xref: /openbmc/linux/drivers/net/ethernet/sfc/efx.c (revision cc976614)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /****************************************************************************
3  * Driver for Solarflare network controllers and boards
4  * Copyright 2005-2006 Fen Systems Ltd.
5  * Copyright 2005-2013 Solarflare Communications Inc.
6  */
7 
8 #include <linux/module.h>
9 #include <linux/pci.h>
10 #include <linux/netdevice.h>
11 #include <linux/etherdevice.h>
12 #include <linux/delay.h>
13 #include <linux/notifier.h>
14 #include <linux/ip.h>
15 #include <linux/tcp.h>
16 #include <linux/in.h>
17 #include <linux/ethtool.h>
18 #include <linux/topology.h>
19 #include <linux/gfp.h>
20 #include <linux/aer.h>
21 #include <linux/interrupt.h>
22 #include "net_driver.h"
23 #include <net/gre.h>
24 #include <net/udp_tunnel.h>
25 #include "efx.h"
26 #include "nic.h"
27 #include "io.h"
28 #include "selftest.h"
29 #include "sriov.h"
30 
31 #include "mcdi.h"
32 #include "mcdi_pcol.h"
33 #include "workarounds.h"
34 
35 /**************************************************************************
36  *
37  * Type name strings
38  *
39  **************************************************************************
40  */
41 
42 /* Loopback mode names (see LOOPBACK_MODE()) */
43 const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
44 const char *const efx_loopback_mode_names[] = {
45 	[LOOPBACK_NONE]		= "NONE",
46 	[LOOPBACK_DATA]		= "DATAPATH",
47 	[LOOPBACK_GMAC]		= "GMAC",
48 	[LOOPBACK_XGMII]	= "XGMII",
49 	[LOOPBACK_XGXS]		= "XGXS",
50 	[LOOPBACK_XAUI]		= "XAUI",
51 	[LOOPBACK_GMII]		= "GMII",
52 	[LOOPBACK_SGMII]	= "SGMII",
53 	[LOOPBACK_XGBR]		= "XGBR",
54 	[LOOPBACK_XFI]		= "XFI",
55 	[LOOPBACK_XAUI_FAR]	= "XAUI_FAR",
56 	[LOOPBACK_GMII_FAR]	= "GMII_FAR",
57 	[LOOPBACK_SGMII_FAR]	= "SGMII_FAR",
58 	[LOOPBACK_XFI_FAR]	= "XFI_FAR",
59 	[LOOPBACK_GPHY]		= "GPHY",
60 	[LOOPBACK_PHYXS]	= "PHYXS",
61 	[LOOPBACK_PCS]		= "PCS",
62 	[LOOPBACK_PMAPMD]	= "PMA/PMD",
63 	[LOOPBACK_XPORT]	= "XPORT",
64 	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
65 	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
66 	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
67 	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
68 	[LOOPBACK_GMII_WS]	= "GMII_WS",
69 	[LOOPBACK_XFI_WS]	= "XFI_WS",
70 	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
71 	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
72 };
73 
74 const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
75 const char *const efx_reset_type_names[] = {
76 	[RESET_TYPE_INVISIBLE]          = "INVISIBLE",
77 	[RESET_TYPE_ALL]                = "ALL",
78 	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
79 	[RESET_TYPE_WORLD]              = "WORLD",
80 	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
81 	[RESET_TYPE_DATAPATH]           = "DATAPATH",
82 	[RESET_TYPE_MC_BIST]		= "MC_BIST",
83 	[RESET_TYPE_DISABLE]            = "DISABLE",
84 	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
85 	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
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 /* UDP tunnel type names */
93 static const char *const efx_udp_tunnel_type_names[] = {
94 	[TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN] = "vxlan",
95 	[TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE] = "geneve",
96 };
97 
98 void efx_get_udp_tunnel_type_name(u16 type, char *buf, size_t buflen)
99 {
100 	if (type < ARRAY_SIZE(efx_udp_tunnel_type_names) &&
101 	    efx_udp_tunnel_type_names[type] != NULL)
102 		snprintf(buf, buflen, "%s", efx_udp_tunnel_type_names[type]);
103 	else
104 		snprintf(buf, buflen, "type %d", type);
105 }
106 
107 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
108  * queued onto this work queue. This is not a per-nic work queue, because
109  * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
110  */
111 static struct workqueue_struct *reset_workqueue;
112 
113 /* How often and how many times to poll for a reset while waiting for a
114  * BIST that another function started to complete.
115  */
116 #define BIST_WAIT_DELAY_MS	100
117 #define BIST_WAIT_DELAY_COUNT	100
118 
119 /**************************************************************************
120  *
121  * Configurable values
122  *
123  *************************************************************************/
124 
125 /*
126  * Use separate channels for TX and RX events
127  *
128  * Set this to 1 to use separate channels for TX and RX. It allows us
129  * to control interrupt affinity separately for TX and RX.
130  *
131  * This is only used in MSI-X interrupt mode
132  */
133 bool efx_separate_tx_channels;
134 module_param(efx_separate_tx_channels, bool, 0444);
135 MODULE_PARM_DESC(efx_separate_tx_channels,
136 		 "Use separate channels for TX and RX");
137 
138 /* This is the weight assigned to each of the (per-channel) virtual
139  * NAPI devices.
140  */
141 static int napi_weight = 64;
142 
143 /* This is the time (in jiffies) between invocations of the hardware
144  * monitor.
145  * On Falcon-based NICs, this will:
146  * - Check the on-board hardware monitor;
147  * - Poll the link state and reconfigure the hardware as necessary.
148  * On Siena-based NICs for power systems with EEH support, this will give EEH a
149  * chance to start.
150  */
151 static unsigned int efx_monitor_interval = 1 * HZ;
152 
153 /* Initial interrupt moderation settings.  They can be modified after
154  * module load with ethtool.
155  *
156  * The default for RX should strike a balance between increasing the
157  * round-trip latency and reducing overhead.
158  */
159 static unsigned int rx_irq_mod_usec = 60;
160 
161 /* Initial interrupt moderation settings.  They can be modified after
162  * module load with ethtool.
163  *
164  * This default is chosen to ensure that a 10G link does not go idle
165  * while a TX queue is stopped after it has become full.  A queue is
166  * restarted when it drops below half full.  The time this takes (assuming
167  * worst case 3 descriptors per packet and 1024 descriptors) is
168  *   512 / 3 * 1.2 = 205 usec.
169  */
170 static unsigned int tx_irq_mod_usec = 150;
171 
172 /* This is the first interrupt mode to try out of:
173  * 0 => MSI-X
174  * 1 => MSI
175  * 2 => legacy
176  */
177 static unsigned int interrupt_mode;
178 
179 /* This is the requested number of CPUs to use for Receive-Side Scaling (RSS),
180  * i.e. the number of CPUs among which we may distribute simultaneous
181  * interrupt handling.
182  *
183  * Cards without MSI-X will only target one CPU via legacy or MSI interrupt.
184  * The default (0) means to assign an interrupt to each core.
185  */
186 static unsigned int rss_cpus;
187 module_param(rss_cpus, uint, 0444);
188 MODULE_PARM_DESC(rss_cpus, "Number of CPUs to use for Receive-Side Scaling");
189 
190 static bool phy_flash_cfg;
191 module_param(phy_flash_cfg, bool, 0644);
192 MODULE_PARM_DESC(phy_flash_cfg, "Set PHYs into reflash mode initially");
193 
194 static unsigned irq_adapt_low_thresh = 8000;
195 module_param(irq_adapt_low_thresh, uint, 0644);
196 MODULE_PARM_DESC(irq_adapt_low_thresh,
197 		 "Threshold score for reducing IRQ moderation");
198 
199 static unsigned irq_adapt_high_thresh = 16000;
200 module_param(irq_adapt_high_thresh, uint, 0644);
201 MODULE_PARM_DESC(irq_adapt_high_thresh,
202 		 "Threshold score for increasing IRQ moderation");
203 
204 static unsigned debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
205 			 NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
206 			 NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
207 			 NETIF_MSG_TX_ERR | NETIF_MSG_HW);
208 module_param(debug, uint, 0);
209 MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
210 
211 /**************************************************************************
212  *
213  * Utility functions and prototypes
214  *
215  *************************************************************************/
216 
217 static int efx_soft_enable_interrupts(struct efx_nic *efx);
218 static void efx_soft_disable_interrupts(struct efx_nic *efx);
219 static void efx_remove_channel(struct efx_channel *channel);
220 static void efx_remove_channels(struct efx_nic *efx);
221 static const struct efx_channel_type efx_default_channel_type;
222 static void efx_remove_port(struct efx_nic *efx);
223 static void efx_init_napi_channel(struct efx_channel *channel);
224 static void efx_fini_napi(struct efx_nic *efx);
225 static void efx_fini_napi_channel(struct efx_channel *channel);
226 static void efx_fini_struct(struct efx_nic *efx);
227 static void efx_start_all(struct efx_nic *efx);
228 static void efx_stop_all(struct efx_nic *efx);
229 static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog);
230 static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp);
231 static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs,
232 			u32 flags);
233 
234 #define EFX_ASSERT_RESET_SERIALISED(efx)		\
235 	do {						\
236 		if ((efx->state == STATE_READY) ||	\
237 		    (efx->state == STATE_RECOVERY) ||	\
238 		    (efx->state == STATE_DISABLED))	\
239 			ASSERT_RTNL();			\
240 	} while (0)
241 
242 static int efx_check_disabled(struct efx_nic *efx)
243 {
244 	if (efx->state == STATE_DISABLED || efx->state == STATE_RECOVERY) {
245 		netif_err(efx, drv, efx->net_dev,
246 			  "device is disabled due to earlier errors\n");
247 		return -EIO;
248 	}
249 	return 0;
250 }
251 
252 /**************************************************************************
253  *
254  * Event queue processing
255  *
256  *************************************************************************/
257 
258 /* Process channel's event queue
259  *
260  * This function is responsible for processing the event queue of a
261  * single channel.  The caller must guarantee that this function will
262  * never be concurrently called more than once on the same channel,
263  * though different channels may be being processed concurrently.
264  */
265 static int efx_process_channel(struct efx_channel *channel, int budget)
266 {
267 	struct efx_tx_queue *tx_queue;
268 	struct list_head rx_list;
269 	int spent;
270 
271 	if (unlikely(!channel->enabled))
272 		return 0;
273 
274 	/* Prepare the batch receive list */
275 	EFX_WARN_ON_PARANOID(channel->rx_list != NULL);
276 	INIT_LIST_HEAD(&rx_list);
277 	channel->rx_list = &rx_list;
278 
279 	efx_for_each_channel_tx_queue(tx_queue, channel) {
280 		tx_queue->pkts_compl = 0;
281 		tx_queue->bytes_compl = 0;
282 	}
283 
284 	spent = efx_nic_process_eventq(channel, budget);
285 	if (spent && efx_channel_has_rx_queue(channel)) {
286 		struct efx_rx_queue *rx_queue =
287 			efx_channel_get_rx_queue(channel);
288 
289 		efx_rx_flush_packet(channel);
290 		efx_fast_push_rx_descriptors(rx_queue, true);
291 	}
292 
293 	/* Update BQL */
294 	efx_for_each_channel_tx_queue(tx_queue, channel) {
295 		if (tx_queue->bytes_compl) {
296 			netdev_tx_completed_queue(tx_queue->core_txq,
297 				tx_queue->pkts_compl, tx_queue->bytes_compl);
298 		}
299 	}
300 
301 	/* Receive any packets we queued up */
302 	netif_receive_skb_list(channel->rx_list);
303 	channel->rx_list = NULL;
304 
305 	return spent;
306 }
307 
308 /* NAPI poll handler
309  *
310  * NAPI guarantees serialisation of polls of the same device, which
311  * provides the guarantee required by efx_process_channel().
312  */
313 static void efx_update_irq_mod(struct efx_nic *efx, struct efx_channel *channel)
314 {
315 	int step = efx->irq_mod_step_us;
316 
317 	if (channel->irq_mod_score < irq_adapt_low_thresh) {
318 		if (channel->irq_moderation_us > step) {
319 			channel->irq_moderation_us -= step;
320 			efx->type->push_irq_moderation(channel);
321 		}
322 	} else if (channel->irq_mod_score > irq_adapt_high_thresh) {
323 		if (channel->irq_moderation_us <
324 		    efx->irq_rx_moderation_us) {
325 			channel->irq_moderation_us += step;
326 			efx->type->push_irq_moderation(channel);
327 		}
328 	}
329 
330 	channel->irq_count = 0;
331 	channel->irq_mod_score = 0;
332 }
333 
334 static int efx_poll(struct napi_struct *napi, int budget)
335 {
336 	struct efx_channel *channel =
337 		container_of(napi, struct efx_channel, napi_str);
338 	struct efx_nic *efx = channel->efx;
339 	int spent;
340 
341 	netif_vdbg(efx, intr, efx->net_dev,
342 		   "channel %d NAPI poll executing on CPU %d\n",
343 		   channel->channel, raw_smp_processor_id());
344 
345 	spent = efx_process_channel(channel, budget);
346 
347 	xdp_do_flush_map();
348 
349 	if (spent < budget) {
350 		if (efx_channel_has_rx_queue(channel) &&
351 		    efx->irq_rx_adaptive &&
352 		    unlikely(++channel->irq_count == 1000)) {
353 			efx_update_irq_mod(efx, channel);
354 		}
355 
356 #ifdef CONFIG_RFS_ACCEL
357 		/* Perhaps expire some ARFS filters */
358 		mod_delayed_work(system_wq, &channel->filter_work, 0);
359 #endif
360 
361 		/* There is no race here; although napi_disable() will
362 		 * only wait for napi_complete(), this isn't a problem
363 		 * since efx_nic_eventq_read_ack() will have no effect if
364 		 * interrupts have already been disabled.
365 		 */
366 		if (napi_complete_done(napi, spent))
367 			efx_nic_eventq_read_ack(channel);
368 	}
369 
370 	return spent;
371 }
372 
373 /* Create event queue
374  * Event queue memory allocations are done only once.  If the channel
375  * is reset, the memory buffer will be reused; this guards against
376  * errors during channel reset and also simplifies interrupt handling.
377  */
378 static int efx_probe_eventq(struct efx_channel *channel)
379 {
380 	struct efx_nic *efx = channel->efx;
381 	unsigned long entries;
382 
383 	netif_dbg(efx, probe, efx->net_dev,
384 		  "chan %d create event queue\n", channel->channel);
385 
386 	/* Build an event queue with room for one event per tx and rx buffer,
387 	 * plus some extra for link state events and MCDI completions. */
388 	entries = roundup_pow_of_two(efx->rxq_entries + efx->txq_entries + 128);
389 	EFX_WARN_ON_PARANOID(entries > EFX_MAX_EVQ_SIZE);
390 	channel->eventq_mask = max(entries, EFX_MIN_EVQ_SIZE) - 1;
391 
392 	return efx_nic_probe_eventq(channel);
393 }
394 
395 /* Prepare channel's event queue */
396 static int efx_init_eventq(struct efx_channel *channel)
397 {
398 	struct efx_nic *efx = channel->efx;
399 	int rc;
400 
401 	EFX_WARN_ON_PARANOID(channel->eventq_init);
402 
403 	netif_dbg(efx, drv, efx->net_dev,
404 		  "chan %d init event queue\n", channel->channel);
405 
406 	rc = efx_nic_init_eventq(channel);
407 	if (rc == 0) {
408 		efx->type->push_irq_moderation(channel);
409 		channel->eventq_read_ptr = 0;
410 		channel->eventq_init = true;
411 	}
412 	return rc;
413 }
414 
415 /* Enable event queue processing and NAPI */
416 void efx_start_eventq(struct efx_channel *channel)
417 {
418 	netif_dbg(channel->efx, ifup, channel->efx->net_dev,
419 		  "chan %d start event queue\n", channel->channel);
420 
421 	/* Make sure the NAPI handler sees the enabled flag set */
422 	channel->enabled = true;
423 	smp_wmb();
424 
425 	napi_enable(&channel->napi_str);
426 	efx_nic_eventq_read_ack(channel);
427 }
428 
429 /* Disable event queue processing and NAPI */
430 void efx_stop_eventq(struct efx_channel *channel)
431 {
432 	if (!channel->enabled)
433 		return;
434 
435 	napi_disable(&channel->napi_str);
436 	channel->enabled = false;
437 }
438 
439 static void efx_fini_eventq(struct efx_channel *channel)
440 {
441 	if (!channel->eventq_init)
442 		return;
443 
444 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
445 		  "chan %d fini event queue\n", channel->channel);
446 
447 	efx_nic_fini_eventq(channel);
448 	channel->eventq_init = false;
449 }
450 
451 static void efx_remove_eventq(struct efx_channel *channel)
452 {
453 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
454 		  "chan %d remove event queue\n", channel->channel);
455 
456 	efx_nic_remove_eventq(channel);
457 }
458 
459 /**************************************************************************
460  *
461  * Channel handling
462  *
463  *************************************************************************/
464 
465 /* Allocate and initialise a channel structure. */
466 static struct efx_channel *
467 efx_alloc_channel(struct efx_nic *efx, int i, struct efx_channel *old_channel)
468 {
469 	struct efx_channel *channel;
470 	struct efx_rx_queue *rx_queue;
471 	struct efx_tx_queue *tx_queue;
472 	int j;
473 
474 	channel = kzalloc(sizeof(*channel), GFP_KERNEL);
475 	if (!channel)
476 		return NULL;
477 
478 	channel->efx = efx;
479 	channel->channel = i;
480 	channel->type = &efx_default_channel_type;
481 
482 	for (j = 0; j < EFX_TXQ_TYPES; j++) {
483 		tx_queue = &channel->tx_queue[j];
484 		tx_queue->efx = efx;
485 		tx_queue->queue = i * EFX_TXQ_TYPES + j;
486 		tx_queue->channel = channel;
487 	}
488 
489 #ifdef CONFIG_RFS_ACCEL
490 	INIT_DELAYED_WORK(&channel->filter_work, efx_filter_rfs_expire);
491 #endif
492 
493 	rx_queue = &channel->rx_queue;
494 	rx_queue->efx = efx;
495 	timer_setup(&rx_queue->slow_fill, efx_rx_slow_fill, 0);
496 
497 	return channel;
498 }
499 
500 /* Allocate and initialise a channel structure, copying parameters
501  * (but not resources) from an old channel structure.
502  */
503 static struct efx_channel *
504 efx_copy_channel(const struct efx_channel *old_channel)
505 {
506 	struct efx_channel *channel;
507 	struct efx_rx_queue *rx_queue;
508 	struct efx_tx_queue *tx_queue;
509 	int j;
510 
511 	channel = kmalloc(sizeof(*channel), GFP_KERNEL);
512 	if (!channel)
513 		return NULL;
514 
515 	*channel = *old_channel;
516 
517 	channel->napi_dev = NULL;
518 	INIT_HLIST_NODE(&channel->napi_str.napi_hash_node);
519 	channel->napi_str.napi_id = 0;
520 	channel->napi_str.state = 0;
521 	memset(&channel->eventq, 0, sizeof(channel->eventq));
522 
523 	for (j = 0; j < EFX_TXQ_TYPES; j++) {
524 		tx_queue = &channel->tx_queue[j];
525 		if (tx_queue->channel)
526 			tx_queue->channel = channel;
527 		tx_queue->buffer = NULL;
528 		memset(&tx_queue->txd, 0, sizeof(tx_queue->txd));
529 	}
530 
531 	rx_queue = &channel->rx_queue;
532 	rx_queue->buffer = NULL;
533 	memset(&rx_queue->rxd, 0, sizeof(rx_queue->rxd));
534 	timer_setup(&rx_queue->slow_fill, efx_rx_slow_fill, 0);
535 #ifdef CONFIG_RFS_ACCEL
536 	INIT_DELAYED_WORK(&channel->filter_work, efx_filter_rfs_expire);
537 #endif
538 
539 	return channel;
540 }
541 
542 static int efx_probe_channel(struct efx_channel *channel)
543 {
544 	struct efx_tx_queue *tx_queue;
545 	struct efx_rx_queue *rx_queue;
546 	int rc;
547 
548 	netif_dbg(channel->efx, probe, channel->efx->net_dev,
549 		  "creating channel %d\n", channel->channel);
550 
551 	rc = channel->type->pre_probe(channel);
552 	if (rc)
553 		goto fail;
554 
555 	rc = efx_probe_eventq(channel);
556 	if (rc)
557 		goto fail;
558 
559 	efx_for_each_channel_tx_queue(tx_queue, channel) {
560 		rc = efx_probe_tx_queue(tx_queue);
561 		if (rc)
562 			goto fail;
563 	}
564 
565 	efx_for_each_channel_rx_queue(rx_queue, channel) {
566 		rc = efx_probe_rx_queue(rx_queue);
567 		if (rc)
568 			goto fail;
569 	}
570 
571 	channel->rx_list = NULL;
572 
573 	return 0;
574 
575 fail:
576 	efx_remove_channel(channel);
577 	return rc;
578 }
579 
580 static void
581 efx_get_channel_name(struct efx_channel *channel, char *buf, size_t len)
582 {
583 	struct efx_nic *efx = channel->efx;
584 	const char *type;
585 	int number;
586 
587 	number = channel->channel;
588 
589 	if (number >= efx->xdp_channel_offset &&
590 	    !WARN_ON_ONCE(!efx->n_xdp_channels)) {
591 		type = "-xdp";
592 		number -= efx->xdp_channel_offset;
593 	} else if (efx->tx_channel_offset == 0) {
594 		type = "";
595 	} else if (number < efx->tx_channel_offset) {
596 		type = "-rx";
597 	} else {
598 		type = "-tx";
599 		number -= efx->tx_channel_offset;
600 	}
601 	snprintf(buf, len, "%s%s-%d", efx->name, type, number);
602 }
603 
604 static void efx_set_channel_names(struct efx_nic *efx)
605 {
606 	struct efx_channel *channel;
607 
608 	efx_for_each_channel(channel, efx)
609 		channel->type->get_name(channel,
610 					efx->msi_context[channel->channel].name,
611 					sizeof(efx->msi_context[0].name));
612 }
613 
614 static int efx_probe_channels(struct efx_nic *efx)
615 {
616 	struct efx_channel *channel;
617 	int rc;
618 
619 	/* Restart special buffer allocation */
620 	efx->next_buffer_table = 0;
621 
622 	/* Probe channels in reverse, so that any 'extra' channels
623 	 * use the start of the buffer table. This allows the traffic
624 	 * channels to be resized without moving them or wasting the
625 	 * entries before them.
626 	 */
627 	efx_for_each_channel_rev(channel, efx) {
628 		rc = efx_probe_channel(channel);
629 		if (rc) {
630 			netif_err(efx, probe, efx->net_dev,
631 				  "failed to create channel %d\n",
632 				  channel->channel);
633 			goto fail;
634 		}
635 	}
636 	efx_set_channel_names(efx);
637 
638 	return 0;
639 
640 fail:
641 	efx_remove_channels(efx);
642 	return rc;
643 }
644 
645 /* Channels are shutdown and reinitialised whilst the NIC is running
646  * to propagate configuration changes (mtu, checksum offload), or
647  * to clear hardware error conditions
648  */
649 static void efx_start_datapath(struct efx_nic *efx)
650 {
651 	netdev_features_t old_features = efx->net_dev->features;
652 	bool old_rx_scatter = efx->rx_scatter;
653 	struct efx_tx_queue *tx_queue;
654 	struct efx_rx_queue *rx_queue;
655 	struct efx_channel *channel;
656 	size_t rx_buf_len;
657 
658 	/* Calculate the rx buffer allocation parameters required to
659 	 * support the current MTU, including padding for header
660 	 * alignment and overruns.
661 	 */
662 	efx->rx_dma_len = (efx->rx_prefix_size +
663 			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
664 			   efx->type->rx_buffer_padding);
665 	rx_buf_len = (sizeof(struct efx_rx_page_state) + XDP_PACKET_HEADROOM +
666 		      efx->rx_ip_align + efx->rx_dma_len);
667 	if (rx_buf_len <= PAGE_SIZE) {
668 		efx->rx_scatter = efx->type->always_rx_scatter;
669 		efx->rx_buffer_order = 0;
670 	} else if (efx->type->can_rx_scatter) {
671 		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
672 		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
673 			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
674 				       EFX_RX_BUF_ALIGNMENT) >
675 			     PAGE_SIZE);
676 		efx->rx_scatter = true;
677 		efx->rx_dma_len = EFX_RX_USR_BUF_SIZE;
678 		efx->rx_buffer_order = 0;
679 	} else {
680 		efx->rx_scatter = false;
681 		efx->rx_buffer_order = get_order(rx_buf_len);
682 	}
683 
684 	efx_rx_config_page_split(efx);
685 	if (efx->rx_buffer_order)
686 		netif_dbg(efx, drv, efx->net_dev,
687 			  "RX buf len=%u; page order=%u batch=%u\n",
688 			  efx->rx_dma_len, efx->rx_buffer_order,
689 			  efx->rx_pages_per_batch);
690 	else
691 		netif_dbg(efx, drv, efx->net_dev,
692 			  "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
693 			  efx->rx_dma_len, efx->rx_page_buf_step,
694 			  efx->rx_bufs_per_page, efx->rx_pages_per_batch);
695 
696 	/* Restore previously fixed features in hw_features and remove
697 	 * features which are fixed now
698 	 */
699 	efx->net_dev->hw_features |= efx->net_dev->features;
700 	efx->net_dev->hw_features &= ~efx->fixed_features;
701 	efx->net_dev->features |= efx->fixed_features;
702 	if (efx->net_dev->features != old_features)
703 		netdev_features_change(efx->net_dev);
704 
705 	/* RX filters may also have scatter-enabled flags */
706 	if (efx->rx_scatter != old_rx_scatter)
707 		efx->type->filter_update_rx_scatter(efx);
708 
709 	/* We must keep at least one descriptor in a TX ring empty.
710 	 * We could avoid this when the queue size does not exactly
711 	 * match the hardware ring size, but it's not that important.
712 	 * Therefore we stop the queue when one more skb might fill
713 	 * the ring completely.  We wake it when half way back to
714 	 * empty.
715 	 */
716 	efx->txq_stop_thresh = efx->txq_entries - efx_tx_max_skb_descs(efx);
717 	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;
718 
719 	/* Initialise the channels */
720 	efx_for_each_channel(channel, efx) {
721 		efx_for_each_channel_tx_queue(tx_queue, channel) {
722 			efx_init_tx_queue(tx_queue);
723 			atomic_inc(&efx->active_queues);
724 		}
725 
726 		efx_for_each_channel_rx_queue(rx_queue, channel) {
727 			efx_init_rx_queue(rx_queue);
728 			atomic_inc(&efx->active_queues);
729 			efx_stop_eventq(channel);
730 			efx_fast_push_rx_descriptors(rx_queue, false);
731 			efx_start_eventq(channel);
732 		}
733 
734 		WARN_ON(channel->rx_pkt_n_frags);
735 	}
736 
737 	efx_ptp_start_datapath(efx);
738 
739 	if (netif_device_present(efx->net_dev))
740 		netif_tx_wake_all_queues(efx->net_dev);
741 }
742 
743 static void efx_stop_datapath(struct efx_nic *efx)
744 {
745 	struct efx_channel *channel;
746 	struct efx_tx_queue *tx_queue;
747 	struct efx_rx_queue *rx_queue;
748 	int rc;
749 
750 	EFX_ASSERT_RESET_SERIALISED(efx);
751 	BUG_ON(efx->port_enabled);
752 
753 	efx_ptp_stop_datapath(efx);
754 
755 	/* Stop RX refill */
756 	efx_for_each_channel(channel, efx) {
757 		efx_for_each_channel_rx_queue(rx_queue, channel)
758 			rx_queue->refill_enabled = false;
759 	}
760 
761 	efx_for_each_channel(channel, efx) {
762 		/* RX packet processing is pipelined, so wait for the
763 		 * NAPI handler to complete.  At least event queue 0
764 		 * might be kept active by non-data events, so don't
765 		 * use napi_synchronize() but actually disable NAPI
766 		 * temporarily.
767 		 */
768 		if (efx_channel_has_rx_queue(channel)) {
769 			efx_stop_eventq(channel);
770 			efx_start_eventq(channel);
771 		}
772 	}
773 
774 	rc = efx->type->fini_dmaq(efx);
775 	if (rc) {
776 		netif_err(efx, drv, efx->net_dev, "failed to flush queues\n");
777 	} else {
778 		netif_dbg(efx, drv, efx->net_dev,
779 			  "successfully flushed all queues\n");
780 	}
781 
782 	efx_for_each_channel(channel, efx) {
783 		efx_for_each_channel_rx_queue(rx_queue, channel)
784 			efx_fini_rx_queue(rx_queue);
785 		efx_for_each_possible_channel_tx_queue(tx_queue, channel)
786 			efx_fini_tx_queue(tx_queue);
787 	}
788 	efx->xdp_rxq_info_failed = false;
789 }
790 
791 static void efx_remove_channel(struct efx_channel *channel)
792 {
793 	struct efx_tx_queue *tx_queue;
794 	struct efx_rx_queue *rx_queue;
795 
796 	netif_dbg(channel->efx, drv, channel->efx->net_dev,
797 		  "destroy chan %d\n", channel->channel);
798 
799 	efx_for_each_channel_rx_queue(rx_queue, channel)
800 		efx_remove_rx_queue(rx_queue);
801 	efx_for_each_possible_channel_tx_queue(tx_queue, channel)
802 		efx_remove_tx_queue(tx_queue);
803 	efx_remove_eventq(channel);
804 	channel->type->post_remove(channel);
805 }
806 
807 static void efx_remove_channels(struct efx_nic *efx)
808 {
809 	struct efx_channel *channel;
810 
811 	efx_for_each_channel(channel, efx)
812 		efx_remove_channel(channel);
813 
814 	kfree(efx->xdp_tx_queues);
815 }
816 
817 int
818 efx_realloc_channels(struct efx_nic *efx, u32 rxq_entries, u32 txq_entries)
819 {
820 	struct efx_channel *other_channel[EFX_MAX_CHANNELS], *channel;
821 	u32 old_rxq_entries, old_txq_entries;
822 	unsigned i, next_buffer_table = 0;
823 	int rc, rc2;
824 
825 	rc = efx_check_disabled(efx);
826 	if (rc)
827 		return rc;
828 
829 	/* Not all channels should be reallocated. We must avoid
830 	 * reallocating their buffer table entries.
831 	 */
832 	efx_for_each_channel(channel, efx) {
833 		struct efx_rx_queue *rx_queue;
834 		struct efx_tx_queue *tx_queue;
835 
836 		if (channel->type->copy)
837 			continue;
838 		next_buffer_table = max(next_buffer_table,
839 					channel->eventq.index +
840 					channel->eventq.entries);
841 		efx_for_each_channel_rx_queue(rx_queue, channel)
842 			next_buffer_table = max(next_buffer_table,
843 						rx_queue->rxd.index +
844 						rx_queue->rxd.entries);
845 		efx_for_each_channel_tx_queue(tx_queue, channel)
846 			next_buffer_table = max(next_buffer_table,
847 						tx_queue->txd.index +
848 						tx_queue->txd.entries);
849 	}
850 
851 	efx_device_detach_sync(efx);
852 	efx_stop_all(efx);
853 	efx_soft_disable_interrupts(efx);
854 
855 	/* Clone channels (where possible) */
856 	memset(other_channel, 0, sizeof(other_channel));
857 	for (i = 0; i < efx->n_channels; i++) {
858 		channel = efx->channel[i];
859 		if (channel->type->copy)
860 			channel = channel->type->copy(channel);
861 		if (!channel) {
862 			rc = -ENOMEM;
863 			goto out;
864 		}
865 		other_channel[i] = channel;
866 	}
867 
868 	/* Swap entry counts and channel pointers */
869 	old_rxq_entries = efx->rxq_entries;
870 	old_txq_entries = efx->txq_entries;
871 	efx->rxq_entries = rxq_entries;
872 	efx->txq_entries = txq_entries;
873 	for (i = 0; i < efx->n_channels; i++) {
874 		channel = efx->channel[i];
875 		efx->channel[i] = other_channel[i];
876 		other_channel[i] = channel;
877 	}
878 
879 	/* Restart buffer table allocation */
880 	efx->next_buffer_table = next_buffer_table;
881 
882 	for (i = 0; i < efx->n_channels; i++) {
883 		channel = efx->channel[i];
884 		if (!channel->type->copy)
885 			continue;
886 		rc = efx_probe_channel(channel);
887 		if (rc)
888 			goto rollback;
889 		efx_init_napi_channel(efx->channel[i]);
890 	}
891 
892 out:
893 	/* Destroy unused channel structures */
894 	for (i = 0; i < efx->n_channels; i++) {
895 		channel = other_channel[i];
896 		if (channel && channel->type->copy) {
897 			efx_fini_napi_channel(channel);
898 			efx_remove_channel(channel);
899 			kfree(channel);
900 		}
901 	}
902 
903 	rc2 = efx_soft_enable_interrupts(efx);
904 	if (rc2) {
905 		rc = rc ? rc : rc2;
906 		netif_err(efx, drv, efx->net_dev,
907 			  "unable to restart interrupts on channel reallocation\n");
908 		efx_schedule_reset(efx, RESET_TYPE_DISABLE);
909 	} else {
910 		efx_start_all(efx);
911 		efx_device_attach_if_not_resetting(efx);
912 	}
913 	return rc;
914 
915 rollback:
916 	/* Swap back */
917 	efx->rxq_entries = old_rxq_entries;
918 	efx->txq_entries = old_txq_entries;
919 	for (i = 0; i < efx->n_channels; i++) {
920 		channel = efx->channel[i];
921 		efx->channel[i] = other_channel[i];
922 		other_channel[i] = channel;
923 	}
924 	goto out;
925 }
926 
927 void efx_schedule_slow_fill(struct efx_rx_queue *rx_queue)
928 {
929 	mod_timer(&rx_queue->slow_fill, jiffies + msecs_to_jiffies(10));
930 }
931 
932 static bool efx_default_channel_want_txqs(struct efx_channel *channel)
933 {
934 	return channel->channel - channel->efx->tx_channel_offset <
935 		channel->efx->n_tx_channels;
936 }
937 
938 static const struct efx_channel_type efx_default_channel_type = {
939 	.pre_probe		= efx_channel_dummy_op_int,
940 	.post_remove		= efx_channel_dummy_op_void,
941 	.get_name		= efx_get_channel_name,
942 	.copy			= efx_copy_channel,
943 	.want_txqs		= efx_default_channel_want_txqs,
944 	.keep_eventq		= false,
945 	.want_pio		= true,
946 };
947 
948 int efx_channel_dummy_op_int(struct efx_channel *channel)
949 {
950 	return 0;
951 }
952 
953 void efx_channel_dummy_op_void(struct efx_channel *channel)
954 {
955 }
956 
957 /**************************************************************************
958  *
959  * Port handling
960  *
961  **************************************************************************/
962 
963 /* This ensures that the kernel is kept informed (via
964  * netif_carrier_on/off) of the link status, and also maintains the
965  * link status's stop on the port's TX queue.
966  */
967 void efx_link_status_changed(struct efx_nic *efx)
968 {
969 	struct efx_link_state *link_state = &efx->link_state;
970 
971 	/* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
972 	 * that no events are triggered between unregister_netdev() and the
973 	 * driver unloading. A more general condition is that NETDEV_CHANGE
974 	 * can only be generated between NETDEV_UP and NETDEV_DOWN */
975 	if (!netif_running(efx->net_dev))
976 		return;
977 
978 	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
979 		efx->n_link_state_changes++;
980 
981 		if (link_state->up)
982 			netif_carrier_on(efx->net_dev);
983 		else
984 			netif_carrier_off(efx->net_dev);
985 	}
986 
987 	/* Status message for kernel log */
988 	if (link_state->up)
989 		netif_info(efx, link, efx->net_dev,
990 			   "link up at %uMbps %s-duplex (MTU %d)\n",
991 			   link_state->speed, link_state->fd ? "full" : "half",
992 			   efx->net_dev->mtu);
993 	else
994 		netif_info(efx, link, efx->net_dev, "link down\n");
995 }
996 
997 void efx_link_set_advertising(struct efx_nic *efx,
998 			      const unsigned long *advertising)
999 {
1000 	memcpy(efx->link_advertising, advertising,
1001 	       sizeof(__ETHTOOL_DECLARE_LINK_MODE_MASK()));
1002 
1003 	efx->link_advertising[0] |= ADVERTISED_Autoneg;
1004 	if (advertising[0] & ADVERTISED_Pause)
1005 		efx->wanted_fc |= (EFX_FC_TX | EFX_FC_RX);
1006 	else
1007 		efx->wanted_fc &= ~(EFX_FC_TX | EFX_FC_RX);
1008 	if (advertising[0] & ADVERTISED_Asym_Pause)
1009 		efx->wanted_fc ^= EFX_FC_TX;
1010 }
1011 
1012 /* Equivalent to efx_link_set_advertising with all-zeroes, except does not
1013  * force the Autoneg bit on.
1014  */
1015 void efx_link_clear_advertising(struct efx_nic *efx)
1016 {
1017 	bitmap_zero(efx->link_advertising, __ETHTOOL_LINK_MODE_MASK_NBITS);
1018 	efx->wanted_fc &= ~(EFX_FC_TX | EFX_FC_RX);
1019 }
1020 
1021 void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
1022 {
1023 	efx->wanted_fc = wanted_fc;
1024 	if (efx->link_advertising[0]) {
1025 		if (wanted_fc & EFX_FC_RX)
1026 			efx->link_advertising[0] |= (ADVERTISED_Pause |
1027 						     ADVERTISED_Asym_Pause);
1028 		else
1029 			efx->link_advertising[0] &= ~(ADVERTISED_Pause |
1030 						      ADVERTISED_Asym_Pause);
1031 		if (wanted_fc & EFX_FC_TX)
1032 			efx->link_advertising[0] ^= ADVERTISED_Asym_Pause;
1033 	}
1034 }
1035 
1036 static void efx_fini_port(struct efx_nic *efx);
1037 
1038 /* We assume that efx->type->reconfigure_mac will always try to sync RX
1039  * filters and therefore needs to read-lock the filter table against freeing
1040  */
1041 void efx_mac_reconfigure(struct efx_nic *efx)
1042 {
1043 	down_read(&efx->filter_sem);
1044 	efx->type->reconfigure_mac(efx);
1045 	up_read(&efx->filter_sem);
1046 }
1047 
1048 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
1049  * the MAC appropriately. All other PHY configuration changes are pushed
1050  * through phy_op->set_settings(), and pushed asynchronously to the MAC
1051  * through efx_monitor().
1052  *
1053  * Callers must hold the mac_lock
1054  */
1055 int __efx_reconfigure_port(struct efx_nic *efx)
1056 {
1057 	enum efx_phy_mode phy_mode;
1058 	int rc;
1059 
1060 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
1061 
1062 	/* Disable PHY transmit in mac level loopbacks */
1063 	phy_mode = efx->phy_mode;
1064 	if (LOOPBACK_INTERNAL(efx))
1065 		efx->phy_mode |= PHY_MODE_TX_DISABLED;
1066 	else
1067 		efx->phy_mode &= ~PHY_MODE_TX_DISABLED;
1068 
1069 	rc = efx->type->reconfigure_port(efx);
1070 
1071 	if (rc)
1072 		efx->phy_mode = phy_mode;
1073 
1074 	return rc;
1075 }
1076 
1077 /* Reinitialise the MAC to pick up new PHY settings, even if the port is
1078  * disabled. */
1079 int efx_reconfigure_port(struct efx_nic *efx)
1080 {
1081 	int rc;
1082 
1083 	EFX_ASSERT_RESET_SERIALISED(efx);
1084 
1085 	mutex_lock(&efx->mac_lock);
1086 	rc = __efx_reconfigure_port(efx);
1087 	mutex_unlock(&efx->mac_lock);
1088 
1089 	return rc;
1090 }
1091 
1092 /* Asynchronous work item for changing MAC promiscuity and multicast
1093  * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
1094  * MAC directly. */
1095 static void efx_mac_work(struct work_struct *data)
1096 {
1097 	struct efx_nic *efx = container_of(data, struct efx_nic, mac_work);
1098 
1099 	mutex_lock(&efx->mac_lock);
1100 	if (efx->port_enabled)
1101 		efx_mac_reconfigure(efx);
1102 	mutex_unlock(&efx->mac_lock);
1103 }
1104 
1105 static int efx_probe_port(struct efx_nic *efx)
1106 {
1107 	int rc;
1108 
1109 	netif_dbg(efx, probe, efx->net_dev, "create port\n");
1110 
1111 	if (phy_flash_cfg)
1112 		efx->phy_mode = PHY_MODE_SPECIAL;
1113 
1114 	/* Connect up MAC/PHY operations table */
1115 	rc = efx->type->probe_port(efx);
1116 	if (rc)
1117 		return rc;
1118 
1119 	/* Initialise MAC address to permanent address */
1120 	ether_addr_copy(efx->net_dev->dev_addr, efx->net_dev->perm_addr);
1121 
1122 	return 0;
1123 }
1124 
1125 static int efx_init_port(struct efx_nic *efx)
1126 {
1127 	int rc;
1128 
1129 	netif_dbg(efx, drv, efx->net_dev, "init port\n");
1130 
1131 	mutex_lock(&efx->mac_lock);
1132 
1133 	rc = efx->phy_op->init(efx);
1134 	if (rc)
1135 		goto fail1;
1136 
1137 	efx->port_initialized = true;
1138 
1139 	/* Reconfigure the MAC before creating dma queues (required for
1140 	 * Falcon/A1 where RX_INGR_EN/TX_DRAIN_EN isn't supported) */
1141 	efx_mac_reconfigure(efx);
1142 
1143 	/* Ensure the PHY advertises the correct flow control settings */
1144 	rc = efx->phy_op->reconfigure(efx);
1145 	if (rc && rc != -EPERM)
1146 		goto fail2;
1147 
1148 	mutex_unlock(&efx->mac_lock);
1149 	return 0;
1150 
1151 fail2:
1152 	efx->phy_op->fini(efx);
1153 fail1:
1154 	mutex_unlock(&efx->mac_lock);
1155 	return rc;
1156 }
1157 
1158 static void efx_start_port(struct efx_nic *efx)
1159 {
1160 	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
1161 	BUG_ON(efx->port_enabled);
1162 
1163 	mutex_lock(&efx->mac_lock);
1164 	efx->port_enabled = true;
1165 
1166 	/* Ensure MAC ingress/egress is enabled */
1167 	efx_mac_reconfigure(efx);
1168 
1169 	mutex_unlock(&efx->mac_lock);
1170 }
1171 
1172 /* Cancel work for MAC reconfiguration, periodic hardware monitoring
1173  * and the async self-test, wait for them to finish and prevent them
1174  * being scheduled again.  This doesn't cover online resets, which
1175  * should only be cancelled when removing the device.
1176  */
1177 static void efx_stop_port(struct efx_nic *efx)
1178 {
1179 	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
1180 
1181 	EFX_ASSERT_RESET_SERIALISED(efx);
1182 
1183 	mutex_lock(&efx->mac_lock);
1184 	efx->port_enabled = false;
1185 	mutex_unlock(&efx->mac_lock);
1186 
1187 	/* Serialise against efx_set_multicast_list() */
1188 	netif_addr_lock_bh(efx->net_dev);
1189 	netif_addr_unlock_bh(efx->net_dev);
1190 
1191 	cancel_delayed_work_sync(&efx->monitor_work);
1192 	efx_selftest_async_cancel(efx);
1193 	cancel_work_sync(&efx->mac_work);
1194 }
1195 
1196 static void efx_fini_port(struct efx_nic *efx)
1197 {
1198 	netif_dbg(efx, drv, efx->net_dev, "shut down port\n");
1199 
1200 	if (!efx->port_initialized)
1201 		return;
1202 
1203 	efx->phy_op->fini(efx);
1204 	efx->port_initialized = false;
1205 
1206 	efx->link_state.up = false;
1207 	efx_link_status_changed(efx);
1208 }
1209 
1210 static void efx_remove_port(struct efx_nic *efx)
1211 {
1212 	netif_dbg(efx, drv, efx->net_dev, "destroying port\n");
1213 
1214 	efx->type->remove_port(efx);
1215 }
1216 
1217 /**************************************************************************
1218  *
1219  * NIC handling
1220  *
1221  **************************************************************************/
1222 
1223 static LIST_HEAD(efx_primary_list);
1224 static LIST_HEAD(efx_unassociated_list);
1225 
1226 static bool efx_same_controller(struct efx_nic *left, struct efx_nic *right)
1227 {
1228 	return left->type == right->type &&
1229 		left->vpd_sn && right->vpd_sn &&
1230 		!strcmp(left->vpd_sn, right->vpd_sn);
1231 }
1232 
1233 static void efx_associate(struct efx_nic *efx)
1234 {
1235 	struct efx_nic *other, *next;
1236 
1237 	if (efx->primary == efx) {
1238 		/* Adding primary function; look for secondaries */
1239 
1240 		netif_dbg(efx, probe, efx->net_dev, "adding to primary list\n");
1241 		list_add_tail(&efx->node, &efx_primary_list);
1242 
1243 		list_for_each_entry_safe(other, next, &efx_unassociated_list,
1244 					 node) {
1245 			if (efx_same_controller(efx, other)) {
1246 				list_del(&other->node);
1247 				netif_dbg(other, probe, other->net_dev,
1248 					  "moving to secondary list of %s %s\n",
1249 					  pci_name(efx->pci_dev),
1250 					  efx->net_dev->name);
1251 				list_add_tail(&other->node,
1252 					      &efx->secondary_list);
1253 				other->primary = efx;
1254 			}
1255 		}
1256 	} else {
1257 		/* Adding secondary function; look for primary */
1258 
1259 		list_for_each_entry(other, &efx_primary_list, node) {
1260 			if (efx_same_controller(efx, other)) {
1261 				netif_dbg(efx, probe, efx->net_dev,
1262 					  "adding to secondary list of %s %s\n",
1263 					  pci_name(other->pci_dev),
1264 					  other->net_dev->name);
1265 				list_add_tail(&efx->node,
1266 					      &other->secondary_list);
1267 				efx->primary = other;
1268 				return;
1269 			}
1270 		}
1271 
1272 		netif_dbg(efx, probe, efx->net_dev,
1273 			  "adding to unassociated list\n");
1274 		list_add_tail(&efx->node, &efx_unassociated_list);
1275 	}
1276 }
1277 
1278 static void efx_dissociate(struct efx_nic *efx)
1279 {
1280 	struct efx_nic *other, *next;
1281 
1282 	list_del(&efx->node);
1283 	efx->primary = NULL;
1284 
1285 	list_for_each_entry_safe(other, next, &efx->secondary_list, node) {
1286 		list_del(&other->node);
1287 		netif_dbg(other, probe, other->net_dev,
1288 			  "moving to unassociated list\n");
1289 		list_add_tail(&other->node, &efx_unassociated_list);
1290 		other->primary = NULL;
1291 	}
1292 }
1293 
1294 /* This configures the PCI device to enable I/O and DMA. */
1295 static int efx_init_io(struct efx_nic *efx)
1296 {
1297 	struct pci_dev *pci_dev = efx->pci_dev;
1298 	dma_addr_t dma_mask = efx->type->max_dma_mask;
1299 	unsigned int mem_map_size = efx->type->mem_map_size(efx);
1300 	int rc, bar;
1301 
1302 	netif_dbg(efx, probe, efx->net_dev, "initialising I/O\n");
1303 
1304 	bar = efx->type->mem_bar(efx);
1305 
1306 	rc = pci_enable_device(pci_dev);
1307 	if (rc) {
1308 		netif_err(efx, probe, efx->net_dev,
1309 			  "failed to enable PCI device\n");
1310 		goto fail1;
1311 	}
1312 
1313 	pci_set_master(pci_dev);
1314 
1315 	/* Set the PCI DMA mask.  Try all possibilities from our genuine mask
1316 	 * down to 32 bits, because some architectures will allow 40 bit
1317 	 * masks event though they reject 46 bit masks.
1318 	 */
1319 	while (dma_mask > 0x7fffffffUL) {
1320 		rc = dma_set_mask_and_coherent(&pci_dev->dev, dma_mask);
1321 		if (rc == 0)
1322 			break;
1323 		dma_mask >>= 1;
1324 	}
1325 	if (rc) {
1326 		netif_err(efx, probe, efx->net_dev,
1327 			  "could not find a suitable DMA mask\n");
1328 		goto fail2;
1329 	}
1330 	netif_dbg(efx, probe, efx->net_dev,
1331 		  "using DMA mask %llx\n", (unsigned long long) dma_mask);
1332 
1333 	efx->membase_phys = pci_resource_start(efx->pci_dev, bar);
1334 	rc = pci_request_region(pci_dev, bar, "sfc");
1335 	if (rc) {
1336 		netif_err(efx, probe, efx->net_dev,
1337 			  "request for memory BAR failed\n");
1338 		rc = -EIO;
1339 		goto fail3;
1340 	}
1341 	efx->membase = ioremap_nocache(efx->membase_phys, mem_map_size);
1342 	if (!efx->membase) {
1343 		netif_err(efx, probe, efx->net_dev,
1344 			  "could not map memory BAR at %llx+%x\n",
1345 			  (unsigned long long)efx->membase_phys, mem_map_size);
1346 		rc = -ENOMEM;
1347 		goto fail4;
1348 	}
1349 	netif_dbg(efx, probe, efx->net_dev,
1350 		  "memory BAR at %llx+%x (virtual %p)\n",
1351 		  (unsigned long long)efx->membase_phys, mem_map_size,
1352 		  efx->membase);
1353 
1354 	return 0;
1355 
1356  fail4:
1357 	pci_release_region(efx->pci_dev, bar);
1358  fail3:
1359 	efx->membase_phys = 0;
1360  fail2:
1361 	pci_disable_device(efx->pci_dev);
1362  fail1:
1363 	return rc;
1364 }
1365 
1366 static void efx_fini_io(struct efx_nic *efx)
1367 {
1368 	int bar;
1369 
1370 	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1371 
1372 	if (efx->membase) {
1373 		iounmap(efx->membase);
1374 		efx->membase = NULL;
1375 	}
1376 
1377 	if (efx->membase_phys) {
1378 		bar = efx->type->mem_bar(efx);
1379 		pci_release_region(efx->pci_dev, bar);
1380 		efx->membase_phys = 0;
1381 	}
1382 
1383 	/* Don't disable bus-mastering if VFs are assigned */
1384 	if (!pci_vfs_assigned(efx->pci_dev))
1385 		pci_disable_device(efx->pci_dev);
1386 }
1387 
1388 void efx_set_default_rx_indir_table(struct efx_nic *efx,
1389 				    struct efx_rss_context *ctx)
1390 {
1391 	size_t i;
1392 
1393 	for (i = 0; i < ARRAY_SIZE(ctx->rx_indir_table); i++)
1394 		ctx->rx_indir_table[i] =
1395 			ethtool_rxfh_indir_default(i, efx->rss_spread);
1396 }
1397 
1398 static unsigned int efx_wanted_parallelism(struct efx_nic *efx)
1399 {
1400 	cpumask_var_t thread_mask;
1401 	unsigned int count;
1402 	int cpu;
1403 
1404 	if (rss_cpus) {
1405 		count = rss_cpus;
1406 	} else {
1407 		if (unlikely(!zalloc_cpumask_var(&thread_mask, GFP_KERNEL))) {
1408 			netif_warn(efx, probe, efx->net_dev,
1409 				   "RSS disabled due to allocation failure\n");
1410 			return 1;
1411 		}
1412 
1413 		count = 0;
1414 		for_each_online_cpu(cpu) {
1415 			if (!cpumask_test_cpu(cpu, thread_mask)) {
1416 				++count;
1417 				cpumask_or(thread_mask, thread_mask,
1418 					   topology_sibling_cpumask(cpu));
1419 			}
1420 		}
1421 
1422 		free_cpumask_var(thread_mask);
1423 	}
1424 
1425 	if (count > EFX_MAX_RX_QUEUES) {
1426 		netif_cond_dbg(efx, probe, efx->net_dev, !rss_cpus, warn,
1427 			       "Reducing number of rx queues from %u to %u.\n",
1428 			       count, EFX_MAX_RX_QUEUES);
1429 		count = EFX_MAX_RX_QUEUES;
1430 	}
1431 
1432 	/* If RSS is requested for the PF *and* VFs then we can't write RSS
1433 	 * table entries that are inaccessible to VFs
1434 	 */
1435 #ifdef CONFIG_SFC_SRIOV
1436 	if (efx->type->sriov_wanted) {
1437 		if (efx->type->sriov_wanted(efx) && efx_vf_size(efx) > 1 &&
1438 		    count > efx_vf_size(efx)) {
1439 			netif_warn(efx, probe, efx->net_dev,
1440 				   "Reducing number of RSS channels from %u to %u for "
1441 				   "VF support. Increase vf-msix-limit to use more "
1442 				   "channels on the PF.\n",
1443 				   count, efx_vf_size(efx));
1444 			count = efx_vf_size(efx);
1445 		}
1446 	}
1447 #endif
1448 
1449 	return count;
1450 }
1451 
1452 static int efx_allocate_msix_channels(struct efx_nic *efx,
1453 				      unsigned int max_channels,
1454 				      unsigned int extra_channels,
1455 				      unsigned int parallelism)
1456 {
1457 	unsigned int n_channels = parallelism;
1458 	int vec_count;
1459 	int n_xdp_tx;
1460 	int n_xdp_ev;
1461 
1462 	if (efx_separate_tx_channels)
1463 		n_channels *= 2;
1464 	n_channels += extra_channels;
1465 
1466 	/* To allow XDP transmit to happen from arbitrary NAPI contexts
1467 	 * we allocate a TX queue per CPU. We share event queues across
1468 	 * multiple tx queues, assuming tx and ev queues are both
1469 	 * maximum size.
1470 	 */
1471 
1472 	n_xdp_tx = num_possible_cpus();
1473 	n_xdp_ev = DIV_ROUND_UP(n_xdp_tx, EFX_TXQ_TYPES);
1474 
1475 	vec_count = pci_msix_vec_count(efx->pci_dev);
1476 	if (vec_count < 0)
1477 		return vec_count;
1478 
1479 	max_channels = min_t(unsigned int, vec_count, max_channels);
1480 
1481 	/* Check resources.
1482 	 * We need a channel per event queue, plus a VI per tx queue.
1483 	 * This may be more pessimistic than it needs to be.
1484 	 */
1485 	if (n_channels + n_xdp_ev > max_channels) {
1486 		netif_err(efx, drv, efx->net_dev,
1487 			  "Insufficient resources for %d XDP event queues (%d other channels, max %d)\n",
1488 			  n_xdp_ev, n_channels, max_channels);
1489 		efx->n_xdp_channels = 0;
1490 		efx->xdp_tx_per_channel = 0;
1491 		efx->xdp_tx_queue_count = 0;
1492 	} else {
1493 		efx->n_xdp_channels = n_xdp_ev;
1494 		efx->xdp_tx_per_channel = EFX_TXQ_TYPES;
1495 		efx->xdp_tx_queue_count = n_xdp_tx;
1496 		n_channels += n_xdp_ev;
1497 		netif_dbg(efx, drv, efx->net_dev,
1498 			  "Allocating %d TX and %d event queues for XDP\n",
1499 			  n_xdp_tx, n_xdp_ev);
1500 	}
1501 
1502 	if (vec_count < n_channels) {
1503 		netif_err(efx, drv, efx->net_dev,
1504 			  "WARNING: Insufficient MSI-X vectors available (%d < %u).\n",
1505 			  vec_count, n_channels);
1506 		netif_err(efx, drv, efx->net_dev,
1507 			  "WARNING: Performance may be reduced.\n");
1508 		n_channels = vec_count;
1509 	}
1510 
1511 	n_channels = min(n_channels, max_channels);
1512 
1513 	efx->n_channels = n_channels;
1514 
1515 	/* Ignore XDP tx channels when creating rx channels. */
1516 	n_channels -= efx->n_xdp_channels;
1517 
1518 	if (efx_separate_tx_channels) {
1519 		efx->n_tx_channels =
1520 			min(max(n_channels / 2, 1U),
1521 			    efx->max_tx_channels);
1522 		efx->tx_channel_offset =
1523 			n_channels - efx->n_tx_channels;
1524 		efx->n_rx_channels =
1525 			max(n_channels -
1526 			    efx->n_tx_channels, 1U);
1527 	} else {
1528 		efx->n_tx_channels = min(n_channels, efx->max_tx_channels);
1529 		efx->tx_channel_offset = 0;
1530 		efx->n_rx_channels = n_channels;
1531 	}
1532 
1533 	efx->n_rx_channels = min(efx->n_rx_channels, parallelism);
1534 	efx->n_tx_channels = min(efx->n_tx_channels, parallelism);
1535 
1536 	efx->xdp_channel_offset = n_channels;
1537 
1538 	netif_dbg(efx, drv, efx->net_dev,
1539 		  "Allocating %u RX channels\n",
1540 		  efx->n_rx_channels);
1541 
1542 	return efx->n_channels;
1543 }
1544 
1545 /* Probe the number and type of interrupts we are able to obtain, and
1546  * the resulting numbers of channels and RX queues.
1547  */
1548 static int efx_probe_interrupts(struct efx_nic *efx)
1549 {
1550 	unsigned int extra_channels = 0;
1551 	unsigned int rss_spread;
1552 	unsigned int i, j;
1553 	int rc;
1554 
1555 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++)
1556 		if (efx->extra_channel_type[i])
1557 			++extra_channels;
1558 
1559 	if (efx->interrupt_mode == EFX_INT_MODE_MSIX) {
1560 		unsigned int parallelism = efx_wanted_parallelism(efx);
1561 		struct msix_entry xentries[EFX_MAX_CHANNELS];
1562 		unsigned int n_channels;
1563 
1564 		rc = efx_allocate_msix_channels(efx, efx->max_channels,
1565 						extra_channels, parallelism);
1566 		if (rc >= 0) {
1567 			n_channels = rc;
1568 			for (i = 0; i < n_channels; i++)
1569 				xentries[i].entry = i;
1570 			rc = pci_enable_msix_range(efx->pci_dev, xentries, 1,
1571 						   n_channels);
1572 		}
1573 		if (rc < 0) {
1574 			/* Fall back to single channel MSI */
1575 			netif_err(efx, drv, efx->net_dev,
1576 				  "could not enable MSI-X\n");
1577 			if (efx->type->min_interrupt_mode >= EFX_INT_MODE_MSI)
1578 				efx->interrupt_mode = EFX_INT_MODE_MSI;
1579 			else
1580 				return rc;
1581 		} else if (rc < n_channels) {
1582 			netif_err(efx, drv, efx->net_dev,
1583 				  "WARNING: Insufficient MSI-X vectors"
1584 				  " available (%d < %u).\n", rc, n_channels);
1585 			netif_err(efx, drv, efx->net_dev,
1586 				  "WARNING: Performance may be reduced.\n");
1587 			n_channels = rc;
1588 		}
1589 
1590 		if (rc > 0) {
1591 			for (i = 0; i < efx->n_channels; i++)
1592 				efx_get_channel(efx, i)->irq =
1593 					xentries[i].vector;
1594 		}
1595 	}
1596 
1597 	/* Try single interrupt MSI */
1598 	if (efx->interrupt_mode == EFX_INT_MODE_MSI) {
1599 		efx->n_channels = 1;
1600 		efx->n_rx_channels = 1;
1601 		efx->n_tx_channels = 1;
1602 		efx->n_xdp_channels = 0;
1603 		efx->xdp_channel_offset = efx->n_channels;
1604 		rc = pci_enable_msi(efx->pci_dev);
1605 		if (rc == 0) {
1606 			efx_get_channel(efx, 0)->irq = efx->pci_dev->irq;
1607 		} else {
1608 			netif_err(efx, drv, efx->net_dev,
1609 				  "could not enable MSI\n");
1610 			if (efx->type->min_interrupt_mode >= EFX_INT_MODE_LEGACY)
1611 				efx->interrupt_mode = EFX_INT_MODE_LEGACY;
1612 			else
1613 				return rc;
1614 		}
1615 	}
1616 
1617 	/* Assume legacy interrupts */
1618 	if (efx->interrupt_mode == EFX_INT_MODE_LEGACY) {
1619 		efx->n_channels = 1 + (efx_separate_tx_channels ? 1 : 0);
1620 		efx->n_rx_channels = 1;
1621 		efx->n_tx_channels = 1;
1622 		efx->n_xdp_channels = 0;
1623 		efx->xdp_channel_offset = efx->n_channels;
1624 		efx->legacy_irq = efx->pci_dev->irq;
1625 	}
1626 
1627 	/* Assign extra channels if possible, before XDP channels */
1628 	efx->n_extra_tx_channels = 0;
1629 	j = efx->xdp_channel_offset;
1630 	for (i = 0; i < EFX_MAX_EXTRA_CHANNELS; i++) {
1631 		if (!efx->extra_channel_type[i])
1632 			continue;
1633 		if (j <= efx->tx_channel_offset + efx->n_tx_channels) {
1634 			efx->extra_channel_type[i]->handle_no_channel(efx);
1635 		} else {
1636 			--j;
1637 			efx_get_channel(efx, j)->type =
1638 				efx->extra_channel_type[i];
1639 			if (efx_channel_has_tx_queues(efx_get_channel(efx, j)))
1640 				efx->n_extra_tx_channels++;
1641 		}
1642 	}
1643 
1644 	rss_spread = efx->n_rx_channels;
1645 	/* RSS might be usable on VFs even if it is disabled on the PF */
1646 #ifdef CONFIG_SFC_SRIOV
1647 	if (efx->type->sriov_wanted) {
1648 		efx->rss_spread = ((rss_spread > 1 ||
1649 				    !efx->type->sriov_wanted(efx)) ?
1650 				   rss_spread : efx_vf_size(efx));
1651 		return 0;
1652 	}
1653 #endif
1654 	efx->rss_spread = rss_spread;
1655 
1656 	return 0;
1657 }
1658 
1659 #if defined(CONFIG_SMP)
1660 static void efx_set_interrupt_affinity(struct efx_nic *efx)
1661 {
1662 	struct efx_channel *channel;
1663 	unsigned int cpu;
1664 
1665 	efx_for_each_channel(channel, efx) {
1666 		cpu = cpumask_local_spread(channel->channel,
1667 					   pcibus_to_node(efx->pci_dev->bus));
1668 		irq_set_affinity_hint(channel->irq, cpumask_of(cpu));
1669 	}
1670 }
1671 
1672 static void efx_clear_interrupt_affinity(struct efx_nic *efx)
1673 {
1674 	struct efx_channel *channel;
1675 
1676 	efx_for_each_channel(channel, efx)
1677 		irq_set_affinity_hint(channel->irq, NULL);
1678 }
1679 #else
1680 static void
1681 efx_set_interrupt_affinity(struct efx_nic *efx __attribute__ ((unused)))
1682 {
1683 }
1684 
1685 static void
1686 efx_clear_interrupt_affinity(struct efx_nic *efx __attribute__ ((unused)))
1687 {
1688 }
1689 #endif /* CONFIG_SMP */
1690 
1691 static int efx_soft_enable_interrupts(struct efx_nic *efx)
1692 {
1693 	struct efx_channel *channel, *end_channel;
1694 	int rc;
1695 
1696 	BUG_ON(efx->state == STATE_DISABLED);
1697 
1698 	efx->irq_soft_enabled = true;
1699 	smp_wmb();
1700 
1701 	efx_for_each_channel(channel, efx) {
1702 		if (!channel->type->keep_eventq) {
1703 			rc = efx_init_eventq(channel);
1704 			if (rc)
1705 				goto fail;
1706 		}
1707 		efx_start_eventq(channel);
1708 	}
1709 
1710 	efx_mcdi_mode_event(efx);
1711 
1712 	return 0;
1713 fail:
1714 	end_channel = channel;
1715 	efx_for_each_channel(channel, efx) {
1716 		if (channel == end_channel)
1717 			break;
1718 		efx_stop_eventq(channel);
1719 		if (!channel->type->keep_eventq)
1720 			efx_fini_eventq(channel);
1721 	}
1722 
1723 	return rc;
1724 }
1725 
1726 static void efx_soft_disable_interrupts(struct efx_nic *efx)
1727 {
1728 	struct efx_channel *channel;
1729 
1730 	if (efx->state == STATE_DISABLED)
1731 		return;
1732 
1733 	efx_mcdi_mode_poll(efx);
1734 
1735 	efx->irq_soft_enabled = false;
1736 	smp_wmb();
1737 
1738 	if (efx->legacy_irq)
1739 		synchronize_irq(efx->legacy_irq);
1740 
1741 	efx_for_each_channel(channel, efx) {
1742 		if (channel->irq)
1743 			synchronize_irq(channel->irq);
1744 
1745 		efx_stop_eventq(channel);
1746 		if (!channel->type->keep_eventq)
1747 			efx_fini_eventq(channel);
1748 	}
1749 
1750 	/* Flush the asynchronous MCDI request queue */
1751 	efx_mcdi_flush_async(efx);
1752 }
1753 
1754 static int efx_enable_interrupts(struct efx_nic *efx)
1755 {
1756 	struct efx_channel *channel, *end_channel;
1757 	int rc;
1758 
1759 	BUG_ON(efx->state == STATE_DISABLED);
1760 
1761 	if (efx->eeh_disabled_legacy_irq) {
1762 		enable_irq(efx->legacy_irq);
1763 		efx->eeh_disabled_legacy_irq = false;
1764 	}
1765 
1766 	efx->type->irq_enable_master(efx);
1767 
1768 	efx_for_each_channel(channel, efx) {
1769 		if (channel->type->keep_eventq) {
1770 			rc = efx_init_eventq(channel);
1771 			if (rc)
1772 				goto fail;
1773 		}
1774 	}
1775 
1776 	rc = efx_soft_enable_interrupts(efx);
1777 	if (rc)
1778 		goto fail;
1779 
1780 	return 0;
1781 
1782 fail:
1783 	end_channel = channel;
1784 	efx_for_each_channel(channel, efx) {
1785 		if (channel == end_channel)
1786 			break;
1787 		if (channel->type->keep_eventq)
1788 			efx_fini_eventq(channel);
1789 	}
1790 
1791 	efx->type->irq_disable_non_ev(efx);
1792 
1793 	return rc;
1794 }
1795 
1796 static void efx_disable_interrupts(struct efx_nic *efx)
1797 {
1798 	struct efx_channel *channel;
1799 
1800 	efx_soft_disable_interrupts(efx);
1801 
1802 	efx_for_each_channel(channel, efx) {
1803 		if (channel->type->keep_eventq)
1804 			efx_fini_eventq(channel);
1805 	}
1806 
1807 	efx->type->irq_disable_non_ev(efx);
1808 }
1809 
1810 static void efx_remove_interrupts(struct efx_nic *efx)
1811 {
1812 	struct efx_channel *channel;
1813 
1814 	/* Remove MSI/MSI-X interrupts */
1815 	efx_for_each_channel(channel, efx)
1816 		channel->irq = 0;
1817 	pci_disable_msi(efx->pci_dev);
1818 	pci_disable_msix(efx->pci_dev);
1819 
1820 	/* Remove legacy interrupt */
1821 	efx->legacy_irq = 0;
1822 }
1823 
1824 static int efx_set_channels(struct efx_nic *efx)
1825 {
1826 	struct efx_channel *channel;
1827 	struct efx_tx_queue *tx_queue;
1828 	int xdp_queue_number;
1829 
1830 	efx->tx_channel_offset =
1831 		efx_separate_tx_channels ?
1832 		efx->n_channels - efx->n_tx_channels : 0;
1833 
1834 	if (efx->xdp_tx_queue_count) {
1835 		EFX_WARN_ON_PARANOID(efx->xdp_tx_queues);
1836 
1837 		/* Allocate array for XDP TX queue lookup. */
1838 		efx->xdp_tx_queues = kcalloc(efx->xdp_tx_queue_count,
1839 					     sizeof(*efx->xdp_tx_queues),
1840 					     GFP_KERNEL);
1841 		if (!efx->xdp_tx_queues)
1842 			return -ENOMEM;
1843 	}
1844 
1845 	/* We need to mark which channels really have RX and TX
1846 	 * queues, and adjust the TX queue numbers if we have separate
1847 	 * RX-only and TX-only channels.
1848 	 */
1849 	xdp_queue_number = 0;
1850 	efx_for_each_channel(channel, efx) {
1851 		if (channel->channel < efx->n_rx_channels)
1852 			channel->rx_queue.core_index = channel->channel;
1853 		else
1854 			channel->rx_queue.core_index = -1;
1855 
1856 		efx_for_each_channel_tx_queue(tx_queue, channel) {
1857 			tx_queue->queue -= (efx->tx_channel_offset *
1858 					    EFX_TXQ_TYPES);
1859 
1860 			if (efx_channel_is_xdp_tx(channel) &&
1861 			    xdp_queue_number < efx->xdp_tx_queue_count) {
1862 				efx->xdp_tx_queues[xdp_queue_number] = tx_queue;
1863 				xdp_queue_number++;
1864 			}
1865 		}
1866 	}
1867 	return 0;
1868 }
1869 
1870 static int efx_probe_nic(struct efx_nic *efx)
1871 {
1872 	int rc;
1873 
1874 	netif_dbg(efx, probe, efx->net_dev, "creating NIC\n");
1875 
1876 	/* Carry out hardware-type specific initialisation */
1877 	rc = efx->type->probe(efx);
1878 	if (rc)
1879 		return rc;
1880 
1881 	do {
1882 		if (!efx->max_channels || !efx->max_tx_channels) {
1883 			netif_err(efx, drv, efx->net_dev,
1884 				  "Insufficient resources to allocate"
1885 				  " any channels\n");
1886 			rc = -ENOSPC;
1887 			goto fail1;
1888 		}
1889 
1890 		/* Determine the number of channels and queues by trying
1891 		 * to hook in MSI-X interrupts.
1892 		 */
1893 		rc = efx_probe_interrupts(efx);
1894 		if (rc)
1895 			goto fail1;
1896 
1897 		rc = efx_set_channels(efx);
1898 		if (rc)
1899 			goto fail1;
1900 
1901 		/* dimension_resources can fail with EAGAIN */
1902 		rc = efx->type->dimension_resources(efx);
1903 		if (rc != 0 && rc != -EAGAIN)
1904 			goto fail2;
1905 
1906 		if (rc == -EAGAIN)
1907 			/* try again with new max_channels */
1908 			efx_remove_interrupts(efx);
1909 
1910 	} while (rc == -EAGAIN);
1911 
1912 	if (efx->n_channels > 1)
1913 		netdev_rss_key_fill(efx->rss_context.rx_hash_key,
1914 				    sizeof(efx->rss_context.rx_hash_key));
1915 	efx_set_default_rx_indir_table(efx, &efx->rss_context);
1916 
1917 	netif_set_real_num_tx_queues(efx->net_dev, efx->n_tx_channels);
1918 	netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels);
1919 
1920 	/* Initialise the interrupt moderation settings */
1921 	efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000);
1922 	efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true,
1923 				true);
1924 
1925 	return 0;
1926 
1927 fail2:
1928 	efx_remove_interrupts(efx);
1929 fail1:
1930 	efx->type->remove(efx);
1931 	return rc;
1932 }
1933 
1934 static void efx_remove_nic(struct efx_nic *efx)
1935 {
1936 	netif_dbg(efx, drv, efx->net_dev, "destroying NIC\n");
1937 
1938 	efx_remove_interrupts(efx);
1939 	efx->type->remove(efx);
1940 }
1941 
1942 static int efx_probe_filters(struct efx_nic *efx)
1943 {
1944 	int rc;
1945 
1946 	init_rwsem(&efx->filter_sem);
1947 	mutex_lock(&efx->mac_lock);
1948 	down_write(&efx->filter_sem);
1949 	rc = efx->type->filter_table_probe(efx);
1950 	if (rc)
1951 		goto out_unlock;
1952 
1953 #ifdef CONFIG_RFS_ACCEL
1954 	if (efx->type->offload_features & NETIF_F_NTUPLE) {
1955 		struct efx_channel *channel;
1956 		int i, success = 1;
1957 
1958 		efx_for_each_channel(channel, efx) {
1959 			channel->rps_flow_id =
1960 				kcalloc(efx->type->max_rx_ip_filters,
1961 					sizeof(*channel->rps_flow_id),
1962 					GFP_KERNEL);
1963 			if (!channel->rps_flow_id)
1964 				success = 0;
1965 			else
1966 				for (i = 0;
1967 				     i < efx->type->max_rx_ip_filters;
1968 				     ++i)
1969 					channel->rps_flow_id[i] =
1970 						RPS_FLOW_ID_INVALID;
1971 			channel->rfs_expire_index = 0;
1972 			channel->rfs_filter_count = 0;
1973 		}
1974 
1975 		if (!success) {
1976 			efx_for_each_channel(channel, efx)
1977 				kfree(channel->rps_flow_id);
1978 			efx->type->filter_table_remove(efx);
1979 			rc = -ENOMEM;
1980 			goto out_unlock;
1981 		}
1982 	}
1983 #endif
1984 out_unlock:
1985 	up_write(&efx->filter_sem);
1986 	mutex_unlock(&efx->mac_lock);
1987 	return rc;
1988 }
1989 
1990 static void efx_remove_filters(struct efx_nic *efx)
1991 {
1992 #ifdef CONFIG_RFS_ACCEL
1993 	struct efx_channel *channel;
1994 
1995 	efx_for_each_channel(channel, efx) {
1996 		cancel_delayed_work_sync(&channel->filter_work);
1997 		kfree(channel->rps_flow_id);
1998 	}
1999 #endif
2000 	down_write(&efx->filter_sem);
2001 	efx->type->filter_table_remove(efx);
2002 	up_write(&efx->filter_sem);
2003 }
2004 
2005 
2006 /**************************************************************************
2007  *
2008  * NIC startup/shutdown
2009  *
2010  *************************************************************************/
2011 
2012 static int efx_probe_all(struct efx_nic *efx)
2013 {
2014 	int rc;
2015 
2016 	rc = efx_probe_nic(efx);
2017 	if (rc) {
2018 		netif_err(efx, probe, efx->net_dev, "failed to create NIC\n");
2019 		goto fail1;
2020 	}
2021 
2022 	rc = efx_probe_port(efx);
2023 	if (rc) {
2024 		netif_err(efx, probe, efx->net_dev, "failed to create port\n");
2025 		goto fail2;
2026 	}
2027 
2028 	BUILD_BUG_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_RXQ_MIN_ENT);
2029 	if (WARN_ON(EFX_DEFAULT_DMAQ_SIZE < EFX_TXQ_MIN_ENT(efx))) {
2030 		rc = -EINVAL;
2031 		goto fail3;
2032 	}
2033 	efx->rxq_entries = efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
2034 
2035 #ifdef CONFIG_SFC_SRIOV
2036 	rc = efx->type->vswitching_probe(efx);
2037 	if (rc) /* not fatal; the PF will still work fine */
2038 		netif_warn(efx, probe, efx->net_dev,
2039 			   "failed to setup vswitching rc=%d;"
2040 			   " VFs may not function\n", rc);
2041 #endif
2042 
2043 	rc = efx_probe_filters(efx);
2044 	if (rc) {
2045 		netif_err(efx, probe, efx->net_dev,
2046 			  "failed to create filter tables\n");
2047 		goto fail4;
2048 	}
2049 
2050 	rc = efx_probe_channels(efx);
2051 	if (rc)
2052 		goto fail5;
2053 
2054 	return 0;
2055 
2056  fail5:
2057 	efx_remove_filters(efx);
2058  fail4:
2059 #ifdef CONFIG_SFC_SRIOV
2060 	efx->type->vswitching_remove(efx);
2061 #endif
2062  fail3:
2063 	efx_remove_port(efx);
2064  fail2:
2065 	efx_remove_nic(efx);
2066  fail1:
2067 	return rc;
2068 }
2069 
2070 /* If the interface is supposed to be running but is not, start
2071  * the hardware and software data path, regular activity for the port
2072  * (MAC statistics, link polling, etc.) and schedule the port to be
2073  * reconfigured.  Interrupts must already be enabled.  This function
2074  * is safe to call multiple times, so long as the NIC is not disabled.
2075  * Requires the RTNL lock.
2076  */
2077 static void efx_start_all(struct efx_nic *efx)
2078 {
2079 	EFX_ASSERT_RESET_SERIALISED(efx);
2080 	BUG_ON(efx->state == STATE_DISABLED);
2081 
2082 	/* Check that it is appropriate to restart the interface. All
2083 	 * of these flags are safe to read under just the rtnl lock */
2084 	if (efx->port_enabled || !netif_running(efx->net_dev) ||
2085 	    efx->reset_pending)
2086 		return;
2087 
2088 	efx_start_port(efx);
2089 	efx_start_datapath(efx);
2090 
2091 	/* Start the hardware monitor if there is one */
2092 	if (efx->type->monitor != NULL)
2093 		queue_delayed_work(efx->workqueue, &efx->monitor_work,
2094 				   efx_monitor_interval);
2095 
2096 	/* Link state detection is normally event-driven; we have
2097 	 * to poll now because we could have missed a change
2098 	 */
2099 	mutex_lock(&efx->mac_lock);
2100 	if (efx->phy_op->poll(efx))
2101 		efx_link_status_changed(efx);
2102 	mutex_unlock(&efx->mac_lock);
2103 
2104 	efx->type->start_stats(efx);
2105 	efx->type->pull_stats(efx);
2106 	spin_lock_bh(&efx->stats_lock);
2107 	efx->type->update_stats(efx, NULL, NULL);
2108 	spin_unlock_bh(&efx->stats_lock);
2109 }
2110 
2111 /* Quiesce the hardware and software data path, and regular activity
2112  * for the port without bringing the link down.  Safe to call multiple
2113  * times with the NIC in almost any state, but interrupts should be
2114  * enabled.  Requires the RTNL lock.
2115  */
2116 static void efx_stop_all(struct efx_nic *efx)
2117 {
2118 	EFX_ASSERT_RESET_SERIALISED(efx);
2119 
2120 	/* port_enabled can be read safely under the rtnl lock */
2121 	if (!efx->port_enabled)
2122 		return;
2123 
2124 	/* update stats before we go down so we can accurately count
2125 	 * rx_nodesc_drops
2126 	 */
2127 	efx->type->pull_stats(efx);
2128 	spin_lock_bh(&efx->stats_lock);
2129 	efx->type->update_stats(efx, NULL, NULL);
2130 	spin_unlock_bh(&efx->stats_lock);
2131 	efx->type->stop_stats(efx);
2132 	efx_stop_port(efx);
2133 
2134 	/* Stop the kernel transmit interface.  This is only valid if
2135 	 * the device is stopped or detached; otherwise the watchdog
2136 	 * may fire immediately.
2137 	 */
2138 	WARN_ON(netif_running(efx->net_dev) &&
2139 		netif_device_present(efx->net_dev));
2140 	netif_tx_disable(efx->net_dev);
2141 
2142 	efx_stop_datapath(efx);
2143 }
2144 
2145 static void efx_remove_all(struct efx_nic *efx)
2146 {
2147 	rtnl_lock();
2148 	efx_xdp_setup_prog(efx, NULL);
2149 	rtnl_unlock();
2150 
2151 	efx_remove_channels(efx);
2152 	efx_remove_filters(efx);
2153 #ifdef CONFIG_SFC_SRIOV
2154 	efx->type->vswitching_remove(efx);
2155 #endif
2156 	efx_remove_port(efx);
2157 	efx_remove_nic(efx);
2158 }
2159 
2160 /**************************************************************************
2161  *
2162  * Interrupt moderation
2163  *
2164  **************************************************************************/
2165 unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs)
2166 {
2167 	if (usecs == 0)
2168 		return 0;
2169 	if (usecs * 1000 < efx->timer_quantum_ns)
2170 		return 1; /* never round down to 0 */
2171 	return usecs * 1000 / efx->timer_quantum_ns;
2172 }
2173 
2174 unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks)
2175 {
2176 	/* We must round up when converting ticks to microseconds
2177 	 * because we round down when converting the other way.
2178 	 */
2179 	return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000);
2180 }
2181 
2182 /* Set interrupt moderation parameters */
2183 int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs,
2184 			    unsigned int rx_usecs, bool rx_adaptive,
2185 			    bool rx_may_override_tx)
2186 {
2187 	struct efx_channel *channel;
2188 	unsigned int timer_max_us;
2189 
2190 	EFX_ASSERT_RESET_SERIALISED(efx);
2191 
2192 	timer_max_us = efx->timer_max_ns / 1000;
2193 
2194 	if (tx_usecs > timer_max_us || rx_usecs > timer_max_us)
2195 		return -EINVAL;
2196 
2197 	if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 &&
2198 	    !rx_may_override_tx) {
2199 		netif_err(efx, drv, efx->net_dev, "Channels are shared. "
2200 			  "RX and TX IRQ moderation must be equal\n");
2201 		return -EINVAL;
2202 	}
2203 
2204 	efx->irq_rx_adaptive = rx_adaptive;
2205 	efx->irq_rx_moderation_us = rx_usecs;
2206 	efx_for_each_channel(channel, efx) {
2207 		if (efx_channel_has_rx_queue(channel))
2208 			channel->irq_moderation_us = rx_usecs;
2209 		else if (efx_channel_has_tx_queues(channel))
2210 			channel->irq_moderation_us = tx_usecs;
2211 		else if (efx_channel_is_xdp_tx(channel))
2212 			channel->irq_moderation_us = tx_usecs;
2213 	}
2214 
2215 	return 0;
2216 }
2217 
2218 void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs,
2219 			    unsigned int *rx_usecs, bool *rx_adaptive)
2220 {
2221 	*rx_adaptive = efx->irq_rx_adaptive;
2222 	*rx_usecs = efx->irq_rx_moderation_us;
2223 
2224 	/* If channels are shared between RX and TX, so is IRQ
2225 	 * moderation.  Otherwise, IRQ moderation is the same for all
2226 	 * TX channels and is not adaptive.
2227 	 */
2228 	if (efx->tx_channel_offset == 0) {
2229 		*tx_usecs = *rx_usecs;
2230 	} else {
2231 		struct efx_channel *tx_channel;
2232 
2233 		tx_channel = efx->channel[efx->tx_channel_offset];
2234 		*tx_usecs = tx_channel->irq_moderation_us;
2235 	}
2236 }
2237 
2238 /**************************************************************************
2239  *
2240  * Hardware monitor
2241  *
2242  **************************************************************************/
2243 
2244 /* Run periodically off the general workqueue */
2245 static void efx_monitor(struct work_struct *data)
2246 {
2247 	struct efx_nic *efx = container_of(data, struct efx_nic,
2248 					   monitor_work.work);
2249 
2250 	netif_vdbg(efx, timer, efx->net_dev,
2251 		   "hardware monitor executing on CPU %d\n",
2252 		   raw_smp_processor_id());
2253 	BUG_ON(efx->type->monitor == NULL);
2254 
2255 	/* If the mac_lock is already held then it is likely a port
2256 	 * reconfiguration is already in place, which will likely do
2257 	 * most of the work of monitor() anyway. */
2258 	if (mutex_trylock(&efx->mac_lock)) {
2259 		if (efx->port_enabled)
2260 			efx->type->monitor(efx);
2261 		mutex_unlock(&efx->mac_lock);
2262 	}
2263 
2264 	queue_delayed_work(efx->workqueue, &efx->monitor_work,
2265 			   efx_monitor_interval);
2266 }
2267 
2268 /**************************************************************************
2269  *
2270  * ioctls
2271  *
2272  *************************************************************************/
2273 
2274 /* Net device ioctl
2275  * Context: process, rtnl_lock() held.
2276  */
2277 static int efx_ioctl(struct net_device *net_dev, struct ifreq *ifr, int cmd)
2278 {
2279 	struct efx_nic *efx = netdev_priv(net_dev);
2280 	struct mii_ioctl_data *data = if_mii(ifr);
2281 
2282 	if (cmd == SIOCSHWTSTAMP)
2283 		return efx_ptp_set_ts_config(efx, ifr);
2284 	if (cmd == SIOCGHWTSTAMP)
2285 		return efx_ptp_get_ts_config(efx, ifr);
2286 
2287 	/* Convert phy_id from older PRTAD/DEVAD format */
2288 	if ((cmd == SIOCGMIIREG || cmd == SIOCSMIIREG) &&
2289 	    (data->phy_id & 0xfc00) == 0x0400)
2290 		data->phy_id ^= MDIO_PHY_ID_C45 | 0x0400;
2291 
2292 	return mdio_mii_ioctl(&efx->mdio, data, cmd);
2293 }
2294 
2295 /**************************************************************************
2296  *
2297  * NAPI interface
2298  *
2299  **************************************************************************/
2300 
2301 static void efx_init_napi_channel(struct efx_channel *channel)
2302 {
2303 	struct efx_nic *efx = channel->efx;
2304 
2305 	channel->napi_dev = efx->net_dev;
2306 	netif_napi_add(channel->napi_dev, &channel->napi_str,
2307 		       efx_poll, napi_weight);
2308 }
2309 
2310 static void efx_init_napi(struct efx_nic *efx)
2311 {
2312 	struct efx_channel *channel;
2313 
2314 	efx_for_each_channel(channel, efx)
2315 		efx_init_napi_channel(channel);
2316 }
2317 
2318 static void efx_fini_napi_channel(struct efx_channel *channel)
2319 {
2320 	if (channel->napi_dev)
2321 		netif_napi_del(&channel->napi_str);
2322 
2323 	channel->napi_dev = NULL;
2324 }
2325 
2326 static void efx_fini_napi(struct efx_nic *efx)
2327 {
2328 	struct efx_channel *channel;
2329 
2330 	efx_for_each_channel(channel, efx)
2331 		efx_fini_napi_channel(channel);
2332 }
2333 
2334 /**************************************************************************
2335  *
2336  * Kernel net device interface
2337  *
2338  *************************************************************************/
2339 
2340 /* Context: process, rtnl_lock() held. */
2341 int efx_net_open(struct net_device *net_dev)
2342 {
2343 	struct efx_nic *efx = netdev_priv(net_dev);
2344 	int rc;
2345 
2346 	netif_dbg(efx, ifup, efx->net_dev, "opening device on CPU %d\n",
2347 		  raw_smp_processor_id());
2348 
2349 	rc = efx_check_disabled(efx);
2350 	if (rc)
2351 		return rc;
2352 	if (efx->phy_mode & PHY_MODE_SPECIAL)
2353 		return -EBUSY;
2354 	if (efx_mcdi_poll_reboot(efx) && efx_reset(efx, RESET_TYPE_ALL))
2355 		return -EIO;
2356 
2357 	/* Notify the kernel of the link state polled during driver load,
2358 	 * before the monitor starts running */
2359 	efx_link_status_changed(efx);
2360 
2361 	efx_start_all(efx);
2362 	if (efx->state == STATE_DISABLED || efx->reset_pending)
2363 		netif_device_detach(efx->net_dev);
2364 	efx_selftest_async_start(efx);
2365 	return 0;
2366 }
2367 
2368 /* Context: process, rtnl_lock() held.
2369  * Note that the kernel will ignore our return code; this method
2370  * should really be a void.
2371  */
2372 int efx_net_stop(struct net_device *net_dev)
2373 {
2374 	struct efx_nic *efx = netdev_priv(net_dev);
2375 
2376 	netif_dbg(efx, ifdown, efx->net_dev, "closing on CPU %d\n",
2377 		  raw_smp_processor_id());
2378 
2379 	/* Stop the device and flush all the channels */
2380 	efx_stop_all(efx);
2381 
2382 	return 0;
2383 }
2384 
2385 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
2386 static void efx_net_stats(struct net_device *net_dev,
2387 			  struct rtnl_link_stats64 *stats)
2388 {
2389 	struct efx_nic *efx = netdev_priv(net_dev);
2390 
2391 	spin_lock_bh(&efx->stats_lock);
2392 	efx->type->update_stats(efx, NULL, stats);
2393 	spin_unlock_bh(&efx->stats_lock);
2394 }
2395 
2396 /* Context: netif_tx_lock held, BHs disabled. */
2397 static void efx_watchdog(struct net_device *net_dev)
2398 {
2399 	struct efx_nic *efx = netdev_priv(net_dev);
2400 
2401 	netif_err(efx, tx_err, efx->net_dev,
2402 		  "TX stuck with port_enabled=%d: resetting channels\n",
2403 		  efx->port_enabled);
2404 
2405 	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
2406 }
2407 
2408 static unsigned int efx_xdp_max_mtu(struct efx_nic *efx)
2409 {
2410 	/* The maximum MTU that we can fit in a single page, allowing for
2411 	 * framing, overhead and XDP headroom.
2412 	 */
2413 	int overhead = EFX_MAX_FRAME_LEN(0) + sizeof(struct efx_rx_page_state) +
2414 		       efx->rx_prefix_size + efx->type->rx_buffer_padding +
2415 		       efx->rx_ip_align + XDP_PACKET_HEADROOM;
2416 
2417 	return PAGE_SIZE - overhead;
2418 }
2419 
2420 /* Context: process, rtnl_lock() held. */
2421 static int efx_change_mtu(struct net_device *net_dev, int new_mtu)
2422 {
2423 	struct efx_nic *efx = netdev_priv(net_dev);
2424 	int rc;
2425 
2426 	rc = efx_check_disabled(efx);
2427 	if (rc)
2428 		return rc;
2429 
2430 	if (rtnl_dereference(efx->xdp_prog) &&
2431 	    new_mtu > efx_xdp_max_mtu(efx)) {
2432 		netif_err(efx, drv, efx->net_dev,
2433 			  "Requested MTU of %d too big for XDP (max: %d)\n",
2434 			  new_mtu, efx_xdp_max_mtu(efx));
2435 		return -EINVAL;
2436 	}
2437 
2438 	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
2439 
2440 	efx_device_detach_sync(efx);
2441 	efx_stop_all(efx);
2442 
2443 	mutex_lock(&efx->mac_lock);
2444 	net_dev->mtu = new_mtu;
2445 	efx_mac_reconfigure(efx);
2446 	mutex_unlock(&efx->mac_lock);
2447 
2448 	efx_start_all(efx);
2449 	efx_device_attach_if_not_resetting(efx);
2450 	return 0;
2451 }
2452 
2453 static int efx_set_mac_address(struct net_device *net_dev, void *data)
2454 {
2455 	struct efx_nic *efx = netdev_priv(net_dev);
2456 	struct sockaddr *addr = data;
2457 	u8 *new_addr = addr->sa_data;
2458 	u8 old_addr[6];
2459 	int rc;
2460 
2461 	if (!is_valid_ether_addr(new_addr)) {
2462 		netif_err(efx, drv, efx->net_dev,
2463 			  "invalid ethernet MAC address requested: %pM\n",
2464 			  new_addr);
2465 		return -EADDRNOTAVAIL;
2466 	}
2467 
2468 	/* save old address */
2469 	ether_addr_copy(old_addr, net_dev->dev_addr);
2470 	ether_addr_copy(net_dev->dev_addr, new_addr);
2471 	if (efx->type->set_mac_address) {
2472 		rc = efx->type->set_mac_address(efx);
2473 		if (rc) {
2474 			ether_addr_copy(net_dev->dev_addr, old_addr);
2475 			return rc;
2476 		}
2477 	}
2478 
2479 	/* Reconfigure the MAC */
2480 	mutex_lock(&efx->mac_lock);
2481 	efx_mac_reconfigure(efx);
2482 	mutex_unlock(&efx->mac_lock);
2483 
2484 	return 0;
2485 }
2486 
2487 /* Context: netif_addr_lock held, BHs disabled. */
2488 static void efx_set_rx_mode(struct net_device *net_dev)
2489 {
2490 	struct efx_nic *efx = netdev_priv(net_dev);
2491 
2492 	if (efx->port_enabled)
2493 		queue_work(efx->workqueue, &efx->mac_work);
2494 	/* Otherwise efx_start_port() will do this */
2495 }
2496 
2497 static int efx_set_features(struct net_device *net_dev, netdev_features_t data)
2498 {
2499 	struct efx_nic *efx = netdev_priv(net_dev);
2500 	int rc;
2501 
2502 	/* If disabling RX n-tuple filtering, clear existing filters */
2503 	if (net_dev->features & ~data & NETIF_F_NTUPLE) {
2504 		rc = efx->type->filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);
2505 		if (rc)
2506 			return rc;
2507 	}
2508 
2509 	/* If Rx VLAN filter is changed, update filters via mac_reconfigure.
2510 	 * If rx-fcs is changed, mac_reconfigure updates that too.
2511 	 */
2512 	if ((net_dev->features ^ data) & (NETIF_F_HW_VLAN_CTAG_FILTER |
2513 					  NETIF_F_RXFCS)) {
2514 		/* efx_set_rx_mode() will schedule MAC work to update filters
2515 		 * when a new features are finally set in net_dev.
2516 		 */
2517 		efx_set_rx_mode(net_dev);
2518 	}
2519 
2520 	return 0;
2521 }
2522 
2523 static int efx_get_phys_port_id(struct net_device *net_dev,
2524 				struct netdev_phys_item_id *ppid)
2525 {
2526 	struct efx_nic *efx = netdev_priv(net_dev);
2527 
2528 	if (efx->type->get_phys_port_id)
2529 		return efx->type->get_phys_port_id(efx, ppid);
2530 	else
2531 		return -EOPNOTSUPP;
2532 }
2533 
2534 static int efx_get_phys_port_name(struct net_device *net_dev,
2535 				  char *name, size_t len)
2536 {
2537 	struct efx_nic *efx = netdev_priv(net_dev);
2538 
2539 	if (snprintf(name, len, "p%u", efx->port_num) >= len)
2540 		return -EINVAL;
2541 	return 0;
2542 }
2543 
2544 static int efx_vlan_rx_add_vid(struct net_device *net_dev, __be16 proto, u16 vid)
2545 {
2546 	struct efx_nic *efx = netdev_priv(net_dev);
2547 
2548 	if (efx->type->vlan_rx_add_vid)
2549 		return efx->type->vlan_rx_add_vid(efx, proto, vid);
2550 	else
2551 		return -EOPNOTSUPP;
2552 }
2553 
2554 static int efx_vlan_rx_kill_vid(struct net_device *net_dev, __be16 proto, u16 vid)
2555 {
2556 	struct efx_nic *efx = netdev_priv(net_dev);
2557 
2558 	if (efx->type->vlan_rx_kill_vid)
2559 		return efx->type->vlan_rx_kill_vid(efx, proto, vid);
2560 	else
2561 		return -EOPNOTSUPP;
2562 }
2563 
2564 static int efx_udp_tunnel_type_map(enum udp_parsable_tunnel_type in)
2565 {
2566 	switch (in) {
2567 	case UDP_TUNNEL_TYPE_VXLAN:
2568 		return TUNNEL_ENCAP_UDP_PORT_ENTRY_VXLAN;
2569 	case UDP_TUNNEL_TYPE_GENEVE:
2570 		return TUNNEL_ENCAP_UDP_PORT_ENTRY_GENEVE;
2571 	default:
2572 		return -1;
2573 	}
2574 }
2575 
2576 static void efx_udp_tunnel_add(struct net_device *dev, struct udp_tunnel_info *ti)
2577 {
2578 	struct efx_nic *efx = netdev_priv(dev);
2579 	struct efx_udp_tunnel tnl;
2580 	int efx_tunnel_type;
2581 
2582 	efx_tunnel_type = efx_udp_tunnel_type_map(ti->type);
2583 	if (efx_tunnel_type < 0)
2584 		return;
2585 
2586 	tnl.type = (u16)efx_tunnel_type;
2587 	tnl.port = ti->port;
2588 
2589 	if (efx->type->udp_tnl_add_port)
2590 		(void)efx->type->udp_tnl_add_port(efx, tnl);
2591 }
2592 
2593 static void efx_udp_tunnel_del(struct net_device *dev, struct udp_tunnel_info *ti)
2594 {
2595 	struct efx_nic *efx = netdev_priv(dev);
2596 	struct efx_udp_tunnel tnl;
2597 	int efx_tunnel_type;
2598 
2599 	efx_tunnel_type = efx_udp_tunnel_type_map(ti->type);
2600 	if (efx_tunnel_type < 0)
2601 		return;
2602 
2603 	tnl.type = (u16)efx_tunnel_type;
2604 	tnl.port = ti->port;
2605 
2606 	if (efx->type->udp_tnl_del_port)
2607 		(void)efx->type->udp_tnl_del_port(efx, tnl);
2608 }
2609 
2610 static const struct net_device_ops efx_netdev_ops = {
2611 	.ndo_open		= efx_net_open,
2612 	.ndo_stop		= efx_net_stop,
2613 	.ndo_get_stats64	= efx_net_stats,
2614 	.ndo_tx_timeout		= efx_watchdog,
2615 	.ndo_start_xmit		= efx_hard_start_xmit,
2616 	.ndo_validate_addr	= eth_validate_addr,
2617 	.ndo_do_ioctl		= efx_ioctl,
2618 	.ndo_change_mtu		= efx_change_mtu,
2619 	.ndo_set_mac_address	= efx_set_mac_address,
2620 	.ndo_set_rx_mode	= efx_set_rx_mode,
2621 	.ndo_set_features	= efx_set_features,
2622 	.ndo_vlan_rx_add_vid	= efx_vlan_rx_add_vid,
2623 	.ndo_vlan_rx_kill_vid	= efx_vlan_rx_kill_vid,
2624 #ifdef CONFIG_SFC_SRIOV
2625 	.ndo_set_vf_mac		= efx_sriov_set_vf_mac,
2626 	.ndo_set_vf_vlan	= efx_sriov_set_vf_vlan,
2627 	.ndo_set_vf_spoofchk	= efx_sriov_set_vf_spoofchk,
2628 	.ndo_get_vf_config	= efx_sriov_get_vf_config,
2629 	.ndo_set_vf_link_state  = efx_sriov_set_vf_link_state,
2630 #endif
2631 	.ndo_get_phys_port_id   = efx_get_phys_port_id,
2632 	.ndo_get_phys_port_name	= efx_get_phys_port_name,
2633 	.ndo_setup_tc		= efx_setup_tc,
2634 #ifdef CONFIG_RFS_ACCEL
2635 	.ndo_rx_flow_steer	= efx_filter_rfs,
2636 #endif
2637 	.ndo_udp_tunnel_add	= efx_udp_tunnel_add,
2638 	.ndo_udp_tunnel_del	= efx_udp_tunnel_del,
2639 	.ndo_xdp_xmit		= efx_xdp_xmit,
2640 	.ndo_bpf		= efx_xdp
2641 };
2642 
2643 static int efx_xdp_setup_prog(struct efx_nic *efx, struct bpf_prog *prog)
2644 {
2645 	struct bpf_prog *old_prog;
2646 
2647 	if (efx->xdp_rxq_info_failed) {
2648 		netif_err(efx, drv, efx->net_dev,
2649 			  "Unable to bind XDP program due to previous failure of rxq_info\n");
2650 		return -EINVAL;
2651 	}
2652 
2653 	if (prog && efx->net_dev->mtu > efx_xdp_max_mtu(efx)) {
2654 		netif_err(efx, drv, efx->net_dev,
2655 			  "Unable to configure XDP with MTU of %d (max: %d)\n",
2656 			  efx->net_dev->mtu, efx_xdp_max_mtu(efx));
2657 		return -EINVAL;
2658 	}
2659 
2660 	old_prog = rtnl_dereference(efx->xdp_prog);
2661 	rcu_assign_pointer(efx->xdp_prog, prog);
2662 	/* Release the reference that was originally passed by the caller. */
2663 	if (old_prog)
2664 		bpf_prog_put(old_prog);
2665 
2666 	return 0;
2667 }
2668 
2669 /* Context: process, rtnl_lock() held. */
2670 static int efx_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2671 {
2672 	struct efx_nic *efx = netdev_priv(dev);
2673 	struct bpf_prog *xdp_prog;
2674 
2675 	switch (xdp->command) {
2676 	case XDP_SETUP_PROG:
2677 		return efx_xdp_setup_prog(efx, xdp->prog);
2678 	case XDP_QUERY_PROG:
2679 		xdp_prog = rtnl_dereference(efx->xdp_prog);
2680 		xdp->prog_id = xdp_prog ? xdp_prog->aux->id : 0;
2681 		return 0;
2682 	default:
2683 		return -EINVAL;
2684 	}
2685 }
2686 
2687 static int efx_xdp_xmit(struct net_device *dev, int n, struct xdp_frame **xdpfs,
2688 			u32 flags)
2689 {
2690 	struct efx_nic *efx = netdev_priv(dev);
2691 
2692 	if (!netif_running(dev))
2693 		return -EINVAL;
2694 
2695 	return efx_xdp_tx_buffers(efx, n, xdpfs, flags & XDP_XMIT_FLUSH);
2696 }
2697 
2698 static void efx_update_name(struct efx_nic *efx)
2699 {
2700 	strcpy(efx->name, efx->net_dev->name);
2701 	efx_mtd_rename(efx);
2702 	efx_set_channel_names(efx);
2703 }
2704 
2705 static int efx_netdev_event(struct notifier_block *this,
2706 			    unsigned long event, void *ptr)
2707 {
2708 	struct net_device *net_dev = netdev_notifier_info_to_dev(ptr);
2709 
2710 	if ((net_dev->netdev_ops == &efx_netdev_ops) &&
2711 	    event == NETDEV_CHANGENAME)
2712 		efx_update_name(netdev_priv(net_dev));
2713 
2714 	return NOTIFY_DONE;
2715 }
2716 
2717 static struct notifier_block efx_netdev_notifier = {
2718 	.notifier_call = efx_netdev_event,
2719 };
2720 
2721 static ssize_t
2722 show_phy_type(struct device *dev, struct device_attribute *attr, char *buf)
2723 {
2724 	struct efx_nic *efx = dev_get_drvdata(dev);
2725 	return sprintf(buf, "%d\n", efx->phy_type);
2726 }
2727 static DEVICE_ATTR(phy_type, 0444, show_phy_type, NULL);
2728 
2729 #ifdef CONFIG_SFC_MCDI_LOGGING
2730 static ssize_t show_mcdi_log(struct device *dev, struct device_attribute *attr,
2731 			     char *buf)
2732 {
2733 	struct efx_nic *efx = dev_get_drvdata(dev);
2734 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2735 
2736 	return scnprintf(buf, PAGE_SIZE, "%d\n", mcdi->logging_enabled);
2737 }
2738 static ssize_t set_mcdi_log(struct device *dev, struct device_attribute *attr,
2739 			    const char *buf, size_t count)
2740 {
2741 	struct efx_nic *efx = dev_get_drvdata(dev);
2742 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
2743 	bool enable = count > 0 && *buf != '0';
2744 
2745 	mcdi->logging_enabled = enable;
2746 	return count;
2747 }
2748 static DEVICE_ATTR(mcdi_logging, 0644, show_mcdi_log, set_mcdi_log);
2749 #endif
2750 
2751 static int efx_register_netdev(struct efx_nic *efx)
2752 {
2753 	struct net_device *net_dev = efx->net_dev;
2754 	struct efx_channel *channel;
2755 	int rc;
2756 
2757 	net_dev->watchdog_timeo = 5 * HZ;
2758 	net_dev->irq = efx->pci_dev->irq;
2759 	net_dev->netdev_ops = &efx_netdev_ops;
2760 	if (efx_nic_rev(efx) >= EFX_REV_HUNT_A0)
2761 		net_dev->priv_flags |= IFF_UNICAST_FLT;
2762 	net_dev->ethtool_ops = &efx_ethtool_ops;
2763 	net_dev->gso_max_segs = EFX_TSO_MAX_SEGS;
2764 	net_dev->min_mtu = EFX_MIN_MTU;
2765 	net_dev->max_mtu = EFX_MAX_MTU;
2766 
2767 	rtnl_lock();
2768 
2769 	/* Enable resets to be scheduled and check whether any were
2770 	 * already requested.  If so, the NIC is probably hosed so we
2771 	 * abort.
2772 	 */
2773 	efx->state = STATE_READY;
2774 	smp_mb(); /* ensure we change state before checking reset_pending */
2775 	if (efx->reset_pending) {
2776 		netif_err(efx, probe, efx->net_dev,
2777 			  "aborting probe due to scheduled reset\n");
2778 		rc = -EIO;
2779 		goto fail_locked;
2780 	}
2781 
2782 	rc = dev_alloc_name(net_dev, net_dev->name);
2783 	if (rc < 0)
2784 		goto fail_locked;
2785 	efx_update_name(efx);
2786 
2787 	/* Always start with carrier off; PHY events will detect the link */
2788 	netif_carrier_off(net_dev);
2789 
2790 	rc = register_netdevice(net_dev);
2791 	if (rc)
2792 		goto fail_locked;
2793 
2794 	efx_for_each_channel(channel, efx) {
2795 		struct efx_tx_queue *tx_queue;
2796 		efx_for_each_channel_tx_queue(tx_queue, channel)
2797 			efx_init_tx_queue_core_txq(tx_queue);
2798 	}
2799 
2800 	efx_associate(efx);
2801 
2802 	rtnl_unlock();
2803 
2804 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2805 	if (rc) {
2806 		netif_err(efx, drv, efx->net_dev,
2807 			  "failed to init net dev attributes\n");
2808 		goto fail_registered;
2809 	}
2810 #ifdef CONFIG_SFC_MCDI_LOGGING
2811 	rc = device_create_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2812 	if (rc) {
2813 		netif_err(efx, drv, efx->net_dev,
2814 			  "failed to init net dev attributes\n");
2815 		goto fail_attr_mcdi_logging;
2816 	}
2817 #endif
2818 
2819 	return 0;
2820 
2821 #ifdef CONFIG_SFC_MCDI_LOGGING
2822 fail_attr_mcdi_logging:
2823 	device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2824 #endif
2825 fail_registered:
2826 	rtnl_lock();
2827 	efx_dissociate(efx);
2828 	unregister_netdevice(net_dev);
2829 fail_locked:
2830 	efx->state = STATE_UNINIT;
2831 	rtnl_unlock();
2832 	netif_err(efx, drv, efx->net_dev, "could not register net dev\n");
2833 	return rc;
2834 }
2835 
2836 static void efx_unregister_netdev(struct efx_nic *efx)
2837 {
2838 	if (!efx->net_dev)
2839 		return;
2840 
2841 	BUG_ON(netdev_priv(efx->net_dev) != efx);
2842 
2843 	if (efx_dev_registered(efx)) {
2844 		strlcpy(efx->name, pci_name(efx->pci_dev), sizeof(efx->name));
2845 #ifdef CONFIG_SFC_MCDI_LOGGING
2846 		device_remove_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
2847 #endif
2848 		device_remove_file(&efx->pci_dev->dev, &dev_attr_phy_type);
2849 		unregister_netdev(efx->net_dev);
2850 	}
2851 }
2852 
2853 /**************************************************************************
2854  *
2855  * Device reset and suspend
2856  *
2857  **************************************************************************/
2858 
2859 /* Tears down the entire software state and most of the hardware state
2860  * before reset.  */
2861 void efx_reset_down(struct efx_nic *efx, enum reset_type method)
2862 {
2863 	EFX_ASSERT_RESET_SERIALISED(efx);
2864 
2865 	if (method == RESET_TYPE_MCDI_TIMEOUT)
2866 		efx->type->prepare_flr(efx);
2867 
2868 	efx_stop_all(efx);
2869 	efx_disable_interrupts(efx);
2870 
2871 	mutex_lock(&efx->mac_lock);
2872 	down_write(&efx->filter_sem);
2873 	mutex_lock(&efx->rss_lock);
2874 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2875 	    method != RESET_TYPE_DATAPATH)
2876 		efx->phy_op->fini(efx);
2877 	efx->type->fini(efx);
2878 }
2879 
2880 /* This function will always ensure that the locks acquired in
2881  * efx_reset_down() are released. A failure return code indicates
2882  * that we were unable to reinitialise the hardware, and the
2883  * driver should be disabled. If ok is false, then the rx and tx
2884  * engines are not restarted, pending a RESET_DISABLE. */
2885 int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
2886 {
2887 	int rc;
2888 
2889 	EFX_ASSERT_RESET_SERIALISED(efx);
2890 
2891 	if (method == RESET_TYPE_MCDI_TIMEOUT)
2892 		efx->type->finish_flr(efx);
2893 
2894 	/* Ensure that SRAM is initialised even if we're disabling the device */
2895 	rc = efx->type->init(efx);
2896 	if (rc) {
2897 		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
2898 		goto fail;
2899 	}
2900 
2901 	if (!ok)
2902 		goto fail;
2903 
2904 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
2905 	    method != RESET_TYPE_DATAPATH) {
2906 		rc = efx->phy_op->init(efx);
2907 		if (rc)
2908 			goto fail;
2909 		rc = efx->phy_op->reconfigure(efx);
2910 		if (rc && rc != -EPERM)
2911 			netif_err(efx, drv, efx->net_dev,
2912 				  "could not restore PHY settings\n");
2913 	}
2914 
2915 	rc = efx_enable_interrupts(efx);
2916 	if (rc)
2917 		goto fail;
2918 
2919 #ifdef CONFIG_SFC_SRIOV
2920 	rc = efx->type->vswitching_restore(efx);
2921 	if (rc) /* not fatal; the PF will still work fine */
2922 		netif_warn(efx, probe, efx->net_dev,
2923 			   "failed to restore vswitching rc=%d;"
2924 			   " VFs may not function\n", rc);
2925 #endif
2926 
2927 	if (efx->type->rx_restore_rss_contexts)
2928 		efx->type->rx_restore_rss_contexts(efx);
2929 	mutex_unlock(&efx->rss_lock);
2930 	efx->type->filter_table_restore(efx);
2931 	up_write(&efx->filter_sem);
2932 	if (efx->type->sriov_reset)
2933 		efx->type->sriov_reset(efx);
2934 
2935 	mutex_unlock(&efx->mac_lock);
2936 
2937 	efx_start_all(efx);
2938 
2939 	if (efx->type->udp_tnl_push_ports)
2940 		efx->type->udp_tnl_push_ports(efx);
2941 
2942 	return 0;
2943 
2944 fail:
2945 	efx->port_initialized = false;
2946 
2947 	mutex_unlock(&efx->rss_lock);
2948 	up_write(&efx->filter_sem);
2949 	mutex_unlock(&efx->mac_lock);
2950 
2951 	return rc;
2952 }
2953 
2954 /* Reset the NIC using the specified method.  Note that the reset may
2955  * fail, in which case the card will be left in an unusable state.
2956  *
2957  * Caller must hold the rtnl_lock.
2958  */
2959 int efx_reset(struct efx_nic *efx, enum reset_type method)
2960 {
2961 	int rc, rc2;
2962 	bool disabled;
2963 
2964 	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
2965 		   RESET_TYPE(method));
2966 
2967 	efx_device_detach_sync(efx);
2968 	efx_reset_down(efx, method);
2969 
2970 	rc = efx->type->reset(efx, method);
2971 	if (rc) {
2972 		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
2973 		goto out;
2974 	}
2975 
2976 	/* Clear flags for the scopes we covered.  We assume the NIC and
2977 	 * driver are now quiescent so that there is no race here.
2978 	 */
2979 	if (method < RESET_TYPE_MAX_METHOD)
2980 		efx->reset_pending &= -(1 << (method + 1));
2981 	else /* it doesn't fit into the well-ordered scope hierarchy */
2982 		__clear_bit(method, &efx->reset_pending);
2983 
2984 	/* Reinitialise bus-mastering, which may have been turned off before
2985 	 * the reset was scheduled. This is still appropriate, even in the
2986 	 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
2987 	 * can respond to requests. */
2988 	pci_set_master(efx->pci_dev);
2989 
2990 out:
2991 	/* Leave device stopped if necessary */
2992 	disabled = rc ||
2993 		method == RESET_TYPE_DISABLE ||
2994 		method == RESET_TYPE_RECOVER_OR_DISABLE;
2995 	rc2 = efx_reset_up(efx, method, !disabled);
2996 	if (rc2) {
2997 		disabled = true;
2998 		if (!rc)
2999 			rc = rc2;
3000 	}
3001 
3002 	if (disabled) {
3003 		dev_close(efx->net_dev);
3004 		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
3005 		efx->state = STATE_DISABLED;
3006 	} else {
3007 		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
3008 		efx_device_attach_if_not_resetting(efx);
3009 	}
3010 	return rc;
3011 }
3012 
3013 /* Try recovery mechanisms.
3014  * For now only EEH is supported.
3015  * Returns 0 if the recovery mechanisms are unsuccessful.
3016  * Returns a non-zero value otherwise.
3017  */
3018 int efx_try_recovery(struct efx_nic *efx)
3019 {
3020 #ifdef CONFIG_EEH
3021 	/* A PCI error can occur and not be seen by EEH because nothing
3022 	 * happens on the PCI bus. In this case the driver may fail and
3023 	 * schedule a 'recover or reset', leading to this recovery handler.
3024 	 * Manually call the eeh failure check function.
3025 	 */
3026 	struct eeh_dev *eehdev = pci_dev_to_eeh_dev(efx->pci_dev);
3027 	if (eeh_dev_check_failure(eehdev)) {
3028 		/* The EEH mechanisms will handle the error and reset the
3029 		 * device if necessary.
3030 		 */
3031 		return 1;
3032 	}
3033 #endif
3034 	return 0;
3035 }
3036 
3037 static void efx_wait_for_bist_end(struct efx_nic *efx)
3038 {
3039 	int i;
3040 
3041 	for (i = 0; i < BIST_WAIT_DELAY_COUNT; ++i) {
3042 		if (efx_mcdi_poll_reboot(efx))
3043 			goto out;
3044 		msleep(BIST_WAIT_DELAY_MS);
3045 	}
3046 
3047 	netif_err(efx, drv, efx->net_dev, "Warning: No MC reboot after BIST mode\n");
3048 out:
3049 	/* Either way unset the BIST flag. If we found no reboot we probably
3050 	 * won't recover, but we should try.
3051 	 */
3052 	efx->mc_bist_for_other_fn = false;
3053 }
3054 
3055 /* The worker thread exists so that code that cannot sleep can
3056  * schedule a reset for later.
3057  */
3058 static void efx_reset_work(struct work_struct *data)
3059 {
3060 	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
3061 	unsigned long pending;
3062 	enum reset_type method;
3063 
3064 	pending = READ_ONCE(efx->reset_pending);
3065 	method = fls(pending) - 1;
3066 
3067 	if (method == RESET_TYPE_MC_BIST)
3068 		efx_wait_for_bist_end(efx);
3069 
3070 	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
3071 	     method == RESET_TYPE_RECOVER_OR_ALL) &&
3072 	    efx_try_recovery(efx))
3073 		return;
3074 
3075 	if (!pending)
3076 		return;
3077 
3078 	rtnl_lock();
3079 
3080 	/* We checked the state in efx_schedule_reset() but it may
3081 	 * have changed by now.  Now that we have the RTNL lock,
3082 	 * it cannot change again.
3083 	 */
3084 	if (efx->state == STATE_READY)
3085 		(void)efx_reset(efx, method);
3086 
3087 	rtnl_unlock();
3088 }
3089 
3090 void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
3091 {
3092 	enum reset_type method;
3093 
3094 	if (efx->state == STATE_RECOVERY) {
3095 		netif_dbg(efx, drv, efx->net_dev,
3096 			  "recovering: skip scheduling %s reset\n",
3097 			  RESET_TYPE(type));
3098 		return;
3099 	}
3100 
3101 	switch (type) {
3102 	case RESET_TYPE_INVISIBLE:
3103 	case RESET_TYPE_ALL:
3104 	case RESET_TYPE_RECOVER_OR_ALL:
3105 	case RESET_TYPE_WORLD:
3106 	case RESET_TYPE_DISABLE:
3107 	case RESET_TYPE_RECOVER_OR_DISABLE:
3108 	case RESET_TYPE_DATAPATH:
3109 	case RESET_TYPE_MC_BIST:
3110 	case RESET_TYPE_MCDI_TIMEOUT:
3111 		method = type;
3112 		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
3113 			  RESET_TYPE(method));
3114 		break;
3115 	default:
3116 		method = efx->type->map_reset_reason(type);
3117 		netif_dbg(efx, drv, efx->net_dev,
3118 			  "scheduling %s reset for %s\n",
3119 			  RESET_TYPE(method), RESET_TYPE(type));
3120 		break;
3121 	}
3122 
3123 	set_bit(method, &efx->reset_pending);
3124 	smp_mb(); /* ensure we change reset_pending before checking state */
3125 
3126 	/* If we're not READY then just leave the flags set as the cue
3127 	 * to abort probing or reschedule the reset later.
3128 	 */
3129 	if (READ_ONCE(efx->state) != STATE_READY)
3130 		return;
3131 
3132 	/* efx_process_channel() will no longer read events once a
3133 	 * reset is scheduled. So switch back to poll'd MCDI completions. */
3134 	efx_mcdi_mode_poll(efx);
3135 
3136 	queue_work(reset_workqueue, &efx->reset_work);
3137 }
3138 
3139 /**************************************************************************
3140  *
3141  * List of NICs we support
3142  *
3143  **************************************************************************/
3144 
3145 /* PCI device ID table */
3146 static const struct pci_device_id efx_pci_table[] = {
3147 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0803),	/* SFC9020 */
3148 	 .driver_data = (unsigned long) &siena_a0_nic_type},
3149 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0813),	/* SFL9021 */
3150 	 .driver_data = (unsigned long) &siena_a0_nic_type},
3151 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0903),  /* SFC9120 PF */
3152 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
3153 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1903),  /* SFC9120 VF */
3154 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
3155 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0923),  /* SFC9140 PF */
3156 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
3157 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1923),  /* SFC9140 VF */
3158 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
3159 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0a03),  /* SFC9220 PF */
3160 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
3161 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1a03),  /* SFC9220 VF */
3162 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
3163 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x0b03),  /* SFC9250 PF */
3164 	 .driver_data = (unsigned long) &efx_hunt_a0_nic_type},
3165 	{PCI_DEVICE(PCI_VENDOR_ID_SOLARFLARE, 0x1b03),  /* SFC9250 VF */
3166 	 .driver_data = (unsigned long) &efx_hunt_a0_vf_nic_type},
3167 	{0}			/* end of list */
3168 };
3169 
3170 /**************************************************************************
3171  *
3172  * Dummy PHY/MAC operations
3173  *
3174  * Can be used for some unimplemented operations
3175  * Needed so all function pointers are valid and do not have to be tested
3176  * before use
3177  *
3178  **************************************************************************/
3179 int efx_port_dummy_op_int(struct efx_nic *efx)
3180 {
3181 	return 0;
3182 }
3183 void efx_port_dummy_op_void(struct efx_nic *efx) {}
3184 
3185 static bool efx_port_dummy_op_poll(struct efx_nic *efx)
3186 {
3187 	return false;
3188 }
3189 
3190 static const struct efx_phy_operations efx_dummy_phy_operations = {
3191 	.init		 = efx_port_dummy_op_int,
3192 	.reconfigure	 = efx_port_dummy_op_int,
3193 	.poll		 = efx_port_dummy_op_poll,
3194 	.fini		 = efx_port_dummy_op_void,
3195 };
3196 
3197 /**************************************************************************
3198  *
3199  * Data housekeeping
3200  *
3201  **************************************************************************/
3202 
3203 /* This zeroes out and then fills in the invariants in a struct
3204  * efx_nic (including all sub-structures).
3205  */
3206 static int efx_init_struct(struct efx_nic *efx,
3207 			   struct pci_dev *pci_dev, struct net_device *net_dev)
3208 {
3209 	int rc = -ENOMEM, i;
3210 
3211 	/* Initialise common structures */
3212 	INIT_LIST_HEAD(&efx->node);
3213 	INIT_LIST_HEAD(&efx->secondary_list);
3214 	spin_lock_init(&efx->biu_lock);
3215 #ifdef CONFIG_SFC_MTD
3216 	INIT_LIST_HEAD(&efx->mtd_list);
3217 #endif
3218 	INIT_WORK(&efx->reset_work, efx_reset_work);
3219 	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
3220 	INIT_DELAYED_WORK(&efx->selftest_work, efx_selftest_async_work);
3221 	efx->pci_dev = pci_dev;
3222 	efx->msg_enable = debug;
3223 	efx->state = STATE_UNINIT;
3224 	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
3225 
3226 	efx->net_dev = net_dev;
3227 	efx->rx_prefix_size = efx->type->rx_prefix_size;
3228 	efx->rx_ip_align =
3229 		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
3230 	efx->rx_packet_hash_offset =
3231 		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
3232 	efx->rx_packet_ts_offset =
3233 		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
3234 	INIT_LIST_HEAD(&efx->rss_context.list);
3235 	mutex_init(&efx->rss_lock);
3236 	spin_lock_init(&efx->stats_lock);
3237 	efx->vi_stride = EFX_DEFAULT_VI_STRIDE;
3238 	efx->num_mac_stats = MC_CMD_MAC_NSTATS;
3239 	BUILD_BUG_ON(MC_CMD_MAC_NSTATS - 1 != MC_CMD_MAC_GENERATION_END);
3240 	mutex_init(&efx->mac_lock);
3241 #ifdef CONFIG_RFS_ACCEL
3242 	mutex_init(&efx->rps_mutex);
3243 	spin_lock_init(&efx->rps_hash_lock);
3244 	/* Failure to allocate is not fatal, but may degrade ARFS performance */
3245 	efx->rps_hash_table = kcalloc(EFX_ARFS_HASH_TABLE_SIZE,
3246 				      sizeof(*efx->rps_hash_table), GFP_KERNEL);
3247 #endif
3248 	efx->phy_op = &efx_dummy_phy_operations;
3249 	efx->mdio.dev = net_dev;
3250 	INIT_WORK(&efx->mac_work, efx_mac_work);
3251 	init_waitqueue_head(&efx->flush_wq);
3252 
3253 	for (i = 0; i < EFX_MAX_CHANNELS; i++) {
3254 		efx->channel[i] = efx_alloc_channel(efx, i, NULL);
3255 		if (!efx->channel[i])
3256 			goto fail;
3257 		efx->msi_context[i].efx = efx;
3258 		efx->msi_context[i].index = i;
3259 	}
3260 
3261 	/* Higher numbered interrupt modes are less capable! */
3262 	if (WARN_ON_ONCE(efx->type->max_interrupt_mode >
3263 			 efx->type->min_interrupt_mode)) {
3264 		rc = -EIO;
3265 		goto fail;
3266 	}
3267 	efx->interrupt_mode = max(efx->type->max_interrupt_mode,
3268 				  interrupt_mode);
3269 	efx->interrupt_mode = min(efx->type->min_interrupt_mode,
3270 				  interrupt_mode);
3271 
3272 	/* Would be good to use the net_dev name, but we're too early */
3273 	snprintf(efx->workqueue_name, sizeof(efx->workqueue_name), "sfc%s",
3274 		 pci_name(pci_dev));
3275 	efx->workqueue = create_singlethread_workqueue(efx->workqueue_name);
3276 	if (!efx->workqueue)
3277 		goto fail;
3278 
3279 	return 0;
3280 
3281 fail:
3282 	efx_fini_struct(efx);
3283 	return rc;
3284 }
3285 
3286 static void efx_fini_struct(struct efx_nic *efx)
3287 {
3288 	int i;
3289 
3290 #ifdef CONFIG_RFS_ACCEL
3291 	kfree(efx->rps_hash_table);
3292 #endif
3293 
3294 	for (i = 0; i < EFX_MAX_CHANNELS; i++)
3295 		kfree(efx->channel[i]);
3296 
3297 	kfree(efx->vpd_sn);
3298 
3299 	if (efx->workqueue) {
3300 		destroy_workqueue(efx->workqueue);
3301 		efx->workqueue = NULL;
3302 	}
3303 }
3304 
3305 void efx_update_sw_stats(struct efx_nic *efx, u64 *stats)
3306 {
3307 	u64 n_rx_nodesc_trunc = 0;
3308 	struct efx_channel *channel;
3309 
3310 	efx_for_each_channel(channel, efx)
3311 		n_rx_nodesc_trunc += channel->n_rx_nodesc_trunc;
3312 	stats[GENERIC_STAT_rx_nodesc_trunc] = n_rx_nodesc_trunc;
3313 	stats[GENERIC_STAT_rx_noskb_drops] = atomic_read(&efx->n_rx_noskb_drops);
3314 }
3315 
3316 bool efx_filter_spec_equal(const struct efx_filter_spec *left,
3317 			   const struct efx_filter_spec *right)
3318 {
3319 	if ((left->match_flags ^ right->match_flags) |
3320 	    ((left->flags ^ right->flags) &
3321 	     (EFX_FILTER_FLAG_RX | EFX_FILTER_FLAG_TX)))
3322 		return false;
3323 
3324 	return memcmp(&left->outer_vid, &right->outer_vid,
3325 		      sizeof(struct efx_filter_spec) -
3326 		      offsetof(struct efx_filter_spec, outer_vid)) == 0;
3327 }
3328 
3329 u32 efx_filter_spec_hash(const struct efx_filter_spec *spec)
3330 {
3331 	BUILD_BUG_ON(offsetof(struct efx_filter_spec, outer_vid) & 3);
3332 	return jhash2((const u32 *)&spec->outer_vid,
3333 		      (sizeof(struct efx_filter_spec) -
3334 		       offsetof(struct efx_filter_spec, outer_vid)) / 4,
3335 		      0);
3336 }
3337 
3338 #ifdef CONFIG_RFS_ACCEL
3339 bool efx_rps_check_rule(struct efx_arfs_rule *rule, unsigned int filter_idx,
3340 			bool *force)
3341 {
3342 	if (rule->filter_id == EFX_ARFS_FILTER_ID_PENDING) {
3343 		/* ARFS is currently updating this entry, leave it */
3344 		return false;
3345 	}
3346 	if (rule->filter_id == EFX_ARFS_FILTER_ID_ERROR) {
3347 		/* ARFS tried and failed to update this, so it's probably out
3348 		 * of date.  Remove the filter and the ARFS rule entry.
3349 		 */
3350 		rule->filter_id = EFX_ARFS_FILTER_ID_REMOVING;
3351 		*force = true;
3352 		return true;
3353 	} else if (WARN_ON(rule->filter_id != filter_idx)) { /* can't happen */
3354 		/* ARFS has moved on, so old filter is not needed.  Since we did
3355 		 * not mark the rule with EFX_ARFS_FILTER_ID_REMOVING, it will
3356 		 * not be removed by efx_rps_hash_del() subsequently.
3357 		 */
3358 		*force = true;
3359 		return true;
3360 	}
3361 	/* Remove it iff ARFS wants to. */
3362 	return true;
3363 }
3364 
3365 static
3366 struct hlist_head *efx_rps_hash_bucket(struct efx_nic *efx,
3367 				       const struct efx_filter_spec *spec)
3368 {
3369 	u32 hash = efx_filter_spec_hash(spec);
3370 
3371 	lockdep_assert_held(&efx->rps_hash_lock);
3372 	if (!efx->rps_hash_table)
3373 		return NULL;
3374 	return &efx->rps_hash_table[hash % EFX_ARFS_HASH_TABLE_SIZE];
3375 }
3376 
3377 struct efx_arfs_rule *efx_rps_hash_find(struct efx_nic *efx,
3378 					const struct efx_filter_spec *spec)
3379 {
3380 	struct efx_arfs_rule *rule;
3381 	struct hlist_head *head;
3382 	struct hlist_node *node;
3383 
3384 	head = efx_rps_hash_bucket(efx, spec);
3385 	if (!head)
3386 		return NULL;
3387 	hlist_for_each(node, head) {
3388 		rule = container_of(node, struct efx_arfs_rule, node);
3389 		if (efx_filter_spec_equal(spec, &rule->spec))
3390 			return rule;
3391 	}
3392 	return NULL;
3393 }
3394 
3395 struct efx_arfs_rule *efx_rps_hash_add(struct efx_nic *efx,
3396 				       const struct efx_filter_spec *spec,
3397 				       bool *new)
3398 {
3399 	struct efx_arfs_rule *rule;
3400 	struct hlist_head *head;
3401 	struct hlist_node *node;
3402 
3403 	head = efx_rps_hash_bucket(efx, spec);
3404 	if (!head)
3405 		return NULL;
3406 	hlist_for_each(node, head) {
3407 		rule = container_of(node, struct efx_arfs_rule, node);
3408 		if (efx_filter_spec_equal(spec, &rule->spec)) {
3409 			*new = false;
3410 			return rule;
3411 		}
3412 	}
3413 	rule = kmalloc(sizeof(*rule), GFP_ATOMIC);
3414 	*new = true;
3415 	if (rule) {
3416 		memcpy(&rule->spec, spec, sizeof(rule->spec));
3417 		hlist_add_head(&rule->node, head);
3418 	}
3419 	return rule;
3420 }
3421 
3422 void efx_rps_hash_del(struct efx_nic *efx, const struct efx_filter_spec *spec)
3423 {
3424 	struct efx_arfs_rule *rule;
3425 	struct hlist_head *head;
3426 	struct hlist_node *node;
3427 
3428 	head = efx_rps_hash_bucket(efx, spec);
3429 	if (WARN_ON(!head))
3430 		return;
3431 	hlist_for_each(node, head) {
3432 		rule = container_of(node, struct efx_arfs_rule, node);
3433 		if (efx_filter_spec_equal(spec, &rule->spec)) {
3434 			/* Someone already reused the entry.  We know that if
3435 			 * this check doesn't fire (i.e. filter_id == REMOVING)
3436 			 * then the REMOVING mark was put there by our caller,
3437 			 * because caller is holding a lock on filter table and
3438 			 * only holders of that lock set REMOVING.
3439 			 */
3440 			if (rule->filter_id != EFX_ARFS_FILTER_ID_REMOVING)
3441 				return;
3442 			hlist_del(node);
3443 			kfree(rule);
3444 			return;
3445 		}
3446 	}
3447 	/* We didn't find it. */
3448 	WARN_ON(1);
3449 }
3450 #endif
3451 
3452 /* RSS contexts.  We're using linked lists and crappy O(n) algorithms, because
3453  * (a) this is an infrequent control-plane operation and (b) n is small (max 64)
3454  */
3455 struct efx_rss_context *efx_alloc_rss_context_entry(struct efx_nic *efx)
3456 {
3457 	struct list_head *head = &efx->rss_context.list;
3458 	struct efx_rss_context *ctx, *new;
3459 	u32 id = 1; /* Don't use zero, that refers to the master RSS context */
3460 
3461 	WARN_ON(!mutex_is_locked(&efx->rss_lock));
3462 
3463 	/* Search for first gap in the numbering */
3464 	list_for_each_entry(ctx, head, list) {
3465 		if (ctx->user_id != id)
3466 			break;
3467 		id++;
3468 		/* Check for wrap.  If this happens, we have nearly 2^32
3469 		 * allocated RSS contexts, which seems unlikely.
3470 		 */
3471 		if (WARN_ON_ONCE(!id))
3472 			return NULL;
3473 	}
3474 
3475 	/* Create the new entry */
3476 	new = kmalloc(sizeof(struct efx_rss_context), GFP_KERNEL);
3477 	if (!new)
3478 		return NULL;
3479 	new->context_id = EFX_EF10_RSS_CONTEXT_INVALID;
3480 	new->rx_hash_udp_4tuple = false;
3481 
3482 	/* Insert the new entry into the gap */
3483 	new->user_id = id;
3484 	list_add_tail(&new->list, &ctx->list);
3485 	return new;
3486 }
3487 
3488 struct efx_rss_context *efx_find_rss_context_entry(struct efx_nic *efx, u32 id)
3489 {
3490 	struct list_head *head = &efx->rss_context.list;
3491 	struct efx_rss_context *ctx;
3492 
3493 	WARN_ON(!mutex_is_locked(&efx->rss_lock));
3494 
3495 	list_for_each_entry(ctx, head, list)
3496 		if (ctx->user_id == id)
3497 			return ctx;
3498 	return NULL;
3499 }
3500 
3501 void efx_free_rss_context_entry(struct efx_rss_context *ctx)
3502 {
3503 	list_del(&ctx->list);
3504 	kfree(ctx);
3505 }
3506 
3507 /**************************************************************************
3508  *
3509  * PCI interface
3510  *
3511  **************************************************************************/
3512 
3513 /* Main body of final NIC shutdown code
3514  * This is called only at module unload (or hotplug removal).
3515  */
3516 static void efx_pci_remove_main(struct efx_nic *efx)
3517 {
3518 	/* Flush reset_work. It can no longer be scheduled since we
3519 	 * are not READY.
3520 	 */
3521 	BUG_ON(efx->state == STATE_READY);
3522 	cancel_work_sync(&efx->reset_work);
3523 
3524 	efx_disable_interrupts(efx);
3525 	efx_clear_interrupt_affinity(efx);
3526 	efx_nic_fini_interrupt(efx);
3527 	efx_fini_port(efx);
3528 	efx->type->fini(efx);
3529 	efx_fini_napi(efx);
3530 	efx_remove_all(efx);
3531 }
3532 
3533 /* Final NIC shutdown
3534  * This is called only at module unload (or hotplug removal).  A PF can call
3535  * this on its VFs to ensure they are unbound first.
3536  */
3537 static void efx_pci_remove(struct pci_dev *pci_dev)
3538 {
3539 	struct efx_nic *efx;
3540 
3541 	efx = pci_get_drvdata(pci_dev);
3542 	if (!efx)
3543 		return;
3544 
3545 	/* Mark the NIC as fini, then stop the interface */
3546 	rtnl_lock();
3547 	efx_dissociate(efx);
3548 	dev_close(efx->net_dev);
3549 	efx_disable_interrupts(efx);
3550 	efx->state = STATE_UNINIT;
3551 	rtnl_unlock();
3552 
3553 	if (efx->type->sriov_fini)
3554 		efx->type->sriov_fini(efx);
3555 
3556 	efx_unregister_netdev(efx);
3557 
3558 	efx_mtd_remove(efx);
3559 
3560 	efx_pci_remove_main(efx);
3561 
3562 	efx_fini_io(efx);
3563 	netif_dbg(efx, drv, efx->net_dev, "shutdown successful\n");
3564 
3565 	efx_fini_struct(efx);
3566 	free_netdev(efx->net_dev);
3567 
3568 	pci_disable_pcie_error_reporting(pci_dev);
3569 };
3570 
3571 /* NIC VPD information
3572  * Called during probe to display the part number of the
3573  * installed NIC.  VPD is potentially very large but this should
3574  * always appear within the first 512 bytes.
3575  */
3576 #define SFC_VPD_LEN 512
3577 static void efx_probe_vpd_strings(struct efx_nic *efx)
3578 {
3579 	struct pci_dev *dev = efx->pci_dev;
3580 	char vpd_data[SFC_VPD_LEN];
3581 	ssize_t vpd_size;
3582 	int ro_start, ro_size, i, j;
3583 
3584 	/* Get the vpd data from the device */
3585 	vpd_size = pci_read_vpd(dev, 0, sizeof(vpd_data), vpd_data);
3586 	if (vpd_size <= 0) {
3587 		netif_err(efx, drv, efx->net_dev, "Unable to read VPD\n");
3588 		return;
3589 	}
3590 
3591 	/* Get the Read only section */
3592 	ro_start = pci_vpd_find_tag(vpd_data, 0, vpd_size, PCI_VPD_LRDT_RO_DATA);
3593 	if (ro_start < 0) {
3594 		netif_err(efx, drv, efx->net_dev, "VPD Read-only not found\n");
3595 		return;
3596 	}
3597 
3598 	ro_size = pci_vpd_lrdt_size(&vpd_data[ro_start]);
3599 	j = ro_size;
3600 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3601 	if (i + j > vpd_size)
3602 		j = vpd_size - i;
3603 
3604 	/* Get the Part number */
3605 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "PN");
3606 	if (i < 0) {
3607 		netif_err(efx, drv, efx->net_dev, "Part number not found\n");
3608 		return;
3609 	}
3610 
3611 	j = pci_vpd_info_field_size(&vpd_data[i]);
3612 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3613 	if (i + j > vpd_size) {
3614 		netif_err(efx, drv, efx->net_dev, "Incomplete part number\n");
3615 		return;
3616 	}
3617 
3618 	netif_info(efx, drv, efx->net_dev,
3619 		   "Part Number : %.*s\n", j, &vpd_data[i]);
3620 
3621 	i = ro_start + PCI_VPD_LRDT_TAG_SIZE;
3622 	j = ro_size;
3623 	i = pci_vpd_find_info_keyword(vpd_data, i, j, "SN");
3624 	if (i < 0) {
3625 		netif_err(efx, drv, efx->net_dev, "Serial number not found\n");
3626 		return;
3627 	}
3628 
3629 	j = pci_vpd_info_field_size(&vpd_data[i]);
3630 	i += PCI_VPD_INFO_FLD_HDR_SIZE;
3631 	if (i + j > vpd_size) {
3632 		netif_err(efx, drv, efx->net_dev, "Incomplete serial number\n");
3633 		return;
3634 	}
3635 
3636 	efx->vpd_sn = kmalloc(j + 1, GFP_KERNEL);
3637 	if (!efx->vpd_sn)
3638 		return;
3639 
3640 	snprintf(efx->vpd_sn, j + 1, "%s", &vpd_data[i]);
3641 }
3642 
3643 
3644 /* Main body of NIC initialisation
3645  * This is called at module load (or hotplug insertion, theoretically).
3646  */
3647 static int efx_pci_probe_main(struct efx_nic *efx)
3648 {
3649 	int rc;
3650 
3651 	/* Do start-of-day initialisation */
3652 	rc = efx_probe_all(efx);
3653 	if (rc)
3654 		goto fail1;
3655 
3656 	efx_init_napi(efx);
3657 
3658 	down_write(&efx->filter_sem);
3659 	rc = efx->type->init(efx);
3660 	up_write(&efx->filter_sem);
3661 	if (rc) {
3662 		netif_err(efx, probe, efx->net_dev,
3663 			  "failed to initialise NIC\n");
3664 		goto fail3;
3665 	}
3666 
3667 	rc = efx_init_port(efx);
3668 	if (rc) {
3669 		netif_err(efx, probe, efx->net_dev,
3670 			  "failed to initialise port\n");
3671 		goto fail4;
3672 	}
3673 
3674 	rc = efx_nic_init_interrupt(efx);
3675 	if (rc)
3676 		goto fail5;
3677 
3678 	efx_set_interrupt_affinity(efx);
3679 	rc = efx_enable_interrupts(efx);
3680 	if (rc)
3681 		goto fail6;
3682 
3683 	return 0;
3684 
3685  fail6:
3686 	efx_clear_interrupt_affinity(efx);
3687 	efx_nic_fini_interrupt(efx);
3688  fail5:
3689 	efx_fini_port(efx);
3690  fail4:
3691 	efx->type->fini(efx);
3692  fail3:
3693 	efx_fini_napi(efx);
3694 	efx_remove_all(efx);
3695  fail1:
3696 	return rc;
3697 }
3698 
3699 static int efx_pci_probe_post_io(struct efx_nic *efx)
3700 {
3701 	struct net_device *net_dev = efx->net_dev;
3702 	int rc = efx_pci_probe_main(efx);
3703 
3704 	if (rc)
3705 		return rc;
3706 
3707 	if (efx->type->sriov_init) {
3708 		rc = efx->type->sriov_init(efx);
3709 		if (rc)
3710 			netif_err(efx, probe, efx->net_dev,
3711 				  "SR-IOV can't be enabled rc %d\n", rc);
3712 	}
3713 
3714 	/* Determine netdevice features */
3715 	net_dev->features |= (efx->type->offload_features | NETIF_F_SG |
3716 			      NETIF_F_TSO | NETIF_F_RXCSUM | NETIF_F_RXALL);
3717 	if (efx->type->offload_features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
3718 		net_dev->features |= NETIF_F_TSO6;
3719 	/* Check whether device supports TSO */
3720 	if (!efx->type->tso_versions || !efx->type->tso_versions(efx))
3721 		net_dev->features &= ~NETIF_F_ALL_TSO;
3722 	/* Mask for features that also apply to VLAN devices */
3723 	net_dev->vlan_features |= (NETIF_F_HW_CSUM | NETIF_F_SG |
3724 				   NETIF_F_HIGHDMA | NETIF_F_ALL_TSO |
3725 				   NETIF_F_RXCSUM);
3726 
3727 	net_dev->hw_features |= net_dev->features & ~efx->fixed_features;
3728 
3729 	/* Disable receiving frames with bad FCS, by default. */
3730 	net_dev->features &= ~NETIF_F_RXALL;
3731 
3732 	/* Disable VLAN filtering by default.  It may be enforced if
3733 	 * the feature is fixed (i.e. VLAN filters are required to
3734 	 * receive VLAN tagged packets due to vPort restrictions).
3735 	 */
3736 	net_dev->features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
3737 	net_dev->features |= efx->fixed_features;
3738 
3739 	rc = efx_register_netdev(efx);
3740 	if (!rc)
3741 		return 0;
3742 
3743 	efx_pci_remove_main(efx);
3744 	return rc;
3745 }
3746 
3747 /* NIC initialisation
3748  *
3749  * This is called at module load (or hotplug insertion,
3750  * theoretically).  It sets up PCI mappings, resets the NIC,
3751  * sets up and registers the network devices with the kernel and hooks
3752  * the interrupt service routine.  It does not prepare the device for
3753  * transmission; this is left to the first time one of the network
3754  * interfaces is brought up (i.e. efx_net_open).
3755  */
3756 static int efx_pci_probe(struct pci_dev *pci_dev,
3757 			 const struct pci_device_id *entry)
3758 {
3759 	struct net_device *net_dev;
3760 	struct efx_nic *efx;
3761 	int rc;
3762 
3763 	/* Allocate and initialise a struct net_device and struct efx_nic */
3764 	net_dev = alloc_etherdev_mqs(sizeof(*efx), EFX_MAX_CORE_TX_QUEUES,
3765 				     EFX_MAX_RX_QUEUES);
3766 	if (!net_dev)
3767 		return -ENOMEM;
3768 	efx = netdev_priv(net_dev);
3769 	efx->type = (const struct efx_nic_type *) entry->driver_data;
3770 	efx->fixed_features |= NETIF_F_HIGHDMA;
3771 
3772 	pci_set_drvdata(pci_dev, efx);
3773 	SET_NETDEV_DEV(net_dev, &pci_dev->dev);
3774 	rc = efx_init_struct(efx, pci_dev, net_dev);
3775 	if (rc)
3776 		goto fail1;
3777 
3778 	netif_info(efx, probe, efx->net_dev,
3779 		   "Solarflare NIC detected\n");
3780 
3781 	if (!efx->type->is_vf)
3782 		efx_probe_vpd_strings(efx);
3783 
3784 	/* Set up basic I/O (BAR mappings etc) */
3785 	rc = efx_init_io(efx);
3786 	if (rc)
3787 		goto fail2;
3788 
3789 	rc = efx_pci_probe_post_io(efx);
3790 	if (rc) {
3791 		/* On failure, retry once immediately.
3792 		 * If we aborted probe due to a scheduled reset, dismiss it.
3793 		 */
3794 		efx->reset_pending = 0;
3795 		rc = efx_pci_probe_post_io(efx);
3796 		if (rc) {
3797 			/* On another failure, retry once more
3798 			 * after a 50-305ms delay.
3799 			 */
3800 			unsigned char r;
3801 
3802 			get_random_bytes(&r, 1);
3803 			msleep((unsigned int)r + 50);
3804 			efx->reset_pending = 0;
3805 			rc = efx_pci_probe_post_io(efx);
3806 		}
3807 	}
3808 	if (rc)
3809 		goto fail3;
3810 
3811 	netif_dbg(efx, probe, efx->net_dev, "initialisation successful\n");
3812 
3813 	/* Try to create MTDs, but allow this to fail */
3814 	rtnl_lock();
3815 	rc = efx_mtd_probe(efx);
3816 	rtnl_unlock();
3817 	if (rc && rc != -EPERM)
3818 		netif_warn(efx, probe, efx->net_dev,
3819 			   "failed to create MTDs (%d)\n", rc);
3820 
3821 	(void)pci_enable_pcie_error_reporting(pci_dev);
3822 
3823 	if (efx->type->udp_tnl_push_ports)
3824 		efx->type->udp_tnl_push_ports(efx);
3825 
3826 	return 0;
3827 
3828  fail3:
3829 	efx_fini_io(efx);
3830  fail2:
3831 	efx_fini_struct(efx);
3832  fail1:
3833 	WARN_ON(rc > 0);
3834 	netif_dbg(efx, drv, efx->net_dev, "initialisation failed. rc=%d\n", rc);
3835 	free_netdev(net_dev);
3836 	return rc;
3837 }
3838 
3839 /* efx_pci_sriov_configure returns the actual number of Virtual Functions
3840  * enabled on success
3841  */
3842 #ifdef CONFIG_SFC_SRIOV
3843 static int efx_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
3844 {
3845 	int rc;
3846 	struct efx_nic *efx = pci_get_drvdata(dev);
3847 
3848 	if (efx->type->sriov_configure) {
3849 		rc = efx->type->sriov_configure(efx, num_vfs);
3850 		if (rc)
3851 			return rc;
3852 		else
3853 			return num_vfs;
3854 	} else
3855 		return -EOPNOTSUPP;
3856 }
3857 #endif
3858 
3859 static int efx_pm_freeze(struct device *dev)
3860 {
3861 	struct efx_nic *efx = dev_get_drvdata(dev);
3862 
3863 	rtnl_lock();
3864 
3865 	if (efx->state != STATE_DISABLED) {
3866 		efx->state = STATE_UNINIT;
3867 
3868 		efx_device_detach_sync(efx);
3869 
3870 		efx_stop_all(efx);
3871 		efx_disable_interrupts(efx);
3872 	}
3873 
3874 	rtnl_unlock();
3875 
3876 	return 0;
3877 }
3878 
3879 static int efx_pm_thaw(struct device *dev)
3880 {
3881 	int rc;
3882 	struct efx_nic *efx = dev_get_drvdata(dev);
3883 
3884 	rtnl_lock();
3885 
3886 	if (efx->state != STATE_DISABLED) {
3887 		rc = efx_enable_interrupts(efx);
3888 		if (rc)
3889 			goto fail;
3890 
3891 		mutex_lock(&efx->mac_lock);
3892 		efx->phy_op->reconfigure(efx);
3893 		mutex_unlock(&efx->mac_lock);
3894 
3895 		efx_start_all(efx);
3896 
3897 		efx_device_attach_if_not_resetting(efx);
3898 
3899 		efx->state = STATE_READY;
3900 
3901 		efx->type->resume_wol(efx);
3902 	}
3903 
3904 	rtnl_unlock();
3905 
3906 	/* Reschedule any quenched resets scheduled during efx_pm_freeze() */
3907 	queue_work(reset_workqueue, &efx->reset_work);
3908 
3909 	return 0;
3910 
3911 fail:
3912 	rtnl_unlock();
3913 
3914 	return rc;
3915 }
3916 
3917 static int efx_pm_poweroff(struct device *dev)
3918 {
3919 	struct pci_dev *pci_dev = to_pci_dev(dev);
3920 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3921 
3922 	efx->type->fini(efx);
3923 
3924 	efx->reset_pending = 0;
3925 
3926 	pci_save_state(pci_dev);
3927 	return pci_set_power_state(pci_dev, PCI_D3hot);
3928 }
3929 
3930 /* Used for both resume and restore */
3931 static int efx_pm_resume(struct device *dev)
3932 {
3933 	struct pci_dev *pci_dev = to_pci_dev(dev);
3934 	struct efx_nic *efx = pci_get_drvdata(pci_dev);
3935 	int rc;
3936 
3937 	rc = pci_set_power_state(pci_dev, PCI_D0);
3938 	if (rc)
3939 		return rc;
3940 	pci_restore_state(pci_dev);
3941 	rc = pci_enable_device(pci_dev);
3942 	if (rc)
3943 		return rc;
3944 	pci_set_master(efx->pci_dev);
3945 	rc = efx->type->reset(efx, RESET_TYPE_ALL);
3946 	if (rc)
3947 		return rc;
3948 	down_write(&efx->filter_sem);
3949 	rc = efx->type->init(efx);
3950 	up_write(&efx->filter_sem);
3951 	if (rc)
3952 		return rc;
3953 	rc = efx_pm_thaw(dev);
3954 	return rc;
3955 }
3956 
3957 static int efx_pm_suspend(struct device *dev)
3958 {
3959 	int rc;
3960 
3961 	efx_pm_freeze(dev);
3962 	rc = efx_pm_poweroff(dev);
3963 	if (rc)
3964 		efx_pm_resume(dev);
3965 	return rc;
3966 }
3967 
3968 static const struct dev_pm_ops efx_pm_ops = {
3969 	.suspend	= efx_pm_suspend,
3970 	.resume		= efx_pm_resume,
3971 	.freeze		= efx_pm_freeze,
3972 	.thaw		= efx_pm_thaw,
3973 	.poweroff	= efx_pm_poweroff,
3974 	.restore	= efx_pm_resume,
3975 };
3976 
3977 /* A PCI error affecting this device was detected.
3978  * At this point MMIO and DMA may be disabled.
3979  * Stop the software path and request a slot reset.
3980  */
3981 static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
3982 					      enum pci_channel_state state)
3983 {
3984 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
3985 	struct efx_nic *efx = pci_get_drvdata(pdev);
3986 
3987 	if (state == pci_channel_io_perm_failure)
3988 		return PCI_ERS_RESULT_DISCONNECT;
3989 
3990 	rtnl_lock();
3991 
3992 	if (efx->state != STATE_DISABLED) {
3993 		efx->state = STATE_RECOVERY;
3994 		efx->reset_pending = 0;
3995 
3996 		efx_device_detach_sync(efx);
3997 
3998 		efx_stop_all(efx);
3999 		efx_disable_interrupts(efx);
4000 
4001 		status = PCI_ERS_RESULT_NEED_RESET;
4002 	} else {
4003 		/* If the interface is disabled we don't want to do anything
4004 		 * with it.
4005 		 */
4006 		status = PCI_ERS_RESULT_RECOVERED;
4007 	}
4008 
4009 	rtnl_unlock();
4010 
4011 	pci_disable_device(pdev);
4012 
4013 	return status;
4014 }
4015 
4016 /* Fake a successful reset, which will be performed later in efx_io_resume. */
4017 static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
4018 {
4019 	struct efx_nic *efx = pci_get_drvdata(pdev);
4020 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
4021 
4022 	if (pci_enable_device(pdev)) {
4023 		netif_err(efx, hw, efx->net_dev,
4024 			  "Cannot re-enable PCI device after reset.\n");
4025 		status =  PCI_ERS_RESULT_DISCONNECT;
4026 	}
4027 
4028 	return status;
4029 }
4030 
4031 /* Perform the actual reset and resume I/O operations. */
4032 static void efx_io_resume(struct pci_dev *pdev)
4033 {
4034 	struct efx_nic *efx = pci_get_drvdata(pdev);
4035 	int rc;
4036 
4037 	rtnl_lock();
4038 
4039 	if (efx->state == STATE_DISABLED)
4040 		goto out;
4041 
4042 	rc = efx_reset(efx, RESET_TYPE_ALL);
4043 	if (rc) {
4044 		netif_err(efx, hw, efx->net_dev,
4045 			  "efx_reset failed after PCI error (%d)\n", rc);
4046 	} else {
4047 		efx->state = STATE_READY;
4048 		netif_dbg(efx, hw, efx->net_dev,
4049 			  "Done resetting and resuming IO after PCI error.\n");
4050 	}
4051 
4052 out:
4053 	rtnl_unlock();
4054 }
4055 
4056 /* For simplicity and reliability, we always require a slot reset and try to
4057  * reset the hardware when a pci error affecting the device is detected.
4058  * We leave both the link_reset and mmio_enabled callback unimplemented:
4059  * with our request for slot reset the mmio_enabled callback will never be
4060  * called, and the link_reset callback is not used by AER or EEH mechanisms.
4061  */
4062 static const struct pci_error_handlers efx_err_handlers = {
4063 	.error_detected = efx_io_error_detected,
4064 	.slot_reset	= efx_io_slot_reset,
4065 	.resume		= efx_io_resume,
4066 };
4067 
4068 static struct pci_driver efx_pci_driver = {
4069 	.name		= KBUILD_MODNAME,
4070 	.id_table	= efx_pci_table,
4071 	.probe		= efx_pci_probe,
4072 	.remove		= efx_pci_remove,
4073 	.driver.pm	= &efx_pm_ops,
4074 	.err_handler	= &efx_err_handlers,
4075 #ifdef CONFIG_SFC_SRIOV
4076 	.sriov_configure = efx_pci_sriov_configure,
4077 #endif
4078 };
4079 
4080 /**************************************************************************
4081  *
4082  * Kernel module interface
4083  *
4084  *************************************************************************/
4085 
4086 module_param(interrupt_mode, uint, 0444);
4087 MODULE_PARM_DESC(interrupt_mode,
4088 		 "Interrupt mode (0=>MSIX 1=>MSI 2=>legacy)");
4089 
4090 static int __init efx_init_module(void)
4091 {
4092 	int rc;
4093 
4094 	printk(KERN_INFO "Solarflare NET driver v" EFX_DRIVER_VERSION "\n");
4095 
4096 	rc = register_netdevice_notifier(&efx_netdev_notifier);
4097 	if (rc)
4098 		goto err_notifier;
4099 
4100 #ifdef CONFIG_SFC_SRIOV
4101 	rc = efx_init_sriov();
4102 	if (rc)
4103 		goto err_sriov;
4104 #endif
4105 
4106 	reset_workqueue = create_singlethread_workqueue("sfc_reset");
4107 	if (!reset_workqueue) {
4108 		rc = -ENOMEM;
4109 		goto err_reset;
4110 	}
4111 
4112 	rc = pci_register_driver(&efx_pci_driver);
4113 	if (rc < 0)
4114 		goto err_pci;
4115 
4116 	return 0;
4117 
4118  err_pci:
4119 	destroy_workqueue(reset_workqueue);
4120  err_reset:
4121 #ifdef CONFIG_SFC_SRIOV
4122 	efx_fini_sriov();
4123  err_sriov:
4124 #endif
4125 	unregister_netdevice_notifier(&efx_netdev_notifier);
4126  err_notifier:
4127 	return rc;
4128 }
4129 
4130 static void __exit efx_exit_module(void)
4131 {
4132 	printk(KERN_INFO "Solarflare NET driver unloading\n");
4133 
4134 	pci_unregister_driver(&efx_pci_driver);
4135 	destroy_workqueue(reset_workqueue);
4136 #ifdef CONFIG_SFC_SRIOV
4137 	efx_fini_sriov();
4138 #endif
4139 	unregister_netdevice_notifier(&efx_netdev_notifier);
4140 
4141 }
4142 
4143 module_init(efx_init_module);
4144 module_exit(efx_exit_module);
4145 
4146 MODULE_AUTHOR("Solarflare Communications and "
4147 	      "Michael Brown <mbrown@fensystems.co.uk>");
4148 MODULE_DESCRIPTION("Solarflare network driver");
4149 MODULE_LICENSE("GPL");
4150 MODULE_DEVICE_TABLE(pci, efx_pci_table);
4151 MODULE_VERSION(EFX_DRIVER_VERSION);
4152