xref: /openbmc/linux/drivers/net/ethernet/sfc/efx.c (revision dea54fba)
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 (count > EFX_MAX_RX_QUEUES) {
1375 		netif_cond_dbg(efx, probe, efx->net_dev, !rss_cpus, warn,
1376 			       "Reducing number of rx queues from %u to %u.\n",
1377 			       count, EFX_MAX_RX_QUEUES);
1378 		count = EFX_MAX_RX_QUEUES;
1379 	}
1380 
1381 	/* If RSS is requested for the PF *and* VFs then we can't write RSS
1382 	 * table entries that are inaccessible to VFs
1383 	 */
1384 #ifdef CONFIG_SFC_SRIOV
1385 	if (efx->type->sriov_wanted) {
1386 		if (efx->type->sriov_wanted(efx) && efx_vf_size(efx) > 1 &&
1387 		    count > efx_vf_size(efx)) {
1388 			netif_warn(efx, probe, efx->net_dev,
1389 				   "Reducing number of RSS channels from %u to %u for "
1390 				   "VF support. Increase vf-msix-limit to use more "
1391 				   "channels on the PF.\n",
1392 				   count, efx_vf_size(efx));
1393 			count = efx_vf_size(efx);
1394 		}
1395 	}
1396 #endif
1397 
1398 	return count;
1399 }
1400 
1401 /* Probe the number and type of interrupts we are able to obtain, and
1402  * the resulting numbers of channels and RX queues.
1403  */
1404 static int efx_probe_interrupts(struct efx_nic *efx)
1405 {
1406 	unsigned int extra_channels = 0;
1407 	unsigned int i, j;
1408 	int rc;
1409 
1410 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
1411 		if (efx->extra_channel_type[i])
1412 			++extra_channels;
1413 
1414 	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1415 		struct msix_entry xentries[EFX_MAX_CHANNELS];
1416 		unsigned int n_channels;
1417 
1418 		n_channels = efx_wanted_parallelism(efx);
1419 		if (efx_separate_tx_channels)
1420 			n_channels *= 2;
1421 		n_channels += extra_channels;
1422 		n_channels = min(n_channels, efx->max_channels);
1423 
1424 		for (i = 0; i < n_channels; i++)
1425 			xentries[i].entry = i;
1426 		rc = pci_enable_msix_range(efx->pci_dev,
1427 					   xentries, 1, n_channels);
1428 		if (rc < 0) {
1429 			/* Fall back to single channel MSI */
1430 			netif_err(efx, drv, efx->net_dev,
1431 				  "could not enable MSI-X\n");
1432 			if (efx->type->min_interrupt_mode >= EFX_INT_MODE_MSI)
1433 				efx->interrupt_mode = EFX_INT_MODE_MSI;
1434 			else
1435 				return rc;
1436 		} else if (rc < n_channels) {
1437 			netif_err(efx, drv, efx->net_dev,
1438 				  "WARNING: Insufficient MSI-X vectors"
1439 				  " available (%d < %u).\n", rc, n_channels);
1440 			netif_err(efx, drv, efx->net_dev,
1441 				  "WARNING: Performance may be reduced.\n");
1442 			n_channels = rc;
1443 		}
1444 
1445 		if (rc > 0) {
1446 			efx->n_channels = n_channels;
1447 			if (n_channels > extra_channels)
1448 				n_channels -= extra_channels;
1449 			if (efx_separate_tx_channels) {
1450 				efx->n_tx_channels = min(max(n_channels / 2,
1451 							     1U),
1452 							 efx->max_tx_channels);
1453 				efx->n_rx_channels = max(n_channels -
1454 							 efx->n_tx_channels,
1455 							 1U);
1456 			} else {
1457 				efx->n_tx_channels = min(n_channels,
1458 							 efx->max_tx_channels);
1459 				efx->n_rx_channels = n_channels;
1460 			}
1461 			for (i = 0; i < efx->n_channels; i++)
1462 				efx_get_channel(efx, i)->irq =
1463 					xentries[i].vector;
1464 		}
1465 	}
1466 
1467 	/* Try single interrupt MSI */
1468 	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1469 		efx->n_channels = 1;
1470 		efx->n_rx_channels = 1;
1471 		efx->n_tx_channels = 1;
1472 		rc = pci_enable_msi(efx->pci_dev);
1473 		if (rc == 0) {
1474 			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1475 		} else {
1476 			netif_err(efx, drv, efx->net_dev,
1477 				  "could not enable MSI\n");
1478 			if (efx->type->min_interrupt_mode >= EFX_INT_MODE_LEGACY)
1479 				efx->interrupt_mode = EFX_INT_MODE_LEGACY;
1480 			else
1481 				return rc;
1482 		}
1483 	}
1484 
1485 	/* Assume legacy interrupts */
1486 	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1487 		efx->n_channels = 1 + (efx_separate_tx_channels ? 1 : 0);
1488 		efx->n_rx_channels = 1;
1489 		efx->n_tx_channels = 1;
1490 		efx->legacy_irq = efx->pci_dev->irq;
1491 	}
1492 
1493 	/* Assign extra channels if possible */
1494 	j = efx->n_channels;
1495 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++) {
1496 		if (!efx->extra_channel_type[i])
1497 			continue;
1498 		if (efx->interrupt_mode != EFX_INT_MODE_MSIX ||
1499 		    efx->n_channels <= extra_channels) {
1500 			efx->extra_channel_type[i]->handle_no_channel(efx);
1501 		} else {
1502 			--j;
1503 			efx_get_channel(efx, j)->type =
1504 				efx->extra_channel_type[i];
1505 		}
1506 	}
1507 
1508 	/* RSS might be usable on VFs even if it is disabled on the PF */
1509 #ifdef CONFIG_SFC_SRIOV
1510 	if (efx->type->sriov_wanted) {
1511 		efx->rss_spread = ((efx->n_rx_channels > 1 ||
1512 				    !efx->type->sriov_wanted(efx)) ?
1513 				   efx->n_rx_channels : efx_vf_size(efx));
1514 		return 0;
1515 	}
1516 #endif
1517 	efx->rss_spread = efx->n_rx_channels;
1518 
1519 	return 0;
1520 }
1521 
1522 static int efx_soft_enable_interrupts(struct efx_nic *efx)
1523 {
1524 	struct efx_channel *channel, *end_channel;
1525 	int rc;
1526 
1527 	BUG_ON(efx->state == STATE_DISABLED);
1528 
1529 	efx->irq_soft_enabled = true;
1530 	smp_wmb();
1531 
1532 	efx_for_each_channel(channel, efx) {
1533 		if (!channel->type->keep_eventq) {
1534 			rc = efx_init_eventq(channel);
1535 			if (rc)
1536 				goto fail;
1537 		}
1538 		efx_start_eventq(channel);
1539 	}
1540 
1541 	efx_mcdi_mode_event(efx);
1542 
1543 	return 0;
1544 fail:
1545 	end_channel = channel;
1546 	efx_for_each_channel(channel, efx) {
1547 		if (channel == end_channel)
1548 			break;
1549 		efx_stop_eventq(channel);
1550 		if (!channel->type->keep_eventq)
1551 			efx_fini_eventq(channel);
1552 	}
1553 
1554 	return rc;
1555 }
1556 
1557 static void efx_soft_disable_interrupts(struct efx_nic *efx)
1558 {
1559 	struct efx_channel *channel;
1560 
1561 	if (efx->state == STATE_DISABLED)
1562 		return;
1563 
1564 	efx_mcdi_mode_poll(efx);
1565 
1566 	efx->irq_soft_enabled = false;
1567 	smp_wmb();
1568 
1569 	if (efx->legacy_irq)
1570 		synchronize_irq(efx->legacy_irq);
1571 
1572 	efx_for_each_channel(channel, efx) {
1573 		if (channel->irq)
1574 			synchronize_irq(channel->irq);
1575 
1576 		efx_stop_eventq(channel);
1577 		if (!channel->type->keep_eventq)
1578 			efx_fini_eventq(channel);
1579 	}
1580 
1581 	/* Flush the asynchronous MCDI request queue */
1582 	efx_mcdi_flush_async(efx);
1583 }
1584 
1585 static int efx_enable_interrupts(struct efx_nic *efx)
1586 {
1587 	struct efx_channel *channel, *end_channel;
1588 	int rc;
1589 
1590 	BUG_ON(efx->state == STATE_DISABLED);
1591 
1592 	if (efx->eeh_disabled_legacy_irq) {
1593 		enable_irq(efx->legacy_irq);
1594 		efx->eeh_disabled_legacy_irq = false;
1595 	}
1596 
1597 	efx->type->irq_enable_master(efx);
1598 
1599 	efx_for_each_channel(channel, efx) {
1600 		if (channel->type->keep_eventq) {
1601 			rc = efx_init_eventq(channel);
1602 			if (rc)
1603 				goto fail;
1604 		}
1605 	}
1606 
1607 	rc = efx_soft_enable_interrupts(efx);
1608 	if (rc)
1609 		goto fail;
1610 
1611 	return 0;
1612 
1613 fail:
1614 	end_channel = channel;
1615 	efx_for_each_channel(channel, efx) {
1616 		if (channel == end_channel)
1617 			break;
1618 		if (channel->type->keep_eventq)
1619 			efx_fini_eventq(channel);
1620 	}
1621 
1622 	efx->type->irq_disable_non_ev(efx);
1623 
1624 	return rc;
1625 }
1626 
1627 static void efx_disable_interrupts(struct efx_nic *efx)
1628 {
1629 	struct efx_channel *channel;
1630 
1631 	efx_soft_disable_interrupts(efx);
1632 
1633 	efx_for_each_channel(channel, efx) {
1634 		if (channel->type->keep_eventq)
1635 			efx_fini_eventq(channel);
1636 	}
1637 
1638 	efx->type->irq_disable_non_ev(efx);
1639 }
1640 
1641 static void efx_remove_interrupts(struct efx_nic *efx)
1642 {
1643 	struct efx_channel *channel;
1644 
1645 	/* Remove MSI/MSI-X interrupts */
1646 	efx_for_each_channel(channel, efx)
1647 		channel->irq = 0;
1648 	pci_disable_msi(efx->pci_dev);
1649 	pci_disable_msix(efx->pci_dev);
1650 
1651 	/* Remove legacy interrupt */
1652 	efx->legacy_irq = 0;
1653 }
1654 
1655 static void efx_set_channels(struct efx_nic *efx)
1656 {
1657 	struct efx_channel *channel;
1658 	struct efx_tx_queue *tx_queue;
1659 
1660 	efx->tx_channel_offset =
1661 		efx_separate_tx_channels ?
1662 		efx->n_channels - efx->n_tx_channels : 0;
1663 
1664 	/* We need to mark which channels really have RX and TX
1665 	 * queues, and adjust the TX queue numbers if we have separate
1666 	 * RX-only and TX-only channels.
1667 	 */
1668 	efx_for_each_channel(channel, efx) {
1669 		if (channel->channel < efx->n_rx_channels)
1670 			channel->rx_queue.core_index = channel->channel;
1671 		else
1672 			channel->rx_queue.core_index = -1;
1673 
1674 		efx_for_each_channel_tx_queue(tx_queue, channel)
1675 			tx_queue->queue -= (efx->tx_channel_offset *
1676 					    EFX_TXQ_TYPES);
1677 	}
1678 }
1679 
1680 static int efx_probe_nic(struct efx_nic *efx)
1681 {
1682 	int rc;
1683 
1684 	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1685 
1686 	/* Carry out hardware-type specific initialisation */
1687 	rc = efx->type->probe(efx);
1688 	if (rc)
1689 		return rc;
1690 
1691 	do {
1692 		if (!efx->max_channels || !efx->max_tx_channels) {
1693 			netif_err(efx, drv, efx->net_dev,
1694 				  "Insufficient resources to allocate"
1695 				  " any channels\n");
1696 			rc = -ENOSPC;
1697 			goto fail1;
1698 		}
1699 
1700 		/* Determine the number of channels and queues by trying
1701 		 * to hook in MSI-X interrupts.
1702 		 */
1703 		rc = efx_probe_interrupts(efx);
1704 		if (rc)
1705 			goto fail1;
1706 
1707 		efx_set_channels(efx);
1708 
1709 		/* dimension_resources can fail with EAGAIN */
1710 		rc = efx->type->dimension_resources(efx);
1711 		if (rc != 0 && rc != -EAGAIN)
1712 			goto fail2;
1713 
1714 		if (rc == -EAGAIN)
1715 			/* try again with new max_channels */
1716 			efx_remove_interrupts(efx);
1717 
1718 	} while (rc == -EAGAIN);
1719 
1720 	if (efx->n_channels > 1)
1721 		netdev_rss_key_fill(&efx->rx_hash_key,
1722 				    sizeof(efx->rx_hash_key));
1723 	efx_set_default_rx_indir_table(efx);
1724 
1725 	netif_set_real_num_tx_queues(efx->net_dev, efx->n_tx_channels);
1726 	netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels);
1727 
1728 	/* Initialise the interrupt moderation settings */
1729 	efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000);
1730 	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
1731 				true);
1732 
1733 	return 0;
1734 
1735 fail2:
1736 	efx_remove_interrupts(efx);
1737 fail1:
1738 	efx->type->remove(efx);
1739 	return rc;
1740 }
1741 
1742 static void efx_remove_nic(struct efx_nic *efx)
1743 {
1744 	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1745 
1746 	efx_remove_interrupts(efx);
1747 	efx->type->remove(efx);
1748 }
1749 
1750 static int efx_probe_filters(struct efx_nic *efx)
1751 {
1752 	int rc;
1753 
1754 	spin_lock_init(&efx->filter_lock);
1755 	init_rwsem(&efx->filter_sem);
1756 	mutex_lock(&efx->mac_lock);
1757 	down_write(&efx->filter_sem);
1758 	rc = efx->type->filter_table_probe(efx);
1759 	if (rc)
1760 		goto out_unlock;
1761 
1762 #ifdef CONFIG_RFS_ACCEL
1763 	if (efx->type->offload_features & NETIF_F_NTUPLE) {
1764 		struct efx_channel *channel;
1765 		int i, success = 1;
1766 
1767 		efx_for_each_channel(channel, efx) {
1768 			channel->rps_flow_id =
1769 				kcalloc(efx->type->max_rx_ip_filters,
1770 					sizeof(*channel->rps_flow_id),
1771 					GFP_KERNEL);
1772 			if (!channel->rps_flow_id)
1773 				success = 0;
1774 			else
1775 				for (i = 0;
1776 				     i < efx->type->max_rx_ip_filters;
1777 				     ++i)
1778 					channel->rps_flow_id[i] =
1779 						RPS_FLOW_ID_INVALID;
1780 		}
1781 
1782 		if (!success) {
1783 			efx_for_each_channel(channel, efx)
1784 				kfree(channel->rps_flow_id);
1785 			efx->type->filter_table_remove(efx);
1786 			rc = -ENOMEM;
1787 			goto out_unlock;
1788 		}
1789 
1790 		efx->rps_expire_index = efx->rps_expire_channel = 0;
1791 	}
1792 #endif
1793 out_unlock:
1794 	up_write(&efx->filter_sem);
1795 	mutex_unlock(&efx->mac_lock);
1796 	return rc;
1797 }
1798 
1799 static void efx_remove_filters(struct efx_nic *efx)
1800 {
1801 #ifdef CONFIG_RFS_ACCEL
1802 	struct efx_channel *channel;
1803 
1804 	efx_for_each_channel(channel, efx)
1805 		kfree(channel->rps_flow_id);
1806 #endif
1807 	down_write(&efx->filter_sem);
1808 	efx->type->filter_table_remove(efx);
1809 	up_write(&efx->filter_sem);
1810 }
1811 
1812 static void efx_restore_filters(struct efx_nic *efx)
1813 {
1814 	down_read(&efx->filter_sem);
1815 	efx->type->filter_table_restore(efx);
1816 	up_read(&efx->filter_sem);
1817 }
1818 
1819 /**************************************************************************
1820  *
1821  * NIC startup/shutdown
1822  *
1823  *************************************************************************/
1824 
1825 static int efx_probe_all(struct efx_nic *efx)
1826 {
1827 	int rc;
1828 
1829 	rc = efx_probe_nic(efx);
1830 	if (rc) {
1831 		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1832 		goto fail1;
1833 	}
1834 
1835 	rc = efx_probe_port(efx);
1836 	if (rc) {
1837 		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1838 		goto fail2;
1839 	}
1840 
1841 	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
1842 	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
1843 		rc = -EINVAL;
1844 		goto fail3;
1845 	}
1846 	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1847 
1848 #ifdef CONFIG_SFC_SRIOV
1849 	rc = efx->type->vswitching_probe(efx);
1850 	if (rc) /* not fatal; the PF will still work fine */
1851 		netif_warn(efx, probe, efx->net_dev,
1852 			   "failed to setup vswitching rc=%d;"
1853 			   " VFs may not function\n", rc);
1854 #endif
1855 
1856 	rc = efx_probe_filters(efx);
1857 	if (rc) {
1858 		netif_err(efx, probe, efx->net_dev,
1859 			  "failed to create filter tables\n");
1860 		goto fail4;
1861 	}
1862 
1863 	rc = efx_probe_channels(efx);
1864 	if (rc)
1865 		goto fail5;
1866 
1867 	return 0;
1868 
1869  fail5:
1870 	efx_remove_filters(efx);
1871  fail4:
1872 #ifdef CONFIG_SFC_SRIOV
1873 	efx->type->vswitching_remove(efx);
1874 #endif
1875  fail3:
1876 	efx_remove_port(efx);
1877  fail2:
1878 	efx_remove_nic(efx);
1879  fail1:
1880 	return rc;
1881 }
1882 
1883 /* If the interface is supposed to be running but is not, start
1884  * the hardware and software data path, regular activity for the port
1885  * (MAC statistics, link polling, etc.) and schedule the port to be
1886  * reconfigured.  Interrupts must already be enabled.  This function
1887  * is safe to call multiple times, so long as the NIC is not disabled.
1888  * Requires the RTNL lock.
1889  */
1890 static void efx_start_all(struct efx_nic *efx)
1891 {
1892 	EFX_ASSERT_RESET_SERIALISED(efx);
1893 	BUG_ON(efx->state == STATE_DISABLED);
1894 
1895 	/* Check that it is appropriate to restart the interface. All
1896 	 * of these flags are safe to read under just the rtnl lock */
1897 	if (efx->port_enabled || !netif_running(efx->net_dev) ||
1898 	    efx->reset_pending)
1899 		return;
1900 
1901 	efx_start_port(efx);
1902 	efx_start_datapath(efx);
1903 
1904 	/* Start the hardware monitor if there is one */
1905 	if (efx->type->monitor != NULL)
1906 		queue_delayed_work(efx->workqueue, &efx->monitor_work,
1907 				   efx_monitor_interval);
1908 
1909 	/* Link state detection is normally event-driven; we have
1910 	 * to poll now because we could have missed a change
1911 	 */
1912 	mutex_lock(&efx->mac_lock);
1913 	if (efx->phy_op->poll(efx))
1914 		efx_link_status_changed(efx);
1915 	mutex_unlock(&efx->mac_lock);
1916 
1917 	efx->type->start_stats(efx);
1918 	efx->type->pull_stats(efx);
1919 	spin_lock_bh(&efx->stats_lock);
1920 	efx->type->update_stats(efx, NULL, NULL);
1921 	spin_unlock_bh(&efx->stats_lock);
1922 }
1923 
1924 /* Quiesce the hardware and software data path, and regular activity
1925  * for the port without bringing the link down.  Safe to call multiple
1926  * times with the NIC in almost any state, but interrupts should be
1927  * enabled.  Requires the RTNL lock.
1928  */
1929 static void efx_stop_all(struct efx_nic *efx)
1930 {
1931 	EFX_ASSERT_RESET_SERIALISED(efx);
1932 
1933 	/* port_enabled can be read safely under the rtnl lock */
1934 	if (!efx->port_enabled)
1935 		return;
1936 
1937 	/* update stats before we go down so we can accurately count
1938 	 * rx_nodesc_drops
1939 	 */
1940 	efx->type->pull_stats(efx);
1941 	spin_lock_bh(&efx->stats_lock);
1942 	efx->type->update_stats(efx, NULL, NULL);
1943 	spin_unlock_bh(&efx->stats_lock);
1944 	efx->type->stop_stats(efx);
1945 	efx_stop_port(efx);
1946 
1947 	/* Stop the kernel transmit interface.  This is only valid if
1948 	 * the device is stopped or detached; otherwise the watchdog
1949 	 * may fire immediately.
1950 	 */
1951 	WARN_ON(netif_running(efx->net_dev) &&
1952 		netif_device_present(efx->net_dev));
1953 	netif_tx_disable(efx->net_dev);
1954 
1955 	efx_stop_datapath(efx);
1956 }
1957 
1958 static void efx_remove_all(struct efx_nic *efx)
1959 {
1960 	efx_remove_channels(efx);
1961 	efx_remove_filters(efx);
1962 #ifdef CONFIG_SFC_SRIOV
1963 	efx->type->vswitching_remove(efx);
1964 #endif
1965 	efx_remove_port(efx);
1966 	efx_remove_nic(efx);
1967 }
1968 
1969 /**************************************************************************
1970  *
1971  * Interrupt moderation
1972  *
1973  **************************************************************************/
1974 unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs)
1975 {
1976 	if (usecs == 0)
1977 		return 0;
1978 	if (usecs * 1000 < efx->timer_quantum_ns)
1979 		return 1; /* never round down to 0 */
1980 	return usecs * 1000 / efx->timer_quantum_ns;
1981 }
1982 
1983 unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks)
1984 {
1985 	/* We must round up when converting ticks to microseconds
1986 	 * because we round down when converting the other way.
1987 	 */
1988 	return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000);
1989 }
1990 
1991 /* Set interrupt moderation parameters */
1992 int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
1993 			    unsigned int rx_usecs, bool rx_adaptive,
1994 			    bool rx_may_override_tx)
1995 {
1996 	struct efx_channel *channel;
1997 	unsigned int timer_max_us;
1998 
1999 	EFX_ASSERT_RESET_SERIALISED(efx);
2000 
2001 	timer_max_us = efx->timer_max_ns / 1000;
2002 
2003 	if (tx_usecs > timer_max_us || rx_usecs > timer_max_us)
2004 		return -EINVAL;
2005 
2006 	if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 &&
2007 	    !rx_may_override_tx) {
2008 		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
2009 			  "RX and TX IRQ moderation must be equal\n");
2010 		return -EINVAL;
2011 	}
2012 
2013 	efx->irq_rx_adaptive = rx_adaptive;
2014 	efx->irq_rx_moderation_us = rx_usecs;
2015 	efx_for_each_channel(channel, efx) {
2016 		if (efx_channel_has_rx_queue(channel))
2017 			channel->irq_moderation_us = rx_usecs;
2018 		else if (efx_channel_has_tx_queues(channel))
2019 			channel->irq_moderation_us = tx_usecs;
2020 	}
2021 
2022 	return 0;
2023 }
2024 
2025 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
2026 			    unsigned int *rx_usecs, bool *rx_adaptive)
2027 {
2028 	*rx_adaptive = efx->irq_rx_adaptive;
2029 	*rx_usecs = efx->irq_rx_moderation_us;
2030 
2031 	/* If channels are shared between RX and TX, so is IRQ
2032 	 * moderation.  Otherwise, IRQ moderation is the same for all
2033 	 * TX channels and is not adaptive.
2034 	 */
2035 	if (efx->tx_channel_offset == 0) {
2036 		*tx_usecs = *rx_usecs;
2037 	} else {
2038 		struct efx_channel *tx_channel;
2039 
2040 		tx_channel = efx->channel[efx->tx_channel_offset];
2041 		*tx_usecs = tx_channel->irq_moderation_us;
2042 	}
2043 }
2044 
2045 /**************************************************************************
2046  *
2047  * Hardware monitor
2048  *
2049  **************************************************************************/
2050 
2051 /* Run periodically off the general workqueue */
2052 static void efx_monitor(struct work_struct *data)
2053 {
2054 	struct efx_nic *efx = container_of(data, struct efx_nic,
2055 					   monitor_work.work);
2056 
2057 	netif_vdbg(efx, timer, efx->net_dev,
2058 		   "hardware monitor executing on CPU %d\n",
2059 		   raw_smp_processor_id());
2060 	BUG_ON(efx->type->monitor == NULL);
2061 
2062 	/* If the mac_lock is already held then it is likely a port
2063 	 * reconfiguration is already in place, which will likely do
2064 	 * most of the work of monitor() anyway. */
2065 	if (mutex_trylock(&efx->mac_lock)) {
2066 		if (efx->port_enabled)
2067 			efx->type->monitor(efx);
2068 		mutex_unlock(&efx->mac_lock);
2069 	}
2070 
2071 	queue_delayed_work(efx->workqueue, &efx->monitor_work,
2072 			   efx_monitor_interval);
2073 }
2074 
2075 /**************************************************************************
2076  *
2077  * ioctls
2078  *
2079  *************************************************************************/
2080 
2081 /* Net device ioctl
2082  * Context: process, rtnl_lock() held.
2083  */
2084 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
2085 {
2086 	struct efx_nic *efx = netdev_priv(net_dev);
2087 	struct mii_ioctl_data *data = if_mii(ifr);
2088 
2089 	if (cmd == SIOCSHWTSTAMP)
2090 		return efx_ptp_set_ts_config(efx, ifr);
2091 	if (cmd == SIOCGHWTSTAMP)
2092 		return efx_ptp_get_ts_config(efx, ifr);
2093 
2094 	/* Convert phy_id from older PRTAD/DEVAD format */
2095 	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
2096 	    (data->phy_id & 0xfc00) == 0x0400)
2097 		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;
2098 
2099 	return mdio_mii_ioctl(&efx->mdio, data, cmd);
2100 }
2101 
2102 /**************************************************************************
2103  *
2104  * NAPI interface
2105  *
2106  **************************************************************************/
2107 
2108 static void efx_init_napi_channel(struct efx_channel *channel)
2109 {
2110 	struct efx_nic *efx = channel->efx;
2111 
2112 	channel->napi_dev = efx->net_dev;
2113 	netif_napi_add(channel->napi_dev, &channel->napi_str,
2114 		       efx_poll, napi_weight);
2115 }
2116 
2117 static void efx_init_napi(struct efx_nic *efx)
2118 {
2119 	struct efx_channel *channel;
2120 
2121 	efx_for_each_channel(channel, efx)
2122 		efx_init_napi_channel(channel);
2123 }
2124 
2125 static void efx_fini_napi_channel(struct efx_channel *channel)
2126 {
2127 	if (channel->napi_dev)
2128 		netif_napi_del(&channel->napi_str);
2129 
2130 	channel->napi_dev = NULL;
2131 }
2132 
2133 static void efx_fini_napi(struct efx_nic *efx)
2134 {
2135 	struct efx_channel *channel;
2136 
2137 	efx_for_each_channel(channel, efx)
2138 		efx_fini_napi_channel(channel);
2139 }
2140 
2141 /**************************************************************************
2142  *
2143  * Kernel netpoll interface
2144  *
2145  *************************************************************************/
2146 
2147 #ifdef CONFIG_NET_POLL_CONTROLLER
2148 
2149 /* Although in the common case interrupts will be disabled, this is not
2150  * guaranteed. However, all our work happens inside the NAPI callback,
2151  * so no locking is required.
2152  */
2153 static void efx_netpoll(struct net_device *net_dev)
2154 {
2155 	struct efx_nic *efx = netdev_priv(net_dev);
2156 	struct efx_channel *channel;
2157 
2158 	efx_for_each_channel(channel, efx)
2159 		efx_schedule_channel(channel);
2160 }
2161 
2162 #endif
2163 
2164 /**************************************************************************
2165  *
2166  * Kernel net device interface
2167  *
2168  *************************************************************************/
2169 
2170 /* Context: process, rtnl_lock() held. */
2171 int efx_net_open(struct net_device *net_dev)
2172 {
2173 	struct efx_nic *efx = netdev_priv(net_dev);
2174 	int rc;
2175 
2176 	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
2177 		  raw_smp_processor_id());
2178 
2179 	rc = efx_check_disabled(efx);
2180 	if (rc)
2181 		return rc;
2182 	if (efx->phy_mode & PHY_MODE_SPECIAL)
2183 		return -EBUSY;
2184 	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
2185 		return -EIO;
2186 
2187 	/* Notify the kernel of the link state polled during driver load,
2188 	 * before the monitor starts running */
2189 	efx_link_status_changed(efx);
2190 
2191 	efx_start_all(efx);
2192 	if (efx->state == STATE_DISABLED || efx->reset_pending)
2193 		netif_device_detach(efx->net_dev);
2194 	efx_selftest_async_start(efx);
2195 	return 0;
2196 }
2197 
2198 /* Context: process, rtnl_lock() held.
2199  * Note that the kernel will ignore our return code; this method
2200  * should really be a void.
2201  */
2202 int efx_net_stop(struct net_device *net_dev)
2203 {
2204 	struct efx_nic *efx = netdev_priv(net_dev);
2205 
2206 	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
2207 		  raw_smp_processor_id());
2208 
2209 	/* Stop the device and flush all the channels */
2210 	efx_stop_all(efx);
2211 
2212 	return 0;
2213 }
2214 
2215 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
2216 static void efx_net_stats(struct net_device *net_dev,
2217 			  struct rtnl_link_stats64 *stats)
2218 {
2219 	struct efx_nic *efx = netdev_priv(net_dev);
2220 
2221 	spin_lock_bh(&efx->stats_lock);
2222 	efx->type->update_stats(efx, NULL, stats);
2223 	spin_unlock_bh(&efx->stats_lock);
2224 }
2225 
2226 /* Context: netif_tx_lock held, BHs disabled. */
2227 static void efx_watchdog(struct net_device *net_dev)
2228 {
2229 	struct efx_nic *efx = netdev_priv(net_dev);
2230 
2231 	netif_err(efx, tx_err, efx->net_dev,
2232 		  "TX stuck with port_enabled=%d: resetting channels\n",
2233 		  efx->port_enabled);
2234 
2235 	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2236 }
2237 
2238 
2239 /* Context: process, rtnl_lock() held. */
2240 static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
2241 {
2242 	struct efx_nic *efx = netdev_priv(net_dev);
2243 	int rc;
2244 
2245 	rc = efx_check_disabled(efx);
2246 	if (rc)
2247 		return rc;
2248 
2249 	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2250 
2251 	efx_device_detach_sync(efx);
2252 	efx_stop_all(efx);
2253 
2254 	mutex_lock(&efx->mac_lock);
2255 	net_dev->mtu = new_mtu;
2256 	efx_mac_reconfigure(efx);
2257 	mutex_unlock(&efx->mac_lock);
2258 
2259 	efx_start_all(efx);
2260 	efx_device_attach_if_not_resetting(efx);
2261 	return 0;
2262 }
2263 
2264 static int efx_set_mac_address(struct net_device *net_dev, void *data)
2265 {
2266 	struct efx_nic *efx = netdev_priv(net_dev);
2267 	struct sockaddr *addr = data;
2268 	u8 *new_addr = addr->sa_data;
2269 	u8 old_addr[6];
2270 	int rc;
2271 
2272 	if (!is_valid_ether_addr(new_addr)) {
2273 		netif_err(efx, drv, efx->net_dev,
2274 			  "invalid ethernet MAC address requested: %pM\n",
2275 			  new_addr);
2276 		return -EADDRNOTAVAIL;
2277 	}
2278 
2279 	/* save old address */
2280 	ether_addr_copy(old_addr, net_dev->dev_addr);
2281 	ether_addr_copy(net_dev->dev_addr, new_addr);
2282 	if (efx->type->set_mac_address) {
2283 		rc = efx->type->set_mac_address(efx);
2284 		if (rc) {
2285 			ether_addr_copy(net_dev->dev_addr, old_addr);
2286 			return rc;
2287 		}
2288 	}
2289 
2290 	/* Reconfigure the MAC */
2291 	mutex_lock(&efx->mac_lock);
2292 	efx_mac_reconfigure(efx);
2293 	mutex_unlock(&efx->mac_lock);
2294 
2295 	return 0;
2296 }
2297 
2298 /* Context: netif_addr_lock held, BHs disabled. */
2299 static void efx_set_rx_mode(struct net_device *net_dev)
2300 {
2301 	struct efx_nic *efx = netdev_priv(net_dev);
2302 
2303 	if (efx->port_enabled)
2304 		queue_work(efx->workqueue, &efx->mac_work);
2305 	/* Otherwise efx_start_port() will do this */
2306 }
2307 
2308 static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2309 {
2310 	struct efx_nic *efx = netdev_priv(net_dev);
2311 	int rc;
2312 
2313 	/* If disabling RX n-tuple filtering, clear existing filters */
2314 	if (net_dev->features & ~data & NETIF_F_NTUPLE) {
2315 		rc = efx->type->filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);
2316 		if (rc)
2317 			return rc;
2318 	}
2319 
2320 	/* If Rx VLAN filter is changed, update filters via mac_reconfigure */
2321 	if ((net_dev->features ^ data) & NETIF_F_HW_VLAN_CTAG_FILTER) {
2322 		/* efx_set_rx_mode() will schedule MAC work to update filters
2323 		 * when a new features are finally set in net_dev.
2324 		 */
2325 		efx_set_rx_mode(net_dev);
2326 	}
2327 
2328 	return 0;
2329 }
2330 
2331 static int efx_get_phys_port_id(struct net_device *net_dev,
2332 				struct netdev_phys_item_id *ppid)
2333 {
2334 	struct efx_nic *efx = netdev_priv(net_dev);
2335 
2336 	if (efx->type->get_phys_port_id)
2337 		return efx->type->get_phys_port_id(efx, ppid);
2338 	else
2339 		return -EOPNOTSUPP;
2340 }
2341 
2342 static int efx_get_phys_port_name(struct net_device *net_dev,
2343 				  char *name, size_t len)
2344 {
2345 	struct efx_nic *efx = netdev_priv(net_dev);
2346 
2347 	if (snprintf(name, len, "p%u", efx->port_num) >= len)
2348 		return -EINVAL;
2349 	return 0;
2350 }
2351 
2352 static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid)
2353 {
2354 	struct efx_nic *efx = netdev_priv(net_dev);
2355 
2356 	if (efx->type->vlan_rx_add_vid)
2357 		return efx->type->vlan_rx_add_vid(efx, proto, vid);
2358 	else
2359 		return -EOPNOTSUPP;
2360 }
2361 
2362 static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid)
2363 {
2364 	struct efx_nic *efx = netdev_priv(net_dev);
2365 
2366 	if (efx->type->vlan_rx_kill_vid)
2367 		return efx->type->vlan_rx_kill_vid(efx, proto, vid);
2368 	else
2369 		return -EOPNOTSUPP;
2370 }
2371 
2372 static int efx_udp_tunnel_type_map(enum udp_parsable_tunnel_type in)
2373 {
2374 	switch (in) {
2375 	case UDP_TUNNEL_TYPE_VXLAN:
2376 		return TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN;
2377 	case UDP_TUNNEL_TYPE_GENEVE:
2378 		return TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE;
2379 	default:
2380 		return -1;
2381 	}
2382 }
2383 
2384 static void efx_udp_tunnel_add(struct net_device *dev, struct udp_tunnel_info *ti)
2385 {
2386 	struct efx_nic *efx = netdev_priv(dev);
2387 	struct efx_udp_tunnel tnl;
2388 	int efx_tunnel_type;
2389 
2390 	efx_tunnel_type = efx_udp_tunnel_type_map(ti->type);
2391 	if (efx_tunnel_type < 0)
2392 		return;
2393 
2394 	tnl.type = (u16)efx_tunnel_type;
2395 	tnl.port = ti->port;
2396 
2397 	if (efx->type->udp_tnl_add_port)
2398 		(void)efx->type->udp_tnl_add_port(efx, tnl);
2399 }
2400 
2401 static void efx_udp_tunnel_del(struct net_device *dev, struct udp_tunnel_info *ti)
2402 {
2403 	struct efx_nic *efx = netdev_priv(dev);
2404 	struct efx_udp_tunnel tnl;
2405 	int efx_tunnel_type;
2406 
2407 	efx_tunnel_type = efx_udp_tunnel_type_map(ti->type);
2408 	if (efx_tunnel_type < 0)
2409 		return;
2410 
2411 	tnl.type = (u16)efx_tunnel_type;
2412 	tnl.port = ti->port;
2413 
2414 	if (efx->type->udp_tnl_del_port)
2415 		(void)efx->type->udp_tnl_del_port(efx, tnl);
2416 }
2417 
2418 static const struct net_device_ops efx_netdev_ops = {
2419 	.ndo_open		= efx_net_open,
2420 	.ndo_stop		= efx_net_stop,
2421 	.ndo_get_stats64	= efx_net_stats,
2422 	.ndo_tx_timeout		= efx_watchdog,
2423 	.ndo_start_xmit		= efx_hard_start_xmit,
2424 	.ndo_validate_addr	= eth_validate_addr,
2425 	.ndo_do_ioctl		= efx_ioctl,
2426 	.ndo_change_mtu		= efx_change_mtu,
2427 	.ndo_set_mac_address	= efx_set_mac_address,
2428 	.ndo_set_rx_mode	= efx_set_rx_mode,
2429 	.ndo_set_features	= efx_set_features,
2430 	.ndo_vlan_rx_add_vid	= efx_vlan_rx_add_vid,
2431 	.ndo_vlan_rx_kill_vid	= efx_vlan_rx_kill_vid,
2432 #ifdef CONFIG_SFC_SRIOV
2433 	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
2434 	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
2435 	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
2436 	.ndo_get_vf_config	= efx_sriov_get_vf_config,
2437 	.ndo_set_vf_link_state  = efx_sriov_set_vf_link_state,
2438 #endif
2439 	.ndo_get_phys_port_id   = efx_get_phys_port_id,
2440 	.ndo_get_phys_port_name	= efx_get_phys_port_name,
2441 #ifdef CONFIG_NET_POLL_CONTROLLER
2442 	.ndo_poll_controller = efx_netpoll,
2443 #endif
2444 	.ndo_setup_tc		= efx_setup_tc,
2445 #ifdef CONFIG_RFS_ACCEL
2446 	.ndo_rx_flow_steer	= efx_filter_rfs,
2447 #endif
2448 	.ndo_udp_tunnel_add	= efx_udp_tunnel_add,
2449 	.ndo_udp_tunnel_del	= efx_udp_tunnel_del,
2450 };
2451 
2452 static void efx_update_name(struct efx_nic *efx)
2453 {
2454 	strcpy(efx->name, efx->net_dev->name);
2455 	efx_mtd_rename(efx);
2456 	efx_set_channel_names(efx);
2457 }
2458 
2459 static int efx_netdev_event(struct notifier_block *this,
2460 			    unsigned long event, void *ptr)
2461 {
2462 	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2463 
2464 	if ((net_dev->netdev_ops == &efx_netdev_ops) &&
2465 	    event == NETDEV_CHANGENAME)
2466 		efx_update_name(netdev_priv(net_dev));
2467 
2468 	return NOTIFY_DONE;
2469 }
2470 
2471 static struct notifier_block efx_netdev_notifier = {
2472 	.notifier_call = efx_netdev_event,
2473 };
2474 
2475 static ssize_t
2476 show_phy_type(struct device *dev, struct device_attribute *attr, char *buf)
2477 {
2478 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2479 	return sprintf(buf, "%d\n", efx->phy_type);
2480 }
2481 static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
2482 
2483 #ifdef CONFIG_SFC_MCDI_LOGGING
2484 static ssize_t show_mcdi_log(struct device *dev, struct device_attribute *attr,
2485 			     char *buf)
2486 {
2487 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2488 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2489 
2490 	return scnprintf(buf, PAGE_SIZE, "%d\n", mcdi->logging_enabled);
2491 }
2492 static ssize_t set_mcdi_log(struct device *dev, struct device_attribute *attr,
2493 			    const char *buf, size_t count)
2494 {
2495 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2496 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2497 	bool enable = count > 0 && *buf != '0';
2498 
2499 	mcdi->logging_enabled = enable;
2500 	return count;
2501 }
2502 static DEVICE_ATTR(mcdi_logging, 0644, show_mcdi_log, set_mcdi_log);
2503 #endif
2504 
2505 static int efx_register_netdev(struct efx_nic *efx)
2506 {
2507 	struct net_device *net_dev = efx->net_dev;
2508 	struct efx_channel *channel;
2509 	int rc;
2510 
2511 	net_dev->watchdog_timeo = 5 * HZ;
2512 	net_dev->irq = efx->pci_dev->irq;
2513 	net_dev->netdev_ops = &efx_netdev_ops;
2514 	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0)
2515 		net_dev->priv_flags |= IFF_UNICAST_FLT;
2516 	net_dev->ethtool_ops = &efx_ethtool_ops;
2517 	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2518 	net_dev->min_mtu = EFX_MIN_MTU;
2519 	net_dev->max_mtu = EFX_MAX_MTU;
2520 
2521 	rtnl_lock();
2522 
2523 	/* Enable resets to be scheduled and check whether any were
2524 	 * already requested.  If so, the NIC is probably hosed so we
2525 	 * abort.
2526 	 */
2527 	efx->state = STATE_READY;
2528 	smp_mb(); /* ensure we change state before checking reset_pending */
2529 	if (efx->reset_pending) {
2530 		netif_err(efx, probe, efx->net_dev,
2531 			  "aborting probe due to scheduled reset\n");
2532 		rc = -EIO;
2533 		goto fail_locked;
2534 	}
2535 
2536 	rc = dev_alloc_name(net_dev, net_dev->name);
2537 	if (rc < 0)
2538 		goto fail_locked;
2539 	efx_update_name(efx);
2540 
2541 	/* Always start with carrier off; PHY events will detect the link */
2542 	netif_carrier_off(net_dev);
2543 
2544 	rc = register_netdevice(net_dev);
2545 	if (rc)
2546 		goto fail_locked;
2547 
2548 	efx_for_each_channel(channel, efx) {
2549 		struct efx_tx_queue *tx_queue;
2550 		efx_for_each_channel_tx_queue(tx_queue, channel)
2551 			efx_init_tx_queue_core_txq(tx_queue);
2552 	}
2553 
2554 	efx_associate(efx);
2555 
2556 	rtnl_unlock();
2557 
2558 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2559 	if (rc) {
2560 		netif_err(efx, drv, efx->net_dev,
2561 			  "failed to init net dev attributes\n");
2562 		goto fail_registered;
2563 	}
2564 #ifdef CONFIG_SFC_MCDI_LOGGING
2565 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2566 	if (rc) {
2567 		netif_err(efx, drv, efx->net_dev,
2568 			  "failed to init net dev attributes\n");
2569 		goto fail_attr_mcdi_logging;
2570 	}
2571 #endif
2572 
2573 	return 0;
2574 
2575 #ifdef CONFIG_SFC_MCDI_LOGGING
2576 fail_attr_mcdi_logging:
2577 	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2578 #endif
2579 fail_registered:
2580 	rtnl_lock();
2581 	efx_dissociate(efx);
2582 	unregister_netdevice(net_dev);
2583 fail_locked:
2584 	efx->state = STATE_UNINIT;
2585 	rtnl_unlock();
2586 	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2587 	return rc;
2588 }
2589 
2590 static void efx_unregister_netdev(struct efx_nic *efx)
2591 {
2592 	if (!efx->net_dev)
2593 		return;
2594 
2595 	BUG_ON(netdev_priv(efx->net_dev) != efx);
2596 
2597 	if (efx_dev_registered(efx)) {
2598 		strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
2599 #ifdef CONFIG_SFC_MCDI_LOGGING
2600 		device_remove_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2601 #endif
2602 		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2603 		unregister_netdev(efx->net_dev);
2604 	}
2605 }
2606 
2607 /**************************************************************************
2608  *
2609  * Device reset and suspend
2610  *
2611  **************************************************************************/
2612 
2613 /* Tears down the entire software state and most of the hardware state
2614  * before reset.  */
2615 void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2616 {
2617 	EFX_ASSERT_RESET_SERIALISED(efx);
2618 
2619 	if (method == RESET_TYPE_MCDI_TIMEOUT)
2620 		efx->type->prepare_flr(efx);
2621 
2622 	efx_stop_all(efx);
2623 	efx_disable_interrupts(efx);
2624 
2625 	mutex_lock(&efx->mac_lock);
2626 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2627 	    method != RESET_TYPE_DATAPATH)
2628 		efx->phy_op->fini(efx);
2629 	efx->type->fini(efx);
2630 }
2631 
2632 /* This function will always ensure that the locks acquired in
2633  * efx_reset_down() are released. A failure return code indicates
2634  * that we were unable to reinitialise the hardware, and the
2635  * driver should be disabled. If ok is false, then the rx and tx
2636  * engines are not restarted, pending a RESET_DISABLE. */
2637 int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2638 {
2639 	int rc;
2640 
2641 	EFX_ASSERT_RESET_SERIALISED(efx);
2642 
2643 	if (method == RESET_TYPE_MCDI_TIMEOUT)
2644 		efx->type->finish_flr(efx);
2645 
2646 	/* Ensure that SRAM is initialised even if we're disabling the device */
2647 	rc = efx->type->init(efx);
2648 	if (rc) {
2649 		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2650 		goto fail;
2651 	}
2652 
2653 	if (!ok)
2654 		goto fail;
2655 
2656 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2657 	    method != RESET_TYPE_DATAPATH) {
2658 		rc = efx->phy_op->init(efx);
2659 		if (rc)
2660 			goto fail;
2661 		rc = efx->phy_op->reconfigure(efx);
2662 		if (rc && rc != -EPERM)
2663 			netif_err(efx, drv, efx->net_dev,
2664 				  "could not restore PHY settings\n");
2665 	}
2666 
2667 	rc = efx_enable_interrupts(efx);
2668 	if (rc)
2669 		goto fail;
2670 
2671 #ifdef CONFIG_SFC_SRIOV
2672 	rc = efx->type->vswitching_restore(efx);
2673 	if (rc) /* not fatal; the PF will still work fine */
2674 		netif_warn(efx, probe, efx->net_dev,
2675 			   "failed to restore vswitching rc=%d;"
2676 			   " VFs may not function\n", rc);
2677 #endif
2678 
2679 	down_read(&efx->filter_sem);
2680 	efx_restore_filters(efx);
2681 	up_read(&efx->filter_sem);
2682 	if (efx->type->sriov_reset)
2683 		efx->type->sriov_reset(efx);
2684 
2685 	mutex_unlock(&efx->mac_lock);
2686 
2687 	efx_start_all(efx);
2688 
2689 	if (efx->type->udp_tnl_push_ports)
2690 		efx->type->udp_tnl_push_ports(efx);
2691 
2692 	return 0;
2693 
2694 fail:
2695 	efx->port_initialized = false;
2696 
2697 	mutex_unlock(&efx->mac_lock);
2698 
2699 	return rc;
2700 }
2701 
2702 /* Reset the NIC using the specified method.  Note that the reset may
2703  * fail, in which case the card will be left in an unusable state.
2704  *
2705  * Caller must hold the rtnl_lock.
2706  */
2707 int efx_reset(struct efx_nic *efx, enum reset_type method)
2708 {
2709 	int rc, rc2;
2710 	bool disabled;
2711 
2712 	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
2713 		   RESET_TYPE(method));
2714 
2715 	efx_device_detach_sync(efx);
2716 	efx_reset_down(efx, method);
2717 
2718 	rc = efx->type->reset(efx, method);
2719 	if (rc) {
2720 		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2721 		goto out;
2722 	}
2723 
2724 	/* Clear flags for the scopes we covered.  We assume the NIC and
2725 	 * driver are now quiescent so that there is no race here.
2726 	 */
2727 	if (method < RESET_TYPE_MAX_METHOD)
2728 		efx->reset_pending &= -(1 << (method + 1));
2729 	else /* it doesn't fit into the well-ordered scope hierarchy */
2730 		__clear_bit(method, &efx->reset_pending);
2731 
2732 	/* Reinitialise bus-mastering, which may have been turned off before
2733 	 * the reset was scheduled. This is still appropriate, even in the
2734 	 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
2735 	 * can respond to requests. */
2736 	pci_set_master(efx->pci_dev);
2737 
2738 out:
2739 	/* Leave device stopped if necessary */
2740 	disabled = rc ||
2741 		method == RESET_TYPE_DISABLE ||
2742 		method == RESET_TYPE_RECOVER_OR_DISABLE;
2743 	rc2 = efx_reset_up(efx, method, !disabled);
2744 	if (rc2) {
2745 		disabled = true;
2746 		if (!rc)
2747 			rc = rc2;
2748 	}
2749 
2750 	if (disabled) {
2751 		dev_close(efx->net_dev);
2752 		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2753 		efx->state = STATE_DISABLED;
2754 	} else {
2755 		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2756 		efx_device_attach_if_not_resetting(efx);
2757 	}
2758 	return rc;
2759 }
2760 
2761 /* Try recovery mechanisms.
2762  * For now only EEH is supported.
2763  * Returns 0 if the recovery mechanisms are unsuccessful.
2764  * Returns a non-zero value otherwise.
2765  */
2766 int efx_try_recovery(struct efx_nic *efx)
2767 {
2768 #ifdef CONFIG_EEH
2769 	/* A PCI error can occur and not be seen by EEH because nothing
2770 	 * happens on the PCI bus. In this case the driver may fail and
2771 	 * schedule a 'recover or reset', leading to this recovery handler.
2772 	 * Manually call the eeh failure check function.
2773 	 */
2774 	struct eeh_dev *eehdev = pci_dev_to_eeh_dev(efx->pci_dev);
2775 	if (eeh_dev_check_failure(eehdev)) {
2776 		/* The EEH mechanisms will handle the error and reset the
2777 		 * device if necessary.
2778 		 */
2779 		return 1;
2780 	}
2781 #endif
2782 	return 0;
2783 }
2784 
2785 static void efx_wait_for_bist_end(struct efx_nic *efx)
2786 {
2787 	int i;
2788 
2789 	for (i = 0; i < BIST_WAIT_DELAY_COUNT; ++i) {
2790 		if (efx_mcdi_poll_reboot(efx))
2791 			goto out;
2792 		msleep(BIST_WAIT_DELAY_MS);
2793 	}
2794 
2795 	netif_err(efx, drv, efx->net_dev, "Warning: No MC reboot after BIST mode\n");
2796 out:
2797 	/* Either way unset the BIST flag. If we found no reboot we probably
2798 	 * won't recover, but we should try.
2799 	 */
2800 	efx->mc_bist_for_other_fn = false;
2801 }
2802 
2803 /* The worker thread exists so that code that cannot sleep can
2804  * schedule a reset for later.
2805  */
2806 static void efx_reset_work(struct work_struct *data)
2807 {
2808 	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2809 	unsigned long pending;
2810 	enum reset_type method;
2811 
2812 	pending = ACCESS_ONCE(efx->reset_pending);
2813 	method = fls(pending) - 1;
2814 
2815 	if (method == RESET_TYPE_MC_BIST)
2816 		efx_wait_for_bist_end(efx);
2817 
2818 	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
2819 	     method == RESET_TYPE_RECOVER_OR_ALL) &&
2820 	    efx_try_recovery(efx))
2821 		return;
2822 
2823 	if (!pending)
2824 		return;
2825 
2826 	rtnl_lock();
2827 
2828 	/* We checked the state in efx_schedule_reset() but it may
2829 	 * have changed by now.  Now that we have the RTNL lock,
2830 	 * it cannot change again.
2831 	 */
2832 	if (efx->state == STATE_READY)
2833 		(void)efx_reset(efx, method);
2834 
2835 	rtnl_unlock();
2836 }
2837 
2838 void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
2839 {
2840 	enum reset_type method;
2841 
2842 	if (efx->state == STATE_RECOVERY) {
2843 		netif_dbg(efx, drv, efx->net_dev,
2844 			  "recovering: skip scheduling %s reset\n",
2845 			  RESET_TYPE(type));
2846 		return;
2847 	}
2848 
2849 	switch (type) {
2850 	case RESET_TYPE_INVISIBLE:
2851 	case RESET_TYPE_ALL:
2852 	case RESET_TYPE_RECOVER_OR_ALL:
2853 	case RESET_TYPE_WORLD:
2854 	case RESET_TYPE_DISABLE:
2855 	case RESET_TYPE_RECOVER_OR_DISABLE:
2856 	case RESET_TYPE_DATAPATH:
2857 	case RESET_TYPE_MC_BIST:
2858 	case RESET_TYPE_MCDI_TIMEOUT:
2859 		method = type;
2860 		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
2861 			  RESET_TYPE(method));
2862 		break;
2863 	default:
2864 		method = efx->type->map_reset_reason(type);
2865 		netif_dbg(efx, drv, efx->net_dev,
2866 			  "scheduling %s reset for %s\n",
2867 			  RESET_TYPE(method), RESET_TYPE(type));
2868 		break;
2869 	}
2870 
2871 	set_bit(method, &efx->reset_pending);
2872 	smp_mb(); /* ensure we change reset_pending before checking state */
2873 
2874 	/* If we're not READY then just leave the flags set as the cue
2875 	 * to abort probing or reschedule the reset later.
2876 	 */
2877 	if (ACCESS_ONCE(efx->state) != STATE_READY)
2878 		return;
2879 
2880 	/* efx_process_channel() will no longer read events once a
2881 	 * reset is scheduled. So switch back to poll'd MCDI completions. */
2882 	efx_mcdi_mode_poll(efx);
2883 
2884 	queue_work(reset_workqueue, &efx->reset_work);
2885 }
2886 
2887 /**************************************************************************
2888  *
2889  * List of NICs we support
2890  *
2891  **************************************************************************/
2892 
2893 /* PCI device ID table */
2894 static const struct pci_device_id efx_pci_table[] = {
2895 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2896 	 .driver_data = (unsigned long) &siena_a0_nic_type},
2897 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2898 	 .driver_data = (unsigned long) &siena_a0_nic_type},
2899 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
2900 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2901 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903),  /* SFC9120 VF */
2902 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2903 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923),  /* SFC9140 PF */
2904 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2905 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923),  /* SFC9140 VF */
2906 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2907 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03),  /* SFC9220 PF */
2908 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2909 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03),  /* SFC9220 VF */
2910 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2911 	{0}			/* end of list */
2912 };
2913 
2914 /**************************************************************************
2915  *
2916  * Dummy PHY/MAC operations
2917  *
2918  * Can be used for some unimplemented operations
2919  * Needed so all function pointers are valid and do not have to be tested
2920  * before use
2921  *
2922  **************************************************************************/
2923 int efx_port_dummy_op_int(struct efx_nic *efx)
2924 {
2925 	return 0;
2926 }
2927 void efx_port_dummy_op_void(struct efx_nic *efx) {}
2928 
2929 static bool efx_port_dummy_op_poll(struct efx_nic *efx)
2930 {
2931 	return false;
2932 }
2933 
2934 static const struct efx_phy_operations efx_dummy_phy_operations = {
2935 	.init		 = efx_port_dummy_op_int,
2936 	.reconfigure	 = efx_port_dummy_op_int,
2937 	.poll		 = efx_port_dummy_op_poll,
2938 	.fini		 = efx_port_dummy_op_void,
2939 };
2940 
2941 /**************************************************************************
2942  *
2943  * Data housekeeping
2944  *
2945  **************************************************************************/
2946 
2947 /* This zeroes out and then fills in the invariants in a struct
2948  * efx_nic (including all sub-structures).
2949  */
2950 static int efx_init_struct(struct efx_nic *efx,
2951 			   struct pci_dev *pci_dev, struct net_device *net_dev)
2952 {
2953 	int rc = -ENOMEM, i;
2954 
2955 	/* Initialise common structures */
2956 	INIT_LIST_HEAD(&efx->node);
2957 	INIT_LIST_HEAD(&efx->secondary_list);
2958 	spin_lock_init(&efx->biu_lock);
2959 #ifdef CONFIG_SFC_MTD
2960 	INIT_LIST_HEAD(&efx->mtd_list);
2961 #endif
2962 	INIT_WORK(&efx->reset_work, efx_reset_work);
2963 	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
2964 	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
2965 	efx->pci_dev = pci_dev;
2966 	efx->msg_enable = debug;
2967 	efx->state = STATE_UNINIT;
2968 	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
2969 
2970 	efx->net_dev = net_dev;
2971 	efx->rx_prefix_size = efx->type->rx_prefix_size;
2972 	efx->rx_ip_align =
2973 		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
2974 	efx->rx_packet_hash_offset =
2975 		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
2976 	efx->rx_packet_ts_offset =
2977 		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
2978 	spin_lock_init(&efx->stats_lock);
2979 	mutex_init(&efx->mac_lock);
2980 	efx->phy_op = &efx_dummy_phy_operations;
2981 	efx->mdio.dev = net_dev;
2982 	INIT_WORK(&efx->mac_work, efx_mac_work);
2983 	init_waitqueue_head(&efx->flush_wq);
2984 
2985 	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2986 		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
2987 		if (!efx->channel[i])
2988 			goto fail;
2989 		efx->msi_context[i].efx = efx;
2990 		efx->msi_context[i].index = i;
2991 	}
2992 
2993 	/* Higher numbered interrupt modes are less capable! */
2994 	if (WARN_ON_ONCE(efx->type->max_interrupt_mode >
2995 			 efx->type->min_interrupt_mode)) {
2996 		rc = -EIO;
2997 		goto fail;
2998 	}
2999 	efx->interrupt_mode = max(efx->type->max_interrupt_mode,
3000 				  interrupt_mode);
3001 	efx->interrupt_mode = min(efx->type->min_interrupt_mode,
3002 				  interrupt_mode);
3003 
3004 	/* Would be good to use the net_dev name, but we're too early */
3005 	snprintf(efx->workqueue_name, sizeof(efx->workqueue_name), "sfc%s",
3006 		 pci_name(pci_dev));
3007 	efx->workqueue = create_singlethread_workqueue(efx->workqueue_name);
3008 	if (!efx->workqueue)
3009 		goto fail;
3010 
3011 	return 0;
3012 
3013 fail:
3014 	efx_fini_struct(efx);
3015 	return rc;
3016 }
3017 
3018 static void efx_fini_struct(struct efx_nic *efx)
3019 {
3020 	int i;
3021 
3022 	for (i = 0; i < EFX_MAX_CHANNELS; i++)
3023 		kfree(efx->channel[i]);
3024 
3025 	kfree(efx->vpd_sn);
3026 
3027 	if (efx->workqueue) {
3028 		destroy_workqueue(efx->workqueue);
3029 		efx->workqueue = NULL;
3030 	}
3031 }
3032 
3033 void efx_update_sw_stats(struct efx_nic *efx, u64 *stats)
3034 {
3035 	u64 n_rx_nodesc_trunc = 0;
3036 	struct efx_channel *channel;
3037 
3038 	efx_for_each_channel(channel, efx)
3039 		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
3040 	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
3041 	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
3042 }
3043 
3044 /**************************************************************************
3045  *
3046  * PCI interface
3047  *
3048  **************************************************************************/
3049 
3050 /* Main body of final NIC shutdown code
3051  * This is called only at module unload (or hotplug removal).
3052  */
3053 static void efx_pci_remove_main(struct efx_nic *efx)
3054 {
3055 	/* Flush reset_work. It can no longer be scheduled since we
3056 	 * are not READY.
3057 	 */
3058 	BUG_ON(efx->state == STATE_READY);
3059 	cancel_work_sync(&efx->reset_work);
3060 
3061 	efx_disable_interrupts(efx);
3062 	efx_nic_fini_interrupt(efx);
3063 	efx_fini_port(efx);
3064 	efx->type->fini(efx);
3065 	efx_fini_napi(efx);
3066 	efx_remove_all(efx);
3067 }
3068 
3069 /* Final NIC shutdown
3070  * This is called only at module unload (or hotplug removal).  A PF can call
3071  * this on its VFs to ensure they are unbound first.
3072  */
3073 static void efx_pci_remove(struct pci_dev *pci_dev)
3074 {
3075 	struct efx_nic *efx;
3076 
3077 	efx = pci_get_drvdata(pci_dev);
3078 	if (!efx)
3079 		return;
3080 
3081 	/* Mark the NIC as fini, then stop the interface */
3082 	rtnl_lock();
3083 	efx_dissociate(efx);
3084 	dev_close(efx->net_dev);
3085 	efx_disable_interrupts(efx);
3086 	efx->state = STATE_UNINIT;
3087 	rtnl_unlock();
3088 
3089 	if (efx->type->sriov_fini)
3090 		efx->type->sriov_fini(efx);
3091 
3092 	efx_unregister_netdev(efx);
3093 
3094 	efx_mtd_remove(efx);
3095 
3096 	efx_pci_remove_main(efx);
3097 
3098 	efx_fini_io(efx);
3099 	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
3100 
3101 	efx_fini_struct(efx);
3102 	free_netdev(efx->net_dev);
3103 
3104 	pci_disable_pcie_error_reporting(pci_dev);
3105 };
3106 
3107 /* NIC VPD information
3108  * Called during probe to display the part number of the
3109  * installed NIC.  VPD is potentially very large but this should
3110  * always appear within the first 512 bytes.
3111  */
3112 #define SFC_VPD_LEN 512
3113 static void efx_probe_vpd_strings(struct efx_nic *efx)
3114 {
3115 	struct pci_dev *dev = efx->pci_dev;
3116 	char vpd_data[SFC_VPD_LEN];
3117 	ssize_t vpd_size;
3118 	int ro_start, ro_size, i, j;
3119 
3120 	/* Get the vpd data from the device */
3121 	vpd_size = pci_read_vpd(dev, 0, sizeof(vpd_data), vpd_data);
3122 	if (vpd_size <= 0) {
3123 		netif_err(efx, drv, efx->net_dev, "Unable to read VPD\n");
3124 		return;
3125 	}
3126 
3127 	/* Get the Read only section */
3128 	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size, PCI_VPD_LRDT_RO_DATA);
3129 	if (ro_start < 0) {
3130 		netif_err(efx, drv, efx->net_dev, "VPD Read-only not found\n");
3131 		return;
3132 	}
3133 
3134 	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
3135 	j = ro_size;
3136 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3137 	if (i + j > vpd_size)
3138 		j = vpd_size - i;
3139 
3140 	/* Get the Part number */
3141 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "PN");
3142 	if (i < 0) {
3143 		netif_err(efx, drv, efx->net_dev, "Part number not found\n");
3144 		return;
3145 	}
3146 
3147 	j = pci_vpd_info_field_size(&vpd_data[i]);
3148 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3149 	if (i + j > vpd_size) {
3150 		netif_err(efx, drv, efx->net_dev, "Incomplete part number\n");
3151 		return;
3152 	}
3153 
3154 	netif_info(efx, drv, efx->net_dev,
3155 		   "Part Number : %.*s\n", j, &vpd_data[i]);
3156 
3157 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3158 	j = ro_size;
3159 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "SN");
3160 	if (i < 0) {
3161 		netif_err(efx, drv, efx->net_dev, "Serial number not found\n");
3162 		return;
3163 	}
3164 
3165 	j = pci_vpd_info_field_size(&vpd_data[i]);
3166 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3167 	if (i + j > vpd_size) {
3168 		netif_err(efx, drv, efx->net_dev, "Incomplete serial number\n");
3169 		return;
3170 	}
3171 
3172 	efx->vpd_sn = kmalloc(j + 1, GFP_KERNEL);
3173 	if (!efx->vpd_sn)
3174 		return;
3175 
3176 	snprintf(efx->vpd_sn, j + 1, "%s", &vpd_data[i]);
3177 }
3178 
3179 
3180 /* Main body of NIC initialisation
3181  * This is called at module load (or hotplug insertion, theoretically).
3182  */
3183 static int efx_pci_probe_main(struct efx_nic *efx)
3184 {
3185 	int rc;
3186 
3187 	/* Do start-of-day initialisation */
3188 	rc = efx_probe_all(efx);
3189 	if (rc)
3190 		goto fail1;
3191 
3192 	efx_init_napi(efx);
3193 
3194 	rc = efx->type->init(efx);
3195 	if (rc) {
3196 		netif_err(efx, probe, efx->net_dev,
3197 			  "failed to initialise NIC\n");
3198 		goto fail3;
3199 	}
3200 
3201 	rc = efx_init_port(efx);
3202 	if (rc) {
3203 		netif_err(efx, probe, efx->net_dev,
3204 			  "failed to initialise port\n");
3205 		goto fail4;
3206 	}
3207 
3208 	rc = efx_nic_init_interrupt(efx);
3209 	if (rc)
3210 		goto fail5;
3211 	rc = efx_enable_interrupts(efx);
3212 	if (rc)
3213 		goto fail6;
3214 
3215 	return 0;
3216 
3217  fail6:
3218 	efx_nic_fini_interrupt(efx);
3219  fail5:
3220 	efx_fini_port(efx);
3221  fail4:
3222 	efx->type->fini(efx);
3223  fail3:
3224 	efx_fini_napi(efx);
3225 	efx_remove_all(efx);
3226  fail1:
3227 	return rc;
3228 }
3229 
3230 static int efx_pci_probe_post_io(struct efx_nic *efx)
3231 {
3232 	struct net_device *net_dev = efx->net_dev;
3233 	int rc = efx_pci_probe_main(efx);
3234 
3235 	if (rc)
3236 		return rc;
3237 
3238 	if (efx->type->sriov_init) {
3239 		rc = efx->type->sriov_init(efx);
3240 		if (rc)
3241 			netif_err(efx, probe, efx->net_dev,
3242 				  "SR-IOV can't be enabled rc %d\n", rc);
3243 	}
3244 
3245 	/* Determine netdevice features */
3246 	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
3247 			      NETIF_F_TSO | NETIF_F_RXCSUM);
3248 	if (efx->type->offload_features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
3249 		net_dev->features |= NETIF_F_TSO6;
3250 	/* Check whether device supports TSO */
3251 	if (!efx->type->tso_versions || !efx->type->tso_versions(efx))
3252 		net_dev->features &= ~NETIF_F_ALL_TSO;
3253 	/* Mask for features that also apply to VLAN devices */
3254 	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
3255 				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
3256 				   NETIF_F_RXCSUM);
3257 
3258 	net_dev->hw_features = net_dev->features & ~efx->fixed_features;
3259 
3260 	/* Disable VLAN filtering by default.  It may be enforced if
3261 	 * the feature is fixed (i.e. VLAN filters are required to
3262 	 * receive VLAN tagged packets due to vPort restrictions).
3263 	 */
3264 	net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3265 	net_dev->features |= efx->fixed_features;
3266 
3267 	rc = efx_register_netdev(efx);
3268 	if (!rc)
3269 		return 0;
3270 
3271 	efx_pci_remove_main(efx);
3272 	return rc;
3273 }
3274 
3275 /* NIC initialisation
3276  *
3277  * This is called at module load (or hotplug insertion,
3278  * theoretically).  It sets up PCI mappings, resets the NIC,
3279  * sets up and registers the network devices with the kernel and hooks
3280  * the interrupt service routine.  It does not prepare the device for
3281  * transmission; this is left to the first time one of the network
3282  * interfaces is brought up (i.e. efx_net_open).
3283  */
3284 static int efx_pci_probe(struct pci_dev *pci_dev,
3285 			 const struct pci_device_id *entry)
3286 {
3287 	struct net_device *net_dev;
3288 	struct efx_nic *efx;
3289 	int rc;
3290 
3291 	/* Allocate and initialise a struct net_device and struct efx_nic */
3292 	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
3293 				     EFX_MAX_RX_QUEUES);
3294 	if (!net_dev)
3295 		return -ENOMEM;
3296 	efx = netdev_priv(net_dev);
3297 	efx->type = (const struct efx_nic_type *) entry->driver_data;
3298 	efx->fixed_features |= NETIF_F_HIGHDMA;
3299 
3300 	pci_set_drvdata(pci_dev, efx);
3301 	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
3302 	rc = efx_init_struct(efx, pci_dev, net_dev);
3303 	if (rc)
3304 		goto fail1;
3305 
3306 	netif_info(efx, probe, efx->net_dev,
3307 		   "Solarflare NIC detected\n");
3308 
3309 	if (!efx->type->is_vf)
3310 		efx_probe_vpd_strings(efx);
3311 
3312 	/* Set up basic I/O (BAR mappings etc) */
3313 	rc = efx_init_io(efx);
3314 	if (rc)
3315 		goto fail2;
3316 
3317 	rc = efx_pci_probe_post_io(efx);
3318 	if (rc) {
3319 		/* On failure, retry once immediately.
3320 		 * If we aborted probe due to a scheduled reset, dismiss it.
3321 		 */
3322 		efx->reset_pending = 0;
3323 		rc = efx_pci_probe_post_io(efx);
3324 		if (rc) {
3325 			/* On another failure, retry once more
3326 			 * after a 50-305ms delay.
3327 			 */
3328 			unsigned char r;
3329 
3330 			get_random_bytes(&r, 1);
3331 			msleep((unsigned int)r + 50);
3332 			efx->reset_pending = 0;
3333 			rc = efx_pci_probe_post_io(efx);
3334 		}
3335 	}
3336 	if (rc)
3337 		goto fail3;
3338 
3339 	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
3340 
3341 	/* Try to create MTDs, but allow this to fail */
3342 	rtnl_lock();
3343 	rc = efx_mtd_probe(efx);
3344 	rtnl_unlock();
3345 	if (rc && rc != -EPERM)
3346 		netif_warn(efx, probe, efx->net_dev,
3347 			   "failed to create MTDs (%d)\n", rc);
3348 
3349 	rc = pci_enable_pcie_error_reporting(pci_dev);
3350 	if (rc && rc != -EINVAL)
3351 		netif_notice(efx, probe, efx->net_dev,
3352 			     "PCIE error reporting unavailable (%d).\n",
3353 			     rc);
3354 
3355 	if (efx->type->udp_tnl_push_ports)
3356 		efx->type->udp_tnl_push_ports(efx);
3357 
3358 	return 0;
3359 
3360  fail3:
3361 	efx_fini_io(efx);
3362  fail2:
3363 	efx_fini_struct(efx);
3364  fail1:
3365 	WARN_ON(rc > 0);
3366 	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
3367 	free_netdev(net_dev);
3368 	return rc;
3369 }
3370 
3371 /* efx_pci_sriov_configure returns the actual number of Virtual Functions
3372  * enabled on success
3373  */
3374 #ifdef CONFIG_SFC_SRIOV
3375 static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
3376 {
3377 	int rc;
3378 	struct efx_nic *efx = pci_get_drvdata(dev);
3379 
3380 	if (efx->type->sriov_configure) {
3381 		rc = efx->type->sriov_configure(efx, num_vfs);
3382 		if (rc)
3383 			return rc;
3384 		else
3385 			return num_vfs;
3386 	} else
3387 		return -EOPNOTSUPP;
3388 }
3389 #endif
3390 
3391 static int efx_pm_freeze(struct device *dev)
3392 {
3393 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3394 
3395 	rtnl_lock();
3396 
3397 	if (efx->state != STATE_DISABLED) {
3398 		efx->state = STATE_UNINIT;
3399 
3400 		efx_device_detach_sync(efx);
3401 
3402 		efx_stop_all(efx);
3403 		efx_disable_interrupts(efx);
3404 	}
3405 
3406 	rtnl_unlock();
3407 
3408 	return 0;
3409 }
3410 
3411 static int efx_pm_thaw(struct device *dev)
3412 {
3413 	int rc;
3414 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3415 
3416 	rtnl_lock();
3417 
3418 	if (efx->state != STATE_DISABLED) {
3419 		rc = efx_enable_interrupts(efx);
3420 		if (rc)
3421 			goto fail;
3422 
3423 		mutex_lock(&efx->mac_lock);
3424 		efx->phy_op->reconfigure(efx);
3425 		mutex_unlock(&efx->mac_lock);
3426 
3427 		efx_start_all(efx);
3428 
3429 		efx_device_attach_if_not_resetting(efx);
3430 
3431 		efx->state = STATE_READY;
3432 
3433 		efx->type->resume_wol(efx);
3434 	}
3435 
3436 	rtnl_unlock();
3437 
3438 	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
3439 	queue_work(reset_workqueue, &efx->reset_work);
3440 
3441 	return 0;
3442 
3443 fail:
3444 	rtnl_unlock();
3445 
3446 	return rc;
3447 }
3448 
3449 static int efx_pm_poweroff(struct device *dev)
3450 {
3451 	struct pci_dev *pci_dev = to_pci_dev(dev);
3452 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3453 
3454 	efx->type->fini(efx);
3455 
3456 	efx->reset_pending = 0;
3457 
3458 	pci_save_state(pci_dev);
3459 	return pci_set_power_state(pci_dev, PCI_D3hot);
3460 }
3461 
3462 /* Used for both resume and restore */
3463 static int efx_pm_resume(struct device *dev)
3464 {
3465 	struct pci_dev *pci_dev = to_pci_dev(dev);
3466 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3467 	int rc;
3468 
3469 	rc = pci_set_power_state(pci_dev, PCI_D0);
3470 	if (rc)
3471 		return rc;
3472 	pci_restore_state(pci_dev);
3473 	rc = pci_enable_device(pci_dev);
3474 	if (rc)
3475 		return rc;
3476 	pci_set_master(efx->pci_dev);
3477 	rc = efx->type->reset(efx, RESET_TYPE_ALL);
3478 	if (rc)
3479 		return rc;
3480 	rc = efx->type->init(efx);
3481 	if (rc)
3482 		return rc;
3483 	rc = efx_pm_thaw(dev);
3484 	return rc;
3485 }
3486 
3487 static int efx_pm_suspend(struct device *dev)
3488 {
3489 	int rc;
3490 
3491 	efx_pm_freeze(dev);
3492 	rc = efx_pm_poweroff(dev);
3493 	if (rc)
3494 		efx_pm_resume(dev);
3495 	return rc;
3496 }
3497 
3498 static const struct dev_pm_ops efx_pm_ops = {
3499 	.suspend	= efx_pm_suspend,
3500 	.resume		= efx_pm_resume,
3501 	.freeze		= efx_pm_freeze,
3502 	.thaw		= efx_pm_thaw,
3503 	.poweroff	= efx_pm_poweroff,
3504 	.restore	= efx_pm_resume,
3505 };
3506 
3507 /* A PCI error affecting this device was detected.
3508  * At this point MMIO and DMA may be disabled.
3509  * Stop the software path and request a slot reset.
3510  */
3511 static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
3512 					      enum pci_channel_state state)
3513 {
3514 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3515 	struct efx_nic *efx = pci_get_drvdata(pdev);
3516 
3517 	if (state == pci_channel_io_perm_failure)
3518 		return PCI_ERS_RESULT_DISCONNECT;
3519 
3520 	rtnl_lock();
3521 
3522 	if (efx->state != STATE_DISABLED) {
3523 		efx->state = STATE_RECOVERY;
3524 		efx->reset_pending = 0;
3525 
3526 		efx_device_detach_sync(efx);
3527 
3528 		efx_stop_all(efx);
3529 		efx_disable_interrupts(efx);
3530 
3531 		status = PCI_ERS_RESULT_NEED_RESET;
3532 	} else {
3533 		/* If the interface is disabled we don't want to do anything
3534 		 * with it.
3535 		 */
3536 		status = PCI_ERS_RESULT_RECOVERED;
3537 	}
3538 
3539 	rtnl_unlock();
3540 
3541 	pci_disable_device(pdev);
3542 
3543 	return status;
3544 }
3545 
3546 /* Fake a successful reset, which will be performed later in efx_io_resume. */
3547 static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
3548 {
3549 	struct efx_nic *efx = pci_get_drvdata(pdev);
3550 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3551 	int rc;
3552 
3553 	if (pci_enable_device(pdev)) {
3554 		netif_err(efx, hw, efx->net_dev,
3555 			  "Cannot re-enable PCI device after reset.\n");
3556 		status =  PCI_ERS_RESULT_DISCONNECT;
3557 	}
3558 
3559 	rc = pci_cleanup_aer_uncorrect_error_status(pdev);
3560 	if (rc) {
3561 		netif_err(efx, hw, efx->net_dev,
3562 		"pci_cleanup_aer_uncorrect_error_status failed (%d)\n", rc);
3563 		/* Non-fatal error. Continue. */
3564 	}
3565 
3566 	return status;
3567 }
3568 
3569 /* Perform the actual reset and resume I/O operations. */
3570 static void efx_io_resume(struct pci_dev *pdev)
3571 {
3572 	struct efx_nic *efx = pci_get_drvdata(pdev);
3573 	int rc;
3574 
3575 	rtnl_lock();
3576 
3577 	if (efx->state == STATE_DISABLED)
3578 		goto out;
3579 
3580 	rc = efx_reset(efx, RESET_TYPE_ALL);
3581 	if (rc) {
3582 		netif_err(efx, hw, efx->net_dev,
3583 			  "efx_reset failed after PCI error (%d)\n", rc);
3584 	} else {
3585 		efx->state = STATE_READY;
3586 		netif_dbg(efx, hw, efx->net_dev,
3587 			  "Done resetting and resuming IO after PCI error.\n");
3588 	}
3589 
3590 out:
3591 	rtnl_unlock();
3592 }
3593 
3594 /* For simplicity and reliability, we always require a slot reset and try to
3595  * reset the hardware when a pci error affecting the device is detected.
3596  * We leave both the link_reset and mmio_enabled callback unimplemented:
3597  * with our request for slot reset the mmio_enabled callback will never be
3598  * called, and the link_reset callback is not used by AER or EEH mechanisms.
3599  */
3600 static const struct pci_error_handlers efx_err_handlers = {
3601 	.error_detected = efx_io_error_detected,
3602 	.slot_reset	= efx_io_slot_reset,
3603 	.resume		= efx_io_resume,
3604 };
3605 
3606 static struct pci_driver efx_pci_driver = {
3607 	.name		= KBUILD_MODNAME,
3608 	.id_table	= efx_pci_table,
3609 	.probe		= efx_pci_probe,
3610 	.remove		= efx_pci_remove,
3611 	.driver.pm	= &efx_pm_ops,
3612 	.err_handler	= &efx_err_handlers,
3613 #ifdef CONFIG_SFC_SRIOV
3614 	.sriov_configure = efx_pci_sriov_configure,
3615 #endif
3616 };
3617 
3618 /**************************************************************************
3619  *
3620  * Kernel module interface
3621  *
3622  *************************************************************************/
3623 
3624 module_param(interrupt_mode, uint, 0444);
3625 MODULE_PARM_DESC(interrupt_mode,
3626 		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
3627 
3628 static int __init efx_init_module(void)
3629 {
3630 	int rc;
3631 
3632 	printk(KERN_INFO "Solarflare NET driver v" EFX_DRIVER_VERSION "\n");
3633 
3634 	rc = register_netdevice_notifier(&efx_netdev_notifier);
3635 	if (rc)
3636 		goto err_notifier;
3637 
3638 #ifdef CONFIG_SFC_SRIOV
3639 	rc = efx_init_sriov();
3640 	if (rc)
3641 		goto err_sriov;
3642 #endif
3643 
3644 	reset_workqueue = create_singlethread_workqueue("sfc_reset");
3645 	if (!reset_workqueue) {
3646 		rc = -ENOMEM;
3647 		goto err_reset;
3648 	}
3649 
3650 	rc = pci_register_driver(&efx_pci_driver);
3651 	if (rc < 0)
3652 		goto err_pci;
3653 
3654 	return 0;
3655 
3656  err_pci:
3657 	destroy_workqueue(reset_workqueue);
3658  err_reset:
3659 #ifdef CONFIG_SFC_SRIOV
3660 	efx_fini_sriov();
3661  err_sriov:
3662 #endif
3663 	unregister_netdevice_notifier(&efx_netdev_notifier);
3664  err_notifier:
3665 	return rc;
3666 }
3667 
3668 static void __exit efx_exit_module(void)
3669 {
3670 	printk(KERN_INFO "Solarflare NET driver unloading\n");
3671 
3672 	pci_unregister_driver(&efx_pci_driver);
3673 	destroy_workqueue(reset_workqueue);
3674 #ifdef CONFIG_SFC_SRIOV
3675 	efx_fini_sriov();
3676 #endif
3677 	unregister_netdevice_notifier(&efx_netdev_notifier);
3678 
3679 }
3680 
3681 module_init(efx_init_module);
3682 module_exit(efx_exit_module);
3683 
3684 MODULE_AUTHOR("Solarflare Communications and "
3685 	      "Michael Brown <mbrown@fensystems.co.uk>");
3686 MODULE_DESCRIPTION("Solarflare network driver");
3687 MODULE_LICENSE("GPL");
3688 MODULE_DEVICE_TABLE(pci, efx_pci_table);
3689 MODULE_VERSION(EFX_DRIVER_VERSION);
3690