xref: /openbmc/linux/drivers/net/ethernet/sfc/efx.c (revision bc5aa3a0)
1 /****************************************************************************
2  * Driver for Solarflare network controllers and boards
3  * Copyright 2005-2006 Fen Systems Ltd.
4  * Copyright 2005-2013 Solarflare Communications Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License version 2 as published
8  * by the Free Software Foundation, incorporated herein by reference.
9  */
10 
11 #include <linux/module.h>
12 #include <linux/pci.h>
13 #include <linux/netdevice.h>
14 #include <linux/etherdevice.h>
15 #include <linux/delay.h>
16 #include <linux/notifier.h>
17 #include <linux/ip.h>
18 #include <linux/tcp.h>
19 #include <linux/in.h>
20 #include <linux/ethtool.h>
21 #include <linux/topology.h>
22 #include <linux/gfp.h>
23 #include <linux/aer.h>
24 #include <linux/interrupt.h>
25 #include "net_driver.h"
26 #include "efx.h"
27 #include "nic.h"
28 #include "selftest.h"
29 #include "sriov.h"
30 
31 #include "mcdi.h"
32 #include "workarounds.h"
33 
34 /**************************************************************************
35  *
36  * Type name strings
37  *
38  **************************************************************************
39  */
40 
41 /* Loopback mode names (see LOOPBACK_MODE()) */
42 const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
43 const char *const efx_loopback_mode_names[] = {
44 	[LOOPBACK_NONE]		= "NONE",
45 	[LOOPBACK_DATA]		= "DATAPATH",
46 	[LOOPBACK_GMAC]		= "GMAC",
47 	[LOOPBACK_XGMII]	= "XGMII",
48 	[LOOPBACK_XGXS]		= "XGXS",
49 	[LOOPBACK_XAUI]		= "XAUI",
50 	[LOOPBACK_GMII]		= "GMII",
51 	[LOOPBACK_SGMII]	= "SGMII",
52 	[LOOPBACK_XGBR]		= "XGBR",
53 	[LOOPBACK_XFI]		= "XFI",
54 	[LOOPBACK_XAUI_FAR]	= "XAUI_FAR",
55 	[LOOPBACK_GMII_FAR]	= "GMII_FAR",
56 	[LOOPBACK_SGMII_FAR]	= "SGMII_FAR",
57 	[LOOPBACK_XFI_FAR]	= "XFI_FAR",
58 	[LOOPBACK_GPHY]		= "GPHY",
59 	[LOOPBACK_PHYXS]	= "PHYXS",
60 	[LOOPBACK_PCS]		= "PCS",
61 	[LOOPBACK_PMAPMD]	= "PMA/PMD",
62 	[LOOPBACK_XPORT]	= "XPORT",
63 	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
64 	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
65 	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
66 	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
67 	[LOOPBACK_GMII_WS]	= "GMII_WS",
68 	[LOOPBACK_XFI_WS]	= "XFI_WS",
69 	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
70 	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
71 };
72 
73 const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
74 const char *const efx_reset_type_names[] = {
75 	[RESET_TYPE_INVISIBLE]          = "INVISIBLE",
76 	[RESET_TYPE_ALL]                = "ALL",
77 	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
78 	[RESET_TYPE_WORLD]              = "WORLD",
79 	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
80 	[RESET_TYPE_DATAPATH]           = "DATAPATH",
81 	[RESET_TYPE_MC_BIST]		= "MC_BIST",
82 	[RESET_TYPE_DISABLE]            = "DISABLE",
83 	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
84 	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
85 	[RESET_TYPE_RX_RECOVERY]        = "RX_RECOVERY",
86 	[RESET_TYPE_DMA_ERROR]          = "DMA_ERROR",
87 	[RESET_TYPE_TX_SKIP]            = "TX_SKIP",
88 	[RESET_TYPE_MC_FAILURE]         = "MC_FAILURE",
89 	[RESET_TYPE_MCDI_TIMEOUT]	= "MCDI_TIMEOUT (FLR)",
90 };
91 
92 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
93  * queued onto this work queue. This is not a per-nic work queue, because
94  * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
95  */
96 static struct workqueue_struct *reset_workqueue;
97 
98 /* How often and how many times to poll for a reset while waiting for a
99  * BIST that another function started to complete.
100  */
101 #define BIST_WAIT_DELAY_MS	100
102 #define BIST_WAIT_DELAY_COUNT	100
103 
104 /**************************************************************************
105  *
106  * Configurable values
107  *
108  *************************************************************************/
109 
110 /*
111  * Use separate channels for TX and RX events
112  *
113  * Set this to 1 to use separate channels for TX and RX. It allows us
114  * to control interrupt affinity separately for TX and RX.
115  *
116  * This is only used in MSI-X interrupt mode
117  */
118 bool efx_separate_tx_channels;
119 module_param(efx_separate_tx_channels, bool, 0444);
120 MODULE_PARM_DESC(efx_separate_tx_channels,
121 		 "Use separate channels for TX and RX");
122 
123 /* This is the weight assigned to each of the (per-channel) virtual
124  * NAPI devices.
125  */
126 static int napi_weight = 64;
127 
128 /* This is the time (in jiffies) between invocations of the hardware
129  * monitor.
130  * On Falcon-based NICs, this will:
131  * - Check the on-board hardware monitor;
132  * - Poll the link state and reconfigure the hardware as necessary.
133  * On Siena-based NICs for power systems with EEH support, this will give EEH a
134  * chance to start.
135  */
136 static unsigned int efx_monitor_interval = 1 * HZ;
137 
138 /* Initial interrupt moderation settings.  They can be modified after
139  * module load with ethtool.
140  *
141  * The default for RX should strike a balance between increasing the
142  * round-trip latency and reducing overhead.
143  */
144 static unsigned int rx_irq_mod_usec = 60;
145 
146 /* Initial interrupt moderation settings.  They can be modified after
147  * module load with ethtool.
148  *
149  * This default is chosen to ensure that a 10G link does not go idle
150  * while a TX queue is stopped after it has become full.  A queue is
151  * restarted when it drops below half full.  The time this takes (assuming
152  * worst case 3 descriptors per packet and 1024 descriptors) is
153  *   512 / 3 * 1.2 = 205 usec.
154  */
155 static unsigned int tx_irq_mod_usec = 150;
156 
157 /* This is the first interrupt mode to try out of:
158  * 0 => MSI-X
159  * 1 => MSI
160  * 2 => legacy
161  */
162 static unsigned int interrupt_mode;
163 
164 /* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
165  * i.e. the number of CPUs among which we may distribute simultaneous
166  * interrupt handling.
167  *
168  * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
169  * The default (0) means to assign an interrupt to each core.
170  */
171 static unsigned int rss_cpus;
172 module_param(rss_cpus, uint, 0444);
173 MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");
174 
175 static bool phy_flash_cfg;
176 module_param(phy_flash_cfg, bool, 0644);
177 MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");
178 
179 static unsigned irq_adapt_low_thresh = 8000;
180 module_param(irq_adapt_low_thresh, uint, 0644);
181 MODULE_PARM_DESC(irq_adapt_low_thresh,
182 		 "Threshold score for reducing IRQ moderation");
183 
184 static unsigned irq_adapt_high_thresh = 16000;
185 module_param(irq_adapt_high_thresh, uint, 0644);
186 MODULE_PARM_DESC(irq_adapt_high_thresh,
187 		 "Threshold score for increasing IRQ moderation");
188 
189 static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
190 			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
191 			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
192 			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
193 module_param(debug, uint, 0);
194 MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
195 
196 /**************************************************************************
197  *
198  * Utility functions and prototypes
199  *
200  *************************************************************************/
201 
202 static int efx_soft_enable_interrupts(struct efx_nic *efx);
203 static void efx_soft_disable_interrupts(struct efx_nic *efx);
204 static void efx_remove_channel(struct efx_channel *channel);
205 static void efx_remove_channels(struct efx_nic *efx);
206 static const struct efx_channel_type efx_default_channel_type;
207 static void efx_remove_port(struct efx_nic *efx);
208 static void efx_init_napi_channel(struct efx_channel *channel);
209 static void efx_fini_napi(struct efx_nic *efx);
210 static void efx_fini_napi_channel(struct efx_channel *channel);
211 static void efx_fini_struct(struct efx_nic *efx);
212 static void efx_start_all(struct efx_nic *efx);
213 static void efx_stop_all(struct efx_nic *efx);
214 
215 #define EFX_ASSERT_RESET_SERIALISED(efx)		\
216 	do {						\
217 		if ((efx->state == STATE_READY) ||	\
218 		    (efx->state == STATE_RECOVERY) ||	\
219 		    (efx->state == STATE_DISABLED))	\
220 			ASSERT_RTNL();			\
221 	} while (0)
222 
223 static int efx_check_disabled(struct efx_nic *efx)
224 {
225 	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
226 		netif_err(efx, drv, efx->net_dev,
227 			  "device is disabled due to earlier errors\n");
228 		return -EIO;
229 	}
230 	return 0;
231 }
232 
233 /**************************************************************************
234  *
235  * Event queue processing
236  *
237  *************************************************************************/
238 
239 /* Process channel's event queue
240  *
241  * This function is responsible for processing the event queue of a
242  * single channel.  The caller must guarantee that this function will
243  * never be concurrently called more than once on the same channel,
244  * though different channels may be being processed concurrently.
245  */
246 static int efx_process_channel(struct efx_channel *channel, int budget)
247 {
248 	struct efx_tx_queue *tx_queue;
249 	int spent;
250 
251 	if (unlikely(!channel->enabled))
252 		return 0;
253 
254 	efx_for_each_channel_tx_queue(tx_queue, channel) {
255 		tx_queue->pkts_compl = 0;
256 		tx_queue->bytes_compl = 0;
257 	}
258 
259 	spent = efx_nic_process_eventq(channel, budget);
260 	if (spent && efx_channel_has_rx_queue(channel)) {
261 		struct efx_rx_queue *rx_queue =
262 			efx_channel_get_rx_queue(channel);
263 
264 		efx_rx_flush_packet(channel);
265 		efx_fast_push_rx_descriptors(rx_queue, true);
266 	}
267 
268 	/* Update BQL */
269 	efx_for_each_channel_tx_queue(tx_queue, channel) {
270 		if (tx_queue->bytes_compl) {
271 			netdev_tx_completed_queue(tx_queue->core_txq,
272 				tx_queue->pkts_compl, tx_queue->bytes_compl);
273 		}
274 	}
275 
276 	return spent;
277 }
278 
279 /* NAPI poll handler
280  *
281  * NAPI guarantees serialisation of polls of the same device, which
282  * provides the guarantee required by efx_process_channel().
283  */
284 static int efx_poll(struct napi_struct *napi, int budget)
285 {
286 	struct efx_channel *channel =
287 		container_of(napi, struct efx_channel, napi_str);
288 	struct efx_nic *efx = channel->efx;
289 	int spent;
290 
291 	if (!efx_channel_lock_napi(channel))
292 		return budget;
293 
294 	netif_vdbg(efx, intr, efx->net_dev,
295 		   "channel %d NAPI poll executing on CPU %d\n",
296 		   channel->channel, raw_smp_processor_id());
297 
298 	spent = efx_process_channel(channel, budget);
299 
300 	if (spent < budget) {
301 		if (efx_channel_has_rx_queue(channel) &&
302 		    efx->irq_rx_adaptive &&
303 		    unlikely(++channel->irq_count == 1000)) {
304 			if (unlikely(channel->irq_mod_score <
305 				     irq_adapt_low_thresh)) {
306 				if (channel->irq_moderation > 1) {
307 					channel->irq_moderation -= 1;
308 					efx->type->push_irq_moderation(channel);
309 				}
310 			} else if (unlikely(channel->irq_mod_score >
311 					    irq_adapt_high_thresh)) {
312 				if (channel->irq_moderation <
313 				    efx->irq_rx_moderation) {
314 					channel->irq_moderation += 1;
315 					efx->type->push_irq_moderation(channel);
316 				}
317 			}
318 			channel->irq_count = 0;
319 			channel->irq_mod_score = 0;
320 		}
321 
322 		efx_filter_rfs_expire(channel);
323 
324 		/* There is no race here; although napi_disable() will
325 		 * only wait for napi_complete(), this isn't a problem
326 		 * since efx_nic_eventq_read_ack() will have no effect if
327 		 * interrupts have already been disabled.
328 		 */
329 		napi_complete(napi);
330 		efx_nic_eventq_read_ack(channel);
331 	}
332 
333 	efx_channel_unlock_napi(channel);
334 	return spent;
335 }
336 
337 /* Create event queue
338  * Event queue memory allocations are done only once.  If the channel
339  * is reset, the memory buffer will be reused; this guards against
340  * errors during channel reset and also simplifies interrupt handling.
341  */
342 static int efx_probe_eventq(struct efx_channel *channel)
343 {
344 	struct efx_nic *efx = channel->efx;
345 	unsigned long entries;
346 
347 	netif_dbg(efx, probe, efx->net_dev,
348 		  "chan %d create event queue\n", channel->channel);
349 
350 	/* Build an event queue with room for one event per tx and rx buffer,
351 	 * plus some extra for link state events and MCDI completions. */
352 	entries = roundup_pow_of_two(efx->rxq_entries + efx->txq_entries + 128);
353 	EFX_BUG_ON_PARANOID(entries > EFX_MAX_EVQ_SIZE);
354 	channel->eventq_mask = max(entries, EFX_MIN_EVQ_SIZE) - 1;
355 
356 	return efx_nic_probe_eventq(channel);
357 }
358 
359 /* Prepare channel's event queue */
360 static int efx_init_eventq(struct efx_channel *channel)
361 {
362 	struct efx_nic *efx = channel->efx;
363 	int rc;
364 
365 	EFX_WARN_ON_PARANOID(channel->eventq_init);
366 
367 	netif_dbg(efx, drv, efx->net_dev,
368 		  "chan %d init event queue\n", channel->channel);
369 
370 	rc = efx_nic_init_eventq(channel);
371 	if (rc == 0) {
372 		efx->type->push_irq_moderation(channel);
373 		channel->eventq_read_ptr = 0;
374 		channel->eventq_init = true;
375 	}
376 	return rc;
377 }
378 
379 /* Enable event queue processing and NAPI */
380 void efx_start_eventq(struct efx_channel *channel)
381 {
382 	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
383 		  "chan %d start event queue\n", channel->channel);
384 
385 	/* Make sure the NAPI handler sees the enabled flag set */
386 	channel->enabled = true;
387 	smp_wmb();
388 
389 	efx_channel_enable(channel);
390 	napi_enable(&channel->napi_str);
391 	efx_nic_eventq_read_ack(channel);
392 }
393 
394 /* Disable event queue processing and NAPI */
395 void efx_stop_eventq(struct efx_channel *channel)
396 {
397 	if (!channel->enabled)
398 		return;
399 
400 	napi_disable(&channel->napi_str);
401 	while (!efx_channel_disable(channel))
402 		usleep_range(1000, 20000);
403 	channel->enabled = false;
404 }
405 
406 static void efx_fini_eventq(struct efx_channel *channel)
407 {
408 	if (!channel->eventq_init)
409 		return;
410 
411 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
412 		  "chan %d fini event queue\n", channel->channel);
413 
414 	efx_nic_fini_eventq(channel);
415 	channel->eventq_init = false;
416 }
417 
418 static void efx_remove_eventq(struct efx_channel *channel)
419 {
420 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
421 		  "chan %d remove event queue\n", channel->channel);
422 
423 	efx_nic_remove_eventq(channel);
424 }
425 
426 /**************************************************************************
427  *
428  * Channel handling
429  *
430  *************************************************************************/
431 
432 /* Allocate and initialise a channel structure. */
433 static struct efx_channel *
434 efx_alloc_channel(struct efx_nic *efx, int i, struct efx_channel *old_channel)
435 {
436 	struct efx_channel *channel;
437 	struct efx_rx_queue *rx_queue;
438 	struct efx_tx_queue *tx_queue;
439 	int j;
440 
441 	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
442 	if (!channel)
443 		return NULL;
444 
445 	channel->efx = efx;
446 	channel->channel = i;
447 	channel->type = &efx_default_channel_type;
448 
449 	for (j = 0; j < EFX_TXQ_TYPES; j++) {
450 		tx_queue = &channel->tx_queue[j];
451 		tx_queue->efx = efx;
452 		tx_queue->queue = i * EFX_TXQ_TYPES + j;
453 		tx_queue->channel = channel;
454 	}
455 
456 	rx_queue = &channel->rx_queue;
457 	rx_queue->efx = efx;
458 	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
459 		    (unsigned long)rx_queue);
460 
461 	return channel;
462 }
463 
464 /* Allocate and initialise a channel structure, copying parameters
465  * (but not resources) from an old channel structure.
466  */
467 static struct efx_channel *
468 efx_copy_channel(const struct efx_channel *old_channel)
469 {
470 	struct efx_channel *channel;
471 	struct efx_rx_queue *rx_queue;
472 	struct efx_tx_queue *tx_queue;
473 	int j;
474 
475 	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
476 	if (!channel)
477 		return NULL;
478 
479 	*channel = *old_channel;
480 
481 	channel->napi_dev = NULL;
482 	memset(&channel->eventq, 0, sizeof(channel->eventq));
483 
484 	for (j = 0; j < EFX_TXQ_TYPES; j++) {
485 		tx_queue = &channel->tx_queue[j];
486 		if (tx_queue->channel)
487 			tx_queue->channel = channel;
488 		tx_queue->buffer = NULL;
489 		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
490 	}
491 
492 	rx_queue = &channel->rx_queue;
493 	rx_queue->buffer = NULL;
494 	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
495 	setup_timer(&rx_queue->slow_fill, efx_rx_slow_fill,
496 		    (unsigned long)rx_queue);
497 
498 	return channel;
499 }
500 
501 static int efx_probe_channel(struct efx_channel *channel)
502 {
503 	struct efx_tx_queue *tx_queue;
504 	struct efx_rx_queue *rx_queue;
505 	int rc;
506 
507 	netif_dbg(channel->efx, probe, channel->efx->net_dev,
508 		  "creating channel %d\n", channel->channel);
509 
510 	rc = channel->type->pre_probe(channel);
511 	if (rc)
512 		goto fail;
513 
514 	rc = efx_probe_eventq(channel);
515 	if (rc)
516 		goto fail;
517 
518 	efx_for_each_channel_tx_queue(tx_queue, channel) {
519 		rc = efx_probe_tx_queue(tx_queue);
520 		if (rc)
521 			goto fail;
522 	}
523 
524 	efx_for_each_channel_rx_queue(rx_queue, channel) {
525 		rc = efx_probe_rx_queue(rx_queue);
526 		if (rc)
527 			goto fail;
528 	}
529 
530 	return 0;
531 
532 fail:
533 	efx_remove_channel(channel);
534 	return rc;
535 }
536 
537 static void
538 efx_get_channel_name(struct efx_channel *channel, char *buf, size_t len)
539 {
540 	struct efx_nic *efx = channel->efx;
541 	const char *type;
542 	int number;
543 
544 	number = channel->channel;
545 	if (efx->tx_channel_offset == 0) {
546 		type = "";
547 	} else if (channel->channel < efx->tx_channel_offset) {
548 		type = "-rx";
549 	} else {
550 		type = "-tx";
551 		number -= efx->tx_channel_offset;
552 	}
553 	snprintf(buf, len, "%s%s-%d", efx->name, type, number);
554 }
555 
556 static void efx_set_channel_names(struct efx_nic *efx)
557 {
558 	struct efx_channel *channel;
559 
560 	efx_for_each_channel(channel, efx)
561 		channel->type->get_name(channel,
562 					efx->msi_context[channel->channel].name,
563 					sizeof(efx->msi_context[0].name));
564 }
565 
566 static int efx_probe_channels(struct efx_nic *efx)
567 {
568 	struct efx_channel *channel;
569 	int rc;
570 
571 	/* Restart special buffer allocation */
572 	efx->next_buffer_table = 0;
573 
574 	/* Probe channels in reverse, so that any 'extra' channels
575 	 * use the start of the buffer table. This allows the traffic
576 	 * channels to be resized without moving them or wasting the
577 	 * entries before them.
578 	 */
579 	efx_for_each_channel_rev(channel, efx) {
580 		rc = efx_probe_channel(channel);
581 		if (rc) {
582 			netif_err(efx, probe, efx->net_dev,
583 				  "failed to create channel %d\n",
584 				  channel->channel);
585 			goto fail;
586 		}
587 	}
588 	efx_set_channel_names(efx);
589 
590 	return 0;
591 
592 fail:
593 	efx_remove_channels(efx);
594 	return rc;
595 }
596 
597 /* Channels are shutdown and reinitialised whilst the NIC is running
598  * to propagate configuration changes (mtu, checksum offload), or
599  * to clear hardware error conditions
600  */
601 static void efx_start_datapath(struct efx_nic *efx)
602 {
603 	netdev_features_t old_features = efx->net_dev->features;
604 	bool old_rx_scatter = efx->rx_scatter;
605 	struct efx_tx_queue *tx_queue;
606 	struct efx_rx_queue *rx_queue;
607 	struct efx_channel *channel;
608 	size_t rx_buf_len;
609 
610 	/* Calculate the rx buffer allocation parameters required to
611 	 * support the current MTU, including padding for header
612 	 * alignment and overruns.
613 	 */
614 	efx->rx_dma_len = (efx->rx_prefix_size +
615 			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
616 			   efx->type->rx_buffer_padding);
617 	rx_buf_len = (sizeof(struct efx_rx_page_state) +
618 		      efx->rx_ip_align + efx->rx_dma_len);
619 	if (rx_buf_len <= PAGE_SIZE) {
620 		efx->rx_scatter = efx->type->always_rx_scatter;
621 		efx->rx_buffer_order = 0;
622 	} else if (efx->type->can_rx_scatter) {
623 		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
624 		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
625 			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
626 				       EFX_RX_BUF_ALIGNMENT) >
627 			     PAGE_SIZE);
628 		efx->rx_scatter = true;
629 		efx->rx_dma_len = EFX_RX_USR_BUF_SIZE;
630 		efx->rx_buffer_order = 0;
631 	} else {
632 		efx->rx_scatter = false;
633 		efx->rx_buffer_order = get_order(rx_buf_len);
634 	}
635 
636 	efx_rx_config_page_split(efx);
637 	if (efx->rx_buffer_order)
638 		netif_dbg(efx, drv, efx->net_dev,
639 			  "RX buf len=%u; page order=%u batch=%u\n",
640 			  efx->rx_dma_len, efx->rx_buffer_order,
641 			  efx->rx_pages_per_batch);
642 	else
643 		netif_dbg(efx, drv, efx->net_dev,
644 			  "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
645 			  efx->rx_dma_len, efx->rx_page_buf_step,
646 			  efx->rx_bufs_per_page, efx->rx_pages_per_batch);
647 
648 	/* Restore previously fixed features in hw_features and remove
649 	 * features which are fixed now
650 	 */
651 	efx->net_dev->hw_features |= efx->net_dev->features;
652 	efx->net_dev->hw_features &= ~efx->fixed_features;
653 	efx->net_dev->features |= efx->fixed_features;
654 	if (efx->net_dev->features != old_features)
655 		netdev_features_change(efx->net_dev);
656 
657 	/* RX filters may also have scatter-enabled flags */
658 	if (efx->rx_scatter != old_rx_scatter)
659 		efx->type->filter_update_rx_scatter(efx);
660 
661 	/* We must keep at least one descriptor in a TX ring empty.
662 	 * We could avoid this when the queue size does not exactly
663 	 * match the hardware ring size, but it's not that important.
664 	 * Therefore we stop the queue when one more skb might fill
665 	 * the ring completely.  We wake it when half way back to
666 	 * empty.
667 	 */
668 	efx->txq_stop_thresh = efx->txq_entries - efx_tx_max_skb_descs(efx);
669 	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;
670 
671 	/* Initialise the channels */
672 	efx_for_each_channel(channel, efx) {
673 		efx_for_each_channel_tx_queue(tx_queue, channel) {
674 			efx_init_tx_queue(tx_queue);
675 			atomic_inc(&efx->active_queues);
676 		}
677 
678 		efx_for_each_channel_rx_queue(rx_queue, channel) {
679 			efx_init_rx_queue(rx_queue);
680 			atomic_inc(&efx->active_queues);
681 			efx_stop_eventq(channel);
682 			efx_fast_push_rx_descriptors(rx_queue, false);
683 			efx_start_eventq(channel);
684 		}
685 
686 		WARN_ON(channel->rx_pkt_n_frags);
687 	}
688 
689 	efx_ptp_start_datapath(efx);
690 
691 	if (netif_device_present(efx->net_dev))
692 		netif_tx_wake_all_queues(efx->net_dev);
693 }
694 
695 static void efx_stop_datapath(struct efx_nic *efx)
696 {
697 	struct efx_channel *channel;
698 	struct efx_tx_queue *tx_queue;
699 	struct efx_rx_queue *rx_queue;
700 	int rc;
701 
702 	EFX_ASSERT_RESET_SERIALISED(efx);
703 	BUG_ON(efx->port_enabled);
704 
705 	efx_ptp_stop_datapath(efx);
706 
707 	/* Stop RX refill */
708 	efx_for_each_channel(channel, efx) {
709 		efx_for_each_channel_rx_queue(rx_queue, channel)
710 			rx_queue->refill_enabled = false;
711 	}
712 
713 	efx_for_each_channel(channel, efx) {
714 		/* RX packet processing is pipelined, so wait for the
715 		 * NAPI handler to complete.  At least event queue 0
716 		 * might be kept active by non-data events, so don't
717 		 * use napi_synchronize() but actually disable NAPI
718 		 * temporarily.
719 		 */
720 		if (efx_channel_has_rx_queue(channel)) {
721 			efx_stop_eventq(channel);
722 			efx_start_eventq(channel);
723 		}
724 	}
725 
726 	rc = efx->type->fini_dmaq(efx);
727 	if (rc && EFX_WORKAROUND_7803(efx)) {
728 		/* Schedule a reset to recover from the flush failure. The
729 		 * descriptor caches reference memory we're about to free,
730 		 * but falcon_reconfigure_mac_wrapper() won't reconnect
731 		 * the MACs because of the pending reset.
732 		 */
733 		netif_err(efx, drv, efx->net_dev,
734 			  "Resetting to recover from flush failure\n");
735 		efx_schedule_reset(efx, RESET_TYPE_ALL);
736 	} else if (rc) {
737 		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
738 	} else {
739 		netif_dbg(efx, drv, efx->net_dev,
740 			  "successfully flushed all queues\n");
741 	}
742 
743 	efx_for_each_channel(channel, efx) {
744 		efx_for_each_channel_rx_queue(rx_queue, channel)
745 			efx_fini_rx_queue(rx_queue);
746 		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
747 			efx_fini_tx_queue(tx_queue);
748 	}
749 }
750 
751 static void efx_remove_channel(struct efx_channel *channel)
752 {
753 	struct efx_tx_queue *tx_queue;
754 	struct efx_rx_queue *rx_queue;
755 
756 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
757 		  "destroy chan %d\n", channel->channel);
758 
759 	efx_for_each_channel_rx_queue(rx_queue, channel)
760 		efx_remove_rx_queue(rx_queue);
761 	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
762 		efx_remove_tx_queue(tx_queue);
763 	efx_remove_eventq(channel);
764 	channel->type->post_remove(channel);
765 }
766 
767 static void efx_remove_channels(struct efx_nic *efx)
768 {
769 	struct efx_channel *channel;
770 
771 	efx_for_each_channel(channel, efx)
772 		efx_remove_channel(channel);
773 }
774 
775 int
776 efx_realloc_channels(struct efx_nic *efx, u32 rxq_entries, u32 txq_entries)
777 {
778 	struct efx_channel *other_channel[EFX_MAX_CHANNELS], *channel;
779 	u32 old_rxq_entries, old_txq_entries;
780 	unsigned i, next_buffer_table = 0;
781 	int rc, rc2;
782 
783 	rc = efx_check_disabled(efx);
784 	if (rc)
785 		return rc;
786 
787 	/* Not all channels should be reallocated. We must avoid
788 	 * reallocating their buffer table entries.
789 	 */
790 	efx_for_each_channel(channel, efx) {
791 		struct efx_rx_queue *rx_queue;
792 		struct efx_tx_queue *tx_queue;
793 
794 		if (channel->type->copy)
795 			continue;
796 		next_buffer_table = max(next_buffer_table,
797 					channel->eventq.index +
798 					channel->eventq.entries);
799 		efx_for_each_channel_rx_queue(rx_queue, channel)
800 			next_buffer_table = max(next_buffer_table,
801 						rx_queue->rxd.index +
802 						rx_queue->rxd.entries);
803 		efx_for_each_channel_tx_queue(tx_queue, channel)
804 			next_buffer_table = max(next_buffer_table,
805 						tx_queue->txd.index +
806 						tx_queue->txd.entries);
807 	}
808 
809 	efx_device_detach_sync(efx);
810 	efx_stop_all(efx);
811 	efx_soft_disable_interrupts(efx);
812 
813 	/* Clone channels (where possible) */
814 	memset(other_channel, 0, sizeof(other_channel));
815 	for (i = 0; i < efx->n_channels; i++) {
816 		channel = efx->channel[i];
817 		if (channel->type->copy)
818 			channel = channel->type->copy(channel);
819 		if (!channel) {
820 			rc = -ENOMEM;
821 			goto out;
822 		}
823 		other_channel[i] = channel;
824 	}
825 
826 	/* Swap entry counts and channel pointers */
827 	old_rxq_entries = efx->rxq_entries;
828 	old_txq_entries = efx->txq_entries;
829 	efx->rxq_entries = rxq_entries;
830 	efx->txq_entries = txq_entries;
831 	for (i = 0; i < efx->n_channels; i++) {
832 		channel = efx->channel[i];
833 		efx->channel[i] = other_channel[i];
834 		other_channel[i] = channel;
835 	}
836 
837 	/* Restart buffer table allocation */
838 	efx->next_buffer_table = next_buffer_table;
839 
840 	for (i = 0; i < efx->n_channels; i++) {
841 		channel = efx->channel[i];
842 		if (!channel->type->copy)
843 			continue;
844 		rc = efx_probe_channel(channel);
845 		if (rc)
846 			goto rollback;
847 		efx_init_napi_channel(efx->channel[i]);
848 	}
849 
850 out:
851 	/* Destroy unused channel structures */
852 	for (i = 0; i < efx->n_channels; i++) {
853 		channel = other_channel[i];
854 		if (channel && channel->type->copy) {
855 			efx_fini_napi_channel(channel);
856 			efx_remove_channel(channel);
857 			kfree(channel);
858 		}
859 	}
860 
861 	rc2 = efx_soft_enable_interrupts(efx);
862 	if (rc2) {
863 		rc = rc ? rc : rc2;
864 		netif_err(efx, drv, efx->net_dev,
865 			  "unable to restart interrupts on channel reallocation\n");
866 		efx_schedule_reset(efx, RESET_TYPE_DISABLE);
867 	} else {
868 		efx_start_all(efx);
869 		netif_device_attach(efx->net_dev);
870 	}
871 	return rc;
872 
873 rollback:
874 	/* Swap back */
875 	efx->rxq_entries = old_rxq_entries;
876 	efx->txq_entries = old_txq_entries;
877 	for (i = 0; i < efx->n_channels; i++) {
878 		channel = efx->channel[i];
879 		efx->channel[i] = other_channel[i];
880 		other_channel[i] = channel;
881 	}
882 	goto out;
883 }
884 
885 void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
886 {
887 	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(100));
888 }
889 
890 static const struct efx_channel_type efx_default_channel_type = {
891 	.pre_probe		= efx_channel_dummy_op_int,
892 	.post_remove		= efx_channel_dummy_op_void,
893 	.get_name		= efx_get_channel_name,
894 	.copy			= efx_copy_channel,
895 	.keep_eventq		= false,
896 };
897 
898 int efx_channel_dummy_op_int(struct efx_channel *channel)
899 {
900 	return 0;
901 }
902 
903 void efx_channel_dummy_op_void(struct efx_channel *channel)
904 {
905 }
906 
907 /**************************************************************************
908  *
909  * Port handling
910  *
911  **************************************************************************/
912 
913 /* This ensures that the kernel is kept informed (via
914  * netif_carrier_on/off) of the link status, and also maintains the
915  * link status's stop on the port's TX queue.
916  */
917 void efx_link_status_changed(struct efx_nic *efx)
918 {
919 	struct efx_link_state *link_state = &efx->link_state;
920 
921 	/* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
922 	 * that no events are triggered between unregister_netdev() and the
923 	 * driver unloading. A more general condition is that NETDEV_CHANGE
924 	 * can only be generated between NETDEV_UP and NETDEV_DOWN */
925 	if (!netif_running(efx->net_dev))
926 		return;
927 
928 	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
929 		efx->n_link_state_changes++;
930 
931 		if (link_state->up)
932 			netif_carrier_on(efx->net_dev);
933 		else
934 			netif_carrier_off(efx->net_dev);
935 	}
936 
937 	/* Status message for kernel log */
938 	if (link_state->up)
939 		netif_info(efx, link, efx->net_dev,
940 			   "link up at %uMbps %s-duplex (MTU %d)\n",
941 			   link_state->speed, link_state->fd ? "full" : "half",
942 			   efx->net_dev->mtu);
943 	else
944 		netif_info(efx, link, efx->net_dev, "link down\n");
945 }
946 
947 void efx_link_set_advertising(struct efx_nic *efx, u32 advertising)
948 {
949 	efx->link_advertising = advertising;
950 	if (advertising) {
951 		if (advertising & ADVERTISED_Pause)
952 			efx->wanted_fc |= (EFX_FC_TX | EFX_FC_RX);
953 		else
954 			efx->wanted_fc &= ~(EFX_FC_TX | EFX_FC_RX);
955 		if (advertising & ADVERTISED_Asym_Pause)
956 			efx->wanted_fc ^= EFX_FC_TX;
957 	}
958 }
959 
960 void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
961 {
962 	efx->wanted_fc = wanted_fc;
963 	if (efx->link_advertising) {
964 		if (wanted_fc & EFX_FC_RX)
965 			efx->link_advertising |= (ADVERTISED_Pause |
966 						  ADVERTISED_Asym_Pause);
967 		else
968 			efx->link_advertising &= ~(ADVERTISED_Pause |
969 						   ADVERTISED_Asym_Pause);
970 		if (wanted_fc & EFX_FC_TX)
971 			efx->link_advertising ^= ADVERTISED_Asym_Pause;
972 	}
973 }
974 
975 static void efx_fini_port(struct efx_nic *efx);
976 
977 /* We assume that efx->type->reconfigure_mac will always try to sync RX
978  * filters and therefore needs to read-lock the filter table against freeing
979  */
980 void efx_mac_reconfigure(struct efx_nic *efx)
981 {
982 	down_read(&efx->filter_sem);
983 	efx->type->reconfigure_mac(efx);
984 	up_read(&efx->filter_sem);
985 }
986 
987 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
988  * the MAC appropriately. All other PHY configuration changes are pushed
989  * through phy_op->set_settings(), and pushed asynchronously to the MAC
990  * through efx_monitor().
991  *
992  * Callers must hold the mac_lock
993  */
994 int __efx_reconfigure_port(struct efx_nic *efx)
995 {
996 	enum efx_phy_mode phy_mode;
997 	int rc;
998 
999 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
1000 
1001 	/* Disable PHY transmit in mac level loopbacks */
1002 	phy_mode = efx->phy_mode;
1003 	if (LOOPBACK_INTERNAL(efx))
1004 		efx->phy_mode |= PHY_MODE_TX_DISABLED;
1005 	else
1006 		efx->phy_mode &= ~PHY_MODE_TX_DISABLED;
1007 
1008 	rc = efx->type->reconfigure_port(efx);
1009 
1010 	if (rc)
1011 		efx->phy_mode = phy_mode;
1012 
1013 	return rc;
1014 }
1015 
1016 /* Reinitialise the MAC to pick up new PHY settings, even if the port is
1017  * disabled. */
1018 int efx_reconfigure_port(struct efx_nic *efx)
1019 {
1020 	int rc;
1021 
1022 	EFX_ASSERT_RESET_SERIALISED(efx);
1023 
1024 	mutex_lock(&efx->mac_lock);
1025 	rc = __efx_reconfigure_port(efx);
1026 	mutex_unlock(&efx->mac_lock);
1027 
1028 	return rc;
1029 }
1030 
1031 /* Asynchronous work item for changing MAC promiscuity and multicast
1032  * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
1033  * MAC directly. */
1034 static void efx_mac_work(struct work_struct *data)
1035 {
1036 	struct efx_nic *efx = container_of(data, struct efx_nic, mac_work);
1037 
1038 	mutex_lock(&efx->mac_lock);
1039 	if (efx->port_enabled)
1040 		efx_mac_reconfigure(efx);
1041 	mutex_unlock(&efx->mac_lock);
1042 }
1043 
1044 static int efx_probe_port(struct efx_nic *efx)
1045 {
1046 	int rc;
1047 
1048 	netif_dbg(efx, probe, efx->net_dev, "create port\n");
1049 
1050 	if (phy_flash_cfg)
1051 		efx->phy_mode = PHY_MODE_SPECIAL;
1052 
1053 	/* Connect up MAC/PHY operations table */
1054 	rc = efx->type->probe_port(efx);
1055 	if (rc)
1056 		return rc;
1057 
1058 	/* Initialise MAC address to permanent address */
1059 	ether_addr_copy(efx->net_dev->dev_addr, efx->net_dev->perm_addr);
1060 
1061 	return 0;
1062 }
1063 
1064 static int efx_init_port(struct efx_nic *efx)
1065 {
1066 	int rc;
1067 
1068 	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1069 
1070 	mutex_lock(&efx->mac_lock);
1071 
1072 	rc = efx->phy_op->init(efx);
1073 	if (rc)
1074 		goto fail1;
1075 
1076 	efx->port_initialized = true;
1077 
1078 	/* Reconfigure the MAC before creating dma queues (required for
1079 	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1080 	efx_mac_reconfigure(efx);
1081 
1082 	/* Ensure the PHY advertises the correct flow control settings */
1083 	rc = efx->phy_op->reconfigure(efx);
1084 	if (rc && rc != -EPERM)
1085 		goto fail2;
1086 
1087 	mutex_unlock(&efx->mac_lock);
1088 	return 0;
1089 
1090 fail2:
1091 	efx->phy_op->fini(efx);
1092 fail1:
1093 	mutex_unlock(&efx->mac_lock);
1094 	return rc;
1095 }
1096 
1097 static void efx_start_port(struct efx_nic *efx)
1098 {
1099 	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1100 	BUG_ON(efx->port_enabled);
1101 
1102 	mutex_lock(&efx->mac_lock);
1103 	efx->port_enabled = true;
1104 
1105 	/* Ensure MAC ingress/egress is enabled */
1106 	efx_mac_reconfigure(efx);
1107 
1108 	mutex_unlock(&efx->mac_lock);
1109 }
1110 
1111 /* Cancel work for MAC reconfiguration, periodic hardware monitoring
1112  * and the async self-test, wait for them to finish and prevent them
1113  * being scheduled again.  This doesn't cover online resets, which
1114  * should only be cancelled when removing the device.
1115  */
1116 static void efx_stop_port(struct efx_nic *efx)
1117 {
1118 	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1119 
1120 	EFX_ASSERT_RESET_SERIALISED(efx);
1121 
1122 	mutex_lock(&efx->mac_lock);
1123 	efx->port_enabled = false;
1124 	mutex_unlock(&efx->mac_lock);
1125 
1126 	/* Serialise against efx_set_multicast_list() */
1127 	netif_addr_lock_bh(efx->net_dev);
1128 	netif_addr_unlock_bh(efx->net_dev);
1129 
1130 	cancel_delayed_work_sync(&efx->monitor_work);
1131 	efx_selftest_async_cancel(efx);
1132 	cancel_work_sync(&efx->mac_work);
1133 }
1134 
1135 static void efx_fini_port(struct efx_nic *efx)
1136 {
1137 	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1138 
1139 	if (!efx->port_initialized)
1140 		return;
1141 
1142 	efx->phy_op->fini(efx);
1143 	efx->port_initialized = false;
1144 
1145 	efx->link_state.up = false;
1146 	efx_link_status_changed(efx);
1147 }
1148 
1149 static void efx_remove_port(struct efx_nic *efx)
1150 {
1151 	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1152 
1153 	efx->type->remove_port(efx);
1154 }
1155 
1156 /**************************************************************************
1157  *
1158  * NIC handling
1159  *
1160  **************************************************************************/
1161 
1162 static LIST_HEAD(efx_primary_list);
1163 static LIST_HEAD(efx_unassociated_list);
1164 
1165 static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right)
1166 {
1167 	return left->type == right->type &&
1168 		left->vpd_sn && right->vpd_sn &&
1169 		!strcmp(left->vpd_sn, right->vpd_sn);
1170 }
1171 
1172 static void efx_associate(struct efx_nic *efx)
1173 {
1174 	struct efx_nic *other, *next;
1175 
1176 	if (efx->primary == efx) {
1177 		/* Adding primary function; look for secondaries */
1178 
1179 		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
1180 		list_add_tail(&efx->node, &efx_primary_list);
1181 
1182 		list_for_each_entry_safe(other, next, &efx_unassociated_list,
1183 					 node) {
1184 			if (efx_same_controller(efx, other)) {
1185 				list_del(&other->node);
1186 				netif_dbg(other, probe, other->net_dev,
1187 					  "moving to secondary list of %s %s\n",
1188 					  pci_name(efx->pci_dev),
1189 					  efx->net_dev->name);
1190 				list_add_tail(&other->node,
1191 					      &efx->secondary_list);
1192 				other->primary = efx;
1193 			}
1194 		}
1195 	} else {
1196 		/* Adding secondary function; look for primary */
1197 
1198 		list_for_each_entry(other, &efx_primary_list, node) {
1199 			if (efx_same_controller(efx, other)) {
1200 				netif_dbg(efx, probe, efx->net_dev,
1201 					  "adding to secondary list of %s %s\n",
1202 					  pci_name(other->pci_dev),
1203 					  other->net_dev->name);
1204 				list_add_tail(&efx->node,
1205 					      &other->secondary_list);
1206 				efx->primary = other;
1207 				return;
1208 			}
1209 		}
1210 
1211 		netif_dbg(efx, probe, efx->net_dev,
1212 			  "adding to unassociated list\n");
1213 		list_add_tail(&efx->node, &efx_unassociated_list);
1214 	}
1215 }
1216 
1217 static void efx_dissociate(struct efx_nic *efx)
1218 {
1219 	struct efx_nic *other, *next;
1220 
1221 	list_del(&efx->node);
1222 	efx->primary = NULL;
1223 
1224 	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
1225 		list_del(&other->node);
1226 		netif_dbg(other, probe, other->net_dev,
1227 			  "moving to unassociated list\n");
1228 		list_add_tail(&other->node, &efx_unassociated_list);
1229 		other->primary = NULL;
1230 	}
1231 }
1232 
1233 /* This configures the PCI device to enable I/O and DMA. */
1234 static int efx_init_io(struct efx_nic *efx)
1235 {
1236 	struct pci_dev *pci_dev = efx->pci_dev;
1237 	dma_addr_t dma_mask = efx->type->max_dma_mask;
1238 	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1239 	int rc, bar;
1240 
1241 	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1242 
1243 	bar = efx->type->mem_bar;
1244 
1245 	rc = pci_enable_device(pci_dev);
1246 	if (rc) {
1247 		netif_err(efx, probe, efx->net_dev,
1248 			  "failed to enable PCI device\n");
1249 		goto fail1;
1250 	}
1251 
1252 	pci_set_master(pci_dev);
1253 
1254 	/* Set the PCI DMA mask.  Try all possibilities from our
1255 	 * genuine mask down to 32 bits, because some architectures
1256 	 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
1257 	 * masks event though they reject 46 bit masks.
1258 	 */
1259 	while (dma_mask > 0x7fffffffUL) {
1260 		rc = dma_set_mask_and_coherent(&pci_dev->dev, dma_mask);
1261 		if (rc == 0)
1262 			break;
1263 		dma_mask >>= 1;
1264 	}
1265 	if (rc) {
1266 		netif_err(efx, probe, efx->net_dev,
1267 			  "could not find a suitable DMA mask\n");
1268 		goto fail2;
1269 	}
1270 	netif_dbg(efx, probe, efx->net_dev,
1271 		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1272 
1273 	efx->membase_phys = pci_resource_start(efx->pci_dev, bar);
1274 	rc = pci_request_region(pci_dev, bar, "sfc");
1275 	if (rc) {
1276 		netif_err(efx, probe, efx->net_dev,
1277 			  "request for memory BAR failed\n");
1278 		rc = -EIO;
1279 		goto fail3;
1280 	}
1281 	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1282 	if (!efx->membase) {
1283 		netif_err(efx, probe, efx->net_dev,
1284 			  "could not map memory BAR at %llx+%x\n",
1285 			  (unsigned long long)efx->membase_phys, mem_map_size);
1286 		rc = -ENOMEM;
1287 		goto fail4;
1288 	}
1289 	netif_dbg(efx, probe, efx->net_dev,
1290 		  "memory BAR at %llx+%x (virtual %p)\n",
1291 		  (unsigned long long)efx->membase_phys, mem_map_size,
1292 		  efx->membase);
1293 
1294 	return 0;
1295 
1296  fail4:
1297 	pci_release_region(efx->pci_dev, bar);
1298  fail3:
1299 	efx->membase_phys = 0;
1300  fail2:
1301 	pci_disable_device(efx->pci_dev);
1302  fail1:
1303 	return rc;
1304 }
1305 
1306 static void efx_fini_io(struct efx_nic *efx)
1307 {
1308 	int bar;
1309 
1310 	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1311 
1312 	if (efx->membase) {
1313 		iounmap(efx->membase);
1314 		efx->membase = NULL;
1315 	}
1316 
1317 	if (efx->membase_phys) {
1318 		bar = efx->type->mem_bar;
1319 		pci_release_region(efx->pci_dev, bar);
1320 		efx->membase_phys = 0;
1321 	}
1322 
1323 	/* Don't disable bus-mastering if VFs are assigned */
1324 	if (!pci_vfs_assigned(efx->pci_dev))
1325 		pci_disable_device(efx->pci_dev);
1326 }
1327 
1328 void efx_set_default_rx_indir_table(struct efx_nic *efx)
1329 {
1330 	size_t i;
1331 
1332 	for (i = 0; i < ARRAY_SIZE(efx->rx_indir_table); i++)
1333 		efx->rx_indir_table[i] =
1334 			ethtool_rxfh_indir_default(i, efx->rss_spread);
1335 }
1336 
1337 static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1338 {
1339 	cpumask_var_t thread_mask;
1340 	unsigned int count;
1341 	int cpu;
1342 
1343 	if (rss_cpus) {
1344 		count = rss_cpus;
1345 	} else {
1346 		if (unlikely(!zalloc_cpumask_var(&thread_mask, GFP_KERNEL))) {
1347 			netif_warn(efx, probe, efx->net_dev,
1348 				   "RSS disabled due to allocation failure\n");
1349 			return 1;
1350 		}
1351 
1352 		count = 0;
1353 		for_each_online_cpu(cpu) {
1354 			if (!cpumask_test_cpu(cpu, thread_mask)) {
1355 				++count;
1356 				cpumask_or(thread_mask, thread_mask,
1357 					   topology_sibling_cpumask(cpu));
1358 			}
1359 		}
1360 
1361 		free_cpumask_var(thread_mask);
1362 	}
1363 
1364 	/* If RSS is requested for the PF *and* VFs then we can't write RSS
1365 	 * table entries that are inaccessible to VFs
1366 	 */
1367 #ifdef CONFIG_SFC_SRIOV
1368 	if (efx->type->sriov_wanted) {
1369 		if (efx->type->sriov_wanted(efx) && efx_vf_size(efx) > 1 &&
1370 		    count > efx_vf_size(efx)) {
1371 			netif_warn(efx, probe, efx->net_dev,
1372 				   "Reducing number of RSS channels from %u to %u for "
1373 				   "VF support. Increase vf-msix-limit to use more "
1374 				   "channels on the PF.\n",
1375 				   count, efx_vf_size(efx));
1376 			count = efx_vf_size(efx);
1377 		}
1378 	}
1379 #endif
1380 
1381 	return count;
1382 }
1383 
1384 /* Probe the number and type of interrupts we are able to obtain, and
1385  * the resulting numbers of channels and RX queues.
1386  */
1387 static int efx_probe_interrupts(struct efx_nic *efx)
1388 {
1389 	unsigned int extra_channels = 0;
1390 	unsigned int i, j;
1391 	int rc;
1392 
1393 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
1394 		if (efx->extra_channel_type[i])
1395 			++extra_channels;
1396 
1397 	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1398 		struct msix_entry xentries[EFX_MAX_CHANNELS];
1399 		unsigned int n_channels;
1400 
1401 		n_channels = efx_wanted_parallelism(efx);
1402 		if (efx_separate_tx_channels)
1403 			n_channels *= 2;
1404 		n_channels += extra_channels;
1405 		n_channels = min(n_channels, efx->max_channels);
1406 
1407 		for (i = 0; i < n_channels; i++)
1408 			xentries[i].entry = i;
1409 		rc = pci_enable_msix_range(efx->pci_dev,
1410 					   xentries, 1, n_channels);
1411 		if (rc < 0) {
1412 			/* Fall back to single channel MSI */
1413 			efx->interrupt_mode = EFX_INT_MODE_MSI;
1414 			netif_err(efx, drv, efx->net_dev,
1415 				  "could not enable MSI-X\n");
1416 		} else if (rc < n_channels) {
1417 			netif_err(efx, drv, efx->net_dev,
1418 				  "WARNING: Insufficient MSI-X vectors"
1419 				  " available (%d < %u).\n", rc, n_channels);
1420 			netif_err(efx, drv, efx->net_dev,
1421 				  "WARNING: Performance may be reduced.\n");
1422 			n_channels = rc;
1423 		}
1424 
1425 		if (rc > 0) {
1426 			efx->n_channels = n_channels;
1427 			if (n_channels > extra_channels)
1428 				n_channels -= extra_channels;
1429 			if (efx_separate_tx_channels) {
1430 				efx->n_tx_channels = min(max(n_channels / 2,
1431 							     1U),
1432 							 efx->max_tx_channels);
1433 				efx->n_rx_channels = max(n_channels -
1434 							 efx->n_tx_channels,
1435 							 1U);
1436 			} else {
1437 				efx->n_tx_channels = min(n_channels,
1438 							 efx->max_tx_channels);
1439 				efx->n_rx_channels = n_channels;
1440 			}
1441 			for (i = 0; i < efx->n_channels; i++)
1442 				efx_get_channel(efx, i)->irq =
1443 					xentries[i].vector;
1444 		}
1445 	}
1446 
1447 	/* Try single interrupt MSI */
1448 	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1449 		efx->n_channels = 1;
1450 		efx->n_rx_channels = 1;
1451 		efx->n_tx_channels = 1;
1452 		rc = pci_enable_msi(efx->pci_dev);
1453 		if (rc == 0) {
1454 			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1455 		} else {
1456 			netif_err(efx, drv, efx->net_dev,
1457 				  "could not enable MSI\n");
1458 			efx->interrupt_mode = EFX_INT_MODE_LEGACY;
1459 		}
1460 	}
1461 
1462 	/* Assume legacy interrupts */
1463 	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1464 		efx->n_channels = 1 + (efx_separate_tx_channels ? 1 : 0);
1465 		efx->n_rx_channels = 1;
1466 		efx->n_tx_channels = 1;
1467 		efx->legacy_irq = efx->pci_dev->irq;
1468 	}
1469 
1470 	/* Assign extra channels if possible */
1471 	j = efx->n_channels;
1472 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++) {
1473 		if (!efx->extra_channel_type[i])
1474 			continue;
1475 		if (efx->interrupt_mode != EFX_INT_MODE_MSIX ||
1476 		    efx->n_channels <= extra_channels) {
1477 			efx->extra_channel_type[i]->handle_no_channel(efx);
1478 		} else {
1479 			--j;
1480 			efx_get_channel(efx, j)->type =
1481 				efx->extra_channel_type[i];
1482 		}
1483 	}
1484 
1485 	/* RSS might be usable on VFs even if it is disabled on the PF */
1486 #ifdef CONFIG_SFC_SRIOV
1487 	if (efx->type->sriov_wanted) {
1488 		efx->rss_spread = ((efx->n_rx_channels > 1 ||
1489 				    !efx->type->sriov_wanted(efx)) ?
1490 				   efx->n_rx_channels : efx_vf_size(efx));
1491 		return 0;
1492 	}
1493 #endif
1494 	efx->rss_spread = efx->n_rx_channels;
1495 
1496 	return 0;
1497 }
1498 
1499 static int efx_soft_enable_interrupts(struct efx_nic *efx)
1500 {
1501 	struct efx_channel *channel, *end_channel;
1502 	int rc;
1503 
1504 	BUG_ON(efx->state == STATE_DISABLED);
1505 
1506 	efx->irq_soft_enabled = true;
1507 	smp_wmb();
1508 
1509 	efx_for_each_channel(channel, efx) {
1510 		if (!channel->type->keep_eventq) {
1511 			rc = efx_init_eventq(channel);
1512 			if (rc)
1513 				goto fail;
1514 		}
1515 		efx_start_eventq(channel);
1516 	}
1517 
1518 	efx_mcdi_mode_event(efx);
1519 
1520 	return 0;
1521 fail:
1522 	end_channel = channel;
1523 	efx_for_each_channel(channel, efx) {
1524 		if (channel == end_channel)
1525 			break;
1526 		efx_stop_eventq(channel);
1527 		if (!channel->type->keep_eventq)
1528 			efx_fini_eventq(channel);
1529 	}
1530 
1531 	return rc;
1532 }
1533 
1534 static void efx_soft_disable_interrupts(struct efx_nic *efx)
1535 {
1536 	struct efx_channel *channel;
1537 
1538 	if (efx->state == STATE_DISABLED)
1539 		return;
1540 
1541 	efx_mcdi_mode_poll(efx);
1542 
1543 	efx->irq_soft_enabled = false;
1544 	smp_wmb();
1545 
1546 	if (efx->legacy_irq)
1547 		synchronize_irq(efx->legacy_irq);
1548 
1549 	efx_for_each_channel(channel, efx) {
1550 		if (channel->irq)
1551 			synchronize_irq(channel->irq);
1552 
1553 		efx_stop_eventq(channel);
1554 		if (!channel->type->keep_eventq)
1555 			efx_fini_eventq(channel);
1556 	}
1557 
1558 	/* Flush the asynchronous MCDI request queue */
1559 	efx_mcdi_flush_async(efx);
1560 }
1561 
1562 static int efx_enable_interrupts(struct efx_nic *efx)
1563 {
1564 	struct efx_channel *channel, *end_channel;
1565 	int rc;
1566 
1567 	BUG_ON(efx->state == STATE_DISABLED);
1568 
1569 	if (efx->eeh_disabled_legacy_irq) {
1570 		enable_irq(efx->legacy_irq);
1571 		efx->eeh_disabled_legacy_irq = false;
1572 	}
1573 
1574 	efx->type->irq_enable_master(efx);
1575 
1576 	efx_for_each_channel(channel, efx) {
1577 		if (channel->type->keep_eventq) {
1578 			rc = efx_init_eventq(channel);
1579 			if (rc)
1580 				goto fail;
1581 		}
1582 	}
1583 
1584 	rc = efx_soft_enable_interrupts(efx);
1585 	if (rc)
1586 		goto fail;
1587 
1588 	return 0;
1589 
1590 fail:
1591 	end_channel = channel;
1592 	efx_for_each_channel(channel, efx) {
1593 		if (channel == end_channel)
1594 			break;
1595 		if (channel->type->keep_eventq)
1596 			efx_fini_eventq(channel);
1597 	}
1598 
1599 	efx->type->irq_disable_non_ev(efx);
1600 
1601 	return rc;
1602 }
1603 
1604 static void efx_disable_interrupts(struct efx_nic *efx)
1605 {
1606 	struct efx_channel *channel;
1607 
1608 	efx_soft_disable_interrupts(efx);
1609 
1610 	efx_for_each_channel(channel, efx) {
1611 		if (channel->type->keep_eventq)
1612 			efx_fini_eventq(channel);
1613 	}
1614 
1615 	efx->type->irq_disable_non_ev(efx);
1616 }
1617 
1618 static void efx_remove_interrupts(struct efx_nic *efx)
1619 {
1620 	struct efx_channel *channel;
1621 
1622 	/* Remove MSI/MSI-X interrupts */
1623 	efx_for_each_channel(channel, efx)
1624 		channel->irq = 0;
1625 	pci_disable_msi(efx->pci_dev);
1626 	pci_disable_msix(efx->pci_dev);
1627 
1628 	/* Remove legacy interrupt */
1629 	efx->legacy_irq = 0;
1630 }
1631 
1632 static void efx_set_channels(struct efx_nic *efx)
1633 {
1634 	struct efx_channel *channel;
1635 	struct efx_tx_queue *tx_queue;
1636 
1637 	efx->tx_channel_offset =
1638 		efx_separate_tx_channels ?
1639 		efx->n_channels - efx->n_tx_channels : 0;
1640 
1641 	/* We need to mark which channels really have RX and TX
1642 	 * queues, and adjust the TX queue numbers if we have separate
1643 	 * RX-only and TX-only channels.
1644 	 */
1645 	efx_for_each_channel(channel, efx) {
1646 		if (channel->channel < efx->n_rx_channels)
1647 			channel->rx_queue.core_index = channel->channel;
1648 		else
1649 			channel->rx_queue.core_index = -1;
1650 
1651 		efx_for_each_channel_tx_queue(tx_queue, channel)
1652 			tx_queue->queue -= (efx->tx_channel_offset *
1653 					    EFX_TXQ_TYPES);
1654 	}
1655 }
1656 
1657 static int efx_probe_nic(struct efx_nic *efx)
1658 {
1659 	int rc;
1660 
1661 	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1662 
1663 	/* Carry out hardware-type specific initialisation */
1664 	rc = efx->type->probe(efx);
1665 	if (rc)
1666 		return rc;
1667 
1668 	do {
1669 		if (!efx->max_channels || !efx->max_tx_channels) {
1670 			netif_err(efx, drv, efx->net_dev,
1671 				  "Insufficient resources to allocate"
1672 				  " any channels\n");
1673 			rc = -ENOSPC;
1674 			goto fail1;
1675 		}
1676 
1677 		/* Determine the number of channels and queues by trying
1678 		 * to hook in MSI-X interrupts.
1679 		 */
1680 		rc = efx_probe_interrupts(efx);
1681 		if (rc)
1682 			goto fail1;
1683 
1684 		efx_set_channels(efx);
1685 
1686 		/* dimension_resources can fail with EAGAIN */
1687 		rc = efx->type->dimension_resources(efx);
1688 		if (rc != 0 && rc != -EAGAIN)
1689 			goto fail2;
1690 
1691 		if (rc == -EAGAIN)
1692 			/* try again with new max_channels */
1693 			efx_remove_interrupts(efx);
1694 
1695 	} while (rc == -EAGAIN);
1696 
1697 	if (efx->n_channels > 1)
1698 		netdev_rss_key_fill(&efx->rx_hash_key,
1699 				    sizeof(efx->rx_hash_key));
1700 	efx_set_default_rx_indir_table(efx);
1701 
1702 	netif_set_real_num_tx_queues(efx->net_dev, efx->n_tx_channels);
1703 	netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels);
1704 
1705 	/* Initialise the interrupt moderation settings */
1706 	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
1707 				true);
1708 
1709 	return 0;
1710 
1711 fail2:
1712 	efx_remove_interrupts(efx);
1713 fail1:
1714 	efx->type->remove(efx);
1715 	return rc;
1716 }
1717 
1718 static void efx_remove_nic(struct efx_nic *efx)
1719 {
1720 	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1721 
1722 	efx_remove_interrupts(efx);
1723 	efx->type->remove(efx);
1724 }
1725 
1726 static int efx_probe_filters(struct efx_nic *efx)
1727 {
1728 	int rc;
1729 
1730 	spin_lock_init(&efx->filter_lock);
1731 	init_rwsem(&efx->filter_sem);
1732 	mutex_lock(&efx->mac_lock);
1733 	down_write(&efx->filter_sem);
1734 	rc = efx->type->filter_table_probe(efx);
1735 	if (rc)
1736 		goto out_unlock;
1737 
1738 #ifdef CONFIG_RFS_ACCEL
1739 	if (efx->type->offload_features & NETIF_F_NTUPLE) {
1740 		struct efx_channel *channel;
1741 		int i, success = 1;
1742 
1743 		efx_for_each_channel(channel, efx) {
1744 			channel->rps_flow_id =
1745 				kcalloc(efx->type->max_rx_ip_filters,
1746 					sizeof(*channel->rps_flow_id),
1747 					GFP_KERNEL);
1748 			if (!channel->rps_flow_id)
1749 				success = 0;
1750 			else
1751 				for (i = 0;
1752 				     i < efx->type->max_rx_ip_filters;
1753 				     ++i)
1754 					channel->rps_flow_id[i] =
1755 						RPS_FLOW_ID_INVALID;
1756 		}
1757 
1758 		if (!success) {
1759 			efx_for_each_channel(channel, efx)
1760 				kfree(channel->rps_flow_id);
1761 			efx->type->filter_table_remove(efx);
1762 			rc = -ENOMEM;
1763 			goto out_unlock;
1764 		}
1765 
1766 		efx->rps_expire_index = efx->rps_expire_channel = 0;
1767 	}
1768 #endif
1769 out_unlock:
1770 	up_write(&efx->filter_sem);
1771 	mutex_unlock(&efx->mac_lock);
1772 	return rc;
1773 }
1774 
1775 static void efx_remove_filters(struct efx_nic *efx)
1776 {
1777 #ifdef CONFIG_RFS_ACCEL
1778 	struct efx_channel *channel;
1779 
1780 	efx_for_each_channel(channel, efx)
1781 		kfree(channel->rps_flow_id);
1782 #endif
1783 	down_write(&efx->filter_sem);
1784 	efx->type->filter_table_remove(efx);
1785 	up_write(&efx->filter_sem);
1786 }
1787 
1788 static void efx_restore_filters(struct efx_nic *efx)
1789 {
1790 	down_read(&efx->filter_sem);
1791 	efx->type->filter_table_restore(efx);
1792 	up_read(&efx->filter_sem);
1793 }
1794 
1795 /**************************************************************************
1796  *
1797  * NIC startup/shutdown
1798  *
1799  *************************************************************************/
1800 
1801 static int efx_probe_all(struct efx_nic *efx)
1802 {
1803 	int rc;
1804 
1805 	rc = efx_probe_nic(efx);
1806 	if (rc) {
1807 		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
1808 		goto fail1;
1809 	}
1810 
1811 	rc = efx_probe_port(efx);
1812 	if (rc) {
1813 		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
1814 		goto fail2;
1815 	}
1816 
1817 	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
1818 	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
1819 		rc = -EINVAL;
1820 		goto fail3;
1821 	}
1822 	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1823 
1824 #ifdef CONFIG_SFC_SRIOV
1825 	rc = efx->type->vswitching_probe(efx);
1826 	if (rc) /* not fatal; the PF will still work fine */
1827 		netif_warn(efx, probe, efx->net_dev,
1828 			   "failed to setup vswitching rc=%d;"
1829 			   " VFs may not function\n", rc);
1830 #endif
1831 
1832 	rc = efx_probe_filters(efx);
1833 	if (rc) {
1834 		netif_err(efx, probe, efx->net_dev,
1835 			  "failed to create filter tables\n");
1836 		goto fail4;
1837 	}
1838 
1839 	rc = efx_probe_channels(efx);
1840 	if (rc)
1841 		goto fail5;
1842 
1843 	return 0;
1844 
1845  fail5:
1846 	efx_remove_filters(efx);
1847  fail4:
1848 #ifdef CONFIG_SFC_SRIOV
1849 	efx->type->vswitching_remove(efx);
1850 #endif
1851  fail3:
1852 	efx_remove_port(efx);
1853  fail2:
1854 	efx_remove_nic(efx);
1855  fail1:
1856 	return rc;
1857 }
1858 
1859 /* If the interface is supposed to be running but is not, start
1860  * the hardware and software data path, regular activity for the port
1861  * (MAC statistics, link polling, etc.) and schedule the port to be
1862  * reconfigured.  Interrupts must already be enabled.  This function
1863  * is safe to call multiple times, so long as the NIC is not disabled.
1864  * Requires the RTNL lock.
1865  */
1866 static void efx_start_all(struct efx_nic *efx)
1867 {
1868 	EFX_ASSERT_RESET_SERIALISED(efx);
1869 	BUG_ON(efx->state == STATE_DISABLED);
1870 
1871 	/* Check that it is appropriate to restart the interface. All
1872 	 * of these flags are safe to read under just the rtnl lock */
1873 	if (efx->port_enabled || !netif_running(efx->net_dev) ||
1874 	    efx->reset_pending)
1875 		return;
1876 
1877 	efx_start_port(efx);
1878 	efx_start_datapath(efx);
1879 
1880 	/* Start the hardware monitor if there is one */
1881 	if (efx->type->monitor != NULL)
1882 		queue_delayed_work(efx->workqueue, &efx->monitor_work,
1883 				   efx_monitor_interval);
1884 
1885 	/* If link state detection is normally event-driven, we have
1886 	 * to poll now because we could have missed a change
1887 	 */
1888 	if (efx_nic_rev(efx) >= EFX_REV_SIENA_A0) {
1889 		mutex_lock(&efx->mac_lock);
1890 		if (efx->phy_op->poll(efx))
1891 			efx_link_status_changed(efx);
1892 		mutex_unlock(&efx->mac_lock);
1893 	}
1894 
1895 	efx->type->start_stats(efx);
1896 	efx->type->pull_stats(efx);
1897 	spin_lock_bh(&efx->stats_lock);
1898 	efx->type->update_stats(efx, NULL, NULL);
1899 	spin_unlock_bh(&efx->stats_lock);
1900 }
1901 
1902 /* Quiesce the hardware and software data path, and regular activity
1903  * for the port without bringing the link down.  Safe to call multiple
1904  * times with the NIC in almost any state, but interrupts should be
1905  * enabled.  Requires the RTNL lock.
1906  */
1907 static void efx_stop_all(struct efx_nic *efx)
1908 {
1909 	EFX_ASSERT_RESET_SERIALISED(efx);
1910 
1911 	/* port_enabled can be read safely under the rtnl lock */
1912 	if (!efx->port_enabled)
1913 		return;
1914 
1915 	/* update stats before we go down so we can accurately count
1916 	 * rx_nodesc_drops
1917 	 */
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 	efx->type->stop_stats(efx);
1923 	efx_stop_port(efx);
1924 
1925 	/* Stop the kernel transmit interface.  This is only valid if
1926 	 * the device is stopped or detached; otherwise the watchdog
1927 	 * may fire immediately.
1928 	 */
1929 	WARN_ON(netif_running(efx->net_dev) &&
1930 		netif_device_present(efx->net_dev));
1931 	netif_tx_disable(efx->net_dev);
1932 
1933 	efx_stop_datapath(efx);
1934 }
1935 
1936 static void efx_remove_all(struct efx_nic *efx)
1937 {
1938 	efx_remove_channels(efx);
1939 	efx_remove_filters(efx);
1940 #ifdef CONFIG_SFC_SRIOV
1941 	efx->type->vswitching_remove(efx);
1942 #endif
1943 	efx_remove_port(efx);
1944 	efx_remove_nic(efx);
1945 }
1946 
1947 /**************************************************************************
1948  *
1949  * Interrupt moderation
1950  *
1951  **************************************************************************/
1952 
1953 static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int quantum_ns)
1954 {
1955 	if (usecs == 0)
1956 		return 0;
1957 	if (usecs * 1000 < quantum_ns)
1958 		return 1; /* never round down to 0 */
1959 	return usecs * 1000 / quantum_ns;
1960 }
1961 
1962 /* Set interrupt moderation parameters */
1963 int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
1964 			    unsigned int rx_usecs, bool rx_adaptive,
1965 			    bool rx_may_override_tx)
1966 {
1967 	struct efx_channel *channel;
1968 	unsigned int irq_mod_max = DIV_ROUND_UP(efx->type->timer_period_max *
1969 						efx->timer_quantum_ns,
1970 						1000);
1971 	unsigned int tx_ticks;
1972 	unsigned int rx_ticks;
1973 
1974 	EFX_ASSERT_RESET_SERIALISED(efx);
1975 
1976 	if (tx_usecs > irq_mod_max || rx_usecs > irq_mod_max)
1977 		return -EINVAL;
1978 
1979 	tx_ticks = irq_mod_ticks(tx_usecs, efx->timer_quantum_ns);
1980 	rx_ticks = irq_mod_ticks(rx_usecs, efx->timer_quantum_ns);
1981 
1982 	if (tx_ticks != rx_ticks && efx->tx_channel_offset == 0 &&
1983 	    !rx_may_override_tx) {
1984 		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
1985 			  "RX and TX IRQ moderation must be equal\n");
1986 		return -EINVAL;
1987 	}
1988 
1989 	efx->irq_rx_adaptive = rx_adaptive;
1990 	efx->irq_rx_moderation = rx_ticks;
1991 	efx_for_each_channel(channel, efx) {
1992 		if (efx_channel_has_rx_queue(channel))
1993 			channel->irq_moderation = rx_ticks;
1994 		else if (efx_channel_has_tx_queues(channel))
1995 			channel->irq_moderation = tx_ticks;
1996 	}
1997 
1998 	return 0;
1999 }
2000 
2001 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
2002 			    unsigned int *rx_usecs, bool *rx_adaptive)
2003 {
2004 	/* We must round up when converting ticks to microseconds
2005 	 * because we round down when converting the other way.
2006 	 */
2007 
2008 	*rx_adaptive = efx->irq_rx_adaptive;
2009 	*rx_usecs = DIV_ROUND_UP(efx->irq_rx_moderation *
2010 				 efx->timer_quantum_ns,
2011 				 1000);
2012 
2013 	/* If channels are shared between RX and TX, so is IRQ
2014 	 * moderation.  Otherwise, IRQ moderation is the same for all
2015 	 * TX channels and is not adaptive.
2016 	 */
2017 	if (efx->tx_channel_offset == 0)
2018 		*tx_usecs = *rx_usecs;
2019 	else
2020 		*tx_usecs = DIV_ROUND_UP(
2021 			efx->channel[efx->tx_channel_offset]->irq_moderation *
2022 			efx->timer_quantum_ns,
2023 			1000);
2024 }
2025 
2026 /**************************************************************************
2027  *
2028  * Hardware monitor
2029  *
2030  **************************************************************************/
2031 
2032 /* Run periodically off the general workqueue */
2033 static void efx_monitor(struct work_struct *data)
2034 {
2035 	struct efx_nic *efx = container_of(data, struct efx_nic,
2036 					   monitor_work.work);
2037 
2038 	netif_vdbg(efx, timer, efx->net_dev,
2039 		   "hardware monitor executing on CPU %d\n",
2040 		   raw_smp_processor_id());
2041 	BUG_ON(efx->type->monitor == NULL);
2042 
2043 	/* If the mac_lock is already held then it is likely a port
2044 	 * reconfiguration is already in place, which will likely do
2045 	 * most of the work of monitor() anyway. */
2046 	if (mutex_trylock(&efx->mac_lock)) {
2047 		if (efx->port_enabled)
2048 			efx->type->monitor(efx);
2049 		mutex_unlock(&efx->mac_lock);
2050 	}
2051 
2052 	queue_delayed_work(efx->workqueue, &efx->monitor_work,
2053 			   efx_monitor_interval);
2054 }
2055 
2056 /**************************************************************************
2057  *
2058  * ioctls
2059  *
2060  *************************************************************************/
2061 
2062 /* Net device ioctl
2063  * Context: process, rtnl_lock() held.
2064  */
2065 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
2066 {
2067 	struct efx_nic *efx = netdev_priv(net_dev);
2068 	struct mii_ioctl_data *data = if_mii(ifr);
2069 
2070 	if (cmd == SIOCSHWTSTAMP)
2071 		return efx_ptp_set_ts_config(efx, ifr);
2072 	if (cmd == SIOCGHWTSTAMP)
2073 		return efx_ptp_get_ts_config(efx, ifr);
2074 
2075 	/* Convert phy_id from older PRTAD/DEVAD format */
2076 	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
2077 	    (data->phy_id & 0xfc00) == 0x0400)
2078 		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;
2079 
2080 	return mdio_mii_ioctl(&efx->mdio, data, cmd);
2081 }
2082 
2083 /**************************************************************************
2084  *
2085  * NAPI interface
2086  *
2087  **************************************************************************/
2088 
2089 static void efx_init_napi_channel(struct efx_channel *channel)
2090 {
2091 	struct efx_nic *efx = channel->efx;
2092 
2093 	channel->napi_dev = efx->net_dev;
2094 	netif_napi_add(channel->napi_dev, &channel->napi_str,
2095 		       efx_poll, napi_weight);
2096 	efx_channel_busy_poll_init(channel);
2097 }
2098 
2099 static void efx_init_napi(struct efx_nic *efx)
2100 {
2101 	struct efx_channel *channel;
2102 
2103 	efx_for_each_channel(channel, efx)
2104 		efx_init_napi_channel(channel);
2105 }
2106 
2107 static void efx_fini_napi_channel(struct efx_channel *channel)
2108 {
2109 	if (channel->napi_dev) {
2110 		netif_napi_del(&channel->napi_str);
2111 		napi_hash_del(&channel->napi_str);
2112 	}
2113 	channel->napi_dev = NULL;
2114 }
2115 
2116 static void efx_fini_napi(struct efx_nic *efx)
2117 {
2118 	struct efx_channel *channel;
2119 
2120 	efx_for_each_channel(channel, efx)
2121 		efx_fini_napi_channel(channel);
2122 }
2123 
2124 /**************************************************************************
2125  *
2126  * Kernel netpoll interface
2127  *
2128  *************************************************************************/
2129 
2130 #ifdef CONFIG_NET_POLL_CONTROLLER
2131 
2132 /* Although in the common case interrupts will be disabled, this is not
2133  * guaranteed. However, all our work happens inside the NAPI callback,
2134  * so no locking is required.
2135  */
2136 static void efx_netpoll(struct net_device *net_dev)
2137 {
2138 	struct efx_nic *efx = netdev_priv(net_dev);
2139 	struct efx_channel *channel;
2140 
2141 	efx_for_each_channel(channel, efx)
2142 		efx_schedule_channel(channel);
2143 }
2144 
2145 #endif
2146 
2147 #ifdef CONFIG_NET_RX_BUSY_POLL
2148 static int efx_busy_poll(struct napi_struct *napi)
2149 {
2150 	struct efx_channel *channel =
2151 		container_of(napi, struct efx_channel, napi_str);
2152 	struct efx_nic *efx = channel->efx;
2153 	int budget = 4;
2154 	int old_rx_packets, rx_packets;
2155 
2156 	if (!netif_running(efx->net_dev))
2157 		return LL_FLUSH_FAILED;
2158 
2159 	if (!efx_channel_try_lock_poll(channel))
2160 		return LL_FLUSH_BUSY;
2161 
2162 	old_rx_packets = channel->rx_queue.rx_packets;
2163 	efx_process_channel(channel, budget);
2164 
2165 	rx_packets = channel->rx_queue.rx_packets - old_rx_packets;
2166 
2167 	/* There is no race condition with NAPI here.
2168 	 * NAPI will automatically be rescheduled if it yielded during busy
2169 	 * polling, because it was not able to take the lock and thus returned
2170 	 * the full budget.
2171 	 */
2172 	efx_channel_unlock_poll(channel);
2173 
2174 	return rx_packets;
2175 }
2176 #endif
2177 
2178 /**************************************************************************
2179  *
2180  * Kernel net device interface
2181  *
2182  *************************************************************************/
2183 
2184 /* Context: process, rtnl_lock() held. */
2185 int efx_net_open(struct net_device *net_dev)
2186 {
2187 	struct efx_nic *efx = netdev_priv(net_dev);
2188 	int rc;
2189 
2190 	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
2191 		  raw_smp_processor_id());
2192 
2193 	rc = efx_check_disabled(efx);
2194 	if (rc)
2195 		return rc;
2196 	if (efx->phy_mode & PHY_MODE_SPECIAL)
2197 		return -EBUSY;
2198 	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
2199 		return -EIO;
2200 
2201 	/* Notify the kernel of the link state polled during driver load,
2202 	 * before the monitor starts running */
2203 	efx_link_status_changed(efx);
2204 
2205 	efx_start_all(efx);
2206 	efx_selftest_async_start(efx);
2207 	return 0;
2208 }
2209 
2210 /* Context: process, rtnl_lock() held.
2211  * Note that the kernel will ignore our return code; this method
2212  * should really be a void.
2213  */
2214 int efx_net_stop(struct net_device *net_dev)
2215 {
2216 	struct efx_nic *efx = netdev_priv(net_dev);
2217 
2218 	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
2219 		  raw_smp_processor_id());
2220 
2221 	/* Stop the device and flush all the channels */
2222 	efx_stop_all(efx);
2223 
2224 	return 0;
2225 }
2226 
2227 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
2228 static struct rtnl_link_stats64 *efx_net_stats(struct net_device *net_dev,
2229 					       struct rtnl_link_stats64 *stats)
2230 {
2231 	struct efx_nic *efx = netdev_priv(net_dev);
2232 
2233 	spin_lock_bh(&efx->stats_lock);
2234 	efx->type->update_stats(efx, NULL, stats);
2235 	spin_unlock_bh(&efx->stats_lock);
2236 
2237 	return stats;
2238 }
2239 
2240 /* Context: netif_tx_lock held, BHs disabled. */
2241 static void efx_watchdog(struct net_device *net_dev)
2242 {
2243 	struct efx_nic *efx = netdev_priv(net_dev);
2244 
2245 	netif_err(efx, tx_err, efx->net_dev,
2246 		  "TX stuck with port_enabled=%d: resetting channels\n",
2247 		  efx->port_enabled);
2248 
2249 	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2250 }
2251 
2252 
2253 /* Context: process, rtnl_lock() held. */
2254 static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
2255 {
2256 	struct efx_nic *efx = netdev_priv(net_dev);
2257 	int rc;
2258 
2259 	rc = efx_check_disabled(efx);
2260 	if (rc)
2261 		return rc;
2262 	if (new_mtu > EFX_MAX_MTU)
2263 		return -EINVAL;
2264 
2265 	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2266 
2267 	efx_device_detach_sync(efx);
2268 	efx_stop_all(efx);
2269 
2270 	mutex_lock(&efx->mac_lock);
2271 	net_dev->mtu = new_mtu;
2272 	efx_mac_reconfigure(efx);
2273 	mutex_unlock(&efx->mac_lock);
2274 
2275 	efx_start_all(efx);
2276 	netif_device_attach(efx->net_dev);
2277 	return 0;
2278 }
2279 
2280 static int efx_set_mac_address(struct net_device *net_dev, void *data)
2281 {
2282 	struct efx_nic *efx = netdev_priv(net_dev);
2283 	struct sockaddr *addr = data;
2284 	u8 *new_addr = addr->sa_data;
2285 	u8 old_addr[6];
2286 	int rc;
2287 
2288 	if (!is_valid_ether_addr(new_addr)) {
2289 		netif_err(efx, drv, efx->net_dev,
2290 			  "invalid ethernet MAC address requested: %pM\n",
2291 			  new_addr);
2292 		return -EADDRNOTAVAIL;
2293 	}
2294 
2295 	/* save old address */
2296 	ether_addr_copy(old_addr, net_dev->dev_addr);
2297 	ether_addr_copy(net_dev->dev_addr, new_addr);
2298 	if (efx->type->set_mac_address) {
2299 		rc = efx->type->set_mac_address(efx);
2300 		if (rc) {
2301 			ether_addr_copy(net_dev->dev_addr, old_addr);
2302 			return rc;
2303 		}
2304 	}
2305 
2306 	/* Reconfigure the MAC */
2307 	mutex_lock(&efx->mac_lock);
2308 	efx_mac_reconfigure(efx);
2309 	mutex_unlock(&efx->mac_lock);
2310 
2311 	return 0;
2312 }
2313 
2314 /* Context: netif_addr_lock held, BHs disabled. */
2315 static void efx_set_rx_mode(struct net_device *net_dev)
2316 {
2317 	struct efx_nic *efx = netdev_priv(net_dev);
2318 
2319 	if (efx->port_enabled)
2320 		queue_work(efx->workqueue, &efx->mac_work);
2321 	/* Otherwise efx_start_port() will do this */
2322 }
2323 
2324 static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2325 {
2326 	struct efx_nic *efx = netdev_priv(net_dev);
2327 	int rc;
2328 
2329 	/* If disabling RX n-tuple filtering, clear existing filters */
2330 	if (net_dev->features & ~data & NETIF_F_NTUPLE) {
2331 		rc = efx->type->filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);
2332 		if (rc)
2333 			return rc;
2334 	}
2335 
2336 	/* If Rx VLAN filter is changed, update filters via mac_reconfigure */
2337 	if ((net_dev->features ^ data) & NETIF_F_HW_VLAN_CTAG_FILTER) {
2338 		/* efx_set_rx_mode() will schedule MAC work to update filters
2339 		 * when a new features are finally set in net_dev.
2340 		 */
2341 		efx_set_rx_mode(net_dev);
2342 	}
2343 
2344 	return 0;
2345 }
2346 
2347 static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid)
2348 {
2349 	struct efx_nic *efx = netdev_priv(net_dev);
2350 
2351 	if (efx->type->vlan_rx_add_vid)
2352 		return efx->type->vlan_rx_add_vid(efx, proto, vid);
2353 	else
2354 		return -EOPNOTSUPP;
2355 }
2356 
2357 static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid)
2358 {
2359 	struct efx_nic *efx = netdev_priv(net_dev);
2360 
2361 	if (efx->type->vlan_rx_kill_vid)
2362 		return efx->type->vlan_rx_kill_vid(efx, proto, vid);
2363 	else
2364 		return -EOPNOTSUPP;
2365 }
2366 
2367 static const struct net_device_ops efx_netdev_ops = {
2368 	.ndo_open		= efx_net_open,
2369 	.ndo_stop		= efx_net_stop,
2370 	.ndo_get_stats64	= efx_net_stats,
2371 	.ndo_tx_timeout		= efx_watchdog,
2372 	.ndo_start_xmit		= efx_hard_start_xmit,
2373 	.ndo_validate_addr	= eth_validate_addr,
2374 	.ndo_do_ioctl		= efx_ioctl,
2375 	.ndo_change_mtu		= efx_change_mtu,
2376 	.ndo_set_mac_address	= efx_set_mac_address,
2377 	.ndo_set_rx_mode	= efx_set_rx_mode,
2378 	.ndo_set_features	= efx_set_features,
2379 	.ndo_vlan_rx_add_vid	= efx_vlan_rx_add_vid,
2380 	.ndo_vlan_rx_kill_vid	= efx_vlan_rx_kill_vid,
2381 #ifdef CONFIG_SFC_SRIOV
2382 	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
2383 	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
2384 	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
2385 	.ndo_get_vf_config	= efx_sriov_get_vf_config,
2386 	.ndo_set_vf_link_state  = efx_sriov_set_vf_link_state,
2387 	.ndo_get_phys_port_id   = efx_sriov_get_phys_port_id,
2388 #endif
2389 #ifdef CONFIG_NET_POLL_CONTROLLER
2390 	.ndo_poll_controller = efx_netpoll,
2391 #endif
2392 	.ndo_setup_tc		= efx_setup_tc,
2393 #ifdef CONFIG_NET_RX_BUSY_POLL
2394 	.ndo_busy_poll		= efx_busy_poll,
2395 #endif
2396 #ifdef CONFIG_RFS_ACCEL
2397 	.ndo_rx_flow_steer	= efx_filter_rfs,
2398 #endif
2399 };
2400 
2401 static void efx_update_name(struct efx_nic *efx)
2402 {
2403 	strcpy(efx->name, efx->net_dev->name);
2404 	efx_mtd_rename(efx);
2405 	efx_set_channel_names(efx);
2406 }
2407 
2408 static int efx_netdev_event(struct notifier_block *this,
2409 			    unsigned long event, void *ptr)
2410 {
2411 	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2412 
2413 	if ((net_dev->netdev_ops == &efx_netdev_ops) &&
2414 	    event == NETDEV_CHANGENAME)
2415 		efx_update_name(netdev_priv(net_dev));
2416 
2417 	return NOTIFY_DONE;
2418 }
2419 
2420 static struct notifier_block efx_netdev_notifier = {
2421 	.notifier_call = efx_netdev_event,
2422 };
2423 
2424 static ssize_t
2425 show_phy_type(struct device *dev, struct device_attribute *attr, char *buf)
2426 {
2427 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2428 	return sprintf(buf, "%d\n", efx->phy_type);
2429 }
2430 static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
2431 
2432 #ifdef CONFIG_SFC_MCDI_LOGGING
2433 static ssize_t show_mcdi_log(struct device *dev, struct device_attribute *attr,
2434 			     char *buf)
2435 {
2436 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2437 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2438 
2439 	return scnprintf(buf, PAGE_SIZE, "%d\n", mcdi->logging_enabled);
2440 }
2441 static ssize_t set_mcdi_log(struct device *dev, struct device_attribute *attr,
2442 			    const char *buf, size_t count)
2443 {
2444 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
2445 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2446 	bool enable = count > 0 && *buf != '0';
2447 
2448 	mcdi->logging_enabled = enable;
2449 	return count;
2450 }
2451 static DEVICE_ATTR(mcdi_logging, 0644, show_mcdi_log, set_mcdi_log);
2452 #endif
2453 
2454 static int efx_register_netdev(struct efx_nic *efx)
2455 {
2456 	struct net_device *net_dev = efx->net_dev;
2457 	struct efx_channel *channel;
2458 	int rc;
2459 
2460 	net_dev->watchdog_timeo = 5 * HZ;
2461 	net_dev->irq = efx->pci_dev->irq;
2462 	net_dev->netdev_ops = &efx_netdev_ops;
2463 	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0)
2464 		net_dev->priv_flags |= IFF_UNICAST_FLT;
2465 	net_dev->ethtool_ops = &efx_ethtool_ops;
2466 	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2467 
2468 	rtnl_lock();
2469 
2470 	/* Enable resets to be scheduled and check whether any were
2471 	 * already requested.  If so, the NIC is probably hosed so we
2472 	 * abort.
2473 	 */
2474 	efx->state = STATE_READY;
2475 	smp_mb(); /* ensure we change state before checking reset_pending */
2476 	if (efx->reset_pending) {
2477 		netif_err(efx, probe, efx->net_dev,
2478 			  "aborting probe due to scheduled reset\n");
2479 		rc = -EIO;
2480 		goto fail_locked;
2481 	}
2482 
2483 	rc = dev_alloc_name(net_dev, net_dev->name);
2484 	if (rc < 0)
2485 		goto fail_locked;
2486 	efx_update_name(efx);
2487 
2488 	/* Always start with carrier off; PHY events will detect the link */
2489 	netif_carrier_off(net_dev);
2490 
2491 	rc = register_netdevice(net_dev);
2492 	if (rc)
2493 		goto fail_locked;
2494 
2495 	efx_for_each_channel(channel, efx) {
2496 		struct efx_tx_queue *tx_queue;
2497 		efx_for_each_channel_tx_queue(tx_queue, channel)
2498 			efx_init_tx_queue_core_txq(tx_queue);
2499 	}
2500 
2501 	efx_associate(efx);
2502 
2503 	rtnl_unlock();
2504 
2505 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2506 	if (rc) {
2507 		netif_err(efx, drv, efx->net_dev,
2508 			  "failed to init net dev attributes\n");
2509 		goto fail_registered;
2510 	}
2511 #ifdef CONFIG_SFC_MCDI_LOGGING
2512 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2513 	if (rc) {
2514 		netif_err(efx, drv, efx->net_dev,
2515 			  "failed to init net dev attributes\n");
2516 		goto fail_attr_mcdi_logging;
2517 	}
2518 #endif
2519 
2520 	return 0;
2521 
2522 #ifdef CONFIG_SFC_MCDI_LOGGING
2523 fail_attr_mcdi_logging:
2524 	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2525 #endif
2526 fail_registered:
2527 	rtnl_lock();
2528 	efx_dissociate(efx);
2529 	unregister_netdevice(net_dev);
2530 fail_locked:
2531 	efx->state = STATE_UNINIT;
2532 	rtnl_unlock();
2533 	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2534 	return rc;
2535 }
2536 
2537 static void efx_unregister_netdev(struct efx_nic *efx)
2538 {
2539 	if (!efx->net_dev)
2540 		return;
2541 
2542 	BUG_ON(netdev_priv(efx->net_dev) != efx);
2543 
2544 	if (efx_dev_registered(efx)) {
2545 		strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
2546 #ifdef CONFIG_SFC_MCDI_LOGGING
2547 		device_remove_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2548 #endif
2549 		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2550 		unregister_netdev(efx->net_dev);
2551 	}
2552 }
2553 
2554 /**************************************************************************
2555  *
2556  * Device reset and suspend
2557  *
2558  **************************************************************************/
2559 
2560 /* Tears down the entire software state and most of the hardware state
2561  * before reset.  */
2562 void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2563 {
2564 	EFX_ASSERT_RESET_SERIALISED(efx);
2565 
2566 	if (method == RESET_TYPE_MCDI_TIMEOUT)
2567 		efx->type->prepare_flr(efx);
2568 
2569 	efx_stop_all(efx);
2570 	efx_disable_interrupts(efx);
2571 
2572 	mutex_lock(&efx->mac_lock);
2573 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2574 	    method != RESET_TYPE_DATAPATH)
2575 		efx->phy_op->fini(efx);
2576 	efx->type->fini(efx);
2577 }
2578 
2579 /* This function will always ensure that the locks acquired in
2580  * efx_reset_down() are released. A failure return code indicates
2581  * that we were unable to reinitialise the hardware, and the
2582  * driver should be disabled. If ok is false, then the rx and tx
2583  * engines are not restarted, pending a RESET_DISABLE. */
2584 int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2585 {
2586 	int rc;
2587 
2588 	EFX_ASSERT_RESET_SERIALISED(efx);
2589 
2590 	if (method == RESET_TYPE_MCDI_TIMEOUT)
2591 		efx->type->finish_flr(efx);
2592 
2593 	/* Ensure that SRAM is initialised even if we're disabling the device */
2594 	rc = efx->type->init(efx);
2595 	if (rc) {
2596 		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2597 		goto fail;
2598 	}
2599 
2600 	if (!ok)
2601 		goto fail;
2602 
2603 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2604 	    method != RESET_TYPE_DATAPATH) {
2605 		rc = efx->phy_op->init(efx);
2606 		if (rc)
2607 			goto fail;
2608 		rc = efx->phy_op->reconfigure(efx);
2609 		if (rc && rc != -EPERM)
2610 			netif_err(efx, drv, efx->net_dev,
2611 				  "could not restore PHY settings\n");
2612 	}
2613 
2614 	rc = efx_enable_interrupts(efx);
2615 	if (rc)
2616 		goto fail;
2617 
2618 #ifdef CONFIG_SFC_SRIOV
2619 	rc = efx->type->vswitching_restore(efx);
2620 	if (rc) /* not fatal; the PF will still work fine */
2621 		netif_warn(efx, probe, efx->net_dev,
2622 			   "failed to restore vswitching rc=%d;"
2623 			   " VFs may not function\n", rc);
2624 #endif
2625 
2626 	down_read(&efx->filter_sem);
2627 	efx_restore_filters(efx);
2628 	up_read(&efx->filter_sem);
2629 	if (efx->type->sriov_reset)
2630 		efx->type->sriov_reset(efx);
2631 
2632 	mutex_unlock(&efx->mac_lock);
2633 
2634 	efx_start_all(efx);
2635 
2636 	return 0;
2637 
2638 fail:
2639 	efx->port_initialized = false;
2640 
2641 	mutex_unlock(&efx->mac_lock);
2642 
2643 	return rc;
2644 }
2645 
2646 /* Reset the NIC using the specified method.  Note that the reset may
2647  * fail, in which case the card will be left in an unusable state.
2648  *
2649  * Caller must hold the rtnl_lock.
2650  */
2651 int efx_reset(struct efx_nic *efx, enum reset_type method)
2652 {
2653 	int rc, rc2;
2654 	bool disabled;
2655 
2656 	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
2657 		   RESET_TYPE(method));
2658 
2659 	efx_device_detach_sync(efx);
2660 	efx_reset_down(efx, method);
2661 
2662 	rc = efx->type->reset(efx, method);
2663 	if (rc) {
2664 		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2665 		goto out;
2666 	}
2667 
2668 	/* Clear flags for the scopes we covered.  We assume the NIC and
2669 	 * driver are now quiescent so that there is no race here.
2670 	 */
2671 	if (method < RESET_TYPE_MAX_METHOD)
2672 		efx->reset_pending &= -(1 << (method + 1));
2673 	else /* it doesn't fit into the well-ordered scope hierarchy */
2674 		__clear_bit(method, &efx->reset_pending);
2675 
2676 	/* Reinitialise bus-mastering, which may have been turned off before
2677 	 * the reset was scheduled. This is still appropriate, even in the
2678 	 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
2679 	 * can respond to requests. */
2680 	pci_set_master(efx->pci_dev);
2681 
2682 out:
2683 	/* Leave device stopped if necessary */
2684 	disabled = rc ||
2685 		method == RESET_TYPE_DISABLE ||
2686 		method == RESET_TYPE_RECOVER_OR_DISABLE;
2687 	rc2 = efx_reset_up(efx, method, !disabled);
2688 	if (rc2) {
2689 		disabled = true;
2690 		if (!rc)
2691 			rc = rc2;
2692 	}
2693 
2694 	if (disabled) {
2695 		dev_close(efx->net_dev);
2696 		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
2697 		efx->state = STATE_DISABLED;
2698 	} else {
2699 		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
2700 		netif_device_attach(efx->net_dev);
2701 	}
2702 	return rc;
2703 }
2704 
2705 /* Try recovery mechanisms.
2706  * For now only EEH is supported.
2707  * Returns 0 if the recovery mechanisms are unsuccessful.
2708  * Returns a non-zero value otherwise.
2709  */
2710 int efx_try_recovery(struct efx_nic *efx)
2711 {
2712 #ifdef CONFIG_EEH
2713 	/* A PCI error can occur and not be seen by EEH because nothing
2714 	 * happens on the PCI bus. In this case the driver may fail and
2715 	 * schedule a 'recover or reset', leading to this recovery handler.
2716 	 * Manually call the eeh failure check function.
2717 	 */
2718 	struct eeh_dev *eehdev = pci_dev_to_eeh_dev(efx->pci_dev);
2719 	if (eeh_dev_check_failure(eehdev)) {
2720 		/* The EEH mechanisms will handle the error and reset the
2721 		 * device if necessary.
2722 		 */
2723 		return 1;
2724 	}
2725 #endif
2726 	return 0;
2727 }
2728 
2729 static void efx_wait_for_bist_end(struct efx_nic *efx)
2730 {
2731 	int i;
2732 
2733 	for (i = 0; i < BIST_WAIT_DELAY_COUNT; ++i) {
2734 		if (efx_mcdi_poll_reboot(efx))
2735 			goto out;
2736 		msleep(BIST_WAIT_DELAY_MS);
2737 	}
2738 
2739 	netif_err(efx, drv, efx->net_dev, "Warning: No MC reboot after BIST mode\n");
2740 out:
2741 	/* Either way unset the BIST flag. If we found no reboot we probably
2742 	 * won't recover, but we should try.
2743 	 */
2744 	efx->mc_bist_for_other_fn = false;
2745 }
2746 
2747 /* The worker thread exists so that code that cannot sleep can
2748  * schedule a reset for later.
2749  */
2750 static void efx_reset_work(struct work_struct *data)
2751 {
2752 	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
2753 	unsigned long pending;
2754 	enum reset_type method;
2755 
2756 	pending = ACCESS_ONCE(efx->reset_pending);
2757 	method = fls(pending) - 1;
2758 
2759 	if (method == RESET_TYPE_MC_BIST)
2760 		efx_wait_for_bist_end(efx);
2761 
2762 	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
2763 	     method == RESET_TYPE_RECOVER_OR_ALL) &&
2764 	    efx_try_recovery(efx))
2765 		return;
2766 
2767 	if (!pending)
2768 		return;
2769 
2770 	rtnl_lock();
2771 
2772 	/* We checked the state in efx_schedule_reset() but it may
2773 	 * have changed by now.  Now that we have the RTNL lock,
2774 	 * it cannot change again.
2775 	 */
2776 	if (efx->state == STATE_READY)
2777 		(void)efx_reset(efx, method);
2778 
2779 	rtnl_unlock();
2780 }
2781 
2782 void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
2783 {
2784 	enum reset_type method;
2785 
2786 	if (efx->state == STATE_RECOVERY) {
2787 		netif_dbg(efx, drv, efx->net_dev,
2788 			  "recovering: skip scheduling %s reset\n",
2789 			  RESET_TYPE(type));
2790 		return;
2791 	}
2792 
2793 	switch (type) {
2794 	case RESET_TYPE_INVISIBLE:
2795 	case RESET_TYPE_ALL:
2796 	case RESET_TYPE_RECOVER_OR_ALL:
2797 	case RESET_TYPE_WORLD:
2798 	case RESET_TYPE_DISABLE:
2799 	case RESET_TYPE_RECOVER_OR_DISABLE:
2800 	case RESET_TYPE_DATAPATH:
2801 	case RESET_TYPE_MC_BIST:
2802 	case RESET_TYPE_MCDI_TIMEOUT:
2803 		method = type;
2804 		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
2805 			  RESET_TYPE(method));
2806 		break;
2807 	default:
2808 		method = efx->type->map_reset_reason(type);
2809 		netif_dbg(efx, drv, efx->net_dev,
2810 			  "scheduling %s reset for %s\n",
2811 			  RESET_TYPE(method), RESET_TYPE(type));
2812 		break;
2813 	}
2814 
2815 	set_bit(method, &efx->reset_pending);
2816 	smp_mb(); /* ensure we change reset_pending before checking state */
2817 
2818 	/* If we're not READY then just leave the flags set as the cue
2819 	 * to abort probing or reschedule the reset later.
2820 	 */
2821 	if (ACCESS_ONCE(efx->state) != STATE_READY)
2822 		return;
2823 
2824 	/* efx_process_channel() will no longer read events once a
2825 	 * reset is scheduled. So switch back to poll'd MCDI completions. */
2826 	efx_mcdi_mode_poll(efx);
2827 
2828 	queue_work(reset_workqueue, &efx->reset_work);
2829 }
2830 
2831 /**************************************************************************
2832  *
2833  * List of NICs we support
2834  *
2835  **************************************************************************/
2836 
2837 /* PCI device ID table */
2838 static const struct pci_device_id efx_pci_table[] = {
2839 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
2840 		    PCI_DEVICE_ID_SOLARFLARE_SFC4000A_0),
2841 	 .driver_data = (unsigned long) &falcon_a1_nic_type},
2842 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE,
2843 		    PCI_DEVICE_ID_SOLARFLARE_SFC4000B),
2844 	 .driver_data = (unsigned long) &falcon_b0_nic_type},
2845 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
2846 	 .driver_data = (unsigned long) &siena_a0_nic_type},
2847 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
2848 	 .driver_data = (unsigned long) &siena_a0_nic_type},
2849 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
2850 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2851 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903),  /* SFC9120 VF */
2852 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2853 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923),  /* SFC9140 PF */
2854 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2855 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923),  /* SFC9140 VF */
2856 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2857 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03),  /* SFC9220 PF */
2858 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
2859 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03),  /* SFC9220 VF */
2860 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
2861 	{0}			/* end of list */
2862 };
2863 
2864 /**************************************************************************
2865  *
2866  * Dummy PHY/MAC operations
2867  *
2868  * Can be used for some unimplemented operations
2869  * Needed so all function pointers are valid and do not have to be tested
2870  * before use
2871  *
2872  **************************************************************************/
2873 int efx_port_dummy_op_int(struct efx_nic *efx)
2874 {
2875 	return 0;
2876 }
2877 void efx_port_dummy_op_void(struct efx_nic *efx) {}
2878 
2879 static bool efx_port_dummy_op_poll(struct efx_nic *efx)
2880 {
2881 	return false;
2882 }
2883 
2884 static const struct efx_phy_operations efx_dummy_phy_operations = {
2885 	.init		 = efx_port_dummy_op_int,
2886 	.reconfigure	 = efx_port_dummy_op_int,
2887 	.poll		 = efx_port_dummy_op_poll,
2888 	.fini		 = efx_port_dummy_op_void,
2889 };
2890 
2891 /**************************************************************************
2892  *
2893  * Data housekeeping
2894  *
2895  **************************************************************************/
2896 
2897 /* This zeroes out and then fills in the invariants in a struct
2898  * efx_nic (including all sub-structures).
2899  */
2900 static int efx_init_struct(struct efx_nic *efx,
2901 			   struct pci_dev *pci_dev, struct net_device *net_dev)
2902 {
2903 	int i;
2904 
2905 	/* Initialise common structures */
2906 	INIT_LIST_HEAD(&efx->node);
2907 	INIT_LIST_HEAD(&efx->secondary_list);
2908 	spin_lock_init(&efx->biu_lock);
2909 #ifdef CONFIG_SFC_MTD
2910 	INIT_LIST_HEAD(&efx->mtd_list);
2911 #endif
2912 	INIT_WORK(&efx->reset_work, efx_reset_work);
2913 	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
2914 	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
2915 	efx->pci_dev = pci_dev;
2916 	efx->msg_enable = debug;
2917 	efx->state = STATE_UNINIT;
2918 	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
2919 
2920 	efx->net_dev = net_dev;
2921 	efx->rx_prefix_size = efx->type->rx_prefix_size;
2922 	efx->rx_ip_align =
2923 		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
2924 	efx->rx_packet_hash_offset =
2925 		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
2926 	efx->rx_packet_ts_offset =
2927 		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
2928 	spin_lock_init(&efx->stats_lock);
2929 	mutex_init(&efx->mac_lock);
2930 	efx->phy_op = &efx_dummy_phy_operations;
2931 	efx->mdio.dev = net_dev;
2932 	INIT_WORK(&efx->mac_work, efx_mac_work);
2933 	init_waitqueue_head(&efx->flush_wq);
2934 
2935 	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
2936 		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
2937 		if (!efx->channel[i])
2938 			goto fail;
2939 		efx->msi_context[i].efx = efx;
2940 		efx->msi_context[i].index = i;
2941 	}
2942 
2943 	/* Higher numbered interrupt modes are less capable! */
2944 	efx->interrupt_mode = max(efx->type->max_interrupt_mode,
2945 				  interrupt_mode);
2946 
2947 	/* Would be good to use the net_dev name, but we're too early */
2948 	snprintf(efx->workqueue_name, sizeof(efx->workqueue_name), "sfc%s",
2949 		 pci_name(pci_dev));
2950 	efx->workqueue = create_singlethread_workqueue(efx->workqueue_name);
2951 	if (!efx->workqueue)
2952 		goto fail;
2953 
2954 	return 0;
2955 
2956 fail:
2957 	efx_fini_struct(efx);
2958 	return -ENOMEM;
2959 }
2960 
2961 static void efx_fini_struct(struct efx_nic *efx)
2962 {
2963 	int i;
2964 
2965 	for (i = 0; i < EFX_MAX_CHANNELS; i++)
2966 		kfree(efx->channel[i]);
2967 
2968 	kfree(efx->vpd_sn);
2969 
2970 	if (efx->workqueue) {
2971 		destroy_workqueue(efx->workqueue);
2972 		efx->workqueue = NULL;
2973 	}
2974 }
2975 
2976 void efx_update_sw_stats(struct efx_nic *efx, u64 *stats)
2977 {
2978 	u64 n_rx_nodesc_trunc = 0;
2979 	struct efx_channel *channel;
2980 
2981 	efx_for_each_channel(channel, efx)
2982 		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
2983 	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
2984 	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
2985 }
2986 
2987 /**************************************************************************
2988  *
2989  * PCI interface
2990  *
2991  **************************************************************************/
2992 
2993 /* Main body of final NIC shutdown code
2994  * This is called only at module unload (or hotplug removal).
2995  */
2996 static void efx_pci_remove_main(struct efx_nic *efx)
2997 {
2998 	/* Flush reset_work. It can no longer be scheduled since we
2999 	 * are not READY.
3000 	 */
3001 	BUG_ON(efx->state == STATE_READY);
3002 	cancel_work_sync(&efx->reset_work);
3003 
3004 	efx_disable_interrupts(efx);
3005 	efx_nic_fini_interrupt(efx);
3006 	efx_fini_port(efx);
3007 	efx->type->fini(efx);
3008 	efx_fini_napi(efx);
3009 	efx_remove_all(efx);
3010 }
3011 
3012 /* Final NIC shutdown
3013  * This is called only at module unload (or hotplug removal).  A PF can call
3014  * this on its VFs to ensure they are unbound first.
3015  */
3016 static void efx_pci_remove(struct pci_dev *pci_dev)
3017 {
3018 	struct efx_nic *efx;
3019 
3020 	efx = pci_get_drvdata(pci_dev);
3021 	if (!efx)
3022 		return;
3023 
3024 	/* Mark the NIC as fini, then stop the interface */
3025 	rtnl_lock();
3026 	efx_dissociate(efx);
3027 	dev_close(efx->net_dev);
3028 	efx_disable_interrupts(efx);
3029 	efx->state = STATE_UNINIT;
3030 	rtnl_unlock();
3031 
3032 	if (efx->type->sriov_fini)
3033 		efx->type->sriov_fini(efx);
3034 
3035 	efx_unregister_netdev(efx);
3036 
3037 	efx_mtd_remove(efx);
3038 
3039 	efx_pci_remove_main(efx);
3040 
3041 	efx_fini_io(efx);
3042 	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
3043 
3044 	efx_fini_struct(efx);
3045 	free_netdev(efx->net_dev);
3046 
3047 	pci_disable_pcie_error_reporting(pci_dev);
3048 };
3049 
3050 /* NIC VPD information
3051  * Called during probe to display the part number of the
3052  * installed NIC.  VPD is potentially very large but this should
3053  * always appear within the first 512 bytes.
3054  */
3055 #define SFC_VPD_LEN 512
3056 static void efx_probe_vpd_strings(struct efx_nic *efx)
3057 {
3058 	struct pci_dev *dev = efx->pci_dev;
3059 	char vpd_data[SFC_VPD_LEN];
3060 	ssize_t vpd_size;
3061 	int ro_start, ro_size, i, j;
3062 
3063 	/* Get the vpd data from the device */
3064 	vpd_size = pci_read_vpd(dev, 0, sizeof(vpd_data), vpd_data);
3065 	if (vpd_size <= 0) {
3066 		netif_err(efx, drv, efx->net_dev, "Unable to read VPD\n");
3067 		return;
3068 	}
3069 
3070 	/* Get the Read only section */
3071 	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size, PCI_VPD_LRDT_RO_DATA);
3072 	if (ro_start < 0) {
3073 		netif_err(efx, drv, efx->net_dev, "VPD Read-only not found\n");
3074 		return;
3075 	}
3076 
3077 	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
3078 	j = ro_size;
3079 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3080 	if (i + j > vpd_size)
3081 		j = vpd_size - i;
3082 
3083 	/* Get the Part number */
3084 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "PN");
3085 	if (i < 0) {
3086 		netif_err(efx, drv, efx->net_dev, "Part number not found\n");
3087 		return;
3088 	}
3089 
3090 	j = pci_vpd_info_field_size(&vpd_data[i]);
3091 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3092 	if (i + j > vpd_size) {
3093 		netif_err(efx, drv, efx->net_dev, "Incomplete part number\n");
3094 		return;
3095 	}
3096 
3097 	netif_info(efx, drv, efx->net_dev,
3098 		   "Part Number : %.*s\n", j, &vpd_data[i]);
3099 
3100 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3101 	j = ro_size;
3102 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "SN");
3103 	if (i < 0) {
3104 		netif_err(efx, drv, efx->net_dev, "Serial number not found\n");
3105 		return;
3106 	}
3107 
3108 	j = pci_vpd_info_field_size(&vpd_data[i]);
3109 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3110 	if (i + j > vpd_size) {
3111 		netif_err(efx, drv, efx->net_dev, "Incomplete serial number\n");
3112 		return;
3113 	}
3114 
3115 	efx->vpd_sn = kmalloc(j + 1, GFP_KERNEL);
3116 	if (!efx->vpd_sn)
3117 		return;
3118 
3119 	snprintf(efx->vpd_sn, j + 1, "%s", &vpd_data[i]);
3120 }
3121 
3122 
3123 /* Main body of NIC initialisation
3124  * This is called at module load (or hotplug insertion, theoretically).
3125  */
3126 static int efx_pci_probe_main(struct efx_nic *efx)
3127 {
3128 	int rc;
3129 
3130 	/* Do start-of-day initialisation */
3131 	rc = efx_probe_all(efx);
3132 	if (rc)
3133 		goto fail1;
3134 
3135 	efx_init_napi(efx);
3136 
3137 	rc = efx->type->init(efx);
3138 	if (rc) {
3139 		netif_err(efx, probe, efx->net_dev,
3140 			  "failed to initialise NIC\n");
3141 		goto fail3;
3142 	}
3143 
3144 	rc = efx_init_port(efx);
3145 	if (rc) {
3146 		netif_err(efx, probe, efx->net_dev,
3147 			  "failed to initialise port\n");
3148 		goto fail4;
3149 	}
3150 
3151 	rc = efx_nic_init_interrupt(efx);
3152 	if (rc)
3153 		goto fail5;
3154 	rc = efx_enable_interrupts(efx);
3155 	if (rc)
3156 		goto fail6;
3157 
3158 	return 0;
3159 
3160  fail6:
3161 	efx_nic_fini_interrupt(efx);
3162  fail5:
3163 	efx_fini_port(efx);
3164  fail4:
3165 	efx->type->fini(efx);
3166  fail3:
3167 	efx_fini_napi(efx);
3168 	efx_remove_all(efx);
3169  fail1:
3170 	return rc;
3171 }
3172 
3173 /* NIC initialisation
3174  *
3175  * This is called at module load (or hotplug insertion,
3176  * theoretically).  It sets up PCI mappings, resets the NIC,
3177  * sets up and registers the network devices with the kernel and hooks
3178  * the interrupt service routine.  It does not prepare the device for
3179  * transmission; this is left to the first time one of the network
3180  * interfaces is brought up (i.e. efx_net_open).
3181  */
3182 static int efx_pci_probe(struct pci_dev *pci_dev,
3183 			 const struct pci_device_id *entry)
3184 {
3185 	struct net_device *net_dev;
3186 	struct efx_nic *efx;
3187 	int rc;
3188 
3189 	/* Allocate and initialise a struct net_device and struct efx_nic */
3190 	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
3191 				     EFX_MAX_RX_QUEUES);
3192 	if (!net_dev)
3193 		return -ENOMEM;
3194 	efx = netdev_priv(net_dev);
3195 	efx->type = (const struct efx_nic_type *) entry->driver_data;
3196 	efx->fixed_features |= NETIF_F_HIGHDMA;
3197 	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
3198 			      NETIF_F_TSO | NETIF_F_RXCSUM);
3199 	if (efx->type->offload_features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
3200 		net_dev->features |= NETIF_F_TSO6;
3201 	/* Mask for features that also apply to VLAN devices */
3202 	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
3203 				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
3204 				   NETIF_F_RXCSUM);
3205 
3206 	net_dev->hw_features = net_dev->features & ~efx->fixed_features;
3207 
3208 	/* Disable VLAN filtering by default.  It may be enforced if
3209 	 * the feature is fixed (i.e. VLAN filters are required to
3210 	 * receive VLAN tagged packets due to vPort restrictions).
3211 	 */
3212 	net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3213 	net_dev->features |= efx->fixed_features;
3214 
3215 	pci_set_drvdata(pci_dev, efx);
3216 	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
3217 	rc = efx_init_struct(efx, pci_dev, net_dev);
3218 	if (rc)
3219 		goto fail1;
3220 
3221 	netif_info(efx, probe, efx->net_dev,
3222 		   "Solarflare NIC detected\n");
3223 
3224 	if (!efx->type->is_vf)
3225 		efx_probe_vpd_strings(efx);
3226 
3227 	/* Set up basic I/O (BAR mappings etc) */
3228 	rc = efx_init_io(efx);
3229 	if (rc)
3230 		goto fail2;
3231 
3232 	rc = efx_pci_probe_main(efx);
3233 	if (rc)
3234 		goto fail3;
3235 
3236 	rc = efx_register_netdev(efx);
3237 	if (rc)
3238 		goto fail4;
3239 
3240 	if (efx->type->sriov_init) {
3241 		rc = efx->type->sriov_init(efx);
3242 		if (rc)
3243 			netif_err(efx, probe, efx->net_dev,
3244 				  "SR-IOV can't be enabled rc %d\n", rc);
3245 	}
3246 
3247 	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
3248 
3249 	/* Try to create MTDs, but allow this to fail */
3250 	rtnl_lock();
3251 	rc = efx_mtd_probe(efx);
3252 	rtnl_unlock();
3253 	if (rc && rc != -EPERM)
3254 		netif_warn(efx, probe, efx->net_dev,
3255 			   "failed to create MTDs (%d)\n", rc);
3256 
3257 	rc = pci_enable_pcie_error_reporting(pci_dev);
3258 	if (rc && rc != -EINVAL)
3259 		netif_notice(efx, probe, efx->net_dev,
3260 			     "PCIE error reporting unavailable (%d).\n",
3261 			     rc);
3262 
3263 	return 0;
3264 
3265  fail4:
3266 	efx_pci_remove_main(efx);
3267  fail3:
3268 	efx_fini_io(efx);
3269  fail2:
3270 	efx_fini_struct(efx);
3271  fail1:
3272 	WARN_ON(rc > 0);
3273 	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
3274 	free_netdev(net_dev);
3275 	return rc;
3276 }
3277 
3278 /* efx_pci_sriov_configure returns the actual number of Virtual Functions
3279  * enabled on success
3280  */
3281 #ifdef CONFIG_SFC_SRIOV
3282 static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
3283 {
3284 	int rc;
3285 	struct efx_nic *efx = pci_get_drvdata(dev);
3286 
3287 	if (efx->type->sriov_configure) {
3288 		rc = efx->type->sriov_configure(efx, num_vfs);
3289 		if (rc)
3290 			return rc;
3291 		else
3292 			return num_vfs;
3293 	} else
3294 		return -EOPNOTSUPP;
3295 }
3296 #endif
3297 
3298 static int efx_pm_freeze(struct device *dev)
3299 {
3300 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3301 
3302 	rtnl_lock();
3303 
3304 	if (efx->state != STATE_DISABLED) {
3305 		efx->state = STATE_UNINIT;
3306 
3307 		efx_device_detach_sync(efx);
3308 
3309 		efx_stop_all(efx);
3310 		efx_disable_interrupts(efx);
3311 	}
3312 
3313 	rtnl_unlock();
3314 
3315 	return 0;
3316 }
3317 
3318 static int efx_pm_thaw(struct device *dev)
3319 {
3320 	int rc;
3321 	struct efx_nic *efx = pci_get_drvdata(to_pci_dev(dev));
3322 
3323 	rtnl_lock();
3324 
3325 	if (efx->state != STATE_DISABLED) {
3326 		rc = efx_enable_interrupts(efx);
3327 		if (rc)
3328 			goto fail;
3329 
3330 		mutex_lock(&efx->mac_lock);
3331 		efx->phy_op->reconfigure(efx);
3332 		mutex_unlock(&efx->mac_lock);
3333 
3334 		efx_start_all(efx);
3335 
3336 		netif_device_attach(efx->net_dev);
3337 
3338 		efx->state = STATE_READY;
3339 
3340 		efx->type->resume_wol(efx);
3341 	}
3342 
3343 	rtnl_unlock();
3344 
3345 	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
3346 	queue_work(reset_workqueue, &efx->reset_work);
3347 
3348 	return 0;
3349 
3350 fail:
3351 	rtnl_unlock();
3352 
3353 	return rc;
3354 }
3355 
3356 static int efx_pm_poweroff(struct device *dev)
3357 {
3358 	struct pci_dev *pci_dev = to_pci_dev(dev);
3359 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3360 
3361 	efx->type->fini(efx);
3362 
3363 	efx->reset_pending = 0;
3364 
3365 	pci_save_state(pci_dev);
3366 	return pci_set_power_state(pci_dev, PCI_D3hot);
3367 }
3368 
3369 /* Used for both resume and restore */
3370 static int efx_pm_resume(struct device *dev)
3371 {
3372 	struct pci_dev *pci_dev = to_pci_dev(dev);
3373 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3374 	int rc;
3375 
3376 	rc = pci_set_power_state(pci_dev, PCI_D0);
3377 	if (rc)
3378 		return rc;
3379 	pci_restore_state(pci_dev);
3380 	rc = pci_enable_device(pci_dev);
3381 	if (rc)
3382 		return rc;
3383 	pci_set_master(efx->pci_dev);
3384 	rc = efx->type->reset(efx, RESET_TYPE_ALL);
3385 	if (rc)
3386 		return rc;
3387 	rc = efx->type->init(efx);
3388 	if (rc)
3389 		return rc;
3390 	rc = efx_pm_thaw(dev);
3391 	return rc;
3392 }
3393 
3394 static int efx_pm_suspend(struct device *dev)
3395 {
3396 	int rc;
3397 
3398 	efx_pm_freeze(dev);
3399 	rc = efx_pm_poweroff(dev);
3400 	if (rc)
3401 		efx_pm_resume(dev);
3402 	return rc;
3403 }
3404 
3405 static const struct dev_pm_ops efx_pm_ops = {
3406 	.suspend	= efx_pm_suspend,
3407 	.resume		= efx_pm_resume,
3408 	.freeze		= efx_pm_freeze,
3409 	.thaw		= efx_pm_thaw,
3410 	.poweroff	= efx_pm_poweroff,
3411 	.restore	= efx_pm_resume,
3412 };
3413 
3414 /* A PCI error affecting this device was detected.
3415  * At this point MMIO and DMA may be disabled.
3416  * Stop the software path and request a slot reset.
3417  */
3418 static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
3419 					      enum pci_channel_state state)
3420 {
3421 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3422 	struct efx_nic *efx = pci_get_drvdata(pdev);
3423 
3424 	if (state == pci_channel_io_perm_failure)
3425 		return PCI_ERS_RESULT_DISCONNECT;
3426 
3427 	rtnl_lock();
3428 
3429 	if (efx->state != STATE_DISABLED) {
3430 		efx->state = STATE_RECOVERY;
3431 		efx->reset_pending = 0;
3432 
3433 		efx_device_detach_sync(efx);
3434 
3435 		efx_stop_all(efx);
3436 		efx_disable_interrupts(efx);
3437 
3438 		status = PCI_ERS_RESULT_NEED_RESET;
3439 	} else {
3440 		/* If the interface is disabled we don't want to do anything
3441 		 * with it.
3442 		 */
3443 		status = PCI_ERS_RESULT_RECOVERED;
3444 	}
3445 
3446 	rtnl_unlock();
3447 
3448 	pci_disable_device(pdev);
3449 
3450 	return status;
3451 }
3452 
3453 /* Fake a successful reset, which will be performed later in efx_io_resume. */
3454 static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
3455 {
3456 	struct efx_nic *efx = pci_get_drvdata(pdev);
3457 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3458 	int rc;
3459 
3460 	if (pci_enable_device(pdev)) {
3461 		netif_err(efx, hw, efx->net_dev,
3462 			  "Cannot re-enable PCI device after reset.\n");
3463 		status =  PCI_ERS_RESULT_DISCONNECT;
3464 	}
3465 
3466 	rc = pci_cleanup_aer_uncorrect_error_status(pdev);
3467 	if (rc) {
3468 		netif_err(efx, hw, efx->net_dev,
3469 		"pci_cleanup_aer_uncorrect_error_status failed (%d)\n", rc);
3470 		/* Non-fatal error. Continue. */
3471 	}
3472 
3473 	return status;
3474 }
3475 
3476 /* Perform the actual reset and resume I/O operations. */
3477 static void efx_io_resume(struct pci_dev *pdev)
3478 {
3479 	struct efx_nic *efx = pci_get_drvdata(pdev);
3480 	int rc;
3481 
3482 	rtnl_lock();
3483 
3484 	if (efx->state == STATE_DISABLED)
3485 		goto out;
3486 
3487 	rc = efx_reset(efx, RESET_TYPE_ALL);
3488 	if (rc) {
3489 		netif_err(efx, hw, efx->net_dev,
3490 			  "efx_reset failed after PCI error (%d)\n", rc);
3491 	} else {
3492 		efx->state = STATE_READY;
3493 		netif_dbg(efx, hw, efx->net_dev,
3494 			  "Done resetting and resuming IO after PCI error.\n");
3495 	}
3496 
3497 out:
3498 	rtnl_unlock();
3499 }
3500 
3501 /* For simplicity and reliability, we always require a slot reset and try to
3502  * reset the hardware when a pci error affecting the device is detected.
3503  * We leave both the link_reset and mmio_enabled callback unimplemented:
3504  * with our request for slot reset the mmio_enabled callback will never be
3505  * called, and the link_reset callback is not used by AER or EEH mechanisms.
3506  */
3507 static const struct pci_error_handlers efx_err_handlers = {
3508 	.error_detected = efx_io_error_detected,
3509 	.slot_reset	= efx_io_slot_reset,
3510 	.resume		= efx_io_resume,
3511 };
3512 
3513 static struct pci_driver efx_pci_driver = {
3514 	.name		= KBUILD_MODNAME,
3515 	.id_table	= efx_pci_table,
3516 	.probe		= efx_pci_probe,
3517 	.remove		= efx_pci_remove,
3518 	.driver.pm	= &efx_pm_ops,
3519 	.err_handler	= &efx_err_handlers,
3520 #ifdef CONFIG_SFC_SRIOV
3521 	.sriov_configure = efx_pci_sriov_configure,
3522 #endif
3523 };
3524 
3525 /**************************************************************************
3526  *
3527  * Kernel module interface
3528  *
3529  *************************************************************************/
3530 
3531 module_param(interrupt_mode, uint, 0444);
3532 MODULE_PARM_DESC(interrupt_mode,
3533 		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
3534 
3535 static int __init efx_init_module(void)
3536 {
3537 	int rc;
3538 
3539 	printk(KERN_INFO "Solarflare NET driver v" EFX_DRIVER_VERSION "\n");
3540 
3541 	rc = register_netdevice_notifier(&efx_netdev_notifier);
3542 	if (rc)
3543 		goto err_notifier;
3544 
3545 #ifdef CONFIG_SFC_SRIOV
3546 	rc = efx_init_sriov();
3547 	if (rc)
3548 		goto err_sriov;
3549 #endif
3550 
3551 	reset_workqueue = create_singlethread_workqueue("sfc_reset");
3552 	if (!reset_workqueue) {
3553 		rc = -ENOMEM;
3554 		goto err_reset;
3555 	}
3556 
3557 	rc = pci_register_driver(&efx_pci_driver);
3558 	if (rc < 0)
3559 		goto err_pci;
3560 
3561 	return 0;
3562 
3563  err_pci:
3564 	destroy_workqueue(reset_workqueue);
3565  err_reset:
3566 #ifdef CONFIG_SFC_SRIOV
3567 	efx_fini_sriov();
3568  err_sriov:
3569 #endif
3570 	unregister_netdevice_notifier(&efx_netdev_notifier);
3571  err_notifier:
3572 	return rc;
3573 }
3574 
3575 static void __exit efx_exit_module(void)
3576 {
3577 	printk(KERN_INFO "Solarflare NET driver unloading\n");
3578 
3579 	pci_unregister_driver(&efx_pci_driver);
3580 	destroy_workqueue(reset_workqueue);
3581 #ifdef CONFIG_SFC_SRIOV
3582 	efx_fini_sriov();
3583 #endif
3584 	unregister_netdevice_notifier(&efx_netdev_notifier);
3585 
3586 }
3587 
3588 module_init(efx_init_module);
3589 module_exit(efx_exit_module);
3590 
3591 MODULE_AUTHOR("Solarflare Communications and "
3592 	      "Michael Brown <mbrown@fensystems.co.uk>");
3593 MODULE_DESCRIPTION("Solarflare network driver");
3594 MODULE_LICENSE("GPL");
3595 MODULE_DEVICE_TABLE(pci, efx_pci_table);
3596