3375 		if (rc)
3376 			return rc;
3377 		else
3378 			return num_vfs;
3379 	} else
3380 		return -EOPNOTSUPP;
3381 }
3382 #endif
3383 
3384 static int efx_pm_freeze(struct device *dev)
3385 {
3386 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3387 
3388 	rtnl_lock();
3389 
3390 	if (efx->state != STATE_DISABLED) {
3391 		efx->state = STATE_UNINIT;
3392 
3393 		efx_device_detach_sync(efx);
3394 
3395 		efx_stop_all(efx);
3396 		efx_disable_interrupts(efx);
3397 	}
3398 
3399 	rtnl_unlock();
3400 
3401 	return 0;
3402 }
3403 
3404 static int efx_pm_thaw(struct device *dev)
3405 {
3406 	int rc;
3407 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3408 
3409 	rtnl_lock();
3410 
3411 	if (efx->state != STATE_DISABLED) {
3412 		rc = efx_enable_interrupts(efx);
3413 		if (rc)
3414 			goto fail;
3415 
3416 		mutex_lock(&efx->mac_lock);
3417 		efx->phy_op->reconfigure(efx);
3418 		mutex_unlock(&efx->mac_lock);
3419 
3420 		efx_start_all(efx);
3421 
3422 		efx_device_attach_if_not_resetting(efx);
3423 
3424 		efx->state = STATE_READY;
3425 
3426 		efx->type->resume_wol(efx);
3427 	}
3428 
3429 	rtnl_unlock();
3430 
3431 	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
3432 	queue_work(reset_workqueue, &efx->reset_work);
3433 
3434 	return 0;
3435 
3436 fail:
3437 	rtnl_unlock();
3438 
3439 	return rc;
3440 }
3441 
3442 static int efx_pm_poweroff(struct device *dev)
3443 {
3444 	struct pci_dev *pci_dev = to_pci_dev(dev);
3445 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3446 
3447 	efx->type->fini(efx);
3448 
3449 	efx->reset_pending = 0;
3450 
3451 	pci_save_state(pci_dev);
3452 	return pci_set_power_state(pci_dev, PCI_D3hot);
3453 }
3454 
3455 /* Used for both resume and restore */
3456 static int efx_pm_resume(struct device *dev)
3457 {
3458 	struct pci_dev *pci_dev = to_pci_dev(dev);
3459 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3460 	int rc;
3461 
3462 	rc = pci_set_power_state(pci_dev, PCI_D0);
3463 	if (rc)
3464 		return rc;
3465 	pci_restore_state(pci_dev);
3466 	rc = pci_enable_device(pci_dev);
3467 	if (rc)
3468 		return rc;
3469 	pci_set_master(efx->pci_dev);
3470 	rc = efx->type->reset(efx, RESET_TYPE_ALL);
3471 	if (rc)
3472 		return rc;
3473 	rc = efx->type->init(efx);
3474 	if (rc)
3475 		return rc;
3476 	rc = efx_pm_thaw(dev);
3477 	return rc;
3478 }
3479 
3480 static int efx_pm_suspend(struct device *dev)
3481 {
3482 	int rc;
3483 
3484 	efx_pm_freeze(dev);
3485 	rc = efx_pm_poweroff(dev);
3486 	if (rc)
3487 		efx_pm_resume(dev);
3488 	return rc;
3489 }
3490 
3491 static const struct dev_pm_ops efx_pm_ops = {
3492 	.suspend	= efx_pm_suspend,
3493 	.resume		= efx_pm_resume,
3494 	.freeze		= efx_pm_freeze,
3495 	.thaw		= efx_pm_thaw,
3496 	.poweroff	= efx_pm_poweroff,
3497 	.restore	= efx_pm_resume,
3498 };
3499 
3500 /* A PCI error affecting this device was detected.
3501  * At this point MMIO and DMA may be disabled.
3502  * Stop the software path and request a slot reset.
3503  */
3504 static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
3505 					      enum pci_channel_state state)
3506 {
3507 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3508 	struct efx_nic *efx = pci_get_drvdata(pdev);
3509 
3510 	if (state == pci_channel_io_perm_failure)
3511 		return PCI_ERS_RESULT_DISCONNECT;
3512 
3513 	rtnl_lock();
3514 
3515 	if (efx->state != STATE_DISABLED) {
3516 		efx->state = STATE_RECOVERY;
3517 		efx->reset_pending = 0;
3518 
3519 		efx_device_detach_sync(efx);
3520 
3521 		efx_stop_all(efx);
3522 		efx_disable_interrupts(efx);
3523 
3524 		status = PCI_ERS_RESULT_NEED_RESET;
3525 	} else {
3526 		/* If the interface is disabled we don't want to do anything
3527 		 * with it.
3528 		 */
3529 		status = PCI_ERS_RESULT_RECOVERED;
3530 	}
3531 
3532 	rtnl_unlock();
3533 
3534 	pci_disable_device(pdev);
3535 
3536 	return status;
3537 }
3538 
3539 /* Fake a successful reset, which will be performed later in efx_io_resume. */
3540 static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
3541 {
3542 	struct efx_nic *efx = pci_get_drvdata(pdev);
3543 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3544 	int rc;
3545 
3546 	if (pci_enable_device(pdev)) {
3547 		netif_err(efx, hw, efx->net_dev,
3548 			  "Cannot re-enable PCI device after reset.\n");
3549 		status =  PCI_ERS_RESULT_DISCONNECT;
3550 	}
3551 
3552 	rc = pci_cleanup_aer_uncorrect_error_status(pdev);
3553 	if (rc) {
3554 		netif_err(efx, hw, efx->net_dev,
3555 		"pci_cleanup_aer_uncorrect_error_status failed (%d)\n", rc);
3556 		/* Non-fatal error. Continue. */
3557 	}
3558 
3559 	return status;
3560 }
3561 
3562 /* Perform the actual reset and resume I/O operations. */
3563 static void efx_io_resume(struct pci_dev *pdev)
3564 {
3565 	struct efx_nic *efx = pci_get_drvdata(pdev);
3566 	int rc;
3567 
3568 	rtnl_lock();
3569 
3570 	if (efx->state == STATE_DISABLED)
3571 		goto out;
3572 
3573 	rc = efx_reset(efx, RESET_TYPE_ALL);
3574 	if (rc) {
3575 		netif_err(efx, hw, efx->net_dev,
3576 			  "efx_reset failed after PCI error (%d)\n", rc);
3577 	} else {
3578 		efx->state = STATE_READY;
3579 		netif_dbg(efx, hw, efx->net_dev,
3580 			  "Done resetting and resuming IO after PCI error.\n");
3581 	}
3582 
3583 out:
3584 	rtnl_unlock();
3585 }
3586 
3587 /* For simplicity and reliability, we always require a slot reset and try to
3588  * reset the hardware when a pci error affecting the device is detected.
3589  * We leave both the link_reset and mmio_enabled callback unimplemented:
3590  * with our request for slot reset the mmio_enabled callback will never be
3591  * called, and the link_reset callback is not used by AER or EEH mechanisms.
3592  */
3593 static const struct pci_error_handlers efx_err_handlers = {
3594 	.error_detected = efx_io_error_detected,
3595 	.slot_reset	= efx_io_slot_reset,
3596 	.resume		= efx_io_resume,
3597 };
3598 
3599 static struct pci_driver efx_pci_driver = {
3600 	.name		= KBUILD_MODNAME,
3601 	.id_table	= efx_pci_table,
3602 	.probe		= efx_pci_probe,
3603 	.remove		= efx_pci_remove,
3604 	.driver.pm	= &efx_pm_ops,
3605 	.err_handler	= &efx_err_handlers,
3606 #ifdef CONFIG_SFC_SRIOV
3607 	.sriov_configure = efx_pci_sriov_configure,
3608 #endif
3609 };
3610 
3611 /**************************************************************************
3612  *
3613  * Kernel module interface
3614  *
3615  *************************************************************************/
3616 
3617 module_param(interrupt_mode, uint, 0444);
3618 MODULE_PARM_DESC(interrupt_mode,
3619 		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
3620 
3621 static int __init efx_init_module(void)
3622 {
3623 	int rc;
3624 
3625 	printk(KERN_INFO "Solarflare NET driver v" EFX_DRIVER_VERSION "\n");
3626 
3627 	rc = register_netdevice_notifier(&efx_netdev_notifier);
3628 	if (rc)
3629 		goto err_notifier;
3630 
3631 #ifdef CONFIG_SFC_SRIOV
3632 	rc = efx_init_sriov();
3633 	if (rc)
3634 		goto err_sriov;
3635 #endif
3636 
3637 	reset_workqueue = create_singlethread_workqueue("sfc_reset");
3638 	if (!reset_workqueue) {
3639 		rc = -ENOMEM;
3640 		goto err_reset;
3641 	}
3642 
3643 	rc = pci_register_driver(&efx_pci_driver);
3644 	if (rc < 0)
3645 		goto err_pci;
3646 
3647 	return 0;
3648 
3649  err_pci:
3650 	destroy_workqueue(reset_workqueue);
3651  err_reset:
3652 #ifdef CONFIG_SFC_SRIOV
3653 	efx_fini_sriov();
3654  err_sriov:
3655 #endif
3656 	unregister_netdevice_notifier(&efx_netdev_notifier);
3657  err_notifier:
3658 	return rc;
3659 }
3660 
3661 static void __exit efx_exit_module(void)
3662 {
3663 	printk(KERN_INFO "Solarflare NET driver unloading\n");
3664 
3665 	pci_unregister_driver(&efx_pci_driver);
3666 	destroy_workqueue(reset_workqueue);
3667 #ifdef CONFIG_SFC_SRIOV
3668 	efx_fini_sriov();
3669 #endif
3670 	unregister_netdevice_notifier(&efx_netdev_notifier);
3671 
3672 }
3673 
3674 module_init(efx_init_module);
3675 module_exit(efx_exit_module);
3676 
3677 MODULE_AUTHOR("Solarflare Communications and "
3678 	      "Michael Brown <mbrown@fensystems.co.uk>");
3679 MODULE_DESCRIPTION("Solarflare network driver");
3680 MODULE_LICENSE("GPL");
3681 MODULE_DEVICE_TABLE(pci, efx_pci_table);
3682 MODULE_VERSION(EFX_DRIVER_VERSION);
3683