xref: /linux/drivers/net/ethernet/sfc/efx_common.c (revision 348cb5dc4d702e071efb4f70c4dacfe2473c815d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /****************************************************************************
3  * Driver for Solarflare network controllers and boards
4  * Copyright 2018 Solarflare Communications Inc.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms of the GNU General Public License version 2 as published
8  * by the Free Software Foundation, incorporated herein by reference.
9  */
10 
11 #include "net_driver.h"
12 #include <linux/module.h>
13 #include <linux/netdevice.h>
14 #include "efx_common.h"
15 #include "efx_channels.h"
16 #include "efx.h"
17 #include "mcdi.h"
18 #include "selftest.h"
19 #include "rx_common.h"
20 #include "tx_common.h"
21 #include "nic.h"
22 #include "io.h"
23 #include "mcdi_pcol.h"
24 
25 static unsigned int debug = (NETIF_MSG_DRV | NETIF_MSG_PROBE |
26 			     NETIF_MSG_LINK | NETIF_MSG_IFDOWN |
27 			     NETIF_MSG_IFUP | NETIF_MSG_RX_ERR |
28 			     NETIF_MSG_TX_ERR | NETIF_MSG_HW);
29 module_param(debug, uint, 0);
30 MODULE_PARM_DESC(debug, "Bitmapped debugging message enable value");
31 
32 /* This is the time (in jiffies) between invocations of the hardware
33  * monitor.
34  * On Falcon-based NICs, this will:
35  * - Check the on-board hardware monitor;
36  * - Poll the link state and reconfigure the hardware as necessary.
37  * On Siena-based NICs for power systems with EEH support, this will give EEH a
38  * chance to start.
39  */
40 static unsigned int efx_monitor_interval = 1 * HZ;
41 
42 /* How often and how many times to poll for a reset while waiting for a
43  * BIST that another function started to complete.
44  */
45 #define BIST_WAIT_DELAY_MS	100
46 #define BIST_WAIT_DELAY_COUNT	100
47 
48 /* Default stats update time */
49 #define STATS_PERIOD_MS_DEFAULT 1000
50 
51 const unsigned int efx_reset_type_max = RESET_TYPE_MAX;
52 const char *const efx_reset_type_names[] = {
53 	[RESET_TYPE_INVISIBLE]          = "INVISIBLE",
54 	[RESET_TYPE_ALL]                = "ALL",
55 	[RESET_TYPE_RECOVER_OR_ALL]     = "RECOVER_OR_ALL",
56 	[RESET_TYPE_WORLD]              = "WORLD",
57 	[RESET_TYPE_RECOVER_OR_DISABLE] = "RECOVER_OR_DISABLE",
58 	[RESET_TYPE_DATAPATH]           = "DATAPATH",
59 	[RESET_TYPE_MC_BIST]		= "MC_BIST",
60 	[RESET_TYPE_DISABLE]            = "DISABLE",
61 	[RESET_TYPE_TX_WATCHDOG]        = "TX_WATCHDOG",
62 	[RESET_TYPE_INT_ERROR]          = "INT_ERROR",
63 	[RESET_TYPE_DMA_ERROR]          = "DMA_ERROR",
64 	[RESET_TYPE_TX_SKIP]            = "TX_SKIP",
65 	[RESET_TYPE_MC_FAILURE]         = "MC_FAILURE",
66 	[RESET_TYPE_MCDI_TIMEOUT]	= "MCDI_TIMEOUT (FLR)",
67 };
68 
69 #define RESET_TYPE(type) \
70 	STRING_TABLE_LOOKUP(type, efx_reset_type)
71 
72 /* Loopback mode names (see LOOPBACK_MODE()) */
73 const unsigned int efx_loopback_mode_max = LOOPBACK_MAX;
74 const char *const efx_loopback_mode_names[] = {
75 	[LOOPBACK_NONE]		= "NONE",
76 	[LOOPBACK_DATA]		= "DATAPATH",
77 	[LOOPBACK_GMAC]		= "GMAC",
78 	[LOOPBACK_XGMII]	= "XGMII",
79 	[LOOPBACK_XGXS]		= "XGXS",
80 	[LOOPBACK_XAUI]		= "XAUI",
81 	[LOOPBACK_GMII]		= "GMII",
82 	[LOOPBACK_SGMII]	= "SGMII",
83 	[LOOPBACK_XGBR]		= "XGBR",
84 	[LOOPBACK_XFI]		= "XFI",
85 	[LOOPBACK_XAUI_FAR]	= "XAUI_FAR",
86 	[LOOPBACK_GMII_FAR]	= "GMII_FAR",
87 	[LOOPBACK_SGMII_FAR]	= "SGMII_FAR",
88 	[LOOPBACK_XFI_FAR]	= "XFI_FAR",
89 	[LOOPBACK_GPHY]		= "GPHY",
90 	[LOOPBACK_PHYXS]	= "PHYXS",
91 	[LOOPBACK_PCS]		= "PCS",
92 	[LOOPBACK_PMAPMD]	= "PMA/PMD",
93 	[LOOPBACK_XPORT]	= "XPORT",
94 	[LOOPBACK_XGMII_WS]	= "XGMII_WS",
95 	[LOOPBACK_XAUI_WS]	= "XAUI_WS",
96 	[LOOPBACK_XAUI_WS_FAR]  = "XAUI_WS_FAR",
97 	[LOOPBACK_XAUI_WS_NEAR] = "XAUI_WS_NEAR",
98 	[LOOPBACK_GMII_WS]	= "GMII_WS",
99 	[LOOPBACK_XFI_WS]	= "XFI_WS",
100 	[LOOPBACK_XFI_WS_FAR]	= "XFI_WS_FAR",
101 	[LOOPBACK_PHYXS_WS]	= "PHYXS_WS",
102 };
103 
104 /* Reset workqueue. If any NIC has a hardware failure then a reset will be
105  * queued onto this work queue. This is not a per-nic work queue, because
106  * efx_reset_work() acquires the rtnl lock, so resets are naturally serialised.
107  */
108 static struct workqueue_struct *reset_workqueue;
109 
110 int efx_create_reset_workqueue(void)
111 {
112 	reset_workqueue = create_singlethread_workqueue("sfc_reset");
113 	if (!reset_workqueue) {
114 		printk(KERN_ERR "Failed to create reset workqueue\n");
115 		return -ENOMEM;
116 	}
117 
118 	return 0;
119 }
120 
121 void efx_queue_reset_work(struct efx_nic *efx)
122 {
123 	queue_work(reset_workqueue, &efx->reset_work);
124 }
125 
126 void efx_flush_reset_workqueue(struct efx_nic *efx)
127 {
128 	cancel_work_sync(&efx->reset_work);
129 }
130 
131 void efx_destroy_reset_workqueue(void)
132 {
133 	if (reset_workqueue) {
134 		destroy_workqueue(reset_workqueue);
135 		reset_workqueue = NULL;
136 	}
137 }
138 
139 /* We assume that efx->type->reconfigure_mac will always try to sync RX
140  * filters and therefore needs to read-lock the filter table against freeing
141  */
142 void efx_mac_reconfigure(struct efx_nic *efx, bool mtu_only)
143 {
144 	if (efx->type->reconfigure_mac) {
145 		down_read(&efx->filter_sem);
146 		efx->type->reconfigure_mac(efx, mtu_only);
147 		up_read(&efx->filter_sem);
148 	}
149 }
150 
151 /* Asynchronous work item for changing MAC promiscuity and multicast
152  * hash.  Avoid a drain/rx_ingress enable by reconfiguring the current
153  * MAC directly.
154  */
155 static void efx_mac_work(struct work_struct *data)
156 {
157 	struct efx_nic *efx = container_of(data, struct efx_nic, mac_work);
158 
159 	mutex_lock(&efx->mac_lock);
160 	if (efx->port_enabled)
161 		efx_mac_reconfigure(efx, false);
162 	mutex_unlock(&efx->mac_lock);
163 }
164 
165 int efx_set_mac_address(struct net_device *net_dev, void *data)
166 {
167 	struct efx_nic *efx = netdev_priv(net_dev);
168 	struct sockaddr *addr = data;
169 	u8 *new_addr = addr->sa_data;
170 	u8 old_addr[6];
171 	int rc;
172 
173 	if (!is_valid_ether_addr(new_addr)) {
174 		netif_err(efx, drv, efx->net_dev,
175 			  "invalid ethernet MAC address requested: %pM\n",
176 			  new_addr);
177 		return -EADDRNOTAVAIL;
178 	}
179 
180 	/* save old address */
181 	ether_addr_copy(old_addr, net_dev->dev_addr);
182 	ether_addr_copy(net_dev->dev_addr, new_addr);
183 	if (efx->type->set_mac_address) {
184 		rc = efx->type->set_mac_address(efx);
185 		if (rc) {
186 			ether_addr_copy(net_dev->dev_addr, old_addr);
187 			return rc;
188 		}
189 	}
190 
191 	/* Reconfigure the MAC */
192 	mutex_lock(&efx->mac_lock);
193 	efx_mac_reconfigure(efx, false);
194 	mutex_unlock(&efx->mac_lock);
195 
196 	return 0;
197 }
198 
199 /* Context: netif_addr_lock held, BHs disabled. */
200 void efx_set_rx_mode(struct net_device *net_dev)
201 {
202 	struct efx_nic *efx = netdev_priv(net_dev);
203 
204 	if (efx->port_enabled)
205 		queue_work(efx->workqueue, &efx->mac_work);
206 	/* Otherwise efx_start_port() will do this */
207 }
208 
209 int efx_set_features(struct net_device *net_dev, netdev_features_t data)
210 {
211 	struct efx_nic *efx = netdev_priv(net_dev);
212 	int rc;
213 
214 	/* If disabling RX n-tuple filtering, clear existing filters */
215 	if (net_dev->features & ~data & NETIF_F_NTUPLE) {
216 		rc = efx->type->filter_clear_rx(efx, EFX_FILTER_PRI_MANUAL);
217 		if (rc)
218 			return rc;
219 	}
220 
221 	/* If Rx VLAN filter is changed, update filters via mac_reconfigure.
222 	 * If rx-fcs is changed, mac_reconfigure updates that too.
223 	 */
224 	if ((net_dev->features ^ data) & (NETIF_F_HW_VLAN_CTAG_FILTER |
225 					  NETIF_F_RXFCS)) {
226 		/* efx_set_rx_mode() will schedule MAC work to update filters
227 		 * when a new features are finally set in net_dev.
228 		 */
229 		efx_set_rx_mode(net_dev);
230 	}
231 
232 	return 0;
233 }
234 
235 /* This ensures that the kernel is kept informed (via
236  * netif_carrier_on/off) of the link status, and also maintains the
237  * link status's stop on the port's TX queue.
238  */
239 void efx_link_status_changed(struct efx_nic *efx)
240 {
241 	struct efx_link_state *link_state = &efx->link_state;
242 
243 	/* SFC Bug 5356: A net_dev notifier is registered, so we must ensure
244 	 * that no events are triggered between unregister_netdev() and the
245 	 * driver unloading. A more general condition is that NETDEV_CHANGE
246 	 * can only be generated between NETDEV_UP and NETDEV_DOWN
247 	 */
248 	if (!netif_running(efx->net_dev))
249 		return;
250 
251 	if (link_state->up != netif_carrier_ok(efx->net_dev)) {
252 		efx->n_link_state_changes++;
253 
254 		if (link_state->up)
255 			netif_carrier_on(efx->net_dev);
256 		else
257 			netif_carrier_off(efx->net_dev);
258 	}
259 
260 	/* Status message for kernel log */
261 	if (link_state->up)
262 		netif_info(efx, link, efx->net_dev,
263 			   "link up at %uMbps %s-duplex (MTU %d)\n",
264 			   link_state->speed, link_state->fd ? "full" : "half",
265 			   efx->net_dev->mtu);
266 	else
267 		netif_info(efx, link, efx->net_dev, "link down\n");
268 }
269 
270 unsigned int efx_xdp_max_mtu(struct efx_nic *efx)
271 {
272 	/* The maximum MTU that we can fit in a single page, allowing for
273 	 * framing, overhead and XDP headroom + tailroom.
274 	 */
275 	int overhead = EFX_MAX_FRAME_LEN(0) + sizeof(struct efx_rx_page_state) +
276 		       efx->rx_prefix_size + efx->type->rx_buffer_padding +
277 		       efx->rx_ip_align + EFX_XDP_HEADROOM + EFX_XDP_TAILROOM;
278 
279 	return PAGE_SIZE - overhead;
280 }
281 
282 /* Context: process, rtnl_lock() held. */
283 int efx_change_mtu(struct net_device *net_dev, int new_mtu)
284 {
285 	struct efx_nic *efx = netdev_priv(net_dev);
286 	int rc;
287 
288 	rc = efx_check_disabled(efx);
289 	if (rc)
290 		return rc;
291 
292 	if (rtnl_dereference(efx->xdp_prog) &&
293 	    new_mtu > efx_xdp_max_mtu(efx)) {
294 		netif_err(efx, drv, efx->net_dev,
295 			  "Requested MTU of %d too big for XDP (max: %d)\n",
296 			  new_mtu, efx_xdp_max_mtu(efx));
297 		return -EINVAL;
298 	}
299 
300 	netif_dbg(efx, drv, efx->net_dev, "changing MTU to %d\n", new_mtu);
301 
302 	efx_device_detach_sync(efx);
303 	efx_stop_all(efx);
304 
305 	mutex_lock(&efx->mac_lock);
306 	net_dev->mtu = new_mtu;
307 	efx_mac_reconfigure(efx, true);
308 	mutex_unlock(&efx->mac_lock);
309 
310 	efx_start_all(efx);
311 	efx_device_attach_if_not_resetting(efx);
312 	return 0;
313 }
314 
315 /**************************************************************************
316  *
317  * Hardware monitor
318  *
319  **************************************************************************/
320 
321 /* Run periodically off the general workqueue */
322 static void efx_monitor(struct work_struct *data)
323 {
324 	struct efx_nic *efx = container_of(data, struct efx_nic,
325 					   monitor_work.work);
326 
327 	netif_vdbg(efx, timer, efx->net_dev,
328 		   "hardware monitor executing on CPU %d\n",
329 		   raw_smp_processor_id());
330 	BUG_ON(efx->type->monitor == NULL);
331 
332 	/* If the mac_lock is already held then it is likely a port
333 	 * reconfiguration is already in place, which will likely do
334 	 * most of the work of monitor() anyway.
335 	 */
336 	if (mutex_trylock(&efx->mac_lock)) {
337 		if (efx->port_enabled && efx->type->monitor)
338 			efx->type->monitor(efx);
339 		mutex_unlock(&efx->mac_lock);
340 	}
341 
342 	efx_start_monitor(efx);
343 }
344 
345 void efx_start_monitor(struct efx_nic *efx)
346 {
347 	if (efx->type->monitor)
348 		queue_delayed_work(efx->workqueue, &efx->monitor_work,
349 				   efx_monitor_interval);
350 }
351 
352 /**************************************************************************
353  *
354  * Event queue processing
355  *
356  *************************************************************************/
357 
358 /* Channels are shutdown and reinitialised whilst the NIC is running
359  * to propagate configuration changes (mtu, checksum offload), or
360  * to clear hardware error conditions
361  */
362 static void efx_start_datapath(struct efx_nic *efx)
363 {
364 	netdev_features_t old_features = efx->net_dev->features;
365 	bool old_rx_scatter = efx->rx_scatter;
366 	size_t rx_buf_len;
367 
368 	/* Calculate the rx buffer allocation parameters required to
369 	 * support the current MTU, including padding for header
370 	 * alignment and overruns.
371 	 */
372 	efx->rx_dma_len = (efx->rx_prefix_size +
373 			   EFX_MAX_FRAME_LEN(efx->net_dev->mtu) +
374 			   efx->type->rx_buffer_padding);
375 	rx_buf_len = (sizeof(struct efx_rx_page_state)   + EFX_XDP_HEADROOM +
376 		      efx->rx_ip_align + efx->rx_dma_len + EFX_XDP_TAILROOM);
377 
378 	if (rx_buf_len <= PAGE_SIZE) {
379 		efx->rx_scatter = efx->type->always_rx_scatter;
380 		efx->rx_buffer_order = 0;
381 	} else if (efx->type->can_rx_scatter) {
382 		BUILD_BUG_ON(EFX_RX_USR_BUF_SIZE % L1_CACHE_BYTES);
383 		BUILD_BUG_ON(sizeof(struct efx_rx_page_state) +
384 			     2 * ALIGN(NET_IP_ALIGN + EFX_RX_USR_BUF_SIZE,
385 				       EFX_RX_BUF_ALIGNMENT) >
386 			     PAGE_SIZE);
387 		efx->rx_scatter = true;
388 		efx->rx_dma_len = EFX_RX_USR_BUF_SIZE;
389 		efx->rx_buffer_order = 0;
390 	} else {
391 		efx->rx_scatter = false;
392 		efx->rx_buffer_order = get_order(rx_buf_len);
393 	}
394 
395 	efx_rx_config_page_split(efx);
396 	if (efx->rx_buffer_order)
397 		netif_dbg(efx, drv, efx->net_dev,
398 			  "RX buf len=%u; page order=%u batch=%u\n",
399 			  efx->rx_dma_len, efx->rx_buffer_order,
400 			  efx->rx_pages_per_batch);
401 	else
402 		netif_dbg(efx, drv, efx->net_dev,
403 			  "RX buf len=%u step=%u bpp=%u; page batch=%u\n",
404 			  efx->rx_dma_len, efx->rx_page_buf_step,
405 			  efx->rx_bufs_per_page, efx->rx_pages_per_batch);
406 
407 	/* Restore previously fixed features in hw_features and remove
408 	 * features which are fixed now
409 	 */
410 	efx->net_dev->hw_features |= efx->net_dev->features;
411 	efx->net_dev->hw_features &= ~efx->fixed_features;
412 	efx->net_dev->features |= efx->fixed_features;
413 	if (efx->net_dev->features != old_features)
414 		netdev_features_change(efx->net_dev);
415 
416 	/* RX filters may also have scatter-enabled flags */
417 	if ((efx->rx_scatter != old_rx_scatter) &&
418 	    efx->type->filter_update_rx_scatter)
419 		efx->type->filter_update_rx_scatter(efx);
420 
421 	/* We must keep at least one descriptor in a TX ring empty.
422 	 * We could avoid this when the queue size does not exactly
423 	 * match the hardware ring size, but it's not that important.
424 	 * Therefore we stop the queue when one more skb might fill
425 	 * the ring completely.  We wake it when half way back to
426 	 * empty.
427 	 */
428 	efx->txq_stop_thresh = efx->txq_entries - efx_tx_max_skb_descs(efx);
429 	efx->txq_wake_thresh = efx->txq_stop_thresh / 2;
430 
431 	/* Initialise the channels */
432 	efx_start_channels(efx);
433 
434 	efx_ptp_start_datapath(efx);
435 
436 	if (netif_device_present(efx->net_dev))
437 		netif_tx_wake_all_queues(efx->net_dev);
438 }
439 
440 static void efx_stop_datapath(struct efx_nic *efx)
441 {
442 	EFX_ASSERT_RESET_SERIALISED(efx);
443 	BUG_ON(efx->port_enabled);
444 
445 	efx_ptp_stop_datapath(efx);
446 
447 	efx_stop_channels(efx);
448 }
449 
450 /**************************************************************************
451  *
452  * Port handling
453  *
454  **************************************************************************/
455 
456 /* Equivalent to efx_link_set_advertising with all-zeroes, except does not
457  * force the Autoneg bit on.
458  */
459 void efx_link_clear_advertising(struct efx_nic *efx)
460 {
461 	bitmap_zero(efx->link_advertising, __ETHTOOL_LINK_MODE_MASK_NBITS);
462 	efx->wanted_fc &= ~(EFX_FC_TX | EFX_FC_RX);
463 }
464 
465 void efx_link_set_wanted_fc(struct efx_nic *efx, u8 wanted_fc)
466 {
467 	efx->wanted_fc = wanted_fc;
468 	if (efx->link_advertising[0]) {
469 		if (wanted_fc & EFX_FC_RX)
470 			efx->link_advertising[0] |= (ADVERTISED_Pause |
471 						     ADVERTISED_Asym_Pause);
472 		else
473 			efx->link_advertising[0] &= ~(ADVERTISED_Pause |
474 						      ADVERTISED_Asym_Pause);
475 		if (wanted_fc & EFX_FC_TX)
476 			efx->link_advertising[0] ^= ADVERTISED_Asym_Pause;
477 	}
478 }
479 
480 static void efx_start_port(struct efx_nic *efx)
481 {
482 	netif_dbg(efx, ifup, efx->net_dev, "start port\n");
483 	BUG_ON(efx->port_enabled);
484 
485 	mutex_lock(&efx->mac_lock);
486 	efx->port_enabled = true;
487 
488 	/* Ensure MAC ingress/egress is enabled */
489 	efx_mac_reconfigure(efx, false);
490 
491 	mutex_unlock(&efx->mac_lock);
492 }
493 
494 /* Cancel work for MAC reconfiguration, periodic hardware monitoring
495  * and the async self-test, wait for them to finish and prevent them
496  * being scheduled again.  This doesn't cover online resets, which
497  * should only be cancelled when removing the device.
498  */
499 static void efx_stop_port(struct efx_nic *efx)
500 {
501 	netif_dbg(efx, ifdown, efx->net_dev, "stop port\n");
502 
503 	EFX_ASSERT_RESET_SERIALISED(efx);
504 
505 	mutex_lock(&efx->mac_lock);
506 	efx->port_enabled = false;
507 	mutex_unlock(&efx->mac_lock);
508 
509 	/* Serialise against efx_set_multicast_list() */
510 	netif_addr_lock_bh(efx->net_dev);
511 	netif_addr_unlock_bh(efx->net_dev);
512 
513 	cancel_delayed_work_sync(&efx->monitor_work);
514 	efx_selftest_async_cancel(efx);
515 	cancel_work_sync(&efx->mac_work);
516 }
517 
518 /* If the interface is supposed to be running but is not, start
519  * the hardware and software data path, regular activity for the port
520  * (MAC statistics, link polling, etc.) and schedule the port to be
521  * reconfigured.  Interrupts must already be enabled.  This function
522  * is safe to call multiple times, so long as the NIC is not disabled.
523  * Requires the RTNL lock.
524  */
525 void efx_start_all(struct efx_nic *efx)
526 {
527 	EFX_ASSERT_RESET_SERIALISED(efx);
528 	BUG_ON(efx->state == STATE_DISABLED);
529 
530 	/* Check that it is appropriate to restart the interface. All
531 	 * of these flags are safe to read under just the rtnl lock
532 	 */
533 	if (efx->port_enabled || !netif_running(efx->net_dev) ||
534 	    efx->reset_pending)
535 		return;
536 
537 	efx_start_port(efx);
538 	efx_start_datapath(efx);
539 
540 	/* Start the hardware monitor if there is one */
541 	efx_start_monitor(efx);
542 
543 	/* Link state detection is normally event-driven; we have
544 	 * to poll now because we could have missed a change
545 	 */
546 	mutex_lock(&efx->mac_lock);
547 	if (efx->phy_op->poll(efx))
548 		efx_link_status_changed(efx);
549 	mutex_unlock(&efx->mac_lock);
550 
551 	if (efx->type->start_stats) {
552 		efx->type->start_stats(efx);
553 		efx->type->pull_stats(efx);
554 		spin_lock_bh(&efx->stats_lock);
555 		efx->type->update_stats(efx, NULL, NULL);
556 		spin_unlock_bh(&efx->stats_lock);
557 	}
558 }
559 
560 /* Quiesce the hardware and software data path, and regular activity
561  * for the port without bringing the link down.  Safe to call multiple
562  * times with the NIC in almost any state, but interrupts should be
563  * enabled.  Requires the RTNL lock.
564  */
565 void efx_stop_all(struct efx_nic *efx)
566 {
567 	EFX_ASSERT_RESET_SERIALISED(efx);
568 
569 	/* port_enabled can be read safely under the rtnl lock */
570 	if (!efx->port_enabled)
571 		return;
572 
573 	if (efx->type->update_stats) {
574 		/* update stats before we go down so we can accurately count
575 		 * rx_nodesc_drops
576 		 */
577 		efx->type->pull_stats(efx);
578 		spin_lock_bh(&efx->stats_lock);
579 		efx->type->update_stats(efx, NULL, NULL);
580 		spin_unlock_bh(&efx->stats_lock);
581 		efx->type->stop_stats(efx);
582 	}
583 
584 	efx_stop_port(efx);
585 
586 	/* Stop the kernel transmit interface.  This is only valid if
587 	 * the device is stopped or detached; otherwise the watchdog
588 	 * may fire immediately.
589 	 */
590 	WARN_ON(netif_running(efx->net_dev) &&
591 		netif_device_present(efx->net_dev));
592 	netif_tx_disable(efx->net_dev);
593 
594 	efx_stop_datapath(efx);
595 }
596 
597 /* Context: process, dev_base_lock or RTNL held, non-blocking. */
598 void efx_net_stats(struct net_device *net_dev, struct rtnl_link_stats64 *stats)
599 {
600 	struct efx_nic *efx = netdev_priv(net_dev);
601 
602 	spin_lock_bh(&efx->stats_lock);
603 	efx->type->update_stats(efx, NULL, stats);
604 	spin_unlock_bh(&efx->stats_lock);
605 }
606 
607 /* Push loopback/power/transmit disable settings to the PHY, and reconfigure
608  * the MAC appropriately. All other PHY configuration changes are pushed
609  * through phy_op->set_settings(), and pushed asynchronously to the MAC
610  * through efx_monitor().
611  *
612  * Callers must hold the mac_lock
613  */
614 int __efx_reconfigure_port(struct efx_nic *efx)
615 {
616 	enum efx_phy_mode phy_mode;
617 	int rc = 0;
618 
619 	WARN_ON(!mutex_is_locked(&efx->mac_lock));
620 
621 	/* Disable PHY transmit in mac level loopbacks */
622 	phy_mode = efx->phy_mode;
623 	if (LOOPBACK_INTERNAL(efx))
624 		efx->phy_mode |= PHY_MODE_TX_DISABLED;
625 	else
626 		efx->phy_mode &= ~PHY_MODE_TX_DISABLED;
627 
628 	if (efx->type->reconfigure_port)
629 		rc = efx->type->reconfigure_port(efx);
630 
631 	if (rc)
632 		efx->phy_mode = phy_mode;
633 
634 	return rc;
635 }
636 
637 /* Reinitialise the MAC to pick up new PHY settings, even if the port is
638  * disabled.
639  */
640 int efx_reconfigure_port(struct efx_nic *efx)
641 {
642 	int rc;
643 
644 	EFX_ASSERT_RESET_SERIALISED(efx);
645 
646 	mutex_lock(&efx->mac_lock);
647 	rc = __efx_reconfigure_port(efx);
648 	mutex_unlock(&efx->mac_lock);
649 
650 	return rc;
651 }
652 
653 /**************************************************************************
654  *
655  * Device reset and suspend
656  *
657  **************************************************************************/
658 
659 static void efx_wait_for_bist_end(struct efx_nic *efx)
660 {
661 	int i;
662 
663 	for (i = 0; i < BIST_WAIT_DELAY_COUNT; ++i) {
664 		if (efx_mcdi_poll_reboot(efx))
665 			goto out;
666 		msleep(BIST_WAIT_DELAY_MS);
667 	}
668 
669 	netif_err(efx, drv, efx->net_dev, "Warning: No MC reboot after BIST mode\n");
670 out:
671 	/* Either way unset the BIST flag. If we found no reboot we probably
672 	 * won't recover, but we should try.
673 	 */
674 	efx->mc_bist_for_other_fn = false;
675 }
676 
677 /* Try recovery mechanisms.
678  * For now only EEH is supported.
679  * Returns 0 if the recovery mechanisms are unsuccessful.
680  * Returns a non-zero value otherwise.
681  */
682 int efx_try_recovery(struct efx_nic *efx)
683 {
684 #ifdef CONFIG_EEH
685 	/* A PCI error can occur and not be seen by EEH because nothing
686 	 * happens on the PCI bus. In this case the driver may fail and
687 	 * schedule a 'recover or reset', leading to this recovery handler.
688 	 * Manually call the eeh failure check function.
689 	 */
690 	struct eeh_dev *eehdev = pci_dev_to_eeh_dev(efx->pci_dev);
691 	if (eeh_dev_check_failure(eehdev)) {
692 		/* The EEH mechanisms will handle the error and reset the
693 		 * device if necessary.
694 		 */
695 		return 1;
696 	}
697 #endif
698 	return 0;
699 }
700 
701 /* Tears down the entire software state and most of the hardware state
702  * before reset.
703  */
704 void efx_reset_down(struct efx_nic *efx, enum reset_type method)
705 {
706 	EFX_ASSERT_RESET_SERIALISED(efx);
707 
708 	if (method == RESET_TYPE_MCDI_TIMEOUT)
709 		efx->type->prepare_flr(efx);
710 
711 	efx_stop_all(efx);
712 	efx_disable_interrupts(efx);
713 
714 	mutex_lock(&efx->mac_lock);
715 	down_write(&efx->filter_sem);
716 	mutex_lock(&efx->rss_lock);
717 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
718 	    method != RESET_TYPE_DATAPATH)
719 		efx->phy_op->fini(efx);
720 	efx->type->fini(efx);
721 }
722 
723 /* Context: netif_tx_lock held, BHs disabled. */
724 void efx_watchdog(struct net_device *net_dev, unsigned int txqueue)
725 {
726 	struct efx_nic *efx = netdev_priv(net_dev);
727 
728 	netif_err(efx, tx_err, efx->net_dev,
729 		  "TX stuck with port_enabled=%d: resetting channels\n",
730 		  efx->port_enabled);
731 
732 	efx_schedule_reset(efx, RESET_TYPE_TX_WATCHDOG);
733 }
734 
735 /* This function will always ensure that the locks acquired in
736  * efx_reset_down() are released. A failure return code indicates
737  * that we were unable to reinitialise the hardware, and the
738  * driver should be disabled. If ok is false, then the rx and tx
739  * engines are not restarted, pending a RESET_DISABLE.
740  */
741 int efx_reset_up(struct efx_nic *efx, enum reset_type method, bool ok)
742 {
743 	int rc;
744 
745 	EFX_ASSERT_RESET_SERIALISED(efx);
746 
747 	if (method == RESET_TYPE_MCDI_TIMEOUT)
748 		efx->type->finish_flr(efx);
749 
750 	/* Ensure that SRAM is initialised even if we're disabling the device */
751 	rc = efx->type->init(efx);
752 	if (rc) {
753 		netif_err(efx, drv, efx->net_dev, "failed to initialise NIC\n");
754 		goto fail;
755 	}
756 
757 	if (!ok)
758 		goto fail;
759 
760 	if (efx->port_initialized && method != RESET_TYPE_INVISIBLE &&
761 	    method != RESET_TYPE_DATAPATH) {
762 		rc = efx->phy_op->init(efx);
763 		if (rc)
764 			goto fail;
765 		rc = efx->phy_op->reconfigure(efx);
766 		if (rc && rc != -EPERM)
767 			netif_err(efx, drv, efx->net_dev,
768 				  "could not restore PHY settings\n");
769 	}
770 
771 	rc = efx_enable_interrupts(efx);
772 	if (rc)
773 		goto fail;
774 
775 #ifdef CONFIG_SFC_SRIOV
776 	rc = efx->type->vswitching_restore(efx);
777 	if (rc) /* not fatal; the PF will still work fine */
778 		netif_warn(efx, probe, efx->net_dev,
779 			   "failed to restore vswitching rc=%d;"
780 			   " VFs may not function\n", rc);
781 #endif
782 
783 	if (efx->type->rx_restore_rss_contexts)
784 		efx->type->rx_restore_rss_contexts(efx);
785 	mutex_unlock(&efx->rss_lock);
786 	efx->type->filter_table_restore(efx);
787 	up_write(&efx->filter_sem);
788 	if (efx->type->sriov_reset)
789 		efx->type->sriov_reset(efx);
790 
791 	mutex_unlock(&efx->mac_lock);
792 
793 	efx_start_all(efx);
794 
795 	if (efx->type->udp_tnl_push_ports)
796 		efx->type->udp_tnl_push_ports(efx);
797 
798 	return 0;
799 
800 fail:
801 	efx->port_initialized = false;
802 
803 	mutex_unlock(&efx->rss_lock);
804 	up_write(&efx->filter_sem);
805 	mutex_unlock(&efx->mac_lock);
806 
807 	return rc;
808 }
809 
810 /* Reset the NIC using the specified method.  Note that the reset may
811  * fail, in which case the card will be left in an unusable state.
812  *
813  * Caller must hold the rtnl_lock.
814  */
815 int efx_reset(struct efx_nic *efx, enum reset_type method)
816 {
817 	bool disabled;
818 	int rc, rc2;
819 
820 	netif_info(efx, drv, efx->net_dev, "resetting (%s)\n",
821 		   RESET_TYPE(method));
822 
823 	efx_device_detach_sync(efx);
824 	efx_reset_down(efx, method);
825 
826 	rc = efx->type->reset(efx, method);
827 	if (rc) {
828 		netif_err(efx, drv, efx->net_dev, "failed to reset hardware\n");
829 		goto out;
830 	}
831 
832 	/* Clear flags for the scopes we covered.  We assume the NIC and
833 	 * driver are now quiescent so that there is no race here.
834 	 */
835 	if (method < RESET_TYPE_MAX_METHOD)
836 		efx->reset_pending &= -(1 << (method + 1));
837 	else /* it doesn't fit into the well-ordered scope hierarchy */
838 		__clear_bit(method, &efx->reset_pending);
839 
840 	/* Reinitialise bus-mastering, which may have been turned off before
841 	 * the reset was scheduled. This is still appropriate, even in the
842 	 * RESET_TYPE_DISABLE since this driver generally assumes the hardware
843 	 * can respond to requests.
844 	 */
845 	pci_set_master(efx->pci_dev);
846 
847 out:
848 	/* Leave device stopped if necessary */
849 	disabled = rc ||
850 		method == RESET_TYPE_DISABLE ||
851 		method == RESET_TYPE_RECOVER_OR_DISABLE;
852 	rc2 = efx_reset_up(efx, method, !disabled);
853 	if (rc2) {
854 		disabled = true;
855 		if (!rc)
856 			rc = rc2;
857 	}
858 
859 	if (disabled) {
860 		dev_close(efx->net_dev);
861 		netif_err(efx, drv, efx->net_dev, "has been disabled\n");
862 		efx->state = STATE_DISABLED;
863 	} else {
864 		netif_dbg(efx, drv, efx->net_dev, "reset complete\n");
865 		efx_device_attach_if_not_resetting(efx);
866 	}
867 	return rc;
868 }
869 
870 /* The worker thread exists so that code that cannot sleep can
871  * schedule a reset for later.
872  */
873 static void efx_reset_work(struct work_struct *data)
874 {
875 	struct efx_nic *efx = container_of(data, struct efx_nic, reset_work);
876 	unsigned long pending;
877 	enum reset_type method;
878 
879 	pending = READ_ONCE(efx->reset_pending);
880 	method = fls(pending) - 1;
881 
882 	if (method == RESET_TYPE_MC_BIST)
883 		efx_wait_for_bist_end(efx);
884 
885 	if ((method == RESET_TYPE_RECOVER_OR_DISABLE ||
886 	     method == RESET_TYPE_RECOVER_OR_ALL) &&
887 	    efx_try_recovery(efx))
888 		return;
889 
890 	if (!pending)
891 		return;
892 
893 	rtnl_lock();
894 
895 	/* We checked the state in efx_schedule_reset() but it may
896 	 * have changed by now.  Now that we have the RTNL lock,
897 	 * it cannot change again.
898 	 */
899 	if (efx->state == STATE_READY)
900 		(void)efx_reset(efx, method);
901 
902 	rtnl_unlock();
903 }
904 
905 void efx_schedule_reset(struct efx_nic *efx, enum reset_type type)
906 {
907 	enum reset_type method;
908 
909 	if (efx->state == STATE_RECOVERY) {
910 		netif_dbg(efx, drv, efx->net_dev,
911 			  "recovering: skip scheduling %s reset\n",
912 			  RESET_TYPE(type));
913 		return;
914 	}
915 
916 	switch (type) {
917 	case RESET_TYPE_INVISIBLE:
918 	case RESET_TYPE_ALL:
919 	case RESET_TYPE_RECOVER_OR_ALL:
920 	case RESET_TYPE_WORLD:
921 	case RESET_TYPE_DISABLE:
922 	case RESET_TYPE_RECOVER_OR_DISABLE:
923 	case RESET_TYPE_DATAPATH:
924 	case RESET_TYPE_MC_BIST:
925 	case RESET_TYPE_MCDI_TIMEOUT:
926 		method = type;
927 		netif_dbg(efx, drv, efx->net_dev, "scheduling %s reset\n",
928 			  RESET_TYPE(method));
929 		break;
930 	default:
931 		method = efx->type->map_reset_reason(type);
932 		netif_dbg(efx, drv, efx->net_dev,
933 			  "scheduling %s reset for %s\n",
934 			  RESET_TYPE(method), RESET_TYPE(type));
935 		break;
936 	}
937 
938 	set_bit(method, &efx->reset_pending);
939 	smp_mb(); /* ensure we change reset_pending before checking state */
940 
941 	/* If we're not READY then just leave the flags set as the cue
942 	 * to abort probing or reschedule the reset later.
943 	 */
944 	if (READ_ONCE(efx->state) != STATE_READY)
945 		return;
946 
947 	/* efx_process_channel() will no longer read events once a
948 	 * reset is scheduled. So switch back to poll'd MCDI completions.
949 	 */
950 	efx_mcdi_mode_poll(efx);
951 
952 	efx_queue_reset_work(efx);
953 }
954 
955 /**************************************************************************
956  *
957  * Dummy PHY/MAC operations
958  *
959  * Can be used for some unimplemented operations
960  * Needed so all function pointers are valid and do not have to be tested
961  * before use
962  *
963  **************************************************************************/
964 int efx_port_dummy_op_int(struct efx_nic *efx)
965 {
966 	return 0;
967 }
968 void efx_port_dummy_op_void(struct efx_nic *efx) {}
969 
970 static bool efx_port_dummy_op_poll(struct efx_nic *efx)
971 {
972 	return false;
973 }
974 
975 static const struct efx_phy_operations efx_dummy_phy_operations = {
976 	.init		 = efx_port_dummy_op_int,
977 	.reconfigure	 = efx_port_dummy_op_int,
978 	.poll		 = efx_port_dummy_op_poll,
979 	.fini		 = efx_port_dummy_op_void,
980 };
981 
982 /**************************************************************************
983  *
984  * Data housekeeping
985  *
986  **************************************************************************/
987 
988 /* This zeroes out and then fills in the invariants in a struct
989  * efx_nic (including all sub-structures).
990  */
991 int efx_init_struct(struct efx_nic *efx,
992 		    struct pci_dev *pci_dev, struct net_device *net_dev)
993 {
994 	int rc = -ENOMEM;
995 
996 	/* Initialise common structures */
997 	INIT_LIST_HEAD(&efx->node);
998 	INIT_LIST_HEAD(&efx->secondary_list);
999 	spin_lock_init(&efx->biu_lock);
1000 #ifdef CONFIG_SFC_MTD
1001 	INIT_LIST_HEAD(&efx->mtd_list);
1002 #endif
1003 	INIT_WORK(&efx->reset_work, efx_reset_work);
1004 	INIT_DELAYED_WORK(&efx->monitor_work, efx_monitor);
1005 	efx_selftest_async_init(efx);
1006 	efx->pci_dev = pci_dev;
1007 	efx->msg_enable = debug;
1008 	efx->state = STATE_UNINIT;
1009 	strlcpy(efx->name, pci_name(pci_dev), sizeof(efx->name));
1010 
1011 	efx->net_dev = net_dev;
1012 	efx->rx_prefix_size = efx->type->rx_prefix_size;
1013 	efx->rx_ip_align =
1014 		NET_IP_ALIGN ? (efx->rx_prefix_size + NET_IP_ALIGN) % 4 : 0;
1015 	efx->rx_packet_hash_offset =
1016 		efx->type->rx_hash_offset - efx->type->rx_prefix_size;
1017 	efx->rx_packet_ts_offset =
1018 		efx->type->rx_ts_offset - efx->type->rx_prefix_size;
1019 	INIT_LIST_HEAD(&efx->rss_context.list);
1020 	efx->rss_context.context_id = EFX_MCDI_RSS_CONTEXT_INVALID;
1021 	mutex_init(&efx->rss_lock);
1022 	efx->vport_id = EVB_PORT_ID_ASSIGNED;
1023 	spin_lock_init(&efx->stats_lock);
1024 	efx->vi_stride = EFX_DEFAULT_VI_STRIDE;
1025 	efx->num_mac_stats = MC_CMD_MAC_NSTATS;
1026 	BUILD_BUG_ON(MC_CMD_MAC_NSTATS - 1 != MC_CMD_MAC_GENERATION_END);
1027 	mutex_init(&efx->mac_lock);
1028 #ifdef CONFIG_RFS_ACCEL
1029 	mutex_init(&efx->rps_mutex);
1030 	spin_lock_init(&efx->rps_hash_lock);
1031 	/* Failure to allocate is not fatal, but may degrade ARFS performance */
1032 	efx->rps_hash_table = kcalloc(EFX_ARFS_HASH_TABLE_SIZE,
1033 				      sizeof(*efx->rps_hash_table), GFP_KERNEL);
1034 #endif
1035 	efx->phy_op = &efx_dummy_phy_operations;
1036 	efx->mdio.dev = net_dev;
1037 	INIT_WORK(&efx->mac_work, efx_mac_work);
1038 	init_waitqueue_head(&efx->flush_wq);
1039 
1040 	efx->tx_queues_per_channel = 1;
1041 	efx->rxq_entries = EFX_DEFAULT_DMAQ_SIZE;
1042 	efx->txq_entries = EFX_DEFAULT_DMAQ_SIZE;
1043 
1044 	efx->mem_bar = UINT_MAX;
1045 
1046 	rc = efx_init_channels(efx);
1047 	if (rc)
1048 		goto fail;
1049 
1050 	/* Would be good to use the net_dev name, but we're too early */
1051 	snprintf(efx->workqueue_name, sizeof(efx->workqueue_name), "sfc%s",
1052 		 pci_name(pci_dev));
1053 	efx->workqueue = create_singlethread_workqueue(efx->workqueue_name);
1054 	if (!efx->workqueue) {
1055 		rc = -ENOMEM;
1056 		goto fail;
1057 	}
1058 
1059 	return 0;
1060 
1061 fail:
1062 	efx_fini_struct(efx);
1063 	return rc;
1064 }
1065 
1066 void efx_fini_struct(struct efx_nic *efx)
1067 {
1068 #ifdef CONFIG_RFS_ACCEL
1069 	kfree(efx->rps_hash_table);
1070 #endif
1071 
1072 	efx_fini_channels(efx);
1073 
1074 	kfree(efx->vpd_sn);
1075 
1076 	if (efx->workqueue) {
1077 		destroy_workqueue(efx->workqueue);
1078 		efx->workqueue = NULL;
1079 	}
1080 }
1081 
1082 /* This configures the PCI device to enable I/O and DMA. */
1083 int efx_init_io(struct efx_nic *efx, int bar, dma_addr_t dma_mask,
1084 		unsigned int mem_map_size)
1085 {
1086 	struct pci_dev *pci_dev = efx->pci_dev;
1087 	int rc;
1088 
1089 	efx->mem_bar = UINT_MAX;
1090 
1091 	netif_dbg(efx, probe, efx->net_dev, "initialising I/O bar=%d\n", bar);
1092 
1093 	rc = pci_enable_device(pci_dev);
1094 	if (rc) {
1095 		netif_err(efx, probe, efx->net_dev,
1096 			  "failed to enable PCI device\n");
1097 		goto fail1;
1098 	}
1099 
1100 	pci_set_master(pci_dev);
1101 
1102 	/* Set the PCI DMA mask.  Try all possibilities from our
1103 	 * genuine mask down to 32 bits, because some architectures
1104 	 * (e.g. x86_64 with iommu_sac_force set) will allow 40 bit
1105 	 * masks event though they reject 46 bit masks.
1106 	 */
1107 	while (dma_mask > 0x7fffffffUL) {
1108 		rc = dma_set_mask_and_coherent(&pci_dev->dev, dma_mask);
1109 		if (rc == 0)
1110 			break;
1111 		dma_mask >>= 1;
1112 	}
1113 	if (rc) {
1114 		netif_err(efx, probe, efx->net_dev,
1115 			  "could not find a suitable DMA mask\n");
1116 		goto fail2;
1117 	}
1118 	netif_dbg(efx, probe, efx->net_dev,
1119 		  "using DMA mask %llx\n", (unsigned long long)dma_mask);
1120 
1121 	efx->membase_phys = pci_resource_start(efx->pci_dev, bar);
1122 	if (!efx->membase_phys) {
1123 		netif_err(efx, probe, efx->net_dev,
1124 			  "ERROR: No BAR%d mapping from the BIOS. "
1125 			  "Try pci=realloc on the kernel command line\n", bar);
1126 		rc = -ENODEV;
1127 		goto fail3;
1128 	}
1129 
1130 	rc = pci_request_region(pci_dev, bar, "sfc");
1131 	if (rc) {
1132 		netif_err(efx, probe, efx->net_dev,
1133 			  "request for memory BAR[%d] failed\n", bar);
1134 		rc = -EIO;
1135 		goto fail3;
1136 	}
1137 	efx->mem_bar = bar;
1138 	efx->membase = ioremap(efx->membase_phys, mem_map_size);
1139 	if (!efx->membase) {
1140 		netif_err(efx, probe, efx->net_dev,
1141 			  "could not map memory BAR[%d] at %llx+%x\n", bar,
1142 			  (unsigned long long)efx->membase_phys, mem_map_size);
1143 		rc = -ENOMEM;
1144 		goto fail4;
1145 	}
1146 	netif_dbg(efx, probe, efx->net_dev,
1147 		  "memory BAR[%d] at %llx+%x (virtual %p)\n", bar,
1148 		  (unsigned long long)efx->membase_phys, mem_map_size,
1149 		  efx->membase);
1150 
1151 	return 0;
1152 
1153 fail4:
1154 	pci_release_region(efx->pci_dev, bar);
1155 fail3:
1156 	efx->membase_phys = 0;
1157 fail2:
1158 	pci_disable_device(efx->pci_dev);
1159 fail1:
1160 	return rc;
1161 }
1162 
1163 void efx_fini_io(struct efx_nic *efx)
1164 {
1165 	netif_dbg(efx, drv, efx->net_dev, "shutting down I/O\n");
1166 
1167 	if (efx->membase) {
1168 		iounmap(efx->membase);
1169 		efx->membase = NULL;
1170 	}
1171 
1172 	if (efx->membase_phys) {
1173 		pci_release_region(efx->pci_dev, efx->mem_bar);
1174 		efx->membase_phys = 0;
1175 		efx->mem_bar = UINT_MAX;
1176 	}
1177 
1178 	/* Don't disable bus-mastering if VFs are assigned */
1179 	if (!pci_vfs_assigned(efx->pci_dev))
1180 		pci_disable_device(efx->pci_dev);
1181 }
1182 
1183 #ifdef CONFIG_SFC_MCDI_LOGGING
1184 static ssize_t show_mcdi_log(struct device *dev, struct device_attribute *attr,
1185 			     char *buf)
1186 {
1187 	struct efx_nic *efx = dev_get_drvdata(dev);
1188 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
1189 
1190 	return scnprintf(buf, PAGE_SIZE, "%d\n", mcdi->logging_enabled);
1191 }
1192 
1193 static ssize_t set_mcdi_log(struct device *dev, struct device_attribute *attr,
1194 			    const char *buf, size_t count)
1195 {
1196 	struct efx_nic *efx = dev_get_drvdata(dev);
1197 	struct efx_mcdi_iface *mcdi = efx_mcdi(efx);
1198 	bool enable = count > 0 && *buf != '0';
1199 
1200 	mcdi->logging_enabled = enable;
1201 	return count;
1202 }
1203 
1204 static DEVICE_ATTR(mcdi_logging, 0644, show_mcdi_log, set_mcdi_log);
1205 
1206 void efx_init_mcdi_logging(struct efx_nic *efx)
1207 {
1208 	int rc = device_create_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
1209 
1210 	if (rc) {
1211 		netif_warn(efx, drv, efx->net_dev,
1212 			   "failed to init net dev attributes\n");
1213 	}
1214 }
1215 
1216 void efx_fini_mcdi_logging(struct efx_nic *efx)
1217 {
1218 	device_remove_file(&efx->pci_dev->dev, &dev_attr_mcdi_logging);
1219 }
1220 #endif
1221 
1222 /* A PCI error affecting this device was detected.
1223  * At this point MMIO and DMA may be disabled.
1224  * Stop the software path and request a slot reset.
1225  */
1226 static pci_ers_result_t efx_io_error_detected(struct pci_dev *pdev,
1227 					      enum pci_channel_state state)
1228 {
1229 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
1230 	struct efx_nic *efx = pci_get_drvdata(pdev);
1231 
1232 	if (state == pci_channel_io_perm_failure)
1233 		return PCI_ERS_RESULT_DISCONNECT;
1234 
1235 	rtnl_lock();
1236 
1237 	if (efx->state != STATE_DISABLED) {
1238 		efx->state = STATE_RECOVERY;
1239 		efx->reset_pending = 0;
1240 
1241 		efx_device_detach_sync(efx);
1242 
1243 		efx_stop_all(efx);
1244 		efx_disable_interrupts(efx);
1245 
1246 		status = PCI_ERS_RESULT_NEED_RESET;
1247 	} else {
1248 		/* If the interface is disabled we don't want to do anything
1249 		 * with it.
1250 		 */
1251 		status = PCI_ERS_RESULT_RECOVERED;
1252 	}
1253 
1254 	rtnl_unlock();
1255 
1256 	pci_disable_device(pdev);
1257 
1258 	return status;
1259 }
1260 
1261 /* Fake a successful reset, which will be performed later in efx_io_resume. */
1262 static pci_ers_result_t efx_io_slot_reset(struct pci_dev *pdev)
1263 {
1264 	struct efx_nic *efx = pci_get_drvdata(pdev);
1265 	pci_ers_result_t status = PCI_ERS_RESULT_RECOVERED;
1266 
1267 	if (pci_enable_device(pdev)) {
1268 		netif_err(efx, hw, efx->net_dev,
1269 			  "Cannot re-enable PCI device after reset.\n");
1270 		status =  PCI_ERS_RESULT_DISCONNECT;
1271 	}
1272 
1273 	return status;
1274 }
1275 
1276 /* Perform the actual reset and resume I/O operations. */
1277 static void efx_io_resume(struct pci_dev *pdev)
1278 {
1279 	struct efx_nic *efx = pci_get_drvdata(pdev);
1280 	int rc;
1281 
1282 	rtnl_lock();
1283 
1284 	if (efx->state == STATE_DISABLED)
1285 		goto out;
1286 
1287 	rc = efx_reset(efx, RESET_TYPE_ALL);
1288 	if (rc) {
1289 		netif_err(efx, hw, efx->net_dev,
1290 			  "efx_reset failed after PCI error (%d)\n", rc);
1291 	} else {
1292 		efx->state = STATE_READY;
1293 		netif_dbg(efx, hw, efx->net_dev,
1294 			  "Done resetting and resuming IO after PCI error.\n");
1295 	}
1296 
1297 out:
1298 	rtnl_unlock();
1299 }
1300 
1301 /* For simplicity and reliability, we always require a slot reset and try to
1302  * reset the hardware when a pci error affecting the device is detected.
1303  * We leave both the link_reset and mmio_enabled callback unimplemented:
1304  * with our request for slot reset the mmio_enabled callback will never be
1305  * called, and the link_reset callback is not used by AER or EEH mechanisms.
1306  */
1307 const struct pci_error_handlers efx_err_handlers = {
1308 	.error_detected = efx_io_error_detected,
1309 	.slot_reset	= efx_io_slot_reset,
1310 	.resume		= efx_io_resume,
1311 };
1312 
1313 int efx_get_phys_port_id(struct net_device *net_dev,
1314 			 struct netdev_phys_item_id *ppid)
1315 {
1316 	struct efx_nic *efx = netdev_priv(net_dev);
1317 
1318 	if (efx->type->get_phys_port_id)
1319 		return efx->type->get_phys_port_id(efx, ppid);
1320 	else
1321 		return -EOPNOTSUPP;
1322 }
1323 
1324 int efx_get_phys_port_name(struct net_device *net_dev, char *name, size_t len)
1325 {
1326 	struct efx_nic *efx = netdev_priv(net_dev);
1327 
1328 	if (snprintf(name, len, "p%u", efx->port_num) >= len)
1329 		return -EINVAL;
1330 	return 0;
1331 }
1332